[gmx-developers] PBC with p2sub1 from pbc_dx() modifications

[chris.neale at utoronto.ca]

Quoting hessb at mpip-mainz.mpg.de:

> Several years ago I had thought of implementing this,
> but I thought I was too much work.
> 
> pbc_dx is (almost) not used at all during normal simulations in Gromacs.
> pbc for non-bonded interactions is handled by the grid neighbor search,
> there one would have to change the shifts of the i-particles,
> but I wouldn't know of an elegant way to fix the shift forces
> used in the pressure calculation.
> The bonded interactions are handled by the graph code.
> This graph can be circumvented by using pbc=full
> in which case pbc_dx is used for bonded interactions.
> 
> Berk.
> 

Thanks for the pointers. It would have taken me forever to figure that out. It 
looked easy enough in pbc_dx(), so I will give it a try in the proper places. My 
coding is good, but it may take some time to actually find the (i)grid neighbour 
search, (ii)pressure, (iii) graph code, (iv) PME, (v) anything else you can 
think of, routines as this is my first adventure into the gromacs code. If 
anybody knows the names of the functions off-hand it would be a great help. I 
will post the modifications as soon as I have it running = well before any 
publication. I realize that I am fishing for some rather detailed direction. If 
this is not an appropriate type of question then I sincerly appologize.

Background: The image for an angle 90 90 90 and (x=y)!=z system that is in the 
first quadrant and is aligned with the x/y/z axes can be computed as x'=y-box[x
], y'=x, z'=-z, giving a p2sub1 in the second quadrant. This is identical to 
the rotated and shifted primary cell to which was applied (x'=-x,y'=y+1/2box[y],
z'=-z) for charmm and appears to be a necessary difference since gromacs wants 
the x axis aligned with the box[x] vector. Chirality will be maintained. And it 
really should be possible, even if I need to incorporate the first p2sub1 image 
into the primary box. It will still run way faster than my current system-times-
four that I am using to mimic the images and thus statistically but not 
instantaneously reduce the leaflet-register stress. To me it seems like a 
necessary thing to take into consideration when one is studying protein-protein 
interactions in a membrane or docking/pulling a ligand out of an enzyme, etc. 
(especially calculating PMF or similar) in order to avoid stressing both 
leaflets reciprocally in surface tension space.