[CP2K-user] [CP2K:20868] All-electron Geometry Optimization of EMIBF4

[Jürg Hutter]

Hi

a few worrying points from your input/output

1) You specify a functional with 100% PBEX and 54% HFX. Most likely that is not your intention?
     See the examples on how to correctly define hybrid functionals.

2) You are starting with SCREEN_ON_INITIAL_P TRUE but it seems there was no restart wavefunction
     available (see output). This will cause trouble as integrals are screen on the initial guess.

3) Your basis (6-31G*) is very small. In fact it is even smaller than the ADMM basis you use (admm-2).
     This is against the very idea of ADMM. With such a small basis you don't need ADMM.
     However, I would use a reasonable basis to start with. If you insist on doing all-electron calculations
     then I would use def2-svp or even def2-tzvpp. The ADMM basis should be chosen accordingly.

4) If you use OT, the mixing section will be ignored. I prefer to delete not used sections to avoid confusion.

5) Your box size (50 Ang) is on the large side for this system size. You probably loose time there too.

regards
JH

________________________________________
From: cp2k at googlegroups.com <cp2k at googlegroups.com> on behalf of Nicholas Laws <nicholaslaws8 at gmail.com>
Sent: Wednesday, November 6, 2024 4:46 PM
To: cp2k
Subject: Re: [CP2K:20867] All-electron Geometry Optimization of EMIBF4

Hi Matthias,

Thank you for the insight! I have since updated my input script and run geometry optimizations for EMIBF4, first with the PBE exchange-correlation functional and then, with the optimized structure, with the WB97X-D exchange-correlation functional. Then, I performed this same workflow for EMI+. Now, I am attempting to perform an NVT ensemble at 200 K on both the geometry optimized EMIBF4 and EMI+, in the same simulation. I have run several versions of the attached .INP file, by adding/removing the &MIXING, with METHOD BROYDEN_MIXING, and turning the EPS_DEFAULT, EPS_SCF, and EPS_SCHWARZ knobs. I have attached .OUT file that describes the lack of energy convergence in the first optimization step and the large run-times per SCF iteration (between 40 and 90 seconds). I was wondering if there was any suggestions on how to improve convergence and runtime for this setup? Also, would it be better to run these molecules in an NVT ensemble separately? The ultimate goal for this workflow is to run an NVE ensemble where the EMI+ ion acts as a fast moving molecule, colliding with the slow moving EMIBF4 neutral. I am trying to study the charged fragmentation products of the EMI+-EMIBF4 collision event. I am currently running this system on a 32-core machine with a 256 GB limit. Thank you for your time and I look forward to hearing from you.

All my best,
Nick

On Tuesday, November 5, 2024 at 4:24:27 AM UTC-5 Krack Matthias wrote:
Hi Nick

You set “RESTART_HESSIAN TRUE” but, most likely, you do not have a .hessian file.

Best

Matthias

From: cp... at googlegroups.com <cp... at googlegroups.com> on behalf of Nicholas Laws <nichol... at gmail.com>
Date: Monday, 4 November 2024 at 20:56
To: cp2k <cp... at googlegroups.com>
Subject: Re: [CP2K:20853] All-electron Geometry Optimization of EMIBF4
Dear Professor Hutter,

I hope you are doing well! I have since updated the .inp file to reflect your suggestions. I am currently facing a leading blanks issue during the first optimization step:

*** WARNING in common/cp_files.F:376 :: The file name <> has leading ***
 *** blanks.

I was wondering if you or the community had any suggestions to fix this problem. I have attached the updated .inp file and the .out file from the job submission. Please let me know if you need any additional information and thank you for your time.

All my best,
Nick

On Thursday, October 24, 2024 at 3:54:02 AM UTC-4 Jürg Hutter wrote:
Hi
you should do the following steps to get this running:

1) Use OT optimizer
&SCF
MAX_SCF 50
SCF_GUESS RESTART
EPS_SCF 1E-6
&OT
PRECONDITIONER FULL_ALL
MINIMIZER DIIS
&END OT
&END SCF
2) Use EPS_Defualt 1.E-10
3) Optimize the structure (at least SCF, maybe geometry) first with a GGA functional (e.g. PBE)
4) Restart from that orbitals and use SCREEN_ON_INITIAL_P TRUE
5) Increase MAX_MEMORY (if possible) so that all integrals are kept incore
6) USE ADMM
&AUXILIARY_DENSITY_MATRIX_METHOD
ADMM_TYPE ADMMQ
EXCH_CORRECTION_FUNC PBEX
&END AUXILIARY_DENSITY_MATRIX_METHOD
with (add it)
BASIS_SET_FILE_NAME BASIS_ADMM_ae
and
BASIS_SET AUX_FIT admm-2
7) Use BFGS for geometry optimizer, maybe even restart the Hessian from a previous GGA optimization
8) if the geometry optimization is not converging you might have to tighten the thresholds
EPS_DEFAULT, EPS_SCF, EPS_SCHWARZ

With these changes, you need about 40 Gb of memory and on 36 core a geometry step takes 1-2 minutes.

regards
JH


________________________________________
From: cp... at googlegroups.com <cp... at googlegroups.com> on behalf of Nicholas Laws <nichol... at gmail.com>
Sent: Wednesday, October 23, 2024 5:30 PM
To: cp2k
Subject: Re: [CP2K:20806] All-electron Geometry Optimization of EMIBF4

Thank you for the advice, Professor Hutter. I have since referenced the all-electron examples in the tests/QS sections and adjusted my input script accordingly (as seen in the attached .inp file). I have decided to separate the geometry optimizations into two separate simulations (one for EMIBF4 and one for EMI+). The content of this post refers to the geometry optimization of EMIBF4. However, after testing several SCF and GEO_OPT configurations, it seems that there are energy instabilities that occur after 5 energy evaluations as depicted here:

SCF WAVEFUNCTION OPTIMIZATION

Step Update method Time Convergence Total energy Change
------------------------------------------------------------------------------

HFX_MEM_INFO| Est. max. program size before HFX [MiB]: 91626
HFX_MEM_INFO| Number of cart. primitive ERI's calculated: 150682718957
HFX_MEM_INFO| Number of sph. ERI's calculated: 54286634411
HFX_MEM_INFO| Number of sph. ERI's stored in-core: 29731009090
HFX_MEM_INFO| Number of sph. ERI's stored on disk: 0
HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 24555625321
HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: 64051
HFX_MEM_INFO| Whereof max-vals [MiB]: 193
HFX_MEM_INFO| Total compression factor ERI's RAM: 3.54
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]: 4
HFX_MEM_INFO| Size of buffers [MiB]: 5
HFX_MEM_INFO| Est. max. program size after HFX [MiB]: 91793

1 Pulay/Diag. 0.20E+00 105.6 93.15134943 -786.8938839625 -7.87E+02
2 Pulay/Diag. 0.20E+00 59.5 1424.78843686 -787.9112236525 -1.02E+00
3 Pulay/Diag. 0.20E+00 59.6 4.0852E+05 -804.4927698695 -1.66E+01
4 Pulay/Diag. 0.20E+00 58.8 2.6603E+05 -1012.0913997227 -2.08E+02
5 Pulay/Diag. 0.20E+00 60.1 1.2150E+05 -3042.9246578949 -2.03E+03
6 Pulay/Diag. 0.20E+00 58.7 1.3437E+05 -4156.9505747502 -1.11E+03
7 Pulay/Diag. 0.20E+00 59.4 58430.75278087 -4430.8543850289 -2.74E+02
8 Pulay/Diag. 0.20E+00 58.9 47799.10199568 -4482.5049850590 -5.17E+01
9 Pulay/Diag. 0.20E+00 59.9 42678.37923542 -4460.2090853173 2.23E+01
10 Pulay/Diag. 0.20E+00 58.6 29018.77397837 -4448.4858203716 1.17E+01
11 Pulay/Diag. 0.20E+00 60.1 30422.75295948 -4458.1250435109 -9.64E+00
12 Pulay/Diag. 0.20E+00 58.8 20157.27665011 -4462.7197837500 -4.59E+00
13 Pulay/Diag. 0.20E+00 60.2 20697.03326037 -4465.4866843212 -2.77E+00
14 Pulay/Diag. 0.20E+00 58.6 14035.03955462 -4464.6656749474 8.21E-01
15 Pulay/Diag. 0.20E+00 59.8 13859.56913168 -4455.4582585163 9.21E+00
16 Pulay/Diag. 0.20E+00 58.6 9844.63604592 -4439.6169244281 1.58E+01
17 Pulay/Diag. 0.20E+00 59.7 9086.46419396 -4429.7590093298 9.86E+00
18 Pulay/Diag. 0.20E+00 59.0 6252.68491596 -4451.0446941090 -2.13E+01
19 Pulay/Diag. 0.20E+00 59.8 5836.44669985 -4446.6123587806 4.43E+00
20 Pulay/Diag. 0.20E+00 58.7 4369.58978109 -4434.7404123014 1.19E+01
21 Pulay/Diag. 0.20E+00 59.8 3709.18259540 -4422.7286506587 1.20E+01
22 Pulay/Diag. 0.20E+00 58.8 2919.47259531 -4423.1759843663 -4.47E-01
23 Pulay/Diag. 0.20E+00 59.8 2418.12012615 -4420.5319920017 2.64E+00
24 Pulay/Diag. 0.20E+00 58.8 1984.95398922 -4419.3227006816 1.21E+00
25 Pulay/Diag. 0.20E+00 60.0 1707.00425122 -4420.3727826353 -1.05E+00
26 Pulay/Diag. 0.20E+00 59.0 1274.71842881 -4420.2666697595 1.06E-01
27 Pulay/Diag. 0.20E+00 60.3 1417.13141141 -4425.4603136343 -5.19E+00
28 Pulay/Diag. 0.20E+00 58.8 1104.03431671 -4418.0031624370 7.46E+00

I was wondering, if there is any guidance or suggestions on how to handle these energy fluctuations when using the aug-cc-pvtz basis set and WB97X-D XC-functional? Additionally, I noticed that runtimes are approximately 1 minute per SCF iteration, I was curious if there was any advice on how to improve this runtime? I've attached a sample SLURM submission script that I used to generate the results located in the attached .out file. Please let me know if there is any additional information that I can provide and I greatly appreciate the support that you and the community provide.

All my best,
Nick

On Tuesday, October 22, 2024 at 3:48:22 AM UTC-4 Jürg Hutter wrote:
Hi

you are missing the &HF section in your specification of the hybrid functional.
Libxc only covers the density functional part, see the many examples in the tests/QS
sections.

regards
JH

________________________________________
From: cp... at googlegroups.com <cp... at googlegroups.com> on behalf of Nicholas Laws <nichol... at gmail.com>
Sent: Monday, October 21, 2024 4:14 PM
To: cp2k
Subject: [CP2K:20794] All-electron Geometry Optimization of EMIBF4

Hi all,

I am trying to do a geometry optimization for energy minimization of EMIBF4 using the aug-cc-pvtz basis set (attached below) and WB97X-D XC-functional. It seems that my optimization requires hundreds of SCF steps before convergence (as seen in the attached .out file) and I was wondering if there are any recommendations for doing all-electron geometry optimizations, especially for the one I discuss in this post (current implementation can be viewed in the attached .inp file)? Please let me know if there any additional information that I can clarify.

Thank you, and I look forward to hearing from you.

All my best,
Nick

--
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<mailto:cp2k+uns... at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/9ae46c98-afd4-4625-b769-6cf39a3fb5d2n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/9ae46c98-afd4-4625-b769-6cf39a3fb5d2n%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+uns... at googlegroups.com<mailto:cp2k+uns... at googlegroups.com>.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/dbffa255-5afa-43c7-8bd5-875420594c7cn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/dbffa255-5afa-43c7-8bd5-875420594c7cn%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+uns... at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/25f28c2f-4d10-47df-8a12-09b26799171bn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/25f28c2f-4d10-47df-8a12-09b26799171bn%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<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/4c856dd6-9910-42ec-9407-a9622d5823b0n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/4c856dd6-9910-42ec-9407-a9622d5823b0n%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 visit https://groups.google.com/d/msgid/cp2k/ZR0P278MB07593B105612F46CC579AEE39F5C2%40ZR0P278MB0759.CHEP278.PROD.OUTLOOK.COM.