DeepQMC implements variational quantum Monte Carlo for electrons in molecules, using deep neural networks written in PyTorch as trial wave functions. Besides the core functionality, it contains implementations of the following ansatzes:
- PauliNet: https://doi.org/ghcm5p
- DeepErwin: http://arxiv.org/abs/2105.08351
Install and update using Pip.
pip install -U deepqmc[wf,train,cli]
from deepqmc import Molecule, evaluate, train
from deepqmc.wf import PauliNet
mol = Molecule.from_name('LiH')
net = PauliNet.from_hf(mol).cuda()
train(net)
evaluate(net)Or on the command line:
$ cat lih/param.toml
system = 'LiH'
ansatz = 'paulinet'
[train_kwargs]
n_steps = 40
$ deepqmc train lih --save-every 20
converged SCF energy = -7.9846409186467
equilibrating: 49it [00:07, 6.62it/s]
training: 100%|███████| 40/40 [01:30<00:00, 2.27s/it, E=-8.0302(29)]
$ ln -s chkpts/state-00040.pt lih/state.pt
$ deepqmc evaluate lih
evaluating: 24%|▋ | 136/565 [01:12<03:40, 1.65it/s, E=-8.0396(17)]
- Documentation: https://deepqmc.github.io