r28 - 12 Jun 2006 - 17:41:28 - GeorgeRedlingerYou are here: TWiki >  Projects Web > SusyRoadMap

Working Document to define roadmap for SUSY analysis

I'm no longer updating this page. I'm finding editing wiki pages to be painful; I much prefer to edit web pages with a WYSIWYG web page editor. Further updates will be at http://www.phy.bnl.gov/~redlinge/Public/susy/notes.html (GR)

Signal Search Strategy

The goal is to assess the current readiness for an inclusive search with the first 100 pb-1. Two possible methods are initially considered for having the best reach in SUSY space, especially with a limited amount of data: a) Jets + EtMiss, b) Jets + EtMiss + 1 lepton. Tri-lepton signatures have not received much attention and should also be looked at in preparation for larger integrated luminosities. For signal characteristics/acceptance, we presently focus on the SU1 (point in the co-anhillation region) and SU3 points (bulk region). More details of the mSugra SUSY points can be found here. GMSB models have gotten less scrutiny; attempt by Jinnouchi to revisit these models (Apr 06) is described here.

A list of studies to define the search strategy include:

  • What is the signal region? High Meff? High Etmiss? Box in the Meff/Etmiss plane? Other?
    • LBL group starting to look at strategies to optimize inclusive SUSY search. Talk by S.Vahsen from Apr 06 is here.
    • Study by J.Conrad and F.Tegenfeldt on signa/bkg discrimination using RuleFit. Writeup is hep-ph/0605106
  • It would be good to do a blind search (example: jets+Etmiss search at CDF). What are the subtleties?
  • How will the background levels in the signal region be estimated? We should assess how far we can go in estimating the backgrounds from the data with minimal reliance on Monte Carlo.
    • Z (--> nu nubar) + jets
      • Tag with Z --> ee (mu mu) +jets in low Etmiss region; discard leptons to extrapolate to signal region. This only accounts for the contribution of Z->nu nu to the Etmiss distribution. Other sources of Etmiss such as detector pathologies or neutrinos in jets would not be accounted for.
    • W (--> l nu) + jets
      • Use Z--> ll + jets, discard one lepton; correct for W/Z cross section ratio (measured separately) and differences in acceptance (from Monte Carlo).
    • QCD multijet
      • Not straightforward; needs thought. Looking for rare mismeasurement effects of something with a huge cross section (or the presence of neutrinos in jets).
      • One possible approach: Measure multi-jet cross sections at low Etmiss. Measure jet mismeasurement characteristics with co-planar dijets. Convolute to extrapolate into the high Etmiss region.
      • Neutrinos in jets? Need b/c jet tag with leptons. Heavy-flavor content in multijets could be different than in dijets....
    • ttbar
      • Isolate top with mass constraints (semi-leptonic decay mode). In principle one an measure the top background level directly in the signal (high Etmiss) region. However, this top sample will also have non-top combinatoric contribution. To estimate its size, must go off top mass peak in low Etmiss region.
      • See Dan Tovey talk here
      • If SUSY decays have lots of top in them (e.g. SU2, focus point region), this method will not work.....
    • Dibosons
      • Small? Included automatically in the Z/W+jets measurement from the data?
      • Measure with fully leptonic modes?
    • First look at a data-driven approach to bkg estimation, using "dual cut" approach is described by G.Redlinger here.
  • Relative merits of 0-lepton and 1-lepton modes.
  • Should we be considering b-tagging tools? Prejudice is that b-tagging will not be available early; could it be that b-tagging will be easier to understand than calorimeter performance?

The following are some initial steps to take in studying the backgrounds


  • Studies with existing Rome Sherpa Z+jets samples.
    • Z(ee)+jets studied by G.Redlinger. Low electron reconstruction efficiency (in the 50% range). Known problem shown by O.Jinnouchi and T.Lari. Claim by B.Mellado was that MC Truth was the problem; doesn't really seem to hold up... Efficiencies in ESD also bad (Redlinger), but CBNT reportedly ok (Mellado,Jinnouchi).
    • Test of bkg estimation technique using Z(ee)+jets sample, removing reconstructed leptons. No statistics after application of simple SUSY cuts. Need to enhance with generator level cuts; looks like this is being done better for CSC samples. However, in the interim, distributions of Etmiss and Meff were checked with looser cuts on #of jets and jet pt. Zee provides a good estimate of Etmiss and Meff in Znn, but only after removing calorimeter pathologies.

  • Studies with ALPGEN. Japanese group have produced a sample of Z+jets with ALPGEN.

  • Produce evgen files (running through Herwig?) with the above ALPGEN output. Currently need to switch off underlying event; understood? See talk by Jinnouchi here

  • For the short term study while waiting for the full simulation production, run fast simulation on above files. (Jinnouchi et al). Status as of Apr06 is here. Wiki page.

  • What studies with fast simulation output?


  • Studies again need to be performed with ALPGEN. For the short term, use the above Z+jets and throw away a lepton to study W+Jets background. Longer term, we need to pursue studies with full simulation of W+Jets

  • Define specific studies with names here.

QCD multijet

  • Studies with ATLFAST transferring calorimeter resolution functions from G4. (A. Gupta). Events with pt(hard) as low as 250 GeV contribute to the background. Not enough statistics to study the impact of jets with pt lower than this. A talk is here.
  • Study the multijet+Etmiss convolution technique.
  • Reassess standard SUSY cuts, especially on hardest jet Et. May want to change it to scale with Meff.
  • How many jets have to be mismeasured in an event to create a fake signal?


  • Independent study of the technique. (J.Cochran, F.Tegenfeldt, J.Conrad). See this talk. Concludes that the technique should work but points out that the method relies on good JetE and Etmiss resolutions; plan to study the effect of sub-optimal resolution. Also making a study of other variables that distinguish SUSY from top. Using RuleFit. Initial results suggest roughly factor of 5 loss in top rejection for a fixed signal (SU1 and SU3 were studied) acceptance.
  • Errors on background estimates from uncertainty in the energy scale.


  • Define studies

Simulation and Digitization

Monte Carlo Event Generators

  • Some subtleties in combining matrix element and parton shower generators. Is it under control? (S.Asai? F.Paige?)

Detector Simulation

  • Understand the large negative energy in LAr hits in strips that causes significant Missing ET tails.
    • Problem seems to have gone away in CSC data samples as reported for example here

  • Better fast simulation would be desirable for high statistics studies (e.g. QCD multijets)
    • Improve ATLFAST
      • ATLFAST performace (circa Mar 06) summarized in this talk by K. Jakobs.
    • Speed up the full simulation
      • Parametrization of e/gamma showers, see here. Parametrizations are available for e/gamma showers in the region abs(eta)<3.2, based on hep-ex/0001020. Factor of 5-10 speedup over full sim. See slides for comparisons of shower profiles; looks like it still needs some tuning....
      • Shower libraries for jets? Was used in D0, vague memories of factor O(100) speedup. (S.Snyder? J.Cochran?)

  • Need tool to merge Monte Carlo data with real data (K. Cranmer)



  • Lots of work on calibrations by others. Possible areas we could focus on:
    • Understand the impact of miscalibrations on S/B.
    • Evaluation/validation of calibration. Need to come up with a "figure of merit" variable for each calibration to assess independently how well the calibration has been done.
    • Understand which issues are specific to SUSY as opposed to Atlas in general.

EM Calibration

  • Inter-region calibration with Z->ee
  • J/psi calibration?

Hadronic Calibration

  • Local Hadronic Calibration (MPI)
    • Status as of Mar 06 described here by P.Schacht.
  • E/p with single pions?
    • Alberta group looks into E/p of isolated tracks in tau decays.

In-Situ Calibration

  • dijet balance (who?)
  • gamma+Jets (Barcelona)
    • Status as of Mar 06 in talk by S.Jorgensen. Focuses mainly on studies of underlying event.
  • Z+Jets (Proudfoot)
  • W->jj from ttbar (Clermont Ferrand)

Reconstruction software

Analysis Tools

  • Migration of SUSYPlot to use EventView
    • Some suggested links to get started:
    • More recent developments are described here and at this tutorial
    • EventView? customised for SUSY analysis: SUSYView

  • Use of Distributed Analysis for job submission (T.Maeno, K.Assamagan).


  • soft electron optimization (J.Cochran?)
  • Offline efficiency, optimize isEM()
  • Use of Likelihood (K.Benslama) or H-matrix (Hyeon Jim Kim) approach for e selection
  • identify conversions. (D. Joffe, "early" conversions; somebody needed for "late" conversions?)
  • incorporate brem-recovery (nobody working on this?)
  • use of topo clustering
  • incorporate e-gamma corrections. Some early work by S.Snyder.
  • underlying event contribution to electrons (postpone for later)


  • soft muon algorithms. K.Assamagan mentions MuonBoy already handles soft muons.
  • Comparison of different algorithms: MOORE, MuonBoy?
    • First look by K.Assamagan (Mar 06) is described here


  • What is the optimal cone size for jet finding in SUSY events?
  • Comparison of different algorithms, impact on energy resolution, Etmiss tails


  • Study tau1p3p for soft taus (K.Assamagan?)
  • A study by Oye et al. of soft tau reconstruction in co-annihilation region ATL-COM-PHYS-2006-030


  • Workshop at BNL 12/20-21/2005
  • Comparison of two B-tag algorithms (lhSig vs IP3D? +SV1) on Rome data for a variety of samples (ttbar, SUSY SU2 and W+jets) can be seen in this talk by T. Lari. Rejection of light jets in SUSY is only about a factor of 1.3 worse compared to ttbar.

Missing ET

  • Use of topoClusters with local had calibration for Missing ET calculation. (D.Cavalli)
  • Object-based Etmiss (Wisconsin group). Talk by B. Mellado from Sept 05. Not really clear how this compares in performance with more standard Etmiss techniques in Atlas. A writeup is reportedly in the works.
  • Recognize and correct for pathological events that contribute to fake Missing ET: for e.g. bad cells, lost e or mu's, etc.
  • Explore other noise suppression techniques
  • Cryostat corrections (K. Cranmer has some info on it)
  • Study the tradeoffs between Z+jets versus Z balance with sumPt for Etmiss studies

Performance Studies

  • Long list of Jet/Etmiss performance notes that are to be produced with CSC data. See this wiki page

  • EtMiss Resolutions
    • Use of topocluster cells improves Etmiss resolution over simple sum of calo cells above 2-sigma threshold. See plot in this talk. Typical improvement of 1 GeV? in sigma(Etmiss).
    • Situation with Rome data and release 10 summarized by D.Cavalli
      • Etmiss resolution goes roughly as 0.48*sqrt(sumEt)
      • Conclusion on Etmiss tails is that after fixing pathological LAr cells and fake Moore muons dominant problem is energy loss from jets... Needs larger jet samples for detailed study.
      • Etmiss resolution depends not only on SumEt but on physics process?? Due to underlying phi asymmetry?
      • Problem with energy calibration in low-Etmiss region; differences in H1 calibration with topocluster-based calib

  • Jet energy resolutions

  • electron efficiency and rejection vs selection cuts
    • Optimize selection cuts
    • Study dependence on jet multiplicity (using Z+jets sample)
    • Status as of Jan 06 summarized by K.Benslama at LBL BSM workshop.
      • Rome analysis. Full simulation of 25 GeV? electrons (single particle). Background sample = QCD dijets with pt > 17 GeV? . No TRT info available in Rome data.
        Selection efficiency jet rejection
        "tuned" isEM 0.83 1.2 E5
        Likelihood 0.83 1.2 E5
        Neural Net 0.90 1.2 E5
      • (Would be nice to see as a function of pt and eta)
      • R vs A curves?
      • Need to find results from more complicated samples (top, W/Z+jets)
    • H-matrix technique studied by Hyeon Jin Kim. Some performance comparisons shown at Physics Week Oct 2005. See here. Unfortunately difficult to interpret because comparisons are not made at a fixed value of acceptance or rejection. More recent work focuses on 10x10 H matrix; report from Mar 06 is here. Claim is H10 has slightly superior performance, but not so obvious since comparisons are again not made at a fixed value of acceptance or rejection. Could also be some bias problems with low statistics? Need to recalibrate H matrix for CSC data.

  • muons

  • same as above for tau's, jets and b's.

Trigger software and Performance

  • Study trigger configurations and rates/acceptance
    • What is the current proposed trigger mix? (K. Cranmer)
    • What additional trigger configurations are required?
    • Are prescales/cuts optimal for early low-lumi running?
    • All these studies require trigger simulation
      • Calorimeter trigger simulation tutorial from Jan 06. Wiki page.
  • Develop trigger software algorithms
  • Which are the important triggers for signal? for background samples?

  • Status of trigger analysis tools
    • L1 simulation seems to be well advanced. A variety of software classes exist as of Oct 05; see talk by S. George. However, only available on ESD's. AOD tools supposed to be available in release 11. It turns out that even at the ESD level only a small subset of the Rome samples have L1 information; available samples described here. Other samples will require running private reconstruction following the instructions here. Probably more useful to start with latest tutorial
    • Major step forward expected starting with Monte Carlo samples produced with release 11.0.5. Significant trigger information becomes available in the ESD/AOD's without having to run reconstruction privately. See this talk from Mar 06 by A. Krasznahorkay for info on the trigger EDM.
  • L1 Status
    • D.Damazio studying electron trigger in L1. Report from Jan 06 is here. In minbias data he sees a trigger rate of ~20 kHZ for the L1 e25i trigger (at what luminosity?? assuming 1 minbias event per crossing, so something like 5E32). Clusters are present in the calorimeter, but there seem to be no MC truth particles pointing to these clusters. Doesn't seem consistent with this study where the e25i rate at L1 is 5 kHz for Rome dijet (pt>17 GeV) data.
    • High L1 electron trigger rate was attributed to the Geant4 bug described here. However, a subsequent study after the bug fix suggests that the problem has not gone away... Hmmm....
    • A few plots from Jan 06 on egamma trigger performance in W->e nu in talk by A.Watson
    • Limited study of L1 jet rates in talk by R.Stamen. J4 sample, release 11.0.3, J180 rate estimated at 100Hz for L=10^33 but neglects feed down from lower Et samples. Would be nice to have a plot of jet trigger rate versus Et threshold....
    • L1 Etmiss rates in talk by S.Rieke. Rome min bias data sample, release 11.0.3, no pileup. Rates of 40, 10, 1, 0.1 MHz for thresholds of 0,5,10,20 GeV? , then starting to run out of statistics. (Recall, max. allowed L1 Accept rate is 75kHz.) A luminosity of 2E33 is assumed here.
    • Apr06 study by G.Redlinger using CSC 11.0.41 samples (minbias+dijets) is here. Scaled to 2E33, the L1 Etmiss rates are 20 and 1.6 MHz at 10,20 GeV? thresholds respectively. Need to resolve the difference with the Rieke study...
    • L1 muon rates/efficiency in barrel discussed by F.Conventi.
    • L1 barrel muon trigger writeup: ATL-COM-DAQ-2006-022
  • Level 2 and EF
    • Broad overview as of Oct 05 in talk by C.Schiavi
    • Egamma
      • e25i trigger optimization studied with single electron sample for signal and dijet sample for bkg. Status as of Oct 05 summarized by C.Santamarina-Rios. See efficiency vs rate curves in the talk. An update (Mar 06) is here; includes results of efficiency studies for physics channels such as Z->ee, W->e nu, H->gamma gamma, H->4e.
      • Automated tools for trigger optimization starting to become available: PESAsim
      • Status of rates from jets and eff for single electrons as of Feb 06 summarized by M.Wielers. Explored use of TRT high threshold hits in L2/EF; not a big effect. Use of updated filter efficiencies results in 25% increase in trigger rate compared to Rome; still under investigation. 60-70% of the trigger rate comes from real electrons; not much room for big gains.
      • Study of egamma trigger efficiency for SUSY (bulk region) by I. Aracena from Mar 06 is here. 90% trigger efficiency for electrons which pass the offline cuts; pt=[25,150]GeV, limited statistics. Absolute trigger eff. (L1+L2+EF) is ~58%; see the transparencies for the breakdown at different trigger levels.
    • Muons
      • Status as of Oct 05
        • L2/EF algorithms available for barrel; work started on endcap
        • Tools to match muon track to inner detector are available for L2/EF
      • Muon isolation studies with calorimeter started
        • Status as of Jan06 in talk by G. Usai.
      • Status of muon L2 as of Mar 06 in talk by A.diMattia
      • Status of muon EF as of Mar 06 in talk by A. Ventura
    • Jets/etmiss
      • Oct 05: Work is just starting (K. Cranmer; status)
      • Basic issue for Etmiss is that it takes too long to unpack the calorimeter data. 3 possible solutions being investigated. 1) process the calo data in the RODs, but requires reprogramming DSP's and has low priority among DSP experts. 2) Etmiss algorithms using L1 trigger tower info; however, some technical issues accessing L1 info from L2. 3) Etmiss calculation based on jet ROI's; currently this doesn't improve over L1 Etmiss.
      • First pass at jet algorithm for EF available (Jan 06). TrigCaloRec unpacks cells, does clustering. TrigJetRec finds jets. A study by E.Segura comparing TrigJetRec? with offline results is here. Update from Mar 06 is here; first look at efficiency looks reasonable.
      • Talk (Jan 06) in context of SUSY is here.
        • Crude trigger simulation, using offline algorithms with some added resolution smearing. Efficiency of ~0.92 for SU1 and SU3 points achieved with an OR of jet, Etmiss and SumEt triggers (no lepton triggers considered here); heavy reliance on Etmiss trigger. Trigger thresholds taken from "Triggermenus short" of Schorner-Sadenius and Tapprogge for L=2E33. What are the trigger rates for this menu?? For acceptance studies, should perhaps try SU4 (low mass point) as it may be more realistic/pessimistic.
      • First look (Mar 06) at L2 jets (TrigT2CaloJet? ) described here by O. Jinnouchi. Some early comparisons with truth and offline jets look ok.
      • Work started by LeCompte? on creating jet trigger table: wiki page
    • Tau
      • First look (Mar 06) at calorimeter part of L2 taus is described here by C.Osuna.
      • First look (Mar 06) at tracking part of L2 described here by O.Igonkina.
      • Status of EF part (TrigTauRec? ) as of Mar 06 summarized here by R.Soluk. Wiki page.
    • Inner detector tracks
      • Tracking with Si data ready at L2/EF. Summary of recent EF activities (Jan 06) in talk by I.Grabowska-Bold.
      • Tools available for standalone TRT tracking and track extension to the TRT
      • Short description of some of the ID algorithms in the context of egamma can be seen here.
      • Performance of IDscan algorithm: ATL-COM-DAQ-2006-021

-- SriniRajagopalan, GeorgeRedlinger - 18 Nov 2005

About This Site

Please note that this site is a content mirror of the BNL US ATLAS TWiki. To edit the content of this page, click the Edit this page button at the top of the page and log in with your US ATLAS computing account name and password.


Powered by TWiki
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback