Three-Crystal Array Cuts SPS Extraction Losses Fourfold

Slow extraction from CERN’s Super Proton Synchrotron is constrained less by the number of protons available than by where the unavoidable losses land. In third-integer resonant extraction, part of the circulating beam intercepts the electrostatic-septum wires; that loss activates the equipment and limits delivery to the North Area experiments. A single bent silicon crystal installed at a non-local location had already cut those losses by about a factor of two, but the future operating budget calls for a factor of four.

The paper reports a 3-crystal Multi-Volume Reflection Array (MVRA) installed in SPS Long Straight Section 4. Rather than attempting to channel every relevant proton in one precisely aligned crystal, the array gives particles several opportunities for volume-reflection deflection. At its measured optimum, the installation reaches the intended fourfold reduction in septum losses. It has also been held in routine physics production with a median factor-3.6 reduction across 0.1 million extraction cycles relative to the pre-shadowing baseline.

Core Contribution

The contribution is an operational beam-loss mitigation system, not just an optics proposal. The authors combine a compact multi-crystal geometry with an optimization procedure that searches over practical crystal configurations, then test the resulting device with beam-position and angle scans. The important distinction from the earlier single-crystal shadowing scheme is redundancy: a particle missed or weakly deflected by one element can still receive a useful kick at a later crystal.

The array is designed to shadow the electrostatic septum from a non-local position. That makes the phase-space transport between the array and the extraction region part of the design problem; the useful quantity is not merely the deflection from an individual crystal, but the resulting displacement and distribution at the septum. The paper therefore treats the three elements as an array whose relative geometry must be optimized together.

Technical Approach

The design uses Multi-Fidelity Bayesian Optimization (MFBO). A multilayer-perceptron surrogate supplies a cheaper approximation during the search, while full particle tracking supplies the higher-fidelity evaluations needed to retain beam-dynamics realism. This is a sensible fit for a layout problem in which evaluating all combinations of positions and angles with detailed tracking would be expensive, yet small alignment changes can materially alter which particles are intercepted downstream.

The authors then compare measured loss scans against simulations adjusted to the measured positions and angles. The agreement is reported to be within a few percent, which matters because it ties the optimization and tracking model to the installed, imperfect device rather than only to an ideal geometry. The work also adds an extremum-seeking controller to keep the system on its operating plateau during production. That operational layer is consequential: a narrow scan optimum would have limited value if ordinary drift pushed the array away from it between tuning periods.

Results and Analysis

The central measured result is a factor-four loss reduction at the optimum of the as-built three-crystal array, matching the design prediction. This improves materially on the factor-two reduction from the earlier single-crystal installation and meets the stated activation-driven target for the planned higher-intensity regime. The simulation-to-scan agreement within a few percent gives the result more weight than a single loss-monitor comparison, since it tests whether the predicted dependence on beam and crystal settings is right.

Production data are less idealized and therefore especially informative. Across 0.1 million extraction cycles, the median reduction is 3.6 rather than 4. That gap is modest, but it is the relevant gap for operations: the device can approach its tuned performance while being regulated in a changing machine, rather than only during a dedicated optimization scan. The paper’s evidence supports the claim that the three-crystal system is deployable today.

The larger projected gains remain predictive. The MFBO and tracking study indicates up to a tenfold reduction for an ideally aligned four- or five-crystal array, but the present paper measures three crystals, not that upgrade. The fourfold result is therefore established by beam data; the tenfold figure should be read as a design direction contingent on additional hardware and alignment performance.

Limits for the Upgrade

The present array leaves little margin beyond the factor-four target under its best settings, while production performance is lower at a 3.6 median reduction. The authors identify the installed geometry and alignment as constraints and discuss a four- or five-crystal successor. That next step is technically plausible given the validated tracking model, but its predicted tenfold suppression has not yet received the same beam-based validation.