Custom Usage#

MUFASA’s default pipeline is built using modular components that can be reconfigured for custom usage and pipeline builds. Here, we will walk you through the purpose and usage of these components.

Modular Fitters#

The modular functions behind MUFASA’s default pipeline operate primarily within the following methodological categories: signal boosting, quality masking, residual searching, and spatial expansion. These categories are not necessarily mutually exclusive and all use corrected Akaike Information Criterion (AICc) to assess the fit quality. Let’s have a closer look at MUFASA’s modular line fitters within these categories:

Signal Boosting#

iter_2comp_fit() achieves signal boosting by first fitting a spatially convolved cube using moment-based initial guesses. This function then undergoes a second iteration of fits using results from the initial fit for the subsequent guesses.

Use when:

  • Wanting to detect emissions deeper into the noise

  • Interested in fainter, larger-scale emissions

Omit when:

  • The targeted emission already has a strong signal

  • The targeted emission can not be resolved by the convolved beam

Quality Masking#

Quality masking involves identifying where a model is poorly fitted using AICc-derived relative likelihood values to compare a model to its simpler counterpart, accounting for extra parameters against over-fitting.

refit_bad_2comp() specifically flags where the log-likelihood of a two-component model relative to a one-component model (i.e., \(\ln{K^2_1}\)) is below a specified threshold (default -5) and uses get_refit_guesses() to find the best-fitted neighbours of the flagged (i.e., bad) pixels and use those neighbours’ fits as guesses to refit bad pixels. The refitting of these bad pixels is carried out by replace_bad_pix().

The usage of get_refit_guesses() in tandem with replace_bad_pix() is a powerful and versatile tool for refitting bad pixels using the information from the good fits nearby. Use these two functions when:

  • attempting to recover poorly fitted pixels near decent fits

  • attempting to fit new pixels in the proximity of already-fitted pixels (see Spatial Expansion)

Residual Searching#

Residual searching looks for signals in the fit residual by attempting to fit another single-component model to the residual using moment guesses. Residual searching performed by refit_2comp_wide() also employees signal boosting by convolving the residual first before attempting the refit. The residual’s fitted model is subsequently used as guesses for the ‘’missing’’ second component alongside results of the original one-component fit as guesses for the other component in attempts to recover the potentially missing component.

Use when:

  • Looking for a spectral component with a significantly different velocity than that of the pre-existing model

  • Attempting to fit an additional spectral component on top of pre-existing model

Spatial Expansion#

Spatial expansion is closely related to quality masking in that they both use existing models in the nearby pixels to help (re)fit pixels using get_refit_guesses() alongside replace_bad_pix(). Spatial expansion, however, is distinct from quality masking in that it expand the fits into unfitted surroundings rather than attempting to recover pre-existing poor fits.

refit_marginal() is MUFASA’s higher-level wrapper that performs spatial expansion, which employs expand_fits() internally to fit the surroundings iteratively. MUFASA’s default pipeline uses these tools to expand fits from robust, high signal-to-noise regions into their fainter surroundings. For quality assurance, MUFASA further employs refit_marginal() to first refit pixels with models that are only marginally better than their simpler counterpart before using them as initial guesses for the spatial expansion.

Use when:

  • Trying to fit the fainter emissions using pre-existing fits

  • Wanting to improve fitting efficiency by leveraging pre-existing fits for guesses over spectral moments

Avoid when:

  • The surrounding emission potentially has gas properties (e.g., line-of-sight velocity) that are significantly different from those of the already-fitted emission