Fun With Science / Globe Deconstruction / Rockets in Vacuum
Answering Shape Debate's Claim #2 — that Newton's third law can't apply to a rocket in vacuum, because thrust needs something to push off.
The burning gas pushes on the inside of the engine. The back of the engine has a hole in it.
Claim #2 on shapedebate.com, quoted below, points to a pending vacuum-chamber test by Alex Kampf, “Propulsion Tester,” results not yet uploaded. Newton's third law is a statement about how mass affects mass; for exhaust gas to be exempt, gas would have to have no mass. This page builds that up from first principles, checks it against a public webcast, and shows the same result at three unrelated scales.
Newton intactClaim does not follow
Air does act on a rocket — the wrong way
The claim: thrust needs a medium to push against, so a rocket cannot work in vacuum. The answer: that describes a propeller or a jet engine, not a rocket. A rocket's third-law pair is the rocket and its own exhaust — mass it carries and throws overboard — not the surrounding air. Thrust is chamber pressure (about 97 bar in a Merlin 1D) pushing on the closed front of the engine; the air's only contribution is ambient pressure bearing inward on the outside, a loss. That is why the same Falcon 9 first stage is rated 7,607 kN at sea level and 8,227 kN in vacuum: thrust goes up, not to zero, as the air thins. A $15 garage load cell, a combat-verified missile intercept and a sixty-year-old vacuum chamber trace the same curve; engines measured firing in real space, several physically unrelated ways at once, agree to a few per cent.
What the claim gets right: the chamber-size caution on its own page. A chamber small relative to thrust and burn duration re-pressurizes from its own exhaust faster than it can be pumped down, and thrust can then look as if it failed in vacuum when what failed was the vacuum. Chamber size has a second face, which neither side has raised: a chamber the exhaust crosses almost instantly makes any timing comparison uninformative, whichever model is right. Numbers for both are in §7 and §8. The referenced experiment is pending upload, so this page argues from the claim as stated and from data already public.
A sealed pressure vessel goes nowhere: gas presses outward on every wall and the pushes cancel. Cut a hole in one end and the cancellation stops. Pressure still bears on the closed forward wall of the combustion chamber; across the open end there is nothing to bear on, so nothing pushes back. That unbalanced push on the vehicle's own structure is the thrust, and it is present before a single molecule leaves the nozzle. What the exhaust does afterwards — whether it meets air, or a wall, or nothing at all — is no longer the rocket's business.
Newton's third law isn't a law about air. It's a law about how mass affects mass.
Take a large block sitting still. Bolt a machine to one edge that continuously throws off chunks of the block's own mass — say 1% at a time — sideways, at a fixed speed. The moment the first chunk leaves, the block moves the other way: momentum balanced at zero before the machine fired and has to go on balancing. Nothing in that story needs air, a wall, or anything nearby. Keep the machine running and F = ma gives the rest: constant force on a shrinking mass, so acceleration climbs the whole time — slowly at first, sharply near the end. The only way a gas exhaust could be special is if gas had no mass. A kilogram of combustion gas leaving a nozzle at 3,000 m/s carries exactly the momentum of a one-kilogram block thrown at 3,000 m/s.
Toy units, not real rocket numbers — the point is the shape of the curve. Watch the acceleration figure climb as the remaining mass (the shrinking square) burns down.
Put the same block-and-machine inside a column of air. It must now shove air out of its way — that is drag, a real loss, but a loss on the vehicle, not a reduction in thrust, and it shrinks as the air thins. Drag is not why a nozzle is rated for more thrust in vacuum. The real reason cuts against the claim.
For the technically inclined: the equation engineers use to size a nozzle is F = ṁ·Ve + Ae(pe − p0) (NASA Glenn, Thrust Equation). The first term is momentum thrust, medium-independent — §2's result. The second is the pressure term above, and as p0 falls it grows: removing the atmosphere entirely adds exactly patmAe, whatever the motor does internally (§7 works the number for small motors). Nothing in it refers to what the exhaust meets after it leaves.
Same machine, same throwing rate, climbing through thinning air (dark = dense, clear = thin). The percentage compares achieved acceleration against the drag-free ideal of §2 — drag loss on the vehicle, not thrust efficiency, which does not change. It climbs toward 100% as the air thins.
The chart below is the §2 demo at real Falcon 9 Block 5 numbers: 9× Merlin 1D, sea-level thrust 7,607 kN, vacuum thrust 8,227 kN, ~411 t propellant, main engine cutoff (MECO, the end of the first-stage burn) at ~162 s, run against the U.S. Standard Atmosphere 1976 density model. It is a simulation, plotted against a hypothetical “air pusher” in which thrust is simply scaled by ambient air density — the literal version of “thrust needs a medium” — through the identical fuel burn and mass-loss schedule, so the only thing that differs is what creates the thrust.
The real curve climbs all the way to MECO; its one flattening, between T+50s and T+82s, is the Max‑Q throttle-down (peak aerodynamic load), a schedule fed into the model (method notes). The “air pusher” has already fallen to roughly 60–65% of sea-level thrust by T+50s, a few kilometres up. Speed is where the two predictions diverge most plainly:
The modelled vehicle keeps building speed to MECO at T+2:42, leaving the window at roughly 3,050 m/s in this run, on the way to orbital velocity (~7,800 m/s) on the second stage. The “air pusher” peaks at about 7.9 km around T+1:35, sinks, burns out at the same T+2:42, and about seventeen seconds later — around T+2:59 — is back at sea level at over 1,000 m/s, its entire propellant load spent without approaching orbital speed. That is what “thrust needs a medium” predicts once you run it instead of asserting it, and it is not what any of the thousands of public orbital launches on record has shown.
None of the inputs has to be taken on faith. SpaceX webcasts every launch with an on-screen speed/altitude readout (this Falcon 9 Starlink launch, for example), archived by mission name on SpaceX's own YouTube channel. Acceleration is the rate those readouts change, so it can be backed out with no claimed thrust or propellant figure anywhere in the calculation. Enthusiasts on the NASASpaceflight.com forum hand-log the readouts frame by frame, and Flight Club, a trajectory simulator built and run by a private individual with no SpaceX affiliation, integrates thrust, drag and gravity itself. Both land on the climbing shape above.
“No medium to push against” is not an exotic condition unique to a vacuum chamber or deep space; it is the normal state of the sky well below where second-stage burns and orbital insertion happen. The chart extends the same atmospheric model past the Kármán line (100 km, the conventional edge of space) to ISS altitude (~423 km).
Pressure at 80 km is already about 0.001% of sea level (1.05 Pa); at the Kármán line, 0.032 Pa — roughly three ten-millionths of sea level, a good laboratory vacuum rather than anything resembling interplanetary space, which is more than ten orders of magnitude lower again. NASA Glenn's VF-5 chamber (§6) at 10−7 torr (a torr is 1/760 of an atmosphere) is a harder vacuum than the Kármán line by three orders of magnitude. If thrust depended on ambient air it would have failed by the time a Falcon 9 second stage lights, tens of kilometres below where anyone claims spacecraft do their real work.
The usual reply is that the Kármán line is a fiction and that upper stages are never filmed from outside. Neither helps. The line is only a name for an altitude; the pressure at 80 km is a property of air, and the argument goes through there whatever the line is called: on the air-pusher model thrust at 80 km is 0.001% of its sea-level value, on the standard model it is above the sea-level value, on its way to 8,227 kN. Whether space-agency footage is authentic is a question this review does not assess, and nothing here depends on it: the intercepts in §6 are tracked from the ground by more than one country, the altitude cells are on the ground, and the 1916 chamber was in a laboratory.
Garage scale. Set NASA aside. Amateur rocketry has measured motor thrust on load cells for decades. A sea-level static fire does not settle the vacuum question — nobody disputes that rockets work in air — but it settles the momentum budget: take any motor's total impulse (thrust summed over the whole burn) from ThrustCurve.org (an independent, community-run database of over a thousand certified motor curves, maintained by a private individual), divide by propellant mass, and you get an effective exhaust velocity of roughly 1.8–2.3 km/s for ammonium-perchlorate composite propellant — what combustion chemistry and nozzle theory predict from the exhaust alone. The band is wide, so this rules out the atmosphere supplying a large share of the thrust, which is what the claim requires, not a few per cent; and the measured figure sits at the low end of theory, the wrong direction for the claim, since at sea level the ambient term of §3 is subtracting. Richard Nakka's DIY strain-gauge load cell puts the sensor at about $10–20 in parts and gives a directly measured thrust-vs-time curve on a home stand.
Bench scale, without combustion. The live claim names “a pressurized gas release,” and that is exactly what a cold-gas thruster is: a tank of nitrogen or butane vented through a nozzle, no flame, flown for attitude control on cubesats built by universities and private companies with no space-agency involvement, and sized with the §3 equation. A cubesat that could not turn would be noticed by its owner. Cheaper still is the school-lab bell-jar version — a balloon-driven or compressed-air cart in a bell jar under a pump — where the one question a careful run must answer is the one conceded at the top: how far the released gas raises the jar's pressure (§8 gives numbers).
Battlefield scale. Publicly documented missile intercepts show the same physics under combat conditions: Arrow-2 interceptors reaching roughly 50 km altitude by design; Arrow-3, exoatmospheric ceiling over 100 km, used repeatedly since its first combat interception in November 2023, including during the April 2024 Iranian attack; and a first combat-verified THAAD intercept in December 2024, a system rated to 150 km. Each is cross-checked by the intercepting country, the country that launched the threat, and outside observers, and catalogued by the non-governmental CSIS Missile Defense Project.
Only then NASA, as one more point on a curve already traced. Glenn's Vacuum Facility 5 is 4.6 m in diameter and 18.3 m long, pumped to roughly 1×10−7 torr, sized so a firing thruster's own exhaust cannot re-pressurize it. Ion and Hall-effect thrusters have been hot-fired there since the 1960s, for thousands of hours with thrust read on a stand; the NEXT ion engine ran over 48,000 hours, roughly five and a half years of continuous firing. If the objection is that those are not gas rockets — though the book files them under this claim at p. 183 — chemical upper-stage engines are fired in altitude cells outside NASA too: DLR Lampoldshausen's P4.1 for the Vinci engine, the USAF Arnold Engineering Development Complex J-4 and J-6 cells, and JAXA Kakuda, with thrust read at the engine mount and cell pressure logged, upstream of any diffuser (the exhaust duct that keeps the cell pumped down while the engine runs).
Every source above except the intercepts is measured on the ground, where a chamber can always be accused of not being empty enough. Engines have also been measured firing in real space, several unrelated ways at once: a 1970s ion thruster's thrust was determined from an onboard accelerometer, from the change in its orbit, and from beam telemetry calibrated on a ground stand, and the three agree to a few per cent. That record is on its own page: Thrust, Measured in Flight. The chamber artefact is real, is about one per cent, and runs in the direction that makes chambers flatter engines slightly — two orders of magnitude from what the claim needs.
Eight years before the book, a YouTube channel called Warped built a vacuum chamber for exactly this purpose and fired real solid rocket motors inside it (Rockets in a Vacuum Chamber — Newton’s Third Law of Motion Visualized, 17 February 2018, about sixteen minutes). He says at the outset that flat-earth material had got into his head and he wanted to challenge the third law properly. A solid motor ejects gas; one burning in vacuum with visible thrust is a direct counter-instance to Kampf's Law, and it stood for eight years before the claim was printed.
Three of its five tests failed, and the creator says so. Attempts one, two and three did not ignite — reviewing the first, he says there is some truth to what the conspiracy theorists think, about ignition specifically. The fix on the fourth was a rupture disk over the nozzle, holding a pocket of atmospheric pressure inside the casing until it lit; read uncharitably, he put the air back. But ignition and thrust are different problems. Solid propellant needs chamber pressure to light and spread flame across the grain; near vacuum the igniter's gases expand away first. Flight motors meant to ignite in space carry nozzle closures — a diaphragm sealing the nozzle until burn-through — for exactly this reason. The disk seals the inside of the motor and does nothing about the vacuum outside; if ambient gas produced thrust, no closure would help.
He gives no thrust figure, and notes as a curiosity that the main flash seemed to happen out in the chamber. Frame by frame, the fourth test's high-speed sequence shows a small orange jet at the nozzle throat extending downstream over fifty milliseconds as a narrow plume anchored at the nozzle, with the chamber-filling fireball arriving about a hundred milliseconds later, well downstream — a plume expanding with no back-pressure to confine it, expected rather than measured. The motor sits in line with a Pelouze spring dial reading to 50, so the rig carries a thrust scale; whether the needle deflects we could not establish — our attempt to measure its angle kept locking onto the orange igniter lead crossing the shot, and three frames from a slow-motion replay cannot rule either way. We claim nothing about what the scale read, and anyone citing this video for a thrust value is going beyond what it shows.
The chamber also stops being a vacuum during the burn: 40 g of propellant at a mean molar mass near 25 g/mol, in about 0.2 m³, releases around 1.6 moles — 0.2 atm cooled, roughly half an atmosphere hot, order-of-magnitude figures from estimated dimensions. Only the first moments of a burn are a vacuum test, which is the interval the creator points at when he says thrust was produced at initial ignition. In the p. 80 rig this confounds wall proximity with pressure rise, since moving the wall changes the free volume behind it; separating them needs a much larger volume or a pressure trace during the run.
The quantitative point runs the opposite way from the claim. Ambient pressure enters the thrust equation with a minus sign, so removing the atmosphere adds patmAe:
| Nozzle exit | Exit area | Thrust gained in vacuum |
|---|---|---|
| model motor, 10 mm | 7.85×10−5 m² | 7.96 N (811 gf) |
| small solid motor, 20 mm | 3.14×10−4 m² | 31.8 N (3.2 kgf) |
An Estes D12 averages about 12 N of thrust at sea level; in vacuum it gains roughly 8 N, about two-thirds more. This is why vacuum specific impulse (thrust per unit of propellant flow) is quoted higher than sea-level for every engine ever flown, and why upper stages use expansion ratios — nozzle exit area over throat area — that would be useless in air. The sceptical prediction is not merely wrong in sign; the effect it gets backwards is a routine engineering number.
The book does not treat Claim #2 as pending: p. 82 names Kampf's Law as a law, and p. 183 says the experiment “conclusively proved these animations are nothing more than CGI.” The design on p. 80 is a syringe on a linear track inside a tube, with the end wall at two distances, scoring “a more delayed reaction… as we increase the length of the chamber” as the discriminator; the run itself is not in the book. Here is what standard physics expects, set down in advance so the result can be judged against it rather than interpreted afterwards. The book gives no volume, gas, pressure, mass or exhaust speed, so the 60 mL of air, 50 g carriage and few-hundred-metres-per-second exhaust below are illustrative values of the right order, not his; every prediction scales with them.
The timing argument. p. 20 applies the same reading to The Action Lab's vacuum-chamber launch: the vehicle starts moving when the exhaust reaches the far wall. That video exists, so we measured it. The delay is real: the exhaust reaches the wall about two seconds of playback before the syringe moves. But the ordering carries no information — a push too small to see and a push not yet arrived look identical on video — and the footage rules out the wall as the cause: the gap between contact and motion is 25 to 220 times too slow for a pressure wave, the emission history is featureless so returning gas has nothing to copy, and the acceleration answers the propellant flare at the nozzle within one or two frames. Frames, arithmetic and caveats: The Action Lab Footage, Frame by Frame.
In a tube the ordering is not just uninformative but forced. The gas moves visibly and fast while the vehicle appears to do nothing; that asymmetry is the third law displayed, not violated, because equal and opposite refers to momentum, not speed. The ratio of speeds is the inverse ratio of masses — 73 mg against 50 g is 685 to 1, so exhaust leaving at 500 m/s leaves the vehicle at 0.73 m/s, and only at the end of the burn. At 0.05 N on 50 g the carriage needs about 45 milliseconds to travel a millimetre, the first displacement that reads as motion on video; in those 45 ms an exhaust front at 500 m/s covers 22 metres. The vehicle has moved half a micron when the cloud reaches a wall at 0.5 m, 8 microns at 2 m, 0.2 mm at 10 m.
Run right, the timing test is a good one. The two models disagree sharply about the onset of motion: standard physics says onset is set by thrust and friction and is independent of chamber length — the same 45 ms at half a metre and at two — while the p. 82 mechanism gates it on gas compressing at the wall and pushing back, a round trip that scales with length: 2 ms against 8 ms across that pair of distances at 500 m/s. Double the tube, double the delay: a clean, pre-registered, falsifiable difference that does not need a 22-metre chamber. It needs the right instrument, and that is not a camera, which measures displacement — the double integral of a force still building — and in a tube even at 2,000 fps sees a fraction of a pixel by the time the gas lands. A force transducer (a piezo or strain-gauge load cell between vehicle and mount, kilohertz bandwidth) registers the onset promptly, and a second on the target wall puts both events on one clock. Two clearly different wall distances, pressure logged, a force trace: an afternoon's work.
A new, independently run vacuum-chamber test is worth doing. Whether the Kampf test lands as evidence either way turns on whether the chamber and pump-down are large enough, relative to thrust and burn duration, that chamber pressure stays low throughout the firing — the sizing problem the facilities in §6 were built to avoid. Thrust holding steady while the chamber stays at a documented low pressure is consistent with standard physics; thrust collapsing while chamber pressure is simultaneously rising is a chamber-sizing result, not a third-law result. If the claimed evidence is timing, the onset has to be read from a force trace at two wall distances in a chamber the exhaust does not cross in a frame; an onset that scales with tube length on such a trace, with pressure logged and flat, is the result this page does not expect. The caveat on our own side is in §7: we claim nothing about what Warped's scale read.
In 1919 Robert Goddard published A Method of Reaching Extreme Altitudes, having already — in 1916 — fired rockets inside evacuated chambers, precisely to test whether they needed air to push against. They did not; they worked better without it, for the pressure-term reason in §3. On 13 January 1920 the New York Times ran an editorial saying that Goddard “does not know the relation of action to reaction, and of the need to have something better than a vacuum against which to react.” That is Claim #2, in a national newspaper, word for word, a century early.
The Falcon 9 charts. The plotted quantity is net acceleration: thrust less drag and the gravity component along the flight path, divided by the falling vehicle mass. The real curve runs only through T+162s (MECO), the window where the webcast overlay and independent frame-by-frame trackers are both watching the same first-stage burn; second-stage flight is a different engine in a different regime, with the same physics, and is not part of the comparison. The flattening between roughly T+50s and T+82s, where the simulated acceleration plateaus around 1.4–1.9g, is Max‑Q — peak dynamic pressure ½ρv², which builds as the vehicle accelerates and falls away as the air thins, topping out around 11–13 km; Falcon 9 is deliberately throttled down through it to limit airframe loads. That plateau is a throttle schedule fed into the model, not an emergent result, and it begins only after the air-pusher curve has already collapsed (§4). The modelled MECO speed of roughly 3,050 m/s is higher than a webcast shows because the simulation spends the full propellant load while a real booster reserves ten to fifteen per cent for boostback and landing, putting actual Starlink MECO nearer 2.2–2.5 km/s.
The air-pusher trajectory. After its altitude peaks, its total speed keeps climbing, for a reason that is the model's core flaw playing out a second time: descending into denser air means more thrust in a thrust-follows-density model. By the time the shared pitch-over program (the same one steering the real ascent) has tipped that thrust to about 15° off horizontal, the vehicle is diving through dense low air at up to roughly 75% of sea-level thrust, pouring nearly all of it into horizontal speed. That stops only when the shared propellant load runs out, at the identical instant as the real MECO because both curves share one fuel schedule; freefall and drag then take over.
The Warped chamber estimate in §7 (40 g, 25 g/mol, 0.2 m³) uses dimensions estimated off the video, not measured; the 1.6 moles, 0.2 atm cold and roughly half an atmosphere hot follow from the ideal gas law. The frame timings (fifty milliseconds for the jet to extend, about a hundred more before the fireball) are read from the on-screen high-speed clock.