Skip to content

SOURCE_AUDIT.md

Latest commit

 

History

History
92 lines (75 loc) · 5.81 KB

SOURCE_AUDIT.md

File metadata and controls

92 lines (75 loc) · 5.81 KB
Raw

Source audit, 2026-10-08

The user approved the first implementation slice and a new public Denolle-Lab/gesc-python repository. Handoff recommendations were not treated as proof of capability. Static audits below are distinct from numerical tests of this new package. No external backend was built or timed.

Source Pinned revision Reuse decision
mdenolle/gesc 92bf1655e5e52a9b204fbb274b1a0853a6cdac3f Mathematical/history reference; no source vendored
Denolle-Lab/PyGesc 5acbfa2550af6c98c6cc71cbc815c1d21d96b236 Unfinished translation; no validated backend reused
Denolle-Lab/surface_wave_sensitivity ec500e58650abf30a9f86b4ff1b265d7154ab88a disba-based example/model ideas, not a monitoring inverse
nqdu/SpecSWD e5dc8ce6138c29384a12ec180c74eace6a8773c9 Production candidate after independent validation/license check
keurfonluu/disba 1a9e26d49dbaadead84cdf427d6f51b3f5c3707b Optional isotropic diagnostic; CPS-derived algorithm

Paper

Denolle et al. (2012), DOI 10.1785/0120110183, was read from the supplied PDF. SHA256: 1d7aebf9b2d34da54ddd056b7bddd8c1642fcc5d8f0ab5a0734f322d77bf07ab. It matches the audited GESC repository's paper copy. Printed pp. 1215-1217 describe singular A u=k B u, modified stresses, interfaces, locked bottom, and MATLAB/LAPACK QZ. Pp. 1218-1221 give analytical/layered/CPS tests and far-field waveforms. P. 1223 explicitly excludes implemented anelasticity. These examples do not establish Scholte/general-anisotropy support or a measured CPS speedup. The paper is not redistributed with this software.

MATLAB GESC

Paths below are relative to gesc/sub at the pinned revision.

  • EigRW.m:36,80 and EigLW.m:21,56: full eigendecomposition before requested-mode retention.
  • EigRW.m:8-11, EigLW.m:11-14: nonfinite matrix entries replaced by zero.
  • EigRW.m:102-106, EigLW.m:69-71: normalization by a surface component.
  • gesc.m:186-193: skips failures and couples family count through a minimum.
  • gesc.m:119-127,181: fundamental-only adaptation and point-count early return.
  • gesc.m:143-145: half-space coefficients taken from profile maxima.
  • RayInnerMat.m:10, LoveInnerMat.m:16: inverse mu; no valid zero-shear acoustic block.
  • RayBC.m:53-57: water branch makes no active change; water edits elsewhere are commented.
  • get_integrals_sw.m:34-47: dry energy integrals and phase/group quantities exist.
  • get_sw_gf.m:132: multimode Rayleigh entry uses Love source eigenfunction uyi.
  • get_sw_gf.m:186, get_sw_dc_farfield.m:244: multimode inverse FFT selects one frequency.
  • get_sw_gf_farfield.m:89-96: structure variable C overwritten; S2 is undefined.
  • Source/moment/modal Green routines exist beyond the paper's far-field tests; their existence is not validation of all branches, amplitudes, or geometries.

No explicit license file or automated numerical test suite was found in that tree. The new backend rederives the equations and does not copy these routines.

Existing lab Python work

PyGesc has an explicit MIT license. gesc.py:395-408 calls missing EigRW, EigLW, LoveInnerMat, LoveBC, and get_integrals_sw routines. The driver also retains MATLAB identifiers, dict/attribute mismatches, and undefined bigC construction. gesc.py:430,435-437 uses elementwise multiplication where the MATLAB expression requires matrix products. It cannot serve as a validated end-to-end backend.

The sensitivity repository has MIT licensing and two SensKernels notebooks. They invoke disba PhaseDispersion, EigenFunction, and PhaseSensitivity for isotropic layered examples. Useful examples exist, but no estimator-aware probabilistic dv/v-to-depth API was found in those inspected notebooks. Geospatial datasets and notebook graphics are not copied into this repository.

SpecSWD and disba

SpecSWD includes fluid/coupling assembly (src/vti/assemble.cpp), fluid and solid displacement conversion (src/vti/transform.cpp), anisotropic assembly, and phase/group kernel code. Tests include analytical dry Rayleigh roots and finite-difference acoustic/elastic kernel comparisons in tests/unit_tests/rayl_op.cpp. Tests were inspected, not run. src/shared/schur.cpp:113-127 uses Euclidean eigenvector normalization; physical excitation residues need separate validation. No complete physical waveform-synthesis API was identified in the inspected Python package.

Its build requires C++17, Fortran, Eigen, LAPACK/LAPACKE, HDF5, and pybind11. Python metadata requires NumPy/SciPy/h5py/Matplotlib. Metadata declares MIT, but no explicit license file was found and separately attributed code is present. Adoption requires upstream clarification and file-level provenance checks. GoogleTest is fetched from unpinned main by CMake; reproducible evaluation must pin it.

disba has an explicit BSD-3-Clause license; dispersion, eigenfunction, ellipticity, and sensitivity APIs and water-layer tests exist. Its runtime includes NumPy, Numba, Matplotlib, and importlib_metadata. It is not an anisotropic waveform solver.

CPS

The official site lists the April 25, 2026 distribution and an explicit MIT license. The program inventory documents isotropic/VTI modal synthesis and generalized-reflection eigenfunctions. Ocean examples demonstrate water-layer dispersion. The source archive has not been audited/built here, and pressure/source normalization parity remains a benchmark gate.