Fortran comparison status

Current status

The current fixed fixture comparison shows exact agreement between Python and Fortran for:

  • uk

  • vk

  • wk

  • pp

  • cu

  • cv

  • cw

  • u_after

  • v_after

  • w_after

The seeded sampling comparison also shows exact agreement for:

  • L3

  • L

  • M

  • final RNG state

The deterministic advancement and initialization comparison shows exact or near-machine-precision agreement for:

  • BAdv

  • BEqnStep

  • BInitRun

  • BInitIter

  • BExp RNG state and waiting-time behavior

The reduced multi-trial scheduling comparison shows agreement for:

  • BLowerdt

  • multi-trial dt / td adaptation

  • nstat / ntseg sub-interval scheduling

  • centerline series accumulation

  • final state arrays and final RNG state

The repeated-realization statistics/output comparison shows agreement for:

  • niter-level repeated scheduled realizations

  • BStats-style cstat accumulation

  • BSeries aggregation across realizations

  • BWriteSeries time/variance output

  • final RNG state after multiple realizations

The postprocessing/change-statistics comparison shows agreement for:

  • BSetOld

  • BChange

  • direct BRecord

  • direct XRecord

  • original BSnap xmgrace-style output products on multiple controlled fixtures, including a second istat case

  • patched-legacy BSnap intercomparison products on a controlled fixture

  • dedicated patched-legacy BSnap intercomparison comparison via scripts/compare_bsnap_intercomparison.py

The smaller helper/config compatibility comparison shows agreement for:

  • BReadOptions

  • BReadPars

  • BReadConfig

  • BInitStats

  • BAddTerm

  • BrngGet / BrngPut

Scope caveat

This still does not yet imply full equivalence with the original odt1 solver. However, the validated scope now includes:

  • the minimal sampled/accepted eddy path,

  • deterministic advancement and initialization,

  • a reduced but faithful multi-trial realization schedule.

BRecord compatibility has now been investigated: the crash was caused by passing a literal argument to a routine that mutates N via N=abs(N). Under Fortran pass-by-reference semantics, writing through a constant argument can segfault. A minimal reproducer is provided in scripts/investigate_brecord.py. When called correctly with an integer variable, BRecord runs and matches the Python implementation.

A separate caveat remains for the original unmodified BSnap intercomparison mode (ioptions(1)=0): a minimal reproducer in scripts/investigate_bsnap_intercomparison.py shows that this path segfaults under the local toolchain, while the patched legacy intercomparison path and the original xmgrace mode (ioptions(1)=1) run and match the Python implementation. The failure is tied to the negative-N header-writing convention around BRecord in the original intercomparison path. The external legacy sources in this repository have therefore been minimally patched to preserve the intended output semantics while avoiding the in-place negative-N mutation.

At the program-orchestration level, pyodt1.legacy.run_legacy_case() now reproduces the usual legacy output bundle more closely by writing T.dat, the selected mode’s A*–I* files (including empty placeholders for unused legacy file slots), and a fort.11 progress log. This narrows the remaining gap to exact line-for-line Bodt.f procedural parity rather than missing output products.

For a precise record of the vendored Fortran source changes, see Legacy source patches.