Novel methods for high-throughput density-functional theory using DFTK
Kohn-Sham density-functional theory (DFT) simulations have become one of the standard computational approaches to predict physical or chemical properties of molecules and solids. A recent focus in this field has been the development of specialised frameworks for high-throughput calculations where the aim is to systematically compute the properties of a large number of systems. For application domains such as catalysis, battery research or crystal structure prediction, where experiments are expensive or time-consuming, such large-scale screening studies offer to automatically boil down a large list of candidate compounds to a tractable set for more detailed follow-up investigation.
From a theoretical point of view the high-throughput regime also poses novel challenges. A particular issue is that the accuracy, runtime and robustness of DFT methods is influenced by a sizeable number of parameters, e.g. basis size, convergence thresholds, DFT functional, and so on. Selecting all these parameters manually for each calculation is no longer practical in such a screening context. The state-of-the-art approach to tackle this is to rely on elaborate heuristics for selecting such parameters based on prior experience. Unfortunately this bears the risk to be either too conservative, causing non-optimal runtime, or not conservative enough, risking frequent breakdowns of the simulation procedure (and the associated manual intervention).
Convergence of a selection of self-consistent field methods for an elongated lattice of aluminium (grey) and silica (red/yellow), shown at the top. The LDOS mixing scheme we recently proposed is able to locally adapt to the material (metal or insulator) and as a result shows a much faster convergence compared to the other approaches. At the same time it is completely parameter-free. |
In this project I develop novel DFT approaches where such empirical heuristics are either replaced by mathematical rigour or where the number of parameters is inherently reduced by employing physically sound models. For this purpose I collaborate with mathematicians, namely Antoine Levitt and Eric Cancès from the Matherials team at CERMICS, École des Ponts and Inria Paris. With Antoine Levitt I am also co-developing the density-functional toolkit (DFTK), a code for DFT simulations in solids, which offers an extremely accessible code base (around 5000 lines), but at the same time delivers a performance comparable to established DFT packages. It has been designed for multidisciplinary collaboration like the ones required in this project and solves the problem to bundle the research efforts of multiple communities in a single software platform, such that both fundamental research on novel models and algorithms as well as production calculations targeting applications can be performed in the same code. For suitable applications I collaborate with Venkat Viswanathan and coworkers from the ACED Project (involving Carnegie Mellon University, Citrine Informatics, Julia computing and MIT). The idea is to employ DFTK as one backend in the developed pipeline to accelerate the discovery of novel compounds for electrochemical systems and catalysis.
Highlighted publications
- Michael F. Herbst and Antoine Levitt. Black-box inhomogeneous preconditioning for self-consistent field iterations in density-functional theory. Journal of Physics: Condensed Matter (2020). DOI 10.1088/1361-648X/abcbdb [code] Blog article.
- Michael F. Herbst, Antoine Levitt and Eric Cancès. A posteriori error estimation for the non-self-consistent Kohn-Sham equations. Faraday Discussions (2020). DOI 10.1039/D0FD00048E [code] Blog article.
- Related blog articles