Accurate Trace Estimation with Fewer Random Bits via Recursive TensorSketch
Abstract
We consider the problem of estimating the trace of an implicit matrix that can only be accessed through matrix-vector products queries. The \textit{Hutchinson trace estimator}% \cite{Girard1987algorithme, article-hutchinson} is a classical sketching method for this problem. Their estimator, , and , satisfies the following guarantees: (i) , and (ii) . Generating one query vector requires random bits; thus, queries require random bits, which can be prohibitive in large-scale applications. Recent work by Meyer et al.\cite{meyer2025hutchinsonsestimatorbadkroneckertraceestimation} proposes a variant of the Hutchinson trace estimator in which each query vector in is constructed as the Kronecker product of random vectors in , requiring random bits for query vectors. The estimator of~\cite{meyer2025hutchinsonsestimatorbadkroneckertraceestimation} is unbiased; however, its variance grows exponentially with . In this work, we address this limitation by proposing a sketching-based estimator that requires random bits, yields an unbiased estimate of the trace, and simultaneously achieves a variance bound that grows polynomially with .