[gmx-users] Force Calculation

[Mark Abraham]

Justin M. Shorb wrote:
> Greetings!
> 
> I have managed to hack through all the Gromacs code and at the bottom of 
> it all are machine-run look up tables. Does anyone know of where I can 
> find the organization of these tables and the way they are called 
> through the program? Or possibly, if anybody has written their own code 
> to extract data from the look-up tables on the fly? Luckily, Gromacs has 
> already sorted out the different contributions to the forces so I think 
> I can pull out the data I need, but I'm afraid I don't even know what 
> sort of files (or code) the table functions are written in.

cd $GROMACS/src; grep table */*.[ch] | grep make

> Where I've gotten is to: /src/gmxlib/nonbonded/nonbonded.c  which calls 
> the machine specific kernels. I am more than willing to work through 
> this myself, but can anybody point me to a good book/site that explains 
> what is actually going on when Gromacs looks up these forces?

Try reading src/gmxlib/nonbonded/nb_kernel/nb_kernel310_c.c This is 
auto-generated code, so it isn't commented, but it illustrates the table 
lookups in non-optimized code.

> Sorry for the imprecise language! Basically I just need to extract the 
> SR electrostatic force from the look-up tables on the fly, and can't 
> find any references out there that explain what goes on in nonbonded.c

The look-up tables are computed at the start of the runs for nonbonded 
interactions that have requested them. There's nothing magical about the 
mathematical functions that fill them - they're documented in the manual.

It's very unclear what you want SR electrostatic force between - pairs 
of atoms? Groups? Is computing them yourself an easier way to get the 
forces you want?

Mark