Skip to content

fix: static equilibria over non-planar bathymetry - #16

Merged
SMI-Lab-Inha merged 5 commits into
devfrom
fix/statics-bathymetry
Oct 15, 2026
Merged

SMI-Lab-Inha merged 5 commits into
devfrom
fix/statics-bathymetry

Conversation

@SMI-Lab-Inha

Copy link
Copy Markdown
Owner

Summary

Static equilibria over real (non-planar) bathymetry. A statics robustness campaign (352 decks plus a 384-run rough-seabed sweep) found two failures that every survey bathymetry can trigger.

1. Finite-EI lines over a non-planar bathymetry failed, even with 2 cm roughness (39 of 192 sweep runs; mostly on a 25 m survey grid).

  • Cause: the catenary seed is laid on the plane fitted along the chord, and was rejected as soon as a node sat about 1e-6 (L + |z|) below the real bed. The anchor walk was not allowed on a non-planar bed, and the mesh-sequenced fallback ended on a kinked branch.
  • Fix:
    • Seed nodes deeper below the real bed than their static penalty penetration are draped onto it; the bending-stiffness continuation settles the contact.
    • The anchor walk also runs on a non-planar bed, with the walked anchor following the bed at its own height above it. A failed walk there no longer marks the deck inadmissible, so the mesh-sequenced route still runs, and its message no longer claims the line is too long.
    • Planar beds keep the old checks exactly. The final state is audited, stability-tested and kink-checked as before.

2. EI = 0 lines resting in a crease of the bilinear bed failed (a trench bottom, a valley of a 2D grid).

  • Cause: the contact normal jumps across a grid line, so a node in a concave crease is always pushed back across it and the discrete problem has no equilibrium. Trench depths 20–26 m and 30 m failed; 3 of 4 lines on a ±0.5 m 2D random grid failed.
  • Fix: each interior grid line is rounded within a band of 5 % of the smaller adjacent spacing.
    • Fillet: the C2 fillet J (δ − |u|)³ / (6 δ²) of the slope jump J, as a tensor product, so height, gradient and curvature are continuous and valleys stay convex.
    • Size of the change: the surface moves by at most |J| δ / 6 along one line (1.7 cm for a 0.2 slope jump on a 10 m grid). At a crossing the two roundings add.
    • What stays exact: exactly planar grids and patches without a crease stay bilinear bit for bit. The rounding errors of a plane written to 2–3 decimals are tiny creases, rounded by a fraction of a millimetre.
    • Contact force in a band: the gradient of the penalty potential, with its Hessian as tangent (up to the bilinear twist, as before). The same surface is used in statics and dynamics, friction, outputs and the Python Bathymetry.
    • Fixed points: a Fixed point at the tabulated depth of a creased node is placed on the rounded surface.

Results

Set Before After
Rough sweep, finite-EI 153/192 192/192
Rough sweep, EI = 0 191/192 192/192
Trench depth 2–30 m (EI = 0) 5/11 11/11
Bathymetry families (B + finite-EI on bathymetry) 51/75 65/75 (the EI = 0 ones left have anchors below the bed)
Long lines walked into place on rough beds 0/5 5/5 (FairTen within 0.02 % of the flat-bed walk)
  • Correctness:
    • Newly converged decks agree with their 4× mesh twins within 1 % (FairTen).
    • In a trench the slack pile gives AnchTen 73.15 kN against w·Δz = 73.24 kN.
    • Independent nodal force balance on the rounded surface: residual ≤ 1e-4 of the nodal weight.
  • Changed results:
    • Rough-bed decks that solved before change by at most 0.04 % in FairTen (one deck settling in another valley: 0.40 %) and up to 8.5 % in AnchTen.
    • This comes from removing the spurious horizontal reactions of one-sided normals. The deck-to-4×-mesh AnchTen spread on 2D-rough beds roughly halves.
  • Unchanged: all 52 examples and the driver test decks are byte-identical (none uses a creased bathymetry).
  • Dynamics: surge/sway across grid lines runs on both line families with no new chatter. On dev the 2D-rough and trench decks could not initialise.

Tests

  • cable_statics_matrix: ROUGH_1001–1003 with mesh twins, all failing on the base.
  • test_bathymetry.f90:
    • C2 continuity across lines, crossings and band edges;
    • z_xx, z_yy and z_xy against finite differences;
    • exact bilinear outside the bands, bounded rounding inside;
    • planar grids bit-identical;
    • the contact tangent against finite differences.
  • test_statics_robustness.py: an independent re-implementation of the rounded surface with a nodal balance, plus the trench and 2D-rough decks, all failing on the base.
  • python/tests/test_clearance.py: fillet values, continuity, planar exactness.
  • Full suite, Windows, gfortran, Release, slow tier included: 257/257. Python: 1926 passed. pre-commit run --all-files passes.

Checklist

  • pre-commit run --all-files passes
  • ctest --test-dir build --output-on-failure passes
  • python -m pytest tests python/tests passes (if Python or validation records changed)
  • VALIDATION.md updated if a validation result changed — no validation record uses a creased bathymetry
  • CHANGELOG.md updated under [Unreleased] (heading added if missing) for user-visible changes
  • Documentation in doc/ updated for changed inputs, options, or outputs (theory, file formats, troubleshooting, Python post-processing)
  • Conventional Commit title (feat:, fix:, docs:, ...)
  • Commits are signed off (git commit -s) under the
    Developer Certificate of Origin

A finite-EI line over any structured bathymetry that is not a plane
along its chord failed its static solve, even with 2 cm roughness: 39 of
192 runs of a rough-seabed sweep failed, most of them on a 25 m survey
grid. The catenary seed is laid on the plane fitted along the chord and
was rejected as soon as a node sat more than about 1e-6 (L + |z|) below
the real bed; on a non-planar bed the anchor walk was not allowed, and
the mesh-sequenced fallback ended on a kinked branch.

Over a non-planar bathymetry the seed nodes deeper below the real bed
than their static penalty penetration (nodal weight over kBot Diam
length) are now draped onto the bed instead of rejecting the seed, and
the bending-stiffness continuation, which queries the full surface,
settles the contact. The anchor walk also runs there, with the walked
anchor kept at its height above the bed under it; its failure no longer
marks the deck inadmissible, so the mesh-sequenced route still runs.
Planar beds keep the old checks exactly, and the final equilibrium is
audited, stability-tested and kink-checked as before.

Rough-seabed sweep, finite-EI: 153/192 -> 192/192; long lines walked
into place on rough beds: 0/5 -> 5/5 (FairTen within 0.02 % of the
flat-bed walk); converged twins on a 4x mesh agree within 1 %. Decks
that already solved on a rough bed and now take the primary route
change by at most 0.03 % in FairTen and 2.2 % in AnchTen (0.5 m
roughness, where the deck and 4x meshes differ by up to 13 %). All
examples and driver test decks are byte-identical. New statics-matrix
decks ROUGH_1001-1003 with mesh twins fail on the base and pass.

EI = 0 lines resting in a crease of the bilinear bed (a trench bottom,
a valley of a 2D grid) still fail: the contact normal jumps across a
grid line, so the discrete problem has no equilibrium there. The
troubleshooting page documents it.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The bilinear bathymetry surface has a slope jump across every interior
grid line. A node resting in a concave crease (a trench bottom, a valley
of a 2D survey grid) took one cell's normal, so its contact force always
pointed back across the crease and the discrete problem had no
equilibrium there: EI = 0 lines whose slack length piles in a trench
failed their static solve for trench depths of 20-26 and 30 m, and 3 of
4 lines on a +-0.5 m 2D random grid failed.

Each interior grid line is now rounded within a band of 5 % of the
smaller adjacent spacing by the C2 fillet h(u) = (delta - |u|)^3 /
(6 delta^2) times the slope jump, as a tensor product across both
directions, so height, gradient and curvature are continuous; valleys
stay convex. The surface moves by at most |J| delta / 6 (1.7 cm for a
0.2 slope jump on a 10 m grid). A crease that would move the surface by
less than 1 um is not rounded, so planar grids stay bilinear bit for
bit. Inside a band the contact force is the gradient of the penalty
potential with its consistent tangent, in the statics and the dynamics
of both line families, the friction anchors, end forces and outputs;
Python's cabledyn.Bathymetry mirrors the surface to 1.4e-14. A Fixed
anchor at the tabulated depth of a creased grid node is placed on the
rounded surface.

Trench-depth sweep 5/11 -> 11/11, EI = 0 rough sweep 191/192 ->
192/192, bathymetry families 52/75 -> 65/75. In a trench the slack pile
gives AnchTen 73.15 kN against w dz = 73.24 kN. Previously converged
rough-bed decks change by at most 0.04 % in FairTen (0.40 % for one
deck settling in another valley) and up to 8.5 % in AnchTen, from
removing the spurious horizontal reaction of one-sided normals; the
deck-to-4x-mesh AnchTen spread on 2D rough beds roughly halves. All
examples and driver test decks are byte-identical. A rejected anchor
walk over a non-planar bed no longer claims the line is too long for
its span.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
- The anchor walk over a non-planar bathymetry moved the anchor on the
  straight line between its start and the deck anchor, which a ridge or
  dip between them could put below or far above the bed. The walked
  anchor now keeps its height above the bed under it at every step, as
  documented, and the predictor starts from the current anchor.
- crease_blend read an uninitialised flag when the second row of an
  x band had no crease.
- The docs and comments now state the rounding bound per line (the
  roundings of two lines add at a crossing), that only an exactly planar
  grid stays bilinear bit for bit, that the friction capacity uses the
  normal force without the in-band curvature term, and that the energy
  agrees with the force up to the bilinear twist; the walk failure no
  longer asserts the line is too long.
- The C2 test checks z_xy against finite differences.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 15, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-15T19:37:20.719323Z 327bf15 Manual request
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@read-the-docs-community

read-the-docs-community Bot commented Oct 15, 2026 •

Copy link
Copy Markdown

Documentation build overview

📚 CableDyn | 🛠️ Build #35175753 | 📁 Comparing 327bf15 against latest (f27c5e3)

  🔍 Preview build  

16 files changed · ± 16 modified

± Modified

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b802b9f4f2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread src/CableDyn_DeckDriver.f90 Resolved
A Fixed point in a crease was accepted anywhere between the rounded
seabed and the bilinear grid surface and moved onto the rounded one, so
an anchor well below the actual seabed but above the grid surface was
silently lifted. It is now accepted only within the seabed tolerance of
either surface: at the tabulated depth it rests on the seabed and is
placed on the rounded surface; between the two but near neither it is
judged by its distance to the rounded seabed alone. New driver tests on
a V valley: an anchor at the tabulated depth initialises, one 1 cm below
the rounded floor is rejected.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
@SMI-Lab-Inha

Copy link
Copy Markdown
Owner Author

@codex review

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c5c71c219c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread src/CableDyn_Static.f90 Resolved
Inside a crease band the EI = 0 static residual and its friction
anchors took the friction capacity from the norm of the band contact
force, which includes the curvature term of the penalty potential,
while the Hermite statics and the dynamics use the scalar normal law at
the normal penetration. Both EI = 0 sites now use the scalar law in a
band, so the capacity is the same across routes and at the dynamic
hand-off; outside the bands the computation is unchanged.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
@SMI-Lab-Inha

Copy link
Copy Markdown
Owner Author

@codex review

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. 🚀

Reviewed commit: 327bf15c5e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@SMI-Lab-Inha
SMI-Lab-Inha merged commit 1446317 into dev Oct 15, 2026
15 checks passed
@SMI-Lab-Inha
SMI-Lab-Inha deleted the fix/statics-bathymetry branch October 15, 2026 20:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant