The Lattice Structure of The Set of Stable Outcomes of The Multiple Partners Assignment Game
The Lattice Structure of The Set of Stable Outcomes of The Multiple Partners Assignment Game
The Lattice Structure of The Set of Stable Outcomes of The Multiple Partners Assignment Game
9
999
99
1. Introduction
The multiple partners assignment game of our title is one of the models for a
two-sided matching market presented in Sotomayor (1992). The participants
belong to two ®nite and disjoint sets, which will be denoted by F and W. For
each pair
fi ; wj in FxW there is a non-negative number, aij , representing the
gain that the coalition f fi ; wj g can get if fi and wj form a partnership. Each
* This work has been partially supported by S. Guggenheim Foundation, FAPESP and by the
Department of Economics of University of Pittsburgh.
I am grateful to David Gale for his careful reading of this paper, his comments and sugges-
tions. I am also indebted to Alvin Roth for his helpful suggestions in an earlier draft of this work
and to an anonymous referee.
568 M. Sotomayor
player has a quota, that is, a positive integer number representing the maxi-
mum number of partnerships he/she can enter. An outcome for this game
is any set of partnerships
fi ; wj which do not violate the quotas of the
players (a matching), together with some splitting uij and vij of the gain aij .
Thus, uij vij aij if fi and wj form a partnership.
Imagine an economic scenario in which the sets F and W are sets of ®rms
and workers, respectively. The amount of income which fi and wj can gener-
ate if they work together is aij . The problem is then to split aij among fi and
wj . That is, if ®rm fi hires worker wj at a salary vij it will receive a pro®t of
uij aij ÿ vij .
The natural questions are then:
Appendix I. In Sotomayor (1992) we also show the existence, but not the
uniqueness, of an optimal stable payo¨ for each side of the market. In the
present paper we show the (to us) rather unexpected result that, again, stable
payo¨s form a convex and compact lattice and consequently there exists a
unique optimal stable payo¨ for each side of the market (Theorems 3 and 4).
The polarization of interests between the two sides within the whole set of
stable outcomes does not carry over to our model, when we compare the total
payo¨s of the players (see Example 3 of Sotomayor (1992)). However we
prove here that this result holds when we compare the individual payo¨s of
the players in some convenient way (Proposition 3).
Indeed, to prove the results above, we face the problem that the uij 's and
vij 's are indexed according to the current matching, which may di¨er from one
stable outcome to the other. Thus a player with a quota greater than one may
face the problem of comparing two unordered sets of payo¨s!
In shirt, it is not clear how to solve the following di½culties:
The solution is obtained here via Theorem 1. This result has no parallel in
the special case of Shapley and Shubik, since its conclusions are addressed to
the players who form more than one partnership. Its statement makes use of
the following terminology: a matching is called optimal if it maximizes the
gain of the whole set of players; a partnership
fi ; wj is called nonessential if
it occurs in some but not all optimal matchings and essential if it occurs in all
optimal matchings. Therefore two matchings di¨er only by their nonessential
partnerships. Theorem 1 states that:
(i) In every stable outcome a player gets the same payo¨ in any nonessential
partnership; furthermore this payo¨ is less than or equal to any other payo¨
the player gets under the same outcome.
(ii) Given a stable outcome
u; v; x and a di¨erent optimal matching x 0 , we can
reindex the uij 's and vij 's according to x 0 and still get a stable outcome.
non-obvious result is that the matching obtained in this way is not only feasi-
ble, but optimal. Moreover, the resulting outcome gives to the players of the
opposite side their lowest payo¨s. In this case the proof is not so straightfor-
ward as the previous proofs, instead it makes use of a sort of decomposition of
the set of nonessential partnerships (Lemma 3).
This paper is organized as follows: Section 2 describes the model; section 3
gives the proof of Theorem 1; the results about the algebraic structure of the
set of stable payo¨s are proved in section 4. An illustrative example of the
results of the paper is presented at the end of section 4. In Appendix I we
prove two important results for the present work which were already proved
in Sotomayor (1992). In Appendix II we discuss the robustness of the con-
clusions of Theorem 1.
There are two disjoint and ®nite sets of players, F f f0 ; f1 ; . . . ; fmÿ1 g and
W fw0 ; w1 ; . . . ; wnÿ1 g, where f0 and w0 are dummy players. Each player
from a given set is allowed to form partnerships with the players from the
opposite set. Each player fi A F may form at most ri partnerships and each
player wj A W may form at most sj partnerships. For each pair
fi ; wj there is
a non-negative number, aij , representing the gain that fi and wj can generate if
they work together. If a partnership
fi ; wj is formed, fi will receive a payo¨
uij V 0 and wj will receive a payo¨ vij aij ÿ uij V 0. The dummy players are
included for technical convenience. We assume that ai0 a0j 0 for all fi A F
and wj A W . As for the quotas, a dummy player can form as many partner-
ships as needed to ®ll up the quotas of the non-dummy players. Thus, for ex-
ample, if player fi A F has 5 un®lled positions, then fi forms 5 partnerships
with w0 . This market is denoted by M
F ; W ; a; r; s or simply by M
a.
A feasible matching is a set of partnerships of the kind
fi ; wj ;
fi ; w0 or
f0 ; wj , for
fi ; wj in FxW , Which do not violate the quotas of the players.
Formally:
If xij > 0 (resp. xij 0) it means that fi and wj are (resp. are not) matched
at x. The set of all fi 's partners at x, with xi0 repetitions of w0 , is denoted
by C
fi ; x. Thus, C
fi ; x fw1 ; w2 ; w3 ; w0 ; w0 g denotes that player fi , with
quota ri 5, has partners w1 ; w2 and w3 under the matching x and has two
positions un®lled. C
wj ; x is similarly de®ned for all wj A W . We will repre-
sent by jAj the number of elements, including the repetitions, of the set A.
Under this notation jC
fi ; xj ri and jC
wj ; xj sj , for all non-dummy
players fi A F and wj A W .
An outcome for this market will be determined by specifying a matching
and the way in which the income within each partnership is divided among its
members. That is:
The lattice structure of the set of stable outcomes 571
De®nition 4. The feasible outcome
u; v; x is stable if mi nj V aij for all
fi ; wj
with xij 0.
In this section we will prove our key result. For its statement and proof
we de®ne M 1 f
fi ; wj A FxW ; xij 0 0g and M 0 1 f
fi ; wj A FxW ; xij0 0 0g.
We also use the following abbreviations: C
fi ; x 1 C
fi ; C
fi ; x 0 1 C 0
fi ;
C
wj ; x 1 C
wj and C
wj ; x 0 1 C 0
wj .
X X
uij vij aij :
fi ; wj A MÿM 0
fi ; wj A MXM 0
Here, the second equality is due to the feasibility of
u; v; x. But
X X X X X
uij vij uij vij
fi ; wj A MÿM 0 fi A F wj A C
fi ÿC 0
fi wj A W fi A C
wj ÿC 0
wj
X X X X
V mi nj
1
fi A F wj A C
fi ÿC 0
fi wj A W fi A C
wj ÿC 0
wj
X X
jC
fi ÿ C 0
fi jmi jC
wj ÿ C 0
wj jnj
fi A F wj A W
X X
jC 0
fi ÿ C
fi jmi jC 0
wj ÿ C
wj jnj
fi A F wj A W
X
mi nj
fi ;wj A M 0 ÿM
X
V aij
2
fi ;wj A M 0 ÿM
where inequality (1) follows from the de®nitions mi min uij and nj min vij
and inequality (2) follows from the stability of
u; v; x. Therefore,
X X X
aij xij V aij aij
fi ; wj A FxW
fi ; wj A M 0 ÿM
fi ; wj A MXM 0
X
aij xij0
3
fi ; wj A FxW
u 0 ; v 0 ; x 0 is clearly stable. 9
It is easy to see that the set of stable payo¨s for a given matching x forms
a lattice. That is, if
u; v; x and
u 0 ; v 0 ; x are stable outcomes, and if uij
maxfuij ; uij0 g and vij minfvij ; vij0 g, for all matched pairs
fi ; wj , then
u ; v; x is stable and, symmetrically, so is
u; v ; x.
If
u; v and
u 0 ; v 0 are associated to di¨erent matchings then there is no
meaning to consider supfu; u 0 g or inffu; u 0 g, since u and u 0 are de®ned on
di¨erent sets of indices. However, by using Theorem 1, we can represent the
set of stable payo¨s of a player as a vector in a Euclidean space, whose
dimension is the quota of the given player. This representation is independent
of the matching. By ordering the players in F (resp. W), we can immerse the
set of stable payo¨s for these players in a Euclidean space, whose dimensions
is the sum of the quotas of all players in F (resp. W). Then the natural partial
order of this Euclidean space induces a partial order in the set of stable
payo¨s. Now it makes sense to ask if the set of stable payo¨s is a compact and
convex lattice.
Before proceeding, let us pause to consider what we have learned from
Theorem 1. Consider a stable outcome
u; v; x. If player fi has quota 5 and
forms partnerships with w1 ; w2 ; w3 ; w4 and w5 , then the set of payo¨s of fi
is given by fui1 ; ui2 ; ui3 ; ui4 ; ui5 g. In case all partnerships are essential, we can
®x any ordering we want and represent all sets of payo¨s of player fi by vec-
tors in the same Euclidean space. Recall that part (a) of Theorem 1 assures
that under a given stable outcome all the nonessential partners of a player
contribute with the same payo¨. Therefore, for instance, if the only essential
partners of player fi are w1 and w2 , we can choose the following ordering for
the 5-tuples of uij 's:
ui1 ; ui2 ; mi ; mi ; mi , where mi mi3 mi4 mi5 . Thus under
any other stable outcome
u 0 ; v 0 ; x 0 the vector of payo¨s of player fi can be
represented by
ui10 ; ui2 0
; mi0 ; mi0 ; mi0 . Moreover, if
u 0 ; v 0 ; x 0 is the outcome as-
sociated with
u; v; x by Theorem 1-b, ui10 ui1 ; ui2 0
ui2 and mi0 mi . That
is, whatever optimal matching is being considered, the vector of payo¨s for
player fi under
u; v; x is the same as under
u 0 ; v 0 ; x 0 .
From now on the arrays of numbers u and v will be represented by a pair
of vectors, still denoted by u and v. Thus we will keep the same notation
u; v; x of the outcome under the new representation. Theorem 1 has now the
desired meaning:
574 M. Sotomayor
De®nition 7. We say that the payo¨ vector
u; v is stable if there is some fea-
sible matching x such that
u; v; x is a stable outcome. In this case x is said to
be compatible with
u; v and vice-versa.
Proof: The set of stable payo¨s is the same for any optimal matching. Let x be
an optimal matching. The set of stable payo¨s is the solution of a system of
linear non strict inequalities associated with x, so it is closed and convex. That
it is bounded follows from the fact that for all matched pairs
fi ; wj under x,
0 U uij U aij and 0 U vij U aij :
Hence the set of stable payo¨s is convex and compact. 9
a) If
fi ; wj is an essential partnership, uij > uij0 if and only if vij < vij0
b) If
fi ; wj is a nonessential partnership, mi > mi0 if and only if nj < nj0 .
Proof: The result follows easily from the fact that if
fi ; wj is essential then
uij vij aij uij0 vij0 , and if
fi ; wj is nonessential then
u; v and
u 0 ; v 0
are compatible with any optimal matching, so mi nj aij mi0 nj0 . 9
Observe that Lemma 1-a) does not require the existence of more than one
optimal matching.
By Lemma 1-b
mi V mi0 , nj0 V nj ; if
fi ; wj is nonessential:
2
Given the stable payo¨s
u; v and
u 0 ; v 0 , de®ne
u ; v and
u; v as
follows: For all essential partnerships
fi ; wj :
It is clear that uij V mi , uij V mi , vij V nj and vij V nj , for all
fi ; wj .
It now makes sense to ask: If
u; v and
u 0 ; v 0 are stable payo¨s, are
u ; v and
u; v also stable payo¨s? The answer is a½rmative:
Theorem 2. Let
u; v and
u 0 ; v 0 be stable payo¨s. Then the payo¨s
u ; v
and
u; v , as de®ned above, are stable.
Proof: We will show that
u ; v is stable; the other assertion follows dually.
Let x be some optimal matching. We already know that
u; v and
u 0 ; v 0 are
compatible with x. To see that there is no pair causing instabilities, we have to
show that mi nj V aij , for all fi A F and wj A W with xij 0. Due to the
stability of
u; v and
u 0 ; v 0 the only cases we have to check are those where
We have either:
mi nj mi nj0 V mi0 nj0 V aij or mi nj mi0 nj V mi nj V aij :
Since the payo¨s are clearly non-negative, it remains to show that if xij 0 0
the gain of
fi ; wj is aij . But it is immediate from Lemma 1-a that, if
fi ; wj is
essential, then
uij vij uij vij aij or uij vij uij0 vij0 aij :
Given two stable payo¨s
u; v and
u 0 ; v 0 , the construction of the payo¨s
u ; v and
u; v leads naturally to the following relations:
supF f
u; v;
u 0 ; v 0 g inf W f
u; v;
u 0 ; v 0 g and
u; v inf F f
u; v;
u 0 ; v 0 g
supW f
u; v;
u 0 ; v 0 g.
Theorem 3. The set of stable payo¨s is a convex and complete lattice under the
partial orders VF and VW .
Theorem 4. There exists one and only one F (resp. W)-optimal stable payo¨.
Furthermore, every wj A W (resp. fi A F ) weakly prefers every stable payo¨ the
F (resp. W)-optimal stable payo¨.
Proof: We will show that a stable payo¨ is the maximal element of the lattice
under VF if and only if it is an F-optimal stable payo¨. Since every complete
lattice has one and only one maximal element, we will have proved the exis-
tence and uniqueness of the F-optimal stable payo¨. The other assertion will
follow dually. Then let
u ; vÿ be the maximal element of the lattice under
VF . Let
u; v be any F-optimal stable payo¨. The maximality of
u ; vÿ
implies that
u ; vÿ VF
u; v, so
(1) u
ij V uij for all essential partnerships
fi ; wj and
(2) m
i V mi for all fi A F . Hence
(3) s
i V si for all fi A F .
Our last result is related to situations in which all the players on a given
side have to choose the partnerships corresponding to their highest payo¨s,
respecting their quotas, among those presented to them by two stable out-
comes. It is not obvious that the resulting set of partnerships will be feasible
and also an optimal matching. Furthermore, this matching gives to the play-
ers of the opposite side their partners corresponding to their lowest payo¨s.
We need the following:
Proof: We will prove that x is optimal. The other assertion follows dually.
Then, by Proposition 1 it is enough to show that
u ; v; x is a stable out-
come.
First observe that C
fi ; x C
fi ; x if fi A F 1 and C
fi ; x C
fi ; x 0
if fi B F 1 . Thus jC
fi ; x j ri for all fi . By the construction of x it follows
that all essential partners of wj A W are also matched to wj under x . From
Lemma 2 it follows that all nonessential partners of wj A W 2 (resp. wj B W 2
are in F 1 (resp. F 2 W F 0 , so wj is matched at x according to x (resp. x 0 ).
Hence jC
wj ; x j jC
wj ; xj sj (resp. jC
wj ; x j jC
wj ; x 0 j sj ) and so
x is feasible.
To see that
u ; v; x is feasible use the de®nition of x and
u ; v to
get that the gain of the pair
fi ; wj is aij when xij > 0. (One can consider the
cases i)
fi ; wj is essential, ii)
fi ; wj is nonessential with fi A F 1 and iii)
fi ; wj is nonessential with fi B F 1 ).
578 M. Sotomayor
To see what we have learned with the results presented here, consider the
following example:
0 1 0 1
0 0 0 0 0 0 0 0 0 0 0 0
B C B C
B0 1 1 1 0 0C B0 0 1 0 1 1C
xB C x0 B C
B0 0 1 0 C
0 0A B0 0 1 0 0 0C
@ @ A
0 0 0 0 1 1 0 1 0 1 0 0
0 1
0 0 0 0 0 0
B C
B0 0 1 1 1 0C
x 00 B
B0
C
@ 0 1 0 0 0C
A
0 1 0 0 0 1
P P 0
We
P
can compute that
fi ; wj A FxW aij xij
fi ; wj A FxW aij xij
00
a x
fi ; wj A FxW ij ij 25. The essential partnerships are:
f1 ; w2 and
f2 ; w2 . The nonessential partnerships are:
f1 ; w1 ;
f1 ; w3 ;
f3 ; w4 ;
f3 ; w5 under the matching x,
f1 ; w4 ;
f1 ; w5 ;
f3 ; w1 ;
f3 ; w3 under the matching x 0 and
f1 ; w3 ;
f1 ; w4 ;
f3 ; w1 ;
f3 ; w5 under the matching x 00 .
From Theorem 1 we can conclude that any outcome
u; v; x with u11 0
u13 , or u34 0 u35 , or u11 > u12 , for example, is unstable. We can verify the
conclusions of Theorem 1 in the following stable outcomes:
u; v; x;
u 0 ; v 0 ; x 0
and
u 00 ; v 00 ; x 00 , where the payo¨s are given by:
The lattice structure of the set of stable outcomes 579
u 0
u12
0
5; m10 0; m10 0; u22
0
4; m30 4; m30 4;
u 00
u12
00
0; m100 0; m100 0; u22
00
3; m300 4; m300 4;
v 0
n10 2; v12
0 0
0; v22 0; n30 3; n40 1; n50 2;
v 00
n100 2; v12
00 00
5; v22 1; n300 3; n400 1; n500 2;
Theorem 1 implies that
u; v;
u 0 ; v 0 and
u 00 ; v 00 are all compatible with the
matchings x; x 0 and x 00 .
Under the vectorial representation, the set of payo¨s for player f1 is
represented by the vector
3; 1; 1 in the ®rst outcome, by
5; 0; 0 in the
second outcome and by
0; 0; 0 in the third one. In this way p1 can ®nd
the supremum and the in®mum of two of these vectors by comparing each
coordinate.
With Theorem 2 we learned to construct new stable payo¨s:
u 1 ; v 1
supF f
u; v;
u 0 ; v 0 g, where u 1
u12 1
5; m11 1; m11 1; u22 1
4; m31 5;
m3 5 and v
n1 1; v12 0; v22 0; n3 2; n4 0; n5 1;
u 2 ; v 2
1 1 1 1 1 1 1 1
Theorem 5 asserts that the players can also choose their best partnerships.
By comparing
u; v with
u 0 ; v 0 the players in F will choose the matching x;
the best partners for the players in W will form the matching x 0 .
To see the decomposition in the set of nonessential partnerships given by
Lemma 2, consider
u; v and
u 0 ; v 0 . Then
F 1 f fi A F ; mi > mi0 and S i 0 fg f f1 ; f3 g;
W 2 fwj A W ; nj0 > vj0 and Sj 0 fg fw1 ; w3 ; w4 ; w5 g and
F 2 f and W 1 f because S 2 f and S2 f:
Note that any nonessential partnership is formed with players in F 1 and
W 2 . Now consider
u 0 ; v 0 and
u 00 ; v 00 . We can see that
F 0 f fi A F ; mi00 mi0 and S i 0 fg f f1 ; f3 g and
W 0 fwj A W ; nj00 nj0 and Sj 0 fg fw1 ; w3 ; w4 ; w5 g: 9
Appendix I
In this section we reproduce the proofs of the existence of stable outcomes and
of Proposition 1. Both results appear in Sotomayor (1992).
Set F 0 F ÿ f f0 g, W 0 W ÿ fw0 g. Then jF 0 j m ÿ 1 and jW 0 j n ÿ 1.
Consider the primal linear programming problem
P1 of ®nding a matrix
xij which
P
(a) maximizes
fi ; wj A FxW aij xij
subject to:
P
(b) Pwj A W 0 xij U ri , for all fi A F 0 ,
0
(c) fi A F 0 xij U sj , for all wj A W and
(d) 0 U xij U 1, for all
fi ; wj A F 0 xW 0 .
subject to:
By the Duality Theorem there is a solution of
P1 and
P1 such that
a
a P
F 0 W W 0 , the payo¨ of the coalition F 0 W W 0 . Furthermore, it can
be shown that there is a matrix
xij with integer entries which is an optimal
solution of
P1 . AnyP such solution will be called P an optimal assignment. It is
easy to check that if
fi ; wj A F 0 xW 0 aij xij V
fi ; wj A F 0 xW 0 aij x0
ij for all feasible
assignments x0 , then x is an optimal assignment.
If
xij is any optimal assignment, then by the Linear Programming Equi-
librium Theorem (see Gale, 1960),
y; z is an optimal dual vector if and only
if
The lattice structure of the set of stable outcomes 581
P
(A) wj A W 0 xij < ri implies yi 0,
P
(B) fi A F 0 xij < sj implies zj 0,
(C) xij 0 implies yi zj V aij and
(D) xij 1 implies yi zj V aij .
Proof: Take any optimal dual vector
y; z for
P1 . De®ne, for all fi A F 0 and
wj A W 0 , vij zj , uij aij ÿ zj if xij 1 and ui0 u0j vi0 v0j 0, where
x
xij is any optimal matching. Then, uij vij aij if xij > 0. Clearly vij V
0 and uij aij ÿ zj V yi V 0, by (D), if xij xij 1. Then
u; v; x is feasible.
To prove stability, suppose xij 0. We have that mi uik for some wk A
C
fi ; x and nj zj . Then, mi nj uik zj V yi zj V aij , where the ®rst
inequality follows from the de®nition of uik and (D) and the last inequality
follows from (C). 9
Lemma A. The feasible outcome
u; v; x is stable if and only if for all feasible
matchings x 0 and all fi A F we have that
X
si V
aij ÿ vij0 xij0
wj A W
where vij0 nj if xij 0 and vij0 vij if xij > 0. Symmetrical result applies to
every wj A W .
Proof:
P Adding P up in P Lemma A yields: P P
0 0 0 0 0
fi A R s i V fi A R wj A S
aij ÿ vij xij V
fi ; wj A RxS aij xij ÿ
fi ; wj A RxS vij xij
P P P P P
aij x 0 ÿ 0 0
fi A F vij xij V
fi ; wj A RxS aij xij ÿ
0
P
fi ; wj A RxS P ij wj A S
0
P P P wj A S
fi A F vij xij
fi ; wj A RxS aij xij ÿ wj A S tj . Then, s i t j V
P 0
fi A R wj A S
fi ; wj A RxS aij xij , which completes the proof. 9
Proof
P of Proposition P 1: Take P R F. P S W in Lemma B. Now
0
fi ; wj A FxW aij xij fi A F s i wj A W t j V
fi ; wj A FxW aij xij for all feasible
0
matchings x . 9
Our conclusions for several optimal matchings are not less robust than our
conclusions for only one optimal matching. This fact is clari®ed below:
0
Theorem 2*. Let
u; P v; x be some stable P outcome. Let0 x P be some feasible
00
matching so that
fi ; wj A FxW aij xij V
fi ; wj A FxW aij xij V
fi ; wj A FxW aij xij
00 0
for all feasible
P matchings x 0Px. (That is, x is a ``second'' optimal matching).
Suppose
fi ; wj A FxW aij xij ÿ
fi ; wj A FxW aij xij0 l. If fi A F and C
fi ; x 0
C
fi ; x 0 , then uij ÿ uik U l for all wj A C
fi ; x ÿ C
fi ; x 0 and all k A C
fi ; x.
Symmetrical results apply to any wj A W .
Proof: If l 0 then x and x 0 are optimal matchings and the result follows
from Theorem 1. Suppose l > 0. Assume that wj A C
fi ; x ÿ C
fi ; x 0 and
uij > uik l for some k A C
fi ; x. Set uij0 uij ÿ l and utm
0
utm for all
ft ; wm 0
fi ; wj , with xtm > 0. Then,
When there are several optimal matchings for the market M
a, we can
obtain a market, M
a 0 , by changing the matrix
aij so that M
a 0 has only
one optimal matching x. Theorem 2* implies that when a 0 is obtained by a
small perturbation of the matrix a, by taking a very small l; uij cannot be
``very di¨erent'' from uik , when
fi ; wj and
fi ; wk are nonessential partner-
ships at x. (That is, juij ÿ uik j < l). Furthermore, if
fi ; wq is an essential
The lattice structure of the set of stable outcomes 583
References
Blair C (1988) The lattice structure of the set of stable matchings with multiple partners. Mathe-
matics of Operations Research 13:619±628
Crawford VP, Knoer EM (1981) Job matching with heterogeneous ®rms and workers. Econo-
metrica 49:437±450
Gale D, Shapley L (1962) College admissions and stability of marriage. American Mathematical
Monthly 69:9±15
Gale D (1960) The theory of linear economic models. McGraw-Hill Book Company, Inc.
Gale D, Sotomayor M (1985) Some remarks on the stable matching problem. Discrete Applied
Mathematics 11:223±232
Kelso ASJr, Crawford VP (1982) Job matching, coalition formation, and gross substitutes.
Econometrica 50:1483±1504
Roth A (1984) Stability and polarization of interest in job matching. Econometrica 52:47±57
Roth A (1985) The college admission problem is not equivalent to the marriage problem. Journal
of Economic Theory 36:277±288
Roth A, Sotomayor M (1990) Two-sided matching. A study in game theoretic modeling and
analysis. Econometric Society Monograph Series, no. 18, Cambridge University Press
Roth A, Sotomayor M (1989) The college admissions problem revisited. Econometrica 57:559±
570
Shapley L, Shubik M (1972) The assignment game fi : The core. International Journal of Game
Theory 1:111±130
Sotomayor M (1992) The multiple partners game. In: Majumdar M (ed.) Equilibrium and dy-
namics: Essays in Honor to David Gale Macmillian, pp. 322±336