Effects of color reconnection on $t\bar{t}$ final states at the LHC

TL;DR

This work assesses color reconnection as a dominant systematic in top-quark mass determinations at the LHC and shows that the conventional CR-on vs CR-off comparison underestimates the true uncertainty for Pythia 8. By introducing two broad classes of CR schemes—one targeting only top-decay products and another affecting all final-state partons—the authors quantify how different reconnection scenarios shift the reconstructed top mass after tuning to LHC data. They find that, with realistic tuning, the CR-induced uncertainty in the top mass is about 0.5 GeV, though extreme variants can produce larger shifts; this motivates more stringent constraints from future $t\bar{t}$ measurements. The paper proposes specific hadron-level observables in semi-leptonic $t\bar{t}$ events to discriminate CR models and outlines paths to reduce the modeling uncertainty through data-driven tuning and expanded theoretical modeling, including future improvements to the color treatment in event generators.

Abstract

The modeling of color reconnection has become one of the dominant sources of systematic uncertainty in the top mass determination at hadron colliders. The uncertainty on the top mass due to color reconnection is conventionally estimated by taking the difference in the predictions of a model with and a model without color reconnection. We show that this procedure underestimates the uncertainty when applied to the existing models in {\sc Pythia}~8. We introduce two new classes of color reconnection models, each containing several variants, which encompass a variety of scenarios that could be realized in nature and we study how they affect the reconstruction of the top mass. After tuning the new models to existing LHC data, the remaining spread of predictions is used to derive a more realistic uncertainty for the top mass, which is found to be around 500 MeV. We also propose how future LHC measurements with $t\bar{t}$ events can be used to further constrain these models and reduce the associated modeling uncertainty.

Figures (9)

• Figure 1: An example of a gluon emitted from a top decay product $g_t$ exchanging colors with a gluon from another interaction $g_r$. A possible string configuration for this final state is shown on the right. The gluons $g_t,g_r$ are identified with the kinks $j,m$ respectively, while the top and bottom quarks and the incoming and outgoing $q$ from the additional interaction are identified with the string endpoints $i,k,l$ and $n$ respectively. The double lines indicate the configuration that results from the perturbative color flow and the dashed lines indicate a configuration with a smaller $\lambda$, resulting from the exchange of colors and anticolors between gluons $j$ and $m$.
• Figure 2: In the 'move' model, a gluon $j$ originally attached to string piece $ik$ can be moved to a different string piece $lm$ if it leads to a smaller total string length $\lambda$. Solid lines indicate the original configuration and dashed lines indicate the resulting configuration after moving the gluon.
• Figure 3: Illustration of the effect of the 'flip' model. The same process as in figure \ref{['fig:CR']}, shown with the underlying string configuration. The solid lines indicate the initial configuration and the dashed lines represent a flip in the string pieces $ij$ and $lm$, resulting from the exchange of one of the color indices between gluons $j$ and $m$. The figure represents a case where a flip reduces the total string length $\lambda$. We note that after the flip, the $b$ quark from the top decay (endpoint $k$) is color connected to quark $l$ from a separate MPI.
• Figure 4: Most of the models predict a shift of $m_{\mathrm{top}}$ to lower values (left) but there are models that predict a positive shift (right). Models that force the top decay products to reconnect have large effects (left).
• Figure 5: Models of CR that lead to a negative (positive) mass shift usually predict broader (narrower) jets (left). For some of the models this is not the case (right).