Optimal Randomized Comparison Based Algorithms for Collision.pdfVIP

Optimal Randomized Comparison Based Algorithms for Collision Riko Jacob? Computer Science Department, Technische Universita?t Mu?nchen jacob@in.tum.de Abstract. We consider the well known problem of finding two identical elements, called a collision, in a list of n numbers. Here, the (very fast) comparison based algorithms are randomized and will only report exist- ing collisions, and do this with (small) probability p, the success probabil- ity. We find a trade-off between p and the running time t, and show that this trade-off is optimal up to a constant factor. For worst-case running time t, the optimal success probability is p = Θ (min{t/n, 1}t/(n log t)). For expected running time t, the success probability is p = Θ (t/(n logn)). 1 Introduction We consider the problem of finding two equal elements, called a collision, in a list of numbers. This is a fundamental problem in computational complexity theory and among the first decision problems for which non-trivial lower bounds were known. It is usually formulated as the decision if all elements of a list are different, and is hence called “element uniqueness” or “element distinctness.” One way to guarantee that no element is repeated is to provide a linear sequence of strict inequalities. Actually, in many models of computation where numbers are treated as atoms, the lower bound of ?(n log n) for sorting carries over to this decision problem. The problem has been studied in all kinds of machine models, for example in the stronger algebraic decision tree model [1], in a general decision tree setting [2], or on a quantum computer [3, 4]. Motivation The investigation in this paper was triggered by a question in cryp- tography, namely the precise complexity of computing discrete logarithms in a cyclic group of roughly n ≈ 2h numbers, where the input can be specified us- ing O(h) bits. Here, cryptographic primitives are build on the assumption that exponentiation is a good one-way function, i.e., it is easy to compute