The root orientation of a body in flight, from its angular momentum about
the centre of mass, constant while nothing touches it, and the shape the
body takes at each instant (#3426).
A multi-segment body in torque-free flight keeps its whole-body angular
momentum L about the centre of mass. Split into the root's own rotation
and the segments' motion relative to the root, L = R (I_b w_b + h_b),
with R the root's orientation, I_b the composite inertia about the
centre in the root's frame, w_b the root's angular velocity in that frame
and h_b the angular momentum of the segments' motion relative to the
root, so the root turns at w_b = I_b⁻¹ (Rᵀ L - h_b): the angular
momentum balance of a multibody, the model #3416 attributes to Yeadon,
Atha and Hales 1990 ("The simulation of aerial movement IV", J Biomech
23(1)) and Yeadon 1993 ("The biomechanics of twisting somersaults", J
Sports Sci 11), neither yet read at the source. I_b and h_b come from the
shape's history (integrateAirborneRotation.history), and the orientation
is integrated by the engine's trajectory integrator on its quaternion block
(integrateMotionTrajectory), so the result is a value read by time,
independent of the times read.
The centre of mass is not integrated: in flight it follows the parabola
the take-off gives in closed form, and a caller places the root so the
shape's centre, turned by orientation, lies on it. The shape's bones are
in the root's own frame (root at the origin, unturned), so the result does
not depend on where in the world or at which heading the flight happens.
Requested momentum variations are integrated alongside it as body-frame
rotation tangents: the forward sensitivity equations s' = f_y s + f_p
(SUNDIALS CVODES, https://sundials.readthedocs.io/en/latest/cvodes/Mathematics_link.html).
Their 3-by-k same-unit matrix uses a Frobenius error block, independently
of the orientation's original three-component RMS. Taking the larger
normalized error keeps the physical orientation budget unchanged. For
zero absolute tolerance this block is invariant under orthogonal body or
parameter-basis changes and a common parameter-unit rescaling. Independent
rescaling of different columns has no such guarantee. The embedded local
estimate does not certify a global Jacobian error or Newton convergence.
An explicit momentumTolerance additionally controls sampled dense angular
momentum directly in kg m²/s, independently of the orientation's RMS.
A constructor-proved constant history may instead use its supported exact
commuting solution. Unsupported controls and derivative requests retain
the numerical integrator, including explicit dense physical admission.
The root orientation of a body in flight, from its angular momentum about the centre of mass, constant while nothing touches it, and the shape the body takes at each instant (#3426).
A multi-segment body in torque-free flight keeps its whole-body angular momentum
Labout the centre of mass. Split into the root's own rotation and the segments' motion relative to the root,L = R (I_b w_b + h_b), withRthe root's orientation,I_bthe composite inertia about the centre in the root's frame,w_bthe root's angular velocity in that frame andh_bthe angular momentum of the segments' motion relative to the root, so the root turns atw_b = I_b⁻¹ (Rᵀ L - h_b): the angular momentum balance of a multibody, the model #3416 attributes to Yeadon, Atha and Hales 1990 ("The simulation of aerial movement IV", J Biomech 23(1)) and Yeadon 1993 ("The biomechanics of twisting somersaults", J Sports Sci 11), neither yet read at the source.I_bandh_bcome from the shape's history (integrateAirborneRotation.history), and the orientation is integrated by the engine's trajectory integrator on its quaternion block (integrateMotionTrajectory), so the result is a value read by time, independent of the times read.The centre of mass is not integrated: in flight it follows the parabola the take-off gives in closed form, and a caller places the root so the shape's centre, turned by
orientation, lies on it. The shape's bones are in the root's own frame (root at the origin, unturned), so the result does not depend on where in the world or at which heading the flight happens. Requested momentum variations are integrated alongside it as body-frame rotation tangents: the forward sensitivity equationss' = f_y s + f_p(SUNDIALS CVODES, https://sundials.readthedocs.io/en/latest/cvodes/Mathematics_link.html). Their 3-by-k same-unit matrix uses a Frobenius error block, independently of the orientation's original three-component RMS. Taking the larger normalized error keeps the physical orientation budget unchanged. For zero absolute tolerance this block is invariant under orthogonal body or parameter-basis changes and a common parameter-unit rescaling. Independent rescaling of different columns has no such guarantee. The embedded local estimate does not certify a global Jacobian error or Newton convergence. An explicit momentumTolerance additionally controls sampled dense angular momentum directly in kg m²/s, independently of the orientation's RMS. A constructor-proved constant history may instead use its supported exact commuting solution. Unsupported controls and derivative requests retain the numerical integrator, including explicit dense physical admission.