THE HISTORY AND ENVIRONMENT OF A FADED QUASAR: HUBBLE SPACE TELESCOPE OBSERVATIONS OF HANNY'S VOORWERP AND IC 2497^*

Wide Field Camera 3 (WFC3) images were obtained in three bands selected to minimize the contribution of emission lines, in order to study the structure of IC 2497 and seek evidence for star clusters (of whatever ages) in the Voorwerp or elsewhere in the massive H i tidal stream found by Józsa et al. (2009). For the UVIS CCD images, two exposures with 2 pixel dither offsets in each detector coordinate were obtained in each of F225W and F814W. After some experimentation for the best balance between typical signal-to-noise ratio (S/N) and cosmic-ray rejection, we used a multidrizzle product (Koekemoer et al. 2002) with 2σ rejection for cosmic rays and a 2 pixel growing radius around identified cosmic rays. With only two exposures, a large number of pixels are still affected by cosmic-ray events; these were patched interactively for display purposes.

The near-IR F160W image used multiple readouts (in the STEP100 sequence) and a three-point dither pattern to yield a well-sampled combined image with excellent dynamic range and cosmic-ray rejection.

We used the tunable ramp filters on the Advanced Camera for Surveys (ACS) to isolate [O iii] λ5007 and Hα at the system redshift, with filter half-transmission width nominally 2% (105 and 138 Å, respectively). The orientation was constrained so that the inner parts of IC 2497, as well as all parts of Hanny's Voorwerp identified from ground-based imaging, fell within the 40'' × 80'' monochromatic field of view. Ground-based data showed the weakness of the continuum compared to these lines, so no matching continuum data were obtained. Processing these images took special care; cumulative radiation damage to the CCDs has resulted in charge-transfer tails from cosmic-ray events affecting much of the area of each image. The best correction for this effect has been presented by Anderson & Bedin (2010): a constrained deconvolution process, which has the net effect of restoring charge in the near-exponential tails back to its original pixel. We used the PixCteCorr routine within PyRAF to apply this correction on the individual ACS images, followed by drizzle combination with cosmic-ray rejection. Restoration of the charge-transfer efficiency (CTE) charge to its original pixel made cosmic-ray rejection much more effective.

As an early check on the role of charge-transfer trails, before the Anderson & Bedin (2010) routine was released, we used a multistep heuristic process, incorporating ground-based images to restore structure at low surface brightness (SB). We used similar drizzle parameters to combine the images. Then a large median filter in a blank region of the images was used to model and remove low-level additive banding appearing in the images. The cosmic-ray trails were removed by subtracting the result of a 31 pixel (15) median filter along the axis closely matching detector y, applied only to pixels with values lower than 0.01 count s⁻¹ (roughly S/N < 4). This step removed real structures at lower SB. To restore these, we used ground-based images in V (dominated by [O iii]) and Hα (Lintott et al. 2009) as follows: the ACS images were convolved with Gaussian filters to approximate the ground-based point-spread functions (PSFs), and the difference between (scaled) ground-based and ACS data was masked in the bright regions where the filtering did not apply. Then this masked difference was added back to the processed ACS images. Finally, residual cosmic rays were patched interactively. This last process was applied only to the region around Hanny's Voorwerp; the central regions of IC 2497 were much brighter than the cosmic-ray residuals and were not so treated. This processing sequence left faint residual streaks from the cosmic-ray response in the Voorwerp, aligned with the columns of the chip (position angle (P.A.) of 112°). However, PixCteCorr gave superior results, with no low SB artifacts evident on comparison of the two results, so we present further analysis based on the CTE-corrected data.

The relative calibration of the emission-line images was extracted from the passbands generated by the STSDAS calcband task. In energy units, the intensity ratio of the two lines corresponding to equal count rates is the inverse ratio of peak throughputs (since the emission lines were placed at the band peaks) multiplied by the ratio of photon energies. From this calibration, equal count rates in the two lines correspond to an intensity ratio in energy units [O iii]/(Hα+[N ii]) = 0.60.

To improve the registration of the HST images for comparison with the radio reference frame, we compared the positions of stars retrieved from the WFC3 images with their SDSS coordinates, rejecting stars whose offset exceeds 01 due to proper motion or unresolved duplicity (two in the UVIS field and one in the smaller IR field). We find coordinate corrections (from the HST to SDSS systems) given by Δα = 0.0147 ± 00016^s, Δδ = −0.057 ± 0046 for the F814 image, and Δα = 0.0299 ± 00021^s, Δδ = −0.127 ± 0034 for the F160W image. The absolute accuracy of the SDSS coordinate frame against the USNO astrometric network is 0045 (Pier et al. 2003), which is the limiting factor in our comparison against the radio reference frame.

We obtained moderate-resolution spectra using the Space Telescope Imaging Spectrograph (STIS) with gratings G430L and G750L. The 02 slit (aperture 52X0.2) was used to ease acquisition tolerances in a complex region, and improve SB sensitivity to emission lines around the galaxy nucleus. A 100 s "white-light" acquisition image was used for slit placement, with the options ACQTYPE=DIFFUSE, DIFFUSE-CENTER=FLUX-CENTROID specified so as to center the slit even if the central light distribution proved to be asymmetric (as it did). The 5 × 5 arcsec acquisition image, in the very broadband redward of 5500 Å provided by the long-pass filter, also proved useful in comparison with the narrowband images of the center of IC 2497. For the red spectrum, a fringe flat was obtained during Earth occultation immediately after the F750L exposures. The slit orientation was along celestial P.A. 598, dictated by scheduling and power constraints; slit positions including both the nucleus and either the southwestern star-forming regions within IC 2497 or the Voorwerp itself were not feasible. Fortuitously, this orientation did sample an extended emission-line loop near the nucleus (Section 3.3).

Each grating had two exposures within an orbital visibility period (again, a compromise between S/N and efficiency of cosmic-ray rejection); for display purposes, we interactively patched residual cosmic rays on a single-pixel basis. Our redshift values incorporate the use of vacuum wavelengths in STIS spectra, optical, and ultraviolet.

2.5.1. Ground-based Imagery

2.5.2. Ground-based Spectroscopy

2.5.3. GALEX Images and Spectra

We have analyzed Galaxy Evolution Explorer (GALEX) data obtained as part of the Nearby Galaxy Survey (Gil de Paz et al. 2007), with long exposures both in direct and spectroscopic modes (Table 1). Because Hanny's Voorwerp is such an extended object, we extracted its integrated spectrum directly from the two-dimensional dispersed images, using dispersion relations and effective-area values from Morrissey et al. (2007). A small error in flux calibration across emission lines is introduced by treating pixels within the line as having wavelengths appropriate to their offset from the object center (rather than all these pixels having effectively the same wavelength) but this error falls well within the Poisson errors of the spectra. The GALEX spectra are combined in Figure 1. The C iv λ1548 + 1550, He ii λ1640, and C iii] λ1909 lines are clearly detected; [Ne iv] λ2425, sometimes seen in the extended emission of radio galaxies, may be present at the ≈2σ level. Their integrated fluxes are compared to optical lines (from Lintott et al. 2009) in Table 2.

Zoom In Zoom Out Reset image size

Table 2. Integrated UV/Optical Line Fluxes from Hanny's Voorwerp

Line	Flux
	(10−15 erg cm−2 s−1)
C iv λ1549	51
He ii λ1640	50
C iii] λ1909	49
[Ne iv] λ2425	10
He ii λ4686	6.7
[O ii] λ3727	27
[O iii] λ5007	191
Hα	62

Download table as:  ASCIITypeset image

In both the optical and UV ranges, the strength of He ii lines is notable. The He ii λ4686/Hβ ratio is in the upper part of range seen for extended emission-line regions around QSOs (Fu & Stockton 2009), perhaps suggesting that the ionizing continuum from IC 2497 is at the hard end of the range seen for QSOs. The He ii λ1640/λ4686 ratio in AGNs can be affected by resonant effects, since H Lyα nearly coincides in energy with the excitation level of n = 4 in He⁺ (MacAlpine et al. 1985). Such effects should be much reduced in the low-density, kinematically quiet environment here; the energy difference between H and He energy levels translates to the Doppler shift from motions at 120 km s⁻¹, much larger than local changes seen in our long-slit spectra. This makes the low-density value for λ1640/λ4686 appropriate for the intrinsic line ratio, so we can set a limit to the reddening. When resonant effects are negligible, the line ratio is 6.8 in energy units (using values from Osterbrock & Ferland 2006). The estimated error on the integrated optical line flux from Lintott et al. (2009) is 12%, while Morrissey et al. (2007) quote a flux accuracy of 3% for spectral regions near the center of the field (as applied here); photon statistics are a larger contribution in this case, with S/N = 12 for the line. The observed line ratio is then 7.46 ± 1.1, implying a reddening limit E_4686–1640 < 0.25 mag. For a typical Galactic extinction law, this would be A_V < 0.18; foreground Galactic extinction in this direction would be only A_V = 0.04 following Schegel et al. (1998). This strongly limits any role for imbedded dust in the emission-line regions, as might be expected at the low densities n_e < 50 cm⁻³ found from [S ii] emission-line ratios.

To quantify the roles of line and continuum contributions, we use the GALEX spectrum to estimate the fractional contribution of emission lines in each UV filter. We did so by converting the GALEX spectrum to photon units, multiplying by each filter's response curve, and measuring the difference in total count rate when emission lines are removed by interpolation. This yields emission-line fractions of 0.14 in the GALEX far-UV (FUV) image, 0.15 in GALEX near-UV (NUV), and 0.06 in the WFC3 F225W filter. In each case, the UV emission is dominated by continuum processes, and strongly so for F225W. These data confirm that the WFC3 F225W image samples mainly the continuum, as intended. This continuum might have contributions from recombination processes, an embedded population of hot stars, or scattered AGN radiation.

Given the dust lanes that are prominent on the southeastern side of the galaxy disk, we want to know how much optical obscuration there is toward the nucleus of IC 2497. On large scales, we can address this via comparison of the optical and near-IR HST images, as well as through the astrometric registration of the two VLBI sources reported by Rampadarath et al. (2010).

The nucleus (as defined by peak intensity) at F160W is found within 004 of VLBI component C2, with an external error of 005 from the scatter in stellar coordinates and systematic error limits on the SDSS frame. This suggests, first, that dust lanes do not provide strong obscuration toward the nucleus itself at this wavelength, and second, that VLBI component C2 rather than C1 (025 away) represents the AGN in this galaxy. This makes sense given that C2 has a flatter spectrum from 6 to 18 cm and is surrounded by diffuse emission roughly aligned with the galaxy disk. As noted by Rampadarath et al. (2010), C1 could be a compact knot in a jet directed toward P.A. 215°, roughly aligned with both the kiloparsec-scale smooth extension in the radio emission (Józsa et al. 2009) and the Voorwerp.

Obscuration arising close to the nucleus (in a traditional "torus") would apply to a large solid angle and be accompanied by strong far-IR (FIR) reradiation, allowing us to construct an energy budget; in contrast, obscuration by a foreground cloud, possibly far from the nucleus, need not have such an FIR signature. Such "accidental" obscuration may occur in some type 2 Seyfert galaxies, as suggested by the dust morphologies shown by Malkan et al. (1998). This analysis suggests that large-scale foreground obscuring structures are not dense enough to hide an AGN powerful enough to ionize Hanny's Voorwerp.

The F160W near-IR structure shows that the extinction in these dust lanes is modest at least at this wavelength, so that we have a view to the nucleus which is relatively unhindered by spatially resolved absorbing features. The F814W and F160W images are compared in Figure 2, resampled to the pixel scale of the F814W data, with a color map based on convolving the F814W image with the nearest Gaussian equivalent to match the PSFs (FWHM 6.1 pixels or 0.23 arcsec). The flux ratio, in particular, traces intricate filaments of dust near the nucleus. At the F160W resolution, the deep dust lane near the nucleus passes close enough to leave the amount of extinction in F814W ambiguous, especially when including the potential 006 positioning error with respect to the core radio source C2 (Figure 3). When smoothed to the lower resolution, the implied extinction at F814W ranges from 0.8to 1.0 mag across the error circle of the VLBI core position. The centroid derived from elliptical isophotes at F160W, at radii up to 0.8 arcsec, has offsets in the range of 0.06 ± 0.06 arcsec from C2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Using intensity profiles of IC 2497 and a bright field star, we find that subtracting a nuclear point source brighter than 1.7 × 10⁻¹⁸ erg cm⁻² s⁻¹ Å⁻¹ at 1.6 μm would leave a starlight profile with a central depression, so we take this as an upper limit to the emerging intensity directly from an AGN component. This is a slightly less stringent limit than would be given by extrapolation of the X-ray power-law detection by Schawinski et al. (2010). There is also a near-point source of emission in the F225W near-ultraviolet filter, but position registration shows its centroid to be displaced by 0.07 arcsec (1.6 WFC pixels) away from the dust lane compared to the peak in F814W, so its observed structure and flux are strongly influenced by the dust lane. A typical reddening curve with R = 3.1 would have an extinction at 2550 Å of ≈ 8 mag for our estimated F814W extinction of 1 mag, so that the apparent UV nucleus is either surrounding structure or is seen through patchy extinction that we have no good way to quantify.

The only way to hide a luminous AGN in this object without exceeding the observed FIR flux appears to be through beaming, as distinct from obscuration by nearby material, as proposed for the emission-line filaments along the jet of Centaurus A by Morganti et al. (1992).¹⁴ The cone angle involved in IC 2497 would roughly span 55° in diameter to encompass all the detected [O iii] emission, which is too broad to suppress the putative nuclear emission in our direction to unobservable levels. Relativistic beaming would give this half-power width at a Lorentz factor γ ≈ 2, bulk velocity v/c = 0.86, in which case the Doppler deboosting in our direction (about 125° from the axis) would be a modest factor 0.3, leaving emission from the core very prominent. Thus, neither obscuration by a torus nor beaming offers an appealing explanation for why Hanny's Voorwerp sees a luminous AGN that is virtually absent along our line of sight (providing further support for our interpretation in Lintott et al. 2009; Schawinski et al. 2010).

The STIS spectra of the nucleus of IC 2497 show line ratios characteristic of a Seyfert nucleus, on the low-ionization side of the category in common diagnostic diagrams (Table 3 and Figure 4). With reduced contamination from bulge starlight compared to our earlier ground-based data, the line equivalent widths are larger, shrinking the error bars due to uncertainty in the underlying starlight (particularly in Hβ absorption; with the observed equivalent width of 25 Å, corrections for underlying stellar absorption will not exceed 30% (8 Å), and would lower the derived [O iii]/Hβ and ionization level. The Doppler widths are substantial, enough to completely blend Hα and the adjacent [N ii] lines. We separate the contributions using Gaussian deblending with the line separations fixed. There may be a contribution to Hα from a low-luminosity broad-line region, since the FWHM needed to fit the Hα+[N ii] blend is twice that of the unblended [O iii] λ5007 line. Combining these features, the nuclear spectrum would be classified as a Seyfert 2 or Seyfert 1.9 galaxy (depending on the detection of a broad-line region contribution at Hα).

Zoom In Zoom Out Reset image size

The two-dimensional profiles of [O iii] and Hβ show some ionization and velocity structure close to the resolution of the spectra.

The line ratios let us derive an ionization parameter for photoionization, following the procedure in Bennert et al. (2006). We deredden the ratios taking the case B Balmer decrement of Hα/Hβ = 2.87; if a distinct broad component of Hα contributes significantly, the dereddening correction would be smaller. For the nuclear emission region, the oxygen lines give log U = −3.5, somewhat lower than our estimate of −3.2 in Lintott et al. (2009). The difference is due to both tighter bounds on Hβ, since contamination from stellar absorption is much smaller, and better sensitivity and calibration accuracy into the blue.

The nuclear emission region is small—FWHM 4 pixels along the slit (020). Thus, the characteristic distance of the ionized region from the core is <010 (100 pc) if this region is centered on the core source. This size measurement makes the constraints on the ionizing luminosity of the nucleus from Lintott et al. (2009) even stronger. In a straightforward scaling, the gas near the nucleus is 150 times closer to the AGN than the Voorwerp, while the density n_e = 560 cm⁻³ (Lintott et al. 2009) is 10–40 times higher. Thus, the space density of ionizing photons is 600–2500 times smaller in the circumnuclear gas than it is upon reaching the Voorwerp. This remarkable mismatch is nontrivial, since producing these line ratios from photoionization requires a radiation field which is as hard in spectral shape as typical AGN continua. An obscured AGN would not provide this; lines of sight sufficiently clear to see ionized gas with these line ratios would be sufficiently unreddened to reveal a significant fraction of the core luminosity in their ionization parameter.

A second distinct emission-line region appears in the STIS spectra, 05 to the northeast of the nucleus (485 pc in projection at an angular-diameter distance of 198 Mpc). The Hα ACS image shows this as a coherent loop or ring passing through the nucleus, of which only a partial arc is strong in [O iii]; its definition can be improved by using the STIS broadband acquisition image as a continuum estimate (Figure 5).

Zoom In Zoom Out Reset image size

Compared to the nucleus, this region has larger redshift, and somewhat higher levels of both ionization and excitation (Table 3). The line width from the [N ii]+Hα blend is comparable to that from [O iii] in the loop, in contrast to the nucleus, and the [O iii] line width is smaller in the loop.

Can this structure be fairly described as a flat ring, or could it be a roughly spherical shell or bubble such as is seen on larger scales around a few AGNs? The SB profiles in both Hα and [O iii] are highly symmetric across the loop, after subtracting a sloping background from inside and outside. The radial emission profile is not well resolved, with FWHM ≈ 015 from the Hα image (020 from the STIS spectrum after continuum subtraction) about the peak radius. Comparison with simple numerical models of a spherical shell shows emission inside the ring at a higher level than we observe; this structure must be more like a flat ring than a complete shell. An additional emission-line feature appears beyond the ring to the northeast, most obviously interpreted as a component with strong [N ii] at velocities gradually returning to nearly the central value over a span of 8 pixels (032, 300 pc). This supports an interpretation of the ring as a local kinematic structure within more quiescent material.

For an expanding ring viewed at an angle i to its normal, the characteristic expansion age is given from observables (line-of-sight expansion velocity v_los and apparent minor diameter D_app) by $T = D_{{\rm app}} \cot i / v_{{\rm los}}$. For v_los = 310 km s⁻¹, this becomes $1.5 \times 10^6 \cot i$ years. (Since our slit was not aligned with the loop as seen in direct images, we do not sample v_los on the minor axis of the loop; this value could be larger and the loop correspondingly younger.) If the ring is in the plane of IC 2497, i ≈ 65°, so T ≈ 7 × 10⁵ years. Expansion perpendicular to the plane is ruled out by noting that it is projected on the far side of the plane as marked by the dust lanes and redshifted relative to the nucleus. The closer a ring lies to the plane of the sky, the shorter its expansion age for a given radial-velocity span. Of course, more complex nonradial motions and a nonlinear expansion history are allowed by our very limited sampling of its velocity (making our age estimate likely an upper limit).

The expansion timescale of this feature could be broadly comparable to the inferred time since the AGN faded; an intriguing possibility is that it was powered by a change in accretion mode to one in which much of the energy emerged in kinetic rather than radiative form ("radio mode"). Such a mode of accretion power has been discussed in connection with radio-loud AGNs having very weak continuum emission at higher energies (Churazov et al. 2005; Merloni & Heinz 2007). In particular, Merloni & Heinz (2008) find that a kinetically dominated mode dominates at low Eddington ratio, so that such a switch could indeed be associated with a drop in accretion rate.

The relatively large velocity dispersion of the emission lines in the ring suggests that the kinematics are more complicated than simple expansion in the galaxy's plane. With σ_v = 325 km s⁻¹ from Gaussian fits, the internal spread is a large fraction of the systematic Doppler component, yet the emission ring is narrower than would be expected from free expansion about this mean. Interpretations might be that motions are largely constrained to be normal to the ring (as at an interface between an outflow and dense disk), or that the emissivity drops rapidly away from the ring so that high-velocity gas disappears from the spectrum before moving far away.

Applying the Bennert et al. (2006) fitting forms, the line ratios in this loop give log U = −3.27, somewhat higher than the nucleus. With only a few strong emission lines detected, and lines profiles broad enough to accommodate the shock velocities associated with its ionization level, shock ionization remains a possibility for this feature.

Lack of more detailed information on the physical conditions in the expanding loop limits our ability to estimate the energy required to produce it. The blended [S ii] doublet has a mean wavelength implying that the doublet ratio is near unity, taking the line redshift to match [N ii] emission in the loop. At a typical temperature of 10⁴ K, this gives electron density n_e ≈ 900 cm⁻³. Very crudely, if we take an astrophysically typical filling factor 10⁻⁴ and fully ionized gas distributed in a ring of thickness comparable to its width, the mass involved would be ≈3000 solar masses. The kinetic energy of such a mass in a ring expanding at 300 km s⁻¹ is then ≈3 × 10⁵¹ erg. Compared to the ionizing luminosity needed to ionize the distant gas in the Voorwerp, this could be supplied by only a small fraction of the energy output when spread over even a few thousand of years. In fact, even compared to the estimated current luminosity of the AGN near 10⁴⁰ erg s⁻¹, this remains a tiny fraction of the total output over the lifetime of the ring. These could be additional outflows at temperatures not sampled by our data, such as a hot interior analogous to the Fermi bubbles in the Milky Way (Su et al. 2010); this situation would increase the energy requirements beyond our estimate for the loop itself.

Possibly related to the origin of the ionized-gas loop is the structure seen in the UV F225W image to the north of the nucleus. As shown in Figure 6, there is a larger plume with some of the brightest emission at about the same P.A. as the Hα loop. The brightest region near the nucleus occurs at roughly the same P.A. as the middle of the loop. The origin of this emission is poorly constrained; the filter is dominated by continuum processes, and the detection of H ii regions elsewhere in IC 2497 confirms that the exposure is deep enough to see bright star-forming regions, but there is no correlation between the UV and Hα near the nucleus. This is not synchrotron emission from a radio jet because the observed jet goes in the opposite direction, so that any radio emission on this side would be far too weak to produce detectable UV emission. Similarly, scattered light from an otherwise obscured and luminous AGN would occupy such a small solid angle in this feature as to suggest that most of its radiation is absorbed near the nucleus, violating the constraints on far-IR luminosity. The roughly radial UV structures are elongated roughly in P.A. 30°, which is not opposite either the Voorwerp or the radio jet knots. The data would be satisfied with a stellar population which is young, but still old enough not to ionize any nearby gas. The dust lanes make it clear that the UV feature is projected against the far side of the disk, and that any counterpart on the south side of the nucleus would be invisibly faint behind the dust.

Zoom In Zoom Out Reset image size

Using data obtained since the preparation of Lintott et al. (2009), we can revisit bounds on the bolometric luminosity of the AGN in IC 2497. We consider the X-ray results of Schawinski et al., and the recently released mid-IR values from Wide-field Infrared Survey Explorer (WISE).

As noted by Lintott et al. (2009), the FIR parameter, estimating the total output from 42–122 μm from IRAS fluxes at 60 and 100 μm, gives a far-IR luminosity of 6 × 10⁴⁴ erg s⁻¹ for IC 2497. With the addition of WISE detections (Wright et al. 2010) from near- to mid-IR, we can fill in the spectral energy distribution (SED) with measurements tracing very hot dust (which turns out not to be energetically important in this galaxy). Zero points of the WISE magnitude scales were taken from Jarett et al. (2011). IC 2497 has a near-IR spectral slope which does not indicate dominance by the AGN, using the color criterion from Stern et al. (2012); they present evidence that this would include even Compton-thick objects at high luminosity.

We have done a more detailed integration of the far-IR luminosity, combining IRAS and WISE measurements and interpolating between measured bands with piecewise power-law shapes. Extrapolation beyond 100 μm used a typical galaxy shape, a 20 K modified Planck law with emissivity scaling as λ⁻², and the whole composite spectrum was taken to peak at 150 μm. The resulting IR luminosity from 3.4 to 122 μm is 8.0 × 10⁴⁴ erg s⁻¹, about 25% larger than given by the FIR parameter. Extrapolation to longer wavelengths using our simple model would raise the total to 1.0 × 10⁴⁵ erg s⁻¹. As usual, this is an upper limit to reprocessed AGN radiation, since star formation also contributes (particularly at longer wavelengths). The implied ionizing luminosity alone for the AGN is twice this limit, and obscuration deep enough to hide the AGN so completely in the optical would also include non-ionizing radiation (very roughly three times its energy contribution for a typical radio-quiet AGN energy distribution. This means that the most generous accounting for reradiation from absorbed AGN emission gives a shortfall by a factor ≈6. The magnitude of the shortfall is increased when including an FIR contribution from star formation in IC 2497, the lack of an AGN color signature in the mid-IR, and the much stronger limits from emission-line gas in its nucleus.

The gaseous structure of Hanny's Voorwerp is best traced by the strong [O iii] emission line. As shown in Figure 9, it is strongly filamentary, with structures on all scales down to the ACS resolution limit. Among major features are the prominent "hole" surrounded by bright filaments, a small-scale loop or shell at the northern end toward IC 2497, a region with embedded star clusters to the northwest, and outlying regions of low SB spanning much of the east–west extent of the filter's monochromatic field. At the low densities derived from the [S ii] line ratio, the recombination timescale (αn_e)⁻¹ (where α is the recombination coefficient) is long compared to the light-travel time across all but the largest of these structures. For hydrogen, n_e < 50 cm⁻³ (Lintott et al. 2009) gives t_rec > 2400 years under nebular "Case B" (Osterbrock & Ferland 2006). Since the recombination coefficient α differs for various species, when the ionizing radiation is turned off, we would see different ionic species recombine and disappear from the spectrum at different times (as calculated by Binette & Robinson 1987). For instance, O^{2 +} recombines ≈100 times faster then H⁺ (Crenshaw et al. 2010). The density dependence of these values means that we might see preferential fading of dense regions after a decline in ionizing flux, which would act to flatten the intensity profile observed from a centrally concentrated cloud. This effect would change prominent line ratios such as [O iii]/Hα in the same sense as changes in ionization parameter solely due to density, changes which we do not observe (Section 5.2.2); both processes are likely overshadowed by the role of unresolved fine structure in the emission-line gas.

Zoom In Zoom Out Reset image size

Earlier imaging, in particular the WIYN V-band data, led us to speculate on features possibly resembling bow shocks. This resemblance is less striking in the HST data, but one structure of particular interest may show interaction with an external medium at the northern end of Hanny's Voorwerp. This is a nearly circular shell or loop, 19 in diameter and 06 thick (Figure 9). However, it could also result from a local explosive event, or be a chance configuration of filaments. We do not yet have kinematic data to test any of these possibilities. This ring does have areas of relatively low [O iii]/Hα around much of its circumference, lower than found elsewhere except for the star-forming regions to its west (Section 5.3). However, we do not see continuum sources here such as mark young star clusters in the western region.

If the pattern of ionized gas traces an ionization cone, it is a ragged one. There are filaments of low SB detected on either side of the obvious structures, and the bright area would trace a rough cone only if the gas is oriented in three dimensions with significant depth along the line of sight.

In the northern part of the Voorwerp, there are filaments which are edge-brightened on the side toward IC 2497, a situation which does not persist farther to the south. These filaments include a region where we see recent star formation (Section 5.3). The combination of morphology and association with young stars, whose formation could be triggered by compression, fits with a picture in which these structures in the gas are interacting with an outflow from IC 2497, such as shown in this direction by the Westerbork continuum data from Józsa et al. (2009). In this respect, an interaction has occurred analogous to that seen in such systems as Minkowski's Object and star-forming regions adjacent to the inner jet of Centaurus A. In contrast, what appears to be a robust jet/galaxy interaction in 3C 321 has not led to a starburst response (Evans et al. 2008), so the properties of both outflow and target gas play roles in the outcome.

The low density measured from [S ii] lines, and the modest role of the outflow in shaping the morphology of Hanny's Voorwerp, indicate that this wind from IC 2497 has a low ram pressure. In contrast, both the star-forming regions and Hα emission in Minkowski's Object show structures indicating strong interaction with the radio jet, with filaments in the downstream direction (Croft et al. 2006). We see this only to a very limited extent, and only in confined regions, in Hanny's Voorwerp, as shown in the [O iii] image (Figure 9). The best morphological case for such interaction—gas being entrained by an outflow—is in the series of "fingers" south of the star-forming regions. They do not point to a single point, but do point roughly away from the galaxy, arise from a single emission-line feature, and align with the star-forming area and the nucleus of IC 2497, so that entrainment makes sense if the outflow has enough ram pressure to overcome their internal pressure. This would suggest that the outflow has a typical ram pressure comparable to the internal pressure in the emission-line gas, so that local variations make the difference between significant and negligible roles for entrainment.

The star-forming regions and the emission-line "fingers" are oriented in P.A. 198° from the nucleus of IC 2497 and span a projected cone angle of width ≈10°, compared to the P.A. 209° of the inner jet as traced by the VLBI core and bright knot (Rampadarath et al. 2010). It is thus plausible that these traces of interaction with an outflow show us a narrow stream which is radio-bright only in its inner kiloparsec. This narrow outflow would be substantially misaligned with the emission-line hole (Section 5.1.1), making its origin in the same outflow unlikely.

5.1.1. The Emission-line Hole

A prominent feature of the Voorwerp, noted by Lintott et al. (2009) and indeed visible in the SDSS g image, is the dark "hole" in its southeastern part. Potential explanations include passage of a narrow jet, an explosion, a shadow cast by an intervening cloud of very large column density or happenstance in the distribution of ionized filaments. The new ACS data address these only in a negative sense—revealing no structural traces of why this is here. The ionization level does not increase toward its edges (Section 5.2.2), but some filaments do seem to go around its edges rather than crossing its edge and vanishing. An explosive origin seems unlikely, since our data show no X-ray emission from any hot bubble that would remain, excess ionization at its edges, or (within our limited spectroscopic information) a kinematic signature of much larger radial velocities at its edges.

If the hole traces the location of a (past) radio jet, it has left no trace in the emission-line material. Filaments do not turn "downstream" within the hole, as is seen in, for example, Minkowski's Object, where jet interaction is clear. The best estimate of the direction of a radio jet comes from the MERLIN and EVLBI observations of the core and inner knot by Rampadarath et al. (2010), which are aligned along P.A. 209°. In contrast, the direction from the nuclear component to the center of the hole lies along P.A. 181°, with the entire structure ranging from 173° to 190°. A jet would have to curve by at least 28° in projection to match the two locations.

A somewhat similar "hole" structure, within a larger region of bright filaments, is seen among the emission-line patches along the northeastern jet of Centaurus A. In the images by Graham & Price (1981) and Keel (1989), the "necklace" region likewise shows a roughly circular region surrounded by emission filaments. For comparison, this region is illustrated in Figure 10. In contrast to Hanny's Voorwerp, however, the ionization of these filaments in Centaurus A is attributed (mostly) to shocks, whose kinematic signature appears in the wide velocity range and broad lines of individual features (Sutherland et al. 1993). In Cen A as well, it is not clear whether the "hole" has direct physical significance, or results only from the arrangement of denser filamentary regions around it.

Zoom In Zoom Out Reset image size

An interpretation as a shadow from a cloud near the AGN does not violate any of the data, but we cannot make a definite statement on the origin of this feature.

The small-scale structure revealed by the ACS images allows us to probe the ionization structure within the gas and improve our limits on the AGN luminosity required to ionize it. This situation, with the same external radiation field at essentially the same intensity impinging on large areas of gas, allows us to apply basic photoionization and recombination principles to probe both the ionizing source and small-scale structure of the gas.

5.2.1. Ionizing Luminosity

As noted by Lintott et al. (2009) and Keel et al. (2012), the SB in recombination lines provides a lower limit to the luminosity needed to maintain the ionization. The new images reveal bright features that were smeared out in our ground-based images, increasing the peak SB and derived ionizing luminosity at each projected radius from the galaxy core. We have considered several ways to evaluate this peak SB. It is straightforward to find the peak pixel values, but these could be affected by overlap of multiple structures or seeing a feature which happens to be elongated along the line of sight. To allow for the effects of overlapping structures, or dense regions which happen to be elongated along the line of sight and therefore appear brighter than a simple calculation for a sheet or spherical structure would indicate, we consider two measures of the brightest regions in histograms of Hα SB in several bins of distance from the center of IC 2497 (Figure 11). The values labeled Max in Figure 11 are for the highest contiguously populated bin in Hα SB, while the values given as 99% are the 99th percentile in SB among pixels more than 2σ above the sky level. (Other measures, such as 1% of peak value, show similar behavior.) As expected, the lower limits to ionizing luminosity are higher when we see finer structure; the average derived from the four plotted zones is L_ion > 7 × 10⁴⁵ erg s⁻¹. In addition, while the peak SB by each measure declines for zones projected farther from IC 2497, this decline is shallower than the r⁻² expected for identical gas parcels illuminated by the same source. This mismatch could be explained either through geometry or through history of the ionizing source (systematic changes in characteristic gas density are disfavored by the ionization structure). Specifically, the quantity r²_projf for each flux measure f increases monotonically with r_proj (with one slight violation between the outer two zones in the 99% value). A geometric explanation would have the gas lying close to a plane which is highly inclined to the plane of the sky and not radial to IC 2497, so that the mapping between r_proj and r differs for each zone. For example, if the southernmost zone had r_proj = r, the gas would lie in a plane tilted by at least 45° to the plane of the sky.

Zoom In Zoom Out Reset image size

Alternatively, the behavior of Hα SB with projected radius could result from long-term variations in ionizing luminosity, reflected through different values of the light-travel time delay. For a gas distribution viewed "face-on" (in the plane of the sky), the ionizing luminosity would have decreased by a factor ≈2 over a timespan of 40,000 yr. The uncertainty in interpretation here points up the importance of the system geometry; this might be addressed by kinematic mapping and modeling of the entire H i stream.

5.2.2. Density and Ionization Structure

Point by point in a nebula, the SB in a recombination line scales with the emission measure

$${\rm EM} = \int {n_e}^2 dl.$$

We have an independent tracer of changes in n_e through the ionization parameter U; in this case, small-scale variations in U must trace changes in local density since adjacent pixels are at essentially the same distance from the ionizing source. If the features seen are due only to density contrast, SB should correlate negatively with U, as deduced from emission-line ratios. We see at most a very mild anticorrelation, so weak that most of the structure we see must result from changes in line-of-sight depth (or equivalently, number of comparable features projected along the line of sight). To put this on a concrete basis, we use the slope of the relation between U and [O iii]/Hβ obtained for an AGN spectrum by Netzer (1990), scaled to [O iii]/Hα with an intrinsic Balmer decrement of 2.9.

The [O iii]/Hα line-ratio map (Figure 12) shows an interesting variety of structures. There are small, discrete regions of low [O iii]/Hα, associated with the continuum objects in the northern part of the Voorwerp. We identify these as star-forming regions, possibly triggered by compression of the gas (Section 5.3). Beyond this, the excitation is not well correlated with emission-line structures in the object (Figure 13); the regions plotted are identified in Figure 14. Ionization level is not correlated with Hα SB (which is driven by density), [O iii] SB (which combines ionization parameter and density), location within a filament center to edge, or location near the "hole." The highest line ratio values occur in several broad, roughly parallel stripes crossing the object, at a skew angle to the projected direction of the IC 2497 nucleus.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In the idealized case of a matter-bounded nebula, the SB in a recombination line (such as the Balmer series) from a given volume follows SB ∝n_e², while the ionization parameter U∝n_e⁻¹. If we are seeing a collection of isolated and internally uniform clouds or filaments, we would expect to see U∝SB^−1/2. Pixel by pixel, we see a much weaker, marginal average relation. At least one of these simplifying assumptions does not apply. The relative strength of [O i] suggests that there are denser regions within the gas, since this transition is favored in partially ionized zones where O⁰ and H⁺ coexist. One way to account for the decoupling of SB and U would be for the gas to have spatially unresolved fine structure (on scales smaller than the ACS resolution limit of ≈100 pc), so that the SB we observe represents more closely the number of clouds and filaments in a particular direction than their mean particle density. These might be analogous to the magnetically confined structures inferred by Fabian et al. (2008); since we find that the outflow from IC 2497 has played only a minor role in the structure of the gas we see, an implication would be that the rest of the H i tail is similarly structured, independent of action of the AGN. This fine structure means that we cannot detect any differential recombination as the ionizing source fades; if a region with a density gradient were spatially resolved, we would expect to see species with faster recombination timescales, such as O⁺, fade faster than longer-lived ones such as H⁺ (Binette & Robinson 1987). Since the detailed structure, in particular lack of filaments radial to IC 2497 in most of the object, indicates that reshaping of the gas by the galaxy's radio outflow has been modest, this suggests that the rest of the H i tail in this system is similarly filamentary. Gas in tidal streams may retain fine structure far beyond the spatial resolution of current data.

These stripes of higher excitation (and almost certainly higher ionization) have projected widths of 07–12 (700–1200 pc). In the ionization-echo picture, these could represent periods of higher ionizing flux and would be significantly smeared by the recombination timescale t_rec ≈ (αn_e)⁻¹ for recombination coefficient α. As noted above, the emission-line limit on density n_e < 50 cm⁻² sets this timescale to be 2400 years or more, scaling with n⁻¹_e (Lintott et al. 2009). This light-travel time ct_rec translates to 07 or more spatially, so that the fine structure we see must be dominated by structure in the gas rather than ionization history, and the filaments must be physically about as thick as they appear. This fits with the H i column density if the ionized region is matter-bounded and not optically thick in the Lyman continuum; thus the lower bound on ionizing luminosity from recombination and energy budget from Lintott et al. (2009) should indeed be higher.

If these stripes do represent ionizing outbursts, they suggest a geometry for the gas—in the light-echo geometry, the intersection of the light-time ellipsoid with the ionized cloud is skew from the direction to the nucleus when the surface is tilted both to our line of sight and to the radius vector toward the ionizing source.

We have little direct information on the three-dimensional geometry of the Voorwerp. The small number of distinct filaments may indicate that it is geometrically thin, rather than being only a thin "skin" of ionized gas on the inner edge of a much thicker H i structure. Since the dominant process in its excitation is photoionization rather than shocks (Lintott et al. 2009), we expect there to have been little mass motion associated with its production, so that if it is thin, so is the H i stream. The mean H i column density in this region is (8 ± 2) × 10¹⁹ atoms cm⁻² (Józsa et al. 2009). The depth into which an external UV source would ionize this cloud depends on both the local density and duration of exposure, which are poorly constrained. Emission-line diagnostics (Lintott et al. 2009) give an upper limit to the density from the [S ii] line ratio, and energy balance shows that at least one-fourth of the impinging ionizing photons are absorbed.

Where does the Voorwerp lie in distance with respect to IC 2497? Escaping radiation would be more likely near the poles of IC 2497, in view of the dusty disk features seen in our images, and a minimum distance would give the most conservative ionizing luminosity. These factors suggest a somewhat longer time delay than the projected separation. If the ionized region is polar to IC 2497, its separation from the nucleus would make an angle θ ≈ 125° to the line of sight (about 35° "behind" the plane of the sky). The true distance between the nucleus of IC 2497 and the Voorwerp would then be r_proj/sin θ. The difference in light-travel time would then be

$$\begin{equation} \Delta t = {{r_{{\rm proj}}} \over {c \sin \theta }} (1 - \cos \theta) \end{equation} \tag{ 1 }$$

(Keel et al. 2012). For θ = 125°, this is greater than the plane-of-the-sky component of time delay r_proj/c by a factor of 1.9, spanning 97,000–230,000 years from the inner to outer regions of [O iii] emission. It is possible that the entire emission-line structure forms a fairly thin, wrinkled sheet roughly perpendicular to the incoming photons; in this case, the stripes of highest ionization level might record the same peak in core luminosity. However, this geometry would require systematic changes in characteristic density of the (unresolved) emitting structures to fit with the changes in SB behavior seen across the Voorwerp (Section 5.2.1). It may therefore be more likely that the ionization pattern records a complex luminosity history of the AGN.

The ionization level and temperature of the gas in Hanny's Voorwerp require the dominant ionization mechanism to be photoionization by a continuum extending well into the far-UV. However, we find evidence of star formation in a few isolated regions both from continuum and emission-line properties.

In all three broad bands (F225W, F814W, and F160W) there are several continuum sources embedded in the small bright northern part of the Voorwerp, nearest IC 2497. Each is associated with an area of very low [O iii]/Hα (although there are other nearby regions of similar line ratio without an obvious embedded continuum object); both lines show local maxima at their locations. The brightest of these are detected in the mid-UV image. These are spatially resolved in images including line emission, but correcting the F814W image for this (using the Hα image as a template for the Paschen continuum and weak emission lines) shows the brightest to be unresolved (Figure 15). This nominally pure-continuum image suggests that additional continuum objects are present within this single 2'' region at lower S/N. The resolution limit is roughly 2 F814W pixels in WFC3, or 008 (80 pc). The associated Hα emission regions have been investigated using the surrounding [O iii]/Hα values to correct the Hα image for AGN-ionized gas. These Hα regions have FWHM = 4.3–4.9 pixels, or 3.8–4.5 pixels (150–180 pc) after making a Gaussian correction for instrumental resolution.

Zoom In Zoom Out Reset image size

Table 4 presents aperture photometry (within a matched radius of 038) for the three brightest regions, noting that they exhibit various degrees of substructure down to the resolution limit. The colors are very similar, especially I₈₁₄ − H₁₆₀ where the errors are small. They also lie close to the continuum shape for a young stellar population. Figure 16 compares their broadband SEDs to the predictions of the Starburst99 models (Leitherer et al. 1999). The continuum slope we observe allows virtually any age (for either a short burst or ongoing star formation) up to ≈2 × 10⁸ years. The Hα emission associated with these star-forming regions shows that star formation continues at a significant level, favoring either smaller ages or extended star-forming timescales. This uncertainty translates into very poor constraints on the total stellar mass formed.

Zoom In Zoom Out Reset image size

We see no comparable blue objects elsewhere in the field outside IC 2497; no luminous star-forming regions or bright stellar clusters of greater age appear in the Voorwerp or in the region we cover of the H i tail. The star formation is uniquely associated with a small region, at least approximately aligned with the small-scale jet and larger outflow seen in radio observations as well as the emission-line "fingers."

We note a very red object at the western end of this brightest region, briefly mentioned by Lintott et al. (2009). Unlike the other continuum structures, it is not associated with an emission-line peak (or any detected emission-line structure). Several additional objects with similar extent and color are seen in both F814W and F160W, possibly background galaxies.

In contrast to Minkowski's Object, the overall ionization here is maintained by an AGN rather than the local stellar populations; only with the spatial resolution of HST have we been able to distinguish their effects. The star formation, although less dominant energetically, may have shared the same general origin; as discussed by Fragile et al. (2004) and Croft et al. (2006) for Minkowski's Object, the external pressure of a radio jet can compress pre-existing material, even if hot and at low density, strongly enough to drive cooling, fragmentation, and presumably star formation. The H i cloud around IC 2497 would provide a rich target for such a process, cooler than envisioned in their calculations and thus needing less compression to begin collapse. In having such a pre-existing reservoir of cool H i, the situation here may be more like that in Centaurus A, where the jet impinges on dense clouds. The distinction has been stressed by Croft et al. (2006), noting different morphological relations between the young stars and H i in Cen A and Minkowski's Object.

The level of star formation we find in Hanny's Voorwerp is modest both in comparison to IC 2497 and to such environments as Minkowski's Object. Making an approximate allowance for fainter Hα regions beyond the apertures in Table 4 adding as much as an additional 30%, the Hα luminosity of these star-forming regions is in the range 0.9–1.4 × 10⁴⁰ erg s⁻¹, or 14%–21% of the value in Minkowski's Object (Croft et al. 2006). We do note that additional star formation, if not collected into clusters, could be hidden in the diffuse UV emission from the gas; the clusters account for only a few percent of the integrated flux found by Swift and GALEX, which is predominantly continuum. This continuum certainly has a substantial contribution in this band from nebular processes and might also include scattered AGN light from embedded dust.