New release 1.4 of the SSCHA

[mesonepigreco]

Release 1.4

We are thrilled to announce the release of version 1.4 of our code, featuring significant enhancements and exciting improvements:

Key Highlights

• Code Speedup: In this release, we have achieved a remarkable speedup, with performance gains ranging from 10x to 100x in specific scenarios. This substantial acceleration is due to the new integration with Julia and a comprehensive algorithm rewrite, enabling all calculations in Fourier space during the free energy minimization.
• Enhanced Code Parallelization: Version 1.4 introduces improved code parallelization. This enhancement extends to the SSCHA minimization (now parallelizing both the evaluation of energies and forces, the gradient, and stochastic reweighting utilizing MPI) and the spectral function.
• AiiDA Integration with Quantum Espresso: Thanks to the contributions of Lorenzo Bastonero (@bastonero), the sscha code now offers a seamless integration with AiiDA and Quantum Espresso. In this way, the SSCHA leverages AiiDA to deal with cluster communication, allowing easier integration with non-SLURM or more exotic HPC configurations.

Do not worry; the old plain cluster module remains a supported feature, and we care about backward compatibility, so there is no need to change your input scripts or install AiiDA to submit on a cluster!

How to update

If you installed from conda or PyPi, just run:

pip install -U cellconstructor python-sscha tdscha

And all the libraries will be updated to the latest version.

Note that tdscha, cellconstructor, and python-sscha all require to match the minor version (i.e. must be all from 1.4.* branch) as they are developed together, and some API changes that one package may need to improve the efficiency of the code is implemented simultaneously in other packages.

Detailed changes

In addition to these significant advances, we have diligently addressed various minor issues and bugs. A more complete list follows.

Julia is not a mandatory requirement to run the new version of the SSCHA. However, while the code offers a fallback mode if Julia is unavailable, we strongly recommend having Julia installed, as failure to do so may result in a significant slowdown, reducing performance gains achieved through this release (without julia the code utilizes the same implementation as in version 1.3).

For updated installation instructions and to experience the full benefits of release 1.4, visit http://sscha.eu/download/

We're excited about this release and the opportunities it offers. Your feedback is invaluable to us, and we look forward to your continued support and engagement with our code.

Thank you for being a part of this journey.

Best regards,
Lorenzo Monacelli, on behalf of the developers

What's Changed

• Timer by @mesonepigreco in Timer #118
• add Slurm QOS parameter support by @eliheady in add Slurm QOS parameter support #120
• Fast parallelization of gradient calculation by @flibbi in Fast parallelization of gradient calculation #111
• Add a more advanced timing system that keeps tracks of nested functions and produce a good report by @mesonepigreco in Add a more advanced timing system that keeps tracks of nested functions and produce a good report #117
• Update Tutorial_PbTe.ipynb by @diegomartinez2 in Update Tutorial_PbTe.ipynb #125
• Add an early interface with the AiiDA infrastructure by @bastonero in Add an early interface with the AiiDA infrastructure #134
• Fourier gradient by @mesonepigreco in Fourier gradient #136
• Fix: props not computed are not poped (AiiDA) by @bastonero in Fix: props not computed are not poped (AiiDA) #158
• Add a new explanation for the VC Relax by @mesonepigreco in Add a new explanation for the VC Relax #150
• Examples: add a sscha-aiida example by @bastonero in Examples: add a sscha-aiida example #148
• Added an error if no stress tensor is found by @mesonepigreco in Added an error if no stress tensor is found #166

New Contributors

• @eliheady made their first contribution in add Slurm QOS parameter support #120
• @flibbi made their first contribution in Fast parallelization of gradient calculation #111
• @bastonero made their first contribution in Add an early interface with the AiiDA infrastructure #134

Full Changelog: v1.2...v1.4

---

This discussion was created from the release v1.4.