Abstract
Coarse-Mesh Rebalance (CMR) is an effective and low-memory accelerator for 3D whole-core neutronics calculations and other neutron transport or diffusion calculations. But usually, it is applied on a fixed schedule: (1) it is triggered every fixed number of accelerated sweeps. (2) Its coarse-mesh problem is solved to a fixed convergence tolerance. Both of these choices are made problem-independently, which limits the performance of the acceleration.Thus, a dynamically scheduled CMR (DS-CMR) is proposed in this paper to adapt the two control knobs at almost no extra cost. (1) an inexact coarse-mesh solve with convergence tolerance tied to the current fine-mesh residual, and (2) a smoothing-stall adaptive trigger driven by the measured convergence rate. The guardrail on the trigger guarantees worst-case performance no worse than the fixed schedule.Sensitivity analysis was conducted on the 2D problem. Numerical results indicate that both control knobs can effectively accelerate exact CMR. Compared to exact CMR, DS-CMR achieves a further 2x speedup. Furthermore, applying the parameters derived from the 2D problem to the 3D problem brings about a 1.57x speedup.In summary, the proposed DS-CMR effectively improves the speedup of exact CMR, thereby further enhancing the computational efficiency of pin-by-pin calculations.
| Original language | English |
|---|---|
| Article number | 112980 |
| Journal | Annals of Nuclear Energy |
| Volume | 242 |
| DOIs | |
| State | Published - Feb 2027 |
Keywords
- Acceleration
- Coarse-mesh Rebalance
- Dynamically Scheduled
- Pin-by-pin
Fingerprint
Dive into the research topics of 'A dynamically scheduled coarse-mesh rebalance acceleration for 3D whole-core pin-by-pin neutronics calculations'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver