Astron. Astrophys. 333, 841-863 (1998)

4. The properties of intermediate redshift damped Ly systems

4.1. H I column density and metal abundances

In the following, we discuss the determination of and metal abundances for each DLAS. Regarding metal lines, the type of data presented here are not appropriate to decide whether they are optically thin or not. We then rely on the velocity distribution inferred from high resolution optical spectra (for species with a similar ionization level), when available. In order to get measurements or limits for several metals from a given system, we use as much as possible already published optical spectra. A curve of growth analysis is performed in cases where no optically thin line has been detected. For consistency with previous studies, we adopt the f values given by Morton (1991) and follow the revisions proposed by Tripp et al. (1996). Abundances are expressed in terms of where represents Solar abundances taken from Anders & Grevesse (1989).
In order to improve our ability to measure the abundance of minor elements, we adapt the stacking technique already used to search for specific features within a whole class of absorption systems (e.g. C IV lines from Ly forest systems: Lu 1991). Some elements like Ni II display several transitions in the same wavelength range with comparable f values and which are likely optically thin. It turns out that in the FOS spectra presented in this paper, the resolution and S/N achieved are such that these lines are occasionally seen individually but at a marginal significance level. We then extract a portion of the spectrum centered on the position where each such line from a given ion is expected and apply appropriate shifts so as to bring the various features at the same wavelength (the shifts are computed a priori, from the absorption redshift and rest wavelengths of the transitions). The individual spectra associated with each line are then averaged. In the optically thin limit, the column density can be obtained from the formula

where , and denotes the oscillator strength, rest equivalent width and wavelength (in  Å) of the transitions considered (note that in the averaged spectrum, the "composite line" is characterized by an equivalent width where n is the number of lines). For undetected lines, all the quoted upper limits are values.

Table 4. Absorption lines detected in the spectrum PKS 1229-021

Table 5. Absorption lines detected in the spectrum of 3C 286

4.1.1. EX 0302-223

The estimate of in this DLAS is complicated by the presence of the QSO O VI /Ly emission line, just redward of the damped Ly line at . Thus, before fitting the latter, the continuum has been defined in a way that leaves a reasonably symmetrical Ly absorption line profile. We get , where the quoted error is dominated by the uncertainty in the definition of the continuum level and by the possible presence of blended Ly forest lines (see Fig. 7). Our result is compatible with an independent estimate based on the same spectrum performed by Pettini & Bowen (1997); these authors have measured the Zn II and Cr II column densities and derived [Zn/H] and [Cr/H] .

Fig. 7. Damped Ly line at in the spectrum of EX 0302-223. The best fit is plotted (dashed line; ) as well as the fits corresponding to a deviation (dotted lines; and 20.51)

In their high resolution spectra, Petitjean & Bergeron (1990) have detected substructure in the Fe II (2586 and 2600 Å), Mg II and Mg I lines and determine . The HST spectrum provides measurements for the 1144 and 1608 Å Fe II lines. Since in the high resolution data N(Fe II) is dominated by one single component, a curve of growth analysis is appropriate. The equivalent width of the 1144 and 1608 Å lines appear to be well consistent with each other (both transitions have comparable f values); when all four lines are considered, no unambiguous solution can be obtained and only a lower limit to the Fe II column density can be inferred from the data,  cm^-2 (i.e. ; a better fit is obtained with in the range 14.6 - 14.8 but the 1144 and 1608 Å line measurements are not accurate enough to provide a reliable upper limit). We adopt the latter value in the following although it is larger than that given by Petitjean & Bergeron (1990): their results are based on the saturated 2586 and 2600 Å Fe II lines and are therefore subject to large uncertainties (e.g. related to the unknown exact zero intensity level). We then infer [Fe/H] . In the HST spectrum, Ni II 1317 and 1370 are marginally present ( detection) with  Å ; the corresponding upper limit is [Ni/H] (the stacking procedure is useless in this case since the portions of the spectra involved are crowded). Similarly, from the non detection of Mn II by Petitjean & Bergeron (1990), one gets [Mn/H] . Several lines from Si II are present but unfortunately, all have comparable and large oscillator strength values and cannot be used to reliably determine .

4.1.2. PKS 0454+039

When fitting the damped Ly profile (Fig. 8), we have excluded a narrow feature near 2268 Å (presumably a Ly -only line) which induces some asymmetry. We then get , a value slightly smaller than that obtained by Steidel et al. (1995), . The difference is likely to arise from our exclusion of the 2268 Å feature; the latter tends to broaden the profile and is less visible on the lower S/N spectrum analyzed by Steidel et al. (1995). These values are compatible given the formal errors quoted above (which, moreover, do not include the uncertainty in positioning the continuum). Steidel et al. (1995) observed several weak transitions from Fe II, Zn II and Cr II from which they determine the abundance of these three elements. Lu et al. (1996) present high resolution and high S/N data on various lines from this system. Several components are detected, spread over 160 km s^-1 ; the major one has a width (FWHM) of about 30 km s^-1. Lu et al. (1996) confirm that the 2249 and 2260 Å Fe II lines used by Steidel et al. (1995) are optically thin.

Fig. 8. Same as Fig. 7 for the damped Ly line at in the spectrum of PKS 0454+039. The three fits shown correspond to , 20.69 (best fit) and 20.75

Lines from Ni II at 1710, 1742 and 1752 Å are found to be marginally present in our spectrum and since they lie in reasonably clean parts of the spectrum, we use the stacking technique discussed above. As can be seen in Fig. 9, a feature is present at 3146.0 Å, the expected wavelength of Ni II 1742 at 0.8596 with 0.18 0.06 Å.

Fig. 9. Composite Ni II (1710, 1742 and 1752 Å) line at in PKS 0454+039. Three portions of the original spectrum have been combined; the one including Ni II 1710 and 1752 have been shifted by 59.5 and -19.5 Å before averaging. The tick mark indicates the expected position of the composite line

This leads to a column density, , and a relative abundance [Ni/H] . The same procedure cannot be used in the more crowded region where Ni II 1317 and 1370 are expected; nevertheless, we check that these two features are individually marginally present (at about a level) with a strength compatible with the previous estimate. For Mn II, Lu et al. (1996) get [Mn/H] 1.36.

4.1.3. 3C 196

As mentioned above, the damped Ly line at 0.437 coincides with a Lyman edge and, in such a circumstance, the low resolution data available poorly constrains . To better assess how far the value already inferred by Cohen et al. (1996) depends on assumptions underlying the fitting procedure, we obtain an independent estimate of based on our new spectrum. In the latter, the damped Ly line goes down to zero at its center and there is no need to correct for the presence of scattered light. Our approach has been to introduce the minimum number of parameters. Three at least are required: the N and b values for H I at and at (given the large expected, the Ly profile is not expected to depend on the velocity distribution). We do not attempt to fit the Ly line at since Cohen et al. have shown that this requires the introduction of an additional parameter - the fraction f of the broad line region covered by the absorbing gas (which has to be less than 1); the relative contribution of emission lines shortward of  Å is expected to be small, so f is no longer relevant. The strength of Ly , Ly , Ly and of the Lyman discontinuity can be used to constrain and b at 0.871. We compute synthetic spectra for various (, b) values, degrade them to the resolution of the G160L spectra and compare them to the data.

We find that and  km s^-1 roughly account for the strength of Ly , Ly , Ly and for the possibly non-zero flux seen shortward of 1700 Å (note that the Cohen et al.'s spectrum also suggests a similar non-zero flux level, although the poor S/N ratio does not allow to be conclusive on this point). In attempting to fit the observed spectrum, we find that the match is not quite so good for the wavelengths of Ly and Ly ; since this is not the case for the Cohen et al.'s spectrum, we believe that this problem arises from a lower S/N ratio or from distortions in the wavelength calibration and, to improve the fit, we allow slight wavelength shifts for these features. This solution is certainly not unique (higher and lower b are also acceptable given the uncertainty in the flux level shortward of 1700 Å; one must also keep in mind that a single Gaussian may be a crude approximation of the real velocity distribution); however, this is not critical since any choice within the acceptable range of values gives about the same shape for the edge when seen at our resolution. We then compute profiles for the Ly line at 0.437, multiply these by the synthetic Lyman edge profile and compare the result to the data. We thus estimate (the corresponding fits are displayed in Fig. 10).

Fig. 10. Same as Fig. 7 for the damped Ly line at in the new (1994) spectrum of 3C 196. The three fits shown correspond to , 20.8 (best fit) and 21.4. The error is displayed and has been shifted downwards for clarity (bottom panel)

This is notably larger than the value of 20.2 derived by Cohen et al. (1996). As we understand it, the difference comes from two reasons. Firstly, the two spectra show departures which, although relatively small, have large effects on the results. In the earlier spectrum, the damped Ly line is less deep and an intensity peak is present just shortward of it (near 1740 Å) while this is much less clear in the latest data (this may be due to different line spread functions: a detailed comparison of the two spectra indicates that indeed, the latest has a significantly higher resolution). Secondly, in both spectra, the continuum is seen to fall off just shortward of 1805 Å, which cannot be due to the Lyman edge (the latter depresses the continuum only at Å). In their (low ) solution, Cohen et al. gives a large weight to metal lines at (S VI 933-944) and (N V 1238-1242) which induce the strong extra absorption required around 1766 and 1800 Å. In our solution, this is naturally produced by the red wing of the damped Ly line itself. We find the large solution more realistic because we doubt S VI at 0.871 and N V at 0.437 can be as strong as required (note e.g. that the Si IV doublet at 0.437 is not detected). Further, it appears unlikely that the strength of these lines be precisely such that their cumulative effect produces the observed coherent fall off. However, although we favor a value above 20.5, we admit that the value cannot be unambiguously determined with the present spectra; only higher resolution data could allow to better model the Lyman edge at 0.871, assess the role of metal and Ly forest lines and determine the true profile of the damped Ly line around its core which really constrains .

Regarding metals, Foltz et al. (1988) have detected in the optical Mg II and Fe II lines but the degree of saturation of the latter is such that they are useless for abundance estimates. One can nevertheless get an upper limit on the Fe II abundance from the non-detection of Fe II 2367. With and a 3 upper limit on of 0.33 Å, we get [Fe/H] . The weak features from Mn II and Ca II are well resolved in the Foltz et al.'s spectrum (line widths exceed 100 km s^-1) and are therefore likely to be optically thin. In this limit, we get from the intermediate strength 2595 Å transition, thus [Mn/H] (the two other Mn II transitions give consistent results; we also checked that the measurements performed by Aldcroft et al. (1994), although less accurate, are in acceptable agreement with those of Foltz et al. 1988). Similarly, from the Ca II K line, we get .

4.1.4. Q 1209+107

Despite the low resolution and S/N, the spectrum constrains well the H I column density at and an acceptable fit to the Ly profile is obtained for (Fig. 11). Acceptable fits can also be obtained by simultaneously decreasing N(H I) and increasing b; however, such solutions are ruled out by the profile of the Lyman edge (Boissé et al. 1998). To our knowledge, the only metal lines from which an abundance can be derived for this system are those of Fe II (Young et al. 1982). From a curve of growth analysis applied to five transitions, we get , thus [Fe/H] . Although no high resolution spectrum is available for Q 1209+107, we believe that this determination is approximately correct because the line strengths clearly indicate that Fe II 2374 lies close to the linear part of the curve of growth (assuming this line to be thin yields the strict lower limit ). Young et al. (1982) do not detect the Mn II triplet; we have used their values to derive the limit .

Fig. 11. Same as Fig. 7 for the damped Ly line at in the spectrum of Q 1209+107. Fits are shown for 19.9, 20.2 (best fit) and 20.5. The error is also given

4.1.5. PKS 1229-021

As in 3C 196, the determination of is complicated by the presence of an abrupt decrease of the continuum flux near the position of the expected damped Ly line due to a Lyman edge from the system. However, the situation is more favorable than for 3C 196 because the spectral resolution is higher. We proceed as above and compute the profile for both the Lyman series/edge at 0.831 and the damped Ly at 0.3950. After assigning a zero intensity level to the core of the damped Ly (which requires a 7% correction; this is the only case for which the offset is negative), it is apparent that the Lyman edge is not completely opaque. On the normalized spectrum, the level attained shortward of 1670 Å is 0.17 from which we derive . We then attempt to reproduce Ly , Ly , ... lines from this system by varying b: a single component does not provide a good fit to the data, the observed lines being slightly too broad for their depth (the fit for   km s^-1 is shown in Fig. 12). However, given the small velocity range involved, we have not attempted multi-component fits because this would not affect the edge profile. In the HST spectrum, the fall off of the intensity begins at  Å : this is too large to be assigned to the edge but rather corresponds to the red wing of the damped Ly line. In fact, at our resolution, the red half of the damped Ly line is nearly unaffected by the Lyman edge which is favorable for the determination of . After successive trials, we get .

Fig. 12. Same as Fig. 7 for the damped Ly line at in the spectrum of Q 1229-021. Fits are shown for 20.54, 20.75 (best fit) and 20.96

PKS 1229-021 has been observed at high spectral resolution by Lanzetta & Bowen (1992). The velocity distribution appears complex and includes narrow components spanning over 200 km s^-1. In the FOS spectra, several lines that could be used for metal abundance measurements are expected. From Si II, only Si II 1808 and Si II 1526 are of interest, other transitions being heavily blended. Assuming Si II 1808 to be optically thin, we get , which is to be considered as a lower limit. Including Si II 1808 and Si II 1526 in a curve of growth analysis suggests that the former line is nearly thin and leads to  km s^-1, a value roughly consistent with the profile observed for unsaturated lines by Lanzetta & Bowen (1992) and . However, since this estimate may be affected by the presence of saturated narrow components in the Si II 1526 line, we adopt for [Si/H] the thin limit, [Si/H] .

Lanzetta & Bowen (1992) observed the two Fe II 2586 and Fe II 2600 lines. Since the latter are strongly saturated, their estimate of heavily depends on the assumed velocity distribution (sum of discrete Gaussian components) and within this assumption, on the number of subcomponents introduced (in such a case, the approach developed by Levshakov & Kegel 1997 to infer column densities may be interesting to consider). Indeed, their would lead to an unrealistically large Fe relative abundance and is in contradiction with the non-detection in our spectrum of Fe II 2249 and 2260. By stacking the (assumed optically thin) two latter features, we get the upper limit . The corresponding relative abundance is [Fe/H] . We also stacked the Ni II 1317, 1370 and 1454 Å lines (the Ni II lines above 1700 Å fall near strong features and cannot be used) and the Cr II 2056 and 2066 Å lines. Ni II is clearly detected (Fig. 13) with  Å , which corresponds to and [Ni/H] . On the other hand, Cr II is not present; we get a 3 upper limit  Å for the composite line which corresponds to or [Cr/H] . Shallow features are seen at the expected position of Zn II 2026 and 2062 with 0.06 Å and 0.09 Å respectively (Fig. 14). Using the first measurement which is both more accurate and uncontaminated by absorption from other species (Mg I 2026 is not expected to contribute significantly contrary to Cr II 2062), we get and [Zn/H] . From unsaturated Mn II lines, Lanzetta & Bowen (1992) derive (sum of all subcomponents) which corresponds to [Mn/H] (since the three lines used are nearly thin over most of the profile, this estimate is not subject to the large uncertainties previously mentioned for Fe II ; the thin limit yields , 13.0 and 13.3 for Mn II 2576, 2594 and 2606 respectively). Ca II H and K lines have been detected at 2 Å resolution by Steidel et al. (1994a). These lines are also seen in an unpublished higher resolution spectrum (0.35 Å FWHM) that P. Petitjean kindly made available to us, with two components at and 0.39516. Equivalent width values from these two spectra are in good agreement, and from the average of Ca II 3934 (0.27 Å), we get .

Fig. 13. Same as Fig. 9 for the composite Ni II (1317, 1370 and 1454 Å) line at in the spectrum of PKS 1229-021. The portions of the original spectrum including Ni II 1317 and 1454 have been shifted by 74 and -118 Å before averaging

Fig. 14. Portion of the PKS 1229-021 G270H spectrum (binned to 1 Å) comprising the Zn II and Cr II lines expected from the DLAS. The three strong lines near 2800 and 2850 Å are from Galactic Mg II and Mg I

4.1.6. 3C 286

The normalization of the spectrum near the damped Ly line is uncertain due to the presence of adjacent emission lines. Therefore, when fitting the profile, we give much weight to the core of the line and get which is in good agreement with the value derived by Cohen et al. (1994) from G160L data (Fig  15).

The DLAS in 3C 286 might seem to be a good case for abundance determinations since the velocity distribution comprises one single component with  km s^-1, a value which is consistent with both 21 cm and Fe II data (see Meyer & York 1992). However, such a low b value implies high line opacities, even with abundances as low as 1/100 Solar. As a result, many of the new lines detected here lie well beyond the linear part of the curve of growth despite their weakness. Since we have some a priori information on the velocity distribution we nevertheless attempt to derive N(Si II) using the few transitions for which reliable measurements could be made (Si II 1260, 1304 and to a lesser extent, Si II 1526). A single Gaussian component with  km s^-1 is clearly inconsistent with the data. Si II 1260 and 1304 could be accounted for if  km s^-1 and but such a large difference between and appears unlikely. Further, the corresponding relative Si abundance would be extremely low (). It seems also that for any velocity distribution, Si II 1190 (which is barely seen) and especially Si II 1193 (undetected) should be notably stronger than observed, as compared to Si II 1260 or 1304; we may therefore suspect that one of the latter is affected by blending with a Ly -only feature and conclude that the present data do not allow to estimate N(Si II) properly. S II 1259 is clearly present on the blue wing of Si II 1260 but higher resolution data would be needed to extract .

Fig. 15. Same as Fig. 7 for the damped Ly line at in the spectrum of 3C 286. The three fits shown correspond to , 21.25 (best fit) and 21.31.

Constraints on can be derived from the absence of Mn II 2576 in the spectrum obtained by Cohen et al. (1994). With , one gets in the thin (i.e. large b) limit; adopting  km s^-1 instead yields a limit of 12.59 indicating that saturation effects might be not negligible in this case. We therefore adopt the latter value which implies [Mn/H] . Regarding Fe II, our measurement of Fe II 1608 appears fully consistent with the curve of growth analysis given by Meyer & York (1992) who derive . The latter authors also give the abundance of Ca II, . Finally, the tighter constraint that we can get on comes from our non-detection of Ni II 1317, which is expected on the blue side of the Ly QSO emission line where the noise level is low. The limit on is 0.080 Å; the thin limit cannot be used in this case and adopting  km s^-1, we find (instead of 13.55 in the thin limit) or [Ni/H] .

4.2. Temperature of the 21 cm absorbing gas

When both Ly and 21 cm absorptions are detected, useful constraints can be derived on the spin temperature of the gas, . The absorbers toward PKS 0454+039 and 3C 286 have been already discussed by Steidel et al. (1995) and Cohen et al. (1994). In the former, 21 cm absorption has not been detected by Briggs & Wolfe (1983) and therefore, only a lower limit on could be inferred, K (Steidel et al. 1995). The high resolution optical data recently obtained by Lu et al. (1996) can be used to get an even tighter constraint. Indeed, they find that the b value for the main component is about 20 km s^-1 (see e.g. the unsaturated Mn II lines). The b parameter relevant to the 21 cm absorbing H I cannot be larger which implies K adopting a single temperature model. Using a similar assumption, Cohen et al. (1994) infer  K for the DLAS in 3C 286. Our result on at in 3C 196 cannot be used to constrain for that absorber because the radio source is essentially extended (and then probes lines of sight distinct from the optical one).

In PKS 1229-021, the situation is more favorable since a significant fraction of the flux at about 1 GHz (the frequency of the redshifted 21 cm line is 1018 MHz) originates from a compact component coincident with the optical quasar (see radio maps published by Kronberg et al. 1992). Following Brown & Spencer (1979) we assume that 50% of the 1GHz flux is emitted by the compact component and that the latter is completely covered by the absorber (this corresponds to a size larger than 30pc). We then derive = 170K. Part of the extended emission could also be covered by the 0.3950 absorber which would result in an increase of . However, such an effect is unlikely to significantly affect the previous estimate because i) the 21 cm line is narrow (FWHM  km s^-1) which suggests that the size of the absorbing region is much smaller than that of a whole galaxy and ii) the absorber candidate (object #3 in Fig. 12 of Paper I) does not cover the extended emission regions. On the opposite, part of the H I inducing the Ly absorption could be at relatively high temperature (e.g. 1000K) and would then be inefficient in producing 21 cm absorption, which would imply an even lower value for the rest of the gas (see the discussion by Wolfe et al. 1985). For instance, if 75% of the gas is at a temperature higher than 1000K, the remaining 25% has to be at less than 49K. We can therefore confidently conclude that, contrary to the DLAS in PKS 0454+039 and 3C 286 (see also Briggs & Wolfe 1983), a significant fraction of the absorbing gas is at a low temperature, typical of H I clouds in the Galactic disk.

4.3. Neutral carbon

If physical conditions in the gas associated with the DLAS studied here were similar to those prevailing in the interstellar medium of our own Galaxy, neutral species should be present in detectable amounts. C I especially can be searched for through its strong 1277, 1328, 1560 or 1656 Å transitions. The latter have been detected in some high redshift DLAS (see e.g Blades et al. 1982; Ge et al. 1997) but, in several cases, stringent upper limits have been obtained (Meyer & Roth 1990; Black et al. 1987). In order to investigate the presence of neutral gas, we made a specific search for C I lines. For EX 0302-223 and PKS 1229-021, no useful constraint could be obtained because the features are expected in regions where either there are strong lines or is too large. On the opposite, C I 1560 in PKS 0454+039 and C I 1328 in 3C 286 are expected right onto one of the QSO emission line (N V and Ly respectively) where the spectra are locally of excellent quality. In PKS 0454+039, we do see a weak line at 2901.55 Å () with Å (Fig. 16).

Fig. 16. Portion of the G270H spectrum of PKS 0454+039 comprising the 1560 line from the DLAS. Dashed tick marks indicate the position expected for Si IV lines from the weak C IV system at . The strong C IV doublet from the DLAS is also shown

An alternative identification could be Si IV 1402 (at ) from the weak metal system. Unfortunately, Si IV 1393 coincides with C IV from the DLAS and cannot be used to estimate the strength of Si IV 1402. The wavelength match strongly favors an identification with C I 1560 and we consider the latter as likely; higher resolution data are needed to definitely establish the correct identification and the presence of C I. Similarly, in the G190H spectrum of 3C 286, there is a feature at 2248.87 Å () with 0.022 Å (Fig. 17). We consider the identification as certain because the line is also seen in the G270H spectrum (although with a lower S/N) and because the wavelength match is excellent.

Fig. 17. Portion of the G190H normalized spectrum of 3C 286 comprising the C I 1328 and C II 1334 lines from the DLAS. Note the low noise level around 2250 Å which corresponds to the top of the QSO Ly emission line

Assuming these lines to be optically thin we get and  cm^-2 for PKS 0454+039 and 3C 286 respectively. In the second case, the quoted value is a lower limit because of possible saturation effects (we get  cm^-2 assuming instead  km s^-1).

In order to compare the physical conditions in these absorbers to those in our Galaxy we consider the plot given by Jenkins & Shaya (1979). The DLAS in PKS 0454+039 appears close to that in Q 0013-004 (Ge & Bechtold 1997) and to Galactic gas. On the other hand, the DLAS in 3C 286 is more like that in MC 1331+170 (Chaffee et al. 1988) and PHL 957 (Black et al. 1987), i.e. significantly deficient in C I with respect to Galactic gas. However, given the low metal abundance seen in the absorber toward 3C 286, the inferred C I /H I ratio suggests that physical conditions in the absorbers are relatively similar to those in our Galaxy and therefore, that there is enough dust to provide the required shielding from UV photons with energy higher than 11.26 eV.

4.4. Dust grains

Up to now, the evidence for dust associated with DLAS has been mostly statistical in nature, QSOs with DLAS showing in average steeper spectra than QSOs devoid of strong systems (Pei et al. 1991). The overall pattern of metal abundances also strongly suggests that selective depletion onto dust grains is effective in the absorbers (Pettini et al. 1997b; Kulkarni et al. 1997) although the interpretation of these data is still controversial (Lu et al. 1996; Prochaska & Wolfe 1996). The 2175 Å feature would be a less ambiguous signature and can be searched for in specific QSOs with DLAS. However, no clear detection has been obtained in any individual QSO (see e.g. Boissé & Bergeron 1988); this is generally taken as evidence for SMC or LMC-type extinction curves which display a less prominent feature. Recently, Malhotra (1997) found evidence for this feature in a composite spectrum of QSOs with Mg II absorption.

Regarding our targets, the 2175 Å feature could be seen in PKS 0454+039 between the C IV and C III QSO emission lines (near 4020 Å) in the excellent flux-calibrated spectrum obtained by Steidel & Sargent (1992). A shallow depression is present centered about 40 Å blueward of the expected position and with a full width of about 200 Å. Comparison with the composite spectrum computed by Zheng et al. (1997) reveals that this feature is most likely intrinsic to the QSO. In PKS 1229-021, it is expected at 3030 Å near the end of our G270H spectrum: no broad depression with an amplitude larger than 10% is seen over a 300 Å width interval. Finally, in 3C 286, some break is seen near 3680 Å in the flux-calibrated spectrum presented by Aldcroft et al. (1994) which could be accounted for by a redshifted 2175 Å feature with a depth of 15 to 20%. The bluest part of the spectrum is noisy and probably affected by intrinsic broad absorption; thus, the reality of that feature is difficult to assess. The spectral index measured for PKS 1229-021 and 3C 286 between the Ly and C IV emission lines are 0.9 and 0.8 respectively which suggests little reddening. In the former, some bending is seen shortward of the O VI emission line but again, comparison with the composite spectrum of Zheng et al. (1997) indicates an intrinsic origin.

4.5. H₂ molecules

H₂ and CO molecules have been searched for in the spectrum of QSOs with high redshift DLAS (see e.g. Black et al. 1987; Lanzetta et al. 1989). H₂ has been detected in two cases only: at in PKS 0528-250 (Foltz et al. 1988) and recently at in Q 0013-004 (Ge & Bechtold 1997). The former system is peculiar as it is at , so the latter case is the only clear detection of H₂ from gas which is likely to be disk material. Aside from these two positive cases, low upper limits have been inferred for f, the fractional abundance of H₂ molecules (typically - ; Black et al. 1987). One major difficulty encountered in these studies is that H₂ lines are expected in the dense Ly forest where they can hardly be distinguished from Ly -only features. At lower redshift, the Ly forest becomes less crowded and the situation is more favorable.

Among the four QSOs from our sample for which G190H or G270H spectra are available, three - EX 0302-223, PKS 0454+039 and 3C 286 - could display H₂ features (all the strong ones occur at  Å). As emphasized by Black et al. (1987), anticoincidences are most significant and, in the spectrum of the three QSOs mentioned above, we have searched for windows which look free of any significant absorption and where strong H₂ lines are expected (Morton & Dinerstein 1976; Foltz et al. 1988). Such regions can indeed be found (e.g. around 2035 Å in PKS 0454+039 or around 1757 Å and 1852 Å in 3C 286: see Fig. 2 in Boissé et al. 1998) which indicates that, at our detection limit, H₂ is not present. The 3 upper limit on for unresolved H₂ lines in the three QSOs is about 0.15 - 0.20 Å. Unfortunately, this value cannot be translated easily into a limit on because the excitation temperature and b parameter are unknown. We can nevertheless obtain an upper limit by comparing the data to synthetic spectra computed by e.g. Foltz et al. (1988) and Lanzetta et al. (1989). For - 15 km s^-1 and in the range 100 - 1000 K,  cm^-2 appears as a conservative upper limit on . Such a column density implies an upper limit on f of , and for EX 0302-223, PKS 0454+039 and 3C 286 respectively.

4.6. Associated gas of high ionization

Among the four DLAS studied at 1.5 - 2 Å resolution, all have strong C IV - Si IV lines except that in 3C 286 (weak C IV, Si IV undetected). The O VI doublet from the DLAS could have been seen in PKS 0454+039 and 3C 286; it is present in the former absorber only, which displays an extensive range of ionization levels. N V lines are undetected in EX 0302-223, PKS 1229-021 and 3C 286 while they are possibly present in PKS 0454+039, blended with a group of Ly -only lines. We note that in the latter case, the line of sight to the QSO probes the halo of a compact galaxy. The upper limits on for undetected lines are in the range 0.15 - 0.3 Å. More data on C IV, N V and O VI absorption lines from low z identified absorbers are needed to investigate the relation between the strength of high ionization features and the properties of the intervening galaxies.

© European Southern Observatory (ESO) 1998

Online publication: April 28, 1998

helpdesk.link@springer.de