Astron. Astrophys. 332, 459-478 (1998)

4. PCA and spectral sequence: test on Kennicutt galaxies

Connolly et al. (1995) have shown using the spectra from Kinney et al. (1996), that the first 2 projections of the PCA define a sequence tightly correlated with the morphological type. Folkes et al. (1996) and Sodre & Cuevas (1997) have demonstrated this property using a larger sample of spectra of local galaxies, namely the sample of Kennicutt (1992a). Here we use again the Kennicutt sample to complement the previous studies and to serve as comparison sample for the ESS sample. We have selected 27 normal Kennicutt galaxies from Hubble types E0 to Im, by discarding peculiar morphological types, and excluding spectra of galaxies with a particular spatial sampling (strong HII regions or high extinction zones). Table 2 lists the ID, the names and morphological types of the selected galaxies. We apply the PCA to these spectra restricted to the spectral range 3700 to 6800 Å, with a pixel size of 5 Å, which is the highest possible resolution for that sample (see Kennicutt 1992a).

Table 2. Kennicutt galaxies selected for PCA.

Left panel of Fig. 3 shows the angles and [see Eqs. (4a) and (4b)] for the 27 chosen Kennicutt spectra (see Table 2), showing the tight sequence strongly correlated with the morphological type, already shown by Sodre & Cuevas (1994), Connolly et al. (1995) and Folkes et al. (1996), using different coordinates.

With only the first 3 PC's, we can reconstruct, on average, 98% of the signal of each Kennicutt spectrum in Table 2. PC's of superior order do not contribute more than 2% to the signal. This was already demonstrated by Connolly et al. (1995), using the observed spectra of Kinney et al. (1996). The physical reason for this striking feature is closely related to the fact that the fundamental spectral features of normal galaxies can be described by a reduced number of stellar spectra, namely spectral types AV and M0III. This was first suggested by Aaronson (1978), using UVK color-color diagrams (see also Bershady 1993, 1995). To probe this effect using the PCA approach, we project stellar spectra (from Sviderkien 1988) of stars with types A0, A2, G0, and K0 of the main sequence, and two spectra corresponding to giants M0 and M1, onto the first 3 PC's from the Kennicutt sample and derive the values of and . Symbols other than points in the left panel of Fig. 3 show that the A stars and the M giants mark the extreme regions (or the extrapolation) of the Hubble sequence, whereas the G and K stars are located inside the sequence. In addition, the right panel of Fig. 3 shows the surprising similarity between the second PC of the Kennicutt sample (with the emission lines eliminated) and the second PC from the stellar spectral sample. This extends and further demonstrates the results of Aaronson (1978) and allows us to conclude that the spectra of nearby galaxies with normal Hubble types, may be described with a reduced number of stellar spectra (2 types), at least in the spectral range which is considered here. Because the position of the observed galaxies along the axis accounts for the relative contributions of the red and blue stellar populations in the observed galaxies, we adopt the parameter to describe the spectral sequence.

Fig. 3a and b. The Kennicutt spectra of normal Hubble types (left figure, dots), onto the classification plane. Red or early type galaxies are to the left with 0 and blue or late types are to the right with 0. The deviation in the parameter is mainly related to the emission lines. The circles and squares indicate the position of the spectra of main sequence stars and giant stars, respectively. The right panel shows a comparison between the second PC from the sample of Kennicutt normal galaxies (thin line) and the PC from the sample of stars appearing in the left panel (thick line).

As a complement, the parameter conveniently characterizes the presence of emission lines. The emission lines play an important role in the spectral classification of galaxies. They serve to characterize the strength of star formation, the nuclear activity and abundances, using for example the ratio between the strength of different emission lines. Francis et al. (1992), apply successfully the PCA technique to understand the systematic properties of QSO's. The role of is demonstrated by truncating the emission lines from all the Kennicutt spectra. This is done by fitting a polynomial of degree one to the adjacent continuum for each line. The resulting values of and are shown in Fig. 4. The ordering of the Hubble sequence along the axis remains the same as for the sample with emission lines. However, all spectra now have smaller values.

Fig. 4. The figure shows the Kennicutt templates with the emission lines removed, projected onto the spherical space (, ), with the same scale as in Fig. 3.

Fig. 4 therefore shows that the emission lines increase the dispersion in the (, ) plane, placing galaxies with strong emission lines far from the equator defined by = 0. The known correlation between star formation and/or activity and the Hubble type for morphologically normal galaxies explains the observed correlation between and in Fig. 3.

Folkes et al. (1996) have studied in detail the reconstruction error as a function of the S/N in the input spectra, using simulated spectra constructed from the Kennicutt sample. They also demonstrated the greatly improved capability of the PCA for filtering the noise over other standard techniques. Here we also find that the noise in the input spectra has no effect onto the classification space (, ) and that the observed sequence remains unchanged when adding arbitrarily high noise onto the Kennicutt spectra of Table 2: decreasing the S/N of the spectra down to 10% of their original value yields a change of on the average ( changes by for galaxies without emission lines, that is types E0 to Sa, with ; for types Sc and Sm/Im, with ).

© European Southern Observatory (ESO) 1998

Online publication: March 23, 1998
helpdesk.link@springer.de