Randomised composite linear-combination-of-unitaries: its role in quantum simulation and observable estimation

TL;DR

This paper analyzes the role of randomised linear-combination-of-unitaries in quantum simulation, introduces methods to realize and estimate unphysical states, and connects these techniques to shadow tomography for efficient observable estimation.

Contribution

It introduces a quantum instrument for realizing non-completely-positive maps and constructs unbiased estimators, advancing the use of randomised LCU in quantum simulation and observable estimation.

Findings

Developed quantum circuits for randomised composite LCU

Constructed unbiased estimators for unphysical states

Demonstrated efficiency in Hamiltonian simulation and eigenstate tasks

Abstract

Randomisation is widely used in quantum algorithms to reduce the number of quantum gates and ancillary qubits required. A range of randomised algorithms, including eigenstate property estimation by spectral filters, Hamiltonian simulation, and perturbative quantum simulation, though motivated and designed for different applications, share common features in the use of unitary decomposition and Hadamard-test-based implementation. In this work, we start by analysing the role of randomised linear-combination-of-unitaries (LCU) in quantum simulations, and present several quantum circuits that realise the randomised composite LCU. A caveat of randomisation, however, is that the resulting state cannot be deterministically prepared, which often takes an unphysical form $U \rho V^\dagger$ with unitaries $U$ and $V$. Therefore, randomised LCU algorithms are typically restricted to only estimating the expectation value of a single Pauli operator. To address this, we introduce a quantum instrument that can realise a non-completely-positive map, whose feature of frequent measurement and reset on the ancilla makes it particularly suitable in the fault-tolerant regime. We then show how to construct an unbiased estimator of the effective (unphysical) state $U \rho V^\dagger$ and its generalisation. Moreover, we demonstrate how to effectively realise the state prepared by applying an operator that admits a composite LCU form. Our results reveal a natural connection between randomised LCU algorithms and shadow tomography, thereby allowing simultaneous estimation of many observables efficiently. As a concrete example, we construct the estimators and present the simulation complexity for three use cases of randomised LCU in Hamiltonian simulation and eigenstate preparation tasks.