Simplified likelihoods using linearized systematic uncertainties

TL;DR

This paper introduces a simplified likelihood framework using linearized systematic uncertainties, enabling faster computations and accurate modeling of non-Gaussian effects in LHC data analysis.

Contribution

It proposes a new simplified likelihood approach based on linearized uncertainties, improving computational efficiency while maintaining accuracy for complex data scenarios.

Findings

Significant reduction in computation time for likelihood evaluations.

Accurate modeling of non-Gaussian effects from low event counts.

Effective handling of correlated uncertainties in data combinations.

Abstract

This paper presents a simplified likelihood framework designed to facilitate the reuse, reinterpretation and combination of LHC experimental results. The framework is based on the same underlying structure as the widely used HistFactory format, but with systematic uncertainties considered at linear order only. This simplification leads to large gains in computing performance for the evaluation and maximization of the likelihood function, compared to the original statistical model. The framework accurately describes non-Gaussian effects from low event counts, as well as correlated uncertainties in combinations. While primarily targeted towards binned descriptions of the data, it is also applicable to unbinned models.