Recommendations for the Modeling of Smooth Backgrounds

In data analyses that exploit a distribution of the data, backgrounds are often modeled using a continuous description of the distribution shape. This technique is used in particular for situations involving narrow signal peaks and wide sidebands of regular backgrounds, for example in the study of Higgs boson decays or in searches for narrow resonances. This note reviews the main techniques in use within ATLAS for smooth background modeling: closed-form functions, Gaussian processes and Functional decomposition. In all cases, the chosen model must provide a sufficiently accurate description of the background distributions. Systematic uncertainties should also be included to accounts for possible residual mismodeling effects. Criteria used to select appropriate models and methods to define the corresponding modeling uncertainties are described, and recommendations applicable to various analysis configurations are provided.