AMBER Archive (2005)

Subject: Re: AMBER: TI with electrostatic decoupling at sander

From: David A. Case (case_at_scripps.edu)
Date: Wed Jan 26 2005 - 23:51:32 CST

On Wed, Jan 26, 2005, yipinl_at_umich.edu wrote:

>
> In the tutorial, a formula:
> [sample-rms]*sqrt[correlation-time/length-of-simulation]
> is used to estimate the expected error in the mean value of DV/DL. My
> understanding is that this is a statistical estimate of the error term. So for
> the correlation-time/length-of-simulation term, should I calculate it using
> (1+2?)/n
> where n is the number of data points and ? is a correlation length calculated
> from the autocrrelation coefficients: ? = sum(1-k/n)?, according to the paper
> by Dr. Pearlman at J. Comp. Chem. 1994, 105?

The mail encoding I got made it difficult to read your question. Estimating
the number of "independent samples" as simulation-length/correlation time is
qualitative: depending exactly on how you calculate the autocorrelation of the
<dV/dl> values, this value might better be calculated with an extra sqrt(2) or
so. So, you should treat such estimates with care; a better method is
probably to independently re-do the simulation multiple times.

>
> For the removing of charge in the vacuum, the tutorial said:
> "Hence, the expected error in the mean is about 0.028 * sqrt[ 0.1/50 ] = 0.001
> kcal/mol. "
> But the RMS fluctuation of DV/DL in the output is 0.0807, not 0.028.

My mistake...I copied the wrong number from the output. I'll update the
tutorial with this change.

...thanks...dave case

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu