Next generation quantum based molecular dy-namics: challenges and perspectives
CECAM-DE-MM1P, University of Bremen, Germany
Molecular dynamics (MD) simulation is a powerful computational tool that is widely used in ma-terials science, chemistry and biology. MD simulations provide a general approach to understand and analyze a wide range of material systems and serve as a valuable complement to support and interpret experiments. MD simulations allow us to directly “see” and manipulate phenomena at the atomic scale and evaluate how properties can be modified to optimize specific designs .
The field of MD is currently dominated by classical simulation methods, where the interactions between atoms are described through classical force fields in which the complex electronic structure of molecules and solids is reduced to simple, parameterized interactions between at-oms that are fitted to experiments and/or electronic structure calculations. Unfortunately, classi-cal MD simulations can be applied only to a limited set of materials, and are often not reliable outside the initial scope of parameterization. This is especially problematic in situations when there is significant interatomic charge transfer between atoms or covalent bond formation. In fact, for many important problems a meaningful parameterization may not even be possible [2,3]. To investigate properties such as photo-excitation, electronic charge distributions, quan-tum-size effects in nanosystems, magnetic phase transitions, Peierls distortions of a crystal lat-tice, electron correlation and 5f localization in actinides, or quantum response properties such as the polarizability and the electrical conductivity, we require a quantum mechanical description of the electrons [1-3]. A quantum based MD therefore offers not only a more accurate and trustworthy replacement of classical MD, which is critically needed, but it also provides an es-sential path to modeling systems and phenomena that are clearly beyond the reach of existing, classical methods. Unfortunately, QMD simulations remain intractable for most systems be-cause of their computational complexity.
The primary objective of the proposed workshop is to interconnect different communities in-volved in quantum-based molecular dynamics modelling working on large scale atomistic simu-lations of materials and devices, regardless the specific scientific subarea of interest. The work-shop will focus on computational and theoretical challenges and techniques, to encourage a strong multidisciplinary approach. We aim for encouraging dialogue between ab-initio, semi-empirical, empirical and classical methods communities. In order to foster such an exchange internationally leading groups in multi-scale physics, chemistry and materials science approach-es will be invited. Given the strongly interdisciplinary scope of the workshop, only theoretical modelling groups will be invited in order to ensure a sharp focus on theoretical method devel-opments for next generation high performance quantum-based MD models for bridging scales in materials/systems sizes and time evolution. About 35 worldwide leading experts in the fields of
● Fast quantum based MD methods (Stable, energy conserving algorithms that reduce the computational overhead of SCF optimization that efficiently can be combined with O(N) schemes and run on large-scale computers.)
● Order-N electronic structure algorithms (Methods that have a computational complexity that scales only linearly with the system size instead of cubically as in traditional methods.)
● Approximate DFT-methods (Self-consistent charge density functional based tight-binding methods and semi-empirical schemes that reduces the computational overhead in direct DFT methods.)
● Quantum-derived many-body potentials (Bond-order, Learn-on-the-fly and Neural Network potentials)
● QM/MM hybrid approaches (Multi-scale approaches where the effect and response of the extended surrounding is modeled with more approximate but computationally faster classical force fields.)
● Accelerated MD for bridging time scales (Techniques to extend the effective time scale of MD simulations as well as methods to boost MD driven sampling of the phase space.)
● Advanced integration schemes for MD (Multiple time step algorithms, parallel integrators)
● Challenges in Computational Science / hybrid architectures (Formulation and design of algo-rithms for large-scale QMD simulations that take full advantage of advanced massively paral-lel hybrid architectures.)
will be invited, and give keynote talks on the related areas to set the stage and define the targets for future developments.
We define the following sub-objectives:
1. To share the state of the art at a computational level, regarding performance, scalability, par-allelization and hence range of applicability of different quantum based MD techniques.
2. To define common targets for method developments between different communities, focus-ing on hierarchical and domain multi-scale coupling techniques, interconnection between electronic structure methods, and classical methods.
3. To share the perspectives for the next years of research in the field, in particular referring to novel emerging issues such as time-dependent and non-equilibrium phenomena in complex systems.
The intended format of the workshop is seven half-days, 40 minutes talks including ample time for discussion after each talk. To ensure full and active participation of also the young emerging researchers attending the meeting a poster session will be organized.
Thomas Frauenheim (University of Bremen) - Organiser
Juerg Hutter (University of Zurich) - Organiser
Benjamin Hourahine (University of Strathclyde) - Organiser
Anders M. N. Niklasson (Los Alamos National Laboratory) - Organiser & speaker