Fix convergence error handler to prevent exceeding niter_max of 500

[t-reents]

This closes #48.

@bastonero To discuss: Once the limit of 500 is reached, should we try to further reduce alpha_mix or just do a last run with niter_max = 500 and the current value of alpha_mix?

[bastonero]

Honestly I don't think increasing niter_max is a good solution either way. Possibly setting its maximum value to e.g. 200 is already enough. Then, we should try to reduce alpha_mix and set nmix to e.g. 8.

[bastonero]

In the meantime: could you also provide a test for the new logic?

[t-reents]

So your idea would be keeping the current logic of reducing alpha_mix and adding something like?

if parameters.get('niter_max', 200) < 200: 
    # In case niter_max was specified and smaller than 200, increase it
    parameters['niter_max'] = min(parameters['niter_max'] * self.defaults.delta_factor_niter_max, 200)
elif 'niter_max' in parameters: 
    # This is the case where niter_max is already present and already set to the maximum of 200. 
    # Increase nmix to 8
    parameters['nmix'] = 8
else: 
    # In case niter_max was not specified in the beginning
    parameters['niter_max'] = 200

[bastonero]

Yeah, something like that. Nevertheless, we don't have sufficient data to prove this will work. You can see also recent discussions for the pw.x counter part in aiida-quantumespresso.

Possibly we can set niter_max=200, nmix=8 and alpha_mix=0.4 in the protocols, and then just reduce the alpha_mix as a solution. We don't have many strategies to play around with, and w/o proofs I think this is by now a safe option.

Pointless to keep increasing the number of iterations if then it doesn't help at all convergence.

[t-reents]

Sounds reasonable. I adapted the handler and moved the new parameters to the protocol. The tests fail somehow during the data regression, although they pass locally. I had to adapt the reference files due to the new protocols, so not sure whether the CI compares to the previous version or something like that. @bastonero Would be great if you could have a look

[bastonero]

Thanks @t-reents , looks good. I guess for some reason you don't have to specify num_mpiprocs_per_machine: 1 in resources. Try to add them and commit - it should solve the issue.

[bastonero]

All good, thanks @t-reents , just the two extra lines to remove.

[t-reents]

Thank you @bastonero, tests pass now

[bastonero]

Thanks @t-reents !