The reactor that runs on rock

Rock, refined, into a reactor

A reactor is a chain. Every atom that splits throws out neutrons, and those neutrons have to hit the next atom or the chain stops. Fresh out of a split they are far too fast to be caught, so you slow them down by letting them bounce off something light. Water is the obvious choice.

The problem is that the hydrogen in ordinary water does not only slow neutrons down. It eats them. Enough of them that natural uranium, which is only 0.72% of the isotope that splits, cannot hold a chain going in ordinary water at all.

Everywhere else the fix is to change the fuel: concentrate U-235 from under one percent up to three, four, five, which needs an enrichment plant. Canada changed the water instead.

What the water eats, over a path you drag

Swap hydrogen for deuterium, its heavier twin with one extra neutron in the nucleus, and the absorption falls off a cliff. These are the published thermal absorption cross sections, and the ratio between them is the entire argument for the design.

nucleus	thermal absorption cross section
hydrogen, H-1	0.332 barns
deuterium, H-2	0.00052 barns
ratio	638 to 1

Cross sections are per nucleus, so to get what a tank of water actually does you multiply by how many nuclei are in the way. That gives a macroscopic cross section, in units of one over length, and the surviving fraction falls exponentially along the path:

Drag the path and watch the two liquids separate. Same neutrons, same distance, same physics.

 

The 638 to 1 in the table is the ratio of the cross sections themselves, and it is what the ratio below reads over a short path. Drag the path out and that ratio falls, because ordinary water is running out of neutrons to eat while heavy water has barely started: the fractions converge on 100% and 0.3%, and the gap between them stops being a multiple and starts being the whole beam.

Number densities: ordinary water 3.34 x 10^22 molecules per cm3 at 1.00 g/cm3, heavy water 3.32 x 10^22 at 1.107 g/cm3 and 20 g/mol. Two nuclei of hydrogen per molecule in each. Real moderators are hotter and less dense than this and a real neutron takes a random walk rather than a straight line, so treat the path as a distance travelled, not a tank width.

The vessel that eats while it runs

Natural uranium has almost no reactivity to spare, so you cannot load a year of fuel and coast down as it burns. The reactor has to be fed continuously, which is why a CANDU looks nothing like the pressure vessel you have seen in a diagram.

The fuel sits in hundreds of horizontal pressure tubes running clean through a big drum of cool heavy water. The water inside the tubes is the hot coolant; the water around them is the moderator, and it stays cold. A fuelling machine clamps onto each end of a channel, and between them they push six to ten fresh bundles per channel through, at full power, and take the spent ones out the far side. The reactor never stops to eat.

One channel is drawn opened: calandria tube, gas annulus, pressure tube, and the string of twelve bundles inside it. The fresh bundle is tinted cool and the spent one warm. Channel counts differ by design, so none is claimed here.

Half a metre, twenty kilos, thirty-seven elements

A bundle is small enough to lift. That is the point: the whole refuelling scheme works because the unit of fuel is a thing a machine can push down a tube, not a core you have to open a vessel to reach.

Thirty-seven elements on the real ring counts: one in the centre, then six, twelve and eighteen. 0.495 m long, about 19.6 kg.

The two calculations that run the plant

Continuous refuelling turns the core into something you have to keep track of. Every day, someone runs two models, and between them they are the plant.

One tracks the burn of every bundle in the core and decides which channels get fed tomorrow. Feed the wrong ones and the power tilts across the core.

One checks every channel's flow against the point where the water stops cooling. Past that point the coolant stops wetting the sheath, the heat has nowhere to go, and the margin you are protecting disappears quickly. More power in a channel needs more flow, so the limit rises to the right and every channel has to sit above it.

Neutronics and thermal-hydraulics. Both are done with validated codes because the regulator requires it. The numbers on the core face here are illustrative; the shape of the check is not.

This is the layer we are building at Reflux: the models that sit between a process and the people who have to decide something about it tomorrow morning.