Re: [AMBER] Lambda Exchange

From: Jason Swails <>
Date: Wed, 08 Oct 2014 14:30:40 -0400

On Wed, 2014-10-08 at 12:23 -0500, Jeff Wereszczynski wrote:
> Hi Jason,
> Thanks for your response. To follow up:
> 1. I can surely edit the prmtop files as you suggest. After I sent
> this email it occurred to me that this will of course modify not only
> the inter-molecular electrostatics, but also the intra-molecular
> electrostatics. Is there a straightforward way to keep the
> intra-molecular electrostatics on while scaling the energy terms
> between the ligand and the rest of the system?

No, but why is this necessary? I think the thermodynamics works out
fine if the charges scale completely (I actually don't know if that's
true for your proposed scheme). In fact, with PME this would be very
difficult to accomplish anyway, since the reciprocal sum is not
decomposable this way at all. (Effectively what you'd have to do is
subtract out the reciprocal space contribution that's "wrong" and add
back in the missing part of the direct-space -- akin to how the
self-interaction energy and particle exclusions are done).
> 2. You said "to use softcore TI," however I can't use the normal TI
> commands with H-REMD (this is the root of most of my issues). When I
> try to do this, I get the following error:
> | ERROR: TI is incompatible with REMD

Ah, then I see the confusion. :) I was not aware that error check was
there. This is, I believe, a side effect of disjoint, over-cautious
development efforts by graduate students. Joe Kaus wrote the TI
implementation while I wrote the REMD implementation (~1 yr. earlier).

So one of the following two things happened:

1) Joe was uncertain whether TI and REMD would work together in pmemd,
since they didn't in sander. So he erred on the side of caution and
prevented them from running together.

2) Joe knows something I don't. Namely, exactly why the two are
incompatible in pmemd.

So what I would suggest is disable this warning (comment that part of
the code out in $AMBERHOME/src/pmemd/src/mdin_ctrl_dat.F90 and
recompile). Just know that you are embarking into uncharted waters
here. Waters I've wanted to investigate myself, but just haven't found
the time.

I don't see anything glaringly wrong with your input (you may need more
than 2 lambda replicas -- you'll have to have a look at the exchange
success rate to tell that).

You can also use multi-dimensional REMD with H-REMD in one dimension
(the lambda dimension) and T-REMD in the other... Just a thought.

Good luck!

P.S. I would be interested to hear any successes or challenges you have
with this approach.

Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Wed Oct 08 2014 - 12:00:02 PDT
Custom Search