Uncovering structure-property relationships of materials by subgroup discovery

Abstract

Subgroup discovery (SGD) is presented here as a data-mining approach to help find interpretable local patterns, correlations, and descriptors of a target property in materials-science data. Specifically, we will be concerned with data generated by density-functional theory calculations. At first, we demonstrate that SGD can identify physically meaningful models that classify the crystal structures of 82 octet binary (OB) semiconductors as either rocksalt or zincblende. SGD identifies an interpretable two-dimensional model derived from only the atomic radii of valence s and p orbitals that properly classifies the crystal structures for 79 of the 82 OB semiconductors. The SGD framework is subsequently applied to 24 400 configurations of neutral gas-phase gold clusters with 5-14 atoms to discern general patterns between geometrical and physicochemical properties. For example, SGD helps find that van der Waals interactions within gold clusters are linearly correlated with their radius of gyration and are weaker for planar clusters than for nonplanar clusters. Also, a descriptor that predicts a local linear correlation between the chemical hardness and the cluster isomer stability is found for the even-sized gold clusters.

From DFT to machine learning: recent approaches to materials science–a review

“…These include the Δ-approach to ML, that in order to increase prediction accuracy, it uses as the learning target the difference between a lower-quality model to the property of interest [448]. Another technique is the subgroup discovery, which finds local structure in data, as opposed to mapping a unique global relationship [449]. And the recent multi-fidelity learning which aims to be applied to small datasets, where in order to enhance the sampling and therefore learning capacity, one can combine lower precision data to overcome the scarcity of higher precision data [450].…”

The NOMAD laboratory: from data sharing to artificial intelligence

“…They concern topics like crystal-structure prediction, property prediction, error estimates, classification of materials, and more. Some of these notebooks correspond to peer-reviewed publications [23][24][25][26][27][28][29][30][31][32][33][34].…”

Telling Cause from Effect Using MDL-Based Local and Global Regression

“…The target of interest is the energy difference δ E between rocksalt and zincblende crystal structures. It is still an open challenge to find that combination of physical properties that can fully explain δ E [5], [6]. For causal inference, it suffices to know that the energy difference is influenced by the descriptor variables and not the other way round.…”