PaperlayerView exact revision on arXiv

Source-linked AI summary

Delphes, a framework for fast simulation of a generic collider experiment

S. Ovyn, X. Rouby, V. Lemaitre

arXiv:0903.2225v3hep-phhep-ex

TL;DR

Phenomenological collider studies need fast but realistic detector estimates without the full complexity of detailed detector simulation. Delphes addresses this with a configurable fast simulation and reconstruction framework, producing analysis objects, trigger outputs, and visualisation data; its validation reports good agreement with CMS and ATLAS jet-resolution curves and with experiment-compatible MET-resolution parameters.

  • Problem

    Phenomenological studies require fast but realistic estimates of collider signal signatures and backgrounds, whereas detailed detector simulation accounts for extensive material, inefficiency, and geometry effects.

  • Method

    Delphes simulates configurable central and forward detector responses from event-generator output, reconstructs analysis objects, and supports trigger emulation and event visualisation.

  • Results

    Delphes jet-resolution curves show good agreement with CMS and ATLAS results, while its MET-resolution parameters are α = 0.63 for CMS-like and 0.53 for ATLAS-like detectors.

  • Takeaways & Limitations

    Delphes is intended as a tool for feasibility studies that gauge the observability of model predictions in collider experiments.

  • Takeaways & Limitations

    The framework uses idealised uniform, beam-axis-symmetric geometry and neglects cracks, dead material, secondary interactions, multiple scattering, photon conversion, and bremsstrahlung.

Abstract

from arXiv · show

This paper presents a new C++ framework, DELPHES, performing a fast multipurpose detector response simulation. The simulation includes a tracking system, embedded into a magnetic field, calorimeters and a muon system, and possible very forward detectors arranged along the beamline. The framework is interfaced to standard file formats (e.g. Les Houches Event File or HepMC) and outputs observables such as isolated leptons, missing transverse energy and collection of jets which can be used for dedicated analyses. The simulation of the detector response takes into account the effect of magnetic field, the granularity of the calorimeters and subdetector resolutions. A simplified preselection can also be applied on processed events for trigger emulation. Detection of very forward scattered particles relies on the transport in beamlines with the HECTOR software. Finally, the FROG 2D/3D event display is used for visualisation of the collision final states.

PROGRAM SUMMARY

Delphes is a C++ framework for fast, realistic estimates of collider-event signatures, combining parametrised detector response with reconstruction, trigger emulation, and event visualisation. It models central and forward detector components while exposing configurable coverage, resolutions, and running conditions.

  • Purpose: Delphes provides fast detector simulation for phenomenological studies that need realistic signal and background estimates.It starts from event-generator output and smears final-state kinematics according to subdetector resolutions.
  • Detector response: The framework includes central and forward detector geometry, magnetic-field effects on tracks, and calorimeter response based on segmentation and resolution.The default system contains tracking, ECAL, HCAL, FCAL, and an enclosing muon system.
  • Limitations: The simulation idealises detector geometry and omits secondary interactions, multiple scattering, photon conversion, and bremsstrahlung.These assumptions define an important boundary on the realism of the fast simulation.
  • Workflow: Delphes accepts standard event-generator inputs and produces ROOT analysis data, with optional trigger results and 3D visualisation files.The output can include generator-level data, reconstructed objects, trigger-emulation results, and optional LHCO text output.
  • Configurability: Detector parameters such as coverage, resolutions, thresholds, and jet settings are controlled through a detector card, enabling configurations beyond the default LHC-like setup.The default calorimeter coverage is |η| < 3 for ECAL and HCAL and 3.0–5.0 for FCAL.
  • Detector response: Tracks are reconstructed for charged particles above configurable thresholds, while calorimeter cells sample energy deposits for downstream reconstruction.The default track reconstruction efficiency is 90% for pT > 0.9 GeV/c and |η| ≤2.5.

3. High-level reconstruction

Delphes constructs high-level analysis objects from generator-identified particles and reconstructed detector information, including leptons, isolation variables, jets, and hadronic τ candidates. Its object definitions use configurable thresholds and simplified detector assumptions.

  • High-level objects: Delphes high-level output provides four-momenta and object-specific properties such as charge, isolation, b-tagging, and time-of-flight.The listed reconstructed objects include photons, leptons, jets, b-jets, τ-jets, and missing transverse energy.
  • Leptons and photons: Electron, muon, and photon collections contain generator-level particles that pass fiducial and reconstruction cuts, so they contain no fake candidates by default.Fake candidates can instead be added later at analysis level.
  • Muons: Muon candidates require acceptance and transverse-momentum thresholds, receive Gaussian pT smearing, and omit multiple scattering and calorimeter punch-through.The defaults are −2.4 ≤η ≤2.4 and pT > 10 GeV/c.
  • Isolation: Charged-lepton isolation requires no other charged particle above 2 GeV/c within a cone of ΔR < 0.5, while track-pT sums and calorimetric ratios are also recorded.The magnetic field is taken into account when determining nearby particles.
  • Jets: Jets are reconstructed from calorimeter cells using six FastJet schemes, with optional energy flow subtracting track-associated energy before clustering.The algorithms differ in sensitivity to soft particles, collinear splittings, and computing speed.

Cone algorithms

Delphes provides multiple cone and recombination-based jet algorithms, with configurable defaults and detector-aware inputs. The framework also supports τ-jet identification using track multiplicity, jet collimation, and isolation criteria.

  • Jet algorithms: Six jet algorithms are available: three cone algorithms and three recombination algorithms based on successive calorimeter-cell merging.The cone choices are CDF Jet Clusters, CDF MidPoint, and SISCone; the recombination choices are longitudinally invariant kt, Cambridge/Aachen, and Anti kt.
  • Jet algorithms: The Anti kt algorithm produces hard jets that are exactly circular in the (y, φ) plane.
  • Jet algorithms: Recombination algorithms are insensitive to soft radiation and collinear splittings, differing mainly in their definitions of merging distances.
  • Jet algorithms: CDF cone reconstruction is the default, and jets are stored only when their transverse energy exceeds 20 GeV.
  • Detector-aware reconstruction: Energy flow subtracts track-pointing energy from calorimeter cells and smears it separately before jet reconstruction.

Tracking isolation

Delphes applies tracking-based selection and reconstruction procedures to identify τ-jets and other analysis objects, while simulating detector effects and trigger decisions. Its forward-detector components extend coverage beyond the central detector using dedicated beamline transport and acceptance models.

  • Tracking isolation: 66% typical efficiency is obtained for hadronically decaying τ leptons after electromagnetic collimation, tracking isolation, and a candidate-pT threshold.The isolation requirement rejects multi-prong candidates, while the final threshold purifies the collection.
  • Tracking isolation: Delphes computes missing transverse energy from calorimeter cells, excluding muons and neutrinos, while retaining the inputs for posterior specialised reprocessing.The same calorimetric cells also provide inputs to jet reconstruction.
  • Tracking isolation: Trigger emulation uses a fully parametrizable table applied to reconstructed analysis objects, including logical combinations of object-level threshold conditions.This differs from real triggers, whose online selection uses lower-resolution reconstructed information than final analysis data.

6. Validation

Delphes is validated against CMS- and ATLAS-like expectations using jet and missing-transverse-energy resolutions. The reconstructed resolutions show good agreement with the corresponding experimental results.

  • Jet resolution: Jet resolution is validated with pp → gg events generated by MadGraph/MadEvent and hadronised with Pythia.The validation covers CMS-like and ATLAS-like detector configurations.
  • Jet resolution: CMS-like jet resolutions from Delphes and CMS are in good agreement.Jets use JetCLU with cone radius R = 0.7 and are matched to generator-level jets.
  • Jet resolution: ATLAS-like jet resolutions from Delphes agree well with the fitted result and the corresponding ATLAS curve.The evaluation uses a kT algorithm with R = 0.6 and maximal matching distance ∆R = 0.2.
  • MET resolution: MET performance is assessed through the resolution of the horizontal component of missing transverse energy across bins of total visible transverse energy.The resolution is obtained by fitting the difference between Delphes and generator-level Emiss_x with a Gaussian in each ΣET bin.
  • MET resolution: The MET resolution parameter α is 0.63 for CMS-like and 0.53 for ATLAS-like detectors, matching the cited experimental expectations.CMS expects α = 0.6–0.7, while ATLAS expects 0.53–0.57 for the corresponding studies.

7. Visualisation

Delphes integrates FROG to visualize detector geometry and collision event topologies in two and three dimensions. Colour coding and interactive kinematics support interpretation of reconstructed objects, while example displays contrast event activity across processes.

  • Detector configuration: FROG provides two- and three-dimensional displays of Delphes detector configurations, illustrating the geometric coverage of different detector subsystems.The generic geometry includes central tracking, central calorimeters, and forward calorimeters; additional forward detectors are not shown in the example layout.
  • Event topology: Colour coding distinguishes reconstructed electrons, muons, taus, jets, and transverse missing energy in event displays.This enhances the visibility of each object set for examining collision final states.
  • Event topology: Interactive inspection exposes kinematic information for individual reconstructed objects through simple mouse actions.An associated photoproduction event of a W boson and a top quark illustrates this capability.
  • Event topology: The Wt photoproduction example shows an isolated muon, a reconstructed fake τ-jet around the electron, and missing energy in the display.The event also contains a surviving proton with a forward hemisphere lacking hadronic activity.
  • Event topology: The inclusive gluon-pair example contains many jets, especially along the beam axis, unlike the illustrated Wt photoproduction event.The paper attributes this beam-axis activity to the destruction of the interacting protons in the collision.
  • Role in Delphes: Delphes uses visualization alongside fast detector simulation, trigger emulation, and reconstructed analysis objects for collider feasibility studies.The framework is intended to gauge the observability of model predictions and can be extended beyond its CMS-based parameters to ATLAS and other experiments.

Internal code references

Delphes uses standard event and ROOT-based analysis interfaces, with configurable classes and detector-card parameters supporting reconstruction, trigger, and visualization workflows.

  • Standard StdHEP and HepMC structures, ROOT-based input conversion, and ROOT output utilities connect Delphes to common event-data workflows.The framework also supports LHEF and ROOT inputs through dedicated converter classes.
  • Generator-level, reconstructed analysis, and trigger-emulation data are organized in separate GEN, Analysis, and Trigger trees.The particle list for invisible objects is maintained through the PdgParticle class.
  • Detector response and analysis behavior are implemented through classes and parameters for smearing, track propagation, jet energy flow, tagging, triggers, and FROG visualization.The configuration includes controls for magnetic-field propagation, very-forward detectors, trigger selection, and event-display preparation.
  • Detector-card parameters configure calorimeter segmentation, isolation, jet reconstruction, b-tagging, thresholds, and very-forward detector settings.FastJet modifications support changes to calorimeter-cell patterns in (η, φ) space.
  • The detector card exposes constants for b-tagging and mistagging, while trigger, forward-detector resolution, and FROG settings are controlled through dedicated parameters.The listed b-tagging constants distinguish real b-jet efficiency from c-jet and light-jet mistag efficiencies.

A. User manual

The user manual describes Delphes as a packaged, configurable fast detector simulation whose detector card controls geometry, resolutions, reconstruction, and optional processing modules.

  • Delphes is distributed as a C++ package containing ExRootAnalysis, Hector, FastJet, FROG, and converters for standard event formats.The package assumes a running ROOT installation and includes the listed external components within the distribution.
  • A detector card configures coverage, resolutions, thresholds, jet algorithms, magnetic-field propagation, forward detectors, trigger selection, and output options.Six flags control magnetic-field propagation, very-forward simulation, forward taggers, trigger selection, FROG preparation, and LHCO output.
  • Without a user-supplied datacard, Delphes uses default smearing and running parameters.Users can replace these defaults by supplying detector and trigger cards for different configurations.
  • The default detector comprises a tracker, central and endcap ECAL and HCAL regions, forward calorimeters, and an enclosing muon system.Default coverage and resolution parameters are specified for these subsystems.
  • Calorimeter response uses parameterized electron/photon and hadron resolutions of the form σ/E = C + N/E + S/√E.Separate stochastic, noise, and constant terms are supplied for central, endcap, forward, and zero-degree calorimeters.
  • Delphes supports configurable jet reconstruction, b-tagging, trigger and event-display processing, and Hector-based very-forward detector settings.The manual provides parameters for jet energy flow, magnetic-field coverage, ZDC and Roman Pot locations, beam optics, and FROG event counts.

A.2.2. Running the code

Running Delphes requires detector and trigger cards, a homogeneous list of supported input files, and an output filename supplied to the executable.

  • Users first create detector and trigger cards, then prepare an input list whose files share one supported extension.Supported extensions include .lhe, .hepmc, .root, and .hep.
  • The executable is run with an input list and output filename, with detector and trigger cards available as optional arguments.The documented usage is ./Delphes input_file output_file [detector_card] [trigger_card].
  • Invoking ./Delphes without the required arguments displays the correct command usage.

A.3. Getting the Delphes information

Delphes stores generator, reconstructed-analysis, and trigger information in ROOT trees containing collections of physics objects, kinematics, detector-cell data, missing energy, and forward-detector hits.

  • The ROOT output is divided into GEN, Analysis, and Trigger trees containing generator data, reconstructed objects, and trigger results.Branches store related quantities such as particle energies, momenta, and pseudorapidities.
  • The Analysis tree includes collections of tracks, calorimeter cells, electrons, photons, muons, jets, and missing transverse energy.The corresponding branches use TRootTracks, TRootCalo, TRootElectron, TRootPhoton, TRootMuon, TRootJet, and TRootETmis classes.
  • Missing-transverse-energy records contain its azimuthal angle, magnitude, and transverse momentum components.
  • The output includes dedicated branches for ZDC, RP220, and FP420 hits, with ZDC records identifying whether a hit is hadronic.Hit positions are computed relative to the beam center.
  • Common particle records provide energy, momentum components, transverse momentum, pseudorapidity, and azimuthal angle.Specific electron, muon, and jet branches add isolation, calorimeter, track-count, and b-tagging properties.
  • Very-forward hit branches record detector side, time of flight, measured or smeared energy, and generator-level particle information.Roman Pot branches additionally provide detector position, hit coordinates and angles, and reconstructed momentum transfer q2.

A.4. Deeper description of jet algorithms

Delphes interfaces six jet algorithms through FastJet, covering cone and sequential-recombination approaches with different sensitivities to soft particles, collinear splittings, and computational speed.

  • Six jet algorithms are interfaced through FastJet: three cone algorithms and three sequential-recombination algorithms.All use calorimetric cells as inputs for clustering.
  • Cone algorithms: CDF Jet Clusters seeds jets from cells inside a circular cone above a transverse-energy threshold but is sensitive to soft particles and collinear splittings.It is described as a fast reconstruction algorithm.
  • Cone algorithms: CDF MidPoint reduces infrared and collinear sensitivities by adding energy-barycentre midpoints to the cone-seed list.
  • Cone algorithms: SISCone is insensitive to additional soft particles and collinear splittings while remaining fast enough for experimental analyses.
  • Sequential recombination algorithms: The three sequential-recombination algorithms are safe against soft radiation and collinear splittings, successively merging calorimeter-cell pairs.They differ mainly in the distance definitions used during merging, including pairwise and beam distances.
  • Sequential recombination algorithms: The longitudinally invariant kt and anti-kt algorithms are included among the sequential-recombination options; anti-kt produces exactly circular hard jets in the (y, φ) plane.

A.5. Running an analysis on your Delphes events

Delphes ROOT outputs can be explored interactively or analyzed with supplied and automatically generated C++ code. ROOT trees, branches, leaves, and trigger results are inspectable and drawable.

  • The Analysis_Ex.cpp example provides a simple way to analyze a Delphes ROOT ntuple, although ROOT TBrowser and MakeClass are also available.These analysis aids are optional and intended to facilitate access to the output.
  • Inspecting ROOT output: A Delphes ROOT file may contain GEN, Analysis, and Trigger trees, and GetEntries() reports the number of entries in a selected tree.The example output lists these three tree names and returns 200 Analysis entries.
  • Inspecting ROOT output: ROOT commands list trees, branches, and leaves, with wildcard filtering available for displaying selected parts of the output.The example filters leaves ending in .E.
  • Inspecting ROOT output: Individual leaves can be drawn through TBrowser or a tree’s Draw method, including trigger acceptance results.
  • Inspecting ROOT output: ROOT supports C++-syntax mathematical operations on multiple leaves, such as products and square-root expressions involving muon components.
  • Generating analysis code: MakeClass generates .h and .C files that provide access to all branches and leaves of a corresponding tree.
  • Running the example: Analysis_Ex requires an input list of Delphes ROOT files and an output ROOT filename, and the resulting source can be edited, modified, and compiled.The documented command is ./Analysis_Ex input_file.list output_file.root.

A.5.1. Adding the trigger information

Delphes provides separate tools for trigger processing, event visualization, and LHCO-based inspection of reconstructed objects. The ROOT output supports richer analysis, while LHCO is compact but data-limited.

  • Adding trigger information: Trigger_Only runs trigger selection separately from general detector simulation on a Delphes ROOT file and appends a tree containing trigger-result data.It requires the ROOT input file and a trigger datacard.
  • Visualizing events: When FLAG_FROG is enabled, Delphes creates .vis and .geom files containing the information needed to run FROG.FROG is compiled once for unchanged geometry and visualization files, then launched from the Utilities/FROG directory.
  • LHCO output: LHCO is a text-ASCII format containing only final high-level objects, arranged in columns with one row per object and comment lines beginning with #.
  • LHCO output: LHCO rows encode event and object information through line number, type, pseudorapidity, azimuth, transverse momentum or energy, invariant mass, tracks, b-tagging, and additional columns.The event ends with a missing-transverse-energy row of type 6.
  • LHCO output: LHCO records track counts and b-tag information, while the ninth column stores calorimetric energy ratios or muon-isolation quantities depending on object type.The final two columns are currently unused.
  • LHCO output: LHCO contains only a fraction of available data and can incur larger file sizes and longer creation times, so ROOT output is preferred when possible.
Loading 0903.2225v3…