4 Line identification

The first step in the process of line identification is to match observed with laboratory wavelengths. This can be done once the radial velocity of each feature is known. By using 200 lines well-identified in previous works, we derived a radial velocity of $rv\,=\,-40\pm 5$km s^-1 for the centroid of the narrow line components (radial velocities are in heliocentric system through this paper, as indicated in Sect. 3). This is in excellent agreement with results of Zanella et al. (1984) who obtained $rv\,=\,-40\pm 2$km s^-1. High spatial resolution observations with the HST show that this is the velocity of the discrete "Weigelt blobs" near the star (Davidson et al. 1995, 1997). The uncertainty of present data, $\sigma$=5km s^-1, is larger than derived by Zanella et al. in spite of our larger spectral dispersion, more numerous lines, and broader spectral coverage. This indicates that there is a real scatter in radial velocity between the emitting clouds.

We shifted the observed line wavelengths by +40km s^-1 to derive the expected laboratory wavelengths displayed in Column 4 of Table 1. We searched for candidates of atomic transitions in the works of Hamann et al. (Aller & Dunham 1966; Thackeray 1967, 1969; Viotti 1968; Hamann et al. 1994; McKenna et al. 1997; and in the unpublished Fe II oscillator strengths of Kurucz. Candidate transitions were rejected if departing more than 3$\sigma$ (=15km s^-1) from the average radial velocity ($rv\,=\,-40$ km s^-1). Several suggested identifications in previous papers were ruled out. We searched carefully for [O II], [O III], and He II and found no traces, definitively rouling out the presence of these lines in the central knot of $\eta$ Carinae. A complete identification of the measured features is out of the scope of the present work. However, as too many lines remained unidentified in our list, we felt compelled to provide as much new information as possible.

The second step in the process of identification was to compare the strength of the candidate transition to what should be expected from other well known lines. This is very important in the case of ions rich in transitions, such as Fe⁺. We identified new lines of [Fe III], [Fe II], Fe II, N I, [N II] Si II, and Ca II. The line [S II]$\lambda$4068 is much stronger than expected from the intensity of [S II]$\lambda$4076 (Fig. 1a). In fact, a Fe II transition in the Kurucz semiempirical list was found to be blended with [S II]$\lambda$4068, what excludes this line for reddening measurements in the method proposed by Rodgers & Searle (1967). The [S II]$\lambda$ 6729 line was too strong, compared to [S II]$\lambda$ 6716, and we found it to be blended with a Fe II line. After deblending and measuring the ratio between the two [S II] lines, we derived a density of $1.2\,\pm 0.4 10^{4}$electronscm^-3 for the narrow line region, which is substantially lower than those derived by other authors (Hamann et al. 1994 and references therein). The infrared Ca II triplet was expected to be detected, due to the presence of Mg II, Na I, N I, Fe II, and other Ca II and [Ca II] lines in the spectrum of $\eta$ Car and similar objects. The identification, however, was tricky, because an unfortunate sequence of blends in the Paschen lines produces an impression of double-peaked profiles in the higher members of the series (Fig. 1k). Actually, the false double peaks disappeared abruptly in Paschen lines at wavelengths longer than H 3-12 and are not shown in H 3-18 and H 3-17.

 

Figure 2: The Ca II $\lambda$8542 Å  line is represented by a dotted line, blended with H 3-15. The Paschen-15 line after deblending (dashed line) shows the same line profile as Paschen-12 (continuous line)

A third step in identification was to look for variability in the line between the high and low states and to compare it with other lines of the same ion. For example, in the case of the infrared Ca II triplet, the false violet peaks in H 3-15 and H 3-13 didn't fade in low state (Fig. 2). As can be seen in Fig. 1k, the red line peak disappears in low state in H3-18, H3-12, and H3-11, and in the higher members of Balmer series. After deblending from the proposed Ca II line, H 3-15 shows a narrow component at -40km s^-1 seated on a broad component, a line profile very similar to H 3-12, as seen in Fig. 2. Another example is that of Ti II lines. Almost all lines of this ion are stronger on low than in high state, so only candidates behaving in the same way were accepted as reliable Ti II. In the same scheme, highly variable Ni II candidates were ruled out. Variability in the broad line component is also a powerful tool for confirming the identification of some features. For example, the enhancement of the broad component in the line at 4068Å  in low state confirms that the the line in blending is very probably from Fe II. Another important line for diagnostic, [N II] $\lambda$5755 (Fig. 1f), also clearly shows the presence of [Fe II]$\lambda$5747 in blending, when looking at variability from high to low state. Several lines in Table 1 have an interrogation mark in Column 2 , because showing variability discrepant from the majority of other lines of the same ion, as is the case of: Ti II$\lambda$4307.90; [Fe II]$\lambda$4533.00, $\lambda$4664.45, $\lambda$4792.28, $\lambda$5900.51, $\lambda$6043.16, $\lambda$6044.16, $\lambda$6944.91, and $\lambda$6966.32; Fe II$\lambda$4666.75, $\lambda$4670.17, $\lambda$8595.19, and $\lambda$10245.58; Cr II$\lambda$4558.66 and $\lambda$4571.24; [Ni II] $\lambda$6007.30; and N I$\lambda$6008.48. However, because these lines are faint and may be affected by measurement errors or unknown blends, the correct identification is still pending. We suggested N II as possible identification for the lines at 5676.02, 5679.56, and 5686.21Å  based on wavelength coincidence, the presence of these lines in the spectrum of the star P Cygni, the presence of other N⁺ transitions in $\eta$ Car, and the disappearance of the narrow component in low state which is typical of high excitation lines.

The previous identifications of Sc II$\lambda$5667.16, [V II] $\lambda$6040.31, and [O I]$\lambda$6300.23 are doubtful, because they were based only on wavelength coincidence. Regarding the former two lines, it is unlikely that these ions show only one transition each across the entire optical/NIR spectrum. For [O I]$\lambda$6300, the identification seems suspicious because the expected stronger [O II] and [O III] are completely absent from $\eta$ Car's spectrum.

We did not revise the identifications of Cr II, [Cr II], and [Co II] lines. Only a few faint [Cr II] and [Co II] lines are present in $\eta$ Car's spectrum. In addition to matching the observed wavelengths, the Cr II lines proposed in earlier works show a coherent variability from high to low state, indicating that the identifications in previous works should be correct.

The identification of some features, such as P Cygni profiles, shoulders, absorption components, and double peaks, is a complex procedure and involves the comparison of several lines of related transitions of the same ion. The basic criteria are: the consistency in radial velocities, variability pattern from the high to low state, and broadness of the feature. P Cygni profiles generally drop below the level of the stellar continuum, have large radial velocities, and are much stronger in low then in high state spectrum, except for Ca II and Na ID lines. The high/low ratio plot is a powerful tool when used to identify P Cygni components. As can be seen in Fig. 1, H and Fe II lines (multiplets 27, 28, 42, 49, 74) present prominent features in the high/low plot at positions corresponding to P Cygni components. Relatively narrow lines appear on the blue edge of P Cygni components in some H lines. No identification was found for them and we labeled these features bemis? in Table 1, as they seem to be emission components blueshifted by -600 to -760km s^-1. In some lines the blue-displaced absorption components is superimposed on the emission profile. The largest number of these features is shown by Na I. Many of them are likely to be multiple P Cygni components, but we cannot exclude that some may be valleys in between multiple emission components. In the case of He I$\lambda$10830 and some Balmer lines, we also measured the individual peaks, labeled by pV and pR in Table 1, in spite of the fact that we do not favor the interpretation of these features as real emission. High excitation lines show a shoulder in the blue edge of the broad component, which has been attributed to independent transitions in some previous works. For example, the feature at 6305Å (Fig. 1i), previously identified as Fe II 200, faded together with the narrow component of [S III]$\lambda$6312, indicating that it is a part of the [S III], rather than an independent line. This is confirmed by the line profile of other high excitation lines.