One improvement through blossom contraction #
The whole algorithm of [KKR18], arXiv:1602.03124v6, Algorithm 1 (printed p. 10) and Section 5.3 (printed pp. 11–13), organized to return one improvement. Each call performs one fresh search and at most one recursive child call. A child improvement is lifted and returned immediately. A child optimum transfers through Lemma 42; the preparation has the original defect count. This arrangement avoids recursively optimizing a contracted child before restarting a local optimization loop.
The explicit depth parameter makes the function total even on malformed input. For an edge
instance with even delta-matroid constraints and valid labels, a depth exceeding
contractionMeasure suffices. This module proves semantic correctness and termination;
encoded native resource bounds belong to the executable implementation.
The recursion organization and its bookkeeping are original; the search, preparation, contraction and lifting are the existing constructions from the cited source.
Return one valid labeling with exactly two fewer defects, or certify optimality by
returning none. The latter interpretation requires the depth and model hypotheses of
improve_spec; depth exhaustion on arbitrary raw inputs simply returns none.
Instances For
An adequately funded call either certifies optimality or removes exactly two defects. All contraction and lifting obligations are derived internally from the fresh search.
Sufficient contraction budgets give exactly the same deterministic improvement, including its outside changes. Depth affects only malformed or inadequately funded executions.