We decided not to use the same
procedure to straighten the orders, because of the presence of a coloured
filter in the beam. This
filter is necessary to balance the exposures at the red end and the violet
end of the tungsten lamp.
But it implies that the spectrum is not flat along the wavelength coordinate.
Therefore, polynomials of order 7 to 19 were fitted to the continuum of
four very metal-poor (almost line-free)
stars : HD 221170, [Fe/H] = -2.10,
HD 216143, [Fe/H] = -2.15, HD 108317,
[Fe/H] = -2.36, HD 140283, [Fe/H] = -2.53 (Soubiran et al. 1998, paper II).
For each order k, the response
curve Bk(x) is given by the polynomial which is less affected by the spectral lines,
with respect to the three others.
Instead of using the tungsten response, the observed spectra are straightened by
dividing their flux,
order by order, pixel by pixel, by Bk(x), normalized at 1 at its maximum.
For further uses all the pixels that correspond to a response value lower
than 0.5 at the edges of the orders are rejected by giving them the flag value -100.0.
Orders 138 to 157 (the bluest spectral orders) were rejected, because
they were underilluminated with respect to the rest of the orders,
and too much degraded by noise.
If the spectral energy distribution (SED) of the target star is very similar to that of the star which was used for straightening the order, no significant slope in the SED of the straightened order is expected. If the two stars have different colours, a residual slope is expected. Initially, the adjustement of the slope between the target and a reference star was performed in the least square fit of the flux levels. However a better consistency was attained in removing this adjustment. In practice the slope remains very small over an order: about 1% from edge to edge for a 1200 K difference in effective temperature. The ``successful'' comparison star has of course a temperature difference considerably smaller than this, with respect to the target star, so the slope adjustment is not necessary. Indeed it is even unwise, because if slope adjustment is provided, the software try to compensate the presence of a strong line at the edge of an order in a comparison star, by inventing an unphysical strong slope in the target star, when the line is non-existent or much weaker in the target star (or vice versa). Figure 1 shows order 100 of HD 25329 before and after straightening.