Re: [ESPResSo-devel] Helfand moments

[Axel Arnold]

Hi Josh,

yes, that would be interesting to integrate (ifdef'd, of course).

Although, I want to add some warning: I encountered EH-fits in another 
context (dielectric constant/conductivity). The fits look indeed much 
better that the current-current autocorrelation. But it turns out that 
in fact both approaches need equally loooooong trajectories, just that 
EH-fits hide this fact much better. Therefore, I consider them a bit 
dangerous (or misleading).

Axel

On 28.08.13 13:56, Josh Berryman wrote:
> Hello everyone,
>
> I've made a crude patch to the function add_nonbonded_pair_virials() 
> to calculate Helfand moments (sometimes called Einstein-Kubo-Helfand 
> moments).
>
> This is basically a slight adition to the stress tensor which allows 
> direct calculation of shear viscosities in the form:
>
> \eta  \propto <G^2>
>
> by analogy with the usual method for calculating diffusion constants 
> D  \propto <x^2>.   I've been working from Erpenbeck PRE 1995 for the 
> formula, and validating my answers against Meier et al. JCP 2004.
>
> Espresso already provides a means to find the shear viscosity, of 
> course, using a Green-Kubo method starting from the stress-tensor 
> autocorrelation function, but this is quite a messy and difficult 
> calculation, given the tendency of the acf to quite often have a long 
> and noisy tail.
>
> EKH doesn't seem to be any cheaper than GK, but it is simpler for the 
> user: there is less onus to e.g. fit something to the acf and 
> integrate the fitted function.  (Or for that matter to notice in the 
> Espresso docs that online acfs are available).
>
> The disadvantage is that it needs information from the verlet 
> nonbonded force calculation, including the *un-imaged* particle 
> coordinates, as well as the force and the imaged pairwise distance.  
> I've been using the function
>
>      get_particle_data(p1->p.identity, &pp1);
>
> to retrieve the un-imaged coordinates, however this of course 
> introduces a performance and scaling penalty, meaning that the 
> functionality will have to be hash-deffed out by default if it is 
> integrated with the main build.
>
> Is there any interest in having this?  Shall I create a branch and add 
> it in?
>
> Josh


--
JP Dr. Axel Arnold
ICP, Universität Stuttgart
Allmandring 3
70569 Stuttgart, Germany
Email: address@hidden
Tel: +49 711 685 67609