Amit Chaudhari

Amit Chaudhari

Researcher in machine learning for electronic structure and materials modelling

Posts by Collection

portfolio

publications

Polymorph-Induced Reducibility and Electron Trapping Energetics of Nb and W Dopants in TiO2

Published in The Journal of Physical Chemistry C, 2025

We investigate the atomic-level origin of the experimentally observed formation of paramagnetic Nb4+ and W5+ species in Nb- and W-doped rutile TiO2 vs. non-magnetic Nb5+ and W6+ species in doped anatase TiO2. This is achieved using a combination of theory (Hubbard corrected density functional theory) and experiment (powder electron paramagnetic resonance spectroscopy).

Citation: A. Chaudhari, A. J. Logsdail and A. Folli, Polymorph-Induced Reducibility and Electron Trapping Energetics of Nb and W Dopants in TiO2, The Journal of Physical Chemistry C, 2025, 129 (34), 15453-15461, DOI: 10.1021/acs.jpcc.5c04364

Machine Learning Generalised DFT+U Projectors in a Numerical Atom-Centred Orbital Framework

Published in Digital Discovery, 2025

We present machine learning-based workflows using symbolic regression and support vector machines to simultaneously optimise Hubbard U values and projectors, enabling accurate and efficient simulations of defects and polarons in transition metal and rare-earth oxides.

Citation: A. Chaudhari, K. Agrawal and A. J. Logsdail, Machine learning generalised DFT+U projectors in a numerical atom-centred orbital framework, Digital Discovery, 2025, DOI: 10.1039/D5DD00292C

Mixture-of-Experts Transformers for Faithfully Deorbitalized Meta-GGA Density Functionals

Published in ChemRxiv, 2025

We adopt a physics-informed deep learning approach to reparameterise popular semi-local meta-GGA density functionals into non-local deorbitalized surrogates that more faithfully mimic their orbital-dependent parent functionals. This is achieved using a Mixture-of-Experts transformers architecture that uses multi-head self-attention and automatic differentiation to accurately predict exchange energy densities and partial derivatives across a diverse set of molecules and materials. The architecture is designed to replace the expensive feature-based non-locality from the orbital-dependent kinetic energy density with architectural non-locality implemented using attention.

Citation: A. Chaudhari and A. J. Logsdail, Mixture-of-Experts Transformers for Faithfully Deorbitalized Meta-GGA Density Functionals, ChemRxiv, 2025, DOI: 10.26434/chemrxiv-2025-mrgzj-v2

Ab Initio Insights into Support-Induced Sulfur Resistance of Ni-Based Reforming Catalysts

Published in Catalysis Science & Technology, 2026

We investigate the metal oxide support effects controlling the sulfur tolerance of Ni-based catalysts for methane steam reforming. This is achieved using a range of theoretical methods; including DFT-parameterised Monte Carlo sampling of a lattice model of S and O adsorption on Ni(111), adlayer validation with a fine-tuned machine learned interatomic potential and DFT+U simulations of the defect chemistry in common support materials. The predictions are used to rationalise the results of experimental characterisation and methane steam reforming activity testing for H2S-poisoned Ni nanoparticle catalysts.

Citation: A. Chaudhari, P. Stishenko, A. Hiregange, C. Hawkins, M. Sarwar, S. Poulston and A. J. Logsdail, Ab Initio Insights into Support-Induced Sulfur Resistance of Ni-Based Reforming Catalysts, Catalysis Science & Technology, 2026, DOI: 10.1039/D5CY01279A

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.