Light scattering within the spectrograph causes unwanted exchange of light between the o and e beams. This results in loss of contrast between the beams (i.e. scale error in degree of polarization) which must be corrected. As discussed in the section on sky subtraction, scattered light from the object also affects the light level in the adjacent sky apertures and the apparent sky polarization will be wrong.
Both sky and lamps flood the entire slit area and there could be excess scattered light (bypassing the slit area components), as a diffuse background overlying the whole frame, a stellar image causing less of such diffuse scattered light. This could upset the determination of pixel sensitivities. Our preliminary impression is that such diffuse scattering, though present, is negligible.
To measure cross talk between the o and e spectra for your specific instrumental setup, observe a star through a calibration polarizer with the halfwave plate set to such an angle that (almost) all the light gets into only one of the beams. The light detected between spectra must be due to scattered light from the illuminated parts of the slit and can be used to estimate the crosstalk level from one spectrum to another. Tests of this kind during commissioning showed a few tenths of a percent of the light in one beam was scattered into the adjacent beam (depending on wavelength).
It should be noted that scattered light from other parts of the spectra (other wavelengths) can have consequences, particularly if P( ) is structured.
Optimal solutions of the scattered-light problem will involve modeling ISIS' scattering properties (which will change with time and depend on incident spectrum and spectrograph settings). In due course, we shall be able to give more useful advice, at present the best we can do is alert you to the problem. In many cases, the solutions suggested in this section and that on sky subtraction will suffice. Refer to Appendix C for more detail; as we gain experience, that appendix will evolve.
Recapitulating, for point source observations the following applies:
[ TIFF ]
Figure 7: Linear polarization as a vector quantity. In each case, the complete circle denotes the rms error, the circular arcs are loci of constant degree of polarization. See text.