4 Data reduction

The spectroscopic reductions were made using the Image Reduction Analysis Facility (IRAF) packages CCDRED, TWODSPEC and ONEDSPEC. For each night of data, the following steps were performed: (a) interpolation over bad columns, dead and hot pixels; (b) bias subtraction; (c) division of each frame by a flat-field exposure to remove multiplicative gain and illumination variations across the chip; (d) extraction of one dimensional spectra for each observation from the two-dimensional image by summing the pixels within the aperture at each point along the dispersion axis and subtracting out the sky background; (e) wavelength calibration and subsequent resampling of the data on a linear wavelength grid; (f) flux calibration of the extracted spectra, using flux standard stars from the KPNO standards sample; (g) correction for extinction using a standard atmospheric extinction law.

4.1 Basic reductions

Pixel-to-pixel variations in the response were removed through division by appropriately normalized exposures of the dome illuminated by a hot, spectroscopically featureless lamp. The dark counts were so low that their subtraction was not performed. Cosmic ray events in each input image were detected and replaced by the average of the four neighbors with the IRAF task cosmicrays. The remaining cosmic ray hits were flagged manually and were subsequently removed by interpolation.

We extracted the spectra using objectively defined apertures. The peak of the galaxy light distribution was used to trace the extraction aperture. For the object spectra, we used a fixed aperture, 6 pixels along the slit centered on the brightest pixel in the portion of the spectrum between 5300-5600 Å. The background sky level was determined from areas as close to the nucleus as possible, taking care not to include the contribution from extended emission near the nucleus. One-dimensional spectra of the standard stars were extracted in exactly the same manner as for the galaxy spectra, using an effective slit length to contain all the stellar light.

Wavelength calibration was carried out by fitting a cubic spline to unblended emission lines of He, Ar, and Fe in the comparison lamp spectra. These spectra were also used to measure the spectral resolution as a function of position on the CCDs. More than 20 lines were used to establish the wavelength scale. Typical rms residuals in the cubic spline fits were 0.3 Å. The accuracy of the wavelength calibration was better than 1.5 Å.

The extracted spectra were flux calibrated on a relative flux scale using more than two KPNO standard stars. Cubic spline sensitivity functions of ninth order were fit interactively for each of the standard star observations. The sensitivity function relates the measured intensity to the (calibrated) flux density (in ergs s^-1 cm^-2 Å^-1) as a function of wavelength, after removing atmospheric extinction.

Atmospheric extinction was corrected using the mean extinction coefficients for the Xinglong station (BAO), that were measured in the Beijing-Arizona-Taiwan-Connecticut (BATC) multi-color survey (Kong et al. 2000). There is little error introduced by this procedure, since the observations were restricted to small air masses, usually less than 1.2 and always less than 1.7.

The telluric O₂ absorption lines near 6280 and 6860 Å (the "B band'') were removed through division by normalized, intrinsically featureless spectra of the standard stars. Large residuals caused by mismatches at the sharp, deep band were not removed.

4.2 Additional reductions

To measure the rest-frame spectral line properties of the galaxies, we first measured the recession velocity of each galaxy by averaging the recession velocities, $V_{\rm o} =c \Delta \lambda / \lambda_0$ of different lines, where $\lambda_0$ is the rest-frame wavelength of the line.

For galaxies with emission lines, recession velocities were determined from the average of five measurable emission lines, [O  II]$\lambda$3727, H$\beta$$\lambda$4861, [O  III]$\lambda$4959, [O  III]$\lambda$5007, H$\alpha$$\lambda$6563. For those objects without emission lines, velocities were obtained from the average of 5 measurable absorption lines, Ca K$\lambda$3935, H$\delta$$\lambda$4101, G band $\lambda$4306, Mg  I+Mg  H$\lambda$5177 and Na  I$\lambda$5896. For the galaxy, IVZw 67, that has neither strong emission nor absorption lines, we adopted the redshift listed by NED. The accuracy of the recession velocities in our sample ranges from very good (3-55 km s^-1 uncertainties) when the galaxy has strong emission lines, to relatively poor (7-67 km s^-1 uncertainties) when the galaxy has only absorption lines. The recession velocity distribution for all the sample galaxies with spectroscopic observations from this study are presented in Fig. 3. All galaxies in the sample have velocities below 9000 km s^-1. On the basis of these recession velocities , the calibrated spectra were shifted to rest-frame wavelengths. Columns 8-9 of Table 3 list the galaxy recession velocities in km s^-1, the uncertainty of the recession velocity.

Figure 3: Histogram of the recession velocity distribution for all the 97 sample galaxies.

The foreground reddening cause by our Galaxy was corrected using the values in Col. 7 of Table 1. The wavelength dependence of the extinction was assumed to follow the empirical selective extinction function of Cardelli et al. (1989), with R_V=A_V/E(B-V)=3.1.

The rest frame spectra were normalized to the average flux in a 50 Å interval centered at 5500 Å and subsequently resampled on a uniform wavelength grid spanning the range 3580-7600 Å.

For each galaxy, multiple exposures taken with the same setting, sometimes over several epochs, were combined in a weighted average, with the weights determined by the signal-to-noise ratios.