6. Results on the ISO-HDF
Figs. 9 and 10 present the reduced map of the ISO-HDF LW2 and LW3. Many sources are clearly visible on the latter, while the former appears almost empty. Various sensitivity levels have been overlayed on the maps, showing that the noise is very inhomogeneous. These sensitivity levels were computed by comparing the noise level at each scale with the wavelet coefficients of a point source image, the source being centered on a pixel. Thus, these levels give only an indication. If a source fell between two pixels, its wavelet coefficients at lower scales (high spatial frequencies) is lower, while those at higher scales (low spatial frequencies) are higher. Since the noise decreases when considering higher scales (Starck et al. 1998), sources can in fact be detected at fluxes lower that the levels shown on Figs. 9 and 10.
6.1. Source catalog
Table 3 gives the catalog of detected sources at 6.7 µm (LW2 filter) and 15 µm (LW3) above a secure of photon and readout noise. The first column is our ISO-HDF identificator, the second and third column are the right ascension and declination of the object. The pointing accuracy is of 1 pixel on the final map, i.e. 3" for the LW3 image and 1.5" for the LW2 image. Columns 4 and 5 give the objects fluxes in the LW2 band in µJy and in ADU/G/s (analog-to-digital unit per second divided by the instrument gain), and columns 7 and 8 give the same in the LW3 band. Columns 6 and 9 give the detection level in the LW2 (resp. LW3) band. Column 10 identifies the field where the object is observed (HDF or flanking field). Column 11 gives the spectrometric redshift when available. Finally, column 12 give the morphological type of the object.
Table 3. PRETI main source list for the ISOHDF observations at m (LW2) and m (LW3). All sources are detected at above readout and photon noise. (1): Source identification `PM3' is for PRETI Main list LW3, `PM2' for PRETI Main list LW2. (2), (3): source position for the J2000.0 equinox. (4): source flux in microJansky in the LW2 filter with error bar or upper limit (). A `-' indicates that the source is not in the LW2 field. (5): source flux in ADU/G/s in a 3" aperture. (6): detection level () in the LW2 image. (7): source flux in microJansky in the LW3 filter with error bar or upper limit (). (8): source flux in ADU/G/s in a 6" aperture. (9): detection level () in the LW3 image. (10): Identificator of the HST field in which the source falls (hd stands for HDF, others for flanking fields). (11): spectroscopic redshift with source indicated by a note. (c): data from Cohen et al. (1996), (p): data from Phillips et al.(1997), (i): compiled in Cowie et al. (1997), (h) data from Hogg et al. (1977b). (12): morphological type of the object. `(i)' indicates a type compiled in Cowie et al. (1997), `(p)' type from Phillips et al. (1997), otherwise determinated by us. E: Elliptical or S0 galaxy. S: spiral. M: merger. P: peculiar/irregular, C: compact object from Phillips et al. (1997) sample, A: active nuclei galaxy, G: source falls on a group. U: unknown type. N: no optical counterpart. A `?' indicate a dubious type identification.
The redshift and morphological type of each ISOCAM source depend on the identification of its optical counterpart. Hopefully, our astrometric correction is good enough to allow us a good superposition with optical images, at a scale of 1 pixel. This is clearly visible on Figs. 11 and 12 where we have overlayed our ISOCAM maps on the KPNO IRIM deep K exposure 2 (Dickinson et al. in prep), as well as in Fig. 13 where we have overlayed the LW3 ISOCAM map and the LW2 detections on the color composite image of the HDF observed by HST from Williams et al. (1996). In most cases, the ISOCAM source matches an IR or optical source, and the identification is easy. However, due to the large PSF (15" at 15µm) of ISOCAM, sometimes more than one object could be associated with our detection. Fig. 14 displays an example of such a case: the main peak of source HDF_PM3_6 correspond to two objects. In this particular case, we have taken the brightest object in K as counterpart. When K data are not available, we have used the `G' (for group) symbol in our table, because we could not determine the right counterpart.
Looking closely at Fig. 14, one can see that the object HDF_PM3_6, detected as a whole, displays substructures. Our detection program does not separate the various components of the source because it is not robust against blending: to avoid detection of cosmic ray residuals, we choose to perform the detection starting from the second scale of the wavelet transform of our map, i.e. on scales of the order of 3 pixels that prevent us to separate very closeby sources, especially if they are much weaker than the main detection. However, in some cases, this separation can be done by hand, when blending is weak enough to allow for the separation of the source (Rayleigh criteria). This is the case for source HDF_PM3_6, for which 4 extensions are visible on Fig. 14, South and East of the main peak. These extensions are in our supplementary list (see Table 4) with the names HDF_PS3_6a, HDF_PS3_6b, etc... . Fig. 15 displays another case where the separation is possible, for source HDF_PM3_17. Fig. 16 displays a closeup on source HDF_PM3_27: the source is extended, with multiple counterparts, but we failed to separate the various contributions. We note however that this source is the radio source 3639+1313 in the catalog of Fomalont at al. (1997), whose contours in the radio map (Richards et al. 1997) are also clearly extended.
Table 4. PRETI supplementary source list for the ISOHDF observations at m (LW2) and m (LW3). All LW3 sources are at least detected at above readout and photon noise, while all LW2 sources are detected at least above . (1): Source identification `PS3' is for PRETI supplementary list LW3 and `PS2' for LW2. Sources with letter at the end, like `6a' are deblended neighbours of the source with same number in the main list. (2), (3): source position for the J2000.0 equinox. (4): source flux in microJansky in the LW2 filter with error bar or upper limit (). A `-' indicates that the source is not in the LW2 field. (5): source flux in ADU/G/s in a 3" aperture. (6): detection level () in the LW2 image.(7): source flux in microJansky in the LW3 filter with error bar or upper limit (). (8): source flux in ADU/G/s in a 6" aperture. (9): detection level () in the LW3 image. (10): Identificator of the HST field in which the source falls (hd stands for HDF, others for flanking fields, `-' means outside of all HST fields). (11): spectroscopic redshift with source incated by a note. (c): data from Cohen et al. (1996), (p): data from Phillips et al. (1997), (i): compiled in Cowie et al. (1997), (h) data from Hogg et al. (1977b). (12): morphological type of the object. `(i)' indicates a type compiled in Cowie et al. (1997), `(p)' type from Phillips et al. (1997), otherwise determinated by us. E: Elliptical or S0 galaxy. S: spiral. M: merger. P: peculiar/irregular, C: compact object from Phillips et al. (1997) sample, A: active nuclei galaxy, G: source falls on a group. U: unknown type. N: no optical counterpart. K: counterpart only in IRIM K image. st: star-like object. A `?' indicates a dubious type identification.
We produced a supplementary list of sources with these extensions of reliable sources, together with results of PRETI at lower detection levels (6 and 5 ), that is listed in Table 4. However, this catalog may contain spurious detections, as is certainly the case for sources HDF_PS3_13 and HDF_PS3_29, with fluxes below 10 µJy. In the future, we expect to obtain data of a long staring observation on an empty field. Using this observation as a basis for simulations, we hope to be able to obtain higher reliability on the sources at these low level of flux.
6.2. Catalog properties
Our main source list consist in 49 objects detected above a detection threshold in both filters. 42 sources are detected at 15 µm only, 4 at 6.5 and 15 µm, and 3 at 6.5 µm only. Since the LW3 field is larger than the LW2 field, this last number means that 21 of the LW3 sources are not detected in LW2. The supplementary list adds 51 sources to this catalog for a total of 100 objects. 47 are observed in the LW3 filter only, and 4 in the LW2 filter only. None are detected in both filters, 15 of the LW3 sources that could be visible in the LW2 field remain undetected. For these missed detection, we have computed upper limit as follows: a test image containing only the source is computed by projecting a point source as observed at each raster pointing in a copy of the final raster map. This ensures that the modification to the PSF produced by microscan are taken into account. The b-spline wavelet transform of this image is then calculated and compared to 5 times the noise map at each scale. The minimum flux ensuring the detection is then converted in microJansky, using our calibration from simulations.
All sources from the main list within the HDF have optical counterparts, thanks to the depth of the HST map, but 6 sources within the flanking fields do not have visible counterparts in the much less deep F814W images. However, these 6 sources are very reliable and were also found by other teams (see below). The number of sources without counterpart rises to 11 in the supplementary list, of which 1 lays in the HDF. Source HDF_PS3_6e does not have any counterpart in the `iw' HST image, but has a counterpart in the IRIM K image. This example shows that the reliability of the source should not be established on the existence of counterparts in optical images only, especially in the flanking fields. It is remarkable that all but one (3652+1444) of the radio sources observed at 8.4 GHz by Fomalont et al. (1997) in the field of view, are detected, mostly in the LW3 filter (Table 6), and one in the LW2 filter only (Table 5).
Table 5. Comparison of the LW2 catalog obtained here with PRETI, with the catalogs by IC (Goldschmidt et al. 1997) and IAS (Désert et al. 1998), as well as radio sources detected at 8.4 GHz (Fomalont et al. 1997)
Table 6. Comparison of the LW3 catalog obtained here with PRETI, with the catalogs by IC (Goldschmidt et al. 1997) and IAS (Désert et al. 1998), as well as radio sources detected at 8.4 GHz (Fomalont et al. 1997)
Other teams have used different algorithms to produce source lists. We have compared our results to those of the Imperial College group (hereafter IC) (Goldschmidt et al. 1997) and to those of IAS (Désert et al. 1998). The common source lists are given in Table 6 for the LW3 filter, and in Table 5 for the LW2 one, as well as the names of the radio sources observed at 8.4 GHz by Fomalont et al. (1997) matching ISOCAM sources.
PRETI and the "triple beam switch method" of IAS (Désert et al. 1998) give very similar results (see Table 5 and 6), especially in the LW3 band where out of the 41 sources detected by them, we find 37. We do not detect HDF_ALL_LW3_8,14,29,40 that are at 3.4, 3.3, 3.9 and 3.1 respectively, that is among the faintest of their list. Both number counts agrees above 200 µJy where PRETI reaches 99.9% completeness: the thwo methods detect 24 sources. Fig. 17 displays a comparison between the fluxes derived by the two methods for the common source list. The fluxes derived by the "triple beam switch method" are lower than our fluxes by a systematic factor of 0.82. We note that both methods relie on the ISOCAM cookbook values for the correction of ADUs into milli-Janskies. For one part, we can explain the difference by an implicit color correction that we have introduced in our simulations: our PSFs are simulated assuming a rising spectrum with spectral index of 3, while standard ISOCAM PSFs, used by Désert et al. 1998, are measured on stars, thus assuming a spectral index of (-2). When comparing our PSFs to standard ones, a mean difference of 10% is found. Moreover, source fluxes are measured by us with aperture photometry while the "triple beam switch method" uses fits of gaussians of fixed width. Since at 6", the ISOCAM PSF is only roughly approximated by a gaussian, this can account for the remaining discrepancy. At these faint flux levels such differences are not very important, considering that the source lists of the two catalogues match very well. However, PRETI allows to reach deeper sensitivity, as we consider sources at 40 µJy as reliable. For the LW2 filter, the "triple beam switch method" detects 6 sources, 3 of which are also detected with PRETI. Of the 3 undetected objects, 2 lie on the edges of the map where we do not allow for detection. The last one is measured by Désert et al. at 58 µJy where we are not complete.
The IC group have made the original data reduction of the ISO-HDF, when some of the latest refinements of ISOCAM calibration such as field distortion were not available. We find 21 of the 22 sources detected by them in their LW3 image (Goldschmidt et al. 1997). We could not find the source J123659.4+621337 in our map. Source photometry differs from our catalog: on the average, their flux estimates are higher than ours by a factor of 1.5, as shown in Fig. 18. They performed simulations, and concluded that they should be 70% complete at a level of 225 µJy. According to the 1.5 factor of difference in photometry, they should have detected at least 26 of our 37 sources above 150 µJy, where they only find 19. This indicates that they are only 50% complete at this level, while PRETI reaches 95% at a level of 150 µJy, according to our simulations, if source confusion is ignored. Comparing the LW2 detections, we only find 5 of the 27 sources in their list (3 from their main list of 6 objects and 2 from their supplementary list of 21 objects), as shown in Table 5. Moreover, we measure much fainter fluxes for these objects. Since this observation is made with 10s integration time, so that pixels affected by cosmic ray glitches are more numerous, two explanations can account for this: either we set too heavy a deglitching criterion and we have cleaned real sources out, or they did not clean their data enough against cosmic rays. Our analysis lead us to favor the second explanation, because we do not find that we are missing sources in the simulations. Here also the results of Désert at al. (1998) are in good agreement with us.
6.3. Number counts
The small number of LW2 sources does not allow to compute reliable number counts, as the poissonian fluctuations introduce an uncertainty of plus or minus 40% 1 when using the reliable 7 source list. Nevertheless, we have, at the level of 100 µJy, 5 reliable detections in an area of , that is: sources per square degree, a value compatible with the prediction of a pure passive luminous evolution model of number counts (Franceschini et al. 1997), as shown on Fig. 19.
The more numerous detections in the LW3 band allow us to compute some points of the Log(N)-Log(S) curve, and to try do determine its slope. We determined two curves, using the 7 and 5 detection thresholds, for which a few more spurious detections are expected.
In order to compute reliable number counts, we have taken into account the following parameters:
This method leads us to derive the number counts that are plotted on Figs. 20 and 21, with error bars corresponding to 1 poissonian fluctuations (as the square root of the number of events). Moreover, we have also plotted on these figure two extreme limits:
Finally, we have also plotted on these figures the prediction of a model with no evolution (Franceschini et al. 1997).
In both 7 and 5 sample, the straightforward result is that the counts derived from ISO-HDF in the LW3 band show a strong excess: nearly a factor of ten at 200 µJy, with respect to the prediction of the no evolution model. This excess is higher than the one obtained by Oliver et al. (1997) from the first reduction of the ISO-HDF data. This might be due to the underestimation of the completeness of the sample.
In order to test our simple method to account for blending, we have plotted in Fig. 21 the number counts obtained with the 5 sample with the method described when blending is not taken into account, but adding to our sample the unblended sources in our supplementary list. On this last plot, the slope of counts is constant down to 60 µJy, and slightly steeper than predicted in the framework of the no evolution model.
© European Southern Observatory (ESO) 1999
Online publication: February 22, 1999