A self-avoiding walk (SAW) is extendable [10,13] if it can be extended into an infinite SAW. We give a simple proof that, for every lattice, extendable SAWs admit the same connective constant that the general SAWs and we give an optimal linear algorithm to generate random extendable SAWs. Our algorithm can generate every extendable SAW in dimension 2. For dimension d > 2, it generates only a subset of the extendable SAWs. We conjecture that this subset is "large" and has the same connective constant that the extendable SAWs. Our algorithm produces a kinetic distribution of the extendable SAWs, for which the critical exponent ν ≈ .57 for d = 2, .51 for d = 3 and .50 for d = 4, 5, 6. Keywords: Self Avoiding Walks, Connective Constant, Critical Exponent ν, Random Generation.
IntroductionA walk on a lattice is self-avoiding if it never passes twice through the same vertex (see Figure 1). Self-avoiding walks (SAWs) appeared as a model for polymers [7]. They also have applications in statistical physics [8] and in probability theory [15]. Formally speaking, for instance on the square grid Z × Z on plane, a walk is a sequence W = (w 0 , w 1 , . . . w n ) of vertices such that, for all i < n, w i and w i+1 are neighbors (i.e. if w i = (x i , y i ) and w i+1 = (x i+1 , y i+1 ), then (x i+1 = x i and y i+1 = y i ± 1) or (y i+1 = y i and x i+1 = x i ± 1)). The walk W is self-avoiding if i = j ⇒ w i = w j .Despite a very simple and natural definition, and although they are very closed to random walks (which are standard objects, very well studied and known) self-avoiding walks ask many problems, both theoretical and practical.Given a lattice (for instance, the hexagonal lattice on plane or the cubic one Z × Z × Z in space), the two main theoretical questions concerning SAWs on that lattice are:• What is the number c n of SAWs of length n?• Given a distribution on the SAWs (for instance the uniform distribution on the SAWs of length n), what is the average Euclidean distance d n between 1 Part of this work was done while one of the authors (P.P.) was visiting the University of Cape Town. It is strongly believed that the critical exponents γ and ν depend only on the dimension (and not on the lattice) while the connective constant µ also depends on the lattice. It is known that the connective constant (which is equal to lim n→∞ c 1/n n ) exists for every lattice. Despite decades of efforts, these questions remain unsolved, especially in low dimensions. It has recently been shown [6] ≈ 2.64). For two-dimensional lattices, it is conjectured that γ = 43/32 and that, for the uniform distribution, ν = 3/4.From a practical point of view, it is very interesting to generate random self-avoiding walks since these walks arise as models for various physical phenomena. In addition, generating random self avoiding walks yields an approximation to the critical exponents and the connective constant. There are many algorithms to generate random self-avoiding walks; among them the pivot algorithm [14,12], the Berretti-Sokal algorithm ...