Schr\"odingerization for quantum linear systems problems with near-optimal dependence on matrix queries

TL;DR

This paper introduces a quantum algorithm for solving linear systems by transforming them into Schr"odinger-type problems, enabling near-optimal complexity and broad applicability to positive definite and Hermitian matrices.

Contribution

It develops a Schr"odingerization-based quantum algorithm for linear systems with near-optimal dependence on matrix queries, including a block preconditioning technique for improved efficiency.

Findings

Achieves near-linear scaling in condition number.

Provides detailed implementation and error analysis.

Extends Schr"odingerization to general Hermitian matrices.

Abstract

We develop a quantum algorithm for linear algebraic equations $ A\bb{x} = \bb{b} $ from the perspective of Schr\"odingerization-form problems, which are characterized by a system of linear convection equations in one higher dimension. When $ A $ is positive definite, the solution $ \bb{x} $ can be interpreted as the steady-state solution to a system of linear ordinary differential equations (ODEs). This ODE system can be solved by using the linear combination of Hamiltonian simulation (LCHS) method in \cite{ACL2023LCH2}, which serves as the continuous implementation of the Fourier transform in the Schr\"odingerization method from \cite{JLY22SchrShort, JLY22SchrLong}. Schr\"odingerization transforms linear partial differential equations (PDEs) and ODEs with non-unitary dynamics into Schr\"odinger-type systems via the so-called warped phase transformation that maps the equation into one higher dimension. When $ A $ is a general Hermitian matrix, the inverse matrix can still be represented in the LCHS form in \cite{ACL2023LCH2}, but with a kernel function based on the Fourier approach in \cite{Childs2017QLSA}. Although this LCHS form provides the steady-state solution to a system of linear ODEs associated with the least-squares equation, applying Schr\"odingerization to this least-squares system is not appropriate, as it results in a much larger condition number. We demonstrate that in both cases, the solution $ \bb{x} $ can be expressed as the LCHS of Schr\"odingerization-form problems. We provide a detailed implementation and error analysis. Furthermore, we incorporate a block preconditioning technique to achieve nearly linear scaling in the condition number, thereby attaining near-optimal query complexity.