Multiscale Molecular Dynamics with MiMiC: Optimizing the Performance on Modern Supercomputers
Location: CECAM-DE-JUELICH
Organisers
With the emergence of exascale supercomputers comes the potential for new scientific breakthroughs in the simulation-driven fields, not least in the biomolecular domain. To efficiently exploit the vast resources offered by the complex hybrid CPU/GPU technologies present in modern supercomputers, requires careful attention to the setup and execution of the simulations. This is especially true for multiscale simulations that employ multiple programs with different performances and parallelization requirements.
This school will train the participants on the efficient use of modern high-performance computing (HPC) facilities to solve problems in the field of computational biochemistry and biophysics using multiscale methods to model complex (bio)molecular systems, ranging from molecules in solution to membrane-embedded proteins. The large size of such systems and the long time scale of relevant phenomena necessitates the use of a combination of complementary techniques that allow spanning multiple temporal and spatial scales as well as the use of large-scale HPC facilities.
The participants will be taught advanced molecular dynamics (MD) simulation techniques for coupling multiple spatial resolutions using quantum mechanics/molecular mechanics (QM/MM) models [1] and multiple time scales through multiple time step (MTS) algorithms [2, 3], as well as the use of enhanced sampling methods [4, 5].
Theoretical lectures will be complemented by practical sessions where students will use the high-performance multiscale modeling framework MiMiC [6, 7] to run multiscale QM/MM MD simulations employing a variety of coupled external programs. MiMiC currently supports CPMD, CP2K, and Quantum ESPRESSO as QM clients and GROMACS and OpenMM as MM clients for electrostatic-embedding QM/MM MD simulations. The PLUMED library will be employed for enhanced-sampling methods.
Throughout the school, a substantial emphasis will be placed on the crucial role of HPC in enabling efficient and accurate multiscale QM/MM simulations of large and complex biomolecular systems. This focus is closely connected with the MiMiC framework, which is designed to enable fast and highly parallelized multiscale MD simulations allowing users to exploit modern heterogeneous supercomputing architectures.
The school will be held at Forschungszentrum Jülich (FZJ). The HPC aspect will be extensively covered in theoretical lectures, including lecturers from the Jülich Supercomputing Center, where the first European exascale supercomputer JUPITER is being installed, and the Italian supercomputing centre CINECA, host of the pre-exascale supercomputer LEONARDO. Practical sessions will provide participants with valuable hands-on training in the efficient utilization of HPC resources, including how to optimize single- and multi-node efficiency of CPU-only and hybrid CPU/GPU nodes. Hence, as a result of attending this school, participants will not only gain an understanding of valuable aspects of multiscale MD simulations but also achieve proficiency in the efficient use of state-of-the-art computational resources.
The school covers accommodation in shared double rooms for five nights (Sunday–Friday) as well as breakfast and lunch.
References
Jógvan Magnus Haugaard Olsen (Technical University of Denmark) - Organiser & speaker
Germany
Davide Mandelli (FZJ) - Organiser & speaker
Italy
Simone Meloni (University of Ferrara) - Organiser & speaker
Switzerland
Ursula Roethlisberger (EPFL) - Organiser & speaker