The landscape is rapidly changing in the world of electronic structure calculations. When I started in the area, even the optimization of a low symmetry crystal structure with a first-principles approach was a significant computational challenge. Now with computing cores become cheaper and more abundant, density functional theory calculations, especially with local or semi-local functionals, are becoming routine and can be tackled even with a (powerful) desktop processor such as the Intel i7. So what should we do with our high-performance computers?
This week there were two good examples of combinatorial computations, each different in their approach.
"New Cubic Perovskites for Single- and Two-Photon Water Splitting using the Computational Materials Repository" led by Karsten Jacobson
Theory: RPBE (structures); GLLB-SC (band gaps).
Concept: Take a single structure type and substitute in 19000 elemental combinations; screen for properties related to photoelectrochemistry.
Result: 20 candidate materials.
Comment: An effective brute force approach, with the main limitation being the assumption of a single crystal structure (perovskite). This work is part of the Computational Materials Repository project.
"Prediction of A2BX4 metal-chalcogenide compounds via first-principles thermodynamics" led by Alex Zunger
Theory: PBE / PBE+U.
Concept: Take a single material stoichiometry (A2BX4) and investigate 429 unreported materials in 40 structure types; screen for thermodynamic stability / accessibility.
Result: 100 new and theoretically stable materials.
Comment: Technically this is the more creative approach as the crystal structure is not constrained. In addition to 40 known structure types, a global structure optimization method is used to assess some compounds, and a range of magnetic configurations are also included for transition metals. As a result, the computational cost quickly elevates from 429 material systems to > 70000 calculations. This work is part of the Center for Inverse Design.
These studies form part of a wider trend, with a number of codes and databases appearing over the past few years:
CALYPSO (Global structure optimization)
USPEX (Global structure optimization)
XTALOPT (Global structure optimization)
CompES (Database)
Materials Project (Database)
It was quickly realised after the initial hype for experimental combinatorial chemistry that it is not a cure-all approach, but as we have access to supercomputers with 100000s of processing cores, combinatorial computational chemistry is becoming an increasingly powerful tool.