e.g.

paper.md

@@ -36,7 +36,7 @@ The PSF characterization methods used by astronomers fall into two main classes:
 Shear Optimization with `ShOpt.jl` introduces modern techniques, tailored to James Webb Space Telescope (JWST) NIRCam imaging, for empirical PSF characterization across the field of view. ShOpt has two modes of operation: approximating stars with analytic profiles, and a more realistic pixel-level representation. Both modes take as input a catalog with image cutouts -- or "vignettes" -- of the stars targeted for analysis. 
 
 # Statement of need
-Empirical PSF characterization tools like Point Spread Function Extractor (PSFEx [@2011ASPC]) and Point Spread Functions in the Full Field of View (PIFF [@Jarvis_2020]) are widely popular in astrophysics. However, the quality of PIFF and PSFEx models tends to be quite sensitive to the parameter values used to run the software, with optimization sometimes relying on brute-force guess-and-check runs. PIFF is also notably inefficient for large, well-sampled images, taking hours in the worst cases. The JWST's Near Infrared Camera (NIRCam) offers vast scientific opportunities [e.g. @casey2023cosmosweb]; at the same time, this unprecedented data brings new challenges for PSF modeling:
+Empirical PSF characterization tools like Point Spread Function Extractor [PSFex @2011ASPC] and Point Spread Functions in the Full Field of View [PIFF @Jarvis_2020] are widely popular in astrophysics. However, the quality of PIFF and PSFEx models tends to be quite sensitive to the parameter values used to run the software, with optimization sometimes relying on brute-force guess-and-check runs. PIFF is also notably inefficient for large, well-sampled images, taking hours in the worst cases. The JWST's Near Infrared Camera (NIRCam) offers vast scientific opportunities [e.g. @casey2023cosmosweb]; at the same time, this unprecedented data brings new challenges for PSF modeling:
 
 (1) Analytic functions like Gaussians are incomplete descriptions of the NIRCam PSF, as evident from Figures 1 and 2. Figure 1 shows that to get a good PSF model we need to model a large dynamic range over a large box size. Figure 2 illustrates that the residuals between an idealized forward model and a Gaussian approximation can be quite severe. This calls for well-thought-out, non-parametric modeling and diagnostic tools that can capture the full dynamic range of the NIRCam PSF. `ShOpt` provides these models and diagnostics out of the box.