AMBER Archive (2009)
Subject: Re: [AMBER] Partial interactions
From: Ignacio J. General (ijgeneral_at_gmail.com)
Date: Fri Apr 24 2009 - 09:49:07 CDT
Hi Thomas. Thanks very much for your answers. I think I now understand how
TI is implemented in Amber.
If you don't mind, I still have a couple of questions, now on the
application part: I need to calculate the free energy of some small system
when it dissapears. I have started to do this in the following way:
a) Use ifsc=1, defining V0 as the entire system and V1 as only one atom of
the original ones (so that I don't have an empty prmtop file which I think
Amber can't handle).
b) After starting doing 1), I read something about using ifsc=2, another
approach to make a system dissapear.
My questions are:
1) From the implementation point of view, are both approaches equivalent?
2) In approach a), and as it is usualy done, I first make the charges
dissapear, and in a second step I use softcore potentials. In this second
step, Sander complaints if I try to use what sounds reasonable to me, ifsc=1
for V0 (the dissapearing system) and ifsc=0 for V1 (where nothing appear nor
dissapear). So I imagine I just have to set ifsc=1 on both, even when V1
doesn't really use soft core potentials. Is this right?
On Wed, Apr 22, 2009 at 10:55 PM, <steinbrt_at_rci.rutgers.edu> wrote:
> Hi Ignacio,
> I am not sure if you mean the actual code implementation here or the
> resulting model. As for the later, Amber applies the mixed Hamiltonian
> V(L) at every single MD step and propagates the system with forces
> resulting from it. In the end, there are no 'two' MD simulations, just one
> at the given value of Lambda. Also, V(L) is not just linear mixing as you
> write under 1) because the softcore potentials are highly non-linear in
> As far as the code implementation goes, two independent processes compute
> the forces & energies according to V0 and V1, from which the actual
> forces, energies and DVDL at the given lambda value are computed in
> thermo_int.f. This is complicated a bit if softcore potentials are used,
> because then some bonded interactions are removed from the Amber energy
> array, not scaled and not contributing to DVDL, plus there is some extra
> code in short_ene.f and ew_directe3/4.h to handle the new potential form.
> But the important thing is, the processes communicate the mixed forces and
> energies and their coordinates progress in sync. So in the end both
> processes propagate copies of the same simulation (producing some
> superfluous output, but making for an easy implementation).
> Kind Regards,
> On Wed, April 22, 2009 3:02 pm, Ignacio J. General wrote:
> > Thanks for your reply Thomas. I think I'll turn to Tinker since I will
> > probably need to do several changes in the code in the near future.
> > Before doing that, I still want to do some TI using Amber. But there are
> > few details in the workings of Amber on TI, that I am not sure I
> > understand. So, can you please tell me if the following procedure is the
> > one
> > used by Amber:
> > (L means lambda, DL means delta lambda)
> > 1) V_L = (1-L)V0 + L V1. This is the new Hamiltonian.
> > 2) Run an MD of system V0, with forces scaled by L.
> > 3) Run an MD of system V1, with forces scaled by (1-L).
> > The simulations above are independent of each other.
> > 4) At a given step in the runs, calculate E_V0 - E_V1 m, divide by DL.
> > This
> > is the dv/dl that appears in the *.out file.
> > Questions:
> > 1) L (and 1-L) are applied to every force (on atoms in corresponding
> > masks),
> > or just to Coulomb and LJ forces?
> > 2) The evolution of each system, V0 and V1, in their own simulation, are
> > guided only by (1-L)V0 or LV1, respectively, right? In other words, the
> > Hamiltonian for each simulation is not the whole Hamiltonian V_L, right?
> Dr. Thomas Steinbrecher
> BioMaps Institute
> Rutgers University
> 610 Taylor Rd.
> Piscataway, NJ 08854
> AMBER mailing list
AMBER mailing list