DECODENSE: Mean-Field Density Matrix Decompositions
Name of applicant
Janus Juul Eriksen
Institution
University of Bristol
Amount
DKK 1,739,366
Year
2020
Type of grant
Reintegration Fellowships
What?
The main objective of the DECODENSE project is the development of new, rapid simulation tools in the field of quantum chemistry. In particular, I will be concerned with the acceleration of contemporary, state-of-the-art electronic structure methods by means of modern machine learning. Through the formulation of novel theory and its efficient implementation within optimized computational workflows, I am proposing new manners by which to decompose quantum-chemical simulations. In turn, these decompositions will make the underlying theory increasingly befitting to the deployment of machine learning in inferring physical properties from chemical structures alone. Importantly, these mappings will be made without recourse to explicit and exceedingly expensive electronic structure treatments.
Why?
The inner workings of molecules and materials may these days be probed and modelled by advanced simulation tools on modern computer architectures. These simulations not only serve as a complement to traditional empirical explorations, but also as practical extensions in cases where these prove infeasible, hazardous, or exceedingly expensive. The crux of the matter, however, is the fact that today's primary computational workhorse - known as Kohn-Sham density functional theory (KS-DFT) - is inherently too computationally costly to warrant large-scale simulations of, e.g., potential drug prospects or emerging solid-state batteries. To that end, my project will ultimately be concerned with circumventing the prohibitive scaling wall of KS-DFT by deploying machine learning to the problem.
How?
A Carlsberg Foundation Reintegration Fellowship will enable me to return home to Denmark after two successful postdoctoral stays abroad. The fellowship will see me integrate into one of our country's most vibrant and interdisciplinary academic environments at DTU Chemistry, which is certain to offer me both the necessary support and the scientific independence to drive my ongoing endeavours towards the development of more accurate takes on machine-learned quantum chemistry to new levels. With a sound core of theory development and high-performance computing in roughly equal parts, my project is bound to offer a manifold of physical and chemical insight as well as a firm basis for forming my own research lab at DTU in the years to come.