Is setting ISYM = 0 recommended for structure optimizations?

[asrosen]

When optimizing a structure with VASP, is it recommended to use ISYM = 0 or not? I do not see any mention about specifying ISYM = 0 during a structure optimization in the VASP manual, but one would naively think that if symmetry is enabled then it is possible for the structure to retain its symmetry over the course of the optimization such that a lower energy state might not be found. It is a common recommendation in other electronic structure packages to disable symmetry during a structure optimization, but I am curious what the VASP team's suggestions are here.

[michael_wolloch]

Hi Andrew,

Indeed, we would not recommend setting ISYM = 0 for structure optimization. While this limits the search space and could prevent the structure from reaching a different minimum, we believe that turning symmetry off is not the correct way to search for a global minimum due to the relatively large amount of noise added. Instead, one should either try to optimize different, but related, structures (as mentioned in the structure optimization article on the wiki), or follow imaginary/soft phonon modes.

For example, if a tetragonal distortion is considered possible in a cubic system, it is better to start the relaxation with a tetragonally distorted structure with symmetry turned on than to start from a cubic structure with symmetry turned off.

Nevertheless, you are correct that the information on the wiki does not adequately address the implications of symmetry on relaxations. I will update the structure optimization article together with Martin Schlipf (who wrote it) next week, considering this.

Thanks for the feedback, and please feel free to share your opinions on the matter in this thread.

All the best, Michael

[asrosen]

Thanks for your input!

I definitely agree that ISYM = 0 will not magically get the user a global energy minimum (if only, right?). In a high-throughput setting, there is perhaps an argument to be made about whether setting ISYM = 0 would be worthwhile or not (since one may not know a priori the different phases accessible to the system) in the hope that you are not trapped in some spurious high-symmetry configuration. But this is perhaps a matter of opinion.

I appreciate the update to the documentation. That was indeed the key part I was wondering --- does ISYM impact relaxations, and it seems that it does. Very useful to know. Thank you!

[michael_wolloch]

Dear Andrew,

I agree that for high-throughput, the situation is debatable and probably depends also on the set of initial structures used. I am glad you agree with the documentation update, and thank you again for bringing this to our attention.

Cheers, Michael