Classifications_ABD_v1.2.fits

Alyssa Drake & Dan Smith - 12 July 2024

[Changes since v1.1: r_50 cut implemented, z-Ha luminosity cut implemented (see paper text)]

In these classifications we have taken a probabilistic approach to
determine the likelihood of each object falling in a particular class,
given the measurements and uncertainties that we have available. This
can be particularly advantageous since it enables different science
goals (e.g. do you wish to have a sample with high purity, or high
completeness). To this end, we have used Monte Carlo simulations to
probabilistically determine (i) whether each source has a radio
excess, and (ii) what constraints we are able to put on the emission
line classifications. To do this, we produce 1000 realisations of each
measurement (whether spectral line or radio flux density) and run each
realisation through the workflow described below. Finally we bring
this information together to produce overall classifications which
combine the two sets of ingredients -- (i.e. radio excess and BPT
class) where possible -- considering each realisation is equally
likely, to infer the likelihood of a source falling in each class
(broadly SFG, RQAGN, HERG, and LERG - though the latter class is
complicated due to the prevalence of sources without emission lines,
as we discuss below). We have used as input the Portsmouth catalogue
based on DR8 of SDSS and BOSS DR12 as described in Thomas et al. 2013,
together with the LoTSS DR2 catalogue (v1.1) as described in
Hardcastle et al. 2023. Below we have produced a summary of some key
features - the key reference document describing this work is the
forthcoming Drake et al (in prep).

Radio excess workflow:

Radio excess measurements are implemented for all objects with an Ha
measurement in the Portsmouth catalogue, by comparing with the 150 MHz
luminosity. If Hbeta information is also available, the Halpha
luminosity is Balmer corrected assuming an intrinsic line ratio of
2.86 and that the difference is due to dust. If for a given
realisation the implied line ratio suggests negative A_{Ha}, we do not
apply the correction. For those sources that have Halpha measurements
but not Hbeta, we have attempted to produce an average correction, by
randomly drawing A_{Ha} values from the distribution recovered for
sources with 5sigma detections in both lines. To identify these
sources where an average Balmer correction has been applied for the
end user, we have introduced a flag on the radio excess measurement
(corresponding exactly to each row where Hb=0). We do not consider
objects without an Ha detection in the Portsmouth catalogue, and
consider only redshifts where Halpha and NII6583 are detectable in the
SDSS/BOSS spectrographs.

Emission line classifcation workflow:

For the subset of objects which have all 4 BPT lines measured, we have
computed probabilistic emission line classifications again by
marginalising over the 1000 MC realisations of the Portsmouth
catalogue obtained by perturbing the measured fluxes by drawing random
numbers from a standard normal distribution multiplied by the
catalogue measurement errors. In this way, by marginalising over the
realisations we have determined the likelihood of each source
occupying each of different sections of the BPT diagram.

The catalogue contains the following columns relating to the above
classification workflows:

For all objs in this cat:

RADIO_Excess (value between 0 and 1 indicating the fraction of
realisations that have a radio luminosity in excess of what can be
expected on the basis of the Balmer-Corrected Halpha emission, at >99%
confidence)

BALMER_CORR_WARNING - indicates no Hb measurement. Balmer correction
was randomly drawn from the distribution of AHa values that arise from
all 5 sigma Ha sources.

CLASS_z_WARNING - we have noted that the z_best in the LoTSS catalogue
differs from the Portsmouth redshift by >= 0.001. Exercise extreme
caution if you wish to consider objects with this flag set to 1.

For objs with 4 BPT lines: BPT classifications

BPT_SFG - fraction of realisations for each object falling to the left of the Kauffman 03 line
BPT_COMP - fraction falling between the Kauffman and Kewley lines
BPT_SEY - fraction of realisations that are Kauffman AGN and which fall to the left of the SEYFERT line from REFERENCE
BPT_LIN - fraction of realisations with are Kauffman AGN and to the right of the SEYFERT line (i.e. LINERS)
BPT_CSEY - fraction of realisations falling between the Kauffman and Kewley lines and to the left of the SEYFERT line
BPT_CLIN - fraction of realisations falling between the Kauffman and Kewley lines and to the right of the SEYFERT line

BPT_ML - maximum likelihood BPT classification (i.e. using fluxes
measured in Portsmouth catalogue). Note that this is included for
completeness only, we do not recommend using this column other than
for checks, and considering in conjunction with the Zscore
(below). Key: s = SFG, S = Seyfert, L = Liner, X = Composite Seyfert,
Y = Composite LINER

zscore - the significance of the deviation from the null hypothesis
that the maximum likelihood classification is correct (i.e. consistent
with the monte carlo realisations). We recommend that you consider
only those sources with have a zscore < 2.5 since emission line ratios
can misbehave at low statistical significance (as discussed
in Drake et al.).

Anything without all 4 BPT lines was set to NaN in all of the
classification columns (though a radio excess probability is still
determined if Halpha is present).

With these ingredients, we have produced the following classifications:

Overall Classifications:
CLASS_SFG - fraction of realisations with no radio excess & BPT SFG classification
CLASS_RQAGN - fraction of realisations with no radio excess & BPT AGN (Kauffman)
CLASS_HERG - fraction with radio excess & BPT SEYFERT
CLASS_LINELERG - fraction with radio excess & (BPT SFG or BPT LINER)

Given that the end user has a large set of likelihoods available, you
may wish to define e.g. the subset of star-forming LERGs (i.e. those
which have BPT SF line ratios). You can do this e.g. by:

threshold = 0.99
CLASS_LINELERG > threshold & BPT_SFG > threshold & zscore < 2.5 & CLASS_z_WARNING == 0

However, we note that defining LERGs using these outputs is not
straightforward, since they don’t (in general) have bright lines. It
is possible to define a sample of LERG-like sources e.g. by combining
high chance of radio excess (e.g. 99% of realisations) plus NOT
classified as a HERG - however the choice of threshold is highly
subjective, and such decisions are left to the end user.