All Figures

$\begin{figure} \par\includegraphics[width=17.05cm,clip]{1974fig1.ps} \end{figure}$ Figure 1: Examples of observed spectra. The left panels show the spectra as obtained by the blue grism, the right column that of the red grism. The upper panel shows spectra of a red giant star, the lower for a main sequence star in M 55, i.e. [Fe/H] = -1.8 dex. Note the larger number of absorption lines in the spectrum of the cool giant. In the lower plot the wavelength ranges used for the neural network analysis are indicated.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=8.5cm,clip]{1974fig2.eps} \end{figure}$ Figure 2: Distributions of the velocity errors of the red grism ( left column) and blue grism ( right column) for the spectra of M 55 ( top panels) and $\omega$ Cen ( bottom panels). The errors in the red grism are somewhat larger because of a smaller number of prominent absorption lines available for the cross-correlation procedure.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=6.55cm,clip]{1974fig3.ps} \end{figure}$ Figure 3: Photometry in M 55. The upper plot is the pre-imaging Johnson BV photometry for a selected range of luminosities. Two isochrones both with 13.5 Gyr and [Fe/H] = -1.8 dex are overplotted. The left one has solar scaled $\alpha$ abundances, the right one has [ $\alpha$ /Fe] = 0.3 dex. The lower plot shows v-y Strömgren photometry and an isochrone with [Fe/H] = -1.8 dex and 16 Gyr (the old age scale). The heavy points show the stars for which spectra have been obtained.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=8.2cm,clip]{1974fig4.ps}\end{figure}$ Figure 4: A sketch to help visualize the procedure to find $\log~g$ and $T_{\rm eff}$ values from the Strömgren photometry. The solid line is an isochrone in the three-dimensional magnitude space, while the ellipsoid is a three-dimensional error box about the measured point. The size of the error ellipsoid has been magnified by a factor 15 and displaced for clarity. The surface gravity and temperature for the measured point is found from the shortest distance between this point and the isochrone. The errors for these two stellar parameters are found by projecting the error ellipsoid onto the isochrone and taking the difference between the extreme values of $T_{\rm eff}$ and of $\log~g$ . This is shown in the two-dimensional projection on the b,v plane.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=8.5cm,clip]{1974fig5.eps} \end{figure}$ Figure 5: Parametrization results for $\log~g$ ( upper line) and $T_{\rm eff}$ ( lower line) for stars in M 55. The left panels shows the parameter values as determined from the isochrone versus the corresponding computed value from the network for the two parameters. The solid line is the identity. The horizontal and vertical error bars for each point are calculated from the photometric error ellipse (see Fig. 4) and from 30 bootstrap networks, respectively. The right panel shows the boxplots of the residuals for each parameter, Delta = (computed value - isochrone value). The solid line in the box shows the median value of the residuals while the borders of the box are the first and third quartiles. Single points which have a larger distance to the median than 1.5 times the boxlength and are identified as outliers. The overall parametrization errors as defined in Eq. (3) are 0.32 dex and 190 K for the two parameters.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=8.05cm,clip]{1974fig6.eps} \end{figure}$ Figure 6: Distribution of the metallicities for stars (combined MSSGB and RGB sample) in M 55 ( left panel) and $\omega$ Cen ( right panel) as determined by our method. The dashed line is centred at -1.80 dex (the literature value of [Fe/H] for M 55) while the solid line denotes the mean of the distribution. There is only a small offset of 0.06 dex. The overall error for the determined metallicities in M 55 as defined in Eq. (3) is 0.15 dex.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=7.8cm,clip]{1974fig7.eps} \end{figure}$ Figure 7: As in Fig. 5 but for the "standard'' stars showing the network outputs compared to the averaged literature values for each of the parameters [Fe/H], $\log~g$ and $T_{\rm eff}$ ( from top to bottom). The errors for the literature values are found from the scatter about the unweighted mean value for the different stated values (see Table 2), while the computed errors are found from 30 bootstrap networks. Note that the metallicity range on which the networks were trained was limited to values larger than -2.5 dex, so we cannot expect good parametrization performances for metallicities below this value. Excluding the single point with the literature value [Fe/H] = -2.95 dex, we find an overall average error in [Fe/H] as defined in Eq. (3) of 0.15 dex, while that for $\log~g$ and $T_{\rm eff}$ are 0.42 dex and 145 K respectively.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=7.5cm,clip]{1974fig8.ps} \end{figure}$ Figure 8: The spectroscopically observed stars in $\omega$ Cen are marked as heavy dots in the Strömgren colour magnitude diagram from Hilker & Richtler (2000). Based on the radial velocities, filled circles were identified as members, while open circles denote non-members.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=16.75cm,clip]{1974fig9.ps} \end{figure}$ Figure 9: A comparison of two observed standard star spectra (solid lines) with the corresponding synthetic spectra (dotted lines) as predicted by the neural network. The network predicts only the stellar parameters, but from these the synthetic spectra were generated through interpolation of the training template grid of synthetic spectra. The upper plot is for a star with stellar parameter residuals of $\Delta$ ([Fe/H]) = -0.04 dex, $\Delta$ ( $\log~g$ ) = 0.12 dex and $\Delta$ ( $T_{\rm eff}$ ) = -89 K, where $\Delta$ is the difference (network output - literature value). The corresponding network outputs are [Fe/H] = -1.06 dex, $\log~g$ = 4.16 dex and $T_{\rm eff}$ = 5811 K. The lower plot shows an example were the difference between the $\log~g$ prediction of the network and the literature value is rather large. For this star, we find $\Delta$ ([Fe/H]) = -0.01 dex, $\Delta$ ( $\log~g$ ) = 1.11 dex and $\Delta$ ( $T_{\rm eff}$ ) = 109 K and networks outputs of [Fe/H] = -2.15 dex, $\log~g$ = 4.03 dex and $T_{\rm eff}$ = 4984 K. The S/N of the synthetic spectra was decreased to 150, i.e. similar to that of the observed spectra.
Open with DEXTER

In the text

$\begin{figure} \par\includegraphics[width=8.5cm,clip]{1974fig2.eps} \end{figure}$	Figure 2: Distributions of the velocity errors of the red grism ( left column) and blue grism ( right column) for the spectra of M 55 ( top panels) and $\omega$ Cen ( bottom panels). The errors in the red grism are somewhat larger because of a smaller number of prominent absorption lines available for the cross-correlation procedure.
Open with DEXTER

$\begin{figure} \par\includegraphics[width=7.5cm,clip]{1974fig8.ps} \end{figure}$	Figure 8: The spectroscopically observed stars in $\omega$ Cen are marked as heavy dots in the Strömgren colour magnitude diagram from Hilker & Richtler (2000). Based on the radial velocities, filled circles were identified as members, while open circles denote non-members.
Open with DEXTER