[gmx-users] Minimising forces for vibrational normal mode analysis

[Nash, Anthony]

Hi all,

I would like to pull out the vibrational normal modes using gromacs over a
customised fragment to compare back with the original QM frequency
analysis. 

I¹ve performed an integrator=cg over my structure, and monitored the
potential energy which converges. The forces also converge beneath the
requested precision (as 0.0001, as per gromacs manual). The message:

‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹

Polak-Ribiere Conjugate Gradients:
   Tolerance (Fmax)   =  1.00000e-04
   Number of steps    =      1000000
   F-max             =  7.35939e+03 on atom 79
   F-Norm            =  1.68505e+03

writing lowest energy coordinates.

Polak-Ribiere Conjugate Gradients converged to Fmax < 0.0001 in 8350 steps
Potential Energy  = -1.67001242395943e+03
Maximum force     =  3.72351263315387e-05 on atom 34
Norm of force     =  1.20696019277311e-05


‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹


When I then come to run integrator=nm I get a maximum force at odds with
the maximum force reported at the final stage of my energy minimisation:

‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹

Maximum force: 6.02183e+02
The force is probably not small enough to ensure that you are at a minimum.
Be aware that negative eigenvalues may occur
when the resulting matrix is diagonalized.

‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹


 
When I decrease the force tolerance in the integrator=cg energy
minimisation to 0.00001 I end up with a poorer force convergence (although
the potential energy is almost the same, but also the integrator=nm will
result in the same measure of maximum force):

‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹

Polak-Ribiere Conjugate Gradients:
   Tolerance (Fmax)   =  1.00000e-05
   Number of steps    =      1000000
   F-max             =  7.35939e+03 on atom 79
   F-Norm            =  1.68505e+03

Energy minimization has stopped, but the forces have not converged to the
requested precision Fmax < 1e-05 (which may not be possible for your
system).
It stopped because the algorithm tried to make a new step whose size was
too
small, or there was no change in the energy since last step. Either way, we
regard the minimization as converged to within the available machine
precision, given your starting configuration and EM parameters.

writing lowest energy coordinates.

Polak-Ribiere Conjugate Gradients converged to machine precision in 9839
steps,
but did not reach the requested Fmax < 1e-05.
Potential Energy  = -1.67001242392665e+03
Maximum force     =  1.63690887039393e-03 on atom 34
Norm of force     =  3.34598700357485e-04
‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹‹



I don¹t want to end up with any imaginary values when I diagonalise the
hessian, any idea how to improve this performance? I am concerned with the
output "Energy minimization has stopped, but the forces have not converged
to the requested precision Fmax < 1e-05 (which may not be possible for
your system).² a a possible indication to the computational limitation of
my machine. 

Many thanks
Anthony