The Chernobyl accident, step by step

Chernobyl 4 was an RBMK-1000: 3,200 thermal megawatts, 1,000 electrical, a graphite core twelve metres across crossed by 1,661 vertical pressure tubes carrying light water that boils inside the channel itself. Graphite provides most moderation; water cools, absorbs neutrons and also contributes to moderation. That split of duties defines everything that followed.

In a PWR or a BWR the water is both moderator and coolant: if bubbles form, there is less moderation, fission falls and power drops by itself. In the RBMK the moderator is solid and stays where it is. Bubbles only remove water, which was also absorbing neutrons, so removing it adds reactivity. That is positive void feedback. In that state, the full coolant voiding effect — a reactivity, rather than a coefficient per unit void — was of the order of +4 to +5 β: more steam, more power, more steam.

Voiding is only part of it. What governs a change in power is the power coefficient of reactivity, which adds up every feedback at once, and in the RBMK it has three contributions with different signs. Fuel heating is the one that pushes back: as the temperature rises, Doppler broadening of the uranium-238 resonances captures more neutrons and reactivity falls, and it does so promptly, at the pace of the power itself. Coolant voiding adds. Graphite heating adds as well. At full power the fuel term dominates and the sum is negative; in the state Chernobyl 4 was in that night it did not dominate, and the power coefficient itself became positive.

On top of that, the reactor had no containment building. It sat in an industrial hall with a light roof, because the worst accident the design contemplated was the rupture of one or two tubes, not of the whole core.

The test itself was reasonable: check that, after a loss of offsite power, the coasting turbogenerator could feed the cooling pumps until the diesels came up. It had been postponed since 1982 and was run at the end of a long day, with the reactor on its way to a scheduled shutdown.

On 25 April the reactor came down from 3,200 to 1,600 MWt and stayed there for some nine hours at the grid dispatcher's request. At 00:28 on the 26th, during a transfer between regulating systems, power collapsed to about 30 MWt. In a reactor that has been at high power, xenon-135 — a fission product that devours neutrons — builds up exactly when power falls: the reactor poisons itself and will not come back up.

Getting it to the 200 MWt at which the test was finally run meant pulling rods, and pulling rods, and pulling rods. At 01:22:30 the calculation gave an operating reactivity margin of 8 equivalent rods, against a regulatory minimum of 15. That margin was neither displayed to the operator nor wired into the protection system: it had to be requested. The reactor was left at low power, with xenon in the middle, with almost every rod withdrawn and with very little steam flow: precisely the state in which the void coefficient is most dangerous.

At 01:23:04 the turbogenerator inlet valves closed and the machine began to slow. Four of the eight main circulation pumps were supplied by that generator and slowed down; the other four remained on grid power. Less water through each channel means the water heats up more on its way through, and with the reactor already close to saturation that means more steam.

Positive void feedback made this state dangerous. At 01:23:40 AZ-5 was pressed, commanding the protection rods to insert. INSAG-7 does not establish with certainty why the button was pressed or a preceding power excursion: it identifies positive scram as the most likely final trigger, while retaining uncertainty about the detailed reconstruction.

Of the 211 control rods, 187 entered from above at 0.4 metres per second: eighteen to twenty seconds to cross the seven metres of core. The other 24 are shortened rods that come up from below, and their job is to flatten the power profile in the lower half; the ones that matter here are those from above. Because an RBMK rod was not only absorber: nearly all of them carried a 4.5-metre graphite displacer hanging from the boron carbide, and it was there for neutron economy. With the rod withdrawn, the displacer filled the channel so it would not fill with water, which absorbs neutrons and makes fuel more expensive, and it sat centred in the fuelled region, leaving 1.25 metres of water above it and 1.25 metres of water below.

There is the mechanism. When the rod starts down, the first thing to enter that bottom metre and a quarter is not the absorber: it is the graphite tip, pushing the water out of that part of the channel. In an instant, at the bottom of the core, an absorber — water — was removed and a moderator — graphite — was put in its place. Local reactivity rises. The rod keeps going down and seconds later the boron shuts that region off, but for those seconds the emergency shutdown begins by adding reactivity instead of removing it. INSAG-7 calls it the positive scram effect, and notes that its magnitude depends on how power is distributed through the core.

That night it was distributed in the worst possible way. With xenon poisoning the middle and the rods withdrawn, the axial power profile had upper and lower peaks separated by a depressed central region. Initial displacer insertion particularly affected the lower peak. The yardstick for this is β, the fraction of neutrons that come not from fission itself but from the later decay of some fragments, about six thousandths of the total: while reactivity stays under 1 β — one dollar — the chain depends on those late neutrons and the reactor is governed in seconds. Cross the dollar and the chain sustains itself on prompt neutrons alone, and the characteristic time becomes milliseconds.

Initial reactivity insertion in the lower region and positive void feedback made an extremely rapid excursion possible. The prompt-critical threshold applies to the effective reactivity of the whole reactor, not to a local-reactivity reading. From that point no rod travelling at 0.4 metres per second was going to arrive in time: the reactor was no longer governed by mechanisms, it governed itself, by destroying itself.

At 01:23:43 power-excursion signals activated; the system recorded more than 530 MWt. Destruction followed within seconds. The final peak depends on reconstructions and is not precisely known: the diagram shows a qualitative trend, not a measured curve or a uniform time scale.

Rapid energy deposition damaged and fragmented the fuel, transferring heat to coolant and generating steam violently. Channel rupture and pressure rise destroyed the reactor and displaced its upper shield. Two explosions were heard; the precise contribution of steam and chemical reactions to each is less certain than the recorded signal sequence.

With no containment building, that energy took the roof off the hall and left the core open to the air. The graphite burned for ten days and the fire's thermal plume carried fission products kilometres up: some 14 EBq of total radioactivity, 5,200 PBq in iodine-131 equivalent. Two workers died in the explosions and 28 more of acute radiation syndrome in the following weeks, most of them firefighters and plant staff. Pripyat, three kilometres away, was evacuated on the afternoon of the 27th; the 30-kilometre zone, some 116,000 people, during 1986.

The IAEA's first report, INSAG-1 in 1986, put the blame on the operators. In 1992, with the full Soviet data on the table, INSAG-7 corrected it: the operators did break rules, but the accident happened because the reactor could reach a state in which its own emergency shutdown blew it up, and that was in no operating manual.

The RBMKs that kept running were modified. Enrichment went from 2.0 to 2.4% and later to 2.8% with erbium as a burnable poison; between 80 and 90 fixed absorbers were installed in fuel channels; the minimum rod margin went from 26-30 to 43-48 equivalent rods; the rods were redesigned so that, withdrawn, no water was left beneath the displacer; shutdown time went from 18 to 12 seconds and a fast system of 24 rods was added that inserts at least 2 β of negative reactivity in under 2.5 seconds. The void coefficient ended up of the order of +β.

Unit 4 was covered by the sarcophagus in November 1986 and since 2016 sits under the New Safe Confinement, a 108-metre-tall steel arch assembled alongside and slid over the building. The other three units kept generating: unit 2 until 1991, unit 1 until 1996 and unit 3 until December 2000.