[CP2K:5228] Semiempirical (SE) MD vs BLYP simulation cost

[hut... at chem.uzh.ch]

Hi

the time for a calculation is split between the calculation of
the Kohn-Sham/Hamiltonian matrix and solving the 'eigenvalue'
equation. In your example DFTB is about 30x faster in generating
the Hamiltonian matrix than BLYP. The eigenvalue problem dominates
the DFTB calculation and because you are doing more SCF iterations,
the total time are closer. 
Test for your specific system which optimiser, OT, Diagonalization,
linear scaling is the fastest. You should especially see what happens
in a MD or GEO_OPT (depending what you want to do).
You can also optimize the different accuracy settings, convergence 
criteria and finally the preconditioner.

regards

Juerg 
--------------------------------------------------------------
Juerg Hutter                         Phone : ++41 44 635 4491
Institut für Chemie                  FAX   : ++41 44 635 6838
Universität Zürich                   E-mail: hut... at chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------

-----cp... at googlegroups.com wrote: -----
To: cp... at googlegroups.com
From: bharat 
Sent by: cp... at googlegroups.com
Date: 05/01/2014 05:33PM
Subject: [CP2K:5228] Semiempirical (SE) MD vs BLYP simulation cost

Dear experts,

I am doing semiempirical MD simulation using CP2K. Normally SE calculations are very fast compared to DFT calculations. But in CP2K, SE MD simulation is only FOUR times faster than BLYP. According the TIMING printed in output file, SE spends 78% of total time in "cp_dbcsr_multiply_d" subroutines, but BLYP only spends 24% of total time there. In the particular calculation, numbers of calls are of course different (29655 (SE) vs 17081 (BLYP)). cp_dbcsr_multiply_d is a part of linear scaling iterative algorithm.
Please find the attached file for cost comparison for typical calculation for both methods.

My question is: is there any way to improve the computational cost for SE calculation?

Any suggestions and comments are welcome.

Thank you.

Sincerely,
Bharat Sharma
   
  -- 
 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+uns... at googlegroups.com.
 To post to this group, send email to cp... at googlegroups.com.
 Visit this group at http://groups.google.com/group/cp2k.
 For more options, visit https://groups.google.com/d/optout.
 

[attachment "Computational Cost.docx" removed by Jürg Hutter/at/UZH]