[gmx-users] minimization force criteria

[Justin Lemkul]

On 7/26/12 8:30 AM, tarak karmakar wrote:
> Dear All,
>
> While running minimization I imposed the the condition for the
> minimization as to be converged only at Fmax < 10 . But I got the
> following
>
>
> Stepsize too small, or no change in energy.
> Converged to machine precision,
> but not to the requested precision Fmax < 10
>
> Double precision normally gives you higher accuracy.
> You might need to increase your constraint accuracy, or turn
> off constraints alltogether (set constraints = none in mdp file)
>
> Steepest Descents converged to machine precision in 15318 steps,
> but did not reach the requested Fmax < 10.
> Potential Energy  = -3.3397742e+06
> Maximum force     =  9.8232233e+02 on atom 6057
> Norm of force     =  3.6321018e+00
>
>
> Then according to the above instruction I increased the stepsize up to
> 0.5 [ I didn't use any constraint during minimization] but got the
> same. So do I need to increase the stepsize more or is there other
> procedure to get to the desired convergence criteria ?
> The minimization .mdp file is as follows
>

Have you consulted the following?

http://www.gromacs.org/Documentation/Errors#Stepsize_too_small.2c_or_no_change_in_energy._Converged_to_machine_precision.2c_but_not_to_the_requested_precision

It may simply not be possible to energy minimize the system further without 
using double precision or a better EM algorithm.  Steepest descents converges 
very quickly but the results can often be improved by additional rounds of CG or 
L-BGFS minimization.  For most purposes, one does not require such a low maximum 
force, but if you need Fmax < 10, then you likely need a significantly more 
robust protocol.

-Justin

>
> ; 7.3.2 Preprocessing
> ;define                  = -DFLEXIBLE    ; defines to pass to the preprocessor
>
> ; 7.3.3 Run Control
> integrator              = steep         ; steepest descents energy minimization
> nsteps                  = 100000         ; maximum number of steps to integrate
>
> ; 7.3.5 Energy Minimization
> emtol                   = 10            ; [kJ/mol/nm] minimization is
> converged when max force is < emtol
> emstep                  = 0.01           ; [nm] initial step-size
> nstcgsteep              = 10            ;  freq of performing one
> steepest descent step
>
> ; 7.3.8 Output Control
> nstxout                 = 500           ; [steps] freq to write
> coordinates to trajectory
> nstvout                 = 500           ; [steps] freq to write
> velocities to trajectory
> nstfout                 = 500           ; [steps] freq to write forces
> to trajectory
> nstlog                  = 100           ; [steps] freq to write
> energies to log file
> nstenergy               = 100           ; [steps] freq to write
> energies to energy file
> energygrps              = System        ; group(s) to write to energy file
>
> ; 7.3.9 Neighbor Searching
> nstlist                 = 1             ; [steps] freq to update neighbor list
> ns_type                 = grid          ; method of updating neighbor list
> pbc                     = xyz           ; periodic boundary conditions
> in all directions
> rlist                   = 0.8           ; [nm] cut-off distance for
> the short-range neighbor list
>
> ; 7.3.10 Electrostatics
> coulombtype             = PME           ; Particle-Mesh Ewald electrostatics
> rcoulomb                = 0.8           ; [nm] distance for Coulomb cut-off
>
> ; 7.3.11 VdW
> vdwtype                 = cut-off       ; twin-range cut-off with
> rlist where rvdw >= rlist
> rvdw                    = 0.8           ; [nm] distance for LJ cut-off
> DispCorr                = Ener          ; apply long range dispersion
> corrections for energy
>
> ; 7.3.13 Ewald
> fourierspacing          = 0.12          ; [nm] grid spacing for FFT
> grid when using PME
> pme_order               = 4             ; interpolation order for PME, 4 = cubic
> ewald_rtol              = 1e-5          ; relative strength of
> Ewald-shifted potential at rcoulomb
>
>
> Thanks,
>
> Tarak
>

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================