From methane and iron nanoparticles to “Turquoise” H2 and carbon nanotubes: simulating the whole catalytic process inside a reactor
- Name: From methane and iron nanoparticles to “Turquoise” H2 and carbon nanotubes: simulating the whole catalytic process inside a reactor
- EuroHPC machine used: Leonardo
- Topic: Physical sciences, Materials Engineering
Overview of the project
This project focuses on the atomistic simulation of turquoise hydrogen production, a process in which hydrogen is obtained from methane decomposition while carbon is captured and converted into solid carbon nanotubes instead of being released as greenhouse gases. Because the underlying mechanisms involve bond breaking and bond formation, they require a quantum-mechanical description. At the same time, the large size of the systems makes direct large-scale quantum simulations impractical. For this reason, the project relies on machine-learning force fields, which can achieve near ab initio accuracy at the computational cost of classical molecular dynamics.
A key step in developing these force fields is the generation of large amounts of accurate ab initio reference data. To support this, the group has developed Tribchem, a workflow generation and management tool for automated high-throughput DFT calculations on solid surfaces and interfaces. While DFT calculations, typically performed with GPU-optimized codes, have traditionally been the most computationally demanding part of these workflows, the increasing system size has made Python-based pre- and post-processing routines a major bottleneck. The project therefore also aims to optimize these functions so that the available HPC resources can be used efficiently throughout the entire workflow.
How did EPICURE support the project and what were the benefits of the support?
The project requested support from EPICURE to address performance bottlenecks in TribChem, a Python-based high-throughput platform developed in their research group to study solid interfaces at the DFT level. Although the DFT calculations themselves run efficiently on HPC systems such as Leonardo, increasing the system size and treating more disordered or lower-symmetry systems caused significant slowdowns in two specific Python routines. In particular, support was requested for improving the CPU parallelization of these functions on the nodes reserved on Leonardo.
“EPICURE helped us by quickly identifying the main critical issues affecting performance and by suggesting concrete improvements to make better use of the available HPC resources. Their support led to measurable gains in code efficiency and helped us better understand how to optimize our workflow for large-scale calculations. Thanks to the support of the EPICURE team, we achieved substantial performance improvements. They successfully identified the bottlenecks and obtained 20x and 6x speed-ups on the two most critical Python routines. This optimization was crucial for our research, as it had previously been nearly impossible to study large or complex interfaces, which typically require large supercells and lack high symmetry.” – Prof. M. Clelia Righi