1 of 38

Journey to the �CMS EGM Object Review

Bongho Tae

(Kyungpook National University)

on behalf of

The CMS Collaboration

2025 KCMS Workshop

1^st July 2025

2 of 38

BONGHO TAE (KNU)

KCMS 2025 Workshop

EGM

2

3 of 38

Outline

Case Study
“Who” do you contact for an object review?
“Where” can you find the official approval guidelines?
“What” do you need to prepare for object approval?
“How” do you validate and approve a new photon ID?
Summary

BONGHO TAE (KNU)

KCMS 2025 Workshop

3

Photons from LLP decays may arrive at the detectors significantly later compared to promptly produced Standard Model photons. As illustrated on the left, an invisible Beyond Standard Model particle travels some distance before decaying into photons, resulting in delayed photon detection in ECAL.

In our analysis, we specifically focus on displaced photon signatures—highlighted here—which are photons emitted significantly later than the initial proton-proton collision. Detecting these delayed photons allows us to identify potential LLP decays and experimentally test various BSM theories.

Delayed photon signatures offer unique advantages in our search:

First, delayed photons can be clearly separated from prompt photon backgrounds.
Second, background noise is significantly reduced due to the precise timing measurement capabilities provided by the CMS ECAL.

4 of 38

CASE STUDY

5 of 38

Search for long-lived particle using delayed photon signature

Snowmass Points and Slopes 8 (SPS8) benchmark model:

GMSB scenario with neutralino NLSP.
Long-lived neutralino decay into photons & gravitino.
Parametrized by neutralino’s lifetime cτ and SUSY breaking scale Λ.

Event signature: Delayed photons and MET

BONGHO TAE (KNU)

KCMS 2025 Workshop

(n1)

5

6 of 38

New Id for Delayed Photon

BONGHO TAE (KNU)

KCMS 2025 Workshop

NEW ID

6

7 of 38

Status

Analysis PAG (Physics Analysis Group): EXO (Exotica)
Main object: Delayed Photon

POG (Physics Object Group): EGM (Egamma)

Photon Collection: SlimmedPhoton + SlimmedOOTPhoton
DNN based photon ID
Trigger: Displaced single photon trigger

HLT_Photon60_R9Id90_CaloIdL_IsoL_DisplacedIdL_PFHT350MinPFJet15

It is very important to precisely follow the object review procedure that matches your research topic and analysis goals.

BONGHO TAE (KNU)

KCMS 2025 Workshop

7

Photons from LLP decays may arrive at the detectors significantly later compared to promptly produced Standard Model photons. As illustrated on the left, an invisible Beyond Standard Model particle travels some distance before decaying into photons, resulting in delayed photon detection in ECAL.

In our analysis, we specifically focus on displaced photon signatures—highlighted here—which are photons emitted significantly later than the initial proton-proton collision. Detecting these delayed photons allows us to identify potential LLP decays and experimentally test various BSM theories.

Delayed photon signatures offer unique advantages in our search:

First, delayed photons can be clearly separated from prompt photon backgrounds.
Second, background noise is significantly reduced due to the precise timing measurement capabilities provided by the CMS ECAL.

8 of 38

“Who”

do you contact for an object review?

9 of 38

EGM Object Review - Contacts

PAG Contacts

POG Contacts

To conduct an object review, always start by contacting your PAG representative.
For official approvals (new ID, SF, WP, etc.), reach out to the POG contacts.
Coordinate with the POG contacts to schedule your presentation at the EGM meeting.

BONGHO TAE (KNU)

KCMS 2025 Workshop

PAG_contact_list

9

10 of 38

EGM Object Review - Contacts

PAG Contacts

POG Contacts

To conduct an object review, always start by contacting your PAG representative.
For official approvals (new ID, SF, WP, etc.), reach out to the POG contacts.
Coordinate with the POG contacts to schedule your presentation at the EGM meeting.

BONGHO TAE (KNU)

KCMS 2025 Workshop

PAG_contact_list

10

11 of 38

“Where”�can you find the official approval guidelines?

12 of 38

Physics Object Approval Process

All information required for the physics object approval review is available on this page.
Since each Physics Analysis Group (PAG) and POG follows its own specific procedures, it is essential to carefully review and follow the guidelines that are relevant to your analysis.
Making sure you are following the correct process from the start will save significant time and effort during the approval.

BONGHO TAE (KNU)

KCMS 2025 Workshop

Physics Object Approval Process

12

13 of 38

EGM Object Review

Dataset, Trigger, ID, SF, Smearing, Scale, etc.

BONGHO TAE (KNU)

KCMS 2025 Workshop

POG_review_instructions

Egamma Questionnaire

EgammaPOG

13

14 of 38

“What”

do you need to prepare for object approval?

15 of 38

EGamma Questionnaire

The Egamma Questionnaire is a required form when introducing a used electron or photon object to your analysis.�
It collects essential information—such as analysis goals, object details, and working points—so that the POG can review and approve your proposal.�
Filling out this form carefully is the first step to official approval and smooth communication with the EGM group.

BONGHO TAE (KNU)

KCMS 2025 Workshop

15

16 of 38

Key Points of the Egamma Questionnaire

Clearly state any special conditions or phase space covered in your analysis

High pT photon region (pT > 100 GeV)
Barrel region only (|η| < 1.44)
No low pT or BParking datasets used

Clearly document the data, objects, and triggers used in your analysis

2016,17 and 18 pre-legacy samples
Only Photon (Slimmed photon + Slimmed OOT Photon)
Trigger: Displaced single photon trigger
Official ID or Custom ID

Describe in detail how you derive efficiencies (scale factors) and estimate systematics

Method: Tag and Probe, Orthogonal
systematics: Alternative fits, Varying tag selection criteria, Alternative MC samples

BONGHO TAE (KNU)

KCMS 2025 Workshop

16

17 of 38

What The (scale) Factor?

BONGHO TAE (KNU)

KCMS 2025 Workshop

17

18 of 38

EGM Meeting

If you plan to apply a new ID—or any new object or method—to your analysis, you must present it at the EGM General Meeting.

This presentation is required to:

Receive expert feedback and suggestions,
Validate the object’s performance and implementation,
Ensure the object meets CMS standards before official approval and widespread use.

BONGHO TAE (KNU)

KCMS 2025 Workshop

EGM POG

18

19 of 38

“How”

do you validate and approve a new photon ID?

By Jihun Kim

20 of 38

Scale factor

What is a scale factor?

Scale factor is a correction applied to simulated (MC) events to make them better match what is observed in real data.
It is defined as the ratio of the efficiency measured in data to the efficiency measured in MC:

Why do we need scale factors?

Simulations (MC) do not perfectly reproduce all detector effects and real data conditions.
By applying scale factors, we correct for small mismatches, ensuring our results and measurements are as accurate and unbiased as possible.
Without scale factors, the results from MC and data could be systematically different, leading to incorrect physics conclusions.

BONGHO TAE (KNU)

KCMS 2025 Workshop

20

21 of 38

Tag and Probe method

The Tag and Probe method is a data-driven technique widely used in CMS to measure the efficiency of object identification, trigger, and reconstruction.
Step 1:

Select Tag and Probe candidates from events (e.g., using well-identified electrons as tags and testing candidates as probes).
Define "passing probes" and probe variables using simple cuts or external modules.

Step 2:

Create T&P trees, then use them to fit the distributions and extract efficiencies.
Efficiencies are stored and can be used for scale factor calculations or further analysis.

BONGHO TAE (KNU)

KCMS 2025 Workshop

21

22 of 38

CMS tools for Tag and Probe

The CMS EGM group provides dedicated tools for each step of the Tag and Probe workflow:

EgammAnalysis-TnPTreeProducer: Creation of Tag and Probe trees in CMSSW
egm_tnp_analysis: RooFit-based fitting, efficiency extraction & SF calcualtion

BONGHO TAE (KNU)

KCMS 2025 Workshop

EgammaAnalysis-TnPTreeProducer

egm_tnp_analysis

22

23 of 38

EgammAnalysis-TnPTreeProducer

set up the input samples, triggers, corrections, and various options (AOD/miniAOD, MC/data), and outputs the necessary TnP analysis root files

specify which features (kinematic variables, ID, isolation, MVA scores, etc.) will be saved for each probe and tag object in the output TnP trees.

BONGHO TAE (KNU)

KCMS 2025 Workshop

EgammaAnalysis-TnPTreeProducer

23

24 of 38

egm_tnp_analysis

BONGHO TAE (KNU)

KCMS 2025 Workshop

egm_tnp_analysis

24

25 of 38

Scale factor for delayed photon ID

BONGHO TAE (KNU)

KCMS 2025 Workshop

25

26 of 38

Systematics

BONGHO TAE (KNU)

KCMS 2025 Workshop

• statData/statMC: Variation due to statistical uncertainty in data/MC

• altSignalModel: Use alternative signal PDF (e.g., double Crystal Ball)

• altBkgModel: Use alternative background model (e.g., polynomial instead of exponential)

• altMCEff: Change MC efficiency calculation method

• altTagSelection: Use alternative tag selection to estimate robustness

26

27 of 38

Orthogonal method

BONGHO TAE (KNU)

KCMS 2025 Workshop

Selection + Muon Trigger + Delayed photon trigger

Selection + Muon Trigger

Orthogonal Method

Samples

Data :SingleMuon
MC : WGToLNuG

Select to use samples containing photon and muon

Muon Trigger : HLT_IsoMu27_v*
Delayed Photon Trigger :

HLT_Photon60_R9Id90_CaloIdL_IsoL_DisplacedIdL_PFHT350MinPFJet15

When Tag and Probe is challenging, the orthogonal method provides a robust way to measure efficiency using independent triggers and selections.

27

28 of 38

Trigger Efficiencies & SF

BONGHO TAE (KNU)

KCMS 2025 Workshop

Data eff.

MC eff.

scale factor

Efficiencies

2018

Data eff.

MC eff.

2017

28

29 of 38

CMS Pub Talk

CMS Pub Talk is the official communication platform for CMS collaboration members.
Here, you can ask questions, request object reviews, follow approval processes, and discuss various analysis topics.
If you need help or want to stay updated on important discussions and official announcements, always check or use CMS Pub Talk.

BONGHO TAE (KNU)

KCMS 2025 Workshop

29

30 of 38

CMS Pub Talk

BONGHO TAE (KNU)

KCMS 2025 Workshop

30

31 of 38

Summary

“Who” do you contact for an object review?

PAG Convener

“Where” can you find the official approval guidelines?

Physics Object Approval Process

“What” do you need to prepare for object approval?

EGamma Questionnaire, Object review in EGM meeting

“How” do you validate and approve a new photon ID?

Calculation SF with systematics

BONGHO TAE (KNU)

KCMS 2025 Workshop

31

Photons from LLP decays may arrive at the detectors significantly later compared to promptly produced Standard Model photons. As illustrated on the left, an invisible Beyond Standard Model particle travels some distance before decaying into photons, resulting in delayed photon detection in ECAL.

In our analysis, we specifically focus on displaced photon signatures—highlighted here—which are photons emitted significantly later than the initial proton-proton collision. Detecting these delayed photons allows us to identify potential LLP decays and experimentally test various BSM theories.

Delayed photon signatures offer unique advantages in our search:

First, delayed photons can be clearly separated from prompt photon backgrounds.
Second, background noise is significantly reduced due to the precise timing measurement capabilities provided by the CMS ECAL.

32 of 38

Thank you for attending my presentation!

Wishing you success and smooth approval for your analysis!

33 of 38

Backup

34 of 38

Physics Object Review

Physics Object Review is the official process in CMS to validate and approve new or updated physics objects—such as electrons, photons, jets, and more—for use in physics analyses.

This review ensures that all objects meet the required performance standards, are well-understood, and are documented for use by the entire collaboration.

The process involves detailed validation, expert feedback, and formal approval through dedicated working groups (POGs), guaranteeing the reliability and reproducibility of CMS physics results.

BONGHO TAE (KNU)

KCMS 2025 Workshop

34

35 of 38

CMS Statistics Questionnaire

35

36 of 38

Q: Since this is a single photon trigger, Why do we need the two separate training for leading and subleading photon

Answer: The leading photon is almost always the triggering photon, which imposes tighter requirements on that photon. The subleading photon is typically unbiased by the trigger, and therefore, the DNN training can be performed on a much more inclusive set of photon objects, allowing more optimization. We can obtain a more optimal signal selection by training the two types of objects separately.

36

37 of 38

Q: Why do you have different training for 2016 and 17/18? Does the detector change between the years?

Answer: We have different training because the triggers used in 2016 and 17/18 are different. The training acts on a different set of photon objects, and that has a significant impact on the ID performance. Therefore the different training for 2016 dataset and 17/18 dataset results in more optimal discriminant and better signal to background discrimination.

37

38 of 38

Q: On slide [6], why are the shaky ROC curves? If you merge the years, it might help you. Is your major background QCD?

Answer: Yes, QCD is our major background and due to the limited size of the QCD MC sample, the ROC has shaky curves when the cut becomes very tight.

38