Rigid-body flight
See the equations that fly a rocket from rail to ground as a rigid body, free to turn, that lightens as its motor burns.
In short
- What it models: a rocket’s flight from the pad to the ground, rail included. The rocket is a rigid body free to move and turn in every direction (six degrees of freedom). It gets lighter as its motor burns.
- Sources: the equations of motion in RocketPy’s technical documentation (RocketPy 1.13.0). The RocketPy paper (Ceotto et al., 2021) is cited but was not fetched.
- How well it is validated: by analytic and unit tests, against RocketPy, and on eight of NASA’s check cases. In a vacuum, a tumbling rocket’s center of mass stays on the exact parabola to 1.7e-6 m over 22 s. On six of the check cases, a free body on the same equations differs from the codes NASA published by more than they differ from each other. It ends 19.51 mm (0.77 in) lower after a 4.4 km (2.7 mi) fall, turns up to 0.01633 deg/s apart, and puts a dragged sphere up to 105.0 mm (4.1 in) away. A named difference between the models brings each inside the codes’ spread at each code’s largest difference. For one code in the sphere’s cases it does so without explaining the gap. Read at every sample, case 2’s tumbling brick’s rates are left just past the spread. Two shots from sea level come out inside it as flown. There the codes sit about 2 m (6.6 ft) apart (a free body against NASA’s check cases). Five of RocketPy’s example rockets were flown from the pad to the ground by both codes, with the same declared drag. They agree on height, speed, time and acceleration within 3%. The largest scored difference is +1.783%, a peak acceleration on the rail (validation report, whole flights against RocketPy). The path agrees as well, except the apogee drift of NDRT 2020 and of rockets that leave the rail slowly in a wind. In those, HPR Sim’s apogee drift is −4.333% to −38.158% from RocketPy’s (validation report). The causes are HPR Sim’s body lift, which RocketPy’s normal force leaves out, and its later release from the rail (ADR-026). For Juno III, another cause is HPR Sim’s simpler fin model. A sixth, Prometheus 2022, passes Mach 1 on its drag table and agrees as well. Its drifts again differ from RocketPy’s through body lift. With HPR Sim’s own drag, against RocketPy flying the drag its examples ship, HPR Sim’s heights differ from RocketPy’s by −7.280% to +10.302% (validation report). The gaps are larger where the two drags differ most. HPR Sim’s is well below the example’s for two rockets. For Prometheus 2022 it is above the example’s at high speed. That rocket has flown through Mach 1 on HPR Sim’s drag since the drag through Mach 1 was built (Accuracy). Twelve public flights have been compared with their altitude logs at the apogee, seven of them along the climb too. 60 more have been compared with their logs, on apogee alone (Accuracy: real flights).
- What it leaves out: staging and delayed ignition (both on Staging), turbulence and thrust misalignment. Tip-off (the pivot as the rocket leaves the rail) is left out unless a flight asks for it (Tip-off), and that phase is held to RocketPy’s on RocketPy’s normal force, and on HPR Sim’s own in still air only. RocketPy’s tip-off phase puts the pivot’s cost at up to 3.765% of the apogee in the landing point and 0.414% in the apogee, one code against another with no flight to check it (below). Its small-angle aerodynamics are used at every angle of attack (the angle between the rocket’s axis and its path through the air), with no stall. A flight that reaches Mach 5, the top of the aerodynamic models, stops with an error. The rotational equations leave out the Earth’s rate; on NASA’s tumbling brick that moves a body rate by up to 0.01633 deg/s within 30 s (#1521).
About the model
What the equations do
At each instant of a flight, HPR Sim adds up every force on the rocket and every turning effect (a moment). They come from the thrust, the weight, the air and the propellant burning away. From those it works out how fast the rocket speeds up and how fast its turning changes. The integrator (Time integration) then carries the rocket forward, one short step at a time.
Burning propellant makes the rocket lighter, moves its center of mass, and lets the exhaust carry away some of the rocket’s turning (jet damping). The equations keep all three. They are RocketPy’s, from its technical documentation ([RP-EOM], under Code and sources), and are written out under Equations of motion.
Code and sources
Code: hpr_sim::{dynamics, flight, free_body, histories, rail, recorder, state}. Decisions:
ADR-011 (equations of motion, aerodynamic coupling, rail, phases and termination) and
ADR-353 and ADR-356 (the differences from NASA’s check cases, traced), and
ADR-357 and ADR-358 (tip-off: its cost, the phase, and what is held). The integrator and
events are in Time integration, and the frames in Frames.
Sources:
-
[RP-EOM] RocketPy technical documentation, “Equations of Motion” v0 (Kane’s method with the Reynolds transport theorem) and v1 (the form solved). They are
docs/technical/equations_of_motion.rstandequations_of_motion_v1.rstin RocketPy 1.13.0 (refs/rocketpy, MIT). -
[C21] G. H. Ceotto, R. N. Schmitt, G. F. Alves, L. A. Pezente and B. S. Carmo, “RocketPy: Six Degree-of-Freedom Rocket Trajectory Simulator”, J. Aerosp. Eng. 34(6), 2021, doi:10.1061/(ASCE)AS.1943-5525.0001331. Cited, not fetched.
-
[N15] D. G. Murri, E. B. Jackson and R. O. Shelton, Check-Cases for Verification of 6-Degree-of-Freedom Flight Vehicle Simulations, NASA/TM-2015-218675, both volumes, 2015, with the errata of September 8, 2025, and the codes’ time histories posted with it (https://nescacademy.nasa.gov/flightsim/2015). Cases 1 to 3 and 6 to 10 are flown (below).
-
[RP-TIP] RocketPy’s tip-off phase,
Flight.udot_rail2, in its development branch (MIT; no release carries it). Its pull request #920 added the phase and #1217 rewrote it. By its description the rewrite simplifies the phase’s function, changes what RocketPy reports inside the phase, and corrects how the phase ends: it finds the moment each button leaves by a root search on the solver’s own interpolation of a step, and it corrects what happened when a parachute opened inside the phase or the rail was too short for it. It also took away the phase’s switch. HPR Sim’s tip-off phase takes its form from the first: the free equations plus the force and the moment the last guide adds. The comparisons below fly the rewrite.
The equations
State
- The state holds the nose tip ’s position and velocity in the launch frame . It also holds the attitude quaternion , four numbers that give the rocket’s orientation, turning body axes into ’s. Last come the body rates (how fast it turns about each body axis) relative to , in body axes: 13 components in all.
- The state follows the nose tip because it is fixed in the body and is the body origin (the design-tree decision). The center of mass (from , body axes) moves as propellant burns.
- The norm of the quaternion (its length, 1 for a pure rotation) drifts slightly between steps. Every use normalizes it, and it is never reset at an event (Time integration).
Equations of motion
[RP-EOM] v0 derives the motion of a variable-mass system (one whose mass changes as it flies) about a point fixed in the body. It uses two standard tools:
- Kane’s equations, a systematic way to write the equations of motion, for the rigid parts;
- the Reynolds transport theorem, which accounts for mass flowing across a boundary, for the propellant and the gas leaving the nozzle.
Its assumptions:
- the gas flow inside the motor is quasi-steady, changing slowly enough to treat as steady at each instant. So the integrals over the space it flows through, the control volume, don’t change;
- the flow is axisymmetric: the same all round the rocket’s axis;
- the exhaust momentum is lumped into the thrust at the nozzle exit.
v1 solves the result for and , the rates of change of the velocity and of the body rates. The lines below are that form, in body axes, with primes for time derivatives seen from the rocket. , , and are labels carried over from [RP-EOM] v1, so each line can be matched against it:
- comes from the momentum of mass moving inside the rocket: gas flowing to the nozzles, and the center of mass shifting as propellant burns. The rotation turns that momentum into a force, , as it does for anything moving inside a turning body (a Coriolis force).
- is the thrust with the propellant’s internal-momentum terms (Measured thrust curves and the internal-momentum terms, below).
- gathers every force: , the weight , the air force , and two terms from the rotation. They are and , the second because the center of mass is not at .
- gathers every moment about . They come from the rotation itself, from the jet and the changing inertia (), and from the weight, the air and the thrust. The last term of , , is for parts that move along the airframe. Here is their angular momentum about , relative to the airframe. The term is zero for a part on the axis, and zero when nothing moves (Moving mass).
- The last lines give the angular acceleration and the nose tip’s acceleration (and from it in ). Then comes the rate at which the attitude changes.
-
Mass terms:
- , and are the mass, its rate and each motor’s mass acceleration.
- and are the inertia about and about the center of mass ().
- is motor ’s nozzle exit.
-
Forces and moments:
- is the thrust along , the motors’
thrust_at_pressure_n, with moment about . - is the weight: normal gravity (centrifugal term included) plus the Coriolis force , both acting at the center of mass.
- and are the aerodynamic force and its moment about (below).
- is the thrust along , the motors’
-
The nozzle gyration tensor describes how the exhaust leaving motor ’s exit disc is spread about . Times the mass flow , it gives the jet damping in :
It is integrated here from v0’s boxed rotational equation. The jet term there is over the exit disc of radius , with a uniform jet. For a disc at on the axis, this gives
- RocketPy 1.13.0’s code (
rocket.py:984-985,evaluate_nozzle_gyration_tensor) uses for the transverse distance term instead of . Its is measured from the center of dry mass, not from the nose tip, so the entries can’t be compared directly. The RocketPy comparison below corrects that term to . - A motor that gives no nozzle exit radius contributes .
- RocketPy 1.13.0’s code (
-
For an axisymmetric rocket turning slowly about a transverse axis, the equations reduce exactly to the classical jet damping about the center of mass,
with from the center of mass to the nozzle exit. The part damps, and the falling inertia partly offsets it (
dynamics::tests::jet_damping_matches_the_classical_form). -
Time derivatives of the mass properties:
- , , and come from central differences of
Assembly::mass_propertieswith a half-width of 1e-4 s. - The stencil, the times each difference samples, is kept inside the current integration interval. Every thrust-curve knot and burnout is a stop time, so no difference straddles a change in the thrust’s slope. Near an interval’s end the derivative is taken at the nearest valid center and extended linearly.
- differences each motor’s mass flow the same way.
- After burnout (an interval starting past every burnout) the rates are zero. They are also zero over intervals shorter than 2e-5 s, where differences would be rounding noise.
- A part that moves along the airframe adds its own rates in closed form, not by differences (Moving mass).
- , , and come from central differences of
-
A motor that burns through an interval is evaluated at its one-sided limit inside the burn, clamped to . The last stage of the step ending at burnout (or the first after ignition) then sees the burning motor. It sees the pressure correction too, which switches off at . Without this, fixed-step RK4 converged at first order (5.1 mm (0.20 in) of apogee at 2 ms).
-
Measured thrust curves and the internal-momentum terms. A static test measures , where is the internal momentum of the burning propellant, so a
.engcurve already contains it. subtracts again.- HPR Sim keeps the terms as RocketPy does, for parity in the RocketPy comparison. The form is exact for the [RP-EOM] model, not for a measured curve.
- The size of the double count on Valetudo: it lifts off at 1.56 ms with 73 N of thrust against 95 N of weight. At liftoff 21 N comes from . The double count changes the burnout speed by at most 0.05 m/s (0.16 ft/s).
-
Earth’s rotation enters only through the Coriolis force. The rotational equations use relative to , which differs from the inertial rate by at most 7.3e-5 rad/s (Frames and the flight decision). On a tumbling body of three unequal moments the effect of leaving it out grows past that: NASA’s check cases 2 and 3 measure it.
Aerodynamics in flight
hpr-aero gives coefficients at a flow condition (Aerodynamics). The engine applies them as follows.
- The air velocity uses the wind at the center of mass’s height. Wind vectors are taken in ’s axes.
- The axial force is , from the whole rocket’s drag. It is taken at the center of mass’s
airspeed, Mach number, angle of attack and Reynolds number per meter (). The drag is power-on
(
DragConditions::thrusting, with the burning motors’ cross-section) while any motor burns in the interval.Simulation::with_full_base_drag_under_powerkeeps the whole base’s drag instead, as OpenRocket does, for comparisons with it (base drag under power). The force acts along the axis, so it has no moment about . - Normal and side forces are taken component by component. Each body and fin set is evaluated at
its own local flow: at the station the aerodynamics gives it
(
AeroModel::component_station_m). That station is the component’s small-angle center of pressure wherever one model carries the component. Faster than sound, where the shock-expansion method and slender-body theory share one, it is not. There the station is joined between the two models while the force joins their slopes and moments. The two agree only at the ends of the join. A body can sit between the models at every speed (a lip riding half in its boattail’s wake). There a tube’s station can sit two calibers from its own center of pressure (issue #106). That moves the lever arm the damping below uses by that much. The margins don’t use the station. The apogee and the drift may lean either way, so a flight inside the join warns of it (the flight-path warnings). A component’s acts along the crossing air and its along (Frames), with moments about from its moment coefficients. - At any angle of attack, a fin set’s normal force follows the crossflow . Its
model’s is used as , the substitution Niskanen keeps for bodies (eq.
3.16–3.17). No source in hand gives a large-angle fin model, and stall is not modeled.
- At small angles nothing changes (1.5% less at 0.3 rad).
- The force now vanishes for axial flow from the tail as well as from the nose.
- A force linear in stays large at and flips with rounding noise in the lateral
velocity. A calm vertical flight falling tail first after apogee then collapsed the step size
and never landed (
tests::a_calm_vertical_flight_falls_tail_first_and_lands).
- The rotation’s contribution to each local flow is the only aerodynamic pitch and
yaw damping. At a component’s station the rotation adds a velocity , which changes the
angle at which the air meets that component. The station is the lever arm. Where it is not the
center of pressure (above), the damping is levered from the wrong point by the same distance.
- Only components with a normal-force slope damp this way: nose cones, transitions and fin sets. A boattail’s slope is negative, so it takes some away.
- Body tubes give none at small angles: their own slope is 0, and their body lift grows with .
hpr-aerohas no pitch or yaw damping coefficients. They would have to replace the local-flow damping, not add to it.- A flight on another tool’s normal force keeps this damping. The table (The normal force from RASAero II) gives the normal force at the center of mass’s airflow. Each component adds only its force in its own local flow minus its force in the center of mass’s airflow: the part the rotation makes. Without rotation that part is exactly zero (ADR-032).
- The fins’ cant drives the roll and the roll rate damps it. The moment about is
, at the center of mass’s Mach number. Here is the dynamic pressure, and the
reference area and diameter, the cant’s rolling moment, the damping and
. The cant’s forcing follows the axial flow,
, so it vanishes broadside and reverses tail first
(
AeroModel::roll, Roll: forcing and damping). The damping is written , so it fades smoothly as the airspeed does. The roll rate also changes through inertia coupling (turning about one axis driving turning about another) when a rocket’s mass isn’t symmetric about its axis. While a motor burns, it also changes through the jet and the falling inertia in . At constant speed a canted rocket spins up to the steady rate where cant and damping balance, as the closed form gives (tests::canted_fins_spin_to_the_analytic_balance). - Limits:
- The models are small-angle: no stall, and body lift and fins extended by . They
overstate the forces at large . In normal flights large occurs near apogee, where the
dynamic pressure is small, and off the rail in strong crosswinds.
Sample::angle_of_attack_radshows where it happens. - The normal force and the drag buildup cover Mach 0 to 5
(Fins through Mach 1,
Drag through Mach 1, the drag’s decision). A fin
set’s CP moves with the Mach number. Each component takes its local airflow at its CP for the
center of mass’s Mach number. At Mach 5 or faster they refuse the
flow, and the flight stops with
SimError::Aero. A drag override table covers drag at any Mach, but the normal force still stops at Mach 5.
- The models are small-angle: no stall, and body lift and fins extended by . They
overstate the forces at large . In normal flights large occurs near apogee, where the
dynamic pressure is small, and off the rail in strong crosswinds.
Running a flight
Phases
-
Pad.
- The rocket starts at rest with the rail’s attitude. Its aft end is at the rail’s foot: the aft-most point it stands on, of its body, of any part on its outside or of its nozzle exits. A swept fin whose tip’s trailing edge reaches past the tube counts there, as do a tube fin’s and a pod’s aft ends. So the rail’s length runs from that point to the rail’s top, and a standoff under the rocket shortens it.
- It stays until the force along the rail, with , exceeds friction (the liftoff event). is the rail’s reaction: the weight, the aerodynamic force and the mass terms across the axis.
-
Rail. One degree of freedom, along the rail, with no rotation:
- It ends when the aft edge of the aft-most rail button or launch lug passes the top of the rail. In a flight that asks for tip-off it ends sooner, when the guide next ahead of that one does. With no guides, it ends when the aft end does.
- If the speed along the rail falls to zero, the rocket returns to the pad phase, at rest where it stopped. It is held there, even if the force down the rail exceeds friction and a real rocket would slide back.
-
Tip-off. Only when the flight asks for it: between the rail and free flight, the rocket slides on its last guide alone and pivots about it (Tip-off).
-
Free. Six degrees of freedom until the ground.
-
Rail geometry.
- A button’s axial extent is its outer diameter, and a lug’s is its length.
- The rocket leaves after traveling . RocketPy ends its rail phase when
the forward button reaches the top (
flight.py:1716-1730,effective_1rl). HPR Sim keeps the rocket guided to the last guide (Loft lesson L26). - The pivot about the last guide (“tip-off”) is left out unless the flight asks for it (Tip-off). Without it the rocket leaves the rail with no angular velocity. On thirteen example designs, RocketPy’s tip-off phase moves the landing point by up to 3.765% of the apogee and the apogee by up to 0.414%, one code against another (what leaving tip-off out costs).
- Friction is Coulomb friction: a user coefficient times the net reaction , the force pressing the guides onto the rail. With a moment across two buttons (a crosswind at low speed), the true friction is larger. No source gives a default coefficient, so the default is zero.
Tip-off
A flight that asks for it flies one more phase between the rail and free flight. That phase is
tip-off, the turn a rocket picks up on its last guide alone. A guide is a rail button or a
launch lug, and the last one is the aft-most. Ask for the phase with
Simulation::with_tip_off in the Rust library. The command line and the Python package don’t
offer it yet. It is off by default, so every flight on this
site and in the reports flies without it unless it says otherwise.
What it does. Once the guide ahead of the last one leaves the top of the rail, the last guide holds the rocket alone. The rocket keeps sliding up the rail on it, free to pitch and yaw about it, with roll held. Gravity turns the nose down, and a crosswind turns it downwind when the air’s force acts ahead of the guide. When the last guide’s aft edge passes the top, the rocket is free, with the turning rate it has picked up.
The equation. The free equations give the acceleration of the nose tip and the angular acceleration the rocket would have with nothing holding it. The guide adds a force on the pivot, square to the rail, and a moment about the long axis . Write for the pivot’s position from the center of mass and for the center of mass’s from the nose tip. is the inertia about the center of mass. The two change the motion by
Three conditions fix the three unknowns, the two parts of across the rail and . The pivot’s acceleration has no part across the rail, in either direction, and the angular acceleration has none about the long axis. With no friction the three equations are linear, and friction is solved with them (below). Source: [RP-TIP], written for this page’s equations about the nose tip.
- The pivot is on the rocket’s axis, at the middle of the aft-most guide.
- Friction is the rail’s, on that force: on the pivot along the rail, against the sliding. It is solved with the force, since it turns the rocket a little. RocketPy’s rail has none.
- The phase starts when the aft edge of the guide next ahead of the last one passes the top of the rail. It ends when the last guide’s does. A rocket with one guide, or none, has no such stretch and flies as before. A rail too short to hold both guides pivots the rocket from liftoff.
- A rocket that stops on its last guide returns to the pad phase, at rest, as on the rail.
- Flying the phase by default is a later increment.
For a burnt-out rocket in no air the three conditions reduce to a pendulum hung from the guide. Write for the distance along the axis from the last guide to the center of mass. is the inertia about the center of mass for a turn in the rail’s vertical plane, and is the rail’s angle above the horizontal. The nose turns down at
A test holds the phase to that within 1e-9 of its size on four rails, 5° to 60° off the vertical.
A worked example. Take a 20 kg (44 lb) rocket on a rail 5° off the vertical, so . Its center of mass is 0.80 m (2.6 ft) ahead of its last guide. Its inertia about the center of mass is 9.0 kg m².
That is 35.9 deg/s². If the last guide holds the rocket alone for 0.07 s, about what Juno III’s does, it leaves turning nose-down at about 2.5 deg/s. Juno III, a real example, is under the tip-off phase against RocketPy’s.
What it leaves out. A real button sits a body radius off the axis and has play in its slot, and a lug slides on a rod. Here the pivot is a pin on the axis that lets nothing through across the rail. The rail does not flex or whip. The rocket cannot slide back down, as on the rail. No real flight checks any of it.
Events and termination
-
Thrust-curve knots, burnouts and the time cap are stop times. A burnout that is reached records a
Burnoutevent. -
Liftoff is the pad force margin rising through zero. It is checked also at each interval’s start, for a thrust that jumps at a knot. Stall is the speed along the rail falling through zero.
-
Rail exit is the travel along the rail reaching the exit travel.
-
Apogee is the center of mass’s ellipsoidal-height rate, , falling through zero. is the ellipsoid normal at its position. A flight that has let a part go records one apogee (Released mass).
-
Ground hit is the center of mass’s ellipsoidal height reaching the launch site’s, descending (Frames, Loft lesson L35).
-
A user event is a function of the
Sample, in free flight. -
A mass shift (
Shift) is a part starting to move along the airframe, and a release (MassRelease) a part leaving it. Each comes on its trigger (Moving mass, Released mass). -
Atmosphere and wind heights are , the height above sea level. is the ellipsoidal height. The geoid undulation , the height of sea level above the ellipsoid, is given in
Environment(no geoid model). -
The flight ends in exactly one of these ways (Loft lesson L25):
GroundHit;NoLiftoff(the last burnout passes on the pad);StalledOnRail(it lifted off, then stopped on the rail after burnout);TimeCap;StepLimit.
Any other failure is an error.
-
Turbulence and thrust misalignment are not modeled in a flight. Recovery is modeled, and has its own page. So are motors lit at their own times and a powered separation, on Staging.
Integration settings
-
FlightSettings::default()uses Dormand-Prince 5(4) withrtol = atol = 1e-8and unit weights, a one-hour cap and 10⁶ steps. -
Accuracy against cost was measured on Valetudo (K400C) in a 5 m/s (16 ft/s) wind from the west. Each apogee is compared with the one at 1e-11 (873.98272 m (2,867.3974 ft)):
tolerance, rtol = atol(ratio; state units)apogee error run time (release build) 1e-4 2.3e-2 m 0.36 ms 1e-6 6.8e-5 m 0.59 ms 1e-8 1.1e-6 m 1.13 ms 1e-10 1e-7 m 2.72 ms
Thrust and mass from another program
A flight can take its thrust, and its mass with its center of mass, from another program, as
numbers against time, in place of its own (Simulation::with_thrust_history and
Simulation::with_mass_history, since M2.3c11). It is
for one job: splitting the gap between HPR Sim’s apogee and OpenRocket’s by cause, one input
swapped at a time, as a drag table swaps the drag (ADR-349).
The split is in
Accuracy: drag is the largest part on
61 of the 68 flights both programs fly. It is not a model, and a flight on
it is not a prediction. It is in the Rust library only, for the validation harness: no hpr
command, Python call or file option takes a history.
-
Between samples. A history is a list of samples. Between two of them, and , it is the straight line
Before the first sample and after the last it holds that sample’s value. So its rate is the slope of the piece it is on, finite, with a jump at each corner, and its second derivative is zero inside a piece. Every sample where the slope changes is a stop time, where an integration step must end (time integration), so no step straddles a corner.
-
Thrust. The history’s value is the thrust as flown, along the axis . It takes the place of the motors’ thrusts with their correction for the air’s pressure, which isn’t applied to it again. It acts at the mean of the lit motors’ nozzle exits, so a symmetric cluster’s thrust turns the rocket no more than HPR Sim’s own does. An asymmetric cluster’s turning moment is left out.
-
What stays HPR Sim’s under a thrust history. The motors still set the burnout event and the base area a burning motor frees (base drag). They also set the mass flow behind jet damping and the nozzle terms and of the equations above. So those terms’ push, 21 N at Valetudo’s liftoff (internal momentum), stays on top of the history’s thrust.
-
Mass. The history gives the whole rocket’s mass and its center of mass , in meters aft of the nose tip, the body origin . In body axes the center of mass is , with HPR Sim’s own small offsets off the axis. Then and are the history’s slopes and .
-
What stays HPR Sim’s under a mass history. The inertia about the center of mass . The inertia about comes from it by the parallel-axis theorem with the history’s mass and center, , with the identity matrix, so the inertia’s difference from OpenRocket’s lands in the split’s remainder. The motors’ mass flow stays theirs in the nozzle terms, as under a thrust history. Where the history’s mass still falls after the motors burn out, its is in the turning terms with no nozzle flow to match, so jet damping is not the classical form there. That too lands in the remainder. A structure-mass scale changes only the inertia under a mass history, since the history gives the mass.
-
Time. A history’s clock is the flight’s: 0 at launch. OpenRocket’s clock starts at the same instant: its launch and ignition events are both at 0 s.
-
What is refused, with the reason. An empty sample list, lists of different lengths, times that don’t rise, a number that isn’t finite, a negative thrust, a mass that isn’t positive, or a first sample after launch. A thrust that doesn’t end at zero, since it would stop while the motor still pushed. A thrust history on a flight with no motor lit at a known time. Either history on a flight that splits, by a separation or an ejection, or that moves or releases a part: a history is the whole stack’s, with no part’s own.
Verification
Unit tests in hpr_sim::tests, hpr_sim::dynamics::tests and hpr_sim::rail::tests, at the
default settings. The numbers were measured on September 17, 2026.
-
A tumbling, spinning Valetudo in a vacuum starts from the nose tip at 500 m (1,640 ft) with m/s.
- The center of mass stays on the closed-form parabola to 1.7e-6 m over 22 s.
- The angular momentum stays constant to 7.8e-7 (relative).
- Apogee and ground contact are within 4.7e-7 s and 1.1e-8 s of the parabola’s. Those times are located (root-found) on the exact ellipsoidal height.
-
Falling nose down through uniform air with , the speed follows to 5.5e-9 . It is the terminal velocity to 1e-5 after 150 s.
-
Torque-free precession is the wobble of a spinning body with no torque on it. Valetudo ( = 0.0019) spinning at 25 rad/s with a transverse rate turns in body axes at = −24.95 rad/s. The rate matches to 4.7e-7 rad/s over 3 s, and the angular momentum in holds to 1.5e-6.
-
The pitch oscillation is checked against linear theory at 100 m/s (328 ft/s), with no drag or gravity. The two-state linear model (path turning, restoring moment , rotational damping ) predicts a 1.44954 s period. The flight measures 1.44965 s (8e-5), and the decay per half period is within 0.2% of the model’s.
-
The flight equations give the classical jet damping to 1e-6. The mass rate equals the motor’s to 1e-6, including stencils clipped by an interval’s end.
-
A powered vertical climb in vacuum runs from 0.5 s to 3.0 s. It ends at the speed of the axial equation, integrated independently from the motor and the assembly, to 4.3e-8 m/s.
-
On a rail at 60°, friction with removes exactly from the acceleration along the rail.
-
Valetudo weathercocks (Loft lesson L20). In a 5 m/s (16 ft/s) wind from the west, its unit axis has an east component of −0.12 at burnout. Its apogee is 96 m (315 ft) upwind, against 1.0 m (3.3 ft, Earth rotation) in calm air.
-
In a calm vertical flight with no Earth rotation, the rocket falls tail first after apogee and still lands, in under 20,000 evaluations.
-
RK4 at 2, 1 and 0.5 ms gives the same apogee as Dormand-Prince to 1.2e-5 m and 2.1e-7 s.
-
A user event on the thrust fires at burnout, and burnout is still recorded once.
-
A 2,000 m (6,562 ft) rail gives
StalledOnRail, with only liftoff and burnout recorded. -
Observer errors end the flight with that error. Starts before ignition or underground are refused. A cleared recorder records the same rows again.
Simulation,EnvironmentandRecorderareSend + Sync. -
Two runs of one
Simulationrecord the same bits for every channel at every step (Loft lesson L24). So does a secondSimulationbuilt the same way. -
A normal flight hits the ground (Loft lesson L25). A rail with at 60° gives
NoLiftoffat rest. A 5 s cap givesTimeCapat exactly 5 s, and a 40-step limit givesStepLimit. -
A 3 m (9.8 ft) rail is tilted to 1.3 rad (Loft lesson L26). The rail exit comes at the last button’s travel to 1e-6 m. Across the rail the rocket stays within 1e-9 m, with no rotation. Friction () delays the exit and slows it.
-
From a 1 m (3.3 ft) rod tilted 5 to 20 degrees, HPR Sim’s rocket is on OpenRocket’s bearing at apogee to within 0.03 degrees. It loses the same apogee to within 0.27 percentage points (
.ork: a tilted launch rod). -
Events come in order: liftoff, rail exit, burnout, apogee, ground hit. Apogee’s vertical speed is below 1e-6 m/s and ground contact’s height below 1e-6 m. Recorder rows fall on the interval or at events.
-
The synthetic 54 mm (2.1 in) rocket on an I175 passes Mach 1 and lands. It does both on HPR Sim’s own drag and on a constant drag table.
-
A Valetudo flight to the ground costs about 1.3 ms, measured on October 9, 2026 (
docs/perf.md). -
Thrust and mass histories (
histories_fly_the_closed_form_of_a_varying_mass_under_constant_thrust). In a vacuum, with gravity of 9.80665 m/s² (32.174 ft/s²), long after its own motor’s burnout, Valetudo rises straight up at a speed of 50 m/s (164 ft/s) on a thrust history of 300 N. Its mass history, not Valetudo’s own mass, starts at , 10 kg, and falls at , 1.5 kg/s, so . Its center of mass moves aft at 0.05 m/s (0.16 ft/s), so is −0.05 m/s (−0.16 ft/s) along , which points to the nose. After a time , the nose tip rises atwhich reaches 101.686 m/s (333.62 ft/s) after 2 s. The flight follows it to 1e-7 m/s. The term is the axial part of the equations’ . Without it the speed would be 0.036 m/s (0.12 ft/s) higher. Each history alone follows its own closed form to 1e-7 m/s too: the thrust on the design’s burnt-out mass, 8.26 kg, gains , 53.0 m/s (174 ft/s) in 2 s, and the mass alone the closed form above with . With a corner 1 s in, where the mass flow halves and the center’s rate drops to 0.02 m/s (0.066 ft/s), a step ends at the corner and each piece follows its own closed form to 1e-7 m/s.
-
Under a mass history, the inertia about the nose tip and its rate are the parallel-axis formula and its derivative, with the history’s mass and center and the design’s own inertia about its center of mass, to 1e-12 and 1e-6 of their sizes (
a_mass_history_moves_the_assembly_s_inertia_by_the_parallel_axis_theorem). -
A mass history read back at 1 s gives the history’s mass and center along the axis, with the design’s offsets off the axis and inertia. A separation, before or after a history, refuses it with the reason, and so do a mass shift and a mass release.
-
The own-history test (
cargo test -p xtask histories). Two public designs are flown, sampled at the times of OpenRocket 24.12’s own flight of each (validation/fixtures/ork/openrocket-histories.json), and flown again on those samples. Both are a 60 mm (2.4 in) airframe with a conical nose and three fins:pods-noneon an AeroTech H128W, the control of a tilted launch rod, andmach-probeon a J570W, made for this test. Onpods-none, a 251-sample flight to Mach 0.76, the apogee goes from 782.7197 m (2,567.978 ft) to 782.7214 m (2,567.984 ft), +0.00022%. Onmach-probe, a 567-sample flight to Mach 1.73, it goes from 2,030.4623 m (6,661.622 ft) to 2,030.4496 m (6,661.580 ft), −0.00063%. The bound is 0.05%, about a thirty-fifth of the 1.73% mean absolute gap between the two programs’ apogees on 60 real flights (accuracy). On the same flights, 5% more thrust raises the apogee by 2.8% and 3.0%, and 20% more mass delays the rail exit by 7.9% and 7.2%: both histories are flown.
A free body against NASA’s check cases
NASA’s Engineering and Safety Center published check cases for flight simulations that move a body in all six degrees of freedom [N15]. Each case is a simple body flown from a stated start. Six codes flew the cases in the air, labeled SIM 1 to SIM 6 without their names: five of NASA’s own and JSBSim, an open-source simulator. Five or six posted a time history for each case, beside the report. The report names no code as right: the codes’ agreement with each other is the measure.
HPR Sim flies eight of the cases with a free body, hpr_sim::FreeBody: a mass and an inertia,
with no design, no motor and no aerodynamics. It is not a second model. Its gravity and
Coriolis force come from the flight’s Earth model, its air and wind from the flight’s
environment at the height a rocket’s flight reads them, its accelerations from the equations
above, and its steps from the integrator at the default tolerances. A case’s damping or drag
is a load the test adds. At every step it forms a force and a moment from the state, the air
and the velocity through the air. A test
(the_derivative_is_a_coasting_rockets) evaluates a coasting rocket by the equations a flight
integrates, and the free body of that rocket’s mass and inertia at the same state: the two
agree to rounding.
- Case 1, the dragless sphere. A sphere of 1 slug (14.59 kg) is dropped from 30,000 ft (9,144 m) over 0° N, 0° E, at rest relative to the Earth, and falls for 30 s with no drag. Its altitude is compared.
- Case 2, the tumbling brick. A brick with three unequal moments of inertia starts at the same place, turning at 10, 20 and 30 deg/s about its three axes relative to the stars, with no aerodynamics. Its three body rates relative to the stars (inertial rates) are compared over 30 s.
- Case 3, the tumbling brick with damping. Case 2 with a damping moment about each axis. The report gives three coefficients of −1, a reference area, a span and a chord, but no formula. The test takes the usual one: about each axis , which is the dynamic pressure times times the rate made dimensionless, . Here is the speed through the air, the rate relative to the stars about that axis, and the span for roll and yaw and the chord for pitch. The codes split on the rate. SIM 4 damps the one relative to the stars, and SIM 1 comes nearer it than the other. SIM 5 and SIM 6 damp the one relative to the Earth. SIM 2 fits neither. The test takes the first. The inertial rates are compared.
- Cases 6 to 10, the sphere with drag. Case 1’s sphere with a drag coefficient of 0.1, acting along its velocity through the air whichever way it points. It is dropped from 30,000 ft in still air (case 6), in a steady 20 ft/s (6.1 m/s) wind from the west (7), and in a wind from the west that grows by 3 ft/s every 1,000 ft of height, 70 ft/s (21 m/s) at the start (8). In cases 9 and 10 it is shot from sea level at 1,000 ft/s (305 m/s) up and 1,000 ft/s east or north. Its position is compared: the distance between HPR Sim’s and the code’s. Each code’s position is read from the latitude, longitude and altitude in its file, since two of the codes write no Earth-fixed position.
How to read the table. A difference is HPR Sim’s value less a code’s, the largest over the 30 s; for a position, the distance. The spread is the largest difference between any two codes. One code is left out when leaving it out cuts the spread by more than half. As the report numbers the codes, that is SIM 1 in cases 1 and 3, SIM 6 in case 2, SIM 3 in cases 6 to 8 and SIM 2 in case 10. The model difference is a named difference between HPR Sim’s model and the codes’, worked out apart from the flight. The last column is what that leaves of the largest difference from any kept code.
| Quantity | Codes kept | The codes’ spread | HPR Sim’s largest difference from a kept code (size) | Model difference | Most left after the model difference |
|---|---|---|---|---|---|
| Case 1, altitude | 5 of 6 | 0.1424 mm (0.0056 in) | 19.51 mm (0.77 in) | −19.50 mm (−0.77 in) | 0.1364 mm (0.0054 in) |
| Case 2, inertial roll rate | 4 of 5 | 0.0008991 deg/s | 0.01077 deg/s | +0.01077 deg/s | 0.0008818 deg/s |
| Case 2, inertial pitch rate | 4 of 5 | 0.001242 deg/s | 0.01633 deg/s | −0.01633 deg/s | 0.001005 deg/s |
| Case 2, inertial yaw rate | 4 of 5 | 0.0003532 deg/s | 0.005500 deg/s | +0.005500 deg/s | 0.0001268 deg/s |
| Case 3, inertial roll rate | 4 of 5 | 0.003609 deg/s | 0.004873 deg/s | −0.003755 deg/s | 0.003176 deg/s |
| Case 3, inertial pitch rate | 4 of 5 | 0.003792 deg/s | 0.005633 deg/s | −0.005609 deg/s | 0.001123 deg/s |
| Case 3, inertial yaw rate | 4 of 5 | 0.003503 deg/s | 0.005102 deg/s | +0.004433 deg/s | 0.0007793 deg/s |
| Case 6, position | 5 of 6 | 85.03 mm (3.3 in) | 102.6 mm (4.0 in) | 180.2 mm (7.1 in) | 77.57 mm (3.1 in) |
| Case 7, position | 5 of 6 | 85.24 mm (3.4 in) | 102.8 mm (4.0 in) | 180.4 mm (7.1 in) | 77.63 mm (3.1 in) |
| Case 8, position | 5 of 6 | 87.74 mm (3.5 in) | 105.0 mm (4.1 in) | 182.3 mm (7.2 in) | 77.68 mm (3.1 in) |
| Case 9, position | 6 of 6 | 1,989 mm (78 in) | 1,980 mm (78 in) | 1,253 mm (49 in) | 1,991 mm (78 in) |
| Case 10, position | 5 of 6 | 1,986 mm (78 in) | 1,977 mm (78 in) | 1,251 mm (49 in) | 1,988 mm (78 in) |
For a position the model difference is a distance, and it differs from code to code; the table gives the largest. Distances don’t subtract, so the last column need not be the difference of the two before it.
Cases 1 to 3 and 6 to 8 come out past the codes’ spread. Each is traced. A named difference between the models, computed apart from the flight, is taken off at each code’s largest difference. What is left is inside the spread (the decisions that trace them, and the later cases, the report). In cases 1 to 3 the difference explains the gap; in cases 6 to 8 it does not explain SIM 1’s, below. A stricter reading, below, leaves case 2 just past the spread. Cases 9 and 10 come out inside it as flown.
- Case 1 is the gravity model. HPR Sim’s sphere ends 19.51 mm (0.77 in) lower after a fall of 4.4 km
(2.7 mi). It falls in WGS 84’s normal gravity
(Gravity), which holds every term of the model Earth’s field. The codes fall
in a field cut off after its second term, , the one for the Earth’s bulge at the equator.
On the equator the next term, , pulls harder by about 4.4 parts in a million of the
attraction, 4.34e-5 m/s². Held for 30 s, that acceleration moves a body 19.5 mm (0.77 in),
which a test of the tool works by hand (
the_gravity_gap_is_j4s_over_the_fall). The tool integrates the exact gap along the fall and gets 19.50 mm (0.77 in). - Case 2 is the Earth’s rate. The codes turn the brick by Euler’s equations in its rate relative to the stars. HPR Sim’s rotational equations use the rate relative to the launch frame, which turns with the Earth, and add the Earth’s rate back only when an inertial rate is read out. The rates the two sets of equations turn start one Earth’s rate apart, 0.0042 deg/s, and the inertial rates read out start the same. On a tumbling body of three unequal moments the inertial rates then drift apart, by up to 0.01633 deg/s in pitch within the 30 s, 3.9 times the Earth’s rate. Integrating both sets of equations apart from the flight gives the differences in the table. What the missing term costs a rocket’s flight is not measured: that, and whether to carry the Earth’s rate in the rotational equations, is #1521.
- Case 3 is the Earth’s rate again. The same two sets of equations are damped alike, on the rate relative to the stars, along the same fall. They leave at most 0.003176, 0.001123 and 0.0007793 deg/s, each inside the spread. Read at every sample, roll and yaw stay inside. Pitch comes to the spread itself, 0.003791 against 0.003792 deg/s. Against SIM 4, which damps as the test does, at most 0.0002691 deg/s is left. Read at every sample, the most is left in roll and pitch against SIM 5 and SIM 6. Those damp the rate relative to the Earth. In yaw the most is left against SIM 2.
- Cases 6 to 8 are the gravity and the codes’ own air. HPR Sim’s sphere ends within 21 mm (0.8 in) of three codes and 102.6 to 105.0 mm (4.0 to 4.1 in) from SIM 1. The codes don’t fly the same air. Set against HPR Sim’s 1976 standard atmosphere, SIM 4’s density is within 4.1e-7 of it. SIM 1’s is up to 1.35e-3 denser: more drag, so its sphere falls less. The model difference flies a point mass twice. Once it falls in the codes’ gravity through the code’s own density, read from its file; once in HPR Sim’s gravity and atmosphere. Less that, HPR Sim sits 0.004812 to 0.01569 mm (0.00019 to 0.00062 in) from the nearest code. It sits 77.57 to 77.68 mm (3.05 to 3.06 in) from SIM 1, inside the spread. SIM 1 is brought inside, not explained. Its own air, flown, would put its sphere 180.2 mm (7.1 in) above HPR Sim’s; its file is 102.6 mm (4.0 in) above. With the gravity alone, cases 6 and 7 were left 0.01 mm (0.0004 in) past the spread, so the air is part of the trace (the decision).
- Cases 9 and 10 are inside the spread, as flown. Shot kilometers downrange, HPR Sim ends 1,980 and 1,977 mm (78 in) from the farthest kept code, SIM 3. That is inside spreads of 1,989 and 1,986 mm (78 in): the codes themselves sit about 2 m (6.6 ft) apart. HPR Sim ends 15.60 mm (0.61 in) from SIM 4 and 13.68 mm (0.54 in) from SIM 5 in both. Less the same model difference it sits 0.03830 mm (0.0015 in) and 0.04217 mm (0.0017 in) from SIM 5. But it sits 1,991 and 1,988 mm (78 in) from SIM 3, just past the spread, so no cause is claimed for these cases. Here the launch frame’s flat axes carry the body far from the pad. So the cases hold the conversion that gives apogee and landing points, to the codes’ own agreement (the report).
A stricter reading. The last column takes the model difference at the moment of each code’s largest difference. Taken at every sample of the 30 s instead, what is left is 0.1369 mm (0.0054 in) in case 1, inside the spread, and 0.0008993, 0.001242 and 0.0003533 deg/s in case 2: over the spread by 0.0000001701, 0.00000007201 and 0.00000007634 deg/s. In case 2 the spread is set by one kept code, SIM 2, against two that agree with each other, so a flight that sat exactly on those two would differ from SIM 2 by the spread and no less. HPR Sim, less the model difference, sits 0.0000005866, 0.0000009229 and 0.0000002790 deg/s from the code nearest it. That distance is the integrator’s step error at the default tolerances: at tolerances of 1e-12 it falls to 0.000000002126, 0.000000003051 and 0.000000001089 deg/s. By this reading case 2 is not inside the spread: the excess is small, named and left open. The report’s “traced” is the first reading’s.
What the comparison does not see.
- The Coriolis force’s sign, in case 1. The force carries the falling sphere east, and that eastward speed is in turn pushed up, so the fall ends higher whichever way the Earth is taken to turn. Case 1’s altitude holds the size of the term, 7.0 mm (0.28 in) of the fall, which a test pins on the free body against its closed form. It does not hold the sign. Cases 6 to 10 compare whole positions, the eastward drift with them, so they do.
- What SIM 1 does in cases 6 to 8. 77.6 mm (3.1 in) of its difference is not accounted for, only inside the spread. SIM 3, left out there, is about 171 mm (6.7 in) off.
- The attitude. The codes’ files also hold the body’s Euler angles, which are not compared (#759). An error in the attitude reaches the inertial rates only through the Earth’s rate turned into the body.
- An error the kept codes share. The codes are the reference, so it would pass unseen.
The codes’ files carry no stated terms, so they are not committed: cargo xtask check-cases
reads them from an archive that cargo xtask refs fetch downloads and checks, and the
committed report holds only differences and spreads.
Whole flights against RocketPy
cargo xtask validate flies six of RocketPy’s example rockets from the pad to the ground. It
compares fifteen numbers of each with RocketPy 1.13.0’s own flight
(ADR-021, the whole-flight comparison). Both codes fly one declared drag coefficient,
a constant of 0.5 on the same reference area. So what is compared is the equations of
motion, the motor and the air, not the drag. The rest is RocketPy’s where HPR Sim has it: its gravity
formula, standard atmosphere and frictionless rail. So are the example’s rail angles, parachutes
and motor, and a declared wind.
| result | value |
|---|---|
| cases scored | 6, one of them (Prometheus 2022) past Mach 1, since M1.8a |
| height, speed, time, acceleration | all scored, all within 3% of RocketPy’s |
| largest of those | +1.783%, Bella Lui’s peak acceleration, on the rail |
| largest in apogee | +1.208%, Prometheus 2022; RocketPy flown with HPR Sim’s body lift and rail release comes within 0.01% (case file) |
| path without wind (drift of apogee and landing) | all scored, within 1.9% (largest −1.811%, Valetudo’s landing, in still air) |
| path in wind | Calisto’s scored (largest +1.257%), and NDRT 2020’s landing; Juno III’s, Bella Lui’s and Prometheus 2022’s, and NDRT 2020’s apogee drift, differ by 4.3 to 38% and are reported, not scored: HPR Sim’s body lift and rail release, and Juno III’s fin slope (ADR-026) |
What the two codes still do differently, and how much it moves:
- The wind. A rocket that leaves the rail slowly in a wind meets the air at a steep angle. Juno III, at 18 m/s (59 ft/s) in an 8.5 m/s (28 ft/s) wind, is 26° off the airflow. There HPR Sim’s normal force includes body lift (Aerodynamics), which RocketPy’s leaves out. Much of it acts ahead of the center of mass, the nose’s above all. So it moves the center of pressure forward and weakens the turn into the wind. HPR Sim turns into it less: Juno III’s apogee is 245.3 m (805 ft) from the pad, against 396.6 m (1,301 ft) in RocketPy. HPR Sim cannot model the airfoil lift curve Juno III’s example gives its fins, so it flies a flat-plate fin slope. Given HPR Sim’s body lift, its rail release and that fin slope, RocketPy puts Juno III’s apogee 248.3 m (815 ft) out. Every windy drift then comes within 1.3% of HPR Sim’s (ADR-026). HPR Sim’s growth of drag with the angle of attack moves no drift by more than 0.1%.
- RocketPy’s equations, corrected. HPR Sim’s equations of motion follow RocketPy’s technical documentation, which measures the center of mass from the center of dry mass. RocketPy 1.13.0’s code reads that vector the other way round. So during the burn it takes the turning moments about the wrong point, and its rockets turn into the wind too far. The fix is a pull request to RocketPy, merged on November 16, 2026. It is built on an earlier one, merged too. No release carries either yet. RocketPy 1.13.0 as installed still has the error, and the comparison applies both fixes. Without them, HPR Sim’s drifts in wind were up to −60.8% short of RocketPy’s at apogee and +151% beyond it at landing (ADR-026, issue #50).
- The rail. HPR Sim’s rail equation keeps the terms for the center of mass moving inside the
body as the propellant burns. RocketPy’s rail equation (
udot_rail1) leaves them out. At a sharp ignition spike, with thrust and mass the same to five digits, HPR Sim’s acceleration is 1.2 to 1.3 m/s² higher. HPR Sim also guides the rocket until its last rail button leaves, where RocketPy frees it at the first. - Calisto’s two peaks. Its acceleration peaks twice, 0.9% apart, and the rail terms make HPR Sim’s maximum the other peak. So the time of the peak (−96.8%) is reported but not scored.
- The parachutes. RocketPy counts the air a canopy drags along (added mass). HPR Sim has none. So the peak deceleration as NDRT 2020’s main opens is 83% higher in HPR Sim, and that number is reported but not scored. The speeds at landing agree within 0.03%.
The metrics are measured as RocketPy defines them. Speeds and accelerations are at the center of
dry mass. Heights are from that point’s height at launch, since RocketPy’s starts at the ground.
The rail exit comes when the rocket has traveled RocketPy’s effective_1rl, the forward button at
the top of the rail. HPR Sim’s own rail-exit event waits for the last one. Each case file argues
its tolerances. Accuracy gives every result.
What leaving tip-off out costs
HPR Sim flies without tip-off unless a flight asks for it (Tip-off). RocketPy’s tip-off phase puts the cost at up to 3.765% of the apogee in the landing point and 0.414% in the apogee. RocketPy flew thirteen example designs. Both are measured from a rocket held straight to its last guide, as HPR Sim holds it. The landing point moves farther from the pad in most flights, and the largest apogee move is down. That is one code against another: the phase is in RocketPy’s development branch, in no release, and no flight checks it. The landing point’s cost is past the plan’s bar of 1% of the apogee, so HPR Sim gained a model of the pivot. It is held to RocketPy’s on RocketPy’s normal force and in still air, not yet in a crosswind on HPR Sim’s own (below).
Tip-off is the turn a rocket picks up between its guides leaving the rail. Once the forward guide is off, the last one holds the rocket alone, and gravity and the wind turn it about that guide. By default HPR Sim keeps the rocket straight until the last guide leaves, then frees it with no turning rate.
cargo xtask tip-off flies the comparison and writes the report
(the decision record):
- What flies. RocketPy, at the merge of its
pull request #1217 (
65b9838f), flies the thirteen public designs built from its examples and tests, on its own inputs. They are twelve rockets; Calisto comes twice, from RocketPy’s getting-started notebook and from its tests. A Calisto id in the report ends in the motor’s position in meters, in the rocket’s own coordinates as its source sets it. Each source measures from its own zero point, and the two zero points are 0.118 m (4.6 in) apart. So in both the nozzle sits 2.533 m (8.31 ft) behind the nose’s tip. The fins differ, and the notebook’s rail buttons and its center of mass without the motor sit 0.118 m (4.6 in) farther back. Each design flies from its example’s rail on a constant drag coefficient of 0.5. It flies in still air and in a 20 mph (8.94 m/s) wind square across the rail’s heading. - Three ways. With the phase off, RocketPy frees the rocket where its forward button leaves the rail. With it on, the rocket slides on, free to pitch and yaw about its last button, until that button leaves too. Guided, the phase is off on a rail longer by the distance between the buttons. That holds the rocket straight to where its last button leaves the real rail, as HPR Sim holds it.
- What HPR Sim’s release leaves out is the phase on less the guided flight.
- The apogee moves by at most −0.414%: −10.9 m (−36 ft) for Juno III in still air. None of the 26 apogees moves by 1%.
- The landing point moves by more than 1% of the apogee in 6 of the 26 flights. They are on three rockets: Valetudo, Juno III and Prometheus 2022. The largest share is 3.765% of the apogee, 28.5 m (93 ft) for Valetudo in the crosswind. The largest distance is 92.0 m (302 ft), for Juno III in the crosswind.
- With the phase on, the rocket lands farther from the pad in 23 of the 26 flights. One of them, the vertical rail’s in still air, moves by 6 mm (0.2 in), which the report’s table rounds to zero. In those 23, a rocket guided to its last button lands nearer the pad than one that pivots. For a landing-area estimate that understates how far the rocket travels: the flattering side. In the other three, all in the crosswind, the guided rocket lands 0.3 to 3.6 m (1 to 12 ft) farther from the pad.
- The pivot itself is small. With the phase on, a rocket leaves its last button turning at up to 2.38 deg/s, its axis at most 0.083° off the rail’s.
- The plan’s rule set the phase on against the phase off. That moves 9 of the 26 landing points by more than 1% of the apogee, by up to 4.585%, and one apogee by +1.165%. Part of that is RocketPy’s early release and not the pivot. Bella Lui, freed at its forward button in the crosswind, has turned into the wind at 15.78 deg/s by the time its last button leaves. With the phase on it turns at 0.60 deg/s.
- A worked example. Juno III leaves a 5.20 m (17.1 ft) rail at 85°, its buttons 1.410 m (4.63 ft) apart. In still air its last button leaves at 21.53 m/s (70.6 ft/s), 0.071 s after the forward one. With the phase on, gravity has it turning nose-down at 2.26 deg/s by then, its axis 0.079° off the rail’s line; guided, it isn’t turning. Guided, it reaches 2,623.9 m (8,609 ft). With the phase on it reaches 2,613.1 m (8,573 ft) and lands 91.3 m (300 ft) farther from the pad. Each figure is rounded on its own; unrounded, the apogee moves down by 10.86 m (35.6 ft).
- The reference moved once. The report flies one fixed commit of RocketPy’s source. At
first that was the merge of RocketPy’s
pull request #920, which added the
phase. RocketPy then rewrote the phase ([RP-TIP] says what changed), and the report now
flies the rewrite (the decision record). No figure quoted on this page changed.
Between the two commits, a turning rate where the last button leaves moved by at most
0.000020 deg/s, and no flight’s apogee or landing point by more than 1.04 mm (0.041 in).
Each of the four differences HPR Sim’s phase is held to is under a hundredth of the bar it
is held to. The flights without the phase moved as far as those with it, so these are
differences between two commits of RocketPy, not the measured effect of the rewrite: the
report does not test which change made each. Its last table gives each flight, and the
earlier report’s data is kept as
validation/fixtures/flight/rocketpy-tip-off-at-pull-920.json. - What the comparison leaves out. RocketPy’s pivot is a frictionless pin with roll held; a real button has play in its slot. The guided flight frees the rocket at its last button’s position, and HPR Sim half a button’s length later, at the button’s aft edge. The thirteen fly one drag coefficient, and a wind that does not change with height. One rocket’s still-air flights open the apogee parachute with no lag, and the report says which and why.
The tip-off phase against RocketPy’s
HPR Sim’s tip-off phase is inside every bar below on all 26 flights when both codes fly RocketPy’s normal force. On HPR Sim’s own normal force it is on the 13 still-air flights and not on the 13 crosswind flights. That is one code against another, with no flight to check either.
cargo xtask tip-off-model flies the comparison and writes the report. The
gate and CI hold it (--check). The decision record gives what is
held and what stays open.
- What flies. The same thirteen designs, rails, sites, winds, parachutes and drag coefficient as RocketPy’s flights above. HPR Sim flies each guided, as it does by default, and with the phase. A design whose dry mass, reference area, motor or distance between its last two guides differs from what RocketPy flew is refused.
- Twice over. On RocketPy’s normal force, HPR Sim flies the force RocketPy flew. That is RocketPy’s slope at the Mach number times the angle of attack, at RocketPy’s center of pressure, read from RocketPy with each flight. The phase then meets one drag and one normal force in both codes. Damping in pitch and yaw stays each code’s own. On its own, HPR Sim flies its own aerodynamics.
- The bars are the tip-off decision record’s. Where the last guide leaves, the turning rate is within 10% of RocketPy’s or 0.1 deg/s, whichever is larger. The angle off the rail is within 10% or 0.005°. Each is compared as a pair. Nose-down is below the rail’s line in the rail’s vertical plane. Nose-right is to the right of the rail’s heading, which is downwind in the crosswind. The two codes’ apogee moves are within 0.2% of RocketPy’s guided apogee, and their landing-point moves, as vectors, within 1% of it.
- On RocketPy’s normal force, 26 of 26 flights are inside every bar. The largest difference in the turning rate is 0.003 deg/s against a bar of 0.238 deg/s, and in the angle 0.0001° against 0.0083°. In the apogee move it is 0.002% of the apogee, and in the landing-point move 0.029%.
- On its own normal force, 13 of 26 are: all 13 in still air, none in the crosswind.
- All 13 crosswind flights are past the turning rate’s bar, by up to 4.010 deg/s against a bar of 0.100 deg/s. 10 are past the angle’s, by up to 0.0671°. 3 are past the apogee move’s, by up to 0.379% of the apogee. 4 are past the landing-point move’s, by up to 4.082%.
- The cause is the normal force. In the crosswind the air meets the rocket at 9.7° to 25.6° as the last guide leaves. There HPR Sim’s own normal-force coefficient is 1.23 to 2.71 times RocketPy’s, and its moment about the last guide 0.35 to 15.20 times. The turn follows the moment. On 10 of the 13 the force acts ahead of the guide, and the nose turns downwind, farther on HPR Sim’s own. On 3 it acts behind the guide, and the nose turns upwind, less far on HPR Sim’s own. In signed terms the rate is farther nose-right on all 13, by 0.136 to 4.011 deg/s.
- Which force is nearer a real rocket’s at those angles, no reference here says (issue #1538). HPR Sim’s adds body lift, which RocketPy leaves out. Holding these 13 to the bars is an open increment.
- Which way leaving the phase out errs. With the phase, HPR Sim lands farther from the pad in 22 of the 26 flights on RocketPy’s normal force, as RocketPy does. On its own it does in 20. A move of under 0.1 m (0.33 ft) either way is not counted here. The count of 23 under What leaving tip-off out costs includes the vertical rail’s still-air flight, which moves 6 mm (0.2 in). For a landing-area estimate, a flight without the phase lands nearer the pad than one with it in most of these.
- A worked example. Juno III in still air, on HPR Sim’s own normal force, leaves its last guide at 21.5 m/s (70.7 ft/s), turning nose-down at 2.254 deg/s. RocketPy’s turns at 2.255 deg/s. Its axis is 0.0788° below the rail’s line, and RocketPy’s 0.0789°. The phase lowers HPR Sim’s apogee by 11.33 m (37.2 ft) and RocketPy’s by 10.86 m (35.6 ft). It lands HPR Sim’s rocket 95.4 m (313 ft) farther from the pad, and RocketPy’s 91.3 m (300 ft).
- What the comparison leaves out. Everything the comparison above does: one drag coefficient, a wind that does not change with height, and a frictionless rail. It also takes RocketPy’s normal force as RocketPy states it for a rocket at rest in the air stream. In flight RocketPy works each surface’s angle from the air’s speed at that surface. And in the crosswind the rocket leaves the rail at 9.7° to 25.6° to the air. Both codes carry small-angle models there, and 7 of the 13 are past the 15° the accuracy work assumes.
The tip-off phase on the real flights
This compares HPR Sim with itself, the phase on against the phase off; no flight’s log checks the phase. It moves each real flight’s apogee by under 1%, and each set’s mean error against the logs by a tenth of a percentage point or less. The compared first flights are the seven public flights compared as soon as they were flown; the predicted first are the five whose predictions were published before their comparison. On the seven compared first the phase brings the mean absolute apogee error down from 6.065% to 5.974%. On the five predicted first, and on the private collection’s 60 logged flights, it raises it by 0.0014 and 0.0033 percentage points. No real flight’s landing point is compared yet, so this measures apogees only. The phase is still off unless a flight asks for it; the next increment may make it the default.
cargo xtask tip-off-flights writes the report (the decision
record); the gate holds it (--check).
- What flies. Every real flight HPR Sim compares, each as its own report flies it: the twelve public flights of the real-flight report, and the private collection’s flights, given in aggregate only. Each flies guided, held straight to its last guide and then freed, as by default, and with the phase. A design needs two guides (rail buttons or launch lugs) at different stations for the phase to move it: 10 of the 12 public flights have them, and 23 of the collection’s 60. Each flies again in still air: the same air with its wind taken out, so that what the phase moves there is gravity’s turn about the last guide alone.
- Gravity lowers; the wind’s part is open. In still air the phase lowers 9 of the 10 public flights with two guides by more than 1 cm (0.4 in). The tenth, NDRT 2020, flies from a vertical rail, where gravity has no moment about the last guide; it moves 1 mm (0.04 in). On a tilted rail gravity turns the rocket nose-down about its last guide, so it leaves pointing lower. In the day’s wind the phase raises 2 of the 10 public apogees. That wind’s part is the part of the phase not yet held to RocketPy’s on HPR Sim’s own normal force; on Juno III in that comparison’s crosswind, HPR Sim’s phase raises the apogee and RocketPy’s lowers it.
- Why two errors rise. Of the predicted flights, the phase lowers Erebus 11, which HPR Sim already flies 8.05% low. On the collection it raises the mean signed error from +8.007% to +8.015%, on a set HPR Sim flies 8.0% high on average, and moves 14 of the 23 flights with two guides away from their logs and 9 toward them. How many it raises is held back, so that no figure gives a few flights’ own moves. Both rises are smaller than the hundredth of a point the reports print. The decision record traces each to its flights and lets the next increment make the phase the default; the fall over the compared-first flights rests on Juno III’s rise, so it is not counted for the phase.
- A worked example. Juno III’s log reads 3,151.5 m (10,340 ft). Guided, HPR Sim’s apogee reads 2,950.2 m (9,679 ft), 6.39% low. With the phase it reads 2,976.7 m (9,766 ft), 5.55% low: 26.5 m (87 ft) higher. In still air the phase lowers it by 16.7 m (55 ft). That is more than the 11.33 m (37.2 ft) of the comparison with RocketPy, which flies Juno III on one drag coefficient, not on the real-flight report’s drag and launch-day air; how much of the gap that explains is not measured. Taking out the wind changes the guided flight too, so the still-air and wind moves are not two parts of one sum.
- What it leaves out. The phase moves apogees by under 1%, against mean absolute errors of 6% to
13%, so no flight checks it. Landing points, where RocketPy’s
flights show the larger cost, wait for landing points against
the tracked landings. Two tests on the Valetudo pin which way
a landing point errs without the phase; the direction depends on the rail and the wind. On a 5.3°
tilted rail in still air the guided flight lands about 51 m (167 ft) short, held between 50.5 m
(166 ft) and 52 m (171 ft):
without_tip_off_a_landing_point_errs_nearer_the_pad_on_a_tilted_rail_in_still_air. On a vertical rail in a 5 m/s (11 mph) crosswind it lands about 26 m (85 ft) farther from the pad, held between 25 m (82 ft) and 26.5 m (87 ft):without_tip_off_a_landing_point_errs_farther_from_the_pad_on_a_vertical_rail_in_a_crosswind. That second direction comes from the wind’s part of the turn, the part not yet held to RocketPy’s on HPR Sim’s own normal force.