4 New atomic line data and their evaluation

 The new line lists for VALD-2 primarily consist of laboratory data and data improved or derived from astrophysical applications. For some species not present in the previous version of our data base we include new calculations. In the following subsections we report on the new data (essentially element by element) and we also explain our ranking when several data sources are available for the same transitions. A survey of all ions for which data has been acquired for VALD in addition to the Kurucz (1993b,c) lists is given in Fig.1 (all the ions discussed in this Section are marked). To improve both old line lists (such as BELLHEAVY from Kurucz 1993b) and new lists described in Sect.4.4 Landé factors for the neutral and first ions of all rare earth elements were added (this includes Tb and is not specifically marked in Fig.1).

The following lists were already part of the first installation of VALD: BELLLIGHT, BELLHEAVY, and NLTELINES from Kurucz (1993b), the GFIRON data on iron peak elements from Kurucz (1993c), lines with observed energy levels taken from Kurucz & Peytremann (1975, not distributed any more), and a list on rare earth elements based on Meggers et al. (1975). They are part of the VALD-2 installation as well but will not be discussed here except for comparison with new data. A few additional lists with highly accurate oscillator strength data had been included into VALD-1 which were redistributed to new VALD-2 linelists. The latter have been compiled separately for each element. The line data redistributed to new lists are compared here with new data and references are given to both the original sources and Paperi.

To make the ensuing discussion more readable we use abbreviations for the references. The same abbreviations are also used in the "reference field'' (number 15) of the VALD extraction output and consist of up to four letters derived from the initials of the respective authors. Uniqueness of these references is guaranteed within an individual element, which is sufficient as all the new line lists of VALD-2 are grouped according to elements and a compact reference format is mandatory when dealing with millions of lines.

4.1 Light elements

  While this paper was being refereed, we implemented new atomic data for C, N, and O (Wiese et al. 1996) into VALD which were provided by NIST. A detailed description of these data sets has to be postponed to a paper on the next implementation of the data base (VALD-3). Already before this extension, we had included new data for 62 lines of SiII and Pi into the data base.

Silicon. Experimental transition probabilities for SiII lines were measured by Bergeson & Lawler (1993a - BLa), by Calamai et al. (1993 - CSB) and by Blanco et al. (1995 - BBC). BLa used a combination of the lifetimes measured by time-resolved laser-induced fluorescence (LIF) and emission branching ratios. CSB measured transition probabilities of 3s²3p(²P⁰) - 3s3p²(⁴P) intersystem lines with an ion-trapping technique. BBC measurements were based on emission line intensities from laser-produced plasma. The accuracy (relative error) of the experimental oscillator strengths lies between 10 and 20%. For two lines in the near ultraviolet and three lines in the red region the experimental data were obtained for the first time. The new list contains 35 spectral lines and for 15 of them we included calculated Stark damping constants taken from Lanz et al. (1988).

Phosphorus. Radiative lifetimes for the 4p excited states of neutral phosphorus were measured experimentally for the first time by Berzinsh et al. (1997). Measurements for 6 levels were combined with calculations for another 7 levels and with theoretical branching ratios to deduce oscillator strengths for 27 lines belonging to the 4s - 4p transition of Pi. Using this set of oscillator strengths a new value for the solar phosphorus abundance of 5.49 $\pm$ 0.04 was obtained which decreases the discrepancy between earlier determinations and the meteoritic value (Berzinsh et al. 1997).

4.2 Iron group elements

  From the near ultraviolet to the near infrared, iron peak elements contribute the bulk part of line opacity in most stars of spectral types B to K. In VALD-1 most of the experimental data was based on the NBS compilations by Martin et al. (1988) and by Fuhr et al. (1988) in the revised version by A. Gulliver -- with Fei as the only major exception (see Paperi for details). Meanwhile, for most species of lower ionization stage new atomic data of high accuracy and -- in many cases -- with individual error estimates have become available. Consequently, we were able to include new data for almost 12000 lines from iron peak elements into VALD-2.

Scandium. New absolute transition probabilities for 182 Sci and for 64 ScII lines were measured by Lawler & Dakin (1989-LD) combining emission branching ratios with radiative lifetimes from time-resolved laser-induced fluorescence (LIF) with a relative error not more than 10%. These data include 141 new Sci and 25 new ScII lines in addition to the NIST compilation (Martin et al. 1988). The overall agreement between the new and the NIST data is within 0.1dex, although the absolute scale is slightly shifted. A few lines show a larger discrepancy, for example 3 lines of the ⁴F - ⁴D⁰ multiplet of Sci, two weak lines of Sci at 2719.13Å and 6306.0Å, and two lines of ScII at 3645.31Å and 3666.53Å. The two last lines have an accuracy of about 50% in the NIST list. If we ignore the latter, we obtain the following difference between NIST and LD transition probabilities:

Sc

i: $\log gf$$_{\rm NIST}$ - $\log gf$$_{\rm LD} = -0.04\pm 0.04$(27 lines)

Sc

II: $\log gf$$_{\rm NIST}$ - $\log gf$$_{\rm LD} = +0.04\pm 0.08$(37 lines).

We also checked the new data with abundance calculations for the Sun using Kurucz's (1993a) standard solar atmosphere model with $T_{\rm eff}$=5777K, $\log\,{g}$=4.44, and $\xi_t$=1.5kms^-1 (for the actual line calculations a value of $\xi_t$=0.85kms^-1 was used). The equivalent widths of the solar lines were taken from the solar atlas of Moore et al. (1966). The mean solar scandium abundance deduced from 21 very weak (1-14 mÅ) Sci lines is 3.12$\pm$0.24, and from 26 ScII lines it is 3.22$\pm$0.14. No weights were introduced for calculating the standard deviations. The corresponding values obtained for the same sample of solar lines with the available NIST oscillator strengths are 3.24$\pm$0.33 (12 Sci lines), and 3.19$\pm$0.17 (10 ScII lines). The adopted solar abundance is 3.17$\pm$0.10 (Grevesse et al. 1996). The new data are in good agreement with the Kurucz (1993c) semi-empirical calculations. Note that in the NIST compilation Sci $\lambda$5301.96 has a wrong wavelength (5302.98Å).

Titanium and Manganese. Raassen & Uylings (1997 - RU) and Uylings & Raassen (1997 - UR) published new calculations of the transition probabilities for TiIII and MnIII. The authors used the orthogonal operator description for odd and even energy levels. This method allows a more accurate evaluation of the wave functions which leads to an order of magnitude better accuracy for the transition probabilities in comparison with the semi-empirical method used by Kurucz (1993c). Not surprising, the new calculations for TiIII agree better with the experimental intensities (Raassen & Uylings 1997). Both data sets are included in VALD-2 with higher ranking than older lists.

Vanadium. New measurements of lifetimes and transition probabilities for VII were performed by Biémont et al. (1989 - BGF) based on LIF lifetime measurements and emission branching fractions. They measured 147 lines of which 85 are in common with the NIST compilation (Martin et al. 1988). Most of the VII data in the NIST table are coming from Karamatskos et al. (1986). A comparison of NIST and BGF lists shows an agreement within 15% ($\log gf$$_{\rm K}$ - $\log gf$$_{\rm BGF} = -0.014\,\pm \,0.065$) with the exception of 6 lines which were taken from Wujec & Musielok (1986) and from Roberts et al. (1973). For these lines the difference exceeds 0.2dex and amounts to up to 0.6dex. Both groups, Karamatskos et al. and BGF, claim identical accuracies ($\leq$ 10%) in most of the cases. We prefer the BGF line list, because it contains more lines with accurate transition probabilities.

Chromium. New measurements of the transition probabilities for 12 CrII lines from the 3d⁴(⁵D)4p z⁶P⁰ levels were published by Bergeson & Lawler (1993b - BL) with an estimated accuracy of about 10%. Only one of these lines was previously included in the NIST compilation (Martin et al. 1988). The determinations are based on the combination of emission branching ratios with LIF lifetime measurements.

Ekberg (1997) analyzed a spectrum of doubly ionized chromium with a low voltage spark discharge and the normal incidence spectrograph at NIST. He observed 143 new energy levels of the 3d³4d and 3d³5s configurations leading to a classification of 721 new CrIII lines. Using the Cowan (1981, 1995) codes Ekberg calculated transition probabilities for 1893 lines in the wavelength region from 736Å to 2675Å. Figure2 compares Kurucz's semi-empirical calculations with Ekberg's results. A large dispersion of $\pm$3dex is observed for the lines with lower level excitation energy between 8 and 9eV. We prefer the new calculations by Ekberg, because they are based on a larger sample of observed energy levels and therefore provide more accurate wavelengths and transition probabilities.

Iron. New accurate measurements of the transition probabilities for neutral iron became available after the NIST compilation by Fuhr et al. (1988) appeared. The most extensive set of measurements, which contains 1814 lines in the range of 2250 - 26660Å, was produced by O'Brian et al. (1991 - BWL) and already included into VALD-1. They used their LIF lifetime measurements in combination with emission branching fractions. For 640 lines transition probabilities were found by interpolating level populations in the inductively coupled plasma source (ICP). For most of these lines the accuracy is better than 10%, which is supported by a comparison with the high accuracy Oxford absorption oscillator strengths. Bard et al. (1991 - BKK) and Bard & Kock (1994 - BK) measured transition probabilities for 230 Fei lines using the same technique. Both works have 80 lines in common with O'Brian et al. (1991). There is no difference in the absolute scale of both sets and they agree within 25%.

We merged all three sets in one new VALD line list and averaged data with the same accuracy (according to the authors); otherwise we tabulated the oscillator strengths with the higher accuracy. We also included in the new list theoretical $\log gf$-values calculated for neutral iron lines in the IR which were classified by Johansson et al. (1994b - JNG) and by Schoenfeld et al. (1995 - SCG) based on laboratory and solar analyses. These lines belong to the 4f - 5g supermultiplet (25445Å - 25700Å), to the 4f - 5g supermultiplet (38730Å - 39280Å), and to the 5g - 6h supermultiplet (73700Å - 74100Å) of the 3d⁶4s(⁶D) configuration. All theoretical calculations were compared to $\log gf$-values derived from the solar spectrum and they were in good agreement. The new file contains latest data on wavelengths, level energies and classification - when available - from the New Fei Multiplet Tables (Nave et al. 1994). The final VALD-2 file consists of 2962 Fei lines of which half have a relative error in $\log gf$ of not more than 10%.

Likewise, accurate LIF lifetime measurements with an uncertainty of less than 5% are now available for FeII (Biémont et al. 1991; Guo et al. 1992; Hannaford et al. 1992). They were used to transform high precision emission branching ratios to absolute transition probabilities. The most recent work by Bergeson et al. (1996 - BMW) includes 67 lines from the 3d⁶(⁵D)4p subconfiguration in the spectral region from 2249Å to 2762Å. All but four lines have an accuracy between 3 and 10%. Another work by Mullman et al. (1997 - MSL) provides absolute absorption oscillator strengths for 7 vacuum-UV lines of FeII in the 1608Å - 1640Å spectral region with an accuracy better than 10%. We decided to use the BMW list as a reference for the comparison between different lists of FeII transition probabilities. First, we compared BMW with the data by Bridges (1973), by Whaling (1985 - W), by Kroll & Kock (1987 - KK), and by Pauls et al. (1990 - PGH). The last list has only two lines in common with BMW and they agree within 12%. The list of Bridges contains three lines in common with BMW and the agreement is better than 10%. A comparison between BMW and W data is shown in Fig. 3a, and between BMW and KK in Fig. 3b. The 46 lines which are in common with Whaling's list agree within 12%. The only systematic difference between both sets is a shift of -0.04dex in the absolute scale of Whaling's $\log gf$-values. Therefore, we apply this shift to all Whaling data. BMW and KK data agree within 10% with no difference between the absolute scales, but there is a systematic dependence of the $\log gf$-difference on the oscillator strength which causes an error of less than 25% for the whole range of -2.0 $\leq$ $\log gf$ $\leq0.5$. As a result, we give the highest priority in the spectral region of 1600Å to 3000Å to MSL and BMW data. We supplement them with Whaling's data corrected by +0.04dex, with Bridges' data (second priority), and with PGH and KK data (third and fourth priority). For a few lines we averaged KK and PGH oscillator strengths.

  

Figure 3: A comparison between FeII oscillator strengths measured by Bergeson et al. and by Whaling a), and by Bergeson et al. and by Kroll & Kock b)

For $\lambda\geq$ 3000Å the main sources for FeII oscillator strengths are: Bridges (1973), Baschek et al. (1970, corrected by +0.16 according to Fuhr et al. 1988), Whaling (1985), Hannaford et al. (1992 - HLGN), Kroll & Kock (1987), Heise & Kock (1990 - HK), Pauls et al. (1990), and Blackwell et al. (1980 - BSS). The HK and PGH data were slightly corrected to fit the best available lifetime measurements by HLGN. The solar oscillator strengths obtained by BSS have a good relative accuracy, but they were based on a solar iron abundance of log(Fe/H)=-4.31. The best present estimate gives log(Fe/H)=-4.50 (e.g. HLGN) and we therefore applied a +0.19dex correction to the BSS oscillator strengths. A comparison of the corrected data with the data from other sets showed a good agreement. In total we obtained 84 lines in the 3000Å - 7712Å spectral region with oscillator strengths having an accuracy of 25% or better. For roughly half of the lines oscillator strengths from 2 to 4 different sources were averaged and they may be considered as the most reliable.

We corrected oscillator strengths for the forbidden ($\Delta S=2$)transitions of FeII, which stem from an anomaly originating from an indirect level mixing of w²P_3/2 and x⁶P_3/2 (Johansson et al. 1995 - JBL). We also included new oscillator strengths for 222 lines of the 4f - 5g supermultiplet of FeII calculated with the Cowan code (Rosberg & Johansson 1992 - RJ) and for 76 lines of the lowest 5g - 6h supermultiplet of FeII calculated in the framework of the relativistic Hartree-Fock approximation by Biémont et al. (1997 - BJP). The final VALD-2 list contains 522 lines of FeII. Further information on the input data for iron can be found in Ryabchikova et al. (1999b).

Cobalt. Accurate transition probabilities for 15 Coi lines with w⁴D_3/2 as the upper level were published by Lawler et al. (1990 - LWG). The measurements were based on LIF lifetime measurements and on emission branching ratios. They have an accuracy of 10 - 12%. Previously, only Kurucz's semi-empirical data were available for these lines.

Two new experiments on CoII transition probabilities were published since the paper by Salih et al. (1985) which was the only source for this ion in the NIST compilation. Crespo López-Urrita et al. (1994b - CUNJ) measured emission branching ratios and converted them into absolute transition probabilities using the lifetimes published by Salih et al. (1985) and by Pinnington et al. (1973). A comparison between the oscillator strength measurements of CUNJ and of Salih et al. (1985) shows a remarkable agreement in the absolute scale within 1% and with a standard deviation of 10% for 26 lines with log(gf)>-1.0. Only one line, $\lambda$2694.68Å, is significantly outside these error limits. For weaker lines we find $\log gf$-values from CUNJ to be systemetically larger by 0.22dex. There are a few lines from Salih et al. which were not included in the NIST compilation due to their apparently low accuracy. Mullman et al. (1998 - MCL) reported transition probabilities for 28 lines combining LIF and emission branching ratio measurements. The error estimates for most of these lines do not exceed 10%. They have 8 lines in common with CUNJ and after correcting the latter for new lifetime measurements they agree within 22%. We attributed the highest rank to MCL data and trusted the error estimates provided by the authors. Based on our comparisons we averaged for most of the lines data from Salih et al. and from CUNJ and compiled them to a new VALD-2 file, together with the MCL and the other new lines from CUNJ. For the lines with $\log gf$>-1.0 we estimate the error to be 10% (0.04dex), and for the rest of the lines we give the errors as quoted by the authors. In total, the new list contains 89 CoII lines.

Nickel. Fuhr et al. (1988) give the highest priority to two sets of experimental transition probabilities for Nii presented by Huber & Sandeman (1980) and by Doerr & Kock (1985). In the mean time, two new sets of experimental measurements were published. Blackwell et al. (1989 - BBPL) used the Oxford spectroscopic furnace to measure relative oscillator strengths for 75 low-lying lines with a very high precision of 0.7%. They converted them to an absolute scale using lifetimes mainly from Becker et al. (1974, 1981). Wickliffe & Lawler (1997a - WLa) reported transition probabilities for 76 lines connected to high-lying, even-parity levels, using emission branching ratios and new LIF lifetime measurements (Bergeson & Lawler 1993c). WLa also checked the Oxford absolute scale with the new lifetimes and found it to be accurate to within 2% after applying an offset of 0.015dex. Hence, one may expect an accuracy for individual lines of the "Oxford measurements'' by BBPL of about 5%. The same accuracy is reported for most lines from WLa. Thus, both lists combined give a total of 151 lines with accurate transition probabilities which we included in VALD-2 with a high ranking. A comparison of the WLa list with the NIST compilation (35 lines in common) shows good agreement. If we reject 3 lines for which the accuracy is marked with a "D'' in the NIST table, we obtain $\Delta$$\log gf$$_{\rm NIST-WLa} = -0.05\pm0.04$. A similar comparison of NIST and BBPL is less comforting. For 74 common lines we obtain $\Delta$$\log gf$$_{\rm NIST-BBPL} = 0.06\pm0.13$.The excess scatter probably is due to the Doerr & Kock's data, because for lines with $\log gf$>-2.0 the transition probabilities from Huber & Sandeman are in excellent agreement with BBPL measurements. Even for weaker lines the differences are still within the errors quoted by Huber & Sandeman (see Blackwell et al. 1989 for a discussion). We used 38 lines from the solar spectrum with equivalent widths from 3 to 110 mÅ and the same solar model atmosphere as in the case of scandium to check the new data. Without attributing weights to individual lines we obtain log(Ni/H) = -5.74$\pm$0.10 which agrees perfectly with the solar and meteoritic value of -5.75 (Grevesse et al. 1996).

4.3 Elements of the fourth and fifth periods

  For elements of the fourth and fifth periods the data for the previous release of VALD has almost entirely been taken from the BELLHEAVY compilation (Kurucz 1993b). New experiments and calculations for these species allow improvement and extension of the contents of VALD for about 1000 spectral lines. For several of these ions no data was available before.

Copper. Most of the data for CuII described in Paperi were taken from the BELLHEAVY line list (Kurucz 1993b) and actually date back to the compilation of Kurucz & Peytremann (1975). In the meantime, experimental absolute transition probabilities were derived by Kono & Hattori (1982) using the delayed-coincidence technique, and by Crespo López-Urritia et al. (1994a) with special high frequency hollow electrode discharge and emission measurements. The results from the last two groups agree within the expected errors and we merged them into the new file for CuII after averaging oscillator strengths for the lines which were in common. The final list contains data for 71 spectral lines with an accuracy for the $\log gf$-values of the order of 15-25%. Old VALD and the new data agree within 25%, without any shift in absolute scales. We recommend to use the new data also because they have individual error estimates.

Zinc. New absolute transition pobabilities for 2 resonance lines of ZnII at 2025.5Å and 2062.0Å were measured by Bergeson & Lawler (1993b) combining emission branching ratios and LIF lifetime measurements. The accuracy of the new data is 7%. The new oscillator strengths are higher by 0.08dex than those available in VALD-1.

Yttrium and Zirconium. Previously, there was no information in VALD on the second ions for any of the elements of the Sr-Y-Zr group. However, transition probabilities for the most prominent lines of YIII and ZrIII were calculated by Redfors (1991) using the Cowan code with estimated uncertainties of about 10%. Later on, Reader & Acquista (1997 - RA) measured and classified 482 ZrIII spectral lines in the 630-4610Å region. For 4 lines they gave double or multiple classifications. The observed energy levels were interpreted theoretically with the Cowan code and the oscillator strengths were calculated for all observed transitions. Maniak et al. (1994) measured the lifetimes of five levels of YIII which were converted to oscillator strengths using their theoretical calculations of the branching ratios. Both sets of data for YIII agree quite well for 5p - 5d and 5p - 6s lines, while the experimental values on average are smaller by 25% for 5s - 5p and 4d - 5p lines than those from Redfors (1991). Our final list consists of 39 YIII lines in the 1280Å to 3020Å spectral region for which the oscillator strengths were taken from Maniak et al. (1994) and supplemented by the data from Redfors (1991).

The Reader & Aquista (1997) line list for ZrIII has 75 lines in common with Redfors' list. A comparison between both sets of data shows that, with the exception of a few lines for which RA oscillator strengths are smaller by 0.3dex, the agreement for most of the lines is within 10-15%. We prefer the RA list for VALD-2, supplemented by 3 lines from Redfors (1991). The final list contains 493 ZrIII lines.

Ruthenium. Accurate transition probabilities for 482 Rui lines were derived by Wickliffe et al. (1994). They combined LIF lifetime measurements with the emission branching ratios. For most of the measurements the precision is better than 10%. The new data included in VALD-2 show systematically lower gf-values than the Corliss & Bozman (1962) data. For 114 Rui lines we had no oscillator strength values in the previous version of VALD. Absolute transition probabilities for 18 UV lines of RuII were determined by Johansson et al. (1994a). The accuracy of the measurements is better than 25%, which could be confirmed by the Ru abundance analysis in the atmosphere of the HgMn star $\chi$Lup (Johansson et al. 1994a).

Xenon. The energy level classification of XeII lines in VALD (see Paperi) was checked and when necessary corrected according to the extensive analysis of XeII by Hansen & Persson (1987).

4.4 Rare earth elements (REE)

  The atomic data for neutral atoms and first ions of REE included in the first version of VALD were mainly based on the BELLHEAVY line list supplemented with experimental data from Komarovskij (1991) and data based on Meggers et al. (1975) for some of the first ions. A new set of experimental data for Pri, Smi, Eui, Hoi, and Ybi was created from the compilation of Komarovskij (1991) and included into VALD-2. No information was available for the second ions which are dominant among rare earth elements in stellar atmospheres with a $T_{\rm eff}$ greater than 7500K. In particular, for Ap stars with overabundant REE the second ions may contribute significantly to the total line absorption. Lines of the REE appear prominently in spectra of magnetic Ap stars which implies the necessity to include Zeeman splitting parameters. For the neutral atoms and the first ions of all REE, except for PrII and NdII, Landé factors of the lower and upper levels determined by experiments were extracted from the AEL section of the NIST atomic line data base (Martin et al. 1978) and included into VALD-2. For PrII and NdII the measurements by Ginibre (1989) and by Blaise & Wyart (1984) were used. Moreover, new $\log gf$-values as well as improved wavelength calibrations and level classications for 1800 lines from rare earth elements have been included into the data base, which we describe in the following.

Lanthanum. A new calibration of the intensities published by Meggers et al. (1975) was proposed by Bord et al. (1996). They provide oscillator strengths in agreement with the laser-beam results by Arnesen et al. (1977). The latter was used by Gratton & Sneden (1994) for the recent solar lanthanum abundance determination. Hence, the intensity calibration of Bord et al. (1996) was adopted to compile a new list of LaII lines for VALD.

Cerium. Bord et al. (1997) calculated the transition probabilities for CeIII lines using the atomic structure code of Cowan and they found good agreement between CeII and CeIII abundances for the silicon star HD200311. While the relative accuracy of the calculations can be estimated as $\pm$0.15dex, the absolute scale may be too large by 0.25dex. A total of 43 CeIII lines in the 2840Å to 6061Å spectral region was added to VALD.

Neodymium. For Ndi we again used the compilation of Komarovskij (1991). Oscillator strengths for NdIII lines became available from calculations by Cowley & Bord (1998) based on the Cowan code. The authors provide data for 54 lines in the 3280Å to 6870Å spectral region. The estimated relative and absolute errors are similar to CeIII.

Europium. An abundance analysis of this element forced us to implement oscillator strengths data in VALD which were determined from stellar and/or solar spectra. A few lines of EuIII were identified in spectra of magnetic Ap stars by Ryabchkova et al. (1999a). For 4 lines they determined astrophysical oscillator strengths with a relative accuracy of $\pm$0.2dex. The absolute scale could be more inaccurate, because of the dependency of the present determination on the ionization balance in stellar atmospheres.

Gadolinium. Komarovskij & Smirnov (1992) revised transition probabilities for Gdi on the basis of new lifetime and branching ratio measurements and extended the list of lines with experimentally measured oscillator strengths. Similar measurements by Bergstrom et al. (1988) provide accurate oscillator strengths for 23 lines of GdII which were used to improve the solar abundance of gadolinium. The authors give an error estimate for the stronger lines of 10%.

Dysprosium. Komarovskij & Smirnov (1994) deduced absolute oscillator strengths for 35 Dyi and 28 DyII lines using lifetime and branching ratio measurements with an estimated accuracy of 25%. For 13 Dyi lines the experimental data were obtained for the first time. The new Dyi list of VALD contains 42 lines and it also includes data extracted from the compilation of Komarovskij (1991).

Similar to Komarovskij & Smirnov (1994 - KS) absolute oscillator strengths measurements for DyII were carried out by Biémont & Lowe (1993 - BL) on the basis of LIF lifetime measurements and relative intensities taken from Kusz (1992). A total number of 63 lines was measured and with two exceptions the lifetimes determined by KS and BL agree very well. The oscillator strengths from BL are considered to be slightly more accurate, because the branching ratios used by BL were measured on spectra with higher resolution than those used by KS. Smirnov (private communication) corrected a few lines which showed the largest difference (in addition to those, for which lifetime measurements in KS were certainly wrong), and after this correction the difference between KS and BL measurements reduced to 0.01dex with a dispersion of 0.1dex. This dispersion corresponds exactly to the 25% accuracy claimed by KS. For the final list of oscillator strengths we used data from BL and one additional line from KS.

Erbium. Absolute experimental transition probabilities for 41 Eri lines were obtained by Komarovskij & Smirnov (1993) with an accuracy of about 25% using lifetimes and branching ratio measurements. For 22 lines experimental data were presented for the first time.

The spectrum of ErIII was re-analyzed by Wyart et al. (1997), who argued for new energy levels and transition probabilities based on a comparison of their line list with the spectrum of the Ap star HR465. The number of known energy levels was increased from 45 to 115 and the number of classified lines to 470. Oscillator strengths for 304 lines calculated with the Cowan code are included in VALD-2.

Thulium New LIF lifetime measurements (Andersen et al. 1996) together with the emission branching ratios resulted in oscillator strengths for 376 lines of Tmi and 146 lines of TmII in the 2500Å to 10000Å spectral region (Wickliffe & Lawler 1997b). The new measurements increase the existing data by 241 Tmi and by 30 TmII lines and yield a significant improvement of $\log gf$-values over the BELLHEAVY list. A new classification was possible for 3 lines of TmII, and for most of the measured lines the typical uncertainty is less than 10%. A comparison between BELLHEAVY and the new oscillator strengths is shown in Fig.4. The new data provide systematically lower $\log gf$-values than the old ones and the difference increases for weaker lines. For the TmII lines at 3362.6Å and 3462.2Å, used for the determination of the solar abundance of thulium (Andersen & Sørensen 1974), the new oscillator strengths agree within the claimed accuracy with the old values.

  

Figure 4: A comparison of the experimental oscillator strengths for TmII determined by Wickliffe & Lawler (1997b) with the BELLHEAVY data

Lutetium. Bord et al. (1998) reported oscillator strengths calculations for 24 lines of LuII using the Cowan code. They found on average their $\log gf$-values to be smaller by 0.27dex than those given by Corliss & Bozman (1962) for $\lambda<$ 3000Å, while for three lines with $\lambda\gt$ 5400Å their $\log gf$-values are larger by $\approx$0.5dex. Den Hartog et al. (1998) reported oscillator strengths for 3 lines of LuII at 3507.4Å, 5983.9Å and 6221.9Å obtained from LIF lifetime measurements combined with emission branching ratios with an accuracy of 0.06dex or better. Only the last line is in common with the list of Bord et al. (1998) and its measured oscillator strength is smaller by 0.16dex than the calculated value. A comparison between the new measurements and those by Corliss & Bozman proved the new $\log gf$'s to be larger by about 0.4dex for red and blue lines. On the other hand, the results of Bord et al. (1998) agree rather well with theoretical calculations of Migdalek & Baylis (1988). We therefore included the experimental oscillator strengths by Den Hartog et al. (1998) in the new list and supplemented it with the data from Bord et al. (1998). The expected errors of the latter data are at least 0.1dex with a possible shift of their absolute scale of about +0.15dex.

4.5 Elements heavier than Lu (Z > 71)

  Heavy elements such as Au, Pt, and Hg are important tracers for the study of diffusion processes in chemically peculiar stars, in particular for those of HgMn type. Recent laboratory data and new theoretical calculations allow now for more reliable abundance analyses using several ions of the same element. A total of about 1600 lines for elements heavier than Lu has been included into VALD-2.

Rhenium. The oscillator strengths for singly ionized rhenium used for VALD-1 were based on Corliss & Bozman (1962) data. Wahlgren et al. (1997) determined an accurate oscillator strength for one (2275.25Å) out of three lines of the ReII UV1 multiplet, using LIF lifetime measurements and emission branching ratios. The new value of $\log gf$ is included in VALD-2. It is smaller by 0.6dex than the corresponding value from Corliss & Bozman (1962). This fact has to be kept in mind, because we do not propose any corrections for the other lines from Corliss & Bozman (1962). Wahlgren et al. (1997) provide for all three lines of the UV1 multiplet also wavelength and intensity data for the hyperfine components of both stable isotopes of rhenium.

Platinum. Calculations of the transition probabilities for Pti lines in the 1730Å to 2540Å spectral region using the Cowan code were carried out by Wahlgren et al. (1995). Wavelengths and energy levels were taken from Blaise & Wyart (1992). In the same paper, Wahlgren et al. report on calculated oscillator strength for the PtII line at 2144.25Å which turned out to be slightly larger than the relative astrophysical $\log gf$-value derived by Dworetsky et al. (1984). Soon after, Wyart & Blaise (1995) published an extensive study of the PtII spectrum based on the atlas of Sansonetti et al. (1992). Wyart & Blaise calculated the transition probabilities for 112 lines in the 1380Å to 2800Å spectral region and also provided theoretical Landé factors for the odd energy levels. Their $\log gf$-value for the 2144.25Å line agrees nicely with the value of Dworetsky et al. and we therefore included all the data from Wyart & Blaise (1995) into VALD-2 together with the astrophysical $\log gf$-values from Dworetsky et al. (1984) for lines in the optical spectral region. For Pti the new list contains the oscillator strengths from Wahlgren et al. (1995). Ryabtsev et al. (1993) classified more than 800 PtIII lines in the range of 559Å to 2020Å and provided the transition probabilities for 666 lines, calculated with the Cowan code. Oscillator strength data for the Pt lines in three ionization stages were used in the abundance analysis of the HgMn star $\chi$Lup and gave a satisfactory agreement for the platinum abundance obtained from the lines of the different ions (Wahlgren et al. 1995). In total, we have included new line data for 14, 119, and 666 lines of Pti, PtII, and PtIII, respectively.

Gold. Similar to Pt, VALD contained information only for the lines of neutral gold. At the same time, lines of AuII and AuIII were also observed in the spectra of HgMn stars. The oscillator strength for the AuII 1740Å line was obtained by combining theoretical branching ratios with measured lifetimes (Wahlgren et al. 1995).

Rosberg & Wyart (1997) performed an extensive study of the AuII spectrum in the 800Å to 8000Å spectral region. They identified more than 500 spectral lines and calculated oscillator strengths for 497 lines using the Cowan codes. Their $\log gf$-value for the 1740Å line is higher by 0.3dex than the corresponding experimental value from Wahlgren et al. (1995). Using Rosberg & Wyart data for the optical lines Ryabchikova (1998a) recalculated the Au abundance in the atmosphere of $\chi$LupA and obtained log(Au/H) = -6.69 which agrees with the log(Au/H) = -6.74 deduced by Wahlgren et al. (1995). It means that the absolute oscillator strength scale for the optical AuII lines agrees better with the experimental oscillator strength for the 1740Å line than the scale for the UV lines. The reason for the 0.3dex difference is not clear yet. Thus, we included in VALD-2 all the data from Rosberg & Wyart with the exception of the AuII line at 1740Å for which we used the experimental $\log gf$-value from Wahlgren et al. (1995).

Wyart et al. (1996) identified more than 1000 lines of AuIII and calculated transition probabilities for 175 lines in the 800Å to 2000Å spectral region. Their $\log gf$-value for the 1746Å line is lower by 0.18dex than the corresponding value from Wahlgren et al. who have calculated the oscillator strength for this line. For VALD-2 we preferred the data from Wyart et al. (1996).

Mercury. The previous VALD version provided data for Hgi and HgII extracted from the BELLHEAVY line list. During the analysis of HgMn stars we found substantial inaccuracies of the wavelengths used in BELLHEAVY. Therefore, we produced a correction file for HgII lines with wavelengths taken from Reader & Sansonetti (1986) measured for the terrestrial mixture of Hg isotopes. In the new list of HgIII lines we included oscillator strengths for 42 spectral lines in the 740 - 3100Å spectral region calculated by Uylings et al. (1993) using the orthogonal operator method with configuration interaction.

Lead. VALD originally contained 11 lines of PbII with theoretical oscillator strengths mainly taken from Migdalek (1976). Meanwhile, Miller et al. (1979) and Alonso-Medina (1996, 1997) provided experimental measurements of the PbII oscillator strengths using emission branching ratios and lifetime measurements. Two different experiments of Alonso-Medina (hollow-cathode discharge and laser induced plasma) gave oscillator strengths which agree within 10%. The cited accuracy of Miller et al. data is lower. Five out of nine lines common to both lists agree within 15% while for 4 lines the difference in transition probabilities may be 2-3 times larger. We included in the VALD-2 list 37 PbII lines for which oscillator strengths were taken from Alonso-Medina (1996, 1997). Two lines, at 3665.5Å and 3945.7Å were rejected, because their wavelengths do not correspond to the proposed classification. Because the accuracy of the wavelengths is not good in Alonso-Medina, we used central wavelengths from Reader & Corliss (1980) whenever possible. Atomic energy levels and experimental Landé factors were taken from Moore (1958).