An undecidable problem in finite combinatorics

Extract

Problems of computing probabilities of statements about large, finite structures have become an important subject area of finite combinatorics. Within the last two decades many researchers have turned their attention to such problems and have developed a variety of methods for dealing with them. Applications of these ideas include nonconstructive existence proofs for graphs with certain properties by showing the properties occur with nonzero probabilities (for examples see Erdös and Spencer [ES] and Bollobás [Bo]), and determination of average running times for sorting algorithms by computing asymptotic probabilities of statements about permutations (see Knuth [Kn]). Two types of techniques recur in solutions to such problems: probabilistic techniques, such as those used by Erdös and Spencer [ES], and classical assymptotic techniques, such as those surveyed by Bender [Be] and de Bruijn [Br]. Studying this body of techniques, one notices that characteristics of these problems suggest certain methods of solution, in much the same way that the form of an integrand may suggest certain substitutions. The question arises, then, as to whether there is a systematic way to approach these problems: is there an algorithm for computing asymptotic probabilities? I will show that the answer is “no”—for any reasonable formulation, the problem of computing asymptotic probabilities is undecidable.

The main theorem of the paper is Theorem 1.6, which says that there is a finitely axiomatizable class in which every first order sentence has an asymptotic probability of 0 or 1—i.e., is almost always true or almost always false in finite structures—but for which the problem of deciding whether a sentence has asymptotic probability 0 or 1 is undecidable. Heretofore, classes known to have such a 0-1 law have had decidable asymptotic probability problems (see Lynch [Ly] for examples and a discussion of previous work in the area).