[gmx-users] more detailed problem from 3.2.1 to 3.3 version

[Mark Abraham]

Adriana Pietropaolo wrote:
> On Wed, 8 Mar 2006, David van der Spoel wrote:
> 
> 
>>Adriana Pietropaolo wrote:
>>
>>>>did you use pme_order=8? Before 3.3.1, you shall use 4.
>>>>
>>>>Yang Ye
>>>
>>>
>>>yes I used 8, as in the 3.2.1, so I have to use 4?
>>>ok, thank you very much!
>>>I'll try...
>>>
>>
>>this is a known bug. a fix is available on the ftp site.
>>
>>
> 
> Hi all,
> with the pme order 4 is now ok,
> but why this? 
> 8 is better than 4,or not? 

Eight is different from four. It could be more expensive for little or 
no gain, or it could be critical to get a reasonable error, depending on 
the other parameters.

> this order is related to spline 
> interpolation, is right?

It describes the order of the polynomial used to interpolate between 
points on the FFT grid. Brief description in the manual.

There is a complex interplay between rcoulomb, ewald_rtol, pme_order, 
fourierspacing/fourier_n?, accuracy of the total electrostatic forces 
and computational cost. To choose these other than randomly, you will 
need to read the smooth PME papers (e.g. Essman) and some subsequent 
ones that describe how to measure the errors in forces, Petersen, I 
think. Then you will need to experiment for your system.

Mark