Dear Anja, in my experience, large negative frequencies are either trivial "noise"(too low mesh cutoff - apparently not your case), or, more frequently, a hint that you are not really at equilibrium. A good idea would be to look at the eigenvectorof the "most negative" mode, extract the relative displacements(not to forget the sqrt of masses) and give your atoms a small kickin this direction before resuming the relaxation. The negative (imaginary)frequency means - you displace the atoms in this mode, and the energy lowers.This might help you to find a better energy minimum,calculate there the frequencies again, etc.Another trouble (or, the reason for the behavior described above)could be the dipole-dipole interactions between your moleculesif the box is not large enough. However, the d-d interaction is long-ranged, so you'll have some amount of trouble at any box size.I don't remember if there are efficient switches to cancel it in this context.For a molecule like H2S, you would normally get 3 rigid displacementand 3 rigid rotation modes, all at zero frequency. It is relatively easyto bring the displacement modes to zero (by brute-force increasingthe mesh cutoff), whereas bringing the rotation modes to zero can bea hell, apparently because the "Siesta space" is not isotrope enough (interaction with replicas of the molecule, the rigid structure of mesh, ..?)Best regardsAndrei Postnikov ----- Anja Foerster <foers...@msu.edu> a écrit :
>Dear all, > I performed a geometry optimization of H2S and then used the Vibra tool and a > FC run to calculate the frequency. For the optimized structure I receive 5 > negative (imaginary) frequencies (-228.9332 -135.1068 -46.3373 > -0.1270 -0.0328). Since it is a small molecule, I varied both the > H-S-distance and the H-S-H angle. As you can see in the attached pdf, 1.38 > Ang is the H-S distance that corresponds to the energetic minimum. However, > only when the H-S-distance is increased to 1.40Ang are the frequencies > getting close enough to 0 for the structure to be labeled as "ground state" > based on the frequency results. > I already tried to increase the meshcutoff value to 800, which did not solve > the problem. My other critical values: MD.MaxForceTol is 0.001 eV/Ang and > DM.Energy.Tolerance is 1.0d-5 eV. I set the MD.MaxForceTol to a stricter > 0.0001 eV/Ang in the geometry optimization calculation, however even after > 400 steps is not converging. For MD.MaxForceTol 0.001 eV/Ang I receive “Tot > 0.002653 0.001420 -0.001683” in the siesta output for “ siesta: Atomic > forces (eV/Ang)”. So the forces should be okay for a frequency run. > Does anyone have an explanation why the energetically lowest state has 5 > imaginary frequencies whereas an energetically higher geometry has > effectively no imaginary frequencies? > Thank you for your help, > Anja > P.S.: I'm using siesta version 3.1 >
