Multi-plane (and source) Cluster Lens Injection and Modelling

[aesar77]

Hi again,

In the past weeks, I tried to implement multi-plane and multi-source cluster lens modelling in SlSim. I attached the example notebook (cluster_lens_demo.ipynb). However, it won't run in the current version of the module due to the modifications I made. The outputs are displayed and hopefully can demonstrate the process. I have some questions in the notebook and below. I'd appreciate any comments, if things I did make sense or not.

Changes I made

• lens.py

  1. added multi_plane as a choice for Lens() initialization. If it is set to True, in lenstronomy_kwargs the LensModel object is cinstructed to be multi_plane. Since this option needs the redshifts if each deflector, deflector_redshift function now records and returns redshifts for each subhalo besides the main deflector.
  2. shear and convergence LOS choices for initialization, which were previously True by default (I did this because I am not sure if convergence is required for clusters with substructures?). If they are True, they inherit the main halo's redshift and added to the LensModel() during its construction.
  3. modified the first condition of the validity check if all of the deflectors are in front of the source if multi_plane is chosen.
  • [Question] In validity test, should we strictly satisfy conditions for the nfw deflector einstein radius or is it enough if they are satisfied for any of the deflectors?

• lensed_system_base.py

  1. minor edits to inherit choices added to Lens() class above.

• cluster_deflectors.py

  1. added assign_galaxy_redshift as a choice for ClusterDeflectors() initialization. If it is True, assign_similar_galaxy function assigns the galaxy redshift sampled in the function to the members, instead of the assigning the cluster redshift to each of them, so that we an have deflectors at various redshifts to create multi-plane lens.
  2. added get_deflector function so that I can chose a deflector by their ID rather than drawing a random lens.
  • [Feedback 1] I think draw_deflector does not check for if index is set to None or not, so it draws it randomly even if it is given? (I did not want to modify this since I figured there might be other functions relying on the randomness.
  • [Feedback 2] draw_members() called by draw_deflectors to create suibhalo population inherits kwargs_draw_members{} arguments, but when it is called directly it takes the default values in the function inputs unless the arguments are given during the call, hence subhalo population and number of members does not match.

• deflector.py

  1. theta_e_infinity function is modified to accomodate multi_plane option for LensModel() creation.

• nfw_cluster.py

  1. added subhalo_redshifts function to return subhalo redshifts (required by the theta_e_infinity function mentioned above)

• lsst_science_pipelines.py

  1. created lens_injection_to_coadd function. I am not sure in which platform SlSim was intented to run, but I realized lens injection notebooks rely on the DC2 database and its Butler. I previously worked with Lsst Science Pipelines, that I used the example database and its Bulter which is for HSC data. In my experience, installing the DC2 on local was not useful since it required the specific sql database and buter. I configured the local data access option for the Rubin Portal but only catalog data access was possible, not images. Therefore I ended up using Rubin Science Platform servers to use get_dp0_images() function and query its instances (coadd, coadd_nImage, mag_zero_visit, variance_map in r, g, i bands) from visits that coincides with the positions from a target list of my interest, and saved them as .fits to use them in SlSim on my local machine. I'd appreciate any suggestion if there is a way to do that. Otherwise, this function might be useful for other users too. It is similar to the lens_injection_fast function, except it takes arguments supplied by get_dp0_images as inputs without needing a butler.

There might be a few minor changes that I forgot to mention, but I hope the notebook I attached makes more sense with these pointed out.

Thanks!

[sibirrer]

Hi @aesar77 thank you very much for posting this very exciting work!
I agree that the current version of SLSim not optimally supports multi-source position rendering. Can you make a pull request of your changes? (don't worry if at this stage other things are breaking, it's just easier to see and discuss how your modifications can smoothly be incorporated)

Your questions in the notebook:
Q1: Yes, we should support more than two sources and it will be interesting to see how many are behind clusters.
Q2: There is no particular limit of the cutout area. But note, the lens equation solver requires a finite spacing/grid and with larger area, the grid will be more sparse (if not adapted to) and this might lead to missing certain images. Larger area while keeping the grid spacing small is computationally expensive. That is why the area is first restricted relative to a conservative estimate of the Einstein radius for the current scenarios of single galaxy deflectors.

[aesar77]

Sure! I have just created the pull request.
For the second question, I offered the cuotut region (that I chose considering the average size of clusters, it might be incorrect) to reposition the sources since multiple deflectors could lens them, unlike single deflector lenses. I think the updated position function is designed considering single deflectors, so it covers a region around a given point that I chose as the main deflector. Maybe if we can give multiple reference points, positions of all deflectors, we can cover finer and smaller regions around them instead of the whole cutout.