Reconfiguration of Satisfying Assignments and Subset Sums: Easy to Find, Hard to Connect

TL;DR

This paper proves that certain simple-looking reconfiguration problems, involving transforming solutions while maintaining constraints, are computationally hard (PSPACE-complete), even when the problems are restricted to easy instances.

Contribution

The paper establishes PSPACE-completeness for several reconfiguration problems with simple constraints, highlighting their computational difficulty.

Findings

Reconfiguration of NAE 3-SAT assignments is PSPACE-complete.

Subset sum reconfiguration with limited element changes is PSPACE-complete.

Path finding in polytopes within the hypercube is PSPACE-complete.

Abstract

We consider the computational complexity of reconfiguration problems, in which one is given two combinatorial configurations satisfying some constraints, and is asked to transform one into the other using elementary transformations, while satisfying the constraints at all times. Such problems appear naturally in many contexts, such as model checking, motion planning, enumeration and sampling, and recreational mathematics. We provide hardness results for problems in this family, in which the constraints and operations are particularly simple. More precisely, we prove the PSPACE-completeness of the following decision problems: $\bullet$ Given two satisfying assignments to a planar monotone instance of Not-All-Equal 3-SAT, can one assignment be transformed into the other by single variable `flips' (assignment changes), preserving satisfiability at every step? $\bullet$ Given two subsets of a set S of integers with the same sum, can one subset be transformed into the other by adding or removing at most three elements of S at a time, such that the intermediate subsets also have the same sum? $\bullet$ Given two points in $\{0,1\}^n$ contained in a polytope P specified by a constant number of linear inequalities, is there a path in the n-hypercube connecting the two points and contained in P? These problems can be interpreted as reconfiguration analogues of standard problems in NP. Interestingly, the instances of the NP problems that appear as input to the reconfiguration problems in our reductions can be shown to lie in P. In particular, the elements of S and the coefficients of the inequalities defining P can be restricted to have logarithmic bit-length.