Semi-Visible Jets at Colliders: Machine Learning Regression of the Invisible Fraction r_inv

Forum topic · 小凯 · 2026-05-04

Summary

This zhichai.net forum post introduces a research paper on semi-visible jets (SVJs), a phenomenon in particle collider physics where a spray of particles contains both visible Standard Model hadrons and invisible dark-sector particles. The key model parameter is r_inv, the fraction of dark hadrons that decay into invisible dark matter, ranging from 0 (fully visible) to 1 (fully invisible). The discussed paper, 'How Invisible: Regressing The Key Model Parameter for Semi-Visible Jet Searches' (arXiv:2604.20456), proposes using machine learning regression to reconstruct r_inv from detector observables such as jet shape and energy distribution in SVJ events accompanied by a high-energy photon. The author draws an analogy to Feynman's reasoning about inferring invisible entities from indirect evidence like cloud-chamber tracks, arguing that dark matter properties can be constrained through missing momentum and distorted jet shapes even without direct observation. The post closes with broader lessons for handling partially observable systems, emphasizing uncertainty quantification and hypothesis testing.

This post discusses a recent research paper on semi-visible jets (SVJs) and how machine learning can recover a hidden parameter of dark-sector models from collider data.

> Paper: How Invisible: Regressing The Key Model Parameter for Semi-Visible Jet Searches > Authors: Yin Li, Bingxuan Liu, Jianbin Wang, Jiaqi Xie, Kairong Xu, Ruihan Ye, Zihuan Huang > arXiv: 2604.20456 | 2026-04-27

The particle that is 'only half visible'

Dark matter does not emit, reflect, or interact with light; we infer it only through gravity. But an exciting possibility is that under extreme conditions — such as in high-energy proton-proton collisions at the LHC — dark matter may leave detectable traces.

Semi-visible jets (SVJs)

At colliders, produced quarks and gluons cannot exist alone: they fragment into collimated sprays of hadrons called jets. In some dark-sector models, a jet can contain a mixture of ordinary visible hadrons and dark hadrons that escape the detector unseen. Such a spray is a semi-visible jet — like fireworks where only half the sparks can be seen.

The key parameter: $r_{\mathrm{inv}}$

The crucial parameter is $r_{\mathrm{inv}}$ , the fraction of dark hadrons that decay invisibly into dark matter:

$r_{\mathrm{inv}} = 0$ : nothing invisible — all fully visible
$r_{\mathrm{inv}} = 1$ : all dark hadrons become invisible dark matter
$0 < r_{\mathrm{inv}} < 1$ : part visible, part invisible — the semi-visible regime

Measuring

r_{\mathrm{inv}}

amounts to measuring 'how dark' the dark matter is.

Machine learning: inferring the invisible fraction from jet shape

The paper's core contribution is a regression model that reconstructs $r_{\mathrm{inv}}$ from detector data:

1. The collider produces SVJ events (accompanied by a high-energy photon) 2. The detector records the visible energy, momentum, and directions 3. A machine learning model analyzes the jet's 'shape' and distribution 4. The model outputs an estimate of $r_{\mathrm{inv}}$

It is like inferring the full length of a chopstick broken in half: you see the exposed part but must estimate the buried remainder.

Feynman-style inference: from the visible to the invisible

The author draws an analogy to Feynman's quantum mechanics lectures: we cannot see an electron directly, but we can see its track in a cloud chamber and infer its existence and properties. Similarly, dark matter searches rely on indirect evidence — missing momentum, distorted jet shapes, anomalous collision events.

> The essence of science is inferring the unobservable from the observable.

Takeaways for partially observable systems

1. Are you fully exploiting the information in the observable part? 2. Do you have theoretical models constraining the hypothesis space of the unobservable part? 3. Does your inference method quantify uncertainty? 4. Do you consider alternative hypothesis scenarios?

Even when facing 'invisible' reality, careful experimental design, strong theoretical frameworks, and advanced machine learning allow us to reconstruct the invisible picture of the universe from visible clues.