All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
1.2.0 - 2024-09-10
- Implementation of excited state simulation
- Spin-penalty for targeting spin states
- Evaluation of HF-forces, spin and other observables
- Type hints and type checking
- Support of folx forward Laplacian
- Expanded configuration through hydra
- Utilities for parsing result files
- Improved logging
- NaN-restart mechanism
- Avoid executing code on import
- Rename pp (pseudo potential) to ecp (effective core potential) to be consistent across codebase
- Support for cuda <= 11
- Support for python 3.9
- Deprecated PauliNet ansatz from 10.1038/s41557-020-0544-y
1.1.2 - 2023-11-20
- Add script to run common configs before release
- Change location of hydra output dir
- Extend development workflow with codespell and mypy
- Update workflows
- Add pre-commit hooks
- Extend documentation
- Run tests in double precision
1.1.1 - 2023-10-20
- Refactor and code cleanup
- Refactor parallel utils for more efficient pmaps
- Errors when executing tutorial
- Ansatz config files
- Tensorboard logging
- Bug crashing energy evaluation
- Saving of final checkpoint
1.1.0 - 2023-09-25
- Refactor and simplify the PauliNet
- Generalize the GNN framework
- Remove wave function state and cutoff mechanism
- Remove defaults from code and fully transition to ansatz initialization through hydra
- Implement pseudo potentials
- Implement training on multiple molecular configurations (undocumented)
- Add configuration files for PauliNet, FermiNet, DeepErwin and PsiFormer
- Add multi-GPU support
- Remove deepcopys slowing down the execution
- Improve logging
- Explicit creation of logging directories
- Remove specification of orbital configurations in wave function ansatz
- Remove GraphNeuralNetwork base class
- PauliNet -> NeuralNetworkWaveFunction
- SchNet -> ElectronGNN
1.0.1 - 2023-01-02
- Issues related to sampler initialization and synchronization with parameter update
- Minor issues with logging, rewinding and checkpointing mechanisms
- Improve compatibility of cli and slurm
1.0.0 - 2022-12-06
- Complete rewrite of the DeepQMC package in JAX
- Support for PyTorch < 1.10
0.3.1 - 2021-12-12
- Version incompatibility of PySCF and H5PY
- Minor issues with walker initialization in some systems
- Minor numerical instability causing rare NaNs
0.3.0 - 2021-01-27
PauliNet
:- Mean-field Jastrow and backflow
- H2O and H4 rectangle systems
- Support for Python 3.9
PauliNet
/OmniSchNet
:- API
PauliNet
:- Separate Jastrow and backflow factories
mo_factory
, functionality replaced with a mean-field backflow- Real-space backflow
- Systems with n_up/n_down = 0
0.2.0 - 2020-08-19
- Command-line interface
PauliNet
:- Nuclear and electronic cusp corrections on by default
omni_kwargs
accepted byPauliNet()
instead offrom_hf()
- All keyword arguments to
from_hf()
are passed toPauliNet()
Sampler
:from_mf()
changed tofrom_wf()
, doesn't use PySCF object by default
0.1.1 - 2020-07-24
Rerelease of 0.1.0 with added package metadata.
0.1.0 - 2020-07-24
This is the first official release of DeepQMC.
At this moment, DeepQMC should be still considered a research code.
- Core functionality to run variational quantum Monte Carlo with Pytorch
- PauliNet, a deep neural network ansatz