[gmx-users] Again: Non-integer charges in tpr-files

[Marc Saric]

Hi all,

I bothered you approx. two weeks ago with a problem I have with grompp 
(or my input to grompp):

If I create a inputfile for GROMACS 3.1.4 (single-precision) on a SGI 
IRIS 6.5) or Linux (Kernel 2.4.18) machine with the following commands:

pdb2gmx -f protein.pdb -p protein.top -sort -ter -o protein.gro

I used the GROMACS forcefield (0) set all termini to 0

editconf -bt dodecahedron -f protein.gro -c -d 0.7 -o
genbox -cp out -cs -p protein.top -o protein.gro
grompp -v -f em.mdp -c protein.gro -o em -p protein.top

em.mdp looks like this:

title               =  protein
cpp                 =  /lib/cpp
define              =  -DFLEX_SPC
constraints         =  none
integrator          =  steep
dt                  =  0.002    ; ps !
nsteps              =  1000
nstlist             =  10
ns_type             =  grid
rlist               =  0.9
coulombtype             =  PME
rcoulomb            =  0.9
rvdw                =  0.9
fourierspacing          =  0.12
fourier_nx              =  0
fourier_ny              =  0
fourier_nz              =  0
pme_order               =  4
ewald_rtol              =  1e-5
optimize_fft            =  yes
;
;       Energy minimizing stuff
;
emtol               =  1000.0
emstep              =  0.01


I always get a warning about non-zero charges from grompp (which is ok 
in general, as one can add counter-ions). The problem is: The charge is 
non-integer (WARNING 1 [file "protein.top", line 41]:
   System has non-zero total charge: -2.399999e+01) which cannot be a 
rounding-error.

I carefully checked the input-files (pdb-file in spdbv and by hand) and 
could not find an error. All atoms are there, all chains are properly 
terminated, there are no unusual or nonstandard aminoacids in them, the 
atom-nomenclature is correct, the coordinates are reasonably similar to 
ideal bondlenght/angles, there are no severe clashes, all hydrogens have 
been removed prior to conversion, etc.

David v.d. Spoel gave me the advice to check the GROMACS topology, which 
I did, but I found no non-integer charges after adding up all 
qtot-values as reported in the top-file (or itp if you have more than 
one strand) either per chargegroup or in total.

I therefore dumped the tpr-file with gmxdump and had a look at the 
charges reported there ("q", "qB", anything else???) but did not find 
any unusual numbers, neither in total nor in single chargegroups (except 
5 or so groups which seemed to have rounding errors (non-zero total 
charges of order 5e-17)).

So: *What is going wrong here?????* What does grompp do beside packing 
all information into one single binary file? Why does this charge-error 
does not show up in the decompiled tpr-file (or have I overlooked it 
-but where)?

I would be very thankfull if someone could answer this question!

Thanks for reading so far. ;-)

-- 
Bye,  Marc Saric

Max-Planck-Institut fuer molekulare Physiologie
Otto-Hahn-Str.11  44227 Dortmund  phone:0231/133-2168