Large Language Model Successfully Ports Complex Ocean Simulation Model to Modern Computing Languages
Researchers used an AI-assisted coding tool to translate FESOM2, a 74,000-line ocean-sea-ice simulation model, from Fortran to C and then to C++/Kokkos for GPU acceleration. The translation preserved the model's physics accuracy while achieving 1.6-3.7x speedup on GPU hardware compared to traditional CPU processing. This demonstrates that LLMs can reliably port production-grade scientific software while maintaining computational accuracy, a capability previously unproven at this scale.
A team of researchers successfully used a large language model (LLM) coding assistant, guided by domain experts, to port FESOM2—a production-grade unstructured mesh ocean and sea-ice model containing approximately 74,000 lines of Fortran code—into C and subsequently into C++/Kokkos for performance portability across CPUs and GPUs. The two-stage translation approach separated the reproduction of numerical accuracy from the introduction of parallelism, with strict validation at each step. The C port reproduced the original Fortran model's long-term simulation statistics over five-year runs, while the Kokkos port achieved bit-for-bit identity on CPUs and statistical equivalence on GPUs over multi-year simulations. On large eddy-rich meshes with up to 7.4 million surface vertices, a single NVIDIA A100 GPU node achieved 1.6-3.7x faster performance than a CPU node, reaching the 1-2 simulated-years-per-day throughput required for production use. The researchers identified three critical practices: staged translation separating numerics from parallelism, strictly literal translation without unauthorized improvements, and validation against stage-appropriate acceptance criteria.
What's missing
The study does not discuss computational costs (e.g., energy consumption or carbon footprint) of GPU versus CPU execution, nor does it address the generalizability of this approach to other scientific software domains beyond ocean modeling. The paper also does not provide detailed comparison with alternative code modernization approaches or discuss potential limitations of LLM-assisted translation for other types of scientific codes with different numerical properties.
What different sources said
- arXiv physicsCenter
An Ocean Model Ported by a Large Language Model: Experience and Lessons from FESOM2 (Fortran to C to C++/Kokkos)
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.