Solving density-of-states occupation in disordered systems using neural networks
6 affiliations
- 1Centre Énergie Matériaux Télécommunications
- 2Department of Physics
- 3Concordia University
- 4Department of Chemistry and Biochemistry
- 5Centre for Research in Molecular Modeling (CERMM
- 6Centre for NanoScience Research (CeNSR
Abstract
Describing density-of-states occupation with high precision in disordered systems is crucial for modeling charge transport and understanding doping mechanisms or interfacial energy-level alignment. Beyond fundamental insight from a mechanistic understanding of these physical phenomena, accurate modeling is pivotal for optimizing technological applications in optoelectronic and thermoelectric devices. However, the required calculations are computationally demanding and inherently complex, preventing systematic computational-experimental approaches. Here, we leverage neural networks to perform these calculations precisely and efficiently. We first demonstrate this approach for the Fermi-Dirac occupation of Gaussian densities of states, achieving accelerations of up to three orders of magnitude, a substantial advantage for advanced modeling workflows that require repeated evaluation of the underlying integrals. We then introduce a neural network for doped molecular materials, such as organic semiconductors, that directly predicts doping levels and Fermi level shifts, circumventing conventional recursive schemes and accelerating computations by nearly five orders of magnitude.
Related
- Electronic Properties of Ladder-Type Phenylenes Studied by Thermally-Assisted-Occupation Density Functional Theory
- Revisiting Percolation and Molecular Diffusion in Disordered Pore Networks
- Ordered-to-disordered transfer learning with graph neural networks for formation-energy and HOMO-LUMO gap prediction in high-entropy perovskite oxides
- Scalable Density-Based Bootstrap Embedding with Perturbative Triples for Large Molecular Systems
- Altering Moisture Transport Mechanisms in Multilayer Material Systems via Open Channel Networks for Enhanced Comfort
BibTeX
@misc{berteaurainville2026solving,
title = {{Solving density-of-states occupation in disordered systems using neural networks}},
author = {Melissa Berteau-Rainville and Emanuele Orgiu and Ingo Salzmann},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15009841/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009841/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Solving density-of-states occupation in disordered systems using neural networks
AU - Melissa Berteau-Rainville
AU - Emanuele Orgiu
AU - Ingo Salzmann
PY - 2026
DA - 2026/10/05
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009841/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009841/v1
AB - Describing density-of-states occupation with high precision in disordered systems is crucial for modeling charge transport and understanding doping mechanisms or interfacial energy-level alignment. Beyond fundamental insight from a mechanistic understanding of these physical phenomena, accurate modeling is pivotal for optimizing technological applications in optoelectronic and thermoelectric devices. However, the required calculations are computationally demanding and inherently complex, preventing systematic computational-experimental approaches. Here, we leverage neural networks to perform these calculations precisely and efficiently. We first demonstrate this approach for the Fermi-Dirac occupation of Gaussian densities of states, achieving accelerations of up to three orders of magnitude, a substantial advantage for advanced modeling workflows that require repeated evaluation of the underlying integrals. We then introduce a neural network for doped molecular materials, such as organic semiconductors, that directly predicts doping levels and Fermi level shifts, circumventing conventional recursive schemes and accelerating computations by nearly five orders of magnitude.
ER -