IMAGINE

— A Bayesian Framework —

Coordinators: Torsten Enßlin, Marijke Haverkorn, Jörg Rachen

Project Outline
A core part of our team proposal has been the development of a software package to support the convergence of the research areas of Galactic magnetism and ultra-high-energy cosmic rays. Preliminarily coined the “Bayesian Machine”, it has been conceived as a tool which

  • allows to generate all standard observables of a variety of GMF models,
  • can sample the likelihood function, resp. in the presence of non-flat priors, the Bayesian posterior over the model parameter space,
  • allows to export the best fit model parameters and their covariance for usage in other tools (like CRPropa), and, in its later development stages,
  • link to external observable generators and use their likelihood (posterior) to achieve optimizations also over non-standard, heterogeneous datasets.

With these properties, it would be able to reflect the growth of knowledge in all relevant fields for our common understanding of the Galactic magnetic field and the origin of ultra-high energy cosmic rays. The development of the “Bayesian Machine” has been a driver through both meetings, but only in the latter a software concept was developed and work packages assigned. There it also received its name: IMAGINE – Interstellar MAGnetic field Inference Engine.

IMAGINE-FlowchartThe diagram shows the design outline for IMAGINE. It is realized in Python, wrapping the latest version of the Hammurabi code (C++) as its observable generator, which also brings in a set of actual magnetic field parametrizations, among them JF12 and Jaffe13 (see Planck Collaboration, 2016). For Bayesian parameter sampling, the PyMultiNest package (Buchner et al. 2014) is used. The IMAGINE pipeline is currently tested against Mock data produced from JF12 realizations, and shall be used for testing their parameter optimizations against the most recent data sets. Later it shall be used to optimize the dynamo based magnetic field model developed in the framework of this team.

Project Status
The development and test of IMAGINE is a central part of the PhD thesis of Theo Steininger (MPI for Astrophysics, Garching) and the Master project of Ellert van der Velden (Radboud University). A publication in a refereed journal is foreseen during 2017. A first presentation of IMAGINE to the community and the planning for its further development is subject of an upcoming workshop at the Lorentz Center in Leiden on 20-24 March 2017.