A Diffusive Model for Radio-Frequency Knock-Out Slow Extraction Preprint: APS/123-QED P. A. Arrutia Sota, E. C. Cortés García, V. A. Sansipersico Abstract Slow resonant extraction from synchrotrons via radio-frequency knock-out is a well-established technique to deliver charged particle beams for various applications. In this contribution, we present explicit analytical expressions for calculating the number of particles slowly extracted over time—commonly referred to as spills. The proposed formulation enables the semi-analytical determination of an amplitude modulation curve to be applied to the radio-frequency exciter, which flattens the spill macrostructure, a feature of high relevance to all users requiring uniform beam delivery. I. Introduction. Charged particle beams prepared in synchrotrons are often required to be delivered over extended timescales, typically from a few hundred to several thousand milliseconds. To achieve this slow extraction—much longer than the revolution period and independent of the bunch structure—the resonant extraction technique is employed. Among the possible resonance orders, the third-order resonance has become the standard working point 4 , widely adopted for applications such as electronics irradiation 8 , cancer therapy 10 , and fixed-target experiments. Several variations of this method exist; one particularly effective approach for enhanced control and quality of the extracted particle counts, so called spills, is the radio-frequency knock-out (RF-KO) technique 9 . In RF-KO extraction, the beam optics are kept constant while the betatron motion of particles is deliberately driven by an external excitation. This induces a controlled emittance growth, a process extensively studied over the past decades to optimize both the macro- and micro-spill structures (see e.g. 6 ; 5 ; 1 ). Owing to the complexity of single-particle dynamics near resonance, comprehensive analytical models remain out of reach. Thus, optimization relies on particle tracking and/or dedicated machine development campaigns for fine tuning. In this contribution, we adopt a simplified diffusive-model approach to RF-KO extraction 11 ; 2 and derive analytical expressions to describe the spill. These are applied to reconstruct the spill profile and to compute an amplitude modulation curve for the excitation, aiming to achieve a flat spill. The results are validated through experimental campaigns conducted at the CERN Proton Synchrotron (PS). The paper is organized as follows: Section II summarizes the established effective model describing the beam dynamics during RF-KO extraction. Section III presents the diffusive model and its analytical formulation, followed by a comparison between the two approaches. Section IV discusses the amplitude modulation required to obtain a constant spill flux. Section V presents the experimental data validating the results, and Section VI concludes with a summary of the main findings. II. Map-based model. Numerical simulations of the RF-KO process are typically performed by employing map-based models, i.e. tracking codes. In this contribution, we will use a simple ‘Hénon-like’ map as our baseline model, which will later be used to benchmark the proposed diffusive model. The map updates the particle’s normalized transverse offset X and momentum P every turn n according to, \begin{bmatrix}X_{n+1}\\ P_{n+1}\end{bmatrix}=R(\mu)\begin{bmatrix}X_{n}\\ P_{n}+SX^{2}_{n}+k(t)\end{bmatrix} (1) where S is the resonance driving term, k(t) is the time-dependent exciter kick, and R(\mu) is the 2-dimensional rotation matrix by an angle \mu , \mu=2\pi\left(m+\frac{1}{3}+\delta Q\right),\ \ m\in\mathbb{Z}, (2) with \delta Q being the tune distance to the third-integer tune, which is taken to be the same for all particles. For the CERN PS, m=6 and S=77\text{m}^{-\frac{1}{2}} . When k(t)=0 , the dynamics may be smoothed and averaged over three turns to obtain the so-called ‘Kobayashi’ Hamiltonian 4 , H=3\pi\delta Q\left(P^{2}+X^{2}\right)+\frac{S}{4}\left(3XP^{2}-X^{3}\right), (3) which defines the phase-space geometry shown in Fig. 1 . As it can be seen, a triangular separatrix divides the space into a stable and an unstable region. The exciter kicks may then be seen as perturbations, transporting particles from the stable to the unstable region to deliver the desired extraction flux. Figure 1: Hamiltonian contours (greens) and time evolution of a single initial condition under RFKO kicks (black), plotted every three turns. III. Diffusive model. Motivated by the phase-space geometry and dynamics described above, alternatively we propose to model the extraction process with a diffusion equation in cylindrical coordinates 13 : \partial_{t}u=\frac{1}{r}\partial_{r}(rD\partial_{r}u), (4) where u(r,t) represents the beam distribution, with r=\sqrt{X^{2}+P^{2}} . The angular dynamics have been ignored, even though the geometry in Fig. 1 is not perfectly rotational-symmetric (especially near the separatrix boundary). In general, we will allow the diffusion coefficient D to be time-dependent and define it by the relationship 2D\cdot t=\langle r^{2}\rangle . For example, if the exciter kicks are drawn independently from a zero-mean Gaussian distribution with standard deviation k , we obtain D=k^{2}/4 . In practice, exciters will have limited bandwidth, but the formalism will still remain useful, as we will demonstrate in later sections. To represent the impact of the separatrix, we impose an absorbing boundary condition u(r=R,t)=0 at the outer radius R , computed by equating the area inside the stable region (triangle) to the area of a circle of radius R , i.e. R=\left(48\sqrt{3}\pi\right)^{1/2}\frac{\delta Q}{S} . A.Time-dependent solution. A general solution to Eq. 4 can be obtained by the method of separation of variables 7 : u(r,t)=\sum_{i=1}^{\infty}a_{i}\,\exp\left[-\lambda_{i}^{2}\,\Theta(t)\right]J_{0}\left(\lambda_{i}r\right), (5) where \Theta(t)\equiv\int_{0}^{t}D(\tau)\,d\tau , \lambda_{i}=\zeta_{i}/R , J_{0} is the Bessel function of the first kind of order zero, and \zeta_{i} are the positive zeros of J_{0} (i.e., J_{0}(\zeta_{i})=0 ). The coefficients a_{i} are obtained by projecting the initial condition u_{0}(r) onto the basis functions: a_{i}=\frac{2}{R^{2}[J_{1}(\zeta_{i})]^{2}}\int_{0}^{R}u_{0}(r)J_{0}\left(\lambda_{i}r\right)rdr, (6) where J_{1} is the Bessel function of the first kind of order one and \int_{0}^{R}u_{0}(r)rdr=N , the initial number of particles. A typical initial condition is the truncated Gaussian distribution: \displaystyle u_{0}(r)=\begin{cases}\frac{1}{A}\exp\left[-\frac{1}{2}\left(\frac{r}{R\sigma}\right)^{2}\right]\quad&\text{for }r