Does unstable modes in svsl matters?

Hi all:

Recently days, I am learning how to consider vibrational contribution in cluster expansion. When I get the static calculation result and fit successfully, I run ‘foreachfile str_relax.out pwd ; svsl -d’. However the program inform me a lot of unstable configurations. I know from manual that we need to find out this unstable mode and generate perturbations along the unstable directions. However what I concern is, does those modes matters in vibrational free energy of a structure using the Stiffness VS Length method? Because the configurations I selected in fitting stiffness are all ground states, they are tought to be stable by program. Those configurations who are unstable are acutally above ground states. In principle, they should not influence the fitting process and vibrational energy calculation, right? So, in this case, can I ignore those unstable modes and force svsl to continue on all structures including those unstable modes by svsl -fn?

svsl uses an approximation (transferability of the force constants) to calculation the phonon dos quickly. This may cause fictitious unstable structures. 3 solutions:

  1. ignore them: the default (without -fn) is to not output the free energy and so clusterexpand will not include those in the fit. This is the simplest if there are enough structures to get decent fit.
  2. include them using the -fn option of svsl. There is a failsafe mechanism: the free energy ignore the unstable modes, but the contribution of the other modes are scaled so that there still 3N-3 vibrational modes.
  3. Use the more expensive fitfc. There, for sufficiently long-range springs, a truly stable structure, should show up as stable.