Question about the final cell update in SSCHA vc_relax

[zhaowb12138]

In the final iteration of vc_relax, if the stress evaluated for the current cell already satisfies the pressure convergence criterion within its statistical uncertainty, UpdateCell() is still performed once more before the relaxation terminates.

This means that the final reported cell is the updated cell, while the stress used to establish convergence was evaluated for the cell before that last update.

Would it be more appropriate in this case to stop without performing the final cell update once the pressure convergence criterion has already been satisfied?

[mesonepigreco]

Hi,
Yes this was originally a bit controversial. I agree that it would be much nicer to have a final dynamical matrix that matches the cell of the last ensemble, which can then be used automatically for linear-response calculations without computing a new ensemble.
The reason why it is so is that in some cases (small ensembles), you also want to use the last gradient to further optimize the cell, otherwise, what happens is that if you raise the temperature slowly (let's say 1 K per step), your stress will most likely be converged so you will not see any change in volume until the jump in temperature is enough to cause a stress above your stochastic threshold. This results in jumps in the thermal expansion calculation. In that case, it is useful to have one extra cell step, which gives you a final, slightly more accurate cell value, since you have the stress. I know it is below the stochastic threshold, but there is still a signal that can be extracted from it.

Now, probably we could just save both?