Workshop: 10th International Workshop on Software Correctness for HPC Applications (Correctness '26)
Authors: Youngjun Lee (Argonne National Laboratory (ANL)), Anshu Dubey (RIKEN Center for Computational Science), and Jan Hückelheim (Argonne National Laboratory (ANL))
Abstract: Verifying a numerical algorithm in a large scientific simulation is challenging: the framework is too big to model-check, and the unit tests exercise only sampled inputs. We report a case study developing a new cloud-in-cell (CIC) deposition algorithm for Flash-X, a multiphysics simulation framework, keeping the CIVL model checker in the loop. Rather than verify the algorithm within Flash-X, we extract only the interfaces the algorithm needs into a small, self-contained C model, and check the algorithm there. Treating particle positions as symbolic values, CIVL proves two physical properties---mass conservation and the deposition location---over the continuum of admissible positions, and verifies memory safety and freedom from MPI deadlocks and data races across rank distributions within bounds. Writing the verifying properties first and rechecking them at each stage turned CIVL into a design guardrail. It also surfaced a defect that our random-seed tests missed. We report its benefits, cost, and tradeoffs.
Back to 10th International Workshop on Software Correctness for HPC Applications (Correctness '26) Archive Listing
Back to Full Workshop Archive Listing