Notes on L-/M-convex functions and the separation theorems

The concepts of L-convex function and M-convex function have recently been introduced by Murota as generalizations of submodular function and base polyhedron, respectively, and discrete separation theorems are established for L-convex/concave functions and for M-convex/concave functions as generalizations of Frank’s discrete separation theorem for submodular/supermodular set functions and Edmonds’ matroid intersection theorem. This paper shows the equivalence between Murota’s L-convex functions and Favati and Tardella’s submodular integrally convex functions, and also gives alternative proofs of the separation theorems that provide a geometric insight by relating them to the ordinary separation theorem in convex analysis. Received: November 27, 1997 / Accepted: December 16, 1999?Published online May 12, 2000     

A capacity scaling algorithm for convex cost submodular flows

This paper presents a scaling scheme for submodular functions. A small but strictly submodular function is added before scaling so that the resulting functions should be submodular. This scaling scheme leads to a weakly polynomial algorithm to solve minimum cost integral submodular flow problems with separable convex cost functions, provided that an oracle for exchange capacities is available.     

The expressive power of binary submodular functions

We investigate whether all Boolean submodular functions can be decomposed into a sum of binary submodular functions over a possibly larger set of variables. This question has been considered in several different contexts in computer science, including computer vision, artificial intelligence, and pseudo-Boolean optimisation. Using a connection between the expressive power of valued constraints and certain algebraic properties of functions, we answer this question negatively.Our results have several corollaries. First, we characterise precisely which submodular polynomials of arity 4 can be expressed by binary submodular polynomials. Next, we identify a novel class of submodular functions of arbitrary arities which can be expressed by binary submodular functions, and therefore minimised efficiently using a so-called expressibility reduction to the Min-Cut problem. More importantly, our results imply limitations on this kind of reduction and establish, for the first time, that it cannot be used in general to minimise arbitrary submodular functions. Finally, we refute a conjecture of Promislow and Young on the structure of the extreme rays of the cone of Boolean submodular functions.     

Optimization over the polyhedron determined by a submodular function on a co-intersecting family

A greedy algorithm solves the problem of maximizing a linear objective function over the polyhedron (called the submodular polyhedron) determined by a submodular function on a distributive lattice or a ring family. We generalize the problem by considering a submodular function on a co-intersecting family and give an algorithm for solving it. Here, simple-minded greedy augmentations do not work any more and some complicated augmentations with multiple exchanges are required. We can find an optimal solution by at most 1/2n(n – 1) augmentations, wheren is the number of the variables and we assume a certain oracle for computing multiple exchange capacities.     

Clique partitioning with value-monotone submodular cost

We consider the problem of partitioning a graph into cliques of bounded cardinality. The goal is to find a partition that minimizes the sum of clique costs where the cost of a clique is given by a set function on the nodes. We present a general algorithmic solution based on solving the problem variant without the cardinality constraint. We obtain constant factor approximations depending on the solvability of this relaxation for a large class of submodular cost functions which we call value-monotone submodular functions. For special graph classes we give optimal algorithms.     

Generalized polymatroids and submodular flows

Polyhedra related to matroids and sub- or supermodular functions play a central role in combinatorial optimization. The purpose of this paper is to present a unified treatment of the subject. The structure of generalized polymatroids and submodular flow systems is discussed in detail along with their close interrelation. In addition to providing several applications, we summarize many known results within this general framework.Supported by a grant from the Alexander von Humboldt Stiftung and by the Institut für Ökonometrie und Operations Research of the University of Bonn, Bonn, Nassestr. 2, Federal Republic of Germany.     

On the subdifferential of a submodular function

The author recently introduced a concept of a subdifferential of a submodular function defined on a distributive lattice. Each subdifferential is an unbounded polyhedron. In the present paper we determine the set of all the extreme points and rays of each subdifferential and show the relationship between subdifferentials of a submodular function and subdifferentials, in an ordinary sense of convex analysis, of Lovász's extension of the submodular function. Furthermore, for a modular function on a distributive lattice we give an algorithm for determining which subdifferential contains a given vector and finding a nonnegative linear combination of extreme vectors of the subdifferential which expresses the given vector minus the unique extreme point of the subdifferential.     

Polybasic polyhedra: structure of polyhedra with edge vectors of support size at most 2

It has widely been recognized that submodular set functions and base polyhedra associated with them play fundamental and important roles in combinatorial optimization problems. In the present paper, we introduce a generalized concept of base polyhedron. We consider a class of pointed convex polyhedra in R^V whose edge vectors have supports of size at most 2. We call such a convex polyhedron a polybasic polyhedron. The class of polybasic polyhedra includes ordinary base polyhedra, submodular/supermodular polyhedra, generalized polymatroids, bisubmodular polyhedra, polybasic zonotopes, boundary polyhedra of flows in generalized networks, etc. We show that for a pointed polyhedron PR^V, the following three statements are equivalent:

(1) P is a polybasic polyhedron.

(2) Each face of P with a normal vector of the full support V is obtained from a base polyhedron by a reflection and scalings along axes.

(3) The support function of P is a submodular function on each orthant of R^V.

This reveals the geometric structure of polybasic polyhedra and its relation to submodularity.     

Minimization on submodular flows

Edmonds and Giles [1] discuss functions on the arcs of a digraph satisfying submodular set constraints. We call such functions submodular flows, show that the difference of group-valued submodular flows is a network circulation in a certain auxiliary digraph, and derive a criterion for the existence of group-valued submodular flows. We develop a method for maximizing the value of a group-valued submodular flow in a specified arc and a negative circuit method for minimizing certain functions on ring-valued submodular flows. In particular, algebraic linear functions over modules and certain semimodules as well as quotients of linear functions over totally ordered, commutative fields can be minimized.     

Supermodularity in Various Partition Problems

Supermodular and submodular functions have attracted a great deal of attention since the seminal paper of Lovász. Recently, supermodular functions were studied in the context of some optimal partition problems. We completely answer a question arisen there whether a certain partition function is supermodular.