Jet-ID Projects


Satus color chart: Red = Unmanned, Blue = On-going, Green = Done
Main Project:  Jet-Id certification

Other Projects:
1 - L1 confirmation
2 - Jet-ID efficiency at low pT and performance at high luminosity
3 - Jet-ID efficiency in the ICR optimisation
4 - Jet-ID for jets with tracking information
5 - Jet-ID using detailed calorimeter information
6 - Jet-ID for b-quark jets
7 - Jet-ID efficiency in the forward region
8 - Additional jet information in CAF
9 - Properties of cone jet algorithms in hadron colliders
 
Jet-ID certification

In an ideal world, there would be no fake jets due to noise or hardware problems in the calorimeter or pile-up events at high luminosity, the D0 detector simulation  would just be perfect and everyone would just apply jet algorithms on Monte Carlo. In short,  there would be no Jet-ID group!

In the real world:
- Jet-ID criteria are needed to reject fake jets and select good jets with the maximum efficiency,
- Jet-ID differences in efficiency between data and Monte Carlo need to be understood or, at least, determined and corrected for.

This is what we call Jet-ID certification.

The Jet-ID criteria & efficiencies must be studied (by order of priority)
  1. for p17 data in CAF,
  2. using an automated process, i.e. a new (to be written) jetcert package in the D0 framework which will
    1. determine the efficiency scale factor needed to correct for differences between data and Monte Carlo,
    2. provide a certified way of downscaling efficiency for Monte Carlo jets (to be included later on in d0correct)
  3. more specifically in the ICR, where standard Jet-ID cuts are far from optimal,
  4. for different types of jets, depending on which tracking information is used (with or without vertex-track confirmation, taggable jets, Eflow jets),
  5. as a function of more detailed jet (and event) characteristics (e.g. number of associated tracks, jet shape, proximity to the closest other jet and so on...),
  6. for b-quark jets,
  7. in the forward region,
  8. for R=0.7 cone jets,
  9. for kT jets.

Manpower & time scale:
For items 1&2: Jens Konrath, Bertrand Martin, Murilo Rangel, Helio Nogima, Rene Luna-Garcia
For item 3: 1 Ph.D/postdoc x 2 months, see also Project 3 below: ==> (Amnon Harel)
For item 4: 1 Ph.D/postdoc x 2-3 months, see also Project 4 below ==> (Petr Vokac) Jens Konrath
For item 5: 1 Professor/Ph.D./postdoc x 6 months, see also Project 5 below ==> please volunteer!
For item 6: 1 Ph.D/postdoc x 2-3 months,  see also Project 6 below ==> please volunteer!  (but note that this project can start only when Project 4 & Project 5 are completed)
For item 7: 1 Ph.D/postdoc x 2 months, see also Project 7 below ==> Yann Coadou
For items 8,9: 1 Ph.D/postdoc x 1-2 months/each project ==> please volunteer!

Contacts: Bernard, Amnon, Nils Gollub, James Heinmiller, Ariel Schwartzman (for new jet-vertex cuts).

L1 confirmation implementation

Current L1 confirmation is arguably the most important jet quality cut. It performs well and is relied on in all D0 analyses. It will retain its crucial role in the future, as track matching provides mainly complementary information. However, several shortcuts and arbitrary decisions had to be made by the original implementors of L1 confirmation in order to prove its viability within their constraints. Hence, the existing implementation contains several significant approximations, which can (and should) be avoided. They result in the large spread of the jet's L1 readout vs. the jet's relevant pT. A good yardstick for the problems in the current implementation is also that L1 remapping in the ICR did not improve the performance. We expect that a much tighter spread is achievable, that the ICDs can be gainfully used and that the dependance on the event's activity can be removed.

We have thus identified several issues where the implementation should be improved, which should result in better jet quality in all the D0 analyses using jets, especially for those where high missing ET is required. Here they are in a rough order of importance:
The final stage of improving the confirmation is choosing the official cut.


Manpower & time scale: 1 Ph.D./postdoc x 3-6 months ==> Robert Wagner

Contacts: Amnon, Bob Kehoe, new L1Cal leader.

Jet-ID x reco efficiency at low pT
and performance at high luminosity

Jets at low pT (< 15 GeV) are of prime importance in many physics channels, most importantly in Top and Higgs physics. Potential sources of inefficiency have been identified, as detailed in D0-note 4457, the main ones being:
  1. cut at pT_min at the end of jet finding (pT_min = 8 GeV in p14, 6 Gev in p17)
  2. cut on pT of proto-jet during proto-jet formation (0.5 x pT_min)
  3. cut on pT of the tower used for seed (Simple Cone) formation (0.5 GeV)
  4. cut on pT of the seed used for proto-jet formation (1 GeV)
  5. cut on the minimal distance between a seed used for proto-jet formation and already found proto-jets (0.5 x R_cone)
While cut 1 is in some way unavoidable (even if its value can be adjusted to the lowest pT to be studied), it should be noted that cuts 2&5 are specific "features" of the RunII Cone jet algorithm used by D0, compared to the algorithm agreed upon with CDF in RunII workshop (see hep-ex/0005012). Cut 2 during proto-jet formation is particularly dangerous, because good low pT proto-jets may not be formed, which could either be merged to higher pT proto-jets or could be used to define valid midpoints with other proto-jets.We thus might want to remove these two cuts, if possible.

While it is desireable to try to improve jet finding efficiency and energy measurement at low pT, this task is more and more difficult as the luminosity (and thus the average number of interactions per bunch crossing) rises, with the risk of being overwhelmed by fake jets. It could thus be helpful for performance at high luminosity to introduce a new cut at the end of proto-jet formation, in order to avoid excessive merging of many low pT fake proto-jets into a fake jet. This cut was used at RunI and was recommended in RunII workshop (see hep-ex/0005012), but was not implemented in RunII cone algorithm for some (unknown) reason. Note that this new cut is similar, but not strictly equivalent, to cut 2 (cut on pT of proto-jet during proto-jet formation at 0.5 x pT_min) mentioned above: cut 2 implies this new cut, since it means that a proto-jet of pT < pT_min/2 (4 GeV in p14) cannot be formed, but the reverse is not true. This new cut should only be applied on proto-jets once they are formed and have been already used to define midpoints, in order to decide which ones should enter the merging/splitting step.

There are several options which can be considered and studied, not just applying a cut on pT of the proto-jets at the beginning of the merging/splitting step. For instance, it seems valid to try to merge a proto-jet, even of very low pT, with a proto-jet of sufficient pT, since the former is then most probably the tail of a good jet. Conversely, merging two low pT proto-jets should be avoided. So the cut could be a function of the pTs of both proto-jets that enter a merging/splitting step or of their ratio or of any useful function of these. One can also imagine many more options if the simple ones are not sufficient.

Eventually, depending on luminosity, a compromise might have to be found between  the jet finding efficiency at low pT, the computing time per event and the contamination of fake jets, formed from an excessive merging of very low pT energy clusters in pile-up events.


Manpower & time scale: 2 Ph.D./postdoc x 2-3 months
Improvement of Jet-ID x reco efficiency at low pT ==> unmanned (Marine Michaut made already significant progress in the past)
Performance of Jet-ID at high luminosity ==> unmanned (Marc-Andre Pleier made already significant progress in the past)

Contacts: Bernard, Emmanuel Busato, Ariel Schwartzman (for vertex-track confirmation of jets at high lumi).

Optimize Jet-ID cuts in the ICR region

The current jet-id cut of EMF>0.05 cuts away too many good energetic jets with 1.0<|eta|<1.5, as can be seen in Christophe's unfinalized note (bottom of fig. 24, right now). This is since the hard jets punch through the few EM layers and leave there a small fraction of their total energy. The CHF<0.4 is also problematic in the ICR. The jet-id in the ICR needs to be reexamined (regardless of track matching).

Manpower & time scale: 1 student / Ph.D. x 2 months ==> Amnon Harel (for p17)

Contacts: Amnon, Bernard, Mikko Voutilainen, JES conveners, Lee Sawyer, Ariel Schwartzman.

Study Jet-ID for jets with tracking information

The use of tracking information will be more and more important for all jet measurements (vertex-track confirmed jets, taggable jets, Eflow jets...).  Of prime interest in view of high luminosity running is the vertex-track confirmation of jets, i.e. the confirmation that a given jet comes from the primary vertex. Since this will be probably used by default in most, if not all, forecoming analyses, the determination of the efficiency of this vertex-track confirmation should be considered an integral part of Jet-ID certification. It can be done by using back-to-back dijet events, where both jets are good jets (from the Jet-ID point of view) and one of them, which is vertex-track confirmed, is used as a tag while the other serves as a probe.

The tracking information could also be used to provide a more detailed description of Jet-ID efficiency. This is equivalent for the tracking to the project 5: Jet-ID using detailed calorimeter information explained below for the calorimeter. Two examples of studies that could make a profiteable use of tracking information are given here:
Eventually, the tracking information will also prove useful to study b-quark Jet-ID efficiency (see below project 6: Jet-ID for b-quark jets).


Manpower & time scale: 1 student / Ph.D. x 2-3 months ==> Jens Konrath (Petr Vokac made significant progress in the past)

Contacts: Amnon, Bernard, Ariel Schwartzman.

Improve Jet-ID efficiency description using calorimeter information 

In Run I, it has been observed that the Jet-ID efficiency for low pT jets can be lower in "hard" events (where these jets happen mainly by radiation from the leading jets) than in "soft" events (where these low pT jets can be leading jets), see D0-note 3324 figs. 5 and 8 and section 2.1. In Run II, it has been observed that the Jet-ID efficiency in Monte Carlo is different if jets have undergone or not splitting and merging (see D0-note 3985, fig. 18 p.37). It is also known that the energy shower development of jets measured with the calorimeter in data is not reproduced in Monte Carlo simulation (e.g. the distribution of energy amongst layers has been shown, in JES group, to differ between data and MC even when the jet pT distributions agree, as noted by Christophe Royon).

At the moment, the Jet-ID efficiency is parameterized as a function of pT and eta only, so these dependencies are not taken into account. Detailed Monte Carlo studies are thus needed to understand these observations and to improve the Monte Carlo agreement with data, with the ultimate goal of applying directly Jet-ID on Monte Carlo. If this best solution cannot be reached soon, a "physics sample dependent" Jet-ID efficiency should be determined in the meantime, in order to provide a more refined description of the behaviour of Jet-ID efficiency in data.

A first (non-exhaustive) list of studies which seem the most promising towards the defined goals is given below:
The studies listed above are based on calorimeter information only. Similar studies based on tracking information are part of a more general project devoted to the use of tracking (see above project 4: Jet-ID for jets with tracking information). At the end of the day, the results of both studies should be combined to make an optimal use of all information available, one of the applications with highest priority being to obtain a Jet-ID efficiency parameterization specific to b-quarks (see below project 6: Jet-ID for b-quark jets).


Manpower & time scale: 1 Professor/Ph.D./postdoc x 6 months ==> Unmanned

Contacts: Bernard, Amnon, Thorsten Kuhl & Christian Zeitnitz (for Geant tuning).

Study Jet-ID efficiency for b-quark jets

The Jet-ID efficiency for b-quark jets has been shown in Monte Carlo to be (slightly) different from the Jet-ID efficiency for light quarks (see D0-note 4479, section 5.1 and fig. 4). Note, however, that this study was made with a Monte Carlo simulation known to have, for instance, a value for the e/pi ratio lower than the one observed in data. Before this observation of a difference between Jet-ID efficiency of b-quarks compared to light quarks jets can be confirmed on data, detailed studies are thus needed to improve the understanding of the detector and of its Monte Carlo simulation. Two projects have been defined for that purpose, one using tracking (see above project 4: Jet-ID for jets with tracking information)  and one using calorimeter (see above project 5: Jet-ID using detailed calorimeter information).  Once these studies will be completed, their results should be applied on the more specific case of b-quark jets in data, first to confirm the observation made on Monte Carlo, then to obtain a Jet-ID efficiency parameterization specific to b-quarks, if needed.


Manpower & time scale: 1 Professor/Ph.D./postdoc x 2-3 months ==> Unmanned

Contacts: Bernard, Amnon, b-ID conveners.

Verify (optimize?) Jet-ID cuts in the forward region

The JES will be extended to at least |eta|<=3.0, official jet-id eff. should follow suit. Do the current cuts make sense there (both for low pT jets used in single top, and for high pT jets used in NP)?


Manpower & time scale: 1 student / Ph.D. x 2 months ==> Yann Coadou

Contacts: Amnon, Bernard, JES conveners.

Store new information about jets in CAF

Two new types of information will be stored in CAF:
  1. Information about the geometrical shape of jets:

    By "shape of a jet" we mean its exact geometrical description, in other words the list of towers of which the jet is made. Typical examples from various jet algorithms can be seen here.
    This will be useful for non-conical jets, that is, not only kT jets but also a significant fraction of the cone-jets, in the contexts of track matching, jet-track matching (e.g. taggability and b-tagging), L1 confirmation, muon isolation, out-of-cone showering and whatever else analysers will come up with once the information is available.

  2. Information about the jet shapes:

    By "jet shapes" (not to be confused with previous term...) we mean its usual meaning in QCD literature, i.e, variables which describe the repartition of transverse energy density inside the cone.
    This will be useful for detailed comparison with detector simulation and for QCD studies . This might also be useful as new variables for jet-ID efficiency and jet energy calibration, and for quark-gluon separation in physics analyses.

Manpower & time scale: 1 student / Ph.D. x 2 months ==> Zdenek Hubacek

Study detailed properties of cone jet algorithm in comparison with CDF

Since RunI, CDF has observed what they call "dark jets" and what should be called precisely "large, collimated and unclustered transverse energy". This is why they introduced "ratcheting" (i.e. keeping any tower in a proto-jet as soon as it has been part of the corresponding cone during the search for a stable cone process) in RunI and "smaller search cone" (i.e. finding proto-jets or stable cones using a smaller radius than the one used to actually build the final proto-jet once the stable position is found) in Run II.

More recently (e.g. in Tev4LHC workshop) CDF reported the observation of "fat jets", i.e. jets with very large spatial extension formed from the merging of many proto-jets.

Even if these "problems" seem to occur at a much lower rate in D0, there is in principle no reason why these should not be observed at all. At least, the situation must be clarified and the origin of the differences (if any) between D0 and CDF must be understood. This work is also of importance in view of the definition of jet algorithms for LHC experiments.


Manpower & time scale: 1 student / Ph.D. x 2 months ==> unmanned (Zdenek Hubacek made significant progress in the past)

Contacts: Bernard, Amnon, Michael Begel, Emmanuel Busato, Alexander Kupco, Markus Wobisch, .

This page is maintained by Bernard Andrieu
Last modified: March 30th, 2006