Motivation
- Public datasets generally focus on “high-level” objects
- Jets, electrons, muons, etc
- This makes perfect sense for general use for most education
- However, this isn’t sufficient for teaching about reconstruction
- How and why do we use certain reconstruction techniques?
- What are the trade-offs of using different types of detectors?
- How does pileup impact object reconstruction?
- We propose a new dataset which can be used to teach jet reconstruction
- Most important contents: topoclusters and tracks (per-vertex)
- Sample: Pythia8 dijets, ∼100k events/slice =⇒ O(1M) events
- Merge slices to one sample, flat in pT
Examples of studies supported by such a dataset
- Small-R pileup studies:
- How does jet multiplicity depend on pileup?
- How can tracks be used to suppress pileup jets?
- Why don’t we build jets from tracks instead of clusters?
- Large-R substructure studies:
- How does the jet algorithm impact the event interpretation?
- How does the jet algorithm impact the jet mass?
- How does the jet algorithm impact pileup stability?
- Many other interesting educational studies are possible
- Such a dataset would have a large variety of pedagogical uses!
- Lots of possible extentions if this is well received
Dataset overview
- The dataset is now ready
- Ten files of 102.9k events each (total 1.029M events), total 21 GB
- Each file is 343 bins with 300 events each, and is 2.1 GB
- Also this is only MC, no data (more on this later)
- You are welcome to check out the dataset
- One of the ten files is here (password is “JetReco”)
- A pair of exercises are also available here (password is “JetReco”)
- Following slides contain:
- Details on the dataset contents
- Examples of what can be done with these files (from the exercises)
- A discussion on choices made and possible follow-ups
Details on the dataset contents
Dataset contents, high level
- This is the high-level overview of the contents
- A detailed overview of each aspect is on the following slides
- General event-level information (NPV, etc)
- Tracks, with standard jet selections applied
- Clusters, with standard jet selections applied
- Truth particles, with standard jet selections applied
- Various types of reconstructed jets
- All reconstructed jets can be rebuilt from the dataset
- They are included to help the user get started and for validation
Dataset contents, event-level info
- EventNumber [unsigned long long]
- RunNumber [unsigned long long]
- Not very useful for MC, but useful in case data is added later
- EventWeight [float]
- Complex quantity, described later, but it is the only weight needed
- mu average [float]
- mu actual [float]
- The same as mu average in MC, but for data we may want both
- NPV [unsigned int]
- Typical definition used for jet calibration: require two tracks
- Note that some common variables are NOT included
- No mcChannelNumber as we merged several samples together here
- No individual weights as it’s all stuck together into EventWeight
- No pileup weight - not needed for MC only
Dataset contents, tracks
- Jet reco in ATLAS applies some standard selections to tracks
- pT > 500 MeV, good quality, etc
- These are also applied to the tracks in this dataset
- Vertex matching done using standard track-to-vertex association tools
- Tracks pt [vector]
- Tracks eta [vector]
- Tracks phi [vector]
- Tracks m [vector]
- I checked, it’s always fixed to the pion mass as expected
- ROOT compresses by factor of 2 compared to pt/eta/phi
- Supports using the same commands regardless of the object type
- Tracks vtx [vector]
- User has to loop over tracks and identify “vertices” using these indices
- Decided to do this to save space rather than storing vertices
- int not unsigned as tracks not associated to a vertex have index -1
Dataset contents, clusters
- Jet reco in ATLAS applies a cluster energy > 0 cut
- This is applied to the clusters written out here
- They are all written out at LCW scale (mitigates lack of jet calibration)
- Clusters pt [vector]
- Clusters eta [vector]
- Clusters phi [vector]
- Clusters m [vector]
- I checked, and it’s always zero as expected
- ROOT compresses by factor of ∼90 compared to pt/eta/phi
- Supports using the same commands regardless of the object type
Dataset contents, truth particles
- Jet reco in ATLAS applies a filter on truth particles
- Only stable particles (cτ > 10 mm), excluding muons and neutrinos
- This is applied using standard ATLAS tools here as well
- Particles pt [vector]
- Particles eta [vector]
- Particles phi [vector]
- Particles m [vector]
- Particles pdgID [vector]
- Allows for studies of pions vs kaons vs etc
- Not strictly needed, but useful and could support other studies
Dataset contents, reconstructed jets
- Jets are stored with moderate pT cuts to save space
- They can be rebuilt down to arbitrarily low pT with the input objects
- Saved without any calibration: students can perfectly rebuild them
- RecoJets R4 {pt,eta,phi,m,jvf}: [vector]x5
- Built from topoclusters, stored for pT > 15 GeV
- TrackJets R4 {pt,eta,phi,m}: [vector]x4
- Built from tracks from the leading PV, stored for pT > 10 GeV
- RecoJets R10 {pt,eta,phi,m,D2beta1,tau32wta}: [vector]x6
- Built from topoclusters, stored for pT > 150 GeV
- RecoJets R10 Trimmed {pt,eta,phi,m,D2beta1,tau32wta}: [vector]x6
- Stored for pT > 150 GeV on the ungroomed jet (keep index parallelism)
- TruthJets R4 {pt,eta,phi,m}: [vector]x4
- Built from truth particles, stored for pT > 5 GeV
- TruthJets R10 {pt,eta,phi,m,D2beta1,tau32wta}: [vector]x6
- Built from truth particles, stored for pT > 100 GeV
- TruthJets R10 Trimmed {pt,eta,phi,m,D2beta1,tau32wta}: [vector]x6
- Stored for pT > 100 GeV on the ungroomed jet (keep index parallelism)
For more information click here
Disclaimer
This dataset is provided by the ATLAS Collaboration only for educational purposes and is not suited for scientific publications.
- The ATLAS Open Data are released under the Creative Commons CC0 waiver.
- Neither ATLAS nor CERN endorses any works produced using these data, which is intended only for educational use.
- All data sets will have a unique DOI that you are requested to cite in any (non-scientific) applications or publications.
- Despite being processed, the high-level primary datasets remain complex, and selection criteria need to be applied in order to analyse them, requiring some understanding of particle physics and detector functioning.
- The large majority of the data cannot be viewed in simple data tables for spreadsheet-based analyses.
