After a simple gaussian centering the y-positions of the orders can be fit very accurately as a function of x as the input for the later data reduction routines like the extraction of spectra. Here the orders were fit with an observed star, but for FORS2 we will do the centering with the spectrum of the internal flat field lamps through the pinhole.
Now the position of the lines from the wave length calibration lamps can be estimated with the corrected position in y and the forecasted positions in x. Here we used a Thorium Argon lamp in combination with a catalog for which we have selected the brightest lines only. Here the scale is different from the earlier figures. It seam to me that the focal length of the camera is longer then the value used for the calculation.
The first step of the wavelength calibration was now to apply the global fit as a function of the x-position and the order only. After the first centering at the forecasted position and a second iteration of the line search with the new dispersion solution, we have already got a dispersion solution with a standard deviation of 0.11Å which is acceptable for almost all applications. All identified lines are marked blue, the white symbols are all lines of the input catalog. The input catalog should be better optimized for the spectral resolution of DFOSC.
The residual deviation of the fit from the laboratory wave length show the good result. Red symbols denote lines which have been rejected as blends and wrong identifications. Many lines are not taken into account, because the input catalog was not well selected for intermediate resolution spectra
The lines are now centered along the slit and again a residual of 0.11Å was obtained for the 3D-fit as a function of X, order and delta y. The parameters of the global solution in two dimensions were hold.
The residuals of the 3D chi square optimization are presented in the next figure:
With the known order positions the raw spectra of all orders can be extracted now. Note that the bad columns were not replaced yet. Here a simple linear rebin can be done because all pixels are shifted in y only
The raw flat field can be normalized for the later reduction steps like the optimum extraction and the fit of the terrestric night sky lines.
After the division by the normalized flat the sky can be subtracted either fitting polynoms or calculating the median over a window, on the part of the slit without light from the target. The next two figures show the rectified spectrum before and after the subtration of the night sky lines. The sky spectrum is calculated at the position of the raw spectrum, before rectifying the frames, similar to the normalized flat field
Now the sky subtracted frame can be rebined to the same wavelength steps. Only the usefull part of the spectra is kept again
Finally the spectra of the single orders can be extracted now using a modified optimum extraction method which considers sub-pixel sampling effects. Note that the very strong cosmic along the spatial direction in order 12 was not removed. Below you see the rebinned spectra of the WR+O binary HD 5980 which showed a remarkable outburst several years ago. The spectrum was devided by the approximate blaze function, which was calculated for the flat field normalization. Note that the fringes were unexpectedly removed from the data, although no standard star observation was reduced yet.