Knowledge Discovery from Atomic Structures using Feature Importances

Joakim LinjaJoonas H\"am\"al\"ainenAntti Pihlajam\"akiPaavo NieminenSami MalolaHannu H\"akkinenTommi K\"arkk\"ainen

Joakim LinjaJoonas H\"am\"al\"ainenAntti Pihlajam\"aki ...+3 Tommi K\"arkk\"ainen

Feb 2023

摘要

Molecular-level understanding of the interactions between the constituents ofan atomic structure is essential for designing novel materials in variousapplications. This need goes beyond the basic knowledge of the number and typesof atoms, their chemical composition, and the character of the chemicalinteractions. The bigger picture takes place on the quantum level which can beaddressed by using the Density-functional theory (DFT). Use of DFT, however, isa computationally taxing process, and its results do not readily provide easilyinterpretable insight into the atomic interactions which would be usefulinformation in material design. An alternative way to address atomicinteractions is to use an interpretable machine learning approach, where apredictive DFT surrogate is constructed and analyzed. The purpose of this paperis to propose such a procedure using a modification of the recently publishedinterpretable distance-based regression method. Our tests with a representativebenchmark set of molecules and a complex hybrid nanoparticle confirm theviability and usefulness of the proposed approach.

展开全部

图表提取