ProofComplete

Proof of KΓΆnig's Theorem

KΓΆnig's theorem establishes the fundamental duality between matchings and vertex covers in bipartite graphs. It is a cornerstone of combinatorial optimization and linear programming duality.


Statement

Theorem6.3KΓΆnig's Theorem

In a bipartite graph G=(XβˆͺY,E)G = (X \cup Y, E), the maximum matching size equals the minimum vertex cover size: Ξ½(G)=Ο„(G)\nu(G) = \tau(G).


Proof

Proof

Ξ½(G)≀τ(G)\nu(G) \leq \tau(G). For any matching MM and vertex cover SS: each edge of MM has at least one endpoint in SS, and these endpoints are distinct (since MM is a matching). So ∣Mβˆ£β‰€βˆ£S∣|M| \leq |S|, hence ν≀τ\nu \leq \tau.

Ξ½(G)β‰₯Ο„(G)\nu(G) \geq \tau(G). We construct a vertex cover of size Ξ½(G)\nu(G) from a maximum matching MM.

Let MM be a maximum matching. Define U={x∈X:x is not saturated by M}U = \{x \in X : x \text{ is not saturated by } M\} (unmatched vertices on the left). Let ZZ be the set of all vertices reachable from UU by alternating paths (paths that alternate between non-matching and matching edges).

Define: L=Z∩XL = Z \cap X (reachable left vertices), R=Z∩YR = Z \cap Y (reachable right vertices).

Claim: S=(Xβˆ–L)βˆͺRS = (X \setminus L) \cup R is a minimum vertex cover.

SS is a vertex cover. Consider any edge xyxy with x∈X,y∈Yx \in X, y \in Y:

  • If xβˆ‰Lx \notin L (i.e., x∈Xβˆ–Lx \in X \setminus L): x∈Sx \in S, so the edge is covered.
  • If x∈Lx \in L: xx is reachable from UU by an alternating path. If xyβˆ‰Mxy \notin M: the path to xx extended by xyxy reaches yy, so y∈RβŠ†Sy \in R \subseteq S. If xy∈Mxy \in M: xx is matched to yy. Since x∈Lx \in L is reachable, and xβˆ‰Ux \notin U (matched vertices are not in UU), xx was reached via some yβ€²βˆˆRy' \in R with xyβ€²βˆ‰Mxy' \notin M, then yβ€²xβ€²βˆˆMy'x' \in M... Actually, the alternating path to xx ends with a matching edge into xx, meaning the previous vertex yβ€²y' is in RR, and yβ€²x∈My'x \in M. But xx is matched to yy by MM, so yβ€²=yy' = y, giving y∈RβŠ†Sy \in R \subseteq S.

∣S∣=∣M∣|S| = |M|. Each matching edge contributes exactly one vertex to SS: if xy∈Mxy \in M with x∈L,y∈Rx \in L, y \in R: y∈Sy \in S (and xβˆ‰Xβˆ–Lx \notin X\setminus L). If xβˆ‰L,yβˆ‰Rx \notin L, y \notin R: x∈Sx \in S (and yβˆ‰Ry \notin R). We cannot have x∈L,yβˆ‰Rx \in L, y \notin R: since xx is matched and reachable, the alternating path goes β‹―β†’yβ€²β†’x\cdots \to y' \to x (with yβ€²x∈My'x \in M, so yβ€²=yy' = y), meaning y∈Ry \in R, contradiction. We cannot have xβˆ‰L,y∈Rx \notin L, y \in R: yy reachable means an alternating path reaches yy via a non-matching edge from some xβ€²βˆˆLx' \in L, then if yxβ€²β€²βˆˆMyx'' \in M the path continues to xβ€²β€²x'', so xβ€²β€²βˆˆLx'' \in L. Since yy is matched to xx in MM, x=xβ€²β€²βˆˆLx = x'' \in L, contradiction.

So each matching edge has exactly one endpoint in SS, and no vertex of SS is unmatched (unmatched vertices in XX are in UβŠ†LU \subseteq L, hence not in Xβˆ–LX \setminus L; unmatched vertices in YY are not reachable, hence not in RR). Thus ∣S∣=∣M∣=Ξ½(G)|S| = |M| = \nu(G). β–‘\square

β– 

RemarkLP Duality Perspective

KΓΆnig's theorem is a special case of LP duality. The maximum matching LP: maxβ‘βˆ‘xe\max\sum x_e subject to βˆ‘eβˆ‹vxe≀1\sum_{e \ni v} x_e \leq 1, xeβ‰₯0x_e \geq 0. The dual: minβ‘βˆ‘yv\min\sum y_v subject to yu+yvβ‰₯1y_u + y_v \geq 1 for uv∈Euv \in E, yvβ‰₯0y_v \geq 0. For bipartite graphs, the LP polytope is integral (totally unimodular constraint matrix), so LP optimum = IP optimum = Ξ½=Ο„\nu = \tau.