[CP2K-user] [CP2K:19713] Re: Temperature during NVE run

[Léon Luntadila Lufungula]

P.S. I have tried several ways to find the best way to calculate the drift 
before I ended up with this formula. However, as the cumulative average of 
the constant of motion oscillates up and down, this way of calculating the 
drift is still not that accurate as it will very much depend on where in 
the oscillation you measure the values. Therefore, I also wanted to ask if 
someone knows a more elegant way to calculate the drift?

Thanks in advance,
Léon

On Tuesday 9 January 2024 at 11:59:24 UTC+1 Léon Luntadila Lufungula wrote:

> Dear all,
>
> I'm running some ab initio NVE calculations to determine accurate settings 
> for EPS_SCF, EPS_DEFAULT, CUTOFF and TIMESTEP for my production runs. I 
> think I have found good values for each as I see little to no drift in the 
> energy after 16ps of simulating and the oscillations also seem reasonable 
> (see below for details). However, I see that the average temperature during 
> my simulation is around 157 K, while I set TEMPERATURE to 298.15 K. (see 
> input and ener files in attachment)
> [image: T.png]
> I know that NVE does not keep the temperature constant, but is it normal 
> that it is nearly half of the set temperature? Any comments on this would 
> be greatly appreciated!
>
> MD convergence details (comments on this part are always welcome):
> The extrapolated drift per ns is: 0.0001415478264154313 a.u 44.697268403725126 
> K The oscillations (standard deviation) of the energy is: 2.6854768188532883e-05 
> a.u. The average kinetic energy is: 0.06610813204255318 a.u. The average 
> CPU time is: 8.164207377190335 The drift per ps is 0.0002 % of the 
> kinetic energy. The oscillations are 0.0406 % of the kinetic energy.
> [image: E.png][image: U_and_E.png]
> The time until equilibration was chosen to be 5ps based on the figures 
> above and all mentioned quantities above have been calculated from 5ps to 
> the final step of the simulation (~16ps).
> Drift was calculated by taking the difference in the cumulative average of 
> the constant of motion over the equilibrium period (5-16ps) and dividing by 
> the timerange (~11ps) to get the drift per ps. This was then multiplied by 
> 1000 to get the drift per ns.
> The oscillations were calculated by taking the standard deviation of the 
> constant of motion over the equilibrium period.
> As instructed by the manual 
> <https://manual.cp2k.org/trunk/methods/sampling/molecular_dynamics.html> in 
> the section of MD Convergence, the drift per ps and oscillations are 
> calculated as percentages of the kinetic energy. As these values are well 
> below 1% and the constant of motion seems to be quite stable in the 
> pictures, I can conclude that the settings are appropriate to maintain 
> stable dynamics. EPS_SCF 6 gave a significant drift, therefore EPS_SCF 7 
> was chosen. I'm also looking into lower values for EPS_DEFAULT to reduce 
> the CPU time.
>
> If someone has comments or suggestions on my way of calculating and 
> interpreting the drift and oscillations, I'd be happy to hear them!
>
> Best wishes for 2024!
>
> Kind regards,
> Léon
>
>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ea9877fd-656a-4fa7-b355-8e4c05e03787n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240109/96558320/attachment.htm>