Characterization of JWST NIRCam PSFs and Implications for AGN+Host Image Decomposition

We present a detailed analysis of the point spread function (PSF) of JWST NIRCam imaging in eight filters: F070W, F115W, F150W, F200W, F277W, F356W, F444W, and F480M, using publicly available data. Spatial variations in the PSF FWHM generally decrease with wavelength: the maximum and RMS fractional variations are $\sim20\%$ and $5\%$ in F070W, reduced to $\sim3\%$ and $0.6\%$ in F444W. We compare three commonly-used methods (SWarp, photutils, and PSFEx) to construct model PSFs and conclude that PSFEx delivers the best performance. Using simulated images of broad-line AGNs, we evaluate the impact of PSF mismatches on the recoverability of host galaxy properties. Host fluxes are generally overestimated when adopting mismatched PSF models, with larger overestimation for more AGN-dominated systems. Broader PSFs tend to produce less concentrated hosts while narrower PSFs tend to produce more concentrated and compact hosts. Systematic uncertainties in host measurements from PSF and model mismatches are generally larger than the formal fitting uncertainties for high signal-to-noise ratio data. Image decomposition can also lead to an artificial offset between the AGN and host centroids, which is common (e.g., $>1\sigma$ [$3\sigma$] detection in $\sim 80%$ [$\sim 20-30\%$] of systems), and scales with the mean host surface brightness. Near the surface brightness limit, this artificial offset can reach as large as $\sim80\%$, $26\%$, and $7\%$ of $R_e$ in systems with $R_e=$0.12", 0.48", and 1.92", respectively. We demonstrate our PSF construction and image decomposition methods with an example broad-line quasar at $z=1.646$ in the CEERS field.