Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Source. Section 4.14 and its extension, printed p. 458 (published PDF).
Statement. If an odd circuit is contracted, every matching of lifts to a matching of by adding edges of . If the quotient vertex is exposed, the lift may leave any specified vertex of exposed.
More generally, a complete pseudovertex expansion on original vertices has a near-perfect matching omitting any specified vertex. Every quotient matching lifts across by adding exactly internal edges. Consequently any increase in quotient matching size lifts to the same increase.
The extension printed after 4.14 (p. 458) asserts only a matching of leaving exactly one exposed vertex and compatible with a given matching of the contracted graph. The choice of an arbitrary exposed vertex in a complete expansion is the form Section 6.5 (pp. 464–465) invokes when it cites 4.14; it is proved below.
Proof. A quotient matching has at most one edge incident to the contracted vertex. If there is such an edge, let be its original endpoint in ; otherwise choose any . The odd-circuit matching omitting is compatible with all edges of , giving the first claim.
For a nested expansion, induct on the number of remembered contractions. An ordinary vertex has the empty matching. At the top contraction, let the odd circuit have child blocks , each already possessing the asserted property. To omit a specified original vertex , choose the circuit matching omitting the child containing . For every other child, its incident selected circuit edge has a particular endpoint in that child. Match internally all but that endpoint, using induction. In the omitted child, match internally all but .
These matchings and circuit edges are disjoint. Every original vertex except is covered, so the internal edge count is . The remembered edge identities supply exactly the required attachment vertices, even when parallel edges are present.
For a quotient matching meeting in one crossing edge, choose to be that edge's original endpoint. If it has no such edge, choose freely. This proves compatibility and the fixed edge-count increment. Applying it to disjoint current blocks or successively reversing nested contractions proves the final assertion.
No optimality converse for an arbitrary quotient is asserted. That converse requires the hypotheses in Section 4.15.