wec-sim-python

cmudrc fork status: This is an active parity effort, not yet a complete wave energy converter simulator. A case-driven dynamics runner now covers Sphere heave free decay, RM3 regular-wave coupled motion, and OSWEC hinged pitch. Other mechanical layouts and force models remain unsupported. See PARITY.md for verified behavior, current MATLAB reference revision, and the remaining work.

Install from a clone with Python 3.12:

python3 -m venv .venv
.venv/bin/python -m pip install -e .
.venv/bin/python -c 'from wecsim import WEC; print(WEC)'

For a regular install without cloning, use python -m pip install git+https://github.com/cmudrc/wec-sim-python.git. Both methods install the wecsim import and the wecsim and wecsim-reference commands.

For development, run the production-code parity checks with:

.venv/bin/python -m pip install pytest
.venv/bin/python -m compileall -q wecsim
.venv/bin/python -m pytest -q tests/test_wave_parity.py tests/test_body_io.py tests/test_oswec_standalone.py tests/test_rm3_standalone.py tests/test_reference_cli.py tests/test_general_dynamics.py tests/test_case_dynamics.py tests/test_pto_connections.py tests/test_python_api.py

WEC-Sim-Python is Sungjun Won's Python port of WEC-Sim, the MATLAB/Simulink wave energy converter simulator. This fork is developing and checking the Python code against MATLAB WEC-Sim while preserving the original author's work.

The installable code is in wecsim/, and runnable cases and bundled RM3 inputs are in examples/. Sungjun Won's early sketches remain available in Git history.

Goal of WEC-Sim-Python

WEC-Sim-Python aims to help researchers, start-up companies, and enthusiasts without access to MATLAB in order to use the open-source code provided by NREL and Sandia lab. Also, with growing research in the field of machine learning, WEC-Sim-Python could be more convenient for those who develop machine learning projects utilizing Python.

Current status

Wave generation, RM3 and OSWEC hydrodynamic input, force preprocessing, and the supported device dynamics have paired MATLAB checks. The Python API is the primary way to configure a linearized device. It constructs bodies, named motions, attachment points, PTOs, and waves as Python objects, then returns NumPy arrays directly. Import it as wecsim after installation:

from wecsim import NoWave, WEC, WorldPoint

wec = WEC("Heaving float")
float_body = wec.body("float", "path/to/hydro.h5")
wec.coordinate("heave", float_body.move("heave"))
wec.pto("main", float_body.at(0, 0, 0), WorldPoint(0, 0, 10),
        damping=1_200_000)
result = wec.run(NoWave(), dt=0.1, end_time=20,
                 radiation_memory=15,
                 initial_coordinate={"heave": 1})
heave = result.bodies["float"].position[:, 2]
pto_force = result.ptos["main"].force

The full Python example defines a two-body device with a shared pitch pivot and body-local PTO points. Run it with python -m examples.configurable_rm3_pto. Relative HDF5 paths in the Python API resolve from the current directory unless base_dir is passed to wec.run. Body order must match the HDF5 hydrodynamic body order. The Python builder currently covers the linear_subspace layout with regular, regularCIC, or no waves; its named coordinates use small-motion kinematics and fixed-axis PTOs. The other validated reference layouts remain available through the case runner and focused solver functions. For a regular-wave declutching PTO, pass control=DeclutchingControl(gain=232_020, declutch_time=0.8) to wec.pto instead of constant damping. Import DeclutchingControl from wecsim. The controller disengages at a velocity sign change and reengages after the configured interval; minimum_on_time defaults to 0.2 s. The published Sphere case is paired against MATLAB at a 0.01 s step. Other geometries and wave settings are not yet paired. For the published Sphere latching case, pass control=LatchingControl(gain=49_181, latch_damping=37_308_296, latch_time=2.4). The controller applies the larger damping for the timed interval after a velocity reversal, then returns to its normal gain. It is a finite damping force, not a rigid lock. minimum_normal_time defaults to 0.2 s. The control settings serialize through WEC.to_case for reproducible saved cases.

The JSON runner remains available for saved and reproducible cases. It accepts simulation, wave, body, constraint, and PTO settings. The included RM3 example runs with the historical bundled HDF5:

python -m wecsim examples/rm3.json --output results/rm3.csv

The runner saves body position and velocity, wave elevation where applicable, PTO force or torque, and a JSON provenance record beside the CSV. HDF5 paths in the case file resolve relative to that file. For a comparison with current MATLAB WEC-Sim, point the case at the HDF5 file from the pinned reference run. The provenance record hashes the case, hydrodynamic files, and any replayed phase CSV.

Supported combinations are:

Constraint kind	Wave type	Bodies and PTO	Integration
heave	none	One equilibrium-mass body, initial heave displacement, no PTO	Radiation convolution
fixed_hinge	pm or pm_multi	Hydrodynamic flap, optional fixed hydrodynamic base, pitch PTO	Directional PM excitation and radiation convolution; pm_multi sums independently phased seas
fixed_hinge	spectrumImportFullDir	Hydrodynamic flap, optional fixed hydrodynamic base, pitch PTO	Imported frequency-dependent directional spectrum and radiation convolution
fixed_hinge	regular	One hydrodynamic flap, optional fixed nonhydrodynamic base, pitch PTO	Regular-wave excitation and constant-frequency radiation
floating_joint	regular or regularCIC	Two equilibrium-mass bodies, pitch inertias, relative-heave PTO; optional body_to_body	Constant-frequency radiation, impulse-response convolution, or sampled FIR radiation
floating_joint	elevationImport	Two equilibrium-mass bodies, relative-heave PTO, optional joint surge spring	Imported MAT elevation and radiation convolution
floating_joint	none	Two equilibrium-mass bodies, named initial coordinates and speeds, relative-heave PTO	Radiation convolution for paired free decay; sampled FIR is also available
linear_subspace	regular, regularCIC, or none	Any number of six-DOF hydrodynamic bodies; named motions or 6-by-N maps; optional linear or rotational PTOs; selected mean-drift coefficients for regular waves	Constant-frequency radiation or convolution

The published OSWEC passive-yaw cases use one yaw coordinate and a torsional PTO. Set passive_yaw=True on the moving body to interpolate excitation at the wave heading relative to its current yaw angle:

from wecsim import RegularWave, WEC, WorldPoint

wec = WEC("Passive-yaw OSWEC")
flap = wec.body("flap", "oswec.h5", mass=12_700,
                inertia=(1.85e6,) * 3, passive_yaw=True)
wec.body("base", "oswec.h5", mass=999, inertia=(999,) * 3)
yaw = wec.coordinate("yaw", flap.move("yaw", pivot=WorldPoint(0, 0, -8.9)))
wec.rotational_pto("hinge", yaw, damping=120_000)
result = wec.run(RegularWave(2.5, 8, direction=10),
                 dt=0.01, end_time=600, ramp_time=100)
angle = result.bodies["flap"].position[:, 5]
absorbed_power = result.ptos["hinge"].absorbed_power

The rotational PTO reports angle in radians as stroke, angular speed in rad/s as velocity, and torque in N m as force. Passive yaw currently supports one pure-yaw hydrodynamic body, stationary additional bodies, regular waves, and independent radiation with full-circle BEM headings. The Python excitation interpolates continuously; the pinned MATLAB case holds its heading coefficients until the angle changes by 0.01 degree.

For the published Sphere Mean_Drift application, select the control-surface coefficient in the generated HDF5 file and use convolution radiation:

from wecsim import RegularCICWave, WEC

wec = WEC("Sphere mean drift")
sphere = wec.body("sphere", "Mean_Drift/hydroData/sphere.h5",
                  inertia=(837.75804096,) * 3,
                  mean_drift="control_surface")
for axis in ("surge", "heave", "pitch"):
    wec.coordinate(axis, sphere.move(axis))
result = wec.run(RegularCICWave(0.1, 2), dt=0.01, end_time=100,
                 ramp_time=20, radiation_memory=10)
drift_force = dict(result.raw.extra_outputs)["body1_mean_drift_force"]
excitation_force = dict(result.raw.extra_outputs)["body1_excitation_force"]

mean_drift="momentum_conservation" selects that HDF5 coefficient when present. An absent selected dataset raises an error. The regular-wave drift force is the selected coefficient times wave amplitude squared and the wave force ramp. The published unmoored linear Sphere run travels more than 16 m in surge over 100 s; the paired result reproduces that source model but does not establish physical accuracy at such a large displacement.

The runner rejects unsupported layouts and settings. For the PM case, supply wave.height, wave.period, optional directions and spreading, and either an integer seed or a phase_file CSV to replay a MATLAB realization. simulation accepts dt, end_time, optional ramp_time, rho, g, and radiation_memory for radiation-memory cases. For an RM3 regularCIC floating joint, set simulation.radiation_method to "fir" to use the published discrete FIR calculation; "convolution" remains the default. The same setting is available for no-wave free decay, without a paired FIR baseline. Both use radiation_memory (60 s by default). The general dynamics module assembles the supported body and PTO forces; it does not parse Simscape models. The inherited WaveClass now uses the pinned MATLAB PM and JONSWAP spectrum definitions, including height-dependent PM energy and JONSWAP's inferred gamma when it is unspecified. Its seeded phases use a local NumPy generator; MATLAB's Threefry substreams produce different realizations for the same integer seed. To replay a MATLAB realization, assign its frequency-by-direction phase matrix before waveSetup:

import numpy as np
from wecsim.waveClass import WaveClass

wave = WaveClass("irregular")
wave.T, wave.H = 8, 2.5
wave.spectrumType = "PM"
wave.freqDisc, wave.numFreq = "Traditional", 64
wave.waveDir, wave.waveSpread = [0, 30, 90], [0.1, 0.2, 0.7]
wave.phaseData = np.loadtxt("pm_phase.csv", delimiter=",")
wave.waveSetup([0.4, 2.0], "infinite", 1, 0.1, 20, 9.81, 1000, 2)

The historical BS fixture remains a compatibility check for the original Python port; current MATLAB WEC-Sim rejects BS inputs. For this RM3 convolution layout, the solver follows the published pitched slider geometry and uses an implicit effective added mass by default. To reproduce the pinned MATLAB/Simulink numerical trajectory, set simulation.added_mass_scheme to "simulink_delay" (or pass that keyword to solve_rm3_regular and the RM3 MCR runners). This explicitly selects the source model's mass split and 1e-7 s acceleration delay; the delay is a numerical setting, not a WEC property. The published Cases 5 and 6 use a suspect fitted state-space radiation model and remain unsupported. The paired Cases 3 and 4 tests check the ordinary implicit solver separately from this opt-in numerical comparison; neither path uses the Cases 5 and 6 fit. For the fixed_hinge layout, an optional second body can be fixed. The regular-wave case accepts a nonhydrodynamic base declared with nonhydro: true, fixed: true, and a three-component center_gravity; PM cases can instead use a hydrodynamic base with fixed: true. Its stationary motion appears in the response. With this base, constraint.location is its ground attachment and pto.location is the flap's hinge attachment. The base's constraint reaction forces are not yet calculated. The regularCIC floating-joint path has paired MATLAB checks for RM3 body-to-body Cases 3 and 4 and all eight physical settings in the published RM3 Multiple Condition Runs Option 1 sweep. The published RM3 radiation options case also checks constant, convolution, and FIR dynamics over 500 s. MATLAB's fitted radiation state-space Cases 5 and 6 remain outside that validated path. The programmatic RM3 solver also accepts mooring_surge_stiffness for a linear spring at its floating joint and excitation_force for sampled imported waves. The pinned MooringMatrix case pairs these forces and the full 400 s body, PTO, and mooring trajectories with MATLAB. Its source body block applies the wave ramp again after the imported-elevation convolution; the paired test reproduces that source-specific step explicitly. The same setup can be passed to the public Python case runner without precomputing a force array:

from wecsim import run_case

case = {
    "simulation": {"dt": 0.01, "end_time": 400, "ramp_time": 40,
                   "radiation_memory": 60},
    "wave": {"type": "elevationImport", "file": "Mooring/MooringMatrix/etaData.mat",
             "variable": "etaData", "reapply_force_ramp": True},
    "bodies": [
        {"hydro_file": "_Common_Input_Files/RM3/hydroData/rm3.h5",
         "hydro_body": 1, "mass": "equilibrium", "pitch_inertia": 21_306_090.66},
        {"hydro_file": "_Common_Input_Files/RM3/hydroData/rm3.h5",
         "hydro_body": 2, "mass": "equilibrium", "pitch_inertia": 94_407_091.24},
    ],
    "constraint": {"kind": "floating_joint", "location": [0, 0, 0],
                   "initial_coordinate": {"spar_heave": -0.21}},
    "pto": {"kind": "relative_heave", "damping": 1_200_000},
    "mooring": {"kind": "joint_surge_spring", "stiffness": 100_000},
}
response = run_case(case, base_dir="path/to/WEC-Sim_Applications")

reapply_force_ramp=True reproduces the pinned MATLAB body block's second force ramp; its default is False. The current mooring setting is a surge spring at this joint. Arbitrary mooring matrices and attachment locations remain unsupported.

The RM3 floating-joint solver accepts optional PTO hard stops in Python:

from wecsim import LinearHardStops, solve_rm3_regular

stops = LinearHardStops(
    lower_bound=-0.6, upper_bound=0.6,
    lower_stiffness=100_000_000, upper_stiffness=100_000_000,
)
response = solve_rm3_regular(
    "rm3.h5", pto_hard_stops=stops, dt=0.025, end_time=400,
)
print(response.pto_stroke, response.pto_stop_force)

The case runner also accepts these names under pto.hard_stops for a floating_joint with regular waves. Hard stops select adaptive integration, constant-frequency radiation, and implicit added mass. Ordinary RM3 cases retain their existing solver. The published MATLAB End_Stops run is time-step sensitive after contact. Paired motion, force, and energy checks pass through 400 s against the 0.0125 s MATLAB run; a 0.025 s run gates source time-step convergence. See PARITY.md for the limits and the published 0.1 s discrepancy.

RM3 multiple-condition runs can be configured in Python without a JSON input:

from wecsim import mcr_grid, run_rm3_mcr

conditions = mcr_grid(
    heights=[1.5, 2.5], periods=[6, 8],
    damping_values=[1_200_000, 2_400_000],
)
result = run_rm3_mcr("rm3.h5", conditions)
power = result.power_matrix(damping=1_200_000)
print(power.periods, power.heights, power.absorbed_power)
first_trajectory = result.traces[0].response

mcr_wave_statistics(path, damping_values) reads the published Option 2 Excel grid; mcr_mat_file(path) reads the Option 3 MAT-file case table. run_mcr(conditions, simulate, averaging_start_time=...) accepts a Python callback for another configured WEC. The callback returns an MCRTrace with time samples and signed absorbed PTO power; an optional response object keeps the full trajectory available. The published RM3 example averages from 199.9 s through 400 s and reports positive absorbed power, the opposite of MATLAB's signed PTO power column. A spring may return stored energy, so an individual absorbed-power sample may be negative.

The paired MATLAB workflow executes the actual Option 1, 2, and 3 wecSimMCR drivers separately. For each option it compares all eight body and PTO trajectories, average powers, and power matrices. The Option 1 physical conditions are also paired as scalar MATLAB runs. Phase-seed sweeps, multiple PTOs, and other MCR postprocessing remain unverified.

For the published three imported-spectrum RM3 sea states, pass the MAT-file table to the Python runner. Paired MATLAB checks cover the incident waves, excitation, body and PTO trajectories, and absorbed power. Across the three 400 s runs, the largest position differences are 14.4 mm surge, 1.03 mm heave, and 0.000638 rad pitch; mean absorbed powers differ by at most 29 W:

from wecsim import run_rm3_spectrum_mcr

sea_states = run_rm3_spectrum_mcr(
    "rm3.h5", "RM3_MCROPT3_SeaState/mcrExample.mat",
)
print(sea_states.mean_absorbed_power)
first_wave = sea_states.wave_elevation[0]
first_float = sea_states.traces[0].response.body_position[:, 0, :]

The three spectrumData*.mat files must sit beside mcrExample.mat. Their third column supplies the exact phase used for each wave component; no random seed is needed. The runner uses the same 60 s convolution radiation model as the validated RM3 cases 3 and 4.

For a no-wave floating joint, constraint.initial_coordinate and constraint.initial_speed accept four values or dictionaries keyed by surge, float_heave, spar_heave, and pitch. In the published RM3 PTO extension examples, float_heave: 5 or spar_heave: -5 creates the same initial +5 m relative PTO stroke while moving a different body. For linear_subspace, the map's rows are surge, sway, heave, roll, pitch, and yaw; its columns are independent generalized coordinates. A two-body heave case, for example, maps the first body's heave to coordinate 1 and the second body's heave to coordinate 2. constraint.initial_coordinate and initial_speed set those coordinates, and pto.damping_matrix and stiffness_matrix apply generalized linear forces. This layout assumes small rotations and a constant coordinate map. The matrices may be signed to represent active linear feedback. For example, the published Sphere reactive PI controller uses a heave-coordinate PTO with stiffness_matrix: [[-573350]] and damping_matrix: [[49181]], giving force = -49181 * heave_speed + 573350 * heave_displacement in the runner's force convention. This published case has no PTO stroke limit; the paired trajectory comparison does not validate hardware feasibility.

Configuring body motions and PTO attachments

The configurable two-body example shows a WEC definition with named bodies, independent motion coordinates, and a PTO connection. Run it with:

python -m wecsim examples/configurable_rm3_pto.json --output results/configurable-rm3.csv

In constraint.coordinates, each named coordinate lists the body motions it drives. Assigning the same coordinate to two bodies gives them a shared motion; separate coordinates let them move independently. A motion names one of surge, sway, heave, roll, pitch, or yaw and may include a scale (default 1). For a rotational motion, pivot can specify {"world": [x, y, z]} or {"point": [x, y, z]} in that body's local center-of-gravity frame. The rotation then moves the body center about that point. With no pivot, the body rotates about its center of gravity. The example's shared pitch rotates both RM3 bodies about world [0, 0, 0]. initial_coordinate and initial_speed may be vectors or dictionaries keyed by coordinate name. The original 6-by-N coordinate_map remains available for advanced cases. These maps describe constant, small-motion kinematics, not an arbitrary multibody joint solver.

ptos is a list of linear actuators. Each PTO has a unique name, from and to endpoints, and damping and/or stiffness. A body endpoint uses {"body": "float", "point": [x, y, z]}: the point is in that body's local frame relative to its HDF5 center of gravity. A fixed world anchor uses {"ground": [x, y, z]}. axis gives a fixed world-space force direction; if omitted, the axis follows the line between endpoints at the reference pose and then stays fixed. Stroke is projected displacement along that axis from the reference pose. Moving an attachment point changes its moment arm when the body rotates. The CSV includes each PTO's stroke, velocity, force, and power absorbed by its damper, as well as named coordinate motion. This force is positive along the axis on the to endpoint and opposite on from. The reported absorbed power is damping times stroke velocity squared for a constant damper and zero while a declutching PTO is disengaged. This connection model is linearized for small rotations; it does not update the actuator's axis as its endpoints move. Use either ptos or the older pto matrix in one case. In the example, each body-local PTO point has a vertical coordinate that places it at world z = 0 in the reference pose; its horizontal offset sets the pitch moment arm.

PTO tuning

The fixed_hinge and floating_joint layouts accept scalar pto.damping and optional pto.stiffness. They also accept either pto.equilibrium_position or pto.pretension. For hinge pitch, the position is an angle in radians and forces are torques; for the floating joint, it is relative heave in meters and forces are newtons. The implemented law is F = -damping * velocity - stiffness * (position - equilibrium_position). As in MATLAB WEC-Sim, pretension sets the equivalent equilibrium position to -pretension / stiffness; nonzero pretension or equilibrium position requires positive stiffness. These fields default to zero, preserving the paired reference cases. For example, add "stiffness": 100000, "equilibrium_position": 0.1 to the RM3 example's pto object to shift its neutral relative heave by 0.1 m.

The linear_subspace layout uses pto.stiffness_matrix and pto.damping_matrix, with optional pto.equilibrium_coordinate (N values). Its force law is F = -K @ (q - q_eq) - C @ q_dot. A nonzero offset must produce a spring force. The zero-offset cases have paired MATLAB checks; nonzero offsets have analytical and case-level tests. Each actuator in ptos can instead set damping, stiffness, and either equilibrium_position or pretension. Connection geometry and nonzero offsets have case-level and analytical checks. A paired MATLAB Sphere case covers nonzero stiffness, extra damping while specifying a body-local attachment shifted 1 m in x. Its heave-only motion cannot validate attachment-location dynamics or a rotational moment arm. Force limits, hard stops, controllers other than the published declutching and reactive PI laws, and hydraulic PTO models are not implemented.

The published Sphere free-decay cases can also be calculated with the focused solver:

from wecsim.linearHeave import solve_heave_free_decay

response = solve_heave_free_decay("path/to/sphere.h5", initial_displacement=1.0)
# response.time, response.position, response.velocity, response.force_total

This solver assumes one heave-only body, zero incident waves, and no PTO, mooring, or nonlinear force. Generate sphere.h5 with the published WEC-Sim_Applications Sphere bemio.m, or download the HDF5 artifact from the MATLAB reference-model run. The RM3 regular-wave heave subsystem can be calculated with the same module:

from wecsim.linearHeave import solve_two_body_regular_heave

response = solve_two_body_regular_heave(
    "path/to/rm3.h5", wave_height=2.5, wave_period=8.0,
    pto_damping=1_200_000.0,
)
# response.position and response.velocity each have two body columns;
# response.pto_force is the PTO force acting on body 1.

This calculation includes only vertical translation, a linear relative-motion PTO, and frequency-dependent hydrodynamic coefficients at the incident wave frequency. The coupled RM3 reference model also predicts both body surge motions and their shared pitch:

from wecsim.rm3Regular import solve_rm3_regular

response = solve_rm3_regular("path/to/rm3.h5")
# response.body_position and response.body_velocity have shape
# (time_steps, 2 bodies, 6 DOFs); response.pto_force is the heave PTO force.

This model uses the published RM3 inertias, regular wave, floating-joint geometry, and linear PTO. It covers the canonical case's active degrees of freedom; it does not supply general Simscape joint dynamics. The OSWEC reference can generate a seeded Python wave realization and pass its six-component excitation history to the hinge-pitch solver:

from wecsim.hingePitch import solve_hinged_pitch_from_excitation
from wecsim.irregularWave import (
    pm_equal_energy_components, synthesize_irregular_response,
)

components = pm_equal_energy_components(
    "path/to/oswec.h5", significant_height=2.5, peak_period=8,
    directions=[0, 30, 90], spreading=[0.1, 0.2, 0.7], seed=7,
)
wave = synthesize_irregular_response(
    "path/to/oswec.h5", components, dt=0.1, end_time=400, ramp_time=100,
)
response = solve_hinged_pitch_from_excitation(
    "path/to/oswec.h5", wave.excitation_force, hinge_z=-8.9,
    body_mass=127_000, pitch_inertia=1.85e6, pto_damping=12_000,
)
# response.angle is in radians; wave.elevation is the incident elevation.

It models pitch about a fixed hinge and the radiation-memory force. PM equal-energy bins, directional excitation, and pitch are checked against current MATLAB WEC-Sim using the same saved random phase matrix. A Python integer seed creates a reproducible Python realization, with a different random sequence from MATLAB.

For an imported full-directional spectrum, configure the Python case with wave.type = "spectrumImportFullDir", a MAT wave.file, and either a wave.phase_file CSV or an integer wave.seed. The default includes each heading-bin width in both elevation and force, preserving the spectrum's integrated energy. The published MATLAB Full_Directional_Waves force block omits that width even though its elevation includes it. To reproduce that specific source trajectory, set wave.force_quadrature to "matlab_omitted" and wave.excitation_interpolation to "spline_frequency"; simulation.added_mass_scheme = "simulink_delay" is a separate opt-in Simulink numerical comparison. The ordinary defaults remain integrated forcing and implicit added mass.

The same wave calculation is available directly in Python:

from wecsim import (imported_full_directional_components,
                    synthesize_full_directional_response)

components = imported_full_directional_components(
    "path/to/oswec.h5", "path/to/fullDirSpectrum.mat", seed=7,
)
incident = synthesize_full_directional_response(
    "path/to/oswec.h5", components,
    dt=0.05, end_time=400, ramp_time=100,
)
# incident.elevation and incident.excitation_force are NumPy arrays.

To run a supported case without writing Python code:

python -m wecsim.reference rm3 --h5 path/to/rm3.h5 --output results/rm3.csv
python -m wecsim.reference rm3 --h5 path/to/rm3.h5 --output results/rm3-b2b.csv --b2b
python -m wecsim.reference oswec --h5 path/to/oswec.h5 --output results/oswec.csv --seed 7
python -m wecsim.reference sphere --h5 path/to/sphere.h5 --output results/sphere.csv --initial-displacement 1

Each command writes a numeric CSV and an adjacent JSON file with the model settings, HDF5 SHA-256 hash, NumPy version, and Git revision/dirty state. For RM3, --b2b includes cross-body hydrodynamic coupling as in the published B2B Case 2; the default matches B2B Case 1. The original README projected completion in August 2022; that date is no longer applicable. See PARITY.md for the tested scope and next reference case.

Design direction

The Python dynamics engine uses independent coordinates for each supported constraint layout, avoiding numerical joint drift. New layouts and force models need explicit MATLAB comparisons before they are called supported. ParaView file output and BEMIO conversion remain future work.