[gmx-users] emtol criterion

[Justin Lemkul]

On 11/27/19 12:39 PM, Christos Deligkaris wrote:
> dear all,
>
> I am trying to decide on a reasonable cutoff for emtol.
>
> The Gromacs manual (section 3.10 on energy minimization) states that a
> reasonable criterion for emtol can be found from the rms force a harmonic
> oscillator would exhibit at room temperature (Equation 3.125). For a weak
> oscillator (100 cm^-1) a force of 7.7 kJ/mol/nm is obtained. This number is
> consistent with the default criterion of 10.0 kJ/nm/mol. It is not clear to
> me how the value of 100cm^-1 was chosen in this example though.

I can't comment directly on this aspect, but a value of 100 cm^-1 is 
about 10x lower than most bond vibrations in biological systems, so this 
would correspond to some slower, low-amplitude motion in, e.g. a 
protein. So in essence, the back-of-the-envelope math suggests forces 
converging below this value would put you at an energy minimum with only 
very low-frequency modes. In practice, this is a very strict criterion 
and often difficult to achieve in mixed precision (and sometimes even in 
double).

> Using a frequency of 1900cm^-1 and 7 amu reduced mass for CO stretching I
> got 122 kJ/mol/nm which appears to be more consistent with the criterion of
> 1000 kJ/nm/mol used in the lysozyme tutorial:
> http://www.mdtutorials.com/gmx/lysozyme/05_EM.html
>
> In the literature I see the number of minimization steps (nsteps) reported
> but not a specific criterion.

In my experience, this generally varies by either (1) software or (2) 
the precision with which the authors report their settings. You can say 
that there were "5000 steps of steepest descent minimization" in 
GROMACS, but more often than not, that's what is set as nsteps but not 
actually used. Other software (CHARMM, AMBER, etc) will gladly perform 
exactly as many steps of SD as you tell it to rather than bailing out 
after a convergence criterion is reached. For GROMACS, I think the emtol 
value should be reported, but this varies by personal preference in most 
papers, unfortunately.

-Justin

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

Justin A. Lemkul, Ph.D.
Assistant Professor
Office: 301 Fralin Hall
Lab: 303 Engel Hall

Virginia Tech Department of Biochemistry
340 West Campus Dr.
Blacksburg, VA 24061

jalemkul at vt.edu | (540) 231-3129
http://www.thelemkullab.com

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