Inside a Medical Cyclotron Target: A Lay Summary

A thimble of water, a beam of protons, and a reactor that breathes.

Somewhere in a hospital basement, a machine the size of a car fires a beam of protons into a thimble of water. A few minutes later that water is radioactive, and within the hour it is inside a patient, lighting up a cancer on a scanner. This is how most PET tracers are made. It is also, quietly, one of the least efficient steps in all of nuclear medicine. My PhD, done at TRIUMF, was about understanding why, and what to do about it.

beam energy 1.00

gas density 0.85

Each dot is a packet of protons. They fly in through the foil (left), scatter into a widening cone, and slow as they lose energy, so they bunch up and pile into a bright cluster where they stop: the Bragg peak. They deposit only a little along the way and most of it right there, which is exactly why a target is so hard to cool. Raise the beam energy or drop the gas density and the beam reaches farther, until the pile-up hits the cooled wall and that energy is lost to metal instead of fluid.

why this matters

Every year more than forty million people are scanned with short-lived radioisotopes like fluorine-18, carbon-11, and technetium-99m. You inject a tiny amount, it collects where the body is busiest (a tumour, a starving patch of heart muscle), and a PET or SPECT scanner watches it decay. For decades, over eighty percent of these isotopes came from a handful of aging nuclear research reactors. When one of them, the Chalk River reactor in Canada, went down, a large part of the world’s medical isotope supply went with it. That scare pushed the field toward a decentralized alternative: make the isotopes on demand, in a hospital, with a compact cyclotron.

There are nearly a thousand of these machines installed worldwide. The catch is that the place where the isotope is actually made, the target, was largely designed by trial and error. Gas targets deliver only about half of their theoretical yield. Liquid targets can run thirty percent below. We were leaving a lot of medicine on the table, and nobody could fully say why.

what a target actually is

Strip away the jargon and a target is a tiny, sealed, pressurized chemical reactor. A thin metal foil holds back a chamber of fluid; a proton beam punches through the foil and into the fluid, where it drives a nuclear reaction. The classic one turns heavy oxygen into fluorine,

{}^{18}\mathrm{O}(p,n)\,{}^{18}\mathrm{F}

and for a fixed beam energy, the amount of isotope you make is set by how much beam current you can pour in. The yield climbs toward a saturation value as you irradiate,

Y = Y_s\, I \,\bigl(1 - e^{-\zeta t}\bigr)

where I is the beam current and ζ the decay rate of the isotope. The obvious move is to crank up I. The problem is that all of that beam power becomes heat, deposited inside a few millilitres of fluid, and the fluid fights back.

A regular liquid water target mounted on the target selector of TRIUMF's TR13 cyclotron, surrounded by cooling lines, valves, and signal wiring. — Regular liquid target on TR13 cyclotron’s selector

the feedback that ruins everything

Here is the heart of it. A proton does not deposit its energy evenly. It barrels through the fluid losing energy slowly, then dumps almost everything in a sharp spike right before it stops, the Bragg peak. How far it travels before that happens, its range, depends on how dense the fluid is, through the Bethe-Bloch law,

-\frac{dE}{dx} = c_1\,\frac{\rho}{E}\,\ln(c_2 E)

Now close the loop. The beam dumps heat where it stops. That heat warms the fluid. Warm fluid is less dense. Less dense fluid lets the beam travel farther, which moves where it stops, which moves where the heat goes. The target’s behaviour is not set by any single quantity but by this circle of cause and effect chasing itself.

The coupling that makes targets hard. The proton range depends on the fluid density; the density depends on temperature; the temperature depends on where the beam dumped its energy. Close the loop and a simple reactor can do surprising things.

When the fluid thins enough, the beam can punch all the way to the back wall of the chamber and bury its Bragg peak in metal. The wall is a good conductor, so that energy is simply carried away and lost. Worse, the isotope made near a hot wall tends to stick to it and never come out. This single effect, the beam over-running its own target, is the largest source of the missing yield.

where the beam stops

You can feel the whole problem in one figure. Push the beam power up and the fluid thins, so the peak slides toward the wall. Push the density up and the beam stops short. The fraction of energy that actually lands in the fluid, λ, holds at one until the peak crosses the wall, then falls off a cliff.

beam power 0.50

fluid density 0.80

Raise the power and the fluid thins, so the beam reaches farther; raise the density and it stops sooner. Most of the energy lands in the Bragg peak, so once that peak pushes past the wall (terracotta), it is deposited in metal, not fluid, and the fraction captured, λ, falls fast. This is the single biggest loss in a real target.

Everything downstream, every bit of pressure, temperature, and yield, depends on getting λ right. So the first job of any honest model is to predict where the beam stops.

a model from two conservation laws

The full physics here is a coupled, multiphase, reacting flow: a problem for a supercomputer, and one that gives you a wall of numbers rather than understanding. I wanted something you could actually reason about. The trick is volume averaging: instead of tracking the temperature and density at every point, you track their averages over the whole chamber and follow only how those averages change in time. A forest of partial differential equations collapses into a handful of ordinary ones.

What survives is a balance of mass and a balance of energy. The energy balance is the one that matters, and it reads almost like plain bookkeeping:

m\,C\,\frac{dT}{dt} \;+\; \lambda_g\,\frac{dm_v}{dt} \;=\; \lambda S_o \;-\; hA\,(T - T_\infty)

On the left, the fluid stores energy by getting hotter (the first term) and by boiling (the second). On the right, the beam pours energy in (the fraction λ of the beam power it actually captures) and the cooling system carries energy out. Add the ideal gas law for pressure, a force balance for the bulge of the foil, and the chemistry of the beam tearing water apart, and you have one model that, in different limits, describes every kind of target. I then spent four studies testing it against real experiments on TRIUMF’s TR13 cyclotron.

inert gas first

Inert gas is the easy case: one phase, no boiling, no chemistry to speak of. The model predicts that the pressure rise should depend on the deposited power and the kind of gas, but not on the starting pressure, and that the curve should bend over as the power climbs. That bend is the fingerprint of natural convection: as the gas heats, it churns, and churning carries heat to the wall faster. Folding that into the heat transfer coefficient gives a clean closed form,

\bar P = 1 + \psi_m\,\frac{\bar S}{h_o + h_1\,\bar S}

Steady-state pressure rise vs scaled beam power for argon, nitrogen, and neon. The solid lines are the closed-form model above, with the heat-transfer coefficients fitted per gas (from the TR13 runs); the open circles are representative measurements scattered about each curve. The fit explains over 99.7% of the variance (R² in the legend). Drag the dashed line to read the model at any beam power, or toggle a gas to isolate it. Notice the pressure rise depends on the gas, not the starting pressure: neon insulates best, argon worst.

and it fits the argon, nitrogen, and neon experiments almost perfectly, better than ninety-nine percent of the variance explained. The practical lesson was blunt: if convection is what saves you, engineer more of it. Put a fan in the chamber. Colleagues at TRIUMF did exactly that, and raised the yield by half.

then water, and a surprise

Water is where it gets strange. Now the fluid can boil, so vapour and radiolysis gas share the chamber with the liquid, and the pressure is mostly vapour pressure. At steady state the model still works beautifully, and it confirmed that liquid and vapour stay in thermodynamic equilibrium the whole time. But steady state was hiding the real story. When I watched the pressure as a function of time, the same target, under the same beam, behaved in three completely different ways depending on how full it was.

beam current µA 10

void fraction % 20

Same target, same beam, three behaviours. A nearly-full target (low void) settles smoothly or overshoots once and relaxes. Open up the void fraction and the beam-fluid coupling tips the system into a limit cycle: a self-sustained pressure oscillation that never settles. The model here is a van der Pol oscillator standing in for the full beam-density coupling, but the regimes, and where they switch, match the cyclotron.

A nearly-full target rises and settles, or overshoots once and relaxes: tame. But open up the empty space above the liquid and the target starts to breathe. The pressure climbs, falls, climbs, falls, a self-sustained oscillation that never settles down, for as long as the beam is on. Nothing in the input is oscillating. The target is doing it to itself.

why a reactor breathes

It turns out there are two separate ways to make a target oscillate, and the thesis pinned down both. The first is geometric, and it is the feedback loop from the start of this piece, with a twist. As the fluid boils and churns, the gas it makes can sit in the chamber in more than one arrangement (a quiet layer, or a roiling plume), and each arrangement lets a different fraction of the beam through. So the captured power λ depends not just on the average density but on the fluid’s recent history. That memory, a hysteresis in λ, is exactly what a self-sustained oscillator needs. Feed it into the model and the breathing appears, matching the experiments.

The second mechanism is chemical. The beam rips water into hydrogen, oxygen, and peroxide, and those reactions release heat of their own. Treat the target as a small chemical reactor with that heat source and run a proper stability analysis (eigenvalues, the same tools you would use for an industrial reactor) and something striking falls out: the steady state need not be unique. There can be a low-temperature solution and a high-temperature one, with an unstable rung between them, and ignition and extinction points where the system jumps. Past a threshold, the only attractor left is a limit cycle: a stable, self-sustained oscillation, born purely from the coupling of heat and chemistry. The breathing target was sitting on one.

what to build instead

A model earns its keep when it tells you what to change. For gas targets, the message was to chase convection: fans, turbulence, anything that moves heat to the wall. For liquid targets, the stability analysis pointed at one number, the mean residence time, the average time a parcel of fluid spends sitting in the beam. The destructive oscillations live at long residence times. Shorten it, by giving the fluid somewhere to flow (an expansion chamber, or an outright flow-through loop), and you do three good things at once: you kill the limit cycle, you drop the pressure so you can run more current, and you keep fresh, dense fluid in the beam so less of it escapes to the wall. The best existing designs already hint at this; the model says there is another twenty percent to be had.

the part I still find beautiful

What I love about this problem is how little it takes to make it rich. One sentence, where the beam stops depends on the density, which depends on the temperature, which depends on where the beam stopped, generates the entire zoo: multiple steady states, ignition and extinction, hysteresis, and a reactor that breathes on its own. And the same observed oscillation has two distinct causes, one about geometry and one about chemistry, that you can only tell apart with the model in hand. A thimble of water and a beam of protons turn out to be a small, honest laboratory for nonlinear dynamics, with cancer scans riding on the answer.

This work was my doctoral thesis in Chemical and Biological Engineering at the University of British Columbia, carried out at TRIUMF, Canada’s particle accelerator centre, across four papers on the steady-state and transient behaviour of gas and liquid cyclotron targets.