From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Fri Apr 20 2012 - 17:50:26 CDT

On Fri, Apr 20, 2012 at 3:49 PM, Ivan Vyalov <vyalov_at_mis.mpg.de> wrote:
> On 04/20/2012 09:11 PM, Axel Kohlmeyer wrote:
>>
>>
>> On Apr 20, 2012, at 6:43 AM, Ivan Vyalov<vyalov_at_mis.mpg.de>  wrote:
>>
>>> Hi all!
>>>
>>>
>>> I have a question related to the normalization of rdf in VMD. I've seen
>>> the previous thread about it
>>> www.ks.uiuc.edu/Research/vmd/mailing_list/vmd-l/18223.html
>>> but it seems that problem of opener has disappeared but mine is still
>>> here. I get the same problem with limiting behaviour of g(r).
>>
>> Which version of VMD are you using?
>>
>> How do you compute the integral?
>> The value of g(r) is binned and the r is taken as the center of the bin.
>>
>> Axel
>>
>>> The system is 4169 SPC/E water molecules at 306 K in the box with cell
>>> length 50 \AA{}.
>>> What I need is to calculate Kirkwood-Buff integral. h(r) looks well in
>>> general:
>>> img846.imageshack.us/img846/6460/56853809.png
>>> but its integral multiplied by r^2 diverges(here it's just a sum h(r)r^2
>>> not multiplied by dr and is a little bigger than the proper integral, but it
>>> doesn't change the problem):
>>> img812.imageshack.us/img812/2722/handintegral.png
>>>
>>> At first, I equilibrated system for 1ns, but when I've obtained this
>>> behaviour I continued to equilibrate for 2 ns more with the same result.
>>> Here is the tail of h(r) which is noisy but definitely lies above zero in
>>> average.
>>> img210.imageshack.us/img210/9803/htail.png
>>> If I average more taking wider bins I get the following picture:
>>> img252.imageshack.us/img252/8083/htailbroadbin.png
>>>
>>>
>>> This looks quite strange even though I know about difficulties with such
>>> calculations.
>>> The question is obvious, is everything alright with the normalization of
>>> g(r) in VMD?
>>>
>>> However, it can be something else rather than normalization because
>>> functions of different pairs behave differently:
>>> img191.imageshack.us/img191/8205/handintegralall.png
>>> This means that OO and HH have positive component in h(r) and OH --
>>> negative.
>>>
>>> Any help and ideas are much appreciated!
>>>
>>> thanks in advance,
>>> Ivan
>
>
> Hello Axel,
>
> I used VMD 1.9 and 1.9.1 and they both give the same result.
>
> Here's the better plot of integral calculated as (scipy)
> cumsum(h*r**2*(r[1]-r[0]))
> http://img62.imageshack.us/img62/9466/handrightintegral.png
>
> The problem comes from positive tail in h, can it be from PME(I took grid
> spacing equal to 1\AA{})? From the other hand if I take the average of h(r)
> from 15 to 25 \AA{} and subtract it from h(r), integral converges and it
> seems to me that the error is constant.
> http://img401.imageshack.us/img401/871/hshiftedandrightintegra.png

can you please try running your g(r) calculation with a smaller
bin size, say 0.01 instead of the default of 0.1?
and let us know if that makes matters better or
worse or has no impact at all.

thanks,
    axel.

-- 
Dr. Axel Kohlmeyer
akohlmey_at_gmail.com  http://goo.gl/1wk0
College of Science and Technology
Temple University, Philadelphia PA, USA.