It has recently been shown that local search is surprisingly good at finding satisfying assignments for certain classes of CNF formulas [24]. In this paper we demonstrate that the power of local search for satisfiability testing can be further enhanced by employinga new strategy, called "mixed random walk", for escaping from local minima. We present experimental results showing how this strategy allows us to handle formulas that are substantially larger than those that can be solved with basic local search. We also present a detailed comparison of our random walk strategy with simulated annealing. Our results show that mixed random walk is the superior strategy on several classes of computationally difficult problem instances. Finally, we present results demonstrating the effectiveness of local search with walk for solving circuit synthesis and diagnosis problems.

Abstract—We introduce a simple erasure recovery algorithm for codes derived from cascades of sparse bipartite graphs and analyze the algorithm by analyzing a corresponding discrete-time random process. As a result, we obtain a simple criterion involving the fractions of nodes of different degrees on both sides of the graph which is necessary and sufficient for the decoding process to finish successfully with high probability. By carefully designing these graphs we can construct for any given rate and any given real number a family of linear codes of rate which can be encoded in time proportional to ��@I A times their block length. Furthermore, a codeword can be recovered with high probability from a portion of its entries of length @IC A or more. The recovery algorithm also runs in time proportional to ��@I A. Our algorithms have been implemented and work well in practice; various implementation issues are discussed. Index Terms—Erasure channel, large deviation analysis, lowdensity parity-check codes. I.

Given a monotone graph property P , consider p (P ), the probability that a random graph with edge probability p will have P . The function d p (P )=dp is the key to understanding the threshold behavior of the property P . We show that if d p (P )=dp is small (corresponding to a non-sharp threshold), then there is a list of graphs of bounded size such that P can be approximated by the property of having one of the graphs as a subgraph. One striking consequences of this result is that a coarse threshold for a random graph property can only happen when the value of the critical edge probability is a rational power of n.

We propose a definition of `constrainedness' that unifies two of the most common but informal uses of the term. These are that branching heuristics in search algorithms often try to make the most "constrained" choice, and that hard search problems tend to be "critically constrained". Our definition of constrainedness generalizes a number of parameters used to study phase transition behaviour in a wide variety of problem domains. As well as predicting the location of phase transitions in solubility, constrainedness provides insight into why problems at phase transitions tend to be hard to solve. Such problems are on a constrainedness "knife-edge", and we must search deep into the problem before they look more or less soluble. Heuristics that try to get off this knife-edge as quickly as possible by, for example, minimizing the constrainedness are often very effective. We show that heuristics from a wide variety of problem domains can be seen as minimizing the constrainedness (or proxies ...

ABSTRACT: Let � be a random Boolean formula that is an instance of 3-SAT. We consider the problem of computing the least real number � such that if the ratio of the number of clauses over the number of variables of � strictly exceeds �, then � is almost certainly unsatisfiable. By a well-known and more or less straightforward argument, it can be shown that ��5.191. This upper bound was improved by Kamath et al. to 4.758 by first providing new improved bounds for the occupancy problem. There is strong experimental evidence that the value of � is around 4.2. In this work, we define, in terms of the random formula �, a decreasing sequence of random variables such that, if the expected value of any one of them converges to zero, then � is almost certainly unsatisfiable. By letting the expected value of the first term of the sequence converge to zero, we obtain, by simple and elementary computations, an upper bound for � equal to 4.667. From the expected value of the second term of the sequence, we get the value 4.601�. In general, by letting the

this paper, in order to estimate ! better, a new `syntactic' approach is proposed: This is suggested, e.g., simply by practical experience with a computer-based random generator and solver of formulae. The generator is completely ignorant of semantics, yet for large n, it produces (within a realistic timeframe) either only satisfiable formulae, for values of c a little below 4:25, or only unsatisfiable formulae, for values of c a little above 4:25. Only a certain class of formulae is likely to come out of the generator, the `typical' formulae. If we could describe (to a sufficient degree) their syntactic structure, the convergence to 0 or to +1 of the expectation of the number of solutions, computed only for these typical formulae, should give a very good indication as to the value of the threshold.

Recently there has been a great amount of interest in Random Constraint Satisfaction Problems, both from an experimental and a theoretical point of view. Rather intruigingly, experimental results with various models for generating random CSP instances suggest a "threshold-like" behaviour and some theoretical work has been done in analyzing these models when the number of variables is asymptotic. In this paper we show that the models commonly used for generating random CSP instances suffer from a wrong parameterization which makes them unsuitable for asymptotic analysis. In particular, when the number of variables becomes large almost all instances they generate are, trivially, overconstrained. We then present a new model that is suitable for asymptotic analysis and, in the spirit of random SAT, we derive lower and upper bounds for its parameters so that the instances generated are "almost surely" over- and underconstrained, respectively. Finally, we apply the technique introduced in [19], to one of the popular models in Artificial Intelligence and derive sharper estimates for the probability of being overconstrained as a function of the number of variables. 1

We introduce a new set of probabilistic analysis tools based on the analysis of And-Or trees with random inputs. These tools provide a unifying, intuitive, and powerful framework for carrying out the analysis of several previously studied random processes of interest, including random loss-resilient codes, solving random k-SAT formula using the pure literal rule, and the greedy algorithm for matchings in random graphs. In addition, these tools allow generalizations of these problems not previously analyzed to be analyzed in a straightforward manner. We illustrate our methodology on the three problems listed above. 1 Introduction We introduce a new set of probabilistic analysis tools related to the amplification method introduced by [12] and further developed and used in [13, 5]. These tools provide a unifying, intuitive, and powerful framework for carrying out the analysis of several previously studied random processes of interest, including the random loss-resilient codes introduced ...

4, and Toby Walsh 5

Let F 3 (n; m) be a random 3-SAT formula formed by selecting uniformly, independently, and with replacement, m clauses among all 8 \Gamma n 3 \Delta possible 3-clauses over n variables. It has been conjectured that there exists a constant r 3 such that for any ffl ? 0, F 3 (n; (r 3 \Gamma ffl)n) is almost surely satisfiable, but F 3 (n; (r 3 + ffl)n) is almost surely unsatisfiable. The best lower bounds for the potential value of r 3 have come from analyzing rather simple extensions of unit-clause propagation. Recently, it was shown [2] that all these extensions can be cast in a common framework and analyzed in a uniform manner by employing differential equations. Here, we determine optimal algorithms expressible in that framework, establishing r 3 ? 3:26. We extend the analysis via differential equations, and make extensive use of a new optimization problem we call "max-density multiple-choice knapsack". The structure of optimal knapsack solutions elegantly characterizes the choi...