[gmx-users] Problem with incorrect GB-Polarization Energy Value

[Mark Abraham]

On 28/08/2012 2:33 AM, jesmin jahan wrote:
> Dear All,
>
> I am using Gromacs 4.5.3 for GB-Polarization Energy Calculation. As I
> am not interested to any other energy terms right now, I have set all
> the non-bonded parameters to 0.
>
> I am also calculating GB polarization energy using other available
> Molecular Dynamic Packages and doing a comparative study between them
> (say: Accuracy Vs. Speed Up).
> I have already used Gromacs for calculating GB-energy for 168
> different Protein molecules and the energy values reported were more
> or less the same as reported by others.
>
> Now, I am using a virus shell as input in this process. It contains
> 1.5 million atoms. Unfortunately, this time, the energy reported is
> almost half of the value reported by others.
> So, I am a little bit confused. Am I doing something wrong? I have
> heard previously that there is no max size for Gromacs.

Probably you're comparing apples with oranges. Your previous reports 
were choosing to use a cut-off in the non-bonded interactions. Perhaps 
some of the other codes are not or cannot. That would greatly affect 
performance and the value of the result, and is part of why it is 
meaningless to try to extract the time for just a part of a calculation 
without the timing of its support infrastructure (such as constructing 
the pair list from the cut-off). Also, since you are computing on 
proteins, you probably have bonded and VDW interactions in your force 
field also, and your timing comparisons will be misleading if you ignore 
those. The time taken to execute an implementation of a complex 
algorithm is rarely additive in its parts. For example, software that 
makes good or bad use of the cache in a real calculation won't have the 
opportunity to show that if the whole problem size fits in the L1 cache 
because it's been stripped down.

Mark