Skip to content

BuSca. Buffer Scanner for LLP searches

Valerii Kholoimov requested to merge Downstream_BuSca into 2024-patches

image

Require lhcb-datapkg/ParamFiles!90 (merged), Moore!3846 (merged) to run CI tests

Developers: V. Kholoimov, A. Oyanguren, B. Jashal, J. Zhuo, V. Svintozelskyi

This is a prototype algorithm for searches at the HLT1 level. It is based on the reconstruction of downstream tracks and vertices, selecting candidates and projecting them in 2D histograms of flight distance (FD) and mass at 30 MHz. The main feature of the BuSca algorithm is that it does not persist the data and therefore saves the bandwidth. Hot spots are searched for indicating a probability that new particles are encountered in that region, providing a guide for new HLT1 and HLT2 lines development.

BuSca structure

Filling 2D histograms online: FD vs mass and Armenteros-Podolanski plot.

1. BuSca monitoring system

Selection procedure

BuSca uses one hidden layer neural network, based on tracks and vertex quality. NN is trained on MinBias and signal channels.

NN structure NN Output
image image.png

BuSca histograms

  1. The mass vs FD plot: gives the opportunity to see not only the mass of the possible new particle, but also the mean place where it decayed. SM particles with masses greater than 1.5 GeV are not expected to decay in the downstream region. This is due to the fact that most of the resonances produced in proton-proton collisions or in B decays have very short lifetimes and decay inside the VELO detector.
  2. The Armenteros-Podolanski plot: it is a representation of the p_{T} of the parent particle in a two-body decay, versus the asymmetry of the longitudinal momentum of the decay products. They appear as semi-ellipses, whose parameters provide information on the masses of the parent and the child particles.

Allen implementation is based on !1556 (merged)

Binning scheme (FD vs Mass)

Adaptative binning is used based on the mass and FD resolutions:

\sigma_m = 0.02\cdot m ; [MeV]

\sigma_{FD} = 80 - 0.02\cdot FD; [mm]

Bin width is configurable in terms of number of \sigma: BW = k\cdot\sigma

See details: https://indico.cern.ch/event/1361613/contributions/5852549/attachments/2822381/4929073/BuSca_19_03.pdf

Output example

FD vs mass Armenteros-Podolanski
image image
The left peak corresponds to K^0_S and Λ^0 particles, while an example of a new particle is shown at 3 GeV. K_S^0 pairs create a bottom big ellipse, with center at (0; 0), while Λ^0 and \bar{\Lambda}^0 form two symmetric ellipses at left and right parts of the plot. The resulting ellipses are additionally highlighted with coloured lines. The top ellipse indicates a new particle with mass 3GeV.

2. Trigger mechanism

HLT1DownstreamBuSca

Line is prepared and turned off until a signal is detected.

Configuration: switch mechanism, mass and FD windows.

image

Turning on in mass = [2800; 3200] MeV and FD = [500; 2300] mm.

image

Armenteros-Podolansi plot for triggered events:

image

BuSca Extra Lines for background studies

  • Downstream same sign vertexing + Same Sign BuSca
  • BuSca Postscaled line (with veto on KS and Lambda)
  • BuSca MuMu and ElEl lines

Bandwidth effect based on MEP dumps:

run 300855
2024-patches:
Inclusive:                                      64111/1000000, ( 1923.33 +/-     7.35) kHz

with BuSca lines:
Hlt1DownstreamBuSca:                                0/1000000, (    0.00 +/-     0.00) kHz  # only monitoring line                                                                                                                                                  
Hlt1DownstreamBuScaSameSign:                        0/1000000, (    0.00 +/-     0.00) kHz  # only monitoring line                                                                                                                                                  
Hlt1DownstreamBuScaPreScaled:                       1/1000000, (    0.03 +/-     0.03) kHz                                                                                                                                                    
Hlt1DownstreamBuScaMuonLine:                        3/1000000, (    0.09 +/-     0.05) kHz                                                                                                                                                    
Hlt1DownstreamBuScaElectronLine:                    5/1000000, (    0.15 +/-     0.07) kHz                                                                                                                                                    
Inclusive:                                      64124/1000000, ( 1923.72 +/-     7.35) kHz

run 301222
2024-patches:
Inclusive:                                      65578/1000000, ( 1967.34 +/-     7.43) kHz

with BuSca lines:
Hlt1DownstreamBuScaPreScaled:                       4/1000000, (    0.12 +/-     0.06) kHz                                                                                                                                                    
Hlt1DownstreamBuScaMuonLine:                        3/1000000, (    0.09 +/-     0.05) kHz                                                                                                                                                    
Hlt1DownstreamBuScaElectronLine:                    3/1000000, (    0.09 +/-     0.05) kHz    
Inclusive:                                      65583/1000000, ( 1967.49 +/-     7.43) kHz

Throughput effect

(measured with testbench over 300855 run including monitoring and hlt1_pp_forward_then_matching_and_downstream sequence)
2024-patches:
INFO     Average total throughput: Evts/s = 83162.7

including BuSca:
INFO     Average total throughput: Evts/s = 82511.7

expected decrease ~ 0.7%

Full testbench output: https://cernbox.cern.ch/s/eUVvA6LZGWnYUAY

HLT2

A passthrough line will is prepared: Moore!3846 (merged)

Edited by Valerii Kholoimov

Merge request reports