imcgreer
commented
6 years ago To answer the question, yes, the emission line parameters are based on absorbed spectra. They were derived by fitting gaussian profiles to composite spectra derived from BOSS quasars, and thus at a mean redshift of ~2.5. In terms of the lost flux, that is compensated for by a rescaling of the line flux here: https://github.com/imcgreer/simqso/blob/master/simqso/sqmodels.py#L140
def BossDr9_EmLineTemplate(*args,**kwargs):
kwargs.setdefault('scaleEWs',{'LyAb':1.1,'LyAn':1.1,However, there are some caveats in trying to understand the behavior you report in desihub/redrock#104. First, while the line parameters are derived from absorbed spectra, the line profiles are symmetric gaussians. If no forest is applied, a Lyman alpha line from simqso will look the same at z=2 and z=5. I'm guessing this is not the case for the PCA spectra, which are redshift-dependent, and that is why you see a linearly increasing trend in the figure attached to desihub/redrock#104.
With the simqso templates you see a constant offset. I believe the issue is a difference in the ground truth between simqso and the redshift fitting method. For simqso, the composite spectra used to derive emission line parameters were obtained by transforming BOSS quasar spectra to the rest-frame using their catalog redshifts (and this was a fairly early version of the redshifts). This means they carry both the systematic biases and the uncertainties present in the BOSS redshifts. Thus the line centers may be systematically blueshifted or redshifted, and the line profiles are likely artificially broadened due to redshift errors. Because simqso uses line wavelengths derived from the composite spectra, it could be that the Lya line is somewhat artificially redshifted due to the presence of absorption in the composites.
That said, where does the ground truth come from in the fitted redshifts? If it is also a PCA method, then likely the agreement between input and output just comes from circular reasoning, i.e., you're getting back what you put in. The constant offset between simqso and the fitted redshifts could instead be a bias in the fitted redshifts.
This snippet shows how to retrieve the (observed) rest-frame wavelengths of the broad and narrow Lya line components used by simqso, in three absolute magnitude bins (M_1450=-26,-24,-22):
In [11]: from simqso.sqgrids import *
In [12]: emLineVar = generateBEffEmissionLines(array([-26.,-24.,-22.]),OnlyLines=['LyAb','LyAn'])
In [13]: emLineVar(3)[:,:,0]
Out[13]:
array([[ 1216.23815918, 1216.49499512],
[ 1216.29370117, 1216.7064209 ],
[ 1216.30151367, 1216.76159668]], dtype=float32)This could be compared to Lya line properties for low-z quasar spectra unaffected by IGM, e.g. this catalog http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/MNRAS/400/1493/sp_fit from this work http://adsabs.harvard.edu/abs/2009MNRAS.400.1493K
If it is found that the line wavelengths in simqso have systematic biases then a workaround could be introduced to modify the template wavelengths according to user specification. This is what was done with the scaleEWs option will allows users to arbitrary rescale the equivalent widths of individual lines.
I have used SIMQSO spectra to assert desi's redshift fitter performance. In PR https://github.com/desihub/redrock/pull/104, I seem to see that the parameters you use for all emission lines blue of the Lya (including Lya) have been extracted from absorbed spectra. Is it true? How do you get the parameters giving your Lya emission line in SIMQSO? Would it be easy to get these parameters from unabsorbed data? i.e. data corrected for absorption.