Cluster parallel session

Rubin/LSST-France dec. 2023 - CC Lyon

Introduction

​

​

• New(ish) to LSST France Cluster?

​

• Move to another lab and/or position ?

2

Clusters in Rubin and LSST France

Science release!

3

https://www.lsst.fr/amas/

Intra cluster light with HSC Astronomical Data/Image: M. Montes (Instituto de Astrofísica de Canarias); Artistic Enhancement: J. Pinto (Rubin Observatory).

Cluster cosmology in the optical – basic idea

1. take photometric and shape catalogs
2. detect clusters : position, redshift, richness (# of galaxies)
3. get the mass from weak lensing
4. compare to theoretical predictions
5. infer constraints on cosmological parameters 🎉

​

4

CosmoPalooza Fall 2023, 2-5 October 2023

​

DESC: What are the goals for Year 1 Data

5

Not shown here :

base galaxy catalog, PSF model, shear catalog (should be similar to 3x2pts)

DESC: What are the goals for Year 1 Data

6

Not shown here :

blinding (should be similar to 3x2pts)

​

DESC: Current cosmo pipeline diagram

DESC related CL activities in France

Organization : C. Combet (CL working group co-convener)

Cluster detection and characterization :

• Cluster detection algorithms : WaZP, YOLO
• Cluster catalog characterization : + redMaPPer, AMICO
• Algorithms comparison : redMaPPer, WaZP, AMICO

Cluster lensing mass measurements :

• CLMM Cluster Lensing Mass Modeling code development

Cluster cosmological pipeline :

• Overall design
• Data vector with TxPipe
• Theoretical covariances and modeling (TJPCov)

Blending in cluster fields

Multiwavelength analyses

​

8

LSST France expertise outside of DESC

9

Brainstorming publication plan for 2024 / before first data

DESC specific

Cluster detection and characterization :

• Redmapper performances on DC2
• WaZP, Amico, YOLO?
• Algorithm comparison project?

​

• Cluster mass measurements
• CLMM v2
• Cluster cosmology pipeline
• Test of the CL cosmo pipeline on DC2
• Test of the CL cosmo pipeline on real data (HSC, DES ?), from catalogs to data vector or data vector to constraints?
• Blending in cluster field
• ?
• Multiwavelength analyses
• DC2 baryon pasting maps ?
• … other?

Other CL (Rubin) projects outside of DESC?

​

​

10

In any case : lots of work and projects and not so much person-power

Is there a will to have a special framework at the LSST France level?

LSST France framework

• If you’re new to LSST and want to get started, come talk to us
• Help is welcome developing & testing the pipeline elements

​

• Different project on detection do not really work together : it’s a problem. Is there a will to really work as a team? If so, need to clarify project scope and authorship, put in common data sets and analysis tools and meet regularly.

​

• More generally, we could put in place “retreat days” rotating between LAPP, LPSC & APC ?

​

• Opportunities to collaborate on real data analysis to prepare LSST

​

11

Cluster detection and catalog comparison

• Common data/method :
• Different catalogs on cosmoDC2 & DC2, with low “richness” cuts
• Halo catalog
• Matching
• Metrics
• Survey depth and masks maps

​

• Individual CF projects vs CFC goal : What exactly do we want to learn? What metrics we want to do?
• CFC : need a common richness estimator?

12

Computing and data storage needs

(Dominique)

13

CLcosmo analyses in CLWG

Cluster catalog: RM, WaZP, AMICO, YOLO (lot of work and projects)

Ongoing or completed projects in the CLWG related to CLcosmo

• Cluster-only analyses (with LSST data only)
• Binned N + stacked WL approach
• Counts in richness-redshift bins
• Stacked WL profiles or WL mass in each bins 94, 134, 244, 271, 288
• Joint fit cosmo and/or richness-mass relation 168, 264, 243 (CLCosmo_Sim)
• Unbinned approach 139
• Each cluster (M,z) is a datapoint
• Need individual cluster masses → high-mass end only
• Can be combined with binned approach
• Cluster clustering 142, 163
• 2-pt correlation
• Complementary approach to counts
• 3x2pt + N → mentioned several times, but work still to be started in the WG

​

14

Suggest to start coordinating work for this approach as starting point

Some of these developments will serve implementation of other approaches

From C. Combet

The DESC Cl Pipeline

Binned N + WL analysis

​

• Dedicated 1 week sprint before July meeting

​

• 4 dedicated telecons before the July meeting (currently ¾)

​

• Goal : make significant progress on the pipeline infrastructure

15

TXPipe

GitHub: https://github.com/LSSTDESC/TXPipe

Primary Cluster Contacts: Marina

Summary: TXPipe is originally developed as the DESC 3x2pt Pipeline implementation. Suggestion to use it as CL shear catalog generator (see Joe's slides 27-31)

What is needed for the Pipeline:

• ease of installation: need beta testers
• Decision regarding the implementation (specify the output : e.g. directly shear profiles, or source catalog only? And the shear covariance matrices).

​

​

CLMM (Cluster Mass Modeling)

GitHub: https://github.com/LSSTDESC/CLMM

Primary Cluster Contacts: Michel, Marina

Summary: DESC tool consisting of a Python library for performing galaxy cluster mass reconstruction from weak lensing observables

What is needed for the Pipeline:

• Miscentering implementation
• Optimize (interpolate) theoretical computations
• Triaxiality? (individual cluster fit)

SACC

GitHub: https://github.com/LSSTDESC/SACC

Primary Cluster Contacts: Eduardo, Sandro

Summary: SACC (Save All Correlations and Covariances) is a format and reference library for general storage of summary statistic measurements for the Dark Energy Science Collaboration (DESC).

What is needed for the Pipeline:

• Migrate a part of the code from Firecrown to SACC
• Example notebooks (different ordering, diff data types in same file)

​

​

​

​

​

TJPCov

GitHub: https://github.com/LSSTDESC/TJPCov/, Documentation

Primary Cluster Contacts: Matt K.

Summary: TJPCov is a general covariance calculator interface to be used within LSST DESC. Given richness and redshift bins from a sacc file, computes and saves covariance back to sacc.

What is needed for the Pipeline:

• Validation/Benchmarking
• Additional geometries beyond full sky?
• Unified source of model+cosmology parameters

​

Firecrown

GitHub: https://github.com/LSSTDESC/firecrown

Primary Cluster Contacts: Sandro

Summary: Firecrown is a Python package that provides the DESC framework to implement likelihoods. Suggestion is to keep the CLcosmo prediction module in FireCrown (for the time being)

What is needed for the Pipeline:

• Example notebooks
• Integration improvement
• Purity implementation
• Weak lensing profile prediction

​

CLCosmo module (cluster counts prediction, proxy, completeness)

Firecrown cluster likelihood

(different mass functions)

Cluster detection projects

• Kyrill, Michel, Simona, Vincent

​

• News :
• [MA] New WaZP catalog on DC2 (flexZBoost)
• [KG] Yolo detections on DC2 with both SDSS and DC2 training
• Problem encountered :
• Next expected milestone :
• [VR] Demo implementation of Monte-Carlo-Tree-Search (MCTS) based cluster detection algorithm

21

Cluster catalogue characterisation projects

• Nathan, Michel, Thibault, Marina, Simona, Rance, Dominique

​

• News :
• [TG] First cosmoDC2 and DC2 AMICO catalogs
• [TG] New WaZP version successfully run at CC-IN2P3 on DC2
• Problem encountered :
• [MR] lack of time to wrap up the project
• [MA] lack of time to wrap up the project
• [TG] Unclear place of the comparison project with respect to individual cluster algorithm projects (code and publication plans)
• Next expected milestone :
• [MR] hopefully a draft
• [TG] Comparison project: focus on a (short) DC2 results publication (redMaPPer/WaZP/AMICO)?

22

Cluster blending projects

• Manon, Marine, Cyrille, Gabriel, Marina, Simona
• News :
• [MR] Gabriel Frugier doing M1 internship on impact of blending on richness determination, other intern working on blending in the APC team
• [SM] We are finding cluster members with the YOLO algorithm, for the moment the results are preliminary and will not present them, however we are asking ourselves: if it will be useful to identify blended galaxies independently with this network.
• [MRa] Development of a new matching algorithm (friendly) which uses a graph structure to gather metrics we can use to better define the (un)recognized blends and study their impact on WL measurements
• Problem encountered :
• ​

23

Cluster blending projects

24

• Next expected milestone :
• [MR] Preliminary results on impact of blending on richness after the internship
• [MRa] Complete version of friendly. Implementation of more metrics. Use of Neural Network to extract properties on blends (cf. Cyrille)

Cluster WL mass

• Céline, Michel, Rance, Calum, Marina…

​

• News :
• [MA]: Preparing for different types redshift inputs
• [MA]: Improve stacking functionality
• [MA]: Several optmizations (cosmo obj., interp., redundancies)
• Problem encountered :
• [MR] Integration taking too much time
• [MR] Lack of emulation, would be good to apply CLMM to data
• Next expected milestone :
• [MR] Integration with other tools during July sprint week

​

25

Cluster cosmology pipeline

• Céline, Constantin, Eduardo, Michel, Marina, Simona

​

• News :
• [MA]: Pipeline diagram
• [MA]: Theoretical prediction code (firecrown and external)
• Problem encountered :
• [MR] Code installation at CCin2p3 (TXPipe)
• Next expected milestone :
• [MR] Progress during sprint week

26

DESC commissioning - cluster

• Céline, Cyrille, Simona, Marina….

​

• News :
• [MR] Commissioning Science Validation Bootcamp 2023 and pre-bootcamp activities by Chris Walter.
• Problem encountered :
• [MR] Couldn’t really follow because I couldn’t have an USDF account, don’t know where to look for information (too many channels), I don’t know how to start and what will be expected from us. I became a DP0 delegate by mistake.
• Next expected milestone :
• [SM] I suggest to meet among ourselves more often to exchange on this process, for the moment each one is by herself/himself, there could be more discussion

27

Photometric redshifts with machine learning

from galaxy images

• Valentin Brekke, Simona Mei, Francois Lanusse

Connected with the photometric redshift impact on cosmology (see Michel Aguena’s talk)

• News : testing of CNN, Resnet, testing and integration of the inception network Pasquet et al. and testing of transformer networks. Creation and coordination of project DESC Project 318: Photometric redshift estimation from galaxy images with machine learning

Working Groups: Photometric Redshifts, 14 participants

• Problem encountered : software integration is progressing in steps, the new transformer networks are less efficient than we expected
• Next expected milestone : inception integration in RAIL: validation on field and cluster galaxies, evaluation of the network on blended galaxies in clusters and propagation to cosmology by the project presented by Michel Aguena.

28

Cluster selection function

• Calum, Jim, Michel?, Anaïs (future PhD student at the APC with Jim)
• Current status:
• A potential LSST project
• Current work being on the Euclid cluster selection function
• Project overview:
• Use numerical simulations (DC2) to look at correlations between cluster properties (both halo and observed)

​

29

Cluster properties

Richness

Redshift

Member concentration

Ellipticity

Inclination

Environment defined with respect to observed galaxies

Miscentering (offset with respect to halo centre)

Halo properties

SOD mass

Lensing mass

Concentration

Inclination

Ellipticity

Environment defined with respect to dark matter distribution

Kinematic mass

Dynamical state

Other projects

• Tbd…

​

• News :
• Problem encountered :
• Next expected milestone :

30

Other projects

• Tbd…

​

• News :
• Problem encountered :
• Next expected milestone :

31