Systematic error in quasar redshift due to Lya forest

[londumas]

The following plot shows 1000 quasars generated with desisim.templates.QSO() and fitted with redrock. Blue points give results with Lya forests, orange points give results without Lya forests The idea is to test to convolve the first eigenspectra of quasar templates with the PDF (lya+lyb+...) and see if the systematic error can be removed or reduced.

[londumas]

Tested with a simple model, and seems to work well. A more complete model should take the resolution of pixels into account, the presence of Ly-beta, Ly-gamma, ...

def mean_transmitted_flux_fraction(lRF,zpix):
     f = np.ones(lRF.size)
     w = lRF<1215.67
     f[w] = np.exp(-0.0023*(1.+zpix[w])**3.64)
     return f

Needs to modify redrock at two different places:

- py/redrock/templates/DistTemplate

        for i, z in enumerate(myz):
            for k in list(self._dwave.keys()):
                lRF = self._dwave[k]/(1.+z)
                zpix = self._dwave[k]/1215.67-1.
                data[i][k][:,0] *= mean_transmitted_flux_fraction(lRF,zpix)

- py/redrock/fitz/fitz()

            for k in list(dwave.keys()):
                lRF = dwave[k]/(1.+z)
                zpix = dwave[k]/1215.67-1.
                binned[k][:,0] *= mean_transmitted_flux_fraction(lRF,zpix)

[londumas]

Here are two plots giving:

• blue: master ( no correction )
• orange (correcting for the transmitted flux fraction)
• green (correcting for the transmitted flux fraction + capping the data weight blue-ward of the forest) The black vertical lines give the redshift when the CIV and Lya lines appear in the data (i.e. lObs>3600.)

[londumas]

Same result using desisim.templates.SIMQSO()

[londumas]

@sbailey, here is the current status on the issue: The current fix works very well on desisim, but it fails on BOSS data. The reason is that the current PCA base is not corrected for the Lyman-series absorption. See issue https://github.com/desihub/desisim/issues/323. The solution would be to produce a new quasar PCA base where the data would be corrected for the absorption in the Lyman-series and in the Lyman-limit. What do you think?

[moustakas]

Why not just re-derive coefficients from a QSO template set generated using simqso?

[londumas]

@moustakas, why not. Could you link me to the doc or a paper explaining simqso? I need to know how the spectra are build and if they are unabsorbed or have the Lyman-series and the Lyman-limit.

[sbailey]

Progress has been made on this via #104, but more work is to be done on the template side, so leaving this open for now and moving from 18.3 to 18.6 project.

[londumas]

Similar issue as https://github.com/desihub/redrock/issues/106