Look Into Starting Alignment

tomeichlersmith commented 2 years ago

SDF

/sdf/group/hps/users/bravo/run/ali21

KF steering file: alignmentDriver_2021.lcsim
GBL/ST momentum constraint: alignmentDriver_PC.lcsim
GBL/ST alignment plots: alignmentDriver_gblplots.lcsim

Email Nathan Balt*, Stepan, Matt G, and Cam about getting added to ifarm at JLab.

tomeichlersmith commented 2 years ago

HPS "Qualification Task" taken from https://indico.slac.stanford.edu/event/7319/

PF comments

Global KF within ACTS
Re-fit tracks found with KF using global fitter so we can obtain global covariance matrix for alignment
- need to construct global chi2 fitter for hps
Use GBL (General Broken Line) on global-scale interfacing with Millipede for alignment
- this is the path we are going take - in principle, should be "easy"

Code already drafted by PF (needs to be tested, validated, cleaned up) - ask PF about questions related to mathematics/algorithms Issues:

[ ] alignment not converging always
[ ] some memory issue leading to a crash which may or may not be related
[ ] :shit: code

Bkgd Material

Need to "tune" detector sensor positions to their real position based on collected data.

Correcting assumed geometry to match the real one.

Want to get uncertainty down to O(100 um).
Easy to end up in local minimum
Tracks defined by track parameters
Normlized by prediction (assuming helical path through mag field) and uncertainty of measurement

Alignment Steps

Build large linear system
- will change
Solve large matrix equation
- MilliPede will do this for us

Test algorithm by inserting known misalignments in MC and checking if we can resolve that

SimpleGBLTrajAliDriver in hps-java

tomeichlersmith commented 2 years ago

Weak Modes

transformations of detector which leaves $\chi^2$-function unchanged but drastically changes the track parameters.
Most difficult to deal with.
local minimum in minimization scheme

Unbiased Residuals

Fit track to five layers -> extrapolate fit to 6th layer -> difference between this projection (predicted) and measurement

Unbiased residuals thin gaussians around zero is a good sign, but still need to check physical track parameters to make sure we are not in a Weak Mode

check hole side with negative charge (mostly electrons) for FEE shoulder
check slot side with positive charge (mostly positrons) sharply falling distributions
check slot slide with negative charge (mostly FEE electrons) for FEE peak

slot (hole) side : three hits in back of detector on slot (hole) slide

Getting out of Weak Modes

Apply constraints on track parameters (e.g. impose a certain track momentum i.e. curvature)
Try to use different set of tracks (avoid local minimum but still aligning same detector) i.e. trying to have different sets correct each others' wick modes
Try to select high quality tracks with high enough momentum to avoid multiple-scattering effects

New geometry -> recompute global position of hits (don't need to re-decode or cluster) Re-define local->global transformation Always run on lcio as long as we cleanup computed variables No worries about evio -> lcio decoding

java -DdisableSvtAlignmentConstants -Djna.library.path="/sdf/group/hps/users/pbutti/sw/GeneralBrokenLines/cpp/install/lib/" -XX:+UseSerialGC -Xmx5000m -jar /sdf/group/hps/users/pbutti/sw/hps-java/distribution/target/hps-distribution-5.1-SNAPSHOT-bin.jar -R 10031 -d HPS_Nominal_201920216SensorSurvey_iter0 -D outputFile=data_events_aliRecon /sdf/group/hps/users/pbutti/run/alignmentDriver_chi2.lcsim -n 10 -i /sdf/group/hps/data/physrun2019/hps_010031/HPS_PhysicsRun2019-v2-FEE-Pass0/data_events_141.slcio

java should point to Cam's install of 15.0.1 JDK

First Goal: Run Kalman tracking by itself

SensorSetup - associate hits with sensors TrackerHitDriver - associate clusters with tracker modules Kalman can skip a lot of the (Helical, Track strategies, track merge) SimpleGBLTrajAliDriver - pass completed Kalman tracks

/sdf/group/hps/users/pbutti/sw/hps-java/steering-files/src/main/resources/org/hps/steering/recon/

tomeichlersmith / hps