4 Neutron-capture elements

In this section we discuss our adopted linelist, and examine the possible sources of error affecting the abundance determination of neutron-capture elements. We then explore the abundance patterns of elements in the three r-process peaks. Eight elements with atomic numbers $38 \leq Z \leq 47$ were measured in the region of the first peak. The second peak is the best constrained, with thirteen elements measured in the range $56 \leq Z \leq 72$. The third peak and the actinides are probed by four elements with measured abundances in the region $76 \leq Z \leq 92$.

4.1 Linelists and physical data

Lines from 28 neutron-capture elements were observed in CS 31082-001, mainly concentrated in the blue and UV parts of the spectrum. The full linelist for heavy elements is provided in an Appendix to this paper, which also lists the references for our adopted oscillator strengths. When available from the paper of Sneden et al. (1996), the same oscillator strengths were adopted in order to make the comparison to the star CS 22892-052 easier. However, more lines were measured here, hence we had to supplement this compilation with additional information. In particular, we note that new results on the lifetimes and branching factors of both uranium (Nilsson et al. 2001a) and thorium (Nilsson et al. 2001b) transitions are now available; we make use of them here (Table 4). The Appendix also lists equivalent widths of the lines in CS 31082-001 and the individual derived abundances. In cases when blending was severe, or hyperfine structure was important, abundances were determined by comparing the observations directly to synthetic spectra (in these cases, no equivalent widths are listed in the Appendix). Hyperfine structure was included for the Ba and Eu lines. The mean abundances obtained for each element are listed in Table 5. Figures 4 to 10 are representative examples of the quality of the observed spectrum, and the fits to synthetic spectra.

 

 

Table 4: Th and U lines used in the abundance determination.

$\lambda$(Å)	$\chi_{{\rm ex}}$(eV)	$\log gf$	$\log \epsilon$
Th II
3351.229	0.188	-0.600	-0.88
3433.999	0.230	-0.537	-0.80
3435.977	0.000	-0.670	-0.85
3469.921	0.514	-0.129	-0.93
3675.567	0.188	-0.840	-0.83
4019.129	0.000	-0.228	-1.04
4086.521	0.000	-0.929	-0.96
4094.747	0.000	-0.885	-1.05
U II
3859.571	0.036	-0.067	-1.92

Figure 6: The observed Eu 4129 and Hf II 3999 Å line in CS 31082-001. Symbols as in Fig. 2.

 

 

Table 5: Neutron-capture-element abundances in CS 31082-001.

El.	Z	$\log \epsilon$(X)	$\sigma$	$\Delta \log \epsilon$	$N_{{\rm lines}}$	[X/Fe]
				(X/Th)
Sr	38	0.72	0.03	0.08	3	+0.65
Y	39	-0.23	0.07	0.06	9	+0.43
Zr	40	0.43	0.15	0.09	5	+0.73
Nb	41	-0.55		0.15	1	+0.93
Ru	44	0.36	0.10	0.14	5	+1.42
Rh	45	-0.42	0.03	0.13	3	+1.36
Pd	46	-0.05	0.10	0.15	3	+1.16
Ag	47	-0.81	0.17	0.22	2	+1.15
Ba	56	0.40	0.17	0.11	6	+1.17
La	57	-0.60	0.04	0.06	5	+1.13
Ce	58	-0.31	0.10	0.04	9	+1.01
Pr	59	-0.86	0.12	0.06	6	+1.33
Nd	60	-0.13	0.17	0.05	18	+1.27
Sm	62	-0.51	0.16	0.06	9	+1.38
Eu	63	-0.76	0.11	0.05	9	+1.63
Gd	64	-0.27	0.15	0.06	9	+1.51
Tb	65	-1.26	0.07	0.04	7	+1.74
Dy	66	-0.21	0.13	0.07	6	+1.55
Er	68	-0.27	0.08	0.09	5	+1.70
Tm	69	-1.24	0.10	0.08	4	+1.66
Hf	72	-0.59		0.17	2	+1.43
Os	76	0.43	0.17	0.16	3	+1.30
Ir	77	0.20		0.11	2	+1.75
Pb	82	<-0.2:			1
Th	90	-0.98	0.05	0.11	8	+1.83
U	92	-1.92		0.11	1	+1.49

   
4.2 Error budget

Table 6 summarizes the various sources of uncertainties affecting the derived neutron-capture-element abundances in CS 31082-001. Stochastic errors ($\Delta$(obs) listed in Col. 6) arise from random uncertainties in the oscillator strengths (gf values) and in the measured equivalent widths. The magnitude of this error is estimated as $\sigma$/ $\sqrt{N-1}$(where $\sigma$ is the rms around the mean abundance) when $N \geq 2$ lines of a given element are observed, otherwise as the quadratic sum of the estimated error on the adopted gf value and the fitting uncertainty. Systematic uncertainties include those which exist in the adopted oscillator strengths, in the equivalent width measurements, mostly related to continuum location, and in the adopted stellar parameters. The first is extremely difficult to assess and is not considered explicitly here, although it might be significant (gf values from various sources may be cross-checked, but often only one source is available). The second should be negligible, given the very high quality of our data. Hence we have examined here only those errors linked to our choice of stellar parameters. These were estimated by varying $T_{{\rm eff}}$ by +100 K, $\log g$ by +0.3 dex, and $\xi$ by +0.2 km s^-1in the stellar atmosphere model (Cols. 2, 3, and 4, respectively). The quantity $\Delta$(T, $\log g$, $\xi$) listed in Col. 5 is the total impact of varying each of the three parameters, computed as the quadratic sum of Cols. 2, 3, and 4. The total uncertainty $\Delta$(total) (Col. 7) on the absolute abundance of each element ( $\log \epsilon$(X)) is computed as the quadratic sum of the stochastic and systematic errors. Columns 8 and 9 list the total uncertainties on the abundance ratios X/U and X/Th. Note that, due to the similarity of the response of a given set of elements to changes in the stellar parameters, systematic errors largely cancel out in the measured ratios of these elements, reducing the uncertainty affecting the relative abundances. However, for the few species that are determined from neutral lines, the stellar parameters uncertainties impact on the [X/Th] or [X/U] ratio is not negligible anymore (e.g. Ru, Rh, Pd).

In the following discussion, since we are mostly concerned by the relative abundance ratios, we chose to consider the total uncertainties on the X/Th ratio as representative of the uncertainty on the abundance pattern (in Figs. 11 and 12, and listed in column $\Delta \log \epsilon$(X/Th) of Table 5).

 

 

Table 6: Error budget for neutron-capture elements.

El.	$\Delta T$	$\Delta\log g$	$\Delta \xi$	$\Delta$(T, $\log g$, $\xi$)	$\Delta$(obs)	$\Delta$(total)	$\Delta \log \epsilon$(X/U)	$\Delta \log \epsilon$(X/Th)
	+100 K	+0.3 dex	+0.2 km s-1
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Sr II	0.064	0.045	-0.047	0.092	0.030	0.10	0.13	0.08
Y II	0.068	0.082	-0.057	0.121	0.023	0.12	0.13	0.06
Zr II	0.066	0.087	-0.034	0.114	0.075	0.14	0.14	0.09
Nb II	0.044	0.091	-0.019	0.128	0.150	0.20	0.19	0.15
Ru I	0.079	-0.021	-0.012	0.159	0.050	0.17	0.18	0.14
Rh I	0.080	-0.019	-0.007	0.162	0.021	0.16	0.17	0.13
Pd I	0.081	-0.023	-0.024	0.166	0.071	0.18	0.19	0.15
Ag I	0.081	-0.022	-0.020	0.166	0.170	0.24	0.24	0.22
Ba II	0.096	0.046	-0.075	0.122	0.076	0.14	0.16	0.11
La II	0.080	0.083	-0.056	0.128	0.020	0.13	0.12	0.06
Ce II	0.078	0.089	-0.013	0.119	0.035	0.12	0.12	0.04
Pr II	0.078	0.089	-0.012	0.119	0.054	0.13	0.12	0.06
Nd II	0.078	0.086	-0.024	0.119	0.041	0.13	0.12	0.05
Sm II	0.082	0.089	-0.010	0.122	0.057	0.13	0.12	0.06
Eu II	0.078	0.090	-0.014	0.120	0.039	0.13	0.12	0.05
Gd II	0.074	0.088	-0.032	0.120	0.053	0.13	0.13	0.06
Tb II	0.078	0.091	-0.016	0.121	0.029	0.12	0.11	0.04
Dy II	0.076	0.088	-0.023	0.118	0.058	0.13	0.13	0.07
Er II	0.078	0.080	-0.088	0.142	0.040	0.15	0.15	0.09
Tm II	0.092	0.053	-0.043	0.115	0.058	0.13	0.14	0.08
Hf II	0.042	0.086	-0.028	0.123	0.170	0.21	0.20	0.17
Os I	0.078	0.011	-0.008	0.157	0.120	0.20	0.19	0.16
Ir I	0.065	0.045	-0.024	0.140	0.090	0.17	0.15	0.11
Th II	0.048	0.090	-0.008	0.132	0.020	0.13	0.11	...
U II	0.046	0.093	-0.002	0.131	0.110	0.17	...	0.11

4.3 The lighter elements, 38 $\leq$ Z $\leq$ 48

This group includes the classically observed elements Sr, Y, and Zr, but also the less well-studied species Nb, Ru, Rh, Pd, and Ag, whose lines are weak and lie in the near-UV part of the spectrum, hampering their detection in normal metal-poor halo giants. The only metal-poor star in which all these elements have been previously detected is CS 22892-052 (Sneden et al. 2000a), thanks to its large enhancement of neutron-capture elements. In CS 31082-001 we detected even more lines of these elements (due to the slightly larger metallicity, reduced blending by CH, CN and NH molecules, and better spectrum quality); Figs. 4 and 5 show examples of the quality of the fits obtained. However, one element detection remains inconclusive - even the strongest expected transition of Cd I in our wavelength domain (3261.05 Å) is too severely blended to be useful for abundance determinations.

In Solar System material, the four lighter elements are dominated by products of the main s-process (with a possible contribution from the weak s-process as well; see Prantzos et al. 1990; Tsujimoto et al. 2000), while the elements Ru, Rh, Pd, and Ag may contain a large r-process fraction (54% to 86%). However, in a star as metal-poor as CS 31082-001, it is expected that the s-process contribution should be negligible, both from a theoretical point of view (the main s-process takes place in AGBs, where the contribution depends on metallicity via the abundances of both seed nuclei and neutron sources, e.g. Prantzos et al. 1990), and from an observational point of view (Burris et al. 2000 and references therein). Thus, all these elements should represent r-process material produced in the early Galaxy.

Figure 7: The observed Os I 4261 Å, 4420 Å, and Ir I 3513 Å lines in CS 31082-001. Symbols as in Fig. 2.

In Fig. 11 we compare the observed abundance pattern in CS 31082-001 to the Solar System r-process abundances, scaled to the abundance of CS 31082-001 (see Sect. 4.4 for details of the scaling procedure). These Solar System r-process abundances are obtained from a de-composition of the Solar abundances (Anders & Grevesse 1989) into their s- and r-process fractions, by subtraction of computed main s-process yields (AGB yields) from the total abundances to obtain the r-process fraction. We show here two sources for this de-composition, illustrating the uncertainties involved. The dashed line in Fig. 11 follows the compilation of Burris et al. (2000), which uses yields from Käppeler et al. (1989) and Wisshak et al. (1996), while the solid line uses yields of AGB models from Arlandini et al. (1999).

It is clear from inspection of Fig. 11 that, although one can argue the case for general agreement in the region of the second r-process peak, the abundances of CS 31082-001 in the region $38 \leq Z \leq 48$ are not all compatible with the Solar System r-process pattern. This effect is best seen in the middle and lower panels of Fig. 11, where the abundance difference $\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}$ between CS 31082-001 and the Solar System (SS) r-process are displayed. When the Burris el al. (2000) de-composition is used, the difference appears as a stronger odd-even effect in CS 31082-001, in addition to a lower mean abundance: $<\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}>_{38\leq Z\leq 48}~=~-1.47$ (rms 0.33) for the lighter elements vs. $<\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}>_{56\leq Z\leq 69}$ = -1.25 (rms 0.10) for the heavier group. If the Arlandini et al. (1999) de-composition is used, the result is very similar for Ru, Rh, and Ag. Note that in this case, the Y abundance is no longer discrepant, while Nb is significantly more abundant than the Solar r-value. As a result, the mean offsets between the lighter and second-peak elements over large intervals in atomic number are in fact quite similar, < $\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}$ $>_{38\leq Z\leq 48}~=~-1.38$ (rms 0.33) compared to $<\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}>_{56\leq Z\leq 69}=-1.28$ (rms 0.08).

Independently of the decomposition used, the most discrepant element is silver, for which the solar system r-process scaled abundance exceeds the CS 31082-001 observed value by 0.8 dex! This low abundance of Ag was also observed in CS 22892-052 by Sneden et al. (2000a), and the good agreement between the CS 31082-001 and CS 22892-052 silver abundances can be seen in Fig. 12). In contrast, the only other metal-poor stars (4 halo stars with -2.15 $\leq$ [Fe/H] $\leq$ -1.3 dex) in which it was observed so far (Crawford et al. 1998) demonstrated a mild [Ag/Fe] enhancement, in agreement with the mild enhancement of the second r-process peak elements in these stars. This illustrates that not everything is understood in the way the r-process builds up elements in the wide atomic mass-range in which it is at work.

  
4.4 The second-peak elements, 56 $\leq$ Z $\leq$ 72

The group of elements between barium and hafnium is the best studied mass-range of the neutron-capture elements, thanks to the relatively strong lines in the visible region of elements such as Ba, Eu, and La. In CS 31082-001 we were able to detect lines from all stable elements between Z=56 and Z=72. However, three of them cannot be used for abundance determinations because of poorly known atomic physics - Ho and Lu have very strong hyperfine structure which is not well quantified; and while three lines of Yb were detected, two are severely blended (3289 Å and 3476 Å), and the strongest one (3694 Å) yields a very large abundance ( $\log \epsilon$(Yb) = 0.18), which we believe is due to the unaccounted hyperfine structure, which acts to de-saturate this strong line (123 mÅ). We are thus left with accurate abundance determinations for 13 elements in the second peak of the r-process. Figure 6 are example of the fit respectively of a Hf and a Eu line. Not that the europium hyperfine structure used here is from Kurucz (Kurucz 1993), although the oscillator strength was taken from the more recent work of Lawler et al. (2001b). The Eu isotope composition adopted was 47.8$\%$ of ¹⁵¹Eu and 52.2$\%$ of ¹⁵¹Eu, as in the solar system, and in accordance to the new measurement of Sneden et al. (2002) who measured this isotopic ratio to be solar in the two metal-poor r-process rich stars BD +17 3248, 115 and CS 22892-052. The europium isotopic composition of CS 31082-001 will be investigated in a forthcoming paper, together with the rest of our sample of extremely metal poor stars.

Figure 8: The observed Pb I 4057 Å lines in CS 31082-001 (upper panel) and CS 22892-052 (lower panel). Symbols as in Fig. 2.

The abundances of the second peak elements are displayed in Fig. 11, and compared to the Solar System r-process, scaled by the mean abundance difference with CS 31082-001: $<\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}>_{56\leq Z\leq 69}$. Both the Burris et al. (2000) and the Arlandini et al. (1999) de-compositions are very similar in this mass range, and the mean underabundances computed for CS 31082-001 are $<\log \epsilon_{*} - \log \epsilon_{r{\rm SS}}>_{56\leq Z\leq 69}$ =-1.25 (rms 0.10) and -1.28 (rms 0.08), respectively. The remarkable agreement of the abundance ratios in halo stars with the Solar r-process pattern in this atomic mass-range has been noted already in several papers (Sneden et al. 1996, 2000a; Westin et al. 2000; Johnson & Bolte 2001), and is also seen in giants of the globular cluster M15 (Sneden et al. 2000b). In this respect CS 31082-001 resembles other metal-poor stars, both mildly r-process-enhanced (115 and others, see Johnson & Bolte 2001), and the extreme r-process-enriched star CS 22892-052. In fact, Fig. 12 shows that the neutron-capture-element pattern in CS 31082-001 (this paper) and CS 22892-052 (Sneden et al. 2000a) are virtually indistinguishable (CS 22892-052, abundances have been scaled by the mean difference between the two stars $<\log \epsilon_{{\rm CS~31082-001}} - \log \epsilon_{{\rm CS~22892-052}}>_{56\leq Z\leq 69}$ =+0.17 (rms 0.10)). Note that while the absolute r-process abundances are larger in CS 31082-001, given the metallicity difference between the two stars ([Fe/H] $_{{\rm CS~31082-001}}=-2.9$ and [Fe/H] $_{{\rm CS~22892-052}}=-3.2$), the total [r/Fe] ratio in CS 31082-001 in the mass range $56 \leq Z \leq 69$ is 0.13 dex lower than in CS 22892-052.

  
4.5 The third-peak elements and the actinides, 76 $\leq$ Z $\leq$ 92

The third r-process peak (near the magic number N=126) is sampled in CS 31082-001 by Os and Ir. The two heaviest species detected are the radioactive actinides Th and U, the use of which as chronometers we discuss in Sect. 6.

Figure 9: The observed Th II 3351 Å, 3434 Å, 4019 Å  and 4086 Å lines in CS 31082-001. Dots: observations; line: synthetic spectrum computed for the mean Th abundance, $\log \epsilon$(Th) =-0.98.

Figure 10: The observed U II 3859 Å line in CS 31082-001. Symbols as in Fig. 2. The best fit is found for $\log \epsilon$(U) =-1.92 (thick line).

Osmium and iridium
Figure 7 shows two of the three osmium detections, and one of the two iridium detections. It was suggested in our preliminary results (Hill et al. 2001) that these two elements were overabundant with respect to the Solar r-process by around +0.3 dex. We now revise this statement slightly. In particular, Ir falls back to the same abundance scale as the $56 \leq Z \leq 69$ elements. The reason for this revision is connected with the code used to derive abundances. In our preliminary analysis (Hill et al. 2001) we were using the code of Spite (1967), while in the present analysis we have switched to a more self-consistent approach, the synthesis code by Plez et al. (1993), which employs the same algorithms to compute the model atmosphere and the synthetic spectrum. The main difference between the two codes lies in the continuous opacity computations and the source function assumptions (a diffusive term is added in the latter). These codes provide identical results above 4000 Å (less than 0.02 dex difference), but the present, presumably more reliable approach, yields systematically lower abundances in the bluest part of the spectrum (with a maximum effect of $\sim$0.2 dex). In the case of Ir, the two lines at 3512 Å and 3800 Å gave discrepant results in our preliminary analysis, but now agree, with $\log \epsilon({\rm Ir})=0.2$ dex (instead of the earlier 0.37 dex which came from the 3512 Å line alone).

Osmium, on the other hand, still seems to be overabundant with respect to both the scaled Solar r-process fraction and to CS 22892-052. If Os in CS 31082-001 was enhanced to the same level as the second r-process peak elements, we would expect $\log \epsilon$(Os) = 0.15 dex, and $\log \epsilon$(Os) = 0.12 dex if it was enhanced similarly to CS 22892-052, whereas we observe $\log \epsilon$(Os) = 0.43 dex (rms 0.17, from 3 lines). We investigated the possible source of differences between our analysis and that of Sneden et al. (1996, 2000a), which could explain this difference: (i) The three lines used here are the same as those used by Sneden et al. (same wavelengths, same excitation potential and same oscillator strengths), (ii) the ionisation potential of Os I adopted here is from a measurement by Colarusso et al. (1997) of 8.44eV, compared to 8.35eV adopted by Sneden et al. (1996), which could account for at most $\sim$0.1 dex difference in the final abundance but in the wrong direction, (iii) the difference in the codes used for the abundance determination cannot play a role only for osmium, leaving all the other abundance determinations unaffected. Hence, at this point, we cannot account for the abundance discrepancy by any obvious differences in the analysis. We are therefore left with the possibility that the Os content of CS 31082-001 is indeed larger than predicted by a scaled Solar r-process. On the other hand, the large abundance dispersion observed from the three lines (rms 0.17) is a hint that there may be hidden problems in the determination of Os abundance from these lines, so that any strong conclusion would be premature at this stage.

Lead
Recently, accurate abundances of lead in metal-poor CH stars (the Pb having likely been transferred from a now-extinct AGB companion) were reported by Aoki et al. (2000) and Van Eck et al. (2001), using the 4057.8 Å line. In CS 31082-001, this line is not visible. Hence, we can only assign an upper limit of $\log \epsilon$(Pb) < -0.2 dex ( $\log \epsilon$(Pb) $=-0.4^{+0.2}_{-\infty}$), as shown in Fig. 8. However, even this upper limit is of great interest, since it is already below the expected abundance of the scaled Solar System r-process fraction (Fig. 11).

Noting from Fig. 12 that our derived abundance of Pb is drastically lower than the detection reported by Sneden et al. (2000a) for CS 22892-052, we re-assessed also the abundance of Pb in CS 22892-052 from spectra taken with VLT+UVES during the commissioning of the instrument (http://www.eso.org/science/uves_comm/). The spectrum was acquired with a resolution of R= 55 000; the S/N of the co-added spectrum (total exposure time of 4.5 h) is $\sim$140 per (0.013 Å) pixel at 4100 Å (i.e., $S/N \sim$ 330 per resolution element). The model for CS 22892-052 is an OSMARCS model with $T_{{\rm eff}} =$ 4700 K, $\log g =$ 1.5, [Fe/H] =-3.0 and [$\alpha$/Fe] = +0.40 (following Sneden et al. 2000a). The syntheses were generated with $\xi=$ 2.1 km s^-1, [Fe/H] = -3.2, and individual abundances from Sneden et al. (1996). C and N abundances were determined, through a fit of the 3850-3900 Å region (CH and CN bandheads), to be $\log \epsilon$(C) = 6.07 ([C/Fe] = +0.75) and $\log \epsilon$(N) = 5.42 ([N/Fe] = +0.70). Then the Pb region was synthesized (some gf values of nearby atomic lines were adjusted to fit the observed spectrum). The Pb line lies in the red wing of a CH line, which is nicely fitted with the abundances derived from the 3850-3900 Å region. Spectra were computed for $\log \epsilon$(Pb) =-0.5, 0.0, 0.5, and no Pb. From inspection of Fig. 8 we derive an upper limit for the Pb abundance of $\log \epsilon$(Pb)< 0.0 ( $\log \epsilon$(Pb)  $= -0.25^{+0.25}_{-\infty}$). The Pb contents of CS 22892-052 and CS 31082-001, therefore, do not seem to be very different, and the reality of the measurement by Sneden et al. (2000a) appears open to question.

Thorium and uranium
The oscillator strengths of the single U II line and eight of the Th II lines that we have measured in CS 31082-001 (Table 4), as well as many others, have been re-determined with superior accuracy by Nilsson et al. (2001a, 2001b). The associated change in the U/Th abundance ratio is quite significant, as the oscillator strengths of the Th II lines decrease by 0.07 dex, on average, whereas the $\log gf$ of the U II line increases from -0.20 to -0.067, i.e., by 0.13 dex. Moreover, the uncertainties associated with the oscillator strengths have been reduced drastically (to better than 0.08 dex for the individual Th lines, and to only 0.06 dex for the U II line), thus they are a negligible source of error compared to the uncertainties in the actual fit of the data. Figure 9 displays the synthesis of a selection of the observed thorium lines, all plotted with a thorium abundance equal to the mean of the eight lines ( $\log \epsilon$(Th) =-0.98). The observed uncertainty in this case was estimated as $\sigma/\sqrt{N-1}$, where $\sigma$is the dispersion around the mean and N, the number of lines, hence leading to a $\log \epsilon({\rm Th})$ $=-0.98\pm 0.02$.

Figure 10 shows the region of the U II 3859 Å line and an enlargement of the uranium line itself, together with synthetic spectra for four different uranium abundances. The accuracy of the fit is estimated to be around 0.1 dex, obtained by testing the influence of several potential sources of error on the fitting procedure, including placement of the continuum and blending by neighboring lines (mainly Fe I 3859.9 Å). The uncertainty arising from the blending of the Fe I line is linked to uncertainties on the oscillator strength, but also the broadening factor of the line. The Van der waals broadening factor was taken from Barklem et al. (1998), and the oscillator strength was increased by 0.1 dex with respect to the VALD2 recommended value to give a best fit to the observed line Fe I 3859.9 Å. Attempts to vary the Barkelem damping constant by 5% or more induced changes in the U abundance of less than 10%. The unidentified features on the red side of the line (at 3860.77 and 3860.9 Å respectively) hamper the cosmetics of the fit but have no influence on the uranium line region. We would like to point out that the 3860.77 Å unidentified line also appears in the spectrum of CS 22892-052, whereas it does not in other giants of similar metallicity and temperature which have no excess of neutron-capture elements. We therefore tentatively attribute this absorption feature to an neutron-capture element, and encourage atomic physicists to work on the identification of this feature. We also note here that there are numerous features in the whole UV part of the spectrum of neutron-capture enriched giants, that are still in need of identification. A more detailed analysis of the 5-line feature (Fe I 3859.21, Nd II 3859.4, U II 3859.57, CN 3859.67 and Fe I 3859.9 Å) is forseen is a near future, involving 3D hydrodynamical models. Adding the 0.06 dex uncertainty associated with the $\log gf$ of the line results in $\log \epsilon(U)=-1.92\pm 0.11$. As seen from Table 6, the overall uncertainty of the U/Th ratio is totally dominated by that of the U line fitting procedure, while errors in the stellar parameters cancel out completely.

Finally, we note that, using the newly determined oscillator strength value for the uranium line, the upper limit deduced by Gustafsson & Mitzuno-Wiedner (2001) for CS 22892-052 becomes $\log \epsilon$(U) $\leq -2.54$.