Muon HLT group scratchpad (Ingo's private notes)

Place to keep notes for getting started on the muon HLT

As a great overview start with Zoltan's Trigger review talk:

Recent errors from the HLT

110321 Error in online running, causes high error message traffic

 ERROR %MSG-e MissingTransientHit: Could not get a tsos on the hit surface. We will miss a tracking hit. L3MuonProducer:hltL3MuonsOIHit 21-Mar-2011 16:42:41 CET Run: 160998 Event: 135753192 
  • code was protected against this failure in 2009:twiki entry
  • Adam and Hwidong are hunting the reason for the error
  • If shown to be less severe, downgrade to WARNING

110321 HLT offline DQM is not using outdated code and needs to URGENTLY be fixed HLT paths are hardcoded

  • slipped through the cracks - Jeff Klukas is working on producing a patch
  • needs to be included fast

110322 Error in the validation of 5e32 v6.0 menu

 %MSG-w Muon:  L3MuonProducer:hltL3MuonsIOHit 22-Mar-2011 04:17:12 CET  Run: 149291 Event: 290181741 
Propagation to the PCA using TSCPBuilderNoMaterial failed! This can cause a severe bug. 
  • Noone yet hunting this bug

Immediate toDo

  • Set up meeting to discuss muon HLT efficiency measurements in detail (cuts, differences -> Establish baseline)
  • Answer Hwidong's mail about TP plots for next Monday
  • Answer Jonathan's mail about DoubleMu4_Exclusive rates

  • Follow up on Issues with TSGFromL2Muon:hltL3TrajSeedIOHit with Adam (written email)
    • Check setting of "OnDemand" in ConfDB - is it 1?
  • Tick off the action items on
    • one item asks for relative isolation
  • DONE (by Slava) Answer Manuel's reply for HLTiso - technical help
  • DONE: Comments to Jeff's DQM plots:
    • Eta and pt cuts are applied to fill the efficiency plots. Would be good to quote them in the histogram title so they are not "hidden"
    • Maybe a linear (x axis) plot of the turn on curves might also be useful (maybe binned like the plots on page 6 of Zoltan's review talk). Current ones might well be enough.
    • Would be great to have a plot of trigger rate vs lumi section normalised to the lumi. Complicated?
            Hi Ingo,
            > > - In your mail you mention some eta and pt cuts you applied to fill the
            > > efficiency plots - I think it would be good to quote them in the histogram
            > > title so they are not "hidden"
            I agree with you that it would be nice to find a way to display this.
            It turns out to be technically difficult, since these efficiency
            histograms are the result of a postprocessor, which can only make
            generic titles that will be the same for all the different paths.
            There may be reasonably easy ways, however, to put this information in
            the "describe" box in the GUI.  I'll do another round of investigating
            and see if I come up with a better solution this time.
            > > - I was pondering if a linear (x axis) plot of the turn on curves might also
            > > be useful (maybe binned like the plots on page 6 of Zoltan's review talk). I
            > > assume that the ones that are there are good enough though.
            I'll consider it.
            > > Also it would be great to have a plot of trigger rate vs lumi section
            > > normalised to the lumi, but that might be complicated?
            I don't know how hard that would be.  I'll try to take a look at what
            would be involved.
            Thanks for the input!
  • DONE: Answer TP HN mail about MuX efficiency measurements
  • DONE Get in contact with Silvia to Start implementing paths and test them on data
  • DONE: Set up agenda for Feb4th - potentially:
    1. Manuel Zeise: update on HLT isolation
    2. Javi: PU dimuon triggers
    3. Preview of the Trigger Review talk
    4. High pT muon triggers
    5. Implementation and tests of the 5e32 ++ muon triggers

Some of my questions

  • muHLT rate budget was reduced from 20 to 13 Hz for 5e32. What was the motivation? Will it go back to 20 in the real world? i.e. plan for 13 and get 20?
  • Are the L1/L2 sent by Silvia for 5e32 final? What does need to be done if not?
    • have the 5e32 paths been implemented in ConvDB?
  • The track iso inefficiency for HLT iso needs to be solved fast - Manuel is making great progress. Archie willing to share the load.
  • Davide Pagano <> and high pT triggers - did you meet with him on skype (last email 101204?)
  • "L2 Cleaner from Daniele" - context? (Seems not related to the L1 CSC doubles issue)
  • for DQM, what is FV and TP (TP = trigger performance?)?
  • I have to be in Berlin on the 08.02.2011, the trigger review. Can someone step in for me?
  • We need a MuHLT certification expert on duty? Shifts? Pool of people? Attend PVT meetings? Report on DQ? What about release validation?
  • What is Adam's involvement at the moment?
  • News from Nicola A. about the HLT Note editorship?
  • Should we work on a looser Iso trigger for high Pileup? Matter of optimisation? Tighter ID and looser Iso? Higher Pt and looser Iso?
  • Detailed:
    • on slide 8 (here (Zoltan) data and MC agree for Mu15. Sinead shows disagreement between data and MC for the same trigger. Is this due to different input samples / pt sprectra? Else why?

  • DONE: HLT Isolation - how do Manuel's and Jeff's studies overlap? Answer: Jeff did not have time to continue this study
  • DONE: What is the best documentation available for MuX efficiency? - needed for TP HN question...

General guidelines:

  • Backup triggers should be >=2x lower than the trigger they back up to make sense
  • HLT vs L1 thresholds: generally L is 90% eff at its threshold, so e.g. use HLT 12 GeV for L1 10 GeV (did I get that right??)
  • keep L2 loose, far from L3, i.e. currently at L1 cut

External Info

L1 trigger menus

Outstanding / Open requests / Tasks

  1. establish muon HLT DQM certification shift / team

Missing triggers


  1. Monitoring Triggers (e.g. for T&P)

Would be nice

  1. Mu+MET or +HT (EXO, listed as missing at
  2. high pt ele/mu + HT trigger (EXO)
  3. L2 Muon without L1 seed - need manpower for study - Daniele/Riccado??? Stoyan

Need to follow up / Ping

  1. Silvia Goy Lopez: Set HLT trigger thresholds for >1e33
    1. Can we be allowed to implement a set of (several) backup triggers per main trigger?
    2. Muon trigger rates from past TDRs - comparison to todays triggers and reasons for differences...
  2. L2/3 cleaner for ghosts. Daniele has code, but it was never validated and he has no time? Find someone to test it or do myself.
  3. Get muon HLT person name from SUSY (was Tom Danielson, but he has no time. They ask us for a name that would fit the bill)
  4. Mailed 110201: Davide Pagano <>: Interested in HighPt muon triggering for EXO group
  5. Zhang Jinzhong <> for muon trigger inefficiency study for low pT muons
  6. Zoltan about "the TWiki on the muon-HLT offline DQM"
  7. NoVertex Trigger: the Princeton Group (Valerie Halyo and Paul Lujan) are working on it
  8. TripleMuon trigger: Massimiliano / SUSY

Check out these:

  1. Mu17_Ele8_CaloId
  2. Ele17_IdIso_Ele8_IdIso
  3. / displaced mu+mu- DiL2Mu35 studies on way to improve this

Outstanding "external" requests

Muon triggers for 5e33?

Outstanding internal requests

Adam: need manpower

  1. L2/L3 cleaners: Could do with some help here
  2. high pT triggering - same here

Barrel Trigger for luminosity measurements

Michael Schmitt / EWK

[...] we are thinking seriously about a
 barrel-only single-muon trigger to be used for
 collecting W->munu with no MET cut and a reasonably
 low pT threshold.  Such events would be used for
 luminosity monitoring (or measurement) and ultimately
 for a W mass measurement. 

Sort out Muon HLT note?

The note is in SVN as notes/MUO-10-003

Important (recent) numbers

Effect of isolation

summarise findings from last muHLT meeting

Effect of pileup (using MC)

summarise findings from last muHLT meeting


Fridays 17:00, biweekly (I call for it) Muon HLT, e.g.

Thursdays 16:00, TSG (plenary), e.g.


L1 prescales
page 2 and 4:
CSC L1 changes
HLT Tools
HLT Coding guidelines
HLT Tutorial
General (Muon) HLT
Muon SW guide
Most recent instructions on how to test a menu change
Summaries / Strategy
Silvia Goy Lopez
TGS 28.02.11, Primary datasets for 2011
Updates in MuonHLT meeting 110218
C. Botta
Gian Piero
Sinéad Walsh, TOP group
G.Abbiendi, A.Fanfani, G.Pallaver
Higgs HN message
Patricia Lobelle Pardo, HIGGS
Manuel Zeise, EWK
Jeff Klukas also looked into the data/MC comparison
Jeff Klukas, muPOG

Find HLT Iso Code at (configured as seen in ConfDB):

Dear all,

Note that the twiki page where we collect and link information relevant for
the trigger reviews,
is regularly updated.

Latest updates:

 - the page with plots of HLT rates vs PT thresholds has been extended to also
   show 2D rates for a few cross-triggers :

 - the page with TMD recipes and samples has been updated:

 - A first indication of the L1 menu for the various scenarios, as shown
   by Ivan at the last TSG meeting, is included :

 - Instructions are given for studying the rate of new L1 triggers using
   the nanoDST files and the code from the L1Trigger DPG :

 - Darin has linked slides with possible luminosity scenarios for 2011 :


This is just my personal list, not meant to be anything official - am trying to get to know people!

muHLT Isolation

Manuel Zeise <> EWK
Jeff Klukas also looked into the data/MC comparison

DQM / Certification / Validation

Offline DQM: Jeff Klukas, Jason Slaunwhite
Online DQM: Hwidong Yoo

Future, data certification, release/tags validation, improvements: Javier Fernandez Menendez (and Patricia Lobelle?)

Pileup Efffects on efficiency

Javier Fernandez Menendez (and Patricia Lobelle?)

HLT/ L1 efficiency

Find out where we loose efficiency: Sinead, Benjamin (both top group)

Open task - work on muon HLT generic

Pablo Martinez Ruiz del Arbol <>
Rainer Wallny <>,

Study muon trigger inefficiencies

Zhang Jinzhong <> interested in low pT ineff (then maybe high pt+showers)


Young Kyu, works with Hwidong, should ask for update in next HLT meeting (110203), plan to finish purity then move to something else (tbd)

Preliminary Trigger Thresholds

From Zoltan's Review talk

Preliminary Single Muon

Lumi 2E32 5E32 1E33 2E33 5E33
L1 pT (GeV) 7 10 10 14 20
L1 Rate (Hz) 2300 3000 6000 8000 12000
HLT Iso pT (GeV) 13 15 18 25 40
HLT Iso Rate (Hz) 10 20 20 20 20
HLT pT (GeV) 15 20 25 35 50
HLT Rate (Hz) 20 20 20 20 20

Preliminary Double Muon

Lumi 2E32 5E32 1E33 2E33 5E33
L1 pT (GeV) 0 3 3.5 4 5
L1 Rate (Hz) 2500 3500 2000 2500 3800
HLT pT (GeV) 4 5 6 7 9
HLT Rate (Hz) 20 20 20 20 20

Meeting Notes

110126 Meet Silvia


- Silvia will be gone 05.02 - 12.02.2011

- High Lumi Triggers:
  - S. will contact Emmanuelle about high pT triggers already now - can we put them in?

E.'s Answer:
Concerning higher thresholds: 

The reasons why we do not want to have an infinite number of HLT paths, even if
they are just "backup" paths, are :

 - maintenance. You know the drills...

 - CPU. Agreed, SingleMu50 does not perform any new operation
   compared to SingleMu30 (besides for the pt filter which cost "no" time).
   However, when you unpack the L1 data, you run this module called
   hltL1GtObjectMap, which tells which L1objects did cause the trigger.
   This module takes quite some time (O(5 ms) from the top of my head),
   and this time increases fast (I do not remember the dependence) with
   the number of paths.

 - "making predictions is difficult, especially when they concern the future"
   (the original sentence was probably in better english). Last year we tried
   that a few times for the electron triggers : e.g. while we were running unp'ed
   Ele17_EleId, we also included in the menu a path Ele17_TightEleId, which
   was supposed to become the main path at the next "jump" in lumi. But it happened
   that, in the meantime, people did revisit their path, realized that TightEleId would
   not be enough, such that the path that finally ran unp'ed was something else,
   TightEREleId (this is just to give an example). Hence, introducing "triggers for
   higher luminosity" ended up in "doing some trigger development in the HLT farm"..

This being said, I agree that when we are "pretty sure" that the proposed trigger
would indeed be the one running at a higher lumi, it can be useful to add it.
I would not buy the argument for SingleMuXX, though. Because it is
trivial to "emulate" SingleMu50 in a dataset where you had SingleMu30; you
just need to look at the muons in TriggerEvent and require that the leading one
has a PT above 50 GeV. The argument holds for more tricky conditions, like
these EleId things, that you can not "emulate" from the (limited) information that
is stored in TriggerEvent.

- Monitoring triggers:
  - They are all in the 2e32 menue
! - NONE are in the 5e32 menue!!!!!! 
   - No L2 passthrough triggers are there - do we need them?
     - if we do not have them we need to rereco from RAW in order
       to evaluate intermediate trigger steps (slow)
     - Silvia asked Christina Botta if she had a proposal for a trigger path
       ending on L2 and is waiting for reply
     - need to urgently find out *which are needed* and *if they would still 
       work in 5e32* (is L1 prescaled away?) 
   -> Contact Zoltan and meet with Christina Botta

- HLT timing with PU
  - none of the muHLT paths have been newly tested for timing issues.
  - this can be specifically dangerous with PU
  - needs to be done!

- muHLT rate budget was reduced from 20 to 13 Hz for 5e32. What was the motivation? Will it go back to 20 in the real world?
  i.e. plan for 13 and get 20?

- L1/2/3 thresholds: L2 tresholds might be an issue for for L3 timing
     - currently have L2==L1. This was fine in the past. 
     - Could be problematic for L3 timing at higher Lumi, as
       there will be more seeds for L3.
     - The only trigger that had different cuts on L2 was Mu5 L1:3, L2:4, L3:5 GeV
    -> Need to check high lumi timing
    -> could use increased L2 cuts to improve L3 timing! :)
- DoubleMu4_Exclusive is taken care of by J. Hollar and is DONE

- TripleMu5 is wide open. Needs new code. Silvia has no time to 
  do it. She will meet with J-R, find out details and let me know 
  of the outcome. 
  -> Who requested this trigger? SUSY?
  -> have to find someone to put it in, if needed

The TripleMu path was requested by the SUSY group. Please contact Massimiliano
Chiorbolli and Alex Mott, they have experience with confDB and should be able to
help you.

- MuX + Y: All the cross triggers are PAG responsability.
  muPOG should keep an eye on them and provide guidance if needed. 

- BPAG is trying to figure how to work their triggers without Zoltan.
  Hermine was going to check out the situation for triggers, but is busy in 
  an ARC. Concentrated on signal triggers so far. Triggers for monitoring 
  and or measureing efficiencies are not taken care of so far

Help for me:
- S. will send me all sorts of links for technically getting 
  stuff into confDb and the likes
- S. will contact Emanuele about high pT triggers already now - can we put them in?
- S. will cc me on mails about:
  - TripleMu5 + code development needed - she cannot do it, will talk with J-R
  - *_NoVTX: Exo group, waiting for feedback from Davide Pagano

110126 Meet Zoltan (skype)

Q- When will he be at CERN?
       no trip planned yet. Atlas week end of Feb. not planning to go.
       But probably week after that. 

Q- how can I take over all tools:
    - OpenHLT
    look at Zoltans UserCode,

    - what he used for extracting rates 
    Surprisingly simple, most important to avoid trigger bias.
    Take trigger that is looser (Mu5 for Mu7), or L2 passthrough for L3 
    or go as loose as L1MuOpen. Make sure dataset includes that reference 
    trigger was included actively (not ran along)
    Try to use as tight as possible looser trigger.
    Run reference trigger and new trigger, that gives you reduction factor.
    You get the trigger rate from WBM -> pick a rate at the beginning or end of run
    and also get the inst lumi for beginning or end of run. Then scale all.
    An example cfg card for this:

    process.option.wantsummary = yes for trigger report

    Use this path:
   HLT_L1MuOpen_v2   140   140   140   140   140   140      L1_SingleMuOpen

    -- Important when using a prescaled reference trigger:
    to prepend a filter to path:
    make a filter, then prepend it to each path a la:
    look for hltMinBiasBSC in
    for filter def.
    Then to prepend it add this at the end of the config file.
    for path in process.paths: 
       getattr(process,path)._seq = process.hltMinBiasBSC*getattr(process,path)._seq
    !!VERY important in case of using a prescaled trigger path.

    - for testing paths
      check rate
      check timing how?
      add to ConfDB - tutorial 
      defend in trigger meeting

    - ConfDB

Q-  Ask Zoltan if he could produce some documentation. 
No time :(

Q- Maybe ask him about his ideas on Offline mu HLT DQM too
  - He had notes, but lost them in the move, he thinks it is lost for good.
  - Slava should also have notes
  - most was just common sense. should be reproducible within a few minutes wit Slava.

for new menues -> ConfDB -> zgesce-> something 2011, see his work in progress
I will finish it and ask questions

-> Zoltans argument to put the triggers in:
   - we put them in now, no changes were seen in the past, non expected.
   - they are very reliable
   - he does not believe they actually add cpu time - the init Emmanuelle describes he thinks should be done once, not once per filter (? find out!)
   - should have written them down right away, now I do not recall...

Zoltan for L3 crash:
  - the run boundary crash has happened before, Zoltan found it, so far
    unknown why it happens. Not an issue for online running of course
  - happened in a skim 
  - trigger tracker has a protectrion for time, no more than 5 
  - he forwarded an email about the crash in the past ( Fwd: Muon L3 IOHit crashes on run boundaries due to pixel stuff )
L3 tracking people Adam E., Jean-Roch, H. Yoo (responsible within Purdue)


-- IngoBloch - 20-Jan-2011

Edit | Attach | Watch | Print version | History: r43 < r42 < r41 < r40 < r39 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r43 - 2011-05-24 - IngoBloch
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Main All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright &© 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback