From: Floris Buelens (floris_buelens_at_yahoo.com)
Date: Fri Jan 30 2009 - 03:36:00 CST
The implementations in 2.6 and the CVS code should reduce to the same if you apply the same parameters. However there are some discrepancies between the defaults given in the CVS version of the documentation and those applied in the current source code - this should be sorted out soon, but in the meantime, if you want to verify agreement between the CVS and 2.6 implementations, you should to add:
to your configuration files for the CVS code.
Note that although the two codes should sample from the same ensemble with these settings, the results won't be binary identical - i.e. even if you run with the same random seed, the trajectories will diverge after a small number of steps. This is because the lambda-scaled interactions are calculated in a slightly different way and in a different order in the new code. To check agreement, you should therefore monitor convergence over a large number of samples.
More broadly, for any measurement involving the insertion or removal of atoms, the new code (CVS) together with a non-zero value of VdwShiftCoef (e.g. 2) and ElecLambdaStart (e.g. 0.5) is really the only one that should be used. The 2.6 code suffers from an age-old issue that leads to unbounded energies and forces when lambda-coupled atoms interact only very weakly with their environment; this is resolved in the new code.
----- Original Message ----
From: Christopher Hartshorn <cmhartshorn_at_wsu.edu>
Sent: Friday, 30 January, 2009 6:54:54
Subject: namd-l: FEP on NAMD2.6 versus latest CVS NAMD build
Hi all. I am running some tests to decide the optimal lambda differences and nsteps. In the process I noticed that when running the exact same system with the exact same FEP and config file parameters, that I received totally different results between precompiled NAMD 2.6 and the latest CVS build, that I compiled myself. The difference is on the order 30 Kcal. Thus the reason I am curious. Is there a difference between the two versions on how the FEP calculation is performed? If there is then which values would I trust? Old or new?
I might add that the reason for two different versions is that I have access to a linux server (the precompiled NAMD is on this one) and we have several individual 8-core Intel Macs which I found it better to compile myself. Thus, why I am curious because if one is more accurate I will need to compile for the server or revert the Macs to ver.2.6.
Thank you for any help
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:19 CST