deletions | additions       diff --git a/Analysis Plan.tex b/Analysis Plan.tex  index a05b7e5..9fbe576 100644  --- a/Analysis Plan.tex  +++ b/Analysis Plan.tex 
...
  Extraction of galaxy structural components through automatic fitting has been developed in a variety of codes, among the most widely-used of which is GALFIT \citep{Peng_2002}. The original version of GALFIT models azimuthal 2-D profiles for an arbitrary number of components, and but is also capable of handling a variety of radial functional forms in addition to more irregular morphologies (such as warps or boxy components). Advanced versions of GALFIT \citep{Peng_2010} provide additional profile components often seen in high-resolution images of galaxies, such as truncated shapes, rings, irregular morphologies, and power-law spirals.    Fitting parametric models automatically to galaxy images are an extremely powerful method of quantifying the morphology and the relationship between the galaxy's dynamical state and its luminosity. One of the major challenges to this approach, however, is running unsupervised decomposition codes on large samples of images. While a number of error minimization algorithms can be used to determine the ``best-fit'' of a model to an image, images of galaxies with extensive structure (especially for HST images with sub-arcsecond resolution) often mean that the appropriate number of model components can be quite large (and impossible to determine \textit{a priori}). In addition, the accuracy of fitting the model parameters (eg, from $\Chi^2$ $\chi^2$  minimization) can be sensitive to relatively small changes in the initial conditions. For example, a model may pick unusual and likely un-physical radial profiles to get a formal minimization of the residuals if the center position of the galaxy is off, sometimes by as little as a few pixels. Finally, galaxies are real objects that are observed surrounding (and sometimes obscuring or overlapping) many other objects; this can include image artifacts, foreground stars in the Milky Way, and nearby companion galaxies. Deciding which of the objects should be modeled (and which components should be used) is a non-trivial task for real data.   This approach benefits significantly from the interaction of humans with the modeling software; while the profile-fitting code is capable of quickly matching the light and providing instant feedback (by viewing both the output model and the residuals after subtracting it from the image), the pattern recognition capabilities of humans are critical for a good solution by specifying the number and type of components and by keeping the values for the model within reasonable physical bounds. Citizen scientists have already demonstrated the ability to distinguish between complicated models of galaxies for $N$-body simulations of mergers \citep{Holincheck_2016}, for example.