Académique Documents
Professionnel Documents
Culture Documents
Preface 1 Introduction and Background 1.1 Newtonian uids . . . . . . . . . . . . . . . . . . . . . 1.2 The Navier-Stokes equations . . . . . . . . . . . . . . . 1.2.1 The continuity equation . . . . . . . . . . . . . 1.2.2 The Euler equation . . . . . . . . . . . . . . . . 1.2.3 The Navier-Stokes equation . . . . . . . . . . . 1.2.4 Dynamical similarity and the Reynolds number 1.2.5 Vorticity, two-dimensional ows . . . . . . . . . 1.3 Riemann map and Carathodory kernel convergence . e 1.4 Hele-Shaw ows . . . . . . . . . . . . . . . . . . . . . . 1.4.1 The Stokes-Leibenzon model . . . . . . . . . . 1.4.2 The Polubarinova-Galin equation . . . . . . . . 1.4.3 Local existence and ill/well-posedness . . . . . 1.4.4 Regularizations . . . . . . . . . . . . . . . . . . 1.4.5 Numerical treatment . . . . . . . . . . . . . . . 1.5 Complex moments . . . . . . . . . . . . . . . . . . . . 1.6 Further remarks on the Polubarinova-Galin equation . 1.7 The Schwarz function . . . . . . . . . . . . . . . . . . ix 1 1 2 3 4 5 5 7 9 11 13 15 19 20 21 22 23 24 27 27 27 29 35 37 38 41 44
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
2 Explicit Strong Solutions 2.1 Classical solutions . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Polubarinova and Galins cardioid . . . . . . . . . . . 2.1.2 Rational solutions of the Polubarinova-Galin equation 2.1.3 Saman-Taylor ngers . . . . . . . . . . . . . . . . . . 2.2 Corner ows . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Mathematical model . . . . . . . . . . . . . . . . . . . 2.2.2 Logarithmic perturbations of the trivial solution . . . 2.2.3 Self-similar bubbles . . . . . . . . . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
5 Capacities and Isoperimetric 5.1 Conformal invariants an 5.1.1 Modulus of a fam 5.1.2 Reduced modulu 5.1.3 Integral means a 5.2 Hele-Shaw cells with ob 5.2.1 Robins capacity 5.2.2 A problem with 5.3 Isoperimetric inequality 5.4 Melting of a bounded cr
6 General Evolution Equations 6.1 The Lwner-Kufarev eq o 6.2 Quasiconformal maps an 6.2.1 Quasiconformal m 6.2.2 The universal Te
7.4
7.3.3 Central extension Hele-Shaw worldsheet 7.4.1 Integrable hierar 7.4.2 Integrable struct 7.4.3 Hele-Shaw world 7.4.4 Variation of the 7.4.5 Virasoro generat
matter physics, material science, The next important step Yakovlevna Polubarinova-Kochin (1912 1981) in 1945 [117], [262], to deal with non-gravity Hele-S idea was to apply the Riemann (the unit disk in most situations) boundary. The equation for this struct many explicit solutions a geometric function theory to inv to this equation in the case of ad connected domains which have be lent functions. The Lwner-Kufar o study (Charles Loewner or Kare Parfenevich Kufarev, 1909 19 Lwner-Kufarev one, having som o related analytically. The Polubar the corresponding subordination
dalgo, Yu. Hohlov, S. Howison, J Ch. Pommerenke, D. Prokhorov, pecially want to thank their wive inspire our work. Irina Markina second author. The project has b the Gran Gustafsson Foundation o grants #1030373, # 7030011 and grant of the University of Bergen
as the coecient of viscosity and Certain uids undergo very large pressures. Such a uid is ca sity to be constant). In uid dy than incompressible uids. A lam move microscopically in straight tuations, satises Newtons Visco stress in the direction x of ow i orthogonal direction y as
The coecient of proportionality viscosity. Many common uids s are Newtonian. A non-Newtonian proportional solely to the velocit Non-Newtonian uids may not h polymers etc. are not Newtonian.
and the domain of integration de We always assume that a u Let us consider a uid that occu surface S(t). Let N (t) be an ex momentum, or energy. Let x = time. We denote by (x , t) the co the extensive property per unit o N (t) =
V (t)
Reynolds Transport Theorem st time t is equal to the rate of ch rate of ux of N across the contr dN dt =
sys
V(
dt
=
sys
V (t)
sys V (t)
d dt
V d
sys V (t)
ization of the Euler equation wh surface forces. Replacing the stre tain the Navier-Stokes equation. T in vector form for a Newtonian i
DV = Dt
If body forces are negligible, the (1.7) are called the Navier-Stoke
Letting L be a representative scal enclosing boundaries), and U be as the steady speed of a rigid bo x Lx
Let us choose a scaling for the pressure as p U 2 p. Substituting these new values into the Navier-Stokes equation (with Fb = 0) we have DV 1 = p + V , (1.8) Dt R where R = U L/ is the Reynolds number. This equation is just the Navier-Stokes equation in dimensionless variables. Taking into account units = kg , m3 U= m , s L = m, = kg , ms
we reach the conclusion that the Reynolds number is a non-dimensional number. Nondimensionalization, being a seemingly supercial step, becomes important when considering dierent ows with the same Reynolds number. A threeparameter family of solutions for a specic ow is equivalent to a one-parameter family for some modelling ow. Two ows with the same Reynolds number and the same geometry are called dynamically similar. There are two dierent types of real uid ow: laminar and turbulent. A well-ordered ow, free of macroscopic velocity uctuations, is said to be laminar. Fluid layers are assumed to slide over one another without uid being exchanged between the layers. In turbulent ow, secondary random motions are superimposed on the principal ow and there is an exchange of uid from one adjacent segment to another. More important, there is an exchange of momentum such that slowly moving uid particles speed up and fast moving particles give up their momentum to the slower moving particles and slow down themselves. In an experiment in 1883, Reynolds demonstrated that, under certain circumstances, the ow in a tube changes from laminar to turbulent over a given region of the tube. He used a large water tank that had a long tube outlet with a tap at the end of the tube to control the ow speed. The tube went smoothly into the tank. A thin lament of coloured uid was injected into the ow at the mouth as is shown in Figure 1.1. When the speed of the water owing through the tube was low, the lament of colored uid maintained its identity for the entire length of the tube. However, when the ow speed was high, the lament broke up into the turbulent ow that existed throughout the cross section. Thus, laminar ow occurs when the Reynolds number R is not too large. When R is suciently large, then turbulence comes into consideration. It is observed empirically that the ow becomes turbulent whenever the Reynolds number exceeds a certain value R which is critical. The Landau theory of the transition from steady laminar ow to turbulence suggests another limiting critical number R > R . Passing R the ow becomes unstable and bifurcations occur until it arrives at turbulence passing R . For the water ow R = 2, 300 and R = 40, 000 in Reynolds experiment.
water
dye
Hence the ux of volume across any curve joining two points is equal to the dierence between the values of at these points, the function is constant along a streamline, and it is called the stream function. The curl V = is called the vorticity of the uid. In terms of the stream function, = x3 . Taking the curl of Navier-Stokes equation ((1.8)) the term p disappears and one gets an equation in alone D 1 = , Dt R or for the stream function 1 D = (). (1.9) Dt R Equation (1.9) has several benets. For example, it is a scalar equation rather than a vector one. As we have remarked, the ow is laminar until the Reynolds number reaches its rst critical value, or it can be thought of as a slow ow. When the Reynolds number passes its second critical value the ow becomes turbulent and it can be either steady or unsteady. Even though it may be generated by a globally steady
process, such as a steady volume ow through a pipe, turbulent ow is never a locally steady ow. We can see that V can be considered to be the sum of a time averaged value V and a time variable increment V that is usually signicantly smaller than the time-averaged value: V = V + V ,
t+T
1 V = T
t
V (x , )d.
Note that the time-averaged value of V is automatically zero. The random component V of the velocity has some of the characteristics of random noise signals, such as electrical noise in electronic circuits. Obviously, there are small amplitude, high frequency, random motions involved in turbulent ow, the details of which are very dicult to calculate or to predict. Adding the so-called Reynolds turbulent stress into the Navier-Stokes equation gives the equation of turbulent ow DV 1 = p + V V V , Dt R where V V means the vector with k-th coordinate V Vk , k = 1, 2, 3.
40
30
20
10
10
20
30
40
Figure 1.2: Kolmogorovs ow An external force Fext added to equation (1.8) or (1.9) in dierent forms can generate interesting ows. For example, Andrei Nikolaevich Kolmogorov (1903 1987) presented in 1959 a seminar in which he suggested a toy problem with which theorists might explore the transition to uid turbulence in two dimensions. The ow is conceptually simple, and exhibits several shear instabilities before becoming fully turbulent. This ow is governed by the incompressible Navier-Stokes equation (1.8) in two dimensions with a forcing term that is periodic in one spatial direction
and steady in time: Fext = F0 sin(2x2 )x1 . Periodic boundary conditions are assumed in both directions of a rectangular box [0, 1] [0, 1]. Equation (1.9) with the term corresponding to Fext added becomes 8 3 D 1 = () + F0 cos(2x2 ). Dt R R
1 The stationary solution is just st = 2 cos(2x2 ). For small values of the forcing parameter F0 the uid develops a steady state spatial prole corresponding to the spatial prole of the forcing. This ow was named the Kolmogorov ow. Above a critical value of the forcing parameter F0 , the ow becomes unstable to small velocity perturbations perpendicular to the direction of forcing. The resulting ow is a steady cellular pattern of vorticies. More generally, the external force can be chosen to be sin(2nx1 ) cos(2mx2 ) F = F0 . cos(2nx1 ) sin(2mx2 )
For a weak forcing, i.e., for a small value of F0 , the 2n2n array of counterrotating vortices (for the case n = m see Figure 1.2) is the only time-asymptotic state.
10
Remark 1.3.2. Generally, a domain whose universal covering is conformally equivalent to the unit disk is called hyperbolic. So the domain in the above theorem is hyperbolic. If f : UR is the map in Theorem 1.3.1 (or the Riemann map), then the number R = R(, a) is called the conformal radius of the domain with respect to the point a. In the case a = it is more natural to let f map onto the exterior of a disk || > R. Then R = R(, ) is uniquely determined by taking the expansion at innity as f (z) = z + a0 + a1 /z + . . . . One of the principal tools to study evolution of domains is the Carathodory e kernel convergence. Constantin Carathodory (18731950) gave in 1912 [51] a e complete characterization of convergence of univalent maps in terms of convergence of the images of a canonical domain under these maps. Its formulation is found also in [13], [92], [270]. Let {n } be a sequence of domains in the Riemann sphere C such that n=1 a xed point z0 belongs to all n excluding possibly a nite number of them. A domain is said to be the kernel of {n } , n=1 = Ker z=z0 {n }, if satises the following three conditions: z0 ; any compact set of belongs to all n starting with a certain number N ; any domain satisfying the preceding conditions is a subset of . If the point z0 belongs to all n , starting with a certain number N (z0 ), but there is no neighbourhood of z0 that is contained in all n for n > N , then Ker z=z0 {n } = z0 and the kernel degenerates. For the kernel with respect to the origin we write simply = Ker {n }. A sequence {n } is said to converge to the kernel with respect to z0 n=1 if every subsequence {nk } has as its kernel. This type of convergence is k=1 called kernel convergence. If n is decreasing and 0 is the set of interior points of = n=1 n , 0 , then n converges to the component of 0 that contains 0 if 0 0 , or to {0} if 0 0 . Theorem 1.3.3. (Carathodory kernel theorem) Let the functions fn () be analytic e and univalent in U U1 , fn (0) = 0, fn (0) > 0, and let n = fn (U ). Then the sequence fn converges locally uniformly in U if and only if n converges to its kernel , = C, with respect to the origin. If Ker n = {0}, then the limiting function is a univalent map of U onto . If Ker n = {0}, then limn fn (z) 0. The kernel convergence can be generalized to continuous intervals as follows. Let {(t)}, t [a, b] be a one-parameter family of domains in the Riemann sphere C such that a xed point z0 belongs to all (t). Consider rst the case t0 [a, b],
11
and let there be a neighbourhood of z0 that belongs to all (t), t = t0 . A domain is said to be the kernel of {(t)} with respect to z0 , if satises the following three conditions: z0 ; for any compact set D of there is a small positive number , such that D (t) for all 0 < |t t0 | < ; any domain satisfying the preceding conditions is a subset of . If there is no such neighbourhood, then we say that the kernel degenerates and Ker z=z0 {(t)} = {z0 }. A generalized Carathodory kernel theorem states that if the functions f (, t) e are analytic and univalent in U , f (0, t) = 0, f (0, t) > 0, (t) = f (U, t), then the family f (, t) converges locally uniformly in U if and only if (t) converges to its kernel , = C, as t t0 with respect to the origin. If Ker (t) = {0}, then the limiting function is a univalent map of U onto . If Ker (t) = {0}, then limtt0 f (z, t) 0.
12
injection/suction of uid
suggested considering a ow between two parallel horizontal plates with a narrow gap between them. Later a model with slightly dierent geometry appeared in [117], [262], [263], and [283], see Figure 1.3. In this model the viscous uid occupies a bounded phase domain with free boundary and more uid is injected or removed through a point well. The free boundary starts moving due to injection/suction. Similar problems appear in metallurgy in the description of the motion of phase boundaries by capillarity and diusion [241]; in the dissolution of an anode under electrolysis [114]; in the melting of a solid in a one-phase Stefan problem with zero specic heat [66], etc. This book will expose some of the developments in two-dimensional HeleShaw theory that have taken place over the last few decades. Several other models, methods, and applications exceed the scope of our work, such as the treatment of the rectangular dam by Polubarinova-Kochina [264] who gave solutions in terms of the Riemann P -function [67], [181]; mathematical treatment of rotating Hele-Shaw cells [63], [105]; some nice analytical and numerical results found in [54], [55], [56], [247]; a study of Hele-Shaw ows on hyperbolic surfaces [164] [165]; applications to electromagnetic problems [69], [114]; models of diusion-limited aggregation
13
[52], [345], [346]; Hele-Shaw ows with multiply connected phase domains [285]; development of singularities in non-smooth free boundary problems [170], [197], [198]; connections between Stokes and Hele-Shaw ows [68] (a large collection of references on Hele-Shaw and Stokes ows is found in [125]), two-phase Muskat problems [1], [180], [311]; some applications of quasiconformal maps are found in [41], [232]. Recently, it was shown [3], that the semiclassical dynamics of an electronic droplet conned in the plane in a quantizing inhomogeneous magnetic eld in the regime when the electrostatic interaction is negligible is similar to the HeleShaw problem in the plane. Further development of these ideas and applications to the complex moments are found in [206], [231], [344].
x1
Figure 1.4: The section of a Hele-Shaw cell in the x1 -direction the Navier-Stokes equations (1.2), (1.7), which, neglecting gravity, become V 1 + (V )V = (p + V ), t V = 0. (1.10)
14
We assume that the injection of uid is slow enough for the ow to be approximately steady and parallel. This means that V = 0, t These assumptions reduce (1.10) to
V1 x1 + V2 x2 V1 V1 x1 + V2 x2 V2 1 p = x1 + V1 , 1 p = x3 , 1 p = x2 + V2 ,
V3 = 0.
= V2
x3 =0,h
= 0.
If h is suciently small and the ow is slow, then we can assume that the derivatives of V1 and V2 with respect to x1 and x2 are negligible compared to the derivatives with respect to x3 . Therefore, we can simplify the system by putting V1 V2 2 V1 2 V2 = = 2 = x2 = 0, xj xj xj j which gives the system p 2 V1 = , x1 x2 3 p 2 V2 = , x2 x2 3 0= p . x3 j = 1, 2,
The last equation in the system shows that p does not depend on x3 , whence V1 , V2 are polynomials of degree at most 2 as functions of x3 . The boundary conditions then imply V1 = 1 p 2 x1 hx3 x2 3 , V2 = 1 p 2 x2 hx3 x2 3 .
1 V1 = h
h2 p V1 dx3 = , 12 x1
1 V2 = h
V2 dx3 =
0
h2 p , 12 x2
15
(1.11)
Here V and p depend only on x1 and x2 , so we may consider (1.11) as a purely twodimensional equation. Thus equation (1.11) describes a two-dimensional potential ow for which the potential function is proportional to the pressure. By incompressibility (1.2) the pressure is a harmonic function. Equation (1.11) is called the Hele-Shaw equation. It is of the same form as Darcys law, which governs ows in porous media. In the sequel we write just V instead of V . The Stokes-Leibenzon model 0 0 suggests a point sink/source (x1 , x2 ) of constant strength within the system. The rate of area (or mass) change is given as V n ds = const,
U
where U = {(x, y) : (x1 x0 )2 + (x2 x0 )2 < 2 } for suciently small. Equality 1 2 (1.11) and Greens theorem imply (
U
for any . So p = Q(x0 ,x0 ) for some constant Q, where (x0 ,x0 ) is Diracs distri1 2 1 2 bution, and the potential function p has a logarithmic singularity at (x0 , x0 ). 1 2 On the uid boundary the balance of forces in the three-dimensional view gives that p = exterior air pressure + surface tension. The air pressure can be taken to be constant while the surface tension is roughly proportional to the curvature of the boundary. If the gap h is suciently small, then the curvature in the x1 , x2 plane is negligible compared to the curvature in the x3 direction. Due to capillary forces the boundary prole in the x3 direction will be somewhat similar to the graph in Figure 1.4 which is more or less the same everywhere. Hence, the surface tension eect on p is more or less constant (at least with respect to x1 , x2 ). Finally, rescaling p we can take p = 0 on the boundary.
16
resulting equation is known as the Polubarinova-Galin equation (see e.g. [179], [171]) (see a survey on the Polubarinova-Kochina2 contribution and its inuence in natural sciences and industry in [253]). We denote by (t) the bounded simply connected domain in the phase zplane occupied by the uid at instant t, and we consider suction/injection through a single well placed at the origin as a driving mechanism (Figure 1.5). We assume y
(t) x 0 (t)
Figure 1.5: (t) is a bounded simply connected phase domain with the boundary (t) and the sink/source at the origin the sink/source to be of constant strength Q which is positive (Q > 0) in the case of suction and negative (Q < 0) in the case of injection. The dimensionless pressure p is scaled so that 0 corresponds to the atmospheric pressure. We put (t) (t) and assume that it is given by the equation (x1 , x2 , t) (z, t) = 0, where z = x1 + ix2 . The initial situation is represented at the instant t = 0 as (0) = 0 , and the boundary 0 = (0) 0 is dened by an implicit function
2 Pelageia Yakovlevna Polubarinova-Kochina, 18991999. She attended the Pokrovskii Womens Gymnasium in St Petersburg, and after the 1917 Revolution, the University of Petrograd. After graduating, Pelageia Polubarinova worked at the Main Geophysical Laboratory under A. A. Friedmann. In 1925 Pelageia Polubarinova and Nikolai Yevgrafovich Kochin, 19011944, got married. Kochin graduated from the University of Petrograd in 1923. He was appointed to the Leningrad State University (former Petrograd) in 1924 and taught mathematics and mechanics there until 1934. The family moved to Moscow University in 1934. Kochin became the head of the mechanics section of the Mechanics Institute of the USSR Academy of Sciences from 1939 to 1944. The German armies reached the outskirts of Moscow in late 1941 and Pelageia and two daughters were evacuated to Kazan. Kochin, however, remained in Moscow carrying out military research. By 1943 the German army had suered defeats by Soviet troops and Moscow was safe enough for Kochina to return. This she did, but Kochin became ill and died before the end of the war. At the time of his death he had been in the middle of lecture courses and Kochina took over the courses and completed delivering them.
1.4. Hele-Shaw ows (x1 , x2 , 0) = 0. The potential function p is harmonic in (t) \ {0} and p = Q0 (z), z = x1 + ix2 (t),
17
(1.12)
where 0 (z) is the Dirac distribution supported at the origin. The zero surface tension dynamic boundary condition is given by p(z, t) = 0 as z (t). (1.13)
The resulting motion of the free boundary (t) is given by the uid velocity V on (t). This means that the boundary is formed by the same set of particles all the time. The normal velocity in the outward direction is vn = V
(t)
n (t),
where n (t) is the unit outer normal vector to (t). Rewriting this law of motion in terms of the potential function, and using (1.11) after suitable rescaling, we get the kinematic boundary condition p = vn , n (1.14)
p where n = n p denotes the outward normal derivative of p on (t). Let us consider the complex potential W (z, t), Re W = p. For each xed t it is a multi-valued analytic function dened in (t) whose real part solves the Dirichlet problem (1.12), (1.13). Making use of the Cauchy-Riemann conditions we deduce that W p p = i . z x1 x2 Since Greens function solves (1.12), (1.13), we have the representation
W (z, t) =
(1.15)
where w0 (z, t) is an analytic regular function in (t). To derive the equation for the free boundary (t) we consider an auxiliary parametric complex -plane, = + i. The Riemann Mapping Theorem yields a unique conformal univalent map f (, t) from the unit disk U = { : |z| < 1} onto the phase domain f : U (t), f (0, t) = 0, f (0, t) > 0. The function f (, 0) = f0 () parameterizes the initial boundary 0 = {f0 (ei ), [0, 2)} and the moving boundary is parameterized by (t) = {f (ei , t), [0, 2)}. The normal velocity vn of (t) in the outward direction is given by (1.14). From now on and throughout the monograph we use the notation f = f /t, f = f /. The normal outward vector is given by the formula n = f , U. |f |
18
Because of the conformal invariance of Greens function we have the superposition (W f )(, t) = and by taking the derivative we get W Q f (, t) = . z 2 On the other hand, in general for a moving boundary, we have vn = Re [f f /|f |], and nally deduce that Q Re f(, t)f (, t) = , 2 = ei . (1.16) Q log , 2
Galin [117] and Polubarinova-Kochina [262], [263] rst derived the equation (1.16), so (1.16) is known as the Polubarinova-Galin equation (see e.g. [179], [171], [253]). From (1.16) one can derive a Lwner-Kufarev type equation by the Schwarzo Poisson formula:
2
Q f(, t) = f (, t) 2 4
ei + d, |f (ei , t)|2 ei 1
(1.17)
where U . The equation (1.17) is equivalent to the kinematic condition (1.16) on the free boundary. Namely, one can take a limit in (1.17) as tends to a point on the unit circle, and implementing the Sokhotski -Plemelj formulae [242], the equation (1.17) reduces to (1.16). We call (1.17) a LwnerKufarev type equation because of the analogy with o the linear partial dierential equation that describes monotone deformations of simply connected univalent domains (see e.g. [13], [92], [270]). In the classical Lwner-Kufarev equation the integral in the right-hand side of (1.17) is to be o replaced by an arbitrary time dependent analytic function with positive real part. This equation produces subordination Lwner chains whose properties have been o deeply studied. Unlike the classical Lwner-Kufarev equation, the equation (1.17) o is not even quasilinear and produces a special type of chains.
19
20
Another typical scenario is ngering that was rst described in the classical work by Saman and Taylor [293]. Recently it has became clear that in the model with injection ngering does not occur [146].
1.4.4 Regularizations
There are several proposals for regularization of the ill-posed problem. One of them is the kinetic undercooling regularization [172], where the condition (1.13) is replaced by p + p = 0, on (t), > 0. n It was shown in [172], [282] that there exists a unique solution locally in time (even a strong solution) in both the suction and injection cases in a simply connected bounded domain (t) with an analytic boundary. We remark that at the conference about Hele-Shaw ows, held in Oxford in 1998, V. M. Entov suggested using a nonlinear version of this condition motivated by applications. Reissig and Pleshchinskii discussed this model in [258] where local existence and uniqueness were obtained as well as some numerical results were presented. Another proposal is to introduce surface tension as a regularization mechanism. The model with nonzero surface tension is obtained by modifying the boundary condition for the pressure p to be the product of the mean curvature of the boundary and surface tension > 0. Let us rewrite the problem (1.12)(1.14) with this new condition: p = Q0 (z), in z (t), on z (t), (1.18) (1.19) (1.20)
p = vn =
(z), p , on z (t). n
A similar problem appears in metallurgy in the description of the motion of phase boundaries by capillarity and diusion [241]. The condition (1.19) is found in [234] (it is known as the Gibbs-Thomson law or the Laplace-Young condition). Pierre-Simon Laplace (17491827) and Thomas Young (1773-1829) obtained this law independently in 1805. Later Josiah Willard Gibbs (18391903) and William Thomson (Lord Kelvin) (18241907) in the 1870s derived an analogous relation. It takes into account how surface tension modies the pressure through the boundary interface. The problem of existence of a solution in the non zero surface tension case is more dicult. Duchon and Robert [91] proved the local existence in time for weak solution for all . Recently, Prokert [273] obtained even global existence in time and exponential decay (in the case of ow driven by surface tension) of the solution near equilibrium for bounded domains. The results are obtained in Sobolev spaces H s with suciently big s. We refer the reader to the works by Escher and Simonett [106], [107] who proved the local existence, uniqueness and regularity of strong
analytically in many cases and s merical scheme, which uses a n [10]. In addition, the possibility o resolution. The cusp singularitie singularities in the presence of s the low error level can lead to components [35], [71]. The bound gration is typically employed, se technique of Hou et al [174] to re by surface tension. A good revie [175]. Some other relevant refere There are several works wh the free boundary problem is r Numerical simulations are also p strated. For example, the Hele-Sh a non analytic perturbation of a contour method in order to give
d dt M (t)
=
(t)
= =
(t)
(t)
Integrating we obtain
(z)d
(t)
for all t [0, t0 ). It is easy to see (see, e.g., [144]) to show that a s is a strong solution to the Heleholds for every analytic and inte
(1.22). To this end, following [60 independent variables and an Then (u, v (, and (1.22) becomes
(u (
=+ t=
where = (, ) is an arbitrar 1(
where the function S(z, t) is den function satises the consistency with respect to an arc length par
for the unit tangent vector on The map z S(z, t) has t ection in (t). Therefore, if ( point S(z, t) moves, for xed z, w direction into account this gives
It is possible to construct many e the nonlinear Polubarinova-Galin form of the parametric univalen expansion/shrinking of the disk c when the uid can be completely obvious form f (
Here 0 is a disk centered on th (Q < 0) and t [0, |0 |/Q] in th it is possible to start with 0 =
-2.5
-5
-7.5
-7.5
-5
-2
reaches the sink. This phenomeno that Galin knew that, but did no [171]. Considering a general poly
n
f (, t) =
k=1
After this rst non-trivial Polub tions were constructed. Among Taylor that will be discussed in t channel. In this section we shall univalent functions. One nds the The rst explicit rational solutio [214]. Unlike the previous case ra the free boundary reaches the sin Let Q > 0 and consider the
f (, t where a(t) = 24 2 23
Figure
f (, t
with the parameters a, b, c chos consists of two equal disks touch give only a sketch in Figure 2.3.
it is convenient to introduce the call a family of domains {(t)} s for each t, and the normal veloci (t). Then a strong solution of th family {(t) : 0 t < t0 } su (the function p will be uniquely (t)). If the domains (t) are Polubarinova-Galin equation (1.1 In the above denition the in or half-open interval. Given a domain (0) with posed case Q < 0 there exists a [0, t0 ). In the ill-posed case Q > 0 (see, e.g., [330]). Since we do not know any r results we shall not include any
G(f )() =
Q 4
1. If f is holomorphic in UR f UR .
It is important for these conclusi U and has no zeros there (where To prove 1) and 2) we rst G(f )() =
Q 4
one at innity may move around exception that the pole at innity
Remark 2.1.2. We shall see later Q < 0, the radius of analyticity R f (, t) is holomorphic in UR , is exists for all 0 t < we shall poles of f (, t) will not cause any down in nite time it will be beca f (, t) reaching U or of f (, t)
Proof. In order to avoid too man only two poles, one nite pole an
f (, 0) =
for suitable coecients Bk (t) an the formula (2.8), that these n coecients of f (see [143] for de Now the Polubarinova-Gali
f(
Inserting here the above expres tifying coecients gives a system and one sees immediately from t solved for the time derivatives a Polubarinova-Galin equation red dierential equations of standard solution, at least for a short twoAnsatz (2.10) was successful so same form for a little while. Th about collision, which will be dis
Figure 2.4:
type of boundary conditions after the French mathematical physic [36], appeared in connection with the Neumanns). Robin completed a doct probable that this attribution does no name in this context is widely used no
Im
where c1 is some real constant. zontal shift, is h() = (c2 To determine the constants we obtain
AR
Having handled these rst steps solutions. We should say that mos and rational functions, or else, lo gers. Another type of explicit solu Cummings [67], who reduced the inating time by applying the Ba making use of the Riemann P-fu Corner ows of an inviscid i in [191], [192], [235], [256], [332] we mention here papers [17], [32 constructed by Kadano [188] is In this section we shall con arbitrary angle such that the visc extends to innity along it and h at the other wall. These solution trivial (circular) solution. In the [188]. Then we present an analo
Let us mention here that th tension and the macroscopic cont can then be dierent from /2. moving contact line obeys intere
We suppose that the motion is d Since the angle between the wall innity as Q p where Q corresponds to the co (Q > 0). Finally, we assume that We introduce Robins comp where is the stream function. -plane, = + i. We set
D D3 D1 D2 D
= {
= {
= { = {
= D
near innity and a0 (t) > 0. The domain (t) by j = {z : z = f ( The normal unit vector in t n = f on 3 , |f |
vn = V n =
p = n
that produces a circular dynamic trivial solution by a function ind in the form
f (, t
near innity. The branch is chos the reection of D, is real at real g satises Re (g()
g() + g
0+0
Therefore, C = . We present nu The special case with angle hypergeometric function reduces
f (, t) = ( 4Qt/
This function maps the domain domain bounded by the imagina of the real axis and an analytic see Figure 2.8.
By the analogy with an in source (see Figure 2.9). The phas the vertex of the corner which i the walls lies on the real axis an origin. We set G = { : || < 1,
for
2|Q|t
Remark 2.2.1. By the proposed m solutions even for more general o be a known solution to the problem Galin equation Re (f f ) =const circular component of the param of the form f (, t) = f0 (, t) + g
So one must solve the equation vanishing real part at the points
In this subsection we discuss def ow in which there is a replacem cous and the other is eectively drop-shaped solutions in a corner those constructed in [17], [328] a cous ngering in corners applying potential. In the symmetric case terms of hypergeometric function solution was given. The bubbles are assumed to bubble-wall contact angles are one of the walls and x the angle
Re (f
Im (fei ) = 0 for
We are going to construct the corner ows (self-dilating dro solutions were discovered rst in values of angles in [32], [33], [34] possible self-similar solutions, and symmetric drops. To simplify matters we sca us analyse the auxiliary mapping
where g() is an analytic functio The boundary 3 starts an contact angles (0, ), and for the function g() can be represen
g()
g g
2(2 + 1) 4( 1) + 2 2 1 ( 1)2
G(
although other forms may be po ferential equation for which we c the form
(1 w)wY + (1 (
is constant, we can choose C1 , C satises the equation (2.20) in th that the function f maps the ray respectively. In order to check th Q and f of the form (2.23), we c The function f is starlike with r univalent. If the constant C2 va is easily veried. This means tha bisectrix of the phase angle, nam In Figures 2.13, 2.14 we pr (a,c), as well as the symmetric ca In the case = 1/2 the hyp h1 () = 1 2
h2 () =
1 2(1 2)
and we have f (, t) =
t A( + 1)1
where (0, 1/2), Q = 4AB(1 that the map f (, t) becomes no The function f (, t) obviou D onto (t) that is a complem 3 starts and ends at the origi forms a self-similar drop-shaped with respect to the bisectrix of th the solution is known [17], [328]. (Figures 2.13(a,c), 2.14(a,c), 2.16 wall angles are the same, we hav ble developments of ngers. For
k=1
1 1 + 1 +k
This type of transformation, wit was used by C. Baiocchi in [20 of the so-called dam problem. F inequality formulations (in some 1980 by Elliott, Janovsk [102], S y To arrive at the concept o of (1.12)(1.14) with Q < 0, i.e (1.12)(1.14) hold. For simplicity function of (t) with the singula
p(z, t) =
dz =
(t)
(t)
always a weak solution. In the cas we may dierentiate (3.6) with r Using that u = |u| = 0 on ( (t), it also follows that u = t function u is indeed the Baiocch There are other types of s Crandall and Lions [62] introdu successfully used to study nonli and Vzquez [50] proved the ex a for the porous medium equation, Hele-Shaw problem.
(z) =
1 2
(0)
and set
We can think of as an obstacle ing the smallest superharmonic f from general potential theory (se lower semicontinuous regularizati functions . Superharmonic fu continuous.
In order to have nite integr out 0 a little (e.g., replace 0 by settings then are that one works case of u). Both problems (i) an characterized by their variationa The variational formulation
UR
u + (
u +
which means that we have establ that (t) was dened to be satu
It is clear from the way the conc solution always is a weak soluti solution by the uniqueness of th be a strong one, e.g., because a w which is not allowed even in the
there. Thus (s) (t). (d) Fix s > 0, t > 0, and set w
(e) By (b) (0) (t) so (3.2 amounts exact to (3.2) hol z C, and () = log |
Proof. So assume (t) is simply map, z = f (, t). Using the pote Schwarz function, dened in (t)
We see immediately from (3.6) t continuously dierentiable away C \ (t), and |u| = 0 there, S(z (t). The conjugate of S(z) can (t) and we use it to extend f by f (1
In the above curve integrals we t Thus, the function f (, t) is anal For any pair of numbers s function h(, s, t) f 1 (f (, s), t A simple application of the Schw
Therefore, r(t) r(s), hence R(s We have f (U, t) = (t) univalent on U . Therefore, (t)
Remark 3.4.3. More generally, t component of (t). A dierent [147]. There it was shown that th transform directly gives a real an
= {the largest
1 The idea of balayage (sweeping i Friedrich Gauss, 17771855, see [84] fo sweeping operations was invented by P
Bal ((s
Bal (1 + 2 ,
Bal (1 + 2 ,
min (
Bal ((0)
corresponding to some kind of m As a useful application of ( r > 0 so small that r2 < t and 1 = |U be the Dirac measure swept out
> 1, (3.1
Bal (t0 + 0 , 1)
Ur
and a measure on D which sa = 0 outside D, and for which B outline the construction, more d Kovalevskaya theorem we solve t u = u = u =
u(z) =
for some domains (t), t < One may notice, using (3.17 Bal ( + t, 1) = = =
so the family (t) really is an construction of , taking t = 0 in for t > 0 one gets the usual forw on all t < .
Remark 3.6.2. By the formula ( t = 0, the solution {(t)} repre sitting compactly in (0). Moreo [47], [304], that the solution {( Therefore, the solution obtained w problem.
Figure 3.1:
Remark 3.6.4. If the initial dom then the strong solution exists loca
Therefore, w 0 in G(t). He subharmonic in the connected se w 0 in C \ H(t) by the maximu big norm, only the second option by the equation (3.22) implies th
By Proposition 3.3.1 (e) we the saturated sets which satisfy |H(t)| = |G(t)| = 0, it follows f This ends the proof.
Problem 3.6.5. What happens w direction (suction) breaks down? it possible to continue the solutio time? Compare [170], [177].
holds for all integrable subharm (3.2), says that (t) is a quadra measure (t) = (0) + t0 and family of domains (t) is a weak assumed saturated) is a quadratu if Bal (, 1) = . If Bal (, 1) is term c , then there exists no q . In case is harmonic the both and are then subha equality we may also consider an particularly rich theory then aris for measures with support in a n to be an arbitrary distribution wi at the following classical concep is called a (classical) quadrat z1 , . . . , zm and coecients ck
If is simply connecte
Clearly the function S(z) w it can be shown [4], [142] that th algebraic curve. This curve may which in the simply connected ca f is not univalent in a full neig for another 1 U ). Away from neighbourhood of , hence is a only a one-sided Schwarz functi
ge (z) = 1
S(z) =
k=
m n
f () =
k=1 j
Here k U are the points which zk . The expressions for the coe complicated because of the nonli Quadrature domain theory erties of Hele-Shaw evolutions, a
(t)
0 t /Q. The correspondin functions (2.3) with one pole at z = 0, and one nite pole = point 1/ = c(t) is mapped onto perfectly ne, and the same argu of the Polubarinova-Galin equati quadrature domain with 0 However, these rational solu cause for polynomial solutions th of the kind
dz = (c0 Qt)
(t)
k=1 (z
zk )n
To get hold of the remaining po quite dicult in general. See [64] of applications of quadrature do [65].
The governing law (Darcys law) Shaw cell makes sense, in its mat three space dimensions (and hig
[115] combined with geometric r true [146]). Classical (strong) solutions with and without surface tensio example [21], [107], [108], [330]. prove is the existence and uniqu strong solutions under the assum analytic boundary. Under certain domain, it is possible even to pro time direction (injection) [152]. A A generalization of the Hele the Laplace operator by a more g allows for considering irregular medium interpretation of the mo is [30], [124], [226], [232], [279].
Let us consider the problem parameterized by a univalent fun normalized as f (, t) = a(t) + a Using the Lwner-Kufarev t o a(t) =
Q 2
a or
a2
dist((
any clear geometric interpretatio Brannan, Kirwan [44] and Stank said to be strongly starlike of ord
The set of all such functions is better geometric description: ev S () is reachable from outside b The inequalities (4.14.3) a tion f to be univalent (see [19] lence). Kaplan [190] proved that Re
We dene a subclass HR () of i), (, )} are reachab co-directed with R . We call th direction (see Figure 4.3). The ne function f to be convex of order
arg f () <
Proof. If we consider f in the clo be replaced by () where equalit The proof is based on consi image of U under the map f ( say there exist t 0 and 0 = ei arg
0 f f(
ei f f (e
Im
f (, t)
f (, t) f t
|f (, t)|2 Im
= Im (f (, t)f(
ar t =
Proof. Let t0 be such that the s [0, t0 ), t0 > 0. Since all function a neighbourhood of U during th (1.16), their derivatives f (, t) a f (, t) are starlike in U (see Th (t) 1, such that f (, t) S ( Let us x t [0, t0 ) and co f (,t ) | arg f (,t ) | = /2. First, we d
ar
(, t) f (, t)
Im
A+ ,t=t
The right-hand side of this equ because of (4.13), (4.15). Therefo max
A+
ar t
f (, f (, t
(t) :
Similar argumentation shows tha If A = (or A+ =), the suciently small. We set the function (t) A =. The so-dened function
Since f (ei , t) = 0, we have that the image of the unit disk U und axis at the point 0 = ei0 , we de
Re 1
Im 1
level lines of a close-to-convex fun to be CR . Here we give an exampl do not preserve the property of t It is known [190] that close-t assertion: for any 1 , 2 such tha
2
1 + Re
1
rei f
H(1 , 2 , f (, t)) =
1
t=0
f0 () =
0
exp
() = 3 (1 + sin ( ( ))), = 2 satises the condition f0 () = f obtain that the right-hand side in Im
ei f0 (ei ) f0 (ei ) 2
= 0.
1
4.3. Hereditary shape of phase domains Calculate, integrating by parts ei f0 (ei ) Im f0 (ei )
2 1
87
32 = 2 2
(4.25)
From the obvious inequality < 2 < it easily follows that the right-hand side 2 of (4.25) is strictly negative and remains negative for f (, t) in some time interval t [0, ). To complete the proof we show that f0 () is close-to-convex and univalent, verifying the condition Re f0 () 0. This condition is equivalent to the inequality 2
1 Re 2 2
(()
0
ei + ) d . 2 ei 2
2
The right-hand side inequality is equivalent to 0 (() )P (, )d 0, where P (, ) Re S(, ) is the Poisson kernel. The sign is obviously veried. The left-hand side inequality can be considered analogously.
where is the curvature of the boundary and is surface tension. The problem of the existence of the solution in the non-zero surface tension case was discussed in Section 1.4.4. Now we obtain the equation for the free boundary using an auxiliary parametric univalent map. To derive it let us consider the complex potential W (z, t), Re W = p. For each xed t this is an analytic function dened in (t) which solves the Dirichlet problem (4.26)4.27 in the sense that its real part induces the same distribution as the solution of the problem (4.264.27). In the neighbourhood of the origin we have the expansion W (z, t) = Q log z + w0 (z, t), 2 (4.29)
88
where w0 (z, t) is an analytic regular function in (t). To derive the equation for the free boundary (t) we use the same arguments as in Sections 1.4.1, 1.4.2 and consider the Riemann conformal univalent map f (, t) from the unit disk U = { : |z| < 1} into the phase plane f : U (t), f (0, t) = 0, f (0, t) > 0. Then the moving boundary is parameterized by (t) = {f (ei , t), [0, 2)}. The normal velocity vn of (t) in the outward direction is given by vn = p/n. The normal outer vector is given by the formula n = f , U. |f |
The superposition (W f )(, t) is an analytic function in the unit disk. Since the Laplacian is invariant under a conformal map, the solution to the Dirichlet problem (4.264.27) is given in terms of the -plane as
2
Q log + 2 2
(ei , t)
0
ei + d + iC, ei
(4.30)
Re 1 + ei f (ei , t)/f (ei , t) , [0, 2). |f (ei , t)| Im e2i Sf (ei , t) (ei , t) = , |f (ei , t)|
2
W Q f (, t) = z 2 2i
89
On the other hand, we have vn = Re f f /|f |, and applying the Sokhotski Plemelj formulae [242] we, nally, get 2 Sf (, t) Q H iIm (), Re f(, t)f (, t) = 2 |f (, t)| = ei , where the Hilbert transform in (4.31) is of the form
2
(4.31)
1 p.v.
For = 0 we just have equation (1.16). From (4.31) one can derive a Lwner-Kufarev type equation by the Schwarzo Poisson formula:
2
1 f = f 2
1
0
|f
(ei , t)|2
where U . An interesting question appears when 0. It turns out that the solution in the limiting -surface-tension case need not always be the corresponding zero surface tension solution (see the discussion in [310], [324], [330]). This means that starting with a domain (0) = (0, ) we come to the domain (t, ) at an instant t using surface tension and to the domain (t) at the same instant t in the zero surface tension model. Then the domain lim (t, ) = (t, 0) is not necessarily the
0
same as (t) (see numerical evidence in [277]). If the boundary is highly curved, then the condition (4.27) must be used even though is small. Obviously, the nonzero surface tension model never develops cusps. Thus, solutions and geometric behaviour of the free boundary for small are of particular interest.
f () f ()
f () f ()
() + Im f () Im 2 Sf () f
|f ()|
90 Proof. We denote
2
i () =
Then, by the Sokhotski -Plemelj formulae we deduce that lim () = i Im e2i Sf (ei ) Im e2i Sf (ei ) H i (). i , t)| |f (e |f (ei , t)|
(10)ei
We note that the second term in the above relation is real, which can be easily seen by the denition of and the Schwarz integral formula. The left-hand side in (4.31) is dierentiable in and that is why one can calculate the derivative in question as the limit Im e2i Sf (ei ) H i () = Im |f (ei , t)| We calculate i () =
2
(1 )ei
lim
().
() =
A(ei )
0
d . ei
Thus, we apply the Sokhotski -Plemelj formulae once again and get the assertion of Lemma 4.3.5. Theorem 4.3.6. Let Q < 0 and the surface tension be suciently small. If the initial domain (0) is strongly starlike of order , then there exists t = t() t0 , such that the family of domains (t) (in the sequel, the family of univalent functions f (, t)) preserves this property during the time t [0, t()]. Remark 4.3.7. For = 0 we have the result of Theorem 4.3.2. Proof. If we consider f in the closure of U , then the inequality sign in (4.3) can be replaced by () where equality can be attained for || = 1. We suppose that there exists a critical map f S () of exact order , that is the image of U under the map f (, t)/f (, t), || 1 touches the boundary rays l of the angle arg w [ , ], say there exist such t 0 and 0 = ei0 , 2 2 that 0 f (0 , t ) = (or ), arg (4.33) 2 f (0 , t ) 2
4.3. Hereditary shape of phase domains and for any > 0 there is such t > t and (0 , 0 + ) that arg ei f (ei , t) f (ei , t) 2 (or ). 2
91
For deniteness we put the sign (+) in (4.33). Without loss of generality, let us assume t = 0. Since f (ei , t) = 0, our assumption about the sign in (4.33) yields that 0 f (0 , 0) > 0, (4.34) Im f (0 , 0) (the negative case is considered similarly). Since 0 is a critical point and the image of the unit disk U under the mapping f (,0) + i0 , we deduce that f (,0) touches the ray l at the point 0 = e ei f (ei , 0) arg f (ei , 0) rei0 f (rei0 , 0) arg r f (rei , 0) We calculate Re 1 + 0 f (0 , 0) 0 f (0 , 0) = 0, f (0 , 0) f (0 , 0) 0 f (0 , 0) 0 f (0 , 0) 0. f (0 , 0) f (0 , 0) (4.35) =
=0
0, 0.
r=1
(4.36)
f (, t) f (, t) arg = Im log = Im t f (, t) t f (, t)
t f
(, t) f (, t) t f (, t) f (, t)
(4.37)
We now dierentiate the equation (4.31) with respect to using Lemma 4.3.5, Im f (, t) f (, t) f (, t)f(, t) 2 f (, t)f(, t) t = H[iA](), (4.38) for = ei . This equality is equivalent to the following: |f (, t)|2 Im
t f (, t) f (, t) t f (, t) f (, t)
= Im (f (, t)f(, t))
f (, t) f (, t) +1 f (, t) f (, t)
H[iA]().
92
Substituting (4.31) and (4.35) in the latter expression we nally have f (, t) arg t f (, t) Im = 1 |f (ei0 , 0)|2 e2i Sf (ei , t) Q + H iIm (0 ) 2 |f (ei , t)| H[iA(ei )](0 ) . |f (ei0 , 0)|2 (4.39)
=ei0 ,t=0
The right-hand side of this equality is strictly negative for small because of (4.34), (4.36). Therefore, ei f (ei , t) arg < i , t) f (e 2 for t > 0 (close to 0) in some neighbourhood of 0 . This contradicts the assumption that (t) fails to be starlike for t > 0 and ends the proof for the class S ().
Problem 4.3.8. Let (0) be starlike of order (not strongly), f (, 0) S , and Q < 0. Do the domains of the Hele-Shaw evolution (t) remain starlike of the same order?
Problem 4.3.9. Let (0) be convex. It is known that for Q < 0 the Hele-Shaw evolution (t) is not necessarily convex. Are there conditions that guarantee convexity of (t) for all the time? Problem 4.3.10. If Q = 0 and = 0, then the driving mechanism is only surface tension. Let (0) be convex. Are the domains (t) convex for all t?
93
(t) x 0 (t)
Figure 4.4: (t) is the complement to a bounded simply connected bubble with the boundary (t) and the sink/source at innity
where Q is the rate of bubble release caused by air extraction, Q < 0 in the case of a contracting bubble, Q > 0 otherwise. For this choice of Q the case of Q < 0 is stable whereas Q > 0 is not. Mathematical treatment for the case of a contracting bubble was presented in [104]. In particular, the problem of the limiting conguration was solved. It was proved that the moving boundary tends to a nite number of points which give the minimum to a certain potential. There an interesting problem was posed: to describe domains whose dynamics present only one limiting point. Howison [178] proved that a contracting elliptic bubble has a homothetic dynamics to a point (in particular, this is obvious for a circular ` one). Entov, Etingof [104] (see also [333]) have shown that a contracting bubble which is convex at the initial instant preserves this property up to the moment when its boundary reduces to a point. These domains are called simple in [104]. Now our parametric domain is the exterior part of the unit disk, and there exists a unique conformal univalent map F (, t) from the domain U = { : || > 1} into the phase plane F : U (t), F (, t) = a + a0 + a1 + , a > 0. By a shift we assume 0 closure(F (U , t)). We repeat the calculations of the preceding subsections taking into account that the normal vector is calculated as n = F /|F |, || = 1,
Q . Re [F (, t)F (, t)] = 2
(4.43)
The Lwner-Kufarev realization of this equation is easily obtained by analogy with o the equation (1.16). Hereditary properties We call the problem with injection at innity the outer problem. In other words we consider the stable case of a contracting bubble, Q < 0. Here we prove that a contracting starlike bubble preserves the property of starlikeness and directional convexity. Let us suppose that the complement of the uid domain contains the origin and is starlike with respect to the origin at the initial instant. Therefore, if a function F () = a + a0 + a1 / + is dened outside of the unit disk, then the function f () = 1/F (1/) is holomorphic in U . Then the equation (4.43) can be rewritten in terms of this holomorphic function as Q|f (, t)| Re [f(, t) f (, t)] = , 2
4
|| = 1,
Q < 0.
(4.44)
The condition of starlikeness Re F /F > 0, || > 1 is equivalent to Re f /f > 0, || < 1. We must control the sign of the functional t arg f /f . Dierentiating (4.44) with respect to we obtain f (, t) arg = Im t f (, t) = 1 Im |f |2 f f 1 f f + f f f f f f +4 f Q|f |4 f 2
Re 1 + Im 1 + Finally, we have
0 f (0 , 0) 0 f (0 , 0) = 0, f (0 , 0) f (0 , 0) 0 f (0 , 0) 0 f (0 , 0) 0. f (0 , 0) f (0 , 0)
(1
f f f + )+6 f f f
(4.45)
95
The right-hand side of (4.45) is strictly negative due to the previous chain of inequalities. Therefore, we have the armative answer in the case of a contracting bubble. Theorem 4.3.11. Let Q < 0. If the domain of a contracting bubble D0 = C \ 0 is starlike (or strongly starlike) with respect to a point z0 D0 at the initial instant, then the family of functions f (, t) and the family of the domains D(t) = C \ (t) preserve the same property as long as the solution exists and z0 D(t). In particular, a convex domain D0 is starlike with respect to any point from D0 , and therefore, the convex dynamics is also preserved, as was proved earlier in [104], [333]. Remark 4.3.12. Let us give a remark concerning the above result. It can be formulated as follows: if we nd a point z0 in D0 with respect to which D0 is starlike, then the domains D(t) are also starlike with respect to the same point z0 during the existence of the solution or up to the time when z0 (t). This means that if D0 is simple (in the terminology of [104]), z0 is a limiting point in which D(t) contracts, and D0 is starlike with respect to z0 , then D(t) remains starlike up to the instant when all air is removed (there exist non-convex simple domains, see [104], [333]). Similarly we establish the following. Theorem 4.3.13. If the initial domain is convex in the direction of R and symmetric with respect to the imaginary axis, then the contracting bubble preserves this property as long as the solution exists. Let us set up in the table the information about the dynamics for bounded and unbounded domains for the inner and outer stable (well-posed) problems known at the moment using special univalent functions. Here yes means that the property is preserved whereas no means that is not. For the outer problem we consider the complement to (t). Class of univalent functions starlike (or strongly starlike) convex close-to-convex convex in the direction of R (with the condition for level lines) Inner problem yes no no yes Outer problem yes yes no yes
For the ill-posed case fewer results are known. Injected air forms a bubble which grows as time increases. It was shown [178] that three kinds of behavior can occur. Firstly, the solution may cease to exist in nite time; secondly, the solution may exist for all time and the free boundary may have one or more limit points
96
y (t)
0 (t) x
Figure 4.5: (t) is an innite domain with the boundary (t) extending to innity and the sink/source at the innity
as t tends to innity; and thirdly, the bubble may exist for all time and ll the whole space as t tends to innity. Making use of quadrature domains it has been proved that the only solutions of the third kind are those in which the bubble is always elliptical. The multidimensional case was treated in [82].
97
We assume that the velocity tends to a constant value Q as x , that is positive when uid is removed to the right and negative otherwise. In terms of the potential p we have p(x, y, t)/x Q as x for any t xed. The problem of the existence was discussed in Subsection 1.4.4. It is noteworthy that for this case the local solvability and uniqueness was proved by Kimura [195] in presence of surface tension. We consider the auxiliary parametric complex -plane, = + i. The Riemann Mapping Theorem yields that there exists a conformal univalent map f (, t) of the right half-plane H + = { : Re > 0} into the phase plane f : H + (t). The half-plane H + is a natural parametric domain for . The function f (, 0) = f0 () produces a parametrization of 0 . The smoothness of the boundary (t) and its behavior in the neighbourhood of allows us to assume the normalization f (, ) = a + a0 + a1 + , , a > 0, i.e., the function f has an analytic continuation on the imaginary axis H + near . Thus, the moving boundary is parameterized by (t) = f (H + , t). The normal velocity vn of (t) in the outward direction is given by vn = p/n. The normal exterior vector is given by the formula 1 f f n = , H + . The harmonic function p is a linear one. The normalization about innity implies that W f = Q and the Polubarinova-Galin equation is of the form Re f(, t)f (, t) = Q, Re = 0. (4.49)
The application of the Schwarz integral formula enables us to deduce a Lwner o Kufarev type equation in the right-hand half-plane f 1 f = t
Q d , H +, |f (i , t)|2 i
(4.50)
with the initial condition f (, 0) = f0 (). Taking into account surface tension this equation becomes iIm Sf (), Re = 0, Re f(, t)f (, t) = Q + H |f | with the Hilbert transform dened as 1 p.v. i
(i )d = H[]().
98 Hereditary properties
We are going to prove that if the initial interface possesses the property to be convex of order in the negative direction (HR ()), then the free boundary remains convex in the negative direction of the same order in so far as the solution to the Hele-Shaw problem exists in the case Q < 0 (the liquid moves to the left). An important remark is that the level lines of a function from HR () remain convex in the negative direction. We are going to prove the following statement. Theorem 4.3.14. Let Q < 0 and (0) be a domain convex in the negative direction of order (). Let the solutions to the equation (4.49) exist during the time t [0, t0 ]. Then, for all t [0, t0 ] the family of functions f (, t) and the family of domains (t) preserve the same property of convexity. Proof. Let us again suppose the contrary. In other words, there exists 0 = ei0 that satises the equality arg f (0 , 0) = 2 (or ) 2 (4.51)
at the initial instant t = 0 and for any > 0 there is such t > 0 and (0 , 0 + ) that arg f (ei , t) (or ). For deniteness we put the 2 2 sign (+) in (4.51). This means that the mapping f (, 0) is critical for the property of convexity in the negative direction. Since the free boundary (t) tends to a vertical line as , we can consider nite critical points 0 = i0 and the set of such critical points lies in the compact subset on the imaginary axis. (t) is analytic, the function f is analytically extendable onto the imaginary axis, and the derivative f (0 , t) = 0. Suppose that Im f (0 , 0) > 0 (for Im f (0 , 0) < 0 the proof is similar). Now we show that f (0 , 0) = 0. If not, then the point 0 is a branch point of the function f (, 0) and in a neighbourhood of this point in H + the quantity arg f (, 0) /2 admits both positive or negative values. This contradicts the assumption that the function f (, 0) is convex in the negative direction. The image of the right half-plane H + under the map f (, 0) touches the ray arg w = at the point f (0 , 0). Thus, the following statements are true: 2 arg f (i, 0) = 0,
=0
arg f ( + i0 , 0)
=0
0.
Calculation of the left-hand sides of these formulae leads to the following: Re f (i0 , 0) f (i0 , 0) = 0, Im < 0. f (i0 , 0) f (i0 , 0) (4.52)
99
We dierentiate (4.49) with respect to and at the point 0 using (4.52) we obtain |f (i, 0)|2 Im f (i, 0) f (i, 0) = Im f(i, 0)f (i, 0) f (i, 0) . = Im f(i, 0)f (i, 0) f (i, 0) From (4.52) and (4.53) it follows that arg f (i0 , t) t
t=0
(4.53)
= QIm
f (i0 , 0) . f (i0 , 0)
(4.54)
The inequality in (4.52) with Q < 0 implies that the right-hand side of (4.54) is strictly negative for t > 0 close to 0 and the inequality arg f (i0 , t) < holds 2 in some neighbourhood of i0 . This contradicts the assumption that (t) fails to be convex in the negative direction for t 0 and ends the proof. Finally, in this section we should say that there are several other processes that involve planar dynamics in Hele-Shaw cells. Let us refer the reader, for example, to the papers [39], [110], [176], [181], [257], [273] and the references therein. We mention here a 600-paper bibliography of free and moving boundary problems for Hele-Shaw and Stokes ow since 1898 up to 1998 collected by Gillow and Howison [125]. Let us mention also a recent work [164] where the authors study the Hele-Shaw ow on hyperbolic surfaces. Most of the results presented in this section are found in [151], [173], [275], [337], [334], [336].
100
Theorem 4.4.1. Let the strong solution to (1.16) with Q = 1 exist during the time interval [0, t0 ), 0 < t0 < , (t) = f (U, t), and let the initial function f (, 0) be analytic and univalent in a neighbourhood of the closure of the unit disk U . Then the function f (, t) is analytic in UR(t) , where the radius of analyticity R(t) > 1 is a nondecreasing function in t [0, t0 ]. The function f (, t) is univalent in U , and possibly f (, t0 ) has a vanishing derivative at some points of the unit circle U or is not univalent on U . It follows that (t0 ) f (U, t0 ) is a simply connected domain with an analytic boundary (t0 ) with possible analytic singularities in the form of nitely many cusps and double points. In the case there are no singularities the strong solution can be extended to some time interval [0, t0 + ). Proof. By the Carathodory Kernel Theorem the domain e (t0 ) =
t[0,t0 )
(t)
is just the same as in the formulation of the theorem and (t0 ) is a simply connected domain. It follows from Proposition 3.3.1 that (t0 ) is also the same as the domain at time t0 for the weak solution. We note also that since the normal velocity on the boundary never vanishes, we have the strict monotonicity of the subordination chain of domains: (s) (t) for s < t and s, t (0, t0 ). (4.55)
Letting t t0 we see that (4.55) holds for t = t0 , i.e., for (t0 ) as well. The strong solution exists in the time interval t [0, t0 ) and coincides with the weak one. Therefore, the statements about f (, t0 ) and (t0 ) follows directly from Theorem 3.4.1. Let us prove the existence of the extension of the solution to the time interval [0, t0 + ) when there are no singularities on (t0 ). Construct the subordination chain of mappings f2 (, t) satisfying the Polubarinova-Galin equation (1.16) with the initial data f2 (, 0) f (, t0 ). The strong solution exists and is unique locally in time, say t [0, ). Moreover, we have lim f (, t) = lim f2 (, t) = f (, t0 )
tt0 0 t0+0
and lim f (, t) = lim f2 (, t) = f (, t0 ) locally uniformly in U1+ . We recall equation (1.17) (with Q = 1):
2 tt0 0 t0+0
1 f(, t) = f (, t) 2 4
ei + d, |f (ei , t)|2 ei 1
t [0, t0 ), || < 1.
A similar equation is valid for the chain f2 (, t) in the time interval [0, ). Taking the limit in the above equation as t t0 0 we observe that there exists the one sided limit f(, t0 0). Similarly, there exists the one-sided limit f2 (, 0+ ) and they are equal. Let us dene f (, t) f2 (, t t0 ) in the interval t [t0 , t0 + ). Above observations yield that the so-extended function is continuous in the interval t
101
[0, t0 + ), analytic, univalent and starlike in some neighbourhood of U . Moreover, it is dierentiable at the point t = t0 , and being extended onto the unit circle, satises the equation (1.16). Thus, it is a unique strong solution in the interval t [0, t0 + ). This nishes the proof of the theorem. Lemma 4.4.2. Under the assumptions of the previous theorem, if 0 is starlike (f0 S ) then the limiting domain (t0 ) has no singularities on the boundary. Proof. The function f (, t) belongs to the class S ((t)) with (t) < 1 for any t (0, t0 ) due to Theorem 4.3.2, where (t) strictly decreases with respect to t. Dene the limiting function f (, t0 ) = lim f (, t), where the limit is taken
tt0 0
locally uniformly in U . The function f (, t0 ) is univalent, strongly starlike of order (t0 ) = lim (t) < 1. According to the geometric characterization of the class S ((t0 )), the boundary of the domain (t0 ) = f (U, t0 ) is reachable by the radial external angles (1 (t0 )), which implies that there is no cusp or a double point on the boundary of (t0 ). This completes the proof. Theorem 4.4.3. Starting with a starlike phase domain 0 with an analytic boundary, the life-span of the strong Hele-Shaw starlike dynamics (t) is innite. Proof. Indeed, if the strong solution exists during the nite interval t [0, t0 ) and does not in t [t0 , t0 + ) for any > 0, then this contradicts Theorem 4.4.1 and Lemma 4.4.2. Problem 4.4.4. Of course, any result on life-span of the strong solution would be welcomed. Are there other geometric conditions that guarantee innite life-span? Given some geometric characteristics of the initial domain can we estimate life-span of the strong solution? Can we estimate life-span of polynomial solutions of degree n depending on n?
tt0 0
102
problem that appears for the complex potential of the unfrozen ow is governed by Darcys law which takes into account additional equations for the temperature eld. One of the typical features of this problem is that there is not, in general, uniqueness of the solution. At the same time the existence can be proved in a usual way. Let us formulate the governing equations. In the exterior part (t) of the crystal cross-section we introduce the complex coordinate z = x + iy and the complex ow potential W = + i, where is the velocity potential and is the stream function. We consider a dimensionless model such that = p refers to the pressure and gravity is neglected. Let us denote the temperature eld by (z) (x, y) in (t) and suppose that the phase transition is taken under the temperature = 0 on (t) = (t). A suitable scaling leads to the condition that (x, y) = 1 if x , where we suppose that the uid moves to the right and (+) corresponds to melting and (-) to crystallization with supercooling. Within strong Hele-Shaw assumptions, the mathematical model is described by the following equations [14], [75], [70], [133], [204]: V = 0, V = , P e V = , = 0, vn = n , n = 0, z (t); z (t). (4.56)
In this system the Pclet number P e is a measure of the intensity of heat transe fer by convection compared with conduction. We note that this model is timereversible [70]. In fact, reversing the sign of the temperature changes only the sign condition for limx and for the kinematic boundary condition which are both reversed.
The Boussinesq transformation [42] applied to the convective heat transfer equation (4.56) leads to uncoupling of the problem and permits us to apply analytic univalent functions. Uncoupling means that the initial problem (4.56) may be split into two independent tasks ([204], [227], [228]), the rst of which is the problem of heat exchange, the second refers to the free boundary nature of the problem. In fact, the Boussinesq transformation is equivalent to the existence of a conformal univalent map from the phase domain (t) onto the plane of the complex potential W = + i. Under this transformation the boundary of the crystal cross-section
103
is mapped onto the slit directed along the positive real axis = 0, > 0 in the W -plane. Thus, the problem admits the form: Pe = , W D, (4.57)
lim = 1,
= 0, lim
= 0, W D.
(4.58)
Now let us introduce the auxiliary parametric complex -plane, = + i. The Riemann Mapping Theorem yields that there exists a conformal univalent map f (, t) of the left half-plane H = { : Re < 0} onto the phase domain f : H (t). The parabolic shape of (t) implies the normalization f (, ) = 2 + a1 + a0 + a1 / + . Fortunately, for the problem (4.57)4.58 the method of separating variables is applicable. First we introduce the map H D given by W = 2 . Then we are looking for a similarity solution = g(). Elementary calculation leads to the relations = on the slit D, where = P e/, and = |W (z)|. n The unit normal outer vector to the moving interface is n = f (i, t)/|f (i, t)|, and the normal velocity, hence, is of the form f (i, t) , vn = Re f(i, t) |f (i, t)| Besides, we have i W (z) = vn n, = 0, > 0. (, ).
Changing variables, W = 2 we come to the Polubarinova-Galin type equation for the free boundary: Re f(i, t)f (i, t) = 2, = 0, (, ). (4.59)
Using the same argumentation as in the above subsections we prove the following statement. Theorem 4.5.1. If the initial crystal interface possesses the property to be convex of order in the negative direction (HR ()), then the free boundary remains convex in the negative direction of the same order as long as the solution to the problem (4.57)(4.58) (or (4.59)) exists ( > 0 and the liquid moves to the right). The result for = 1 is obtained in [205].
104
1 r2 (1 (0) ) . r r
Here the left member is (r u ) and the right member is non-negative due to the r starlikeness of (0). Thus r u is subharmonic in (t) \ {0}, and r u = 0 on (t). r r t At the origin u has the positive singularity u 2 log |z|, hence r u < 0 r near the origin, and therefore, in all (t) \ {0}, from which the starlikeness follows.
105
Ur(a) (a)
for suitable r(a) > 0. Proof. Set = (0) + (t) and write = Bal (, 1) = + u, where u is the smallest function satisfying u 0, u 1 . We rst assume that K lies in the lower half-plane K {y 0}, i.e., that (0) {y < 0}, and we shall study the geometry of + = {y > 0}. Let u be the reection of u with respect to the real axis, i.e., u (x+iy) = u(xiy), and set v = u inf(u, u ) = (u u )+ . Since u 1 1 everywhere, we have u 1. Therefore, inf(u, u ) 1 everywhere. Indeed, the inmum of two superharmonic functions is again superharmonic, so 1 1 1 inf(u, u ) 1 = (inf(u, u ) |z|2 ) = inf(u |z|2 , u |z|2 ) 0. 4 4 4
106
Since u = 1 in + it follows that v 0 in + . Moreover, v = 0 on (+ ). The maximum principle now shows that v 0 in + . This means that u u + in , i.e., that u is smaller (or at least not larger) at any point in the upper halfplane, than in the reected point in the lower half-plane. On the real axis this gives u (x, 0) 0, (4.60) y and in general it shows that the reection of + in the real axis is contained in : (+ ) . (4.61)
Now since u = 0 on ()+ and ( u ) = y u = 0 in + we can apply y u the maximum principle again, now to y , to obtain that u 0 in + . This y inequality is everywhere strict because if we had equality at some point, then it would follow that u = 0 in a whole component of + ; and this is impossible because 1 = = + u = u in + . The conclusion now is that u(x + iy) is a strictly decreasing function of y > 0 in + . Therefore, since u = 0 outside , every vertical line L in the upper halfplane intersects + in at most one segment (L \ + is connected). It follows that ()+ is a graph of a function, say
()+ = {z = x + iy, y = g(x)}. The domain of denition of the function g may consist of more than one interval. It follows from the general regularity theory (e.g., [46], [297]) that g is real analytic. Next, with K still in the lower half-plane we shall obtain a similar convexity statement, but for semicircles instead of vertical lines. Let (r, ) be the polar coordinates. In the proof of Theorem 4.6.1 we studied r here we shall study u u u = y +x , x y in + . Since u = 1 in + and the coecients in u = do not depend on ,
u
u u u =x +y , r x y
(4.62)
1 r r
u r
1 2u r2 2
is harmonic, u = u = 0 in + .
107 on (+ ), we have
u = 0 on ()+ . By (4.60), (4.62) we have that u 0 u 0 on the real axis. Now consider a circular arc CR = {z = rei , r = R, 0 < < } in the upper half-plane. We shall prove that CR \+ consists of at most one segment (more precisely, that it is connected), and we shall argue by contradiction. So suppose CR \ + has at least two components. Then there are points z1 = Rei1 and z2 = Rei2 with 0 < 1 < 2 < , such that z1 , z2 ()+ , and z = Rei + for all 1 < < 2 . Since u(z1 ) = u(z2 ) = 0 and u(z) > 0, we have, integrating along CR ,
z
Therefore, there must be points z = Rei with 1 < < 2 arbitrarily close to 1 for which u > 0. Similarly, there must be points z = Rei with 1 < < 2 arbitrarily close to 2 for which u < 0. Now we apply the maximum principle. Every component of { u > 0} must reach some part of the negative real axis, because we know that u 0 on all other possible parts of the boundary of that component. Similarly, every component of { u < 0} must reach some part of the positive real axis. But it is obviously topologically impossible to have components of { u > 0} stretching from points arbitrarily close to z1 (the rightmost end point of the described component of CR ) to the negative real axis, and simultaneously, components of { u > 0} stretching from points arbitrarily close to z2 on the positive real axis. This contradiction shows that CR \ + actually is connected. The same reasoning applies with the center of the polar coordinates at any point of the real axis. Thus for any semicircle C in the upper half-plane with its center on the real axis, C \ + is connected. Together with the rst part of the proof, saying that L \ +
108
is connected for any vertical semiline L, what we have proved can be expressed by saying that the complement of + in the upper half-plane is convex with respect to the Poincar metric in the upper half-plane. e Next, as for Euclidean convexity, C + \ + being convex (in the Poincar e metric) implies that it is the intersection of Poincar half-planes, or that, turning e to the complements, + is the union of such. This means that + =
aR
Ur(a) (a)
(4.63)
for suitable radii r(a) 0. Let z ()+ and let Nz be the inward normal ray at z. Since ()+ is a graph of a function, Nz intersects the real axis at a point p(z), which we call the foot point of the normal. In terms of the equation y = g(x) for ()+ we have z = x + ig(x) and one easily computes p(z) = x + g(x)g (x). The fact (4.63) that + is a union of semidisks (Ur(a) (a))+ implies that actually (U|zp(z)| (p(z)))+ + . (4.64)
Let z1 = z2 . Assume that two normals Nz1 and Nz2 intersect in C+ , say w Nz1 Nz2 , where w C+ . We may assume that |z1 w| |z2 w|, for example. Then z1 is contained in the closure of the disk with the center at w and the radius |z2 w|, and therefore, in the interior of the larger disk U|z2 p(z2 )| (p(z2 )). But z1 ()+ , so this disk must contain points outside + as well. This contradicts (4.64), and we conclude that Nz1 and Nz2 actually cannot intersect in C+ . One easily sees that the inner ball property (4.63), or the fact that the inner normals do not intersect in C+ , is equivalent to the foot point p(z) = x + g(x)g (x) being a monotone increasing function of x. So far we have assumed that K {y < 0}. Adapting the results we have obtained to all half-planes containing K easily gives the statements of the theorem. We give some details below. (i) We already know that \ K is analytic but with possible singularities. But since near any point of \ K, will be a graph seen from several dierent angles (choosing dierent half-planes containing K) there can be no singular points. (ii) If the inward normal Nz at a point z \ K did not intersect K we could nd a half-plane H K which does not meet Nz , contradicting that \ H is a graph when seen from H. (iii) Similarly, if w Nz1 Nz2 , w \ K, z1 , z2 \ K, z1 = z2 , we choose a half-plane H K, such that w H. By (ii), both Nz1 and Nz2 intersect
109
K. This shows that z1 H, z2 H, and we are back in the situation with H = {y < 0}, i.e., we have a contradiction. Thus Nz1 and Nz2 do not meet before they reach K. (iv) From what we already did it follows that \K =
aK
Ur(a) (a)
\K
Obviously, r(a) = 0 for a K \ . The points from K can be trivially covered by the disks Ur(a) (a) with a K. Now (iv) follows, and the proof of the theorem is complete.
and let be the saturated set (3.15) for Bal (, 1). Then Ur()+R , and if r() 2R we moreover have Ur()R . Proof. First we recall from (3.20) that Bal (, 1) = + C\ . From this it follows that d
UR
dz +
C\
for all functions in C which are integrable and subharmonic in . In particular, taking = 1, || d.
110
The upper bound UR+r() is actually a direct consequence of the Inner Normal Theorem. By that theorem is a union of disks with centers in UR , so if contained points outside UR+r() , then it would contain a disk of radius greater than r(), which is impossible because || d = |Ur() |. Next assume r() 2R and let z . We shall show that |z| r() R, / hence that Ur()R . By a rotation we may assume that z = 0. If 0 < < R we choose () = G(, ) + c,
R 1 where G(, ) = 2 log | R() | is Greens function of UR and c > 0. Since , / is subharmonic in . The level lines of G(, ) and are circles, so
2
{ : () > 0} = { : G(, ) > c} = Ut (a) for some disk Ut (a), which contains UR because c > 0. Now choose c so that G(, ) d = 0.
Ut (a)
By a straightforward computation (see section 6 in [300]), this gives R < t < 2R. Thus, using that Ut (a) is exactly the set where is positive we get cr()2 = c
UR
d
UR
(G(, ) + c) d() =
UR
d c d = ct2 .
d +
C\
d =
Ut (a) Ut (a)
Hence r() t < 2R, contrary to our assumption in the beginning. Thus there is no 0 < < R with . So UR and Bal (, 1) = . / If R then we take instead to be the Poisson kernel for the disk UR+ (R) and with the pole at : () = (R + )2 | + R|2 (R + )2 | + R|2 = . |(R + ) ( + R)|2 | |2
Since , is subharmonic in . Also in this case the level lines of are / circles. The circle where = 1 passes through the center R of UR+ (R) and through the pole = , hence 1 inside that circle, in particular 1 in UR (0). Moreover, UR+ (R) is exactly the set where 0. All this, combined with the mean value property of in UR+ (R), gives that r()2
UR
d
UR
111
UR+ (R)
Thus R r() as required. Corollary 4.6.4. In the one point injection Hele-Shaw case = (0) + ta . This gives, if a UR , (0) UR , that (t) U(|(0)|+t)/+R . If, in addition, t 4R2 |(0)|, then U(|(0)|+t)/R (t).
the story that Queen Dido of Ty by the (straight) Mediterranean land as could be enclosed by a b a more algebraic form which ind Denote the perimeter and area o 4A L2 . The equality only ho example, if S is a surface area wh 36V 2 S 3 (see, e.g., [59], [90]). Pappus of Alexandria (ca 3 which is useful to them, that the h and will hold more honey for th each. But we, claiming a greater a somewhat wider problem, nam gures having the same perimete always greater, and the greatest o to them. Probably, the most represen ious aspects of mathematical ph
m(
Then, m(, ) =
1 2
log R.
Example 5.1.3. Let be an annu the family of curves in that co 2 m(, ) = log R .
M (, a) :=
is said to be the reduced modulu point a, where M ( ) is the mo respect to the family of curves th
as u ranges over the class Lip (C K1 , u(z) 1 in K2 . A condenser C is said to be a in C function (z), 0 (z) 1 for z K1 , (z) = 1 for z K Dirichlet principle yields that in in the case of an admissible cond potential function . Then by G cap C =
C\{K1
1 cap CR
cap K =
exists and is said to be the logarit (5.4) is also known as Pugers t Next we briey summarize rithmic capacity of a compact se consider n (K) =
We consider the zero surface ten source at the origin and with a b Let f be a univalent funct the unit disk U and let S be the call the radius-area estimate, is S S = (a2 + k=2 k|ak |2 ). Equali is no upper estimate of S, namel the classical result by Plya and o estimate is in general and 8R Let S(t) be the area of the the conditions of Section 1.4.2. A that the rate of the area change From (1.17) we deduce that
2
1 a = a 2 4
Q d |f (ei , t)|2
|f (ei )|2
|h
0
We have |c0 | 2, and by the Are hand side of (5.6) is 6/a2 . T which are convex at an instant t
Problem 5.1.4. If the domain (t of the solution of (1.16) convex to nd a geometric condition tha existence of the solution of (1.1 mean (5.5) from above.
g(z) S g(z) = z
where () is a non-decreasing fu
(1 eik z)2(1)k
k=1
Using known properties of Stieltj that the set of functions (5.7) is there exists a sequence {gn (z)} to g(z) in U . Therefore, we need
where S(t) means the area of th that the outer Hele-Shaw proble estimates for integral means can
Recent studies of Robins funct connections with several problem length and minimal energy cons interpretation that contains the shall connect the rate of area cha of Robins reduced modulus of th
provided by means of the modul Let Cr (z0 ) = {z : |z z0 | = a nite z0 let us consider th Cr (z0 ). Then the limit lim
r0
exists and is exactly Robins red and r suciently large we connect A with Cr . Then the lim lim
exists and is Robins reduced mo that Robins modulus is change following rule: for nite points w w = f (z) = w0 + a(z z0 ) +
(w)
= = =
The arc = {ei , [ the slit [1, r]. Robins fore, making use of the fo 1 MU (A, 0) = 2 log 1/ 2 +
vn
= 0,
The complex potential is exactly course R depends on t). Richardson [286], [286], con circular initial evolution. We consider the case when of consideration. Observation of and (t) suggests the contact an boundary (t) we involve an au The Riemann Mapping Theorem valent map f (, t) from the uni r(t) (0, 1], U = { : || < f (0, t) = 0, a(t) = f (0, t) > (t) = {f (), (1 0i, r(t) corresponds to a stagnation point parametrization of (0) = (0)
The length of the radial slit 1r( (t) with respect to the origin is If (t) is the union of interv minus several slits onto the phas component of (t). Now we apply Robins redu phase domain of the injecting u recall that the boundary (t) o a free part (t) and the solid pa 0 (t). The parametric functio satises the equations (5.11)(5.1 A simple application of Gr change is expressed as S = Q injection we have Q < 0. Let the Pick function = (w, t) =
a 1 = a 2 Obviously, 2 cos 2
radius. Finally, let us remark that ponent (t) can be treated in th Riemann-Hilbert problem yields sidered only the simplest case.
In this section we shall obtain an Theorem 5.1.2 for the corner ow and derived the governing equatio phase domain. Here we use a slig our concrete purpose. Similarly to the model anal where the viscous uid occupies plane whose boundary (t) at an the corner and a free interface 3
Im (fei ) = 0 for
Hence, we can rewrite the cond problem for the analytic function
which is known for = 2 (see T To make inequality (5.17) related to the free boundary 3 . in C with three nite xed boun boundary. Denote by D the dom U (a, ) = {z : |z a| < }. Deno D joining the boundary arc of U with the leg of the triangle D w the circle so that it can be conne the limit
M (D, a) =
exists, where a = sup a is the in D at a, then it is called the re for the reduced modulus to exist
SR
In Section 3.4 have already discu In this section we consider a bou The uid moves to the right and t the crystal. The governing equati
lim
The Boussinesq transformation a (4.56) leads to uncoupling of th from the phase domain (t) onto Under this transformation the b
Q=2
1
is the total heat ux. Obviously, sign of the normal velocity. Ther for all [0, 2). A simple appl of the area change of the nucleu with (-): S = Q. In fact, we ha 2S(t) =
(t)
Im (f df )
Then, dS = 2 dt
(c
0
double point (Theorem 4.4.1) oc This chapter is devoted to g to the Lwner-Kufarev equation o mal maps that admit quasiconfo associated quasidisks are bounde Hele-Shaw ows of viscous uids As usual, U denotes the un of all holomorphic univalent func U , and by , the class of c1 + malized by f () = + c0 + with c0 = 0. These classes have complex analysis for a long time posed by Bieberbach1 in 1916 [38
1 Ludwig Georg Elias Moses Bieberb He received his doctorate in 1911. Bieb in Switzerland, Frankfurt in Germany, geometry and where he developed a re
In 1959 Shah Dao-Shing [30 formal automorphisms of U . In a by Gehring and Reich [122], and an analogous equation for classe tions (to be more precise, in ter method for conformal maps, its a so much attention. Several attempts were laun to obtain conformal maps that [26], [156]).
The principal goal of this ch mal maps with quasiconformal ex smoothly extendable onto the un
He was converted to the views of the persecuted his Jewish colleagues (e.g., II in 1945 Ostrowski invited him to lec contribution as a mathematician. It is Ostrowski Prize for solving the Bieber
conformal radius of (t). A rec known nowadays as the Lwner o Lwner [224] and Kufarev [212]. o Geometrically, it readily co (t). Indeed, supposing an ana outward direction is n = f (f(, t))/(f (, t)) the normal vn = Re
f(, t)f ( |f (, t
and is positive. Therefore, Re p( case of a nonanalytic boundary r What does p(, t) look like
or when the domain (t) is no lon fail to be a quasidisk as t t . 0 universal Teichmller space, whi u
We ask the following question: g Hele-Shaw evolution (t), what k Teichm ller space? u A general scheme of the pro
6.2 Quasiconformal m
The theory of quasiconformal m eth century. At that time, quasi their denition was based on the so-called regular quasiconformal gested by Ahlfors and Beurling, systems of dierential equations i tant applications to various eld
surfaces of the same genus modu Teichm ller brought togeth u maps, and relevant quadratic di the well-known theory of Teichm thoroughly substantiated by Ahl
f + x
2 Paul Julius Oswald Teichm ller, 1 u active member of the Nazi party. He d
complex parameter t; f (z, ) is a to z, f < 1. If f is n-dier is Hlder continuous of order o to the equation (6.4) is (n + 1)-di the same Hlder condition [31], o a continuous derivative wheneve shown by the example f (z) = z( Lipschitz continuous at z = 0. B a Hlder continuous f for any 0 o The dependence on the para or continuous function with respe then the same is true for f .
Let us consider the family F of such map f satises the Beltram tional sense, where f is a meas
can be thought of as the family o quasiconformal extension. Moreov maps f of U that admit quasico some k. As we mentioned above, a n U can have a quasiconformal ext unit circle f may behave quite i f (S 1 ) can have a Hausdor dime
Remark 6.2.1. Given a bounded K let N (, ) denote the minimal n cover . Let (K) = sup l
denote the supremum of the Mink all bounded K-quasicircles. The by (K) (see [27]). In [27] several e.g., (K) 2 cK 3.41 .
manifold modelled on B(U ). Th vector space that can be though point (represented by 0). Mo homeomorphism of the unit disk in a direction . Let us construct to a small parameter :
f () =
Sf =
(w w
of L (U ) that annihilates the o identify the tangent space to T at It is natural to relate it to a sub identically on A(U ) due to (6.6) . Thus, the operator splits
0 N
6.3 Di S 1 /Rot S 1 em
We denote the Lie group of C s S 1 by Di S 1 . Each element of Di increasing, C real-valued funct Lie algebra for Di S 1 is identi tangent vector elds to S 1 with
[1 ,
unitary representations of Di S Khlerian manifold M = Di S 1 a tations of S 1 . The group Di S 1 M with Rot S 1 as a stabilizer. T Kirillov and Yuriev in [199]. Th particular, in the space of smoot the space of quadratic dierentia based on the class 0 of function U of U , are supposed to be sm local uniform topology of U . Let S denote the class of a g() = c0 + c1 + c2 2 + whi have f (U ) and there is an The superposition g 1 f restric for each element of M there exist curve is a quasicircle if and only i quasiconformal extension to U . B bedded into the universal Teichm
146
on M corresponds to the Khlerian structure on T given by the Weil-Petersson a metric. The Goluzin-Schier variational formulae lift the actions from the Lie algebra Vect S 1 onto 0 . Let f 0 and let d(ei ) be a C real-valued function in 1 (0, 2] from Vect S making an innitesimal action as + d(ei ). Let us consider a variation of f given by d f () = 1 2i
S1
wf (w) f (w)
wd(w)dw . f (w) f ()
(6.7)
Kirillov and Yuriev [199], [200] have established that the variations d f () are closed with respect to the commutator and the induced Lie algebra is the same as Vect S 1 . Moreover, Kirillovs result [201] states that there is the exponential map Vect S 1 Di S 1 such that the subgroup Rot S 1 coincides with the stabilizer of the map f () from 0 .
(z) w 1 w(z)
1 ||
2 2 |dz|
| z|
= 0.
(6.8)
An advantage of this extension is that if , Mb(U ), then the extension of o is given by h , which is not true for the Beurling-Ahlfors extension. The three-point boundary normalization of F0 can be always attained, and thus, the Douady-Earle extension is compatible with the denition of the universal Teichm ller space. Later, in 1988, another proof of Douady-Earles result appeared u in [220] where the authors worked with the inverse function. The functional equation (6.8), in particular, implies that a C mapping representing an element from the manifold M has a real analytic extension h F which is C on S 1 .
3 Arne Karl-August Beurling, 19051986, Lars Valerian Ahlfors, 19071996; this extension appeared in a 1956 paper [37] with the nonalphabetic listing of the authors as Ahlfors had insisted because of the contribution made by Beurling. Ahlfors ... felt mostly like a secretary; the main ideas of the paper were due to Beurling (see [28]).
147
Let f 0 represent an element from M . Let g S be the adjoint map, g f 1 = . If h is the Douady-Earle extension of , then g h S 1 f S 1 S and g h is a quasiconformal extension of f 0 . Given M we construct the mapping f that satises the normalization of the class 0 and whose Beltrami coecient is F Fw F Fw f () = , w = h(), U, (6.9) F Fw F Fw
1
with f () = 0 for U . The equivalence class [f ] is a point of the universal Teichm ller space T . So the Douady-Earle extension denes an explicit embedding u of M into T .
w (wd(w))dw , w
(6.10)
where the distributional derivative w is given in the unit disk U , d(w) is a con tinuous extension of the C function d(ei ) Vect S 1 into U that has Ls (U ) distributional derivatives in U , s > 2, and dw is the area element in U . Thus, one can extract the elements from H that are of the form () = d(), where means / . We are going to deduce an exact form of using the Douady-Earle extension. For this we start with the variation of the element (ei , ) = ei (1 + id(ei )) + o( ), M, d Vect S 1 ,
and is small. The Beltrami coecient of the extended quasiconformal map h has its variation as h () = () + o( ), where
F Fw F Fw F Fw F Fw =0, w= 2 2 |dz|
() = where F (, w) =
, U,
(6.11)
1 2
S1
(z, ) w 1 w(z, )
1 ||
| z|
= 0.
(6.12)
148
Thus, () depends only on d(ei ). We will give explicit formulae in the next section. They can be obtained substituting (ei , 0) = ei , and taking into account that
F Fw F Fw =0, w=
= 0.
The Lie algebra Vect S 1 is embedded into the Lie algebra of H by (6.11), (6.12). Hence, a ow given on M corresponding to a vector d Vect S 1 is represented as a ow on the universal Teichm ller space T corresponding to the vector H u given by (6.11).
3 () = 2
1 ||2 (1 ei )2
(6.13)
Proof. Let (, ) = ei(+ d(e )) and h(, ) be the Douady-Earle extension of into the unit disk U , U by means of (6.12). If = 0, then h(, 0) . We calculate
2
F (, w)
1 2 1 2 1 2 1 2
0 2
ei ( ei )2 d, | ei |4 ei ( ei )2 d, | ei |4 1 ||2 d, | ei |2 1 ||2 d. | ei |2
F (, w)
0 2
w F (, w)
0 2
w F (, w)
149
1 , 1 ||2
F (, w) w F (, w)
= 0,
=0,w=
=
=0,w=
1 , 1 ||2
w F (, w)
=0,w=
= 0.
We will use the properties of the Douady-Earle extension. Let us x a point 0 U and choose two Mbius transformations , of U such that (0) = 0 and o (0) = h(0 , ). We set g = 1 h . Then, g(0, ) = 0, g(0, ) = 0 and g(0, ) = h(0 , ) g(0, ) = h(0 , ) So we see that (0) , (0) (0) . (0)
By the property of the Douady-Earle extension we have that the function g(, ), U is the extension of g(ei , ) by means of (6.12). If = 0, then g(, 0) . Now we put (ei , ) = g(ei , ) in (6.12) and calculate variations in ,
2
F (, w)
1 = =0,w==0 2 = 1 2 1 2 1 2
0 2
ei (1 ei )3 ei (ei , 0)d,
0 2
w F (, w)
=
=0,w==0
0 2
150
Then, we can obtain the explicit form of the variation of the Beltrami coecient by (6.11) as
g(0, ) g(0, ) 2
=0
3 = 2
ei (ei , 0)d.
0
(6.14)
The Mbius transformation does not depend on whereas does. Explico itly, we put h((), ) h(0 , ) 1 h () = , 1 h((), )h(0 , ) where () = ( + 0 )(1 + 0 )1 . We denote ei = (ei ). Therefore, denoting ei = (ei ) = we have g(ei , 0) = Then, ei d = h(ei , 0)(1 |0 |2 ) h(0 , 0)(1 0 ei ) + h(0 , 0)ei (ei 0 ) . (1 0 ei )2 1 |0 |2 i e d, (1 ei 0 )2
2 2
ei + 0 , 1 + 0 ei
=0
3 = 2
1 |0 |2 (1 ei 0 )2
Taking into account that (0) = 1 we come to the statement of the theorem. Corollary 6.4.2. If q = max |d(ei )|, then
[0,2]
|()| 3
1 + ||2 q. 1 ||2
3 () = 2
3 () = 2
(6.15)
151
As we see, the given estimate is good enough when || is not close to 1. Let us now give an asymptotic estimate for |()| in the case || 1. Corollary 6.4.3. There exists a constant M independent of such that |()| M 1 ||2 . ||2
In particular, |()| = O(1 ||2 ) as || 1. Proof. We integrate by parts the right-hand side in the formula (6.13) twice and come to the expression (1 ||2 ) () = 4 2
2
1 ||2 (1 ei )2
d.
(6.16)
The absolute value of the above integral is bounded because of the Poisson kernel in it and due to the smoothness of the function d.
152
In particular, (0) = 1. This semi-ow is generated by a vector eld v() if for each U the function w = (, ), 0 is a solution of an autonomous dierential equation dw/d = v(w) with the initial condition w| =0 = . The semi-ow can be extended to a symmetric interval (t, t) by putting = 1 (, ). Certainly, the latter function is dened on the set (U , ). Admitting this restriction for negative we have a one-parameter family for (t, t). For a semi-ow , 0 on G there is an innitesimal generator at = 0 constructed by the following procedure. Any element is represented by a conformal map (, ) that satises the Schwarz Lemma for the maps U U . Hence, Re 1, U , (, ) (, ) where the equality sign is attained only for 0 = id following limit exists (see, e.g., [134], [135], [309]) (, ) = Re lim Re 0 (, ) and the representation (, )
=0
(, 0) . Therefore, the
(, ) =0
0,
= p()
holds, where p() = p0 + p1 / + is an analytic function in U with positive real part, and ( ) = p0 . (6.17) =0 In [136] it was shown that is even C with respect to . The function p() is an innitesimal generator for at = 0, and the following variational formula holds: (, ) = + p() + o( ), ( ) = 1 + p0 + o( ). (6.18) The convergence is thought of as local uniform. We rewrite (6.18) as (, ) = (1 + p0 ) + (p() p0 ) + o( ) = ( ) + (p() p0 ) + o( ). (6.19) Now let us proceed with the semigroup G qc G of quasiconformal automorphisms of C. A quasiconformal map representing an element of G qc satises the Beltrami equation in C, = () , with the distributional derivatives and , where () is a measurable func tion vanishing in U and essentially bounded in U by = ess sup | ()| k < 1,
U
6.5. Parametric representation of univalent maps for some k. If k is suciently small, then the function tional formula (see, e.g., [221]) () b0 1 =
U
(w)dw + o(k), w
(6.20)
where dw stands for the area element in the w-plane. Now for each small and G qc the mapping h(, ) = (, )b0 ( ) is ( ) from qc and represents an equivalence class [h ] T . Consider the one-parameter 0 curve x T that corresponds to [h ] and a velocity vector () H (that is not trivial), such that h (, ) = (, ) = () + o( ). We take into account that (, 0) in U and is extended up to the identity map of C. The formula (6.20) can be rewritten for (, ) as (, ) b0 ( ) = ( )
U
(w)dw + o( ). w
(6.21)
(w)dw + o( ). w
(6.22)
(w)dw . (w )
(6.23)
The constants p0 , p1 and the function must be such that Re p(z) > 0 for all z U . We summarize these observations in the following theorem. Theorem 6.5.1. Let be a semi-ow in G qc . Then it is generated by the vector eld v() = p(), (w)dw 1 p1 , p(z) = p0 + (w )
U
where () H is a harmonic Beltrami dierential and the holomorphic function p() has positive real part in U . This theorem implies that at any point 0 we have (, ) = (, )p((, )).
154
155
The function (, t, ) is embedded into an evolution family in G. It is differentiable with regard to and t in [0, t0 ), and (, t, 0) = . Fix t and let D = 1 (U , t, ) \ U . Then, there exists H such that the Beltrami coecient is of the form (, t, ) = (, t)+o( ) in U \D , (, t, ) = f (, t ) in D , and vanishes in U . We make suciently small such that (, t, ) vanishes in D too. Therefore, = lim 0 (, t, ) locally uniformly in C and (, t, ) is embedded now into an evolution family in G qc . The identity map is embedded into a semi-ow G qc (which is smooth) as the initial point with the same velocity vector (, t, ) = p(, t), U , =0 that leads to equation (6.1) (the semi-ow is tangent to the evolution family at the origin). Actually, the dierentiable trajectory f (, t) generates a pencil of tangent smooth semi-ows with starting tangent vectors p(, t) (that may be only measurable with respect to t). The projection to the universal Teichm ller space u is shown in Figure 6.3.
Figure 6.3: The pencil of tangent smooth semi-ows The requirement of non-triviality makes it possible to use the variation (6.21). Therefore, the conclusion is that the function f (, t) satises the equation (6.1) where the function p(, t) is given by p(, t) = p0 (t) + 1 p1 (t)
U
(w, t)dw , (w )
and has positive real part. The existence of p0 (t), p1 (t) comes from the existence of the subordination chain. We can assign the normalization to f (, t) controlling
156
the change of the conformal radius of the subordination chain by et . In this case, changing variables we obtain p0 = 1, p1 = 0. Summarizing the conclusions about the function p(, t) we come to the following result. Theorem 6.5.2. Let f (, t) be a subordination chain of maps in U that exists for t [0, t0 ) and satises the conditions (iv). Then, there are a real-valued function p0 (t) > 0, a complex-valued function p1 (t), and a harmonic Beltrami dierential (, t), such that Re p(, t) > 0 for U , p(, t) = p0 (t) + 1 p1 (t)
U
(w, t)dw , (w )
U ,
and f (, t) satises the dierential equation f (, t) f (, t) = p(, t), t in t [0, t0 ). In the above theorem the function (, t) belongs to the space of harmonic dierentials. We ask now about another but equivalent form of as well as whether one can extend the equation (6.24) onto the whole complex plane. Writing w = f (, t ), (, t, ) = f 1 (w, t) we calculate the dilatation of the function (, t, ) in U . Note that is dierentiable by t, . f w w wz f 1 1 fw w + fw w w + f 1 w f wz = = 1 = = . 1 f w w + f 1 w w fw w + fw w 1 f w w f We use that f 1 f = f f /f . Finally, f , f , f are dierentiable by t almost everywhere in t [0, t0 ), locally uniformly in U , and f f f t f = , 0 (, t) = lim 0 f 1 |f |2 where the limit exists a.e. with respect to t [0, t0 ) locally uniformly in U , or in terms of the inverse function f 1 1 fw t f (, t). 0 (, t) = 1 fw 1 |f 1 |2 U , (6.24)
157
Sometimes, it is much better to operate just with dilatations, avoiding functions, so we can rewrite the last expression as log 1 f t 0 (z, t) = f (z, t) f (z, t) . 1 |f 1 |2
Remark 6.5.3. The function (, t) in Theorem 6.5.2 may be replaced by the function 0 (, t) that belongs to the same equivalence class in H.
Let us consider one-parameter families of maps in U normalized by f (, t) = a1 (t) et + + . The inverse result to the Lwner-Kufarev equation states that o given a holomorphic function p(, t) = 1 + p1 (t)/ + in U with positive real part the solution of the equation (6.24) presents a subordination chain (see, e.g., [270]). This enables us to give a condition for 0 that guarantees a normalized one-parameter non-trivial family of maps f (, t) to be a subordination chain.
Theorem 6.5.4. Let f (, t) be a normalized one-parameter non-trivial family of maps for U which satises the conditions (iiv) and is dened in an interval [0, t0 ). Let each f (, t) be a homeomorphism of C which is meromorphic in U , a1 (t) + , and satises (6.24). Let the quasiis normalized by f (, t) = et + conformal extension to U be given by a Beltrami coecient f = (, t) which is dierentiable with respect to t almost everywhere in t [0, t0 ). If
< 4
1 0
sK (s)ds
0.706859 . . . ,
where 0 (, t) is as above and K () is the complete elliptic integral, then f (, t) is a normalized subordination chain.
158
0 (w, t)dw (w )
0
U
dw |1 w/z|
1 2
0 0 0 0 0 4 0
0 1 2
0 1 2
1+
r2 2
r 2 cos
1/ 2
0 0 1 2
0 1
sK (s)ds < 1.
Then Re p(z, t) > 0 that implies the statement of the theorem. Remark 6.5.5. If 0 (, t)
q, then
1 + |(, 0)| 1 + |(, t)| e2tq . 1 |(, t)| 1 |(, 0)| This obviously follows from the inequality |f 1 | |f | = |f 1 |. t t Remark 6.5.6. Let us remark that the function 0 can be unilateraly discontinuous on S 1 in U , therefore, it is not possible, in general, to use the Borel-Pompeiu formula to reduce the integral in p to a contour integral. The equation (6.24) is just the Lwner-Kufarev equation in partial derivatives o with a special function p(z, t) given in the above theorems. Now we discuss the possibility of extending the equation (6.24) to all of C. We dierentiate the function (, t, ) with respect to when U U . It
159
f f f + f =: G(, t). 2 2 |f |2 |f |f | |f | |2
This formula can be rewritten in the form f(, t) = (f G(, t) + f G(, t)). Taking into account the equation (6.24) in U we have in the whole plane f(, t) = (f G(, t) + f G(, t)), f p(, t), for U, for U , (6.25)
where p(, t) is a holomorphic in U function with the positive real part by Theorem 6.5.2. The variational formula (6.22) and dierentiation of the singular integral imply that G (, t) = 0 (, t), U . Now let us clarify what is G. Let us consider U . The Pompeiu formula leads to G(, t) = h(, t) 1
U
0 (w, t)dw , w
where h(, t) is a holomorphic function with respect to . The function G is continuous in U and by the Cauchy theorem h(, t) = 1 2i
S1
G(w, t) dw. w
To obtain the boundary values of the function G(w, t), |w| = 1, we shall use the second line in (6.25). Unfortunately, in general, it is not possible to use the same function f in both lines of (6.25) to obtain boundary values of G. Indeed, the mapping f (, t) is dierentiable regarding t a.e. in t [0, t0 ) locally uniformly in C, and continuous in C for almost all t [0, t0 ). Therefore, the function (f G(, t)+f G(, t)), U is an extension of f (, t)p(, t), U , whereas f , U is not necessarily an extension of f , U . A simple example of this situation is as follows. Let us consider the function t e , for U , c + c f (, t) = 1 t e , for U , c + c where c > 1. This mapping forms a subordination chain with the dilatation () that vanishes in U and is the constant 1/c2 in U . This chain is trivial, but it is
160
not important for our particular goal here because we do not use at this stage the crucial variation. Then, for U , , c2 2 + 1 G(, t) = 2 2 , for U , c 1 and it splits into two parts that can not be glued on S 1 . The same is true for the derivatives f in U and f in U . If (, t) satises the condition (v) in a neighbourhood of S 1 in U, then the derivatives f , f , U have a continuation onto S 1 and F (, t) = f f p(, t) f f p(, t) , 2 |f |2 |f | S 1,
where f p(, t) is thought of as the angular limits that exist a.e. on S 1 . Moreover, in a neighbourhood of S 1 the derivative f vanishes and the function F (, t) can be written on S 1 as F (, t) = p(, t). In turn, h(, t) = 1 2i
S1
wp(w, t) dw. w
This information allows us to formulate the following theorem. Theorem 6.5.7. Let f (, t) be a subordination non-trivial chain of maps in U that exists for t [0, t0 ) and satises the conditions (iv). (i) For U there exists a holomorphic function p(, t) given by Theorem 6.5.2 such that f(, t) = f (, t)p(, t). (ii) For U there exists a continuous in function F (, t) given by F (, t) = 1 2i
S1
wp(w, t) 1 dw w
0 (w, t)dw , w
161
(6.26)
where the function w = g(, t) is a solution of the equation dw = wp(w, t), (6.27) dt almost everywhere in t [0, ), with the initial condition g(, 0) = . The function p(, t) = 1 + p1 (t)/ + is analytic in U , measurable with respect to t [0, ), and its real part Re p(, t) is positive for almost all t [0, ). The equation (6.27) is known as the Lwner-Kufarev ordinary dierential equation. The solutions to o (6.27) form a retracting subordination chain g(, t), i.e., it satises the condition g(U , t) U , g(U , t) g(U , s) for t > s, and g(, 0) . The connection between (6.24) and (6.27) can be thought of as follows. Solving (6.24) by the method of characteristics and assuming s as the parameter along the characteristics, we have dt d df = 1, = p(, t), = 0, ds ds ds with the initial conditions t(0) = 0, (0) = 0 , f (, 0) = f0 (), where 0 is in U . We see that the equation (6.27) is exactly the characteristic equation for (6.24). Unfortunately, this approach requires the extension of f0 (w1 (, t)) into U 1 (w means the inverse function) because the solution to (6.24) is the function f (, t) given as f0 (w1 (, t)), where = w(z, s) is a solution of the initial value problem for the characteristic equation (6.27) that maps U into U . Therefore, the solution of the initial value problem for the equation (6.24) may be non-univalent. Solutions to the equation (6.27) are univalent functions w(z, t) = et + a1 (t)/ + in the exterior of the unit disk that map U into itself. Conversely, every function from the class 0 can be represented by the limit (6.26) where there exists a function p(, t) with positive real part for almost all t 0, such that w(, t) is a solution to the equation (6.27) (see [270, pages 159163]). Each function p(, t) generates a unique function from the class 0 . The reciprocal statement is not true. In general, a function g 0 can be determined by dierent functions p. From [270, page 163] it follows that we can guarantee the univalence of the solutions to the Lwner-Kufarev equation in partial derivatives (6.24) assuming o the initial condition f0 () given by the limit (6.26) with the function p(, t) chosen to be the same in the equations (6.24) and (6.27), see [275]. The Lwner partial derivative equation (6.24) and the ordinary derivative o equation (6.27) admit the Hamiltonian interpretation. Let us rewrite equation (6.24) in the form f (, t) H(, f , t) = , (6.28) t
162
where H(, f , t) = f (, t)p(, t), and the derivative f is taken with respect to the complex variable . Interpreting the function H(, f , t) as a Hamiltonian we must write H(, f , t) = = p(, t). (6.29) t f Equations (6.28) and (6.29) constitute the conjugate pair of Hamiltons relations, assuming the initial conditions f (, 0) = f0 () and (, 0) = and rewriting = w(, t), U . Our goal is to deduce a form of the function p on the case of the subclass qc . Let a one-parameter family of maps w = g(, t), g qc , satisfy the following 0 0 conditions. (i) The maps g(, t) form a retracting subordination chain g(U , 0) U . (ii) The map g(, t) is meromorphic in U , g(, t) = (t) + a1 (t)/ + , where (t) > 0 and dierentiable with respect to t. (iii) The map g(, t) is a quasiconformal homeomorphism of C. (iv) The chain of maps g(, t) is not trivial. (v) The Beltrami coecient g (, t) of this map is dierentiable with respect to t locally uniformly in U . Note that in this case we do not need a strong assumption (v) in Section 6.5.2. Set H(, t, ) = g(g(, t), ) = ( )w + b0 ( ) + b1 ( ) + , w
where w = g(, t). For each xed t the mapping g(, t) generates a smooth semiow H in G qc which is tangent to the path g(, t + ) at = 0. Therefore, we use the velocity vector wp(w, t) (that may be only measurable with regard to t) with w = g(, t) and obtain H(, t, ) = g(, t)p(g(, t), t).
=0
As before, the trajectory g(, t) generates a pencil of tangent smooth semi-ows with the tangent vectors wp(w, t), w = g(z, t). Since g(U , t) U for any t > 0, we can consider the limit
0
lim
(6.30)
163
where p(, t) = p0 (t) + p1 (t)/ + is an analytic function in U that has positive real part for almost all xed t. The equation dened by (6.30) is an evolution equation for the path g(, t) and the initial condition is given by g(, 0) = . We suppose that all g(, t) admit real analytic quasiconformal extensions and the family is non-trivial in the above sense. The function g(w, ) = (H(, t, ) b0 ( ))/( ) can be extended to a function from qc and it represents an equiva0 lence class [g ] T . There is a one-parameter path y T that corresponds to a tangent velocity vector (w, t) such that g (w, ) = (w, t) + o( ), w = g(z, t).
We calculate explicitly the velocity vector making use of the Beltrami coecient for a superposition: (w, t) = lim or (w, t) =
0
g(w, ) g(, t) 1 H(,t, ) g(,t) g (, t) = lim , 0 1 g(,t) H(,t, ) g (, t) g g 1 (w, t), 1 |g(,t) |2 g
g(,t) t
U.
(6.31)
It is natural to implement an intrinsic parametrization using the Teichmller disu tance T (0, [g t ]) = t, and assume the conformal radius to be (t) = et that implies p0 = 1. The assumption of non-triviality allows us to use the variational formula (6.22) to state the following theorem. Theorem 6.5.8. Let g(, t) be a retracting subordination chain of maps dened in t [0, t0 ) and U . Each g(, t) is a homeomorphism of C which is meromorphic in U , g(, t) = et + b1 / + , with an e2t -quasiconformal extension to U given by a Beltrami coecient (, t) that is dierentiable with regard to t a.e. in [0, t0 ). The initial condition is g(, 0) . Then, there is a function p(, t) such that Re p(, t) > 0 for U , and p(w, t) = 1 1
g(U,t)
w g(U , t),
where (u, t) is given by the formula (6.31), solution to the dierential equation dw = wp(w, t), dt with the initial condition g(, 0) = . 1
U
w g(U , t),
(6.32)
Remark 6.5.9. Taking into account the superposition we have p(g(, t), t) = 1 where u U , U .
164
Remark 6.5.10. The function wp(w, t) has a continuation into g(U, t) given by dw = F (w, t), dt where the function F (w, t) is a solution to the equation
2 g g F g 1 (w, t). = w |g |2 |g |2
In contrary to the Lwner-Kufarev equation in partial derivatives, the function o F is the continuation of p in U through S 1 . The solution exists by the Pompeiu integral and can be written as F (w, t) = h(w, t) 1
g(U,t) 2 g g d u g 1 (u, t) 2 |g |2 |g | uw
1 = h(w, t)
g(U,t)
(u, t)d u , uw
where w g(U, t), h(w, t) is a holomorphic function with respect to w, that can be written as 1 up(u, t) h(w, t) = du. 2i uw
g(U,t)
Reciprocally, given a function F (u, t), u g(U, t), we can write the function p(w, t) as 1 Fu (u, t)du p(w, t) = 1 , w(u w)
g(U,t)
U ,
6.5. Parametric representation of univalent maps and f (, t) satises the dierential equation f (, t) f (, t) = p(, t), t U .
165
Theorems 6.5.2 and 6.5.11 are linked as follows. For a given subordination chain of maps f (, t) , that exists for t [0, t0 ) and U , there is a C function d(ei , t) by Theorem 6.5.11 and we can construct the function (, t) by the Douady-Earle extension and the formula (6.11). Then, the function f (, t) satises the equation of Theorem 6.5.2 with p(, t) dened by such (, t). Let us consider the ordinary Lwner-Kufarev equation for the functions o smooth on S 1 . If the retracting chain g(, t) is smooth on S 1 , then we use again the embedding of M into T and reach a similar result. Theorem 6.5.12. Let g(, t) be a retracting non-trivial subordination chain of normalized maps that exists for t [0, t0 ) and U . Each g(, t) is meromorphic in b1 (t) + , (t) > 0. An additional U , smooth on S 1 , and g(, t) = (t) +b0 (t)+ assumption is that g : U U for each xed t. Then, there are a real-valued function p0 (t), a complex-valued function p1 (t), and a smooth real-valued function d(ei , t), such that Re p(, t) > 0 for U , p(, t) = p0 (t) + 1 p1 (t) 2i
S1
zg (z, t) g(z, t)
U ,
and w = g(, t) is a solution to the dierential equation dw = wp(w, t), dt with the initial condition g(, 0) = . Remark 6.5.13. If we work with normalized functions g(, t) = et + then p0 (t) 1, p1 (t) 0. b1 (t) + , w g(U , t)
166
1 p0 (t) = 2 and
1
0
|f
(ei , t)|2
d,
1 p1 (t) =
ei d, |f (ei , t)|2
2 . |f (ei , t)|2 Apart from the trivial case of growing ellipses there are no self-similar solutions, and therefore the Hele-Shaw dynamics f (, t) generates a non-trivial path in T . Thus, given a Hele-Shaw evolution (t) = f (S 1 , t) we observe a dierentiable non-trivial path on T , such that at any time t the tangent vector is a harmonic Beltrami dierential given by d(ei , t) = 3 (, t) =
2
(, t) =
(1
ei )4 |f
167
Figure 6.4: DLA clusters undergoing Brownian motion in some uid and letting them adhere irreversibly on contact with another one bring us to the basics of DLA. Fix a seed particle at the origin and start another one from innity letting it perform a random walk. Ultimately, that second particle will either escape to innity or contact the seed, to which it will stick irreversibly. Next another particle starts at innity to walk randomly until it either sticks to the two-particle cluster or escapes to innity. This process is repeated to an extent limited only by the modelers patience. The clusters generated by this process are highly branched and fractal (see Figure 6.4). The DLA model was introduced in 1981 by Witten and Sander [345], [346]. It was shown to have relations to dielectric breakdown [248], one-phase uid ow in porous media [57], electro-chemical deposition [130], medical sciences [303], etc. A new conformal mapping language to study DLA was proposed by Hastings and Levitov [162], [163]. They showed that two-dimensional DLA can be grown by iterating stochastic conformal maps. Later this method was thoroughly handled in [72]. For a continuous random walk in 2-D the diusion equation provides the law for the probability u(z, t) that the walk reaches a point z at the time t, u = u, t where is the diusion coecient. When the cluster growth rate per surface site is negligible compared to the diusive relaxation time, the time dependence of the relaxation may be neglected (see, e.g., [346]). With a steady ux from innity and the slow growth of the cluster, the left-hand side derivative can be neglected and we have just the Laplacian equation for u. If K(t) is the closed aggregate at
168
Figure 6.5: Kochs snowake the time t and (t) is the connected part of the complement of K(t) containing innity, then the probability of the appearance of the random walker in C \ (t) is zero. Thus, the boundary condition u(z, t) (t) = 0, (t) = (t) is set. The only source of time dependence of u is the motion of (t). The problem resembles the classical Hele-Shaw problem, but the complex structure of (t) does not allow us to dene the normal velocity in a good way although it is possible to do this in the discrete models. Now let us construct a Riemann conformal map f : U C which is meroa1 (t) morphic in U , f (, t) = (t) + a0 (t) + + , (t) > 0, and maps U onto (t). The boundary (t) need not even be a quasidisk, as considered earlier. While we are not able to construct a dierential equation analogous to the PolubarinovaGalin one on the unit circle, the retracting Lwner subordination chain still exists, o and the function f (, t) satises the equation f(, t) = f (, t)pf (, t), U , (6.33)
e where pf (, t) = p0 (t) + p1 (t)/ + is a Carathodory function: Re p(z, t) > 0 for all U and for almost all t [0, ). A dierence from the Hele-Shaw problem is that the DLA problem is well-posed on each level of discreteness by construction. An analogue of DLA model was treated by means of Lwner chains o by Carleson and Makarov in [52]. In this section we follow their ideas as well as those from [170]. Of course, the fractal growth phenomena can be seen without randomness. A simplest example of such growth is the Koch snowake (Helge von Koch, 1870 1924) (see Figure 6.5). DLA-like fractal growth without randomness can be found, e.g., in [73]. A new development, called Schramm-Lwner Evolution (previously called o Stochastic Lwner Evolution by O. Schramm [306]) provides another probao bilistic view on the Lwner chains and a new interpretation of the traditional o conformal eld theory approach. The basic idea is to replace the control function p in the classical Lwner equation by a function with driving parameter a o one-dimensional Brownian motion (see [289], [306]). Returning to the fractal growth we want to study a rather wide class of models with complex growing structure. We note that (t) = cap K(t) = cap (t). Let M (0, 2) be the class of positive measures on [0, 2]. The control function
169
pf (, t) =
0
ei + dt (), ei
and p0 (t) = t , where t () M (0, 2) for almost all t 0 and absolutely continuous in t 0. Consequently, (t) = (t) t . There is a one-to-one corre spondence between one-parameter (t) families of measures t and Lwner chains o (t) (in our case of growing domains C \ (t) we have only surjective correspondence). Example 1. Suppose we have an initial domain (0). If the derivative of the measure t with respect to the Lebesgue measure is the Dirac measure dt () 0 ()d, then ei0 + , pf (, t) i0 e and (t) is obtained by cutting (0) along a geodesic arc. The preimage of the endpoint of this slit is exactly ei0 . In particular, if (0) is a complement of a disk, then (t) is (0) minus a radial slit. Example 2. Let (0) be a domain bounded by an analytic curve (t). If the derivative of the measure t with respect to the Lebesgue measure is 1 dt () = , d 2|f (ei , t)|2 then 1 pf (, t) = 2
2
ei + d, |f (ei , t)|2 ei 1
and letting tend to the unit circle we obtain Re [f f ] = 1, which corresponds to the classical Hele-Shaw case, for which the solution exists locally in time. In the classical Hele-Shaw process the boundary develops by uid particles moving in the normal direction. In the discrete DLA models, either lattice or with circular patterns, the attaching particles are developed in the normal direction too. However, in the continuous limit it is usually impossible to speak of any normal direction because of the irregularity of (t). In [52, Section 2.3] this diculty was circumvented by evaluating the derivative of f occurring in t in the above Lwner model slightly outside the boundary o of the unit disk. Let (0) be any simply connected domain, (0), 0 (0). The derivative of the measure t with respect to the Lebesgue measure is 1 dt () = , d 2|f ((1 + )ei , t)|2
170
with suciently small positive . In this case the derivative is well dened. It is worth mentioning that the estimate cap (t) = (t) t 1
would be equivalent to the Brennan conjecture (see [271, Chapter 8]) which is still unproved. However, Theorem 2.1 [52] states that if R(t) = max |f ((1 + )ei , t)|,
[0,2)
R(t + t) R(t) C , t
for some absolute constant C. Carleson and Makarov [52] were, with the above model, able to establish an estimate for the growth of the cluster or aggregate given as a lower bound for the time needed to multiply the capacity of the aggregate by a suitable constant. This is an analogue of the upper bound for the size of the cluster in two-dimensional stochastic DLA given by [193].
Chapter 7
172
Chapter 7. Hele-Shaw Evolution and Strings repulsion between like electrical charges or the interaction of bar magnets. It can be attractive or repulsive, and acts only between pieces of matter carrying electrical charge.
The gravitational force is weak, but very long ranged. It is always attractive, and acts between any two pieces of matter in the Universe since mass is its source. It binds together galaxies in clusters and governs the planets motion. Einsteins great dream (Albert Einstein, 18791955) was to come up with a unied theory of all forces. He believed that all forces were dierent aspects of the same force. Modern research shows that for objects at very high energy, the weak force and the electromagnetic force behave similarly. At the same time, the strong force appears to grow weaker. This caused the appearance of the socalled Theory of Everything and the Big Bang Theory on the global structure and history of the Universe. They suggest that the Universe originated from a single location of extremely high energy and matter densities. All forces were united in one superforce and perhaps it was acting on only one type of particles. After a huge explosion the Universe expanded, the energy decreased, unstable particles decayed, stable matter formed. Theories that postulate the unication of the strong, weak, and electromagnetic forces are called Grand Unied Theories. Theories that add gravity and try to unify all four fundamental forces into a single force are called Superunied Theories. The theory that describes the unied electromagnetic and weak interactions is called the Standard Electroweak Theory, or sometimes just the Standard Model. Grand Unied and Superunied Theories remain theoretical speculations so far as yet unproved, but there is strong experimental evidence for unication of the electromagnetic and weak interactions in the Standard Electroweak Theory. Furthermore, although Grand Unied Theories are not proved experimentally, there is strong circumstantial evidence to suggest that a theory at least like a Grand Unied Theory is required to make sense of the Universe. The central paradox is the apparent incompatibility of two main theoretical foundations of particle physics. The rst foundation is Einsteins general theory of relativity, which relates the force of gravity to the structure of space and time. The second theoretical foundation is quantum mechanics formulated for three of the four known forces of nature: the strong, weak and electromagnetic interactions, which can account for the atomic and subatomic world. Moreover, there have been discovered force-carrying particles for these three forces. They are for example, photons (the electromagnetic force), mesons and gluons (the strong force), and intermediate vector bosons or W and Z bosons (the weak force). No elementary particle that carries gravity (hypothetic graviton) have been observed yet. The concept of what is elementary, i.e. not built of anything else, changes as science progresses. Thirty years ago, it was believed that the proton was an elementary particle, but it has long since been known that it is built from three quarks. One of the pretenders to be a unied theory is a so-called Higgs eld. In the standard model of particle physics all the quarks and leptons turn out to be
173
massless. Peter Ware Higgs (born May 29, 1929), FRSE, FRS, until recently the holder of a personal chair in theoretical physics at the University of Edinburgh and now an emeritus professor, proposed in 1964 a eld that (like Ether!) permeates all space. All particles interact with this eld and it gives them mass. The Higgs particle responsible for such a eld has not yet been observed although attempts will be launched to try to observe the Higgs particle directly. Until recently there seemed to be little hope that Einsteins theory of gravity, the fourth fundamental force, could be united with the precepts of quantum mechanics. In the last decade elementary particle physicists have become optimistic [74], [129], [160] that the theoretical impasse might be resolved. The optimism is based on striking developments in a new kind of theory known as string and superstring theory. The far-reaching claims made for this theory would, if correct, provide the much sought-after Theory of Everything, the unication of physics. The familiar elementary particles are understood as dierent modes of a single string. Superstring theory combines string theory with a mathematical structure called supersymmetry. It may be considered as the fth element of modern physics.
174
tions can not be determined any longer. However, we may assign a probability density to each path which is the higher, the shorter is the path, or in the setup of Lagrangian mechanics, the smaller the action is. All elementary particles are topological points but in Feynmans interpretation of quantum mechanics they possess some new internal structure. This can be expressed in the following denitions. A string is a 1-dimensional object moving in D-dimensional space-time M. The rst coordinate is treated as time. As the string moves in time, it sweeps a 2-dimensional surface in M which is called the worldsheet (WS) of the string (analogously with the worldline of a Newtonian particle). We need to construct the probability density over the space of all surfaces connecting the initial and nal position of the string. The density must be the higher the smaller is the action (least action or least energy principle). If we denote by S[] the action where is a quantity characterizing the surface (the metric further on), the idea of Gaussian distribution yields the simplest Feynman (path) integral1 n D e ini or the partition function, or the correlator, where stands for the Planck constant and D is the most sophisticated part of this integral, which is roughly speaking a local product xM d((x)) of measures on M. The symbol D here is a concise way to represent the innite-dimensional integral over all possible eld congurations on all of space-time. The problem with the denition of D is that there is no translation-invariant measure on an innite-dimensional vector space. The path integral is rather formal and may be given strict mathematical sense only in some simple situations. Up to now, only in dimension D = 26 is the theory consistent and the integral (7.1) mathematically well dened as an integral over the Teichm ller space with respect to the Weil-Petersson measure. u Since Feynman formally wrote down [113] the rst path integral formulation of quantum mechanics, several denitions have been proposed from dierent perspectives. For example, Albeverio and Hoegh-Krohn [11] develop a general theory of oscillatory integrals on Hilbert spaces which contains as a particular case a denition of Feynman path integrals for nonrelativistic quantum mechanics, quantum statistical mechanics, and quantum eld theory. Their formalism can be used for computing path integrals by computers more readily than for computing path integrals analytically. A good mathematical and historical treatment of Feynmans integral is found in [185]. The partition function (7.1) represents the transition probability between two observed points. We shall parameterize WS by means of two parameters (, ). Let x be phase coordinates = 0, 1, . . . , D 1 of M, and let M be at. The relative spacetime position of a point on the string WS is given by the functions x (, ). The
1 Throughout
S[]
(7.1)
175
parameters (, ) have no physical signicance, however the functions x (, ) tell us how the string moves in M, and hence, they are the dynamical variables. Strings may be closed if the dynamical variables are periodic with regard to or open otherwise. The parameter is normally taken in the interval (, ) or [0, ). We shall consider only closed strings. Of course, 2-dimensional objects, worldsheets, may be generalized to any dimension. Edward Witten proposed the existence of so-called M-theory at a conference at the University of Southern California in 1995. According to Witten himself, M stands for magic, mystery or membrane, depending on ones taste. He proposed a multidimensional object usually called p-brane (derivative of membrane) referring to its spatial dimension p (for example, a string is a 1-brane and a membrane is a 2-brane). M-theory is a solution proposed for the unknown theory of everything which would combine all ve superstring theories and 11-dimensional supergravity together. According to Witten, mathematical tools which have yet to be invented are needed in order to fully understand it.
7.1.2 Metrics
We shall use: the ambient Euclidean (+,+) or Lorenzian (,+) metric dsM of M, 1 0 = ... 0 0 1 ... 0 ... ... ... ... 0 0 ; ... 1
ds2 = dx dx , M
the intrinsic metric dsc of the canonical cylinder, which is also Euclidean or Lorenzian 0 = ds2 = d d , , , = 0, 1. c 1 = The intrinsic metric does not depend on the embedding of the worldsheet into M; the induced metric dsws on the worldsheet ds2 = ws x x d d = g d d .
an unknown apriori (variable) metric ds on the worldsheet as on a 2-D manifold, such that ds2 = h (, ) d d .
176
7.1.3 Dynamics
Dynamical variables x (, ) satisfy the equations of motion derived by making use of the least action principle. In Lagrangian mechanics, a system with a conguration space N is characterized by its Lagrangian L a smooth, real-valued functional on the direct product of the tangent bundle T N and the time axis. If P N stands for the path space (which is an innite-dimensional Frchet manifold), the e action functional S is the real-valued integral functional dened by S = ini Ldt, where t is the time parameter. Turning to the space-time conguration space M and a worldsheet in M we work with the action functional S = ws (x)dx where the Lagrangian is included in to this spatial integral. So dened action must have some physical or geometrical relevance. In classical mechanics, action is the dierence between kinetic and potential energies. Yichir Nambu (1921present, from o o the University of Chicago, 1994/95 Wolf prize in Physics) in 1970 [243] and Tetsuo Got (from the Nihon University Tokyo) in 1971 [128] proposed to choose the area o of the worldsheet as the simplest action. This action, known as the Nambu-Goto action, admits the form SN G = T
ws
dx = T
C
d 2
| det g |,
where C is the canonical cylinder and the quantity T , the string tension, has dimension mass per unit length. The string undergoes geodesic motion, i.e., the dynamical variables satisfy the Euler-Lagrange equation 1 | det g | | det g | g x = 0, = 0, . . . , D 1,
1 where g = g . The metric g is dened on the worldsheet embedded into M, thus it depends on the mapping x(, ), and as the equation for x is highly non linear. Moreover, the square root make it unpleasant to think of the path integrals as of Gaussian integrals. To overcome these problems Alexander Markovich Polyakov (1945present, formerly at the Landau Institute in Moscow, currently at Princeton University, 1986 Dirac Medal, 1994 Lorentz Medal) proposed to introduce Lagrange multipliers in 1981 [267]. He considered the variable metric h as an additional variable and the Dirichlet integral
SP =
T 2
C
d 2
| det h | h
x x
as Polyakov action. The Euler-Lagrange equation (with regard to the variation of the worldsheet) for the Polyakov action SP is formally the same as for the Nambu-Goto action SN G , 1 | det h | | det h | h x = 0, = 0, . . . , D 1, (7.2)
177
but h does not depend on x any more. So we have linear equations for x(, ). The Euler-Lagrange equation must be supplemented by a boundary condition such that the mapping x(, ) preserves the innitesimal boundary structure (tangent vectors to C go to tangent vectors to W S). There are three extra degrees of freedom in h but they can be removed by using the equation of motion for h, SP = 0, h where the left-hand side is a functional derivative. The two-dimensional energymomentum tensor is dened as T = 2 T SP 1 , h | det h |
and the equation of motion implies T = 0. Equation (7.2) yields that x(, ) is conformal, i.e., that the metric h on the canonical cylinder C is proportional to the induced metric on W S = x(C). Moreover, SP = SN G under T = 0, whereas in general, SP SN G .
7.1.4 Symmetries
A guiding principle of modern physics is that there is a close connection between the standard conservation laws and the symmetries of space-time. There is no single theorem describing such a connection, however this principle is sometimes referred to as the Noether theorem, although the Noether2 1918 theorem [250] itself contains only a very partial statement of it. By symmetries for the Polyakov action SP we mean transformations keeping the action invariant. Global symmetries Euclidean isometries or Poincar invariance in the Lorenzian case e (Poincar group) applied to WS: e x x + b ;
x x + x ; where = are innitesimal Euclidean or Lorenzian rotations.
2 Emmy Noether, (born Nther), 18821935, achieved the doctorate in 1907 under Paul Goro dan. She was one of the most talented mathematicians of the early 20th century but she was refused permission to teach by the university of Gttingen, and her colleague, David Hilbert, o had to advertise her courses in the universitys prospectus under his own name. Finally, she was admitted to the faculty in 1919.
reparametrization invariance (dieomorphisms in 2-D): (, ) implies the invariance of the area element d 2 Weyl rescaling h d d e(,) h d d . Weyl rescaling leaves | det h| h invariant Translations from the Euclidean isometries or the Poincar group led to the e conservation of energy and momentum, while innitesimal Euclidean or Lorenz rotations led to the conservation of angular momentum. The Euclidean isometries and the Poincar group appear as a global internal symmetry in M but not in the e 2-D space-time C. The conformal or Weyl invariance implies that the energy-momentum tensor is always traceless: h T = tr T = T = 0. The metric h is dened up to a (,) conformal factor e , so we can choose a conformal gauge, i.e., we can choose a general coordinate transformation that brings the metric to the form h = e(,) with the metric , and then we can apply Weyl scaling to eliminate the conformal factor. We consider the complex coordinates d = id + d and d = id + d, ds2 = 4(d 2 + d 2 ). In this gauge, the Polyakov action describes a free two-dimensional conformal eld theory and the equation of motion for x(, ) = x(, ) is simply the free wave equation 2 2 + 2 2 x(, ) = 0. | det h| = d 2 | det h|.
The general solution to this equation is x(, ) = ( + ) + ( ). After the change of variables i (Wick rotation) we get a complex form of it, x(, ) = () + (), where () and () describe the left-moving and the right For closed strings the worldsheet theory splits into moving components of x(, ). two independent components. Wick rotation is a method of nding a solution to a problem in Minkowski space from a solution to a related problem in Euclidean space, by analytic continuation. This solution may then, under reverse substitution, yield a solution to the original problem. Analogously, one may consider the Euclidean metric and space instead which is more familiar than the Minkowskii space. The equation of motion becomes the Laplacian 2 x(, ) = 0. In this case we get x(, ) = () + () as a general solution. So in both cases the motion splits into the left-moving and the right-moving components of x(, ).
7.2. Correlator
179
However, they are not completely equivalent from the mathematical point of view, because in the Laplacian case the solution must be harmonic, and therefore necessarily have strong regularity properties. In the Euclidean case one may interpret the above procedure in the following way. We consider the complex coordinates d = id + d and d = id + d, the surface S dieomorphic locally to the canonical cylinder C, and the conformally invariant metric e(,) with the Kronecker delta. Expressed in complex coordinates we have the invariant form e(,) d 2 = 4
x d
which is a holomorphic quadratic dierential in terms of the local parameter . The surface S is a critical point for SP with respect to variations by families of dieomorphisms. In general, the canonical surface C is a Riemann surface.
7.2 Correlator
7.2.1 Correlator and partition function
In probability theory and statistics, a correlator, also called a correlation coefcient, indicates the strength and direction of a linear relationship between two random variables. In general statistical usage, correlation or co-relation refers to the departure of two variables from independence. The correlator X,Y between two random variables X and Y with expected values E(X) and E(Y ) is X,Y = E(XY ) E(X)E(Y ) E(X 2 ) E 2 (X) E(Y 2 ) E 2 (Y ) .
The complete information on X,Y is got by calculating Xk Xj = E(Xk Xj ), X1 = X, X2 = Y , j, k = 1, 2. Often, the quantity Xk Xj is called a correlator instead of X,Y . Quantum particles, unlikely classical ones, can roughly speaking randomly deviate from their classical trajectories. Therefore, given the position and velocity of a particle at a given moment, we can not predict its position at a later time. The particle motion in M in a potential eld M RD is described by a stochastic process (t) = (0 (t), . . . , D1 (t)). By the Kolmogorov extension theorem [203], this process is completely described by the correlators j1 (t1 ) . . . jn (tn ) , n = 0, . . . , D 1,
which by denition are the expected values of the products of random variables j1 (t1 ) . . . jn (tn ). A typical eld of application of this idea is Brownian motion. A quantum particle is much less intuitive and much more dicult to visualize. But Feynman pointed out that the behaviour of a quantum particle in a potential eld is correctly described by the same model by appropriate choosing of the
180
weight, or in other words, of the correct measure in terms of integration. So the complete information on the dynamics of a quantum particle can be expressed by the correlator j1 (t1 , x1 ) . . . jn (tn , xn ) = normalized as D e
D j1 (t1 , x1 ) . . . jn (tn , xn )e 1
S[]
S[]
= 1,
where D is a measure on the space of all possible paths in M, and S[] is an action. It is clear that such a denition is not satisfactory yet from the mathematical point of view because of the innite dimensional integration. In the case of Brownian motion a justication is possible within the standard Lebesgue measure theory (theory of Wiener integrals). In quantum mechanics this setting is less satisfactory. However, physicists made several interesting conclusions using this path integral formalism, conrmed to be true, which gives certain expectations for a corresponding mathematical justication to be developed. One of the important goals of any eld theory, including conformal eld theory, is to calculate the full set of partition functions. In the Euclidean setup of the bosonic string theory the heuristic idea is to make sense of the path integral Dx Dh eSP [h] , where we have put = 1 for simplicity. The integration is taken over the space of all mappings x : C M (with prescribed boundary conditions) and over all metrics h on the canonical dierentiable surface C with boundary (which we took previously to be a cylinder). Considering equivalent metrics we arrive at the space of deformations, the Teichm ller space. Finally, the measure is decomposed into a u measure on the space of conformal factors (due to Weyl3 symmetry) and the WeilPetersson measure on the nite Teichm ller space. The details of this construction u exceed the scope of our book so we refer the interested reader to [186].
7.2. Correlator
181
respect to innite dimensional metric factors from the Weyl symmetry. The latter locus is called the conformal anomaly. In string theory, conformal symmetry on the worldsheet, the local Weyl symmetry, must cancel if the theory is to be consistent. The required cancellation implies that the spacetime dimensionality must be equal to the critical dimension, which is either 26 in the case of bosonic string theory or 10 in the case of superstring theory. The classical theory has no mass scale so there is a conformal symmetry, but this is broken by a conformal anomaly. In classical 2D gravity, the metric is purely gauge, because it has 3 components and there are 3 gauge symmetries, 2 dieomorphisms and 1 Weyl rescaling of the metric. But after quantization, the trace of the metric becomes physical. In lightcone quantization, the physical Hilbert space becomes unexpectedly large, because you cannot factor out the conformal symmetry unless the number of scalar elds equals D = 26, due to the conformal anomaly. Provided that D < 26, the theory is consistent in spite of the anomaly, because the Hilbert space has a positive-denite inner product (the spectrum is ghost-free). Alternatively, one can introduce the trace of the metric as a dynamical eld already at the classical level; it is then known as the Liouville eld (Joseph Liouville, 18091882). The parameters can be ne-tuned so that the Liouville eld has central charge c = 26 D, canceling the matter and ghost contributions, and proper conformal invariance is restored. However, the physical Hilbert space is still larger than expected classically, because it now includes the Liouville eld. Thus, the associated Liouville action is that of a soluble QFT, although the integration measure is not the usual one in the functional approach to quantum eld theory. It is a theory of conformal anomaly. There is no particular reason for the length parameter to drop out of the problem, i.e., to apply Weyl invariance. Therefore, the string is non-critical in the sense that we use the metric e(,) instead of . The correlator is the path integral A =
C(S)
D A() e
S[]
We take S to be an n-punctured spherec and C(S) consists of all smooth conformal metrics ds2 = e(,) |d 2 | with asymptotics e(,) (| j | log | j |)2 near the punctures j . This concrete situation was thoroughly treated by Takhtajan and Zograf [318], [348], [349]. The correct Liouville action for the non-critical strings becomes S[] = lim
0 S()
(7.3)
where S() is obtained by removing -disks centered on the punctures. The least action principle leads to the Liouville equation 2 1 = 2 e .
182
The geometric meaning of this equation is that the conformal metric e(,) |d|2 on S has constant negative curvature 1. The asymptotics guarantee that the metric is complete and the area of the surface S is nite due to the Gauss-Bonnet formula. The additional terms in the formula for S[] are present because the integrand (the rst term) does not keep the local 2-form | |2 d d invariant under the change of the local parameters on charts of S.
7.2.3 Superstrings
We already mentioned classical symmetries. Supersymmetry is a hypothetical symmetry that relates bosons and fermions. In supersymmetric theories, every fundamental fermion has a bosonic superpartner and vice versa. There is no direct experimental evidence so far that supersymmetry exists in the real world. However, there is some indirect evidence which suggests that supersymmetry may be found at energies not too far above those accessible by todays particle accelerators. When we include fermions in the worldsheet theory of the string, we automatically get a new type of symmetry called supersymmetry which relates bosons and fermions. Fermions and bosons are grouped together into supermultiplets which are related under the symmetry. This is the reason for the super in superstrings. A consistent quantum eld theory of superstrings exists only in 10 spacetime dimensions. We do not enter this topic and refer an interested reader to [189].
kj ak = aj .
k=
This property is similar to one of the main properties of the Dirac delta function:
183
and in fact Diracs delta was named after the Kronecker delta because of this analogous property. For real x and y we have
n=1
sin nx sin ny
n=1
n=
(z w) =
n=
wn z 1n = +
1 w z 1 + + + 2 + ... w2 w z z
For an analytic function f (z), Diracs delta exhibits the standard properties (z w) = (w z), (z w)(z w) = 0, Res z=0 (z w)m n! (z w) n!
m n
2 S [h] = P x . x
To analyse this equation in a distributional sense we insert arbitrary test functions dened in a punctured neighbourhood of and analogously get 2 x insertion = 0,
184
or that x (, ) is harmonic. For the partition function x we obtain 0= = or DxDh SP [h] x e x 2 x (, )x (w, w) ,
DxDheSP [h] 2 ( w, w) +
2 2 x (, )x (w, w) = ( w, w),
in a distributional sense, where 2 (, ) = ()(). Let us consider the innitesimal innite dimensional conformal coordinate transformation d d (1 + ())(1 + ())d d with () = 0. The con dition of the conformal symmetry implies that the energy-momentum tensor of the worldsheet theory T = T (, ) is traceless and symmetric. There are only two independent components which can be chosen to be T () = T = T11 T22 2iT12 and T () = T = T11 T22 + 2iT12 (the components T and T vanish). So, the energy-momentum tensor T (, ) splits into left-moving and right-moving parts T (, ) = T () + T (). The conservation of T (, ) is expressed as T () = 0, T () = 0,
Considering only the holomorphic component T () we have also T () = 0 that corresponds to an innite number of conserved quantities. Now we continue working with the initial postulate. The 2-D coordinate invariance implies n
j1 (1 ) . . . jk (k ) . . . jn (n ) =
k=1
1 2i
d T ()j1 (1 ) . . . jn (n ) ().
This is the basic Ward identity of 2-D QFT (additional terms appear for highergenus Riemann surfaces, see [101]). The contour integration is performed enclosing possible singularities of T (). From the Ward identities, one may get useful identities, operator product expansions and transformation laws. Identities: T () = T () = 0, T = T = 0.
185
under conformal change of variables (). With jk () = (k + )jk (), jk () transform as primary (conformal) holomorphic elds of conformal weight k , k jk () jk ( ). Here S () is the Schwarzian derivative and c appears as a parameter of the theory and is called the central charge. It can not be determined from any general principle. The points and w are taken in the same holomorphic chart. Only the holomorphic part of Diracs participates. Let us introduce the operators Ln , n = 0, 1, 2, . . . as the coecients of the formal Laurent expansion T () = Then Ln = Ln . n+2 n=
d n+2 T (), 2i
where integration is performed over any contour encircling the origin. The operators Ln are called Virasoro generators4 [342]. The operator product of the Virasoro generators Ln Lm can be expressed as Ln Lm =
0
d n+1 T () 2i
dw n+1 w T (w), 2i
where |w| > ||. This product does not commute and the Poisson brackets are [Ln , Lm ] = Ln Lm Lm Ln = (n m)Ln+m + c n(n2 1)n,m . 12
Here the contours cancel except for a contribution as w, and we take into account the operator product T ()T (w). This is the famous Virasoro algebra of string theory.
4 Miguel Angel Virasoro, born in Argentina in 1940. In 1995, Miguel Virasoro from the Univer sity of Rome was appointed to succeed the Nobel Laureate in Physics Abdus Salam (19261996) as the director of the International Centre for Theoretical Physics, Trieste, Italy. He was the Centres director until the end of May 2002.
186
The Virasoro algebra is the central extension of the Witt algebra. To analyze and construct this central extension we rst consider the real situation over the unit circle. Let us consider the Lie algebra Vect S 1 of smooth vector elds on S 1 , d () , d where ( + 2) = (). The Poisson bracket in Vect S 1 is given as [1 , 2 ] = 1 2 1 2 . Let I and G be Lie algebras. An exact sequence is a sequence of objects and morphisms between them such that the image of one morphism equals the kernel of the next. Let us consider the exact sequence of Lie algebras 0 I E G 0. E is called the central extension of G by I if I belongs to the center of E. The central extension is given as E G I. A simple example is [x + a]E = [x, y]G + [a, b]I . The Virasoro algebra is the unique (up to isomorphism) non-trivial central extension of Vect S 1 by R given by the Gelfand-Fuchs cocycle:
2 f g
1 (1 , 2 ) = 2
(1 2 1 2 )d.
0
The Virasoro algebra V ir is a Lie algebra over the space Vect S 1 R dened by the commutator [(1 , a), (2 , b)]V ir = ([1 , 2 ]Vect
S1 ,
c (1 , 2 )), 12
187
where a and b are elements of the center, ab ba vanishes, and c is the central charge. Integration by parts leads to the 2-cocycle condition (1 , [2 , 3 ]) + (2 , [3 , 1 ]) + (3 , [1 , 2 ]) = 0, and 1 (1 , 2 ) = 4
2
(1 + 1 )2 d.
0
Correspondingly, we consider the group Di S 1 of dieomorphisms of the circle preserving its orientation f (), where (mod 2) is a parameter on S 1 , f ( + 2) = f (). The Virasoro-Bott group is the unique (up to isomorphism) non-trivial central extension of Di S 1 given by the Thurston-Bott cocycle [40]
2
1 (f, g) = 2
log((f g) )d log(g ).
0
c (f, g)). 12
We shall identify Vect S 1 with the functions with vanishing mean value over S . So we have
() =
n=1
an cos n + bn sin n.
J() =
n=1
an sin n + bn cos n.
iJ() =
and the latter extends into the unit disk as a holomorphic function. Taking the basis of Vect S 1 C as en = iein we get [en , em ] = (n m)en+m + c n(n2 1)n,m . 12
So this shows that the Virasoro algebra is realizable both as a central extension of the Witt algebra and as an algebra of the Virasoro generators of CFT.
188
z k dz ,
k 1;
C0 =
interior
dz .
Here z = x + iy and it is understood that the point z = 0 is inside the domain, whereas the point z = is outside. The integrals at k = 1, 2 are assumed to be properly regularized. In order to nd the harmonic moments of the interior Ck =
interior
z k dz ,
k 1,
the set of moments of the exterior (and their complex conjugate) are identied with the times of the dispersionless 2D Toda hierarchy as follows: t t0 = C0 , tk = Ck , k Ck tk = , k k 1.
These parameters are treated as independent variables. The derivatives of the har monic moments of the interior Ck with respect to the tj , tj satisfy the symmetric lattice equation Ck Cj Ck Cj = , = . tj tk tj tk Wiegmann and Zabrodin [344] constructed a real-valued -funcion, a solution of the so-called Hirota equation, such that Ck log = , tk The Hirota equation reads as Sf 1 (z) = 6 z2
k,n=1
log Ck = , tk
k 1.
2 log , z n+k tk tn 1
189
where z = f () is the parametric map of the unit disk onto the exterior phase domain. The left-hand side is just the analytic component of the energy-momentum tensor.
d =
(t)
(z)
G(z, a) ds n
for every smooth function , and where G(z, ) := G(t) (z, ) is the Green function of the domain. Then, by denition, ((a)F )(D) := d dt F ((t)).
t=0
One can show that any innitesimal domain variation of a smooth domain D can be approximated by linear combinations of (a) with a D. See [301] for a general theorem in this direction. Taking a close to the boundary, the variation corresponding to (a), i.e., Hele-Shaw injection at a, produces a bump on the boundary at a. Example 7.4.1. Let h be a xed harmonic function dened in a neighbourhood of the closure of D and dene F by F (D) =
D
hd.
hd =
D
h(z)
t=0 k
GD (z, a) ds = h(a). n
Taking, e.g., h(z) = z we get moments such as in the previous subsection as special cases of the functional F , with (a)
D
z k d = ak
190 for k 0.
Example 7.4.2. As a second example, choose two points b, c D and take F to be F (D) = GD (b, c). For a general domain variation, the change of this functional is given by the Hadamard variational formula: following classical tradition, let n denote an innitesimal change of the boundary in the normal direction and let GD denote the corresponding change of the Green function. Then the Hadamard formula is GD (b, c) =
D
On dividing with a corresponding innitesimal time interval t we may view n/t as the normal velocity of the boundary. In our case (Hele-Shaw injection at a) we have n G(, a) = . t n Thus, we get the beautiful formula (a)GD (b, c) =
D
(7.4)
In particular we see that (a)GD (b, c) is completely symmetric in a, b and c. The symmetry in the above formula (7.4) calls for an explanation. This will be provided in the next example, where an explicit version of the -function appears. We rst remark that the Richardson conservation of moments for Hele-Shaw evolutions, which in a stronger form is built into the concept of weak solution and into the balayage formulation, contains an integrability statement: Hele-Shaw injection at a point a followed by injection at a point b gives the same result as if the injections had been performed in the opposite order. In innitesimal form this statement becomes (a)(b) = (b)(a). Therefore, the above symmetry would be explained if we could nd a functional E, such that G(b, c) = (b)(c)E + constant. Such a functional indeed exists and can be realized as a renormalized energy of the complementary domain, and which can also be identied with the logarithm of the -function. Example 7.4.3. Let B be a xed bounded domain containing the closure of D and dene 1 E(D) = log |z |dz d . 4 B\D B\D
191
In this functional the domain D occurs symmetrically at two places. Thus, the result of (c) acting on D in E(D) will be twice the result of its action on only one of the occurrences. This gives (c)E(D) = 1 2 1 d 2 dt log |z |d dz
t=0 B\D B\(t)
log |z |
B\D D
GD (, c) ds dz n
1 = 2
Here we used that log |z | is harmonic in D in the variable when z B \ D. Recall that U B\D denotes the potential of the set B \ D. In order to continue we decompose the Green function G(z, ) as G(z, ) = 1 log |z | + H(z, ) 2
with H(z, ) harmonic in each variable. Then, applying (b) to the above and using that G(z, c) = 0 on D we obtain (b)(c)E(D) = = 1 2 1 2 log |z c|
D
G(, b) ds dz n
G(, b) ds dz n
1 log |b c|. 2
(b)(c)E(D) = GD (b, c) + C(b, c), where C(b, c) is a constant with respect to variations of D. Applying nally (a) we get (a)(b)(c)E(D) = (a)GD (b, c), and the symmetry of (a)GD (b, c) is explained. The -function [206], originally introduced in the context of integrable hierarchies, can in the present situation be chosen so that log (D) = E(D). One may also consider higher order derivatives, like dening (k) (a) by (k) (a)F (D) = k (a)F (D). ak
z j dz = k!kj ,
D
z k dz .
To connect to the results in the preceding subsection one should take D to be the exterior (unbounded) domain. Then the derivatives tk will correspond to (k) () (suitably interpreted) and the statements there may be derived using the commutativity of the -derivatives.
= 0,
(7.6)
so it is the Green function of the domain (t) with a singular point at the origin. The non-trivial motion regarding time t is given by the kinematic condition on (t). We have W p p = i , z = x + iy. z x y Since Greens function solves (7.5)(7.6), we have the representation W (z, t) = log z + w0 (z, t), (7.7)
where w0 (z, t) is an analytic regular function in (t). Because of the conformal invariance of Greens function we have the superposition (W f )(, t) = log ,
193
and the conformally invariant complex velocity is just W (z, t) = f 1 (z, t), where f = f 1 (z, t) is the inverse to our parametric function f and prime means the complex derivative. Rewriting this relation we get (W (z, t) dz)2 = d 2 . 2 (7.8)
The velocity eld (p) is the conjugation of (W ). In other words the velocity eld is directed along the trajectories of the quadratic dierential in the left-hand side of (7.8) for each xed moment t. The equality (7.8) implies that the boundary (t) is the orthogonal trajectory of the dierential (W (z, t) dz)2 with a double pole at the origin. The dependence on t yields that the trajectory structure of this dierential changes in time, and in general, the stream lines are not inherited in time. These lines are geodesic in the conformal metric |W (z, t)||dz| generated by this dierential. We plan to construct a eld theory for the Hele-Shaw evolution somehow analogous to the Liouville theory. However, the Liouville theory itself does not work directly any longer in our case. It is not directly connected with the classical Green function; symmetries given by the Poincar group are not those relevant to e the Hele-Shaw dynamics. Moreover, observe that being thought of as a Riemann surface, the phase domain (t) \ {0} is hyperbolic and it admits the Poincar e metric with constant negative curvature ds2 = |f 1 |2 |dz|2 . |f 1 |2 log2 |f 1 |
The asymptotics about the origin and close to the hyperbolic boundary are such that the standard expression (7.3) (the principal term of it) for the Liouville action can not be used any more. Moreover, the action integral for the hyperbolic metric in the punctured unit disk has the following asymptotics: (|z |2 + e )dz
U ()
where U () = {z : 1 < |z| < 1 2 }. Therefore, the corresponding integral for (t) \ {0} will have a similar asymptotics plus terms containing f (0, t) and the boundary distortion (in our case the boundary derivatives) at the unit circle, which makes it dicult to operate with. Let us instead use the at logarithmic metric generated by (7.8) which is natural for the Laplacian growth ds2 = |f 1 |2 |dz|2 = |W |2 |dz|2 . |f 1 |2
194
The hyperbolic boundary is not singular for this metric whereas the origin is. But it is a parabolic singularity which can be easily regularized. The density of this metric satises the usual Laplacian equation zz = 0 in (t) \ {0}, where |f 1 |2 ds2 = e(z) |dz|2 , (z) = log 1 2 . |f | Obviously, the Laplacian equation is the Euler-Lagrange equation for the variational problem dened by the Dirichlet integral |z |2 dz ,
D
locally for any measurable set D (t) \ {0}. However, this functional cannot be dened globally in (t) \ {0} because of the parabolic singularity at the origin. To overcome this obstacle we dene the classical action in the following way. Let (t) = (t) \ {z : |z| } for a suciently small , U () = { : < || < 1}. The function possesses the asymptotics log Therefore, the nite limit lim 0 1 , |z|2 |z |2 1 |z|2 as z 0.
|z |2 dz + 2 log
=: S[]
(7.9)
(t)
exists and we call it the logarithmic action for the Laplacian growth. Remark 7.4.4. As in the Liouville theory we have no reason to discard the conformal factor (Weyl rescaling) and we have the rotation invariance instead of the Poincar group. Another important issue is that the innitesimal structure at the e boundary is invariant under the mapping f , i.e., the stream lines within the unit disk are radial, orthogonal to the unit circle and that they remain orthogonal under the conformal map f . This means that the boundary conditions are satised. Remark 7.4.5. The logarithmic action represents the reduced area of the Hele-Shaw worldsheet in the Riemannian metric ds2 . Remark 7.4.6. The potential (z) possesses a particular geometric meaning. It gives rise to a vector eld , whose projection onto each normal outward vector to the level line of Greens function is the curvature of this line, /n = . Lemma 7.4.7. The Euler-Lagrange equation for the variational problem for the logarithmic action S[] is the Laplacian equation = 40 (z), z (t), where 0 (z) is the Dirac distribution supported at the origin, where is taken from the class of twice dierentiable functions in (t) \ {0} with the asymptotics log |z|2 as z 0.
195
|z |2 dz =
C
(t) |z |2 dz ,
where (t) is the characteristic function of (t). Then, due to Greens theorem, lim S [ + hu] S [] h = = 2
C
h0
(t) Re z uz dz 1 2
(t)
u dz +
1 2
(t)
ds, n
(7.10)
for every C (C) test function u supported in (t). On the other hand, we have /n 2/ as 0 and u = 0 on (t). Therefore, the expression (7.10) tends to 1 udz 2u(0), 2
(t)
as 0, and the latter must vanish, that is equivalent to the Laplacian equation mentioned in the statement of the lemma. Obviously, the logarithmic term in the denition of S[] does not contribute to the variation. Remark 7.4.8. The boundary condition for the potential is (ei ) = log vn , where vn is the normal boundary velocity of the Hele-Shaw evolution. Remark 7.4.9. The above Dirichlet problem is related to an interesting property of plane domains discussed in [305]. If (z) is harmonic in a domain and satises on its boundary the relation 2 , = 2 s n then 2 i 2 dz 2 z is a quadratic dierential on . Returning to the calculation of we get z = 1 f 1 f + f f 1 (z, t).
Hence, the action S can be expressed in terms of the parametric function as 2 f 1 S[] S[f ] = lim + d + 2 log + 2 log |f (0, t)|, (7.11) 0 f
U
196
f 1 + f
1 ||2
(7.12)
The functional (7.12) resembles the universal Liouville action dened by Takhtajan and Teo in [319][321] for quasicircles, which is based on conformal maps from the unit disk and from its exterior. However, (7.12) can not be directly reduced to the Liouville action. Observe that the classical kinetic energy E for the harmonic potential p is calculated by the Dirichlet integral |2pz |2 dz =
D D
|W |2 dz ,
locally for any measurable set D (t) \ {0}. However, this integral again cannot be dened globally in (t) \ {0}. Treating E in the same way as S we come to the nite limit E := lim |W |2 dz + 2 log , 0
(t)
or in terms of the parametric function f , E = E[f ] = 2 log |f (0, t)|. The latter expression allows us to think of E as a capacity which exactly corresponds to the physical sense of E as minimal energy.
is equivalent to the fact that the function f (, t) satises the non linear rst-order partial dierential equation Re f f = 1, || = 1, (7.13)
197
with the initial condition f (, 0) = f0 (). We denote by Sf the Schwarzian derivative 2 f 3 f Sf = , f 2 f and by (, t) = Re 1 +
ei f f
|f (ei , t)|
the curvature of the boundary (t) at the point f (ei , t). Theorem 7.4.10. Let z = f (, t) be the parametric function for the Laplacian growth, E[f ] the kinetic energy, and S[f ] the logarithmic action. Then
2
1
0
|f
(ei , t)|2
d,
(, t) d +
0 0
Proof. Making use of the Cauchy-Schwarz representation we extend equation (7.13) into the unit disk f = f p(, t), (7.14) where 1 p(, t) = 2 Immediately, we obtain that
2 2
ei + d. |f (ei , t)|2 ei 1
(7.15)
1
0
|f
(ei , t)|2
d,
f 1 + f
(1 +
f )p(, t) + p (, t) f
+
0
1 d. |f (ei , t)|2
(7.16)
198
The integral in the rst term of the right-hand side of (7.16) we rewrite by Greens theorem as 1 I= 2i
S1 2
f 1 + f
f (1 + )p(, t) + p (, t) f
1 d 2
1 d, |f (ei , t)|2
taking into account a singularity at the origin. Applying the Cauchy-Schwarz formula to the rst term in I containing p we arrive at
2
2 Re I
=
0
1 + ei
2
f (ei , t) f (ei , t)
d |f (ei , t)|2
2
+Re
0
1 + ei
f (ei , t) f (ei , t)
ei p (ei , t)d
0
1 d, |f (ei , t)|2
or
2 2
d S[f ] = 2 Re I + dt
d |f (ei , t)|2
=
0
1 + ei
2
f f
+
0 2
Re
1 + ei
+
0
Im
1 + ei
Im ei p (ei , t) d.
These equalities are thought of as limiting values making use of the analyticity of f on the boundary and f (, t) = 0 for all in the closure of the unit disk. Let us denote in the latter expression by J2 the last integral and by J1 the intermediate one. We have,
2 2
J2 =
0
Im
1+e
i f
1 Im 2
and
1 |f (ei , t)|2
2 |f (ei , t)|2
Im
ei f . f
Im
1 + ei
f f
d . |f (ei , t)|2
|f (ei , t)|2
d S[f ] = dt
1 |f (ei , t)
2
1 |f
(ei ,
We observe that (,
d 3 1 1 S[f ] = (1 + + + n ) dt 2 4 4
2
3 1 2 4n
2
1 |f (ei , t)|2
+
0 2
(1
d S[f ] = 2 dt
(, t) d +
0 0
Vect S 1 onto S. Let f S and let from Vect S 1 making an innites variation of f given by L [f ]() =
f 2 ( 2i
Kirillov and Yuriev [199, 200] h closed with respect to the commu Vect S 1 . Moreover, Kirillovs resu Vect S 1 Di S 1 such that the map f () from S. It is convenient [202] to ext Vect S. Taking k = ieik , k expressions for Lk = f , f S L0 (f ) = f (
P () =
where Sf () is the Schwarzian d associated right-invariant tangen k = ieik , one constructs th g k . By {k } we denote the dual g on the vector k is given as
where means the natural proj [9] in the context of the construc
time
Figure 7.
[12] S. Albeverio, J. Jost, S. P to string theory, Variationa London Mathematical Soci sity Press, Cambridge, 199
[22] G. Baker, M. Siegel, S. T isolated conformal map sin 120 (1995), no. 2, 348364.
223259.
ies in complex analysis and 2 (Halle, 1988), Pitman R Harlow, 1991, 160164.
[136] V. V. Goryainov, One-par compositional analogue of i Applied Mathematics and Nats. Akad. Nauk Ukrainy Russian)
[138] L. Gromova, A. Vasilev, O coecient functional for un 712. [139] H. Grtzsch, Uber einige E o II, Ber. Verh.- Schs. Akad a 376.
[168] W. Hengartner, G. Schobe in one direction, Collection occasion of his 70th birthd
[169] G. Herglotz, Uber die analy anziehenden Massen, Gekr zu Leipzig, 56 pp.
[170] Yu. E. Hohlov, S. D. Howis A model for non-smooth fre Appl. Math., 47 (1994), 10
[182] G. P. Ivantzov, The tempera crystal growing in a coolin 567569. (in Russian)
[214] P. P. Kufarev, The problem circle with an arbitrary nu 75 (1950), 507510. (in Ru
[215] O. Kuznetsova, On polynom Mat. Zh. 42 (2001), no. 5, (2001), no. 5, 907915.
[232] O. Martio, B. ksendal, Fl mal map can produce fracta no. 1, 110.
[248] L. Niemeyer, L. Pietroner dielectric breakdown, Phys. [249] T. Nishida, A note on a (1977), no. 4, 629633.
[281] M. Reissig, F. H bner, An u models with and without (Graz, 1997), Int. Soc. Ana
[282] M. Reissig, S. V. Rogosin, model for Hele-Shaw movi cooling regularization, Euro
[311] M. Siegel, R. E. Caisch, S and illposedness for the Mu no. 10, 13741411.
3197.
[331] Yuhai Tu, Saman-Taylor all orders (La Jolla, CA, 1 Plenum, New York, 1991, 1
[348] P. G. Zograf, L. A. Takhtaj ters and the geometry of T 0, Mat. Sb. (N.S.) 132(17 USSR-Sb. 60 (1988), no. 1,
[349] P. Zograf, L. Takhtajan, H cessory parameters and K a 355 (2003), no. 5, 1857186
dz Bal dist (, a) S 0 0 qc qc 0 S
characteristic funct area element in z-p partial balayage distance from set class of univalent fu normalized by f () class of univalent fu normalized by f () subclass of of fun subclass of of fun subclass of 0 of fu subclass of of fun extension into U subclass of 0 of fu extension into U class of starlike fun
the norm Banach space norm in the u Banach space with the nite tangent space Lie group of C of the unit cir group of rotat Lie algebra of
capacity, 116 Carathodory, 10 e Carathodory kernel, 11 e Carathodory Kernel Theorem, 1 e Cauchy integral, 25 Cauchy transform, 25, 70 Cauchy-Kovalevskaya Theorem, 1 central charge, 185 central extension, 186 classical solution, 19 close-to-convex function, 77, 85 closed string, 175 coecient of viscosity, 1 complex moment, 22 condenser, 116 conformal anomaly, 181 conformal invariance, 115 conformal radius, 116
kinematic boundary condition, 1 kinematic viscosity, 2 kinetic undercooling regularizatio Koch snowake, 168 Koebe, 9 Koebe theorem, 75 Kolmogorov, 8 Kolmogorov ow, 9 Kufarev, 19, 29, 136
Lwner, ix o Lwner-Kufarev equation, 18, 75 o 136, 137, 157, 158, 161, Lambs method, 13 laminar ow, 6 Laplace, 20 Laplace-Young condition, 20 Laplacian growth model, 24
semi-ow, 151, 154 semigroup, 151 shear stress, 1 Sobolev space, 57, 58 Sokhotski -Plemelj formula, 89, 9 starlike function, 76, 79 Stokes-Leibenzon model, 15 stream function, 7 stress tensor, 5 string, 174 string theory, 173 strong interaction, 171 strong solution, 19, 31, 65, 66 subordination chain, 137, 154 superstrings, 182 Szeg, 114 o Teichm ller, 139 u