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

6. Analysis

6.1. Classifying the galaxies

In this paragraph we perform the spectral classification of the ESS galaxies, using the sequence. Our goal is not an indirect morphological classification using the spectral classification. The objective is to establish a link between the spectral classification and the known Hubble sequence in order (1) to test the reliability of the PCA classification by comparison with the technique, and (2) to compare the ESS classification with that for other redshift surveys. The major assumption is that the spectral trends for the observed galaxies have the same nature as the spectral trends followed by the Kennicutt sequence, in which we know a priori the morphological type. In order to assign discrete types for comparison with the Hubble sequence, we define a type- relationship using the Kennicutt templates.

Fig. 8 shows the galaxies of sample 1 in the (, ) plane (dots), and 6 galaxies of known Hubble type, which are the average of several of the Kennicutt templates from Table 2 with the same type (open circles). The EL type is the average of galaxies # 1, 2, 3 and 4 of Table 2. The other types are S0 (average over objects # 5, 7 and 8), Sa (# 9, 10, and 12), Sb (# 14, 15 and 26), Sc (# 16, 17, 20, 21, 22, and 23) and Sm/Im (# 24 and 25). These average spectra are then projected onto the PC's derived from the ESS sample 1. The different averaged Kennicutt spectra in Fig. 8 are not equally separated in the (, ) plane. It was already known that spiral galaxies have larger differences in their spectra than ellipticals (Morgan & Mayall 1957, Wyse 1993). Fig. 8 provides a quantitative demonstration of this effect: the change by one morphological type for early-type galaxies in the averaged Kennicutt galaxies, corresponds to a small variation in , compared to the late-types. In Fig. 8, the density of objects is significantly higher for low values of than for high values of : gal/deg for and gal/deg for . The large distances between the Sb and Sc and between the between the Sc and the Sm/Im leave space for the intermediate types Sbc, Scd, etc..., and for the Sd type, respectively, for which there are no templates in the Kennicutt's sample.

Fig. 8. Position of galaxies from sample 1 in the (, ) plane (), and of the 6 averaged Kennicutt templates () projected onto the PC's of sample 1. The spectral sequence is binned in two different ways: a variable binning in (marked as I, II, etc...) which follows the position of the Kennicutt templates, and a uniform binning in (marked as I', II', etc...). Vertical lines indicate the boundaries of the classes.

The observed position of the average Kennicutt templates along the axis, provides a natural binning for correlating the spectral sequence and the Hubble type, which we define by regions I, II, III, IV, V and VI, separated by vertical lines in Fig. 8. The boundaries of these regions are the averaged value between 2 adjacent average Kennicutt templates. This variable binning accounts for the varying density which, as mentioned above, is inherent to the frequency of the spectral properties of the ESS galaxies. We also define a uniform binning in denoted I', II', III', IV', V', and VI'. The length for each bin in this case is the total span in divided by the number of morphological/spectral types. Table 4 shows the bin values in for the uniform and non-uniform binning. Also shown in Table 4 are the number of galaxies per bin and the corresponding fractions from the total of 310 galaxies for the combination of samples 1 and 2.

Table 4. PCA classification for samples 1 and 2. N(PCA) indicates the numbers and fractions of galaxies for the defined spectral types, using the position of the spectra along the axis (see Fig. 8). We show the results for uniform and variable bins in . The variable binning is obtained from the projected position of the averaged Kennicutt templates in Fig. 8 (see text).

A bootstrap test shows that the r.m.s. deviation in the number of galaxies within each class in Table 4 is , that is 0.7% in the fractional number per type given in Table 4. The largest source of error comes from the uncertainties in the flux calibrations of the spectra. The 7% external errors in the calibrations curves induce a 5% error in the number of galaxies per spectral class. Using bootstrap experiments we also derive an r.m.s. deviation in of , and an r.m.s. uncertainty in varying from at , to at .

Table 5 shows the results of the (see Sect. 3) applied to the galaxies of samples 1, 2 and 3 (347 in total) with the 6 averaged Kennicutt galaxies as templates. Considering the discrete nature of the method, it is important to establish an indicator of the error which we make in the association of a type. We define it as the fraction of total galaxies for each type for which the second closest template has a value differing by less than 20% from the with the closest template (column 3 of Table 5). We choose 20% as a conservative threshold: given the 7% uncertainty in the flux calibration, we consider that cumulative squared differences less than 20% between a spectrum and 2 templates is not significant.

Table 5. Results of the spectral classification method over samples 1, 2 and 3, using averaged Kennicutt templates described in the text.

The histogram of Fig. 9 shows the corresponding distribution of galaxy types for the sum of samples 1, 2 and 3 using the method (solid lines). The dotted and dashed lines represent the distributions of types derived from the PCA using the uniform and non-uniform binnings in , respectively, as defined in Table 4. The reader should recall that the test is performed over the whole spectral range covered by each ESS spectrum, and therefore the comparison between the PCA and method also provides a test of the influence of the spectral range on the spectral classification. For a uniform spanning of types in (dotted line) there are large differences with respect to the method (solid line), for almost all types. On the other hand, the non-uniform binning based on the averaged Kennicutt templates (dashed line) gives a good agreement between the PCA and the results, especially for late types, thus further demonstrating the reliability of the PCA technique in classifying galaxy spectral types. Note that the normalization changes the type fractions given in Tables 4 and 5 by less than 0.1%. In Sect. 7.2 we compare the type fractions of Tables 4 and 5 with the results of other major surveys.

Fig. 9. Histogram showing the distribution of spectral types derived by the method (solid line, over samples 1, 2 and 3) and the PCA (samples 1 and 2 included). Dashed and dotted lines indicate the results using a non-uniform and uniform binning in , respectively (see Tables 4 and 5).

The fact that the test over the whole sample of 347 galaxies, using the largest possible spectral range for each spectrum (see Table 2) produces nearly the same ¹ fractions of spectral types (see Fig. 18) than the PCA over sample 1 restricted to the spectral range (3700-5250 Å), confirms that this spectral interval is wide enough for application of both techniques.

6.2. Filtering effect of the reconstruction and type dependence

We now illustrate the filtering capability of the PCA using the ESS sample. Fig. 10 shows the S/N of the reconstructed spectra of sample 1 using 3 PC's, as a function of the original S/N, for the different spectral types. The filtering effect of reconstructing the spectra with 3 PC's is striking. Whereas the S/N of the input spectra range from 6 to 40, the reconstructed spectra have S/N between 35 and 80. The gain in S/N is strongly dependent on the spectral type. For late types, the increase in S/N has the largest values, which reaches more than a factor of 4 for 70% of the types V and VI. This is due to the low S/N ratio in the weak continuum of the original spectra, which allows a relatively large improvement in S/N. For most of the early types (I to II), the gain in S/N is larger than 1.5 times, and can be as high as a factor 5. The range of variation for the S/N of the reconstructed spectra is related to the S/N of the PC's. The PC's have an intrinsic level of noise, and there is a minimum and maximum S/N achieved with the permitted linear combinations of the first 3 PC's for the defined spectral sequence. In Fig. 10, the dashed line indicates the S/N of the first PC (55).

Fig. 10. S/N of the input spectra and their reconstructions using the first 3 PC's. The different symbols indicate spectral type: , II(), III(), IV(+), V(), VI(). Lines indicate the boundary of a gain in S/N equal to factors of 4, 2 and no gain. The dashed line indicate the S/N of the first PC.

We conclude that the noise carried by the original spectra can be reduced to an interval of well known S/N values, if one uses the reconstructed spectra. The S/N in the original spectra is a function of the apparent magnitude of the objects and the observing conditions, whereas the S/N in the reconstructed spectra depends only on the systematic and statistically significant spectral characteristics of the objects. Of course, in the limiting case when the noise is so high as to hide all the spectral features, the PCA error [Eq. (7)] is large. With data of sufficient S/N ratio, the possibility of reconstruction allows to transform the original sample of spectra into a sample with a reduced noise level. The filtered spectra can then be used for follow-up study of the various galaxy populations, and for comparison of the spectral features with models of spectro-photometric evolution. Note however that the details of the line properties, like equivalent width or line strength, are not linear and cannot be described in detail by the linear PCA approach.

6.3. The emission line galaxies

We now examine the properties of the [OII] emission line, present in of the ESS spectra, in relation to the PCA classification. Fig. 11 shows the (, ) values for the galaxies of sample 1, and indicates the points with measured equivalent widths (W, hereafter) of [OII] (3727 Å) satisfying W([OII]) 30 Å (asterisks), and 15 Å W([OII]) 30 Å (filled dots).

Fig. 11. Galaxies of sample 1 (277 galaxies) with W([OII]) 15 Å (dots), with 15 W([OII]) 30 Å (filled circles), and with W([OII]) 30 Å (stars). Open circles indicate the peculiar galaxies discussed in Sect. 6.6, ordered following Table 8. Triangles denote emission line galaxies discussed in Sect. 6.3.

In Fig. 11, most of the galaxies with (types V/Sc and later), have W([OII]) 15 Å. There are only 4 galaxies with W([OII]) 30 Å and types I/E or II/S0. Table 6 shows the number and fraction of galaxies with 15 W([OII]) 30 Å and W([OII]) Å for different spectral types. We give the mean redshift and the mean value of for each sub-sample and the corresponding standard deviations. Table 6 first displays the well known trend between the spectral type and the frequency of strong emission lines: the later the type, the larger the fraction with strong emission lines, and the stronger the emission lines. For a given spectral class (IV, V, VI), is systematically larger for galaxies with larger W([OII]) (see Table 6). This confirms the relationship between (i.e., the power of the third eigenvector for each spectrum), and the strength of emission lines. Fig. 12 shows that within each subsample of Table 6, the objects span most of the redshift range for the ESS ().

Table 6. Information on the emission-line galaxies .

Fig. 12. Equivalent widths of [OII] for galaxies with W([OII]) 10 Å, as a function of redshift and spectral type.

The equivalent width of [OII] allows us to examine the possibility of galaxy evolution using the average value of W([OII]), as a function of redshift: W([OII]) is a direct measure of the degree of star formation in a galaxy (see for example Osterbrock (1989) and references therein), because this radiation is produced by the interstellar medium (ISM) which is excited by the ultraviolet (UV) radiation from hot stars. Several scenarios of galaxy evolution predict an increase in the star formation rate with increasing look back times. Observational evidence is provided by the excess of blue galaxies in deep number counts (Couch & Sharples 1987, Colless et al. 1993, Metcalfe et al. 1995 and references therein) as well as the increasing density of emission line galaxies in deep redshift surveys (Broadhurst et al. 1988). The Butcher-Oemler effect shows signs of recent evolution in clusters of galaxies (at ) which can be partly interpreted in terms of increased star formation (Butcher & Oemler 1978, Dressler & Gunn 1983, Lavery & Henry 1988). One of the current issues is whether analogous effects occur in the field and at which redshift. Hammer et al. (1997) show that the fraction of bright emission-line galaxies in the field gradually increases with redshift, from 34% to 75% in the redshift range . Their [OII] luminosity density of field galaxies increases only weakly from to (by a factor 1.6), and by a large factor (8.4) between and . Similar results are found from other data samples, using different selection criteria and/or different spectroscopic techniques (Heyl et al. 1997). In the ESS sample analyzed here, we do not detect any significant evidence for an increase of W([OII]) with redshift up to (see Fig. 15). This result agrees with those given by Hammer et al. (1997) and by Heyl et al. (1997).

We also note the presence of [OII] in several early type galaxies of the ESS sample. In particular, there are 4 galaxies with types I-II/E-S0 which have W([OII]) Å (with and marked with an asterisk in Fig. 11). The nature of this emission is not fully clear. However, some agreement exists (see for example Dorman 1997) to point out that very hot post-AGB stars present in such galaxies could be the source responsible for the [OII] emission, probably also related to the so-called Ultraviolet Upturn Phenomenon ("UVX"); NGC 1399 a well-known example of this effect (Dorman 1997). It also was suggested that the environment could play an important role in the "UVX" phenomenon (Ellis 1993). Note that 2 of 4 early-types galaxies with emission lines in the ESS sample belong to the same group of galaxies (at ) and the other 2 have and 0.31.

Fig. 13 shows the spectra of the six galaxies with and 5.0 (open triangles in Fig. 11). They all have W([OII]) Å, and, except one, have spectral types later than IV/Sb. The spectra show clear signatures of strong star formation or activity. If we place these 6 emission-line galaxies onto diagnostic diagrams of log([OIII] 5007/H ) versus log([NII] 6584/H ) or log([OIII] 5007/H ) versus log([SII] 6716+ 6731/H ) (Villeux & Osterbrock 1987), we find that galaxies # 1, 2, 5 and 6 are most likely HII galaxies (i.e. have a high current stellar formation rate). Only galaxy # 3 lies clearly inside the Seyfert 2 region. Galaxy # 4 was impossible to classify due the absence of H from the spectrum. Note that the ESS fraction of AGN () is in marked disagreement with the large fraction () found by Tresse et al. (1996) at in the Canada France Redshift Survey (CFRS).

Fig. 13a-f. Spectra of the galaxies in the ESS sample 1 (at rest-wavelength) with and . The galaxies are marked as open triangles in Fig. 11 and have W([OII]) Å. For each object, the spectral type and W([OII]) in Å are: #1, IV/Sb, 52; #2, V/Sc, 44; #3, V/Sc, 38; #4, V/Sc, 50; #5, VI/Sm-Im, 46; #6, VI/Sm-Im, 51.

6.4. Redshift distribution and completeness

To the limiting magnitude = 20.5, the spectroscopic sample used for the spectral classification in this paper (sample 1) represents 41% of the complete magnitude-limited sample. For a given magnitude bin, the inverse fraction of galaxies having a measured redshift gives the completeness correction to apply for that bin. Bottom panel of Fig. 14 shows the histogram of galaxies per 0.5 magnitude bin. The solid line represents the total number of galaxies in the ESS spectroscopic sample with (669 galaxies). The dashed line represents the histogram of the 277 galaxies of sample 1 used in most of the analysis. Upper panel of Fig. 14 shows the completeness as a function of magnitude for sample 1 (in 0.5 magnitude bins). We then correct the number of galaxies per type which are obtained in Sect. 6.1 and Table 4 by using the inverse of the completeness curve. The resulting type fractions are nearly identical for all spectral types, with absolute changes 1% in the type fractions. The small variations result mainly from the homogeneous spread of different types as function of apparent magnitude.

Fig. 14a and b. Bottom panel: histograms showing the total number of galaxies per bin of 0.5 magnitudes and with (total 669, solid), and the number of galaxies used for the present analysis (the 277 galaxies of sample 1, dashed). Upper panel: the fraction of the total number of galaxies per 0.5 magnitudes in in sample 1.

Fig. 15 shows the distribution of types in redshift space (for 0.1 ), using the PCA spectral classification. The type population is stable as a function of redshift for , with spectra of type IV(Sb) as the dominant type, followed by type V(Sc) with no clear indication of evolution in the type populations with z. Recall that the absolute errors in the population fractions are (note that the last bin has a small number of objects, so the errors in the population fractions are larger). Fig. 15 indicates however a significant excess in the fraction of early types at : the local density of galaxies with types I-II/E-S0 is 3.1 above the average value for (using ). This effect could be caused by the presence of an elliptical-rich group of galaxies. The complete redshift sample is necessary for further investigation of this feature.

Fig. 15. Fraction of galaxies of each spectral type, per redshift interval. The spectral type is provided by the PCA over sample 1 (277 galaxies). The bin size is 0.1. The absolute 1 errors in the type fractions are in and for , this last value due to the reduced number of galaxies.

6.5. Morphology-spectral relationship for the ESS sample

We now examine the morphology-spectral relationship for the ESS sample by testing whether our spectral classification procedure is consistent with the morphology of some of the objects. We have performed a visual morphological classification of the 35 brightest galaxies in samples 1 and 2. CCD images of these objects in the R filter (Arnouts et al. 1997) are given in Fig. 16, in decreasing order of brightness along with the orientation of the slit used to obtain the spectra. The redshift, magnitude, morphological and PCA spectral types are listed in Table 7. These galaxies span the magnitude range = 15.82-18.58 and the redshift range 0.10-0.25, with one galaxy having 0.42 (# 28). The morphological classification is inspired from that for the Revised Shapley Ames Catalog (Sandage & Tammann 1981). It was performed by GG in two steps. The first step was to make a rough classification based on the search for three features in each galaxy: disc and/or bulge and/or spiral arms. If a galaxy could not be included in any of these 3 categories, it was assigned a "peculiar" morphology. Note was taken of signs of merging or interaction when present. The second step was to define sub-classes within ellipticals (bulges with/without discs) and spirals (disc and/or spiral arms). This task is difficult (but not impossible!) because the objects are small (). The discs are therefore poorly visible and the contrast of the spiral arms is weak. In Fig. 16, the spiral arms are clearly visible in object # 1, and discs are visible in # 17, 30, 31, etc... Careful visual inspection of the images of the galaxies using variable contrast allows to discover the presence of spiral arms in many cases. This can be done for 18.0 (objects # 1 to 17), for which the typical apparent diameter of the galaxies is . For fainter magnitudes, visual detection of the spiral arms is very difficult. We note that the spectral type of each galaxy was kept unknown prior to the morphological classification, in order to avoid a psychological bias. The morphological classification was repeated one month later after sorting at random the galaxies to be classified. In general, the second morphological type does not differ from the first assigned type by more than one morphological type (see Fig. 17).

Fig. 16. The 35 brightest galaxies (filter ) is the ESS sample listed in Table 7. The spatial extension of galaxy # 1 is and it is for galaxies labeled # 2 to 38. ID number correspond to those in Table 7, where some galaxies have more than one spectroscopic measure (see also Fig. 17).

Table 7. The redshift, R magnitude, visual morphological type and PCA spectral type for the 35 brightest galaxies (and their 38 spectra).

Fig. 17. Comparison between the morphological and spectral classification for the 35 galaxies listed in Table 7 and displayed in Fig. 16. There are 35 galaxies and 38 spectra, some galaxies having more than one spectroscopic measure: dashed lines for the morphological classification indicate that the galaxy is the same as the preceding one, and the 2 (or 3) spectra provide 2 measures of the spectral type.

Fig. 17 shows that there is good agreement between the visual morphological classification and the PCA spectral classification, for most of the selected galaxies. The mean difference between spectral and morphological type is 0.8 type, with an r.m.s. type dispersion of 0.5. We stress that the good agreement between spectral type and morphological type for the objects in Table 7 suggests that we have been successful in our visual classification in reproducing the typical morphological criteria used in the Kennicutt sample. This also suggests that the low redshift members of the ESS survey () have a similar morphological-to-spectral relationship than the nearby galaxies () in the Kennicutt sample. Note that the morphological classification of the Kennicutt local sample is done in the B band, and this is consistent with our classification being performed on the R images: the B filter "redshifts" to the R filter at (the average redshift of the objects in Table 7).

6.6. Peculiar objects revealed by the PCA

One of the useful features of the PCA is its capability to detect objectively objects which deviate from the general trend. In Fig. 11, we mark with open circles galaxies which lie outside of the main (, ) sequence and outside the sequence of emission-line galaxies (see Sect. 6.3) in order to show that the degree of peculiarity is related to the departure from the main sequence. Although galaxy #5 has a close neighbor in the (, ) plane, we only show object # 5 because two spectra with values differing by less than 5% are indistinguishable. The spectra of the selected objects are shown in Fig. 18. Redshift, magnitudes, colors, spectral and morphological types of these objects are listed in Table 8. Visual examination of the R CCD images of these objects has been performed. In general, the spectral and morphological type agree within the uncertainties in determining the morphology (see Sect. 6.6). It is interesting to note that the images of the six of nine galaxies show clear signs of peculiarities and/or merging. One galaxy even shows a jet-like feature (# 7). Galaxy # 1 shows a very red continuum, steeper than in any Kennicutt galaxy, a strong break, as well as strong Na 5892 and H emission lines. Note that the image shows a regular morphology, and therefore the spectral classification provides additional physical information. Galaxy # 2 presents a very low S/N ratio which is just at the chosen limit for sample 1 (S/N 5.0). The unusual continuum shape of this spectrum is probably responsible for the deviation from the sequence but the low S/N ration does not allow any further study. Galaxies # 3, 4, 6 and 7 exhibit Markarian signatures (galaxies with bursts of star formation). The continuum shape and absorption bands, typical of these galaxies, are clearly seen in Fig. 18 and are comparable to other known spectra (see for example Contini 1996). Such objects have in general a spiral morphology (Contini 1996), which is in marked contrast with the I(E) spectral type of objects # 3 and 4. This shows that, for galaxies which significantly deviate from the PCA spectral sequence, the spectral/morphological relationship is no longer valid. Object # 5 is a star which was willingly introduced in the sample as a test, and as we can see, it is far from the main sequence. It provides an additional test of the capability of the PCA to distinguish abnormal spectra. Galaxy # 8 is a typical galaxy with strong star formation. Note that H was blanketed by a sky line. Galaxy # 9 resembles a QSO at , if one identifies the broad emission line with MgII.

Table 8. Peculiar objects revealed by the PCA.

Fig. 18a-i. Galaxies deviating from the general trends of the ESS spectral characteristics, in rest-wavelength.

Other objects generally labeled as peculiar are the AGN (Seyfert, LINER's, N galaxies, etc...) and the E+A galaxies (see Zabludoff et al. 1996). As discussed in Sect. 6.3, the AGN galaxies are included in the "emission-line sample". The PCA is an excellent tool to detect these objects and quantify their spectral features via the PC. The E+A galaxies are however difficult to identify from the PCA classification. Because classical E+A galaxies have strong Balmer absorption lines, but no emission lines in the region 3500 to 7000 Å, these galaxies lie inside the normal sequence of early to intermediate-type galaxies (see Fig. 3). In order to identify the E+A galaxies in the ESS sample, we have measured the equivalent width of H , H and H for the galaxies of sample 1, using a similar criterium to that used by Zabludoff et al. (1996): W Å and no sign of [OII] emission. We found 9 galaxies which satisfy the selection criterium. The PCA spectral type of these galaxies is, except in one case, V/Sb. These galaxies imply a 3% fraction of E+A galaxies in the ESS in the redshift range . This is in marked contrast with the 21 E+A out of 11113 field galaxies found by Zabludoff et al. in the Las Campanas Redshift Survey to , which gives of E+A galaxies. Two E+A candidates of sample 1 have , and the fraction does not change if we consider the 66 galaxies of sample 1 with . Because of the large redshift of these objects (), the CCD images of these galaxies subtend too small solid angle for a detailed morphological study. Although we do not see evidence of merging, four galaxies form two different pairs, and so we cannot discard the possibility of interaction between some of the E+A galaxies and their neighbors. The other E+A do not show evidence of interaction.

We have verified that all of the objects which strongly deviate from the sequence in which lie the normal galaxies are indeed peculiar objects. On the other hand, the spectra of the objects which lie inside the spectral sequence were visually inspected and do not show signs of any peculiarity, except the possibility that they could have strong absorption lines (for example E+A galaxies). The PCA is therefore efficient at detecting in a quantitative manner both the normal and most frequent objects, and the rare and peculiar objects.

© European Southern Observatory (ESO) 1998

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