problem in the frequence calculation of free molecule

Message

rghomari · #1 Post by **rghomari** » Thu Jul 30, 2009 9:19 pm

Hello users VASP.
I had problem when i calculate the frequences of free molecule (2-cyclohexenone ) in large box (13.59*9.41*20)
i found 39 frequencies positive and 5 imaginary frequencies,
but when i do this calculation in Gaussian program i found only the positive frequences.
1-where it come from these imaginary frequencies?
2-the same problem in the calculation of surface (2layers of Rh (111)) i got 18 imaginary frequencies.
My INCAR file :
System = frequence calculation
ISMEAR=0
PREC=Accurate
GGA=91
ENCUT=400
IBRION=5
POTIM=0.015
EDIFF=1e-04
EDIFFG=1e-03
NSW=20
LREAL=Auto
Thank you in advance.

graeme · #2 Post by **graeme** » Fri Jul 31, 2009 3:09 am

Try systematically reducing sigma and POTIM and look for convergence of the normal mode frequencies.

rghomari · #3 Post by **rghomari** » Fri Jul 31, 2009 9:21 pm

thanks, i will try it

physicalattraction · #4 Post by **physicalattraction** » Wed Aug 12, 2009 7:00 am

If you have N atoms in your molecule, than you have 3N degrees of freedom, so your dynamical matrix has the size 3N x 3N. However, it can be shown that, when you calculate these eigenvalues, 6 eigenvalues will be 0, so you are left with 3N - 6 nonzero eigenvalues for your dynamical matrix, which means 3N - 6 vibrational modes. In practice however, the eigenvalues are not always exactly zero, but always small, in some cases even so small that their square is negative, hence the imaginary frequencies.

I have however the same problem. I try to calculate the vibrational modes of a dimethyl sulfide molecule, but the frequencies heavily depend on my choice for POTIM. I have reduced it from 0.020 to 0.001 already, and the frequencies do not converge. Changing SIGMA from 0.05 to 0.01 does not change the frequencies for all choices of POTIM, so that one is okay. It also heavily depends on my choice for NFREE (I have tried 4 and

. How can I reliably calculate frequencies of a molecule in the gas phase? Is POTIM = 0.001 still not low enough?

graeme · #5 Post by **graeme** » Wed Aug 12, 2009 12:26 pm

Go even smaller with SIGMA, by powers of 10 (e.g. 1e-4 or 1e-5). Also, use EDIFF=1e-8 to generate accurate forces. Then look for convergence of the frequencies with POTIM from 0.01 to 0.001. I have never seen the need to use a smaller finite difference step size than 0.001 Ang.

physicalattraction · #6 Post by **physicalattraction** » Fri Aug 14, 2009 9:28 am

Thanks for the tip. I tried systematically changing EDIFF = 1E-4, 1E-6 and 1E-8, SIGMA = 1E-2, 1E-3, 1E-4 and 1E-5 and POTIM = 0.015, 0.005 and 0.001. Only the results for EDIFF = 1E-8 were converged, but the values for the frequencies were already correct for the highest values of SIGMA and POTIM. Conclusion: I have to set EDIFF = 1E-8 to obtain the correct frequencies.