Dominik Farhan

Tutte's theorem for matching

Definitions

We will assume G=(V,E)G = (V, E) to be an undirected graph.

Definition (Matching). MM \subseteq is a matching if for different e,fM\forall e,f \in M ef=e\cap f = \emptyset.

Definition (maximum matching). Matching of maxM\max |M|.

Definition (perfect matching). matchig is perfect if for each vVeM:evv \in V \exists e \in M: e\cap v \neq \emptyset.

Definition (M-covered vertex). Vertex is MM-covered if there is an edge in MM containing it.

Definition (M-exposed vertex). Vertex is M-exposed if it is not in any edge of the matching.

Definition (deficiency of G and M). def(G)def(G) is the minimum number of vertices exposed in any matching of GG. Similarly def(M)def(M) is the number of vertices that are MexposedM-exposed for this particular matching.
Definition (Inessential vertex). Vertex is inessential if there is a maximum matching not containing it.

Definition (alternating path). Path is alternating (in a particular matching) if edges on it are alternately in or not and the first and the last vertex are both MM-exposed.

Theorems

Theorem (Augmenting path and matching). Matching is maximum iff there is no augmenting path.

Proof.     \implies

By complementarity. Suppose there is an augmenting path. Then we can swap edges on that path, thus the matching cannot be maximum.

Proof (\Longleftarrow).

Again by complementariy. Assume that MM is not maximum and there is NN with greater size. Let F=MΔNF = M \Delta N. Observe that G[F]G[F] contains only cycles and paths. Because edges have to alternate between matchings, all cycles are of even length. However we know that N>M|N| > |M| so there then must be a path of odd length which contains more edges from NN than from MM. This is an alternating path in MM.

Theorem (Little observation). For any MM and all AVA \subseteq V it must hold that def(M)odd(GA)Adef(M) \geq odd(G \setminus A) - |A|
Proof (Little observation).

Let's have the matching MM and some AA. When we remove AA from the graph we are left with some even and odd components. In the best possible scenario MM pairs all vertices in even components. We are left with the odd. In each odd components, in the best possible scenario, matching elements all but one vertex. This gives us odd(GA)odd(G \setminus A) vertices we need to pair.

Again, in the best possible case all vertices in AA can be paired with the vertices in the odd components. We are left with odd(GA)Aodd(G \setminus A) - |A| vertices. Which completes the proof.

Theorem (Week Tutte-Berge). For any AVA\subseteq V maxMM=minM12(Vdef(M))minAV12(Vodd(GA)+A)\max_{M} |M| = \min_{M} \frac{1}{2}(|V| - def(M)) \leq \min_{A \subseteq V} \frac{1}{2}(|V| - odd(G\setminus A) + |A|)
Theorem (Tutte-Berge). For any AVA\subseteq V maxMM=minAV12(Vodd(GA)+A)\max_{M} |M| = \min_{A\subseteq V} \frac{1}{2}(|V| - odd(G\setminus A) + |A|)

There are two proofs of this theorem. One non-constructive which I will not attempt to carry out. However can be found here. The other is constructive and assumes reader to be fammiliar with Edmond's max matching algorithm. This proof is scatched without many import details bellow.

Proof (Tutte-Berge (bipartite case)).

Assume GG to be bipartite. Then we can apply Edmonds algorithm. No odd cycle can appear, so we can never get a blossom which we would need to contract. We are going to end up with matching that is maximal and some number of vertices that are deficient. So we are essentialy constructing the solution so that the equality always holds.

Proof (Tutte-Berge (general case)). Here, an edge between vertices with even distance from the root creates an odd cycle (blossom, denote it as CC). Notice that there is an MM-augmenting path in GG iff there is an MM-augmenting path in G×CG \times C. So the idea is to contract any blossom and reduce it to previous case.
Theorem (Tutte). GG has a perfect matching iff for all AA subsets of vertices odd(GA)A.odd(G \setminus A) \leq |A|.
This theorem is direct corollary of Tutte-Berge. However, in reality this special case was proven before Tutte-Berge. Again only rough description of the proof.
Proof (    \implies). Let's turn things around and assume that there is AA such that odd(GA)>A.odd(G \setminus A) > |A|. Then if there was a perfect matching it means that there is a matched edged between each odd component and some vertex in AA which is clearly not possible.
Proof. \Longleftarrow'

This can be proved by rather complicated induction. Mayber I'll add this later.

Theorem (Gallai-Edmonds decomposition).