Conveners archive

https://groups.cern.ch/group/cms-tracker-alignment-conveners/Lists/Archive/

EOS transition of MPproduction area

• migration not finalized due to missing support for HTCondor submission
  • for details, see CERN-IT ticket: https://cern.service-now.com/service-portal/view-request.do?n=RQF0845905
• missing pieces on our side:
  • resync AFS and EOS area -> make script available
  • update MPS templates to directly point to new EOS area
  • move AFS area to a backup location, e.g. /afs/cern.ch/cms/CAF/CMSALCA/ALCA_TRACKERALIGN/MP/MPproduction.backup
    • only temporary, until all dependencies on old AFS location are fixed, e.g. all-in-one tool feature mp = 1234
  • after moving the AFS area, make the old AFS path a symlink to the new EOS area:

    /afs/cern.ch/cms/CAF/CMSALCA/ALCA_TRACKERALIGN/MP/MPproduction/ -> /eos/cms/store/group/alca_millepede/MPproduction
    

Various methods to build a TrackerAlignmentRcd

Obviously by running an alignment. But especially at start-up it is desired to construct a record from various inputs and assumptions.

Ideal alignment

This can be done with a plugin and the corresponding configuration file.

In the configuration file you have to set the following option:

process.createIdealTkAlRecords.alignToGlobalTag = False

Otherwise, you override (at least parts of) your ideal alignment with the content of the loaded alignment record (either from the global tag or from an ESPrefer).

Now you have created an sqlite file with an ideal TrackerAlignmentRcd, TrackerAlignmentErrorExtendedRcd and TrackerSurfaceDeformationRcd.

Ideal alignment only in parts of the detector

If you want to align the content of your created payloads to the content your loaded alignment payloads based on DetId matching, you can set the above option in the configuration file to True:

process.createIdealTkAlRecords.alignToGlobalTag = True

This does not make a lot of sense, if the loaded alignment and the new payload belong to the same geometry, but was quite useful when initial alignment payloads with the latest strip alignment from the phase-0 detector, but with an ideal phase-1 pixel. As the DetIds are mutually exclusive for the phase-0 and phase-1 pixel modules, the above option did exactly that:

Ideal pixel + realistic strip.

If you want to achieve the same for cases where both input and output belong to the same geometry, you can use this option in the configuration file

process.createIdealTkAlRecords.skipSubDetectors = cms.untracked.vstring("P1PXB", "P1PXEC")

Apply random smearing

Define the input alignment in the configuration file and the smearing in the scenario file and run the tool, e.g., like this:

cmsRun createRandomlyMisalignedGeometry_cfg.py myScenario=MisalignmentScenario_PhaseI_PseudoAsymptotic

Interplay of GlobalPositionRcd (GPR) and TrackerAlignmentRcd

The TrackerAlignmentRcd contains an Alignments objects which contains a vector of AlignTransform for each module.

The way the alignment is applied is the following:

• take the position and rotation from the GlobalPositionRcd which contains an AlignTransform object for each subdetector, i.e. among others also for the tracker
• take for each module the AlignTransform object from the TrackerAlignmentRcd
  • combine those two transformations 1
• these transformations are identical to absolute positions and orientations if you apply them to (0,0,0) (position) and (0,0,0) (orientation)

• the position and orientation of the module are then set (not moved) to the above absolute position and orientation 2

To make a long story short: real position = alignment

How to run a private rereco

The tool used to run the 2017 private rereco is available here.

Basically one job per RAW input file is submitted and that's it.

If you want to change the input data you just change the file list in 1.

How to manually recenter a geometry

Define the input alignment within the configuration file and specify the to-be-centered reference system with this parameter:

process.TrackerGeometryCompare.setCommonTrackerSystem = "P1PXBBarrel" # for MC
# process.TrackerGeometryCompare.setCommonTrackerSystem = "TOBBarrel"  # for Data

Afterwards you run the tool like this:

cmsRun GeometryCentering_cfg.py

How to measure ALCARECO rates and event sizes

You can use the script written by Joze Zobec:

https://github.com/cms-trackeralign/ALCARECO-statistics

Gero's legacy thoughts

https://indico.cern.ch/event/243934/#22-closing-words https://twiki.cern.ch/twiki/bin/view/CMS/TkAlignmentLegacyThoughts

How to exclude BPIX hits

In universalConfigTemplate.py one has to replace the following snippet (at the bottom of the file):

if setupAlgoMode == "mille":
    import Alignment.MillePedeAlignmentAlgorithm.alignmentsetup.MilleSetup as mille
    mille.setup(process,
                input_files        = readFiles,
                collection         = setupCollection,
                json_file          = setupJson,
                cosmics_zero_tesla = setupCosmicsZeroTesla,
                cosmics_deco_mode  = setupCosmicsDecoMode)

with this snippet:

if setupAlgoMode == "mille":
    import Alignment.MillePedeAlignmentAlgorithm.alignmentsetup.MilleSetup as mille
    mille.setup(process,
                input_files        = readFiles,
                collection         = setupCollection,
                json_file          = setupJson,
                cosmics_zero_tesla = setupCosmicsZeroTesla,
                cosmics_deco_mode  = setupCosmicsDecoMode)
    process.TrackerTrackHitFilter.commands.remove("keep PXB") # remove BPIX hits
    process.TrackerTrackHitFilter.commands.append("drop PXB") # for real!

Note, that it is important to have the two new lines within the if branch, i.e. mind the indentation. I am also not sure, if only the second line (marked with # for real!) is sufficient, but using both lines works for sure.

Switching to generic CPE

If you want to switch from the default template CPE to the generic CPE you have to change a few configuration parameters.

For the validation you have to set the following two parameters for all validations that use data:

ttrhbuilder = WithTrackAngle
usepixelqualityflag = false

The default values are listed on the All-in-One Tool Twiki.

For MillePede one has to add another parameter to the mille.setup() call:

if setupAlgoMode == "mille":
    import Alignment.MillePedeAlignmentAlgorithm.alignmentsetup.MilleSetup as mille
    mille.setup(process,
                input_files        = readFiles,
                collection         = setupCollection,
                json_file          = setupJson,
                cosmics_zero_tesla = setupCosmicsZeroTesla,
                cosmics_deco_mode  = setupCosmicsDecoMode,
                TTRHBuilder        = "WithTrackAngle")

There is no need to specify something for the pixel quality flag because this is only defined when using templates. The method detects if templates are not used and automatically sets this parameter to False, which is not true for the validation. For the validation one has to explicitly write (as done above):

usepixelqualityflag = false

If one uses the integrated pixel Lorentz angle calibration one has to change also the following parameter when using generic CPE:

siPixelLA.lorentzAngleLabel = ""

This assumes you have included the configuration fragment like this:

from Alignment.CommonAlignmentAlgorithm.SiPixelLorentzAngleCalibration_cff \
    import SiPixelLorentzAngleCalibration as siPixelLA

By default this label is set to "fromAlignment", i.e. if one wants to override the Lorentz angle in iterations or validations, one needs to make sure that the label is consistent.

When one overrides conditions, one often omits the label because it is most of the time an empty string. If it is not empty, one has to explicitly use it.

About LA/BP corrections

How LA are handled together with pixel templates

If templates are used one has to set DoLorentz to True. This is the default:

https://github.com/cms-sw/cmssw/blob/master/RecoLocalTracker/SiPixelRecHits/python/PixelCPETemplateReco_cfi.py#L16

If the LA is zero (currently true for LA from alignment), we would get a unit vector in negative z direction here: https://github.com/cms-sw/cmssw/blob/master/RecoLocalTracker/SiPixelRecHits/src/PixelCPEBase.cc#L524-L532

This is used here: https://github.com/cms-sw/cmssw/blob/master/RecoLocalTracker/SiPixelRecHits/src/PixelCPEBase.cc#L213-L214

And the members for each DetParam object that are set here: https://github.com/cms-sw/cmssw/blob/master/RecoLocalTracker/SiPixelRecHits/src/PixelCPEBase.cc#L544-L545

are then finally used here: https://github.com/cms-sw/cmssw/blob/master/RecoLocalTracker/SiPixelRecHits/src/PixelCPETemplateReco.cc#L399-L400

Assignment of IOVs

Question to Nazar Bartosik

You assigned different IOVs to the different sub-detectors. Currently we use the same IOVs for BPIX,FPIX,TIB,TID,TEC which are basically determined by movements of the pixel high-level structures that we observe with PCL and by known magnet ramps.

Could you elaborate on how you determined the IOVs per sub-detector?

Nazar's answer

If you mean the purely technical side, the configuration of different sets of IOVs to different subdetectors should be achieved by extending the process.AlignmentProducer.RunRangeSelection list. See the MP campaigns listed below.

If you’re asking about the way we determined those boundaries, I think it was done by ry running alignment with very fine time granularity (probably run-by-run) and looking at determined alignment corrections vs time. The last time that I’m aware of, it was done by Nastya, as you may see here: https://indico.cern.ch/event/243933/contributions/1561044/attachments/416110/578058/Thursday_13.06.13.pdf

On the requirement of running BP together with LA or if LA-only is enough

Question to Nazar Bartosik

If I am not mistaken you need peak/deco data to determine BP corrections and 0T/3.8T data to determine the LA. As both of them are connected, I was wondering if you tried to determine only the LA under the assumption of fixed BP corrections. I am asking because we lack peak data at the moment.

Nazar's answer

Of course you can. That’s what we did before the Backplane-calibration code was implemented. In theory it is going to be less precise, but in practice those backplane corrections looked strange in the end. So if you have no peak data, you can just use the default BP corrections and let the alignment compensate for possible BP miscalibration. Residual effect on hit positions should anyway be much smaller compared to pure alignment without LA calibration.

MP campaigns to look at

• mp1337 - alignment + LA
• mp1338 - alignment + LA + BP
• both are archived on /eos/cms/store/group/alca_trackeralign/MPproduction

Final status report from Gregor

https://indico.cern.ch/event/688840/?showDate=all&showSession=all#35-la-calibration-in-alignment

Automatic determination of dataset weights in MillePede

The idea is to check the track numbers after running the mille jobs and assign default optimal weights based on the track statistics per track topology.

• requires part of the automatic validation to be run before the pede job -> easy
• use machine learning methods to learn from many combinations tracks -> time consuming part
  • perform study with MC
  • try different sets
    • MinBias + ZMuMu (done often during 2017)
    • MinBias + Cosmics (CRUZET/CRAFT) (likely scenario at start-up)
    • MinBias + ZMuMu + IsoMu + Cosmics (CRUZET/CRAFT) + UpsilonMuMu (EOY style)
  • try different numbers of tracks
  • try different weights
  • try different initial misalignment scenarios (but not too extreme)
• possible figures of merit:
  • fit of c in delta_z = c * z for each subdetector (each side for endcaps)
  • sine wave fit of A in delta_X = A * sin(phi + C) where X can be any of the variables in the geometry comparison plots; mainly relevant for barrel

TODO-List for newcomers

• LSF rights for ALCA_TRACKERALIGN and ALCA_MILLEPEDE you can get them via the TkAl conveners -> also space at /store/caf/user/${USER} -> also AFS rights: _calca_:ata (check with command pts m -nameorid ${USER})
• HTCondor rights -> cms-millepede
• LSF rights for ALCA_EXPRESS from AlCa/DB conveners -> cmsexpress
• the EOS group space via e-group subscription and hence from e-group admins -> cms-eos-alca-trackeralign -> cms-eos-alca-millepede

Additional steps for new conveners

• Get admin rights for ALCA_TRACKERALIGN and ALCA_MILLEPEDE -> granted by AlCa/DB conveners or previous TkAl conveners via web interface
• Ask Tracker DPG conveners to be added to cms-tracker-alignment-conveners e-group