Band Gaps of Halide Perovskites from an Orbital Cluster Expansion: Mechanism, Iodide Bias, and a Hybridization Channel

Linear descriptors are cheap, interpretable, and trainable on hundreds of labels. What they lack is an account of why they work and where they break. We address both for the band gap of halide perovskites. We label 1784 CsPbX3-type supercells, 1241 of which are distinct under symmetry, with scalar-relativistic density-functional theory and fit an orbital cluster expansion: a linear model over counted atomic figures and distance-binned electrostatics. Under cross-validation grouped by the A/B-cation family, with the ridge penalty selected on a grouped split too, it reaches 0.151 eV (95% interval 0.140–0.160), 0.20 of the target standard deviation, and a 29% error reduction relative to Magpie. The path there is nonmonotonic: bonded angular figures make the fit worse than the composition, and only the distance-binned electrostatic channel recovers it─an ionic perovskite’s gap responds to the electrostatic environment, not to the bonded geometry. Densities of states identify the mechanism: the halide dominates the gap (effect size η2 = 0.386) and the B cation modulates it; the scalar-relativistic Pb/Ge gap ratio jumps from 1.26× in the fluoride to about 5× in the three heavier halides. The residual is biased in one halide only: iodide gaps are underestimated by a median of 0.085 eV, while the other halides are centered within 0.03 eV. A second-order hybridization channel removes that bias and improves the global error by 6.1%, significant with the families fixed but not when they are resampled; per-class regularization, the principled remedy, degrades the fit. Symmetry-equivalent copies in the labeled set, identified during revision, are excluded throughout.

Authors

Institutions

Publication Details

Journal
The Journal of Physical Chemistry C
Published
2026-11-21
DOI
https://doi.org/10.1021/acs.jpcc.6c05221
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Band Gaps of Halide Perovskites from an Orbital Cluster Expansion: Mechanism, Iodide Bias, and a Hybridization Channel

David L. Azevedo
The Journal of Physical Chemistry C
Machine Learning in Materials Science
article

Band Gaps of Halide Perovskites from an Orbital Cluster Expansion: Mechanism, Iodide Bias, and a Hybridization Channel

David L. Azevedo
article en

Abstract

Linear descriptors are cheap, interpretable, and trainable on hundreds of labels. What they lack is an account of why they work and where they break. We address both for the band gap of halide perovskites. We label 1784 CsPbX3-type supercells, 1241 of which are distinct under symmetry, with scalar-relativistic density-functional theory and fit an orbital cluster expansion: a linear model over counted atomic figures and distance-binned electrostatics. Under cross-validation grouped by the A/B-cation family, with the ridge penalty selected on a grouped split too, it reaches 0.151 eV (95% interval 0.140–0.160), 0.20 of the target standard deviation, and a 29% error reduction relative to Magpie. The path there is nonmonotonic: bonded angular figures make the fit worse than the composition, and only the distance-binned electrostatic channel recovers it─an ionic perovskite’s gap responds to the electrostatic environment, not to the bonded geometry. Densities of states identify the mechanism: the halide dominates the gap (effect size η2 = 0.386) and the B cation modulates it; the scalar-relativistic Pb/Ge gap ratio jumps from 1.26× in the fluoride to about 5× in the three heavier halides. The residual is biased in one halide only: iodide gaps are underestimated by a median of 0.085 eV, while the other halides are centered within 0.03 eV. A second-order hybridization channel removes that bias and improves the global error by 6.1%, significant with the families fixed but not when they are resampled; per-class regularization, the principled remedy, degrades the fit. Symmetry-equivalent copies in the labeled set, identified during revision, are excluded throughout.

The Journal of Physical Chemistry C
Universidade de Brasília (BR)
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Band Gaps of Halide Perovskites from an Orbital Cluster Expansion: Mechanism, Iodide Bias, and a Hybridization Channel — David L. Azevedo · The Journal of Physical Chemistry C (2026) | TGRS Research Map | TGRS