all frequencies real in a dymmatrix calculation

[mr.precog]

Hello everybody!

I am studying a bcc system with a monovacancy. The goal is to find the prefactor, so I first perform a NEB calculation, by means of the compiled VASP TST binary, in order to find the saddle point. All calculations are well-converged both at the minimum and the saddle point (EDIFF = 1E-6 and EDIFFG = -0.001). Next I run the dymseldsp script to obtain the DISPLACECAR file, and then proceed to run VASP again. However, when I try to use the dymmatrix script in order to get the required frequencies at the saddle point, these are all real. I have tried first to set EDIFF = 1E-8 and EDIFFG = -1E-8 in my INCAR and subsequently increased EDIFFG to -1E-10, keeping EDIFF = 1E-8. And the result is always the same. The number of atoms in the Hessian calculation is 126, since I keep one fixed to avoid translational modes, and their corresponding displacement 0.001. What else could I try? Any suggestion is very much welcome.

Thanks in advance
Sergio L. Palacios

[graeme]

If you post your NEB calculation, I can take a look. I would start by checking the convergence of the saddle.

[mr.precog]

Dear Graeme,

Thank you very much for your help. Please find attached the original NEB calculation with the minimum amount of files that I think are necessary to check it. Please remember that I re-relaxed the equilibrium and saddle points with much more demanding conditions subsequently.

I look forward to read from you again

Regards,

Sergio L. Palacios

[graeme]

The force on the saddle is still significant. How did you subsequently re-relax it?

[mr.precog]

As I told you in my first post, I relaxed the saddle point with EDIFF = 1E-6 and EDIFFG = -0.001. The INCAR file used was:

NPAR = 2
ISTART = 0
ENCUT = 479
PREC = Acc
EDIFF = 1E-6
EDIFFG = -0.001

NSW = 100
IBRION = 2
ISIF = 2
ISYM = 0
SMASS = 2
ISMEAR = 2
SIGMA = 0.2
POTIM = 0.25
IOPT = 2
RWIGS = 1.455

[graeme]

Ok, but you can't relax to a saddle point with conjugate gradients. This will take you farther from the saddle. You need a method which converges to a saddle, such as a min-mode following method. Perhaps simpler is to continue your NEB calculation. I do, however, recommend compiling vasp with the VTST code so that you can use the climbing image NEB method. This will bring the highest energy image to the saddle point upon convergence.

[mr.precog]

Thank you very much for your response and help. I will try again the calculation taking your advice and recommendation into account.

Regards

[mr.precog]

[quote=graeme post_id=8885 time=1491307651 user_id=2]
Ok, but you can't relax to a saddle point with conjugate gradients. This will take you farther from the saddle. You need a method which converges to a saddle, such as a min-mode following method. Perhaps simpler is to continue your NEB calculation. I do, however, recommend compiling vasp with the VTST code so that you can use the climbing image NEB method. This will bring the highest energy image to the saddle point upon convergence.
[/quote]


Dear Graeme,

Following your advice I have performed the NEB calculation with climbing image (please find attached the files). After that I have relaxed the saddle point with the same requirements that I told you in my previous messages. Then I run the dymseldsp and dymmatrix scripts again finding all frequencies real at the saddle point one more time. The head of my freq.dat file is as follows:

12.724775 cm^{-1} ... 0
12.931559 cm^{-1} ... 0
13.003913 cm^{-1} ... 0
69.598437 cm^{-1} ... 0
70.093844 cm^{-1} ... 0
70.570899 cm^{-1} ... 0
71.976124 cm^{-1} ... 0
72.044974 cm^{-1} ... 0
72.451618 cm^{-1} ... 0
72.630259 cm^{-1} ... 0

Any suggestion? What am I doing wrong?

Regards,

Sergio L. Palacios

[graeme]

The NEB calculation looks ok to me. One small suggestion: it would make sense to use an odd number of images for a symmetric pathway. This will help to identify the climbing image and not rely on the calculation to break symmetry.

I am more concerned about your statement that you relaxed the saddle point after the NEB calculation. The point of the climbing image calculation is that one image (03 in your case) will converge to the saddle. You should not do a relaxation of this configuration, unless you use a method such as the dimer method which is designed to converge to a saddle point. In your case, the NEB forces are low and so there is no need for this.

Finally, make sure that the forces are sufficiently accurate (e.g. EDIFF=1e-8) and your finite difference displacement is sufficiently small (e.g. < 0.01 Ang). You can also try the linear response method, which is now implemented in vasp, for calculating frequencies.

[mr.precog]

Thank you very much for your help. I will proceed as you tell me hoping that everything will run ok now.

Regards,

Sergio L. Palacios