Correcting a measured distribution for detector smearing to recover the true one — an ill-posed inversion that needs regularisation.Unfolding is the process of correcting a measured distribution for the smearing effects of a detector. In particle physics, for example, a detector does not record the true energy or momentum of a particle; it records a blurred version, with resolution limits and efficiencies that distort the underlying physics. Unfolding aims to invert this response function to recover the true distribution, which is essential for comparing experimental data with theoretical predictions.
The problem has been recognized since the early days of experimental physics, with formal methods developed in the 1960s and 1970s, notably by G. D'Agostini in the 1990s for the iterative Bayesian approach. The term 'unfolding' became standard in high-energy physics.
Mathematically, the measured distribution is the convolution of the true distribution with a detector response matrix. Unfolding attempts to invert this convolution. However, this inversion is ill-posed: small statistical fluctuations in the measured data can lead to enormous, oscillatory errors in the unfolded result. To stabilize the solution, regularisation is applied — for example, by adding a penalty for large fluctuations (Tikhonov regularisation) or by using an iterative Bayesian approach that incorporates prior knowledge and smoothness constraints.
Unfolding is used in nearly every particle physics measurement, from cross-section determinations at the LHC to neutrino oscillation analyses, and also in astrophysics for deconvolving telescope images.
The ill-posed nature of unfolding is the same mathematical issue that plagues image deblurring in astronomy and medical imaging, where naive inversion amplifies noise into wild artifacts.
No articles yet