[gmx-users] Buckingham Potential

[Mark Abraham]

Andrew Paluch wrote:
> To whom this may concern,
> 
> I am performing simulations for a system that is modeled with a 
> Buckingham potential and PME for the electrostatic interactions.  I have 
> developed my forcefield file such that non-bonded cross terms are listed 
> in a table, and also 1-4 interactions are listed in a table with an LJ 
> type of potential.

Be aware that "table lookups" for non-bonded interactions refer to 
tables in which GROMACS does lookups of complex functions of r for 
non-bonded interactions. See various manual sections for details. I'm 
not sure to what your usage of "table" refers.

> On 16 processors, I am able to run my system for 0.5ns in a day.  If I 
> map the potential to an LJ, I am able to achieve 2.5ns in a day.  Does 
> anyone have any idea as to where my performance loss may be occurring?   
> Are there any special compilation or run-time flags that I should be using?

Buckingham requires the evaluation of an exponential function in r, 
which is presumably noticeably more costly than

r_twelfth = r_sixth * r_sixth;

So, if you are not using table lookups, this will be considerably 
slower. Even if you were using VDW table lookups, that will also be 
slower than LJ.

Mark