Our extensive discussion of the circumstellar properties of S stars confirms the dichotomy extrinsic/intrinsic S stars and leads to the following conclusions.
Extrinsic S stars have the lowest mass loss rates among S stars (
y-1), and undetectable circumstellar shells. This is consistent
with the hypothesis that these stars are much less evolved than intrinsic S stars,
populating the RGB or Early-AGB rather than the TP-AGB like intrinsic S stars.
The binary character of these stars does not seem to increase their mass loss
rate, as it has sometimes been suggested, e.g., in discussions relating to
symbiotic or barium stars. Note, however, that the few binary intrinsic S stars
are among the stars with the largest mass loss rates in our sample.
Among intrinsic S stars, various subclasses must be distinguished:
1. A few Tc-rich S stars (e.g., HR Peg) have very low mass-loss rates, similar
to those of the extrinsic S stars, with undetectable circumstellar shells [Region
A of the (K - [12], [25]-[60]) color-color diagram]. These are S stars with weak
chemical peculiarities, barely distinguishable from normal M giants.
2. S stars with weak chemical peculiarities and tenuous O-rich circumstellar
shells (small K - [12] excess), fed by a small
albeit measurable mass loss rate of a few 10-7 y-1 (Region B).
These are short-period (
to 150 d) SR variables or short-period (P <
500 d) Mira variables. This region may be contaminated by (possibly normal M)
supergiants of variability type SRc (T Cet, RS Cnc, Y Lyn). These SRc supergiants
have resolved shells at 60 
m, and for RS Cnc, a double wind.
3. S stars with strong chemical peculiarities and dense O-rich circumstellar
shells (as indicated by SiO maser emission, the 9.7 m silicate feature and large
K - [12] indices), populating the lower part of Region C. Mass loss rates range
from several 10-7 to 10-5 y-1. Almost all of them
are long-period (P > 300 d) Mira variables.
4. S stars with strong chemical peculiarities (often classified as SC in the
optical), with neither SiO maser emission nor 9.7 m silicate
emission, with featureless IRAS spectra and moderate to large [25] - [60] indices.
These stars populate Region
D and the upper part of Region C (like optical carbon stars) and have moderate
mass loss rates (a few 10-8 to a few 10-7 y-1) with small
wind velocities (generally 
km s
). They are mainly semi-regular (with
periods ranging from 60 to 360 d) or irregular variables, with a few short-period
(P < 370 d) Mira variables. These properties are reminiscent of carbon
stars. All their features may equally well be explained by a
carbon-rich circumstellar shell, or by a detached shell.
The relationship between stars in Regions C and D is far from being clear.
It may be a difference of galactic populations, as suggested by their different
properties as variable stars, or stars might oscillate between Regions C and D at
different phases of their variability cycle, as observed for some variable IRAS
sources in our sample, or they may be on different parts of the loop
associated with the detachment of a dust shell.
5. A few S stars have well-resolved (and thus very extended) IR shells (also
visible in the optical in the case of RZ Sgr), probably detached from their parent
star, as indicated by strong 60 and 100 m excesses (Region E). The mass loss
rates span a wide range, from a few 10-7 to several 10-6 y-1,
as do the wind velocities (25 km s for TT Cen to
2.8 km s for FU Mon). These stars with extreme properties are often SC or CS
stars, and may be experiencing a loop in the IRAS color-color diagram as first
proposed by Willems & de Jong (1988), triggered by their C/O ratio being very
close to unity. In TT Cen, the dominant spectral features are ZrO bands at some
times, and C2 bands at other times. This is a rare and short-lived phase, and
thus does not face the difficulties generally opposed to the Willems & de Jong
scenario. Some SC or CS stars in Region C, like BH Cru (a sister case of TT Cen)
and UY Cen, may be at the end of the counter-clockwise loop in the
IRAS
color-color diagram, entering the region of heavily-obscured IR carbon stars.
Region E is probably fed by stars leaving Region C when their C/O ratio reaches
values close to unity. However, there is no need that all S stars experience such
a loop.
Acknowledgements
We thank the staff at CSO, especially Ken Young (Taco), Antony Schinkel, Maren Purves and Tom Phillips, for the observing time and for their help with the observations. Astronomical research at the CSO is supported by the National Science Foundation via grants AST96-15025. We thank the staff at IPAC for swiftly processing the numerous requests for IRAS archive data sent over the Internet. We especially thank Ron Beck for providing us with the IRAS template data. We thank Princeton University, the Fonds National de la Recherche Scientifique (Belgium) and the National Science Foundation (U.S.A.), via grant AST96-18503, for partial support of this work. The CO line formation modeling was based on code by Mark Morris, and the figures were drawn using software by Robert Lupton and Patricia Monger. This research has made use of the Simbad data base, operated at CDS, Strasbourg, France. A.J. is Research Associate, F.N.R.S. (Belgium).