2 Observations and data reduction

2.1 Time-series spectroscopy

We obtained time resolved longslit spectra for the sdBV PG 1605+072 during one observing run. The observations were carried out with the 3.5-m telescope at the DSAZ, Calar Alto, Spain and the TWIN spectrograph (two SITe-CCDs with 2048 $\times$ 800 pixels à 15 $\mu$m). The time allocation committee awarded two nights on May 14 and 15, 2001. Because of rain during the first night data were only obtained during the second night (5 h 27 m of data). The passage of clouds and variable seeing affected the quality of the data considerably. In order to achieve an adequate time resolution observations were carried out in trailing mode which means that the telescope tracking is tuned such that the star moves slowly along the slit in N-S direction. The drift velocity was chosen at 270 $^{\prime\prime}$ per hour so that a time resolution of 15 s per pixel resulted. The time resolution is dependent on the seeing conditions and thus varied during the night. The slit width was 1.5 $^{\prime\prime}$. We chose the gratings T05 and T06 (36 ${\rm\AA}$/mm) corresponding to a spectral resolution of $\approx$1 ${\rm\AA}$. The spectral ranges from 3860-4960 ${\rm\AA}$ in the blue and from 5880-6920 ${\rm\AA}$ in the red were covered. Moreover, to decrease the readout time and the noise level the data were binned (1$\times$2).

2.2 Multi-band photometry

Simultaneously with the time resolved spectroscopy of the target star its light curve was measured in four wavelength bands at the 2.2-m telescope at Calar Alto. At this telescope five nights between May 14 and 18, 2001 were awarded. No data could be taken during the first night and from the middle of the fourth night until the end of the run due to bad weather conditions. 12 h 27 m of pure observation time resulted in 880 measurements. The telescope was equipped with the new multi-band photometer BUSCA (Reif et al. 1999; Cordes et al. in prep.) which is able to measure in four photometric bands simultaneously. Three beamsplitters split the incoming light into four beams that feed four 4k$\times$4k CCDs. Only Strømgren filters were available for BUSCA. Use of these filters would have further reduced the S/N. Therefore, no filters were inserted and the beamsplitters served as non-overlapping broad band filters. The transmission curves of the four bands are shown in Fig. 1. These curves convoluted with the quantum efficiency curves of the CCDs provide the pass bands for our observations. The four wavelength bands are denoted by " $UV_{\rm {B}}$'', " $B_{\rm {B}}$'', " $R_{\rm {B}}$'' and " $NIR_{\rm {B}}$'' throughout the paper following the notation of Cordes et al. (in prep.). We chose an exposure time of 15 s which provides a good S/N and a reasonable coverage of the main frequencies. 2$\times$2 binning (resolution: 0.17 $^{\prime\prime}$ per unbinned pixel) and the windowing option (reading out 346$\times$400 pixels of the CCDs) helped to reduce the full cycle time to 51 s.

Figure 1: Transmission curves of the four BUSCA wavelength bands. The four ranges are denoted by " $UV_{\rm {B}}$'' (central wavelength: 3600 ${\rm\AA}$), " $B_{\rm {B}}$'' (4800 ${\rm\AA}$), " $R_{\rm {B}}$'' (6300 ${\rm\AA}$) and " $NIR_{\rm {B}}$'' (8000 ${\rm\AA}$) although these are not comparable to the classical Johnson UBVRI system.

The brightness variations of PG 1605+072 were detected by relative photometry. For this purpose two stars of similar brightnesses in the R band ( R₁=13.2 and R₂=13.4) next to our target star were used.

2.3 Data reduction

Data reduction was done using the IDL based software SPEX (Long slit SPectrum EXtraction package), and TRIPP (Time Resolved Imaging Photometry Package, see Schuh et al. 1999). TRIPP is based on the CCD photometry routines written by R. D. Geckeler (1998) and performs aperture photometry. The most important step is the determination of the relative flux of the target star with one or more stars as comparison objects. The detection of variations of the order of a few mmag is only possible when comparison stars and sky background are recorded simultaneously (especially under comparatively poor conditions).

SPEX is designed for the reduction of long slit spectra. The reduction comprises of the standard procedures flatfielding, biasing, cosmic ray event extraction, and 2-D wavelength calibration. The latter is very important because we extract spectra from single CCD rows in order to establish the time-series. Therefore, we have to ensure that the position of the comparison lines perpendicular to the dispersion direction doesn't change. In fact, these positions vary and thus we correct for that by fitting a polynomial. After the calibration of one CCD row the resulting dispersion relation is applied to every single CCD row.

One crucial point was the calibration of the time axis. As a consequence of the trailing mode observation the seeing disk moves slowly across the CCD and therefore the signal smears over the trailing direction. The initial point of time of the measurement and the cycle time were determined by fitting Gaussian profiles to the rising flank of the star's signal. Finally, radial velocities were derived by fitting Lorentz profiles to the Balmer lines ${\rm H}_{\alpha}$- ${\rm H}_{8}$ as well as to He I 4471 ${\rm\AA}$ and He II 4686 ${\rm\AA}$.