3 Observational technique

  The magnification giving a pixel scale of 0.05'' has been selected to ensure an adequate sampling of the PSF, at H band. The field was thus 12.8 $\times$ 12.8''; for the study of extended sources larger than the field size it is obviously necessary to employ several pointings and mosaic the resulting images after basic data reduction. ADONIS has a limit of 30'' for the radial extent of the offset sky so values less than [30 - half detector size] ('') must be employed in order to have unvignetted sky background frames. Special care has been taken in the selection of the offset sky position to avoid any overlapping with the extended object observed. For all sources, object and chopped sky images were obtained at each position of the polarizer. A data cube of 256 $\times$ 256 spatial pixels $\times$ M frames, where M is the number of object and sky frames, was acquired. Table 1 lists the details of the ADONIS polarimetry observations of the science and calibration sources.

Sets of chopped images were obtained at nine different positions of the polarizer, each 22.5$^\circ$ apart, from 0 to 180$^\circ$. The minimum number of frames required to determine the linear polarization and its position angle is 3 (spanning more than 90$^\circ$ in position angle). By effectively oversampling the polarization curve (viz. the variation of detected signal with polarizer rotation angle) one can at least hope to average out shorter term variations in atmospheric transmission in order to improve the quality of the polarization measurement. Expressed in terms of the Stokes parameters I, Q and U (see e.g. Azzam & Bashara 1987), I depends on the total signal whilst Q and U depend on the difference in signals between images taken at polarizer angles of 0, 90, 45 and 135$^\circ$.Then the linear polarization p is given by, $p(\%) = 100 \times \sqrt(q^2 + u^2)$ where q=Q/I and u=U/I. The position angle of linear polarization is, $\theta(^\circ) = 28.648 \times {\rm tan}^{-1}(U/Q)$    (Serkowski 1962). Determining the polarization from $\geq$ twice as many images as necessary leads to improvement in polarization accuracy provided that any photometric variations are on timescales different from the exposure time of individual images at each polarizer angle. The worst case scenario is when photometric variations occur on a timescale similar to the exposure times, so that the measured difference signals vary wildy - the polarization determined by fitting a cosine curve then approaches zero. The chosen exposure times per polarizer angle were in the range 1 to 50 s depending on the source brightness (see Table 1). Observing a polarized source with the polarizer at 0 and 180$^\circ$ polarizer positions should give the same detected counts and is therefore a direct way to monitor the photometric variations during the observational sequence. Column 8 of Table 1 lists half the difference (in percentage) between integrated counts in the star profile for the 0 and 180$^\circ$ images (i.e. rms on the mean of the 0 and 180$^\circ$ signal values). For R Monocerotis, the semi-stellar peak of the reflection nebula NGC 2261, the aperture covers the central extended source (full extent 8''), whilst for OH 0739-14, a reflection nebula around an embedded young star, an area 10'' in size was used for the statistics.