Collaborative Research: Frameworks: Scalable Modular Software and Methods for High-Accuracy Materials and Condensed Phase Chemistry Simulation

电子在材料中的排列方式产生了各种各样的不同行为。我们可以观察这些行为，并在各种技术中使用它们。然而，这些行为的预测是一个严峻的挑战。这使得新材料的成功设计更加困难。“材料基因组计划”的目标是根据量子物理定律，利用计算机模拟来模拟电子。这将使研究人员能够设计出具有理想性能的新材料。该项目旨在建立快速准确的计算机程序来模拟这些新材料。这些程序结合了计算机科学、量子化学和凝聚态物理的进步。它们将在基于python的开源社区代码中实现。此分布模型允许其他研究人员使用此代码并贡献新特性。本研究通过开发低尺度、高精度方法的新型并行实现，解决了量子材料模拟中现有软件网络基础设施的空白。特别是，平均场计算的新技术将被开发，这将作为周期性耦合簇和量子蒙特卡罗方法的基础。最先进的稀疏和张量分解技术将被用来实现良好的系统大小缩放，同时在每个数值方案中保持精度。关键的是，这些方法将使用高效的高级软件抽象来开发，在PySCF中作为python级模块实现，利用Cyclops库进行大规模并行执行。库软件基础设施也将被扩展，以通过源代码到源代码的自动区分来最大限度地提高生产力，并支持在新兴的基于gpu的超级计算架构上执行稀疏内核。该奖项由美国国家科学基金会高级网络基础设施办公室、美国国家科学基金会数学和物理科学理事会内的材料研究部和化学部联合支持。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

How electrons are arranged in materials gives rise to a large variety of different behaviors. We can observe these behaviors and use them in various technologies. However, the prediction of these behaviors is a serious challenge. This makes the successful design of new materials harder. The goal of the Materials Genome Initiative is to use computer simulations to model electrons according to the laws of quantum physics. This will allow researchers to design new materials with desired properties. This project aims to build fast and accurate computer programs which simulate those new materials. These programs combine advances in computer science, quantum chemistry, and condensed-matter physics. They will be implemented in an open-source Python-based community code. This distribution model allows other researchers to use this code and to contribute new features.This research addresses gaps in existing software cyberinfrastructure in quantum materials simulation, by developing novel parallel implementations of low-scaling, high-accuracy methods. In particular, new techniques for mean-field calculations will be developed, which will act as groundwork for periodic coupled-cluster and quantum Monte Carlo methods. State-of-the-art techniques in sparsity and tensor decomposition will be employed to achieve good system-size scaling while retaining accuracy within each of these numerical schemes. Critically, the methods will be developed using efficient high-level software abstractions, implemented as Python-level modules within PySCF that leverage the Cyclops library for massively-parallel execution. The library software infrastructure will also be extended to maximize productivity via source-to-source automatic differentiation, as well as to enable execution of sparse kernels on emerging GPU-based supercomputing architectures. This award is jointly supported by the NSF Office of Advanced Cyberinfrastructure, and the Division of Materials Research and the Division of Chemistry within the NSF Directorate of Mathematical and Physical Sciences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.