2 Structure and the new format of VALD line data

  The basic structure of VALD as described in Paperi remains unchanged: the data base is built from several lists of atomic line data published by various providers. These input lists (or source lists) are preserved separately and are at first ranked according to their known performance in applications (predominantly in astrophysics), as well as according to error estimates provided by the authors of the original data. Comparisons are made by the VALD team to decide on the final ranking. Lists are also re-ranked according to their actual performance as part of the data base or whenever new data are included which supply information about the same atomic (or, in the near future, also molecular) species.

Each individual request to the data base is handled by merging data from all relevant lists available. The merging procedure is performed by selecting each datum for an individual spectral line from the most highly ranked list which provides information on the particular atomic parameter (see Paperi). This selection is preferred over averaging of data, because individual errors of line data from different sources can vary dramatically, like oscillator strengths ($\log gf$-values) obtained from semi-empirical calculations (cf. Fig. 2) which would then be mixed up with high accuracy laboratory data. Moreover, for most of the line data individual error estimates are not available. Weighted averaging is recommended only when such estimates are known for all individual lines from two or more line lists which are supposed to be merged. On several occassions this has been done to create new VALD line lists which appear to the merging procedure of a VALD request as a single line list, although they were composed from several individual source lists (see Sect.4). Sometimes, line data had to be corrected for systematic deviations (usually but not always known from the literature) before becoming part of the VALD archive. Thus, it was possible to enlarge, for instance, the number of lines for which reliable oscillator strengths with individual error estimates are available.

  

Figure 2: Dependence of the differences in log(gf) calculated by Ekberg and by Kurucz for CrIII lines on the excitation energy of the lower level

The data base can be accessed by extraction programs (tools) described in Paperi. At the time of publication, VALD had already been upgraded substantially in comparison with its original installation (cf. the overview of Piskunov 1996). Besides of adding new line data this upgrade also included the application of compression techniques to some of the VALD tools and provision to treat an archive of many million spectral lines with limits only set by disk space and the speed of the host computer. More details about these and related improvements are given below. These enhancements have been indispensable for the computation of opacity distribution functions (Kupka & Piskunov 1998; Piskunov & Kupka 1999) and provide the basis for including molecules into the data base for both spectroscopic work and opacity calculations. In addition, more detailed information on the accuracy of oscillator strengths has become available for many spectral lines. As a consequence, it became necessary to extend the information stored with each spectral line archieved in VALD and a spectral line now is characterized by the following parameters (we refer to the introduction in Paperi):

1.

Central wavelength in Å.

2.

Species identifier. Provides element (or molecule) name and ionization stage.

3.

$\log gf$ - logarithm of the oscillator strength f times the statistical weight g of the lower energy level.

4.

E_i - excitation energy of the lower level (in eV).

5.

J_i - total angular momentum quantum number of the lower energy level.

6.

E_k - excitation energy of the upper level (in eV).

7.

J_k - total angular momentum quantum number of the upper energy level.

8.

g_i - Landé factor of the lower energy level; default value is 99, if no value can be provided.

9.

g_k - Landé factor of the upper energy level; default value is 99, if no value can be provided.

10.

$\log \Gamma_{\rm r}$ - logarithm of the radiation damping constant in s^-1; default value is 0, if no value can be provided.

11.

$\log \Gamma_{\rm s}$ - logarithm of the Stark damping constant in (s $N_{\rm e})^{-1}$ (i.e. per perturber) at 10000K; default value is 0, if no value can be provided.

12.

$\log \Gamma_{\rm w}$ - logarithm of the vanderWaals damping constant in (s $N_{\rm H})^{-1}$ (i.e. per perturber) at 10000K; default value is 0, if no value can be provided.

13.

Spectroscopic terms of lower and upper energy levels.

14.

Accuracy for $\log gf$ in dex (where available).

15.

Comments, e.g. multiplet number as in Martin et al. (1988).

16.

Flags. Will be used to provide a link to information on Zeeman patterns, on autoionization lines, to information available for computing more accurate Stark and vanderWaals broadening parameters, to supplement the main quantum numbers for hydrogen lines, and more.

We refer to this new format as "version 3.0 format'', because it already includes all the features necessary to handle molecular line data. The first 12 entries contain numerical data and occupy 52 bytes if they are stored as a sequence of uncompressed IEEE floating point numbers. The new species identifier allows for more flexibility and avoids a cumbersome extension of the ion identifier used in previous versions of VALD when molecules are included at a later stage (see Sect. 5). Following the 12 numerical parameters the next three parameters originally consisted of plain (ASCII) text fields for a total of 30 characters. Field 13 was enlarged to provide 24 characters altogether for both lower and upper energy levels to ensure that term designations for molecules can be stored properly (Greek letters will be indicated by a preceeding $\backslash$). As more precise estimates of oscillator strengths have become available since the first release of VALD, it was decided to replace the letter identifier for the accuracy of $\log gf$ as used in Martin et al. (1988) by indicating the error $\Delta$$\log gf$ in dex. Hence, the associated parameter field 14 now holds a numerical value (two bytes). Also, the "comments field'', now number 15, had to be enlarged significantly from 6 to 16 characters. It is used to provide

• accurate source descriptions for input lists compiled with data from different authors,
• multiplet designation (where available), and also to
• store now outdated, but still commonly used accuracy descriptors, like letters which were provided by the previous VALD version.

The parameter fields 13 and 15 are filled with blanks when no information is available. A value of -1 is inserted for the accuracy descriptor (parameter field 14) when no information is available. Finally, we decided to add two bytes for flag values. These will be used to indicate the availability of additional information in external databases like, for example, Zeeman patterns, Stark and vanderWaals broadening, and to mark special cases (as, for example, Hydrogen lines where Stark and vanderWaals constant fields are actually used to store energy levels, or autoionization lines for which the same fields contain parameters of the Fano profile). Typically, such information is available or needed for a small sub-set of lines. Thus, instead of an unnecessary increase of the data base or change of its internal format, we will provide a link to other data bases. Future VALD extraction tools are expected to use these flags when assembling the reply to a request but this transition will be transparent for the end user.

The impact of all these changes on the default output for VALD-EMS requests is minimal and guarantees as much continuity for the users as possible.