LiCl is running slower than KCl

Hung Nguyen

unread,

Sep 9, 2024, 11:47:00 AMSep 9

to cp2k

Dear CP2K community,

I am running PBE0-D3 (50% HFX exchange) on several molten salt systems and I found that LiCl system is running much slower than KCl (I used identical input files for them, except coordinate-related information).

The following is HFX information the log file reports:

For LiCl:

HFX_INFO| Replica ID: 1

HFX_INFO| FRACTION: 0.5000000000

HFX_INFO| Interaction Potential: TRUNCATED

HFX_INFO| Cutoff Radius [angstrom]: 6.0000000000

HFX_INFO| EPS_SCHWARZ: 1.0E-06

HFX_INFO| EPS_SCHWARZ_FORCES 1.0E-06

HFX_INFO| EPS_STORAGE_SCALING: 1.0E+00

HFX_INFO| NBINS: 64

HFX_INFO| BLOCK_SIZE: 4

HFX_INFO| NUMBER_OF_SHELLS: AUTO

HFX_INFO| Number of periodic shells considered: -1

HFX_INFO| Number of periodic cells considered: 27

HFX_MEM_INFO| Est. max. program size before HFX [MiB]: 973

HFX_MEM_INFO| Number of cart. primitive ERI's calculated: 1492589796730

HFX_MEM_INFO| Number of sph. ERI's calculated: 1180851126208

HFX_MEM_INFO| Number of sph. ERI's stored in-core: 1180851126208

HFX_MEM_INFO| Number of sph. ERI's stored on disk: 0

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 0

HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: 891545

HFX_MEM_INFO| Whereof max-vals [MiB]: 30785

HFX_MEM_INFO| Total compression factor ERI's RAM: 10.11

HFX_MEM_INFO| Total memory consumption ERI's disk [MiB]: 0

HFX_MEM_INFO| Total compression factor ERI's disk: 0.00

HFX_MEM_INFO| Size of density/Fock matrix [MiB]: 372

HFX_MEM_INFO| Size of buffers [MiB]: 78

HFX_MEM_INFO| Number of periodic image cells considered: 27

HFX_MEM_INFO| Est. max. program size after HFX [MiB]: 12886

For KCl:

HFX_INFO| Replica ID: 1

HFX_INFO| FRACTION: 0.5000000000

HFX_INFO| Interaction Potential: TRUNCATED

HFX_INFO| Cutoff Radius [angstrom]: 6.0000000000

HFX_INFO| EPS_SCHWARZ: 1.0E-06

HFX_INFO| EPS_SCHWARZ_FORCES 1.0E-06

HFX_INFO| EPS_STORAGE_SCALING: 1.0E-01

HFX_INFO| NBINS: 64

HFX_INFO| BLOCK_SIZE: 4

HFX_INFO| NUMBER_OF_SHELLS: AUTO

HFX_INFO| Number of periodic shells considered: -1

HFX_INFO| Number of periodic cells considered: 27

HFX_MEM_INFO| Est. max. program size before HFX [MiB]: 1489

HFX_MEM_INFO| Number of cart. primitive ERI's calculated: 986338569460

HFX_MEM_INFO| Number of sph. ERI's calculated: 678423895453

HFX_MEM_INFO| Number of sph. ERI's stored in-core: 500193908814

HFX_MEM_INFO| Number of sph. ERI's stored on disk: 0

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 178229986639

HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: 608162

HFX_MEM_INFO| Whereof max-vals [MiB]: 6047

HFX_MEM_INFO| Total compression factor ERI's RAM: 6.27

HFX_MEM_INFO| Total memory consumption ERI's disk [MiB]: 0

HFX_MEM_INFO| Total compression factor ERI's disk: 0.00

HFX_MEM_INFO| Size of density/Fock matrix [MiB]: 396

HFX_MEM_INFO| Size of buffers [MiB]: 90

HFX_MEM_INFO| Number of periodic image cells considered: 27

HFX_MEM_INFO| Est. max. program size after HFX [MiB]: 9582

I would appreciate it if someone could help me to point out why LiCl is running much slower KCl.

Best,

Hung

Frederick Stein

unread,

Sep 9, 2024, 11:52:26 AMSep 9

to cp2k

Dear Hung,

What basis sets are you using? Do you apply ADMM?

Best,
Frederick

Hung Nguyen

unread,

Sep 9, 2024, 12:02:15 PMSep 9

to cp2k

Dear Frederick,

I am using TZVP basis set. I am also applying ADMM.

I attached input files I am using.

Best,

Hung

SP-KCl-12480.inp

SP-LiCl-11026.inp

Jürg Hutter

unread,

Sep 9, 2024, 12:04:08 PMSep 9

to cp...@googlegroups.com

Hi

please quantify "much slower".

Odd thing is that according to your output for LiCl you store all integrals incore:
LiCl

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 0

KCl

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 178229986639

However, the total amount of integrals is much larger for LiCl (Li has a more difuse basis than K).

regards
JH

________________________________________
From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Hung Nguyen <hungngu...@gmail.com>
Sent: Monday, September 9, 2024 5:46 PM
To: cp2k
Subject: [CP2K:20669] LiCl is running slower than KCl

--
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<mailto:cp2k+uns...@googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/db784987-f0fe-4f12-88ed-13d1b0c4568dn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/db784987-f0fe-4f12-88ed-13d1b0c4568dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

Debmalya Ray

unread,

Sep 9, 2024, 12:05:58 PMSep 9

to cp...@googlegroups.com

Is your convergence for LiCl worse than KCl? I found systems containing Na and Mg converges slower and needs higher cutoff.

--

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/4911945f-e0c0-41ef-86d5-3367895b3599n%40googlegroups.com.

Hung Nguyen

unread,

Sep 9, 2024, 12:25:40 PMSep 9

to cp2k

Dear Prof. Hutter,

Thank you so much for pointing it out! I have just figured out that in the HFX data, I showed in the prior message, I did use a different amount of memory. My sincere apologies for the confusion!

I have been running a few systems. Some of them have memory calculated on the fly, and some of them not. The following is the result I obtain for LiCl and KCl when I use a lot of memory to make sure that no ERI is calculated on the fly:

For LiCl
1 OT CG 0.80E-01 1919.1 0.00842150 -4418.1754324748 -4.42E+03
2 OT LS 0.13E+00 447.5 -4456.4755208811

For KCl
1 OT CG 0.80E-01 687.7 0.00530909 -8394.4355734905 -8.39E+03
2 OT LS 0.32E+00 197.3 -8440.0336970755

Also, the following is result when I used less memory in KCl and ended up having some ERI calculated on the fly:

1 OT CG 0.80E-01 683.5 0.01650856 -6317.2855863300 -6.32E+03
2 OT LS 0.32E+00 375.8 -6856.0434752195
3 OT CG 0.32E+00 400.3 0.00885142 -7865.8954808822 -1.55E+03

Best,

Hung

Hung Nguyen

unread,

Sep 9, 2024, 12:28:49 PMSep 9

to cp2k

Hello Debmalya!

I have not yet been able to converge any jobs with LiCl since they ran too slowly.

However, I do have jobs running with MgCl2 and they run faster with less ERI calculated than KCl and LiCl.