Condensation Parity Walkthrough Ownership Record
This independent condensation walkthrough record is the source of truth for routing deferred capabilities from the bounded direct-condensation walkthrough.
Supported evidence boundary
Section titled “Supported evidence boundary”The walkthrough source compares an independent fp64, fixed-four-substep NumPy oracle with the low-level direct GPU kernel. Its walkthrough regression test keeps that bounded comparison executable.
Physics, conservation, and energy are separately evaluated evidence. Warp CPU
is the supported baseline when installed; CUDA is optional, additive evidence.
This record does not establish high-level CPU strategy parity, Runnable
parity, or general CPU workflow parity. The
fixed-four direct-kernel evidence boundary
states the related production-hook evidence limit.
Scope boundaries and current owners
Section titled “Scope boundaries and current owners”This walkthrough establishes direct fixed-four-substep kernel evidence only. It does not qualify the following independent paths:
| Capability | Canonical owner and evidence requirement |
|---|---|
thermal_work consumption, temperature feedback, adaptive stepping |
Future approved numerical-method work with its own numerical/validation contract |
| Backend selection, native/runnable integration, resident loops | Execution; use its explicit supported paths and separate tests, without inferring general CPU-strategy parity |
| Graph capture and host-validation separation | Graph-capture runbook; capture qualification requires separate evidence |
| Performance and memory budget | Resident benchmarks and qualification gates; no reviewed measurement is supplied by the walkthrough |
| Broad state/multi-step autodiff | Pending initial-state optimization; distinct from the bounded raw-rate interior probe |
| Phase-aware surface tension and BAT activity in the direct kernel | Future approved physics expansion and validation contract |
Focused reproduction commands
Section titled “Focused reproduction commands”python docs/examples/gpu/direct/gpu_condensation_parity_walkthrough.pypytest examples_tests/gpu_condensation_parity_walkthrough_test.py -q -Werrorpytest particula/gpu/kernels/tests/condensation_test.py -q -Werrorpytest particula/gpu/kernels/tests/condensation_stiffness_test.py -q -Werrorpytest particula/gpu/kernels/tests/condensation_graph_capture_test.py -q -Werrorpytest particula/gpu/kernels/tests/condensation_autodiff_test.py -q -WerrorExplore this topic: Scientific and numerical correctness