Online material

Appendix A: Details on the observations and data reduction

A.1. ADONIS NIR imaging

The ESO ADaptive Optics Near Infrared System (ADONIS; Beuzit et al. 1997a) was available on the 3.6 m telescope at ESO La Silla until September 2002. Typical Strehl ratios were around 0.1 in J band and 0.3 in H band. ADONIS was coupled with the near-IR camera SHARPII+ which operates in the J to K band. Our data were acquired on June 8, 2000 in the H and SK (short K) filter with total exposure times up to 600 s. We attached a fully opaque coronographic mask in front of the Lyot (pupil) stop to reject the peak of the PSF for the brightest source, to increase the integration time and sensitivity in order to reveal fainter structures. Details on the coronograph and its performance are given in Beuzit et al. (1997b). For HD 155448, a mask of 1.4′′ and a lens scale with 0.1′′/pix were used. We co-added ten exposures of 60 s in both filters. The mask’s fixed position meant that no jitter offsets were possible. Data cube clean-up, dark and flatfield correction, as well as sky subtraction were performed in standard ways. Figure 1 shows an ADONIS H band exposure. No PSF subtraction was applied in this case.

Despite the high sensitivity of the ADONIS data, no reliable photometry could be obtained. There were problems both with the calibrator stars and with the camera’s ADU conversion factor during this night, and weather conditions were not very stable.

A.2. EFOSC2 optical imaging and spectroscopy

A.2.1. Photometry

Broadband photometry in the B, V, and R filters was obtained on February 26, 2006 with EFOSC2 (Buzzoni et al. 1984) at the 3.6 m telescope in La Silla. These imaging data have a binning of 1 (i.e. 0.157′′/pix). To better resolve the emission between the components, the data were re-sampled to 0.079′′/pix. Eight frames of 1 s integration time were co-added for each filter. Despite the short exposures, the A component was saturated in the V and R filters. The B2 source is not resolved from B1 in the optical filters. Components A, B, and C have a flux error of 0.1 mag due to the problem in resolving their fluxes. The aperture photometry of source C does not include the arc-shaped emission region. Landolt standard fields (Landolt 1992) for zero points were observed just prior to the science data. In addition, we used SExtractor to extract the photometry and colour determination of all  ~2000 sources in the EFOSC2 field (about 5.5′ ×  5.5′), in order to check whether more objects belong to the HD 155448 system. An R band EFOSC2 image is shown in Fig. 2.

A.2.2. Spectroscopy

EFOSC2 spectra were obtained on March 1 and April 20, 2006, as well as on August 19, 2007 and August 12, 2008. The spectra obtained in 2006 were taken with grism 11 and a 1.0″ slit, covering 3380–7520 Å at a dispersion of 2.04 Å/pix and a mean resolution R ~ 400. Slit orientations were applied as shown in Fig. 1, to avoid a contamination by the neighbouring components as much as possible and to resolve the C source and the arc individually. In 2007, we re-observed the C component with a 1.0″ slit and EFOSC2 grisms 5 (5200–9350 Å, 2.06 Å/pix, R ~ 300) and 18 (4700–6770 Å, 1.0 Å/pix, R ~ 600), to obtain higher resolution and a redder spectral coverage. The slit orientation during the 2007 campaign was east-west, to cover only the C component (“Slit C only” in Fig. 1). When the high-resolution holographic grisms 19 (4441–5114 Å, 0.34 Å/pix, R ~ 3000) and 20 (6047–7147 Å, 0.55 Å/pix, R ~ 2500) became available, we observed the system again with a 0.5″ slit in orientations as shown in Fig. 1 and Table 2. Spectrophotometric standard stars were taken from the list of Hamuy (1992, 1994) and observed with a wider 5.0″ slit. Bias and flats were taken in a standard way. The wavelength calibration was done by correlating to a HeAr lamp spectrum. The calibrated final spectra were averaged for consecutive exposures in each setting and the continuum normalised.

Further data processing was performed for the R ~ 3000 spectra. The 3σ error in the wavelength calibration obtained is  ~4 km s^-1 (σ = 1.2 km s^-1). To produce the position velocity diagrams discussed in Sect. 2.3, the two-dimensional wavelength calibrated spectra were corrected for the curvature induced by the instrument along the spatial direction and the trace in the PSF along the dispersion direction. The curvature along the spatial direction was determined by fitting a second degree polynomial to the centre position (determined by a Gaussian fit) of the bright sky [O i] emission line near 6300 Å along the spatial axis. Then the spectra at each spatial position was shifted such that the centre of the sky [O i] emission is always at the same pixel. The reference pixel was the pixel in which the [O i] sky line crosses the peak of the star’s continuum. The PSF trace was determined by fitting Gaussians to the PSF at several positions in the continuum and then fitting a second degree polynomial to the PSF centre positions. The spectra then were shifted such that the centre of the PSF is always located at the same pixel. The reference pixel is the median of the PSF centre positions.

A.3. SOFI NIR imaging

We acquired JHK broadband images with SOFI (Moorwood et al. 1998) at the La Silla NTT during May 7, 2004. SOFI’s small field scale with 0.144′′/pix was used. For each filter, ten exposures with random jitter box width of 20″ were coadded. The total integration time in each passband amounts to 100 s. Darks and dome flats were taken in a standard way. Unfortunately, it was not possible to get separate photometry for the B1 and B2 components, owing to the low spatial resolution. The photometry of the C component is derived without the arc, placing the aperture – as closely as possible – only around the star. The derived magnitudes of source C might be slightly fainter than they really are, since with circular apertures it cannot be avoided that the very closely located arc contributes partly to the 2 pixel thin sky annulus, thus resulting in a higher value for the sky. This uncertainty is considered in the corresponding magnitude errors. Zero points (ZPs) were derived from NICMOS photometric standard stars, which were observed two hours after HD 155448. We estimate a final photometric accuracy of 0.1 mag, as we do not accurately know the quality of sky transparency during these observations that were obtained from the ESO science archive. The estimate of 0.1 mag is derived from the maximal difference in our ZPs to those of a photometric SOFI night.

A.4. NACO NIR imaging

Additional adaptive optics data were obtained with NAOS-CONICA (NACO; Lenzen et al. 2003; Rousset et al. 2003) at the Paranal Observatory in service mode during ESO period 75 (i.e. between April and September 2005). We acquired images in narrow band filters (NB) at the central wavelengths 1.64 μm, 2.12 μm, 3.74 μm, and 4.05 μm, with a spatial resolution of 0.027′′/pix and total integration times up to 435 s. Dark, flatfield, and sky correction were made in the standard way. Ten individual exposures for each filter, with a jitter box width of 10″, were shifted and co-added.

We think that some problems must have occurred during these observations, because we observed strong flux losses when comparing the derived photometry to the expected values, as could already be recognised from the raw data. We immediately see that the exposures are not very sensitive, so that even the arc northeast of the C component is not seen in any of the NACO images. Nevertheless, we could derive useful relative photometry and astrometry for the narrow band filters NB_1.64 and NB_2.12. Unfortunately, no standard star was observed for this service mode programme. Therefore, we first derived relative photometry and then scaled it to the 2MASS H and K band flux of the D component, since this component likely has the most accurate 2MASS flux, which is recognisable from the flags in the 2MASS catalogues, while its larger distance to the neighbouring stars will avoid contamination. The resulting NACO photometry matches that of SOFI within the error ranges.

A.5. VISIR MIR imaging and spectroscopy

A.5.1. Photometry

We observed the HD 155448 system between 8 and 19 μm with the mid-IR instrument VISIR (Lagage et al. 2004) at Paranal in service mode in period 75. In the N band, dual imaging around the wavelength of PAH emission was performed with the VISIR filters PAH2 (11.25 μm), and PAH2_ref2 (11.88 μm) as the nearest continuum reference. The Q band exposures were taken with the Q2 filter (18.72 μm), which – for VISIR – has the best Q band sensitivity. The spatial resolution at both wavelengths is 0.075′′/pix in the VISIR small field. All mid-IR exposures were performed with a standard chopping and nodding technique with a throw of 10′′. Mid-IR standard stars for flux calibration were selected by the observatory staff from a list of mid-IR standard stars by Cohen (1998), and were observed close in time and airmass to the science observations.

Data cube chop and nod co-addition was done in the standard ways. Where necessary, we corrected for known artifacts (in the form of stripes) of the VISIR detector. The co-added exposure time for each PAH observation amounts to 1085 s. For the final image, we selected the five best out of six observations, spread over several months, resulting in 5425 s (1.5 h) total integration. The resulting PAH on/off images have the same exposure time and S/N. Both images were re-centred with an accuracy of 0.1 pix before subtraction. In the Q band, each co-added observation sums to 1987 s. For the final image, the four best observations (out of five) were selected, resulting in a total exposure time of 7948 s (2.2 h).

The photometric accuracy is limited by the following uncertainties: (1) uncertainty in the aperture, which is negligible in our case with errors smaller than 5 mJy; (2) uncertainty in background subtraction: by co-addition of many exposures this influence got also strongly suppressed; and (3) uncertainty in the flux calibration. The latter may be significant, because the atmospheric transmission often varied in the short time between observation of the science target and the consecutively observed standard star. Since several calibrated images are co-added, this effect may be neutralised, but nevertheless we estimate the photometric accuracy to be only within 10%.

A.5.2. Spectroscopy

Low-resolution N band spectra (R ~ 350 at 10 μm) were obtained between 8 and 13 μm with a 1.0′′ slit. We aimed at having both the signal of the star and its arc-shaped circumstellar matter in the slit, so we rotated the slit to an orientation of 50° position angle (cf. Figs. 3 and 8). Our three VISIR spectral settings, centred at 8.8 μm, 11.4 μm, and 12.4 μm, cover approximately the wavelength ranges 8.0–9.4 μm and 10.4–13.1 μm, after removing the outer range of each setting.

The spectroscopic frames were debiased and checked for a correct alignment of the spectrum with the x-axis of the frame. Two peaks are distinguishable in the spectral profile: the position of the star and that of the circumstellar arc. We extracted the signal separately for both regions, with “a neutral zone” left in between, and coadded the signal of corresponding nod positions. In addition, data from various observations between June and September 2005 were averaged. An initial ad-hoc wavelength calibration, derived from the central wavelength of each spectral setting and the spectral resolution per pixel, was refined by cross-correlating with atmospheric lines. The final error in wavelength accuracy δλ/λ is  <10^-3.

Mid-IR standard stars for telluric correction were selected by the observatory staff and executed immediately before or after the science data. Flux was calibrated in the standard way by dividing the target through the calibrator spectrum and multiplying the result with a profile of the standard star taken from Cohen (1998). Due to the varying atmospheric transparency in the time between observation of the science target and the standard star (typically less than 1 h), the flux calibration is not accurate so we need to cross-correlate the spectral flux to our aperture photometry. Moreover, we have wavelength dependent slit losses: while we covered the entire stellar neighbourhood (silicate-dominated), the emission of the outer part of the circumstellar arc (PAH-dominated) was not included in the 1.0″ slit. Therefore, we scaled the 11.4 μm silicate spectrum of the stellar vicinity to the aperture photometry of 0.44 Jy at 11.88 μm (cf. Table 4), motivated by the dominance of this spectrum by silicates. Similarly, we scaled the 11.4 μm spectrum of the PAH-dominated circumstellar arc to the photometry for the arc of 0.41 Jy at 11.25 μm, which was obtained as the difference of the total flux and the flux in the vicinity of the star (cf. Table 4). We corrected most of the slit losses in the PAH emitting wing by this approach. Finally, the 12.4 μm settings, which overlap with the 11.4 μm ones, were scaled to match the flux of the latter. We note that this flux calibration is not a unique solution, but the best effort possible for this data. We estimate a possible deviation from the actual flux within 10%. More difficult is the adjustment of the 8.8 μm settings, which do not overlap with the more longward settings and for which we do not have aperture photometry at shorter wavelengths. In the absence of other calibration options, the 8.8 μm settings are adjusted to match the longward spectra. The approximate error in flux calibration for the 8.8 μm silicate spectrum is another 10%, while the absolute intensity of the 8.6 μm PAH band remains uncertain.

A.6. Spitzer MIR spectroscopy

HD 155448 was observed with the infrared spectrograph (IRS, Houck et al. 2004) onboard the Spitzer Space Telescope on March 22, 2005 (PID: 3470, PI: Jeroen Bouwman). The source was measured using Short Low (5.2–14.5 μm), Short High (9.9–19.5 μm), and Long High (18.7–37.2 μm) modules. The integration time was 30 s for the Short High module and six seconds for the other modules, respectively, and at least two cycles were used for redundancy. A high-accuracy PCRS peak-up was used to acquire the target in the spectrograph slit. The spectra are based on the droopres and rsc products, processed through the S18.7.0 version of the Spitzer data pipeline for the low- and high-resolution modules, respectively. For the details of the data reduction procedure we refer to Juhász et al. (2010) and Acke et al. (2010). The brief description of the data reduction steps is as follows. For the Short Low spectrum the associated pairs of imaged spectra were subtracted in order to correct for the background emission, stray-light contamination and anomalous dark currents. For the high-resolution spectra the background has been removed by fitting a local continuum underneath the source profile.

Pixels flagged by the data pipeline as being “bad” were replaced with a value interpolated from an 8-pixel perimeter surrounding the flagged pixel. In the case of the Short Low module, the spectra were extracted using a 6.0-pixel fixed-width aperture in the spatial dimension. For the high-resolution modules, spectral extraction was done by fitting the source profile with the known PSF in the spectral images. Low-level fringing at wavelengths  >20 μm was removed using the irsfringe package (Lahuis & Boogert 2003). The spectra were calibrated using a spectral response function derived from IRS spectra and MARCS stellar models for a suite of calibrators provided by the Spitzer Science Centre. To remove any effect of pointing offsets, we matched orders based on the point spread function of the IRS instrument, correcting for possible flux losses.

© ESO, 2011