Ensemble setup for QSpaceHessian

[tveskaeg]

Hi there!
Congratulations to the new release of v1.6!
I'm very curious to try the new approach QSpaceHessian, which sounds amazing for larger systems.

Before getting started, I wanted to be sure to set up everything correctly. I'm following the short example from https://sscha.eu/tdscha/qspace-hessian/ and I noticed that the ensemble is not updated by the SSCHA minimized final dyn.
In earlier versions (get_free_energy_hessian, or StaticHessian) this update was always done by calling sth like:

ensemble.update_weights(final_dyn, T)

Is this not needed for the QSpaceHessian? Or is that included automatically by ensemble.load_bin? My previous calculations are not stored as binary format, but in a folder by calling
ensemble.save(folder, ...)
...
minimizer.dyn.save_qe("dyn_end_population" ... )

Best,
Roman

[mesonepigreco]

Hi, I would not use the new QSpaceHessian for now; we found a severe bug in the implementation that prevents it from getting the correct result. We are preparing a new release with a fixed version, which we hope to release in 1 month (or at least merge into the master branch).

Version 1.7 should fix it definitively and make it the default mode when calling the standard get_free_energy_hessian; however, as it stands now, if q != -q + G points are present in the mesh, the calculation is unreliable (unfortunately, this means that it passes tests on small supercells like 2x2x2, giving the false impression of reproducing the correct results).

We have a working branch that we will hopefully merge in a few weeks, together with a paper about the algorithm, benchmarks, and correctness proof.