Pristine-State Predictors and Defect Response in Oxygen Vacancy Formation on Compositionally Modified LiCoO₂(001) Surfaces

Surfaces mediate many important chemical transformations, yet their local structure and electronic state can evolve substantially during reaction, making it difficult to connect the properties of an intact surface to the energetics of its subsequent transformation. Here, we use oxygen vacancy formation in complex metal oxides as a model defect process to ask how much of this energetic response is encoded in the pristine surface before oxygen removal. Density functional theory (DFT) calculations within a consistent GGA-PBE baseline are combined with descriptor-based machine learning across a systematically constructed dataset of 531 oxygen-stoichiometric, compositionally modified LiCoO₂(001) surfaces. Using descriptors calculated exclusively from the oxygen-stoichiometric structures, a random-forest model achieves a repeated group-aware out-of-fold mean absolute error of 0.188 eV for the corresponding DFT oxygen vacancy formation energies. Controlled representation and transfer tests show that the predictive information is largely captured by direct descriptors of the pristine local chemical and electronic environment, with limited additional information gained from either a computed slab formation energy or reference binary oxide defect values and reduced transferability to chemically distinct local environments. Analysis of feature recurrence, held-out permutation importance, and accumulated local effects consistently identifies the charge of the oxygen selected for removal and descriptors of the local metal oxygen electronic environment as important predictors of vacancy formation energy. Transfer is less reliable for chemically distinctive local-composition families, establishing an important domain-of-applicability limitation. Explicit post-vacancy electronic structure analysis reveals composition-dependent redistribution of occupied projected electronic states, but this response does not map uniquely onto vacancy formation energy, and compact response descriptors do not provide a statistically resolved general correction to the residual prediction error within the 48-structure subset examined. Together, these results show that pristine-state descriptors encode substantial information about oxygen-vacancy energetics while also identifying where that representation becomes insufficient, establishing predictive relationships and domain-of-applicability limits.

Authors

Institutions

Publication Details

Journal
Electronic Structure
Published
2026-11-19
DOI
https://doi.org/10.1088/2516-1075/aec4ae
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

Pristine-State Predictors and Defect Response in Oxygen Vacancy Formation on Compositionally Modified LiCoO₂(001) Surfaces

Amy C. Marschilok, Jennifer L. Bjorklund, Sara Elizabeth Mason, Esther S. Takeuchi et al.
Electronic Structure
Machine Learning in Materials Science
article

Pristine-State Predictors and Defect Response in Oxygen Vacancy Formation on Compositionally Modified LiCoO₂(001) Surfaces

Amy C. Marschilok, Jennifer L. Bjorklund, Sara Elizabeth Mason, Esther S. Takeuchi, Kailey Elizabeth Holen, Dorothea Fennell, Xiaohui Qu, Natalia C. Mena‐Santiago, Kenneth Takeuchi
article en

Abstract

Surfaces mediate many important chemical transformations, yet their local structure and electronic state can evolve substantially during reaction, making it difficult to connect the properties of an intact surface to the energetics of its subsequent transformation. Here, we use oxygen vacancy formation in complex metal oxides as a model defect process to ask how much of this energetic response is encoded in the pristine surface before oxygen removal. Density functional theory (DFT) calculations within a consistent GGA-PBE baseline are combined with descriptor-based machine learning across a systematically constructed dataset of 531 oxygen-stoichiometric, compositionally modified LiCoO₂(001) surfaces. Using descriptors calculated exclusively from the oxygen-stoichiometric structures, a random-forest model achieves a repeated group-aware out-of-fold mean absolute error of 0.188 eV for the corresponding DFT oxygen vacancy formation energies. Controlled representation and transfer tests show that the predictive information is largely captured by direct descriptors of the pristine local chemical and electronic environment, with limited additional information gained from either a computed slab formation energy or reference binary oxide defect values and reduced transferability to chemically distinct local environments. Analysis of feature recurrence, held-out permutation importance, and accumulated local effects consistently identifies the charge of the oxygen selected for removal and descriptors of the local metal oxygen electronic environment as important predictors of vacancy formation energy. Transfer is less reliable for chemically distinctive local-composition families, establishing an important domain-of-applicability limitation. Explicit post-vacancy electronic structure analysis reveals composition-dependent redistribution of occupied projected electronic states, but this response does not map uniquely onto vacancy formation energy, and compact response descriptors do not provide a statistically resolved general correction to the residual prediction error within the 48-structure subset examined. Together, these results show that pristine-state descriptors encode substantial information about oxygen-vacancy energetics while also identifying where that representation becomes insufficient, establishing predictive relationships and domain-of-applicability limits.

Electronic Structure
University of Puerto Rico at Cayey (PR), The University of Texas at El Paso (US), Brookhaven National Laboratory (US), Stony Brook University (US)
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