Machine learning-based virtual diagnostics of dielectric laser acceleration

T. Egenolf, O. Boine-Frankenheim · 2026

Read Machine learning-based virtual diagnostics of dielectric laser acceleration by T. Egenolf, O. Boine-Frankenheim (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-28549.txt.

Introduction

Egenolf and Boine-Frankenheim build a virtual tool to diagnose dielectric laser acceleration. Such devices could make electron accelerators much smaller. Yet laser and electron contact is hard to measure directly. Their system learns the reverse link from electron energy spectra to laser settings. It trains a neural network on synthetic data from a tracking code called DLAtrack6D. In tests, it recovers pulse tilt near one degree. It also finds phase offsets with small error. Fast output could support live control at the ARES linac.

Catalog record

CREATORT. Egenolf, O. Boine-Frankenheim
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31173
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.28549

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine learning-based virtual diagnostics of dielectric laser acceleration T. Egenolf, O. Boine-Frankenheim Abstract We present the development of a digital twin–based reconstruction framework for dielectric laser acceleration (DLA) based on machine-learning-assisted inversion of single-shot electron energy spectra. DLA as a promising candidate for compact electron accelerator designs using optical near-fields in dielectric nanostructures lacks on direct diagnostics on the laser-electron interaction. Thus, the outgoing electron energy distribution is one of the few experimentally accessible observables. To exploit this information, the DLA interaction and the mapping on the downstream spectrometer are treated as nonlinear measurement device whose response is described by the symplectic six-dimensional tracking code DLAtrack6D. This forward simulation model serves as a digital twin mapping laser-electron interaction parameters onto the resulting energy spectrum. For diagnostics, we are interested in the inverse mapping which is represented by a neural network trained by synthetic datasets generated with the forward simulation model. The reconstruction performance and parameter identifiability of the inverse model are evaluated for parameter ranges relevant to planned experiments at the ARES linac in the SINBAD facility at DESY. Simulation studies demonstrate that the method can recover pulse front tilt angles with an accuracy of about 1 ° 1\text{\,}\mathrm{\SIUnitSymbolDegree} and phase offsets with an RMSE of about 0.36 rad 0.36\text{\,}\mathrm{rad} corresponding to a difference of 0.4 fs 0.4\text{\,}\mathrm{fs} in arrival time. Training with noisy spectra further improves robustness against spectrometer noise. The trained surrogate model evaluates in the millisecond range, enabling shot-to-shot parameter estimation compatible with the 50 Hz 50\text{\,}\mathrm{Hz} repetition rate of ARES. The approach effectively transforms the DLA interaction region into a virtual in situ diagnostics for otherwise inaccessible laser parameters during experiment and provides a foundation for real-time monitoring and control of future DLA experiments. I Introduction Dielectric laser acceleration (DLA) has emerged as a promising approach toward compact accelerator technology by exploiting optical near-fields in nanostructured dielectric materials 10 ; 17 ; 1 . DLA structures are driven by infrared or optical laser pulses, enabling accelerating gradients in the multi-GV/m range due to the high damage thresholds of dielectric materials at optical frequencies 17 ; 12 . Experimental demonstrations of laser-driven acceleration in nanophotonic structures have confirmed the feasibility of the concept at subrelativistic and relativistic electron beam energies 19 ; 3 ; 5 ; 2 . A characteristic feature of DLA structures is the sub-wavelength spatial periodicity following the synchronicity condition \lambda_{z}=m\beta\lambda_{0} , where m is the order of the accelerating spatial harmonic, \beta is the particles’ velocity and \lambda_{0} is the central laser wavelength 15 ; 14 ; 16 . Typically, the period lengths are on the order of a few hundreds of nanometers up to a few micrometers and interaction lengths ranging from hundreds to thousands of grating periods. The accelerating fields are confined to near-field regions inside nanofabricated dielectric geometries, rendering direct field diagnostics during operation impractical. Unlike RF cavities, where field probes and pickup antennas provide access to amplitude and phase information, DLA structures do not allow internal electromagnetic measurements. Consequently, the primary experimentally accessible observable in current DLA experiments is the outgoing electron energy spectrum measured downstream of the interaction region using a magnetic spectrometer 13 ; 6 . The spectrometer thus represents the only direct beam diagnostic that encodes information about the laser–electron interaction inside the nanophotonic structure on a shot-to-shot base during the experimental run. However, the mapping between laser parameters and the measured spectrum is highly nonlinear and depends on multiple spatio-temporal degrees of freedom, including pulse front tilt, spectral phase, arrival time offset, and transverse alignment. Precise control of these parameters is essential for stable DLA operation. In particular, pulse front tilt (PFT) is required to extend interaction lengths 4 ; 7 . Similarly, femtosecond-level timing synchronization between the electron bunch and the laser pulse critically determines the achievable net acceleration. Small deviations in shaping parameters can lead to significant reductions in effective accelerating gradient or to asymmetric spectral distortions 6 . The absence of direct internal diagnostics therefore poses a significant instrumentation challenge for DLA experiments. In conventional accelerator facilities, beam-based alignment and RF phase scans are routinely used to infer cavity parameters indirectly. A similar philosophy can be adopted for DLA systems: the interaction region and downstream spectrometer can be interpreted as a nonlinear measurement device whose response function is determined by beam dynamics in the laser-driven near field. Under this perspective, the measured spectrum represents the output of a deterministic forward operator that maps a reduced set of effective laser parameters to an observable energy distribution. The reconstruction of laser and interaction parameters from the measured spectrum constitutes an inverse problem. Due to the nonlinear dependence of the spectrum on shaping variables and the finite resolution of the spectrometer, the inversion is generally ill-posed in the full parameter space. However, when restricted to a physically motivated and experimentally bounded parameter domain, the problem becomes tractable. In this work, we present the design and performance evaluation of a model-based virtual beam diagnostic for DLA experiments, e.g. at the ARES linac at DESY 13 ; 8 . A fast six-dimensional tracking model is employed to compute the forward instrument response of a realistic DLA–spectrometer system 15 . Based on this forward model, synthetic datasets covering the experimentally relevant parameter space are generated and used to train a neural network that approximates the inverse mapping from measured spectra to effective interaction parameters. The proposed approach is not intended as a replacement for physical modeling but as a computationally efficient surrogate for iterative fitting procedures and shot-to-shot parameter estimations, enabling real-time monitoring of effective on-chip laser parameters. The method transforms the DLA interaction into a virtual diagnostic instrument capable of providing indirect access to otherwise inaccessible quantities. The present paper focuses on the instrumentation aspects of this concept. We quantify the sensitivity of the spectrometer response to relevant shaping parameters, analyze identifiability and noise robustness, and evaluate the achievable reconstruction accuracy under realistic experimental conditions. While experimental validation is pending commissioning of the DLA laser setup at ARES, the results provide quantitative performance benchmarks and establish the methodological framework for real-time spectral inversion in nanophotonic accelerator experiments. Table 1 : Simulation parameters according to experimental planes at ARES for a working point including velocity bunching 13 . Parameter Value Mean beam energy 99.1\text{\,}\mathrm{MeV} Initial rms energy spread 43.8\text{\,}\mathrm{keV} Initial bunch length 268\text{\,}\mathrm{nm} DLA structure length 1\text{\,}\mathrm{mm} DLA period length 2.05\text{\,}\mathrm{\SIUnitSymbolMicro m} Central laser wavelength 2.052\text{\,}\mathrm{\SIUnitSymbolMicro m} Acceleration gradient 1\text{\,}\mathrm{GV}\text{\,}{\mathrm{m}}^{-1} Repetition rate 50\text{\,}\mathrm{Hz} II Forward Model and Instrument Response Definition To enable quantitative reconstruction of interaction parameters, the spectrometer response must be described as a deterministic forward operator that maps a reduced set of effective laser and alignment parameters onto the measured energy distribution. This requires a beam dynamics model that captures the essential features of the laser–electron interaction within the nanophotonic structure while remaining computationally efficient enough for large-scale dataset generation. Figure 1 : Computing time of the Forward model using DLAtrack6D as function of the number of used CPU kernels (red and blue line for 10^{5} and 2.5\times 10^{5} macroparticles, respectively) showing the parallelization properties of the applied tracking code, and as function of the number of simulated DLA periods showing the almost linear dependence. In

Read the complete text