[gmx-users] How to set fourierspacing?

[Mark Abraham]

wuxiao wrote:
> Dear colleagues,
>   I use PME to consider coulomb interactions. It seems that the 
> fourierspacing is an important parameter. When it is set a larger value, 
> the calculation can be much faster. But what is the disadvantages of 
> using a larger value? I wish to get you help.

Faster, yes, but also less accurate. There's a complex interaction of
* your hardware,
* your FFT library,
* the GROMACS kernel performance,
* rcoulomb,
* ewald_rtol,
* fourierspacing,
* pme_order and
* (in parallel) your number of PME nodes
that determines the accuracy and speed of the PME approximation to the
full electrostatic interactions. Carsten Kutzner has previously posted
to this list some heuristics for varying fourierspacing. Search for them.

Mark