- Indico style
- Indico style - inline minutes
- Indico style - numbered
- Indico style - numbered + minutes
- Indico Weeks View
BOOST 2023 is the 15th conference of a series of successful joint theory/experiment workshops that bring together the world's leading experts from theory and LHC/RHIC experiments to discuss the latest progress and develop new approaches on the reconstruction of and use of jet substructure to study Quantum Chromodynamics (QCD) and look for physics beyond the Standard Model.
The conference will cover the following topics:
Previous editions:
N.B. Berkeley et al. was originally scheduled to host in 2021, but this was postponed to 2022 and then again to 2023 because of Covid19.
Note: All talks will be in the building 50 auditorium. Wednesday afternoon is open for participants to collaborate and/or explore the SF Bay Area! We will end at lunchtime on Friday.
Every bunch crossing at the LHC causes not just one proton-proton interaction, but several. These additional collisions are called "pileup". With the increasing luminosity of the LHC also the number of pileup interactions per bunch crossing increased in the past years and it will reach up to 140 during high-luminosity LHC operation. Removing the pileup from an event is essential, because it does not only affect the jet energy but also other event observables as for example the missing transverse energy, the jet substructure, jet counting and the lepton isolation. In addition, jets as an experimental signature of energetic quarks and gluons produced in high energy processes, need to be calibrated in order to have the correct energy scale. A detailed understanding of both the energy scale and the transverse momentum resolution of jets at the CMS is of crucial importance for many physics analyses. In this talk we present recent developments in terms of jet energy scale and resolution, substructure techniques and pileup mitigation techniques.
Experimental uncertainties related to hadronic object reconstruction can limit the precision of physics analyses at the LHC, and so improvements in performance have the potential to broadly increase the impact of results. Recent refinements to reconstruction and calibration procedures for ATLAS jets and MET result in reduced uncertainties, improved pileup stability and other performance gains. In this contribution, highlights of these developments will be presented.
We will overview the usage of boosted jet taggers within CMS. We will also discuss how their performance is calibrated for usage in analyses. This includes a new Lund Plane reweighting technique that can calibrate the performance of high-prong jets for which there is no standard model equivalent which can be used as a proxy.
Hadronic object reconstruction is one of the most promising settings for cutting-edge machine learning and artificial intelligence algorithms at the LHC. In this contribution, highlights of ML/AI applications by ATLAS to particle and boosted-object identification, MET reconstruction and other tasks will be presented.
Physics measurements in the highly Lorentz-boosted regime are a critical part of the LHC physics program. In the CMS Collaboration, various boosted-jet tagging algorithms, designed to identify hadronic jets originating from the decay of a massive particle to bb̅ or cc̅, have been developed and deployed in a variety of analyses. This talk summarises the performance of these algorithms with Run-2 data collected with √s = 13 TeV. Three control regions are studied, including jets from g→bb̅ (cc̅) decays or Z decays. The algorithms are calibrated using a combination of measurements in these three regions.
Jet substructure has been successful in broadening our understanding of fundamental physics and QCD. In this talk, I will introduce a variety of new energy correlator based observables, specifically the two and three point heavy energy correlators, which measure correlations of energy flow at collider experiments on heavy quarks. These observables provide new insights into jet substructure, specifically allowing for direct access to hadronization and intrinsic mass effects before confinement. This opens the door to a new class of precision, heavy flavored based measurements at the LHC and beyond.
The energy correlators inside jets are measured for the first time at the CMS experiment. AlphaS is determined from the ratio of 3-point and 2-point correlators.
We will describe an approach to resolving the scales of the QGP using the recently introduced energy correlator observables, showing how these observables enable one to cleanly extract the dynamics of the plasma as a function of scale. We consider the case of both massless and heavy quark jets, and highlight how heavy quark energy correlators provide a particularly sensitive probe of the QGP due to an interplay of the suppression of collinear radiation due to mass effects, and medium induced radiation. We also discuss future theory advances and prospects for experimental measurements with energy correlators.
In recent years, we have seen a rapid development in the description of jets evolving in the presence of structured QCD matter, as the one produced in heavy ions collisions. So far, such theoretical progress has been discussed only at the level of quantities which can not be directly measured. In this talk, I will present the first steps towards understanding the impact of the medium's structure in jet observables. I will show leading order calculations for jet shapes and jet angularities, in the limit of dilute and dense matter at leading order in hydrodynamical matter gradients. Lastly, I will argue that medium anisotropies can, in principle, be detected using energy correlators within jets. I will also introduce some other developments along this direction.
The future Electron-Ion Collider (EIC) will provide the first electron-nucleus collisions for a variety of nuclei species with high luminosity, wide kinematic coverage, and excellent detector resolution, providing new insights into cold nuclear matter effects and transport phenomena. As jets are an accurate proxy of the struck quark that traverses the nucleus, the signature of the nuclear interaction is imprinted into its structure. Recently, driven by significant theoretical advances, an exciting program has been launched to reframe jet substructure in terms of multipoint correlation functions of energy flow operators used to study the dynamics of jets, with many interesting phenomenological applications. In this talk, we explore energy correlators as jet substructure at the EIC and demonstrate how various energy and length scales of the nucleus are cleanly imprinted on the structure of correlations within the jet. Using the eHIJING event generator, we study the 2-point energy-energy correlator (EEC) in various electron-nucleus collisions such as e-C, e-Cu, e-Au, and e-U at the EIC energy. We demonstrate that a clear modification is seen with respect to the electron-proton baseline. In addition, the magnitude of the modification shows a clear nuclear size dependent trend, which can be used to test our knowledge of jet-medium interactions with high precision. This work initiates the exciting program to study in-medium jet evolution and hadronization with EEC and to use higher-point correlators to map out a more differential picture of the structure of the nucleus.
Jet sub-structure observables are known to be sensitive to the effects
due to mass of partons produced via hard scattering QCD interactions.
For example, QCD predicts the suppression of collinear emission around a
massive quark called dead-cone effect and it was recently observed by
the ALICE collaboration at the LHC.
In this talk, we discuss how the quark mass affect the theoretical
computations of an event shape observable such as energy-energy
correlation functions. In particular, we consider and resum the large
logarithms involving the quark mass up to next-to-leading logarithmic
approximation and study the differences between parton shower approaches
to QCD radiation by massive quarks as implemented into Pythia and Herwig
Monte Carlo event generators.
Characterising double-Higgs production has been a major part of the LHC physics program in Run 2 and beyond. We discuss new techniques and results in boosted, hadronic final states in CMS, with a focus on wide-radius jet taggers and data-driven multi-jet background estimation, as well as measurements of gluon-gluon- and vector-boson-fusion HH production in the 4 beauty quark final state in 138fb^{-1} of data at \sqrt{s} = 13 TeV, which observed (expected) a cross section of 9.9 (5.1) relative to the SM prediction and excluded the quartic VVHH coupling \kappa_{2V} = 0 for the first time. Finally, we look ahead to possible new final states and improvements to triggers and techniques in Run 3.
The CMS search program for Vector-Like Quarks uses a wide variety of tools and techniques to identify the massive bosons and heavy quarks resulting from their decays. Since VLQ masses are expected to be in the TeV scale, their decaying daughter particles are highly boosted. CMS searches for VLQs have benefitted greatly from novel jet classifiers designed for high-momentum regimes. This poster will summarize jet identification tools used in CMS and their applications to VLQ searches, in addition to touching upon the outlook of the CMS VLQ search program.
Accurate and precise calorimeter modeling presents one of the most significant computational bottlenecks in modern high-energy physics simulation chains. For this reason, extensive work has been done to speed up calorimeter simulation and make it more computationally efficient. A highly promising method for achieving this speed-up is generative machine learning (ML) models. However, most approaches investigated so far are limited to fixed calorimeter geometries and resolutions. This work presents a major breakthrough in the field of ML calorimeter simulation by, for the first time, directly generating a point cloud of a few thousand space points with energy depositions in the detector in 3D space without relying on a fixed-grid structure. This is achieved through the use of a generative point cloud diffusion model. We showcase the performance of this approach using the specific example of simulating photon showers in the planned electromagnetic calorimeter of the International Large Detector (ILD) and achieve overall good modeling of physically relevant distributions.
The pursuit of detecting high-energy Higgs boson decays into a pair of heavy quarks is a prominent focus within the ATLAS experiment's physics program. In this study, we introduce an innovative tagger that leverages graph networks and employs tracks as input constituents. Our approach demonstrates a substantial improvement when compared to the previous boosted Higgs boson tagger employed by the ATLAS experiment, as observed through extensive Monte Carlo sample analyses. We will present the significant improvements achieved and delve into the details of the training procedure, emphasizing techniques employed to mitigate the tagger's dependency on the reconstructed jet mass.
The BEST Searches for Vector Like Quarks at CMS
Abstract: The Boosted Event Shape Tagger is a multi-class jet tagger optimized for the diverse final states inherent to all-hadronic decays of Vector-Like Quarks. Its architecture is a simple DNN whose discriminating power benefits from physics-driven observables calculated in the lab frame, but also in a series of Lorentz-boosted frames aiming to provide the network with over/rest/under-boosted frame information. The tagger is applied to a search for pair-produced Vector-Like Quarks, each which decay to a third-generation quark and a massive boson, for T- and B-like interpretations in the all hadronic channel.
State-of-the-art (SoTA) deep learning models have achieved tremendous improvements in jet classification performance while analyzing low-level inputs, but their decision-making processes have become increasingly opaque. We propose an analysis model (AM) that amalgamates several phenomenologically inspired neural networks to address this interpretability issue without compromising classification performance. Our methodology incorporates networks that scrutinize two-point energy correlations, generalizations of particle multiplicities via Minkowski functionals, and subjet momenta. Regarding top jet tagging at the hadronic calorimeter angular resolution scale, our AM's performance is on par with SoTA models, such as the ParticleTransformer and ParticleNet.
Subsequently, we explore the generator systematics of top versus QCD jet classification among Pythia (PY), Vincia (VIN), and Herwig (HW) samples using both SoTA models and our AM. Both models can accurately discern differences between simulations, enabling us to adjust the systematic differences via classifier output-based reweighting. Furthermore, AMs equipped with partial high-level inputs (AM-PHLIPs) can be utilized to identify relevant high-level features; if critical features are omitted from the AM inputs, reweighting is affected adversely. We also visualize our correction method, focusing on important variables in top jet tagging, as identified by the DisCo method.
The LHC has unlocked a previously unexplored energy regime. Dedicated techniques have been developed to reconstruct and identify boosted top quarks. Measurements in boosted top quark production test the Standard Model in a region with a strongly enhanced sensitivity to high-scale new phenomena. In this contribution, several new measurements of the ATLAS experiment are presented of the differential cross section and asymmetries in this extreme kinematic regime. The measurements are based on the complete 139/fb run-2 data set of proton-proton collisions at 13 TeV collected in run 2 of the LHC. The measurements are interpreted within the Standard Model Effective Field Theory, yielding stringent bounds on the Wilson coefficients of two-light-quark-two-quark operator.
Identifying boosted hadronic decays of W/Z bosons is central to many LHC physics analyses. This poster presents the performance of constituent-based W/Z boson taggers using large-radius boosted jets reconstructed from Unified Flow Objects (UFOs) in simulated collisions at sqrt(s)=13 TeV. Several taggers which consider all the information contained in the kinematics of the jet constituents are studied. A comparison between these taggers and the current generation of ATLAS W/Z taggers is also provided. Several constituent based taggers are found to improve performance across the wide kinematic range of interest. The dependence of each tagger's performance on physics modeling is also studied.
The basic signals in the ATLAS calorimeter are clusters of topologically connected cell signals. These 'topo-clusters' provide signal features that are sensitive to the differences between electromagnetic and hadronic shower developments. These features are presently used for a local hadronic calibration of these clusters employing a sequence of scale factors retrieved from binned multi-dimensional lookup tables with some observed inefficiencies related to the binning and the corresponding loss of correlations. A new machine-learning-based calibration trains a deep neural network to learn the response of the topo-clusters at the basic signal scale. The resulting predicted response is then the basis of the (smooth) calibration function applied cluster by cluster. First results are shown for the obtained signal linearity and resolution, all at the local level of topo-clusters. In addition, successful network designs and configurations are discussed and motivated by aspects of calorimeter signal formation, the incoming energy flow in jets and environmental conditions like pile-up at the proton-proton collisions at the LHC. This new local calibration promises improved performance in jet substructure reconstruction as well as jet tagging, for both jets reconstructed with the calorimeter only and with particle flow objects.
Calorimeter topo-clusters are the basic ingredient to the reconstruction of jets, electrons photons and tau leptons in the ATLAS experiment. Due to the long integration time, the calorimeters are susceptible to energy deposits from adjacent out-of-time collisions. Reducing the impact of Pile-Up on calorimeter signal reconstruction is fundamental to improve reconstruction efficiency as well as computing resources consumption. This poster presents an improved version of the ATLAS topo-cluster making algorithm, by including a cut on signals??? timing, as measured by the ATLAS calorimeters. The new algorithm performance in hadronic signals reconstruction is evaluated on both Monte Carlo simulation and Run 2 ATLAS data. The new algorithm is found to significantly reduce the effect of Out-Of-Time Pile-Up and has been adopted for topo-cluster building in Run 3.
We overview a new technique to improve the modeling of boosted multi-prong jets. The technique is based on reclustering a multi-prong such that each prong is contained in a separate subjet. The substructure of each prong is then corrected via data-driven reweighting of splittings in the Lund Jet Plane. This is a generic technique that can be applied to jets with high numbers of prongs for which there is currently no known calibration procedure. This will enable future searches utilizing high prong jets.
Event selection and background estimation are the backbone of every search for new physics at the Large Hadron Collider. The TIMBER and 2DAlphabet software frameworks are designed to simplify both procedures and facilitate the development of modular, general-purpose code for physics analyses using Ntuple-like data formats. TIMBER is a Python library designed for streamlining event selection by providing tools for the slimming/skimming of large datasets, automatic application of corrections to Monte Carlo, visualization of analysis cutflows, simple plotting, and the production of histograms for use in shape-based background estimates. By making use of ROOT’s high-level columnar analysis framework, RDataFrame, TIMBER achieves greatly reduced computation times while retaining the readability and simplicity of Python scripting. The 2DAlphabet framework is also presented as a tool for performing data-driven background estimates using a two-dimensional extension to the ABCD (alphabet) method. This Python package automatically constructs the workspace and input to the two-dimensional binned likelihood, and provides the user with tools to plot the resulting distributions and perform a battery of statistical tests to validate the results.
Fast simulations which can accurately model jet substructure are will be of utmost importance for boosted jet analyses at the HL-LHC. There has been significant development recently in generative models for accelerating LHC simulations, but less explored are methods for validating these simulations. We present a rigorous study on evaluation metrics, and discuss the novel Frechet and kernel physics distances as highly sensitive, quantitative metrics for validating not only ML, but potentially also traditional GEANT-based, simulations. We finally introduce our graph network and novel attention-based generative models, which have excellent qualitative and quantitative performance in generating LHC jets, as a case study for the use of these metrics.
The FORMOSA detector at the proposed Forward Physics Facility is a scintillator based detector proposed to search for signatures from high momentum millicharged particles produced in the forward region of the LHC. This talk will cover the challenges of operating such a detector in the forward region, impressive potential sensitivity, and plans for a FORMOSA demonstrator to prove the feasibility of the detector design. In addition, first results will be shown from the scintillator-based Run 3 milliQan detector in the CMS cavern, which is taking data now to search for millicharged particles.
We deploy machine learning techniques to design the hadronic calorimeter for future Electron-Ion Collider (EIC). Tradition method of detector design relies on computationally expensive simulation using Geant4 package. Furthermore, the output of these simulations is not differentiable, and thus cannot be optimized using gradient-based technique. To overcome this hurdle, we are using generative models that can mimic the Geant4 simulation. These models will be differentiable enabling gradient-based optimization. Our detector design approach requires two interconnected components: a generator that mimics the detector, and reconstruction algorithm that uses output of detector to produce physics objects.
This presentation will be focused on AI based reconstruction of highly granular sampling hadronic calorimeter. The proposed Fe/Sc sampling calorimeter is based on SiPM-on-tile technology and capable of providing “5D” information of a shower, energy, position (x, y, and z), and time. For our study we are using the simulated showers from single particles (e-/π+).
Hadronic showers are particularly complex as the shower contain both an electromagnetic and hadronic component, each of which elicit a different detector response. We show how potentially learning from the shape of showers can improve the model performance in reconstructing the energy and polar angle of the incoming particle. Our AI based models have been proven to be more efficient and easier to adopt compared to the traditional approach. We will showcase the results of our AI based results for energy regression using the calorimeter cell information and compare them with the results obtained using more traditional approaches.
The LHC has unlocked a previously unexplored energy regime. Dedicated techniques have been developed to reconstruct and identify boosted top quarks. Measurements in boosted top quark production test the Standard Model in a region with a strongly enhanced sensitivity to high-scale new phenomena. In this contribution, several new measurements of the ATLAS experiment are presented of the differential cross section and asymmetries in this extreme kinematic regime. The measurements are based on the complete 139/fb run-2 data set of proton-proton collisions at 13 TeV collected in run 2 of the LHC. The measurements are interpreted within the Standard Model Effective Field Theory, yielding stringent bounds on the Wilson coefficients of two-light-quark-two-quark operator.
We present searches for additional Higgs bosons produced at high momenta with data collected by the CMS experiment. Searches include additional scalar particles at high mass and searches for decays of the 125 GeV Higgs boson into a pair of light scalars.
Detailed measurements of Higgs boson properties and its interactions can be performed using highly boosted objects, where the detector signatures of two or more decay products overlap. The talk will present several ATLAS analyses targeting these topologies, using collision data collected during Run 2 of the LHC. The talk will present studies of the properties of Higgs boson production at high transverse momentum, where the Higgs boson and associated states such as a weak vector boson or a top quark-antiquark pair are reconstructed as boosted jets. The presentation will also highlight tests of the CP nature of Higgs boson interactions in these topologies. Finally, the talk will present searches for new high-mass Higgs-like resonances decaying into highly boosted Z bosons producing merged di-electron final states.
A first search is conducted for boosted Higgs boson production via vector boson fusion in the H(bb) decay channel at the LHC proton-proton collider. The result is based on the full 13 TeV dataset collected by the CMS detector in 2016, 2017, and 2018, corresponding to an integrated luminosity of 138 fb^{-1}. Jet kinematics are used to define independent regions targeting vector boson fusion (VBF) and gluon fusion (ggF) production of Higgs bosons with p_{T}>450 GeV. The H(bb) decay is isolated by selecting large-radius jets and exploiting jet substructure and dedicated heavy flavour taggers for boosted resonances based on advanced machine learning techniques. The ggF and VBF signal strengths are extracted simultaneously by performing a fit to data in the large-radius jet mass.
Many new-physics signatures at the LHC produce highly boosted particles, leading to close-by objects in the detector and necessitating jet substructure techniques to disentangle the hadronic decay products. This talk will illustrate the use of these techniques in recent ATLAS searches for heavy W' and Z' resonances in top-bottom and di-top final states, as well as in searches for vector-like quarks or dark matter, using the full Run 2 dataset. Additionally, an analysis searching for semi-visible jets, with a significant contribution to missing transverse momentum, is presented. This type of topologies can arise in strongly-interacting dark sectors.
We present searches for new particles in final states with high boost at CMS. Results include searches for resonances such as, for example, heavy resonances coupling to multi-bosons, third-generation particles, leptoquarks or vector-like quarks. We focus on describing new results that use boosted reconstruction techniques for objects in the final state (including MET).
Searches for new resonances in two-body invariant masses are performed using an unsupervised anomaly detection technique in events produced in $pp$ collisions at a center-of-mass energy of 13 TeV recorded by the ATLAS detector at the LHC. An autoencoder network is trained with 1% randomly selected collision events and anomalous regions are then defined which contain events with high reconstruction losses. Studies are conducted in data containing at least one isolated lepton. Nine invariant masses ($m_{jX}$) are inspected which contain pairs of one jet ($b$-jet) and one lepton ($e$, $\mu$), photon, or a second jet ($b$-jet). No significant deviation from the background-only hypothesis is observed after applying the event-based anomaly detection technique. The obtained model-independent limits are shown to have a strong potential to exclude generic heavy states with complex decays.
Searches for new resonances in two-body invariant mass distributions are performed using an unsupervised anomaly detection technique in events produced in pp collisions at a center of mass energy of 13 TeV recorded by the ATLAS detector at the LHC. Studies are conducted in data containing at least one isolated lepton. An autoencoder network is trained with 1% randomly selected collision events and anomalous regions are then defined which contain events with high reconstruction losses from the decoder. Nine invariant mass distributions are inspected which contain pairs of one light jet (or one $b$-jet) and one lepton ($e$, $\mu$), photon, or a second light jet ($b$-jet). The 95% confidence level upper limits on contributions from generic Gaussian signals are reported for the studied invariant mass distributions. The obtained model-independent limits show strong potential to exclude generic heavy states with complex decays.
Anomaly, or out-of-distribution, detection is a promising tool for aiding discoveries in particle physics. However, existing detection methods perform detection against only a single background process, primarily using boosted objects. Moreover, most algorithms can depend on nuisance features such as jet mass to falsely tag known physics as an anomaly or ignore new physics as an already known background process. In this work, we build detection algorithms for anomalous particle decays against multiple background types. In addition, we generalize the notion of decorrelation to the multi-background setting by combining representation learning techniques for robustness to nuisances and classifier-based out-of-distribution detection algorithms. We demonstrate the benefit of such nuisance-aware out-of-distribution detection algorithms on a high-dimensional dataset of particle decays at the Large Hadron Collider.
In this talk, we will discuss the consequences of models where dark sector quarks could be produced at the LHC, which subsequently undergo a dark parton shower, generating jets of dark hadrons that ultimately decay back to Standard Model hadrons. This yields collider objects that can be nearly indistinguishable from Standard Model jets, motivating the reliance on substructure observables to tease out the signal. However, substructure predictions are sensitive to the details of the incalculable dark hadronization. We will show how the Lund jet plane can be an effective tool for designing observables that are resilient against the unknown impact of dark hadronization on the substructure properties of dark sector jets.
Jet taggers based on substructure are exploited in many areas of the LHC physics program. The ability to select jets stemming from different incident particles, so-called jet tagging, is afflicted with systematic uncertainties of both theoretical and experimental nature. This results in taggers that are ultimately less powerful than they could be. We propose an algorithm that creates a feature space that is insensitive to the parton shower model. This approach effectively minimizes the impact of jet-related systematic uncertainties, which are oftentimes among the dominant ones in the analysis of collider data. This is achieved by considering augmented data corresponding to a set of systematic uncertainties: partons are re-showered by systematically tuning parameters in Pythia8 and Herwig7. The resulting jets are embedded in a contrastive space via self-supervised deep neural networks. We consider graph and transformer-based architectures. In this talk, we demonstrate that our method achieves comparable tagging performance to established benchmarks for jet tagging in Higgs to bottom quarks, but with greatly reduced uncertainties. The algorithm presents a crucial step towards more robust searches for new physics involving particle jets, and paves the way for ultimate-precision jet measurements.
Despite the recent proliferation of symmetry-based machine learning methods in jet physics, the preference for smaller symmetry groups and highly custom architectures negatively impacts explainability and generalizability. In this work, we present an update to our own algorithm, which delivers both significant improvements in the top-tagging performance and the capability to perform full four-vector-regression. PELICAN is a fully Lorentz equivariant network which acts on input four-vectors in a permutation equivariant manner. The incorporation of these symmetries yields a network with unique explainability features and visualization capabilities. We investigate the generalizability of our network with respect to jet mass regression tasks in the context of Lorentz-boosted decays of top quarks and W bosons, performance in infrared and collinear safe regimes, and other frame-dependent detector effects. Finally, we propose an interpretation of PELICAN as a soft clustering algorithm and its potential use for Lorentz-invariant latent representations of jets.
Discrepancies between real and simulated collider events are a significant source of uncertainty in LHC physics, especially in the context of training machine learning models, where even small differences in input variable distributions can degrade the performance of a simulation-trained model. In this work we present a deep learning implementation of “chained quantile morphing”: a technique to correct Monte Carlo (MC) mis-modeling of data using normalizing flows. A collection of variables is corrected from MC to data using an iterative procedure, which ensures that their single-variable conditional densities match at each step. This approach is able to account for and preserve inter-variable correlations, and minimally shifts values in each sample. While originally conceived to correct MC to match data, the technique can be applied to map between any two distributions. We demonstrate the method using pairs of 2D toy datasets, jet substructure observables from different BSM signals, and events simulated by different MC generators. We also apply it to correct physically meaningful embeddings of collider events in a contrastive space.
We study the effectiveness of theoretically-motivated high-level jet observables in the extreme context of jets with a large number of hard sub-jets (up to N=8). Previous studies indicate that high-level observables are powerful, interpretable tools to probe jet substructure for N≤3 hard sub-jets, but that deep neural networks trained on low-level jet constituents match or slightly exceed their performance. We extend this work for up to N=8 hard sub-jets, using deep particle-flow networks (PFNs) and Transformer based networks to estimate a loose upper bound on the classification performance. A fully-connected neural network operating on a standard set of high-level jet observables, 135 N-subjetiness observables and jet mass, reach classification accuracy of 86.90%, but fall short of the PFN and Transformer models, which reach classification accuracies of 89.19% and 91.27% respectively, suggesting that the constituent networks utilize information not captured by the set of high-level observables. We then identify additional high-level observables which are able to narrow this gap, and utilize LASSO regularization for feature selection to identify and rank the most relevant observables and provide further insights into the learning strategies used by the constituent-based neural networks. The final model contains only 31 high-level observables and is able to match the performance of the PFN and approximate the performance of the Transformer model to within 2%.
As machine learning begins to play an increasingly larger role in high energy physics, it is important to understand and interpret what precisely these models learn. In this work, we propose Moment Pooling architectures, which generalizes the summation in standard Deep Sets architectures to an arbitrary multivariate moments. This can be used to drastically decrease latent space sizes, significantly improving the model's interpretability while maintaining performance. We show that this is particularly useful in jet physics, where many existing useful jet observables can be naturally expressed in this form. We then show several examples of how the Moment Pooling architecture may be used in jet tagging, as well as how this structure can provide insight on the complexity of jet observables.
In this talk, we present several variations, applications, and analytic interpretations of the event-shape observable event isotropy. We explore how the the observable can be altered, e.g. by varying the distance metric, the reference topology, the underlying geometry, etc., to be more or less sensitive to features of the event. With these studies one can improve their discrimination power in searches for rare SM processes or BSM phenomena.
We present a class of Neural Networks which extends the notion of Energy Flow Networks (EFNs) to higher-order particle correlations. The structure of these networks is inspired by the Energy-Energy Correlators of QFT, which are particularly robust against non-perturbative corrections. By studying the response of our models to the presence and absence of non-perturbative hadronization, we can identify and design networks which are insensitive to the simulated hadronization model, while still optimized for a given performance objective. Moreover, the trained models can give surprising insights into the physics of the problem, for example by spontaneously learning to identify relevant energy scales. We demonstrate our method by training an effective tagger for boosted bosons with minimal sensitivity to theory systematics, which are notoriously difficult for experimentalists to quantify.
In this talk, we present a new proposal on how to measure quark/gluon jet properties at the Large Hadron Collider (LHC). The main advantage of this approach is that our construction of an observable allows a single set of experimental cuts to be used to select jets, keeping all detector parameters unchanged, and in this way, reducing many systematic uncertainties. We will discuss the details of the measurement strategy, including the theoretical basis and the experimental techniques for measuring them. We will also present preliminary results from our phenomenological analysis of samples generated by Herwing and Pythia event generators using this approach. Overall, our proposed measurement strategy provides a promising new avenue for studying quark/gluon jet properties at the LHC.
Jet angularities provide a detailed representation of the radiation pattern within high energy quark and gluon jets. While there has been major advances in jet substructure studies at hadron colliders, the precision achievable by collisions involving electrons is superior, since several of the complications from hadron colliders are absent.
In this contribution jets are analyzed which were produced in deep-inelastic electron-proton scattering and recorded by the H1 experiment at HERA (DESY). The measurement is unbinned and multi-dimensional, making use of machine learning to correct for detector effects. Results are presented after unfolding the data to particle level for events in the fiducial volume of momentum transfer Q²>150 GeV² and jet transverse momentum $p_\text{T}^\text{jet}>10\,\text{GeV}$. All of the available object information in the events is used to achieve the best precision through the novel use of graph neural networks (GNN). The networks were trained at the new Perlmutter supercomputer at Berkeley Lab with a large number of Graphical Processing Units (GPUs).
After a hard scattering event, an outgoing parton will radiate gluons which fragment into final-state hadrons. To study the radiation patterns of light and heavy partons, we look at the Lund jet plane (LJP), an observable where various types of emissions such as soft-collinear, hard-collinear, and non-perturbative emissions as well as initial-state radiation and the underlying event can be separately identified. To study fragmentation into hadrons, transverse-momentum-dependent fragmentation functions (TMD FFs) go beyond traditional collinear non-perturbative FFs and provide multidimensional information on the hadronization process. By measuring the LJP aimed at the parton level, and TMDs aimed at the hadron level, we obtain a more complete picture of the formation and evolution of jets. Recent results measuring TMD jet FFs for identified charged pions, kaons, and protons in a predominantly light quark jet sample will be presented, as well as the status of analyses of the LJP for light-, charm-, and beauty-initiated jets at the LHCb experiment, a forward (2 < $\eta$ < 5) detector well-optimized for studying heavy flavor physics.
Jets are collimated sprays of final-state particles produced from initial high-momentum-transfer partonic scatterings in particle collisions. Since jets are multi-scale objects that connect asymptotically free partons to confined hadrons, jet substructure measurements can provide insight into the parton evolution and the ensuing hadronization processes. Compared to the jets at the LHC, jets produced in $\sqrt{s} = 200$ GeV $pp$ collisions at RHIC have lower transverse momenta and are therefore more susceptible to non-perturbative effects. The jet substructure measurements in the STAR experiment, therefore, provide complementary information about different regimes of quantum chromodynamics. In addition to the inclusive and SoftDrop groomed jet observables, such as jet mass ($M$), jet charge ($Q$), groomed jet mass ($M_{\mathrm{g}}$), groomed jet radius ($R_{\mathrm{g}}$) and shared momentum fraction ($z_{\mathrm{g}}$), the STAR collaboration has also recently measured the correlations between various substructure observables.
We extend the previous studies of multi-dimensional jet substructure observables by studying the correlation between SoftDrop and CollinearDrop groomed jet observables, the latter of which have an enhanced sensitivity to soft radiation within jets. Such correlation measurements reveal the interplay between different stages of the parton shower. In this talk, we present the first measurements of the CollinearDrop groomed jet mass and its correlation with $R_{\mathrm{g}}$ and $z_{\mathrm{g}}$, in $pp$ collisions at $\sqrt{s} = 200$ GeV. The measurements are fully corrected for detector effects with MultiFold, a novel machine learning method which preserves the correlations in the multi-dimensional observable phase space. We compare our fully corrected measurements with predictions from event generators such as PYTHIA and HERWIG.
Modern machine learning (ML) techniques allow us to rethink how the the high dimensional features of jets can be optimally used to probe the strong interaction. Recently a new class of jet substructure observables, the energy correlators (EECs), have been introduced to study the statistical properties inside jets and enable first principle ways to do physics in the complicated LHC environment. We demonstrate the high dimensional potential of a recently proposed ML-based analysis method, OmniFold, to experimentally study the field theoretic observables EECs. In particular, we show with Dire simulation that EECs can probe the higher order effects of DGLAP evolution, which can be precisely analyzed through OmniFold.
At leading order in positron-proton collisions, a lepton scatters off a quark through virtual photon exchange, producing a quark jet and scattered lepton in the final state. The total transverse momentum of the system is typically small, but deviations from zero can be attributed to perturbative initial and final state radiations in the form of soft gluon radiation when the transverse momentum difference, $|\vec{P}_\perp|$, is much greater than the total transverse momentum of the system, $|\vec{q}_\perp|$. The soft gluon radiation comes exclusively from the jet, and should result in a measurable azimuthal asymmetry between $|\vec{P}_\perp|$ and $|\vec{q}_\perp|$. Quantifying the contribution of soft gluon radiation to this asymmetry should serve as a novel test of perturbative QCD as well as an important background estimation for measurements of the lepton-jet imbalance that have recently garnered intense investigation. The measurement is performed in positron-proton collisions from HERA Run-II measured with the H1 experiment. A new machine learning method is used to unfold eight observables simultaneously and unbinned. The final measurement, the azimuthal angular asymmetry, is then derived from these unfolded and unbinned observables. Results are compared with parton shower Monte Carlo predictions as well as soft gluon radiation calculations from a Transverse Momentum Dependent (TMD) factorization framework.
Additionally, a multi-differential measurement of the momentum imbalance between outgoing jets and the scattered positron is reported, which provides a useful test of pQCD in the regime where collinear and transverse-momentum-dependent frameworks overlap.
Dinner will be at Skates, a restaurant in Berkeley right on the San Francisco Bay. There will be a shuttle running from 5-7 in a loop from the building 50 auditorium at LBNL to downtown Berkeley and then to the restaurant.
Measuring the jet substructure in heavy-ion collisions provides exciting new opportunities to study detailed aspects of the dynamics of jet quenching in the hot and dense QCD medium created in these collisions. In this talk, we present new comprehensive ATLAS measurements of jet suppression and substructure performed using various jet radii and grooming techniques. We will present new results of the jet substructure, which use Soft-Drop grooming procedure to identify the hardest parton splitting in the jet. The measurements are performed using different jet constituents such as charged tracks, smaller R calorimeter jets, and novel objects reconstructed using tracker and calorimeter information. The jet suppression is characterized using RAA and presented as a function of its transverse momentum ($p_{\mathrm{T}}$), the angle of the hardest splitting ($r_{g}$), and the corresponding transverse momentum scale ($\sqrt{d_{12}}$). These new measurements, along with theory comparisons, will elucidate the mechanisms of jet suppression, medium effects, and energy recovery in the QCD medium.
Event shape observables provide incisive probes of QCD, both its perturbative and nonperturbative aspects. Grooming techniques have been developed to separate perturbative from non-perturbative components of jets in a theoretically well-controlled way, and have been applied extensively to jet measurements in hadronic collisions.
In this contribution, the first application of grooming techniques to event shape measurements in electron-proton collisions is presented, utilizing data from the H1 experiment at HERA (DESY). The analysis is based on the novel Centauro jet clustering algorithm, which is designed specifically for the event topologies in deep inelastic scattering. Cross-section measurements of groomed event 1-jettiness and groomed invariant jet mass are shown, as well as a measurement of the ungroomed 1-jettiness event shape observable. The measurements are compared to Monte Carlo models, fixed order QCD predictions, and to a theoretical calculation based on Soft Collinear Effective Theory.
Jet substructure observables are incisive probes of quantum chromodynamics (QCD), providing insight into perturbative and non-perturbative processes, and probing the structure and dynamics of the quark-gluon plasma (QGP). The jet shower is sensitive to multiple scales during its evolution, encoding the physics into correlated angular and momentum space phenomena which cannot be fully characterized by a single observable. This multidimensional complexity requires both new approaches via new observables and detailed studies of existing measurement techniques to fully disentangle and understand the encoded information. In this talk, we report several recent ALICE jet substructure measurements in pp and Pb--Pb collisions at $\sqrt{s_{\mathrm{NN}}} = 5.02$ TeV, both with and without grooming. We present a new measurement of the energy-energy correlators (EEC) in pp collisions at $\sqrt{s}$ = 5.02 TeV. The EEC is a momentum-weighted angular correlation of jet constituents which probes intra-jet energy flow, with clear separation between perturbative and non-perturbative regimes. We also report new studies of the soft drop and dynamical grooming algorithms, benchmarking their performance using the groomed relative transverse momentum, $k_{\mathrm T,g}$. These new measurements include the first application of dynamical grooming in heavy-ion collisions, and enable a search for excess $k_{\mathrm{T,g}}$ emissions arising from point-like scatters in the QGP, providing new constraints on its quasi-particle nature.
Jets are powerful probes used to improve our understanding of the strong force.
A useful way of understanding the radiation pattern of the jet is via the Lund jet plane, a representation of the phase space constructed using iterative Cambridge/Aachen declustering. In this talk, we discuss recent jet substructure measurements in pp and PbPb collisions based on Cambridge/Aachen declustering.
Diverse studies of jets in heavy-ion collisions promise a consistent picture of QCD medium interactions and a path towards further differentiating energy loss mechanisms. Some results, however, remain disjoint: the jet mass and jet angularities, including girth and thrust, are strongly-correlated observables which have given seemingly conflicted answers on the angular quenching of jets traversing the QGP. New systematic measurements of the perturbatively-calculable angularities, using consistent definitions for the first time, resolve the long-standing girth-mass problem and reveal quenching effects at broad angles. Concurrently, applying soft drop grooming isolates the narrowing in the core of quenched jets. New comparisons of the jet axis differences between groomed and ungroomed jets using various recombination schemes highlights the quenching contributions from soft radiation. Pushing to lower transverse momentum allows these studies to illuminate enhanced quenching effects at small $Q^2$. Similarly, tagging jets with an external probe provides information about jet energy and momentum modifications with respect to an initial hard-scattered parton, improving access at low momentum. A novel mixed-event approach has also pushed jet-medium studies down to zero transverse momentum, providing a potential opportunity to reduce bias on future jet measurements. We compare recent results to assorted jet quenching models, providing information on medium interactions as a function of angle and momentum.
With the newly upgraded sPHENIX detector capable of performing high precision jet substructure measurements, we present a comprehensive and systematic jet substructure study at Relativistic Heavy Ion Collider. The study includes a variety of key jet substructure variables such as jet angularities with and without soft-drop or collinear-drop grooming, as well as recoil-free di-jet and photon+jet azimuthal angle decorrelation using the Winner-Take-All (WTA) recombination scheme to define the jet directions. We employ various event generators in the study, including Pythia, Herwig and Sherpa for the proton-proton collision baseline. The Caeser framework is used to perform semi-numerical calculations for these observables. With the well-defined perturbative precision, non-perturbative contributions can be robustly extracted. For the medium jet substructure study, we investigate different Monte Carlo implementations of quenching models such as JETSCAPE, Jewel and QPythia. The jet observables have different sensitivities to physics ranging from parton shower to soft radiation, allowing us to quantify model differences and thereby point to modification patters through which future experimental data may shed light on the nature of jet-medium interaction.
The properties of partonic fragmentation in QCD depend on the flavors of the partons involved in the 1→2 splitting processes that drive parton showers. These dependencies arise from the differences in the Casimir factors of quarks and gluons, as well as the mass of heavy quarks. To explore these flavor dependencies, we use heavy-flavor jets as an experimental tool, particularly at low and intermediate transverse momenta where mass effects are significant. In this talk, we present the recent results of charm-tagged jets (reconstructed D$^0$-meson) by ALICE at the LHC. These results include the first direct measurement of the dead-cone effect at colliders, using the comparison of Lund planes of charm-tagged jets and inclusive jets and the first measurement of the jet angularity which can be tuned in their sensitivity to mass and Casimir effects. Additionally, the groomed momentum fraction and opening angle of the first splitting are reported, which link to fundamental ingredients of the splitting functions. Comparisons to flavour-untagged jet sample will probe both the flavour depdendeces due of the mass of the charm quark, as well as the high purity quark nature of the D$^0$-tagged jet sample. Further comparisons to different MC generators will access the role of these flavour dependencies in different parton shower prescriptions.
The jet charge is an old observable that has proven uniquely useful for discrimination of jets initiated by different flavors of light quarks, for example. In this Letter, we propose an approach to understanding the jet charge by establishing simple, robust assumptions that hold to good approximation non-perturbatively, such as isospin conservation and large particle multiplicity in the jets, forgoing any attempt at a perturbative analysis. From these assumptions, the jet charge distribution with fixed particle multiplicity takes the form of a Gaussian by the central limit theorem and whose mean and variance are related to fractional-power moments of single particle energy distributions. These results make several concrete predictions for the scaling of the jet charge with the multiplicity, explaining many of the results already in the literature, and new results we validate in Monte Carlo simulation.
Recently, spurred by significant theoretical advances, an exciting program has been launched to reframe jet substructure in terms of multipoint correlation functions of energy flow operators. These energy correlators successfully image the intrinsic and emergent scales of QCD as a function of scale with many interesting phenomenological applications.
In this talk, we present an exciting vision of the future, where we go beyond the simple energy flow paradigm and expand the landscape of asymptotic observables / detectors to gain insight into how different quantum numbers flow. Using charge flow as an example, we demonstrate how charge energy correlators nontrivially modify the scaling behaviors in the perturbative scattering region, thereby illustrating nontrivial changes in the hadronization transition rate. Furthermore, we will discuss the insight that can be gained from this approach and initiate the exciting program to generalize to even wider class of asymptotic observables.
Due to its large mass the top quark plays an important role in consistency checks of the Standard Model and new-physics searches. Studies concerning precise theoretical predictions of the top production and its decay are commonly based on the narrow-width (NW) limit of the top quark propagator or on full off-shell computations. The NW limit, where the top quark is treated as an on-shell particle, allows for a convenient factorization of the top production and decay dynamics and is mostly used for current experimental analyses. On the other hand, off-shell computations consistently include non-resonant effects and non-factorizable corrections, but are more complicated to apply and determine in practice, particularly when QCD corrections are accounted for. In this talk I present a novel approach for boosted top quarks that allows to combine the factorization property of the NW limit and off-shell effects. The approach is valid in the resonance region for boosted top production and applies methodology known from SCET to the electroweak theory. The approach allows to incorporate resummed QCD corrections for differential top decay observables using boosted HQET generalizing results known from semileptonic $B$ decays.
We provide a quantitative interpretation for direct top quark mass measurements in terms of a field-theoretic mass scheme. A relation between the top quark mass parameter in Monte Carlo generators and the MSR mass at a scale of R = 3 GeV is derived. This is achieved by fitting Monte Carlo templates for the simulated jet mass distribution of large-radius jets containing a hadronically-decaying top quark of large transverse momentum with a first-principles, hadron level calculation at NNLL accuracy. The result confirms that the MC mass parameter is numerically close to the MSR and pole masses.
Several jet flavor identification strategies have been employed to identify jets originating from heavy-flavor quarks and from the decay of heavy Lorentz-boosted objects. Often, separate tagging strategies are utilized for different Lorentz-boost regimes of the target physics object. For instance, in searches for the Higgs boson decaying into a pair of heavy-flavor quarks, a division is made between the low-boost (resolved) regime, where two separate thin jets are flavor-tagged, and the high-boost (merged) regime, where a single large-radius jet is flavor-tagged. Although the latter approach effectively exploits correlations between the two decay products thereby rejecting background more strongly, it fails to effectively perform in low-boost regimes, which encompass the majority of signal events. In this talk, we propose a novel tagging strategy that optimally exploits available information across all Lorentz-boost regimes. The new approach employs clustered thin jets as seeds to define unconventional jets, referred to as PAIReD jets, which are subsequently input into existing machine learning-based algorithms to discriminate between signal and background events. As a result, we achieve a 2-3x stronger background rejection compared to conventional strategies in low Lorentz-boost regimes, without the need for additional clustering algorithms. The momenta, masses, and tagger scores of these PAIReD jets can be calibrated using standard calibration methods employed in hadron collider experiments.
Decays of ultra-heavy resonances of masses up to 8 TeV represent some of the heaviest particles that could be produced at the Run II LHC. Many such models describe these heavy resonances decaying to pairs of vector-like quarks (VLQs) with the fully-hadronic channels generally having the largest branching fraction. While giving the largest event yield, the extreme level of jet collimation and complex final states of fully-hadronic decays pose a substantial challenge to reconstructing the underlying resonance and daughter pair. To study these types of decays, a jet sorting algorithm has been developed that uses event geometry and a series of Lorentz boosts to reconstruct the daughter pair for these fully hadronic events. The application of this technique on simulations of a model containing a scalar diquark decaying to VLQ pairs is shared.
We describe a new jet clustering algorithm named SIFT (Scale-Invariant Filtered Tree) that maintains the resolution of substructure for collimated decay products at large boosts. The scale-invariant measure combines properties of kT and anti-kT by preferring early association of soft radiation with a resilient hard axis, while avoiding the specification of a fixed cone size. Integrated filtering and variable-radius isolation criteria block assimilation of soft wide-angle radiation and provide a halting condition. Mutually hard structures are preserved to the end of clustering, automatically generating a tree of subjet axis candidates. Excellent object identification and kinematic reconstruction for multi-pronged resonances are realized across more than an order of magnitude in transverse energy. The clustering measure history facilitates high-performance substructure tagging, which we quantify with the aid of supervised machine learning. These properties suggest that SIFT may prove to be a useful tool for the continuing study of jet substructure.
We present results using an optimized jet clustering with variable $R$, where the jet distance parameter $R$ depends on the mass and transverse momentum $p_T$ of the jet. The jet size decreases with increasing $p_T$, and increases with increasing mass. This choice is motivated by the kinematics of hadronic decays of highly Lorentz boosted top quarks, W, Z, and H bosons. The jet clustering features an inherent grooming with soft drop and a reconstruction of subjets in one sequence. These features have been implemented in the Heavy Object Tagger with Variable R (HOTVR) algorithm, which we use to study the performance of jet substructure tagging with different choices of grooming parameters and functional forms of $R$.
Jets are the most familiar and complex physical objects in high-energy physics experiments.
Since jets invariably appear in critical elementary particle processes, identifying their origin has become an essential technique for physical analysis, including new particle searches and precision measurements.
Many particles are produced in jets, detected by particle detectors, and reconstructed as flight trajectories or calorimetric clusters.
In recent years, machine learning, mainly neural networks, has been studied and used for the flavor classification of jets.
Graph neural networks, in particular, have been well studied and have produced excellent results, as they are well suited to representing the particles in jets.
Transformer, which has been successful in the development of large-scale language models in recent years, can also be used for flavor classification of jets due to its versatile architecture.
Thus, we adopt the latest techniques of the Transformer family instead of the traditional graph neural networks.
Thus, it is possible to apply the shortcomings of graph neural networks and the latest research results of the Transformer family.
This presentation will present research results on jet flavor classification using Transformer as architecture.