QMC-HAMM uses Aurora to study the relationship between the very microscopic and macroscopic. Pictured here is the transition of hydrogen from a liquid to a solid at very high pressures relevant in planet interiors and fusion experiments. We precisely determine the exact conditions of the transition by simulating quantum correlations between electrons. Credit: Isabel Zhou and Shubhang Goswami, University of Illinois Urbana-Champaign
In this project, the team will develop community software that integrates high-accuracy quantum Monte Carlo calculations with multiscale and machine-learning models to surpass density functional theory in predictive power. This will target correlated materials problems including high-pressure hydrogen, advanced battery materials, and charge density waves, while generating benchmark-quality data for the broader materials community.
There has been a sea change in materials modeling as machine learning models have come to dominate the description of materials. These models have benefited from the easy availability of data from first principles density functional theory calculations, which approximately describe the quantum mechanical properties of fundamental electrons and nuclei. Advances in machine learning have resulted in models that are near density functional theory in accuracy, resulting in a situation where the underlying data itself is the limiting factor in predictivity. This project centers around the development of models which use high accuracy data from quantum Monte Carlo to further enhance the ability to predict the properties of materials. We call the project "High Accuracy Multiscale Models using Quantum Monte Carlo" (QMC-HAMM).
The QMC-HAMM project is devoted to creating a set of community-serving software tools that enable one to link highly accurate many- electron microscopic quantum simulations with multiscale modeling that can achieve large length and time scales. A key distinction of our work from most of the field is that electron correlation plays a critical role in the materials physics but is a crucial weakness for widely-used computational tools such as density functional theory. By starting with higher accuracy microscopic calculations and using modern methods of computing coarse-grained models, our platform will enable higher accuracy large- scale models than the standard techniques based on DFT. It is critical that we address a diverse set of problems to ensure our methodology is transferable.
With the objective of diversity in mind, we consider three application areas, chosen for impact and for fundamental interest. The first is hydrogen at high pressure, which is relevant for fusion science and planetary interiors, and has an intricate phase diagram which is reminiscent of other materials. The second is the development of machine learning interatomic potentials for several materials, including lithium phosphates and two dimensional materials. These materials are relevant for new battery and other electronic applications. Finally, we will address a fundamental question in the physics of materials, examining the fundamental mechanisms of charge density waves, a phenomenon in which a material breaks symmetry spontaneously, often suddenly changing its properties significantly.
At a broader scale, this proposal will enable cutting edge methodology that moves beyond the state of the art in materials modeling. As a result of this work, high accuracy models will be produced, and improved practices will be developed. Each computation campaign will produce high accuracy data on the selected materials, which will be a valuable resource for other materials modelers; such data is scarce at best. Our collaboration has a track record of producing both high accuracy data and models, in particular for bilayer graphene and hydrogen, as well as excited states in color centers.
