FAR-ULTRAVIOLET SENSITIVITY OF THE COSMIC ORIGINS SPECTROGRAPH

From the inception of the Cosmic Origins Spectrograph (COS) (Morse et al. 1998; Green et al. 2003) as an instrument for the Hubble Space Telescope (HST), it was appreciated that the combination of a single bounce reflection grating with the large geometric collecting area (∼40,000 cm²) and the ∼15% reflectivity at 1000 Å of the magnesium fluoride over-coated aluminum (MgF₂/Al) mirrors (Hunter et al. 1971) could provide useful sensitivity in the far-UV bandpass between the rather sharp MgF₂/Al discontinuity at ∼1150 Å and the Lyman edge at 912 Å. The possibility that COS could work below 1150 Å was intriguing because of the potential for new science. A low spectral resolution far-UV channel could be used for new studies of the steeply increasing populations of faint galaxies and quasars not detectable with the higher spectral resolution channels offered by the Far-Ultraviolet Spectroscopic Explorer (FUSE). Moreover, FUSE is no longer operational and as such the community has lost its window on a set of critical spectral diagnostics, such as for H₂ and O vi, that are only available below 1150 Å.

The COS instrument team found that the windowless detector design and the low resolution G140L grating R(≡λ/Δλ)∼ 2000 lent itself easily to the inclusion of a non-optimized far-UV channel in detector Segment B, spanning ≈100–1100 Å. Its expected effective area was anticipated to be about 100–200 times lower (Froning & Green 2009) than that of COS longward of 1150 Å. However, this estimate was uncertain. It was known that upon return to Earth the WFPC-1 pickoff mirror had been severely contaminated during its time in orbit with a normal incidence reflection of ≈75% at 1500 Å, falling to ∼1% at 1216 (Tveekrem et al. 1996). Consequently, the use of the G140L Segment B was unsupported for the HST Cycle 17 round of proposals, because of the unknown level of contamination that the HST Optical Telescope Assembly (OTA) had suffered in the nearly 20 years since its launch.

Here, we present the first evidence that any OTA mirror contamination in the far-UV is inconsequential and that in fact the sensitivity below 1000 Å rivals that of the silicon carbide channels on FUSE (Dixon & Kruk 2009). We discuss some new science opportunities enabled by this new capability and conclude that "Lyman limited" instrumentation is a feasible option for future large aperture UV telescopes.

On 2009 September 04, the hot DA white dwarf WD 0320-539 (GSC 08493-00891, T_eff = 32.9 K, log g = 7.89, V = 14.9; Vennes et al. 2006) was observed in the G140L channel of COS as part of an external detector flat-field program, number 11491. The data were acquired in three focal plane positions (FP-POS) observations, each 640 s long, yielding a total on target time of 1920 s. The time tagged photon tables have file prefixes of la9h02nsq, la9h02nsu, la9h02nsz. We report here only on the Segment B spectrum.

Standard pipeline processes (Kaiser et al. 2008) were used to correct geometric distortion in the raw image photon tables (rawx, rawy), producing corrected time tagged photon tables (xcorr, ycorr). The pipeline feature that coaligns the three separate FP-POS spectra into individual "*_corrtag_b.fits" two-dimensional images was found to produce spectral and spatial smearing in the, so-called, (xfull, yfull) photon lists. Consequently, custom software was used to coalign, extract, and combine the three FP-POS spectra into a single one-dimensional spectrum using the (xcorr, ycorr) photon lists. The individual spectra exhibited rather prominent correlated fixed pattern from the detector over the last 1000 pixels, where the signal was highest, with rms fixed pattern variation ≈8%. To remove this variation, the extracted spectra were coadded without coaligning the spectra and normalized with a low-order spline function, to produce a flat-field template. The template was divided into the individual extractions before coaligning on the stellar features and coadding.

The spectra for WD 0320-539 were extracted from the two-dimensional images using a rectangular aperture 30 pixels high. Counts with pulse heights <4 and >30 were eliminated. A detector background region, located at the edge of the detector ∼160 pixels below the object region, was similarly extracted. The G140L Segment B wavelength scale is 0.08 Å pixel⁻¹, and there are ≈6 pixels per spectral resolution element. Spectral pixels are 6 μm wide and cross-dispersion pixels are 24 μm high. Further information on the COS instrument can be found in J. C. Green et al. (2010, in preparation) or in the COS instrument handbook for Cycle 18 (Dixon 2010).

This object has also been observed several times by FUSE at a resolving power of R∼ 20,000. The spectrum, provided through the Multimission Archive at the Space Telescope Science Institute (MAST), incorporates the final FUSE calibration (Dixon & Kruk 2009) and will serve as our absolute calibration reference. The specific FUSE files used in this work are "D0230201103*ttagfcal.fit" where the asterisk is a wildcard for the channel labels: 1bsci, 1alif, 2alif.

In Figure 1, we show the count pixel⁻¹ spectrum on a linear scale plotted from 900 to 1050 Å, using a linear wavelength solution. The progression of the pressure broadened Lyman series characteristic of a high-gravity hot white dwarf is evident. The FUSE spectrum, rebinned to the G140L Segment B wavelength grid and scaled by a constant factor of 8 × 10¹³ counts (erg cm⁻² s⁻¹ Å⁻¹)⁻¹, is overplotted in blue. The count spectrum can be faithfully converted to the stellar flux for WD 0320-539 over most of the displayed bandpass using this constant factor, demonstrating the effective area is essentially constant in this region. The overall agreement is excellent.

Zoom In Zoom Out Reset image size

The effective area curve, shown on a semi-logarithmic scale in Figure 2, was derived simply by dividing the FUSE spectrum into the COS count spectrum and applying the appropriate time (t = 1920 s), energy (hc/λ), and dispersion (dλ = 0.08 Å pixel⁻¹) scalings. A spline fit based on 13 points averaged over 3 Å intervals, at wavelengths chosen to avoid the Lyman series line cores, was employed to estimate the effective area. The fit intervals are marked in red on Figure 1. Points from the spline fit and associated errors, sampled on 20 Å intervals starting at 920 Å  are presented in Table 1. For comparison, we also show the effective area curves for the SiC1b, LiF1a, and LiF2a channels of FUSE as dot-dashed lines in Figure 2.

Zoom In Zoom Out Reset image size

We can estimate the amount of contamination that the OTA has suffered by comparing a model of the effective area to these data. Recall that the effective area is the product of the telescope geometric area, the reflectivity of each telescope mirror, the grating absolute efficiency, and the detector efficiency. We use the component efficiencies displayed in Figure 3. The triangles are the measured MgF₂/Al reflectance (250 Å of MgF₂) from Hunter et al. (1971). The squares are the COS Segment B detected quantum efficiencies measured by the instrument team (Vallerga et al. 2001) for the CsI-coated detector. The points from 1150 Å and longward are preflight measurements from the flight detector. The three points shortward of 1150 Å were obtained from a similar detector with long wavelength response similar to that of the flight detector. The over plotted statistical error bars are small. The asterisks are absolute efficiency measurements for the G140L flight grating made by the instrument team (Osterman et al. 2002), except for the 1026 Å point which was obtained from the flight spare. The over plotted statistical error bars are also small. G140L absolute efficiency measurements were not available for the full far-UV wavelength range, so we used the available measurements to constrain a model computed as the product of the MgF₂/Al reflectance and a groove efficiency for a trapezoidal-shaped profile (McCandliss et al. 2001). The result is shown as the solid line that has reasonable agreement with the asterisks.

Zoom In Zoom Out Reset image size

Linear interpolation was used to represent the reflectance, grating efficiency, and quantum efficiency (QE) at wavelengths other than the measured points for computing the modeled effective area shown as the dashed line on Figure 2. Multiplying our model by a factor of ∼0.6 produces a reasonable match to the G140L Segment B effective area at 1000 Å. If we assume that the OTA carries the bulk of the disagreement and that the original reflectivity was similar to that of Hunter et al. (1971), then after accounting for the two reflections, we conclude the individual mirrors of the OTA retain ≈80% of their original efficiency.

With the effective area in hand, we are now able to determine the limiting background for the G140L Segment B channel. The G140L has a resolving power of R∼ 2000, which is ∼10 times lower than the FUSE resolution. We expect that the background limit should be correspondingly lower, provided the effective area, detector dark rate, and spectral extraction heights are similar to that realized by FUSE.

A background spectrum was extracted parallel to the spectral region at the bottom edge of the detector, so as to avoid as much as possible light scattered in the cross-dispersion direction. The extraction region was 30 × 3421 pixels², which corresponds to an area on the detector of 0.72 × 20.5 mm². The spectrum, converted to flux and rebinned to a 0.48 Å wide pixel, is shown as the black line in the top panel of Figure 4. The purple histogram is the same spectrum rebinned to a 2.4 Å pixel. The mean dark count rate per cm² from the three spectra for the extracted region was ≈2.5 counts s⁻¹ cm⁻², which is higher than the nominal rate for this detector of ∼1.5 counts s⁻¹ cm⁻² (Dixon 2010). We note the first two spectra had higher dark rates (2.7 counts s⁻¹ cm⁻²) than the last one (2 counts s⁻¹ cm⁻²), which was likely due to a transition from orbital day to night. The mean rate corresponds to 1.1 × 10⁻⁴ counts s⁻¹ pixel⁻¹. Applying the appropriate energy and wavelength binning per pixel to this number, and dividing by the effective area we derive the solid red line in the upper panel of Figure 4, which is in good agreement with the data.

Zoom In Zoom Out Reset image size

In the lower panel, we compare the G140L Segment B modeled background with that for FUSE (dot-dashed lines). The FUSE backgrounds were derived, following the method described by McCandliss (2003), where the height of the extraction window varies as a function of wavelength to follow the spectral astigmatism variation. For comparison, we overplot (solid blue line) the spectrum of the He ii Lyα forest toward HE 2347-4342 (Zheng et al. 2004) as observed by FUSE rebinned to 2.4 Å, along with the estimated continuum for this object. We note FUSE invested 431Ks of night time observation to this object.

The effective area of COS at wavelengths below 1000 Å is comparable to the effective area of FUSE over the same bandpass and it exceeds that of FUSE in the 1090–1180 bandpass (Kruk et al. 2009). At present, the error bars at the short wavelength end are somewhat uncertain because of the monotonically decreasing count rate in our calibration object, caused by the overlap in the converging pressure broadened Lyman series. This situation will improve as further calibrations are acquired. Aside from a small window around 1020 Å, the COS G140L Segment B background limit is smaller than that of FUSE, meaning that COS can observe fainter targets in a significantly shorter amount of time longward of 1030 Å and shortward of 1000 Å, albeit at a lower spectral resolution. The combination of reasonable effective area with low detector background will allow COS to pursue science left behind by FUSE when the planned low spectral resolution channel was descoped. We briefly discuss some of these science opportunities here.

For example, analysis of He ii "Gunn–Peterson" absorption in the intergalactic medium (IGM) carried out by FUSE was conducted primarily on two quasar lines of sight, HE 2347-4342 and HS 1700+6416 (Kriss et al. 2001; Shull et al. 2004; Zheng et al. 2004; Fechner et al. 2006). The FUSE observations in combination with H i absorption line studies from the ground demonstrated that fine-grained fluctuations in the metagalactic ionizing background are associated with source hardness, IGM density voids and peaks. However, there is a critical need for multiple lines of sight to better constrain the cosmic variance in the evolution of the IGM density and in the redshift associated with the initiation of the He ii reionization epoch near z_r (He ii) = 2.8 ± 0.2 (Shull 2006).

The higher sensitivity at low flux will also be beneficial to programs seeking to detect and quantify the fraction of Lyman continuum photons that escape from star-forming galaxies at low redshift. Low redshift observations are essential (McCandliss et al. 2008) to further our understanding of the physical processes responsible for the H i reionization epoch near z_r (H i) ∼ 6 thought to be caused by Lyman continuum photons escaping from the first star-forming galaxies (Yan & Windhorst 2004).

COS will be used to probe clouds in the interstellar medium with total extinction A_V > 3, where the long sought after transition from translucent to dense medium is expected to occur (Rachford et al. 2009). We would expect such sight lines to have total hydrogen columns ∼6 × 10²¹ cm², assuming the standard conversion (Bohlin et al. 1978), which by definition would have molecular fraction (f(H₂) ≡ 2N(H₂)/(2N(H₂) + N(H))) near 1. One of the problems that these observing programs faced with the demise of FUSE and the uncertain COS far-UV response was the lack of guaranteed access to the dominate cloud species, H₂, which has ground-state absorption bands in the far-UV shortward of ∼1110 Å. This is no longer an issue, as the effective area of COS G140L Segment B near 1100 Å is ∼200 cm². H₂ absorption will be dominated by the J = 0, 1 rotational levels and even for molecular fractions as low as 0.01, these lines will be broad and easily detectable at the expected resolving power of R = 2000, as they will be on the damping part of the curve of growth (McCandliss 2003). We note that the Hopkins Ultraviolet Telescope, which had a resolving power of R∼ 500, could easily detect H₂ columns to 10¹⁹ cm² (McCandliss et al. 1993). This will enable, in conjunction with CO measurements made using the high and/or medium resolution modes of the Space Telescope Imaging Spectrograph or COS, direct studies of the CO/H₂ ratio (Burgh et al. 2007) in dense regimes more typical of those probed by CO observations at radio wavelengths and could lead to more accurate estimates of molecular gas mass in star-forming galaxies.

We have demonstrated that there is significant response in the G140L Segment B channel of COS shortward of the MgF₂/Al discontinuity near 1150 Å extending all the way to the Lyman edge. This low resolution far-UV capability is a unique instrumental mode that has never been widely available to the astronomical community and as such has great potential for scientific discovery.

The COS far-UV demonstration has implications for future facility class missions. We note that no effort was made to optimize the far-UV response of this channel. The use of a SiC-coated grating would provide a factor of 2 gain in efficiency, albeit at the expense of the long wavelength response. An additional factor of 2 would result through the use of a dual-order spectrograph design (Lupu et al. 2008) or a blazed ion etched grating. We estimate that the efficiency of a properly optimized spectrograph on a general purpose 10 meter telescope, having a two bounce OTA with standard MgF₂/Al mirrors, has the potential to achieve an effective area of ∼500–1000 cm² in the far-UV, easily breaking the desired factor-of-10 threshold commonly used to assess the discovery potential of new observing modes (Moos et al. 2004). The COS observations presented here prove that such an approach is a viable option for large facility class missions and suggests that including a windowless far-UV channel below 1150 Å on future missions should be seriously considered.

The authors acknowledge useful and encouraging discussions with the following people: Cynthia S. Froning, J. Michael Shull, John T. Stocke, Theodore P. Snow, S. Alan Stern, Derck L. Massa, Oswald H. W. Siegmund, John V. Vallerga, Brian Fleming, Roxana E. Lupu, Eric B. Burgh, David J. Sahnow, Wei Zheng, Jeffrey W. Kruk, W. Van Dixon, William P. Blair, Kenneth Sembach, B-G Andersson, Paul D. Feldman, and H. Warren Moos. This work has been supported by NASA contract NAS5-98043 to the University of Colorado.

Facilities: HST (COS) - Hubble Space Telescope satellite.