From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Wed Jan 05 2011 - 12:07:39 CST
On 5 January 2011 18:13, BIN ZHANG <zhngbn_at_gmail.com> wrote:
> Dear Jerome:
>
> Thanks a lot for your response.
>>
>> For convergence analysis, we need a histogram of deltaU_ba
>> sampled at point b, which we don't have. That is the job of the
>> backwards calculation. (In the old linear paradigm, that was not
>> necessary because there was a simple, linear relationship between
>> deltaU_ba and deltaU_bc).
>
> Is this an implementation issue in NAMD, or true in theory also? Since
> deltaU_ba = Ua - Ub, I don't understand why you cannot calculate it as you
> do for deltaU_bc = Uc - Ub. From the trajectory sampled in Ub, I thought one
> should be able to calculate both Ua, and Uc.
In a sense, it is implementation-related. NAMD can only compute two
values of the potential at the same time: the reference and one
target. In principle, as you wrote, additional values could be
computed on the fly (at an increased computational cost). There is
another benefit in running two-way transformations, though: it is a
way good test of incomplete sampling and drift in configuration space,
which can be detected in the form of hysteresis in the free energy as
a function of lambda.
Cheers,
Jerome
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:31 CST