Dave -
I will probably do TI this way. I have been thinking about a
"multi-pmemd"
for my own purposes (I don't particularly want to get into replica
exchange,
etc., but I do potentially want to be able to queue up one job that uses a
pile of processors to handle a series of tasks - like maybe 1000
minimizations or what have you).
Regards - Bob
----- Original Message -----
From: "David A. Case" <case.scripps.edu>
To: <amber-developers.scripps.edu>
Sent: Wednesday, June 29, 2005 4:44 PM
Subject: amber-developers: new method for free energy calculations in
Amber
9
> FYI:
>
> I have turned on some new code in Amber 9 for doing free energy
> calculations
> with thermodynamic integration.
>
> The new method uses multisander to run two groups (each group itself
have
> several cpu's). The first group computes the lambda=0 energy and
forces,
> and
> the second group computes the same for lambda=1. There is no longer any
> need
> for a "pert" prmtop file: each group reads a regular prmtop
corresponding
> to
> one of the endpoints.
>
> Then, at each step, the energies and forces and communicated and
combined
> to give the actualy Hamiltonian.
>
> The advantage is that there is no need for a perturbed prmtop file,
which
> should eliminate lots of confusion. You just create two prmtop files
for
> the
> endpoints (which you can test and interrogate in the usual way). When
you
> are
> satisfied with the endpoints, you can compute the free energy difference
> between them. Also, there is more flexibility about choosing the
states,
> since the current code has only a single bond-list for both states.
This
> means that you cannot convert a 1-3 interaction (say) into a 1-4
> interaction
> in any simple fashion.
>
> The two prmtop files *must* have the same number of atoms in the same
> order;
> the coordinates on the two processes are kept synchronized. Also, a
minor
> disadvantage is that you must be running MPI to do free energy
> calculations.
> But I think the advantages will greatly outweigh the disadvantages.
>
> If you want the old way of doing things, add -DOLD_TI to your AMBERBUILD
> flags. But the old code may go away sometime in the future. Also, I
have
> turned off the old tests for TI in amber9/test/Makefile, since they no
> longer
> work. But I have not yet inserted new test cases: that should come
soon.
>
> Most people should not have to do anything, other than perhaps turning
on
> OLD_TI if you want to do free energy calcs in sander9 without risking
> using
> the new code.
>
> ...dac
>
>
Received on Wed Apr 05 2006 - 23:49:55 PDT
