On 5/06/2012 6:46 PM, Laurence Leherte wrote:
Hello,
Thank you very much for your reply. I actually carried out very basic
MDs of a peptide in vacuum (no pbc, cut-off for electrostatics and vdw
= "infinity", niter = 10^6). The computing results are given in the
two tables below. As they are identical, I am assuming that there is
only one neighbor list that is created for the calculation of both the
electrostatic and vdW interactions.
Those settings trigger a different kind of non-bonded kernel from ones
normally used in explicit solvent calculations. These kernels apparently
don't bother to construct Q-only or LJ-only versions. It's probably not
worthwhile for only a handful of atoms in the whole system.
Since I want to calculate all vdW terms, I suppose that all Coulomb
terms are automatically calculated too. Am I right ?
What if I create a group with all the nul-charge atoms and, by some
means (e.g., a different fudgeQQ value - or something else- for that
group ???), avoid to compute the electrostatic interactions with all
the other charges particules ?
Similarly to the problem I reported in my mail, I also want to add
some kind of charged virtual sites. In that case, those sites should
not contribute to the vdW interactions of the system. I found no
problem at all to define such virtual sites and to carry out MD
simulations, but I suspect vdW terms to be calculated even if epsilon
and sigma are set equal to zero for such "particles".
Yes, but the whole thing is so cheap it's not worth thinking about.
Mark
Thank you in advance for any help
Laurence
For the regular MD (all atoms bear a non-zero charge) :
Computing: M-Number M-Flops % Flops
-----------------------------------------------------------------------------
All-vs-All, Coul + LJ 19404.019404 737352.737 72.1
Outer nonbonded loop 197.000197 1970.002 0.2
1,4 nonbonded interactions 518.000518 46620.047 4.6
Bonds 105.000105 6195.006 0.6
Angles 363.000363 60984.061 6.0
Propers 674.000674 154346.154 15.1
Virial 24.200242 435.604 0.0
Stop-CM 19.700197 197.002 0.0
Calc-Ekin 197.000394 5319.011 0.5
Lincs 96.000288 5760.017 0.6
Lincs-Mat 468.001404 1872.006 0.2
Constraint-V 192.000384 1536.003 0.2
Constraint-Vir 9.600096 230.402 0.0
-----------------------------------------------------------------------------
Total 1022818.053 100.0
For the modified system (most of the atoms, i.e. 169 over 197, bear a
nul charge) :
Computing: M-Number M-Flops % Flops
-----------------------------------------------------------------------------
All-vs-All, Coul + LJ 19404.019404 737352.737 72.1
Outer nonbonded loop 197.000197 1970.002 0.2
1,4 nonbonded interactions 518.000518 46620.047 4.6
Bonds 105.000105 6195.006 0.6
Angles 363.000363 60984.061 6.0
Propers 674.000674 154346.154 15.1
Virial 24.200242 435.604 0.0
Stop-CM 19.700197 197.002 0.0
Calc-Ekin 197.000394 5319.011 0.5
Lincs 96.000288 5760.017 0.6
Lincs-Mat 468.001404 1872.006 0.2
Constraint-V 192.000384 1536.003 0.2
Constraint-Vir 9.600096 230.402 0.0
-----------------------------------------------------------------------------
Total 1022818.053 100.0
-----------------------------------------------------------------------------
On 04/06/2012 16:37, Mark Abraham wrote:
On 5/06/2012 12:08 AM, Laurence Leherte wrote:
Dear Gromacs users,
I am using the Amber99 FF in MD simulations of peptides (and
proteins). In a first stage to the design a different charge
distribution, most of the atomic charges were set equal to zero
(i.e., all charges but the C and O backbone atoms).
It appeared that the calculation times observed for the original
all-atom charges and the modified system are similar.
My question is thus the following one. In order to save calculation
time (and whatever the FF is), how is it possible to avoid that the
atoms bearing a nul charge are considered in electrostatic
calculations ? I should specify here that I want these atoms to be
considered in the vdW non-bonding interactions.
IIRC GROMACS neighbour searching already identifies atoms with zero
charge and/or LJ parameters and uses non-bonded code that does not
compute contributions that are known to be zero. You should be able
to see this from the differences in the flop accounting at the end of
your .log files when you have different numbers of zero-charge atoms.
If the total calculation times are similar, then the number of atoms
for which time was saved was negligible. This would be normal for a
peptide in a much larger quantity of water. You will have to judge
the truth of this from the timing and flop breakdown at the end of
the .log file.
Mark
--
Laurence Leherte
Laboratoire de Physico-Chimie Informatique
Unité de Chimie Physique Théorique et Structurale
Facultés Universitaires Notre-Dame de la Paix (University of Namur)
Rue de Bruxelles, 61
B-5000 Namur
Belgique (Belgium)
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at
http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
