From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Wed Dec 09 2009 - 10:25:14 CST
Indeed, so far it has not seemed necessary to implement this in the
new code. The benefits were dubious: the "deleterious effects" we
mentioned at the time were mostly hypothetical, and since then we have
found no clear evidence of them (our 2005 paper was not conclusive on
this, apart from the expected result that too much smoothing is
harmful). As a matter of fact, averaging of force data from
neighbouring bins does effectively occur over time, as the coordinate
diffuses across bin boundaries. We have not found short-timescale
variations of the biasing force per se to cause numerical problems. At
some point, we all but stopped using dSmooth.
So for the new code, I decided to focus on basic functionality, and
wait to see how much demand there would be for smoothing before I
implemented it. There has not been any demand so far... Let me know if
you have reasons to think it is needed.
2009/12/9 Hugh Martin <hughtendo_at_gmail.com>:
> In the NAMD2.7b1 implementation of ABF, I notice that neither the "dSmooth"
> option, nor an equivalent, is present. It is also not included in the
> revised tutorial. It's purpose was initially:
> "to alleviate the deleterious effects due to abrupt variations of the force,
> the corresponding ï¬‚uctuations are smoothed out, using a weighted running
> average over a present number of adjacent bins, in lieu of the average on
> the current bin"
> Does the new implementation avoid these effects without the 'dSmooth'
> Many thanks,
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:34 CST