[gmx-users] Diffusion coefficient depenting on time interval

[David van der Spoel]

Erik Marklund wrote:
> Hi users,
>
> I have simulated a box of NVT-water, more specifically the flexible 
> urey-bradley-like TIPS-reparametrisation by Dang and Pettitt (1987). 
> What I notice is that the diffusion coefficient increases when I 
> calculate it from a longer snippet from my trajectory. If I use only 
> 100 ps; starting at 100, 200 or 300 ps; the caculated coefficient is 
> lower than if I use 400 ps (100-500 ps). Is this a known issue, 
> inherent in the Einstein relation somehow?
>
This is more a sampling issue I would think. Make your simulations even 
longer.