Electronic configuration vectors provide transferable elemental representations for machine learning across the periodic table
Abstract
Machine learning models typically represent elements as discrete identities, limiting inference for elements absent from training. We introduce the electronic configuration vector (ECV), a physically grounded representation derived from the Aufbau principle that continuously encodes orbital occupancies and energetics across elements. Principal component analysis reveals low-dimensional manifolds that organize elements by atomic number, with extrema corresponding to block transitions. Across adsorption energy, formation energy, and metallicity prediction, ECV yields accurate, lightweight, interpretable models while revealing chemical structure. ECV marginally outperforms the Darwin LLM in metallicity classification on the Matbench benchmark (ROC-AUC 0.9614 vs. 0.9598), despite far lower complexity. It also generalizes to unseen elements, achieving 0.16 eV MAE for perovskite formation energies, compared with 0.39 eV for one-hot encoding and 0.18 eV for elemental property vectors. Unlike one-hot encoding, ECV does not impose elemental orthogonality and, unlike property vectors, is definable for every element. These results establish ECV as a coordinate system for transferable machine learning across the periodic table.
Related
- Learning Electron Density with Virtual Atoms for Transferable Atomistic Representations
- Hybridizing expert descriptors and learned embeddings yields transferable Mat2Vec-* element representations
- Parity-equivariant prediction of conformer-resolved electronic circular dichroism spectra accelerates absolute configuration assignment
- Response-state Learning for Transferable Vibrational Spectroscopic Characterization with Electron Prior
- Development of a Deep Reinforcement Learning Approach for Transition State Search in Molecular Systems: A Transferable Model Across Structural and Dimensional Variations
BibTeX
@misc{rajagopalan2026electronic,
title = {{Electronic configuration vectors provide transferable elemental representations for machine learning across the periodic table}},
author = {Raghavendra Rajagopalan and Ananth Govind Rajan},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008244/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008244/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Electronic configuration vectors provide transferable elemental representations for machine learning across the periodic table
AU - Raghavendra Rajagopalan
AU - Ananth Govind Rajan
PY - 2026
DA - 2026/09/03
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008244/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008244/v1
AB - Machine learning models typically represent elements as discrete identities, limiting inference for elements absent from training. We introduce the electronic configuration vector (ECV), a physically grounded representation derived from the Aufbau principle that continuously encodes orbital occupancies and energetics across elements. Principal component analysis reveals low-dimensional manifolds that organize elements by atomic number, with extrema corresponding to block transitions. Across adsorption energy, formation energy, and metallicity prediction, ECV yields accurate, lightweight, interpretable models while revealing chemical structure. ECV marginally outperforms the Darwin LLM in metallicity classification on the Matbench benchmark (ROC-AUC 0.9614 vs. 0.9598), despite far lower complexity. It also generalizes to unseen elements, achieving 0.16 eV MAE for perovskite formation energies, compared with 0.39 eV for one-hot encoding and 0.18 eV for elemental property vectors. Unlike one-hot encoding, ECV does not impose elemental orthogonality and, unlike property vectors, is definable for every element. These results establish ECV as a coordinate system for transferable machine learning across the periodic table.
ER -