Aiming for Chemical Accuracy in Ground-state Density Functional Theory

在化学系化学理论、模型和计算方法项目的支持下，加州大学欧文分校的Kieron Burke教授将努力改进基态密度泛函理论（DFT）。每年，有超过5万篇科学论文使用DFT来预测分子和固体的性质，并设计新的药物和材料。美国大约三分之一的超级计算机用于这项任务。但由于近似的性能较差，这种计算受到限制。这项工作的目的是开发一种新的数学框架，有望在DFT计算的准确性和速度方面取得突破。在此建议的范围内，DFT计算速度或准确性的改进将改变大多数DFT应用程序的性质和功能。为了促进教育和推广，参与该项目的研究生和本科生将在一个多元化和跨国的研究小组中接受培训，以获得DFT的高水平知识。一门由Burke小组开发的物理科学机器学习自学在线课程将向公众开放，该课程已经向加州大学欧文分校的所有学生和教师开放。该建议结合了半经典方法（h的幂展开，普朗克常数），用于分离能量误差和密度驱动误差的密度校正DFT，以及用于从数据中寻找近似泛函的机器学习。目的是从原理证明到可以应用于日常计算的近似密度函数。如果获得成功，在该奖项下进行的研究有可能(a)将所有DFT计算的基本起点（广义梯度近似）转化为一种新的形式；（二）将材料中的官能团与化学中的官能团统一起来；(c)为平衡状态下的强键提供化学精度，(d)应用于非相互作用动能，从而影响无轨道DFT。简而言之，这里的成功可能会对目前使用DFT计算的所有领域做出贡献，因此，有可能实现深远而广泛的科学影响。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。

With support from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry, Professor Kieron Burke of the University of California, Irvine will endeavor to improve ground state density functional theory (DFT). Each year, more than 50,000 scientific papers use DFT to predict the properties of molecules and solids, and to design new pharmaceutical drugs and materials. About 1/3 of US supercomputer use is devoted to this task. But such calculations are limited due to poor performance of approximations. The aim of this work is to develop a new mathematical framework that has the promise of breakthroughs in both accuracy and speed of DFT calculations. Improvements in speed or accuracy of DFT calculations on the scale of this proposal would transform both the nature and capabilities of most DFT applications. Contrbuting to education and outreach, graduate and undergraduate students involved in this project will be trained to acquire high level knowledge of DFT, in a diverse and multi-national research group. A self-guided online course in Machine Learning in the Physical Sciences, developed by the Burke group and already available to all UC-Irvine students and faculty, will be made available to the public.This proposal combines semiclassical methods (expansions in powers of h, Planck's constant), density­-corrected DFT for separating energy errors from density-driven errors, and machine learning for finding approximate functionals from data. The aim is to go from proof-of-principle to approximate density functionals that can be applied to routine calculations. If successful, the research being undertaken under this award has the potential (a) to transform the basic starting point (generalized gradient approximations) of all DFT calculations into a new form; (b) to unite the functionals used in materials with those in chemistry; (c) to produce chemical accuracy for strong bonds at equilibrium, and (d) to apply to the non-interacting kinetic energy, and so impact orbital-free DFT. In short, success here would likely contribute to all fields currently using DFT calculations, and thus, the potential to achieve far-reaching broader scientific impact.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.