3 A general view of the results

The optical and near-infrared (NIR) spectrum of $\eta$ Car is very rich in emission lines, with ionization potentials of atomic species ranging from $\sim$6 eV to more than 60 eV. It is a mixture of permitted and forbidden lines with narrow components (fwhm$\sim$20-80km s^-1) seated on $\sim$8 times broader components. H$\alpha$ is the strongest line, peaking at $I_{\rm peak}/I_{\rm c}$$\sim$200 in high state, followed closely by He I$\lambda$10830. Single ionized iron shows the most numerous atomic transitions, with 20% more Fe II than [Fe II] lines, ranging from intermediate to low intensities. A group of highly excited lines ([Ne III], [Fe III], [Ar III], and [S III]) and some nitrogen lines also have intermediate strength. Numerous faint lines, mainly from Fe II, Ti II, Cr II and N I, make the spectrum of $\eta$ Car very crowded, with plenty of blends and only a few regions free of lines to sample the real stellar continuum.

In Fig. 1, we present the spectrum from 4040 to 10970Å, excluding the wavelength regions heavily affected by telluric absorption (7200 - 8400Å  and 8900 - 9900Å). In the upper part of figure, we display the high state spectrum (solid line) superimposed on the low state (dotted line). The intensity relative to the stellar continuum was transformed to a logarithmic scale, in order to enhance the visibility of the broad line components and their variations. The S/N in the stellar continuum is displayed for some wavelengths. Numbers above the spectrum plot are for 1995 and below for June 1992. We show S/N $\ge$ 100 in the red region (6000 - 7000Å) decreasing to S/N$\sim$ 20-40 at the blue and near-infrared extremities. The S/N is higher in the emission lines than in the continuum, obviously. It is important to emphasize that photon noise is not the main source of uncertainty in the line measurements. Errors are dominated by the placement of the continuum level and by uncertainties in deblending the components.

The ratio between the 1995 and 1992 spectra is labeled high/low in Fig. 1 and is displayed in linear scale at the bottom of each plot window. It is useful for a quick look at line variability, line profiles and for guessing what component belongs to which transition. For example, in Fig. 1c, the [Fe III]$\lambda$4769 line is faint and blended, but the high/low plot shows a line profile very similar to the neighboring isolated [Fe III]$\lambda$4750, $\lambda$4701, and $\lambda$4654 lines. Similarity between lines of Balmer series is very clear in the high/low ratio plot, although difficult to see in the original spectrum because of blendings. Before dividing the 1995 and 1992 spectra, we performed a 5-point triangular smoothing to avoid spikes due to small wavelength mismatches. In this way, the high/low ratio plot tends to underestimate the real amplitude of variations from high to low states. The horizontal dashed-line in Fig. 1 (ordinate equal 1) indicates the level of unchanged features; points above this line indicate that the feature had higher flux (relative to the continuum) in high state than on low, and vice-versa. For example, a line that was strong in 1995 but faint in June 1992 generates a high peak in the high/low ratio. This is the case of P Cygni components that are deeper in low than in high state. A complete description of the features displayed in Fig. 1 will be presented in the next sections.

In Table 1, we present the measured line parameters. A short dash indicates that the corresponding measurement was not performed because it was meaningless or technically unfeasible. In Column 1 , a number designates the sequential order in wavelength of a spectral line or blend of lines. Letters label individual features inside a blend. Column 2 displays the identification of atomic transitions: notidf stands for not identified feature, an interrogation mark (?) for a doubtful identification, and IS indicates an interstellar absorption. Column 3 comments on the appearance of the feature: sgl stands for a single isolated line, comp for an isolated transition composed with more than one component, nar for a narrow line component, bro a broad line component, bld for a blending of different atomic transitions, P Cygni for a blue shifted absorption, pV and pR for the violet and red components of a double-peaked line, abs for an absorption component superimposed on the emission profile, bshd for an enhancement of the blue extremity of the broad line component, bem? for a feature that looks like a blueshifted emission present in some lines, an exclamation mark (!) for a remarkable feature, and an interrogation mark (?) for doubt about the reality of the feature. As it is difficult to know a priori if a double-peaked line is produced by an absorption feature or by two emission components, we measured both the peaks and the valley. Column 4 displays the laboratory wavelength of identified transitions (inÅ); numbers inside brackets are the suggested laboratory wavelength for transitions where we didn't find a reliable identification. The suggested laboratory wavelength was obtained by shifting the measured line center by +40km s^-1. Columns 5 through 9 refer to the 1995 (high state) spectrum and 10 though 14 to 1992 (low state) spectrum. In Columns 5 and 10 we present the observed wavelength barycenter of the line components (in Å), measured through Guassian fit; Columns 6 and 11 the heliocentric radial velocity (km s^-1); Columns 7 and 12 the line peak intensity relative to the local stellar continuum; Columns 8 and 13 the line flux normalized to the continuum (in units of Å) - this coincides with the classical definition of equivalent width when the base of the line merges into the local continuum. A minus sign in Columns 7, 8, 12 or 13 indicates that the feature in absorption. In Columns 9 and 14 we present the fwhm of the feature (in km s^-1), corrected for the instrumental profile. Column 15 displays the percent variation of the line flux, relative to that in 1995 (high state), derived from the expression:

$$var\,\,=\,100\,\frac{\mathrm{flux}_{1995} - \mathrm{flux}_{1992}}{\mathrm{flux}_{1995}}.$$ (1)

A line that fades completely during the low state results in a number 100 in Column 15 , while negative numbers are for lines that are stronger on that phase than in low state. P Cygni components result in largest negative numbers in Column 15 , as most of them were faint or absent in 1995. In order to not have an undetermined value in Column 15 , we adopted a flux of -0.1 in Column 8 when the feature could not be measured reliably in the 1995 spectrum.

The way we defined the index of variability (var ) in Column 15 of Table 1 is analogous to the high/low ratio plotted in Fig. 1. The main differences between the two ways of defining variability of the spectral features are: a) a nonvariable feature results in var =0 while high/low =1; b) variations in the P Cygni profiles result in var $\le$0 while high/low $\ge$0. The advantage of the var parameter over the high/low ratio is that it doesn't depend on the line intensity, allowing a direct comparison of the degree of variability between strong and faint lines. Unfortunately, the high/low ratio couldn't be defined in the same way, as it would result in a very noisy plot, disabling the evaluation of variability by a quick look.

 

Figure 1: a-c)

 

Figure 1: d-f)

 

Figure 1: g-i)

 

Figure 1: j-l)

  

Figure 1: a-n) The solid line at the top of each panel represents the high state spectrum and the superimposed dashed line the low state spectrum. The scale is logarithmic to enhance the broad line features. The ratio between the two spectra high/low, is plotted at the bottom, in linear scale Figure 1: m-n )