The Missing Three Thousandths

In 1972, an analyst at a French nuclear fuel plant ran a routine check on a uranium sample and got 0.7171%.

The number should have been 0.7202%. That’s the fraction of uranium-235 in natural uranium, and it’s one of the most boring numbers in the industry, in the best sense. It barely varies from one mine to another. You could dig uranium out of Canada or Niger or Australia and measure it, and the U-235 would sit at 0.720-something percent like it had been told to. It is the kind of number that you check because checking it is the procedure, not because you expect to learn anything.

The gap between the two figures is three thousandths of a percentage point. In most jobs, that’s noise. In nuclear materials accounting, it’s an alarm. U-235 is the isotope that makes bombs, and the entire safeguards system rests on the idea that every gram of it can be accounted for. A sample that’s light on U-235 means one of two things: an instrument is wrong, or someone has taken some out.

Neither was true. What had actually happened was much stranger, and it took the French Atomic Energy Commission a couple of months to believe it: the uranium had already been through a nuclear reactor. A natural one. About two billion years ago.

---

The audit

The plant was at Pierrelatte, in the Rhône valley, and the analyst was named H. Bouzigues. (I should say that the credit isn’t perfectly settled. A CEA chemist named J.-F. Dozol has given his own account in which the anomaly was caught as uranium was being prepared for shipment, and his mass spectrometry was what found the first fingerprint of fission. Popular retellings often also have the physicist Francis Perrin, a former head of French atomic energy, exclaiming “This cannot be possible,” which is a good line that I can’t trace to a primary source. Big discoveries collect claimants and quotes the way ships collect barnacles.)

What everyone agrees on is how the investigation went, because it went the way any good audit does: follow the number upstream.

The low-U-235 uranium led back to a mill at Mounana, near Franceville in Gabon, which processed ore from a mine called Oklo. Depleted shipments had been leaving since 1970 without anyone noticing. By the middle of 1972 they added up to roughly 700 tons of uranium, short by around 200 kilograms of U-235.

Two hundred kilograms of U-235 is enough for something like half a dozen bombs. You can imagine the mood in the room.

But the missing isotope hadn’t been extracted. When the CEA analyzed ore from the deposit, they found fission products: the particular mix of lighter elements and isotopes that you only get when uranium nuclei split. In some cores the U-235 fraction had dropped as low as 0.44%. The uranium hadn’t been taken anywhere. It had been burned, where it lay, by a chain reaction. The books were short because the rock had spent the money.

The CEA went public on September 25, 1972.

The recipe

What’s satisfying about Oklo is that it wasn’t a total surprise to everyone. It had been predicted, quietly, sixteen years earlier.

In 1956, Paul Kuroda, a chemist at the University of Arkansas, published a two-page paper called “On the Nuclear Physical Stability of the Uranium Minerals.” In it he worked out what a uranium deposit would need in order to go critical by itself. There are four ingredients.

The deposit has to be thick enough. A fission neutron travels on the order of two-thirds of a meter before it’s absorbed, so a thin seam leaks neutrons out the sides before they can do anything useful.

There has to be enough U-235. This is the fuel; U-238, the other 99.3% of natural uranium, mostly just sits there.

There has to be a moderator. This is the part people usually skip, so here’s the short version. The neutrons that come out of a fission are fast, and fast neutrons tend to fly straight past U-235 nuclei. Slow them down, by bouncing them off light atoms like the hydrogen in water, and they become much more likely to get captured and split the next nucleus. A moderator is a speed bump for neutrons.

And there must be very little in the way of “neutron poisons,” elements like boron and lithium that soak up neutrons and kill the chain.

Today, nowhere on Earth can hit the second requirement. At 0.72% U-235, you can’t get a sustained chain reaction with ordinary water as a moderator. That’s why the fuel in a normal power reactor is enriched to around 3–5%.

But the ratio hasn’t always been 0.72%. U-235 decays roughly six times faster than U-238 (its half-life is about 704 million years), so the further back you go, the more of it there was. Roll the clock back two billion years and natural uranium was about 3% U-235. That’s light-water reactor fuel, made by nothing but time.

Oklo had the rest of the recipe too. Its reactor zones were absurdly rich, 20 to 30% uranium by some measurements, against roughly half a percent in the ore around them. They were thick. And they were wet.

The geyser

Here’s the part of Oklo I think about most.

A reactor that runs away is dangerous, and a big share of nuclear engineering is about designing feedback that stops a reaction from growing out of control. Oklo had that feedback built in, and it was extremely simple.

In 2004, Alex Meshik and his colleagues at Washington University in St. Louis worked out the reactor’s rhythm from xenon isotopes trapped inside mineral grains from one of the reaction zones. Xenon is a fission product, and different xenon isotopes get produced, decay and escape at different rates, so the mix trapped in a grain is a kind of timing record. Their reading of it: groundwater seeped into the ore and slowed the neutrons, and the reaction started. The reaction heated the rock until the water boiled off. Without water, there was no moderator, so the reaction stopped. The rock cooled, the water came back, and it started again.

About 30 minutes on. About two and a half hours off. For what may have been a few hundred thousand years.

This is called negative feedback, and engineers would kill for a version this reliable. The reactor’s own output removed the thing it needed to keep going. It was physically incapable of a runaway, because the moment it got too hot it lost its moderator. It ran like a geyser, breathing in water and puffing it out as steam, on a schedule nobody set.

It also wasn’t very powerful. Meshik estimated average output at probably under 100 kilowatts, which he described as “a few dozen toasters.” But it kept going, on and off, long enough to produce something like 15,000 megawatt-years of energy in total. Nobody knows exactly how long it ran. Estimates range from about 150,000 years to several hundred thousand, which is an awkward amount of uncertainty until you remember the error bars are smaller than the whole of human history.

As mining went on, more zones turned up. Six had been identified by the mid-1970s. Today the count is sixteen in the Oklo and Okélobondo area, plus one at Bangombé, about 35 kilometers away.

The receipts

The rock kept records of everything it did, which turned out to be the most useful thing about it.

It made plutonium. More than two tons of plutonium-239 formed in the reaction zones as U-238 caught neutrons. Plutonium-239 has a half-life of about 24,000 years, so it’s long gone, but it decayed back into U-235 — so much that almost half the U-235 remaining in the reactor zones today was once plutonium. The reactor partially refueled itself, on a delay of geological time.

It trapped its own exhaust. Grains of aluminum phosphate in the zones hold the highest concentration of fission xenon ever measured in a natural material, gas that has been sitting in the same crystals for nearly half the age of the Earth. That’s what Meshik read.

And it ran a two-billion-year experiment in nuclear waste disposal, with no supervision. Most of the heavy, non-volatile fission products and actinides stayed within centimeters of where they were made, which is why people designing geological repositories study Oklo as a natural analogue. I want to be fair to the evidence here, though, because this gets oversold. Some elements did move, and significantly: krypton, xenon, cesium, strontium, barium, molybdenum and iodine. The honest summary is that the rock held onto the worst of it, not all of it.

A couple of the old drill cores are in the Natural History Museum in Vienna. At five centimeters away, they give off about the radiation dose you’d pick up from cosmic rays on an eight-hour flight from Vienna to New York. The reactor’s been off for two billion years and it’s still faintly warm, in the radiological sense.

The second audit

Now the part where the essay comes back to where it started.

Bouzigues’s question, at Pierrelatte, was: is this number that’s supposed to be constant actually constant? It wasn’t, and the reason was a reactor.

In 1976, a young Soviet physicist named Alexander Shlyakhter took the reactor and asked the same question about the universe.

Some nuclear reactions are extremely picky about energy. Samarium-149 is one of them. It swallows slow neutrons enormously well, but only because of a resonance, a narrow energy window where capture is very likely. Where that window sits depends on the balance of forces inside the nucleus, including electromagnetism. And the strength of electromagnetism is set by a number called the fine-structure constant, about 1/137, which shows up throughout physics and which nobody can derive from anything else. If it had been even a tiny bit different two billion years ago, samarium’s resonance would have sat somewhere else, and the Oklo reactor would have eaten through its samarium at a different rate.

So you can look at how much samarium-149 is left in the reaction zones and work backward. Shlyakhter did, in a short paper in Nature, and found that the resonance couldn’t have moved by more than a tiny amount. Twenty years later, Thibault Damour and Freeman Dyson redid the analysis carefully and tightened the bound. Over two billion years, they found, the fine-structure constant can’t have changed by more than about one part in ten million.

(Again, in the spirit of full disclosure: not every analysis agrees. In 2004 a pair of physicists used Oklo data to argue that the constant had shifted slightly. Most of the work since has come down on the side of no change. Oklo isn’t a closed case so much as an unusually good witness.)

Think about what this is. In 1972, a lab tech checks a number in a ledger, sees it’s off by three thousandths, and follows it upstream to a reactor that predates the first animals by more than a billion years. Four years later someone uses the same reactor’s leftovers to check a different ledger: whether the rules of physics on the day the reactor ran were the same as the rules on the day he looked.

They were. To one part in ten million, two billion years apart, the universe had been charging the same price for everything.

---