[gmx-developers] Inconsistency with "foreign lambda" calculations

[Matt Wyczalkowski]

Hi --

I found what I believe is an inconsistency in the results of foreign
lambda calculations, as implemented in the version VERSION
3.3.99_development_20071120.  For a small system where lambda controls
the charges of a peptide (lambda=0 charges off, lambda=1 charges are
OPLS-AA charges), the "foreign" energies as reported in deltaE.edr are
inconsistent with those reported during simulations.

For a given initial configuration of the system, I perform two
different simulations:
(a) init_lambda = 1, deltaE-lambda = 0 1
(b) init_lambda = 0, deltaE-lambda = 0 1

For simulation (a), I get the following potential energies
U = 1552.29 (consistent in log file, energy.edr and deltaE.edr)
dU = -27.39 (this is the energy difference between lambda=1 and
lambda=0, as reported l=0 column of deltaE.edr)

For simulation (b), I get,
U = 1546.42
dU = 27.39  (from l=1 column of deltaE.edr)

Note that U + dU from (a) do not match U from (b), nor vice versa.
Its my understanding that for a given configuration, they should in
fact match; that is, U(a) + dU(a) = U(b), since dU reports on the
difference between the native energy, and what the energy would be for
a different lambda.

I'm attaching a TAR file which reproduces this problem.

Regards,

Matt
-------------- next part --------------
A non-text attachment was scrubbed...
Name: deltaE_test.tar.gz
Type: application/x-gzip
Size: 6117 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20080229/023c4f56/attachment.gz>