frequency at saddle point

[lxs234]

Hi everyone,
I am studying the solute (Al, Si, Ge, Ga, In, In) diffusion in hcp Ti using 8 frequency model.
for each jumps i have found the transition state using NEB with a 3x3x3 (54atom) supercell and calculated the frequency using IBRION=5.
However, for the jump frequency of the solvent where it exchanges its position with the nearest neighboring vacant site in the same basal plane when the solute is in the adjacent plane (this jumps is referred as wb'), when I have calculated the frequency at the saddle point, I haven’t obtained the imaginary frequency.
I started my Neb calculation using 5 images, Ediff=10-6 and Ediffg=-10-2 and using 13x13x7 kpoint mesh, I have used the same Ediff and Ediffg and kpoint mesh for the frequency calculation, then I repeat the frequency calculation using Ediff=10-8, and potim 0.001 but still I couldn’t get the imaginary frequency.
Any suggestion? Is there anyone who study the same system or have the same problem?

Thank
Lucia