论文Publications

Accelerating Atom Simulations with Variable-Block Sparse Matrix Library

A distributed variable-block sparse matrix library for atomistic simulations, evaluated on block-sparse benchmarks and a million-atom-scale InP nanoparticle.

Moiré Phonon Condensation in Magic-Angle Twisted Bilayer Graphene

A softening layer-symmetric moiré phonon mode explains a structural reconstruction near the magic angle. Two A1 modes capture over 99.5% of the displacement ...

Deep neural network for phonon-assisted optical spectra of semiconductors at finite temperatures

A DeePMD plus DeePTB workflow transfers HSE-level electronic information to large supercells to predict finite-temperature phonon-assisted optical spectra.

Neural-quantum-states impurity solver for quantum embedding problems

A graph-transformer neural-quantum-states impurity solver for the ghost Gutzwiller quantum embedding framework, benchmarked against exact diagonalization on ...

Deep learning accelerated quantum transport simulations in nanoelectronics: from break junctions to field-effect transistors

DeePTB-NEGF connects machine-learned Hamiltonians and quantum transport for over 10,000 break-junction configurations and carbon-nanotube field-effect transi...

Deep Learning-Based Quantum Transport Simulations in Two-Dimensional Materials

DeePTB-NEGF couples learned tight-binding Hamiltonians with nonequilibrium Green's function calculations for graphene, h-BN and MoS2.

Understanding neural networks with logarithm determinant entropy estimator

A matrix-based logarithm-determinant entropy estimator for high-dimensional information measurements and neural-network analysis.

Pitfalls of exchange–correlation functionals in description of magnetism: Cautionary tale of the FeRh alloy

A comparison of exchange–correlation approximations in density-functional calculations of FeRh, illustrating the sensitivity of magnetic predictions to latti...

Deep learning tight-binding approach for large-scale electronic simulations at finite temperatures with ab initio accuracy

DeePTB learns environment-dependent tight-binding Hamiltonians and enables finite-temperature electronic-property calculations in million-atom GaP systems.

Learning local equivariant representations for quantum operators

SLEM: a strictly-localized equivariant neural network that predicts DFT quantum operators (Hamiltonian, overlap and density matrices) with high accuracy, str...

Automatic differentiable nonequilibrium Green's function formalism: An end-to-end differentiable quantum transport simulator

An end-to-end differentiable Non-equilibrium Green's Function quantum transport simulator implemented in PyTorch, providing gradients for sensitivity analysi...