Adaptable phase retrieval for coherent transition radiation spectroscopy based on differentiable physics information

Coherent transition radiation (CTR) spectroscopy is a critical diagnostic for characterizing the longitudinal structure of relativistic electron bunches in laser-plasma and conventional accelerators. Recovering the bunch profile from a measured CTR spectrum is, however, an ill-posed phase-retrieval problem traditionally addressed using Gerchberg–Saxton (GS)-type iterative algorithms, whose explicit constraint-enforcement steps can be difficult to extend to more complex experimental forward models. Here, we introduce a differentiable framework for CTR phase retrieval. We consider a density-parametrized gradient-descent method (GD-Amp) and propose a phase-only formulation (GD-Phase) that fixes the calibrated spectral amplitude while optimizing the Fourier phase under physical real-space priors. Synthetic benchmarks spanning multi-peaked and strongly modulated bunches show that GS remains a highly accurate and computationally efficient baseline, while GD-Amp and GD-Phase provide distinct differentiable formulations with high reconstruction fidelity and different ensemble-conditioning behavior. We further extend the framework to detector-level inversion using direct instrument-aware density optimization (GD-IA) and an untrained implicit neural representation (INR-IA). The detector-level tests show that missing spectral information can admit multiple detector-consistent solutions, with the selected solution depending on the chosen representation, while explicit low-frequency priors and neural representation bias regularize this ambiguity differently. The resulting framework provides a composable route for incorporating detector models, physical priors, and multi-diagnostic constraints. It also offers a natural path toward higher-dimensional, multimodal, and uncertainty-aware diagnostics in realistic experimental settings.

Authors

Institutions

Publication Details

Journal
Machine Learning Science and Technology
Published
2026-11-18
DOI
https://doi.org/10.1088/2632-2153/aec40e
Primary Topic
Particle Accelerators and Free-Electron Lasers
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Adaptable phase retrieval for coherent transition radiation spectroscopy based on differentiable physics information

Ulrich Schramm, Michael Bussmann, Ritz Ann Aguilar, Maxwell LaBerge et al.
2 links
Machine Learning Science and Technology
Particle Accelerators and Free-Electron Lasers
article

Adaptable phase retrieval for coherent transition radiation spectroscopy based on differentiable physics information

Ulrich Schramm, Michael Bussmann, Ritz Ann Aguilar, Maxwell LaBerge, Jeffrey Kelling, Alexander Debus, Arie Irman, Andreas Doepp, Fabia Dietrich, Zewu Bi
article en

Abstract

Coherent transition radiation (CTR) spectroscopy is a critical diagnostic for characterizing the longitudinal structure of relativistic electron bunches in laser-plasma and conventional accelerators. Recovering the bunch profile from a measured CTR spectrum is, however, an ill-posed phase-retrieval problem traditionally addressed using Gerchberg–Saxton (GS)-type iterative algorithms, whose explicit constraint-enforcement steps can be difficult to extend to more complex experimental forward models. Here, we introduce a differentiable framework for CTR phase retrieval. We consider a density-parametrized gradient-descent method (GD-Amp) and propose a phase-only formulation (GD-Phase) that fixes the calibrated spectral amplitude while optimizing the Fourier phase under physical real-space priors. Synthetic benchmarks spanning multi-peaked and strongly modulated bunches show that GS remains a highly accurate and computationally efficient baseline, while GD-Amp and GD-Phase provide distinct differentiable formulations with high reconstruction fidelity and different ensemble-conditioning behavior. We further extend the framework to detector-level inversion using direct instrument-aware density optimization (GD-IA) and an untrained implicit neural representation (INR-IA). The detector-level tests show that missing spectral information can admit multiple detector-consistent solutions, with the selected solution depending on the chosen representation, while explicit low-frequency priors and neural representation bias regularize this ambiguity differently. The resulting framework provides a composable route for incorporating detector models, physical priors, and multi-diagnostic constraints. It also offers a natural path toward higher-dimensional, multimodal, and uncertainty-aware diagnostics in realistic experimental settings.

Machine Learning Science and Technology
Helmholtz-Zentrum Dresden-Rossendorf (DE), Center for Advanced Systems Understanding (DE), Ludwig-Maximilians-Universität München (DE)
Helmholtz-Gemeinschaft
Particle Accelerators and Free-Electron Lasers
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon