Math776: Graph Theory (I)

Fall, 2017
Homework 3 solution
1. [page 31, #39 ] Prove Gallai’s theorem that the edge set of any graph G
can be written as a disjoint union E(G) = C ∪ D with C ∈ C(G) and
D ∈ C ∗ (G).
Proof: Let G be an arbitrary graph. Suppose for |G| < n there is a
partition of G such that G[V1 ] and G[V2 ] both have even degree and D =
{ab ∈ E(G)|a ∈ V1 , b ∈ V2 }. Consider G with |G| = n. If deg(v) is even
for all v ∈ V (G), then we are done as G = D and C = ∅. So, let v ∈ V (G)
such that deg(v) is odd. Construct G0 = G\{v} by adding the edge ab
if ab ∈
/ E(G) and deleting the edge ab if ab ∈ E(G) for a, b ∈ N (v). We
note that constructing the edges of G0 in this way preserves the parity
of the vertices in N (v). By the induction hypothesis, there is a partition
of G0 such that G0 [V1 ] and G0 [V2 ] both have even degree. As deg(v) is
odd, there are an odd number of neighbors of v in one of V1 or V2 and
an even number in the other. WLOG, suppose there are an odd number
of neighbors of v in V1 . Then we add v to V2 and reconstruct the edges
between v and its neighbors and delete the edges that were added. Then
G[V1 ] and G[V2 ] both have even degree. Thus, E(G) = C ∪ D.
2. [page 54, #11 ] Let G be a bipartite graph with bipartition {A, B}. As-
sume that δ(G) ≥ 1, and that d(a) ≥ d(b) for every edge ab with a ∈ A.
Show that G contains a matching of A.
Solution: Assume G has a minimal set S such that S does not satisfy
the marriage condition. In other words |N (S)| < |S|. Remove one vertex
of S, call it S 0 . Since S was minimal we are now guaranteed a matching
in S 0 . Also the |N (S 0 )| = |N (S)| orPelse we would have
P had a matching in
S. The edges from S 0 to N (S 0 ) = ai ∈S 0 d(ai ) = bj ∈N (S 0 ) d(bj ). So for
each ai ∃bj such that d(ai ) = d(bj ). When we add our one vertex back into
S 0 , it will be connected to one of the vertices in N (S 0 ). This will disrupt
the equality above and ∃ a bj such that d(ai ) < d(bj ).
So there is no minimal set that violates the marriage condition. So all
subsets have the marriage condition. So we have a matching on A. 

3. [page 55, #5 ] Derive the marriage theorem from König’s theorem.

Solution: The König’s theorem says that in a bipartite graph G, max|M | =
min|K|. where M is a matching, and K is a vertex cover of edges. We
use this theorem to prove the Hall’ theorem which says that G contains a
matching of A if and only if |N (S)| ≥ |S| for all S ⊆ A. We use contradic-
tion method, given a graph G, it satisfies the Hall condition, but has no
matching of A, then according to König’s theorem, the size if cover U is

1
 0 0 0
less than size of A, i.e. |U | < |A|. We say U = A + B with A ⊆ A and
0 0 0 0 0 0
B ⊆ B. Then |A | + |B | = |U | < |A|. Hence |B | < |A| − |A | = |A \ A |.
0 0
Since U is a cover of edges, there is no edge between A \ A and \AB ,
0 0 0
so, |N (A \ A )| ≤ |B | < |A \ A |, which contradicts to the hall condition
0
when set S = A \ A .
4. [page 55, #8 Find an infinite counterexample to the statement of the mar-
riage theorem.
Proof: Let G be a bipartite graph with partition classes A and B. Enu-
merate the vertices in each of these partition classes. Then, for a fixed
index i, let ai be adjacent to bi−1 where i ≥ 2, and let a1 be adjacent to
every vertex in B (see picture below). Notice, when the size of A and B
are infinite, any subset of the vertices in A has at least as many neighbors
as the size of the subset. Therefore, |N (S)| ≥ |S| for all S ⊆ A. Notice,
this is not the case when the sizes of A and B are finite. However, for
i ≥ 2, ai has only one edge incident, and thus, this edge must be used when
considering a matching for A. Therefore, choosing each of these edges, one
must then find an independent edge that is incident to a1 , which is not
possible. Hence, no matching of A exists.

5. [page 55, #9 ] Let A be a finite set with subsets A1 , . . . , An , and let

d1 , . . . , dn ∈ N. Show that there are disjoint subsets Dk ⊂ Ak , with
|Dk | = dk for all k ≤ n if and only if
X
|∪i∈I Ai | ≥ di
i∈I

for all I ⊂ {1, . . . , n}.

Solution: We construct a bipartite graph. In the left hand partition L,
we place di copies of a vertex labeled Di for i = 1, . . . , n, and into the
right hand partition R we place each of the distinct elements of ∪ni=1 Ai .
For i = 1, . . . , n, make each copy of Di adjacent to each a ∈ Ai .

2
 If there exist disjoint subsets Dk ⊆ Ak , with |Dk | = dk for all k ≤ n,
then this is equivalent to there being a matching in the bipartite graph we
constructed. Hence, by the Marriage Theorem, |∪i∈I Ai | = |N (Di i∈I )| ≥
P P
i∈I |Di | = i∈I di for all I ⊆ {1, . . . , n}.
P
On the other hand, if |∪i∈I Ai | ≥ i∈I di for all I ⊆ {1, . . . , n}, then we
P P
have |N (Di i∈I )| = |∪i∈I Ai | ≥ i∈I di = i∈I |Di | for all I ⊆ {1, . . . , n}.
Furthermore, if we take P an S ⊂ L without the full number of copies of
some Di , then |S| < i∈I |Di | ≤ |N (Di i∈I )| = |N (S)|. Therefore, by
the Marriage Theorem, there exists a matching, and we have constructed
disjoint subsets Dk ⊆ Ak , with |Dk | = dk for all k ≤ n.

2. [page 55, #14 ] Show that all stable matchings of a given graph cover the
same vertices. (In particular, they have the same size.)
Solution: Let M, M 0 be two stable matchings of G. For a contradiction,
suppose ∃ v0 ∈ M 0 \ M . Then v0 has a neighbor v1 with v0 v1 ∈ M 0 .
Note that v1 must be matched in M , otherwise we may add v0 v1 to M
to get a larger stable matching, a contradiction. Since v1 is matched in
M , then v1 has a neighbor v2 with v1 v2 ∈ M . We have that v0 v1 v2 is a
path with edges alternately in M 0 and M . Continue in this manner to
get a full path P = v0 v1 · · · vn (for some n ∈ N) and consider vn−1 . We
have the preferences vn−2 <vn−1 vn in M , but vn <vn−1 vn−2 in M 0 , a
contradiction.
Thus, such a v0 cannot exist, so M and M 0 must cover the same vertices.