Optimizing the Beam's Path Through an Isotope Production Target

Facilities that make rare, short-lived isotopes and accelerate them work in a fixed order. A proton beam strikes a hot uranium-carbide target and splits uranium nuclei. The fragments diffuse out of the hot target, are ionized to a single positive charge, and are passed to a charge breeder, a device that strips more electrons so the ions can be accelerated efficiently. The whole chain is very lossy. At SPES the end-to-end efficiency is estimated at about 2% (Future Perspectives of the Legnaro National Laboratories: The SPES project). Much of the downstream loss sits inside the breeder, but the first step decides how many isotopes ever reach it. That first step is set by temperature. The target must be hot, around 2000 °C, so isotopes diffuse out before they decay, but never so hot that it sinters or melts. Today the beam spot that heats the target is essentially static. Heat management is handled by geometry: SPES splits the target into seven thin disks so they radiate 8–10 kW to the chamber walls (The SPES project at the INFN- Laboratori Nazionali di Legnaro).

This note proposes to treat the beam's path through the target as a trajectory to be optimized in three dimensions. Two of those dimensions are the transverse position of the spot, moved by fast steering magnets. The third is the depth at which power is deposited, controlled by beam energy. A rotating energy degrader would vary it quickly, and the SPES driver cyclotron's 35–70 MeV range would vary it slowly. The goal is to keep as much of the target volume as possible in the narrow temperature window where short-lived isotopes escape quickly. A second goal is to keep the singly charged ion flux steady, because that flux feeds the charge breeder. The claim is specific and could fail. A time-varying path that also uses depth should beat the best static spot and the best fixed circular sweep on release of short-lived isotopes at equal peak temperature. If it does not, the static design is already near-optimal and we will say so.

Core Claim

Each thin disk heats and cools on its own timescale. Fission heating and proton stopping deposit energy unevenly along the beam axis, so the front disks and back disks sit at different temperatures under a fixed spot. Spreading the beam transversely flattens radial gradients but cannot fix the disk-to-disk gradient. Changing deposition depth can. Our hypothesis is that the release-weighted yield is limited by the coolest useful volume, not the average, and that only a path that moves in all three coordinates can raise that floor without raising the hottest point.

We write the problem as

$$\max_{\mathbf{u}(t)} \; \frac{1}{T}\int_0^T \!\!\int_V \rho_f(\mathbf{r},t)\, \eta\big(\Theta(\mathbf{r},t);\, \tau_{1/2}\big)\, dV\, dt \quad \text{s.t.} \quad \Theta \le \Theta_{\max},\;\; \mathrm{Var}_t[\Phi_{1+}(t)] \le \delta,$$

Here $\mathbf{u}(t)$ is the beam path (transverse position $x,y$ and deposition depth set by energy). $\rho_f$ is the local fission rate density, and $\Theta$ is the local temperature from a transient heat model with disk-to-disk radiative exchange. $\eta$ is the fraction of an isotope with half-life $\tau_{1/2}$ that diffuses and effuses out before decay, using an Arrhenius diffusion coefficient. $\Theta_{\max}$ is the material limit. $\Phi_{1+}$ is the singly charged flux sent to the breeder, with $\delta$ a bound on its fluctuation. $T$ is the averaging time.

The optimization reaches the heat model through two routes. $\Theta$ depends on $\mathbf{u}$ directly. Depth control also changes $\rho_f$, because the fission cross-section depends on proton energy. So the optimizer trades yield against temperature shaping rather than treating heat as a nuisance.

Relation to Prior Work

The thermal side motivates and constrains the design. SPES fixed the geometry: seven radiating UCx disks at about 2000 °C under 8–10 kW from a C70 cyclotron demonstrated at 40 MeV/201 µA and 70 MeV/500 µA (The SPES Radioactive Ion Beam facility of INFN; The SPES radioactive ion beam project of LNL: status and perspectives). We keep this geometry and add only a control input. JUNA's rotating cold targets sweep a fixed spot over a large track, handling up to 20 kW/cm² (Progress of Jinping Underground laboratory for Nuclear Astrophysics (JUNA)). The ILC rotating positron target shows sweeping also works with purely radiative cooling (The ILC positron target cooled by thermal radiation). Both sweep to lower the peak temperature. We sweep to shape a temperature window in volume and depth, which is a different objective. Diamond-sandwich conduction (Abrahamyan et al.) does not fit a target that must stay near 2000 °C, so it constrains the design rather than informing it.

On the breeding side, the SPES chain reports more than 85% simulated transmission from breeder through ALPI for ¹³²Sn, with final intensities of 10⁷–10⁹ pps (Status of the SPES Exotic Beam Facility; ALPI Setup as the SPES Accelerator of Exotic Beams). That transmission is high, so the leverage lies upstream of the breeder. The ALPI paper also notes that intensities this low cannot be read by standard monitors. This favors an open-loop, model-based path over live feedback on the exotic beam itself.

The literature search returned no papers on 3D trajectory optimization itself, and no work at this exact intersection. We use standard direct-collocation trajectory optimization as the method and claim no novelty for it.

Research Program

We will build a reduced transient thermal model of the seven-disk target. It will combine energy deposition from proton stopping and fission heating, conduction within each disk, and radiative exchange between disks and with the water-cooled chamber. Before any optimization, it must reproduce the published SPES operating point of about 2000 °C at 8–10 kW for a centered static beam.

On top of the thermal model we couple a diffusion–effusion release model for a short-lived and a longer-lived reference isotope. We then add a simple input model that converts release into a $\Phi_{1+}(t)$ time series. Breeder physics is not modeled beyond this injected flux, because the source material gives no injection-stability data.

The path $\mathbf{u}(t)$ is parameterized in three ways, all with hardware-realistic limits on steering rate and degrader thickness:

  • a periodic transverse pattern;
  • a periodic depth schedule;
  • their joint form.

We will optimize by direct collocation over one thermal period and by model-predictive control for transients such as beam ramp-up and recovery after a beam trip.

Baselines

The naive reference point is the static centered Gaussian spot at the SPES design energy. This is what the target was designed around. Every method is compared at the same beam power and the same peak temperature, so gains cannot come from simply running hotter.

The strongest static competitor is the defocused static spot. Its width and the static energy are tuned by grid search with the same number of thermal-model evaluations as the trajectory optimizer gets.

The strongest existing dynamic method is a fixed circular sweep. It is the target-frame equivalent of JUNA- or ILC-style rotation (Progress of Jinping Underground Nuclear Astrophysics Experiment (JUNA) project; The ILC positron target cooled by thermal radiation). Its radius and frequency get the same tuning budget.

No reported result was found on Lacuna for temperature uniformity or release fraction under any sweep pattern on an ISOL target. The only published reference figure is the SPES design point of about 2000 °C at 8–10 kW. We use it as a calibration check, not a score to beat.

Two ablations isolate the key ingredient: transverse-only optimization and depth-only optimization. Beating the baselines means at least a 20% gain in release-weighted yield for an isotope with sub-second half-life over the best tuned circular sweep, at equal $\Theta_{\max}$, with flux fluctuation no worse. The 20% threshold is our choice, sized to exceed plausible model error. The joint path must also beat both ablations, or the "three-dimensional" claim is dropped.

Why not just…?

Why not just defocus the static spot? This is the most serious threat for steady-state operation. A wider spot alone may flatten the radial profile enough. It cannot change the front-to-back gradient set by stopping and fission along the axis. The depth-only ablation and the tuned-defocus baseline measure exactly how much of the gain that gradient accounts for. If the defocused spot comes within model error of the joint path, we will report that the static design is sufficient.

Why not just reshape the disks, for example with graded thicknesses or spacings? Geometry is how SPES solved the problem, and a fixed geometry can be tuned for one operating point. It cannot adapt across driver energies, during ramp-up, or after a beam trip. Our model-predictive experiments target those transient regimes specifically. We will also run one passive graded-thickness geometry through the same model as an extra comparison, so that a geometric win would be visible.

Why not just rotate the target, as JUNA and the ILC do? In the target frame, rotation is thermally a circular sweep, and the circular-sweep baseline covers it. Rotation also complicates the sealed path from target to ion source, which the ISOL release depends on. Steering the beam leaves the target stationary.

Decisive Checks

The first gate is calibration. If the model cannot reproduce about 2000 °C at 8–10 kW for a centered spot, nothing downstream is trusted. The main check is the equal-peak-temperature comparison of release-weighted yield across the static, defocused, circular, transverse-only, depth-only and joint paths. It is repeated for two isotopes of different half-life and at 40 and 70 MeV driver settings.

A third check is the fluctuation of $\Phi_{1+}(t)$. A path with high yield but large thermal cycling would hurt the breeder, so it does not count as a win.

Finally, a bench test checks that sweep-induced temperature dynamics are captured before any claim is made about real targets. A graphite disk mock-up is heated by a steered electron beam and read with infrared pyrometry. The measured temperatures, time constants and gradients are compared with the model's predictions.