Nuclear power is the only way of making electricity where stopping the reaction does not stop the heat. Everything distinctive about a reactor — the emergency cooling, the redundant power supplies, the containment, the years a fuel assembly spends underwater — follows from that one fact. This page covers the physics and the engineering, and it is careful with the numbers that are usually quoted wrong.
A uranium-235 nucleus absorbs a slow neutron, becomes U-236 in a highly excited state, and within about a millionth of a millionth of a second splits into two lighter nuclei. The fragments fly apart carrying most of the released energy as kinetic energy, which becomes heat when they stop in the fuel. Two or three free neutrons come with them, and those neutrons are what make a reactor possible.
The recoverable energy is about 200 MeV per fission. Burning one carbon atom releases about 4 eV. The ratio is roughly fifty million to one, and it is the entire reason nuclear fuel logistics look nothing like fossil fuel logistics. A plant that burns coal receives unit trains. A plant that fissions uranium receives a truck.
| About 3.6 million to one | Complete fission of a kilogram of pure U-235 releases about 82 TJ thermal. Correct, and a bound no reactor comes near, because a fuel assembly is a few percent U-235 and is discharged long before that is used up. | Physics |
| About 21,000 to one | One tonne of natural uranium through a once-through light water reactor yields roughly 44 million kilowatt-hours, which the World Nuclear Association puts at over 20,000 tonnes of coal. This is the number that describes an actual plant. | Engineering |
The two differ by a factor of about 170. Both get quoted as "a kilogram of uranium equals X of coal", and the first is routinely used while the second is being described. If you are comparing fuel logistics for a real plant, the engineering number is the one you want.
A chain reaction is described by k-effective, the ratio of neutrons in one generation to the last. Below one the reaction dies away; at exactly one it sustains itself; above one it grows. A reactor at steady power is critical, k equals one, and it is held there.
If that were the whole story a reactor would be uncontrollable. The prompt neutron lifetime in a thermal reactor is on the order of tens of microseconds; a chain reaction running on prompt neutrons alone would double in power faster than any mechanical system could respond. What makes reactor control physically possible is that a small fraction of the neutrons do not arrive with the fission at all.
| β for U-235 | About 0.0065 to 0.0068 for thermal fission, depending on the evaluated nuclear data library and the incident neutron energy. Figures of 0.64, 0.65, 0.66 and 0.70 percent are all in circulation and are not simply errors. | The isotope |
| β-effective for the core | Weights delayed neutrons by how effective they are at causing further fission. They are born colder than prompt neutrons, around 0.5 MeV against 2 MeV, which changes their survival. It is β-effective, not β, that governs kinetics. | The reactor |
| β-effective at end of cycle | Falls through the cycle. Pu-239 has a delayed fraction of about 0.23 percent against U-235 at 0.66, and by end of cycle plutonium supplies roughly a third of the power, so the composite drops toward about 0.005. | The moment |
This is the reason an end-of-cycle core is measurably twitchier than a fresh one, and it is a real operational effect rather than a curiosity. "About two-thirds of one percent" is an honest statement for U-235. A four-digit β-effective without a named core state and data library is not.
A nuclear weapon requires a supercritical mass of highly enriched fissile material assembled fast enough that the chain reaction runs to completion on prompt neutrons before the assembly blows itself apart. Every one of those conditions is absent from a power reactor, and not by accident.
A well-designed reactor pushes back against its own power increases without anyone doing anything. Three feedbacks do most of that work, and their signs are the single most important design characteristic a reactor has.
| Feedback | Mechanism | Sign in a light water reactor |
|---|---|---|
| Fuel temperature, or Doppler | As fuel heats, resonance absorption peaks in U-238 broaden and capture more neutrons. Acts within milliseconds, in the fuel itself, with no delay for heat transfer. | Negative. Always, and it is the fastest protection the core has |
| Moderator temperature | Hotter water is less dense, so it moderates less well and absorbs less. In an undermoderated lattice the net effect is fewer thermal neutrons. | Negative in normal operation, though it can go positive at high boron concentration, which is why beginning-of-cycle boron is limited |
| Void coefficient | What happens to reactivity when coolant boils and voids form. In a light water reactor the water is the moderator, so voiding removes moderation. | Negative. Boiling shuts a light water reactor down |
Fission neutrons are born fast, around 2 MeV. U-235's fission cross-section is far larger for slow neutrons than for fast ones, so a thermal reactor has to slow them down — ideally with light nuclei, which take more energy per collision, and without absorbing them on the way.
Those two requirements pull against each other, and the choice of moderator sets the entire rest of the design. Light water is an excellent moderator and a significant absorber, so a light water reactor must make up the loss with enriched fuel. Heavy water and graphite absorb far less, so reactors built around them can run on natural uranium — which is why countries without enrichment capability built CANDUs and gas-cooled reactors.
| Moderator | Neutron economy | Consequence for the design |
|---|---|---|
| Light water | Excellent slowing down, meaningful absorption | Requires 3 to 5 percent enrichment. Compact core. Coolant and moderator are the same fluid, which gives the negative void coefficient for free |
| Heavy water | Good slowing down, very low absorption | Runs on natural uranium. Larger core, on-load refuelling, and an expensive initial heavy water inventory |
| Graphite | Slower per collision, very low absorption | Runs on natural or slightly enriched fuel. Very large core. If cooled by water, the void coefficient can be positive — see the RBMK above |
Coolant is a separate choice, and it need not be the moderator at all. Light and heavy water, carbon dioxide, helium, liquid sodium, molten salt and lead have all been used or seriously proposed. A fast reactor dispenses with moderation entirely and uses a coolant chosen specifically not to slow neutrons down, which is why sodium and lead appear there and water does not.
Reactivity control operates on three quite different timescales, using three different mechanisms, and it is worth keeping them apart.
A scram stops the fission chain reaction in about two seconds. It does not stop the heat. The fission products already in the fuel keep decaying, and they do not care what the control rods are doing. Immediately after shutdown a reactor is still producing on the order of 6 to 7 percent of full thermal power, and it will keep producing heat, in declining amounts, for years.
Made concrete: one hour after the earthquake scram on 11 March 2011, Fukushima Daiichi Unit 1 was producing about 22 MW of decay heat, and Units 2 and 3 about 33 MW each. Thirty-three megawatts is the thermal output of a large industrial boiler, being generated inside a sealed pressure vessel with no way to turn it off. Two months later Unit 1 was still at 1.8 MW. Five years later the three units together were down to about 1 MW, at which point cooling injection could be interrupted for a couple of days.
One caution on the headline percentage. The often-quoted 6 percent is the fission-product decay contribution during steady operation; the 7 percent is total heat immediately after scram, including residual fissions from delayed neutrons still arriving in the first seconds. Both are right, and the actual value depends on how long the core has been running and at what burnup. It is not a figure to quote to more than two significant digits without stating the operating history.
Everything after the core is a steam plant. Heat raises steam, steam turns a turbine, the turbine turns a generator, and the exhaust is condensed and pumped back. It is a Rankine cycle, and its efficiency is set by the temperature and pressure of the steam it can make.
| Design | How the steam is made | What that costs you |
|---|---|---|
| Pressurized water reactor | Primary water is kept liquid at about 155 bar and passes through steam generators, boiling a separate secondary loop. Two loops, so the turbine side stays clean. | Steam generators are large, expensive, and historically a leading source of plant problems. Secondary steam is cooler than the primary water |
| Boiling water reactor | Water boils in the core and the steam goes straight to the turbine. One loop. | No steam generators, simpler and cheaper, but the turbine hall is part of the radiological envelope because the steam has been through the core |
| What is usually said | A nuclear plant is about 33 percent efficient and a modern combined-cycle gas plant is about 60 percent. The comparison is everywhere and it is not apples-to-apples. | Common |
| Why it is not comparable | The 60 percent is a lower-heating-value figure for a new H-class machine at ISO conditions. US statistics use higher heating value, and 60 percent LHV is about 54 percent HHV. Uranium contains no hydrogen, so a reactor efficiency is neither LHV nor HHV, it is simply thermal. | The basis |
| Same basis, same year | EIA tested heat rates for 2024, net, fleet average: nuclear 32.7 percent against combined-cycle gas at 45.2 percent. Coal, for reference, is 34.1 percent — very slightly better than nuclear. | Honest |
The gap between nuclear and gas is real and large. It is just not 33 against 60. The reason a reactor gives up efficiency is steam temperature: a light water reactor is limited by the pressure at which you are willing to keep water liquid, so it makes saturated steam at roughly 285 degrees Celsius, where a supercritical fossil boiler runs far hotter.
Efficiency also matters far less economically for nuclear than for gas. Uranium is a negligible share of a reactor's generating cost, so a 33 percent efficient reactor is wasting something nearly free. A 45 percent efficient gas plant is wasting its dominant operating cost. What low efficiency does cost the reactor is capital: two thirds of three thousand megawatts thermal has to be rejected to the environment, which means a bigger condenser, more cooling water or a bigger tower, and a bigger low-pressure turbine.
| Type | Coolant and moderator | Fuel | Distinguishing characteristic |
|---|---|---|---|
| PWR | Light water, both | 3–5% enriched UO₂ | The world workhorse, and the basis of naval propulsion. Two loops. Negative void coefficient. VVER is the Russian PWR line and is usually counted with PWRs |
| BWR | Light water, both | 3–5% enriched UO₂ | Direct cycle, no steam generators. Control rods enter from below, since the top is full of steam separators |
| PHWR / CANDU | Heavy water, both | Natural uranium | Runs on unenriched fuel and refuels on load through pressure tubes rather than a single pressure vessel. Positive void coefficient, offset by other design features |
| Magnox / AGR | CO₂ cooled, graphite moderated | Natural (Magnox) or ~3% enriched (AGR) | The British line. Gas coolant allows higher outlet temperature and better thermal efficiency than an LWR |
| RBMK | Light water cooled, graphite moderated | ~2% enriched | Channel type, no full containment as Western practice understands it, and a void coefficient that could go strongly positive. The Chernobyl design |
| Fast reactor | Sodium or lead, no moderator | Higher enrichment or MOX | No moderation at all, so it can fission U-238 and transmute actinides. Sodium burns in air and reacts violently with water, which drives the entire plant design |
Counts by type are harder to state cleanly than they look. IAEA's Power Reactor Information System is the canonical source but publishes through an application that cannot be read as a static page, and the World Nuclear Association's own pages disagree with themselves on the number of operating RBMKs. Neither PRIS nor WNA breaks out VVERs separately from PWRs. If you need a type census for anything that matters, pull PRIS directly rather than trusting a secondary table, including this one.
| Enrichment level | U-235 content | What it is for |
|---|---|---|
| Natural uranium | 0.711% | CANDU and Magnox fuel, and the feed for everything else |
| Low enriched uranium | Under 5% | Essentially all commercial light water fuel |
| HALEU | 5 to 20% | Most advanced reactor designs. Western supply barely exists |
| Highly enriched uranium | Over 20% | Research and naval reactors. A proliferation category, not a performance one |
| Weapons grade | About 90% | Not a civil fuel at any point in any commercial cycle |
| Once through | Fuel is used once and the assembly is stored intact. The United States has done this since a 1977 policy decision against commercial reprocessing, taken on proliferation grounds, which was never reversed in practice. | US, Sweden, Canada |
| Closed, or partly closed | Spent fuel is chemically separated, usually by the PUREX process, and the recovered plutonium and uranium are made into mixed oxide fuel. Recovers energy and reduces the volume of high-level waste, at the cost of separating plutonium. | France, Russia, Japan |
Be careful with world reprocessing capacity figures. The World Nuclear Association table still lists the Sellafield THORP and Magnox plants, both of which have closed, which overstates the total substantially. This is a live error in a widely cited source rather than an old edition.
| Quantity | What it measures | SI unit | Older unit |
|---|---|---|---|
| Activity | Disintegrations per second in the source. Says nothing about anyone being exposed | Becquerel (Bq) | Curie (Ci) |
| Absorbed dose | Energy deposited per unit mass of tissue: one joule per kilogram | Gray (Gy) | Rad |
| Equivalent dose | Absorbed dose weighted for how damaging the radiation type is | Sievert (Sv) | Rem |
| Effective dose | Equivalent dose weighted again for the sensitivity of each organ irradiated | Sievert (Sv) | Rem |
The weighting factors are why a becquerel figure tells you almost nothing on its own. Photons and beta particles carry a factor of 1; alpha particles carry 20; neutrons vary with energy. An alpha emitter is nearly harmless outside the body, because the particles cannot penetrate the dead outer layer of skin, and is among the most dangerous things there is once inhaled or ingested. External and internal hazard are different problems with different controls.
| Exposure | Approximate effective dose |
|---|---|
| US average annual dose, all sources | About 620 mrem (6.2 mSv). Roughly half is natural, and radon is the largest single natural contributor |
| Annual public dose limit from a licensed facility | 100 mrem (1 mSv), 10 CFR 20.1301 |
| Annual occupational limit, whole body | 5 rem (50 mSv) TEDE, 10 CFR 20.1201 |
| Occupational limit, lens of the eye | 15 rem (150 mSv) — note that ICRP recommends far lower and the NRC has not adopted it |
| Declared pregnant worker, whole gestation | 0.5 rem (5 mSv) to the embryo or fetus, 10 CFR 20.1208 |
| Onset of acute radiation syndrome, short-term whole body | Around 100 rem (1 Sv) |
| LD50 without medical care, short-term whole body | Roughly 400 to 500 rem (4 to 5 Sv) |
At high doses the relationship between radiation and cancer risk is measured and not in serious dispute. At low doses — below roughly 100 mSv — the excess risk, if any, is smaller than the statistical noise in any feasible study. What to assume in that region is a genuine scientific and regulatory argument, and the honest thing is to lay out the positions rather than pick one.
| Linear no-threshold | Risk is proportional to dose all the way down, with no threshold. Adopted by the NRC, ICRP and the National Academies BEIR VII committee as a deliberately conservative basis for radiation protection. | Regulatory basis |
| Threshold or non-linear | Below some dose the repair mechanisms handle the damage and there is no excess risk. The French Academy of Sciences and Academy of Medicine argued in 2005 that LNT is not supported by the biology at low doses. | Contested |
| Hormesis | Very low doses are net beneficial by stimulating repair. A minority position with some experimental support and no acceptance in radiation protection standards. | Minority |
What matters practically is not who is right but what LNT may be used for. ICRP and UNSCEAR endorse it for optimizing protection and explicitly warn against using it to compute death tolls by multiplying tiny individual doses across very large populations. Nearly every headline casualty projection you will read for Chernobyl or Fukushima does exactly that, and the bodies whose risk model is being used say it should not be done.
Reactor safety is organized around three functions that must be maintained: control reactivity, cool the fuel, and contain the radioactivity. Everything else is implementation. The physical protection is layered so that no single failure reaches the public.
Since Three Mile Island the industry has quantified all of this rather than arguing about it qualitatively. Probabilistic risk assessment estimates core damage frequency and large early release frequency per reactor-year. The NRC's guidance treats a core damage frequency around one in ten thousand reactor-years as the reference point for current plants, and the best operating plants are assessed nearer one in a million. These are model outputs with real uncertainty attached, not measurements, and they should be read as the outputs of an analysis of record for a specific plant.
On comparative risk: several published analyses put deaths per terawatt-hour far lower for nuclear than for coal or oil, and the broad conclusion is robust. The specific decimal is not. The most-circulated figure derives its nuclear numerator largely from Chernobyl projections and from Fukushima evacuation deaths, which means the numerator is dominated by deaths that radiation did not cause. Cite the band, not the decimal.