Discriminating QCD Compton and Quark-Antiquark Annihilation Processes in $γ$ + Jets Using Interpretable Machine Learning

TL;DR

The paper addresses the challenge of distinguishing between QCD Compton and quark–antiquark annihilation in $pp$ collisions that produce a photon plus jets, by leveraging jet substructure observables. It trains interpretable classifiers (BDT and MLP) on labeled quark- and gluon-initiated jets from dijet events and applies them to $\gamma + {\rm jet}$ samples to infer the underlying production mechanism. The analysis finds that jet multiplicity and jet girth provide the strongest discrimination, with jet mass offering a smaller contribution and jet charge contributing little. The observed separation saturates at high $p_{T,jet}$ due to intrinsic QCD radiation, establishing a physics-driven baseline for precision jet studies across $pp$, $ep$/A, and heavy-ion collisions.

Abstract

We investigate how effectively final-state jet substructure can discriminate between QCD Compton and quark-antiquark annihilation processes from photon-jet production in $pp$ collisions at $\sqrt{s}=13$ TeV. Using infrared- and collinear-safe jet observables, multivariate classifiers -- boosted decision trees and multilayer perceptrons -- are trained on labeled quark- and gluon-initiated jets from dijet events and applied to photon-jet samples. Observables probing soft and wide-angle radiation, in particular jet multiplicity and jet girth, dominate the discrimination. The jet mass provides a complementary but weaker contribution, while the jet charge exhibits negligible discriminating power. A comparison of the two classifiers demonstrates that the achievable separation is limited primarily by QCD radiation effects rather than by classifier complexity. These findings quantify the extent to which information about the underlying hard process survives hadronization and realistic jet reconstruction, providing a physics-driven baseline for precision jet measurements in $pp$, $ep/$A, and heavy-ion collisions.

Figures (8)

• Figure 1: Seven input features for Quark (green) vs. Gluon (orange) Jet used for training: $p_{\rm T,jet}$, $\eta_{\rm jet}$, $\phi_{\rm jet}$, $N_{\rm const}$, $M_{\rm jet}$, $Q^{\rm ch}_{\kappa}$. These quantities are calculated from $pp$ collisions at $\sqrt{s}$=13 TeV using PYTHIA8 Detroit tune dijet sample with $p_{\rm T,jet}$$>$ 30 $\rm GeV/{\it c}$.
• Figure 2: The MLP score distribution from $\gamma + {\rm jet}$ sample. The MLP Score $>0.5$ ($<0.5$) is used for gluon-like (quark-like) jet.
• Figure 3: The BDT score distribution from $\gamma + {\rm jet}$ sample. The BDT Score $>0.2$ ($< -0.2$) is used for quark-like (gluon-like) jet.
• Figure 4: Using the MLP model: $p_{\rm T,jet}$, $girth$, $M_{\rm jet}$, $N_{\rm const}$, and $Q^{\rm ch}_{\kappa}$ distributions are shown for $\gamma + {\rm jet}$ with 50 $<$$p_{\rm T,jet}$$<$ 100 $\rm GeV/{\it c}$ . Here, the $y$-axis is normalized by the total number of jets ($N$). The blue hashed region and blue marker represent the true quark and predicted quark-like jet (from QCD Compton process); the same for red color represents gluon jet (from the annihilation process).
• Figure 5: Same for the BDT model.