[gmx-users] Strange behaviour during energy minimization - gmx 3.1

[Marc Baaden]

Hi,

I just witnessed something very stranged (and reproducible)
with Gromacs 3.1 double precision:

When I minimize (steepest descent) my simple organic molecule I get
the following "1-4 interaction" warning

>> Step=    0, Dmax= 1.0e-02 nm, Epot=  1.54212e+03 Fmax= 2.96052e+04,
>> atom= 34
>>   Back Off! I just backed up em_steep1.trr to #em_steep1.trr# Warning:
>> 1-4 interaction at distance larger than 1 These are ignored for the
>> rest of the simulation turn on -debug for more information Step=   34,
>> Dmax= 1.2e-12 nm, Epot=  2.01579e+04 Fmax= 1.19160e+05, atom= 30
>> Stepsize too small (5.82077e-13 nm)Converged to machine precision, but
>> not to the requested precision (0.0001)


Now I try to be a good boy and turn on the "-debug" option, but I get
no more warning now ! Minimization runs fine :

>> Step= 8917, Dmax= 1.1e-12 nm, Epot= -8.31014e+02 Fmax= 8.68832e+00,
>> atom= 38 Stepsize too small (5.51051e-13 nm)Converged to machine
>> precision, but not to the requested precision (0.0001)


Now trying Gromacs 3.0.5 (single prec) without debug yields no error and
converges (to machine precision) after 2659 steps.


3.0.5 (double prec) does not converge during the requested 50000 steps:
>> Step = 50000, Dmax = 5.76e-09 nm, Epot = -8.31273e+02 Fmax = 1.41460e-02 


Any ideas ?
(At first sight I would suspect a problem with the 3.1 version ?)
Shouldn't both versions behave exactly similarly (for a simple
steepest descent minimization) ?

  Marc

-- 
 Dr. Marc Baaden - Laboratory of Molecular Biophysics, Oxford University
 mailto:baaden at smplinux.de  - ICQ# 11466242 -  http://www.marc-baaden.de
 FAX/Voice +49 697912 39550  -  Tel: +44 1865 275380  or  +33 609 843217