VMD-L Mailing List

From: Jérôme Hénin (jerome.henin_at_ibpc.fr)
Date: Tue Feb 19 2019 - 11:48:04 CST

Hi Amy,

In the fepout file I see the last window:
#NEW FEP WINDOW: LAMBDA SET TO 1 LAMBDA2 1 LAMBDA_IDWS 0.99

This is what confuses the deinterleaving program. The last window should
just be a one-way reverse transformation, not IDWS, going from 1 to 0.99.
That is what the "official" FEP script should generate (if it doesn't, it's
a bug that needs fixing).

In your file, because lambda and lambda2 are identical, all the "forward"
energy differences are zero:
FepEnergy: 5 -81621.4223 -81621.4223 7507.7985
     7507.7985 -0.0000 -0.0000 310.6220 -0.0000
FepE_back: 10 -81662.1280 -81657.6108 7531.8894
     7531.4122 4.0399 -0.0000 311.2123 -0.0000
FepEnergy: 15 -81731.5264 -81731.5264 7579.5508
     7579.5508 0.0000 -0.0000 310.9899 -0.0000

The simplest fix I see for this is re-running the last window without IDWS
(if that isn't too expensive). You can then make a copy of the current
fepout, deleting the last window, and supply the newly created,
single-window fepout separately to deinterleave_idws.py.

Best,
Jerome

On Tue, 19 Feb 2019 at 18:12, Amy Rice <arice3_at_hawk.iit.edu> wrote:

> Hi Brian,
>
> Thank you for the response. I went back and re-ran both calculations with
> the default temperature in the GUI (300 in both cases).
>
> Version 1.9.3
> 300 K: total free energy change is -89.32071464053809 , total error is
> 0.09901395668211348
> 310 K: total free energy change is -89.32071464053809 , total error is
> 0.09901395668211348
>
> Version 2.1
> 300 K: total free energy change is -98.6952146581652 , total error is
> 0.063764981495484
> 310 K: total free energy change is -98.7037132059109 , total error is
> 0.07093475906066159
>
> The older version does not seem to accept the temperature change, as you
> suggested, while version 2.1 gives a slightly different result with the
> higher temperature provided. Should I take the result from version 2.1 to
> be the more reliable one? I am also sending my fepout files to Jerome
> off-list for further evaluation of this difference.
>
> Thank you,
> Amy
>
> On Mon, Feb 18, 2019 at 4:31 PM Brian Radak <brian.radak_at_gmail.com> wrote:
>
>> Hi Amy,
>>
>> I've noticed this also. I believe there were some bugs in the BAR
>> estimator (and fepout file parsing) that were addressed between 1.9 and
>> 2.1. I'm surprised that the magnitude of difference is so large, but it's
>> quite reasonable that they are different.
>>
>> Older versions also have some problems with modifying temperature from
>> the GUI (they get "stuck" at the default 298 K). Please double check and
>> run the calculation with the default settings and then change to 310 K to
>> make sure this is properly working. If the result doesn't change, then
>> there is likely a bug.
>>
>> Brian
>>
>> On Mon, Feb 18, 2019 at 4:48 PM Amy Rice <arice3_at_hawk.iit.edu> wrote:
>>
>>> Hi all,
>>> I noticed that using different versions of ParseFEP (1.9.2 vs 2.1) gives
>>> different free energy output when supplied with the same fepout files and
>>> temperature selection. With 1.9.2, the free energy change was calculated as
>>> -89.320 with a total error of 0.099. The same files in version 2.1 resulted
>>> in a free energy change of -98.704 and a total error of 0.071. Is this
>>> expected behavior? In both cases, I set the temperature to 310 K and
>>> selected to use the BAR-estimator method. Was there a change in how the
>>> calculation is performed between these two versions? It is a little
>>> disconcerting to have such a large difference between the two. I have
>>> attached the log file for both cases.
>>>
>>> Best,
>>> Amy
>>>
>>> --
>>> Amy Rice
>>> Ph.D. Candidate
>>> Physics Department
>>> Illinois Institute of Technology
>>>
>>
>
> --
> Amy Rice
> Ph.D. Candidate
> Physics Department
> Illinois Institute of Technology
>