GAMESS - Release Notes | LibXC Interface - Release Notes

LibXC Interface - Release Notes

June 30, 2021 R1 Public Release

LibXC 5.1.5 is used.

Functionals

  1. These functionals were added as pre-defined:
  2. These functionals were fixed in the LibXC interface:
  3. These functionals were fixed in the LibXC library:
  4. Functional mixing can be enabled in two ways:
  5. Kinetic functionals were removed: GAMESS(US) code should be more flexible before they can be enabled
  6. Double-hybrid functionals will fail when trying to calculate gradients

  7. Discrepancies between GAMESS(US) and LibXC energies of functionals (more than 1e-6 Hartree):

    Functional GAMESS(US) Energy LibXC Energy Difference Status
    rdft-HCTH120 -154.101432826 -154.101426930 5.9e-06 Unknown
    rdft-HCTH407 -154.108748553 -154.108651459 9.7e-05 Unknown
    rdft-MOHLYP2 -153.934665050 -153.934632077 3.3e-05 Unknown
    rdft-REVTPSS -154.225528646 -154.085255196 1.4e-01 Broken in GAMESS
    rdft-X3LYP -154.041899667 -154.041917751 1.8e-05 Unknown
    udft-HCTH120 -192.591417621 -192.591410532 7.1e-06 Unknown
    udft-HCTH407 -192.582774257 -192.582670977 1.0e-04 Unknown
    udft-MOHLYP2 -192.474560872 -192.474526664 3.4e-05 Unknown
    udft-REVTPSS -192.746790511 -192.577317509 1.7e-01 Broken in GAMESS
    udft-X3LYP -192.551370381 -192.551389248 1.9e-05 Unknown

Documentation

  1. Functionals sorted by their place in Jacob's ladder and their family
  2. Fix: MGGA functionals are not included in GGA sections

Generators

  1. Generators were merged into one file
  2. The Generators now use the Python API to LibXC (pylibxc) instead of parsing headers

Compilation and installation

  1. download-libxc.csh and download-libxc-tests.csh were expanded: loading of LibXC library and LibXC tests archives does not necessarily require an active internet connection. If no arguments are given, the scripts will try to get the archives from ${GMS_PATH}. If this procedure is not successful, the scripts will try to download from the Internet. If one argument is given, the scripts will try to use the argument as the path where the library or tests are.
  2. For Windows, download-libxc.csh and download-libxc-tests.csh were updated.

Program-related changes

  1. mod_nameio's interface was changed: use only group_input -- other needed routines for reading input are in this class
  2. mod_nameio can check allowed keys
  3. mod_nameio can show the location where an input error happened. It shows the key where the processing stopped.
  4. Fix: mod_nameio can correctly process groups with similar names (for example, GGA_XC and MGGA_XC)
  5. Bibliography indexes in .log are now sorted correctly
  6. functional.src provides a functional class so that, theoretically, several functionals can be constructed in one GAMESS(US) run
  7. Functional module only supports routines with arrays
  8. Into the add_functional routine of functional class, the optional arguments hfex, alpha, beta, and omega were added for getting coefficients of hybrid functionals from LibXC library
  9. Fixed memory problem in the interface when TD calculations are
  10. Added utils_strings with some routines which work with strings
  11. Arrays in functionals.src are deallocated when GAMESS(US) ends
  12. Fix: GAMESS(US) terminates normally if there are errors produced by the LibXC interface
  13. All double precision type declarations were changed into real(kind=fp)
  14. Since FPE errors in LibXC have been fixed, NaNs checks were removed

Known issues

  1. LibXC interface works incorrectly if it is built using IBM XL compilers

June 30, 2020 R1 Public Release

LibXC 5.0.0 is used.

Functionals

Set DFTTYP=USELIBXC in the $CONTRL group in the input file for enabling the LibXC interface. The choosing of functionals is available in two ways: 1. Pre-defined functionals with fixed parameters. The full list of available functionals can be found in the manual. For example, SCAN0 (0.25 HF + 0.75 SCAN_X + 1.00 SCAN) will be selected:

 $LIBXC FUNCTIONAL=SCAN0 $END
  1. User-defined functionals. Do not set a FUNCTIONAL keyword! $LIBXC HFEX=0.25 MP2OS=0.1 MP2SS=0.05 $END $MGGA_X SCAN=0.65 $END $GGA_X PBE=0.1 $END $MGGA_C SCAN=0.95 $END Here, the functional has the formula 0.25 HF + 0.65 SCAN_X + 0.1 PBE_X + 0.95 SCAN_C + 0.1 MP2OS + 0.05 MP2SS, where HFEX is a fraction of exact exchange (100% is 1.00), MP2OS is a fraction of non-local correlation energy for opposite spin, MP2SS is a fraction of non-local correlation energy for same spin. The MP2 (it sets MP2SS=MP2 and MP2OS=MP2) keyword also exists, but MP2OS and MP2SS have priority. Be careful when mixing MP2 and MP2SS or MP2OS! The full list of available groups and functionals is available in the docs-input.txt. Happy mixing!

The total energies of both functional for He atom (3-21G/300.302) are: 1. -2.8763983686 (SCAN0) 2. -2.8726429406 (MIXED)

Compilation and installation

For using LibXC, set GMS_LIBXC in install.info and Makefile to true. The total compilation process is the following:

./config
./tools/libxc/download-libxc.csh
make ddi
make libxc -j$(nproc)
make modules
make -j$(nproc)

For Windows, see instruction here: https://github.com/gms-bbg/gamess-issues/issues/43

Program-related changes

  1. A new input file parser (mod_nameio.src) that is used by the LibXC interface. The input file parser uses Fortran 2003 features that lead to a broken compilation when using GCC 4.9. The smallest checked version is GCC 5.5, and I hope that GCC 5.0 will be enough. The new module supports reading of the fortran data types logical, integer, double, character and arrays composed of these data types (single array fields are separated by comma) in free format. The XMARKR is not implemented. The maximum length of keywords and the values assigned to keywords (arrays can be entered) are 256 and 256*16 characters, respectively.
  2. The LibXC interface. It is composed of two files: functionals.src and libxc.src. The file functionals.src holds variables that contain information about chosen functionals. This file is the glue between libxc.src and the LibXC library. It calls the LibXC Fortran 2003 API. The file libxc.src contains subroutines for reading and converting arrays between GAMESS(US) and LibXC. Inside libxc.src the code-generated subroutine libxc_choose_functionals exists. Do not edit this routine by hand! The script for generating (gen_function.py) is in tools/libxc/.

Known issues

  1. There functionals are broken (Fixed in 2021R1):

  2. Double-Hybrid functional's gradients are not stopped. Results, gotten by this way, are not correct! (Fixed in 2021R1)

  3. Discrepancies between GAMESS(US) and LibXC energies of functionals (more than 1e-7 Hartree):

    Functional LibXC Energy GAMESS(US) Energy Difference
    rdft-APF -153.985351865 -153.985352102 -0.000000236
    rdft-B3LYP -154.116250921 -154.020768865 0.095482056
    rdft-B3LYPV5 -153.706080859 -154.020768865 -0.314688006
    rdft-B3PW91 -154.061351344 -154.061351912 -0.000000567
    rdft-B97-1 -153.147441042 -154.060554422 -0.913113380
    rdft-CAMQTP00 -154.056370566 -154.056370142 0.000000423
    rdft-HCTH120 -154.101426930 -154.101432826 -0.000005895
    rdft-HCTH147 -154.114536741 -154.092996196 0.021540544
    rdft-HCTH407 -154.108651459 -154.108748553 -0.000097093
    rdft-HCTH76 -155.314847211 -150.731191405 4.583655805
    rdft-HLYP2 -153.572621732 -153.509483856 0.063137875
    rdft-HLYP -153.998025496 -154.400683586 -0.402658089
    rdft-M11-L -153.973653255 -153.973653069 0.000000186
    rdft-MN12-SX -153.703301863 -151.026324056 2.676977806
    rdft-MOHLYP2 -153.934632077 -153.934665050 -0.000032972
    rdft-OP -154.006461178 -154.000471655 0.005989522
    rdft-PBE0M -153.988704195 -153.988704901 -0.000000705
    rdft-PW91 -154.045812806 -154.045813492 -0.000000686
    rdft-REVTPSS -154.085255196 -154.225528646 -0.140273449
    rdft-SOGGA11 -154.407295817 -154.085024608 0.322271208
    rdft-THCTHHYB -153.620194439 -154.093062019 -0.472867579
    rdft-WB97X -154.071539470 -154.071539351 0.000000118
    rdft-X3LYP -154.041917751 -154.041899667 0.000018083
    udft-B3LYP -192.622685369 -192.533460315 0.089225054
    udft-B3LYPV5 -192.227420564 -192.533460315 -0.306039751
    udft-B3PW91 -192.551658365 -192.551658596 -0.000000230
    udft-B97-1 -191.674241984 -192.564035505 -0.889793520
    udft-CAMQTP00 -192.560219070 -192.560218607 0.000000462
    udft-HCTH120 -192.591410532 -192.591417621 -0.000007089
    udft-HCTH147 -192.596962628 -192.575529088 0.021433540
    udft-HCTH407 -192.582670977 -192.582774257 -0.000103279
    udft-HCTH76 UNCONVERGED SCF -189.422648754 ---
    udft-HLYP2 -192.085115537 -192.062433628 0.022681908
    udft-HLYP -192.496262805 -192.889962762 -0.393699957
    udft-MN12-SX -192.198980680 -189.226130389 2.972850290
    udft-MOHLYP2 -192.474526664 -192.474560872 -0.000034207
    udft-OP -192.480853894 -192.488800904 -0.007947009
    udft-PBE0M -192.479410006 -192.479410292 -0.000000286
    udft-PW91 -192.555722859 -192.555723140 -0.000000280
    udft-REVTPSS -192.577317509 -192.746790511 -0.169473002
    udft-SOGGA11 -192.893402626 -192.590418386 0.302984240
    udft-THCTHHYB -192.155470213 -192.577752629 -0.422282416
    udft-WB97X -192.573067117 -192.573067003 0.000000113
    udft-X3LYP -192.551389248 -192.551370381 0.000018866

  4. mGGA functionals can not be used for DFT hessian and TD-DFT gradient calculation

  5. Some of the mGGAs require the Laplacian of the electron density, but GAMESS(US) can not provide it

  6. mod_nameio.src does not show the location of input errors (Partially fixed in 2021R1)

  7. It is not possible to use double-hybrid functionals with different fractions of same-spin and opposite-spin energy components for the U-SCS-MP2 scheme

  8. Some of the flags in -ffpe-trap=invalid,zero,overflow are incompatible with the LibXC library (Fixed in 2021R1)

  9. Some of the LibXC functionals can produce NaNs when energies or their derivatives are calculated. For energies, the NaN is checked, and if, in the point, the NaN is found this point will be skipped. The derivatives are not checked. (Fixed in 2021R1)

  10. Sometimes GAMESS(US) produces a local kinetic energy (tau) that is less than zero. This leads to NaNs for some mGGA functionals. In order to fix that, the absolute value of tau is taken.

  11. The LibXC interface can not be used together with FMO

  12. The LibXC interface works incorrectly if it is built using PGI compilers (Fixed in 2021R1)

Citation

Igor S. Gerasimov, Federico Zahariev, Sarom S. Leang, Anton Tesliuk, Mark S. Gordon, Michael G. Medvedev,
Introducing LibXC into GAMESS (US),
Mendeleev Commun., 2021, 31, 302–305
DOI: 10.1016/j.mencom.2021.05.008
https://doi.org/10.1016/j.mencom.2021.04.008