The first one is a "classical'' iterative LTE analysis. Measured equivalent widths of 10 different elements (among which 28 Fe I lines, spanning a broad interval in excitation potential and equivalent widths, and 6 Fe II transitions) were converted in abundances, using the MOOG (Sneden 1973) software packages. The complete list of lines used is given in Table 2, while their selection is discussed in Sect. 3.2. MOOG uses model atmospheres and atomic data (wavelength, excitation potential, gf values) to compute theoretical curves of growth, from which the abundances are derived. Kurucz LTE plane parallel models (Kurucz 1993a, see Sect. 3.5) and Kurucz atomic data (Kurucz & Bell 1995) were used, except for gf values which were, in majority, adjusted by comparison with the solar spectrum (Sect. 3.3).

The atmospheric parameters and abundances were obtained by iteratively modifying the effective temperature, surface gravity, metallicity, mean alpha element overabundance and micro-turbulence velocity of the input model and rederiving the abundances until (i) the Fe I transition abundances exhibited no trend with excitation potential or logarithm of reduced equivalent width, (ii) Fe I and Fe II lines gave the same average abundances (ionisation equilibrium) and (iii) the iron and alpha elements average abundances were consistent with those of the input model atmosphere.

Each of these diagnostics is sensitive to changes of several of the input atmospheric parameters. Given a set of observed equivalent widths, the slope of Fe I abundance as a function of excitation potential decreases with effective temperature, is very slightly sensitive to surface gravity and metallicity and increases with micro-turbulence. The slope of Fe I abundances as a function of logarithm of reduced equivalent width increases with effective temperature, decreases with surface gravity, is very slightly sensitive to metallicity and strongly decreases with micro-turbulence. Around the convergence point, the mean Fe I abundance is weakly sensitive to changes of the atmospheric parameters, while the mean Fe II abundance decreases significantly with effective temperature and increases significantly with surface gravity.

"Low excitation potential'' neutral iron group transitions have been reported to yield, in giant stars, lower abundances than their "high excitation'' counterparts (Ruland et al. 1980; Drake & Smith 1991). This behaviour is usually attributed to non-LTE effects due to the low density of the photosphere. As presented in Sect. 5, the first analysis of IS Vir and V851 Cen leads to the result that both of them are giants. It was therefore necessary to test whether the results of the first analysis had been affected by non-LTE effects, via the "low excitation potential'' transitions. Both stars were then reanalysed without the "low excitation potential'' transitions. Following the Ruland et al. (1980) study of LTE departure as a function of excitation potential, all Fe I lines with $\chi < 3.5$ eV were discarded from the initial selection. This, of course, makes it impossible to rely on the slope of the Fe I transition abundances as a function of excitation potential to constrain the atmospheric parameters, because the remaining interval, $3.6 \leq \chi \leq 5.0$ eV, is too limited. Another "diagnostic'' was therefore needed to avoid the LTE analysis solution being degenerated.

In IS Vir and V851 Cen second analysis, photometric colour indices were used. The effective temperatures were derived from the B-V index for IS Vir, and from the B-Vand V-I indices for V851 Cen (the two colours leading to different temperatures, V851 Cen was analysed two times, using each index separately). Surface gravities, micro-turbulences and abundances were estimated in the same way as in the first method, using the same set of lines (but excluding the "low excitation potential'' Fe I transitions). In the case of the B-V/temperature transformation, which is metallicity sensitive, the two steps were iterated until convergence. The conversion of photometric indices into effective temperatures, as well as the consistency of different colour indices as RS CVn stars temperature estimator is discussed in more details in Sect. 3.6.

As discussed in Sect. 3.6, the photometric indices present a major drawback: in several RS CVn stars, different colours yield different effective temperatures. Therefore, analyses based on photometric indices are not consistent and reliable enough to identify any possible non-LTE effect that might affect the first method. IS Vir and V851 Cen were thus analysed a third time.

Instead of the "low excitation potential'' Fe I transitions or the photometric colour indices, the third method makes use of information contained in the wings of the 6162 Å Ca I transition. In IS Vir and V851 Cen, the 6162 Å Ca I line is a "strong'' line, and therefore its wing profiles are strongly sensitive to surface gravity. As the 6162 Å Ca I wing profiles are also sensitive to effective temperature, micro-turbulence velocity and iron and calcium abundances, the analysis was performed in an iterative way. The atmospheric parameters determined with the second method were used as starting parameters. In a first step, the surface gravity was derived by comparing the 6162 Å Ca I line wing profiles to a library of synthetic profiles (this step is detailed in Sect. 3.7). In a second step, the measured equivalent widths of the set of lines used in the second method were converted to abundances, using the LTE analysis program MOOG. The surface gravity of the MOOG input atmospheric model was the one derived during the first step. Effective temperature, surface gravity, micro-turbulence velocity and abundances were obtained by iteratively modifying the input $T_{\rm eff}$ , $\xi$ , ${\rm [Fe/H]}$ and ${\rm [Ca/H]}$ and rerunning the two steps until (i) the Fe I transitions exhibited no trend with logarithm of reduced equivalent width, (ii) Fe I and Fe II lines gave the same average abundances and (iii) the iron and calcium average abundances were consistent with the input abundances.

Another "classical diagnostic'' is often used to derive the effective temperature: the adjustment of the Balmer lines wing profiles. In F- and G-type stars, the wings of the Balmer lines are very sensitive to the effective temperature (the width of the wings monotically increasing with $T_{\rm eff}$ ), while very slightly sensitive to the surface gravity or the metallicity. As a consequence Balmer lines are very good temperature indicators above $T_{\rm eff} \simeq 5000$ K (Cowley & Castelli 2002), and were effectively used by, e.g., Ottmann et al. (1998) in their analysis of hotter active binaries. Below this threshold, Balmer line wings become less temperature sensitive and are no longer useful as estimators. Since most of the 18 SB1 stars in our present sample are classified early K-type stars, Balmer line profile adjustment could not be used to derive their temperatures. However, even at higher temperature, the presence of activity (filled in lines, or emission) makes it necessary to be very careful in the use of Balmer lines as temperature indicators in RS CVn stars.

3.2 Lines selection

Both IS Vir and V851 Cen are relatively cool and therefore exhibit a lot of blended features. A first selection of unblended or "slightly'' blended lines (lines that can be reliably deblended) has therefore been performed starting from a synthetic spectrum of parameters $T_{\rm eff} = 4500$ K, $\log g = 4.0$ and ${\rm [Fe/H]} = 0.0$ , covering the wavelength range 5000-8000 Å (the bluer part of the spectrum suffers so heavily from blending as to be effectively useless in the present context). This spectrum was computed with Piskunov's SYNTH program using atomic parameters from the VALD database (Piskunov et al. 1995; Kupka et al. 1999; Kupka et al. 2000) and a Kurucz model atmosphere (Kurucz 1993a). The lines chosen in the synthetic spectrum were then examined, one by one, in the two sample spectra. For each star, lines which appeared asymmetric or showed an unusually large width, were assumed blended with unidentified lines and therefore discarded from the initial sample. Lines whose profiles overlapped with the position of one of the telluric lines listed in "Rowland's table of the solar spectrum'' (Moore et al. 1966) were also discarded. Note that the lines concerned are not necessarily the same from star to star because of the radial velocity differences.

Ideally, one would like to rely only on weak lines (lines on the linear part or close to the knee of the curve of growth, i.e. approximately W < 70 mÅ) to perform detailed analysis. Unfortunately, in the case of "cool'' stars exhibiting a moderate rotational velocity, the number of reliable and unblended lines is rather small. Even for the two slow rotators analysed here, IS Vir and V851 Cen, Na and Mg are derived from "strong'' lines (respectively around 100 and 150 mÅ). In order to quantify the effect of "strong'' lines on the results, V851 Cen has been analysed (method 3) with two sets of Fe I lines: the first made of 20 lines with $W \leq 70$ mÅ, the second made of these plus 11 lines with $70 < W \leq 163$ mÅ (see Table 2). The two studies give very similar results, with negligible differences ( $\Delta T_{\rm eff} = 0$ , $\Delta \log g = +0.03$ , $\Delta \xi = +0.1$ km s^-1, $\Delta {\rm [Fe/H]} = +0.01$ and a few hundredths of dex for the other elements; the largest difference being ${\rm [Ca/Fe]} = -0.05$ dex). The results presented in Table 7 have been obtained with the large set of lines: "weak'' plus "strong''.

As many damping constants are poorly known, all the lines that may have been significantly van der Walls broadened ( $\log (W/\lambda) \geq -4.55$ ) were discarded from the analysis. In the remaining transitions, the van der Waals damping process is a minor contributor to the equivalent widths. MOOG's Unsöld approximation option was used to take into account van der Waals damping in the computation of theoretical curves of growth.

3.3 $\mathsfsl{log~gf}$ adjustment

Many of the lines used in this analysis don't have accurate oscillator strengths reported in the literature. We have thus determined, from a solar spectrum, the oscillator strengths of most of the lines (see below) used in the present analysis. Their equivalent widths were measured in a high S/N ratio ( $S/N \simeq 250$ ) moon spectrum extracted from the archive of the first FEROS commissioning period (fall 1998). Their oscillator strengths were then adjusted using a Kurucz model atmosphere with $T_{\rm eff} = 5777$ K, $\log g = 4.44$ and micro-turbulence $\xi = 1.0$ km s^-1, so that a spectral analysis made with the same approach as for IS Vir and V851 Cen stars (i.e. deriving abundances with the MOOG LTE analysis software packages) reproduced the Kurucz solar abundances. In order to be consistent with Kurucz models and opacities, we kept the old Fe I solar abundance, $\log {\rm Fe} = 7.67$ , instead of the meteoritic value $\log {\rm Fe} = 7.51$ . As the analysis performed here is purely differential with respect to the Sun, the use of the "old'' Fe value has no consequence on the results derived. In order to avoid degrading the accuracy of the oscillator strengths with equivalent widths' measurement errors, only lines with $W_\odot > 10$ mÅ were calibrated on the Sun. For the weaker lines, we kept the Kurucz theoretical values (Kurucz & Bell 1995). The characteristics of the selected lines are listed in Table 2, with their wavelengths, excitation potentials and adopted oscillator strengths. When available, oscillator strengths from the literature, extracted from Edvardsson et al. (1993), Neuforge-Verheecke & Magain (1997) and Feltzing & Gonzalez (2001) are also listed.

3.4 Continuum fitting and line's measurement

The FEROS spectra were straightened by segments of 200 to 400 Å. In each of them, intervals of the continuum were selected by comparison with a Kurucz synthetic spectrum. Those series of intervals were adjusted by low degree polynomials, which in turn were used to straighten the different domains. In slow rotators, such as IS Vir and V851 Cen, the fitting of the continuum is relatively simple. This is not true for intermediate rotators which display very few intervals at continuum level. There is then a risk to underestimate the position of the continuum. An underestimation of 0.5% of the continuum, would lead to an error of 5% on the equivalent widths and 0.05 dex on the abundances (on "weak'' lines).

The equivalent widths were measured with the IRAF (Jacoby 1998) task splot, assuming Gaussian profiles. They are compiled in Table 2.

3.5 Atmospheric models

The analysis were performed using Kurucz' LTE plane parallel atmospheric models, computed without the overshooting option and for a value of the length of the convective cell over the pressure scale height $\alpha = l / H_{\rm p} = 0.5$ . This later value has been shown to give a better description of Balmer H $\alpha$ and H $\beta$ line profiles than the previous value of 1.25 (Fuhrmann et al. 1993; van't Veer-Menneret & Megessier 1996).

Table 3: Characteristics of the 9 test stars: galactic coordinates, distances (pc), colour excesses, B-V, $V-R_{\rm c}$ and $V-I_{\rm c}$ indices and the corresponding effective temperatures.
HD Num. l b d E(B-V) $E(V-R_{\rm c})$ $E(V-I_{\rm c})$ B-V $V-R_{\rm c}$ $V-I_{\rm c}$ $T_{\rm eff}$ $T_{\rm eff}$ $T_{\rm eff}$

(B-V) (V-R) (V-I)

HD 26354 262 -46 35 0.03 0.02 0.04 0.88 0.52 0.99 4963 4927 4824

HD 34802 289 -32 180 0.06 0.04 0.08 1.03 0.57 1.09 4746 4602 4546

HD 61245 258 -12 111 0.03 0.02 0.04 1.01 0.54 1.02 4785 4704 4687

HD 72688 255 3 131 0.02 0.02 0.03 0.93 0.47 0.91 4945 4973 4936

HD 81410 254 19 120 0.04 0.03 0.06 0.98 0.53 1.02 4844 4740 4687

HD 106225 287 53 125 0.01 0.00 0.01 0.99 0.58 1.13 4824 4569 4472

HD 119285 309 1 76 0.03 0.02 0.05 1.05 0.60 1.15 4708 4507 4436

HD 136905 356 40 95 0.07 0.04 0.09 0.96 0.53 1.01 4884 4740 4708

**Table 3:** Characteristics of the 9 test stars: galactic coordinates, distances (pc), colour excesses, B-V, $V-R_{\rm c}$ and $V-I_{\rm c}$ indices and the corresponding effective temperatures.
HD Num.	l	b	d	E(B-V)	$E(V-R_{\rm c})$	$E(V-I_{\rm c})$	B-V	$V-R_{\rm c}$	$V-I_{\rm c}$	$T_{\rm eff}$	$T_{\rm eff}$	$T_{\rm eff}$
										(B-V)	(V-R)	(V-I)
HD 26354	262	-46	35	0.03	0.02	0.04	0.88	0.52	0.99	4963	4927	4824
HD 34802	289	-32	180	0.06	0.04	0.08	1.03	0.57	1.09	4746	4602	4546
HD 61245	258	-12	111	0.03	0.02	0.04	1.01	0.54	1.02	4785	4704	4687
HD 72688	255	3	131	0.02	0.02	0.03	0.93	0.47	0.91	4945	4973	4936
HD 81410	254	19	120	0.04	0.03	0.06	0.98	0.53	1.02	4844	4740	4687
HD 106225	287	53	125	0.01	0.00	0.01	0.99	0.58	1.13	4824	4569	4472
HD 119285	309	1	76	0.03	0.02	0.05	1.05	0.60	1.15	4708	4507	4436
HD 136905	356	40	95	0.07	0.04	0.09	0.96	0.53	1.01	4884	4740	4708

3.6 Effective temperature from photometric indices

The use of photometric indices to derive effective temperature is a classical technique commonly used to analyse non-active stars. In the present work we are however considering stars with a very high level of magnetic activity, which also produces large photospheric spots. Their high activity level could thus affect their spectral energy distribution and therefore their colour indices. In the case of the metallicity, large discrepancies between photometric and spectroscopic estimates were reported by Gimenez et al. (1991) and Favata et al. (1997).

In order to check the relative consistency of different colour indices, we derived, for 8 SB1 objects from our program (hereafter referred to as test stars), with good quality Hipparcos parallaxes and B-V, $V-R_{\rm c}$ and $V-I_{\rm c}$ indices either from Cutispoto (1998) or Cutispoto et al. (2001), the effective temperatures corresponding to the three colours. There was no recent measurement of the V-R or V-I colour indices reported in the literature for IS Vir, which, therefore, has not been included in the test stars sample. The visual extinction (i.e. A(V)) was derived using the model of Arenou et al. (1992) and converted in colour excesses using the coefficients : E(B-V) = 0.302 A(V), $E(V-R_{\rm c}) = 0.194~ A(V)$ and $E(V-I_{\rm c}) = 0.415~ A(V)$ (Schlegel et al. 1998). Galactic coordinates, distances (pc) and colour excesses of the tests stars are summarised in Table 3.

The colours were converted into effective temperatures using the empirical calibration for F0V to K5V main sequence stars of Alonso et al. (1996) (HD 26354) and for F0 to K5 giant stars of Alonso et al. (1999) (for the 7 others), initially assuming solar metallicities. The 8 test stars were classified as either main sequence or evolved stars, by comparing their absolute magnitudes with the Bertelli et al. (1994) isochrones. To convert the $V-R_{\rm c}$ and $V-I_{\rm c}$ indices measured by Cutispoto (1998) and Cutispoto et al. (2001) into the $V-R_{\rm J}$ and $V-I_{\rm J}$ indices used by Alonso et al. (1996) and Alonso et al. (1999), we used the Bessell (1979) transformations, $V-R_{\rm J} = (V-R_{\rm c} + 0.03) / 0.73$ and $V-I_{\rm J} = V-I_{\rm c} / 0.778$ . The colour indices as well as the corresponding effective temperatures are summarised in Table 3.

In all cases, the $V-I_{\rm c}$ index lead to temperatures systematically cooler (from 10 to 350 K with a mean of 175 K) than the B-V index. In 6 cases out of 8, the $V-R_{\rm c}$ indices give temperatures significantly cooler than B-V (and slightly warmer than $V-I_{\rm c}$ ), with differences ranging from 80 to 255 K.

In order to check if the discrepancies between the effective temperatures derived from B-V, and those derived from V-I, were artefacts of the adopted transformations, we compared Alonso et al. (1999) calibrations with those for giant stars of McWilliam (1990). In the ranges of colour relevant to our test stars: $0.85 \leq B-V \leq 1.10$ and $0.90 \leq V-I_{\rm c} \leq 1.20$ , the differences between the 2 systems never exceed 40 and 20 K. We also computed the mean differences $T_{\rm eff}(B-V) - T_{\rm eff}(V-I)$ over 7 evolved test stars (i.e. all test stars but HD 26354) with the 2 calibrations and found respectively 181 K (Alonso et al. 1999) and 180 K (McWilliam 1990), in perfect agreement.

Alonso et al. (1996) and Alonso et al. (1999) calibrations have been derived for single stars and their validity is established only for single systems. A binary system, in which both components contribute significantly to the total flux, and at the same time differ in colours, will deviate from the colour/temperature relations. Of course, in most cases, a large difference in colour also means a large difference in brightness and therefore a small contribution of the secondary to the overall system colours.

The possible role of binarity in the discrepancies observed in the test stars, between the B-Vand V-I temperatures, was investigated via a synthetic grid of binary systems (computed using the isochrones of Bertelli et al. 1994, for solar metallicity). This grid contains all possible arrangements of primary and secondary components ranging from the top of the giant branch to K7 dwarf ( $\simeq$ 3900 K) by steps of 50 K. Several system ages have been considered: 0.5, 1, 2, 3, 4, 5, 8, 10 and 12 Gyr. Each arrangement is characterised by the effective temperature, evolutionary stage (dwarf or giant), visual absolute magnitude and B-V and V-I colour indices of each component and of the global system.

For each of the 8 test stars, we searched the grid for the synthetic binary system with similar B-V temperature and visual absolute magnitude (i.e. within $\pm$ $1 \sigma$ of the test star parameters) that showed the largest difference between the temperatures derived from B-V and V-I indices. To take into account that all the test stars are single line systems, only systems with secondary component at least 2.5 mag fainter than the primary were considered.

In the case of the dwarf star HD 26354, the largest discrepancy between the B-V and V-I diagnostics is around 150 K (assuming a 1 Gyr old system with a 5200 K subdwarf as primary component and a 3900 K dwarf as secondary component). For the 7 evolved test stars, the "strongest differences'' range from 18 to 67 K with a mean of 41 K. This mean value of 41 K is small compared to the mean difference of 181 K derived from the real colours of the same 7 test stars. Moreover, the above 18 to 67 K differences represent "extreme'' cases among a large number of combinations of primary and secondary components, whose B-V and V-I colour indices lead to very similar effective temperatures. Therefore, the binarity may play a minor role in the discrepancy between the B-V and V-I temperatures, but is most likely not the dominant effect.

Unlike the $V-I/T_{\rm eff}$ transformation, both the B-V and V-R vs. temperature relations of Alonso et al. (1999) are sensitive to metallicity. To test whether the above discrepancy can be solved by adjusting the iron abundances, we performed a full detailed analysis on 4 of the previous stars, namely HD 26354, HD 34802, HD 106225 and HD 119285, using the second method together with the B-V index. The derived metallicities and corresponding B-V effective temperatures are presented in Table 4. V-I temperatures are also listed for comparison. The metallicities of HD 34802 and HD 119285 are close to solar and therefore their temperatures are fairly similar to those listed in Table 3. HD 26354 exhibits an iron overabundance which increases the difference between the B-V and V-I temperatures by about 120 K. The moderate deficiency of HD 106225, reduces by 100 K the discrepancy between B-V and V-I, but the difference is still about 250 K. Therefore the metallicity is unlikely to explain the temperature discrepancy derived through the two indices, discrepancy which may be due to stellar activity.

Table 4: Metallicities, B-V and V-I temperatures of 4 of the test stars. The errors given for $T_{\rm eff}$ correspond to the propagation of the errors on the metallicity.
HD Num. ${\rm [Fe/H]}$ $T_{\rm eff}$ $T_{\rm eff}$

(B-V) (V-I)

HD 26354 $+0.36 \pm 0.22$ 5085⁺⁸⁵_-77 4824

HD 34802 $+0.17 \pm 0.21$ 4793⁺⁶⁵_-57 4546

HD 106225 $-0.40 \pm 0.16$ 4723⁺³⁸_-33 4472

HD 119285 $-0.14 \pm 0.12$ 4674⁺²⁹_-27 4436

**Table 4:** Metallicities, B-V and V-I temperatures of 4 of the test stars. The errors given for $T_{\rm eff}$ correspond to the propagation of the errors on the metallicity.
HD Num.	${\rm [Fe/H]}$	$T_{\rm eff}$	$T_{\rm eff}$
		(B-V)	(V-I)
HD 26354	$+0.36 \pm 0.22$	5085⁺⁸⁵_-77	4824
HD 34802	$+0.17 \pm 0.21$	4793⁺⁶⁵_-57	4546
HD 106225	$-0.40 \pm 0.16$	4723⁺³⁸_-33	4472
HD 119285	$-0.14 \pm 0.12$	4674⁺²⁹_-27	4436

3.7 Surface gravities by profile fitting

The wings of "strong'' lines are collisionally broadened and therefore gravity sensitive. In the third method, we take advantage of this behaviour to derive the surface gravity of our stars. The wings of one of those "strong'' lines are compared to a library of synthetic lines spanning a large range in gravities. In a first step the grid is searched for the most similar synthetic profile. In a second step the difference of gravity between the best synthetic match and the observed profile is estimated. Variations of this method have been successfully used by several authors (e.g. Drake & Smith 1991). Several "strong'' lines are present in the FEROS spectral interval: the magnesium green triplet ( $\lambda = {5167.3, 5172.7, 5183.6}$ Å), the neutral calcium red triplet ( $\lambda = {6102.7, 6122.2, 6162.2}$ Å) and the ionised calcium infrared triplet ( $\lambda = {8498.0, 8542.1, 8662.1}$ Å). We used the 6162 Å neutral calcium line whose profile is more sensitive to surface gravity than the two other transitions of the triplet. This line looses relatively quickly its sensitivity to gravity below $\log g = 2$ . For more evolved stars it is more accurate to rely on the Ca II triplet lines, whose strengths increase with decreasing gravity. A grid of about 1600 synthetic spectra was computed with the Kurucz' ATLAS9 (Kurucz 1993a) and SYNTHE (Kurucz 1993b) software packages. It covers the parameter space $4250 \leq T_{\rm eff} \leq 6000$ K (by steps of 250 K), $0.5 \leq \log g \leq 5.0$ (by steps of 0.5), $\xi = 0, 1, 2, 4$ km s^-1, ${\rm [Fe/H]} = -1, -0.5, -0.3, 0.0, 0.3$ and $[\alpha/\rm Fe] = 0.0$ and +0.4. Each spectrum ranges from 6000 to 6250 Å with an initial resolution of 300 000.

The 6162 Å calcium line is not only sensitive to gravity, but also to effective temperature, metallicity, micro-turbulence and, of course, calcium abundance. As a consequence, the first step in the estimation of gravity is to fix those four parameters. The synthetic grid is then interpolated to compute a new sub-grid at the chosen values of $T_{\rm eff}$ , $\xi$ , ${\rm [Fe/H]}$ and ${\rm [Ca/H]}$ , keeping $\log g$ as the only degree of freedom. The 10 spectra of this sub-grid are then compared with the observed spectrum. For each comparison, the flux of the synthetic and observed spectra are adjusted by least squares, finding the coefficients a₀ and a₁ that minimise the expression:

Once the most similar synthetic spectrum (hereafter referred as pivot) is identified, the difference of gravity between this "twin'' synthetic profile and the observed one is derived, using a variant of the atmospheric parameters optimal extraction method (Cayrel et al. 1991). The 6162 Å Ca I profile is projected on the grid, assuming that between two neighbouring synthetic spectra, the gravity evolved linearly with the logarithm of the flux: $\log F_{\rm obs} = (1-u) \log F_{\rm piv} + u \log F_{\rm nei}$ , where $F_{\rm obs}$ , $F_{\rm piv}$ and $F_{\rm nei}$ are respectively the observed and pivot fluxes and the flux of the one out of the pivot's two synthetic neighbours that is most similar to the observed profile. The projection is performed by least squares, searching for the coefficients u, a₀ and a₁ that minimise the expression:

Table 5: Parameters of the reference stars used to estimate the accuracy of calcium line fitting technique : signal to noise ratios (S/N), effective temperatures ( $T_{\rm eff}$ ), surface gravities ( $(\log g)_{\rm bib}$ ), metallicities ( ${\rm [Fe/H]}$ ) and micro-turbulences ( $\xi$ ). ( $\log g$ ) gives the gravities derived by profile fitting.
Id. S/N $T_{\rm eff}$ $(\log g)_{\rm bib}$ ${\rm [Fe/H]}$ $\xi$ $\log g$

Sun 380 5777 4.44 0.00 $\$ 1.0 4.42

Sun 270 " " " $\$ " 4.39

Sun 140 " " " $\$ " 4.43

Sun 160 " " " $\$ " 4.43

Sun 120 " " " $\$ " 4.41

Sun 200 " " " $\$ " 4.48

Sun 220 " " " $\$ " 4.46

HD 1835 140 5771 4.44 0.15 $\$ 1.0 4.56

HD 10307 200 5882 4.33 0.02 $\$ 1.0 4.24

HD 24040 150 5594 4.50 0.07 $\$ 1.0 3.95

HD 28099 110 5761 4.50 0.17 $\$ 1.0 4.48

HD 85503 180 4540 2.20 0.29 $\$ 1.2 2.14

HD 86728 130 5742 4.21 0.12 $\$ 1.0 4.37

HD 95128 180 5855 4.25 0.00 $\$ 1.0 4.14

HD 113226 140 4990 2.70 0.11 $\$ 1.6 2.33

HD 114710 90 5979 4.40 0.08 $\$ 1.0 4.22

HD 115383 110 5920 3.96 0.10 $\$ 1.3 4.16

HD 146233 240 5803 4.34 0.03 $\$ 1.0 4.48

HD 184406 80 4450 2.47 $-0.13\ \$ 1.8 2.64

HD 186408 160 5820 4.26 0.07 $\$ 1.0 4.31

HD 186427 140 5762 4.38 0.06 $\$ 1.0 4.39

HD 187691 160 6101 4.22 0.09 $\$ 1.0 4.21

HD 217014 160 5757 4.23 0.15 $\$ 1.0 4.24

HD 219134 150 4727 4.50 0.05 $\$ 1.0 4.55

**Table 5:** Parameters of the reference stars used to estimate the accuracy of calcium line fitting technique : signal to noise ratios (S/N), effective temperatures ( $T_{\rm eff}$ ), surface gravities ( $(\log g)_{\rm bib}$ ), metallicities ( ${\rm [Fe/H]}$ ) and micro-turbulences ( $\xi$ ). ( $\log g$ ) gives the gravities derived by profile fitting.
Id.	S/N	$T_{\rm eff}$	$(\log g)_{\rm bib}$	${\rm [Fe/H]}$	$\xi$	$\log g$
Sun	380	5777	4.44	0.00 $\$	1.0	4.42
Sun	270	"	"	" $\$	"	4.39
Sun	140	"	"	" $\$	"	4.43
Sun	160	"	"	" $\$	"	4.43
Sun	120	"	"	" $\$	"	4.41
Sun	200	"	"	" $\$	"	4.48
Sun	220	"	"	" $\$	"	4.46
HD 1835	140	5771	4.44	0.15 $\$	1.0	4.56
HD 10307	200	5882	4.33	0.02 $\$	1.0	4.24
HD 24040	150	5594	4.50	0.07 $\$	1.0	3.95
HD 28099	110	5761	4.50	0.17 $\$	1.0	4.48
HD 85503	180	4540	2.20	0.29 $\$	1.2	2.14
HD 86728	130	5742	4.21	0.12 $\$	1.0	4.37
HD 95128	180	5855	4.25	0.00 $\$	1.0	4.14
HD 113226	140	4990	2.70	0.11 $\$	1.6	2.33
HD 114710	90	5979	4.40	0.08 $\$	1.0	4.22
HD 115383	110	5920	3.96	0.10 $\$	1.3	4.16
HD 146233	240	5803	4.34	0.03 $\$	1.0	4.48
HD 184406	80	4450	2.47	$-0.13\ \$	1.8	2.64
HD 186408	160	5820	4.26	0.07 $\$	1.0	4.31
HD 186427	140	5762	4.38	0.06 $\$	1.0	4.39
HD 187691	160	6101	4.22	0.09 $\$	1.0	4.21
HD 217014	160	5757	4.23	0.15 $\$	1.0	4.24
HD 219134	150	4727	4.50	0.05 $\$	1.0	4.55

The reliability of the method was tested on a subsample of the ELODIE library of standard stars (Soubiran et al. 1998; Prugniel & Soubiran 2001). This library, in addition to the spectra, provides values of $T_{\rm eff}$ , $\log g$ and ${\rm [Fe/H]}$ for each star. The ELODIE spectra are relatively similar in characteristics to FEROS spectra, with a resolution of 42 000 and a wavelength domain ranging from 3900 to 6800 Å. We applied the gravity determination method to 24 spectra from the library: 7 spectra of the Sun (Moon), 17 main sequence star spectra and 4 evolved star spectra. We assumed a calcium abundance proportional to the solar one ( ${\rm [Ca/Fe]} = 0.0$ ) for all stars, a micro-turbulence of 1 km s^-1 for the main sequence stars and searched the literature for the micro turbulence of the evolved objects. Table 5 list the stars used, their adopted atmospheric parameters and the surface gravities estimated with our method. The average difference between the estimated and the bibliographic gravities is -0.02 with a dispersion of 0.16. A test performed on IS Vir has shown that systematically shifting the calcium line fitting gravity diagnostic by $\Delta \log g = -0.16$ modifies the third method result by $\Delta T_{\rm eff} = -140$ K, $\Delta \log g = -0.38$ , $\Delta {\rm [Fe/H]} = -0.13$ dex and $\Delta \xi = +0.1$ km s^-1. This error is accounted for in the error bars given for IS Vir and V851 Cen in Tables 6 and 7.

Table 6: IS Vir: number of transitions used to derive the abundances of the different elements (n), mean values (<>) and error bars ( $\sigma$ ) of the atmospheric parameters and abundances, with the three methods.
meth. 1 meth. 2 meth. 3

n <> $\sigma$ n <> $\sigma$ n <> $\sigma$

$T_{\rm eff}$ 4720 100 4690 150 4780 185

$\log g$ 2.65 0.25 2.55 0.41 2.75 0.45

$\xi$ 1.65 0.08 1.65 0.11 1.60 0.13

${\rm [Fe/H]}$ 44 +0.04 0.08 36 +0.02 0.13 36 +0.09 0.16

${\rm [Na/Fe]}$ 1 +0.23 0.04 1 +0.23 0.04 1 +0.23 0.04

${\rm [Mg/Fe]}$ 1 +0.17 0.06 1 +0.18 0.06 1 +0.18 0.06

${\rm [Al/Fe]}$ 3 +0.18 0.06 3 +0.19 0.07 3 +0.16 0.08

${\rm [Si/Fe]}$ 7 +0.06 0.05 7 +0.07 0.08 7 +0.02 0.11

${\rm [Ca/Fe]}$ 4 +0.11 0.04 4 +0.10 0.04 4 +0.13 0.06

${\rm [Sc/Fe]}$ 1 +0.05 0.05 1 +0.03 0.09 1 +0.08 0.12

${\rm [Ti/Fe]}$ 2 +0.05 0.04 2 +0.04 0.04 2 +0.06 0.05

${\rm [Co/Fe]}$ 2 -0.04 0.07 2 -0.04 0.07 2 -0.04 0.08

${\rm [Ni/Fe]}$ 9 -0.10 0.03 9 -0.10 0.04 9 -0.10 0.04

**Table 6:** IS Vir: number of transitions used to derive the abundances of the different elements (n), mean values (<>) and error bars ( $\sigma$ ) of the atmospheric parameters and abundances, with the three methods.
	meth. 1		meth. 2	meth. 3
	n	<>	$\sigma$	n	<>	$\sigma$	n	<>	$\sigma$
$T_{\rm eff}$		4720	100		4690	150		4780	185
$\log g$		2.65	0.25		2.55	0.41		2.75	0.45
$\xi$		1.65	0.08		1.65	0.11		1.60	0.13
${\rm [Fe/H]}$	44	+0.04	0.08	36	+0.02	0.13	36	+0.09	0.16
${\rm [Na/Fe]}$	1	+0.23	0.04	1	+0.23	0.04	1	+0.23	0.04
${\rm [Mg/Fe]}$	1	+0.17	0.06	1	+0.18	0.06	1	+0.18	0.06
${\rm [Al/Fe]}$	3	+0.18	0.06	3	+0.19	0.07	3	+0.16	0.08
${\rm [Si/Fe]}$	7	+0.06	0.05	7	+0.07	0.08	7	+0.02	0.11
${\rm [Ca/Fe]}$	4	+0.11	0.04	4	+0.10	0.04	4	+0.13	0.06
${\rm [Sc/Fe]}$	1	+0.05	0.05	1	+0.03	0.09	1	+0.08	0.12
${\rm [Ti/Fe]}$	2	+0.05	0.04	2	+0.04	0.04	2	+0.06	0.05
${\rm [Co/Fe]}$	2	-0.04	0.07	2	-0.04	0.07	2	-0.04	0.08
${\rm [Ni/Fe]}$	9	-0.10	0.03	9	-0.10	0.04	9	-0.10	0.04

Table 7: V851 cen: number of transitions used to derive the abundances of the different elements (n), mean values (<>) and error bars ( $\sigma$ ) of the atmospheric parameters and abundances, with the three methods.
meth. 1 meth. 2 (B-V/V-I) meth. 3

n <> $\sigma$ n <> $\sigma$ n <> $\sigma$ n <> $\sigma$

$T_{\rm eff}$ 4700 80 4670 150 4440 100 4780 160

$\log g$ 3.00 0.27 2.90 0.42 2.25 0.32 3.23 0.41

$\xi$ 1.50 0.07 1.50 0.12 1.60 0.10 1.40 0.13

${\rm [Fe/H]}$ 46 -0.13 0.07 36 -0.14 0.12 36 -0.31 0.10 36 -0.04 0.15

${\rm [Na/Fe]}$ 1 +0.30 0.03 1 +0.29 0.04 1 +0.28 0.04 1 +0.27 0.04

${\rm [Mg/Fe]}$ 1 +0.41 0.07 1 +0.39 0.07 1 +0.44 0.07 1 +0.35 0.04

${\rm [Al/Fe]}$ 3 +0.46 0.07 3 +0.46 0.08 3 +0.51 0.10 3 +0.41 0.08

${\rm [Si/Fe]}$ 6 +0.15 0.04 6 +0.14 0.07 6 +0.26 0.06 6 +0.08 0.09

${\rm [Ca/Fe]}$ 4 +0.22 0.04 4 +0.21 0.06 4 +0.14 0.05 4 +0.22 0.06

${\rm [Sc/Fe]}$ 2 +0.24 0.10 2 +0.22 0.13 2 +0.06 0.10 2 +0.30 0.15

${\rm [Ti/Fe]}$ 2 +0.19 0.04 2 +0.17 0.05 2 +0.12 0.05 2 +0.18 0.05

${\rm [Co/Fe]}$ 2 +0.07 0.04 2 +0.06 0.04 2 +0.04 0.04 2 +0.06 0.05

${\rm [Ni/Fe]}$ 9 -0.06 0.03 9 -0.07 0.04 9 -0.09 0.03 9 -0.06 0.04

3 Data analysis

3.1 Methods

3.2 Lines selection

3.3 $\mathsfsl{log~gf}$ adjustment

3.4 Continuum fitting and line's measurement

3.5 Atmospheric models

3.6 Effective temperature from photometric indices

3.7 Surface gravities by profile fitting

	meth. 1			meth. 2 (B-V/V-I)						meth. 3
	n	<>	$\sigma$	n	<>	$\sigma$	n	<>	$\sigma$	n	<>	$\sigma$
$T_{\rm eff}$		4700	80		4670	150		4440	100		4780	160
$\log g$		3.00	0.27		2.90	0.42		2.25	0.32		3.23	0.41
$\xi$		1.50	0.07		1.50	0.12		1.60	0.10		1.40	0.13
${\rm [Fe/H]}$	46	-0.13	0.07	36	-0.14	0.12	36	-0.31	0.10	36	-0.04	0.15
${\rm [Na/Fe]}$	1	+0.30	0.03	1	+0.29	0.04	1	+0.28	0.04	1	+0.27	0.04
${\rm [Mg/Fe]}$	1	+0.41	0.07	1	+0.39	0.07	1	+0.44	0.07	1	+0.35	0.04
${\rm [Al/Fe]}$	3	+0.46	0.07	3	+0.46	0.08	3	+0.51	0.10	3	+0.41	0.08
${\rm [Si/Fe]}$	6	+0.15	0.04	6	+0.14	0.07	6	+0.26	0.06	6	+0.08	0.09
${\rm [Ca/Fe]}$	4	+0.22	0.04	4	+0.21	0.06	4	+0.14	0.05	4	+0.22	0.06
${\rm [Sc/Fe]}$	2	+0.24	0.10	2	+0.22	0.13	2	+0.06	0.10	2	+0.30	0.15
${\rm [Ti/Fe]}$	2	+0.19	0.04	2	+0.17	0.05	2	+0.12	0.05	2	+0.18	0.05
${\rm [Co/Fe]}$	2	+0.07	0.04	2	+0.06	0.04	2	+0.04	0.04	2	+0.06	0.05
${\rm [Ni/Fe]}$	9	-0.06	0.03	9	-0.07	0.04	9	-0.09	0.03	9	-0.06	0.04

3 Data analysis

3.1 Methods

3.2 Lines selection

3.3 adjustment

3.4 Continuum fitting and line's measurement

3.5 Atmospheric models

3.6 Effective temperature from photometric indices

3.7 Surface gravities by profile fitting

3.3 $\mathsfsl{log~gf}$ adjustment