NIMBLE (Non-parametrIc jeans Modeling with B-spLinEs)

NIMBLE is a tool for inferring the cumulative mass distribution of a gravitating system from full 6D phase space coordinates of its tracers via spherical Jeans modeling. Spherical Jeans modeling inherently assumes the system is spherically symmetric and in dynamical equilibrium, however, Rehemtulla et al. 2022 show that these conditions are not completely necessary for an accurate mass profile estimate when applied mock Milky Way-like galaxies.

Rehemtulla et al. 2022 (https://arxiv.org/abs/2202.05440) gives much more detail on the routines included here and extensive tests using them.

NIMBLE also includes codes for performing related tasks:

• Creating a variety of equilibrium mock galaxies using the Agama library (https://github.com/GalacticDynamics-Oxford/Agama)
• Creating mock Gaia & DESI observational data of the Latte suite of FIRE-2 cosmological hydrodynamic zoom-in simulations (Wetzel+2016, https://ui.adsabs.harvard.edu/abs/2016ApJ...827L..23W)

Testing with halo alone datasets

NIMBLE includes a bash script titled run_halo_alone.sh which will generate, run Jeans modeling on, and plot a figure for halo alone type datasets of up to 3 provided axis ratios.

The halo alone datasets are generated using Agama (https://github.com/GalacticDynamics-Oxford/Agama) through halo_alone.py located in equilibrium_models/. For example, to create a halo alone dataset with q=0.8 run the following:

python3 equilibrium_models/halo_alone.py 0.8

This will create halo_alone_0.8_prejeans.csv and halo_alone_0.8_true.csv in data/halo_alone, which contain kinematic information of an N-body representation of this model. These files are used in the following step where the NIMBLE inverse modeling Jeans routine is executed on the dataset using jeans_bspline.py. To run it, provide file paths to the _prejeans.csv and _true.csv files created in the previous step.

python3 jeans_bspline.py data/halo_alone/halo_alone_0.8_prejeans.csv data/halo_alone/halo_alone_0.8_true.csv

This will create an assortment of files in results/halo_alone_0.8/ including plots of the velocity, density, velocity anisotropy, and mass enclosed profiles. To create a figure comparing the results of multiple halo alone runs, similar to Fig. 3 in Rehemtulla+2022, run fig3-5.py located in figures/ with the argument halo_alone.

Testing with halo_disk_bulge datasets

Running halo_disk_bulge is very similar to running halo_alone. The run_halo_disk_bulge.sh script will generate, run Jeans modeling on, and plot a figure for the original dataset and its two variants (described in Sec. 3.1 of Rehemtulla+2022).

To generate these datasets manually, use halo_disk_bulge.py in equilibrium_models/ as follows:

python3 equilibrium_models/halo_disk_bulge.py OM

In this example I've optionally added OM which creates the variant with a Cuddeford-Osipkov-Meritt velocity anisotropy profile. Omit OM and you'll generate the original HDB dataset alongside its disk contamination variant, again described in Sec. 3.1 of Rehemtulla+2022.

The process of running NIMBLE's inverse modeling Jeans routine on these and plotting a comparison figure is done identically for these and the halo_alone datasets. Do note that the disk contamination variants share a _true.csv file with their non-disk contamination parents so there will be no HDB_DC_true.csv file.

Testing with Latte FIRE-2 galaxies (without observational effects)

Running the inverse modeling routine on error-free Latte data comes with the additional complication of having to download the Latte data. run_latte_errorfree.sh automates downloading it from yt Hub but it is quite large so it will still take some time. Once the data is downloaded for each Latte galaxy of interest, run_latte_errorfree.sh will prepare the mocks, run the inverse modeling Jeans routine on them, and plot a figure of the resulting mass profiles.

As usual, all these steps can be performed manually with the individual scripts if desired.

python3 read_latte.py m12f

Running the above command will create the m12f_prejeans.csv and m12f_true.csv files used for Jeans modeling. This will also require the gizmo_read package, also available at yt Hub and also automatically set up by run_latte_errorfree.sh. You can then run the inverse modeling Jeans routine in jeans_bspline.py similarly to halo_alone datasets by providing the paths to these two .csv files. The Jeans routine will automatically select the knot configurations shown in Rehemtulla+2022 but can still be configured to use custom knots. Creating a figure comparing the results from the three Latte galaxies can be done by running figures/fig3-5.py with the argument latte.

Testing with Latte FIRE-2 galaxies (with observational effects and deconvolution)

Running the deconvolution routine with observational selection functions and errors imposed on Latte data is a little more involved than running the inverse modeling routine. deconv.py will do most of the work, however. It handles imposing the observational effects, deconvolving them, and outputting the Jeans estimated mass proifle. It's output can then be used in figures/fig7.py to make a copy of Fig. 7 in Rehemtulla+2022. Before running deconv.py, one must download and prepare the Latte data (as described in the previous section) and install a few packages.

The output of each MCMC trial and the final deconvolution output will be saved in a directory for each run. Taking m12f LSR1 with DR4 errors as an example, output will be saved to results/deconv_m12f_LSR1_DR4/. Before the first MCMC trial and after each following one, deconv.py will create figures showing the true, error-imposed, and MCMC-deconvolved density and velocity dispersion profiles. It will also make a corner plot and an MCMC parameter evolution plot after each trial. After convergence, a few text files are written. Most importantly, Menc_beta_final.csv will contain a radial grid and the 16-50-84 percentiles for the enclosed mass and velocity anisotropy profiles.

There is a bash shell script titled run_deconv.sh included to automate the deconvolution routine. When using it, make sure that the configurations deconv.py is run on match those chosen to be plotted in fig7.py (i.e. sims in run_deconv.sh matches sims_to_plot in fig7.py and similarly for lsrs)

To run deconv.py and fig7.py manually use the following commands. Again taking m12f LSR1 with DR4 errors as an example

python3 deconv.py m12f lsr1 dr4
python3 fig7.py dr4

Requirements

Matplotlib 3.0.0 or newer Python 3.6 or newer