🤖 AI Summary
本文解决了最大割问题中SDP精确性识别的复杂性问题，通过构造具有多项式界整数权重的平方和对偶证书证明了加权图和简单无权图的强NP难性。
📝 Abstract
The Semidefinite Programming (SDP) relaxation of the Maximum Cut (Max-Cut) problem is exact when its optimal value equals the integer maximum cut, geometrically corresponding to a rank-1 optimal solution. While the pioneering work of Delorme and Poljak established that recognizing exactness is NP-hard, their reduction relied on exponentially scaling edge weights. This established only weak NP-hardness and explicitly left open the complexity for simple, unweighted graphs. In this paper, we resolve the computational complexity of the exactness property. First, we prove that deciding SDP exactness for weighted graphs is strongly NP-hard by constructing a sum-of-squares dual certificate with polynomially bounded integer weights. Second, we extend this hardness to simple, unweighted graphs via a geometric embedding of restricted Not-All-Equal 4-SAT into edge-disjoint clique structures. Both proofs utilize geometric locking mechanisms that force the continuous SDP relaxation to an absolute global minimum, decoupling the continuous bounds from the underlying combinatorial hardness.