Disjoint Stable Matchings

Aadityan Ganesh, Prajakta Nimbhorkar, Geevarghese Philip, Vishwa

Prakash HV
January 6, 2021
Chennai Mathematical Institute

1/39
Outline

ˆ Background
ˆ Stable Matchings
ˆ Gale-Shapley Algorithm
ˆ Extended Gale-Shapley Algorithm
ˆ The Lattice Structure
ˆ Disjoint Stable Matchings
ˆ Disjoint Perfect Matchings
ˆ Disjoint Stable Matchings
ˆ Algorithm to find Disjoint Stable Matchings
ˆ Correctness and Running time of the Algorithm
ˆ Rotations
ˆ Rotational poset
ˆ Properties of Rotational poset
ˆ Analogous Algorithm

2/39
Stable Matchings

Marriage Matching Instance

A marriage matching instance of size n involves two disjoint sets
of size n, the men and the women. Associated with each person
is a strictly ordered preference list containing all the members of
the opposite sex. Person p prefers q to r , where q and r are of
the opposite sex of p, if and only if q precedes r on p’s
preference list.

w4 , w3 , · · · , wn m1 w1 m1 , m3 , · · · , mn
w1 , w6 , · · · , w9 m2 w2 m5 , m2 , · · · , m9
.. .. .. ..
. . . .
w5 , w7 , · · · , wn mn wn m5 , m7 , · · · , mn

Set of men Set of women

3/39
Stable Matchings

Marriage Matching Instance

A marriage matching instance of size n involves two disjoint sets
of size n, the men and the women. Associated with each person
is a strictly ordered preference list containing all the members of
the opposite sex. Person p prefers q to r , where q and r are of
the opposite sex of p, if and only if q precedes r on p’s
preference list.

w4 , w3 , · · · , wn m1 w1 m1 , m3 , · · · , mn
w1 , w6 , · · · , w9 m2 w2 m5 , m2 , · · · , m9
.. .. .. ..
. . . .
w5 , w7 , · · · , wn mn wn m5 , m7 , · · · , mn

Set of men Set of women

3/39
Stable Matchings

Marriage Matching Instance

A marriage matching instance of size n involves two disjoint sets
of size n, the men and the women. Associated with each person
is a strictly ordered preference list containing all the members of
the opposite sex. Person p prefers q to r , where q and r are of
the opposite sex of p, if and only if q precedes r on p’s
preference list.

w4 , w3 , · · · , wn m1 w1 m1 , m3 , · · · , mn
w1 , w6 , · · · , w9 m2 w2 m5 , m2 , · · · , m9
.. .. .. ..
. . . .
w5 , w7 , · · · , wn mn wn m5 , m7 , · · · , mn

Set of men Set of women

3/39
Stable Matchigs

If (m, w ) are matched in a matching M, we say m = pM (w ) and

w = pM (m )
Blocking Pair
A man m and a woman w are said to block a matching M, or
the pair (m, w ) is said to be a blocking pair for M, if m and w
are not partners in M, but m prefers w to pM (m ) and w prefers
m to pM (w ). A matching with no blocking pair is called a stable
matching, and is otherwise unstable.

m1 w1

m2 w2
4/39
Stable Matchigs

If (m, w ) are matched in a matching M, we say m = pM (w ) and

w = pM (m )
Blocking Pair
A man m and a woman w are said to block a matching M, or
the pair (m, w ) is said to be a blocking pair for M, if m and w
are not partners in M, but m prefers w to pM (m ) and w prefers
m to pM (w ). A matching with no blocking pair is called a stable
matching, and is otherwise unstable.

m1 w1

m2 w2
4/39
Stable Matchigs

If (m, w ) are matched in a matching M, we say m = pM (w ) and

w = pM (m )
Blocking Pair
A man m and a woman w are said to block a matching M, or
the pair (m, w ) is said to be a blocking pair for M, if m and w
are not partners in M, but m prefers w to pM (m ) and w prefers
m to pM (w ). A matching with no blocking pair is called a stable
matching, and is otherwise unstable.

m1 w1

m2 w2
4/39
Stable Matchings

Stable Matching
A matching with no blocking pair

Checking stability: O (n2 )

5/39
Stable and Fixed Pairs

Stable Pair
A pair (m, w ) is called as a stable pair if m and w are partners in
at least one stable matching.

Fixed Pair
A pair (m, w ) is called as a fixed pair if m and w are partners in
at all stable matchings.

6/39
Stable and Fixed Pairs

Stable Pair
A pair (m, w ) is called as a stable pair if m and w are partners in
at least one stable matching.

Fixed Pair
A pair (m, w ) is called as a fixed pair if m and w are partners in
at all stable matchings.

6/39
Gale-Shapley Algorithm

Algorithm 1 Gale-Shapley
1: procedure Find stable matching(M)
2: assign each person to be free
3: while some man m is free do
4: w ← first woman on m’s list to whom m hasn’t proposed
5: if w is free then
6: assign m and w to be engaged to each other
7: else
8: if w prefers m to her current matched partner m0 then
9: assign m and w to be engaged and m0 to be free
10: else
11: w rejects m . m remains free
12: end if
13: end if
14: end while
return Stable matching consisting of n engaged pairs
15: end procedure

7/39
Gale-Shapley Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

8/39
Gale-Shapley Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

8/39
Gale-Shapley Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

8/39
Gale-Shapley Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

8/39
Gale-Shapley Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

8/39
Gale-Shapley Algorithm - Key Results

1. Every marriage instance has a stable matching.

2. All possible execution of the Gale-Shapley algorithm yields the
same result.
3. It results in “Man-optimal” stable matching.
Man-optimal: Every man is matched with his most favored
partner among all stable partners.
4. Reversing roles, i.e, women proposing, results in
“Woman-optimal” stable matching.
Woman-optimal: Every woman is matched with her most
favored partner among all stable partners.
5. The man-optimal stable matching is woman-pessimal, and
vice-versa.
9/39
Gale-Shapley Algorithm - Key Results

1. Every marriage instance has a stable matching.

2. All possible execution of the Gale-Shapley algorithm yields the
same result.
3. It results in “Man-optimal” stable matching.
Man-optimal: Every man is matched with his most favored
partner among all stable partners.
4. Reversing roles, i.e, women proposing, results in
“Woman-optimal” stable matching.
Woman-optimal: Every woman is matched with her most
favored partner among all stable partners.
5. The man-optimal stable matching is woman-pessimal, and
vice-versa.
9/39
Gale-Shapley Algorithm - Key Results

1. Every marriage instance has a stable matching.

2. All possible execution of the Gale-Shapley algorithm yields the
same result.
3. It results in “Man-optimal” stable matching.
Man-optimal: Every man is matched with his most favored
partner among all stable partners.
4. Reversing roles, i.e, women proposing, results in
“Woman-optimal” stable matching.
Woman-optimal: Every woman is matched with her most
favored partner among all stable partners.
5. The man-optimal stable matching is woman-pessimal, and
vice-versa.
9/39
Gale-Shapley Algorithm - Key Results

1. Every marriage instance has a stable matching.

2. All possible execution of the Gale-Shapley algorithm yields the
same result.
3. It results in “Man-optimal” stable matching.
Man-optimal: Every man is matched with his most favored
partner among all stable partners.
4. Reversing roles, i.e, women proposing, results in
“Woman-optimal” stable matching.
Woman-optimal: Every woman is matched with her most
favored partner among all stable partners.
5. The man-optimal stable matching is woman-pessimal, and
vice-versa.
9/39
Gale-Shapley Algorithm - Key Results

1. Every marriage instance has a stable matching.

2. All possible execution of the Gale-Shapley algorithm yields the
same result.
3. It results in “Man-optimal” stable matching.
Man-optimal: Every man is matched with his most favored
partner among all stable partners.
4. Reversing roles, i.e, women proposing, results in
“Woman-optimal” stable matching.
Woman-optimal: Every woman is matched with her most
favored partner among all stable partners.
5. The man-optimal stable matching is woman-pessimal, and
vice-versa.
9/39
Extended Gale-Shapley Algorithm

Algorithm 2 Extended Gale-Shapley

1: procedure GS-Extended(M) . M is an SM instance
2: assign each person to be free
3: while some man m is free do
4: w ← first woman on m’s list
5: if some man p is engaged to w then
6: assign p to be free
7: end if
8: assign m and w to be engaged to each other
9: for each successor m0 of m on w ’s list do
10: delete w on m0 ’s list
11: delete m0 on w ’s list . deleting the pair (m0 , w )
12: end for
13: end while
return Stable matching consisting of n engaged pairs
14: end procedure

10/39
Run of Extended GS Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
Run of Extended GS Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1 m3
m3 : w2 w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
Run of Extended GS Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1
m3 : w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
Run of Extended GS Algorithm

m1 : w2 w3 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1
m3 : w1 w3 w3 : m3 m2 m1

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
Run of Extended GS Algorithm

m1 : w2 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1
m3 : w1 w3 w3 : m3 m2

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
Run of Extended GS Algorithm

m1 : w2 w1 w1 : m1 m2 m3
m2 : w3 w1 w2 w2 : m2 m1
m3 : w1 w3 w3 : m3 m2

Men’s Preference Women’s Preference

m1 w1

m2 w2

m3 w3

11/39
GS-lists

MGS-list
Final preference lists generated by the extended Gale-Shapley
algorithm with men as proposers are called as man-oriented
Gale-Shapley lists or MGS-lists.

WGS-list
Final preference lists generated by the extended Gale-Shapley
algorithm with women as proposers are called as woman-oriented
Gale-Shapley lists or WGS-lists.

12/39
GS-lists

MGS-list
Final preference lists generated by the extended Gale-Shapley
algorithm with men as proposers are called as man-oriented
Gale-Shapley lists or MGS-lists.

WGS-list
Final preference lists generated by the extended Gale-Shapley
algorithm with women as proposers are called as woman-oriented
Gale-Shapley lists or WGS-lists.

12/39
GS-lists

GS-list
Intersection of MGS-list and WGS-list.

Note: GS-lists can be obtained by applying man-oriented extended

Gale-Shapley algorithm to get MGS-lists and then, starting with
the MGS-lists, applying woman-oriented extended GS algorithm.

13/39
Extended GS Algorithm Key Results

1. all stable matchings are contained in the GS-lists.

2. no matching (stable or otherwise) contained in the GS-lists

can be blocked by a pair that is not in the GS-lists.

3. In the man-optimal (respectively woman-optimal) stable

matching, each man is partnered by the first (respectively
last) woman on his GS-list, and each woman by the last
(respectively first) man on hers.

14/39
Extended GS Algorithm Key Results

1. all stable matchings are contained in the GS-lists.

2. no matching (stable or otherwise) contained in the GS-lists

can be blocked by a pair that is not in the GS-lists.

3. In the man-optimal (respectively woman-optimal) stable

matching, each man is partnered by the first (respectively
last) woman on his GS-list, and each woman by the last
(respectively first) man on hers.

14/39
Extended GS Algorithm Key Results

1. all stable matchings are contained in the GS-lists.

2. no matching (stable or otherwise) contained in the GS-lists

can be blocked by a pair that is not in the GS-lists.

3. In the man-optimal (respectively woman-optimal) stable

matching, each man is partnered by the first (respectively
last) woman on his GS-list, and each woman by the last
(respectively first) man on hers.

14/39
Extended GS Algorithm Key Results

1. all stable matchings are contained in the GS-lists.

2. no matching (stable or otherwise) contained in the GS-lists

can be blocked by a pair that is not in the GS-lists.

3. In the man-optimal (respectively woman-optimal) stable

matching, each man is partnered by the first (respectively
last) woman on his GS-list, and each woman by the last
(respectively first) man on hers.

14/39
The Lattice Structure

A person x is said to prefer a matching M to a matching M 0 if x

prefers pM (x ) to pM 0 (x ).
Domination
A stable matching M is said to dominate a stable matching M 0 ,
written M  M 0 , if every man has at least as good a partner in
M as he has in M 0 .i.e., every man either prefers M to M 0 or is
indifferent between them. M disjointly dominates M 0 (M ≺ M 0 )
if M  M 0 and M ∩ M 0 = ∅.

m1 :
m2 :
m3 :
m4 :
15/39
Men’s Preference
The Lattice Structure

A person x is said to prefer a matching M to a matching M 0 if x

prefers pM (x ) to pM 0 (x ).
Domination
A stable matching M is said to dominate a stable matching M 0 ,
written M  M 0 , if every man has at least as good a partner in
M as he has in M 0 .i.e., every man either prefers M to M 0 or is
indifferent between them. M disjointly dominates M 0 (M ≺ M 0 )
if M  M 0 and M ∩ M 0 = ∅.

m1 :
m2 :
m3 :
m4 :
15/39
Men’s Preference
The Lattice Structure

A person x is said to prefer a matching M to a matching M 0 if x

prefers pM (x ) to pM 0 (x ).
Domination
A stable matching M is said to dominate a stable matching M 0 ,
written M  M 0 , if every man has at least as good a partner in
M as he has in M 0 .i.e., every man either prefers M to M 0 or is
indifferent between them. M disjointly dominates M 0 (M ≺ M 0 )
if M  M 0 and M ∩ M 0 = ∅.

m1 :
m2 :
m3 :
m4 :
15/39
Men’s Preference
Meet and Join

M0
m1 :
m2 :
M1 M2 m3 :
m4 :
Men’s Preference
M 00

M 0 = {(m, w ) | w = best (pM1 (m ), pM2 (m ))}

M 00 = {(m, w ) | w = worst (pM1 (m ), pM2 (m ))}

16/39
Meet and Join

M0
m1 :
m2 :
M1 M2 m3 :
m4 :
Men’s Preference
M 00

M 0 = {(m, w ) | w = best (pM1 (m ), pM2 (m ))}

M 00 = {(m, w ) | w = worst (pM1 (m ), pM2 (m ))}

16/39
Meet and Join

M0
m1 :
m2 :
M1 M2 m3 :
m4 :
Men’s Preference
M 00

M 0 = {(m, w ) | w = best (pM1 (m ), pM2 (m ))}

M 00 = {(m, w ) | w = worst (pM1 (m ), pM2 (m ))}

16/39
Meet and Join

M0
m1 :
m2 :
M1 M2 m3 :
m4 :
Men’s Preference
M 00

M 0 = {(m, w ) | w = best (pM1 (m ), pM2 (m ))}

M 00 = {(m, w ) | w = worst (pM1 (m ), pM2 (m ))}

16/39
The Lattice Structure

Set of all stable matchings form a distributive lattice under the

Domination domination.

17/39
Disjoint Stable Matchings

17/39
Why do we need them?

P1

P2

P3

P4

18/39
Why do we need them?

P1

P2

P3

P4

18/39
Why do we need them?

P1

P2

P3

P4

18/39
Why do we need them?

P1

P2

P3

P4

18/39
Problem Statement

For a given marriage instance, find a largest set S of disjoint stable

matchings.

19/39
Existence of Disjoint Stable Matchings

Does there exist a marriage matching instances with disjoint stable

matchings?
m1 : w1 , w2 , w3 w1 : m2 , m3 , m1
m2 : w2 , w3 , w1 w2 : m3 , m1 , m2
m3 : w3 , w1 , w2 w3 : m1 , m2 , m3

m1 w1 m1 w1 m1 w1
m2 w2 m2 w2 m2 w2
m3 w3 m3 w3 m3 w3

20/39
Existence of Disjoint Stable Matchings

Does there exist a marriage matching instances with disjoint stable

matchings?
m1 : w1 , w2 , w3 w1 : m2 , m3 , m1
m2 : w2 , w3 , w1 w2 : m3 , m1 , m2
m3 : w3 , w1 , w2 w3 : m1 , m2 , m3

m1 w1 m1 w1 m1 w1
m2 w2 m2 w2 m2 w2
m3 w3 m3 w3 m3 w3

20/39
Existence of Disjoint Stable Matchings

Does there exist a marriage matching instances with disjoint stable

matchings?
m1 : w1 , w2 , w3 w1 : m2 , m3 , m1
m2 : w2 , w3 , w1 w2 : m3 , m1 , m2
m3 : w3 , w1 , w2 w3 : m1 , m2 , m3

m1 w1 m1 w1 m1 w1
m2 w2 m2 w2 m2 w2
m3 w3 m3 w3 m3 w3

20/39
Necessary Condition

If the man-optimal and the woman-optimal stable matchings share

a common edge (m, w ), then (m, w ) is in every stable matching.

This is because w is both the best stable partner and the worst
stable partner of m.

So, to have disjoint stable matchings, man-optimal and

woman-optimal matchings must be disjoint.

21/39
Algorithm: Disjoint Stable Matchings

Algorithm 3 Disjoint Stable Matchings

1: procedure Find maximum set of Disjoint stable matchings(M)
2: S ←∅ . S: Set of disjoint matchings. Initialize S to be an empty set
3: M 0 ← M.ReverseRoles . Men renamed as women and women as men
4: Mz ←FindStableMatching(M 0 ) . GS Algorithm: Woman-optimal
5: X ←GS-Extended(M) . calling Algorithm 2 modifies M’s list
6: while X ∩ Mz = ∅ do
7: S ← S ∪ {X }
8: for every man m do
9: Delete first woman w on m’s list . First woman is pX (m)
10: Delete last man on w ’s list . Last man is pX (w )
11: Mark m as free
12: end for
13: X ←GS-Extended(M) . Get a new disjoint matching as X
14: end while
15: S ← S ∪ {Mz }
return S
16: end procedure

22/39
Termination and Time Complexity

In every iteration, we delete at least one entry from the preference

list. As the size of preference list is 2n2 , the algorithm terminates.

For the same reason, the running time of the algorithm is O(n2 ).

23/39
Run of Disjoint GS Algorithm

m1 :

m2 :

m3 :

m4 :

m5 :

m6 :

m7 :

Men’s preference list

24/39
Run of Disjoint GS Algorithm

m1 :

m2 :

m3 :

m4 :

m5 :

m6 :

m7 :

Men’s Preference

24/39
Run of Disjoint GS Algorithm

m1 :

m2 :

m3 :

m4 :

m5 :

m6 :

m7 :

Men’s Preference

24/39
Run of Disjoint GS Algorithm

m1 :

m2 :

m3 :

m4 :

m5 :

m6 :

m7 :

Men’s Preference

24/39
Disjoint Stable Matchings

Lemma 1
Each Mi in the set S = {M0 , M1 , · · · , Mn = Mz } is a perfect
matching.

Note: It does not come freely from Extended GS!

It only guarantees one-one.

25/39
Disjoint Stable Matchings

Lemma 1
Each Mi in the set S = {M0 , M1 , · · · , Mn = Mz } is a perfect
matching.

Note: It does not come freely from Extended GS!

It only guarantees one-one.

25/39
Disjoint Stable Matchings

Lemma 2
All the matchings in the set S are stable matchings.

Lemma 3
If M0 , M1 , · · · , Mn = Mz are the matchings discovered by the
algorithm 3 in this order, then M0 ≺ M1 ≺ · · · ≺ Mn = Mz .

26/39
Disjoint Stable Matchings

Lemma 2
All the matchings in the set S are stable matchings.

Lemma 3
If M0 , M1 , · · · , Mn = Mz are the matchings discovered by the
algorithm 3 in this order, then M0 ≺ M1 ≺ · · · ≺ Mn = Mz .

26/39
Disjoint Stable Matchings

Lemma 4
In any arbitrary execution E of the algorithm 3, for any man m,
pMi (m ) is the best stable partner of m when, for every man,
stable partners from M0 , M1 , · · · , Mi −1 are disallowed.

m1 :

m2 :

m3 :

m4 :

m5 :

m6 :
27/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M00 M10 ··· ··· Mk0

28/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M00 M10 ··· ··· Mk0

28/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M0 M10 ··· ··· Mk0

28/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M0 M10 ··· ··· Mk0

28/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M0 M1 ··· ··· Mk0

28/39
Longest Chain of Disjoint Stable matchings

Lemma 5

The algorithm 3 gives the longest chain of disjoint stable

matchings.

Proof:

M0 M1 ··· Mn

M0 M1 ··· Mn ··· Mk0

28/39
Disjoint Stable Matching

Theorem 6 (Teo, C.-P. and Sethuraman, J. (1998))

Let I be an SM instance and let T be a set of stable matchings

in I . Let αj,T (respectively β j,T ) denote the set of pairs obtained
by assigning each man mi (woman wi ) to pj,T (mi ) (pj,T (wi )),
the jth element in the sorted multiset
PT (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,T } (respectively PT (wi )).
Then, each of αj,T and β j,T is a stable matching.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted multiset

{pM1 (m), pM2 (m), · · · , pMk (m)} }

M10 M20 ··· Mq0

29/39
Disjoint Stable Matching

Theorem 6 (Teo, C.-P. and Sethuraman, J. (1998))

Let I be an SM instance and let T be a set of stable matchings

in I . Let αj,T (respectively β j,T ) denote the set of pairs obtained
by assigning each man mi (woman wi ) to pj,T (mi ) (pj,T (wi )),
the jth element in the sorted multiset
PT (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,T } (respectively PT (wi )).
Then, each of αj,T and β j,T is a stable matching.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted multiset

{pM1 (m), pM2 (m), · · · , pMk (m)} }

M10 M20 ··· Mq0

29/39
Disjoint Stable Matching

Theorem 6 (Teo, C.-P. and Sethuraman, J. (1998))

Let I be an SM instance and let T be a set of stable matchings

in I . Let αj,T (respectively β j,T ) denote the set of pairs obtained
by assigning each man mi (woman wi ) to pj,T (mi ) (pj,T (wi )),
the jth element in the sorted multiset
PT (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,T } (respectively PT (wi )).
Then, each of αj,T and β j,T is a stable matching.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted multiset

{pM1 (m), pM2 (m), · · · , pMk (m)} }

M10 M20 ··· Mq0

29/39
Disjoint Chain

Corollary 7

Let S = {M1 , M2 , · · · , Mk } be a set of disjoint stable

matchings. Let αj,S denote the stable matching obtained by
matching each man mi to pi,S (mi ), the jth woman in the sorted
set PS (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,S }. Then, the stable
matchings from the set C = {α1,S , α2,S , · · · , αk,S } forms a
k-length chain α1,S ≺ α2,S ≺ · · · ≺ αk,S of disjoint stable
matchings.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted set

{pM1 (m), pM2 (m), · · · , pMk (m)} }

30/39
M10 M20 ··· Mk0
Disjoint Chain

Corollary 7

Let S = {M1 , M2 , · · · , Mk } be a set of disjoint stable

matchings. Let αj,S denote the stable matching obtained by
matching each man mi to pi,S (mi ), the jth woman in the sorted
set PS (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,S }. Then, the stable
matchings from the set C = {α1,S , α2,S , · · · , αk,S } forms a
k-length chain α1,S ≺ α2,S ≺ · · · ≺ αk,S of disjoint stable
matchings.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted set

{pM1 (m), pM2 (m), · · · , pMk (m)} }

30/39
M10 M20 ··· Mk0
Disjoint Chain

Corollary 7

Let S = {M1 , M2 , · · · , Mk } be a set of disjoint stable

matchings. Let αj,S denote the stable matching obtained by
matching each man mi to pi,S (mi ), the jth woman in the sorted
set PS (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,S }. Then, the stable
matchings from the set C = {α1,S , α2,S , · · · , αk,S } forms a
k-length chain α1,S ≺ α2,S ≺ · · · ≺ αk,S of disjoint stable
matchings.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted set

{pM1 (m), pM2 (m), · · · , pMk (m)} }

30/39
M10 M20 ··· Mk0
Disjoint Chain

Corollary 7

Let S = {M1 , M2 , · · · , Mk } be a set of disjoint stable

matchings. Let αj,S denote the stable matching obtained by
matching each man mi to pi,S (mi ), the jth woman in the sorted
set PS (mi ) = {wi |(mi , wi ) ∈ M, M ∈ αj,S }. Then, the stable
matchings from the set C = {α1,S , α2,S , · · · , αk,S } forms a
k-length chain α1,S ≺ α2,S ≺ · · · ≺ αk,S of disjoint stable
matchings.

Given stable matchings M1 , M2 , · · · , Mk ,

Mi0 = {(m, w ) | w is the i-th women in the sorted set

{pM1 (m), pM2 (m), · · · , pMk (m)} }

30/39
M10 M20 ··· Mk0
Maximum Size Set of Disjoint Stable Matchings

Theorem 8

For a given stable marriage instance, algorithm 3 gives the

maximum size set of disjoint stable matchings.

31/39
Rotations

31/39
Rotations

For every stable matching M, we define the following:

sM ( m )
For any man m, let sM (m ) denote the first women w on m’s list
such that w strictly prefers m to pM (w )

nextM (m )
For any man m, let nextM (m ) denote pM (sM (m ))

Note: sM (m ) might not exist. Example: Wz .

Both sM (m ) and nextM (m ) can be easily be found using Reduced
Lists.

32/39
Roations

Definition of Rotations
An ordered list of matched pairs
ρ = (m0 , w0 ), (m1 , w1 ), · · · , (mr −1 , wr −1 ) in a stable matching
M is called as a rotation exposed in M if for each i
(0 ≤ i ≤ r − 1), mi +1 is nextM (mi ) where i + 1 is taken modulo
r.
Elimination of a Rotation
If M is a stable matching and
ρ = (m0 , w0 ), (m1 , w1 ), · · · , (mr −1 , wr −1 ) is a rotation exposed
in M, then M/æ is defined to be matching in which each man
who is not in ρ stays married to his partner in M, and each man
mi in M is matched to wi +1 = sM (mi )

That is, M/ρ differs from M by one place cyclic shift of each men
33/39
in ρ
Roations

Definition of Rotations
An ordered list of matched pairs
ρ = (m0 , w0 ), (m1 , w1 ), · · · , (mr −1 , wr −1 ) in a stable matching
M is called as a rotation exposed in M if for each i
(0 ≤ i ≤ r − 1), mi +1 is nextM (mi ) where i + 1 is taken modulo
r.
Elimination of a Rotation
If M is a stable matching and
ρ = (m0 , w0 ), (m1 , w1 ), · · · , (mr −1 , wr −1 ) is a rotation exposed
in M, then M/æ is defined to be matching in which each man
who is not in ρ stays married to his partner in M, and each man
mi in M is matched to wi +1 = sM (mi )

That is, M/ρ differs from M by one place cyclic shift of each men
33/39
in ρ
Representation

m2 m7 m8

m1 m3 m4

m5 m6

Figure 1: Graph H (M )

(mi , mj ) ∈ E (H (m)) if mj = nextM (mi ).

34/39
Properties of Rotations

ˆ M/ρ is a stable matching such that M  M/ρ

ˆ Every stable matching except the women optimal matching
has at least one rotation exposed in it.
ˆ Every path from M0 to Mz in M corresponds to some
permutation of set of all rotations.

35/39
Properties of Rotations

ˆ M/ρ is a stable matching such that M  M/ρ

ˆ Every stable matching except the women optimal matching
has at least one rotation exposed in it.
ˆ Every path from M0 to Mz in M corresponds to some
permutation of set of all rotations.

35/39
Properties of Rotations

ˆ M/ρ is a stable matching such that M  M/ρ

ˆ Every stable matching except the women optimal matching
has at least one rotation exposed in it.
ˆ Every path from M0 to Mz in M corresponds to some
permutation of set of all rotations.

35/39
Properties of Rotations

ˆ M/ρ is a stable matching such that M  M/ρ

ˆ Every stable matching except the women optimal matching
has at least one rotation exposed in it.
ˆ Every path from M0 to Mz in M corresponds to some
permutation of set of all rotations.

35/39
poset of rotations

The set of all rotations forms a partial order under the following
relation.
ρ1 ≺ ρ2 iff in every path from M0 to Mz in M ρ1 gets eliminated
before ρ2 .
ρ1 ρ2

ρ3 ρ4

ρ5

ρ6 ρ7

Figure 2: The Rotational Poset Π((M ))

36/39
poset of rotations

Theorem 9
There is a one-one correspondence between the closed subsets of
Π((M )) and stable matchings in (M )

Theorem 10
S is a closed set of rotations of Π((M )) corresponding to a
stable matching M iff S is the (unique) set of rotations in every
M0 -M chain in (M )

37/39
poset of rotations

Theorem 9
There is a one-one correspondence between the closed subsets of
Π((M )) and stable matchings in (M )

Theorem 10
S is a closed set of rotations of Π((M )) corresponding to a
stable matching M iff S is the (unique) set of rotations in every
M0 -M chain in (M )

37/39
Algorithm based on Π((M ))

Given a stable matching Mi , we can find the next best disjoint

stable matching Mi +1 by doing the following steps.
1. For every (m, w ) ∈ Mi , find R = {ρ|(m, w ) ∈ ρ}
2. Find R̂ = closure of R
3. Return Mi +1 = Mi /R̂.

We use the above steps recursively (until we find a matching that

intersects with Mz ) to get the maximum chain of disjoint stable
matchings.

38/39
Algorithm based on Π((M ))

Given a stable matching Mi , we can find the next best disjoint

stable matching Mi +1 by doing the following steps.
1. For every (m, w ) ∈ Mi , find R = {ρ|(m, w ) ∈ ρ}
2. Find R̂ = closure of R
3. Return Mi +1 = Mi /R̂.

We use the above steps recursively (until we find a matching that

intersects with Mz ) to get the maximum chain of disjoint stable
matchings.

38/39
Algorithm based on Π((M ))

Given a stable matching Mi , we can find the next best disjoint

stable matching Mi +1 by doing the following steps.
1. For every (m, w ) ∈ Mi , find R = {ρ|(m, w ) ∈ ρ}
2. Find R̂ = closure of R
3. Return Mi +1 = Mi /R̂.

We use the above steps recursively (until we find a matching that

intersects with Mz ) to get the maximum chain of disjoint stable
matchings.

38/39
Algorithm based on Π((M ))

Given a stable matching Mi , we can find the next best disjoint

stable matching Mi +1 by doing the following steps.
1. For every (m, w ) ∈ Mi , find R = {ρ|(m, w ) ∈ ρ}
2. Find R̂ = closure of R
3. Return Mi +1 = Mi /R̂.

We use the above steps recursively (until we find a matching that

intersects with Mz ) to get the maximum chain of disjoint stable
matchings.

38/39
Algorithm based on Π((M ))

Given a stable matching Mi , we can find the next best disjoint

stable matching Mi +1 by doing the following steps.
1. For every (m, w ) ∈ Mi , find R = {ρ|(m, w ) ∈ ρ}
2. Find R̂ = closure of R
3. Return Mi +1 = Mi /R̂.

We use the above steps recursively (until we find a matching that

intersects with Mz ) to get the maximum chain of disjoint stable
matchings.

38/39
Future Work

ˆ Disjoint Stable Matchings in the Roommate problem.

ˆ When disjoint stable matchings do not exist, minimize
pairwise intersection.

39/39
Thank You!

39/39