[gmx-users] Re:Re:Re: MSD normalization

[Mark Abraham]

On 9/12/2011 5:37 AM, Sławomir Stachura wrote:
> Hi!
> Thanks Mark for reply. When I calculated fractional diffusion constant from mass weighted MSD of long simulation I get sth about 0.05. But when I do the same for centre of mass, considering that MSD is expressed in nm^2, the result is 1.35. When I look into xvg file  MSD of centre of mass is 10-20 times than for mass-weighted (which is substantially wrong). So assuming that, by any chance, centre of mass MSD is in A^2, calculated fractional diffusion constant is of 0.02 magnitude, which is theoretically more understandable.
> What I can show you is a sample of two MSDs:
>
> - Mass weighted:
>
>           0           0
>          15   0.0150352
>          30   0.0198877
>          45   0.0235849
>          60   0.0267214
>          75   0.0294651
>          90   0.0319853
>         105   0.0342538
>         120   0.0364239
>         135   0.0384275
>         150   0.0403661
>         165   0.0421938
>         180   0.0439495
>         195   0.0456557
>         210   0.0472885
>         225   0.0488768
>         240   0.0503814
>         255    0.051868
>         270   0.0533207
>         285   0.0547436
>         300   0.0561625
>
> -And centre of mass:
>
>           0           0
>          15    0.916885
>          30     1.05944
>          45     1.16649
>          60     1.25004
>          75     1.31444
>          90     1.37307
>         105     1.43301
>         120     1.47879
>         135     1.52776
>         150     1.57137
>         165      1.6211
>         180     1.64439
>         195     1.68883
>         210     1.71222
>         225     1.75035
>         240     1.77255
>         255      1.8117
>         270     1.83702
>         285     1.87376
>         300      1.8958
>
>
> But still there is more thing - do you how the MSD is normalized? Because I have compared mass weighted MSD calculated using Gromacs tool and nMoldyn and I get the same difference for all my cases (simulations of 2ns, 15ns and 150ns)- in the latter  diffusion constant is twice the value of the one obtained in Gromacs. So I expect the difference in normalization.

OK, but we need to see your terminal command lines in order to eliminate 
your use of the tool as a source of the problem. They should be at the 
top of the .xvg file, if nowhere else.

Mark

>
> Cheers,
>     Slawomir
>
>
>
>
>> On 6/12/2011 3:58 AM, S“awomir Stachura wrote:
>>> Hi,
>>>
>>> There is something interesting concerning these MSD calculations. If I treat Centre of Mass MSD  if it'd be expressed not in nm^2/ps , but Angstroms^2/ps, I get ideally the same results as for mass-weighted ones. And it's true for all my three cases, in which I performed simulation of different times. Is that possible there is some bug in MSD? Because as for me it looks so....
>> I would be amazed if GROMACS did this. Can you copy and paste from your
>> terminal to show this?
>>
>> Mark
>>
>>> Hope to have someone join the discussion,
>>>     Slawomir
>>>
>>>
>>>
>>>
>>>
>>> Wiadomo¶æ napisana przez S“awomir Stachura w dniu 2011-12-04, o godz. 22:38:
>>>
>>>> Hi,
>>>>
>>>> I was just wondering how MSDs in Gromacs are normalized. I am calculatiing MSDs for my lipid system of different timesteps of simulation and I get much higher results for centre of mass calculations in comparison to mass-weighted one.  And I thought about normalization issue, but I couldn't find any informations concerning that topic in manual.
>>>> I'd be grateful if someone could explain how mean square displacement is normalized.
>>>>
>>>> Cheers,
>>>>    Slawomir