Skip to content

release: CableDyn 0.4.3 - #189

Open
SMI-Lab-Inha wants to merge 47 commits into
mainfrom
release/0.4.3
Open

SMI-Lab-Inha wants to merge 47 commits into
mainfrom
release/0.4.3

Conversation

@SMI-Lab-Inha

Copy link
Copy Markdown
Owner

CableDyn 0.4.3

This PR merges dev with #144 to #188 into main, together with the version bump and the release records. #144 to #188 cover VIV parity, finite-EI damping, the coupled speed-ups, CDOutDt and the validation cases L3-C, D, W, Wh, N, F, O and K.

Commits of the release branch

  • release: CableDyn 0.4.3 sets the version and date (2026-11-20) in:

    • the build, the Fortran banner and the OpenFAST module;
    • the C API and its header (patch 3; ABI 1.4 unchanged);
    • the Python package and CITATION.cff.

    It also dates the CHANGELOG section and writes the release note on the documentation front page. The README and VALIDATION.md now name the 0.4.2 release driver as the driver the L5 wall-clock records were measured with (see below).

  • chore(release): rerun the experiment records with 0.4.3: Holcombe and Bergdahl reproduce their summaries, metrics, refinement figures and the hashed Holcombe static profiles byte for byte.

Gates on this tree

The record validation/RELEASE_0_4_3.md follows once the final assets are checked.

Gate Result
GNU Release CTest, five OpenBLAS kernels (Prescott, Nehalem, Sandybridge, Haswell, Zen) 433 of 434 on each before the experiment rerun; the remaining validation_records row passes on this tree; 0 warnings
GNU Debug CTest (-fcheck=all) 433 of 434, the same row; 0 warnings
Example fallback scan, five kernels all 66 decks, no fallback note
Python tests 3232 passed, 47 skipped
Distributions (build, twine check) pass; licence in both; the wheel holds only cabledyn
pre-commit, release metadata, tree audit, git diff --check pass
OpenFAST build gate (gfortran, double precision) pass, OpenFAST-v5.0.0-CableDyn-0.4.3
Linearisation (release single-precision openfast.exe and double build) pass, with the figures of 0.4.2
Clean-machine check of the preflight assets (run 44735202382) 14 of 14: all 66 decks with the release driver, both CompMooring = 5 restarts bit-identical, the processor guard, and the wheel in a fresh conda environment
Bodies suite 102 of 102

Example outputs and timing against 0.4.2

Each comparison runs every shipped deck with both versions and compares every output file.

  • Outputs: 64 of the 65 decks both versions run give byte-identical outputs, with the GNU builds and with the release drivers. The one change is viv_taut_line_crossflow.dat: its VIV phase is now second order in the time step (fix: the coupled VIV response converges at second order in the time step #171), and the fairlead tension changes by up to 0.34 %. viv_taut_cable_inline.dat uses the in-line VIV section, which 0.4.2 refuses.
  • Wall time:
    • GNU builds: geometric-mean ratio 0.981 (132.2 s to 132.0 s); the two torsion decks are 22-24 % faster.
    • Release drivers: ratio 1.013 (90.2 s to 90.8 s); the largest ratios are on decks that run in 0.03 s or less.

L5 records

The raw L5 results of the external rows were lost from the release host, so the records stay those of the 0.4.2 drivers, and the README and VALIDATION.md now say so. A same-session control run on a busy host timed the 0.4.2 release driver before and after the 0.4.3 preflight driver:

Case 0.4.2 driver 0.4.3 driver
Lazy-wave spans 1.43-1.45 / 2.54-2.57 / 5.35-5.44 s 1.45 / 2.57 / 5.46 s
One-hour VolturnUS-S chain 4.62 s 4.71 s
WD0050 chain 0.205 s 0.207 s

Every model reproduces its reference response.

@codex review

Finite-EI cables had axial (BA) and torsional (GJ|K|Bt) Kelvin-Voigt
damping but no transverse structural damping. The LINE TYPES EI column
now takes EI|Bb, a damping moment Bb [N m^2 s] times the curvature
rate, or EI|-zeta|f, the damping ratio of a pure bending mode of
frequency f (Bb = zeta EI/(pi f)). A mode is damped in its bending
share, zeta = omega Bb/(2 EI) * EI k^2/(EI k^2 + T).

The rate is the curvature rate in the frame carried with the tangent,
(I - t t^T) dkappa/dt, zero for a rigid motion whose rotation is normal
to the tangent; under active line control it includes the change of
the unstretched length. Force and Jacobians are exact. One helper
returns the axial and bending damping for the assembly, the end forces
and the node recovery (unchanged without bending damping); the viscous
moment stays out of the elastic moment and curvature outputs while the
end forces carry it. Both deck readers refuse a damping on EI = 0 and
the malformed forms by name; the Python builder writes and reads both.

Tests: kernel Jacobians and the length-rate term against central
differences and an independent dissipation function; free decays of a
pinned cable against the bending-share damping ratio to 0.05 %; the
node recovery free of the viscous moment; the deck route.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The synchronisation model carried the cross-flow lift only. A finite-EI
line type now takes the in-line shedding force of Ulveseter, Thorsen,
Saevik and Larsen (Mar. Struct. 60, 2018), an optional VIV section row
TypeName ClX [alpha]: per unit length 1/2 rho D ClX cos(phi_x) U v_n
along the relative normal flow, the direction of the drag, with its
phase slaved to twice the cross-flow one,
d phi_x/dt = 2 (d phi/dt) [1 + alpha sin(phi_xd - phi_x)], phi_xd the
phase of the in-line motion from the running RMS values (alpha 0.15 by
default). ClX > 0 needs the type's cross-flow Cl > 0.

The kernels give the node and per-length forces with exact Jacobians and
the in-line motion and phase rate. The Hermite cable integrates the force
with the lift at its Gauss points; the phase is explicit with the Heun
commit (its end rate from the end cross-flow rate) and sits in the
step-recovery, snapshot and HFMF checkpoint state (3 nn more values in
the VIV block). Both deck readers take the section and refuse malformed
rows, unknown or repeated types, ClX on a type without the lift, alpha
outside [0, 1) and in-line VIV on an EI = 0 line, which follows later;
the Python builder writes and reads it.

Tests: kernel forces and Jacobians against central differences; on the
taut line of the lock-in benchmark the in-line response runs at twice
the cross-flow frequency (2.000 on the Hermite march, 1.95 through the
deck) with A/D 0.19 against 0.03 from the lift alone; ClX 0 reproduces
the cross-flow-only run bit for bit; snapshot replay, the recovering
step and the mirror round trip are bit-exact. New example
viv_taut_cable_inline.dat.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
#145 gave finite-EI cables the in-line shedding force of Ulveseter et
al. (2018) and refused it on EI = 0 lines (the two-node elements with
consistent mass). Those lines now take it too, so a VIV section row
acts wherever the cross-flow lift does.

On an EI = 0 line the in-line force is a node force like the lift,
1/4 rho d C_x (l_i + l_i+1) U cos(phi_x) v_n, with the lift's wetted
fractions, node tangent and neighbour-averaged kinematics, and its
exact velocity and neighbour-position Jacobians. The in-line phase
starts at twice the cross-flow phase and advances with the Heun
commit, its end-state rate slaved to the end cross-flow rate, as on
the cable.

The in-line phase and mean squares follow the cross-flow state
everywhere it goes: the sub-step slots, step recovery, the system's
rollback and frame snapshot, model copies, and the OpenFAST line
checkpoint (3 more reals per node).

Checks:
- the deck of #145 as an EI = 0 line matches the finite-EI cable:
  CF A/D 0.8228 vs 0.8219, IL A/D 0.1915 vs 0.1905, the same
  frequencies;
- coupled restart and rewind with in-line VIV continue bit for bit,
  and a rebuild without the VIV state does not;
- the sub-stepping restore of tests/test_model carries the in-line
  state;
- full ctest 412/412 on the default and Prescott kernels, Debug
  -fcheck=all on the VIV, model and system tests, pytest, the example
  fallback scan clean.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
A VIV study needs the energy ledger of the response: what the VIV loads
put into a line against what the drag and the structural damping take
out, along the line and over the run. The OUTPUTS channels
VIVPow<L>, DragPow<L> and DampPow<L> [W] report the power these loads
deliver to line L at the committed state, and <name><L>N<J> [W/m] the
same per unit unstretched length about node J.

The power of a load on an element is its nodal (finite-EI:
generalised) load times the element's velocities, the exact rate of
work of the load the solver applies; a node load delivers f . v at its
node. A node takes half of each of its elements' power, so the node
values sum to the line total, and an element's damping power, hence a
node's, is never positive.

- VIV: the cross-flow lift and the in-line force of a VIV section row.
- Drag: the Morison drag, and on a finite-EI line its ATTACHMENTS'
  drag.
- Damping: the axial dashpot BA, and the bending dashpot Bb on a
  finite-EI line. The torsional Bt and the viscoelastic and Syrope
  laws' internal damping are not included (documented).

They run on EI = 0 lines and finite-EI cables, on the single-line,
point-system, multibody, coupled (OpenFAST, FAST.Farm) and C API
routes; a static run reports 0. The residual-only Morison drag call of
the finite-EI element, written out three times, is one helper now
(element_drag_load), with the same arguments as before.

Checks:
- tests/test_hermite_load_power: the work of the three loads closes a
  weightless cable's energy balance within 0.04 % of the largest work
  in a bending-dashpot decay, an axial-dashpot decay, a still-water
  drag decay and a VIV run in a current;
- tests/test_load_power_deck: on the locked-in in-line VIV line the
  means balance within 0.04 % of the VIV power as a finite-EI cable,
  an EI = 0 line and two EI = 0 lines joined at a free point; the
  per-length channels of all nodes sum to the line total; malformed
  channels are refused by name;
- tests/test_viv_coupled: on the coupled route both lines balance
  within 0.01 %;
- the Python deck reader and builder mirror the channels (native
  parity of 13 spellings);
- full ctest 414/414 on the default and Prescott kernels, Debug
  -fcheck=all on the new and touched tests, pytest, the example
  fallback scan clean.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
A VIV study describes a response by its cross-flow and in-line
amplitude along the span, the response frequencies against the flow,
the excited modes, how far it is a travelling wave and its orbits.
viv_response computes them from a line's node positions (a per-line
.p.out file, a MoorDyn-F line file, or the L<L>N<J>p channels of a
main output) over a window.

Each node's motion about its mean position is resolved into its local
flow frame: the mean tangent t from End A towards End B, the in-line
axis along the flow normal to t, and the cross-flow axis t x in-line;
a node whose tangent lies along the flow has no frame (nan). The
VIVResponse gives:
- RMS and half peak-to-peak displacement over D, for either direction;
- each node's and the line's dominant frequency (Hann periodogram,
  parabolic peak) and f D/U_n against the synchronisation band;
- modal weights at every sample, on the sines of a taut line with
  fixed ends or on given shapes, with the dominant mode and the
  unexplained share;
- the complex Fourier shape at a frequency and its travelling-wave
  index (the travelling index of Feeny's complex orthogonal
  decomposition, J. Sound Vib. 310, 2008, 77-90);
- orbits and the in-line/cross-flow phase difference;
- a CSV export and envelope and orbit plots.

Review fixes:
- the default sines need the whole line (a main output with only some
  nodes takes given shapes); a response with no dominant frequency has
  a nan travelling-wave index;
- the nodes of a main output keep their source numbers (export, node
  lookup);
- a modal fit on shapes that are not independent at the nodes is
  refused with the rank;
- a node or component that does not move (range at round-off of its
  position) has no frequency, no Fourier shape and no orbit phase.

Checks: synthetic standing and travelling waves with known amplitudes,
frequencies, modes, phases and index (python/tests/test_viv.py); the
end-to-end run of examples/viv_taut_cable_inline.dat reproduces the
deck's A/D 0.82 cross-flow and 0.19 in-line at twice the frequency,
modes 1 and 2, f D/U_n inside the band, a standing wave. Full pytest,
ruff, mypy strict, Sphinx -W.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The stress recovery for fatigue took the curvature magnitude at the two
extreme fibres, T/A +/- E kappa r, so it did not follow the bending
plane. Under combined cross-flow and in-line VIV the plane turns, and a
whirling section, whose curvature magnitude stays constant, showed no
cycles and no damage, although every point around it reverses once a
revolution.

section_stress takes the curvature vector on the section axes, the
CurvX/CurvY channels of a finite-EI line (bending strain -(kx x + ky y)),
and returns the stress at points around the section,

  sigma(theta) = T/A - E r (kappa_x cos theta + kappa_y sin theta).

damage_along_arc(..., quantity="section", points=16) counts each point,
keeps the largest damage and reports its angle (critical_angle on
DamageProfile, None for the other quantities). The node-channel reader
now knows the CurvX and CurvY prefixes.

Checks: the stress of a known curvature vector point by point; in a
fixed bending plane the extreme points are cable_stress's two fibres;
a whirling section has no damage under "stress" and its point-by-point
damage under "section"; missing or mismatched channels and a T-N curve
are refused. pytest, ruff, mypy strict, Sphinx -W.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…150)

A stepped-current study holds the current at a sequence of speeds and
lets each step start from the state the previous one left, so the
response follows its branch up and back down; separate runs would each
start from rest. The standalone driver has no restart file, but its
MoorDyn-C Currents 2 table (a depth profile in time) carries the whole
sweep in one run.

current_steps(speeds, hold=, ramp=, direction=) returns a CurrentSteps
schedule: each speed held, then ramped linearly to the next. Its write()
puts current_profile_dynamic.txt in the deck folder (uniform over two
depths), duration is the TMax to give, windows(settle) the analysis
window of every hold, and responses() the viv_response of each. The
table is resampled at its smallest interval, the ramp, so a hold must be
a whole number of ramps (refused otherwise).

Checks: the schedule, the file rows and the refusals
(python/tests/test_viv.py); end to end, the in-line VIV example under
0.10 -> 0.18 -> 0.10 m/s: the lock-in frequency rises with the current
and returns with it. Full pytest, ruff, mypy strict, Sphinx -W.

Review fix: the file holds the last speed two ramps past TMax, since the resampled
table takes its first sample over its last interval (and a rounded step can drop the
last sample), which had returned the run to the first speed within the last hold.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The synchronisation model with in-line VIV resolves the cross-flow
response at the shedding frequency and the in-line one at twice it.
Riser measurements show cross-flow components at three times the
frequency as well, and they dominate curvature fatigue at high modes.
Ulveseter, Thorsen, Saevik and Larsen (Appl. Ocean Res. 90, 2019,
101856, Eq. 6) add a higher-harmonic force of the lift's form at the
sum of the two phases,

  f_hh = 1/2 rho D C_hh cos(phi + phi_x) U (t x v_n),

which acts across the flow at 3 omega (and at omega and 5 omega
through the turn of the relative flow), and along it at 2 and 4 omega.

A fourth VIV section column Chh (default 0; Chh > 0 needs ClX > 0)
adds it on EI = 0 lines and finite-EI cables:
- the cable adds k_hh cos(phi + phi_x) to the node values of the lift
  scale its elements integrate;
- the EI = 0 line adds a second node lift at the phase sum.
It needs no state of its own, and its derivatives enter the tangent
with the lift's. The Python deck reader and builder take the column.

Checks:
- tests/test_viv_inline_deck D5: Chh 1.0 raises the quarter-span
  in-line response at four times the cross-flow frequency from A/D
  0.0017 to 0.0070, the EI = 0 line within 2 % of the cable; Chh 0 is
  the row without the column to the printed digits; Chh without
  in-line VIV, a negative Chh and five columns are refused;
- tests/test_hermite_load_power L4 now carries Chh 1: the load work
  still closes the cable's energy balance (3e-5);
- native/Python parity of nine row forms;
- full ctest on the default and Prescott kernels, Debug -fcheck=all
  on the VIV, model and system tests, pytest, Sphinx -W.

A second commit documents the published coefficient sets as deck columns: Ulveseter
et al. (2018) Table 3 and (2019) Table 1 with the C_hh of their Section 4, mapped onto the
LINE TYPES and VIV columns (cF = f0, dF = df, Ca = C_M - 1), checked against the papers.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
A line restrained in torsion takes GJ and Irn from the 14-column
LINE TYPES row (GAs GJ Irt Irn), and that row could not carry
MoorDyn's cross-flow VIV columns: a torsion-restrained power cable had
no way to get VIV. The row now takes Cl, or Cl dF cF, after its
fourteen columns (15 or 17 columns), with MoorDyn's dF = 0.08 and
cF = 0.18 where only Cl is given and the checks of the 11- and
13-column rows. The VIV wiring does not depend on the row width or on
the nodal twist, so nothing else changes.

The Python deck reader takes the widths, reads Cl dF cF after the
fourteenth column and names them cl, df, cf in the field paths; the
builder writes them after the wide row instead of refusing the
combination.

Checks (tests/test_viv_inline_deck.f90, D6): a 17-column row free in
torsion is the 13-column deck to the printed digits; a 15-column row
is the 17-column one with 0.08 and 0.18; on the line rigid in torsion
at both ends (nodal twist solved) the lift locks the cable in at A/D
0.66 and 0.292 Hz, the amplitude and frequency of the line free in
torsion, as it must be with no torque-tension coupling. Python: read,
kept and refused like the native reader (test_deck_file.py), builder
round trip (test_builder.py). Full ctest under both BLAS kernels,
pytest, ruff, mypy, Sphinx -W, pre-commit.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The Delta Flume tests of Chaplin et al. (2005, J. Fluids Struct. 21,
3-24) are the stepped-current VIV benchmark of a 13.12 m vertical
tension riser. Before any VIV comparison the riser model itself has to
be right: its mass, added mass, bending stiffness and the tension that
falls with the submerged weight down the line. Table 2 of the paper
gives the first seven still-water natural frequencies at five top
tensions, so CableDyn's modal analysis can be checked against 35
measured values with no fitted parameter.

validation/experiments/chaplin/chaplin_natural_frequencies.py writes a
deck per series (EI 29.9 N m^2, mass ratio 3.0 with the internal water,
which reproduces the paper's 12.1 N/m submerged weight, pinned ends,
the unstretched length that gives the Table 2 top tension), runs the
modal analysis about the static state and compares. Assumed: fresh
water, Ca 1, and EA 5.5e6 N, which the paper does not give.

Result: 1.87 % high on average over 34 values, mean absolute error
2.28 %, largest 4.88 % (series A mode 1). The 35th, series C mode 2,
is measured at 1.423 Hz against 1.619 Hz; its ratio to the first mode
(1.82) departs from every other series (1.97-2.16), so it is reported
apart and kept in the all-35 statistic (2.61 %). Ca 1.1 would bring the
mean to +0.62 %; the default stays 1. 128 elements and ten times EA
change no frequency by more than 3e-6 and 1.2e-4.

Added to VALIDATION.md as L3-C, to the experiments index and the
references; the reference is checked against Crossref.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…n) (#154)

CableDyn's VIV follows the synchronisation model of Thorsen et al. The
other model class in common use is the wake oscillator: a van der Pol
oscillator stands for the near wake and drives the lift. A model-form
comparison needs both on the same line.

An optional WAKE OSCILLATOR section row TypeName CL0 [St [eps [A]]]
gives a line type the wake oscillator of Facchinetti, de Langre and
Biolley (J. Fluids Struct. 19, 2004), distributed along the line as in
Violette et al. (2010):

  q'' + eps om (q^2 - 1) q' + om^2 q = (A/D) y'',  om = 2 pi St U_a/D,
  lift per unit length 1/4 rho D CL0 q U^2 across the relative flow,

with y'' the node's cross-flow acceleration (their acceleration
coupling), U_a the ambient flow speed normal to the line (their free
stream, so the cable's own cross-flow velocity does not modulate the
wake frequency), U the relative normal flow speed of the lift (as the
synchronisation lift), and the defaults St 0.2, eps 0.3, A 12 they fix from
forced-wake data (checked in the PDF). Their fluid damping
gamma om rho D^2 y', gamma = Cd/(4 pi St), is the linearised Morison
drag CableDyn applies in full, so the wake adds the lift only. It
replaces the synchronisation lift of the type, which keeps Cl 0; a line
takes one VIV model.

It is a mode of the cable's VIV configuration: Set_VIV takes the
per-element (St, eps, A), the lift scale is 1/4 rho D CL0, and the
element lift takes q at its end nodes in place of cos(phi), with the
exact Jacobians at fixed q. The state (q, q') per node is advanced
beside the cable like the phase: a predictor under the step-start
acceleration for the step's lift, then from the committed state with
the acceleration linear to the converged end one, at the step-start
wake frequency (classical Runge-Kutta in sub-steps with
om max(1, eps q^2) dt <= 0.1, so a stiff eps stays stable, redone with
twice the sub-steps should the rate grow on the way,
CD_VIV_Wake_Advance). q starts on its limit cycle at the VIV phase
seed, q = 2 cos(phi_0). The state is in the step-recovery, snapshot and
checkpoint state and the OpenFAST mirror (2 nn values after the
in-line block). Finite-EI cables only for now; an EI = 0 line with the
wake oscillator is refused by name, as is a line mixing the models.

The Python deck reader takes the section with the native checks, the
builder writes it (LineType.wake_cl0, wake_st, wake_eps, wake_a).

Checks:
- tests/test_viv_wake_kernel.f90: the free oscillator settles on q = 2
  at its period (the van der Pol shift eps^2/16); forced at its own
  frequency it follows the lift magnification of Facchinetti's Eq. 17
  within 0.2 % at y0 = 0.05 and 1.9 % at 0.5 (the third harmonic their
  one-harmonic balance leaves out; halving the step changes K by
  2e-5), the sub-stepped advance converges at fourth order, and with
  eps 20 from (2, 0) 0.1 s calls stay finite and match 1 ms calls
  within 1e-13;
- tests/test_viv_wake_deck.f90: on the taut cable of the in-line VIV
  gate with Cl 0 and CL0 0.3 the wake locks the cable in at 0.282 Hz,
  below its own 0.300 Hz, at a steady A/D 0.585; uncoupled (A 0) it
  drives the cable at 0.298 Hz at A/D 0.099; over the lock-in the drag
  and damping take out the wake lift's 2.84 W within 0.21 %; CL0 0 is
  the deck without the section to the printed digits; eleven refusals;
- tests/test_viv_coupled.f90: on the coupled route the cable's
  checkpoint block grows by 2 nn, the restarted run continues bit for
  bit, a rebuild without the VIV state does not, and a rewound step
  repeats;
- Python: rows read, kept and refused as natively; the builder writes
  the values given; five native/Python parity decks agree.
Full ctest under both BLAS kernels, Debug -fcheck=all on the new
tests, pytest, ruff, mypy, Sphinx -W, pre-commit.

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

Thorsen, Saevik and Larsen (J. Fluids Struct. 61, 2016) make the
cross-flow excitation coefficient of the synchronisation model a
function of the vibration amplitude, C_v(y0/D), fitted to riser tests;
drag amplification laws of vibrating cylinders (Chaplin et al. 2005,
Eq. 4) are functions of the same amplitude.

An optional VIV AMPLITUDE section, rows TypeName AD Cv [CdAmp], gives a
line type a table over the amplitude ratio A/D of the coefficient Cv
that replaces its LINE TYPES Cl and a factor CdAmp (default 1) of its
normal drag coefficient, linear between rows and held beyond the end
ones (2 to 32 rows, A/D strictly increasing, the type with Cl > 0).

Each node estimates its amplitude as Thorsen et al. do (Eq. 8 and 9):
half the travel between two zero crossings of the cross-flow velocity
(trapezoidal, crossings placed by linear interpolation; the travel is
carried signed, so a reversal through a sample at zero ends a half
cycle and a touch does not), smoothed by a tenth of the difference per
step (k_a = 0.1/dt). The velocity is taken across the ambient flow
normal to the line, their incoming flow: the phase's axis across the
relative flow turns with the node's own velocity and read the amplitude
16 % low at A/D 0.68. An element reads its table at the mean of its free
nodes over its diameter, at the committed estimate, so the Jacobians
only scale. The estimate is updated at the end of each converged step
from the committed one.

The estimate is four more rows of the per-node VIV motion statistics
(viv_ms, CD_VIV_NMS_AMP rows with a table, CD_VIV_NMS without, so a
deck without a table runs and checkpoints exactly as before), carried
by the step-recovery, snapshot and system rewind paths (the system
snapshot and the sub-step slots follow a line that gains tables, a
restore across that change is refused) and appended to the line VIV
checkpoint blocks (4 nn values). VIVAmp<L>N<J> writes the smoothed
estimate [m] on a line with a table. The Python deck reader mirrors the
section and channel checks; the builder writes the table
(LineType.viv_amplitude).

Checks:
- tests/test_viv_amplitude_kernel.f90: a harmonic velocity gives the
  amplitude within the trapezoidal (om dt)^2/12 (1.95e-6; halving the
  step quarters it), the smoothing is Eq. 9 step by step, a touch of
  zero ends no half cycle while a reversal through a zero sample does
  (a triangular velocity on its zeros: nine half cycles at y0 = 1
  exactly), a growing oscillation gives each half
  cycle's mean amplitude within 0.1 %, tables interpolate and hold,
  a NaN amplitude stays inside a table, and bad tables are refused by
  name;
- tests/test_viv_amplitude_deck.f90, on the taut gate line as an EI = 0
  line and a finite-EI cable: Cv = Cl with CdAmp 1 is the deck without
  the section bit for bit; VIVAmp follows the steady mid-span response
  within 0.03 % (A/D 0.68); a Cv falling to zero at A/D 0.3 holds the
  response at 0.28; Cv 0 with CdAmp 2 settles at the static offset of
  Cd 2.4 to six digits; thirteen refusals;
- tests/test_viv_coupled.f90: with tables on both lines the checkpoint
  blocks grow by 4 nn, the restarted run continues bit for bit, a
  rebuild without the VIV state does not, and a rewound step repeats;
- tests/test_system.f90: a snapshot after a line gains tables is resized
  and restores, a restore of one taken before is refused untouched;
- tests/test_model.f90: tables set on a sub-stepped VIV line after it
  has stepped, the sub-step slots take the six rows and 100 more steps
  converge;
- Python: rows read, kept and refused as natively, the builder writes
  them, six native/Python parity decks agree.
Full ctest under both BLAS kernels, Debug -fcheck=all on the new and
changed tests, pytest, ruff, mypy, Sphinx -W, pre-commit.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…DIAMETER section) (#156)

A lazy-wave cable's buoyancy modules are ATTACHMENTS rows on the bare
cable: their mass, volume and drag are point loads, and the cable's line
type keeps its own diameter for its drag, added mass and buoyancy. Its
VIV then also took that diameter, while the shedding of a module zone is
set by the larger module diameter.

An optional VIV DIAMETER section row TypeName Dviv gives the VIV model
of a line type a diameter of its own: the shedding frequency and band
(or the wake frequency and its coupling A/D), the lift, in-line and
higher-harmonic forces and the amplitude ratio of a VIV AMPLITUDE table
take Dviv, the drag, added mass, buoyancy and contact keep Diam. A
module zone is a section with a copy of the cable's type, Dviv the
module diameter and its own coefficients; the docs say how, and that the
shedding of alternating modules and bare cable has no established model.
The type needs VIV (Cl > 0 or a WAKE OSCILLATOR row); one row a type.

EI = 0 lines take it through the diameter CD_Set_Model_VIV already
has; finite-EI cables through Set_VIV dviv (per element, 0 = the drag
diameter, kept across a later Set_Drag), which viv_drag_terms and the
amplitude-table helpers read. The Python deck reader mirrors the checks,
the builder writes the row (LineType.viv_diameter).

Checks:
- tests/test_viv_diameter_deck.f90, on the taut gate line as an EI = 0
  line and a finite-EI cable: Dviv = Diam is the deck without the
  section bit for bit; with the synchronisation lift the line locks in
  at 0.287 Hz, and with Dviv 0.2 m its band (0.075-0.195 Hz) lies below
  the mode and the phase runs at its top, 0.196 Hz; an uncoupled wake
  oscillator drives the cable at 0.149 Hz against St U/Dviv = 0.15 Hz;
  six refusals;
- Python: rows read, kept and refused as natively, the builder writes
  them, six native/Python parity decks agree.
Full ctest under both BLAS kernels, Debug -fcheck=all on the VIV tests,
pytest, ruff, mypy, Sphinx -W, pre-commit.

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

An optional VIV DAMPING section row "TypeName a b [c]" splits the normal
drag of a finite-EI line type in the frame of the ambient normal flow:
Morison's drag in line (with the VIV AMPLITUDE drag factor) and, across
it, 1/2 rho D [C_dy |w_y| + c Cdn |w|] w_y with C_dy = a + b y0/D, the
cross-flow hydrodynamic damping of Thorsen et al. (2016, Eq. 1; 0.31
0.89 is their fit) at the amplitude estimate y0. Below an ambient
normal flow of 0.01 m/s the force blends to Morison's, where the axes
are undefined.

CD_Morison_Drag_Split_Per_Length_Jac computes the split force with
exact Jacobians in the relative velocity and the tangent. The dynamic
finite-EI drag element takes it in the residual, the tangent and the
element drag load. Statics is unchanged: at rest in the flow the split
force is Morison's. Set_VIV takes cdy (2 or 3, n_elem), checks it and
rolls it back with the rest of the VIV state.

"0 0 1" is Morison's drag with the drag factor on the in-line part
alone, so an amplified drag no longer damps the cross-flow VIV motion
twice. Thorsen's quadratic form drops the linear U y' part of
Morison's cross-flow damping and suits cross-flow-only VIV.

The section needs Cl > 0 and a VIV AMPLITUDE table; EI = 0 lines are
refused. The Python deck reader and builder take the section, with
parity tests against the native reader.

Tests: test_viv_cf_damping_kernel (Morison limits, the cross-flow
law, the blend, Jacobians against central differences, refusals) and
test_viv_cf_damping_deck (0 0 1 equals the line without the row,
1.5e-8 in Release and checked to 1e-4; the in-line-only factor gives
the Cd 2.4 static offset; Thorsen's form raises A/D 0.68 to 1.02;
power balance within 0.2 %; six refusals).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…158)

A finite-EI dynamic step with bending damping (EI|Bb) could fail on a
soft line: Newton stagnated just above the tolerance and the step
recovery's smaller steps made it worse. Found on a 20 m riser on a
20 kN/m tensioner (EA_eff 4e5 N, 80 elements, Bb 500 N m^2 s, dt 1e-3 s)
in VIV: the residual stopped at 1.06 times the tolerance.

The bending damping takes the curvature rate from the Newmark velocity,
whose rounding is gamma/(beta dt) times that of the coordinates, so its
nodal forces round at about Bb gamma/(beta dt) eps |r|/l0^3. The force
scale's round-off floor covered the axial forces alone,
(EA + BA gamma/(beta dt)) eps |r|/l0. It now takes the bending forces,
(EI + Bb gamma/(beta dt)) eps |r|/l0^3, too; that term binds only where
it is the larger, so every example and test deck gives bit-identical
outputs (1685 arrays over 70 decks).

Test: test_hermite_bending_damping_dynamic B6, a soft line (EA 4e5 N)
with Bb 1e4 N m^2 s at dt 1e-4 s. On the old floor its first step
stagnates at 2.2e-7 against the tolerance 1e-8; now 200 steps converge
and the motion stays bounded.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
A finite-EI line could not end on a Free point: the single-cable route
held both end positions and refused the point's loads, and with a Free
point the deck went to the multibody march, which holds cable ends. A
free-hanging cable or riser with a clump weight (forced-oscillation and
tow tests of power cables, installation, a dropped line) had no deck.

A Free point that ends exactly one finite-EI line as its End B, from a
Coupled or Vessel End A, on a standalone deck whose lines are all
finite-EI and that has no BODIES, RODS or other Connect/Free point, is
now that line's free end:
- the point's Mass, Vol, CdA and CA act on the End B node as an end
  attachment (weight and buoyancy, added mass, fluid inertia, drag);
- the static solve holds the end first on the chord from End A towards
  the deck position, at the line's length stretched by its hanging
  weight (the taut seed; a vertical span of exactly the line's length
  has no catenary seed), then releases it with the attachment polish's
  warm-started Newton solve (new option release), so the run starts at
  the end's rest position;
- the dynamics hold End A alone;
- Point<P>p channels read the line's End B node.
The line must be pinned in bending at both ends and the end net heavy;
a torsion restraint at one end is noted and not solved. A deck in which
any Free point is not such an end, and a host-driven deck, keeps its
behaviour and its refusal messages. The Python deck reader takes the
same rule.

Checks (tests/test_free_end_deck.f90, a 7 m cable with a 200 kg clump):
- still water: top tension = submerged weight to 1e-6 relative, the end
  one stretched length below the top to 1e-7 m, nothing moves;
- 0.5 m/s current: the end rests where the EI = 0 route puts the same
  deck's Free point, within 2e-5 m at EI 3 N m^2 (0.134357 against
  0.134353 m), converging as EI falls;
- 0.05 m sway of the top at 0.05 Hz: the end follows in phase, the
  amplitude ratio within 1 % of the Rayleigh estimate (1.106 against
  1.104);
- a clamped end, a buoyant end and a second Free point keep their
  refusal messages;
- python/tests/test_deck_native_parity.py: the reader and the driver
  agree on the accepted and refused decks.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…nt (L3-W) (#160)

The SJTU truncated steel catenary riser of Wang et al. (2014, Marine
Structures 39, 131-156) in a 0.2 m/s uniform in-plane current is the
case Thorsen et al. (2017, Marine Structures 51, 134-151) use to verify
the time-domain synchronisation model that CableDyn's cross-flow VIV
implements. It checks the VIV term on a curved, partly grounded riser
with its own published coefficients, so no parameter is fitted.

validation/experiments/wang_scr/wang_scr_uniform_current.py writes the
deck (23.71 m riser between pinned joints, the top 21.0425 m out and
9 m up, D 0.024 m, 0.69 kg/m, EI 10.5 N m^2, EA 6.66e5 N; seabed
1 (kN/m)/m with friction 0.2; Cv 0.7, f0 0.18, df 0.08, CD 1.2, CM 2,
cross-flow only), runs it and compares:
- the r.m.s. dynamic cross-flow strain at the 25 stations of Thorsen et
  al. Fig. 17, read from the vector figure (axes calibrated on its tick
  marks; the figure's simulated maximum reads 99.6 % of the measured,
  as their text states);
- the dominant strain frequency 8.25 m from the lower end.

Result (200 elements, dt 0.005 s, 150 s):
- frequency 0.992 Hz against the measured 1.00 Hz (their model 1.02 Hz,
  Davies et al. 2026 1.04 Hz);
- largest r.m.s. strain 7.3 % above the measured largest (theirs 0.4 %
  below), the mean over the stations 3.6 % above;
- one r.m.s. peak more along the span than the test (7 against 6), as
  theirs, so the station-by-station ratio scatters;
- 400 elements and half the time step change the largest r.m.s. strain
  by 0.2 % and 0.1 % and leave the frequency unchanged.

Added to VALIDATION.md as L3-W, to the experiments index and the
references (Wang et al. 2014, Thorsen et al. 2017, Davies et al. 2026,
checked against Crossref).

Review fix: the script resolves a relative executable path against the
invocation directory (or PATH) and stops with its name when missing.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The theory manual gains "Published coefficient sets": each set with its
source, the tests it was fitted to and the line-type columns it maps to:
- Thorsen et al. (2016): the excitation curve Cv(y0/D) fitted to the 44
  NDP riser tests (0.35 at 0, a maximum of 1.00 at 0.46, zero from
  0.66) and the cross-flow damping 0.31 + 0.89 y0/D;
- Thorsen et al. (2017): the Wang et al. (2014) riser set (L3-W);
- Ulveseter et al. (2018): the in-line Cx 0.85, alpha 0.15;
- Andersen et al. (2024a): the forced-oscillation set of their Table 3
  (Cv 1.2, CD 1.1, Ca 1.0, band 0.11-0.26, i.e. dF 0.075 about cF
  0.185 here) and the unified Cv of about 2.0;
- Andersen et al. (2024b): the cross-flow average of their Table 3
  (1.28, 0.131, 0.066; their in-line set belongs to another in-line
  synchronisation form, which is said);
- Facchinetti et al. (2004): CL0 0.3, St 0.2, eps 0.3, A 12.

Thorsen et al.'s curve, read from the vector figure, is given as a
VIV AMPLITUDE table on 0.05 steps. Every value was read from the
papers; the two Andersen references were checked against Crossref.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The WAKE OSCILLATOR section now acts on EI = 0 lines as on finite-EI
cables; a deck with the wake oscillator on an EI = 0 line was refused.

At each free node next to a lifting element the van der Pol wake
variable q drives the lift 1/4 rho D CL0 q U^2 over the node's
tributary length, in the direction of the synchronisation lift. The
forcing is the node's smoothed cross-flow acceleration, and the wake
frequency that of the ambient flow normal to the line. The step follows
the finite-EI cable: q predicted at the step's evaluation point under
the step-start forcing, then advanced from the committed state with the
acceleration linear over the step (the shared CD_VIV_Wake_Advance).

The state (q, dq/dt) is held in the line's VIV phase and first motion
statistic, so the rollback, step-recovery, sub-step and checkpoint
copies carry it unchanged; the checkpoint layout of an EI = 0 line does
not change. CD_Set_Model_VIV takes the per-element (St, eps, A) as an
optional wake argument; it refuses in-line VIV and amplitude tables
with it, as the finite-EI cable does.

Gates (tests/test_viv_wake_deck.f90, test_viv_coupled.f90):
- the taut line of the wake gate with EI = 0 locks in at 0.282 Hz and
  A/D 0.5856, against 0.5850 for the finite-EI cable (0.1 %), and
  responds uncoupled (A 0) at the wake's 0.298 Hz, A/D 0.0989 against
  0.0986;
- the drag and damping take out the wake-lift power within 0.3 %;
- CL0 0 is the deck without the section, bit for bit;
- on the coupled route a checkpoint restart continues bit for bit with
  the EI = 0 wake state, a rebuild without it does not, and a rewound
  step repeats;
- a line mixing the two models and an EI = 0 type that keeps its Cl
  are refused by name.

The Python deck reader accepts the wake oscillator on EI = 0 lines.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…face (#163)

The statics of a finite-EI line through the free surface could fail at
isolated currents. Each Gauss point's hydrodynamic load switched on and
off at the waterline, so the residual jumped and an equilibrium with a
point at the surface had no root. A light lazy-wave cable (OD 31 mm,
0.919 kg/m, EI 0.217 N m^2, its top 0.75 m above the surface) in a
cross current solved at 0.399 and 0.401 m/s and failed every static
route at 0.400 m/s: the load stepping stopped at a drag fraction of
0.997, and Newton with the live drag stagnated at the same residual
from every start. 99, 101 and 120 elements solved; at 0.400 m/s the
third Gauss point of the element through the surface sat at z = 0
(-0.16 mm at 0.399 m/s, +0.13 mm at 0.401 m/s).

The drag, added mass, wave Froude-Krylov load and VIV lift of a Gauss
point now carry the immersed fraction of its section, linear across one
diameter about the surface (CD_HermiteCable_Immersion), with its
derivative in the Jacobians. A point deeper than D/2 has the fraction 1
exactly, so fully submerged elements are unchanged to the bit; the EI =
0 path already used a continuous wetted fraction. The VIV lift element
takes the diameter as an optional argument (without it, the old switch).

Gates:
- test_hermite_drag_dynamics E: an element through the surface: the
  position Jacobian matches central FD (1.6e-10), the load changes by
  at most 3.3e-4 of its size per 0.1 mm as the element moves through
  the surface (a switch jumped by a Gauss weight, 5/18 or 8/18), and a
  fully submerged element is unchanged to the bit;
- test_static_surface_crossing: the cable above at 21 currents,
  0.390-0.410 m/s: every speed solves, the deflection and top tension
  grow monotonically, and 0.400 m/s lies between its neighbours (the
  deck fails on dev at 0.400 m/s only).

The full suite passes unchanged under both BLAS kernels.

Review fix: in waves, a Gauss point above the instantaneous surface but
inside the band takes the wave velocity and acceleration at the surface
(the Wheeler kinematics vanish above it), so the FK load and the wave
drag stay continuous (test_hermite_wave_forcing E: at most 2.7e-4 of
the load per D/2000 step).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…164)

A net-buoyant ATTACHMENTS module on a finite-EI line had no static
position near the free surface. Its buoyancy switched off at its node's
height, so its weight in air took it below the surface and its weight
in water above it. The static start stopped with "attachment <k> ...
crossed the still-water level".

The buoyancy, Froude-Krylov force and drag of an attachment now carry
the immersed fraction of the sphere of its volume (of its drag area
without one), the formula a Rigid6 body already uses:
h = clamp(eta - z + r, 0, 2r), phi = h^2 (3r - h)/(4 r^3). The load is
continuous with its slope through the surface.

- Dynamics: the residual takes phi, with the height derivative in the
  position tangent. A node deeper than r keeps phi = 1 exactly, so its
  loads are unchanged to the bit.
- Statics: CD_HermiteCable_Static_Solve takes optional att_node,
  att_buoyancy and att_radius. It adds the buoyancy lost at the surface,
  rho g V (1 - phi), in each Newton iterate, with its stiffness and
  potential. The polish no longer needs the wet/dry fixed point or the
  crossing error. The torsion stage keeps its fixed point over the
  smooth loads.
- Deck waves: a node above the instantaneous surface takes the wave
  kinematics at the surface instead of zero, as the internal waves of a
  cable already did (#163). Line Gauss points interpolated from such
  nodes therefore change too.

The added mass stays in the constant mass matrix. The modal analysis
leaves out the surface's hydrostatic stiffness, a module's as a line's
own.

Gates:
- test_attachment_surface:
  - K1, the fraction: its slope and potential match central FD (1e-8)
    and are continuous at the band edges; r = 0 keeps the old switch.
  - K2, the static solve: with the node in the band the residual gains
    rho g V (1 - phi) and the tangent column matches central FD
    (8.7e-9). Deeper than r the residual and tangent are bit-identical
    to the solve without the term.
- test_static_dry_weight gate 9, the module of the old failure
  (200 kg, 2 m^3, on a slack cable):
  - it now floats 0.348 m below the surface (phi 0.81);
  - the march holds that height over 3 s (drift 0);
  - in 1.5 m, 6 s Airy waves it heaves between -1.21 and +0.48 m
    through its band for 18 s, every step converged.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…red motion (#165)

The wake oscillator took its frequency from the ambient flow normal to
the line. A cable moved through still water, by a tow or by its floater,
therefore never shed, and the model could not drive floater-induced VIV.

A new optional section VIV FRAME, rows `TypeName Tf`, sets the flow
frame of a line type's VIV model. With Tf > 0 [s] the free stream is the
ambient flow relative to the node's velocity, low-pass filtered over Tf:

  U_a = |(u - vbar)_n|,
  vbar_{n+1} = vbar_n + (1 - exp(-dt/Tf)) (v_{n+1} - vbar_n)

This is the exact first-order filter of the converged end velocity held
over the step, started at the node's initial velocity. The slow motion
counts as flow, and the VIV velocity, much faster than Tf, is filtered
out.

In this commit the frame sets the wake oscillator's frequency, and a row
needs a WAKE OSCILLATOR row. The synchronisation model's ambient-frame
quantities (the amplitude estimate, the cross-flow damping split) can
take the same frame later without changing the input. With Tf 0 or no
row, runs are unchanged to the bit.

- Finite-EI cables: wo_s grows from 2 to 5 rows per node when a node has
  Tf > 0. The recovery, snapshot and checkpoint copies carry the new
  rows, and the mirror packs SIZE(wo_s).
- EI = 0 lines: the filtered velocity takes rows 3-5 of the VIV motion
  statistics (CD_VIV_NMS_AMP rows; a wake line has no amplitude table),
  so the state get/set, rollback and checkpoint already carry it.
- The core setters take wake(4, ne) = Tf as an optional fourth row.
- Deck reader, Python reader and builder (viv_frame_tf), docs and
  CHANGELOG.

Gates:
- test_viv_wake_deck W6: a finite-EI taut cable towed at 0.15 m/s
  through still water, with Tf 10 s, locks in at A/D 0.588, 0.283 Hz.
  The same cable held in a 0.15 m/s current gives 0.585, 0.282 Hz.
  Without Tf the towed cable reaches A/D 0.042.
  On the held cable Tf changes A/D by 0.03 % on both cores, and the
  cores agree within 0.1 %.
- test_viv_wake_deck W7: VIV FRAME refusals by name (malformed row,
  unknown or repeated type, negative Tf, a type without a wake
  oscillator).
- test_viv_coupled: with a VIV FRAME the checkpoint blocks grow by the
  filtered velocity; the restarted run continues bit for bit, a rebuild
  without the VIV state does not, and a rewound step repeats.
- Python reader, builder and native-parity cases for the section.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…amping split and phase input (#166)

With a VIV FRAME row and Tf > 0, the synchronisation model's ambient-frame features (the amplitude
estimate feeding VIV AMPLITUDE Cv and CdAmp, and the axes and blend of the VIV DAMPING split) and
the cross-flow and in-line motion fed to its phases now use the node velocity relative to its
velocity low-pass filtered over Tf, on finite-EI cables and EI = 0 lines. The relative flow, and so
the axes and the shedding speed, are unchanged. A cable moved through still water by a tow or its
floater now responds as the same cable held in the moving water: a pipe in oscillating flow gives
peak A/D 0.819 moving against 0.819 held (0.711 against 0.850 without the frame), and a towed taut
cable 0.7905 against 0.7906 held in the current.

The filtered velocity is state (viv_ms rows 7-9) in recovery, snapshot, checkpoint and the System
pack. A VIV FRAME row needs LINE TYPES Cl > 0 or a WAKE OSCILLATOR row. Runs without the row are
bit-identical to dev. The frame check gates the RMS at 1 % and the peak, a single extreme that
ranges 0.811-0.819 over builds, kernels and platforms, at 2 %.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
Validation case L3-D: cross-flow and in-line VIV of the Delta Flume riser (Chaplin et al. 2005a,
2005b) in the nine blind-trial stepped currents, U 0.16-0.95 m/s over the lower 5.94 m of a 13.12 m
pinned riser with the top spring in series. The VIV set is the synchronisation model with nothing
fitted to these data: Cl 0.8 in f0 0.18 +/- 0.08, in-line Cx 0.85 and alpha 0.15, CD 1.2 and
Chaplin's in-line drag amplification. Measurements are the experimenters' processed results from
the blind-trial presentation, read from the vector graphics through the tick labels.

Dominant mode right in 5 of 9 cases and within one in all; mean in-line displacement within 17 %
in 7 of 9; CF amplitude -44 % to +4 % of the summary maxima. dt/2 and 128 elements move the std and
mean metrics by at most 2.6 % except the beating case 5 and the two-response case 9, both
documented. The mode projection uses a row-wise trapezoid so NumPy 1.x runs it.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…ges (#168)

A finite-EI cable could fail its first dynamic step at a small time
step: Newton stagnated just above the tolerance, and the step
recovery's smaller steps only raised it. Found on a 13 m riser (64
elements) at rest in a 0.95 m/s current taking the VIV lift at dtM
1.25e-4 s: the residual stopped at 1.06 times the tolerance, and each
4-fold smaller substep raised the stagnant residual about 16-fold
(3.8e-8, 6.2e-7, 8.6e-6, 1.3e-4, 2.6e-3, 4.7e-2 N), to 3.4e4 times the
tolerance at 1024 substeps.

The 16-fold growth per 4-fold step is the inertia's 1/dt^2, not the
1/dt of a velocity-coupled force (drag, VIV lift): the inertia takes
the acceleration (q - q_n)/(beta dt^2) - ... from coordinates stored
to their last bits, so its residual cannot be settled below about eps
sum_j |M_ij| |q_j| / (beta dt^2). The rows that stalled were the axial
rows of nodes 8.5 to 12 m deep, not the cross-flow rows the lift
loads: the Newton correction they needed, a few 1e-16 m, is below the
1.8e-15 m spacing of those coordinates. Without the lift the step
starts at the static equilibrium and converges at its predictor; with
it the line near rest sets no inertial force scale, and tol times the
force scale (here its axial round-off floor) lies below the inertia's
floor. A floor of the force scale built from that bound, where it
binds, would accept residuals up to 16 times the bound, some 60 times
the floor Newton reaches (the bound sits about 4 times above it:
1.7e-7 against 3.8e-8 N on the riser, 4.7e-3 against 1.2e-3 N in the
test below). The fix is therefore a stall test, as on the EI = 0
route: when no step length of a fresh Newton correction reduces the
residual, the residual is below 1e-4 of the force scale, and the
correction moves no free coordinate by more than 16 eps times its
scale (the largest coordinate magnitude or element length; 1 or the
largest tangent or twist magnitude for those DOFs), the iterate is
converged.

The riser now runs at dtM 1.25e-4 s, and over 2 s its node positions
and fairlead tension continue the time-step sequence of dtM 5e-4 and
2.5e-4 s (differences 1.3 mm and 0.55 mm, 5.5 N and 2.9 N). A step
that meets the tolerance is unchanged: every example and test-data
deck gives bit-identical outputs on the old and new driver (229 files
over 118 decks), and through the C API five coupled steps of every
deck give bit-identical line arrays (1685 arrays over 70 decks), and
the full test suite (424 tests; 426 after the rebase onto 6c469fe) passes on the default and the Prescott
OpenBLAS kernel, where the 66 example decks also run with no fallback
note.

Test: test_hermite_viv round-off floor, the test's 100 m line hung
vertically from 50 to 150 m depth, from rest with the lift at dt
2.5e-5 s. Before, the first step stagnated at 2.4 times the tolerance
through its 50-iteration budget and failed; now each step stops on the
floor after two iterations, and the cross-flow displacement at 0.02 s
matches the same line's at dt 1e-3 s within 1 % (to 5 digits).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…ts reversed (#169)

A finite-EI line with little numerical damping could fail its march a
few steps in. Found on the lock-in line of the long-run VIV checks (a
taut 20 m line, 10 elements, EA 5e7 N, 3.8 kN, pinned in a 0.3 m/s
cross current with the lift) at rho_inf 1 and 50 steps per VIV cycle
(dt 0.037 s): step 2 converged in 3 quadratic iterations to a state with
every nodal tangent reversed (m_x -1.13 against +1.00015), 2.4e8 J of
strain energy against 3.5 J, and step 6 stagnated at 1e-3 through its
50 iterations. The same happened at rho_inf 0.8 and 0.9 (failing at
steps 10 and 7), at 25 steps per cycle (steps 5, 13 and 4 at rho_inf
0.8, 0.9 and 1; at 0.5 the run went on through reversed roots to a
mid-span A/D of 5.7), and without the lift (step 38). It is not the
lift, the force cache or the cross-step factor reuse (the reuse is off
under VIV, and the step fails the same with it disabled), and not an
instability of the time integration: the energy was steady up to the
jump.

The cause is the Newton predictor. The line's seed has unit tangents
while its elements are stretched by T/EA = 7.6e-5, so the tangents'
stretch, the line's stiffest motion (EA against their small consistent
mass, omega ~ 8e3 rad/s, omega dt ~ 300), starts 7.6e-5 off its
balance. At rho_inf 1 nothing damps it: it rings between 1 and
1 + 2 T/EA with a committed acceleration of 4.9e3 1/s^2 on the
tangents, and the constant-acceleration and Newmark guesses
(dt^2/2 a_n = 3.4 and dt^2/4 a_n = 1.7 of the tangent's length) carried
every tangent across zero, next to the root with each element folded
into a Z along its chord. With rho_inf 0.5 the same mode carries a
velocity overshoot (dt v_n = 1.7 at 25 steps per cycle). Seeded at its
stretch (tangents 1 + T/EA) the line runs at every rho_inf and step.

Fix: where either guess would change the length of a free nodal tangent
by more than a quarter (a strain change of 0.25 in one step, which no
step resolves), that tangent starts at its committed value, from which
the stiff mode's true increment is a Newton step away. The test is on
the length alone, since a guess that turns a tangent fast is often
right: the line-control test's seabed cable turns two tangents 19
degrees in its first 0.01 s step and does not converge from held
tangents. Holding only the length and keeping the turn also converged,
but took 132 steps over 12 iterations against 31 on the 25-step runs.
And a step that ends with a nodal tangent turned by 90 degrees or more
is now a failed attempt with its own message (the next attempt or the
step recovery re-solves it) instead of a converged step: with the
predictor guard disabled, the line stops at step 2 with that message
rather than marching on through reversed roots.
A node with some tangent components fixed (a planar model fixes the
out-of-plane one) is guarded on its free components: the length change
is that of their increment, and only they are held.

The line now runs at rho_inf 0.5 to 1 and 25 and 50 steps per cycle.
Over 20 cycles its largest mid-span A/D is that of the line seeded at
its stretch within 0.75 % and its largest strain energy within 1.2
times; at 50 steps per cycle and rho_inf 1 it runs the 400 cycles of
the long-run check (largest A/D 0.568 against 0.567 seeded at the
stretch; energy mismatch 7.3e-4 of the VIV work, 8.3e-5 at 100 steps
per cycle). The guard changes only a step that would stretch a tangent
by a quarter: every example and test-data deck gives bit-identical
outputs on the old and new driver (134 decks, the
console logs equal up to their progress clock), the full test suite
passes on the default and the Prescott OpenBLAS kernel (427 tests), and
the 66 example decks run on both kernels with no fallback note.

Test: test_hermite_tangent_predictor, the taut line seeded with unit
tangents against the same line seeded at its stretch, at rho_inf 0.5,
0.8, 0.9 and 1 and 25 and 50 steps per cycle over 20 cycles, and
without the lift at rho_inf 1, also kept in its plane (r_z and m_z
fixed at every node). Before, 7 of the 9 runs failed and one
ran through reversed roots (9 assertions), and the planar run stopped
at step 2 while the guard skipped partly constrained nodes; now every
step converges
(at most 5.9 iterations per step on average, 23 at most), the strain
energy stays within 1.5 times the reference's and the largest A/D
within 2 %.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The bending damping Bb is stiffness-proportional, zeta = omega Bb/(2 EI),
so a ratio set at one frequency cannot hold across a VIV band. The LINE
TYPES EI column now takes EI|-zeta|f1|f2: beside Bb = zeta EI/(pi (f1 +
f2)) it adds a mass-proportional transverse damping alpha = 4 pi zeta f1
f2/(f1 + f2) [1/s], so zeta(omega) = alpha/(2 omega) + omega Bb/(2 EI)
equals zeta at f1 and f2 for a mode whose stiffness is all bending.

The new force is alpha m (I - t t^T) v, alpha times the line's own mass
on the velocity normal to the tangent: alpha M v in the two transverse
directions of a straight span, zero along the axis (axial damping stays
with BA). The velocity is absolute, as in the usual Rayleigh model, so
rigid transverse and pendulum motion is damped too, increasingly below
f1; the theory page says so and why the filtered-velocity frame of the
VIV FRAME section is not used (it would add a filter state, lag the
damping and break the two-frequency ratio). No state is added. Force and
both Jacobians are exact; the term joins the structural damping load, so
the end forces, the node recovery, DampPow, the consistent initial
acceleration and the OpenFAST linearisation (central differences of the
same equations) carry it. The API setter is
CD_HermiteCable_Dyn_Set_Mass_Damping (and CD_HFMF_Set_Mass_Damping).
EI = 0 lines refuse the form, as the other damping forms. Both deck
readers refuse a positive ratio, a non-positive frequency and a fifth
part by name; the Python builder writes and reads the form.

Tests: kernel Jacobians against central differences (3e-10), the force
against an independent dissipation function, alpha M v on a straight
element and no damping of axial sliding; free decays of modes 1, 2 and
3 of a pinned cable for alpha alone, Bb alone and both against the
Rayleigh ratio (largest error 0.05 %), and the two-frequency pair at
the target at f1 and f3 (0.02 %); the energy ledger with alpha and Bb
closes to 2.8e-4 of the dissipated work; the deck route and refusals.
Every example and test-data deck (134) writes bit-identical outputs.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…tep (#171)

A taut line locked in by the synchronisation model changed its
lock-in frequency by 0.74 %, 0.41 % and 0.27 % per halving of the
step from 25 to 200 steps per cycle (the long-run VIV checks), worse
than first order, while the same phase update under prescribed
motion was second order. The phase sampled the committed
generalised-alpha acceleration a_n for its cross-flow phase (and the
wake oscillator for its forcing). That acceleration is the line's at
t_n + (alpha_m - alpha_f) dt only, a first-order error at rho_inf <
1, and at rho_inf = 1 it carries an undamped alternating mode of the
integrator: on the 20 m line the mid-span a_n alternated by +-11 m/s^2
against a physical acceleration of about 1.3 m/s^2. The lock-in at
rho_inf = 1 therefore converged to a frequency 3.4 % low (3.940 at
3200 steps per cycle against 4.0797 rad/s), and at rho_inf 0.8 and
0.4 it was first order (0.06/n and 0.25/n at n steps per response
period). Prescribed motion imposes the exact acceleration, which is
why it did not show there.

Candidates switched off one at a time on that line, 25 to 3200 steps
per cycle: a second-order running mean square (exact exponential with
trapezoidal samples) changed nothing; a fixed normalisation in place
of the RMS left it first order; a tighter Newton tolerance changed
nothing; the lift at a fixed shedding rate (no phase feedback) was
second order (ratios 3.8 to 4.0); the acceleration recomputed from
the equations of motion after every step moved the limit to 4.0797
rad/s; and the phase on the step's mid-point motion with the
acceleration from the converged velocities restored second order
(ratios 3.6 to 5.0, then 4.01 to 4.04 from 400 steps on) at rho_inf 1
and 0.8 alike. No amplitude table, flow frame or in-line VIV is
involved in that case.

The phase now advances by the implicit mid-point rule, one
fixed-point pass, on the node's cross-flow motion at the step's
mid-point: the mean position, the velocity (q_n+1 - q_n)/dt and the
acceleration (v_n+1 - v_n)/dt, the step means of the velocity and of
its derivative (second order at every rho_inf, free of the committed
acceleration's lag and alternating mode, and keeping the phase of a
harmonic motion exactly). The step's lift takes the phase predicted
with the rate of the previous step's mid-point, a new row of the VIV
motion statistics (two with in-line VIV), so the step recovery,
sub-step slots, snapshots and checkpoints carry it and a restart stays
bit-identical; a newly installed VIV model seeds it from the present
state. The running mean squares advance by the mid-point rule
(1 - exp(-dt/T_w) on the mid-point sample), the in-line phase in the
same way slaved to the mid-point cross-flow rate, the flow frame's and
the wake filter's filtered velocities exactly for the velocity linear
over the step, and the wake oscillator under the mid-point
acceleration (held over the step, its integral exact) at the mean of
the start and end wake frequencies, that acceleration being its
predictor (row 2 of its statistics). The time discretisation of the
VIV phase therefore changes on purpose; this is not a round-off
change.

Measured on the same line from 25 to 3200 steps per cycle (EI = 0
line and rho_inf 0.8 unless noted), the ratio of successive changes
per halving of the step, before and after:

  frequency, rho_inf 1:          limit 3.4 % low;  3.6-4.9
  frequency, rho_inf 0.8 / 0.4:  2.0-3.4 / 1.5-3.2;  3.6-4.8 / 3.8-4.5
  mid-span A/D, in-line VIV:     2.2-2.3;  2.6-4.3
  frequency, flow frame:         1.6-2.0;  3.1-4.5
  frequency, amplitude table:    1.9-2.3;  3.0-4.5
  wake oscillator, frequency:    2.0-2.5;  3.8-4.0
  finite-EI cable, frequency:    1.7-2.3;  3.1-4.1

The Richardson limits at rho_inf 1, 0.8 and 0.4 agree to 1e-9 (4.0797304
rad/s), and the errors are about 2.2/n^2 (rho_inf 1, 0.8) and 3.4/n^2
(0.4) in frequency and 1.6/n^2 and 3.4/n^2 in mid-span A/D at n steps
per response period, 3.5/n^2 on the finite-EI cable. With the cable
seeded at its static stretch, rho_inf 1 reaches the rho_inf 0.8 limit;
from unit tangents the undamped tangent-stretch mode rings and moves
it at fine steps (the seed, not the VIV update). Under prescribed
motion the phase error of the long-run check after 400 cycles at 100
steps per cycle is 0.0167 rad against 0.0334 before, still second
order (3.99). The energy closure of that check at 100 steps per cycle
is 2.0e-4 of the VIV work against 8.3e-5 before, second order (ratio
4.0): the lift's predicted phase now takes the rate of the previous
step's mid-point, half a step behind the old step-start rate.

The two VIV example decks change at their default dtM 0.02 s (about
170 steps per response period, rhoInf 0.4) and move toward their
results at dtM 0.00125 s. viv_taut_line_crossflow, against dtM
0.00125 s: the mid-span RMS 0.31 % high before, 0.02 % low after; the
lock-in frequency 0.15 % low before, 0.018 % after; the fairlead
tension RMS 0.29 % high before, 0.027 % low after. viv_taut_cable_inline:
the quarter-span in-line RMS 1.56 % low before, 0.34 % after; the
cross-flow frequency 0.10 % high before, 0.022 % after; the largest
fairlead tension 0.70 % high before, 0.23 % low after; the mid-span RMS
0.21 % and 0.20 % low. Over the 90 s the phase moves, so the time series
differ pointwise by up to 12 % (crossflow, fairlead tension) and 36 %
(inline, quarter-span y) of the channel's range. Every other example
and test-data deck (132 of 134) gives bit-identical output files and
console logs. The full test suite (430 tests, on dev with #169 and #170) passes on the default
and the Prescott OpenBLAS kernel, and the 66 example decks run with no
fallback note on both.

Test: viv_step_order, the lock-in frequency of the line second order
from 100 to 400 steps per cycle at rho_inf 1 (ratio 3.99) and 0.4
(3.67) and within 1e-4 of each other (4.3e-5), the wake oscillator's
crossing frequency (3.90) and the finite-EI cable from 200 to 800
steps per cycle (4.39); on the previous code every check fails (ratios
2.44, 2.27, 2.18 and 2.76, the two rho_inf 3.1 % apart). The VIV
state layouts in test_model, test_system, test_viv_coupled,
test_hermite_viv_frame and test_hfmf_mirror carry the predictor rows.

With the mid-point VIV update the predictor check of the lifted
taut line fails one cell, rho_inf 1 at 25 steps per cycle (21 per
response period): from the unit-tangent seed z max/D 0.5903 against
the balanced seed's 0.5641 (+4.6 %) and Es max 1.51 times it. Every
step converges (6.15 iterations on average, at most 21).

At rho_inf 1 nothing damps the unit seed's stretch ringing, and at
that resolution it leaves a near-Nyquist velocity in the trajectory:
the mid-node (v_n+1 - v_n)/dt alternates by 4.8 m/s^2 against a VIV
acceleration of 3.2 m/s^2 (a velocity of about 0.09 m/s, 27 % of the
VIV velocity; the committed a_n alternates by 32 against 17, 19
against 10.5 on the old update), 10 to 20 times less at 50 steps per
cycle. The response then depends on how the VIV update treats that
velocity: four equally consistent variants (the last two mid-point
accelerations averaged, a three-point second-order Nyquist-free
acceleration, an averaged predictor rate, both of the latter) give
z max/D 0.623, 0.579, 0.589 and 0.578, +0.8 % to +10 % from the
converged 0.5639 with the old update's +0.8 % among them. That value
is the fine-step one (200 to 800 steps per cycle), the same at
rho_inf 0.8 with either seed and at rho_inf 1 with the balanced seed.
The new update reaches it from 50 steps per cycle with either seed
(unit seed 0.5632 at 50, 0.5641 at 100, 0.5639 from 200 on), while
the old update settled at rho_inf 1 0.23 % high from the unit seed and
0.1 % low from the balanced one. The result is deterministic (Newton
tolerance 1e-6 to 1e-10 and a 1e-7 seed tilt change it by under
0.03 %).

So that cell measures the VIV update's treatment of an unresolved
spurious velocity, not the predictor. It keeps G1 (every step
converges) and G4 (iterations) and skips G2 and G3, and a new gate G7
holds the convergence claim: the unit seed at rho_inf 1 and 50 steps
per cycle within 0.5 % of the converged z max/D 0.5639 (measured
0.5632). The other cells keep G2 and G3 (measured at most 1.07 times
and 0.09 %). doc/theory.rst and the CHANGELOG note that at rho_inf 1
from an unbalanced seed the VIV response is resolution-sensitive below
about 50 steps per cycle: keep rhoInf at or below 0.95 or seed the
tangents at the stretch.

The nine cases and their sensitivity runs (half the time step, 128
elements, 80 s for cases 5, 6, 8 and 9, the 38.1 kN/m top spring) rerun
on the second-order VIV phase update. The counts are unchanged: the
dominant mode right in 5 of 9 and within one in all, the mean in-line
displacement within 17 % in 7 of 9 (-25 % to +34 %). The cross-flow
amplitude is -45 % to +4 % against the maxima (-40 % to +18 % against
the envelopes; within the 12 % measurement scatter in cases 2, 4, 6 and
7), the cross-flow curvature -39 % to +26 %, the in-line curvature
-53 % to +111 %.

Half the time step now changes the peak amplitude of cases 1-4, 7 and 8
by at most 2.7 % (4.5 % before) and their in-line curvature by at most
1.2 % (10 % before). Case 5 keeps its peak amplitude within 0.6 % (it
moved by up to 20 % before) while its in-line curvature moves by up to
25 %. The peak amplitude of case 6 now depends on the step and the mesh:
-11 % at half the time step, kept over an 80 s run, and -13 % with 128
elements, its other metrics within 2 % (in-line curvature 12 %). Case 9
still settles in mode 7 at half the time step. README, provenance.json
(new 80 s runs of cases 6 and 8 at both steps), summary.json,
response_comparison.csv, VALIDATION.md and the changelog entry follow.

The mid-point VIV phase update changes every coupled-VIV result, so
the validation case L3-W and the numbers the documents quote from
coupled VIV runs are recomputed with it, each as its text describes
(deck, options, window, output flag).

L3-W (validation/experiments/wang_scr), rerun as its README says, and
its two sensitivity runs: the largest r.m.s. strain over the measured
largest 1.0729 -> 1.0733, the station mean 1.0362 -> 1.0361, the
r.m.s. log error 0.339 -> 0.338, the dominant frequency 0.992 Hz
(unchanged), 7 r.m.s. peaks along the span (unchanged); 400 elements
1.0753 -> 1.0759 and half the time step 1.0742 -> 1.0747 (0.24 % and
0.13 % from the base run). Every number and sentence of the README,
VALIDATION.md and the CHANGELOG still holds at its printed precision;
summary.json, the comparison CSV and provenance.json (commit, date,
sensitivity) are updated.

Numbers changed:

- doc/outputs.rst, examples/viv_taut_cable_inline.dat: the VIV loads'
  power over 60-90 s 8.52 -> 8.45 W, the drag (and damping) 8.56 ->
  8.48 W (balance within 0.5 %, unchanged).
- doc/python_postprocessing.rst: the travelling-wave index 0.05 -> 0.04.
- doc/theory.rst: the EI = 0 wake-oscillator line locks in at 0.283 Hz
  (was "the same 0.282 Hz"); the amplitude estimate within 0.04 % (0.1 %
  on the EI = 0 line; was 0.03 %); Thorsen's damping A/D 1.02 -> 1.01;
  the moving/held pipe table (peak 0.819/0.819 -> 0.811/0.818, RMS
  0.446/0.447 -> 0.446/0.446; without the row 0.711/0.850 ->
  0.700/0.851 and 0.349/0.450 -> 0.345/0.450; Cl alone 0.713/0.714 ->
  0.713/0.712 and 0.372/0.371 -> 0.372/0.370; without 0.711/0.736 ->
  0.700/0.733 and 0.349/0.375 -> 0.345/0.374), the RMS agreement
  within 0.2 % -> 0.6 % (0.02 % with the table); the towed cable 0.7905
  / 0.7906 -> 0.7902 / 0.7903, 0.665 -> 0.663, 0.6764 / 0.6764 ->
  0.6758 / 0.6758, Tf on the held cable at most 0.23 % -> 0.24 %.
- doc/driver_format.md: towed within 0.01 % -> 0.02 %, at most 0.23 %
  -> 0.24 %.
- VALIDATION.md: L2-VIV zero seed 0.701 -> 0.699 against 0.699;
  L2-VIV-EI 0.662/0.421 against 0.663/0.421 -> 0.661/0.420 against
  0.661/0.420; the coupled VIV row 0.6645 / 0.6646 -> 0.6631 / 0.6625
  (standalone 0.665 -> 0.663). The L2-VIV band against MoorDyn-F,
  rerun over 150-300 s at U 0.06-0.42 m/s: mode 1 at 0.09-0.18 m/s
  within 0.2 %, mode 3 at 0.39-0.42 m/s within 1.4 %, and at 0.36 m/s
  MoorDyn-F settles in mode 2 and CableDyn in mode 3 (the earlier phase
  update gives the same branches, its A/D within 0.5 % of the new one at
  every speed, so the discrepancy predates this change); the row said
  "mode 3 at U >= 0.36 m/s within 1.2 %".
- CHANGELOG.md: the wake entry's EI = 0 frequency, the amplitude
  estimate 0.03 % -> 0.04 %, the frame entry's pipe and tow numbers as
  in the theory, and the #169 entry's largest A/D within 0.75 % ->
  0.1 % except at rho_inf 1 and 25 steps per cycle.
- tests/test_viv_frame_deck.f90 and tests/test_hermite_viv_frame.f90
  (header comments only): the measured values of their gates.

Every other quoted coupled-VIV number still holds at its precision
(the example decks' A/D 0.66, 0.82, 0.19 and 0.03, their frequencies,
the in-line and higher-harmonic gates, the wake, diameter and Thorsen
gates, the riser case). The Delta Flume case was rerun before; the
NDP case is on another branch.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The bare riser of the Norwegian Deepwater Programme tests (Trim et al.
2005, J. Fluids Struct. 21(3), 335-361) towed horizontally at its 22
uniform-flow speeds, 0.3-2.4 m/s, responding at up to 12.5 Hz across
the flow. It tests high-mode cross-flow VIV against four coefficient
sets on one model, one of them blind.

validation/experiments/ndp_riser/ndp_riser_viv.py writes a deck per set
and speed: 38 m between universal joints, D 27 mm, EI 598.8 N m^2, EA
8.19e6 N, 0.933 kg/m flooded (Fontaine et al. 2011, Table 1), fresh
water (the mass ratio 1.62 implies 1006 kg/m^3), the clump-weight rig
as an axial spring in series, EA_eff = L/(1/k_ax + L/EA), with T0 4 kN
in still water and k_ax 3.5 kN/m derived from the reported 4-6 kN range,
not tuned. 200 elements, dt 1 ms, 25 s, statistics after 10 s, CurvX
and CurvY at every node. It runs the sets in parallel (--sets,
--speeds; the default 25 runs take about two minutes, the full
110-run sweep of the committed record about twelve on 24 cores) and
compares the maximum r.m.s. cross-flow strain (fundamental, band-passed
about the dominant frequency as Thorsen et al. 2016 filter the
measurement), the dominant frequency f D/U and the spatial-mean sigma/D.
CurvX is the cross-flow curvature (the section axis starts vertical
at End A); the dominant frequency is searched for in f D/U 0.05-0.30,
which keeps a slow swing and the third harmonic out.

Sets, each as its source gives it, checked against the articles:
- b1, the generic set of the Delta Flume case (Cl 0.8, cF 0.18, dF
  0.08, ClX 0.85, alpha 0.15, Chaplin's drag amplification on the
  in-line drag via VIV DAMPING 0 0 1): blind;
- and24b, Andersen et al. (2024b), Table 3, cross-flow average (Cl 1.28,
  cF 0.131, dF 0.066, Cd 1.16), fitted to the sheared-flow tests of the
  same riser;
- uls19, Ulveseter et al. (2019), Table 1 (Cl 0.85, cF 0.144, dF 0.064,
  ClX 0.75, alpha 0.15, Chh 1.5), fitted to these tests;
- th16, Thorsen et al. (2016), the C_v curve of their Fig. 7 and their
  cross-flow damping (VIV DAMPING 0.31 0.89 0), fitted to these tests;
  th16c keeps Morison's linear cross-flow damping (a variant).

Measurements: the maximum r.m.s. cross-flow strain of each test with
Thorsen et al.'s own simulation, read from the vector graphics of their
Fig. 8 (extract_thorsen2016.py, which also reads the Fig. 7 curve), and
sigma/D and the dominant frequencies of Trim et al.'s Figs. 16 and 18,
read by eye on enlarged guided crops of the raster figures
(digitise_trim2005.py draws them; +-0.01 in sigma/D, +-0.3 Hz).

Result on the second-order VIV phase (22 speeds; r.m.s. log error of the
strain, f D/U against the measured 0.143, sigma/D CF against 0.436):
- uls19 0.178, 0.145, 0.461: the only set that matches strain,
  frequency and amplitude together (sigma/D within 6 % CF and 3 % IL);
  Thorsen et al.'s own fitted simulation scores 0.187;
- and24b 0.186, 0.134, 0.687 (sigma/D 57 % high: a lower mode at a
  larger amplitude gives the same curvature);
- b1 0.277, 0.192, 0.429 (blind; frequency a third high);
- th16 0.363, 0.193; th16c 0.365, 0.153: the damping form sets the
  frequency.

At 1.0 m/s half the time step and 400 elements move the strain of four
sets by at most 0.8 % (uls19 +1.9 % and +18.5 %, at 5.60 Hz instead of
5.80 Hz). At 2.4 m/s the response is not converged speed by speed: the
strain moves by -14 % to +12 % with half the time step and by +0.6 % to
+20 % with 400 elements, and a 55 s run moves th16c by +12.5 %, so a
longer window does not remove the spread (the mix of neighbouring high
modes, of the size of the speed-to-speed scatter). The 22-speed
statistics hold: the full sweep at half the time step gives 0.280,
0.187, 0.184, 0.375 and 0.376, f D/U within 0.001 and sigma/D within
1.4 %. Sea water (U >= 0.5 m/s): 0.299, 0.181, 0.154, 0.434, 0.353.

Added to VALIDATION.md as L3-N, to the experiments index, the
changelog and the references (Trim et al. 2005, Fontaine et al. 2011,
checked against Crossref); the top-level README now names the six
published experiments.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…r and bending-damping speed-ups (#173)

viv_long_run runs the synchronisation model for 400 cycles. Prescribed
motion: every node of a finite-EI cable moves across a 0.3 m/s flow as a
rigid body, so the shedding phase obeys Adler's equation,
d psi/dt = a - b sin(psi), solved in closed form. Outside the band
(5.5 rad/s) the error grows linearly in time (T/(T/2) 2.014 and 1.988 at
50 and 100 steps per cycle) and is second order (4.01 from 50 to 100
steps): 0.0167 rad after 400 cycles at 100 steps per cycle. Inside the
band the lag holds within 3.0e-5 rad of psi*, the averaging-ripple
floor, and the mean rate is the motion's within 5.8e-12.

Lock-in, finite-EI and EI = 0 (taut 20 m line, rho_inf = 1, no added
mass, 100 and 200 steps per cycle): the load powers close E(t) - E(0) to
2.0e-4 / 1.8e-4 of the VIV work, second order (3.98 / 3.67) and
stationary (T against T/2 1.00). On the mid-point phase update of #171
the lift takes the phase predicted from the previous step's mid-point
rate, half a step behind, so the bound is 3e-4. The halves agree in RMS
A/D within 7.3e-4, VIV power within 1.8e-3 and lock-in frequency within
1.9e-4; the frequency changes by 1.6e-4 / 1.9e-4 from 100 to 200 steps
per cycle. Every tolerance and its measured value is in the test header.
57 s Release; label slow, TIMEOUT 1200.

run_study with jobs > 1 runs every case with OMP_NUM_THREADS=1 unless the
environment sets it, and study.json records omp_num_threads. On a 24-core
workstation (gfortran Release) 96 finite-EI VIV cases scale 13.3x at 24
jobs and 15.1x at 48; on the 1024-element lazy-wave example 24 jobs take
16.7 s with one thread each, 17.9 s with four and 19.1 s with the OpenMP
default. The Python manual gives the table and the recommended settings.

Speed-ups, outputs bit-identical:
- a row evaluates a line's node powers once for all its power channels
  (VIVPow, DragPow, DampPow; 19 % of a VIV run) on the Hermite, EI = 0
  and system routes; any failure still goes through the
  channel-at-a-time path. viv_taut_cable_inline.dat 2.39 s to 2.14 s;
- the bending damping force no longer computes the second derivatives
  of the curvature, which only the position Jacobian uses (32 % of the
  Delta Flume case 9 run): 73.2 s to 58.6 s.
load_power_deck P6 checks that a power channel read alone equals the
same channel among its line's other power channels, on all three routes.

Checked: 431/431 tests on the default and the Prescott OpenBLAS kernel;
the 134 example and test-data decks bit-identical to the reference
build; the NDP validation runs of #172 reproduce their committed rows.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…AME scale separation (#174)

The theory page said that VIV DAMPING 0 0 1 with a CdAmp drag factor
amplifies the mean in-line drag alone. The force is
1/2 rho D Cdn A_d |w| w_x on the whole in-line component, so the in-line
damping rises with the factor too (by 2.3-2.8 at A/D 0.6-0.8 with
Chaplin et al. 2005, Eq. 4). The code is unchanged; the theory page and
the input reference now say so.

The VIV FRAME pages said to take Tf several VIV periods long and short
against the motion. They now give the bound: keeping about 95 % of the
VIV velocity (omega Tf >= 3) while leaving at most 10 % of the motion's
velocity in the relative velocity (omega_m Tf <= 0.1) needs a motion
period at least 30 VIV periods long. For a motion only a few VIV periods
long no Tf separates the two, and the response depends on Tf.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The OD 48 mm research cable of the FLOTANT tests at MARIN (de Wilde,
Heijnsdijk & de Ridder 2021, Deliverable D5.5, MARIN report
30995-53-DWB): 6.4 m hanging from the carriage with a 200 kg clump,
towed at the 40 speeds of 0.025-1.0 m/s. It tests cross-flow and in-line
VIV of a real cable section against three published coefficient sets,
none fitted to these tests, on a structure identified only from the
specimen's forced-oscillation RAOs.

validation/experiments/flotant/flotant_viv.py writes a deck per case,
runs them in parallel (reused when deck and executable are unchanged)
and compares:
- tow: A/D_rms = sigma/OD at 1/2L, 3/4L and L, cross-flow and in line,
  at the same test speed, the model's displacement processed as the
  report's accelerometer motions (the g theta/omega^2 tilt term, the
  3.5 rad/s high-pass); Ur with f1 1.77 Hz as Appendix 3; 60 s runs,
  statistics after 20 s, dt 2 ms;
- rao: the in-air (5 and 10 mm) and in-water (5 mm) RAOs of
  Appendices 1 and 2, the structural check.

The line follows the report's FEM tables node by node, in fresh water
(Table 2-2 implies 1000 kg/m^3), with the clump or counterweight as a
20 kN/m axial spring in series. What the tests leave open is identified
from the RAOs: cable EI 1650 N m^2 and bending damping 130 N m^2 s from
the in-air resonance (the tables' EI 3000 needs MARIN's empirical factor
1.10), and in water a link chain bending with EI 100 N m^2 and the clump
held laterally by 32.5 kN/m and 20 N s/m, which give the measured split
of the first in-water mode (r.m.s. log error 0.46 in surge against 1.19
for MARIN's fixed clump and 1.17 for a free one).

Sets, checked against the articles:
- uls19, Ulveseter et al. (2019), Table 1 (Cl 0.85, cF 0.144, dF 0.064,
  ClX 0.75, alpha 0.15, Chh 1.5), fitted to the NDP riser: blind here;
- b1, the generic set of the Delta Flume case (Cl 0.8, cF 0.18, dF 0.08,
  ClX 0.85, alpha 0.15, Chaplin's drag amplification, VIV DAMPING
  0 0 1): blind;
- wake, Facchinetti et al. (2004) (CL0 0.3, St 0.2, eps 0.3, A 12):
  blind.

Result on the second-order VIV phase update (#171) (cross-flow r.m.s.
log error over the three stations, Ur >= 3):
- uls19 0.731: the first peak at 1/2L is 0.755 at Ur 5.30 against 0.776
  at 5.58, but the rise starts at a lower speed (0.42 against 0.18 at
  Ur 3.53), the lock-in releases at Ur 7.06 against 5.88 (1.2 Ur late)
  and the upper branch stays at 0.64-0.71 above Ur 9 where the test
  falls to 0.18-0.44;
- b1 0.594: 38 % low at the peak (0.483), releases at the measured Ur;
- wake 1.208: 62 % low (0.296).
With MARIN's fixed clump or a free clump (End B Free) uls19 scores
0.721 and 0.692 but releases at Ur 8.83 and 9.12 and misses the RAOs.

extract_d55.py reads the data from the report's PDF (FEM tables from the
text, the RAO charts from the vector graphics, Appendix 3 from the raster
images calibrated on the grid lines; every point maps to a tested speed
within 0.0085 m/s). The caveats of the data are stated: the Ur basis
(1.77 Hz against the 1.3 Hz first in-water mode of Sec. 5.3), the bottom
tension (1436 N in Table 6-1 against about 1700 N in the FEM table) and
the towing direction alternating with speed.

Half the time step, two elements per table element and 120 s runs change
no A/D_rms by more than 0.006 at five speeds; the RAO scores move by at
most 0.001. The default command run into a new work directory reproduces
the committed record byte for byte.

Added to VALIDATION.md as L3-F, to the experiments index, the changelog
and the references (de Wilde et al. 2021, from the report's PDF; no DOI);
the top-level README now names the seven published experiments.

flotant_viv.py and ndp_riser_viv.py now remove a folder's result, executable
stamp and output before writing new inputs, so a rerun that fails or is
interrupted cannot leave the previous model's result to be reused; a tow
selection entirely below Ur 3 prints n/a for its scores instead of failing
before the comparison files are written.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
gfortran 16.2 (-Wall -Wextra) reported seven warnings on dev, none a
defect. Each site now gives the same result for every input, NaN, signed
zeros and infinities included:

- CD_VIV_Check_Amp_Table compared a row's A/D with amp(1, i - 1, e)
  behind IF (i > 1), which -Wdo-subscript cannot see through ("out of
  bounds (0 < 1)"). It now compares with the previous row's value, held
  from -1 below every valid first row, in the same order of checks, so
  every table gets the same message.
- The half-cycle start of the amplitude estimate, .NOT. (p /= 0), is
  ABS(p) <= 0 (true for +-0 only, false for NaN, as before).
- The three immersed-fraction Jacobian terms, dwf /= 0, are
  .NOT. (ABS(dwf) <= 0) (true for NaN, as before).
- The attachment-weight fixed point, ALL(f_att_now == f_att), is
  ALL(f_att_now <= f_att .AND. f_att_now >= f_att), IEEE equality.
- test_model declares its sub-step event counter with the kind of the
  counter it copies (INTEGER(int64)).

All 138 example and test-data decks write byte-identical outputs and the
same messages as the dev build; 431/431 tests pass on the default and
the Prescott OpenBLAS kernel.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…, not its heading (#177)

Past the pitchfork of a current along the vertical plane of a finite-EI
line (the grounded run turned sideways on the seabed), the static solve
adds a cross-flow of 1 % of the horizontal current to leave the plane,
then polishes at the exact current. The cross-flow was the left normal of
the line's local frame, and that frame is the identity for any chord
without a y component. A chord from the anchor along -x therefore took
the cross-flow at +y (to its right), a chord a round-off away from -x at
-y (to its left), and a deck rotated by 180 degrees about the vertical
left its plane to the same global side as the original.

The cross-flow now points to the left of the horizontal chord from the
anchor (End B) to the fairlead (End A) whatever its heading, +y for a
vertical chord as before, so the rotated deck reaches the rotated
equilibrium. Of the 138 example and test-data decks one changes,
tests/data/fei_hairpin_current.dat (chord along -x): it reaches the
mirror image of its previous equilibrium, x, z, tension and curvature
identical to the bit, y negated, the same end forces with Fy reversed.

The statics matrix gains a side check (the node farthest from y = 0 on
the recorded side) and mirror twins (the mirror image of the parent to
1e-6): CUR_1013 and its 180-degree rotation leave to opposite sides (the
rotation failed before), and a cross current of 2 % of the deck's own,
toward +y or -y, picks the side, the two mirror images of each other.
The theory page states the rule and that a deck can pick the side with a
small cross component of its own current.

431/431 tests pass on the default and the Prescott OpenBLAS kernel.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…Wh) (#178)

The truncated steel catenary riser of Wang et al. (2014) in both cases
Thorsen, Saevik & Larsen (2017) use to verify the synchronisation model,
each with their published coefficient set, nothing fitted:

- uniform current, 0.2 m/s (L3-W, Sec. 5.1): f at 8.25 m 0.992 Hz
  against the measured 1.00 Hz (their model 1.02 Hz); largest r.m.s.
  strain 1.073 times the measured largest (theirs 0.996), station mean
  1.036 (theirs 0.970), r.m.s. log error 0.338 (theirs 0.360). The
  numbers of the earlier record are unchanged.
- heave-induced VIV in still water (L3-Wh, Sec. 5.2): the upper end
  follows two sinusoids of 5.96 s fitted to the measured top motion
  (their Fig. 19: x 0.0485 m, z 0.1305 m, x lagging z by 83 deg); the
  strain is band-passed from 4 times the motion frequency to 50 Hz as
  Wang et al. process the measurement. Without a VIV FRAME row, the
  published model's phase input: f 1.019 Hz (measured close to 1 Hz),
  largest strain 1.208 times the measured largest (theirs 1.056),
  station mean 1.334 (theirs 1.087), lower / upper half of the span
  1.71 / 0.93 (theirs 1.44 / 0.71).

The heave period is about 6 VIV periods, below the scale separation
the VIV FRAME filter needs (about 30), and the heave result depends on
Tf. With Tf 0.35-0.5 s the response stays at 1.01-1.02 Hz (largest
strain 1.07-1.13); with 0.75-1 s it moves to 0.88 Hz (0.96-0.97,
upper half 0.60-0.63); at Tf 0.5 s half the time step moves it from
the first branch to the second. It is reported as a sensitivity.

Convergence on this build: half the time step and 400 elements change
the largest strain by +0.13 % and +0.24 % in the current, -1.0 % and
-0.7 % in heave (station mean -2.9 % and -2.3 %), frequencies
unchanged.

validation/experiments/wang_scr/wang_scr_viv.py writes the 11 decks of
the record (primary, convergence and Tf runs), runs them in parallel,
reuses a run whose inputs and executable are unchanged (removing the
result, the executable stamp and the output before new inputs are
written) and compares; the default command run into a new work
directory reproduces response_comparison.csv and summary.json byte for
byte. It replaces wang_scr_uniform_current.py. extract_thorsen2017.py
reads Figs. 17 and 22 (vector) and the Fig. 19 top motion (raster)
from the article's PDF; Thorsen et al.'s simulated lines peak at
99.6 % and 105.6 % of the measured largest strain, where their text
gives 99.6 % and 6 % high.

VALIDATION.md, the experiments indexes, the README, the CHANGELOG and
the theory manual's table of published coefficient sets name the new
case.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The tests of Fu, Wang, Baarholm, Wu & Larsen (2014): a 4 m flexible
cylinder (D 24 mm, EI 10.5 N m^2, pretension 500 N) moved across its axis
through still water, A_m 0.68 m (KC 178) at periods 16.5, 10.2 and 8.45 s
(U_R,max 4, 6.5 and 7.9). The relative flow reverses twice per cycle, so
the case tests the VIV models through flow reversal and on the
accelerating and decelerating halves of each half-cycle, where Fu et al.
measure different amplitudes (hysteresis).

validation/experiments/fu_oscillatory/fu_oscillatory_viv.py writes a deck
per run, runs them in parallel (reused when deck and executable are
unchanged) and compares the cross-flow A/D at mid-span (station CF4)
against the instantaneous U_R on the two paths with the measured paths of
Figs. 15-17, and the share of each half-cycle above sqrt(2)/2 of its
peak with the article's lock-in shares (about 17, 34 and 50 %).
extract_fu2014.py reads the paths from the figures, raster images
calibrated on the plot frame and traced by colour; the traced gaps between
the paths (0.164 at U_R 3.5, 0.222 at 4) match the figures' annotations
0.15D and 0.2D.

Three published sets, none fitted to these tests (uls19, b1, wake, as in
the FLOTANT and NDP cases), each with a VIV FRAME row: the cylinder moves
through still water, where without it the wake oscillator never sheds.
The primary runs take Tf 0.2 s; the motion period is 23-44 periods of f1,
around the 30 the frame's two conditions need, so the record reports the
runs without the row and with Tf 0.1 and 0.3 s.

Results (A/D r.m.s. error over both paths; largest A/D):
- every run converges through the flow reversals;
- U_R,max 4: the wake oscillator gives the measured largest A/D (0.280
  against 0.286), the generic set is 65 % high, uls19 barely responds;
- U_R,max 6.5 and 7.9: every set holds the lock-in while the flow slows,
  the mean gap between the paths 2-5 times the measured one and the
  largest A/D 1.3-2.3 times the measured; the wake oscillator comes
  closest (0.177 and 0.132).
Without the row the generic set scores 0.034 at U_R,max 4, the best of
any run, but keeps the deceleration hold at 6.5 and 7.9.

Half the time step and 80 elements change the largest A/D by at most
0.1 % (generic set, wake oscillator) and 4.1 % (uls19). The default
command run into a new work directory reproduces response_comparison.csv
and summary.json byte for byte.

Added to VALIDATION.md as L3-O, to the experiments indexes, the top-level
README, the changelog and the references (Fu et al. 2014, from the
article's first page and footers).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…17) (#179)

An optional VIV STOCHASTIC section, rows TypeName sigma c [Nf [seed]],
makes the centre cF of a line type's synchronisation band the slowly
varying random process of Ulveseter, Thorsen, Saevik & Larsen (2017,
Eq. 6 and Fig. 4):

  F0 = cF + sum_i a sin(p_i + 2 pi f_i theta),  theta = integral U/D dt,

Nf components (default 100) at the mid-points f_i = (i - 1/2) c cF/Nf of
a flat spectrum over (0, c cF], amplitude a = sigma sqrt(2/Nf), so F0
has the mean cF and the standard deviation sigma. Each component
advances at its non-dimensional frequency times U/D, as the shedding
phase does (U the shedding speed of the phase rate). The start phases
are uniform on [0, 2 pi), drawn per node from a minimal-standard stream
that the seed, the node and the line select: the nodes are independent,
as in the article, a run repeats its realisation and two lines of one
type differ. The band half-width dF stays constant.

theta is the one new state: one more row of each node's VIV motion
statistics, so step recovery, snapshots, rollback and checkpoints
(standalone and the OpenFAST mirror) carry it; the phases are rebuilt
from the seed. The step advances theta by dt U/D at the mid-point and
takes F0 at theta + dt U/(2D), the mid-point rule of the phase (#171).
On EI = 0 lines and finite-EI cables, standalone and coupled, with or
without in-line VIV; refused on a type without Cl > 0 or with a WAKE
OSCILLATOR row.

Checks (test_viv_stochastic, test_viv_coupled):
- the process with the article's NDP values (cF 0.144, sigma 0.016,
  c 0.07, 100 components) over theta 0-4e6 has the mean cF to 2e-6, the
  standard deviation sigma to four digits and an excess kurtosis of
  -0.18; the phases repeat per seed and node and differ between them;
- a sigma 0 row is the deck without it, bit for bit, on both cores;
- with sigma 0.02 a run repeats bit for bit, another seed differs, and
  the taut cable stays locked in (A/D within 1 % of the constant centre's
  over 50-80 s on both cores);
- the coupled checkpoint restart continues bit for bit with theta in
  both VIV blocks (one more value per node), a rebuild without the VIV
  state does not, and a rewound step repeats;
- refusals: no Cl, a wake-oscillator type, c out of (0, 1], a
  non-integer Nf, a seed out of range.

The EI = 0 model's explicit copy and move routines and its VIV row
count (CD_Model_VIV_NMS) carry the new fields. Python reader and
builder: the section, its native checks and a round trip. Driver
format, theory (Stochastic band centre), capabilities, changelog and
the reference (from the article's first page).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…181)

The two Debug (-O0) jobs held every pull request for about 2 h 10 min
against 30-50 min for the rest of CI. Their purpose is the -O0 fault
class (an absent OPTIONAL reached through a non-short-circuited
.AND./.OR.), which lives in code the fast tests already reach; most of
their time went to the 18 tests labelled slow (the long lazy-wave,
mooring and statics campaigns, about 6500 of 21900 CPU-s on the Windows
Debug job).

A pull request now runs the Debug tier with -LE slow (an estimated
64 min at four tests at once, in step with the Release jobs). The slow
tests still run on every pull request in Release on both platforms and
on the Prescott kernel, and every push to dev and main runs the full
suite at -O0, so each merge stays hardened without holding the PR.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
… (L3-K) (#182)

Add validation/experiments/chaplin_catenary: the 5.36 m, 56 mm model riser
of Chaplin & King (2018), EI 1.3 N m^2, pinned on the floor of a 2 m flume
and 0.2 m above the surface, at both mass ratios in current of 0.2-1.4 m/s
in both directions, against the measured mean top tension and its standard
deviation (Figs. 10, 13), cross-flow spectral peak frequencies (Figs. 9,
15) and straight-part inclination (Fig. 12), read from the article's raster
figures by extract_chaplin2018.py (frame and tick calibration, circle Hough
markers, error-bar caps, checks on the drawn f d/U = 0.11 and W h lines).

Three published coefficient sets, none fitted to these tests, and a run
without VIV. Every set gives one frequency along the suspended span in
positive current, as measured, at f d/U 0.121 (Ulveseter et al. 2019),
0.138 (generic set) and 0.17 (wake oscillator) against 0.111. The generic
set's drag amplification gives the light riser's tension rise (9.8 %
r.m.s.; 88.8 against 88.6 N to 1.4 m/s) and the heavy riser's inclination
in negative current (5.0 deg r.m.s., implied Cd 2.55 against 2.47). The
heavy riser's still-water tension is 37 % above the measured with the
article's geometry, reported as a caveat; tensions are also compared as
increments. Convergence, structural-damping and floor-friction runs are in
the record; the default command reproduces it byte for byte.

Update VALIDATION.md, the experiment lists, CHANGELOG.md and the
references (ref-chaplin2018).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
#183)

#181 took the slow-labelled tests out of the pull-request Debug tier,
but its own run still held the Debug jobs for about 1 h 45 min: the
remaining 414 tests cost 18,700 CPU-s at -O0 on the Windows runner, and
27 of them (the VIV deck gates, the coupled and threaded C API runs,
the azimuth and graded-mesh statics, among others) take 200-1030 CPU-s
each, about two thirds of the total.

CMakeLists.txt gives those 27 tests the label debug_long (only where the
test exists in the build), and a pull request's Debug tier runs
ctest -LE 'slow|debug_long': 387 of 432 tests, about 25 minutes of tests
at four at once. The skipped tests still run on every pull request in
Release on both platforms and on the Prescott kernel, and the full suite
runs at -O0 on every push to dev and main.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
…#184)

Every Twist<L>N<J> channel of a finite-EI cable re-evaluated the twist
of every element of the line (CD_HermiteCable_Dyn_Element_Twist, with
the twist reference and its Hessian per element), and every Torq<L>N<J>,
CurvX/CurvY and VIV amplitude channel re-evaluated its own line-wide
array.
On the coupled OpenFAST route the channels are evaluated in every
CalcOutput (about two per step) and again for the committed-step row of
the .CD.out, so a deck with a few twist channels spent most of its
CalcOutput time on them.

The channels of one output row now share these values: CD_ChannelRow
(the former load-power row cache) also carries the torsion state, the
element and end twists, the node torques, the section curvature and the
VIV amplitude of a cable. The first channel of a row that needs a value
evaluates it with the same call the channel made before, and the cable's
other channels of the row read it, so every value is bit-identical. A
row is valid for one state only: CD_AGG_EvalChannels (new) clears it at
the start and the end of every row, the standalone routes per row.

- CD_CalcOutput, its committed-step row and lin_eval evaluate the
  channels with CD_AGG_EvalChannels. A supercycled module's row takes
  the channels the same call has just evaluated on the committed state
  (the re-evaluation repeated them bit for bit).
- The standalone mixed route (write_mixed_row), the multibody march and
  the cubic-Hermite writer pass the row through
  CD_Eval_Aggregate_Channel / eval_cable_channel.
- A channel that fails still ends the row after the channels before it.

On the IEA-15MW UMaine model with a 600-element lazy-wave cable
restrained in torsion (nine Twist and nine BendMom channels, DT 0.01 s),
CalcOutput takes 12.3 ms per step instead of 26.6 ms (channels 1.0 ms
instead of 9.9 ms, the .CD.out row 0.6 ms instead of 5.1 ms).
The coupled cases, restarts, linearization files, FAST.Farm and the
example decks are bit-identical to dev.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
On the OpenFAST route (CompMooring = 5) every CalcOutput of a
finite-EI cable restrained in torsion with polar inertia (the nodal
twist) did three line-wide assemblies the result did not need:

- The fairlead force (CD_HermiteCable_Dyn_Reaction) and the moment of
  a rigid end connection (CD_HermiteCable_Dyn_EndConnection_Moment)
  each assembled the residual and the mass of the whole line, because
  the condensed torsion's single twist couples every element. The
  nodal twist does not: an element's torque reaches only the rows of
  its own nodes.
- The transfer of the current fairlead kinematics checked the twist
  of every element twice before writing (the aggregate preflight and
  the cable's own trial pass), although the module reloads its
  committed state on any failure of that transfer.
- The restore after the evaluation transferred the committed fairlead
  kinematics into the cable, with the same checks and a twist
  assembly, and then reloaded the cable's committed state from its
  mirror, which rewrites the coupled end, its frames and the twist.

Now:

- node_reaction (the rows of one node from the elements that reach
  it) also serves a line with the nodal twist: assemble_twist takes
  node_only and adds the torques of the node's elements and end spring
  as the full assembly adds them, without touching the model's twist
  scratch. The condensed torsion keeps the full assembly. The rows are
  assembled in a buffer of their own (ws_Rnode) instead of an
  automatic array, so the call allocates nothing.
  EndConnection_Moment takes full_assembly, as Reaction does.
- An output probe (CD_AGG_UpdateStates_Moving, output_probe) skips the
  preflight of a cable with the nodal twist, and
  CD_HFMF_SetCoupledKinematics takes caller_restores to skip its trial
  pass there: the write itself checks the same conditions with the
  same message, and the probe's caller restores the whole state on
  failure. The condensed torsion keeps both checks.
- The restores of CD_CalcOutput and lin_eval pass cables_from_mirror:
  the transfer leaves the cables to the mirror reload that follows.

test_hermite_node_reaction gains nodal-twist cables (End B held or on
a torsional spring, torsional damping, drag and added mass) at
scattered states: the node reaction at every node and the moment at
both rigid ends equal the full assembly bit for bit and leave the
scratch untouched.

On the IEA-15MW UMaine model with the three chains and a lazy-wave
cable of 184 elements restrained in torsion (DT = 0.025 s, 150 s),
CalcOutput takes 11.2 s instead of 30.4 s (1.9 ms per step instead of
5.1 ms) and the run 65 s instead of 81 s; .out, .CD.out and
.CD.static.out are bit-identical to dev.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The coupled .CD.out record takes a row after every committed CableDyn
step. Each row re-evaluates the deck channels on the committed state
and writes them, a large share of the module's output cost on a cable
with many station channels at a small step (40-50 rows per second at
DT = 0.02-0.025 s).

OPTIONS CDOutDt [s] (alias cd_out_dt) writes a row at every whole
multiple of it instead; 0, the default, keeps a row per committed step.
It must be a whole multiple of the CableDyn step (dtM, or DT without
dtM), or the run stops at initialisation. The OpenFAST shell counts
the glue steps per row (nsuper times the CableDyn steps), the rewind of
an abandoned branch trims the rows that branch wrote, and a restart
continues on the same grid. The channels OpenFAST itself writes follow
its DT_Out as before. The Python deck reader accepts and range-checks
the option.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The nodal twist evaluates every element's reference twist and holonomy,
with their 12 x 12 Hessians on each tangent assembly. On a power cable
with the nodal twist in a coupled OpenFAST run, these kernels are a
large share of the cable step. A good part of that time went on
run-time library calls and on Hessian entries nobody reads:

- skew, outer, outer9 and identity3 used RESHAPE and SPREAD, which
  gfortran turns into calls into its run-time library (spread,
  reshape_r8) on every Gauss point and solid angle. They now set their
  entries one by one, with the same products.
- The solid angle of a reference triangle formed its full 9 x 9 Hessian,
  while CD_HermiteTwist_Reference reads only the blocks between the
  triangle's free vertices (the node tangents; the committed tangents
  and end directors are fixed). With the new optional argument free,
  solid_angle forms those blocks alone, each entry with the same
  expression as before.
- element_holonomy evaluated the shape derivatives twice per Gauss
  point (once directly, once inside element_ab). It also formed the two
  blocks of its 9 x 9 intermediate Hessian that are never read (rows of
  the unit tangent against the r' and r'' columns).
- The finiteness checks of the reference vertices and of the Hessian
  are one rank-3 and one rank-2 test instead of 6 and 12 calls per
  element.

Every entry is formed by the same floating-point operations in the same
order, so the results are bit-identical.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
The VIV FRAME section takes an optional third column, rows TypeName Tf
[Method]: filter (the default, the first-order filter as before) or cycle.
With cycle the frame velocity of a synchronisation type is the mean of the
node velocity over the window Tf plus half its change over it,

  vbar(t) = (1/Tf) int_{t-Tf}^{t} v dt' + (v(t) - v(t-Tf))/2,

which drops a component of period Tf and its harmonics (the cross-flow VIV
and the in-line VIV at twice its frequency) and passes the slow motion of a
towed or floater-driven cable without the filter's lag; Tf is then one
period of the response. The filter's lag leaves the frame faster than the
cable on a decelerating motion and slower on an accelerating one.

The kernel (CD_VIV_Cycle_Init, CD_VIV_Cycle_Update, CD_VIV_Cycle_Valid)
samples the velocity's integral every Tf/64, exactly for the velocity linear
over a step, in a ring of 66 samples held relative to the integral at the
committed time; I(t - Tf) and v(t - Tf) come from the quadratic through the
three samples around t - Tf. Its state, 2 + 3 x 66 rows per node, sits in
the motion statistics after the frame rows on EI = 0 lines and finite-EI
cables, so step recovery, snapshots and checkpoints carry it, and a
restored state with a sample time off [0, 1] or a slot off the ring fails
closed. A line takes one method; rows without the column are unchanged to
the bit.

Tests: viv_cycle_frame (a period-Tf component and its second harmonic drop
out within 0.25 %, a ramp passes to round-off, a quadratic gives the
T^2/12 error, long steps); hermite_viv_frame F5 (moving and fixed pipe
agree, the frame follows the imposed velocity within 0.17 % against 1.7 %
for the filter, snapshot replay bit for bit); hfmf_mirror (6) checkpoint
round-trip and an invalid slot failing closed; viv_coupled (checkpoint
restart and rewind with the ring on both line kinds); viv_frame_deck G5-G7
(Method filter is the row without it to the bit, Method cycle towed and
held, refusals); viv_wake_deck W7 (the four-column row, an unknown Method
and Method cycle on a wake-oscillator type refused by name). The Python
deck checker and builder follow (viv_frame_method).

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
Set the version to 0.4.3 (2026-11-20) in the build, the Fortran banner,
the OpenFAST module, the C API and its header, the Python package and
CITATION.cff, and date the CHANGELOG section. The release note of the
documentation front page names what 0.4.3 changes. The README and
VALIDATION.md name the 0.4.2 release driver as the one the L5 wall-clock
records were measured with.

Signed-off-by: Jae Hoon Seo <jaehoon.seo@inha.ac.kr>
Holcombe and Bergdahl rerun with the 0.4.3 GNU Release build. Both
reproduce their summaries, metrics, refinement figures and the hashed
Holcombe static profiles byte for byte; provenance.json names the
release, the date and the executable. The Bergdahl waveform comparison
still stands from the 0.3.0 run. The bodies suite passes all 102
metrics with the same build.

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 commented Nov 20, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-20T14:11:46.785397Z 706ab87 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

Copy link
Copy Markdown

Documentation build overview

📚 CableDyn | 🛠️ Build #36044157 | 📁 Comparing 706ab87 against latest (789cb6d)

  🔍 Preview build  

24 files changed · ± 24 modified

± Modified

@chatgpt-codex-connector

Copy link
Copy Markdown

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

Reviewed commit: 706ab87df9

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".

This branch is waiting to be deployed

1 waiting deployment
testpypi — 706ab87d Waiting Nov 20, 2026 by SMI-Lab-Inha via publish-testpypi #237
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