Research

Computational chemistry for bonding, pressure, and materials discovery.

I study how bonding and electronic structure change under pressure and reduced dimensionality, and how machine learning can accelerate the search for new materials.

My work combines first-principles electronic-structure calculations, crystal-structure prediction, high-throughput workflows, and machine learning to understand materials where conventional chemical intuition becomes strained.

Electronic structure High pressure Structure discovery ML workflows

Scientific montage showing a compressed crystal, an electron-localization field, and a machine-learning workflow — From atomic structure to real-space electronic descriptors to materials discovery.

Atomic structure Template discovery DFT under pressure Hydrogen metallization Real-space fields ELF prediction Materials insight 2D electrides

Research Themes

Scientific questions connected by bonding and electronic structure.

Real-space electronic structure

Electron localization, ELF prediction, interstitial bonding, and field-based descriptors.

High-pressure materials

Superhydrides, molecular hydrogen, pressure-induced bonding, and superconductivity.

Data-driven discovery

Chemical templates, high-throughput DFT, ML screening, and structure prioritization.

Reduced-dimensional chemistry

2D electrides, TMDCs, vacancies, adsorption, and tunable interstitial electrons.

Featured Projects

Artifacts that show the research program forming.

Poster thumbnail for symmetry-aware prediction of electron localization functions

Real-space electronic structure

ELFNet and electronic-structure prediction

Real-space electronic-structure prediction from superposed atomic densities.

Scientific question: Can electron localization be screened before full electronic-structure post-processing?
Approach: Periodic 3D CNNs map superposed atomic density grids to DFT electron localization functions.
Main idea: Fast ELF prediction can expose bonding and interstitial localization early in screening.

PyTorch 3D CNN Periodic grids VASP ELF

Two-step chemical-template workflow for discovering metal superhydrides

Chemical-template discovery

Metal superhydrides from chemical templates

Template-guided machine learning for complex high-pressure hydrides.

Scientific question: Can chemical templates reduce the search space for complex high-pressure hydrides?
Approach: Machine-learning screening, convex-hull validation, and first-principles calculations.
Main idea: Localized interstitial electrons can organize metal sublattices into new hydride prototypes.

DFT Convex hulls Templates ML screening

Pressure-driven bonding

Molecular hydrogen metallization

Electronic delocalization in compressed molecular hydrogen phases.

Scientific question: How does molecular hydrogen move from an insulating molecular solid toward delocalization?
Approach: Electronic structure, band gaps, orbital character, and real-space localization analysis.
Main idea: Pressure can stretch H2 units and push charge into interstitial regions before full dissociation.

HSE06 Band structure ELF Pressure

Poster thumbnail for tunable electrides on monolayer transition-metal dichalcogenides

Reduced-dimensional chemistry

Electrides and interstitial electrons in 2D materials

Vacancy-tuned interstitial electrons in monolayer transition-metal dichalcogenides.

Scientific question: Can vacancies and reduced dimensionality stabilize chemically useful interstitial electrons?
Approach: DFT, ELF, vacancy engineering, and adsorption-energy trends across monolayer TMDCs.
Main idea: Localized interstitial electrons can tune surface reactivity and hydrogen-evolution trends.

TMDCs Vacancies ELF HER

Crystal-structure machine-learning workflow diagram with pretraining, validation, and sampling stages

Reproducible computation

High-throughput DFT and structure-search workflows

Scalable workflows for structure generation, relaxation, post-processing, and screening.

Scientific question: How can structure prediction and electronic-structure analysis be made reproducible at scale?
Approach: Python, Bash, Slurm, VASP, CALYPSO, and PyTorch workflows across local and HPC systems.
Main idea: Reliable automation turns many calculations into a traceable materials-discovery campaign.

Python Slurm VASP CALYPSO PyTorch

Methods and Tools

The computational stack behind the projects.

DFT VASP CALYPSO AIRSS ELF HSE06 r2SCAN PyTorch 3D CNNs Slurm Python High-throughput screening

Selected Research Artifacts

Where the details live.

Publications Journal and workshop papers Posters Conference research summaries Slides Talks and thesis deck Blog Short technical write-ups Code Public repositories