Projects
Research code, open-source tools, and academic resources.
GradTDDFT
A JAX-based differentiable implementation of time-dependent density functional theory (TDDFT). Provides end-to-end differentiable SCF calculations, Casida equations, and neural network exchange-correlation functional training. Supports GPU acceleration via CUDA J/K integral experiments.
QHformer
A SO(3)-equivariant neural network for predicting quantum Hamiltonian matrices from molecular geometries. Combines hybrid attention mechanisms with equivariant architectures for accurate Hamiltonian prediction in quantum chemistry applications.
🧠 3D Model Architecture Graph
OPTXC
A machine learning toolkit for optimizing exchange-correlation functionals in linear-response TDDFT calculations. Supports multiple functionals (LC-wPBE, wB97XD) and optimization models (XGBoost, Random Forest, Gradient Boosting, CatBoost). Transforms SMILES strings into molecular descriptors for functional optimization.