Homework 5 Solutions

Problem 1. If G is a graph with a maximum matching of size 2k, what is the smallest
possible size of a maximal matching in G?
Solution: The answer is k. To construct such a graph, take a graph with k components, each
of which is a three edge path. The unique maximum matching uses two edges from each
component, but there is a maximal matching using just one from each component. To prove
that this is best possible, we need to prove that every graph G with a matching M ∗ of size
2k has the property that every maximal matching M has size at least k. To see this, note
that in order to be maximal, the matching M must cover at least one endpoint from each
edge of M ∗ (otherwise we could just add this edge to M , thus contradicting maximality). It
follows that M must cover a set of 2k vertices, so it must have size at least k.

Problem 2. Prove or disprove: Every tree has at most one perfect matching (a perfect
matching is a matching covering every vertex).
Solution: This is true. Let M, M 0 be perfect matchings in the tree T = (V, E) and consider
the graph on V with edge set M ∪ M 0 . Since M and M 0 both cover all the vertices, every
component of this new graph is either a single edge (common to both M and M 0 ) or a cycle.
Since T is a tree, there can be no cycle, so we conclude that M = M 0 .

Problem 3. Let G be a simple 2n vertex graph and assume that every vertex has degree
≥ n + 1. Show that G has a perfect matching.
Solution: It follows from Theorem 1.15 that G has a Hamiltonian cycle. Taking every second
edge of this cycle yields a perfect matching.

Problem 4. Let G be a bipartite graph with bipartition (A, B), let S ⊆ A and let T ⊆ B.
Assume there exist matchings M and M 0 so that M covers S and M 0 covers T , and then
prove that there exists a matching M ∗ which covers S ∪ T .
Solution: Consider the graph H = (V (G), M ∪ M 0 ). Each component of H is either an
isolated vertex, an edge which is contained in M ∩ M 0 , a cycle with edges alternately in
M and M 0 , or a path where edges are alternately from M and M 0 . Let H1 , . . . , H` be the
components of H and choose a matching Mi from each Hi as follows. If Hi is an isolated
vertex, then it is not in S ∪ T and we let Mi = ∅. If Hi is either an edge in M ∩ M 0 or a cycle,
 2
or a path of odd length, then Hi has a matching Mi which covers V (Hi ), so in particular
it covers (S ∪ T ) ∩ V (Hi ). Finally, we consider the case that Hi is a path of even length.
Here the edges must alternate between M and M 0 , so one end of the path is incident with
an edge in M and the other in M 0 . However, there must be an odd number of vertices in
this path, so either both ends are in A or both ends are in B. In the former case we let
Mi = M ∩ E(Hi ) and in the latter we set Mi = M 0 ∩ E(Hi ). In either case we again have
that all vertices in S ∪ T which are contained in Hi are covered by Mi . So, now ∪`i=1 Mi is a
matching in G covering S ∪ T as desired.

Problem 5. Let X be a finite set and let A1 , A2 , . . . , Am be subsets of X. Prove that one
of the following is true

1. There exists a set I ⊆ {1, 2, . . . , m} so that | ∪i∈I Ai | < |I|.

2. There exist distinct elements a1 , a2 , . . . , am ∈ X so that ai ∈ Ai for every 1 ≤ i ≤ m.

Hint: turn this into a graph theory problem.

Solution: Define a simple bipartite graph G with vertex set {1, 2, . . . , m} ∪ X and bipartition
({1, 2, . . . , m}, X) by the rule that i ∈ {1, 2, . . . , m} and x ∈ X are adjacent if and only if
x ∈ Ai . If there exists a matching M in G which covers {1, 2, . . . , m}, then for every
1 ≤ i ≤ m let ai ∈ X be the element which is paired with i by M . Now, by construction
a1 , a2 , . . . , am are distinct and ai ∈ Ai for every 1 ≤ i ≤ m. If there is no such matching, then
by Hall’s Marriage Theorem, there must exist a set I ⊆ {1, 2, . . . , m} so that |N (I)| < |I|.
However, then we have |I| < |N (I)| = | ∪i∈I Ai | so the first outcome holds.

Problem 6. Prove that if man m is paired with woman w in some stable marriage, then
w does not reject m in the Gale-Shapley Algorithm. Hint: consider the first occurrence of
such a rejection.
Solution: Let M be a stable marriage, and suppose for a contradiction that during the Gale-
Shapley algorithm, some man m is rejected by a woman w for which m and w are paired
in M . Consider the first step of the algorithm during which such a rejection occurs. Since
m is rejected by w on this step, w must receive a proposal from some man m0 whom she
prefers to m on this step. Since, by assumption M is a stable marriage, it follows that m0
must be paired with a woman w0 in M with the property that m0 prefers w0 to w. However,
 3
since m0 is proposing to w at this step of the Gale-Shapley algorithm, he must already have
been rejected by w0 , but this contradicts our assumption that this was the first step of the
algorithm on which a rejection of the given type occurs.

Problem 7. Generalizing Tic-Tac-Toe A positional game consists of a set X of positions and

a family W1 , W2 , . . . , Wm ⊆ X of winning sets (Tic-Tac-Toe has 9 positions corresponding
to the 9 boxes, and 8 winning sets corresponding to the three rows, three columns, and
two diagonals). Two players alternately choose positions; a player wins when they collect a
winning set.
Suppose that each winning set has size at least a and each position appears in at most b
winning sets (in Tic-Tac-Toe a = 3 and b = 4). Prove that Player 2 can force a draw if a ≥ 2b.
Hint: Form a bipartite graph G with bipartition (X, Y ) where Y = {W1 , W2 , . . . , Wm } ∪
{W10 , W20 , . . . , Wm0 } with edges xWj and xWj0 whenever x ∈ Wj . How can Player 2 use a
matching in G?
Solution: Let Y 0 ⊆ Y and define S to be the set of all edges incident with a vertex in Y 0 .
Since every vertex in Y has degree at least a we must have |S| ≥ a|Y 0 |. On the other hand,
every vertex in X has degree at most 2b, so we must have 2b|N (Y 0 )| ≤ |S|. Combining these
gives us a|Y 0 | ≤ 2b|N (Y 0 )| and together with the assumption a ≥ 2b we find |N (Y 0 )| ≥ |Y 0 |.
So, it now follows from Hall’s Theorem that there is a matching M which covers Y .
For every 1 ≤ i ≤ m let Qi be the set consisting of the two vertices in X which are
matched to Wi and Wi0 . Now the sets Q1 , . . . , Qm are disjoint two element subsets of X
and every Wi contains Qi . Here is a strategy which will guarantee the second player a draw
(or better). For each move made by the first player, if the first player chooses a position
x ∈ Qi for some 1 ≤ i ≤ m then the second player responds by choosing the other position
in Qi if it is available. Otherwise, the second player just plays arbitrarily. It follows from a
straightforward induction that after every turn of the second player, there is no set Qi for
which player 1 has chosen one element, and player 2 none. It follows from this that player
1 can never choose both members of a set Qi , and from this that player 1 cannot choose all
members of any Wi . Thus, player 1 cannot win when player 2 adopts this strategy.