[CP2K-user] [CP2K:18231] Re: How to Gaussian type calculations and Performance

[Jürg Hutter]

Hi

maybe you are missing the correlation part of the M06 functional?

MGGA_C_M06_2X

I assume this was included in the Gaussian calculation.
Also check that you have the same number of basis functions.
Gaussian is not very consistent in using spherical/cartesian functions.
CP2K always uses spherical.

regards
JH

________________________________________
From: cp2k at googlegroups.com <cp2k at googlegroups.com> on behalf of DMT <dobromirak at gmail.com>
Sent: Tuesday, December 20, 2022 3:27 PM
To: cp2k
Subject: [CP2K:18230] Re: How to Gaussian type calculations and Performance

Without the MIXING the calculation time dropped to only 5x - 6x times that of Gaussian (from 10x).

The subroutines' timings seem proportionate to those before, just less total time.

The total Energy is exactly the same a before: -238.613 Hartree, compared to Gaussian's -240.219 Hartree. Without HF Ex for m062x -219.981.

On Tuesday, December 20, 2022 at 4:15:17 PM UTC+2 DMT wrote:
Dear Marcella,

!!!> Yes, I had posted the wrong input file. Now I am attaching the real one + the output.

Thank you for clarifying the MIXING option.

On an average a single micro-iteration takes 6.2 seconds, the total number of micro-iterations for convergence is 33.

The first 11 subroutines seem to take the most time:

 CP2K                                 1  1.0    0.046    0.046  210.153  210.153
 qs_energies                          1  2.0    0.000    0.000  209.392  209.392
 scf_env_do_scf                       1  3.0    0.000    0.000  208.603  208.603
 scf_env_do_scf_inner_loop           33  4.0    0.002    0.002  208.603  208.603
 qs_ks_update_qs_env                 33  5.0    0.000    0.000  182.059  182.059
 rebuild_ks_matrix                   33  6.0    0.000    0.000  182.057  182.057
 qs_ks_build_kohn_sham_matrix        33  7.0    0.003    0.003  182.057  182.057
 hfx_ks_matrix                       33  8.0    0.000    0.000  110.000  110.000
 integrate_four_center               33  9.0    1.361    1.361  109.993  109.993
 integrate_four_center_main          33 10.0    0.180    0.180  106.232  106.232
 integrate_four_center_bin         1610 11.0  106.052  106.052  106.052  106.052,

the rest take below 30 average total time.

As for the column ASL, the largest are:
pw_scatter_s                       562 13.2<-ASL (the second number)
fft3d_s                           1134 12.7
fft_wrap_pw1pw2_200                597 11.9
integrate_four_center_bin         1610 11.0
xc_pw_derive                       198 11.0
fft_wrap_pw1pw2                   1133 10.7
integrate_four_center_main          33 10.0
xc_rho_set_and_dset_create          33 10.0
xc_pw_divergence                    33 10.0
.
On Tuesday, December 20, 2022 at 3:36:27 PM UTC+2 Marcella Iannuzzi wrote:
Dear Dobomir,

The posted input refers to a pseudo potential calculation using GPW and not to an ALL electron calculation using GAPW.
In this case the total energies of Gaussian and CP2K are not comparable.
But maybe you posted the wrong file.
Anyway, for molecules please remove the Broyden mixing, which is a G-space mixing.
Concerning the timings, is the single iteration time too long or the number of iterations to convergence too large?
In the final timings written in the output it is possible to check which part of the calculation is taking large portion of the allocated resources.

Kind regards
Marcella



On Tuesday, December 20, 2022 at 2:01:25 PM UTC+1 DMT wrote:
Hello,

I am new to CP2K and this is my first post in the group. I have heard many positive things about the program and I want to study it. First I want to learn how to do the kind of calculations I am already used to use other software for.

I have several questions and I'm mainly interested in organic systems.

I picked a very simple metalo-organic complex for test of accuracy, properties and speed (my resources are not unlimited).

The Gaussian calculation I chose to repeat is very simple - Single Point (Energy run) with m062x functional and def2-tzvp basis set. I used the CP2K Energy tutorial and managed to put together the attached input file.

To my understanding I have to use QS > METHOD GAPW in order to apply the basis set, which I downloaded in CP2K format from the Basis Set Exchange web site.
Since the system is not PERIODIC I used POISSON_SOLVER WAVELET.
I have left DIAGONALIZATION ON with ALGORITHM STANDARD.
I also left MIXING T and used BROYDEN_MIXING with ALPHA 0.4 (default) and NBROYDEN 8 (from input examples).
I assume in MGRID: NGRIDS 5 and CUTOFF 400 are good values (from other inputs)
In the &XC_FUNCTIONAL I used &HYB_MGGA_X_M06_2X and in the &HF section I placed FRACTION 0.54 (54% HF Exchange is the default in m062x).
In the KIND X sections I wrote BASIS_SET def2-tzvp and POTENTIAL ALL, which to my understanding forces no core potential at all, but instead uses all electrons from the basis. I noticed that in the &DFT section I had to place POTENTIAL_FILE_NAME POTENTIAL, because although I'm not using a potential CP2K gives an error if I don't provide a potential file. I took the file from the "data" dir of the cp2k source code. <<< Is this the correct way to do it, when there is no potential ?
My ABC setting in &CELL is 12 12 12, although the system is quite a bit smaller in distances between nuclear coordinates - otherwise I get a warning about density at the box edges and convergence is even slower.

Are those setting fine, if I simply want to repeat a Gaussian SP job for testing ? Have I picked the DFT functional correctly ? How about the HF part ? Can you tell me in short what exactly is this MIXING and when to use it ?

CP2K performed quite a few times slower than Gaussian (about 10x). I don't want to state the exact timings (I checked with the command `time'), because Gaussian License prohibits publishing of timing information.
Without the &HF section the performance increases 2x to 3x, but I guess that is for the lack of HF Exchange, which simply makes m062x wrong. The Energy without HF Ex is considerably higher. Wither way Gaussian gives a much lower result for the total Energy. What is going on and am I doing it the right way ?

I understand CP2K has its strengths and in no way I want to speak against it, I just want to learn how to use it and before I get into PERIODIC systems, surface interactions / reactions, solvents and molecular crystals (where I bet CP2K is much better than Gaussian) I want to learn the basics and repeat the kind of calculations I am used to.

Any comment can be of help. Anything to get to a well done calculation and with an increased performance, without sacrificing accuracy too much.

Best Regards,
Dobromir

--
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<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/0f21b966-0a60-4921-bbb1-f01da159a908n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/0f21b966-0a60-4921-bbb1-f01da159a908n%40googlegroups.com?utm_medium=email&utm_source=footer>.

-- 
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/ZR0P278MB07598D0FA4FB1422BF1BA3B09FEA9%40ZR0P278MB0759.CHEP278.PROD.OUTLOOK.COM.