Supersymmetric cross sections using neural networks

Searching for supersymmetry at the Large Hadron Collider one of the most important ingredients is knowledge of the cross section for supersymmetric particle production. In quantum field theory this can be calculated as a series expansion in the coupling (of the strong force), where complications in evaluating the terms grow unreasonably beyond the first term. Results are know for the second term, however, in the form of computationally challenging integrals that must be re-evaluated for every value of the parameters of the supersymmetric model. Typically, an evaluation of all relevant cross sections for a given parameter point would take on scale of hours. This makes investigations of the supersymmetric parameter space brutal in terms of the necessary computing power.

We have a long term project ongoing where we use methods from machine learning to perform regression on data sets consisting of these cross sections calculated on superclusters. The goal is to allow fast evaluation of the cross sections -- typically within one second -- and a reliable estimate of the associated error in the regression. In the past we have seen great success with boosted decision trees and Gaussian processes.

The current project will focus on the implementation of neutral networks for regression. This will require both an understanding of the quantum field theory behind the cross section calculation, to understand the physics of the processes, and know-how on the use of neural networks. Good programming skills in python and/or C++ will be a useful starting-point.

Publisert 7. mai 2019 17:38 - Sist endret 7. mai 2019 17:38

Veileder(e)

Omfang (studiepoeng)

60