dynmat calculation to get prefactor for vacancy jump

[erhart]

First of all, thanks for kindfully providing your codes. They have really made my work much easier. So far I have been doing quite a few CI-NEB calculations to obtain the energy barriers for diffusion which worked fine. More recently I tried to obtain prefactors via the Vineyard equation but encountered problems when trying to analyze the dynamical matrix calculations: They gave several imaginary frequencies for the initial state as well as the saddle point configuration. I repeated the calculations for a perfect cell and still got one imaginary frequency. I tested displacements between 0.001 and 0.1A and varied the size of the dynamical matrix between 15x15 and 93x93 (with 31 atoms in the full cell). I also fitted several matrices corresponding to different displacements. In all cases I got imaginary frequencies. The system of interest is a vacancy jumping on a wurtzite lattice. Currently, I am using 32-atom supercells. Larger supercells would be desirable but are beyond the computational capacity at my disposal. What else do I have to check? Could the problem be related to the lattice symmetry? Any hint would be appreciated. Thanks.

[graeme]

Are all atoms allowed to move in your calculation? If so, there should be three low translational modes. Noise in the forces can make these modes positive or negative. Also, the sqrt[k/m] tends to accentuate the value of low modes.

Also, make sure you have a small ediff. Try ediff=1e-8 to get accurate forces.

You minima should be well converged, using at least ediffg=-0.01.

Make sure that the masses are correct in your massfile.

One last issue could be the smearing function. I had problems trying to find the frequencies of CH4. It turns out that the default M-P smearing was causing problems and the Gaussian smearing works better. This will show up as something odd in the density of states. There was a helpful message about this on the vasp forum:

http://cms.mpi.univie.ac.at/vasp-forum/ ... .php?4.138