The satisfiability threshold of random linear equations over finite commutative rings

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

University of Waterloo

Abstract

In this thesis we study systems of random linear equations as a random constraint satisfaction problem (CSP). Determining the satisfiability threshold is one of the most fundamental problems in the study of any random CSP: how many constraints can a typical CSP have before it can no longer be satisfied? This question has been answered in the setting of linear equations over a finite field [7]. We investigate natural generalizations of [7] in the setting of finite commutative rings. Under very mild assumptions, the satisfiability threshold of a random CSP is either linear or sublinear in the number of variables. We provide a complete characterization of the linear and sublinear regimes for our model of random linear equations over a ring. In the sublinear regime, we show that the threshold is coarse and determine its exact value. In the linear regime, we determine the exact threshold under the assumption that the underlying ring is a principal ideal ring. We show that the threshold over principal ideal rings is sharp, and coincides exactly with the satisfiability threshold over fields. This extends a striking phenomenon in [7], where it is observed that the satisfiability threshold over finite fields is the same for every choice of field. We show by example that this phenomenon does not extend to non-principal ideal rings. We construct a non-principal ideal ring R, and determine the threshold for a family of distributions of linear equations over R. Not only does the satisfiability threshold differ from linear equations over a field, we find that the satisfiability threshold depends on parameters of the distribution in ways that are not observed for fields and principal ideal rings. The final section of this thesis is devoted to contiguity results among models of random bipartite graphs. In particular, we show that the models of [7] and [14] are contiguous for some choice of parameters, and we correct a mistake made in [13]. These results are used in earlier sections of the thesis, and may be of independent interest.

Description

Citation

Endorsement

Review

Supplemented By

Referenced By