Parametrized classifiers for optimal EFT sensitivity

Siyu Chen,Alfredo Glioti,Giuliano Panico,Andrea Wulzer

doi:10.1007/jhep05(2021)247

Abstract

We study unbinned multivariate analysis techniques, based on Statistical Learning, for indirect new physics searches at the LHC in the Effective Field Theory framework. We focus in particular on high-energy ZW production with fully leptonic decays, modeled at different degrees of refinement up to NLO in QCD. We show that a considerable gain in sensitivity is possible compared with current projections based on binned analyses. As expected, the gain is particularly significant for those operators that display a complex pattern of interference with the Standard Model amplitude. The most effective method is found to be the “Quadratic Classifier” approach, an improvement of the standard Statistical Learning classifier where the quadratic dependence of the differential cross section on the EFT Wilson coefficients is built-in and incorporated in the loss function. We argue that the Quadratic Classifier performances are nearly statistically optimal, based on a rigorous notion of optimality that we can establish for an approximate analytic description of the ZW process.

Highlights

The second step is to turn the new physics theory into concrete predictions
Suboptimal performances are shown in the figure in order to outline more clearly, in section 6, that our method is systematically improvable as long as larger and larger Monte Carlo samples are available
Our results with the MadGraph Next to Leading Order (NLO) Monte Carlo are reported in figure 3 and in table 1

Summary

Introduction

The second step is to turn the new physics theory into concrete predictions. These should be sufficiently accurate, since the EFT operator effects are often a small correction to the pure SM predictions. It should be mentioned that this program occasionally fails It could be impossible for the Monte Carlo codes to provide a sufficiently accurate representation of all the components of the data distribution, for instance of reducible backgrounds from misidentification. A strategy based on intermediate high-level observables is unavoidably suboptimal It would approach optimality only if the fully differential distribution was measured for all the relevant variables, with sufficiently narrow binning. There are often too many discriminating variables to measure their distribution fully differentially, and, even if this was feasible, one would not be able to predict accurately the cross section in too many bins In this situation, the sensitivity to the presence (or absence) of the EFT operators could be strongly reduced and it could be impossible to disentangle the effect of different operators and resolve flat directions in the parameter space of the EFT Wilson coefficients. The free parameters of the phenomenological modeling of the transfer functions are fitted to Monte Carlo samples

Objectives

Results

Conclusion