Study of pp->HH-> bbγγ production at the LHC using Neural Networks in simulated events from the ATLAS experiment

After the announcement of the Higgs boson's discovery in 2012, its properties have been the focus of multiple studies, with determining the Higgs potential being the main goal. Higgs boson pair production is the key process to probe the Higgs self-coupling, which defines the shape of the Higgs potential. Simulated ATLAS Run 2 events, corresponding to an integrated luminosity of 140/fb at sqrts=13TeV, are used to perform the event selection and calculate the standard cutflow efficiency, followed by a comparison of two b-tagging algorithms and several working points. Four neural network classifiers are developed to enhance signal-to-background discrimination for both Run 2 and Run 3 (sqrts=13. 6TeV at 168/fb), in the low-mass (mbbγγ*=350GeV) regions. Their performance is validated using the binned Asimov significance and bootstrap stability tests. The total binned Asimov significance is found to be 0. 487+/-0. 022 for Run 2 and 0. 660+/-0. 029 for Run 3. A Poisson likelihood function is constructed to extract the signal strength parameter (μ). By performing 10, 000 pseudo-experiments, applying Poisson fluctuations around the Monte Carlo number of events as the observed data, the behavior of real data is emulated. The resulting 68% confidence level intervals of μ are found to be -0. 962, 3. 099 for Run 2 and -0. 529, 2. 649 for Run 3. An optimal neural network score cutoff is deterimined to define an enriched signal region. The low- and high-mass models are then combined and the Run 3 results are extrapolated to the expected luminosity of the High-Luminosity LHC of approximately 3000/fb. The total binned Asimov significance is calculated 2. 963+/-0. 183, accompanied by a 68% confidence level interval of μ 0. 614, 1. 387, demonstrating promising future analysis potential.