If we assume that the true continuum level is not wildly different from the fitted continuum the error in measured equivalent width can at the most be as large, in difficult cases, as 2 mÅ for weak lines, i.e.

20%, and 4-5 mÅ for stronger lines, i.e.

10%. This translates to typically 0.08 dex in the resulting abundance derived from a line of strength 10 mÅ and 0.04 dex for a line of strength
50 mÅ. Apart from the effects of the continuum errors and blends, the error in derived abundances due to errors in the measurement of the equivalent width of a line is negligible (< 0.01 dex). In general, the lines and continua are, due to the high S/N and high resolution, easy to fit and the errors given above should be regarded as upper limits.

5.2. Oscillator strengths

The oscillator strengths, derived from the observed solar spectrum, can be affected by misidentification, by undetected blends and by errors in continuum fitting and measurements of equivalent widths in the solar spectrum. As in the stellar spectra, location of the continuum is a much larger source of error than the actual measurement of a line. (Note, however, that the solar spectra have higher S/N, usually

, and thus, identification of the continuum becomes easier as well as identification of lines. The Sun is also more metal-poor than the programme stars, which makes identification of the continuum in the Sun easier.) Using the results in Sect. 5.1 (click here) we find that errors in

-values may be as large as 0.08 dex, but a more characteristic number is 0.04 dex.

Since we perform a purely differential analysis errors due to misplaced continua, neglected blends, etc. should partly cancel in the first approximation and not affect the resulting differential abundances very much, as long as we study stars similar to the Sun.

5.3. Blends

In the selection of lines we have carefully avoided all lines that could be subject to blending with nearby lines as given in Moore et al. (1966). For ions with several lines measured we have also looked for lines which produces abnormally high abundances as compared with the majority of the lines. This led us to exclude three FeII lines from our final analyses: 6383.71, 6383.45, and 6627.32 Å.

5.4. Fundamental parameters of the model atmospheres

Edvardsson et al. (1993a) estimate the error in the effective temperature due to errors in b - y to range from
-50 K to +100 K and the corresponding error in

to be

0.2 dex.

The effects of such errors in

and

are exemplified in Table 3 (click here). As expected, abundances derived from ions are most sensitive to changes in surface gravity while abundances derived from atoms are most sensitive to changes in effective temperature. In general, errors in derived abundances are smaller than 0.10 dex for atoms when varying the effective temperature by

K and less than 0.02 dex when the surface gravity is varied by

dex; they are smaller than 0.02 dex for ions when the effective temperature is varied and less than 0.10 dex when the surface gravity is varied.

5.5. Photometric versus spectroscopic metallicities

For dwarf stars that are significantly more metal-rich than the Sun (e.g. [Me/H]

0.2 dex), the metallicity used in the calculation of the model atmosphere is important, since it governs the line blanketing and thus the temperature structure of the model atmosphere. We may expect that final derived abundances are sensitive to this parameter. Following Edvardsson et al. (1993a) we decreased (and in a few cases increased) the metallicities to the values derived for [Fe/H] in a preliminary abundance analysis and reiterated the abundance determination. (We have determined iron abundances from lines arising from both FeI and FeII. For FeI usually more than 30 lines were analysed and for FeII three to four lines. The formal error in the mean [Fe/H] derived from FeI for a certain star is usually smaller than 0.02 dex.) In the mean we find that we had to reduce the abundances by 0.07 dex from the initial photometric values, with a spread of 0.01 dex, Table 1 (click here). Apparently, our photometric metallicities tend to overestimate the metal content in these metal-rich stars as compared to the iron abundance derived from spectral abundance analysis, Fig. 2 (click here).

Figure 2: Metallicities determined from photometry vs. iron abundances derived in our spectral analysis. The one-to-one relation is indicated by a dotted line and a least-square fit to the data points, taking the error in [Fe/H] to be

0.02 and the error in the photometric metallicities to be

0.1, is also shown, dashed line. The

-proabability for this fit is 0.92. A fit made without taking the errors into account is shown by a long-dashed line. The stars from Barbuy & Grenon (1990),

, and HD 171999A and HD 175518, denoted by

symbols, were excluded from the fit (see Sect. 5.5 (click here))

Usually, the differences between the metallicities estimated from photometry, are close enough to those derived from spectroscopy that we have not found it necessary to reiterate the determination of effective temperature and

. (The determination of metallicity was, however, changed according to the spectroscopic result so that the final model used in the analysis had [Me/H] consistent with the resulting spectroscopic [Fe/H].) However, for some of the stars the photometry indicates rather extreme metallicities compared with the spectroscopic iron abundances. For HD 171999A we have only measured 6 FeI equivalent widths (since this star was only observed with one CCD setting, see Sect. 3.1 (click here)) and thus the spectroscopically determined iron abundance is not as good as for the other stars. However, we note that the line-to-line scatter is small, 0.03 dex. For HD 175518 it is questionable if its photometric metallicity is realistic. In any large catalogue there will always be a few stars with

errors in the photometry. Since spectroscopic iron abundances indicate a much lower metallicity this is most probably an example of that.

We have studied, for HD 175518, the effects on derived abundances if [Me/H] is lowered to 0.2 dex as indicated by the spectroscopy, thus affecting the estimates of the rest of the fundamental parameters. The lower metallicity implies a lower effective temperature; [Me/H] = 0.22 dex corresponds to a decrease of

200 K. From Table 3 (click here) we see that most elements will change by

0.1 dex and thus the star will mainly move horizontally, by

0.1 dex, in our relative abundance diagrams. HD 175518 is an extreme case in our sample and the abundances of the other stars and general abundance trends for the whole sample should not be affected by comparable amounts.

5.6. Effects of hyperfine structure

Some of the elements analysed are well known to be subject to hyperfine structure. We have not taken this into account when deriving the chemical abundances. Instead, for those elements in particular we have aimed at selecting weak enough lines, so that the neglect of hyperfine structure in the abundance analysis should not affect the calculated equivalent widths and thus not the derived abundances.

Copper.

Figure 3 (click here) shows copper abundances derived from the two lines used in our study as functions of equivalent width. Our data clearly show that the stronger copper line is subject to hyperfine structure and that these lines should be analysed using synthetic spectroscopy taking the hyperfine structure splitting into account. Results by Summers (1994) may suggest departures from LTE in the population of levels in the copper atom. We do not, however, have enough data to make further empirical investigations of such departures from LTE. We omit copper abundances from the following discussion.

Figure 3: Copper abundances derived from the lines at 5220 Å,

symbols, and 7933 Å,

symbols. The widths of these lines are 16 and 36 mÅ, respectively, in the solar flux spectrum

Manganese and Cobalt.

The manganese and cobalt lines used in this study are not saturated. When plotting abundances derived from each line as a function of equivalent width no distinct pattern was found; indicating that the omission of hyperfine structure in the treatment of the lines is not problematic. In many stars lines with equivalent widths of about 20 and 60 mÅ, respectively, gave manganese abundances that are in excellent mutual agreement.

r- and s-process elements.

The lines used to derive abundances for the heavy s- and r-process elements are sufficiently weak in our programme stars to be safely used as abundance criteria in spite of being subject to hyperfine structure. Among the light s-process elements we note, however, that only YII, and maybe YI, have lines strong enough and secure enough that derived abundances can be used with confidence. The

-values for zirconium are very uncertain because of the faintness of the lines in the Sun.

5.7. Checks on Boltzmann and Saha equilibria

Deviations from the Boltzmann excitation equilibrium, which might suggest an error in the effective temperature, can be traced by studying abundances derived from individual lines as a function of the lower excitation potential for the lines. 32 FeI lines measured in most of the stars, marked in Table 2 (click here), were selected for this purpose. For all stars (excluding four stars with too few lines observed) the abundance of each iron line was plotted as a function of the excitation energy of the lower level,

, and a least square linear fit was made to the points of the form [Fe/H]=

, Fig. 4 (click here). The mean value of the slope coefficients, k, is 0.00 (excluding the K dwarf stars), suggesting that the systematic errors in effective temperatures are small.

Figure 4: Two examples of how the slope coefficients in Fig. 5 (click here) were obtained. Examples are for a) HD 91204 and b) HD 125968. The dashed lines represent linear least-square linear fits

Next, these linear coefficients were plotted as functions of effective temperature, surface gravity and spectroscopically derived iron abundance, Fig. 5 (click here). We have compared our results with those obtained by Edvardsson et al. (1993a) (their Figs. 9a-f) and find that the two studies span the same range of excitation energy slopes, k. There are small but systematic deviations from the excitation equilibrium, varying with effective temperature. A change in

of +100 K results in a

of +0.006 dex eV^-1. The change of excitation energy slope with surface gravity estimates seems to reflect the variation of surface gravity with effective temperature (surface gravity increases as effective temperature decreases).

Figure 5: The slope coefficients from the excitation energy balance diagram for each star plotted vs. effective temperature, surface gravity and spectroscopically derived iron abundances. The star with the lowest k is HD 180890. Linear least square fits yield: a)

, b)

, c)

[Fe/H]

In many studies surface gravities are determined by requiring ionization equilibrium. This is typically made by changing the surface gravity until the iron abundances derived from FeI and FeII lines yield the same abundance. We have not, as discussed earlier, used this method. As we will see this has led us to discover what appears to be a case of significant overionization in K dwarf stars and an opposite smaller effect for the hotter stars (see Fig. 7 (click here)).

What would the effects be if we assumed ionization equilibrium, and adjusted the surface gravities accordingly? We can estimate changes in the stellar abundances from the results of Table 3 (click here) and Table 11 (click here). From this we find that half of the stars should have their surface gravities increased by 0.25 - 0.35 dex to achieve ionization equilibrium for iron. This means that the the iron abundance will change with

. Abundances of other elements will change with similar amounts but with differing signs, which means that for some elements [X/Fe] will change by up to 0.1 dex and for others not at all. However, we note that the oxygen abundances are very sensitive to the surface gravity and may change by up to 0.2 dex. As a comparison we estimate the maximum error in the derived oxygen abundance caused by incorrectly set continua to be less than 0.1 dex.

An adjustment downwards of the gravities by about 0.3 dex would increase the conflict with the gravity estimates from the CaI 6162 Å line wings. We consider such a revision less probable.

5.8. K dwarf stars - overionization

Our results admit a comparison for five elements (scandium, vanadium, chromium, iron and yttrium) of abundances derived from ions to abundances derived from the corresponding atom, as function of effective temperature, within a rather wide range of effective temperature.

We find an apparent overionization as compared to expectations from LTE calculations for the five K dwarf stars in our study, Fig. 7 (click here). Our results are at first sight unexpected, especially for the two K dwarf stars with iron abundances of

0.3 dex relative to the solar iron abundance, as derived from FeI lines. In stars more metal-rich than the Sun the photoionizing non-local UV-flux is more efficiently blocked than in more metal-poor stars. However, as discussed in Sect. 5.9 (click here), the stronger blocking may be more than compensated by the increased temperature gradient which may enhance the non-locality of the radiation field.

We have carefully inspected the CrII and FeII lines in the K dwarf spectra and excluded all lines which may be subject to severe blends in these cool stars. For FeII we used the lines at 6456.39 and 6516.39 Å and for CrII the lines at 5305.86 and 5310.69 Å. In spite of using blend free lines the apparent overionization remains. The internal consistency between abundances derived from the two FeII lines is very good and this is also the case for CrII.

For scandium, vanadium and yttrium blends remain a possible source of error, but the similarity with the trends for iron and chromium suggests a common cause of the apparent overionization for all these elements.

The iron abundances derived from the atom show no obvious trend with effective temperature, see Fig. 6 (click here).

Figure 6: Iron abundances, derived in this study, as a function of effective temperatures

A probable reason for these effects is overionization (see Sect. 5.9 (click here)). Before discussing this, however, we shall explore the possibility that errors in the temperature scale could also contribute significantly.

5.8.1. Errors in effective temperatures for K dwarf stars

The calibration of the photometry in Olsen (1984) is, for

, based on a sample of 15 K and M dwarf stars using stellar parameters from Cayrel de Strobel & Bentolila (1983). The K and M dwarf stars in Olsen's sample span a small range in

and

. This is reflected in the change in the calibration at

. For

the calibration is degenerate in metallicity. Olsen (1984) quotes an error of

100 K for the effective temperature as derived from b-y.

Additional photometry is scarce for our K dwarf stars. However, UBV photometry exists and we have checked our effective temperatures using the calibration of B-V by Johnson (1966), Table 4 (click here). These effective temperatures agree well with those obtained from

photometry. We note, however, that the increased blocking for metal-rich stars, as compared to the calibration stars used by Johnson (1966), may cause the present effective temperatures to be underestimated. We have also derived (crude) effective temperatures from calibrations of the MK classification. Comparing the effective temperatures derived from photometry with calibrations of effective temperatures as functions of spectral classification (Bell & Gustafsson 1989 and Johnson 1966) we estimate

-values that deviate as compared with our standard values as indicated in Table 4 (click here), Cols. 4 and 5.

We have searched the literature for independent derivations of the effective temperature for these stars. Those found agree well with the photometric results, Table 4 (click here). As described earlier we have also used the excitation energy balance to check our effective temperatures. A change in effective temperature of +200 K for HD 32147 changed the slope of the least-square fit to the data points in the diagram abundance-versus-excitation energy from +0.04 to +0.00. A change of -200 K brought about a similar change but in the opposite direction. Thus, the excitation equilibrium indicates that +200 K is an acceptable change of the effective temperature for this star. The line-to-line scatter in derived abundances from FeI lines for this star is among the largest,

0.13. The same changes in effective temperatures give similar values for HD 61606A. Since the other three K dwarf stars are in the same effective temperature range and have similar surface gravities as the two stars discussed here, and since HD 32147 is the most metal-rich and HD 61606A is the most metal-poor of the five K dwarf stars, changes in the fundamental parameters of the remaining three stars will produce similar changes in abundances.

To conclude, a change in effective temperature of +200 to +400 K may be allowed as judged from the excitation equilibrium. As is obvious from Fig. 7 (click here), a change of this order of magnitude would restore the LTE ionization balance for iron and chromium. The large line-to-line scatter in derived abundances for lines with high excitation energies makes attempts to derive effective temperatures from excitation equilibria very dependent on one or two points in the lower end of the excitation energy range, spanned by the lines as illustrated in Fig. 4 (click here). Effective temperatures derived in other studies and from B-V colours deviate by less than this from our values.

Figure 7: Abundance ratios of chromium and iron as functions of [Fe/H]. [CrII/CrI], denotes [Cr/H]

as determined from CrII lines minus [Cr/H]

as determined from CrI lines, and similarly for iron. Results for chromium are shown in panel a) and for iron in panel b)

symbols denote the K dwarf stars and

symbols denote the stars from Barbuy & Grenon (1990). We exemplify, with HD 61606A (

)and HD 103932 (

), how the resulting loci of the K dwarf stars will be shifted in the diagram if the parameters of the stellar model atmospheres are changed; changes of

K (solid lines),

dex (dashed lines) and [Fe/H]

dex (dotted lines) are shown. Filled triangles denote increased values of the respective parameters while filled squares denote decreased values

As already noted the surface gravities for the K dwarf stars seem rather well determined, see Table 1 (click here). We also note that our

value for HD 131977A agrees well with that given by Morell (1994). We conclude that realistic errors in

are not enough to account for the departure from ionization equilibrium.

To conclude, we cannot from our analysis exclude that the apparent pattern of overionization, at least partially, is due to a temperature scale that is several hundred K too low. However, our analysis, together with evidence from other studies, suggest that deviations from LTE is a more plausible cause for the effects.

5.9. Non-LTE

No detailed study has been devoted to the non-LTE effects on abundance determinations for metal-rich dwarf stars, cooler than the Sun. A general result of the available studies for solar-type stars is, however, that several different effects are at play and may counteract each other, and this makes all extrapolation to the present study of metal-rich dwarf stars from studies of other types of stars or studies of other elements questionable.

Among the significant effects are (cf. Bruls et al. 1992) resonance line scattering, photon suction, ultraviolet overionization, (infra)red over-recombination and optical line pumping. The resonance-line scattering effects, in which photon losses cause the source functions of resonance lines to drop far below the Planck function at depths greater than those where the line optical depth is unity, may lead to severe overestimates of abundances - e.g., Carlsson et al. (1994) find that in Li-rich cool stars the Li abundance may well be overestimated by a factor of 3 as a result of this.

Photon suction may, for metal-rich cases in particular, lead to overpopulation of, e.g., the ground state and thus inhibit the effects of overionization. This is the result of a compensation of photon losses in the upper photosphere in resonance lines, as well as in connected ladders of transitions, by a downward population replenishment flow from the continuum reservoir. It is of great significance for atoms with a majority of corresponding ions and with pronounced cascade ladders. For complex atoms it should be of greatest significance for the high-lying levels that thus can compensate population depletion processes at lower excitation energy, e.g. caused by overionization.

The ultraviolet overionization has been a major worry in analyses of late type stellar spectra for two decades. It arises because the mean intensity

drops below the Planck function

in the line-forming regions of the atmosphere on the blue side of the spectrum peak. Overionization is known to occur for FeI in the Sun from levels a few electron volts below the continuum (see Rutten 1988 and references therein) and may more or less effect other metals as well (see, e.g., Baumueller & Gehren 1996; Bruls 1993). Overionization was suggested by Auman & Woodrow (1975) to be significant for a number of elements with lower ionization energies in cool stars. Major problems in modeling it are, however, the difficulties in predicting the ultraviolet flux of late-type stars with the crowding of spectral lines and the possible existence of an "unknown opacity'' (cf., e.g. Gustafsson 1995), as well as the dependence of the results of the notoriously uncertain collision cross sections, e.g. for collisions with H atoms (cf. Steenbock & Holweger 1984). Empirically, #M&Mäckle et al. (1975) and Ruland et al. (1980), found a tendency for the low-excitation lines (

eV) of FeI and TiI in K giant star spectra to give systematically lower abundances than the high-excitation lines (

eV). The abundance difference was typically found to be 0.15 dex. Subsequently, Steenbock (1985) succeeded in reproducing this result with statistical-equilibrium calculations. He found the effect to mainly reflect an overionization in upper layers (notably

) of the atmospheres, where the low excitation lines are formed. The effect is much smaller in the solar spectrum, leading to systematic errors in a differential analysis where red giants are compared with the Sun.

For metal-rich stars, the blocking by the crowd of spectral lines in the ultraviolet could be expected to - at first sight - strongly reduce the overionization effect, but this may be compensated for by a steeper temperature gradient in their atmospheres as a result of line blanketing effects, as in the case of LiI, Carlsson et al. (1994), or of CaI, Drake (1991). The latter study is particularly instructive for judging the results of the present investigation. Drake finds that for G and K-type stars the overionization effects on CaI abundances increase with decreasing effective temperature, with increasing acceleration of gravity and with increasing metallicity. At least the last two results may seem contrary to intuition. They reflect the significance of HI absorption shortwards of the Balmer discontinuity, which blocks more of the ionizing UV flux for the giants than for the dwarfs, and the afore-mentioned effects of metal-line blanketing on the temperature structure. For the K dwarf stars the effects on CaI abundances may, according to Drake's results, well result in an underestimate by a factor of two or more if LTE is assumed.

Over-recombination is important for photoionization transitions from levels close to the continuum (i.e. in the infrared), since for them the angle-averaged

may drop below the local Planck function deep in the photosphere. This may produce net recombinations, and overpopulation of the upper levels.

Optical (ultraviolet) pumping occurs in strong lines, e.g. the resonance lines and is analogous to overionization in that it is driven by

. It is important, not the least when it occurs in ultraviolet resonance lines and excites the atoms to states which may be much more easily photoionization due to a much richer radiation fields available at longer wavelengths, as was early suggested by Aumann & Woodrow (1975). This is most important for trace elements, and for metal-poor stars.

The complex interplay between these different mechanisms affects most levels of the atom, at great atmospheric depths, for atoms where the strong lines get efficiently optically thin in the photosphere, i.e. for relatively rare elements like the alkalis, while for the more abundant atoms like Fe and Mg the stronger, e.g. resonance, transitions are in detailed balance through most of the photosphere. For these, the relatively simple overionization phenomenon is probably dominating, except for transitions very close to the continuum for which e.g. photon suction may be significant.

In a recent study Gratton et al. (in prep.) have used detailed statistical-equilibrium calculations to explore the departures from LTE for solar-type dwarfs as well as for red giants of different metallicities, and their effects on abundance determinations for O, Na, Mg, and Fe. These authors find relatively small effects for stars cooler than the Sun for OI - LTE abundances from the IR triplet lines should be corrected downwards by less than 0.1 dex for stars with

. For NaI the subordinate lines are weakened by overionization and cascade by about 0.1 dex for the solar-type dwarfs. The dominating effect for MgI is overionization, and the non-LTE abundance corrections are thus generally positive. Typically the corrections are 0.1 dex in the dwarf stars. For FeI, where again overionization is dominating the abundance corrections

dex. Most of these effects are found to be greater for

. It should be noted, however, that models for metal-rich dwarfs with

were not included in this study.

Summing up the discussion of non-LTE we conclude that the effects on abundances are expected to be mainly due to overionization for most of the elements. For the alkali atoms, as well as for the rare earths, more complex effects may also be significant. Typically, errors of about 0.1 dex may be expected in the differential results but the complexity of the interplay between different effects, and in particular the results obtained by Drake (1991) for CaI, suggest that greater effects may be present, in particular for the metal-rich K dwarf stars.

5.10. Collecting errors

We have shown that errors in fundamental parameters give errors in mean resulting abundances of less than 0.1 dex. For elemental abundances derived from several lines this may be the dominating error, while for abundances derived from one single line errors due to blends and fitting of continua may be the main contributors to the overall error. Deviations from LTE in the excitation and ionization balance may also be of importance, probably more so for abundances based on few lines, in particular for the K dwarf stars. We collect our best estimates of errors due to different sources in Table 5 (click here).

5.11. Comparison of results for stars in common with other studies

The majority of our stars have not been studied before through spectroscopic abundance analysis.

For those of our stars (HD 30562, HD 32147, HD 67228, HD 1319777, HD 182572, HD 186427) that are in the catalogue by Cayrel de Strobel et al. (1997) the agreement between iron abundances derived in this study and those listed in the catalogue is good, cf. Table 6 (click here).

HD 32147 has been given much attention in the discussion of Super Metal Rich (SMR) stars. SMR stars have been defined as stars with

dex (for a discussion and references on SMR stars see Taylor 1996). Low resolution work and photometric determinations of [Fe/H] have been carried out for this stars, but this is, to our knowledge, the first high dispersion analyses of the star. Our [Fe/H] of 0.28 dex implies that this star is really an SMR star.

Our results for HD 182572 are compared with the results of the detailed analysis by McWilliam (1990) in Table 7 (click here). We note that considerable discrepancies remain even after correcting for the difference in effective temperatures.

	NaI	AlI	SiI	CaI
Friel et al.	0.07	0.12	0.060.03	0.070.05
This work	0.130.04	0.120.04	0.100.04	0.020.09
	TiI	FeI	FeII	NiI
	0.100.04	0.050.03	0.020.04	0.050.03
	0.070.10	0.050.05	-0.060.04	0.040.10

Table 8: Comparison of derived abundances, [X/H], between Friel et al. (1993) and this work for HD 186427, 16 Cyg B. Friel et al. use the following stellar parameters (T

/log g/[Fe/H])=(5770/4.30/0.05) and we use (5773/4.17/0.12)

We have analysed three stars previously studied by Edvardsson et al. (1993a). The results are compared in Table 9 (click here). For HD 30562 and HD 144585 the results agree to within the errors quoted. The results for HD 67228 show larger discrepancies than the other stars for magnesium, silicon, calcium and iron. A higher microturbulence parameter, as used in Edvardsson et al. (1993a), would decrease our results by 0.01 or 0.02 dex (ionized iron by
0.04 dex). Edvardsson et al. (1993a) find that a change in

by +100 K gives an iron abundance 0.06 higher for their stars. Thus, the increase of 52 K needed to transform the results of Edvardsson et al. (1993a) to our temperature scale means an increase of their iron abundance by 0.03 dex. Also the silicon and calcium abundances are affected in the same way as iron while all the other abundances remain as before, within the errors, in the two studies. The loggf-values agree well (0.06 difference, we have the higher value). The spectrum we have obtained for this star is of high quality (S/N

200). We do not find the discrepancy between the two studies alarming. HD 67228 has also been studied by Andersen et al. (1984) in a study on lithium isotope ratios in F and G dwarf stars. They derive an [Fe/H] of 0.05 from spectral lines using a model with (

/log g/[Fe/H])=(5850/4.2/0.05).

Two stars from our sample, HD 30562 and HD 67228, were recently analysed in detail by Tomkin et al. (1997) on the basis of different spectra; however, obtained with the same instrument and analysed independently with model atmospheres computed with the same computer program. The results are compared with those of our analyses in Table 10 (click here). In view of the errors in these analyses we find the agreement satisfactory.

HD 186427 (16 Cyg B) have been extensively studied, in particular in connection with searches for solar twins. A recent spectroscopic study has been performed by Friel et al. (1993). The results are in good agreement, Table 8 (click here). However, we find lower iron abundances derived from FeII lines than Friel et al. (1993) do. This difference is probably mainly due to the different surface gravities used.

To conclude, we find that, for those few stars in our programme in common with other studies, abundances are rather well reproduced. This gives confidence when we now apply our results to the exploration of the chemical evolution of the Galaxy.

ID /	[OI]	NaI	MgI	AlI	SI	CaI	TiI	Cr I	CrII	FeI	FeII	NiI	EuII
HD 72946	0.16	0.22	0.24	0.23	0.13	0.22	0.38	0.24	0.09	0.24	0.00	0.18	0.09
+100 K	+0.01	+0.04	+0.04	+0.06	+0.01	+0.06	+0.08	+0.06	-0.03	+0.05	+0.01	+0.05	0.00
+0.2 dex	+0.09	-0.03	-0.01	-0.01	-0.01	-0.02	0.00	0.00	+0.08	-0.01	+0.12	+0.01	+0.08
HD 103932	0.28		0.30	0.14	0.27	-0.12	+0.03	0.01	0.44	0.14	0.57	0.41	0.32
+200 K	+0.05		-0.08	+0.12	-0.18	+0.18	+0.24	+0.12	-0.18	-0.04	-0.32	-0.05	-0.01
+0.4 dex	+0.16		+0.02	-0.04	+0.10	-0.04	-0.02	+0.02	+0.19	+0.07	+0.21	+0.09	+0.15
HD 110010	-0.04	0.53	0.48	0.40	0.51	0.35	0.29	0.30	0.22	0.35	0.33	0.31	0.16
+100 K	+0.01	+0.05	+0.04	+0.04	+0.01	+0.06	+0.08	+0.06	+0.03	+0.06	-0.05	+0.05	0.00
+0.2 dex	+0.09	-0.02	-0.01	-0.02	-0.01	-0.01	0.00	0.00	+0.08	-0.01	-0.03	0.00	+0.09

		Spectral	Bell &	Johnson	Johnson		Neff	Morell	Arribas &
		class	Gustafsson				et al.		Martinez Roger
ID					B-V
HD 32147	4625	K3V	+200	+100	1.06	4619	4570		4670150
HD 61606A	4833	K2V	+300	+100	0.96	4863
HD 103932	4510	K5V	-300	-100	1.12	4473
HD 131977A	4585	K4V	-100	0	1.10	4522	4575	4570
HD 136834	4765	K3V	+100	0	1.00	4765

Source of error	Error in resulting relative abundance
Measurement of	negligible
Continuum fitting	< 0.09 dex, usually 0.05 dex
	< 0.1 dex
Non-LTE effects	0.1 - 0.2 dex?
Oscillator strengths	<0.1 dex

	Mc William	This work
		5739 K	5380 K
FeI	0.31	0.42	0.24
SiI	0.28	0.51	0.47
CaI	-0.11	0.42	0.21
ScII	0.14	0.36	0.36
TiI	-0.02	0.50	0.21
VI	-0.02	0.44	0.12
CoI	0.18	0.58	0.29
NiI	0.00	0.46	0.24
EuII	0.18	0.13	0.13

	HD 30562		HD 67228		HD 144585
		Diff.		Diff.		Diff.
[OI]			0.21
NaI	0.21	+0.01	0.23	+0.06	0.36	0.00
MgI	0.33	-0.01	0.23	+0.08
AlI	0.25	-0.07	0.22	+0.03
SiI	0.21	-0.02	0.27	+0.11	0.24	-0.03
CaI	0.17	-0.01	0.15	+0.12	0.24	-0.03
TiI	0.13	-0.02	0.09	+0.02	0.29	+0.02
FeI	0.19	+0.05	0.16	+0.12	0.27	+0.04
FeII	0.16	+0.10	0.22	+0.06	0.11	-0.05
NiI	0.17	0.00	0.14	-0.01	0.32	+0.06
YII	0.09		0.12	0.00	0.06	+0.06
NdII			-0.10		-0.14	-0.06

5. Errors in resulting abundances

5.1. Measurement of equivalent widths and fitting of continua