Research On Evolution Equations Compendium Volume 1

RESEARCH ON EVOLUTION EQUATION COMPENDIUM.
VOLUME 1
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in rendering legal, medical or any other professional services.
RESEARCH ON EVOLUTION EQUATION COMPENDIUM. VOLUME 1
GASTON M. N'GURKATA
EDITOR
Nova Science Publishers, Inc.

New York
Copyright 2009 by Nova Science Publishers, Inc.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without the written permission of the Publisher. For permission to use material from this book please contact us: Telephone 631-231-7269; Fax 631-231-8175 Web Site: http://www.novapublishers.com NOTICE TO THE READER The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers use of, or reliance upon, this material. Any parts of this book based on government reports are so indicated and copyright is claimed for those parts to the extent applicable to compilations of such works. Independent verification should be sought for any data, advice or recommendations contained in this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to persons or property arising from any methods, products, instructions, ideas or otherwise contained in this publication. This publication is designed to provide accurate and authoritative information with regard to the subject matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other professional services. If legal or any other expert assistance is required, the services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS. LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA
ISBN: 978-1-61209-404-5 (E-Book)
Published by Nova Science Publishers, Inc.
New York
CONTENTS
Preface Double Scale Convergence and Homogenization of Quasilinear Parabolic Equations Xuming Xie Second Grade Fluids with Enhanced Viscosity as Dynamical Systems Ahmed S. Bonfoh Periodic Solutions of Some Evolution Equations with Infinite Delay Khalil Ezzinbi and James H.Liu Internal Pollution and Discriminating Sentinel in Population Dynamics Problem O. Nakoulima and S. Sawadogo Center Manifold and Stability in Critical Cases for Some Partial Functional Differential Equations Mostafa Adimy,Khalil Ezzinbi and Jianhong Wu Stability Radii of Positive Linear Functional Differential Systems in Banach Spaces Pham Huu Anh Ngoc, Nguyen Van Minh and Toshiki Naito Effects of Nonlocality and Phase Shift Definitions in Generalizing Levinsons Theorem S.B.Qadri, B. Mulligan, M.F. Mahmood and J. Y. Al-Khal Asymptotic Behavior of the Fitzhugh-Nagumo System Weishi Liu and Bixiang Wang A Partial Differential Equation with Nonautonomous Past Delay in L1-Phase Space S. Boulite, G.Fragnelli, M. Halloumi and L. Maniar Solving the Hyperbolic Problem Obtained by Transmutation Operator Hikmet Koyunbakan ix 1 11 19
29
47
75
99 129
165 183
vi
Contents
Elliptic Operators with Variable Coefficients Generating Fractional Resolvent Families Miao Li, Fu-Bo Li and Quan Zheng Existence Results for Pseudo Almost Periodic Differential, Functional, and Neutral Integral Equations Toka Diagana An Asymptotic Model of a Nonlinear Kelvin-Voigt Viscoelastic Plate Robert P.Gilbert and Robert J. Ronkese Mathematical Analysis of a Bilateral Obstacle Problem for a Class of Second-Order Operators Laurent Lvi and Guy Vallet Application of the Picard Operators to Second Order ODEs Alexandru Bica, Loredana Galea and Sorin Muresan Doubly Nonlinear Degenerate Parabolic Equations on Carnot Groups Junqiang Han and Pengcheng Niu Some Scalar Conservation Laws with Discontinuous Flux Julien Jimenez A New Exact Solution to the Delayed Diffusion Equation P. M. Jordan Level Sets for Reaction Diffusion Equations Amin Boumenir On the Almost Periodicity of the Superposition of Functions Stanislaw Stoiski Time Periodic Solutions for Quasigeostrophic Motion and Their Stability Mei-Qin Zhan Jacobian Feedback Loops Analysis II: Stability and Instability Bourama Toni Existence of Oscillating Solution for Nonlinear State-Dependent Delay Differential Equation Zhixian Yu and Rong Yuan Semilinear Abstract Differential Equations with Deviated Argument Ciprian G. Gal Singular Solutions of a Semi-Linear Elliptic Equation on Nonsmooth Domains Lotfi Riahi
195
205 235
255 277 287 297 317 327 333
343 355
367 381
387
Contents Existence of Weighted Pseudo Almost Periodic Solutions to Some NonAutonomous Differential Equations Toka Diagana A Krasnoselskii-Type Fixed Point Theorem for Multifunctions Defined on a Hyperconvex Space Marcin Borowski Index
vii
397
411 417
PREFACE
Chapter 1 - The behavior of inhomogeneous material, where the inhomogeneities are on a small scale, is of considerable interest in material sciences, mechanical engineering and in many technological applications, especially those involving composite materials. The theory of homogenization was created to model and predict the behavior of such material, when the inhomogeneities are on such a scale much smaller than the linear dimension of the system. In these situations, continuum models, in which rapid oscillation of the material properties is averaged, are sufcient to describe the behavior of the system. Moreover, they have the advantage of avoiding extensive numerical computations needed when the small scale behavior is taken into account. There have been numerous publications on homogenization theory and its applications in the past decades, the authors refer to the books by Bensoussan et al., Sanchez-Palencia and Jikov et al. For homogenization of nonlinear equations, a standard approach is using energy methods by Tartar and Murat; however, in some cases, it is difcult to work out energy method sin was rst introduced by Nguetseng, later elaborated by Allaire; the advantage of this method is that the homogenized equations and proof of convergence are obtained in a single step. Recently Caffarelli proposed an alternative approach in which the homogenization be considered as a viscosity process. Following the idea by Caffarelli, Garcia-Azorero et al. proved that the uniform limit of the solutions at scale e to quasilinear parabolic equations is a viscosity solution to the homogenized equation. The use of viscosity solution methods was rst considered by Evansor divergence structure PDE. In this paper, the authors shall use the method of double-scale convergence to derive the homogerized equation of the quasilinear parabolic equations considered in the references they site. Chapter 2 - The authors consider a modied 2d second grade incompressible uid with a higher order viscosity term, and show that this nonlinear evolution system possesses an exponential attractor. This object gives a more precise description of the asymptotic behaviour of dynamical systems than the global attractor. Chapter 3 - In this work, the authors study the existence of periodic solutions for some partial functional differential equations with innite delay. The authors assume that the linear part is not necessarily densely dened and satises the Hille-Yosida condition, and the phase space is chosen to be C for some decreasing function g from (-,0] to [1,). The authors also present a related Massera type result, namely the existence of a bounded solution on R+ implies the existence of a periodic solution.
Gaston M. NGuerekata
Chapter 4 - The so called Lions sentinel method is applied to some time varying system with two time scale variables and with missing data. As an example, some population dynamics problem with missing data is considered. The two time scale variables represent the running time variable and the age of individuals. This insight seems to be new. Building the object sentinel leads to some null-controllability problem with restricted control variables and two time scale variables. Chapter 5 - In this work, the authors prove the existence of a center manifold for some partial functional differential equations, whose linear part is not necessarily densely dened but satises the Hille-Yosida condition. The attractiveness of the center manifold is also shown when the unstable space is reduced to zero. The authors prove that the ow on the center manifold is completely determined by an ordinary differential equation in a nite dimensional space. In some critical cases, when the exponential stability is not possible, the authors prove that the uniform asymptotic stability of the equilibrium is completely determined by the uniform asymptotic stability of the reduced system on the center manifold. Chapter 6 - In this paper the authors study stability radii of positive linear functional differential systems in Banach spaces under multi-perturbations and multi-afne perturbations. The authors prove that for the class of positive systems, complex stability radius, real stability radius and positive stability radius of positive systems under multiperturbations (or multi-afne perturbations) coincide and they are computed via a simple formula. The authors illustrated the obtained results by an example. Chapter 7 - This paper considers Levinsons Theorem for nonlocal potentials. The effect of continuum bound states is claried, and the presence of spurious states is introduced. Differences between denitions of the phase shift are discussed, with special attention given to discontinuities in the phase shift. Examples from nuclear physics emphasize the utility of this theorem and illustrate some of the more delicate aspects of the derivation. A summary is provided which analyzes certain crucial points in the discussion as they relate to previousattempts to obtain a generalized Levinsons theorem. NUCLEAR REACTIONS Levinsons Theorem, nonlocal potentialscontinuum bound states, spurious states, Jost functions, Fredholm determinants, denition of the phase shift. Chapter 8 - For the FitzHugh-Nagumo system dened on IR, the authors prove the existence of a compact global attractor in a weighted Sobolev space which contains bounded solutions, in particular, traveling wave and spatially periodic solutions. The authors also study the behavior of the global attractors as a parameter goes to zero. Although the limiting system for =0does not possess a bounded attracting set, the authors show that there exists a constant 0 such that global attractors for 0 < = are all contained in a compact subset of the phase space. Furthermore, the authors construct a compact local attractor for the limiting system and establish the upper semicontinuity of the global attractors of perturbed system and the local attractor of the limiting system. Chapter 9 - In this paper, the authors consider partial differential equations with nonautonomous past delay in L1-phase space. Here, the authors consider general delay operators given by Stieltjes integrals. Using the Miyadera-Voigt perturbation result, the authors show the wellposedness of these equations. The authors study also the robustness of some asymptotic properties, as asymptotic stability, almost periodicity and almost automorphy, under the nonautonomous past delay effect. As an application, the authors give a dynamical population equation.
Preface
xi
Chapter 10 - In this paper, the authors solve the problem of constructing the kernels of transmutation operator and give the generalized transmutation operator for a singular problem. Chapter 11 - In this paper, the authors show that under very general conditions, elliptic operator with zero boundary condition can generate fractional resolvent families. Chapter 12 - This paper is a survey devoted to the authors recent results related to the existence and uniqueness of pseudo almost periodic solutions to some abstract differential equations and partial differential equations. Some of those results will be slightly modied and applied to some concrete problems. As main tools, the authors will make extensive use of the method of analytic semigroups, and both the Banach and Zimas xed-point principles. Applications include the study of pseudo almost periodic solutions to some boundary-value problems such as the heat equation with delay as well as the logistic differential equation. Chapter 13 - In the 1970s, Cowin and Hegedus introduced an adaptive elasticity model for bone deposition and reabsorption (remodeling). In 1998, Figueiredo and Trabucho published a paper on an asymptotic model of an adaptive elastic rod. Afterwards, Monnier and Trabucho proved the existence and uniqueness of a solution for this model; Trabucho himself extended this result to a model that contains non-linear strain terms along with linear strain terms in a rate remodeling equation. Here, it is proposed to introduce a Kelvin-Voigt model for the plate that includes non-linear terms rate remodelling terms with respect to strain and the time derivative of strain. As many biomechanists consider cancellous bone to be structure consisting of both rode and plates, adding the study of the plate permitsre a more realistic modeling of the trabeculae of spongy bone. Recent studies have suggested that the dissipation of acoustic energy in cancellous bone interrogation via ultrasound is due to the viscous properties of the trabeculae, rather than that of interstitial blood and marrow. This implies use of a Kelvin-Voigt model for both the plate-like and rod-like trabeculae. Chapter 14 The authors investigate some inner bilateral obstacle problems for a class of strongly degenerate parabolic-hyperbolic quasilinear operators associated with homogeneous Dirichlet data in a multidimensional bounded domain. The authors rst introduce the concept of an entropy process solution, more convenient and generalizing the notion of an entropy solution. Moreover, the boundary conditions are expressed by using the background of Divergence Measure Fields. The authors ensure that proposed denition warrants uniqueness. The existence of an entropy process solution is obtained through the vanishing viscosity and penalization methods. Chapter 15 - Using the Perovs xed point theorem and the ber contractions theorem are obtained sufcient conditions for the smooth dependence by the end points of the solution of the two point boundary value problem for second order ODEs. Finally, an application in the metabolism control is presented. Chapter 16 - In this paper the authors use variational methods to study the nonexistence of positive solutions for the following doubly nonlinear degenerate parabolic equations on Carnot groups:
xii
Gaston M. NGuerekata
Here is a bounded domain with smooth boundary in a Carnot group G G is the horizontal gradient on G, T > 0, V L1loc (), u0 is not identically zero, m R, 1 0. Chapter 17 - The authors deal with the scalar conservation law in a one dimensional bounded domain : tu+ x(k( x) g(u)) = 0, associated with a bounded initial value u0. The function k is supposed to be bounded, discontinuous at { x0= 0} , and with bounded variation. A weak entropy formulation for the Cauchy problem has been introduced by J.D Towers. The existence and the uniqueness is proved by N. Seguin and J. Vovelle through a regularization of the function k. The authors generalize the denition of J.D Towers and the authors adapt the method developed in to establish an existence and uniqueness property in the case of the homogeneous Dirichlet boundary conditions. Chapter 18 - A new analytical solution to the one-dimensional delayed diffusion equation is derived in terms of the Lambert W-function. A new critical value of the thermal lag time parameter is also noted. The effects of varying this parameter are examined using numerical methods. Chapter 19 - The authors are concerned with the shape of the level sets of solution of reaction diffusion equations. Using the maximum principle the authors nd sufcient conditions for their concavity. Chapter 20 - In this note the authors present some theorems on the superposition of a function which satises the Hlder condition and an (NSp)-almost periodic (a.p.) function, as well an Sp-a.p. function. Moreover, the authors prove a theorem on the superposition of a continuous function and an (NH)-a.p. function. In the following, the authors give a theorem on the superposition of a differentiable function and a Vp-a.p. function. Finally, the authors prove a theorem on the superposition of a differentiable function and an (NC(1))-a.p. function. Chapter 21 - In this article, the authors study the quasi geostrophic equation, which is a prototypical geophysical uid model. The authors will show the existence of time-periodic solutions for any the Coriolis parameter and the Ekman dissipation constant with nonhomogeneous boundary conditions. Chapter 22 - The authors investigate the loop stability conditions of dierential systems, that is, the conditions of invariance of the Jacobian spectrum under any variation of entries that leaves unchanged the loop structure. The authors use the dynamical properties of Jacobian Loops described by the products of the matrix entries under cyclic permutations of the indices. It appears that all k-order Feedback Loops given by the union of disjoint simple loops involving k variables must be positive for any asymptotic stable behavior. The authors also conjecture that the Loop structure requires a negative Feedback Loop of the system dimension for the onset of chaotic behavior. Chapter 23 - In this paper sufcient conditions for oscillation of all solutions are given for the equations
Preface
xiii
Chapter 24 - In this paper the authors prove local and global existence results for semilinear differential equations with deviated argument in Banach spaces. Chapter 25 - Under general conditions on the signed Radon measure the authors prove the existence of positive singular solutions for the problem u+up=0, p >1 on bounded NTA domains in Rn,n 2. These results extend the recent ones proved by some authors to more general classes of potentials and domains. A new proof based on a simple xed point argument is also given. Chapter 26 - The paper studies the so-called weighted pseudo almost periodic functions recently introduced by the author. Properties of those weighted pseudo almost periodic functions are discussed including a composition result of weighted pseudo almost periodic functions. The obtained results, subsequently, are utilized to study the existence and uniqueness of a weighted pseudo almost periodic solution to some non-autonomous abstract differential equations. Chapter 27 - The authors present a xed point theorem for a sum of two convex-valued multi functions acting on a weakly compact, hyper convex subset of a normed space. The theorem is a multivalued version of a result of D. Bugajewski.
All the articles have been previously published in Journal of Evolution Equations, Volume 2 by Nova Science Publishers. It was submitted for appropriate modifications in an effort to encourage wider dissemination of research.
In: Research on Evolution Equation Compendium. Volume 1 ISBN: 978-1-61209-404-5 Editor: Gaston M. NGuerekata 2009 Nova Science Publishers, Inc.
D OUBLE S CALE C ONVERGENCE AND H OMOGENIZATION OF Q UASILINEAR PARABOLIC E QUATIONS

Xuming Xie Department of Mathematics, Morgan State University 1700 E. Cold Spring Lane, Baltimore, MD 21251
Abstract We study the homogenization of the nonlinear parabolic equation x t (u )t div x (F(Dx u , , 2 )) = f (t, x) where F(, y) = F(, y + m, s + k) for any integer vectors k, m Zn . We shall use double-scale convergence to derive the homogenized equations and prove a corrector result.
AMS Subject Classication: 35A35 Key Words: Nonlinear parabolic equations; Homogenization; Double-scale convergence; Monotonicity
Introduction
The behavior of inhomogeneous material, where the inhomogeneities are on a small scale, is of considerable interest in material sciences, mechanical engineering and in many technological applications, especially those involving composite materials. The theory of homogenization was created to model and predict the behavior of such material, when the inhomogeneities are on such a scale much smaller than the linear dimension of the system. In these situations, continuum models, in which rapid oscillation of the material properties is averaged, are sufcient to describe the behavior of the system. Moreover, they have the advantage of avoiding extensive numerical computations needed when the small scale behavior is taken into account. There have been numerous publications on homogenization theory and its applications in the past decades, we refer to the books by Bensoussan et al [2], Sanchez-Palencia [11] and Jikov et al [7]. For homogenization of nonlinear equations, a standard approach is using energy methods by Tartar and Murat; however, in some cases, it is difcult to work out energy method since the construction of test functions could be tricky. A two scale convergence method
E-mail
address: xxie@morgan.edu
Xuming Xie
was rst introduced by Nguetseng [10], later elaborated by Allaire [1]; the advantage of this method is that the homogenized equations and proof of convergence are obtained in a single step. Recently Caffarelli [3] proposed an alternative approach in which the homogenization be considered as a viscosity process. Following the idea by Caffarelli, Garcia-Azorero et al [5] proved that the uniform limit of the solutions at scale to quasilinear parabolic equations is a viscosity solution to the homogenized equation. The use of viscosity solution methods was rst considered by Evans [4] for divergence structure PDE. In this paper, we shall use the method of double-scale convergence to derive the homogerized equation of the quasilinear parabolic equations considered in [5]. Double-scale convergence was introduced in [6] and [8]. It is useful for homogenization of linear parabolic equations when the scale of time variable is different from that of space variable. Since it is a generalization of two scale convergence method, the double convergence method has the same advantage as two scale convergence method and is simpler than viscosity process used in [5]. We obtain that solutions to the microscopic equations doublescale converges to solutions of the same homogenized equations as in [5]; we also obtain a corrector result.
Formulation of the Problem
We consider the following nonlinear parabolic equation x t (u )t div x (F(Dx u , , 2 )) = f (t, x) in U where F(, y, s) is a vector eld in Rn Rn R satisfying the following conditions: (1) Periodicity: F(, y, s) is 1-periodic in y and s, i.e., F(, y + m, s + k) = F(, y, s) for all m Zn , k Z. (2) Monotonicity: | |2 < F(, y, s) F(, y, s), > | |2 (2.2) (2.1)
for some positive constants and which are independent of , for all , Rn and for a.e. y and s. (3) Growth: There exists C > 0 such that |F(, y, s)| C|| where C is positive and independent of . (4) regularity: F(, y, s) is continuous in and measurable in y and s. Remark 2.1. Conditions (1) - (3) are the same as in [5], the regularity condition (4) is the well known Caratheodory condition, which is weaker than the regularity condition required in [5]. In [5], F(, y, s) is required to satisfy Lipschitz condition in for a.e (y, s) Rn R. (2.3)
Double Scale Convergence and Homogenization of Quasilinear ...
Lemma 2.2. For f (t, x) L ([0, T ], L2 (U)), there exists a unique solution 1 u L2 ([0, T ], H0 (U))C([0, T ], L2 (U)), t (u ) L2 ([0, T ], H 1 (U)) to equation (2.1) with initial and boundary condition u |t=0 = 0, Furthermore, we have following estimates u
L2 ([0,T ]U)
u |U = 0.
(2.4)
C,
x u
L2 ([0,T ]U)
C,
(2.5) (2.6)
x t F(x u , , 2 )
L2 ([0,T ]U)
C,
where C is a positive constant and independent of . Proof. The existence and uniqueness and (2.5) follow from the standard Galerkin Method, 1 see Theorem 30.A of [12] with evolution triple V = H0 (U), H = L2 (U),V = H 1 (U). (2.6) follows from (2.5) and the growth condition (2.3). Our goal in this paper is to investigate the behavior of the sequence u (t, x) of solutions to (2.1) as 0. We expect that u (t, x) converges to a function u(t, x) in a double scale sense , which will be explained shortly. We are going to derive the effective equations that the function u(t, x) will satisfy. We are also going to estimate how fast that u (t, x) will converge to u(t, x), i.e the so called corrector results.
Double Scale Convergence and the Main Theorem
To investigate the behavior of the sequence u (t, x) as 0, we are going to use the notion of double-scale convergence which is a generalization of two-scale convergence. It was introduced by Holmbom [6] and Goudon and Poupaud [8] to study linear parabolic equations. Let C# (Y ) denote the space of those C functions periodic on Y , where Y = [0, 1] [0, 1]n . Let dZ = dt dx ds dy. Denition 3.1. {u (t, x)} L2 ([0, T ] U) double-scale converges to u(t, x, s, y) L2 ([0, T ] U Y ) iff for any (t, x, s, y) C ([0, T ] U,C# (Y )),
T
0 0
lim
u (t, x) (t, x,
t x , )dxdt = 2
T 0
U Y
Remark 3.2. The above denition uses C test functions ; however as pointed out in [1, 6], test function (t, x, s, y) could be any admissible test function. Functions in spaces 2 L2 ([0, T ] U,C# (Y )), or L# (Y,C([0, T ] U)) are admissible.
1 1 Recall the space W21 ([0, T ], H0 (U), L2 (U)) = {u(t, x) : u L2 ([0, T ], H0 (U)), t u L2 ([0, T ], H 1 (U))}. The following compactness theorem ([6],Theorem 3.1) will be essential:
u(t, x, s, y) (t, x, s, y)dZ.
Xuming Xie
Theorem 3.3. (i) If {u (t, x)} is a bounded sequence in L2 ([0, T ] U)), then there exists 2 u(t, x, s, y) L2 ([0, T ] U, L# (Y )) such that a subsequence of {u (t, x)} double-scale converges to u(t, x, s, y) in the sense of Denition 3.1. 1 (ii) If {u (t, x)} is a bounded sequence in W21 ([0, T ], H0 (U), L2 (U)), then there exist u(t, x) 2 ([0, T ], H 1 (U)) and u (t, x, s, y) L2 ([0, T ] U, L2 ([0, 1], H 1 ([0, 1]n ))) such that a subseL 1 # quence of {u (t, x)} double-scale converges to u(t, x) and a subsequence of x u doublescale converges to x u + y u1 in the sense of Denition 3.1. Proof. See Theorem 3.1 in [6]. Remark 3.4. In [8], double-scale convergence is dened for a sequence of measures; a compact theorem is also obtained (see Proposition 4.4 in [8]). The lemma below is essential in the proof of the main homogenization theorem.
1 Lemma 3.5. Assume that {u } is a bounded sequence in W21 ([0, T ]; H0 (U), L2 (U)), and u(t, x) and u1 (t, x, s, y) are as in Theorem 3.3, then T
0 0
lim
t x 1 (u (t, x) u(t, x))(t, x, 2 , )dt dx =
T 0
U Y
for any (t, x, s, y) = b(t)c(x)p(y)q(s), where b(t) 2 2 L# ([0, 1]n ), q(s) L# ([0, 1]). Proof. See Corollary 3.3 in [6].
C0 ([0, T ]), c(x)
Now we are ready to prove the following main theorem: Theorem 3.6. The solution sequence u (t, x) as in Lemma 2.2 double-scale con1 verges to a function u(t, x) in L2 ([0, T ], H0 (U)), and x u double-scale converges to 1 x u(t, x) + y u1 (t, x, s, y) where (u, u1 ) is the unique solution in L2 ([0, T ], H0 (U)) 2 [0, T ] U; L2 ([0, 1], H 1 ([0, 1]n )) of the homogenized problem L # ut div x
F(x u(t, x) + y u1 , y, s)dyds = f
(u1 )s div y F(x u(t, x) + y u1 , y, s) = 0 u|t=0 = 0, u|U = 0 u1 (t, x, s, y) Y-periodic in (s, y). Proof. The varational formulation of (2.1) is following
T 0
u (T, x)(T, x)dx
u t dx dt
T 0
x t F(u , , 2 ) dx dt = U
T 0
1 for (t, x) W21 ([0, T ], H0 (U), L2 (U)).
u1 (t, x, s, y)(t, x, s, y)dZ

C0 (U), p(y)
(3.1)
in U,
(3.2) (3.3) (3.4) (3.5)
in U Y,
f dx dt (3.6)
From Lemma 2.2 and Theorem 3.3, up to a subsequence, there are u(t, x), u1 (t, x, s, y) and F0 (t, x, y, s) such that u (t, x) double-scale converges to u(t, x) x u (t, x) double-scale converges to x u(t, x) + y u1 (t, x, s, y) x t F(x u , , 2 ) double-scale converges to F0 (t, x, y, s). Let x t (t, x) = b(t)1 (x) + b(t)1 (x, , 2 ), (3.8) (3.7)
1 where 1 (x, y, s) = c(x)p(y)q(s), b(t) C0 ([0, T ]), 1 (x), c(x) H0 (U)), p(y) ([0, 1]n ), q(s) C ([0, 1]). C# # Plugging (3.8) into (3.6) T 0
Note x t u t [b(t)1 (x, , 2 )]dxdt U 0 T x t 1 u b(t)(1 )s (x, , 2 )dxdt + = U 0 T x t 1 (u u)b(t)(1 )s (x, , 2 )dxdt + = U 0

T
T 0
x t u b (t)1 (x, , 2 )dxdt U 0 T x t 1 u(x,t)b(t)(1 )s (x, , 2 )dxdt U 0 T x t + u b (t)1 (x, , 2 )dxdt 0 U T x t 1 x t (u u)b(t)(1 )s (x, , 2 )dxdt + t [b(t)u(x,t)]1 (x, , 2 )dxdt U 0 T x t u b (t)1 (x, , 2 )dxdt. (3.10) + U 0

Taking limit 0 and using Lemma 3.5, we have

T
0 0
lim
x t u t [b(t)1 (x, , 2 )]dxdt = lim 0 U =
T 0
u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ. (3.11)
x t u t [b(t)1 (x) + b(t)1 (x, , 2 )]dxdt U T x t x t F(u , , 2 ) [b(t)x 1 + b(t)y 1 (x, , 2 )]dx dt + U 0 T x t f [b(t)1 (x) + b(t)1 (x, , 2 )]dx dt. (3.9) = 0 U
1 x t (u u)b(t)(1 )s (x, , 2 )dxdt
Xuming Xie Taking double-scale limit in (3.9) and using (3.11), we have
T 0 T 0 T 0
ut b1 (x)
U Y
F0 (t, x, y, s) (b(t)1 + b(t)y 1 (t, x, y, s))dZ =
T 0
In above equation, let 1 = 0 , we obtain:

T 0 T 0
ut b1 (x) +
U Y
b(t)F0 (t, x, y, s) 1 dZ =
T 0
Letting 1 = 0 in (3.12), we have

T 0 T 0
(3.14) Now the question is to identify F0 (t, x, y, s) in terms of u, u1 and F. To this end, for > 0 , (t, x, y) C0 (U Y [0, T ]) and functions uk (t, x, s, y) C0 (U Y [0, T ]) such that 1
1 uk (t, x, s, y) u1 (t, x, s, y) strongly in L2 ([0, T ] U, L2 ([0, 1], H# ([0, 1]n ))), 1
u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ +
U Y
b(t)F0 (t, x, y, s) y 1 (x, y, s)dZ = 0.
let Then
x t x vk (t, x) = x [u(t, x) + uk (t, x, , 2 )] + (t, x, ). 1 vk (t, x) double-scale converges to x u(t, x) + y uk (t, x, y, s) + (t, x, y) vk (t, x, y). 1 (3.16) From monotonicity condition
T 0
x t x t F(u , , 2 ) F(vk , , 2 ) (x u vk ) 0.
Using equation (3.6) with = u ,

T 0
1 x t F(u , , 2 ) u dxdt = 2 U
u2 (T )dx +
T 0
Combining (3.17) and (3.18), 1 2 u2 (T, x) dx +

T 0
x t F(vk , , 2 ) u dxdt U U 0 0 T x t x t F(u , , 2 ) vk dxdt + F(vk , , 2 ) vk dxdt 0. (3.19) U U 0 f u dxdt

Since u (T, x) , up to a subsequence, weakly converges to u(T, x) in L2 (U), it follows

0 U
lim
u2 (T, x) dx
that
u2 (T, x) dx.
U
U U
u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ f b(t)1 (t, x)dtdx. (3.12)
f b(t)1 (t, x)dtdx. (3.13)
(3.15)
(3.17)
f u dx dt.
(3.18)
(3.20)
Double Scale Convergence and Homogenization of Quasilinear ... Taking the double scale limit in (3.19) and using (3.20), we have 1 2 u2 (T, x) dx +
U T 0 T 0 T 0
f u dxdt
U
U Y
U Y
F0 (t, x, y, s) vk dZ +
T 0
U Y
By growth, regularity conditions and Lebesgues dominated convergence theorem, and (3.15)
T
k 0
lim
U Y
F(vk , y, s) [x u + y u1 ]dZ
T 0
U Y
F(x u + y u1 + , y, s) [x u + y u1 ] dZ, (3.22)
k 0
lim
F(vk , y, s) vk dZ
T 0
U Y
U Y
F(u + y u1 + , y, s) [u + y u1 + s] dZ, (3.23)
k 0
lim
U Y
F0 (t, x, y, s) vk dZ
T 0
U Y
F0 (t, x, y) [u + y u1 + ] dZ. (3.24)
Leting k in (3.21) and using (3.22)-(3.24), we have 1 2 u2 (T )dx +

U T 0 T 0 T 0
f u dxdt
U
U Y
F0 (t, x, y, s) x u dZ
+
T 0
U Y
F0 (t, x, y, s) y u1 dZ
U Y
[F(u + y u1 + , y, s) F0 (t, x, y, s)] dZ 0. (3.25)
Replacing in (3.6) by u(t, x), we have

T 0
u (T, x) u(T, x)dx +
u t u dxdt +
T 0
T 0
f u dxdt
U
x t F(u , , 2 ) x u dydxdt = 0. (3.26) U
Taking the double-scale limit in the above

T 0 T 0 T 0
u(T, x)u(T, x)dx+

U
ut u dxdt +
f udxdt
U
U Y
F(vk , y, s) [x u + y u1 ] dZ F(vk , y, s) vk dZ 0. (3.21)
F0 (t, x, y, s)x udZ = 0, (3.27)
8 which is equivalent to 1 2 u2 (T )dx +

U T 0
Xuming Xie
T 0
f u dxdt
U
U Y
Consider a sequence of functions of b(t)1 (x, y, s) that converges strongly to u1 (t, x, s, y) 1 in L2 ([0, T ] U, L2 ([0, 1], H# ([0, 1]n ))). Passing limit in (3.14), we have
T 0
U Y
F0 (t, x, y, s) y u1 dZ =
T 0
U Y n+1
From (3.28) and above , (3.25) becomes

T 0
U Y
[F(u + y u1 + , y, s) F0 (t, x, y, s)] dZ 0.
Letting 0 we have
T 0
U Y
[F(u + y u1 , y, s) F0 (t, x, y, s)] dZ 0,
which implies F0 (t, x, y, s) = F(x u + y u1 , y, s). (3.32)
Finally, (3.2) follows from (3.13) and (3.32); and (3.3) follows from (3.14) and (3.32).
Remark 3.7. Our homogenized equation (3.2) is the same as (II.7) in [5], (3.3) is the same as (II.3) in [5].
1 Theorem 3.8. Assume that u1 (t, x, s, y) L2 ([0, T ] U,C# (Y )), then
lim u (t, x) u(t, x) u1 (x,t,
t x , ) 2
Proof. We consider x t v = x u(t, x) + u1 (x,t, , 2 ) , which double-scale converges to u + y u1 . By monotonicity condition:

T 0
x t x t F(u , , 2 ) F(v , , 2 ) (u v )dtdx
T 0
As in the proof of Theorem 3.6 , the left hand side of (3.33) goes to zero, which implies n that the sequence x u u(t, x) u1 (x,t, x , t2 ) converges to zero in L2 ([0, T ] U) .
F0 (t, x, y, s) x u dZ = 0.
(3.28)
u1 s u1 dZ = 0.
(3.29)
(3.30)
(3.31)
L2 ([0,T ],H 1 (U))
= 0.
|u v |2 dxdt. (3.33)
References
[1] G. Allaire, Homogenization and two-scale convergence, SIAM J. Math. Anal. 23.6 (1992) 1482-1518. [2] A. Bensoussan, J.L. Lions, G. Papanicolau. Asymptotic Analysis for Periodic Structures. (1978) Amsterdam, North Holland. [3] Caffarelli, L. A. A note on nonlinear homogenization. Comm. Pure. Appl. Math. LII (1999) 829-838. [4] L.C. Evans. Periodic homogenization of certain fully nonlinear partial differential equations, Proc. Roy. Soc. Edinbeugh sect. A, 120 (1992) 245-265. [5] J. Garcia-Azorero, C.E. Gutierrez, I. Peral, Homogenization of quasilinear parabolic equations in periodic media, Comm. Partial Diff. Equa. vol 28 (2003) 1887-1910. [6] A. Holmbom, Homogenization of parabolic equations an alternative approach and some corrector-type results. Applications of Mathematics 42(1997) 321-342. [7] V. Jikov, S. Koslov, O. Oleinik, Homogenization of Differential Operators and Integral Functions, (1994) Berlin, Springer. [8] T. Goudon, F. Poupaud, Homogenization of transport equations: weak mean eld approximation. SIAM J Math. Anal. 36(2004) 856-881. [9] F. Murat, L. Tartar. H-convergence, topics in the mathematical modeling of composite materials, L. Cherkaev, R.V. Kohn ed., Progress in nonlinear differential equations and their applications, Birkhauser, Boston, 1998, 21-43. [10] G, Nguetseng, A general convergence result for a functional related to the theory of homogenization, SIAM J. Math. Anal. 20 (1989) 608-623. [11] E. Sanchez-Palencia, Non-homogenous Media and Vibration theory. Lecture notes in physics 127 Berlin, Springer. [12] E. Zeidler, Nonlinear Functional Analysis and Its Application, IIB (1990), Berlin, Springer. [13] X. Xie, A note on homogenization of quasilinear parabolic equations, preprint, 2004.
S ECOND G RADE F LUIDS WITH E NHANCED V ISCOSITY AS DYNAMICAL S YSTEMS

Ahmed S. Bonfoh The Abdus Salam International Centre for Theoretical Physics, Mathematics Section, Strada Costiera 11, 34014 Trieste, Italy
Abstract We consider a modied 2d second grade incompressible uid with a higher order viscosity term, and show that this nonlinear evolution system possesses an exponential attractor. This object gives a more precise description of the asymptotic behaviour of dynamical systems than the global attractor.
Keywords: modied second grade uids, global attractor, exponential attractors. 2000 AMS MS Classications: 35A05, 35B45, 35B40
Introduction
The study of the longtime behaviour of solutions to nonlinear evolution equations arising from mechanics and physics is a capital issue, as it is important, to understand and predict their evolution. Many dissipative systems possess a global attractor, which is a compact set lying in the phase space, and which attracts uniformly the trajectories starting from bounded sets when time goes to innity. In other words, the global attractor is the unique largest bounded invariant set which embodies all the permanent regimes corresponding to all possible initial data. However, the global attractor may present some major defaults for practical purposes. Indeed, the rate of attraction of the trajectories may be small and very sensitive to perturbations. In order to give a more precise description of the longtime behaviour of such systems, C. Foias et al. have introduced a new object called an exponential attractor. This object is a compact and positively invariant set which contains the global attractor, has nite fractal dimension and attracts all the trajectories starting from bounded sets at a uniform exponential rate. Moreover, exponential attractors are more robust than global attractors with respect to approximations and perturbations (see [6]). In this paper, we consider a modied 2d uid of second grade and investigate the existence of exponential attractors.
E-mail
address: abonfoh@ictp.it, Phone: +390402240220, Fax: +39040224163
12
Ahmed S. Bonfoh
Setting of the Problem
We consider a modied second grade incompressible uid in two-dimensional space by adding the higher order viscosity term 2 u, that is,
2 t (u u) + u u + curl (u u) u
= f + p, (2.1)
div u = 0, u| = 0, u|t=0 = u0 .
The function u = u(x,t) is the velocity, p = p(x,t) is the pressure, the density of the uid is supposed to be = 1, f = f (x) is the given external body force, and the parameters , and are given strictly positive constants ( is the kinematic viscosity). The case = 0 corresponds to the usual uids of second grade, which has been studied in [3], [4], [5] and in the references therein. For both cases = 0 and = 0, (2.1) reduces to the Navier-Stokes equations. Let us remark that a superviscosity term is often added for the numerical solution of the turbulent Navier-Stokes equations (see [7]). We assume that the uid lls a simply-connected bounded open set , with smooth and connected boundary of class C 4,1 . We denote by n the unit normal vector to , directed outside . We denote by . and (., .) the usual norm and scalar product in L2 () (and also in L2 ()2 ). We dene the following functional sets: V = {u C 0 ()2 , div u = 0}, H = the 2 ()2 , and V = the closure of V in H 1 ()2 . V is a Hilbert space with closure of V in L 0 the scalar product (u, v)V = (u, v) + (u, v). We also dene the Hilbert space W = {u V, curl(u u) L2 ()} endowed with the scalar product (u, v)W = (u, v)V + (curl(u u), curl(v v)). Note that, in two spatial dimension, the operator curl is a scalar. 1 2 1 2 That is, for u = (u1 , u2 ), we have curl u = u1 u2 and u = ( u1 , u2 ). It has been x x x x 3 () H 1 ()]2 , div u = 0} and also the norms q shown in [3] that W = {u [H H 3 ()2 0 and curl(q q) are equivalent on W . Throughout this paper, the same letter c (and sometimes ci , i = 0, 1, 2, ...) denotes positive constants which may change from line to line.
Existence of the Global Attractor
The existence and uniqueness of solutions of problem (2.1) with = 0 have been investigated, for instance, in [3] and [4]. These results allowed I. Moise et al. to consider the semigroup {S(t)}t0 of operators S(t) : W W, u0 u(t), u(t) being the solution at time t of (2.1) with = 0. They proved that this semigroup is asymptotically compact (it is not compact), weakly continuous and possesses the global attractor in W via energy equations (cf. [9]). Now, when > 0, we can show that the semigroup {S(t)}t0 generated by (2.1) is uniformly compact, continuous, Lipschitz on bounded sets of H 4 ()2 , and possesses the global attractor in W , which is bounded in H 4 ()2 if f H 2 ()2 (see, e.g., [1] and [10] for more details about the theory of global attractors). We state the following results.
Second Grade Fluids with Enhanced Viscosity as Dynamical Systems
13
Theorem 3.1 If u0 W and f H 1 ()2 , then there exists a unique solution u to problem (2.1) satisfying u L (R+ ;W )L2 (0, T ;W H 4 ()2 ), with u L (R+ ;V ), for any T > 0. t Theorem 3.2 We assume that f H 1 ()2 . Then, the semigroup {S(t)}t0 possesses the global attractor A in W . Furthermore, if f H 2 ()2 , then the attractor A is bounded in H 4 ()2 . The proofs of Theorem 3.1 and 3.2 are based on showing, rstly, that curl(u u) L (R+ ; L2 ()) L2 (0, T ; H 1 ()), and then using the equivalence of norms. We can note that the norms curl(u u) H 1 () and u H 4 ()2 are also equivalent on W H 4 ()2 . Here, a new property is the boundedness of the attractor A in H 4 ()2 . This result comes from the existence of a bounded absorbing set B2 for {S(t)}t0 in H 4 ()2 . This is a consequence of showing that, for any u satisfying Theorem 3.1, curl(u u) belongs to a bounded absorbing set of H 1 (), for all time t t2 , for some t2 > 0. Let us give a proof. We set b = curl(uu) and remark that curl(bu) = u.b, for all u W H 4 ()2 . If u is the solution of (2.1) satisfying Theorem 3.1, then b is the unique function in L (R+ ; L2 ()) L2 (0, T ; H 1 ()), for all T > 0, solution of the following system (see [4]):
b t b + u.b + ( ) curl u = curl f , b|t=0 = curl(u0 u0 ).
(3.1)
We consider a special basis {w j } j1 of H 1 (), which is orthogonal in L2 (), where w j , j N , is the solution of the homogeneous Neumann problem:
w j w j = j w j , w j n = 0,
(3.2)
and where 0 < 1 2 3 ..., j + for j +, { j } j1 being the family of eigenvalues associated with the family of eigenfunctions {w j } j1 . We can note that w j is also the solution of the following variational problem:
(w j , v) + (w j , v) = j (w j , v), v H 1 ().
(3.3)
Since is regular enough, w j belongs to H 2 (), and, the norms w w and w H 2 () are equivalent on H 2 (). We now implement a Galerkin method with the basis {w j } j1 . Noting that (u.b, b) = 0 (see [3]), we formally obtain 1 d 2 dt ( b 2 + b 2 ) + b b 2 + ((u.b), b) +( )[ ( curl u, b) + ( curl u, b)] = b 2 + b 2 + ( curl f , b) + (curl f , b).
(3.4)
Therefore,
d dt ( b 2 + b 2 ) + c1 b 2 2 () c2 ( b 2 + b 2 ) H + u 2 () b 2 2 () + c3 u 2 4 ()2 + c4 f 2 2 ()2 , L H H H
(3.5)
14
Ahmed S. Bonfoh
for any > 0. If u0 W and f H 1 ()2 , then u satises Theorem 3.1, and u(t) L () c1 u(t) L () c2 u(t) W c, t 0. Using this latter fact together with a proper choice of , equation (3.5) reduces to
b 2 + b 2 ) + c1 b c2 ( b 2 + b 2 ) + c3 f d dt ( 2 H 2 () 2 . H 2 ()2
(3.6)
If f H 2 ()2 , then we deduce the existence of a bounded absorbing set B2 for {S(t)}t0 in H 4 ()2 , by applying the uniform Gronwall lemma to (3.6), and using the equivalence of norms.
Existence of Exponential Attractors
Let E be a metric space, X a compact subset of E and consider a continuous semigroup {S(t)}t0 , on E, mapping X into X. We recall the denition of an exponential attractor for {S(t)}t0 . Denition 4.1 A compact set M is called an exponential attractor for {S(t)}t0 for the topology of E if: - M contains the global attractor; that is, A M ; - M is positively invariant under S(t); that is, S(t)M M , t 0; - the fractal dimension of M is nite; - there exists a constant c0 > 0 such that, for every bounded subset B X, there exists a constant c1 (B) > 0 such that distE (S(t)B, M ) c1 ec0t , t 0, where distE is the Hausdorff semi-distance with respect to the metric of E: distE (A, B) = sup inf a b
aA bB E.
Sufcient conditions ensuring the existence of exponential attractors in Hilbert spaces are given in [6]. It depends on a dichotomy principle called the squeezing property. Denition 4.2 The semigroup {S(t)}t0 veries the squeezing property on X if, for a real 1 number belonging to [0, 4 [, there exists a projection PN : E E, with nite rank N (), and a time t such that if (, ) X 2 , (I PN )(S(t ) S(t )) E PN (S(t ) S(t )) then S(t ) S(t ) E E .
E,
(4.1)
Proposition 4.1 If {S(t)}t0 satises the squeezing property on X and if S(t ) is Lipschitz on X with Lipschitz constant L, then there exists an exponential attractor M for {S(t)}t0 on X such that the fractal dimension of M is bounded as follows: dF (M ) N max(1, ln(16L+1) ) . ln 2 (4.2)
15
In the order to verify the squeezing property, Babin and Nicolaenko introduced in [2] a method based on a decomposition of the difference of two trajectories. A consequence of their result is stated as follows (see [8]). Proposition 4.2 Let E and V be two Hilbert spaces such that the inclusion V E is compact. Let X E be a closed set and S(t) : X X be a semigroup. Let us furthermore assume that there exists a projection PN with nite rank N such that (I PN )y E c(N ) y V , y V , where c(N ) 0 as N +. If there exists 1 , 2 satisfying S(t) S(t) = 2 1 (t) + 2 (t) and 1 (t) 2 d(t) 2 , 2 (t) V h(t) 2 , , X, where E E E d(t) is continuous and satises limt+ d(t) = 0 and h(t) is continuous, then {S(t)}t0 enjoys the squeezing property on X for the topology of E. We now want to apply this result to show the existence of an exponential attractor on the W set X = tt1 S(t)B2 , where B2 is a bounded absorbing set for {S(t)}t0 in H 4 ()2 and t1 is such that S(t)B2 B2 , t t1 . The set X is compact and positively invariant by S(t). Let u1 and u2 be two solutions of (2.1) satisfying Theorem 3.1. We then write u = u1 u2 as sum of two functions u = u1 + u2 , where u1 and u2 are the solutions of the following problems:
and
2 2 2 2 t (u u ) (u u ) + curl(u1 u1 ) u1 curl(u2 u2 ) u2 + ( )u = 0, 2 = 0, div u = 0, u2 |t=0 = 0,
respectively. Similarly to the decomposition of u, we consider the difference b = b1 b2 of the two corresponding solutions of (3.1) and write b = b1 + b2 , where
b1 1 1 t + b b = 0, 1| b t=0 = curl(u0 u0 ),
and
b2 t + b2 b2 + u1 .b1 u2 .b2 + ( ) curl u b = 0, 2| b t=0 = 0.
We formally take the L2 scalar product of (4.5) with b1 , and obtain

d dt
Therefore,
d dt
1 1 1 1 t (u u ) (u u ) 1 = 0, div u
= 0, (4.3)
u1 | = 0, u1 |t=0 = u0 ,
(4.4)
u2 |
(4.5)
(4.6)
b1
2 +2
b1
2 + 2
b1
= 0.
(4.7)
b1
2 +c
b1
0,
(4.8)
16 and
Ahmed S. Bonfoh
Finally, we deduce from the equivalence of norms previously noticed that

W
We now note that u1 .b1 u2 .b2 = u1 .b + u.b2 , and equation (4.6) becomes
b2 t + b2 b2 + u1 .b + u.b2 + ( ) curl u b = 0, b2 |t=0 = 0.
Still working with the basis {w j } j1 considered in Section 3, we obtain, by proceeding as for (3.4),
1 d 2 dt ( b2 2 + b2 2 ) + b2 b2 2 + ((u1 .b), b2 ) +(u1 .b, b2 ) + ((u.b2 ), b2 ) + (u.b2 , b2 ) +( )[ ( curl u, b2 ) + ( curl u, b2 )] = (b, b2 ) + (b, b2 ). 2+ b2 2 ) + c1 b2 2 2 () c2 ( b2 2 + b2 2 ) H +1 ( u1 2 () + u1 2 () ) b 2 2 () L L H +2 ( u 2 () + u 2 () ) b2 2 2 () L L H +3 u 2 4 ()2 + 4 ( b 2 + b 2 ), H 2 H 4 ()2
Therefore,
d dt (
b2
for any 1 , 2 , 3 , 4 > 0. We note that u
where Y = H 1 () or H 2 () . According to the results in Section 3, if u01 , u02 X, then u1 (t), u2 (t) X, t 0. We recall that X is a bounded set of H 4 ()2 . This is equivalent to the fact that b1 (t) and b2 (t) belong to a bounded set of H 1 (). We also have t 2 0 bi H 2 () dt c(t), i = 1, 2, for all t 0. Using these boundedness and choosing properly constants 1 , 2 , 3 and 4 , we get
b2 2 + b2 2 ) + c1 b2 2 2 () H c2 (1 + b2 2 2 () )( b2 2 + b2 2 ) H +c3 b1 2 b2 2 2 () + c4 b1 2 2 () . H H d dt (
Finally, we obtain by applying the Gronwall lemma that

2 ds], H 2 ()
Noting (4.9) and the equivalence of norms, we get u2 (t)

H 4 ()2
c1 e
Both estimates (4.10) and (4.16) allow us to say that {S(t)}t0 enjoys the squeezing property on X. Moreover, S(t) is Lipschitz on X; and we consider PN as the projection on the space generated by the N rst eigenfunctions of the operator on V H 2 ()2 . We then have the following result. Theorem 4.1 The semigroup {S(t)}t0 possesses an exponential attractor M on X.
c2
t 0 (1+
b2
2 )d H 2 ()
b2 (t)
2 H 1 ()
c1 e 0 t + 0 b1
c2
(1+ b2
u1 (t)
c1 ect u0
W,
b1 (t) c1 ect b0 ,
t 0
b1 u1
2 d H 1 ()
c b0 2 , t 0. c2 u0
2 W,
(4.9)
t 0
2 d H 4 ()2
t 0.
(4.10)
(4.11)
(4.12)
(4.13)
c b
2 , H 1 ()
and b
2 Y
b1
2 Y
+ b2
2 Y,
(4.14)
2 )d H 2 ()
t [ 0 b1 t > 0.
b2
2 ds H 2 ()
(4.15)
u0
W,
t > 0.
(4.16)
17
Acknowledgments
I am currently an Abdus Salam International Centre for Theoretical Physics Post-Doctoral Research Fellow, and I would like to thank the Centre for all its support.
References
[1] A.V. Babin and M.I. Vishik, Attractors of evolution equations, North-Holland, Amsterdam, London, New York, Tokyo, 1992. [2] A. Babin and B. Nicolaenko, Exponential attractors of reaction-diffusion systems in an unbounded domain, J. Dyn. Diff. Equ. 7 (1995), 567-589. [3] D. Cioranescu and E.H. Ouazar, Existence and uniqueness for uids of second grade, Nonlin. Part. Diff. Equ., Coll` ge de France, Pitman 109 (1984), 178-197. e [4] D. Cioranescu and V. Girault, Weak and classical solutions of a family of second grade uids, Int. J. Non-Linear Mechanics 32(2) (1997), 317-335. [5] J.E. Dunn and R.L. Fosdick, Thermodynamics, stability and boundedness of uids of complexity two and uids of second grade, Arch. Rat. Mech. Anal. 56(3) (1974), 191-252. [6] A. Eden, C. Foias, B. Nicolaenko and R. Temam, Exponential attractors for dissipative evolution equations, Masson, Paris, 1994. [7] J.L. Lions, Quelques M thodes de R solutions des Probl` mes aux Limites Non e e e Lin aires, Dunod, Paris, 1969. e [8] A. Miranville, Exponential attractors for a class of evolution equations by a decomposition method, C. R. Acad. Sci. Paris, 328(1) (1999), 145-150. [9] I. Moise, R. Rosa and X. Wang, Attractors for non-compact semigroups via energy equations, Nonlinearity 11(7) (1998), 1369-1393. [10] R. Temam, Innite Dimensional Dynamical Systems in Mechanics and Physics, (2nd Edition) Springer-Verlag, Berlin, Heidelberg, New York, 1997.
P ERIODIC S OLUTIONS OF S OME E VOLUTION E QUATIONS WITH I NFINITE D ELAY

Khalil Ezzinbi 1and James H. Liu 2 des Sciences Semlalia, D epartement de Math matiques, B.P. 2390, Marrakesh, Morocco e 2 Department of Mathematics and Statistics James Madison University, Harrisonburg, VA 22807, USA
1 Facult e
Abstract In this work, we study the existence of periodic solutions for some partial functional differential equations with innite delay. We assume that the linear part is not necessarily densely dened and satises the Hille-Yosida condition, and the phase space is chosen to be Cg for some decreasing function g from (, 0] to [1, ). We also present a related Massera type result, namely the existence of a bounded solution on R+ implies the existence of a periodic solution.
Keywords and phrases : Periodic solutions, innite delay, Kuratowskis measure of noncompactness, condensing map.
1 Introduction
The purpose of this work is to study the existence of periodic solutions for the following partial functional differential equation with innite delay u (t) = Au(t) + f (t, ut ), t > 0, u(s) = (s), s 0, (1.1)
where A is an unbounded linear operator on a Banach space X. Here, the operator A is not necessarily densely dened on X, and satises the Hille-Yosida condition: there exist M0 1 and 0 R such that (0 , +) (A) and |R (, A)n | M0 , for n N and > 0 , ( 0 )n
where (A) is the resolvent set of A and R (, A) = ( A)1 . The initial data belongs to some phase space Cg to be dened later.
This
research is supported by a TWAS Research Grant under contract: No 03-030 RG/MATHS/AF/AC address: ezzinbi@ucam.ac.ma E-mail address: liujh@jmu.edu
E-mail
20
Khalil Ezzinbi and James H. Liu
The function f : R+ Cg X is continuous and T periodic in t. As usual, the history function ut Cg is dened by ut (s) = u(t + s), s 0. Periodic solutions are derived for Eq.(1.1) by [5] in a general phase space with axioms when A is densely dened; by [6, 7] in Cg when A is time dependent and densely dened; and by [4] in a general phase space with axioms when A is non-densely dened. In this work we use the techniques developed in [4, 5, 6, 7] to present an extension of [4, 5, 6, 7] by studying Eq.(1.1) in Cg when A is non-densely dened. This work is organized as follows: In Section 2, we recall some results about the existence, uniqueness, and estimations of solutions. In Section 3, we prove that the Poincar e map in condensing, which is then used to derive the existence of periodic solutions by using certain boundedness of solutions. In section 4, we prove a Massera type theorem when f is linear with respect to the second argument, showing that the existence of a bounded solution on IR+ is sufcient for the existence of periodic solutions.
Solutions and Their Estimations
First, we make the following assumptions: (H1 ) A satises the Hille-Yosida condition. (H2 ) f is continuous, and Lipschitzian with respect to the second argument. Lemma 2.1. [3] Let A0 be the part of the operator A in D(A), which is dened by D(A0 ) = {x D(A) : Ax D(A)}, A0 x = Ax. Then A0 generates a C0 -semigroup (T0 (t))t0 on D(A). Now, we dene the phase space Cg for Eq.(1.1) as in [6]. Lemma 2.2. [6] There exists an integer K0 > 1 such that 1 ( )K0 1 M0 < 1, 2
T where M0 = supt[0,T ] |T0 (t)|. Next, let w0 = K0 , then there exists a function g on (, 0] such that g(0) = 1, g() = , g is decreasing on (, 0], and for d w0 one has
sup
s0
1 g(s) . g(s d) 2
For the function g given in the above lemma, we dene the continuous functions space Cg Cg = : C((, 0]; X) and lim |()| =0 . g()
Periodic Solutions of Some Evolution Equations with Innite Delay If we provide Cg with the norm ||g = sup
s0
21
|(s)| , Cg , g(s)
then Cg is a Banach space. Next, we recall some results concerning existence and uniqueness of integral solutions for Eq.(1.1). Denition 2.3. [1] Let Cg . A function u : R X is called an integral solution of Eq.(1.1) if the following conditions hold: (i) u is continuous on [0, ), (ii) u0 = , t (iii) 0 u (s) ds D (A), for t 0, t t (iv) u (t) = (0) + A 0 u (s) ds + 0 f (s, us )ds, for t 0. Theorem 2.4. [2] Assume that (H1 ) and (H2 ) hold. Then, for all Cg such that (0) D (A), Eq.(1.1) has a unique integral solution u. Moreover, let u and v be two integral solutions of Eq.(1.1) on (, L], L > 0. Then for t [0, L], |ut vt |g M1 |u0 v0 |g ek1t , where M1 and k1 are some positive constants. The integral solution of Eq.(1.1) is given by the following variation of constants formula t T0 (t) (0) + lim T0 (t s) B f (s, us )ds , t 0, u (t) = + 0 (t), t 0,

where B = ( A)1 , and u(t) D(A), t 0. If D(A) = X, then the integral solutions coincide with the known mild solutions given in [5] and references therein. Later on, we will follow [4, 5, 6] and call integral solutions as solutions. As an immediate consequence of the above theorem, we conclude that solutions of Eq.(1.1) are locally bounded in t and . We can get more information on the estimation of solutions. Lemma 2.5. [6] Let u be a continuous function on (, T ] such that u0 Cg . Then for any 0 h < r T with r h w0 (w0 is from Lemma 2.2), one has |ur |g max sup |u(s)| ,
s[h,r]
1 |uh |g . 2
In the sequel, we examine the Kuratowskis measure of non-compactness, which will be used in the next section to study periodic solutions via xed points of a condensing operator. The Kuratowskis measure of non-compactness (or the measure) for a bounded set H of a Banach space Y with norm | |Y is dened as (H) = inf d > 0 : H has a nite cover of diameter < d . The following are some basic properties of the measure of non-compactness.
22
Lemma 2.6. Let B1 and B2 be bounded sets of a Banach space Y . Then (B1 ) dia(B1 ). (dia(B1 ) = sup{|x y|Y : x, y B1 }.) (B1 ) = 0 if and only if B1 is precompact. (B1 ) = ||(B1 ), . (B1 = {x : x B1 }) (B1 B2 ) = max{(B1 ), (B2 )}. (B1 + B2 ) (B1 ) + (B2 ). (B1 + B2 = {x + y : x B1 , y B2 }) (B1 ) (B2 ) if B1 B2 . In the sequel, we assume that (H3 ) T0 (t) is compact on D(A) whenever t > 0. 0 0 Next, dene Cg = { Cg : (0) D(A)}. For D Cg and u() the unique solution with u0 () = , we dene Wl (D) = {ul () : D} and W[h,r] (D) = {u[h,r] () : D}, where u[h,r] means the restriction of u on [h, r].
0 Lemma 2.7. [6] Assume that (H1 ), (H2 ) and (H3 ) are satised and let D Cg . Then for any 0 h < r T with r h w0 (w0 is from Lemma 2.2), one has
(Wr (D)) max (W[h,r] (D)),
1 (Wh (D)) . 2
0 Lemma 2.8. Let D Cg be bounded. Then (W[h,r] (D)) = 0 for any 0 < h < r T .
Proof. Let 0 < h < r T. For D and t [h, r] , one has u(t, ) = T0 (t)(0) + lim Dene w by w(t, ) = lim Then
t + 0
+ 0
T0 (t s)B f (s, us (, ))ds, for D for t [h, r] .
+ 0
lim
T0 (t s)B f (s, us (, ))ds = T0 () lim + lim
+ 0 t
+ t
Since T0 () is compact, there exists a compact set W such that T0 ()

+ 0
lim
T0 (t s)B f (s, us (, ))ds : D W .
From the boundedness of f , there exists a positive constant a such that lim
+ t
T0 (t s)B f (s, us (, ))ds a, uniformly in D.
T0 (t s)B f (s, us (, ))ds.
T0 (t s)B f (s, us (, ))ds
T0 (t s)B f (s, us (, ))ds.
Periodic Solutions of Some Evolution Equations with Innite Delay
23
We deduce that the set {w(t, ) : D} is totally bounded and therefore is relatively compact in X. To prove the equicontinuity, let 0 < h t0 < t r. Then |w(t, ) w(t0 , )| lim
+ (T0 (t t0 ) I) lim
+ 0
There exists a positive constant b such that lim
+ t0
T0 (t s)B f (s, us (, ))ds b(t t0 ), uniformly in D.
Moreover, W0 =
t0
+ 0
lim
T0 (t0 s)B f (s, us (, ))ds : D
and its well known that

h0+
lim (T0 (h) I)u = 0, uniformly in u W0 .
On the other hand, T0 (t)(0) = T0 (t h)T0 (h)(0). (2.1) Since T0 (h) is compact, using (2.1) we get that the family {T0 ()(0) : D} is equicontinuous in C([h, r] ; X). Consequently, we deduce that
tt0 0+
lim |u(t, ) u(t0 , )| = 0, uniformly in D.
It follows that u(, )|[h,r] : D is equicontinuous. By Arz la-Ascolis theorem, we e deduce that u(, )|[h,r] : D is relatively compact in C([h, r] ; X). Hence the result is true by using Lemma 2.6.
Existence of Periodic Solutions
In this section, we study periodic solutions of Eq.(1.1). For this purpose, we use the Poincar map P dened by e 0 P() = uT (, ), Cg , where u(, ) denotes the (integral) solution of Eq.(1.1) with u0 (, ) = . In the sequel, we will show that P has a xed point. We note that a xed point of P gives rise to a periodic solution. First, we prove that the operator P is condensing. Theorem 3.1. Assume that (H1 ), (H2 ) and (H3 ) hold. Then the Poincar map P is cone 0 densing in Cg .
+ t0
T0 (t s)B f (s, us (, ))ds

t0
T0 (t0 s)B f (s, us (, ))ds .
is relatively compact
24 Proof. Since
|P(1 ) P(2 )|g = |uT (1 ) uT (2 )|g M1 ek1 T |1 2 |g ,

0 it implies that P is continuous and takes bounded sets into bounded sets. Let D Cg be bounded with (D) > 0. Then, using lemmas 2.7 and 2.8 (w0 is from Lemma 2.2),
(P(D)) = (WT (D)) max (W[T w0 ,T ] (D)),
1 (WT w0 (D)) 2
1 = (WT w0 (D)) 2 1 1 max (W[T 2w0 ,T w0 ] (D)), (WT 2w0 (D)) 2 2 1 2 = ( ) (WT 2w0 (D)) 2 1 2 1 ( ) max (W[T 3w0 ,T 2w0 ] (D)), (WT 3w0 (D)) 2 2 1 3 = ( ) (WT 3w0 (D)) 2 ...... 1 1 ( )K0 1 max (W[0,w0 ] (D)), (D) . 2 2 Next, for [0, w0 ], W[0,w0 ] (D) {T0 ()(0) : D} + And for t [0, w0 ], |T0 (t)1 (0) T0 (t)2 (0)| M0 |1 (0) 2 (0)| M0 |1 2 |g ,
+ 0
(3.1)
lim
where M0 = supt[0,T ] |T0 (t)|. Consequently, using an argument similar to that in [6], we see that for [0, w0 ], one has {T0 ()(0) : D} M0 (D). Also, we can see that for [0, w0 ], { lim Therefore we have that (W[0,w0 ] (D)) M0 (D). Thus, we have 1 1 (P(D)) ( )K0 1 max M0 (D), (D) 2 2 1 ( )K0 1 M0 (D) < (D). 2 This completes the proof. To derive periodic solutions, we list the following xed point theorems.
+ 0
T0 ( h)B f (h, uh ())dh : D} = 0.
T0 ( h)B f (h, uh ())dh : D .
(3.2)
(3.3)
25
Theorem 3.2 (Sadovskiis xed point theorem). [8] Let P be a condensing operator on a Banach space Y , i.e., P is continuous and takes bounded sets into bounded sets, and (P(B)) < (B) for every bounded set B of Y with (B) > 0. If P(H) H for a convex, closed and bounded set H of Y , then P has a xed point in H. Theorem 3.3. [7] Suppose S0 S1 S2 are convex bounded subsets of a Banach space Y , S0 and S2 are closed, and S1 is open in S2 , and suppose P is a condensing operator in Y . If P j (S1 ) S2 , j 0, and there is a number N(S1 ) such that Pk (S1 ) S0 , k N(S1 ), then P has a xed point. Note that the Sadovskiis xed point theorem requires that the Poincar operator maps e a bounded set into itself. Thus some kind of boundedness of the solutions is required. As for Theorem 3.3, some asymptotic boundedness of the solutions is required. Thus we list the following denitions. Denition 3.4. [6] The solutions of Eq.(1.1) are said to be locally strictly bounded if there exists a constant B > 0 such that ||g B implies that its solution satises u(t, ) B for t [0, T ]. Denition 3.5. [7] The solutions of Eq.(1.1) are said to be ultimate bounded if there is a bound B > 0, such that for each B3 > 0, there is a K > 0, such that ||g B3 and t K imply that its solution satises u(t, ) < B. We now study the relationship between the boundedness and the periodicity of solutions of Eq.(1.1). By using Theorem 3.2 and Denition 3.4, we have Theorem 3.6. Let (H1 ), (H2 ) and (H3 ) be satised. If the solutions of Eq.(1.1) are locally strictly bounded (or assume that solutions are non-increasing in norm on [0, T ]), then Eq.(1.1) has a T periodic solution.
0 Proof. Let H = { Cg : ||g B} with B from Denition 3.4. Then H is convex, closed 0 and bounded in Cg . Next, for u() = u(, ) with H, the locally strictly boundedness implies that u(t) B for t [0, T ]. Then we obtain from Lemma 2.5 that
|P()|g = |uT ()|g max max B,
sup
s[0,T ]
u(s) ,
1 ||g 2 (3.4)
1 B = B. 2
Thus the result is true by using Theorem 3.2. By using Theorem 3.3 and Denition 3.5, we have the following result, whose proof is omitted here since it is similar to one in [7]. Theorem 3.7. Let (H1 ), (H2 ) and (H3 ) be satised. If the solutions of Eq.(1.1) are ultimate bounded, then Eq.(1.1) has a T periodic solution.
26
A Massera Type Theorem
In this section, we study the existence of a periodic solution for the following nonhomogeneous partial functional differential equation d x(t) = Ax(t) + L(t, xt ) + h(t), t > 0, dt x0 = , (4.1)
where L is a continuous function from IR+ Cg into X, linear with respect to the second argument and T -periodic in t; h is a continuous T -periodic function with values in X. We will obtain a Massera type theorem for Eq. (4.1). More precisely, we have Theorem 4.1. Let (H1 ) and (H3 ) be satised. If Eq. (4.1) has a bounded solution y (with y(0) D(A)) on IR+ in the sense that sup |y(t)| < +, then it has a T -periodic solution.
tIR+
Proof. Let y be a bounded solution of Eq. (4.1) on IR+ . From the denition of Cg , we know that {yt : t 0} is also bounded in Cg . Let
D := co {ynT : n IN} ,
where co denotes the closure of the convex hull. Then D is a nonempty bounded closed 0 convex subset of Cg . Let D . Then there exists a sequence (k )k co {ynT : n IN} such that k = k ynk T , k 0, k = 1, and k i i i
i=1
i
nk
nk
i=1
0 as k .
Then, one has xT (, k ) = k xT (, ynk T ) i = k ynk T +T . i

i=1
i
nk
i=1 nk
Thus, xT (, k ) D . Since xT (, ) xT (, k ) g 0 as k +, and D is closed, we deduce that P() = xT (, ) D for all D , which implies that P(D ) D . By Theorem 3.2, we conclude that Eq. (4.1) has a T -periodic solution.
References
[1] M. Adimy, H. Bouzahir and K. Ezzinbi, Existence for a class of partial functional differential equations with innite delay, Nonlin. Anal., 46(2001), 91-112, [2] M. Adimy, H. Bouzahir and K. Ezzinbi, Local existence and stability for a class of partial functional differential equations with innite delay, Nonlin. Anal., 48(2002), 323-348.
27
[3] W. Arendt, C.J.K. Batty, M. Hieber and F. Neubrander, Vector Valued Laplace Transforms and Cauchy Problems, Monographs in Mathematics. Vol. 96, Birkhauser Verlag, 2001. [4] R. Benkhalti and H. Bouzahir and K. Ezzinbi, Existence of Periodic solutions for some partial functional differential equations with innite delay, J. Math. Anal. Appl., 256(2001), 257-280. [5] H. Henriquez, Periodic solutions of quasi-linear partial functional differential equations with unbounded delay, Funkcial. Ekvac., 37(1994), 329-343. [6] J. Liu, Periodic solutions of innite delay evolution equations, J. Math. Anal. Appl., 247(2000), 627-644. [7] J. Liu, T. Naito and N. Minh, Bounded and Periodic solutions of innite delay evolution equations, J. Math. Anal. Appl., 286(2003), 705-712. [8] B. Sadovskii, On a xed point principle, Funct. Anal. Appl., 2(1967), 151-153.
I NTERNAL P OLLUTION AND D ISCRIMINATING S ENTINEL IN P OPULATION DYNAMICS P ROBLEM

1 Universit e
O. Nakoulima 1and S. Sawadogo 2 des Antilles et de la Guyane, Facult e des Sciences Exactes et Naturelles, D epartement de Math ematiques/Informatiques, ` Campus de Fouillole, 97159 Pointe a Pitre Cedex, France. 2 Universit de Ouagadougou, UFR/Sciences Exactes et Appliqu ees, e D partement de Math ematiques/Informatiques, e 03 BP 7021 Ouagadougou 03-Burkina Faso
Abstract The so called Lions sentinel method is applied to some time varying system with two time scale variables and with missing data. As an example, some population dynamics problem with missing data is considered. The two time scale variables represent the running time variable and the age of individuals. This insight seems to be new. Building the object sentinel leads to some null-controllability problem with restricted control variables and two time scale variables.
AMS Subject Classication: 93B05, 92D25, 35B37, 35K05 Key Words: population dynamics, controllability, sentinels.
1 Introduction
1.1 Population Dynamics Problem with Missing Data
The unknown real function y depends on variables t, a, x where t (T0 , T ) stands for the running time, a (0, A) for the age of individuals and x RN for the space variable. The number y (t, a, x) is the distribution of a year old individuals at time t at the point x.
For relevant notation we agree on the following. If some function depends on several variables (x1 , x2 , ..xn ) then we denote by (xk ) or (xk , xl )the function of the other variables. Possible confusion is avoided in precising the denition domain. The function y has to satisfy the following two time scale varying equation y y t + a y + y = f + f in Q = (T0 , T ) (0, A) = U y = 0 in = (T0 , T ) (0, A) = U (1) y (T0 ) = y0 + y0 in QA = (0, A) y (0) = 0A (a) y (a) da in QT = (T0 , T )
E-mail E-mail
address: onakouli@univ-ag.fr address: sawasom@yahoo.fr
30
O. Nakoulima and S. Sawadogo
It is assumed that is open and bounded with C 2 boundary = and (t, a, x) 0, (t, a, x) 0. The parameters of the problem have the following sense : the bound T > 0 is the horizon of the problem (the lower bound of running time scale is T0 in order to avoid confusion with the lower bound of age time scale which is actually 0), the bound A is the expectation of life, the weight is the natural fertility rate, the function = (t, x, a) is the natural death rate of ayear old individuals at time t > 0 and in the position x, the function f corresponds to external ow and y0 = y0 (a, x) is the initial distribution of individuals. Convenient assumptions for mesurability and integrability of functions are made. In particular f L2 (Q). The data of the state equation (1) are incomplete in the following sense : (a) f L2 (Q) and y0 L2 (QA ) are unknown functions with f
L2 (Q)
1, y0
L2 (QA )
In the model (1), the term y0 represents missing data where as f stands for pollution term. The general question we want to address is for any given extra observation of the system (1), can we evaluate f , without any attempt at computing y0 ? (H0 )
(b) R and R are unknown numbers and small enough.
Lions[8] rst investigated parabolic patterns. Here we construct sentinels when the supports of the observation function and of the control function are included in two different open subsets of RN . This point of view has already been proposed by Nakoulima [10]. The sentinels theory relies on three features: the state equation, the observation system and some particular evaluation function : the sentinel itself. State equation: The equation (1) governs the state y. We assume the following [1] : (H1)
L+ ((0, A) (T0 , T ) ), (0, A) s.t. (a) = 0 for a (, A);
(H2) Lloc ([0, A); L ((T0 , T ) )), 0 a.e. in QA ;
t, A < t < T, x
For the biological comments about the model and for the basic existence of the solution to (1) we refer to [2, 7, 11]. For the sake of simplicity, we use indifferently y, y(t, a, x; , ) or y(, ) to denote the unique solution of (1). It is relevant since and are xed parameters. Observation system: the observation is the knowledge, along some time period, of some function yobs which is dened on some strip over some nonempty open subset O , called observatory, as follows yobs = m0 + i mi .
i=1 M
(H3)
t, T0 < t < T0 + A, x , lim
t (, a t + , x)d = +, aA 0 a , lim 0 (t a + , , x)d = +. aA
(2)
Sentinels for Population Dynamics
31
where the functions m0, m1 , ..., mM are given in L2 (U O) but where the real coefcients i are unknown. We assume the coefcients i are small. We refer to the terms i mi as the interference terms. We can assume without loss of generality that the functions mi are linearly independent. Sentinel. Let and let , open and nonempty and set G =U . For any control function w L2 (G ), set S(, ) = h0 y(t, a, x; , )dtdadx + wy(t, a, x; , )dtdadx.
U O U
(3)
h0 L2 (U O)
(4)
Choose now w L2 (G ) such that the following holds: S (0, 0) = 0 y0 ;
(i). S is stationary with respect to the missing terms y0 , i.e. (6)
(ii). S is stationary with respect to interference terms i mi ,i.e.

h0 mi dtdadx +
wmi dtdadx = 0, 1 i M;
and (iii). the norm w satisfy the above conditions

L2 (U)
is minimal among control functions in L2 (G ) which w

L2 (G )
= minimum.
Remark 1.1 According to (8), the function S if it exists, is unique. We refer to function S as the sentinel. Remark 1.2 If the functions mi are null functions, no information comes from (7). The sentinel S is dened only by (6) and (8). If mi = 0, the sentinel S is dened by (6), (7) and (8) and it is called a discriminating sentinel. Remark 1.3 The support sup p (mi ) of functions mi is assumed to be included in O. Sup/ pose O = 0 then U wmi dtdadx = 0. There are then no contraints on the control w. Therefore, it sufces to choose h0 such that h0 is orthogonal to each mi . So, without loss of generality, it may be assumed that O. (9)

1.2
Equivalence to Null-Controllability with Constraints on the Control
Here it will be shown that the existence of such a control function w dealing (6)-(8) is equivalent to null-controllability property for the system with constrained control. Consider the function y = y where y corresponds to parameter values = 0, = 0. The function y is the solution of the problem
(7)
(8)
32

y y t + a y + y = 0 y = 0 on ; y (T0 ) = y0 in QA ; y (0) = 0A (a)y (a)da
in Q; (10) in QT .
Under the assumptions (H1 ) (H3 ) the linear problem (10) gets one only solution y such that y (t, A, x) = 0. For the details of the proof we refer to [2, 7, 11]. We now consider the stationary condition (6). The stationary condition holds if and only if

h0 y dtdadx +
wy dtdadx = 0, y0 , y0
L2 (QA )
1.
(11)
where O and are indicator functions for the open sets O and respectively. There is one only solution in L2 (Q) as some consequence of the xed point theorem for contracting mapping [2, 3]. The so called adjoint state q depends on the unknown w and its utility comes from the following process. First, multiply both members of the differential equation in (12) by y , and integrate by parts over Q
A U O U 0

In order to transform the equation (11), introduce now the classical adjoint state. More precisely, consider the solution q = q(t, a, x) of the linear problem q q t a q + q = q(0) + h0 O + w in Q, q=0 in , (12) q(T ) = 0 in QA , q(A) = 0 in QT ,
Thus, the condition (6) (or (11) ) holds if and only if q(0, a, x) = 0 a.e (a, x) (0, A) . (14)
Then, consider the constraints (7). Let K L2 (U ) be the subspace generated by the family {mi }1iM K =span {mi |1 i M} . (15) There is one unique k0 K such that
U O U

Denoting by K if and only if
h0 y dtdadx +
wy dtdadx =
q(0, a, x)y0 dadx ,
(13)
y0 L2 (QA ) y0
L2 (QA )
1.
h0 mi dtdadx +
k0 mi dtdadx = 0, 1 i M.
(16)
the orthogonal subspace of K in L2 (U ) , the condition (7) holds w k0 = v K . (17)
33
The above considerations show that nding the control w such that the pair (w, S) satises (6)(8) comes to nd the control v such that the pair (v, q) satises the following system v K , q L2 (Q) (18) q q t a q + q = q(0) + h0 O + k0 + v in Q, q = 0 on , (19) q(T ) = 0 in QA , q(A) = 0 in QT , q(T0 ) = 0 in QA , v
L2 (U)
(20)
= minimum.
(21)
Problem (18)-(21) stands for some null-controllability problem with constrained control variable v.
1.3
Controllability Problem
For the problem (18)-(21), two matters are considered. The rst one consists in solving the null-controllability problem, and the second one consists in characterizing the optimal solution (21) by some optimality system. The problem (18)-(21) is solved when K ={0} (i.e. setting without constraints or free constraints) in several issues by various methods [1], [4]. In the present paper both points are considered in the general setting K = {0}.
Null-Controllability with Constraints on the Control
Consider the following general null-controllability problem. Find some pair (v, q) such that v K , q L2 (Q); = = = = q(0) + h + v in Q, 0 in , 0 in QA , 0 in QT , (22)
q q t a q + q q q(T ) q(A) v
(23)
q(T0 ) = 0 in QA , = minimum,
(24) (25)
L2 (U)
where h L2 (Q) (for h = h0 O + k0 we recognize (18)-(21)).
2.1
Observability Inequality Adapted to Constraints
The observability inequality we are looking for is some consequence of Carlemans inequality. In order to state Carlemans inequality, introduce now some objects and notations. Choose rst some auxiliary function C2 () which satises the following conditions : (x) > 0 x , (x) = 0 x , |(x)| = 0 x 0 ,
34
where 0 denotes any open set such that 0 (for example 0 can be some small enough open ball). Such a function exists according to A. Fursikov and O. Yu. Imanuvilov [6]. For any positive parameter value dene then the following weight functions (t, a, x) = e(x) e2 e(x) , (t, a, x) = , at(T t) at(T t) (26)
Now the inequality can be formulated. There exist numbers 0 > 1 and s0 > 1 and there exists some number C > 0 which only depends on the open sets and , such that for any > 0 , for any s s0 = s0 (, T, ) and for any V the following holds:
and adopt the following notations L = t + a + I, L = t a + I, (Q), = 0 on . V = C
(27)
e2s |t + a |2 + ||2 dtdadx + Q s
s2 e2s ||2 dtdadx
T0
The above inequality is the global Carlemans inequality for which we refer to [1] and [12]. We are now concerned with our own inequality. Since does not vanish, set es 1 = or = es . Then C2 (Q) and
is bounded. By substitution in (28) the following inequality holds 1

Q

1 As a consequence of the boundeness of and 3 s13 4 , some rst observability inequality comes 1 ||2 dtdadx C( |L|2 dtdadx + ||2 dtdadx). (31) 2 Q Q U
Another observability inequality follows which is adapted to the constraints setting using the subspace K (15). Recall rst that K is nite dimensional. (32) Denote by P = the orthogonal projection operator from L2 (G ) on K . The following lemma is the key for our results (33)
1 ||2 dtdadx C( 2 Q
2 3 s3 4
|L|2 dtdadx +
+ C
s3 4 3 e2s ||2 dtdadx

T A
e2s |L|2 dtdadx +
s3 4 3 e2s ||2 dtdadx . (28)
(29)
1 ||2 dtdadx). 2
(30)
Sentinels for Population Dynamics Lemma 2.1 Assume any function k K L2 (U, H 1 ()) such that k k t + a k + k = 0 in G is identically zero in G . Then there exists some positive constant C > 0 such that :

35
(34)
1 ||2 dtdadx C( 2 Q
|L|2 dtdadx +
| P|2 dtdadx) V .
(35)
Remark 2.2 The above assumption has been already set by Lions [8] for the parabolic system. Proof 2.1 Suppose that (35) does not hold. Then 1 2 j N , j V , j dtdadx = 1, 2 U 1 1 2 2 L j dtdadx and ( j P j ) dtdadx . j j U G 1 2 j dtdadx 0. 2 k Q

The forthcomming proof consists in extracting some subsequence ( jk )k such that the following contradiction holds
k
Denote by (h|g)L2 (G ) the natural scalar product in the Hibert space L2 (G ) . Let {k1 , k2 , ..., kM } be some orthonormal basis of K . 1) We show rst that for any i = 1, 2, ..., M the numerical sequence (( j |ki )L2 (G ) ) jN is bounded or equivalently that the sequence Start with the norm inequality 1 2 P j dtdadx 2
1 2
Since
1 2
is bounded and by (36) it follows that there is some number
j N,
1 2 P j dtdadx . 2
Since K is nite dimensional, norms are equivalent. Particularly the mappings

|k|2 dtdadx and k
are equivalent norms on K . There is then some number
j N,
P j dtdadx .
(36)
lim
P j
2 L2 (G )
is bounded
1 2 j dtdadx 2 +
1 2
1 2 ( j P j ) dtdadx 2
1 2
. (37)
(38)
1 2 |k| dtdadx 2
36
O. Nakoulima and S. Sawadogo The relation (( j P j ) K , j N means the following (( j P j ) |ki )L2 (G ) = 0 i, 1 i M, j N . Thus P j = (P j |ki )L2 (G ) ki = ( j |ki )L2 (G ) ki .
i=1 i=1 M M
(39)
(40)
and from orthonormality P j dtdadx = ( j |ki )L2 (G )

2 i=1 2 L2 (G ) M 2
= P j
2 L2 (G )
(41)
thus P j 2) Since (P j ) jN is bounded and
then the sequence ( j ) jN is bounded. There is some weakly convergent subsequence still denoted by ( j ) jN such that j g weakly in L2 (G ) . (45)
Since subsequences have the same limit as convergent sequence j P j 0 strongly in L2 (G ) . (46)
Since K nite dimensional, there is some possibly sub-subsequence still denoted by ( j ) jN such that P j strongly in L2 (G ) . (47) As a consequence of (46) and (47),(45) j g = strongly in L2 (G ) . Then (46) P j g = strongly in L2 (G ) . Since P is some compact operator and by (45) P j Pg strongly in L2 (G ) . Therefore Pg = g and so g K . 3) There is however some better knowledge about the function g. Actually g = 0. From (36) L j 0 strongly in L2 (Q) . Then L j 0 strongly in L2 (G ) . Thus L j 0 weakly in D (G ) and so Lg = 0. Following (34) g = 0 on G .
(43)
( j P j )
2 L2 (G )
( j P j ) dtdadx 0,
Sentinels for Population Dynamics 4) The two preceding steps deal the conclusion j 0 strongly in L2 (G ) . Since j V , the observability inequality (31) applies as follows:
Q

37
j N ,
1 2 j dtdadx C 2 Q lim
j
L j dtdadx +
j dtdadx .
G
(48)
It is clear then that
1 2 j dtdadx = 0. 2 Q
The proof is now completed.
2.2
Existence of Optimal Control Variable
Consider now the following symetric bilinear form

V , V , a(, ) =
LLdtdadx +
( P)( P)dtdadx.
(49)
According to Lemma 3, this symetric bilinear form is a scalar product on V . Let V be the completion of V with respect to the related norm :
V
a(, ).
(50)
The closure of V is the Hilbert space V. Remark that the norm . V is related with the right member of the inequality (35). Similarly, the left member of (35) leads to the norm V , || = 1 ||2 dtdadx 2 Q
1 2
The completion of V is the weighted Hilbert space usually denoted by L2 . The inequal1 ity (35) has the following meaning V , || C
V
This inequality extends to V . This shows that V is continuously imbedded in L2 . 1 We can state the following Lemma 2.3 Let us consider h such that h L2 (Q) and h L2 (Q). The linear form dened by
V is continuous.
(51)
(52)
hdtdadx
Q
38
Proof 2.2 From Cauchy-Schwarz inequality in L2 (Q)

V ,
hdtdadx (
2 |h|2 dtdadx) 2 (
L2
1 1 ||2 dtdadx) 2 2 Q
(53)
V ,
hdtdadx h
|| C h
L2
Since V = V and the function is linear V , the inequality extends to V :
V, that was to be proved.
hdtdadx C
(54)
By the Lax-Milgram theorem [5], for any function h as above, there exists one and only one solution V of the variational equation:
a( , ) = Set
hdtdadx V.
(55)
v = ( P ) then q = L .
(56) (57)
Remark 2.4 In the statement of the null-controllability problem, there are boundary and initial or end conditions. These conditions concern the values of the control or state functions at the points of the boudary for example. The solutions dealt by means of functionnal analysis are not functions but elements of function spaces which are equivalence classes. As a consequence boundary or initial or end values of the solutions have to be considered in function spaces. Such a question has been adressed by Lions-Magenes. We refer to [9] for deriving the following trace theorems in regular opens sets . Assume q L2 (U ) L2 U, L2 () and q H 1 (U, L2 ()). Then q|U H 1 (U, H 2 ()). The meaning of q| , the trace of q on , is clear. q Assume q L2 (U ) L2 [T0 , T ] [0, A] , L2 () and q + a L2 (U, H 2 ()). t Then q C ([0, A], L2 ([T0 , T ] , H 2 ())) C ([T0 , T ], L2 ([0, A] , H 2 ())). That means there exists some function q : [T0 , T ] [0, A] L2 ([T0 , T ] , H 2 ()) standing for q L2 (U ) which is separately continuous so that the following values in L2 ([T0 , T ] , H 2 ()) get sense (t, a) U, q (t, a) = q (t, a)
1
Sentinels for Population Dynamics and q(T ) L2 ([0, A] , H 2 ())
39
q(A) L2 ([T0 , T ] , H 2 ()) Proposition 2.1 Keep the notations in the statement of the null-controllability problem. Assume (34) and the function h as in(52). Then there exists some couple (v, q) is such that (22)-(24) hold. Proof 2.3 The couple (, q) = ( , q ) dened above from stands for the solution. Since V then v L2 (G ) and q L2 (Q). Since P K then v = ( P ) K . By direct substitution in the formulas (49), (55) and (57) it follows

q(0) L2 ([T0 , T ] , H 2 ())
V,
q Ldtdadx +
( P )( P)dtdadx =
hdtdadx.
Q
(58)
Because P K , it reduces to
Q Q

V, i.e.
q Ldtdadx =
hdtdadx
( P )dtdadx, v dtdadx.
(59)
In the duality frame D (Q),D (Q) (60) means that L q = h + v in D (Q). (61)
Besides h + v L2 (Q), then L q L2 (Q). Since q L2 (Q) and q H 1 (U, L2 ()) and by the above remark q|U 1 H 1 (U, H 2 ()). Similarly, since q L2 (Q)) and q + q L2 (U, H 2 ()) q (0) t a L2 ([0, A] , H 2 ()), q (T ) L2 ([0, A] , H 2 ()); q (0) L2 ([T0 , T ] , H 2 ()) and q (A) L2 ([T0 , T ] , H 2 ()). Taking into account (61), integrate by parts V q Ldtdadx +
T
q ,
H 2 (),H 2 ()
dtda
H 2 (),H 2 () ]dt H 2 (),H 2 () ]da
0 A
[ q (t, 0), (t, 0)
H 2 (),H 2 () H 2 (),H 2 ()
q (t, A), (t, A)
[ q (T0 , a), (T0 , a)
q (T, a), (T, a)
V,
q Ldtdadx =
hdtdadx +
(60)
(h + v )dtdadx. (62)
40
O. Nakoulima and S. Sawadogo By (60) since V V , it follows
V ,
U T
q ,
H 2 (),H 2 () H 2 (),H 2 () H 2 (),H 2 ()
dtda q (t, A), (t, A)

H 2 (),H 2 () ]dt H 2 (),H 2 () ]da
0 A
[ q (t, 0), (t, 0)
[ q (T0 , a), (T0 , a)
q (T, a), (T, a)
= 0. (63) Then, successively, we get q = 0 on , q (T0 ) = 0 and q (T ) = 0 in QA , q (0) = 0 and q (A) = 0 in QT . Since q (0) = 0 we have L q = q (0) + h + v . Hence the proof is completed. Remark 2.5 Adapted observability inequality (35) shows that the choice of the scalar product on V is not unique. Thus, there exists an innity of control functions v such that (22)-(24) hold. Consider the set of control variables v such that (22)-(24) hold. By Proposition 4 this set is nonempty and it is clearly convex and closed in L2 (G ) . Therefore, there exists a unique control variable v of minimal norm in L2 (G ) . We state then the following : Theorem 2.6 Assume (34). For every h satisfying (52), there exists some control variable v such that (22)-(24) hold. Moreover, we can get a unique control v such that (25) holds. Now, we are concerned with the optimality system for v .
Optimality System for the Optimal Control Variable
Let q be the solution of system (23) submitted to the optimal control v . A classical way to derive the optimality system for the couple (v , q ) is the penalization method by Lions [8]. Describe now the method. Let > 0 . Dene the functional 1 J (v, q) = v 2 2 (G ) L 2 2 (64) q q 1 q + q q(t, 0, x) h v + 2 t a L2 (Q) for any couple (v, q) such that 2 v K , q L (Q), q q t a q + q q(0) L2 (Q), q = 0 on ; q(T ) = 0 in QA ; q(A) = 0 in QT ; q(T0 ) = 0 in QA . Consider the minimization problem inf{J (v, q) | (v, q) subject to (65)}.
(65)
(66)
41
Proposition 3.1 Under the assumptions of Theorem 8, the minimization problem has an optimal solution. There exists (v , q ) such that J (v , q ) = min{J (v, q) | (v, q) subject to (65)}. (67)
Proof 3.1 Let (vn , qn ) be a minimizing sequence satisfying (65). The sequence (J (vn , qn ))n is bounded from above J (vn , qn ) () (68) then vn
L2 (G ) C() qn t qn qn + qn qn (0) h vn 2 a L (Q)
C().
(69)
There is some subsequence of (vn )n , still denoted by (vn )n , such that vn v weakly in L2 (G ). (70)
As a consequence (65) the (sub)sequence (qn )n is bounded qn

L2 (Q)
C.
(71)
There is some subsequence of (qn )n , still denoted by (qn )n such that qn Then lim inf J (vn , qn ) J (v , q ). We deduce that (v , q ) is a unique optimal control, from the strict convexity of J . Now, we study the convergence of (v , q ) . Proposition 3.2 Let ((v , q )) be the family of solutions of (67). Then lim0 v = v weakly in L2 (U ) lim0 q q weakly in L2 (Q). (74) (73) q weakly in L2 (Q). (72)
Proof 3.2 (v , q ) satises (22)-(24), then from the structure (64) of J (v, q), we have of course v L2 (G ) C (75) q q t a q + q q (t, 0, x) h v 2 C .
L (Q)
Where C s are various constants independent of . q veries (65) (75) then q L2 (Q) C.
(76)
42 L2 (Q)
O. Nakoulima and S. Sawadogo There are a subseqence of (v , q ) , again denoted by (v , q ) , v0 L2 (G ) and q0 such that v v0 weakly in L2 (G ), v0 K ; q q0 weakly in L2 (Q). (77)
Then q q0 weakly in D (Q) and by the weak continuity of the operator L in D (Q) it follows L q L q0 weakly in D (Q). Moreover the traces functions are continuous, then the pair (v0 , q0 ) satises the system q q t0 a0 q0 + q0 = q0 (0) + h + v0 in Q, q0 = 0 on , (78) q0 (T ) = 0 in QA , q0 (A ) = 0 in QT . q0 (T0 ) = 0 in QA . v we get v0
2 L2 (G ) 2 L2 (G )
(79)
From the following estimate J (v , q ) (80) (81)
lim inf J (v , q ).
Since the pair (v , q ) satises (22)-(24) lim inf J (v , q ) v Thus v0 and then v0
L2 (G ) L2 (G ) 2 L2 (G ) .
v = v
L2 (G )
L2 (G ) .
Hence v0 = v . Since (78) has a unique solution, it follows q0 = q . Let us express the optimality conditions satised by (v , q ), as optimal solution of (67). Proposition 3.3 The assumptions are as in Theorem 8. The couple (v , q ) is optimal solution of problem (67) if and only if there is one function such that the triplet {v , q , } satises the following so called optimality system q q t a q + q = q (0) + h + v + in Q, q = 0 in , (82) q (T ) = 0 in QA , q (A) = 0 in QT , q (T0 ) = 0 in QA , (82)
t + a + = 0 in Q, = 0 in , (0) = 0A (a) (a)da in QT ,
(83) (84)
v = ( P ) .
43
Proof 3.3 Express the Euler-Lagrange optimality conditions which characterize (v , q ). For any (v, ) such that (65) the following holds v vdtdadx+

(
Q
q q q + q q (0) h v ) t a + (0) v )dtdadx = 0. (85) ( t a
Dene the adjoint state = 1 q q q + q q (0) h v . t a (86)
Then (82) holds . For any (v, ) such that (65), (85) becomes
v vdtdadx +
+ (0) v )dtdadx = 0. t a
(87)
and
Integrate by parts in (87). As a consequence the couple (v , ) is shown to satisfy t + a + = 0 in Q (88) = 0 on A (0) = 0 (a) (a)da
G
(v + ) vdtdadx = 0, v K .
Hence v + K . Since v K then v + = P (v + ) = P and thus v = ( P ) . (90) Hence the assertion follows. Remark 3.1 There is no available information concerning (A) in QT , (0) in QA , (T ) in QA . A priori estimates for approximate adjoint state ( ) are now looked for. This is the essential point. From (90) and (75) ( P ) L2 (G ) C. (91) Since L = 0, then
V
Therefore there are a subseqence ( ) still denoted by ( ) and V such that weakly in V. (93)
(89)
C.
(92)
44
O. Nakoulima and S. Sawadogo Following the lines of the proof of Lemma 3, we conclude with the weak convergence in L2 (G ). Thus P 0 strongly in L2 (G ) (93)
so that 0 K . By (91) and (93) there is some 1 K
P so that = 0 + 1 . Thus 0 = P and P
1 weakly in V
P weakly in L2 (G ).
(94)
The above argument leads to the following statement : Theorem 3.2 The assumptions are as in Theorem 8. The couple (v , q ) is the optimal solution of problem (22)-(25) if and only if there is a function such that the triplet (v , q , ) is solution of the following optimality system q q t a q + q = q (0) + h + v in Q, q = 0 on , q (T ) = 0 in QA , q (A) = 0 in QT , q (T0 ) = 0 in QA , + = 0 in Q, on , (0) = A (a) (a)da in Q , T 0 v = ( P ) .
t + a = 0
v K , q L2 (Q), V
(95)
(96)
(97) (98) (99)
Discriminating Sentinels
It is clear, when h = h0 O + k0 that the problem (18)-(21) and the problem (22)-(25) are the same. The results concerning the former apply to the latter. Consider then the results obtained in the previous sections. More precisely, assume that K is dened as in (15) and h0 , k0 are such that (52) holds. Assume also (9). Let (v , q , ) be dened as in Theorem 13. Then v = ( P ) . (100) Denition - The function S dened by

S(, ) =
h0 y(, )dtdadx +
is called discriminating sentinel dened by the open subsets O and , and by the function h0 .
(k0 ( P ))y(, )dtdadx
(101)
45
Return to the problem (H0 ). The above function S allows to lay down the relevant question. Because of (6) we can write S(, ) = S(0, 0) + S (0, 0) for and small, (102)
where S is dened by (101). On the other hand, by (2) and (7)

S(, ) = So that
h0 m0 dtdadx +
v m0 dtdadx.
(103)
We also have

S (0, 0) = where y is given by
h0 y dtdadx +
y y t + a y + y y = 0 in ,
Let q (h0 ) be the solution (uniquely dened by h0 ) of (96). Multiplying (96) by y it follows h0 y dtdadx + v y dtdadx = q (h0 ) f dtdadx. (107)
U O U
y (T0 ) = 0 in QA , y (0) = 0A (a)y (a)da in QT .

Therefore (104) becomes
h0 (m0 y0 )dtdadx +
v (m0 y0 )dtdadx =
This equation allows to evaluate the right hand side of (108) and therefore it is an available information concerning f . A perturbation (or a pollution) f will be stealthy for the sentinel dened by h0 if
Q
q (h0 ) f dtdadx = 0.
Conclusion
As a conclusion we announce two lines in prolongation of our work. The rst one is concerned with a generalization of the idea of sentinel which tackles new observation or control designs. Indeed it has been shown above that sentinel by J. L. Lions can hoock on some two-time scale varying system with both interior observation and control. But several other designs are realic. For example, only one part of the boundary
h0 (m0 y0 )dtdadx +
v (m0 y0 )dtdadx =
S (0, 0).
(104)
v y dtdadx
(105)
= f in Q, (106)
q (h0 ) f dtdadx.
(108)
(109)
46
may be observed, nd the control may be constrained and/or restricted to the boundary. Different mixing are possible which could be solved by means of some setting so called Carmeman inequality. This the aim of some forthcomming paper. The second line in process is concerned with furtivity as it is dened at the end of above section.
Acknowledgments
Authors wish to express theirs gratitudes to Robert Janin for his many helpful suggestions.
References
[1] B.Ainseba, 2002, Exact and approximate controllability of age and space population dynamics structured model, J. Math.Anal.Appl. [2] B.Ainseba and M.Langlais, 2000, On a population dynamics control problem with age dependence and spatial structure, J. Math. Anal. Appl. 248,455474. [3] B. Ainseba et M.Langlais, 1996, Sur un probl` me de contr le dune population struce o tur e en age et en espace.C.R.Acad.Sci.Paris,t.323,serie I, 269274. e [4] B.Ainseba and S.Anita, 2001, Local exact controllability of the age-dependent population dynamics with diffusion.Abstract Appl.Anal.,6, 357-368. [5] H.Brezis: Analyse fonctionnelle.Th orie et application. Masson, Paris 1983. e [6] A.Fursikov, O.Imanuvilov,1996, Controllability of evolution equation.Lecture Notes Series 34,RIM-GARC, Seoul National University. [7] M.Giovanna and M.Langlais, 1982, Age-dependent population diffusion with external constraint., J.Math.Biology, 14, 7-94. [8] J.L.Lions: Sentinelles pour les syst` mes distribu s a donn es incompl` tes.Masson, e e ` e e Paris 1992. [9] J.L. Lions et M. Magenes . (1968) Probl` mes aux limites non homog` nes et applicae e tions. Paris, Dunod,Vol. 1 et 2. [10] O. Nakoulima, 2004, Contr labilit a z ro avec contraintes sur le contr le. C. R. Acad. o e` e o Sci. Paris, Ser.I 339/6 405-410. [11] A. Ou draogo, O. Traor , 2003, Sur un probl` me de dynamique des populations, e e e Imhotep. Vol 4 n 1. [12] S.Sawadogo, 2005, Th` se unique.Universit de Ouagadougou. e e
C ENTER M ANIFOLD AND S TABILITY IN C RITICAL C ASES FOR S OME PARTIAL F UNCTIONAL D IFFERENTIAL E QUATIONS
Mostafa Adimy1 Khalil Ezzinbi2 and Jianhong Wu 3 , 1 Universit de Pau et des Pays de lAdour, Laboratoire e de Math matiques Appliqu es CNRS UMR 5142 e e Avenue de luniversit 64000, Pau, France e 2 Universit Cadi Ayyad, Facult des Sciences Semlalia, e e D partement de Math matiques, B.P. 2390, Marrakesh, Morocco e e 3 York University, Department of Mathematics and Statistics, Faculty of Pure and Applied Sciences, 4700 Keele Street North York, Ontario, Canada, M3J 1P3
Abstract In this work, we prove the existence of a center manifold for some partial functional differential equations, whose linear part is not necessarily densely dened but satises the Hille-Yosida condition. The attractiveness of the center manifold is also shown when the unstable space is reduced to zero. We prove that the ow on the center manifold is completely determined by an ordinary differential equation in a nite dimensional space. In some critical cases, when the exponential stability is not possible, we prove that the uniform asymptotic stability of the equilibrium is completely determined by the uniform asymptotic stability of the reduced system on the center manifold.
Key Words: Hille-Yosida operator, integral solution, semigroup, variation of constants formula, center manifold, attractiveness, reduced system, critical case, asymptotic stability, approximation. 2000 Mathematical Subject Classication: 34K17, 34K19, 34K20, 34K30, 34G20, 47D06.
This research is supported by Grant from CNCPRST (Morocco) and CNRS(France) Ref. SPM 17769, by TWAS Grant under contract Ref. 03-030 RG/MATHS/AF/AC, by the Canada Research Chairs Program, by Natural Sciences and Engineering Research Council of Canada, and by Mathematics for Information Technology and Complex Systems. E-mail address: mostafa.adimy@univ-pau.fr E-mail address: ezzinbi@ucam.ac.ma: to whom all correspondence should be sent E-mail address: wujh@mathstat.yorku.ca
48
Mostafa Adimy, Khalil Ezzinbi and Jianhong Wu
Introduction
The aim of this paper is to study the existence of a center manifold and stability in some critical cases for the following partial functional differential equation d u(t) = Au(t) + L(ut ) + g(ut ), t 0 dt u0 = C := C ([r, 0] ; E) , (1.1)
where A is not necessarily densely dened linear operator on a Banach space E and C is the space of continuous functions from [r, 0] to E endowed with the uniform norm topology. For every t 0 and for a continuous u : [r, +) E, the function ut C is dened by ut () = u(t + ) for [r, 0] . L is a bounded linear operator from C into E and g is a Lipschitz continuous function from C to E with g(0) = 0. In this work, we assume that A is a Hille-Yosida operator: there exist lR and M0 1 such that (, ) (A) and (I A)n M0 for and n N, ( )n
where (A) is the resolvent set of A. In [21], the authors proved the existence, regularity and stability of solutions of (1.1) when A generates a strongly continuous semigroup, which is equivalent by Hille-Yosida Theorem to that A is a Hille-Yosida operator and D(A) = E. In [3], the authors used the integrated semigroup approach to prove the existence and regularity of solutions of (1.1) when A is only a Hille-Yosida operator. Moreover, it was shown that the phase space of equation (1.1) is given by Y := C : (0) D(A) . Assume that the function g is differentiable at 0 with g (0) = 0. Then the linearized equation of (1.1) around the equilibrium zero is given by d v(t) = Av(t) + L(vt ), t 0 dt v0 = C. (1.2)
If all characteristic values (see section 2) of equation (1.2) have negative real part, then the zero equilibrium of (1.1) is uniformly asymptotically stable. However, if there exists at least one characteristic value with a positive real part, then the zero solution of (1.1) is unstable. In the critical case, when exponential stability is not possible and there exists a characteristic value with zero real part, the situation is more complicated since either stability or instability may hold. The subject of the center manifold is to study the stability in this critical case. For differential equations, the center manifold theory has been extensively studied, we refer to [6], [7], [8], [12], [13], [14], [15], [16], [19], [20] and [24]. In [17] and [22], the authors proved the existence of a center manifold when D(A) = E.
Center Manifold and Stability in Critical Cases for Some Partial Functional ...
49
They established the attractiveness of this manifold when the unstable space is reduced to zero. In [11], the authors proved the existence of a center manifold for a given map. Their approach was applied to show the existence of a center manifold for partial functional differential equations in Banach spaces in the case when the linear part generates a compact strongly continuous semigroup. Recently, in [18], the authors studied the existence of invariant manifolds for an evolutionary process in Banach spaces and in particularly for some partial functional differential equations. For more details about the center manifold theory and its applications in the context of partial functional differential equations, we refer to the monograph [27]. Here we consider equation (1.1) when the domain D(A) is not necessarily dense in E. The nondensity occurs, in many situations, from restrictions made on the space where the equations are considered (for example, periodic continuous solutions, H"older continuous functions) or from boundary conditions ( the space C1 with null value on the boundary is not dense in the space of continuous functions). For more details, we refer to [1], [2], [3], [4] and [5]. The organization of this work is as follows: in section 2, we recall some results of integral solutions and the semigroup solution and we describe the variation of constants formula for the associated non-homogeneous problem of (1.2). We also give some results on the spectral analysis of the linear equation (1.2). In section 3, we prove the existence of a global center manifold. In section 4, we prove that this center manifold is exponentially attractive when the unstable space is reduced to zero. In section 5, we prove that the ow on the center manifold is governed by an ordinary differential equation in a nite dimensional space. In section 6, we prove a result on the stability of the equilibrium in the critical case. We also establish a new reduction principal for equation (1.1). In section 7, we study the existence of a local center manifold when g is only dened and C1 -function in a neighborhood of zero. In the last section, we propose a result on the stability when zero is a simple characteristic value and no characteristic value lies on the imaginary axis.
Spectral Analysis and Variation of Constants Formula
In the following we assume (H1 ) A is a Hille-Yosida operator. Denition 2.1. A continuous function u : [r, +) E is called an integral solution of equation (1.1) if i)
t t 0
ii) u(t) = (0) + A iii) u0 = .
u(s)ds + L
0
us ds +
We will call, without causing any confusion, the integral solution the function ut , for t 0. Let A0 be the part of the operator A in D(A) which is dened by D(A0 ) = x D(A) : Ax D(A) A0 x = Ax for x D(A0 ).
u(s)ds D(A) for t 0,

0 t
g(us )ds for t 0,
50
Mostafa Adimy, Khalil Ezzinbi and Jianhong Wu The following result is well known (see [3]).
Lemma 2.2. A0 generates a strongly continuous semigroup (T0 (t))t0 on D(A). For the existence and uniqueness of an integral solution of (1.1), we need the following condition. (H2 ) g : C E is Lipschitz continuous. The following result can be found in [3]. Proposition 2.3. Assume that (H1 ) and (H2 ) hold. Then for Y, equation (1.1) has a unique global integral solution on [r, ) which is given by the following formula t T0 (t)(0) + lim T0 (t s)B (L(us ) + g(us )) ds, t 0 u(t) = (2.1) (t), t [r, 0] , 0
where B = (I A)1 for .
Assume that g is differentiable at zero with g (0) = 0. Then the linearized equation of (1.1) at zero is given by equation (1.2). Dene the operator U(t) on Y by U(t) = vt (., ), where v is the unique integral solution of equation (1.2) corresponding to the initial value . Then (U(t))t0 is a strongly continuous semigroup on Y . One has the following linearized principle. Theorem 2.4. Assume that (H1 ) and (H2 ) hold. If the zero equilibrium of (U(t))t0 is exponentially stable, in the sense that there exist N0 1 and 0 such that |U(t)| N0 et for t 0, then the zero equilibrium of equation (1.1) is locally exponentially stable, in the sense that there exist 0, 0 and k 1 such that |xt (., )| ket || for Y with || and t 0, where xt (., ) is the integral solution of equation (1.1) corresponding to initial value . Moreover, if Y can be decomposed as Y = Y1 Y2 where Yi are U-invariant subspaces of Y , 1 Y1 is a nite-dimensional space and with 0 = lim log |U(h)|Y2 | we have h h inf {|| : (U(t)|Y1 )} e0t for t 0, where (U(t)|Y1 ) is the spectrum of U(t)|Y1 , then the zero equilibrium of equation (1.1) is unstable, in the sense that there exist 0 and sequences (n )n converging to 0 and (tn )n of positive real numbers such that |xtn (., n )| .
51
The above theorem is a consequence of the following result. For more details on the proof, we refer to [3]. Theorem 2.5. [9] Let (V (t))t0 be a nonlinear strongly continuous semigroup on a subset of a Banach space Z and assume that x0 is an equilibrium of (V (t))t0 such that V (t) is differentiable at x0 , with W (t) the derivative at x0 of V (t) for each t 0. Then, (W (t))t0 is a strongly continuous semigroup of bounded linear operators on Z. If the zero equilibrium of (W (t))t0 is exponentially stable, then x0 is locally exponentially stable equilibrium of (V (t))t0 . Moreover, if Z can be decomposed as Z = Z1 Z2 where Zi are W -invariant 1 subspaces of Z, Z1 is a nite-dimensional space and with 1 = lim log |W (h)|Z2 | we have h h inf {|| : (W (t)|Z1 )} e1t for t 0, then the equilibrium x0 is unstable in the sense that there exist 0 and sequences (yn )n converging to x0 and (tn )n of positive real numbers such that |V (tn )yn x0 | . Some informations of the innitesimal generator of (U(t))t0 can be found in [5]. For example, we know that Theorem 2.6. The innitesimal generator AU of (U(t))t0 on Y is given by D(AU ) = AU = C1 ([r, 0] ; E) : (0) D(A), (0) D(A) and (0) = A(0) + L() for D(AU ).
We now make the next assumption about the operator A. (H3 ) The semigroup (T0 (t))t0 is compact on D(A) for t 0. Theorem 2.7. Assume that (H3 ) holds. Then, U(t) is a compact operator on Y for t r. Proof. Let t r and D be a bounded subset of Y . We use Ascoli-Arzelas theorem to show that U(t) : D is relatively compact in Y . Let D, [r, 0] and 0 such that t + 0. Then (U(t)) () = T0 (t + )(0) + lim Note that
+ 0
t+
t+ 0
T0 (t + s)B L(U(s))ds +
T0 (t + s)B L(U(s))ds
t+ t+
t+
T0 (t + s)B L(U(s))ds.
T0 (t + s)B L(U(s))ds
52 and
t+ + 0
lim
T0 (t +s)B L(U(s))ds = T0 () lim
+ 0
The assumption (H3 ) implies that T0 ()

+ 0
t+
lim
T0 (t + s)B L(U(s))ds : D
is relatively compact in E. As the semigroup (U(t))t0 is exponentially bounded, then there exists a positive constant b1 such that lim
t+
+ t+
T0 (t + s)B L(U(s))ds b1 for D.
Consequently, the set lim
t+
+ t+
T0 (t + s)B L(U(s))ds : D
is totally bounded in E. We deduce that (U(t)) () : D is relatively compact in E, for each [r, 0]. For the completeness of the proof, we need to show the equicontinuity property. Let , 0 [r, 0] such that 0 . Then (U(t)) () (U(t)) (0 ) = (T0 (t + ) T0 (t + 0 )) (0) + lim lim Furthermore,
t+ 0 + 0
t+
T0 (t + s)B L(U(s))ds T0 (t + 0 s)B L(U(s))ds.
t+0
+ 0
T0 (t + s)B L(U(s))ds =
t+0 0
T0 (t + s)B L(U(s))ds T0 (t + s)B L(U(s))ds.
t+ t+0
+ Consequently,
|(U(t)) () (U(t)) (0 )| |T0 (t + ) T0 (t + 0 )| |(0)| + lim + lim

+ 0
t+0
(T0 (t + s) T0 (t + 0 s)) B L(U(s))ds
t+
+ t+0
T0 (t + s)B L(U(s))ds .
t+
T0 (t +s)B L(U(s))ds.
53
Assumption (H3 ) implies that the semigroup (T0 (t))t0 is uniformly continuous for t 0. Then lim |T0 (t + ) T0 (t + 0 )| = 0.
0
The semigroup (U(t))t0 is exponentially bounded. Consequently, there exists a positive constant b2 such that
+ t+0
lim
= (T0 ( 0 ) I) lim
+ 0
We have proved that there exists a compact set K0 in E such that

+ 0
t+0
lim
T0 (t + 0 s)B L(U(s))ds K0 for D.
Using Banach-Steinhauss theorem, we obtain

0
lim (T0 ( 0 ) I)x = 0 uniformly in x K0 .
This implies that

+ 0
lim (U(t)) () (U(t)) (0 ) = 0 uniformly in D.
We can prove in similar way that

0
lim (U(t)) () (U(t)) (0 ) = 0 uniformly in D.
By Ascoli-Arzelas theorem, we conclude that U(t) : D is compact for t r. Now, we consider the spectral properties of the innitesimal generator AU . We denote by E, without causing confusion, the complexication of E. For each complex number , we dene the linear operator () : D(A) E by () = I A L(e I), where e I : E C is dened by e x () = e x, x E and [r, 0] . Denition 2.8. We say that is a characteristic value of equation (1.2) if there exists x D(A)\{0} solving the characteristic equation ()x = 0. (2.2)
+ 0
lim
and
t+0
(T0 (t + s) T0 (t + 0 s)) B L(U(s))ds

t+0
t+
T0 (t + s)B L(U(s))ds b2 ( 0 )
T0 (t + 0 s)B L(U(s))ds.
54
Since the operator U(t) is compact for t r, the spectrum (AU ) of AU is the point spectrum p (AU ). More precisely, we have Theorem 2.9. The spectrum (AU ) = p (AU ) = { C : ker () = {0}} . Proof. Let p (AU ). Then there exists D(AU )\{0} such that AU = , which is equivalent to () = e (0), for [r, 0] and (0) = A(0) L() with (0) = 0. Consequently ()(0) = 0. Conversely, let C such that ker () = {0} . Then there exists x D(A)\{0} such that ()x = 0. If we dene the function by () = e x for [r, 0] , then D(AU ) and AU = , which implies that p (AU ). The growth bound 0 (U) of the semigroup (U(t))t0 is dened by 0 (U) = inf 0 : sup et |U(t)| < .
t0
The spectral bound s(AU ) of AU is dened by s(AU ) = sup {Re() : p (AU )} . Since U(t) is compact for t r, then it is well known that 0 (U) = s(AU ). Consequently, the asymptotic behavior of the solutions of the linear equation (1.2) is completely obtained by s(AU ). More precisely, we have the following result. Corollary 2.10. Assume that (H1 ), (H2 ) and (H3 ) hold. Then, the following properties hold, i) if s(AU ) < 0, then (U(t))t0 is exponentially stable and zero is locally exponentially stable for equation (1.1); ii) if s(AU ) = 0, then there exists Y such that |U(t)| = || for t 0 and either stability or instability may hold; iii) if s(AU ) 0, then there exists Y such that |U(t)| as t and zero is unstable for equation (1.1). As a consequence of the compactness of the semigroup U(t) for t r and by Theorem 2.11, p.100, in [10], we get the following general spectral decomposition of the phase space Y. Theorem 2.11. There exist linear subspaces of Y denoted by Y , Y0 and Y+ respectively with Y = Y Y0 Y+ such that i) AU (Y ) Y , AU (Y0 ) Y0 , and AU (Y+ ) Y+ ; ii) Y0 and Y+ are nite dimensional;
55
iii) (AU |Y0 ) = { (AU ) : Re = 0} , (AU |Y+ ) = { (AU ) : Re 0}; iv) U(t)Y Y for t 0, U(t) can be extended for t 0 when restricted to Y0 Y+ and U(t)Y0 Y0 ,U(t)Y+ Y+ for t lR; v) for any 0 < < inf {|Re | : (AU ) and Re = 0} , there exists K 0 such that |U(t)P | Ket |P | for t 0, |U(t)P0 | Ke 3 |t| |P0 | for t lR, |U(t)P+ | Ket |P+ | for t 0, where P , P0 and P+ are projections of Y into Y ,Y0 and Y+ respectively. Y ,Y0 and Y+ are called stable, center and unstable subspaces of the semigroup (U(t))t0 . The following result deals with the variation of constants formula for equation (1.1) which are taken from [5]. Let X0 be dened by X0 = {X0 c : c E} , where the function X0 c is dened by (X0 c) () = 0 if [r, 0) , c if = 0.
We introduce the space Y X0 , endowed with the following norm | + X0 c| = || + |c| . The following result is taken from [5]. Theorem 2.12. The continuous extension AU of the operator AU dened on Y X0 by: D(AU ) = C1 ([r, 0] ; E) : (0) D(A) and (0) D(A) AU = + X0 (A(0) + L() (0)), is a Hille-Yosida operator on Y X0 : there exist lR and M0 1 such that (, ) (AU ) and M0 (I AU )n for and n N, ( )n with (AU ) the resolvent set of AU . Moreover, the integral solution u of equation (1.1) is given for Y, by the following variation of constants formula ut = U(t) + lim
0
where B = (I AU )1 for . Remarks. i) Without loss of generality, we assume that M0 = 1. Otherwise, we can renorm the space Y X0 in order to get an equivalent norm for which M0 = 1. ii) For any locally integrable function : lR E, one can see that the following limit exists:
s
lim
U(t )B X0 ()d for t s.
U(t s)B (X0 g(us )) ds for t 0,
(2.3)
56
Global Existence of the Center Manifold
Theorem 3.1. Assume that (H1 ) and (H3 ) hold. Then, there exists 0 such that if Lip(g) = sup |g(1 ) g(2 )| < , |1 2 | 1 =2
then, there exists a bounded Lipschitz map hg : Y0 Y Y+ such that hg (0) = 0 and the Lipschitz manifold Mg := { + hg () : Y0 } is globally invariant under the ow of equation (1.1) on Y . Proof. Let B = B (Y0 ,Y Y+ ) denote the Banach space of bounded maps h : Y0 Y Y+ endowed with the uniform norm topology. We dene
F = {h B : h is Lipschitz, h(0) = 0 and Lip(h) 1} .
Let h F and Y0 . Using the strict contraction principle, one can prove the existence of vt solution of the following equation vt = U(t) + lim
0
We now introduce the mapping Tg : F B by Tg (h) = lim
+ lim
The rst step is to prove that Tg maps F into itself. Let 1 , 2 Y0 and t lR. Suppose that Lip(g) < . Then vt 1 vt 2 Ke 3 |t| |1 2 | + 2K |P0 | By Gronwalls lemma, we get that e 3 |t| vt 1 vt 2 K |1 2 | e2K|P0 ||t| and
vt 1 vt 2 K |1 2 | e[ 3 +2K|P0 |]|t| for t lR.
If we choose such that then
2K |P0 | < , 6
vt 1 vt 2 K |1 2 | e 2 |t| for t lR.
U(t ) B X0 g(v + h(v ))
d, t lR.
(3.1)
U() B X0 g(v + h(v )) U() B X0 g(v + h(v ))

d
+
d.
t 0
e 3 |t| v 1 v 2 d .
(3.2)
(3.3)
Center Manifold and Stability in Critical Cases for Some Partial Functional ... Moreover, |Tg (h)1 Tg (h)2 |
57
|U()P | g(v 1 + h(v 1 )) g(v 2 + h(v 2 )) d

+
+ Consequently,
0
|Tg (h)1 Tg (h)2 | 2
Ke |P | v 1 v 2 d + 2
Using inequality (3.3), we obtain |Tg (h)1 Tg (h)2 | 2K 2 |P | |1 2 | It follows that |Tg (h)1 Tg (h)2 | If we choose such that 4 2 K (|P | + |P+ |) |1 2 | .
0
e 2 d + 2K 2 |P+ | |1 2 |
4 2 K (|P | + |P+ |) < 1,
then Tg maps F into itself. The next step is to show that Tg is a strict contraction on F . Let h1 , h2 F . For Y0 and for i = 1, 2, let vti denote the solution of the following equation vti = U(t) + lim Then, vt1 vt2 K |P0 | and vt1 vt2 2K |P0 |
t 0 0
U(t ) B X0 g(vi + hi (vi ))
d for t lR.
t 0
e 3 |t| v1 v2 + h1 (v1 ) h1 (v2 ) + h1 (v2 ) h2 (v2 ) d ,
e 3 |t| v1 v2 d + K |P0 | |h1 h2 |
t 0
e 3 |t| d .
By Gronwalls lemma, we obtain that vt1 vt2 By (3.2), we obtain vt1 vt2
3K |P0 | |h1 h2 | e 2 |t| for all t lR. 3K |P0 | |h1 h2 | e[ 3 +2K|P0 |]|t| for t lR.
|U()P+ | g(v 1 + h(v 1 )) g(v 2 + h(v 2 )) d.
+ 0
Ke |P+ | v 1 v 2 d.
+ 0
e 2 d.
58
Mostafa Adimy, Khalil Ezzinbi and Jianhong Wu For i = 1, 2, we have Tg (hi ) = lim

+ lim It follows that
|Tg (h1 ) Tg (h2 )| 2K |P |
K |P | 62 K |P0 | |h1 h2 | + |h1 h2 | 2 K |P+ | 62 K |h1 h2 | . + 2K |P+ | 2 |P0 | |h1 h2 | + K 12K |P0 | + 1 |h1 h2 | .
Consequently, |Tg (h1 ) Tg (h2 )| (|P | + |P+ |) We choose such that (|P | + |P+ |) 12K K |P0 | + 1 < 1.
Then Tg is a strict contraction on F , and consequently it has a unique xed point hg in

F : Tg (hg ) = hg .
Finally, we show that Mg := { + hg () : Y0 } is globally invariant under the ow on Y . Let Y0 and v be the solution of equation (3.1). We claim that t vt + hg (vt ) is an integral solution of equation (1.1) with initial value + hg (). In fact, we have Tg (hg )(vt ) = hg (vt ), t lR. Moreover, for t lR, one has Tg (h)(vt ) = lim

+ lim which implies that hg (vt ) = lim
+ lim Then, for t lR, we have
vt + hg (vt ) = U(t) + lim + lim

t +
0 t
U(t ) B X0 g(v + hg (v )) U(t ) B X0 g(v + hg (v ))

+ lim
0 t
U() B X0 g(vi + hi (vi ))

0
d
+
U() B X0 g(vi + hi (vi ))
d.
U() B X0 g(vt+ + hg (vt+ ))

d
+
U() B X0 g(vt+ + hg (vt+ ))
d,
U(t ) B X0 g(v + hg (v ))

d
+
U(t ) B X0 g(v + hg (v ))
d.
U(t ) B X0 g(v + hg (v ))

d
+
d.
Center Manifold and Stability in Critical Cases for Some Partial Functional ... For any t a, we have

59
lim
= lim
t a
+ lim
and lim
= U(t a)
lim
By the same argument as above, we obtain lim
= lim
+ t a
+ lim and
+
lim
U(t ) B X0 g(v + hg (v )) = U(t a)

+
lim
Note that hg (v ) = lim a

+
U(a ) B X0 g(v + hg (v ))
a
+ lim and in particular
U(a ) B X0 g(v + hg (v ))
vt = U(t a)v + lim a
Consequently, for any t a, we obtain vt + hg (vt ) = U(t a) (v + hg (v )) + lim a a + lim

a a a
U(t ) B X0 g(v + hg (v ))
t
+ lim
U(t ) B X0 g(v + hg (v ))
U(t ) B X0 g(v + hg (v ))
a
U(t ) B X0 g(v + hg (v ))

U(t ) B X0 g(v + hg (v ))
U(t ) B X0 g(v + hg (v ))
a
U(t ) B X0 g(v + hg (v ))
a
U(t ) B X0 g(v + hg (v ))

U(t ) B X0 g(v + hg (v ))

d,
U(a ) B X0 g(v + hg (v ))
d .
d
+
d
+
d,
d
+
U(a ) B X0 g(v + hg (v ))
d .
d,
U(t ) B X0 g(v + hg (v ))
d.
U(t ) B X0 g(v + hg (v ))
+
d,
60
which implies that for any t a, vt + hg (vt ) = U(t a) (v + hg (v )) + lim a a

a
Finally, we conclude that vt + hg (vt ) is an integral solution of equation (1.1) on lR with initial value + hg (). Theorem 3.2. Let v be the solution of equation (3.1) on lR. Then, for t lR vt K || e 2 |t| and vt + hg (vt ) 2K || e 2 |t| . Conversely, if we choose such that 2K (|P | + |P+ | + 3 |P0 |) < 1, then for any integral solution u of equation (1.1) on lR with ut = O(e 2 |t| ), we have ut Mg for all t lR. Proof. Let v be the solution of equation (3.1). Then, using the estimate (3.3), we obtain that vt K || e 2 |t| for t lR and from the fact that Lip(h) 1 and hg (0) = 0, we obtain vt + hg (vt ) 2K || e 2 |t| for t lR. Let u be an integral solution of equation (1.1) such that ut = O(e 2 |t| ). Then there exists a positive constant k0 such that |ut | k0 e 2 |t| for all t lR. Note that ut = U(t s)us + lim On the other hand, ut+ = U(t s)u+ + lim s
s s

U(t )B X0 g(u )d for t s.
U(t ) B X0 g(u )
Moreover, for s t and s 0, we have U(t s)u+ Ke(ts) |P+ us | k0 K |P+ | e(ts) e 2 |s| = k0 K |P+ | et e 2 s . s Therefore,
s

lim U(t s)u+ = 0. s
It follows that ut+ = lim Similarly, we can prove that ut = lim
U(t ) B X0 g(u )
U(t ) B X0 g(u )
U(t ) B X0 g(v + hg (v )) d.
d for s t.
d.
d.
Center Manifold and Stability in Critical Cases for Some Partial Functional ... We conclude that ut = ut+ + ut +U(t)u0 + lim 0
0
61
Let Y0 such that = u0 . By Theorem 3.1, there exists an integral solution w of equation (1.1) on lR with initial value + hg () such that wt Mg for all t lR and wt = U(t) + lim + lim + lim Then, for all t lR, we have |ut wt | lim
0 t 0 t
U(t ) B X0 g(w ) U(t ) B X0 g(w )
U(t ) B X0 (g(u ) g(w ))

t
+ lim + lim This implies that |ut wt | K |P0 | + |P |
U(t ) B X0 (g(u ) g(w )) U(t ) B X0 (g(u ) g(w ))
t 0
e 3 |t| |u w | d
t
e(t) |u w | d + |P+ |
Let N(t) = e 2 |t| |ut wt | for all t lR. Then, N = sup N(t) < . On the other hand, we tl R have
t 0
N(t) KN |P0 | Finally we arrive at
e 6 |t| d + |P |
e 2 (t) d + |P+ |
2K (3 |P0 | + |P | + |P+ |) N.
Consequently, N = 0 and ut = wt for t lR.
Attractiveness of the Center Manifold
In this section, we assume that there exists no characteristic value with a positive real part and hence the unstable space Y+ is reduced to zero. We establish the following result on the attractiveness of the center manifold.
U(t ) B X0 g(u )
d for t lR.
U(t ) B X0 g(w )
+
d.
d
+
d .
e(t) |u w | d .
e 2 (t) d .
62
Theorem 4.1. There exist 0, K1 0 and 3 , such that if Lip(g) < , then any + satises integral solution ut () of equation (1.1) on R
ut () hg (ut0 ()) K1 et hg (0 ) for t 0. Proof. The proof of this theorem is based on the following technically lemma.
(4.1)
Lemma 4.2. There exist 0, K0 0 and 3 , such that if Lip(g) < , then there is a continuous bounded mapping p : lR+ Y0 Y Y such that any integral solution ut () of equation (1.1) satises
ut () = p(t, ut0 (), ) for t 0. Moreover, |p(t, 1 , 1 ) p(t, 2 , 2 )| K0 |1 2 | + et |1 2 | for all 1 , 2 Y0 , 1 , 2 Y and t 0.
(4.2)
(4.3)
Idea of the proof of Lemma 4.2. The proof is similar to the one given in [22]. Let Y0 and Y . For t 0 and 0 t, we consider the system q(,t, , ) = U(t) lim and p(t, , ) = U(t) + lim
0
Using the contraction principle, we can prove the existence of q and p. The expression (4.2) and the estimate (4.3) are obtained in a completely similar fashion to that in [22]. Proof of Theorem 4.1. Let Mg be the center manifold of equation (1.1). Then any integral solution lying in Mg must satisfy (4.2). Let ut = ut ( + 0 ) be an integral solution of equation (1.1) on R+ with initial value + 0 . Let 0. Then, u0 + hg (u0 ) Mg and the corresponding integral solution exists on R and lies on Mg . This solution can be considered as an integral solution of equation (1.1) starting from + 0 at 0. Let vt = vt ( + 0 ) be the integral solution corresponding to + 0 . Using Lemma 4.2, we conclude that u hg (u0 ) = p(, u0 , ) p(, u0 , ), which implies that u hg (u0 ) K0 e . Since Lip(h) 1, we have u hg (u0 ) K0 e hg (0 ) + 0 0 . (4.4)
U(s) B X0 g (q(s,t, , ) + p(s, q(s,t, , ), ))
ds,
U(t s) B X0 g(q(s,t, , ) + p(s, q(s,t, , ), ))
ds.
63
The initial values 0 and 0 correspond to the solutions of the following equations for 0s vs = U(s )v + lim

v = U(s )v + lim s
Note that v = v . It follows, for 0 s , that |vs v | K(1+K0 ) |P0 | s

s
e 3 (s) |v v | d+KK0 |P0 |
Then by Gronwalls lemma, we deduce that there exists a positive constant which depends only on constants , K, K0 and such that, for 0 s , we have |vs v | . s If we assume that Lip(g) is small enough such that < 1, then 0 0 hg (0 ) + hg (0 ) hg (0 ) , which gives that 0 0 We conclude that ut hg (ut0 ) 1 K0 et hg (0 ) for t 0. 1 hg (0 ) . 1
As an immediate consequence, we obtain the following result on the attractiveness of the center manifold. Corollary 4.3. Assume that Lip(g) is small enough and the unstable space Y+ is reduced to zero. Then the center manifold Mg is exponentially attractive. We also obtain. Proposition 4.4. Assume that Lip(g) is small enough and the unstable space Y+ is reduced to zero. Let w be an integral solution of equation (1.1) that is bounded on R. Then wt Mg for all t R. Proof. Let w be a bounded integral solution of equation (1.1). Since, the equation (1.1) is autonomous, then for 0, wt + is also an integral solution of equation (1.1) for t 0 with initial value w at 0. It follows by the estimation (4.1) that wt+ () hg (wt0 + ()) K1 et w hg (w0 ) for t 0. Let t . Then wt hg (wt0 ) K1 e(t) w hg (w0 ) for t . (4.5)
Since w is bounded on R, letting , we obtain that wt = hg (wt0 ) for all t R.
U(s ) B X0 g(v + p(, v , )) U(s ) B X0 g(v + p(, v , ))
d,
0
d.
e 3 (s) e d .
64
Flow on the Center Manifold
In this section, we establish that the ow on the center manifold is governed by an ordinary differential equation in a nite dimensional space. In the sequel, we assume that the function g satises the conditions of Theorem 4.1. We also assume that the unstable space Y+ is reduced to zero. Let d be the dimension of the center space Y0 and = (1 , ...., d ) be a basis of Y0 . Then there exists d-elements = ( , ...., ) in Y , the dual space of Y, such 1 d that , j := ( j ) = i j , 1 i, j d, i i and = 0 on Y . Denote by the transpose of ( , ...., ). Then the projection operator i 1 d P0 is given by P0 = , . Since U 0 (t) t0 is a strongly continuous group on the nite dimensional space Y0 , by Theorem 2.15, p. 102 in [10], we get that there exists a d d matrix G such that U 0 (t) = eGt for t 0.
Let n N, n n0 and i {1, ...., d} . We dene the function xni on E by xni (y) = , Bn (X0 y) . i Then xni is a bounded linear operator on E. Let xn be the transpose of (xn1 , ..., xnd ), then xn , y = , Bn (X0 y) .
Consequently,
nn0 sup |xn | < ,
which implies that (xn )nn0 is a bounded sequence in L (E, Rd ). Then, we get the following important result. Theorem 5.1. There exists x L (E, Rd ) such that (xn )nn0 converges weakly to x in the sense that xn , y x , y as n for y E.
For the proof, we need the following fundamental theorem [25, pp. 776] Theorem 5.2. Let X be a separable Banach space and (z )n0 be a bounded sequence in n X . Then there exists a subsequence zk k0 of (z )n0 which converges weakly in X in n n the sense that there exists z X such that zk , y z , y as k for x X. n
Proof of Theorem 5.1. Let Z0 be a closed separable subspace of E. Since (xn )nn0 is a bounded sequence, by Theorem 5.2 there is a subsequence xnk kN which converges weakly to some xZ0 in Z0 . We claim that all the sequence (xn )nn0 converges weakly to in Z . This can be done by way of contradiction. Namely, suppose that there exists a xZ0 0
subsequence xn p pN
65
of (xn )nn0 which converges weakly to some xZ0 with xZ0 = xZ0 . Let
ut (., , , f ) denote the integral solution of the following equation d u(t) = Au(t) + L(ut ) + f (t), t dt u = C, where f is a continuous function from R to E. Then by using the variation of constants formula and the spectral decomposition of solutions, we obtain P0 ut (., , 0, f ) = lim and
P0 Bn X0 f () = , Bn X0 f () = xn , f () .
n+
It follows that P0 ut (., , 0, f ) = lim

n+
= lim
n+
For a xed a Z0 , set f = a to be a constant function. Then

t k+
lim
e(t)G xnk , a d = lim
p+
which implies that

t e(t)G xZ0 , a d =
e(t)G xZ0 , a d for a Z0 .
Consequently xZ0 = xZ0 , which yields a contradiction. We now conclude that the whole sequence (xn )nn0 converges weakly to xZ0 in Z0 . Let Z1 be another closed separable subspace of X. By using the same argument as above, we obtain that (xn )nn0 converges weakly to xZ1 in Z1 . Since Z0 Z1 is a closed separable subspace of = x in Z Z . For any y E, we dene x by E, we get that xZ1 0 1 Z0 x , y = xZ , y ,
where Z is an arbitrary given closed separable subspace of E such that y Z. Then x is well dened on E and x is a bounded linear operator from E to Rd such that
|x | sup |xn | < , nn0 and (xn )nn0 converges weakly to x in E.
As a consequence, we conclude that
U (t ) Bn X0 f ()
d,
t t
e(t)G , Bn X0 f () d,
e(t)G xn , f () d.
e(t)G xn p , a d for a Z0 ,
66
Corollary 5.3. For any continuous function f : R E, we have

t n+
lim
U (t ) Bn X0 f ()
d =
Let Y0 such that + hg () Mg . From the properties of the center manifold, we know that the integral solution starting from + h() is given by vt + hg (vt ), where vt is the solution of vt = U(t) + lim
0
U(t ) B X0 g(v + hg (v ))

Let z(t) be the component of vt . Then z(t) = vt fort lR. By Theorem 5.1 and Corollary 5.3, we have z(t) = vt = eGt z(0) + We conclude that z satises z(t) = eGt z(0) + lim
n 0
t 0
e(t)G x , g(v + hg (v )) d fort lR.
eG(t) xn , g(v + hg (v )) d fort lR.
Finally we arrive at the following ordinary differential equation, which determines the ow on the center manifold z (t) = Gz(t) + x , g(z(t) + hg (z(t))) for t lR. (5.1)
Stability in Critical Cases
In this section, we suppose that Lip(g) < , where is given by Theorem 4.1. Here we study the critical case where the unstable space Y+ is reduced to zero and the exponential stability is not possible, which implies that there exists at least one characteristic value with a real part equals zero. Theorem 6.1. Assume that Lip(g) is small enough. Then there exists a positive constant K2 such that for each Y, there exists z0 Rd such that ut0 z(t) + ut hg (z(t)) K2 et hg (0 ) for t 0, (6.1)
where z is the solution of the reduced system (5.1) with initial value z0 and u is the integral solution of equation (1.1) with initial value . Proof. Let Y and u be the corresponding integral solution of equation (1.1). Then u0 = U(s t)ut0 + lim s
t
U(s ) B X0 g(u + u0 )
Also let s ws,t be the solution of the following equation ws,t = U(s t)ut0 + lim
t
U(s ) B X0 g(w,t + hg (w,t ))
e(t)G x , f () d for t, R.
d fort lR.
d for 0 s t.
d for 0 s t.
Center Manifold and Stability in Critical Cases for Some Partial Functional ... Then, for 0 s t, we have u0 ws,t 2K |P0 | s It follows that e 3 s u0 ws,t 2K |P0 | s
67
t s
e 3 (s) u0 w,t d + K |P0 |
t s
e 3 u0 w,t d + K |P0 |
Using Gronwalls lemma, we obtain that u0 ws,t 2K |P0 | s where = 2K |P0 | +

3.
t s
e(s) u hg (u0 ) d for 0 s t,
By Theorem 4.1, we obtain
t s
u0 ws,t 2KK1 |P0 | s Let us recall that Consequently,

3,
e()(s) d
u hg (u0 ) for 0 s t. s s
. Then we can choose small enough such that < 0. 2KK1 |P0 | us hg (u0 ) for 0 s t, s 2KK1 |P0 | u0 hg (u0 ) for t 0. 0
u0 ws,t s and for s = 0, we have
u0 w0,t 0
We deduce that {w0,t : t 0} is bounded in Y0 and then there exists a sequence tn as n and Y0 such that w0,tn as n . Let w(., ) be the solution of the following equation wt (., ) = U(t) + lim
0
U(t ) B X0 g(w (., ) + hg (w (., )))
By the continuous dependence on the initial data, we obtain, for all s 0 ws (0, ) = lim ws (0, w0,tn ) = lim ws (0, wtn (0, utn )),
n n
= lim wstn (0, utn ) = lim ws,tn .

n n
By (4.4) and (6.2), there exists a positive constant K2 such that u0 ws (0, ) K2 es u hg (u0 ) for all s 0. s 0 0 If we put z(t) = wt (0, ) for t lR, z(t) = eGt z(0) +
then
t 0
e(t)G x , g(z() + hg (z())) d for t lR.
Finally, the estimation (6.1) follows from (4.4).
e 3 (s) u hg (u0 ) d.
t s
e 3 u hg (u0 ) d.
(6.2)
d for t 0.
68
Now, we can state the following result on the stability by using the reduction to the center manifold. Theorem 6.2. If the zero solution of equation (5.1) is uniformly asymptotically stable (unstable), then the zero solution of equation (1.1) is uniformly asymptotically stable (unstable). Proof. Assume that 0 is uniformly asymptotically stable for equation (5.1). For 0, let B = + 0 Y Y0 : + 0 < , and Mg B for some 0, be the region of attraction of 0 for equation (5.1). First, we prove that 0 is stable for equation (1.1). Let 0. Then there exists < such that |z(t)| < for t 0, provided that |z(0)| < , where z is a solution of (5.1). As 0 is assumed to be uniformly asymptotically stable for equation (5.1), there exists t0 = t0 () such that |z(t)| < 2 , for t t0 . Without loss of generality, we can choose and t0 so that 1 max(K1 , K2 )et0 < 2 . By the continuous dependence on the initial value for equation (1.1), there exists 1 < 2 such that if + 0 V1 := + 0 Y Y0 : 0 < 1 1 , hg (0 ) < 2 2 ,
then the corresponding integral solution ut = ut ( + 0 ) of equation (1.1) satises ut B for t [0,t0 ] . Moreover, ut hg (ut00 ) < 0 1 . 2
Furthermore, by Theorem 6.1, there exists z0 Rd such that ut0 z(t) K2 et hg (0 ) for t 0, (6.3)
where z is a solution of the reduced system (5.1) with initial value z0 such that |z0 | < . It follows that ut00 < . Consequently, ut0 B and ut must be in B for all t 0. This completes the proof of the stability. Now we deal with the local attractiveness of the zero solution. For a given integral solution u(., ) of equation (1.1) which is assumed to be bounded for t 0, it is well known that the -limit set () is nonempty, compact, invariant and connected since the map ut (, ) is compact for t r. For the attractiveness of 0, let V be chosen as above and V . Then the integral solution u of equation (1.1) starting from lies in B . The -limit set () of u is nonempty and invariant and must be in Mg B . Since the equilibrium 0 of (1.1) is uniformly asymptotically stable, we deduce by Theorem 11.4, p. 111 [23] and by the LaSalle invariance
69
principle that the only invariant set in Mg B must be zero. Consequently, the -limit set () is zero and ut (., ) 0 as t 0. Assume now that the zero solution of the reduced system (5.1) is unstable. Then there exist 1 0, a sequence (tn )n of positive real numbers and a sequence (zn )n converging to 0 such that |z(tn , zn )| 1 , where z(., zn ) is a solution of (5.1). On the other hand z(., zn ) + hg (z(., zn )) is an integral solution of equation (1.1) and |z(tn , zn ) + hg (z(tn , zn ))| (1 Lip(hg )) |z(tn , zn )| . Moreover, Lip(hg ) can be chosen such that 1 Lip(hg ) 0. It follows that |z(tn , zn ) + hg (z(tn , zn ))| (1 Lip(hg )) 2 , for some 2 0. Consequently, the zero solution of equation (1.1) is unstable.
Local Existence of the Center Manifold
In this section we prove the existence of the local center manifold when g is only dened in a neighborhood of zero. We assume that (H4 ) There exists 1 0 such that g : B(0, 1 ) E is C1 -function, g(0) = 0 and g (0) = 0, where B(0, 1 ) = { C : || < 1 } . For < 1 , we dene the cut-off function g : C E by g () = g() if || , g( || ) if || .
We consider the following partial functional differential equation d u(t) = Au(t) + L(ut ) + g (ut ) for t 0 dt u0 = C. (7.1)
Theorem 7.1. Assume that (H1 ), (H3 ) and (H4 ) hold. Then there exist 0 < < 1 and Lipschitz continuous mapping hg : Y0 Y Y+ such that hg (0) = 0 and the local Lipschitz manifold Mg = + hg () : Y0 is globally invariant under the ow associated to equation (7.1). Proof. Using the same arguments as in [26], Proposition 3.10, p.95, one can show that g is Lipschitz continuous with Lip(g ) 2 sup g () .
||<
70
It follows that Lip(g ) goes to zero when goes to zero. According to Theorem 3.1, we deduce that there exist 0 and mapping hg : Y0 Y Y+ with hg (0) = 0 such that the Lipschitz manifold Mg = + hg () : Y0 is globally invariant under the ow of equation (7.1). Denition 7.2. The local center manifold associated to equation (1.1) is dened by Mg = + hg () : Y0 B(0, ). We deduce that the local center manifold contains all bounded integral solutions of equation (1.1) which are bounded by . Moreover, the following interesting results hold. Theorem 7.3. (Attractiveness). Assume that (H1 ), (H3 ), (H4 ) hold and the unstable space Y+ is reduced to zero. Then there exist 0 < < 1 , K3 0 and 3 , such that any integral solution ut () of equation (1.1) with initial data B(0, 1 ), exists on R+ and satises ut () hg (ut0 ()) K3 et hg (0 ) for t 0. (7.2) Theorem 7.4. (Reduction principle and stability). Assume that (H1 ), (H3 ), (H4 ) hold and the unstable space Y+ is reduced to zero. If the zero solution of equation (5.1) is uniformly asymptotically stable (unstable), then the zero solution of equation (1.1) is uniformly asymptotically stable (unstable).
Application
In this section, we assume that (H5 ) (AU ) { C : Re() 0} = {0} and 0 is a simple eigenvalue of AU . In this case, the reduced system (5.1) on the center manifold becomes z (t) = (z(t)), where is the scalar function given by (z) = x , g(z + hg (z)) for z R. Theorem 8.1. Assume that (H1 ), (H3 ), (H4 ), (H5 ) hold and satises (z) = am zm + am+1 zm+1 + .... (8.1)
If m is odd and am < 0, then the zero solution of equation (1.1) is uniformly asymptotically stable. If am > 0 then the zero solution of equation (1.1) is unstable. Proof. The proof is based on Theorem 6.2 and on the following known stability result.
Center Manifold and Stability in Critical Cases for Some Partial Functional ... Theorem 8.2. [6] Consider the scalar differential equation z (t) = am zm + am+1 zm+1 + ....
71
(8.2)
If m is odd and am < 0, then the zero solution of equation (8.2) is uniformly asymptotically stable. If am > 0, then the zero solution of equation (8.2) is unstable. Concluding remark. Assumption (8.1) is natural and it is a consequence of the smoothness of the center manifold, which states that if g is a Ck -function, for k 1, then hg is also a Ck -function. Consequently if g is a C -function, then the center manifold hg is also a C -function. Assumption (8.1) can be obtained by using the approximation of the center manifold hg . The proof of the smoothness result is omitted here and it can be done in similar way as in [11].
Acknowledgements
A part of this work has been done when the second author was visiting the Abdus Salam International Centre for Theoretical Physics, ICTP, Trieste-Italy. He would like to acknowledge the centre for the support.
References
` [1] M. Adimy and K. Ezzinbi, Semi groupes int gr s et equations diff rentielles a retard e e e en dimension innie, C. R. Acad. Sci. Paris, t. 323, s rie I, 481-486, (1996). e [2] M.Adimy and K.Ezzinbi, A class of linear partial neutral functional differential equations with non-dense domain, Journal of Differential Equations, 147, 285-332, (1998). [3] M. Adimy and K. Ezzinbi, Local existence and linearized stability for partial functional differential equations, Dynamic Systems and Applications, Vol. 7, 389-403, (1998). [4] M. Adimy and K. Ezzinbi, Existence and stability of solutions for a class of partial neutral functional differential equations, Hiroshima Mathematical Journal, Vol. 34, No. 3, 251-294, (2004). [5] M. Adimy, K. Ezzinbi and M. Laklach, Spectral decompostion for some partial neutral functional differential equations, Canadian Applied Mathematics Quarterly, Vol. 9, No. 4, 1-34, (2001). [6] J. Carr, Applications of Center Manifold Theory, Applied Mathematical Sciences, Springer-Verlag, Vol. 35, (1981). [7] S. N. Chow and K. Lu, Invariant manifolds for ows in Banach spaces, Journal of Differential equations, 74, 285-317, (1988). [8] G. Da Prato and A. Lunardi, Stability, instability and center manifold theorem for fully nonlinear autonomous parabolic equations in Banach spaces, Archive for Rational Mechanics and Analysis, Vol. 101, 115-141, (1988).
72
[9] W. Desch and W. Schappacher, Linearized stability for nonlinear semigroups, in Differential equations in Banach spaces, (A. Favini and E. Obrecht, Eds), Lecture Notes in Mathematics, Springer-Verlag, Vol. 1223, 61-73, (1986). [10] O. Diekmann, S. A. van Gils, S. M. Verduyn Lunel and H-O Walther, Delay Equations, Functional Complex and Nonlinear Analysis, Applied Mathematical Sciences, Springer-Verlag, (1995). [11] T. Faria, W. Huang and J. Wu, Smoothness of center manifolds for maps and formal adjoints for semilinear FDEs in general Banach spaces, SIAM, Journal of Mathematical Analysis, 34, 173-203, (2002). [12] J. K. Hale, Critical cases for neutral functional differential equations, Journal of Differential Equations, 10, 59-82, (1971). [13] J. K. Hale, Theory of Functional Differential Equations, Springer-Verlag, (1977). [14] D. Henry, Geometric Theory of Semilinear Parabolic Equations, Lecture Notes in Mathematics, Springer-Verlag, Vol. 840, (1981). [15] A. Keller, The stable, center-stable, center, center-unstable and unstable manifolds, Journal of Differential Equations, 3, 546-570, (1967). [16] A. Keller, Stability of the center-stable manifold, Journal of Mathematical Analysis and Applications, 18, 336-344, (1967). [17] X. Lin, J. W. H. So and J. Wu, Centre manifolds for partial differential equations with delays, Proceedings of the Royal Society of Edinburgh, 122, 237-254, (1992). [18] N. V. Minh and J. Wu, Invariant manifolds of partial functional differential equations, Journal of Differential Equations, 198, 381-421, (2004). [19] K. Palmer, On the stability of the center manifold, Journal of Applied Mathematics and Physics, (ZAMP), Vol. 38, 273-278, (1987). [20] S. N. Shimanov, On the stability in the critical case of a zero root for systems with time lag, Prikl. Mat. Mekh. 24, 447-457, (1960). [21] C. C. Travis and G. F. Webb, Existence and stability for partial functional differential equations, Transactions of the American Mathematical Society, 200, 395-418 (1974). [22] J. W. H. So, Y. Yang and J. Wu, Center manifolds for functional partial differential equations: Smoothness and attractivity, Mathematica Japonica, 48, 67-81, (1998). [23] T. Yoshizawa, Stability Theory and the Existence of Periodic Solutions and Almost Periodic Solutions, Applied Mathematical Sciences, Springer-Verlag, Vol. 14, (1975). [24] A. Vanderbauwhede and S. A. Van Gils, Center manifolds and contractions on a scale of Banach spaces, Journal of Functional Analysis, 72, 209-224, (1987).
73
[25] E. Zeidler, Nonlinear Functional Analysis and its Applications, Tome I, Fixed Point theorems, Springer-Verlag, (1993). [26] G. F. Webb, Theory of Nonlinear Age-dependent Population Dynamics, Marcel Dekker, (1985). [27] J. Wu, Theory and Applications of Partial Functional Differential Equations, Applied Mathematical Sciences, Springer-Verlag, Vol. 119, (1996).
S TABILITY R ADII OF P OSITIVE L INEAR F UNCTIONAL D IFFERENTIAL S YSTEMS IN BANACH S PACES

Pham Huu Anh Ngoc 1 Nguyen Van Minh 2 and Toshiki Naito 3 , 1 Department of Mathematics, University of Hue, 32 Le Loi Street Hue City, Vietnam 2 Department of Mathematics, State University of West Georgia, Carrollton, GA 30118 3 Department of Mathematics, The University of Electro-Communications, Chofu, Tokyo 182-8585. Japan
Abstract In this paper we study stability radii of positive linear functional differential systems in Banach spaces under multi-perturbations and multi-afne perturbations. We prove that for the class of positive systems, complex stability radius, real stability radius and positive stability radius of positive systems under multi-perturbations (or multi-afne perturbations) coincide and they are computed via a simple formula. We illustrated the obtained results by an example.
Key Words: linear functional differential equation in Banach space, positive system, stability radius, multi-perturbation, multi-afne perturbation. AMS Subject Classication: 34 K20, 93 D09
1 Introduction
Over the last decades there has been an increasing interest in the robust stability of dynamical systems that has many applications in control engineering. In the analysis of robust stability, the notion of stability radius shows to be an appropriate tool. By denition, the stability radius of a given asymptotically stable system x(t) = Ax(t) is the maximal > 0 for which all the systems of the form x(t) = (A + DE)x(t), <
are asymptotically stable. Here, is a unknown disturbance matrix, D and E are given matrices dening the structure of the perturbations.
E-mail E-mail
address: phanhngoc@yahoo.com. Corresponding author address: vnguyen@westga.edu E-mail address: naito@e-one.uec.ac.jp
76
Pham Huu Anh Ngoc, Nguyen Van Minh and Toshiki Naito
In general, one may consider the disturbance matrix as real or complex. Accordingly, the maximal is called complex or real stability radius, respectively. The basic problem in the study of robustness of stability of the system is to characterize and compute these radii in terms of given matrices A, D, E. It is worth mentioning that these two stability radii are in general different. The analysis and computation of the complex stability radius for systems under structured perturbations was done rst in [8] and extended later in many subsequent papers (see [9] for a survey up till 1990) while the computation of the real stability radius, being a much more difcult problem, was solved only recently, see e.g. [24]. The situation is much simpler for the class of positive systems. It is shown (see [11], [26]) that if A is a Metzler matrix (i.e. all off-diagonal entries of A are nonnegative) and D, E are nonnegative matrices, then the complex and the real stability radii coincide and can be computed directly by a simple formula. These results have been extended recently to innite dimensional case (see [1], [2]), for positive continuous time-delay systems (see [27]-[30]) and for discrete time-delay systems (see [13], [19]). It is worth noticing that the notion of stability radius can be extended to various perturbation types [9]. Among perturbation types, two of the following perturbation types A A + Di i Ei
i=1 N
(Multi-perturbation)
(1)
A A + i A i
i=1
(Multi-afne perturbation)
(2)
are most well-known in control theory that include the perturbation types studied in the literature. The problem of computing the stability radii of positive linear systems without delays under multi-perturbations was studied in [20]. Recently, in [21], the authors successfully computed the stability radii of positive linear functional differential equations under multi-perturbations and multi-afne perturbations. Although recently there have been many works dedicated to extending the abovementioned results to the innite dimensional case (see e.g. [1], [2], [3], [10], [23], [31]), the problem of computing the stability radii of positive linear functional differential equations in Banach spaces under multi-perturbations and multi-afne perturbations as considered in [21] has not been studied yet. And the main purpose of this paper is to extend the results of [21] to general retarded systems described by linear functional differential equations in Banach spaces of the form x(t) = A0 x(t) +
0 h
d[()]x(t + ),
where A0 is a generator of a compact C0 -semi-group and () is a function of bounded variation on [h, 0] with values in a Banach space. Formulae will be derived for the stability radii of these general systems in terms of the operator A0 and the function which dene the initial system. The organization of the paper is as follows. In the next section, we summarize some notations and give preliminary results on C0 - semigroups on Banach spaces and on Banach
Robust Stability of Linear Functional Differential Equations
77
lattice theory which will be used in the sequel. In Section 3, a lower and upper bound for the complex stability radius of a general functional differential equation under multiperturbations are given. Then, for the class of positive functional differential equations it will be shown that the complex, real, positive stability radii under multi-perturbations coincide and a simple formula for their computation can be established. In Section 4, we study stability radii of positive functional differential equations under multi-afne perturbations. We also prove that the positive, real, complex stability radii coincide and can be computed by an explicit formula. An example is given to illustrate the obtained results.
Preliminaries
Let (X, ) be a Banach space. An X-valued function () : [, ] X is said to be of bounded variation if Var(; , ) := sup
P[,] k
(k ) (k1 ) < +,
(3)
where the supremum is taken over the set of all nite partitions of the interval [, ]. The set BV([, ], X) of all X-valued functions () of bounded variation on [, ] satisfying () = 0 is a Banach space endowed with the norm = Var(; , ). Let Y,U be Banach spaces. Throughout this paper, we denote by L(Y,U), (L (Y,U)) the Banach space of all linear operators (bounded linear operators) from Y to U. Given () BV([, ], L (Y,U)) then for any continuous functions C([, ], K) and C([, ],Y ), the integrals

()d[()] and
p exist and are dened respectively as the limits of S1 (P) := k (k )((k ) (k1 )) and p S2 (P) := k ((k ) (k1 ))(k ) as d(P) := maxk |k k1 | 0, where P = {1 = 2 n = } is any nite partition of the interval [, ] and k [k1 , k ] (see e.g. [5]). It is immediate from the denition that ()d[()]
max[,] |()| , (4) max[,] () .
d[()]()
L =
denes a bounded linear operator from C([, ],Y ) to U. In the subsequent sections the following subspace of BV([h, 0], X) will be used frequently : NBV([h, 0], X) := { BV([h, 0], X); (h) = 0, is c.f.l. on (h, 0)}. (6)
It is clear that NBV([h, 0], X) is closed in BV([h, 0], X) and thus it is a Banach space with the norm = Var(; h, 0). Denote by C([h, 0], X) the Banach space of all continuous functions on [h, 0] with values in X normed by the supremum norm.
Then,
d[()](), C([h, 0],Y ),
d[()]()
(5)
78
Let (T (t))t0 be a strongly continuous semigroup (or shortly, C0 -semigroup) of bounded linear operators on complex Banach space (X, ). Denote by A the generator of the semigroup (T (t))t0 and D (A) its domain. That is,
D (A) := x X : lim
T (t)x x X t0 t
and
T (t)x x , x D (A). t0 t Since A is a closed operator, D (A) is a Banach space with the graph norm Ax = lim x
D (A)
:= x + Ax ,
x D (A).
(7)
The resolvent set (A), by denition, consists of all C for which (IX A) has a bounded linear inverse in X. The complement of (A) in C is called the spectrum of A and denoted by (A). With the C0 -semigroup (T (t))t0 , we associate the following quantities: 1. The spectral bound, s(A) := sup{Re (A)}, where (A) is spectrum of the linear operator A. 2. The abscissa of uniform boundedness of the resolvent of A, s0 (A) := inf R : {Re > } (A) and sup
Re>
R(, A) < .
3. The growth bound 1 (A), 1 (A) := inf{ R : there exists M > 0 such that T (t)x Met x 4. The uniform growth bound 0 (A), 0 (A) := inf{ R : there exists M > 0 such that T (t) Met for all and t 0}. It is well-known that s(A) 1 (A) s0 (A) 0 (A) +, see, e.g [16], [22]. Next, the C0 -semigroup T (t)t0 is called : 1. Hurwitz stable if (A) C := { C : Re < 0}, 2. Strictly Hurwitz stable if s(A) < 0, (8)
D(A)
for all x D (A) and t 0}.
Robust Stability of Linear Functional Differential Equations 3. Exponentially stable if 1 (A) < 0, 4. Uniformly exponentially stable if 0 (A) < 0. It is well-known that for an eventually norm continuous semigroup, that is,
tt0
79
lim T (t) T (t0 ) = 0, for some t0 0,
s(A) = 1 (A) = s0 (A) = 0 (A), see e.g. [16]. So, the notations of strictly Hurwitz stability, exponentially stability, uniformly exponentially stability coincide. To make the presentation self-contained, we give some basic facts on Banach lattices which will be used in the sequel (see, e.g. [15]). Let X = {0} be a real vector space endowed with an order relation . Then X is called an ordered vector space. Denote the positive elements of X by X+ := {x X : 0 x}. If furthermore the lattice property holds, that is, if x y := sup{x, y} X, for x, y X, then X is called a vector lattice. It is important to note that X+ is generating, that is, X = X+ X+ . Then, the modulus of x X is dened by |x| := x (x). If is a norm on the vector lattice X satisfying the lattice norm property, that is, if |x| |y| x y , x, y X, (9)
then X is called a normed vector lattice. If, in addition, (X, ) is a Banach space then X is called a (real) Banach lattice. We now extend the notion of Banach lattices to the complex case. For this extension all underlying vector lattices X are assumed to be relatively uniformly complete, that is, if for every sequence (n )nN in R satisfying |n | < + and for every x X and every n=1 sequence (xn )nN in X it holds that 0 xn n x sup
nN i=1
xi
X.
Now let X be a relatively uniformly complete vector lattice. The complexication of X is dened by XC = X + X. The modulus of z = x + y XC is dened by |z| = sup |(cos )x + (sin )y| X.
02
(10)
A complex vector lattice is dened as the complexication of a relatively uniformly complete vector lattice endowed the modulus (10). If X is normed then x := |x| , x XC (11)
denes a norm on XC satisfying the lattice norm property. If X is a Banach lattice then XC endowed the modulus (10) and the norm (11) is called a complex Banach lattice. Throughout of this paper, for simplicity of presentation, we denote X, XR instead of XC , X, respectively. Let ER , FR be real Banach lattices and T L(ER , FR ). Then T is called positive and denoted by T 0 if T (E+ ) F+ . By S T we mean T S 0, for T, S L(ER , FR ).
80
An operator T L(E, F) is called real if T (ER ) FR . Then a operator T L(E, F) is called positive (T 0) if T is real and T (E+ ) F+ . We introduce the notations LR (E, F) := {T L(E, F) : T real} ; L R (E, F) := {T L (E, F) : T real}, L+ := {T L(E, F) : T 0} ;
L + (E, F) := {T L (E, F) : T 0}.
(12) (13)
For T L + (E, F), we emphasize the simple but important fact T = sup
xE+ , x =1
Tx .
(14)
Finally, NBV([, ], L R (E, F)) is called increasing if (1 ) (2 ) 0 for 1 / 2 . In this paper, we always dene inf 0 = +, 01 = +.
3
3.1
Stability Radii of Linear Abstract Functional Differential Equations under Multi-perturbations

Abstract Functional Differential Equations
We let X be a complex Banach lattices which makes C([h, 0], X) a Banach lattice as well. We consider a linear retarded system described by the following general functional differential equation x(t) = A0 x(t) + Lxt , t 0, x(t) X (15) x() = 0 (), [h, 0]. where, for each t 0, xt C := C([h, 0], X) is dened by xt () = x(t + ), [h, 0], A0 is the generator of a C0 -semigroup on X and L : C([h, 0], X) X is a bounded linear operator dened by L =
0 h
d[()](), t 0, C([h, 0], X).
(16)
Here () NBV([h, 0], L (X)) is a given L (X)-valued function of bounded variation on [h, 0] such that vanishes at h and is c.f.l. on [h, 0]. We shall extend the denition of to R by setting () = (h) = 0 for all h, () = (0) for all 0. Denition 3.1. A function x C([h, 0], X) is called a solution of the linear retarded system (15)-(16), if a) x is right-size differentiable at 0 and continuously differentiable for t > 0. b) x(t) D (A0 ) for t 0. c) (15) is satised for t 0. To (15)-(16), we associate the following operator A on the Banach space C([h, 0], X). Let A be the differential operator A f = f , D (A) := { f C1 ([h, 0], X) : f (0) D (A0 ), f (0) = A0 f (0) + L f }. (17)
81
Theorem 3.2. [16] The operator A dened in (17) is the generator of a C0 -semigroup (T (t))t0 on C([h, 0], X) satisfying the translation property T (t) f (s) = f (t + s) if t + s 0 T (t + s) f (0) if t + s > 0, (18)
f C([h, 0], X). Moreover, for 0 D (A) dene x : [h, ) X by x(t) = 0 (t) if h t 0 T (t)0 (0) if t > 0. (19)
Then x is the unique solution of the linear retarded system (15)-(16) . The C0 -semigroup (T (t))t0 in the above theorem is called solution semigroup of linear retarded system (15)-(16) and its operator A is called the solution operator of the linear retarded system (15)-(16). Denition 3.3. The linear retarded system (15)-(16) is called, respectively, Hurwitz stable, strictly Hurwitz stable, exponentially stable, uniformly exponentially stable if its solution semigroup (T (t))t0 is Hurwitz stable, strictly Hurwitz stable, exponentially stable, uniformly exponentially stable. H(s) := sIX A0
Dene
0 h
es d[()].
(20)
H(s) is called the characteristic operator of the linear retarded system (15)-(16). By the retarded resolvent set (A0 , ), we understand the set of all s C for which H(s) has a bounded linear inverse in X. The complement of (A0 , ) in C is called the retarded spectrum and denoted by (A0 , ). Then, the following fact can be found in [16], page 224. (A) = (A0 , ). (21)
Assuming that A0 generates a C0 -compact semigroup (U(t))t0 of bounded linear operators, it was shown in [32, 33, 17, 18] that the initial value problem u(t) = U(t)(0) + u0 = C ,
0 t
U(t s)L(us )ds,
t 0,
has a unique continuous solution u(t; ) for t h, and {T (t)}t0 , T (t) = ut (; ), is a C0 -semigroup of linear (and compact for t > h) operators on C , with innitesimal generator A given as above. Moreover, one has (A) = P (A) = {s C : H(s)y = 0, for some y dom(A0 ) \ {0}} where p (A) denotes the point spectrum of A. In this case, since (T (t))t0 is eventually compact, it is eventually norm continuous. Hence, the above concepts of stability are actually the same if A0 generates a compact semigroup. For more information on the theory and applications of abstract functional differential equations we refer the reader to [6, 7, 17, 18, 33].
82
3.2
Stability Radii
Let X, Y, Ui0 ,Ui1 (i N := {i {1, 2, ..., N}) be complex Banach lattices and D0 i L (Ui0 , X), D1 L (Ui1 , X) (i N ), E L (X,Y ). i Assume that the retarded system (15)-(16) is Hurwitz stable and A0 , are subject to multi-perturbations of the type A0 A0 = A0 + N D0 i E, i=1 i = + N D1 i E, i=1 i i L (Y,Ui0 ), i N (22) i NBV([h, 0], L (Y,Ui1 )), i N and thus the perturbed system is described by
x(t) = (A0 +N D0 i E)x(t)+ i=1 i x() = 0 (), [h, 0].
0 N 1 h d[() + i=1 Di i ()E]x(t + ),
t 0 (23)
Here D0 L (Ui0 , X), D1 L (Ui1 , X)(i N ) and E L (X,Y ) are xed and describe the i i structure of perturbations, while i L (Y,Ui0 ) and i () NBV([h, 0], L (Y,Ui1 )), i N are unknown disturbances. We shall measure the size of each perturbation := [, ] where := (0 , ..., N ), := (0 , ..., N ), i L (Y,Ui0 ), i NBV([h, 0], L (Y,Ui1 )), i N , by the norm := i + i ,
i=1 i=1 N N
i := Var(i ; h, 0), i N.
(24)
Set D C := { = [, ] : i L (Y,Ui0 ), i NBV([h, 0], L (Y,Ui1 )), i N}, D R := { = [, ] : i L R (Y,Ui0 ), i NBV([h, 0], L R (Y,Ui1 )), i N}, D + := { = [, ] : i L + (Y,Ui0 ), i NBV([h, 0], L R (Y,Ui1 )), i is increasing for every i N} Then D C , D R , D + is called respectively the class of complex, real, nonnegative with respect to the perturbation structure (22). Denote by A , ((T (t))t0 the solution opera tor, solution semigroup of the perturbed system (23), respectively. To study robustness of stability of the retarded system (15)-(16) we introduce the following. Denition 3.4. Let the retarded system (15)-(16) be Hurwitz stable. The complex, real, positive stability radius of the system with respect to perturbations of the form (22), measured by the norm (24), is dened respectively by rC = inf{ : D C , (A ) C }, rR = inf{ : D R , (A ) C }, r+ = inf{ : D + , (A ) C }. (25) (26) (27)
83
We dene the associated transfer functions of the perturbed system (23) by setting Gk (s) = EH(s)1 Dk i i
0 where H(s)1 := sIX A0 h es d[()] We need the following technical lemmas.
i N , k {0, 1},
1
(28)
, Re s > s(A).
Lemma 3.5. Let X,Y be Banach spaces and S L (Y, X), T L (X,Y ). Then, IX ST is invertible if and only if so is IY T S. Proof. Assume that IX ST is invertible. Setting R := (IX ST )1 , we can check that (IY T S)(IY + T RS) = IY = (IY + T RS)(IY T S). By the same way, we can prove that if IY T S is invertible then so is IX ST. Lemma 3.6. Assume that the supsC,Res0 H(s)1 < +. Then system (15)-(16) is Hurwitz stable and
(i) If supsC,Res0 G00 (s) = 0 for some i0 N then, for every > 0, there exists a complex i perturbation := [, ] D C , such that < and (A ) C . (30) (ii) If supsC,Res0 G10 (s) = 0 for some i0 N then, for every > 0, there exists a complex i perturbation := [, ] D C , such that < and (A ) C . (32) (iii) In particular, if G0 (0) L + (Ui0 ,Y ), G1 (0) L + (Ui1 ,Y ) for every i N and i i max{maxiN G0 (0) , maxiN G1 (0) } = 0 then, for every > 0, there exists a noni i negative perturbation D + satisfying < and (A ) C . (34) 1 max{maxiN G0 (0) i , maxiN G1 (0) } i + . (33) 1 + , supsC,Res0 G00 (s) |ehs | i (31) 1 supsC,Res0 G00 (s) i + , (29)
84
Proof. (i) It follows from supsC,Res0 H(s)1 < + that supsC,Re s 0 G00 (s) < +. i For > 0, there exists s0 C, Re s0 0 such that 1 G00 (s0 ) i Let u0 Ui0 u0 = 1 satisfy 0 < 1 supsC,Re s 0 G00 (s) i <
0
Theorem, there exists y (Y ) , y = 1 satisfying 0 0
1 G0 (s0 )u0 i
1 G0 (s0 ) i
0
+ 2 . Then, by the Hahn-Banach
y (G00 (s0 )u0 ) = G00 (s0 )u0 . 0 i i Dene 00 := i It is easy to see that 00 = i 00 Ex0 = u0 . Therefore, i This implies ((A0 + D00 00 E) + i i We dene := [, ] = ((0 , ..., N ), (0 , ..., N )) , where i = 00 if i = i0 otherwise i = 0, i N and i = 0 for all i N . i Then, we have = 00 < i 1 supsC,Re s 0 G00 (s) i + and s(A ) Re s0 0.
0 h
1 G00 (s0 )u0 i

1 G0 (s0 )u0 i
0
u0 y () L (Y,Ui0 ). 0 0
. Setting x0 := H(s0 )1 D00 u0 D (A0 ), we have i
0 = x0 = H(s0 )1 D00 00 Ex0 . i i
es0 d[()])x0 = s0 x0 ,
x0 D (A0 ), x0 = 0.
(ii) The proof is similar that of (i). For > 0, there exists s1 C, Re s1 0 such that 1 G10 (s1 ) i |es1 h | < 1 supsC,Re s 0 G10 (s) i <
0
|esh |
Let u1 Ui1 , u1 = 1 satisfy 0
Theorem, there exists y (Y ) , y = 1 satisfying 1 1
1 G1 (s1 )u1 i
1 G1 (s1 ) i
0
+ 2 |ehs1 |. By the Hahn-Banach
y (G10 (s1 )u1 ) = G10 (s1 )u1 . 1 i i Dene 10 := i It is easy to see that 10 = i 1 G10 (s1 )u1 i
1 G1 (s1 )u1 i
0
u1 y () L (Y,Ui1 ). 1 0
. We now consider the following step function
i0 () =
0 if = h, 10 ehs1 if (h, 0]. i
Robust Stability of Linear Functional Differential Equations It is clear that i0 NBV([h, 0], L (Y,Ui1 )), 0 1 = Var(1 ; h, 0) = 10 |ehs1 |, i

0 h
85
es1 d[1 ()] = 10 . i = 10 Ex1 = u1 . i
Setting x1 := H(s1 )1 D10 u1 D (A0 ), we have i Therefore, 0 = x1 = H(s1 )1 D1 0 This implies (A0 +
0 h
0 s1 h e d[i0 ()]Ex1
es1 d[() + D10 i0 E])x1 = s1 x1 , i
We dene := [, ] = ((0 , ..., N ), (0 , ..., N )) , where i = i0 if i = i0 otherwise i = 0, i N and i = 0 for every i N . Then, we have = 10 |es1 h | < i 1 supsC,Re s 0 G10 (s) |esh | i + and s(A ) Re s1 0.
(iii) If G0 (0) L + (Ui0 ,Y ) and G1 (0) L + (Ui1 ,Y ) for every i N then, by (14), i i = supu(Ui0 )+ , u =1 G0 (0)u , G1 (0) = supu(Ui1 )+ , u =1 G1 (0)u . Thus we can i i i choose u0 (Ui0 )+ , u1 (Ui1 )+ u0 = u1 = 1 such that G0 (0) i G0 (0)u0 = G0 (0) , G1 (0)u1 = G1 (0) . i i i i Since G0 (0)u0 0, G1 (0)u1 0, by the Hahn-Banach theorem for positive operators i i (see e.g. [34] page 249), there exist positive linear forms y (Y ) , y (Y ) of dual 0 1 norms y = y = 1 such that y Gi (0)u0 = Gi (0)u0 , y Gi (0)u1 = Gi (0)u1 . Hence 0 1 0 1 the perturbation constructed as in (i), (ii) (where s0 = s1 = 0) is nonnegative. The proof is complete. Using Lemma 3.6 we obtain the following estimates for the complex radius rC . Theorem 3.7. Let the system (15)-(16) be Hurwitz stable and supsC,Res0 H(s)1 < +. Assume that A0 , are subjected to multi-perturbations of the form (22). Then, we have maxiN { supsC,Res0 G0 (s) i 1 rC }, maxiN { supsC,Res0 G1 (s) } i
max
max
maxiN { supsC,Res0 G0 (s) i
1 . (35) }, maxiN { supsC,Res0 G1 (s) |ehs |} i
Proof. Assume that rC < +. Let := [, ], := (0 , ..., N ), := (0 , ..., N ), i L (Y,Ui0 ), i NBV([h, 0], L (Y,Ui1 )), i N
0 h
es1 d[i0 ()]Ex1 .
x1 D (A0 ), x1 = 0.
86
satisfy (A ) C . By (21), there exists s0 C, Res0 0 such that s0 (A0 + N N 0 E)+ 0 s0 1 i=1 Di i h e d[()+ i=1 Di i ()E] . Since the system (15)-(16) is Hurwitz sta0 s0 ble, H(s0 ) = (s0 IX A0 h e d[()]) has a bounded linear inverse. Since s0 IX (A0 + D0 i E) i
i=1
IX
i=1
D0 i D1 i i
i=1 N N
and since this operator is not invertible, the operator IX
i=1
D0 i D1 i i
i=1
is not invertible. Then, it follows from Lemma 3.5 that the operator IY EH(s0 )1 ( D0 i + D1 i i
i=1 i=1
is not invertible. Hence EH(s0 )1
i=1
D0 i + D1 i i
i=1
Using the inequalities in (4), we get max{max G0 (s0 ) , max G1 (s0 ) } i i

iN iN
Hence max max {

iN
sup
sC,Res0
G0 (s) }, max { i
iN
or equivalently, 1 . max maxiN { supsC,Res0 G0 (s) }, maxiN { supsC,Res0 G1 (s) } i i
By the denition of the complex stability radius rC , we have rC max maxiN { supsC,Res0 G0 (s) i 1 }, maxiN { supsC,Res0 G1 (s) } i
It remains to prove that rC 1 . max maxiN { supsC,Res0 G0 (s) }, maxiN { supsC,Res0 G1 (s) |ehs |} i i
0 h
es0 d[i ()]
i=1
sup
sC,Res0
0 h 0
es0 d[() + D1 i ()E] i

i=1
es0 d[i ()] EH(s0 )1 H(s0 ),
0 h
es0 d[i ()] EH(s0 )1 ,
0 h
es0 d[i ()]
1.
i + i
i=1
1.
G1 (s) } i
1,
(36)
Robust Stability of Linear Functional Differential Equations Fix > 0. It follows from (36) that max max {
iN
87
sup
sC,Res0
G0 (s) }, max { i
iN
sup
sC,Res0
G1 (s) } = 0. i
Taking the Lemma 3.6 (i), (ii) into account, there exists a complex perturbation := [, ] D C such that < and (A ) C . By the denition of the complex stability radius rC , rC < 1 + . max maxiN { supsC,Res0 G0 (s) }, maxiN { supsC,Res0 G1 (s) |ehs |} i i max maxiN { supsC,Res0 G0 (s) i 1 + , }, maxiN { supsC,Res0 G1 (s) |ehs |} i
This implies rC max maxiN { supsC,Res0 G0 (s) i 1 . }, maxiN { supsC,Res0 G1 (s) |ehs |} i
If rC = + (for example, E is zero operator) then by Lemma 3.6, max{

iN
sup
sC,Res0
G0 (s) } = max{ i
iN
sup
sC,Res0
G1 (s) } = 0. i
Then the equalities in (35) hold. This completes our proof. The problem of computation of the real stability radius rR is much more difcult and has been solved rst for the ordinary linear systems of the form x(t) = A0 x(t), t 0, where the system matrix A0 is subjected to single perturbations (i.e. N = 1), see, e.g [24]. Recently, this result has been extended to linear time invariant time delay systems [14]. We note that, by denition, rC rR , so rC can be accepted as the lower bound for rR . Unfortunately, as shown in many previous papers (see, e.g. [9] ) these two stability radii can be arbitrarily distinct. Therefore, it is an interesting problem to nd classes of systems of practical interest for which these two stability radii coincide. Motivated by the results of [11], [12], [28], [29], [30], in this section, we show that for the class of positive functional differential equations, the positive, real and complex stability radius under multi-perturbations coincide and can be computed by a simple formula. We now restrict ourselves to positive systems. Denition 3.8. System (15)-(16) is called positive if its solution semigroup (T (t))t0 is a positive semigroup, that is, T (t) L + (C[h, 0], X) for every t 0. Proposition 3.9. [16] The system (15)-(16) is positive if A0 generates a positive semigroup and L L (C[h, 0], X), X is a positive operator. / Recall that, by denition, a linear operator A0 has a compact resolvent if (A0 ) = 0 and 1 is compact for every s (A ). In order to give characterization of the real (sIX A0 ) 0 stability radius of the positive systems, we need some technical lemmas.
88
Lemma 3.10. [16] Suppose that A0 generates a positive semigroup and L L (C[h, 0], X), X is a positive operator. Then 0 (i) The spectral bound function s(A0 + h e d[()]) is decreasing and continuous from the left on R. (ii) If, additionally, A0 has a compact resolvent and there exists a 0 R with (A0 + 0 0 0 / e d[()])) = 0, then s(A0 + h e d[()]) is continuous on R and the spectral h bound s(A) of the solution operator of the system (15)-(16) is a unique solution of equation = s(A0 +
0 h
e d[()])).
Moreover s(A0 + s(A0 +
0 h
e d[()]) < s(A) < , e d[()]) > s(A) > .
0 h
In the rest of this paper, we suppose that Assumption 3.11. A0 generates a positive compact C0 -semigroup (U(t))t0 of bounded / linear operators on X, L L (C[h, 0], X , X is a positive operator and (A) = 0. Remark 3.12. As noted in the beginning of this section, if A0 generates a C0 -compact semigroup then the solution semigroup (T (t))t0 of the system (15)-(16) is eventually compact. Hence, the concepts of stability in Denition 3.3 are actually the same. Furthermore, if Assumption 3.11 is fullled then it is easy to see that all conditions of Lemma 3.10(ii) is satised.
0 Lemma 3.13. Let H(s) = sIX A0 h es d[()]. Then i) H(t1 )1 H(t2 )1 0 f or t2 > t1 > s(A).
(ii) Let s = + C, > s(A). Then |H(s)1 x| H()1 |x|, x X. (38)
Proof. (i) It follows from L L (C[h, 0], X), X being a positive operator and the equality
0 h
et d[()])x =
0 h
(et x)d[()], x X,
0 that h et d[()] L + (X) for every t R. Since A0 is a generator of a positive C0 0 0 semigroup on X and h et d[()] L + (X), A0 + h et d[()] also generates a positive 0 C0 -semigroup on X. By Lemma 3.10 (ii), if t > s(A) then t > s(A0 + h et d[()]). From Theorem 1.1 (C-III), page 292 [16], we have 0 h
(tIX A0
et d[()])1 = H(t)1 0
for every t > s(A).
(37)
(39)
Robust Stability of Linear Functional Differential Equations It is easy to verify that the following resolvent equation holds H(t1 )1 H(t2 )1 = (t2 t1 )H(t1 )1 H(t2 )1 + H(t1 )1 It follows from (39) that H(t1 )1 H(t2 )1 0 f or t2 > t1 > s(A).
0 h
89
(et1 et2 )d[()]H(t2 )1 .
(ii) Let A0 be the generator of the positive semigroup (U(t))t0 on X. Set B(s) := 0 s h e d[()], s C. It is well-known that A0 + B(s) with domain D (A0 + B(s)) = D (A0 ) is the generator of a C0 -semigroup (Vs (t))t0 satisfying

t t n Vs (t)x = lim U( )e n B(s) x, n n see e.g [16], page 44. It is easy to see that
0 h
for t 0, x X,
(40)
|(
es d[()])x| (
where s C, Res > s(A). Using the above inequality, we could check that 0 (A0 + 0 s 0 s(A) d[()]), s C, Res > s(A), where 0 () is growth h e d[()]) 0 (A0 + h e bound of a C0 -semigroup with the generator (). On the other hand, since A0 is a generator 0 0 of a compact C0 -semigroup, so do (A0 + h es d[()]), (A0 + h es(A) d[()]), see, e.g. [22], page 79. It follows from the spectral mapping theorem that 0 (A0 +
0 h 0 h
es d[()]) = s(A0 +
0 h
es d[()]),
0 (A0 +
es(A) d[()]) = s(A0 +
0 h
es(A) d[()]).
0 From (21), s = ( + ) (A0 + h es d[()]), for every s C, Re s > s(A). By 0 0 Lemma 3.10 (ii) and (42), s(A) = s(A0 + h es(A) d[()]) = 0 (A0 + h es(A) d[()]). 0 s(A) Then, we derive that if Re s = > s(A) then Re s > 0 (A0 + h e d[()]) 0 (A0 + 0 s h e d[()]). So, we can represent the following
H(s)1 x = (sIX A0
0 h
es d[()])1 x =
est Vs (t)xdt.
From the formulas (40), (43), it is easy to see that |H(s)1 x| et |Vs (t)||x|dt
0 0
et V (t)|x|dt = H()1 |x|.
0 h
es(A) d[()])|x|,
x X,
(41)
(42)
(43)
90
Remark 3.14. It follows from the above lemma that if the system (15)-(16) is Hurwitz stable and Assumption 3.11 is fullled then supsC,Res0 H(s) < +. In fact, by (37) and (38), we have |H(s)1 x| H(Re s)1 |x| H(0)1 |x|, x X, for every s C, Re s 0. From the lattice norm property (9) and the denition of the operator norm, we get H(s)1 H(0)1 , for every s C, Res 0. We are now in the position to prove the main result of this section. Theorem 3.15. Assume that the linear retarded system (15)-(16) is Hurwitz stable and A0 , are subjected to multi-perturbations of the form (22) where D0 L + (Ui0 , X), D1 i i L + (Ui1 , X) (i N ) and E L + (X,Y ). Then, we have rC = rR = r+ = 1 max{maxiN G0 (0) i , maxiN G1 (0) } i . (44)
Proof. Suppose that rC < +, as otherwise there is nothing to show. We prove that rC = 1 max{maxiN G0 (0) i , maxiN G1 (0) } i . (45)
It follows from (37), (38) and D0 L + (Ui0 , X), D1 L + (Ui1 , X) (i N ) that i i |G0 ()x| = |(EH()1 D0 )x| (EH(Re)1 D0 )|x| (EH(0)1 D0 )|x|, i i i i |G1 ()y| = |(EH()1 D1 )y| (EH(Re)1 D0 )|y| (EH(0)1 D1 )|y|, i i i i x Ui0 , i N , y Ui1 , i N ,
for every C, Re 0. From the lattice norm property (9) and the denition of the operator norm, we have G1 () G1 (0) i i and G1 () G1 (0) , i N , i i (46)
for every C, Re 0. Then (45) now follows from (35) and (46). On the other hand, by D0 L + (Ui0 , X), D1 L + (Ui1 , X) (i N ) and (37), we have i i G0 (0) L + (Ui0 ,Y ), G1 (0) L + (Ui1 ,Y ), i N . i i It follows from Lemma 3.6 and the denition of r+ that r+ 1 max{maxiN G0 (0) i , maxiN G1 (0) } i . (47)
Finally, (44) follows from (45), (47) and the inequalities rC rR r+ . This completes our proof. We illustrate the above result by an example.
Robust Stability of Linear Functional Differential Equations Example 3.16. Consider the equation
u(t,x) t
91
2 u(t,x) 2 x
d(x)u(t, x) + b(x)u(t 1, x)
(t 0, x [0, 1])
(48)
with boundary condition

u(t,x) x |x=0
=0=
u(t,x) x |x=1
(t 0)
(49)
and initial condition u(s, x) = (s, x), s [1, 0], x [0, 1].
Let X be the function space C[0, 1] equipped with sup-norm, and let B be dened by Bh = h , with domain D(B) := {h C2 [0, 1] : h (0) = h (1) = 0}. Denote by Mb , Md the respective multiplication operators for b, d X, b, d 0. Then (48)-(49) takes the abstract time-delay system of the form u(t) = Bu(t) Md u(t) + Mb u(t 1), u0 = C([1, 0] [0, 1]). t 0, (50)
It is well-known that B generates a positive compact semigroup, see [16], page 230. The same is true for the operator B Md , see [22], page 79. Moreover, if maxx[0,1] (b(x) d(x)) < 0 then the delay system (50) is uniformly exponentially stable, see [16], page 230. This is equivalent to the uniform exponential stability of the solutions of (48)-(49). We now assume that the functions b, d are subjected to perturbations of the form b(x) b(x) + 0 (x), d(x) d(x) + 1 (x), 0 , 1 X.
From Theorem 3.15, the solutions of the following perturbed equation (d(x) + 1 (x))u(t, x) + (b(x) + 0 (x))u(t 1, x) (t 0, x [0, 1]) (51) with boundary condition =
u(t,x) x |x=0 u(t,x) t 2 u(t,x) 2 x
=0=
u(t,x) x |x=1
(t 0)
(52)
and initial condition u(s, x) = (s, x), are uniformly exponentially stable if max |0 (x)| + max |1 (x)| <
x[0,1]
s [1, 0], x [0, 1],
x[0,1]
1 . (B + Mb Md )1
92
Stability Radii of Positive Linear Functional Differential Equations under Multi-afne Perturbations
In this section, we study the robust stability of the positive system (15)-(16) where A0 , are subjected to multi-afne perturbations of the following form A0 A0 + N0 i Ai i=1 (53) + N1 j j . j=1 Here Ai L (X), i N 0 := {1, 2, ..., N0 }; j NBV ([h, 0], L R (X)), j N 1 := {1, 2, ..., N1 } are given and i , j , i N 0 , j N 1 are unknown scalar parameters. We note that the class of multi-afne perturbations of this type has been considered rst in [11], [12] for positive linear systems with no time delays in Rn and then in [13], [29] for positive time-delay systems in Rn . The problem of computing the stability radius of positive linear functional differential equations in Rn under multi-afne perturbations has just been done recently in [21]. In this section, we extend these results to general functional differential equations in Banach lattices. We assume that Assumption 3.11 holds. Denote by A((i )iN 0 ,( j ) jN 1 ) solution operator of perturbed systems with respect to multi-afne perturbation (53). In view of positivity of the system under consideration, we assume that Ai L + (X), i N 0 and j NBV ([h, 0], L R (X)) are increasing, for all j N 1 . We dene the complex and the real stability radii of the linear retarded system (15)-(16) under multi-afne parameter perturbations (53) by setting, for K = C and, respectively, K = R,
a rK = inf{max(max |i |; max | j |) : i , j K, i N 0 , j N 1 ; A((i )iN0 ,( j ) jN 1 ) C }. iN 0 jN 1
(54) a Similarly, the positive stability radius r+ is obtained by restricting, in the above denition, the disturbances (, ) := ((i )iN 0 , ( j ) jN 1 ) to be nonnegative. We emphasize that by (21), A((i )iN 0 ,( j ) jN 1 ) can be replaced by (A0 + N0 i Ai , + i=1
N1 j=1 j j ) in the formula (54). It is clear that a a a rC rR r+ .
(55)
The following theorem shows that the equalities in (55) hold and gives us a computable a a formula of stability radii rK , r+ , (K = C, R). Theorem 4.1. Suppose the positive linear retarded system (15)-(16) is Hurwitz stable and A0 , are subjected to multi-afne perturbations of the form (53). If the stability radii of the system are given by (54) then
a a a rC = rR = r+ =
s (A0
N0 0 0 N1 1 h d[()]) (i=1 Ai + j=1 h d[ j ()])
(56)
Proof. We rst show that

a r+ =
s (A0
N0 0 0 N1 1 h d[()]) (i=1 Ai + j=1 h d[ j ()])
(57)
93
a Assume that r+ < +. Let (, ) = ((i )iN 0 , ( j ) jN 1 ) be an arbitrary nonnegative destabilizing perturbation. That is
(A0 + i Ai , + j j ) C .
i=1 j=1
N0
N1
Therefore, there exists s0 C, Re s0 0 such that s0 (A0 + N0 i Ai + i=1

0 s0 h e d[( N
+ N1 j j )()]). This is equivalent to 1 H(s0 )1 N0 i Ai + j=1 i=1 . By Lemma 3.13 (ii), we have
|H(s0 )1
N0
i=1
i A i + j
j=1
H(Re s0 )1 |
i=1
i A i + j
j=1
Since Ai L + (X), i N 0 ; j NBV ([h, 0], L R (X)), j N 1 being increasing, using Lemma 3.13 (i), we derive max(max |i |; max | j |) H(0)1
iN 0 jN 1 N0
i=1 N1
Ai +
0 h
j=1 h
H(Re s0 )1 |
i=1
i A i + j
j=1 N1
es0 d[ j ()] x| es0 d[ j ()] x|,
|H(s0 )1
i=1
i A i + | j
j=1
operator, using the Gelfands formula for spectral radius r() of a bounded linear operator () and the lattice norm property (9), we get
N0 N1
r H(0)1
i=1
Ai +
j=1 h
d[ j ()]
max(maxiN 0 |i |; max jN 1 | j |)
0 s0 h e d[ j ()]
r H(0)1
i=1
Ai +
j=1 h
N0
N1
d[ j ()]
1 . max(maxiN 0 |i |; max jN 1 | j |)
It follows from 1 H(s0 )1 N0 i Ai + N1 j i=1 j=1
r H(s0 )1 N0 i Ai + N1 j i=1 j=1
for every x X. Since H(0)1 N0 Ai + N1 i=1 j=1
0 h d[ j ()]
N0
0 h
N1
0 h
es0 d[ j ()] x|,
N0
1 j=1 j
0 s0 h e d[ j ()]
N0
N1
0 h
es0 d[ j ()] x|
for every x X.
N1
d[ j ()]
|x|
is a positive bounded linear
0 s0 h e d[ j ()]
, that
(58)
94
On the other hand, by Perron-Frobenius Theorem for positive bounded linear operator (see. e.g [15]), s H(0)1 r H(0)1
i=1
Ai +
j=1 h
i=1
Ai +
j=1 h
d[ j ()]
H(0)1
i=1
Ai +
j=1 h
This yields, by virtue of (58), max(max i ; max j )

iN 0 jN 1
s H(0)1 N0 Ai + N1 i=1 j=1
0 h d[ j ()]
a By the denition of the positive stability radius r+ , a r+
H(0)1
N0 0 N1 i=1 Ai + j=1 h d[ j ()]
using again Perron-Frobenius Theorem for positive bounded linear operator, we have H(0)1 This implies
N0 1 0 (A0 + Ai ), i=1 N1 1 d[ + j ()] . h j=1 0
i=1
Ai +
j=1 h
It means that nonnegative perturbation ( , ) dened by = 1/, i N 0 , = j i a 1/, j N 1 , is destabilizing. By the denition of r+ , we have
a r+
s H(0)1 (N0 Ai + N1 i=1 j=1
Thus, we obtain
a r+ =
s H(0)1 (N0 Ai + N1 i=1 j=1
a It is important to note that by the above argument, if r+ = +, then
s H(0)1 ( Ai +
i=1
j=1 h
So the equality (57) is obvious.
N0
N1
d[ j ()])
N0
Setting := s H(0)1 N0 Ai + N1 i=1 j=1
0 h d[ j ()]
. To prove the inverse inequality,
N1
d[ j ()]
0 h d[ j ()])
0 h d[ j ()])
= 0.
N0
N1
N0
N1
d[ j ()]
N0
=
N1 0
d[ j ()]
95
a a a a We are now ready to show that rC = rR = r+ . Suppose rC < + and (, ) = ((i )iN 0 , ( j ) jN 1 ) is an arbitrary complex destabilizing perturbation. By a similar argument as in the beginning, we have
1 := s H(0)1 Since
i=1
|i |Ai + | j |
j=1
it follows from Perron-Frobenius Theorem that 1 H(0)1 It follows that 0 (A0 + |i | Ai ), i=1 1
N0 0 h
i=1
|i |Ai + | j |
j=1
d[ +
| j | j ()] . j=1 1
N1
It means that nonnegative perturbation ( , ) dened by i = |i |/1 , i N 0 , j = a a |i |/1 , j N 1 , is destabilizing. By the denition of rC , r+ , we conclude
a a r+ rC . a a a a a a In combining with the inequalities rC rR r+ , this implies rC = rR = r+ . This completes our proof.
References
[1] A. Fischer, Stability radii of innite-dimensional positive systems, Math. Control Signals Systems 10 (1997) 223-236. [2] A. Fischer, D. Hinrichsen and N.K. Son, Stability radii of Metzler operators, Vietnam Journal of Math. 26 (1998) 147-163. [3] A. Fischer and J.M.A.M. van Neerven, Robust stability of semi-groups and apllication to stability of delay systems, Journal of Mathematical Analysis and Applications 226 (1998) 82-100. [4] J. Hale, Theory of Functional Differential Equations, Acad. Press, New York, 1977. [5] E. Hille and R. S. Phillips, Functional Analysis and Semigroups, Providence, RI: American Mathematical Society, 1957. [6] Y. Hino, S. Murakami, T. Naito, Nguyen Van Minh, A variation of constants formula for abstract functional differential equations in the phase space, J. Differential Equations 179 (2002) 336-355.
N0
H(0)1 N0 |i |Ai + N1 | j | i=1 j=1
0 h d[ j ()]
N1
N0
N1
0 h
d[ j ()]
1.
is a positive bounded operator,
0 h
d[ j ()]
96
[7] Y. Hino, S. Murakami, T. Naito, Functional Differential Equations with Innite Delay, Lecture Notes in Math. 1473, Springer-Verlag, Berlin-New York, 1991. [8] D. Hinrichsen and A.J. Pritchard, Stability radius for structured perturbations and the algebraic Riccati equation, Systems & Control Letters 8 (1986) 105-113. [9] D. Hinrichsen and A.J. Pritchard, Real and complex stability radii: a survey, in : D. Hinrichsen and B. M rtensson, editors, Control of Uncertain Systems, volume 6 of a Progress in System and Control Theory, Basel. Birkh user, 1990, 119-162. a [10] D. Hinrichsen and A.J. Pritchard, Robust stability of linear evolution operators on Banach spaces, SIAM Journal on Control Optimization 32 (1994) 1503-1541. [11] D. Hinrichsen and N.K. Son, - analysis and robust stability of positive linear systems, Appl. Math. and Comp. Sci. 8 (1998) 253-268. [12] D. Hinrichsen and N.K. Son, Stability radii of positive discrete-time systems under afne parameter perturbations, Inter. J. Robust and Nonlinear Control 8 (1998) 1169-1188. [13] D. Hinrichsen, N.K. Son and P.H.A. Ngoc, Stability radii of positive higher order difference systems, Systems & Control Letters 49 (2003) 377-388. [14] G. Hu and E. J. Davison, Real stability radii of linear time-invariant time-delay systems, Systems & Control Letters 50 (2003) 209-219. [15] P. Meyer-Nieberg, Banach Lattices, Springer-Verlag, Berlin, 1991. [16] R. Nagel (Ed.), One-Parameter Semigroups of Positive Operators, Springer-Verlag, Berlin, 1986. [17] T. Naito, J.S. Shin, S. Murakami, On solution semigroups of general functional differential equations, Nonlinear Anal. 30 (1997) 45654576. [18] T. Naito, J.S. Shin, S. Murakami, The generator of the solution semigroup for the general linear functional differential equation, Bull. Univ. of Electro-Communications 11 (1998) 2938. [19] P.H.A. Ngoc and N.K. Son, Stability radii of positive linear difference equations under afne parameter perturbations, Applied Mathematics and Computation 134 (2003) 577-594. [20] P.H.A. Ngoc, N.K. Son, Stability radii of linear systems under multi-perturbations, Numer. Funct. Anal. Optim. 25 (2004) 221-238. [21] P.H.A. Ngoc and N.K. Son, Stability radii of positive linear functional differential equations under multi-perturbations, To appear in SIAM Journal on Control and Optimization, (2004). [22] A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations, Springer-Verlag, Berlin, 1983.
Robust Stability of Linear Functional Differential Equations [23] A.J. Pritchard and S. Townley, Robustness of linear systems, Equations 77 (1989) 254-286.
97
Journal Differential
[24] L. Qiu, B. Bernhardsson, A. Rantzer, E.J. Davison, P.M. Young and J.C. Doyle, A formula for computation of the real stability radius, Automatica 31 (1995) 879-890. [25] J. S. Shin, T. Naito, Nguyen Van Minh, On stability of solutions in linear autonomous functional differential equations, Funkcial. Ekvac. 43 (2000) 323337. [26] N.K. Son and D. Hinrichsen, Robust stability of positive continuous-time systems, Numer. Funct. Anal. Optim. 17 (1996) 649-659. [27] N.K. Son and P.H.A. Ngoc, Complex stability radius of linear time-delay systems, Vietnam Journal of Mathematics 26 (1998) 379-383. [28] N.K. Son, P.H.A. Ngoc, Stability radius of linear delay systems, in Proceedings of the American Control Conference, San Diego, California (1999) 815-816. [29] N.K. Son, P.H.A. Ngoc, Robust stability of positive linear time delay systems under afne parameter perturbations, Acta Mathematica Vietnamica 24 (1999) 353-372. [30] N.K. Son, P.H.A. Ngoc, Robust stability of linear functional differential equations, Advanced Studies in Contemporary Mathematics 3, No.2 (2001) 43-59. [31] S. Clark, Y. Latushkin, S. Montgomery-Smith, T. Randolph, Stability radius and internal versus external stability in Banach spaces: an evolution semigroup approach, SIAM J. Control Optim. 38 (2000) 17571793. [32] C.C. Travis, G.F. Webb, Existence and stability for partial functional differential equations, Trans. Amer. Math. Soc. 200 (1974) 394-418. [33] J. Wu, Theory and Applications of Partial Functional Differential Equations, Applied Math. Sci. 119, Springer, Berlin-New York, 1996. [34] A.C. Zaanen, Introduction to Operator Theory in Riesz Spaces, Springer-Verlag, Berlin, 1997.
E FFECTS OF N ONLOCALITY AND P HASE S HIFT D EFINITIONS IN G ENERALIZING L EVINSON S T HEOREM

S.B. Qadri and B. Mulligan Department of Physics, The Ohio State University Columbus, Ohio 43210 USA M.F. Mahmood and J.Y. Al-Khal Department of Mathematics, Howard University Washington, DC 20059, USA
Abstract This paper considers Levinsons Theorem for nonlocal potentials. The effect of continuum bound states is claried, and the presence of spurious states is introduced. Differences between denitions of the phase shift are discussed, with special attention given to discontinuities in the phase shift. Examples from nuclear physics emphasize the utility of this theorem and illustrate some of the more delicate aspects of the derivation. A summary is provided which analyzes certain crucial points in the discussion as they relate to previous attempts to obtain a generalized Levinsons theorem. NUCLEAR REACTIONS Levinsons Theorem, nonlocal potentials continuum bound states, spurious states, Jost functions, Fredholm determinants, denition of the phase shift.
Introduction
In this paper, we consider scattering interactions which can be described by a spinindependent, central potential, and focus on the connection between the zero-energy phase shift and the phase shift as the energy of the projectile becomes arbitrarily large. For simplicity, we restrict discussion to the = 0 partial wave. When the scattering potential V is local, a well-known relation exists between the phase shift and the number of bound states of the scattering center. Assuming that V satises the restrictions r |V (r)| dr < and r 2 |V (r)| dr < , (1)
0 0
Levinson [1] showed that phase shift (E) obeys the condition (0) () = n,
Naval
(2)
Reasearch Laboratory, Washington, DC 20375 USA
Corresponding Author
100
S.B. Qadri, B. Mulligan, M.F. Mahmood et al.
where n is the number of bound states of V . This relation, which has come to be known as Levinsons theorem, was rst generalized by Swan [2] to include a situation in which the interaction is described by a nonlocal potential. Such a circumstance occurs when the incident particle is identical to or contains particles identical to those of the target, as with collisions of nucleons with light nuclei or electrons with atoms. Under these conditions Swan observed that, in addition to bound states, states excluded due to the Pauli principle also contribute to the zero-energy phase shift. This results in values of n in Eq. (2) not equal to the number of bound states. Specically, Swan showed that the modied version of Levinsons theorem for the case of potentials with nonlocality which arises from antisymmetrization is (0) () = (n + m) (3)
where n is (as before) the number of bound states of the target and incident particle and m is the number of states from which the incident particle is excluded by the Pauli principle. Temkin [3] and Rosenberg and Spruch [4] have also investigated Levinsons theorem in the presence of antisymmetrization, and agree with Swans conclusions. Subsequently, Jauch [5] extended Levinsons theorem to a large class of potentials for general scattering systems, and Gourdin and Martin [6] discussed the theorem in the context of nonlocal potentials as a class. Jauchs generalization was made under the assumption that a nonlocal potential has no bound states in the continuum. Although this assumption is strictly true for a short range local potential [7], it cannot be made for nonlocal interactions. For example, Martin [8] examined the phase shift due to a nonlocal, separable, central potential and concluded that bound states in the continuum are indeed possible and that one must take such states into account in Levtnsons theorem. In particular, he showed that (0) ()is equal to times the number of bound states plus the number of continuum bound states. In this respect, Martins results are consistent with the earlier conclusions of Swan. In discussing states excluded by the Pauli principle, Swan noted that these states appear as solutions of the scattering integral equation at all scattering energies. Such redundant states are simply continuum bound states which appear in the spectrum of the scattering solution at every positive energy [9], and can be expected in any scattering calculations which take into account antisymmetrization between the incident particle and a target described by a antisyrnmetrized single-particle state. They are a consequence of the fact that any row of a Slater determinant can be added to any other row of the determinant without changing the value of the determinant. Thus, the scattering solution is nonunique because any arbitrary amount of lled states can be added to the scattering state. Additional constraints must be placed on the scattering solution to obtain wavefunction orthogonal to the lled states, as dictated by physical considerations. Many of the problems associated with the presence of redundant states have been extensively studied and discussed [9-15]. In particular, it has been pointed out that the presence of redundant states is possible only in the case of nonlocal potentials. Also, since the early work, Levinsons theorem has been subjected to repeated examination and extension. For example, Ida [16] has discussed its importance with respect to resolution of nonuniqueness of solutions of the Low equation. Levinsons theorem in the multichannel case has been investigated by a number of authors [17-28].
Effects of Nonlocality and Phase Shift Denitions...
101
Martins result for Levinsons theorem has been replicated by Bertero, et al. [29]. Their derivation was obtained for a large class of nonlocal interactions. On the other hand, Buslaev [30] proved that in spite of the presence of continuum bound states Levinsons theorem holds in its original form for a large class of potentials which is similar to that considered by Bertero, et al. That this apparent discrepancy is due to the denition of the phase shift has been shown by Dreyfus [31]. His discussion follows similar observations by Bolsterli [32] and by Beregi [33], who pointed out that there is an ambiguity in the denition of the phase shift when a continuum bound state is present. Additional work on the generalization of Levinsons theorem to nonlocal potentials also has been carried out by Mills and Reading [34], Horwitz and Marchand [35], Gl kle and LeTourneux [36], W llenberg [37], and o o Newton [38]. A review paper on Levinsons theorem was published by Beregi, et al. [39]in 1973. If for a non-singular local potential the initial conditions imposed on the radial part of the wavefunction u(r) were u(0) = u (0) = 0, the solution would be zero everywhere. In the case of a nonlocal potential, however, it has been shown [40,41] that a nontrivial solution with u(0) = u (0) = 0 can exist. This anomalous type of solution [2,42,43] found with nonlocal potentials is not due to a continuum bound state. A (non-normalizable) state whose wavefunction satises the condition u(0) = u (0) = 0 has been named a spurious state. As is the case with a continuum bound state, a spurious state also results in an ambiguity in the denition of the phase shift at the energy of the spurious state. In each of the generalizations of Levinsons theorem for nonlocal potentials discussed above, there are important aspects which the authors have not considered. The purpose of the present work is to examine Levinsons theorem in detail and to look into these aspects with the purpose of clarication. All our discussions will be limited to nonlocal potentials which are symmetric, and for which a thorough analysis of solutions of the radial equation has been presented [40]. The extension of the results of Ref. [40] to nonsymmetric nonlocal potentials is rather complicated, and has just been begun [44]. The focus here has been to avoid these complications, and rather to concentrate on fundamental questions of Levinsons theorem as it relates to such basic features of non-locality as the presence of continuum bound states and spurious states.
Method of Approach
The procedure which we use here for discussing Levinsons theorem follows the approach of Swan [2]. As pointed out in Eq. (2), for a short range local potential (0) () is equal to times the number of bound states of the potential. Swan noted that in the complex kplane the Jost function L +(k)for a local potential has a zero on the positive imaginary axis associated with each bound state. Furthermore, for a local potential L +(k)can be shown to be analytic in the upper half plane, and to have no zeros in addition to those associated with the bound states. On this basis, it is possible to use a well-known theorem from the theory of complex variables to form a contour integral which will have as its value times the number of bounds states. This theorem concerns a function f (z) which is analytic inside and on a simple closed contour C, except for a nite number of poles inside C, and for which f (z) = 0 on C.If N
102
and P are, respectively, the number of zeros and poles inside C , then 1 2i f (z) dz = ln f |C = N P. f (z) (4)
applying this theorem to the function L + (k) using the contour C indicated in Fig. 1 yields
L + (k) dz = 2in. + C L (k)
(5)
Figure 1. Contour C for integration of in the case of a local potential. Zeros of L +(k)are indicated by the symbol o. For a local potential it can also be demonstrated that
L + (k) =| L +(k) | ei(k) .

Carrying out explicitly the line integral in Eq. (5) gives
(6)
L + (k) dz = 2i[(0) ()], + C L (k)
(7)
where we have used the fact that (k) = (k). Combining Eqs. (5) and (7) immediately yields Eq. (2). In obtaining the result in Eq. (5), the integral around the semicircular contour in Fig. 1 has been assumed to go to zero. However, if L +(0) = 0 the explicit integral in Eq. (7) around the portion of the path C will yield i. In that case, the right-hand side of Eq. (7) is replaced by 2i[(0) ()] ip. Then, instead of Eq. (2) we obtain the well-known result 1 (8) (0) () = (n + ). 2
Extension to Nonlocal Potentials
In order to adapt this approach to a discussion of Levinsons theorem for nonlocal potentials it is necessary to investigate in detail the Jost function for a nonlocal potential. Jost and
103
Pais [45] have shown that, for a local potential, instead of arriving at the properties of the potential by starting with the Jost function L +(k), one could equally well have started with the integral equation for the physical solution and obtained the properties of the potential from the Fredholm determinant associated with the kernel of that equation. As one introduces the concept of nonlocality, this shift of the focus of discussion toward integral equations and the Fredholm determinants related to the kernels of these integral equations becomes very useful. When the potential is nonlocal, the = 0 radial equation is of the form d2 + k2 u(r) = V (r, r ) u(r ) dr . (9) dr 2 0 This integro-differential equation can be converted into integral equations by the use of Greens functions which satisfy appropriate boundary conditions. Several integral equations, their solutions, and related quantities are dened in the following paragraph, in analogy with the local potential case. The regular solution (k, r) is dened by the boundary conditions (k, 0) = 0; (k, 0) = 1. (10).
The integral equation for the regular solution (k, r) is (k, r) = k1 sin kr +
r 0 0
G(k, r, r )V (r , s) (k, s) ds dr ,
(11)
where G(k, r, r ) is the Greens function and is equal to k1 sin k(r r ). The Fredholm determinant associated with the kernel of this integral equation is denoted by D(k). The physical solution + (k, r) is dened by the mixed boundary conditions that + (k, r) have the asymptotic form i (12) + [eikr S+ (k)eikr ] 2 as r , and that + (k, r) be regular at r = 0. S+ (k) is the s-wave scattering matrix element. The physical solution + (k, r) and its conjugate (k, r) satisfy the integral equations (k, r) = sin kr +
0
G (k, r, r )V (r , s) (k, s) ds dr ,
0
(13)
where G(k, r) = k1 eikr> sin kr< with r> = max(r, r ) and r< = min(r, r ). The asymptotic form of (k, r) for large r is given by Eq. (12) with S+ replaced by S (k) = [S+ (k)] where represents complex conjugation. The Fredholm determinants associated with the kernels of Eq. (13) are denoted by D(k). The integral equations for the Jost solutions are f (k, r) = eikr
r 0
G(k, r, r )V (r , s) f (k, s) ds dr ,
(14)
The Fredholm determinant associated with the kernel of Eq. (14) is denoted by (k). The Fredholm determinants have the following properties: D (k), D(k), and (k) are even functions of k, while D (k) is an odd function of k. Both D(k) and (k) are real for real k, and D(k) = (k) for symmetric nonlocal potentials [46].
104
For a local potential, the reason that one can start either from a Jost function point of view or from a physical integral equation Fredholm determinant point of view lies in the fact [45] that for a local potential the Jost functions L (k) and the Fredholm determinants D (k) are identically equal,
L (k) = D (k)
[ local potential ].
(15)
Also, with the exception of Eq. (13), for a local potential the integral equations become Volterra equations. Thus, the Fredholm determinants D(k) and (k) become unity in the case of a local potential and can play no explicit role in the description of the scattering process. The Fredholm determinants D (k) are not unity in the case of a local potential, consistent with the fact that they are the only Fredholm determinants needed to establish all of the scattering information. For a nonlocal potential, the integral equations with which D(k) and (k) are associated are Fredholm integral equations rather than Volterra integral equations. Thus, the Fredholm determinants D(k) and (k) are not unity, and it might be anticipated that a complete description of the scattering process would require these determinants in addition to the determinants D (k). Nevertheless, zeros of the Fredholm determinants D (k) and D(k) are not independent. Whenever there is a zero of D (k) for real k there is also associated a zero of D(k) at that same value of k [40]. A zero of D (k) for real k = 0 is called a continuum bound state [6,8], often referred to by the abbreviation CBS. The existence of a continuum bound state results from the fact that when D (k) = 0 for real k = 0 solutions of the homogeneous integral equations associated with Eq.(l3), namely

(k, r) = h
0 0
G (k, r, r )V (r , s) (k, s) ds dr , h
(13)
are normalizable. Trivial solutions (k, r) = 0 are the only solutions allowed when h D (k) = 0. A spurious state [40,43] occurs when D(k) is zero and D (k) is nonzero for a real value of k. It is demonstrated in Ref. 40 that at a spurious state the regular solution (k, r) does not exist, although it is possible to obtain a real solution of Eq. (9) which is regular at the origin. In the case of a local potential, it is customary [47] to dene the Jost functions L (k) in terms of the Jost solutions at r = 0: that is
L (k) = f (k, r)|r=0 .
(17)
No ambiguity exists in showing that this denition leads to Eq. (15). Thus Eq. (15) could equally well be used as a denition of L (k) for a local potential. For a nonlocal potential, difculties can arise in using Eq. (17) as the denition of L (k). The Jost solutions f (k, r) do not exist at a spurious state. At a continuum bound state, f (k, r) may or may not exist. Under circumstances in which there are no difculties with the existence of f (k, r), that is, at values of k for which D(k) = 0, for a symmetric nonlocal potential it has been shown [48-51] that D(k) and L (k) are related by the
Effects of Nonlocality and Phase Shift Denitions... expression
105
L (k) = D(k)/D(k).
[ symmetric nonlocal potential ]
(18)
However, the derivation of Eq. (18) breaks down when f (k, r) do not exist. Correspondingly, the denition of the Jost functions as given in Eq. (17) will no longer hold if f (k, r) do not exist. This is especially a problem in the case of redundant states, since it would mean a failure of the existence of the Jost functions at any energy. However, it has been shown [52], using the Greens function cancellation method of Krause and Mulligan [53] to construct a potential which produces a redundant state, that under these circumstances Eq. (18) can nevertheless be used as a denition of L + (k). In addition, the results in Ref. [52] demonstrate that the Jost functions L (k) can be directly related to the kernels of integral equations by means of Fredholm determinants without referring to the solutions of the integral equations involved. Consequently, in dening the Jost functions, the question of the existence of Jost solutions at a CBS or in the presence of a redundant state can be avoided. As we shall see, by taking the denition of the Jost functions as the ratio of Fredholm determinants D(k) to D(k) and proceeding with the derivation of Levinson s theorem, we run into no problems. There are, nevertheless, other difculties associated with the derivation of Levinsons theorem in the case of a nonlocal potential. Consider Eqs. (l5) and (18). The expression for L (k) for a local potential is a special case of the expression for L (k) for a nonlocal potential. Because of the presence of D(k) in Eq.(18), when considering the denition of the phase shift for a nonlocal potential an ambiguity arises. This ambiguity has been discussed in detail in Ref. [41]. The presence of the Fredholm determinant D(k) results in two possible denitions of the phase shift: D = -phase of D+(k) and L =-phase of L +(k). The phase shift L gives a continuous phase at a CBS and discontinuity of in the phase shift as the energy increases through each spurious state. The phase shift D is continuous at a spurious state, but gives a discontinuity of with increasing energy to the phase shift at a CBS. Since for a local potential the phase of L (k) is the same as the phase of D(k), the phase shifts L and D are identical in that case. For a nonlocal potential, a derivation of Levinsons theorem can proceed in much the same way as that described in Sec. 2 in the case of a local potential. However, because of the two possible denitions of the phase shift, there will be two possible results for Levinsons theorem. In one case, the function f (z) to be employed in Eq. (4) will be the Jost function L +(k) dened in Eq. (l8); in the other case one would use D+ (k).
Properties of D+ (k), D(k), and L + (k) for a Nonlocal Potential
For a local potential, it is known that L + (k) and therefore D+ (k) have the property L +(k) D+ (k) = 0 for real k = 0. As we have already noted, in the case of a local potential the Fredholm determinant D+ (k) cannot have zeros on the real k-axis. Also, we will see that for a nonlocal potential zeros of D(k) [and thus poles of the Jost function L +(k)] are found on and above the real axis in some cases. For these reasons, in the case of a nonlocal potential the techniques from complex analysis [54] used to obtain the character of the Fredholm determinant D+ (k) and the Jost function L + (k) for a local potential are not sufcient for a general discussion of D+(k), D(k), and L +(k). However, we have demonstrated [55] that
106
techniques from functional analysis can be used for a wide class of nonlocal potentials to obtain the analytic properties of D+(k), D(k), and L +(k). These techniques were developed by Iwasaki and Mulligan [56], who focused on the fact that the kernels of integral Eqs. (11), (13), and (14) are not square integrable. Noting that the usual derivations of the Fredholm alternative [57] do not apply in cases where kernels are not square integrable, they make use of Riesz-Schauder theory [58], which demonstrates that the Fredholm alternative easily can be extended to integral equations for which the operators are compact. The conditions on the nonlocal potential V (r, r ) necessary for compactness of the kernels of (11) and (13), are given in Ref. [55] or Ref. [56]. Complete continuity of the kernels of the integral Eqs. (11) and (13) is crucial to establishing the existence of the Fredholm determinants D+ (k) and D(k) for a nonlocal potential. It is shown in Ref. [55] that these kernels are compact for k real, from which it follows that the Fredholm determinants D+ (k) and D(k) exist for k real. Fredholm determinants can then be dened for complex k by extending these denitions, established on the real axis, to the complex plane. The properties of these determinants, established in Ref. [55], are summarized below. The character of D+ (k) and D(k) in the upper half plane are such that D+ (k) when so extended will be analytic in that portion of the k plane, while D(k) may not. Thus with nonlocal potentials, as with local potentials, poles of D+ (k) are conned to the bottom half of the complex plane. Zeros of D+ (k) in the upper half of the complex k plane must lie either on the real axis or on the positive imaginary axis. In the case of a local potential, D+ (k) cannot have a zero for k real [40]; since for a local potential L +(k) = 0 for real k = 0 [47], it follows that for a local potential D+ (k) must not be zero for real k = 0. On the other hand, the modied denition of L +(k) in the case of a nonlocal potential, given by Eq. (l8), allows for a zero of D+ (k) on the real k axis as long as such a zero is accompanied by a corresponding zero of D(k), resulting in a bound state in the continuum. Such zeros of D+ (k) and D(k) on the real k axis will occur in pairs, symmetric about the imaginary axis. That such a zero of D(k) is possible on the real k axis has been established by many examples [40]. However, because the boundary conditions associated with the equation for the regular solution are not the same as those associated with the physical solution, in general it would not be expected that D(k) would be zero for the same values of k as is D+ (k). Indeed, for a local potential D(k) is unity at all zeros of D+ (k). Thus, as mentioned earlier, a zero of D(k) at the same value of k as that for which D+(k) is zero is a special circumstance; usually zeros of D(k) can be expected to occur at values of k for which D+ (k) is not zero. However, zeros of D(k) which occur on the real k axis always occur in pairs, symmetric with respect to the origin. Zeros of D(k) off of the real axis occur in sets of four, symmetric with respect to both the real and imaginary k-axes. Since D(k) is a ratio of two polynomials of the same order, the number of zeros and number of poles in the entire complex plane are equal. We already have pointed out that D(k) will not have poles on the real axis. Its poles will be symmetrically distributed in the upper and lower half planes. If a pair of zeros of D happen to occur on the real axis, then the number of poles in either of the half planes, excluding the real axis, exceeds the number of zeros by 1. If 2m zeros of D(k) lie on the real axis (symmetrically located about the imaginary axis) then in either half plane the number of poles exceeds the number of zeros by m, excluding the real axis.
107
A zero of D+ (k) on the real axis can be of any order. That is, a potential can be constructed which at a given real value k0 of k produces any specied number of continuum bound states. It has been demonstrated [59] that if the order of the zero of D+ (k) at k = k0 is m, then the order of the zero of D(k) at k = k0 is at most m + 1 and at least m. Furthermore, a zero of D(k) at k = k0 which occurs for real k0 = 0 and for which D+ (k0 ) = 0 must be a simple zero [59]. That is, a zero of D(k) at k = k0 of order m where m > 1 must be accompanied by a zero of D+ (k0 ) of at least order m 1. As discussed earlier, the Jost function L + (k)is given by the ratio of D+(k)to D(k). In taking this ratio, poles of D+(k)can be cancelled by poles of D(k) and zeros of D+ (k)can be cancelled by zeros of D(k). Thus, as already has been pointed out [52], the Jost function for a nonlocal potential does not usually contain as complete a set of information about the system as the determinants D+ (k)and D(k). The zeros of D(k) which do not cancel become poles of L +(k)and the poles of D(k) which do not cancel become zeros of L +(k). Thus, when m continuum bound states occur, there will be 2 m zeros of D+(k)and 2m zeros of D(k) located on the real k axis. These zeros will cancel, and as a result of the m continuum bound states the Jost function L +(k)will, for k 0 have m zeros more than the number of its poles. Since bound states result in zeros of D+ (k)on the positive imaginary axis and since these zeros are not cancelled by zeros of D(k), these zeros of D+ (k)also will be zeros of L + (k)in the upper half plane. Thus if a potential is such that it has m continuum bound states and n bound states, the Jost function L +(k)for that potential will have (m + n) more zeros in the upper half plane than its number of poles in the upper half plane. For a spurious state, D(k) = 0 at the value of k at which the state occurs and D(k) = 0 at that same k. As discussed above, when D+ (k) = 0 only a rst order zero of D(k) can occur. Thus at any given value of k there can be only one spurious state. If there are 2s such isolated zeros of D(k) on the real axis, then in the upper half plane (and, for that matter, in the lower half plane as well) D(k) will have s more poles than the number of zeros, excluding the real axis. Since D+(k)is analytic in the upper half-plane, if there are no bound states it follows that L + (k)will have s more zeros than its number of poles (excluding the poles on the real axis) in the upper half of the complex k-plane. Thus, if in addition to s spurious states there are n bound states, in the upper half plane L +(k)will have (s + n) more zeros than the number of poles. We are now in a position to discuss Levinsons theorem in terms of L +(k)or in terms of + (k). Before beginning this discussion, however, we investigate the properties of (k) and D D (k) at k=0, and associate these properties with solutions of Eq.(9).
Solutions at k = 0
Newton [60] has pointed out that for a nonsingular, central, local potential L + (k)can be zero at k = 0. Such a zero must be simple, and does not correspond to a normalizable solution of the radial equation. Newton refers to the solution of the radial equation under these circumstances as a half-bound state [61]. In the case of a nonlocal potential, such a state is also possible. It will be characterized by a zero of D+ (k)but not a zero of D(k); L +(k)thus will be zero under this circumstance. It is also possible in the case of a nonlocal potential to have normalizable states at k = 0. The Saito potential is an example of a nonlocal potential with a zero-energy bound state, although this case is complicated by the fact that the Saito
108
potential is not symmetric [62]. A nonlocal potential can have more than one of these states, which are characterized by zeros of both D+ (k)and D(k), and thus no zero of L +(k). Furthermore, if there are q of these states at k = 0, then the order of the zeros of D+ (k)and D(k) is 2q. This can easily be seen by realizing that such a state is a continuum bound state which happens to occur at k = 0 rather than some positive value of k. A CBS at k0 , where k0 > 0, is characterized by a pair of zeros of D+(k), one at k0 and the other at k0 , and a pair of zeros of D(k) at those same positions; zeros of D+ (k)and D(k) at k = 0 clearly will occur as double zeros. The Levinsons theorem contour for a half-bound state, since it implies a zero of
L +(k)and D+ (k), is handled for both L and D in exactly the same manner as for a local potential. Integration of either L + (k) /L +(k) or D+ (k) /D+ (k) clockwise around a
semicircle located at the origin yields i, as discussed in Sec. 2. The treatment of the contour for Levinsons theorem in the case of a zero energy bound state differs depending upon whether one is considering L or D . For a continuum bound state at zero energy, L +(k)will not have a zero at k = 0, and thus no adjustment of the contour is necessary to accommodate the presence of the state. As discussed above, however, D+ (k)will have a double zero at k = 0. In order to evaluate the integral of D+ (k) /D+ (k) around an unambiguous contour in terms of the zero energy phase shift we can slightly displace the double zeros about the origin by and let 0 as shown in Fig. 2. The contribution from each of the contours in Fig. 2 will be i, resulting in a total contribution as 0 of 2i. It is also necessary to consider how the presence of the pair of zeros of D+ (k)affects the phase shift D at k = 0. This question, however, is deferred until the next section.
Figure 2. Contour C for integration of D(k)+/D(k) in the case double zeros of D+(k) at the origin. These zeros, displaced by in order to dene an unambiguous contour, are indicated by the symbol o.
109
6
6.1
Derivation of the Generalized Levinsons Theorem for the Phase Shift

Case of n Bound States and One Half-Bound State
For a nonlocal potential with n bound states and one half-bound state, the Jost function L +(k)will have n zeros on the positive real axis and one zero at the origin, as shown in Fig. 3. There will be no poles on or above the real axis. Thus, making use of Eq. (4) for the contour C of Fig. 3 gives L +(k) dk = 2in (19) + C L (k) As discussed in the case of a local potential in Sec. 2, the explicit line integral yields
L +(k) dk = 2i[L (0) L ()] i. + C L (k)

Thus we obtain the same result as in the case of a local potential, namely 1 L (0) L () = (n + ). 2
(20)
(21)
Figure 3. Contour C for the derivation of the Generalized Levinsons Theorem for L when there are n bound states and a half-bound state at k = 0. Zeros of L + (k)are indicated by the symbol o.
6.2
Case of n Bound States and m Continuum Bound States
As discussed in Sec. 4, the Jost function L +(k)for a potential with n bound states and m continuum bound states has n + m more zeros than poles in the upper half plane, and has no zeros or poles on the real axis. This set of conditions is indicated in Fig. 4, and applies whether the continuum bound states are at k = 0 or k > 0.
110
S.B. Qadri, B. Mulligan, M.F. Mahmood et al. Calculating the logarithmic derivative of L + (k)by the two different methods yields
L +(k) dk = 2i(n + m) + C L (k)

and
(22)
L +(k) dk = 2i[L (0) L ()]. + C L (k)
(23)
Figure 4. Contour C for the derivation of the Generalized Levinsons Theorem for L when there are n bound states and m continuum bound states. Zeros and poles of L +(k)are indicated by o and , respectively. At least n zeros of L +(k)must be on the positive imaginary axis. Additional zeros of L +(k)and the poles of L +(k)in the upper half plane may be either on the positive imaginary axis or off the axis in pairs symmetrically located with respect to the imaginary axis. The assertion that there be n + m more zeros than poles in the upper half plane is a condition which takes into account all of the zeros and poles. Thus, in this case L (0) L () = (n + m). If in addition there is a zero of L +(k)at k = 0 we get 1 L (0) L () = (n + m ). 2 (25) (24)
6.3
Case of n Bound States and p Spurious States
As discussed in Sec. 4, the Jost function L + (k)for a potential with n bound states and s spurious states has n + s more zeros than poles above the real axis. In addition, L +(k)is characterized by 2s poles located in pairs on the real axis. This situation is shown in Fig. 5. Using Eq. (4) for the contour C of Fig. 5 we get
L +(k) dk = 2i(n + s). + C L (k)
(26)
111
Figure 5. Contour C for the derivation of the Generalized Levinsons Theorem for L when there aren bound states and s spurious states. Zeros and poles of L + (k)are indicated by o and respectively. At least n zeros of L +(k)must be on the positive imaginary axis and 2s poles of L +(k)must be on the real axis in pairs symmetric with respect to the imaginary axis. The additional zeros and poles of L +(k)in the upper half plane must either be on the positive imaginary axis, or in pairs located symmetrically with respect to that axis. The assertion that there be n + s more zeros than poles in the upper half plane is a condition which takes into account all of the zeros and poles of L + (k)above the real axis. Direct integration of the logarithmic derivative of L + (k)around C yields
s L +(k) dk = 2i L (0) L () + [L (ki + 0) L (ki 0)] + 2is, + C L (k) i=1
(27)
where the term 2is arises from integrating around the 2s semicircles on the real axis, as shown in Fig. 5. When a spurious state occurs at k = ki there is a phase change of in going from k = ki to k = ki + in the limit as 0. This is L (ki + 0) L (ki 0) = . Thus Eq. (27) becomes (28)
L + (k) dk = 2i[L (0) L () s] + 2is. + C L (k)

Taken together, Eqs. (26) and (29) imply L (0) L () = (n + s). If, in addition, there is a zero of L + (k)at k = 0 then we get 1 L (0) L () = (n + s ). 2
(29)
(30)
(31)
112
6.4
Summary
We have seen that if we use L as the denition of the phase shift, then Levinsons theorem becomes (32) L (0) L () = (n + p), where n is the number of bound states and p is the number of continuum bound states (including those at k = 0) and/or spurious states. If, also, we have a zero of L + (k)at k = 0 then Eq. (32) becomes 1 (33) L (0) L () = (n + p + ), 2
7
7.1
Derivation of the Generalized Levinson s Theorem for the Phase Shift

Case of n Bound States and One Half-Bound State
With respect to bound states and half-bound states, the structure of the poles and zeros of D+ (k)is the same as that of L +(k). Thus, as shown in Fig. 6, there are no poles of D+(k)on or above the real axis, n zeros of D+ (k)on the positive imaginary axis corresponding to n bound states, and a simple zero of D+ (k)at k = 0 corresponding to a half-bound state. Applying Eq. (4) to the logarithmic derivative of D+ (k)for the contour C shown in Fig. 6, we get D+(k) dk = 2in. (34) + C D (k) Direct integration yields D+ (k) dk = 2i[D(0) D ()] i. D+ (k) 1 D (0) D () = (n + ). 2 (35)
Therefore
(36)
7.2
Case of n Bound States and m Continuum Bound States
As discussed in Sec. 4, D+ (k) for a nonlocal potential has no poles on or above the real axis. If the potential has n bound states, then there will be n zeros of D+(k)on the positive imaginary axis, as shown in Fig. 7. If, in addition, the potential has m continuum bound states, then there will be 2m zeros of D+ (k)on the real axis, in pairs symmetric about k = 0. This also is shown in Fig. 7. Using Eq. 4 and integrating around the contour C shown in Fig. 7 yields D+(k) dk = 2in. (37) + C D (k) From the line integral around the contour we get
m D+(k) dk = 2i D (0) D () + [D (ki + 0) D (ki 0)] 2mi, D+ (k) i=1
(38)
113
Figure 6. Contour C for the derivation of the Generalized Levinsons Theorem for D when there are n bound states for k = 0 and a half-bound state at k = 0. Zeros of D+ (k) are indicated by o and respectively.
Figure 7. Contour C for the derivation of the Generalized Levinsons Theorem for D when there are n bound states and m continuum bound states. Zeros of D+ (k) are indicated by o. where the term 2mi results from the integrals around the 2m semicircles along the real axis. Since D (ki + 0) D (ki 0) = . (39) Eq. (39) becomes D+ (k) dk = 2i[D(0) D () + m] 2mi, + C D (k) Combining Eqs. (37) and (40) gives D (0) D () = n. (41) (40)
If, in addition, there is a zero of D+ (k)of order q at k = 0, it is necessary to discuss this derivation further. A continuum bound state at k = 0 requires a double zero of D+(k). Thus, if q is odd, there will be a half-bound state at k = 0. Since only one such state is
114
possible according to the result in part A of this section, an odd value of q will add 1 2 to the right hand side of Eq. (41). There will be no contribution to the right hand side of Eq. (41), however, in the case of a zero-energy continuum bound state. As discussed in Sec. 3 one can move the pair of zeros to give a zero at k = and one at k = ; integrating around these two semicircles will yield 2i. But, as indicated above, the phase shift is now discontinuous at k = and one at k = , and this discontinuity yields 2 i. This cancels the contribution of the integrals around the semicircles. Thus for n bound states, m continuum bound states (either at k = 0 or k > 0), and a half-bound state, Levinsons theorem for D (k) is 1 (42) D (0) D () = (n + ). 2 Actually, there is an ambiguity in the zero energy phase shift D . As the parameters of a nonlocal potential are adjusted, a zero of D+ (k)can move from a position on the positive imaginary axis down to k = 0. At that point, the bound state becomes a continuum bound state, at zero energy. Whereas the bound state contributed to the zero energy phase shift D the continuum bound state will contribute nothing. Thus a very small change in the potential parameters can result in a discontinuity of the phase shift D by at zero energy. Such a discontinuity did not occur in the case of the phase shift L under these circumstances.
7.3
Case of n Bound States and p Spurious States
Corresponding to n bound states there will be n zeros of D+ (k)on the positive imaginary axis, as shown in Fig. 8. Since p spurious states correspond to p zeros of D(k), without corresponding zeros of D+ (k)on or above the real axis, the logarithmic derivative of D+ (k)for the contour C of Fig. 8 will not be affected by the p spurious states. Applying the argument principle then will give D+(k) dk = 2in. (43) + C D (k)
Figure 8. Contour C for the derivation of the Generalized Levinsons Theorem for when there are n bound states and p spurious states. Zeros of D+ (k) are indicated by the symbol o.
Effects of Nonlocality and Phase Shift Denitions... Direct integration yields D+(k) dk = 2i[D (0) D ()]. D+ (k) D (0) D () = n.
115
(44)
Therefore (45)
7.4
Summary
We have seen that if we use D as the denition for the phase shift, then Levinsons theorem becomes (46) D (0) D () = n. where n is the number of bound states. The phase shift does not depend upon the number of continuum bound states and/or spurious states. If, also, we have a rst-order zero of D+ (k)at k = 0, which would be the case if and only if a half-bound state is present, then Eq. (46) becomes 1 (47) D (0) D () = (n + ). 2
Examples
In this section we consider two examples of nonlocal potentials and calculate the phase shift (0)() using both of the denitions, D (k) = |D (k)|e iD and L (k) = |L (k)|e iL . Both examples will make use of nonlocal potentials of the form V (r, r ) = g(r)g(r ). D+ (k)and (48)
These two examples have been chosen because the behavior in the complex plane of D(k) for these potentials has been analyzed in a previous paper [55].
8.1
Yamaguchi Potential
In 1954 Yamaguchi [63] introduced a one-term separable nonlocal potential to describe nucleon-nucleon scattering. The Yamaguchi potential is of the form shown in Eq. (48) with g(r) = er . (49)
The expressions for D+ (k)and D(k) for this potential are given in Refs. [40] and [55]. They can be written in the form (50) D+(k) = N + (k)/M +(k), where N + (k) = 2k2 + i42 k ( + 23 ) and M +(k) = 2(k + i)2 (52) (51)
116 and
D(k) = N(k)/M(k), where N(k) = 2k2 ( 23 ) and M(k) = 2(k i)(k + i). From these expressions we see that the roots of N + (k) are k1,2 = i . 2
(53) (54) (55)
(56)
No values of , will make D+(k)zero for real values of k; therefore no continuum bound state can be associated with the Yamaguchi form factor. The roots of N(k) are k1,2 = 2 . 2 (57)
Thus D(k) can be zero for a wide range of values of and . Although the values of and used by Yamaguchi do not generate a spurious state at any energy, if > 23 a spurious state will occur. The zeros of M +(k) are k1,2 = i. The zeros of M(k) are k1,2 = i. 8.1.1 Phase Shift Calculation Using L (59) (58)
We now use the denition L +(k) = |L +(k)|eiL (k) to obtain the phase shift in two cases. Case 1. When < 23 , there is no spurious state. L +(k)has one zero in the upper half plane, at k = i, and a pole at k = i (23 )/2. Using Eq. (4) for L + (k)to obtain the integral around the contour C shown in Fig. 9 yields L + (k) dk = 0. (60) + C L (k) The line integral around the contour C gives
L +(k) dk = ln L +(k) + C L (k)

or
+
0 0
L +(Kei ) Kiei d + ln L +(k) L + (Kei )

0
.
K
(61)
L +(k) L + (K) L + (0) dk = ln + + ln + + + + L (K) L (0 ) C L (k)
L +(Kei ) Kiei d. L +(Kei )
(62)
117
Figure 9. Contour C for integration for Levinsons Theorem for L , showing the poles and zeros of L + (k)in the upper half of the k plane for the Yamaguchi potential with parameters for which there is no spurious state. [There are no zeros or poles of L +(k)on the real axis.] Zeros and poles of L + (k)for this potential are indicated by o and , respectively. The integral on the right hand side of Eq. (62) vanishes as K . Using the fact that (k) = (k), Eq. (62) implies lim
L +(k) dk = 2i[L (0) L ()]. + C L (k)

L (0) L () = 0.
(63)
Therefore we get (64)
Case 2. When > 23, there is a spurious state. The Jost function has one zero at k = i and two poles, at k = ( 23)/2 = k1 , on the real axis. Therefore, integrating around the contour C shown in Fig. 10 yields
L +(k) dk = 2i. + C L (k)
(65)
In the limits 0 and K , direct integration around the contour shown in Fig. 10 gives
L +(k) L + (k1 0) L +() L + (0) L + (k1 +0) dk = ln ln + +ln + ln +2i. + L +(k1 +0) L (+) L (+0) L +(k1 0) C L (k)
(66) or
L + (k) dk = 2i [L (0) L () + L (k1 +0) L (k1 0)] + 2i. + C L (k)
(67)
Because of the spurious state at k = k1, there is a phase change of in going from k = k1 to k = k1 + That is, in the limit as 0, L (k1 +0) L (k1 0) = . (68)
118
Figure 10. Contour C for integration for Levinsons Theorem for L , showing the poles and zeros of L + (k)in the upper half of the k plane and on the real axis for the Yamaguehi potential with parameters for which there is a spvrious state at k = k1 . Zeros and poles of L +(k)for this potential are indicated by o and , respectively. Combining Eqs. (65), (67), and (68) yields 2i = 2i [L (0) L () ] + 2i, which implies L (0) L () = . 8.1.2 Phase Shift Calculation Using D (70) (69)
Next we use the denition D+(k) = |D+(k)| eiD (k) to get the zero-energy phase shift. In both the case > 23 and < 23 the result will be the same, since for all values of and the zeros of D+ (k) lie in the lower half plane. Using Eq. (4) for D+(k)and integrating around the contour C shown in Fig. 11, we get D+(k) dk = 0. D+ (k) (71)
The line integral around the contour yields, in the limit as K and 0, D+ (k) dk = ln D+ (0) ln D+() + ln D+() ln D+ (0). + C D (k) Combining Eqs. (7l) and (72) gives D (0) D () = 0. (73) (72)
8.2
Beregi Potential
The second example to be considered here is the Beregi potential. Beregi [33] suggested a one-term separable nonlocal potential of the form given by Eq. (48) with g(r) = e1 r ae2 r . (74)
119
Figure 11. Contour C for integration for Levinsons Theorem for D the Yamaguchi potential. [Note that there are no poles or zeros of D+ (k) in the upper half-plane.] This potential yields a continuum bound state at 259 .3 MeV and a bound state at 2.225 MeV [41]. The parameters for the Beregi potential are = 302.73 f m3 , 1 = 2.67 f m1, 2 = 5.34 f m1, a = 3.0854. The Fredholm determinants for this potential are given in Refs. [41] and [55]. They can be written as N + (k) (75) D+ (k) = + M (k) where N + (k) = k4[212(1 + 2 )] + ik3 (1 + 2 )[422 + 412 ] 1 2 + k2 [(1 + 2 )(232 21 3 2 a2 1 82 2 + 4a12 ] 1 2 1 2 + ik[(1 + 2 )(4a12 2a22 42 3 43 2 22) 1 1 2 1 2 2 + 8a2 2 + 8a1 2] 1 2 + [(1 + 2 )(233 + 3 + a2 3 ) 4a2 2] 1 2 2 1 1 2 and M +(k) = 21 2 (1 + 2 )(k + i1)2(k + i2 )2 and D(k) = N(k) M(k) (78) (77) (76)
120
S.B. Qadri, B. Mulligan, M.F. Mahmood et al. Table 1. Zeros of N + (k) and M + (k) for the Beregi Potential ( f m1 ) Zeros of N(k)+ (1.77,0) (1.77,0) (0,0.23) (0,-16.25) Zeros of M +(k) (0,-2.67) (0,-2.67) (0,-5.34) (0,-5.34)
Table 2. Zeros of N(k) and M(k) for the Beregi Potential ( f m1 ) Zeros of N(k) (1.77,0) (-1.77,0) (0,11.49) (0,-11.49) where N(k) = k4 [212(1 + 2 )] + k2 [(1 + 2 )(213 + 23 2 2 a2 1) + 41 2 a] 2 1 + [(1 + 2 )(233 3 + 21 2 a(1 + 2 ) a2 3) 42 2 a] 1 2 2 1 1 2 and M(k) = 212(1 + 2 )(2 + k2 )(2 + k2 ) 1 2 (80a) The zeros of N + (k) and M +(k) are tabulated in Table I, whi1e the zeros of N(k) and M(k) are tabulated in Table II. The information from these tables is combined in Table III to give the zeros and poles of L + (k)for this potential. 8.2.1 Phase Shift Calculation Using L (k) (80) Zeroso f M(k) (0,2.67) (0,-2.67) (0,5.34) (0,-5.34)
The zero-energy phase shift c l associated with the Beregi potential can be calculated as discussed in Sec. 6. Using Eq. (4) for L +(k)and integrating around the contour C of Fig. 12, we have L +(k) dk = 4i. (81) + C L (k) Explicit calculation of the line integral around the contour C yields
L +(k) dk = 2i[L (0) L ()]. + C L (k)

Table 3. Zeros of L +(k) for the Beregi Potential Zeros (0,0.23) (0,2.67) (0,5.34) (0,-16.25) Poles (0,11.49) (0,-11.49) (0,-2.67) (0,-5.34)
(82)
121
Figure 12. Contour C for integration for Levinsons Theorem for L , showing the poles of L +(k)in the upper half-plane for the Beregi potential [There are no zeros or poles of L +(k)on the real axis.] Zeros and poles are indicated by o and , respectively. This gives L (0) L () = 2 8.2.2 Phase Shift Calculation Using L (83)
The information necessary for the calculation of L for the Beregi potential is given in Table I. Using the argument principle for the contour C given in Fig. 13, we obtain D+ (k) dk = 2i. + C D (k) (84)
Figure 13. Contour C for integration for Levinsons Theorem for D showing the zeros of D+ (k) on the real axis for the Beregi potential. These zeros are indicated by o. Taking k1 = 1.77 f m1, the explicit line integral yields D+ (k) dk = 2i[D (0) D () + D (k1 +0) D (k1 0)] 2i. + C D (k) (85)
122
S.B. Qadri, B. Mulligan, M.F. Mahmood et al. The discontinuity in the phase shift at k = k1 is given by D (k1 +0) D (k1 0) = . From Eqs. (84), (85), and (86) we get 2i[D (0) D ()] = 2i (87) (86)
or D (0) D () = . (88)
Summary and Conclusion
In this section we compare the proofs of Levinsons theorem given in Secs. 6 and 7 with other proofs in the literature. In so doing, it is important to note that in derivations of Levinsons theorem in the literature some authors have used L + (k)or D+ (k) and others L (k) or D (k). The use of L (k) or D (k) instead of L +(k)or D+(k) leads to a corresponding change in the contour integral considered from the upper half-plane to the lower half-plane. As mentioned in the introduction, one of the earliest derivations of the generalized Levinsons theorem for a nonlocal potential was that due to P. Swan [2]. The problem he was considering was that of nonlocality due to antisynimetrization of the incident particle with respect to particles in the target. He found that the natural denition to use was the phase of L (k), that is the phase shift L . [In his paper the notation for L (k) is f (k).] The contour used for his calculation and the pertinent zeros and poles are displayed in Fig. 5 of Ref. [2] and the phase shift calculation is carried out on p. 23 of Ref. [2]. It is interesting to note that although he showed that L (k) could have poles on the real axis, he obtained a zero energy phase shift of zero. Using the approach presented here, for his case we would obtain L (0) = instead of L (0) = 0. That is, although he recognized the presence of poles on the real axis, Swan did not take into consideration a discontinuity of in the phase shift as the energy increases through a spurious state. Also, in his derivation of the generalized Levinsons theorem redundant states occur in his Eq. (54), which reads as follows: f (k) 0 f (k) 0
0 (r 0 (r
C1
) f (k, r ) dr dk = ) f (k, r ) dr
C1
f (k) dk. f (k)
(87a)
He assumes the existence of the Jost solution f (k, r) under these conditions. As pointed out in Ref. [52], a Jost solution may or may not exist when a continuum bound state or a redundant state occurs. Thus his subsequent derivation is applicable only to those cases in which Jost solutions exist. Jauchs [5] generalized Levinsons theorem was undertaken with the assumption that a nonlocal potential does not have bound states in the continuum. In his generalization of Levinsons theorem, Jauch used the denition of the phase shift as the phase of the S-matrix, that is, (88a) S(k) = e2i(k) . As pointed out in Ref. 41, this denition is not sensitive to the presence of bound states in the continuum or to discontinuities in the phase shift due to poles of L +(k)on the real
123
axis. Thus Jauch left out the important consideration of the existence of continuum bound states and spurious states in the case of a nonlocal potential. Shortly after Jauchs paper, Gourdin and Martin [6] noted that a continuum bound state could be associated with a nonlocal potential and that such a state required a modication of Levinsons theorem. Martin [8] showed that for a nonlocal potential with continuum bound states the index in Eq. (2) should be replaced by + ), where is the number of bound states with energy less than zero and is the number of continuum bound states. He used the denition L in his derivation, and his result can be expressed in the following form, (0) () = ( + ). (89)
His method does not take into consideration possible zeros of D(k) and thus the possible presence of spurious states. The derivation of Levinsons theorem presented by Bertero, Talenti, and Viano [29] is based on consideration of the Fredholm determinant D+ (k). Their derivation, which begins on p.635 of Ref. [29], considers the phase shift associated with D+ (k) [denoted in the paper by F(k)] and the contour is closed in the upper half-plane. They consider N zeros in the upper half-plane and 2N zeros on the real axis symmetrically situated about the imaginary axis. There are also q zeros at the origin. In the nomenclature of Bertero et al. a continuum bound state is called a spurious state and states labeled spurious states in our nomenclature are not considered. Figure 1 on p. 637 of Ref. [29] displays the zeros mentioned above. The contour integration in Ref.[ 29] yields the following equation
N F(k) dk = 2i (0) eta() + [(ki +0) (ki 0] 2iN iq. C F(k) i=1
(90)
However, although F(k) [and thus (k)] is not continuous at a continuum bound state, Bertero et al. took the phase shift to be continuous at the continuum bound states (which are in their terminology spurious states), and obtained the following relation q (0) () = (N + N + ). 2 (91)
Had they considered the phase shift to be discontinuous at the continuum bound states they would have obtained the relation (36), but with an ambiguity at zero energy as to whether one should consider zeros of D+ (k) as zero-energy continuum bound states or zeroenergy bound states. Also, in addition to not taking into consideration the phase change at continuum bound states, their proof does not refer to the possibility of a spurious state occurring. Gl ckle and LeTourneux [36] studied the generalized Levinsons theorem for composite o particles in the framework of the Saito model. They discussed the zeros of D+ (k) [which they designated by D(k)] and used the phase of D+ (k) for getting the zero energy phase shift. In their proof, which is given on p. 21 of Ref. [36], they assume m bound states and n continuum bound states. The zeros of D+ (k) corresponding to n continuum bound states lie on the real axis symmetrically about the imaginary axis. They used the integral in Eq. (4)
124
and obtained the contribution 2in from the 2n semicircles around the zeros on the real axis. However in arriving at their Eq. (38), which reads as follows,
dk
d ln D(k) = 2i[(0) ()], dk
(92)
they did not take into account the phase change at a continuum bound state. Thus they obtained the relation (0) () = (m + n). (93) Since there is discontinuity of in the phase of D(k) at each continuum bound state, their result should have been (0) () = m. (94) For a large class of nonlocal potentials giving rise to continuum bound states, Buslaev obtained Levinsons theorem in its original form. Buslaevs work is consistent with Bolsterlis results. It is also consistent with our derivation of Levinsons theorem using the denition D when continuum bound states are present. Dreyfus has also studied Levinsons theorem for nonlocal interactions. He pointed out that owing to the two dentions of the scattering phase shift we get two different versions of Levinsons theorem; one of them takes the continuum bound states into account while the other does not. However, neither Buslaev nor Dreyfus take into consideration the possibility of the existence of spurious states. Finally, Newton has considered the point k = 0 and proved a generalized Levinsons theorem which includes the possiblity of a bound state or a half-bound state at k = 0. His derivation is based on the Fredholm determinant D+ (k). By using the phase of D+ (k) to dene a phase shift continuous at a continuum bound state, he obtained a statement of Levinsons theorem which is the same as that which we nd here for L . He also presents strong arguments as to why this is the proper statement of the theorem. He does not, however, consider the possibility of spurious states in his derivation. In studying Levinsons theorem for a nonlocal potential we have taken into account the fact that redundant states, continuum bound states, and spurious states are all possible in the case of a nonlocal potential. This has been done for both of the possible denitions, D and L , of the phase shift for scattering by a nonlocal potential outlined in Ref.[ 41]. Thus the present work discusses a variety of aspects of Levinsons theorem not previously treated in the case of a nonlocal potential. An important aspect of our ability to extend Levinsons theorem to nonlocal potentials rests upon the results of Ref. [55], in which we introduced functional analytic techniques for determining the character of the Fredholm determinants D+ (k) and D(k) associated with integral equations for scattering solutions of the radial equation. In discussing the derivation of the generalized Levinsons theorem for both the denitions of the phase shifts we have seen that the generalized Levinsons theorem associated with denition D does not take into account the presence of continuum bound states or spurious states. Also there is ambiguity in D when there are zeros of D(k) at the origin. The denition L takes into account the presence of continuum bound states as well as spurious states and half-bound states or zero-energy bound states. The denition D fails completely in the presence of redundant states, while the denition L gives consistent results. All things considered, we nd that the phase shift L seems to be more useful in all
125
the cases considered for the generalization of Levinsons theorem. In this respect we agree with the conclusion of Ref. [41] which prefers the denition L on the basis of different considerations.
References
[1] N. Levinson, Kgl. Danski Videnskab. Selskab. Mat. Fys. 25 (1949) No. 9. [2] P. Swan, Proc. Roy. Soc. London, Sect. A228 (1955) 10. [3] A. Temkin, J. Math. Phys. 2 (1961) 336. [4] L. Rosenberg and L. Spruch, Phys. Rev. 121 (1961) 1720. [5] J. M. Jauch, Helv. Phys. Acta 30 (1957) 143. [6] M. Gourdin and A. Martin, Nuovo Cimento 6 (1957) 757. [7] Barry Simon, Comm. Pure and Appl. Math. 21 (1967) 531. [8] A. Martin, Nuovo Cimento 7 (1958) 607. [9] I. 0. Krause and B. Mulligan, Ann. Phys.(N.Y.) 94 (1975) 31. [10] E. Feenberg, Phys. Rev. 42 (1932) 17. [11] S. Hochberg, H. S. W. Massey, and L. H. Underhill, Proc. Phys. Soc. London A67 (1954) 957. [12] S. Saito, Prog. Theor. Phys. 40 (1968) 893; 41 (1969) 705. [13] R. Tamagaki, Prog. Theo. Phys. Suppl. Extra Number (1968), p.242 [14] K. Wildermuth and Y. C. Tang, A Unical Theory of the Nucleus (vieweg, Baunschweig, Germany, 1977). [15] B. Buck, H. Friedrich, and C. Wheatley, Nucl. Phys. A275 (1977) 246. [16] M. Ida, Prog. Theor. Phys. 21 (1959) 625. [17] R. Haag, Nuovo Cimento 5 (1957) 203. [18] C. Koniski and T. Ogimoto, Prg. Theo. Phys. 22 (1959) 807. [19] J. C. Polkighorne, Proc. Cambr. Phil. Soc. 54 (1959) 560. [20] M. T. Vaughn, R. Aaron, and R. D. Amado, Phys. Rev. 124 (1961) 1258.
126
[21] G. C. Ghirardi, M. Pauli, and A. Rimini, Ann. Phys. (N.Y.) 21 (1963) 401. [22] R. L. Warnock, Phys. Rev. 131 (1963) 1320. [23] J. M. Charap, Nuovo Cimento 36 (1965) 419. [24] M. Kato, Ann. Phys. (N.Y.) 31 (1965) 130. [25] D. Atkinson, K. Dietz, and D. Morgan, Ann. Phys. (N.Y.) 37 (1966) 77. [26] V. G. Kadyskevskii, R. M. Mir-Kasimov., and N. B. Skachkov, Nuovo Cimento 55 (1967) 233. [27] K. L. Nagy, Acta Phys. Hung. 24 (1968) 433. [28] J. B. Hartle and C. E. Jones, Ann. Phys.(N.Y.) 38 (1969) 348. [29] M. Bertero, G. Talenti, and C. A. Viano, Mud. Phys. A113 (1968) 625. [30] V. S. Buslaev, Probl. Math. Phys. (Leningrad) 4 (1969) 43. [31] T. Dreyfus, thesis, University of Geneva, 1976, unpublished preprint, and J. Phys. A; Math. Gen. 9 (1976) L187. [32] M. Bolsterli, Phys. Rev. 182 (1969) 1095. [33] P. Beregi, Nucl. Phys. A206 (1973) 217. [34] R. L. Mills and J. F. Reading, J. Math. Phys. 10 (1969) 321. [35] L. Horwitz and J. P. Marchand, Rocky Mount. J. Math. 1 (1971) 225. [36] W. Gltickle and Jean LeTourneux, Nucl. Phys. A269 (1976) 16. [37] M. Von W 511enberg, Math. Nachr. 78 (1977) 223; 78 (1977) 369. [38] R. C. Newton, J. Math. Phys. 18 (1977) 1582. [39] P. Beregi, B. N. Zakharev and S. A. Niyazgulov, Soy. J. Particles Nucl. 4 (1973) 217. [40] B. Mulligan, L. G. Arnold, B. Bagchi, and T. 0. Krause, Phys. Rev. C 13 (1976) 2131. [41] B. Bagehi, T. 0. Krause, and B. Mulligan, Phys. Rev. C 15 (1977) 1623. [42] M. Coz, L. C. Arnold, and A. D. MacKellar, Ann. Phys. (N.Y.) 59 (1970) 219. [43] L. C. Arnold and A. D. MacKellar, Phys. Rev. C 3 (1971) 1095.
Effects of Nonlocality and Phase Shift Denitions... [44] B. Bagchi and B. Mulligan, Phys. Rev. C 20 (1979) 1973. [45] R. Jost and A. Pais, Phys. Rev. 82 (1951) 840. [46] L. C. Arnold, B. Bagchi, and B. Mulligan, unpublished.
127
[47] R. G. Newton, Scattering Theory of Waves and Particles (McGraw-Hill, New York 1966). [48] M. Bertero, G.. Talenti, and C. A. Viano, Nuovo Cimento 46 (1966) 337. [49] C. S. Warke and R. K. Bhaduri, Nucl. Phys. A162 (1971) 289. [50] Y. Singh and C. S. Warke, Can. J. Phys. 49 (1971) 1029. [51] S. S. Ahmed, Nuovo Cimento 23A (1974) 362. [52] B. Bagchi, B. Mulligan, and S. B. Qadri, Phys. Rev. C 20 (1979) 1251. [53] T. 0. Krause and B. Mulligan, J. Math. Phys. 15 (1974) 770. [54] H. M. Nussenzveig, Causality and Dispersion Relations (Academic Press, New York 1972). [55] B. Mulligan and S. B. Qadri, Phys. Rev. C 24 (1981) 874. [56] M. Iwasaki and B. Mulligan, Bull. Am. Phys. Soc. 22 (1977) 1030; 23 (1978) 629; and to be published. [57] F. Smithies, Integral Equations (Cambridge University Press, Cambridge 1958). [58] K. Yosida, Functional Analysis (Spring-Verlag, Berlin 1974), pp.279-286. [59] B. Mulligan and A. D. Wolfe, to be published. [60] R. G. Newton, J. Math. Phys. 1 (1960) 319. [61] R. G. Newton, J. Math. Phys. 18 (1977) 1348. [62] B. Bagchi, B. Mulligan, and S. B. Qadri, Prog. Theor. Phys. 60 (1978) 765. [63] Y. Yamaguchi, Phys. Rev. 95 (1954) 1628.
A SYMPTOTIC B EHAVIOR OF THE F ITZ H UGH -N AGUMO S YSTEM

Weishi Liu1and Bixiang Wang 2 Department of Mathematics, University of Kansas, Lawrence, KS 66045, USA 2 Department of Mathematics, New Mexico Institute of Mining and Technology Socorro, NM 87801, USA
1
Abstract For the FitzHugh-Nagumo system dened on I we prove the existence of a comR, pact global attractor in a weighted Sobolev space which contains bounded solutions, in particular, traveling wave and spatially periodic solutions. We also study the behavior of the global attractors as a parameter goes to zero. Although the limiting system for = 0 does not possess a bounded attracting set, we show that there exists a constant 0 are all contained in a compact subset 0 such that global attractors for 0 < of the phase space. Furthermore, we construct a compact local attractor for the limiting system and establish the upper semicontinuity of the global attractors of perturbed system and the local attractor of the limiting system.
AMS Subject Classication: Primary 35B40. Secondary 35B41, 35K57, 92C55 Key Words: FitzHugh-Nagumo system, asymptotic behavior, global attractor
1.
Introduction
v t w t 2v + h(v) w, x2 (v w),
We are interested in the long-time behavior of the FitzHugh-Nagumo system on I R: = = (1.1) (1.2)
E-mail address: wliu@math.ku.edu, supported in part by NSF Grant DMS-0406998 and a General Research Fund of University of Kansas. E-mail address: bwang@nmt.edu, supported in part by the Start-up Fund of New Mexico Institute of Mining and Technology.
130
Weishi Liu and Bixiang Wang
where, and are positive constants, h is a given nonlinear function (typically, a cubic like function). The FitzHugh-Nagumo system was introduced as a simplication of the Hodgkin-Huxley equation that was derived as a model for the propagation of action potentials in the giant nerve axon of the squid (see, for example, [6, 10, 21]). The nonlinearity in the Hodgkin-Huxley is obtained by assuming a certain form and then tting the nonlinearity to experimental data by nonlinear least squares or by solving an inverse problem [30], and the cubic like function h in the FitzHugh-Nagumo equation is a simplication that captures some main qualitative behavior of the Hodgkin-Huxley equation. System (1.1)-(1.2) has been studied extensively along two major directions: the existence and stability of traveling waves, and the long-time behavior of the solutions. The existence of traveling waves for this system was studied in detail through bifurcation analysis in [8], a combination of center manifold reduction and the Lyapunov-Schmidt reduction in [25], and the geometric singular perturbation method in [14, 16]. The stability of the traveling waves was rst studied in [13], where the Evans function method was successfully developed. The method has been extended in [1, 25, 26] and becomes predominant in the investigation of stability problems. The asymptotic behavior of solutions for system (1.1)(1.2) has been studied in [18, 19, 27]) when the system is dened in a bounded domain. In the case of an unbounded domain, the existence of a global attractor for this system was R R proved in [23] in the standard L2(I n ) L2 (I n) space, which is compact in the phase space but does not contain traveling wave solutions. The existence of a global attractor in a locally uniform Sobolev space was showed in [7], which contains traveling wave solutions but is not compact with respect to the norm topology of the phase space. As a consequence, the global attractor in [7] attracts solutions only in a weak topology, but not in the norm topology, of the phase space. In the present paper, as a necessary step toward the understanding of the global asymptotic behavior, we consider existence of global attractors and study the behavior of these attractors (at the level of invariant sets) as the parameter 0. Note that, if is set to zero in (1.1)-(1.2), then the component w is conserved, and hence, there cannot be any bounded set that attracts all solutions of the limiting system. However, we will establish, in a suitably weighted space, the existence of a compact global attractor that contains all bounded solutions for all positive but small and show that the union of all global attractors of the perturbed system (1.1)-(1.2) for positive but small is actually contained in a compact subset in the phase space. Furthermore, we will construct a compact local attractor for the limiting system and establish the upper semicontinuity of global attractors for the perturbed system at the compact local attractor for the limiting system. Our method is based on the idea of tail ends estimates on solutions, by showing that the solutions of system (1.1)(1.2) are uniformly small when space and time variables are sufciently large. Actually, along with other things, the uniform smallness of tail ends of solutions for large time is not only sufcient, but also necessary for the asymptotic compactness of solution operators of dissipative evolution equations. The idea of tail ends estimates on solutions was used in [31] for the Reaction-Diffusion equation and in [23] for the FitzHugh-Nagumo system in the standard L2(I n ) space, respectively. This paper demonstrates that the method of tail R ends estimates is effective not only for the standard Sobolev spaces, but also for weighted Sobolev spaces. The existence of global attractors for dissipative equations on unbounded domains can be found in [2, 4, 7, 9, 11, 17, 20, 22, 24] and the references therein.
Asymptotic Behavior of the FitzHugh-Nagumo System
131
In order to include bounded solutions, the weight we choose decays with a polynomial order at innity. A clear drawback of any non-trivial weight is that the weighted norm destroys the transition (in space) invariance and some traveling waves (those connects zero steady-state to other steady-states) approaches zero in a decaying weighted norm as t . Nonetheless, some dynamics of traveling waves are kept in the weighted norm space and, most importantly, the compactness and upper semicontinuity of perturbed global attractors provide a good platform for further study of asymptotic behavior of the dynamics of the FitzHugh-Nagumo system. For example, an important question is how one denes the semi-ow of the slow dynamicsthe limiting system resulting from a time re-scaling by on the compact local attractor to capture the essence of asymptotic behavior of the system for > 0 small. An answer to this question and other related ones will be our further investigation. Another interesting question is what happens if the parameter equals zero. Note that > 0 is a dissipative condition for w. While for = 0 the system still denes a continuous semi-group in the weighted space (Theorem 2.1), our approach does require > 0 for existence of compact global attractors. We believe that a certain dissipative condition for w is needed and would also like to examine the situation more carefully in the future. The paper is organized as follows. In the next section, we describe our main results. In Section 3, the existence and uniqueness of solutions for system (1.1)-(1.2) are proved for R)H 1(I We then extend the solution operator to dene a continuous R). initial data in H 1 (I dynamical system in a weighted Sobolev space. Section 4 is devoted to uniform estimates of solutions as t , which are necessary to establish the point dissipativeness and asymptotic compactness of solutions. Particularly, we show that the solutions are uniformly small when space and time variables are large enough. In Section 5, we prove the existence of a global attractor for system (1.1)-(1.2) in a weighted Sobolev space. We rst establish the asymptotic compactness of the solutions, and then conclude the existence of a global attractor by a standard result. The behavior of the global attractors as 0 is discussed in the last three sections. We will show that not only global attractors are bounded uniformly in (in Section 6) but also their union for small > 0 is contained in a compact subset of the phase space (in Section 7). Finally, in Section 8, we construct a compact local attractor for the limiting system and prove the upper semicontinuity of global attractors as 0.
2.
Main results
Consider the following FitzHugh-Nagumo equations: v t w t = = 2v + h(v) w, x2 (v w), (2.1) (2.2)
for (t, x) (0, ) I with the initial data R v(0, x) = v0 (x), w(0, x) = w0(x), xI R (2.3)
132
where and are positive constants, h is a smooth function that satises, for some positive constants a, and and for all v I R, h(0) = 0, h(v)v av 2 + , h (v) . (2.4)
As a particular example, the cubic function h = v(1 v)(v a) for a > 0 satises (2.4) with 27 a2 a + 1 (a + 1)4 and = . = 256 3 To present our results, we introduce the notation used in the paper. Fix > 1/2 and dene the weighted Sobolev space L2 (I as R) L2 (I = R) with the norm
1/2
u : u is measurable,
(1 + |x|2) |u(x)|2dx <
(1 + |x|2) |u(x)|2dx
1 R) We also denote H (I L2 (I the weighted space with the norm dened by R)
2 1,
= u
u x
2 .
Due to the weight (1+|x|2) , a function in L2 (I can have a certain growth as |x| ; R) R) in particular, the space L2 (I contains all bounded measurable functions such as traveling wave and spatially periodic solutions. For later purpose, we dene a function : (x) = (1 + |x|2) , where = xI R, (2.5)
1 min{1, a} where a is the number in (2.4). 2
(2.6)
Then satises the following inequalities: | and u

2
R, (x)| (x), x I x (x)|u(x)|2dx 2 u

2 .
(2.7)
(2.8)
1/2
(x)|u(x)|2dx Inequality (2.8) shows that the norm is equivalent to the norm 2 (I 2 (I for u L R). As usual, we denote the norm of L R) by . We now describe the main results of the paper and delay their proofs to later sections. We emphasize that we will always assume that h satises (2.4) and > 1/2. Our rst result shows that one can dene a continuous dynamical system for problem (2.1)-(2.3) in the space L2 (I L2 (I R) R). More precisely, we have
133
Theorem 2..1. For > 0, one can associate problem (2.1)-(2.3) with a continuous semigroup S(t)t0 in L2 (I L2 (I such that, when the initial datum (v0, w0) R) R) 1 1 H (I H (I S(t)(v0, w0) is the unique solution of problem (2.1)-(2.3) in the sense R) R), of (3.19)-(3.21). Our next result is on the existence of a global attractor of the dynamical system S(t)t0 . Theorem 2..2. For > 0, the dynamical system S(t)t0 has a global attractor in L2 (I R) 2 (I L R), which is a compact invariant set and attracts every bounded set with respect to the norm topology of L2 (I L2 (I R) R). In the sequel, we denote by A the global attractor corresponding to a given . An interesting question is: what happens to the set A as 0 ? Naturally, one would like to compare the dynamics for > 0 with that for = 0. When = 0, problem (2.1)-(2.3) reduces to v t w t = 2v + h(v) w, x2 (2.9) (2.10)
= 0,
for (t, x) (0, ) I with the initial data R v(0, x) = v0 (x), w(0, x) = w0(x). (2.11)
Given initial data (v0, w0), by (2.10), we see w(t) = w0 for all t 0. Therefore, problem (2.9)-(2.11) cannot have a bounded set that attracts all solutions in the space L2 (I R) 2 L (I R). Interestingly enough, we will show that all global attractors A for small > 0 are 1 R) 1 R)). uniformly bounded in L2 (I L2 (I (in fact, in H (I H (I R) R) Theorem 2..3. The global attractors A are uniformly bounded in for 0 < < a a 1 1 min{1, 2 } in the space H (I H (I R) R). More precisely, for all 0 < < min{1, 2 } and (v, w) A , (2.12) (v, w) M1 , and (v, w) where M1 and M2 are given by M1 = and M2 = 1 ( 2 + 1)e
2 ++2
1,
2 2 M1 + M2 ,
(2.13)
(a2 + 2a + 2) a2
dx,
(2.14)
( + 1) 4(a + 1)(a + 1) + ( 2 + + 2)2 3a2
dx, (2.15)
where a and are the constants in (2.4) and is the constant given by (2.6).
134
The next result shows that the union of all global attractors A is not only bounded, but also compact in L2 (I L2 (I R) R). More precisely, we have the following conclusion. Theorem 2..4. Let
0 a = min{1, 2 }. Then the union
A of all global attractors is

0< <
0
precompact in the space L2 (I L2 (I R) R). Finally, we establish the upper semicontinuity of global attractors at = 0.
Theorem 2..5. There exists a local compact attractor A0 for the limiting system (2.9)-(2.10) in L2 (I L2 (I such that R) R) lim dL2 L2 (A , A0) = 0,
0
where dL2 L2 (A , A0) = sup inf a b

aA bA .
3.
A semigroup associated with problem (2.1)-(2.3)
In this section, we prove Theorem 2..1 to establish the dynamical system S(t)t0 associated with problem (2.1)-(2.3). To this end, we rst consider the FitzHugh-Nagumo equations dened on a bounded interval (L, L) and derive some estimates on solutions uniformly in L. We then take the limit of a convergent subsequence of the solutions dened on (L, L) and show that the limit is a solution of problem (2.1)-(2.3). R) R). Consider the We denote by L = (L, L) and assume (v0 , w0) L2 (I L2 (I following boundary initial value problem on L : v t w t = = 2v + h(v) w, x2 (v w), (3.1) (3.2)
for (t, x) (0, ) L with the initial data v(0, x) = v0(x), and the boundary conditions v(t, x) = 0, w(t, x) = 0, (t, x) (0, ) L . (3.4) w(0, x) = w0(x), x L , (3.3)
It follows from known results (see, e.g., [18]) that problem (3.1)-(3.4) has a unique solution (v, w) C([0, ), L2(L ) L2 (L)). In order to obtain a solution of problem (2.1)-(2.3), we next formally derive the uniform estimates in L for solutions of problem (3.1)-(3.4). The estimates hold for smooth functions and will become rigorous by a limiting process. We begin with the estimates in L2(L ) L2 (L).
135
Lemma 3..1. Suppose (v0, w0) L2 (I L2 (I with (v0 , w0) M . Then, for every R) R) T > 0, there is a constant C depending only on , , T and M (but independent of L), such that, the solution (v, w) of problem (3.1)-(3.4) satises v(t)
L2(L )
+ w(t)
T
L2 (L )
C, C.
t [0, T ],
(3.5) (3.6)
v(t)
0
2 H 1(L ) dt
In the sequel, we denote by C any positive constant which may change values from line to line. Proof. Multiplying (3.1) and (3.2) by v and w, respectively, and integrating with respect to x over L , we obtain 1d 2 dt and |v|2dx +
L L
|x v|2dx =
L
h(v)vdx
L
vwdx,
(3.7)
1 d 2 dt
|w|2dx =
L L
vwdx
L
|w|2dx.
(3.8)
It follows from (3.7) and (3.8) that 1d 2 dt |v|2dx

L
+
L
|w|2dx +
L
|x v|2dx vwdx
L L
=
L
h(v)vdx + ( 1)
|w|2dx. (3.9)
By (2.4), we have the estimates for the rst term on the right-hand side of (3.9): h(v)vdx =
L L
(h(v) h(0)) vdx =

L
h ()|v|2dx
L
|v|2dx.
(3.10)
Substituting (3.10) into (3.9) and dropping the negative term on the right-hand side of (3.9), we nd that there exists a constant C1 depending on and such that, for all t 0: d dt +2
L
|v|2dx +
L L
|w|2dx |v|2dx + |w|2dx .

L
(3.11)
|x v|2dx C1
L
By Gronwalls Lemma, for all t 0, we have |v(t)|2dx +

L L
|w(t)|2dx eC1 t
L
|v0 |2dx +
L
|w0|2dx |w0|2dx ;
I R
eC1 t
I R
|v0|2 dx +
136
in particular, for t [0, T ], the following holds: |v(t)|2dx +

L L
|w(t)|2dx eC1 T M 2.
(3.12)
Integrating (3.11) with respect to t and using (3.12), we nd that there exists a constant C2 depending on , , T and M such that
T
|x v|2dxdt C2 ,
0 L
which and (3.12) conclude the proof. Next, we derive estimates on solutions of problem (3.1)-(3.4) in the space H 1 H 1. Lemma 3..2. Suppose (v0, w0) H 1(I H 1(I with (v0, w0) H 1H 1 M . Then, R) R) for every T > 0, there is a constant C depending only on , , T and M (but independent of L), such that, the solution (v, w) of problem (3.1)-(3.4) satises (v(t), w(t)) and
0 H 1(L )H 1 (L )
C,
w t
T
H 1(L )
C,
t [0, T ],
(3.13)
v(t)
2 H 2(L ) dt
C,
0
v t
2 L2 (L )dt
C.
(3.14)
Proof. Multiplying (3.1) and (3.2) by xx v and xx w, respectively, and integrating with respect to x over L , one has 1d 2 dt and |x v|2dx +
L L
|xxv|2dx =
L
h(v)xxvdx
L
x v x wdx,
1d 2 dt 1 d 2 dt =
L
|x w|2dx =
L L
x v x wdx
L
|xw|2dx.
Therefore, |xv|2dx +
L L
|x w|2dx +
L
|xxv|2dx |x w|2dx.
L
h(v)xxvdx + ( 1)
L
x v x wdx
(3.15)
Using (2.4) to estimate the rst term on the right-hand side of (3.15), we obtain
L
h(v)xxvdx
L
|x v|2dx.
(3.16)
Then, it follows from (3.15) and (3.16) that, for all t 0, d dt |x v|2dx
L
+
L
|xw|2dx + 2
L
|xxv|2dx |x w|2dx ,
L
C1
L
|xv|2dx +
(3.17)
Asymptotic Behavior of the FitzHugh-Nagumo System where C1 is a constant depending on t 0, |xv(t)|2dx +

L L
137
and . By Gronwalls Lemma, we nd that, for
|x w(t)|2dx eC1 t
L
|x v0 |2dx +
L
|x w0|2dx |xw0 |2dx ,

I R
eC1 t
I R
|x v0|2 dx +
which shows that, for t [0, T ], |x v(t)|2dx +

L L
|xw(t)|2dx eC1 T M 2.
(3.18)
Integrating (3.17) with respect to t and using (3.18), we obtain

T
|xxv|2dxdt C,
0 L
which, along with (3.1)-(3.2), (3.18) and Lemma 3..1, implies (3.13) and (3.14). We are now in a position to establish the existence of solutions for problem (2.1)-(2.3). Suppose (v, w) C([0, ), L2(I L2(I R) R)). Then we say (v, w) is a solution of problem (2.1)-(2.3) if, (v(0), w(0)) = (v0, w0) and, for every T > 0, R)), v L2 ((0, T ), H 1(I v w R)), R)), (3.19) L2((0, T ), H 1(I L2 ((0, T ), L2(I t t
and (v, w) satises the following equalities, for any C0 ([0, T ] I R): T 0
v , t
T (H 1,H 1 )
dt +
0 I R
x v x dxdt =
0 T I R
(h(v) w) dxdt,
(3.20)
T 0 I R
w dxdt = t
(v w) dxdt,
0 I R
(3.21)
R) R). where (H 1 , H 1) denotes the duality pairing between H 1 (I and H 1(I Lemma 3..3. Suppose (v0, w0) H 1(I H 1 (I R) R). Then problem (2.1)-(2.3) has a solution (v, w) in the sense of (3.19)-(3.21). In addition, v L ((0, T ), H 1(I L (I R) R)) L2((0, T ), H 2(I R)), v R)), L2((0, T ), L2(I t
w L ((0, T ), H 1(I L (I R) R)),
w L ((0, T ), H 1(I R)). t
Proof. Let be a smooth function satisfying 0 (s) 1 for s 0, and (s) = 1 for 0 s 1 ; 2 (s) = 0 for s 1.
138
Then, given (v0, w0) H 1(I H 1(I we denote by R) R), (v0,L(x), w0,L(x)) = It is evident that (v0,L, w0,L) (v0, w0) in H 1 (I H 1(I as L . R) R) Let (vL, wL) be the solution of problem (3.1)-(3.4) dened on L with the initial datum (v0,L, w0,L). Then, we extend (vL, wL) to I by dening (vL (t, x), wL(t, x)) = (0, 0) for R |x| > L. From Lemma 3..2, for any given T > 0, there is a constant C depending only on , , T and (v0 , w0) H 1H 1 (but independent of L) such that (vL (t), wL(t)) and
0 1 T H 1 (I R)H 1 (I R)
|x|2 |x|2 )v0(x), ( 2 )w0(x) , L2 L
xI R.
C,
wL t
T
H 1 (I R)
C,
t [0, T ],
(3.22)
vL (t)
2 R) H 2(I dt
C,
0
vL t
2 R) L2 (I dt
C.
(3.23)
R) R) By embedding H (I L (I and (3.22), we have vL (t)

L (I R)
+ wL(t)
L (I R)
C,
t [0, T ].
(3.24)
It follows from (3.22)-(3.24) that there exists a pair of functions (v, w) satisfying R) R)) v L ((0, T ), H 1(I L (I L2((0, T ), H 2(I R)), v L2((0, T ), L2(I R)), t (3.25) (3.26)
w R)), L ((0, T ), H 1(I t and a subsequence of (vL , wL) (still denoted by (vL , wL)) such that, as L , R) R)), w L ((0, T ), H 1(I L (I vL v, wL w, vL v R)), weakly in L2 ((0, T ), L2(I t t wL w R)). star-weakly in L ((0, T ), H 1(I t t
Next, we prove (v, w) is a solution of problem (2.1)-(2.3). For that purpose, we take a test R). Then, by (3.1)-(3.2) we have function C0 ([0, T ] I
T 0 I R
vL dxdt + t
T
x vL x dxdt =
0 I R 0 I R
(h(vL ) wL ) dxdt, (3.27)
T wL (vL wL) dxdt. (3.28) dxdt = 0 0 I R t I R We now deal with the nonlinear term on the right-hand side of (3.27). Choose a constant K large enough such that supp() [0, T ] K . Then, by a standard compactness argument (see, e.g., [29]), from (3.25), we infer that
vL v in L2((0, T ), L2(K )).
(3.29)
Asymptotic Behavior of the FitzHugh-Nagumo System Therefore, by (3.24)-(3.25) the following holds for the nonlinear term h:
T T T
139
|
0 I R
h(vL ) dxdt
0 T I R
h(v) dxdt|
0 I R
|(h(vL) h(v)) | dxdt
0 K T
|h ()(vL v)| || dxdt C

0 T K T
|vL v| || dxdt |vL v|2dxdt

0 K 0 K
||2 dxdt (3.30)
0,
as L .
Taking the limits of (3.27)-(3.28) as L , we nd that (v, w) satises

T 0 I R
v dxdt + t
T 0 I R
x v x dxdt =
0 I R T 0 I R
(h(v) w) dxdt,
(3.31)
w dxdt = t
(v w) dxdt,
0 I R
(3.32)
By a standard continuity argument (see, e.g., [28]), from (3.25)-(3.26) we can prove R) R)) and (v(0), w(0)) = (v0 , w0). (v, w) C([0, T ], H 1(I H 1(I Then, by (3.24)-(3.26) and (3.31)-(3.33), Lemma 3..3 follows. Note that for any > 1/2 (in fact for any 0), the standard space L2 (I is R) contained in the weighted space L2 (I R). Thus the solutions obtained in Lemma 3..3 are actually functions in L2 (I L2 (I R) R). Next, we show that the solutions are Lipschitz continuous in the space L2 (I L2 (I R) R). As an immediate consequence, the solutions in Lemma 3..3 are unique. R) R). Then problem (2.1)-(2.3) possesses Lemma 3..4. Suppose (v0 , w0) H 1(I H 1(I a unique solution satisfying (3.19)-(3.21). Furthermore, for every T > 0, there exists a constant C depending only on , , and T such that any two solutions (v1, w1) and (v2, w2) satisfy the inequality, for t [0, T ], v1 (t) v2 (t)
(3.33)
+ w1(t) w2 (t)
C ( v1 (0) v2 (0)
+ w1(0) w2(0) ) . (3.34)
Proof. Let V = v1 v2 and W = w1 w2. By Lemma 3..3, both (v1 , w1) and (v2 , w2) satisfy equations (2.1) and (2.2) in the distribution sense. Therefore, V and W satisfy V t W t = = 2V + (h(v1) h(v2 )) W, x2 (V W ). (3.35) (3.36)
140
Multiplying (3.35) by 1 V where 1 is the function given by (2.5) with = 1 and integrating the resulting identity with respect to x, we obtain 1d 2 dt 1 |V |2dx + 1 |x V |2 = (h(v1 ) h(v2 )) 1 V dx 1 W V dx. (3.37)
x 1 (xV ) V dx
The rst term on the right-hand side of (3.37) can be estimated from (2.4) as (h(v1 ) h(v2 )) 1V dx = h ()1|V |2dx 1 |V |2dx. (3.38)
Using (2.7), the second term on the right-hand side of (3.37) satises | x 1 (x V ) V dx| 1 |x V | |V |dx 1 2 1 1|x V |2dx + 2 2 1 |V |2dx. (3.39) For the last term on the right-hand side of (3.37), we have | 1 W V dx| 1 2 1 |W |2dx + 1 2 1 |V |2 dx. (3.40)
It follows from (3.37)-(3.40) that there is a constant C1 depending on and such that d dt 1 |V |2 dx C1 1 |V |2 dx + 1 |W |2dx . (3.41)
Similarly, multiplying (3.36) by 1 W and integrating, we nd there is a constant C2 depending on such that d dt Therefore, we have d dt 1 |V |2dx + 1 |W |2 C 1 |V |2 dx + 1 |W |2dx . 1 |W |2dx C2 1|V |2dx + 1 |W |2dx . (3.42)
Thus, (3.34) follows from Gronwalls inequality. The proof is complete. Next, we prove Theorem 2..1 and dene a dynamical system for problem (2.1)-(2.3). Proof of Theorem 2..1. Given T > 0, by Lemma 3..4, there is a mapping G from 1 R) H 1(I H 1 (I into C([0, T ], L2 (I L2 (I R) R) R) R)) such that for each (v0, w0 ) H (I 1 R), G(v0, w0) is the unique solution of problem (2.1)-(2.3) in the sense of (3.19)H (I (3.21). It follows from (3.34) that the mapping G is continuous from H 1(I H 1(I R) R) L2 (I R)L2 (I into C([0, T ], L2 (I R) R)L2 (I R)). Since H 1(I R)H 1 (I is a dense subR) set of L2 (I L2 (I G can be extended uniquely to a mapping G from L2 (I L2 (I R) R), R) R) 2 2 into C([0, T ], L(I R)L (I R)). We then dene a semigroup S(t)t0 : L2 (I R)L2 (I R) L2 (I L2 (I such that, for every t 0 and for (v0 , w0) L2 (I L2 (I R) R) R) R), 0 , w0)(t). It follows from (3.34) that S(t)t0 is a continuous semiS(t)(v0, w0) = G(v group.
141
4.
Uniform estimates in time
In this section, we derive a priori estimates on solutions of problem (2.1)-(2.3) as t . We rst establish uniform estimates in time in the space L2 (I L2 (I and then improve R) R) 1 1 the estimates in H (I H (I R) R). Finally, we show that all solutions are uniformly small when space and time variables are sufciently large. Lemma 4..1. Suppose (v0 , w0)
R and let (v(t), w(t)) = S(t)(v0, w0). Then,
(v(t), w(t)) and

t+d
M,
for all t T,
(4.1)
x v(t)
t
2 dt
C(1 + d),
for any t T and d > 0,
(4.2)
where M is a constant depending only on the data ( , , a, , ), T depending on the data ( , , a, , ) and R, while C depending on the data ( , , a, , ), but not on d. Proof. Multiplying (2.1) and (2.2) by v and w, respectively, and integrating with respect to x, we nd 1d 2 dt and 1d 2 dt Therefore, 1d 2 dt |v|2dx +
I R I R
|v|2dx+
I R I R
vx v x dx+
I R
|x v|2dx =
I R
h(v)vdx
I R
wvdx,
|w|2dx +
I R I R
|w|2dx =
I R
vwdx.
|w|2dx
+
I R
vx v x dx +
I R
|x v|2dx h(v)vdx. (4.3)

I R
I R
|w|2dx =
Since h satises (2.4), we have the following estimates for the right-hand side of (4.3): h(v)vdx a
I R I R
|v|2dx +
I R
dx.
(4.4)
By (4.3) and (4.4), we get 1 d 2 dt |v|2dx +

I R I R
|w|2dx
+
I R
|x v|2dx a
I R
|v|2dx +
I R
|w|2dx
I R
dx
I R
vx v x dx. (4.5)
142
Next, we estimate the last term on the right-hand side of (4.5). By (2.6) and (2.7) we have |
I R
vxv x dx|
I R
|v||xv| dx v 2dx +
I R
1 2
v 2 dx +
I R I R
(x v)2dx (4.6)
1 a 4
1 4
(x v)2dx.
I R
Note that the last inequality is obtained by (2.6). From (4.5)-(4.6), it follows that 1 d 2 dt |v|2dx
I R
+
I R
|w|2dx + 1 a 2
2
1 2
|x v|2dx
I R
|v| dx +
I R I dx. R I R
|w|2dx
I R
dx. (4.7)
Let C0 = min{a, 2 } and C1 = 2 d dt |v|2dx +

I R I R
Then from (4.7) we get |v|2dx + |w|2dx C1 , (4.8)

I R
|w|2dx + C0
I R
Applying Gronwalls inequality, by (v0 , w0) |v(t)|2dx +

I R I R
R and (2.8), we obtain |v0|2 dx +

I R I R C0 t
|w(t)|2dx eC0 t
2
|w0|2dx +
2
C1 C0
C1 (4.9) C0 C 2C1 2 R2 max{ , 1}eC0 t + 1 , (4.10) C0 C0 max{ , 1}e (v0, w0) +

1 for t T = C0 (ln(C0 2 R2 max{ , 1}) ln C1 ). Note that (4.10) implies (4.1). Integrating (4.7) with respect to t, (4.2) follows from (4.10). The proof is complete. a Notice that if min{1, 2 }, then the above constant C0 = min{a, 2 } = 2 . In this case, for all t 0, it follows from (4.9) that
|v|2dx +
I R I R
|w|2dx 2 e2
v0
+ w0
dx,
I R
(4.11)
which proves useful to establish uniform bounds of global attractors, independent of small , in Section 6. In the sequel, we denote by B the set: B = (v, w) L2 (I L2 (I : R) R) (v, w)
M ,
(4.12)
where M is the constant in (4.1). Then Lemma 4..1 implies the set B is an absorbing set for the dynamical system S(t)t0 . Since B itself is bounded, it follows from Lemma 4..1 that there exists a constant TB depending on the data ( , , a, , ) such that S(t)B B, for all t TB , (4.13)
which shows that B is positively invariant for large time, and is useful when we investigate the large-time behavior of solutions. 1 R). Next, we estimate the component v of the solution in H (I
143
Lemma 4..2. Suppose (v0 , w0) R and let (v(t), w(t)) = S(t)(v0, w0). Then we have (4.14) x v(t) M, for all t T, where M is a constant depending only on the data ( , , a, , , ), T depending on the data ( , , a, , ) and R. Proof. Multiplying (2.1) by 1 xx v and integrating, we get 1 Note that 1 and 1h(v)xx vdx = h(v)x 1 x vdx 1 x (h(v)) x vdx. (4.17) v 1d xx vdx = t 2 dt 1 |x v|2dx v x 1 x vdx, t (4.16) v xx vdx = t 1 |xx v|2dx + 1 h(v)xxvdx 1 wxxvdx. (4.15)
Then, it follows from (4.15)-(4.17) that 1d 2 dt 1 |xv|2dx+ + By (2.1) we have v h(v) = xx v w. t Using (2.4) and (4.19), from (4.18) we nd 1 d 2 dt 1 |x v|2dx+ + By (2.7) we have | xx v x 1 x vdx| |xx v| |1 | |xv|dx 1 2 1 |xx v|2dx 1|x v|2dx. (4.21) 1 |xx v|2dx wx 1 x vdx + xx v x 1 x vdx 1 |xv|2dx + 1w xx vdx. (4.20) (4.19) 1 |xxv|2dx = h 1|x v|2dx + v h(v) x 1 x vdx t 1 w xx vdx. (4.18)
1 + 2 2
The second and the last term on the right-hand side of (4.20) satisfy | wx 1 x vdx| 1 |1 ||w||xv|dx 2 1 1 w2 dx + 2 1|x v|2dx, (4.22)
144 and |
Weishi Liu and Bixiang Wang 1 2 1|w|2dx + 1 2 1|xx v|2dx.
1 w xx vdx|
(4.23)
By (4.20)-(4.23) we get 1d 2 dt that is, d (4.24) xv 2 2 + + 2 x v 2 + ( + 1) w 2 . dt Applying the uniform Gronwalls Lemma (see, e.g., [28]), by (4.1)-(4.2) we obtain x v 2 C, for all t T + 1, where T is the constant in (4.1). Next, we improve the estimates given by (4.1) for the component w of the solution. Since equation (2.2) has no smoothing effect on w, we cannot obtain an estimate for w in 1 R) the space H (I if the initial datum w0 L2 (I R). Nevertheless, we can prove w is a sum 1 R) of two functions: one is regular in the sense it belongs to H (I and the other converges to zero uniformly as t . This decomposition technique has been used by several authors for the FitzHugh-Nagumo equations (see, for example [18]). We split w as w = w1 + w2 where w1 is the solution of the initial value problem w1 = w1, t and w2 is the solution of w2 = w2 + v, t w2 (0) = 0. (4.26) w1(0) = w0, (4.25) 1 |x v|2dx 1 2 1 + + 2 2 1 |x v|2dx + 1 ( + 1) 2 1 |w|2dx,
It is evident the following energy equation holds for w1: w1
= e
w0
for all t 0,
(4.27)
which implies that w1 converges to zero uniformly in bounded initial data when t . 1 R) Next, we show that w2 is bounded in the space H (I for large time. Lemma 4..3. w2 satises xw2 (t)
M,
for all t T,
(4.28)
where M is a constant depending only on the data ( , , a, , , ), T depending on the data ( , , a, , ) and R when (v0, w0) R. Proof. Multiplying (4.26) by 1 xx w2 and then integrating, we get 1 w2 xx w2 dx = t w2 1xx w2 dx + v1xx w2 dx. (4.29)
Note that the left-hand side of (4.29) can be rewritten as 1 1 d w2 xx w2dx = t 2 dt 1 |xw2 |2dx w2 x 1 x w2dx. t (4.30)
145
We now deal with the right-hand side of (4.29). For the rst term on the right-hand side, we have w2 1xx w2 dx = 1 |xw2 |2dx + w2 x 1 x w2dx. (4.31)
Integrating by parts, we get the identity for the last term on the right-hand side: v1 xx w2dx = By (4.29)-(4.32) we nd 1d 2 dt 1 |x w2|2dx + 1 |x w2|2dx = 1 x vxw2 dx (4.33) 1 x v x w2dx vx 1 x w2dx. (4.32)
w2 + w2 v x 1x w2dx. t
By (4.26), the last term on the right-hand side of (4.33) is zero. While, the rst term on the right can be estimated as follows. 1 x v x w2dx From (4.33) and (4.34) we obtain d dt 1 |x w2|2 + 1 |xw2 |2dx 1|x v|2dx C, for t T, (4.35) 1 2 1 |x w2|2 dx + 1 |x v|2dx. (4.34)
where the last inequality is obtained by (4.14). Then, (4.28) follows from (4.35) and Gronwalls Lemma. In what follows, we derive estimates on solutions for large time and large space variables. Actually, we show that the solutions are uniformly small for large time in the complement of a bounded set, which is crucial for the asymptotic compactness of the dynamical system S(t)t0 associated with problem (2.1)-2.3). The uniform estimates outside a bounded set are stated as follows. Lemma 4..4. For every > 0, there exists T () depending only on , ( , , a, , ) and R when (v0, w0) R, such that, for all t T (), the following inequality holds: 1|v|2 + 1|w|2 dx ,
|x|K()
(4.36)
where K() depends only on and ( , , a, , , ). Proof. Let be a smooth function satisfying 0 (s) 1 for s 0, and (s) = 0 for 0 s 1; (s) = 1 for s 2.
146
Then, there is a positive constant C such that | (s)| C for all s 0. Multiplying (2.1) 2 by ( |x| ) v and then integrating, we nd k2 1d 2 dt ( |x|2 ) |v|2dx = k2 +
2
xx v ( (
|x|2 ) vdx k2 ( |x|2 ) wvdx.(4.37) k2
|x|2 ) h(v)vdx k2
Multiplying (2.2) by ( |x| ) w and integrating, we get k2 1d 2 dt ( |x|2 ) |w|2dx = k2 ( |x|2 ) vwdx k2 ( |x|2 ) |w|2dx k2 (4.38)
By integration by parts, we have the following identity for the rst term on the right-hand side of (4.37): xx v ( |x|2 ) vdx = k2 By (4.37)-(4.39) we obtain 1d 2 dt + = Using (2.4) we have 1d 2 dt |x|2 |x|2 ) |v|2dx + ( 2 ) |w|2dx k2 k 2 |x|2 |x|2 |x| + ( 2 ) |x v|2dx + a ( 2 ) |v|2dx + ( 2 ) |w|2dx k k k |x|2 |x|2 |x|2 ( 2 )v x v x dx v x v x ( 2 ) dx. ( 2 ) dx k k k (4.41) ( |x|2 |x|2 ) |v|2dx + ( 2 ) |w|2dx 2 k k 2 |x|2 |x| ( 2 ) |xv|2dx + ( 2 ) |w|2dx k k 2 |x|2 |x| ( 2 )v x v x dx ( 2 ) h(v)vdx k k 2 |x| v x v x ( 2 ) dx. k ( |x|2 |x|2 ) |x v|2dx ( 2 )v x v x dx k2 k |x|2 (4.39) v x v x ( 2 ) dx. k (
(4.40)
Next, we estimate terms on the right-hand side of (4.41). By (2.7) and (2.6), we have the
Asymptotic Behavior of the FitzHugh-Nagumo System following bounds for the second term on the right-hand side. | ( |x|2 |x|2 )v x v x dx| ( 2 ) |v||xv|dx 2 k k 1 |x|2 ( 2 ) |x v|2dx 4 k |x|2 1 a ( 2 ) |v|2dx. + 4 k
147
(4.42)
By properties of , the last term on the right-hand side of (4.41) is bounded by | |x|2 2x |x|2 v x v ( 2 ) 2 dx| v x v x ( 2 ) dx| = | k k k C2 2 C1 2 C1 x v v , (4.43) | ||v||xv|dx k k k k|x| 2k
for all t T , where the last inequality is obtained by Lemmas 4..1 and 4..2, C2 and T are constants independent of k. By (4.42) and (4.43), it follows from (4.41) that 1d 2 dt + 1 2 |x|2 |x|2 ) |v|2dx + ( 2 ) |w|2dx k2 k 2 |x| |x|2 1 ( 2 ) |x v|2dx + a ( 2 ) |v|2dx + k 2 k C2 |x|2 , for all t T. ( 2 ) dx + k k (
|x|2 ) |w|2dx k2 (4.44)
Let C3 = min{a, 2 }. Then we get from (4.44) that, for all t T , d dt + C3 2 |x|2 |x|2 ) |v|2dx + ( 2 ) |w|2dx k2 k 2 |x|2 |x| ( 2 ) |v|2dx + ( 2 ) |w|2dx k k 2 |x| 2 C2 . ( 2 ) dx + k k (
(4.45)
Next, we manipulate the rst term on the right-hand side of (4.45), which is bounded by |2 ( |x|2 ) dx| 2 k2 (
|x|k
|x|2 ) dx 2 k2
dx.
|x|k
(4.46)
Therefore, by (4.45) we nd there exists a constant K() such that for k K() and t T, d dt +C3 ( |x|2 |x|2 ) |v|2dx + ( 2 ) |w|2dx k2 k 2 |x|2 |x| ( 2 ) |v|2dx + ( 2 ) |w|2dx k k
2 .
(4.47)
148
By Gronwalls Lemma, for all t T and k K(), we get |x|2 |x|2 ) |v(t)|2dx + ( 2 ) |w(t)|2dx k2 k I R I R |x|2 |x|2 2 eC3 (tT ) ( 2 ) |v(T )|2dx + ( 2 ) |w(T )|2dx + k k C3 I R I R 2 C4 eC3 (tT ) + . (4.48) C3 (
C Note the last inequality is obtained by Lemma 4..1. Let T1() = T + 1/C3 ln C3 4 . Then, by (4.48) we obtain, for t T1() and k K(),
(
I R
|x|2 ) |v(t)|2dx + k2
(
I R
|x|2 3 ) |w(t)|2dx , k2 C3
which implies (4.36) since (s) = 1 for s 2. The proof is complete. As an immediate consequence of Lemma 4..4 and (4.27), we have the following estimates for (v, w2), which are useful when we establish the asymptotic compactness of the solution in the next section. Corollary 4..5. For every > 0, there exists T () depending only on , ( , , a, , ) and R when (v0, w0) R, such that, for all t T (), the following inequality holds: 1 |v|2 + 1 |w2|2 dx ,
|x|K()
where K() depends only on and ( , , a, , , ).
5.
Existence of global attractors
In this section, we show that the dynamical system S(t)t0 has a global attractor in the space L2 (I L2 (I R) R). It is known that the global attractor exists if S(t)t0 is asymptotically compact and has a bounded absorbing set. We have already proved that S(t)t0 has a bounded absorbing set, which is given by (4.12). The next lemma establishes the asymptotic compactness of S(t)t0. Lemma 5..1. The dynamical system S(t)t0 is asymptotically compact, that is, if (vn , wn) 2 R) and t , then S(t )(v , w ) is precompact in L2 (I is bounded in L2 (I n n n n R)L (I R) 2 L (I R). Proof. Assume (vn , wn) is bounded in L2 (I L2 (I and tn . We want to prove R) R) the sequence {S(tn )(vn , wn)} has a convergent subsequence in L2 (I L2 (I R) R). To this end, we decompose {S(tn )(vn , wn)} as a sum of two sequences: one is precompact and the other converges to zero as n . Given (v0 , w0) L2 (I L2 (I denote by w1 and R) R), w2 the solutions of problem (4.25) and (4.26), respectively. Then, for t 0, we dene two 2 R) into itself such that S (t)(v , w ) = (0, w (t)) maps S1 (t) and S2(t) from L2 (I 1 0 0 1 R)L (I
149
and S2 (t)(v0, w0) = (v(t), w2(t)) for (v0, w0) L2 (I L2 (I R) R). Thus, S(t)(v0, w0 ) = S1 (t)(v0, w0) + S2 (t)(v0, w0). Particularly, for the sequence {S(tn )(vn, wn )}, we have S(tn )(vn , wn) = S1(tn )(vn , wn) + S2 (tn )(vn , wn). Since (vn , wn) is bounded in L2 (I L2 (I there exists R > 0 such that R) R), (vn , wn) It then follows from (4.27) that S1(tn )(vn , wn)

(5.1)
R,
for n 1.
(5.2)
tn
wn
Re
tn
0,
as n .
(5.3)
Thus, if S2 (tn )(vn , wn) has a convergent subsequence, then so does S(tn )(vn , wn). We now prove the sequence S2 (tn )(vn , wn) is precompact in L2 (I L2 (I that is, for any R) R), > 0, the set {S2 (tn )(vn , wn)} has a nite covering of balls of radii less than . For a positive constant K, lets denote K = {x : |x| K} and c = {x : |x| > K}. K Then by Corollary 4..5, given > 0, there exist K() and T () such that S2(t)(vn , wn)
L2 (c )L2 (c ) K() K()
, 4
for t T ().
(5.4)
Since tn , there exists N1() such that tn T () for n N1. Therefore, S2(tn )(vn , wn)
L2 (c )L2 (c ) K() K()
, 4
for n N1 .
(5.5)
On the other hand, by Lemmas 4..1-4..3, there exist M1 and T1 such that S(t)(vn , wn)
1 R)H 1 (I H (I R)
M1
for t T1.
(5.6)
Take N2 large enough such that tn T1 for n N2. Then it follows from (5.6) that S(tn )(vn , wn)
1 R)H 1 (I H (I R)
M1
for n N2.
(5.7)
Taking (5.3) into account, by (5.7) we nd that there exists M2 such that S2(tn )(vn , wn) Note (5.8) implies S2(tn )(vn , wn)
1 1 H (K() )H (K()) 1 R)H 1 (I H (I R)
M2
for n N2 .
(5.8)
M2
for n N2,
(5.9)
where K() is the constant in (5.4). Since K() is bounded, the norm H (K()) is 1 equivalent to the norm H 1 (K()) . Therefore, by (5.9) we see that there exists M3 such that S2(tn )(vn , wn ) H 1(K())H (K() ) M3 for n N2. (5.10) 1
150
By the compactness of embedding H 1 (K()) L2 (K()), the sequence S2 (tn )(vn , wn) 2 ( 2 ( is precompact in L K()) L K()) and hence in L2 (K()) L2 (K()). This implies that, for the given > 0, {S2(tn )(vn , wn)} has a nite covering in L2 (K()) L2 (K()) of balls of radii less than /4, which along with (5.4) shows that {S2(tn )(vn , wn)} has a nite covering in L2 (I L2 (I of balls of radii less than . In R) R) other words, the sequence {S2 (tn )(vn , wn)} is precompact in the space L2 (I L2 (I R) R). R)L2 (I Therefore, by (5.1) and (5.3), we know {S(tn )(vn , wn)} is precompact in L2 (I R). The proof is complete. We are now ready to complete Theorem 2..2 which is concerned with the existence of a global attractor. Proof of Theorem 2..2. By a standard result (see, for example, [3, 5, 12, 15, 28]), we know the dynamical system S(t)t possesses a global attractor if it is asymptotically compact and has a bounded absorbing set. In our case, the asymptotic compactness of S(t)t0 was established by Lemma 5..1 and the bounded absorbing set was given by (4.12).
6.
Uniform bounds of global attractors in

0. We
In this section, we show that the global attractors A are uniformly bounded as rst derive the uniform bounds in L2 (I L2 (I R) R).
a Lemma 6..1. For any < min{1, 2 } the solution (v, w) of (2.1)-(2.3) satises, for t 0,
w and v
2
2 e2
v0
+ w0
dx,
(6.1)
a(a 2 ) + 2 2 2 e a(a 2 )
v0
+ w0
2(1 + a) a2
dx,
(6.2)
where a and are the constants in (2.4), and is the constant given by (2.6). Furthermore, for t 0,
t+1
x v
t
2 dt
2(a + 2)(a 2 ) + 4 2 2 e 3a(a 2 ) 4(a + 1)(a + 1) dx. 3a2
v0
+ w0
(6.3)
Proof. Multiplying (2.1) by v and integrating with respect to x, we get 1d 2 dt |v|2dx+ |x v|2dx = vx v x dx+ h(v)v vwdx. (6.4)
By inequality (2.4) for h, we obtain from (6.4) that 1 d 2 dt |v|2dx + a |x v|2dx |v|2dx + vx v x dx dx vwdx. (6.5)
151
We now estimate the terms on the right-hand side of (6.5) as follows. By (2.6) and (2.7) we see the rst term on the right-hand side is bounded by | vx v x dx| 1 a 4 |v||xv|dx |v|2dx + 1 4 1 2 |v|2dx + |x v|2dx (6.6)
|x v|2dx.
For the last term on the right-hand side of (6.5), we have the bounds: | vwdx| a 4 a 2 |v|2dx + 1 a |w|2dx. (6.7)
It follows from (6.5)-(6.7) that 1 d 2 dt |v|2dx+ 3 4 |x v|2dx+ |v|2dx dx+ 1 a |w|2dx. (6.8)
a Since < min{1, 2 }, (4.11) implies
|w|2dx 2 e2
I R
v0
+ w0
dx,
for t 0.
(6.9)
By (6.8) and (6.9) we get, for t 0, d dt |v|2dx + 1 a 3 2 |x v|2dx + a

t
|v|2dx v0
2
2 1 +
2 dx + 2 e2 a
+ w0
(6.10)
Dropping the second term on the left side , by Gronwalls inequality we obtain |v|2dx 2 a + Since < eat
a 2
1+
1 a
dx + eat
t
|v0|2 dx
2
2 2 e2 a(a 2 )
v0
+ w0
(6.11)
by assumption, we have
t
|v0 |2dx e2
|v0 |2dx 2 e2
v0
+ w0
(6.12)
It follows from (6.11) and (6.12) that

|v|2dx 2 a 1+ 1 a dx + 1 + 2 a(a 2 ) 2 e2
t
v0
+ w0
(6.13)
By (6.9) and (6.13) we get (6.1) and (6.2), respectively. Next, we prove (6.3). Integrating (6.10) from t to t + 1, we obtain 3 2
t+1
x v
t
2 dt
|v(t)|2dx + 2 1 + 2 + 2 a v0
2
1 a
t+1 t
dx e2
t
+ w0
dt.
(6.14)
152
We now deal with the last term on the right-hand side of (6.14). Notice that
t+1
e2
t
dt e2
for all > 0 and t 0.
(6.15)
Substituting (6.13) and (6.15) into (6.14), then (6.3) follows. The proof is complete. Notice that by adding (6.1) and (6.2) together we get the inequality: S(t)(v0, w0)
2
K1e2
v0
+ w0
+ K2,
for all t 0, (6.16)
where K1 and K2 are given by K1 = 2 (1 + a(a 2 )) 2 (a2 + 2a + 2) , K2 = a(a 2 ) a2 dx. (6.17)
The uniform boundedness of A in L2 (I L2 (I can now be easily obtained. R) R) Proof of Theorem 2..3. Suppose (v, w) A and tn . Then by invariance of A , there exists a sequence of (vn , wn ) A such that (v, w) = S(tn )(vn, wn ) for all n 1. Since A B, where B is the bounded absorbing set given by (4.12), we have (vn , wn )
(6.18)
M,
for all n 1,
(6.19)
where M depends only on the data ( , , a, , ). On the other hand, by applying (6.16) to S(tn )(vn , wn) we get S(tn )(vn , wn)
2
K1e2
tn
(vn , wn)
+ K2 ,
for all n 1,
(6.20)
where K1 and K2 are constants given by (6.17). It follows from (6.19) and (6.20) that S(tn )(vn , wn)
2
K1 M 2e2
tn
+ K2,
for all n 1,
which along with (6.18) implies that for any (v, w) A , the following inequality holds (v, w)
2
K1 M 2e2
tn
+ K2,
for all n 1.
(6.21)
Taking the limit as n , we nd (v, w) 2 K2 from which (2.12) follows. To complete the proof, we also need to establish the estimates claimed in (2.13). In what 1 R) 1 R) follows, we rst derive various uniform bounds in for the solution in H (I H (I via a sequence of lemmas, and then complete the proof of Theorem 2..3 in the end of this section.
a Lemma 6..2. For any < min{1, 2 }, the solution (v, w) of (2.1)-(2.3) satises
x v(t)
K3 e2
v0
+ w0
+ K4 ,
for t 1,
(6.22)
Asymptotic Behavior of the FitzHugh-Nagumo System where K3 and K4 are given by K3 = 2 e2 and K4 = e
2 ++2
153
+ 2 ++2
+1 2(a + 2)(a 2 ) + 4 + 3a(a 2 ) (2 + 2 + + 2)2
, (6.23)
( 2 +
( + 1) 4(a + 1)(a + 1) + 2 + 2) 3a2
dx,
(6.24)
where a and are the constants in (2.4), and is the constant given by (2.6). Proof. We start with inequality (4.24): d x v dt
2
2 + + 2
xv
+ ( + 1) w
2 ,
for t 0.
(6.25)
Substituting (6.1) into (6.25) we get, for all t 0, d x v dt

2
k x v
( + 1) 2 e2
2
v0
+ w0
( + 1)
dx.
(6.26) Here and after, we denote by = + + 2. Let t and s be positive numbers satisfying 0 t s t + 1. Then multiplying (6.26) by et and integrating with respect to t between s and t + 1, we obtain x v(t + 1)
2 (t+1) e t+1 2
es x v(s) + ( + 1)
+ ( + 1) 2
t+1
v0
+ w0
2 s
e(+2
)t
dt (6.27)
dx
s
et dt.
The two integrals on the right-hand side of (6.27) have the estimates
t+1
e(+2
s
)t
dt
1 e(+2 +2
t+1 )s
and
s
et dt
1 s e .
(6.28)
Substituting (6.28) into (6.27), and then multiplying the resulting inequality by e(t+1) we nd xv(t + 1)
2
e(t+1s) x v(s) + + 1 2 +2
+
2
( + 1) (t+1s) e + w0
2
dx
s
v0
e(t+1s)2
(6.29)
Since 0 t s t + 1, integrating (6.29) with respect to s between t and t + 1 we have

t+1
x v(t + 1)
e(t+1s) xv(s) + ( + 1) 2 +2 v0
2
2 ds
( + 1)
t+1 t
t+1
e(t+1s) ds
t s
dx (6.30)
+ w0
e(t+1s)2
ds.
154
We now deal with the right-hand side of (6.30). By (6.3), the rst term on the right-hand side of (6.30) satises
t+1 t+1
e(t+1s) xv(s)
t
2 ds
e
t
x v(s)
2 ds
2(a + 2)(a 2 ) + 4 2 2 t+ e 3a(a 2 ) 4(a + 1)(a + 1) + e dx. 3a2
v0
+ w0
(6.31)
The integrals in the last two terms of the right-hand side of (6.30) can be estimated
t+1
e(t+1s)2
t
ds
1 e2 +2
t+1 t+
and
t
e(t+1s) ds
1 e .
(6.32)
It follows from (6.30)-(6.32) that, for all t 0,

x v(t + 1)
2
+1 2(a + 2)(a 2 ) + 4 2 e2 + 3a(a 2 ) ( + 2 )2 ( + 1) 4(a + 1)(a + 1) + + e dx, 2 3a2
t+
v0
+ w0
which along with = 2 + + 2 implies (6.22). As an immediate consequence of Lemma 6..2, we have the following estimates for global attractors.
a Lemma 6..3. Let 0 < < min{1, 2 }. If (v, w) A , then 2
x v
2 ++2
( 2 +
4(a + 1)(a + 1) ( + 1) + 2 + 2) 3a2
dx,
(6.33)
where a, and are the constants in (2.4) and is the constant given by (2.6). Next, we establish the uniform bounds in
1 R). H (I
for the solution w2 of problem (4.26) in
Lemma 6..4. Let 0 < (4.26) satises: x w2

2
a < min{1, 2 }. If (v0 , w0) A , then the solution w2 of problem
2 ++2
( 2 +
( + 1) 4(a + 1)(a + 1) + 2 + 2) 3a2
dx,
(6.34)
where a, and are the constants in (2.4) and is the constant given by (2.6). Proof. We begin with inequality (4.35): d dt 1 |xw2 |2 + 1|x w2|2 dx 1 |x v|2dx, for t 0. (6.35)
155
Since (v0, w0) A , we know for all t 0, (v(t), w(t)) = S(t)(v0, w0) A . Then it follows from Lemma 6..3 that 1|x v(t)|2dx C, where C is given by C = e
2 ++2
for all t 0,
(6.36)
( 2
( + 1) 4(a + 1)(a + 1) + 2 + + 2) 3a2
dx,
(6.37)
By (6.35) and (6.36) we get d dt 1|x w2|2 + 1 |x w2|2dx C , for t 0. (6.38)
Applying Gronwalls Lemma to (6.38), we obtain 1 |xw2 (t)|2dx Then (6.34) follows from (6.39) and (6.37). As an immediate consequence of Lemmas 6..3 and 6..4, we have the following bounds 1 R) 1 R) in the space H (I H (I for the dynamical system S(t)t0.
a Corollary 6..5. If (v0, w0) A for 0 < < min{1, 2 }, then the solution (v, w) of problem (2.1)-(2.3) and the solution w2 of problem (4.26) satisfy:
C , 2
for all t 0.
(6.39)
x v(t) where K5 is given by K5 = ( 2 + 1)e 2

2 ++2
+ x w2(t)
K5 ,
for all t 0,
(6.40)
4(a + 1)(a + 1) ( + 1) + ( 2 + + 2)2 3a2
dx, (6.41)
where a, and are the constants in (2.4) and is the constant given by (2.6). We are now ready to complete the proof of Theorem 2..3. Proof of Theorem 2..3 (Continued). By invariance of A , there exist tn and (vn , wn) A such that (v, w) = S(tn )(vn , wn ), Decompose S(t) as in (5.1): S(t)(v0, w0) = S1 (t)(v0, w0)+S2 (t)(v0, w0),
2 for (v0 , w0) L2 (I R), (6.43) R)L (I
for all n 1.
(6.42)
where S1 (t)(v0, w0) = (0, w1(t)), S2 (t)(v0, w0) = (v(t), w2(t)), w1 and w2 are solutions of problem (4.25) and problem (4.26), respectively. It follows from (6.42) and (6.43) that (v, w) = S1 (tn )(vn , wn) + S2(tn )(vn , wn), for all n 1. (6.44)
156 Since (vn , wn) A , we nd
(vn , wn)
C,
for all n 1,
(6.45)
where C is the constant on the right-hand side of (2.12). By (4.27) and (6.45) we get S1 (tn )(vn , wn) On the other hand, S2(tn )(vn, wn ) where K6 and K7 are given by K6 = and K7 = 1 ( 2 + 1)e
2 ++2 1 H
tn
wn
Ce
tn
0,
as n .
(6.46)
2 2 K6 + K7 ,
(6.47)
(a2 + 2a + 2) a2
dx,
(6.48)
( 2 +
( + 1) 4(a + 1)(a + 1) + 2 + 2) 3a2
dx.
(6.49) 1 R) 1 R). ThereBy (6.47) we see the sequence {S2 (tn )(vn , wn)} is bounded in H (I H (I 1 1 fore there exists (, w) H (I H (I such that v R) R)
1 1 v R) R). S2(tn )(vn , wn) (, w) weakly in H (I H (I
(6.50)
Furthermore, by (6.47) and (6.50) we have (, w) v

1 1 H H
lim inf S2(tn )(vn , wn)

n
1 1 H H
2 2 K6 + K7 .
(6.51)
For any (f, g) L2 (I L2 (I it follows from (6.44) that R) R), (v, w), (f, g) = S1 (tn )(vn , wn), (f, g) + S2(tn )(vn , wn), (f, g) , (6.52)
where , denotes the inner product of L2 (I L2 (I R) R). Taking the weak limit of (6.52) in L2 (I L2 (I as n , by (6.46) and (6.50) we get R) R) (v, w), (f, g) = (, w), (f, g) , v By (6.53) we nd
1 1 (v, w) = (, w) H (I H (I v R) R).
for all (f, g) L2 (I L2 (I R) R).
(6.53)
(6.54)
It follows from (6.51) and (6.54) that (v, w) (6.48) and (6.49), concludes the proof.
1 1 H H
2 2 K6 + K7 , which, together with
157
7.
Compactness of the union of global attractors
In this section, we show that all global attractors are contained in a compact subset of L2 (I L2 (I The idea is to prove that the functions in global attractors are uniformly R) R). small in when the space variable is sufciently large.
a Lemma 7..1. Suppose 0 < < min{1, 2 }. Then for every > 0, there exist a constant K() depending only on and the data (, a, , , ), but independent of , such that for all (v, w) A , the following inequality holds:
1|v|2dx + 1 |w|2dx .
|x|K()
(7.1)
Proof. Let tn . Since (v, w) A , there exists a sequence of (vn , wn) A such that (v, w) = S(tn )(vn , wn), for n 1. (7.2)
Set (vn (t), wn(t)) = S(t)(vn, wn ). Then Theorem 2..3 implies that there exists a constant C independent of such that for all n 1 and t 0: (vn (t), wn(t))
1 1 H H
C.
(7.3)
Proceeding as the proof of Lemma 4..4 but using (7.3) instead of Lemmas 4..1 and 4..2, we nd that, given > 0, there exists a constant K1() depending only on and the data (, a, , , ) such that for k K1() and t 0: d dt +2 ( |x|2 |x|2 ) |vn (t)|2 dx + ( 2 ) |wn(t)|2dx k2 k 2 |x| |x|2 ( 2 ) |vn (t)|2 dx + ( 2 ) |wn(t)|2dx k k
a 2 .
2 , (7.4) By Gronwalls
which is the analogue of (4.47) with C3 = min{a, 2 } = 2 since < Lemma, we can verify from (7.4) that, for t 0 and k 2K1(): |wn (t)|2dx C1 e2
|x|k t
(7.5)
Next, we derive the estimates similar to (7.5) for vn (t). For that purpose, multiplying (2.1) 2 by ( |x| ) vn (t) we get k2 1d 2 dt + ( |x|2 |x|2 ) |vn (t)|2dx = xx vn (t) ( 2 ) vn (t)dx k2 k 2 |x|2 |x| ( 2 ) h(vn (t))vn (t)dx ( 2 ) wn (t)vn (t)dx. k k
(7.6)
By Youngs inequality and (7.5), we obtain the following bound for the last term on the right-hand side of (7.6). | ( |x|2 ) wn (t)vn (t)dx| k2 a 4 a 4 |x|2 ) |vn (t)|2 dx + k2 |x|2 ( 2 ) |vn (t)|2 dx + k ( |x|2 1 ( 2 ) |wn (t)|2dx a k C1 2 t e , (7.7) + a a
158
for all t 0 and k 2K1 (), where C1 is independent of and the last inequality is obtained by (7.5). Dealing with the rst two terms on the right-hand side of (7.6) as the proof of Lemma 4..4 but using (7.3) instead of Lemmas 4..1 and 4..2, after detailed computations, by (7.7) we nd that there exists K2 () such that for t 0 and k K2(): d dt ( |x|2 ) |vn (t)|2 dx + a k2 ( |x|2 ) |vn (t)|2dx k2 1+ 2 a + 2C1 2 e a
t
By Gronwalls Lemma, for all t 0 and k K2(), we get: |x|2 2 ) |vn (t)|2 dx 1 + 2 k a I R which implies for t 0 and k 2K2 (): ( |vn (t)|2dx
|x|k
2C1 + e2 a a(a 2 )
1+
2 a
2C1 + e2 a a(a 2 )
(7.8)
Let K3 () = 2 max{K1(), K2()}. Then it follows from (7.2), (7.5) and (7.8) that for all n 1 and k K3 (): |w|2dx =
|x|k |x|k
|wn(tn )|2dx C1 e2
tn
(7.9)
and |v|2dx =
|x|k |x|k
|vn (tn )|2dx
1+
2 a
2C1 + e2 a a(a 2 )
tn
. (7.10)
Taking the limits of (7.9) and (7.10) as n , then (7.1) follows. We now establish the compactness of the union of all global attractors. A has a nite Proof of Theorem 2..4. Given > 0, we want to show that the set
0< <
0
covering of balls of radii less than . For a positive constant K, denote by K = {x : |x| K} and c = {x : |x| > K}. K Lemma 7..1 implies that there exists a constant K() depending only on and the data A: (, a, , , ) such that for all (v, w)
0< <
0
(v, w)
L2 (c )L2 (c ) K() K()
. 4
(7.11)
On the other hand, by Theorem 2..3, the set

0< <
0
1 R) 1 R), A is bounded in H (I H (I
1 1 1 and hence bounded in H (K()) H (K()). Since the embedding H (K()) 1( K()) L2 (K()) L2 (K()) is compact, we nd that, for the given , the set H A has a nite covering of balls of radii less than /4 in L2 (K()) L2 (K()), 0< <
0
which and (7.11) imply that the set

0< <
0
A has a nite covering of balls of radii less than
in L2 (I L2 (I as desired. R) R),
159
8.
Upper semicontinuity of global attractors
The upper semicontinuity of global attractors as 0 is established in this section. Note that the limiting system (2.9)-(2.11) has no global attractors in L2 (I L2 (I R) R). But we will construct a local attractor A0 for the limiting system such that A is upper semicontinuous to A0 as 0. In order to dene a local attractor for the limiting system, we denote by B the limiting set of A , that is,
0< <
0
B = {(v, w) L2 L2 : there are (v n , w n ) A such that (v n , w n ) (v, w)}. Let W be the projection of B to the w-space:
with
W = {w L2 (I : there is v L2 (I such that (v, w) B}. R) R) Then it follows from Theorem 2..4 that B is compact in L2 (I L2 (I R) R), and therefore 2 (I W is compact in L R). The following lemma states that the limiting system has a local attractor in L2 (I W. R) Lemma 8..1. The limiting system (2.9)-(2.10) has a local attractor A0 which is a compact invariant set and attracts every bounded subset of L2 (I W with respect to the norm R) topology. Further, if we denote by Aw the global attractor of equation (2.9) for a xed w W, then A0 can be characterized by A0 = {(v, w) L2 (I L2 (I : w W and v Aw }. R) R) Proof. The proof is similar but simpler than Theorem 2..2, and the details are omitted. The relationships between the solutions of system (2.1)-(2.2) and system (2.9)-(2.10) are necessary for proving the upper semicontinuity of global attractors. In this respect, we have
Lemma 8..2. Suppose 0 < < min{1, 2 }. If (v , w ) is the solution of system (2.1)-(2.2) R) R), and (v, w) is the solution of system with the initial datum (v1 , w1) L2 (I L2 (I 2 (I (2.9)-(2.10) with the initial datum (v2, w2) L R) L2 (I then we have, for t 0, R),
v (t) v(t) eCt v1 v 2

2
2 2
+ w (t) w(t)
t
+ w1 w2
+
0
eC(ts) v(s) w(s) 2 ds,
where C is a constant depending on and , but independent of . Proof. Let V = v v and W = w w. Then it follows from (2.1)-(2.2) and (2.9)-(2.10) that 2V dV = h(v ) h(v) W, (8.1) dt x2
160
dW = (V W ) + (v w). (8.2) dt Similar to the proof of Lemma 3..4, by (8.1)-(8.2) we can verify that V and W satisfy d V 2 + W 2 C V 2 + W 2 + v w dt Then Lemma 8..2 follows from Gronwalls lemma and inequality (8.3).
2 .
(8.3)
To distinguish the semigroups associated with the perturbed and the unperturbed system, we denote by S (t)t0 the semigroup for problem (2.1)-(2.3) with > 0, and S(t)t0 for = 0. By a continuity argument, it follows from Lemma 8.2 that if n 0 and (v n , w n ) (v, w) in L2 (I L2 (I then R) R), S n (t)(v n , w n ) S(t)(v, w) in L2 (I L2 (I R) R). The following lemma is crucial for establishing the upper semicontinuity of attractors. Lemma 8..3. Let (v n , w n ) A n and n 0. Then there exists (v0 , w0) A0 and a subsequence (v nk , w nk ) of (v n , w n ) such that (v nk , w nk ) (v0 , w0) in L2 (I R) 2 L (I R).
(8.4)
Proof. It follows from Theorem 2..4 that the set

n=1
is precompact. Therefore, there
exist (v0 , w0) L2 (I L2 (I and a subsequence of (v n , w n ) (still denoted by R) R) (v n , w n )) such that (v n , w n ) (v0 , w0) in L2 (I L2 (I R) R). Take a sequence {tm } with tm . By the invariance of A m=1 S n (tm )(v n , w n ) A
n n
(8.5) we know
for all n, m 1.
Therefore, for m = 1, it follows from the precompactness of

n=1
that there exist
(v1, w1) and a subsequence (v S

n1
n1
,w
n1
) of (v n , w n ) such that
(t1 )(v
n1
,w
n1
) (v1 , w1) in L2 (I L2 (I R) R) as n1 .
n2
Similarly, for m = 2, there exist (v2, w2) and a subsequence (v such that S
n2
,w
n2
) of (v
n1
,w
n1
(t2 )(v
n2
,w
n2
) (v2 , w2) in L2 (I L2 (I R) R) as n2 .
Repeating this procedure for all m 1, it follows from a standard diagonal argument that there exists a subsequence (v nk , w nk ) of (v n , w n ) such that, for all m 1, S
nk
(tm )(v
nk
,w
nk
) (vm , wm) in L2 (I L2 (I R) R) as k .
(8.6)
By (8.4) and (8.6) we nd that, as k : (v

nk
,w
nk
)=S
nk
(tm ) S
nk
(tm )(v
nk
,w
nk
) S(tm )(vm, wm).
(8.7)
It follows from (8.5) and (8.7) that, for all m 1, (v0 , w0) = S(tm )(vm , wm), which, along with tm , implies (v0 , w0) A0 . The proof is complete.
161
We are now in a position to prove that the global attractors A are upper semicontinuous at = 0. Proof of Theorem 2..5. We argue by contradiction. If Theorem 2..5 is not true, then there exist > 0 and a sequence (v n , w n ) A n with n 0 such that dL2 L2 ((v n , w n ), A0) . On the other hand, it follows from Lemma 8..3 that there exists a subsequence (v of (v n , w n ) such that dL2 L2 (v
nk nk
(8.8) ,w
nk
,w
nk
), A0 = 0,
a contradiction, which concludes the proof. For simplicity, we only discuss the one-dimensional FitzHugh-Nagumo system in the paper. It is evident that all estimates obtained in Sections 4 and 6 are also valid even the space dimensions are higher than one, as long as the nonlinear function h satises (2.4). This means that if there is a dynamical system associated to problem (2.1)-(2.3), and condition (2.4) is satised, then Theorems 2..2-2..5 still hold regardless of the space dimensions.
References
[1] J. Alexander, R. Gardner, and C. Jones. A topological invariant arising in the stability analysis of traveling waves. J. Reine Angew. Math. 410 (1990), 167-212. [2] F. Antoci and M. Prizzi. Attractors and global averaging of non-autonomous ReactionDiffusion equations in I N . Topological Methods in Nonlinear Analysis , 20 (2002), R 229-259. [3] A.V. Babin and M.I. Vishik. Attractors of Evolution Equations . North Holland, Amsterdam, 1991. [4] A.V. Babin and M.I. Vishik. Attractors of partial differential evolution equations in an unbounded domain. Proceedings Royal Society Edinburgh 116A (1990), 221-243. [5] J.M. Ball. Continuity properties and global attractors of generalized semiows and the Navier-Stokes equations. J. Nonlinear Sci. 7 (1997), 475-502. [6] J. Bell. Some threshold results for models of myelinated nerves. Mathematical Biosciences 54 (1981), 181-190. [7] A.N. Carvalho and T. Dlotko. Partially dissipative systems in locally uniform spaces. Cadernos de Matematica 02 (2001), 291-307. [8] B. Deng. The existence of innitely many traveling front and back waves in the FitzHugh-Nagumo equations. SIAM J. Math. Anal. 22(1991), 1631-1650. [9] E. Feireisl, Ph. Laurencot, and F. Simondon. Global attractors for degenerate parabolic equations on unbounded domains. J. Differential Equations 129 (1996), 239-261.
162
[10] R. FitzHugh. Impulses and physiological states in theoretical models of nerve membrane. Biophys. J. 1 (1961), 445-466. [11] O. Goubet and R. Rosa. Asymptotic smoothing and the global attractor of a weakly damped KdV equation on the real line. J. Differential Equations 185 (2002), 25-53. [12] J.K. Hale. Asymptotic Behaviour of Dissipative Systems. Providence R.I., 1988. [13] C. Jones. Stability of the traveling wave solution of the FitzHugh-Nagumo system. Trans. Amer. Math. Soc. 286 (1984), 431-469. [14] C. Jones, N. Kopell, and R. Langer. Construction of the FitzHugh-Nagumo pulse using differential forms. Patterns and dynamics in reactive media (Minneapolis, MN, 1989), 101-115, IMA Vol. Math. Appl. 37, Springer, New York, 1991. [15] O. Ladyzhenskaya, Attractors for semigroups and evolution equations . Cambridge University Press, Cambridge, 1991. [16] W. Liu and E. Van Vleck. Turning points and traveling waves in FitzHugh-Nagumo equations. J. Differential Equations (to appear). [17] K. Lu and B. Wang. Global attractors for the Klein-Gordon-Schr dinger equation in o unbounded domains. J. Differential Equations 170 (2001), 281316. [18] M. Marion. Finite-dimensional attractors associated with partly dissipative reactiondiffusion systems. SIAM J. Math. Anal. 20 (1989), 816-844. [19] M. Marion. Inertial manifolds associated to partly dissipative reaction-diffusion systems. J. Math. Anal. Appl. 143 (1989), 295-326. [20] I. Moise, R. Rosa, and X. Wang. Attractors for non-compact semigroups via energy equations. Nonlinearity 11 (1998), 1369-1393. [21] J. Nagumo, S. Arimoto, and S. Yosimzawa, An active pulse transmission line simulating nerve axon. Proc. J. R. E. 50 (1964), 2061-2070. [22] M. Prizzi. A remark on reaction-diffusion equations in unbounded domains. Discrete Contin. Dynam. Syst. 9 (2003), 281-286. [23] A. Rodriguez-Bernal and B. Wang. Attractors for partly dissipative reaction diffusion systems in I n . J. Math. Anal. Appl. 252 (2000), 790-803. R [24] R. Rosa. The global attractor for the 2D Navier-Stokes ow on some unbounded domains. Nonl. Anal. TMA 32 (1998), 71-85. [25] B. Sandstede. Stability of multiple-pulse solutions. Trans. Amer. Math. Soc. 350 (1998), 429472. [26] B. Sandstede. Stability of N -fronts bifurcating from a twisted heteroclinic loop and an application to the FitzHugh-Nagumo equation. SIAM J. Math. Anal. 29 (1998), 183 207.
163
[27] Z. Shao. Existence of inertial manifolds for partly dissipative reaction diffusion systems in higher space dimensions. J. Differential Equations 144 (1998), 1-43. [28] R. Temam. Innite Dimensional Dynamical Systems in Mechanics and Physics (2nd Ed.). Springer-Verlag, New York, 1997. [29] R. Temam, Navier-Stokes Equations, Theory and Numerical Analysis (3rd Ed.). North-Holland, Amsterdam, 1984. [30] G. J. Wang and J. Beaumont. Parameter estimation of the Hodgkin-Huxley gating model: an inversion procedure. SIAM J. Appld. Math. 64 (2004), 1249-1267. [31] B. Wang. Attractors for reaction diffusion equations in unbounded domains, Physica D 128 (1999), 41-52.
A PARTIAL D IFFERENTIAL E QUATION WITH N ONAUTONOMOUS PAST D ELAY IN L1 - PHASE S PACE

S. Boulite,1 G. Fragnelli,2 M. Halloumi1 and L. Maniar1 1 University Cadi Ayyad, Faculty of Sciences Semlalia, Department of Mathematics, B.P. 2390, 40000 Marrakesh, Morocco. 2 Dipartimento di Ingegneria dellInformazione, Universit` degli Studi di Siena, via Roma 56, 53100 Siena, Italy a
Abstract
In this paper, we consider partial differential equations with nonautonomous past delay in L1 -phase space . Here, we consider general delay operators given by Stieltjes integrals. Using the Miyadera-Voigt perturbation result, we show the wellposedness of these equations. We study also the robustness of some asymptotic properties, as asymptotic stability, almost periodicity and almost automorphy, under the nonautonomous past delay effect. As an application, we give a dynamical population equation.
AMS Subject Classication: 34G10, 35B15, 35B20, 35B40, 47D06, 47N60 Key Words: Partial differential equations with delay, Miyadera-Voigt perturbation, semigroup, asymptotic stability, almost periodicity, almost automorphy, population equation
E-mail address: sboulite@ucam.ca.ma E-mail address:fragnelli@dii.unisi.it E-mail address: Halloumi@menara.ma E-mail address: maniar@ucam.ac.ma
166
S. Boulite, G. Fragnelli, M. Halloumi and L. Maniar
1.
Introduction
After their introduction by R. Nagel and S. Brendle [5], partial differential equations with nonautonomous past delay u u (t) = Bu(t) + (t), u(0) = x, u0 = g, t 0, (1.1)
have attracted many interest. Here the operator (B, D(B)) is a generator of a strongly continuous semigroup (S(t))t0 in a Banach space X, the delay r +, and : D() Lp ([r, 0], X) X, p 1, is an unbounded linear operator. Moreover, the modied history function ut is dened by ut(s) := U (s, 0)u(t + s) U (s, s + t)g(t + s) if t + s > 0, if t + s 0,
where (U (t, s))rts0 is an exponentially bounded backward evolution family on X. In the case p > 1, this equation was studied in [14, 16] with general delay operators following the study done in [1, 2] for autonomous past delay (i.e., U (, ) = Id). More precisely, the authors have considered the delay operator
0
(f ) :=
r
d()f (),
(1.2)
where : [r, 0] L(X) is a function of bounded variation. In particular, in these papers, they have studied the wellposedness of (1.1), by showing that the delay operator is a small Miyadera-Voigt perturbation. More recently, in [15, 16, 22], the authors have considered the critical case p = 1, but with the particular unbounded delay operator
0
(f ) :=
r
(s)f (s)ds,
L1([r, 0]).
(1.3)
They have showed that these operators are also small Miyadera-Voigt perturbations; thus, by the Miyadera-Voigt perturbation Theorem [17, 24], the wellposedness of (1.1) follows. In this paper, we consider the case p = 1 and the general linear delay operator given in (1.2). In particular, in Section 3, by using the same idea as in [25], we show that this operator is still a small Miyadera-Voigt perturbation, and this yields the wellposedness of the delay equation (1.1). Moreover, in Section 4, we study the persistence of some asymptotic properties, as the asymptotic stability, the almost periodicity, the almost automorphy and the ergodicity, under the nonautonomous past delay effect. More exactly, we give conditions under which the trajectories of the semigroup (S(t))t0 and the modied history functions t ut (and then the mild solution t u(t) X) have the same asymptotic properties. These results have been obtained for autonomous past delay by Casarino-Maniar-Piazzera [6, 7, 8] in the nite delay case. Here, assuming the exponential stability of the backward evolution family U (, ) (which cannot happen in the autonomous past delay case, since U (, ) = Id), we
A Partial Differential Equation with Nonautonomous Past Delay
167
are able to show this also in the innite delay case. These asymptotic properties have been extensively studied, see, e.g., [4, 10, 12, 19, 20] Finally, in Section 5, we illustrate our abstract results by the dynamical population equation r u (t, x) = N u(t, x) du(t, x) + 0 b(a)v(t, a, x)da b1 v(t, r, x), t 0, x , v (t, a, x) = v(t, a, x) + v(t, a, x) dv(t, a, x) b(a)v(t, a, x), t 0, D a x , 0 a r, v(t, 0, x) = f (x)u(t, x), t 0, x , (1.4) where u(t, x) is the density of the population at time t and position x , and v(t, a, x) is the density of the subpopulation of pregnant individuals with time of gestation a, that at time t is in position x. This population equation with b1 = 0 was studied in [15], by transforming it in a partial differential equation with nonautonomous past delay (1.1), where the delay operator is of the form (1.3). Here, the term b1 v(t, r, x) represents a foreign action on the new born individuals which were born at the time of gestation r. For more explanation of the biological signication of this equation and precise data, see Section 5 and also [15]. We show how to write (1.4) as a partial differential equation with nonautonomous past delay, with a delay operator of the form (1.2) and then we study its wellposedness and its asymptotic behavior.
2.
Preliminaries
Recall some denitions and basic results needed for the elaboration of this work. Denition 2..1. A family (U (t, s))ts,t,sI of bounded linear operators on a Banach space X is called an (exponentially bounded backward) evolution family if (i) U (t, )U (, s) = U (t, s), U (t, t) = Id for all t s, t, , s I, (ii) the mapping (t, s) U (t, s) is strongly continuous, (iii) U (t, s) M ew(st) for some M 1, w R and all t s, t, s I.
Here I = [r, 0] or I = (, 0]. We will use evolution semigroup techniques for which we refer to [9, 21]. To that purpose, we extend (U (t, s))ts,t,sI to an evolution family (U (t, s))ts on R. Denition 2..2. (1) The evolution family (U (t, s))ts,t,sI on X is extended to an evo lution family (U(t, s))ts by setting U (r, r), t s < r, (if r < ) U (r, s), t < r, s I, (if r < ) (t, s) := U (t, s), t s, t, s I, U U (t, 0), t 0 s, U (0, 0), 0 t s.
168
(2) On the space E := Lp (R, X), we dene the corresponding evolution semigroup (T (t))t0 by (T (t)f )(s) := U (s, s + t)f (s + t), f E, s R, t 0. It is easy to prove that the semigroup (T (t))t0 is strongly continuous on E . We denote D(G)), where D(G) is a dense subset of C0 (R, X). its generator by (G, Since (G, D( G)) is a local operator see [23, Theorem 2.4], we can restrict it to the space p (I, X) by the following denition. E := L Denition 2..3. Take D(G) := {f|I : f D(G)} and dene Gf := (Gf )|I for f = f|I D(G). This operator G is not a generator on E. However, if one identies E with the subspace {f E : f (s) = 0, s I}, then E remains invariant under (T (t))t0. As a consequence / we obtain the following lemma. Lemma 2..1. The semigroup (T0(t))t0 induced by (T (t))t0 on E is (T0(t)f )(s) = 0, s + t > 0, U (s, s + t)f (s + t), s + t 0, (2.5)
and its generator is given by G0 = G, D(G0) = {f D(G) : f (0) = 0}. We recall also some needed asymptotic properties for functions. Let J be an interval of I Let BC(J, X) be the Banach space of all bounded continuous functions from J R. to X, endowed with the uniform norm, and BU C(J, X) the Banach space of all bounded uniformly continuous functions from J to X. A function g BC(R, X) is said to be almost periodic if for every > 0 there exists l( ) > 0 such that for every a I the R interval [a, a + l( )] contains some for which the inequality g(t + ) g(t) holds for all t I We note that g AP (I X). R. R, A function f BC(R+ , X) is called asymptotically almost periodic if there exist a unique g AP (I X) and a unique h C0(R+ , X) such that f = h + g|R+ . R, A function f BU C(R+ , X) is said to be (uniformly) ergodic if the limit
+ 0
lim
0
et f ( + t)dt
exists in BU C(R+ , X), and totally uniformly ergodic if the function ei f () is uniformly ergodic for all R. A function f BC(R, X) is called almost automorphic (f AA(I X)) if for every R, R sequence (n )nN in I there exists a subsequence (sn )nN (n )nN such that
n,m+
lim
f (t + sn sm ) = f (t)
for each t I R.
This is equivalent to g(t) := lim f (t + sn ) and f (t) = lim g(t sn )

n+ n+
169
are well dened for each t I The function g is measurable, but not necessarily conR. tinuous. Clearly, if the convergence in the denition above is uniform in t I then R, f AP (I X). That is AP (I X) AA(I X). R, R, R, A function f BC(R+ , X) is called asymptotically almost automorphic if there exist a unique g AP (I X) and a unique h C0 (R+ , X) such that f = h + g|R+ . R,
3.
Wellposedness of partial differential equations with nonautonomous past delay
Consider the abstract equation with nonautonomous past delay u (t) = Bu(t) + (t), t 0, u u(0) = x, u0 = g L1 ([r, 0], X), (3.6)
where (B, D(B)) generates a C0 -semigroup (S(t))t0 on a Banach space X, r + and the delay operator : D() L1 ([r, 0], X) X is linear. The modied history function ut is dened as ut (s) := U (s, 0)u(t + s) U (s, s + t)g(t + s) if t + s > 0, if t + s 0,
where U := (U (t, s))rts0 is an exponentially bounded backward evolution family on X (with constants , M ). It is shown in [1, 14, 16] that the wellposedness of (3.6) is equivalent to show that the operator matrix A := on the domain D(A) := x f D(B) D(G) : f (0) = x B 0 G
generates a strongly continuous semigroup in the product Banach space E1 = X L1 ([r, 0], X). Hence, we concentrate now on the last fact. To obtain this claim, we use Miyadera-Voigt perturbation result. For this, we write A in the form A = A0 + B := B 0 0 + 0 G 0 0
with domain D(A0) = D(A) and B : X D() E1 . In [14], see also [22] for nite delay, it was shown that the operator (A0 , D(A0)) generates a strongly continuous semigroup (T0 (t))t0 on E1 given by T0 (t) = S(t) 0 , St T0 (t) t 0,
170
where St : X L1([r, 0], X) is given by (St x)() = U (, 0)S(t + )x 0 for + t > 0, for + t 0,
and (T0(t))t0 is given by (2.5). By using the Miyadera-Voigt perturbation Theorem [24, Theorem 2.2], we have the following main result. Theorem 3..1. Assume that is bounded linear from C([r, 0], X) L1 ([r, 0], X) to X satisfying the following condition: there exist t0 > 0, q1 ]0, 1[ and c1 > 0 such that
t0
(M )
0
(Stx + T0(t)f ) dt q1 x + c1 f
x for all f D(A). Then, the matrix operator A generates a strongly continuous semigroup (T (t))t0 on E1 satisfying the variation of constants formula
T (t) ( x ) = T0 (t) ( x ) + f f
0 x for all ( f ) D(A) and t 0.
T0(t s)BT (s) ( x ) ds f
(3.7)
x Proof. Let c > c1, and the new norm f c = x + c f to the original norm E1 . Hence, by (M ) t0 0
L1
on E1, which is equivalent
BT0(t) ( f )
t0 c
=
0
(Stx + T0(t)f ) dt
q1 x + c1 f L1 c1 x (f ) max q1 , c Setting q = max q1 , c1 , we have q < 1 and c

t0 0 x BT0 (t) ( f ) c
c.
q (x) f
c.
(3.8)
Since A0 is a generator of a strongly continuous semigroup on (E1, E1 ), and being E1 equivalent to c , one has that A0 is a generator on (E1, c ). By assumptions, B L(D(A0), (E1, c )) and, by (3.8), B is a small Miyadera-Voigt perturbation of the semigroup T0 . Hence, the Miyadera-Voigt perturbation Theorem implies that also A = A0 + B generates a strongly continuous semigroup on E1 and satises (3.7). Observe that, for r = +, C([, 0], X) denotes the space C0((, 0], X)( i.e. the space of continuous functions vanishing at ). Remark 3..1. In the references [1, 2, 15, 16, 22], the above result was shown under a different Miyadera-Voigt condition (M ) : there is t0 > 0, q ]0, 1[ such that
t0 0
(Stx + T0(t)f ) dt q ( x ) f
E1 .
The Miyadera-Voigt condition (M ) has appeared for the rst time in [25] in the autonomous past delay case.
171
In the following lemma we give a class of operators which satises the MiyaderaVoigt assumption (M ). Lemma 3..2. Let r and : [r, 0] L(X) be of bounded variation such that ||([r, 0]) < , where || is the positive Borel measure of [r, 0] dened by the total variation on . Let : C([r, 0], X) L1([r, 0], X) X the operator given by
0
(f ) :=
r
d()f ().
(3.9)
Then, fullls the condition (M ). Proof. Let 0 < t0 < r. We have

t0 t0 0
(Stx + T0(t)f ) dt =
0 t0 0 0 r
d()(Stx + T0(t)f )() dt

t0 t
0 t0 t 0 t
d()U (, 0)S( + t)x dt +

0 t0 0 r
d()U (, + t)f ( + t) dt
t
d||() U (, 0) S( + t)x dt +
r
d||() U (, + t)
f ( + t) dt.
By Fubinis theorem and a change of variables, we obtain

t0 0
(Stx + T0 (t)f ) dt M e||t0

0
t0
d||() S( + t)x dt
t 0 0
+ M e||t0
r
d||() f (t) dt S(t) t0||([r, 0]) x + ||([r, 0]) f S(t) x + f

0tt0 L1 ) L1 )
M ( sup
0tt0
Mw ||([r, 0])(t0 sup q1 x + c1 f with Mw := M e|w|t0 , q1 := sup

0tt0 L1 ,
S(t) Mw ||([r, 0])t0 and c1 := Mw ||([r, 0]).
Choosing t0 small so that q1 < 1, the condition (M ) is then satised. Corollary 3..1. Let given as in Lemma 3..2. Then, (A, D(A)) generates a strongly continuous semigroup (T (t))t0 on E1 satisfying the variation of constants formula T (t) ( x ) = T0 (t) ( x ) + f f
0 x t
T0(t s)BT (s) ( x ) ds f
(3.10)
for all ( f ) D(A) and t 0. Moreover, there is a unique mild solution u of the equation (3.6) and it is given by u(t) = 1T (t) ( f ) , t 0, f (t), r t 0.
x
172
4.
Asymptotic behaviour of partial differential equations with nonautonomous past delay
We study the persistence of some asymptotic properties of the semigroup (S(t))t0 under the modied delay term. More precisely, we give conditions under which the trajectories of the semigroups S() and T () have the same asymptotic behavior. First, we show that this holds for S() and T0 (), and by perturbation techniques the claim will be obtained. In the following theorem, we show rst that S() and (T0(t))t0 have some common asymptotic properties. The nite delay case can be shown similarly as in [7]; for the asymptotic almost automorphy see [6]. We give here the proof for the innite delay case. Theorem 4..1. Assume that the evolution family (U (t, s))ts0 is exponentially stable, and let x X. If the map t S(t)x is (i) in C0 (R+ , X), the space of functions vanishing at innity, or (ii) asymptotically almost periodic, or (iii) asymptotically almost automorphic, or (iv) uniformly ergodic, or (v) totally uniformly ergodic. R then t T0(t) ( x ) has the same property for every f L1(I , X). f Proof. Let f L1(I , X). Since R T0(t) ( f ) =
x S(t)x St x+T0 (t)f
we have only to verify that the map t St x + T0(t)f has the same property as S()x. R To show (i), we should show that t St x + T0(t)f C0(R+ , L1(I , X)), i.e.,
0
St x() + T0 (t)f () d 0.
t
We have
0 0 0
Stx() + T0(t)f () d
0
Stx() d +
0
T0(t)f () d
1t+0 U (, 0)S(t + )x d +
t
1t+0 U (, t + )f (t + ) d U (, t + )f (t + ) d.
1t U (, 0)S(t + )x d +
Since S()x is vanishing at +, there exists M > 0 such that sup S(t)x M , and
t0
then 1t U (, 0)S(t + )x M M ew := g().
A Partial Differential Equation with Nonautonomous Past Delay The function g is integrable on R and 1t U (, 0)S(t + )x
t+
173
a.e. R .
Hence, the Lebesgue dominated convergence Theorem implies that

0
U (, 0)S(t + )x d 0 .
t t+
We have also
t 0
T0(t)f
L1
U (, t + )f (t + ) d = Me
U ( t, )f () d
tw
L1
t+
0 .
Assume now (ii). It is clear that T0()f is asymptotically almost periodic. Now it is sufcient to show that R+ t St x is asymptotically almost periodic. From denition there exist a unique g AP (R, X) and a unique h C0 (R+ , X) such that S(t)x = h(t) + g(t), Hence, St x() = = U (, 0)S(t + )x, + t > 0 0, +t0 U (, 0)g(t + ) + U (, 0)h(t + ), + t > 0 0, + t 0. t 0. (4.11)
Since h C0 (R+ , X) and U (, 0) is exponentially stable, then, as in (i), h1 (t)() := 1t U (, 0)h(t + ) C0(R+ , L1(R , X)). Set g1 (t)() := U (, 0)g(t + ) and h2 (t)() := h1 (t)() 1t+0 U (, 0)g(t + ) for all t R and R . Then, (g1(t) + h2(t))() = g1(t)() + h1 (t)() g1(t)() g1(t)() = 0 for + t > 0, for + t 0 (4.12)
=St x(). Since h1 C0(R+ , L1(R , X)) and

0 t
1t+0 U (, 0)g(t + ) d g
U (, 0) d 0 ,
t+
it follows that h2 C0(R+ , L1(R , X)). Since g is almost periodic, then for all there exists l( ) > 0 such that for all a R, there exists [a, a + l( )] such that g(t + + ) g(t + ) for all t R, I , R
> 0,
174 and moreover
g1(t + ) g1 (t)
L1
=
0
U (, 0)g(t + + ) U (, 0)g(t + ) d U (, 0)
0
g(t + + ) g(t + ) d U (, 0)
L1 .
U (, 0) d =
This means that g1 is almost periodic, and nally t St x is asymptotically almost periodic. For the asymptotic almost automorphy, assume that S(t)x = h(t) + g(t), t 0, where g AA(R, X) and h C0 (R+ , X). From (4.12), Stx = g1 (t) + h2 (t), t 0, with R. h2 C0(R+ , X). To conclude we have to show that g1 is almost automorphic on I For this, consider a sequence (n )nN in I Since g AA(I X), there is a subsequence R. R, (sn )nN such that lim g(t + + sn sm ) = g(t + )
n,m+
for each t I I . Therefore, as for the almost periodicity, one can see that R, R g1 AA(I L1(I , X)). As (T0(t))t0 is exponentially stable then it is asymptotiR, R cally almost automorphic.
+
For (iv), we have that the limit F ()
lim
0
et S( + t)xdt exists in
BU C(R+ , X) for each x X, and then F () BU C(R+ , X). Put Ft () = U (, 0)F (t + ) U (, 0)F (0) for + t > 0 for + t 0.
Show rst that t Ft BU C(R+ , L1(R , X)). For h > 0, we have

0
Fs+h () Fs () d
0
1s+h+0 U (, 0)F (s + h + ) 1s+0 U (, 0)F (s + ) d 1s+h+0 U (, 0)F (0) 1s+0 U (, 0)F (0) d
s 0
U (, 0)F (s + h + ) d +
sh s s
U (, 0)
F (s + h + ) F (s + ) d
+
sh
U (, 0)F (0) d
0
2h U (, 0)
+
s
U (, 0)
F (s + h + ) F (s + ) d.
Since F () BU C(R+ , X) then, for > 0 there exists h > 0 such that F (s + h + ) F (s + ) for all s + 0 and 0 < h < h . Hence, Fs+h () Fs ()
h U (, 0)
U (, 0)
L1 .

+
175
The same for h < 0. To achieve (iv) we show that the limit lim
0
et S+t xdt = F
0
exists in BU C(R+ , L1(R , X)). Let s 0.

0 +
0 0
et Ss+t xdt() Fs () d
+
=
0
0 +
et Ss+t x()dt Fs () d 1s+t+0 et U (, 0)S(s + t + )xdt 1s+>0 U (, 0)F (s + )

0
1s+0 U (, 0)F (0) d

0 +
s s
U (, 0) U (, 0)
+
et S(s + t + )xdt F (s + ) d
+
et S(s + t + )xdt F (0) d U (, 0)

L1
sup
s0 s 0
et S(s + t)xdt F (s)

+
+
s
e(s+) U (, 0) U (, 0)
et S(t)xdt F (0) d
+ Hence
F (0) e(s+) F (0) d.
et Ss+t xdt() Fs () d
+
2 sup
s0 0 0
et S(s + t)xdt F (s) e (1 e )d,
U (, 0)
L1
+ M F (0)
and this ends the proof of (iv). The same proof yields the assertion (v). Using [3, Lemma 7.2] and the variation of constants formula (3.7), we obtain the asymptotic properties of the perturbed semigroup T () and then the ones of the trajectories of the solutions to equation (3.6). Theorem 4..2. Assume that the evolution family (U (t, s))ts0 is exponentially stable, and that is bounded linear from C([r, 0], X) L1([r, 0], X) to X. Assume also that S()x satises one of the asymptotic properties (i)-(v) for all x X, and there are 0 q1 < 1 and c1 > 0 such that
(Stx + T0 (t)f ) dt q1 x + c1 f
0
(4.13)
x for all ( f ) D(A0). Then, T () ( x ) has the same asymptotic properties for all ( x ) E1. f f
176
Proof. As in the proof of Theorem 3..1, one can nd 0 q < 1 and c > 0 such that
BT0(t) ( x ) f
0
q (x) f
(4.14)
x for all ( f ) D(A), where c is a norm equivalent to E1 . Fix now ( x ) D(A). f x From Theorem 4..1, the trajectory T0() ( f ) has the same asymptotic properties of S()x with respect to the new norm c . From Theorem 3..1, we have
T (t) ( x ) = T0 (t) ( x ) + f f
0
T0(t s)BT (s) ( x ) ds, f
t 0.
(4.15)
Hence, it is sufcient to show that the function

t
I + R
t f (t) :=
T0(t s)g(s)ds,
satises the same properties, where g is dened by g(s) := BT (s) ( f ) , s 0, 0, s 0,

x
For assertions (i)-(ii)-(iv)-(v), by [3, Lemma 7.2] it sufces to show g L1(I (E1, C )), R, and this can be deduced from (4.14) and (3.7). As the semigroup S() is bounded, the assertion (iii) follows from [6, Proposition 2.3] and (ii). Finally, all the assertions follow x for all ( f ) E1 by density of D(A). The asymptotic behavior of the solutions of (3.6) can be now obtained. Corollary 4..1. Assume that S()x satises one of the asymptotic properties (i)-(v) for all x X and that is a bounded linear from C([r, 0], X) L1 ([r, 0], X) to X satisfying (4.13). Then, t ut and t u(t) have the same asymptotic properties as S()x.
5.
Application
In this section, we use the abstract results of the previous ones to study the wellposedness and asymptotic behavior of the dynamical population equation r u (t, x) = N u(t, x) du(t, x) + 0 b(a)v(t, a, x)da b1 v(t, r, x), t 0, x , v (t, a, x) = v(t, a, x) + v(t, a, x) dv(t, a, x) b(a)v(t, a, x), t 0, D a x , 0 a r, v(t, 0, x) = f (x)u(t, x), t 0, x , (5.16) where D and N denote the Laplace operators whit Dirichlet and Neumann conditions, respectively. Here t, r are positive, a [0, r], d, b1 > 0 and b L1 ([0, r], I +). The space R n variable x is supposed to vary in R where is open, connected and bounded with smooth boundary. The condition u(t, x) = 0 in , n (5.17)
177
where n denotes, as usual, the outward normal, states that the population cannot cross the boundary. The condition v(t, a, x) = 0 if x , t 0, a [0, r] (5.18) says that no pregnant individual reaches the borderline. Moreover the condition f (x) = 0 for x is required. The initial conditions u(0, x) = u0 (x), u(, x) = g(, x), (r, 0), v(0, a, x) = v0 (a, x), v0 (a, x) = f (x)g(a, x), a [0, r], where v0 , g and u0 are given functions. (5.19) (5.20)
5.1.
The population equation as an abstract equation with nonautonomous past delay
In this subsection, we show in details how to transform the population equation (5.16)(5.20) in a partial differential equation with nonautonomous past delay, and then study its wellposedness and asymptotic behavior. To this purpose, we start by solving the second equation along the characteristic lines in the plane (t, a), namely in the strip [0, +)[0, r]. Set V (s, x) = v(t0 + s, a0 + s, x), where a0 [0, r], t0 0 are xed, while x varies in and s in [0, +). Rewriting the second equation in (5.16) with V (s, x), one obtains V (s, x) = v(t0 + s, a0 + s, x) + v(t0 + s, a0 + s, x) s t a = D v(t0 + s, a0 + s, x) dv(t0 + s, a0 + s, x) b(a0 + s)v(t0 + s, a0 + s, x) = D V (s, x) dV (s, x) b(a0 + s)V (s, x). To solve the above equation we follow the abstract approach choosing X = L1() as a Banach space and denoting D(D ) the domain of D on X. For t < a, putting t0 = 0, we obtain V (s, x) = D V (s, x) dV (s, x) b(a0 + s)V (s, x) V (0, x) = v(0, a0, x) = v0 (a0, x), which has the unique solution V (s, ) = e
s 0 b(a0 +)d
es(D d) V (0, ),
where d denotes the multiplication operator and et(D d) the strongly continuous semigroup generated by the linear operator D d on X with the appropriate domain. Hence v(s, a0 + s, ) = e =e
s 0 b(a0 +)d a0 +s a0
es(D d) v(0, a0, ) v0 (a0, ).
b()d s(D d)
178
Setting t = s and a0 + s = a, this becomes v(t, a, ) = e For t > a, putting a0 = 0, we obtain V (s, x) = D V (s, x) dV (s, x) b(s)V (s, x) V (0, x) = v(t0 , 0, x) = f (x)u(t0, x). Solving this equation we obtain v(t0 + s, s, ) = V (s, ) = e =e =e Setting t0 + s = t, this becomes v(t, a, ) = e
a 0 b()d s 0 s 0 s 0 a at
b()d s(D d)
v0 (a t, ).
b()d s(D d)
e e e
V (0, ) v(t0, 0, ) f ()u(t0, ).
b()d s(D d) b()d s(D d)
ea(D d) f ()u(t a, ).
Substituting v(t, a, x) in the rst equation of (5.16), one has, for t > r,
r
u (t, x) = (N d)u(t, x) +
0 r
b(a)v(t, a, x)da b1v(t, r, x) b(a)[e

0
a 0
= (N d)u(t, x) + b1 e
r 0
b()d a(D d)
f (x)u(t a, x)]da
b()d r(D d)
f (x)u(t r, x)
0
= (N d)u(t, x) +
r
b(a)[e
a 0
b()d a(D d)
f (x)u(t + a, x)]da
b1 e
r 0
b()d r(D d)
f (x)u(t r, x).
For t < r, we have

r
u (t, x) = (N d)u(t, x) +
0 t
b(a)v(t, a, x)da b1 v(t, r, x)

r
= (N d)u(t, x) +
0 0
b(a)v(t, a, x)da +
t
b(a)v(t, a, x)da b1v(t, r, x) e f(x)u(t + a, x)]da

r rt
= (N d)u(t, x) +
t t
b(a)[e
b()d t(D d)
a 0
b()d a(D d)
+
r
b(a)[e
a at
v0 (a t, x)]da b1e
b()d t(D d)
v0 (r t, x).
We put U (t, s) = e and () =

r 0
t s b()d
e(st)(D d) ,
t s 0,
b(s)f ()(s)ds b1f ()(r),
C([r, 0], X).
A Partial Differential Equation with Nonautonomous Past Delay It is easy to verify that (U (t, s))ts0 is an evolution family. The modied history function ut is ut (s, ) := Hence, (t()) = u
r
179
U (s, 0)u(t + s, ) U (s, s + t)g(t + s, )

0
if t + s > 0 if t + s 0.
b(s)f ()t(s, )ds b1f ()t(r, ). u u
Since for t > r, ut (r, ) = U (r, 0)u(t r, ) and s r, t + s t r 0 = ut (s, ) = U (s, 0)u(t + s, ), through
0
(t) = u
r
b(s)f ()U (s, 0)u(t + s, )ds b1f ()U (r, 0)u(t r, ).
If we set B = N d, then
0
Bu(t, ) + (t) = (N d)u(t, ) + u

r
b(a)f ()U (a, 0)u(a + s, )da
b1f ()U (r, 0)u(t r, ) = u (t, ). Thus, by setting u(t) := u(t, ), we obtain u u (t) = Bu(t) + (t ). For t r, ut (r, ) = U (r, t r)g(t + s, ) and
0
(t) = u =
r 0
b(s)f ()t(s, )ds b1f ()U (r, t r)g(t r, ) u

r t
b(s)f ()U (s, s + t)g(t + s, )ds b(s)f ()U (s, 0)u(t + s, )ds b1 f (r)U (r, t r)g(t r, ).
t
Summarizing, the population equation is written as the abstract equation with nonautonomous past u u (t) = Bu(t) + (t), t 0, (5.21) u(0) = x, u0 = g, where the operator B = N d generates an exponentially stable C0 -semigroup S := (S(t))t0 on X = L1().
5.2.
Wellposedness and asymptotic behavior of the population equation
We have transformed the population equation (5.16)- (5.20) into the abstract equation with nonautonomous past u u (t) = Bu(t) + (t), t 0, (5.22) u(0) = x, u0 = g,
180
where the operator B = N d generates an exponentially stable C0 -semigroup S := (S(t))t0 on X = L1 () and the backward evolution family U (t, s) = e s b()d e(st)(D d) , t s 0, is exponentially stable. Moreover, the operator , given by
0
t
() =
r
b(s)f ()(s)ds b1f ()(r),
can be written as a Stieltjes integral

0
() =
r
d(s)(s),
where : [r, 0] L(X) is the function with a bounded variation on [r, 0] given by
s
(s) = b1f ()1[r,0] +

0
b( )f ()d.
Here, we have considered a delay operator with one discrete delay, but one can consider several discrete delays and the delay operator
0 n
() =
r
b(s)f ()(s)ds
i=1
bi f ()(ri),
with 0 r1 < r2 < ... < rn = r. In this case is the bounded variation function
n s
(s) =
i=1
bi f ()1[ri,0] +
b( )f ()d.
0
Hence, by Lemma 3..2, these operators satisfy the Miyadera-Voigt condition (M ). Therefore, by Corollary 3..1, there is a unique mild solution of equation (5.22). Moreover, we have the following result. Theorem 5..1. There is a unique mild solution u : I + L1() of the population R equation (5.16)-(5.20). Moreover, if ||([r, 0]) < d then u(t) L1() 0, t +. Proof. One has
+ + 0
(Stx + T0(t)f ) dt =
0 + 0 0 r
d()(Stx + T0 (t)f )() dt

r t
0 + t 0 t
d()U (, 0)S( + t)x dt +

0 r 0 r
d()U (, + t)f ( + t) dt
t
d||() U (, 0) S( + t)x dt +
r
d||() U (, + t)
f ( + t) dt
1 ||([r, 0]) x + ||([r, 0]) f . d
1 Hence, by assumption q1 = ||([r, 0]) < 1, and since S is exponentially stable, by d Theorem 4..2, the result follows.
181
References
[1] A. B tkai and S. Piazzera, Semigroups and linear partial differential equations with a delay, J. Math. Anal. Appl. 264 (2001), 1-20. [2] A. B tkai and S. Piazzera, Semigroups for Delay Equations on LP -phase Spaces, a book manuscript, Research Notes in Mathematics 10, A. K. Peters, Ltd., Wellesley, MA, (2005). [3] C. J. K. Batty and R. Chill, Bounded convolutions and solutions of inhomogeneous Cauchy problems, Forum Math. 11 (1999), 253277. [4] S. Boulite, L. Maniar and G. M. NGu r kata, Almost automorphic solutions for hyee perbolic semilinear evolution equations, Semigroup Forum 71 (2005), 231240. [5] S. Brendle and R. Nagel, Partial functional differential equations with nonautonomous past, Disc. cont. Dyn. Syst. 8 (2002), 1-24. [6] V. Casarino, Almost automorphic groups and semigroups. Rend. Accad. Naz. Sci. XL Mem. Mat. Appl. 24 (2000), 219235. [7] V. Casarino and S. Piazzera, On the stability of asymptotic properties of perturbed C0 -semigroups, Forum Math. 13 (2001), 91-107. [8] V. Casarino, L. Maniar and S. Piazzera, The asymptotic behaviour of perturbed evolution families, Differential Integral Equations 15 (2002), 567586 [9] C. Chicone and Y. Latushkin, Evolution Semigroups in Dynamical Systems and Differential Equations, Mathematical Surveys and Monographs 70, American Mathematical Society, Rhode Island, (1999). [10] C. Corduneanu, Almost Periodic Functions , 2nd Edition, Chelsea-New York, 1989. [11] K.J. Engel and R. Nagel, One-parameter Semigroups for Linear Evolution Equations , Graduate Texts in Mathematics 194, Springer-Verlag (2000). [12] A. M. Fink, Almost Periodic Differential Equations, Lecture Notes in Mathematics 377, Springer-Verlag, New York, 1974. [13] G. Fragnelli, Classical solutions for PDEs with nonautonomous past in Lp spaces, Bulletin of the Belgian Mathematical Society 11 (2004), 133148. [14] G. Fragnelli and G. Nickel, Partial functional differential equations with nonautonomous past in Lp phase spaces, Diff. Int. Equ. 16 (2003), 327348. [15] G. Fragnelli and L. Tonetto, A population equation with diffusion, J. Math. Anal. Appl. 289 (2004), 9099. [16] G. Fragnelli, An age dependent population equation with diffusion and delayed birth process, Int. J. Math. Math. Sci. 20 (2005), 32733289.
182
[17] I. Miyadera, On perturbation theory for semi-groups of operators, T hoku Math. J. 18 o (1966), 299310. [18] R. Nagel and G. Nickel, Well-posedness for nonautonomous abstract Cauchy problems, Progr. Nonlinear Differential Equations Appl. 50, Birkh user, (2002), 279293. a [19] G. M. NGu r kata, Almost Automorphic Functions and Almost Periodic Functions in ee Abstract Spaces, Kluwer Academic / Plenum Publishers, New York, 2001. [20] G. M. NGu r kata, Topics in Almost Automorphy , Springer-Verlag, New York, 2005. ee [21] G. Nickel, Evolution semigroups for nonautonomous Cauchy problems, Abstr. Appl. Anal. 2 (1997), 7395. [22] G. Nickel and A. Rhandi, Positivity and Stability of Delay Equations with Nonautonomous Past, Math. Nachr. 278 (2005), 113. [23] F. R biger, A. Rhandi, R. Schnaubelt and J. Voigt, Non-autonomous Miyadera pertura bations, Diff. Int. Eq. 13 (2000), 341368. [24] J. Voigt, On the perturbation theory for strongly continuous semigroups, Math. Ann. 229 (1977), 163171. [25] M. Stein, H. Vogt, J. Voigt, The modulus semigroup for linear delay equations III. J. Funct. Anal. 220 (2005), 388400.
S OLVING THE H YPERBOLIC P ROBLEM O BTAINED BY T RANSMUTATION O PERATOR

Hikmet Koyunbakan Firat University, Department of Mathematics, 23119 Elazig, TURKEY
Abstract In this paper, we solve the problem of constructing the kernels of transmutation operator and give the generalized transmutation operator for a singular problem.
Key Words: Inverse Problem, Hyperbolic Equation AMS Subject Classication: 31A25, 35L20.
1.
Introduction
The standard problem on spectral theory is to solve (1.1) for y with appropriate boundary conditions, for a given potential q(x). In the unusual application all that can be observed is the asymptotic behaviour of y for large |x| , which is taken for the whole spectrum comprising at the scattering data for q(x). Inverse eigenvalue problems, usually for equations of the form y + q(x)y = y (1.1)
with a variety of boundary conditions, have an extensive literature [3], [4], [5], [6], [14] . Unless q(x) is constrained, one needs the spectra from two sets of boundary conditions, or one spectrum and a sequence of norming constants to uniquely determine q(x). In the past, important works on these problems were done by Gelfand and Levitan [4], Levinson [8] and Levitan [9]. These authors show that the inverse problem is solved by q(x) = 2 d H(x, x), dx
E-mail address: hkoyunbakan@gmail.com
184
Hikmet Koyunbakan
where H(x, t) is nucleus function of transmutation operator. One topic of inverse problem for spectral analysis is transmutation operator [1], [2], [7], [15]. In [5], [11] Gilbert examined some problems for boundary value problems by using this method, in [7], Hryniv and Mykytyuk, constructed transformation operators for Srurm Liouville operators with singular potentials, in [1], Boumenir and Tuan proved the existence of a transmutation operator between two weighted Sturm-Liouville operator and in [15] Volk solved Hyperbolic equation having singularity type operator.
m2 1 4 x2
at zero by using transmutation

m2 1
In this paper, we obtain the problem which is singularity type k + x2 4 and solve x it by using the Riemann method. Before giving the main results, we will mention some well-known trues. We consider the conuent hypergeometric equation d2 W dW aW = 0 (1.2) + [c x] 2 dx dx in which a and c are parameters. In this equation, by considering transmutation w = 1 exp( x )xm+ 2 W, we get 2 x where k =
c 2
1 k m2 d2 w + + dx2 4 x x2
1 4
w=0
(1.2 )
a, m =
c 2
1 . Standard solution of (1.2 ) is 2
1 x 1 M (k, m, x) = exp( )xm+ 2 F (m k + , 2m + 1, x), (1.3) 2 2 where F is solution of (1.2) and sometimes known as Kummer s function. [13, p:255-261] Let E be a topological linear space and L1 and L2 be two linear, but not necessarily continuous, operators from E to E, where E1 and E2 are closed subspaces in E.
Denition 1.1 A linear invertible operator, X, dened in the whole space E and acting from E1 to E2 is called a transmutation operator for a pair of operators L1 and L2 if it satises the following two conditions [9, p:4] i) The operator X and its inverse X 1 are continuous in space E. ii) The operator equation is of the form L1X = XL2 or L1 = XL2X 1. We consider the problem 1 k m2 d2 w + + dx2 4 x x2
1 4
w = 0,
0<x<1
(1.4) (1.5)
w(0) = 0.
Equation (1.4) is a Whittaker equation and its solution is Whittaker function M (k, m, x). [13] Furthermore, considering 1 k m2 d2w + + dx2 4 x x2
1 4
+ q(x) w = 0,
0<x<1
(1.6)
Solving the Hyperbolic Problem Obtained by Transmutation Operator
185
w(0) = 0,
(1.7)
where q(x) is a real-valued integrable function on (0, 1). This function is called potential function in spectral theory.
2.
Existence of the Transmutation Operator
Theorem 2.1 Let take M (k, m, x) and (x) as solutions of (1.4),(1.5) and (1.6), (1.7) respectively. When the transmutation operator X = XL1 ,L2 can be realized as
x
X [M (k, m, x)] = (x) = M (k, m, x) +

0
H (x, t) M (k, m, t)dt,
(2.1)
the kernel H (x, t) is solution of the problem
k m2 2H (x, t) + x2 x x2
1 4
+ q(x) H (x, t) =
2 H (x, t) k m2 + t2 t t2
1 4
H (x, t) (2.2)
H (x, 0) = 0,
(2.3)
dH (x, x) = q(x). dx
(2.4)
Conversely, if function H(x, t) is a solution of equation (2.2) and satises conditions (2.3), (2.4) then the operator XL1 ,L2 dened by formula (2.1) is a transmutation operator for the pair of L1 and L2 . Proof: We dene two operators d2 k m2 + + dx2 x x2
1 4
L1 =
+ q(x) ,
L2 = Hence, we have
d2 k m2 + + dx2 x x2
1 4
186
Hikmet Koyunbakan
L1 X [M (k, m, x)] = L1 = + = M (k, m, x)
k m2 + x x2
1 4
+ q(x)
dH (x, x) M (k, m, x) dx H (x, t) M (k, m, t) M (k, m, x)H (x, x) x t=x

x
m2 k 2 H (x, t) M (k, m, t)dt + + x2 x x2

1 4 x
1 4
+ q(x) M (k, m, x)
k m2 + + x x2 On the other hand,
+ q(x)
0
H (x, t) M (k, m, t)dt.
(2.5)
XL2 [M (k, m, x)] = M (k, m, x) +

x
k m2 + x x2
1 4
M (k, m, x)
0 x
H (x, t) M (k, m, t)dt k m2 H (x, t) + t t2

1 4
+
0
M (k, m, t)dt.
(2.6)
For third term, integrating by parts

x
H (x, t) M (k, m, t)dt =

0
H (x, t) M (k, m, t) t=0

x
t=x
H (x, t) M (k, m, t) t t=x
+
0
2H (x, t) M (k, m, t)dt. t2
Inserting this value in the (2.6), we obtain k m2 + x x2

1 4
XL2 [M (k, m, x)] = M (k, m, x) + H (x, t) M (k, m, t) t=0 +

x t=x
M (k, m, x) (2.6 ) M (k, m, t)dt.
H (x, t) M (k, m, t) t
x
t=x 1 4
2 H (x, t) M (k, m, t)dt + t2

0
k m2 H (x, t) + t t2
187
The operator X is a transmutation operator and satises Denition 1.1 (i). Hence, from (2.5) and (2.6 ) we get the problem
m2 k 2 H (x, t) + x2 x x2
1 4
+ q(x) H (x, t) =
2 H (x, t) k m2 + t2 t t2
1 4
H (x, t) , (2.7)
H(x, 0) = 0,
(2.8)
dH (x, x) = q(x). (2.9) dx Thus, if an operator X = XL1 ,L2 can be realized in form (2.1), then its kernel, i.e., H(x, t) is solution of problems (2.2)-(2.4). Conversely, if a function H(x, t) is solution of problems (2.2)-(2.4), then operator X = XL1 ,L2 constructed from (2.1) is a transmutation operator for the pair of L1 and L2 . To verify this, it sufces to repeat the above computations in reverse order. This completes the proof. 2
3.
Riemann Method
In this section, we solve the problem (2.7)-(2.9) by using Riemann method. Equation (2.7) has a singularity at x = t = 0. Thus, under the following substitution (x t)2 (x + t)2 , s= , 4 4 the equation (2.7) is equivalent to z= H (x, t) = (z s) 4 U (z, s) ,
1
(3.1)
1 U 1 U 1 2U + qU + R (z, s) U = 0, zs 4 (z s) z 4 (z s) s 4 zs
(3.2)
where R (z, s) is a function depending on variables z and s. Equation (3.2) is a Euler-Poisson-Darboux differential equation. For t = 0 in (3.1), because of z = s, equation (3.2) has a singularity on the characteristic line z = s. For > 0, we consider characteristic z = s + . (Figure 3.1) Hence, the problem (2.7)-(2.9) transforms to 2U 1 U 1 U 1 + qU + R (z, s) U = 0, zs 4 (z s) z 4 (z s) s 4 zs q U dU + 3 = 1 , dz 2z 4 4z 2 U (z, z ) = 0. (3.3)
(3.4) (3.5)
In this section, we shall describe an approach which has considerable afliation with the Greens function and which is restricted in its application to hyperbolic equation. Now,
188
Hikmet Koyunbakan
Figure 3.1: Singularity of (3.3) on plane (z,s).
we shall nd out solution of problem (3.3)-(3.5) by Riemann method. We consider two operators L [U ] = 1 U 1 U 2U + + R (z, s) U, zs 4 (z s) z 4 (z s) s
L [V ] =
2V 1 V 1 V 1 V + R (z, s) V, + zs 4 (z s) z 4 (z s) s 2 (z s)2
where L is the adjoint operator of L and V is a solution of conjugate equation. If the rst equation multiply by V and the second equation multiply by U , after subtracting and integrating on , it yields
V L [U ] U L [V ] dzds
1 2 z
U V 1 U UV s s 2 (z s)
dzds
1 2 s
U V 1 U + UV z z 2 (z s)
dzds.
Using the Green theorem, we yield
189
V L [U ] U L [V ] dzds =
AM BA
U V 1 1 V U U V dz 2 z z 2 (z s) (3.6)
1 U V 1 V U U V ds. 2 s s 2 (z s)
As shown in the Figure 3.1, since s and z are xed on AM and M B respectively, it follows that 1 U V 1 V U U V ds = 0 2 s s 2 (z s)
AM
and
MB
U V 1 1 V U + U V dz = 0. 2 z z 2 (z s)
Furthermore, we have 1 U V 1 1 1 V U + U V dz = U V (M ) + U V (A) 2 z z 2 (z s) 2 2 +

AM
AM
V V dz z 4 (z s)
and 1 U V 1 V U + U V ds = 2 s s 2 (z s)
MB
1 1 U V (B) U V (M ) 2 2
MB
V V + ds. s 4 (z s)
Inserting these values in (3.6), it is obtained that {V L [U ] U L [V ]} dzds =

BA
1 U V UV V U dz 2 z z 2 (z s)
1 U V UV 1 + V U ds U V (M ) 2 s s 2 (z s) 2 V 1 V 1 dz + U V (B) U + U V (A) + 2 z 4 (z s0 ) 2
AM
1 U V (M ) 2
MB
V V + ds. s 4 (z0 s)
(3.7)
190
Hikmet Koyunbakan
On the right hand, the values of integrals on AM and M B are not known and on these lines, M is not vanished, then we have to investigate solution of V the following problem. L [V ] = 0, V V = 0 on line AM , z 4 (z s0 ) V V + = 0 on line M B, s 4 (z0 s) V = 1 at M point. Solution of this problem is called Riemann-Green function and it is in the form [12] V (z, s, z0, s0) = where = (s s0 ) (z z0 ) (s z0 ) (z s0 ) (s z) 2 [(s0 z) (s z0 )]
1 4 1
(3.8) (3.9)
(3.10) (3.11)
1 1 , , 1, , 4 4
and F is a Hypergeometric function such that F Hence, we have U V (A) + U V (B) + 2
1 1 , , 1, 4 4
= 1+
1 4 + . . .. 1! 1
1 4
U (M ) =
V L [U ] dzds
+
BA
U V UV 1 V U dz 2 z z 2 (z s) 1 U V UV V U ds. 2 s s 2 (z s) (3.12)
Because of being z = s + on BA line, 1 U V UV 1 U V UV V U = V U . 2 z z 2 (z s) 2 s s 2 (z s)

1 qU. 4 zs
dz = ds and
On the other hand L [U ] =
From this point of view and (3.12) we obtain

z z0
U V (A) + U V (B) U (M ) = + 2
s0 s0 +
1 qU V dzds. 4 zs
(3.13)
191
Last equation is a Volterra- type integral equation. Therefore, its solution is unique and the solution can be proved by successive approximations . Now, we generalize transmutation operator for the problem (1.6),(1.7). Taking q1 (x), q2(x), . . . , qn (x) as real-valued integrable functions on (0, 1), we show that the solution of the following problem can be established by using transmutation operator d2 w 1 k m2 + dx2 4 x x2
1 4 n
+
k=0
qk (x) w = 0,
0<x<1
(3.14)
w(0) = 0. If the solution of the problem 1 k m2 d2w + dx2 4 x x2

1 4
+ q1 (x) w = 0,
w(0) = 0 is 1 (x), then, we obtain

x
1 (x) = M (k, m, x) +
0
H1 (x, t) M (k, m, x)dt.
In the identical manner, when the solution of the problem d2 w 1 k m2 + dx2 4 x x2

1 4
+ q1 (x) + q2 (x) w = 0,
w(0) = 0 is 2 (x), then, we have

x
2 (x) = X [1 (x, )] = 1 (x, ) +

x
H2 (x, t) 1 (t) dt.

0 x
= M (k, m, x) +
0
H2 (x, t) + H1 (x, t) +
t
H1 (x, s) H2 (s, t) ds M (k, m, t)dt.
or
x
2 (x) = M (k, m, x) +
0
K1 (x, t) M (k, m, t)dt,
where
192
Hikmet Koyunbakan
x
K1 (x, t) = H2 (x, t) + H1 (x, t) +

t
H1 (x, s) H2 (s, t) ds .
If this process is continued for q1 (x), q2(x), ...qn(x), the solution of the problem 1 k m2 d2 w + dx2 4 x x2
1 4 n
+
k=0
qk (x) w = 0,
0<x<1
w(0) = 0 is
x
n (x) = X n1 (x) = M (k, m, x) +

0
Kn1 (x, t) M (k, m, t)dt.
(3.15)
In this case, the kernel Kn1 (x, t) is solution of the problem 2Kn1 (x, t) k m2 + + x2 x x2 = 2K
n1 (x, t) t2 1 4 n
+
k=0
qk (x) Kn1 (x, t) (3.16)
k + + t
m2
t2
1 4
Kn1 (x, t) ,
Kn1 (x, 0) = 0,
x
(3.17)
1 Kn1 (x, x) = 2
qk (s)ds.
0 k=0
(3.18)
References
[1] Boimenir, A. and Tuan Vu Kim, Existence and construction of the transmutation operator, Journal of Math. Phys., 45, 7(2004) 2833-2843 [2] Carrol, R. W., Transmutation theory and applications , North-Holland, 1985 [3] Gasymov, M.G., On the determination of the Sturm-Liouville equation with peculiarity in zero on two spectrum. Dan SSSR, vol.16, 2(1965), 271-276. [4] Gelfand, I.M. and Levitan, B.M., On the determination of a differential equation from its spectral function, Amer. Math. Soc.Trans., ser. 2,1(1955), 253-304. [5] Gilbert, Robert P. and Wood, David H. A transmutation approach to underwater sound propagation. Wave Motion 8 (1986), no. 5, 383397.
193
[6] Hochstadt H., The Inverse Sturm-Liouville Problem, Commun. on Pure and Appl. Math.1973, vol.XXVI, 715-729. [7] Hryniv, R. O. and Mykytyuk Ya. V., Transformation operators for Sturm-Liouville operators with singular potentials, Math. Phys. Anal. Geom., 7(2004), 119-149 [8] Levinson, N., The inverse Sturm-Liouville problem, Mat.Tidsskr.B., 1949, pp. 25-30. [9] Levitan, B. M., Inverse Sturm-Liouville problems , VNU science press, Netherland, 1987. [10] Levitan, B. M. and Sargsjan, I. S, Introduction to spectral theory: Self-Adjoint ordinary differential operators. Am. Math. Soc Trans. 39, 1975. [11] Lin, Zhongyan; Gilbert, R. P. Numerical algorithm based on transmutation for solving inverse wave equation. Math. Comput. Modelling 39 (2004), 13, 14671476. [12] Mackie, A.G., Boundary value problems, Oliver and Boyd Ltd., Edinburgh and London, 1965. [13] Olver, F.W.J., Asymptotics and Special Functions , Academic Press, New York and London, 1974. [14] Panakhov E. S.and Koyunbakan H., Inverse problem for singular Sturm-Liouville operator, Proceeding of IMM of NAS of Azerbaijan , vol. XVIII( 2003), p.113-126. [15] Volk, V. Y., On Inverse formulas for a differential equation with a singularity at x = 0, Usp. Math. Nauk. 8 (56), 1953, 141-151.
E LLIPTIC O PERATORS WITH VARIABLE C OEFFICIENTS G ENERATING F RACTIONAL R ESOLVENT FAMILIES

Miao Li,1 Fu-Bo Li 1 and Quan Zheng 2 1 Department of Mathematics, Sichuan University, Chengdu, Sichuan 610064, P.R. China 2 Department of Mathematics, Central China (Huazhong) Normal University Wuhan 430079, P. R. China and Department of Mathematics Huazhnog University of Science and Technology Wuhan 430074, P. R. China
Abstract In this paper, we show that under very general conditions, elliptic operator with zero boundary condition can generate fractional resolvent families.
Key Words: elliptic operator, fractional resolvent family, Volterra equations AMS Subject Classication: Primary 45D05, 34G10; Secondary 47D06, 47D09.
Introduction
The theory for the abstract Volterra equations has been developed rapidly due to its applications to many problems in mathematical physics, such as viscoelasticity and heat conduction in materials; see [10] and the references therein. A Volterra-type equation on a complex Banach space X takes the form
t
u(t) = f (t) +
0
a(t s)Au(s) ds,
t0
(0.1)
E-mail address: limiao1973@hotmail.com; The rst and second author were supported by the NSF of China (Grant No. 10501032), and the third by TRAPOYT and the NSF of China (Grant No. 10371046). E-mail address: lifubomath@yahoo.com.cn E-mail address: hqzheng@hust.edu.cn
196
M. Li, F-B. Li and Q. Zheng
where a L1 (R+ ) a scalar kernel, A is a closed linear operator on X with dense domain loc D(A) X, and f : R+ X is continuous. In 1980, Da Prato and Iannelli [3] rst introduced the notion of resolvent families, which now plays a central role in the theory of Volterra equations [6, 8, 10]. A strongly continuous family of bounded linear operators on X, {R(t)}t0, is called a resolvent family for (0.1) if R(t) commutes with A and satises the resolvent equation
t
R(t)x = x +
0
a(t s)R(s)Ax ds,
t 0, x D(A).
If the solution of (0.1) exists, it can be represented by the variation of parameters formula u(t) = d dt
t
R(t s)f (s) ds.

0
It is well known that when A generates a C0 -semigroup {S1 (t)}t0, then the abstract Cauchy problem u (t) = Au(t)(t 0), u(0) = x has a unique solution with initial value at x D(A), and this is saying,
t
S1 (t)x = x +
0
S1 (s)Ax ds,
x D(A),
which means that A generates a resolvent family with kernel a1 (t) 1; and similarly, when A generates a cosine function, then the second order equation u (t) = Au(t)(t 0), u(0) = x, u (0) = 0 has unique solutions which form a resolvent family
t
S2(t)x = x +
0
(t s)S2 (s)Ax ds,
x D(A)
with kernel a2 (t) = t. This motivates us to consider the resolvent family for kernel a(t) = t1 . This class of resolvent families was studied recently by Bajlekova [2], and we call it fractional (or -times) resolvent family in this paper. When 1 < < 2, such resolvent families interpolate between C0 -semigroups and cosine functions. For A = on R, the Volterra equation corresponding to such resolvent family just interpolates the heat equation and the wave equation (see [5, 7]). A number of papers are devoted to studying the generation of analytic semigroups by elliptic operators (see, e.g. [9, 11, 12, 13]). Since the class of fractional resolvent families is a generalization of semigroups and cosine functions, it is therefore natural to consider the generation of fractional resolvent families by elliptic operators. In this paper we consider elliptic operators on L2(), where is a bounded domain in Rn . The G rdings inequality and the numerical range method will be used. We show that a under very general conditions, elliptic operators with zero boundary condition can generate -times resolvent family for some (1, 2). And in some cases, such can be specied.
Preliminaries on fractional resolvent families
In this section we will introduce the notion of fractional resolvent families and quote some of their basic properties as lemmas, for the proofs we refer to [2]. Throughout this section, the operator A is densely dened and > 0.
Elliptic Operators with Variable Coefcients
197
Denition 1.1. A family {S(t)}t0 B(X) is called an -times resolvent family for A if the following conditions are satised: (a) S (t) is strongly continuous for t 0 and S (0) = I; (b) S(t)D(A) D(A) and AS (t)x = S(t)Ax for x D(A) and t 0; (c) for x D(A), S (t)x satises
t
S(t)x = x +
0
(t s)1 S(s)Ax ds, ()
t 0.
(1.1)
Denition 1.2. (a) An -times resolvent family {S(t)}t0 B(X) is called exponentially bounded if there are constants M 1 and 0 such that S(t) M et , t 0. (1.2)
If there is an -times resolvent family S(t) for A and satisfying (1.2), we write (A, S()) C (M, ) or A C (M, ). Also, set C () = M 0C (M, ) and C = 0 C (). (b) An -times resolvent family {S(t)}t0 is called analytic if S (t) admits an analytic extension to a sector 0 := {z C|| arg z| < 0 } for some 0 (0, /2]. A A (0 , 0) means that there is an -times resolvent family S (t) for A which can be extended analytically to 0 and for each < 0 and > 0 there is M = M (, ) such that S (z) M eRez , z . In addition, denote A () = 0 A (, ) and A = 0</2 A (). It is clear that the resolvent family S1 (t) is a C0 -semigroup and S2 (t) is a cosine function. Moreover, we have Lemma 1.3. Assume that A C for some > 2, then A B(X). The following is the generation theorem for -times resolvent families. Lemma 1.4. Let (0, 2], the following are equivalent: (a) A C (M, ); (b) ( , ) (A) and M n! dn 1 ( R(, A)) , n d ( )n+1 > , n N {0};
(c) ( , ) (A) and there exists a strongly continuous family {S(t)}t0 B(X) satisfying S(t) M et for t 0 such that
1R(, A)x =
0
et S(t)x dt,
> , x X.
And the generation and perturbation for analytic -times resolvent family are as follows
198
Lemma 1.5. Let (0, 2). Then A generates a bounded analytic -times resolvent family of angle 0 if and only if (A) (/2+0 ) and for all 0 < < 0 R(, A) C , || (/2+).
Lemma 1.6. Let (0, 2) and A A (, ). Let B be a closed operator satisfying D(B) D(A) and Bx a Ax + b x , x D(A). There exists > 0 such that if 0 a then A + B A . The following subordination principle is very important in the theory of -times resolvent families. Lemma 1.7. Let 0 < < 2, = /, 0. If A C () then A C ( 1/ ) and the following representation holds:
S(t)x =
0
t, (s)S (s)x ds,
t > 0,
where t, (s) := t (st ) and (z) is the Wright function dened by
(z) :=
n=0
(z)n 1 = n!(n + 2 ) 2i
1 exp( z ) d,
where 0 < < 1 and is a contour which starts and ends at and encircles the origin counterclockwise. Moreover, S (t) admits an analytic extension to the sector min{(),} , where () = (1/ 1)/2. Remark 1.8. When = 1 and = 2, it is in fact the Weierstrass formula relating C0 semigroup S1 with the cosine function S2 1 S1(t) = t
0
es
2 /4t
S2(s) ds,
t > 0.
Elliptic differential operators on L2
Throughout this section, we suppose that is a bounded domain in Rn with boundary of class C 2m . We will use the following notations. If = (1 , , n ) is any n-tuple of non-negative integers, we set || = 1 + n . Let x = (x1 , , xn ) be a vector in Rn , we set x = x1 xn . n 1 Similarly,
D = D1 1 Dnn ,
where Dk = i
, xk
k = 1, , n.
199
Let be a domain in Rn , u(x) be a complex valued function dened in . Various norm will be used in the sequel: u u and we will denote (u, v)m :=
||m 2
:= (
m,2
|u(x)|2 dx)1/2, D u 2 ,
||m
:=
D uD v dx.
Consider the differential operator of order 2m, A(x, D) =

||2m
a (x)D,
(2.1)
where the coefcients a satises (A1) for || = 2m, a C m (); for || 2m, a (x) have derivatives of order || m and uniformly bounded; (A2) the operator A(x, D) is strongly elliptic, that is, there exists a constant C > 0 such that a (x) C||2m , x , Rn . (2.2) Re (1)m
||=2m
Denition 2.1. For the strongly elliptic operator A(x, D), we dene the operator A by Au := A(x, D)u,
m for u D(A) := H 2m () H0 ().
Lemma 2.2. (G rdings inequality) If A(x, D) satises (A1) and (A2), then there exist a m constants c0 > 0 and 0 0 such that for u H 2m() H0 () Re(Au, u)0 c0 u
2 m,2
0 u
2 0,2 .
Lemma 2.3 ([9]). Let A(x, D) be strongly elliptic of order 2m. Then for every satisfying m 0 and every f L2 (), there exists a unique u H 2m () H0 () such that A(x, D)u + u = f . If B is a linear operator in a Banach space X its numerical range S(B) is the set S(B) = { Bx, x : x D(B), x, x = x = x = 1}. Lemma 2.4 ([9]). Let B be a closed linear operator with dense domain in X. Let S(B) be the numerical range of B and let be the complement of S(A) in C. If then B is one-to-one and has dense range. Moreover, if 0 is a component of satisfying (B) 0 = then the spectrum of B is contained in the complement S0 of 0 and R(, B) 1 d(, S(B)) ,
where d(, S(B)) is the distance of from S(B).
200
Lemma 2.5 ([9]). Let A(x, D) satisfy (A1) and (A2). Then A generates an analytic C0 -semigroup on L2 (). The main object of this section is to show that A also generates an -times resolvent family for some (1, 2). The proof follows the same lines of Theorem 2.7 in [9]. Theorem 2.6. Let A(x, D) satisfy (A1) and (A2). Then there is some (1, 2), such that A A . Proof. Let A0 = A + 0I, where 0 is given by Lemma 2.2, then we have Re(A0 u, u)0 c0 u
2 m,2,
m for u D(A0 ) = H 2m () H0 (). By integration by parts, we have for some constant b > 0, |Im(A0 u, u)0| |(A0 u, u)0| b u 2 . m,2
So the numerical range of A0 S(A0 ) = {(A0 u, u)0 : u D(A), u

2
= 1} 1 ,
where 1 = arctan(b/c0) < /2, where c0 is given by Lemma 2.2. Choose 1 < < /2, then for ( )c {0}, here ( )c means the complement of , we have d(, S(A0)) c ||. By Lemma 2.3, R( A0 ) = L2 () for > 0; and by Lemma 2.4, for > 0, A0 is c injective, so (, 0) (A0 ). Then ( )c {0} is in S(A0 ) and its intersection with (A0 ) is not empty. Again by Lemma 2.4, we obtain that (A0 ) ( )c {0} and for ( )c {0}, 1 R(, A0 ) d(, S(A0))1 . c || Since 2(1 1 /) > 1, if we pick 1 < < 2(1 1 /), then there exists > 0 such that (A0 ) (/2+) and R(, A0 ) 1 , c || (/2+) ,
for some constant c, so we have by Lemma 1.5 that A0 A , and thus A A by Lemma 1.6. From the proof of Theorem 2.6 and the subordination principle (Lemma 1.7), we have Theorem 2.7. Let A(x, D) satisfy (A1) and (A2), and the numerical range of A, S(A), be contained in + for some R and (0, /2). Then A A for all 0 < < 2(1 /). If the operator A is of the form Au =
||,||m
(1)||D (a (x)D u),
(2.3)
by using G rdings inequality in a quadratic integral form (see [1]), the smoothing condia tions can be improved.
201
Theorem 2.8. If A is given by (2.3) and satises (a) a is uniformly continuous on , || = || = m; and a is bounded and measurable for || + || 2m; (b) A is strongly elliptic, which in this case means that Re
||=||=m
a (x) + C||2m ,
x , Rn ,
for some constant C > 0; (c) the numerical range of A is contained in + for some R and (0, /2). Then A A for all 0 < < 2(1 /). Next we consider the strongly elliptic operator of second order in divergence form:
n n
Au =
j,k=1
D (ajk (x)D u) +
j=1
bj (x)Dj u + c(x)u,
(2.4)
where Dj u means the j-th derivative of u. Fattorini ([4]) has the following result: Lemma 2.9. Let A be given by (2.4) and strongly elliptic. Suppose that ajk , bj , c L () and (ajk (x)) are real symmetric matrices, then A generates a cosine function on L2 (). As a direct consequence, we have Corollary 2.10. Under the assumptions of Lemma 2.9, A generates an -times resolvent family for all (0, 2]. For second order elliptic operator with the highest coefcients not necessarily real generating -times resolvent family, we make the following assumption. Let A be given by (2.4). (S1) A is strongly elliptic. (S2) the coefcients ajk (x) C(), and there exists 0 < < /2 such that for all x , a(x) := (ajk (x)) = B(x) + D(x), where D(x) = diag(d1(x), , dn(x)) is a diagonal matrix with elements dj (x) (j = 1, , n), and B(x) is conjugate symmetric, i.e., B(x) = B(x) and the eigenvalues of B(x) are nonnegative. (S3) bj , c are complex measurable functions and bounded on . Theorem 2.11. Suppose that A satises (S1) (S3), then for all 0 < < 2(1 /), A A . Proof. Let P u =
n j k j,k=1 D (ajk (x)D u),
then ajk (x)Dk uDj u dx.
(P u, u)0 =
For xed x , let B = (bjk ) = B(x), its eigenvalues are 1, , n 0, and the corresponding eigenvectors are 1 , , n , respectively. Let j = j /|j |, which has the
202
form j = (j1, , jn ) for j = 1, , n, then 1, , n are orthogonal vectors and Bj = j j for j = 1, , n. Let T = (1, , n), then T is an invertible matrix and T BT = (1, , n ) B(1 , , n) = (1, , n ) (11, , nn ) = diag(1, , n) =: E, so that
n
bjk (x)j k = (1, , n)B(1 , , n )

j,k=1
= (1, , n)(T )1 ET 1(1, , n) = (T 1 (1, , n ) ) ET 1 (1, , n )

n
=
j=1
j |yj |2 0
where (y1 , , yn ) = T (P u, u)0 =
(1 , , n ) . On the other hand, since ajk (x)Dk uDj u dx
bjk (x)Dk uDj u dx +
dj (x)|Dj u|2 dx,
we know the numerical range of P falls into , thus it follows from Theorem 2.8 that P A when < 2(1 /). Next, by G rdings inequality, a |(P u, u)0| c1 u
1 so for u H0 () H 2() with u 2 2 1,2
c2 u 2, 2
= 1, we have
2
Pu Let Qu =
= sup{|(P u, v)0| : v L2(), v |(P u, u)0| c1 u

2 1,2 c2 .
= 1}
n j j=1 bj (x)D u
+ c(x)u, for every > 0,

2
Qu
= sup{(Qu, v)0 : v L2(), v sup{ + c
= 1}
bj u
2 1,2
Du
2
v
2
2 2
v
1,2
: v L (), v
2 )
= 1}
bj
+ c
1 ( 4c1 P u 2 + C() c1 u + which implies that Qu for all u

1 H0 () 2
bj
+ c
Pu
+ C() u
H 2(),
thus by Lemma 1.6, A = P + Q A .
203
Remark 2.12. The strongly elliptic condition (S1) is equivalent to saying that the matrices Re a(x) are positive denite, while the eigenvalues condition in (S2) also requires some 2 i , A2 = information on the imaginary part of ajk (x). For example, let A1 = i 2 2 3i , then Re A1 = Re A2 = diag(2, 2) are positive denite, but the eigenvalues 3i 2 of A1 are 3 and 1, while the eigenvalues of A2 are 5 and -1.
References
[1] S. Agmon, Lectures on Elliptic Boundary Value Problems, Van Nostrand Math. Studies 2, 1965. [2] E. Bajlekova, Fractional Evolution Equations in Banach Spaces, Ph.D. Dissertation, Einhoven University of Technology, 2001. [3] G.Da Prato and M. Iannelli, Linear integro-differential equation in Banach space, Rend. Sem. Math. Padova 62 (1980), 207-219. [4] H.O. Fattorini, Second Order Linear Differential Equations in Banach Spaces , North-Holland, 1985. [5] Y. Fujita, Integrodifferential equations which interpolate the heat equation and the wave equation, Osaka J. Math. 27 (1990), 309-321. [6] C. Lizama, Regularized solutions for Volterra equations, J. Math. Anal. Appl. 243 (2000), 278-292. [7] F. Mainardi, Fractional relaxation-oscillation and fractional diffusion-wave phenomena, Chaos, Solitons and Fractals 7 (1996), 1461-1477. [8] H. Oka, Linear Volterra equations and integrated solution families, Semigroup Forum 53 (1996), 278-297. [9] A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations , Springer-Verlag, Berlin, 1983. [10] J. Pr ss, Evolutionary Integral Equations and Applications, Monographs in Math. 87, u Birkha ser Verlag, 1993. u [11] H.B. Stewart, Generation of analytic semigroups by strongly elliptic operators, Trans. Amer. Math. Soc. 199 (1974), 141-162.
204
[12] H.B. Stewart, Generation of analytic semigroups by strongly elliptic operators under general boundary conditions, Trans. Amer. Math. Soc. 259 (1980), 299-310. [13] H. Tanabe, Equations of Evolution, Monographs and Studies in Math. 6, 1979.
In: Research on Evolution Equation Compendium. Volume 1 ISBN:978-1-61209-404-5 Editor: Gaston M. NGuerekata 2009 Nova Science Publishers, Inc.
E XISTENCE R ESULTS FOR P SEUDO A LMOST P ERIODIC D IFFERENTIAL , F UNCTIONAL , AND N EUTRAL I NTEGRAL E QUATIONS
Toka Diagana Department of Mathematics, Howard University, 2441 6th Street N.W., Washington D.C. 20059, USA
This article is dedicated to the memory of my dear uncle El Hadj Bocar Baba Diagana for having taught us the basic knowledge and for having guided our rst footsteps in the right direction.
Abstract This paper is a survey devoted to the authors recent results related to the existence and uniqueness of pseudo almost periodic solutions to some abstract differential equations and partial differential equations. Some of those results will be slightly modied and applied to some concrete problems. As main tools, we will make extensive use of the method of analytic semigroups, and both the Banach and Zimas xed-point principles. Applications include the study of pseudo almost periodic solutions to some boundary-value problems such as the heat equation with delay as well as the logistic differential equation.
Key Words: pseudo almost periodic function; almost periodic function; existence and uniqueness of a pseudo almost periodic solution; semilinear differential equation, analytic semigroup; Zima xed-point theorem, Banach xed-point theorem, logistic equation with delay, integral equation of advanced type, heat equation, functional differential equation AMS Subject Classication: 43A60; 35B15; 47B55.
E-mail address: tdiagana@howard.edu
206
Toka Diagana
1.
Introduction
Let (X, ) be a complex Banach space. An original question consists of studying pseudo almost periodic solutions to the semilinear differential equation u (t) = Au(t) + f (t, u(t)), t R, (1.1)
where A is a densely dened closed linear operator on X, and the forcing term f : R X X is a jointly continuous function. Now when A is the innitesimal generator of an analytic semigroup (respectively, a Hille-Yosida operator) and f : R X X is pseudo almost periodic, both the Banach xed-point principle and the method of analytic semigroups were used in [3, 9, 28] as a main tool to determine sufcient conditions for the existence and uniqueness of pseudo almost periodic solutions to Eq. (1.1). In contrast with the above-mentioned contributions, we will make use of the so-called Zimas xed-point principle [42] to deal with the existence of pseudo almost periodic solutions to Eq. (1.1). A slightly different version of Eq. (1.1) which has some interesting applications, especially in the delay case, consists of the functional-differential equation u (t) = Au(t) + f (t, u(h(t))), t R, (1.2)
where A is a densely dened closed (possibly unbounded) linear operator acting on X, h : R R is a continuous function with h(R) = R, and f : R X X is a jointly continuous function. In the case when A is the innitesimal generator of an analytic semigroup and f : R X X is pseudo almost periodic, the Zimas xed-point principle will be used to study the existence of pseudo almost periodic solutions to Eq. (1.2). One should stress that the study of pseudo almost periodic solutions to Eq. (1.2) was recently considered in Diagana [18] in the case when A is the innitesimal generator of an exponentially stable c0-semigroup. Here, we consider the case when A is the innitesimal generator of an analytic semigroup. Furthermore, various other formulations of Eq. (1.2) had been studied by several authors; among them are [22], [24], [29], [30], and [37] and the references therein. In the particular case when h(t) = t p with p > 0, one obtains the delay case, which is important in many applications. As a generalization of both Eq. (1.1) and Eq. (1.2) we also consider the existence of pseudo almost periodic solutions to their perturbed semilinear equation u (t) = Au(t) + f (t, Bu(t)), t R, (1.3)
where A is the innitesimal generator of an analytic semigroup acting on X, B : D(B) X X is a densely dened closed linear operator, and f : R X X is a jointly continuous function. As previously, under some additional assumptions on A, B, and f , the existence and uniqueness of a pseudo almost periodic solution to Eq. (1.3) is obtained by combining both semigroup methods and the Zimas xed-point principle. One should indicate that Eq. (1.3) was recently considered by Diagana et al.[17] with the help of both fractional powers of operators and the Banach xed-point theorem. Here, we consider its related abstract setting and use the Zimas xed point principle rather than the Banachs one to deal with those existence results.
Pseudo Almost Periodic Solutions
207
In addition to Eq. (1.1), Eq. (1.2), and Eq. (1.3) we also study pseudo almost periodic solutions to more general equation, which consists of the following abstract partial evolution equation d [u(t) + f (t, Bu(t))] = Au(t) + g(t, Cu(t)), t R dt (1.4)
where A is the innitesimal generator of an analytic semigroup acting on X, B, C are arbitrary densely dened closed linear operators on X, and f, g are some jointly continuous functions. Upon some appropriate assumptions, the existence and uniqueness of a pseudo almost periodic (mild) solution to Eq. (1.4) is obtained. One should mention that various formulations of Eq. (1.4) were extensively studied by Hern ndez et al. [25, 26, 27]. In a particular, in [27], both the method of fractional powers of operators and the Banach xedpoint principle were combined to study almost periodic and asymptotically almost periodic solutions to a slightly different version of Eq. (1.4) corresponding to the delay case. More recently, Diagana [15] had studied sufcient conditions for the existence and uniqueness of pseudo almost periodic solutions to Eq. (1.4) by combining both semigroup methods and the Banach xed-point principle. However, there, the operator A was assumed to be the innitesimal generator of an exponentially stable c0-semigroup. Moreover, the assumptions made there require much more regularity for the linear operator A, i.e., it would have been more realistic for applications to assume that A is the innitesimal generator of an analytic semigroup. Here, the latter assumption will be made. One should also point out that Eq. (1.4) has various applications. More precisely, one can make use of it to characterize the existence and uniqueness of pseudo almost periodic solutions to the class of partial differential equations of the form [u(t, x) + f (t, Bu(t, x))] = Au(t, x) + g(t, Cu(t, x)) t u(t, )| = 0 (1.5) (1.6)
for all t R with Rn is a (bounded) subset. Furthermore, Eq. (1.4) may also enable us to characterize pseudo almost periodic solutions to partial neutral functional differential equations [25], integro-differential equations [10, 26], and many others. At the end of this survey, we consider another problem, which is very interesting for applications, but not directly linked to the previous ones. Namely, we study some sufcient conditions for the existence of pseudo almost periodic solutions to the abstract integral equation of the form
u(t) = f (u(h1 (t))) +

t
Q(s, u(s), u(h2(s)))C(t s)ds + g(t)
(1.7)
for each t R, where f, g, h1, h2, C : R R are continuous functions with hi (R) = R for i = 1, 2, and Q : R R R R is jointly continuous. Setting, h1 (t) = h2 (t) = t p where p > 0 is a constant, in Eq. (1.7), one obtains the so-called neutral delay integral equation of advanced type
u(t) = f (u(t p)) +

t
Q(s, u(s), u(s p)C(t s)ds + g(t),
(1.8)
208
Toka Diagana
which was introduced in the literature by Burton [7] as an intermediate step while studying the existence and uniqueness of (periodic) bounded solutions to the logistic differential equation delay given by u (t) = au(t) + u (t p) q(t, u(t), u(t p)) (1.9)
where a > 0, 0 || < 1, and p > 0 are respectively constants. Here, we study pseudo almost periodic solutions to Eq. (1.7) and use it to deal with pseudo almost periodic to Eq. (1.9). These results were recently obtained in Diagana et al.[19]. Applications include among others, the existence and uniqueness of pseudo almost periodic solutions to some boundary value problems related to the heat with delay. The existence of almost automorphic, asymptotically almost automorphic, almost periodic, asymptotically almost periodic, and pseudo almost periodic solutions is among the most attractive topics in the qualitative theory of differential equations due to their signicance and applications in physics, mathematical biology, and many others. The concept of pseudo almost periodicity (p.a.p), which is the central question in this paper was rst initiated by Zhang in [39, 40, 41] and is a natural generalization of the of the classical (Bochner) almost periodicity (a.p). Thus this new concept is welcome to implement another existing generalization of almost periodicity, the so-called notion of asymptotically almost periodicity due to Fr chet, see, e.g., [23, 32, 33]. Moreover, let us e emphasis on the fact that the concept of pseudo almost periodicity is a special case of the well known Besicovitch almost periodicity, i.e., the space B1 (R, X), see, e.g., [34]. More details on those concepts and related issues can be found in [1, 2, 3, 9, 11, 23, 28, 33, 38, 39, 40, 41] and the references therein. In addition to [3, 9, 28] one should mention that the study of pseudo almost periodic solutions to differential equations, partial differential equations, functional differential equations with delay, has been of a great interest for many mathematicians in the last few years, see, e.g., [1, 2, 4, 11, 15, 16, 17, 19, 39, 40, 41] and the references therein. Here, we examine some sufcient conditions, which do ensure the existence and uniqueness of a pseudo almost periodic solution to Eq. (1.1)-Eq. (1.2)- Eq. (1.3), Eq. (1.4), Eq. (1.7), and Eq. (1.8) using the above-mentioned techniques. Let us recall some denitions and notations that we shall use in the sequel.
2.
Preliminaries
2.1. Pseudo Almost Periodic Functions

Let (X, ) be a Banach space and let (BC(X), ) denote the Banach space of all bounded continuous functions : R X endowed with the sup norm dened by
:= sup .
tR
Similarly, BC(R) where X is an open subset, denotes the space of all bounded continuous functions : R X.
209
Denition 1. [8] A function f BC(X) is called almost periodic if for each > 0, there exists l > 0 such that every interval of length l contains a number with the following property: f (t + ) f (t) < , t R. The number above is then called an -translation number of f , and the collection of such functions will be denoted AP (X). Example 1. The function f (t) = sin t+sin 2t is a classical example of an almost periodic function on R. To construct more almost periodic functions on R one can use the following. Proposition 1. Let f AP (R). If g L1 (R), then f g, the convolution of f with g on R, is almost periodic. Proof. Since f is continuous and g L1 (R), it is not hard to see that the function t (f g)(t) is continuous. Moreover, |(f g)(t)| f g 1 for each t R, where g 1 is the L1-norm of g, and therefore f g BC(R). It remains to prove that f g is almost periodic. First of all, note that when g 0 there is nothing to prove. From now on, we suppose g 0. Since f AP (R), for every > 0 there exists l > 0 such that for all R there exists [, + l ] with |f ( + ) f ()| In particular, the following holds g g for each R.
1
|f (t s + ) f (t s)| Now
+
for each = t s R.
1
(2.10)
(f g)(t + ) (f g)(t) =
{f (t + ) f (t )}g()d, t R.
Thus from Eq. (2.10) and the fact that g L1(R) it easily follows that (f g) (f g) and hence f g AP (R). Example 2. Combining Example 1 and Proposition 1 it easily follows that for each function K L1 (R), the function dened by
+
L(t) :=
K(t s)(sin s + sin 2s)ds, (t R)
is almost periodic.
210 Similarly
Toka Diagana
Denition 2. A function F BC(R ) is called almost periodic in t R uniformly in any K bounded subset if for each > 0 there exists l > 0 such that every interval of length l > 0 contains a number with the following property F (t + , x) F (t, x) < , (t R, x K).
Here again, the number above is called an -translation number of F, and the class of such functions will be denoted AP (R ). More details on properties of almost periodic functions : R X and as well as those of the form F : R X X can be found in the literature, especially in [8, 23, 32, 33, 38] and the references therein. Throughout the rest of the paper, we suppose that = X and set AP0 (X) := {f BC(X) : lim 1 T 2T
T
f (s) ds = 0}.
T
Similarly, let AP0 (R X) denote the collection of functions F BC(R X) such that 1 T 2T lim uniformly in u X. Example 3. Both f (t) = t2 1 with = 0, and g(t) = e|t| belong to AP0 (R). + 2
T
F (t, u) dt = 0
T
Proposition 2. Let f AP0 (R). If g L1 (R), then f g, the convolution of f with g on R, belongs to AP0 (R). Proof. From f AP0 (R) and g L1 (R) it is clear that f g BC(R). By assumption, 1 T 2T lim Now setting J(T ) := it follows that 1 2T
T T
|f (t)| dt = 0.
T +
1 2T
T T
|f (t s)| |g(s)| ds dt
|(f g)(t)| dt J(T )

T
= =
1 2T
|f (t s)| |g(s)| ds dt
T +
|g(s)|
+
1 2T 1 2T
|f (t s)| dt
T T s
ds ds
=
+
|g(s)|
|f (r)| dr
T s
|g(s)| T (s) ds,
Pseudo Almost Periodic Solutions where fT (u) =
211
T u 1 |f (r)| dr. 2T T u Clearly, fT (u) 0 as T . Next, since fT is bounded and g L1(R), using the Lebesgue dominated convergence theorem, it follows that lim J(T ) = 0, and hence T
f g AP0 (R). Example 4. Combining Example 3 and Proposition 2 it easily follows that for each function K L1 (R), the function dened by
+
M (t) :=
K(t s) ds, (t R) s2 + 2
belongs to AP0 (R). Denition 3. A function f BC(X) is called pseudo almost periodic if it can be expressed as f = g + , where g AP (X) and AP0 (X). The collection of such functions will be denoted by P AP (X). Note that the functions g and appearing in Denition 3 are respectively called the almost periodic and the ergodic perturbation components of f. Furthermore, the decomposition in Denition 3 is unique [39, 40, 40] Example 5. The function f (t) = L(t)+M (t) ( = 0) where L and M are respectively the given in Example 1 and Example 3, is pseudo almost periodic. As a consequence of Proposition 1 and Proposition 2 the following holds. Proposition 3. Let f P AP (R). If g L1 (R), then f g, the convolution of f with g on R, belongs to P AP (R). We now equip P AP (X) the collection of all pseudo almost periodic functions from R into X with the sup norm. It is well known that (P AP (X), . ) is a Banach space, see, e.g., [28]. Denition 4. A function f BC(R X) is called pseudo almost periodic in t R uniformly in x X if it can be expressed as f = g + , where g AP (R X) and AP0 (R X). The collection of such functions will be denoted by P AP (R X).
2.2. The Zimas Fixed-Point Theorem

We now recall the so-called Zimas xed-point theorem [42] that we shall use in the sequel. For that, consider a given Banach space (Y, Y ) equipped with both a binary relation and a mapping m : Y Y such that: (i) The relation is transitive;
212 (ii) The norm (iii)

Y
Toka Diagana is monotonic, i.e.,

Y
v, then u
for all u, v Y;
m(u) and m(u)
= u
for each u Y.
Y,
Theorem 4. [42] In the Banach space (Y, B : Y Y be mappings such that
, m) given above, let : Y Y and
(iv) B is a bounded linear operator with spectral radius r(B) < 1; (v) If u v, then Bu Bv for all u, v Y;
(vi) m(u v)
Bm(u v) for all u, v Y.
Then the equation u = u has a unique solution. Observe that Theorem 4 is more general than the classical Banach xed-point theorem and has recently found several applications, especially those dealing with the existence and uniqueness of solutions to differential equations, partial differential equations, functional differential equations, and integro-differential equations, see, e.g., Bugajewski and Zima [5, 6].
3.
Throughout the rest of this paper (X, ) stands for a (complex) Banach space.
3.1. Pseudo Almost Periodic Solutions To Eq. (1.1)

To deal with the existence and uniqueness of pseudo almost periodic solutions to Eq. (1.1), we require the following assumptions: (H.1) The function f : R X X, (t, u) f (t, u) is pseudo almost periodic in t R uniformly in u X, i.e., f = g + where g AP (R X) and AP0 (R X); and f satises Lipschitz condition in u X for each t R in the following sense f (t, u) f (t, v) L(t) . u v , t R, u, v X, where L : R [0, ) is a continuous function satisfying
+
L()d < ;
(H.2) The operator A is the innitesimal generator of an analytic semigroup T (t) such that there exist M > 0 and > 0 with T (t) M et for all t 0;
Pseudo Almost Periodic Solutions (H.3) If r() denotes the spectral radius of the bounded linear operator dened by
t
213
(u)(t) = M .
L()u()d, t R, u P AP (X),
then r() < 1, where M is the constant appearing in (H.2). Theorem 5. Suppose that assumptions (H.1)-(H.2)-(H.3) hold. Then Eq. (1.1) has a unique pseudo almost periodic solution. Proof. Let u be a bounded solution to Eq. (1.1). It is well known [3] that u can be explicitly expressed by
t
u(t) :=
T (t s)f (s, u(s)ds, t R.
Now if u P AP (X), then using (H.1) it follows that u = (u) + (u), where
t
(u)(t) :=
T (t s)g(s, u(s))ds, and

t
(u)(t) :=
T (t s)(s, u(s))ds,
where g AP (R X) and AP0 (R X). First of all, note that since f P AP (R X) and satises the Lipschitzs condition (H.1), for each v P AP (X), f (, v()) belongs to P AP (X), see, e.g., [9]. Moreover, g(, v()) and (, v(h())) are respectively the almost periodic and ergodic perturbation components of f . Next we show that (u) AP (X). Since g(, v()) AP (X), therefore, for each > 0 there exists > 0 such that for all there is [, + ] such that g(s + , u(s + )) g(s, u(s)) < M . , s R.
Considering (u)(s + ) (u)(s) and using (H.2) it easily follows that (u)(t + ) (u)(t) < , t R, and hence t (u)(t) is almost periodic. The next step consists of showing that t (u)(t) is in AP0 (X). Clearly, t (u)(t) is a bounded continuous function. Thus the remaining task consists of showing that 1 T 2T lim Using (H.2) it follows that 1 T T lim
T T
(u)(t) dt = 0.
T
(u)(t) dt I + J,
T
214 where I = lim M T 2T M T 2T

T
Toka Diagana
t
dt
T T T
e(ts) (s, u(s)) ds , and

T
J = lim
dt
T
e(ts) (s, u(s)) ds = 0.
It remains to show that I = J = 0. For that, we will make use of the following arguments: (A1) (, u()) AP0 (X), as we mentioned it above; (A2) K = sup (t, u(t)) < .
tR
Now I = = M T 2T lim M T 2T lim

T t
(t, u(t)) dt
T T T
e(ts) ds 1 [1 e(t+T )]
(t, u(t)) dt
T T
M 1 . lim T 2T = 0, by (A1). Similarly, M T 2T lim lim
(t, u(t)) dt
T
es (s, u(s)) ds
T
et dt
M . K T s e ds T 2T M .K 1 e2T = lim T 2T 2 = 0, by (A2), and hence (u) AP0 (X). Dene the nonlinear operator
t
1 T [e eT ]
(u)(t) :=
T (t s)f (s, u(s))ds, u P AP (X), t R.
In view of the above, maps P AP (X) into itself, and (u) and (u) are respectively the almost periodic and ergodic perturbation components of (u). To complete the proof, we have to show that : P AP (X) P AP (X) has a unique xed-point. For that, we consider the Banach space Y = (P AP (X), ) equipped with both the relation and the mapping m : P AP (X) P AP (X) dened by: If u, v P AP (X), u v if and ony if u(t) v(t) , t R,
Pseudo Almost Periodic Solutions and m(u) = u , i.e., (m(u))(t) = u(t) for all t R. Now (1) The binary relation is transitive; (2) m(u) and m(u) = u for each u P AP (X); and (3) The norm is monotonic, that is, u v, then u For all u, v P AP (X),
t
215
u, v P AP (X).
(u)(t) (v)(t)
T (t s) (f (s, u(s)) f (s, v(s))) ds M e(ts) L(s) u(s) v(s) ds

t
M.
L(s) u(s) v(s) ds,
and hence m((u) (v)) m(u v). We also observe that is increasing, that is, if u v, then u v for all u, v P AP (X). Therefore, using Theorem 4 it follows that the operator has a unique xed-point in P AP (X), which obviously is the only pseudo almost periodic solution to Eq. (1.1).
3.2. Pseudo Almost Periodic Solutions to Eq. (1.2)

To deal with the existence and the uniqueness of pseudo almost periodic solutions to Eq. (1.2) we need assumptions (H.1)-(H.2) given previously and the following: (H.4) The function h : R R is continuous, h(R) = R, and u(h) P AP (R) whenever u P AP (X) Theorem 6. Suppose that assumptions (H.1)-(H.2)-(H.3)-(H.4) hold. Then, Eq. (1.2) has a unique pseudo almost periodic solution. The proof of Theorem 6 is similar to that of Theorem 5. However, for the sake of clarity, we will give its main steps. Proof. Let u be a bounded solution to Eq. (1.2). It is well known [3] that u can be explicitly expressed by
t
u(t) :=
T (t s)f (s, u(h(s))ds, t R.
Now if u P AP (X), then using (H.1) it follows that u = (u) + (u), where
t
h (u)(t) :=
T (t s)g(s, u(h(s))ds, and

t
h (u)(t) :=
T (t s)(s, u(h(s))ds,
216
Toka Diagana
where g AP (R X) and AP0 (R X). As in the proof of Theorem 5, for each v P AP (X), the function f (, v(h())) belongs to P AP (X), by (H.4). Moreover, the functions g(, v(h())) and (, v(h())) are respectively the almost periodic and ergodic perturbation components of f . Next we show that (u) AP (X). Since g(, v(h())) AP (X), therefore, for each > 0 there exists > 0 such that for all there is [, + ] such that h (u)(t + ) h (u)(t) < , t R, and hence t (u)(t) is almost periodic. The next step consists of showing that t (u)(t) is in AP0 (X). Clearly, t (u)(t) is a bounded continuous function. Thus the remaining task consists of showing that 1 T 2T lim Using (H.2) it follows that 1 T 2T lim
T T T
h (u)(t) dt = 0.
T
(u)(t) dt I + J,
T t
where I = lim M T 2T
dt
T T T
e(ts) (s, u(h(s))) ds , and

T
J = lim
M T 2T
dt
T
e(ts) (s, u(h(s))) ds.
Using similar arguments as in the proof of Theorem 5 one can easily show that I = J = 0. Dene the nonlinear operator
t
h (u)(t) :=
T (t s)f (s, u(h(s)))ds, u P AP (X), t R.
Clearly, h maps P AP (X) into itself, and h (u) and h (u) are respectively the almost periodic and ergodic perturbation components of h (u). Consider the Banach space Y = (P AP (X), ) equipped with both the relation and the mapping m : P AP (X) P AP (X) dened by: If u, v P AP (X), u v if and ony if u(t) v(t) , t R,
and m(u) = u , i.e., (m(u))(t) = u(t) for all t R. Proceeding as previously, for all u, v P AP (X),
t
h (u)(t) h (v)(t)
T (t s) (f (s, u(h(s))) f (s, v(h(s)))) ds M e(ts) L(s) u(h(s)) v(h(s)) ds

t
M.
L(s) u(h(s)) v(h(s)) ds,
217
and hence m(h (u) h (v)) m(u v). Therefore, by Theorem 4 it follows that the operator h has a unique xed-point in P AP (X), which obviously is the only pseudo almost periodic solution to Eq. (1.2). Example 6. For p > 0, consider the heat equation with delay given by u 2u (t, x) + f (t, u(t p, x)) , t R, x [0, ] (3.11) (t, x) = t x2 u(t, 0) = u(t, ) = 0, t R, (3.12) where f (t, u) = a(t) sin u + b(t) sin( 2u) with a, b P AP (R) L1(R). Let X = L2[0, ] equipped with its natural topology and let A be the operator dened by D(A) = {u L2 [0, ] : u L2[0, ], u(0) = u() = 0} Au = u , u D(A). It is well-known [27] that A is the innitesimal generator of an analytic semigroup T (t) on L2[0, ] with T (t) et , t 0. Using [9, Proposition 2.2] it can be easily checked that f satises assumption (H.1) with L(t) = |a(t)| + 2|b(t)|. Now set t |a(s)| + 2|b(s)| u(s)ds ( u)(t) :=
for all t R and u P AP (R). Corollary 7. The boundary value problem Eq. (3.11)-Eq. (3.12) has a unique pseudo almost periodic solution whenever r( ) < 1.
4.
Pseudo Almost Periodic Solutions To Some Partial Evolution Equations
Throughout the rest of the paper, (Y, Y) denotes an arbitrary Banach space, which is continuously embedded into the ground space (X, ). Furthermore, the constant > 0 will denote the bound of such an embedding, i.e., u . u
Y,
u Y.
Examples of such Y include among others the following three cases: (i) The domain (D(A), norm dened by for each u D(A).
D(A))
of a closed operator A where

D(A)
D(A)
is the graph
:= u + Au
218
Toka Diagana
(ii) Let (0, 1). Suppose that A is a densely dened closed invertible linear operator on X. If A is the innitesimal generator of an analytic semigroup, as we previously mentioned, the fractional powers A of A exist. In this event, one sets Y = (D(A), ), where u for each u D(A ). (iii) Y = X. In this event, the linear operators B appearing in Eq. (1.3) must be bounded linear operators on X (see assumption (H.6) below).
:= u + Au

This subsection is devoted to the existence and uniqueness of a pseudo almost periodic solution to (1.3). In addition to assumption (H.1), our present setting requires the following assumptions: (H.1) The function f : R Y X, (t, u) f (t, u) is pseudo almost periodic in t R uniformly in u Y, and f : R Y Y satises Lipschitz condition in u Y for each t R in the following sense f (t, u) f (t, v)
Y
L(t) . u v , t R, u, v Y,
where L : R [0, ) is a continuous function satisfying

+
L()d < ;
(H.5) The operator A is the innitesimal generator of an analytic semigroup T (t) such that there exist constants M > 0 and > 0 with T (t)
B(Y)
M et, t 0.
B(Y,X)
(H.6) The linear operator B B(Y, X). Set B
(H.7) If r() denotes the spectral radius of the bounded linear operator dened by
t
(u)(t) := M . t R, u P AP (Y), then r() < 1.
. .
L(s)u(s)ds,
Lemma 8. Suppose that assumptions (H.1)-(H.5)-(H.6) hold. Dene the nonlinear operator by: For each P AP (Y),
t
()(t) :=
T (t s)f (s, B(s))ds.
Then maps P AP (Y) into itself.
219
Proof. First of all, let us check that is well-dened. Indeed, if P AP (Y), then t B(t) is in P AP (Y), by (H.6). And hence, by the composition theorem of pseudo almost periodic functions [3], the function () = f (, B()) belongs to P AP (Y) whenever P AP (Y). In particular, N = sup f (t, B(t)) Y < .
tR
Now
t
()(t)
=
t
T (t s)f (s, B(s))ds T (t s)f (s, B(s))

t
M.
Yds
e(ts) . f (s, B(s))

t
Yds
= M .N . = M .N .
0
e(ts) ds
+
e d
M .N < ,
and hence t ()(t) is well-dened. Since P AP (Y), one can write = p + q where p AP (Y) and q AP0 (Y), and hence () = (p) + (q) where
t
p(t) :=
T (t s)p(s)ds, and
t
q(t) :=
T (t s)q(s)ds.
Clearly, (p) AP (Y). Indeed, since p AP (Y), for every > 0 there exists > 0 such that for all there is [, + ] with p p where =
Y,
<.
. M Thus, using (H.1) it easily follows that (p)(t + ) (p)(t) Y < for each t R, and hence (p) (p) Y, < , that is, (p) AP (Y). Next, we show that t (q)(t) is in AP0 (Y). First, note that s (q)(s) is a bounded continuous function. Thus we have to show that 1 T 2T lim Clearly, 1 T 2T lim
T T
(q)(t)
T
Y dt
= 0.
(q)(t)
T
Ydt
I +J
220 where I := lim and J := lim M r 2T M T 2T
Toka Diagana
T t
dt
T r
e(ts) q(s)
Yds
dt
T
e(ts) q(s)
Yds.
We must show that I = J = 0. To this end, notice that C = q Clearly, I = M T 2T lim

T t
Y,
< .
q(t)
T T
Ydt
e(ts) ds
T t+T
M = lim T 2T M lim T 2T M = lim r 2T =
q(t)
T T
Ydt 0 +
e d e d
0
q(t)
T T
Ydt
q(t)
T T
Y dt
M 1 . lim T 2T = 0. Similarly, J = M .C T 2T lim M .C T 2T lim lim

T
q(t)
T
Y dt
dt
T T t+T +
e d e d
2T
dt
T 2T
M .C .e
= 0, and hence (q) belongs to AP0 (Y). Theorem 9. Suppose that assumptions (H.1)-(H.5)-(H.6)-(H.7) hold. Then Eq. (1.3) has a unique pseudo almost periodic (mild) solution. Proof. From Lemma 8 it is clear that : P AP (Y) P AP (Y) where
t
(u)(t) =
T (t s)f (s, Bu(s))ds
is well-dened.
221
Consider the Banach space M = (P AP (Y), Y,) equipped with both the relation and the mapping m : P AP (Y) P AP (Y) dened by: If u, v P AP (Y), u v if and ony if u(t)
Y Y
v(t)
Y,
t R,
and m(u) = u Y, i.e., (m(u))(t) = u(t) Now, for all , P AP (Y),

t
for all t R.
()(t) ()(t)
=
t
T (t s) [f (s, B(s)) f (s, B(s))] ds

Y
T (t s) [f (s, B(s)) f (s, B(s))]

t
Y ds
M. M. M.
L(t)e(ts) (s) (s) ds

t
. .
t
L(t)e(ts) (s) (s) L(t) (s) (s)

Yds
Yds
. .
and hence m((u) (v)) m(u v). We also observe that is increasing, that is, if u v, then u v for all u, v P AP (Y). Therefore, by Theorem 4 it follows that the operator has a unique xed-point in P AP (Y), which obviously is the only pseudo almost periodic solution to Eq. (1.3). Corollary 10. Suppose that assumptions (H.1)-(H.5)-(H.6)-(H.7) hold in the case where Y = X, and B B(X). Then Eq. (1.3) has a unique pseudo almost periodic (mild) solution.

In what follows, BC(R, Y) will denote the collection of bounded continuous functions. It is equipped with the Y-sup norm dened by u ,Y := sup u(t) Y .
tR
Denition 5. A function u BC(R, X) is said to be a bounded mild solution to Eq. (1.4) provided that the function s AT (t s)f (s, Bu(s)) is integrable on (, t) for each t R and
t t
u(t) = f (t, Bu(t))
AT (t s)f (s, Bu(s))ds +
T (t s)g(s, Cu(s))ds
for each t R. The existence result of this subsection requires assumptions (H.8)-(H.9)-(H.10) given below.
222
Toka Diagana
(H.8) The operator A is the innitesimal generator of an exponentially stable semigroup (T (t))t0 such that there exist constants M > 0 and > 0 with T (t)
B(Y)
M et, t 0.
Furthermore, the function AT () dened from (0, ) into B(Y) is strongly (Lebesgue) measurable and there exist a nonincreasing function : R [0, ) and
a constant > 0 with :=

0
es (s)ds < such that

B(Y,X)
AT (s)
es . (s), s > 0.
(H.9) The functions f, g : R Y X, (t, u) f (t, u), g(t, u) are pseudo almost periodic in t uniformly in u, and f, g : R Y Y are Lipschitz, i.e., there exist constants Kf , Kg > 0 such that f (t, u) f (t, v)
Y
Kf . u v , and
Y
g(t, u) g(t, v) for all t R, and u, v Y.
Kg . u v
(H.10) The linear operators B, C B(Y, X) with max( B

B(Y,X),
B(Y,X))
In addition to Lemma 8 (where f and B should be respectively replaced by g and C), the proof of the main result of this subsection requires the following technical lemma: Lemma 11. Suppose that assumptions (H.8)-(H.9)-(H.10) hold. Dene the nonlinear operator 1 by: For each P AP (Y),
t
( )(t) =
AT (t s)f (s, B(s))ds
Then maps P AP (Y) into itself. Proof. Let P AP (Y). Since B B(Y, X) it follows that the function t B(t) belongs to P AP (Y). Again, using the composition theorem of pseudo almost periodic functions [3] it follows that () = f (, B()) is in P AP (Y) whenever P AP (Y). In particular, N = sup f (t, B(t)) Y < .
tR
Arguing as in the proof of Lemma 8 it can be easily shown that 1 is well-dened. Now write = w + z, where w AP (Y) and z AP0 (Y), and hence 1 = (w) + (z) where
t t
w(t) :=
AT (t s)w(s)ds, and z(t) :=
AT (t s)z(s)ds.
223
Clearly, (w) AP (Y). Indeed, since w AP (Y), for every > 0 there exists > 0 such that for all there is [, + ] with the property: w w Y < . where 1 = . Arguing as in the proof of Lemma 8 it can be easily follows that (w) AP (Y). Now 1 2r
r
(z)(t)
r
Y dt
1 2r 1 2r 1 2r +
r r r
AT (t s)
t
B(Y,X)
z(s)
Y ds dt
e(ts) . (t s) z(s)
r r r
Y ds dt
e(ts) (t s) z(s)
r r r
Y dt ds
1 e(ts) (t s) z(s) Ydtds 2r r s rs z ,Y r (r s) . e d ds 2r rs r + z(s) Y dt 2r r . z ,Y r z(s) Y dt. + 2r 2r r Clearly, lim 1 r 2r

r
(z)(t)
r
Y dt
= 0, and hence z AP0 (Y).
Theorem 12. Suppose that assumptions (H.8)-(H.9)-(H.10) hold. Then Eq. (1.4) has a unique pseudo almost periodic (mild) solution whenever := M . Kg + Kf .
1+
0
es (s)ds
< 1.
Proof. Dene the nonlinear operator : P AP (Y) P AP (Y) by

t
(u)(t) = f (t, Bu(t))

t
AT (t s)f (s, Bu(s))ds
T (t s)g(s, Cu(s))ds.
For all , P AP (Y), f (t, B(t)) f (t, B(t))

Y
Kf . B(t) B(t) Kf . Kf . . (t) (t) .

,Y , Y,
by (H.9)-(H.10).
224
Toka Diagana
Using the composition theorem of pseudo almost periodic functions [3] it follows that the operator dened by (u)(t) = f (t, Bu(t)) maps P AP (Y) into P AP (Y). In view of the above, and both Lemma 8 and Lemma 11 it easily follows that the nonlinear operator is well-dened, that is, it maps P AP (Y) into P AP (Y) . To complete the proof, we will successively approximate the expressions and in norm Y and next apply the classical Banach xed-point principle.
t
( )(t) ( )(t)
=
t
AT (t s) [f (s, B(s)) f (s, B(s))] ds

Y
AT (t s) [f (s, B(s)) f (s, B(s))]

t
ds
Kf . = Kf . = Kf . Similarly,
t
. .
,Y
.
+
e(ts) (t s)ds e ()d

0
,Y
. .
,Y .
()(t) ()(t)
=
t
T (t s) [g(s, C(s)) g(s, C(s))] ds

Y
T (t s) [g(s, C(s)) g(s, C(s))]

t
Y ds
M . Kg . = M . Kg .
. .
,Y . ,Y .
e(ts) ds
Clearly, for all , P AP (Y), () ()

,Y
,Y .
Consequently, if < 1, then the operator has a unique xed-point, which obviously is the only pseudo almost periodic solution to Eq. (1.4).
Corollary 13. Suppose that assumptions (H.8)-(H.9) hold in the case where Y = X, and B, C B(X). Then Eq. (1.4) has a unique pseudo almost periodic (mild) solution whenever M . Kg + Kf . 1 + := es (s)ds <1 0 where = max( B , C ).
225
Example 7. X = L2[0, ] equipped with its natural topology. For p > 0, consider the boundary value problem with delay given by 2u [u(t, x) + f (t, u(t p, x))] = (t, x) + g (t, u(t p, x)) t x2 u(t, 0) = u(t, ) = 0, t R,
(4.13) (4.14)
where f, g : R L2 [0, ] L2 [0, ] are jointly continuous functions. To illustrate Theorem 12 we examine the existence and uniqueness of pseudo almost periodic solutions to the heat equation given by the system Eq. (4.13)- Eq. (4.14). For that, as in Example 6, suppose that A is the linear operator dened by D(A) = {u L2 [0, ] : u L2[0, ], u(0) = u() = 0} Au() = u = u (), u() D(A). and take Y = D(A) equipped with the graph norm of A. Note that the linear operator A has many properties, in particular has a discrete spectrum with eigenvalues of the form n = n2 , n N with corresponding normalized eigen2 sin(nt). Moreover, the following properties hold: functions given by zn (t) := (a) {zn : n N} is an orthonormal basis for L2[0, ];
(b) Au =
n=1
n2 u, zn zn , for each u D(A);
(c) T (t)u =
n=1
en
2t
u, zn zn , for each u L2 [0, ];
(d) Let X denote the space D((A)) endowed with the graph norm . It can be shown that X X is continuous for 0 < 1 and there exist some constants C , > 0 such that T (t)
B(X,X)
C e t , t > 0. t
Moreover, it is possible to dene fractional powers of A as a closed linear operator, see, e.g., [25, 26, 27, 31]. In particular,
(e) For u L2 [0, ] and (0, 1), (A) u =

n=1
1 u, zn zn ; n2
(f) The operator (A) : D((A)) L2[0, ] L2[0, ] is given by
(A) u =
n=1
n2 u, zn zn ,
226 for each u D((A)) where
Toka Diagana
D((A) ) = {u() L2 [0, ] :

n=1
n2 u, zn zn L2 [0, ]}.
Theorem 14. Under assumptions (H.9), the heat-like equation with delay given by the system Eq. (4.13)-Eq. (4.14) has a unique solution whenever Kg + 2Kf < 1.
5.
Case of the Logistic Equation With Delay
Throughout the rest of this subsection, we suppose that X = R equipped the classical absolute value. Note however that when we deal with the pseudo almost periodicity of the kernel Q it would be more convenient to choose X = R R, see (H.13). Our setting requires the following assumptions: (H.11) The function f, g : R R are pseudo almost periodic and f satises, |f (x) f (y)| . |x y|, for all x, y R; (H.12) The function hi : R R is continuous, hi (R) = R, and u(hi ) P AP (R) (i = 1, 2) whenever u P AP (R); (H.13) The function Q : R (R R) R, (t, x, y) Q(t, x, y) is pseudo almost periodic in t R uniformly if (x, y) R R. Setting Q = Q1 + Q2 with Q1 AP (R R R, R) and Q2 AP0 (R R R, R), we suppose that Q2 (, v(), v(h2())) L1 (R) for each v P AP (R) where h2 is the function appearing in (H.12). Furthermore, there exists 0 k 1 such that |Q(t, x, y) Q(t, w, z)| (k . |x w| + (1 k) . |y z|) for all x, y, z, w R;
0 < 1,
(H.14) 0 <
0
|C(s)|ds = C0 < .
Our main result requires the following technical lemma: Lemma 15. Under assumptions (H.12)-(H.13)-(H.14), the function dened by
u(t) :=
t
Q(s, u(s), u(h2(s)))C(t s)ds
maps P AP (R) into itself.
227
Proof. Let u P AP (R). First of all, note that t u(h2 (t)) is pseudo almost periodic, by (H.12). Using (H.13) it follows that s Q(s, u(s), u(h2(s))) is pseudo almost periodic, see, e.g., [3, 9]. Now write Q = Q1 + Q2 where Q1 AP (R R2 , R) and Q2 AP0 (R R2 , R). Consequently, u = 1 (u) + 2 (u) where
1 u(t) :=
t
Q1 (s, u(s), u(h2(s)))C(t s)ds
and 2 u(t) :=
t
Q2 (s, u(s), u(h2(s)))C(t s)ds.
To complete the proof, it remains to prove that 1 u AP (R) and 2 u AP0 (R). Since Q1 (, u(), u(h2())) AP (R), for each > 0, there exists > 0 such that for all , there is [, + ] with |Q1(s + , u(s + ), u(h2(s + ))) Q1(s, u(s), u(h2(s)))| < for each s R. Note that 1 u(t + )) =
t
C0
(5.15)
Q1(r + , u(r + ), u(h2(r + )))C(t r)dr, by setting
r = s . Considering 1 u(t + ) 1 u(t) it easily follows that |1 u(t + ) 1 u(t)| < , t R,
by Eq. (5.15) and (H.14), and hence 1 (u) AP (R). The next step consists of showing that 2 u AP0 (R). It is clear that s 2 (u)(s) is a bounded continuous function. Thus, it remains to show that 1 r 2r lim Clearly, 1 r 2r lim where I := lim 1 r 2r
r r r r
|2 u(t)| dt = 0.
r
|2 u(t)| dt I + J,
r
dt
r r t
|Q2(s, u(s), u(h2(s)))| . |C(t s)|ds , and
J := lim
1 r 2r
dt
r r
|Q2(s, u(s), u(h2(s)))| . |C(t s)|ds.
To show that I = J = 0, we make use of the following arguments:

s r+s
(A0)
r
|C(t s)|dt =
0
|C(v)|dv C0 for all r + s 0;
(A1) Q2(, u(), u(h2())) AP0 (R);
228 (A2) Q2(, u(), u(h2())) L1 (R).
Toka Diagana
Indeed, by changing the order of integration we obtain: I = 1 r 2r C0 lim r 2r = 0, lim

r s
|Q2(s, u(s), u(h2(s)))| ds

r r r
|C(t s)|dt
|Q2(s, u(s), u(h2(s)))| ds

r
by (A0)-(A1). Similarly, J = 1 r 2r 1 = lim r 2r lim

r r s+r r
|Q2(s, u(s), u(h2(s)))| ds

r
|C(t s)|dt
r s+r
|Q2(s, u(s), u(h2(s)))|ds

r sr
|C(v)|dv
= 1 2r
lim
|Q2(s, u(s), u(h2(s)))| . r (s) ds,
where r (s) =
|C(v)|dv.
sr
s+r C0 , by |C(v)|dv C0 for all s r. And hence r (s) 0 2r sr as r . Since Q2 (, u() u(h2 ())) L1 (R) it follows that
Clearly, r (s)
r r
lim
|Q2(s, u(s), u(h2(s)))| . r (s) ds = 0,
by (A.2) and the Lebesgue dominated convergence theorem. And therefore 2 u AP0 (R). Theorem 16. Under assumptions (H.11)-(H.12)-(H.13)-(H.14), Eq. (1.7) has a unique pseudo almost periodic solution whenever + C0 < 1. Proof. Let u P AP (R). Dene the nonlinear operator
(u)(t) := f (u(h1 (t))) +

t
Q(s, u(s)u(h2(s)))C(t s)ds + g(t),
t R.
First of all, let us mention that f (u(h1 ())) P AP (R), which follows immediately from the composition theorem of pseudo almost periodic functions in [3, 9]. Thus, in view of the previous facts and Lemma 15, it easily follows that maps P AP (R) into itself and that 1 u and 2 u are respectively the almost periodic and ergodic perturbation components of (u). To complete the proof, we must show that : P AP (R) P AP (R) has a unique xed-point. For u, v P AP (X),
Pseudo Almost Periodic Solutions |(u)(t) (v)(t)|
229
uv
+
t
|(Q(s, u(s), u(h2(s))) Q(s, v(s), v(h2(s))))| . |C(t s)|ds
uv uv and hence
+
t
[k|u(s) v(s)| + (1 k)|u(h2(s) v(h2(s))] . |C(t s)|ds

. t
+ uv
|C(t s)|ds, ( + C0 ) . u v
.
(u) (v)
Therefore, by the Banach xed-point principle, the operator has a unique xed point whenever + C0 < 1, which obviously is the only pseudo almost periodic solution to Eq. (1.7).
Setting h1 (t) = h2 (t) = t p, one can easily see that (H.12) holds, and hence the next corollary is a straightforward consequence of Theorem 16. (In assumption ( H.13), we suppose that the ergodic component Q2 of Q is given such that Q2 (, v(), v(p)) L1 (R) for each v P AP (R).) Corollary 17. Under assumptions (H.11)-(H.13)-(H.14), Eq. (1.8) has a unique pseudo almost periodic solution whenever + C0 < 1. The rest of this subsection is devoted to the existence and uniqueness of pseudo almost periodic solutions to Eq. (1.9). In what follows we dene the function q (t, x, y) := q(t, x, y) ay for all t R and x, y R, where a, and q are respectively the constants and the function appearing in Eq. (1.9). We require the following assumption: (H.5) The function q : R(RR) R, (t, x, y) q (t, x, y) is pseudo almost periodic in t R uniformly if (x, y) RR. Setting q = q1 +q2 where q1 AP (RRR, R) and q2 AP0 (R R R, R), we suppose that q2 (, v(), v( p))) L1 (R) for each v P AP (R). Furthermore, suppose that , a > 0, and |q(t, x, y) q(t, w, z)| (1 a)|x w| for all t, x, y, z, w R. Theorem 18. Under assumption (H.5), the logistic equation, Eq. (1.9), has a unique pseudo 1 almost periodic solution whenever + < 1. a
230
Toka Diagana
Proof. One follows along the same lines as in [7]. We are interested in bounded solutions only. Thus if u is a bounded solution to Eq. (1.9), then d [(u(t) u(t p)) eat] = [au(t p) q(t, u(t), u(t p))]eat. dt Clearly,
u(t) = u(t p) +
t
[q(s, u(s), u(s p)) au(s p)]ea(ts) ds,
(5.16)
for each t R, by lim [(u(t) u(t p)) eat] = 0 (u is bounded).

t
To complete the proof, in Eq. (1.7), take f (t) = t, h1 (t) = h2 (t) = t p, C(t) = eat , g(t) = 0, and Q(t, u(t), u(t p)) = q(t, u(t), u(t p)), t R, and follow along the same lines as in the proof of Theorem 16.
References
[1] E. Ait Dads, K. Ezzinbi, and O. Arino, Pseudo Almost Periodic Solutions for Some Differential Equations in a Banach Space, Nonlinear Anal (TMA). 28 (1997), no. 7, pp. 11411155. [2] E. Ait Dads and O. Arino, Exponential Dichotomy and Existence of Pseudo Almost Periodic Solutions of Some Differential Equations, Nonlinear Anal (TMA). 27 (1996), no. 4, pp. 369386. [3] B. Amir and L. Maniar, Composition of Pseudo-Almost Periodic Functions and Cauchy Problems with Operator of Nondense Domain, Ann. Math. Blaise Pascal. 6 (1999), no. 1, pp. 111. [4] D. Bugajewski, T. Diagana, and C. M. Mahop, Asymptotic and Pseudo Almost Periodicity of Convolution Operator and Applications to Differential and Integral Equations, Zeitschrift fur Analysis und ihre Anwendungen , 2005, to appear. [5] D. Bugajewski and M. Zima, On the Darboux Problem of Neutral Type, Bull. Austral. Math. Soc. 54 (1996), no. 3, pp. 451458. [6] D. Bugajewski and M. Zima, On the Spectral Radius of Linearly Bounded Operators and Existence Results for Functional-Differential Equations. Dynamical systems and differential equations (Wilmington, NC, 2002). Discrete Contin. Dyn. Syst. 2003, suppl., pp. 147155. [7] T. A. Burton, Basic Neutral Integral Equations of Advanced Type, Nonlinear Anal. (TMA) 31 (1998), no. 3/4, pp. 295310.
231
[8] C. Corduneanu, Almost Periodic Functions, Second Edition, Chelsea - New York, 1989. [9] C. Cuevas and M. Pinto, Existence and Uniqueness of Pseudo Almost Periodic Solutions of Semilinear Cauchy Problems with Non-dense Domain, Nonlinear Anal (TMA). 45 (2001), pp. 7383. [10] W. Desch, R. Grimmer, Ronald, and W. Schappacher, Well-Posedness and Wave Propagation for a Class of Integro-differential Equations in Banach Space. J. Differential Equations 74 (1988), no. 2, pp. 391411. [11] T. Diagana, Pseudo Almost Periodic Solutions to Some Differential Equations, Nonlinear Anal (TMA). 60 (2005), no. 7, pp. 12771286. [12] T. Diagana, Some Remarks on Some Second-Order Hyperbolic Differential Equations, Semigroup Forum. 68 (2004), no. 3, pp. 357364. [13] T. Diagana, C. M. Mahop and G. M. NGu r kata, Pseudo Almost Periodic Solution to ee Some Semilinear Differential Equations. Mathematical and Computer Modelling 43 (2006), No. 1-2., pp. 8996. [14] T. Diagana and G. M. NGu r kata, Pseudo Almost Periodic Mild Solutions To Hyperee bolic Evolution Equationa in Abstract Intermediate Banach Spaces. Applicable Analysis (2006), to appear. [15] T. Diagana, Existence and Uniqueness of Pseudo Almost Periodic Solutions to Some Classes of Partial Evolution Equations. Nonlinear Analysis (TMA), 2005, to appear. [16] T. Diagana, Existence and Uniqueness of Pseudo Almost Periodic Solutions to Semilinear Differential Equations and Applications. Nonlinear Analysis (TMA), 2005, to appear. [17] T. Diagana, C. M. Mahop, G. M. NGu ekata, and B. Toni, Existence and Uniqueer ness of Pseudo Almost Periodic Solutions to Some Classes of Semilinear Differential Equations and Applications. Nonlinear Analysis (TMA), 2005, to appear. [18] T. Diagana, Existence and Uniqueness of Pseudo Almost Periodic Solutions to Some Functional Differential Equations. Functional Differential Equations, 2005, to appear. [19] T. Diagana, C. M. Mahop, Pseudo Almost Periodic Solutions to A Neutral Delay Integral Equation. Cubo Journal, 2005, to appear. [20] T. Diagana, G. M. NGu r kata, and N. V. Minh, Almost Automorphic Solutions of ee Evolution Equations, Proc. Amer. Math. Soc. 134 (2004), no. 11, pp. 32893298. [21] T. Diagana and G. M. NGu r kata, Almost Automorphic Solutions to Semilinear ee Evolution Equations, Functional Differential Equations, 2005, to appear. [22] A. Feldstein, A. Iserles, D. Levin, Embedding of Delay Equations into an Innite Dimensional ODE System, J. Differential Equations. 117 (1995), no.1, pp. 127150.
232
Toka Diagana
[23] A. M. Fink, Almost Periodic Differential Equations , Lecture Notes in Mathematics 377, Springer-Verlag, New York - Berlin, 1974. [24] L. Fox, D. F. Mayers, J. R. Ockendon, A. B. Tayler, On a Functional Differential Equation, J. Inst. Math. Appl. 8 (1971), p. 271307. [25] E. Hern ndez and H. R. Henrquez, Existence of Periodic Solutions of Partial neutral a Functional Differential Equations with Unbounded Delay. J. Math. Anal. Appl 221 (1998), no. 2, pp. 499522. [26] E. Hern ndez, Existence Results for Partial Neutral Integro-differential Equations a with Unbounded Delay. J. Math. Anal. Appl 292 (2004), no. 1, pp. 194210. [27] E. Hern ndez M., M. L. Pelicer, and J. P. C. dos Santos , Asymptotically almost peria odic and almost periodic solutions for a class of evolution equations, Electron. J. Diff. Eqns 2004(2004), no. 61, pp. 115. [28] H. X. Li, F. L. Huang, and J. Y. Li, Composition of pseudo almost-periodic functions and semilinear differential equations. J. Math. Anal. Appl. 255 (2001), no. 2, pp. 436 446. [29] Z. Jackiewicz, A Note on Existence and Uniqueness of Solutions to Neutral Functional Differential Equations with State Dependent Delays, Comment. Math. Univ. Carolin. 36 (1995), no. 1, p. 1517. [30] T. Kato and J. B. McLeod, The Functional-Differential Equation y (x) = ay(x) + by(x), Bull. Amer. Math. Soc. 77 (1971), p. 891937. [31] A. Lunardi, Analytic semigroups and optimal regularity in parabolic problems , PNLDE Vol. 16, Birkh auser Verlag, Basel, 1995. a [32] G. M. NGu r kata, Almost Automorphic Functions and Almost Periodic Functions in ee Abstract Spaces, Kluwer Academic / Plenum Publishers, New York-London-Moscow, 2001. [33] G. M. NGu r kata, Topics in Almost Automorphy, Springer, New York, 2005. ee [34] A. A. Pankov, Bounded and Almost Periodic Solutions of Nonlinear Operator Differential Equations. (Translated from Russian by V. S. Zjackovski and A. A. Pankov). Mathematics and Applications (Russian Series), v. 55. Kluwer Academic Publishers, 1985. [35] A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations, Springer-Verlag, 1983. [36] M. Renardy and R. C. Rogers, An Introduction to Partial Differential Equations , Texts in Appl. Math. 13, Springer-Verlag, New York-Berlin- Heidelberg-LondonParis, 1992. [37] B. Rzepecki, On Some Functional-Differential Equations. Glas. Mat. Ser. III 19(39) (1984), no. 1, pp. 7382.
233
[38] S. Zaidman, Topics in Abstract Differential Equations Pitman Research Notes in Mathematics Ser. II John Wiley and Sons, New York, 1994-1995. [39] C. Y. Zhang, Pseudo Almost Periodic Solutions of Some Differential Equations, J. Math. Anal. Appl. 151 (1994), pp. 6276. [40] C. Y. Zhang, Pseudo Almost Periodic Solutions of Some Differential Equations II, J. Math. Anal. Appl. 192 (1995), pp. 543561. [41] C. Y. Zhang, Integration of Vector-Valued Pseudo Almost Periodic Functions, Proc. Amer. Math. Soc. 121 (1994), pp. 167174. [42] M. Zima, A Certain Fixed-Point Theorem and Its Application to Integral-Functional Equations, Bull. Austral. Math. Soc.48 (1992), pp. 179186.
A N A SYMPTOTIC M ODEL OF A N ONLINEAR K ELVIN -VOIGT V ISCOELASTIC P LATE

Robert P. Gilbert and Robert J. Ronkese Department of Mathematical Sciences University of Delaware, Newark, DE 19716
Abstract In the 1970s, C OWIN and H EGEDUS [5, 6] introduced an adaptive elasticity model for bone deposition and reabsorption (remodeling). In 1998, F IGUEIREDO and T RA BUCHO [8] published a paper on an asymptotic model of an adaptive elastic rod. Afterwards, M ONNIER and T RABUCHO [9] proved the existence and uniqueness of a solution for this model; T RABUCHO [10] himself extended this result to a model that contains non-linear strain terms along with linear strain terms in a rate remodeling equation. Here, it is proposed to introduce a Kelvin-Voigt model for the plate that includes non-linear terms rate remodelling terms with respect to strain and the time derivative of strain. As many biomechanists consider cancellous bone to be structure consisting of both rode and plates, adding the study of the plate permitsre a more realistic modeling of the trabeculae of spongy bone. Recent studies have suggested that the dissipation of acoustic energy in cancellous bone interrogation via ultrasound is due to the viscous properties of the trabeculae, rather than that of interstitial blood and marrow [1]. This implies use of a Kelvin-Voigt model for both the plate-like and rod-like trabeculae.
1.
Introduction
The model of C OWIN and H EGEDUS [5, 6] has a rate remodeling (deposition and reabsorption) equation of the form: = a() + Akm ()ekm where a() is a constitutive function, Akm is a rate remodeling coefcient and ekm is the strain tensor. This equation was coupled with equations describing the equilibrium of forces, both of the body and of the surfaces, to be found on the bone material. Later, an existence and uniqueness theorem was postulated in detail in a paper by M ONNIER and T RABUCHO [9]. However, a subsequent paper by T RABUCHO [10] suggested adding a nonlinear term cijkl ekl eij for the strain tensor in the above equation and then proved existence and uniqueness. In this paper, a Kelvin-Voigt model of a nonlinear viscoelastic plate is considered with the more complicated rate remodeling equation:
kl kl = a ()+A ()e (u )+A ( )e (u)+c ()e (u )e (u )+d ( )e (u)e (u ) kl kl kl ijkl kl ij ijkl ij
This research was supported in part by NSF grant INT 0438765 and by an Alexander v. Humboldt Senior Scientist Award at the Ruhr University Bochum.
236
Robert P. Gilbert and Robert J. Ronkese
with Akm as a rate remodeling coefcient and dijkl as a viscosity coefcient that are grouped with tensors that are the time derivatives of the strain. This model includes the time rate of change of strain, which we believe represents a truer picture of the remodeling process. This paper introduces the asymptotic model and shows how asymptotic expansions of both the displacement vector and the stress tensor can be simplied. Equations of the displacement-stress model are derived as well as a characterization of the rst term of the displacement expansion. The rst few terms of the asymptotic expansion of the time rate of change of the volume fraction of spongy bone material is derived.
2.
Set-up of the Problem
The middle-surface of the three dimensional plate is the interior, , of a simply connected curve, , in the x1 x2 plane. The plate, which is to have a thickness 2, will be designated by . The so called reference plate will have a thickness of L which will be designated by . Thus, = (, ) and = (0, L). The lateral boundary of is = [, ]. The lateral boundary of is 0 = [0, L]. The boundary on the top 0 and bottom of is = {} and = {}, respectively. For the top and + bottom of , there is + = {L} and = {0}. It is assumed that the constant > 0 and is much smaller than the dimensions of . The plate is assumed to be clamped so that a displacement vector u vanishes on the lateral boundary. The notation for subscripts
+
x 3
x=L = x (0, L)

x 2 +
x=0
x =
, = x ( ) x =
Plate Models in Different Domains
Figure 1. The plates and . is as follows: Latin indices take on values of {1,2,3} while Greek indices take on values of
An Asymptotic Model of a Nonlinear Kelvin-Voigt Viscoelastic Plate
237
{1,2}. For repeated indices, the summation convention will be employed. A system of inertia associated with the coordinate system Ox x x is chosen so that 1 2 3 the axis Ox passes through the centroid of each cross section despite the thickness of 2. 3 Also: x d = x x d = 0, = {1, 2} 1 2
Differential operators will be either represented by xi or simply by i . The unit normal n = (n1 , n2, 0) on the lateral boundary, while n = (0, 0, 1) on the top and n = (0, 0, 1) on the bottom. The plate is composed of a nonlinear, viscoelastic, homogenous material where the Lam constants of isotropic materials, which are related to Youngs modulus and Poissons e ratio [2, pp. 59, 60], are replaced in the anisotropic case by the fourth-order tensors c ijkm and d ijkm used in the denition of the stress tensor as seen below. Volume forces, denoted by f = (fi ) : R3 , act in the interior of and surface forces, denoted by g = (gi ) : R3 , act on the boundaries = . + 0 The following equations of equilibrium for the nonlinear viscoelastic clamped plate as given in C OWIN and H EGEDUS [5, 6], are used to nd a u = (u ) : R3 such that: i j ij (u ) = (0 + M (u ))fi , on ij (u )n = gi , on j
where = ; is the lateral boundary of the plate; is the boundary + + 0 0 of the plate on the top and bottom; = density of the material; 0 = the reference change in the volume fraction of material; M = mollication operator from C 0 ( ) into C (R3); P 1 = truncation operator of class C ; fi = body forces; gi = surface forces; e (u ) = ij 1 ( u + j u ) i 2 i j
is the strain tensor; the stress tensor for a Kelvin-Voigt material has the form:
ij (u ) = ij (u ) = c ( )e (u) + d ( )e (u) ijkm ij ijkm
and the clamping conditions of u = 0 apply on the lateral boundary . This is a Kelvin0 Voigt model for homogeneous materials. It is assumed, as in a paper by F IGUEIREDO and T RABUCHO [8, pp. 3, 4], that the denition of the stress tensor contains the modied elasticity coefcient c ( ) and the ijkl modied viscosity coefcient dijkl ( ) which are dependent upon , the change in the volume fraction of the elastic material. aijkl is a elasticity coefcient and bijkl is a viscosity coefcient. c = (0 + M P ((u))) aijkl , d = (0 + M P ((u))) bijkl ijkl ijkl 0 is the original change in the volume fraction of elastic material. M is a mollier. P is a truncation operator. Due to symmetry about the Ox3 axis, c ( ) = c ( ) = c ( ), c3( ) = c333( ) = 0 ijkl jikl klij
238
Robert P. Gilbert and Robert J. Ronkese d ( ) = d ( ) = d ( ), d3( ) = d333( ) = 0 ijkl jikl klij
It is also assumed that the coefcients cijkl are positive denite for all symmetric tensors ij , and so there is an elliptic condition that is satised: c > 0 : c ( )ij kl c ij kl ij R3x3, ij = ji ijkl Hence, we obtain the following proposed remodeling rate equation:
kl kl = a ()+A ()e (u )+A ( )e (u)+c ()e (u )e (u )+d ( )e (u)e (u ) kl kl kl ijkl kl ij ijkl ij
a ( ) is a constitutive function. A ( ) is a rate remodeling coefcient. kl c ( )e (u )e (u ) is a nonlinear strain term. A ( )e (u ) is a linear time derivative kl ij ijkl kl kl ( )e (u )e (u ) is a nonlinear time derivative of strain term. The kl of strain term, and dijkl ij modied elasticity coefcients cijkl may also be considered as rate remodeling coefcients. The same could be construed for the modied viscosity coefcient dijkl . FIGUEIREDO and T RABUCHO justify this by mentioning that the driving mechanism associated with the remodeling rate equation is the strain energy per unit volume ij e (u ) as some evidence ij suggests [8, pp. 4, 5]. For any function g C 0 [0, T ]C 0( ) whose extension to R3 is g(t), the mollication operator is dened as follows: M (g(t))(x) = g (t)(x) =
R3
(x y)(t)(y)dy g
(x ) is a mollier derived from the function : (x ) = (

R3
(y)dy)1
x 1 ( ); n
(x) = exp(1/( x
2 R3
1)) if x
R3
<1
= 0, otherwise For d = the change in volume fraction and real numbers 0 < << 1, 0 < << 1, the truncation operator P of class C 1 is dened as: 0 (x ) + 2 if d (x ) 0 (x ) + 2 P (d )(x ) = d (x ) if 0(x ) d (x ) 1 0 (x ) 1 0 (x ) if d (x ) 1 0 (x ) with 0 < (0 + P (d))(x ) 1 2 There is a variational formulation of this problem. One seeks to nd a solution u = (u ) such that i
u V ( ) = {v = (vi ) [H 1( )]3 : v = 0 }, + ij (u )e (v )dx = ij fi vi dx + gi vi ds +
h vi d i
for all v V ( ). It is assumed that fi L2 ( ) and gi L2 ( ) which makes the problem wellposed.The left side of the above equation is symmetric and continuous. The right side is continuous as well.
239
3.
Scalings of Unknowns
For the scaling of various elements in the equations, the basic relationship of functions in the reference domain, , to the those in the other, , comes from CIARLET [3, pp. 32, 33]: x = (x ) = (x1 , x2, x3) i
= and 3 = 1 3 v (x ) = v (x) and v3 (x ) = 1 v3 (x) x f (x ) = f (x) and f3 (x ) = f3 (x) x g(x ) = g(x) and g3 (x ) = 2 g3(x) x 0
h (x ) = h (x) and h (x ) = h3 (x) x + 3

(x ) = (u()(x)), 3(x ) = (u()(x)), 33 (x )(x) = 2 33(u()(x))
e (u()) = e (u())(x), e (u ) = 1 e3 (u())(x), e (u()) = 33
1 e (u())(x) 2 33
where and represent the top and the bottom of the plate respectively. + First, the aforementioned scalings convert the terms in the proposed rate remodeling equation to their counterparts in the domain which are then substituted back into the equation: () = a ( ) + A e () + ce ()e () + 2 [A3 e3 () + A3 e3 ()] 1 + 2 [A33e33() + A33 e33()] + 4c33 e3 ()e3 () + 4d33 e3 ()e3() + c33e33 ()e () + d33e33 ()e ()] + 1 [c3333e33()e33() + d3333e33()e33()] 4 Next, a Kelvin-Voigt model for the plate will be derived. A general approach will be attempted with fourth-order tensors and time derivatives in the denition of the stress tensor. Using as the volume of integration,
ij e dx = ij
[cijkle e + dijkl e e ]dx kl ij kl ij
[ce e + de e ]dx [c33e e + d33e e ]dx 33 33 [c33 e e + d33 e e ]dx 3 3 3 3
+4
240
Robert P. Gilbert and Robert J. Ronkese +
[c33 e e + d33 e e ]dx 33 33 [c3333e e + d3333e e ]dx 33 33 33 33
Now, reverting back to the reference domain as the volume of integration: ij eij dx =
[ce e + d e e ]dx [c33e33e + d33e33 e ]dx
+ + +
1 2 4 2 1 2 1 4
[c33 e3 e3 + d33 e3 e3]dx
[c33 e e33 + d33 e e33]dx
[c3333e33e33 + d3333e33e33 ]dx
Asymptotic expansions of the following: u() = u0 + 1 u1 + 2 u2 + ... eij (u()) = e0 + e1 + e2 + ... ij ij ij eij (u()) = e0 + e1 + e2 + ... ij ij ij
2 1 2 1 () = 2 + 1 + + 1 + ... 2 1 2 1 3 () = 2 3 + 1 3 + 3 + 1 3 + ... 4 3 2 1 33 () = 4 33 + 3 33 + 2 33 + 1 33 + ...
can be inserted into the force balance equation: ij ()eij (v)dx =

fi vi dx +
0
gi vi dx +
+
hi vi dx
(1)
where upon the comparison of this equation with the above equation of fourth-order tensors and negative powers of yields: 4 3 33 33 33 = c3333e0 + d3333e0 and 33 = c3333e1 + d3333e1 33 33
p 33 = c33 ep+2 + d33 ep+2 + c3333ep+4 + d3333ep+4 , p 2 33 33 p 3 = 2c33 ep+2 + 2d33 ep+2 , p 2 3 3

2 1 = c33e0 + d33e0 and = c33e1 + d33e1 33 33 33 33 p = c33ep+2 + d33ep+2 + cep + dep , p 0 33 33
241
Setting the non-zero powers of equal to zero, one gets the following equations: 4 :
4 33 e33(v)dx = 0, 3 : p ij eij (v)dx = 0 3 33 e33(v)dx = 0
For p = {2, 1, 1, 2, 3, ...}, 0 :
0 hi vi dx
+
0 ij eij (v)dx =
fi vi dx +
gi vi dx +
One is now in a position to characterize many of the quantities in the above equations leading to signicant simplications in the asymptotic expansions. Lemma 1. Given (t) = 1 dijkl ekl ekl dx with dijkl > 0, 2 if t (t) 0 and ekl (0) = 0, then ekl (t) = 0. Proof. 1 dijkl ekl ekl dx = (t) 0 (t) K 2 t t However, eij (0) = 0 (0) = 0 K = 0 and (t) 0; but assuming dijkl > 0, (t) 0. Combining the two inequalities concerning leads to: (t) = 0 eij (t) = 0 Theorem 1. If
4 4 2 33 e33(v)dx = 0, then e0 = 0, 33 = 0, = 0 33
Proof. Choose v = u0.

4 33 e33 (v)dx = 0 = 4 33 e0 dx 33
[c3333e0 e0 + d3333e0 e0 ]dx 33 33 33 33 [c3333e0 e0 + d3333 33 33 1 0 0 (e e )]dx 2 t 33 33
Using Lemma 1, Theorem 2. If
e0 33
4 2 = 0 e0 = 0 33 = 0, = 0 33
3 3 1 33 e33(v)dx = 0, then e1 = 0, 33 = 0, = 0 33

3 33 e33 (v)dx = 0 = 3 33 e1 dx 33
[c3333e1 e1 + d3333e1 e1 ]dx 33 33 33 33 [c3333e1 e1 + d3333 33 33 1 1 1 (e e )]dx 2 t 33 33
Using Lemma 1,
e1 33
3 1 = 0 e1 = 0 33 = 0, = 0 33
242

2 ij eij (v)dx = 0, then 2 33 e33(v)dx = 0
2 Theorem 3. If = 0, e0 = 0 and 33
2 e3 = 0, 3 = 0,

2 ij eij (v)dx = 0 = 2 ij e0 dx ij
2 2 2 [ e0 + 3 e0 + 33 e0 ]dx = 0 3 33 2 3 e0 dx = 0 3
2 From Theorem 1, = 0 = e0 33
[c33 e0 e0 + d33 e0 e0 ]dx = 0 3 3 3 3 1 0 0 (e e )]dx = 0 2 t 3 3
[c33 e0 e0 + d33 3 3
Using Lemma 1,
e0 3
2 = 0 e0 = 0 3 = 0 3 2 ij eij (v)dx = 0 2 33 e33(v)dx = 0
1 Theorem 4. If = 0, e1 = 0 and 33
1 ij e1 dx = 0, then ij 1 33 e33(v)dx = 0
1 e1 = 0, 3 = 0, 3
1 Proof. Choose v = u1 so that ij e1 dx = 0. ij 1 1 From Theorem 2, = 0 = e1 3 e1 dx = 0 33 3
[c33 e1 e1 + d33 e1 e1 ]dx = 0 3 3 3 3 c33 e1 e1 + d33 3 3 1 1 1 (e e )]dx = 0 2 t 3 3
Using Lemma 1,
e1 3
1 = 0 e1 = 0 3 = 0 3 1 ij eij (v)dx = 0 1 33 e33(v)dx = 0
2 1 To nd out more about 33 and 33 , a technique used by C IARLET and D ESTUYNDER , [4, p. 330] in the case for the linear elastic plate is employed here and can work in general.

n n n Theorem 5. For n = {2, 1}, if = 0 = 3 , then 33 = 0
243
Proof. For n = {2, 1},

n ij eij dx = 0 = n n n n ( e + 3 e3 + 33e33)dx
From Theorem 1 and Theorem 2,
=0
n n (33e33 + 3 e3 )dx = 0
Using Greens formula, 0=

n (33
v3 n v3 + 3 )dx = x3 x +
n n 33 3 + )v3 dx x3 x n 33v3 d + n 3 v3 d n
n n Since (33e33 + 3 e3 )dx = 0, one looks for a solution where x33 + x3 = 0 in 3 n and 33 = 0 on + . From Theorem 3 and Theorem 4 for n = {2, 1}, 3 = 0,
so
n 3 x
=0 0=
n 33 x3
n = 0. Thus, 33 is a constant with respect to x3 . n 33v3 dx = n 33v3 dx n 33v3 dx v3 (+ )
n So 33 = 0 on + . However,
n 33 x3
n = 0, so 33 = 0 in .
Corollary 1. For the viscoelastic case, all negative powers of ij are zero.
n n n Corollary 2. For any positive odd integer n, if 3 = 0 = , then 33 = 0 n n Proof. For any positive odd integer n, ij eij (v)dx = 0. Given that = 0, this n n simplies to (33e33 + 3 e3 )dx = 0. Now, Greens formula can be applied to this equation just as it was above in the cases of n = {2, 1}. 2 Theorem 6. Since 33 = 0, there exists a u2 such that
e2 33
c33 0 t 0 [( d3333 e
d33 0 s c3333 d3333 e )exp( 0 d3333 dq)]ds t c exp( 0 d3333 ds) 3333
2 2 Proof. From Theorem 5, 33 = 0 ij = 0 and there would exist a u2 such that 2 33 = c33 e0 + d33 e0 + c3333e2 + d3333e2 = 0 33 33
e2 + 33 e2 = 33
c33 0 d33 0 c3333 2 e = e e d3333 33 d3333 d3333

c33 0 t 0 [( d3333 e
244
1 Theorem 7. Since 33 = 0, there exists a u3 such that c33 1 t 0 [( d3333 e
e3 33
1 1 Proof. From Theorem 5, 33 = 0 ij = 0 and there exists a u3 such that 1 33 = c33 e1 + d33 e1 + c3333e3 + d3333e3 = 0 33 33
e3 + 33 e3 33 =
c33 1 d33 1 c3333 3 e = e e d3333 33 d3333 d3333

c33 1 t 0 [( d3333 e
Now, a simplication of the asymptotic expansions of both the displacement vector u and the stress tensor will be derived. Theorem 8. e1 = 0, u1 = 0 Proof. From Theorem 2, e1 = 0 and from Theorem 4, e1 = 0. Insert this into 33 3 1 1 1 1 ij eij dx = 0. e dx = 0 (ce1 + d e1 )e1 dx = 0
[ce1 e1 + d
1 1 1 (e e )]dx = 0 2 t
Using Lemma 1,
e1
= 0 e1 = 0 e1 = 0 u1 = 0 ij
1 1 Theorem 9. e3 = 0, = 0, e3 = 0, 3 = 0 33 3 1 Proof. From Theorem 5, 33 = 0 and from Theorem 8, e1 = 0. Together, both lead to 1 e3 = 0. Combining this with e1 = 0 leads to = 0. Now, choose v = u3 , and insert 33 1 1 it into the equation ij eij (v)dx = 0. Since = 0 = e3 , this equation becomes: 33
[c33e3 + d33e3 ]e3 dx = 0 3 3 3 1 3 3 (e e )]dx = 0 2 t 3 3
[c33e3 e3 + d33 3 3
Using Lemma 1,
e3 3
1 = e3 = 0 e3 = 0 3 = 0 3 3
A characterization of en+4 in terms of en+2 for any odd integer n 1 suggests further 33 steps in evaluating positive odd powers of the asymptotic expansions of the displacement vector and the stress tensor.
245
Theorem 10. The positive odd powers in the asymptotic expansions of both the displacement vector u and the stress tensor are all zero.
1 1 Proof. From Theorem 9, = 0 = 3 and 1 ij eij (v)dx = 0 1 33e33 (v)dx = 0
n Due to Corollary 2 to Theorem 5, one can set 33 = 0 for n = 1. However, keep the notation with the exponent n. n 33 33 = c33 en+2 + d33 en+2 + c3333en+4 + d3333en+4 = 0 33
Now, take the Laplace transforms of all terms.

c33 L{en+2 } + d33[sL{en+2 } en+2 (0)] + c3333L{en+4 } + d3333[sL{en+4 } en+4(0)] = 0 33 33 33
Since en+2 (0) = 0 = en+4 (0) 33 L{en+4 } = 33 [c33 + d33 s] L{en+2 } [c3333 + d3333s] d33 [s + c33 /d33 ] L{en+2 } = d3333[s + c3333/d3333] c33 d33 s 1 L{en+2 } L{en+2 } = d3333 s + c3333/d3333 d3333 s + c3333/d3333 d33 d33 c3333 c33 1 = = L{en+2 } + ( ) L{en+2 } 2 d3333 (d3333) d3333 s + c3333/d3333 d33 n+2 d33 c3333 c33 e (t) + ( ) d3333 (d3333)2 d3333 e1
t 0
en+4 (t) = 33
en+2 ( )exp{
c3333 (t )}d d3333
Thus, when n = 1, Due to Corollary 2 to Theorem 1 ij = 0 and that: e5 (t) = 33
= 0 leads to e3 = 0. Now, consider the case of n = 1. 33 1 1 1 5, 3 = 0 = leads to 33 = 0. This would mean that
t 0
d33 3 d33c3333 c33 e (t) + ( ) d3333 (d3333)2 d3333
e3 ( )exp{
c3333 (t )}d d3333
This can be put to good use in the following way. Since e3 = 0 = e3 , 33 3

3 ij e3 dx = 0 ij 3 e3 dx = 0
[ce3 + de3 + c33e5 + d33e5 ]e3 dx = 0 33 33
Substitute for e5 : 33 0 = {ce3 + de3 + c33[ d33 e3 (t) + ( 3333

d d33 c3333 (d3333 )2
c33 d3333
t 3 c3333 0 e ( )exp{ d3333 (t
)}d ]
246
d
Robert P. Gilbert and Robert J. Ronkese + d33[ d33 e3 (t) ( 3333

c3333 d33 c3333 d3333 ( (d3333 )2 d33 c3333 c3333 c d33 )e3 exp{ d3333 (0)} (d3333 )2 3333 c33 t 3 c3333 3 d3333 ) 0 e ( )exp{ d3333 (t )}d ]}e dx
0=
1 3 3 3 3 {c e e + d 2 t (e e ) d33 3 3 d33 c3333 c t c + c33[ d3333 e e ( (d3333 )2 d33 ) 0 e3 ( )e3 ( )exp{ d3333 (t 3333 3333 d d c c + d33[ d33 1 t (e3 e3 ) + ( 33 3333 d33 )(e3 e3 ) (d3333 )2 3333 2 3333 c33 t 3 c3333 d33 c3333 3 ( )exp{ c3333 (t )}d ]}dx d3333 ( (d3333 )2 d3333 ) 0 e ( )e d3333
)}d ]
3 33 e3 = e3 = 0 e3 = 0 e5 = 0 e5 = 0 u3 = 0, = 0 33
Substitute v = u5 into
3 ij eij (v)dx = 0 to get:
3 3 e5 dx = 0 = 4 3
[c33e5 + d33e5 ]e5 dx 3 3 3
[c33e5 e5 + d33 3 3
1 5 5 (e e )]dx = 0 2 t 3 3
Using Lemma 1,
e5 3
3 = 0 e5 = 0 3 = 0 3
3 3 3 3 Since = 0 = 3 , ij eij (v)dx = 0 becomes 33e33(v)dx = 0. Due to 3 3 Corollary 2 to Theorem 5, one can set 33 = 0 which leads to ij = 0 and to the same situation as before when n = 1 was inserted into:
en+4 (t) = 33 to produce:
d33 n+2 d33 c3333 c33 e (t) + ( ) d3333 (d3333)2 d3333
t 0
en+2 ( )exp{
c3333 (t )}d d3333
e5 (t) = 33
d33 3 d33c3333 c33 e (t) + ( ) d3333 (d3333)2 d3333
t 0
e3 ( )exp{
c3333 (t )}d d3333
Now, set n = 3 and one obtains: e7 (t) = 33 d33 5 d33c3333 c33 e (t) + ( ) d3333 (d3333)2 d3333
t
e5 ( )exp{
0
c3333 (t )}d d3333
The same pattern of calculations starts anew. Therefore, by induction on the odd integer n 1, the positive odd powers of the asymptotic expansions of the displacement vector u and the stress tensor are equal to zero. The constitutive equation for the general Kelvin-Voigt model on the domain is:
ij = c (0 )e + d (0 )e kl ijkl kl ijkl Next, the coefcients Dijkl (0 ) of the inverse matrix dened by the viscoelastic coefcients d (0) are used to rearrange the constitutive equation into a rst order differential ijkl equation: de kl + Dijkl (0)c (0 )e = Dijkl (0 )ij ijkl kl dt
247
One solves for ekl in the above system of differential equations, noting that e (0) = 0, kl and obtains: t s 0 Dijkl ij exp( t Dijkl cijkl d )ds ekl = t exp( 0 Dijkl c d ) ijkl Inside the above equation, the following can be dened: Tklij (t, s) =
t Dijkl (s)exp( s t Dijkl cijkl d ) t 0 Dijkl cijkl d )
exp(
from which e (t) = 0 Tijkl (t, s)ij (s) kl For any tensor kl , this last equation can be put into the following form: t
[
0
Tklnq nq ds]kldx
e (u )kldx = 0 ij
[
0
T ds + 0 t
T3333ds] dx
[4
0 t
T333 ds]3 dx t T333333ds]33dx 0
[
0
T33 ds +
e kl dx = 0 ij
Now, convert from to making the appropriate changes in variables and scaling:
t t
[
0
T ds] dx + 2
t
[
0
T3333 ds] dx
t
+ 42
[
0 t
T333 ds]3 dx + 2
[
0
T33 ds]33dx
+ 4
[
0
T333333ds]33dx
eij kl dx = 0
This equation is coupled with: ij eij (v)dx =

(0 + P (0 ))fivi dx +
0
gi vi dx +
+
hi vi dx
for all test functions vi , and the two equations characterize the mixed displacement-stress approach. Theorem 11. The rst term u0 = (u0, u0, u0) in the asymptotic expansion 1 2 3 u( ) = u0 +
1 1 2 2 0 3 u + ... satises: u0 = u0 (x1 , x2, t), u0 = (x1, x2, t) x3 x 3 3 2 u0 2 u0 tx ) u0
u +
1 0 Additionally, = 1 T( tx + 2
248
Proof. One takes the 0 terms in the rst of the two equations of the mixed displacementstress approach and creates the following equations:
t
[
0
0 T ds] dx
e0 dx = 0
e0 3 dx = 0, 3
e0 33dx = 0 33
t 0 = 0 T ds and e0 = 0 33 and e0 = 0 3 From which 33 3 u0 0 0 0 3 Now e33 = 0 x3 = 0 u3 = u3 (x1, x2, t) u0 u0 u0 0 3 Furthermore, e0 = 0 x = x3 u0 (x1, x2, t) = (x1, x2, t) x3 x 3 3 t 0 Finally, using the fundamental theorem of calculus, e0 = 0 T ds becomes
e0
0 T =
2 u0 0 1 2 u0 + ) e = ( t 2 tx tx
0 =
2u0 1 1 2 u0 T( + ) 2 tx tx
Theorem 12. The second term u1 = (u1, u1, u1) in the asymptotic expansion 1 2 3 u( ) = u0 +
1 1
u +
2 2
1 3 u + ... satises: u1 = u1 (x1, x2, t), u1 = (x1, x2, t) x3 x 3 3 2 u1 2 u1 tx )
u0
1 1 Additionally, = 1 T( tx + 2
Proof. One takes the previous proof and substitutes the superscript 1 for the superscript 0 wherever the superscript 0 is to be found. The theorem for terms with the superscript 1 results. Considering that the odd terms in the asymptotic expansion for the displacement are zero, this is a moot point. Theorem 13.
0 3 =
1 1 2 1 1 1 0 T33 e3 , = T33 e2 , T e0 = T33 e2 33 4 t t t t 33
Proof. The 2 equation leads to three different equations according to the subscripts involved:
t t 2 T ds + 0 t 0 0 T3333ds] dx
e2 dx = 0
[
0 t
0 T333 ds] dx 0 T33 ds]33dx 0
e2 3 dx = 0 3 e2 33 dx = 0 33
An Asymptotic Model of a Nonlinear Kelvin-Voigt Viscoelastic Plate The 4 equation adds a last equation to those above:
t t 2 T33 ds]33dx + 0
249
[
0
2 T333333ds]33dx
e4 33 dx = 0 33
One should recall that the stress tensor has its own asymptotic expansion of: ( ) = 0 +
2 2
4 4
+ ....
which with the above equations leads to a further characterization of some terms in the expansion by way of the fundamental theorem of calculus: 1 1 1 0 0 3 = T33 e2 , = T33 e2 3 4 t t 33
0 Combining with an earlier expression for , one has: 0 1 = T
0 1 e = T33 e2 t t 33
Assuming that e0 = 0, one sees that e2 = 0 while e0 = 0 = e1 33 33 33

0 Other expressions with can be obtained from the force balance equation as follows:
Lemma 2. For v C 1 () and v3 C 1 (),

0 v dx = 0 f v dx + 0 g v d + 0 h v d
0=
0 f3 v3 dx +
0 g3v3 d +
0 h3v3 d
Proof. Let = (0 + P (0 )). From the 0 equation, one can derive the following: Substitute 3 for 3 in 3 e0 dx = 0 to get 3 e0 dx = 0. 3 3 Substitute for in e0 dx = 0 to get e0 dx = 0. Now, two variants of the force balance equation (1) on page 6 are:
0 0 ( v + 3 3 v )dx = 0 f v dx + 0 g v d + 0 h v d
0 0 (333v3 + 3 v3 )dx =
0 f3 v3 dx +
0 g3 v3 d +
0 h3 v3 d
The two equations can be added together: 0 ( v + 3 e0 + 33e0 )dx = 33 3 0 0 0 0 0 0 (f v + f3 v3 )dx + 0 (g v + g3 v3 )d + + (h v + h3 v3 )d The second and third terms of the left side are zero due to the beginning of this proof. The 0 0 v terms on the right side can be grouped with v dx to form one equation. The 0 v3 terms on the right side can be grouped with zero to form another.
250
Theorem 14. For v C 1 () and C 1 (),
d =
f d +
g ds
with boundary conditions given as:

0 [ v ]+ = 0 h v d
Proof. From the above lemma, the rst of the resulting equations is:
0 v dx = 0 f v dx + 0 g v d + 0 h v d
Integrate by parts on the left side noting that v = 0 on 0 to get 0 0 [ v ]+ v dx. Decompose each integral in the middle three terms as follows:
L L L
0 d]v dx3 =
[
0
0 f d]v dx3 +
[
0
0 g ds]v dx3
with boundary conditions given as:

0 [ v ]+ = 0 h v d
The quantities in the square brackets of the equation with the decomposed integrals can form an equation which is stated as follows:
d =
f d +
g ds
This is the strong formulation of the equation at the start of this proof, which is the rst of the two equations derived from the force balance equation (1) in Lemma 2. It is also a statement of this theorem. Theorem 15. For v C 1 (), f3 C 1 (), g3 C 1 (), h3 C 1 (), 0=
x 3 f3d +
x 3 g3ds
with boundary conditions given as: 0=

+ 0 x (3 h3 )v d
Proof. From the above lemma, the second of the resulting equations is: 0=
0 f3 v3 dx + 0 g3v3 d + 0 h3v3 d

v0
251
0 0 0 From Theorem 3, e0 = 0 v3 = 3 (x , x3, t) x x . Choose 3 (x, x3, t) = 0 3 3 0 so that v3 = x x . 3 v 0
0=
0 x f3 3 v dx +
0 x g3 3v ds +
0 x h3 3 v ds
Integrate by parts noting that v = 0 on 0 . 0=

0 x (3 f3 )v dx + 0 x (3g3 )v d + 0 x (3h3 )v d
Decompose each integral as follows:

L L
0=
0
0 x 3 f3 d]v dx3 + 0
0 x 3 g3ds]v dx3
with boundary conditions given as: 0=

+ 0 x (3 h3 )v d
The quantities in the square brackets of the equation with the decomposed integrals can form an equation which is stated as follows: 0=
x 3 f3d +
x 3 g3ds
This is the strong formulation of the equation at the start of this proof, which is the second of the two equations derived from the force balance (1) in Lemma 2. It is also a statement of this theorem. Next, one considers the time rate of change in the volume fraction and its asymptotic expansion. Theorem 16. The asymptotic expansion for the time rate of change in the volume fraction is given by: = 0 + 1 1 + 2 2 + ... 0 =c 2 e2 +d 2 e2 +A e2 + A e2 +2c 0 2 33 where 0 2 3333e33 33 3333 e33 33 33 33 33 e e33 +2d33e e33 + 33 A e0 + A e0 + ce0 e0 + d e0 e0 + a with 0(0) = 0 and 1 = 2A3 e2 + 2A3 e2 3 3 and 2 = 2c3333e2 e4 + 2d3333e2 e4 + A33e4 + A33 e4 + 2c33(e0 e4 + 33 33 33 33 33 33 33 2e2 e2 ) + 2d33(e0 e4 + 2e2 e2 ) + 8c33e2 e2 + 8d33e2 e2 + A e2 + 33 33 3 3 33 3 3 e2 + c (e0 e2 + +e2 e0 ) + d (e0 e2 + e2 e0 ) A Proof. In the domain , the time rate of change in volume fraction is given by: = e + c e e + d e e a + Akl ekl + Akl kl ijkl ij kl ijkl ij kl + A e + 2A e + A e + A e + 2A e + A e + c =a 33 33 33 33 3 3 3 3 e e +
252
2c e e + 4c e e + c e e + de e + 2d e e + 3333 33 33 33 33 33 3 3 33 33 3 3 33 33 4d e e + d e e 3333 33 The scalings: e (u(x)) = e (u(x )), e3 (u(x)) = e (u(x )), e33(u(x)) = 2e (u(x )) 33 3 convert the expression for into the expression for on the domain , which is the rst expression mentioned in the above theorem. Now, the expression for in the domain can be multiplied by 4 and asymptotic expansions in the form: eij = e0 + 2e2 + 4 e4 + ... can be substituted for the various ij ij ij strain tensors. From Theorems 1,3,10, one may note that 0 = e0 = e0 = e2n+1 for integer n 0. 33 ij 3 4 0 + 5 1 + 6 2 + ... = c3333(e0 + 2 e2 + 4 e4 + ...)(e0 + 2 e2 + 4 e4 + ...) + d3333(e0 + 2 e2 + 4 e4 + 33 33 33 33 33 33 33 33 33 33 33 ...)(e0 + 2 e2 + 4 e4 + ...) 33 + 2 [A33 (e0 + 2 e2 + 4 e4 + ...) + A33(e0 + 2 e2 + 4 e4 + ...) + 2c33(e0 + 33 33 33 33 33 33 2 2 4 4 0 2 2 4 4 0 e + e + ...)(e33 + e33 + e33 + ...) + 4c33(e3 + 2 e2 + 4 e4 + ...)(e0 + 3 3 3 2 e2 + 4 e4 + ...) + 2d33(e0 + 2 e2 + 4 e4 + ...)(e0 + 2 e2 + 4 e4 + ...) + 33 33 33 3 3 3 3 3 3 3 3 4d33(e0 + 2 e2 + 4 e4 + ...)(e0 + 2 e2 + 4 e4 + ...)] + 3 [2A3 (e0 + 2 e2 + 4 e4 + ...) + 2A3 (e0 + 2 e2 + 4 e4 + ...)] 3 3 3 3 3 3 + 4 [a + A (e0 + 2 e2 + 4 e4 + ...) + A (e0 + 2 e2 + 4 e4 + ...) + c(e0 + 2 e2 + 4 e4 + ...)(e0 + 2 e2 + 4 e4 + ...) + d 0 + 2 e2 + 4e4 + ...)(e0 + (e 2 e2 + 4 e4 + ...)] The terms with the lowest power of are those with 4 , and these will be desig nated as 0 . In likewise fashion, those with 5 as a coefcient will be 1, and those with 6 2 . The asymptotic expansion for the time rate of change in the as coefcent will be volume fraction is given by: = 0 + 1 1 + 2 2 + ... 0 , 1 and 2 as given in the statement of this theorem. with
References
[1] Buchanan, J.L., Gilbert, R. P. and Khashanah, K.; Determination of Parameters of Cancellous Bone Using Low-Frequency Acoustic Measurements, Journal of Computational Acoustics , Vol. 12, No. 2, pp. 99-126, 2004. [2] Chou, P.C. and Pagano, N.J.; Elasticity-Tensor, Dyadic and Engineering Approaches , Dover, New York, 1992. [3] Ciarlet, P.G.; Mathematical Elasticity, Volume II: Theory of Plates, North-Holland, Amsterdam, 1997.
253
[4] Ciarlet, P.G. and Destuynder, P.; A justication of the two-dimensional linear plate mdifferential equationl , Journal de M` canique, vol. 18, no.2, 1979. e [5] Cowin, S.C. and Hegedus, D.H., Bone remodeling I: theory of adaptive elasticity, Journal of Elasticity , vol.6, no.3, pp. 313-326, 1976. [6] Cowin, S.C. and Hegedus, D.H., Bone remodeling II: small strain adaptive elasticity, Journal of Elasticity , vol.6, no.4, pp. 337-352, 1976. [7] Cowin, S.C. and Nachlinger, R.R., Bone remodeling III: uniqueness and stability in adaptive elasticity theory, Journal of Elasticity , vol.8, no.3, pp. 285-295, 1978. [8] Figueiredo, I. and Trabucho, L.;Asymptotic Model of a Nonlinear Adaptive Elastic Rod, Mathematics and Mechanics of Solids , vol. 9, no. 4, 2004. [9] Monnier, J. and Trabucho, L.; An Existence and Uniqueness Result in Bone Remodeling Theory, Computer Methods in Applied Mechanics and Engineering , vol. 151, pp. 539-544, 1998. [10] Trabucho, L.; Non-linear Bone Remodelling: An Existence and Uniqueness Result, Mathematical Methods in the Applied Sciences , vol. 23, pp. 1331-1346, 2000. [11] Trabucho, L. and Via o, J.M.;Mathematical Modeling of Rods, found in the Handn book of Numerical Analysis, Ciarlet, P.G. and Lions, J.L.(editors); Volume IV, NorthHolland, Amsterdam, 1996.
M ATHEMATICAL A NALYSIS OF A B ILATERAL O BSTACLE P ROBLEM FOR A C LASS OF S ECOND -O RDER O PERATORS
Laurent L vi and Guy Vallet e Universit de Pau et des Pays de lAdour e Laboratoire de Math matiques Appliqu es - UMR 5142 CNRS e e BP 1155 - 64013 PAU Cedex - FRANCE
Abstract We investigate some inner bilateral obstacle problems for a class of strongly degenerate parabolic-hyperbolic quasilinear operators associated with homogeneous Dirichlet data in a multidimensional bounded domain. We rst introduce the concept of an entropy process solution, more convenient and generalizing the notion of an entropy solution. Moreover, the boundary conditions are expressed by using the background of Divergence Measure Fields. We ensure that proposed denition warrants uniqueness. The existence of an entropy process solution is obtained through the vanishing viscosity and penalization methods.
1.
1.1.
Introduction
Mathematical Framework
Obstacle problems for conservation laws in physics and mechanics have been studied by many authors [6, 12, 16]. In this work, we are especially interested in the second-order quasilinear operator stemming from the theory of uid ows through porous media: P(t, x, .) : u t u + xi i (t, x, u) + (t, x, u) (u),
i=1 p
where is a nondecreasing function; especially may vanish on nonempty intervals of R. Let us given two thresholds a and b that are xed reals such that a 0 b. For any positive T , the bilateral obstacle problem for P on the bounded domain of R p , p 1, may be formally described: nd a measurable function u on ]0, T [ satisfying the formulation of free boundary [a < u < b], a u b in ]0, T [, P(t, x, u) = 0 on the (unknown) zone [a < u < b], (1) P(t, x, u) 0 on the (unknown) zone [a < u = b], P(t, x, u) 0 on the (unknown) zone [a = u < b], u = 0 on ]0, T [, u(0, .) = u0 on ,
256
Laurent L vi and Guy Vallet e
where u0 is a measurable function on such that a u0 b a.e. in . On the one hand, we emphasize that without any assumption on the sign of the source term for P a bilateral obstacle condition on initial data does not a priori pass on to the associated solution. On the other hand, entropy formulations for the Dirichlet problem to strongly degenerate parabolic-hyperbolic operators have been introduced by J.Carrillo in 1999 ([2]). Since then, numerous works have been achieved on this matter [1, 13, 14, 17, 18]. Here, we rst provide in Subsection 1.3. the denition of an entropy solution to (1). However, since we release the smoothness assumptions on the convective term and the reactive one it is more convenient to introduce the concept of an entropy process solution to (1). This notion, highlighted in [18] for the Dirichlet to P, may be considered as an extension to the second order of the framework due to R.Eymard, T.Gallou t & R.Herbin e in [7]. The uniqueness of an entropy process solution to (1) is stated in Section 2. and its existence is obtained in Section 3. via the vanishing viscosity method and by relaxing the bilateral obstacle condition. These two results provide the existence on an entropy solution to (1) and warrant the strong convergence in Lq (]0, T [), 1 q < +, of the sequence of approximate solutions.
1.2.
Main Notations and Assumptions on Data
The reaction term is a continuous function on [0, T ] [a, b]. In addition is Lipschitzian with respect to its third variable with a constant M , uniformly with respect to (t, x) in ]0, T [. The ux term is a W 1,+ -class vector-valued function on ]0, T []a, b[ such that for all i in {1, .., p}, xi i is Lipschitzian with respect to its third variable with a constant Mxi i , uniformly with respect to (t, x) in ]0, T [. Thus we can set M = max Mxi i and dene for any t of [0, T ],
i{1,..,p}
M(t) = max(a, b)eK1 t +
K2 K2 t e 1 , K1
(2)
where K1 = M + M and K2 = max |(t, x, 0) + Div(t, x, 0)|. The diffusive term is a nondecreasing W 1,+ (]a, b[)-class function such that (0) = 0. We note E = {l R, {l} = 1({(l)})}. Hence, if 1 denotes a generalized inverse of , 0 the hypo-inverse for example dened through r Im, 1 (r) = In f 1 ({r}), 0 then (E) is the set of points where 1 is continuous ([2]). 0 We assume that is a bounded subset of R p such that = is Lipschitz-deformable (see [1, 3, 4, 13]). Then the space of L2 -Divergence Measure elds on Q - denoted DM 2 (Q) - is given by
[0,T ]
DM 2 (Q) = V = (v0, v1 , .., v p) (L2(Q)) p+1, Div(t,x)V Mb (Q) ,

where Mb(Q) is the space of bounded Radon measures on Q. For any V in DM 2 (Q), it is useful to dene a linear mapping V on H 1(Q) L (Q) C (Q) through the following
Mathematical Analysis of a Bilateral Obstacle Problem for a Class ... generalized Gauss-Green formula coming from [4]: V () := V,
257
=
Q
V.(t , )dxdt +
Q
d[Div(t,x)V ],
(3)
where d[Div(t,x)V ] denotes the Borelian measure on Q associated with the bounded Radon measure Div(t,x)V . In addition the next property holds (see [13]): let V be an element of DM 2 (Q) such that v0 is continuous at t = 0 and t = T with respect to the L1 ()-norm then, for any in H 1 (Q) L (Q) C (Q),
0+
lim
V .(0, )dxdt
Q
v0 (T, x)(0, x)dx
v0 (0, x)(0, x)dx V, ,
(4)
for any boundary-layer sequence ( )>0 , i.e. a sequence of C 2() C ()-class functions such that lim 0+ = 1 pointwise in , 0 1 in , = 0 on . In (4), if belongs 1 to L2 (0, T ; H0 ()) the right-hand side is equal to 0. We will consider sgn the approximation of the function sgn given for any positive parameter and nonnegative real x by: sgn(x) = min x , 1 and sgn (x) = sgn (x).
To simplify the writing, we refer to the notations F(u, k) = sgn(u k){(t, x, u) (t, x, k)}, G(u, k) = sgn(u k)Divx (t, x, k) + (t, x, u), Uk = (|u k|, |(u) (k)| + F(u, k)), = (t , ). Eventually, for any n in N , H n stands for the n-dimensional Hausdorff measure and for all s of ]0, T ], Qs is the cylinder ]0, s[, s =]0, s[ with the convention Q = QT and = T . The outer normal of is denoted .
1.3.
Two Concepts of Weak Solutions
The existence of possible internal and boundary layers leads us to propose a mathematical formulation for (1) through an entropy inequality inside the studied eld - using the classical Kruzhkov entropy pairs - and on its boundary; the latter is viewed as an extension to the second order of the F.Ottos formulation provided in [15] for hyperbolic rst-order operators and uses - as in [1, 13, 18] - the mathematical framework of Divergence-Measure Fields. In addition, to take into account the bilateral obstacle condition we only consider the Kruzhkov pairs for a parameter k that has to belong to the bounded interval [a, b]. That is why, by referring to a preliminary study of the positiveness obstacle problem for P in [11], it will be said that:
258
Denition 1. A measurable function u on Q is an entropy solution to (1) if: a u b a.e. in Q,

1 (u) L2 (0, T ; H0 ()), t (u) L2 (0, T ; L2 ()), loc t u L2 (0, T ; H 1 ()),
(5) (6) (7)
k [a, b], Uk DM 2 (Q), k [a, b], D (] , T [), 0, Uk .dxdt G(u, k) dxdt +

Q
|u0 k|(0, .)dx 0,
(8)
L (Q) H 1 (Q) C (Q), 0, (T, .) = (0, .) = 0, k [a, b], F(k, 0).d H p Uk , + U0 , .
(9)
Remark 1. Considering k = a and in D (Q), 0, in (8) leads to: ((u a)t (u). + F(u, a). G(u, a))dxdt 0.
Q
We observe that a.e. in Q, F(u, a) = (t, x, u) (t, x, a). So thanks to some integrations by parts one gets: P(t, x, u), D (Q),D (Q) +
Q
(1 sgn(u a))(Div(t, x, a) + (t, x, u))dxdt 0.
Assume now that u is smooth enough so that P(t, x, u) belongs to L1 (Q) and let us choose = (u a)+ with in D (Q), 0. Since (1 sgn(u a))(Div(t, x, a) + (t, x, u))(u a)+ = 0 a.e. on Q, we obtain P(t, x, u)(u a)+dxdt 0.
Q
That means P(t, x, u)(u a)+ 0 a.e. on Q. The same reasoning with k = b leads to P(t, x, u)(u b) 0 a.e. in Q that formally provides (1). However, one of the feature of this work in comparison with [11] is to release the assumptions of regularity for and . So for the existence property stated in Section 3. through the vanishing viscosity and penalization methods, we can only refer to an L (Q)estimate of approximate solutions. That is why we lean on the original presentation of R.Eymard, T.Gallou t & R.Herbin in [7] for rst-order quasilinear hyperbolic operators e that consists - by using the basic tools of Young measure [5] - in introducing a free variable living on ]0, 1[ and a new measurable and bounded unknown ]0, 1[Q, (,t, x), that fullls an entropy formulation on ]0, 1[Q and called an entropy process solution. We adapt this concept when dealing with second order quasilinear operators associated with a forced bilateral constraint. From now, to simplify the writing, we set Q =]0, 1[Q and dq = ddxdt:
Mathematical Analysis of a Bilateral Obstacle Problem for a Class ...

1
259
Denition 2. Let be in L (]0, 1[Q) and u(t, x) =

0
(,t, x)d. We say that is an
entropy process solution to (1) if: a b a.e. in Q , for a.e. (t, x) in Q, ((, t, x)) = (u(t, x)) for a.e. in ]0, 1[, the smoothness properties (6) hold and: k [a, b], k DM 2(Q), k [a, b], D (] , T [), 0, k .dxdt
Q
(10) (11)
(12)
G(, k)dq +
Q
|u0 k|(0, .)dx 0.
(13)
L (Q) H 1 (Q) C (Q), (T, .) = (0, .) = 0, 0, k [a, b], F(k, 0).d H p k , + 0 , ,
(14)
where k =
0 1 1
|(, ., .) k|d,
0
(|()(, ., .) (k)| + F((,., .), k))d .
The concept of an entropy process solution generalize that of an entropy solution since an entropy process solution independent from the variable is an entropy solution. But the rst notion is more convenient that the second one, mainly because it requires a few estimates of approximate solutions and so an existence result is easier to obtain. Moreover, the existence and uniqueness of an entropy solution results from the existence and uniqueness of an entropy process solution. First by reasoning as F.Otto in [15], we may announce: Proposition 1. An entropy process solution to (1) fullls: ess lim
t0+ ]0,1[
|(,t, x) u0(x)|ddx = 0.
(15)
2.
The Uniqueness Theorem
The proof basically relies on a inner comparison property which is an extension to secondorder operators of the usual hyperbolic method of doubling variables due to S.N.Kuzhkov [9]; the contribution of diffusive terms being controlled thanks to an energy inequality in the same spirit as in the original paper of J.Carrillo [2]. To do so, we need some preliminary lemma. The rst one takes into account that may vanish on a nonnegligible subset of R:
260 Lemma 1. (i) k E and a.e. on Q,
sgn(u k) = sgn((u) (k)) = sgn((, ., .) k) for a.e. in ]0, 1[. (ii) () = 0 a.e. on Q0 {(,t, x) Q , (,t, x) E} Proof. For the rst equality in (i) we remark that when k belongs to E, if u(t, x) > k then (u(t, x)) > (k). This way the second equality is a consequence of (11) and uses the same reasoning as for the rst equality with in the place of u. The point (ii) is already emphasized ([2]) with the setting of entropy solutions. It uses the monotony of 1 so 0 that Im \ (E) - the set of points where 1 is discontinuous - is a countable (and thus a 0 Lebesgue-negligible) subset of R. The second lemma may be viewed as an inequality version of the energy equality stated by J.Carrillo in [2]. It permits to determine the sign of diffusive terms appearing in the method of doubling variables. We prove that this inequality is fullled by any entropy process solution to (1) but only for k in E. We emphasize that in [11] the forthcoming relation (16) results from an underlying formulation of the unilateral obstacle problem for P through a strong variational inequality (in the sense of J.L.Lions in [12]). But here, in the context of the bilateral obstacle problem, we have not been able to establish such a formulation and so we directly start from (13). Indeed: Lemma 2. Let be an entropy process solution to (1). Then, for any real k of [a, b] E, for any nonnegative function of D (Q), k .dxdt
Q
G(, k)dq lim sup

Q
0+
sgn(() (k))[()]2dq.
Q
Proof. We consider (13) for any nonnegative in D (Q) and, thanks to a density argument, 1 for any nonnegative in H0 (Q) so as to choose the test-function = |sgn ((u) (k))| with in D (Q), 0. Let us perform the following transformations (by setting S (v) = sgn((v) (k))): For the evolution term, | k|t dq = I1 + I2 where,
Q
I1 =
Q
| k||S(u)|t dq, | k|sgn((u) (k))sgn((u) (k))t (u)dq.

Q
I2 =
In I2 we have taken into account that sgn(S(u)) = sgn((u) (k)) a.e. on Q when k belongs to E. But due to Lemma 1 (i), | k| = sgn(u k)( k) a.e. on Q and sgn((u)
261
(k)) = sgn(u k) a.e. on Q. So the integration with respect to may be calculated. We get: I2 =
Q
sgn ((u) (k))(u k)t (u)dxdt t (S (u))(u k)dxdt

Q
t S (u)(u k)dxdt.
Q
Since t u is an element of L2 (0, T ; H 1 ()),

T
t (S(u))(u k)dxdt =
Q 0
t u, S(u)
1 H 1 ,H0 dt.
An integration by parts in time using the Mignot-Bamberger formula ([8]) gives

T u
t u, S(u)
1 H 1 ,H0 dt
S ()d t dxdt.
We now take the -limit through the Lebesgue dominated convergence Theorem in I1 and I2 . Since k belongs to E, I2 goes to 0 + (Lemma 1 (i)). For the diffusive term, considering that sgn(S()) = sgn(() (k)) a.e. on Q provides
Q
|() (k)|.dq [|() (k)|.]|S()|dq

Q Q
sgn(() (k))[()]2dq.
Concerning the convective term, as a consequence of (11), we write: F(, k).dq =

Q Q
[F(, k).]|S()|dq +
Q
F(, k).|S()|dq.
To take the -limit, we transform the second term in the right-hand side by by taking into account that due to Lemma 1 (i), sgn(S()) = sgn(() (k)) = sgn( k) a.e. on Q . There follows F(, k).|S()|dq
Q
=
Q
sgn (() (k)){(t, x, ) (t, x, k)}.()dq.
Now, we introduce the vector-valued function

r
H (t, x, r) =
(k)
[(t, x, 1()) (t, x, k)]sgn( (k))d, 0
262
so that thanks to Lemma 1 (ii), sgn (() (k)){(t, x, ) (t, x, k)}()dq

Q
=
Q
DivH (t, x, ())dq

()
(k)
Divx (t, x, 1()) Divx (t, x, k) sgn( (k))ddq, 0
The rst term is integrated by parts and for the second one we come back to the denition of sgn to obtain: F(, k).|sgn(() (k))|dq
Q
=
Q
1 1
[(t, x, 1()) (t, x, k)].d dq 0

I
Divx [(t, x, 1()) (t, x, k)]d dq, 0

I
where I = I((), (k)) [(k) , (k) + ] and I(r, p) = [min(r, p), max(r, p)] for all (r, p) of R2 . Since k belongs to E, the generalized function 1 is continuous in (k) ([2]). 0 This way, the previous integral goes to 0 when tends to 0 +. The study of the reaction term does not present any difculty. This way the next Kuzhkov-type relation holds: Proposition 2. Let 1 and 2 satisfying (6,10,11,13). Then for any nonnegative of D (Q Q):
Q Q
|1 2 |(t + t ) (x + x )|(u1) (u2)|.(x + x )dqd q F(1 , 2 ).x + F(2, 1 ).x dqd q (G(1, 2) + G(2, 1))dqd q 0.
Q Q
+
Q Q
Proof. To simplify we set d q = d d xdt and we add a tilde superscript to any function in tilde variables. On the one hand in (16) written in variables (,t, x) for 1 we take k(, t , x) = 2 for a.e. (, t , x) in Q \ Q02 , so that k(, t , x) belongs to E. On the other hand in (13) written in variables (,t, x) for 1 , we take k(, t , x) = 2 for a.e. (, t , x) in u2 Q0 . Each inequality obtained is integrated with respect to the variables , t and x on the corresponding domain. We add and use a version of the Fatous Lemma to deal with the
263
lim sup in the right-hand side. Indeed we observe by coming back to the proof of Lemma 2 for xed, that the function k
Q
sgn (() (k))[()]2dq
is uniformly bounded with respect to and k in [a, b]. Finally we obtain for 1:
Q Q
|1 2 |t dqd q +
Q Q
[x |(1) (2)| F(1 , 2 )].x dqd q
+
Q Q
G(1, 2) dqd q sgn ((1) (2 ))[(1)]2 dqd q,

Q
lim sup
0+
where in the right-had side we have used the lemma 1 (ii) to rewrite the integration eld under the form Q = Q \ Q01 Q \ Q02 . Besides, we integrate over Q the Gauss-Green formula: x (u1).x [sgn((u1) (u2))] dt d x = 0 a.e. (t, x) in Q.
Q
By developing the partial derivatives and taking into account that (u2) is an element of 1 L2 (0, T ; H0 ()), the -limit provides the next equality: sgn((1) (2 ))x(1).x dqd q
Q Q
= =
0+
lim
sgn((1) (2 ))x(1).x (2) dqd q sgn ((1) (2 ))x (1).x(2 ) dqd q.

Q
Q Q
0+
lim
We apply the same reasoning with 2 and bring the results together to obtain:
Q Q
|1 2 |(t + t ) (x + x )|(1) (2 )|.(x + x )dqd q F(1, 2).x + F(2, 1).x dqd q

Q Q
+
Q Q
(G(1, 2 ) + G(2 , 1))dqd q sgn((1) (2)) [x (1) x (2)]2 dqd q.

Q
lim sup
0+
264
Laurent L vi and Guy Vallet e The desired inequality follows.
From Proposition 2 we may state the main result of this section, that is the T Lipschitzian dependence in L1 of an entropy process solution to (1) with respect to the corresponding initial data. For the treatment of boundary terms, the proof follows C.Mascia, A.Porretta & A.Terracinas one ([13]) but it needs to be transcribed in the framework of entropy process solutions. It leads to: Theorem 1. If 1 and 2 are two entropy process solutions to (1) corresponding to initial data u 0,1 and u0,2 respectively then for a.e. t of [0, T ], |1 (,t, x) 2 (,t, x)|dxdd eM t
]0,1[2
|u0,1 u0,2 | dx.
Proof. In Proposition 2 we choose = (t, x, t, x)l (x)m (x) for any boundary-layer sequences (l )l>0 and (m )m>0 and any nonnegative in D ((]0, T[)2). We develop the partial derivatives and we argue that due to (3) and (4),
m0+
lim lim
l0+
Q Q
x |(1) (2 )| F(1 , 2) m x l dqd q
=
Q
12 , d q,
where 12 refers to 2 with = 1. Similarly lim lim

l0+
Q Q
m0+
x |(1) (2 )| F(2 , 1) l x m dqd q
=
Q
21 , dq,
and we obtain:
QQ
12 .(t,x) dxdtd xdt +

Q Q
G(1, 2 )dqd q
+
Q Q
(x |(1) (2 )|.x + x |(1 ) (2 )|.x )dqd q 21 .(t,x) dxdtd xdt +

QQ
G(2, 1 )dqd q (16)
Q Q
1 2 , d q
Q
21 , dq I1 I2 ,
Mathematical Analysis of a Bilateral Obstacle Problem for a Class ... where I1 = lim+
m0
265
x |(1) (2 )|.xm dqd q, x |(1) (2 )|.xl dqd q.
Q Q
I2 = lim +
l0
Q Q
An integration by parts with respect to x and then with respect to x allows us to express the limit with respect to m in I1 . Indeed:
m0+
lim
x |(1) (2 )|.xm dqd q =

Q Q
Divx[|(1) (2 )|.x ]dqd q.
Q Q
An integration by parts with respect to x and then with respect to x provides: I1 =

Q
x |(1)|.dqd H(p,x) . t
With the same arguments, I2 =

Q
p x |(2)|.d qd H(t,x) .
We take now into account (14) for 1 and 2 to have a majoration of the right-hand side of (16) in terms of:
p x |(1)| F(1 , 0) .dqd H(t,x) +
1 , d q 0
Q
+
Q
p (x |(2 )| F(2, 0)).d qd H(t,x) +
0 2, dq.
Q
(17)
We choose n = Wn(x x)wn (t t ) where is a nonnegative element of D (]0, T[), (Wn)n and (wn )n are the standard mollier sequences on R p and R. In addition, n is large enough so that for any t in ]0, T [, t (t)wn(t t ) and for any t in ]0, T [, t (t)wn(t t ) belongs to D (]0, T[). We take the limit with respect to n in (17) by referring to [13] providing that each line goes to 0. For the left-hand side of (16), we use classical techniques and the fact that (t + t )n = (t)Wn(x x)wn (t t ) and (x + x )n = 0. Eventually one gets,
Q ]0,1[
|1 2 | (t)dqd M
Q ]0,1[
|1 2 |dqd ,
1 for any nonnegative in D (]0, T [) and so, by density, for any nonnegative in H0 (]0, T[). Now the conclusion is classical: it uses for a Lipschitzian and piecewise linear approximation of I[0,] , for a.e. in ]0, T [, the initial condition (15) for 1 and 2 and the Gronwall
266 Lemma.
When u0,1 = u0,2 a.e. on , we deduce (see [7] or [5] in the framework of Young measure solutions) the existence of a measurable function on Q, such that 1 (,t, x) = 2 (,t, x) = (t, x) for a.e. and in ]0, 1[ and for a.e. (t, x) in Q. Thus in Denition 2 the integrations with respect to the Lebesgue measure on ]0, 1[ may be performed. So that = u a.e. on Q and u is namely an entropy solution to (1) in the sense of Denition 1. As a consequence, Corollary 1. If (1) has an entropy process solution then it has an entropy solution. In addition, if u 1 and u2 are two entropy solutions corresponding to initial data u 0,1 and u0,2 respectively, then for a.e. t in ]0, T [: |u1 (t, x) u2(t, x)|dx eM t

|u0,1 u0,2 |dx.
3.
The Viscosity and Penalization Methods
Our aim is to obtain an existence result for (1) by introducing some diffusion on the whole domain and by relaxing the bilateral obstacle condition. This leads us to introduce rst, for any function f (or f (t, x, .)), the Lipschitz bounded extension f of f outside [a, b] dened through: f (u) = f (u) if u [a, b], f (u) = f (a) if u a, f (u) = f (b) if u b, and similarly for f (t, x, .) (in this case, observe that xi f = (xi f ) ). Now for any positive parameter intended to tend to 0 + - so that it will be supposed less or equal than a xed value 0 - we dene = + IdR a bilipschitzian function and we consider the nondegenerate penalized parabolic operator P (t, x, .) : u t u + xi i (t, x, u) + (t, x, u) (u) + (u),
i=1 p
1 with (z) = ((z a) + (z b)+). We consider the resulting problem: nd a measur able and bounded function u on Q satisfying P (t, x, u) = 0 on Q, u = 0 on ]0, T [, u (0, .) = u on , 0 where u is a regularization of u0 obtained by the mean of molliers, so that: 0 > 0, u D (), a u b a.e. in , lim u = u0 in L1 (). 0 0 0
0+
(18)
267
3.1.
Some a priori Estimates
We know (see for example [10]) that for a xed positive , the problem (18) has a unique 1 weak solution u in L (Q) H 1 (Q) L(0, T ; H0 ()) with (u ) in L2 (0, T ; H 2 ()). In this section, we are mainly interested in the -uniform a priori estimations for the sequence (u )>0 . Indeed, Proposition 3. There exist some constants C1, C2 and C3 independent from such that t [0, T ], |u(t, .)| M(t) a.e. in , 1 1 (u a) L2 (Q) + (u b)+ L2 (Q) C1 1 (u ) L2(0,T ;H0 ()) + t u L2 (0,T ;H 1 ()) C2 , t t (u ) L2 (Q) C3,
x
(19) (20) (21) (22)
where M(t) is dened by (2) and (x) =

0
( ())1/2 d.
Remark 2. We have not been able to establish any -uniform estimates for (u )>0 in W 1,1 (Q) or even in BV (Q) as in [11] for the positiveness obstacle problem for P when is a W 2,+ -class vector-valued function and a W 1,+-class function. Furthermore even if we have enough compactness for ( (u ))>0 , we cannot take advantage of it since ( )1 may not exist. So to describe the behavior of (u )>0 when goes to 0+ we can only refer to (19), which leads to consider the notion of an entropy process solution to (1). Proof of Proposition 3 - With the notations of Section 1.2., M = M and Mxi i = Mxi i . So the standard maximum principle for u ensures (19), the independence with respect to resulting from the monotonicity of the penalized operator . For (20), we treat each 1 term separately: to estimate (u b)+ we take the L2 (Q)-scalar product between (18) and 1 + (u b) . We observe that 1 1 1 t u (u b)+ dxdt = t [(u b)+]2 dxdt = [(u b)+ ]2(T, x)dx, 2 2
Q Q
since a u b a.e. on . So the contribution of the evolution term is nonnegative and it is 0 the same for the diffusive one, thanks to the Green formula. For the convective and reactive integrals, due to the denition of and , 1 = 1 (Div (t, x, u) + (t, x, u ))(u b)+ dxdt
Q
(Div(t, x, b) + (t, x, b))(u b)+ dxdt

Q 2 L2(Q) +
1 (t, x, b) + (t, x, b) 2
1 1 (u b)+ 2
2 L2 (Q) dxdt.
268 Eventually, we remark that 1
(u )(u b)+ dxdt =

Q
1 (u b)+
2 L2 (Q) .
So we conclude the existence of a constant C, independent from , such that The previ1 ous techniques with (u a) provide (20). The energy estimate (21) results from the L2 (Q)-scalar product between (18) and u . Since (u )u is nonnegative a.e. in Q, it warrants thanks to (19) a -uniform bound for (u ) in L2 (Q) p . Then by coming back to the denition of the norm in L2 (0, T ; H 1 ()) and referring to (20) we derive the estimation of t u . Let us focus on (22). We take the L2 (Q)-scalar product between (18) and tt (u ). This way, tt u t (u )dxdt =
Q
1 tt (u ) 2
2 L2 (Q) .
For the diffusive integral, since w = (u) belongs to the functional space W (0, T ) 1 1 {v L2 (0, T ; H0 () H 2()); t v L2 (Q)}, we use the density of D ([0, T ]; H0 () 2 H ()) into W (0, T ) to carry out calculations with a sequence (wk )k of mollied functions with respect to the time variable. One has:
Q
twk t wk dxdt =
Q
twk .t wk dxdt
Q
t t [wk ]2 dxdt 2
1 2
Q
[wk ]2 dxdt +
T [wk ]2 (T, x)dx. 2
Then, we pass to the limit with k to obtain t (u )t (u )dxdt =

Q
1 2
Q
[ (u )]2 dxdt
T [ (u )]2 (T, x)dx 2
1 2
Q
[ (u )]2 dxdt C,
where C is a constant independent from thanks to (19) and (21). We develop the partial derivatives in the convective term to write: tDivx (t, x, u)t (u )dxdt
Q
=
Q
i=1 p
u i (t, x, u) txi (u ) tt (u) dxdt xi i (t, x, u) t ( ) (u ) tt (u ) dxdt.
+
Q
i=1
Mathematical Analysis of a Bilateral Obstacle Problem for a Class ... The Young inequality with p = 2 and (19) prove that: tDivxi (t, x, u)t (u )dxdt C (u )
Q 2 L2 (Q) p
269
1 tt (u) 4
2 L2 (Q) .
Again, with the Young inequality and (19): t (t, x, u)t (u )dxdt C +
Q
1 tt (u ) 4
2 L2 (Q),
where C doest not depend on thanks to (19). It is the same for the penalized term since, 1 1 due to (20), (u a) and (u b)+ are bounded in L2 (Q) uniformly with respect to , that completes the proof of Proposition 3. We give now a formulation of boundary conditions for the solutions to the nondegenerate relaxed problem that will be the starting point to derive the formulation of boundary conditions for the solution to (1) by taking the limit with respect to . In what follows F , G and Uk, refer to F, G and Uk with , and in the place of , and respectively. This way, by arguing as in [11, 18], we prove that: Proposition 4. For any , the next compatibility condition holds on : F(0, k).d H p Uk, ,

+ U0,,
(23)
for all nonnegative in L (Q) H 1 (Q) C (Q), (T, .) = (0, .) = 0 and any real k. Moreover there exists a constant C4, independent from such that Div(t,x)Uk, Mb(Q) C4. (24)
Proof. First point - We observe that | (u ) (k)| is a bounded Radon measure on Q, for any real k. As in [17], this assertion comes from the Katos Inequality and uses the fact that (u ) is an element of L2 (0, T ; H 2()). As a consequence, Uk, and (0, |(u ) (k)|) belong to DM 2 (Q) for any real k. Thus, by using the boundary condition for u and the relation (4) with (T, .) = (0, .) = 0, we may successively write: Uk,,
0
= lim+
Q
Uk,.(0, ) dxdt | (u ) (k)|.dxdt lim+

0
= lim+
0 Q
F (u , k). dxdt
Q
= lim+
0 Q
(0, |(u ) (k)|).(0, ) dxdt +
F(0, k). d H p
= (0, |(u ) (k)|), +
F(0, k). d H p .
270
Laurent L vi and Guy Vallet e Let us conclude through the technical property:
1 Claim 1. ([18]) Let w be in L2 (0, T ; H0 ()) such that w 0 a.e. on Q and w belongs to Mb (Q). Then, for any nonnegative of L (Q) H 1 (Q) C (Q)
(0, w),
0.
With w = | (u ) (k)| + | (u )| | (k)| provides (23) since, due to (4), (0, | (u))|,
= U0,,
We remark now that u fullls (8) with in the place of . Indeed for any nonnegative function of D (] , T [), we take the L2 (Q)-scalar product between (18) and the testfunction sgn (u k). As soon as k belongs to [a, b], (u)sgn (u k) is nonnegative.
v
After some integration by parts we obtain (with S (v) = sgn(v k) and I (v) = S ()d): I (u )t dxdt +
Q Q
u (t, x, )S()d .dxdt

k
I (u )(0, .)dx 0
Q u
(Div (t, x, k) + (t, x, u)) S (u )dxdt
[Div (t, x, ) Div (t, x, k)]S ()d dxdt

k
S (u ) (u ).dxdt.
(25)
1 For a.e. x, sgn (x) = I],[ (x) and since the function z Div (t, x, z) is continuous uniformly with respect to (t, x) in Q, the third term in the left-hand side vanishes when goes to 0 + . Thus (8) for u is obtained at the limit thanks to the Lebesgue dominated convergence Theorem. We deduce that for any real k of [a, b], k, Div(t,x)Uk, G (u , k) is a nonnegative Radon measure on Q such that k, Mb (Q) =
Q
(26)
dk, =
Q
d[Div(t,x)Uk,]
Q
G (u , k) dxdt.
But by using (3) and (4):

Q
d[Div(t,x)Uk,] F(0, k).d H p +

(0, | (u) (k)|), 1 |u (T, x) k|dx.
|u k|dx 0
271
Then as a consequence of (19) and since u is uniformly bounded with respect to , we 0 claim the existence of a constant C independent from such that:
Q
d[Div(t,x)Uk, ] (0, | (u ) (k)|), 1 +C.
On the one hand, the previous inequality with k = 0 and Claim 1 with w = | (u )| provide
Q
d[Div(t,x)U0, ] C,
that means that (0,)>0 given by (26) is a bounded sequence in Mb (Q) and (Div(t,x)U0, )>0 too. On the other hand for any k in [a, b], Claim 1 with w = | (u ) (k)| + | (u )| | (k)| gives:
Q
d[Div(t,x)Uk,] (0, | (u)|), 1 +C.
But by referring to (3) and (4) and using the homogeneous boundary condition for u , we observe that: (0, | (u )|), 1
= U0,, 1
|u |dx 0
|u(T, x)|dx.
Thus, due to the uniform boundedness for u , there exists a constant C independent 0 from such that: | (0, | (u )|), 1 | | U0,, 1
| +C Div(t,x)U0, Mb (Q) +C.
Eventually the estimate of Div(t,x)U0, highlighted previously ensures that the sequence (k,)>0 is a uniformly bounded in Mb(Q) and so (Div(t,x)Uk, )>0 too. Relation (24) follows, which completes the proof of Proposition 4.
3.2.
Convergence Toward an Entropy Process Solution
Due to Proposition 3, there exists a measurable function u on Q such that, up to a subsequence when goes to 0 +, (u )>0 converges toward u in L (Q) weak and (t u )>0 weakly converges toward t u in L2 (0, T ; H 1 ()). Besides there exists a function 1 such that, up to a subsequence, ( (u ))>0 goes to in L2 (0, T ; H0 ())weak. But 1 (Q) with respect to , we may be sure since (t(u ))>0 is uniformly bounded in H that ( (u ))>0 strongly converges (up to a subsequence) toward in Lq (Q) for any q in [1, +[. In order to connect (u) and , we rst remind some properties of bounded sequences in L :
272
Claim 2. ([7]) Let (un)n>0 be a sequence of measurable functions on an open bounded subset O such that (un )n is uniformly bounded in the L (Q)-norm the by a constant M. Then, there exists a subsequence (u(n))n>0 and a measurable and bounded function on ]0, 1[O such that for all continuous and bounded functions h on O ] M, M[, L1 (O ), lim
n+
h(, u(n))dx =
O
]0,1[O
h(, (, ))dd.
Such a result has found its rst application in the approximation through the articial viscosity method of the Cauchy problem in R p for a scalar conservation law, as one can establish a uniform L -control of approximate solutions. It has also been applied to the numerical analysis of transport equations since Finite-Volume schemes mainly give an L -estimate uniformly with respect to the mesh length of the numerical solution. Here, we refer to this concept when the approximating sequence is (u)>0 and so there exists a function in L (]0, 1[Q) such that thanks to (20), a b a.e. in ]0, 1[Q. Of course, we also have for a.e. (t, x) in Q,
1 1
u(t, x) =
0
(,t, x)d and (t, x) =

0
((t, x, ))d.
Furthermore we remark that |((, t, x)) (t, x)|ddxdt = lim

]0,1[Q 0+
| (u ) (t, x)|dxdt = 0.
Q
Thus we even have, for a.e. (,t, x) of ]0, 1[Q, ((, t, x)) = (t, x). But since is a nondecreasing function, for a.e. (t, x) in Q, 1 ({(t, x)})= [i1, i2], where i1 (.,t, x) i2 a.e. in ]0, 1[. This way, by integrating from 0 to 1 we prove that (i1) = (u(t, x)) = (i2) = (t, x) that is namely (11). The previous developments guide us toward the next statement: Theorem 2. - The obstacle problem (1) admits an entropy process solution. Proof. We have highlighted a function u and a process such that (10,11,6) hold. Moreover due to (24) there exists an element of Mb (Q) - identied in the sense of distributions on Q to Div(t,x) k - such that up to a subsequence when goes to 0 + , (Div(t,x)Uk,)>0 converges toward Div(t,x) k in Mb(Q) weak and (12) follows. To establish (13) we start from (25). For the left hand-side of the -limit only refers to the claim 2 while for the right-hand side of (25) we use the smoothness of u and the Green formula: u
S (u ) (u ).dxdt =
Q Q
S ()() ()d .dxdt
=
Q
K (u )dxdt,

x
273
where K is the the continuous function x

k
S ()( ) ()d. We pass to the limit with
by arguing that |H(x) H(x)| |x k| and referring to Claim 2. It follows

k u0
S ()d t dq +
u (t, x, )S()d.dq
Q k
S ()d (0, .)dx
(Div(t, x, ) + (t, x, ))S()ddq

Q
Q k
[Div(t, x, ) Div(t, x, k)]S()ddq

k
S () ()d dq.
The -limit relies on the Lebesgue dominated convergence Theorem and uses the continuity, uniformly with respect to (t, x) in Q, of z Div(t, x, z) to deal with the third line in the left-hand side. Note that for the right-hand side:
0+ Q]0,1[
lim
S () ()d ddxdt =
|() (k)|ddxdt.
Q]0,1[
An integration by parts gives (13). We establish (14) by passing to the -limit in (23) and using that: Claim 3. - ([18]) - Let ( )>0 be a sequence of Mb(Q), 0, converging toward in Mb (Q) weak . Then, for any nonnegative of L (Q) C (Q), d lim inf +
0 Q Q
d .
This claim with the nonnegative Radon measure k, dened by (26) provide that for any nonnegative of H 1(Q) L (Q) C (Q), with (T, .) = (0, .) = 0, up to a subsequence, dk lim inf
Q 0+ Q
dk, ,
where in the sense of the bounded Radon measures on Q, k = Div(t,x) k + G with G = lim G (u, k) in L (Q) weak (up to a subsequence). This way,
0+
lim sup
0+ Q
d[Div(t,x)Uk,]
Q
d[Div(t,x) k ].
274
Laurent L vi and Guy Vallet e Eventually, because (Uk,)>0 weakly converges in L2 (Q) p+1 toward k , lim sup Uk,,
0+
=
Q
k .dxdt + lim sup

0+ Q
d[Div(t,x)Uk,],
and this way, for any real k of [a, b], lim sup Uk,,
0+
k , .
Relation (14) follows, that completes the proof of Theorem 2. As pointed out in R.Dipernas paper [5] within the framework work of Young measure solutions, the strong convergence of approximate solutions occurs if and only if the process may be identied to a function independent from the variable , so that the associated Young measure reduces to a Dirac mass at almost all points of the physical domain. So, as a consequence of Theorem 2 and Corollary 1 we claim that: Theorem 3. - The bilateral obstacle problem (1) admits a unique entropy solution that is the limit of the whole sequence (u )>0 of solutions to (18) >0 when goes to 0+ in Lq (Q), 1 q < +.
References
[1] R. B urger, H. Frid & K.H. Karlsen, On a free boundary problem for a strongly degenerate quasi-linear parabolic equation with an application to a model of pressure ltration. SIAM J. Math. Anal. 34, No.3, 611-635 (2003). [2] J.Carrillo, Entropy solutions for nonlinear degenerate problems. Arch. Ration. Mech. Anal. 147, No.4, 269-361 (1999). [3] G.Q.Chen & H.Frid, Divergence-measure elds and hyperbolic conservation laws. Arch. Ration. Mech. Anal. 147, No.2, 89-118 (1999). [4] G.Q.Chen & H.Frid, On the theory of divergence-measure elds and its applications. Bol. Soc. Bras. Mat., Nova Sr. 32, No.3, 401-433 (2001). [5] R.J.Diperna, Measure-valued solutions to conservation laws. Arch. Ration. Mech. Anal. 88, 223-270 (1985). [6] G.Duvaut & J.L.Lions, Les in quations en m canique et en physique. Travaux et e e recherches math matiques. 21. Paris: Dunod. XX. (1972). e [7] R.Eymard, T.Gallou t & R.Herbin, Existence and uniqueness of the entropy solution e to a nonlinear hyperbolic equation. Chin. Ann. Math., Ser. B 16, No.1, 1-14 (1995). [8] G.Gagneux & M.Madaune-Tort, Analyse math matique de mod` les non lin aires de e e e ling nierie p troli` re. Math matiques & Applications (Paris). 22. Paris: Springere e e e Verlag. xiv, 188 p. (1995).
275
[9] S.N.Kuzhkov, First order quasilinear equations in several independent variables. Math. USSR, Sb. 10, 217-243 (1970). [10] O.A.Ladyzhenskaya, V.A.Solonnikov, N.N.Uraltseva, Linear and quasi-linear equations of parabolic type. Translations of Mathematical Monographs. 23. Providence (AMS). XI, 648 p. (1968). [11] L. L vi, The Positiveness Problem for a Class of Degenerate Parabolic-Hyperbolic e Operators Adv. Math. Sci. Appl. 15, No 1, 307-333 (2005) [12] J.L.Lions, Quelques m thodes de r solution des probl` mes aux limites non lin aires. e e e e Etudes math matiques. Paris: Dunod; Paris: Gauthier-Villars. XX, 554 p. (1969). e [13] C.Mascia, A.Porretta & A.Terracina, Nonhomogeneous Dirichlet problems for degenerate parabolic-hyperbolic equations. Arch. Ration. Mech. Anal. 163, No.2, 87-124 (2002). [14] A.Michel & J.Vovelle, Entropy formulation for parabolic degenerate equations with general Dirichlet boundary conditions and application to the convergence of FV methods. SIAM J. Numer. Anal. 41, No.6, 2262-2293 (2003). [15] F.Otto, Initial-boundary value problem for a scalar conservation law. C. R. Acad. Sci., Paris, Sr. I 322, No.8, 729-734 (1996). [16] J.F.Rodrigues, Obstacle problems in mathematical physics. North-Holland Mathematics Studies, 134, Notas de Matemtica (114). Amsterdam: North-Holland. XV, 352 p. (1987). [17] E.Rouvre & G.Gagneux, Formulation forte entropique de lois scalaires hyperboliquesparaboliques d g n r es. Ann. Fac. Sci. Toulouse, VI. S r., Math. 10, No.1, 163-183 e e ee e (2001). [18] G.Vallet, Dirichlet problem for a degenerated hyperbolic-parabolic equations. Adv. Math. Sci. Appl. 15, 423-450 (2005)
A PPLICATION OF THE P ICARD O PERATORS TO S ECOND O RDER ODE S

1 Department
Alexandru Bica1 Loredana Galea2 and Sorin Muresan1 , of Mathematics and Informatics, University of Oradea Str. Universit ii No.1, 410087, Oradea, Romania at 2 Faculty of Law and Economics, The Agora University of Oradea Piata Tineretului no.8, 410526, Oradea, Romania
Abstract Using the Perovs xed point theorem and the ber contractions theorem are obtained sufcient conditions for the smooth dependence by the end points of the solution of the two point boundary value problem for second order ODEs. Finally, an application in the metabolism control is presented.
2000 AMS Mathematics Subject Classication: 47H10. Keywords and phrases:: two point boundary value problem, Picard operators, ber contractions theorem, smooth dependence by the end points.
1.
Introduction
Consider the following two point boundary value problem corresponding to the nonlinear second order ODE: x (t) = f (t, x (t), x (t)) , t [a, b] (1) x (a) = , x (b) = , , R The existence and uniqueness of the solution of (1) is studied in [1] and [2]. In [2] is presented a new approach for the problem (1) using the Perovs xed point theorem and considering the equivalence of the problem (1), in C2 [a, b], with the following system of Fredholm integral equations: b x (t) = ta + bt G (t, s) f (s, x (s) , y (s))ds ba ba a (2) b y (t) = G (t, s) f (s, x (s), y (s)) ds , ba t
a
E-mail E-mail
address: abica@uoradea.ro address: loredana.galea@univagora.ro E-mail address: smuresan@uoradea.ro
278
Alexandru Bica, Loredana Galea and Sorin Muresan
where was considered y (t) = x (t). Here, G (t, s) is the well known Green function. The Perovs xed point theorem (see [2], [4], [6], [7], [9], [10], [11] and [13]) generates an efcient technique to approach existence, uniqueness and approximation of the solution of the system of equations (2) (and consequently for the problem (1)). Theorem 1 (Perov, [2], [5], [7] ) Let (X, d) be a generalized metric space (with the metric taking values in Rn ) and A : X X a mapping for which there exist a matrix Q Mn (R+) + such that: d (A (x) , A (y)) Qd (x, y) , x, y X (3) If all eigenvalues of Q lies in the open unit ball from the complex plane, then: (i) the operator A has a unique xed point x X (ii) for any x 0 X, the sequence of successive approximations (xm )mN X dened by xm = A (xm1 ) , m N , is convergent to x (iii) the following inequality holds: d (xm , x ) Qm (In Q)1 d (x0 , x1 ), m N.
Here, In is the unity matrix in Mn (R) and in the inequality (3) the order relation is dened in Rn component-wise, Rn = x = (x1 , ..., xn) Rn |xi 0, i = 1, n . + + In the study of smooth dependence of the solution of the problem (1), by data and by parameters, are very useful the notions of Picard and weakly Picard operators (see [8], [11]), and the ber contractions theorem (see [9], [10], [11], [6]). Below, we investigate the dependence of the solution of (1) by the end points a and b.
2.
Preliminaries
Denition 1 ([8], [10], [11]) Let (X, d) be a metric space. An operator A : X X is Picard operator if there exists x X such that: (i) x is the unique xed point of A . (ii) the sequence (An (x0 ))nN converges to x , for all x 0 X, where A0 = Id (X) and n+1 = A An , n N. A Denition 2 ([8], [10], [11]) Let (X, d) be a metric space. An operator A : X X is weakly Picard operator if the sequence (An (x0 ))nN converges for all x 0 X and the limit (which may depend on x 0 ) is a xed point of A. Applications of the technique of Picard and weakly Picard operators can be viewed in [4], [11], [13]. The above denitions can be extended in the same way for operators on generalized metric spaces (with the metric having vectorial values). Using the Perovs xed point theorem I.A. Rus have obtained the result: Theorem 2 (of ber generalized contractions, Rus [9], [10]) Let (X, d) be a metric space (generalized or not) and (Y, ) be a complete generalized metric space (x, y) Rn . Let + the operators B : X X, C : X Y Y and the continuous operator A : X Y X Y . Suppose that:
Application of the Picard Operators to Second Order ODEs (i) B : X X is weakly Picard operator (ii) A (x, y) = (B (x) ,C (x, y)), for all x X, y Y (iii) there exist a matrix Q Mn (R+) with Qm 0 as m , such that: (C (x, y1 ) ,C (x, y2 )) Q (y1 , y2 ) ,
279
for all x X and y1 , y2 Y . Then, the operator A is weakly Picard operator. Moreover, if B is Picard operator, then A is Picard operator. Remark 1 (see [2]) For a matrix Q Mn (R+ ), the following properties are equivalent: (i) Qm 0 as m (ii) all eigenvalues of Q lies in the open unit ball from the complex plane.
3.
The Main Result
In the following, we will study the smooth dependence by the end points a and b of the solution of the problem (1). In this aim we will use the Picard operators technique. Other way to approach data dependence of the solution of differential equations can be found in [12]. Consider c, d R such that c a < b d and the two point boundary value problem, xt 2 (t; a, b) = f (t, x (t; a, b), xt (t; a, b)) (4) x (a; a, b) = , x (b; a, b) = This problem is equivalent with the following integro-differential equation: bt t a + x (t; a, b) = ba ba
b
G (t, s) f s; x (s; a, b), xt (s; a, b) ds

a
(5)
for all t, a, b [c, d] with a t b, where G is the well-known Green function. We see that, denoting y = xt , the study of the equation (5) it reduces to the study of the following system of integral equations: b x (t; a, b) = ta + bt G (t, s) f (s, x (s; a, b), y (s; a, b)) ds ba ba a (6) b y (t; a, b) = G (t, s) f (s, x (s; a, b) , y (s; a, b)) ds ba t
a
where
(s a) (b t) , if s ba G (t, s) = (t a) (b s) , if s ba
t , t
280
Alexandru Bica, Loredana Galea and Sorin Muresan sa , if s t G ba . (t, s) = bs t , if s t ba Let C [c, d]3 = u : [c, d]3 R u continuous and be the Chebychevs norm on
C [c, d]3 , u = max |u (x, y, z)| : (x, y, z) [c, d]3 . Consider the space X = C [c, d]3 C [c, d]3 with the generalized metric d : X X R2 , d ((x1 , y1 ) , (x2 , y2 )) = ( x1 x2 , y1 y2 ) , (x1 , y1) , (x2 , y2 ) X.
Dene the operator B : X X, B (x, y) = (B1 (x, y), B2 (x, y)) with bt t a B1 (x, y) (t; a, b) = + ba ba B2 (x, y) (t; a, b) = ba [c, d] R2
b b
G (t, s) f (s, x (s; a, b) , y (s; a, b))ds

a
(7)
G (t, s) f (s, x (s; a, b) , y (s; a, b)) ds t
(8)
in the condition f C . In the supplementary condition f (s, , ) C1 R2 , s [c, d] we can dene the operator C : X X X, C ((x, y) , (u, v)) = (C1 ((x, y) , (u, v)) ,C2 ((x, y), (u, v))) with C1 ((x, y) , (u, v)) (t; a, b) =
b
(b t) ( ) (b a)2
+ G (t, a; a, b)
(9)
f (a, x (a; a, b) , y (a; a, b))

a
G (t, s; a, b) f (s, x (s; a, b), y (s; a, b)) +G (t, s; a, b) a
f f (s, x (s; a, b) , y (s; a, b))u (s; a, b) + (s, x (s; a, b) , y (s; a, b)) v (s; a, b) )ds x y C2 ((x, y) , (u, v))(t; a, b) =
b
( ) (b a)
2
G (t, a; a, b) t
(10)
f (a, x (a; a, b) , y (a; a, b))

a
G 2 G (t, s; a, b) f (s, x (s; a, b), y (s; a, b)) + (t, s; a, b) ta t
f f (s, x (s; a, b) , y (s; a, b)) u (s; a, b) + (s, x (s; a, b), y (s; a, b)) v (s; a, b) )ds. x y
Finally, let A : X X X X be the operator dened by A ((x, y) , (u, v)) = (B (x, y) ,C ((x, y) , (u, v))).
Application of the Picard Operators to Second Order ODEs For (x, y) X xed, consider the system, u (t; a, b) := second part of (9) v (t; a, b) := second part of (10)
281
(11)
We will suppose that the following conditions holds: (S1)(continuity): f C [c, d] R2 ; (S2) (Lipschitz): there exist L1 > 0, L2 > 0 such that: | f (s, z1, w1 ) f (s, z2, w2 )| (S3)(smoothness): L1 |z1 z2 | + L2 |w1 w2 |, s [c, d] , z1, z2, w1 , w2 R
f (s, , ) C1 R2 , s [c, d] .
Theorem 3 a) In the conditions (S1), (S2) if L 1 (d c)2 < 1 and L2 (d c) < 3 then the 4 system (6) of integral equations has in X an unique solution (x , y) such that x , y C1 [c, d]3 , x (; a, b) C2 ([c, d]), a, b [c, d] and x = y. Moreover, for all (x0 , y0 ) X t the sequence (xm , ym )mN dened by bt t a + xm+1 (t; a, b) = ba ba ym+1 (t; a, b) = ba
b b
G (t, s) f (s, xm (s; a, b) , ym (s; a, b))ds

a
(12)
G (t, s) f (s, xm (s; a, b) , ym (s; a, b)) ds t
(13)
uniformly converges to (x , y ) for all t , a, b [c, d] and d ((x , y ) , (xm , ym ))

2 1 4 L1 (d c) L1 (d c) m1 2 1 4 L1 (d c) + L2 (d c) 1 1 L1 (d c)2 L2 (d c) 4
(14)
2 1 4 L2 (d c) L2 (d c)
d ((x0 , y0 ), (x1 , y1 )), m N .

3 4,
b) In the conditions (S1), (S2), (S3), if L 1 (d c)2 < 1 and L2 (d c) < x (t, , ), y (t, , ) C1 [c, d] for all t [c, d] and the pair tion of the system (11) for xed (x, y) = (x , y).
2 x y a , a
then
is the unique solu-
Proof 4 a) From (S1) we infer that B (X) X. Let (x1 , y1) , (x2 , y2 ) X. We have: |B1 (x1, y1 ) (t; a, b) B1 (x2 , y2 )(t; a, b)|
b
|G (t, s)| (L1 |x1 (s; a, b) x2 (s; a, b)| + L2 |y1 (s; a, b) y2 (s; a, b)|)ds
a
(d c)2 (L1 x1 x2 + L2 y1 y2 ) 4
282
Alexandru Bica, Loredana Galea and Sorin Muresan |B2 (x1, y1 ) (t; a, b) B2 (x2 , y2 )(t; a, b)|
b
G (t, s) | f (s, x1 (s; a, b) , y1 (s; a, b)) f (s, x2 (s; a, b), y2 (s; a, b))|ds t (d c) (L1 x1 x2 + L2 y1 y2 ).
Consequently, d (B (x1 , y1 ), B (x2, y2 )) where Q=

(dc)2 4 L1 (d c) L1 (dc)2 4 L2 (d c) L2
Q d ((x1 , y1 ), (x2, y2 )),
The eigenvalues of Q are 1 = 0 and 2 = 1 (d c)2 L1 + (d c)L2 . From hypothe4 sis, follows 2 < 1 and therefore Q m 0, as m . Then B is Picard operator. From the Perovs xed point theorem follows that the operator B have an unique xed point (x , y) X and the sequence of successive approximations given in (12) and (13) uniformly converges to (x , y) for all t , a, b [c, d]. Moreover, we have the inequality (14). On the other hand, B (x , y ) = (x , y) means, t a bt + x (t; a, b) = ba ba
b
G (t, s) ( f (s, x (s; a, b), y (s; a, b))) ds

a
(15)
y (t; a, b) =
ba
G (t, s) f (s, x (s; a, b) , y (s; a, b)) ds t
(16)
and from condition (S1) we have that x , y C1 [c, d]3 . If we derive the equality (15) by t, from (16) follows
2
Elementary calculus lead to tx (t; a, b) = f t, x (t; a, b) , x (t; a, b) and 2 t (a; a, b) = , x (b; a, b) = . x Then x is the unique solution of (4). b) From (S1) and (S3) we have C (X X) X. For (x, y) X arbitrary, we consider (u1 , v1 ), (u2, v2) X and have:
x t
= y and x (, a, b) C2 ([c, d]).
|C1 ((x, y) , (u1 , v1)) (t; a, b) C1 ((x, y) , (u2 , v2))(t; a, b)|

b
|G (t, s; a, b)| (
a
f (s, x (s; a, b) , y (s; a, b)) |u1 (s; a, b) u2 (s; a, b)| + x
f (s, x (s; a, b) , y (s; a, b)) |v1 (s; a, b) v2 (s; a, b)|)ds y (d c)2 (L1 u1 u2 + L2 v1 v2 ) 4
Application of the Picard Operators to Second Order ODEs because from (S2) we infer that: f (s, x (s; a, b), y (s; a, b)) x and f (s, x (s; a, b) , y (s; a, b)) y Analogous we obtain: |C2 ((x, y) , (u1 , v1)) (t; a, b) C2 ((x, y) , (u2 , v2))(t; a, b)| (d c)(L1 u1 u2 + L2 v1 v2 ) . Then, d (C ((x, y), (u1 , v1 )) ,C ((x, y) , (u2, v2 ))) Q d ((u1, v1 ), (u2, v2 )) . L1
283
L2 ,
s, a, b [c, d].
We have Qm 0 as m and therefore for xed (x, y) = (x , y ), the operator C ((x , y ) , ) have an unique xed point (u, v). Then, A ((x , y ), (u , v)) = ((x , y ) , (u, v )) and so, A is Picard operator. If we choose x0 C2 [c, d]3 , y0 = x0 , u0 = x0 , v0 = y0 then the sequence given t a a by (An ((x0, y0 ) , (u0, v0 )))n uniformly converges on X X to ((x , y ), (u, v )), that is xn uniformly converges to x , yn = xn uniformly converges to y , un = xn uniformly converges t a to u, vn = yn uniformly converges to v . a Then u = x , v = y and the pair x , y is the unique xed point of the system a a a a (11). An analogous study can be realized for the dependence of the solution of the problem (4) by b.

Corollary 5 In the conditions of the above theorem, the solution of the problem (4) and his derivative are smooth dependent by a and b.
4.
Application
We will apply the above result to study the dependence of the glycaemia prole by the moment of the a jeun blood-glucose measurement. In this sense, the system modeling blood-glucose homeostasis proposed and studied in [3], x = a xy bx my x + G0 , a, b, c, d, m, G0 > 0 (17) y = cx dy
284
Alexandru Bica, Loredana Galea and Sorin Muresan
is equivalent with the second order ODE: x = (b + d) x x2 x (bx + aG0 ) x + G0 x + G0 t [t0, T ] . (18)
[(a + m) c + bd] x2 + G0 (ab + bd + mc) x , x + G0
We investigate here the dependence of the solution of equations (18) and of his derivative by the initial moment t0 . So, we consider an interval [, ] such that t0 < T . For the equation (18) the conditions (S1), (S2), (S3) are fullled with: L1 = and (b a) G0 1 + + 7 (mc + bd) + 4ab + 3ac 25 250 L2 =
aG0 2 9 + b+d + 5 5 50 in the realistic hypothesis x [40, 40], G (t) [50, 130] (in case of the presence of the blood-glucose homeostasis phenomenon, considering G0 (80, 100)). Here G (t) represent the blood-glucose level at the moment t. We remember that in the system (17) x represent the deviation of the blood-glucose level from the homeostasis equilibrium G0 and x represent the speed of modication of the blood-glucose levels. Applying the results from the above section we obtain: Corollary 6 If 2 9 aG0 3 + b+d + ( ) < 5 5 50 4 and (b a) G0 1 + + 7 (mc + bd) + 4ab + 3ac ( )2 < 1 25 250 (19)
(20)
then the deviation of the blood-glucose levels from the homeostasis equilibrium and the speed of modication of the blood-glucose levels are smooth dependent by the moment of the a jeun glycaemia measurement. Proof 7 Follows from Theorem 3. Remark 2 In the conditions of the above Corollary, small deviations in the rst moment of measurement t 0 lead to not signicant difference in the measured blood-glucose levels. Consequently, the choice of the a jeun measurement moment is not so strictly for the patients for which the conditions (19) and (20) holds. Generally, for healthy persons the conditions (19) and (20) holds, since < 1 (for the glycaemia prole the unity of time is the day and for instance , can be = 6AM and = 4AM in the next day) and the positive coefcients a, b, c, d, m have usually small values. Therefore, the physician have the possibility to choose the moment t 0 how he want between the hours 6:30 and 8 AM (according to the particularities of the patients and considering the diet paradigms).
Application of the Picard Operators to Second Order ODEs
285
References
[1] P.B.Bailey, L. F. Shampine, P.E. Waltman, Nonlinear Two Point Boundary Value Problems , Acad. Press, New York, 1968. [2] S.R. Bernfeld and V. Laksmikantham, Introduction to nonlinear boundary value problems , Acad. Press, New York, 1974. [3] A.Bica, Lyapunov function for a bidimensional system-model in the glucose homeostasy and clinical interpretations, Int.J.of Evolution Equations, Vol.1, no.1, 69-79, 2005. [4] G.Dezso, Continuous dependence and derivability with respect of the parameters of the xed point, Bull.Appl.and Computer Math.,Tech. Univ. Budapest , 84,pp.245254,1998. [5] G.Dezso, Fixed point theorems in generalized metric space, PUMA Pure Math. Appl. 11, No.2, 183-186, 2000. [6] V.Muresan, Some application of the ber contraction theorem, Studia Univ. BabesBolyai, 45(4), pp.87-96, 2000. [7] A.I.Perov and A.V.Kibenko, On a general method to study the boundary value problems, Iz.Akad. Nauk.,30, pp249-264, 1966. [8] I.A.Rus, Weakly Picard mappings, Comment. Math. Univ. Caroline, 34 (3), pp.769773, 1993. [9] I.A.Rus, A ber generalized contractions theorem and applications, Mathematica, 41(1), pp.85-90, 1999. [10] I.A.Rus, Fiber generalized operators on generalized metric spaces and an application, Scripta Sci. Math., 1(2), pp.355-363, 1999. [11] I.A.Rus, Picard operators and applications, , Sci. Math.Japon., 58(1), pp.191219,2003. [12] J.Sotomayor, Smooth dependence solutions of differential equation on initial data: a simple proof, Bol. Soc. Brasil., 4 (1), pp.55-59, 1973. [13] A. Tamasan, Differentiability with respect to lag for nonlinear pantograph equations, PUMA, 9, pp.215-220, 1998.
D OUBLY N ONLINEAR D EGENERATE PARABOLIC E QUATIONS ON C ARNOT G ROUPS

Junqiang Han and Pengcheng Niu Department of Applied Mathematics, Northwestern Polytechnical University, Xian, Shaanxi 710072, Peoples Republic of China
Abstract In this paper we use variational methods to study the nonexistence of positive solutions for the following doubly nonlinear degenerate parabolic equations on Carnot groups: u = G um1 |G u| p2 G u +Vum+p2 in (0, T), t u(x, 0) = u0 (x) 0 in , u(x,t) = 0 on (0, T). Here is a bounded domain with smooth boundary in Carnot group G, G is the horizontal gradient on G, T > 0, V L1 (), u0 is not identically zero, m R, loc 1 0.
Key Words: doubly nonlinear degenerate parabolic equations, nonexistence, Carnot group. AMS Subject Classication: 35K55, 35K65
1.
Introduction
We study the following doubly nonlinear degenerate parabolic equations: u = G um1 |G u| p2 G u +Vum+p2 in (0, T ), t u(x, 0) = u0(x) 0 u(x,t) = 0 on
in , (0, T ).
(1.1)
Here is a bounded domain with smooth boundary in Carnot group G, G is the horizontal gradient on G, T > 0, V L1 (), u0 is not identically zero, m R, 1 0. It is well known that Euclidean space R n with its usual Abelian group structure is a trivial Carnot group and that the Heisenberg group H n is the simplest non-commutative
The
project supported by National Natural Science Foundation of China, Grant No. 10371099. address: southhan@163.com E-mail address: pengchengniu@yahoo.com.cn
E-mail
288
Junqiang Han and Pengcheng Niu
Carnot group. The problem (1.1) on R n and H n has been studied by Kombe [11]. Goldstein and Kombe [6], [7], Kombe [12] investigated the problem (1.1) on R n , H n and G when m = 1, respectively. The equation on R n u = div um1 |u|p2 u (1.2) t (m R) is classied as doubly nonlinear (see [14]) or with implicit nonlinearity(see [10]). When p = 2, (1.2) becomes 1 u = um . t m (1.3)
This is the linear heat equation if m = 1, the porous medium equation if m > 1, the fast diffusion equation if 0 < m < 1 and the very fast diffusion equation (super-diffusion) if m 0, respectively. The case m = 1 in (1.2) corresponds to the p-Laplace heat equation u = div |u|p2 u . t To perturb (1.2) by a suitable potential function V , we have u = div um1 |u| p2 u +Vum+p2 . t (1.5) (1.4)
Indeed, it is a good combination of the porous medium equation, the fast diffusion equation, the very fast diffusion equation and the p-Laplace heat equation with a singular lower order term. In this paper we are concerned with the nonexistence of positive solutions of the problem (1.1) on general Carnot groups G and our results recover the previous results in [6], |G N| p [7], [11], [12]. In particular, we consider the singular potential V (x) = N p ( > 0, N is a homogeneous norm in Carnot group) which appears in Hardys inequality on polarizable Carnot groups (see [2]). It is well known that singularities of this type belong to a border line case where both the strong maximum principle and Gaussian bounds in [1], [13] fail. Such V also lies outside the Kato class potentials which have been studied extensively. |G N| p |G N| p 1 We also consider the highly singular, oscillating potential V (x) = N p + N p sin N ( > 0, R\{0} and > 0). We will show that nonexistence of positive solutions is intimately related to Hardys inequality. The plan of this paper is as follows. In Section 2. we introduce some notation and basic facts about Carnot groups. In Section 3. we study problem (1.1).
2.
Notation and Preliminary Result
In this section we recall some notation and basic results about Carnot groups. A Carnot group is a simply connected Lie group, G, with a Lie algebra g, which admits a stratication, g = V1 Vr satisfying [V1,V j ] = V j+1 for j = 1, . . ., r 1 and [V1,Vr] = {0}. The homogeneous dimension of G is Q = j (dimV j ).
j=1 r
Doubly Nonlinear Degenerate Parabolic Equations on Carnot Groups
289
We assume that dimV1 = m 2 and that X = {X1 , X2, . . ., Xm}is an orthonormal basis of V1. The horizontal gradient on G is G = (X1 , , Xm) . The related doubly nonlinear degenerate operator on G is G,m,p = G um1 |G u| p2 G u . When m = 1, it becomes the p-sub-Laplacian on G: G,p u = G |G u|p2 G u ; When m = 1 and p = 2, it becomes the sub-Laplacian on G: G = (2.2) (2.1)
X 2. j
j=1
(2.3)
Folland [3] proved that in any Carnot group G, there exists a homogeneous norm N smooth on G\{0} such that = N 2Q is a fundamental solution of G at 0. The following embedding result due to Folland and Stein [3], [4] holds on G: Theorem 2.1. Let G be an open set. For any 1 0 such that for C0 (), || dx
p
1 p
Sp
|G | dx
p
1 p
(2.4)
where p =
pQ Qp
is the Sobolev exponent relative to p.
It plays a crucial role in this paper. 1 We say that a Carnot group G is polarizable if the homogeneous norm N = u 2Q associated to Follands solution u for the sub-Laplacian G is -harmonic in G\{0}. Recently, DAmbrosio [2] has proved the following L p -Hardys inequality on polarizable Carnot groups. Theorem 2.2. Let 1 < p < Q and G an open set. Assume that G is polarizable. Then we have for C0 (), Q p p
p
|G N| p || dx Np
p
|G |p dx.
(2.5)
It is an open problem whether (2.5) still holds for general Carnot groups. The Hardy inequality on polarizable Carnot groups says that the nonlinear operator L p u =
G p u N p u p1 is a positive operator if Qp . If > Qp then the operap p tor L p is unbounded from below. From this point of view it is natural to consider a sample p G potential such as V (x) = |NN| . p For further discussion on Carnot groups, see [3], [5], [15, Chapter XIII, Section 5], [9], [8, Section 11.3] and the reference therein.
| N| p
290
3.
Nonexistence Results
Throughout this section, is a bounded domain with smooth boundary in Carnot group G, 1 0. We dene the positive solutions in the following sense. Denition 3.1. By a positive local solution continuous off of K, we mean (i) K is a closed Lebesgue null subset of ; (ii) u : [0, T ) L1 () is continuous for some T > 0; (iii) (x,t) u(x,t) C ((\K) (0, T )); (iv) u(x,t) > 0 on (\K) (0, T ); (v) lim u(,t) = u0 in the sense of distributions; (vi) G u Lloc () and u is a solution in the sense of distributions of the PDE.
p t0
Remark 3.2. If 0 < a 0 for (x,t) K0 [a, b] for some 1 > 0. We can weaken (iii), (iv) to be (iii) u(x,t) is positive and locally bounded on (\K) (0, T ); 1 (iv) u(x,t) is locally bounded on (\K) (0, T ). If a solution satises (i), (ii), (iii), (iv), (v) and (vi) then we call it a general positive local solution off of K. This is more general than a positive local solution continuous off / of K. If K = 0, we simply call u general positive local solution. Lemma 3.3. Let be a bounded domain in G and 1 0 there exists a positive constant C () such that
M ||p dx
|G | p dx +C()
|| p dx.
(3.1)
The proof is similar to them in [6] and [7].

p Theorem 3.4. Let m R, 1 0, then the problem (1.1) has no general positive local solution off of K. Proof. We argue by contradiction. Given any T > 0, let u : [0, T ) L1 () be a general positive local solution off of K to (1.1) in (\K) (0, T ) with u0 0 but not identically || p zero. Multiply both sides of (1.1) by the test function um+p2 where C0 (\K), and integrate over , to get 1 d 3 m p dt =
u3mp || p dx
G um1 |G u| p2 G u
|| p um+p2
dx (3.3)
V || p dx.
Doubly Nonlinear Degenerate Parabolic Equations on Carnot Groups Integration by parts gives L = G um1 |G u| p2 G u
291
um1 |G u| p2 G u G
dx um+p2 || p dx. um+p2
||p
(3.4)
A direct computation shows that G = u || p um+p2 1 p || p2 G um+p2 ||p (m + p 2)um+p3 G u 2(m+p2) ||p2 ||p G (m + p 2) m+p1 G u. um+p2 u
= p Therefore
L = p
u1p |G u| p2 || p2 G u G dx
+(m + p 2) (m + p 2) p

up |G u|p || p dx
up |G u| p || p dx (3.5)
u1p |G u| p1 || p1 |G | dx.
Here we can use the following elementary inequality: Let p > 1 and w1 = w2 be two positive different real numbers. Then (p 1)w2 pw2
1 1
p1
w1 > w1 .
(3.6)
Choosing w1 = p |G | and w2 = p(p1) u1 |G u| ||, we have (m + p 2) > 1

up |G u|p ||p dx p
u1p |G u| p1 ||p1 |G |dx (3.7)
|G | p dx.
Substituting (3.7) into (3.5) yields L= G um1 |G u| p2 G u ||p um+p2 dx > 1 |G | p dx. (3.8)
Merging (3.8) into (3.3) and integrating from t1 to t2 , with 0 < t1 < t2 < T , we obtain
V || p dx 1
|G | p dx
<
1 (3 m p)(t2 t1 )
u (x,t2 )3mp u (x,t1)3mp || p dx.
(3.9)
292

(3mp)Q
We know that f (y) = y p is concave when 3 inequality for concave functions, we obtain u (x,ti )
(3mp)Q p
p Q
m + p. Applying Jensens
(3mp)Q p
dx ||
u (x,ti) dx
||
(3mp)Q p
= C (||)
u (x,ti) dx
< .
Here we use the fact that is bounded, whence || is nite. Therefore u (x,ti )3mp L p (),
Q
i = 1, 2.
By Lemma 3.3, we have for every > 0, there exists a positive constant C() such that 1 (3 m p)(t2 t1 )

u (x,t2)3mp u (x,t1 )3mp ||p dx || p dx. (3.10)
|G | dx +C()
p
Replacing (3.9) by (3.10), we get
V || p dx 1
|G | p dx
|G | p dx +C()
|| p dx.
Therefore
0=C0 (\K)
inf
|G |
dx
||
(1 1 )V p
||p dx
dx
(1 1 )C(1) > .
(3.11)
This contradicts our assumption (3.2). This completes the proof of Theorem 3.4. We now focus our attention on some singular potentials. First we treat the following positive singular potential: |G N|p (3.12) V (x) = Np where > 0. It appears in Hardys inequality on polarizable Carnot groups (see [2]). Theorem 3.4 gives the following corollary. Corollary 3.5. Let 0 , m R, 1 1
Qp p p m+p2 p1 p1
and V (x) be dened by
p (3.12). Then the problem (1.1) has no general positive local solution off of K if 3 Q
We consider the following highly singular, oscillating potential for the next corollary: V (x) = |G N| p |G N| p 1 + sin , p p N N N (3.13)
where > 0, R\{0} and > 0. The oscillating part has very large positive and negative oscillating parts. In particular, it oscillates widely, but important cancellations occur between the positive and the negative parts in the quadratic form. Therefore, nonexistence of positive solutions only depends on the size of .
Doubly Nonlinear Degenerate Parabolic Equations on Carnot Groups Corollary 3.6. Let 0 , m R, 1 1
Qp p p m+p2 p1 p1
293
and V (x) be dened by
p (3.13). Then the problem (1.1) has no general positive local solution off of K if 3 Q
To prove Corollary 3.5 and Corollary 3.6, we need to construct a sequence {n} of test functions which satises
|G n | p
dx
(1 1 )V |n| p dx
|n |p dx
as n .
This can be done as in [6]. In the sequel, we study the problem (1.1) through another approach. We transform (1.1) m+p2 to the following problem via change of variable v = u p1 : p1 p1 p1 v m+p2 = t G,p v +V vp1 in (0, T ), m+p2 (3.14) v(x, 0) = v0 (x) 0 in , v(x,t) = 0 on (0, T ),
m+p2
where v0 (x) = u0 (x) p1 . The main point applying this method is that we obtain a different lower bound for m + p. Theorem 3.7. Let m R, 1 0, then the problem (1.1) has no general positive local solution off of K. Proof. The proof is by contradiction. Given any T > 0, let v : [0, T ) L1 () be a general positive local solution off of K to (3.14) in (\K) (0, T) with v0 0 but not identically zero. || p Multiply both sides of (3.14) by the test function v p1 where C0 (\K), and integrate over , to get d 1 3 m p dt v
(p1)(3mp) m+p2
|| p dx = 1
G,p v
|| p v p1
|| p v p1
dx +
V || p dx.
(3.16)
Applying integration by parts for to the proof of Theorem 3.4, we have G,p v
G,p v
dx and using (3.6), which is similar
|| p v p1
dx >
|G | p dx.
(3.17)
294
Substituting (3.17) into (3.16) and integrating from t1 to t2, where 0 < t1 < t2 < T , we obtain 1 (3 m p)(t2 t1 ) > 1

v(x,t2)
(p1)(3mp) m+p2
v(x,t1 )
(p1)(3mp) m+p2
|| p dx (3.18)
|G |p dx +
V || p dx.
p p+Q(p1)
(p1)(3mp)Q
We know that f (y) = y (m+p2)p is concave when 3 Jensens inequality for concave functions, we obtain v (x,ti)
(p1)(3mp)Q (m+p2)p
m + p. Using
dx ||
v (x,ti )dx
(p1)(3mp)Q (m+p2)
|| < .
(p1)(3mp)Q (m+p2)p
= C (||)
v (x,ti ) dx
Here we use the fact that is bounded, whence || is nite. Therefore v (x,ti)
(p1)(3mp) m+p2
L p (),
i = 1, 2.
By Lemma 3.3, we have for every > 0, there exists a positive constant C() such that 1 (3 m p)(t2 t1 )

v(x,t2)
(p1)(3mp) m+p2
v(x,t1 )
(p1)(3mp) m+p2
|| p dx (3.19)
|G | dx +C()
p
||p dx.
Merging (3.19) into (3.18) gives
V || p dx 1
|G | p dx <
|G | p dx +C()
|| p dx,
and so for 1 as in the hypothesis, inf

|G | p
dx
0=C0 (\K)
||
(1 1 )V p
||p dx
dx
(1 1 )C(1) > ,
(3.20)
which contradicts the assumption (3.15). The proof of Theorem 3.7 is now complete. Corollary 3.8. Let 0 , m R, 1 1 Qp . p Corollary 3.9. Let 0 , m R, 1 1 Qp . p
p1 p1
Doubly Nonlinear Degenerate Parabolic Equations on Carnot Groups
295
Remark 3.10. The methods we applied above gave us two different lower bound for m + p. The better lower bound for m + p in each method depends on p. p 2Q (i) If 1 Q+1 then 3 p+Q(p1) m + p is the better lower bound; (iii) If p =
2Q Q+1
then both methods give the same lower bound for m + p.
References
[1] D. G. Aronson, Non-negative solutions of linear parabolic equations. Ann. Scuola Norm. Sup. Pisa, Vol. 22 (1968), 607694. [2] L. DAmbrosio, Hardy type inequalities related to second order degenerate differential operators, to appear. [3] G. B. Folland, Subelliptic estimates and function spaces on nilpotent Lie groups, Ark. Math.. Vol. 13 (1975), 161207. [4] G. B. Folland, E. M. Stein, Estimates for the b complex and analysis on the Heisenberg group, Comm. Pure Appl. Math.. Vol. 27 (1974), 429522. [5] G. B. Folland, E. M. Stein, Hardy spaces on homogeneous groups, volume 28 of Mathematical Notes, Princeton University Press, Princeton, New Jersey, 1982. [6] J. A. Goldstein, I. Kombe, Nonlinear degenerate parabolic equations with singular lower order term, Advances in Differential Equations . Vol. 8 (2003), no. 10, 1153 1192. [7] J. A. Goldstein, I. Kombe, Nonlinear degenerate parabolic equations on the Heisenberg group, International Journal of Evolution Equations . Vol. 1 (2005), 122. [8] P. Hajlasz, P. Koskela, Sobolev met Poincar , Memoirs Amer. Math. Soc.. Vol. 145 e (2000), 688. [9] J. Heinonen, Calculus on Carnot groups, In Fall School in Analysis (Jyv skyl , 1994), a a Ber. Univ. Jyv skyl Math. Inst., Jyv skyl , Vol. 68 (1995), 131. a a a a [10] A. S. Kalashnikov, Some problems of the qualitative theory of second-order nonlinear degenerate parabolic equations, Russian Math. Surveys. Vol. 42 (1987), 169222. [11] I. Kombe, Doubly nonlinear parabolic equations with singular lower order term, Nonlinear Analysis. Vol. 56 (2004), 185199. [12] I. Kombe, The Hardy inequality and nonlinear parabolic equations on Carnot groups, to appear. [13] P. Li, S. T. Yau, On the parabolic kernel of the Schr dinger operator, Acta Math. Vol. o 156 (1986), 153201.
296
[14] J. L. Lions, Quelque methodes de resolution des problemes aux limites nonlineaire , Springer, Berlin. [15] E. M. Stein, Harmonic analysis: real-variable methods, orthogonality, and oscillatory integrals, Princeton University Press, Princeton, NJ, 1993. With the assistance of Timothy S. Murphy, Monographs in Harmonic Analysis, III.
S OME S CALAR C ONSERVATION L AWS WITH D ISCONTINUOUS F LUX

Julien Jimenez Universit de Pau et des Pays de lAdour e Laboratoire de Math matiques Appliqu es - UMR 5142 CNRS e e BP 1155 - 64013 - PAU Cedex - FRANCE
Abstract We deal with the scalar conservation law in a one dimensional bounded domain : t u + x (k(x)g(u)) = 0, associated with a bounded initial value u0 . The function k is supposed to be bounded, discontinuous at {x0 = 0}, and with bounded variation. A weak entropy formulation for the Cauchy problem has been introduced by J.D Towers in [12]. In [11] the existence and the uniqueness is proved by N. Seguin and J. Vovelle through a regularization of the function k. We generalize the denition of J.D Towers and we adapt the method developed in [11] to establish an existence and uniqueness property in the case of the homogeneous Dirichlet boundary conditions.
1.
Introduction
We are interested in the existence and uniqueness properties for a scalar conservation law made of an hyperbolic rst-order quasilinear equation set in a one-dimensional bounded domain , and for any positive nite real T , that can be formally described: Find a bounded measurable function u on Q =]0, T [ such that u + (k(x)g(u)) = 0 in Q =]0, T [, t x u(0, x) = u0(x) on , u= 0 on (a part of) ]0, T [,
(1)
where k is a discontinuous function at a point x0 of . Such an equation arises in the modelling of continuous sedimentation of solid particles in a liquid ([3]) or when one considers a two-phase ow in an heterogeneous porous medium without capillarity effects ([5], [4]). By normalization, we suppose that =]1, 1[. The initial condition u0 belongs to L () and takes values in [m, M] where m and M are two xed reals, m < M. The function k is discontinuous at x0 = 0 and k|[1,0[ is an element of W 1,+ (]1, 0[) while k|]0,1] belongs to W 1,+ (]0, 1[).
298
Julien Jimenez
Thus, thanks to a Cauchy criterion, we can dene: kL = lim k(x)

x0
and
kR = lim+ k(x).
x0
The ux function g is Lipschitzian on R. We suppose also that: g changes no more than once its monotony (2)
Eventually, we suppose that equation (1) is non degenerate in the sense of Panov (see [9], [10]), that is to say: g is not linear on nondegenerate intervals (3) and
L {x , k(x) = 0} = 0.
where L denotes the Lebesgue measure.
(4)
The mathematical formulation for (1) is given in Section 2. through an entropy inequality on the whole Q, using the classical Kruzkov entropy pairs (see [7]) and involving a term that takes into account the jump of k along {x0 = 0}. As soon as we are able to transcript in Section 3. the transmission conditions along the interface included in Denition 1, we are able to state, in Section 4., the uniqueness. To do so strong traces for u along the interface {x0 = 0} will be needed. Finally Section 5. is devoted to the existence property for (1) through a suitable regularization of the function k.
2.
Denition of an Entropy Solution
We propose a denition extending that of J.D. Towers ([12]) - also used by N. Seguin and J. Vovelle ([11]) or F. Bachmann ([1]) - to the case where k depends on the space variable and for the homogeneous Dirichlet problem in a bounded interval of R. So we say that: Denition 1. A function u of L (Q) is an entropy solution to problem 1 if: (i) R , Cc([0, T [) , 0 , (|u(t, x) |t (t, x) + k(x)(u, )x(t, x))dxdt Q k (x)sgn(u )g()dxdt Q T + |u0 |(0, x)dx + |(kL kR )g()| (t, 0)dt 0 ,
0
(5)
where (u, ) = sgn(u )(g(u) g()), (ii) for a.e. t in ]0, T [, for any real , k(1)(sgn(u (t) ) + sgn())(g(u (t)) g()) 1 1 k(1)(sgn(u(t) ) + sgn())(g(u(t)) g()) 0, 1 1 0. (6) (7)
Some Scalar Conservation Laws with Discontinuous Flux
299
In this denition u and u denote the traces of u respectively in (+1) and (1)+ in 1 1 the sense of A. Vasseur [13] (see also Y. Panov [10]). Indeed it follows from [13], Lemma 1. Let u be an entropy solution to (1). If for each (, ) = (0, 0), for a.e. x [1, 1], L ({ | + .k(x)g () = 0}) = 0, there exists two functions u in L (]0, T [) such as, for 1 every compact set K of ]0, T [, esslim
x1 K
|u(t, x) u (t)|dt = 0. 1
(8)
In [10], E. Panov proved the existence of these strong traces with a continuous ux function, when the boundary is not a characteristic hypersurface. The latter condition is satised here under (3) and (4), when we consider the problem (1) separately on ] 1, 0[ and on ]0, 1[. Remark 1. Of course, the statement of Lemma 1 also ensures the existence of strong traces for u, u+ and u in L (]0, T[) along {x0 = 0} respectively at right and at left.
3.
Conditions at the Interface {x0 = 0}
Let us establish that the previous denition ensures the uniqueness. The proof is based on that proposed in [11] and relies essentially on the transmission conditions along {x0 = 0} underlying to entropy inequality (5). Indeed the existence of strong traces for u permits us to state rst: Lemma 2. Let u in L (Q) be an entropy solution to (1). So, for a.e. t in ]0, T [, for all real , kL (u(t), ) kR (u+(t), ) + |(kL kR )g()| 0 . (9)
Proof. Let be a nonnegative element of Cc(Q). We refer to the cut-off function on R, , for > 0, introduced in [11]: if 2 < |x| , 0 |x| + 2 (x) = if |x| 2 , 1 if |x| < . such that (x) 0 if x = 0, and (0) = 1 for all . Thanks to a density argument we may choose as test-function in (5). We pass to the limit when goes to 0+ by using the Lebesgue dominated convergence Theorem providing that all the terms tend to 0 except |kL kR |g() 0T (t, 0)dt (which does not depend on ) and: I = k(x)(u, ) dxdt .
Q
300
Julien Jimenez
By denition of , T 1 T 1 2 I = k(x)(u, ) dxdt + k(x)(u, ) dxdt, 0 2 0 T 1 |k(x)((u, )(t,x) kL(u, )(t, 0))|dtdx, and, by setting L = 0 2 we prove that lim 0+ L = 0 because (., ) is Lipschitzian on [0, 1], and due to the denition of kL and u. As a consequence, we obtain (9). As in [11], a Rankine-Hugoniot condition may be deduced from (9). To do so we need an additional hypothesis on the function g. So we suppose that: 1 R, 1 esssup u, g(1)(kL kR ) 0 2 R, 2 essinf u, g(2)(kL kR ) 0 (10)
Lemma 3. Under (10), for a.e. t in ]0, T [, the following Rankine-Hugoniot condition holds: kL g(u(t)) = kR g(u+(t)) . Proof. We choose = 1 in (9) to obtain: kR g(u+) kL g(u) + g(1 )(kL kR ) + |g(1)(kL kR )| 0 From (10), we deduce that kR g(u+) kL g(u). By choosing = 2 in (9), and using (10), we obtain the reverse inequality. (11)
4.
The Uniqueness Theorem
First we recall that Lemma 4. If a bounded mapping u satises (5), then: esslim +
t0
|u(t, x) u0(x)|dx = 0.
(12)
We are now able to state an uniqueness property for (1) through a T -Lipschitzian dependence in L1 (Q) of a weak entropy solution with respect to corresponding initial data. Theorem 1. Let u and v be two entropy solutions to (1) for initial conditions (u0, v0 ) in (L (]1, 1[)2. Then, under (10):
T 0 1 1
|u(t, x) v(t, x)|dxdt T

1 1
|u0(x) v0 (x)|dx.
(13)
301
Proof. We use the method of doubling variables due to S. N. Kruzkov (see [7]) by reasoning in two steps: we consider rst some test-functions vanishing on a vicinity of {x0 = 0}. That provides a Kruzkov-type inequality between two entropy solutions from which one the former vanishing hypothesis is released by using (11). Lemma 5. Let u and v be two entropy solutions in L (Q) to (1) associated with initial conditions u 0 and v0 in L (]1, 1[). For any nonnegative function in Cc([0, T[), vanishing in a neighborhood of {x0 = 0}, (|u(t, x) v(t, x)|t (t, x) + k(x)(u(t, x), v(t, x))x(t, x))dxdt (14)
|u0 (x) v0 (x)|(0, x)dx 0.
Proof. Let ( j ) jN be a classical sequence of molliers in R, such that j (x) j (x), an element of Cc([0, T [) satisfying the hypotheses of Lemma 5. For j N and (t, x, s, y) Q Q, we set: j (t, x, s, y) = ( t+s , x+y ) j (t s) j (x y). 2 2 To simplify, we denote w = t+s , z = x+y , u = u(t, x), v = v(t, x), v = v(s, y), q = (t, x), 2 2 q = (s, y). By choosing = v in (5) for u (respectively = u in (5) for v) against the test-function j and integrating over Q with respect to q (respectively q), it comes: |u v|t (w, z) j (t s) j (x y)dqd q
QQ
QQ
sgn(u v)(k (x)g(v) k (y)g(u)) jdqd q

t |u0(x) v0 (y)|( 2 , z) j (x y) j (t)dqdy
+2
Q
(15)
+
Q
t (|u u0 | + |v v0 |)( 2 , z) j (x y) j (t)dqdy
+
QQ
(u, v)k(x)(x )(w, z) j (t s) j (x y)dqd q +

QQ
(u, v)(k(y) k(x))y j (q, q) dqd q 0 .
We will just focus on the second and the sixth line. Indeed there is no difculty to pass to the limit when j goes to + in the other lines by referring to the notion of Lebesgue points for an integrable function on Q (and by using (12) for the forth line). Lets study rst the sixth line, denoted I j . Coming back to the denition of j yields I j = I1, j + I2, j , where:
302 I1, j = I2, j =

QQ
Julien Jimenez (u, v)(k(y) k(x))y((w, z)) j(t s) j (x y)dqd q, (u, v)(k(y) k(x))(w, z) j(t s)y ( j (x y))dqd q.
QQ
By using the notion of Lebesgue points, we state that

j+
lim I1, j = 0.
Next we write I2, j = Ia + Ib with: Ia = and, Ib = (u, v)(k(y) k(x))(w, z) j(t s)y ( j (x y))dqd q.
QQ
{(u, v) (u, v)}(k(y) k(x))(w, z) j(t s)y ( j (x y))dqd q
QQ
Let us rst consider Ib . We denote T (q, q) = (u, v)(k(y) k(x))(w, z) j(t s)y ( j (x y)), Q =]0, T []1, 0[ and Q+ =]0, T []0, 1[,
Ib,1 =
Q Q
T (q, q)dqd q , Ib,2 =
T (q, q)dqd q
Q Q+
Ib,3 =
Q+ Q
T (q, q)dqdq and Ib,4 =
T (q, q)dqd q.
Q+ Q+
Then, Ib = Ib,1 + Ib,2 + Ib,3 + Ib,4 , so we just need to study Ib,1 and Ib,2, the arguments for Ib,3 and Ib,4 being similar. We integrate by parts Ib,1 with respect to y to obtain: Ib,1 = 1 2 +
Q 0
Q Q
(u, v)k (y)(w, z) j (t s) j (x y)dqd q (u, v)(k(y) k(x))y(w, z) j(t s) j (x y)dqd q
Q Q T
x (u, v)(kL k(x))(w, ) j (t s) j (x)dqds 2
When j goes to +, the two last terms tend to 0, owing to the continuity of k on ]1, 0[ and to the denition of kL . Moreover, since k|[1,0[ belongs to W 1,+ ([1, 0[) and is continuous, the rst term tends to: (u(t, x), v(t, x))k (x)(t, x)dq.
Q
Some Scalar Conservation Laws with Discontinuous Flux Similarly, lim Ib,4 =
j+
303
(u(t, x), v(t, x))k (x)(t, x)dq.

Q+
By denition of j , Ib,2 is equal to:

T 0 0 1 j 0 T 0
1 j
(u, v)(k(y) k(x))(w, z) j(t s)y ( j (x y))dqd q
As vanishes on a neighborhood of {x0 = 0}, from a certain j0, Ib,2 vanishes and it is the same for Ib,3 . Eventually:
j+
lim Ib =
(u, v)k (x)(t, x)dq
We study now Ia. By using the same decomposition as for Ib , it appears four integrals whose two vanish (because vanishes on a vicinity of {x0 = 0}) and it only leads to consider the term, denoting by Ia,1 : {(u, v) (u, v)}(k(y) k(x))(w, z) j(t s)y ( j (x y))dqd q
Q Q
By using the Lipschitz condition for and k, we highlight a nonnegative constant C1 independent from j, such that: |Ia,1 | C1
Q Q
|v(s, y) v(t, x)||x y| j (t s)|y ( j (x y))|dqd q
This way, due to the denition of j , there exists a nonnegative constant C2 such that: |Ia,1| C2 j2
j+ {|ts| 1 ,|xy| 1 } j j
|v(t, x) v(s, y)|dqd q,
so that, lim Ia,1 = 0, and as a consequence

j+
lim Ia = 0.
To sum up:
j+
lim I j = lim (I1, j + Ia + Ib ) =

j+
(u(t, x), v(t,x))k (x)(t, x)dq.

Q
We study now the j-limit of the second line in (15) that is: Lj = We write L j = L1, j L2, j with L1, j =
QQ
QQ
sgn(u v)(k (x)g(v) k (y)g(u)) jdqd q.
k (x)(u, v) j dqd q and
304 and L2, j =

QQ
Julien Jimenez
g(u)sgn(u v)(k (x) k (y)) j dqd q.
On the one hand, it is clear that: lim L1, j = k (x)(u(t, x), v(t, x))(t, x)dq.
Q
j+
On the other hand, as for the study of Ia and Ib we share L2, j into four terms whose two vanishes ( vanishing on a neighborhood of {x0 = 0}) so that we only consider: L2,a = and, L2,b =
Q Q
g(u(t, x))sgn(u(t, x) v(s, y))(k (x) k (y)) j dqd q
Q+ Q+
g(u(t, x))sgn(u(t, x) v(s, y))(k (x) k (y)) j dqd q.
We observe that: |L2,a| C g where =] 1, 0[. So that, since k belongs to L ([1, 0[), lim L2,a = 0 and it is the same for L2,b .
j+
|k (x) k (y)| j (x y)dxdy,
To summarize, lim L j =
j+
k (x)(u(t, x), v(t, x))(t, x)dq, and (14) follows that comQ
pletes the proof of Lemma 5. Now we state that: Lemma 6. Under (10), the Kruzkov inequality (14) still holds for in
Cc([0, T [), 0.
Proof. Thanks to a density argument we can choose in (14) the test function (1 ) where is dened in the proof of Lemma 2. By taking the -limit, it comes:
Q
(|u v|t + k(x)(u, v)x)dxdt +
|u0 v0 |(0, x)dx J,
with:
T
J=
0
(kL (u, v ) kR (u+, v+))(t, 0)dt.
Inequality (9) shows that J is nonnegative. Indeed let us study, for a.e. t of ]0, T [, the sign of: I = kL (u, v) kR (u+, v+) .
305
We just focus on the case when u+ v+ and u v have an opposite sign. Otherwise due to (11), that is satised because of (10), I = 0. When sgn(u+ v+ ) = sgn(u v ) = 0, by using (11), we have: I = 2kL (u, v ) = 2kR (u+, v+) We suppose that kL kR > 0, v+ < u+ and u < v, the study of the other cases being similar. So, I = 2kL (g(u) g(v )) = 2kR (g(u+) g(v+ )) Here we must consider some different situations. 1. u < v < v+ < u+. Then (9) can be written, for any of [u, u+], kL (g(u) g()) kR (g(u+) g()) + (kL kR )|g()| 0. if g(v) 0, by choosing = v in (16), we have: 2kL (g(u) g(v )) 0, so I 0. if g(v+) 0, by choosing = v+ in (16), we obtain: 2kR (g(u+) g(v+ )) 0, so I 0. if g(v) < 0 and g(v+) > 0, we deduce from (11), as kL kR > 0, that kR < 0 and kL > 0. If we suppose that I = 2kL (g(u) g(v)) < 0, then g(u) > g(v). But we also have I = 2kR (g(u+) g(v+)), that implies that g(u+) < g(v+). Consequently g changes at least twice its monotony that contradicts the assumption (2). 2. u < v+ < v < u+. As in the previous case, if g(v) 0 or g(v+) 0, we have I 0. If g(v) < 0 and g(v+) > 0, there exists in ]v+, v [ such that g() = 0. Choosing = in (9) written successively for u and v yields to 2kR g(u+) 0 and 2kR g(v+) 0. Then I 0. 3. u < v+ < u+ < v. if g(v+) 0 or g(u+) 0, from (9), I 0. if g(v+) < 0 and g(u+) > 0, there exists in ]v+ , u+[, such that g() = 0. By choosing = in (9), written for u and v, we show that I 0. (16)
306
Julien Jimenez All the others cases may be reduced to one of the previous situations.
Now, in order to prove (13), thanks to Lemma 6 we choose in (14) the test-function such that, for any (t, x) in [0, T [, (t, x) = (t)(x), > 0 where Cc([0, T[) and is an element of Cc() such that = 1 on ]1 + , 1[ and || 2 . We obtain:
Q
{|u v| (t)(x) + k(x)(u, v)(t)(x)}dxdt +
|u0 v0 |(0)(x)dx 0.
There is no difculty to pass to the limit when goes to 0 +. We just point out that due to the properties of ( ) and to the denition of v and u , 1 1
T 0
lim+
k(x)(u, v)(t)(x)dq =
{k(1)(u , v ) k(1)(u , v )}(t)dt. 1 1 1 1
It comes |u v| (t)dxdt +
T 0 Q
k(1)(u , v )(t)dt 1 1
T 0
|u0 v0 |(0)dx
k(1)(u , v )(t)dt. 1 1
Lets prove now that:

T 0
k(1)(u , v )(t)dt 1 1
T 0
k(1)(u , v )(t)dt 0 . 1 1
By coming back to Denition 1, we know that, for a.e t in ]0, T [, for all in R: k(1)(sgn(u (t) ) + sgn())(g(u(t)) g()) 0, 1 1 k(1)(sgn(v(t) ) + sgn())(g(v(t)) g()) 0. 1 1
Thus, a.e. on ]0, T [, if u (t) and v (t) have the same sign, we choose = v in the rst inequality (or 1 1 1 = u (t) in the second one) to obtain: 1 k(1)(u (t), v (t)) 0, 1 1 if u (t) and v (t) have an opposite sign, choosing = 0 in the two inequalities gives: 1 1 k(1)(u (t), v (t)) 0. 1 1 Hence, for a.e. t in ]0, T [, k(1)(u (t), v (t))(t) 0. 1 1
Some Scalar Conservation Laws with Discontinuous Flux Similarly, for a.e. t in ]0, T [, k(1)(u , v )(t) 0. 1 1 This way,
Q
307
|u v|t (t)dxdt +
|u0 v0 |(0)dx 0.
The conclusion follows from classical arguments which completes the proof of Theorem 1.
5.
Existence of an Entropy Solution
The proof relies on a suitable regularization k , > 0, of the function k and uses a compactness argument for the sequence (k (u , ))>0, where u is the weak entropy solution to the corresponding mollied problem. We will consider two different situations: R, x , (kL kR )g(x) 0, R+ , x , (kL kR )g(x) 0 g(m) = g(M) = 0. Remark 2. As well (17) as (18) implies (10). (17)
(18)
Remark 3. When we take into account (2) and (3), we observe that (18) includes the class of srictly convex (or strictly concave) functions that vanish at m and M. In addition (17) is fullled as soon as g is strictly monotone and vanishes at a point. In this framework we establish that: Theorem 2. The following assertions hold: (i) Under (17) Problem (1) admits at least one entropy solution u. (ii) Under (18) the problem (1) admits at least one entropy solution u such that, for a.e. (t, x) in Q, m u(t, x) M. We suppose rst that the initial condition u0 is smooth. Then through a Cauchy criterion in L1 (Q) we come back to the situation of u0 in L ().
308
5.1. First Step: u0 Cc ()
Julien Jimenez
We apply the ideas introduced in [11] (also used in [1]) that is to consider a regular approximation of the function k. Let (k ) be a sequence of smooth functions such as, for every positive , k = k out of ] , [ and k is monotone on [, ] (depending on the sign of kL kR ). That implies: x , k(x) k(x) and |k |BV () |k|BV (). Then we denote u the unique entropy solution (see [2]) to the regularized problem: Find a measurable and bounded function u in BV (Q) C ([0, T]; L1()) such that formally u + (k (x)g(u)) = 0 on Q, t x (19) u(0, x) = u0 (x) on , u= 0 on a part of ]0, T [. Lemma 7. (i) Under (17), for any t in ]0, T [, we set: R(t) = ( u0 where Mk = max( k Then |u(t, x)| R(t), a.e. on . (ii) Under (18), m u (t, x) M, for a.e.(t, x) Q. (iii) The sequence (u)>0 is strongly precompact in L 1 (Q). Proof. (i) Let > 0. We introduce the viscous problem associated with (19): Find u, in L2 (0, T ; H 2 ()) C ([0, T]; H 1 ()), t u, L2 (Q) such that, t u, + x (k (x)g(u,)) = xx u, a.e. in Q, u, = u0 on , u(t, 1) = u(t, 1) = 0 for any t ]0, T [.
M kM g t + max(||, ))e
eMk Mg t 1 |g(0)| Mg
L (]1,0[) ,
L (]0,1[) )
and Mg = Lip(g).
(20)
It is known that (20) admits an unique solution that converges toward u in
C ([0, T]; L1()) when tends to 0. We multiply (20) with (u, R(t))+, we integrate
over Qs =]0, s[, s ]0, T [, and perform the following transformations:
Qs
xx u, (u, R(t))+dxdt =
Qs
[x (u, R(t))+]2 ,
2 L2 ()
Qs
t u,(u, R(t))+dtdx =
1 (u,(s, .) R(s))+ 2 +
Qs
R (t)(u, R(t))+dtdx,
Some Scalar Conservation Laws with Discontinuous Flux since u, (0, .) = u0 R(0). Moreover
Qs
309
x (k(x)g(u, ))(u, R(t))+dtdx = +

Qs
Qs
k (x)g(R(t))(u, R(t))+dtdx
x (k (x)(g(u,) g(R(t)))(u, R(t))+dtdx
where we use an integration by parts in the last term on the right-hand side. Then the Young inequality gives: x (k(x)(g(u,) g(R(t)))(u, R(t))+dtdx + Gathering all terms yields to: 1 (u, (s, .) R(s))+ 22 () + [R (t) + k g(R(t))](u, R(t))+dtdx L 2 Qs 2 ( k Mg ) ((u, R(t))+)2 dxdt. 4 Qs Now lets show that the term (t, x) = R (t) + k g(R(t)) is nonnegative. Therefore, thanks to the Gronwalls Lemma the conclusion will follow. On [, ], by denition of R(t), R(t) and R (t) 0. So: if kL kR , k 0, and by (17), g(R(t)) 0. Then (t, x) 0. if kR kL , k 0, and by (17), g(R(t)) 0. Then (t, x) 0. Besides, on ] 1, [[, 1[, k = k. Moreover: R (t) = MgMk R(t) + Mk |g(0)|, and k g(R(t)) Mk Mg R(t) + k g(0). So (t, x) k g(0) + Mk |g(0)| 0. To show that u(t, x) R(t), we multiply (20) with (u, + R(t)) and we use the same techniques as before, especially the rst line in (17). (ii) The proof refers to (i) and basically lies on the fact that g(m) = g(M) = 0. Indeed, the arguments of the proof of (i) still hold for R(t) = M and, in this case, (t, x) = R (t) + k g(R(t)) = 0. Then we apply the same method with R(t) = m. The conclusion follows. (iii) Let 0 > 0. From (i) and (ii), the sequence (u )>0 is bounded in L (Q). So there exists a subsequence, still denoted by (u )>0, such that u converges for the weak- topology in L (Q) to a function u. Moreover, for each < 0 , by denition of k , k g(u) = kg(u) on ]1, 0[]0 , 1[. Then, under (3) and (4) it follows from a result of E. Yu. Panov in [9] that a subsequence (u )<0 converges to u strongly in L1 (]0, T[]1, 0[]0 , 1[). By using a diagonal process we obtain that (u)>0 is strongly precompact in L1 (Q). Remark 4. To prove (iii), one could use the method developed in [11] that lies on the following BV -estimate: |k((u, ))|BV (Q) C(|u0|BV () + |k|BV ()), 1 4 [x (u, R(t))+]2
Qs
Qs
Qs
(k Mg )2((u, R(t))+)2.
310
Julien Jimenez
where C is a positive real. So, (4) and the compactness of the embedding from BV (Q) into L1 (Q) ensure that a subsequence ((u, )) converges to a bounded and BV -function , a.e on Q. Besides we notice that, under (2) and (3), the assumption (17) or (18) on g implies that, for any bounded interval of R, there exists 0 in R such that (., 0) is strictly monotone on this interval. That yields to the convergence of a subsequence (u ) in L1 (Q). Let u be the L1 -limit of a subsequence (u) . Now we have to establish that u is an entropy solution to (1). First we prove that u fullls (5). To this purpose, we introduce the regularized entropy pairs, for any R, and any real : () =

sgn(r )g (r)dr and I () =
sgn(r )dr,
where sgn denotes the Lipschitzian approximation of the function sgn given for any posix tive and any nonnegative real x by sgn(x) = min( , 1) and sgn(x) = sgn (x). By coming back to (20) and considering the test-function v = sgn(u, ), Cc([0, T [), 0, we can take the limit on with classical arguments. So we establish that u fullls the regularized entropy inequality for all in Cc([0, T [),
Q
I (u )t dxdt +
k (x)(u)x dxdt (21)
+
Q
k (x)((u) I (u )g(u))dxdt +
I (u0)(0, x)dx 0.
We want to pass to the limit in (21), rst with respect to and then with respect to . The difculty is only concentrated in the rst term of the second line. That is why we write (with dq = dxdt):
T Q 1
k (x)((u) I (u)g(u))dq
T
k (x)((u) I (u)g(u))dq
T 0 1
(22)
+
0
k (x)((u) I (u)g(u))dq +
k (x)((u) I (u)g(u))dq.
However, owing to the denition of k ,

T 0 1 T 0 1
k (x)((u ) I (u)g(u))dq =
k (x)((u) I (u)g(u))dq, (23)
and
T 1 0 T 1 0
k (x)((u ) I (u)g(u))dq =
k (x)((u ) I (u)g(u))dq.
(24)
In addition, by referring to the denition of and I ,

T 0 T 0 T 0
k (x)((u ) I (u)g(u))dq =
k g()sgn(u )dq
k (
sgn( )g ()d (g(u ) g())sgn(u ))dq.
Some Scalar Conservation Laws with Discontinuous Flux We look for a majoration of the right-hand side of this equality. First,
T 0
311
k g()sgn(u )dq |g()|
T 0
|k |dq.
Now we turn on to the estimate of the term |D(u)| where:

u
D(u ) =
sgn( )g ()d (g(u ) g())sgn(u ).
More precisely, |D(u)| Cg with Cg = 2Mg . Indeed, let us x and . Then a.e. on Q: if u + , + u g ()d + D(u) = g ()d (g(u) g()) + and |D(u)| Mg + |g( + ) g()|, because 0 1. Then we use the Lipschitz condition for g. if u + , u u g ()d (g(u ) g()) D(u) = and |u |2 |D(u)| Mg|u | + Mg 2Mg . if u , g ()d + g(u) g(), g ()d + D(u) = u and |D(u)| Cg , as in the rst case. Eventually we deduce that the term:
T 0
k (
sgn( )g ()d (g(u) g())sgn (u ))dq

T 0
is bounded by: Cg and

T 0
|k |dxdt,

k (x)((u) I (u)g(u))dq (Cg + |g()|)
T 0
|k |dxdt.
(25)
As k is monotone on [, ], for small enough, |k | = sgn(kR kL )k . So,

T 0 T 0
|k |dq = sgn(kR kL )
k dq
Then we integrate by parts to obtain:

T 0 T T
|k |dq = sgn(kR kL ) + sgn(kR kL )
k x dq (26)
(k()(t, ) k()(t, ))dt .
312
Julien Jimenez
Finally, from (21), (22), (23), (24), (25) and (26), for any positive and , we have: sgn(kR kL )(Cg + |g()|)
T T 0
k x dxdt +
k(x) (u)x dxdt
+
0 1 T 1
k (x)((u ) I (u )g(u))dq +
I (u0 )(0, x)dx
+
0
k (x)((u) I (u)g(u))dq +
T 0
I (u )t dxdt
+sgn(kR kL )(Cg + |g()|)
(k()(t, ) k()(t, ))dt 0.
We take now the -limit. Clearly, because (u ) goes to u in L1 (Q) and since I and are Lipschitzian,
0+ Q
lim
(I (u )t + k (x) (u )x )dq =
(I (u)t + k(x) (u)x)dq.
Thanks to the denition of kL and kR and to the continuity of ,

T 0+ 0
lim
(k()(t, ) k()(t, ))dt = (kR kL )
T 0
(t, 0)dt.
Moreover, k x being bounded independently with respect to ,

T 0
lim+
k x dxdt = 0.
So, for any positive , the following inequality holds:

Q
(I (u)t + k(x) (u)x)dxdt +
k (x)((u) I (u)g(u))dxdt
T 0
I (u0)(0, x)dx + (|g()| + Cg)|kR kL |
(t, 0)dt 0.
(27)
We take the limit with respect to through the Lebesgue dominated convergence Theorem, providing that u fullls (5). Lastly, let us establish that u satises (6)-(7). To this purpose, we use the functions H and Q dened in [8] for any , R, by: H (, ) = (dist(, I[0, ]))2 + 2 and Q (, ) =

1 2
1 H (, )g ()d
where I[0, ] denotes the closed interval bounded by 0 and . The sequence (H , Q) converges uniformly to (dist(, I[0, ]), G (,0, )) where:
G (, 0, ) = ((, 0) (, 0) + (, )).
By taking in (20) the test-function 1 H (u, , ), for any function Cc(]0, T[
1 2
313
), and passing to the limit with respect to , we obtain for any positive and the following inequality:
Q Q
H (u , )t dxdt +
k Q (u, )xdxdt
k (x)(Q(u , ) 1 H (u, )g(u))dxdt 0.
If we only consider functions vanishing in a neighborhood of {x0 = 0} containing [, ] (that will not be restictive in the sequel), we can take the -limit without difculty to obtain:
Q Q
H (u, )t dxdt +
kQ (u, )xdxdt (28)
k (x)(Q(u, ) 1 H (u, )g(u))dxdt 0.
Then, for (t, x) ]0, T [, we choose in (28) a sequence of test-functions dened by n (t, x) = (t)n(x) with Cc(]0, T [), 0, and n Cc() such as n 0, n (x)0 on ]1, 1 1 [, n (1) = 1 and n n. On the one hand, by reasoning as in [8] we make n sure that
1 n 1 1 n T 0
lim
n(x)k(x)Q(u, )(t)dtdx exists and is nonnegative.
On the other hand by using the denition of u , 1

1 n 1 1 n T 0
lim
n (x)k(x)Q(u, )(t)dtdx =
T 0
k(1)Q(u (t), )(t)dt. 1
Finally, when goes to 0 +,

T 0
k(1)G (u, 0, )(t)dt 0. 1
To conclude we just emphasize that the previous inequality is equivalent for all in I[0, u ], 1 to: sgn(u )k(1)(g(u) g()) 0, 1 1 that is namely (6) when is reduced to belong to I(0, u ). 1 In the same way, by choosing (t, x) = (t)n (x) in (28), with Cc(]0, T [), 0, and n Cc() such as n 0, n (x) = 0 on ] 1 + 1 , 1[, n (1) = 1 and n n, by using n the denition of u , we establish (7). 1
5.2. Second Step: u0 L ()

We use a mollication process to come back to the rst step. Indeed, for j N, we consider j j j the sequence (u0) j such that u0 belongs to Cc() and (u0) tends to u0 in L1 (). We denote j u j the entropy solution to (1) associated with the initial condition u0 so that, for any j, u j fullls (27) and (28). The comparison result (13) ensures that the sequence (u j ) j is a Cauchy sequence in L1 (Q) and so tends to a limit, denoted u. Then the j-limit in (27) and (28) warrants that u is an entropy solution to (1). To conclude, we point out that (17) or (18) implies (10), so that: Corollary 1. Assume that (17) or (18) holds. Then (1) has an unique entropy solution.
314
Julien Jimenez
6.
Generalisation
In this section we keep the same assumptions on g but we consider that k has a nite number of discontinuities. Let D = {1, ..., n 1}, n = 0, x0 = 1, xn = 1. We suppose that: k is discontinuous at xi , i D, (29)
while k|]xi ,xi+1 [ W 1,+ (]xi, xi+1 [). Of course we need a new denition of an entropy solution which has to be equivalent to Denition 1 when D is reduced to one point. So we say that: Denition 2. Under (29), a function u of L (Q) is an entropy solution to Problem (1) if u satises (6)-(7) and if, R , Cc([0, T[) , 0 , (|u(t, x) |t (t, x) + k(x)(u, )x(t, x))dxdt Q k (x)sgn(u )g()dxdt (30) Q T + |u |(0, x)dx + |(k+ k )g()| (t, xi )dt 0 , 0 i i
iD 0
where ki+ = lim k(x) and ki = lim k(x) +

xxi xxi
We denote u+ and u the strong traces in L (]0, T [) at {x = xi }. By using the same i i techniques as before we can state the following theorem: Theorem 3. Under (18), when k satises (29), there exists a unique entropy solution to (1). Moreover, at every point x i , i D, u satises the Rankine-Hugoniot condition: ki+ g(u+) = ki g(u) i i In addition, we adapt (17) under the form: + + for i, j D, i = j, sgn(ki ki ) = sgn(k j k j ), + R , x , (k1 k1 )g(x) 0, + R+ , x , (k1 k1 )g(x) 0. This condition is satised when g is strictly monotone and the next corollary holds: Corollary 2. Assume that (31) is satised. Then problem (1) has a unique entropy solution, in the sense of Denition 2.
(31)
References
[1] F. Bachmann, Analysis of a scalar conservation law with a ux function with discontinuous coefcients, Advances in Differential equations , 9:11-12, 1317-1338, 2004
315
[2] C. Bardos, A.Y Leroux, J.C Nedelec, First order quasilinear equations with boundary conditions, Comm. in partial differential equations, 4, 1017-1034, 1979. [3] S. Diehl, A conservation law with point source and discontinuous ux function modelling continuous sedimentation, SIAM J. Math. Anal., 56, 388-419, 1996. [4] G. Gagneux, M. Madaune-Tort, Analyse math matique de mod` les nonlin aires de e e e ling nierie p troli` re, Math matiques et Applications , 22, Springer-Verlag, Berlin, e e e e 1996. [5] T. Gimse, N.H. Risebro,Solution of the Cauchy problem for a conservation law with a discontinuous ux function, SIAM J. Math. Anal., 23, 635-648, 1992. [6] E. Godlewski, P.A. Raviart, Hyperbolic systems of conservation laws, Math matiques e et Applications, 3-4, Ellipses, Paris, 1991. [7] S.N. Kruzkov, First-order quasilinear equations with several independent variables, Mat. Sb. 81, 228-255, 1970. zc [8] J. M` lek, J. Ne as, M. Rokyta, M. Ru i ka, Weak and mesure-valued solutions to evoa c lutionary PDEs Applied Mathematics and Mathematical Computation , 13, Chapman & Hall, London, 1996. [9] E. Yu. Panov, Property of strong precompactness for bounded sets of measure valued solutions of a rst-order quasilinear equation, Sbornik: Mathematics, 190:3, 427-446, 1999 [10] E. Yu. Panov, Existence of strong traces for generalized solutions of multidimensional scalar conservation laws, Journal of Hyperbolic Differential Equations , 2:4, 885-908, 2005. [11] N. Seguin, J. Vovelle, Analysis and approximation of a scalar conservation law with a ux fonction with discontinuous coefcients, Math. Models Methods Appl. Sci. 13:2, 221-257, 2003. [12] J.D. Towers, Convergence of a difference scheme for conservation laws with a discontinuous ux, SIAM J. Numer. Anal. 38, 681-698,2000. [13] A. Vasseur,Strong traces for solutions of multidimensional scalar conservation laws, Arch. Rational Mech. Anal. 160, 181-193, 2001.
A N EW E XACT S OLUTION TO THE D ELAYED D IFFUSION E QUATION

P.M. Jordan Code 7181, Naval Research Laboratory, Stennis Space Center, MS 39529, U.S.A.
Abstract A new analytical solution to the one-dimensional delayed diffusion equation is derived in terms of the Lambert W -function. A new critical value of the thermal lag time parameter is also noted. The effects of varying this parameter are examined using numerical methods.
AMS Subject Classication: Primary 80A20 Key Words: Delayed diffusion equation; Fouriers law; thermal relaxation time; Lambert W -function
1.
Introduction
According to Fouriers law, heat conduction in a homogeneous and isotropic solid body is described by the ux relation [1] q = K, (1.1) where q, K(> 0), and = (r, ) are the heat ux vector, the (constant) thermal conductivity, and absolute temperature, respectively, and where r is the position vector and denotes the time. When combined with the conservation of energy law cp + q = 0, (1.2)
where is the mass density and c p is the specic heat at constant pressure, and assuming no heat sources/sinks are present, Eq. (1.1) results in the heat transport equation 2 = 0, (1.3)
where = K/(c p ) is the thermal diffusivity. Since Eq. (1.3), the well-known diffusion equation, is a parabolic PDE, we see that Fouriers law implies that a thermal disturbance at any point in a material body will be felt instantly, but unequally, at all other points of
E-mail address: pjordan@nrlssc.navy.mil. The author thanks Prof. R. E. Mickens for the many enlightening discussions and his helpful suggestions. This work was supported by ONR/NRL funding (PE 061153N)
318
P.M. Jordan
the body; in other word, Fouriers law predicts that thermal signals propagate with innite speed, and thus conicts with the requirements of causality. This drawback of Fouriers law, which Maxwell (by 1871) was clearly aware of, contrary to the claim made in Ref. [2], is most apparent under low temperature and/or high heat-ux conditions [3, 4]. To correct this unrealistic feature, which has come to be known as the paradox of heat conduction, various modications to Eq. (1.1) have been proposed, the best known of these being the MaxwellCattaneo (MC) law [2, 3] 1 + 0 q = K. (1.4)
Here, the thermal relaxation constant 0 (> 0), the value which has been experimentally determined for a number of materials [3, 57], represents the time lag required to establish steady thermal conduction in a volume element once a temperature gradient has been imposed across it. Now the MC law, which has received a great deal of attention in the context of generalized thermoelasticity (see, e.g., [3, 7]), gives rise to the dissipative, hyperbolic heat transport equation 2 + 0 2 2 = 0, (1.5) which is a special case of the telegraph equation known as the damped wave equation, or DWE for short (see, e.g., Refs. [810] and those therein). Thus, according to the MC law, heat conduction occurs via the propagation of damped thermal waves with (nite) propagation speed c0 = /0 , a phenomenon which has come to be known as second sound [2, 3, 11]. As noted by Tzou [12], the expression on the LHS of Eq. (1.4) consists of the rst two terms of the Taylor series expansion of the LHS of the delayed Fourier ux relation q(r, + 0 ) = K(r, ). (1.6)
According to Eq. (1.6), a variation in the temperature gradient at time is felt by the heat ux vector at the later time + 0 . [Here, we note that the 0 0 limiting case of both Eqs. (1.4) and (1.6) is Fouriers law.] Under this ux law, and again assuming that no heat sources/sinks are present, heat transport is governed by the delayed diffusion equation (r,t + 0 ) 2 = 0, (1.7)
which we observe is simply Eq. (1.3) with time delay 0 included in the term on the LHS. The primary aim of this work is to derive a new form of the solution to the onedimensional (1D) case of Eq. (1.7) in the context of a well known initialboundary value problem (IBVP). In the process, a new critical value of the lag time parameter is identied. Additionally, we numerically examine the effects of varying the lag time parameter and compare the maxima of the solutions of Eqs. (1.3) and (1.7) in the 1D case. To this end, the present communication is arranged as follows. In section 2, we formulate the IBVP and give a known exact solution. In section 3, we derive a new form of this solution in terms of the Lambert W -function. In section 4, numerical work is presented and discussed. And nally, in section 5, a summary of our ndings is given.
Delayed Diffusion Equation
319
2.
2.1.
Mathematical Analysis
Problem Formulation
Consider a slender, homogeneous, thermally conducting solid rod of constant crosssectional area and constant thermal diffusivity . The rod occupies the open interval (0, ) along the -axis of a Cartesian coordinate system. Let = (, ) denote the rods absolute temperature, 0 sin[/ ] the rods initial temperature (of course 0 > 0), and for simplicity we assume that the temperature at both ends is maintained at zero. Furthermore, let the lateral face of the rod be fully insulated. Assuming that heat conduction within the rod is governed by Eq. (1.6), the mathematical model of this physical system consists of the following IBVP: (2.1a) (, + 0 ) = , (, ) (0, ) (0, ); (0, ) = 0, ( , ) = 0, > 0; (2.1b) (2.1c) (, ) = 0 sin[/ ], (, ) (0, ) [0, 0].
Employing the following nondimensional quantities: u = /0, x = / , t = (/ 2), (2.2)
we recast our IBVP in dimensionless form as ut (x,t + 0 ) = uxx (x,t), u(0,t) = 0, u(x,t) = sin(x), (x,t) (0, 1) (0, ); t > 0; (2.3a) (2.3b) (2.3c)
u(1,t) = 0,
(x,t) (0, 1) [0 , 0].
Here, 0 = 0 (/ 2) is the (strictly positive) dimensionless thermal lag time.
2.2.
Derivation of a Known Solution
Following [13], we assume u(x,t) = X(x)T (t) and separate variables. Omitting the details, it is easy to show that sin(x), n = 1, (2.4) Xn (x) = 0, n = 2, 3, 4, . . ., and that T must satisfy the delay ODE (see Refs. [1416] and those therein) T (t + 0 ) + 2 T (t) = 0. Here, we note that the initial condition given in Eq. (2.3 c) has reduced to T (t) = 1 for t [0 , 0]. (2.6) (2.5)
Next, we apply the Laplace transform L [ ] to Eq. (2.5). After using Eq. (2.6) and then solving the resulting subsidiary equation, the transform domain solution is found to be [16] 1 2 T (s) = , s s(s + 2es0 ) (2.7)
320
P.M. Jordan
where T (s) L [T (t)] and s is the transform parameter. The inverse of the rst term on the RHS of Eq. (2.7) is, of course, trivial. Obtaining the inverse of the second term, however, is somewhat more difcult. As noted in [16], the easiest way to accomplish this is by rst expanding the second term in Eq. (2.7) in powers of 1/s. Doing so, and then simplifying, yields 1 1 2 T (s) = + s s m=0 s
m+1
exp[s(m0)].
(2.8)
Here, we call attention to the fact that the 1/s factor outside the summation sign was inadvertently omitted in [13, Eq. (18)]. On inverting this series term-by-term using a table of inverses along with the properties of the Laplace transform (see, e.g., Ref. [16]), we obtain, after incorporating the contribution from the rst term on the RHS of Eq. (2.8) and then multiplying by sin(x), the closed-form solution
t/0 +1
u(x,t) =
m=0
(02 )m
[t/0 (m 1)]m m!
sin(x)
(t > 0),
(2.9)
where denotes the greatest integer (or oor) function. Equation (2.9) is the solution given by Su et al. [13].
2.3.
Case of Zero Delay: Diffusion Equation Solution
Finally, we note that the Fourier-based solution, u(x,t) = exp(2t) sin(x) (t > 0), (2.10)
is easily determined by setting 0 0 in IBVP (2.3) and then using separation of variables.
3.
3.1.
Solution in Terms of the Lambert W -function

The Inversion Integral
In this section an alternative form of Eq. (2.9) will be derived. To this end, we observe that by the Laplace inversion theorem [17], T (t) = 1 2i
+i i
est ds s(s + 2es0 )
(t > 0),
(3.1)
where the constant (> 0) is chosen so that the line s = lies to the right of all the singularities of the integrand. Omitting the details, it is not difcult to establish that the integral in Eq. (3.1) can be evaluated using the residue theorem (see, e.g., [17]); in particular, we note that the singularities of the integrand consist only of poles [16]. Determining these poles is, of course, equivalent to nding the roots of the transcendental equation s(s + 2 es0 ) = 0. (3.2)
321
Note that all roots of this equation are simple ones unless 0 = , where (2e)1 , in which case s = 1/ is a double root. Clearly, Eq. (3.2) is satised either when s = 0 or when (3.3) s0 exp(s0) = 2 0 . Let us now introduce the Lambert W -function [15, 18, 19], which is dened to be the solution of the equation W () exp[W ()] = . Using this function, the roots of Eq. (3.3) can be expressed in closed form. Specically, we have s = sr 1Wr (2 0) 0 for 0 = , and s = s1 = s0 1/ and s = sr 2eWr (e1 ) (r = 1, 2, 3, . . .), (3.5) (r = 0, 1, 2, . . .), (3.4)
when 0 = . Here, Wr ( ) denotes the rth (r Z) branch of the W -function [15, 18] with W0( ), also written as simply W ( ) when there is no possibility for confusion, denoting the principal branch. Since the early 1990s, the W -function has appeared in a number of papers from various elds (see, e.g., Refs. [1820] and those therein). It should be noted, however, that W is not a function, strictly speaking, since there are two distinct real values of W () for every (e1 , 0).
3.2.
Exact Time Domain Solution
On calculating and summing the residues of the integrand at each of the poles, the integral in Eq. (3.1) is readily evaluated in terms of an innite series. Consequently, an exact expression for T (t) is obtained and we nd, after simplifying and multiplying by sin (x), that the solution to IBVP (2.3) can also be expressed as u(x,t) = sin(x) where T1 (t) = 20
exp (t/0)W0(2 0) + W0(2 0)[1 +W0(2 0)] m=1
T1 (t), 0 < T2 (t), 0 =
(t > 0),
(3.6)
exp (t/0)Wm(20 ) exp (t/0)Wm(20 ) + Wm(2 0)[1 +Wm(2 0 )] Wm(2 0)[1 +Wm (20 )] and T2 (t) = e1
(3.7)
3t 8 exp(t/ ) 1 + 3 4
exp (t/ )W1(e1 ) W1(e1 )[1 +W1(e1)] . (3.8)
m=2
exp (t/ )Wm(e1 ) exp (t/ )Wm(e1) + Wm(e1 )[1 +Wm(e1)] Wm (e1)[1 +Wm (e1 )]
322
P.M. Jordan
Here, we observe that the restriction 0 (0, ] is necessary since u assumes negative values when 0 > , in contradiction to the fact that u is the absolute temperature. We also observe that Re[Wr (2 0)], where r Z and Re[ ] denotes the real part of a complex quantity, is always negative, meaning that T1,2 0 as t , when 0 (0, ]. And it should be noted that solving Eq. (2.5) in terms of the W -function appears to have been rst discussed in Ref. [15].
3.3.
Analytical Results
W0() 2 , (3.9)
Using the fact that [18] where || e1 is sufciently small, it can be shown that sin(x) lim T1(t) = exp(2t) sin(x);
0 0
(3.10)
i.e., we can recover the Fourier-based solution [Eq. (2.10)] by letting 0 0 in Eq. (3.6). Based on the results given in [14], it is possible to obtain the large- t asymptotic expressions 2 2 0 exp (t/0)W0( 0 ) , < 0 W0(2 0 )[1 +W0(20 )] u(x,t) sin(x) (t ). (3.11) 8 3t 1+ exp(t/ ), 0 = 3e 4 Here, we should mention that the W -function is not explicitly discussed in [14]. Also, for a more detailed treatment of the asymptotic analysis of delay equations, see Ref. [21] and those therein.
4.
Numerical Results
Since the solutions given in Eqs. (2.9) and (2.10) have been numerically compared/contrasted in [13], and Eqs. (2.9) and (3.6) are equivalent, there is no need to reproduce these graphs, which plotted u vs. x for xed values of t, again here. Instead, our numerical work, which was carried out using the software package M ATHEMATICA (Version 5.0), wherein Wr ( ) is implemented as ProductLog[r, ], focuses on understanding the behavior of the difference function (t), which we dene below, as the lag time is varied, including the case of the critical value 0 = = (2 e)1 0.0373. To this end, we have plotted in Fig. 1 the difference function (t) exp(2t) T1 (t), 0 < T2 (t), 0 = (t > 0), (4.1)
which corresponds to subtracting Eq. (3.6) from (2.10) for xed x = 1/2. Here, we see that max[(t)](0), which is a global maximum, is a bounded, increasing function of 0 , as is
Delayed Diffusion Equation the corresponding stationary point tC (0). In addition, it is clear that exp(2t) > T1(t), 0 < T2(t), 0 = (t > 0),
323
(4.2)
which is in agreement with the ndings of [13]. Figure 1 also indicates that, after achieving its maximum, (t) rapidly tends to zero as t , and that the rate of its decay is a decreasing function of 0.
t
0.15
0.1
0.05
t
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Figure 1. (t) vs. t. Bold: 0 = . Thin-solid: 0 = (0.75) . Bold-broken: 0 = (0.5) . Thin-broken: 0 = (0.125) .
5.
Summary
We have presented a new exact solution to the delayed diffusion equation in terms of the Lambert W -function. We have found that = (2e)1 is a critical value of the lag time associated with this new solution and that c < , (5.1)
where c (2)2 0.02533 is the lag time value above which the solution of the corresponding IBVP involving the (1D) DWE can become negative [9, 10]. In addition, we plotted the difference function for several values of 0 . We found that for t > 0, the maximum temperature predicated by Eq. (2.10), the Fourier-based solution, is always greater than that predicted by Eq. (3.6) and that the value of rapidly tends to zero as t . Our numerical studies also indicated that there is a critical value of time, t = tC (0 ), at which this difference is greatest and that both max [](0) and tC (0) are
324
P.M. Jordan
bounded, increasing functions of the lag time 0 , meaning that both max[](0) and tC (0) are at their maximum when 0 = . Finally, in the course of this analysis the following (possibly new) identity was noticed: 1 = Wm ()[1 +Wm()] m=
+
( = 0, e1 ).
(5.2)
References
[1] H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids , 2nd edn., Oxford University Press, New York, 1959. [2] D. D. Joseph and L. Preziosi, Heat waves. Rev. Mod. Phys. 61 (1989), 4173; 62 (1990), 375391. [3] D. S. Chandrasekharaiah, Thermoelasticity with second sound: A review. Appl. Mech. Rev. 39 (1986), 355376. [4] W. Dreyer and H. Struchtrup, Heat pulse experiments revisited. Cont. Mech. Thermodyn. 5 (1993), 350. [5] G. Caviglia, A. Morro, and B. Straughan, Thermoelasticity at cryogenic temperatures. Int. J. Non-Linear Mech. 27 (1992), 251263. [6] P. J. Antaki, Hotter than you think, Machine Design, July 13 (1995), 116118. [7] D. S. Chandrasekharaiah, Hyperbolic thermoelasticity: A review of recent literature. Appl. Mech. Rev. 51 (1998), 705729. [8] A. C. King, D. J. Needham, and N. H. Scott, The effects of weak hyperbolicity on the diffusion of heat. Proc. Roy. Soc. A 454 (1998), 16591679. [9] R. E. Mickens and P. M. Jordan, A positivity-preserving nonstandard nite difference scheme for the damped wave equation. Num. Meth. Partial Diff. Eq. 20 (2004), 639649. [10] R. E. Mickens and P. M. Jordan, A new positivity-preserving nonstandard nite difference scheme for the DWE. Num. Meth. Partial Diff. Eq. 21 (2005), 976985. [11] C. I. Christov and P. M. Jordan, Heat conduction paradox involving second-sound propagation in moving media. Phys. Rev. Lett. 94 (2005), 154301. [12] D. Y. Tzou, Macro- to Microscale Heat Transfer: The Lagging Behavior . Taylor & Francis, Washington, DC, 1997. [13] S. Su et al., Comparison of the solutions of a phase-lagging heat transport equation and damped wave equation. Int. J. Heat Mass Transfer 48 (2005), 22332241. [14] B. K. Drives and R. D. Driver, Simplicity of solutions of x (t) = bx(t 1). J. Math. Anal. Appl. 157 (1991), 591608.
325
[15] D. J. Jeffrey, D. E. G. Hare, and R. M. Corless, Unwinding the branches of the Lambert W function. Math. Scientist 21 (1996), 17. [16] D. Zwillinger, Handbook of Differential Equations , 3rd edn., Academic Press, San Diego, CA, 1997. [17] H. S. Carslaw and J. C. Jaeger, Operational Methods in Applied Mathematics . Dover, New York, 1963. [18] R. M. Corless et al., On the Lambert W function. Advs. Comput. Math. 5 (1996), 329359. [19] B. Hayes, Why W ? Amer. Scientist 93 (2005), 104108. [20] P. M. Jordan, Finite-amplitude acoustic traveling waves in a uid that saturates a porous medium: Acceleration wave formation. Phys. Lett. A 355 (2006), 216221. [21] A. C. Fowler, Asymptotic methods for delay equations. J. Engng. Math. 53 (2005), 271290.
L EVEL S ETS FOR R EACTION D IFFUSION E QUATIONS

Amin Boumenir Department of Mathematics, University of West Georgia, Carrollton, GA 30118 USA
Abstract We are concerned with the shape of the level sets of solution of reaction diffusion equations. Using the maximum principle we nd sufcient conditions for their concavity.
1.
Introduction
Due to the recent advances and applications of level sets in the area of image processing, numerical analysis, and differential equations, [1, 2, 4, 5, 6], we would like to investigate the shape of level sets of classical reaction diffusion equations dened by t u = u + f (u) x Rm t > 0 (1) u(x,t) = 0 x , t > 0 u(x, 0) = (x) x where m 1. The nonlinearity f and initial condition are assumed to satisfy the conditions [A] f C2 , f (u) 0, f (u) > 0, and f (u) < 0 [B] C2 , 0 and f () < < 0. Denote by Q = [0, ) and let Q represent its boundary. We would to study the shape of the level sets of the solution of (1) which are surfaces dened by the points (x, t(x, n)) Q such that u takes the value n, i.e.
:= {(x,t(x, n)) Q, u(x,t(x, n)) = n}.

n
(2)
It is obvious that the geometry of the set {n }nR , which is a subset of Rm would fully describe the solution (x,t, u(x,t)) by level surfaces in Rm+1 . We are interested in nding sufcient conditions for the solution u (.,t) to be, for example, concave down and also for n to be concave up. The level lines provide a phase space type analysis for the solution. For example, they cannot intersect except at a singularity and also play the role of characteristics, as they channel information. Since the gradient u is orthogonal to n , i.e. u n
E-mail
address: boumenir@westga.edu
328
Amin Boumenir
and level sets method are powerful tools for Hamilton-Jacobi equations ut = H (u) . However, in the case of reaction diffusion equations such as (1), the gradient is not available, unless you solve for the solution, which makes the standard computational machinery of the level set method difcult if not impossible to apply. Nevertheless, they still provide interesting geometrical insights for the solution especially when singularities such as blow-up for (1) occur. Condition [A] includes the interesting sub-critical nonlinearity that can be found in [3]. For example blow-up when f (u) = |u| p1 u, occurs for 1< p< for m = 1, 2 and 1 < p < m+2 m2 for m 3.
Thus condition [A] which allows for 1 0 u(a,t) = 0, u(x, 0) = (x) 0 x [a, a]
(3)
Proposition 1. Assume that conditions [A] and [B] hold then the solution u(., t), when it exists, is positive and concave down Proof. First the solution exists locally and is positive. This follows from the maximum 2 principle since t u x2 u = f (u) 0 and on the boundary u 0 when (x,t) Q. Next we prove that u(.,t) is concave down. Since f is smooth, uxx =
2 2 t uxx x2 uxx f (u)uxx = f (u)ux < 0 uxx (a,t) = t u(a,t) f (u) < 0, t > 0 uxx (x, 0) = (x) < 0 x [a, a] 2 u, x2
satises
x [a, a], t > 0 (4)
The maximum principle then implies that uxx < 0 inside Q and thus the solution u(.,t) . is concave down. Proposition 2. Assume that conditions [A] and [B] hold. Then each level lines t(x, n) dened by (2) has one minimum. If n max then min (t(x, n)) = 0 while n > max then min t(x, n) = (n) where
axa axa
axa
max u(x, (n)) = n.
Proof. From the concavity of the solution, it follows that the solution u(.,t), which is positive, has one maximum for each t > 0, say at m(t), u(x,t) < u(m(t),t) for x = m(t)
Level Sets for Reaction Diffusion Equations and clearly ux (x,t) > 0 if a < x < m(t) ux (x,t) < 0 if m(t) > x > a.
329
From (3) it follows that t ut x2 ut f (u)ut = 0 while ut (x, 0) = (x) + f () > 0 for x [a, a] and u(a,t) = 0 for t > 0. Thus ut (x,t) 0 if (x,t) Q, and by the maximum principle we have (5) ut > 0 for (x,t) Q.
2
Then by differentiating implicitly u (x,t(x, n)) = n with respect to x, t(x, n) satises a differential equation ux (x,t(x, n)) (6) t (x, n) = ut (x,t(x, n)) Thus from (6), the sign of ux and ut , we deduce that t (x, n) < 0 if a < x < m(t) while t (x, n) > 0 if m(t) < x < a. (7)
Each level lines t(x, n) then must attain its minimum value which is a function of n. If we denote by m0 = max , then for n m0 we have min t(x, n) = 0 since u(x, 0) m0 . If n > m0 , then (n) = min (t(x, n)) > 0, otherwise t(x, n) would intersect t(x, m0 ) which is impossible. Let now (n) be the rst time where max u(x,t) = u (x, (n)) = n, i.e. u(x,t) < n for t < (n). Then t(x, (n) (n), otherwise u(x,t(x, (n)) < n would not hold. Since it is the rst time then it is the minimum min (t(x), n) = (n) = max u(x, (n)).
axa axa axa axa axa axa
Proposition 3. Assume that conditions [A] and [B] hold, and (4) + 2 f () + f () 2 + f () f () 0. If t (x, n).utx < 0 for (x,t) Q then t (x, n) > 0, i.e. t(x, n) is concave up. Proof. For any xed n, differentiate (6) with respect to x ut (x,t)t = uxx (x,t) utt (x,t)t 2 2utx (x,t)t . (8)
We already have ut > 0 while uxx < 0. Similarly, as done for uxx , we can also show utt < 0 in Q if utt 0 holds on Q. Indeed, we already have (utt )t (utt )xx f (u)utt = f (u)ut2 < 0. Since utt satises the boundary condition utt (a,t) = 0 and from ut = uxx + f (u) on t =0 utt (x, 0) = uxxt + f (u)ut = (4) + 2 f () + f () 2 + f () f () 0. By the maximum principle, we have utt (x,t) < 0 in Q. It remains to see that if t (x, n).utx < 0 then (8) implies t > 0. The condition t .utx < 0 is not needed when we are close to the boundary x = a, and t > 0.
330
Amin Boumenir
Corollary 4. Assume that conditions [A] and [B] hold, and (4) + 2 f () + f () 2 + f () f () 0. If x a, then the level sets are concave up. Proof. Clearly by the previous proposition and (7), we only need to show that uxt > 0 when x a+ and uxt < 0 when x a. Observe that from u(a, t) = 0 it follows ut (a,t) = 0 and by (5) ut (x,t) > 0, we deduce that for x a+ , ut (x,t) is increasing in x, and so uxt (x,t) > 0. Similarly if x a , then uxt < 0. By (7) and in both cases, x a, we have uxt t < 0. The concavity follows from (8). Corollary 5. Assume that conditions [A] and [B] hold, and the solution converges to its steady state as t , then, the level lines are concave up, i.e. t (x, n) > 0. Proof. Consider the region where the level curve is decreasing, see (7), and denote its inverse by x(t, n) say, i.e. u(x(t, n),t) = n. The analogs of (6) and (8) are ux (x(t, n),t) x (t, n) + ut (x(t, n),t) = 0. uxxx + ux .x + 2utx x + utt
2
(9) (10)
= 0.
Since x (t, n) < 0, the function x(t, n), which is decreasing and bounded from below x(t, n) > a, must have a limit as t . Let lim x(t, n) = (n). Thus by continuity of the solution we have lim u((n),t) = lim u(x(t, n),t) and so ut , utt and uxt 0 as t , and t t so by (10) we have (11) uxxx 2 + ux .x = o(1). Since uxx < 0 while ux > 0 it follows that x > 0 and the level curve is concave. For the second branch of t(x, n), when t (x, n) > 0, concavity means that x (t, n) < 0, which follows from (11). Thus by combining both branches of x(t) we obtain t (x) > 0 and so t(x, n) is concave up.
t
3.
The n-Dimensional Case
In the n-dimensional case, dim (Q) = m + 1, the dimension of the level surface, dim n = m 2 and thus we need m parameters for its representation, n := {(x,t(x, n)) : u(x,t(x, n)) = n} . From the smoothness of the solution, we have u t xi 2 u t x2 i u u + f (u) xi xi 2u 2u u = 2 + f (u) 2 + f (u) xi xi xi =
and u satises the equation 2 t u = 2u + f (u)u + f (u) ||u|| u = f (u) < 0 x , t > 0 u(x, 0) = (x) < 0 x .
Level Sets for Reaction Diffusion Equations By condition [A], f (u) < 0, and so u 2 u f (u)u 0 t which, by the maximum principle, implies that u < 0 for (x,t) Q. Thus the solution u(.,t) is concave down and obviously positive.
331
Proposition 6. Assume that conditions [A] and [B] hold, then the solution u(., t) is positive and concave down. We next establish t(x, n) > 0, which means that the level surfaces are concave up. Using implicit differentiation u(x1, x2 , ..., xn,t(x, n)) = n we deduce u + 2ut .t + utt ||t||2 + ut .t = 0. Using similar arguments as in proposition 3, we have Proposition 7. Assume that conditions [A] and [B] hold, and 2 + 2 f () + f () ||||2 + f () f () 0. If for (x,t) Q we have ut .t < 0 then t(x, n) > 0, i.e. t(x, n) is concave up. In dimension one, the concavity of the solution gives an insight on how a single blow-up point develops. Clearly the shape of level lines also gives a hint, in the n-dimensional case, on the geometry of the blow-up set which is still an open question.
References
[1] T. Chan and Vese, L. Active contour without edges, IEEE trans. on Image processing 10, 266-277, (2001). [2] S. Osher and R. Fedkiw, Level Set Methods and Dynamic Implicit Surfaces, Applied Math. Sciences, 153, Springer, (2003). [3] Y. Giga, S. Matsui, and S. Sasayama. Blow up rate for semilinear heat equations with subcritical nonlinearity. Indiana Univ. Math. J. 53, 2, 483514,(2004). [4] G. Sapiro, Geometric Partial Differential equations and Image Analysis , Cambridge U. press, (2001). [5] J. Sethian, Curvature and the evolution of fronts, Comm.in.Math.Physics. 101, 487-499, (1985). [6] J. Sethian, Level Set Methods and Fast Marching Methods: Evolving Interfaces in Computational Geometry, Fluid Mechanics, Computer Vision and Materials Science, Cambridge University Press, (1999).
O N THE A LMOST P ERIODICITY OF THE S UPERPOSITION OF F UNCTIONS

Stanisaw Stoinski Faculty of Mathematics and Computer Science Adam Mickiewicz University, Umultowska 87, 61-614 Pozna , Poland n
Abstract In this note we present some theorems on the superposition of a function which satises the H lder condition and an (NS p )-almost periodic (a.p.) function, as well an o S p -a.p. function. Moreover, we prove a theorem on the superposition of a continuous function and an (NH)-a.p. function. In the following, we give a theorem on the superposition of a differentiable function and a Vp -a.p. function. Finally, we prove a theorem on the superposition of a differentiable function and an (NC(1) )-a.p. function.
2000 Mathematics Subject Classication: 42A75 . Key words and phrases: Almost periodic function, superposition, Hausdorff metric, variation.
1.
Introduction
The continuous functions which are almost periodic (a.p. for short) with respect to the uniform metric, i.e. the uniformly a.p. functions, was dened and studied by H. Bohr (see [5]) in 1925 and later. V.V. Stepanov, H. Weyl and A.S. Besicovitch had studied the properties of functions a.p. with respect to integral metrics (see [4] - [6], [9]). B.M. Levitan had considered some kinde of generalization of Bohrs a.p. function, namely N-a.p. functions (see [9]). However functions a.p. with respect to the Hausdorff metric (see [8]) (which was studied by B. Sendov and others , see [10]), so-called H-a.p. functions was dened and investigated by S. Stoi ski (see [11]-[14]) and a little later by A.S. Djafarov and G.M. n Gasanov (see [7]). The denition and the properties of functions a.p. in variation (Vp -a.p. for short) can be found for example in [15] and [16]. It is well known (see [9], p.21) that if f is a uniformly continuous function on the set Yg of values of a function g, where g is a uniformly a.p., i.e. B-a.p. function, then the superposition f g is B-a.p. The superpositions of different a.p. functions and continuous or uniformly continous functions, was considered by M. Adamczak and S. Stoi ski in [2], n [18] and [19].
E-mail
address: stoi@amu.edu.pl
334
Stanisaw Stoi ski n
This paper is concerned with the superposition of such functions as: Stepanov a.p., NHa.p. or Vp -a.p. and (NC(1) )-a.p. functions and continuous or differentiable functions. The results of this type have application in the eld of differential equations.
2.
Preliminaries
Let F denote the class of closed and bounded with respect to the y axis point sets on the Oxy plane, such that their respective projections on the x axis are identical with the interval (bounded or unbounded), and such that the intersection of every straight line x = x0 , x0 , and F F is a closed interval. If A, B F , then the Hausdorff distance between A and B we call the following number r (A, B) = max(sup inf X Y
XA Y B 0 , sup inf XB Y A
X Y
0 ),
where X Y
0
= X(x1 , y1 ) Y (x2 , y2 )
= max(|x1 x2 |, |y1 y2 |)).
The functional r : F F R+ is a metric in F . The following lemma is true (see [10]): Lemma 1 . Suppose that A, B F . If: (a) for an arbitrary X A there exists Y B such that X Y and (b) for an arbitrary X B there exists Y A such that X Y then r (A, B) . ,
0 0
It is easily seen that if r (A, B) , then the conditions (a) and (b) from Lemma 1 hold. By the complete graph f of the bounded function f : R, we call a set of points (x, y), for which x , I f (x) y S f (x), where I f , S f is the lower and upper Baires function respectively, both with respect to the function f : I f (x) = lim inf
0 |x x|
f (x )
S f (x) = lim sup f (x ).

0 |x x|
The Hausdorff distance between two functions f and g, both dened and bounded on , is dened in the following way: it is the Hausdorff distance between their respective complete graphs, i.e. r ( f , g) = r ( f, g). Let rR = r. A set E R is called relatively dense iff there exists a positive number l such that in every open interval (, + l), R, there is at least one element of the set E. Let f : R R be a bounded function. If for > 0 there is r( f , f ) , where f (x) f (x + ),then the number R is called an (H, )- almost period ((H, ) -a.p.) of the function f . Let us denote the set of (H, )-a.p. of f by EH {; f }. A bounded function f : R R is called H-almost periodic (H-a.p.) iff for each > 0 the set EH {; f } is relatively dense. Let H be the space of H-a.p. functions. Examples, elementary properties of H-a.p. functions and the connection between H-a.p. functions and almost periodic functions of other types can be found in [11]-[14].
On the Almost Periodicity of the Superposition of Functions For f : R R let us write Vp ( f ) = sup (| f (x)| +Vp ( f ; x)), xR where
x
335
Vp ( f ; x) = sup{ | f (ti+1 ) f (ti )| p } p ,

1
n1 i=0
1 p<
x is the partition x 1 = t0 < t1 < t2 < ... < tn = x + 1 of the interval x 1, x + 1 , x R. Let us denote (BVp )loc = { f : R R : Vp ( f ; x) < for every x R}. Let f (BVp )loc . If for > 0 there is Vp ( f f ) , then the number R is called a (Vp , )-almost period of the function f . A continuous function f (BVp )loc is called almost periodic in the p-th variation (Vp -a.p.) iff for each > 0 the set Evp {, f } = { R : Vp ( f f ) } is relatively dense. For p = 1 we obtain a V -a.p. function. Let Vp be the space of Vp -a.p. functions. Elementary properties of functions almost periodic in variation can be found in [15] and [16]. Let us dene for f C(n) (R) the functional D(n) ( f ) = sup (| f (x)| + | f (i) (x)|). i=1 xR A function f C(n) (R) is called C(n) -almost periodic (C(n) a.p.) iff for each > 0 the set E (n) {, f } = { R : D(n) ( f f ) } is relatively dense. By C(n) we denote the space of C(n) -a.p. functions. Basic properties and examples of C(n) -a.p. functions we can nd in [1]. Let f : R R be a continuous function. If for N > 0, > 0 there is DN ( f f ) = max{| f (x) f (x)| : N x N} , then the number R is called an (N, )-almost period ((N, )-a.p.) of the function f . Let NE{; f } denote the set of (N, )-almost periods of f . A continuous function f : R R is called N-almost periodic (N-a.p.) iff there exists a uniformly a.p. function (see [9]), a dominant function of f , such that for any two positive numbers N, there exists a > 0 such that E{; } NE{; f }, where E{; } = { R : sup{|(x) (x)| : x R} }. By N we denote the space of N-a.p. functions. The above denitions were introduced by B. M. Levitan (see [9]). The number R is called an (NH, )-almost period ((NH, )-a.p.) of the bounded on every closed interval function f : R R iff for two positive numbers N, we have r N,N ( f , f ) . Let us denote by NEH {; f } the set of (NH, )-almost periods of f . A bounded on every closed interval function f : R R is called (NH)-almost periodic ((NH)a.p.) iff there exists an H-a.p. function , a dominant function of f , such that for any two positive numbers N, there exists a > 0 such that EH {; } NEH {; f } (see [19]).
n
336
Stanisaw Stoi ski n
The number R is called an (NS p , )- almost period ((NS p , )-a.p.) of the function p f Lloc , where 1 p < , iff for two positive numbers N, we have (NDS p )( f f ) = max (
NxN
Let us denote by NES p {; f } the set of all (NS p , )-almost periods of f . A function p f Lloc is called (NS p )-almost periodic ((NS p )-a.p.) iff there exists an S p -a.p. function (see [9]), a dominant function of f , such that for any two positive numbers N, there exists a > 0 such that ES p {; } NES p {; f }, where ES p {; } = { R : sup ( xR (see [17]). Let us write for f C(n) (R) (ND(n) )( f ) = max (| f (x)| + | f (i) (x)|)
NxN i=1 n
The number R is called an ((ND(n) ), )-almost period (((ND(n) ), )-a.p.) of the function f C(n) (R), where > 0, iff (ND(n) )( f f ) . Let NE (n) {; f } denote the set of ((ND(n) ), )-almost periods of f . A function f C(n) (R) is called (NC(n) )-almost periodic ((NC(n) )-a.p.) iff there exists a C(n) -a.p. function , a dominant function of f , such that for any two positive numbers N, there exists a > 0 such that E (n) {; } NE (n) {; f } (see [3]).
3.
Main Results
The following theorem, which concerns (NS p )-a.p. functions, is true: Theorem 1 . If a function f : R R satises the H lder condition with the exponent o 1 0 < 1 and with the constant M > 0, a function g is (NS p )-a.p., where p , then the p )-a.p. superposition f g is (NS
p Proof. Since f satises the H lder condition and g Lloc , so for an arbitrary interval o < a, b > R the following estimation b a
| f (g(u))| p du 2 p(1+) M p
p where c is a certain constant, holds. Hence, it follows that f g Lloc . p )-a.p., there exists an S p -a.p. function , a dominant funcBecause g is (NS tion of g, such that for any two positive numbers N, there exists a > 0 such that ES p {; } NES p {(/M)1/ ; g}. Hence, for every x N; N and ES p {; } the following estimation x+1 x1
| f (g(u)) f (g (u))| p du
sup (
NxN
1 p
x+1 x1
| f (u) f (u)| p du) p .
x+1 x1
|(u) (u)| p du) p }
for N > 0.
b a
|g(u)| p du + c,
x+1 x1
|g(u) g (u)|p du) p
On the Almost Periodicity of the Superposition of Functions
337
holds, i.e. NES p {; f g}, and so is a dominant function of f g, too. Therefore, the superposition f g is (NS p )-a.p. We prove likewise the following: Theorem 2 . If a function f satises the H lder condition with the exponent 0 < 1, a o p -a.p., where p 1 p function g is S , then the superposition f g is S -a.p. It is known (see [19]) that the superposition of a continuous function and a continuous H-a.p. function is (NH)-a.p. Now, we will prove the following theorem: Theorem 3 . If f is a continuous function on the set Yg of values of a function g, where g is a continuous (NH)-a.p. function, then the superposition f g is (NH)-a.p. Proof. Fix arbitrarily N > 0, > 0. Let us denote mN = min{g(x) : N x N}, MN = max{g(x) : N x N}. By the Darboux property it follows that {g(x) : N x N} = mN ; MN . Because g is (NH)-a.p., there exists an H-a.p. function , a dominant function of g, such that for any two positive numbers N > 0, > 0 there exists a > 0 such that EH { ; } NEH { ; g}. Hence, for EH { ; } we have r N,N (g , g) . Therefore, for x N; N and EH { ; } we obtain g (x) = g(x + 1 (x, ) ) + 2 (x, ) and g(x) = g (x + 1 (x, ) ) + 2 (x, ) , where |i (x, )| 1, |i (x, )| 1, i = 1, 2, x + 1 , x + 1 N, N . If mN = inf{g(x) : x R} and MN = sup{g(x) : x R}, then f is a uniformly continuous function on the set Yg . Let 0 < < min(, ), where a = () > 0 is a number which characterizes the uniform continuity of the function f on the closed interval Yg , and EH { ; }. For an arbitrary X = X(x, y) ( f g) there exists Y = Y (x + 1 , f (g(x + 1 ))) f g, where x N, N , such that X Y 0 . Analogously, for an arbitrary X = X(x, y) f g there exists Y = Y (x + 1 , f (g (x + 1 ))) ( f g) , where x N, N , such that X Y 0 . Hence, from the above estimates we get by Lemma 1 r N,N (( f g) , f g) , and so the superposition f g is an (NH)-a.p. function, because every (H, )-a.p. of is an (NH, )-a.p. of f g, i.e. is a dominant function of f g. In the following we assume that mN , MN Yg . If Yg is a bounded interval, i.e. there exist the constans m = inf{g(x) : x R} and M = sup{g(x) : x R}, then we construct the closed interval 1 mN aN , MN + aN , where 0 < aN = 2 min{mN m, M MN }, if mN > m, MN < M, mN , MN + aN , where 0 < aN = 1 (M MN ), 2 IN = if mN = m, MN < M, mN aN , MN , where 0 < aN = 1 (mN m), 2 if mN > m, MN = M.
338
Stanisaw Stoi ski n In case when Yg is an unbounded interval, then we construct the closed interval if Yg = R, mN 1, MN + 1 , m 1, M + a , where 0 < a = 1 (M M ), N N N N N 2 if MN < M, infYg = , mN 1, MN , if MN = M, infYg = , IN = mN aN , MN + 1 , where 0 < aN = 1 (mN m), 2 if mN > m, supYg = +, mN , MN + 1 , if mN = m, supYg = +.
The function f is uniformly continuous on the closed interval IN (or IN ). Let 0 < = (N, ) min(1, aN ) be a number which characterizes the uniform continuity of a function f on the interval IN (or IN ), 0 < < min(, ) and EH { ; }, where is dened as in the rst part of the proof. Then for an arbitrary X = X(x, y) ( f g) , where x N, N , there exists Y = Y (x + 1 , f (g(x + 1 ))) f g such that X Y 0 . Inversely, for an arbitrary X = X(x, y) f g, where x N, N , there exists Y = Y (x + 1 , f (g (x + 1 ))) ( f g) such that X Y 0 . From the above estimates we get by Lemma 1 r N,N (( f g) , f g) . Hence, the superposition f g is an (NH)-a.p. function, because every (H, )-a.p. of is an (NH, ) a.p. of f g, i.e. is a dominant function of f g. It is known (see [19]), that if the derivative f of a function f :Yg R is uniformly continuous and bounded on Yg , where g is a B-a.p. function such that the derivative g is uniformly continuous, then the superposition f g is V -a.p. The following theorem, which concerns Vp -a.p. functions, is true: Theorem 4 . If the derivative f of a function f :Yg R is uniformly continuous and bounded on the set Yg of values of a function g, where g is a Vp -a.p. function, 1 p < , then the superposition f g is Vp -a.p. Proof. Because f is bounded on Yg , there exists a constant M1 > 0 such that | f (u)| M1 for every u Yg . The Vp -a.p. function g is Vp -bounded (see [16]), and so there exists a constant M2 > 0 such that Vp (g; x) M2 for every x R. Since the derivative f is bounded on Yg , so f is uniformly continuous on Yg . It is known that g is a Vp -a.p. function, and so g is uniformly a.p. Hence, g is uniformly continuous on R (see [9]). Therefore, for an arbitrary > 0 there exist a positive numbers 1 , 2 , 3 such that | f (u ) f (u )| < and |g(t ) g(t )| < 3 6M2 1 3 for |u u | < 1 , u , u Yg ,
for |t t | < 2 , for |u u | < 3 ,
t ,t R,
| f (u ) f (u )| <
u , u Yg ,
(1)
On the Almost Periodicity of the Superposition of Functions where 0 < 3 < min ( 1 , 6M1 , 3
3M1 (2n0 +1)
1/p
339
2 ), n0 = [ 2 ] + 1. Since g (BVp )loc , so for an
arbitrary partition x : x 1 = t0 < t1 < ... < tn = x + 1, where x R, we have by the Lagrange theorem { |( f g)(ti+1 ) ( f g)(ti )| p } p
1
n1 i=0
= { (| f (ui )||g(ti+1 ) g(ti )|) p } p

1
n1 i=0
M1 { |g(ti+1 ) g(ti )| p } p ,
1
n1 i=0
where ui (min(g(ti ), g(ti+1 )), max(g(ti ), g(ti+1 ))). Hence, we obtain for every x R Vp ( f g; x) M1Vp (g, x) M1 M2 < , and so f g (BVp )loc . Since g is a uniformly a.p. function, so for Evp {3 ; g} we have |g(t + ) g(t)| 3 for every t R. From (1) we conclude that | f (g(t + )) f (g(t))| < , 3 t R. (2)
Because g is Vp -a.p., for Evp {3 ; g} we have for every partition x , x R, where the partition x is dened as in the rst part of the proof, { |[g(ti+1 ) g (ti+1 )] [g(ti ) g (ti )]| p } p 3 .
1
n1 i=0
(3)
Now, we present the estimate Vp ( f g ( f g) ; x), x R,
for Evp {3 ; g}. We choose the subsequences (ti ) and (ti ) of the sequence (ti ), i = 0, 1, 2, ..., n 1, such k k that k = 0, 1, 2..., n 1, for ik {0, 1, 2, ...n 1}, |ti +1 ti | < 2
k k
and |ti
k +1
ti | 2
k
for ik {0, 1, 2, ...n 1},
k = 0, 1, 2..., n 1.
Let us write p ( f g ( f g) ; x ) = { |[ f (g(ti+1 )) f (g (ti+1 ))] [ f (g(ti )) f (g (ti ))]| p } p

1
n1 i=0
= { |[ f (g(tik +1 )) f (g (tik +1 ))] [ f (g(tik )) f (g (ti ))]| p

k=0
k
n 1
340
n 1
Stanisaw Stoi ski n +
k=0
|[ f (g(ti
)) k +1
f (g (ti
n 1 k=0
))] [ f (g(ti )) k +1 k
1
f (g (ti ))]| p } p
k
{ Aip } p + {
k
n 1 k=0
Ap } p ,
ik
where Al = |[ f (g(tl+1 )) f (g (tl+1 ))] [ f (g(tl )) f (g (tl ))]| for l = ik or l = ik . By the Lagrange theorem it follows that there exists ui ( min(g(tik ), g(tik +1 )), max(g(tik ), g(tik +1 ))),
k
ui ( min(g (tik ), g (tik +1 )), max(g (tik ), g (tik +1 )))

k
for which Aip (| f (ui ) f (ui )||g (tik +1 )g (tik )|+|[g(tik +1 )g (tik +1 )][g(tik )g (tik )]|| f (uik )|) p ,
k k k
and there exists ul ( min(g(tl ), g (tl )), max(g(tl ), g (tl ))), for which A p (| f (ui
ik
k +1
l = ik , ik + 1,
) f (ui )||g(ti ) g (ti )|

k k k k k k +1
+|[g(ti
k +1
) g (ti
k +1
)] [g(ti ) g (ti )]|| f (ui
)|) p .
Using Minkowskis inequality, we obtain { Aip } p { (| f (ui ) f (ui )||g (ti +1 ) g (ti )|) p } p
1 1
n 1 k=0
n 1 k=0
+{ (|[g(tik +1 ) g (tik +1 )] [g(tik ) g (tik )]|| f (uik )|) p } p .

1
n 1 k=0
For Evp {3 ; g} we have |ui ui | < 1 , and so

k k
| f (ui ) f (ui )| <

k k
, 6M2
k = 0, 1, 2, ...n 1,
and, using the estimation (3), we obtain

n 1 1 { Ap } p < . i 3 k=0 k
In the following we have for Ev {3 ; g}

p n 1
k=0
Ap } p {
ik
n 1 k=0
(| f (ui +1 ) f (ui
k
)||g(ti ) g (ti )|) p } p

k k
On the Almost Periodicity of the Superposition of Functions

n 1
341
1
+{
k=0
(|[g(ti +1 ) g (ti +1 )] [g(ti

k k 1
) g (ti )]|| f (ui

k
)|) p } p +1
p M1 3 (2n0 + 1),
where n0 = [2/2 ] + 1. Therefore, we obtain p ( f g ( f g) ; x ) < From (2) and (4) it follows that Vp ( f g ( f g) ) for Evp {3 ; g}, i.e. the superposition f g is Vp -a.p. It is known (see [2]), that if the derivative f of a function f :Yg R is uniformly continuous and bounded on Yg , where g is a C(1) -a.p. function, then the superposition f g is C(1) -a.p. Now, we will prove the following theorem: Theorem 5 . If the derivative f of a function f :Yg R is continuous on the set Yg of values of a function g, where g is an (NC(1) )-a.p. function, then the superposition f g is (NC(1) )-a.p. Proof. It is known (see [3]) that since g is an (NC(1) )-a.p. function, so g and g are Na.p. Because the superposition of a continuous function and an N-a.p. function is an N-a.p. function (see [19]), the superpositions f g and f g are N-a.p. For every x R we have ( f g) (x) = f (u) g (x) where u = g(x), and hence the product f (g) g is N-a.p. (see [9], p. 146). It is known (see [3]) that then f g is an (NC(1) )-a.p. function.
1 2 p + M1 3 (2n0 + 1) < . 3 3
(4)
Acknowledgment
I would like to express my gratitude to the Reviewer for his invaluable critical remarks and suggestions.
References
[1] M. Adamczak, C(n) -almost periodic functions, Commentat. Math. 37 (1997), 112. [2] M. Adamczak, On compactness of C(n) -almost periodic functions, Demonstr. Math. Vol.32 No 2 (1999), 385393. [3] M. Adamczak and S. Stoi ski, On the (NC(n) )-almost periodic functions, Function n Spaces, Proceedings of the Sixth Conference, Wrocaw 2001, World Scientic Publishing Co. Pte. Ltd. (2003), 3948. [4] A. S. Besicovitch, Almost periodic functions, Cambridge 1932.
342
Stanisaw Stoi ski n
[5] H. Bohr, Zur Theorie der fastperiodischen Funktionen, I Teil, Acta Math. 45 (1925), 29127; II Teil, Acta Math. 46 (1925), 101214. [6] C. Corduneanu, Almost Periodic Functions, Chelsea Publishing Co., New York, 1989. [7] A. S. Djafarov and G. M. Gasanov, Almost periodic functions of the relatively Hausdorff metric and some of their properties, Izv. Acad. Azerb. Sci. Ser. Sci. Phys. Tech. Math. 1 (1977), 5762 (in Russian). [8] F. Hausdorff, Grundz ge der Mengenlehre, Leipzig 1914. u [9] B. M. Levitan, Almost periodic functions, Moscow 1953 (in Russian) . [10] B. Sendov and B. Penkov, -entropy and -capacity of the space of continuous functions, Izv. Math. Inst. Acad. Bulg. Sci. 6 (1962), 2750 (in Bulgarian). [11] S. Stoi ski, H-almost periodic functions, Funct. Approximatio Comment. Math. 1 n (1974), 113122. [12] S. Stoi ski, A connection between H-almost periodic functions and almost periodic n functions of other types, Funct. Approximatio Comment. Math. 3 (1976), 205223. [13] S. Stoi ski, On modulus of non-monotonicity of an H-almost periodic function, n Funct. Approximatio Comment. Math. 4 (1976), 8590. [14] S. Stoi ski, On superposition of the Bohrs almost periodic functions, Funct. Apn proximatio Comment. Math. 5 (1977), 8589. [15] S. Stoi ski, Real-valued functions almost periodic in variation, Funct. Approximatio n Comment. Math. 22 (1993), 141148. [16] S. Stoi ski, Real-valued functions almost periodic in p-variation, Fasc. Math. 26 n (1996), 155162. [17] S. Stoi ski, N-almost periodic in the sense of variation functions, Commentat. Math. n 41 (2001), 195202. [18] S. Stoi ski, A note on N-almost periodic functions and (NI)-almost periodic funcn tions, Commentat. Math. 44(2) (2004), 199204. [19] S. Stoi ski, On superposition of almost periodic functions, International Journal of n Evolution Equations Vol.1 Number 2 (2005), 145151.
T IME P ERIODIC S OLUTIONS FOR Q UASIGEOSTROPHIC M OTION AND T HEIR S TABILITY

Mei-Qin Zhan Department of Mathematics & Statistics University of North Florida Jacksonville, FL 32224, USA
Abstract In this article, we study the quasigeostrophic equation, which is a prototypical geophysical uid model. We will show the existence of time-periodic solutions for any the Coriolis parameter and the Ekman dissipation constant with nonhomogeneous boundary conditions, which improves the results obtained in [4] and [13].
1991 Mathematics Subject Classication: 35B05, 35B30, 35B65 Key words and phrases: quasigeostrophic uid model, time-periodic motion, positive solution, monotonicity
1.
Introduction
t + J(, ) + x = 2 r + f (x, y,t), (1.1)
The two dimensional barotropic quasigeostrophic equation
where (x, y,t) is the stream function, > 0 is the meridional gradient of the Coriolis parameter, > 0 is the viscous dissipation constant, r > 0 is the Ekman dissipation constant, and f (x, y,t) is the wind forcing. Moreover, = xx + yy is the Laplacian operator in the x,y plane, x, y D an arbitrary bounded planar domain with piecewise smooth boundary, and J( f , g) = fx gy fy gx is the Jacobi operator. This equation is derived as an approximation of the rotating shallow water equations by the conventional asymptotic expansion in small Rossby number. Schochet ([9]) has recently shown that quasigeostrophy is a valid approximation of the rotating shallow water equations in the limit of zero Rossby number, i.e. at asymptotically high rotation rate. Assume is C2 , let = , then the quasigeostrophic equation becomes the following system of equations, t + J(, ) + x = r + f (x, y,t) = ,
E-mail
(1.2)
address: mzhan@unf.edu
344
Mei-Qin Zhan
with the following boundary and initial conditions, which satises appropriate compatible conditions, (x, y,t) = g1 (x, y,t) on D, (x, y,t) = g2 (x, y,t) on D (x, y, 0) = 0 (x, y). (1.3)
The time periodic solution of (1.2) has been studied by many authors. In [4], the author studied the quasigeostrophic model (1.2) with the homogeneous boundary conditions, g1 (x, y,t) = g2 (x, y,t) = 0. He showed that (1.2) admits at least one time periodic solution under a rather restrictive technique condition that the Coriolis parameter and the Ekman dissipation constant satisfy r + |D| > 1 ( |D| + 1), here |D| is the area of domain D. In [13], 2 under the same condition on r and , the author shows that the when f (x, y,t) is positive and periodic in time, the barotropic quasigeostrophic equation admits a time-periodic positive solution. Moreover, the author also showed that the solution is stable when its x is bounded in L norm. In this article, by using the method of upper and lower solutions and its associated monotone iteration, we will show that the equation (1.2) admits at least one time-periodic for any the Coriolis parameter and the Ekman dissipation constant with no homogeneous boundary conditions. Thus we are able to improve the result of [4] and [13].
2.
The Main Results
In this section we will state and prove our main results. We rst give a sketch proof of the following existence of classical solution to the quasigeostrophic equations (1.2)-(1.3). Let D = D D, and X = C1,2 ([0, ) D) C([0, )). Lemma 2.1. The quasigeostrophic equations (1.2)-(1.3) have an unique classical smooth solution , X when f (x, y,t) C([0, ),C (D)), g1 (x, y,t), g2 (x, y,t) C([0, )C (D)), and 0 (x, y) C2, (D). Proof. First we transform the equations so that the boundary conditions become nonslippery. Let and be the classical solutions of the following two linear equations, t = r (x, y,t) = g2 (x, y,t) on D (x, y, 0) = 0. and = , (x, y,t) = g1 (x, y,t) on D, Set = and = , we have the following equations, t + J(, ) + x + + J(, ) + J( , ) x = r + f (x, y,t) + J( , ) = , (2.6) (2.5) (2.4)
Time Periodic Solutions and Their Stability with the following non-slippery boundary conditions and initial condition (x, y,t) = 0 on D, (x, y,t) = 0 on D (x, y, 0) = 0 (x, y).
345
(2.7)
The equations (2.6-2.7) have the same nonlinear term as the vorticity form of the 2D Navier-Stokes equations, so the methods in [12, 3, 6, 8] can be applied to obtain the existence and uniqueness of the classical solutions. The main idea uses the nonlinear Galerkin method. By projecting , , , f = f (x, y,t) + J( , ) onto a nite-dimensional subspace(expanded by nite-Fourier series). Let Pn be the projection operator and Gn = Pn G etc to denote the projection of any function G into the subspace, the equations (2.6) become a system of ODEs tn + Pn J(, ) + n + +n + Pn J(n , n ) + Pn J(n , n ) x x = n r + f n (x, y,t) + Pn J(n , n ) n = n , (2.8)
which can be solved globally and to obtain the approximate solutions. To prove the convergence of the approximate solutions we need the following energy estimates. First, multiply both sides of equation for in (2.6) by and integrate by part, notice
J(, )dxdy = 0
and the Poincar inequality, e | we have 1d 2 dt

2
1 |D| x dxdy| ( + 1) 2 D
+C1
C2 (t)
2 dxdy +C3 (t)
Second, from the equation for in (2.6) we have ||2 dxdy 2

D
||2 dxdy
D
Notice that | D J(, ) dxdy| 2 4 (D) (t) C() L L1 ([0, T ))
2 (t) + L4 (D)
and using this and Sobolev embedding, we know that
for any T > 0. Finally, multiply both sides of equation for in (2.6) by and integrate by part, using (2.10) we have
1 d 2 dt
2 +C 4
C5 2 4 (D) (t) 2 4 (D) (t) L L +C6 (t) 2 +C7 (t) 2 +C8 (t)
2 dxdy
(2.9)
2 2 (t) L4 (D)
(2.10)
2 L4 (D)
is in
(2.11)
346
Mei-Qin Zhan
Here Ci (t), i = 1, 8 denote various bounded or integrable functions in t that do not depend on , and . Inequalities (2.9-2.11) enable us to pass to limit to obtain strong solutions in Sobolev space C1 ([0, T ), H 1 (D)) L2 ([0, T ), H 2 (D)). Then regularity results on the parabolic and elliptic equations, one has the classical smooth solutions. The uniqueness is a consequence of the comparison principle. The next result we want to present is the following comparison principle, this result indicates that the quasigeostrophic equation enjoys a comparison principle such that two solutions with different initial data and external forces can be compared. Theorem 2.2. Suppose that f (i) (x, y,t) C (D), g1 (x, y,t), g2 (x, y,t) C (D) i = 1, 2 (1) (2) (1) (2) and f (1) (x, y,t) f (2) (x, y,t), g2 (x, y,t) g2 (x, y,t), 0 (x, y) 0 (x, y), for all x, y,t. Let (i) (x, y,t) X , (i) (x, y,t) X , i = 1, 2, be the solutions of
(i) (i)
t + J((i) , (i) ) + x = (i) r(i) + f (i) (x, y,t) (i) = (i) , with boundary and initial conditions, (i) (x, y,t) = g1 (x, y,t) on D, (i) (i) (x, y,t) = g2 (x, y,t) on D (i) (i) (x, y, 0) = 0 (x, y). We have (1) (x, y,t) (2) (x, y,t) for all x, y,t. Proof. Let (x, y,t) = (1) (x, y,t) (2) (x, y,t), (x, y,t) = (1) (x, y,t) (2) (x, y,t), f (x, y,t) = f (1) (x, y,t) f (2) (x, y,t), (1) (2) g1 (x, y,t) = g1 (x, y,t) g1 (x, y,t), (1) (2) g2 (x, y,t) = g2 (x, y,t) g2 (x, y,t), (1) (2) 0 (x, y) = 0 (x, y) 0 (x, y),
(i)
(i)
(i)
(2.12)
(2.13)
(2.14)
Using the equations (2.12) and (2.13) we have the following equations for the differences dened in (2.14) t + J((1) , ) + J(, (2) ) + x = r + f (x, y,t) = , with boundary and initial conditions, (x, y,t) = g1 (x, y,t) on D, (x, y,t) = g2 (x, y,t) on D (x, y, 0) = 0 (x, y). Now let u(x, y,t) = et (x, y,t), and (x, y,t) = et (x, y,t), then ut = u + et d (x, y,t), dt (2.16) (2.15)
Time Periodic Solutions and Their Stability so ut + u = et d (x, y,t). dt
347
Multiply both sides of equations in (2.15) and (2.16) by et we have ut + u + J((1) , u) + J(, (2) ) + x = u ru + et f (x, y,t) = u. with boundary and initial conditions, (x, y,t) = et g1 (x, y,t) on D, u(x, y,t) = et g2 (x, y,t) on D u(x, y, 0) = u0 (x, y).
(2.17)
(2.18)
Now if (x0 , y0 ,t0 ) < 0, at some point interior point (x0 , y0 ,t0 ) (0, T ) D. We have u reaches its negative minimum at some interior (x1 , y1 ,t1 ) (0, T ) D. So ut (x1 , y1 ,t1 ) = ux (x1 , y1 ,t1 ) = uy (x1 , y1 ,t1 ) = 0, this indicates that, at (x1 , y1 ,t1 ), Furthermore, we have u(x1 , y1 ,t1 ) 0. Using the equation (2.17) for u, we have for all t 0, u(x1 , y1 ,t1 ) + J(, (2) )(x1 , y1 ,t1 ) + x (x1 , y1 ,t1 ) = u(x1 , y1 ,t1 ) ru(x1 , y1 ,t1 ) + et f (x1 , y1 ,t1 ) ru(x1 , y1 ,t1 ) + et f (x1 , y1 ,t1 ) Notice that satised the following second-order linear elliptic equation, = u (x, y,t) = et g1 (x, y,t) on D, so we have, |x (x1 , y1 ,t1 )| + |y (x1 , y1 ,t1 )| C(|u|(t1 ) + |et1 g2 |(t1 )), (2.21) (2.20) J((1) , u) = 0.
(2.19)
here |h|(t) denotes the maximum value of any function h(x, y,t) over D for each xed t. It is easy to see that |J(, (2) ) + x | C(|x |(t) + |y |(t)), Combine (2.21) and (2.22), we have |J(, (2) ) + x | C(|u|(t1 ) + |et1 g2 |(t1 )) (2.23) (2.22)
348
Mei-Qin Zhan
Since u(x1 , y1 ,t1 ) < 0 we can choose large enough in (2.19) so that, using the estimate (2.23), 0 > u(x1 , y1 ,t1 ) + C(|u|(t1 ) + |et1 g2 |) + ru(x1 , y1 ,t1 ), which contradicts with u(x1 , y1 ,t1 ) + C(|u|(t1 ) + |et1 g2 |(t1 )) + ru(x1 , y1 ,t1 ) et f (x1 , y1 ,t1 ) 0. Therefore u, and hence have to be nonnegative. (2) (x, y,t) for all x, y,t. This proves (1) (x, y,t)
Corollary 2.3. Suppose f (x, y,t), g2 (x, y,t), and 0 is nonnegative, then the solution (x, y,t) to equation (1.2-1.3) is nonnegative for all time t > 0. Proof. Choose f (2) (x, y,t) g1 (x, y,t) g2 (x, y,t) 0 in the Theorem 2.2, we have the result. Remark 2.4. Let L() = J(, ) + x + r, this result shows that L is a monotone operator. So it is possible to prove existence of positive time-periodic solutions and stability of time-periodic and almost time-periodic solutions. Hereafter, we require that f (x, y,t) C (D), g1 (x, y,t), g2 (x, y,t) C (D) are all time periodic functions, and D is of class C1+ . We dene the upper and lower solutions for the quasigeostrophic equations below, Denition 2.5. A function u = (, ) is called an upper solution of (1.2) in X if t + J(, ) + x + r f (x, y,t) = , (x, y,t) g1 (x, y,t) on D, (x, y,t) g2 (x, y,t) on D (x, y, 0) (x, y, T ).
(2) (2)
(2.24)
Similarly, u = (, ) is called a lower solution if it satises the inequalities in 2.24 in reversed order. Lemma 2.6. Suppose that (1) f (x, y,t), g2 (x, y,t), and g1 (x, y,t) are positive and (2) there exists a M > 0 such that, (x, y,t) [0, T ] D, M > max{ f (x, y,t) , g1 (x, y,t), g2 (x, y,t)}, r
1 we have u0 = (0, 0) is a lower solution of (1.2)-(1.3) and uM = (M, M( 2 y2 + 1)) is an upper solution of (1.2)-(1.3).
Proof. It is easy to see that u0 = (0, 0) is a lower solution of (1.2)-(1.3). To show that 1 uM = (M, M( 2 y2 + 1)) is an upper solution of (1.2)-(1.3) one just needs to notice that, for 1 2 = M( 2 y + 1) we have = M and x = 0.
Time Periodic Solutions and Their Stability
349
Let J {u = (, ) X ; 0 M, = }, where M is the constant specied in Lemma 2.6. A pair of upper and lower solutions u = (, ), u = (, ) J are said to be ordered in J if on D. Theorem 2.7. Under assumption of Lemma 2.6, the quasigeostrophic equations (1.2)-(1.3) have both maximal time-periodic solution u and a minimal time-periodic solution u that satises, 0 u u. Here we say that u is maximal and u is minimal in the sense that, for any periodic solution u J, we have u u u. Proof. Starting from either u(0) = u0 or u(0) = uM as an initial iteration we construct a sequence {u(m) = ((m) , (m) }, m = 1, 2, 3, by solving t + J((m) , (m) ) + x = (m) r(m) + f (x, y,t) (m) = (m) , (m) (x, y,t) = g1 (x, y,t) on D, (m) (x, y,t) = g2 (x, y,t) on D (m) (x, y, 0) = (m1) (x, y, T ). Denote the sequence by {u(m) } when u(0) = u0 and {u(m) } when u(0) = uM . By Theorem 2.2 and the fact that u0 is lower solution, we have u0 u(1) . Similarly, the fact that uM is an upper solution and u(1) (x, y, 0) = uM (x, y, T ) = 1 uM (x, y, 0) = (M, M( 2 y2 + 1)) indicates that u(1) uM .
1 Since u(1) and u(1) satisfy the same equation with u(1) (x, y, 0) = (M, M( 2 y2 + 1)) > u(1) (x, y, 0) = (0, 0), using Theorem 2.2, we have (m) (m)
(2.25)
u(1) u(1) . Inductively, for all m 1 we have, u0 u(m) u(m+1) u(m+1) u(m) uM . Hence 0 (m) (m) (x, y,t) M for all m. From 0 |(m) (x, y,t)| M for all m, and (m) = (m) , we have |(m) |2 dxdy =
D D
(2.26)
(m) (m) dxdy
|(m) |2 dxdy +
2M ,
350 {(m) (t)}
Mei-Qin Zhan
So {(m) (t)} H 2 (D) is a bounded sequence. By Sobolev embedding we have C1, (D) is a bounded sequence. Applied the regularity results of the linear parabolic equations we can prove that {(m) }, {(m) } are bounded sequences in X , and that 0 |(m) (x, y,t)| M for all m. Combine this with the monotonicity of (2.26), we have u = limm u(m) X and u = limm u(m) X are two classical solutions to (2.25) and u, u J. To show that both u and u are time-periodic solutions, we need to show that (x, y,t) = (x, y,t + T ) for all t > 0, here we use (x, y,t) to denote either (x, y,t) or (x, y,t). Let (x, y,t) = (x, y,t) (x, y,t + T ) we have t + J(, ) + J(, ) + x = r = , with boundary and initial conditions, (x, y,t) = 0 on D, (x, y,t) = 0 on D (x, y, 0) = 0, (2.27)
(2.28)
Now proceed as in Theorem 2.2, we have 0, replace by we can get 0 we have (x, y,t) 0. So (x, y,t) is periodic function in t with period T, hence (x, y,t) is periodic function in t with period T. To prove that u is maximal and u is minimal in the sense that, for any periodic solution u J, we have u u u. To this end, one just need to see that we can regard u as both upper and lower solutions at the same time and we will have u u(m) u0 we have u u. Similarly, we have u u. Remark 2.8. Notice, we just need to assume that f (x, y,t), g1 (x, y,t), and g2 (x, y,t) are bounded, we require that they are non-negative here just to show that the sign of the periodic solution agrees with the sign of f (x, y,t), g1 (x, y,t), and g2 (x, y,t). When | f (x, y,t)| + 1 |g1 (x, y,t)| + |g2 (x, y,t)| M, then u0 = (M, M( 2 y2 + 1)) is the lower solution. In this case we still able to prove the existence of time-periodic solution without impose any conditions on the Coriolis parameter r and the Ekman dissipation constant . It turns out that, under appropriate conditions, the time-periodic solution of (1.2) is stable in the sense stated in the following theorem: Theorem 2.9. Suppose that f (x, y,t) 0, and there exists a b < r such that |y | b, x where ( , ) is the time-periodic solution to the equation t + J(, ) + x = r + f (x, y,t) = , (2.29)
Time Periodic Solutions and Their Stability with boundary and initial conditions, (x, y,t) = g1 (x, y,t) on D, (x, y,t) = g2 (x, y,t) on D (x, y, 0) = 0 (x, y). We have for any 0 (x, y) (x, y, 0) 0 b 0 (x, y) (x, y, 0) + 0 b, if ((x, y,t), (x, y,t)) is solution to the 2.29 with (x, y, 0) = 0 (x, y), then |(x, y,t) (x, y,t)| 0 e(br)t . and |(x, y,t) (x, y,t)| C0 e(br)t , where C depends on and g1 (x, y,t). Proof. Let = 0 e(br)t , = 1 y2 0 e(br)t . 2 It is easy to see that = ( 1 y2 0 e(br)t ) = 0 e(br)t = 2 (x, y,t) = g1 (x, y,t) 0 e(br)t g1 (x, y,t) on D, (x, y,t) = g2 (x, y,t) 0 e(br)t g2 (x, y,t) on D (x, y, 0) = (x, y, 0) 0 = (x, y, T ) 0 (x, y, T ) 0 e(br)T = (x, y, T ). Now, for , we have t + J(, ) + x + r = t + J( , ) + r x +y0 e(br)t + r0 e(br)t (b r)0 e(br)t x = f (x, y,t) + y0 e(br)t r0 e(br)t (b r)0 e(br)t x f (x, y,t).
1 Similarly, let = + 0 e(br)t , = + 2 y2 0 e(br)t ), we can show that
351
(2.30)
(2.31)
(2.32)
= ( 1 y2 0 e(br)t ) 2 = 0 e(br)t = (x, y,t) = g1 (x, y,t) + 0 e(br)t g1 (x, y,t) on D, (x, y,t) = g2 (x, y,t) + 0 e(br)t g2 (x, y,t) on D (x, y, 0) = (x, y, 0) + 0 = (x, y, T ) + 0 (x, y, T ) + 0 e(br)T = (x, y, T ). Now, for , we have t + J(, ) + x + r = t + J( , ) + r x +y0 e(br)t + r0 e(br)t + (b r)0 e(br)t x = f (x, y,t) + y0 e(br)t r0 e(br)t + (b r)0 e(br)t x f (x, y,t).
(2.33)
(2.34)
352
Mei-Qin Zhan Using the comparison result of Theorem 2.2, we have that 0 e(br)t + 0 e(br)t . Elementary result on the elliptic equations show that |(x, y,t) (x, y,t)| C0 e(br)t ,
with C depends on and g1 (x, y,t). Remark 2.10. Note that if f (x, y,t) = 0 and g1 (x, y,t) = g2 (x, y,t) = 0 we have the trivial time-period solution, (x, y,t) = (x, y,t) = 0. So the condition in the Theorem 2.9 is satised and we thus show that when the initial data is small, the solution will decay exponentially. Corollary 2.11. For any |0 (x, y)| 0 b, if ((x, y,t), (x, y,t)) is solution to the equation, t + J(, ) + x = r = , with boundary and initial conditions, (x, y,t) = 0 on D, (x, y,t) = 0 on D (x, y, 0) = 0 (x, y). we have |(x, y,t)| 0 e(br)t . and |(x, y,t)| C0 e(br)t , where C depends only on . Proof. From the remark we see that (x, y,t) = (x, y,t) = 0 are periodic solution, from Theorem 2.9, we have the decaying result.
3.
Conclusion
The quasigeostrophic equation studied here models the geophysical uid dynamics in the case of of innite Rossby deformation radius and with at bottom and is applicable to planetary-scale solutions. The situation with innite Rossby deformation radius is equivalent to the rigid-lid approximation. We have shown that the nonlinear operator dened by quasigeostrophic equation is monotonic in the sense that positive solutions exist when the initial and boundary data is positive. Our result does not impose the conditions on and r,. On the other hand the monotonicity of the operator L() = J(, ) + x + r, will certain enable one to obtain more properties of the long term dynamical behavior of the solutions, including the stability of the time-periodic solutions.
Time Periodic Solutions and Their Stability
353
References
[1] A. J. Bourgeois and J. T. Beale; Validity of the quasigeostrophic model for large-scale ow in the atmosphere and ocean, SIAM, J. Math. Anal. 25(1994), 1023-1068. [2] B. Cushman-Roisin; Introduction to Geophysical Dynamics, Prentice Hall, 1994. [3] Chae, Dongho; Lee, Jihoon Global well-posedness in the super-critical dissipative quasi-geostrophic equations. Comm. Math. Phys. 233 (2003), no. 2, 297311. [4] J. Duan; Time-Periodic Quasigeostrophic Motion under Dissipation and Forcing, Appl. Math. Comput., 102(1999), 121-127. [5] D. Gilbarg and N. S. Trudinger; Elliptic Partial Differential Equations of Second Order, 2nd edition, Springer-Verlag, 1983. [6] Constantin, Peter; Cordoba, Diego; Wu, Jiahong On the critical dissipative quasigeostrophic equation. Dedicated to Professors Ciprian Foias and Roger Temam (Bloomington, IN, 2000). Indiana Univ. Math. J. 50 (2001), Special Issue, 97107. [7] A. Majda, and S. Wang; The Selective Decay Principle for Barostropic Geophysical Flows, MAA. 8 (2001), no. 4, 579594. [8] Mattingly, J. C. and Sinai, Ya. G., An elementary proof of the existence and uniqueness theorem for the Navier-Stokes equations, Commun. Contemp. Math., vol. 1, no. 4, pp. 497516, 1999. [9] S. H. Schochet; Singular limits in bounded domains for quasilinear symetric hperbolic systems having a vorticity equation, J. Diff. Eqns, 68(1987), 400-428 [10] R. Temam, Innite-Dimensional Dynamical Systems in Mechanics and Physics, Springer-Verlag, (2000). [11] G. Wolansky; Existence, uniqueness and stability of stationary barotropic ow with forcing and dissipation, Comm. Pure Appl. Math., 41(1988),19-46. [12] J. Wu; The 2D dissipative quasi-geostrophic equation, Appl. Math. Lett. 15 (2002), no. 8, 925930. [13] M. Zhan; On Quasigeostrophic Motion Under Dissipation and Forcing, submitted.
JACOBIAN FEEDBACK LOOPS ANALYSIS II: STABILITY AND INSTABILITY
Bourama Toni Department of Mathematics & Computer Science Virginia State University, Petersburg VA 23806 E-mail address: btoni@vsu.edu
Abstract. We investigate the loop stability conditions of dierential systems, that is, the conditions of invariance of the Jacobian spectrum under any variation of entries that leaves unchanged the loop structure. We use the dynamical properties of Jacobian Loops described by the products of the matrix entries under cyclic permutations of the indices. It appears that all k-order Feedback Loops given by the union of disjoint simple loops involving k variables, must be positive for any asymptotic stable behavior. We also conjecture that the Loop structure requires a negative Feedback Loop of the system dimension for the onset of chaotic behavior.
1. Introduction Consider the autonomous dierential system xi (t) = dxi (t) = Fi (x1 , x2 , , xn ; k1 , k2 , , kN ) dt x = (x1 , , xn ) Rn , K = (k1 , k2 , , kN ) RN ,
(1.1)
describing a dynamical system with phase space in Rn , and the parameter/control space in RN . The component functions Fi , i = 1, 2, , n of F (x, K) are assumed to be at least C 1 (U ), that is, dierentiable along with their rst partial derivatives on U an open set of Rn . The partial order relation x y xi yi , i = 1, ..., n denes the vector order in Rn . The Jacobian matrix at x = (1 , . . . , xn ) is given by x J() = DF () = x x Fi (F1 , , Fn ) () = x () x (x1 , , xn ) xj = [Jij ]1i,jn ,
1i,jn
(1.2)
and in general depends on the state variables, except for linear systems [1,9]. The relation xi (t) = Fi (x1 , , xj , , xn , K) actually shows how the rate of change in variable xi is dependent on changes in any given variable xj . Therefore the Jacobian entry Fi xj (x) = Jij , for 1 i, j n describes the interaction between the variables xi and xj , as positive (respectively negative, no) interaction for Jij > 0 (respectively Ji,j < 0, Ji,j = 0). Dynamical systems theory is mostly based on quantitative values of the Jacobian entries. But for some systems, mainly in Biosciences such as Biology and Biochemistry, and in Economics, the relevant informations are of qualitative nature, that is, the sign si,j of the nonzero Jacobian entries, si,j {+, }, also called logical Jacobian entries. Combination of
1991 Mathematics Subject Classication. 34C,58F,92B,93D. Key words and phrases. Jacobian Feedback loops, loop equivalence, stability.
356
Bourama Toni
the sign entries si,j describe the Jacobian Feedback loops whose denition and dynamical roles are given below. The theory of Jacobian loops is therefore the analysis of the dynamics (simple and complex) using solely the loop-pattern Jacobian matrix, that is, even when only the signs, not the magnitudes of the Jacobian terms, are known. As such it is certainly an ecient tool in surveying dynamical response of models simulating physico-chemical, biological and economical systems by stressing qualitative understanding as the primary goal rather than numerical prediction. In a series of studies we intend to show that the Jacobian Loop analysis is an ecient tool in the qualitative study of dierential systems, in particular, for dierential models where informations about the underlying dynamics often reside in the rules of construct of the system and not in the absolute values. See for instance [2,6]. Important questions to be addressed include the charaterization of Jacobian loops and their combinations promoting stability/instablity, exchange of stability and Hopf bifurcation, as well as the onset of chaotic behavior in evolution equations. 1.1. Jacobian Loops: Denitions and Notations. For In the set of indices 1, , n we denote by Ik = {i1 , , ik } an ordered subset of k dierent elements of In and by Ik = k (Ik ) = {j1 , , jk }, with k k a permutation of Ik . Recall Card(k ) = k!, i.e., there are k! permutations. Every permutation k may be factored into disjoint circular (cyclic) permutations i , i = 1, , , that is, k = 1 2 . The signature of k , denoted sg(k ), is (1) , the number of inversions in k , that is, the number of pairs (jm , jn ) with jm > jn while im < in , for jm = k (im ), and jn = k (in ). The permutation k , is even (resp. odd) for an even (resp. odd) . There are c e o exactly k! even and exactly k! odd permutations in k . We denote k (resp. k , k ) the 2 2 subset of circular (resp. even, odd) permutations. [3]. The set n is the classic symmetric group of permutations on the set of indices In . We have the following dening concepts. The set of nonzero terms Jij , i Ik , and j Ik , describes a Jacobian loop associated to the nonzero product
l=k
P (k , J) :=
l=1
Jil k (il ) = Ji1 k (i1 ) Ji2 k (i2 ) . . . Jik k (ik )
(1.3)
called a loop product. The loop is called a korder simple Jacobian loop Lk when the permutation k is a kcycle= (i1 , i2 , , ik ) with the loop product pk = Ji1 i2 Ji2 i3 Jik1 ik Jik i1 . (1.4)
Its sign sgn(Lk ) is that of the loop product pk = P (Lk ) := P (k , J). Its length or dimension l(Lk ) = k is the number of loop factors Jil k (il ) involved, as well as the number of related system variables xi . Therefore (1) A simple Jacobian loop Lk is positive (resp. negative) for an even (resp. odd) number of negative loop factors Jil k (il ) in the loop product P (Lk ). (2) The loop graph Lk associated with the loop Lk consists of k distinct vertices given by the system variables xi , i = 1, . . . , k and k edges Eij = (xi , xj , sij ) directed from j to i where sij = sign(Jij ) denotes the nature of the interaction between the variable xj and xi . (3) A positive (resp. negative) loop involving the variables x1 , x2 , . . . , xk is also conveniently denoted L+1 x2 xk (resp. L1 x2 xk ). x x
Jacobian Feedback Loops Analysis II: Stability and Instability
357
A non-circular permutation k yields a union of simple Jacobian loops, called a composite loop L = i= Li = (L1 , . . . , L ) of dimension l(L ) = k given by the sum of the lengths i=1 k k i= of its simple components loops, i.e., k = i=1 d(Li ) = 1 + + . A proper composite loop L of resonance (, k) is a disjoint union of simple loops of k total length k, that is, the component loops do not share a vertex. We denote Pk the loop product of a composite loop Lk . The sign of a composite loop Lk is the sign of Pk , l= or equivalently, sign(L ) = l=1 sign(Li ) = (1) , where is the number of negative k simple loops in L . = (1)+1 is the characteristic of the proper composite loop of k k resonance (, k). Therefore a korder proper composite loop has a negative (resp. positive) resonance, i.e., a negative (resp. positive) characteristic for even (resp. odd). Denition 1.1. A korder Feedback Fk is dened by Fk =
all Pk (1)+1 Pk ,
the loop product of the proper composite loop L . k
Consequently, (see also [8]) (1) A composite loop L is positive (resp. negative) for an even (resp. odd) number of k its negative simple loops. (2) A proper composite loop L with all component simple loops negative has a negative k resonance in the Feedback Fk as dened above. The loop structure (or qualitative structure), denoted LR , corresponding to the region R in the phase space or to a sign-pattern is the set of all Jacobian loops (simple and proper composite) along with their signs. The Jacobian Loop analysis required rst the determination of the loop structure associated to the system in a given region of the phase space, either from the signed entries of a Jacobian matrix evaluated at equilibria or constant at some parameter values or solely from the logical evaluation of the interaction between the variables in terms of positive/negative. Given a general n square matrix A Mn (R), A = [aij ]1i,jn the sign-pattern matrix Al = [sij ]1i,jn , sij {+, } is also called the logical matrix. Denitions 1.2. (1) A matrix B = [bij ]1i,jn is logically or sign equivalent to A = [aij ]1i,jn in the region R if B has the same sign pattern as A, i.e., Al = Bl . We denote A B. (2) A matrix B = [bij ]1i,jn is loop equivalent to A = [aij ]1i,jn in the region R, if B yields the same loop structure LR as A. We denote A B. A denotes the logical equivalence class of matrix A represented by the logical matrix Al . The logical equivalence class is actally represented by a n n array of nonzero sij = +, entries such that, for any matrix A = [aij ] , sign(aij ) = sij . A denotes the loop equivalence class of matrix A. The class = L is a loop structure represented by a set of simple signed loops, such that any matrix A has the loop structure L. Note that the loop equivalence class is the loop structure uncovered from a representative logical matrix. Therefore if A B then they are sign-patterned in a way to provide the same loop structure, though they are not necessarily logically equivalent in the sense sign(aij ) = sign(bij ). Thus matrix operations such as negation, transposition, permutations, and signature similarity allow to derive one sign pattern from the other. Therefore (1) Logical equivalence obviously implies loop equivalence but not inversely.
358
Bourama Toni
(2) Given two matrices A and B loop equivalent, the logical class B may be obtained from that of A by some combination of negation, transposition, permutation, and signature similarity. Clearly, If A B therefore the spectrum (A) is the same as (B). Next, to adress the loop stability analysis, we recall how the role of loops is highly determined by their relation to the characteristic equation. (see also [8]) 2. Loops and Jacobian Spectrum For a matrix A, Jacobian or otherwise, given by A = [aij ]1i,jn the characteristic polynomial is dened by the monic polynomial CA () = |I A| = n + c1 n1 + + ck nk + + cn1 + cn . From Linear Algebra the coecients may be expressed as ck = coef f icient(nk ) = (1)k mk , k = 0, , n 1, (2.2) (2.1)
where the sum extends over all kth order principal minors mk of A. For instance we have cn =(1)n det(A) = (1)n |A|, c1 = Aii = T r(A), for k = n, (2.3)
where Tr(A) is the trace of A.
From the theory of determinant and permutations we may write

i=
mk =
k k
(1)
il Ik
Ail k (il ) =
all
(1)k
i=1
P (i , A) =
all
(1)k P (L ), k
(2.4)
where the permutation k k of the indices 1 i1 < i2 < < ik n factors into the cyclic permutations (1 , , ) yielding the proper composite loop L = (1 , , ) with k loop product P (L ) = Pk as dened in section 1. Consequently we obtain an expression of k the characteristic coecients ck in terms of the proper composite loops. Importantly it was proved in [8] Lemma 2.1. The kth order Feedback Fk involving all the proper composite loops L with k = 1, , k may be expressed in terms of the coecients of the characteristic polynomial by Fk = ck = (1)+1 Pk . (2.5)
all
Note that: (1) The k order Feedback Fk being the korder coecient of characteristic polynomial entails that the loop factors Ai(i) dened in section 1. are the only Jacobian entries contributing to the characteristic equation, and therefore, inuence directly the eigenvalues of the matrix, and consequently the dynamics. (2) From the standard theory of equations it is also known that the coecients ck = Fk are related to the eigenvalues i in a systematic way by the following Viete formulas:
359
(See [3,5]).
i=1
c1 = F1 = (1 + + n ) =
i=1
Aii = T race(A). (Aii Ajj Aij Aji )
c2 = F2 =
i,j=1,i<j
i j =
i,j=1,i<j
= 1 2 + 1 3 + + n1 n . c3 = F3 =
i,j,k=1,i<j<k
(2.6)
i j k
= (1 2 3 + 1 2 4 + + n2 n1 n ). ... cn = Fn = (1)n 1 2 n = (1)n det(A). It is also known that, given a n-degree polynomial with real coecients p(z) := z n + a1 z n1 + + ak z nk + + an1 z + an , a necessary condition for p(z) to have all its zeros z with negative real part, i.e., p(z) is a strongly stable polynomial, is that all the coecients ak > 0, k = 1, , n. Indeed, if all real parts are negative then we have either the form p(z) = or p(z) = (z ) = (z + a), a > 0. By successive multiplication we necessarily obtain p(z) with ak > 0. This immediately implies (see also [8]) Theorem 2.2. A necessary condition to have all eigenvalues with negative real parts Re < 0 is that all kth order Feedback Fk must be positive. For the sake of completeness we recall The following results from [8] along with their proofs. They are used in the loop stability analysis. Lemma 2.3. A proper composite loop L with all components simple loops negative has k a negative resonance in the kth order Feedback, that is, its characteristic is negative. k Proof. Indeed the term (1)+1 P (L ) in the Feedback Fk has the sign (1) (1)+1 = 1. k Hence the claim. Theorem 2.4. If there is no proper composite loop Lk of dimension k n, then the characteristic coecient ck = 0. Moreover at least one proper composite loop Ln of the system dimension is necessary to have a nonsingular jacobian matrix. Proof. From formulas (2.4), (2.5) the characteristic coecient ck can be written as
l=r1 l=r2 l=r
(z ( + i)) =
(z 2 2z + 2 + 2 ) =
(z 2 + az + b),
a > 0,
b > 0,
ck = Fk =
(1)
l=1
Jil k (il )
l=r1 +1
Ail k (il )
l=r1
Ail k (il ) .
(2.7)
360
Bourama Toni
Terms in the expression of ck with one cyclic permutation correspond to r1 = k, those with two cyclic permutations correspond to r1 < k, r2 = k r1 , and so on. Therefore, if there is no proper composite loop Lk of dimension k, then each term of the sum is zero. Hence the claim. For k = n the system dimension, formulas (2.5)-(2.7) yield clearly Det(A) = |A| = 0. We say that the logical equivalent class A or the loop equivalence class A is logically nondegenerate if every matrix in the class is nonsingular in the sense |A| is nonzero. We prove Theorem 2.5. If the loop structure L does contains a composite loop Ln of the dimension of the system, and all such loop Ln have the same sign, then the corresponding Jacobian determinant |A| is nonzero. Proof. Indeed suppose all the composite loops Ln of the dimension of the system have the same sign. Then cn , consisting of nonzero terms of the same sign, is therefore nonzero. Consequently, the Jacobian determinant is nonzero. Moreover cn is positive (resp. negative) if all Ln have an odd (resp. even) number o (resp. e ) of simple loops Li . We also prove Theorem 2.6. A positive simple loop in the loop equivalence class is a necessary condition for the Jacobian matrix to have a positive real eigenvalue. Proof. Recall the characteristic coecients derived from (2.4), (2.5), (2.6), that is, ck =
Lk =(L1 , ,L )
(1) P (1 , A) P (2 , A) P ( , A),
(2.8)
where the simple Jacobian loops Li , i = 1, , are dened by the cyclic permutations i , i = 1, , . Now assume that the region has a negative loop equivalence class, i.e., there c is no positive simple loop in its loop structure. So every simple loop Li dened by i k is negative. Therefore the corresponding nonzero loop product P (i , A) is also negative. Then for a composite loop Lk = (L1 , , L ) we have sign(1) P (1 , A) P (2 , A) P ( , A) = sign((1)2 ) = +. (2.9)
Thus all the characteristic coecients ck are positive. This entails a characteristic polynomial of degree n with only positive coecients. By Descartes rules of sign it cannot have a positive real root. Hence the claim. 3. Loop Stability The loop stability refers to the invariance of the jacobian spectrum under any variation of entries that leave unchanged its loop structure. Jacobian loops and their combinations provide valuable information about the stability of a system even when only the signs, not the magnitudes of the Jacobian terms, are known. Recall that A denotes the spectrum of matrix A, i.e., the set of all eigenvalues of A or zeros of the characteristic polynomial CA (). We summarize the classic characterizations of stability from Routh-Hurwitz and Lyapunov theories. For more details see [1,3,5].
361
Linear Stability Criteria. The necessary and sucient conditions to have all real parts negative are given by: (1) (Lyapunov) There exist a positive denite symmetric matrix Q such that QA+At Q is a negative denite matrix. (2) (Routh-Hurwitz) All the Hurwitz determinants Hi are positive, where H1 =c1 , . . . c1 1 0 Hn = . . . 0 c3 c2 c1 . . . 0 c5 c4 c3 . . . 0 c2n1 c2n2 c2n3 = cn Hn1 , . . . cn for all k, and alternate Hurtwitz determinants up to (3-1) cj = 0, j > n.
(3) (Linard-Chipart) ck > 0, e order n are positive.
Denition 3.1. (1) The matrix A is stable (resp. asymptotically stable) if its characteristic polynomial CA () is stable (resp. strongly stable), that is, for every eigenvalue of A we have Re () 0 (resp. < 0). (2) A is unstable if it is not stable. In other words, there is at least one eigenvalue such that Re () > 0. (3) The matrix A is of saddle-type if Re () > 0 for some eigenvalues, and Re () < 0 for the remainings. Therefore a sink equilibrium has a asymptotically stable Jacobian, whereas a source has an unstable Jacobian. Moreover, if Re () > 0 for every A the instability is said to be strong, e.g., a source is strongly unstable. A weak instability is characterized by Re () > 0, for some eigenvalues, and Re () = 0 for the remainings. Recall that, as a function of the matrix, A is neither additive nor multiplicative. Moreover we have the followings: 1 (1) For a nonsingular matrix A, i.e., |A| = det(A) = 0, A1 = { , A }. So A 1 and A must be stable simultaneously. (2) A = At , At denotes the transpose of matrix A. Thus A and At must be stable simultaneously. Denition 3.2. (1) The matrix A is loop stable (resp. asymptotically stable) if every matrix in the loop equivalence class A is stable (resp. asymptotically stable) . (2) A is loop unstable if every matrix in the loop equivalence class A is unstable. We note that a loop equivalence class A that is not stable is not necessarily unstable. Instead it does have at least one matrix that is unstable, i.e., with an eigenvalue of positive real part in its spectrum. The loop analysis is addressed here for irreducible matrices. Recall a matrix A = [aij ]1i,jn is irreducible or indecomposable if there is no simultaneous row-and-column
362
Bourama Toni
permutation Pr Pc such that A is similar to Pr APc = B C O D . (3-2)
where Pr and Pc are respectively the row and column permutation matrices, and B, C are respectively a pp and a qq block such p+q = n, and O a pq block of zeros. By a Laplace decomposition the spectrum of A is given by A = B + D , thus reducing its eigenvalue analysis to that of the individual diagonal block of lower dimension. Therefore we assume all matrices are irreducible without loss of generality. Actually a Pr Pc permutation amounts to a renumbering of the system variables, and renumbering should certainly not aect the properties of the system in general, and its asymptotic behavior in particular. Theorem 2.2 in section 2. yields in the light of the Linear Stability criteria Theorem 3.4. All kth order Feedback Fk positive is a necessary condition for a stable loop equivalence class. Moreover we obtain Theorem 3.5. If the loop structure L has a positive simple loop Lk then the corresponding loop equivalence class cannot be stable. Proof. Indeed it suces to construct a representative matrix with k eigenvalues arbitrarily close to the k-th roots of unity by continuity, and one of them is a simple positive eigenvalue. Therefore the matrix is unstable. Hence the claim. Similarly one can prove Theorem 3.6. The loop equivalence class is unstable if it has a composite loop Ln of the dimension of the system, positive for n even, and negative for n odd. Proof. From section 1. for any matrix A in such a loop structure, the presence of a composite loop of the dimension of the system ensures a nonzero determinant such that the characteristic coecient cn = Fn = CA (0) is > 0 for n even, and < 0 for n odd. This entails the characteristic polynomial of any matrix in the class has a positive root. Hence the claim. The above Linear Stability criteria yields the following condition in terms of Jacobian loops Theorem 3.7. If an irreducible Jacobian matrix J is loop stable then the loop structure L has at least one simple loop L1 , and at least one proper composite loop Ln of the system dimension, and all simple Jacobian loops Lk , k 2 are negative. We therefore note that negative loops promote stability whereas positive loops promote instability. A positive proper composite loop contributes a negative term to the characteristic polynomial, whereas a negative proper composite contributes a positive term. Now assume that for the parameter vertor K = (k1 , . . ., kN ) the Jacobian matrix is loop stable for the parameter value K = K , and loop unstable for K = K . The system then undergoes an exchange of loop stability at K . Hopf bifurcation is known to ensure the existence of limit cycles or sustained oscillations. A system admits a Hopf bifurcation only if the corresponding Jacobian matrix admits and exchange of stability, that is, Lemma 3.8. In order to admit a Hopf bifurcation, the Jacobian matrix can neither be loop stable nor loop unstable throughout the entire region of the phase space.
363
3.1. Loop Analysis in the Plane. For a 2-dimensional system, we prove the following results. Theoreom 3.9. Any loop structure in the plane consisting of two 1-loop L1 of opposite signs and a negative 2-loop cannot be loop stable or loop unstable. The proof is based on the following lemma: Lemma 3.10. The logical equivalence class given by = + + , (3-3) such A is
cannot be not stable and cannot be not unstable, that is, there is a matrix A unstable or stable.
Proof. The equivalence class above is the so-called 1-striped sign pattern; as such, given any monic quadratic polynomial q(x) = x2 + bx + c, there is matrix M whose characteristic polynomial CA () = q(). Therefore there is certainly one whose spectrum contains an eigenvalue with a positive (resp. negative) real part. Hence the claim. Proof of Theorem 3.9. The proof of thereom 3.9 follows immediately; indeed by combinations such as negation, transposition, permutation, signature similarity, one can construct a matrix M with the loop structure of two 1-loop of opposite signs and a negative 2-loop in a such way that the matrix M is in class of the type in lemma 3.10. Based on the Linear Stability Theory, we can classify the loop equivalence classes in the plane as follows: Theorem 3.11. (1) The loop structure consists of only a negative 2-loop L : The dynamic if that of a 2 linear center, that is, a family of periodic orbits surrounding the origin. (2) The loop structure L consists of a negative 2-loop L and two 1-loop of same sign; 2 the dynamic is that of a focus point, that is, the presence of sustained oscillations around the origin. If the system is bounded then there exists a limit cycle surrounding the origin. (3) The loop structure consists of only two 1-loops of opposite sign (resp. same sign). Then the origin is a saddle point (resp. a node stable for negative 1-loops, and unstable for positive 1-loops). (4) These loop equivalence classes are the only loop equivalence classes in the plane. Proof. Immediate from Linear stability theory. Indeed every 2 2 matrix is amenable, via nonsingular linear transformations to one of the forms: a 0 0 b , a b b a . (3-4)
Therefore we have Corrollary 3.12. A negative 2-loop L is a necessary condition for any periodic behavior 2 such as a center, a focus, or a limit cycle, i.e., an isolated periodic orbit.
364
Bourama Toni
3.2. Loop Analysis for 3-dimensional System. Similarity to the 2-dimensional case, we prove Theorem 3.13. Any 3-dimensional loop structure containing the two negative 2-loops L xy and L , the positive 2-loop L+ , in the absence of any positive 1-loop L+ and negative yz xz y 1-loops L and L is destabilized by the two positive 3-loops whenever they are present. x z Proof. Indeed such a loop may be represented by a logical class whose is similar by the usual operations of signatures and permutations, to +/0 + + /0 . (3-5) + + +/0 One can easily construct a matrix in this class with a positive eigenvalue. Theorem 3.14. For a 3-dimensional sytem the following four loop equivalence classes cannot be stable. (1) The loop structure consists of two 1-loops L of opposite signs, and two negative 1 2-loops L , no 3-loop. 2 (2) The loop structure consists of one positive 1-loop L , and two negative 1-loops x L1 , one negative 2-loop L , and one positive 2-loop L+ , but no 3-loop. 2 2 (3) The loop structure consists of two 1-loops L of opposite signs, and one negative 1 2-loop L , and a positive 3-loop L+ . xyz 2 (4) The loop structure consists of two 1-loops L of opposite signs, one negative 2-loop 1 L , one positive 2-loop L+ , and a negative 3-loop L . xyz 2 2 (5) These loop classes cannot be unstable, and are the only ones that cannot be stable and unstable. (6) These equivalence classes are the ones necessary for any Hopf bifurcation ensuring the existence of limit cycles. Proof. We consider in each case a representative logical equivalence class, that is, a signpattern with such a loop pattern, and then use some results for advanced matrix theory on the so-called arbitrary spectrally sign pattern. By signatures and permatutions, a representative sign-pattern of the classes above is similar respectively the following sign-patterns well-known to be spectrally arbitrary, that is, they contains a matrix whose spectrum has a positive eigenvalue, root of an arbitrary monic cubic polynomial. (1) The rst case corresponds to the sign-pattern: + 0 + 0 (3-6) 0 + (2) Case Two: + + + 0 + 0 (3) Case three: + 0 + 0 + 0 (3-8) (3-7)
365
(4) Case four: + + + + 0 0 (3-9)
The previous results entail the following conjecture illustrated in the jacobian loop analysis of the well-known Lorenz and Rossler systems, paradigms of chaotic dynamics. It is a step toward the necessary and/or sucient conditions in terms of Feedback Loops for the onset of chaos. Conjecture 3.15. The previous loop equivalence classes are some of the necessary loop structures for the onset of any chaotic behavior in a 3-dimensional system. This conjecture is a step toward the determination of necessary and/or sucient conditions in terms of Feedback Loops for the onset of chaos, in particular in Lorenz and Rossler systems. These systems are respectively: 3.3. The Lorenz System. The Lorenz system is described by x = (y x) y = x y xz z = z + xy Its equilibria are the origin O = (0, 0, 0) for all values of the parameter , and E = ( ( 1), ( 1), 1) appearing at > 1 and symmetric with respect to the z-axis. A Lorenz logical Jacobian class in the positive quadrant is represented by + 0 s21 (3-11) whereas along the z-axis we have s21 0 + 0 0 0 (3-12) (3-10)
where s21 is respectively (+), (0), and () under, in, and above the plane z = . This implies a variation of a loop structure in these regions. 3.4. Rossler System. The Rossler system is given by x = y z y = x + ay z = bx + xz cz For a wide range of the parameters a, b, c the system exhibits two unstable equilibria de type saddle-focus periodically repulsive (resp. attractive) in a plane while attractive (resp. repulsive) along a normal direction. (3-13)
366
Bourama Toni
For the positive parameters a Rossler logical Jacobian class is given by 0 + + 0 , + 0
(3-14)
that is, the loop structure contains two negative 2-loops L and L along with two 1-loops xy xz of opposite signs L+ and L . Note, as conjectureed, the presence of two proper 3-dimensional y z composite loops L2 given by the negative (L+ , L ) and the positive (L , L ). y xz z xy 3 References
[1] Bellman, R., Introduction to Matrix Analysis, Classics in Applied Matematics 19, 2nd Ed. SIAM, 1997. [2] Eisenfeld, J. and De Lisi, C., On conditions for qualitative instability of regulatory loops with applications to immunological control loops, Mathematics and Computers in Biomathematical Applications, Elsevier, New York (1994), 39-53. [3] Gantmacher, F.R., Thorie des Matrices, vol. 1,2, Collection Universitaire de Mathmatiques, Dunod e e Paris, 1966. [4] Gouz, J.L., Positive and negative circuits in Dynamical Systems, J.Biol.Syst. 6 (1998), 11-15. e [5] Lancaster, P. and Tismenetsky, M., The Theory of Matrices with Applications, Academic Press, 1985. [6] Quirk, R. and Ruppert, R., Qualitative economics and the stability of equilibrium, Rev. Econ. Studies 32 (1965), 311-326. [7] Thomas, R., The role of Feedback Circuits: positive feedback circuits are a necessary condition for positive eigenvalues in the feedback matrix, Ber.Bunzenges.Phys.Chem. 98 (1994), 1148-1151. [8] Toni, B., Jacobian Feedback Loop Analysis I, International Journal of Evolution Equations. 1(4) (2005). [9] Verhulst, F., Nonlinear Dierential Equations and Dynamical Systems, Springer-Verlag Universitext, 1990.
E XISTENCE OF O SCILLATING S OLUTION FOR N ONLINEAR S TATE -D EPENDENT D ELAY D IFFERENTIAL E QUATION
Zhixian Yu and Rong Yuan School of Mathematical Science, Beijing Normal University, Beijing 100875, Peoples Republic of China
Abstract In this paper sufcient conditions for oscillation of all solutions are given for the equations x(t) + pi (t) f (x(t ri (t))) = 0,
i=1 n
and
x(t) + pi (x(t)) f (x(t ri (x(t)))) = 0,

i=1
where ri (), pi () C(R, R+ ).
1991 Mathematics Subject Classication: 34K11 Key words and phrases: Oscillating solution; state-dependent; delay equation.
1.
Introduction
n
In this paper, we will discuss the oscillation of all solutions for the following delay equations x(t) + pi (t) f (x(t ri (t))) = 0,
i=1
(1.1)
and
x(t) + pi (x(t)) f (x(t ri (x(t)))) = 0,

i=1
(1.2)
where ri (), pi () C(R, R+ ), f (x) is continuous. As a usual, a solution of equation (1.1) (or (1.2)) is said to oscillate if it has arbitrarily large zeros.
Supported E-mail
by National Natural Science Foundation of China address: yzx3411422@163.com E-mail address: ryuan@bnu.edu.cn
368
Zhixian Yu and Rong Yuan
where p C([0, ), [0, )), C([0, ), (0, )), limt (t (t)) = , f C(R, R) and u f (u) > 0 for u = 0, and proved that every solution of Eq.(1.3) is oscillatory.
Existence of oscillating solutions has been extensively discussed in the literatures (see [1-14]) and the references therein. This paper is motivated by the papers [5, 9]. Li and Kuang [9] considered the following rst-order nonlinear nonautonomous delay differential equation x (t) + p(t) f (x(t (t))) = 0, (1.3)
In [5], Ladas and Stavroulakis discussed the oscillation of all solutions of differential equations with deviating arguments of the form y (t) + pi y(t i ) = 0
i=1 n n
(1.4)
and
y (t) pi y(t + i ) = 0
i=1
(1.5)
in the condition that
pi
i
1 n
i=1
i
2
1 2
1 > , e 1 > , e
or 1 n
i=1
(pi i )
where i (i = 1, 2, , n) are positive constants, pi (i = 1, 2, , n) are all positive constants. Some authors (for example, [1, 3, 13]) considered the oscillation problem for the delay differential equation x (t) + qi (t)x(t i (t)) = 0,
i=1 n
(1.6)
where pi , i are continuous and positive functions. We would like to point out that our conditions and approaches are different from those in [1, 3, 13]. When we are back to the equations (1.4) and (1.5), our conditions coincide with these in [5]. In the sequel, we assume the following conditions: (H0). f (x)x > 0 for x = 0, 0 < lim infx0 (H1). For Eq.(1.1),
t f (x) x
= .
t
lim pi (t) = pi , lim ri (t) = ri ,
where ri (t) is bounded, ri (t) < 1 and pi , ri are positive constants, i = 1, 2, . . . n. (H1 ). For Eq.(1.2),
Existence of Oscillating Solution lim pi (x) = pi , lim ri (x) = ri ,
369
x0
x0
where ri (x) is bounded, r (x) > 0, and pi and ri are positive constants, i = 1, 2, . . . n. (H2). (n pi ) n (n ri ) > i i=1 (H3).
1 n
1
1 0 e . 1 0 e ,
n (pi ri ) 2 i=1
>
where 0 is a nite positive number with 0 < 0 .
Now we can formulate our main results as follows. Theorem 1.1 Assume that the conditions (H0), (H1), (H2) hold, then all the solutions of Eq.(1.1) are oscillatory. Theorem 1.2 Assume that the conditions (H0), (H1 ), (H2) hold, then all the solutions of Eq.(1.2) are oscillatory. Theorem 1.3 Assume that the conditions (H0), (H1), (H3) hold, then all the solutions of Eq.(1.1) are oscillatory. Theorem 1.4 Assume that the conditions (H0), (H1 ), (H3) hold, then all the solutions of Eq.(1.2) are oscillatory. The remaining parts of this paper are organized as follows. In section 2 we will give the proof of Theorems 1.1 and 1.2. The proof of Theorems 1.3 and 1.4 will be given in section 3. In the last section we will provide some examples.
2.
Proof of Theorems 1.1 and 1.2
We begin with a Lemma. Lemma 2.1 [5] Assume that g(b1 , b2 , , bn ) = log bi
i=1 n
i=1
qi bi
i=1
ri
is nonnegative for bi 1, i = 1, , n, qi > 0, and ri > 0. Then g(b1 , b2 , , bn ) has a maximum at 1 1 1 . , , , q1 n ri q2 n ri qn n ri i=1 i=1 i=1 Moreover,
qi
i=1
1 n
i=1
ri
1 . e
370 Proof of Lemma 2.1. proof is quoted here. Now observe that
Zhixian Yu and Rong Yuan The proof has been given in [5]. For the readers convenience its 1 g = qi bi bi bi = At the critical point 1 1 1 , , , n n q1 i=1 ri q2 i=1 ri qn n ri i=1
n 2 g a2 ai a j = i 2 i, =1 bi b j i=1 bi n
i=1
ri
= 0,
for
1 , qi n ri i=1
i = 1, 2, . . . , n.
the function g(b1 , b2 , , bn ) has a maximum because the quadratic form
is negative denite. Since g(b1 , b2 , , bn ) 0, the maximum of g(b1 , b2 , , bn ) at the critical point should be nonnegative. So we have
i=1
log qi ri
i=1
i=1
1 n
qi ri
n
1 qi n ri i=1
i=1
ri
= n,
that is,
qi
i
i=1
1 . e
This completes the proof.
Proof of Theorem 1.1 In fact we only need to prove that Eq. (1.1) does not have an eventually positive solution x(t) [we can prove in the similar method that Eq. (1.1) does not have an eventually negative solution x(t)]. By contradiction, then x (t) < 0, since x(t) is an eventually positive solution and condition (H0) . So x(t) has a limitation as t . We claim that
t
lim x(t) = l = 0.
Indeed, if l = 0, then we have l > 0 and according to conditions (H0) and (H1), the following inequality 0 = lim x (t)
t
= lim pi (t) f (x(t ri (t)))

t i=1
= pi f (l) < 0
i=1
Existence of Oscillating Solution
371
holds, which implies that l = 0. On the other hand, by (H1), for arbitrary > 0, there is a large enough T such that if t > T, pi < pi (t) < pi + , ri < ri (t) < ri + (2.7) and f (x(t)) > 0 . x(t) i (t) = and bi = lim inf i (t),
t
(2.8)
Set
x(t ri (t)) , x(t)
t > T, i = 1, 2, . . . n
i = 1, 2, . . . n.
Then i (t) > 1 when t > T and bi 1, i = 1, 2, . . . , n, since x (t) < 0. Dividing x(t) as t > T in the both sides of (1.1), we obtain that
n x (t) f (x(t ri (t))) + pi (t)i (t) = 0, x(t) i=1 x(t ri (t))
t > T, i = 1, 2, . . . , n.
(2.9)
Integrating (2.9) over the intervals [t rk (t),t], k = 1, 2, . . . , n, then we get log x(t) log x(t rk (t)) + that is, log k (t)
pi (s)i (s) tr (t)

k
pi (s)i (s) tr (t)

k
i=1
Now we will consider two cases: Case 1: All numbers bi , i = 1, 2, . . . , n are nite. By (2.10) and (2.8) for large enough t (t > T ), we have log k (t) (0 )
t
trk (t) i=1
according to (H0), limt x(t) = 0 and the boundedness of ri (t). Therefore, combining (2.7) and (2.11), we obtain log bk (0 ) (pi )(rk )bi .
i=1 n
i=1
f (x(s ri (s))) ds = 0, x(s ri (s))
f (x(s ri (s))) ds = 0, x(s ri (s))
k = 1, 2, . . . , n.
(2.10)
pi (s)i (s) ds,
k = 1, 2, . . . , n.
(2.11)
(2.12)
372
Zhixian Yu and Rong Yuan Letting 0, then we have log bk 0 pi rk bi ,

i=1 n
k = 1, 2, . . . n,
(2.13)
which implies that

k=1
log bk 0
i=1
pi bi
k=1
rk
(2.14)
By inequality (2.14) and Lemma 2.1 where we take qi = 0 pi , we have the following inequality
n n n n
k=1
log 0 pk ri
i=1
k=1
0 pk ri
n
1 0 pk n ri i=1
n
k=1
rk
= n.
(2.15)
Thus,
0 pi
i=1
i=1
1 n
1 , en
that is,
pi
i
i=1
ri
1 , 0 e
which contradicts (H2). Therefore Eq. (1.1) could not have an eventually positive solution in case 1. Case 2: At least one of the numbers bi , i = 1, 2, . . . n is innite. Without loss of generalization, we assume that for some i {1, 2, , n}, x(t ri (t)) = +. t x(t) lim According to (H0), for large enough t1 , we have that if t > t1 , x (t) + pi (t) f (x(t ri (t))) < 0. and f (x(t)) > 0 . x(t) Thus, for any t > t1 , we have the following inequality x (t) + (0 )pi (t)x(t ri (t)) < 0. Since limt pi (t) = pi , we have lim
(2.16)
(2.17)
t tri (t)
pi (s)ds = pi ri > 0.
Existence of Oscillating Solution So there exist d > 0 and t2 > 0 such that
t tri (t)
373
pi (s) d
fort > t2 .
Then for any t > t2 , there is (t) > t such that

(t) t
where i (t) = t ri (t). Let t3 = max{t1 , t2 }. For any t > t3 , integrating (2.17) over the interval [t, (t)] and [i ((t)), t] respectively, we nd x((t)) x(t) + (0 ) and x(t) x(i ((t))) + (0 )
(t) t
From x (t) < 0, i (t) = 1 ri (t) > 0, (2.20) and (2.21), we obtain x((t)) x(t) + (0 ) and x(t) x(i ((t))) + (0 )
(t) t
for t > t3 . Since x(t) > 0, omitting the rst terms in (2.22) and (2.23), we conclude x(t) + (0 ) and x(i ((t))) + (0 )
(t) t
pi (s)ds x(i ((t))) < 0
t i ((t))
for t > t3 . From (2.18), (2.19), (2.24) and (2.25), it follows d x(t) + (0 ) x(i ((t))) < 0 2 and d x(i ((t))) + (0 ) x(i (t)) < 0 2 (2.26)
and
d pi (s)ds . 2 i ((t))
pi (s)ds =
d 2
(2.18)
(2.19)
pi (s)x(i (s))ds < 0

t i ((t))
(2.20)
pi (s)x(i (s))ds < 0.
(2.21)
pi (s)ds x(i ((t))) < 0

t i ((t))
(2.22)
pi (s)ds x(i (t)) < 0
(2.23)
(2.24)
pi (s)ds x(i (t)) < 0
(2.25)
(2.27)
374
for t > t3 . From (2.26) and (2.27), we obtain d d2 x(t) > (0 ) x(i ((t))) > (0 )2 x(i (t)). 2 4 Then 4 x(i (t)) < < x(t) (0 )2 d 2
which is a contradiction to the assumption of case 2. From the above two cases, we conclude that equation (1.1) could not have an eventually positive solution. This completes the proof.
Proof of Theorem 1.2 In fact we only need to prove that Eq. (1.2) does not have an eventually positive solution x(t) [we can similarly prove that Eq.(1.2) does not have an eventually negative solution x(t)]. By contradiction, then x (t) < 0, since x(t) is an eventually positive solution and condition (H0). Thus, limt x(t) exists. We claim that
t
lim x(t) = l = 0.
Indeed, if l = 0, then we have l > 0 and according to conditions (H0) and (H1 ), the following inequality 0 = lim x (t)
t
= lim pi (x(t)) f (x(t ri (x(t))))

t i=1
= pi (l) f (l) < 0

i=1
holds, which implies that l = 0. By using the same notations as in the proof of Theorem 1.1, we also consider two cases: Case 1: All numbers bi , i = 1, 2, . . . , n are nite. Case 2: At least one of the numbers bi , i = 1, 2, . . . , n is innite. For case 1, the proof is similar to the proof of Theorem 1.1. So we omit it. Thus we only need to consider case 2, that is, for some i {1, 2, , n}, x(t ri (x(t))) = +. t x(t) lim From (H0), for large enough t1 , we have x (t) + pi (x(t)) f (x(t ri (x(t)))) < 0 and f (x(t)) > 0 x(t) (2.28)
Existence of Oscillating Solution for t > t1 . Thus, for any t > t1 , we have the following inequality x (t) + (0 )pi (x(t))x(t ri (x(t))) < 0. Since limt x(t) = 0 and limt pi (x(t)) = pi , we have lim
375
(2.29)
t tri (x(t))
pi (x(s))ds = pi ri > 0.
So there are d > 0 and t2 > 0 such that
t tri (x(t))
pi (x(s)) d
fort > t2 .
Then for any t > t2 , there is (t) > t such that

(t) t
where i (t) = t ri (x(t)). Claim: i (t) > i (s) for t > s, where s is large enough. In fact, for t > s, s is large enough, we have i (t) i (s) = t ri (x(t)) [s ri (x(s))] = (t s) ri ()x ()(t s)
where x(t) < < x(s) and s < < t. The inequality of the fourth line follows from r (x) > 0 and x (t) < 0. Let t3 = max{t1 , t2 }. For any t > t3 , integrating (2.29) over the interval [t, (t)] and [i ((t)), t] respectively, we nd x((t)) x(t) + (0 ) and x(t) x(i ((t))) + (0 )
(t) t
Combining our claim and (2.32), (2.33), we obtain x((t)) x(t) + (0 )

(t) t
and
d pi (x(s))ds , 2 i ((t))
pi (x(s))ds =
d 2
(2.30)
(2.31)
= (t s) [ri (x(t)) ri (x(s))]
= (t s)[1 ri ()x ()] > 0,
pi (x(s))x(i ((s)))ds < 0

t i ((t))
(2.32)
pi (x(s))x(i (s))ds < 0.
(2.33)
pi (x(s))ds x(i ((s))) < 0
(2.34)
376 and
Zhixian Yu and Rong Yuan x(t) x(i ((t))) + (0 )
t i ((t)) t i ((t))
pi (x(s))ds x(i (t)) < 0.
(2.35)
for t > t3 . Since x(t) > 0, omitting the rst terms in (2.34) and (2.35), we conclude x(t) + (0 ) and x(i ((t))) + (0 )
(t) t
for t > t3 . Combining (2.30), (2.31, (2.36) and (2.37), we have d x(t) + (0 ) x(i ((s))) < 0 2 and d x(i ((t))) + (0 ) x(i (t)) < 0 2 for t > t3 . From (2.38) and (2.39), it follows d d2 x(t) > (0 ) i ((t)) > (0 )2 x(i (t)). 2 4 Then 4 x(i (t)) < < , x(t) (0 )2 d 2 (2.38)
which contradicts to the assumption of case 2. From the above two cases, we obtain that equation (1.2) could not have an eventually positive solution. This completes the proof. Remark: If we use the condition i (t) := t ri (t) = max[0,t] {s ri (s)} instead of ri (t) < 1, then Theorem 1.1 still holds. Theorem 1.2 also holds if we replace r (x) > 0 by r (x) bounded or r (x) > 0 for |x| < since limt x (t) = 0. Corollary 2.1 Let (H0 ).
f (x) x
0 for x = 0, where 0 < 0 < .
Then (i) Assume that (H0 ), (H1), (H2) hold, each solution of (1.1) is oscillatory. (ii) Assume that (H0 ), (H1 ), (H2) hold, each solution of (1.2) is oscillatory.
pi (x(s))ds x(i ((s))) < 0
(2.36)
pi (x(s))ds x(i (t)) < 0.
(2.37)
(2.39)
377
Proof. We will obtain this result in the similar way as in Theorem 1.1. In fact, note that we neednt use limt x(t) = 0 when (H0 ) holds for equation (1.1). We will omit the details.
Theorem 2.1 Assume (H1), (H2) and (H0 ). Each solution of the advanced delay differential equation x(t) pi (t) f (x(t + ri (t))) = 0,
i=1 n
(2.40)
is oscillatory. Proof. We neednt use limt x(t) = 0 when (H0 ) holds for the advanced equations. Thus, we can prove Theorem 2.1 in view of the similar technique. Remark. We consider the advanced equation x(t) pi (x(t)) f (x(t + ri (x(t)))) = 0,
i=1 n
(2.41)
corresponding to(1.2). Note x (t) > 0 when x(t) > 0 for (2.41). So we cant prove limt x(t) = 0 in the condition (H0 ). Thus, we cant conclude if each solution of (2.41) is oscillatory in the similar method. Corollary 2.2 Let (H2 ). n (n pi ) n r > i=1 (H2 ). (n ri ) pi > i=1
1
1 0 e ,
where r = min{r1 , r2 , . . . , rn }
1 0 e ,
where p = min{p1 , p2 , . . . , pn }.
(i) Assume that (H0) or (H0 ), (H1), (H2 ) hold, each solution of (1.1) is oscillatory. (ii) Assume that (H0) or (H0 ), (H1), (H2 ) hold, each solution of (1.1) is oscillatory. (iii) Assume that (H0) or (H0 ), (H1 ), (H2 ) hold, each solution of (1.2) is oscillatory. (iv) Assume that (H0) or (H0 ), (H1 ), (H2 ) hold, each solution of (1.2) is oscillatory.
Corollary 2.3 Assume (H0 ), (H1), in addition, one of the conditions (H2 ) and (H2 ). Then all of solutions of the advanced equation (2.40) oscillate.
3.
Proof of Theorems 1.3 and 1.4
We only prove Theorem 1.3. Theorem 1.4 can be proved similarly. So we omit it. We begin with another lemma. Lemma 3.1 For t > 0, then
logt t
1. e
378
Proof. Let f (t) = logt . Observe that f (t) = 1logt = 0 for t = e. Since f (t) 0 for t t2 0 < t e and f (t) < 0 for t > e, f (t) has the maximum at t = e. This implies that logt 1 . t e Proof of Theorem 1.3. Similar to the proof of Theorem 1.1, we only need to prove that Eq. (1.1) does not have an eventually positive solution x(t) [we can prove in the similar method that Eq. (1.1) does not have an eventually negative solution x(t)]. By contradiction, we also consider two cases: Case 1: All numbers bi , i = 1, 2, . . . n are nite, where bi is as in the proof of Theorem 1.1. Case 2: At least one of the numbers bi , i = 1, 2, . . . n is innite. We only consider case 1: bi < [we can prove Theorems 1.3 and 1.4 in the similar to the method of Theorems 1.1 and 1.2 for bi = ]. Under the the conditions (H0) and (H1), we must have (2.13). By Lemma 3.1, we obtain log bk 1 f or bk 1. (3.42) bk e Combining (2.13) and (3.42), we nd
n 1 bi 0 pi rk , e bk i=1
(3.43)
which implies that

n n bk bi n pi ri + (pi rk + pk ri ). 0 e i=1 bk bi i,k=1
i<k
Now using the result that pi rk we have

1 bi bk + pk ri 2(pi pk ri r j ) 2 , bk bi
n n n 1 1 n pi ri + 2 (pi pk ri rk ) 2 = ( (pi ri ) 2 )2 ), 0 e i=1 i,k=1 i=1

i<k
which contradicts (H3). This completes the proof.
Corollary 3.1 (i) Assume that (H0 ), (H1), (H3) hold, each solution of (1.1) is oscillatory. (ii) Assume that (H0 ), (H1 ), (H3) hold, each solution of (1.2) is oscillatory. Proof. We will obtain this result which is the similar to the proof of Theorem 1.3. Corollary 3.2 Assume (H0 ), (H1) and (H3), each solution of the advanced delay differential equations (2.40) is oscillatory.
379
Proof. We obtain easily the corollary according to the process of theorems 1.1 and 1.3. Corollary 3.3 Let (H3 ). pi ri > (H3 ).
1 n 1 0 e ,
for some i {1, 2, . . . n}.

1 0 e .
(n ri pi ) > i=1
Then, (i) Assume that (H0) or (H0 ), (H1), (H3 ) hold, each solution of (1.1) is oscillatory. (ii) Assume that (H0) or (H0 ), (H1), (H ) hold, each solution of (1.1) is oscillatory. (iii) Assume that (H0) or (H0 ), (H1 ), (H3 ) hold, each solution of (1.2) is oscillatory. (iv) Assume that (H0) or (H0 ), (H1 ), (H3 ) hold, each solution of (1.2) is oscillatory. Corollary 3.4 Assume (H0 ), (H1), in addition, one of the conditions (H3 ) and (H3 ). Then all of solutions of the advanced equation (2.40) oscillate. Remark 3.1 When pi (t), ri (t), pi (x) and ri (x) are constants for i = 1, 2, . . . , n and add f (x) = x, it is obvious that Theorems 1.1, 1.2, 1.3 and 1.4 coincide with those of G.Ladas and I. P.Stavroulakis [5].
4.
Examples
1 arctan2 t))) +
Example 4.1. We consider x (t) + (1 + arctan2 t) arctan x(t (3 + (2 + arctan2 t)) arctan x(t (1 +
1 arctan2 t))) = 0,
1 where p1 (t) = 1 + arctan2 t, p2 (t) = 2 + arctan2 t, r1 (t) = 3 + arctan2 t, r2 (t) = 1 + 1 2 t and f (x) = arctan x. It is easily checked that all the conditions of Theorems arctan 2 1.1 and 1.3 are satised for the above delay differential equation. In fact, p1 = 1 + , p2 = 4 2 2 + , r1 = 3 + , r2 = 1 + , r (t) < 1, limx0 f (x) = 1 > 0, 4 4 4 x
pi
i=1
i=1 2
1 2
ri
>4>
1 e
and 1 2 Example 4.2. We consider

i=1
(ri pi )
5+2 6 1 > . > 2 e
+(2 + arctan2 x(t) arctan(x(t (1 + 2 arctan2 x(t)))) = 0,
x (t) + (1 + arctan2 x(t)) arctan(x(t (3 + arctan2 x(t))))
where p1 (x) = 1 + arctan2 x, p2 (x) = 2 + arctan2 x, r1 (x) = 3 + arctan2 x, r2 (x) = 1 + 2 arctan2 x and f (x) = arctan x. It is also obvious that all the conditions of Theorems 1.2 and 1.4 are satised. We omit the details.
380
References
[1] Y. Cheng, Oscillation in nonautonomous scalar differential equations with deviating arguments, Proc. Amer. Math. Soc. 110 (1990), 711-719. [2] M. Elabbasy, A. S. Hegazi, S. H.Saker, Oscillation of solutions to delay differential equations with positive and negative coefcients. Electron. Electron. J. Differential Equations 2000, No. 13, 13 pp. (electronic). [3] B.R. Hunt, J.A. Yorke,When all solutions of x = qi (t)x(t Ti (t)) oscillate. J. Differential Equations. 53 (1984), 139-145. [4] J. Jiang, Oscillation criteria for rst order nonlinear delay differential equations. Hiroshima Math. J. 31 (2001), 467-476. [5] G.Ladas and I. P.Stavroulakis, Oscillations caused by several retarded and advanced arguments. J. Differential Equations. 44 (1982), 134-152. [6] G.Ladas, Y. G.Scas and I. P.Stavroulakis, Necessary and sufcient conditions for oscillations. Amer. Math. Monthly. 90 (1983),637-640. [7] B. Li, Oscillation of rst order delay differential equations. Proc. Amer. Math. Soc. 12 (1996), 3729-3737. [8] B. Li, Oscillations of delay differential equations with variable coefcients. J. Math. Anal. Appl. 192 (1995), 312-321. [9] B. Li and Y. Kuang, Sharp conditions for oscillations in some nonlinear nonautonomous delay differential equations. Nonlinear Anal. 29 (1997), 1265-1276. [10] Xianyi Li and Deming Zhu, Oscillation and nonoscillation of advanced differential equations with variable coefcients. J. Math. Anal. Appl. 269 (2002), 462-488. [11] Y. Kuang, and H. L. Smith, Slowly oscillating periodic solutions of autonomous statedependent delay equations. Nonlinear Anal. 19 (1992), 855-872. [12] M. R. S.Kulenovi , G.Ladas, A.Meimaridou, Necessary and sufcient condition for c oscillations of neutral differential equations. J. Austral. Math. Soc. Ser. B 28 (1987), 362-375 [13] M. K. Kwong, Oscillation of rst-order delay equations. J. Math. Anal. Appl. 156 (1991), 274-286. [14] X. H.Tang, Oscillation of rst order delay differential equations with distributed delay. J. Math. Anal. Appl. 289 (2004), 367-378.
S EMILINEAR A BSTRACT D IFFERENTIAL E QUATIONS WITH D EVIATED A RGUMENT

Ciprian G. Gal Morgan State University, Department of Mathematical Sciences, Baltimore, Md, 21251, U.S.A.
Abstract In this paper we prove local and global existence results for semilinear differential equations with deviated argument in Banach spaces.
1991 Mathematics Subject Classication: 34G10. Key words and phrases: mild solutions, semigroups of linear operators, semilinear differential equations with deviated arguments.
1.
Introduction
In the very recent papers [2] and [3], the existence and uniqueness of almost automorphic mild solutions with values in Banach spaces X and in locally convex spaces, respectively, for the differential equation with deviated argument u (t) = Au(t) + f (t, u(t), u[(u(t),t)]), t R, were proved, where A is the innitesimal (bounded) generator of a C0 -semigroup of operators (T (t))t0 on a Banach space, satisfying an additional exponential-type conditions of stability of the form ||T (t)|| Ket , t R+ with K > 0, < 0 and (y,t) : X R R is an almost automorphic function with respect to t and Lipschitz in y. These results generalize some known classical results in the literature [4] , [5] , [6] . The goal of the present note is to prove the existence and uniqueness of local and global solutions for the initial value problem with deviated argument u (t) = Au(t) + f (t, u(t), u[(u(t),t)]), t R+ , u (0) = u0 , but without to impose any additional assumption on the semigroup (T (t))t0 . These results generalize in some cases (that is, when A is bounded), the classical ones with usual (non-deviated) argument in [1].
E-mail
address: cgal@morgan.edu
382
Ciprian G. Gal
2.
Basic Results
Let (X, ) be a Banach space. We consider the nonlinear differential equation with deviated argument u (t) = Au(t) + f (t, u(t), u[(u(t),t)]), t R+ , u (0) = u0 , (2.1)
where A is a bounded linear operator in X that generates a (C0 ) semigroup {T (t)}tR+ on X. It is easy to prove that any solution of this problem has the form u(t) = T (t)u0 + and we refer to any continuous u CL (R+ , X) = {u () C (R+ , X) : ||u(t) u(s)|| L|t s|, t, s R+ } satisfying the relation (2.2) as a mild solution of the above problem . Obviously, because of the absence, in general, of its differentiability, a mild solution is not a strong solution of the problem (2.1). A mild solution is thus a kind of generalized solution. We have the following local existence result for (2.1). Theorem 2.1 Let u0 X and let f : R+ X R+ X be continuous in all of its variables and satisfy the Lipschitz condition: for each > 0, there is a K = K () > 0 such that || f (t, u1 , v1 ) f (t, u2 , v2 )|| K(||u1 u2 || + ||v1 v2 ||), u1 , u2 , v1 , v2 X, whenever 0 t . Moreover, assume : X R+ R+ is globally continuous on X R+ and satises (, 0) = 0 and for each > 0, there is a L = L () > 0 such that |(u,t) (v,t)| L ||u v||, u, v X, whenever 0 t . Then, for > 0 sufciently small , (2.1) has a unique solution in CL ([0, ] , X) = {u () C ([0, ] , X) : ||u(t) u(s)|| L|t s|, t, s [0, ]} , for some suitable L = L () > 0. Proof. The idea of proof is that of Theorem 2.4, pp. 88-89 in [1], modied correspondingly to the deviated argument. Let > 0 and consider Y = CL ([0, ] , X). Let E be a closed neighborhood of u0 in X. Dene S by (Su) (t) = T (t)u0 +
t 0
T (t s)[ f (s, u(s), u((u(s), s)))]ds, for t [0, ]
and u M = {v () Y : v (0) = u0 , v [0, ] E} . Note that M is a complete metric space and Su C ([0, ] , X) if u C ([0, ] , X) . We show that Su Y, for a suitable
T (t s)[ f (s, u(s), u((u(s), s)))]ds,
(2.2)
(2.3)
Semilinear Abstract Differential Equations with Deviated Argument
383
constant L = L () > 0. Let u0 X. Since A is bounded, then by semigroup thet ory u (t) = T (t) u0 C1 ([0, ] , X) , so that it is enough to show that v (t) = 0 T (t s)[ f (s, u(s), u((u(s), s)))]ds C1 ([0, ] , X) . It will follow Su CL ([0, ] , X) . We have v (t) = T (0)[ f (t, u(t), u((u(t),t)))] + Av (t) , since A is linear and closed. Let M, such that T (t) Met , for t R+ . Using the assumptions on f , we have for t [0, ] : ||v (t) || sup
t[0,]
f (t, u(t), u((u(t),t))) + sup Av (t)

t[0,]
sup
t[0,]
f (t, u(t), u((u(t),t))) f (0, u0 , u0 ) + sup

t[0,]
f (0, u0 , u0 )
+ sup
||AT (t s)[ f (s, u(s), u((u(s), s)))]||ds, f (t, u0 , u0 )
t[0,] 0
(1 + Me |||A|||) sup
t[0,]
+K () sup ( u(t) u0 + u((u(t),t)) u ((u0 , 0)) ) (1 + Me |||A|||) := M f () ,

t[0,]
which exists because u CL ([0, ] , X) . Here ||| ||| stands for the operator norm. Estimate (2.4) yields for t [0, ] , > 0 sufciently small, sup
t[0,]
d [Su (t)] sup T (t) Au0 + sup v (t) dt t[0,] t[0,] Me Au0 + M f () . (2.5)
Therefore, by the mean value theorem in vector spaces, Su CL ([0, ] , X) , if L is chosen such that L() Me Au0 + M f () (2.6) for sufciently small < 1. Thus, for such chosen L(), the map S is well dened. Next, we show that S (M ) M . For this, we assume without loss of generality that E is bounded. Finally, we have Su u0
Y
= sup Su (t) u0
t[0,]
sup T (t) u0 u0 + sup v (t) := J1 () + J2 () .

t[0,] t[0,]
While J1 () 0 as 0, calculating as in (2.4) and using the assumptions on the functions f , , we also obtain the following estimate for J2 () Me sup
t[0,]
f (t, u (t) , u ( (u (t) ,t)))
(2.4)
384 Me Me sup
t[0,]
Ciprian G. Gal f (t, u0 , u0 ) + sup

t[0,]
f (t, u(t), u((u(t),t))) f (0, u0 , u0 )
sup
t[0,]
f (t, u0 , u0 ) + K () (1 + L () L ()) sup u(t) u0

t[0,]
Thus, it is clear J2 () 0 as 0+ . It follows that S maps bounded sets of M into the same bounded sets (of M ), provided that is sufciently small. It remains to show that S is a contraction. We have from (2.3) : Su Sw Me K ()
Y
= sup Su (t) Sw (t)

t[0,]
Me K ()
Me K ()
Me K ()
Me K () (L () L () + 2) u w
Here obviously we can assume, without loss of generality, that K(), L(), L () are nondecreasing as functions of , which implies that we have Me K () (L () L () + 2) 0 as 0+ . Therefore S is a contraction as desired, for sufciently small, that is, for Me K () (L () L () + 2) < 1. This implies L() < 1 Me K() 2 1 . L (2.8)
Taking into account the inequality satised by L() above in the proof, concerning the fact that S is a well dened map, it follows from (2.6) and (2.8) that the following inequality must hold: 1 1 Me ||Au0 || + M f () < L() < 2 . (2.9) Me K() L Now, since M f () is non-decreasing as function of , we get that Me ||Au0 ||+M f ()
1 M||Au0 || + M f (0+), as 0+ and also Me K() 2 L1 + as 0+ , which implies that for sufciently small < 1, the constant L() exists. Thus, for such a constant
( u (s) w (s) + u ( (u (s) , s)) w ( (w (s) , s)) ) ds
( u (s) w (s) + u ( (u (s) , s)) u ( (w (s) , s)) + u ( (u (s) , s)) w ( (w (s) , s)) ) ds
2 sup u (r) w (r) + L () | (u (s) , s) (w (s) , s)| ds

0rs 0
(L () L () + 2) sup u (r) w (r) ds

0rs Y
(2.7)
Semilinear Abstract Differential Equations with Deviated Argument
385
satisfying (2.9), S is a well dened map from CL() ([0, ] , X) to CL() ([0, ] , X) if is sufciently small. The contraction property of S was veried in (2.7) . Hence, the Banach xed point theorem implies that the desired solution of (2.1) is the xed point of S. The proof is complete. Furthermore, we have the following global existence theorem for problem (2.1) . Theorem 2.2 We have the same assumptions as in Theorem 2.1. Then, the initial value problem (2.1) has a unique solution in CL ([0, ) , X) C1 ((0, ) , X). Proof. The idea of proof is that of Theorem 2.5, pp. 89-90 in [1], modied correspondingly to the deviated argument. Let S be as in (2.3) . We claim that [Met K (t) (L (t) L (t) + 2)t]n S u (t) S w (t) sup u (s) w (s) , n! s[0,t]
n n n for all u, w CL ([0,t] , X) , for all t > 0. Let us denote MS (t) by n MS (t) :=
(2.10)
[Met K (t) (L (t) L (t) + 2)]n > 0, n!
for t > 0, n N. We may assume without loss of generality that

n t K (t) , L (t) , L (t) , MS (t)
(for all n) are monotone nondecreasing functions in R+ . For n = 1, (2.10) holds because of the estimate in (2.9) . We prove our claim using induction on n. Assume (2.10) is true. We have Sn+1 u (t) Sn+1 w (t) =
0 t
T (t s)[ f (s, Sn u(s), Sn u((Sn u(s), s))) f (s, Sn w(s), Sn w((Sn w(s), s)))]ds
0 t
Met K (t)
( Sn u (s) Sn w (s) + Sn u ( (Sn u (s) , s)) Sn w ( (Sn w (s) , s)) ) ds

t 0
Met K (t)
Met K (t)
Met K (t)
n Met K (t) MS (t) (2 + L (t) L (t)) u w
( Sn u (s) Sn w (s) + Sn u ( (Sn u (s) , s)) Sn u ( (Sn w (s) , s)) + Sn u ( (Sn u (s) , s)) Sn w ( (Sn w (s) , s)) ) ds
t 0 n (MS (s) sn u w Y
+ L (s) | (Sn u (s) , s) (Sn w (s) , s)|

Y ) ds Y
n +MS (s) sn u w t 0
n n (2MS (s) sn + L (s) L (s) MS (s)) sn u w t 0
ds
Y
n+1 sn ds MS t n+1 u w
386
Ciprian G. Gal
Thus, by induction, (2.10) holds for all n N. Next, we let t > 0 be xed, but otherwise arbitrary and choose n so large that [Me K (t) (L (t) L (t) + 2)t]n < 1. n! Then by (2.10) , S has a unique xed point in Y = CL ([0,t] , X) and therefore the problem (2.1) has a unique solution in CL ([0,t] , X) , where L is chosen as in (2.9) (see the proof of Theorem 2.1). The result follows since t > 0 is arbitrary. Remark. When the generator A is bounded, Theorem 2.1 generalizes Theorem 2.4, p. 88 in [1], while Theorem 2.2 generalizes Theorem 2.5, p. 88-89 in [1].
References
[1] J. A. Goldstein, Semigroups of Linear Operators and Applications, Oxford University Press, Oxford, 1985. [2] C. G. Gal, Almost automorphic mild solutions to some semilinear abstract differential equations with deviated argument, Journal of Integral Equations and Applications, 17 (2005), no. 4, 391396. [3] C. G. Gal, Almost automorphic mild solutions to some semilinear abstract differential equations with deviated argument in Fr chet spaces, submitteed. e [4] C. G. Gal, S.G. Gal and G.M. NGuerekata, Almost automorphic functions in Fr chet e spaces and applications to differential equations, Semigroup Forum, 75 (2005), No. 2, 23-48. [5] G. M. NGu r kata, Almost Automorphic and Almost Periodic Functions in Abstract ee Spaces, Kluwer Academic/Plenum Publishers, New York, 2001. [6] G. M. NGu r kata, Existence and uniqueness of almost automorphic mild solutions ee to some semilinear abstract differential equations, Semigroup Forum, 69(2004), No. 1, 80-86.
S INGULAR S OLUTIONS OF A S EMI - LINEAR E LLIPTIC E QUATION ON N ONSMOOTH D OMAINS

Lot Riahi Department of Mathematics, National Institute of Applied Sciences and Technology, Charguia 1, 1080, Tunis, Tunisia
Abstract Under general conditions on the signed Radon measure we prove the existence of positive singular solutions for the problem u + u p = 0, p > 1 on bounded NTAdomains in Rn , n 2. These results extend the recent ones proved by some authors to more general classes of potentials and domains. A new proof based on a simple xed point argument is also given.
(2000)AMS Subject Classications: 35J60, 35K55. Key words and phrases: semi-linear elliptic equation, singular solution, Kato class, nonsmooth domain.
1.
Introduction
In this paper we consider the semi-linear elliptic problem u(x) + u p (x) = 0, x D \ {0} u(x) > 0, x D \ {0} (P ) u(x) G(x, 0) for x near 0 and > 0 small u(x) = 0, x D,
where is the Laplacian operator, p > 1, D is a bounded non-tangentially accessible (NTA)domain (see denition in [7]) in Rn , n 2 containing 0; G denotes the Laplacian Green function with Dirichlet boundary condition on D and is a signed Radon measure on D. Recall that, in the special cases = V (x)dx and D a bounded Lipschitz domain, Zhang and Zhao [12] studied this problem in the dimension n 3 and they proved the existence of innitely many positive continuous solutions provided that the function W := G p1 (., 0)V is in the well known Kato class Kn , i.e. W is a Borel measurable function satisfying lim sup
r0 xD
E-mail
|W (y)| dy = 0. n2 D(|xy|<r) |x y|
address: Lot.Riahi@fst.rnu.tn
388
Lot Riahi
The main tool in obtaining the result is the 3G-Theorem proved in [5]. By the same arguments, as indicated in [12, p.779], the result also holds for the problem u + f (., u) = 0 provided that f (x, G(x, 0))/G(x, 0) is in the Kato class Kn . In the two dimensional case the same problem is studied in [8] on Dini-smooth Jordan domains under the condition f (x, G(x, 0))/G(x, 0) in the new functional class K (D) dened by lim sup (y) G(x, y)|W (y)|dy = 0, (x)
r0 xD D(|xy|<r)
where (x) = d(x, D) is the distance from x to the boundary of D. Recall that, for n = 2, the Kato class K2 (see [2] and [4]) is dened as the set of Borel measurable functions W satisfying: 1 lim sup |W (y)|dy = 0. Log r0 xD D(|xy|<r) |x y| From the inequalities in [11] and [8], we know that K2 K (D). The key in proving the result in the two dimensional case is the 3G-Theorem established in [11]. In this paper there are two aims. The rst is to prove the existence of continuous solutions for the problem (P ) when G p1 (., 0) is in a general class of signed Radon measures denoted by K loc (D) covering the classes Kn and K (D), where D is a bounded NTA-domain (see denition in [7]) containing 0 in Rn , n 2. The second aim is to present a new and short proof based on a more simple xed point theorem than the Schauder theorem used in the rst papers and we show the uniqueness of the solution. The idea also could be applied to solve other nonlinear elliptic problems. We also establish characterizations of the classes K loc (D) and K (D) which allow the reader to see the kind of measures admitted for the class K loc (D) and not for the class K (D). At the end of the paper we provide an interesting example of measures in K loc (D) \ K (D). Throughout the paper the letter C will denote a positive constant which may vary in value from line to line. To present the result of the paper, we should rst introduce the class K loc (D). Let (x) = 1 G(x, 0), for all x D. A singed Radon measure on D is said to be in the class K loc (D) if it satises : (y) (i) := sup G(x, y)||(dy) < , xD D (x) (ii) For any compact subset E D, lim sup The result of the paper is the following. Theorem 1.1. Let be a signed Radon measure on D such that G p1 (y, 0)(dy) is in the class K loc (D). Then there exists a number 0 > 0 such that for all ]0, 0 [, there is a unique continuous solution of the problem (P ) satisfying 3 G(x, 0) u(x) G(x, 0), 2 2 and lim for all x D \ {0},
r0 xE D(|xy|<r)
u(x) = . x0 G(x, 0)
G(x, y)||(dy) = 0.
Singular Solutions of a Semilinear Elliptic Equation
389
Remark 1.2. The condition (i) is more weaker than the Green bounded potential condition generated by the Kato class Kn used in [12]. Moreover the local Kato condition (ii) is more weaker than the global Kato conditions used in [12] and [8] and is sufcient to prove the continuity of solutions as for the Schr dinger equation in [2], [3] and [10]. These facts o loc (D) with regards to the classes K and K (D). show the advantage of our class K n Remark 1.3. Note that the measure may be singular with respect to the Lebesgue measure. Moreover, if we denote by Lip, NTA and Jords , the sets of bounded Lipschitz domains, NTA-domains, Dini-smooth Jordan domains, respectively; we have Lip NTA for n 2 and Jords Lip for n = 2. In particular our result extends the ones proved in [12] and [8] to a more general class of potentials and a more general class of domains. Remark 1.4. Let us denote by Uni f and Jor, the sets of bounded uniform domains (see denition in [1]) in Rn , n 2 and bounded Jordan domains (see denition in [4]) in R2 , respectively. We have NTA Uni f for n 2, NTA Jor and Jor NTA for n = 2. In contrast to NTA-domains, the uniform and Jordan domains may be irregular for the Dirichlet problem and their Green functions vanish outside a polar set of D. Replacing the condition u(x) = 0, x D by u(x) = 0 outside a polar set on D and using the same arguments, based on the 3G-Theorems established in [6] and [9], we may prove the existence of positive continuous solutions for the problem (P ) on uniform domains for n 3 and Jordan domains for n = 2.
2.
Preliminaries and Known Results
In this section we need to recall and prove some results which are the basic tools to prove Theorem 1.1. In particular the following 3G-Theorem is recently established for Lipschitz domains [10], for NTA-domains [6] in Rn , n 3 and for Jordan domains in R2 [9]. It is also true for NTA-domains in R2 by similar arguments as in [6]. Theorem 2.1. (3G-Theorem) Let D be a bounded NTA-domain in Rn , n 2. Then, there exists a constant C0 = C0 (n, D) > 0 such that for all x, y, z D, G(x, z)G(z, y) (z) (z) C0 G(x, z) + G(z, y) . G(x, y) (x) (y) Corollary 2.2. Let be in the class K function s on D, we have sup
xD loc (D).
Then, for any nonnegative superharmonic
In particular sup
xD D
s(y) G(x, y)||(dy) 2C0 . s(x) D
(2.1)
G(x, y)||(dy) 2C0 .
(2.2)
390
Lot Riahi
n
Proof. (2.1) holds easily by writing s(x) = sup
( fn )n of nonnegative Borel measurable functions on D, using the 3G-Theorem and the Fubinis theorem. (2.2) holds by taking s = 1 in (2.1). Lemma 2.3. Let be in the class K H(x) =
loc (D).
Then, the function G(x, y)G(y, 0)||(dy)
1 G(x, 0)
is continuous on D and tends to 0 as x 0. Proof. Note rst that by taking s = G(., 0) in (2.1), we have H(x) 2C0 , and so H is a real nite valued function on D. Let x0 D and > 0 so that E = B(x0 , ) D. We will prove the continuity at x0 . Since K loc (D), by the condition (ii), for > 0 there is r ]0, [ small such that sup
xE D(|xy|<r)
G(x, y)||(dy) < .
H(x)
r We will estimate H1 (x), for x B(x0 , 2 ). Let M = sup xE
have H1 (x) C0
(y) G(x, y)||(dy) + r B(x0 , 2 ) (x)

r B(x0 , 2 )
MC0 MC0
G(x, y)||(dy) +
r B(x0 , 2 )
G(x, y)||(dy) +
DB(x,r)
r B(x0 , 2 )
Clearly the local Kato condition (ii) yields lim
r0 B(x ,r)
(see also Proposition 7.1 in [3]). In particular, for r small enough,

r B(x0 , 2 )
G(y, 0)||(dy) < .
Combining (2.3)-(2.5), we obtain H1 (x) 2MC0 . We will estimate H2 (x). We have (n/2 1) n if n 3. 4 2 |x y|n2 G(x, y) 1 (1 Log 1 ) if n = 2. 2 |x y|
1 G(x, y)G(y, 0)||(dy) G(x, 0) D 1 G(x, y)G(y, 0)||(dy) + = r G(x, 0) B(x0 , 2 ) := H1 (x) + H2 (x). =
G(x, y) fn (y)dy, for an increasing sequence
(2.3)
r DBc (x0 , 2 )
G(x, y)G(y, 0)||(dy)
1 . By the 3G-Theorem, we (x)
(y) G(y, 0)||(dy) r B(x0 , 2 ) (0) G(y, 0)||(dy) G(y, 0)||(dy) . (2.4)
G(x, y)||(dy) = 0 for all x, x D
(2.5)

r r It follows that for x B(x0 , 4 ) and y D Bc (x0 , 2 ), G(x, y)
391 where cn is a con-
cn , n1 r
stant depending only on n. Using this inequality,
convergence theorem we easily obtain the continuity of the function x at x0 . Since

r DBc (x0 , 2 )
G(x, y)G(y, 0)||(dy)
1 is continuous on D, H2 (x) is continuous at x0 . Hence H(x) is continuG(x, 0) 2cnC0 r and ous at x0 . Moreover when x0 = 0, we remark that for x B(0, 4 ), H2 (x) G(x, 0) H1 (x) 2MC0 . Thus H(x) 0 as x 0.
3.
Proof of Theorem 1.1.
We will show that there exists a number 0 > 0 such that for all ]0, 0 [, there is a unique function u C(D \ {0}) satisfying, for all x D \ {0}, u(x) = G(x, 0) + and
G(x, y)u p (y)(dy)

D
3 G(x, 0) u(x) G(x, 0). 2 2
Clearly if u satises the above conditions, then u is a distributional solution of the u(x) and = G p1 (., 0) it then sufces to show that problem (P ). By writing w(x) = G(x, 0) there exists a unique function w C(D) satisfying w(x) = + and 1 G(x, 0)
G(x, y)G(y, 0)w p (y)(dy)

D
3 w(x) . 2 2 For > 0, let then consider the set S = w C(D) : 3 . w 2 2

),
S is a closed subset in the Banach space (C(D), . dene the operator T on S by Tw(x) = + 1 G(x, 0)
G(x, y)G(y, 0)w p (y)(dy).

D
By Lemma 2.3, for w S, Tw C(D). Moreover, by the 3G-Theorem, we have Tw(x) 1 p1 2 p+1C0 .
G(y, 0)||(dy) < and the dominated
where w
= sup |w(x)|. Let us

xD
392
Lot Riahi
1
Then by choosing 0 < < (2 p+2C0 ) p1 , we get Tw 2 On the other hand, by the 3G-Theorem, for w, w S, we have, Tw Tw
3 2
and so Tw S.
sup
1 G(x, y)G(y, 0)||(dy) xD G(x, 0) D 2C0 w p w p ww
2 p p p1C0
For 0 < < (2 p pC0 ) p1 we have 2 p p p1C0 < 1. Hence for 0 < < 0 = 1 (2 p+2 pC0 ) p1 , the operator T is a 1 -Lipschitz mapping from S into itself; and so by 4 the well known xed point theorem there exists a unique w S such that Tw = w.
4.
The Classes K
loc (D)
and K (D).
In this section, we prove characterizations of the classes K loc (D) and K (D) which allow us to show that K loc (D) contains more signed Radon measures than the class K (D). Proposition 4.1. Let be a signed Radon measure on D. Then K loc (D) if and only if (y) the function P (x) = G(x, y)||(dy) is in Cb (D). Here Cb (D) is the set of bounded D (x) continuous functions on D. Proof. Let K J(x) =
loc (D).
Then sup P (x) = < and so P is bounded. We will show the

xD
continuity of P on D. Since is continuous on D, it then sufces to show the continuity of

D
(y)G(x, y)||(dy). For any r > 0 small, we have = (y)G(x, y)||(dy) +

J(x)
D(|xy|<r)
:= pr (x) + qr (x). By using the fact that 1{|xy|r} G(x, y) cn ; n1 r
inated convergence theorem, it follows that qr is continuous on D. Now let x0 D and > 0 such that E = B(x0 , ) D. By the assumption (ii), for > 0, there is r ]0, [ small such that sup
xE D(|xy|<r)
Then for x B(x0 , ), we have |J(x) J(x0 )| |pr (x) pr (x0 )| + |qr (x) qr (x0 )| 2 + |qr (x) qr (x0 )|, and so J(x) J(x0 ) as x x0 . Thus J is continuous on D.
wp w p
D(|xy|r)
(y)G(x, y)||(dy)
(y)||(dy) < 2C0 and the dom-
G(x, y)||(dy) < .

xD
393
Conversely assume that P Cb (D). Then = sup P (x) < . As above for r > 0 pr (x) qr (x) qr pr small = P (x) with is continuous on D, and so is continuous on D. (x) (x) pr (x) Since lim = 0 for all x D, by the Dini-theorem, it follows that, for any compact r0 (x) subset E D, pr (x) =0 lim sup r0 xE (x) which means that satises (ii). Now, for n 2, let us dene the class K (D) as the set of signed Radon measures on D satisfying : (y) lim sup G(x, y)||(dy) = 0. r0 xD D(|xy|<r) (x) This class is used in [8] in the special cases of Dini-smooth Jordan domains in R2 and = V (x)dx. Clearly by the compactness of D, if K (D) then < . Hence satises (i) and (ii) and so K loc (D). Combining with Lemma 3.2 in [10] and Lemmas 3.5 and 3.6 in [9], we then have the inclusions: Kn K (D) K loc (D). For interesting examples of functions in K loc (D) \ Kn , we refer the reader to Corollary 4.8 in [10]. In what follows we will prove a characterization of the class K (D). Proposition 4.2. Let be a signed Radon measure on D. Then K (D) if and only if (y) the function P (x) = G(x, y)||(dy) is in C(D). Here C(D) is the set of functions D (x) continuous up to the boundary on D. Proof. Let K (D). Then K loc (D) and so by Proposition 4.1, P Cb (D). It remains to prove that P is continuous up to the boundary D. Let x0 D. For > 0, there is r > 0 small so that (y) sup G(x, y)||(dy) < . xD D(|xy|<r) (x)
r For x D B(x0 , 2 ), we then have
Q1 (x) := Moreover the function
Q2 (x) := is the quotient of
r D(|x0 y| 2 ) r r lutions on D B(x0 , 4 ) vanishing on D B(x0 , 4 ) and so by a limit as x x0 . Thus P (x) = Q1 (x) + Q2 (x) has a limit as
(y)G(x, y)||(dy) and (x) which are two nonnegative soTheorem 7.9 in [7], Q2 (x) has x x0 . Thus P C(D).
Conversely, let be a Borel measurable function such that P C(D). Then = sup P (x) < . For r > 0 small, put
xD
r D(|x0 y|< 2 )
(y) G(x, y)||(dy) < . (x)
(y) G(x, y)||(dy) r D(|x0 y| 2 ) (x)
394

Lot Riahi (y) G(x, y)||(dy) D(|xy|<r) (x)
pr (x) = and qr (x) =
(y) G(x, y)||(dy). D(|xy|r) (x)
Let x0 D. By the boundary Harnack principle (Theorem 5.1 in [7]), there is C > 0 G(x, y) G(x, y) r such that for all x D B(x0 , 2 ) and all y D Bc (x, r), = C. More(x) G(x, 0) G(x, y) = M(y, x0 ), where M(., x0 ) is the Martin-Poisson kernel with p le x0 o over lim xx0 G(x, 0) normalized at 0. Since
xx0
lim qr (x) =
D(|x0 y|r)
and so qr C(D). Thus pr = P qr C(D). Since lim pr (x) = 0, by the Dini-theorem, we

r0
have lim sup pr (x) = 0 and so K (D).

r0 xD
Example 4.3. We will present an interesting example of signed Radon measures in K loc (D) \ K (D) even in simple smooth domains. Let us consider D = B(0, 1) the unit ball in Rn , n 2. From the formula of the Green function or the estimates given in [4] C and [8], we know that G(x,y) |xy|n1 , for all x, y D. Fix a point z0 B(0, 1). For all (x) integer m 1, let us dene the sequence of disjoint balls Bm in D by Bm = B(ym , rm ), where 1 1 ym = (1 4m )z0 and rm = 16m(m+1) . Choose m > 0 such that m G(.,ym ) 2m on D \ Bm (it n1 sufces to take m 2m rm C1 ), and dene pm = m G(., ym ) 2 and p=
Clearly pm C(D) and for all integer k0 1, the series on D \
Bk and so p C(D). Moreover p 3. Then p Cb (D). We will prove that lim sup p(y) lim sup pm (ym ) = 2.
yz0 m
k=k0
p C(D). We have /
On the other hand by choosing zm = ym + 2rm z0 , we clearly have zm / m and p(zm ) 1 which imply lim inf p(y) 1.
yz0
k=1
(y)||(dy) 2C0 , by the dominated convergence theorem
(y)M(y, x0 )||(dy). Hence qr has a limit at any point x0 D
m=1
pm .
m=1
pm converges uniformly
Bk , zm z0 as
Singular Solutions of a Semilinear Elliptic Equation Thus p Cb (D) \C(D). Note that p =
395
on D, and so there is a positive Radon measure on D such that (x)p(x) =

m=1
pm =
m=1
m G(., y) 2 is a -potential
G(x, y)(dy), x D.
D
We then have p(x) =
(y) G(x, y)(dy), x D, (x) D

loc (D) \ K
1 where = . From Propositions 4.1 and 4.2, it follows that K
(D).
Acknowledgment
I want to sincerely thank Professor Wolfhard Hansen for his help in the example 4.3 and Professor Kentaro Hirata for his interesting remarks and comments about the paper.
References
[1] H. Aikawa : Boundary Harnack principle and Martin boundary for a uniform domain, J. Math. Soc. Japan, 53, 1 (2001), 119-145. [2] M. Aizenman and B. Simon : Brownian motion and Harnacks inequality for Schr dinger operators, Comm. Pure Appl. Math., 35 (1982), 209-271. o [3] A. Boukricha, W. Hansen and H. Hueber : Continuous solutions of the generalized Schr dinger equation and perturbation of harmonic spaces, Exposition Math., o 5 (1987), 97-135. [4] K.L. Chung and Z. Zhao: From Brownian motion to Schr dingers equation, Springero Verlag, New York 1995. [5] M. Cranston, E.B. Fabes and Z. Zhao: Conditional gauge and potential theory for the Schr dinger operator, Trans. Amer. Math. Soc., 37, 1 (1988), 171-194. o [6] W. Hansen : Uniform boundary Harnack principle and generalized triangle property, J. Funct. Anal., 226 (2005), 452-484. [7] D. Jerison and C. Kenig : Boundary behavior of harmonic functions in nontangentially accessible domains, Advances in Math., 46 (1982), 80-147. [8] H. M agli and L. M atoug : Singular solutions of a nonlinear equation in bounded a a domains of R2 , Jour. Math. Anal. Appl., 270 (2002), 230-246.
396
Lot Riahi
[9] L. Riahi: A 3G-Theorem for Jordan domains in R2 , Colloquium Math., 101 (2004), 1-7. [10] L. Riahi: The 3G-inequality for general Schr dinger operators on Lipschitz-domains, o Manuscripta Math., 116 (2005), 211-227. [11] M. Selmi: Inequalities for Green function in a Dini-Jordan domain in R2 , Potential Anal., 13 (2000), 81-102. [12] Q.S. Zhang and Z. Zhao: Singular solutions of semilinear elliptic and parabolic equations, Mathematische Annalen, 310 (1998), 777-794.
E XISTENCE OF W EIGHTED P SEUDO A LMOST P ERIODIC S OLUTIONS TO S OME N ON -AUTONOMOUS D IFFERENTIAL E QUATIONS
Toka Diagana Department of Mathematics, Howard University, 2441 6th Street N.W., Washington D.C. 20059, USA
Abstract The paper studies the so-called weighted pseudo almost periodic functions recently introduced by the author. Properties of those weighted pseudo almost periodic functions are discussed including a composition result of weighted pseudo almost periodic functions. The obtained results, subsequently, are utilized to study the existence and uniqueness of a weighted pseudo almost periodic solutions to some non-autonomous abstract differential equations.
1991 Mathematics Subject Classication: 44A35; 42A85; 42A75 Key words and phrases: weight, pseudo almost periodic, weighted pseudo almost periodic, almost periodic, heat equation.
1.
Introduction
In Diagana [4], a new generalization of Bohr almost periodic functions was introduced. Such a new concept is called weighted pseudo almost periodicity and implements in a natural fashion the notion of pseudo almost periodicity due Zhang [13, 14, 15]. To construct those new spaces, the main idea consists of enlarging the so-called ergodic component, utilized in the Zhangs denition of pseudo almost periodicity, with the help of a weighted measure d(x) = (x)dx, where : R (0, ) is a locally integrable function over R, which is commonly called weight. Unlike [4], here we provide with all complete proofs. We basically take a closer look into properties of weighted pseudo almost periodic functions and study their relationship with the Zhangs pseudo almost periodicity. For that, we consider a binary equivalence relation, , on U , the set of weights, which enables us to reorganize those weights into different equivalence classes. Among other things, if two weights 1 and 2 are equivalent, that is, 1 2, then their corresponding weighted pseudo almost periodic spaces coincide. In particular, when a weight is bounded with inf xR (x) > 0, it is then equivalent to the constant function 1, and hence the weighted pseudo almost periodic space with weight coincides with the Zhangs spaces (Corollary 3.4). In addition to
E-mail
address: tdiagana@howard.edu
398
Toka Diagana
the above, a composition result of weighted pseudo almost periodic functions is obtained (Theorem 3.6). Applications include the search of weighted pseudo almost periodic solutions to the abstract non-autonomous differential equation u (t) = A(t)u(t) + f (t, u(t)), t R, (1.1)
where A(t) : D(A(t)) X X is a family of densely dened closed linear operators on a common domain D = D(A(t)), which is independent of t, and f : R X X is a weighted pseudo almost periodic function in the rst variable uniformly in the second one. To illustrate Theorem 4.3 we consider the existence and uniqueness of a weighted pseudo almost periodic solution to the one dimensional heat equation.
2.
Preliminaries
Let (X, ), (Y, Y) be two Banach spaces. The collection of all bounded linear operators from X into Y with be denoted B(X, Y). This is simply denoted B(X) when X = Y. Let U denote the collection of all functions (weights) : R (0, ), which are locally integrable over R such that (x) > 0 for almost each x R. From now on, if U and for T > 0, we then set m(T, ) :=
T T
(x)dx.
As in the particular case when (x) = 1 for each x R, we are exclusively interested in those weights, , for which, lim m(T, ) = . Thus throughout the rest of the paper we
T
consider weights dened by U := U : lim m(T, ) = .

T
Of course, one can always consider those weights U for which lim m(T, ) = .
T
However, those cases do not correspond to the situations that we want to consider here, as we examine some generalizations of the Zhangs pseudo almost periodicity. In addition to the above, the current setting requires the space of weights UB dened by UB := U : is bounded with inf (x) > 0 .
xR
Obviously, UB U U, with strict inclusions. Let BC(R, X) (respectively, BC(R Y, X)) denote the collection of all X-valued bounded continuous functions (respectively, the class of jointly bounded continuous functions F : R Y X). The space BC(R, X) equipped with its natural norm, that is, the sup norm dened by u = sup u(t)
tR
is a Banach space. Furthermore, C(R, Y) (respectively, C(R Y, X)) denotes the class of continuous functions from R into Y (respectively, the class of jointly continuous functions F : R Y X).
Existence of Weighted Pseudo Almost Periodic Solutions ...
399
Denition 2.1. A function f C(R, X) is called (Bohr) almost periodic if for each > 0 there exists l() > 0 such that every interval of length l() contains a number with the property that f (t + ) f (t) < for each t R. The number above is called an -translation number of f , and the collection of all such functions will be denoted AP(X). Denition 2.2. A function F C(R Y, X) is called (Bohr) almost periodic in t R uniformly in y Y if for each > 0 and any compact K Y there exists l() such that every interval of length l() contains a number with the property that F(t + , y) F (t, y) < for each t R, y K. The collection of those functions is denoted by AP(Y, X). To introduce those weighted pseudo almost periodic functions, we need to dene the weighted ergodic space PAP0 (X, ). Weighted pseudo almost periodic functions will then appear as perturbations of almost periodic functions by elements of PAP0 (X, ). Let U . Dene PAP0 (X, ) := 1 T m(T, )
T T
f BC(R, X) : lim
f () () d = 0 .
Obviously, when (x) = 1 for each x R, one retrieves the so-called ergodic space of Zhang, that is, PAP0 (X), dened by PAP0 (X) := f BC(R, X) : lim 1 T 2T
T
f () d = 0 .
T
Clearly, the spaces PAP0 (X, ) are richer than PAP0 (X) and give rise to an enlarged space of pseudo almost periodic functions. In Corollary 3.4, some sufcient condition on the weight U are given so that PAP0 (X, ) = PAP0 (X). In the same way, we dene PAP0 (Y, X, ) as the collection of jointly continuous functions F : R Y X such that F(, y) is bounded for each y Y and 1 T m(T, ) lim
T T
F(s, y) (s) ds = 0
uniformly in y Y. We are now ready to dene the notion of weighted pseudo almost periodicity. Denition 2.3. Let U . A function f BC(R, X) is called weighted pseudo almost periodic (or -pseudo almost periodic) if it can be expressed as f = g + , where g AP(X) and PAP0 (X, ). The collection of such functions will be denoted by PAP(X, ). Remark 2.4. (i) The functions g and in Denition 2.3 are respectively called the almost periodic and the weighted ergodic perturbation components of f .
400
Toka Diagana
).
(ii) The space PAP(X, ) is a closed subspace of (BC(R, X), (PAP(X, ), ) is a Banach space. (iii) PAP(X)
U
This yields
PAP(X, ).
(iv) If UB , then the space PAP(X, ) is translation invariant. Proposition 2.5. Fix UB . Let f PAP0 (R, ) and let g L1 (R). Then f g, the convolution of f and g on R, belongs to PAP0 (R, ). Proof. From f PAP0 (R, ) and g L1 (R) it is clear that f g BC(R). Now setting J(T, ) := it follows that 1 m(T, )
T T
1 m(T, )
T T
| f (t s)|(t)|g(s)| ds dt
|( f g)(t)|(t)dt J(T, )
+
=
+
|g(s)|
1 m(T, )
| f (t s)|(t)dt ds
T
|g(s)|T (s)ds,
where T (s) =
T 1 | f (t s)|(t)dt. m(T, ) T Since PAP0 (R, ) is translation invariant it follows that T (s) 0 as T . Next, using the boundedness of T (|T (s)| f ) and the fact that g L1 (R), the Lebesgue dominated convergence theorem yields + T
lim
|g(s)|T (s)ds = 0,
and hence f g PAP0 (R, ). Similarly, it is clear that if h AP(R) and L1 (R), then the convolution h AP(R). Combining those results one obtains: Proposition 2.6. Fix U . Let f PAP(R, ) and let g L1 (R). Then f g, the convolution of f and g on R, belongs to PAP(R, ). Example 2.7. Let U . Then the function T dened by
T (x) =
K(x y) f (y)dy,
where K L1 (R) and f PAP(R, ). Then T PAP(R, ), by Proposition 2.6.
401
Proposition 2.8. Let U and let p > 1. If | f | p PAP0 (R), then f PAP0 (R, ) provided 1 1 p p m(T, p1 ) 1 L(p) = sup (2T ) p . < . m(T, ) T >0 Proof. It sufces to note that the following inequality holds 1 m(T, )
T
| f (t)|(t)dt
T
1 m(T, )
T T
T T
1/p
| f (t)| pdt
1 1 p
L(p) .
p p1
(t)dt
T T 1/p
1 2T
| f (t)| pdt
Therefore passing to the limit as T in the previous inequality yields the desired result. Denition 2.9. Fix U. A function F C(R Y, X) is called weighted pseudo almost periodic (or -pseudo almost periodic) in t R uniformly in y Y if it can be expressed as F = G + , where G AP(Y, X) and PAP0 (Y, X, ). The collection of such functions will be denoted by PAP(Y, X, ).
3.
Properties of Weighted Pseudo Almost Periodic Functions
Let U. If f , g PAP(X, ) and let R, then f + g is also in PAP(X, ). Moreover, if | f ()| is not even and if (x) = M() < , sup (x) xR then the function f(x) := f (x) for x R is also in PAP(X, ). In particular, if is even, then f belongs to PAP(X, ). Theorem 3.1. Fix U . The decomposition of a weighted pseudo almost periodic function f = g + , where g AP(X) and PAP0 (X, ), is unique. Proof. Let f = g1 + 1 where g1 AP(X) and 1 PAP0 (X, ). Following along the same line as in the proof of [13, Lemma 1.3] it easily follows that g1 (R) f (R). Thus, if f = g2 + 2 where g2 AP(X) and 2 PAP0 (X, ), then 0 = f f = (g1 g2 ) + (1 2 ) PAP(X, ), where (g1 g2) AP(X) and (1 2 ) PAP0 (X, ). Hence, using the argument above, it follows that (g1 g2 )(R) {0}, and therefore, g1 = g2 and 1 = 2 .
402
Toka Diagana 2
Denition 3.2. Let 1, 2 U . One says that 1 is equivalent to 2 and denote it 1 1 if and only if := UB . 2
1 Examples of equivalent weights include and , where is any arbitrary weight in U . Let 1, 2, 3 U . It is clear that 1 1 (reexivity); if 1 2, then 2 2 (symmetry); and if 1 2 and 2 3 , then 1 3 (transitivity). So, is a binary equivalence relation on U . Thus the equivalence class of a given weight U will then be denoted by cl() = { U : }. It is then clear that U =
U
cl().
Theorem 3.3. Let U . If 1 , 2 cl(), then PAP(X, 1) = PAP(X, 2). Proof. Since 1 2 there exist some constants K, K > 0 such that K 2(x) 1(x) K2(x) for each x R. In particular, for each T > 0, 1 1 1 . K . m(T, 2) m(T, 1) K . m(T, 2) Let PAP0 (X, 2). In view of the above it easily follows that 1 m(T, 1)
T T
(s) 1(s)ds =
T 1 1 (s) (s)2(s)ds m(T, 1) T 2 T K (s) 2(s)ds m(T, 1) T T K (s) 2(s)ds. K . m(T, 2 ) T
Consequently, 1 T m(T, 1 ) lim

T T
(s) 1(s)ds = 0,
and hence PAP0 (X, 1 ), that is, PAP0 (X, 2) PAP0 (X, 1). To obtain the other inclusion, one simply uses the fact that the equivalence relation symmetric, that is, 2 1 , too.
is
An immediate consequence of Theorem 3.3 is that PAP(X, ) = PAP(X, cl()). This enables us to identify the Zhangs space PAP(X) with a weighted pseudo almost periodic class PAP(X, ). Corollary 3.4. If UB , then PAP(X, ) = PAP(X, cl(1)) = PAP(X). Proof. It sufces to see that there exist K, K > 0 with K (x) K for each x R, which yields 1. In view of the above, the proof of the next corollary is quite immediate.
403
Corollary 3.5. If 1 2 , then (1) PAP(X, 1 + 2 ) = PAP(X, 1) = PAP(X, 2 ), and (2) 1 PAP(X, ) = PAP(X, cl(1)) = PAP(X). 2 The next theorem implements the theorem of composition of pseudo almost periodic functions given in [1]. Theorem 3.6. Let U and let f PAP(Y, X, ) satisfying the Lipschitz condition f (t, u) f (t, v) L . u v
Y
for all u, v Y, t R.
If h PAP(Y, cl()), then f (, h()) PAP(X, cl()). Prior to proving Theorem 3.6, we give the following immediate consequences. Corollary 3.7. Let 1 , 2 U with 2 cl(1). Let f PAP(Y, X, cl(1)) satisfying the Lipschitzs condition f (t, u) f (t, v) L . u v
Y
for allu, v Y, t R.
If h PAP(Y, 2), then f (, h()) PAP0 (X, cl(1 )). Corollary 3.8. Let UB . Let f PAP(Y, X, cl()) satisfying the Lipschitzs condition f (t, u) f (t, v) L . u v If h PAP(Y, ), then f (, h()) PAP(X). Proof. (Theorem 3.6) We adapt our proof to the one given in [1]. Write f = g + where g AP(Y, X) and PAP0 (Y, X, ). Similarly, write h = h1 + h2 , where h1 AP(Y) and h2 PAP0 (Y, ). It is obvious to see that f (, h()) C(R, X) as the composition of continuous functions. Now decompose f as follows f (, h()) = g(, h1()) + f (, h()) g(, h1()) = g(, h1()) + f (, h()) f (, h1()) + (, h1()). Using the theorem of composition of almost periodic functions, it is easy to see that g(, h1()) AP(X). Now, set F() = f (, h()) f (, h1()). Clearly, F PAP0 (X, ). Indeed, for T > 0, 1 m(T, )
T T Y
for allu, v Y, t R.
F(s) (s)ds =
1 m(T, ) L m(T, ) L m(T, )
T T T T T T
f (s, h(s)) f (s, h1(s)) (s)ds h(s) h1(s) (s)ds h2(s) (s)ds,
404 and hence
Toka Diagana 1 T m(T, ) lim

T T
F(s) (s)ds = 0.
To complete the proof we have to show that 1 T m(T, ) lim

T T
(s, h1(s)) (s)ds = 0.
As h1 AP(Y), h1 (R) is relatively compact. Thus for each > 0 there exists a nite number of open balls Bk = B(xk , ), centered at xk h1(R) with radius for instance 3L 3L
m
with h1(R)
k=1 m
Bk . Thus for 1 k m, the set Uk = {t R : h1(t) Bk } is open and

k1
R=
k=1
Uk . Now, set Vk = Uk
i=1
/ Ui and V1 = U1. clearly, Vi V j = 0 for all i = j.
Since PAP0 (Y, X, ) there exists T0 > 0 such 1 T m(T, ) lim

T T
(s, xk ) (s)ds <
for T T0 3m
(3.1)
and k {1, 2, ..., m}. Moreover, since g (g AP(Y, X)) is uniformly continuous in R h1 (R), one has g(t, xk) g(t, x) < for x Bk , k = 1, 2, .., m. 3 (3.2)
Using above and the following the decompositions (, h1()) = f (, h1 ()) g(, h1()) and (t, xk) = f (t, xk ) g(t, xk ) it follows that 1 m(T, ) = + +
T T
(s, h1(s)) (s)ds (s, h1(s)) (s)ds (s, h1(s)) (s, xk ) (s)ds (s, xk) (s)ds f (s, h1(s)) f (s, xk ) (s)ds (s, xk) (s)ds
m 1 m(T, ) k=1 m 1 m(T, ) k=1 m 1 m(T, ) k=1 m 1 m(T, ) k=1 m 1 m(T, ) k=1
Vk [T,T ]
Vk [T,T ]
Vk [T,T ]
Vk [T,T ]
Vk [T,T ]
Existence of Weighted Pseudo Almost Periodic Solutions ... + + + +

m 1 m(T, ) k=1 m 1 m(T, ) k=1 m 1 m(T, ) k=1
405
Vk [T,T ]
g(s, h1(s)) g(s, xk) (s)ds L h1(s) xk

Y (s)ds
Vk [T,T ]
Vk [T,T ]
g(s, h1(s)) g(s, xk) (s)ds (s, xk ) (s)ds. < for 1 k m. 3L
k=1
m(T, )
Vk [T,T ]
For each s Vk [T, T ], h1 (s) Bk in the sense that h1(s) xk Clearly, from (3.1)-(3.2) it easily follows that 1 m(T, ) for T T0, and hence 1 T m(T, ) lim
T T T T
(s, h1(s)) (s)ds
(s, h1(s)) (s)ds = 0.
4.
Weighted Pseudo Almost Periodic Solutions
Let U. This section is devoted to the search of a -pseudo almost periodic solution to the abstract non-autonomous differential equation (1.1). For that, we suppose that {A(t) : t R} is a family of closed linear dened on a common domain D, which is independent of t, and dense in X. Throughout the rest of the paper we suppose that the system u (t) = A(t)u(t), t s, u(s) = x X, (4.1)
has an associated evolution family of operators {U(t, s) : t s with t, s R}, which is uniformly asymptotically stable. Denition 4.1. A family of linear operators {U(t, s) : t s with t, s R} B(X) is called an evolution family of operators for (4.1) whenever the following conditions hold: (a) U(t, s)U(s,r) = U(t, r) for every r s t; (b) for each x X the function (t, s) U(t, s)x is continuous and U(t, s) B(X, D) for every t > s; and (c) the function (s,t] B(X), t U(t, s) is differentiable with U(t, s) = A(t)U(t, s). t
406
Toka Diagana
More details on evolution families can be found in the book by Lunardi [12]. To deal with the existence and uniqueness of a -pseudo almost periodic solution to (1.1), we require the following additional assumptions: (H.1) The function f : R X X, (t, u) f (t, u) is -pseudo almost periodic in t R uniformly in u X. Furthermore, f satises the Lipschitzs condition: there exists K > 0 such that f (t, u) f (t, v) K . u v
Y,
u, v X, t R.
(H.2) The evolution family U(t, s) is asymptotically stable. Namely, there exist some constants M, > 0 such that U(t, s) Me(ts) for every t s. (H.3) Let > 0. Setting
T
P() := sup
T >0 T
e(T +t) (t)dt ,
then P() < . Remark 4.2. Note that there are weights U for which (H.3) is not satised. For instance, consider the weights dened for each > 0 by (t) = |t|e2t for t R. It is then clear that P() = . Theorem 4.3. Fix U . Suppose that assumptions (H.1)-(H.2)-(H.3) hold. Then (1.1) has a unique weighted pseudo almost periodic solution whenever K < . M Proof. Let u be a bounded solution of (1.1). It is well-known that it can be expressed as
t
(Mu)(t) :=
U(t, s) f (s, u(s))ds for each t R.
Let u PAP(X, ) and write f = g + where g AP(Y, X) and PAP0 (X, X, ). Using (H.1) and the composition result, Theorem 3.6 it follows that f (, u()) = g(, u()) + (, u()) where g(, u()) AP(X) and (, u()) PAP0 (X, ). Rewrite (Mu) as (Mu) = (M1 u) + (M2 u), where
t
(M1u)(t) :=
U(t, s)g(s, u(s))ds for each t R, and

t
(M2u)(t) :=
U(t, s)(s, u(s))ds for each t R.
Since g(, u()) AP(X), for each > 0, there exists l() > 0 such that each every interval of length l() contains a such that g(t + , u(t + )) g(t, u(t)) < . for all t R. M
407
Considering (M1u)(t + ) (M1)(t) and using assumption (H.2) it easily follows that (M1u)(t + ) (M1 u)(t) < for all t R, and hence (M1u)() AP(X). The next step consists of showing that 1 T m(T, ) lim Now it follows from (H.2) that 1 T m(T, ) lim where I() := lim M T m(T, )
T T T T T T
(M2u)(t) (t)dt = 0.
(M2u)(t) (t)dt I() + J(),

t T
(t)dt
T
e(ts) (s, u(s))

T
, and .
J() := lim Now
M T m(T, )
(t)dt
e(ts) (s, u(s))
I() =
T t M (t, u(t)) (t)dt e(ts) ds T m(T, ) T T T 1 M (t, u(t)) (t)dt = lim 1 e(t+T ) T m(T, ) T T M M (t, u(t)) (t)dt . lim T m(T, ) T = 0,
lim
by (, u()) PAP0 (X, ). Similarly,

T T M es (s, u(s)) ds et (t)dt T m(T, ) T P(, ) . M . sup (s, u(s)) sR lim T . m(T, )
J() =
lim
= 0, by (H.3). To complete the proof, we apply the xed-point principle of Banach to the nonlinear operator M. In view of the above, it is clear that M maps PAP(X, ) into itself. Moreover, for all u, v PAP(X, ), it easy to see that MK u v , and hence M has a unique xed-point, which obviously is the unique -pseudo almost periodic solution to (1.1). (Mu) (Mv)
408
Toka Diagana
Remark 4.4. In the particular case when UB , that is, PAP(X, cl()) = PAP(X) by Corollary 3.4, we retrieve the non-weighted situation, since assumption (H. 3) is always achieved in that event. This means that weighted pseudo almost periodic functions are good generalizations of the Zhangs pseudo almost periodic functions.
5.
Example
To illustrate Theorem 4.3 we consider the existence and uniqueness of weighted pseudo almost periodic solutions to the heat equation given by the system u 2 u (t, x) + Q(t, x)u(t, x) + . H(t) sin(u(t, x)) (5.1) (t, x) = t x2 u(t, 0) = u(t, ) = 0, t R (5.2) where H(t) = sint + sin( 2t) + e|t| for each t R, Q : R [0, ] R is a jointly measurable function, and > 0 is a constant. For that, we suppose that X = (L2 [0, ], 2) and dene D(A) = {u L2 [0, ] : u L2 [0, ], u(0) = u() = 0} Au() = u (), u() D(A). Note that the linear operator A dened above is the innitesimal generator of an analytic semigroup (T (t))t0 on L2 [0, ] satisfying T (t) et for each t 0. Dene the class of operators A(t) by setting D(A(t)) = D(A) A(t)v(x) = Av(x) + Q(t, x)v for each v D(A(t)). By assuming that x Q(t, x) is continuous for each t R with Q(t, x) 0 (0 > 0) for all t R, x [0, ], one sees that the system u (t) = A(t)u(t), t s, u(s) = y X, (5.3)
has an associated evolution family (U(t, s))ts on L2 [0, ], which can be explicitly expressed by
t
Q(, y)d U(t, s)v = T (t s)e Consequently, U(t, s) e(1+0 )(ts) for every t s. Let N N. Throughout the rest of this example, we suppose that the weight is given by: (t) = (1 + |t|2)N , t R.
s
v.
409
Setting f (t, u(t, )) = . H(t) sinu(t, ), one can clearly see that f is 3-Lipschitz in the sense that, f (t, u(t, )) f (t,v(t,)) 2 3 . u(t, ) v(t, ) 2 for all u(t, ), v(t, ) L2 [0, ], t R. Furthermore, it can be easily checked that H PAP(R, (1 + |t|2)N ). Consequently, f is pseudo almost periodic in t R uniformly in the second variable. Note that (H.3) is satised. The previous discussion can be formulated as follows. Theorem 5.1. The heat equation with Dirichlet condition in (5.1)-(5.2) has a unique weighted pseudo almost periodic solution whenever 3 < 1. 1 + 0
References
[1] B. Amir and L. Maniar, Composition of pseudo-almost periodic functions and Cauchy problems with operator of nondense domain. Ann. Math. Blaise Pascal 6 (1999), no. 1, pp. 111. [2] D. Bugajewski and T. Diagana, C. M. Mahop, Asymptotic and Pseudo Almost Periodicity of the Convolution Operator and Applications to Differential and Integral Equations, J. Anal. Appl. 25 (2006), 327-340. [3] C. Corduneanu, Almost Periodic Functions , 2nd Edition, Chelsea-New York, 1989. [4] T. Diagana, Weighted Pseudo Almost Periodic Functions and Applications, C. R. Acad. Sci. Paris, Ser I 343 (2006), no. 10, 643-646. [5] T. Diagana, Pseudo Almost Periodic Solutions to Some Differential Equations, Nonlinear Anal 60 (2005), no. 7, pp. 12771286. [6] T. Diagana and E. Hern` ndez M., Existence and Uniqueness of Pseudo Almost Peria odic Solutions to Some Abstract Partial Neutral Functional-Differential Equations and Applications, J. Math. Anal. Appl. 327(2007), no. 2, 776-791. [7] T. Diagana, C. M. Mahop, and G. M. NGu r kata, Pseudo Almost Periodic Solution ee to Some Semilinear Differential Equations, Math. Comp. Modelling 43 (2006), no. 1-2, pp. 89-96. [8] T. Diagana, C. M. Mahop, G. M. NGu r kata, and B. Toni, Existence and Uniqueee ness of Pseudo Almost Periodic Solutions to Some Classes of Semilinear Differential Equations and Applications. Nonlinear Anal. 64 (2006), no. 11, pp. 2442-2453. [9] T. Diagana, Existence and Uniqueness of Pseudo Almost Periodic Solutions to Some Classes of Partial Evolution Equations. Nonlinear Anal. 66 (2007), no. 2, 384-395.
410
Toka Diagana
[10] T. Diagana, An Introduction to Classical and p-adic Theory of Linear Operators and Applications, Nova Science Publishers, New York, 2006. [11] H. X. Li, F. L. Huang, and J. Y. Li, Composition of pseudo almost-periodic functions and semilinear differential equations. J. Math. Anal. Appl 255 (2001), no. 2, pp. 436446. [12] A. L UNARDI, Analytic semigroups and optimal regularity in parabolic problems , PNLDE Vol. 16, Birkh auser Verlag, Basel, 1995. a [13] C. Y. Zhang, Pseudo Almost Periodic Solutions of Some Differential Equations. J. Math. Anal. Appl 181 (1994), no. 1, pp. 6276. [14] C. Y. Zhang, Pseudo Almost Periodic Solutions of Some Differential Equations. II. J. Math. Anal. Appl 192 (1995), no. 2, pp. 543561. [15] C. Y. Zhang, Integration of Vector-Valued Pseudo Almost Periodic Functions, Proc. Amer. Math. Soc 121 (1994), no. 1, pp. 167174.
A K RASNOSELSKII -T YPE F IXED P OINT T HEOREM FOR M ULTIFUNCTIONS D EFINED ON A H YPERCONVEX S PACE
Marcin Borkowski
Abstract We present a xed point theorem for a sum of two convex-valued multifunctions acting on a weakly compact, hyperconvex subset of a normed space. The theorem is a multivalued version of a result of D. Bugajewski.
1.
Introduction
In [3], D. Bugajewski proved the following Krasnoselskii-type theorem in a hyperconvex setting. Theorem 1. Let K be a bounded hyperconvex subset of a normed space (X, ) such that K K for every (0, 1]. Assume that 1. f1 : K X is nonexpansive; 2. f2 : K X is completely continuous; 3. f1 (x) + f2 (y) K for any x, y K; 4. every sequence (xn ) such that xn K for n N and
n
lim xn f (xn ) = 0,
where f := f1 + f2 , has a limit point. Then, f has a xed point. Recall that the assumption that H H can be released, as it was shown in [4]. Recently, M. Ozdemir and S. Akbulut published the paper [6] with a multivalued version of Bugajewskis theorem. Unfortunately, their proof contains some errors. We will state a slightly different version of this theorem and then discuss the errors in [6].
412
Marcin Borkowski
2.
Preliminaries
Let X be a metric space. By B (X) we denote the family of nonempty, bounded and closed subsets of X and by K (X) the set of nonempty compact subsets of X. In what follows, we will use the symbol dX for a metric in the space X and HX for the Hausdorff metric in the hyperspace B (X); we will write d and H if the underlying space is obvious from the context. By A we will denote the complement of the subset A of some space X, i.e., the set X \ A. Denition 1. We call a mapping f : X Y between metric spaces nonexpansive, if d( f (x), f (y)) d(x, y) for each x, y X. Denition 2. Let A be any subset of a metric space X. The Kuratowski measure of noncompactness of the set A, denoted by (A), is the greatest lower bound of the numbers > 0 such that A can be covered by a nite family of sets of diameter not greater than . (We put (A) = + for unbounded sets.) A mapping f : X Y between metric spaces is called -condensing if ( f (A)) (A) for each nonempty A X and ( f (A)) < (A) provided that (A) > 0. The aforementioned notions can be easily transfered to the setting of multivalued mappings. Denition 3. Let X and Y be metric spaces. A multifunction F : X B (Y ) is called nonexpansive, if it is nonexpansive as a mapping from (X, dX ) to (B (Y ), HY ). A multifunction / F : X 2Y \ {0} is called -condensing if (F(A)) (A) for each nonempty A X and (F(A)) < (A) whenever (A) > 0, where F(A) := aA F(a). We will use the following well-known facts about multifunctions (see, e.g., [5]): a nonexpansive, compact-valued multimap is continuous (in particular, upper semi-continuous); moreover, if F1 , F2 are compact-valued and upper semi-continuous multifunctions dened on a subset of a normed space, then so is F1 + F2 . Theorem 2. Let X and Y be metric spaces. Then, F : X K (Y ) is upper semi-continuous if and only if for each sequence (xn ) X converging to some x X and sequence (yn ) Y such that yn F(xn ), there is a subsequence of (yn ) converging to a point in F(x). / Denition 4. Let Z be a nonempty subset of a metric space X and let F : Z 2X \ {0} be a closed-valued multifunction. We call F hemicompact if each sequence (xn ) Z such that limn d(xn , F(xn )) = 0 has a convergent subsequence. Denition 5. Let Z be a subset of a normed space X and let F : Z K (X). We call F strongly continuous if for each sequence (xn ) Z weakly convergent to x Z, every sequence (yn ) X such that yn F(xn ) for n N has a subsequence convergent in norm to some point in F(x). Fact 3 ([8, p. 31, Remark (iii)]). Let Z be a weakly compact subset of a normed space X, and let F : Z K (X) be strongly continuous. Then F(Z) is compact.
Krasnoselskii-Type Theorem for Multifunctions
413
Proof. Indeed, let (yn ) F(Z). For each n N choose xn Z such that yn F(xn ). We can assume that xn x Z. This means that (yn ) has a subsequence convergent to some y F(x) F(Z). Fact 4 ([8, p. 31, Remark (v)]). A strongly continuous map F : Z K (X) dened on a weakly compact subset Z of a normed space X is upper semi-continuous. Proof. Let xn x in X and F(x) V , where V X is open. Assume F(xn )V for inntely / many n N. Passing to a subsequence we can assume that F(xn ) V = 0 for each n N. Choose yn F(xn ) V ; from strong continuity we can assume that yn y F(x). Then y F(x) V contradiction. Now let us recall the denition of a hyperconvex space, introduced by Aronszajn and Panitchpakdi: Denition 6 ([2]). We call a metric space X hyperconvex, if every collection of closed balls {B(xi , ri )}iI in X such that d(xi , x j ) ri + r j for any i, j I has a nonempty intersection. Note that if H is a hyperconvex subspace of a metric space X, then there exists a nonexpansive retraction of X onto H, i.e., a nonexpansive mapping R : X H such that R|H is the identity map on H (see [2]). We will also need a Darbo-type xed-point theorem for multifunctions. In order to state it, we will introduce a so-called U class of multivalued mappings. c First, if X is some class of multivalued mappings, we will denote by X (X,Y ) the sub/ class of mappings in X acting on X and taking values in 2Y \ {0}, and by X c the class of nite compositions of mappings from X . By U we will denote a class of multivalued mappings such that each F in U c is upper semi-continuous and compact-valued (in particular, n n all single-valued, continuous mappings belong to U ) and each F U c (B , B ) has a xed n point, where B is a closed n-dimensional Euclidean ball. Further, given a class U , we will / denote by U (X,Y ) the family of mappings F : X 2Y \ {0} such that for any K K (X), c there exists an FK U c (X,Y ) such that FK (x) F(x) for each x K. We will also note that Kakutani multimappings (i.e., convex- and nonempty-valued, upper semi-continuous multimaps from a subset of a Hausdorff topological vector space to another such space) are an example of U maps (see [7]). c Theorem 5 ([1, Theorem 2.4]). Let X be a hyperconvex bounded metric subspace of a topological vector space and let F U (X, X) be -condensing. Then F has a xed point. c
3.
Main Result
Theorem 6. Let us assume that H is a nonempty, hyperconvex and weakly compact subset of a normed space X, F1 : H K (X) is strongly continuous, F2 : H K (X) is nonexpansive, F := F1 + F2 is hemicompact with convex compact values lying in H. Then, F has a xed point. Proof. Let R : X H be a nonexpansive retraction of X onto H. For each n N, dene / a multifunction Gn : H 2H \ {0} by the formula Gn := R((1 1 )F). Obviously, the map n
414
Marcin Borkowski
F : H K (X) is Kakutani, so it is a U mapping. Now let K H be compact and let c FK U c (H, K (X)) be such that FK (x) F(x) for each x H. Since U c is closed under nite superpositions, R((1 1 )FK ) U c (H, K (X)) and R((1 1 )FK (x)) Gn (x) for each n n x K. This shows that Gn is a U mapping. c Let A H. We have:
1 (Gn (A)) = R((1 n )(F1 + F2 )(A)) 1 ((1 1 )F1 (A) + (1 n )F2 (A)) n 1 (1 1 )(F1 (A)) + (1 n )(F2 (A)) n
(1 1 )(A). n Since weakly compact sets are bounded, we can infer from Theorem 5 that each Gn has a xed point xn . n 1 For each n N, choose xn (1 n )F(xn ) such that xn = R(xn ). Since n1 xn F(xn ) H, and in particular, the sequence (xn ) is bounded, we have:
n d(xn , F(xn )) xn n1 xn = n = R(xn ) R( n1 xn ) n (1 n1 )xn =
1 n1
xn 0.
That means that (passing to a subsequence if necessary) we can assume that xn x H. Choose sequences (yn ) and (zn ) such that yn F1 (xn ), zn F2 (xn ) and xn = R((1 1 n )(yn + zn )). From Theorem 2 we infer that yn y F1 (x) and zn z F2 (x) (again passing to subsequences if necessary). Of course, y + z H. We have:
1 xn = R((1 n )(yn + zn )) R(y + z) = y + z F1 (x) + F2 (x) = F(x)
and the proof is nished.
4.
Remarks
We will now turn our attention to the main result of the paper [6]. It states the following: Conjecture. Let H be a nonempty bounded hyperconvex subspace of a normed space X such that H H for each (0, 1]. Assume that a multivalued mapping F := F1 + F2 : H K (H) is hemicompact. Assume also that F1 , F2 : H K (X) are convex-valued, F1 is strongly continuous, F2 nonexpansive and F1 (H) + F2 (H) H. Then, F has a xed point. Unfortunately, the proof given therein contains a few errors. 1. The authors use the fact that the metric projection of X onto H is single-valued, continuous and nonexpansive. However, none of these statements need be true. Indeed, let X := R2 with the l1 -type norm and H := conv{(0, 0), (0, 1)} conv{(0, 0), (1, 0)};
Krasnoselskii-Type Theorem for Multifunctions
415
then, the metric projection of X onto H fails to be unique at the point (1, 1); moreover, considering the points a := (1 , 1) and b := (1, 1 ) for (0, 1) shows that it is not continuous (and hence also not nonexpansive). (Note that using a nonexpansive retraction instead of the metric projection solves both these problems.) 2. The authors use some properties of strongly continuous multifunctions, analogous to Facts 3 and 4, but without the assumption of weak compactness. The proofs of these facts, however, require that the domain be weakly compact, which went unnoticed in the quoted paper. (In fact, the authors replaced weak convergence by norm convergence in their denition of strong continuity, which resulted in a notion equivalent to upper semi-continuity. Obviously this rendered the analogue of Fact 4 true, but Fact 3 remains false with this new denition.) 3. The authors infer -condensingness of a mapping from its nonexpansiveness; the example of the identity map shows that this inference is false. 4. In the course of the proof, the authors dene multifunctions Gn (n N) as a sum of certain two multivalued mappings; however, these mappings are dened on different domains, so this sum is ill-dened. (It turns out that if we consider this sum on the intersection of the domains of the summands, we arrive at the conclusion that using the metric projection, apart from being wrong, was unnecessary.) In view of the above, the question whether the statement given is true remains open.
References
[1] R. P. Agarwal and D. ORegan, Hyperconvex spaces and xed points, Georgian Math. Journal 9 (2002), no. 2, 199206. [2] N. Aronszajn and P. Panitchpakdi, Extension of uniformly continuous transformations and hyperconvex metric spaces, Pacic J. Math 6 (1956), 405439. [3] D. Bugajewski, Fixed-point theorems in hyperconvex spaces revisited, Math. Comput. Modelling 32 (2000), no. 13, 14571461. [4] D. Bugajewski and R. Espnola, Remarks on some xed point theorems for hyperconvex spaces and absolute retracts, Function Spaces the Fifth Conference (Lecture Notes in Pure and Applied Mathematics Series 213). Edited by H. Hudzik and L. Skrzypczak, 2000, pp. 8592. [5] S. Hu and N. S. Papageorgiou, Handbook of Multivalued Analysis, Vol. I: Theory, Kluwer Academic Publishers, Dordrecht/Boston/London, 1997. [6] M. Ozdemir and S. Akbulut, A xed point theorem for multivalued maps in hyperconvex spaces, Appl. Math. Comput. 157 (2004), 637642. [7] D. ORegan, Admissible spaces and -xed points, Nonlinear Funct. Anal. & Appl. 10 (2005), no. 1, 5564. [8] , New xed-points results for 1-set contractive set-valued maps, Computers Math. Applic. 35 (1999), no. 4, 2734.
INDEX
A
A, 211, 212, 294, 296, 299, 422, 429, 432 Abelian, 305 acoustic, xiv, 247, 348 adjustment, 112 age, xii, 29, 30, 46, 193 algorithm, 206 alternative, xi, 2, 9, 110, 343 ambiguity, 105, 108, 109, 118, 127, 128 amplitude, 348 AMS, 1, 11, 29, 77, 133, 177, 195, 207, 217, 291, 293, 305, 339, 410 Amsterdam, 9, 176, 266, 291 analytic techniques, 128 anomalous, 105 application, xiii, xiv, 46, 175, 177, 195, 200, 288, 290, 291, 293, 302, 356 argument, xvi, 20, 24, 26, 44, 60, 67, 98, 118, 125, 143, 144, 172, 275, 319, 325, 328, 404, 405, 408, 409, 410, 425 assumptions, 20, 30, 32, 41, 42, 44, 182, 213, 218, 219, 224, 225, 227, 230, 232, 233, 234, 235, 236, 238, 240, 241, 270, 272, 335, 406, 408, 430 asymptotic, xii, xiii, xiv, xvi, 11, 25, 47, 54, 107, 133, 134, 135, 154, 157, 160, 177, 178, 179, 180, 184, 186, 187, 188, 189, 191, 193, 195, 247, 248, 253, 256, 258, 259, 260, 261, 262, 265, 266, 345, 365, 377, 384 asymptotically, 12, 48, 70, 72, 77, 157, 160, 180, 181, 184, 185, 186, 219, 220, 365, 383, 429, 430 Asymptotics, 206 atmosphere, 375 atoms, 104 attractiveness, xii, 47, 49, 63, 65, 70 attractors, xiii, 11, 12, 14, 18, 133, 134, 135, 137, 138, 151, 157, 160, 165, 168, 169, 171, 172, 174, 175 automorphy, xiii, 177, 178, 184, 186 averaging, 174 axon, 134, 175 Azerbaijan, 206
B
Banach spaces, xii, xvi, 49, 73, 74, 77, 78, 85, 101, 102, 404, 422 behavior, xi, xiii, xvi, 1, 3, 54, 119, 133, 134, 135, 151, 179, 184, 188, 189, 191, 283, 345, 374, 377, 378, 384, 385, 387, 418 Beijing, 389 bifurcation, 134, 378, 384, 386 birth, 193 blood, xiv, 247, 300, 301 Bohr, 355, 364, 421, 423 borderline, 189 Boston, 9, 439 boundary conditions, xiv, xv, 49, 107, 110, 138, 195, 216, 264, 265, 269, 284, 291, 317, 336, 365, 366, 367 boundary value problem, xiv, 196, 220, 229, 237, 291, 293, 295, 340 bounded linear operators, 51, 79, 80, 179, 208 bounded solution, xii, xiii, 19, 26, 133, 135, 220, 225, 227, 242, 430 bounds, 105, 151, 156, 160, 162, 163, 165, 166, 306 brane, 175 Brownian motion, 418 Burkina Faso, 29
C
calculus, 261, 262, 298
418
Canada, 47 Carnot, viii, xiv, xv, 305, 306, 307, 308, 311, 314 cation, 249, 250, 256, 266, 293, 301, 305, 306, 334, 339 Cauchy problem, xv, 193, 194, 208, 288, 317, 336 causality, 340 CBS, 108, 109, 112 chaos, 387 chaotic behavior, xvi, 377, 387 China, 207, 305, 389 classes, xvi, 38, 90, 385, 386, 387, 410, 411, 412, 415, 421 classical, 18, 32, 40, 220, 221, 224, 236, 238, 271, 281, 318, 321, 328, 331, 349, 366, 367, 368, 372, 404 closure, 12, 26, 37 communication, 340 compatibility, 284 complement, 80, 83, 211, 212, 436 complexity, 18 complications, 105 components, 223, 225, 226, 228, 240, 379, 381, 423 composition, xvi, 231, 234, 236, 240, 421, 422, 427, 430 computation, 78, 79, 90, 102, 309 computing, 30, 78, 95 conduction, 207, 339, 340, 341, 347 confusion, 29, 30, 49, 53, 344 conjecture, xvi, 377, 387 conjugation, 107 conservation, xv, 269, 288, 290, 291, 317, 335, 336, 337, 339 constraints, 32, 33, 34, 104 construction, 1, 205 continuity, 42, 110, 144, 172, 296, 322, 333, 352, 359, 360, 384, 412, 413, 414, 415, 437, 439 contractions, xiv, 74, 293, 294, 302 control, xii, xiv, 29, 30, 31, 33, 38, 40, 41, 45, 46, 77, 78, 288, 293, 377, 388 convective, 270, 276, 283, 284 convergence, xi, 1, 2, 3, 4, 7, 9, 41, 44, 181, 185, 223, 240, 270, 276, 286, 289, 290, 291, 319, 331, 333, 367, 414, 415, 417, 424, 439 convex, xvi, 25, 26, 40, 328, 404, 435, 437, 438 cosine, 208, 209, 210, 213 covering, 159, 160, 169, 170, 411 critical value, xv, 339, 340, 345, 346 cross-sectional, 341 cryogenic, 347 cycles, 384, 386
Index
death rate, 30 decay, 346, 374 decomposition, 15, 54, 67, 153, 323, 384, 425 definition, 3, 14, 26, 29, 77, 79, 80, 82, 90, 93, 97, 99, 103, 105, 108, 109, 110, 116, 119, 120, 122, 126, 127, 128, 129, 180, 181, 185, 355, 378, 410, 411, 412, 421, 437, 439 deformation, 374 degenerate, xiv, 174, 270, 290, 291, 305, 307, 314, 318 density, 12, 179, 188, 249, 275, 281, 284, 319, 325, 339 deposition, xiv, 247 derivatives, 211, 248, 252, 279, 280, 284, 377 deviation, 301 dichotomy, 14 diet, 301 differential equations, xii, xiii, xvi, 9, 19, 26, 27, 47, 48, 49, 73, 74, 78, 79, 90, 95, 100, 102, 177, 193, 219, 220, 224, 242, 244, 259, 295, 349, 356, 403, 409, 421, 434 differentiation, 353 diffusion, xv, 18, 46, 175, 176, 193, 215, 282, 306, 339, 340, 346, 347, 349, 350 diffusivity, 339, 341 Dirichlet boundary conditions, xv, 291, 317, 433 discontinuity, 109, 118, 126, 128 displacement, 248, 256, 258, 259, 260, 261, 262 dissipative system, 11, 174 distribution, 29, 30, 144 divergence, xi, 2, 213, 290 duality, 39, 141 dynamical properties, xvi, 377 dynamical system, xii, 11, 135, 136, 137, 138, 146, 151, 157, 160, 166, 174, 377
E
economics, 388 Eden, 18 Ekman dissipation constant, xv, 365, 366, 372 elaboration, 179 elasticity, xiv, 247, 250, 266 electrons, 104 energy, xi, xiv, 1, 12, 18, 103, 104, 105, 109, 111, 112, 118, 120, 122, 124, 126, 127, 128, 153, 175, 247, 250, 273, 275, 283, 339, 367 entropy, xiv, xv, 269, 270, 271, 272, 273, 274, 275, 280, 282, 283, 288, 290, 317, 318, 319, 320, 321, 328, 329, 331, 334, 335, 364 equality, 33, 91, 98, 274, 275, 279, 298, 332 equilibrium, xii, 47, 48, 49, 50, 51, 70, 247, 249, 301, 383, 388 Euclidean space, 305
D
death, 30
Index
evolution, xi, 3, 11, 18, 27, 46, 101, 102, 134, 174, 175, 178, 179, 180, 181, 184, 187, 191, 192, 193, 219, 244, 276, 283, 353, 378, 429, 430, 432 evolutionary process, 49 expansions, 248, 253, 256, 258, 259, 266
419
F
failure, 109 family, 13, 18, 23, 32, 41, 178, 179, 181, 184, 187, 191, 192, 207, 208, 209, 210, 212, 213, 385, 422, 429, 430, 432, 436, 437 fee, 251 feedback, 388 Feedback Loops, viii, xvi, 377, 379, 381, 383, 385, 387 fertility rate, 30 FitzHugh-Nagumo, xiii flow, 30, 47, 49, 56, 59, 66, 68, 71, 72, 135, 175, 375 fluid, 11, 12, 365, 374 Fourier, 340, 343, 345, 346, 367 Fox, 244 fractal dimension, 11, 14 France, 18, 29, 47 functional analysis, 110 funding, 339
heat, xiv, 207, 208, 215, 217, 220, 229, 237, 238, 306, 339, 340, 341, 347, 353, 421, 422, 432, 433 Heisenberg group, 305 heterogeneous, 317 Hilbert space, 12, 14, 15, 37 Hille-Yosida, xii Hiroshima, 73 Hlder condition, xv Holland, 9, 18, 174, 176, 205, 215, 266, 291 homeostasis, 300, 301 homogenized, xi, 1, 2, 4, 8 homogenous, 9, 249 horizon, 30 hyperbolic, xiv, 200, 269, 270, 271, 272, 273, 290, 291, 317, 340 hyperbolicity, 347 hyperconvex, 435, 437, 438, 439 hypothesis, 301, 313, 320, 321
I
identity, 146, 154, 155, 347, 437, 439 IMA, 175 immunological, 388 inclusion, 15, 426 incompressible, xi, 11, 12 independence, 283 independent variable, 291, 336 Indiana, 353, 375 indices, xvi, 248, 249, 377, 378, 380 induction, 259, 408, 409 inertia, 249 infinite, 19, 26, 27, 78, 99, 179, 184, 374, 395, 397, 401 inhomogeneities, xi, 1 insight, xii, 29, 353 instability, 48, 54, 73, 383, 384, 388 integration, 106, 112, 115, 116, 119, 121, 122, 123, 125, 127, 155, 212, 240, 252, 276, 279, 281, 285, 289, 312, 330 interaction, 104, 377, 378, 379 interactions, 103, 104, 105, 128 interface, 318 interference, 31 interstitial, xiv, 247 interval, 79, 138, 180, 221, 222, 271, 301, 318, 331, 333, 341, 356, 357, 358, 359, 360, 396, 398, 423, 430 inversion, 176, 343 inversions, 378 isotropic, 249, 339 Italy, 11, 73, 177 iteration, 366, 371
G
gauge, 418 Gaussian, 306 gene, 105, 422, 432 generalization, 2, 3, 45, 104, 105, 126, 129, 208, 218, 220, 355, 421 generalizations, 105, 422, 432 generation, 208, 209 Geneva, 130 geophysical, xv, 365, 374 Georgia, 77, 349 Germany, 129 gestation, 179 glucose, 300, 301, 302 graph, 80, 229, 237, 356, 378 Green's function, 107, 109, 200 groups, xiv, 99, 193, 194, 305, 306, 307, 311, 314 growth, 3, 7, 54, 80, 92, 136
H
H1, 30, 251, 270, 271, 276, 282, 287, 289, 329, 390, 391, 392, 394, 397, 399, 400, 401, 402 H2, 30, 282, 284, 329, 391, 395, 399, 400 Hamilton-Jacobi, 350 Harmonic analysis, 315
420
Index
J
Jacobian, viii, xvi, 377, 378, 379, 380, 381, 382, 383, 384, 385, 387, 388 Jacobian matrix, 377, 378, 379, 382, 384 Japan, 77, 418 Jordan, viii, 339, 347, 348, 411, 412, 416, 419
K
kernel, 107, 198, 200, 205, 208, 238, 315, 417 King, 347
L
L1, vii, xiii, 271, 272, 280, 282, 288, 296, 297, 298, 301, 305, 308, 309, 312, 320, 324, 328, 329, 330, 331, 333, 334, 382 L2, 251, 270, 271, 272, 276, 279, 282, 283, 284, 285, 287, 290, 296, 297, 298, 300, 301, 324, 325, 329, 330 Lambert W-function, xv, 339, 341, 344, 346 Laplace transforms, 258 lattice, 79, 81, 82, 93, 96 lattices, 81, 82, 84, 95 law, xv, 288, 291, 317, 335, 336, 337, 339, 340 laws, 269, 290, 336, 337 Lebesgue measure, 282, 318 Levinsons Theorem, vii, xii Lie algebra, 306 Lie group, 306, 314 limitation, 392 linear function, xii, 77, 78, 101, 102 linear systems, 78, 90, 95, 101, 102, 377 Lions sentinel, xii, 29 London, 18, 129, 206, 244, 336, 439 Lyapunov, 134, 302, 383 Lyapunov function, 302 lying, 11, 64, 437
Mathematical Methods, 266 mathematicians, 220 matrices, 388 matrix, xvi, 66, 77, 78, 90, 107, 126, 181, 182, 213, 214, 259, 294, 295, 377, 378, 379, 380, 381, 382, 383, 384, 385, 386, 388 Mb, 270, 284, 285, 286, 287, 288, 289 measurement, 300, 301 measures, 4, 270, 289, 411, 415, 416, 417 media, 9, 175, 269, 347 memory, 217 metabolism, xiv, 293 metric, 14, 294, 296, 302, 355, 356, 364, 405, 436, 437, 438, 439 metric spaces, 294, 302, 436, 439 Mexico, 133 minors, 380 mixing, 46 modeling, xiv, 9, 247, 300 models, xi, 1, 174, 175, 374, 378 modulus, 81, 194, 249, 364 monotone, 328, 329, 331, 332, 335, 366, 370, 408 Morocco, 19, 47, 177 Moscow, 244, 364 motion, 365, 418 multidimensional, xiv, 269, 337 multiplication, 94, 189, 381
N
NAS, 206 nation, 379 natural, 30, 35, 73, 126, 208, 220, 229, 237, 307, 422 nerve, 134, 175 nerves, 174 Neumann condition, 188 New Jersey, 314 New Mexico, 133 New York, 18, 99, 101, 102, 131, 175, 176, 193, 194, 206, 243, 244, 245, 266, 302, 347, 348, 364, 388, 409, 418, 433, 434 Newton, 105, 111, 128, 130, 131 nonlocality, 104, 107, 126 normal, 12, 189, 249, 271, 387 normalization, 317 norms, 12, 13, 14, 16, 35, 87 nuclear, xiii, 103 nuclei, 104 nucleons, 104 nucleus, 196 null-controllability, xii numerical analysis, 288, 349 numerical computations, xi, 1
M
machinery, 350 Madison, 19 manifold, xii, 47, 48, 49, 56, 63, 64, 65, 66, 68, 70, 71, 72, 73, 74, 134, 175, 176 manifolds, 49, 73, 74, 175, 176 mapping, 14, 32, 56, 64, 71, 72, 92, 146, 179, 223, 226, 228, 233, 270, 294, 320, 415, 436, 437, 438, 439 marrow, xiv, 247 Massera type, xii, 19, 20, 26 material sciences, xi, 1 mathematical biology, 220
Index
421
O
observations, 105 Ohio, 103 one dimension, xv, 317, 422 operators, viii, 9, 101, 209, 211, 213, 215, 244, 291, 409, 434 orbit, 385 orthogonality, 315 oscillation, xi, xvi, 1, 215, 384, 385, 389, 390, 403
P
Pacific, 439 pairing, 141 parabolic, xi, xiv, 1, 2, 3, 9, 30, 35, 73, 174, 244, 269, 270, 282, 290, 291, 305, 314, 315, 339, 368, 372, 419, 434 paradox, 340, 347 parameter, xiii, xv, 31, 34, 95, 101, 102, 133, 134, 135, 193, 271, 282, 339, 340, 342, 365, 366, 372, 377, 379, 384, 387 Paris, 18, 46, 73, 244, 290, 291, 336, 388, 433 partial differential equations, xiii, 74, 177, 178, 181, 184, 193, 217, 219, 220, 224, 336 particles, 104, 126, 127, 317 partition, 79, 357, 361 patients, 301 PDEs, 193, 336 periodicity, xiii, 25, 177, 178, 186, 220, 238, 421, 422, 423 perturbation, xiii, 45, 77, 78, 84, 85, 87, 90, 95, 96, 97, 99, 134, 177, 178, 181, 182, 184, 194, 209, 223, 225, 226, 228, 240, 418, 423 perturbation theory, 194 perturbations, xii, 11, 77, 78, 79, 82, 84, 87, 90, 93, 94, 95, 101, 102, 178, 423 phase shifts, 109, 128 phase space, xii, xiii, 11, 19, 20, 48, 54, 100, 133, 134, 135, 177, 193, 349, 377, 379, 384 physics, xiii, 9, 11, 103, 207, 220, 269, 291 physiological, 175 planar, 365 planetary, 374 play, 108, 349 Poisson, 200, 417 Poland, 355 pollution, 30 polynomial, 135, 380, 381, 382, 383, 384, 385, 386 polynomials, 110 population, xii, xiii, 29, 46, 177, 179, 188, 189, 191, 192, 193 porous, 269, 306, 317, 348
porous media, 269 positive feedback, 388 power, 266 powers, 218, 219, 230, 237, 253, 256, 258, 259, 342 prediction, 378 pregnant, 179, 189 press, 206, 353 pressure, 12, 290, 339 propagation, 134, 205, 340, 347 property, 13, 14, 15, 16, 31, 52, 81, 83, 93, 96, 109, 184, 221, 222, 235, 271, 272, 273, 284, 317, 318, 320, 359, 408, 418, 423 proposition, 352, 353 pseudo, xiii, xvi, 217, 218, 219, 220, 223, 224, 225, 227, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241, 244, 421, 422, 423, 425, 426, 427, 429, 430, 431, 432, 433, 434 pulse, 175, 347 PUMA, 302
Q
quasilinear, xi, xiv, 2, 9, 269, 272, 290, 291, 317, 336, 337, 375
R
race, 381 radius, xii, 77, 78, 79, 84, 87, 90, 95, 96, 97, 101, 102, 224, 225, 230, 374, 428 range, 104, 105, 120, 208, 211, 212, 213, 214, 387 real numbers, 50, 51, 71, 250, 309 reasoning, 272, 273, 274, 279, 321, 334 recall, 14, 16, 20, 21, 49, 69, 180, 220, 223, 262, 306, 320, 380, 381, 412, 437 reflexivity, 426 regular, 13, 38, 107, 108, 110, 153, 329 relationship, 25, 251, 421 relationships, 171 relaxation, 215, 339, 340 relaxation time, 339 remodeling, xiv, 247, 248, 250, 251, 266 remodelling, xiv, 247 residues, 344 resolution, 104, 315 retarded systems, 78 Rhode Island, 193 robustness, xiii, 78, 84, 177 Romania, 293 Royal Society, 74, 174 Russian, 244, 314, 364
S
sample, 307
422
scalar, xv, 12, 15, 35, 37, 40, 72, 73, 95, 208, 283, 285, 288, 291, 317, 335, 337, 403 scaling, 135, 251, 260 scattering, 103, 104, 107, 108, 119, 128, 195 search, 133, 422, 429 sedimentation, 317, 336 semicircle, 112 semigroup, 12, 13, 14, 15, 17, 20, 47, 48, 49, 50, 51, 52, 53, 54, 55, 80, 81, 82, 83, 84, 90, 91, 92, 94, 101, 102, 137, 138, 146, 172, 178, 179, 180, 181, 182, 183, 184, 187, 188, 191, 192, 194, 208, 209, 210, 212, 217, 218, 219, 224, 229, 230, 234, 404, 405, 406, 432 separation, 343 series, 340, 342, 344, 367, 378, 417 shape, xv, 349, 353 sign, 270, 275, 325, 327, 329, 342, 351, 372, 377, 378, 379, 381, 382, 385, 386 signals, 340 signs, 378, 379, 382, 385, 386, 388 similarity, 379, 380, 385 singular, xiii, xvi, 105, 134, 195, 196, 206, 306, 310, 311, 314, 343, 410, 412 singularities, 306, 343, 350 smoothing, 153, 175, 212 smoothness, 73, 270, 273, 288, 297, 353 Sobolev space, xiii, 133, 134, 135, 136, 368 software, 345 spatial, 12, 46 spectral analysis, 49, 196 spectrum, xvi, 50, 54, 80, 83, 104, 195, 205, 211, 377, 380, 382, 383, 384, 385, 386 speed, 301, 340 spin, 103 stability, xii, xiii, xvi, 18, 26, 47, 48, 49, 54, 68, 70, 72, 73, 74, 77, 78, 79, 81, 83, 84, 90, 91, 94, 95, 97, 99, 101, 102, 134, 174, 177, 178, 193, 266, 370, 374, 375, 377, 378, 380, 381, 382, 384, 385, 388, 404 steady state, 352 Stieltjes, xiii, 192 strain, xiv, 247, 248, 249, 250, 266 stress, 218, 248, 249, 250, 252, 256, 258, 259, 260, 261, 262 substitutes, 262 substitution, 34, 39, 200 successive approximations, 204, 294, 297 superposition, xv, 355, 356, 358, 359, 360, 363, 364 symmetry, 250 systems, xii, xvi, 11, 18, 74, 77, 78, 90, 95, 99, 101, 102, 104, 174, 175, 242, 336, 375, 377, 378, 387
Index
T
Taylor series, 340 temperature, 339, 340, 341, 345, 346 temperature gradient, 340 thermal conduction, 340 thermal relaxation, 339, 340 threshold, 174 thresholds, 269 time, xii, xiv, xv, 2, 11, 12, 13, 14, 20, 29, 30, 45, 74, 78, 90, 94, 95, 101, 102, 133, 134, 135, 148, 151, 153, 154, 179, 182, 247, 248, 250, 252, 265, 266, 276, 284, 301, 339, 340, 341, 345, 346, 347, 351, 365, 366, 370, 371, 372, 374 time variables, 134, 135, 148 tin, 318, 319, 320, 327 TMA, 175, 242, 243 Tokyo, 18, 77 topological, 174, 196, 437 topology, 14, 15, 48, 56, 134, 137, 171, 229, 237, 330 trabeculae, xiv, 247 trajectory, 188 transcript, 318 transfer, 85 transformation, 196 transformation operators, 196 transformations, 275, 329, 385, 439 transition, 135 translation, 83, 221, 222, 423, 424 transmission, 175, 318, 319 transport, 9, 288, 339, 340, 347 transpose, 66, 383 traveling waves, 134, 135, 174, 175, 348 Tunisia, 410 turbulent, 12 two-dimensional, 266
U
ultrasound, xiv, 247 uniform, xi, xii, 2, 11, 14, 47, 48, 56, 80, 94, 134, 135, 138, 148, 151, 153, 154, 160, 163, 165, 174, 180, 181, 282, 283, 287, 288, 359, 360, 412, 418 USSR, 291
V
values, 13, 26, 31, 38, 48, 65, 78, 79, 104, 108, 110, 120, 122, 139, 202, 203, 248, 294, 301, 317, 344, 345, 346, 355, 359, 360, 363, 377, 378, 379, 387, 404, 437

Research On Evolution Equations Compendium Volume 1

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Research On Evolution Equations Compendium Volume 1

Transféré par

Droits d'auteur :

Formats disponibles

RESEARCH ON EVOLUTION EQUATION COMPENDIUM.

RESEARCH ON EVOLUTION EQUATION COMPENDIUM. VOLUME 1

Nova Science Publishers, Inc.

Copyright 2009 by Nova Science Publishers, Inc.

ISBN: 978-1-61209-404-5 (E-Book)

Published by Nova Science Publishers, Inc.

255 277 287 297 317 327 333

D OUBLE S CALE C ONVERGENCE AND H OMOGENIZATION OF Q UASILINEAR PARABOLIC E QUATIONS

Formulation of the Problem

Double Scale Convergence and Homogenization of Quasilinear ...

Double Scale Convergence and the Main Theorem

u(t, x, s, y) (t, x, s, y)dZ.

t x 1 (u (t, x) u(t, x))(t, x, 2 , )dt dx =

C0 ([0, T ]), c(x)

F(x u(t, x) + y u1 , y, s)dyds = f

u (T, x)(T, x)dx

1 for (t, x) W21 ([0, T ], H0 (U), L2 (U)).

u1 (t, x, s, y)(t, x, s, y)dZ

(3.2) (3.3) (3.4) (3.5)

Double Scale Convergence and Homogenization of Quasilinear ...

Taking limit 0 and using Lemma 3.5, we have

x t u t [b(t)1 (x, , 2 )]dxdt = lim 0 U =

u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ. (3.11)

1 x t (u u)b(t)(1 )s (x, , 2 )dxdt

F0 (t, x, y, s) (b(t)1 + b(t)y 1 (t, x, y, s))dZ =

In above equation, let 1 = 0 , we obtain:

Letting 1 = 0 in (3.12), we have

u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ +

b(t)F0 (t, x, y, s) y 1 (x, y, s)dZ = 0.

Using equation (3.6) with = u ,

Combining (3.17) and (3.18), 1 2 u2 (T, x) dx +

x t F(vk , , 2 ) u dxdt U U 0 0 T x t x t F(u , , 2 ) vk dxdt + F(vk , , 2 ) vk dxdt 0. (3.19) U U 0 f u dxdt

Since u (T, x) , up to a subsequence, weakly converges to u(T, x) in L2 (U), it follows

u1 (x,t, y, s)b(t)(1 )s (x, y, s)dZ f b(t)1 (t, x)dtdx. (3.12)

f b(t)1 (t, x)dtdx. (3.13)

F(x u + y u1 + , y, s) [x u + y u1 ] dZ, (3.22)

F(u + y u1 + , y, s) [u + y u1 + s] dZ, (3.23)

F0 (t, x, y) [u + y u1 + ] dZ. (3.24)

Leting k in (3.21) and using (3.22)-(3.24), we have 1 2 u2 (T )dx +

[F(u + y u1 + , y, s) F0 (t, x, y, s)] dZ 0. (3.25)

Replacing in (3.6) by u(t, x), we have

u (T, x) u(T, x)dx +

x t F(u , , 2 ) x u dydxdt = 0. (3.26) U

Taking the double-scale limit in the above

u(T, x)u(T, x)dx+

F(vk , y, s) [x u + y u1 ] dZ F(vk , y, s) vk dZ 0. (3.21)

F0 (t, x, y, s)x udZ = 0, (3.27)

8 which is equivalent to 1 2 u2 (T )dx +

From (3.28) and above , (3.25) becomes

[F(u + y u1 + , y, s) F0 (t, x, y, s)] dZ 0.

[F(u + y u1 , y, s) F0 (t, x, y, s)] dZ 0,

which implies F0 (t, x, y, s) = F(x u + y u1 , y, s). (3.32)

lim u (t, x) u(t, x) u1 (x,t,

Proof. We consider x t v = x u(t, x) + u1 (x,t, , 2 ) , which double-scale converges to u + y u1 . By monotonicity condition:

x t x t F(u , , 2 ) F(v , , 2 ) (u v )dtdx

L2 ([0,T ],H 1 (U))

Double Scale Convergence and Homogenization of Quasilinear ...

S ECOND G RADE F LUIDS WITH E NHANCED V ISCOSITY AS DYNAMICAL S YSTEMS

address: abonfoh@ictp.it, Phone: +390402240220, Fax: +39040224163

Setting of the Problem

Existence of the Global Attractor

Second Grade Fluids with Enhanced Viscosity as Dynamical Systems