DCTracks: An Open Dataset for Machine Learning-Based Drift Chamber Track Reconstruction

Qian Liyan; Zhang Yao; Yuan Ye; Zhang Zhaoke; Fang Jin; Jiang Shimiao; Zhang Jin; Li Ke

cs.LGFeb 16, 2026

DCTracks: An Open Dataset for Machine Learning-Based Drift Chamber Track Reconstruction

Authors: Qian Liyan, Zhang Yao, Yuan Ye, Zhang Zhaoke, Fang Jin, Jiang Shimiao, Zhang Jin, Li Ke, +6 more

Organizations: Institute of High Energy Physics, Chinese Academy of Sciences No.19B Yuquan Road, Shijingshan, Beijing, China · University of Chinese Academy of Sciences No.19A Yuquan Road, Shijingshan, Beijing, China · Sun Yat-sen University School of Science, Shenzhen Campus of Sun Yat-sen University, Shenzhen, China · China Academy of Space Technology No.104 Youyi Road, Haidian, Beijing, China · Institute of Automation Chinese Academy of Sciences No.95 Zhongguancun East Road, Haidian, Beijing, China · Key Laboratory of Particle Physics and Particle Irradiation (MOE), Institute of Frontier and Interdisciplinary Science, Shandong University, Qingdao, Shandong, China

Abstract

We introduce a Monte Carlo (MC) dataset of single- and two-track drift chamber events to advance Machine Learning (ML)-based track reconstruction. To enable standardized and comparable evaluation, we define track reconstruction specific metrics and report results for traditional track reconstruction algorithms and a Graph Neural Networks (GNNs) method, facilitating rigorous, reproducible validation for future research.

Figures & tables

Panel of Figure 1 — Figure 1: BESIII MDC structure (left) and 3D view of the event (right).

Superlayer	Type	$N_{\mathrm{layer}}$	$N_{\mathrm{wire}}/\mathrm{layer}$	Radius (mm)	Length (mm)
1	U	4	40,44,48,56	$\sim$ 79 – 115	780–816
2	V	4	64,72,80,80	$\sim$ 127 – 162	828 – 864
3	A	4	76,76,88,88	$\sim$ 197 – 246	1092 –1272
4	A	4	100,100,112,112	$\sim$ 262 – 311	1442 – 1612
5	A	4	128,128,140,140	$\sim$ 327 – 375	1782 – 1952
6	U	4	160 $\times$ 4	$\sim$ 400 – 448	2174 – 2192

Table 1: MDC layer structure and geometry parameters.

Event Type	$p_{T}$ [GeV/ $c$ ]	$\cos\theta$	$\phi$ [rad]	Particles
Single-track	$0.15\sim 1.5$	$-0.93\sim 0.93$	$0\sim 2\pi$	$e^{\pm},\mu^{\pm},\pi^{\pm},K^{\pm},p,\bar{p}$
Conventional two-track	$0.15\sim 1.5$	$-0.93\sim 0.93$	$0\sim 2\pi$	$\pi^{+}\pi^{-}$
Close-by two-track	$0.15\sim 1.5$	$-0.93\sim 0.93$	$\Delta\phi=0.2$	$\pi^{+}\pi^{-}$

Table 2: Kinematic settings for single-track and two-track event simulation.

Panel of Figure 2 — Figure 2: Displays of the simulated events in the x-y plane for a single-track event (left), a conventional two-track event (middle) and a close-by two-track event (right).

Panel of Figure 3 — Figure 3: Displays of reconstructed events in the x-y plane for a single-track event (left), a conventional two-track event (middle) and a close-by two-track event (right). A condensation point on the track provides estimates the track parameters. This concept is closely related to the GNN finding method we use.

in %	Event Type	$\bm{\epsilon_{\text{hit}}}$	$\bm{p_{\text{hit}}}$
Baseline Finder	Single-track ( $\pi^{+}$ )	$92.24\begin{subarray}{c}+0.12\\ -0.12\end{subarray}$	$98.58\begin{subarray}{c}+0.05\\ -0.05\end{subarray}$
GNN Finder		$92.20\begin{subarray}{c}+0.12\\ -0.12\end{subarray}$	$98.91\begin{subarray}{c}+0.05\\ -0.05\end{subarray}$
Baseline Finder	Conventional two-track ( $\pi^{+}\pi^{-}$ )	$90.87\begin{subarray}{c}+0.14\\ -0.14\end{subarray}$	$97.93\begin{subarray}{c}+0.07\\ -0.07\end{subarray}$
GNN Finder		$91.62\begin{subarray}{c}+0.13\\ -0.13\end{subarray}$	$98.83\begin{subarray}{c}+0.05\\ -0.05\end{subarray}$
Baseline Finder	Close-by two-track ( $\pi^{+}\pi^{-}$ )	$91.26\begin{subarray}{c}+0.16\\ -0.16\end{subarray}$	$97.95\begin{subarray}{c}+0.08\\ -0.08\end{subarray}$
GNN Finder		$82.68\begin{subarray}{c}+0.21\\ -0.21\end{subarray}$	$97.89\begin{subarray}{c}+0.08\\ -0.08\end{subarray}$

Table 3: Hit efficiency (

\epsilon_{\text{hit}}

) and hit purity (

p_{\text{hit}}

) for different event categories.

Panel of Figure 4 — Figure 4: Hit efficiency and hit purity for tracks found by both the GNN Finder and the Baseline Finder . Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for single-track $\pi^{+}$ events.

Single-track $\pi^{+}$ events
in %	$\epsilon_{\text{track}}$	$\epsilon_{\text{track,q}}$	$R_{\text{clone}}$	$R_{\text{fake}}$	$R_{\text{wrong,q}}$
Baseline Finder	$99.71\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$	$99.69\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$	$0.07\begin{subarray}{c}+0.01\\ -0.01\end{subarray}$	$0.01$	$0.02\begin{subarray}{c}+0.01\\ -0.01\end{subarray}$
GNN Finder	$99.81\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$	$99.55\begin{subarray}{c}+0.03\\ -0.03\end{subarray}$	$0.00$	$0.01$	$0.27\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$
Baseline Fitter	$99.70\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$	$99.68\begin{subarray}{c}+0.03\\ -0.03\end{subarray}$	$0.06\begin{subarray}{c}+0.01\\ -0.01\end{subarray}$	$0.01$	$0.02\begin{subarray}{c}+0.01\\ -0.01\end{subarray}$
GNN Fitter	$99.75\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$	$99.50\begin{subarray}{c}+0.03\\ -0.03\end{subarray}$	$0.00$	$0.01$	$0.25\begin{subarray}{c}+0.02\\ -0.02\end{subarray}$
Conventional two-track $\pi^{+}\pi^{-}$ events

Table 4: Track finding and fitting efficiencies for different event categories.

Panel of Figure 5 — Figure 5: Track efficiency and track charge efficiency for tracks found by both the GNN Finder (orange) and the Baseline Finder (blue) with and without fitting. Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for single-track $\pi^{+}$ events.

Figure 6: Helix parametrization of a particle trajectory.

Panel of Figure 7 — Figure 7: Relative transverse momentum resolution for tracks found by both the GNN Finder and the Baseline Finder with and without fitting. Results are shown as a function of $p_{\mathrm{T}}^{\mathrm{MC}}$ for single-track events (left), conventional two-track events (middle) and close-by two-track (right) events.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 8 — Figure 8: Hit efficiency and hit purity for tracks found by both the GNN Finder and the Baseline Finder . Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for conventional two-track $\pi^{+}\pi^{-}$ events.

Panel of Figure 9 — Figure 9: Hit efficiency and hit purity for tracks found by both the GNN Finder and the Baseline Finder . Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for close-by two-track $\pi^{+}\pi^{-}$ events.

Panel of Figure 10 — Figure 10: Track efficiency and track charge efficiency for tracks found by both the GNN Finder (orange) and the Baseline Finder (blue) with and without fitting. Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for conventional two-track $\pi^{+}\pi^{-}$ events.

Panel of Figure 11 — Figure 11: Track efficiency and track charge efficiency for tracks found by both the GNN Finder (orange) and the Baseline Finder (blue) with and without fitting. Results are shown as functions of $p_{\mathrm{T}}^{\mathrm{MC}}$ (left column) and $\cos\theta^{\mathrm{MC}}$ (right column) for close-by two-track $\pi^{+}\pi^{-}$ events.

Panel of Figure 12 — Figure 12: Track parameters of single-track $\pi^{+}$ events.

Panel of Figure 13 — Figure 13: Track parameters of conventional two-track $\pi^{+}\pi^{-}$ events.

Panel of Figure 14 — Figure 14: Track parameters of close-by two-track $\pi^{+}\pi^{-}$ events.

Explore similar work

Using Line Segment Tracking to enable ML-based Track Reconstruction in CMS
Oct 11, 2026Aashay Arora on behalf of the CMS CollaborationHigh-Energy Physics
University of California San Diego, La Jolla, CA, USA
Vision Transformers and Graph Neural Networks for Charged Particle Tracking in the ATLAS Muon Spectrometer
Mar 26, 2026Jonathan RenuschHigh-Energy Physics Graph Neural Networks
on behalf of the ATLAS Collaboration · CERN, Geneva, Switzerland
Characterization and upgrade of a quantum graph neural network for charged particle tracking
Mar 9, 2026Matteo Argenton, Laura Cappelli, Concezio BozziQuantum Neural Networks High-Energy Physics
Department of Physics and Earth Science, University of Ferrara, Via Saragat 1, Ferrara, 44122, Italy · INFN Ferrara, Via Saragat 1, Ferrara, 44122, Italy

Oct 11, 2026J/K move · Enter open · S save

Using Line Segment Tracking to enable ML-based Track Reconstruction in CMS

Aashay Arora on behalf of the CMS Collaboration

University of California San Diego, La Jolla, CA, USA

High-Energy Physics

Open paper AskPDF arXiv

High-pileup conditions in CMS during the HL-LHC era make charged-particle tracking increasingly challenging as detector occupancy and combinatorics grow. We present a hybrid approach that exploits Line Segment Tracking (LST) objects rather than individual hits to enable the first CMS ML-based track reconstruction algorithm. The LST objects, namely the triplets used in this study, are built according to geometry- and physics-driven criteria and carry richer local structure, geometrical cues, and physics quantities such as transverse momentum, which helps the model incorporate detector-level information from the start. We apply a Transformer encoder with geometrically masked sparse self-attention to embed these objects in a learned latent space, and use object condensation to assemble full tracks in a single inference step. This object-based representation reduces complexity in dense environments and makes the model less sensitive to ambiguities introduced by high pileup. Evaluated on CMS Phase-2 simulation of

t\bar{t}

events at

\langle μ\rangle = 200

, the approach achieves reconstruction efficiency comparable to the LST quintuplet baseline while reducing the fake rate by roughly a factor of three and suppressing duplicates almost entirely, demonstrating the promise of advanced ML architectures for next-generation particle tracking.

Abstract

Explore similar work

Using Line Segment Tracking to enable ML-based Track Reconstruction in CMS

Vision Transformers and Graph Neural Networks for Charged Particle Tracking in the ATLAS Muon Spectrometer

Characterization and upgrade of a quantum graph neural network for charged particle tracking