Bypassing the computational bottleneck of quantum-embedding theories for strong electron correlations with machine learning
Bypassing the computational bottleneck of quantum-embedding theories for strong electron correlations with machine learning
复制标题
DOI：
10.1103/physrevresearch.3.013101
复制
发表时间：
2020-06
影响因子：
4.2
通讯作者：
John Rogers;Tsung-Han Lee;S. Pakdel;Wenhu Xu;V. Dobrosavljevi'c;Yongxin Yao;O. Christiansen;Nicola Lanatà
中科院分区：
文献类型：
--
作者：
John Rogers;Tsung-Han Lee;S. Pakdel;Wenhu Xu;V. Dobrosavljevi'c;Yongxin Yao;O. Christiansen;Nicola Lanatà
对强关联物质进行量子力学模拟的一个主要障碍是，在目前可用的理论工具下，精确的计算往往过于昂贵，根本不可行。在这里，我们设计了一个计算框架结合量子嵌入（QE）方法与机器学习。这使我们能够完全绕过QE算法中计算最昂贵的组件，使其总成本与裸密度泛函理论（DFT）相当。我们进行基准计算的一系列锕系元素的系统，我们的方法准确地描述了相关效应，减少了数量级的计算成本。我们认为，通过产生更大规模的训练数据集，将有可能将我们的方法应用于具有任意化学计量和晶体结构的系统，为凝聚态物理，化学和材料科学中几乎无限的应用铺平道路。
A cardinal obstacle to performing quantum-mechanical simulations of strongly-correlated matter is that, with the theoretical tools presently available, sufficiently-accurate computations are often too expensive to be ever feasible. Here we design a computational framework combining quantum-embedding (QE) methods with machine learning. This allows us to bypass altogether the most computationally-expensive components of QE algorithms, making their overall cost comparable to bare Density Functional Theory (DFT). We perform benchmark calculations of a series of actinide systems, where our method describes accurately the correlation effects, reducing by orders of magnitude the computational cost. We argue that, by producing a larger-scale set of training data, it will be possible to apply our method to systems with arbitrary stoichiometries and crystal structures, paving the way to virtually infinite applications in condensed matter physics, chemistry and materials science.