Luca luca.d... at
Wed May 6 14:40:18 UTC 2015

Hi Florian

I have just found the problem. It was in the wrong amount of memory given 
to cp2k for storing the integrals, so it was forced to calculate on the fly 
part of those. With the right amount of memory (and eps 1E-7) the time for 
rtp is now about 60 sec.

All the bests

On Wednesday, May 6, 2015 at 2:13:45 PM UTC+2, Luca wrote:
> Hi Florian 
> thank you for your remarks and the information about the details of the 
> implementation. However that input was just to test the density_propagation 
> with the admm. My input for thereal-time spectrum contained the right 
> keyword (rt_propagation). I tested the ARNOLDI algorithm and I did not get 
> any speed-up respect to the padè. However as you pointed out, I used a 
> small eps_schwartz  I am going to try 1E-7 and later the truncated exchange.
> Luca   
> On Wednesday, May 6, 2015 at 12:50:55 PM UTC+2, Florian Schiffmann wrote:
>> Hi 
>> A few remarks on your input. 
>> You are running ehrenfest dynamics which is not what you want. For 
>> spectra you need rt_propagation. This will speed up your calculation as no 
>> forces are required. Furthermore integrals only need to be computed at the 
>> very first step and afterwards are read from memory. Its mor than a factor 
>> of 2 at least in speed. 
>> I doubt eps_schwarz needs to be so low. Try 10e-7. This should still be 
>> accurate but again faster rtp steps. 
>> Would be nice if you could post timings with these settings. 
>> Flo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list