Identifying Radio-active Galactic Nuclei among Radio-emitting Galaxies

Our starting sample of radio-emitting galaxies consists of the 32,616 objects presented in the ROGUE I catalog. This catalog was made by selecting galaxies from the SDSS DR7 (see below) having a FIRST radio counterpart within 3'' from the optical position. It provides information about core and total radio flux density at 1.4 GHz, as well as on the morphology of the radio structure and of the optical host galaxy. There are 1537 galaxies clearly connected to extended radio sources, 940 galaxies connected with a possibly extended or one-sided structure, and 902 sources considered as SF, blended, or not related to a galaxy with an SDSS spectrum. As many as 29,237 objects fall into the unresolved or "elongated" (as opposed to "extended"; see Kozieł-Wierzbowska et al. 2020) radio source categories.

The monochromatic radio luminosity at 1.4 GHz is calculated from the total radio flux density (F_1.4) as

$$\begin{eqnarray}&&{L}_{1.4}=\displaystyle \frac{4\pi {D}_{L}^{2}}{{\left(1+z\right)}^{1-\alpha }}{F}_{1.4},\end{eqnarray} \tag{ 1 }$$

where we take the spectral index to be α = 0.75, which is close to the mean spectral index of sources detected at 1.4 GHz (e.g., Condon 1984; Smolčić et al. 2017), and the luminosity distance D_L is calculated from the spectroscopic redshifts from SDSS. We assume a flat ΛCDM cosmology, with Ω₀ = 0.3 and H₀ = 70 km s⁻¹ Mpc⁻¹.

Objects in the ROGUE I catalog have been selected from the SDSS seventh data release (DR7) database (Abazajian et al. 2009) and belong either to the Main Galaxy Sample (Strauss et al. 2002) or to the Luminous Red Galaxy sample (Eisenstein et al. 2001). Note that, in principle, these samples do not contain any Type I AGNs (although Oh et al. 2015 identified around 4000 galaxies with broad Hα lines in the full DR7 galaxy catalog).

The ROGUE I catalog contains only SDSS DR7 galaxies with redshift z larger than 0.002 (which guarantees that luminosity distances are not dominated by peculiar motions; e.g., Ekholm et al. 2001) and S/N in the continuum at 4020 Å of at least 10, in order to allow a meaningful study of their stellar populations. Repeated spectra of the same galaxy were removed by matching galaxies in R.A. and decl. (Δ < 05). From the ROGUE I catalog, we removed the small fraction of galaxies (0.2%) for which the Petrosian half-light radius is negative or the stellar mass is smaller than 10⁷ M_⊙ in the starlight database, as well as the few objects that do appear in the published ROGUE I catalog but whose radio emission was found to be blended with another source or for which the optical galaxy is not the host of the radio emission (objects tagged B or ND in the catalog). Our final sample contains 32,163 galaxies and is referred to as the ROGUE I sample in the remainder of the paper.

The SDSS spectra allow us to derive certain physical quantities such as the redshifts, the total galaxy stellar masses (M_⋆), the black hole masses (M_BH), and D_n(4000). These last quantities are obtained by means of the inverse spectral synthesis code starlight (Cid Fernandes et al. 2005) to fit the observed continuum emission of the galaxies. The black hole masses are derived from the stellar velocity dispersions (σ_⋆) using the relation by Tremaine et al. (2002), which is applicable for σ_⋆ ≥ 70 km s⁻¹. D_n(4000) is taken from the synthetic spectra as explained in Stasińska et al. (2006). Emission-line measurements were performed after subtraction of the synthesized stellar spectrum from the observed SDSS spectrum. All the data were obtained exactly as explained in Section 2 of Kozieł-Wierzbowska et al. (2017) and can be retrieved from the starlight database ⁹ (Cid Fernandes et al. 2009).

WISE observed the entire sky in four infrared bands with central wavelengths 3.4, 4.6, 12, and 22 μm (dubbed W1, W2, W3, and W4, respectively; Wright et al. 2010). The photometric sensitivity at 5σ is 0.068, 0.098, 0.86, and 5.4 mJy for each one of the four bands. The catalog contains positional data of more than 500 million objects with S/N > 5 for at least one of the bands. The photometric calibration was made in the Vega system, with conversion factors to the AB system of 2.699, 3.339, 5.174, and 6.620 for W1, W2, W3, and W4, respectively (Jarrett et al. 2011).

Emission in the W1 and W2 bands is more easily detected because of the sensitivity of the detector and because the data usually have better S/N. The widest band is W3, with a spectral coverage of 7–17 μm, which was designed to observe the strong polycyclic aromatic hydrocarbon (PAH) emission bands that are present in this range (Jarrett et al. 2011). The PAH emission is often associated with the presence of warm dust within molecular clouds, excited by the strong radiation field from the massive young stars immersed in it (da Cunha et al. 2008) and/or from the dust torus around an AGN (Alonso-Herrero et al. 2014). The main features of these kinds of emission are centered around 11 μm and hence detectable inside the W3 band for the whole redshift range probed by the ROGUE I sources (z ≤ 0.6), while thermal emission is found within the W4 band, the most S/N-poor filter of WISE (Wright et al. 2010).

We matched the WISE and the SDSS DR7 catalogs, identifying the most probable pairs within a 1'' radius. To use the WISE measurements, we required S/N_W3 ≥ 2 to exclude spurious measurements and problems due to sensitivity. This is especially needed for early-type galaxies, which have little or no gas and are generally very faint in this spectral range.

Our final sample of galaxies with usable WISE data consists of 23,294 objects and will be referred to as the ROGUE I–WISE sample.

Throughout the text we also use the flux calculated for the W3 band. We obtain fluxes from Vega magnitudes using

$$\begin{eqnarray}&&{F}_{\nu }\,[\mathrm{Jy}]={F}_{\nu 0}{10}^{-0.4{m}_{\mathrm{Vega}}},\end{eqnarray} \tag{ 2 }$$

where F_ν0 is the zero magnitude flux density and m_Vega is the WISE magnitude of the object. We use the zero-points and the relations from Jarrett et al. (2011) ¹⁰ with a zero magnitude flux density F_ν0 = 31.674 for W3 and assuming that the MIR spectral energy distributions do not significantly deviate from a power law (F_ν ∼ ν⁰).

As mentioned in the introduction, we look for a unique criterion that would be able to separate SF galaxies from galaxies where the radio emission is linked to an AGN. The first problem with using several criteria together, as is the case of, e.g., Best & Heckman (2012), is that relevant data are not always available for all the objects, so that for some objects the classification has to rely on only a fraction of the chosen criteria.

We have matched their catalog of 18,268 objects to the starlight database and found 18,074 objects in common. To analyze their data, we used the radio fluxes from the Best & Heckman (2012) catalog and emission lines, stellar masses, and other spectral measurements from the starlight database. Objects that do not exhibit the four emission lines ([O iii], Hβ, [N ii], and Hα) in the Baldwin et al. (1981, BPT) diagram cannot be classified using this criterion. This concerns 11,939 (66%) of their entire sample of ∼18,000 objects.

Second, for those objects with data allowing one to apply more than one criterion, classifications according to the various criteria do not necessarily agree. Let us consider the other two criteria used by Best & Heckman (2012), which are D_n(4000) versus L_1.4/M_⋆ and the relation between the Hα emission-line luminosity and the radio luminosity (L_Hα vs. L_1.4). Among the 9596 objects from Best & Heckman (2012) having all the required data to use the two methods, there are 1085 objects for which the classifications differ.

In the following subsections we present two methods that can be individually applied to radio catalogs depending on data availability.

It has been known almost since the beginning of extragalactic radio astronomy and infrared astronomy (de Jong et al. 1985; Helou et al. 1985; Condon & Broderick 1988; Condon 1992, and references therein) and confirmed in recent studies (e.g., Wang et al. 2019) that there is a tight correlation between the radio luminosity and the far-infrared luminosity of "normal" SF galaxies, i.e., galaxies that do not contain an AGN. The radio emission in these sources consists of free–free radiation from H ii regions and synchrotron radiation from supernova remnants (mostly Type II and Type Ib) whose progenitors are more massive than 8 M_⊙, so it probes stellar populations younger than 5 × 10⁷–10⁸ yr. The far-infrared radiation measures the bolometric luminosity of stars more massive than about 5 M_⊙ reprocessed by circumstellar dust. Neither the radio emission nor the far-infrared emission is affected by dust extinction; thus, the observed radio flux and far-infrared flux are expected to be tightly linked. In the local universe (up to z = 0.15), the relation is very tight, with a dispersion of about 0.25 dex (Yun et al. 2001). It has been used to identify jetted AGNs as objects deviating from this relation (Donley et al. 2005; Park et al. 2008; Del Moro et al. 2013). However, existing far-infrared data are from rather shallow surveys (IRAS, Neugebauer et al. 1984; or Infrared Space Observatory [ISO]) not deep enough to be useful for the characterization of ROGUE I galaxies, while Spitzer (Werner et al. 2004) data are available only for selected objects.

On the other hand, MIR surveys exist for the whole sky (AKARI, Murakami et al. 2007; WISE, Wright et al. 2010). Rosario et al. (2013) showed the luminosity in the W3 band versus L_1.4 for a sample of emission-line AGNs selected from the SDSS galaxy database. They found that these emission-line AGNs were distributed into two branches. Seyfert galaxies gather on the upper branch, in the same region as a control sample of inactive SF galaxies. Low-ionization nuclear emission-line galaxies (LINERs) are split between the upper branch and the lower branch. These authors argued that, for the Seyfert galaxies, the W3 emission is almost entirely due to dust heated by ongoing star formation and highlighted the MIR–radio plane as a useful tool in studies of star formation and accretion properties of AGNs. The MIR–radio plane was also successfully used to study star formation properties of the radio/X-ray-selected sample by Mingo et al. (2016).

Figure 1 shows the ROGUE I–WISE sample on the $\mathrm{log}{F}_{W3}$ versus $\mathrm{log}{F}_{1.4}$ plane, which we dub MIRAD, as it is based on MId-infrared and RADio fluxes. The full sample is plotted as gray error bars, which have been obtained by error propagation from the observed fluxes. Points in blue highlight the 2538 objects that are classified as pure SF galaxies according to Stasińska et al. (2006) on the [N ii] λ6584/Hα versus [O iii] λ5007/Hβ plane (requiring a minimum S/N of 3 in those emission lines). Several features are noteworthy. First, this plot shows two different families of objects, with SF galaxies falling on the upper sequence. Second, uncertainties are much smaller in F_W3 than in F_1.4 for the SF sequence, and the opposite is true for the lower sequence.

Zoom In Zoom Out Reset image size

This diagram shows a very clear dichotomy among the population of galaxies with radio emission. We looked for objective criteria to draw a separation line. It turns out that an attempt to separate the BPT-pure SF galaxies and the Fanaroff–Riley type I (FR I) and type II (FR II) radio sources (Fanaroff & Riley 1974), in which radio emission is unambiguously connected to an AGN, gives an infinity of solutions that are equally valid. We thus chose the simplest of all, namely,

$$\begin{eqnarray}&&{F}_{W3}={F}_{1.4},\end{eqnarray} \tag{ 3 }$$

where both quantities are in mJy. Surprisingly, as was shown by Rusinek et al. (2020), the same criterion separates radio-quiet and radio-loud AGNs in a sample of hard-X-ray-emitting objects from the Swift/BAT catalog of Ricci et al. (2017).

This line is drawn in Figure 2, where the pure SF galaxies are plotted as blue points and the objects corresponding to the FR I and FR II radio galaxies are plotted as red points. All but 3 objects out of a total of 601 from the FR I and FR II class lie below this line. In the remainder of the paper, we consider that all the galaxies below this line are radio-AGNs, meaning that their radio emission is dominated by phenomena related to an AGN (mainly a jet originating from a supermassive black hole) and that only a small fraction of it, if any, is due to SF regions. The three FR II radio galaxies that are above the dividing line are most probably misclassifications. We have checked them, and while their NVSS structures are considered as lobes in ROGUE I, they may in fact be separate sources. Objects above the line are considered to be galaxies where the radio emission is mainly due to star formation. Note that there are, however, 53 objects out of 2539 (2%) that are classified as SF according to the Stasińska et al. (2006) line in the BPT diagram but lie below the separation line, some of them even quite far from it. They may actually not be pure SF galaxies, since one has to recall that the SF demarcation line is actually an upper envelope for ionization by massive stars. Possibly these objects do contain an active AGN. Detailed studies of them should provide more information about their nature.

Zoom In Zoom Out Reset image size

According to our MIRAD dividing line, 10,995/(10,995 + 12,229), i.e., roughly 47% of the objects from the ROGUE I–WISE sample, are radio-AGNs. Note that the MIRAD discriminator can be applied without any knowledge of the redshift, which can be useful for surveys such as LOFAR that do not involve optical spectra.

Among the diagrams used by previous authors to select radio-AGNs, the DLM one is quite efficient and can be applied to objects with or without emission lines, whether or not they have any WISE detections. Different dividing lines have been proposed in the past (Best et al. 2005a; Kauffmann et al. 2008; Best & Heckman 2012; Sabater et al. 2019). Here, since we emphasized the utility of a unique diagram, and since we have shown that the MIRAD diagram works extremely well when no spectra are available and no redshift can be determined, we defined a line for the DLM diagram that is as compatible as possible with our dividing line in the MIRAD diagram. We proceeded as follows. For a given dividing line in the DLM diagram, we defined the completeness C and the R reliability fractions for the SF and AGN classes (Strateva et al. 2001). C_SF is the fraction of galaxies classified as MIRAD-SF that are correctly classified as SF in the DLM diagram, whereas R_SF is the fraction of galaxies classified as SF in the DLM diagram that are also MIRAD-SF. Defining C_AGN and R_AGN likewise for the radio-AGN class, we calculated the figure of merit P (Baldry et al. 2004), which is the product of those four fractions: P ≡ C_SF R_SF C_AGN R_AGN. We calculated P over a wide range of slopes and intercepts for the DLM dividing line, refining our search in the loci of the greatest values of P.

Two possible solutions were found:

$$\begin{eqnarray}&&{D}_{{\rm{n}}}(4000)=-0.28\,{L}_{1.4}/{M}_{\star }+4.9\end{eqnarray} \tag{ 4 }$$

and

$$\begin{eqnarray}&&{D}_{{\rm{n}}}(4000)=-0.23\ {L}_{1.4}/{M}_{\star }+4.3.\end{eqnarray} \tag{ 5 }$$

We chose Equation (5), which yields a marginally greater value of P, for the dividing line in the DLM diagram.

Figure 3 shows the values of D_n(4000) as a function of L_1.4/M_⋆ for the ROGUE I–WISE sample, together with the dividing lines from Kauffmann et al. (2008), those from Sabater et al. (2019), ¹¹ and our own dividing lines. Figure 4 shows the DLM diagram for the ROGUE I sample, together with the dividing line represented by Equation (5). Here the red points correspond to the FR I and FR II radio sources and the blue points correspond to SF galaxies from the BPT diagram, as in Figure 2. We see that our DLM dividing line puts only 3 extended radio-AGNs (out of a total of 999) in the SF zone, and only 27 BPT SF galaxies out of a total of 2568 (i.e., 1%) in the radio-AGN zone. So the DLM diagram with our dividing line seems to work even slightly better than the MIRAD diagram.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The total number of objects that can be classified with the DLM diagram is 32,163 (as opposed to 23,294 in the MIRAD diagram). However, the ratio of the number of radio-AGN to SF galaxies is 1.61 in the DLM diagram for the ROGUE I sample and 0.89 in the MIRAD diagram for the ROGUE I–WISE sample. In other words, many radio-AGNs are missing in the MIRAD diagram. Those are likely the ones that have too small flux in the W3 band to be detected. This means that the MIRAD diagram provides a biased estimate of the total number of radio-AGNs among radio sources. This needs to be remembered when dealing with demography of radio-AGNs. A way to circumvent this would be to use upper limits for the W3 detection in order to cull a sample of objects with low MIR emission.

Figure 5 shows the distribution in the MIRAD diagram of the radio morphologies as listed in the ROGUE I catalog. As expected, almost all the extended radio morphologies (i.e., FR I; FR II; hybrid; bent sources that are wide-angle tail, narrow-angle tail, and head-tail; and restarted radio sources composed of double–double, X-shaped, and Z-shaped sources) are radio-AGNs. Unresolved and elongated radio sources, on the other hand, are found on both sides of the dividing line, in similar proportions. It is precisely for those cases that our diagrams are useful to pinpoint which objects correspond to radio-AGNs.

Zoom In Zoom Out Reset image size

Figure 6 is the equivalent of Figure 5, but for the DLM diagram applied to the ROGUE 1 sample. Qualitatively, the results are similar to those obtained with the MIRAD diagram.

Zoom In Zoom Out Reset image size

It has already been shown in many studies (e.g., Dunlop et al. 2003; Kozieł-Wierzbowska et al. 2017) that the fraction of radio-loud AGNs increases rapidly in galaxies with black holes more massive than $\mathrm{log}{M}_{\mathrm{BH}}/{M}_{\odot }$ ∼ 8. Figure 7 shows the distribution of ROGUE I–WISE galaxies with black hole masses in different ranges in the MIRAD diagram, and Figure 8 shows the same for ROGUE I in the DLM diagram. A total of 1727 objects in the ROGUE I sample and 1707 objects in the ROGUE I–WISE sample have σ_⋆ < 70 km s⁻¹, so they lack M_BH estimates and are not included in any M_BH bin.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We note that galaxies with low M_BH fall preferentially above the MIRAD line and those with high M_BH fall below it, suggesting that there is an optimal M_BH separation between SF and radio-AGN. To find out which M_BH value best corresponds to our MIRAD separation line, we again use the figure of merit P (see Section 3.3), where the completeness C and the R reliability fractions are defined according to classification as SF or AGN by MIRAD and an M_BH limit. Figure 9 shows P as a function of the M_BH threshold. For the MIRAD diagram and the ROGUE I–WISE sample (panel (a)), P reaches its maximum value at $\mathrm{log}{M}_{\mathrm{BH}}/{M}_{\odot }=7.85$. For the DLM diagram applied to the ROGUE I sample (panel (b)), P reaches its maximum value at $\mathrm{log}{M}_{\mathrm{BH}}/{M}_{\odot }=7.82$.

Zoom In Zoom Out Reset image size

Thus, both the MIRAD and the DLM diagrams indicate that radio-AGNs are found almost exclusively among galaxies with black hole masses above 10^7.8 M_⊙.

Figure 10 shows the distribution in the MIRAD diagram of the optical morphological classifications from the ROGUE I catalog. Spiral (S) and barred spiral (bS) galaxies are shown in blue; lenticular (L) and barred lenticular (bL) galaxies are shown in green; elliptical (E) and interacting elliptical (iE) galaxies are shown in red; galaxies with distorted morphology (D), galaxies with mergers (M), interacting spiral galaxies (iS), and interacting lenticular galaxies (iL) are shown in light navy blue; and ring (R) galaxies are marked in purple.

Zoom In Zoom Out Reset image size

Later-type galaxies (spirals and lenticulars) fall above the MIRAD separation line, whereas earlier-type ones (ellipticals) fall below the line. Galaxies showing signs of interaction also lie above the separation line, which suggests that their radio emission is connected to star formation. About 13% of interacting systems, however, are below the line, which is in agreement with the fact that a significant fraction of radio jets are found in mergers and interacting galaxies (see, e.g., Chiaberge et al. 2015).

Figure 11 is similar to Figure 10, but for the DLM criterion applied to the ROGUE I sample. Similarly to the MIRAD diagram, and even more conspicuously, it shows that the vast majority of radio-AGNs are found among elliptical galaxies.

Zoom In Zoom Out Reset image size

Figure 12 shows the distribution of the five categories of galaxies defined by Cid Fernandes et al. (2011) in the W_Hα versus [N ii]/Hα diagram (the WHAN diagram):

• 1.  
  Pure SF galaxies: $\mathrm{log}[{\rm{N}}\,{\rm\small{II}}]/{\rm{H}}\alpha \lt -0.4$ and W_Hα ≥ 3 Å, shown in blue.
• 2.  
  Strong AGNs (i.e., Seyferts): $\mathrm{log}[{\rm{N}}\,{\rm\small{II}}]/{\rm{H}}\alpha \geqslant -0.4$ and W_Hα ≥ 6 Å, shown in dark green.
• 3.  
  Weak AGNs: $\mathrm{log}[{\rm{N}}\,{\rm\small{II}}]/{\rm{H}}\alpha \geqslant -0.4$ and 3 Å ≤ W_Hα < 6 Å, shown in light green.
• 4.  
  Liny retired galaxies (RGs, i.e., fake AGN): 0.5 Å ≤ W_Hα < 3 Å, shown in orange.
• 5.  
  Lineless RGs: W_Hα < 0.5 Å, shown in red.

Zoom In Zoom Out Reset image size

Note that the value of $\mathrm{log}[{\rm{N}}\,{\rm\small{II}}]/{\rm{H}}\alpha =-0.4$, which separates pure SF galaxies from AGNs in the WHAN classification of Cid Fernandes et al. (2011), is based on the line proposed by Stasińska et al. (2006) to separate pure SF galaxies from AGNs in the BPT diagram. Objects with $\mathrm{log}[{\rm{N}}\,{\rm\small{II}}]/{\rm{H}}\alpha \gt -0.4$ may well experience present-day star formation, but the presence of an AGN can be detected in their optical spectrum (as compared to pure SF galaxies), be it at a level of a few percent. Figure 13 is the same as Figure 12, but for the DLM diagram applied to the ROGUE I sample.

Zoom In Zoom Out Reset image size

A few objects in our samples (54 in the ROGUE I–WISE sample and 114 in the ROGUE I sample) cannot be classified with the criteria above: they have W_Hα ≥ 3 Å (i.e., they are not retired) but lack an [N ii]/Hα measurement, meaning that they cannot be classified as being either an optical AGN host or an SF galaxy.

Figures 12 and 13 reveal that the SF and sAGN classes are distributed very similarly in the MIRAD and DLM diagrams. Only a small fraction of sAGNs are identified as radio-AGNs. The proportion is somewhat larger for wAGNs (which can broadly be identified with "true LINERs," i.e., galaxies with low-level nuclear activity). In both the MIRAD and DLM diagrams, one can see that liny and lineless RGs constitute the vast majority of radio-AGNs: 79% in MIRAD and 87% for the DLM diagram. These percentages must not be taken literally, of course, but they show that any classification based on emission lines alone would miss the huge population of radio-AGNs that are lineless RGs.

AGN galaxies that have a radio counterpart are often divided into high- and low-excitation radio galaxies (HERGs and LERGs, respectively; e.g., Laing et al. 1994; Buttiglione et al. 2010). The "strong AGNs" in our radio-AGN sample can be roughly identified with HERGs, while "the weak AGNs," as well as the liny and lineless RGs, would belong to the LERG class. It is to be noted that for the majority of the LERGs it is thus not possible to measure the bolometric luminosity of the AGN using optical emission lines. This obviously concerns the RGs, but even in weak AGNs the emission lines are strongly contaminated by emission due to ionization by hot low-mass evolved stars (HOLMES) and not by an AGN (see Cid Fernandes et al. 2011). In some radio galaxies shocks from radio jets may also play a role (e.g., Davies et al. 2017).

In addition to DLM, several other diagrams have been used in the past to help identify radio-AGNs in radio surveys. These are the BPT diagram, the L_Hα versus L_1.4 diagram (Best & Heckman 2012; Heckman & Best 2014), and the WISE color–color diagram used by Sabater et al. (2019). In Figure 14 we show these diagrams, color-coding the points according to their classification in the MIRAD diagram for the ROGUE I–WISE sample (panels (a)–(c)) and to the DLM diagram for the ROGUE I sample (panels (d)–(f)). For panels (a) and (d) we required S/N > 2 in all emission lines in the BPT, for panels (b) and (e) we required S/N > 2 for Hα, and for panels (c) and (f) we required S/N ≥ 2 in all the WISE bands.

Zoom In Zoom Out Reset image size

As obvious from the previous sections, classifications using emission lines will miss a large number of radio-AGNs. On the other hand, adding information from emission lines to the one obtained from the MIRAD or DLM diagrams will not improve the classification, since the latter two are clear-cut while both the BPT diagrams and the L_Hα versus L_1.4 diagram show a fuzzy separation between radio-AGNs and other objects, as seen in Figure 14. As has been shown by Stasińska et al. (2008, 2015), the BPT diagram, in addition to requiring good S/N in at least four emission lines, cannot distinguish galaxies with low-excitation active nuclei from RGs, i.e., galaxies that have stopped forming stars and are ionized by the HOLMES from their old stellar populations. The WISE color–color diagram was used by Sabater et al. (2019) to classify the sources according to a simple division at a given value of W2 − W3 first proposed by Herpich et al. (2016). Sabater et al. (2019) considered this diagnostic rather crude and used it mainly when their other diagnostics gave intermediate or contradictory classifications.