0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 Fiber diameter/Major axis
0 5 10 15 20 25 30 35 40 45
N
(a) Fiber diameter over the major axis length
0.000 0.005 0.010 0.015 0.020 0.025 0.030 Fiber area/Galaxy area
0 5 10 15 20 25 30 35 40 45
N
(b) Fiber area over the galactic area
Figure 7.8: Distribution of the fiber size with respect to the galaxy size for the galaxies in the SDSS subsample. The first subplot show the distribution of the fiber diameter with respect to the length of the major axis of the galaxy. The second subplot shows the distribution of the area covered by the fiber with respect to the area of the galaxy.
starlight.ufsc.br/. We installed the Starlight binaries in different computers.
Some of them were able to manage batch queues and parallelise the jobs and others did it in a serial way. Depending on the hardware platform of each computer we installed the 32-bit version or the two 64-bit binaries, one compiled with theIntel ifortcompiler and the other compiled withGNU gcc. We used the statically linked versions of the binaries to avoid problems with library dependences.
The program takes as input parameters: a) an array with the wavelength, flux, error in the flux and error mask, b) the selected set of spectral bases, c) the initial values for the velocity shift and dispersion and d) a set of advanced options. The input of the array is done in a file. The wavelength has to be uniformly sampled and the flux must be calibrated. The error is an estimation of the flux error and the error mask signals if the corresponding value is good or has to be flagged as incorrect. We used as the spectral base 45 synthetic spectra, with 3 metallicities and 15 different ages, of Bruzual & Charlot(2003). The 3 metallicities are: 0.004 (0.2 Z); 0.02 (Z) and 0.05 (2.5 Z) and the 15 ages are: 1.0 Myr; 3.16 Myr; 5.01 Myr; 10.0 Myr; 25.12 Myr;
40.0 Myr; 101.52 Myr; 286.12 Myr; 640.54 Myr; 904.79 Myr; 1.434 Gyr; 2.5 Gyr;
5.0 Gyr; 11.0 Gyr and 13 Gyr. These base spectra were distributed with Starlight and are reliable to fit the stellar populations of our galaxies (Cid Fernandes et al. 2005).
We made some programs to adapt the original SDSS spectra to the input of Starlight.
The programs performed these steps:
The first task was to uniformly resample the spectra. We used a spline interpo- lation with a very large sampling rate (small differences in wavelength) to avoid loosing flux and then resampled the spectrum with a spectral resolution of 1 Å.
SDSS wavelengths are given in vacuum wavelengths instead of the normally used air wavelengths. Air wavelengths were calculated from vacuum wave- lengths based on IAU standards (Morton 1991).
The spectra are affected by reddening due to the gas and dust of the Milky Way.
We performed a dereddening of the flux and of the error in the flux based on the dust emission maps ofSchlegel et al.(1998). The extinction curve used was the one ofO’Donnell(1994) which is an improvement in the optical and the near infrared of the extinction curve ofCardelli et al.(1989).
We shifted the spectra to the rest-frame with the redshifts taken from the Prince- ton re-reduction of the SDSS data.
We also took the velocity dispersion from the Princeton re-reduction as the ini- tial value because it was usually more accurate that the one obtained from the SDSS default pipeline.
After some tests we found out that there were differences between the results ob- tained with different binaries. Sometimes one of the binaries did not find a fitting to the stellar population for a galaxy while the others did. There were also differences depending on the mask array that we chose to enter to the program. We used two different types of masks for each galaxy:
The first type flagged all the errors marked in the original SDSS FITS file, i.e.
where the error flag was not set to 0.6Some of these errors were actually minor
6The error mask is an array of the same length of the flux data. It contains the error code assigned
7.4. DETERMINATION OF THE STELLAR POPULATIONS 73 errors, e.g. a low signal to noise ratio in some pixel. But they can significantly affect the number of points to be fitted in the spectra.
The second type flagged only errors that we considered major error, i.e. where the error flag in the SDSS FITS file was set to a number above or equal to 16777216 (0x1000000) and below 67108864 (0x4000000) which correspond to the errors:“No data available in combine B-spline”and“Rejected in combine B-spline”respectively.
In some cases the first type of mask gave better results while in other cases the second type was the best. As we wanted the best fit independently of the mask and the binary used, we performed a fit with each binary (32-bit, 64-bit ifort and 64-bit gcc) and each kind of mask (with all the errors and with only major errors) for each galaxy. After doing the six different fits we compared among them to find the best result for each galaxy.
To compare among the different fits we used theχ2parameter. Starlight estimates this parameter, the exact output parameter isχ2divided by the total number of spectral points that could be fitted by the program. Thus, the parameter takes into account not only the goodness of the fitted zone but also the width of spectral range that was correctly fitted. Although this parameter is a good estimator for the accuracy of the fit we empirically found that a new parameter: χ2/(Nfit)2 is even better to discriminate the best fit. In Figure7.9and Figure7.10are plotted the different fits for two galaxies, CIG 366 and CIG 437. For CIG 437 the better spectral fit usingχ2/(Nfit) is the one obtained using the 64-bitgcccompiled version of Starlight with the mask including all the errors. But is clear from the image that this is not the best fit. This is due to the small number of fitted points. On the other hand the better fit withχ2/(Nfit)2is the one of theifortversion with only the major errors masked. As we said,χ2/(Nfit)2works better thanχ2/(Nfit) to choose the best fit. In the case of CIG 366 thegcccompiled version of Starlight could not find a fit when we used the mask with all the errors.
for each spectral point (what we call error flag) denoted with an hexadecimal code. The error codes are explained in detail in:http://www.sdss.org/dr6/dm/flatFiles/spSpec.htmland are reproduced for reference in AppendixB.1