An upper bound for the connective constant of the square lattice
In plain words
Self-avoiding walks on a square grid never visit a point twice. Their count grows roughly like a fixed number, the connective constant, raised to the number of steps. Its exact value is unknown; estimates put it near 2.638. This paper lowers Couronné's upper bound of 2.662342426 to 2.6600756. It follows bridges, walks that can be extended forever and grow in number at the same rate, keeping a finite memory of part of their past and of points they could enter only by trapping themselves. Two separately written programs check the proof's table of 4,124,965,633 memories and weights.
Abstract
Let μ be the connective constant of self-avoiding walks on the square lattice Z2, the exponential growth rate of the number of n-step self-avoiding walks. We prove μ ≤ 2.6600756. The previous best upper bound, μ ≤ 2.662342426, is due to Couronné. As in the work of Pönitz and Tittmann and of Couronné, the bound comes from a finite automaton whose states remember a bounded part of the past of the walk. Our automaton counts bridges, and, as in Couronné’s automaton, its states may contain cells that the walk has not visited; here these are cells that no infinite self-avoiding continuation of the walk can visit, recognized by three tests that depend only on the state and the step. Distinct bridges, preceded by a fixed infinite history, then give distinct paths in the automaton, so the Hammersley–Welsh theorem turns a bound for the number of paths into a bound for μ. The bound follows from a certificate, consisting of the automaton (4,124,965,633 states) and a vector of integer weights, which a program checks in exact integer arithmetic.