v4.0.0-final

This is the final commit we consider part of v4.0.0. It contains mainly bug fixes for the original v4.0.0 release.
Combine with the newest psiflow container: psiflow:v4.0.0-final_cu124

Expect many breaking changes after this release. Some modules will (temporarily) stop working on the main branch. Please bear with us while we perform some internal housekeeping.

What's Changed

• Fix Github CI by @pdobbelaere in #69
• Pin Parsl version to the one in container 'oras://ghcr.io/molmod/psiflow:v4.0.0_cu118' by @pdobbelaere in #66
• copy default CP2K dict per evaluation by @svandenhaute in #71
• Make client wait for socket to appear by @pdobbelaere in #75
• Functions by @pdobbelaere in #67
• Option to set atomic masses in walkers by @armaet in #79
• Consolidate TMPDIR creation for bash apps by @pdobbelaere in #73
• enforce lock during file creation by @pdobbelaere in #80
• Implement cell optimisations through ASE by @pdobbelaere in #81
• Support for Psiflow driver in i-Pi by @pdobbelaere in #88
• Replica exchange Hamiltonian swaps as bias by @pdobbelaere in #90

New Contributors

• @armaet made their first contribution in #79

Full Changelog: v4.0.0...v4.0.0-final

v4.0.0

What's Changed

• Function by @svandenhaute in #30
• Create symlinks into psiflow_internal directory by @pdobbelaere in #61
• update Parsl to 2024.12.16 and cctools to 7.14.0
• a bunch of smaller bugfixes

New Contributors

• @pdobbelaere made their first contribution in #61

v4.0.0-rc1

Fix Parsl dependency which pointed to a nonexisting PR.

Complete redesign!

see docs.

v3.0.4

• clean up CP2K interface by removing all pymatgen dependencies and using cp2k-input-tools
• speed up Dataset manipulations with simple tricks
• remove deprecated NWChem interface. Cluster calculations should be done through pyscf. In the near future, also ORCA will be supported

v3.0.3

Bugfix in type annotation of SequentialLearning

v3.0.2

• fix wandb logging bug where if mask is all NaNs, the returned value was not a tuple but a float (i.e. NaN)
• ensure committee functionality is robust towards datasets with varying number of atoms

v3.0.1

This release implements a bugfix where training and walker dynamics tasks were not properly terminated when they hit their max_walltime.

psiflow v3.0.0

• PySCF support
• New Weights & Biases interface (click here to see what it looks like!)
• Temperature ramps can now be applied on a smaller number of iterations, where subsequent iterations will perform the sampling continue at the highest (final) temperature
• Execution configurations are now specified in a much easier YAML file format (see configs for examples)
• There is now a comprehensive notebook which explains the basics of Parsl, psiflow, and free energy calculations with ML potentials and pymbar. It's still somewhat preliminary so it might still contain a few typos.

In terms of development, we've added support for Python 3.10, added CI pipelines for basic tests which only require small amounts of CPU compute, and have adopted the black code style and flake8 linting.

v2.0.0

• support for OpenMM --> much faster simulations (CPU x2-x5, GPU x10)
• better active learning workflow --> much less 'unphysical' samples in generated data
• better Weights & Biases logging --> easier to track progress
• support for cluster calculations via NWChem
• support for query-by-committee active learning
• ... and many other small bugfixes!