Deployment of a Matrix Element Method code for the ttH channel analysis on GPU’s platform

Abstract

The observation of the associated production of the Higgs boson with two top quarks in proton-proton collisions is one of the highlights of the LHC Run 2. Driven by the theoretical description of the physics processes, the Matrix Element Method (MEM) consists in computing a probability that an event is compatible with the signal hypothesis (ttH) or with one of the background hypotheses. It is a powerful classifying tool requiring high dimensional integral computations. The deployment of our MEM production code on GPU's platform will be described. What follows will focus on the adaptation of the main components of the computations in OpenCL kernels, namely the Magraph matrix element code generator, VEGAS, and LHAPDF. Finally, the gain obtained on GPU's platforms compared with classical CPU's platforms will be assessed. *

The Matrix Element Method used in the search for the associated production of the Higgs boson with top quarks and decaying into tau leptons

“…All the parton combinations are described by the Σ p term and σ i is a normalization factor tuned to achieve the best signalto-background separation when combining the weights into a likelihood ratio. The computation of these weights are CPU intensive and a GPU-based derivation has been proposed successfully [3]. An excess has been observed with respect to the background-only hypothesis in the multilepton final states: 3.2σ observed significance (2.8σ expected) [2].…”

Matrix element regression with deep neural networks — Breaking the CPU barrier

“…Many developments have been made recently in the context of the MEM by using either parallel computing [22,23] or GPU acceleration [24,25]. In addition methods that bypass the classic numerical integration libraries by using boosted decision trees [26] or neural networks [27], and the recent new applications of normalizing flows for phase-space sampling [28,29] are promising ways to improve the computation time and potentially to avoid the currently required integration variables optimizations such as implemented in MoMEMta.…”

Many-gluon tree amplitudes on modern GPUs: A case study for novel event generators

“…In the past, several conceptual studies have shown substantial performance gains, both for tree-level computations [27][28][29][30][31][32][33][34][35][36] and for one-loop level computations [37,38]. In addition, the evaluation of PDFs [39] and the matrix element method [40][41][42] have been explored on GPUs. Nevertheless, a production-ready GPU-enabled event generator suitable for experimental applications has not yet become available.…”