Increasing Computational Efficiency of CFD Simulations of Reactive Flows at Catalyst Surfaces through Dynamic Load...

Area of Science:

• Computational fluid dynamics (CFD)
• Chemical engineering
• High-performance computing

Background:

• Multiscale simulations of reactive flows at catalyst surfaces are computationally intensive.
• Existing methods face challenges in efficiently distributing computational load, particularly for chemistry solutions.
• Optimizing parallel efficiency is crucial for advancing these simulations.

Purpose of the Study:

• To introduce a numerical strategy using dynamic load balancing (DLB) to enhance computational efficiency in multiscale CFD simulations of reactive flows.
• To improve the distribution of computational workload across processors for chemistry calculations.
• To enable more extensive and efficient use of high-performance computing resources.

Main Methods:

• Implementation of dynamic load balancing (DLB) combined with a hybrid parallelization technique (MPI and OpenMP).
• Application of the strategy to fixed and fluidized bed reactor simulations.
• Assessment of parallel efficiency and computational speed-up compared to simulations without DLB.

Main Results:

• Demonstrated significant improvements in parallel efficiency, from 19% to 87% (fixed bed) and 19% to 91% (fluidized bed).
• Achieved computational speed-ups of 1.9x (fixed bed) and 2.1x (fluidized bed) compared to non-DLB simulations.
• Successfully minimized computational overheads through optimized load distribution.

Conclusions:

• The proposed DLB strategy effectively enhances the computational efficiency of multiscale CFD simulations for reactive flows.
• This approach facilitates more efficient utilization of high-performance computing resources.
• The method expands the scope and feasibility of complex reactive flow simulations.