[gmx-users] grompp single vs double precisions

[Yang Ye]

Mark Abraham wrote:
> Yang Ye wrote:
>> rob yang wrote:
>>
>>> hi list,
>>> i am comparing the relative energy differences between different 
>>> conformations of a same molecule in vaccuum. i am using both gromacs 
>>> and another software to compare the results. the forcefield for both 
>>> softwares is oplsaa (2001), and i've manually verified that they are 
>>> indeed the same for my molecule as well as the partial charges.
>>> the relative energy results obtained from these 2 softwares are 
>>> different. i understand that different softwares have different 
>>> implementations but i think i should be able to get a similiar 
>>> relative energies for these conformations.
>>> so i am thinking maybe the minimizations is where things went wrong. 
>>> i have compiled the double version of mdrun. but i didn't for grompp 
>>> since i didn't think it was necessary. the question i have for you 
>>> guys is then: is the double version of grompp also necessary?
>>
>> I didn't see your problem just now. For your case, definitely you need
>> double version for grompp to the result to be more precise.
>
> If he's noticing a difference in the relative energy between two 
> configurations calculated with gromacs and two configurations 
> calculated with another program, why will the precision to which 
> gromacs is calculating the number "definitely" be relevant? One 
> molecule in a vacuum can't accumulate enough round-off error in 
> calculating energies and forces for this to be important, in my opinion.
>
> If your point is that you need to use double grompp to make double 
> mdrun effective, then I'll take your word for it. :-)
grompp generates a tpr at the precision that grompp is compiled. My 
concern whether the difference introduced by single-precision tpr has 
made the starting point energy different, especially with automatic LJ 
and pair interaction generation.

Of course, for comparison, one shall try both single and double. 
Because, simply, we will choose one precision for production running. 
There shall be some conclusion on for what purpose, use what precision 
(could be platform-dependent).
>
> Mark
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the www 
> interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>
Yang Ye