Submodular maximization and its generalization through an intersection cut lens
Liding Xu,Leo Liberti
DOI: https://doi.org/10.1007/s10107-024-02059-2
IF: 2.7
2024-02-19
Mathematical Programming
Abstract:We study a mixed-integer set S:={(x,t)∈{0,1}n×R:f(x)≥t}$$\mathcal {S}:=\{(x,t) \in \{0,1\}^n \times \mathbb {R}: f(x) \ge t\}$$ arising in the submodular maximization problem, where f is a submodular function defined over {0,1}n$$\{0,1\}^n$$. We use intersection cuts to tighten a polyhedral outer approximation of S$$\mathcal {S}$$. We construct a continuous extension F¯f$$\bar{\textsf{F}}_f$$ of f, which is convex and defined over the entire space Rn$$\mathbb {R}^n$$. We show that the epigraph epi(F¯f)$${{\,\textrm{epi}\,}}(\bar{\textsf{F}}_f)$$ of F¯f$$\bar{\textsf{F}}_f$$ is an S$$\mathcal {S}$$-free set, and characterize maximal S$$\mathcal {S}$$-free sets containing epi(F¯f)$${{\,\textrm{epi}\,}}(\bar{\textsf{F}}_f)$$. We propose a hybrid discrete Newton algorithm to compute an intersection cut efficiently and exactly. Our results are generalized to the hypograph or the superlevel set of a submodular-supermodular function over the Boolean hypercube, which is a model for discrete nonconvexity. A consequence of these results is intersection cuts for Boolean multilinear constraints. We evaluate our techniques on max cut, pseudo Boolean maximization, and Bayesian D-optimal design problems within a MIP solver.
mathematics, applied,computer science, software engineering,operations research & management science