arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.11014cond-mat.dis-nnquant-ph
收藏

用于多体物理的神经网络量子态

Neural-network quantum states for many-body physics

Matija Medvidović, Javier Robledo Moreno

AI总结:

本综述梳理了神经网络量子态在多体物理中的应用,推导各类变分蒙特卡洛方法核心方程,探讨变分流形几何与实现瓶颈,概述相关最新计算成果。

摘要:

近年来,变分量子计算借鉴了机器学习领域的诸多工具与算法。研究人员借助深度学习启发的试探态的强大表达能力与高效的基于梯度的优化方法,已证明这类试探态可准确建模自旋、费米子与量子比特系统中的多体关联现象。本综述推导了不同类型变分蒙特卡洛(VMC)方法的核心方程,涵盖基态搜索、时间演化与重叠优化,并讨论了量子态层析等数据驱动任务。文中重点探讨了变分流形的几何结构以及实际实现中的瓶颈,同时概述了第一性原理基态与实时计算的最新成果。

英文摘要:

Variational quantum calculations have borrowed many tools and algorithms from the machine learning community in the recent years. Leveraging great expressive power and efficient gradient-based optimization, researchers have shown that trial states inspired by deep learning problems can accurately model many-body correlated phenomena in spin, fermionic and qubit systems. In this review, we derive the central equations of different flavors variational Monte Carlo (VMC) approaches, including ground state search, time evolution and overlap optimization, and discuss data-driven tasks like quantum state tomography. An emphasis is put on the geometry of the variational manifold as well as bottlenecks in practical implementations. An overview of recent results of first-principles ground-state and real-time calculations is provided.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑