Re: [AMBER] Statistically estimating the error and data reliability for Jarzynski Method

From: Ramin Salimi <>
Date: Wed, 17 May 2017 17:50:16 +0000

Hi Dr. Roitberg

that is correct, I phrased it wrong. What I meant referred to your 2006 paper: Free energy calculations with non-equilibrium methods: applications of the Jarzynski relationship

In that paper, you pulled the Ace-Alanine 8 -NMe peptide molecule with two different harmonic potentials. If we assume that you fixed the one end with ntr flag, how do you pull the other end? How does your input file look like to pull the other end? And also, how does the physics of the problem change when we have more than one potential? In other word, in SMD, or nmropt, as far as I know, we have the distance as the reaction coordinate and one single potential two pull that distance, but how does the problem change in case of two different potentials on two different atoms of the system?

Moreover, in Fig. 6 of that paper, you have also drawn the exact deltaF. How do you know that is exact?

More generally, how should we know that this pulling rate for this system is near-equilibrium? this rate is near-instantaneous?



From: Adrian Roitberg <>
Sent: Wednesday, May 17, 2017 12:30:22 PM
To: AMBER Mailing List
Subject: Re: [AMBER] Statistically estimating the error and data reliability for Jarzynski Method

Ramin, which two methods ?

SMD is simply a way to pull or push things along a coordinate.

Jarzynski is a statistical mechanics theorem, not a technique.

One can use Jarzynski's theorem to postprocess SMD data in order to get
free energies vs reaction coordinate.

Can you clarify what you meant by the two methods ?


On 5/17/17 1:21 PM, Ramin Salimi wrote:
> Thanks Hannes
> You are correct, but I thought that there might be some numerical mysterious way to calculate the error.
> But what about the other question that I posed:
> How different is it from the Jarzynski method, that basically works on a reaction coordinate i.e. the distance between the atoms or groups of atoms, when we use ntr to freeze one atom or a group of atoms, and pull the other end with a different potential? I mean what difference does it make in results? which one is better to use? I would want to learn how different the physics of the problem would be for I am used to see only constant force constants in SMD.
> When we create the input files for SMD, or nmropt, we should put at least two atoms for iat, but what this case that hold one end with ntr flag? How do we prepare the input file to pull the other end?
> Thanks
> Ramin
> ________________________________
> From: Hannes Loeffler <>
> Sent: Wednesday, May 17, 2017 4:12:44 AM
> To:
> Subject: Re: [AMBER] Statistically estimating the error and data reliability for Jarzynski Method
> On Wed, 17 May 2017 09:00:54 +0000
> Ramin Salimi <> wrote:
>> Hi Hannes
>> You, in fact, are right since the free energy difference is equal to
>> the average work plus the second cumulant expansion, but we do not
>> have the free energy difference so amongst the three quantities, we
>> only have the average work, not either the variance or the free
>> energy difference???
> The equality only holds when the distribution of the work values is a
> Gaussian which e.g. when you operate in a near-equilibrium regime, see
> for instance .
> The variance here is the variance of your work values! E.g. paste all
> work values that you have received from N simulations into a spread
> sheet and ask it to compute the sample variance (like the function
> var.S in Excel/LibreOffice calc) from this data. That's really just
> standard statistics.
> Cheers,
> Hannes.
> _______________________________________________
> AMBER mailing list
> _______________________________________________
> AMBER mailing list

Dr. Adrian E. Roitberg
University of Florida Research Foundation Professor
Department of Chemistry
University of Florida
AMBER mailing list
AMBER mailing list
Received on Wed May 17 2017 - 11:00:03 PDT
Custom Search