THE COSMIC ORIGINS SPECTROGRAPH

In the ultraviolet, each reflective surface has at best 80% reflectivity. Therefore the FUV channel was designed to utilize a single optical element: one grating. This grating would have to disperse the light, re-image the diverging telescope beam, correct the spherical aberration of the Hubble Optical Telescope Assembly (OTA), and correct the inherent aberrations of the quasi-Rowland circle mount. The design of these gratings is discussed in Section 2.4.

Previous HST corrector designs for STIS (Woodgate et al. 1998) and Advanced Camera for Surveys (ACS; Ford et al. 1996) utilized multiple optics to correct the spherical aberration content of the OTA by using a collimating mirror to reform the entrance pupil (or equally, an image of the primary mirror) and then a corrector/reimaging mirror to introduce a wavefront correction of equal and opposite magnitude to the pathlength errors in the HST primary mirror. This is necessary when correcting the aberrations over a field. In such an application (e.g., STIS) a slit can be introduced in the system at the corrected focus, and any post-slit spectrograph can utilize classic designs which image the spectrograph entrance slit. Such a design results in a large number of reflective surfaces. However, by designing COS as a point-source spectrograph and limiting the field of view to a few arc seconds, any point in the diverging beam behind the OTA focus is an effective pupil, and the wavefront correction can be implemented at any point behind the focus. This is the technique utilized in COS.

Most slitted spectrographs have significant throughput losses at the spectrograph entrance slit. Some modes of STIS have up to 50% losses at the slit (Bostroem et al. 2010). The entrance slit in most spectrographs is present to define the width of the imaged point source in the dispersion direction, to limit contamination from the sky and other astronomical sources, and to relax requirements on pointing/tracking so that the spectral resolution does not degrade if the telescope drifts on the sky. However, on HST, these reasons are less relevant. The excellent pointing stability of HST (typically less than 0007 jitter) eliminates the need to correct for sky drift. The spherical aberration of the HST OTA means that a point source does not form a well-defined image at any point without further correcting optics. Therefore, any slit utilized at the OTA focus in a classic spectrograph approach (where the final image is the image of the slit) will either have low transmission if the slit is small, or poor spectral resolution if the slit is wide. COS avoids both of these pitfalls by focusing the spectrograph not at the slit, but on the sky, so that spectral width of a monochromatic point source is only as large as the corrected HST line-spread function (LSF; Kriss 2011) plus any aberrations in the spectrograph. By proper design, these can be made small enough to meet the spectral resolution requirements.

As a consequence, the slit itself will be out of focus at the final spectral image plane, and the sky transmission curve as a function of off-axis angle will not have a sharp cutoff (see Figure 2). Effectively, COS is a slitless spectrograph. However, to minimize sky contribution and potential confusion with nearby point sources, an aperture is included at the minimum waist point in the HST OTA point-spread function (PSF). The minimum waist point is not the same as the minimum encircled energy location, but rather the point at which 100% of the light is included in the minimum diameter. The diameter of the COS aperture is ∼700 μm, which at the focal length of the HST OTA (57.6 m) corresponds to 25 in diameter. This slitless aspect of COS has implications for the observation of diffuse sources, both in the étendue (AΩ) of the instrument to diffuse emission, and to the spectral resolution and spatial resolution of COS in such an application. These implications are discussed in Section 6.6.

Zoom In Zoom Out Reset image size

Since the entrance aperture (effectively an out-of-focus field stop) is not at a location where the telescope forms a real image (due to the spherical aberration), the COS spectrograph does not image the entrance slit, but rather, forms an image of the sky at the spectral focal surface. (In a standard design where the entrance slit is located in a stigmatic real image plane, this distinction is irrelevant.) This is important to remember, as objects outside the field of view of the aperture will pass light through the aperture due to the spherical aberration, but the light will be re-imaged by COS at the appropriate sky location, not the centroid of the transmitted light in the aperture.

It should be noted that the HST OTA, while excellent, is not perfect, and has aberration content due to figure errors other than spherical aberration as well as scattered light from surface imperfections and dust. These cause uncorrectable contributions to the HST PSF which results in the COS slit passing slightly less than 100% of the original beam (approximately 98%) and slightly degrading the final PSF of the final imaged light. These effects on the LSF and the spectral resolution of COS are discussed in Section 6.3.

Transmitting optics can introduce losses in the system, particularly at the shortest FUV wavelengths. Additionally, the dominant source of dark noise in the systems utilizing windowed detectors is frequently UV photons produced in the windows through phosphorescence. This is true for the Multi-Anode Microchannel Array (MAMA) detectors utilized by the COS NUV channels and the STIS FUV and NUV channels. By eliminating windows in the detector systems, the throughput is enhanced and the noise level is greatly reduced, resulting in large gains in S/N, particularly for the faintest targets with F_λ < 1.0 × 10⁻¹⁴ erg cm⁻² s⁻¹ Å⁻¹ at 1200 Å.

Since the COS FUV channel has no transmitting optics at any point in the system, the front surface reflectivity of the HST OTA and FUV gratings provides measurable performance in the 900–1150 Å region (and very limited performance at even shorter wavelengths), a unique capability on HST. This performance is presented in detail in McCandliss et al. (2010). Because we could not be certain that the HST OTA would provide any performance at these wavelengths, the design of aberration control holography did not consider these wavelengths, and the focal surface match of the detector is not maintained at the extreme grating positions that must be employed to access these wavelengths. To have extended the design to these wavelengths would have necessitated reducing performance above 1150 Å, and this seemed unwise given the possibility that the OTA might provide no reflectivity at these wavelengths. Therefore, the spectral resolution becomes substantially degraded at in these sub-1150 Å modes, to a level of approximately R = 2000. Efforts are ongoing to optimize the resolution and efficiency in the sub-1150 Å modes. However, since the Far-Ultraviolet Spectroscopic Explorer (FUSE; Moos et al. 2000) is no longer operational, COS provides the only spectroscopic access to these wavelengths at the moment.

The traditional tradeoff between holographic gratings and mechanically ruled gratings is normally considered to be one of high groove efficiency (mechanical) versus low scatter (holographic). However, proprietary techniques now exist that allow the introduction of a blazed groove profile on holographically ruled gratings. Physical measurements of a COS FUV grating groove profile are shown in Figure 3. This allows the simultaneous development of a high groove efficiency and low scatter grating. To achieve such a profile first requires the exposure of the holographic grating photoresist to a different depth than would be used for a sinusoidal grating, and then exposing the grating to an ion-etch beam at multiple angles, each for a different exposure time. Not all combinations of line density and blaze angle can be reliably implemented. The parameter space in optical design (line density, incoming angle, and outgoing angle) for which the appropriate blaze conditions can be physically implemented is constrained and not well defined except through simulations of the ion-etch process. To simultaneously achieve the appropriate spectral dispersion and have a manufacturable blaze angle, strict constraints are imposed on the available optical design parameter space into which the spectrograph must conform, and even then it is necessary to iterate with the manufacturer to find a reliable solution. Multiple test gratings were produced to optimize the flight production units, and it required 4–8 prototype runs to achieve optimal flight gratings.

Zoom In Zoom Out Reset image size

The use of a holographic grating also permits the introduction of aberration control in the recording process, allowing a single optic to maintain the performance over a significant bandpass. The actual design of the gratings is presented in Table 1.

Table 1. COS FUV Spectrograph Design Parameters

Channel Name	G130M	G160M	G140L
HST Secondary Vertex/Slit (z)	6414.4 mm	6414.4 mm	6414.4 mm
Slit Off Axis	90.49 mm	90.49 mm	90.49 mm
Slit/Grating	1652.57 mm	1652.57 mm	1652.57 mm
α	201	201	740745
β	86466	86466	−404595
α − β	114534	114534	114534
Grating/Detector	1541.25 mm	1541.25 mm	1541.25 mm
Detector Normal/Central Ray	904664	904664	904664
Groove density (l mm−1)	3800.	3093.3	480.
Radius	1652. mm	1652. mm	1613.87 mm
a4	1.45789E-9	1.45789E-9	1.33939E-9
a6	−4.85338E-15	−4.85338E-15	1.48854E-13
γ	−710	−625	100
δ	653512	385004	240722
rc	−4813.92 mm	−4363.6 mm	3674.09 mm
rd	5238.29 mm	4180.27 mm	3305.19 mm
Recording λ	4880 Å	4880 Å	4880 Å

Download table as:  ASCIITypeset image

The FUV G140L grating, which has much lower dispersion, was not ion-etched to create a triangular groove profile, but utilizes a laminar profile instead. While this produces acceptable performance, it is not as efficient as would be predicted with a triangular groove profile. Unfortunately, the level of ion-etching required to produce a 14 blaze angle would remove so much material that we could not guarantee the figure quality of the grating would be preserved. After the production of the flight, laminar profile gratings, a test grating was produced with 14 blaze angles in late 2001. However, it was never tested for figure quality, as the necessary HST-simulating, spherical aberration inducing optical feed system was needed at the end-to-end test level to support the scheduled launch date of 2004.

Essential to the observational efficiency of a spectroscopic instrument like COS is the availability of a large observational bandpass. For IGM studies, the actual wavelength at which the absorbers will appear is not known before the observation is made; therefore, complete coverage across the entire 1150–1775 Å bandpass in as few exposures as possible maximizes the net information gathering capability of the instrument in fixed observing time. Traditionally, a large bandpass is achieved through the use of an echelle spectrograph, such as on STIS. However, echelle spectrographs require multiple gratings and optics, with the subsequent loss of throughput, and mechanically ruled echelle gratings often exhibit significant scatter. (It has currently not been demonstrated that high-angle triangular blaze profiles can be achieved through the ion-etching of holographic gratings.) Therefore, it was decided to utilize a sufficiently large detector to obtain at least one half of the total bandpass in a single exposure in first order. Fortunately, large microchannel plate detectors have recently been developed for FUSE, and these served as the template for the COS detectors. The active area for the COS detectors is two segments, each 85 mm long × 10 mm high, curved to match the focal surface of the system, with a 9 mm gap between the segments. The readout system for each segment is a cross delay line, resulting in the production of 16K × 1K non-physical pixels, which do not arise from any physical boundaries, but from the digitization of a measured time interval. In this way, the COS detectors are very different from a classical CCD or IR focal plane array, or even the STIS MAMA detectors, whose pixel scale is established by a physical set of wires. For a much more complete discussion of the detector development and capabilities please see Vallerga et al. (2001).

Windowed detectors often utilize semi-transparent photocathodes applied to the back side of the windows. However, higher detection quantum efficiencies (DQE) can normally be obtained by the use of opaque photocathodes, which in the case of COS are applied directly to the surface of the microchannel plates. This has resulted in gain of up to four times the quantum efficiency for the photocathode compared to semi-transparent applications (Lampton et al. 1986; Vallerga et al. 2001). While the open-face design results in exposure of the photocathode to the environment of the spectrograph, and subsequent degradation of the efficiency, current measurements indicate a loss of approximately 5% per year from the photocathode, which is small compared to the significant gains realized through the opaque utilization.

The throughput on the FUV channels is excellent and exceeds the performance predictions in the proposal by approximately a factor of two. This is the result of higher grating efficiencies than anticipated and higher detector DQE than assumed in the proposal. The on-orbit effective area of COS is presented in Figure 4, along with the effective area of STIS in its most comparable mode.

Zoom In Zoom Out Reset image size

The on-orbit background for the FUV detector is approximately ∼1 count s⁻¹ cm⁻², consistent with pre-launch predictions (Sahnow et al. 2010). The origin of this background is approximately ∼0.5 counts s⁻¹ cm⁻² from internal radioactive decays in the microchannel plate glass, and ∼0.5 counts s⁻¹ cm⁻² from incident particles that the detector electronics are unable to discriminate from real photon events. This results in ∼0.8 background counts/resolution element in a nominal 2000 s (∼1 orbit) observation. Being photon counting devices, the COS detectors have zero read noise. Except in the vicinity (within ±2 Å) of a bright airglow line, the contribution from scattered light is even less than the dark background. This means that for brighter targets (F_λ > 1.0 × 10⁻¹³ erg cm⁻² s⁻¹ Å⁻¹) the FUV channels are signal limited and have an effective dark background of zero. For fainter targets, correction for background is necessary. However, the correction is much smaller than for previous HST spectrographs. This allows COS to reach reasonable S/N at unprecedented low flux levels. One of the goals stated in the proposal was to break through the 10⁻¹⁴ flux level (10⁻¹⁴ erg cm⁻² s⁻¹ Å⁻¹) which was considered the practical limit for observations with the existing and planned HST instruments. COS has measured line fluxes with equivalent continuum fluxes of 1.0 × 10⁻¹⁷ erg cm⁻² s⁻¹ Å⁻¹ (Froning et al. 2011) such as the black hole X-ray binary seen in Figure 5 and in the intergalactic He ii absorption troughs (Figure 12). Because of the extremely low background in the FUV channel, even fainter fluxes could be observed if the science justified the exposure times necessary to obtain the signal.

Zoom In Zoom Out Reset image size

The on-orbit background performance of the NUV MAMA is dominated by the phosphorescence of the detector window. Ground testing of the COS detector, compared to the STIS flight detector, under similar conditions (utilizing radioactive sources to stimulate the phosphorescence), indicated that the COS detector would have approximately 25% of the dark rate of the STIS detector. As the STIS NUV detector had a dark rate of 1200 counts s⁻¹ (over the entire detector) before failure, it was anticipated that the COS MAMA would have ∼300 counts s⁻¹ dark rate on orbit. To our surprise, immediately after turn on, the count rate was only 60 counts s⁻¹. However, that has been steadily rising on orbit and is now close to 300 counts s⁻¹. The STIS NUV MAMA, on the other hand, had a significantly enhanced background after repair, which has been steadily decreasing since re-commissioning. The rates are displayed in Figure 6. For a more detailed discussion, see Penton & Keyes (2010).

Zoom In Zoom Out Reset image size

The spectral resolving power of COS was originally designed to be R ⩾ 20,000 at all wavelengths, but the on-orbit performance is below this level. In the G185M (NUV) channel, the substitution of a lower density grating has resulted in the resolving power ranging from 16,000–22,000 across the bandpass, with the lowest resolving power at the shortest wavelengths. This reduction in spectral resolution was accepted as preferable to the significantly reduced throughput that would have resulted from using the original gratings. The resolving power of the NUV channels on orbit is presented in Figure 7 and is consistent with pre-launch expectations. The mid-frequency wavefront errors present in the Hubble primary and secondary mirrors which degrade the resolution in the FUV (see below) modes are present in the NUV modes, but at a reduced level which does not reduce the resolving power below specification.

Zoom In Zoom Out Reset image size

The FUV channels have lower resolving power on orbit than measured in thermal vacuum testing, due to limitations of the HST primary and secondary mirrors. Figure errors, primarily in the mid-frequency regime, result in some power moving from the core to the wings of any HST image. This effect is greater at shorter wavelengths and is seen in all HST instruments. It manifests as wings in optical images (at a significantly reduced level due to the longer wavelengths involved), and in the spectrographs when sufficiently wide entrance slits are employed. For example, this same effect is seen in STIS line profiles when the 2'' slit is employed. The convolution of these aberrations with the COS FUV grating and FUV detector responses results in a line profile that is broader than a pure Gaussian with an FWHM equating to 1 part in 20,000 of the wavelength. Additionally, the depth of an absorption line is less from the true LSF than it would be if the COS LSF was a pure Gaussian. Fortunately, the HST OTA performance models (Krist 1995), when convolved with the COS LSF, as measured in ground testing, match the on-orbit profile very well (Ghavamian et al. 2009). In practice, this means that as long as profile fitting is done with the appropriate on-orbit profiles, highly accurate parameters for the absorbing and/or emitting material can be derived. Except for weak, narrow lines, the ability to constrain the physical parameters (as compared to the ideal line profile capabilities) is limited by the S/N of the observation and not the degraded spectral resolving power.

As noted, this effect reduces the sensitivity of the instrument to detect weak (<10 mÅ), narrow features for a fixed S/N (B. Keeney et al. 2011, in preparation). It should be noted that for strong lines, or lines significantly broader than the instrumental profile (greater than ∼50 km s⁻¹), this phenomenon has virtually no effect on the ability to extract information from the spectrum at any S/N level. Although the minimum detectable equivalent width, at a fixed S/N, is reduced from this effect (compared to pre-launch predictions), the greatly increased S/N in COS spectra more than makes up for this shortcoming, making COS vastly more sensitive to detecting weak features against a faint continuum, in a fixed exposure time, when compared to STIS.

Defining the spectral resolution (Δλ) of COS as the FWHM of the instrumental profile (in physical coordinates) multiplied by the dispersion of the instrument yields a resolving power for COS (λ/Δλ) of 16,000–22,000 in the FUV channels, with the highest resolving power at the longest wavelengths in each channel. The resolving power increases primarily because λ increases, but also because the LSF improves at longer wavelengths (see Figure 8). For a detailed discussion of this issue, refer to Kriss (2011).

Zoom In Zoom Out Reset image size

It was not originally anticipated that breakthrough science would result from COS observations due to an increase in resolving power, since spectrographs with significantly higher resolving power have been (GHRS), and continue to be available (STIS). However, the limited sensitivity of these instruments meant that many classes of astrophysical targets could only be observed with their low-dispersion modes. Subsequent observations with COS have, in fact, greatly increased the spectral resolution of the data on these targets, resulting in exciting new science. As an example, in Figure 14 we show the spectrum of the classical T Tauri star AA Tau taken in 9900 s of COS observing time, and the previously best high S/N spectral data on this object, an ACS prism spectrum with a 10,900 s exposure. Clearly the ability to understand the physical processes in the target is greatly enhanced with the COS data, through its greatly improved spectral resolution and S/N.

Even though the design of COS FUV channel retains residual astigmatism in order to optimize the spectral resolution with a single optic, spatial imaging at the approximately 1'' level remains. Two point sources, separated by 1'' in the cross-dispersion direction, will present completely separated spectra in the FUV channel. At the time of this writing, this capability has not been utilized on any observations that are beyond their proprietary period. However, recently executed observations of the inner ring of SN1987A (HST ID 12241, PI - R. Kirshner) should be able to utilize this capability to great scientific benefit.

It should be noted that the calculation of the y-position (cross-dispersion axis) by the detectors is done independently of the x-position and that this calculation should be corrected by the gain of each photon event (the pulse height) to obtain the most accurate cross-dispersion position and best imaging performance. This y-position dependence on gain is referred to as y-walk and has only recently been well quantified (Sahnow et al. 2011). This should soon be implemented in the COS pipeline, as even for point sources, the geometric correction of the detector involves both the x- and y-positions, and the flat-field corrections are improved if this effect is properly accounted for.

In the NUV channel, the image is fully corrected by the use of multiple optics. Therefore, the imaging capabilities are defined solely by the inherent aberrations of the correcting and camera mirrors, and the detector resolution, which is the dominant effect. The cross-dispersion resolution of NUV spectra is approximately 01.

When a flat mirror is placed into the optical path of the NUV channel instead of a diffraction grating, the resulting mode is referred to as Target Acquisition 1 mode (TA1). This results in a highly sensitive imaging mode with high-angular resolution and a very limited field of view. In Figure 9 we show the image of Pluto and Charon, obtained in a 60 s exposure during target acquisition. The separation of the two objects at the time of this exposure was 08.

Zoom In Zoom Out Reset image size

The 25 diameter aperture of COS makes the observation of diffuse objects highly sensitive but with degraded spectral resolution. Because the COS aperture rests at a point that is out of focus, the actual sky acceptance of the aperture does not have a sharp edge, and light from regions outside the nominal angular size of the COS aperture is accepted, while regions within the aperture, but near the edge, are vignetted. Please refer to Figure 2. Full illumination of the aperture reduces the resolving power of the M-modes to about 1500. The étendue (AΩ) of COS at 1200 Å is approximately 3 × 10⁻⁷ cm² sr. As an example of the ability of COS to observe diffuse emission, in Figure 10 we show the early release observations of the oxygen-rich LMC supernova remnant N132D (France et al. 2009), along with the previously best spectrum obtained with the HST Faint Object Spectrograph (Blair et al. 2000).

Zoom In Zoom Out Reset image size

COS was designed to maximize the sensitivity of HST with respect to ultraviolet spectroscopy. While comparisons of instrument efficiency and effective area are informative, in the end it is observing time required to achieve a particular S/N that must be considered the most direct comparison. A full S/N calculation includes not only the collecting power of the instrument, but also the corrections to the final S/N from dark backgrounds and scattered light. For purposes of comparison to STIS, the times to achieve a particular S/N in the COS G130M mode at 1200 Å and 1600 Å are compared to that needed to obtain a comparable S/N in the STIS E140M mode at the same wavelengths. The advantages of COS are greater at the shortest wavelengths, as STIS scattered light performance and slit transmissions affect the final S/N most powerfully at the shortest wavelengths, while the COS noise performance is relatively wavelength independent. These comparisons also depend upon the nature of the spectrum observed. For example, pure emission-line sources are less dependent upon correction for scattered light than observations of continuum sources. In the comparisons below, the source is assumed to be a flat continuum (in F_λ). Since the STIS E140M mode has higher spectral resolution (R = 42,000) than the G130M mode, the comparison is also made at an S/N reduced by $\sqrt{2}$ for STIS E140M, to reflect the ability to bin the higher resolution STIS data to have comparable resolution to COS. It should also be noted that STIS E140M has a bandpass approximately twice the G130M bandpass, so that if observation of the entire 1150–1750 Å region is required, the COS net advantage needs to be reduced by approximately a factor of two. (All "time-to-signal-to-noise" calculations were made using the Space Telescope Cycle 19 Exposure Time Calculator, with the STIS E140M 6'' × 02 slit, and default sky values.)

While COS has a 10–20 times advantage in the effective area compared to STIS echelle modes, its greatly reduced scattered-light levels and low backgrounds magnify its advantages at low flux levels, as can be seen in Table 2. The original design goal of improving time to S/N by a factor of 10 has been substantially exceeded. At the lower flux levels, S/N is dominated by corrections for noise rather than just by signal. The low scatter and low background of COS are just as crucial for the data quality as the enhancements to the collecting power.

COS has broken through the F_λ = 1 × 10⁻¹⁴ erg cm⁻² s⁻¹ Å⁻¹ flux barrier, reducing observing time for fainter systems by a factor of 100 or more, and making practical the systematic observation of hundreds of QSOs. Many Galactic targets of great interest also have flux levels at or below this level. It is this enormous enhancement in observing power that has allowed COS to make revolutionary observations of faint systems, both extragalactic and Galactic. Some of these observations are highlighted in the next section.

At low redshift ∼90% of the normal baryonic matter probably resides in the regions between galaxies with only ∼10% found in luminous objects (stars and galaxies) (Fukugita & Peebles 2004). The cooler 30% of the baryons with T ≈ (1–4) × 10⁴ K have been traced by the Lyα forest (Penton et al. 2004; Lehner et al. 2007; Danforth & Shull 2008) and some of the O vi lines (Tripp et al. 2008; Thom & Chen 2008). The warmer 20%–30% of the baryons with T ≈ (1–10) × 10⁵ K have been traced by broad Lyα (Richter et al. 2006; Lehner et al. 2007; Danforth et al. 2010), O vi absorption (Danforth & Shull 2008; Tripp et al. 2008), and Ne viii absorption (Savage et al. 2005; Narayanan et al. 2009). The remaining ∼40% of the baryons probably reside in hotter gas with T > 10⁶ K. The hot baryons can be studied at X-ray wavelengths through searches for O vii and O viii and other X-ray absorption lines. At UV wavelengths, warm and hot baryons can be studied via redshifted absorption lines of Ne viii, Mg x, Si xii, and O vi and by broad H i Lyα (BLA) absorption.

COS provides a major new facility for studies of the low-redshift IGM. Although COS has a spectral resolution lower than the E140M mode on the STIS, COS is ∼15 times more efficient in collecting photons, making it possible to obtain relatively high S/N spectra of bright QSOs for IGM absorption line studies. The search for Ne viii and/or BLA absorption associated with O vi absorption is greatly facilitated when the absorption line spectra have S/N ∼ 30–40.

The importance of the higher S/N spectra obtained by COS has been demonstrated through the detection of Ne viii (λλ770, 780) absorption in an intervening multi-phase absorption system containing O vi at z = 0.495096 toward PKS 0405-123 (z_em = 0.5726) by Narayanan et al. (2010). The detected Ne viii directly points to high-temperature gas since Ne viii is very difficult to produce by photoionization by the general extragalactic EUV radiation background. Therefore, the Ne viii and associated O vi likely arise in collisionally ionized gas at T ≈ 5 × 10⁵ K with a large baryonic column density of log N_H ≈ 19–20. The multi-phase absorber either traces the gas in the hot halo of a low-luminosity galaxy with an impact parameter of 110 h⁻¹₇₀ kpc detected by Chen & Mulchaey (2009) or in a warm IGM structure connecting to the galaxy.

In another example, Savage et al. (2011) have reported the detection of BLA and strong O vi absorption in a multi-phase absorption system at z = 0.22601 toward HE 0153−4520 (z_em = 0.450). In this system, a BLA with b_H i = 140⁺¹⁴₋₁₆ km s⁻¹ and log N_H i = 13.70^+0.05_−0.08 shown in Figure 11 is likely associated with the O vi having b = 37 ± 1 km s⁻¹ and log N_O vi = 14.21 ± 0.02. The O vi and BLA measurements imply the direct detection of a trace amount of thermally broadened H i in hot gas with log T = 6.07^+0.09_−0.12, [O/H] = −0.28^+0.09_−0.08, and log N_H = 20.41^+0.13_−0.17. The observation illustrates the power of COS to detect and study gas in the low-redshift IGM with log T ≈ 6 and a very large total column density of hydrogen. The observed reference to H i via the BLA is crucial for determining the temperature, metallicity, and total baryonic content of the absorber.

Zoom In Zoom Out Reset image size

Additional examples of the power of high S/N IGM observations with COS include the detection of an H I-free broad O vi absorber at z = 0.1670 toward PKS 0405-124 (Savage et al. 2010) revealing hot gas associated with a pair of galaxies and the BLA and O vi absorber tracing a galaxy filament at z = 0.01025 toward Mrk 290, where log T ≈ 5.14, log N_H ≈ 19.6, and [O/H] ≈−1.7 (Narayanan et al. 2010).

Determining when and how the universe was ionized has been an important question in cosmology for decades (Gunn & Peterson 1965; Fan et al. 2006; Meiksin 2009). The epoch of reionization in hydrogen is a probe of the transition from neutral to ionized hydrogen in the IGM that occurred between redshifts z = 7–12, marking the exit from the cosmic "dark ages." Helium underwent similar reionization from He ii to He iii at z = 2.8 ± 0.2 (Reimers et al. 1997; Kriss et al. 2001; Shull et al. 2004), likely mediated by the harder (E ⩾ 54.4 eV) radiation from quasars and other AGNs. With a 4 ryd ionization potential, He⁺ is harder to ionize than H⁰ and He⁺² and recombines five to six times faster than H⁺ (Fardal et al. 1998). For these reasons and the fact that most hot stars lack strong 4 ryd continua, AGNs are likely to be the primary agents of He ii reionization. Owing to its resilience, He⁺ is much more abundant than H⁰, with predicted column-density ratios η ≡ N(He ii)/N(H i) ≈ 50–100 (Miralda-Escudé et al. 1996; Fardal et al. 1998).

HST/COS obtained ultraviolet spectra (Shull et al. 2010) of the bright (V = 16) high-redshift (z_em ≈ 2.9) quasar, HE 2347−4342, using both the G130M (medium-resolution, 1135–1440 Å) and G140L (low-resolution, 1030–2000 Å) gratings. The spectra exhibit patchy absorption in the Lyα line of He ii λ303.78 (Figure 12) between z = 2.39–2.87 (G140L) and z = 2.74–2.90 (G130M). COS provides better spectral resolution, higher S/N, and much better determined backgrounds than previous studies, with sensitivity to abundance fractions x_He ii ≈ 0.01 in low-density filaments of the cosmic web. The COS spectra provide clear evidence that the epoch of He ii reionization is delayed below z = 3, with intervals of patchy He ii flux transmission, punctuated by three long (5–10 Å) troughs of strong He ii absorption. These troughs, spanning 25–60 Mpc of proper distance between z = 2.751–2.807, z = 2.823–2.860, and z = 2.868–2.892, probably arise from filaments in the baryon distribution and the scarcity of any strong photoionizing source (AGN) within 30–50 Mpc. They are uncharacteristic (Furlanetto & Dixon 2010) of the IGM if He ii was reionized at z ≈ 3, and they suggest that the epoch of He ii reionization extends to z ≲ 2.7. A comparison of He ii and H i absorption in Lyα shows optical-depth fluctuations in the He ii/H i ratio on scales Δz ≈ 0.003–0.01. Corresponding to 4–10 Mpc in comoving distance, these fluctuations likely arise from AGN source variations (1–4 ryd spectral indices) and IGM radiative transfer effects. Probes of the He ii abundance fraction, x_He ii ⩾ (10⁻²)(0.1/δ_He)(τ_He ii/5), at redshifts z = 2.7–3.0 are now possible at optical depths τ_He ii ⩾ 5 in low-density regions (overdensities δ_He ≈ 0.1). The He ii absorption is 50–100 times more sensitive to trace abundances than H i at z = 6, where a neutral fraction x_H i = 10⁻⁴ saturates the Lyα absorption.

Zoom In Zoom Out Reset image size

Absorption spectra in the ultraviolet provide sensitive diagnostics of conditions in the halo of the Milky Way, where the competing process of accretion and outflow determines the evolution of the Galaxy. Recent HST spectra provide considerable insight into the infall of low-metallicity gas onto the disk, an ongoing process that can account for the observed stellar metallicities, star formation rates, and mass–metallicity relations (Pagel 1994). A previous HST/STIS survey of Galactic high-velocity clouds (HVCs) in the sensitive UV absorption line of Si iii λ1206.50 Å (Shull et al. 2009) suggested that the low Galactic halo is enveloped by a sheath of ionized, low-metallicity gas. This gas reservoir can provide a substantial cooling inflow (∼1 M_☉ yr⁻¹) to help replenish star formation in the Galactic disk, recently estimated at 2–4 M_☉ yr⁻¹ (Diehl et al. 2006; Robitaille & Whitney 2010). Galactic disk formation is believed to occur by the gradual accretion of pristine or partially processed material into the interstellar medium. The metallicity of the initial reservoir of gas is enriched by ejecta from star formation, mixed with infalling low-metallicity gas, possibly through "cold-mode accretion" (Dekel & Birnboim 2006; Kereš et al. 2009). This process continues to the present day, regulated in a manner that produces the local G-dwarf metallicity distribution and avoids the overproduction of metal-poor disk stars. The infall of gas from the IGM and Galactic halo also places a chemical imprint on mass–metallicity relations (Erb et al. 2006).

COS has recently obtained high-quality UV spectra of HVCs in Complex M (Yao et al. 2011) and Complex C (Shull et al. 2011). The wide spectral coverage and higher S/N, compared to pre-SM4 HST spectra, provide a better velocity definition of the HVC absorption, additional ionization species, and improved abundances in halo gas. Complex M abundances are high, between 0.85 and 3.5 Z_☉, suggesting that it consists of returning Galactic fountain gas. In contrast, Complex C has a low metallicity (0.1–0.3 Z_☉; Collins et al. 2007) and a wide range of ion states that suggest dynamical and thermal interactions with hot gas in the Galactic halo. COS studied four AGN sight lines (Figure 13) toward Mrk 817, Mrk 290, Mrk 876, and PG 1259+593. Spectra in the COS medium-resolution G130M (1133–1468 Å) and G160M (1383–1796 Å) gratings detect ultraviolet absorption lines from eight elements in low-ionization stages (O i, N i, C ii, S ii, Si ii, Al ii, Fe ii, and P ii) and three elements in intermediate- and high-ionization states (Si iii, Si iv, C iv, and N v). The HVCs exhibit a range of H i and O vi column densities, log N_H i = 19.39–20.05 and log N_O vi = 13.58–14.10, with substantial amounts of kinematically associated photoionized gas. The high-ion abundance ratios are consistent with models of cooling interfaces between photoionized and collisionally ionized gas: N(C iv)/N(O vi) ≈ 0.3–0.5, N(Si iv)/N(O vi) ≈ 0.05–0.11, N(N v)/N(O vi) ≈ 0.07–0.13, and N(Si iv)/N(Si iii) ≈0.2. The thermal pressure of the hot Galactic halo, P_halo/k ≈ 10–20 cm⁻³ K, is comparable to the ram pressure on the HVCs. As the HVCs encounter higher densities in the stratified lower halo, they may be dissipated by interface instabilities, feeding the Galactic halo rather than replenishing the reservoir of star formation in the disk. Individual clumps of Complex C (Figure 13) appear to be near virial equilibrium with pressure confinement. Their masses (⩾10⁵ M_☉) are probably above the threshold for survival against interface instabilities (Heitsch & Putman 2009).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The combination of high-sensitivity, low instrumental backgrounds, and moderate spectral resolution afforded by COS has provided a new window to the physical and chemical state of protoplanetary disks. Previously, only a small sample of the nearest young (age ≲ 10 Myr) low-mass circumstellar disk targets were bright enough for moderate resolution far-ultraviolet spectroscopy (Herczeg et al. 2006). The most extensive ultraviolet surveys were carried out with the G140L mode of STIS and the spectroscopic modes on ACS (e.g., the PR130L mode; Ingleby et al. 2009). The large instrumental background and very low dispersion of these data severely limit their utility for measurements of line identifications, fluxes, and widths. We are in the process of assembling a large sample (>30) of protoplanetary disk systems by combining the guaranteed time observation targets with several GO programs. This work is shedding new light on the molecular composition and physical characteristics of the planet-forming regions (a ≲ 5 AU) in these systems.

Photoexcited molecular hydrogen (H₂) emission has long been recognized as an important probe of protoplanetary disk surfaces (Brown et al. 1981; Valenti et al. 2000), however low-resolution and/or high instrumental backgrounds have prevented a detailed analysis of this emission in all but the brightest sources (Herczeg et al. 2002, 2004; Walter et al. 2003). With COS, we are not only acquiring high S/N, velocity-resolved profiles of tens to hundreds of H₂ emission lines in every gas-rich disk observed (Yang et al. 2011), but we have also begun to push these studies into the regime of protoplanetary disks orbiting brown dwarfs (France et al. 2010a). The ability to separate line and continuum features at very faint flux levels (F_λ ≲ 2 × 10⁻¹⁵ erg cm⁻² s⁻¹ Å⁻¹) has enabled the first detailed study of H₂ excited by non-thermal electrons (France et al. 2011b) in a protoplanetary environment.

Perhaps the most intriguing discovery in the COS circumstellar disk program is the detection and characterization of carbon monoxide (CO) in far-ultraviolet spectra of low-mass protoplanetary systems. France et al. (2011a) presented the first spectrally resolved detections of emission and absorption of lines of CO in these environments. This work showed that CO can account for a significant fraction of the "1600 Å bump" previously thought to be H₂ emission that has been suggested as a potential measure of the mass surface density of protoplanetary disks. Initial modeling results (Figure 14; Schindhelm et al. 2011) indicate that the photoexcited CO population in these disks may be spatially distinct from the well-studied M-band CO fundamental emission (see, e.g., Najita et al. 2007), suggesting a stratification of the molecular structure in low-mass protoplanetary disks not predicted by existing models.

The high sensitivity and low background of COS have allowed for the observation through heavily reddened sightlines. Five stars have been observed, with A_V ranging from 2.7 to 7.6. Such a high-extinction sightline may be simply a long sightline through low-density clouds, as in Cyg OB2 8A (A_V = 4.83; Snow et al. 2010), which shows atomic depletions and molecular content consistent with the diffuse interstellar medium, or may sample translucent clouds, which represent a change in photochemical regimes due to higher densities and heavy shielding from the photodissociating interstellar radiation field (Sheffer et al. 2008). In the latter case, carbon monoxide (CO) becomes the dominant form of gas-phase carbon, and Burgh et al. (2007) showed that the CO/H₂ ratio provides a good indicator of the presence of translucent material.

One sightline in particular may have even probed dense cloud material, an unprecedented look at a regime formerly only accessible to infrared or longer wavelengths. NGC 2024-1 (A_V = 7.6) exhibits a CH column density indicating a molecular fraction near unity and some of the strongest ultraviolet CO absorptions ever observed. The inferred CO/H₂ ratio is near 10⁻⁴, indicating that nearly all the carbon along the line of sight is in CO. In conjunction with ultraviolet absorption spectroscopy taken previously, COS allows us to probe the transition region between the diffuse and the dark clouds in a consistent manner.

The star formation rate densities in starburst galaxies can exceed the values found in normal galaxies by orders of magnitude. The ionizing radiation from the newly formed young stars can lead to prominent Lyα emission due to recombination of hydrogen in the ambient interstellar medium. The Lyα line has become an important spectral signature in young galaxies at high redshift since the Lyα luminosity amounts to a few percent of the total galaxy luminosity concentrated in a small wavelength bin (Stiavelli et al. 2004). However, the assumption of Lyα being a pure recombination line in a gaseous medium is too simple. Resonant scattering by atomic hydrogen significantly increases the likelihood of Lyα destruction by dust, leading to much lower line strengths (Neufeld 1990; Charlot & Fall 1993). Equally important, the kinematic properties of the interstellar medium may very well be the dominant escape or trapping mechanism for Lyα (Mas-Hesse et al. 2003).

The theoretical situation is sufficiently complex that observational tests are called for. While a large body of data exists at high redshift (Hayes et al. 2010, and references therein), there are few (if any) Lyα surveys of starburst galaxies in the local universe. Such surveys are motivated by the claim of a trend of increasing Lyα escape fraction with redshift (Deharveng et al. 2008). Moreover, spectroscopy of local Lyα emitters can be complemented by high-resolution imagery, thus providing additional constraints not available at high redshift. Previous generations of UV spectrographs were quickly pushed beyond their limits when observing suitable candidate galaxies; in order to be able to observe Lyα intrinsic to the galaxy, the redshifts should be high enough to avoid Galactic foreground Lyα absorption, which at the same time implies large distances and rather faint flux levels in the UV.

The vastly superior sensitivity has finally enabled observations of local starburst galaxies at large enough redshift and in statistically meaningful numbers to study their Lyα properties. We selected a sample of 20 star-forming galaxies from the KISS database (Salzer et al. 2000). These galaxies form a homogeneous set selected via their Hα emission, with a wide range of oxygen abundances, dust reddening, and luminosities.

In Figure 15 we show COS G130M spectra obtained for two of the program galaxies. The plethora of spectral features can be grouped into broad stellar-wind lines (e.g., N v λ1240), numerous weak stellar photospheric absorptions, narrow interstellar lines (e.g., C ii λ1335), and the complex Lyα feature (Leitherer et al. 2011). Lyα can be a pure, damped profile, an emission line associated with a galactic outflow, or a combination of both. The rich details provided by COS will allow us to study the physical conditions in the Lyα emitting regions (including the Lyα escape fraction) and their correlations with, e.g., the galaxy kinematics, dust content, and star formation properties. Early results from these efforts are presented in France et al. (2010b).

Zoom In Zoom Out Reset image size