iftpwa - Introduction#
Information Field Theory for Partial Wave Analysis (iftpwa)#
When multiple resonances overlap in phase space, partial wave analysis is needed to disentangle the individual contributions. One can think of it as a generalized Fourier analysis where each basis component can represent a physically interpretable process (caveats!).
See also
Want to learn more about Information Field Theory? Checkout these Notes
Design Requirements#
uncertainty quantification via a Bayesian framework
handles high dimensionality via flexible prior models for regularization
allow analysis of large datasets using MPI and variational inference
complex valued parameters
modular and extensible / collaboration agnostic
This led to the development of the Information Field Theory for Partial Wave Analysis (iftpwa) package by Florian Kaspar (et al.) at the Technical University of Munich (TUM) with support from the Max Planck Institute for Astrophysics (MPA), which develops the Numerical Information Field Theory (NIFTy) probabilistic programming framework for Bayesian inference. GlueX later joined in on the project which I am a part of as of this writing. GlueX uses a polarized photon beam which allows access to additional information on the reaction dynamics. This additional information comes at the cost of increased complexity in the analysis and can lead to unstable results using traditional methods which Bayesian methods like NIFTy could help regulate.
NIFTy provides Metric Gaussian Variational Inference and Geometric Variational Inference methods for optimization. Both of these approaches alternate between optimizing the KL divergence for a specific shape of the variational posterior and updating the shape of the variational posterior, see here. A single iteration contains both these steps.