Skip to content
LumenaryAcademy
Frontier 90 min

Engineering for space and fusion

Reusable rockets and how they land, keeping spacecraft at the right temperature and in one piece, making things in orbit, and the magnets, heat exhaust and neutron-proof materials that fusion power needs — with an honest status as of late 2026.

Code tested · 6AI-draftedAwaiting review by a mechanical engineer

Your guides: Chloe · The Designer, Margaret · The Test Engineer, Gideon · The Checker

MotivationH

Two extreme frontiers

On 21 December 2015 a Falcon 9 first stage, having sent eleven satellites towards orbit, flew back to Cape Canaveral and landed upright on a pad. Fifteen months later a used booster flew again. By the mid-2020s landing and reflying boosters had become routine, individual boosters had flown more than twenty times, and in October 2024 a far larger booster was caught out of the air by mechanical arms on its launch structure in Texas.

Seven years after that first landing, on 5 December 2022, the National Ignition Facility in California focused 2.05 MJ of laser light on a capsule of hydrogen isotopes the size of a peppercorn and released 3.15 MJ of fusion energy, the first time a laboratory fusion experiment produced more energy than was delivered to its fuel.

Space and fusion are the two places where mechanical engineering is pushed hardest at once: enormous power densities, temperatures from a few kelvin to millions, loads that would wreck ordinary structures, and no chance to repair most mistakes. This lesson works through the engineering of each with the tools of the course, and ends with an honest account of where each stands in late 2026.

ComputationC

Reusable launch vehicles: what reuse costs, and what it saves

Everything about a launch vehicle starts with the rocket equation: a stage with exhaust speed ve=Ispg0v_e = I_\text{sp} g_0 that burns from mass m0m_0 to mfm_f gains Δv=veln⁡(m0/mf)\Delta v = v_e \ln(m_0/m_f). Reaching low Earth orbit needs about 9.4 km/s including gravity and drag losses, so even with two stages only a few per cent of the lift-off mass reaches orbit as payload. To land a first stage, it must keep propellant back for the trip home (an entry burn to slow down before the dense atmosphere, a landing burn at the bottom, and for a return to the launch site a boost-back burn to reverse its course) and carry landing legs and steering fins. All of that is taken from the payload.

stage separationsecond stage to orbitentry burngrid fins steerlanding burnship at sealaunch sitelanding zoneboost-back burn(return to site)
The flight of a reusable first stage. After stage separation the second stage continues to orbit. The booster either flips and burns back towards its launch site (dashed) or coasts on to a ship at sea; it then fires an entry burn to slow down, steers through the atmosphere with grid fins, and fires a final landing burn.

The program below uses a generic two-stage launcher with round numbers (not any particular rocket), finds the largest payload it can put in orbit with the first stage thrown away, landed at sea, or returned to its launch site, and then asks what a kilogram in orbit costs if the booster flies many times.

python
import math
g0 = 9.80665
# A generic two-stage launcher (round numbers, not any particular rocket)
prop1, dry1, isp1 = 400.0, 25.0, 300.0      # first stage: propellant (t), dry mass (t), average Isp (s)
prop2, dry2, isp2 = 100.0, 4.5, 345.0       # second stage
dv_orbit = 9.4                              # km/s to low Earth orbit, including gravity and drag losses

def payload(dv_reserve_kms, extra_dry=0.0):
    """Largest payload (t) when the first stage keeps enough propellant to give itself dv_reserve on the way back."""
    d1 = dry1 + extra_dry                                    # legs and grid fins add dry mass
    reserve = d1 * (math.exp(dv_reserve_kms * 1000 / (isp1 * g0)) - 1)
    def dv_total(p):
        m0 = d1 + prop1 + dry2 + prop2 + p
        dv1 = isp1 * g0 * math.log(m0 / (m0 - (prop1 - reserve)))
        dv2 = isp2 * g0 * math.log((dry2 + prop2 + p) / (dry2 + p))
        return (dv1 + dv2) / 1000, dv1 / 1000
    lo, hi = 0.0, 100.0
    for _ in range(60):                                      # bisection: payload that just reaches orbit
        mid = 0.5 * (lo + hi)
        lo, hi = (mid, hi) if dv_total(mid)[0] > dv_orbit else (lo, mid)
    return lo, reserve, dv_total(lo)[1]

base = payload(0.0)[0]
print(" first-stage recovery            reserve (t)  staging dv (km/s)  payload (t)  vs expendable")
for name, dv_res, extra in [("expendable", 0.0, 0.0),
                            ("landing at sea (entry + landing burns)", 1.2, 2.0),
                            ("return to launch site (adds boost-back)", 3.0, 2.0)]:
    p, reserve, dv1 = payload(dv_res, extra)
    print(f" {name:40s} {reserve:6.1f}       {dv1:6.2f}         {p:6.1f}       {p / base:5.0%}")

# Economics, for the sea landing: the first stage is ~65 % of the vehicle's cost; refurbishment ~5 % a flight
p_sea = payload(1.2, 2.0)[0]
print("\n flights per booster   cost per kg of payload, relative to expendable")
for flights in (1, 2, 5, 10, 20):
    cost = 0.65 / flights + 0.05 + 0.35                       # second stage is still thrown away
    print(f"        {flights:3d}                   {cost / p_sea / (1.0 / base):5.0%}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Landing at sea costs 11% of the payload; returning to the launch site, which needs far more propellant to reverse the booster's course, costs 30%. The economics nonetheless favour reuse once a booster flies a few times: with these assumptions the cost per kilogram falls to 59% of the expendable cost at five flights and about half at ten. It then levels off, because the second stage is still thrown away every time. That floor is the reason for the next step, fully reusable vehicles in which the upper stage also returns, and it is why heat shields that survive many re-entries are now as important a research problem as engines.

The numbers are only as good as the turnaround. NASA's Space Shuttle was partly reusable, but its orbiters needed months of inspection and refurbishment between flights, including thousands of individually fitted heat-shield tiles, and its cost per flight never approached the hopes of the 1970s. What changed in the 2010s was designing for fast, cheap refurbishment from the start: engines that can be inspected without removal, robust thermal protection, and a fleet whose data from every flight feeds the next.

ComputationC

Computing it: a landing burn

The program simulates the final burn of a 33-tonne booster falling at 250 m/s, with one engine of 800 kN that can throttle down to half thrust. First it finds the one ignition height at which a full-thrust burn stops the booster exactly at the pad, and tries igniting half a second early and late. Then it uses closed-loop guidance: at every instant the computer commands the thrust that would bring the booster to rest exactly at the pad with constant deceleration, a=v2/2ha = v^2/2h, starting from a height planned for 80% throttle so that it has margin to throttle up or down.

python
g0, g = 9.80665, 9.81
m_dry, m_prop = 27_000.0, 6_000.0              # booster dry mass and landing propellant (kg)
T_max, thr_min, isp = 800e3, 0.5, 282.0        # one engine: full thrust (N), deepest throttle, Isp (s)
v0 = 250.0                                     # falling speed when the landing burn starts (m/s)
print(f"thrust-to-weight at the deepest throttle: {T_max * thr_min / ((m_dry + m_prop) * g):.2f}"
      " (above 1: the booster cannot hover)")

def descend(h0, guidance, dt=0.002):
    """Integrate the landing burn from height h0. guidance(h, v, m) returns the throttle (clipped to
    the engine's range). Stops when the booster reaches the pad or comes to rest above it.
    Drag during the burn is ignored; it would help a little."""
    h, v, m = h0, -v0, m_dry + m_prop
    while h > 0 and v < 0:
        T = min(max(guidance(h, v, m), thr_min), 1.0) * T_max
        v += (T / m - g) * dt
        h += v * dt
        m -= T / (isp * g0) * dt
    return h, v, m_dry + m_prop - m

def outcome(h, v):
    return f"comes to rest {h:5.1f} m above the pad" if h > 0 else f"hits the pad at {-v:5.1f} m/s"

# 1. Open loop at full thrust: find the one ignition height that stops exactly at the pad
full = lambda h, v, m: 1.0
lo, hi = 100.0, 5000.0
for _ in range(40):
    mid = 0.5 * (lo + hi)
    lo, hi = (mid, hi) if descend(mid, full)[0] <= 0 else (lo, mid)
h_full = hi
print(f"\nopen loop, full thrust: ignite at {h_full:.0f} m")
for err in (-0.5, 0.5):
    print(f"  ignite {abs(err)} s {'early' if err < 0 else 'late '}: " + outcome(*descend(h_full - err * v0, full)[:2]))

# 2. Closed loop: each instant, command the thrust that would stop the booster exactly at the pad
#    (constant deceleration v^2 / 2h), starting from the height planned for 80 % throttle.
def guide(h, v, m):
    return m * (v * v / (2 * h) + g) / T_max
h_plan = 2919.0                                # where a steady 80 % burn would start
print(f"\nclosed loop, planned for 80 % throttle: ignite at {h_plan:.0f} m")
for err in (-1.0, -0.5, 0.0, 0.5, 1.0):
    h, v, used = descend(h_plan - err * v0, guide)
    print(f"  ignition error {err:+.1f} s: " + outcome(h, v) + f", propellant used {used:5.0f} kg")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Open loop, the timing is unforgiving. Ignite half a second early and the booster stops 125 m up, still with more thrust than weight, and starts to climb; half a second late and it reaches the pad at 66 m/s. Closed loop, with throttle in reserve, it lands softly for any ignition error of up to a second either way, using between 5.1 and 5.4 tonnes of propellant. The real problem is harder: three dimensions, wind, a moving ship, engine limits, a booster that must stay upright, and the requirement to hit a target a few metres across. Engineers formulate it as an optimisation, minimising propellant subject to all those constraints, and solve it on board many times a second; a formulation as a convex problem, which can be solved reliably and fast, was published by Açıkmeşe and Ploen in 2007 for Mars landing. Lumenary Robotics' trajectory optimisation and model predictive control teach the methods.

Who is doing it, as of 2026. SpaceX lands Falcon 9 and Falcon Heavy boosters routinely and has caught Starship's Super Heavy booster on several flights since October 2024; its fully reusable Starship system was still in flight testing at the end of 2025, with a larger version due in 2026. Blue Origin has flown its suborbital New Shepard reusably since 2015 and landed its first orbital-class New Glenn booster on a ship in November 2025. Rocket Lab has recovered Electron boosters from the sea and is developing the reusable Neutron. In China, several companies and state developers were testing reusable first stages in 2024–25 and attempting booster landings from orbital launches by the end of 2025. India's space agency ISRO completed autonomous landing tests of its winged reusable demonstrator in 2023–24, and Europe and Japan are flying reusable-stage demonstrators. Check the companies' and agencies' own announcements for anything later: this is a fast-moving field.

Note

Rockets are dangerous: their propellants are toxic, cryogenic or explosive, and launch sites keep everyone far away. The safe way to explore rocketry is with commercially made, certified model-rocket motors, under the rules of a national model-rocketry association and a club; never make propellants or motors yourself. Simulations like the one above are the other safe route.

ComputationC

Spacecraft thermal control and structures

In orbit there is no air, so a spacecraft exchanges heat only by radiation: it absorbs sunlight (about 1361 W/m² near Earth), sunlight reflected from Earth (albedo) and Earth's own infrared, adds the heat of its electronics, and radiates to the 3 K background of deep space. Its temperature is set by two surface properties: the solar absorptance α\alpha, the fraction of sunlight absorbed, and the infrared emittance ε\varepsilon, how well it radiates at its own temperature. A plate facing the Sun with an insulated back settles where αS=εσT4\alpha S = \varepsilon \sigma T^4, so its temperature depends on the ratio α/ε\alpha/\varepsilon.

sunlight, αSalbedoEarth infraredelectronics,Q internalMLI blanketradiatorεσT⁴ to deepspace (3 K)
The heat balance of a spacecraft in Earth orbit. Sunlight, sunlight reflected from Earth (albedo) and Earth's infrared arrive; electronics add heat inside; multilayer insulation (MLI) blankets most surfaces, and a radiator facing away from the Sun rejects the heat to deep space.
python
sigma, S = 5.670e-8, 1361.0                    # Stefan-Boltzmann constant; sunlight at 1 AU (W/m^2)

# 1. A flat plate facing the Sun, insulated behind: alpha S = eps sigma T^4
surfaces = {  # (solar absorptance alpha, infrared emittance eps), typical beginning-of-life values
    "white paint":                 (0.20, 0.90),
    "black paint":                 (0.95, 0.88),
    "polished aluminium":          (0.15, 0.05),
    "optical solar reflector":     (0.08, 0.80),
    "gold foil":                   (0.25, 0.03),
}
print(" surface                     alpha   eps   alpha/eps  equilibrium in full sun")
for name, (a, e) in surfaces.items():
    T = (a * S / (e * sigma)) ** 0.25
    print(f" {name:27s} {a:5.2f}  {e:5.2f}   {a / e:6.2f}     {T:5.0f} K = {T - 273.15:5.0f} C")

# 2. Sizing a radiator: the hot case sets the area, the cold case sets the heater power
eps = 0.85
q_hot, Q_hot = 150.0, 600.0        # absorbed sunlight reflected by Earth + Earth's infrared (W/m^2); electronics (W)
q_cold, Q_cold = 60.0, 300.0       # in eclipse only Earth's infrared; payload switched off
T_max = 273.15 + 20                # warmest the radiator may run
A = Q_hot / (eps * sigma * T_max**4 - q_hot)
print(f"\nhot case: radiator area for {Q_hot:.0f} W at 20 C = {A:.2f} m^2")
T_cold = ((Q_cold / A + q_cold) / (eps * sigma)) ** 0.25
print(f"cold case: the same radiator would settle at {T_cold - 273.15:.0f} C")
T_min = 273.15 + 0
print(f"heater power to hold it at 0 C: {A * (eps * sigma * T_min**4 - q_cold) - Q_cold:.0f} W")

# 3. How fast does it cool? 35 minutes of eclipse, starting at 20 C, heaters off
m_c = 150.0 * 900.0                # J/K: 150 kg of mostly aluminium coupled to the radiator
T, dt = T_max, 1.0
for second in range(35 * 60):
    T += (Q_cold + q_cold * A - eps * sigma * A * T**4) / m_c * dt
print(f"after a 35-minute eclipse with heaters off: {T - 273.15:.1f} C")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Surface finish is the thermal designer's main tool. White paint and optical solar reflectors (quartz mirrors silvered on the back) absorb little sunlight but radiate well, so they stay cold in full sun; polished metal and gold absorb little but radiate even less, so they run hot. Most of a spacecraft is wrapped in multilayer insulation, many layers of aluminised plastic film separated by thin netting, which isolates it from the Sun and from deep space, and its heat is rejected deliberately through radiators with a low α/ε\alpha/\varepsilon facing away from the Sun. The extreme example is the James Webb Space Telescope, whose five-layer sunshield keeps the instruments on its cold side at around 40 K.

The radiator calculation shows the central dilemma. The hot case (sunlit, full power) sets the radiator's area, 2.9 m² here; in the cold case (eclipse, payload off) the same radiator would cool the spacecraft to −32 °C, so heaters, about 300 W here, or variable devices (louvres, loop heat pipes, variable-emittance coatings) must hold it warm. Thermal mass helps through a short eclipse, during which the structure cools only from 20 °C to 12 °C, but not through a long fault in which the spacecraft has turned itself off to save power. Ammonia loops, heat pipes and pumped fluid loops move heat from where it is made to where it can be radiated; the International Space Station's external loops reject tens of kilowatts.

Structures. A spacecraft's structure must survive a few minutes of launch, and then mostly carry its own weight in zero gravity for years. Launch brings quasi-static accelerations of several gg, random vibration from the engines and the turbulent boundary layer, intense acoustic noise in the fairing, and pyrotechnic shocks at separation. Each launch provider's user's manual specifies these environments and the minimum natural frequencies a payload must have, so that it does not resonate with the rocket (the coupled analysis of vibrations). Structures are mostly aluminium honeycomb panels with aluminium or carbon-fibre skins: very stiff for their mass. The test campaign follows the principle test as you fly: sine and random vibration on a shaker, acoustic tests in a reverberant chamber, shock tests, and thermal-vacuum cycling, first on a qualification model at loads above those expected, then on the flight unit at the expected levels. The European Cooperation for Space Standardization publishes freely available standards for all of it.

In-space manufacturing, in outline. The first 3-D printer on the International Space Station made its first part in 2014, and the European Space Agency tested a metal printer there in 2024. The case for manufacturing in orbit is strongest where microgravity changes the product: crystals of pharmaceuticals and semiconductors that grow more uniformly without convection and sedimentation, and specialty optical fibres that form fewer defects. In 2024 the US company Varda Space Industries returned a capsule to Earth containing crystals of an antiviral drug grown in orbit. Further ahead, agencies and companies study building on the Moon from local soil, which would avoid launching every kilogram. As of 2026 all of this is at demonstration scale, limited by the cost of getting material up and products back down; cheaper launch is what could change that.

DefinitionsH

Fusion engineering: magnets, heat and neutrons

A tokamak confines a plasma of deuterium and tritium, hotter than 100 million kelvin, in a ring-shaped magnetic field, so that the nuclei fuse: D+T→4He (3.5 MeV)+n (14.1 MeV)\mathrm{D + T \to {}^4He\,(3.5\ MeV) + n\,(14.1\ MeV)}. The helium nucleus stays in the plasma and heats it; the neutron, carrying four-fifths of the energy, flies out through the magnetic field into the surrounding structure, where its energy becomes heat for a power plant. The physics of confinement is a subject of its own; the engineering challenges are mechanical, thermal and materials problems of an extreme kind.

central solenoidplasmaneutronsblanket andfirst wallvacuum vesseltoroidal field coildivertor
A cross-section through one side of a tokamak. The D-shaped plasma sits inside a vacuum vessel lined with a breeding blanket; the divertor at the bottom receives the exhaust of heat and particles; superconducting toroidal-field coils loop around the vessel, and a central solenoid drives current in the plasma. Everything sits inside a cryostat.

Magnets. The field that holds the plasma also pushes on the coils that make it, with a magnetic pressure B2/2μ0B^2/2\mu_0. ITER's 18 toroidal-field coils, wound from niobium–tin superconductor cooled to about 4 K, give 5.3 T on the plasma axis and 11.8 T at their peak, and each weighs some 300 tonnes, most of it steel structure to resist the magnetic forces. A newer route uses high-temperature superconductors, tapes coated with rare-earth barium copper oxide (REBCO), which carry far more current in a high field and can run at about 20 K. In 2021 Commonwealth Fusion Systems and MIT tested a large magnet of this kind at 20 T. Higher field allows a much smaller tokamak for the same performance, which is why several private companies have adopted it; the price is enormous stresses, quench protection (a superconductor that loses superconductivity locally must dump its stored energy safely) and a young supply chain for the tape. Lumenary Materials' lesson on superconductors explains the materials.

Heat exhaust. Particles and heat leaving the plasma are guided onto a small area at the bottom of the vessel, the divertor. ITER's divertor targets are blocks of tungsten bonded to water-cooled copper-alloy pipes, designed for 10 MW/m² in steady operation and 20 MW/m² in slow transients, between a sixth and a third of the heat flux leaving the surface of the Sun.

Materials under neutron load. The 14.1 MeV neutrons knock atoms in the walls out of place (damage is counted in displacements per atom, dpa: the first wall of a power plant might see of order ten per full-power year), and their nuclear reactions make hydrogen and helium inside the metal, which embrittles it. Fusion neutrons make much more helium per displacement than fission neutrons do, so experience from fission reactors is only a partial guide. Structural steels for fusion are designed for reduced activation, with elements that do not leave long-lived radioactivity, such as Europe's EUROFER97 and Japan's F82H, but no material has yet been tested under a fusion-like neutron spectrum at power-plant doses. IFMIF-DONES, under construction in Granada, Spain, will provide such a neutron source.

Fuel. Tritium is radioactive, with a half-life of 12.3 years, and almost absent in nature. A power plant must make its own by surrounding the plasma with a breeding blanket containing lithium, in which neutrons react with lithium-6 to make tritium and helium. The tritium breeding ratio must exceed 1 with margin, a hard target because ports, heating systems and the divertor take space from the blanket.

python
import math
mu0 = 4e-7 * math.pi

# 1. Magnets: the field pushes outwards on its coils with a magnetic pressure B^2 / (2 mu0)
print(" field (T)   magnetic pressure (MPa)   about this many atmospheres")
for B in (5.3, 11.8, 20.0):         # ITER on the plasma axis; ITER's peak on its coils; a 2021 HTS test magnet
    p = B**2 / (2 * mu0)
    print(f"   {B:5.1f}          {p / 1e6:6.0f}                   {p / 101325:7.0f}")

# 2. Heat exhaust: steady heat flux through a tungsten armour block into a water-cooled copper-alloy pipe
T_water, h = 100.0, 1.0e5           # coolant temperature (C) and heat-transfer coefficient (W/m^2 K)
layers = [(6e-3, 120.0), (1.5e-3, 320.0)]   # (thickness m, conductivity W/m K): tungsten, then CuCrZr
for q in (5e6, 10e6, 20e6):
    T_surface = T_water + q / h + sum(t / k * q for t, k in layers)
    print(f"heat flux {q / 1e6:4.0f} MW/m^2: tungsten surface about {T_surface:5.0f} C")

# 3. Fuel: tritium burned per year by 1 GW of D-T fusion power (17.6 MeV per reaction)
E_reaction = 17.6e6 * 1.602176634e-19                 # J
reactions = 1e9 * 365.25 * 86400 / E_reaction
m_T = reactions * 3.01605 * 1.66054e-27              # kg of tritium
print(f"\ntritium burned by 1 GW of fusion power for a year: {m_T:.0f} kg")

# 4. Gain: from plasma gain Q to net electricity (illustrative plant, all powers in MW)
def net_electric(P_fus, Q, heat_eff=0.40, blanket_gain=1.15, thermal_eff=0.35, other=60.0):
    P_heat = P_fus / Q                                   # heating power delivered to the plasma
    gross = thermal_eff * (P_fus * blanket_gain + P_heat)  # neutrons multiply a little in the blanket
    return gross - P_heat / heat_eff - other            # minus heating systems' and plant's own use
print("\n plasma gain Q   net electric power from 500 MW of fusion")
for Q in (1, 5, 10, 20, 30, 50):
    print(f"     {Q:4d}          {net_electric(500.0, Q):7.0f} MW")

# NIF, December 2022: 3.15 MJ of fusion from 2.05 MJ of laser light; the lasers drew roughly 300 MJ
print(f"\nNIF 2022: target gain {3.15 / 2.05:.2f}, but fusion energy / grid energy about {3.15 / 300:.3f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The magnetic pressure at 11.8 T is 55 MPa, over 500 atmospheres, and at 20 T it is 159 MPa, a load the coil structure must carry continuously, cycling every time the magnets are energised. The divertor surface reaches about 750 °C at 10 MW/m² and almost 1400 °C at 20 MW/m², where tungsten begins to recrystallise and lose toughness; most of the temperature drop is across the 6 mm of tungsten itself. A gigawatt of fusion power burns 56 kg of tritium a year, while the world's civilian stock, mostly a by-product of Canadian-designed heavy-water reactors, is a few tens of kilograms: fusion plants will have to breed essentially all their own fuel.

ExplanationH

Honest status, late 2026

Taken from the projects' own reports (check their pages for later news):

  • ITER, the international tokamak under construction in southern France, built by seven members (China, the European Union, India, Japan, Korea, Russia and the United States), revised its baseline in 2024: research operation with hydrogen and deuterium plasmas from 2034, full magnetic energy in 2036, and deuterium–tritium operation from 2039, aiming at 500 MW of fusion power from 50 MW of heating (Q=10Q = 10) for several minutes. Its components (magnets, vacuum-vessel sectors, cryostat) are among the largest precision mechanical assemblies ever made.
  • JET, in the UK, set a record of 69 MJ of fusion energy in its final deuterium–tritium experiments in 2023 before closing. Long-pulse superconducting tokamaks have pushed duration: China's EAST and France's WEST reported plasmas of over 1000 seconds in early 2025. Japan and Europe's JT-60SA, the largest operating superconducting tokamak, made its first plasma in 2023, and Korea's KSTAR and Germany's Wendelstein 7-X stellarator continue to set performance and duration records.
  • Lasers. The National Ignition Facility repeated and improved on ignition in 2023–25, with yields of several megajoules. It is a physics facility firing about once a day, not a prototype power plant; inertial fusion energy would need targets fired several times a second, made cheaply in millions, and lasers far more efficient than NIF's.
  • Private companies had raised billions of US dollars by 2025. Commonwealth Fusion Systems is building SPARC in Massachusetts, a compact high-field tokamak intended to show Q>1Q > 1, and has announced a first power plant, ARC, for Virginia in the early 2030s; others in the UK, Canada, Germany, Japan and China pursue tokamaks, stellarators, magnetised-target and laser approaches, and several have signed electricity purchase agreements for the late 2020s and 2030s. These are company targets, and no fusion device has yet produced net electricity.
  • Materials and fuel: no material has been qualified for a power plant's neutron dose, IFMIF-DONES is under construction to provide the test environment, and tritium breeding has been tested only with small blanket mock-ups; ITER will test several blanket modules.

The honest summary: fusion's physics milestones of the 2020s are real, its engineering problems (magnets at scale, heat exhaust, materials, tritium breeding, remote maintenance of radioactive components, reliability) are now the critical path, and a first electricity-producing plant in the 2030s is an ambitious goal, not a certainty. The organisations themselves, the IAEA and EUROfusion, publish the primary information.

Common misconceptionsH

Common misconceptions

Misconception

"Reusable rockets are free to fly again." Each flight costs propellant, payload capacity (11–30% in this lesson's model), inspection and refurbishment, recovery ships and operations, and the upper stage if it is expended. Reuse pays when refurbishment is fast and cheap and boosters fly many times.

Misconception

"Space is cold, so spacecraft need only heaters." A spacecraft in sunlight with the wrong surfaces can overheat badly, and its electronics generate heat that can only be radiated away. Thermal design balances a hot case and a cold case with coatings, insulation, radiators and heaters.

Misconception

"Fusion produces no radioactivity." It produces no long-lived fission products and cannot run away like a fission chain reaction, but its neutrons activate the surrounding structure and its fuel includes radioactive tritium. Reduced-activation materials aim to make that waste short-lived, which is a real and important advantage.

Misconception

"Ignition at NIF means fusion power is solved." It was a landmark in physics: more fusion energy out than laser energy in. Turning it into electricity requires engineering gain well above 1, repetition several times a second, cheap targets and durable chambers, none of which exist yet.

Exercises

Exercises

1. Engineering for space and fusion

Warm-upHuman

An upper stage has a specific impulse of 350 s and burns from 40 t to 10 t. What velocity change does it give, in m/s, to the nearest 10 m/s? (Take g0=9.80665g_0 = 9.80665 m/s².)

2. Engineering for space and fusion

CoreHuman

A first stage with a dry mass of 25 t (including legs and fins) must keep enough propellant to give itself 1.2 km/s for its entry and landing burns, with an average specific impulse of 300 s. How many tonnes of propellant must it hold in reserve? Give the nearest whole number.

3. Engineering for space and fusion

CoreHuman

Why did the lesson's booster land softly under closed-loop guidance planned for 80% throttle, but not under an open-loop full-thrust burn with the same half-second timing error?

Choose one

4. Engineering for space and fusion

CoreComputational

Write radiator(Q_hot, Q_cold, eps, T_hot_C, T_cold_C, q_hot, q_cold) returning (area, heater_W). The area is set by the hot case: Q_hot watts of electronics must be rejected at T_hot_C while the radiator absorbs q_hot W/m² from the environment. In the cold case (Q_cold watts, q_cold W/m² absorbed), return the heater power needed to hold the radiator at T_cold_C, or 0.0 if no heater is needed. Use σ=5.670×10−8\sigma = 5.670 \times 10^{-8} W/(m²·K⁴).

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

5. Engineering for space and fusion

Warm-upHuman

A fusion magnet's conductor sits in a field of 13 T. What is the magnetic pressure B2/2μ0B^2/2\mu_0 on it, in MPa, to the nearest whole number? (μ0=4π×10−7\mu_0 = 4\pi \times 10^{-7} H/m.)

6. Engineering for space and fusion

CoreHuman

A D–T fusion device produces 500 MW of fusion power, but only for a quarter of the year on average. Each reaction releases 17.6 MeV and consumes one tritium nucleus (3.016 u). How many kilograms of tritium does it burn in a year? Give the nearest whole number.

7. Engineering for space and fusion

StretchComputational

Using the lesson's plant model, net electric power is

Pnet=ηth (gbPfus+Ph)−Phηh−Pother,Ph=PfusQ.P_\text{net} = \eta_\text{th}\,(g_b P_\text{fus} + P_h) - \frac{P_h}{\eta_h} - P_\text{other}, \qquad P_h = \frac{P_\text{fus}}{Q}.

Write q_needed(P_fus, heat_eff=0.40, blanket_gain=1.15, thermal_eff=0.35, other=60.0) returning the smallest plasma gain QQ for which Pnet>0P_\text{net} > 0 (the break-even value), or None if no gain, however large, gives net electricity.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

8. Engineering for space and fusion

CoreHuman

Which statements about fusion gain are correct? Select all that apply.

Choose all that apply

Further reading and resources

  • Autonomous Precision Landing of Space Rockets · Lars Blackmore · 2016The Bridge (US National Academy of Engineering) 46(4). How rocket landing is posed as a convex optimisation problem.
  • Convex Programming Approach to Powered Descent Guidance for Mars Landing · Behçet Açıkmeşe and Scott Ploen · 2007Journal of Guidance, Control, and Dynamics 30(5), 1353–1366.
  • NASA Glenn Research Center, Beginner's Guide to RocketsClear explanations of thrust, specific impulse and the rocket equation.
  • ESA Space Engineering and TechnologyThe European Space Agency's engineering pages, including thermal control and structures.
  • ECSS (European Cooperation for Space Standardization)Freely available standards for space engineering, including thermal control and structural testing.
  • Spacecraft Thermal Control Handbook · David Gilmore (editor) · 2002The Aerospace Corporation's standard reference, volume 1 covering fundamental technologies.
  • ITERThe international fusion project in France, its design, construction and 2024 baseline schedule.
  • EUROfusionThe European fusion research consortium, which ran JET's final experiments and develops the DEMO design.
  • IFMIF-DONESThe neutron source being built in Granada, Spain, to test fusion materials.
  • Lawrence Livermore National Laboratory, National Ignition FacilityReports of ignition in December 2022 and later experiments.
  • Commonwealth Fusion SystemsBuilder of the SPARC tokamak with high-temperature superconducting magnets.
  • IAEA World Fusion Outlook · International Atomic Energy AgencyAn annual survey of public and private fusion projects worldwide.