Journal of Mathematical Economics 11 (1983) 261-266. North-Holland A THEOREM ON THE ADDITIVITY OF THE QUASI-CONCAVE CLOSURE OF AN ADDITIVE CONVEX FUNCTION* Uzi SEGAL Nufleld College, Oxford OX1 INF, UK Received September In this paper a necessary and sufficient of an additive convex function is given. 1982, accepted condition January for the additivity 1983 of the quasi-concave closure 1. Introduction Most of utility theory depends on the assumption that the preference relation is convex, i.e., that for all x the set {ylykx} is a convex set. However, even if we do not assume that the preference relation is convex, some of the results of utility theory still hold. To obtain these results, it is convenient to define the ‘convexed’ relation 2‘ in the following way [see Hildenbrand (1974)]: Given a relation 2 define the convexed relation 2’ by =>[ XE C onv{wlwkz}], where ConvA means x&‘y iff Vz [yEConv{wlw~z}] the convex hull of the set A. However, not all the properties of the unconvexed relation hold for the convexed one as well. In this paper it will be assumed that the relation 2 can be represented by an additive convex function of the form 1 ui(xi) (x,, . . . , x, E R), and it will be shown under what conditions the relation 2’ can be represented by an additive function as well. Note that since the utility function is not a quasi-concave one, we shall not be able to use the technique which was used by Houthakker (1960) to show the conditions under which both direct and indirect utility functions are additive. 2. Definitions and propositions Definitions 1. R”,={(x, ,..., x,)ER”Ix~~O ,..., x,20}, R”, + = {(x 1,..., x,)ER”Ix~>O ,..., x,>O}. *This research was supported Gorman for useful discussion. 03044068/83/$3.00 0 by the A.V. and J.E. Posnansky 1983, Elsevier Science Publishers Trusts. I am also thankful B.V. (North-Holland) to T. 262 U. Segal, An additive convex function 2. A function U: R: +R is quasi-concave (quasi-convex) if for all x, YE R; and AE [O, 11, U(Ax + (1 - 1)~) L min {U(x), U(y)} ( 5 max {U(x), U(y)}). 3. Let U:R” + +R be a continuous strictly increasing function (i.e., [for all i, Xi Lyi, and, for some i, Xi >yJ*U(x)> U(y)). The quasi-concave closure of U will be denoted by U* and is defined by U*(x,, . . ., x,) =min {max {U(yl, . . .,~~)lCUi(yi-~i)=O}l(~,, . . ., a,)~ R: +}. Remark. The min and the max in the last definition continuity and the monotonicity assumptions. 4. The function U: R”, +R is a transformation and h:R+R there are ur,..., u,:R++R is justified by the of an additive function if such that U(X~,...,X,,) = h(C ui(xi)). 5. The set of the indifference curves of a function R}. H(x r,. . . ,x,)E R; IU(x,,. . .,x,)=a}la~ U: R;+R is the set Propositibns 1. A convex function is quasi-convex. such that, for all i, Ui is convex, then U is 2. If U(Xl,..., X,)=CUi(Xi) convex. 3. If U represents relationk”. 4. The indifference curves of the quasi-concave closure of a quasi-convex function are intersections of R: with hyperplanes. 5. the relation If U is a transformation 2, then U* represents the convexed of an additive function, then, for all i #j # k # i, (a/ax,)((aU/axi)/(XJ/axj))~o. 6. If U = h 0 V, then there exists h* such that U* = h* 0 V*. 3. The theorem Theorem. Let u 1,. . . , u, : R + +R be strictly increasing differentiable convex functions. The quasi-concave closure of the function U(x,, . . .,x,) =cui(xi) is a transformation of an additive function if and only if there are u: R+R and positive d,, . . . , d, such that w Proof. 1,. . .) Xn)=CU(diXi). We shall prove first that the condition is a sufficient one. Let ccording to the assumptions on U, u is convex and U(x l,...,xn)=zu(dixi)..A U. Segal, An additive convex function strictly increasing. Obviously, each indifference 263 curve of U* is of the form Given a point XE R”, it is easy to verify that x~H(xd~q) and that u*(x l,...,Xn)=U(Cdixi). As for the necessity of the condition. Let U(x,, . . . ,x,) =cui(xi) be such that, for all i, ui is a strictly increasing differentiable convex function and assume that U*(x,, . . . , x,) = h(~ui(xi)). Assume first that u,(O) =. . . = u,(O) = 0 and define wi=u;roui. Obviously, for all i, W,(O) =O, Wi is a strictly increasing differentiable function and U(x,, . . ., xn) =C”l(wi(xi)) =“l(xl) + i$2 ul(wi(xi))~ Define Yj(z) =(wi(z),O,f ** O), j=l, 9 =(O o,goO), )...) j=i, )...) 0, w,: ‘wyyz)), 0,. . . O), =(O,..., ) U(yi(z)) =. . . = U(yL(z)) hence of the set {y~(z)}jnzI for some Let x=(x,,0 ,..., O,xi,O ,..., and x=~~Lx~~~(z). Obviously, Xi =( 1 -c~~)z. We obtain that j${Li}. each indifference curve of U* is the convex hull iE{l,...,n} and ZER+. 0), ~20 and (c(~,..., c(,)ER;, such that ccli=l for j$(l, i}, olj =O, hence x1 =a,w,(z) and ~1~=(z-xi)/z, hence zwi(z)-xiwi(z)-xxlz=o. (1) For all i #j denote Zij(Xi,Xj)'U*(O,. ..,O,Xi,O,. u*(x r, . . .,x,,) = h(CUi(xi)) see Debreu j(X;,X[i) = Zi .,O,Xj,O,. ..,O). and it is easy to verify that [zij(Xi,XS)=Zij(X:,Xj) Zi . A Zij(Xi,Xji)=Zij(XF,Xj)]* j(X;l, XJ); (1959). Define Ai={(O ,..., O,xi,O ,..., O)eR’!+). 264 U. &gal, An additive convex function On thse se’t L\ A,.U* =U,. hence Let a,flzO. zli(wi(a),O)=z,i(O,cr), Zli(Wi(B),O)=Z,i(O,p), hence In other words, U*(Wi(p)pO,. . .,O, Cty0, * * * ~0)~ U*(Wi(OI),Oy and, by using eq. (l), we obtain that two following equations: ZWi(Z) - aWi(Z) -ZWi(B) there exists * a. yOy~,O,. f*90)~ z=z(ac, fl), which solves the =O, (2) ZWi(Z)-PWi(Z)-ZWi(a)=O. Differentiating these two equations implicitly yields aZ zwi(a) wi(z) aol=Wi(Z)-Wi(~)+(Z-a)w~(z)=Wi(Z)-Wi(cI)+(Z-~)W~(Z)’ aZ WP) wi(z) ag=Wi(z)-Wi(j?)+(Z-CI)W:(Z) =wi(z)-wi(c()+(z-~)w:(z)’ hence Wi(Z)/ZW:(B) =zw{(E)/w~(z) or (Wi(Z))’-z~w~(N)w~(/?) =O. As can be seen from (2), ~(a, 0) = 01,hence, for B = 0, we obtain The s6nii~on ti ‘r?msr?iS5eren’ix$equakon js Wi(ct)= tip/( 1+ Cic(), and, since wi is an increasing function, di ~-0 and ci 20. Define and we obtain that U(X 1,. . .) X,) =CU,(diXi/( 1+ CiXi)). d, = 1, c1 =O, U. Segal, An additive convex function 265 Let ~20 and define xi =z/(di -c~z). For all i, Ul(dixi/(l +cixi)) = each indifference curve of U* is a set of the form Us hence {(%Zl(4-c,z), . . . f O&Z/(&-Cnz))ICai= l, (cI1,.. *>am)E RT}> and the value of U* over this set is ul(z). Let (x,, . . .,x,) E R’!+. &iZ/(di -CiZ) =Xi(di -ciz)/z> =Xi*cli and, since Ccq = 1, we obtain that 1 +CCiXi, CdiXt/z= hence u*(x 1,. . .,X,)=~1(~)=ul(CdiXi/(l By Proposition +CciXi))* 5 we obtain that, for all i # j # k # i, (dic,-d,ci)(dj(l +CClXl)-Cj Cd,xl) -(djCk-dkCj)(dt(l +CC,X,) -ciCdlx,)~O. Let x1=.,. =xn =x, and denote d=xd,, c=Cct, thus Differentiating with respect to x yields and we obtain that ck = cd,/d. c,=O and d,=l, hence c=O and, therefore, c,=...=c,=O, thus U(x,, . . . 7xn) =Cu,(dixi)* If U(0,. . . , 0)# 0, define V(x 1,‘“, By Proposition X,)=U(X1,...,Xn)-U(O,...,O)=C(Ui(Xi)-Ui(O)). 6, V* is also a transformation of an additive function, U. Segal, An additive convexfunction 266 hence, by what has been proved already, Define u(x) = u(x) +xu,(O)/n, Remark. We actually b 1,..., b,ER, such that and obtain that proved that there are u: R-R, dl, . . . , d, > 0 and Ui(Xi) = U(dixi) + bi. References Debreu, G., 1959, Topological methods in cardinal utility theory, in: K.J. Arrow, S. Sarlin and P. SupIpes, e&s., Matnemaficd~ m&to& ‘m ‘he so&h sktences ‘~&i&orb ihitver~try “rress, Stanford, CA). Hildenbrand, W., 1974, Core and equilibria of a large economy (Princeton University Press, Princeton, NJ). Houthakker, H.S, t960, Additive preferences, Econometrica 28,. 244-257.