Wavelet Solver Issues

[Moe El]

Hi Everyone,

I am trying to run a static energy and geometry optimization of an rgd chain; however, I always keep getting an error related to the solver. I followed the tutorial online and tried to use the same methodology for my molecule. I am very new to CP2K and I dont know what I am doing wrong. I would like any feedback for improvement. Thank you all for your time and effort.

[Marcella Iannuzzi]

Hi ...

I could run the input that you posted, rgd_energy.inp,

without any error related to the wavelet solver. The output looks actually OK (see below)

Maybe the installation of cp2k that you are using is the problem

Regards

Marcella

  Step     Update method      Time    Convergence         Total energy    Change

  ------------------------------------------------------------------------------

     1 P_Mix/Diag. 0.40E+00   56.3     1.41372953      -235.4646768605 -2.35E+02

     2 P_Mix/Diag. 0.40E+00   59.9     0.83072771      -236.1830992228 -7.18E-01

     3 P_Mix/Diag. 0.40E+00   59.7     0.51090604      -236.6024408629 -4.19E-01

     4 P_Mix/Diag. 0.40E+00   59.7     0.30502715      -236.8517046496 -2.49E-01

     5 P_Mix/Diag. 0.40E+00   59.5     0.18568352      -237.0010410658 -1.49E-01

     6 P_Mix/Diag. 0.40E+00   59.9     0.11078336      -237.0909866346 -8.99E-02

     7 P_Mix/Diag. 0.40E+00   59.7     0.06776712      -237.1452015485 -5.42E-02

[Augustin Bussy]

Dear Moe,

your inputs are correct, although not optimal (more on this later). I did some testing, and it seems that  you ran into a bug of the wavelet solver when large boxes and high number of OMP threads are involved. We will look into that.

A box of 50A x 50A x 50A is very, very large. For optimal performance, one should pick a box size that is just big enough, such that the electronic density vanishes at its edges. In your case, 18A x 18A x 18A is sufficient. This choice should speed things up significantly, and keep you away from the bug I mentioned earlier.

As a side note, CP2K is better optimized for MPI parallelization rather than OMP. From your output file, I can see that you run with 1 MPI rank and 36 OMP threads. I would suggest something like 9 MPI ranks and 4 OMP threads instead (I get a speedup of 10 that way).

Best,

Augustin

On 1/13/22 21:49, Moe El wrote:

Hi Everyone,

I am trying to run a static energy and geometry optimization of an rgd chain; however, I always keep getting an error related to the solver. I followed the tutorial online and tried to use the same methodology for my molecule. I am very new to CP2K and I dont know what I am doing wrong. I would like any feedback for improvement. Thank you all for your time and effort.

--
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...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/f9c3b3dd-f39b-496b-8c1d-aa5acd2d313an%40googlegroups.com.

-- 
Augustin Bussy
Postdoc Researcher
--------------------------

University of Zurich
Augustin Bussy
Department of Chemistry
Research Group Hutter
Winterthurerstrasse 190
CH-8057 Zurich

www.chem.uzh.ch
augusti...@chem.uzh.ch
--------------------------