Pretraining on boosted boson decays for polarization tagging of hadronic W bosons

summary

The gist

Pretraining on boosted boson decays for polarization tagging of hadronic W bosons investigates where the polarization information of a boosted W boson lies and how much a tagger can recover.

In short

The study investigates where polarization information of boosted W bosons is stored and how much a machine learning tagger can recover it. By pretraining a transformer model on unpolarized jets from various particle decays, researchers found that focusing on boson data helps modestly. They determined that jet containment limits the performance of decay-angle taggers, but using the production angle improves results, suggesting event-level tagging is superior.

Key concepts

Longitudinal and Transverse W Bosons
These describe two different ways a W boson can be moving relative to its direction of travel. Longitudinal bosons are rare and carry crucial polarization information that is hard to measure directly. Transverse bosons are more common, and the study focuses on how to distinguish them using jet properties after they are boosted.
Point-Edge Transformer
This is a specific type of neural network architecture used for analyzing jet substructure. It processes jets by looking at both individual constituents (points) and the relationships between them (edges). It was pretrained on many unpolarized jets to learn how to recognize different particle types.
Jet Containment
This refers to a physical constraint where requiring both decay quarks inside the reconstructed jet limits the information available for tagging. This requirement removes the most asymmetric decays, which are often the most distinctive for separating polarization states in standard jet taggers.

Terminology used across episodes

This episode discusses

The paper

Pretraining on boosted boson decays for polarization tagging of hadronic W bosons · Read on arXiv

Saptaparna Bhattacharya

Southern Methodist University

Transcript

Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.

Kai: We're looking at the paper: "Pretraining on boosted boson decays for polarization tagging of hadronic W bosons".

Mira: Here's the gist of the paper. Pretraining on boosted boson decays for polarization tagging of hadronic W bosons investigates where the polarization information of a boosted W boson lies and how much a tagger can recover.

Kai: The paper constructs a pretraining dataset of unpolarized jets from six classes: W+, W−, Z, H, top, and QCD jets.

Mira: The samples are matched in transverse momentum so that the classes cannot be distinguished by kinematics alone.

Kai: A point-edge transformer is pretrained on this dataset with four objectives: classification of the six process classes (pt cls), label-free generative training (pt gen), both objectives combined (pt both), and regression of nine decay-quark targets (pt reg).

Paper summary: Kai: To recap, the paper "Pretraining on boosted boson decays for polarization tagging of hadronic W bosons" investigates where the polarization information of a boosted W boson actually lies and how much a tagger can recover from it.

Mira: They build this dataset using unpolarized jets from six classes: W plus, W minus, Z, H, top quarks, and QCD jets. The goal is to pretrain a point-edge transformer on this data first.

Lev: This pretraining helps modestly at first because the samples are unpolarized so they can't be distinguished by kinematics alone.

Kai: They test four different training objectives for that transformer: classification, label-free generative training, both objectives combined, and regression of nine decay-quark targets.

Mira: The key finding here is that requiring both decay quarks inside the jet truncates the decay-angle distribution at an edge set by m over p T R.

Lev: That edge actually bounds any hadronic polarization tagger, which is a specific limit on how good those traditional taggers can be, according to the paper.

Kai: Within that acceptance region, they show that the decay angle carries essentially all the polarization information of a jet in W plus jet events.

Conclusion: Kai: So we're wrapping up our look at "Pretraining on boosted boson decays for polarization tagging of hadronic W bosons." The authors are basically showing how you can use AI to learn from unpolarized data—like jets from Z bosons or top quarks—before you even get the polarized samples.

Mira: Right. They’re figuring out where that polarization information is hiding inside a boosted W boson and what kind of recovery we can expect from our detector data, especially when those rare longitudinal bosons only show up in about zero point three percent of events at high boosts.

Lev: The core finding here is that you can’t just rely on the decay angle within one jet; that’s a ceiling for how much you get, but adding the production angle—how the whole event is moving—actually helps push performance up by about zero point three to zero point five in AUC.

Kai: It means that even if our detector only gives us limited information inside a single jet, combining it with the overall event kinematics can give us a better measurement than just looking at the decay angles alone.

Mira: That’s what they are suggesting for future measurements of vector-boson scattering, implying that switching from just tagging individual jets to analyzing the whole event gives us significant gains.

Kai: Overall, this paper shows us that context outside of just the jet decay angle is key to unlocking more information from these rare longitudinal bosons. We need to keep building those event-level tools if we want to extract all that polarization data.

More episodes

← Home