LIBRARY OF THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY EQUILIBRIUM EXISTENCE RESULTS FOR SIMPLE ECONOMIES AND DYNAMIC ^AMES By 3ul R. ^o^ WtinJc tAurcJr Rao( Sevk\ ^leindorfer Murat^ Sertel ?a,o\ December 1972 MASSACHUSETTS TECHNOLOGY 50 MEMORIAL DRIVE IBRIDGE. MASSACHUSFTTR ni INSTITUTE OF MASS. iH3T. TECH. JArn4l973 DEV/tY LIBRARY EQUILIBRIUM EXISTENCE RESULTS FOR SIMPLE ECnNOMIES AND DYNAMIC OAMES By Paul R. J^leindorfer^«^-\ December 1972 ^^^ WonUe.. 631-72 This paper was prepared upon the kind invitation of the IEEE Conference on Decision and Control in New Orleans, 1972, where it was delivered, December 13. It carries Preprint 1/72-22 of the International Institute of Management, D-1000 Berlin 33, Griegstrasse 5. 00. ^5/- 73 RECEIVED JAN M. I. T. 18 1973 LIBRARIES INTRODUCTION AND SUMMARY O. The purpose of this paper is to bring some siirplicity and crenerality to the investigation of equilibriuir existence in certain "simple" dynamic games "" The way we deal with these objects is, essentially, by reducing them to what we term "simple economies", these latter being, from the view. point of equilibrium existence results, at once more general and less cumbersome. In §1, we exhibit some topological facts about a certain rather general class of simple economies, includinc, the fact (1.7, Main Theorem) that they each possess non-empty and compact set of eauilibria. Then, in §2, we define the simple dynamic games of interest, immediately reducing them (2.5, Reduction Lemma) to the simple economies studied. In §3, we show (3.0) that the simple dynamic games in question have non-empty and compact sets of equilibria, using the fact (1.7) that the simple economies to which they reduce have such sets of equilibria. We then illustrate by example (3.3) that a general class of discrete-time, deterministic games with convex performance criteria is covered a by the equilibrium existence results just described. This class includes dynamic games for which certain non-linearities in the next-state map are allowed, and for which controls are restricted to compact reaions, these regions themselves varying as a function of state. Of course, we do not intend 1 ) An excellent bibliography of previous work on dynamic games can be compiled throucrh th- references cited within [Kuhn and Szego, 1971^ and from^the survey paper, [Ho, 197oJ. that results particularized to this quite special example be understood as the main thrust of this paper. Historically, the study of the existence of (competitive) general equilibrium in economies exhibits quite a long-standing and extensive literature. A crucial turning point in that literature is afforded by the Arrow & Debreu fl954j study, benefiting from Debreu' s [1952] earlier investigation and using a FPT (fixed point theorem) of Eilenberg & Montgomery [1946[. This work is generalized in [Sertel, of the more powerful FPT ' s of [Prakash & by use 1971] Sertel, 1971]. Essentially, our present results can easily be demonstrated as corollaries to this last mentioned, but owing to the relative inaccessibility of both this work and the fixed point theory it employs, we restrict ourselves here to what can be done by using the relatively well-known FPT of Fan [1952] which [Prakash & Sertel, 1970] generalizes. Even with these handicaps, our main theorem generalizes the Arrow [1954 1 & Debreu equilibrium existence result. We work in locally convex spaces. Such spaces include normed spaces and the conjugate space of the Banach space of all real-valued continuous functions on a given compact space. This conjugate space, in turn, is the natural habitat of probability measures under the weak (or w*) topology (see Parthasarathy [1967|). Our working in locally convex spaces is motivated by the hope and conjecture that equilibrium existence results for stochastic dynamic games may also be obtained by reducing them to the simple economies intro- duced here. 0.0. Notation and Conventions we denote ^ N = {1,2, m We fix N^ = N ,n}, \J denote the Cartesian product n and denote W e Given any set {O}. we consider the family W, e For the set of natural numbers : = {1,2,...}. = {"'"Z ^Z = TI Z - i | """Z x...x and any Z 1,... ,m} "^z , i=l j^Z „z generic elements being in lower case; , When a given Z. e = z ( z , . . . ,^z) "^z e and bv ^Z is understood, we denote projections and components according to V , V^ = k^ ^""^ N'^Z) = ^z, k = U^) ^ m. Every Cartesian product of topological spaces is understood to carry the product topology. Let be a set. Then Z non-empty subsets of topology, sets of When denotes the set of Z is equipped with a denotes the set of non-empty closed sub- C(Z) Z. Z. [z] When Z lies in a real vector space, denotes the set of non-empty convex subsets of abbreviate C{Z)r\ Finally, P 0,(2) to C{^(Z) We . denotes the set of reals; whenever considered as a topological space, topology. Z. Q.(Z) R carries the usual SIMPLE ECONOMIES AND THETR EQUILIBRIA The main task of this section is to introduce the notion of a simple economy, define equilibria of such, and then establish some fundamental results includ- ing our main theorem giving sufficient conditions under which simple economies possess non-empty and compact sets of equilibria. Apart from the interest of these notions and results in their own right, altogether they provide the main framework and the tools for our analysis of the equilibrium existence question for the dynamic games formulated in the next section. 1.0. 1 .0.0. Definition S A simple economy is an ordered quadruplet : < W, U, T, A > = , where 1.0.1. W = {X^ a ?^ I e A} i^ is a non-empty family of non-empty sets generic elements X = nx A " X° = , x X„, A\{a} ^ for generic elements 1.0.2. U = {u is an u on : X -t- X g n Rjcx (a E , from which we define denoting e with X x e X and x' A); A} associated family of real-valued functions X T 1.0.3. (t^ = X : -* r^^J'"^ ^^ ^ is an associated family of maps non-empty subset 1.0.4. {a = 7^ X° : -^ t &a- ' ° assignino a t^ to each X (x) cT x e X; ^^ ^ is a self-indexed family of maps defined, at each x" by X", e oi(x") = where {x t x" : Pemark the maps x° e e X , x°) > ^Sup u (., x") } o iz e a t a (z a , x")}. It is evident that in a simple economy wherein ; cisely a (X is defined by Px 1 -> = {z t(x°') 1.1. (x°)|u t e are each independent of t t x", (x , = t x") i.e., t treated by Arrow & a for all (x") = t OTT^a X a (a e ^). x x we have pre- , e X and The "economies" Debreu Il954| are all of this sort. For a discussion of some major deficiencies of such a notion of an economy, most of which are present also in that of a "simple economy" •simple'! — we — hence, the oualifier refer the reader to [Sertel, 1971; pp. 18-22] Remark 1.2. To check that : 1 self-contradiction, .O is not a i.e., that simple economies are well-defined, we need only indicate that an object satisfyina 1.0.0. S will permit 1.0.4. so long as, for each index the maps t(.,x):X a e - 1.0.3, A, have fixed points -*-[x| ich u a x") (., attains a supremum on each such set of fixed points. All this will obtain, e.a., when compact spaces, each = t t (x" and oifyCt, X", a e E u t (., x consist of W is upper semi-continuous, ) consists of compact spaces (X) A) To aid in verbalizing thoughts, we borrow the follow- ing nomenclature of [Sertel, 197f]. 1.3. 1.3.1. Nomenclature: X Let be as in 1.0. S of is called the behavior space * a called a behavior of 1.3.2. x" is called the x'^ being called an x" e X"; a iff x a e «, X x a being ; ot g-exclusive behavior space a-exclusive behavior (of S), (of S) iff 1.3.3. is called the collactive behavior space X being called a collective behavior (of X X u is called the utility function of 1.3.5. t (resp., t mation (resp. of a, of effective feasibility trans forr^ation , t^(X) = {t^(x)|x iff ct d t e a and a, a X} e d iff S) ) being called the feasi - being called feasibility a (X); is called the personnel behavor (of a; is called the feasibility transfor - ) bility space of A iff S) X; e 1.3.4. 1.3.6. (of S) a (of S) beina called a a , A. e Regarding the following useful tool dealing with continuity properties of set-valued mappings, the definitions and facts in the appendix may be consulted. 1.4. Theorem a s: : Let u : P x Q -> p be a continuous function on non-empty compact Hausdorff space O -»• C(P) P x fi, be continuous, and define a map let a on by = {p a(q) s(q)|u(p, e Sup u(., q) > q) (q } 0). e s(q) Then maps a Proof upper semi-continuously into Q Given any ; q - closed in the compact s(q) P hence, pact, so that the continuous it. This shows that is compact, being 0, ; x s{q) for each q establish the upper semi-continuity of peatedly use the closed graph theorem the graph >* = r { (q , p)|q e for each q e a(q) Q, hence, closed in p Q, it will follow that P, is com- {q} attains a supremum on u a{q) ^ C(P). we will re- a, and show that A. 2 a(q)} e of is compact, r^ = Trp(({q} ^({q} x X C P P) To Q. e is a so that, r^| H ^^ n is compact, P. We now show that r ^^ >< is closed. As s is continuous, it is upper semi-continuous, so that, by A. its graph 2, is closed. As u(P X Q) ^R r^ = e by 0, graph r^ = C u(P X { (q , Q) By continuity of (p, q) e P X O, p s (q) e } C. P P >* x 0, is compact and, of course, Hausdorff. v(q) X e is continuous on the compact u By A. 4, the function q {(q, p)|q Q, = v: -^ Sup u s(q)x{q} r) |q e 0, is closed. u, P defined, at each is continuous, r = v(q) } Hence, r^ CZ Q so is the graph r = u(p, q)}ClP>' x p = r x ^ u (P so that its v(0) ig closed. {(p, q, x Q) r) of u closed. Thus, (r^ >« Therefore, its projection hence closed. Now r is closed, hence compact r^ P) into G P is compact, Q x is nothing but G which, O^g' clearly, is closed. This completes the proof. * 1.5. Corollary In the last theorem, assume also that : convex in a real vector space, that on X P Then Proof for each {q} maps a Let : a (q) with s(q) {p C(2(P). C(2.(P) All we need to show is that Q. e is convex. Now of s(0)C and that Q, e e is P is quasi-concave upper semi- con tinuously int^ O q q u . a (q) is nothing but the intersection the former of P]u(p, q) > v(q)}, which is convex by assumption, and the latter of which is convex by quasi-concavity of shows that P x {a}. This completing the proof. is convex, a(q) on u # 1.6. Definition S = <W, each = E (x) n A A of point S X iff E : X [xj -* defined, at by X, E is a map T, A> U, x The evolution of a simple economy : X (x) (tt X*^ ) = n a (x") . A is called an equilibrium or contract X e a e E(x). The set is called the contractual set of C of contracts of S. This section culminates with the following: S 10 1.7. Main Theorem Let : = <W, S be an ordered U, T, A> quadruplet such that, using the notation of 1.7.1. W 1.0.1. with each is as in X 1.0., compact and con- a vex in a locally convex Hausdorff topological vector space; 1.7.2. is as in r 1.0.2. continuous on 1.7.3. T is as in T = { (x d T a t 1.7.4. for each a 1.0.3. and, a ' , a for each x") e o x' X, a X t (x") = {(x x") : aa x') , X -*- [X X e X a X Ix' o'a c t where = {x t e t (x aa convex for each 1 is a self-indexed family of maps (x") the graph (x a , x°)} ot (x")|u (x x") , > a ^Sup x°)} , x" e u (., is as in 1.0.4. for each a e A, continuously into t maps C(liX ) ; X of x"; and such that Then (a) is a {x°J x a A, e e u X (., A P, e and quasi-concave on X x", x')|(x , a a and, upper semi- x")}, 11 (b) for each (c) S ^ e maps a , into X C{^(X and ); is a simple economy. furthermore, If, for each 1.7.5. a a e t /^ , is lower semi-continuous, a then also (d) for each (e) the contractual set a P, e is upper semi-continuous; a of C and is non-empty and S compact. T a X T (x t a x (., of ) t to a is convex by assumption and it is closed since {x"} X a of the restriction ) is closed. Thus, t a ( . x° , is into ) CQ^iX a by A. ); 2, it is also upper semi-continuous. Hence, by Fan's FPT (;v.5), of X its graph X a X . a T a intersects (x") the diagonal A o Furthermore, this intersection is one of two closed and convex sets, and so is closed and convex. Being closed in the compact X Moreover, its projection into showina that ^^^Ply \«.X t(x°) e CO(X ^""-^^^^cx' a ). ^"' X it is compact, , is nothing but X Now the graph of ^a^ e X^ X x° X t t Xjx^ (x ) is a = y^}). 12 which is compact, hence closed. Thus, is upper t semi-continuous. So much proves (a). hence compact, so that the continuous restriction of to X X subset of X X a t ) Clearly, this subset is nothing but . since {x"}, proves t a (x°) [x ] proving (c) ( (e) ad (d)): Applying Being upper semi-continuous by 1.7.5, (e) From x" upper semi-continuously into E of S maps so that, by Fan's E and (b) point. In other words, Actually, since and (a) is proved. (ad : 1.2) (Cf. is now continuous. t (d) ) we see that (b) is a simple economy S 1.4., evolution , from ), now assume 1.7.5. , lower semi-continuous by CCIKY.) is convex. This (a), ]OC(l(X [x i.e., that , and (d) As (c)): a(X°')C maps by , on u^ (b) (ad For (x and it is convex by quasi-concavity of a{x°'), u attains a supremum on a (non-empty) closed (x"} S X FPT (d) , for each Q.^{y. a ). A, e a Thus, the upper semi-continuously into [1952], E has a fixed has an eauilibrium and C ?^ is upper semi-continuous, its graph 0. 13 r = { (x, x' ) e X X x|x' pact in the compact space where h e E{x)} X is the diagonal of is closed, hence com- x. x X x compact. This completes the proof. C = As x, C tt (Fp O A), is actually 14 SIMPLE DYNAMIC GAMES The purpose of this section is to define simple dynamic games and their equilibria. The existence of these equilibria will be studied in the next section by reference to simple economies derived from simple dynamic qames The present sec- . tion gives a constructive procedure for obtaining a unique derived simple economy from any given simple dynamic aame. 2.O. Definition A simple dynamic game : ( s .d.g. ) is an ordered octuple. 2.0.0. S = < W, = {X. Y, Y, 6, 0), U, A} 3^ A > T, where 2.0.1. W j^ I non-empty sets d e X. a is a non-empty family of with generic elements from which we define X = nX. " A denoting (a 2.0.2. Y 2.0.3. Ye e x e x and A); is a non-empty set; [y] ; x e x and x" = X. x. e n X., A\{a} 6 a a for aeneric elements 15 2.0.4. 6:XxY-*-Y from which is a function, define the derived functions ^i : ^X x vre Y -> inductively Y and ^6{^x, i) = 6(^x, ^ ^5(^x, i)) and 7 Mj^x, Y.) = X Y <5(^x, ( y;) '• ••/ 6 („x, Y_))J < Z e Y, Ik e N, where j^X : <5 is a map defined by Y -* °6(^x, i) = i 2.0.5. ti) 2.0.6. U = {d). Y : (^x [x.]|a -- an XxYx = {u. j^X, e : n is a family of maps; A} e Y); e i; Y-vPliEA} isa family of ' real-valued utility functions. 2.0.7. T = {t. X : Y X whence, for each derived maps t. K a . : „X n X Y — ^t -»- ->- a . : Hx.] '-K a-' [x.] | i X x A} is a family of maps we inductively define the A, e e y -^ [x.] as follows: (footnote 2) see next page) and 16 = ""^^(n^' y) t,('^x,^-lj(^x, y)) and k-1 t.(^x, = n 1 ^ I^^J = i^(^) Y.) ^t.(^x, i) where ^a '• n^ "" °t^(„x, i) 2.0.8. A a map (^x n^5<n^ ^X, i e (footnote of page 15) x", y) ( ' e Y); e ,y) ^^^^ e x" X Y. X* a x : x" x y -^ Lx J y ^> The reader may find it convenient to assume on first reading that (x defined by is a self-indexed family of mans such that, for any 2) i-s t.(x°',y) = x. for all This assumption implies, in particular, k^i^n^'i) = k^i for all as will become clear below, (^x,i) e ^X X Y, which, corresponds to the case where the feasible control region is fixed. 17 where n^ • n*^a and '^ X "^ (^x°, ^) G is defined for each [n^^J ^x"x Y a e A by In order to facilitate discussion we introduce the follov/inrr 2.1. terminology. Terminology; From here on, whenever we speak of game", we will always mean a The elements of 2.0. 2.1.0, 2.1.1. Let x^ will be called the (^x^) (a-plan space), d-control ' X ^ ( and g-plan x^ iff ) x. t X" (^x" scace (U lan plan iff ) E X J"). space ), X e X ( X x e x. e x" (^x") a-exclusive plan X ) will be called ) iff will be called the control (^X) and ( g-exclusive control space and ( . anana X. e a a-exclu5ive control ^^ be as in S will be called an will be called the (g-exclusive plan space). x'' "dynamic g-control space (^x^) ' (^X") a will be referred to as follows. S will be called the planning horizon n ( s.d.g.. (^x) X) will be called a control 2.1.2. called the state space V7ill be Y a state iff 2.1.3. y e s pace of initial states will be called an initial state iff 2.1.4. the k-extended next state map k-string next-state map 2.1.5. Y will be called the next-state map 6 will be called y Y. will he called the Y and , . ^ k . , and y. X* • 6 will be called will be called the ,6 . will be called the initial control feasibility speci - u) fication , u). a will be called the initial a-control • feasibility specification , and d. e will be [x.^ called an initial a-control feasibility for e i 2.1.6, d. = <;.(i). d will be called the control feasibility specification T will be called the t. _t. 11 iff will be called the utility function of ^ u. a 2.1.7. Y a („t.) n ct d-control feasibility specification will be called the bility specification . and ^d. o-plan ' (j^d.) (effective) feasi— will be called an 19 (effective a-r>lan = d. n a 2,1.8. h (^x, n t. n a feasi hllity for a plan ) o-exclusive plan (^d. = t. n a n a y) i- ( n x"*, will be called the personnel of will be called a x (an and an initial state ^x"") player y) •'^ S , ) ^ iff . while each a e ^ . The connotation of the above denotation will becoire clear as we proceed; however, few points can be irade at a this tir^e. The key to understandina the above formulation of dynamic names is 2.0.8, v;hich Given any initial state ^x , each player computes the set now elucidate briefly. y. ^^'^ a-exclusive plan ^^Y is assumed to develop an a In doing so, player 2.0.8. v;e ^(^x", y;) a-plan obeying takes the given a ( x°', and y^) of solutions to the following optimal control problem: Maximize „X n X { a n x. ana u. y} x"", X n *- ( , n x", y, *- y) n-" Y subject to: /k. k. o k-1 6(x.,ic, y), k. ; y = X. where . E t. n a n^ = ^n^^a' ( n X. a , n'^"^ . keN, x, , n x", y;) ^ = <^^ / k + "^^ o. y=Y, 1 e N ^'^'^ J = ^^ "v^ 20 are the sequences of controls and states over the planning horizon, left superscript beinq the "time" index. Note that, by 2.0.7, x. °t.(^x, ^) = e a-initial control feasibility, while, for we have x. t.( x. e k control recrion for x" , • y) , , qiven a (I).(y), k = 1 , . . . - ,n 1, so that the feasible is determined by the initial state x. and the controls precedino the k We postpone a discussion of the conditions under which s.d.q. a is well-defined until we have demon- S strated that any s.d.q. may be reduced to a simple econofny. Sufficient conditions for the existence of will then S be clear from similar conditions already stated 2.2. Pemark and Notation Let : =<W, When S that S' Y. Y', and S' 6, u, r, Then clearly ^ 0. T, A is also a dynamic game. > are related in this fashion, we say is a restriction of S. will be called the restriction of When y. S is understood, v;e When S denote Y' = (i) C X' Definition S E = < W, : n : Y, S' in this case by The evolution of a simple dynamic game Y, 5, w, XxY-v rXxYl -J — Ln r, T, h > S' to the initial state S(Y). 2.3. and be a s.d.g. specified as S Yl^Y' in 2.0, and suppose S' (1.2. for simple economies. 1.7(c)) is a map defined, at each 21 („x, i) E ^X X by^ i, TV A point iff (^x, y) e called an equillbriuin of vrill be ^^ (f,^' ^^n^' ^^ "^^^ • ^^^ °^ equilibria of called the contractual set of Re mark 2.4. a Sup '^^^ u. ( a X r n x",y} -i k. ; /k. y = 6( x. a ^''^'^x. a 3) Note that i; satisfy A, e '^a^n^a'n^ 'n« ^n^& 'n^" „X. n ct S S. It may be verified that the contractual set of : which, for each ^' S £ ^t. a F ( X n X X. ,„x°,y . '-^ n a n y) n-' Y k-1.. ktot , ^ , X. , k^ ,y) '^ n a n y) , ' k Y ^ ., , , E k+1 N, E = Yr ^' N. is actually a inap into "[n^a^'^'XlC l^^"!' 22 We now show that every simple dynamic qame can be reduced to simple economy whose evolution and, hence, a whose contractual set are identical to those of the given simple dynamic game. 2.5. reduction Lemma: Let S = < W, Y,y there is a simple economy (a) derivable from uniquelv i - S S, , 6 ,aj be a s.d.a.. ,n,T, A > = < W,i\T,7^ >, with X = n x X Y, — and for which: = E(x) (b) E(x) (c) the contractual sets for Proof : (ad (a)): be aiven. Denote by X (x Let a ^ y) ; and S s.d.g. the set and coincide. S and an object .<^ ^^(a). a ^ A Consider the follov7ing sequence of definitions which specify the ele- ments of the simple economy S to be derived from S. 2.5.1 U A \{a} Xt ^ (a e PJ "" ; 23 Define the family of functions 2.5.2. u.(x) = u,(.^.(i), u. (x) = /2, {u. X -> X -<- P a ] e A^} by ^Mx))| X X; e J Define the family of functions 2.5.3. : (t. : rx.'Ma e A.) by t. (x) = t. (X) = t. (x) , n a i A e X Define the family of maps 2.5.4. X; e (x) ir^ {a x''' : [x.!|o -» e A_|_} by = afx") (x. t. (x") |u. (x. ,x a a a e a ) Sup = u.{.,x")}, .a t,(x'») where t. : = {z. t. (x") x" X. iz. e is defined by fx.! -• We now denote e t. (z., yJ^ A = (a) x") and, . for a e A, we denote aeA_|_ X a = X. a , u a = u. a , t a = t.. a These identifications yield 24 (from 1.0.1 S - 1.0.4 .-\nd 2.5.1 - 2.5.4) a unique economy = < W,U,T,A >. ( and S ad (b) and (c) ) Clearly, if the evolutions of (see 1.6 and 2.3). This leaves only that is direct from definitions 1.6 2.6. (b) to be shown, and (compare 1.0.4 (2.^.8) and (2.3) with 2.5.4) Definition : derived from 2.5 S are identical, their contractual sets will coincide (a) Let S S be a s.d.n. The simple economy S via the procedure niven in the proof of will be called the derived (simple) economy of F. 25 Equilibrium Existenc e Results for Dynaniic GamPs ^- . This section applies the equilibrium existence results obtained in section for economies to the case of 1 dynamic names. Given the reduction procedure of 2.5, the only concern is to demonstrate conditions on the elements of a given dynamic name S under which the elements of the derived economy of S satisfy the hypotheses of Theorem 1.7. This we now do. 3.0. Proposition: Let S = < W,Y,Y,«,:,tT,T,^ > be specified as in 2.0, where 3.0.1. X. is a compact and convex subset of a locally 5^ convex Hausdorff topological vector space [so that the product X, (a h e too, is compact and convex in a locally convex Hausdorff topoloaical vector space] 3.0.2. Y is closed and convex subset of a locally 7^ Hausdorff topolooical vector space; 3.O.3. 3.0.4. Y e 6 : is compact and convex; [y] X X Y linear on -^ Y X. is continuous and, x {x°'} x Y; for each x° t x' 26 3.0.5. For each with 3.0.6. ^, e a z l" , for each ( For each w. Y : a a ' ' ^ y. X.' — an yxyx u. n : x", v) e n •'- A, e is a continuous man [x.] -v — a convex for each (y) For each and, 3.n.7. 0). a t. X : map for v;hich the section Y-»-P |x -+ is continuous on quasi-concave ^ y — x X*^ y x n is a continuous I r.(x°) = { (x. ,x° ,y ,xl ) } X. {x"} X a each x" D X x" e x X a Ixl a ' and each t e k (x. a a x" ,y where N, e , . ) a a ct e is convex for is the "D.ID Y closed convex hull of the set ^D = {y e for some y| ^ (n^'i^ 1/ n^ ~^i(^>^>y) Y = Then is a siirple dynamic aame; (a) S (b) for each a e /, maps a continuously into (c) CQ.(j^X^) the contractual set of striction of S ^^x" y upper semi- and ; - S x in fact, of everv re- to a compact, convex Y ' C Y - is non-empty and compact. Proof : (ad (a) and (b) ) be a well-defined s.d.g. : By Proposition 2.5, and a e A S will have the asserted continuity and convexity properties if the will 27 derived economy of of Theorem 1.7. nomy of S fulfills the hvpotheses (1.7.1-1.7.5) <W,U,T,A> S = Let be the derived eco- We verify each of the hypothesis 1.7.1-5 sepa- S. rately. (ad 1.7.1): (ad 1.7.2): 1.7.1 is fulfilled in view of 3.0.1 and 3.O.3. 1.7.2 follows for tinuity of for each n x" a c n x" on 6 tinuity and linearity properties of = a, a '^n^"-^ ^ con- , — For (3.0.4). 6 . For 1.7.3): For » 1.7. map (2.5.2) (ad X. follows from tne con- which, in turn, , from 3.0.6. A\{a} e and linearitv of 6, a 1.7,3 reauires for each A\{a}, £ 1.7.3 follows trivially from 3.O.3. = a, a a that the A e graph n T. = {( a X., „x^, nan kl na of t« e xJ) nana Let Y n ' n x", i' y) X., „x", y)}d„X nana'n-^ n ^t. e n x ( X. (^x^, a and e y.) e ^^^ "" t t. n a „X Y — x „X. ( X., n a' y, — ^ n na T* x ( , ^x", — v) n v) = be convex } ^" be given and note that, by comoactness of A X» x.l xl n a'n a X a e ' be closed and that the graph {(„x., for each „xJ)|( X,, n a a y, -^ , T« is closed if each of its components t« iff is so, t» k + is upper semi- 1 e N. In fact, 28 since t. is a composition uous by 3.0.7), projection by 2.0.4, 3.0.4), continuous, 3.0.5. k Also, „t t. of {contin- t. (continuous ^~'^i °t. is continuous by 2.0.7 and is upper semi-continuous and u ii and is continuous, hence upper semi- t- N. E Thus, (see 2.0.7) tt is T. n a closed. We turn now to the convexity of the sections n^S Fix i^- (n^"' (^x\ that the sections Where ^ ^al^J ^5 ^ ^n^ of (^x\ T, ' = y, , , let So, it is k ^n^'a' "^a ^n^ ^} n^"' ' (ad 1.7.4): f^'x^ v} ^^ ^T. X 'Cy, kp (^.^^ ^T. linear on is u i.^ ^ '^t. nH X. x { a ^n ^^} y allows the is convex, oarUcular, tt, roting, ) ^X (in 2.0.7) of is '^ t. a is convex, observe that (^x., ^i^J ' n^"" y, ' is convex for each 1.7.4 (2.0.7) Using 1}. n ^^^^ 3.0.7) (^:^^ ^) (in (^x^, ^) V That n - {l e linear and recalling the definition that ^ X , ]<•_ of course, that projection /\ X.. ^ N) straightforward to verify that reader to compute that (,x., X ^X- e is convex is direct from the definition which is into see . 1 i^ „x", v)}<r:^X Now the convexity of lb xl) + (k (n^-, 2.0.4 and 3.0.4. k-1 • 5, We first verify y. are convex ^^ ((^^,, x % and 3.0.5. t. (^i", ^) ""t. i) ^^" e Y_) I^ i) ^"^ k + 1 "^^^ e the fact that N. follows directly from 2.5.4 y 29 (ad 1.7.5) The map : t^, 2.5.3), is certainly a e A\{a}. belncr a constant map semi-continuous. Now, fix lov.'er To establish the lower sem.i-continulty of " t = n a n logy of ^x^. is open. This beinq trivially so when It suffices to show that henceforth that V o| t for some ^i. (2.0.8) of e V, k each a ^^"^ n^a keN, '^P contained in ^ k + 1 e ^""^ n^k' = x. , k_ "p aeneric elements each ^'1^ - « and the map n 4) X^ t define writina'^ k. ^ and -- "P ^k^a' r °r ^P k>' ^'""t^f x.,g)}. : j^P x p -v V {^L} [^"^""p] and, for ""px...:^' x e ^ .P (k k^ ^y N e : ) Fo] . y.^ x ^ X k n (n^i'k^i^ a a ^ n^a " a bv -^ k^i' by The reader will, please, excuse our momentary departure, in defining j^P, from our usual notational convention as announced in 0.0. y — as We will con- U. e °P = ^«)' ^i x V. define the function N, ^x n contains a nbd of each V n '^ = and complete our proof by showing that U is a nbd of we assume to rewrite 4^ N => e We now proceed to show that l., n a of its points. Toward that, suppose struct a set V = 0, it is useful to define 0. j^ and, using the definition V = {a a Now let t.. ^^ ^"^ basic open set in the product topo- ^^"^a k^N t ^ we first note that by 2.5.3, ^' (see I.0.4, 30 ^T(^p,q) = ''tj(''x(,^r),q) Froir 2.0,7 (^p,a) we see that, for any H,(^x,,a) = k + N e 1 ^P c Since q ^(^p,q) for soire r e for every 3 , k p, ( k n,..., p)e^^,keN => ( N. g tinuous, since . . . a N, k + 1 e ]^ k + * k set 0, V n = k V (k (following the order N N, e 1 ^ (i._-iP/f5 e is con- x — verification ad is obviously so. Thus, x inductively define open sets for k N) e and, , i W, -^V, ( j e { 1 , . . for n-2,...,0), k = n-1, . ,k } ) and , obeying: n, '"•W p v;ith T(i,_^Pr")) e is certainly lov/er seir.i-continuous t To construct each p (as seen in the is so t. each e V c for each ;^lso, and since 1 6) , as U • 1.7.3, above) k + 0. and for any , From this we are able once more to rewrite U = {nEO| X = nV ^Prn)\ ( n ''"'^V.) j=k+1 n ''T(,p,q) Ji 0} J with k (^P,...,''p,q) e ( n j=l ^v.) ^ X ^nc'^w, ''uCo. ^P, ^v,Cr *" Fto check that this is possible, we use the followina facts inductively (in the order k = n-1, n-2,...,0) : (i) x is 31 A lov/er Kerri-continuous and V. -i=k+1 is open; ''^V (ii) ''^^', n r j=k+1 (""p, . n 1 . . ,'^f ,'^) rC X ^Y Nv; G w and , '"''V so ''iC-f,^), '^ e - usina (iii) W open, so i.'=; -"' and (i) (ii) , is chosen as a basic open set in the product =1 1 topolocry of A U = Set PX...X k P X 0. construction, it is clear that Fror^ its U. V k=o is open and that Cr. V For this, we take any constructinq for every n point a k + 1 ( Since N. e q e p,...,"f>) q V.) =1 V e p T(.p,q) e we have vCZ°\<l, ^ n t( p,q) G ( k r) V j=k Then, f^ Choose 0. as any point of this f> ^p n ) • k-1 - T ~ k + e 1 N, is chosen for every (i^_^P»<l) k e ( 1 , . . ^ e A j=k V. (k e { 1 , . . . ,k } ) and o clear, it follows that ( ^n, . . . ,'^n,q) e( n j ^W ^ f, so that ( and affords an element f) k+1 - p. exists, so that satisfies 1.7.5, U C ^' being =1 - . ^V ^ ^"'^r.) (\ ''\(rP,n) j=k+1 Pf---/"p) e ^ k ( by i^ e ^ " Now a such that rC e intersection. Now assume that, for kp and show I'' - (0 j q It remains to show only that V. c ) x is non-emnty ^ This shows that the desired V^V, establishina that and completina the verification of the hypotheses of theorem 1.7. ~ '^ncz'^W. S . ,k} 32 ( That ad c) has a non-empty and compact contractual S set equilibrium is now a direct consequence of theorem 1.7 and proposition 2.5 (b) That any restriction of . (non-empty) compact and convex set to a S ^^^ ^ non-empty X'CI X and compact contractual set follows from 2.2 and a simple check that the restriction 3. O.I - S' of S so obtained satisfies (where only 3.0.3 is involved). This completes 3.0.7 the proof. 3.1. Remark If, : in 3.0, one takes the restriction of initial state i e Y, then 3.0 an equilibrium for all exists X E X ^ e Y, im.plies that (c) i.e., satisfying 2.4.1 with for each Y. ~ S Z to the S ^ has (y.) there X.' Thus, under Y.- the hypotheses of the last proposition, there is an equilibrium plan for each feasible initial state. Proposition 3.0 also shows that there exist well-defined simple dynamic games, so that 2.0 is not a self-contradiction. Of course, this V7ell- definedness holds under much weaker conditions than 3.0.1-7. Clearly, ^'^n^a^ ^t* """^ ^a' will be a continuous map of ^a' ^"^ ^ ^x"' x any additional convexity or linearitv assumptions. and recalling 1.2 and 2.5, the derived economy hence, S S Piven this, of (and, S itself) will be well-defined if 3.0.1-3 hold, is continuous, continuous into y ^^^ only assumed continuous without and if, for each a e A, w., a t*, a and if u. a * are 33 3.2. Pemark In the sinple dynainic qaires we have considered so : feasible control rerrion to vrhich the far, the k+1 a-control • is constrained to belonn depends on the previous X. a k control k-1 and state X • (See the discussion iinmediately y. preceding 2.2). The reader will note, however, that the case k where the control region in question also depends on naturally under the case that v/e illustrate this by considerina k on • a falls consider generally. Pelow we dependence of these reaions on k and, while this dependence is on y, • y • alone, the y reader will easily be able to extend the siinple idea involved k to the general case where the dependence on k is not through Suppose 3.2.1. : Y E for each a A, e vex (k E N) u. = and "f. a t a '^r{y.) = a {(x.,y) a where , '^D 'V . X. e (y) CO(X.) e ^dIx. x a ' a e I*, a (y) a-controls k+1 . a o for all } con- are con- to their respective control reaions is, S hence, as follows: £ >f • x. a '^'''i. = is as in 3.0.7. The fashion in which the strained in . a [x.] satisfies -> with Y, satisfies 3.0.1-4 W,Y,Y,(5,w,t\T,A > < is a continuous map for v/hich y^ y ^F . • y alone. 0.6 and that, and 3.0.6 v/here k-1 and • y = S • x >f •('^y) , k + 1 E N, a e /. 6 34 Moreover, it is straiahtforward to check that satis- S fies 3.0.5 and 3.0.7; so that the conclusions of Proposition 3.0 are valid. The followina exanple illustrates the application of 3.0 to a salient class of deterministic dynamic games. 3.3. Example {I. Let : N|a A} of a s.d.cr. S 3.3.2. Y = 3.3.3. Ye 3.3.4. 6 3.3.5. w. m be a finite set, and let A e t and m e = < W,Y,Y, 6 {m. e N|a ,(!> ,U,T,A > as follows: R; is compact and convex; [y] satisfies 3.0.4; : where Y -> 4- . is defined by |X.] : Y -» C (X. ) e A), be niven. We specify the elements N w. C^) is defined by = "^ ^^^^ 4'.(y) ' = ^ ^ ^' 35 {x^ < O) X^l^li^ik^rY) e is a given continuous 3.3.6. t^:XxY-> is defined |_x.] t.(x,y) = H'.(6(x,y)), (^x,y,^y) "f. ("y) + k=o {'^f. X : are each a Y X a Y X X (x,y) ^y) , N, a \li (a ^ e * X e "^ ** ^i ^ "* ji a A); for each i by A e Y; X u(^x,i,^y) = by ^^ ^f.i^-'^'^k, E " 3.3.8. ^X e in which , function, °y where e and where Y, " -> P|k + e 1 and {"f. p) E Y : -> P|a e M family of continuous and concave functions; satisfies 2.0.8 (a e P) Suppose, in addition, that the follov;ina condition holds. 3.3.9. For all a e A, conditions on if (ii) ip . a 4'. v/ill be "¥. a satisfies: for each y satisfies 3.2.1. e (i) ij;. Y, \p . a [Note that these satisfied if : X. a x Y is \h . ^a P -» linear or is continuous; f. a is convex on X. x {y}; 36 (iii) for all y there exists Y, e x. X. e ^P^(k^,Y) See, e.n. Hogan (iv) 3. o. 3.1, for each a y S, x,y, u. "^ a n A, e ( "^ n 5( ^f. n n x e i^^^k.,^^^k\^y)) n k*a ;/k6 ( ijj.c'^x., X. , ^ ""y) X k-1-. , < O, y) ^x. , • oy = X' ex., n x,y)) = Subiect to ky = '^D, as specified above, there exists an Y, — e •^ ("f.("y) + Maxii^un. ft x k e N. satis- 1-3.0.7, Therefore, by 3.0, and recallino 2.4 and fies n X. [1971^, Theoreins 10,12.3 It is easily verified that for all is convex on ij/^ v;ith a ct and O; < ., , k e N; k e N, X satisfyina, ' ' 37 4 . Llirltations and Extensions Our main results are the eauilibriuri existence result for siirple econoiries (1.7), the lemira (2.5) reducinc sirrnle dynamic qames to simple economies, and the eauilibrium existence result (3.0) for simple dynamic cames The reader may have been struck by our somewhat curious obstinacy in always referrina to the objects we treat as "simple". Here we try to indicate the deficiencies in the above formulation which gave rise to this standing oualifi^ cation. For simple economies, these deficiencies amount to our havinn irinored the followina: 1. informational and/or perceptional imperfections on the part of behavors; 2. the usefulness of specifying "effective" preferences (utility functions) as dependent upon a desian pa- rameter which may be called an "incentive scheme"; 3. the fact that, in practice, the feasibility trans- formations depend not only on the behavior chosen from a given feasibility, but also on the aiven feasibility itself - the resources one has tomorrow depend not only on what one does today, but also on the resources available today. 38 From the viewpoint of eauilibrium existence, these oininissions turn out not to be crucial, since it can be shdwn, under fairly weak assumptions, that generalized economies (called "social systems" in [Sertel, 1971^), including the above features, retain the property of having non-empty, compact sets of equilibria (see [[Sertel, 197l]]). Nonetheless, from the viewpoint of social system design {e.q., selecting an appropriate incentive scheme) , these characteristics are clearly essential elements of the problem. T^nother element of simplicity in our work and the pre- vious literature is the assumption that the collective feasi- bility is box-shaped, i.e., that what is currently feasible for a. is independent of the current a-exclusive behavior. One would, of course, like to remove this restriction, as many of the problems of central planning (and control theory) are not completely decomposable on the constraint side, even by "dual" m.ethods (incentive schemes) . Turning now to dynamic games, those treated here suffer from the same restrictions discussed above for simple economies. Moreover, there appear to be sionificant difficulties in interpreting the informational and behavioral connotations implied by the evolutions (and their equilibria) of these simple dynamic games. In fact, in the current formulation, each nlayer a e A computes his a-plan given an a-exclusive plan. This can be likened to the situation in an auction hall, where each player calls out his projected plan, and recomputes 39 his plan based on the projected (i.e. called out) plans of all other players. The equilibrium existence question posed, then, is sirply whether there exists a plan which is repeatable in the sense of beino called out twice in succession. We find it difficult, however, to interpret the equilibrium of this "auction" process, unless n = in the case where plans 1, are exhibited only as thev are enacted; i.e., where equilibria are to be interpreted as repeatable in the sense of enactable twice in a row. Finally, as may be seen from 2.4, the equilibria whose existence is established here correspond to open loop equilibria of dynamic names as they are usually defined However, the existence of closed loop equilibria 1970|). |Ho, (see, e.a., can also be studied by similar methods. Essentially, one starts with oiven simple dynamic name a dynamic name of and derives from it S by replacina the d-control spaces by the function space of stratecry maps S laws) S' to X.. a a • Y. (or control from the observed state and control history tine k) k (until The (non-trivial) issue to be resolved is the determination of conditions under which the dynamic qame S' resultina from this transformation will inherit from the convexity, linearity, continuity, and compactness assump- tions required by Proposition 3.0 5) For a P - or, more aenerally. discussion of some aspects of the meaninrrfulness of prrviilibrium in dynamic oames , see IPtarr & Ho, 1969a, 1969bl 40 throuah an appropriate reduction, the assumptions required by Theorem 1.7. In nrincirle, the same apt-roach links the above results to stochastic dynamic names, thouoh clearly much work remains to be done in specifying the transfor- mations reauired for reducina these to the simple economies and dynamic frames introduced here. 41 ^ • Appendi x In the interest of accessibility, here we collect a minirral amount of topoloaical information concerning certain maps whose values or arcruments are non-empty sets. In what follov/s, are topolocrical spaces. Our first de- and P finition is extracted from [[Michael, 1951^, of course. ( N.B see A.I. . For Isf, usf and finite topolorries on a hyperspace, 195lQ.) [Michael, Definition Let : F : P -> is lov;er semi-continuous (use) J [o] (Isc) be a map. We say that iff it is continuous when semi finite (Isf) [Equivalently [[resp. , is aiven the low er [o] upper semi-finite we say that , F is Isc F upper semi-contin u ous [[resp., (usf)^] (resp., use) topolooy, iff the set {p is open (resp. , (resp., closed) continuous when valently, Isc c P F I closed) .3 n V} in P We say that [o] whenever F VC is open is continuous iff it is is qiven the finite topology. say that V7e (p) F [Equi- is continuous iff it is both and use. The proof of the following "closed araph theorem" may be obtained, 3.3 in for example, by combining propositions 3.2 and [Prakash of [Fan, 19 52] : & Sertel, 1970], where 3.3 is also I.emma 2 42 Proposition P. 2. Assume that : Hausdorff, and let f P : Denote the oraphs of f and P -> 0, are both compact O F P : and F by -* a be two maf C (0) and G, s respec- tively: q = <^ = It is iff { ( {p,Ci) 1 (P/q) 1 p c P, q = f (p) }, P e P, q E F(p) (resp., G) q CP }. is closed that x f is continuous (resp., F is use). The followinq fact, which is a part of corollary 3.1.4.3 in [sertel, turns out to be 1971^, useful tool basic to a optimization. A. 3. Proposition w : C(P) w(d,q) Equip P = x be non-emptv and compact Hausdorff, n x u:Pxq->r and let Define Let : be a continuous real-valued function. O - P by Sup u(p,q) ped (deC(P), QcQ) with the finite topoloqy. Then C(P) w is continuous We have the followinn obvious A. 4. Corollary : Let all be as in A. continuous. Defining v(q) v = w(s (rr) v , is then continuous. : q) 3, O - P (a and let by e O) , s : -> C(P) be 43 A. 5. Fixed Point Theorem Theorem 1; Fan IS 52 Given : locally convex Hausdorff topoloqical vector space if X^ is compact and convex and if L is an use point x e transformation, X such that x F : X -^ a L, CQ^iX) then there exists a (fixed) e F(x) . 44 REFERENCES Arrow, K.J. and G. Debreu, Competitive Economy", Econometrica a G. , vol. 22, "A Social Equilibrium Existence Theorem", , 3, in Proceedings of the National Academy of Sciences (U.S.A.), vol. Eilenberg, No. 265-290, 1954. pp. Debreu, "Existence of an Equilibrium for S. 38, pp. , 886-893, 1952. and D. Montgomery, "Fixed Point Theorems for Multi-Valued Transformations", American Journal of Mathematics Fan, K., , vol. 68, dd. 214-222, 1946. "Fixed-point and Minimax Theorems in Locally Convex Topological Linear Spaces", Proceedings of the National Academy of Sciences , (U.S.A.), vol. 38, pp. 121-126, 1952. Ho. Y.C. , "Differential Games, Dynamic Optimization, and Generalized Control Theory", Journal of Optimization Theory and Applications, vol. Hogan, W.F., 6, No. 3, Sept. 1970. "Point to Set Maos in Mathematical Programming", Working Paper No. 170, Western Management Science Institute, UCLA, Feb., 1971. 45 Kuhn, r. and C. Szego, Gaires and ^''ichael, North Holland, 1971. , "Topoloaies on Spaces of Subsets", E. Transactions of the vol. The Theory of Differential (ed.) Pelated Topics 71, Parthasarathy 152-182, pp. K.P., , 7^ir,erican Society , 1951. Probability Measures on Metric Sp aces, /Academic Press, 1967. Prakash, P. and M.^. Sertel, Theoreir.s r''atheriatical "^^ixed Point and Minirax in Semi-Linear Spaces", Workincr paper, Sloan School of Manaaement, Mgc^ciachusetts (484-70) Institute of Technolony, Sent., 1970. Sertel, "Flerients of Eouilihrium Methods for Social ^''.P. /Analysis", TM.T. , Jan. tinpublished Ph.D. Dissertation, , 1971 Starr, A.W. and Y.C. Ho, . "Nonzero-Sun Differential Haires" Journal of Optimization Theory and Applications Starr, vol. 3, 7\.W. and Y.C. Ho, No. 3, March, "Further Properties of Nonzero-Sum Differential Games", and Applications , , 1969. vol. Journal of Optimization Theory 3, No. 4, April, 1969. y^ BaUm- 143 w no.624- 72 Ronald Eu/The accelerated r-J5 ^rv 3 DOO ^OflD summer OM^Sm.,, D'^BKS 53^659 7t)l bfll - l,l:ll:OI:illi:ifil DD3 b7D ^DflD HD28.M414 no.626637015 [i-Pk;; TDflD no HD28.IVI414 Sertel, 3 7 of 782 7Mfl 357 629-72 fundamental continu n-nkS |'}"'l'|/-|Vj TOaO ODD 7M7 755 TDfiD 3 TDflD 111111,11 3 iliin2 QQQ /The Mural 637018 3 72a Zeno'Thp pronomir impact Zannetos, 3 fl^7 003 7D1 TMO DD3 b7D T21 1 TDflD irioijiii DD3 b71 Til 1 iipirtff'liiiii 3 TOfiD 3 TDfiD Q03 7D1 Its 003 b7D T^b -^