Too slow for 1008 Ni atoms, 20 or more minutes per step

[yell... at gmail.com]

Thank you very much for your swift reply.

Task was waiting in the queue yesterday. The simulation time per step is 
reduced obviously.

But the speed gets slower because of the increasing of SCF cycles.

 

# Step Nr. Time[fs] Kin.[a.u.] Temp[K] Pot.[a.u.] Cons Qty[a.u.] UsedTime[s]

 0 0.000000 9.566949152 2000.000000000 -170530.032653392 -170520.455919565 
0.000000000

 1 1.000000 9.631592301 2013.513848184 -170530.197758967 -170520.559182878 
3368.403749943

 2 2.000000 9.844516264 2058.026254206 -170530.614774376 -170520.760553775 
119.362696171

 3 3.000000 10.134210226 2118.587663569 -170531.251818051 -170521.099615591 
140.031235933

 

 103 103.000000 5.131475709 1072.750701983 -170544.302114205 
-170530.046201117 479.899127960

 104 104.000000 5.085703611 1063.181904658 -170544.285813199 
-170530.018294504 466.044290066

 105 105.000000 5.031751308 1051.903010733 -170544.056415561 
-170529.785912487 447.755058050

 

 122 122.000000 6.246773076 1305.907029920 -170546.291929917 
-170529.923306198 866.148313046

 123 123.000000 6.724651351 1405.808945809 -170545.774218367 
-170528.879454790 852.673817873

 124 124.000000 7.087707135 1481.706868681 -170545.226813770 
-170527.918999875 866.536659002

 

Is the following phenomenon normal?

 

*** SCF run converged in     1 steps ***

  Leaving inner SCF loop after reaching    20 steps.

 

Best regards,

Huang

在 2017年4月10日星期一 UTC+8下午5:45:00，zhj... at gmail.com写道：
>
> *1. the system(containing more than 1000 atoms) is too big;*
> *2. using PREFERRED_DIAG_LIBRARY 
> <https://manual.cp2k.org/trunk/CP2K_INPUT/GLOBAL.html#list_PREFERRED_DIAG_LIBRARY> in CP2K_INPUT 
> <https://manual.cp2k.org/trunk/CP2K_INPUT.html> / GLOBAL 
> <https://manual.cp2k.org/trunk/CP2K_INPUT/GLOBAL.html> section can speedup 
> the simulation, set PREFERRED_DIAG_LIBRARY=ELPA, but you should install 
> elpa     library before using it;*
> *3. the cutoff may be too small, the bigger values will be slower;*
> *4. the EPS_SCF is too big,  tunning it to small( for example, 4.0E-7, the 
> simulation will be slower);*
> *5. using the relaxed system can speedup convergence.*
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170412/11e0c34b/attachment.htm>