[gmx-users] Coulomb LR - PME vs. Cut-off

[Mark Abraham]

On 7/04/2010 1:15 AM, Marc Charendoff wrote:
> Hello,
> I am currently running MD on a protein-ligand complex centered in an
> (approx) 70.0 Angstrom ^3 water box with sodium counterions. I have run
> 1.0 ns simulations under two different conditions. The first condition
> with coulombtype "cut-off" I have run to perform LIE calcs as I have
> read PME will cause undesired artifacts. These runs tend to be unstable
> though with only varying success on obtaining completed production runs
> before box collapses.

Using plain cutoffs causes massive artefacts. This doesn't make for good 
inputs to the (fairly crude) LIE approximation. That said, if you've 
copied simulation conditions under which your force field was 
parameterized or has been found to be reliable, then perhaps you could 
benefit from more solvent and/or a better equilibration. See the 
standard advice here 
http://www.gromacs.org/Documentation/Terminology/Blowing_Up

> However, when I run under condition 2 (PME) I find
> that there is no problem at all, though I cannot then run g_lie
> (obviously). Would creating a bigger solvent box help in making the use
> of cut-offs more stable? Is there not a preferred ratio of rcoulomb
> distance to some solvent box dimension that might alleviate this issue?

No. The model physics should match the parameterization (or other 
known-good) conditions.

> Otherwise, what is it about PME exactly that creates issues for g_lie?
> Is there a way around it? Any guidance would be appreciated.

You did read about PME before using it, right? Then you'd know it 
generates an approximation to the full infinite periodic Coulomb sum 
which includes a *modified* term that is a sum over a function of 
distance, and another term that is evaluated in Fourier space. The 
former probably makes LIE more unreliable even if you neglected the 
latter, and the latter by its very nature cannot be decomposed into 
contributions from real-space terms. I've lost track of the number of 
times I've said this on this list regarding LIE :-)

Mark

> Regards, Marc
> *Condition 1*
>
> ns_type = grid
>
> pbc = xyz
>
> rlist = 0.9
>
> coulombtype = cut-off
>
> rcoulomb = 1.2
>
> vdwtype = cut-off
>
> rvdw = 1.4
>
> *Condition 2*
>
> ns_type = grid
>
> rlist = 0.9
>
> coulombtype = PME
>
> rcoulomb = 0.9
>
> vdwtype = cut-off
>
> rvdw = 1.4
>
> fourierspacing = 0.12
>
> fourier_nx = 0
>
> fourier_ny = 0
>
> fourier_nz = 0
>
> pme_order = 6
>
> ewald_rtol = 1e-5
>
> optimize_fft = yes
>