NIFTY

Near-Infrared Fitting for T and Y Dwarfs

Kevin Hainline, Jake Helton, and Evan Chen

NIFTY is a code designed to fit JWST NIRCam/MIRI photometry or NIRSpec prism spectroscopy of cold brown dwarf candidates with the LOWZ (Meisner et al. 2021), ATMO2020 (Phillips et al. 2020), Sonora Flame Skimmer (Mang et al. 2026), Sonora Elf Owl v2 (Mukherjee et al. 2024, Wogan et al. 2025), or earlier Sonora Elf Owl (with PH3, Beiler et al. 2024) atmospheric models, using a Bayesian framework with the emcee sampler (https://emcee.readthedocs.io/en/stable/user/sampler/) or the Nautilus sampler (https://nautilus-sampler.readthedocs.io/en/latest/). This code was described in Hainline et al. (2026) (doi.org/10.48550/arXiv.2510.00111).

NIFTY operates in two modes:

• phot — fit broadband photometry from a catalog of one or more sources.
• spec — fit a single NIRSpec prism spectrum.

In both modes, NIFTY produces a corner plot, a best-fit SED or spectrum plot, and a text file with the 16th, 50th, and 84th percentiles of all fit parameters. Photometry fits typically converge in under 3 minutes per source for ~14 bands, though convergence time varies, especially at high SNR. Spectroscopic fits are slightly longer.

Installation

The installation comes with the NIFTY Python package and the commands nifty and build_model.

The easiest way to install NIFTY is through pip:

pip install astro-nifty

The main pip installation only includes the dependencies required for sampling. If you also want to build your own models, install the optional dependencies:

pip install astro-nifty[build-model]

If you are installing with pip, it is recommended that you install within a virtual environment to keep dependencies clean. See the Python tutorial on virtual environments and packages for more information.

Alternatively, you can clone the NIFTY repository and set up an environment using conda or micromamba:

conda env create -f environment.yml
conda activate NIFTY

Running NIFTY

The easiest way to run NIFTY is through the nifty command. NIFTY runs in two modes: photometry mode and spectroscopy mode, which require slightly different arguments.

To see the available arguments, you can run: nifty phot -h or nifty spec -h to display the help message for that mode.

Photometry mode (nifty phot)

Fit broadband photometry for one or more sources from a catalog.

Single source:

nifty phot \
  --model path/to/model_grid.tar.gz \
  --stub JADES-GS \
  --config BD_NIRCam_MIRI_filters.json \
  --id 20541 \
  path/to/photometry_file.fits

List of IDs from a file (one ID per line):

nifty phot \
  --model path/to/model_grid.tar.gz \
  --stub JADES-GS \
  --config BD_NIRCam_MIRI_filters.json \
  --idlist all_source_IDs.dat \
  path/to/photometry_file.fits

Multiple IDs:

nifty phot \
  --model path/to/model_grid.tar.gz \
  --stub JADES-GS \
  --config BD_NIRCam_MIRI_filters.json \
  --id 20541 --id 452029 --id 430165 \
  path/to/photometry_file.fits

Spectroscopy mode (nifty spec)

Fit a single NIRSpec prism spectrum. The spectrum file should be a whitespace-delimited text file with three columns: wavelength (microns), flux (nJy), and flux error (nJy). The --id argument is used only to label output files.

nifty spec \
  --model path/to/model_grid.tar.gz \
  --stub JADES-GS \
  --id 20541 \
  path/to/spectrum.txt

Optional arguments

Configuration File (--config)

NIFTY uses a single JSON file to describe all filter information needed for photometry fitting. Here is a sample entry:

{
  "filter_columns": {
    "JWST_F115W": {
      "extension": "CIRC",
      "flux": "F115W_CIRC3",
      "error": "F115W_CIRC3_e",
    },
    "JWST_F444W": {
      "extension": "CIRC",
      "flux": "F444W_CIRC3",
      "error": "F444W_CIRC3_e",
    }
  }
}

The filter name must match a valid sedpy filter. If no telescope prefix is provided, the prefix is assumed to be "JWST_".

• extension: the FITS HDU name where this filter's data lives. Ignored for plain-text catalogs.
• flux / error: exact column names in the photometry file.

An example config file for JADES NIRCam + MIRI observations is included as BD_NIRCam_MIRI_filters.json.

Photometry File Format

NIFTY can read:

• FITS files, with filter fluxes and errors in named HDU extensions and columns as specified in the config JSON.
• Plain-text catalogs, whitespace-delimited, one row per object.

For plain-text catalogs, include an ID column (case-sensitive) and flux/error columns whose names match the config JSON exactly. Do not prefix the header line with #. Example:

ID     F115W_CIRC3     F115W_CIRC3_e     F444W_CIRC3     F444W_CIRC3_e
101    0.123           0.010             0.456           0.025

All fluxes and errors should be in nJy. A minimum relative flux error of 5% is applied automatically; errors below this floor are inflated to 0.05 * flux.

Output Files

For each fitted source, NIFTY writes the following files to the output directory:

Each file name is prefixed with <stub>_<ID>_<Model>_.

Model Parameters

All fits assume a source radius of 1 Jupiter radius. Distances are always reported in parsecs.

The ModelGrid Files

While you can create your own ModelGrid files for fitting (see below), we have prepared files for the ATMO2020, LOWZ, Sonora Elf Owl, SonoraElfOwlPH3, and SonoraFlameSkimmerQuickstart fits at this link:

https://arizona.box.com/v/NiftyModelGrids

These files range in size from 2 - 300 MB, and you can point NIFTY at these to run fits. They include all of the JWST NIRCam and MIRI filters, as well as the Roman WFI filters for use with future Roman HLWAS observations.

Creating the ModelGrid Files

Before running NIFTY, download the model grids you want to fit against, then run build_model to build the model grid file that NIFTY uses during fitting.

For a more detailed description of parameters, run build_model -h.

build_model \
    --path path/to/model/files/ \
    --config BD_NIRCam_MIRI_filters.json \
    ModelName

The valid model names are SonoraFlameSkimmerQuickstart, SonoraElfOwl, SonoraElfOwlPH3, ATMO2020, LOWZ. At this time, NIFTY does not support creating the SonoraFlameSkimmer complete models from scratch.

Typically, there are two ModelGrid files outputted. The raw grid is the grid created by reading in all the points that are part of the model. However, many models do not have values at all grid points, so missing points are filled. with NaN. The complete grid has values for all grid points. If the raw grid is missing values, they are filled by a fitting procedure (see code for details). If the raw grid is already complete, only the complete grid is saved.

By default, the output model grid files are named <Model>_raw_ModelGrid.tar.gz and <Model>_complete_ModelGrid.tar.gz.

The Sonora Elf Owl model grid spans several tens of GB and can take a few hours to process. The other grids are considerably smaller.

Expected Model File Layouts

SonoraFlameSkimmerQuickstart — A single zip file:

spectra_R3k_quickstart.zip

SonoraElfOwl — directory of temperature-range tarballs containing NetCDF files:

Sonora_Elf_Owl_v2/
    teff_275_325.tar.gz
    teff_350_400.tar.gz
    ...

SonoraElfOwlPH3 — a single .npz file:

elf_owl_disequilibrium_PH3.npz

ATMO2020 — per-metallicity subdirectories containing .dat files:

meisner_2023/
    grid_m1.0/
        spec_jwst_t700_g3.5_m1.0_kg_g3.5.dat
        ...
    grid_m0.5/
    grid_p0/
    grid_p0.3/

LOWZ — CSV index and models tarball

LOWZ/
    LOWZ_models_index.csv
    models.tar.gz

References

• Meisner et al. 2021 (LOWZ): https://doi.org/10.3847/1538-4357/ac013c
• Phillips et al. 2020 (ATMO2020): https://doi.org/10.1051/0004-6361/201937381
• Mukherjee et al. 2024 (Sonora Elf Owl): https://doi.org/10.3847/1538-4357/ad18c2
• Wogan et al. 2025 (Sonora Elf Owl v2): https://doi.org/10.3847/2515-5172/add407
• Beiler et al. 2024: https://doi.org/10.5281/zenodo.11370829
• Foreman-Mackey et al. 2013 (emcee): https://doi.org/10.1086/670067
• Lange 2023 (nautilus): https://doi.org/10.1093/mnras/stad2441
• Mang et al. 2026 (Sonora Flame Skimmer): https://doi.org/10.48550/arXiv.2608.06454