Low-Overhead Parallelisation of LCU via Commuting Operators

TL;DR

This paper introduces a method to parallelize the Linear Combination of Unitaries (LCU) technique, significantly reducing circuit depth and resource requirements for quantum algorithms by partitioning commuting operators and employing adaptive circuits.

Contribution

It presents a novel parallelization approach for LCU, including SELECT subroutine and QROM circuits, achieving depth reduction with minimal qubit overhead and analyzing implications for fault-tolerant quantum computing.

Findings

Depth savings of approximately n/2 for molecular Hamiltonians

Parallelization reduces T-depth proportionally to the logical algorithm

Significant reduction in overall space-time volume of quantum computations

Abstract

The Linear Combination of Unitaries (LCU) method is a powerful scheme for the block encoding of operators but suffers from high overheads. In this work, we discuss the parallelisation of LCU and in particular the SELECT subroutine of LCU based on partitioning of observables into groups of commuting operators, as well as the use of adaptive circuits and teleportation that allow us to perform required Clifford circuits in constant depth. We additionally discuss the parallelisation of QROM circuits which are a special case of our main results, and provide methods to parallelise the action of multi-controlled gates on the control register. We only require an $O(\log n)$ factor increase in the number of qubits in order to produce a significant depth reduction, with prior work suggesting that for molecular Hamiltonians, the depth saving is $O(n)$, and numerics indicating depth savings of a factor approximately $n/2$. The implications of our method in the fault-tolerant setting are also considered, noting that parallelisation reduces the $T$-depth by the same factor as the logical algorithm, without changing the $T$-count, and that our method can significantly reduce the overall space-time volume of the computation, even when including the increased number of $T$ factories required by parallelisation.