[gmx-developers] more integration

[Michael Shirts]

> Subtracting positions leads to the same Ekin problem I have mentioned 
> before. This causes the discretization error in Ekin to be 4 times larger than 
> it is now in the CVS code.

The simple solution, then, is to just use SHAKE/RATTLE.  Might be a
bit slower than LINCS in some systems, but I rarely have come across
that many systems where the restraints were the speed bottleneck.

> I don't think this is a serious issue.
> Atomic virials just cause a bit more noise on the virial.

This may very well be the case, but though I read something about this
in one of the Tuckemanm/Martyna papers, but I could be misremembering.
 I'll keep investigating.  One thing all the publications on this
subject have demonstrated -- this whole thing is very, very tricky!

Best,
Michael