神经量子态与尖峰分子波函数:诅咒还是福音?
Neural Quantum States and Peaked Molecular Wave Functions: Curse or Blessing?
AI总结:
针对神经量子态在量子化学中受分子波函数尖峰结构、哈密顿量项数多限制的问题,提出无放回自回归采样、低成本局部能量替代等方法,单GPU实现118量子比特分子计算,性能超CCSD(T)基准,效率提升超一个数量级。
摘要:
神经量子态领域近年来取得了巨大进展,已成为计算量子多体物理领域极具竞争力的工具。然而,迄今为止其最大的研究成果大多集中在相互作用自旋系统上,在量子化学中的实用性仍有待验证。该领域面临两大主要难题:一是分子波函数的尖峰结构会阻碍采样过程,二是二次量子化哈密顿量中包含大量项,限制了方法向更大分子规模的扩展。在本文中,我们同时解决了这两个问题,并提出尖峰结构实际上可能是实现计算效率大幅提升的关键。具体而言,我们提出了一种新型的无放回自回归采样算法,以及一种计算成本更低的局部能量替代值的计算流程。我们还对随机重构优化技术进行了定制化改进,并搭配了高度优化的GPU实现方案。最终,我们的计算所需资源大幅减少,与已有研究相比速度提升超过一个数量级。在单块GPU上,我们研究了包含多达118个量子比特的分子,在包含约$10^{15}$个斯莱特行列式的希尔伯特空间中,性能超越了“金标准”CCSD(T)基准,该空间规模比此前实现的结果大出数个数量级。我们认为本研究凸显了NQS(神经量子态)在高难度量子化学计算中的应用前景,也为未来的方法开发提供了良好的基础。
英文摘要:
The field of neural quantum states has recently experienced a tremendous progress, making them a competitive tool of computational quantum many-body physics. However, their largest achievements to date mostly concern interacting spin systems, while their utility for quantum chemistry remains yet to be demonstrated. Two main complications are the peaked structure of the molecular wave functions, which impedes sampling, and large number of terms in second quantised Hamiltonians, which hinders scaling to larger molecule sizes. In this paper we address these issues jointly and argue that the peaked structure might actually be key to drastically more efficient calculations. Specifically, we introduce a novel algorithm for autoregressive sampling without replacement and a procedure to calculate a computationally cheaper surrogate for the local energy. We complement them with a custom modification of the stochastic reconfiguration optimisation technique and a highly optimised GPU implementation. As a result, our calculations require substantially less resources and exhibit more than order of magnitude speedup compared to the previous works. On a single GPU we study molecules comprising up to 118 qubits and outperform the ``golden standard'' CCSD(T) benchmark in Hilbert spaces of $\sim 10^{15}$ Slater determinants, which is orders of magnitude larger than what was previously achieved. We believe that our work underscores the prospect of NQS for challenging quantum chemistry calculations and serves as a favourable ground for the future method development.
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。