[gmx-users] negative eigenvalues occured and not "nearly zero"

[Mark Abraham]

silvester.thu wrote:
> Hi Berk, thanks for your reply, but there are still problems.
>  
> While the first six eigenvalues are corresponding to the three 
> translational and three rotational dimensions of freedom of the whole 
> system, they should be zero (theoretically) or at least not 
> much different from zero (e.g: 1e-3 or -1e-3). But in my problem, I 
> encountered some negative values that are much different from zero, as I 
> listed before. 
>  
> And I have used the g_nmtraj_d to generate and visualize the modes 
> corresponding to those negative eigenvalues, and I found that those 
> modes were neither translational nor rotational movements -- they were 
> actually "oscillational movements". It is unreasonable. So I guess there 
> might be something wrong in my calculation. But, I still can not figure 
> out what is wrong, even after I discussed with members in my group this 
> morning.

I don't have much experience in such a calculation with an MM force
field, but such Hessian eigenvalues in quantum chemistry indicate that
you are at some non-minimum stationary point. That seems unlikely for an
MM EM calculation, unless the PES is very flat. I can't guess what
"oscillational movements" are, but if you perturb the system in the
direction of that eigenvector (easy if it's mostly on a few atoms,
otherwise just add a suitable multiple of the eigenvector to the atom
coordinate vector) you can see if you get into the range of a suitable
local minimum, or return to your existing stationary point.

Mark