The archive has to be untared with
tar xfz DH_benchmark.tar.gz

The files *BASIS* and POTENTIAL_UZH contain the necessary basis sets and pseudopotentials for all calculations.

The directories high, medium and low contain the files of the three water-containing systems. Each directory has the structure
<density>/<basis set>/<method>/<filename>.
With <density> refering to the three densities, <basis set> being a double zeta ("DZ") or triple zeta ("TZ") quality primary basis set, <method> is either "PBE", "no_ADMM" (double-hybrid without ADMM) and "ADMM" (double-Hybrid with ADMM).

Similarly, the directories HCN, NH3 and Ar contain the files for the other three test systems.
After <method>, there is another level refering to the systems to be considered for the calculation of the BSSE-corrected cohesive energies. These are "crystal" for the crystal, "ghost" for the atom or molecule with a suitable number of ghost atoms and ghost molecules, "Mon_bulk" the geometry of a single molecule with the crystal geometry (not for Ar) and "Mon_gas" with the geometry of a single molecule with an optimized geometry.

The directory "mpi" contains the calculations for the strong scaling plot. The subdirectories refer to the number of nodes on the Piz Daint XC50 super computer.

The directories "benzene" and "TiO2" contain the files of the large benchmark system with subdirectory structure
<system>/<method>/<filename>.

The files are either CP2K input files (*.inp), CP2K output files (*.out, does not start with "slurm"), run files (run.sh), Slurm output files (slurm-<JobID>.out) and CP2K wave-function restart files (*.wfn).

All calculations have been carried out with 

- a CP2K binary, self-compiled from the CP2K master trunk (potentially later available in CP2K 9.x).
- Libint 2.6
- COSMA 2.5.1
- cray-libsci and cray-libsci_acc (use the vendor provided BLAS/Lapack/PBLAS/ScaLapack libraries, GPU acceleration recommended)

If you try to reproduce these calculations, adjust all relevant files accordingly to your directory structure and use your available scheduler.
The smaller calculations take a ca. 1 TB of memory and ca. 200 CPU hours.
The large benchmark systems take ca 30-40 TB of memory and ca. 6000 CPU hours.
Preoptimize a PBE wave-function and then start the Double-Hybrid calculation. You might use other Double-Hybrids then us.

Later versions of CP2K or of the libraries might perform better than here.

For further information, ask the authors

Frederick Stein (Frederick.Stein@chem.uzh.ch)
Juerg Hutter (hutter@chem.uzh.ch).

Zurich, December 2021