Track fitting at the full LHC collision rate
Listen
Radio episode about this paper
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: "Track fitting at the full LHC collision rate".
Mira: Here's the gist of the paper. The LHCb experiment at the Large Hadron Collider underwent a major upgrade before the LHC Run 3 data taking period, employing an all-software approach in its trigger system Here we present a fast implementation of a Kalman filter, used in the first trigger stage since the 2025 data taking period, allowing to determine parameter estimates of charged-particle trajectories at the full LHCb collision frequency of 30 MHz This approach replaces computationally expensive magnetic field map lookups and numerical integration methods with fast analytical parameterisations while maintaining the mathematical framework of Kalman filtering Compared to the previous fitter this parameterised Kalman filter shows a significantly improved momentum resolution, resulting in a factor of two improvement in the invariant mass resolutions for reconstructed D0 and J/ψ hadrons It additionally demonstrates greater robustness against detector misalignment effects and substantially sharpens the discrimination between genuine particle trajectories and accidental background, more than doubling the rejection of the latter at no cost to genuine-track efficiency, for a standard selection working point.
Kai: Algorithm Development The core of this work is a fast implementation of a Kalman filter used in the first trigger stage since the 2025 data taking period This algorithm replaces the previous partial track fitting algorithm in the real-time trigger environment at the cost of a 2% increase in processing time.
Mira: The Parameterised Kalman Filter is a Kalman filter as defined in the previous section, but uses simplifications specific to typical implementations in a high-energy physics context which allows it to be executed with minimal time consumption.
Kai: It was developed as a drop-in replacement for the previous track fitting algorithm, the VELO Only Kalman Filter, to provide better momentum (and therefore mass) resolution, more robustness against detector misalignments, and improved rejection against fake tracks (so-called ghosts) in the HLT1 stage of LHCb.
Title and authors: Kai: So we're looking at this paper, "Track fitting at the full LHC collision rate," and it’s about how they rebuilt the first trigger stage for LHCb using a Kalman filter. It seems like they’re moving away from older methods that were too slow for what the detector can actually produce now.
Mira: That's right, Kai, it touches on making sure we can actually keep up with data taking at the full LHC collision frequency of thirty MHz <ref:2607.14793#pg1,at the full LHC collision>. It’s about this fast implementation of a Kalman filter they use to estimate particle paths.
Lev: From an engineering standpoint, replacing the magnetic field map lookups and numerical integration with analytical parameterizations is smart because it needs to run in real time. If you have to solve equations every time a collision happens, that overhead kills your trigger rate.
Kai: Exactly, and the paper says this new approach allows them to determine parameter estimates of charged particle trajectories at that full thirty MHz frequency without slowing down the trigger system by much more than a two percent increase in processing time <ref:2607.14793#pg1,to determine parameter estimates of charged particle trajectories at>.
Mira: But what’s really interesting is how they use this Kalman filter not just for speed, but for something else—improving the quality of the measurements they get from those tracks.
Lev: The paper implies that this isn't just about fitting tracks faster; it’s about getting better momentum resolution because they are incorporating information from all three tracking detectors: the VELO, the UT, and SciFi.
Kai: So, when you look at the results section of this paper, they show a factor of two improvement in the invariant mass resolutions for reconstructed D zero and J psi hadrons compared to what they got with their previous fitter.
Mira: That is a big deal because those mass resolutions directly tell us how well we can measure the properties of the heavy particles we’re looking for, like those mesons.
Lev: And it also speaks to robustness; they found that this parameterised Kalman filter shows greater robustness against detector misalignment effects, which is something that often degrades momentum estimates when you rely only on pattern recognition.
Kai: So, if the magnetic field map shifts slightly in the detector geometry, their estimates for particle momentum don't fall apart as much as they used to. It makes the reconstruction more stable across different conditions.
Mira: And they also mentioned substantially sharpening the discrimination between genuine particle trajectories and accidental background events in that first trigger stage, which means they are rejecting fake tracks much better than before.
Lev: The paper quantifies this rejection improvement by saying it doubles the rejection of those accidental background events at a standard selection working point, and this happens without sacrificing genuine track efficiency.
Title and authors: Kai: That’s a crucial detail for me because efficiency is always a trade-off in these kinds of systems, so having that improvement happen while keeping the real track acceptance high is important.
Mira: It really shows that you can gain better rejection and better resolution simultaneously just by changing how you model the particle movement.
Lev: And they did this by choosing a simple function to parameterize the extrapolation instead of solving the full equations of motion numerically, which is what made it fast enough for real-time hardware.
Kai: So, in short, they took a complex mathematical framework and simplified the physics assumptions just enough to make it work reliably at the full collision rate with good performance gains.
Mira: It points toward this idea that you can achieve offline quality reconstruction right there in the trigger system before you even get to the offline analysis stage.
Lev: And they conclude by saying that this deployment demonstrates a real potential for advanced reconstruction algorithms running on GPUs, which is a step toward getting that high-quality work done at the full collision frequency.
Kai: So we’re looking at this paper, "Track fitting at the full LHC collision rate," which shows how they implemented a fast Kalman filter for track fitting in the first trigger stage since two thousand twenty-five <ref:2607.14793#pg1,Track fitting at the full LHC collision rate>.
Mira: It’s about replacing slow numerical methods with fast analytical parameterizations to get momentum estimates at thirty MHz, leading to a factor of two improvement in mass resolutions for D zero and J psi hadrons <ref:2607.14793#pg1>.
Lev: The key improvements are better momentum resolution, increased robustness against detector misalignment, and substantially sharper discrimination against background events.
Kai: It really shows how much performance can be gained by focusing on parameterizing the extrapolation instead of running heavy numerical integration.
Mira: And the implication for me is that this sets a new benchmark for what we can expect from real-time reconstruction algorithms in high energy physics experiments.
Lev: From my side, it confirms that if you can reduce the computational cost to fit within real-time constraints while maintaining precision, you open up much more possibilities for running complex fits on actual hardware.
Kai: So, to wrap up this discussion on "Track fitting at the full LHC collision rate," we see a fast implementation of a Kalman filter that gets us better momentum resolution and rejection against fake tracks without sacrificing efficiency.
Mira: It’s a solid step toward achieving offline quality reconstruction right in the trigger system, which should boost the statistical precision of future measurements in Run three and beyond <ref:2607.14793#pg1>.
Lev: And for real hardware running, it shows that this level of optimization is achievable, paving the way for even more complex fitting routines on GPUs.
The paper's summary: Kai: So, this paper is about taking track fitting to the full LHC collision rate, which means they’re finally building something that can handle all that data coming in at once.
Mira: It boils down to replacing older tracking methods with a new Kalman filter setup for the first trigger stage since two thousand twenty-five.
Kai: Right, so it’s this fast implementation of a Kalman filter designed to estimate particle paths at thirty mega hertz collision frequency. It’s moving away from partial fitting to full track fitting in real time.
Lev: From my side, that means they’re replacing the old partial fit algorithm entirely. That old thing was limited because it assumed a magnetic field map was perfect and only looked at the VELO segment.
Mira: And this new parameterised version uses the full track information from all three detectors: VELO, UT, and SciFi. They use simple functions to describe how a particle moves between layers instead of solving complex equations every time.
Kai: That’s what they’re doing to keep it fast for the hardware. They replace expensive magnetic field map lookups and numerical integration with these analytical parameterizations while keeping the core Kalman filtering math intact.
Lev: And they claim this change only increases processing time by about two percent, which is pretty good for a real-time trigger system. It shows they managed to keep the overhead low enough for GPU architectures too.
Mira: The main payoff is the momentum resolution. They see a factor of two improvement in the invariant mass resolutions for things like D zero and J psi hadrons compared to what was achieved with the older VELO Only Kalman Filter.
Kai: A factor of two improvement in mass resolution means we can measure those particles much more precisely. That’s significant for how we understand their properties.
Lev: And they also found it’s more robust against detector misalignment effects, which is a big deal because that stuff messes up momentum estimates when you only rely on pattern recognition algorithms.
Mira: Plus, it substantially sharpens the discrimination between real particle trajectories and accidental background events. They doubled the rejection of those fake tracks at a standard selection working point without sacrificing genuine track efficiency.
Kai: So, they’ve got speed, better resolution on mass, and better rejection against noise all in one package for the trigger stage. It’s a lot to fit into that real-time environment.
Lev: The caveat is that this is just for the first trigger stage; they aren't claiming it solves everything offline. They are building a solid foundation for something better down the line, like adding outlier removal or state estimates for particle identification later on.
Mira: It’s a step toward getting an offline-quality reconstruction right there in the trigger system, which is what we’re really aiming for in this kind of high energy physics computing.
Kai: It means more effective event selection at the trigger level, which boosts efficiency and ultimately enhances the statistical precision of LHCb measurements for Run three and beyond.
Lev: So they’ve shown that you can get a lot of performance gain by simplifying the physics model just enough to make it run on hardware efficiently.
Mira: And this sets a benchmark for how we can expect advanced reconstruction algorithms to function at the full collision frequency in experiments like this one.
The paper's improvements: Kai: We’ve looked at how they built this fast Kalman filter for the LHCb trigger, so now we need to talk about what they suggest as improvements to this whole system.
Mira: The paper points out that while the parameterised version is much faster, there are still some areas where they think things could get even better.
Lev: They mention that for a real hardware setup running at the full collision rate, you still need to worry about how those analytical functions handle extreme detector misalignment.
Kai: So they suggest adding a layer of outlier removal right into this process, something that can clean up those noisy estimates before they get used for event selection.
Mira: Right, and they also talk about the need to add state estimates for particle identification using the RICH detectors in the future. That’s a clear roadmap for what comes next after this initial trigger stage is set up.
Lev: From an engineering standpoint, if you want to get closer to that offline quality reconstruction at the full collision frequency, you can't just stop at parameterization; you have to figure out how to integrate those future identification steps into the existing pipeline.
Kai: It sounds like they’re proposing a staged upgrade path then. First, this fast trigger filter gets us running reliably on all three detectors and improving mass resolution significantly.
Mira: And then, in the next iteration of this system, you layer on those more complex physics steps like particle identification to really sharpen the track quality metric.
Lev: The assumption here is that because they’ve mastered the speed and robustness of this Kalman filter implementation now, adding those more complex components won't break the real-time requirements too badly.
Kai: So they’re suggesting that this success opens the door to moving toward a full offline quality reconstruction even at high collision rates, assuming we can manage those future computational demands.
Mira: It really frames this paper as the critical first step in achieving that ultimate goal of getting offline-quality reconstruction done on GPUs.
Lev: The implication for hardware is that this shows the path: get the fast, robust core running now and then tackle the more computationally intensive additions like outlier removal later.
Kai: It’s a clear message to hardware developers that optimizing those analytical parameterizations is a viable way to push reconstruction quality right into the real-time environment.
Conclusion: Tom: So we’re wrapping up this look at the paper, "Track fitting at the full LHC collision rate," which focused on that fast Kalman filter implementation for LHCb data taking.
Kai: To recap, they managed to get a fast track fitting algorithm running at the full collision frequency of thirty mega hertz using analytical parameterizations instead of slow numerical integration.
Mira: And the main result is a factor of two improvement in invariant mass resolutions for D zero and J psi hadrons because they’re using all three detectors.
Lev: I just want to say that even with this speed boost, you still have to deal with the fact that real hardware needs robust methods for detector misalignment.
Kai: True, and the authors did acknowledge that this is a first step; they aren't claiming it solves everything offline yet, but it’s a massive improvement for what’s possible in the trigger system right now.
Mira: Exactly, so we’ve seen how you can use this technique to get better track quality metrics and background rejection without sacrificing efficiency at the working point.
Lev: For me, it shows that if you can keep the overhead low like they did with that two percent increase in processing time, then pushing more complex physics models down into the trigger level becomes much more feasible on actual hardware.
Kai: It sets a clear direction for what we need to build next—that future particle identification step they mentioned is definitely the next big thing after this foundational work.
Mira: We’re seeing a move toward getting offline quality reconstruction right there in the trigger system, which really boosts the statistical precision of LHCb measurements moving into Run three and beyond.
Lev: It confirms that optimizing that analytical parameterization approach is a solid way to push computational limits on real-time systems.
Kai: So this paper, "Track fitting at the full LHC collision rate," shows we can get significant gains in resolution and background rejection by changing how we model particle movement for real-time use.
Mira: It’s a strong demonstration of how theoretical insights about Kalman filtering can be translated into practical hardware performance.
Lev: And it opens up the door for more complex fitting routines to run on real hardware, provided you handle those detector shifts carefully.
Kai: Next week, we’ll be looking at some work on GPU-based kalman filters that take this idea even further and try to push the limits of what's possible in terms of processing power.
Laboratoire de Physique Nucléaire et de Hautes Energies (LPNHE) · Department of Physics, Indiana University · Department of Physics & Astronomy, University of Manchester · Physikalisches Institut, Ruprecht-Karls-Universität Heidelberg
hep-ex, physics.ins-det
Submitted: 2026-07-16
Updated: 2026-11-11
Comments: 16 pages, 10 pretty figures
Journal ref: Eur. Phys. J. C 86, 1247 (2026)
DOI: 10.1140/epjc/s10052-026-16478-2
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 93/100
The gist: The LHCb experiment at the Large Hadron Collider underwent a major upgrade before the LHC Run 3 data taking period…
Key concepts
- Kalman Filter
- A mathematical tool used to estimate the state of a system by combining predictions with actual measurements. In this context, it is adapted to track charged particles through magnetic fields, helping determine their precise trajectories in real-time.
- Parameterised Kalman Filter
- A simplified version of the Kalman filter tailored for high-energy physics. It uses pre-calculated functions instead of complex numerical solving and magnetic field map lookups, allowing it to run very fast on hardware while maintaining the core mathematical benefits of filtering.
- Momentum Resolution
- This refers to how accurately the experiment can measure the speed (momentum) of a particle. The new filter provides a significant improvement, meaning the reconstructed mass peaks for particles like D0 and J/psi are much narrower and sharper than before.
- Real-Time Trigger (HLT1)
- This is the first stage of data processing that decides which events are important enough to keep. The new algorithm allows for full track fitting at the very high collision rate of 30 MHz, enabling better event selection before data is stored.
Terminology
Summary
The LHCb experiment at the Large Hadron Collider underwent a major upgrade before the LHC Run 3 data taking period, employing an all-software approach in its trigger system Here we present a fast implementation of a Kalman filter, used in the first trigger stage since the 2025 data taking period, allowing to determine parameter estimates of charged-particle trajectories at the full LHCb collision frequency of 30 MHz This approach replaces computationally expensive magnetic field map lookups and numerical integration methods with fast analytical parameterisations while maintaining the mathematical framework of Kalman filtering Compared to the previous fitter this parameterised Kalman filter shows a significantly improved momentum resolution, resulting in a factor of two improvement in the invariant mass resolutions for reconstructed D0 and J/ψ hadrons It additionally demonstrates greater robustness against detector misalignment effects and substantially sharpens the discrimination between genuine particle trajectories and accidental background, more than doubling the rejection of the latter at no cost to genuine-track efficiency, for a standard selection working point. p. 0 Compared to the previous fitter this
parameterised Kalman filter shows a significantly improved momentum resolution,
resulting in a factor of two improvement in the invariant mass resolutions for
reconstructed D0 and J/ψ hadrons. It additionally demonstrates greater robustness
against detector misalignment effects and substantially sharpens the discrimination
between genuine particle trajectories and accidental background, more than doubling
the rejection of the latter at no cost to genuine-track efficiency, for a standard
selection working point.Where the paper says it
Algorithm Development
The core of this work is a fast implementation of a Kalman filter used in the first trigger stage since the 2025 data taking period This algorithm replaces the previous partial track fitting algorithm in the real-time trigger environment at the cost of a 2% increase in processing time. The Parameterised Kalman Filter is a Kalman filter as defined in the previous section, but uses simplifications specific to typical implementations in a high-energy physics context which allows it to be executed with minimal time consumption. p. 5 The implementation discussed in the following however is the
first full Kalman filter that was used in a production environment of an LHC experiment
at the full collision rate. The Parameterised Kalman Filter is a Kalman filter as defined in the previous
section, but uses simplifications specific to typical implementations in a high-energy
physics context which allows it to be executed with minimal time consumption. The Parameterised Kalman Filter is a Kalman filter as defined in the previous
section, but uses simplifications specific to typical implementations in a high-energy
physics context which allows it to be executed with minimal time consumption. It was
developed as a drop-in replacement for the previous track fitting algorithm, the VELO
Only Kalman Filter, to provide better momentum (and therefore mass) resolution, more
robustness against detector misalignments, and improved rejection against fake tracks
(so-called ghosts) in the HLT1 stage of LHCb.Where the paper says it
Where the paper says it
Implementation Details
The Parameterised Kalman Filter extends the track fitting to all three tracking detectors: the VELO, UT and SciFi. The standard implementation of such an algorithm includes frequent look-ups of a magnetic field map and detector geometry, as well as intensive calculations to predict the movement of a charged particle in the magnetic field by numerically solving the equations of motion. p. 5 First
described as a fast alternative to a full Kalman filter [23], and prototyped on CPUs, this
approach has proven to be well suited for the GPU-based HLT1 and provides improved
state estimates while still fulfilling the real-time requirements. As a Kalman filter performs extrapolations of particle trajectories between layers, the
standard implementation of such an algorithm includes frequent look-ups of a magnetic
field map and detector geometry, as well as intensive calculations to predict the movement
of a charged particle in the magnetic field by numerically solving the equations of motion. Almost all throughput improvements in
the Parameterised Kalman Filter, compared to a standard implementation as described
above, are achieved in changes to the state prediction step. Instead of extrapolating
the state using numerical methods to solve the propagation of a charged particle in a
magnetic field, a simple function is chosen to parameterise this extrapolation.Where the paper says it
Where the paper says it
Performance Comparison
The performance of the Parameterised Kalman Filter is compared against a standard implementation as described above, which accounts for approximately 14% of the total processing time in HLT2 reconstruction. The Parameterised Kalman Filter makes up roughly 3. p. 8 The Parameterised Kalman Filter performs all calculation in
single floating point precision by default in order to maximize throughput. This makes full
use of the available processor cores optimised for single precision and reduces the memory
resources allocated to a single track and thus increases opportunities for parallelisation. The cost of such an implementation is substantial, in the CPU-based HLT2 reconstruction,
where a full Kalman filter is employed, the track fit accounts for approximately 14% of
the total processing time. This approach is not suited for HLT1 as the field map and
geometry exceed the GPU’s limited fast memory, while the position-dependent lookups and
data-dependent integration steps produce irregular, non-coalesced memory accesses and
control-flow divergence across threads in a warp. This enables having tracks
from different events processed in the same kernel block to fully utilise the available
threads. As a direct consequence the runtime of the algorithm scales linearly with the
number of tracks in an event. First
described as a fast alternative to a full Kalman filter [23], and prototyped on CPUs, this
approach has proven to be well suited for the GPU-based HLT1 and provides improved
state estimates while still fulfilling the real-time requirements. As a Kalman filter performs extrapolations of particle trajectories between layers, the
standard implementation of such an algorithm includes frequent look-ups of a magnetic
field map and detector geometry, as well as intensive calculations to predict the movement
of a charged particle in the magnetic field by numerically solving the equations of motion.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Key Results
The primary benefit of fitting the full track comes from incorporating hits upstream and downstream of the LHCb magnet, which enables precise measurements of particle momenta. The resulting momentum resolution is much better than that achieved by the pattern recognition algorithms (using a parametric approach) and used by the VELO Only Kalman Filter. p. 5 The Parameterised Kalman Filter is a Kalman filter as defined in the previous
section, but uses simplifications specific to typical implementations in a high-energy
physics context which allows it to be executed with minimal time consumption. It was
developed as a drop-in replacement for the previous track fitting algorithm, the VELO
Only Kalman Filter, to provide better momentum (and therefore mass) resolution, more
robustness against detector misalignments, and improved rejection against fake tracks
(so-called ghosts) in the HLT1 stage of LHCb. Implemented
on approximately 500 GPUs within the first-level trigger, the algorithm has replaced
the previous partial track fitting algorithm in the real-time trigger environment at
the cost of a 2% increase in processing time. Compared to the previous fitter this
parameterised Kalman filter shows a significantly improved momentum resolution,
resulting in a factor of two improvement in the invariant mass resolutions for
reconstructed D0 and J/ψ hadrons. Compared to the previous fitter this
parameterised Kalman filter shows a significantly improved momentum resolution,
resulting in a factor of two improvement in the invariant mass resolutions for
reconstructed D0 and J/ψ hadrons. It additionally demonstrates greater robustness
against detector misalignment effects and substantially sharpens the discrimination
between genuine particle trajectories and accidental background, more than doubling
the rejection of the latter at no cost to genuine-track efficiency, for a standard
selection working point. Compared to the previous fitter this
parameterised Kalman filter shows a significantly improved momentum resolution,
resulting in a factor of two improvement in the invariant mass resolutions for
reconstructed D0 and J/ψ hadrons. It additionally demonstrates greater robustness
against detector misalignment effects and substantially sharpens the discrimination
between genuine particle trajectories and accidental background, more than doubling
the rejection of the latter at no cost to genuine-track efficiency, for a standard
selection working point.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Conclusion
The successful deployment of the Parameterised Kalman Filter in LHCb’s production trigger system demonstrates the potential for advanced reconstruction algorithms in real-time high-energy physics computing and is another step towards an offline-quality reconstruction on GPUs. The improved track parameter estimates enable more effective event selection in the trigger system, increasing the efficiency for recording events of physics interest and thereby enhancing the statistical precision of LHCb’s measurements in Run 3 and beyond. p. 0 Uecker4
1Laboratoire de Physique Nucl´eaire et de Hautes Energies (LPNHE), Sorbonne Universit´e, CNRS/IN2P3,
Paris, France
2Department of Physics, Indiana University, Bloomington, USA
3Department of Physics & Astronomy, University of Manchester, United Kingdom
4Physikalisches Institut, Ruprecht-Karls-Universit¨at Heidelberg, Heidelberg, Germany
Abstract
The LHCb experiment at the Large Hadron Collider underwent a major upgrade
before the LHC Run 3 data taking period, employing an all-software approach in its
trigger system. Here we present a fast implementation of a Kalman filter, used in the
first trigger stage since the 2025 data taking period, allowing to determine parameter
estimates of charged-particle trajectories at the full LHCb collision frequency of
30 MHz. The dipole magnet creates a magnetic field, dominantly
in the y direction, with a peak field strength of about 1 T and an integrated bending power
of 4 Tm. In addition, LHCb also operates two Ring Imaging Cherenkov (RICH) detectors,
RICH1 and RICH2. Uecker4
1Laboratoire de Physique Nucl´eaire et de Hautes Energies (LPNHE), Sorbonne Universit´e, CNRS/IN2P3,
Paris, France
2Department of Physics, Indiana University, Bloomington, USA
3Department of Physics & Astronomy, University of Manchester, United Kingdom
4Physikalisches Institut, Ruprecht-Karls-Universit¨at Heidelberg, Heidelberg, Germany
Abstract
The LHCb experiment at the Large Hadron Collider underwent a major upgrade
before the LHC Run 3 data taking period, employing an all-software approach in its
trigger system. Here we present a fast implementation of a Kalman filter, used in the
first trigger stage since the 2025 data taking period, allowing to determine parameter
estimates of charged-particle trajectories at the full LHCb collision frequency of
30 MHz.Where the paper says it
Where the paper says it
Where the paper says it
Acknowledgments
We thank LHCb’s Real-Time Analysis project for the support, for many useful discussions, and for reviewing an early version of this manuscript. We also thank the LHCb Computing, Simulation and Online projects for producing the simulated samples used to test and benchmark the performance of the parameterised Kalman filter and to evaluate the HLT1 throughput. p. 2 In the context of this work, simulation is used to determine the parameters of the
Parameterised Kalman Filter and to evaluate its performance. In the simulation,
pp collisions are generated using Pythia [13] with a specific LHCb configuration [14]. In the context of this work, simulation is used to determine the parameters of the
Parameterised Kalman Filter and to evaluate its performance. In the simulation,
pp collisions are generated using Pythia [13] with a specific LHCb configuration [14].Where the paper says it
Where the paper says it
References
[1] LHCb collaboration, A. A. Alves Jr. et al., The LHCb detector at the LHC, JINST 3 (2008) S08005. p. 1 The upgraded LHCb detector [2] enables data collection at an instantaneous luminosity
of L = 2 × 1033 cm−2
s
−1
, a five-fold increase compared to the instantaneous luminosity
during the previous data taking period from 2015 to 2018. For the upgrade, significant
parts of the detector have been replaced or modified to be read out at the nominal
LHC bunch-crossing frequency of 40 MHz: The tracking system has been fully replaced
and consists of a silicon-pixel vertex detector (VELO) surrounding the pp interaction
region [5], a large-area silicon-strip detector (UT) [6], located upstream of a dipole
magnet with a bending power of about 4 Tm, and three stations of scintillating fibre
detectors (SciFi) placed downstream of the magnet [6].Where the paper says it
[2] LHCb collaboration, R. p. 0 EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH (CERN)
CERN-LHCb-RD-2026-003
July 29, 2026
Track fitting at the full LHC collision
rate
P. Billoir1
, T. This approach has enabled LHCb to record an integrated luminosity larger
than 25 fb−1
at √
s = 13. 6 TeV in 2024 and 2025, more than doubling the dataset collected
during Run 1 and Run 2 of the LHC.Where the paper says it
Where the paper says it
[3] LHCb collaboration, R. p. 0 EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH (CERN)
CERN-LHCb-RD-2026-003
July 29, 2026
Track fitting at the full LHC collision
rate
P. Billoir1
, T. Section 5 describes the implementation on GPUs, and
Sect. 6 analyses its performance in a simulated LHCb detector. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
Where the paper says it
[4] LHCb collaboration, LHCb SMOG Upgrade, CERN-LHCC-2019-005, 2019. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[5] LHCb collaboration, LHCb VELO Upgrade Technical Design Report, CERN-LHCC2013-021, 2013. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[6] LHCb collaboration, LHCb Tracker Upgrade Technical Design Report, CERN-LHCC2014-001, 2014. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[7] M. Adinolfi et al., Performance of the LHCb RICH detector at the LHC, Eur. p. 1 For the upgrade, significant
parts of the detector have been replaced or modified to be read out at the nominal
LHC bunch-crossing frequency of 40 MHz: The tracking system has been fully replaced
and consists of a silicon-pixel vertex detector (VELO) surrounding the pp interaction
region [5], a large-area silicon-strip detector (UT) [6], located upstream of a dipole
magnet with a bending power of about 4 Tm, and three stations of scintillating fibre
detectors (SciFi) placed downstream of the magnet [6]. Charged hadrons are identified
using information from two ring-imaging Cherenkov (RICH) detectors [7, 8]. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[8] LHCb collaboration, LHCb PID Upgrade Technical Design Report, CERN-LHCC2013-022, 2013. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[9] LHCb collaboration, LHCb Upgrade GPU High Level Trigger Technical Design Report, CERN-LHCC-2020-006, 2020. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[10] T. Evans, C. Fitzpatrick, and J. Horswill, An automated bandwidth division for the LHCb upgrade trigger, Comput. p. 0 De Cian3
, L. Uecker4
1Laboratoire de Physique Nucl´eaire et de Hautes Energies (LPNHE), Sorbonne Universit´e, CNRS/IN2P3,
Paris, France
2Department of Physics, Indiana University, Bloomington, USA
3Department of Physics & Astronomy, University of Manchester, United Kingdom
4Physikalisches Institut, Ruprecht-Karls-Universit¨at Heidelberg, Heidelberg, Germany
Abstract
The LHCb experiment at the Large Hadron Collider underwent a major upgrade
before the LHC Run 3 data taking period, employing an all-software approach in its
trigger system. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[11] LHCb collaboration, R. p. 0 EUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH (CERN)
CERN-LHCb-RD-2026-003
July 29, 2026
Track fitting at the full LHC collision
rate
P. Billoir1
, T. De Cian3
, L. Uecker4
1Laboratoire de Physique Nucl´eaire et de Hautes Energies (LPNHE), Sorbonne Universit´e, CNRS/IN2P3,
Paris, France
2Department of Physics, Indiana University, Bloomington, USA
3Department of Physics & Astronomy, University of Manchester, United Kingdom
4Physikalisches Institut, Ruprecht-Karls-Universit¨at Heidelberg, Heidelberg, Germany
Abstract
The LHCb experiment at the Large Hadron Collider underwent a major upgrade
before the LHC Run 3 data taking period, employing an all-software approach in its
trigger system. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
Where the paper says it
[12] R. Aaij et al., Allen: A high level trigger on GPUs for LHCb, Comput. p. 0 This approach replaces computationally expensive magnetic field map
lookups and numerical integration methods with fast analytical parameterisations
while maintaining the mathematical framework of Kalman filtering. Implemented
on approximately 500 GPUs within the first-level trigger, the algorithm has replaced
the previous partial track fitting algorithm in the real-time trigger environment at
the cost of a 2% increase in processing time. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[13] T. Sj¨ostrand, S. Mrenna, and P. Skands, A brief introduction to PYTHIA 8. p. 2 the reconstruction sequence of charged particles, the Parameterised Kalman Filter,
performing a fit using information from all involved tracking detectors. This paper is organised as follows: After a brief introduction of the LHCb detector and
its simulation, the principles of a Kalman Filter are reviewed in Sect. Submitted to European Physics Journal C
CC BY 4. arXiv:2607. In the context of this work, simulation is used to determine the parameters of the
Parameterised Kalman Filter and to evaluate its performance. In the simulation,
pp collisions are generated using Pythia [13] with a specific LHCb configuration [14]. arXiv:2607. 14793v2 [hep-ex] 27 Jul 2026Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[14] I. Belyaev et al., Handling of the generation of primary events in Gauss, the LHCb simulation framework, J. p. 0 Here we present a fast implementation of a Kalman filter, used in the
first trigger stage since the 2025 data taking period, allowing to determine parameter
estimates of charged-particle trajectories at the full LHCb collision frequency of
30 MHz. This approach replaces computationally expensive magnetic field map
lookups and numerical integration methods with fast analytical parameterisations
while maintaining the mathematical framework of Kalman filtering.Where the paper says it
[15] D. J. Lange, The EvtGen particle decay simulation package, Nucl. p. 2 In the simulation,
pp collisions are generated using Pythia [13] with a specific LHCb configuration [14]. Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16].Where the paper says it
[16] N. Davidson, T. Przedzinski, and Z. Was, PHOTOS interface in C++: Technical and physics documentation, Comput. p. 2 In the simulation,
pp collisions are generated using Pythia [13] with a specific LHCb configuration [14]. Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16]. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[17] Geant4 collaboration, J. p. 2 Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16]. The interaction of the generated particles with the
detector, and its response, are implemented using the Geant4 toolkit [17] as described in
Ref. Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16]. The interaction of the generated particles with the
detector, and its response, are implemented using the Geant4 toolkit [17] as described in
Ref. Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16]. The interaction of the generated particles with the
detector, and its response, are implemented using the Geant4 toolkit [17] as described in
Ref. Decays of unstable particles are described by EvtGen [15], in which final-state radiation
is generated using Photos [16]. The interaction of the generated particles with the
detector, and its response, are implemented using the Geant4 toolkit [17] as described in
Ref.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[18] M. Clemencic et al., The LHCb simulation application, Gauss: Design, evolution and experience, J. p. 2 the reconstruction sequence of charged particles, the Parameterised Kalman Filter,
performing a fit using information from all involved tracking detectors. This paper is organised as follows: After a brief introduction of the LHCb detector and
its simulation, the principles of a Kalman Filter are reviewed in Sect.Where the paper says it
[19] R. E. Kalman, A new approach to linear filtering and prediction problems, Transactions of the ASME–Journal of Basic Engineering 82 (1960) 35.
[20] W. D. Hulsbergen, Decay chain fitting with a Kalman filter, Nucl. p. 2 The studies presented here use simulated events of minimum bias pp-collisions,
events containing a D∗
(2010)+ → (D0 → K−π
+)π
+ decay and events containing a
B0
s → (J/ψ → µ
+µ
−)(ϕ → K+K−) decay, together representing a large part of the LHCb
physics programme in terms of kinematics and topology. 3 Track Fitting with a Kalman Filter
Efficiently selecting collision events containing b or c hadron decays requires precise
estimates of track parameters, including track momentum and direction. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[21] X. Ai et al., A gpu-based kalman filter for track fitting, Computing and Software for Big Science 5 (2021) 20. p. 5 4 The Parameterised Kalman Filter
Owing to the parallel nature of track fitting and the parallelisation capabilities of modern
computing architectures, several studies have been performed with Kalman-based track
fitting, see e. The implementation discussed in the following however is the
first full Kalman filter that was used in a production environment of an LHC experiment
at the full collision rate.Where the paper says it
[22] Cerati, Giuseppe et al., Parallelized kalman-filter-based reconstruction of particle tracks on many-core processors and gpus, EPJ Web Conf. p. 5 As a Kalman filter performs extrapolations of particle trajectories between layers, the
standard implementation of such an algorithm includes frequent look-ups of a magnetic
field map and detector geometry, as well as intensive calculations to predict the movement
of a charged particle in the magnetic field by numerically solving the equations of motion. The cost of such an implementation is substantial, in the CPU-based HLT2 reconstruction,
where a full Kalman filter is employed, the track fit accounts for approximately 14% of
the total processing time.Where the paper says it
[23] P. Billoir, M. De Cian, P. A. G¨unther, and S. Stemmle, A parametrized Kalman filter for fast track fitting at LHCb, Comput. p. 5 The Parameterised Kalman Filter
extends the track fitting to all three tracking detectors: the VELO, UT and SciFi. First
described as a fast alternative to a full Kalman filter [23], and prototyped on CPUs, this
approach has proven to be well suited for the GPU-based HLT1 and provides improved
state estimates while still fulfilling the real-time requirements. Submitted to European Physics Journal C
CC BY 4. arXiv:2607.Where the paper says it
Where the paper says it
[24] M. Losasso et al., Tests and Field Map of LHCb Dipole Magnet, Applied Superconductivity, IEEE Transactions on 16 (2006) 1700. p. 2 The stations
are arrayed such that 99% of all tracks from the region around the interaction point and
within the nominal LHCb acceptance cross at least four VELO stations. The UT, a four
layer silicon-strip detector, is situated downstream of the VELO in front of the large dipole
magnet and is used to provide space point measurements before the volume with a large
magnetic field.Where the paper says it
[25] LHCb collaboration, HLT1 RICH decoding algorithm performance, LHCB-FIGURE2024-027, 2024. p. 1 Accurate determination of track
parameters using a track fit is important for the HLT1 selection performance. Up to
the 2024 data-taking campaign, the HLT1 track fitting algorithm performed a simplified
fitting procedure by applying a Kalman filter to only the VELO segment of each track.Where the paper says it
[26] LHCb collaboration, LHCb Framework TDR for the LHCb Upgrade II Opportunities in flavour physics, and beyond, in the HL-LHC era, CERN-LHCC-2021-012, 2022. p. 1 1 Introduction
The LHCb experiment [1–3] is one of the four big experiments at the Large Hadron Collider
(LHC) at CERN. The original experiment and its upgrade are designed as general purpose
detectors covering the forward direction.Where the paper says it
[8] Acknowledgments
We thank LHCb’s Real-Time Analysis project for the support, for many useful discussions, and for reviewing an early version of this manuscript. We also thank the LHCb Computing, Simulation and Online projects for producing the simulated samples used to test and benchmark the performance of the parameterised Kalman filter and to evaluate the HLT1 throughput. M. De Cian acknowledges support from STFC, L. H. Uecker acknowledges support from BMFTR.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Faster track fitting in real-time trigger: The Parameterised Kalman Filter replaces computationally expensive magnetic field map lookups and numerical integration methods with fast analytical parameterisations while maintaining the mathematical framework of Kalman filtering. This allows for determination of parameter estimates at the full LHCb collision frequency of 30 MHz, replacing the previous partial track fitting algorithm in the real-time trigger environment.
-
Improved momentum resolution: The new approach shows a
significantly improved momentum resolution, resulting in a factor of two improvement in the invariant mass resolutions for reconstructed D0 and J/psi hadrons.
This is achieved by using the full track information from all involved tracking detectors (VELO, UT, and SciFi). -
Enhanced background rejection: The algorithm demonstrates "substantially sharpens the discrimination between genuine particle trajectories and accidental background, more than doubling the rejection of the latter at no cost to genuine-track efficiency, for a standard selection working point." This is evidenced by improved reduced chi squared distributions (lower ghost efficiency) compared to the VELO Only Kalman Filter.
-
Robustness against detector misalignment: The Parameterised Kalman Filter
demonstrates greater robustness against detector misalignment effects.
This means that momentum estimates remain accurate even when the geometry of the detectors shifts randomly in simulation, which is a significant advantage over methods relying solely on pattern recognition algorithms. -
GPU-optimized throughput: The implementation is designed for GPU architectures by avoiding costly numerical solving and using parameterisations instead of unique access to magnetic field maps. This allows the algorithm to be executed on approximately 500 GPUs within the first-level trigger while only increasing processing time by about 2% compared to the previous fitter.
-
Real-time hardware replacement: The algorithm enables
full track fitting at the full LHC collision rate
in HLT1, which was previously limited by simplified fitting procedures that only applied to the VELO segment under a vanishing magnetic field assumption. This allows for a transition from simplified fits to a full reconstruction sequence in the real-time trigger environment.
Abstract
The LHCb experiment at the Large Hadron Collider underwent a major upgrade before the LHC Run 3 data-taking period, employing an all-software approach in its trigger system. Here we present a fast implementation of a Kalman filter, used in the first trigger stage since the 2025 data-taking period, allowing to determine parameter estimates of charged-particle trajectories at the full LHCb collision frequency of 30 MHz. This approach replaces computationally expensive magnetic field map lookups and numerical integration methods with fast analytical parameterisations while maintaining the mathematical framework of Kalman filtering. Implemented on approximately 500 GPUs within the first-level trigger, the algorithm has replaced the previous partial track fitting algorithm in the real-time trigger environment at the cost of a 2% increase in processing time. Compared to the previous fitter this parameterised Kalman filter shows a significantly improved momentum resolution, resulting in a factor of two improvement in the invariant mass resolutions for reconstructed D0 and J/ψ hadrons. It additionally demonstrates greater robustness against detector misalignment effects and substantially sharpens the discrimination between genuine particle trajectories and accidental background, more than doubling the rejection of the latter at no cost to genuine-track efficiency, for a standard selection working point.
Sources
- The LHCb upgrade I
- LHCb Detector Performance
- Performance of the LHCb RICH detector at the LHC
- An Automated Bandwidth Division for the LHCb Upgrade Trigger
- A Comparison of CPU and GPU implementations for the LHCb Experiment Run 3 Trigger
- Allen: A high level trigger on GPUs for LHCb
- A Brief Introduction to PYTHIA 8.1
- PYTHIA 6.4 Physics and Manual
- PHOTOS Interface in C++; Technical and Physics Documentation
- Decay Chain Fitting with a Kalman Filter
- A parametrized Kalman filter for fast track fitting at LHCb
Related papers
- DrSai: an agentic AI system toward autonomous physics analysis workflows at BESIII
- CANTON- mu Proposal: A Next-Generation Muon g-2 Measurement at Sub-0.1 ppm Precision
- Search for dark photon dark matter near 1,meV using a superconducting mixer
- ANUBIS detector and its sensitivity to neutral long-lived particles
- How to scale your HEP ML models: A recipe for robust architecture comparisons at scale
- Search for the 16 O(ppp) to 13 C pi+ pi+ e+ Decay Mode in Super-Kamiokande Using Machine Learning Techniques