Quantum Algorithms for Solving Generalized Linear Systems via Momentum Accelerated Gradient and Schrödingerization

Abstract

In this paper, we propose a quantum algorithm that combines the momentum accelerated gradient method with Schrödingerization [S. Jin, N. Liu and Y. Yu, Phys. Rev. Lett, 133 (2024), 230602][S. Jin, N. Liu and Y. Yu, Phys. Rev. A, 108 (2023), 032603], achieving polynomial speedup over its classical counterpark in solving linear systems. The algorithm achieves a query complexity of the same order as the Schrödingerization based damped dynamical system method, namely, linear dependence on the condition number of the matrix, and can overcome the practical limitations of existing non-Schrödingerization-based quantum linear system algorithms. These limitations stem from their reliance on techniques such as VTAA and RM, which introduce substantial quantum hardware resource overhead. Furthermore, it demonstrates both theoretically and experimentally that the auxiliary variables introduced by our method do not dominate the error reduction at any point, thereby preventing a significant increase in the actual evolution time compared to the theoretical prediction. In contrast, the damped method fails to meet this criterion. This gives new perspectives for quantum algorithms for linear systems, establishing a novel analytical framework for algorithms with broader applicability, faster convergence rates, and superior solution quality.