title author publisher isbn10 | asin print isbni3 ebook isbn3 Tanguage subject publication date lee ade subject ‘Mathematical Population Dynamics: Proceedings of the Second International ConferenceLecture Notes in Pure and Applied Mathematics ; V. 130 Arino, Ovide 0824784243 9780824784249 9780585333977 coverPage Mathematical Population Dynamics Proceedings of the Second International Conference Edited by Ovide Arino University of Paw Pau, France David B. Axelrod Rutgers University Piscataway, New Jersey Marek Kimmel Rice University Houston, Texas Marcel Dekker, inc. New York® Basel* Hong Kong page iPage i Library of Congress Cataloging-in-Publication Data Mathematical population dynamics: proceedings of the second international conference / edited by Ovide Arino, David E, Axelrod, Marek Kimmel. Pp. em, (Lecture notes in pure and applied mathematics; v, 131) Includes bibliographical references and index. ISBN 0-8247-8424-3 1. Biomathematics Congresses. 2. Molecular biology Mathematics Congresses. 3. Genetics Mathematics Congresses. 4, Epidemiology Mathematics Congresses. I. Arino, Ovide. IL Axelrod, David E. II]. Kimmel, Marek. IV. International Conference on Mathemati (2nd : 1989 : Rutgers University) V. Series. (QH323.5.M366— 1991 574.0151 de20 91-8703, ar I Population Dynamics This book is printed on acid-free paper Copyright © 1991 by MARCEL DEKKER, INC. All Rights Reserved Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage and retrieval system, without permission in writing from the publisher. MARCEL DEKKER, INC. 270 Madison Avenue, New York, New York, 10016 Current printing (last digit) 10987654321 PRINTED IN THE UNITED STATES OF AMERICAPage i PREFACE This volume is intended for mathematicians, statisticians, biologists, and medical researchers who are interested in recent advances in analyzing changes in populations of genes, cells, and tumors: in the natural history of cancer; and in epidemiological topies such as AIDS. The selection of subjects was motivated by recent developments in the mathematical theory of populations and by new data available from various branches of molecular and cellular biology and biomedical sciences. The topics included have been selected from those presented at the Second International Conference on Mathematical Population Dynamics. No attempt has been made to cover all of the traditional topics in population biology. The chapters are grouped within the following topics: structured populations, ordinary and partial differential equations models, AIDS and theory of epidemics, stochastic models, cell eyele kineties, proliferation and tumor growth, and geneties and molecular biology. This volume appears at a time when there are important developments occurring in mathematics and biology. From the mathematical viewpoint, there has been recent progress in models of structured populations. These are models that identify variables such as age, size, amounts of specific constituents, or other characteristics of individuals (genes, cells, or organisms) to reproduce the time evolution of heterogeneity within populations. On the most abstract level, these new models requite analysis of integral, functional, and partial differential equations (frequently achieved by the analysis of semigroups of operators) replacing the traditional ordinary differential equation formalism. There also has been recent progress in stochastic models involving a discrete or continuous variety of types of individuals within a population. In biology and medicine, new data have been collected within the new framework of molecular and cellular biology and through the use of advanced technologies, including computer-aided data storage and retrieval. New data have revealed that heterogeneity is a common feature of populations and is an important aspect of their dynamic existence. page tiiPage iv These chapters resulted from the Second International Conference on Mathematical Population Dynamics held May 17 20, 1989, at Rutgers University, New Brunswick, New Jersey. The Conference was organized by Ovide Arino (University of Pau and University of Mississippi). David E. Axelrod. (Rutgers University), Marek Kimmel (then at Memorial Sloan-Kettering), James R. Thompson (Rice University), and Glenn F. Webb (Vanderbilt University). There were 82 participants from 13 countries. Financial support for the Conference was provided in part by the United States National Science Foundation (DMS-8818652), the Rutgers University Office of Research and Sponsored Programs, and the Waksman Busch Memorial Fund. ‘The following individuals devoted their time and expertise to the demanding and unrewarding task of refereeing papers published in this volume: W. Arendt, . Arino, D. E. Axelrod, R. Bartoszynski, P, Bates, C. Begg, R. Benkhalti, A. Bertuzzi, S. Busenberg, B. Clarkson, K. Cooke, J. Cushing, R. Day. 0. Dickmann, W. G. Fitzgibbon, B. Flehinger, T. Fokas, A. Gandolfi, J. Garner, R. Gatenby, A. Georgiev, A. Grabosch, K. Hadeler, H. Heijmans, M. Hola, B. Holland, J. Keener, M. Kimmel, W. King, A. Koch, T. Kuczek, G. Ladas, M. Langlais, 8. Maggelakis, J. Mahaffy, Ch. Mode, J. Nedelman, A. Olbrot, M. Parrott, D. Pearl, M. Pierre, D. Rigney, E. Sanchez, F. Schultz, R. Sennerstam, J Skierski, H. Smith, P. Smouse, J. So, G. Swan, P. Takac, W. Tan, B. Tang, H. Thieme, J. Thompson, F, Traganos, S. Tucker, J. Tyson, G. Webb, B. Weir, A. White, J. Wu, and M. Zerner. OVIDE ARINO DAVID E. AXELROD MAREK KIMMEL page_ivCONTENTS Preface Contributors Part I Structured Populations 1 Analysis of a Cell Population Model with Unequal Division and Random Transition Ovide Arino, Marek Kimmel, and Martin Zerner 2 Slow Oscillations in a Model of Cell Population Dynamics Ovide Arino and Abdessamad Mortabit 3 Competing Size-Structured Species JM. Cushing 4 Quiescence in Structured Population Dynamics: Applications to Tumor Growth Mats Gyllenberg and Glenn F. Webb 5 Altruistic Population Model With Sex Differences Ying-Hen Hsieh 6 Remarks on an Epidemic Model with Age Structure Michel Langlais a Heterogeneous Population John Radeliffe and Linda Rass Page xi n 63 5‘Ayo Age San Poplin dee wi an And Sc ara Modeig of Cal Pain Dyin App te Say Clie Seg Renew nd Open Ovens plain Mkt Sie Dependent Pls ‘The tt Rap aon nth Dyn ee Eqns Son tt 4M ese Jo Hen Ser Mer Lae Coat of Nolin Sytem th Actions Anas Pega ames aint Ma Pal Ponti ti Epis Compton Maid Gra ‘Doin Mango Nat alls Coton Mie ost Maing of Pople Dynamic wit Sytem Exel wh ve Ap Sho bly Be Pot [IDS he Ty pis ‘Maing Cag Dice Vale Popo Si and Demo pic ttn Soa art aa ars Kemet Croke, Aeon Cao gi Choe:Pee Scr Gap Roy M Andere Medods for Namercly Speci Dyn StchateNodel of AIDS Epon Heese open Sa omg. ‘Dynamic of Callalar AMbaon with Randomly Disb Elec Sc Ma ninth Diy ei oe Ut ‘SIMEST: Teta fr Mode! Agrepnton with Consens of Cn ‘Sho Maing Heer) Cele Aptis a Nope ‘Reglaze Hates of Cl yee Pama opin ec by Lost rm Flow Comer Da esi Ret i ns ereese Sept Coper ease Come by Matalin fhe Crowd Rae ol Manali Stones Pio Dyno: Doan Etc ote Vai of Cel Cle Deion aking Expres oS Call Che Maa “Agent Est of Noam Nae Cosmin aed Noni oe oma ki ‘uutogs Boe ManowTampesinn Fok Sao Anton Hoge Popmiiow of NmalanéTemoreate " ese it41 Effects of Selection and Mutations On the Probabilities of Identity between Genes in Small Populations Michel Gillois 48 Comparison of Two Models for Initiation of Replication in Escherichia Coli Joseph M. Mahaffy and Judith W. Zyskind 49 Quantitative Shift Model and Generation of Single-Cell Heterogeneity Jerry A. Peterson Index page_ix Page RB 747 167 781Contributors John A. Adam Department of Mathematics and Statistics. Old Dominion University, Norfolk, Virginia 23529 Donald A. Allers Department of Mathematics. Pomona College. Claremont California 91711 Roy M. Anderson Department of Biology, Imperial College. University of, London, London SW7 2BB, England Ovide Arino Department of Mathematics. University of Pau, 64000 Pau France Ds E. Axelrod _Waksman Institute, Rutgers University, Piscataway, New Jersey 08855-0759 Robert Bartoszyriski Department of Statistics, The Ohio State University. Columbus, Ohio 43210-1247 Jacques Bélair Department of Mathematics and Statistics, Université de Montréal. Montreal, Quebec, H3C 337 Canada Alessandro Bertuzzi Istituto di Analisi dei Sistemi ed Informatica. Consiglio Nazionale delle Ricerche, 00185 Rome, Italy Stavros N. Busenberg Department of Mathematics, Harvey Mudd College. Claremont, California 91711 Carlos Castillo-Chavez Biometrics Unit, Cornell University. Ithaca, New York 14853-7801 Chao W. Chen U.S. Environmental Protection Agency. Washington. D.C. 20460 Kenneth L. Cooke Department of Mathematics, Pomona College. Claremont California 91711wi Comriburors Stephen Cooper Department of Microbiology and Immunology. University of, Michigan Medical School. Ann Arbor. Michigan 48109-0620 J.M. Cushing Department of Mathematics. Interdiseiplinary Program in Applied Mathematics. University of Arizona, Tueson, Arizona 85721 Neil Dubin Institute of Environmental Medicine. New York University Medical Center, New York. New York 10010 Zdzislaw Duda Department of Automatic Control. Silesian Technical University. 44101 Gliwice. Poland Katherine Bennett Ensor Department of Statistics. Rice University. Houston. Texas 77251-1802 W. E. Fitzgibbon Department of Mathematics. University of Houston. Houston. Texas 77204 Betty J. Flehinger Department of Mathematical Sciences. T. J. Watson Research Center. IBM Research Division. Yorktown Heights. New York 10598, Alberto Gandolfi_ Istituto di Analisi dei Sistemi ed Informatica. Consiglio Nazionale delle Ricerche, 00185 Rome. Italy Michel Gillois Laboratoire de Genetique Cellulaire. Institut Ni Recherche Agronomique. 31326 Castanet-Tolosan, France jonal de la Kerry W. Go Department of Statistics, Rice Universi 77251-1892 Annette Grabosch Lehrstuhl Biomathemati Tabingen, 7400 Tubingen, Germany Houston. Tex: . Biologie II, Universitat Sunetra Gupta Department of Pure and Applied Biology. Imperial College. University of London, London SW7 2BB. England Mats Gyllenberg* Academy of Finland and Department of Mathe Helsinki University of Technology. SF-02150 Espoo. Finland ates Anton Hagenbeek Institute of Applied Radiobiology and Immunology TNO. 2280 HV Rijswijk. The Netherlands Jack K, Hale Center for Dynamical Systems and Nonlinear Studies. Georgia Institute of Technology. Atlanta, Georgia 30332 Henk J. A. M. Heijmans Centre for Mathematics and Computer Science. 1098 SI Amsterdam. The Netherlands Michael L. Higgins Department of Microbiology and Immunology. Temple University School of Medicine. Philadelphia, Pennsylvania 19140 Cartentafiiation: Department of Applied! Mathematics. Lulea Unisessty of Technology. SSIS? Lule, SwedenContributors iit Marek Hola Department of Chemical Pathology. University College and Middlesex School of Medicine, London WIP 6DB. England Bart K. Holland Division of Biostatistics and Epidemiology. Department of Preventive Medicine. University of Medicine and Dentistry of New Jersey — New Jersey Medical School, Newark, New Jersey 07103-2757 Ying-Hen Hsieh Department of Applied Mathem: University. Taichung, Taiwan, Republic of China ies, National Chung-Hsing Hsu Department of Mathematical Sciences, Memphis State University, Memphis. Tennessee 38152 Marek Kimmelt Department of Pathology, Memorial Sloan-Kettering Cancer Center. New York, New York 10021 Jerzy Klamka Department of Automatic Control, Silesian Technical University, 44101 Gliwice, Poland John P. Klein Department of Statisties, The Ohio State Univers Ohio 43210-1247 y. Columbus, Arthur L. Koch Department of Biology. Indiana University, Bloomington, Indiana 47405 ‘Thomas Kuczek Department of Statistics, Purdue University, West Lafayette. Indiana 47907 Michel Langlais U.F-R. Sciences Humaines Appliquées. Université de Bordeaux II, 33076 Bordeaux, France Sjoerd M. Verduyn Lunel Center for Dynamical Systems and Nonlinear Studies, Georgia Institute of Technology. Atlanta, Georgia 30332 Sophia A. Maggelal Dominion Universit ist Department of Mathematics and Statistics. Old . Norfolk. Virginia 23529 Joseph M. Mahaffy Department of Mathematical Sciences. San Diego State University. San Diego, California 92182 Mario Markus | Max-Planck-Institut fr Ernahrungsphysiologie, D-4600 Dortmund 1. Germany Margaret C. Memory§ Department of Mathematics. University of Alabama, Tuscaloosa, Alabama 34587-0350 Current ations: * Anatomy and Human Biology Group. Biomedical Sciences Division, King’s College London, London WC2R 215, England "Department of Statistics. Rice University. Houston, Texas 77251-1802 {College of Science, Department of Mathematis, Rochester Insitute of Technology, Rochester. Now York [4h23-0867 {Division of Science and Mathematics. Misssippi Unversity for Women. Columbus, Misissippi 9701xiv Comiributors Elizabeth Milcos-Livanos* Waksman Institute. Rutgers University, Piscataway. New Jersey 08855-0759 Charles J. Mode Department of Mathematics and Computer Science. Drexel University, Philadelphia. Pennsylvania 19104 J.J. Morgan Department of Mathematics. Texas A&M University. College Station, Texas 77843 Mitchelle S, Morrison Department of Biostatistics. Epidemiology. and System Science, Medical University of South Carolina. Charleston, South Carolina 29425-2503 Abdessamad Mortabit Department of Mathematics. St. Andrews Presbyterian College. Laurinburg. North Carolina 28352 Irwin Nathan U.S. Marketing Group. Xerox Corporation. Stamford, Connecticut 06904-2345 Dennis K. Pearl Department of Statistics, The Ohio State University. Columbus. Ohio 43210-1247 Jerry A. Peterson John Muir Cancer and Aging Research Institute, Walnut Creek, California 945% John Radcliffe School of Mathematical Sciences. Queen Mary and Westfield College. University of London, London E1 4NS. England Linda Rass School of Mathematical Sciences, Queen Mary and Westfield College. University of London, London E1 4NS. England Frank W. Schultz+ Department of Experimental Tumor Therapy. Radiobiological Institute TNO. 2280 HV Rijswijk. The Netherlands Roland Sennerstam Division of Cell Analysis. Department of Pathology. Karolinska Institute and Hospital. 10401 Stockholm. Sweden Janusz S. Skierski? Department of Biophysics and Biomathematics. Medical Center of Postgraduate Education, Warsaw. Poland Hal L. Smith Department of Mathematics, Arizona State University. Tempe. Arizona 85287 Giuseppe Starace Istituto di Medicina Sperimentale, Consiglio Nazionale delle Ricerche 00185, Rome, Italy Current aftiations: “Department of Pathology, Lineberger Cancer Research Ci ‘Chapel Hull, Chapel Hill, Nosth Carolina 7899-7298 *Department of Dosimetry and Carcinogenesis. Invitute of Applied Radiobiology and Invmunohoes TNO, 2280 HV Ris, The Netherlandh {Experimental Cell Rescarch Laboratory. Memorial Sloan-Kettering Cancer Center, Now Yurk, New York 1021 wer, The University of Nowth Canina atContributors wv sity, Houston, Texas David N, Stivers Department of Statistics, Rice Univ 77251-1892 Andrzej Swierniak Department of Automatic Control. Silesian Technical University, 44101 Gliwice. Poland Wai-Vuan Tan Department of Mathematical Sciences. Memphis State University, Memphis. Tennessee 38152 Horst R. Thieme Department of Mathematics, Arizona State University. “Tempe, Arizona 85287-1808 James R. Thompson Department of Statistics, Rice University. Houston, Texas 77251-1892 ‘A. van Rotterdam* Department of Experimental Tumor Therapy. Radiobiological Institute TNO, 2280 HV Rijswijk, The Netherlands Cristobal Vargas Departamento de Matematicas. Centro de Investigacion y Estudios Avanzados, Mexico City 07000, Mexico Neha I. Vibhakar_ Waksman Institute, Rutgers University. Piscataway, New Jersey 08855-0759 Eberhard 0. Voit Department of Biostatistics, Epidemiology. and System Science. Medical University of South Carolina. Charleston, South Carolina 29425-2503, Gerard Wagemaker Institute of Radiobiology, Erasmus University. 3000 DR, Rotterdam, The Netherlands iy. Greenville, S. J. Waggoner Department of Mathematics, Furman Univers South Carolina 29613, Glenn F. Webb Department of Mathematics. Vanderbilt University. Nashville. Tennessee 37235 Jenne J. Wielengat Department of Experimental Hematology. Radiobiological Institute TNO, 2280 HV Rijswijk, The Netherlands Matthew Witten Balcones Research Center, University of Texas System— Center for High Performance Computing, Austin, Texas 78758-4497 Ting Yang Institute of Environmental Medicine, New York University Medical Center, New York, New York 10010 Martin Zerner Department of Mathematics, Université de Nice. 06034 Nice. France Judith W. Zyskind Department of Biology, San Diego State University. San Diego. California 92182 * Current affiliations Department of Medical Informatics, Instiute of Applied Radiobiology and Immunology TNO. 2280 HV Rijsuijk. The Netherlands *+Depariment of Hematology. Dr. Danie! den Hoed Cancer Center, 3008 AE Rotterdam. The Nether: landsStructured Populations1 Analysis of a Cell Population Model with Unequal Division and Random Transition OVIDE ARINO University of Pate, 64000 Pau, France MAREK KIMMEL* Memorial Sloan-Kewering Cancer Cemer. New York, NY 121 MARTIN ZERNER University of Nice. 06034 Nice, France 1. INTRODUCTION In recent years we have witnessed considerable progress in the development of the theoretical tools of population dynamics. One of the most important is the theory of semigroups of positive linear operators. The foundations and basic results of this theory can be found in the book edited by Nagel (1986) and. in a more application-oriented manner. in a volume edited by Metz and Diekmann (1986) The principal use of semigroup theory in such models is to characterize the long- term behavior of dynamical systems. including a description of the inner structure cof the population In this paper we are interested in those aspects of the theory that concern models of dynamics of cell populations of the type introduced in Kimmel et al (4984) and investigated in Arino and Kimmel (1987. 1989) and in Sanchez et al (1989). The philosophy of constructing these models differs from the usual ap- proach based on partial differential equations. (This difference is clear. for exam- ple. from Example 3 in Webb's (1987) basic paper on cell cycle models,] ‘The models that we are interested in describe the regulation mechanisms of the cell cycle, starting from the birth mass of cells, with unequal division of mass between daughter cells being an important factor. The first analysis of a model of this type. published in Arino and Kimmel (1987), was carried out by these authors in 1984, when this paper was submitted, independent of the collection of uniform- ‘Current affiation: Rice University, Houston, TX, 77251-18924 Arino et al. ized results edited by Nagel (1986). Accordingly, this analysis, although also based on semigroup theory. included many details not covered by the uniformized approach. This model has also been quoted in Nagel (1986) as Example C.1V.3.11, unfortunately with improper assumptions. In the present paper we look at the analysis of the model of Kimmel et al (1984) and Arino and Kimmel (1987), slightly generalized for this paper. from the viewpoint of the general theory of positive semigroups. After preliminaries, we Prove that the model gives rise to a semigroup of positive linear bounded oper- ‘ators, and that this semigroup is strongly continuous, eventually compact. and in some sense irreducible. From these properties we prove the result on asymptotic exponential growth of solutions ll, MODEL DERIVATION, ASSUMPTIONS, AND SUPPORT PROPERTY A. Derivation We proceed as in Kimmel et al. (1984). The most important notion of the model is the distribution of cell mass flux at the beginning of the G, phase (the onset of the cell cycle). It is denoted by n(.y) and treated as an unnormed distribution density of the pair (1,») (see the remark below), The interpretation is that n(t,y) dr dy is equal to the number of cells with mass between y and y+ dy that entered G, in the time inteval from ¢ to 1+ dt, The following assumptions define the model 1. Suppose that a mitotic cell just before division has mass x. The density of probability of the daughter cell's mass y, conditional on x. is denoted by fvat). It is necessary that f(y.x) = 0, whenever y> x, and that f(x — 0) = f(v.n) 2. The fate of the daughter cell produced during division, which reenters the cycle with initial mass y, is described in probabilistic terms: a, The time 7 it spends in the cycle is a random variable with conditional distribution density 7(7.y), given y b. The mass x of this cell when it reenters division is a function &(7.¥) of the time it spends in the cycle and of its birth mass ‘These hypotheses are depicted as follows: Growth of Cell Unequal division —| a | s |G | M j-y~fn— ory) ry) ‘The derivation of the model equation is carried out in several successive steps. Let us first suppose that cells spend in the cycle exactly 7 time units and have birth mass equal to & Then the distribution density of the flux at the beg of the next G, phase is equal toCell Population Model Analysis 5 2fly. (7.0) a But 7 is distributed with density 7(-.£) conditional on & and & is distributed with density n(s,€) at time s. Therefore, the distribution density of the pair (7.€) is equal to yr €)mts.6) @ Consequently, the joint density f(s + 7, y:7,€) (Contribution to the flux density through next G, of cells of size & at their birth which spend time 7 in the cycle and which were born at time s with size y) is equal to the product of (1) and (2). After a change of variables (s,y:7.€) into (t,y:7,€), where = 5 + + (the Jacobian is equal to 1), we obtain Ale yer 8) = 2fly (Sly rE Im(e~ 2,8) 8) Continuity of the flow requires that n(t.y) = f J A(ty:7.€) de dé. Integrating (3) provides the equation of the model, (ey) = af [fy eeireemne- 1.€)d6 dr @ Remark. The derivation outlined above includes intuitive manipulations on intor- mal unnormed densities. These manipulations can be formalized if n(-) and a(-) are treated as densities of the expectations of counting measures of a branching process describing our model [as in Kimmel (1983)] ‘The expression for the total number N (1) of cells present at time ¢ is derived in the following way. The density of cell flux through Gi, including cells born with size y at time s which spend time + in the cycle, is equal to n(s.y)y(7,»). The population at time r includes cells born between 1 ~ 7 and ¢: woe Lf [_aeotesreers . 7 i [oconte- s.y)dsdy where Tis the til of the distribution of cell cycle length fie. Puy) = fi raz) B. Assumptions We proceed to specifying the basic hypotheses on functions f, y, and, which formalize the requirements of cell cycle dynamics, (Hy) fE Lic(RE)s £205 [5.01.0 dy = 1s fOr = ya) = fv.) f(y-2) is non negative and there exists d; € (0.4) such that f(x) is positive if and only if y € (d,x, dx), where dz = 1~ dy (Hy) E Lic(R2)s fo y(t.x) dr = 1; (ty) is nonnegative and there exist two. continuous decreasing functions 7, and rz such that lime. 7,(¢) >0: 7 <7», and y(r,€) is positive if and only if + © (74(£), 74(€)) (He) 6 © Go R2); 620; (-,€) and 4(1,-) is increasing. ‘The assumptions on f express the fact that f(-,x) is the density of the con-6 Arino etal. itional distribution of the mass of daughter cell provided that the mass of the mother cell is x. The support property reflects the fact that the mass partition to daughter cells may not exceed a maximum degree of inequality The assumptiony ony express the fact that 9(-.€) is the density of the con- ditional distribution of the cell eycle duration given the birth mass of the cell & ‘The support property takes into account the following requirements; (1) cell cycle time varies only’ in certain limits: (2) it should be in inverse relationship to the birth mass: and (3) a minimum cell eycle time is required even for cells with large birth mass. ‘The assumptions on ¢ express the fact that the mass at division of the cell is larger for cells with higher birth mass and cells that stay longer in the cycle. ‘The assumptions that f and ¥ are positive on the entire corresponding intervals are technical. They are needed for the proof of Lemma 5. We see that. strictly speaking. fand y are classes of functions. Due attention will be given to this point at the only place where it is needed, which is again the proof of Lemma 5. C. Support Properties It is reasonable from the biological viewpoint to require that the support with respect tothe variable y of the solution nf.) starting from any nonnegative initial data, be confined to a bounded interval J, separated from 0. The interpretation is that the regulation mechanisms of the cell cycle eliminate ces that are very small or very large ‘Generally, to formally construct a solution of equation (4), after time fit is necessary to know it om the set {0.9} :4y~ #\(V) > 42 ty = Faly)oy > O}. Along the solution n. the restriction of nm to 4, = (sv) 25 (t= Ty). f= THY). V > OP comprises the data necessary and sufficient to continue the solution, We will adopt the standard notation (Hale. 1977) nds) = nit s.yi 1200 (SQVEM B= (Gs.0) 9 >0.5€ (709.0) We prove the existence and some properties of the solution in the next section. The fact that n(t.) has a bounded support contained in 1 = (Ay.Aa} implies that the system (4) has maximum and minimum delays 8; and 6. = TAD = TAD) ” Therefore. the initial data can be restricted to 4 and the solution constructed in steps of length 8), Based on hypotheses (H,). (H,). and (Hy). the following implication is true’ supp mols) C= [AAs] DWE [0.6]. suppn(e.) CHEV MET KOLEET (8) (6) where OG) = ddlr (EE i Functions ¢, are not. in general, monotone, which may complicate the analysis of the support properties. We will restrict ourselves to the case corresponding to fone possible variant of cell cycle regulation (see Figure 1) EER, oeCelt Population Model Analysis 7 a a 1 2 Figure 1 Support hypotheses for the model (H*) ; is increasing: $(€) > §. € as where 0 < ay < ay <* are constants We will summarize certain properties of functions 4; in the following lemma {stated without proof) Lemma 1. Suppose that (H*) is satisfied. Let us define two sequences (£,} and {Eu} such that = €o = €ER, and £,.1= b1(&.). Eu-1 = do(G,). Then & a. Ey > ay.asn + % Moreover, byt > Gin€ >> bro < bk <> Eye > En > 2 > Enns < Eu ‘Application of Lemma 1 and implication (8) allow us to state the basic support property. The proof, which is elementary, is omitted, Lemma 2. Let us choose = [A,.Az] such that 0< A, Sa, 5 it 1S Az <%, Then supp rio(s-)C1= [Aras s€ [82.0] then supp n(t) C[4(A). bLAN]CL > 0 Therefore, supp n(t:) is asymptotically contained in [ad] ll SEMIGROUP AND ASYMPTOTIC BEHAVIOR As mentioned before, the solution n(t,») of equation (4) can be uniquely extended by steps of length 6; starting from initial data on A= {(s):. 5€ (=72(9).0).y 1}. Lemma 2 assures that the support of solutions does not leave the strip R. J. We will show that the solution exists in the L" sense. Indeed, integrating (1), we obtain8 Arino et al [ finemararsa s2 a i wr blind 7-6) dE de dr =2ff _lnuts.€)] as de (In the integrations above, the initial data ny are treated as defined on the rectangle {~6:,0] x J. and equal to zero outside A.) The results can be restated as Urtleo.ery <0 = 2a ar (10) ‘We will accept L'(A) as the basic space and call it X. Equation (10) yields Indie = Srl. (0.8) ay Lemma 3. Suppose that hypotheses (H,). (H,). (Hy). and (H*) are satisfied. If ‘ny = 0 belongs to X= L'(A), then there exists @ function n: Q— 2, where = AUB. x 1.20. & Lh(Q): which verifies (11) almost everywhere in 9: and lim, .o1,= ny in X. The solution is unique in the sense of an equivalence class in Li(O). Proof. Existence of solutions is obtained by iteration of the a priori estimate (11). Uniqueness and nonnegativity are obvious. Continuity of mat 1= 00" is implied by continuity of translations in L! Corollary 1. The family of mappings {G(1).1 = 0}. GW): XB nym EX (2 is a strongly continuous semigroup of positive bounded linear operatory on X ‘The next result requires additional hypotheses regarding the conditional distri butions f and (Hy) fe LiF). (Hy) y € LilRE). Lemma 4. Under the hypotheses of Lemma 3 supplemented by (H) and (H5). Gla) is compact from X into X for any > 24s Proof. Under (H) and (H;). mya.y-1 € Li( (0) * 1). Its then enough to ‘observe that the map K: L°((0.8) * 1) —> L'(B2- 82 + 8) % 1), defined by ney) = (Kny(ey) =2 [flr eu- nee Eon .8) d6 dt =a" [et is compact as being defined by the integrable kernel g. Indeed. g can be approxi- mated in the norm of L' by continuous functions with compact support. The E)nl7.€) de de a3)Cell Population Model Analysis 9 Remark. This lemma can also be proved with (Hj) and (Hj) replaced by the weaker hypothesis that the operator K in the proof above is weakly compact. It is, however, difficult to find weaker assumptions on f and + implying the weak compactness of K. We will denote A the generator of G, (A) its domain. We now define Xy (respectively, Xa) as the set of functions in X that vanish when the y component of the variable is outside (respectively. inside) the interval [ay.a2). X is the direct sum of X; and Xz, and X; (not X2) is stable under G (Lemma 2). The definition of an irreducible semigroup can be found in Nagel (1986, Definition C-I11 3.1) Let us notice that Gx, is a semigroup with generator Avjeaynys Lemma 5. The restriction of G to X; is irreducible Remark. From Lemma 2 it follows that (G()} is not irreducible if 1 = (A143) with Ay 0) Hep >), such that inf Glo). u) > 0 aay rn denotes the solution with initial data my such that Glan .8) = nr.) = nle+ 7.6) For any (7,£) with € € (a,.a2) and +> ~7,(¢), we define the set So(t.€) = ((s.y) 5 (7 + 11(E), 7 + t2lE)), YE (dis — 7, €)s dads — 7. EV} and notice some of its properties. It is an open connected set. If (r.€) i a point of density of the set {(s.x): n(s.x) > 0} (we will write, for brevity, a point of density of n), then n is positive on So(r,€). The boundary of So(r.£) contains the points (+ + 7)(£). 4y(€)) and (7 + 1316), da(€)), 80 Sy(r.€) co are joining these two points (endpoints excluded). As a consequence, for every y in (di(€), b2(€)). there is a point (s,y) in Sa(t,€). Finally, So(7.€) contains points (6.») with s positive Let (o.9) and (to.€o) be points of density of uw and no, respectively. We first take some point (11,1) © So(to.Eo) With 7; positive. We then define a sequence of sets and two sequences of numbers in the following way’ Sia (EVE Seo = Ue Solr.) ees = di(as) Bi= 61, Bess = (Bs) ‘nis positive on S,, which is open for k > 1, and for every y € (ag,Bs) there is an s such that (5,9) € Ss For k large enough, 7 belongs to (as,B,) (Lemma 1). 50 that there is a number 4 such that m is positive on a neighborhood of (s,m). Let us set f= 5 ~ 0. which implies that G(s)n is positive on a neighborhood of (a.m) as na(orn) = n(s,n). It can be checked that fois postive, As (on) is a point of density of u, u is positive (on a subset of positive measure of that neighborhood, which in turn implies (14)ww Arino et a. We may now state the main result of this section. Theorem 1. Suppose that hypotheses (H,). (Hj). (H,). (Hs). (Hla). and (H*) are satisfied. Then. for any initial data n, > (0. the semigroup exhibits asymptotic exponential growth. that is. GU) = Cae" n* + ofe*) «sy where C,,>0 is a constant depending on the initial data, A*>0. e*E L! is nonnegative with support [a:.a3). and (A*.4*) satisfy the equation mor 2f i fly. 6r.O)ly(r.g)e “MEE dr, yE (a.a) (1s) with A* > Re A. for all other pairs (A,u) satisfying (16); and A*, yy) are the only solutions of (16) to within a constant multiple of +. Proof. We want to check the hypotheses of the following proposition. where (A) is the spectral bound of s(A) = sup(Re A: in the spectrum of A) Let Tbe an eventually compact semigroup and A its generator. Suppose that ois the only eigenvalue of A with real part s(A) and it is simple. Let us call v and v* the associated eigenvectors of A and A’ [where AT is the generator of the -migroup adjoint to T: see Nagel (1986, Chap. A.1.3.4)]. Then there is a number wo" ys and fic’ fle. yd = 1. The sup: port of f(-.») is the interval [yd. »(1 ~ d)] for some given d (0, 3) (Hi) 6 is nonnegative function of class CJ, +2) such that (0) = 0. 6'(2) > di. for all > 0, and for some positive constant «hi, Moreover. i we denote by dy(5) = d(z). dls) = (1~ dds). there are two numbers ay. ds. DS ay = (resp..< =) for sa (resp.. > 4,). for i= 1.2 (Ha) iva positive function of class C1(|0. + 1) with We) =O and Wits) 0 for s>0, (H,) «is a continuously differentiable function on [0.+%) with a(0)= 1 =) = 0. and 02) = (for all = = 0. (1) AE. It was shown in Arino and Kimmel (1989) that ayymptotically the support in the Structure variable of the solutions is contained in the interval 1 = fa\-as]- andOscillations in Cell Population Dynamics 1s also the property for a function to have its support C [a). a2] is invariant by the solution operator. These facts are still valid in the context of the present chapter. For simplicity. we restrict ourselves to such functions. and thus we will work in the following state space: L"{(— da.),0) x J]. As usual. we will denote by n, the function defined by (7.x) = n(t + 7.x). The asymptotic behavior of nonnegative solutions is as follows: m,—N(=)ra(t). as ¢— + %, where va(x) is the unique solution of the fixed-point problem v(x) = J v(2)flx. &2))dz. normed by Si volz)iz2) dz = 1, and N (=) = 0 for a in (0,3) and N(=) = NY 0 "(1/2A) for ‘in (, 1]. That is, asymptotic extinction of the cell population is expected if the probability of corvect divisions is less than 4, while a stabilized steady state is reached if A is greater than 3. The modified model we consider here provides an example where slow oscil- lations are forced into existence by the introduction of a delay term. Let us write the equation and state the hypotheses needed throughout this paper. The Equation n(t.x) = Dao |N(t— i fos etenyne = W(z).2) dz w wo [fanaa a Hypotheses. Essentially. they are the assumptions introduced above. some of them being reinforced. This will be indicated by a prime added to the old notation, (H,) In addition to (H,). itis assumed that &,(a.) < (a1), (H,)' In addition to (H,). it is assumed that F= Way) andr that 0-<7<2r and 4i By differentiating both sides of (2). and using (1) and the fact that Life. (2) dx = 1, (1)-Q) yields NU) = PA @(N(E~ 1) ~ NAC) 3 where k(t) = Sin(t~ Wz).2)dz. This equation suggests the analysis of (0) glx(— 1)) k(t) in the general setting of nonautonomous functional differential equations verifying conditions motivated directly by those of the model az) are such Il, SLOWLY OSCILLATING SOLUTIONS FOR A NONAUTONOMOUS FUNCTIONAL DIFFERENTIAL EQUATION ‘The aim of this section is to study the oscillatory properties of the solutions of the equation (0 = goat = 1) KC) « ‘The autonomous case 2 = gextt =D) 6) hhas been studied extensively for the last three decades. [See notably Hale (1977)16 Arino and Mortabit ‘or Nussbaum (1978) as survey papers.] For instance. it is well known that if \g'(0)] > 7/2, equation (5) has a nontrivial slowly periodic solution, The proof of this result uses an asymptotic fixed-point theorem due to Browder (1965). How- ever, the nonautonomous case has not yet been studied. Even the existence of slowly oscillating solutions has not been investigated for equations of type (4). In this section it is our intention to show that under conditions on g similar to those in the autonomous case. there are slowly oscillating solutions with sustained oscillations, That is, sufficient conditions on g and k will be provided. ensuring the ejectivity of the trivial solution Hypotheses (H,)_ gis in C(— », +), bounded, and such that xg(x) <0. for x # 0. (H.)_ kis positive continuous function on (0. +%). There exist positive constants, Ky and Kz such that Ky 5 Ji°" k(s) ds = Ks for some rin (0.1) with [1/r] =4 ((-] denoting the integer part). (H,) and (H) will be assumed throughout this section. Let C be the Banach space of continuous fu 1.0}. For each @ in C, there is a unique continuous function x(1, 6) on [~ 1, +) satisfying (4) and such that x, , = @ The solutions depend continuously upon the initial data in C, That s, the nonlinear evolution operator U(r, v)@ = x,, where x(s) = x(¢+ 8) for s in (~ 1.0]. (3) = 3) for s in [= 1,0]. and ¢= ¥ 20 [T()8 = Ul.0)8 = x). is continuous from C into C. Remark 1. If 6 is in C and T(1)@=0 for some +> 0. then @ ‘That is. the solution x(¢, @) may not take the value 0 and remain 0 for a time interval equal to I. unless 9 = 0, which in turn implies that x(¢ ) = 0 for all 12 ~ 1. This is not difficult to conclude from the conditions on g and k Let P be the subset of C consisting of all @ such that (—1) = and 4 is nondecreasing on {~ 1.0] Remark 2. For 8 in P\(0}, by (H,) and (Ha). the corresponding solution x(t. 8) starts decreasing for 1 and keeps doing so unless it erosses the axis. Say at f (frst eross-point), in which ease it will attain its frst local minimum at f+ Land starts increasing. It keeps doing so unless it crosses the F-axis. say at 1. (second cross-point). in which ease it will atain its local maximum at f+ and starts decreasing: and so on, Lemma 1. Suppose that (H,) and (Hy) are satisfied. Let @ be given in P and (0) be the corresponding solution. If for some T> 0, we have x(t. 8) > 0 [resp. (06) < 0] forall ¢= 7. then x(t, 8) —»0.as ¢—» + Proof: When there is no risk of confusion, we will write x(1) instead of X(t). First, we note the fact that by virtue of (F,). we have [Moda oe (6) Second, if x(1)>0 for all r= 7, then, by Remark 2, x(0) is decreasing on (T+ 1, +%). Therefore, (1) ~+x(*) = 0. On the other hand. integrating (4) on (T+ 1), we have f Igix(e — I) lo) dt = A(T 4 1) = xl) <Oscillations in Celt Population Dynamics 7 By (6) it may be concluded that g(x(~)) = 0: that is, (=) = 0. The other situation where x(1) remains negative after T is handled in a similar fashion. This ends the proof. Definition 1. & solution x(¢, 6) of (4) is said to be slowly oscillating if there exists ‘an increasing sequence of positive real numbers (referred to as the the cross-point sequence) (f)j=1. With 1 ~1;>1 such that x(j, 6) = 0, and x(t, 8) is decreasing on (ty; + 1etayri + 1) and increasing on (aje1+ 1s fa2 + 1) for all j= 1 Theorem 1. Suppose that (H,) and (H,) are satisfied and let 6 be given in P\(0). If the condition le" Ki >1 ce) holds, then the solution x(¢, 6) is slowly oscillating. Moreover. if (jt is the eross point sequence. then x(4) #0 and x(¢,+ 1) = 0 for all j= 1 Proof: (i) Suppose that after some time T>0..x(t) remains on one side of the taxis, say the upper one (the other situation is treated in a similar fashion): that is, x(0)>0 for all r=. Then, by Lemma 1, x(2) +0 as 1+. Now. integrating (4) on (t, +%) for some > T+ 1, we have xt) f Ig(ats ~ Iyiae ds = f Ig(x(s ~ D)|k(s) ds =latol f KG) ds = Kile) Dividing both sides by |g(x(1))| and letting r go to +2, we obtain |g'(O)IKy = 1. which contradicts (2,). Therefore, x(t) crosses the axis, say at > 0. In view of Remark 2, this defines a sequence (1),=1 such that x(t) = 0 and x(0) is nondecreas ing on (ta): + 1, tzj+2 + 1) and noninereasing on (t3, + 1s ta)-1 + 1) for each j= 1 (ii) Suppose that for some j= 1, x(4) = 0. By (4) and the positivity of k, we ‘obtain that xi, ~ 1)=0 and therefore, x(0) = 0, for all ¢ in (G,-1.4,~ 1), by the monotonicity of x. This. in turn, implies, by similar arguments. that x(1) = x1 #1) for all rin (i. + 1.4). In particular, we would have x(1) = x(j-1 + 1) #0, which yields a contradiction. (Gli) Suppose that #2) = 0 for some + in (t (4) and the positivity of & imply that x(r~ 1) = 0. If t= 14-1 +1, then x(s) = O for alsin (1,1 ~ 1). By (4) and the positivity of &, this implies that x(s +1) = x(i-1 + 1) for s in the same interval. In particular, we have x()) = x(G,-1 + 1) # 0, a contradiction. If r= 1 >4)-1 +1, then by definition of the sequence t,, we have = 1~ 1. This is impossible since r€ (j,,+ 1). Using the same arguments, we have that ¥(0) # 0 for all rin (i, +1, j-:) for all = I. This ends the proof. Remark 5. Given @ in P\(O}, let x(¢, 6) be the corresponding solution of (4) and ({) be the corresponding cross-point sequence. Integration of (4) on intervals of the form (ty,1,ta;+1 + 1) yields the following estimate: 1) +1) for some j= 1, Equation ay +1.9=[4 Kygx(te, + 1.8) a where [-] is the integer part. Similarly, on intervals of the form (I3j.fa) + 1). we have1s Arino and Moriabir wo onel X(lay-1 + 1,6) 8) We have constructed. in this manner. two sequences of estimates of the relative maxima and minima of x(1). which we denote by £"” and x“. defined as follows: evan. 2 [Yat af aa” 1 ate fa (| Kg) Tris clear that <0 8, then #"" is decreasing to £ (resp.. increasing to x). If 40) <., then (resp... x'"") is increasing to & (resp., decreasing to x). Finally. if 40) =<. then "and x" are constant, equal to and x, respectively Proof: The prot may cay be deed fom ihe dacsion aboveOscillations in Cell Population Dynamics 9 The last part of this paper is devoted to establishing that the oscillations obtained so far are sustained. Let us define the operators B, from P into P by Bia) = Ty +0 where 7(s) is the semigroup solution. Lemma 2. For all j= 1, B, is continuous at the origin Proof: We proceed by induction on j. Let j= 1 (8,A)1= nt + 1. 0)|= J Iette 1.0] Aco a For @ close enough to 0, we have g(x(t~ 1, 6) ~ g'(O(~ 1.8) for all ¢ in (4. + 1). On the other hand, since x is decreasing on (0, 1,). we have 2 , rma tel fw 1 01K dr= xs? lL -t ‘Therefore, By is continuous at 0. Now suppose thatthe property holds for j [Bier A= bles + Doel [ [geste ~ 1. 091k a By the induction assumption, if @ is close enough to 0. so is x(t ~ 1, 6) for rin (ieihe1 +1), So the same arguments as before apply. to obtain ihe [i] “This ends the proot Lemma 3. Suppose that the hypotheses (11) and (Hy) ae satisfied, and more ver assume tha the following Inequalities hol 1B,.1 Ks, 8)| = lye. Ae x, 1. €>0, such that for any 8 in PAO} for which |7 (08h. = €, for all = 0, we have [B14 ~ = ¥1B,0(~ 1) (10) Proof: For the convenience of the proof, we introduce two numbers a and B. 2x()+2)20. Consider ¢ in the interval (5, +.2.4, +2 +r) where xis decreasing, ¢~ 1, then, lies in (+ 1. + 17) where xis also decreasing, and therefore xe Dal +r) a2) Now integrating (4) on (i, + 2, 4, +2 +). using (12) and (9). we obtain My +2) =x 424 NZ arly tlt), (3) This implies that 4, €(y+2.4+2+ 7). Indeed, otherwise. we have x(1, +2) 2 ax(,+ 1+), which by the facts that a> 2, [Vr] = B+ 2. that x is decreasing, yields a contradiction, Hence x(t, +2 +r) <0. Furthermore. a +24 =U + 24a +2) +x +2) bey +24 r)) = ay +2) ~ x +24) — aly +2) Zar tl+n-xy tle using (13) and the fact that x is decreasing. 2 (a- Darien) fa Det, #2-) = My +2] since x is decreasing. = a(a~ 1)/B,4~r)| using (11). Therefore, again by monotoni [Buu A A= ala 1)/B,4-r)] That is, (10) holds in this ease (ii) Suppose that (4) +2 —r) > 0>x(4,+ 2): that is, fy is in the interval (+2 rst, +2). From (11) it may be concluded that either of x, we have Mys2=n $18.a- 0 or sy +2) = S18, n) (1) Suppose that x(t, +2— 1) =" |B 9) (ty V= 5 1B A n) «dyOscillations in Cell Population Dynamics 2 and consider ¢ in (tj .01 + 7). We have that X(¢= 1) min(]B,A— Ahexty + 1+) However, since x is decreasing on (+ 1,;+ 2), we know that XG +14 nea +2—7) and then, by (14), we obtain x=) = |B, (=) as) Now, integrating (4) on (j.1, hei +7). and using (15) and (9), we have [eG.ci + 1 = alB, A= 1). Since x is decreasing on (21 + rs tei + 1a We get 1B,.. &- r)| = @|B, (7) (2) Suppose that x(t, + 2) = (a/2)]B, (~ |. Then, by the fact that.x is decreas- ing, (10) is trivial (ii) The last case to deal with is when t.)€(t,+1 t 124,41 +r then, by the monotonicity of x, we have = ne hy +2-- x42) 2 a|B A 1] using (11). Therefore, (10) holds in this ease Suppose that tj. <4 + 1+ rand let ein (1+ 1,4 + 1+ that is, ¢ Lies in the interval (4+ F) where x is increasing, so t= Dl sagen (16) Integrating (4) on (1, +1. 4,41 +7), and using (9) and (16), we obtain 1 a+ N=2hG +140) (7) 7 ) Now, on cach of the intervals (+14 ir, 4 +1+@+ 1). (un -2, “ris decreasing and for rin each of them, we have XO= DEG +P) 18) Integrating (4) on these intervals successively, and using (18) and (9), we obtain x(t, + 1+ in) ~ x(t +14 + Dr = |B,A- 9) «mes 1 Summing on i, and using the fact that (I/r] - 2= B, we have My tlt =x +2- = alBA-P)] And a fortiori, since x is decreasing, (10) holds. Hence (10) holds for all j= 1. For j=0. the conclusion is trivial if (~ 1) = 0. Otherwise, the foregoing proof applies. This ends the proof. The next theorem provides sufficient conditions for the oscillations to be sustained, that is, conditions on g and k ensuring that 0 is an ejective Fixed point of B, for all j= 122 Arino and Mortabit Theorem 2. We suppose that the same hypotheses as in Lemma 3 are satisfied ‘Then (is an ejective fixed point of B, for all = 1 Proof: Let j= 1. Suppose that the statement above is false. ‘That is. for every € > 0, there exists = te) in P with [4 esuch that for all positive integers 4. [B40] = € (or. |B,,6] = €). By the continuity of the operators By at (for all = I. expressed in Lemma 2, it may be concluded that for every €>0. there exists #= te) in P with || =e, such that for all q = j, we have |B, = €. We choose € small enough so that (10) applies and yields a contradiction, This ends the proof. ll, APPLICATION TO THE MODEL EQUATION Let us recall the equations wore [fh mteindedu (20) State Space. LQ), where 1= (~ 1=7.0) 1 Equations (19)-(20) lead to the following delay differential equation NU) = Raat = 1) ~ HK en where wu) = [me we (22 Results of existence and uniqueness of solutions are provided in Arino and Kimmel (1989) and Arino and Mortabit (submitted). We shall be concerned with the asymptotic behavior only. For A in (0.4). it fs proved in Arino and Mortabit {submitied) that m0 as 1» + =. This may be interpreted as expected extine tion of the cell population if the probability of correct divisions is les than 3. From now on, we assume A in (>. 1). Let x(n) = N()~ NY for 12 0: g(<1 [2Ao(z + N%)— 1]. NY has been introduced in Section I. In view of these no- tations, (21) may be written as. X= alee IDK) where &(1) is given by (22) To apply the results of Section II, we should check if assumptions (#7,) and (Hy) hold. This involves a few technicalities: it has been done in detail in ‘Arino and Mortabit (submitted) and we will not repeat the computations here. We will ‘mainly indicate the steps to follow and then show on an example that the conditions wwe collect in our final theorem can really be met. At this point it may be appropri- ate to discuss the connection between oscillatory properties of 1 and those at N Consider the operator 4: L'(0) —C. defined by oman= ff ndeandean frreOscillations in Cell Population Dynamics 2 Let # denote the subset of L'(2) of all m=0 such that Zn, ~ NXE P. where P isasin Section II. The oscillatory N solutions we are interested in are characterized by the property that N enters the set P at successive times 1,= (No) --1 ~ According to the notation introduced in Section II, we have N,.) = B,Ny. This property can be expressed in terms of only, by introducing the operators AlN) =. where 7, = 1 Ato) + 1 2) We will say that n(x) is oscillating (resp... slowly oscillating) if N (0) is oscillating (resp... slowly oscillating). We can then conclude from the considerations above that slowly oscillating 1 solutions are characterized by the fact that mo € domain of «9, for all j. and sustained oscillating solutions are those for which 1 (26) N and N are the fixed points other than N’ of the function p,°p,. where py is defined as 1 pate) = [aot NE (me [f ) (27) Equations (24) and (25) lead to weno Koya and [a= vo= which, using the fact that [by (Hf,)'7 < 2r. yields w= [“usyars2n,M Arino and Mortabit that is. the integral estimates required in (H1,) with KiaN. K=2N, To obtain positivity of & for r large enough, let us first state the additional assumption that we need on f. Let hy and b; be numbers such that y(a) < by < bs < bs(ay) [yee (H)"]. and denote » = b2). = MB). In view of (Ho) we have res5, We assume now that there exists 6 > (0 such that fi. Lemma 4. [Lemma 4.15 in Arino and Mortabit (submitted)}. Suppose that (28) holds. Then k(t) > 0 for 1=F + 2s. Keeping in mind the notations introduced in this section. we are now in a position to translate in terms of the functions and parameters defining the model the conditions stated in Theorem I and Lemma 3 for sustained oscillation. Theorem 3. Suppose that (H,). (Ha)'. (Ha)'. (Hy), and (Hy) hold, and more (a) 2A {0 (NS) [NED 1 [that is, conc-vion (26)f: then, for each my in? such that (Ano 0) + N72 = N. the corresponding solutions of system (19)-(20) is slowly owillating about the steady-state solution V {¥4(1). (b) If. in addition to the above. we assume 25 torn El cE (bib) (28) any t<2ajorwa)0. A typical example that satisfies (29) is 1 for 0S x= eh) =} N= AN = NT for Nos My 0 forv> Ny It is a simplified example in the sense that it is a piecewise linear function. For that reason it lacks the regularity condition that we assumed on ¢. But the results ‘we will obtain can be extended to functions @ close to oy in the C" topology and satisfying the condition (H,). Claim. Let ory be defined by (30) with No. Ny. and m chosen so thatOscillations in Cell Population Dynamics 2 Aue MS and m>6 BD ‘Then there exists €, 0 <¢=$, so that (29) holds for all A. $< A< $+. To prove the claim, we have first to compute N2. p, and then N. N. The equation for Nis 2Aag(N%) = 1. which for 2A > 1 indicates that N'S€ [Ny Nib So lovs(NV2)] = (Ny ~ Na)". This implies that (29) is reduced to 2N7 8 < 2A(N, — Nu) < 28) '[en = 2) (32) ind N are the lower and upper solutions of the equations p, ps(x) =x. where is given by (27) with o = a». Te turns out that p,° p(x) = py({2AI" Nt), for x [0. No]. and in fact this, value is just Vif it is less than Np. We will prove that this is the case. at least for A close to 3. Let the function ¢ be defined by (x) = p,([2A]"N 1). We have ¢( 3 fo and (3) = (IN, ((m + 1)Ny~ Ni)(Qn ~ IN + Nis $0 that 2G) <0 if and only if (m + 1) Nu > Ni. With m= 4, this last condition is by the first inequality in (31). So N= p, ([2A]"-N 2). which implies that "NE. that is, (29) is reduced to faa ({2A]"N A) ' < 2ACNs = Nu) § < 2[oa((2A"N 3) ‘(r= 2) Pm = lea} It is enough to look at these inequalities at A= $. Then we have AN < (Ny ~ Na" < NHN = 2) which will be verified if 2N, <3Ny and Ny <(m ~2)Ny. This in turn implies that m > 6. Conversely, if m6. then Ny and Ny can be chosen in such a way that the condition (31) can be verified. Finally, we may notice that m (as) faccording 10 (H,)'] ean be assumed arbitrarily small just by assuming v accordingly small. independent of the other functions defining the model. Also, the fact that a(x) = 1 for x = Ny fis the idea that limitations in the cell proliferation process take place only if the rnumber of cells exceeds some threshold REFERENCES Asino, ., and Kimmel, M. (1987). Asymptotic Analysis of a Cell Cycle Model Based on. Unequal Division, SIAM J. Appl, Math, 47: 128-145 Azino, O.,. and Kimmel, M. (1989). Asymptotic Behavior of a Nonlinear Functional- Integral Equation of Cell Kinetics with Unequal Division. J. Marh, Biol. 27- 341-384 Arino, O., and Mortabit, A, (submitted), A Periodicity Result for a Nonlinear Functional Integral Equation, Browder. F. E, (1965). A Further Generalization of the Schauder Fixed Point Theorem. Duke Math. J. 32: 575-578, Hale, J. (1977). Theory of Functional Differential Equations. Springer-Verlag. Berlin. Nussbaum, R, (1978), Periodic Solutions of Nonlinear Autonomous Functional Differential Equations. Lecture Notes in Mathematics 730, Springer-Verlag, Berlin3 Competing Size-Structured Species J. M. CUSHING University of Arizona, Tucson, AZ 857: 1. INTRODUCTION Body size is one of the most important attributes of an individual organism. It is significant in determining an organism's energetic requirements and ability to exploit resources for growth and reproduction and its interaction with its physical and biological environment, including predators, prey, and competitors. Despite these obvious and recognized facts, relatively litle mathematical theory of size- structured population interactions exists and virtually no dynamical models of competing size-structured species can be found in the literature. For a discussion of the importance of size structure in population dynamics and ecological interactions, particularly competitive interactions. see Werner and Gilliam (1984). Zooplankton communities provide one important example in which size struc- ture has been of primary significance in the study of multispecies interactions. The observation that species of zooplankton tend to occur in associations characterized by body size and that large size species tend to predominate in zooplankton communities, atleast in the absence of planktivorous fish, led Brooks and Dodson (1965) to propose the size efficiency hypothesis (SEH). The two basic elements of this hypothesis are that large zooplankton species are more efficient at exploiting resources, which provides the potential for the competitive exclusion of smaller species, and that size-selective predation by large-bodied (vertebrate) predators, Which falls more heavily on the larger zooplankton species, can allow for the survival of smaller species or even in some cases result in the elimination of larger species. This hypothesis has been the main theoretical framework of much zooplankton research since its formulation, and many experimental studies have attempted to test its assumptions and verify its implications (see Hall et al.. 1976). ‘Although the principle that predator-mediated competition plays an important role in shaping zooplankton community structure seems to be widely accepted, attempts to verify the assumption that larger species are more “efficient” competi= tors have been equivocal. It appears that this notion is more complicated than first a728 Cushing thought. Furthermore, other factors, such as invertebrate predation and “juvenile bottlenecks.” have led to modifications of the original SEH (Dodson. 1974): Hall ct al.. 1976: Neill, 1975: Lynch, 1978; DeMott and Kerfoot. 1982; Gerritsen. 1984), ‘The purpose of this paper is to derive and analyze some competition models for the dynamics of n size-structured species competing for a single limiting re- source and to see what implications they have with regard to the relationship between body size and competitive success. In Section II a discrete model is proposed and its dynamics studied. This model is for an “interference” competitive interaction for a single limiting resource that is supplied at a constant rate. In Section IIIa continuous model for an “exploitative” competition for a single limiting and dynamically varying resource is studied. The model is based on the Single-species model of Diekmann et al. (1984). Both models are built on the basic assumption that resource uptake rates are proportional to body surface area. They are therefore particularly relevant to many simple (invertebrate) animals such ‘made up zooplankton communities (e.g. fiter-feeding species of Daphnia). The continuous model is more general in that it includes several features neglected in the discrete model: namely. metabolic demands. model-determined maximal adult size, and resource dynamies. The diserete model. however. is mathematically more tractable ‘One major point that emerges from the analyses is that one must be very careful about relating crucial body size measurements (e-g.. adult size. juvenile size. or size at birth) to competitive success. Among species that are very similar in all other regards, a large body size is an advantage. However. there are many other factors that can compensate for a competitive disadvantage due to small body size ll, DISCRETE SIZE-STRUCTURED COMPETITION MODEL We begin with the description of a general discrete model for the dynamics of a structured population whose individual members are categorized by means of and are allowed to move between a finite number m = I of specified classes. Let s,(1) denote the number or density of individuals in class /, 1= (Sm, at times ¢ 1,2... and let x(0) = col(x,()%) be the column vector of these densities ‘The transitions among all classes that occur between time rand 1+ I yields a new class distribution vector given by Tx(i), where Tis the m x m transition matrix, TMma~fi) Taf Wah vnfn Mfaft d= fs) Tn fomfn Thfifi Uhr : Taf fon Tfoafs — Mefasfs Hap(L = find Here I], is the probability that a j-lass individual survives one unit of time. After ‘one unit of time. a fraction f, of the surviving individuals leave class j, with a fraction f,, of this group moving into class /# j. In addition to these transitions. classes may obtain new members due to births. (Immigrations and emigrations will be ignored.) After one unit of time, a surviving j-class individual gives birthCompeting Size-Structured Species 29 to b,, -class offspring. and consequently, the class distribution vector of offspring at time 1+ 1 is given by Bx(1), where B= (ILp,). [Here the census is assumed taken immediately following reproduction by surviving individuals. If, on the other hand, the census were taken immediately before reproduction, then B would be (11,6,).] If both births and class transitions are taken into account. the class distribution vector at time r+ 1 is the sum Tr(#) + Bx(i) and the dynamics of the class distribution vector are determined by the matrix difference equation HUF I= PHO, P=TH+B w In the event that some parameters are dependent on time + [either explicitly or implicitly through a dependence upon the density vector x(1)] the projection matrix P'is dependent on [i.e.. P= P(t,x()]. Thus (1) can be nonlinear. This general model includes the famous matrix model for age-structured popu: lations of Leslie (1945), In this case the classes are age classes one time unit in length, so that f,= 1. fi, =Oifj ##— 1, and fy = 1. (ie. all surviving individ- uals must advance one age class in one unit of time). P then becomes a Leslie matrix. Another special case occurs when 0 0 with an associated positive eigenvector > 0, which is normalized so that we! : = 1 Also assume that x(t) = 0 is a solution of (2) for which x(0) = 0 (# 0) and p(x) > 0 for £20, Then me) +67 > .as 1 bx This is a generalization of a nonlinear ergodic result of Cushing (1989b) (see Caswell, 1989). The dynamics of total weighted population size p(t) are governed by the equation (ue) + BU) we! LON) LO 3 rue) which can be obtained by forming the inner product of (2) with w! and noting that w! Le(e) = we" Lnteyp(e). By Theorem 1. we! L(t) +10! Le! = A‘ wl = A Thus (3) is “asymptotic” to the “limit equation” ait) + A> BU PLO, w pies and one might expect that the long-time asymptotic dynamics of p(t) can be deduced from this equation in place of (3). These ideas can be made mathema tically rigorous (LaSalle, 1976). but these technicalities will not detain us here We will use (4) to study the dynamics of the total weighted population the competition models below. Let us now briefly consider a specific model of “interference competition based on size classes. [More details and further analysis of this model will appear elsewhere (Cushing, 1990).] Let x,(r) denote the number or density of individuals cof length s lying in the size clas interval [5,1.5,). 0 §< 5; "+ < Sy. Let y be a representative length of an individual from size css [e.g.. the average s, (6, ~§, M2}. An individual of size , will be an individual of size cass i, Thr, is the “inherent” resource uptake rate per unit time per unit body area of an individual of size class i (i... the uptake rate at low population densities in the absence of intraspecific or interspecific competitive effects) and ¢ “denotes the fractional decrease in resource uptake of an individual (per unit of body surface area per unit time) due 10 a unit surface area of competitor. the resource uptake ate of an individual of sizes, is approximately rers5 exp(— dp(t)). Here pt) v's) is the total surface area of all individuals at time ¢ and the weighting vector v« consists of the (average) body surface areas 1 = col(u,s3), The constant «r, > 0 is a constant of proportionality that relates surface area to the square of body length (it clearly depends on body geometry). In this model the resource is supplied at a constant rate and is not dynamically modeled fall individual organisms of size s, have uniform density and body volume scales cubically o length. then an individual's body weight is given by W, = 1,9!Competing Size-Structured Species 31 ‘The weight change ,W of an individual of size i over a unit of time is approxi- mately 34,$3.,s, where 4s is its change in length over one unit of time. On the other hand, weight change can be related to the resource uptake rate by the formula 4,W = x,r,0753 exp{— dp(1))'n,. where x, > 0 is the fraction of consumed resource that individuals of size s; allocate to growth and , >0 is a conversion factor of resource units to body weight. It follows that the growth rate per unit time of an individual of size fs 3,8 = «0:1, expl— dp(O/340, We will assume that no individual can shrink in size and for simplicity that the size class interval lengths 3, = 5, ~ §, are chosen so that in one unit of time an individual can grow no more in length than into the next size class, The transition matrix T= 0 is given by the bidiagonal matrix — a to pin eo Tel 0 oO 0 (5) : eG 0 0 (ee ee Here B, = «7 /3y1 7.8, is defined to be the “growth coefficient." The reason for the 1 in the lower right-hand corner of T i that no individual can grow larger ‘Consider now the fertility matrix B. It will be assumed that all newborns lie in the smallest size class, [5). 51). If the remaining fraction 1 ~ x, of consumed resource is allocated to reproduction andif w, > Ois a conversion factor of resource units to offspring body weight. the birth rate for an individual of size Fis a, expt dp(e)}. where is defined to be the “reproductive coefficient” and where Ww = jus} isthe weight at birth, Then OF o Bane“ 6) 00 0 ‘The dynamics of the population density vector x(t) are governed by equation (2) with a(n=M= Bye), b(n) =e and with L given by the nonnegative Usher matrix By~ Bit a ay tt ty Gy Gy, B By ~ Bi : 0 0 0 0 0 0 0 0 0 0 0 Be~ Bus 0 0 0 0 Bua A,32 Cushing where B, is the largest B,. Clearly, L > 0, 0 a(1) = ay =I, and (1) = 0. In fact. (1) is bounded away from 0. as is required by (H1), because it can be shown that (1) is bounded (see Lemma 2 of the Appendix). Also. if L satisfies (H2) (the Perron-Frobenius theory applies to this question). the dynamics of p(r) are deter mined by the scalar equation (4), that is. by the scalar difference equation ple + 1) = TCL + Be“) ptr) a with p(O) = wx(0) >0, where @= A ~ B,, Since A” >0 is the dominant eigen- value of L. @is that eigenvalue of the matrix Bit a 1 Gn Bi 0 M=L-pIl=| 0 o 0 “s) 0 0 0 0 with the largest real part. The dynamics of (7) can be summarized as follows. There exists a unique positive equilibrium p()=e=d"! Ine’), & =(1~I1V/II, provided that 16' > 1. If 8 > 1. then ple) +0 as r+ =. IF 1 < 16 < @, = expll/(1 ~ M1). then ¢>0 is globally attracting and solutions p(:) are nonoscillatory. If 4, < 616" < 6%, then e > 0 is globally attracting and solutions p(t) are oscillatory. but damped. If 6/@' > 63. then e >0 is repelling and there occurs the familiar cascade of periodic doubling bifurcations of stable limit cycles and ultimately “chaos” as 6/6" is increased. Consider now the interaction of m size-structured species j= 1.2... ..1m. each of whose dynamics is described by a model of the type described above. all of ‘whom utilize a common limiting resource that is supplied to species j at a constant rate rj. Each species is described by an m,-vector ¥)(t) = col(x, (2) of size class densities. It is assumed that the same length scale is used for each species. but not necessarily the same size classes or number of size classes. All model par- ameters above are now subscripted by j to indicate species ay well as size depe' dence. We then obtain a system of n coupled nonlinear matrix equations for the rn class density vectors x,(7). Just as in the single species case considered above. we find that the normalized size distribution of each species asymptotically ap- proaches the normalized positive eigenvector of L,. regardless of the asymptotic dynamics of x(f) or p,(t) and that the dynamics of the total population surface areas p,(¢) are governed by the system of limiting equations pat + A) = TL + ee) pH) 0° Here 6, is the eigenvalue of M, with largest real part. M, has the form (8) with additional subscripts j on its entries to indicate species dependence. It is easy to see that (9) possesses no equilibria for which all species are positive (except in the infinitely unlikely case when two or more nonlinear equi identical solutions), The only nonnegative equilibria are the zero equilibrium and the n axis equilibria for which exactly one species is present. namely p, = 0 for j#kandCompeting Size-Structured Species 33 a Pemex where ex = de'In (20) provided that 6/6 > 1 (i.e., provided that the kth species has a positive equilib- rium in the absence of the other species). In fact. any species for which 0/0 <1 will be eliminated, that is, any species that has no positive equilibrium in the absence of competition cannot survive in the presence of competition. This is easily seen from the inequality 0 = p,(#-+ 1) = 11,(1 + 6,)pa(0). Which follows from (9). Accordingly, we assume that foi fora (j-e., each species has a positive equilibrium in the absence of competition). A simple linearization argument shows that this assumption implies that the zero equilibrium of (9) is a repeller. Thus all species will not be eliminated. ‘The axis equilibrium (10) will be referred to as the “inherent equilibrium of the kth species.” An inherent equilibrium will be called “inherently stable” if that species is stable in the absence of competition fi... if ¢g is @ stable equilibrium of (7) with subscripts k). This occurs if 4/82 < é = exp[ 2/(1 ~ H)]. An inher- ent equilibrium will be called “competitively stable” if itis stable in the presence of competition, {ie.. if it is stable as an equilibrium of (9)]. A competitively stable axis equilibrium implies that the Ath species is immune to competitive exclusion from invasion by small populations of the other species ‘The cigenvalues of the linearization of (9) at the kth equilibrium (10) are 1=(1= Madgex and 11+ 6,exp(~ deeg)), 7 # &. The first lies in the interval (~1, +1) provided that 4/4 < iy. (i.e., the kth species is inherently stable). All the remaining positive eigenvalues lie in (1, + 1) provided that 4 for anja k a 6 We conclude that the inherent equilibrium (10) for species k is locally competi- tively stable provided that it is inherently locally stable and provided that it has the largest ratio 9/0), In particular, the competition equations (9) can have at ‘most one locally stable equilibrium. Under some further parameter restrictions this local stability result can be shown to be global (Cushing, 1990) This stability result is of course consistent with the basic principle in ecology that no more than one species can survive on a single limiting resource. Whereas this “competitive exclusion principle” holds here as far as equilibrium dynamics is concerned, it does not necessarily hold for model (9) if nonequilibrium dynamics are considered, When one species has nonequilibrium inherent dynamics, it is possible for more than one species to coexist on a single resource. This coexistence is by means of a nonequilibrium state and can be in the form of limit eycles or even “chaos.” [This phenomenon has been observed before in nonstructured models for exploitative competition models by McGehee and Armstrong (1977). Butler et al. (1983), and others. Such cases will not be discussed further here, as ‘only equilibrium dynamics will be considered ‘The surviving species is determined by the aggregate parameters 6, which contain the size-specific parameters of the model, and by @ = (I~ I1,VIl,. Witha Cushing regard to the size-specific model parameters, this model implies (insofar as equilib- rium dynamics are concerned) that it is 10 a species’ advantage in competitive situation 10 have a large @, value. To relate competitive success to species body size it is required to explore the relationship between the eigenvalue #, and the entries a. 6, of the matrix M,. It can be shown, under the reasonable assumptions that a ~ 0 (the smallest size class is infertile) and «, > a for all i m, that the eigenvalue @ of a matrix M with form (8) is an increasing function of each of the size-class-specific reproductive coefficients @, (Cushing, 1990). Thus an increase in any of these coefficients can be viewed as advantageous to a species insofar as potential competitive situations are concerned. For example, consider a competitive interaction between species that are similar, or more precisely model identical. except for maximal adult size, In this case all 8, are equal and the winning species has the largest eigenvalue & The matrices M, differ only in the reproductive coefficient of the largest size class m. Asa result the 8, are eigenvalues of matrices (8) differing only in the entry ay. It follows that the surviving species (i.e. the only locally stable species) is the one with the largest m class reproductive coefficient a. which. if all other model parameters are identical. is the species with the largest adult size 5,,. Thus. under these circumstances. our model supports the first tenet of the SEH, at least for ‘competitive interactions between “similar” species. However. if suitable differences exist between the species (i.c.. between other model parameters) it is easily possible for the largest species to be a competitive loser. For example. a smaller species can have a larger ratio 6’ by possessing higher survivability factors, smaller resource conversion factors. different resource allocation factors, smaller weight at birth, or smaller maturation size. A simple case will illustrate this point ‘Suppose that there are no significant differences in the growth coefficients of all juvenile classes or the reproductive coefficients of all adult classes and suppose that once an individual reaches a maturation size it remains fertile. In such a cave all juvenile size classes and all adult size classes can be grouped and a model consisting of only two size classes considered. In this case the matrix, —B, ay Ms (0) is 22 anda study ofthe parameter dependence ofits egensie 4 8B +B + Ba") is relatively straightforward. (For simplicity the unnecessary size class index has been suppressed: 6) = By. a= ay and 5, $2. 6) = Ki eC.) Consider 6, = 6,(s,. ,) as a function of s, “adult size) and 6, (the length of the juvenile size class or simply the “maturation size"). A typical level curve C: #(5,.8) = cis drawn in Fig. 1. From this graph one sees that a change ins, or 8, that moves the point (5,5) to the right (left) of C, will result in an increase (decrease) in 8, With regard to competitive interactions. this observation can be interpreted in the following way. If another species r, with identical parameters. except for s, and 5, is putin competition with species x, then it will competitively ‘exclude this species ifthe point(s. 3,) lies to the right of the curve Cin Fig. 1Competing Size-Structured Species 35 JUVENILE SIZE INTERVAL ADULT SIZE Figure 1 The level curves of the eigenvalue (5.5) = of M, in the 22 case are parabolic in the s,, 8-plane. sis the (average) adult size and 6, is the length of the juvenile size class. Thus, once again, we find that if wo species are otherwise identical, the one with the smaller adult size will suffer competitive exclusion, We also see that if two species are otherwise identical, the species with the larger size at maturation will suffer competitive exclusion, Notice, however, that if both adult and juvenile sizes are changed, a decrease in adult size need not result in competitive exclusion, It is clear from Fig, 1 that an increase in adult size accompanied by an increase in maturation size could possibly result in parameter values to the left of the curve C, and hence a decrease in 8, Consequently, a species with larger adult size could competitively lose out to a species with a smaller adult size if its maturation size is too much larger than that of the smaller species. Further studies of this discrete model can be found in Cushing (1990). We turn now to some models of competing size-structured species in which time and size are continuous variables. WI. CONTINUOUS SIZE-STRUCTURED COMPETITION MODEL We begin with a capsule summary of the model for the dynamics of a single size- structured species given by Diekmann et al. (1984). A full treatment can be found in Metz and Diekmann (1986, Chap. 1.3). The models in Section II are discrete analogs of these continuous models, although there will be some differences (e.g the inclusion of metabolic demands in the continuous model). The basic assumption is again that food ingestion is proportional to body surface area. If s denotes length and R denotes resource or food density, then36 Cushing resource uptake rate is f(R)s*, where the resource uptake rate (per unit area) f(R) is dependent on resource density R. It is assumed that this rate is continuously differentiable and positive whenever the resource density is nonzero: fO)=0, f(R)>O for R>0 Although it is not necessary for much of the analyis below, for simplicity only the case of monotonically increasing rates f(R) will be considered. A typical such relationship is the Michaelis-Menten or Holling type I expression /(R) = Ria > R).c> 0, a>0. It is further assumed that ingested food is allocated between metabolic main- tenance. individual growth. and reproduction. Energy reserves are ignored. If ‘metabolic maintenance needs are assumed proportional to weight (or. up 10 a scaling factor. to volume), ifa fraction x of ingested food is available for individual growth and metabolic demands. and if 1 is a conversion factor relating weight to food units. then one obtains a growth equation ds/dr= g(R.s) with @(R.s) = [xf(R) ~ gs is the metabolic rate {i.e., the rate (per unit volume) of food units needed for metabolic maintenance]. Here [x], = x for y > 0, and 0 otherwise. It is assumed that the remaining fraction I ~ x of ingested food is channeled to reproduction. If s», and s, denote the length at birth of all individuals and the length at which reproduction starts. respectively, and if reproduction ceases when the growth rate drops to zero. the individual birthrate is given by o 4 =s=9(R) (Ros {(R)s* = g3° man(s,.s(R)) = 5 = 5(R) Here w is a reproduction conversion factor relating food units to offspring weight: that is. ew, is the amount of food needed to produce one offspring, where 1, is the weight at birth (if individuals are assumed to have a uniform density. then with an appropriate choice of units. w),= 3). s(R) is the size at which growth stops at resource density R and 5(R) is the size at which all food is needed just for maintenance. A(R) = «f(R) SE F(R) = CR) Finally itis assumed that the individual death rate d > 0 is a constant independent of size. time, and resource density R. ‘The foregoing submodels for growth. birth. and death rates ean be ineorpor- sated into dynamical equations for population level dynamics by using the modeling methodology of Metz and Diekmann (1986) (see also Sinko and Streifer. 1967) If {r,s} denotes population density as @ function of time ¢ and length s. then for 120, a a 4 yt.s) +2 (giR.s)ple.s)) = ~dpltss) for s,5(R) To complete the formulation of the model, the dynamies of the resource must be specified. The simplest assumption is that the food resource density R is somehow held constant over time. This case is investigated in some detail by Metz and Dickmann (1986). The more realistic case when resource density can change due to the feeding activity of the population and to other external causes is considered here. If k(R) denotes the inherent density dependent growih rate of the resource R= R(1) in the absence of feeding by the population p. then 2c) - p09 fa dt A typical resource growth law is k(R) = (Ry ~ Ryd, Ro >0. d>0, which corre- sponds to a constant replenishing of nonreproducing food particles and a constant system washout (by. for example, dilution in a continuous-flow culture such as in a chemostat). Another would be the classical logistic law for a self-renewing resource, We are interested in the dynamics of several competing species of the type meeting the assumptions described above, all of which attempt to utilize the same limiting resource R. It will be assumed that the n interacting species. whose densities will be denoted by p,, 1 in, interact with each other by means of this common limited food resource and in no other way so that the competition is of the type usually referred to as “exploitative,” as opposed to “interference. Thus, for i= 1.....nand 1>0, 2 oite.s) + 2 (EAR sols) =~ dallas). s4<8 <5,(R) an as BARS) P:At55) = ie m(R.s)p.(tes) ds ay = MR)~ BGR] pss? a a together with pi(t,s) = 0, s > 5)(R). All quantities «, £, «and so on, are species specific and are consequently subscripted. In the absence of the other competitors the dynamics of the species p, are governed for = 0 by the equations Falta) + 2 (Re sdots)) =~ dplts)s —54,<5<5(R) a (@) 5 (b) gi(Resn doles», [ maesrotesyas (2, K(R) ~ ie) f pilt.s)s? ds together with p/(t,s) = 0, $>5,(R).38 Cushing We are concerned with both the existence and stability of equilibrium solutions of these equations. A positive equilibrium solution (R().pi(f.8)) = (Ay. p(8)) = 0) oof (12),.is feasible only if s, < 5,(A,). [It is not difficult to show that otherwise the only equilibrium is (4.0), where A > 0 satisfies A(A) = 0.] IF the equilibrium equa forsy, = <3.) o for s s,Q) where y, = £)/n,. is substituted into (2b), and (12c),, one finds that A, > 0 must satisiy 1 (3, (4).0) = (3), in which case tiny aea(stan fo (BEI) eas) ‘The expression Il, in (13), is defined by meee ea (3(2,).0) is the net reproductive rate at resource level A, (i.c.. the expe: number of offspring per individual per lifetime). There exists a positive equilibrium oof (13), if and only if equation (13), has a positive solution A, > 0) for which K(A,) =0. It can be shown that because f, (A) is monotonically increasing in A. so is H,(s,(A).0). Inasmuch as 1, (0,0) = 0. (3), has a unigue positive solution it 11(s,(A).0) > | for large A> 0. ‘A similar analysis ofthe equilibrium equations associated with the competition equations (11) shows that in order to have an equilibrium in which two or more 2 are positive. two or more or the equations I1,(s,(4).0) = I must have a common solution A>0. The improbability of this occurrence is commensurate with the principle of competitive exclusion. which says that it is impossible for two oF more species to coexist on a single resource. Thus we will restrict our investigation of stability to “axis” — equilibria ofthe form. (R-pi.Po.- Px) (A000. 0.68.00... .0). A, > Opi 20. where (A,. 2) isan equilibrium of (12). We are interested in whether or not a species p,. which can successtully exist con the resource R. can be invaded by a competitor p,. k # i. For simplicity let 1. Thus we will assume the existence of a stable positive equilibrium (a1. p/) > of (12), and consider the stability of (pi... ...0) as an equilibrium of (1) Because the competition model (11)) is exploitative (ie.. the coupling of the equations is through and only through the resource equation). the linearization at (Api... 0) is such that given the assumed stability of (Ay. pi) as a solution of (12),. stability is determined by the linearization of the p, equations at (A). pi) for i # 1, These equations are SdCompeting Size-Structured Species 39 ala St (RA. S)p) = = dp, ar? as fo menor (Ay SPIES If we look for nontrivial solutions of the form p, = 6,(s)e". the equations for 4, are Seid bO0N= = (d #28) BAS deon)= [ 1m (A4.8) (3) ds These equations have @ nontrivial solution ¢ if and only if: satisfies the equation 11, (A) 2) = 1. If these “characteristic equations” have. for i# I. only complex roots with Re z <0, the equilibrium is stable, If for some i # | thete exists a root Re = > 0, the equilibrium is unstable. [We have proceeded in a formal fashion in this stability analysis. A rigorous justification of linearized stability for a quite general single-species size-structured model, but without dynamical resource. is given by Tucker and Zimmerman (1988). Their treatment could probably be extended to a system of model equations such as (11) to give rigorous mathematical justification of our results here.] It can be shown that II, has the following properties for o > 0: T,(o.2) 40 asc =real> + (7.3) =1L(o.0) for Rez=0 ‘The following theorem easily follows from these facts. Theorem 2, Assume that equations (12), have a stable positive equilibrium (A. 63}. ETT, (5,(4,).0) < L for all # 2. then the axis equilibrium (A. 97.0... .0) ‘of equations (11) is stable. If Tl(s1(A:).0) > I for some 7 # 2. this equilibrium is, unstable. In the absence of a global analysis of a competition model. questions concern- ing the competitive outcome are usually investigated by means of the local lin- earized stability of the “axis” equilibria such as (A,.p%.0.....0). If such an equilibrium is stable, then the species p, is invulnerable to competitive exclusion by at least small densities of the remaining species. The criterion for this is. by Theorem 2, that the net reproductive rates of the invading species are all less than replacement level at the resource equilibrium Ay. If, on the other hand, at least cone invading species has @ net reproductive rate at resource level A, greater than, 1. species p, is vulnerable to exclusion by this species. ‘The expression 11,(o.0) = LE. al (= : aay Kiawny 63m Jy, So can be integrated to obtain40 Cushing Lee, ne.0) 82 ep race ay at 9 =) + tg ~ 5,9 a ae yd +20 Both p, and q, are increasing functions of their arguments o Thus H,(e0) is increasing in o, On the other hand, for fixed «. 11,(or.0) is decreasing in s, [see aay} The stability criteria in Theorem 2 depend on most model parameters in a rather complicated way and we will not attempt here to study this dependence in any detail. Instead, only one question will be asked: For otherwise similar species. how does the stability of the axis equilibria depend on the relative adult and juvenile sizes of the competing species? By “otherwise similar” is meant the identity of all the model parameters as) =o man & (hence y, = y. Wn, =") Suppose that species p; has the largest adult size at the resource level Ay so. that 5,(A)) = s\(A,). This will occur if species p, has a larger resource uptake rate: f,(y) = f(A)). Furthermore, if the species have the same maturation length (or slightly more generally. if no species has a smaller maturation length than that of species p; (i... ¥, =, for i# 1}. the monotonicity properties of the I1,(7.10) imply the inequalities TH (3,(41).00 5 Hy(s,(A1)-0) = Ts (4.0) ‘Theorem 2 now implies that species p, is locally invulnerable to invasion. Notice that in the case considered here the largest species wins the competition (at least near equilibrium and against invasion by small densities of smaller species), in keeping with the basic tenet of the SEH. ‘A species py that is largest at the equilibrium resource level A; may not be. however. the largest species at another resource level A, associated with the axis equilibrium corresponding to another species p.. That is. 1(A1) > 5,(A1) does not necessarily imply that $\(A,) > s,(A)). I. however. p, is the largest species at all axis equilibria resource levels A,. similar local stability analysis shows that the other axis equilibria are unstable. This. and the absence of any other positive or nonnegative equilibria. would suggest that the axis equilibrium for species py might be globally attracting. This is the case in many exploitative competition models without structure (Butler and Wolkowicz. 1985). On the other hand. if species p, is the largest species at resource level A, while species p, is the largest species at resource level A, the competitive outcome for intial states far from equilibria will presumably depend on initial densities. These questions need investigation. A larger species may not be invulnerable to invasion by smaller species if the smaller species has other model parameter that are significantly different from those of the larger species. For example. suppose that we retain the conditionsCompeting Size-Structured Species 4 (5) (so that the species still differ only in adult and juvenile sizes) and still assume that species pis the largest species at resource level Ay. However. unlike the case just considered, suppose that a smaller species p, has a smaller maturation length $53}. As noted above. the net reproductive rate TI, (s,(A1)-0) is a decreasing function of 5, Consequently. this quantity increases as 5, decreases and there is the possibility that it will exceed I for sufficiently small s,, (but. of course 5, = 5). If this happens. species p, would be vulnerable to invasion and exclusion by this smaller (adult size) species. For example, this will definitely occur if £(1 = ky/wxd > 1. This could be interpreted as a simple example of a juvenile bortleneck-like phenomenon, although a more appropriate model of the juvenile bottleneck would incorporate strong competitive effects on the growth of the juveniles of the larger species from the adults of the smaller species. For a further study of a special case of (11). see Cushing (1989) IV. CONCLUSION We have briefly looked at two models for the dynamics of m size-structured species competing for a single resource. one discrete and one continuous in time and length scales. These models are based on the fundamental assumption that re- source uptake scales to the square of body length. Both models are consistent with the basic competitive exclusion principle as far as equilibrium dynamics are concened (i... that at most one species can survive at equilibrium on a single limiting resource). Parameter relationships were derived which determine the surviving species, and as an application the question of how competitive success depends on adult and juvenile body size was considered. It was seen that in general the connection between competitive success and body size in both models is a complicated one involving a large number of the model parameters and that without further restrictions on parameter ranges. body size is not a predictor of competitive success. However. under the assumption that all species are otherwise “similar,” it was indeed found that the surviving species has the largest adult body size, However. species with smaller adult sizes can eliminate species with larger adult sizes in a number of different ways. including earlier maturation, smaller weight at birth. higher survivability, more efficient resource conversion factors, or different reproduction-to-growth resource allocation patterns, ‘APPENDIX Part of Theorem 1: From (2) it follows pl + 1) = wi" (a(n! + IDL)X(Q) which when divided into (2) yields the equation (oll + Ly wh (—(QT + Lyn(o)” for n(1). where p(t) = a(0)/6(0). It is not difficult to show that the solution of this equation is given by (0) = way" Mr), where wis the solution of YEH Y= (OLE + Lyle), (0) = 140) (Ad) ne + 1= 01,2,2 Cushing For simplicity we give a proof when L has a basis of eigenvectors ¥ Vsc00+¥in In the more general case a similar proof can be constructed using generalized eigenvectors. [See Impagliazzo (1985) for an analogous proof in the age-structured case] The coefficient matrix p(e)l + L in (A.1) has eigenvalues a, + p(t) with eigen vectors v,. Expand (0) = 37", 6,1,. The solution of (A.1) is given by wi) = Se Ta + ptkoe, and (0) is given by She Mis + atk) Sve MV, + lke, It we divide numerator and denominator by HE, (Ar + (A) and define thio Th (2248) then 1a 26 TL (ape ent ny = 7 By Lemma 1 below, we have that [1,(1)—+0 as 1+ * for £22 and thus st) —-cvvadcuel ry = v4 = 0". Note: cy is positive because 1 = ut a0) where 11 is the left eigenvector of L associated with Ay = A”. which is known to be a positive vector (Gantmacher. 1960) Lemma to I AL>[&| for f= 2.....m and 0-5 ple) S py + % for all = 0.1.2... then TH, () +0 as 1+ + * for each Proof: It is easy to see geometrically in the complex plane that Ay >A, implies that |A, + xr +2) = 1 for all x0, Since this ratio is continuous in x. itis bounded away from 1 on bounded x intervals. that is.0-< A, + /(A, +4) 0 and both sequences are bounded for all > 0. Proof: Assume that 0 x(1) # 0. By (H1). pli) =0 for all r and from (1) follows x(1 + 1) =0. An easy investigation of the components of P(2)x*) in reverse order shows that x(¢ +1) =0 implies the contradiction x(0) = 0. Thus x(r-# 0) # (0. By induction (= (0) # O for alr. This in turn implies that p(1) > 0 for all r= 0. Clearly. p(1) is bounded it x(2) is bounded for 120. To show that x(n is bounded. we consider = s(1) = 37", ,(). From (1) and the definitions of T and T in (4) and (5), we obtain s(r+ 1) = Ts(n) + Ure “$7 y,3, (2). from whieh follows for r= 0 Os0(1-+ 1) 5 F(t) Ad with s(0) = .4,(0) > 0, where F(s) = 11 + yte""s and s* = min dys > 0. y* = max ry, =0. F(s) =0 has the following property: There exists real s'=0Competing Size-Structured Species B such that F(s) is monotonically increasing and satisties F(s) <5 for s>s' and mano. F(s) = FAs"). If it happens that s(1)>s' for all ¢=0. then by (A2) we have F(s(0)) 0. (0) = Qu20 6) Here P(0) is the number of proliferating cells at time 1. Q() the number of48 Gyllenberg and Webb quiescent cells at time t, (2) the total number of cells in the tumor at time # and therefore a measure of the size of the tumor: B the birth or division rate. is the death rate of proliferating cells, and jug the death rate of quiescent cells. 1.) is the rate at which proliferating cells become quiescent and r (1) isthe transition rate into the proliferating class from the quiescent class when the tumor size is N. We suppose that is a positive constant and yy and yy are nonnegative constants, 7,(N) is @ continuous nonnegative nondecreasing function of N and 1,(N} isa continuous nonnegative nonincreasing function of N. In the absence of quiescence b:= ~ jp is the specific growth rate or Malthusian parameter of the population. Throughout this section we assume that b is strictly positive It is the increasing tendency of cells to become and to remain quiescent as the tumor becomes larger that constitutes the growth limiting mechanism, The limits eo tim ro(N) a” lim r5(N) (8) of the transition rates as tumor size tends to infinity will therefore play an important role in what follows. Observe that according to our assumptions about r,, and r, 1, and |, are well defined by (7) and (8) and that 0 1, = *.0=1, <= Our interest is in the qualitative behavior of solutions as time progresses. This behavior is atypical Gompertzian growth of total tumor size N(¢) and a diminishing Of the growth fraction Po NO This occurs regardless of the specific choices of transition rates r,(N) and r,(N) as long as r(N) is nondecreasing and r(N)) nonincreasing. The resulty stated below are proved in Gyllenberg and Webb (1989). Theorem 1. Suppose that wg = 0. r,(N) = 0 for all N= 0 and 1, > h. Then N(o) is increasing, bounded. and lim... N(1) = N*. where N'* is the unique solution of Gn, oy buns 20) = f ranyan (10) IF r.(No) > b. then N (0) is concave. but if r,(Nu) b= B~ hep has a natural biological interpretation. [t says that for very’ large tumors the rate at which cells become quiescent or die is treater than the rate of recruitment of new cells. If we suppose that transition to‘Quiescence in Sructured Population Dynamics 9 ‘quiescence and death take place immediately after mitosis, the assumption means that for very large tumors on average more than one of the two cells produced at mitosis will either become quiescent or dc ‘The conclusions of Theorem | tell us that the ultimate size N* of the tumor is finite and depends on the initial state. N (1) grows monotonically to N* and has a unique turning point, which is attained at the moment when the net rate of production of new cells equals the transition into quiescence. Thus the growth curve N(t) has the typial S-shape of the Gompertz and logistic curves, ‘Truly quiescent cells retain their capability to divide, The next theorem, which assumes thatthe transition rate from quiescence to proliferation is positive, applies to this situation, It is still assumed that quiescent cells do not die Theorem 2. Suppose that yxy = 0 and there exis positive constants ry. Ff such that £, = 1,(N) =F, for all N=0. Then N(0) is increasing and lim... N( = = Let G = PUN (1). WEL, <=. then lim,..- Pl) = * and on a EEE + Abt a2) 2 where L:= b =U, ~ ly If lp = %, then NrAN) lim sup P(e) = lim sup NX) (3) Seine eR (N), im int P(r) ro) and consequently, G* = 0. In the case of necrosis described by Theorem 1. the choice of r.(V) had little influence on the shape of N (1). The situation of Theorem 2 is completely different. Since N’ = bP the relations (13) and (14) show that the faster r,(N) tends to infinity as N+, the more slowly N(#) tends to infinity as r—+. If r,(N) tends to infinity faster than a linear function, the slope of N(t) tends to zero and the curve looks very similar to the Gompertz curve. In Theorems 1 and 2 we assumed no mortality in the quiescent class. This is a realistic assumption, since in some tumors the cells that die seem to be the proliferating ones (Tubiana, 1971). In other types of tumors, however. disap- pearing cells are for the most part quiescent (Tannock. 1968). Our last result for the model (3)=(6) is concerned with tumors of this kind. This time we have to assume that r, is not only nondecreasing but actually strictly increasing. The assumption about r, remains the same as before. Theorem 3. Suppose that (i) wg >0, (ii) r4(0)0 (as) p(O.x) = x), Bersn 9) a0. uy, Bors 20) Here p(t.) and q(t.) are the densities with respect to cell size x of the proliferating and quiescent classes, respectively, at time t. The minimum division size is xo and the maximum division size is xy. It is assumed that mother cells divide symmetrically into two daughter cells, so the minimum possible cell size is, -xy/2. We assume that xy/2 on [1/2. x1]. Cell division depends on cell size and is governed by the per capita fission rate b(x). We take (x) to be positive on (1, 81) and 0 outside this interval, This means that cells reaching size x, will never take part in reproduction: they are considered dead. When x > .x/2 the first term to the right of equation (16) is defined to be zer0. s(x) and wo(x) are the size dependent per capita mortality rates of the proliferating and quiescent class. respectively. They are nonnegative. of course. The functions r(x..N) and gv.) describe the interaction between the two classes. When a cell divides. on average r(x, N) of the daughter cells enter quiescence. Thus r(x. N) is a number between 0 and 2. Quiescent cells reenter the proliferating class at a rate governed by ‘x. N). Clearly we have to assume that p(x. N) is nonnegative In addition to the obvious restrictions on the vital rates imposed by their biological interpretations, we need certain smoothness hypotheses. It is, for in- stance, clear that to prove existence and uniqueness of the solutions of the model (16)-(20), we need a Lipschitz condition on r and p. We shall not state these conditions explicitly but only vaguely assume that all occurring functions are “sulficiently” smooth. The interested reader is referred to Gyllenberg and Webb (to appear) for an exact formulation of the hypotheses needed. That paper also contains detailed proofs of all results presented here. It is important to note that the nonlinearity caused by the dependence on total tumor size N enters the equations only through the functions r(x. N) and ptx. N) describing the interaction between the proliferating and quiescent classes, whereas the functions g(x) and 6(x) describing proliferation (j.e.. growth and division. are assumed to be independent of N. This assumption is strongly suggested by numer- ‘ous experiments showing that the mean duration of the cell cycle in solid tumors, is relatively constant and that for a given tumor. variation in the growth rate is due mainly to variations in the ratio between proliferating and quiescent cells, (Tubiana, 1971; Martinez and Griego, 1980). On the other hand, itis quite likely that cell loss depends on tumor size. For instance, the risk of formation of met stases increases with increasing size of the primary tumor (Tubiana. 1982), We have neglected this dependence in our model because we are interested in thes Gyllenberg and Webb consequences of quiescence for tumor growth, and size-dependemt mortality rates ‘would only complicate the model without bringing new insight into the role played by quiescence. We shall consider the model (16)-(20) as an abstract evolution equation in a Banach space Z. As state space Z we choose X x ¥. where X= L! (tw/2. x1). Y= L" (xy/2, x1/2) with fb, wz = ly + llvdy. This is a natural choice, since the norms in X and Y give the total number of proliferating and quiescent cells, respectively, and the norm in Z gives the total number of cells in the tumor Actually, Z is isomettically isomorphic 10 L'(Q2), where the individual state space {is the disjoine union of the permitted size intervals for proliferating and quiescent cells, respectively IV. POSITIVE SEMIGROUPS AND ASYNCHRONOUS EXPONENTIAL GROWTH In the study of structured population models, both linear and nonlinear. the theory of semigroups has turned out to be a very powerful tool (Webb, 1985: Metz and Diekmann, 1986). Semigroups associated with solutions of structured population models have an important property: They are positive, (This is clear since only positive densities are biologically meaningful.) In the analysis of asymptotic prop- erties of semigroups the theory of positive semigroups developed by the group in Tibingen (see Nagel, 1986) often plays a crucial role. In recent years several authors have applied this theory to problems in structured population dynamics [see. e.g. Heijmans (1986), Gyllenberg and Webb (1987). and Greiner and Nagel (1988)]. Next. we give a brief account of some central results in the theory of positive semigroups in relation to structured population dynamics. For a more complete treatment, the reader is referred t0 Nagel (1986) or Clément et al (1987) ‘An important property observed in many cell populations is the asynchronous exponential growth. Roughly speaking. asynchronous exponential growth of a population means that ultimately the total number of individuals in the population grows exponentially. while the size distribution (or age distribution or any other distribution of individual states) converges toward a stable distribution. To be more precise, let T(). 1=0 be a linear C,-Semigroup with generator A on a Banach space X. [One should think of T(0¢ as being the size distribution of the population at time ¢, given the initial size distribution X.] T(t) is said to have asynchronous exponential growth it there exist @ AER and a rank | projection P con X such that ¢~""7(0) tends to P exponentially in the operator norm topology as f tends to infinity. In other words, there should exist positive constants M and fe such that le “7(n) ~ P= Me Ais called the intrinsic growth rate or Malthusian parameter of the population, «nd Ps called the spectral projection The ultimate growth of a linear C,-semigroup can be estimated using the growth bound ay(A) defined by en 1 eA) = lim “ogli7 Ch 2Quiescence in Structured Population Dynamics 3B since for all w > wy(A) there exists an M > 0 such that |T'(n)| = Me™’. The growth bound is related to the location of the spectral values of the infinitesimal generator A. The number (A) := supfRe A: AE ofA) (23) is called the spectral bound of A. One has -* = s(A) 5 w9(A) <=. If s(A) = wo(A), the semigroup is said to have spectrum determined growth. We still define ‘one more quantity. the essential growth bound im fog a(T(H) 4) ony (A) i= where a is the measure of noncompaciness. As we shall see, the order relation between the three numbers defined in (22)-(24) will be crucial for the determina- tion of the asymptotic behavior of the semigroup T(t). Let X be a Banach lattice with positive cone X. and let T(. 1=0 be a positive semigroup on X. ie.. TIX. CX]. T(0) is called irreducible if for all HEX. (0). 6 E XI /0}. there exists a 1>0 such that (T()d. 6% > 0. The biological interpretation of irreducibility in the context of size-structured popula tion dynamics is as follows: Given any initial size distribution, after a sufficiently Jong period of time there must be individuals with sizes (almost) everywhere in the permitted size interval. Thus to obtain an irreducible solution semigroup the individual state space should be chosen as economically as possible without any “holes” to which the offspring of the initial population will not return. ‘We have the following convenient test for asynchronous exponential growth. Theorem 4, (Clément et al., 1987, Th. 9.11, p. 224). If T(t) is an irreducible positive semigroup with generator A on a Banach lattice X and if (A) = «0(A). then (9 has asynchronous exponential growth with Malthusian parameter A equal to the spectral bound s(4), and spectral projection P given P= (by 25) where is a quasiinterior point of X, (i.e. (Us U*)>0 for all wt EXT 10)), which is an eigenvector of A corresponding to the eigenvalue A= s(A) and W* is, a strictly positive functional which is an eigenvector of the adjoint operator A* corresponding to the eigenvalue A = s(A). ‘Theorem 4 may be hard to apply since the growth bounds are defined in terms Of the solution, which of course is not a priori known, and therefore the crucial inequality w:(A) O and if individual cells do not grow exponentially. 7(0) has ynchronous exponential growth with Malthusian parameter Avy, The growth fraction G(1) tends to a strictly positive limit as 1+ A more general (including the case 1g # 0) as well as a more detailed version of this theorem can be found in Gyllenberg and Webb (to appear), VI. LINEAR QUIESCENCE MODEL In this section we discuss the Tinear version of the mode! (16)-(18). that is. the case where the transition functions r and p do not depend on total tumor size N We shall not impose siny condition on the individual growth rate g(x) other than smoothness: in particular. exponential individual growth is allowed ‘Again it turns out to be advantageous to treat the model ay a perturbation of the simpler model of defective mitosis. So let Au be the operator on ¥’ defined bby (31) and define the linear operator iy in Z =X x ¥ by ld. U = [Awd (0 + nod] HA) = HA) *Y wh generates a positive Crsemigroup J. 120, Define the bounded per turbation #: ZZ by Me. Nex) = [pbb 2rVh(20) H20] (39) Since A is a positive operator we find exactly as in the analogous ease treated in section V that the operator = f+ A. #64) = 4(o%) generates a positive G.— semigroup .7(1) satisfying the variation-of-constants formula (35) [vith the new 74(1) and fi]. Our results on the asymptotic behavior of 7(2) in the ease min 70, min p> 0 are summarized in the following theorem. Theorem 7. (a) The spectral bound A; = sto! is the unique real solution on [+ minip + ug}. *) of the characteristic equation May t (40 (8)Quiescence in Structured Population Dynamics 7 where = ['fp- Hoey +a (3) [2 OBE propere arenes (b) F(e) has asynchronous exponential growth with Malthusian parameter Ay. (©) If [b. 4] # [0.0]. then the growth fraction Gi) tends to a strictly positive limit as 1 =. Skeich of proof of (b): F(t) can be solved from the variation- formula (35) using the well-known method of successive approximations. This yields the following generation expansion for F(t), t= 0: a= 3 F0) (42) f Tot — )DIy-(8) ds (3) G To prove that J{t) has asynchronous exponential growth. one applies Theorem 4. First one proves that 3; J, (0) is compact. As a consequence, (FW) 5 a( FAO) § (Full sy where ais the measure of noncompactness. It follows [see (22) and (24)] that oy(s) = en(sho) (35) Positivity of # implies that (oly) = aust) (46) Comparing the characteristic equations (32) and (40). one finds the relation (Aon) * (2h) < (31) : “7 Since the state space Z is an L'-space, Theorem 5 implies that (sh) = v6), SCL) = w(t) 48) and combining this with (45)-(47), one finally arrives at the crucial inequality (al) < wo(at) 9) To apply Theorem 4 one still has to prove that H(t) is irreducible. Let [d,W] EZ, 40}. IF 6 = 0, then W(x) # 0 for all x in a subset of (xo/2..x\/2) with nonzero measure. Since p(x) >0, a fraction of the initially quiescent cells will become proliferating. Thus there exists a 20 such that p(x) >0 for all x ina subset of (o/2,.11) with nonzero measure, Sooner or later some of these proliferat- ing cells will get offspring with size less than x)/2 0 on (%0..%:) and r(x) > 0 on (%9/2-2)/2) this group of individuals will get offspring with sizes in the whole of (t0/2,x1/2), both in the proliferating and quiescent compartment. It follows that for r sufficiently large, J(0)[4, v](x) > 0 a.e, in the permitted size intervals It is interesting to compare the behavior of solutions of the linear quiescence ‘model with the behavior of solutions of the fission model without quiescence [(26)~Se Gyllenberg and Webb (28)]. A most remarkable difference occurs in the case of exponential individual growth, In this case there is without quiescence no convergence toward a stable siz distribution, whereas with quiescence the population has asynchronous exponential growth, One can therefore say that quiescence has a stabilizing effect on the size Uistribution, To determine whether the population grows or decays. it suffices to consider the ordinary fission model without quiescence. Analyzing the character- istic equations (30) and (40). one easily finds that If A, > 0, then 0 Ay < Ab, Tf ay <0, then Ay < Ay This resuit means that in an exponentially growing tumor. introduction of quicse- tence will slow down the growth rate of the tumor. but quiescence alone is not capable of driving the cell population to extinction, In a decaying population {quiescence cannot sive the population, only make it survive a little bit longer. Vil. NONLINEAR QUIESCENCE MODEL In this section we treat the full nonlinear model (16)~(20). Our approach is based on the experimentally supported assumption that in large tumors the probability of going into quiescence is higher than in small ones and that the probability of returning to proliferation decreases as the tumor grows. Mathematically. this is expressed by the requirement that r(x. N’) is increasing in N and pv. N))is deer ing in N. Since the biological interpretation of r and p requires that r(x.N') is bounded above by 2 and ptr.) is nonnegative (see Section IIL) the following limits exist roy fim rN) ext si (9): Tim - (50) p.(xy:= lim pix.N) Meret 51) (x)= lim ph 5 ‘The functions r, and p, describe the interaction between quiescence and protifer- ation in very large tumors. We also introduce the functions misers.0). eye 2 midis pv), exc tt 53) n= plex 0). t which can be interpreted as the transition functions for very small tumors If one replaces the transition functions r(x,.N') and pox, N)) in (16), (17) by vither r,(x) and p.(x). respectively, oF r(x) and p(x). respectively. one obtains the linear quiescence model treated in Section IV. In the investigation of the nonlinear model the idea is to use already proved properties of the linear semi- group corresponding to that problem. We therefore define 3, and Ay as in (41) but with 7(x0N). pix. N) replaced by r(x). pCa and rx). pul}. respectively 7. (1. 10 denotes the linear semigroup corresponding to the quiescence model with transition Funetions r(x) and (0)Quiescence in Structured Population Dynamics 59 Theorem 8. Assume the following (i) min p>0, min r>0. (ii) r@,N) is nondecreasing in N, px XE (X9/2, x2) (iii) MAX 2meey2 {7 (UN) [re(x) ~ ror, NAN < 2, maxy,2~x= «v2 St (UN) [0(,.N) ~ pa(x)] dN<=, Then the following are true (a) 1A.) >, then (pesgu]:= lime €*Tp(t-)eg(ts J] exists in Z, where A. is the unique real root of \.(A.) = 1. Further. Y) is nonincreasing in N for all a> 0 (b) If 4.(0) <1, then (p(t, ).9(¢. )] is bounded in Z. (6) If A-(0) <1 and (0) > 1. there is a nontrivial equilibrium solution (d) If49(0) <1, the trivial equilibrium [0,0] is locally exponentially asymptoti- cally stable, The various types of asymptotic behavior in Theorem 8 are determined by the values of (0) and Ao(0). These quantities correspond to the net reproductive numbers for very large and very small tumors, respectively. If the net reproductive number of @ large tumor is greater than I. then ultimately the solutions will behave like those corresponding to the linear semigroup 7.(0). which according to Theorem 7 has asynchronous exponential growth with Malthusian parameter A. If, on the other hand, the net reproductive number of a large tumor is less than. 1. the tumor will remain bounded. Whether there in that case exists a nontrivial equilibrium or not is determined by the net reproductive number for small tumors. as is the stability of the trivial equilibrium. The proof of Theorem 8 is again based on perturbation techniques and a variation-of-constants formula. Let J(t), 120 be the nonlinear semigroup on Z associated with the solutions of (16)-(20), that is. Fld. 0] = [ptt ). lt} Then one easily finds the representation Ft) = Flt~ ty Hey) + f Tle ~ 5) BN(3)) Ts) ds (ss) where AN Yb. AC) = [2(r(4) ~ r(x. N)PB(2)(2K) + (—ON) = pe CO. = 22) ~ ree. N))B(20 B28) ~ (BEN) = pM] (55) Our last result is a structured analog of the nonlinear necrosis model of ‘Theorem 1.60. Gyllenberg and Webb Theorem 9 Let p=0. wy =0 and min r>0. Assume that the function (+ [p(t.). tt.) is continuously differentiable from [0. ) to Z. If r-x) > 1 ¥€ [rd2. 4/2}. then O(a) is nondecreasing, lim,... Q(t) <. lim,... P(t) = 0, and consequently, fim, .. G(t) = 0. Vill, DISCUSSION The size-structured quiescence model (16)~(20) can be regarded as a generalization of both the unstructured quiescence model (3)-(6) and the size-structured cell fission mode! without quiescence (26)-(28). This generalization has features that fare missing in the simpler models. which come closer to reality and which aa explain some phenomena observed. ‘The most important difference between the linear structured quiescence model and the structured model without quiescence occurs when individual cells are assumed to grow exponentially. In this case the quiescence model exhibits asym chronous exponential growth, whereas in the model without quiescence the popu: lation will grow exponentially but no convergence toward a stable size distribution will take place. The reason for the lack of convergence in the model without quiescence is that when individual cells grow exponentially. two sister cells will together grow at the same rate as the mother cell would have done if she had not divided, Thus it does not matter at what size a mother cell divides: the time at which her daughters reach a given size is predestinated. In other words. there is no dispersion in size. no information is lost and there is no ayynchronization, Mathematically. this amounts to a lack of compactness of the semigroup associated with the solutions. Quiescence provides a dispersal mechanism sufficient to make the semigroup ultimately compact, thus implying asynchronous exponential growth, At cell division some cells become quiescent and some do not. Quiescent cells remain in this state for a time period that is a random variable. Information about the initial population is indeed lost, and asynchronization takes place, In many experiments individual exponential growth and asynchronous expo- rnential growth of the populition have been observed simultaneously. The dis- cussion above shows that quiescence is a mechanism allowing compatibility of these two features. Of course, we do not claim that itis the only possible explana tion, Heijmans (1984) showed that asynchronous exponential individual growth are compatible in a model where fission does not result into to equal daughter cells but where the daughter-mother size ratio is a smooth probability density symmetrie around 1/2. Next we compare the nonlinear structured quiescence model (16)~(20) with its unstructured counterpart (3)-(6). In the case of necrosis we notice that structure does not bring anything new into the model. The structured model ay described by Theorem 9 predicts a behavior quite similar to that predicted by the unstruc- tured model of Theorem 1. Moreover. the assumptions /, > of Theorem 1 and (x) > L of Theorem 9 have the same biological interpretation Looking at the unstructured model with true quiescence [7,(N)) =, = 0] we obverve the decisive role played by uy. If there is no mortality in the quiescent class (uy = 0). the tumor will grow without bound (Theorem 2), but even the smallest positive mortality 4g will imply the existence of a globally asymptotically Mable equilibrium (Theorem 3). Growth is a characteristic feature of an untreatedQuiescence in Structured Population Dynamics a tumor and it seems very hard to find a biological justification for the stabilizing effect of mortality in the quiescent class. Again the structured model seems to come much closer to reality. Unlimited growth, boundedness, and existence of nontrivial equilibria are not determined by wg alone but by the quantities 3.(0) and (0). which involve all the vital rates (Theorem 8). If. for instance. 3..(0) > 1 and the population grows ultimately exponentially, a small perturbation of uo (or of any other of the vital rates) will not change this mode of growth, Similar statements are valid in situations (b), (c). and (d) of Theorem 8. The structured model therefore has a kind of robustness that the unstructured model lacks. ACKNOWLEDGMENT The research by Glenn F. Webb was supported by the National Science Foundation under Grant DMS-8922947, REFERENCES Bell, G. I. and Anderson. E. C. (1967). Cell Growth and Division. I. A Mathematical ‘Model with Applications to Cell Volume Distributions in Mammalian Suspension Cultures. Biophys. J. 7: 329-351 Bertuzzi. A., Gandolfi, A., and Giovenco. M. A. (1981). Mathematical Models of the Cell Cycle with a View t0 Tumor Studies. Math. Biosci. $3: 159-188, Burton. A. C. (1966). Rate of Growth of Solid Tumors as a Problem of Diffusion, Grow:h 30: 157-176, Clément, Ph.. Heijmans, H. J. A. M., Angenent, S.. van Duijn, C.1.. and de Pagter. B. (1987). One-Parameter Semigroups, CWI Monograph 5. North-Holland, Amsterdam, Diekmann O.. Heijmans, H. J. A. M.. and Thieme, H. R. (1984). On the Stability of the Cell Size Disteibution, J. Math. Biol, 19: 227-248, Dickmann, O.. Heijmans, H. J. A. M., and Thieme. H. R. (1986), On the Stability of the Cell Size Distribution. I. Time-Periodic Development Rates. in Hyperbolic Partial Differential Equations 111 (M, Witten, ed.), International Series in Modern Applied Mathematics and Computer Science. Vol. 12. Pergamon Press, Elmsford, N-Y.. pp. 491-812 Feller. W. (1939). On the Logistic Law of Growth and its Empirical Verifications in Biology ‘Acta Biotheor. 5: S166 Frindel, E., and Tubiana, M. (1971). Radiobiology and the Cell Cycle. in The Cell Cycle and Cancer (R. Baserga. ed.). Marcel Dekker, New York. pp. 391-447, Gavosto, P.. and Pileri, A. (1971). Cell Cycle of Cancer in Man, in The Cell Cycle and Cancer (R. Baserga, ed.). Marcel Dekker, New York, pp. 99-128. Greenberg, W.. Protopescu, V., and Van det Mee, C. V.M. (1987). Boundary Value Problems in Abstract Kinetic Theory: Operator Theory: Advances and Applications in Mathematics 23. Birkhauser Verlag, Basel. Greiner, G., and Nagel, R. (1988). Growth of Cell Populations via One-Parameter Semi groups of Positive Operators. in Marhemarics Applied to Science. Academic Press. New York. pp. 79-108. Gyllenberg. M., and Webb. G. F. (1987). Age-Size Structure in Populations with Quiese fence. Math. Biosci, 8: 67-95, Gyllenberg. M.. and Webb. G. F_ (1989). Quiescence as an Explanation of Gopertzian ‘Tumor Growth. Growth Dev. Aging 53 Gyllenberg, M.. and Webb. G. F. (to appear). A Nonlinear Structured Population Model (of Tumor Growth with Quiescence. J. Math. Biol.02 Gyllenbery and Webb Hejjmans, H.J. A. M. (984), On the Stable Size Distribution of Populations Reproducing by Fission into Two Unequal Parts, Math, Bioses, 72: 19-S0, Heijmans. H. J. A. M. (1986). Structured Populations, Linear Semigroups and Positivity Marl. Z. 191; 599-617. Laird, A. K, (1964), Dynamics of Tumor Grovth, Br. J, Cancer 18° 490-802 Laird, A. K. (1965). Dynamies of Tumor Growth: Comparison of Growth Rates and Extrapolation of Growth Curve to One Cell. Br. J. Canter 19: 278-291. Martinez. A. O. and Griego. R. J. (1980). Growth Dynamies of Multicell Spherioids From ‘Thrve Murine Tumors, Growel 44: [12-122 MeCredic. J. A.. Inch, W. R.. Keuuy. J. aad Watson, T. A. (1965), The Rate of Tumor Growth in Animals, Growl 29: 331-347, Mew. J. A. J.. and Diekmann, 0. (1986). The Dynamics of Physiologically Structured Populations. Springer Lecture Notes in Biomathematis 68. Heidelbery. Nagel, R. ed.) (1986). One-Parumeter Semigroups of Positive Operators, Springer Lecture Notes in Mathematics 1184, Springer-Verlag. Heidelberg, Norton, L.. Simon, R.. Brereton. H. D.. and Bogden, A. E, (1976), Predicting the Course ‘of Gompertzian Growth, Nature 264: 542-835, Pacey. A, (1983). Semigroups of Linear Operaiors and Applications tr Partial Differential Eiguations, Springer-Verlag. Heidelberg. Rotenberg. M. (1978). Correlations. Asymptotic Stability and the Gi, Theory of the Cell Cycle. in Biomurhematies and Cell Kinetics, Developments in Cell Biologs, Nol. 2 (4. 4. Valleson and P. D. M, Macdonald. eds). Elsevier! North-Holland Biomedical Press. Amsterdam. pp. §9-67, Rotenberg. M. (1982). Theory of Distributed Quiewent State in the Cell Cyele. J. Theor Biol. 96: 495-509. Rotenbeeg, M. (983). Transport Theory for Growing Cell Populations, J. Theor. Biot 103: INL-199, Sinko. J. W..and Stevifer, W. (1967), A New Model for Age-Size Structure of a Population Feology 48: 910-918. Sinko. J. W.. and Sueifer, W. (1971), A Modet for Populations Reproducing by Fission Ecology 52: 380-38, Skipper. H. E, (1971). The Cell Cycle and Chemotherapy of Cancer. in The Cell Cle ‘and Cancer (R. Baverga. ed.). Marcel Dekker. New York. pp. 388-387, tannock. LF. (1968), The Relation Between Cell Proliteration and the Vascular System in a Transplanted Mouse Mammary Tumour. Br. J. Cunicer 22: 288-273, ‘Touchiya, H. M., Fredrickson, A. G.. and Aris, P. (1966), Dynamies of Microbial Coll Populations. Adv. Chem. Eng. 6! 125-198, Tubiana, M. (1971). The Kinetics of Tumour Cell Proliferation and Radiotherapy. Br: J Radiol, 44; 325-M7. Tubiana, M. (982). L. H, Gray Medal Lecture: Cell Kinetics Int. J Radiat. Oncol. Biol. Phys. 8: \471=1480, Webb. G. F, (1988) Theory of Nonlinear Age-Dependemt Population Dynamics. Marcel Dekker, New York Radi jon Oncology5 Altruistic Population Model with Sex Differences YING-HEN HSIEH National Chung-Hsing University, Taichung. Taiwan, Republic of China 1. INTRODUCTION ‘The evolution of altruism in ecosystems is an intriguing phenomenon in biology We define “evolution” as having a fixed ratio of the population showing one particular characteristic. such as performing altruistic acts. The problem is interest- ing and complicated because (1) the evolution of a self-sacrificing subpopulation seems to be a direct contradiction to the theory of evolution—the altruistic acts usually decreases an individual's fitness to survive, and (2) the altruistic act per- formed can be. at times, age-dependent, sex-dependent. space-dependent. or group-dependent. The literature on altruism includes Hamilton (1964). Maynard Smith (1965. 1980). Trivers (1971). Cavalli-Sforza and Feldman (1978). Feldman and Cavalli-Sforza (1981), and Akin (1984), Population geneticists in general believe that altruism was able to evolve in a community due to one of the three factors: reciprocal altruism, kin selection. oF group selection (see Akin. 1984). but the relative importance of these three factors, is still very much in controversy. The purpose of this work is not to ask why altruism evolves. but how. We will use simple mathematical models to find the circumstances under which altruism is able to evolve in certain communities but rot in others. It is easy to foresee that the conditions will be both physiological and environmental In previous works by the author (Hsieh, 1988, 1989a), a model of coupled Von Foerster equations of the form Dea.t) + ela peta.t) + mara. stat) + ¥(a.0] yilayx"(a.t) a © xa.t) + vat) Dy(a.t) + aslay(a.t) + maysta.d[xta.t) + sa.) xla)v(a. x(a.) x(a.t) + (act) 150, a€ [0.1] 2)6 Hsielt was used to study the possibility of evolution of altruism in a community with the numbers of altruistic and selfish individuals represented by separate differential equations. [For a discussion on models of this type. see Coleman (1978).| The conditions for the persistence of the altruistic group in the population was obtained, under the assumption of synchronized reproduction in a semelparous population ‘where reproduction occurs once in each individuals lifetime, In @ subsequent work. (Hyich, 19895), it was also shown that under the assumption of indiscriminate altruism for all individuals in the community. the evolution of altruism is necessiry but not sufficient for the persistence of the community itself ‘The Von Foerster equation is generally associated with one-sex models where only the female members of the population are considered. However. in some altruistic communities with intricate structure such as social insects. there is usually «difference in the performance of altruistic act between male and female. &s well as different reproductive rates for each sex. One example is the social bees (honeybees and bumblebees). where different forms of altruistic acts are per formed by the (adult) male and female of the species in a complex social structure Headed by the queen bee, whose sole function is to lay eggs (up to 20K) a dy) the community is comprised of female workers. which build cells for the eggs and collect enough pollen to feed the larva until it becomes a flying adult hee: male bees, which fertilize the young queens: and the parasitic (nonaltruistic) bees. whi do not build hives and develop in the cells of the host working bees. In Hsieh (1988), we get around this problem by prescribing a fecundity fune- tion that is general enough to account for the discrepaney in the sex ratio of succeeding generations. But to deal with the different levels of altruism. we introduce separate equations for the male and female members of the altruistic, group so that distinct altruistic functions and fecundity functions can be assigned. accordingly to different sex groups. However. we will not distinguish sex among. the yelfish members of the community’ since both sexes benefit equally from altruistic acts. Therefore. not only is the difference in ability to perform altruistic act considered. but we can also take into account the possible change in sex ratio, from generation to generation. The model in question is as follows: Daxta.t) + alana.) + maysla.).N (a0) _ narnia. nate. + a0} _ » Nat) ‘ Dela.) + asaya. + masta. N(a.0) = =0 ow Dela.t) + aladeta.t) + masta. N(a.0) Asta. + sa] 6. ae [0a] 6) Na voy. ¢ denotes the number of male altruists, female altruists, and nonaltruists. respectively, and N= x + y'+ : is the total population. In Section II we reduce the equations (3)-(S) into simpler system for the altruistic ratio so that in Section III we can yield analytical results on sufficientAlvruistie Population Model with Sex Differences 65 conditions for a fixed sex ratio in the altruistic group. Finally. we will run computer simulations for different cases described previously in the text and discuss the biological and ecological implications of our results. Although the main theme is sex differences in altruistic acts. the model can be used to describe any situations where any three subgroups of a population performing distinct social behavior (not necessary altruism). It also allows different levels of mortality and reproduction for each subgroups. Hence the analysis could have wider application in studying the complicated communities of semelparous insets. Ml, MODEL FOR ALTRUISTIC RATIO Let X,(a). ¥a(a). Z,(a) be the numbers of altruistic males, altruistic females. and selfish individuals, respectively. of age a at the nth generation. By the assumption ‘of semelparous population with synchronized reproduction, equations (3) be simplified (see Coleman. 1978. or Coleman and Hsieh. 197) as follows Xia) + a(a)X,(a) + m(a)X,(a)N,Aa) Hala) + Yala) y ) ean = nla (a) =0 (a 70) Kula) (6a) Ya) + ax(a)¥,a) + w3(a) ¥.ladNala) Kuta) + Yala) = (a) MO Val y (gy = 0 (6% (Naa) ' i Zila) + asla)Z,fa) + msla)Za(0)N,(2) ~ ya) ** yn ela) =0, Osast (6) Here N,(a) = X,(a) + ¥,(a) + Z,(a) is the total population number of the mth generation at age a. the a,’s are mortality functions. the 7's denote the effect of dispersal and migration. the ,’s are “altruistic functions” from [0. 1] + R which describe the effects of altruism on each group (hence a group that incurs a loss in fitness due to altruism performed may have a negative altruistic function). and the ‘generation length is normalized to unity We assume further that the rate of dispersal and migration is the same for all groups (i.e. m= =m). Ih we let X,fa) + Yala) fala) o Nata) be the altruistic ratio in the mth generation of age @ and Yuta sola) = ele) — 6) Xa) + ¥.(@) be the ratio of females among the altruists of the ath generation and age a. we can combine (6a)~(6¢) to get the corresponding equations satisfied by f(a) and gu(a) for 0=a= 1 and n €Z.66 Hsieh Pad ~ fla = fl@Malad ~ e(aigylad + Fulapgaalfnla) — Fala) fala] = 0 (a) gala) + gala ~ gaCa)lldstar ~ Fola)faCa] = 0 (9b) with Kila) = a(a) ~ ax(a) au dia) = asta) ~ ala) ap Fula) = yala) ~ yal) «2p Fa) = yela) ~ yaad a3) ‘To equations (9). we add the fecundity functions Fy and Fy for the altruistic ratio and the ratio of females within the altruistic group. respectively. Fe (0) = Fulfal)) (1s) 00> Fag) (140) The role of female reproduction as an altruistic act does not appear. althou it certainly qualifies as altruism. However, a consideration of female reproduction can be incorporated into the altruistic function for female altruists: hence the females claysified in the selfish group are those females that do not perform altruistic tasks and reproduce no offsprings. We can then look tor sutficient con- ditions for the evolution of altruism as well as a fixed ratio of trvists in the population using the fixed-point theory developed by Coffman and Coleman (1978) male Ill ANALYSIS OF THE MODEL Equations (9) can be rewritten as Pale) + pil Fale). ale)-a) = 00 «say eile) + pal Fala). alata) = 0. neZ (usb) where pult.y.a) = —x(1 = xyfeia) ~ ds(ayy + ¥4(ayay — ¥(arx] (ea) pals. vod) = C1 ~ v)feista) — Fa(aday (166) dare the loss functions for f,, and g,. Since the conditions for evolution of altruism were given in Hyich (1988), the first question we pose is the following: Suppose that the altruistic ratio for adults at cach generation f',(1) becomes fixed and positive at euch generation as 1 — What is the ratio of female altruists in the altruistic group” It we defined the generation growth function G: [0.1] [0.1] for the adult Female ratio among altruists g,(1) as follows: Kall) = Glee MDD EZ an where the initial value g,(Q0) of the IVP (15b) is given by (4b), the conditions forAltruistic Population Model with Sex Differences or the existence of a fixed point of (17) are given as follows: Theorem 1. Suppose that 00 (b)_ In F3(0) > fi[es(a) + |¥2(a)|] da there exists a unique fixed positive point g = G(g). If. on the other hand, Fx(0) = Oand ine) Satisty the inequalities in (20) and (22)-(24). If the fecundity functions F, satisfies either (a) FQ)>0 for ’=1,268 Hoich tb) In FO) >f ay(a) da and we) InAs [ altruism will evolve with a fixed ratio of female altruists in the altruistic group. Conditions (22)~(24) together with (a) or (b) and (b’) in Theorem 2 give sulficient conditions for evolution of altruism in a population moxtel with sex differences, We will now compare these conditions with the corresponding results for a one-sex model. We recall that the convexity conditions for evolution of altruism in a one-sex model (Hsieh, 1988) are a) + |ysta)l] da ila) = wa) = aay (27) is the ditference in mortality of selfish group (2) and altruists (#1). and Fla) = yl) — Yl) es) is the difference in altruistic functions for the selfish members (,) and for alteuists (a) Condition (25) says that the mortality function of altruists must be no worse than that of selfish group at any time. We note that (22) implies en(ay = maxfay(a). asta}. « €[0. 1] ow which means the mortality rate of the selfish group is no lower than either the male or female altruists at any time ~exactly what we have in (25). ‘Conditions (23) and (24) can be rewritten respectively as Fala) = marl (a), de(a) — (a) + F(a} (0) and ~2¥ ala) = marl (ale éala) ~ la) ~ 29 lO}. aE [OA en Hence if ¥.(a) = a(a) = 0, aE [0,1] (ie. we do not distinguish sex among the alteuists), then (30) and (31) together imply: that aa). a E[0.1] (2) (a) = 9a) which is different from (26). It is quite reasonable, since evolution of altruism with a nonzero fixed female altruistic ratio is not altogether the same as evolution of altruism per se. But it is interesting that condition (32). although different from (26), has the similar implication that the net altruistic benefit. whether neyative OF not, should be of moderate magnitude for altruism to evolve. Thus the moral lesson is: There must be a certain balance in nature!Altruistic Population Model with Sex Differences Cc [As a final remark before we get to numerical examples. it is interesting 10 note that when F(a) = d3(a) = 0. (30) and (31) yield = fala) = -&,(a) 3) which is exactly condition (26). ‘One possible interpretation of condition (33) is that when the whole altruistic group has the same mortality function, which is no greater than that of the selfish group but the male altruists benefit as much from altruistic acts as the selfish group. the female altruists can still persist whether or not they are benefitted more from the altruistic acts than the male altruists—as long as the total difference in fitness incurred is of moderate magnitude. In fact, if we interchange sex roles (let X, denote females and Y,, denote males), we then have a situation where male altruists, although possibly making a greater sacrifice in their altruism, will still evolve. Hence the model allows us to study altruistic communities without prejudg- ing which of the three groups is making the greater overall sacrifice 0.2 ° 0 2 2 o 2 GENERATION “=-ALTRUISTIC FREQUENCY "a -ALTRUISTIC FEMALE RATION Figure 1 Altistic frequency f4(1) (circles) and fraction of female altruiss 4 (1) (tri angles) at adulthood with f(0) = gO) = 12: Fila) =x, Fel) = 2, ai(a) = a sa) = =alS, Fula) = Fala) = a0,70 Haieh IV. EXAMPLES AND DISCUSSION In this section we use simulated numerical examples to illustrate our results. I all eases, we use t fourth-order Runge-Kutta method with f= 0.08 to solve f(1) and g,(1) for = 1.2.3... using equations (9a) and (9b) simultaneously. We also use the initial values of f(t) = gu(t) = 00.5 for all simulations. In F let Fis) = 8. FG) = 1/2 (ie. the altruistic group is equally divided between males and females). a,(a) = a. do(a) = ~alS. ¥1(a) = ¥a(a) = a/10, s0 that the functions satisfy the conditions in Theorem 2. As we can see from community approaches a pure altruistic society with a female ratio of 0.7088. A simple biological explanation for condition (20) is: For female altruists to evolve. it is sufficient that they be more competitive physiologically Jasta) = «(a)| and that this competitive edge outweighs the edge that male altruists have as bene: ficiaries of altruistic acts if. indeed. the males have such an edge [7:(«) ~ y.ta)] ‘The convexity condition in Theorem | is sufficient but not necessary. as we shall demonstrate in the next example. For Fig. 2 we let Fux) = cand Pals) = 172 as before, but we let (a) = F2(a) = ~ a. dla) = dla) = ~ al. The functions preseribed here do not satisfy either condition (20) oF the convexity eondition in 02 : — —_—_l ° 10 2 0 “ 30 GENERATION ‘e--ALTRUISTIC FREQUENCY ‘ALTRUISTIC FEMALE RATION Figure 2 Altruistie frequency f,(1) (circles) and fraction of female altruists (1) ttt Angles) at adulthood with f,(0) = gi(0) = U2: Fils) = XP) = 2, lad = ata) = ~alS yulay= FetaAltruistic Population Model with Sex Differences 7 10 0.8 FREQUENCY 0.4 0.2 0 0 10 2 Fy o GOIERATION (= -ALTRUISTIC. FREQUENCY “A-ALTRUISTIC FALE RATION Figure 3 Altruistc frequency f,(1) (Grcles) and fraction of female altruists gy(1) (tr- angles) at adulthood with fu(0) = gu(0) = M2: Fy(x) =x, Fals) = ¥. (a) = al. ds(a) al, ¥4(a) = af10. ¥s(a) = (1 = 2ays ‘Theorem 1. but as one can see from Fig. 2, we still have evolution of pure altruism in the population with a fixed female ratio of 0.6410 as n —> =. The most interesting case in nature could be the instances where altruism evolves in a community with the altruistic act performed by solely one sex (male or female). The following example illustrates that such a phenomenon is possible im our model. In Fig. 3 we let Fi(x) =x, F:(0) =». ila) = al5. c(a) = a! 10. Fala) = ali, ¥2(a) = (1 ~ 2a)/5. (The given fecundity function F: is probably not very realistic biologically. but is nonetheless possible.) Then altruism will evolve with an all-female altruistic group. One example with all-male altruistic group could easily be obtained by interchanging X,(a) and Y,(a) in the original model in (6) In conclusion we point out that itis possible for g,(1) to converge to a nonzero fixed point without evolution of altruism. For Fig. 4 we let F(x) = x. F:ly) = ys 4i(a) = =a, (a) = ~al5, and ¥,(a) = ¥2(a) = al 10; then the altruistic group will eventually consist solely of females, but the altruistic ratio will go to zero. So in reality. the number of female altruists will also go to zero and the altruists become2 Hsieh ° 10 2 30 @ 0 GENERATION ‘eALTRUISTIC. FREQUENCY ‘=-ALTRUISTIC FEMALE RATION Figure 4 Altruistic frequency f,(1) (circles) and fraction of female altrusts ga (1) (ti: ingles) at adulthood with f,(0) = g\(0) = U2: Fs) =. FAW) =. ela) =~ a. eka = ~alS. F(a) = Fa) * a0, extinet. Hence a fixed ratio of female altruists is only meaninglul given that altruism will persist ACKNOWLEDGMENTS ‘This research was supported by a grant from National Science Council of ROC (NSC 78-0208-MO0S~03) for which the author iy grateful. The author alyo wishes to thank the referees for their comments and suggestions. REFERENCES Akin, E. (1984), Evolution: Game Theory and Economics, ia Population Biolog (8. Levin. d.). American Mathematical Society. Providence, Rt.. pp. 37-01 Cavalli'Sforea, L. L.. and Feldman, M. W. (1978). Darwinian Selection sind ~ Altruism. Theor. Popul. Biol. 1: 268-280 Coffman. C. V.. and Coleman, B. D. (1978). On the Growth of Populations with Narrow Spread in Reproductive Age: Il, Conditions of Convesity, J. Math, Biol. 6: 268-W3,Altruistic Population Model with Sex Differences 73 Coleman, B. D. (1978). On the Growth of Populations with Narrow Spread in Reproductive ‘Age: I. General Theory and Examples. J. Math. Biol. 6: 1-19. Coleman, B. D., and Hsien. Y. H. (1979). Theory of the Dependence of Population Levels ‘on Environmental History for Semelparous Species with Short Reproductive Seasons, Proc. Math. Acad. Sci. USA 740): 5407-S310. Feldman, M, W.. and Cavalli-Sforza. L. L, (1981), Further Remark on Darwinian Selection and “Altruism.” Theor. Popul. Biol. 19: 251-240), Hamilton. W. D. (1964). The Genetic Evolution of Social Behavior. J. Theor. Biol. 7: 1~ 16. Hsieh, Y. H. (1988). On the Evolution of Al Comput. Model. 11: 472-475. Hsieh, Y. H. (1989a). Evolution of Altruistic Communities. COBNOSES. 43): 145-147. 1989) Hsieh, Y. H. (19896). Persistence of Altruistic Community under Indiscriminate Altruism. Proc. NSC (ROC) (+0 appear). Maynard Smith, J. (1965). The Evolution of Alarm Calls. Ain. Nat, 99: 89-68, Maynard Smith. J, (1980), Models of the Evolution of Altruism. Theor. Popul, Biol. 18 51-159. Trivers, RL. (1971). The Evolution of Reciprocal Altruism. Q. Rev. Biol. 46: 3S-S7 uism in an Age-Structured Population. Math6 Remarks on an Epidemic Model with Age Structure MICHEL LANGLAIS Université de Bordeaux I, 33076 Bordeaux. France |. INTRODUCTION We exhibit the threshold parameters for a global population and for an infective class in two fairly simple epidemic models with age dependence. The first main feature in our mathematical models is a nonlinear birth and death mechanism that depends on the total population: nevertheless, its structure is quite simple, making the mathematical analysis easier and avoiding the lengthy technicalities to be used in a forthcoming paper of Langlais, where general structure conditions as in Langlais (1988) will be imposed. Second, we do not assume that the global populs. tion has already reached a steady state: It stabilizes together with the infective class. When the global population becomes extinct, the disease is obviously eradi- cated, but when the global population goes to a nontrivial steady state, eradication of the disease depends on the size of such parameters as the contact rate. the recovery rate, and the vertical transmission rate. The models we consider are actually closely related to an earlier model of Webb (1985) and to those called intracohort and intercohort in Busenberg et al (1988): we keep the same names. See also Bailey (1975), Waltman (1974). and ‘Murray (1989) for a generic derivation of epidemic models. ll, THE MODELS We first describe the mathematical model for the global population and next. the two epidemic models ‘A. Dynamics of the Global Population We denote by u = u(t.) the distribution of individuals in a single species popula: tion, later referred to as the global population. having age a > 0 at time ¢ > 0: the total population at time ¢> 0 is 757% Langlais (tay da a a) The evolution of 1 as time evolves is given by the well-known Sharpe. Lotka. MeKendrick. and von Foerster set of relations, fact a= Mad. 10, a0 W(O.a) = ula), a> 0. ut =f BUautradda, 11> 0 Here s(r.a) is the supply of individuals of age a >0 at time 10, uy the initial distribution of individuals at time ¢= 00 and B(1.a) is a maternity function such that the last relation represents the birth mechanism. [More details are given in Gurtin (1973), for example] We assume that the supply y is due mainly to death and takes the following form: Mea = = wPO.aulta). 0, a> 0 (see Gurtin and MacCamy, 1974, or Webb, 1985): (p.a) is the death modulus at age «> 0 when the total population is at level p 201. For the sake of simp- lification we take Hip.) = nAP)+ Hala). — p20. a>0 We choose the fertility function to be Bl) = BAPUN) Bla), 0. a= Putting the ¢ assumptions together, the evolution of 1 is run by the set of equations a+ at Lal PO) + aad) =0, 10, a0 a) = nla), a>0 e 10.0) = BPO f Buwuiayda, 1 >0 When B,(p) = 1 this is the separable model of Busenberg and Lunnelli (1985) B. Intracohort Model Assume that a very mild epidemic disease is spreading through the population under consideration. We may split the global population into two subelasses: the ws and the susceptible class. having i(t.a) and s(1.«) for respective distribution with respect to the time and age variable; by construction, infective uta) = i(t.a) +s(.a). > 0, a> a A very simple model describing the spreading of this mild epidemic disease in the global population and taking into account the interactions between the two classes di + Ot + [we PUD) + Malad} = Yladis~ Bi, > 0 a>Epidemic Model with Age Structure ” as + ays + lalPO) + pala]e=—ylayis + 8 0. a> ) together with the initia distribution in such class at time 1 = 0 10.0) = inlays s(0.a)= sila), a>0 and the vertical transmission of the disease to offspring through birth ate 0) = ea.(Po) | Bylayitteayda, 11> 0 (1.0) = (0) =i). 0 6) Here y. 6. and ¢ are nonnegative parameters: y is the contact rate (i.e.. yi is the probability that a susceptible individual will become infected by contact with an infective of the same age). 8 is the recovery rate, and ¢ gives the fraction of offspring produced by the infective class that are actually infective. Note that only the infective class can produce infective offspring. Upon adding relations (4). (5) and (6). we recover equation (2) C. Intercohort Modet The main difference with the intracohort model lies in the possibility that a susceptible will get infected by any infective individual. The probability for a susceptible to become infected by contact is now wt) wieenty= fod Hence (6) does not change. while equations (4) and (5) read AG + Ai + [ue(PLO) + wa l@)]i= ylayl(nys = Bi, > 0. a> 0 om (oO ays + 84s + [e(PO) + HelIs= —Ylayl(Os + i. The form of the interaction terms we have in equations (4). (5). and (7) is used by many authors: see. for example, Bailey (1975), Waltman (1974), Webb (1981), Busenberg et al. (1988), and Murray (1989), and their references. ll, BASIC ASSUMPTIONS We first suppose that 8. j,. and y are nonnegative functions and further that Oxmlstetx 0=p,la)sf<+ fora=0 Osyasy0 o Mal=y>0: ay= 520 The dynamics of the infective class are governed by the following equations: ait i+ La MAP = YEH] HO 0 ia uo (a), @>0 BB (PIN. 60) «0 ieayda, 10 a In a fortheoming paper we shall show that there exists a unique suitable solution bof this problem verifying Daa sulna. 10, a>0 ay ‘A. Stabilization for the Global Population Stabilization results for the global population are well known in the literature: see Webb (1985), Busenberg and funnelli (1985), oF Langlais (1988), for example, We secall these results here briefly. Further estimates, as well as the outti ‘are supplied in Section VILA to handle the stabilization of the infective class. Using simplifying assumptions (9) we find upon integrating equation (2) from a= (to a= +» that the total population PC) satisfies the nonlinear ordinary differential equation 2 =[BBAPI~ so APP. oO P10) = poo of a proof, We may conclude from the monotonicity ssumptions concerning pant 4, that it B~ £50. then Pip goes 100 ast tends to +7 white it B~ 0, then P(r) goes to pay ftends to +Epidemic Model with Age Structure 9 where p* is the unique postive root of the equation BBA p*) ~~ Help") = 0 (12) Obviously. i ~ jis the threshold parameter for the total population. Thus Lemma 1. If B= si 50. then u(t, -) +0 in L'(0,) as ¢ goes to +. The trivial equilibrium is actually exponentially asymptotically stable in that case and the convergence is uniform on each compact domain [0. A]. With more efforts we have Lemna 2. Set (a) = [+ pape) If A ,£> 0. then u(t.) —+p*¢(-) uniformly on each compact interval (0. A] as ¢ goes to += Remark. & steaightforward calculation shows that when 6 ji>0. the pair (p*.p%e) is the unique nontrivial and nonnegative steady-state solution for (2). namely the unique nonnegative solution of ae t Lit w(gy= 0. a>0 40) 94 B. Eradication of the Disease We first find the threshold parameter for the infective class. If 6~ #= 0. using Lemma I and inequality (11). we get i(t,-) +0 in L'(0, *) as r goes to +2, Hence from now on we assume that B-a>0 3) Integrating the first equation in (10) over all ages shows that / satisfies the differential equation Al = [eBBAPC) + YPC) — taking the initial value (PU) ~ yl) ~ BM. 10 10) f infa) da Let /* be given by po ~ BBA p") + yp ~ = pel p y 1K follows again from the theory of dynamical systems [see Walker. 1980. for example] that either mya 8 aay 0 and 11) +0.as goes to +80. Langlais 10 and +I? ast goes to + ‘Asa first consequence we have Theorem 1. When I* is nonpositive the solution of (10) goes to 0 in L'(0,%) as 1 goes to +2, Remark. Note that [* can be made nonpositive by increasing 8 and decreasing and ¢; that is, the disease is eradicated upon either improving the recovery rate ‘or decreasing the contact and the vertical transmission rates (or both). More work is necessary for analyzing the case /* > 0. C. Nontrivial Infective Equilibrium If we assume (13), a nontrivial equilibrium for the intercohort model is a nonnega- tive solution of aaj + [a+ we(P]i= vp ea) —j]- 3. a>0 HO) = BB p VS cs) en When J is known, j satisfies the linear differential equation a +L * mph) +S + i= pela). a >0 HO) = «BBA p*)F Using the expression of ¢ given in Lemma 2, we may solve this equation explicitly: substituting this exact solution back into we obtain after some calculation that J is @ positive root of Syd + Alla + wel pt) + YS + 8] = JLo + BML B B.C") + yh = 0 and J = ~ 8/y have the wrong sign. The only remaining possibility is €BB.(p*) + yp* = wel p*) = 8 Y which is equal to 1* when it is positive. Hence we have proved Lemma 3. Assume that (13) holds. There is a unique nontrivial infective equilib- rium in the intercohort model if and only if /* given in (14) is positive Remark, The equilibrium is explicitly found as Ha) = «BB. p*expi—[yl* + w+ wel p*) + da} 4 ELE wD] yes ~ expt [yl* + + wel") + ]a)) (expt [a + (P10)Epidemic Model with Age Structure 81 D. Persistence of the Disease Note that /*> 0 implies that A ,Z > 0. In Section VI we shall prove Theorem 2. When I* is postive the solution of (10) is uniformly convergent on each compact damain [0.14] to the nontrivial equilibrium, Remark. When I* is nonpositive, the same generic proof shows that the conver: gence of {0.0 is uniform on each interval (0, A] V. INTRACOHORT MODEL For this second model we impose the following hypothesis on the parameters: sala) = vy(a) +H > 0: there exists a number A, < + such that supp By and supp », are contained in [0. Ai}: (a) Br EC! Bila) >0 in (Az. Ai) for some Ay: {s(a) > 0 in some (Ay. As) (0. Ai) No sign condition is required for v,. provided that y.,(a) 20 for a= 0. Similarly, the special form of 4, is irrelevant as long as we only describe the results of this, section. However, itis needed in their proofs in some places. ‘The dynamics of the infective class in the intracohort model is now run by the set of relations B+ dai + (ana) + we(PO|E= ylahilu~ i] = Bi, 1 >0, a >0 i(0.a)=in(a), a >0 10.0) = aPC) [ Bu(ai(t.a)da, 1 >0 an Again there exists a unique suitable solution of this problem satisfying OSi.a) Sula). 1>0, a> (1s) A. Stabilization for the Global Population Following the plan of Section IV.A, we make a preliminary calculation in order to gain some insight. A steady-state solution for equation (2) is a nonnegative solution of dav + [na(a) + we(Q)] = 0. Jf ve)da=920 (9) a>o 40) = awa Bua) Ka) da Define aeyeo{-[mierde} 220rs Langlais which is the probability for an individual to survive at age @ when the total population is very small. By direct computation we find that a nontrivial nonnega- tive steady-state solution exists if and only if it reads v(a) = qa) where wa) = me da and q is the unique positive root of aw Bula) mayen da ey The monotonicity assumptions on B, and 42. and (16) imply that there is at most ‘one nonnegative root for (20): further. this root is positive if and only if the right- hand side of (20) is bigger than 1 for q = 0. namely if and only if the root of the so-called characteristic equation f Bula) nade "da ev is positive, Let r* be the real root of (21): then either r* <0 and there is no nontrivial solution of (19) or r* > 0 and there is a unique nontrivial solution of (19) We then have as in Langlais (1988): Lemma 4. If (21) has no positive root, then u(t,-)—+0 uniformly on {0 A] for each A >0 and P(t) +0 as 1 + Lemma 5, If (21) has a positive root. let p* be positive root of (20), and let nla) eee $5 maye da Then u(t.-) + p* (-) uniformly on each (0, A} and Pl) — p* as ¢—+% These two stabilization results are actually consistent with those we found in Section IV.A. (a) B. Nontri ial Infective Equilibrium We may assume that r* is positive: otherwise, we know from inequality (18) and Lemma 4 that i(r.") goes to 0 as ¢ goes to +% and it easily follows that there is no nontrivial infective equilibrium for a total population at level 0 An equilibrium in this intracohort model corresponding to @ total population at level p* is a nonnegative solution of aaj + Lala) + ae(W™ = Ylayst* ea) ~j]~ 8. a >0 HO) = €B. wwf B,(a)j(a) da (22)Epidemic Model with Age Structure 83 Set 2a) = y(a)p* ea) ~~ pala) ~ wep?) The solution of the nonlinear differential equation z(a) ~ y(a)j]. a >0 isexplciny given by iy ——— SPU z(@) da] _ 23) 1+ iy Si la) explfi 2(a) do] da For the condition at a = 0 in (22) to be satisfied we must have either jj, = 0 or . xpLfiiz(a) da} A sf (a) Fi zta) de] gg = ey Bet") |, Pol TF Fyay exp eto) do] da The left-hand side of this relation is decreasing with respect to the parameter jy and it goes 10 0 as jy goes to +». Hence for (24) to hold for some positive jy its necessary and sufficient that is left-hand side be larger than | at jy = 0. or introduc- ing ila) Hyde) easton [ Bula) ma) x exp{oe [1a e(a)da~ (6+ ne(}o} do 5) equation (24) has a unique positive solution if and only if (y.8.€) > 1 Lemma 6. ‘There is a unique nontrivial and nonnegative infective equilibrium in the intracohort model if and only if 1(y..5,€) > I: it is given by relation (23) where j isthe positive root of equation (24), Remark. I(y.8.«) > | implies that r*, the root of equation (20). is positive C. Stabilization for the Infective Class ‘The analysis is slightly different from the one for the intercohort model. The uniform boundedness of u with respect to = 0 and a= 0 deduced from equation (2) once P is known to be uniformly bounded in time together with relation (18) shows that / is also uniformly bounded on 10 and a = 0. Define the a-limit set on) = (4) EL (R) such that there is (ta) 0 > +22 and (tc, ») —+(-) weakly in L?(0, A) for each A > Oas k > +} This makes sense and we have Theorem 3. Any clement of w(iy) is a nonnegative steady-state solution, and further the convergence is again uniform on each compact domain (0. A].84 Langlais ‘The outline of the proof is given in Section VIL. As a first consequence, we get. using Lemma 6, Theorem 4. Assume that I(y,8,€)= 1: then the solution of (17) tends to 0 uniformly on each compact domain [0. A] ‘The second ease requires a more specific treatment. Theorem 5. Assume that I(y.8.e) >: then the solution of (17) converges Uniformly to the unique nontrivial equilibrium, which is therefore stable Proof: Because the w-limit set (iy) is connected (Walker. 1980) we have either «(is) = {0} oF (iy) = {the nontrivial equilibrium). We proceed by contradic- tion and show that the frst case is not possible. If i(t.a) tends toward 0, then given any fixed As > A, and any small positive number 7 there is a time (7) >0 such that in the domain ¢ > T(m). 0 | implies that the root of (21) is positive and we may apply Lemma 3 Using the weak maximum principle we get that the solution of (17) is bounded from below in the domain 1> Tm, O TKn) Proceeding as in Langlais (1988, Sec. 4.3), the last wo conditions in (16) on B, and jy imply that 2(T().a) = (T(m).a) > 0, O< a < Tm) Hence the general theory of linear age-dependent population dynamics problem (see Feller. 1941, or Webb, 1985) asserts that the ultimate behavior of = as « tends to + depends on the size of My. 8. €m) = «Bop — of By (a) wa) a) da ar= cule [aerate do= ts me=af-2 [et] For =0 we find I(y.6,€,0) = My.5.€) > 1 by assumption: @ continuity argument shows that for 7 small enough we still have [(y. 5, €. 9) > 1. It follows that <(F,+) goes to + as F goes to +. So does j because it lies above =. whichEpidemic Model with Age Structure 85 is impossible because / is bounded. Hence w(ip) is reduced to the nontrivial steady state. VI. PROOF OF THEOREM 2 The proof follows the main ideas of Langlais (1988) with the simplifications pro- vided by (9) as in Langlais (1989) ‘A. More Estimates on the Global Population Returning to Section 1V.1, we know that P(s) is uniformly bounded in time: OSPHsM0 29) Osu(ra) It follows from (2) that 4,u + 4,u is bounded in ¢>0 and a >0 but we cannot have separately a,u and a, bounded in 1>0 and a>0; otherwise, u would be continuous. As it was already pointed out in Langlais (1988, Sec. 7). we first have Au and a,u lie in L?(00, au EL? ((A,*) x (0A) 0 [ (aan? (oayda= Mya) < + where Ms(A) does not depend on 1> A86 Langlais A similar but more elaborate proof is given for the infective class in Lemma Bi just set e= Le y in its proof (o get that of Lemma 7. Remark. The proof of Lemma 2 is now a consequence of Lemma 7 (see Section VLC with €= 1. 6= y = 0) as in Langlais (1988) B. A Priori Estimates for the Infective Class From Section 1V.8 we may deduce that (1) satisfies OSS M0 G0) Furthermore, using (28) we shall verify in Section VIII that Lemma 8. a1 L2(0.*) We now are to extend Lemma 7 to the infective class Lemma 9. For any A> (0, we have af L(A.*) x (0A) 0 f aittarde= Maye o> en where Ms(A) does not depend on 1 > A. Proof Inequality (29) together with (11) show O< ita) = M, (32) ‘Again / is discontinuous along the characteristic line ¢ = a and we still have 04 and. agi lie in L*(0< a <1< T) for each finite T. itferentiating equation (10) with respect to the time variable in the domain 1 a, we find that 26.4) = ailtea) is a solution of ae + ae + E+ mCP) + 8+ yl = = w(PC)) API + ylki)au + y= Had. int >a 3) 2.0) = 24). 10 We readily see that 2 € L*(0,%). Fix Ay > 0, and let A < Ay < T. Taking the product of (33) with >. integrating by parts over the domain (TA) = (W.a).00}. Fixing A >0 we have from Lemma 9 and estimate (32) that max (| lasdeayida. max lr ad|) < +% ‘The semiorbits are equibounded and equicontinuous on (0. A] for each A > 01. We may now define the «limit set (ig) = GC), there exists 1g + +% such that i(t,.-) + jC) uniformly on each (0, A] ask -» +>} Following an idea of Langlais and Phillips (1985), choose p a smooth function satisfying ary=0.suppoc(-lt}e aspdr=t Let y be another smooth function vanishing after some A >0. Taking the product ‘of equation (10) with p(s ~ %) x(a). integrating by parts over ¢>0 and 0< a=. and setting += 1~ 4 we find after some routine calculations that for m > A +2 fone has Jf terete = tanta + 4 nelPCe + heals + nde da wif [fte-ine racahte +a +f X(A)BBAPT + A) OF + ICT + 4) de Gy We already know that Ply—opt. Kiya tt ast u(t.) p*¢(-) uniformly on (0, A] as ¢—> + ‘Assuming that f(0,.+) + j0-) uniformly on [0, A] as 44 + +288 Langlais we also have Hs + te.) fla) im L*((= 1.1) x (0, A)) as te > + because ai lies in L*((A, *) * (0. A)). With these convergences at hand, letting k go to + in (34), we sce that j is a solution of Ah + (+ mel P= YEU p* ela) — j]— 8. 0, a>0 The differential equation for P yields only APEL(O.T) — for cach T>0Epidemic Model with Age Structure 89. This is sufficient to get Lemna 10. For each finite T. #,u and au lie in L°(00. We first part of the statement is proved upon using forward time differential quotient techniques on equation (2) in 1 >a and the local (in time) estimates already obtained. We now have Lemma I, Given any A>0, au LUA.) (0.4) 05 [GF ia) das Maa) < + where Mx(A) does not depend on 1> A Sketch of the Proof: Compared to the proof of Lemma & we have to work With the two equations derived upon differentiating with respect to the variable 1 = first, equation (17) for u. and next, equation (36) for P. expressed in the form 8,P + [pel P) + ar { [Be(P)B.(a) ~ ayfu(t.ayda, 10 PLO) = po {see hypothesis (16)]. The proof is as in Langlais (1988, Sec. 7). to which we refer: here, kay is represented by J B. A Priori Estimates for the Infective Class From the boundedness of w and inequality (18) we get Ose M<= 120. 020 Next, proceeding as we in Lemma 10 for u, for we still have