search

changhoonhahn / SpenderQ

MIT License
2 stars 0 forks source link

sigma calibration #4

Open changhoonhahn opened 4 weeks ago

changhoonhahn commented 4 weeks ago

Currently, SpenderQ takes in one user input: the number of sigmas used to identify absorption features from the difference between the reconstruction and the spectra.

Different values of sigma produce significantly different reconstruction residuals, as I explored in doc/methods_paper/sigma.ipynb. Here's a figure that summarizes this: image

In principle, the sigma is set by the number of spectral bins that have absorption, which is in turn set by the amount of neutral hydrogen. However, we need a data-driven way to determine the sigma, rather than calibrating it based on sims.

changhoonhahn commented 4 weeks ago

Here's a distribution of $(\tilde{f}_q(\lambda) - f_q(\lambda))/\sigma(\lambda)$ for the Ly$\alpha$ range of 1024 spectra: image I've overplotted 3 component Gaussian Mixture Model (GMM) fits. The GMMs identify the component that corresponds to the continuum and based on this we can see that a $1\sigma$ cut overestimates the continuum, while the $1.5\sigma$ and $2\sigma$ cuts underestimate the continuum.

changhoonhahn commented 4 weeks ago

Based on this GMM-based procedure, the cutoff for LyA should be 1.1-1.2sigma and for LyB should be ~0.8sigma. I'll run SpenderQ with this setup next.