/HubbardNet

Deep neural network-based solution to the ground and excited states of 1D and 2D Bose-Hubbard model

Primary LanguageJupyter Notebook

HubbardNet: efficient predictions of the Bose-Hubbard Model Spectrum with Deep Neural Networks

All scripts have been tested on Google Colab (have options for both CPU/GPU). Can also be run locally.

Citation

For reference of the HubbardNet model, please see and cite the following manuscript:

"HubbardNet: Efficient Predictions of the Bose-Hubbard Model Spectrum with Deep Neural Networks." Ziyan Zhu, Marios Mattheakis, Weiwei Pan, Efthimios Kaxiras. ArXiv: 2212.13678 (2022)

Contact

Ziyan (Zoe) Zhu: ziyanzhu [at] stanford.edu

Please contact me with any issues and/or request.

List of files:

  • matrix_elements.py: functions to construct Bose-Hubbard model
  • HubbardNet_gpu.py: functions to construct the DNN and train the network

Examples:

  • iterative_opt_gpu_for_gs.ipynb: ground state optimization
  • iterative_opt_gpu_for_gs_mult_N.ipynb: ground state optimization with multiple N's in the training set
  • iterative_opt_gpu_spectrum.ipynb: iteratively obtain the full spectrum (all excited states) using multiple U's in the training set