[gmx-users] Rugged dG/dlambda when turning off charges

[van Bemmelen]

Hi Bob,

Yes, when only turning off charges sc_alpha can be safely set to zero.
This will probably give you a much smoother result, although I'm not
sure if it will take care of the enormous fluctuations you are seeing.

Have you done any error analysis? Maybe you're just experiencing very
bad convergence in the <dG/dlambda> values for each separate lambda
point.

Cheers,
Jeroen


>Date: Fri, 25 Jan 2008 18:26:37 -0500
>From: "Robert Johnson" <bobjohnson1981 at gmail.com>
>Subject: Re: [gmx-users] Rugged dG/dlambda when turning off charges
>To: "Discussion list for GROMACS users" <gmx-users at gromacs.org>
>Message-ID:
>	<b302c1a40801251526l138acf61xf20ac0af56b2b390 at mail.gmail.com>
>Content-Type: text/plain; charset=ISO-8859-1
>
>Thanks David,
>Since I'm turning off the VdW parameters as a second step, 
>there really isn't any reason to use the soft core potentials 
>when turning off the charges right? Thus, sc_alpha should be 
>set to zero when turning off charges?
>Thanks,
>Bob
>
>On Jan 25, 2008 5:38 PM, David Mobley <dmobley at gmail.com> wrote:
>> It will only be smooth if you don't use soft core (i.e. 
>sc-alpha =0 ).
>>
>> Best,
>> David Mobley
>> http://www.dillgroup.ucsf.edu/~dmobley
>>
>>
>> On Jan 25, 2008 12:40 PM, Robert Johnson 
><bobjohnson1981 at gmail.com> wrote:
>> > Hello everyone,
>> > As I mentioned in a previous post, I'm trying to compute the free 
>> > energy of binding of a DNA base on a carbon nanotube. My system 
>> > consists of a single DNA nucleotide (base, sugar, phosphate group) 
>> > on a rigid carbon nanotube in aqueous solution. A single Na+ 
>> > counterion is also included for charge neutralization. I'm 
>> > proceeding with this by computing the free energy associated with 
>> > the following
>> > transformations:
>> >
>> > Water + Nanotube + DNA + Na+ -> Water + Nanotube + Nothing Water + 
>> > DNA + Na+ -> Water + Nothing
>> >
>> > For each of these transformations I FIRST turn off the charges on 
>> > the base and counterion over a set of 5 lambda values (0.0, 0.25, 
>> > 0.5, 0.75, 1.0). From literature and other posts on the 
>GMX mailing 
>> > list, it seems like this small set of values should be adequate 
>> > because dG/dlambda tends to be pretty smooth for 
>discharging the molecule.
>> >
>> > Unfortunately, that doesn't seem to be the case in my system. Here 
>> > is a table of the dG/dlambda values I get from running 5 ns 
>> > trajectories of my system:
>> >
>> > lambda     <dG/dlambda>
>> > 0.0           6211.105
>> > 0.25         1055.254
>> > 0.5           1230.675
>> > 0.75         1128.359
>> > 1.0           756.2904
>> >
>> > If you plot this you will see a large drop from lambda=0 to 0.25. 
>> > Then dG/dlambda increase slightly from lambda=0.25 to 0.5 and then 
>> > smoothly decreases after that. This seems like a pretty strange 
>> > result. Also, it seems that something weird is going on around 
>> > lambda=0. If I compute dG/dlambda for lambda=0.05, I get something 
>> > around 1200-1300, which is again a very sharp decrease from the 
>> > value of 6211.105 I get at lambda=0. Thus, it doesn't seem like 
>> > simply adding more lambda values is going to help me.
>> >
>> > Here are the values I'm using for the free energy calculation:
>> > free_energy         =  yes
>> > init_lambda         =  0.05
>> > delta_lambda        =  0
>> > sc_alpha            =  0.5
>> > sc_power            =  1
>> > sc_sigma            =  0.3
>> >
>> > Also, here is an excerpt from my topology file:
>> > [ atoms ]
>> > 1       P       1       DG      P       1       1.1659  30.9700
>> >  P       0.0000  30.9700
>> > 2       O       1       DG      O1P     1       -0.7761 16.0000
>> >  O       0.0000  16.0000
>> > 3       O       1       DG      O2P     1       -0.7761 16.0000
>> >  O       0.0000  16.0000
>> >
>> > I'm not perturbing any of the bonding interactions or VdW 
>parameters 
>> > - I'm just turning off all charges on the molecule.
>> >
>> > Does anyone have any ideas or suggestions about how this 
>can be improved?
>> >
>> > Thanks,
>> > Bob Johnson