Int J Game Theory (2001) 30:177±185

999
20019

Assignment games with stable core*

TamaÂs Solymosi1 and T. E. S. Raghavan2
1 Department of Operations Research, Budapest University of Economic Sciences and Public
Administration, 1828 Budapest, Pf. 489, Hungary (e-mail: tamas.solymosi@opkut.bke.hu.)
Supported by OTKA Grant T030945.
2 Department of Mathematics, Statistics and Computer Science, University of Illinois at
Chicago, 851 S. Morgan, Chicago, IL 60607, USA (e-mail: ter@uic.edu.) Partially funded by
NSF Grant DMS 970-4951.

Final version: April 1, 2001

Abstract. We prove that the core of an assignment game (a two-sided matching

game with transferable utility as introduced by Shapley and Shubik, 1972) is
stable (i.e., it is the unique von Neumann-Morgenstern solution) if and only if
there is a matching between the two types of players such that the correspond-
ing entries in the underlying matrix are all row and column maximums. We
identify other easily veri®able matrix properties and show their equivalence to
various known su½cient conditions for core-stability. By these matrix charac-
terizations we found that on the class of assignment games, largeness of the
core, extendability and exactness of the game are all equivalent conditions, and
strictly imply the stability of the core. In turn, convexity and subconvexity are
equivalent, and strictly imply all aformentioned conditions.

Key words: assignment game, stable core, large core, exact game

1. Introduction

Assignment games (Shapley and Shubik, 1972) are models of two-sided match-
ing markets with transferable utilities where the aim of each player on one side is
to form a pro®table coalition with a player on the other side. Since only such
bilateral cooperations are worthy, these games are completely de®ned by the
matrix containing the cooperative worths of all possible pairings of players from
the two sides.
Shapley and Shubik (1972) showed that the core of an assignment game is
precisely the set of dual optimal solutions to the assignment optimization prob-
lem on the underlying matrix of mixed-pair pro®ts. This result not only implies

* The authors thank the referees for their comments and suggestions.
178 T. Solymosi, T. E. S. Raghavan

that all assigment games have non-empty core but also that the core can be
determined without explicitly generating the entire coalitional function of the
game. The algorithm of Solymosi and Raghavan (1994) demonstrates that for
assignment games the nucleolus can also be directly computed from the data
i†
that induces the game. The bargaining set M1 (Aumann and Maschler, 1964;
Davis and Maschler, 1967) is another solution which is completely known and
easily computable for assignment games, since for these games it coincides with
the core (Solymosi, 1999).
The stable set, the classical solution suggested by von Neumann and Mor-
genstern (1944), need not exist for all TU-games (Lucas, 1968). Besides, it need
not be unique. Indeed, if a game has a stable set then it typically has a multitude
of stable sets. On the other hand, there are classes of special games on which the
stable set exists and is unique. Perhaps the best known such class is that of
convex games for which Shapley (1971) proved that the core is the unique stable
set.
The main purpose of this paper is to identify those assignment games whose
core is stable, hence it is the unique stable set. Since assignment games induced
by diagonal matrices are convex (cf. Theorem 3), there are such games. A sec-
ondary goal is to give the characterization in terms of properties of the under-
lying matrix so that its veri®cation does not require the explicit knowledge of
the entire coalitional function. We achieve both goals by proving that the core
of an assignment game is stable if and only if the entries in an optimal assign-
ment for the grand coalition are row and column maximums in the underlying
matrix (Theorem 1). In other words, to have core-stability it is necessary and
su½cient that there is a matching between the two types of players in which
each player is paired with whom his/her pro®tability is the highest. This implies
that assignment games with a di¨erent number of players on the two sides
cannot have a stable core. It remains an open problem whether or not all as-
signment games have a stable set.
Several su½cient conditions for stability of the core have been discussed in
the literature. Convexity of the game (Shapley, 1971) is a well-known one. Sub-
convexity of the game and largeness of the core were introduced by Sharkey
(1982) who showed that convexity implies subconvexity; subconvexity implies
largeness of the core; which in turn implies stableness of the core. In an un-
published paper Kikuta and Shapley (1986) investigated another condition,
baptized to extendability of the game in (Gellekom et al., 1999), and proved that
it is necessary for core-largeness and still su½cient for core-stability. A uni®ed
proof of all these relations was given by Gellekom et al. (1999). Their valid
counter-examples demonstrate that these conditions are indeed all di¨erent.
Shapley (1971) proved that in a convex game, for each coalition there is
a core allocation which gives exactly its worth to the coalition. This condi-
tion, baptized to exactness of the game in (Schmeidler, 1972), typically neither
implies nor is implied by core-stability (Biswas et al., 1999; Gellekom et al.,
1999). On the other hand, for totally balanced games, exactness is implied by
largeness of the core (Sharkey, 1982) and even by extendability (Biswas et al.,
1999). It follows from the results of Biswas et al. (1999 and 2000) that for totally
balanced symmetric games and for totally balanced games with no more than
four players, core-largeness, extendability, exactness, and core-stability are all
equivalent conditions.
We provide matrix characterizations also for these su½cient conditions
(Theorem 2 and Theorem 3). It turns out that on the class of assignment games,
Assignment games with stable core 179

largeness of the core, extendability and exactness of the game are all equivalent
conditions, but are strictly stronger than stability of the core. Convexity and
subconvexity are also equivalent, and are strictly stronger than all the other
conditions.

2. De®nitions and preliminaries

A transferable utility cooperative game on the nonempty ®nite set P of players
is de®ned by a coalitional function V : 2 P ! R satisfying V q† ˆ 0. The
function V speci®es the worth of every coalition S J P.
Given a game P; V †, a payo¨ allocation x A R P is called e½cient, if x P† ˆ
V P†; individually rational, if xi ˆ x fig† b V fig† for all i A P; coalitionally
rational, if x S†bV S† for all S J P; where, by the standard notation, x S† ˆ
P
i A S xi if S 0 q, and x q† ˆ 0. We denote by I P; V † the imputation set (i.e.,
the set of e½cient and individually rational payo¨s), and by C P; V † the core
(i.e., the set of e½cient and coalitionally rational payo¨s) of the game P; V †.
The game P; V † is called superadditive, if S X T ˆ q implies V S W T† b
V S† ‡ V T† for all S; T J P; balanced, if its core C P; V † is not empty; and
totally balanced, if every subgame (i.e., the game obtained by restricting the
player set to a coalition and the coalitional function to the power set of that
coalition) is balanced. Note that totally balanced games are superadditive.
Given a game P; V †, the excess e S; x† :ˆ V S† x S† is the usual mea-
sure of gain (or loss if negative) to coalition S J P if its members depart from
allocation x A R P in order to form their own coalition. Note that e q; x† ˆ 0
for all x A R P , and

C P; V † ˆ fx A R P : e P; x† ˆ 0; e S; x† a 0 ES H Pg;

i.e., the core is the set of allocations which yield nonpositive excess for all
coalitions.
We say that allocation y dominates allocation x via coalition S if y S†aV S†
and yk > xk Ek A S. Note that an allocation can be dominated only via co-
alitions having positive excess at that allocation. The (nonempty) core C of a
game is called stable if for every imputation x A InC there exists a core allo-
cation y A C and a coalition S such that y dominates x via S.
Given two ®nite sets S and T, we call m J S  T an S; T†-assignment, if
it is a bijection from some S 0 J S to some T 0 J T such that jS 0 j ˆ jT 0 j ˆ
min jSj; jTj†. Trivially, m ˆ q if S ˆ q or T ˆ q. We shall write i; j† A m
as well as m i† ˆ j. We denote by P S; T† the set of all S; T†-assignments.
Obviously, P S; T† ˆ fqg if S ˆ q or T ˆ q.
A game P; V † is called an assignment game if there exists a partition P ˆ
I W J, I X J ˆ q, of the player set and a nonnegative matrix A ˆ ‰aij Ši A I ; j A J
such that
X
V S† ˆ VA S† :ˆ max aij ES J P:
mAP SXI ; SXJ †
i; j† A m

By adding dummy player(s) (i.e. zero rows/columns to the matrix), we can

assume without loss of generality that there are the same number of players of
both types (i.e. the underlying matrix is square). It will be convenient to identify
180 T. Solymosi, T. E. S. Raghavan

the players with the row/column indices. So we shall use N ˆ f1; 2; . . . ; ng for
the set of indices, while I ˆ f1; 2; . . . ; ng and J ˆ f1 0 ; 2 0 ; . . . ; n 0 g for the set of
row and column players, respectively. In other words, we put a prime ( 0 ) on the
index j to distinguish the j-th column player j 0 A J from the j-th row player
j A I . Coalitions of the type fi; j 0 g are called mixed-pair coalitions.
Throughout the paper we assume that the rows and columns of the under-
lying matrix A are arranged suchP that the diagonal assignment f i; i† : i A Ng is
n
maximal in A, i.e., VA I W J† ˆ iˆ1 aii .
To emphasize the bipartite nature of assignment games, we shall write the
payo¨ allocations as u; v† A R n  R n . Let us introduce the notation

eij u; v† :ˆ aij ui vj i; j A N

for the excess of coalition fi; j 0 g at allocation u; v†. We associate with u; v† A

R n  R n the n  n excess matrix E u; v† ˆ ‰eij u; v†Ši; j A N .
The (total) balancedness of assignment games was proved by Shapley and
Shubik (1972). One key point in their characterization of the core of assign-
ment games is that, besides e½ciency, it su½ces to require rationality only for
single-player and mixed-pair coalitions. Formally,

C VA † ˆ f u; v† A I VA † : eii u; v† ˆ 0; eij u; v† a 0 Ei; j A Ng;

where for the imputation set we clearly have

( )
Xn
n n
I VA † ˆ u; v† A R  R : ekk u; v† ˆ 0; ui b 0; vi b 0 Ei A N :
kˆ1

3. Assignment games with stable core

We say that a matrix A has dominant diagonal if all of its diagonal entries are
row and column maximums, i.e., aii b aij and aii b aji for all i; j A N. Note that
A has dominant diagonal if and only if in the assignment game VA the special
imputations

u; v† :ˆ ui ˆ 0; vi ˆ aii : i A N† u; v† :ˆ ui ˆ aii ; vi ˆ 0 : i A N† 1†

are both core allocations.

Now we are ready to prove our main result.

Theorem 1. Let A be a nonnegative n  n-matrix such that its main diagonal is

an optimal assignment, and let VA be the n ‡ n†-player assignment game induced
by A. Then the following are equivalent:

(i) C VA † is stable;
(ii) A has dominant diagonal.

Proof: i† ) ii† Consider the imputation u; v† de®ned in (1). If u; v† B C VA †

then, by the stability of the core, there exists an allocation u 0 ; v 0 † A C VA † such
that u 0 ; v 0 † dominates u; v† via some coalition S. Since then VA S† must be
Assignment games with stable core 181

positive, S contains at least one column player, say the j-th one. Then vj0 >
vj ˆ ajj , a contradiction to vj0 a ajj that holds in any core allocation u 0 ; v 0 †.
We get that u; v† A C VA †. Interchanging the roles of rows and columns and
repeating the above argument with the other special imputation u; v†, we sim-
ilarly get that u; v† A C VA †. It follows that A has dominant diagonal.
ii† ) i† Let u; v† A I VA †nC VA †. We show that if A has dominant diag-
onal then there exists u 0 ; v 0 † A C VA † such that u 0 ; v 0 † dominates u; v† via some
coalition S. Actually, we show domination via a mixed-pair coalition.

Case 1: eii u; v† 0 0 for some i A N.

Since the sum of the diagonal entries in E u; v† is 0, there exists an i A N such

that eii u; v† > 0, i.e., ui ‡ vi < aii . Since u; v† is an imputation, aii > 0. Then
clearly there is a l A 0; 1† such that ui < 1 l†aii and vi < laii . (Take, e.g.,
aii ui ‡ vi
lˆ .) This implies that the allocation u 0 ; v 0 † :ˆ l u; v† ‡ 1 l†

2aii
u; v† dominates u; v† via the diagonal mixed-pair coalition fi; i 0 g. Since ob-
viously u 0 ; v 0 † A C VA †, the claim follows.

Case 2: eii u; v† ˆ 0 for all i A N.

With such an imputation u; v† we associate a directed graph G u; v† with node

set N such that there is an arc i; j† of ``length'' eij :ˆ eij u; v† from any node i
to any node j 0 i. Note that the arc lengths can be positive, negative or zero.
Let us recall some graph terminology. In a directed graph a path from node
s to node t, or an s; t†-path for short, is a sequence s; i1 †, i1 ; i2 †; . . . ; ik ; t† of
arcs. The path is called simple if the visited nodes s; i1 ; . . . ; ik ; t are all distinct.
A cycle is a set of arcs such that exactly one arc goes in and exactly one arc
goes out from each visited node.
Since the diagonal is a maximal assignment in A, the graph G u; v† does not
contain a cycle of positive length. Suppose not, andP the cycle i1 ; iP 2 †; . . . ; ik ; i1 †
k k
is of positive length. Then, with ik‡1 ˆ i1 , we get hˆ1 eih ih‡1 ˆ hˆ1 aih ih‡1
Pk Pk Pk Pk
u
hˆ1 ih v
hˆ1 ih‡1 ˆ a
hˆ1 ih ih‡1 a
hˆ1 ih ih > 0, a contradiction to the op-
timality of the diagonal assignment. (Notice that the graph G u; v† contains a
cycle of zero length if and only if the diagonal is not the only maximal assign-
ment in A.) It follows that for any node k A N, the numbers

lk :ˆ the length of the longest path ending in k; dk :ˆ maxf0; lk g

are both well de®ned. Moreover, dk is nonnegative.

Since A has dominant diagonal, from

ui ‡ vi ˆ aii b aij ˆ ui ‡ vj ‡ eij ; uj ‡ vj ˆ ajj b aij ˆ ui ‡ vj ‡ eij

we get for any i 0 j,

vi vj b eij ; uj ui b eij : 2†

It follows from the telescopic nature of these inequalities that for any two
nodes s and t in G u; v†,
182 T. Solymosi, T. E. S. Raghavan

minfut us ; v s vt g b the length of the longest s; t†-path: 3†

Let us de®ne the allocation u 0 ; v 0 † by ui0 ˆ ui di and vi0 ˆ vi ‡ di for all

i A N. We claim that u 0 ; v 0 † is in the core. Obviously, ui0 ‡ vi0 ˆ aii and vi0 b 0
for all i A N. To see ui0 b 0 in case di > 0, let k be a node such that there is
a k; i†-path of length di . Then from (3) we get ui uk b di , hence ui0 ˆ ui
di b uk b 0. Obviously, ui0 b 0 in case di ˆ 0. It remains to see ui0 ‡ vj0 b aij
for all i 0 j A N. By de®nitions, ui0 ‡ vj0 ˆ ui ‡ vj ‡ dj di and aij ˆ ui ‡ vj ‡ eij ,
so we only need to show dj b di ‡ eij . If di ˆ 0, we are done by noting that arc
i; j† is itself a path ending in node j, so dj b lj b eij . If di > 0, take a longest
path ending in node i. Its length is di ˆ li . Adding to this path arc i; j† gives a
(not necessarily simple) path ending in j of length di ‡ eij , so dj b lj b di ‡ eij .
Therefore, u 0 ; v 0 † is indeed a core allocation.
Since u; v† is not in the core, there is at least one positive arc in the graph
G u; v†. Thus, di > 0 for at least one node i. We claim that there exists an arc
p; q† of positive length such that dq ˆ epq . Let r be a node for which dr > 0,
and let us consider a path of length dr that ends in r. Suppose this path starts
from node p. We can assume without loss of generality that the path is simple
( just leave out the cycles of zero length, if any) and minimal with respect to
inclusion (i.e., there are no paths of zero length ending in p). Clearly, dp ˆ 0
for the starting node p, and our path is a longest p; r†-path. It follows that for
any node t on this path the length of the subpath from p to t is positive. In
particular, the ®rst arc in this path, ending in node q, has positive length.
Therefore, we indeed have an arc p; q† such that dp ˆ 0 and dq ˆ epq > 0.
We ®nish the proof by constructing a core allocation u 00 ; v 00 † which domi-
nates u; v† via this particular mixed-pair coalition f p; q 0 g. To this end, let

1
e :ˆ minfepq ; minfvj0 : vj0 > 0gg:
2
Clearly, e > 0. De®ne u 00 ; v 00 † by

ui00 :ˆ ui0 ‡ e and vi00 :ˆ vi0 e if vi0 > 0

ui00 :ˆ ui0 ˆ aii † and vi00 :ˆ vi0 if vi0 ˆ 0

for all i A N.
We claim that u 00 ; v 00 † is in the core. Since u 0 ; v 0 † A C, we obviously have
ui00 b 0, vi00 b 0 and ui00 ‡ vi00 ˆ aii for all i A N. It remains to check that
ui00 ‡ vj00 b aij for all i 0 j. When both vi0 and vj0 are positive or when both are
zero, we have ui00 ‡ vj00 ˆ ui0 ‡ vj0 b aij . If vi0 > 0 and vj0 ˆ 0 then ui00 ‡ vj00 ˆ
ui0 ‡ e ‡ vj0 b aij ‡ e > aij . On the other hand, if vi0 ˆ 0 and vj0 > 0 then ui00 :ˆ
ui0 ˆ aii and ui00 ‡ vj00 ˆ ui0 ‡ vj0 e > aii b aij . Thus, u 00 ; v 00 † is indeed a core
allocation.
Finally, we show that u 00 ; v 00 † dominates u; v† via the mixed-pair coalition
f p; q 0 g. Firstly, since epq > 0, (2) implies vp > vq b 0, so vp0 ˆ vp † > 0. Then
dp ˆ 0 gives up00 ˆ up0 ‡ e ˆ up dp ‡ e > up . Secondly, it follows from dq ˆ epq
that vq0 ˆ vq ‡ dq > 0. Then e a 12 epq gives vq00 ˆ vq0 e b vq ‡ 12 epq > vq .
Thirdly, by de®nitions, up00 ‡ vq00 ˆ up0 ‡ vq0 ˆ up ‡ vq ‡ epq ˆ apq . Thus, u 00 ; v 00 † A
C indeed dominates u; v† via the mixed-pair coalition f p; q 0 g.
Assignment games with stable core 183

4. Assignment games with large core

A (balanced) game P; V † is said to have a large core if for every coalitionally

rational allocation y there is a core allocation x A C P; V † such that xk a yk for
all k A P. Sharkey (1982) proved that largeness implies stableness of the core.
He also proved that if a game has a large core then the totally balanced cover
of the game is exact. For assignment games this means that largeness of the
core implies exactness of the game. This latter property is de®ned as follows. A
(balanced) game P; V † is said to be exact if for every coalition S there is a core
allocation x A C P; V † such that x S† ˆ V S†. A related property, weaker than
largeness of the core but (for totally balanced games) stronger than exactness,
was investigated by Kikuta and Shapley (1986). They showed that this prop-
erty is also su½cient for stability of the core. A (balanced) game P; V † is said
to be extendable if any core allocation of any subgame can be extended to a
core element of P; V †.
We say that matrix A has doubly dominant diagonal if aii ‡ ajk b aik ‡ aji
for all i; j; k A N. Notice that this property is restrictive only if i is distinct from
j and k. Also note that having a dominant diagonal and having a doubly dom-
inant diagonal are independent properties, i.e., a matrix can have a dominant
but not a doubly dominant diagonal and vice versa.
Now we are ready to prove the main result in this section.

Theorem 2. Let A be a nonnegative n  n-matrix such that its main diagonal is

an optimal assignment, and let VA be the n ‡ n†-player assignment game induced
by A. Then the following are equivalent:

(i) C VA † is large;
(ii) VA is extendable;
(iii) VA is exact;
(iv) A has dominant and doubly dominant diagonal.

Proof: i† ) ii† It holds for any TU-game, see (Kikuta and Shapley, 1986) or
the proof of Proposition 1 in (Biswas et al., 1999).
ii† ) iii† It is straightforward to establish for totally balanced games, see
the proof of Proposition 2 in (Biswas et al., 1999).
iii† ) iv† Let VA be exact.
Take the coalition J of all column players. Then there is a u; v† A C VA †
such that u; v† J† ˆ v J† ˆ VA J† ˆ 0, thus vj ˆ 0 for all j A J. It follows that
ui ˆ aii for every row player i, i.e., u; v† ˆ u; v†. Thus, the core contains one
of the special imputations in (1). The mirror argument gives that the core con-
tains the other special imputation in (1). Therefore, the underlying matrix A has
dominant diagonal.
As we remarked above, when we check that our matrix A has a doubly
dominant diagonal we can assume without loss of generality that i 0 j and
i 0 k. Take coalition f j; k 0 g. Then there is a u; v† A C VA † such that uj ‡ vk ˆ
ajk . Adding it to ui ‡ vi ˆ aii and using coalitional rationality of u; v† give
aii ‡ ajk ˆ ui ‡ vk ‡ uj ‡ vi b aik ‡ aji . Thus, A has doubly dominant diagonal.
iv† ) i† Let A have dominant and doubly dominant diagonal.
Take an allocation u 00 ; v 00 † that is coalitionally rational in VA . Then ui00 b 0,
vj00 b 0 and eij00 :ˆ eij u 00 ; v 00 † a 0 for all i; j A N. Let us de®ne u 0 ; v 0 † by ui0 :ˆ
ui00 minfui00 ; minj eij00 †g and vi0 :ˆ vi00 for all i A N. Clearly, u 0 ; v 0 † is coali-
184 T. Solymosi, T. E. S. Raghavan

tionally rational, and u0i a ui00 for all i A N. Moreover, for each row player i,
either ui0 ˆ 0 or minj eij0 † ˆ 0, where eij0 :ˆ eij u 0 ; v 0 †. Now, let us decrease
the payo¨ of each column player as much as possible without loosing coali-
tional rationality, i.e., let us de®ne allocation u; v† by ui :ˆ ui0 and vi :ˆ vi0
minfvi0 ; minj eij0 †g for all i A N. As before, u; v† is coalitionally rational,
vi a vi0 for all i A N, moreover, for each column player i 0 , either vi ˆ 0 or
minj eij † ˆ 0.
We claim that this u; v† is a core allocation. Since it is coalitionally rational,
we are done in case eii ˆ 0 for all i A N. Suppose eii < 0 for some i A N. Then at
least one of ui and vi is positive, because A is nonnegative. Let us suppose that
ui ˆ 0. Then for column player i 0 , vi > 0 and eii < 0, so there is a row player
j 0 i such that eji ˆ 0, i.e., uj ‡ vi ˆ aji . Subtracting this from ui ‡ vi > aii gives
0 b uj > aii aji , a contradiction to A having dominant diagonal. We obtain
that if eii < 0 for some i A N then both ui and vi are positive. Then however,
there must be indices j 0 i and k 0 i such that eji ˆ 0 and eik ˆ 0. Since u; v† is
coalitionally rational, ejk a 0. It follows that eii ‡ ejk < 0 ˆ eik ‡ eji , a contra-
diction to A having doubly dominant diagonal. Therefore, we must have eii ˆ 0
for all i A N, so u; v† is indeed a core allocation that is componentwise less
than or equal to the arbitrarily chosen coalitionally rational allocation u 00 ; v 00 †.
Hence, C VA † is large.

5. Convex and subconvex assignment games

A game P; V † is called convex if for all S; T J P the coalitional function

satis®es V S† ‡ V T† a V S W T† ‡ V S X T†. Sharkey (1982) proved that
subconvexity, a weaker version of convexity, also implies largeness and hence
stability of the core. Here we use the following alternative de®nition of sub-
convexity that was given by Gellekom et al. (1999).
Given a game P; V †, a bijective map p : P ! f1; 2; . . . ; jPjg will be called
an enumeration of P. The set of predecessors of i A P is de®ned by Predp i† ˆ
f j A P : p j† < p i†g. Given an enumeration p of P, let the payo¨ vector y p
be given by

yip :ˆ maxfV Q W i† V Q† : Q J Predp i†; Q W i 0 Pg:

Namely, yip is the maximal marginal contribution of player i with respect to

a subset of his predecessors. The game P; V † is called subconvex if y p P† a
V P† for all enumeration p of the player set P.
Now we are ready to prove the main result in this section.

Theorem 3. Let A be a nonnegative n  n-matrix such that its main diagonal is

an optimal assignment, and let VA be the n ‡ n†-player assignment game induced
by A. Then the following are equivalent:

(i) VA is convex;
(ii) VA is subconvex;
(iii) A is a diagonal matrix (i.e., aij ˆ 0 if i 0 j).

Proof: i† ) ii† Straightforward (cf. Sharkey, 1982).

ii† ) iii† Suppose A is not diagonal and a12 > 0. Consider an enumeration
Assignment games with stable core 185

p which speci®es the following order of the players: 1; 2 0 ; 2; 1 0 ; . . .†. Clearly,

y1p ˆ 0 and y2p0 ˆ a12 . Since 2 0 is a predecessor of his optimally matched partner
2, we get y2p b a22 . Similarly, column player 1 0 can secure for himself at least
a11 by joining his optimally matched partner 1 who is his predecessor. The same
reasoning and the nonnegativity of y p give that for any pair f j; j 0 g of players
on the diagonal, yjp ‡ yjp0 b ajj holds. It follows that y p P† b a12 ‡ VA P† >
VA P†, a contradiction to subconvexity.
iii† ) i† If A is a diagonal matrix, the value P of a coalition S is simply the
sum of the diagonal entries in S, i.e., VA S† ˆ fi; i 0 gJS aii . Therefore, VA is an
additive set function and so the game is convex.

References

[1] Aumann RJ, Maschler M (1964) The bargaining set for cooperative games. In: Advances
in Game Theory, M Dresher, LS Shapley, AW Tucker (eds.), Princeton University Press,
Princeton, N.J., 443±476
[2] Biswas A, Parthasarathy T, Potters JAM, Voorneveld M (1999) Large cores and exactness.
Games and Economic Behavior 20:1±12
[3] Biswas A, Ravindran G, Parthasarathy T (2000) Stability and largeness of the core of sym-
metric games. International Journal of Game Theory 29:11±22
[4] Davis M, Maschler M (1967) Existence of stable payo¨ con®gurations for cooperative games.
In: Essays in mathematical economics in honour of Oskar Morgenstern, Shubik M (ed.)
Princeton University Press, Princeton, N.J., 39±52
[5] Gellekom JRG van, Potters JAM, Reijnierse JH (1999) Prosperity properties of TU-games.
International Journal of Game Theory 28:211±227
[6] Kikuta K, Shapley LS (1986) Core stability in n-person games. Mimeo, July 1986
[7] Lucas WF (1968) A game with no solution. Bulletin of the American Mathematical Society
74:237±239
[8] Neumann J von, Morgenstern O (1944) Theory of Games and Economic Behavior. Princeton
University Press, Princeton, New Jersey
[9] Schmeidler D (1972) Cores of exact games I. Journal of Mathematical Analysis and Appli-
cations 40:214±225
[10] Shapley LS (1971) Cores of convex games. International Journal of Game Theory 1:11±26
[11] Shapley LS, Shubik M (1972) The assignment game I: The core. International Journal of
Game Theory 1:111±130
[12] Sharkey WW (1982) Cooperative games with large cores. International Journal of Game
Theory 11:175±182
[13] Solymosi T (1999) On the bargaining set, kernel and core of superadditive games. Interna-
tional Journal of Game Theory 28:229±240
[14] Solymosi T, Raghavan TES (1994) An algorithm for ®nding the nucleolus of assignment
games. International Journal of Game Theory 23:119±143