On Quantum Versions of the Yao Principle

Abstract

The classical Yao principle states that the complexity R_epsilon(f) of an optimal randomized algorithm for a function f with success probability 1-epsilon equals the complexity max_mu D_epsilon^mu(f) of an optimal deterministic algorithm for f that is correct on a fraction 1-epsilon of the inputs, weighed according to the hardest distribution mu over the inputs. In this paper we investigate to what extent such a principle holds for quantum algorithms. We propose two natural candidate quantum Yao principles, a ``weak'' and a ``strong'' one. For both principles, we prove that the quantum bounded-error complexity is a lower bound on the quantum analogues of max mu D_epsilon^mu(f). We then prove that equality cannot be obtained for the ``strong'' version, by exhibiting an exponential gap. On the other hand, as a positive result we prove that the ``weak'' version holds up to a constant factor for the query complexity of all symmetric Boolean functions