JSTOR

2.1. ISO 6.7 m Data

The 53W002 field was observed with ISOCAM (Cesarsky et al. 1996) on orbits 500 and 521. The LW2 filter, with half‐power transmission points at 5.0 and 8.5 μm and an effective wavelength of 6.7 μm for a flat spectrum, was used to approximate the emitted K band at (the effective emitted‐frame wavelength actually corresponds to 1.97 μm). Since we were interested in 11 objects in the 53W002 group (including three foreground and background Lyα emitters at ), a trade‐off had to be made between covering a wide enough field and high point source sensitivity. We used the 3 pixel⁻¹ scale, undersampling the LW2 point‐spread function (PSF) in the interest of covering a larger region at once. The final spatial resolution will be somewhat worse than this, since data from many offset pointings are stacked. Two levels of pointing offsets were used for these data. At each of 21 positions, sets of s exposures were done, with the pattern using offsets of 5 pixels (15 ). A set of 21 of these pointings was laid out to cover the area spanning most of the known objects around 53W002, with 45 spacing in both celestial coordinates. To provide scheduling flexibility while also assuring that the right regions were observed, these locations were specified in celestial rather than spacecraft coordinates, resulting in a detector orientation that is skewed to the pointing grid (by about 109° for orbit 500, and 129° for orbit 521, with rotations of 0 25 or less within each orbit). The overlap of these observations is shown in Figure 1. To minimize the time used in microslews between these positions, they were observed in a boustrophedonic order. For clarity in describing the reductions, we will refer to each set of exposures as a raster.

(281 KB)

Fig. 1.— Overlap pattern of individual LW2 rasters superimposed over a version of processed 6.7 μm mosaic. No masking of raster borders was done, so as to preserve the full field for this illustration, and background variations have not been removed. The numbers denote observation sequence numbers: 931–947 are from orbit 500, while 703–729 were observed on orbit 521. The circled cross shows the location of 53W002. The low‐redshift system 53W003 is prominent to the northeast (upper left).

Open New Window

Before each raster observation, a brief exposure was made using the LW1 filter, since experience had suggested that the LW detector response to transients could be improved by a change in the background level. The total exposure time in LW2 was 19,051 s (5.3 hr), of which a typical useful total at a given sky position is 13,000 s in the inner part of the field. The region of best exposure is roughly circular, with a radius of 90 , centered at α = 17^h14^m13 0, δ = +50°15 57 (J2000.0). There is considerable structure at the 10% level in the exposure pattern, even within this region, but this amplitude does not strongly change the detection limit. In comparison with other ISO surveys at this wavelength, this study is thus substantially deeper than the Canada‐France Redshift Survey (CFRS) field examined by Flores et al. (1999), it is comparable in exposure to the LW2 data on the Hubble Deep Field (HDF) presented by Serjeant et al. (1997), and about 3 times shorter than the very deep SSA13 and Lockman Hole observations by Taniguchi et al. (2001).

Reduction of these data began with the ISOCAM CIA package (ver. 4.0, 2000 April release; Delaney 2000). Its routines were used to clean the data stream of incomplete and otherwise invalid exposures, correct for the dark pattern by standard library frames and deglitch and stabilize the individual exposures. Deglitching, which corrects for long‐lasting radiation‐induced events, was found to be most effective using the multiresolution median algorithm in CIA. The stabilization step corrects for the substantial time constant of the detector created when responding to a change in illumination (such as happens at each change of pointing). The data for each raster were combined to form a single mosaic, rejecting pixels masked as uncorrectable in the earlier stage, and each mosaic was projected onto a common celestial‐coordinate frame and expanded to 1 0 pixels using the C routine project, distributed with the CIA package. This expansion preserved the intensity‐flux calibration in millijansky per solid angle. The number of glitches not rejected by these techniques was still substantial, so we carried out further processing and masking using the combine and imedit tasks within IRAF.

Residual flat‐field structures were apparent in the raster products, so these were corrected using a median frame for each orbit’s data. Drifts in background level between rasters were also removed. Glitches that remained after the CIA processing were identified by the comparison of spatially overlapping rasters and were masked before combining. The best results were found by masking the outer parts of each raster as well, leaving those regions that were included in at least six of the nine constituent observations (a 111 arcsec² region in each case). The final data mosaic combined all these masked raster products with 3 σ rejection. The effective resolution of this mosaic should be slightly better than 6 FWHM, a combination of the input 3 pixel⁻¹ scale and stacking of variously offset observations. This value varies significantly, depending on a source’s exact location with respect to the observed pixel grid; some star images have a FWHM as small as 4 .

The mosaicked image still shows residual variations in blank‐sky level that are associated with slight slope changes from one observation to the next. We removed these using a running median filter, which substantially improved the image uniformity and reliability of aperture fluxes while not altering image structure on relevant scales for our small targets. The detection threshold was assessed through statistics of individual pixel values and by the statistics of detected positive and negative peaks after smoothing with a Gaussian effective PSF with 5 FWHM. Nine ISO sources are robust to the details of selection aperture and measuring radius within the 4 –8 range, and to whether single discrepant pixels are rejected or not. The effective 3 σ flux limit across the central high‐exposure area is about 0.027 mJy, below which the non‐Gaussian noise distribution prevents us from saying anything useful about the number of sources. We also estimated the mean noise by photometry with the same 5 radius on a set of (blank) positions derived from the object locations by 10 –20 offsets in each coordinate. The mean 1 σ error found in this way is 0.010 mJy, in good agreement with the estimate from statistics of positive and negative fluctuations.

The nominal astrometric positioning of ISOCAM frames is known to change with motions of the lens wheel (Blommaert et al. 2003, § 4.11), typically by 2 to 3 pixels. The only independent way for us to check the coordinate system is through the bright interacting galaxy 53W003 at , near the north edge of the mosaic. The best match between its optical and mid‐IR structures comes from a 6 shift in each coordinate such that the actual coordinates of an object fall southeast of the values given by the nominal ISOCAM pointing. This offset also aligns two much fainter sources with relatively bright red stars, and two known galaxies with their optical locations, so we adopt this astrometric location. The mean offsets, in the sense of corrections to be applied to the nominal ISO header coordinates, are and Δ , which have been applied to the listed source positions. The possibility remains of a systematic shift between the two orbits’ data, which we can limit by noting that substacks of the region of 53W002 itself, where the two data sets overlap, both recover consistent positions of ISO sources (within the limits set by their flux levels). Any differential offset between the two sets is too small to be detected with such a faint target. This also limits any possible offsets associated with lens‐wheel rotations, which accompanied some of the interspersed LW1 “reset” exposures.

We carried out tests for the reality of possible sources in this field at various stages in the processing, because of the large number of transient events and the availability of many redundant measurements at most locations of interest. Possible bright sources were examined in each raster that included their locations, and any interfering glitches were masked before further combination of the rasters. The sources remaining in the final mosaic image (Fig. 2) pass these straightforward tests. For 53W002 itself, we examined various subsets of the data, none of which shows any problem that could in itself mimic a source detection.

(314 KB)

Fig. 2.— Final mosaicked 6.7 μm image of the 53W002 field, incorporating masking of the edges of individual rasters to reduce artifacts in the clipped mean and median filtering of sky variations though an window. The rapid increase in noise to the edge of the mosaic outside the central 90 radius is still apparent. The maximum extent of the exposed region is 347 north‐south and 354 east‐west. The prominent elongated source at upper left is the low‐redshift merging system 53W003.

Open New Window

For evaluating source counts, the most conservative interpretation is that the LW2 sources with plausible optical, radio, or submillimeter counterparts are real, and an upper bound is that all the statistically significant (>3 σ) sources are real. Given the behavior of the detectors and radiation environment, it is hard to pin them down more tightly, so we will consider both extremes in counting the overall source population.

The list of detected sources is given in Table 1. Only three bright sources are listed from outside the 90 radius of full exposure: 53W003, a red star, and one additional possible star (object 2), listed at the top of the table. The remainder are from the central area of deep exposure and form a 3 σ cut, as determined by taking the strongest peaks after convolution with a Gaussian function of 5 FWHM. One source is a good match for the position of a moderate‐redshift spiral galaxy that is within the WFPC2 frames of Pascarelle et al. (1996).

TABLE 1 Source Detections in LW2 Field

Open New Window

The ISO detectors' gain varied with such factors as the history of energetic particle flux. Accordingly, we have checked the flux calibration for consistency with stellar atmosphere models, as was done by the HDF‐South survey as well (Oliver et al. 2002). In this case, we used catalogs of stellar magnitudes from B through K, taken from images at Palomar (Neuschaefer & Windhorst 1995), Kitt Peak (Keel et al. 1999), and the NASA Infrared Telescope Facility (§ 2.2). These were supplemented by B and R magnitudes from the USNO A‐2 catalog of the Palomar Sky Survey (Monet 1998)5 for the brightest stars, which were saturated in the deep CCD images. The optical and near‐IR fluxes were used to predict a flux at 6.7 μm for comparison with the derived ISO values. Nondetections are important (in addition to detections) in that they verify the detection threshold. Rare stars with dust shells will appear too bright in the mid‐IR, compared to the photospheric prediction, but for a group of stars, this should not change the derived flux limit. As it happened, the red star to the northeast of 53W002 (number 6 in Table 1 and Fig. 3) was especially important, having a color‐derived spectral type of M2 V. The stellar data are consistent with the nominal ISO LW2 flux scale, so we had no need for any correction, adopting source fluxes “as is.” This comparison limits any systematic variation in the ISO flux scale to the ±30% level.

(153 KB)

Fig. 3.— Sources in the 53W002 field, as shown on a Gaussian‐smoothed LW2 image (left, smoothed to 5 FWHM), and a blue‐light image (right, with 4100 Å data from Keel et al. 1999) resampled to the same scale. Sources listed in Table 1 are marked. Number 1 is 53W003, and number 10 is at the position of 53W002 at .

Open New Window

2.2. Supporting Data

We have made use of previously reported data from the ground and space, including WFPC2 images from Pascarelle et al. (1996, 1998), NICMOS images in F110W and F160W filters at 1.1 and 1.6 μm (Keel et al. 2002), wide‐field optical B and V imagery from the Kitt Peak National Observatory (KPNO) 4 m prime‐focus camera (Keel et al. 1999), and deep‐K and narrowband (redshifted Hα) imagery of the immediate environs of 53W002 (Keel et al. 2002).

An additional, shallower K‐band mosaic of the entire full‐exposure ISO field was obtained during one of the same NASA Infrared Telescope Facility (IRTF) observing sessions. The NSFCam system was used at 0 3 pixel⁻¹, for a typical exposure of 540 s for regions outside the deep central observations. The region was covered with nine subrasters using 12 offsets. In many cases, the overlap regions of these subrasters do not include bright enough objects to register them from the K data, so we stacked the images using relative astrometry from optical KPNO CCD images. The intensity scale was set from observations of UKIRT faint standard stars (Hawarden et al. 2001).

A collective mean‐flux limit was set to the submillimeter brightness of the compact objects in the 53W002 grouping by coadding the archival Submillimeter Common‐User Bolometric Array (SCUBA) data, as discussed in § 3.3.

3.1. The Radio Galaxy 53W002

The ISO data suggest a detection of the radio galaxy 53W002 itself at the ∼3 σ level. Depending on the aperture size and details of background subtraction, the derived flux is 0.024–0.039 mJy, or ergs cm⁻² s⁻¹ Å⁻¹. We include a term for error due to “sky” fluctuations as noted in § 2.1, added in quadrature. This is slightly conservative, in the sense that the error range of 0.024–0.039 mJy includes some of the effects of noise in the sky annulus. Thus, we adopt a measurement of ergs cm⁻² s⁻¹ Å⁻¹. This furnishes a constraint on the continuum slope, and therefore age, of the dominant stellar population. The central AGN contributes a significant amount of light at shorter wavelengths, both in emission lines and from its continuum, which can be estimated from spectroscopy and Hubble Space Telescope (HST) image analysis. Concentrating on the longer wavelengths, which are less affected by the most recent star‐forming history, we adopt starlight fractions of 0.77 at F110W, 0.62 at F160W, and 0.78 in K (from PSF fitting in WFPC2 and NICMOS images by Windhorst et al. [1998] and Keel et al. [2002], and consistent with the spectroscopic estimate of an optical spectral index for the AGN by Windhorst et al. [1991]). This correction is small enough not to add to the color error, dominated by the ISO point. The starlight flux gives a ratio in of between emitted wavelengths of 0.65 and 1.97 μm. This translates to emitted , or, adding the flat continuum shape from Keel et al. (2002), .

The emission‐line equivalent widths and image decomposition of the HST data indicate that light from the active nucleus is a significant contributor in the emitted‐frame UV and optical bands (Windhorst et al. 1991, 1998). We thus examine whether or not the AGN contribution (perhaps reddened) might add to the observed ISO flux. Line emission should contribute only a small fraction of the ISO detection, based on scaling from the observed Lyα, Hα, and [O iii] emission. At , the response of the LW2 filter extends from 1.47 to 2.50 μm at half‐peak transmission. This band contains spectral features that include Paα at 1.87 μm, and also Brγ at 2.16 μm and the adjacent H₂ S(1) line, which is comparably strong in some galaxy contexts. Using the constants in Osterbrock (1989, Table 4.4) for low densities at 10⁴ K, the theoretical ratios to Hα are 0.12 for Paα and 0.03 for Brγ. The IRTF data in Keel et al. (2002) give a total intensity for Hα (including some contribution from [N ii]) of ergs cm⁻² s⁻¹. To convert the prediction from scaling based on Hα intensity to an in‐band LW2 flux, we take a simple trapezoidal approximation of the passband shape, giving an approximate detected flux of ergs cm⁻² s⁻¹ in the LW2 passband. Of this, the strong hydrogen recombination lines would contribute at most about 15% so that within these error bars we can safely neglect this correction.

There is a foreground elliptical galaxy at that appears about 7 northwest of 53W002, potentially confusing the detection of 53W002. This is the galaxy labeled number 6 by Windhorst et al. (1991, 1994), not to be confused with object 6 of Pascarelle et al. (1996) and later papers (which is 53W002 itself). We considered whether blending with this object might increase the measured flux of 53W002 in the ISO data, but its effect must be unimportant, for two reasons. The statistics of the data stack allow the possibility of two sources at such a separation, but this would be inconsistent with the astrometry of other sources nearby. In particular, the stellar identification to the northwest of 53W002 would become a blank field. This is important, because this object is so close that it would share whatever detailed pointing errors apply to the area around 53W002 itself. The star image is close to the expected PSF for a point source, which implies that 53W002 (located 85 away) did not suffer pointing errors between rasters, which could degrade the image size. Furthermore, the LW2 passband falls well on the long‐wavelength side of the stellar emission for even old populations at moderate redshifts; we see this galaxy near 4.2 μm in its emitted frame, so the predicted flux is a small fraction of the observed value for 53W002.

Figure 4 shows the broadband photometry of 53W002 compared to population synthesis models from Worthey (1994) for ages 1.0–2.0 Gyr after allowance is made for the contribution of the central AGN, following Windhorst et al. (1998) and Keel et al. (2002). Normalized to the near‐IR points that span the emitted optical range 4700–6600 Å, these models indicate that the error range on the ISO measurements spans the range 0.8–2.1 Gyr in single‐burst age, with the error range centered near 1.5 Gyr. These values indicate a “formation redshift” for the onset of widespread star formation , with the error range centered at . We intentionally neglect reddening and any contribution of the AGN at 6.7 μm, in order to get a maximum possible stellar age. Even this measurement, of quite modest precision, narrows the epoch of widespread star formation in 53W002 to the redshift range over which galaxies are now detected. Age estimates from mid‐IR data should be more robust than estimates from shorter wavelengths, which suffer from the bias toward the youngest populations as observed in the emitted UV, and the potential role of dust associated with the molecular gas (Scoville et al. 1997). Whether this advantage is realized in practice depends on the accuracy of the mid‐IR data. We also note that there is lingering uncertainty over the near‐IR role of asymptotic giant branch (AGB) stars in the 0.2–2 Gyr age range (as highlighted by Maraston 2004), for which synthesis libraries are still poor in empirical tests.

(62 KB)

Fig. 4.— Infrared spectral energy distribution of 53W002 compared to model stellar populations. The near‐IR points, which map to the emitted optical near the B and R bands, are the NICMOS and IRTF data from Keel et al. (2002), incorporating correction for the central AGN. The model stellar populations are from the Worthey (1994) code as implemented on the World Wide Web. The error range for the ISO point spans ages of 1.0–2.0 Gyr, corresponding to redshifts at the onset of significant star formation , with a central value of .

Open New Window

Significant star formation in 53W002 took place over a span of at least 1 Gyr. The emitted‐optical spectrum from Motohara et al. (2001) shows a substantial Balmer jump, indicating a stellar population less than about yr old; a more complete, self‐consistent model would be usefully constrained only by much smaller errors in the mid‐IR data (as expected from forthcoming Spitzer Space Telescope results). Thus, we find that the stars in 53W002 were formed over the cosmically brief but dynamically significant span of 1–2 Gyr. In keeping with the general picture of more massive objects beginning star formation first and being able to bind their own enriched gas, 53W002 was the first luminous system to begin star formation in this group, in contrast to the lower luminosity Lyα emitters (§ 3.3).

3.2. Nondetection of a Reddened AGN in a Submillimeter‐bright Object

Among the five known active nuclei in the 53W002 grouping, one is also identified as a submillimeter‐bright source. Smail et al. (2003) have combined SCUBA mapping with Very Large Array (VLA) continuum data at 1.4 GHz to identify object 18 of Pascarelle et al. (1996) with the submillimeter source, implying a bolometric luminosity of solar luminosities. The origin of this energy output is not clear; all the properties of this object observed in the optical and near‐IR are traceable to a luminous, largely obscured active nucleus and its surrounding emission and reflection nebulae. This interpretation is strengthened by the Chandra spectrum, which shows a heavily absorbed power‐law form (White et al. 2004). This object has an especially extensive Lyα halo whose ionization is most easily explained if our line of sight suffers unusually high extinction (so that the UV continuum that we do observe would be scattered light). The ISO nondetection of this object shows that any such luminous central source must be quite strongly obscured. It must be fainter than 53W002 itself at 6.7 μm, which requires an emitted‐frame color . Such a blue limit implies that any obscured nucleus does not yet dominate the light at 2 μm. A typical quasi‐stellar object continuum must be reddened by to have this color in , and the reddening must be substantially greater in this case, because the continuum at shorter wavelengths is quite blue (as shown by Keel et al. 2002) and thus cannot be reddened AGN light (although scattered light could play a role). The resolved continuum from NICMOS observations also indicates that little direct AGN light emerges in our direction in the μm emitted range. Given the uncertainties in separating any direct starlight from side effects of the AGN, we can limit direct light from the AGN at these shorter wavelengths to less than about 20% of the total.

3.3. Narrow‐Line Lyα Star‐Forming Objects

We set a mean limit on the 6.7 μm flux of the faint blue Lyα emitters in this field by adding regions around each of those registered on the optical positions. The objects stacked were numbers 5, 11, 12, 29, 34, 60, and 113 from Pascarelle et al. (1996). Of these, 5 and 12 are at , 94 is at , and the rest are at (Pascarelle et al. 2004). This mean image gives a flux of formally less than 0.010 mJy, which is slightly better than the noise statistics should allow. The 1 σ noise level for 10 apertures averaged over seven objects gives 0.003 mJy, and the 5 diameter effective apertures used for these unresolved objects would give group mean 1 σ levels close to 0.004 mJy. We therefore take the more conservative mJy value, since the measured formal flux might be unusually low by small‐number statistics. This means that on average, they are at least 3 times fainter than 53W002 itself. The mean spectral slope can be evaluated against observed wavelengths near 1.6 μm, by using NICMOS data (Keel et al. 2002) and the sample mean flux of ergs cm⁻² s⁻¹ Å⁻¹. The mean flux ratio between emitted wavelengths 0.47 and 2.0 μm is then greater than 2.5, implying . This means that these objects could, in principle, have stellar populations as old, and have formed over a span as long, as we infer for 53W002. A limit to dust emission from ongoing star formation can be set from the coadded SCUBA data likewise stacked at the positions of these objects. The formal mean 850 μm flux is mJy. For objects with a spectral energy distribution like that of Arp 220, this flux (0.14 mJy) at would correspond to a total far‐IR luminosity of ∼ solar luminosities, or a mean star formation rate of about 30 M_⊙ per year. This is a few times lower than that implied by the typical mid‐UV continua of these objects (using the star formation form from eq. [1] of Kennicutt 1998), which may attest to the nearly dust‐free nature of the star‐forming regions as seen in the very blue UV‐optical colors (Keel et al. 2002).

In themselves, then, these data do not add to the case that the compact Lyα emitters arise from short (perhaps repeated) bursts, either in small “protogalactic” systems (Pascarelle et al. 1996, 1998; Keel et al. 2002) or as pieces of larger systems whose typical surface brightness is below current detection thresholds (Colley et al. 1996).