Dcdftbmd: Divide‐and‐Conquer Density Functional Tight‐Binding Program for Huge‐System Quantum Mechanical Molecular Dynamics Simulations

Abstract

Dcdftbmd is a Fortran 90/95 program that enables efficient quantum mechanical molecular dynamics (MD) simulations using divide‐and‐conquer density functional tight‐binding (DC‐DFTB) method. Based on the remarkable performance of previous massively parallel DC‐DFTB energy and gradient calculations for huge systems, the code has been specialized to MD simulations. Recent implementations and modifications including DFTB extensions, improved computational speed in the DC‐DFTB computational steps, algorithms for efficient initial guess charge prediction, and free energy calculations via metadynamics technique have enhanced the capability to obtain atomistic insights in novel applications to nanomaterials and biomolecules. The energy, structure, and other molecular properties are also accessible through the single‐point calculation, geometry optimization, and vibrational frequency analysis. The available functionalities are outlined together with efficiency tests and simulation examples. © 2019 Wiley Periodicals, Inc.

Density-functional tight-binding: basic concepts and applications to molecules and clusters

“…Standard DFTB2 is 10 2 to 10 3 times faster than even local functionals, and even more if compared with higher-level functionals such as hybrid, double hybrid or LC-corrected functionals. Algorithmic schemes achieving linear scaling with the number of atoms in solving the DFTB Hamiltonian [21,376,[379][380][381][382] such as the Divide and Conquer techniques [21,381,382] or cluster type algorithms [376] have now proved the feasibility of calculations on extremely large systems up to one million atoms at least for covalent or intermolecular complexes (see Figure 12: a box of 350000 water molecules), even though one should mention that the case of metals remains more delicate due to electronic delocalization. Even if large scale dynamical simulations on such huge systems are not yet practicable, DFTB certainly stands as a promising method to address simulations of systems with up to 10000 atoms on the next generation of High-Performance Computing architectures, which would be quite helpful for theoretical investigation of properties and processes involved in the chemistry and physics of large molecular systems, possibly biomolecules, or in nanoparticle physics.…”

Hierarchical parallelization of divide‐and‐conquer density functional tight‐binding molecular dynamics and metadynamics simulations

“…D cdftbmd [ 65 ] is a DFTB implementation that features efficient energy and force calculations combined with the DC method where its acceleration can be attributed to hybrid parallelization using a message passing interface (MPI) and open multi‐processing (OpenMP). Unbiased MD and (WT)MetaD simulations are already available in the program, as noted in the first section.…”

“…To date, fast large‐scale chemical reaction simulations combining DFTB with DC have been demonstrated for MD [ 57–64 ] and MetaD [ 65,66 ] (DC‐DFTB‐MD/MetaD). Herein, the extension of DC‐DFTB‐MD/MetaD for running simulations simultaneously for a set of geometries in the same system called walkers or replicas is reported to improve sampling and free energy evaluation efficiency.…”

GPU‐Accelerated Large‐Scale Excited‐State Simulation Based on Divide‐and‐Conquer Time‐Dependent Density‐Functional Tight‐Binding

“…The above DC‐TDDFTB energy and energy gradient calculations are implemented in the Dcdftbmd program . In the next subsections, we explain the GPU implementation for these calculations.…”

“…Our group has developed the DC‐based HF, post‐HF theories, their gradient calculations, and excited‐state properties . Recently, quantum mechanical molecular dynamics (QM‐MD) simulations for tens of thousands atoms in the ground states have been accomplished by applying the DC method to the density‐functional tight‐binding (DFTB) method, which is an approximation to DFT with the concept of adopting up to two‐center terms for parameterized integrals and repulsive potential. Furthermore, the combination of the DC method and the time‐dependent (TD) DFTB method was also examined …”