Feasibility studies for HH->bbtautau at the FCC-hh using graph-neural networks

A feasibility study of the di-Higgs production channel at the FCC-hh has been conducted focusing on the $ hh \to b\bar{b} \tau^+ \tau^- $ final state. Both the leptonic-hadronic ($\ell$-had) and fully hadronic (had-had) decays of the taus were considered, leveraging advanced Graph Neural Network (GNN) techniques for event classification.  The analysis employed Graph Attention Networks (GATs), with each event represented as a fully connected graph, where nodes corresponded to reconstructed physics objects, and edges captured their relational properties. Additional complex features—such as invariant masses of b-jet and taus, angular separations, and transverse kinematic variables—were integrated to enhance event classification. This innovative application of GNNs significantly improved the ability to extract signal events, offering a powerful technique for precision Higgs self-coupling measurements at hh colliders. At 84 TeV centre-of-mass energy, the Higgs self-coupling could be constrained at the XXX\% level using the had-had channel alone, and precise cross-section measurements will be possible in different invariant mass regions ($ m_{hh} < 350 $ GeV and $ m_{hh} > 350 $ GeV). At 100 TeV the uncertainty improves by YYY\%.