Research
In our group, we develop the physical algorithm and numerical methods for the scientific calculations as well as accelerating these applications in the highperformance computers. If you find this interesting and exciting, you can find our articles on Google Scholar profile. Topics we are currently working on include:
1. All-electron density functional perturbation theory
We have proposed a computational framework for density functional perturbation theory with all-electron, full-potential accuracy. Pioneered the development and implementation of computational methods simultaneously adaptable to molecules and solids, as well as to external perturbations such as atomic displacements and electric fields. Achieved billion-core scalable all-electron, full-potential first-principles calculations on China’s exascale heterogeneous many-core supercomputers.
Shang#* et al, Comput. Phys. Commun. 215，26 (2017)
Shang# et al, New J. Phys. 20, 073040(2018)
Shang#* et al, SC21 (Gordon Bell Prize Finalist)
Shang#* et al, Comput. Phys. Commun. 258,107613, (2021) Shang* et al, SC23
2. QiankunNet: Neural network quantum state (NNQS) method for quantum chemistry
The fundamental many-electron Schrodinger equation is solved straightforwardly with QiankunNet, a neural network quantum state (NNQS) framework based on generative Transformer architecture along with a batched autoregressive sampling method tailored for this Transformer-based ansatz in quantum chemistry calculations. This approach significantly improves the accuracy and efficiency of first-principles calculations compared to previous fermionic ansatz methods. QiankunNet showcases the power of the Transformer-based language model in achieving unprecedented efficiency in quantum chemistry calculations, opening up new avenues for chemical discovery and demonstrating the potential to solve the large-scale Schrodinger equation with modest computational cost.
Shang# et al, https://arxiv.org/abs/2307.09343
Shang* et al, SC’23 https://dl.acm.org/doi/10.1145/3581784.3607061
Shang* et al, J. Chem. Theory Comput. 20, 14, 6218(2024）