Data-driven relationship of atomic structure and physical properties as the holistic view on the materials science fundamentals

TL;DR

This paper introduces a holistic, data-driven approach to understanding the relationship between atomic structure and physical properties of inorganic materials, integrating experimental, computational, and machine learning data.

Contribution

It presents a novel method combining experimental database, machine learning, and DFT simulations to analyze atomic structure-property relationships in inorganic substances.

Findings

Integrated database and computational methods enhance understanding of structure-property relationships.

Machine learning and DFT data complement experimental data for comprehensive analysis.

The approach offers a unified view of materials science fundamentals.

Abstract

The fundamental relationship of the atomic structure (represented by its atomic property parameters, APPs) and its physical properties of a specific inorganic substance can be realized in the bottom-up data-centric and the top-down knowledge physics-centric ways. Nowadays these two approaches compete and enhance one another qualitatively and quantitatively. We present our own holistic method and implementation, based on the PAULING FILE peer-reviewed inorganic substances database, the world largest materials database containing under one shelter crystallographic structures, phase diagrams and large variety of physical properties of single-phase inorganic substances. In addition we present generated machine-learning data, as well as simulated DFT physics-centered data, which are in close connection and comparison with the PAULING FILE peer-reviewed reference data.