Re: FEP calculations: optimizing the dlambda, number of steps

From: Shyno Mathew (sm3334_at_columbia.edu)
Date: Wed Dec 11 2013 - 15:00:01 CST

Hi all,
Initially while doing FEP, I wasn't outputting the dcd file!!
Now, I did a short run with dcdfreq specification. When I look at the
trajectory, I see the water box is chopped little on each corner.
In the beginning the water box looks fine (rectangular shape). As FEP
progresses, some water molecules are disappearing.
Does this mean something is wrong with the FEP calculations I did?

Look forward to hearing an answer!!

thanks,
Shyno

On Tue, Dec 10, 2013 at 11:28 AM, Shyno Mathew <sm3334_at_columbia.edu> wrote:

> Dear all,
> I would like to confirm the FEP part in my .conf file is correct?
> 1. Here is how it looks like for the forward FEP :
> alch on
> alchType FEP
> alchFile 04collab_wet.fep
> alchCol B
> alchOutFile 05collab.fepout
> alchOutFreq 100
> alchVdwLambdaEnd 1.0
> alchElecLambdaStart 0.5
> alchVdWShiftCoeff 5.0
> alchDecouple off
> alchEquilSteps 5000
>
> set Lambda0 0.0
> set dLambda 0.05
> while {$Lambda0 < 1.0} {
> alchLambda $Lambda0
> set Lambda0 [expr \$Lambda0 + \$dLambda]
> alchLambda2 $Lambda0
> run 15000
> }
>
> 2. Now for the reverse FEP everything is same except the following:
> set Lambda0 1.0
> set dLambda -0.05
> while {$Lambda0 > 0.0}
>
> thanks in advance for your help,
>
> best,
> Shyno
>
>
> On Tue, Dec 3, 2013 at 6:12 PM, Shyno Mathew <sm3334_at_columbia.edu> wrote:
>
>> Hello Jerome,
>> Is there a way to make sure my system is equilibrated at each strata?
>> Currently, I do 10000 steps of equilibration followed by 10000 steps of
>> data collection. But I am not sure 10,000 steps of equilibration is enough?
>> Also, what method/algorithm namd uses to do FEP?
>> I read BAR is another method to calculate free energy.
>> Since the ParseFEPplugin was showing us some errors, we are trying to
>> write a script that calculates BAR.
>>
>> thanks,
>> Shyno
>>
>>
>>
>> On Fri, Nov 22, 2013 at 1:09 PM, Shyno Mathew <sm3334_at_columbia.edu>wrote:
>>
>>> Hi Jerome,
>>> I was using "minimize" incorrectly.
>>> I meant for each window I have 5000 steps of equilibration followed by
>>> 10,000 steps of data collection.
>>> So I should try increasing the equilibration steps.
>>>
>>> thanks,
>>> Shyno
>>>
>>>
>>>
>>> On Fri, Nov 22, 2013 at 12:01 PM, Shyno Mathew <sm3334_at_columbia.edu>wrote:
>>>
>>>> thanks again Jerome. Sorry to bother with multiple questions.
>>>> Here is how the FEP section of my .conf file looks like (I saw it in
>>>> one of the FEP tutorials).
>>>> So for each window it does 5000 step minimization followed by 10,000
>>>> step run. What you are suggesting is just do minimization at the very first
>>>> window only?
>>>>
>>>> alch on
>>>> alchType FEP
>>>> alchFile 04collab_wet.fep
>>>> alchCol B
>>>> alchOutFile 05collab.fepout
>>>> alchOutFreq 100
>>>> alchVdwLambdaEnd 1.0
>>>> alchElecLambdaStart 0.5
>>>> alchVdWShiftCoeff 5.0
>>>> alchDecouple off
>>>> alchEquilSteps 5000
>>>>
>>>> set Lambda0 0.0
>>>> set dLambda 0.05
>>>> while {$Lambda0 < 1.0} {
>>>> alchLambda $Lambda0
>>>> set Lambda0 [expr \$Lambda0 + \$dLambda]
>>>> alchLambda2 $Lambda0
>>>> run 15000
>>>> }
>>>>
>>>> thanks,
>>>> Shyno
>>>>
>>>>
>>>>
>>>> On Fri, Nov 22, 2013 at 11:49 AM, Jérôme Hénin <jerome.henin_at_ibpc.fr>wrote:
>>>>
>>>>> ----- Original Message -----
>>>>> > Thanks for the reply Jerome.
>>>>> > Now, for each window I have 5000 minimization followed by 10000
>>>>> steps. I
>>>>> > did try going to 10^6 steps, but didn't give significant difference
>>>>> in free
>>>>> > energy.
>>>>>
>>>>> In general I wouldn't recommend minimizing between windows. You have
>>>>> to thermalize to room temperature again every time, and discard the
>>>>> additional equilibration time.
>>>>>
>>>>> > So just to be clear, in the previous message by saying "increase the
>>>>> > sampling time for just those two by running extra simulations" you
>>>>> meant
>>>>> > going to smaller dlambda values between 0.45 and 0.55?
>>>>>
>>>>> I meant running more MD for the same lambda and dlambda values.
>>>>>
>>>>> > Also, does it make sense to do FEP just between 0.45 and 0.55 and
>>>>> then use
>>>>> > it with data for all other windows?
>>>>>
>>>>> Yes, you can merge any amount of data from separate runs. How to do it
>>>>> properly depends on what free energy estimator you use. For the FEP
>>>>> estimator, it's just a matter of taking the log of a weighted average of
>>>>> exponenials of deltaGs.
>>>>>
>>>>> Jerome
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Shyno Mathew
>>>> PhD student
>>>> Department of Chemical Engineering
>>>> Columbia University
>>>>
>>>
>>>
>>>
>>> --
>>> Shyno Mathew
>>> PhD student
>>> Department of Chemical Engineering
>>> Columbia University
>>>
>>
>>
>>
>> --
>> Shyno Mathew
>> PhD student
>> Department of Chemical Engineering
>> Columbia University
>>
>
>
>
> --
> Shyno Mathew
> PhD student
> Department of Chemical Engineering
> Columbia University
>

-- 
Shyno Mathew
PhD student
Department of Chemical Engineering
Columbia University

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:59 CST