6 Discussion

The planet around $\iota$ Hor remains the single clear detection of an extrasolar planet by the CES Long Camera survey. This corresponds to a detection rate of $\approx$3%, a value similar to other precise Doppler searches. The discovery of $\iota$ Hor b demonstrated for the first time the feasibility of the RV technique in planet detection in the case of young and thus moderately active stars.

Seven stars (19%) of the CES sample show minor signs of variability (minor in the sense that they pass one but fail at other tests): $\zeta$ Tuc, HR 506, $\zeta ^{2}$ Ret, $\epsilon$ Eri, $\phi ^{2}$ Pav and HR 8883. While for $\epsilon$ Eri this is an indication for the presence of the highly eccentric RV signature of the planet, the cause of variability for the other stars remains unknown since no convincing periodicity or trends were found. For HR 8883 the RV variations can be explained by the high instrinsic activity of this star (high X-ray luminosity).

Low amplitude linear RV trends were found for the following 5 targets (14%): $\beta$ Hyi, $\alpha$ For, HR 6416, $\epsilon$ Ind and HR 8501. For the known binaries $\alpha$ For, HR 6416 and HR 8501 these trends agree well with the expected acceleration by the stellar secondary. The large scatter around the linear trend of $\alpha$ For is probably also due to high stellar activity (again a high X-ray luminosity).

$\beta$ Hyi and $\epsilon$ Ind are identified as candidates for having long-period and probably stellar companions. But could these trends also be caused by planets? If we assume that the minimum orbital period would be 4 times the monitoring time span (i.e. $P\approx20$ years) and that the RV semi-amplitude is of the order of the RV shift over the 5 years ($\beta$ Hyi: $38~{\rm m~s}^{-1}$, $\epsilon$ Ind: $21~{\rm m~s}^{-1}$) and e=0, the observed trends could be caused in the case of $\beta$ Hyi by a planet with  $m\sin i \approx 4~{M}_{\rm Jup}$ at  $a\approx 7.6$ AU and for $\epsilon$ Ind by an  $m\sin i \approx 1.6~{M}_{\rm Jup}$ companion at $a\approx 6.5$ AU. Such planetary systems with a distant giant planet would resemble our Solar System more closely than the extrasolar planetary systems found so far. For e>0 orbits the period can even be much shorter than 20 years and we therefore conclude that although the linearity of the RV trends points towards distant and previously unknown stellar companions, both stars constitute prime targets for follow-up observations by the CES planet search program.

$\phi ^{2}$ Pav has been earlier announced by our team as a possible candidate for having a planetary companion with an orbital period of about 43 days and $m\sin i = 0.7~{M}_{\rm Jup}$(Kürster et al. 1999a). This signal was found with a low confidence level and based on a preliminary analysis of a subset of the Long Camera data, using an early version of the Radial code (Cochran & Hatzes 1990) to obtain the RV measurements. The analysis of the complete data set of $\phi ^{2}$ Pav using the Austral software did not confirm the presence of this companion. The total rms scatter over the entire 5 1/2 years is $35.4~{\rm m~s}^{-1}$, slightly larger than the mean internal error of $31.3~{\rm m~s}^{-1}$ for this star. No apparent Keplerian signal is present. This is consistent with results coming from the Anglo-Australian planet search (Butler et al. 2001), who collected 7 measurements of $\phi ^{2}$ Pav over the course of 1 year, which reveal a total rms scatter of only $5~{\rm m~s}^{-1}$ (the Anglo-Australian planet search uses the UCLES echelle spectrometer which covers a much larger spectral region and the entire $1000~{\rm\AA}$ of the I₂-reference spectrum, hence the higher precision of their results). However, we have identified in our much longer and higher sampled data a periodic signal of $\approx$7 days, again with low confidence (the FAP of this signal is still higher than 0.001 but it appears in both the unbinned original RV data as well as in the nightly averaged results). If the periodic signal is indeed real what could produce such an RV signature? $\phi ^{2}$ Pav belongs to the $\zeta$ Ret stellar kinematic group, a group of metal deficient stars with an age of $\approx$5 Gyr (del Peloso et al. 2000). The iron abundance was determined as [Fe/H] = -0.37 by Porto de Mello & da Silva (1991) and as [Fe/H] = -0.44 by Edvardsson et al. (1993). This low metallicity can account for the observed large RV scatter, since fewer and shallower absorption lines in the small CES bandpass degrade our measurement precision. In fact $\phi ^{2}$ Pav has the second largest internal RV error of the F-type stars in the CES sample. Based on H$\alpha$ emission, $\phi ^{2}$ Pav appears to be slightly more active than the Sun (del Peloso et al. 2000) and the star is already evolving into the subgiant phase (Porto de Mello & da Silva 1991). With this higher level of activity we suspect that the P=7 day RV variation is in fact the stellar rotation period and that our RV measurements are affected by cool spots in the photosphere of $\phi ^{2}$ Pav. These spots would modulate the RV measurements with a typical timescale of  $P_{\rm Rot}$. Since these spots appear and disappear on short timescales compared to the monitoring duration and the overall activity level might change over 5 1/2 years, the amplitude as well as the phase of this modulation varies with time. Such a signal is therefore difficult to detect significantly, which is exactly what we observe here. The expected size of the subsurface convection zone for a low-metallicity F-type star is smaller than for a star of solar metallicity. Even with $P_{\rm Rot}=7$ days such a star would not display a much larger activity level than $\phi ^{2}$ Pav due to the inefficiency of the dynamo. From the $v\sin i = 6.7~{\rm km~s}^{-1}$ and $R_{*}=1.86~R_{\odot}$ (Porto de Mello priv. comm.) and the $P_{\rm Rot}$ value of 7 days we derive a viewing angle of $\approx$ $30^{\circ}$. The continued monitoring of $\phi ^{2}$ Pav will demonstrate whether the P=7 day is robust and can be recovered with a higher confidence level.

Roughly 50% of the targets (18 stars) of the CES Long Camera survey show absolutely no sign of variability or trends in their RV data. Within the given RV precision of the CES Long Camera survey the following stars were found to be RV-constant: HR 209, HR 448, HR 753, $\zeta ^{1}$Ret, $\delta$ Eri, HR 2667, HR 4523, HR 4979, HR 6998, HR 7373, HR 7703, HR 8323 and GJ 433.

In the cases of the binaries $\kappa$ For, HR 2400, HR 3677 and $\alpha$ Cen A & B (see Endl et al. 2001a) no sign of significant periodic signals were found in the RV residuals after subtraction of the binary orbit. Interestingly, $\kappa$ For does not show any excess scatter although based on its $L_{\rm X}$-flux and H$\alpha$-emission (Porto de Mello priv. comm.) it is an active star. Still, the residuals after subtraction of the binary orbit are consistent with our measurement errors.

The CES Long Camera survey is in all cases sensitive to short-period ("51 Peg''-type) planets with orbital separations of a < 0.15 AU. This result confirms the general bias of precise Doppler searches towards short-period companions. For 22 stars of the CES Long Camera survey these mass-limits reach down into the sub-Saturn mass regime at $a \approx 0.04$ AU.

For most stars the region where planets with $m\sin i < 1~{M}_{\rm Jup}$ could have been detected is confined to orbital separations of less than 1 AU. Beyond 2 AUs no planets with $m \sin i \approx 1~{M}_{\rm Jup}$ were found to be detectable around any star of the survey. Subsequently, in order to detect a Solar System analogue the time baseline and (if possible) the RV precision of the CES planet search has to be increased. Within the limitations of our numerical simulations (e=0, P^2/3 sampling) we can rule out the presence of giant planets within 3 AU of the CES survey stars according to the limits presented here (with the exceptions of HR 209, HR 8883 and periods inside the non-detectability windows). Spectral leakage is the main cause for the windows of non-detectability. Even for well observed stars like e.g. $\tau$ Cet or $\beta$ Hyi these windows exist close to the seasonal one year period. This demonstrates how difficult the detection of RV signals with a one year periodicity is.

The average detection threshold $\bar{F}_{\rm CES}$ for the examined 30 Long Camera survey stars is 2.75, meaning that on average detectable planetary signals have K amplitudes which exceed the noise level by a factor of 2.75.

6.1 Outlook

With the decommissioning of the Long Camera in April 1998, phase I of the CES planet search program came to an end. All results based on this homogeneous set of observations are included in this work or were already presented earlier.

Although the CES was modified quite substantially after that, with the installation of the Very Long Camera (VLC) yielding a higher resolving power of $R\approx220~000$ and an optical fibre-link to the 3.6 m telescope being the most significant changes, the CES planet search program was continued using the same I₂-cell for self-calibration. This ensures the capability to merge the RV results from phase I with the newer phase II data set without the need to compensate for velocity zero-point drifts as demonstrated for HR 5568 in Fig. 24. The displayed RV results now cover almost 3 years for this star (compared to 1 year of the Long Camera survey). For the intermediate time when the 1.4 m CAT and the 3.6 m telescope were used in combination with the VLC and the 2K CCD (which meant a reduced bandwidth of $\approx$ $18~{\rm\AA}$ due to the higher spectral dispersion) we observe a small RV offset of $\approx$ $25~{\rm m~s}^{-1}$. This offset can be explained by the difference in spectral regions which were analysed to obtain the RVs. However, after the VLC was equipped with a longer 4K CCD the spectral bandwidth was increased to $36.5~{\rm\AA}$. To assure that the RV results are based on the same spectral regions we analyse both the Long Camera and the VLC data using a stellar template spectrum obtained with the most current instrumental setup (i.e. VLC & 4K CCD). A comparison of the RV results derived with the current CES and Long Camera results does no longer show any velocity offset (see Fig. 24). This demonstrates that the I₂-cell technique successfully compensates even for major instrumental setup changes. This guarantees a high long-term RV precision and allows a smooth continuation of the CES planet search program.

Figure 24: RV monitoring of HR 5668 during the refurbishment of the CES. A comparison of the Long Camera results (full diamonds) with data collected with the new VLC and the 2K CCD (boxes and circles) show a slight offset. This offset disappears with the installation of the 4K CCD (triangles) which increased and equalized the spectral bandwidth (see text for details). The total rms scatter over the 3 years is  $15.5~{\rm m~s}^{-1}$, and $7.5~{\rm m~s}^{-1}$ without the intermediate data (between the vertical dotted lines).

The Very Long Camera at the CES is promising to increase the RV precision of the CES planet search due to several reasons: the resolving power is doubled with respect to the Long Camera while the spectral bandwidth is not reduced by a large amount ( $36.5~{\rm\AA}$ instead of 48.5), and the S/N-ratio of spectra is higher due to the usage of image-slicers and the larger aperture of the 3.6 m telescope. With the successful merging of the new Very Long Camera data with the Long Camera survey and an expected better RV precision the CES planet search might become sensitive to Solar System analogues in the near future.