Re: Querries on eABF

From: Souvik Sinha (souvik.sinha893_at_gmail.com)
Date: Sat Jul 15 2017 - 07:39:34 CDT

Hello,
I am performing some small MW test runs (say 10ns) (semi-eABF) to check
values of thermal width coupling parameter sigma for one colvar (alpha). My
intention is to get the value of sigma where
i) z-marginal (count) is nearly uniform i.e. narrow joint distribution in
"extended variable" vs "real colvar" space and
ii) nearly uniform distribution in 2 colvar [ alpha vs distanceZ (no
extended variable is attached i.e. classical ABF)] space.

There I noticed that as I was decreasing the value of sigma, two of my
goals met up to some level but later on with more smaller sigma, uniformity
in the mentioned distributions is again disturbed. It seems like there are
critical threshold values of sigma that will yield best results for
respective colvars. May be it's a naive approach but I must ask why
stronger coupling constants can't always yield best results in this
scenario?

It's possible that timescale of my test runs is not sufficient to get
anything better than what I already have but my goal was to get the best
result in that timescale and moving on to longer runs for my original
investigation using well-optimized parameters from the test runs.

Thank you.

On Wed, Jul 12, 2017 at 6:04 PM, Souvik Sinha <souvik.sinha893_at_gmail.com>
wrote:

> Ok I got it now. Actually I was getting indivually "count" & "zcount" for
> all the walkers. Now I just checked that all walkers gave same "count" file
> i.e. aggregated over all walkers.
>
> Thank you.
>
> On Wed, Jul 12, 2017 at 5:54 PM, Jérôme Hénin <jerome.henin_at_ibpc.fr>
> wrote:
>
>> Hi Souvik,
>>
>> The ".count" values are aggregated over all walkers, whereas the
>> ".zcount" are specific to each individual walker".
>>
>> What should be uniform (in the long time limit) is "count". "zcount"
>> should be approximately uniform, meaning that eABF should ensure that there
>> are no gaps in sampling. The smaller the sigma parameter, the more uniform
>> it will be (and closer to count).
>>
>> In the end you'll get a separate CZAR estimate of the PMF for each
>> walker, which you can compare to assess convergence. You can combine them
>> using the "inputPrefix" option of ABF.
>>
>> Best,
>> Jerome
>>
>>
>> On 12 July 2017 at 14:15, Souvik Sinha <souvik.sinha893_at_gmail.com> wrote:
>>
>>> Hello
>>>
>>> Sorry to bother again. I am confused at one point. If I do semi-eABF
>>> using MW strategy, what "count" file I should consider to check sampling
>>> along my define 2 RCs? Now "*.zcount" is giving values for "czar estimator"
>>> but normal "*.count" is giving what value exactly? "*.count" values are
>>> much larger than "*.zcount" values for any walker. I am confused that what
>>> file should I consider to check for uniformity of sampling.
>>>
>>> Thanks.
>>>
>>>
>>> On Wed, May 31, 2017 at 9:40 AM, Jérôme Hénin <jerome.henin_at_ibpc.fr>
>>> wrote:
>>>
>>>> That is correct, the error message came from the Tcl script that
>>>> implements the ZY estimator. Good catch.
>>>>
>>>> Best,
>>>> Jerome
>>>>
>>>>
>>>>
>>>> On May 30, 2017 21:19, <yjcoshc_at_gmail.com> wrote:
>>>>
>>>>> The eabf.tcl in NAMD_2.12_Source/lib/eabf is the implementation of
>>>>> Zheng/Yang estimator. It may not support semi-eABF. If you want to use
>>>>> semi-eABF, you can disable it and use czar estimator only.
>>>>>
>>>>> I have asked the author of Z/Y estimator to subscribe this mailing
>>>>> list but he seems to have some trouble with subscribing and sending mails.
>>>>>
>>>>> 在 2017年05月31日 00:01, Souvik Sinha 写道:
>>>>>
>>>>> Actually, the error lines are the following :
>>>>>
>>>>> colvars: Error while executing calc_colvar_forces:
>>>>> colvars: can't read "extendedFluctuation": no such variable
>>>>> FATAL ERROR: Error in the collective variables module: exiting.
>>>>>
>>>>>
>>>>> So, I think it's arriving from the "calc_colvar_forces" process of the
>>>>> tcl script eabf.tcl that I am sourcing during the run. But I couldn't find
>>>>> any sort of warning line written in that tcl script. So may be its from
>>>>> some other Tcl script related to colvars.
>>>>>
>>>>> P.S. eabf.tcl is given with NAMD2.12 package
>>>>>
>>>>> Souvik
>>>>>
>>>>> On Tue, May 30, 2017 at 6:42 PM, Giacomo Fiorin <
>>>>> giacomo.fiorin_at_gmail.com> wrote:
>>>>>
>>>>>> Hi, I think that there must be either a transcription error, or there
>>>>>> is a Tcl script somewhere that prints an error message with the "colvars:"
>>>>>> prefix. That error message doesn't look like it comes from the C++ code.
>>>>>>
>>>>>> Giacomo
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Tue, May 30, 2017 at 8:54 AM, Jérôme Hénin <jerome.henin_at_ibpc.fr>
>>>>>> wrote:
>>>>>>
>>>>>>> Hi Souvik,
>>>>>>>
>>>>>>> The part of the output you quote contains two separate errors.
>>>>>>>
>>>>>>> > colvars: can't read "extendedFluctuation": no such variable
>>>>>>>
>>>>>>> This is actually a Tcl error meaning that some code is trying to
>>>>>>> access a Tcl variable with that name. That is a little surprising, I'd like
>>>>>>> to see the input that gave that. Can you send the NAMD config file and any
>>>>>>> colvars input off-list?
>>>>>>>
>>>>>>> And yes, semi-eABF is applicable when at least one variable provides
>>>>>>> the total force and Jacobian derivative needed for standard ABF. I
>>>>>>> documented semi-eABF for the sake of completeness, but it is not entirely
>>>>>>> clear how useful it is in practice. The final answer would come from
>>>>>>> real-world tests on a large, slow system.
>>>>>>>
>>>>>>> Jerome
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 30 May 2017 at 08:37, Souvik Sinha <souvik.sinha893_at_gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> Thanks for the reply.
>>>>>>>>
>>>>>>>> I think we can use semi-eABF for cases where system force is
>>>>>>>> available for
>>>>>>>> one of the RC (so classical ABF is applicable here) but not the
>>>>>>>> other one
>>>>>>>> (considering 2 RCs). So that other RC can be treated with eABF. But
>>>>>>>> I am
>>>>>>>> not certain about it.
>>>>>>>>
>>>>>>>> Souvik
>>>>>>>>
>>>>>>>> On Tue, May 30, 2017 at 5:56 PM, <yjcoshc_at_gmail.com> wrote:
>>>>>>>>
>>>>>>>> > Hi,
>>>>>>>> >
>>>>>>>> > 1. The r_colvarName is the value of extended variable (or
>>>>>>>> fictitious
>>>>>>>> > particle). In *Smoothed Biasing Forces Yield Unbiased Free
>>>>>>>> Energies with
>>>>>>>> > the Extended-System Adaptive Biasing Force Method*, it's called λ.
>>>>>>>> >
>>>>>>>> > 2. Also I don't figure out when to use semi-eABF. I just read it
>>>>>>>> on the
>>>>>>>> > article and have never use it.
>>>>>>>> >
>>>>>>>> > yjcoshc
>>>>>>>> >
>>>>>>>> > 在 2017年05月30日 15:51, Souvik Sinha 写道:
>>>>>>>> >
>>>>>>>> > Hi,
>>>>>>>> >
>>>>>>>> > 1. I am running a 2D eABF using "alpha" & "distanceZ" colvars as
>>>>>>>> my 2 RCs
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Giacomo Fiorin
>>>>>> Associate Professor of Research, Temple University, Philadelphia, PA
>>>>>> Contractor, National Institutes of Health, Bethesda, MD
>>>>>> http://goo.gl/Q3TBQU
>>>>>> https://github.com/giacomofiorin
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Souvik Sinha
>>>>> Research Fellow
>>>>> Bioinformatics Centre (SGD LAB)
>>>>> Bose Institute
>>>>>
>>>>> Contact: 033 25693275
>>>>>
>>>>>
>>>>>
>>>
>>>
>>> --
>>> Souvik Sinha
>>> Research Fellow
>>> Bioinformatics Centre (SGD LAB)
>>> Bose Institute
>>>
>>> Contact: 033 25693275
>>>
>>
>>
>
>
> --
> Souvik Sinha
> Research Fellow
> Bioinformatics Centre (SGD LAB)
> Bose Institute
>
> Contact: 033 25693275
>

-- 
Souvik Sinha
Research Fellow
Bioinformatics Centre (SGD LAB)
Bose Institute
Contact: 033 25693275

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:26 CST