Wish list of outputs for a QSO VAC

[paulmartini]

This purpose of this ticket is to organize discussion on the desired contents of a QSO VAC, which would be constructed from a QSO catalog.

The desired contents are emission line and continuum measurements that could be useful for the LyA, GQC, and GQP WGs. For example, we should be able to improve quasar redshifts with information about BH mass and luminosity relative to Eddington. This in turn requires measurements of continuum luminosities, parametric and non-parametric line shape information, removal of FeII emission, and power-law fits to the continuum.

The parent QSO catalog would be the official QSO catalog produced by the LSS WG. Therefore this does not include low-luminosity AGN in galaxies whose continua include significant stellar emission. The rationale is they have quite secure redshifts from the stellar features, and they are usually sufficiently low redshift and luminosity that issues like Fe template subtraction do not apply.

Here are some quantities to start a discussion of the contents:

• Key emission lines: CIV, MgII, Hbeta for BH measurement
• Emission lines to add to emlines.ecsv: He II 1640.42, Al III 1857.40 to add to
• Fit/subtract Fe emission blends (esp. from 2250-3000A around Mg II)
• Fit PL continuum
• Non-parametric quantities: first and second moment + FWHM of the "Key emission lines"
• Continuum luminosities at 1450, 1700, 3000, 5100 Angstroms

Some references on Fe templates are:

• Vestergaard & Wilkes (2001)
• Tsuzuki et al. (2006)
• Salviander et al. (2007)

For reference, here is a figure with a fit to the Fe emission around Mg II from a recent paper led by Zhefu Yu:

Caption: The four colored lines are a PL continuum and the three Fe templates referenced above. Also okay to ignore the single-epoch spectrum.

[moustakas]

@paulmartini I'd like to make some progress on this ticket but I could use your help with the templates.

First, you've suggested three possible template sets, V01, T06, and S07, but if we had to select one, which one would you select? I can fit subsets of objects with all three (in separate calls to FastSpecFit), but I suggest we select a fiducial set (or derive our own from DESI observations themselves??).

Could you (or a student) put the templates into a simple FITS format that I can use (and put them somewhere I can access on NERSC)? For example, here are the current templates--
https://data.desi.lbl.gov/public/external/templates/fastspecfit/README.txt
https://data.desi.lbl.gov/public/external/templates/fastspecfit/1.0.0/

Second, can you provide more input regarding the power-law fitting you would like me to perform? I did play around with this previously and used a simple F_nu \propto \nu^(-alpha) or F_lambda \propto \lambda^(alpha-2) model, with alpha ranging between 0.5 and 2.0. Is this what you had in mind, too? For reference, this is the paper I've been looking at, but let me know if you have other favorites--
https://www.aanda.org/articles/aa/pdf/2017/08/aa30378-16.pdf

Finally, do you want me to try to fit the host-galaxy continuum, too, or can we safely ignore it?

I'm looking forward to hearing your thoughts.

[paulmartini]

Hi @moustakas --

Great that you plan to work on this soon!

I've made a fits file with four templates and placed them in a single fits file here:
/global/cfs/cdirs/desi/users/martini/qsospecfit/fetemplates.fits
The first (EXTNAME = DEFAULT) is the variant of the Vestergaard & Wilkes (2001) template we've used the most. The three other templates are another Vestergaard & Wilkes template and the two others mentioned above. The EXTNAMEs should be clear, and there is more information in a COMMENT in each extension. Let me know if you'd like these arranged differently.

I suggest just using the one I called 'DEFAULT.' Zhefu Yu has done experiments with the others and not found big differences. The big disadvantage of the Vestergaard & Wilkes template is that there is not empirical data at the location of MgII, and ignoring an FeII contribution at those wavelengths probably leads to an overestimate of the MgII line flux and FWHM. Yet for scaling relations this may not matter too much, given the calibration process.

The power-law fit you mention is a good match to observations of the ultraviolet continuum, namely alpha = 0.5 to 2.0 where F_nu \propto \nu^(-\alpha).

For quasars we can ignore the host-galaxy continuum. I think it is okay to ignore it.

[moustakas]

@arjundey has recommended adding CII] 2326 to the line-list.

[moustakas]

Some notes / papers for myself:

• https://github.com/imcgreer/simqso/blob/master/examples/SimpleSpecExample.ipynb
• https://github.com/MJTemple/qsogen
• https://iopscience.iop.org/article/10.3847/0004-6256/151/6/155/pdf
• https://arxiv.org/pdf/2109.04472.pdf

[moustakas]

Whittle+85 (https://ui.adsabs.harvard.edu/abs/1985MNRAS.213...33W/abstract) recommend a few non-parametric line-profile measurements that may be useful.

[moustakas]

Some additional AGN/QSO modeling refs (courtesy @bd-j!)--

Lyu+22 added semi-empirical AGN optical-FIR templates, and different SF dust emission templates - this is built off a very old prospector version, and I think uses a custom FSPS:
https://ui.adsabs.harvard.edu/abs/2022ApJ...941..191L/abstract

This adds a power law continuum with attenuation independent from the stars - I hope it will be added to prospector soon.
https://ui.adsabs.harvard.edu/abs/2024arXiv240302304W/abstract
https://ui.adsabs.harvard.edu/abs/2024arXiv240302304W/abstract

This uses Lyu+22 and also a version with just a scaled Polletta 2006 QSO template added on:
https://ui.adsabs.harvard.edu/abs/2024ApJ...968....4P/abstract
https://ui.adsabs.harvard.edu/abs/2024ApJ...963..128B/abstract

This adds a scaled empirical AGN emission line template from Richardson 2014 - this is actually in prospector as AGNSpecModel, but was really just used to explore the potential impact of AGN emission lines on photo-as
https://ui.adsabs.harvard.edu/abs/2022arXiv220802794N/abstract

There's been some other work about including emission lines more flexibly, including produced by AGN-like ionizing continuua, in
https://ui.adsabs.harvard.edu/abs/2024arXiv240504598L/abstract
https://ui.adsabs.harvard.edu/abs/2024ApJ...969L...5L/abstract

[moustakas]

@paulmartini, @yzfkieran817 @wei-jian

Below is the VW01 template (originally provided by @paulmartini), resampled to constant-velocity pixels and then smoothed by a Gaussian kernel ranging in sigma between 150 and 10,000 km/s.

A couple questions:

• I've resampled the template to have constant 75 km/s pixels (the original template has constant-lambda, not constant log-lambda). Is this reasonable?
• Is the 150-10,000 km/s prior range reasonable?
• In the original template there are several wavelength ranges which are identically zero, including around the MgII wavelength range. Do I need to reset these flux values to zero after smoothing by the velocity dispersion factor?

[paulmartini]

Hi @moustakas --
Here are my answers (and @yzfkieran817's):

1. Yes, the resampling to 75 km/s pixels seems reasonable.
2. Yes, the prior range of 150-10,000 km/s looks reasonable.
3. No need to reset the values to zero in the wavelength ranges where they are zero in the template.

[moustakas]

@paulmartini @yzfkieran817

Regarding this request:

Non-parametric quantities: first and second moment + FWHM of the "Key emission lines"

I have implemented something (currently in a branch) but would be grateful if you could check my math and code here and make sure it's what you want / need. In particular, there are a variety of

Specifically, I isolate the pixels in the continuum-subtracted (not yet iron-subtracted, but that's in the works) spectrum centered on each of the following four lines: CIV, MgII (mean wavelength of the doublet), Hbeta, and [OIII] 5007.

The central wavelength and width are taken to be the Gaussian-weighted center (based on my line-fitting) +/- 5-sigma, where "sigma" is again based on the Gaussian fit to the line.

Then, I simply (and non-parametrically) compute:

$M_{1} = \sum_i (W_{i} F_{i}) / \sum_i F_{i}$
$M_{2} = \sum_i (W_{i} - M_{1})^{2} F_{i} / \sum_i F_{i}$
$M_{3} = \sum_i (W_{i} - M_{1})^{3} F_{i} / \sum_i F_{i}$

where $W_i$ and $F_i$ are the wavelengths and fluxes, respectively, and $M_1$, $M_2$, and $M_3$ are the three moments in observed-frame Angstrom, Angstrom$^2$, and Angstrom$^3$, respectively.

There are many definitions of these moments, including the standardized (normalized) moments, which I'm not computing, but maybe this is what you want / prefer?

As a quick example, here's CIV on sv3/dark/25960/39627776256380593

Note that I would like to use these calculations to help identify / pre-select spectra with possible outflows, particularly using [OIII] 5007, so it would be great to use maximally useful definitions.

@stephjuneau @Ragadeepika-Pucha others...

[paulmartini]

This looks good to me, @moustakas. Specifically, I think it is better to compute the moments as you've done, rather than the standardized/normalized versions, and it is better to report observed-frame quantities. These should be more straightforward to use for the primary science drivers of these additions, namely outflows and centroid shifts associated with these emission features. I also looked over the code and the implement matches what I expected. Thanks!

[stephjuneau]

Hi @moustakas , @paulmartini et al.

For [OIII]4959,5007, the most common method for non-parametric measurements that I see in the literature is using the cumulative flux velocities like this figure (from Speranza et al. 2024):

And here are other examples from the QSOFEED project from Bessiere et al. 2024:

I suspect that the reason Chiara is using these measurements in her work is that they're very commonly used so that would allow us to make direct comparisons with the literature. There are many more examples but just another selected highlight is this work by Davies et al. 2020 may be one of the most careful analyses of mass outflow rates for ionized outflows in AGN that I'm aware of. See their Section 3.2 for velocities, where they use v_peak (as the offset between the velocity of [OIII] peak and systemic velocity) to determine if [OIII] is mostly outflow or ISM and v98 (which as they define it is the max of |v02| and v98). Then their definition of v98 is used directly as "v_out" to compute the mass outflow rate and kinetic energy:

For funsies, this figure shows a range of expected profiles for [OIII] bicones depending on inclination i and on dust extinction:

This paper by Bae & Woo (2016) also has a lot of detailed predictions of line profiles from modeling the outflows in 3D, and here's a passage about bulk velocities for outflow properties and other references:

All that said, maybe Paul has read different papers that instead used quantities like you're proposing? Could you please post some examples as I'd be interested in taking a look! IMO, any of the quantities will be interesting but I'm definitely more familiar with the cumulative-flux velocities to infer physical outflow properties.

[moustakas]

@stephjuneau thank you for these additional references and suggestions.

I'll let @paulmartini comment regarding the definitions he prefers, but I'm hopeful that the three moments currently coded up are adequate for at least pre-selecting objects / emission lines for more detailed analysis.

[paulmartini]

Hi @moustakas, @stephjuneau - My motivations were to identify lines centroids that did not match the systematic redshift, as we care about redshift errors for the LyA / QSO correlation analysis, and to identify objects with non-Gaussian line shapes, which might provide further information about the reliability of the redshift measurements. I suggested the three moments because they are conceptually simple and statistically robust. I think that they should suffice to identify targets for more detailed study for other science goals, such as the study of AGN winds. I suspect those will be a small minority of the total sample, although I defer to John on whether or not he wants to add more line measurements for every object.