MARKOV PROCESSES CHARACTERIZATION AND CONVERGENCE STEWART N. ETHIER and THOMAS G. KURTZ JOHN WILEY & SONS New York Chichester Brisbane Toronto Singapore"ANOTE TO THE READER This book hasbeen electronically reproduced from Algal information stored a Jobin Wiley & Sons, Ine ‘Weare pleased that the use of this neve technology will enable us to keep works of enduring scholay ‘ale in pint as longa there i aregonable demand for them. The content ofthis book siden to previous printings [Cony © 1986 Yon Wik & Son, Ie ih reserved Publ sangouly in Canada, Reproduction or treatin fy pat of tht work $end peed Sn 17 rate ofthe copyright ower ula Requests for ‘Termin rurther information shoal be adres 0 the Permisons Department, Jobe Wily & Son In teary of Congress Catalin Pabstion Data “Marko processes (ie soe probity and mathemati! sat) Bibhogaphy: nse nde UE Matow procezes Kurt, Thomas G QAIKTEM 1s siea33 510TH WaNG-Ai-oniees Prine inthe United State of Ameria wosT6s4s PREFACE ‘The original aim ofthis book was a discussion of weak approximation results for Markov processes. The scope has widened with the recognition that each technique for verifying weak convergence is closely tied to a method of charac- terizing the limiting process. The result is @ book with pethaps more pages devoted to characterization than to convergenoe. ‘The Introduction illustrates the three main techniques for proving cone vergence theorems applied to a singe problem. The firs technique is based on ‘operator semigroup convergence theorems. Convergence of generators (in an appropriate sense) implies convergence of the corresponding, semigroups, which in turn implies convergence of the Markov processes. Trotters orginal Work in this afea was motivated in purt by diffusion approximations. The second technique, which is more probabilistic in nature, is based on te mar lingale characterization of Markov processes as developed by Stcoock and Varadhan. Hete again one must verify convergence of generators, but weak ‘compactness arguments and the martingale characterization of the limit are Used to complete the proof. The third technique depends on the representation ofthe processes as solutions of stochastic equations, and is more in the spit of classical analysis. Ifthe equations “converge,” then (one hopes) the sol lions converge. Although the book is intended primarily a8 a reference, problems are included inthe hope that it will lso be useful as a text in a graduate course on stochastic processes. Such a course might include basic material on stochastic Processes and martingales (Chapter 2, Sections I~6), an introduction to weak convergence (Chapter 3, Sections 1-9, omitting some of the more technical results and proof) a development of Markov processes and martingale prob- Jems (Chapter 4, Sections 1~4 and 8), and the martingale central limit theorem (Chapter 7, Section 1). A seletion of applications to particular processes could complete the course,[As an aid to the instructor of such a course, we include a fowehart for all ‘proof in the book. Thus, if one's goal is to cover a particular section, the chart Indicates which of the earlier results can be skipped with impunity (It also reveals tha the course outline suggested above is not entirely sel-contained) ‘Results contained in standard probability texts such as Bilingley (1979) or Breiman (1968) are assumed and used Without reference, as are results from measure theory and elementary functional analysis. Our standard reference here is Rudin (1974). Beyond this, our intent has been to make the book felf-contained (an exception being Chapter 8) At points where this has not feemed feasible, we have included complete references, frequently discussing the needed material in appendixes. ‘Many people contributed toward the completion of this project. Cristina CCostantini, Eimear Goggin, S.J. Sheu, and Richard Stockbridge read Targe portions of the manuscript and helped to eliminate « number of errors, Carolyn Bire, Dee Frana, Diane Reppert, and Marei Kurtz yped the manu Script The National Science Foundation and the University of Wisconsin, through @ Romnes Fellowship, provided support for much of the research in the book. ‘We are particularly grateful to our editor, Beatrice Shube, for her patience ‘and constant encouragement, Finally, we must acknowledge our teachers, Colleagues, and friends at Wisconsin and Michigan State, who have provided the stimulating environment in which ideas germinate and flourish, They con- tributed to this work in many uncredited ways. We hope they epprove of the result Stewart N. Een ‘Tomas G. Kuerz Sat Lake Cy, Wak Maem, Wiconte CONTENTS Introduction 1 1. Operator Semigroups 6 Definitions and Basie Propertics, 6 1 2 The Hille-Yosida Theorem, 10 3 Cores, 16 4 Multivalued Operators, 20 5 Semigroups on Funetion Spaces, 22 6 Approximation Theorems, 28 7 Perturbation Theorems, 37 8 Problems, 42 9 Notes, 47 2. Stochastic Processes and Martingales ° Stochastic Processes, 49 Martingales, 55 Local Martingales, 64 ‘The Projection Theorem, 71 ‘The Doob-Meyer Decomposition, Squate Integrable Martingale, 78 Semigroups of Conditioned Shifts, 80 Martingale Indexed by Directed Sets, 84 Problems, 89 Notes, 93gence of Probability Measures 95 1 The Prohoroy Metric, 96 2. Prohorov's Theorem, 103, 3. Weak Convergence, 107 4 Separating and Convergence Determining Sets, 111 5 The Space D0, «), 116 6 The Compact Sets of D,[O, <0), 122 7 Convergence in Distribution in Dy{0, oo), 127 8 Criteria for Relative Compaciness in D,{0, oo), 132 9. Further Criteria for Relative Compaciness in Dg{0, co), 141 10. Convergence to 8 Process in Cy{0, ©) 147 11 Problems, 190 12 Notes, 154 44 Generators and Markov Processes 155 1 Markov Processes and Transition Functions, 156 2. Markov Jump Processes and Feller Processes, 162 3 The Martingale Problem: Generalities and Sample Path Properties, 173 4 ‘The Martingale Problem: Uniqueness, the Markov Property, and Duality, 182 5 The Martingale Problem: Existence, 196 6 The Martingale Problem: Localization, 216 7 The Martingale Problem: Generalizations, 221 8 Convergence Theorems, 225. 9. Stationary Distributions, 238 10 Perturbation Results, 253 11 Problems, 261 12 Notes, 273, 5 Stochastic Integral Equations ws Brownian Motion, 275 Stochastic Integrals, 279 Stochastic Integral Equations, 290 Problems, 302 Notes, 305 {6 Random Time Changes 306, 1 One-Parameter Random Time Changes, 306 2. Multiparameter Random Time Changes, 311 3 Convergence, 321 7 0 " Markov Processes in 2%, 329 Diffusion Processes, 328 Problems, 332 Notes, 335 Invariance Principles and Diffusion Approximations a7 ‘The Martingale Central Limit Theorem, 338, Measures of Mixing, 345 (Central Limit Theorems for Stationary Sequences, 350 Diffusion Approximations, 354 Strong Approximation Theorems, 356 Problems, 360 Notes, 364 Examples of Generators 365 Nondegenerate Diffusions, 366 Degenerate Diffusions, 371 Other Processes, 376 Problems, 382 Notes, 385 1 2 3 4 5 Branching Processes 306 1 2 3 4 3 6 Galton-Watson Processes, 386 ‘Two-Type Markov Branching Processes, 292 Branching Processes in Random Environments, 396 Branching Markov Processes, 400 Problems, 407 Notes, 409 ic Models a0 ‘The Wright-Fisher Model, 411 1 2 Applications of the Diffusion Approximation, 415 3 Genotypic-Frequency Models, 426 4 Infinitely-Many-Allele Models, 435 5 Problems, 448 6 Notes, 451 Density Dependent Population Processes 452 1 Bxamples, 452 2. Law of Large Numbers and Central Limit Theorem, 455contents 3. Diffusion Approximations, 459 44 Hitting Distributions, 464 5 Problems, 466 6 Notes, 467 12 Random Evolutions 460 Introduction, 468, Daiving Process in a Compact State Space, 472 Driving Process in a Noncompact State Space, 479 Non-Markovian Driving Process, 483 Problems, 491 Notes, 491 Appendixes aa Convergence of Expectations, 492 Uniform Integrability, 493 Bounded Pointwise Convergence, 495 Monotone Class Theorems, 496 Gronwall's Inequality, 498 ‘The Whitney Extension Theorem, 499 Approximation by Polynomials, 500 Bimessures and Transition Functions, $02 Tuloes’s Theorem, 504 ‘Measurable Selections and Measurabilty of Inverses, 506 ‘Analytic Sets, 506 Relerences 08 Index sa Flowchart 529 INTRODUCTION “The development of any stochastic model involves the identification of proper- ties and parameters tha, one hopes, uniquely characterize a stochastic process. ‘Questions concerning continuous dependence on parameters and robustness tnder perturbation arise naturally out of any such characterization. In fact the model may well be derived by some sort of limiting or approximation argu- iment, The interplay between characterization and approximation or con- wergence problems for Markov processes is the central theme of this book. ‘Operator semigroups, martingale problems, and stochastic equations provide approaches to the characterization of Markov processes, and to each of these approaches correspond methods for proving convergence results, "The processes of interest to us here always have values in a complete, separable metric space E, and almost always have sample paths in D,{0, 0), the space of right continuous E-valued functions on [0, <0) having let limits ‘We give DLO, <0) the Skorohod topology (Chapter 3), under which it also becomes a complete, separable metric space. The type of convergence We are usually concerned with is convergence in distribution; that is, for a Sequence of processes (X,} we are interested in conditions under which limes BLS(X)] = ELIA] for every fe C(D,(0, 0) (For a metric space 5, C18) denoies the space of bounded continuous functions on S. Convergence i distribution is denoted by X,—> X) AS an introduction to the methods pre- sented in this book we consider a simple but (we hope) illuminating example. Mateo) weeded ayer eae ] nl) i2 swrmoDUCTION and let ¥, be a birth-and-death process in Z. with transition probabilities satisfying Q PUYE+ A =J+ 11 Y=] = mal2erann and ® poner iy=s—11 140 =1) = m(2)p + as h+0+, In this process, known as the Schldgl model, ¥,(0) represents the ‘number of molecules at time tof a substance R in a volume n undergoing the chemical reactions ® RoR, Ri t2R with the indicated rates. See Chapter 11, Section 1) ‘We resale and renormalize letting 0 XfQ= nor Kyy = 1), =O. “The problem is to show that X, converges in distribution to a Markov process X to be characterized below. "The first method we consider is based on a semigroup characterization of, R X.Let By = (nny — 1:9 € Z,)yand note that © ‘140 fl0) = ELIA) | X40) = 21 defines a semigroup (7)} on BLE) with generator ofthe frm 0 Gefla) = mA + ne MAaNL Sx + 29%) — fl} tse ~ So Be band yl (See Chapter 1) Leting lx) = 1 + 33%, a) 6 fe) = 4°) = 7, 4 Taylor expansion shows that 1G. f0) = GF) + wR A) — AOL + AEH +) —SO} of afd Ma) — WL + AMS — 04) — FD) sparta [a tre 8 763) de de eat [ase mr) = 18) (G+ wi +s) — G+ OID wwamoouction 3 for allf¢ C4) with f” © C{R) and all x € B,. Consequently, for such f (19) lim sup | 6 fe) — Gf) |= Now by Theorem 1.1 of Chapter 8, (AB LL Se C-—, 20] m C%R), Ge C-e, I} is the generator of a Feller semigroup {T\®)} on CC —co, co]. By Theorem 2.7 ‘of Chapter 4 and Theorem 1.1 of Chapter 8, there exists a diffusion process X corresponding to (T()}, that i, a strong Markov process X with continuous sample paths such that (12) ELI) | FE = Me - 9X) for allf¢ C20, 0] and t = 52 0.(FF = o(X(u): <3) "To prove that X,— X (assuming convergence of initial distributions), it suffices by Corollary 8.7 of Chapter 4 to show that (10) holds forall fin a core D for the generator A, that is, forall fin a subspace D of (4) such that A is the closure of the restriction of 4 to D, We claim that 3) Da (St aihge CURLS’ € CAR) (0) © CRI} is a core and that (10) holds for all fe D. To see that D isa core, fist check that (8) 9A)= [fe Cla, co} 0 C4R):f* € CR) XP € CL, 3} Then fet ¢ CHR) satis 4.) 5-5 2,2 and pot hs) = Malm). Given Fea) choose ge D with aig} ¢ C(@)and U — a) € CR) and define Co) =f) 00+ [UF av ‘Then fy + 9 € D for each m, fn +9» f.and Oly + 9) Gf “The second method is based on the characterization of X as the solution of ‘martingale problem. Observe that a9) sore) ~ [6 f0e40) ds is an {9F}-martngle foreach fe (E) with compact support: Conte auenty some subsequence (X,}converees in dsb otha, by the Sntnous mupring theorem (Coralia 19 of Chapter 3) and Problem 7 of Eup, on sexe) ~ | epexe as4 tmooucrion is an (#f)-martingale for each f¢ C3(M), oF in other words, X is a solution of the martingale problem for [Uf Gf):fe CH)}. But by Theorem 23 of Chapter 8, this property characierizes the distribution on Da[0, 0) of X. ‘Therefore, Corollary 8.16 of Chapter 4 gives X= X (assuming convergence of {nial dstibutions), provided we can show that as lim Pf ixizebea T>0 Let g(x) = ef + 7%, and check that there exist constants C,, > 0 such that 6,9 0, and lim. ipsa Cy < co Letina 2 021 (0) | = ewe have 9) * int o()P{ sup 1X401z 8 it oo, an} 5 Hleap (~G,f6yaA Moldtae ATI 5 FLoAX,(00 by Lemma 32 of Chapter 4 and the optional sampling theorem. An additional {tila} assumption on the intial distributions therefore guarantees (18) “Actually we can avoid having to verify (18) by observing that the uniform convergence of G, to Of for fe C2(R) and the uniqueness for the limiting snartingale problem imply (again by Corollary 8.16 of Chapter 4) that X,-~X in Distd, 2) where R° denotes the one-point compactification of R. Com vergence in Daf, 2) then follows from the fact that X, and X have sample paths in Daf, 2) Both ofthe approaches considered so far have involved characterizations in terms of generators. We now consider methods based on stochastic equations. First, by Theorems 3.7 and 3.10 of Chapter 5, we can characterize X as the ‘unique solution ofthe stochastic integral equation. 20) X(Q) = X(0) + 2/29) — f X(s) ds, where W is a standard, one-dimensional, Brownian motion, (In the present Txample, the term 2,/21/() corresponds to the stochastic integral term) A Convergence theory ean be developed using this characterization of X, but we ddo not do so here, The interested reader is referred to Kushner (1974). “The final approach we discuss is based on a characterization of X involving random time changes. We observe first that ¥, satisfies can r4o = 100+. (x [atertron) = wv ater #), wmmopucton 5 whore N, and Ware independent, standard (parameter 1), Poisson processes. Consequently, X, satisfies 2) xem xsoy 09H (me [4 400 ) n(n [nat e944) s+ [a= mtb MO dy, wvnre 14d) = N su) — uw and (a) = N-(0) —w are independent, centered, ‘sandard, Poisson processes. Now itis easy to see that ey (8 8), 298 (0) Ws WO, where W, and W_ are independent, standard, one-timensional Brownian ‘motions. Consequently, if some subsequence {X,} converges in distribution to X, one might expeot that ee fae (Un this simple example, (20) and (24) are equivalent, but they will not be so in general) Clearly, 24} characterizes X, and using the estimate (18) we conclude X,-©X (assuming convergence of intial distributions) from Theorem 5:4 of Chapter 6 For a further discussion of the Schldgl model and related models soe Schldgh (1972) and Malek-Mansour et al (1981). The martingale proof of convergence is from Costantini and Nappo (1982), and the time change proof isfrom Kurtz (19810). Chapters 4-7 contain the main characterization and convergence results| (with the emphasis in Chapters § and 7 on diffusion processes). Chapters 1-3 contain preliminary material on operator semigroups, martingales, and weak convergence, and Chapters &-12 are concerned with applications.1 OPERATOR SEMIGROUPS Operator semigroups provide a primary tol inthe study of Markov pro- (Sites n this chapter we develop tbe bse background for thir ty andthe Sxience and approtimtion revs that are eed ltr asthe basi or xi fone and upprotation theorems for Markov process, Seton 1 ives th Sn ton, and Seon 2h He Yona hear, wih characters the operators that ae goveafors of semigroups. Section 3 concerns the probiam of verjng the hypotheses ofthis theorem, and Sections 4 and 5 ae Exvoted to geneazatons othe concep of the geerator. Setions 6 and 7 preset the apronimation and perturbation eu “Throughout the chapter L denotes aoa Banach space with norm 9-1 1. DEFINITIONS AND BASIC PROPERTIES |A one-parameter family {T(): +20) of bounded tinear operators on a Banach space L is called a semigroup if T(O)= I and Tis + #) = T(9)T() for all 412 0.A semigroup (T(} on Lis said to be strongly continuous iim, .o TWF for every f¢ Ls iti said to be a contraction semigroup if | TU) 1 for all 120. ‘Given a bounded linear operator B on L, define an en Laem 120. 6 1. EMMONS AND BASE PROFERTES. 7 A simple calculation gives e#* = e'¥¢ for all s.t > 0, and hence {eis @ semigroup, which can easly be seen to be strongly continuous, Furthermore wwe have a : 023 sets E eens Leap e ein, 0, ‘An inequality of this type holds in general for strongly continuous semi- sr0ups 11 Proposition Let {To} be a strongly continuous semigroup on L. Then there exist constants M > 1 and > 0 such that 03 IT@Is Me", 20. Prook Note first that there exist constants M'>1 and t >0 such that I T(Ol| O and h> 0, then as) Ire + Ws Test = ETOLT—AL sMe"ITUS—S1, and if < hs, then 09, ITe= mf TON Te HEH $Me“ TOS—Si o 13, Remark Let {7(9} be a strongly continuous semigroup on L such that (13) holds, and put S() = e°*"7() for each ¢ 2 0. Then {S(0} is a strongly continuous semigroup on Z such that wn IsolsM, r20.In particular, if M= 1, then {S()} is a strongly continuous contraction semi- group on L. Let (S{¢)} be a strongly continuous semigroup on L such that (1.7) holds, and define the norm ||: [[on L by as) = sup Huet Then Lf < fil < MILI foreach fe E40 the new norm is equivalent to the orginal norm: aso, with respect to =I {S()) i & strongly continuous contration semigroup on L- ‘Most of the rent in the subsequent sections ofthis chaper are tated in terms of strongly contnaovs contrition semigroups. Using these reductions however, many them canbe reformulated in tersof moncontration semi groups. e ‘A (possibly unbounded) linear operator A on L is linear mapping whose domain 9(4) is a subspace of L and whose range s#(4) les in L. The graph of Ais piven by a9) HA) = (APS HAI} € Lx L Note that L x Lis itself a Banach space with componentwise addition and scalar multiplication and norm {(f,g)1 = [f+ lal 4 is sad to be closed if (A) isa closed subspace of L x L. ‘The (infinitesimal) generator of & semigroup {T()} on Lis the linear oper- ator A defined by (410) Afmtin? (TOFD “The domain 9(A) of A isthe subspace of al L for which this it exits Before indicating some ofthe properties of generators, we brie discuss he L. Let Ci() be te space of continuously die: enable funetons w 8 L IT Ais the fine interval a 8], : A~+ Lis ud oe (Rlemany ineorabe over A i ing Doe ly = fe) et, Whee a= ig SH, StS ey Ssyctea band 8 max (te) the limi i denoted by Jul) dt or (fuloa”ir& (alcy ac doo id to be imegrable over A i ula i integrable over [orb] for cach b> a and im. feu) dl exists; again he Timit is denoted by {4 dt) dt or 2 140 dt ‘We eave the roo ofthe following lemma tothe reader (Problem 3 1. aMAmioNS AND 8ASC PRorERTES 9 14 emma (@) Ife CyB) and fed < co, then xs intarble over ] aan) [moa s [worse 1 particular, if is the Sint interval a,b, then every funetion i bathed [a 6), ry function in C,(A) is 4) “Let be a closed linear operator on L. Suppose tha u € Cd), Mi) © 18) for allt 8 Bue C4) and oth and Bees egal ‘A Then f, u(t) dt 9(B) and ee eects nde oa) 8 [ode f aay a © tee Cifa.b}. then "a wu [fae -ua 15 Propodtion Let (TW) bea strongly continuous semigroup on 1 with @ Hfe Lands 20, then f, Ti)fds © 9A) and uy Tos {resus ) Ife 9(A) and ¢ 2 0, then Tf 9(4) and (uy TWf= ATIOS= THAL (© Ife 94) and 2 0, then (116 Tws-s [arose “ Taar as Prool. (a) Observe that tan terw-nf maraet Hf” otf rosa t [rape for all > O, and as +O the right side of (1.17) converges to Tlof | (Ms + ms TN as noras~ [ nora10 ontarorswceours ©) Since (118) TIME + ng TPL = ATIF = TALS for all h>O, where Ay=A-![TU) 1, it follows that Tinfe 914) and (ajdt) T= ATIOS= TAL Thus, it suffices to check that (dia” Tf = TAF assuming £ > 0), Bu this follows from the identity (119) [news Tos) - TAF t= hil, = AIS+ (Tle W)— TONAL valid for 0< h st (© Thisis a consequence of) and Lemma 1.(0). oD 16 Corollary If A is the generator of a strongly continuous semigroup {T(@} on L, then 94) is dense in L and A is closed. Proof, Since lito. fy Tle) fds = f for every fe L, Proposition 1.5(@) implies that 9(4) is dense in L. To show that A is closed, let {/,} < 94) satisfy rf and Af,» g. Then TWfa—J.= fy TAL, ds for each t > 0, 50, Ietting n+ <0, we find that T(9f—™= fp T(9g ds. Dividing by ¢ and letting 1-10, we conclude that fe 9(4) and 4f = a 2. THE MILLE-YOSIDA THEOREM Let A be closed linear operator on LI for some real 2, 2— A (= 2 ~ A) is ‘one-to-one, A(i — A)= L, and (2 ~ 4)~" is bounded linear operator on L, then 2 is said to belong to the resolvent ser pA) of 4, and Ry = (h— A) is called the resofben at A) of . 2.1 Proposition Let {70} be a strongly continuous contraction semigroup con L with generator A. Then (0, 20) p(d) and en Gates [" e#To a forallg ¢ Land 2> 0 Proof, Let 1>0 be arbitrary. Define U, on L by Ug = [i e*TIQg dt Since @2) Waal s [ei Togk dts 2-%i4 forth 1, Use bonded tna pratt nL Now seg, 29 Lerw—nue=! [etre + hemos Fh, "Me at = eam for every h> 0, so, leting h+0,.we find that Usg © 9(4) and AU, AUs9 ~ othatis, (24) G-AWig=9, gel In addition, if g € D(A), then (using Lemma 1.4(b)) es) UsAg= iE eariag a= | * Ae™* Tog) dt wale “Tog dt = AU, (2.6) UA-Aw=9, ge HA) By (26, 4A is one-to-one, and by (24), @(i~ A)= L. Also, (2 4)" Uy by (24) and (26) s0 1€ p(4). Since > 0 was arbitrary, the proof is complete a Let 4 be a closed linear operator on L. Since (2 (= AXA A) for all 2, 1 © (4), we have (u— A)-YA— Ay (i A)-*,and a simple calculation gives the resolvent identity OD RR ARR AG WRB, He HA) HEd.© (A) and [2 — wl < Ral then Alu A) aay 29 ha ares defines a bounded linear operator that i infact (4 — A)". In particular, this implies that p(4) is open in. A linear operator A on Lis said to be dissipative it Lf — Af) = ALI for every fe 9(4) and 1 > 0. 22 Lemma Let A be a dissipative linear operator on L and let 4 > 0, Th Aisclosed if and only if {4 — 4) i closed Proof. Suppose A is closed. If {f,} < B(A) and (A A)f.—» h then the disi- pativity of implies that {f,} is ‘Cauchy? Thus, there exists fe L such that112 onmaros sewcrours ‘feof and hence Afz> if, Since A is closed, fe 9(A) and b= (A— ADL follows that (i — A) is closed. ‘Senpose Mk A)isclosed If Lf) 0. Given 9 © Lt g,= (2— ANdy — A)” foreach mand rote that, because 4 is dissipative, (29) tim fag — 98 = Him [(2— 22, — Ata tim etl igi =o. Hence @(2-— A) is dense’ in L, but because A is closed and dissipative, FA A) is closed by Lemma 22, and therefore @(2— A)= L. Using the dissipativity of A once again, we conclude that 7 — A is onetb-one and Tandy th & 4! It follows that £¢ p*(A) 80 9°(4) is closed in (0, co), as required o 24 Lemma Let A be a dissipative close linear operator on L, and suppos. 2A Sens dense in L and (0, co} a(A), Then the Yosida approximation A ara Zhchoed for each 2> 0 by A, = ACL ~ A)" has the following proper ties: (a For each 4 > 0, A, is @ bounded linear operator on Land (¢} isa strongly continuous contraction semigroup on L, (0) AA, =A, As forall 3,0 0 liye AsS= Aftor every fe M4). proot, For each A> 0, et R,=(2—A)"? and note that Rut s 4"! Since Ton 94) it fllows that (on AR, = Ton Land RQ) 210) Are BR,- aE on L, 28 and eu Ape ARsA on HA, 2>0. By (2.10) we find that, foreach A > 0, Ais bounded and. np guiny g erHeIEL gt an we 2. THE MULE-YOSDA THEOREM 13. forall 0, proving (Colson) consequence f 210 a for (c), we claim first that : leeds en) lim iR,f=f fel. Noting that [AR f—SE=(R,Afll sA-MAF|—+0 as A+0 for cack Peat is) lows tot te Ye at '(4) Bde nL td TAR, — 11 <2 for all 2>0. Finally, @) is @ consequence of (@.11) and 243). Qa 25 Lemma If B and C are bounded linear opetators on L such that BC = CBand Je] < 1 and [el] <1 for alle 0, then (214) ley esi shar oft forevery fe Land t 2 0 Proof, The result follows from the identity ain eynetn [ Eterereyese [ et —oe [ewe Of ds (Noe that eas ez testy and C) a We ar now eyo prove te Hie-Voni trem 126 Theorem A linear operator A on L isthe generator ofa strongly contin- uous contraction semigroup on L ifand only if feos Proot. The necessity of the conditions (a}() follows from Corollary 1.6 a ae 2.1, We therefore turn to the proof of sufficiency. ° Pete etasn oe we ee ec ae we T,(t) =e, By Lemmas 2.4(b) and 2.5, ee Wy 219 INOF- Tost sthas- Ast14 onaxtor sewarours for al fe L, £20, and 2, > 0. Thus, by Lemma 24(¢, ima TH(0/ exists forall t= 0, uniformly on bounded intervals, forall fe 9(4), hence for every fea ‘Denoting the limit by T(0)fand using the identity QIN Te +9f- TWMOS= (76 +0— Tis + OF + HOLT TOY + (HO ~ TOITOL we conelude that (7(0)} is a strongly continuous contraction semigroup on Ir remains only to show that A is the generator of (T(0)}. By Proposition 1500, 18) Tuos—F= | Beda as for alle L,1 2 0,and 4>0.For each fe 9(A) and > 0, the identity G19) TVAS~ TWAF= TOMAS — AN + (16) ~ TO Af together with Lemma 24(¢ implies that Ti9A,f— TWAS as A-» 2, unix formly in 0 < # < ¢, Consequently, (218) yields 220 Tef-f= [ for all 9(A) and 1 > 0, From this we find that the generator B of {T() is fan extension of A. But for each 2 > 0, 2 ~ Bis one-to-one by the necessity of {by and (1 — A) = L since de p(A). We conclude that B = A, completing the root o “The above proof and Proposition 29 below yield the following result as a by-product, TUAf ds 27 Proposition Let {T(d)} be a strongly continuous contraction semigroup on Liwith generator A, and let A, be the Yosida approximation of 4 (defined in Lemma 24). Then 2) ey TOSI SHAS- AL, Fe BA 1 20,40, 50, for each fe Ly limje4f= T(0f for all ¢2 0, uniformly on bounded intervals, 28 Corollary Let {7() be a strongly continuous contraction semigroup of ‘Lwith generator A, For Mc Ly let 2 Aygo (A> 0: a(a — At: MM}. either (a) M is a closed convex subset of Land Ay is unbounded, oF (b) M is fa closed subspace of L and Ay is nonempty, then 223) TQ:M>M, 120. 2 THE MULEYOSOA THOM 15, Proot, If 2, > O and |1 — p/| <1, then 6. (28)) 29 a art $8 (1-2) garages Consequently, iM is closed convex subset of L, then Le Ay, implies GF} Ay, and if M is a closed subepac of fy then '@ hy implies, 2 © Ag. Therlre under either (a) o (5), we have Ay = (0c) Final, by 210), es) exp {eA} = exp (13 exp (iL2 — A)" oo § say forall 04nd > 0,30 the conlsion allows om Proposition 27.) Sipoupe on L vik mre 4 aed Bo mpestedy Wd tee T(t) = S(t) for all t = 0. : . a Prot, This results consequence fhe net proposition o 210 Proposition Let A be a dissipative lincar operator on L. Suppose that 10, )-oL i continuous i) € 914) for all> 0, a (022) is contin 226 va = a+ [as as for all ¢ > & > 0, Then fxd) < 0) for 120. Prot Let 0 0. root, For the first assertion, it suffices to show that if {f) © 9(4) fr», and Af, 9 ¢ L, then g = 0. Choose {gq} © 9(A) such that gam 9. By the dissipaivity of 4, 228) (2 Adog~ Aa = lim [2 AYO. + Hf lim Alon + Yel = Allon for every 2>0 and each m, Dividing by 4 and letting A» 20, we find that deol 2 lignl foreach m. Letting m+ Ali — A) is obvious, so. to prove equality, we need only show that (2 — A) i closed. But this isan immediate consequence of Lemma 22. o 212 Theorem A linear operator 4 on L is closable and its closure is the ‘generator of a strongly continuous contraction semigroup on L if and only if (A) dense in L, (Ais dissipative (RA A)is dense in L for some A > 0. Proof, By Lemma 211, A satisfies (a)-(6) above if and only i is elosable and A satistos(}-(¢ of Theorem 26. a 2. CORES, In this section we introduce a concept that is of considerable importance in Sections 6 and 7 3 coms 7 Let be aloe ner operator on LA subspace D of 34) said tobe a core for A if the closure of the restriction of A to D is equal to A (ie, if Alp = A), “ : 34 Proposition Let A be the generator of a strongly continuous contraction semigroup on L. Then a subspace D of 24) isa core for A if and only if Dis dense in L and &(2 ~ A) is dense in L for some 2 > 0. 32 Remark A subspace of Lis dense in L if and only if tis weakly dense (Rudin (1973), Theorem 3.12. : oc rool. ‘The sullcieney follows from Theorem 2.12 and from the observation that, A and B generate strongly continuous contraction semigroups on L. and if A isan extension of B, then A'= B. The necessity depends on Lemma 211, o 313 Proposition Let A be the generator ofa strongly continuous contraction semigroup {7()) on L. Let Dy and D be dense subspaces of L with Dy © D = ‘9(A}. (Usually, Dy = D) If T(}: Dy» D for all t= 0, then Disa core for A. Prool, Given fe Dy and A > 0, an fal E omn(treo for: |, 2,.... By the strong continuity of {T(¢)} and Proposition 2.1, 0 nn ayem in £ ermer("a— a = B= Aas s0 RU Alp) > Do. This sullces by Proposition 3.1 since Dy is dense in LO “TMA Aydt Given a dipative nes operir A wth (4 dese in Lon fen wants to show that gears stony ontinowscontastonsemigoup on By Theorem 215s neoary tnd eulcont condion that 4) be Senn in forse > 0, We can vets ble aoe of hating ‘toe fame, (4) fr the generator of x songyconinuus contain rou, exept taut he saon in Proposons 3 and, scr nt poe in drench, te ei of econ Penary concer ith veriyng the Tange sondon ondon () ‘Theorem 2.12, me “ le ve tha he allowing ea genera Propostion 3.318 omtroxsoucrous 44 Froposton Let A be a disiatve near operator on Ly and Dp a sabe touce of (4) tha is See nL. Suppose tha, foreach fe Dy here ex 8 Bicinuown funtion uy {Oso} such tak a0) =H). € 1A) for al Peay: @, so) = Li continons, and : wo-aso= [mie for all 1 £>0. Then A is closuble, the closure of A generates a strongly continuous contraction semigroup {T(0} on L, and T(J = ujlt) forall fe Do and 20 and denote uy by u. Let Prool By Lemma 211, A is clsable. Fix fe Do and denote uy f,> e> 0,and note that ft €~“u0) dt © (A) and . affewoae [erate 5 [Petunanrer— of . [aa “er |" aet- Aus) ds c ete) dt he) — €-Mah Since \jut)[ < 11 for all ¢ 20 by Proposition 2.10, we can let e—-0 and to-+ 09 in (5) t0 obtain {3 ed) dt € (A) and oo =a [" eae drag fe conclude that &(1 — A) > Dg, which by Theorem 26 proves that 4 gener- se emang continuous conection segoup (70) 0 f= Now oF each Fe Pov if An@fes roposition 2.10 for all ¢> ¢>0. Subtracting (33) from this and applying Prop ‘once again, we obtain the second conclusion ofthe proposition. o on Tos Te) The nxt result shows that a slit condition fr to generat i that 4 be ttiangulizable. OF course, this is a very restrictive assumption, but it oceaionallysatsied coms 19 35 Proportion Let be « disipatve linear operator on Land suppose that Lys La Lay isa sequence of iite dimensional subspaces of 20 such that Ot, is dense in Ltt: Ly» Ly for +(2— AXLg| = Ly forall 1 not belonging to the set of igenvalues of lj.. hence for all but at most finitely many 4 > 0. Conse- quently, — A\J-2L) = Uz Le forall bt at most countably many 1 > 0 and in particular for some 4>0. Thus, the conditions of Theorem 212 are satisied. o We turn next to a generalization of Proposition 3.3 in a different direction, The idea is to try (o approximate A sufficiently well by « sequence of gener. «ators for which the conditions of Proposition 3.3 are satisfied. Before stating the result we record the following simple but frequently useful lemma, 36 Lemma Let Ay, Az,..-and A be linear operators on L, Dy a subspace of L, and 4 > 0. Suppose that, for each g € Dg, there exists fg © S(A)n3XA) forn= 1,2,...such that g, == (2 — Af,» gas n-» co and 68) fim H(A, — AYfal = 0. Then BA= A} > Dy Proot. Given g € Dy, choose {f,} and {g,} as in the statement of the lemma, and observe that lim,.lI(2— A)f,~"gql = O by (38) It follows that Jim, (2 ~ A)f,~ ol = 0, giving the desired result a 37 Proposition Let A be a linear operator on 1. and Dy and D, dense subspaces of L satisying Dy < 94) Dy NTOlyll se", £20, and ony Ti): Do HA, 120, Then A is closable and the closure of 4 generates a strongly continuous contraction semigroup on L,20 onasror suctous Proc, Observe fit that (4 i dese nL and, by (38) andthe diay Oreach drei dpa. It therfore suis to very conion (0) of Theorem Bi, Fie > a: Giteng € Dy, let fo LE cmmn(Boe on for each mnzi (CL 1D. Then for mee 2evuy GAD reciiogt dag dem gan mov coy 30 there eit «sequence (mn) of Postvelntepers soc that U= Ace abn» Morea, 6.13) HA, = AV SEF Gag Inn iol 0 as ne by (39) and (3.10), so Lemma 36 gives the desired conclusion. o 38 Corollary Let A be @ linear operator on L with 9() dense in L, and let Il- Ml be a norm on 9(4) with respect to which 9(4) is a Banach space. For n=l, 2... let 7, be a linear -[-contraction on L such that Ty: 9A) BUA), and define A, = n(T, ~ D. Suppose there exist > 0 and a stquence {2} € (0,2) fending to zero such that, for n= 1,2... (3.9) holds and G14 Wool $1 +2 “Then A is elosable and the closure of A generates strongly continuous contraction semigroup on I root. We apply Proposition 3.7 with Dy=D, = (4) For n= 1, 2... ap (t4,}: 94) (4) and G15) exp (64, lvall Sex (nt exp {ot Taloval} < exp C04} for alle 2 0,50 the hypotheses ofthe proposition are satisfied a 4. MULTIVALUED OPERATORS Recall that if A is a linear operator on , then the graph #4) of A is @ subspace of L x L such that (0,9) © $(4) implies g = 0. More generally, we regard an arbitrary subset A of Lx L as a multialued operator on L. with domain (4) = (f: (f, 9) € A for some g} and range AA) = (93. g)eA for some f), A & L x L is said to be linear if A is a subspace of Lx L. If A is Tinea, then A is said to be single-oalued i, g) & A implies g = 0; inthis case, 4 MVD OrtRATORS 21 Aisa graph oa near operator on Lazo denoted by 4,30 ve wet A= git (ged. MW ACLXL is linear, then A is sald to be desponne I La ol A171 forall (ged and > 6 the closure of Ae ceva Jum the closure in Lt. of the. subwpae A. Finale we. Slog B= A= (GH: ea) eA} foreach a> 0 ise gen) a opto is oat aon closure of Gin the above ses) is Sngle-vlued, Consequently, the tere “hosabie” so longer needs . siete We ben by noting that the generator of strongly continuous contraction semigroup isa mavina sieve rulaloe)nesropeatn, 441 Proposition Let be the generator of a tronaly continous contraction seniaroup on L. Let BL x be linar and spate and soppos tha Ac B.Then A = B. mm suppose that trod Lt Uishe# and 20. Then ULif—g)eA—B. See Le AA there exe te S10 ach that th “ih pe He, te An AST 8 by ineay, GRO} CL By by eal time g= ahve ee We turn next to an extension of Lemma 2.11 42 Lemma Let Ac L x L be linear and dissipative, Then ay Ao= (Kade A: 9 ¢ HA} is single-valued and =A} = a2 ~ A) for every A > 0 Froot, Given (0,9) ¢ Ao, We must show that g= 0. By the definition of Ay, there nists a sequence (gf) © A such that gg. For each (by + 39) © A by the inerty Of A, 40124, ~ he-— a1 2 Aba] for every 450° the disiptvty of 4. Dividing by Zand ling 2 co we ind that lig 91 > ful for each m Letting no, we conclude that 9 = 0 “The proof of the second assertion is similar fo that of the second assertion of Lemme 2.11 a Ti main result ofthis section ithe following version ofthe Hill-Yosida 43 Theorem Let ACL x L be linear and dissipative, and define Ay by (4)Then 4g isthe generator ofa strongly continuous contraction semigroup ‘on 9(4) if and only if — A) > (4) for some A > 0. rool. gis single-valued by Lemuma 4.2 and i leary dsipative, so by the Hile-Yosida theorem (Theorem 26, Ay generates a strongly continuous contraction semigroup on (A) if and only if (Aq) is dense in 4} and Id ~ Aq) = 9(A) for some 2 0, The later condition is clearly equivalent to22 oneaaror senciours i= A) =F for some 1>0, which by Lemma 42 is equivalent to SLA) 2 Ae for some 40. Ths 10 complet the proof sles to Bey ihe atte is dens in BUA) assuming that UL A) = 9A) for some tea Dy Lemma 23, @(.—Aq)= A) for every £>0, 90 AEA = aA — A) > HA) for every A> 0. By the dissipativity of A, we may regard (Q— 4)" as a (single-valued) bounded linear operator on (A — A) of norm ‘at most 2! for each 4 > 0. Given (f, 9) € A and > 0, if g ¢ @td— A) and fe BA) c Bac Ml — A), s0 9 ¢ RA — A), and therefore Waa = Arts (2 ~ A)-1g| 0. kn view Of (4.2) this completes the proof a Multivalued operators arise naturally in several ways. For example, the following concept is crucial in Sections 6 and 7. Formal, 2... et Ly, in addition to L, be a Banach space with norm ‘also denoted by {ly and let mg: L—» L, be a bounded linear transformation. ‘Assume that sup, [gl <0. If Ay © Ly x Ly is linear for each m2 1, the ‘extended limit of the sequence (4,} is defined by (43) exlim Ay = {Uf 9) €L x Le there exists (9) € Aa for each nz 1 such that [fy—mfI-+0 and lig, —* ol} We leave it to the reader to show that ex-lim,..g A, i necessarily closed in Lx L (Problem 11), "To see that exdimy-a Ay need not be single-valued even if each Ay is lt y= L. m= Land A, = B+ nC for each n= 1, where B and C are bounded iihear operators on E. Iff belongs to 7(C), the null space of C, and he Ly then A+ (L/n)—+ BY Ch, 80 as) (6 87+ CR: fe WI, he LE} < extn A ‘Another situation in which multivalued operators arise is described in the next section. SEMIGROUPS ON FUNCTION SPACES In this section we want to extend the notion of the generator of a semigroup, Dut to do-s0 we need to be able to integrate functions u: [0,co)—+ L that are | SMGROUPS ON FUNCTION seActS 23 not continuous and to which the Riemann integral of Section 1 doct not apply. For our purposes, the mos elicient way to get around this dilficulty is to restrict the class of Banach spaces L under consideration. We therefore assume in this section that Lis a “function space” that arses in the following way. Let (M, .€) be a measurable space, let Tbe a collection of postive mes tures on and let & be te vector epee of smeuable fons ach a 1) sh ssup [ist dn < en, Note that =| is @ seminorm on but need not be a nom. Let HW = (fe £: || =0} and let L be the quotient space 2/7, that is, Ls the pace of equivaience classes of functions in 2, where f~ gif If g] = 0. As js typically the casein discussions of L-spaces, we do not distinguish between ‘8 Tanction in £ and its equivalence class in L unless necessary Lis a Banach space, the completeness following as for H-spaces. In fact, ify isa ocinte measure on 1 0.1t 6 oy [cero ae [° ere eon forall he L,4>0,and s 0, then a2 ~ A) = Lfor every 2> 0. Prool, Let (fig) @ 4 A> 0,and h = Jf g. Then [evrera- [eaioe a al emreyae~2[) ca [etrone Tt Conese, < 4-M ile roving dinpaiy, nd (6) ols manag (Gh et he Land #0, and deine f= [Se"* That and ten 6 6s) [raed a ¢ [he [earn anar= [nna serie +h du de~ [Th [feenonae [Peer a [ref ave =T0f-5 for all 2 0,s0(f, g)¢ Aand h= Af ge Al Ad o 5 stwwcROUrs ON FNCTON srAces 25 ‘Te following proposition, which is analogous to Proposition 1.S(a), gives 8 useful description of some elements of A. 52. Proposition Let L and (71()} be as in Theorem 51, let he Land u> 0, {and suppose that 10) 70 [Ten dem [re + 9h as forall 0. Then 7 (=) rook, Poty'= [5 TWh deThen 1) ros-s- [res snas— [ron as = [tones fron a [rorya In the present context, given a dissipative closed linear operator A & Lx Ly it may be possible fo find measurable functions w:[0, o)-» Land [0 co)» L such that (u() (2) € for every ¢> O and 6m wus Pind 130 ‘One would expect u to be continuous, and since A is closed and linear, it is reasonable to expect that om (Lua [104)-([aauovo)ea for all ¢> 0. With these considerations in mind, we have the following multi- ‘valued extension of Proposition 2.10. Note that this result isin fact valid for arbitrary L.26 onenaToR seacnours 53. Propostion Let ACL x L be a disiptive closed linear operator. Suppose ut (0. o)-r ie continuous and (fps) ds, i) = (0) A for exch 1>0.Then Gas) [uo s IOI forall: 2 0. Given 4 > 0, define (19 [Fetunes ona ean oe ‘Then (fg) € Aand Af g = 0). root, Fix 0,and foreach €> 0, put at) = 6° fe) ds. Then 1 so wos [Mets 9 — ao) a Since (ud) «(dt +e) — ai) € 4, it follows as in Proposition 2.10 thet uid = fu0)f. Letting 2 0, we obtain (5.15). negrating by parts, [er[ooa (sis) p= emma s0 9) € A by the continuity of w and the fact that 4 is closed and linear: The equation if ~ 9 = uO) follows immediately from the definition offand gD Heuristically, it {S()} has generator B and {T(0)} has generator A + B, then (cf Lemma 62) (519) To) say wane [seat ‘We extend Proposition 5:3 to this setting. 34 Proposition Let L be at in Proposition 5.1, lt Ac Lx L be a diss pative closed linear operator, and let {S()} be a strongly continuous, measur Role, contraction semigroup on L, Suppose u:[0,co)-+ Lis continuous, [0, 0)» Lis bounded and measurable, and 5. sacxours Ov RACTON sthas 27 2H voy sn + [ses forall > 0.1 7 (rs faas)en for every ¢> 0, and (523) [isc ome dem sa [seme a for alg. 0, then (15) hos for all> 0 55 Remark The above result holds in an arbitrary Banach space under the assumption that ois strongly measurable, that is,» can be uniformly approx~ mated by measurable simple functions Proof, Assume first that u:[0, co)-»L is continuously differentiable, [0, oo)=+ Lis continuous, and (9), (0) € A for all 12 0. Let O= tg <1, < << t,= Then, as in the proof of Proposition 2.10, 624) Lah = HMO) + Chee ~ ba = bon +S [veo = fatey = (5 = 0) — Da) ] stad + Scboepd — bey ~ 6 ~ (1) = Dede) — (= tes * (8 — 5949) - Yt t=) al sad + & Chetol — Fed ~~ fae + 1S = DEED +E if. st 1) — Dus) ~~ 969) + 6D zy28 oremaToR sewcaoUrs con the right is where = fy and s” = # fort, <4 < Since the integrand on the rig Pounded and tends to Zero a8 max (tj ~ %,) + 0, we obtain (5.15) in this case Tn the general case, fx ¢ 2 0, and for each « > 0, put (25 40 Q is) da, fg | vt) ds. Thea 29 one! [aerna we [serxoaset[ [arenas wisn [sown ace [e+e one [[[se-meosea = s0fe i Siu) ds + 2°* : Fi se— mtn ar al +e + [se nie a ‘By the special case already treated, jn [somare [ [a-neaal (627 [Mos and leting 6» 0, we obtain (5.15) in general a 6 APPROXIMATION THEOREMS In this section, we adopt the following conventions. For = 1525... Ln it Wdition to Lis a Banach space (with norm also denoted by ff) and seerrais a bounded linear transformation, We assume that sup yl < % We waite fyofiff,€ Lafor each n> 1, Land lity~ol fe ~ 1 Theorem For = 1,2, et (Ti) and (7) esronay conti Srna py ony ind Lith generators Ay and A et D bed se thane flowing oe equal For each fb Ths f+ Te for lO, unforly on Bounded inera 6 APmROMMATION THEOREMS 29 () For each fe L, Tem f> Teffor all ¢ 2 0. (©. For each fc D, there exists fe 94) for each m1 such that Leo fand Ay fe Af ies (G, ANS D) < exci, Ay) ‘The proof of this result depends on the following two lemmas, the first of Which generalizes Lemma 2S. 62 Lemma Fix a positive integer n Let {S,(0} and {S{o} be strongly contin- ‘ous contraction semigroups on L, and L with generators B, and B, Let ‘Je BB) and assume that 1,S()fe9(B,) for all s> 0 and that Byr,S(-)f [0, co)-» Lis continuous. Then, for each ¢ > 0, 6 Sdimf-0y= [54a Bar and therefore 62) bsuoms—m 801s [Mass ~ mB de root. It suffices to note that the integrand in (6.1) is ~(@ds)Se ~ ny SS)F forOssst o 63 Lemma Suppose that the hypotheses of Theorem 6.1 are satisfied together with condition (c) of that theorem. For n= 1, 2,... and > 0, let Ai and A* be the Yosida approximations of A, and A'(cl Lemma 24). Then Abn, f+ AY for every fe L and 2 > 0. Prool. Fix 4>0, Let fe D and g=(i—A)f. By assumption, there exists fre HA for each n> Vsuch that fy-af and AyJo-» Af, and therefore (L— A), vg, Now observe that 63) Labs —mA%Ol = HER = A)" aT g — LA = A)" = AN = BUG AD gg = ma a9 SPA AQ eg Sol +21 — R249 0, fe MA), and g = (2— Af so that f= {5 e*TIOg du, For each nz 1, put fon {e e*T{Omg dt € 94) By (0) and the 0, let (TH@)} and {79} be the strong ty continuous contraction semigroups on L, and L generated by the Yosida eoproximations Az and 42. Givenfe D, choose (J,} as in). Then = TOS = Tih Sf + (Ta ~ TAOS] $+ THO — HD+ THORS ~ He TON +m LTHOf— TOF] for every n = 1 and ¢ 0. Fix fg 2 0. By Proposition 2.7 and Lemma 6.3, 64) THe (65) Tim sup ITAL — THOSE $ Tim tolls Sn — Anh toll An fe BoA + RAT ADE the AY Abe SU + |All SLU) s Kil Af Ail where K = supa. Using Lemmas 62, 63, and the dominated con- vergence theorem, we obiain (66) Tim sup (THO S— TAOS sim iB Wade, 5, AMOS ds = ‘Applying (6), (66), and Proposition 27 0 (64), we find that 6 Tim sup 170, f— MTOM S 2Ktol AY A Since A was arbitrary, Lemma 24(¢) shows thatthe left side of (6.7) is ero But iss valid for lif D, and since Dis dese in L hos For alle T= “There isa diserete-parameter analogue of Theorem 6.1 the proof of which depends on the following lemma. 64 Lemma Let B bea linsar contraction on L, Then. (63) yey el s SalB—S1 forallfe Land n= 0,1, 6 AROBNATION THORENS 31 Proof, Fins Land n = 0. For k=0,1,.-+5 (69) By BYLs |B "Ys |S" earn] slk—al ies. Therefore im ipy—em-ari = [aye BE ay |5e-5] ser Eko as sfer 3 oo BY asi = Vales (Note that the last equality follows from the fact that a Poisson random ‘atiable with parameter n has mean n and variance) 3 G5 Theorem For = 1, 2,..., et Z, be a linear contraction on Ly, let ¢, be ‘positive number, and put ay ~ 6,77, ~ 1). Assume that lim. = 0. Let {(T(0} be a strongly continuous contraction semigroup on L with generator 4, tnd let D be a core for A. Then the fellowing are equivalent Ga) For each fe L, This f-» T(0f forall 2 0, uniformly on bounded imorvals (b) Foreach fe L, Ti, f-> T(Nf or all 2 0. (0 For each fe D, there exists f, € L, for each mz 1 such that f,-»f and Af» Af ity (6 AP): Le D} & exit Ay rool, (arb) Immediate Ge Let £>0,fe DA), and g=(2~ AYf so that f= fg e709 dt, For each n 2 1, put uy hae Mert ng.32 onsaron seceours ‘By (b) and the dominated convergence theorem, f,»/, and a simple celeu- lation shows that (612) G-AMh= 10+ 1, Fad germ te erin FMT ng for every n> 1,50(— Air g = (A — Af I follows that 4, fy-» AL (ema) Givenf« D, choose {,} as in (@). Then (613) TY aS TOS tf) vom forevery n & Land 1 0. Fix fo 2 0. By Lemma 64, 1s Tim sup op [roo LE (ea foofeE]opes-aner and by Theorem 61, wy sup Consequently, «19 lim sup I Tiegh — x, TWOP = 0. But this is valid for all f¢ D, and sinoe Dis dense in Lit holds for all fe Le a 66 Corollary Let (V(: #20} be a family of linear contractions on L with VO)=1, and let {1(0} be a strongly continuous contraction semigroup on L' with generator A. Let D be a core for A. If limp E'[Ve —f) = Affor every fc D, then, foreach fe L, Viemtf—> Teas forall #2 0, uniformly on bounded intervals. root, It suffices to show that if {tis a sequence of positive numbers such that f,-+1 2 0, then UG /n)f—» TWO} for every fe L, But this is an immediate consequence of Theorest 6.5 with T, = V{t/n) and «, = tymfor each m= 1. 6 wrmoNMATION THEOREMS 39 67 Corollary Let (T(}, {S(9}, and {U(@} be strongly continuous contrac tion semigroups on L with generators 4, B, and C, respectively. Let D be a core for A, and assume that D = 9(B) 5 9{C) and that A= B+C on D. ‘Then, for each fe L, wr male jr-m for all 10, uniformly on bounded intervals. Alternatively, if {e,) is a sequence of positive numbers tending to zero, then, for each fe L, (618) lim [seu = Ts forall 0, uniformly on bounded intervals. rool. The first result follows easily from Corollary 66 with V4) = QU) forall = 0, The second follows directly from Theorem 6.5 o 68 Corollary Let (7) be a strongly continuous contraction semigroup on L with generator A. Thea, for each fe, (I~ (/n)A)-f— Tf for all > 0, uniformly on bounded intervals. Alternatively, if (eis a Sequence of positive ‘numbers tending to zero, then, for each fe L, I~ 6,4) PY» Tif for all + 2 0, uniformly on bounded intervals, Proof, ‘The firt result is a consequence of Corollary 66. Simply take V(q)=(U = tA)" foreach ¢ = 0, and note that ife > Oand 2 = e~#, then 19) EMVOS-f} = PU APY i= AL, here Ay isthe Yosida approximation of A (ef Lemma 2.4) The second result follows from (6.19) and Theorem 6.5. a We would now like to generalize Theorem 6.1 in two ways First, we would like to be able to use some extension A, of the generator A, in verifying the ‘conditions for convergence. That is, given (fg) © 4 it may be possible to find U0) € 4, for each n = I such that f,—f and gy» g when itis not possible (GF al least more difficult) t0 find (fg) € A, for each mz 1. Second, we ‘would like to consider notions of convergence other than norm convergence. For example, convergence of bounded sequences of functions pointwise or uniformly on compact sets may be more appropriate than uniform con- vergence for some applications. An analogous generalization of Theorem 6.5 is also given.34 onstaron scmucxours Let LIM denote notion of convergence of certain sequences Jy € ly» na, 2,..., 0 elements /¢ L satisfying the following conditions (620) LIMf=f and LIM o,=o imply LIM (f, + Ban) = af + Bg for all fe R. (621) LIM St =s foreach ket and tim sup 1/2 —IVIS%—F1 =O imply LIM Se =f (622) There exists K > 0 such that for each fe L, there is sequence fg L, with [fl SKIS ls = 1, 2,-+-y satisfying LIM j=. WA, Ly x Lys inca for each m2 1, then, by analogy with (43), we define ex LIM A, = (Ufa) L x Le there exist (f, 09) € Ay for each n> 1 such that LIM f, =f'and LIM g. = 0} 69. Theorem For n= 1, 2acvs let Aye LyX Ly and Ac L x L be linear Sha dissipative with (1 — A) = Ly and (i — 4) = L for some (pence ll) se gn and tet {7(0} and (T(0]_be the corresponding strongly continuous 2 Srenetion semigroups on (A) and BCA) Let LIM satisty (620)-4622) together with (624) (623) LIM f= 0 implies LIM (2 — Ag~'fy= 0 for all 2> 0. la) IA cexLIM Ay, then, for each (fg) € A, there exists Un» do) © Ae for aach n > 1 such that sup,[ fl < 2, SUPyh dal < 2s LIM fz = LIM ge yand LIM Tithfe= Taf forall 2 0 (bit in addition {7,@)} extends to a contraction semigroup (also denoted by (T@)}) on Ly foreach n = 1, and if (625) LIM, = Oimplies LIM Tif, = 0 forall = 0, then, for each f¢ WA), LIM f, = fimplies LIM Tyeh/, = TUOffor all ¢ 2 0 10 Remark Under the hypotheses of the theorem, excLIM 4, is closed {iL x 1 (Problem 16), Consequently, the conclusion of (a) is valid for ot aed. a rook, By renorming Ly, m= 1 2...» if necessary, we can assume K =I in (622) : Tee denote the Banach spas ([]eeit,) x L with norm given by HCE} DI = SOP IL Wf He and et (629 Boa MU Ne 2 LIM Se Conditions (620) and (621) imply that y is a closed subspace of, and ‘Condition (622) (with K = 1) implies tha, for each fe L, there is an element pase Lomi GAN = WE t (627) of = {LUI} anh al © & x 22 Uys 09) © Ay for each nel and (96 A} ‘Then of is linear and dissipative, and &(2— of) = 2 forall 2> 0. The corre sponding strongly continuous semigroup {7(0} on @/)is given by TUK A= (TAO), TOD. We would like to show that (629) (628) Fi LATA LFA, 120. To do so, we need the following observation. If (fg) € A, > 0, h= (Uh) B) € Zo, and (630) Ung == Ay ~hy foreach m > 1, then 31) UIDs Uda] & (Lo x Lo) 0 ‘To prove this, since A < ex-LIM A,, choose (/,, 9.) € 4, for each n> 1 such that LIM f,~ and LIM 9, = 9g. Then LIM (hy — (Uy ~ 04) = 0, 80 by (624, LIM (@—A)-th,—fy= 0. Te follows that LIM f.= LIM (Z— 4" LIM fas and" LIM g,= LIM Qf, h) =47— hp. Also, supal ils Faupy (fl < <0 and sup, gel = 28up, hal <0. Consequently, [} (io) a belongs to Zo x 0, and it clearly also tetonas to ‘Given ({iy} Ne Ze and £°> 0, there exists Ug) A such that 1f— 9 = h Define Uy, 00) € Ay for each mz1 by (620). Then (A= 24) "((hyH = (hf € Baby (631) 50 (632) BHM Lor Lo, 470. By Corollary 28, this proves (6.29), ‘To prove (a), let U, a) A, 2>0, and h= Af~g. By (622) there exists (hyp abe 2o with [((ag),B)| = 1AN> Define (fx, 09) © Ay for each m2 1 by (630, By (631), (629), and (628), ((T(O\s TU) © Lo for all ¢ 0, s0 the sonelusion of (a) is satisfied. "As for (b), observe that, by (a) together with (625), LIM f, =Se 9(4) implies LIM Tof, = Tf for all> 0, Let fe (A) and choose {7%} < 914) such that |" —J1< 2-* for each k > 1. Put f® = 0, and by (6.2), choose