AMBER Archive (2004)

Subject: Re: AMBER: TI with sander

From: David A. Case (
Date: Tue Sep 14 2004 - 14:20:49 CDT

On Tue, Sep 14, 2004, Oliver Hucke wrote:
> o Is it correct that, if atoms disappear, the backward direction can not
> be simulated. I.e. the hysteresis between the forward and the backward
> direction, as an error estimate, can not be computed?
> How should one estimate the error then?

In principle, TI does not have any concept of "direction". Each of the
averages of <dV/Dlambda> can be simulated independent of the others.

If you wish, you could perform a series with lambda = (say) 0.1, 0.3, 0.5, 0.7
and 0.9 (starting each new value from the end-point of the previous value).
Then you could do an additional set with lambda = 0.9, 0.7, 0.5, 0.3 and 0.1.
This could be then divided roughly into a "forward" and a "backward"

But the better to way to estimate errors is to try to determine the
uncertainty in <dV/dlambda> at each lambda point, then propagate these error
estimates into a uncertainty in the final result.

> o What meaning do dV/dL values have for lambda=0 and lambda=1 ?

The same as they mean at all other values of lambda? I clearly don't
understand the question: there is nothing special about the endpoints in terms
of the "meaning" of this value.

> o Where in the source code can I find the paragraphs relevant for TI?

In thermo_int.f (it's not hidden all that much!)


The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to