Skip to content

New version 1.1

Compare
Choose a tag to compare
@mesonepigreco mesonepigreco released this 24 Feb 21:26
· 396 commits to master since this release

A lot of bugs have been solved, and finally, we release the stable version of the python-sscha code.

The focus of this release was to improve the user experience. To this aim, we implemented a better minimization algorithm and a much improved variable cell relaxation algorithm. In particular, we:

  1. Minimization now occurs with automatic step selection (line-minimization), able to account also for imaginary frequencies and too drastic steps due to the stochastic nature of the algorithm. The user does not have to specify the step anymore, as the automatic step selection works extremely well.
  2. The variable cell minimization algorithm uses a similar line-minimization approach, we also fixed a bug in the units of the bulk modulus. Now the algorithm for the cell dynamics is robust and never diverges, even with a noisy ensemble.
  3. Speedup of the code. We removed a lot of redundant diagonalization performed throughout the code that slowed down the minimization, in particular for systems with a large primitive cell.

We have also added some useful extra features to the ensemble class, like the possibility to save into the raw format to interface with DeepMD-kit code (you can use your SSCHA ensemble to train a neural network potential!)

Furthermore, we extended the functionality of cluster calculation, which is more stable. Still, a major update on the cluster interface will come in the near future.

Remember to upload also CellConstructor to version 1.1, as the two codes must always work in paired versions.

What's Changed

New Contributors

Full Changelog: https://github.com/SSCHAcode/python-sscha/commits/v1.1