[CP2K-user] [CP2K:20549] DFT-D4 issues

[Jürg Hutter]

Hi

CP2K uses an external library for D4. Currently, this library is not running in parallel and the
number of parameters that can be adjusted is minimal. You can change the cutoff radius (D4_CUTOFF)
and the coordination cutoff  D4_CN_CUTOFF for possible speedups.
In the CP2K internal implementation for D3, there are some additional features used to speed up
the calculation (besides parallelization), e.g. constant C6 terms for the 3-body terms (REFERENCE_C9_TERM).

regards
JH

________________________________________
From: cp2k at googlegroups.com <cp2k at googlegroups.com> on behalf of mayank... at gmail.com <mayank.dodia at gmail.com>
Sent: Friday, August 9, 2024 6:17 PM
To: cp2k
Subject: [CP2K:20540] DFT-D4 issues

Hi,

I am using CP2K-2024.1 (git:96bff0e) installed via toolchain as

./install_cp2k_toolchain.sh --with-gcc=install --with-openmpi=system --with-openblas=system --with-sirius=no --with-gsl=no --with-spfft=no --with-spla=no --with-spglib=no --with-spla=no --with-dftd4=install

I was testing the dft-d4 module in CP2K for DFT NVT BOMD of 64 water molecules at 300K. The input file is a modified version of H2O-64.inp, which is attached below. For reference, when using the GGA revPBE-D3(BJ) functional, the simulation runs with no issues.

598          299.000000         0.299941422       330.589518254     -1104.202651786     -1103.888925173         6.335769739
599          299.500000         0.301541403       332.352985832     -1104.203091753     -1103.888929962         6.343083243
600          300.000000         0.304829862       335.977460517     -1104.206316541     -1103.888980814         6.294001573
601          300.500000         0.309924126       341.592257168     -1104.211543588     -1103.889060265         6.321073636
602          301.000000         0.316551623       348.896953508     -1104.217560779     -1103.889142976         5.964969439
603          301.500000         0.322822897       355.809028278     -1104.223085700     -1103.889206038         6.120740931
604          302.000000         0.327538416       361.006380378     -1104.227137720     -1103.889238373         5.925923496

The computation time per step here is ~6 sec on a 32-core Intel(R) Xeon(R) Silver 4210. The temperature increase of ~60K is expected due to DFT functional change. However if I switch to dft-d4, then

77           38.500000         0.862684099       950.833395090     -1103.274695245     -1102.386378043        91.672277213
78           39.000000         0.881930821       972.046751741     -1103.282855444     -1102.374307902        90.513970271
79           39.500000         0.882748826       972.948340710     -1103.287092185     -1102.380456423        90.920328229
80           40.000000         0.896756672       988.387512978     -1103.287828468     -1102.367459596        91.310964988
81           40.500000         0.958453301      1056.388319792     -1103.284416570     -1102.302414268        90.900099128
82           41.000000         1.038365692      1144.466180877     -1103.270000131     -1102.208050576        91.362470460
83           41.500000         1.079534923      1189.842095377     -1103.244176025     -1102.140472954        91.729197830
84           42.000000         1.097126299      1209.230962278     -1103.212461950     -1102.090361488        90.937799281
85           42.500000         1.105902625      1218.904056354     -1103.178147196     -1102.046945536        90.644135027
        86           43.000000         1.101998771      1214.601305186     -1103.141896114     -1102.013002638        90.902946438

I can notice the energy divergence and the computation time is stable at around ~91 sec, which in my experience, is more than for the equivalent revPBE0-D3(BJ) ADMM simulation.

The relevant section in the input files are :
DFT-D3

       &VDW_POTENTIAL
         POTENTIAL_TYPE  PAIR_POTENTIAL
         &PAIR_POTENTIAL
           R_CUTOFF     1.0000000000000009E+01
           TYPE  DFTD3(BJ)
           PARAMETER_FILE_NAME ./dftd3.dat
           REFERENCE_FUNCTIONAL revPBE
           CALCULATE_C9_TERM  T
           REFERENCE_C9_TERM  T
           LONG_RANGE_CORRECTION  F
         &END PAIR_POTENTIAL
       &END VDW_POTENTIAL

DFT-D4

       &VDW_POTENTIAL
         POTENTIAL_TYPE  PAIR_POTENTIAL
         &PAIR_POTENTIAL
           R_CUTOFF     1.0000000000000009E+01
           !TYPE  DFTD3(BJ)
           TYPE  DFTD4
           !PARAMETER_FILE_NAME ./dftd3.dat
           REFERENCE_FUNCTIONAL revPBE
           !CALCULATE_C9_TERM  T
           !REFERENCE_C9_TERM  T
           LONG_RANGE_CORRECTION  F
         &END PAIR_POTENTIAL
       &END VDW_POTENTIAL

I haven't used DFT-D4 before, so I am not sure if I need to enter any addtional parameters in the dispersion block. But the 15:1 computation time ratio between dft-d4 and dft-d3bj suggest there is some issue in the simulation setup. Can you suggest any relevant changes I need to make resolve this situation?

Best Regards,
Mayank


--
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/49eaaddb-c3f9-45ed-ac49-c91bdf6c73d7n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/49eaaddb-c3f9-45ed-ac49-c91bdf6c73d7n%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/ZR0P278MB0759E8DDE708B96D57D2E1059F862%40ZR0P278MB0759.CHEP278.PROD.OUTLOOK.COM.