Searcharxiv⌕ Search

arXiv · 2611.11061

Symplectic Normal Coordinates Flow for Generative Modeling of Hamiltonian Systems

Abstract

In conditioned generative models of physical systems, the densities at different values of the conditioning parameters are often related through symplectic transformations. For stable linear Hamiltonian systems, there exists a symplectic transformation to "normal coordinates" such that the dynamics is reduced to rotations in each phase-space plane. This picture extends to the nonlinear case where, away from resonances, the Birkhoff normal form again consists of rotations in phase space, now with amplitude dependence. In this work, we introduce an invertible symplectic flow layer consisting of amplitude-dependent rotations in learned normal coordinates. We derive the condition under which the nonlinear rotation layer is symplectic and detail an architecture that guarantees it. Experiments are carried out on three conditional generative modeling problems: a synthetic chain of coupled nonlinear oscillators and two problems from accelerator physics. We show empirically that normalizing flows with the new layers achieve competitive or superior negative log-likelihoods compared to baseline flows, with gains that grow with the strength of the nonlinearity on the synthetic system.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Christopher M. Pierce. 2026-11-08. Symplectic Normal Coordinates Flow for Generative Modeling of Hamiltonian Systems. https://searcharxiv.com/abs/2611.11061

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Multi-Gaussian Model for Intrabeam Scattering

Conventional analytical models of intrabeam scattering (IBS), such as the Piwinski model and Bjorken-Mtingwa model, generally rely on the assumption of a Gaussian beam distribution, limiting their applicability to non-Gaussian cases. In this work, we develop a multi-Gaussian model to extend analytical IBS calculations to a broader class of non-Gaussian phase-space distributions. The proposed approach represents a non-Gaussian distribution as a superposition of Gaussian components sharing a common covariance matrix but having different mean values. We derived a generalized model for evaluating the momentum second-moment evolution rate and the IBS growth rate. Furthermore, for equally spaced Gaussian components, we developed an infinite bunch train model and introduce a practical truncation rule to reduce computational costs. The model is applied to a representative steady-state microbunching (SSMB) storage ring, where longitudinal overlap between neighboring microbunches is shown to enhance IBS growth rates in all three directions. The proposed framework provides a practical approach to evaluating IBS in non-Gaussian beam distributions and offers a basis for further development toward a more general non-Gaussian IBS theory.

physics.acc-ph↗

Surrogate-assisted beam distribution shaping at the CERN Proton Synchrotron and Super Proton Synchrotron

The Super Proton Synchrotron at CERN delivers protons to the North Area experimental facility at 400 GeV/c. The extracted beam is split twice vertically in the transfer line with Lambertson septa to serve three fixed-target experiments simultaneously. A significant fraction of the beam is lost during these splitting processes, driving radioactive activation of the Lambertson septa. This contribution investigates reducing the splitting losses by optimising the vertical beam distribution before extraction. The developed method relies on resonant excitation with the transverse damper, reshaping the Gaussian vertical profile into a uniform and ultimately bimodal distribution, with the excitation amplitude acting as a single continuous control knob. A loss reduction of up to 23.5 % was measured. In simulations, the vertical amplitude detuning is established as a second knob that reduces the resulting beam envelope. This technique was verified with measurements at the CERN Proton Synchrotron. Surrogate models based on generative deep learning were developed for efficient real-time validation and Bayesian system identification.

physics.acc-ph↗

Surface-muon production for ultraslow-muon sources of $μ^+$-based colliders

Anti-muon colliders based on ultraslow muons, such as $μ$TRISTAN and the $μ^+p$ collider LH$μ$C, need muon intensities well beyond those of present facilities. We study surface-\mup\ production by 3~GeV protons in graphite, molybdenum and tungsten targets with Geant4, for single, tilted and multi-layer target geometries. As single targets, molybdenum and tungsten give comparable yields, more than twice that of graphite, because a larger fraction of the beam interacts. Tilting a target with respect to a narrow beam and splitting it into several layers both raise the yield substantially: five-layer Mo and W targets give 6--8 times the surface-muon yield quoted for the J-PARC MLF graphite target, counted at the production target before capture and transport.

physics.acc-ph↗