Dear Administrator,
I have done few NEB image calculations and finally ended up with one image which is close to TS. And then I decided to do Dimer calculation with this structure as the starting POSCAR and MODECAR derived from the initial and final POSCARS using the modemake.pl script. Finally I ended up with a structure which has zero forces on the atoms and resulted in two imaginary frequencies.
Now I would like to ask you what should I need to do as a next step to remove this small second frequency, because our criteria is to have one imaginary frequency for the normal mode analysis. Please suggest me what should I do to remove this second frequency?
Should I increase the energy convergence criteria from 1e-07 to 1e-10? And EDIFFG is presently -0.001 and needs to be increased to -0.0001?
Thanks
second run of Dimer method
Moderator: moderators
I don't think you should ever need to converge a dimer calculation further than 0.001 eV/Ang. There are a few other things to check though:
(1) The dimer uses a finite difference approximation to calculate curvatures. Because of this, the current position (written to the CONTCAR) can be displaced from the saddle point (center of the dimer) by the distance DNdR. If the run terminates on its own, the CONTCAR should be at the center of the dimer. If you kill it, it could be displaced by this small amount. To deal with this issue, we always write the position of the center of the dimer to the CENTCAR. If this is identical to the CONTCAR, there is no problem, and you can use either for the mode calculation. If they are different - use the CENTCAR.
(2) The most likely problem is with the normal mode calculation. It is very important that the forces are accurate enough for the finite difference step used, and that that this step, in turn, be small enough. Accurate forces can be calculated with ediff=1e-8, and a small enough displacement should be no larger than 0.01 and better if closer to 0.001 Ang. If you recalculate the modes with a different finite difference displacement, and the modes change significantly, there is a problem. Other important variables for accurate forces are lreal=.false. and isym=0.
(3) Finally, there is a remote possibility that the dimer can converge on a second order saddle. We have only seen this in rare cases when there is unusually high symmetry at the saddle point. If this is the case, we would be interested to look at this in detail.
(1) The dimer uses a finite difference approximation to calculate curvatures. Because of this, the current position (written to the CONTCAR) can be displaced from the saddle point (center of the dimer) by the distance DNdR. If the run terminates on its own, the CONTCAR should be at the center of the dimer. If you kill it, it could be displaced by this small amount. To deal with this issue, we always write the position of the center of the dimer to the CENTCAR. If this is identical to the CONTCAR, there is no problem, and you can use either for the mode calculation. If they are different - use the CENTCAR.
(2) The most likely problem is with the normal mode calculation. It is very important that the forces are accurate enough for the finite difference step used, and that that this step, in turn, be small enough. Accurate forces can be calculated with ediff=1e-8, and a small enough displacement should be no larger than 0.01 and better if closer to 0.001 Ang. If you recalculate the modes with a different finite difference displacement, and the modes change significantly, there is a problem. Other important variables for accurate forces are lreal=.false. and isym=0.
(3) Finally, there is a remote possibility that the dimer can converge on a second order saddle. We have only seen this in rare cases when there is unusually high symmetry at the saddle point. If this is the case, we would be interested to look at this in detail.