Case overview
NeuralPOM is focused on one scientific case: the wind-driven Double Gyre. It is the data-generation, training, inference, and long-rollout benchmark for the hybrid AI--physics mainline.
Registry
The central registry in src/neuralpom/cases/catalog.py exposes only:
| Case | problem_id | Grid | Supported cores |
|---|---|---|---|
double_gyre | 4 | 90 x 82 x 16 | implicit, explicit |
python
from neuralpom.cases import get_case_settings
from neuralpom.core_factory import resolve_core_class
settings = get_case_settings("double_gyre")
core_cls = resolve_core_class("implicit")
core = core_cls.from_settings(settings)
core.run_initialization()The default training and long-rollout workflows use the implicit core. The explicit implementation is retained as a direct POM-style reference and as a CUDA fast-path consistency target.
See Double Gyre for the physical and workflow details.