Shu's weekly group discussion

Wednesday 15 Apr 2026, 10:00 → 13:35 Asia/Shanghai

Tsung-Dao Lee Institute/N6F-N630 - Smart Classroom (Tsung-Dao Lee Institute)

点击链接入会，或添加至会议列表：

https://cern.zoom.us/j/4042031778

10:00 → 10:05 Introduction & Announcements

10:05 → 10:15 Status

Speaker: Kang Yu (TDLI, SJTU)

10:15 → 10:25 Status

Speaker: Linjing Duan (TDLI, SJTU)

CEPC Zvvy nTGC

• Discussed with Yuning how to produce MC samples using the nTGC UFO models in MadGraph+Pythia8, and resolved the zero-hit issue in the full simulation by using the correct samples.
• Plan to guide Yuning and others through the next stage of the project, namely histogram production and fitting.

ZZγ

• Studied the HGamCore analysis framework used in the H→γγ* analysis, and summarized which parts can be kept and which parts need to be modified for the ZZγ analysis framework. （this will be discussed with Matthias this afternoon.）
• Spent time learning how to use MadGraphPythia8 to produce MC samples for the pp -> ll + jets process at V+0–3j@NLO precision. I am also still learning how to use POWHEG.
• Next step: confirm and choose the Run 3 samples for our analysis？

10:25 → 10:35 Status

Speaker: Jun Wei Lok (TDLI, SJTU)

• Anomalies and technical issues identified in ANSYS Fluent simulation:
  The contour plot reveals non-physical temperature distributions, within the range of 344C and 166C.
  There is a lack of thermal diffusion between the chips and the PCB board.
  
   
• To improve the ANSYS simulation, will enforce thermal coupling by ensuring all chip-substrate interfaces are set as "Coupled" walls and perform a fresh standard initialization at 293 K to eliminate numerical residuals.
  To avoid being stalled by ANSYS convergence issues, will simultaneously set up the model in COMSOL Multiphysics

10:35 → 10:45 Status

Speaker: Zelin Yan (TDLI, SJTU)

[Today's focus:]

• HGTD monitoring: 
  • Setup OPC-UA server in wincc panel on kth account 
• Wlllv analysis: 
  • Modify analysis framework code for ntuple production with systematics(only contains electron, muon and MET sys, jet sys will be added later)
• H2a4b analysis: 
  • Selection optimization, mainly on MET vars and GN2v01 btag with wp77
  • Data/mc fixed in 0bCR 

[Next plans:]

• HGTD monitoring:
  • Due to high timeout in UI, may swicth to local server
• Wlllv analysis: 
  • Update BDT input and finish ntuple production with sys
  • Fit BDT score with sys 
• H2a4b analysis: 
  • Prepare status report on HLRS meeting this Thursday

10:45 → 10:55 Status

Speaker: Bai-Hong Zhou (TDLI, SJTU)

Today's focus: 

[ssWW]:

• Modified our ParT training codes and retrained the model;
• Used old easyjet framework to run the different sherpa version of WZQCD process;

 

---

Plans: 

 

[ssWW]:

• Check the yields of WZQCD in the WZCR in different sherpa version;
• Use the new ntuples with the DNN WPs to train the ParT model;

[Hyy*]:

• Check the breakdown valiation fit strategy and results;

[Zbbj Calibration]:

• Push all my fitting framework related codes to the main branch;
• Work with non-prompt fit;
• Work with Changqiao for derive the final scale factor;

10:55 → 11:05 Status

Speaker: Xuliang Zhu (TDLI, SJTU)

unet_clusster.pdf

Today's focus: 

[DarksSHINE]:

• UNet Cluster in DarkSHINE DAna framework (For baseline 1.6)

[AI Agent]:

• Skills (You can install it to your claude, codex, openclaw ...)
  • https://github.com/HEPSkills/sm-ana-aod
  • My repo of developing skills for HEP, welcome to add your greate ideas.
  • Using ATLAS Open Data
    • No skills about Athena (not main priority)
    • start from root files and ntuples level are smiilar )
  • Baseline analysis workflow:
    • Hyy (fully tested)
    • Hmumu
    • Zmumu (fully tested)
    • Hbb (fully tested)
    • Hzz4l
    • ttbar
    • wz3l
• Agent installed my skill: https://github.com/hrzhao76/hepex-analysisops-agents/
  • An agent-harness based agent specifically designed for Standard Model experiments
  • Ntuple -> cutflow -> fitting:
    • Analysis framework environment configuration
    • dataset selection
    • event skimming
    • signal region definition (signal region optimization skills will be added later)
    • event selection
    • plotting
    • statistical fitting (lmfit, pyhf (pyhfnot fully tested))
  • A central control agent is responsible for parsing user requests, inferring the most suitable workflow, and dynamically loading relevant physics analysis domain knowledge into the context (e.g., selecting the appropriate ATLAS Open Data version, identifying observables relevant to the analysis, constructing signal/control region logic, and matching the fitting strategy to the target Standard Model process). It then invokes the relevant specialized skills as composable subroutines
  • This agent supports two modes
    • iterative interactive mode
    • fully autonomous mode
  • Usage: docker run  /bin/bash, run oh

docker run -p 9009:9009 \
  -e OPENAI_API_KEY="..." \
  -e HEPEX_AGENT_MODEL="openai/gpt-5" \
  ghcr.io/hrzhao76/hepex-analysisops-agents:latest

• Benchmark: https://github.com/hrzhao76/hepex-analysisops-benchmark 

---

Plans: 

• ONNX speed optimization and new massive sampe production (DAna (reconstruction) only)

11:05 → 11:15 Status

Speaker: Zhiyu Zhao (TDLI, SJTU)

 

Today's focus: 

DarkSHINE:

• ECAL simulation and digitization
  • Integrated the existing optical simulation, detector simulation and digitization code to enable further optimization in the next steps
  • zhiyuzZzZ/detector_simulation_digitization
• ECAL prototype electronics
  • USTC: some updates to the previous preamplifier circuit design, mainly converting the single-ended signal to a differential signal. Now running performance simulations for the new circuit

STCF:

• USTC will soon send us p-CsI crystal for timing-resolution tests coupling with SiPM, to see whether there is an improvement in timing performance compared with APDs
• CsI crystals are hygroscopic, so we need to control the humidity during handling. We are currently procuring a dehumidifier for the new lab, and designing a new readout board for the CsI crystal test

PhD thesis:

• Inserting figures and tables into the draft

Plan:  

DarkSHINE:

• ECAL simulation and digitization: do some parameter scans in  optical simulation, starting with studying the dependence of the detection unit's light yield on the particle's incident angle

STCF:

• Complete the circuit board design and generation, and do some individual SiPM test before receiving the crystal

 

11:15 → 11:25 Status

Speaker: Haidar Mas'ud Alfanda (TDLI, SJTU)

Exotic Higgs decay analysis

• We realized that the default ParT model cannot give a good jet tagging efficiency for 4b final state
• Spent time learning about the Foundation Model
• Next step: Arrange for discussion with Yulei and Xuliang

DarkSHINE

1. HCal R&D:

• The new 75-cm plastic scintillators has arrived
• Performing the cosmic ray test 
• Higher light output (> 200 p.e.)
• The muon detection inefficiency is ~2-3%

2. CDR Paper Review:

• Working on the sections assigned to me
• Reviewing the whole manuscript in depth for general comments
• Get in touch with "not-so-active" members regarding potential contributions

11:25 → 11:35 Status

Speaker: Matthias Tartarin (TDLI, SJTU)

---

Today's focus: 

[Hyy*]:

• Spent (too much) time on debugging my implementation of the weights from MC files, following AC initial directions, and worked from there until the GM on Thursday.
• Still had issues by then. 
  • AC gave me the correct inputs for the ilumni cal tool "Trigger option" (to compute the individual luminosities per data files) -> Corrected the issue I had with Data22 luminosities. 
  • But another issue took the spotlight:  Trouble matching mc23e and Data24 yields in the SR.
  • Danning said everything was good on her side -> Lead to an very explicit printout of each values used as input on my side for the sum of weight, one by one. I finally found an ill-defined value (corresponding to a single file only), which was causing the normalisation factor to blow. After removing this ill-file from the list everything behaved as expected.
• Weight implementation is now finally clean. But all of this work did not solve the shape issue.
  
  [ssWW]: 
• Implemented the NLO correction for the WZ_EW yield (as a function of m_jj) in the reweighing script, according to the results in the article shared by a member of the team.
   

 

---

Immediate Plans: 

 

• [Hyy*]: Now that the weight is not longer a source of issue, this leaves space for investigating the CR templates. It is still not exactly clear how I should implement these. Preparing to present the results I have for the next GM.

 

 

11:35 → 12:30 REU: Undergrad Research Discussions

11:35 Status

Speaker: Peixu Teng (SPA, SJTU)

11:40 Status

Speaker: Zhijing Zhang (CCNU & TDLI/SJTU)

update4.15.pptx

11:45 Status

Speakers: Jiacheng Shi (SPA, SJTU), Yuning Chen (Zhiyuan College, SJTU), Zhengyang Qian (SPA, SJTU)