Tiger Team: tt-electrochem-dftb+
Modeling Electrochemical Systems with DFTB
Installation of DFTB+ quantum mechanical simulation software
package tailored to the underlying hardware architecture
on JUSTUS 2 and finding the optimal parallelization scheme
for the specific electrochemical models system was achieved
in the collaborative effort between the scientists from a group
of theoretical chemistry and the HPC experts of the
Competence Center for Computational Chemistry and Quantum Sciences.
The part of the project was also to link DFTB+ with PLUMED library.
Electrochemistry treats the interface between an electronic conductor,
the electrode, and an ionic conductor, the electrolyte. Any realistic
model must treat both adjoining phases in the same atomic detail.
Most groups use density functional theory (DFT) to model the
interface; this works well for the electrode, but lacks the power
to include a realistic part of the solution.
In this project we use DFTB, a DFT based tight binding model,
which is much faster than DFT and allows the treatment of much
larger systems, including electrode and solution. A new version
of a corresponding software called DFTB+ has just been released,
which makes it possible to perform biased molecular dynamics
(metadynamics) by linking to the package plumed. Installing and
using DFTB+ with plumed is technically challenging, and requires
a close cooperation between experts from bwHPC and theoretical
chemists1.
The installation of the state of the art DFTB+ of version 20.2.1
interfaced with PLUMED of version 2.7.0 was performed based on the
Intel's family compiler 19.1.2 and MPI library 2019.8.
The monitoring of the test jobs revealed relatively low job
efficiency when adopting just the default parameters. Following the
instructions of the code developers, we were able to achieve a
substantial improvement in the job performance while using the same
amount of the CPU resources. This was possible by adjusting
the process groups in the parallel block with respect to
the size of the model, k-points, spins etc. Additional speedup
was obtained by adapting the software to use advanced libraries,
special eigen-solver and further numerical libraries like
EIGENEXA, MAGMA, ARPACK, or ELSI.
With the new installation of DFTB+ on JUSTUS 2 the scientists
achieved a speedup exceeding 1000% in comparison to the installation
and the original workflow reached on the equivalent amount of CPUs
on the preceding cluster system JUSTUS 1.
[1] E. Santos and W. Schmickler, Hydrogen adsorption on doped graphene investigated by a DFT-based tight-binding method, J. Phys.: Condens. Matter 33 (2021) 504001
Members of the Tiger-Team:
Prof. Dr. W. Schmickler, Dr. E. Santos, Institute of Theoretical Chemistry, Ulm University;
Dr. J. Kucera, HPC Competence Center for Computational Chemistry and Quantum Sciences, Ulm University
Status:
finished.