MIT: 18.

211: COMBINATORIAL ANALYSIS

FELIX GOTTI

Lecture 30: Matching and Hall’s Theorem

Hall’s Theorem. Let G be a simple graph, and let S be a subset of E(G). If no two
edges in S form a path, then we say that S is a matching of G. A matching S of G is
called a perfect matching if every vertex of G is covered by an edge of S.
Definition 1. Let G be a bipartite graph on the parts X and Y , and let S be a
matching of G. If every vertex in X is covered by an edge of S, then we say that S is
a perfect matching of X into Y .
For a graph G and a subset T of V (G), we let NG (T ) denote the set of vertices of G
that are adjacent to some vertex in T , that is,
NG (T ) := {v ∈ V (G) | vw ∈ E(G) for some w ∈ T }.
Observe that if G is bipartite on the parts A and B, then NG (T ) ⊆ B for any T ⊆ A.
We proceed to prove the main result of this lecture, which is due to Philip Hall and is
often called Hall’s Marriage Theorem.
Theorem 2. For a bipartite graph G on the parts X and Y , the following conditions
are equivalent.
(a) There is a perfect matching of X into Y .
(b) For each T ⊆ X, the inequality |T | ≤ |NG (T )| holds.
Proof. (a) ⇒ (b): Let S be a perfect matching of X into Y . As S is a perfect matching,
for every x ∈ X there exists a unique yx ∈ Y such that xyx ∈ S. Define the map
f : X → Y by f (x) = yx . Since S is a matching, the function f is injective. Therefore
for any T ⊆ X, we see that |T | = |f (T )| ≤ |NG (T )| because f (T ) ⊆ NG (T ).
(b) ⇒ (a): Conversely, suppose that |T | ≤ |NG (T )| for each T ⊆ X. We will prove
that there exists a perfect matching of X into Y by induction on n := |X|. If n = 1,
then the only vertex x in X must be adjacent to some vertex y in Y by condition (b)
and, therefore, {xy} is a perfect matching of X into Y . Now assume that every bipartite
graph on the parts X 0 and Y 0 with |X 0 | < |X| and satisfying condition (b) has a perfect
matching of X 0 into Y 0 . We split the rest of the proof into two cases.
Case 1: For every nonempty proper subset T of X (that is, T ( X), the strict inequality
|T | < |NG (T )| holds. Take x ∈ X and y ∈ NG ({x}). Let G0 be the bipartite graph we
1
2 F. GOTTI

obtain by removing x and y (and the edges incident to them) from G. Now for every
subset A of X \ {x}, we see that
|NG0 (A)| ≥ |NG (A)| − 1 ≥ |A|,
where the last inequality holds because A is a strict subset of X. By induction hy-
pothesis, there exists a perfect matching S 0 in G0 of X \ {x} into Y \ {y}. It is clear
now that S 0 ∪ {xy} is a perfect matching in G of X into Y .
Case 2: There exists a nonempty proper subset A of X such that |A| = |NG (A)|. Let
G1 be the subgraph of G induced by the set of vertices A ∪ NG (A), and let G2 be the
subgraph of G we obtain by removing A ∪ NG (A) (and their incident edges) from G.
It is clear that G1 = (A, NG (A)) and G2 = (X \ A, Y \ NG (A)) are bipartite graphs.
Let us show that both G1 and G2 satisfy condition (b).
To show that G1 satisfies (b), take T ⊆ A. It follows by the way G1 was constructed
that NG1 (T ) = NG (T ). As a result, |NG1 (T )| = |NG (T )| ≥ |T |. Then G1 satisfies
condition (b). In order to argue that G2 also satisfies condition (b), take T 0 ⊆ X \ A
and observe that NG (T 0 ∪ A) = NG (A) ∪ NG2 (T 0 ), where the union on the right-hand
side is disjoint. Since |NG (T 0 ∪ A)| ≥ |T 0 ∪ A| and |NG (A)| = |A|,
|NG2 (T 0 )| = |NG (T 0 ∪ A)| − |NG (A)| ≥ |T 0 ∪ A| − |A| = (|T 0 | + |A|) − |A| = |T 0 |.
Therefore G2 also satisfies condition (b). Since |A| < |X| and |X \ A| < |X|, our
induction hypothesis guarantees the existence of a perfect matching S1 in G1 of A into
NG (A) and a perfect matching S2 in G2 of X \ A into Y \ NG (A). Then it follows from
the construction of G1 and G2 that S1 ∪ S2 is a perfect matching in G of X into Y ,
which concludes the proof. 
We conclude this lecture characterizing whether a matching on a simple graph has
the maximum number of edges possible. First, we need the following definitions.
Definition 3. Let G be a graph, and let M be a matching of G. A path P = v1 v2 . . . v`
is called M -alternating provided that vi−1 vi ∈ M if and only if vi vi+1 ∈
/ M . An M -
alternating path is called M -augmenting if it starts and ends at vertices that are not
covered by any edge of M .
We can now characterize the maximum-length matching in terms of augmenting
paths.
Theorem 4. Let G be a simple graph with a matching M . Then M is a maximum-
length matching if and only if G has no M -augmenting paths.
Proof. For the direct implication suppose that G has an M -augmenting path, namely, P .
Since P is an M -augmenting path, it has odd length. Write P := v1 v2 . . . v2` . Since P
is M -alternating and none of the vertices v1 and v` is covered by M , we see that
P ∩ M = {v2k v2k+1 | k ∈ [` − 1]}. Then after replacing the subset P ∩ M of M by
 COMBINATORIAL ANALYSIS 3
the subset {v2k−1 v2k | k ∈ [`]}, we would obtain a new matching of G with more edges
than M .
For the reverse implication, suppose that G has no M -augmenting path. Let M 0 be
a maximum-length matching. If M 0 = M , then we are done. Suppose, therefore, that
M 6= M 0 . Let S be the symmetric difference of M and M 0 , that is, the set of edges
in M ∪ M 0 that are not in M ∩ M 0 . Now observe that every connected component of
the graph (V (G), S) is either a path or an (even-length) cycle whose edges alternate
between M 0 and M . Now the maximality of M 0 , along with the non-existence of M -
augmenting paths, guarantees that each connected component of (V (G), S) that is a
path must be a path of even length. Hence |M | = |M 0 |, which implies that M is a
maximum-length cycle. 

Practice Exercises
Exercise 1. [1, Exercise 11.4] Let G be a bipartite graph on the parts X and Y , and
suppose that the inequality deg x ≥ deg y holds for all x ∈ X and y ∈ Y . Prove that X
has a perfect matching into Y .

Exercise 2. [1, Exercise 11.12] Let G be a regular bipartite graph (that is, a graph with
all the vertices having the same degree). Prove that G has a perfect matching.

References
[1] M. Bóna: A Walk Through Combinatorics: An Introduction to Enumeration and Graph Theory
(Fourth Edition), World Scientific, New Jersey, 2017.

Department of Mathematics, MIT, Cambridge, MA 02139

Email address: [email protected]