Improving the Direct Determination of $|V_{ts}|$ using Deep Learning

TL;DR

This work tackles the direct determination of the CKM element $|V_{ts}|$ by exploiting $t\bar t$ dilepton events and a novel s-jet tagging strategy. It introduces the DiSaJa family of Transformer-based discriminators, with DiSaJa-H using multi-domain inputs (jets, leptons, MET) and DiSaJa-L incorporating jet constituents via a jet-constituent encoder, achieving superior event-level discrimination over an input-baseline. Training strategies incorporate signal-only and signal-plus-background configurations, demonstrating substantial gains in significance, with HL-LHC projections indicating potential discovery-level sensitivity and a 95% CL interval for $|V_{ts}|$ substantially tighter than current indirect constraints. The approach leverages full event information and demonstrates robust performance improvements, suggesting broad applicability to other collider analyses involving rare jet-flavor decays.

Abstract

An $s$-jet tagging approach to determine the Cabibbo-Kobayashi-Maskawa matrix component $|V_{ts}|$ directly in the dileptonic final state events of the top pair production in proton-proton collisions has been previously studied by measuring the branching fraction of the decay of one of the top quarks by $t \to sW$. The main challenge is improving the discrimination performance between strange jets from top decays and other jets. This study proposes novel jet discriminators, called DISAJA, using a Transformer-based deep learning method. The first model, DISAJA-H, utilizes multi-domain inputs (jets, leptons, and missing transverse momentum). An additional model, DISAJA-L, further improves the setup by using lower-level jet constituent information, rather than the high-level clustered information. DISAJA-L is a novel model that combines low-level jet constituent analysis with event classification using multi-domain inputs. The model performance is evaluated via a CMS-like LHC Run 2 fast simulation by comparing various statistical test results to those from a Transformer-based jet classifier which considers only the individual jets. This study shows that the DISAJA models have significant performance gains over the individual jet classifier, and we show the potential of the measurement during Run 3 of the LHC and the HL-LHC.

Figures (12)

• Figure 1: Normalized distributions of the transverse momentum, pseudorapidity, and mass of reconstructed jets matched to prompt $s$ and $b$ quarks from top-quark decays in fully matched $t\bar{t} \to sWbW$ and $t\bar{t} \to bWbW$ events after the event selection.
• Figure 2: Architecture of the original SaJa model (left) and the DiSaJa-H (right). $N_{x}$, $D_{x}$, and C denote the number of the object $x \in$ {jets, leptons, MET}, the dimension size of the object $x$, and the number of output categories, respectively.
• Figure 3: Detailed network structure of blocks of Feed-Forward Network, Encoder, and Decoder. The decoder processes the output of the jet embedder as the target input and the output of the event encoder block as the source input.
• Figure 4: Architecture of the jet constituent encoder, which can replace the jet high-level feature encoder. Track and tower features are fed into encoders and the jet constituent encoder learns jet representation.
• Figure 5: Distribution of the highest $\Pqt\to\Pqs\PW$ score used as a discriminant in the signal sample, showing jets that are correctly assigned, wrongly assigned, and unmatched with partons. The ratios for these three categories are reflected in the distributions.