2 Observations and data reduction

2.1 Observations

We observed 12 Galactic globular clusters, 9 of which are located close to the Milky-Way bulge (see Fig. 1). Four globular clusters belong to the halo sub-population with a mean metallicity $\rm [Fe/H]\leq-0.8$ dex (Harris 1996). The other globular clusters with higher mean metallicities are associated with the bulge. Our sample includes the well-studied metal-rich clusters NGC 6553 and NGC 6528, which is located in Baade's Window. Several relevant cluster properties are summarized in Table 1. Our cluster sample was selected to maximize the number of high-metallicity clusters and to ensure a high enough signal-to-noise ratio (S/N) of the resulting spectra.

Long-slit spectra were taken on three nights in July 5th to 7th 1999 with the Boller & Chivens Spectrograph of ESO's 1.52 m on La Silla. We used grating #23 with 600 grooves per mm yielding a dispersion of 1.89 Å/pix with a spectral range from $\sim$3400 Å to $\sim$7300 Å. We used the detector CCD #39, a Loral $2048\times2048$ pix² chip, with a pixel size of 15 $\mu$m and a scale of 0.82 $^{\prime\prime}$/pix. Its readout noise is 5.4 e^- and the gain was measured with 1.2 e^-/ADU. In order to check the dark current we also obtained dark images which resulted in a negligible average dark current of 0.0024 e $^{-}~{\rm s}^{-1}~{\rm pix}^{-1}$. The total slit length of the spectrograph covers 4.5$^\prime$ on the sky. For the benefit of light sampling the slit width was fixed at 3 $^{\prime\prime}$, which guarantees an instrumental resolution ($\sim$6.7 Å) which is smaller than the average resolution ($\ga$8 Å) of the Lick standard system (Worthey et al. 1994; Trager et al. 1998). The mean seeing during the observing campaign varied between 0.8 $^{\prime\prime}$ and 1.6 $^{\prime\prime}$, resulting in seeing-limited spectra. Consequently, the stellar disks are smeared over 1-2 pixel along the spatial axis.

To ensure a representative sampling of the underlying stellar population we obtained several spectra with slightly offset pointings. In general three long-slit spectra were taken for each of our target clusters (see Table 2 for details). The observing pattern was optimized in time (i.e. in airmass) to obtain one spectrum of the nuclear region and spectra of adjacent fields by shifting the telescope a few arc seconds (i.e. $\sim$2 slit widths) to the North and South. Exposure times were adjusted according to the surface brightness of each globular cluster to reach an statistically secure luminosity sampling of the underlying stellar population. Before and after each block of science exposures, lamp spectra were taken for accurate wavelength calibration.

In addition to the globular cluster data, we obtained long-slit spectra of three stellar fields near the Galactic center (see Fig. 1). Two of them are located in Baade's Window. The exposure time for a single bulge spectrum is 1800 s. Five slightly offset pointings have been observed in each field resulting in 15 exposures of 30 min each.

During each night Lick and flux standard stars were observed for later index and flux calibrations. Table 2 shows the observing log of all three nights. Figure 1 gives the positions of all observed globular clusters (filled dots) and bulge fields (open squares) in the galactic coordinate system.

Figure 1: Distribution of galactic globular clusters as seen in the galactic coordinate system. The filled circles are the observed sample globular clusters while open circles mark the position of other known Milky Way globular clusters. All observed globular clusters are appropriately labeled. The positions were taken from the Globular Cluster Catalog by Harris (1996). Large squares show the positions of our three bulge fields for which spectroscopy is also available. Note that two of the three fields almost overlap in the plot.

 

 

Table 2: Journal of all performed observations.

Night	Targets	Exptime	RA(J2000)	Dec (J2000)	l[$^{\circ}$]	b[$^{\circ}$]
5.7.1999	NGC 5927	3$~\times~$600 s	15h 28m 00.5 s	$-50^{\circ}40'22''$	326.60	4.86
	NGC 6388	3$~\times~$600 s	17h 36m 17.0 s	$-44^{\circ}44'06''$	345.56	-6.74
	NGC 6528	3$~\times~$600 s	18h 04m 49.6 s	$-30^{\circ}03'21''$	1.14	-4.17
	NGC 6624	3$~\times~$600 s	18h 23m 40.5 s	$-30^{\circ}21'40''$	2.79	-7.91
	NGC 6981	1$~\times~$1320 s	20h 53m 27.9 s	$-12^{\circ}32'13''$	35.16	-32.68
	Bulge1	5$~\times~$1800 s	18h 03m 12.1 s	$-29^{\circ}52'06''$	1.13	3.78
6.7.1999	NGC 6218	3$~\times~$1200 s	16h 47m 14.5 s	$-01^{\circ}56'52''$	15.72	26.31
	NGC 6441	3$~\times~$600 s	17h 50m 12.9 s	$-37^{\circ}03'04''$	353.53	-5.01
	NGC 6553	3$~\times~$720 s	18h 09m 15.6 s	$-25^{\circ}54'28''$	5.25	-3.02
	NGC 6626	3$~\times~$600 s	18h 24m 32.9 s	$-24^{\circ}52'12''$	7.80	-5.58
	NGC 6981	1$~\times~$1800 s	20h 53m 27.9 s	$-12^{\circ}32'13''$	35.16	-32.68
	Bulge2	5$~\times~$1800 s	18h 05m 21.3 s	$-29^{\circ}58'38''$	1.26	4.23
7.7.1999	NGC 6284	3$~\times~$600 s	17h 04m 28.8 s	$-24^{\circ}45'53''$	358.35	9.94
	NGC 5927	2$~\times~$600 s	15h 28m 00.5 s	$-50^{\circ}40'22''$	326.60	4.86
	NGC 6356	3$~\times~$900 s	17h 23m 35.0 s	$-17^{\circ}48'47''$	6.72	10.22
	NGC 6637	3$~\times~$900 s	18h 31m 23.2 s	$-32^{\circ}20'53''$	1.72	-10.27
	NGC 6981	1$~\times~$1800 s	20h 53m 27.9 s	$-12^{\circ}32'13''$	35.16	-32.68
	Bulge3	5$~\times~$1800 s	17h 58m 38.3 s	$-28^{\circ}43'33''$	1.63	2.35

   
2.2 Data reduction

We homogeneously applied standard reduction techniques to the whole data set using the IRAF platform (Tody 1993). The basic data reduction was performed for each night individually. In brief, a masterbias was subtracted from the science images followed by a division by a normalized masterflat spectrum which has been created from five quarz-lamp exposures. The quality, i.e. the flatness, of the spectra along the spatial axis was checked on the sky spectra after flatfielding. Any gradients along the spatial axis were found to be smaller than $\lesssim$5%.

He-Ne-Ar-Fe lines were used to calibrate all spectra to better than 0.13 Å (rms). Unfortunately, the beam of the calibration lamp covers only the central 3.3$^\prime$ along the slit's spatial axis (perpendicular to the dispersion direction), which allows no precise wavelength calibration for the outer parts close to the edge of the CCD chip. We tried, however, to extrapolate a 2-dim. $\lambda$-calibration to the edges of the long-slit and found a significant increase in the rms up to an unacceptable 0.7 Å. Hence, to avoid calibration biases we use data only from regions which are covered by the arc lamp beam. Our effective slit length is therefore 3.3$^\prime$ with a slit width of 3 $^{\prime\prime}$. For each single pixel row along the dispersion axis an individual wavelength solution was found and subsequently applied to each object, bulge, and sky spectrum. After wavelength calibration the signal along the spatial axis was averaged in $\lambda$-space, i.e. the flux of 3.3$^\prime$ was averaged to obtain the final spectrum of a single pointing.

Finally, spectrophotometric standard stars, Feige 56, Feige 110, and Kopff 27 (Stone & Baldwin 1983; Baldwin & Stone 1984) were used to convert counts into flux units.

   
2.3 Radial velocities

All radial velocity measurements were carried out after the subtraction of a background spectrum (see Sect. 3.1) using cross-correlation with high-S/N template spectra of two globular clusters in M 31 (i.e. 158-213 and 225-280, see Huchra et al. 1982 for nomenclature). Both globular clusters have metallicities which match the average metallicity of our globular cluster sample. We strictly followed the recipe of the Fourier cross-correlation which is implemented in the FXCOR task of IRAF (see IRAF manual for details). Table 1 summarizes the results including the internal uncertainties of our measurements resulting from the fitting of the cross-correlation peak.

Following the rule of thumb, by which 1/10 of the instrumental resolution ($\sim$6.7 Å) transforms into the radial velocity resolution, we estimate for our spectra a resolution of $\sim$40 km s^-1. In order to estimate the real uncertainty we compare the radial velocity measurements of one globular cluster (NGC 6981) which was observed in all three nights. We find a dispersion in radial velocity $\sigma_{\rm v}\approx17$ km s^-1 and a maximal deviation of 32.4 km s^-1. A comparison of measured radial velocities of all our Lick standard stars with values taken from the literature gives a dispersion of $\sigma_{\rm v}\approx40$ km s^-1 which matches the earlier rough estimate. In the case of NGC 6981, the internal error estimate ( $\Delta_{\rm cc} v_{\rm rad}=18.4$ km s^-1) underestimates the real radial velocity uncertainty assumed to be of the order of $\sim$40 km s^-1 by a factor of $\sim$2. Note however, that data of lower S/N will produce larger radial velocity uncertainties. Moreover, taking into account the slit width of 3 $^{\prime\prime}$ the maximum possible radial velocity error for a star positioned at the edge of the slit is $\sim$200 km s^-1. For high surface-brightness fluctuations inside the slit, this would inevitably result in larger radial velocity errors than originally expected from the calibration quality. Since we sum up all the flux along the slit, we most effectively eliminate this surface-brightness fluctuation effect. In fact, after a check of all our single spectra, we find no exceptionally bright star inside the slit aperture, which could produce a systematic deviation from the mean radial velocity.

After all, we estimate that our real radial velocity uncertainties are larger by a factor $\sim$2-4 than the values given in Table 1.

   
2.4 Transformation to the lick system

The Lick standard system was initially introduced by Burstein et al. (1984) in order to study element abundances from low-resolution integrated spectra of extragalactic stellar systems. It has recently been updated and refined by several authors (González 1993; Worthey et al. 1994; Worthey & Ottaviani 1997; Trager et al. 1998). The Lick system defines line indices for specific atomic and molecular absorption features, such as Fe, Mg, Ca and CN, CH, TiO, in the optical range from $\sim$4100 Å to $\sim$6100 Å. The definitions of a line index are given in Appendix A. We implemented the measuring procedure in a software and tested it extensively on original Lick spectra (see Appendix A for details). This code is used for all further measurements.

The Lick system provides two sets of index passband definitions. One set of 21 passband definitions was published in Worthey et al. (1994) to which we will refer as the old set. A new and refined set of passband definitions is given in Trager et al. (1998) which is supplemented by the Balmer index definitions of Worthey & Ottaviani (1997). This new set of 25 indices is used throughout the subsequent analysis. However, we also provide Lick indices based on the old passband definitions (see Appendix D) which enable a consistent comparison with predictions from SSP models which make use of fitting functions based on the old set of passband definitions. Note that indices and model predictions which use two different passband definition sets are prone to systematic offsets. This point will be discussed in the second paper of the series (Maraston et al. 2002).

Before measuring indices, one has carefully to degrade spectra with higher resolution to adapt to the resolution of the Lick system. We strictly followed the approach of Worthey & Ottaviani (1997) and degraded our spectra to the wavelength-dependent Lick resolution ($\sim$11.5 Å at 4000 Å, 8.4 Å at 4900 Å, and 9.8 Å at 6000 Å). The effective resolution (FWHM) of our spectra has been determined from calibration-lamp lines and isolated absorption features in the object spectra. The smoothing of our data is done with a wavelength-dependent Gaussian kernel with the width

$$\sigma_{\rm smooth}(\lambda)=\left(\frac{{\it FWHM}(\lambda)_... ...ambda)_{\rm data}^2}{8~ {\rm ln}~2}\right)^{\frac{1}{2}}\cdot$$ (1)

We tested the shape of absorption lines in our spectra and found that they are very well represented by a Gaussian. Worthey & Ottaviani tested the shape of the absorption lines in the Lick spectra and found also no deviation from a Gaussian. Both results justify the use of a Gaussian smoothing kernel.

The smoothing kernel for the bulge stellar fields is generally narrower since one has to account for the non-negligible velocity dispersion of bulge field stars. A typical line-of-sight velocity dispersion $\sigma_{\rm LOS}\approx 100$ km s^-1 was assumed for the bulge data (e.g. Spaenhauer et al. 1992). We do not correct for the mean velocity dispersion of the globular clusters ( $\sigma_{\rm LOS}\approx 10$ km s^-1 Pryor & Meylan 1993).

Another point of concern for low-S/N spectra ( $S/N\lesssim10$ per resolution element) is the slope of the underlying continuum (see Beasley et al. 2000, for detailed discussion of this effect) which influences the pseudo-continuum estimate for broad features and biases the index measurement. However, since all our spectra are of high S/N($\gtrsim$50 per resolution element), we are not affected by a noisy continuum.

After taking care of the resolution corrections, one has to correct for systematic, higher-order effects. These variations are mainly due to imperfect smoothing and calibration of the spectra. To correct the small deviations 12 index standard stars from the list of Worthey et al. (1994) have been observed throughout the observing run. Figure 2 shows the comparison between the Lick data and our index measurements for all passbands. Least-square fits using a $\kappa$-$\sigma$-clipping (dashed lines) are used to parameterize the deviations from the Lick system as a function of wavelength. The functional form of the fit is

$${EW}_{\rm cal} = \alpha + (1+\beta)\cdot{EW}_{\rm raw},$$

where ${EW}_{\rm cal}$ and ${EW}_{\rm raw}$ are the calibrated and raw indices, respectively. Table 3 summarizes the individual coefficients $\alpha$ and $\beta$. This correction functions are applied to all further measurements. The corresponding coefficients for index measurements using the old passband definitions are documented in Table D.1.

 

 

Table 3: Summary of the coefficients $\alpha$ and $\beta$ for all 1st and 2nd-order index corrections.

index	$\alpha$	$\beta$	rms	units
CN1	-0.0017	-0.0167	0.0251	mag
CN2	-0.0040	-0.0389	0.0248	mag
Ca 4227	-0.2505	-0.0105	0.2582	Å
G4300	0.6695	-0.1184	0.4380	Å
Fe 4384	-0.5773	0.0680	0.2933	Å
Ca 4455	-0.1648	0.0249	0.4323	Å
Fe 4531	-0.3499	0.0223	0.1566	Å
Fe 4668	-0.8643	0.0665	0.5917	Å
H$\beta$	0.0259	0.0018	0.1276	Å
Fe 5015	1.3494	-0.2799	0.3608	Å
Mg1	0.0176	-0.0165	0.0160	mag
Mg2	0.0106	0.0444	0.0112	mag
Mgb	0.0398	-0.0392	0.1789	Å
Fe 5270	-0.3608	0.0514	0.1735	Å
Fe 5335	-0.0446	-0.0725	0.3067	Å
Fe 5406	-0.0539	-0.0730	0.2054	Å
Fe 5709	-0.5416	0.3493	0.1204	Å
Fe 5782	-0.0610	-0.0116	0.2853	Å
NaD	0.3620	-0.0733	0.2304	Å
TiO1	0.0102	0.2723	0.0133	mag
TiO2	-0.0219	0.1747	0.0342	mag
H $\delta _{\rm A}$	-0.1525	-0.0465	1.5633	Å
H $\gamma_{\rm A}$	0.4961	0.0117	0.6288	Å
H $\delta_{\rm F}$	-0.1127	-0.0639	0.4402	Å
H $\gamma _{\rm F}$	-0.0062	-0.0343	0.1480	Å

Note, that most passbands require only a small linear offset, but no offset as a function of index strength. While the former is simply due to a small variation in the wavelength calibration, the latter is produced by over/under-smoothing of the spectra. Absorption lines for which the smoothing pushes the wings outside narrowly defined feature passbands are mostly affected by this non-linear effect. However, for passbands of major interest (such as CN, H$\beta$, Fe 5270, Fe 5335, Mgb, and Mg₂) the Lick indices are satisfactorily reproduced by a simple offset (no tilt) in the index value (see Fig. 2).

Figure 2: Comparison of passband measurements from our spectra and original Lick data for 12 Lick standard stars. The dotted line shows the one-to-one relation, whereas the dashed line is a least-square fit to the filled squares. Data, which have been discarded from the fit because of too large errors or deviations, are shown as open squares. Bold frames indicate some of the widely used Lick indices which are also analysed in this work.