Quantum element-wise transforms

Abstract

Quantum algorithms for basic numerical linear algebraic tasks have proven essential for translating diverse problems to a unified quantum computational context. Many of these tasks -- e.g., applying a polynomial function to the spectrum of a matrix embedded in a unitary process (a so-called block encoding), or taking linear combinations of block encodings -- are well-addressed by techniques like quantum singular value transformation (QSVT) or linear combination of unitaries (LCU). However, there exist useful matrix transforms whose realization by existing quantum algorithms is unclear or inefficient. In this work we construct improved quantum algorithms for some of these transforms, the simplest of which is a polynomial function applied element-wise. We show the space required to compute quantum element-wise transforms can be reduced exponentially in the degree of the applied function compared to prior work, and raise and rectify errors in previous constructions. We present our algorithms alongside applications to machine learning, simulation, and signal processing.