RooStats

RooStats
Name	RooStats
Title	RooStats
Released	2011
Programming language	C++, Python
Operating system	Cross-platform
License	BSD

RooStats is a software toolkit for statistical modeling and inference developed within the CERN ROOT (software) framework. It provides implementations of frequentist and Bayesian methods designed for high-energy physics applications, connecting libraries and tools used at CERN, Fermilab, DESY, SLAC National Accelerator Laboratory, and other particle physics institutions. RooStats supports complex likelihood construction, hypothesis testing, and interval estimation used in collaborations such as ATLAS, CMS, LHCb, and in experiments at Tevatron and LEP.

Introduction

RooStats originated to unify statistical practice across experiments at CERN and to provide interoperable tools for collaborations like ATLAS Collaboration and CMS Collaboration. It was developed by contributors from CERN teams as part of the ROOT (software) project, drawing on statistical research from institutes such as University of Oxford, Harvard University, Princeton University, University of California, Berkeley, and University of Chicago. RooStats interoperates with software ecosystems including ROOT (software), RooFit, HistFactory, BAT (Bayesian Analysis Toolkit), MINUIT, and tools employed by Institute of High Energy Physics, Chinese Academy of Sciences researchers.

Design and Architecture

RooStats architecture builds on the object model of RooFit within ROOT (software), organizing components into calculators, workspaces, and models used by projects at CERN, Fermilab, and DESY. Core classes interface with minimizers such as MINUIT and with sampling engines used by Markov chain Monte Carlo implementations and by packages from Stan (software), PyMC3, and TensorFlow Probability when combined through wrappers. The design emphasizes modularity for collaborations like CMS Collaboration to plug in custom likelihood components and systematic uncertainties handled by frameworks used at Brookhaven National Laboratory, Lawrence Berkeley National Laboratory, and SLAC National Accelerator Laboratory.

Statistical Methods and Tools

RooStats implements hypothesis tests and interval estimation aligned with procedures used in analyses at ATLAS and CMS, including profile likelihood ratio tests related to results reported at Large Hadron Collider conferences and meetings at International Conference on High Energy Physics. It provides calculators for frequentist methods inspired by work from Jerome Neyman and Sir Ronald Fisher as well as Bayesian routines reflecting practices used by groups at Imperial College London and Carnegie Mellon University. RooStats integrates with asymptotic formulae derived in publications circulated at arXiv and presented at workshops like PHYSTAT. Tools support combination procedures analogous to those used in joint results from Tevatron experiments and global fits performed by groups such as Global Electroweak Fit teams.

Workflow and Usage

Typical workflows mirror analysis chains of ATLAS and CMS and often begin in environments maintained by CERN analysis groups, using RooWorkspace objects to store models and datasets. Analysts from institutions such as University of California, San Diego, University of Michigan, ETH Zurich, and University College London construct models with components representing signal hypotheses from generators like PYTHIA, HERWIG, and MadGraph, incorporate detector effects modeled with inputs from GEANT4, and then perform fits using optimizers related to MINUIT2. RooStats outputs test statistics and intervals that feed into review processes at collaboration meetings such as those held by the ATLAS Collaboration and CMS Collaboration and into publication pipelines coordinated by publishers like Physical Review Letters and Journal of High Energy Physics.

Performance and Validation

Performance evaluation and validation of RooStats procedures have been conducted in contexts including statistical workshops at CERN and through software validation campaigns involving teams from Fermilab and DESY. Benchmarks compare asymptotic approximations to toy Monte Carlo sampling similar to studies presented at PHYSTAT and evaluated in reviews from groups at Brookhaven National Laboratory, Lawrence Livermore National Laboratory, and Los Alamos National Laboratory. Validation efforts reference statistical standards discussed by scholars affiliated with University of Cambridge, Columbia University, Yale University, and Stanford University and implemented in collaborative analyses at Large Hadron Collider experiments.

Adoption and Applications

RooStats has been adopted by major particle physics collaborations including ATLAS Collaboration, CMS Collaboration, and LHCb Collaboration for searches such as those for the Higgs boson and for measurements reported at conferences like Moriond and Lepton Photon Conference. It is used in combined measurements with inputs from experiments at Tevatron and in astrophysics analyses at institutions such as Max Planck Institute for Astrophysics and Kavli Institute for Cosmology. RooStats also appears in methodological studies by statisticians at University of Washington, University of Toronto, McGill University, and in training materials for schools organized by CERN and by regional centres like Brookhaven National Laboratory.

Category:Physics software