Long lived particle searches with the

CMS experiment. Present and future

S. Folgueras with input from M. Alcaide, A. Escalante, P. Martínez Ruiz

del Árbol, R. López, and A. Soto.
 My (BSM)
Where research
to look interests
for new physics?
Improve precision of SM tests (i.e. Higgs couplings, 𝑚! )
Target unobserved SM processes (i.e. 𝐻 → 𝐻𝐻; 𝐻 → 𝑐𝑐)
Search for deviations at high momenta (i.e. Effective Field Theories)
1. Improve the precision of SM tests (e.g Higgs couplings, MW)
esearch interests
Probe new2. phase
Target unobserved
space SM processes
(i.e. Long-lived (e.g 4 top, H → cc, H → HH)
particles)
Coupling modifier measurements
3. inSearch
and their evolution time for deviations at high-p T
(e.g SM as EFT)
CMS-EXO-19-019
4. Probe new phase space (e.g Long-lived particles)
the precision of SM tests (e.g Higgs couplings, MW)
nobserved SM processes (e.g 4 top, H → cc, H → HH)
or deviations at high-pT (e.g SM as EFT)

Bulk of searches
Nature 607, 60–68 (2022)

19Escalante @ICTEA Seminar

From A. 2
Long-lived particles appear everywhere
Why long-lived particles?
And their c𝜏 was critical to the design of different experiments in HEP

• The SM is full of LLPs:

● • Kaon
muonphysics (e.g NA62)
(𝜏 = 2.2𝜇𝑠)
c𝜏(K+)==3.71
○ (c𝜏(K+)
• Kaon 3.71mm
• Heavy flavour
● Heavy
• c𝜏(D+)flavor physics
= 311.78 𝜇m (e.g LHCb)
+
c𝜏(D
• ○c𝜏(B+) ) = 311.78
= 491.06 𝜇m 𝜇m
+
○ c𝜏(B ) = 491.06
• There is no reason 𝜇m they won’t be present on
to believe
BSM theories.

● Higgs physics (e.g ATLAS/CMS)

○ 𝜏H < 1.9·10-13 s, (𝜏H= 2· 10-22 s in SM)
○ c𝜏(𝜏) = 87.03 𝜇m (e.g H → 𝜏𝜏)

arXiv:2212.03883

From A. Escalante @ICTEA Seminar 3

 G. Cottin
@LHCP 2023

arXiv:2212.03883v1

4
 Experimental signatures of long-lived particles

https://doi.org/10.1098/rsta.2019.0047
5
Run 3
• After the short data taking year in 2023, LHC is performing well
• 85.5 fb-1 recorded in CMS in 2022+2023+2024 at 13.6 TeV
• Improved trigger strategy!

6
 First Run 3 result: Displaced dimuons at 13.6 TeV

https://cms.cern/news/long-lived-particles-light-lhc-run-3-data
https://home.cern/news/news/physics/cms-collaboration-
cern-presents-its-latest-search-new-exotic-particles

From A. Escalante @ICTEA Seminar

 JHEP 05 (2024) 047
Displaced dimuons at 13.6 TeV. New triggers
• Use the 2022 dataset (36.7 fb-1) recorded with new LLP triggers with thresholds down to pT(𝜇) > 10 GeV
• Re-optimized L1 triggers, including pT without beam spot constraint, and new reconstruction algorithms.
• Use dxy information at trigger level to control the background rate.
• Factor 2-4 more signal efficiency

8
 JHEP 05 (2024) 047
Displaced dimuons at 13.6 TeV
Despite 2.5 smaller dataset, comparable (or better) sensitivity w.r.t. 13 TeV result.

From A. Escalante @ICTEA Seminar 9

 CMS-EXO-23-013

Displaced jets at 13.6 TeV

Despite 2.5 smaller dataset, up to factor 10 improvement.

• Improved trigger strategy: 4-17 times larger signal acceptance

• Improved analysis strategy, using GNNs as taggers to identify
dijets arising from LLP decays
• Using displaced track features and info from the displaced vertex
• the displaced tagger achieves a background rejection factor of 104
when the signal efficiency is ~55%

10
 [CERN-LPCC-2019-
01]
Towards the HL-LHC
• Preparing for the big upgrade of the LHC detectors,
starting 2029.
• HL-LHC upgrade offers an unprecedented opportunity to
explore uncharted lands and achieve scientific progress.
• 10 times more data to what we will have by the end of
Run 3 will facilitate a rich physics program.
• Extend reach of new physics searches: unexplored
signatures (LLPs, HSCPs… ) or regions of the phase-
space will be within reach.
• Improve current understanding of the SM and Higgs
sector by improving existing precision measurements
and accessing rare decays (H → 𝜇𝜇) or production modes
(HH) previously unseen at the LHC.
• However, this physics program will have to overcome
significant challenges to succeed.

11
Improve muon triggers using the existing architecture
hits

tracks

Displaced
signatures

14/06/2023 12
 Extending LLPs coverage (yet in Run-3)

From A. Escalante @ICTEA Seminar 13

Reconstruction of muon showers

x10
Mixing angle

Current trigger (missing

energy-based) strategy
Hit-counting trigger
strategy
JHEP 02 (2023) 011; arXiv:2210.17446

14/06/2023 14
Graph Neural Networks for real-time muon reconstruction

14/06/2023 15
Graph Neural Networks for particle reconstruction
F. Siklér, “Combination of various data analysis techniques for efficient

Representing tracking data using graphs track reconstruction in very high multiplicity events”,
Connecting the Dots conference 2017 (link)
S. Farrell et al., “Novel deep learning methods for track reconstruction”,
proceedings of Connecting the Dots conference 2018 (link)

Charged particles leave hits in the Represent the data using a Goal:
detector graph classify the edges of the graph

• High classification
score
• => high probability
that the edge is part of
a track
• Low classification score
• => low probability that
the edge is part of a
track

One node of the graph = one hit in the detector

Connect two nodes using an edge
if “it seems possible” that the two hits
are two (consecutive) hits on a track

Jan Stark European AI for Fundamental Physics Conference, Amsterdam | April/May 2024 5

14/06/2023 16
Explore capabilities of AI-engines
Provide the necessary throughput and latency for triggering?

x8
performance

x5
throughput

14/06/2023 17
Thanks for listening!
Foreseen
improvements on
detection efficiency
and triggering might
allow the discovery of
BSM physics.

Provide an answer to
fundamental
questions of nature.

14/06/2023 18
backup

19
 Our demonstrator
ICTEA 2024 A. ZABI
52
CMS L1 TRIGGER @ HL-LHC

TESTING AND SYSTEM DEMONSTRATION

Phase-2 Level-1 Trigger system demonstration
‣ Single-board and multiple board tests performed
‣ Integration centers across the globe: larger scale
integration @ CERN (904). Multiple flavour board
tests.
‣ Slice test in Muon Barrel Trigger during Run-3.
Installation @P5: DT—>BMT—>GMT—>GT
‣ Board interconnection: protocol
‣ Links (asynchronous) operation @ 25.78 Gb/s
‣ L1 Trigger boards sending packets only once
(no retransmission) → error proof
‣ Protocols (64/66b or 64/67b) encoding
achieved low error rate, validated recovery
mechanism etc. Building 904 @ CERN

Muon Trigger Slice Test

20
INTREPID
FW R&D Slice Test
Timing, latency System-level
01
M1 and occupancy M3
03 demonstrator 05
M5
Demonstrator Proof of concept

SW R&D GNNs Discovery

Displaced jets in
the barrel M2
02 Hit-based pattern
recognition M4
04 Long-lived particles
at the HL-LHC or
beyond
Kalman filter AI engines with
Versal ACAPs Dark matter?
HL-LHC: challenges

• Expected pileup (PU): ~140 (nominal HL-LHC lumi) • Radiation damage / accumulated dose in detectors
and on-board electronics may result in a progressive
• Motivates/requires:
degradation of the performance.
• Improved granularity wherever possible
• Novel approaches to in-time Pile Up mitigation: Precision • Maintain detector performance in harsh conditions:
Timing detectors (30ps) • The complete replacement of the Tracker and Endcap
• A complete renovation of the Trigger and DAQ systems Calorimeter systems.
for better selectiveness, despite the high PU. • Major electronics overhaul and consolidation of the
Barrel Calorimeters and Muon systems

22
 Refined version using module triplets:
C. Rougier, PhD thesis, Université de Toulouse,
defended September 2023 (link)
Graph building techniques
New data-driven graph construction method:
• build graphs starting from a list of possible connections from a zone to another zone: the module map
• done using 90k simulated tt events at <!> = 200, considering particles with pT > 1 GeV and leaving at least 3 hits

Jan Stark European AI for Fundamental Physics Conference, Amsterdam | April/May 2024 8

14/06/2023 23