[gmx-users] stepsize too small ... but potential energy negative!

Luca Mollica luca.mollica at ibs.fr
Mon May 24 13:52:49 CEST 2010

>> - my system is a protein (with or w/o ligand) in solvent (water SPC).
>> Following your suggestions, I tried to perform an EM on the protein w/o
>> ligand after the editconf step (i.e. I created the topology with 
>> pdb2gmx,
>> created a cubic box with editconf, then I used grompp+mdrun to 
>> perform EM).
>> I received the same error: the minimization starts, performs 37 
>> steps, then
>> stops with the "stepsize too small...." communication. The energy in 
>> this
>> case is -3.06e+4 (one order of magnitude higher than the solvated 
>> system)
>> but still negative, and the maximum force is still "inf on atom 1". So
>> minimizing in vacuo does not help to solve the problem. This atom 1 
>> is the
>> Nter of the protein, it is belonging to a Ser and I did not charge it
>> explicitly with pdb2gmx (i.e. I did not use the flag -ter) but it is 
>> bound
>> to 3 H in the .gro file.
> Were there any messages from pdb2gmx when you produced your topology?  
> How large is your box?  What force field are you using?

The choice of FF and different outcome of GMX could address the problem 
in more precise way.
And yes, does pdb2gmx plainly goes till the end of topology generation ?
Moreover, did you try a larger box, even at great extent with respect to 
protein ?
Are there any warning during the "grompping" step ? A mismatch of atoms 
in principle could lead to such errors at the beginning of calculations.

>> - If I look at the protein with VMD or other visualizing tools, it 
>> seems to
>> me that no major problems are present on the structure. In 
>> particular, atom
>> 1 is not "broken" or something similar, I can't see no "aberrations". I
>> don't know how to check if something goes wrong apart from 
>> visualization, do
>> you know some tools that could automatically check the file? To the 
>> best of
>> my knowledge, gmxcheck performs checks only on trajectories, or can I 
>> use it
>> also on single structures?

VMD could not necesserely tell the truth ... :).
It could be not a problem of visualization.

>> - I also tried to continue anyway after the minimization step with 
>> the PR MD
>> in NVT, but it did not start because of a lot of LINCS error. It 
>> suggests me
>> that some distortions are present in my structure, but if I cannot 
>> minimize
>> it, how to relieve this problem?
> If a system does not minimize, any further steps will be futile.

Minimization looks for a local minimum: maybe some troubles in LINCS 
derive from some problems related to uncorrect distributions of dihedral 
angles or sidechain stereochemistry that
a steepest descent could not properly correct. In such a case, I guess 
Modeller (or whatever) could have generated a structure that is really 
far away from local minima.
Of course, the correct procedure is a. minimization b. positional 
restrained MD c. production run : however, you could try a workaround of 
the problem setting up a short low temperature
simulation of the protein as-it-is in vacuo (few ps at few K degrees, 
for example) and see what happens. In this way you are forcing the 
protein to get in action but with variations that could not (in principle)
result in a so verbose output of LINCS (i.e., it could result in a 
system that does not violate very much the LINCS "rules"). If the thing 
goes fine, it will not be the cleanest way to proced, however you could 
argue which the problem could be, try to fix it and go back to and 
through the standard procedure.
It's a "dirty" suggestion, I know, but at this point it's worth to try ...

> <snip>
>>         G96Bond       G96Angle    Proper Dih.  Improper Dih.          
>> LJ-14
>>     2.40716e+03    1.23569e+03    6.85831e+02    8.70819e+01   
>> 2.11992e-314
>>      Coulomb-14        LJ (SR)        LJ (LR)   Coulomb (SR)   Coul. 
>> recip.
>>    2.11992e-314   -2.19860e+03   -1.52663e+02   -6.70307e+03   
>> -2.59368e+04
>>       Potential Pressure (bar)
>>    -3.05754e+04            nan
> The fact that your LJ-14 and Coulomb-14 energies are essentially 
> numerical infinity (!) suggests some very bad underlying problem.  
> Hard to say at this point what it is though.
> Other useful information would be the actual .mdp file you're using, 
> and answers to my questions above.

Yes, the mdp file at this point is essential, even if the preciously 
attached text should include all the things used and required for better 



More information about the gromacs.org_gmx-users mailing list