[gmx-users] Re: Re: FEP problem with lambda perturbation 1 to 0.95

[Floris Buelens]

> So what you mean is that I should either use soft core for

> the entire transformation or not use it at all. But then I do not
> understand what Carsten said about using at least on lambda values
> close to 0 and 1. I thought that if I used soft core on the last point
> of my transformation (the one that crashed before I used soft core) I
> would eliminate the singularities produced by the end points. I got a
> little confused now.

right, I see the confusion now - I don't agree with the way Carsten's phrased that, for a single free energy calculation you absolutely have to use a single consistent set of parameters. At end points lambda=0 and 1, the soft core potential is identical to standard vdW / electrostatics; this means the free energy difference for switching on a soft core potential at either end point is zero. When lambda is not =0 or =1, this is no longer true, so if you activate the soft core potential part way through there would be a free energy difference which isn't being counted.
hope that helps,

Floris


> Hi Fabricio,
>
> It's not correct to change your value of alpha, you should use a single value for the whole transformation. The soft core potential is constructed so that the end points at lambda=0 and 1 represent the 'native' states (with no influence from the soft core parameters), and that intermediate values produce a smooth transformation.
> Best,
>
> Floris
>
>
> ________________________________
> From: Ragnarok sdf <fabracht1 at gmail.com>
> To: gmx-users at gromacs.org
> Sent: Monday, 10 August, 2009 20:54:35
> Subject: [gmx-users] Re: Re: FEP problem with lambda perturbation 1 to 0.95
>
> I have tried using the soft core correction for values of lambda above
> 0.5, because i read that the overlaping starts usually at this lambda
> value. The problem is that now I values of deltaG where lambda is near
> 0.5 a little bit off-scale. What I mean is that what used to be a nice
> sequencial decrease in values from lambda 0 to lambda 1, now gives me
> a crazy value for lambda 0.5 (by crazy I mean 3 times larger) then for
> lambda 0.55, 0.6, 0.65 etc, I get values 3 times smaller than what
> they used to be. I guess that maybe using soft core correction for
> lambda values starting from 0.5 was a little bit to soon. But then for
> which lambda values should I start using soft core and how would I
> justify my choice?
> Thank you
> Fabrэcio Bracht
> - Ocultar texto das mensagens anteriores -
>
>> I agree with Carsten. See perhaps the discussion at www.alchemistry.org as
>> well.
>>
>> On Fri, Aug 7, 2009 at 2:35 AM, Carsten Kutzner <ckutzne at gwdg.de> wrote:
>>
>>> On Aug 6, 2009, at 10:57 PM, Ragnarok sdf wrote:
>>>
>>>  I am performing FEP do obtain the dimerization of a protein in
>>>> membrane. The lambda intervals i am using are 0.05 for each window.
>>>> After that I rerun each lambda .trr perturbing the system (plus)0.05
>>>> and (minus)0.05 lambda value. Then with g-energy I obtain the deltaG
>>>> for each delta lambda.
>>>> Well, I have encountered a problem when trying to simulate the last
>>>> window (1.0 - 0.95 ). The simulation runs for a while and then dies.
>>>> The log file says
>>>> Step 2200  Warning: Pressure scaling more than 1%.
>>>>
>>>>  Hi Fabricio,
>>>
>>> do you use soft-core? If not, I think you need to, at least for
>>> the intervals next to 0 and 1 to avoid singularities (these
>>> can result in undefined / NaN forces). See chap. 4.5.1 and
>>> 7.3.23 in the manual.
>>>
>>> Carsten
>>>
_______________________________________________
gmx-users mailing list    gmx-users at gromacs.org
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at http://www.gromacs.org/search before posting!
Please don't post (un)subscribe requests to the list. Use the
www interface or send it to gmx-users-request at gromacs.org.
Can't post? Read http://www.gromacs.org/mailing_lists/users.php