[gmx-users] problem going from cut-off to pme

[Mark Abraham]

Yiannis wrote:
> Le 8 mars 06 à 20:35, Mark Abraham a écrit :
> 
>> Yiannis wrote:
>>
>>> Well, my initial guess was something wrong with the structure.  But  
>>> the same gro file gives a correct minimisation with cut-off.
>>> So I tried a 0 step minimisation to see which component is  
>>> different.  Here it is:
>>> ==> cutoff.xvg <==
>>> @ s0 legend "Coulomb (SR)"
>>>     0.000000  -7838.336914
>>> ==> pme4.xvg <==
>>> @ s0 legend "Coulomb (SR)"
>>> @ s1 legend "Coul. recip."
>>>     0.000000  -8141383915228874407936.000000  -9205.502930
>>> Obviously the Coulomb(SR) is aberrant. The only line I changed   
>>> between the two mdp is:
>>> coulombtype              =  cut-off
>>> to
>>> coulombtype              = pme
>>> I don't understand why the Coulomb SR is different and even more  
>>> why  it is illogical.
>>> Could it be pme parameters? I keep the defaults (pme=4 or 8 does   
>>> change a little the reciprocal component but not the SR).
>>
>>
>> Possibly. Please give us the .mdp file. You may have only changed  
>> coulombtype, but there could be other junk there depending where  you 
>> got the original from.
> 
> 
> OK, I attach it to the mail. I got it from somebody who asks me what  to 
> do and I told him I'll ask the list. Something else I wanted to  check 
> is the charge groups as it is a zwiterionic molecule (and the  contacts 
> as Yang Ye suggests).

Yes, do that as well. In the meantime, your integrator is md which isn't 
useful for attempting energy minimization.

>> pme_order != 4 is broken in version 3.3, as discussed many times on  
>> this mailing list. There's a fix to src/mdlib/pme.c on the gromacs  
>> ftp site, or you can wait for the upcoming 3.3.1 release.
> 
> 
> I missed that info, thanks. Anyway, it was set to pme=4, the pme=8  was 
> just a test.

Sure.

>> The Coulomb SR would necessarily be different between cut-off and  
>> PME. Changing only the PME interpolation order necessarily will  only 
>> change the reciprocal-space component. The sum of SR+recip for  PME 
>> will be different from the SR for cut-off. If you don't  understand 
>> why any of these happen, then you need to go and read  some basic 
>> stuff explaining the two methods - the gromacs manual is  a good place 
>> to start.
> 
> 
> I will do some reading as I understand the second and third assesment  
> but not the first one...

The Ewald beta parameter (controlled indirectly in gromacs through 
ewald_rtol) changes the relative weight and accuracy of the direct and 
reciprocal space components in a contrary manner. Sections 4.1.6 and 4.9 
of the manual describe this briefly. There will exist a value that gives 
numerical agreement of SR between PME and Cut-off, but it doesn't mean 
anything.

Mark