Journals of Accelerator Conferences Website (JACoW)

JACoW is a publisher in Geneva, Switzerland that publishes the proceedings of accelerator conferences held around the world by an international collaboration of editors.

URLhttps://doi.org/10.18429/JACoW-IPAC2026-THP5364
TitleAccelerator optimizations using SciBmad with PyTorch
Authors
  • C. Ung
    Cornell University
  • M. Signorelli, G. Hoffstaetter, D. Sagan
    Cornell University (CLASSE)
AbstractSciBmad is a new, fully differentiable software ecosystem for accelerator physics, usable in Julia and/or Python. Full differentiability enables efficient and numerically-reliable optimizations of particle accelerators, in particular with machine learning (ML) methods. One of the most commonly used ML frameworks is PyTorch, which provides powerful, versatile, and straightforward tools for such purposes. We implemented PyTorch bindings with SciBmad, enabling seamless integration of SciBmad’s powerful and differentiable simulations with a standard PyTorch workflow. With these bindings, all of PyTorch’s powerful tools can be used easily with SciBmad. In this paper, we describe the implementation details and present some examples demonstrating SciBmad-PyTorch workflows.
Paperdownload: THP5364.pdf
CiteBibTeX, LaTeX, Text/Word, RIS, EndNote
Conference17th International Particle Accelerator Conference
Series
LocationDeauville, France
Date17-22 May 2026
PublisherJACoW Publishing, Geneva, Switzerland
Editorial BoardEditorial Board
Online ISBN978-3-95450-252-3
Online ISSN2673-5350
Received13 May 2026
Revised06 June 2026
Accepted
Issued20 July 2026
DOI10.18429/JACoW-IPAC2026-THP5364
Pages4810-4813