Chapter 22: Staging, Propellant Management, and Reusability

"If one can figure out how to effectively reuse rockets just like airplanes, the cost of access to space will be reduced by as much as a factor of a hundred." — Elon Musk (widely quoted)

Overview

Every chapter of this book has been, in one way or another, a response to a single exponential. Back in Chapter 3 we met the tyranny of the rocket equation and its first great escape hatch — staging, the trick of throwing away empty mass so that later engines do not have to drag it to orbit. We proved, on a real Falcon 9, that two stages reach orbit where one cannot. That was the intuition. This chapter is where staging grows up.

Here we answer the questions Chapter 3 deferred. How many stages should a rocket have, and how should you divide the job between them so the whole vehicle is as light as possible? What actually goes wrong inside a half-empty propellant tank — the sloshing, the bouncing, the boiling — and what hardware quiets it? What happens when an engine dies mid-ascent, and why does a rocket with nine engines sometimes shrug that off? And then the question that is rewriting the economics of the whole enterprise: once the rocket has done its job, why throw it away? We will engineer a booster that flies itself back down through the atmosphere and lands on its tail, and we will do the arithmetic that decides whether doing so actually saves money.

This is an advanced chapter, and it earns the label by making several ideas interact at once — optimization, fluid dynamics, control, and economics all bearing on the same vehicle. But the spine is still the rocket equation. Optimal staging is the rocket equation asked to give up the least mass; propellant management is what it takes to actually burn the propellant the equation assumes you have; reusability is the rocket equation surrendering a slice of its precious mass ratio in exchange for something the equation never accounted for — a rocket you get to keep.

In this chapter, you will learn to:

• Show, with numbers, why the structural-coefficient ceiling forces staging, and why the payoff from each added stage shrinks until two or three is the sweet spot.
• Split a delta-v budget between stages to minimize lift-off mass, and prove that equal stages should do equal work.
• Name and explain ullage, slosh, POGO, and boiloff — the four ways a tank of propellant misbehaves.
• Quantify engine-out capability and tank pressurization with thrust-to-weight and pump physics.
• Walk the Falcon 9 landing sequence and explain the "hoverslam," then judge when reuse pays for itself.

Learning Paths

🚀 Space Enthusiast: Read 22.1 for the "why," then go straight to 22.5 and 22.6 — propulsive landing and the economics of reuse are the two ideas that will change how you watch a Falcon 9 come home. Skim the optimization algebra in 22.2; keep the result (equal stages do equal work) and the diminishing-returns table.

📐 Engineering Student: This is a core chapter. Do the optimal-staging derivation in 22.2 yourself, then work every ⭐⭐/⭐⭐⭐ exercise. The stage-sizing formula and the payload-fraction bookkeeping here are the tools you will use to size your own vehicle in Chapter 29.

🎮 KSP Player: You have felt every idea in 22.1–22.2 the hard way — asparagus staging is parallel staging, and your lander's inability to hover a stock engine at minimum throttle is the hoverslam of 22.5. Focus on 22.2 (how to split stages) and 22.5 (landing), and try to land a first stage after reading.

🛰️ Industry Prep: 22.4 (engine-out) and 22.6 (reuse economics) are the language of vehicle selection and business cases. Section 22.6's cost model is a simplified version of the spreadsheets that decide whether a launch company lives or dies; the Mission Design Checkpoint asks you to run it on your own mission.

22.1 Why staging: the equation demands it

Let us start by making Chapter 3's argument airtight, because everything else in this chapter grows from it. In Chapter 3 we defined a stage's structural coefficient

$$ \varepsilon = \frac{m_s}{m_s + m_p}, $$

the fraction of a loaded stage (structure $m_s$ plus propellant $m_p$, not counting payload) that is dead structure. A good modern stage has $\varepsilon \approx 0.05$ to $0.10$: tanks, engines, plumbing, and avionics amount to only five-to-ten percent of the stage's fueled mass, and the rest is propellant. (This $\varepsilon$ is the structural coefficient of Chapter 3 — not the nozzle expansion ratio of Chapter 19, which unluckily shares the symbol, and not the specific orbital energy $\varepsilon$ of Chapter 6. Context will keep them apart; when in doubt, "structural coefficient" is a pure fraction between 0 and 1.)

The structural coefficient sets a hard ceiling on what a single stage can do. Imagine a stage carrying no payload at all — just structure and propellant. Its mass ratio is

$$ \frac{m_0}{m_f} = \frac{m_s + m_p}{m_s} = \frac{1}{\varepsilon}, $$

so its absolute best delta-v, achieved only in the useless limit of zero payload, is

$$ \Delta v_{\max} = v_e \ln\!\left(\frac{1}{\varepsilon}\right). $$

Worked Example: the single-stage ceiling. Take a very good kerosene stage, $v_e = 3.4\ \text{km/s}$ and $\varepsilon = 0.08$. Its ceiling is $$\Delta v_{\max} = 3.4\ \text{km/s} \times \ln(1/0.08) = 3.4 \times \ln(12.5) = 3.4 \times 2.526 = 8.59\ \text{km/s}.$$ That is less than the roughly $9.4\ \text{km/s}$ needed to reach low Earth orbit (Chapter 1) — and it assumed no payload. Even shaving the structure to a heroic $\varepsilon = 0.06$ only lifts the ceiling to $3.4 \times \ln(16.7) = 9.57\ \text{km/s}$: barely orbital, and still with nothing aboard. Add a real payload and the mass ratio, and thus the delta-v, collapses. A single kerosene stage simply cannot deliver a useful payload to orbit. This is not an engineering failure to be fixed with better welding; it is a wall built into the arithmetic.

That wall is why we stage. When the first stage's tanks run dry, the empty tanks, the spent engines, and the interstage are just mass — mass the remaining propellant would have to keep accelerating. Staging throws that dead mass overboard so the next engine starts with a fresh, favorable mass ratio. And, as Chapter 3 proved, the stages' delta-vs simply add:

$$ \Delta v_{\text{total}} = \sum_i v_{e,i} \ln\!\left(\frac{m_{0,i}}{m_{f,i}}\right), $$

where each stage's masses include everything stacked above it (the upper stages, still fueled, plus the payload) as its payload.

But if two stages are good, why not ten? Because every stage you add brings its own engines, its own tank domes, its own separation system — new dead mass and new ways to fail. The gain from staging has sharply diminishing returns. We can see this precisely. Suppose we must produce $\Delta v_{\text{total}} = 9.4\ \text{km/s}$ with $N$ identical stages, each with $v_e = 3.4\ \text{km/s}$ and $\varepsilon = 0.08$, splitting the delta-v equally. For each stage the mass ratio is $R = e^{(\Delta v_{\text{total}}/N)/v_e}$, and (as we will derive in 22.2) each stage's payload ratio — the fraction of that stage's lift-off mass that is genuinely useful load above it — is

$$ \pi = \frac{1 - \varepsilon R}{R\,(1 - \varepsilon)}. $$

The overall payload fraction of the whole vehicle is the product $\pi^N$. Here is what that gives:

Read the table from the top. One stage is not merely inefficient — it is impossible: the required mass ratio ($15.9$) exceeds the structural ceiling $1/\varepsilon = 12.5$, so the payload fraction goes negative, which is the arithmetic's way of saying "there is no room for payload, or even for all the structure you asked for." Two stages rescue the mission, delivering a $3.45\%$ payload fraction. Three stages improve it to $4.13\%$ — a real gain. But four stages add only another $0.26\%$, and five only $0.13\%$. Each new stage buys less, while charging the full price of new hardware and a new separation event that must work perfectly or the mission is lost (Chapter 32). Two or three stages is where that trade balances, which is exactly what almost every orbital rocket in history has chosen.

🚪 Threshold Concept. Staging is not a clever add-on; it is a mathematical necessity imposed by the structural coefficient. No chemical stage can be built light enough to reach orbit with a payload, because $1/\varepsilon$ caps its mass ratio below what orbit demands. Once you internalize that the ceiling is $\Delta v_{\max} = v_e \ln(1/\varepsilon)$ and that orbit sits above it for one stage but below the sum for two, you understand why every human who has left Earth rode a rocket that deliberately tore itself apart on the way up. The rocket that reaches orbit is never the rocket that left the pad.

🔄 Check Your Understanding 1. Why does the single-stage ceiling depend on $\varepsilon$ but not on the size of the rocket? 2. In the table, the per-stage payload ratio $\pi$ rises as $N$ increases, yet we still stop at two or three stages. What is being traded against that rising $\pi$?

Answers

1. Because both the mass ratio $1/\varepsilon$ and the delta-v $v_e\ln(1/\varepsilon)$ depend only on fractions of mass (structure-to-propellant, exhaust speed), never on absolute mass. Scale the whole stage up tenfold and every fraction is unchanged, so the ceiling is unchanged — a bigger stage hits the same wall. 2. Each added stage brings its own engines, tank domes, avionics, and a separation event. The rising $\pi$ (better mass ratio per stage because each does less work) is bought with more total dry hardware and more failure points; past three stages the extra payload fraction is tiny and the extra risk and cost are not. The optimum is a balance, not a maximum.

22.2 Serial versus parallel, and the optimal split

Rockets stage in two geometries. Chapter 3 introduced both; let us sharpen the distinction. In serial (or tandem) staging, stages are stacked nose to tail and fire in sequence: the first stage burns out and drops away, then the second ignites. Falcon 9, the Saturn V, and most upper-stage stacks are serial. In parallel staging, two or more boosters fire alongside the core from the moment of lift-off and are jettisoned early, while the core keeps burning. The Space Shuttle's twin solid boosters, Falcon Heavy's two side cores, Ariane 5's strap-ons, and the R-7 that launched Sputnik are all parallel-staged.

Why choose one over the other? The deciding physics is thrust-to-weight at lift-off (Chapter 16). A rocket cannot leave the pad unless its thrust exceeds its weight, $T/W > 1$, and it wants a healthy margin (typically $T/W \approx 1.2$–$1.4$) so it does not waste delta-v hovering against gravity (Chapter 4). A tall, efficient core optimized for the vacuum of high altitude often cannot produce enough sea-level thrust to lift the whole fueled stack. Strapping on boosters adds lift-off thrust exactly when it is needed, then drops them once the vehicle is lighter and higher. Parallel staging also lets a single core design serve many payloads: add boosters for heavy missions, fly bare for light ones. The cost is complexity — all engines light on the ground, and the boosters must separate cleanly from a still-thrusting core.

🔗 Connection: A subtle in-between exists. Falcon Heavy was designed for propellant crossfeed — the side boosters would feed propellant to the center core so that, at separation, the core still had full tanks, mimicking an ideal serial stage. It was never flown; the plumbing and risk outweighed the gain. The original Atlas used "stage-and-a-half": it dropped two of its three engines (and their weight) partway up while keeping the single sustainer and the tanks. These hybrids show that "serial" and "parallel" are the ends of a spectrum, and real vehicles live along it.

The optimal split

Now the central question of this chapter: given a total delta-v to produce, how should you divide it among the stages so that the whole rocket is as light as possible for a given payload? This is optimal staging, and it is the discipline that turns Chapter 3's "just add another stage" into a quantitative design tool.

Definition (optimal staging). The choice of how many stages to use and how to apportion the total delta-v (equivalently, the mass ratio, propellant, and structure) among them so as to minimize the vehicle's lift-off (gross) mass for a required payload and mission delta-v — or, equivalently, to maximize the payload delivered by a vehicle of given lift-off mass.

🧩 Productive Struggle. Before reading the derivation: you have $9\ \text{km/s}$ to produce with two identical stages. Would you have the big first stage do most of the work (say $6\ \text{km/s}$) and the small second stage the rest, or split it evenly, or the other way around? Guess, and hold your guess against the answer below.

Strategy first. We will bookkeep the vehicle as a chain of nested "payloads": stage 1's payload is the entire stack above it (stage 2, stage 3, …, and the real payload); stage 2's payload is everything above it; and so on. The useful payload fraction of the whole rocket is then the product of the per-stage payload ratios. We hold the total delta-v fixed, write each stage's payload ratio in terms of its mass ratio, and ask which set of mass ratios maximizes the product. For equal stages the answer is beautiful and exact; for unequal stages it is a single clean condition solved numerically.

For one stage, write the payload ratio $\pi = m_L / m_0$ (useful load over that stage's gross mass) and recall the structural coefficient $\varepsilon$. A little algebra (done in the exercises) relates the mass ratio $R = m_0/m_f$ to these:

$$ R = \frac{1}{\varepsilon + \pi(1 - \varepsilon)}, \qquad\text{equivalently}\qquad \pi = \frac{1 - \varepsilon R}{R\,(1 - \varepsilon)}. $$

Notice the constraint hiding in the second form: $\pi > 0$ requires $R < 1/\varepsilon$. Ask a stage for a mass ratio above its structural ceiling and the payload ratio goes negative — the stage is impossible, exactly the failure we saw for a single-stage-to-orbit vehicle in 22.1.

For an $N$-stage rocket the overall payload fraction telescopes into a product,

$$ \frac{m_L}{m_0} = \pi_1 \,\pi_2 \cdots \pi_N, $$

and the total delta-v is the sum $\Delta v = \sum_i v_{e,i} \ln R_i$. Minimizing lift-off mass for a fixed payload is the same as maximizing $m_L/m_0$, i.e. maximizing $\sum_i \ln \pi_i$.

The equal-stage result. Suppose every stage has the same exhaust velocity $v_e = c$ and the same structural coefficient $\varepsilon$. Because $\Delta v = c \sum_i \ln R_i = c \ln(\prod_i R_i)$, holding the total delta-v fixed fixes the product $\prod_i R_i = e^{\Delta v / c} \equiv K$. Now watch what we are maximizing. Since

$$ \prod_i \pi_i = \frac{\prod_i (1 - \varepsilon R_i)}{\left(\prod_i R_i\right)(1-\varepsilon)^N} = \frac{\prod_i (1 - \varepsilon R_i)}{K\,(1-\varepsilon)^N}, $$

and $K$ is fixed, we need only maximize $\prod_i (1 - \varepsilon R_i)$ subject to $\prod_i R_i = K$. For two stages, $\prod_i(1 - \varepsilon R_i) = 1 - \varepsilon(R_1 + R_2) + \varepsilon^2 R_1 R_2$; with $R_1 R_2 = K$ fixed, the only free part is $-\varepsilon(R_1 + R_2)$, which is largest when $R_1 + R_2$ is smallest. By the arithmetic–geometric-mean inequality, a sum with a fixed product is smallest when its terms are equal: $R_1 = R_2 = \sqrt{K}$. Equal mass ratios mean equal delta-v. The same argument extends to $N$ stages: identical stages should split the delta-v equally, each taking $\Delta v_{\text{total}}/N$. $\blacksquare$

So the intuition test above has a clean answer: for two identical stages, split evenly. Let us confirm it costs real mass to do otherwise.

Worked Example: the cost of a bad split. Two identical stages, $v_e = 3.4\ \text{km/s}$, $\varepsilon = 0.08$, must deliver $m_L = 1000\ \text{kg}$ with a total $\Delta v = 9.0\ \text{km/s}$. For a chosen split, each stage's mass ratio is $R_i = e^{\Delta v_i / v_e}$, its payload ratio is $\pi_i = (1-\varepsilon R_i)/[R_i(1-\varepsilon)]$, and the lift-off mass is $m_0 = m_L / (\pi_1 \pi_2)$. Trying several splits:

Split $\Delta v_1 / \Delta v_2$ (km/s)	$\pi_1 \pi_2$	Lift-off mass $m_0$
$2.7 / 6.3$ (or $6.3/2.7$)	$0.0337$	$29.6\ \text{t}$
$3.6 / 5.4$ (or $5.4/3.6$)	$0.0392$	$25.5\ \text{t}$
$4.5 / 4.5$ (even)	$0.0410$	$\mathbf{24.4\ \text{t}}$

The even split ($4.5/4.5$) minimizes lift-off mass at $24.4\ \text{t}$. A $60/40$ split costs an extra tonne of vehicle ($25.5\ \text{t}$) for the same payload; a lopsided $70/30$ costs more than five extra tonnes. The penalty is symmetric — it does not matter which stage is overworked — and it grows fast as you move away from the middle. For identical stages, even is optimal, and "close to even" is where you want to live.

The general case. Real stages are not identical — a kerosene first stage and a hydrogen upper stage have different $v_e$ and different $\varepsilon$ — and then the split is not even. Setting up the maximization of $\sum_i \ln \pi_i$ subject to $\sum_i v_{e,i} \ln R_i = \Delta v$ with a Lagrange multiplier (the algebra is in the exercises) yields a strikingly clean condition. There is a single constant $\mu$ (a velocity, the same for every stage) such that each stage's optimal mass ratio is

$$ \boxed{\ R_i = \frac{v_{e,i} - \mu}{v_{e,i}\,\varepsilon_i}\ } $$

and $\mu$ is fixed by requiring the mass ratios to add up to the mission: $\sum_i v_{e,i} \ln R_i = \Delta v_{\text{total}}$. This last equation is transcendental — you solve it numerically, sweeping $\mu$ until the delta-vs sum correctly — but the rule it encodes is intuitive. Because $R_i = (1 - \mu/v_{e,i})/\varepsilon_i$, a stage with a higher exhaust velocity or a lower structural coefficient gets a larger mass ratio, and therefore does more of the delta-v. The efficient, light stage should work hardest; the thirsty, heavy stage should do less. When all stages are identical the formula gives them all the same $R_i$, recovering the equal-split result. (I have verified this condition reproduces the equal-stage optimum; treat the general transcendental form as the standard optimal-staging result — see the further reading — and solve it numerically for a real mixed-propellant vehicle.)

⚠️ Common Misconception: "The biggest stage should do the most work." It feels natural that the giant first stage, with all that propellant, should provide most of the delta-v. It usually provides the least per kilogram of exhaust efficiency. A kerosene first stage has a lower $v_e$ than a hydrogen upper stage, and the optimal-split rule says the lower-$v_e$ stage should take less of the delta-v, not more. The first stage is big because it must lift the entire stack against gravity and punch through the atmosphere at high thrust-to-weight — a job about force, not efficiency. Its size is a thrust decision; the delta-v split is an efficiency decision, and the two do not have to agree. We will see the Saturn V embody exactly this tension in Case Study 1.

🔄 Check Your Understanding 1. For two identical stages needing $8\ \text{km/s}$ total with $v_e = 3.2\ \text{km/s}$, what delta-v and mass ratio should each stage have? 2. In the general rule $R_i = (v_{e,i} - \mu)/(v_{e,i}\varepsilon_i)$, what happens to a stage whose exhaust velocity $v_{e,i}$ is only slightly above $\mu$?

Answers

1. Split evenly: each stage does $\Delta v_i = 4\ \text{km/s}$, so $R_i = e^{4/3.2} = e^{1.25} \approx 3.49$. 2. Its mass ratio $R_i = (v_{e,i}-\mu)/(v_{e,i}\varepsilon_i)$ becomes small (approaching $1$ as $v_{e,i} \to \mu$), so it does very little delta-v. A stage whose exhaust velocity barely exceeds $\mu$ is barely worth staging; if $v_{e,i} \le \mu$ the formula gives $R_i \le 0$, the optimizer's signal that this stage should not exist at all — the delta-v belongs on the more efficient stages.

22.3 Propellant management: the demons in the tank

The rocket equation assumes you can deliver propellant to the engine smoothly, completely, and on schedule. A real tank of cryogenic liquid, half-empty and accelerating, does not cooperate. Four distinct troubles live in that tank, and every one has killed or nearly killed a rocket. This section names them and the hardware that tames them.

Ullage

In vacuum or in the brief weightless coast between stages, liquid propellant does not sit obediently at the bottom of its tank. With no gravity to settle it, surface tension pulls it into blobs that can float anywhere — including away from the outlet where the feed line draws. Start a pump under those conditions and it swallows gas instead of liquid, and an engine fed gas does not start; it chokes, cavitates, or explodes.

Definition (ullage). Ullage is the gas-filled volume above the liquid in a partly full tank (the term is borrowed from winemaking). In rocketry it also names the fix: an ullage maneuver is a small forward acceleration, produced by ullage motors (little solid or hypergolic thrusters) or by the reaction-control thrusters, applied just before main-engine start to settle the liquid propellant over the outlet so the pumps draw gas-free.

Ullage motors are why an upper stage fires a puff of small thrusters a moment before its main engine lights. The Saturn V's third stage carried solid ullage rockets to settle its propellant before the J-2 relit for the trip to the Moon. It is a tiny burn with an outsized job: without it, restarting an engine in space is a coin flip.

Propellant slosh

Now give the tank some sideways motion, as every steering correction does. The liquid sloshes — the same back-and-forth you feel carrying a wide, half-full pan of water.

Definition (propellant slosh). Propellant slosh is the oscillation of liquid propellant back and forth in a partially filled tank. Because the liquid can be a large fraction of the vehicle's mass, its sloshing shifts the center of mass and feeds forces and torques into the guidance and control system; if the slosh frequency lands near a control-loop or structural frequency, the coupling can grow and destabilize the vehicle.

The cure is anti-slosh baffles — rings or vanes fixed inside the tank that break the fluid into smaller masses and add damping, raising the slosh frequency away from the control loop and bleeding energy out of the motion. They are cheap insurance against an expensive failure. That the failure is real is not hypothetical: SpaceX's second Falcon 1 flight, in 2007, reached space but lost control late in the second-stage burn when unchecked propellant slosh, excited by the engine's earlier behavior, coupled into the control system and the stage began to tumble. Baffles and a revised control law followed. Slosh is a quiet demon that only wakes when the tank is half-empty and the vehicle is maneuvering — which is to say, during the most delicate part of the flight.

POGO oscillation

The third demon is a feedback loop that can turn a rocket into a giant, self-destructing musical instrument. It is called POGO, after the pogo stick it makes the vehicle imitate.

Definition (POGO). POGO is a self-excited longitudinal oscillation of a liquid rocket: a closed loop in which vibration along the vehicle's long axis modulates the pressure in the propellant feed lines, which modulates the flow into the chamber, which modulates the thrust, which drives the structural vibration that started the cycle. If the loop gain exceeds one, the oscillation grows — sometimes to amplitudes that damage the engine or injure the crew.

Trace the loop and its menace is clear. The long propellant columns in the feed lines behave like springs of fluid; the vehicle's structure is a spring of metal; the combustion chamber closes the loop by turning flow into thrust. Couple three resonators that feed one another and you have an oscillator. POGO nearly grounded the crewed Titan II of the Gemini program until it was tamed, savaged the uncrewed Apollo 6 Saturn V, and on Apollo 13's ascent shut down the S-II's center engine early when a $16\ \text{Hz}$ oscillation built to dangerous levels — the mission's first emergency, before the more famous one. The standard cure is a POGO accumulator (or suppressor): a small gas-charged cavity plumbed into the feed line that acts as a soft spring, detuning the fluid column so the loop gain never reaches one. It breaks the coupling the way a shock absorber breaks a car's bounce.

📜 From History: POGO earned obsessive attention because it endangered people. When Gemini astronauts were to ride the Titan II, its roughly $11\ \text{Hz}$ longitudinal oscillation — bearable for a warhead, not for a human trying to read instruments — had to be driven below a fraction of a g before the vehicle was crew-rated. Engineers added gas-filled standpipes to the oxidizer lines to soften the fluid springs. The lesson recurs throughout spaceflight (Chapter 32): a phenomenon that is a nuisance for cargo becomes a hard constraint the moment a heartbeat is aboard.

Cryogenic boiloff

The last demon afflicts the highest-performing propellants. Liquid hydrogen boils at $20\ \text{K}$ ($-253\,^\circ\text{C}$), liquid oxygen at $90\ \text{K}$, liquid methane at $112\ \text{K}$. Space, paradoxically, is full of heat sources — sunlight, Earth's infrared glow, the warm structure of the rocket itself — and heat leaks relentlessly into any tank colder than its surroundings.

Definition (boiloff). Boiloff is the continual loss of cryogenic propellant as heat leaking through the tank walls vaporizes the liquid; the vapor must be vented to keep the tank from over-pressurizing, and the vented mass is gone for good. Boiloff limits how long a cryogenic stage can be stored on the pad or coast in space, and it drives the design of insulation and, for long missions, active refrigeration.

The bookkeeping is simple: a heat leak $Q$ boils off propellant at a rate $\dot m = Q / h_{fg}$, where $h_{fg}$ is the propellant's latent heat of vaporization.

Worked Example: how fast does hydrogen boil away? Suppose $Q = 100\ \text{W}$ of heat leaks into a liquid-hydrogen tank (latent heat $h_{fg} \approx 446\ \text{kJ/kg}$). Illustratively (Tier 3 round numbers): $$\dot m = \frac{Q}{h_{fg}} = \frac{100\ \text{W}}{446{,}000\ \text{J/kg}} = 2.24\times10^{-4}\ \text{kg/s} = 19.4\ \text{kg/day}.$$ On a small stage holding a tonne of hydrogen, that is roughly $2\%$ per day, gone — before you have spent a drop on the mission. This is why a hydrogen upper stage must fire within hours, why sending a cryogenic stage on a months-long cruise to Mars is so hard, and why propellant depots are an unsolved engineering problem rather than a solved one. Hydrogen's high $h_{fg}$ actually helps (oxygen's is only about $213\ \text{kJ/kg}$), but hydrogen's ultra-low temperature and low density make its tanks huge and its heat leak stubborn.

Boiloff is fought with multi-layer insulation and reflective sunshields (Chapter 24 treats the thermal physics), by orienting the vehicle to shade the tanks, and, for the most ambitious missions, by active cryocoolers that aim for "zero boiloff." It is also a major reason many deep-space stages abandon cryogens for storable hypergolic propellants (Chapter 18) that sit for years without a chill — trading specific impulse for the simple ability to wait.

🔄 Check Your Understanding 1. Which demon — ullage, slosh, POGO, or boiloff — is most dangerous specifically during a restart in orbit, and why? 2. POGO is called a longitudinal instability. Why does that word matter — how does it differ from slosh?

Answers

1. Ullage. During a coast the propellant floats away from the outlet, so before a restart it must be settled by an ullage burn; skip it and the pump ingests gas and the engine will not light. (Boiloff also acts during coasts, but it is a slow loss, not an ignition failure.) 2. "Longitudinal" means the oscillation is along the vehicle's thrust axis — the whole rocket stretches and compresses like a spring, coupling to the axial feed-line and combustion pressures. Slosh is a lateral motion of liquid that mainly disturbs attitude and the center of mass sideways. Different directions, different loops, different cures (accumulators for POGO, baffles for slosh).

22.4 Tank pressurization and engine-out capability

Two more vehicle-level realities round out the propulsion story before we turn the rocket around and fly it home: keeping the tanks pressurized, and surviving the loss of an engine.

Why tanks must be pressurized

A turbopump (Chapter 17) is a marvel, spinning fast enough to raise propellant pressure a hundredfold — but it cannot suck. Present its inlet with propellant at too low a pressure and the liquid flashes to vapor as it accelerates into the impeller, a destructive condition called cavitation that shreds performance and can wreck the pump in seconds. To prevent it, the tank must hold its propellant above a minimum inlet pressure (the net positive suction head). Tank pressure also gives thin-walled tanks their rigidity — some, like the old Atlas and the Centaur, are "balloon tanks" that would collapse unpressurized — and, on pressure-fed engines with no pumps at all, it is the only thing forcing propellant into the chamber.

The pressure comes from a pressurant gas in the ullage space. Classically that gas is helium, stored cold and dense in high-pressure bottles and warmed as it expands into the tanks. Helium is inert and light but it is its own headache: it must be carried in heavy composite-overwrapped pressure vessels (COPVs), and helium is a finite, expensive resource.

🔧 Engineering Reality: In September 2016 a Falcon 9 exploded on the pad during propellant loading, destroying its payload. The cause was traced to a helium COPV submerged in the second stage's liquid-oxygen tank: super-chilled oxygen had seeped into and frozen within the vessel's overwrap, and the vessel failed. A pressurization component — not an engine, not the airframe — took the whole vehicle. It is a vivid reminder that in a rocket there are no minor systems; the pressurant that merely supports the propellant is as capable of ending the flight as the engine that burns it.

A cleaner alternative is autogenous pressurization: instead of a separate gas, tap the vehicle's own propellants, vaporize them (often using heat from the engine), and feed that gas back to pressurize the tanks — gaseous oxygen over the liquid oxygen, gaseous methane over the liquid methane. Starship's Raptor system does exactly this, eliminating helium entirely. The payoff is not only mass and simplicity but Mars: a vehicle that pressurizes itself from its own tanks needs no supply of a rare gas it cannot make on another planet — a systems choice we will unpack fully in Chapter 38.

Engine-out capability

A rocket with one big engine has a brutal property: if that engine fails, the mission is over. A rocket with many smaller engines can be built to survive a loss.

Definition (engine-out capability). Engine-out capability is a launch vehicle's ability to tolerate the failure of one or more engines during flight and still reach orbit (or execute a safe abort), by throttling up or burning the surviving engines longer and re-planning the trajectory in real time. It requires enough engines that the loss of one still leaves adequate thrust, a propellant reserve to burn longer, engine-to-engine isolation so a failure does not cascade to its neighbors, and guidance authority to compensate.

The first requirement is the sharpest, because at lift-off a stage must keep $T/W > 1$ even after losing an engine.

Worked Example: Falcon 9 loses an engine at lift-off. Falcon 9's first stage has nine Merlin engines, each producing about $845\ \text{kN}$ at sea level (Tier 2). At lift-off the fueled stack masses roughly $549{,}000\ \text{kg}$, so it weighs $$W = 549{,}000\ \text{kg} \times 9.81\ \text{m/s}^2 = 5.39\times10^{6}\ \text{N} = 5{,}386\ \text{kN}.$$ With all nine engines, $T = 9 \times 845 = 7{,}605\ \text{kN}$ and $T/W = 7{,}605/5{,}386 = 1.41$. Lose one engine and $T = 8 \times 845 = 6{,}760\ \text{kN}$, giving $T/W = 6{,}760/5{,}386 = 1.26$. Still above one — the rocket can still climb. It will fly a slightly lofted trajectory and the surviving engines will burn a little longer to make up the lost impulse, but the mission is not lost. That $1.26$ is the whole argument for many small engines instead of one big one.

This is not a paper capability. On the CRS-1 mission in 2012, a Falcon 9 lost engine 1 about $79$ seconds into flight; the remaining eight burned longer and delivered the primary Dragon cargo to the International Space Station. (A secondary payload was stranded in a low orbit, because a NASA-imposed propellant-margin rule forbade the second-stage relight that its higher orbit needed — engine-out is not free, and here it cost the rideshare passenger.) The Saturn V had the same grace: both Apollo 6 and Apollo 13 lost second-stage engines and reached orbit on those remaining.

📜 From History: The counterexample is the Soviet N1 Moon rocket, whose first stage clustered thirty engines. Thirty engines mean thirty times the chances of a failure, and the N1's engine-management system could not diagnose and compensate fast enough; a shutdown could cascade. All four N1 launches (1969–1972) failed, none reaching orbit. Engine-out capability is not automatic with many engines — it is a designed property requiring isolation, fast detection, and adaptive guidance (Chapter 27). Many engines give you the raw material for graceful degradation; without the control system to manage them, they give you only more ways to fail. The difference between Falcon 9 and the N1 is not the engine count — it is what happens in the half-second after one quits.

🔄 Check Your Understanding 1. Why is engine-out at lift-off harder to survive than engine-out late in the first-stage burn? 2. Give one reason many small engines can be worse for reliability than one big engine, and the design feature that flips it back to better.

Answers

1. At lift-off the vehicle is fully fueled and heaviest, so $T/W$ is lowest and least forgiving; losing an engine there may drop $T/W$ near or below one. Late in the burn the stage is far lighter, $T/W$ is already high (often $>3$), and losing one engine barely dents the margin. 2. More engines means more independent chances that some engine fails (worse), but with engine isolation, fast fault detection, and adaptive guidance, the vehicle tolerates those failures instead of being destroyed by them — turning a higher failure rate into a lower mission-loss rate. The N1 had the engines but not the management; Falcon 9 has both.

22.5 Reusability: flying the rocket home

For sixty years, the deepest waste in spaceflight was hidden in plain sight: we built exquisite machines and used each one exactly once. The rocket equation seems to demand it — every kilogram of landing gear or recovery propellant is a kilogram stolen from the mass ratio. But the equation only counts mass; it never counted dollars. Reusability is the decision to spend a little of the equation's precious mass to buy back the whole vehicle.

Definition (reusability). Reusability is designing a launch vehicle, or a stage of it, to be recovered intact and flown again, so that its manufacturing cost is amortized over many flights rather than discarded after one. A vehicle may be expendable (thrown away), partially reusable (recovering the first stage and/or the fairings, as Falcon 9 does), or fully reusable (recovering every stage, the goal of Starship).

How do you get a spent first stage — a $40$-meter tube falling from the edge of space at more than a kilometer per second — back to the ground intact? You could use parachutes (heavy, imprecise, and unkind to engines that must land in salt water) or wings (the Space Shuttle's answer, and a heavy one). SpaceX's answer is to use the engines themselves.

Definition (propulsive landing). Propulsive landing (retropropulsion) is decelerating and landing a stage vertically using its own rocket engines, rather than parachutes, wings, or a splashdown. The stage relights in flight, uses thrust to cancel its velocity, and touches down on deployable legs (or is caught by a tower).

The Falcon 9 booster's return is a precise, four-act sequence, and every act is a piece of physics we have already met:

1. Stage separation. Around two and a half minutes into flight, having spent most of its propellant, the first stage separates from the second and is now a mostly empty tube coasting upward and downrange.
2. Boostback burn (for a return to the launch site). The booster flips $180^\circ$ using cold-gas thrusters and relights several engines to cancel its downrange velocity and send it arcing back toward the pad. For high-energy missions this burn is shortened or skipped, and the booster instead lands far downrange on an autonomous droneship — the difference between a return-to-launch-site profile and a droneship recovery is, at bottom, a delta-v budget.
3. Entry burn. Falling back into the thickening atmosphere at high speed, the booster relights three engines to slow down before the air gets dense. This is not just braking: the deceleration and the engine plume shield the vehicle's base and cut the peak heating and dynamic pressure it must survive (Chapter 7; Chapter 5). Retropropulsion lets SpaceX skip the heavy heat shield that a passive re-entry would demand — the booster's aluminum-lithium skin survives because the engines never let it get too hot.
4. Aerodynamic descent and landing burn. Four grid fins — lattice control surfaces folded during ascent, deployed for descent — steer the falling booster with aerodynamic lift toward a target only meters wide. In the final seconds the center engine (sometimes three, then one) relights for the landing burn, and the legs deploy just before touchdown.

There is a beautiful controllability trap hidden in that last burn, and it is worth doing the numbers, because it explains why the landing looks so alarmingly abrupt.

Worked Example: why a Falcon 9 booster cannot hover. By landing, the booster is nearly empty, dry mass about $25{,}600\ \text{kg}$, so it weighs roughly $$W = 25{,}600 \times 9.81 = 251{,}000\ \text{N} = 251\ \text{kN}.$$ A single Merlin, even throttled to its minimum (about $40\%$ of $845\ \text{kN}$), still produces $$T_{\min} = 0.40 \times 845 = 338\ \text{kN},$$ so the lightest possible thrust gives $T/W = 338/251 = 1.35 > 1$. The engine cannot balance the booster's weight — even at minimum throttle it pushes up harder than gravity pulls down. Therefore the booster cannot hover, hold, and gently settle. It must fall engines-first and ignite at exactly the right instant so that its deceleration brings its velocity to zero precisely as it reaches the ground: a "hoverslam" or "suicide burn." If it lights a moment too early it stops in mid-air and then shoots back up; a moment too late and it hits the ground still moving. Landing a Falcon 9 is a problem of timing to within a fraction of a second, which is why it is done by computer, not by a pilot (Chapter 27).

How hard is the deceleration? If the booster (with reserve propellant, say $30{,}000\ \text{kg}$) arrives at about $250\ \text{m/s}$ and burns one engine at full thrust, the net deceleration is $$a = \frac{845{,}000\ \text{N}}{30{,}000\ \text{kg}} - 9.81\ \text{m/s}^2 = 28.2 - 9.8 = 18.4\ \text{m/s}^2 \approx 1.9\,g,$$ stopping it in $t = 250/18.4 \approx 14\ \text{s}$ over a height of $v^2/2a = 250^2/(2\times18.4) \approx 1{,}700\ \text{m}$. A single, violent, perfectly timed burn from under two kilometers up. (Numbers illustrative, Tier 3; the real profile varies with mission and reserve.)

🔗 Connection: Propulsive landing was only possible because of choices made upstream in Part III. The engine must relight reliably in flight and throttle deeply — properties of the Merlin's gas- generator cycle and injector design (Chapter 17) — and the guidance must solve the timing problem in real time (Chapter 27). A solid motor cannot be relit or throttled, which is one reason no solid-boosted stage has ever landed this way. Reusability is not a bolt-on; it reaches back into the engine cycle.

The next step removes even the legs. Starship's Super Heavy booster returns to its launch tower and is caught in mid-air by two giant arms — first achieved in October 2024 — so the vehicle need not carry landing legs at all, saving mass exactly where the rocket equation is stingiest. That story, and Starship's goal of recovering both stages, is the climax of Chapter 38; here it is enough to see the principle. Every gram of leg or grid fin or reserve propellant is a tax the rocket equation levies on reuse. The bet is that the vehicle you get back is worth far more than the payload you gave up. Whether that bet pays is arithmetic, and we do it now.

🔄 Check Your Understanding 1. Why does the entry burn happen high in the atmosphere rather than waiting until the booster is lower and slower? 2. In one sentence, why can't a solid rocket booster land the way a Falcon 9 first stage does?

Answers

1. Slowing down before the air becomes dense keeps the peak dynamic pressure and heating within what the booster's structure can survive; waiting until it is low means hitting thick air at high speed, exactly the punishing environment retropropulsion is meant to avoid. 2. A solid motor cannot be shut down, relit, or throttled once ignited, so it can neither perform a controlled entry/landing burn nor be timed for a hoverslam — propulsive landing requires an engine you can turn on, off, and down on command.

22.6 The economics of reuse

Here is the theme this whole book has been building toward — reusability is changing everything. The physics of landing a booster is settled; the question that decides whether it matters is economic. Does reuse actually make launch cheaper — and if so, by how much, and after how many flights?

Start with where the money goes. A rocket's propellant is astonishingly cheap: a Falcon 9's load of kerosene and liquid oxygen costs only a few hundred thousand dollars, a fraction of a percent of a launch that lists near sixty million. Almost all the cost is hardware — the engines, tanks, and avionics, thrown away after one flight. The analogy that makes it vivid: if an airline scrapped a $100$-million airliner after every single flight and kept only the fuel, a ticket would cost a fortune, no matter how cheap the jet fuel. Reuse is the proposal to fly the airliner again.

Model it simply. Let the reusable hardware (say the first stage) cost $M$ to build and fly $N$ times. Each flight also has a refurbishment cost $R$ (inspect, clean, replace what wears) and a fixed per-flight cost $F$ that reuse does not recover — the expended upper stage, propellant, range fees, and operations. Then the cost per flight is

$$ C_{\text{reuse}}(N) = \frac{M}{N} + R + F. $$

The manufacturing cost is amortized: spread over $N$ flights, its share $M/N$ shrinks as $N$ grows, and the cost per flight falls toward a floor of $R + F$. Compare that to an expendable vehicle, which pays the full hardware cost every time: $C_{\text{expend}} = M + F$ (roughly; its second stage and ops are the same $F$).

Worked Example: the break-even flight. Use illustrative figures (Tier 3): building the reusable booster costs $M = \$30\ \text{M}$; refurbishment is $R = \$3\ \text{M}$ per flight; the expended upper stage plus operations and propellant is $F = \$25\ \text{M}$; and an expendable version costs $C_{\text{expend}} = \$50\ \text{M}$ per launch. Then $C_{\text{reuse}}(N) = 30/N + 3 + 25 = 28 + 30/N$ (in \$M):

| Flights $N$ | $C_{\text{reuse}}(N)$ (\$M) | vs. expendable (\$50M) | |---|---|---| | 1 | $58$ | more expensive | | 2 | $43$ | cheaper | | 5 | $34$ | much cheaper | | 10 | $31$ | | | 20 | $29.5$ | | | $\infty$ | $28$ (the floor) | |

Three things jump out. First, the very first flight of a reusable design costs more than an expendable one ($\$58$M vs $\$50$M): you paid for legs, grid fins, and recovery, and you gave up payload, all to fly the booster once. Reuse is a loss until you actually reuse. Second, break-even comes fast — around $N \approx 1.4$, so by the second flight of a booster the design is ahead. Third, and most important, the cost does not fall to zero; it asymptotes to a floor of $\$28$M set by the still-expended upper stage and the fixed operations. You cannot amortize your way past the parts you keep throwing away.

That floor is the punchline of the chapter and the setup for the next era. If reusing the first stage alone drops cost to a floor set by the expendable second stage, then the way past the floor is to reuse the second stage too — which is precisely Starship's bet, and precisely why fixing operations cost (fast turnaround, few people, no refurbishment) matters as much as fixing the hardware. Falcon 9 proved reuse works and pays; Starship aims at the floor Falcon 9 cannot cross.

🚪 Threshold Concept. For the whole history of rocketry, the vehicle was a consumable, like a bullet. Reusability reclassifies it as a vehicle, like an aircraft — and that single reclassification resets the economics of everything downstream. A ten-to-hundred-fold drop in launch cost does not just make old missions cheaper; it makes new activities possible that were unthinkable at the old price — thousand-satellite constellations (Chapter 33), propellant depots, routine crewed flight. We are at the beginning of that shift, not the end. Once you see launch as a service you buy repeatedly rather than a rocket you build once, the entire map of what is affordable in space redraws itself.

🐛 Find the Error. An analyst argues: "Falcon 9 recovers its first stage, which is most of the vehicle, so reuse should cut the cost per flight essentially to the cost of propellant — a few hundred thousand dollars. Anything more is inefficiency." Where is the reasoning wrong?

Answer

The analyst ignores the fixed per-flight cost $F$ that reuse never recovers: the expendable second stage (built new every flight), plus refurbishment $R$, range fees, and operations. In the model, $C_{\text{reuse}} \to R + F$ as $N \to \infty$, a floor far above propellant cost — in the worked example, $\$28$M, not $\$0.3$M. Propellant is cheap, but the second stage and operations are not, and you keep paying for them on every flight. Reuse removes the first-stage manufacturing term, not the whole cost. Getting near the propellant-cost limit requires reusing every stage and driving down operations — the reason full reuse, not partial, is the real prize.

Mission Design Checkpoint: staging design and the reuse question

This chapter adds two things to your Mission Design Review: a staged delta-v budget, and a decision about reusability.

The design. Take the delta-v budget you have been building since Chapter 3 and turn it into a staging plan. Decide how many stages your launch needs (for Earth-to-orbit tracks, two or three, per 22.1's diminishing-returns table). If your stages are similar, split the delta-v evenly (22.2); if they differ in propellant, give the higher-$v_e$ stage more of the work. Then size each stage from the top down with Chapter 3's stage-mass formula,

$$ m_{\text{stage}} = m_{\text{above}}\,\frac{R - 1}{1 - \varepsilon R}, \qquad R = e^{\Delta v_{\text{stage}}/v_e}, $$

checking each time that $R < 1/\varepsilon$ so the stage is buildable. Record the lift-off mass; it is the number that will drive your launch-vehicle selection in Chapter 30.

The reuse note. Add a short paragraph to your MDR: would reuse help this mission? Run the 22.6 model on rough numbers. A high-cadence constellation (Track A comsats by the dozen) loves reuse; a one-off deep-space probe (Track C or D) may fly so rarely that the payload penalty of recovery outweighs any amortization — for a single flight, recovery only costs you payload. State your call and your reason.

The code. Extend astrotools/rocket.py, which you began in Chapter 3, with an optimal-split helper. It reuses stage_delta_v and computes the overall payload fraction of an equal-stage design:

import math
G0 = 9.80665

def payload_fraction(dv_total, ve, eps, n):
    """Overall payload fraction of an n-stage rocket with identical stages
    (exhaust velocity ve, structural coefficient eps) splitting dv_total equally."""
    R = math.exp((dv_total / n) / ve)      # per-stage mass ratio (equal split)
    if R >= 1 / eps:                        # stage exceeds its structural ceiling
        return float("nan")                 # not buildable
    pi = (1 - eps * R) / (R * (1 - eps))    # per-stage payload ratio
    return pi ** n

# How many stages to put 9.4 km/s of delta-v up, at ve = 3.4 km/s, eps = 0.08?
for n in (2, 3):
    print(n, "stages ->", round(payload_fraction(9400, 3400, 0.08, n), 4))
# Expected output:
# 2 stages -> 0.0345
# 3 stages -> 0.0413

The outputs are the payload fractions from 22.1's table — $3.45\%$ for two stages, $4.13\%$ for three — now computed rather than tabulated. Point this at your mission's delta-v and chosen engine, and it tells you what fraction of your rocket can be payload, and thus how big the rocket must be. In Chapter 29 this feeds straight into mission.py's vehicle sizing.

Summary

Staging, propellant management, and reusability are the chapter where the rocket equation meets the whole vehicle. Carry these forward:

Key numbers worth remembering: good structural coefficient $\varepsilon \approx 0.05$–$0.10$; single-stage kerosene ceiling $\approx 8.6\ \text{km/s}$ (zero payload); two-stage LEO payload fraction $\approx 3$–$4\%$; Falcon 9 lift-off $T/W \approx 1.4$ (and $\approx 1.26$ with one engine out); reuse break-even $\approx 2$ flights.

Spaced Review

Retrieval strengthens memory. Answer from memory before checking, then look back at the cited section.

1. (§22.1, Ch. 3) Chapter 3 said stage delta-vs add. Using that, explain in one sentence why a two-stage rocket beats a single stage carrying the same propellant and structure.
2. (§22.1, Ch. 3) The structural coefficient is $\varepsilon = m_s/(m_s+m_p)$. What is the maximum mass ratio a single stage can have, and why does that cap force staging?
3. (§22.5, Ch. 17) Which two engine capabilities, both properties of the engine cycle you met in Chapter 17, does propulsive landing absolutely require?
4. (§22.4, Ch. 17) A turbopump needs its inlet pressure kept above the propellant's vapor pressure. What is the failure called if it isn't, and what tank feature prevents it?
5. (§22.2) For two identical stages, how should the total delta-v be split, and what is the one-line reason?

Answers

1. The two-stage rocket drops its heavy empty first-stage structure at staging, so the second stage's engine never has to accelerate that dead mass; the single stage drags it all the way up, wasting propellant on empty tanks. 2. The maximum mass ratio is $1/\varepsilon$ (achieved only with zero payload), so the best single-stage delta-v is $v_e\ln(1/\varepsilon)$; for real $\varepsilon\approx0.06$–$0.10$ that ceiling sits below orbital delta-v, so no single stage can reach orbit with payload — staging is mandatory. 3. Restart (relight in flight) and deep throttling (to control the deceleration and time the landing burn); a solid motor has neither. 4. Cavitation — the propellant flashing to vapor at the pump inlet; it is prevented by keeping the tank pressurized (adequate net positive suction head) with a pressurant gas. 5. Evenly — each stage takes half. With equal $v_e$, holding total delta-v fixes the product of mass ratios, and by AM–GM the payload fraction is maximized when the mass ratios (hence the delta-vs) are equal.

What's Next

We have now pushed the rocket equation as far as the propulsion system can take it: optimally staged, its propellant tamed, an engine to spare, and — for the first time in history — the vehicle flown home to fly again. But every kilometer per second in this chapter has assumed a vehicle that survives the forces of getting there: the crushing acceleration, the shaking, the thermal shock, the loads that staging and landing themselves impose. A rocket that is optimally light is a rocket built to the edge of what its materials can bear. In Chapter 23 we open Part IV and turn to the structure itself — launch loads, materials, factors of safety, and the relentless mass budget — asking not how the vehicle moves, but how it holds together while it does. The enemy, as always, is mass; the new question is how little structure you can get away with and still not break.