do you have an example input that fails and can be run in a reasonable time?
On 06/11/17 15:55, Maxim Skripnik wrote: > Dear developers, > > I'd like to report a strange behavior of neb.x when one sets > tefield=true and dipfield=true. The error of the first image becomes > very large even though the xyz coordinates in FIRST_IMAGE are properly > relaxed. This seems to be not an issue for the actual neb calculation > since neb.p finds the energy barrier anyway. Nevertheless, it leads to a > corrupt .path file, which contains either ********** or/and NaN. When > one wants to restart the neb calculation (e.g. when going from > CI_scheme='no-CI' to 'auto'), is crashes since neb.x reads the corrupted > .path file which contains ******* and NaN. > I took a look at the pw.x calculation of the first image, which is done > only once at the beginning of a neb calculation (stored in > prefix/neb_1/PW.out). And already there one can see the problem: > > end of prefix/neb_1/PW.out: > > ... > ... > ... > > convergence has been achieved in 26 iterations > > Writing output data file neb.save > > Forces acting on atoms (Ry/au): > > atom 1 type 2 force = ************** NaN************** > atom 2 type 2 force = ************** NaN************** > atom 3 type 2 force = ************** NaN************** > atom 4 type 2 force = ************** NaN************** > ... > ... > ... > > Total force = NaN Total SCF correction = 0.000109 > > Interestingly, this doesn't happen with the last image, which was > relaxed with exactly the same pw.x settings. This happens even with > eamp=0, as long as tefield=true and dipfield=true. I was able to > reproduce the behavior with QE 6.0 and 6.1 on two totally different > computer clusters. Any idea why this happens? > > Best regards > Maxim Skripnik > University of Konstanz > > _______________________________________________ > Pw_forum mailing list > [email protected] > http://pwscf.org/mailman/listinfo/pw_forum > -- Lorenzo Paulatto - Paris _______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
