Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://hea-www.harvard.edu/~bradw/cv/papers/proxcen.ps.gz Äàòà èçìåíåíèÿ: Mon Sep 30 22:54:55 2002 Äàòà èíäåêñèðîâàíèÿ: Sat Dec 22 03:16:39 2007 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï ï ï

STRINGENT X­RAY CONSTRAINTS ON MASS LOSS FROM PROXIMA CENTAURI
Bradford J. Wargelin and Jeremy J. Drake
Smithsonian Astrophysical Observatory, 60 Garden Street, Cambridge, MA 02138; bwargelin@cfa.harvard.edu, jdrake@cfa.harvard.edu
Received 2002 April 5; accepted 2002 June 12
ABSTRACT
We have analyzed data from two Chandra imaging observations of Proxima Centauri, searching for an X­
ray halo arising from charge exchange between highly charged ions in its stellar wind and neutral gas in the
surrounding interstellar medium. Based upon our model of Proxima Cen's charge exchange emission, the
absence of any detectable charge exchange signal places a statistical 3 # upper limit of #3 # 10 #13 M # yr #1
(14 _
MM# ) on the mass­loss rate (9 _
MM# for the 2 # limit and 4 _
MM# for 1 #), with a model uncertainty of roughly a
factor of 3. This is orders of magnitude smaller than the upper limits that have been placed on late­type dwarf
stars using radio observations, and it supports a recent mass­loss result for Proxima Cen based on Ly#
absorption profiles. We have also studied the coronal spectrum, both in quiescence and during a prominent
flare. Results are consistent with those obtained in previous X­ray observations, but a firm determination of
coronal metal abundances remains elusive.
Subject headings: stars: coronae --- stars: individual (Proxima Centauri) --- stars: late­type ---
stars: mass loss --- X­rays: stars
On­line material: color figures
1. INTRODUCTION
Proxima Centauri, a flaring M5.5 dwarf, is our nearest
stellar neighbor and lies at a distance of 1:3009 # 0:0005 pc
(Benedict et al. 1999), roughly in the direction of the Galac­
tic center (l ¼ 313=94, b ¼ #1=93). Proxima Cen's associa­
tion with the binary system # Cen is controversial
(Matthews & Gilmore 1993; Anosova, Orlov, & Pavlova
1994), as the 2=2 separation between them corresponds to a
large but poorly determined distance of at least a few thou­
sand AU. However, the usual assumption is that Proxima
Cen is a bound companion and shares the solar or greater
photospheric metal abundances of # Cen AB. A rotation
period of 83:5 # 0:5 days, or perhaps half that, has been
reported by Benedict et al. (1998), while Jay et al. (1996)
have more tentatively suggested a 30 day period. Whichever
period is correct, Proxima Cen is a slow rotator, consistent
with its modest X­ray activity level, which is roughly one­
quarter the saturation limit of LX =L bol # 10 #3 for active
late­type stars (Agrawal, Rao, & Sreekantan 1986; Fleming
et al. 1993).
In addition to having a surprisingly mysterious nature,
Proxima Cen is intriguing as a representative of that most
common class of fully fledged hydrogen­burning stars, the
M dwarfs, and it is one of the few late M dwarfs close
enough to allow characterization of its coronal emission, of
which coronal metallicity is a particularly interesting
parameter. The reviews of, e.g., Feldman & Laming (2000),
Drake (2002), and White (1996) discuss evidence from
EUVE, ASCA, and BeppoSAX spectra of active stars that
indicates their coronae to be significantly metal­poor with
respect to a solar composition, while Drake (2002) also
presents evidence that less active stars can share the solar­
like coronal `` FIP e#ect,'' in which the coronal abundances
of elements with low first ionization potentials (FIP d 10
eV; e.g., Mg, Si, and Fe) are enhanced (relative to photo­
spheric abundances) compared to the abundances of ele­
ments with high FIP (e10 eV; e.g., O, Ne, and Ar). More
recently, abundances derived from high­resolution Chandra
and XMM­Newton spectra have indicated that in very active
stars, the abundances of Ne, Ar, and possibly other high­
FIP elements can be strongly enhanced relative to that of Fe
(Drake et al. 2001; Brinkman et al. 2001; Phillips et al. 2001;
Maggio et al. 2002).
Proxima Cen's stellar wind is also of great interest, since
even a moderate M dwarf mass­loss rate would have impor­
tant implications for the origin of cosmic­ray seed particles,
as well as heavy­element dispersal, kinetic heating, and ion­
ization throughout the interstellar medium (ISM). Mass
loss is also of critical importance in models of angular
momentum loss and other aspects of stellar evolution. With
very few exceptions, however, existing measurements of
stellar mass­loss rates do not extend below a few times 10 #10
M # yr #1 (4 orders of magnitude higher than the solar rate of
#2 # 10 #14 M# yr #1 ) and only apply to high­mass O and B
stars, red giants, and supergiants.
Lim &White (1996) provide a good summary of theoreti­
cal and observational constraints on dwarf star mass­loss
limits prior to 1997. A fairly recent theoretical e#ort is that
of Badalyan & Livshits (1992), who predict that magneti­
cally saturated M dwarf flare (dMe) stars can have winds
with mass­loss rates exceeding #10 #11 M # yr #1 .
Perhaps the strongest observational evidence in favor of
strong late­type dwarf winds comes from V471 Tau, an
eclipsing binary consisting of a K2 V star and a white dwarf.
Mullan et al. (1989) reported observing discrete absorption
features in the UV continuum of the white dwarf and inter­
preted them as arising from a cool (10 4 K) wind with a mass­
loss rate of 10 #11 M # yr #1 . Lim, White, & Cully (1996a),
however, argue that the detection of nonthermal radio emis­
sion from this system implies that any wind from the K
dwarf must be optically thin, which is probably inconsistent
with the temperature and mass­loss rate inferred by Mullan
et al. More recently, Bond et al. (2001) observed transient
absorptions in the Si iii 1206 A š line, which they ascribed to
coronal mass ejections (CMEs). Based on the number of
events and other viewing considerations, they estimate that
the active K star `` emits some 100--500 CMEs day #1 , as
The Astrophysical Journal, 578:503--514, 2002 October 10
# 2002. The American Astronomical Society. All rights reserved. Printed in U.S.A.
E
503

compared to #1--3 day #1 for the Sun. The K dwarf 's mass­
loss rate associated with CMEs is at least Ï5 25÷ # 10 #14
M # yr #1 , but it may well be orders of magnitude higher if
most of the silicon is in ionization states other than Si iii.''
Doyle & Mathioudakis (1991) and Mullan et al. (1992)
have also reported marginal detections of millimeter radia­
tion from several dMe stars, of which YZ CMi is the most
significant example. Based on a 2.2 # measurement at 1.1
mm and data in other wave bands from IRAS and the
VLA, 1 Mullan et al. (1992) inferred that YZ CMi has an
optically thick wind and loses mass at the rate of #5 # 10 #10
M # yr #1 . Houdebine, Foing, & Rodono ‘ (1990) recorded
optical spectra of another dMe star, AD Leo, and argued
that based on the observation of a CME during a flare, the
total flare­related mass loss was between 2:7 # 10 #13 and
4:4 # 10 #10 M # yr #1 . Lim & White (1996), however, argued
that existing radio data on both stars indicate the existence
of nonthermal coronal emission, which must arise near the
stellar surface and then propagate through an optically thin
wind to be detectable. Citing the detection of nonthermal
emission from YZ CMi at 327 MHz (Kundu & Shevgaon­
kar 1988), they derive a more model­dependent upper limit
of 5 # 10 #14 M # yr #1 for a 300 km s #1 10 4 K wind, or up to
#10 #12 M # yr #1 for a wind with temperature of #10 6 K and
velocity between 300 and 600 km s #1 . Van den Oord &
Doyle (1997) similarly derived a limit of no more than
#10 #12 M # yr #1 , based on existing IR and radio observa­
tions of several dMe stars. For the specific case of Proxima
Cen, based on its nondetection at 3.5 cm, Lim, White, & Slee
(1996b) directly placed a 3 # upper limit of #7 # 10 #12
M # yr #1 on its mass­loss rate, assuming a wind with T #
10 4 K and velocity #300 km s #1 (or #10 #11 M # yr #1 for
T # 10 6 K).
In the past few years, several papers utilizing a di#erent
wind detection method have been published, based on the
idea that interaction of an ionized stellar wind with neutral
gas in the ISM, via proton­hydrogen charge exchange (CX),
creates a `` hydrogen wall '' of warm neutral gas (Wood,
Alexander, & Linksy 1996; Gayley et al. 1997; Izmodenov,
Lallement, & Malama 1999). The signature of this gas is a
slight excess of Ly# absorption, which has been detected, at
least tentatively, in seven nearby stars (Wood et al. 2001 and
references therein). In four of those systems, three of which
are composed of late­type, main­sequence stars, the absorp­
tion is strong enough to have permitted mass­loss rate esti­
mates: very roughly, _
MM ¼ 10 _
M# M# for the active RS CVn
type binary # And (G8 IV--III+unknown) and 1 _
MM # for #
Ind (K5 V) (Mu ˜ ller, Zank, & Wood 2001); and _
MM ¼ 2 _
M# M#
for # Cen AB (G2 V+K0 V) and an upper limit of 0.2 _
MM#
for Proxima Cen (Wood et al. 2001). Some of the assump­
tions used to derive those rates, however, particularly
regarding the intrinsic Ly# profiles of the stars in question,
are controversial, and the resulting uncertainties are not
well understood. The D/H abundance ratios used in that
work may also be less secure than was assumed (Vidal­Mad­
jar & Ferlet 2002).
In an earlier paper (Wargelin & Drake 2001), we
described a more direct method of investigating the winds of
late­type dwarf stars, via the CX X­ray emission that results
as highly charged ions in the wind, particularly oxygen,
interact with neutral gas in the ambient ISM. The emission
mechanism is essentially identical to that for comets, first
explained by Cravens (1997), except that the neutral gas is
primarily atomic hydrogen rather than water vapor. To
briefly summarize, when a highly charged ion collides with
neutral gas, an electron can be transferred from the neutral
into an excited energy level of the wind ion, which then
decays and emits an X ray. Two­electron CX occurs roughly
10% of the time in wind­comet interactions (Greenwood et
al. 2000), but this process is unimportant in the H­domi­
nated (single­electron) neutral gas considered here.
Soon after the first X­ray observations of comets, Cox
(1998) pointed out that CX must occur throughout the
heliosphere as the solar wind interacts with neutral gas in
the ISM. Subsequent quantitative analyses (Cravens 2000;
Cravens, Robertson, & Snowden 2001) determined that this
mechanism accounts for a significant fraction of the
observed soft X­ray background, in agreement with indica­
tions from ROSAT observations (Snowden et al. 1995) that
roughly half of the 1
4 keV background comes from a `` local
hot plasma.'' As discussed by Wargelin &Drake (2001), this
same process must occur for any star with a highly ionized
wind residing inside a partially neutral region of the ISM.
By searching for the resulting distinctively profiled CX emis­
sion, stellar winds with mass­loss rates not much greater
than the Sun's can now be detected around nearby stars
with high­resolution, large­area X­ray observatories.
In this paper, we present results from Chandra observa­
tions of Proxima Cen, beginning with an overview of the
data and a discussion of light curves in x 2, followed by spec­
tral analysis of emission in quiescence and during a large
flare in x 3. In x 4, we derive a sensitive upper limit to the stel­
lar mass­loss rate, based on the observed level of CX X­ray
emission, marking the first application of this wind detec­
tion method, and we conclude with a discussion of model
uncertainties and prospects for reducing them.
2. THE OBSERVATIONS
Proxima Cen was observed twice by Chandra as a Guest
Observer target (PI J. Linsky) between 2000 May 7 and 9.
Both observations (observation IDs 49899 and 641) utilized
the ACIS­S CCD array in imaging mode, for 29,856 and
19,036 s, respectively. The intention was to use ACIS­S with
the High Energy Transmission Grating, but a hardware
fault prevented insertion of the grating.
Standard Chandra X­Ray Center pipeline products were
reprocessed to level 2 using the Chandra Interactive Analy­
sis of Observations (CIAO) software version 2.1.3, taking
advantage of recent gain map improvements for the central
(S3) CCD released in CALDB version 2.8. Various energy
filters were then applied to the data to increase the contrast
of secondary X­ray sources against the background. About
two dozen extraneous sources were removed from the S3
chip.
Unfortunately, because of the high counting rate of the
source and the chosen CCD frame time (3.2 s), more than
95% of the events in the main peak were rendered useless
because of pile­up and the associated phenomenon of grade
migration, which lead to distorted spectra and nonlinear
counting rates. To keep these e#ects at a negligible level, we
had to exclude data from the central core of the source.
1 The VLA (Very Large Array) is a facility of the National Radio
Astronomy Observatory (NRAO). The NRAO is a facility of the National
Science Foundation operated under cooperative agreement by Associated
Universities, Inc.
504 WARGELIN & DRAKE Vol. 578

Because so few events were left, we included counts from the
`` readout streak '' for use in spectral and light curve analy­
sis. The streak is an artifact of the CCD readout process,
with a net exposure time per frame equal to the number of
CCD rows (1024) multiplied by the time required to shift
the image by one row during readout (40 ls), or 0.041 s. The
readout streak exposure e#ciency when using a 3.2 s frame
time is therefore 0:041=Ï3:2 × 0:041÷ ¼ 1:26%, and pile­up
is completely negligible.
Light curves for each observation ID (Figs. 1 and 2) were
extracted from the source event files at energies up to 2 keV
using annuli of radii 4 and 30 pixels (2 00 and 15 00 ), centered
on the source, plus a 2 pixel--wide box to include the readout
streak. Background rates were derived from the entire S3
chip, after we excluded extraneous sources, a 200 pixel
radius circle around Proxima Cen, and a narrow strip
around the edges of the chip. The background was statisti­
cally uniform across the chip and showed no significant tem­
poral variability during either observation.
3. SPECTRAL ANALYSIS
3.1. Data Extraction
Based upon the light curves, we divided the observations
into flare and quiescent time ranges for spatial and spectral
analysis. The time ranges are marked as intervals F1, F2,
and Q1--4 and are summarized in Table 1. The latter part of
the flare (the 200 s between F1 and Q2) was not included in
F1 in order to avoid `` contamination '' of the flare's point­
spread function (PSF; see x 4.3). The 1000 s between Q3 and
F2 were also not analyzed.
Readout streaks, typically containing one or two hundred
counts, were used to estimate the true inner­core counting
rate in the absence of pile­up, after we excluded a 50 pixel
radius circle around the main peak and accounted for the
minor e#ects of background and X­ray events in the PSF
wings. Events from the background­subtracted outer core
and wings were then added to give the total corrected rates.
The preflare and postflare quiescent intervals of observation
Fig. 1.---Light curve for observation ID 49899, using 100 s time bins (50 s for flare detail). Zero time corresponds to 2000 May 7, 22:56:24 UT. The spatial fil­
ter is an annulus of radii 4 and 30 pixels, plus the 2 pixel--wide readout streak (radii 4 and 80 pixels, with 3 pixel streak in flare detail). Background is 0.16 counts
per 100 s time bin (0.40 per 50 s in flare detail). The Q1 and Q2 time ranges were included in the composite quiescent spectral analysis, and F1 was used for the
flare analysis.
Fig. 2.---Light curve for observation ID 641, using the same spatial filter
as for observation ID 49899. Zero time corresponds to 2000 May 9,
00:05:54 UT. Background is 0.16 counts per 100 s time bin. Q3 and Q4 were
included in the quiescent spectral analysis.
No. 1, 2002 CONSTRAINTS ON MASS LOSS FROM PROXIMA CEN 505

ID 49899 (Q1 and Q2) have the same count rate, whereas
the analogous intervals of observation ID 641 have higher
rates that di#er from each other by a factor of 2.
To maximize the number of counts available for spectral
analysis, we adjusted the sizes of the annuli and readout
streak boxes for each time interval, using as small an inner
radius as possible on the annuli while keeping pile­up e#ects
negligible. Inner radii were chosen by comparing the PSFs
of low­rate and high­rate data with each other and with cali­
bration models and by studying how spectral shapes and
hardness ratios varied depending upon how much of the
core was included in the extraction region. As shown in
Table 1, we used an inner radius of 3 pixels for quiescent
times and 4 pixels for the flares. Outer radii and readout
streak box widths were chosen to include as many X­ray
events as possible while keeping the relative background
contribution low and were di#erent for each time interval.
Extracted spectra for each of the time ranges are shown in
Figure 3. Data from Q1 and Q2 were combined, since their
rates and spectra were the same.
Spectral analysis was performed with the CIAO
SHERPA `` fitting engine '' (Freeman, Doe, & Siemiginow­
ska 2001), using the plasma emission code MEKAL (Kaas­
tra 1992; Liedahl, Osterheld, & Goldstein 1995). The main
aim of this analysis was to constrain the coronal element
abundances and to investigate the plasma parameters that
characterize the larger flare event. Because of the limited
number of events, data were combined from all times
deemed to be free of significant flaring (Q1+Q2+Q3+Q4),
in order to create a composite quiescent spectrum. For the
flare analysis, we used events from interval F1.
Energies in the range 0.3--2.0 keV were considered; below
this energy range the ACIS­S response becomes uncertain,
while at higher energies the data consist largely of back­
ground events. Parameter estimation was performed using
the modified # 2 statistic (Gehrels 1986) and was verified
using the C statistic (Cash 1979). Results are summarized in
Table 2.
3.2. Quiescent Spectrum
As noted in x 1, the photospheric metallicity of # Cen AB
(and by plausible extension, Proxima Cen) is known to be at
least as high as the Sun's. We therefore first attempted to
match the composite quiescent spectrum with isothermal
models corresponding to the solar photospheric composi­
TABLE 1
Spectral Data Extraction Parameters
Time Range
Annular Radii
(pixels)
Streak Width
(pixels)
Total
Spectrum
Counts
Estimated
Background
Counts
Corrected
X­Ray
Counts
Exposure
Live Time
(s)
Corrected Rate
(counts s #1 )
Q1+Q2 ......... 3, 30 2 983 47 19,330 # 1560 28,688 0.67 # 0.03
F1 ................. 4, 80 3 491 5 10,360 # 950 592 17.5 # 2.6
Q3................. 3, 35 3 475 18 7,690 # 950 7,450 1.03 # 0.13
F2 ................. 4, 60 3 566 11 13,420 # 1130 2,172 6.18 # 0.52
Q4................. 3, 40 3 914 23 16,800 # 1300 8,186 2.05 # 0.16
Note.---Spectra from time ranges Q1--Q4 were summed into a single quiescent spectrum for fitting. The spectrum of the large flare, F1,
was also analyzed. `` Corrected X­ray Counts '' is the estimated total number of X­ray events (not including the readout streak) at energies
below 2 keV if there had been no pile­up or grade migration e#ects. The live­time fraction is 0.987 for all time intervals.
TABLE 2
Summary of Model Parameter Estimation
Quiescent Flare
Model Metallicity a
kT
(keV) Normalization b Metallicity a
kT
(keV) Normalization c
Isothermal, solar abundance ............ [M/H] # 0 0.31 # 0.01 4.1 # 0.1 . . . . . . . . .
Isothermal, abundance­free ............. [M/H] = #1:0 ×0:1
#0:05 0.50 # 0.03 2.0 # 0.1 . . . . . . . . .
Isothermal, grouped abundance....... [Fe/H] = #0.9 # 0.1 0.37 # 0.03 22 # 2 [Fe/H] = #0:6 ×0:5
#0:2 0:81 ×0:4
#0:1 11 ×7
#4
[Mg/H] = #0.2 # 0.2 0.37 # 0.03 22 # 2 [Mg/H] = 0:1 ×0:4
#0:3 0:81 ×0:4
#0:1 11 ×7
#4
[Ne/H] = #0.2 # 0.1 0.37 # 0.03 22 # 2 [Ne/H] < 0.8 0:81 ×0:4
#0:1 11 ×7
#4
[O/H] = #0.7 # 0.1 0.37 # 0.03 22 # 2 . . . . . . . . .
Multithermal, solar abundance ........ [M/H] # 0 1.37 4.6 # 0.3 [M/H] # 0 1.37 92 # 14
0.68 1.4 # 0.3 0.68 16 # 9
0.34 1.4 # 0.2 0.34 8.5 # 7.2
0.17 1.5 # 0.4 . . . . . . . . .
0.085 <3.9 . . . . . . . . .
a All abundances are expressed in the usual logarithmic bracket notation relative to solar photospheric abundances tabulated by Anders &Gre­
vesse 1989. See also x 3.2.
b Normalizations for fits to the quiescent spectrum must be multiplied by 19.2 to account for the loss of events due to pile­up. The adjusted nor­
malizations correspond to the plasma EM in units of 10 #19 =Ï4#D 2
÷ R n e nH dV , where D is the distance to Proxima Cen (1.30 pc) and n e and nH are
the electron and hydrogen number densities, respectively.
c For the flare fits, normalizations must be multiplied by 21.1 to correct for pile­up losses.
506 WARGELIN & DRAKE Vol. 578

tion tabulated by Anders & Grevesse (1989). It should be
noted that this abundance tabulation has been superseded
by subsequent compilations of Grevesse and coworkers
(see, e.g., Grevesse & Sauval 1998), although the di#erences
are generally small (d0.1 dex) for the abundant elements
relevant to our study (N, O, Ne, Mg, Si, S, Ar, and Fe). Two
exceptions are worthy of remark here: first, Fe, for which
Grevesse & Sauval (1998) adopt recent solar photospheric
results that are in agreement with the value obtained from
carbonaceous chondrites, ½Fe=H# ¼ 7:50, instead of
½Fe=H# ¼ 7:67; second, O, for which there is a recent solar
measurement based on a non­LTE analysis of forbidden O i
lines indicating an abundance of ½O=H# ¼ 8:69 # 0:05
(Allende Prieto, Lambert, &Asplund 2001)---0.24 dex lower
than the Anders & Grevesse (1989) value. While noting
these di#erences between currently accepted values of solar
photospheric abundances and those of Anders & Grevesse
(1989), we retain the latter for our model analysis partly for
convenience (this compilation is `` hard­wired '' into the
SHERPA fitting engine), but also to easily enable cross­
comparison of our fitting results with those of other studies.
As will be seen, it turns out that these abundance di#erences
are in any event of little consequence for our modeling.
It was readily apparent that isothermal models were inad­
equate for representing the data in the vicinity of 1 keV,
below 0.5 KeV, and near 0.65 KeV (Fig. 4a). The sense of
the latter discrepancy is a model flux that is significantly
higher than the data indicate; the peak in the model here is
largely caused by the resonance Ly# line of H­like O.
Allowing the global metal abundance parameter to vary
yielded better matches (Fig. 4b), although with systematic
problems near 0.6 keV. The best fit, although still poor, in
this case was for a metal abundance relative to the solar
photosphere of M=H
½ # ¼ #1:0. The plasma temperature
parameter was also significantly di#erent from that for the
solar photospheric abundance case: 0.5kT (5:8 # 10 6 K),
Fig. 3.---Extracted source spectra for each time segment and the background spectrum. The relative contribution of the background (dotted curves) varies
in each source spectrum because of the di#erent extraction areas and source rates for each time segment.
No. 1, 2002 CONSTRAINTS ON MASS LOSS FROM PROXIMA CEN 507

versus 0.3kT (3:5 # 10 6 K) for the latter. An improved fit
(Fig. 4c) was obtained by varying the abundances, which we
grouped together because of the limited number of counts:
C, N, and O; Na, Mg, Al, Si, and S; Ne and Ar; and Fe, Ca,
and Ni. The most likely temperature for this case, 0.37kT
(4:3 # 10 6 K), is similar to that obtained with fixed solar
abundances (0.31kT). Note that in this case, the abundance
parameters for the Ne and Mg groups appear to be some­
what higher than those for the Fe and O groups (see Table
2); this is further discussed below.
While an isothermal model with varying element abun­
dances matches our data within statistical uncertainties, we
also investigated the propriety of multithermal models with
restricted abundance parameters. We find that a model
composed of four isothermal components on a regular loga­
rithmic temperature grid (log T ¼ 6:3, 6.6, 6.9, and 7.2), all
with solar photospheric abundances, also provides an
acceptable match to the observations. The emission mea­
sure (EM) parameters of the di#erent components for the
optimum match are illustrated in Figure 5, together with the
EMs for the di#erent isothermal models.
Fig. 4.---Composite quiescent spectrum, best­fit model, and residuals using the MEKAL radiative loss model with various assumptions. (a) Fixed solar pho­
tospheric abundances ([M/H] = 0.0). (b) Global metallicity scaling allowed to vary. The best­fit model corresponds to a metallicity of [M/H] = #1.0. (c)
Grouped element abundances allowed to vary independently. The most important elements in the four groups are O, Ne, Mg, and Fe. The best­fit model has
Ne and Mg abundance parameters somewhat higher than that for Fe (see text). (d ) Multithermal model with fixed metallicity, [M/H] = 0.0. The optimum
EMs of the di#erent temperature components are illustrated in Fig. 5.
Fig. 5.---ACIS­S EM parameters for the best­fit multithermal models of
flare and quiescent spectra, together with individual EMs from the various
isothermal models. Error bars were estimated using the full error projection
utility in the CIAO tool SHERPA.
508 WARGELIN & DRAKE Vol. 578

3.3. Flare Spectrum
The flare interval F1 contains fewer events than the
period of quiescence, so data were grouped to yield a mini­
mum of 10 counts in a single bin. Because F1 includes data
from the flare rise, peak, and much of the decay, we would
not expect the spectrum to be well fitted with a single tem­
perature. Nevertheless, an isothermal model with varying
abundance parameters (grouped as for the quiescent spec­
tral fits) is found to be acceptable within the fairly large
uncertainties of the data, although an isothermal model
with a solar photospheric abundance parameter is not. The
acceptability of the first fit, however, is likely just a reflection
of the poor statistical quality of the data, and we hesitate to
draw any firm conclusions regarding abundances. One
could argue, however, that if the source were indeed isother­
mal, then there is evidence of a higher Fe abundance during
the flare than during quiescence. Again, however, a multi­
thermal model (log T ¼ 6:6, 6.9, and 7.2) with solar photo­
spheric abundances was found acceptable. Adding another
component at log T ¼ 7:5 did not yield any significant
improvement to the fit. Results for all three model fits---iso­
thermal with solar abundances, isothermal with variable
abundances, and multithermal with solar abundances---are
summarized in Table 2.
3.4. Discussion
It is clear from the formal spectral analysis that the deci­
mation of photon events resulting from pile­up compro­
mises the data to an extent that useful formal constraint of
model abundance parameters is not possible for either qui­
escent or flare emission. However, in both cases, insistence
on an isothermal source model would imply that the Fe
abundance is considerably lower than the solar coronal
value and that Ne in particular is likely enhanced relative to
Fe. Such a pattern would be in keeping with the recent
results based on Chandra and XMM­Newton high­resolu­
tion spectra of the RS CVn type binaries HR 1099 (V711
Tau) and II Peg, in which strong Ne enhancements over Fe
were uncovered (Brinkman et al. 2001; Drake et al. 2001;
Phillips et al. 2001; Maggio et al. 2002). Moreover, Drake et
al. (2001) found from a literature survey that parameter esti­
mation analyses of low­resolution ASCA spectra of many
other active stars tended to yield abundance parameters for
Ne significantly higher than those for Fe, suggesting that
the phenomenon is universal in active stars.
Unfortunately, the quiescent data are also consistent with
a multithermal model with photospheric abundances. We
have no observational evidence to favor one solution over
another, although we do remark that the multithermal
model EMs tend to appear flatter as a function of tempera­
ture than might be expected based on EMs derived for other
stars (see, e.g., Fig. 2 in the review of Bowyer, Drake, &
Vennes 2000). We suggest, then, that the ACIS­S spectrum
presents some evidence that the coronal abundance anoma­
lies uncovered in other active stars are also shared by our
nearest neighbor.
In the case of the analysis of the flare interval F1, we are
again stymied by poor data quality resulting from pile­up.
Similar conclusions apply regarding the equal applicability
of models in which either abundance parameters are
allowed to vary or additional temperature components are
added. In this regard, we note that at some level the method
of parameter estimation using test statistics applied to a
large spectral range becomes meaningless when there are
obvious systematic di#erences between model and data in
any much narrower spectral interval: a test statistic that
indicates statistical concordance between model and data
(e.g., reduced # 2 < 1) does not tell the whole story, and the
propriety of the model becomes dependent on the energy
range adopted for computation of the statistic. In particu­
lar, the variable­abundance parameter and multithermal
models are both equally appropriate, based on the test sta­
tistic applied to the interval shown, but in the range 0.7--1
keV the multithermal model with solar photospheric abun­
dances presents an aesthetically better representation of the
data.
3.5. Quiescent Activity Level in the Context of
Earlier Observations
Haisch et al. (1998) presented a comparison of their
RXTE observations of Proxima Cen with earlier measure­
ments obtained by di#erent satellites. These disparate data
were compared via isothermal EM­versus­temperature loci.
These types of curves are described in detail by Drake
(1999). In brief, the EM­temperature (EM­T) locus is given
by the isothermal plasma EM at temperature T that yields
the observed broadband count rate. The observed count
rates and di#erent responses of the various instruments that
have observed Proxima Cen each define a di#erent locus in
the EM­T plane. Since, in the case of stellar coronae, the
plasma is not likely to be isothermal, the individual loci
actually represent the upper limit to the plasma EM at any
given T. If the true EM distribution is relatively sharply
peaked and the bandpasses in question are largely domi­
nated by lines formed at this peak temperature, then the iso­
thermal approximation can be reasonable.
We have taken the EM­T loci for the di#erent instrument
count rates reported for Proxima Cen from Haisch et al.
(1998) and show these in Figure 6. One small di#erence
between our Figure 6 and Figure 3 of Haisch et al. (1998)
Fig. 6.---Loci of isothermal EM vs. isothermal plasma temperatures that
correspond to di#erent instrument count rates reported for Proxima Cen in
the literature for observations between 1979 and 1996. The diamond corre­
sponds to the isothermal EM and temperature found in the optically thin,
collision­dominated model parameter estimation analysis for the case of
solar photospheric abundances. Here the units of EM have been converted
to the logarithm of the product N 2 e ÏT÷VÏT÷ in units of cm #3 . The arrow
indicates the isothermal plasma temperature obtained by Haisch, Antunes,
& Schmitt 1995 from their ASCA observation.
No. 1, 2002 CONSTRAINTS ON MASS LOSS FROM PROXIMA CEN 509

concerns the Einstein locus, which we represent here (shaded
region) as including the range of apparently quiescent count
rates 0.1--0.3 counts s #1 from the 1979 March 6--7 and 1980
August 20 observations (Haisch et al. 1980, 1983). The point
corresponding to our Chandra ACIS­S EM for an isother­
mal model for solar photospheric abundances overlaps with
this Einstein locus, but lies at the lower end of the range of
EMs previously obtained in the temperature range
log T ¼ 6:5 6:8 that characterizes the quiescent coronal
temperatures found for Proxima Cen. We have used the
solar photospheric abundance point because the other loci
are also based on a radiative loss model with solar photo­
spheric abundances; the EMs obtained when the global met­
allicity and grouped element abundances were allowed to
vary in the parameter estimation process are higher by
about 0.6 and 0.7 dex, respectively, and shift to slightly
higher temperatures. The isothermal plasma temperature
we obtain from our model parameter estimation for solar
photospheric abundances, 0.31 keV (3:6 # 10 6 K;
log T ¼ 6:56), is slightly cooler than the 6:1 # 10 6 K
obtained from the ASCA observation by Haisch et al.
(1995; arrow), but it is very similar to the estimates of
4 # 10 6 K by Haisch et al. (1980, 1983) based on the Einstein
observations. Proxima Cen thus appears to have been in a
relatively low state of activity at the time of the Chandra
observation.
4. CX AND THE STELLAR WIND
4.1. Spatial Distribution
A quantitative assessment of Chandra's ability to detect
stellar winds and a model of the spatial distribution of astro­
spheric CX emission for a spherically symmetric wind have
been described by Wargelin & Drake (2001). For this work,
we have used a more detailed model of the interaction of
Proxima Cen's stellar wind and its local ISM. As noted in
x 1, a `` hydrogen wall '' of neutral gas is expected to form
when an ionized wind interacts with neutral gas in the ISM,
with a strong density enhancement on the `` upwind '' side
when the star and surrounding gas have relative motion.
Proxima Cen and # Cen are surrounded by the `` G
cloud,'' for which Linsky et al. (2000) have derived a value
of 0.10 cm #3 for the neutral H density, n H i , based on
absorption­line spectra obtained by HST and EUVE. The
G cloud is moving roughly toward us from the direction of
Proxima Cen, toward l ¼ 184=5, b ¼ #20=5 (Lallement &
Bertin 1992), with virtually the same velocity and direction
of motion as the Local Interstellar Cloud surrounding the
Sun. Wood et al. (2001; see their Fig. 3) modeled the wind­
cloud interaction for a range of mass­loss rates and showed
that the neutral gas density within the hydrogen wall was
approximately twice that in the undisturbed ISM, while
behind the wall, inside the astrosphere, the neutral density
was about half that in the undisturbed gas.
Because of Proxima Cen's position and the cloud's direc­
tion of motion, the distribution of gas around it should be
approximately symmetric from our viewpoint. We use a
simplified model of the Wood et al. results, taking
nH ¼ 0:10 cm #3 in the undisturbed ISM, 0.20 cm #3 in a
hemispherical shell on the upwind side, and 0.05 cm #3 inside
and behind the shell, in the astrosphere. Several hemisphere
sizes were considered, as shown in Figure 7, corresponding
to mass­loss rates of 2, 1, 0.5, 0.2, and #0.05 _
MM # (the last
case being our own extrapolation). Near the star, n H i is
lower than the asymptotic astrospheric value because of CX
with the densest part of the solar wind (mostly protons), as
well as radiation pressure and photoionization. We use the
simple Cravens (2000) model for neutral gas density within
the astrosphere, n n r
Ï ÷ ¼ n n0 e ## n =r , where n n0 is the neutral
gas density far from the star (0.05 cm #3 ) and # n is the scale
for neutral gas depletion. For the Sun, # n for hydrogen is
roughly 5 AU upstream and 20 AU downstream. Because
Proxima Cen's mass­loss rate _
MM is comparable to or less
than the Sun's and # n scales as roughly _
MM 0.5 , we assume that
# n ¼ 5 AU; the actual value is of little importance to our
results.
As seen in Figure 7, when projected on the sky and
summed in annular bins around the star, the CX emission
profile rises steeply from the center and then slowly falls.
The deficit of emission at small radii is because of the afore­
mentioned neutral gas depletion near the star. Local max­
ima at intermediate radii arise from sight lines through the
sides of the hemispherical hydrogen wall, where CX emis­
sion is enhanced because of the higher neutral density. The
general fallo# in X­ray CX emission at large radii occurs
because highly charged ions are `` used up '' as they change­
exchange with neutral gas, recombining into lower charge
states that cannot emit at X­ray energies. We refer to CX
emission from ions in their initial (coronal) charge state as
`` primary '' emission (i.e., bare!H­like or H­like!He­
like CX), while photons from the second step of bare!H­
like!He­like CX are `` secondary '' emission.
The path length for CX is equal to 1=n n #CX , where n n is
the total neutral gas density and #CX is the cross section for
CX. The He/H abundance ratio is #10%, and He should
have at least as high a fraction of neutral atoms as H
because of its higher ionization potential, so we conserva­
Fig. 7.---Numerical simulations of CX emission distribution around
Proxima Cen. Solid lines are for models with a hemispherical hydrogen wall
with twice the ISM neutral density at various distances from the star and
half the ISM density inside the astrosphere (behind the wall). The short­
dashed curve is for when there is no hydrogen wall and the astrospheric
neutral density is the same as that of the undisturbed ISM. The long­dashed
curve is for the same case, but showing secondary CX emission. Vertical
dotted lines mark our annular search range (13--51 AU), which contains
16%--28% of the primary CX emission and roughly one­third as much sec­
ondary emission. [See the electronic edition of the Journal for a color version
of this figure.]
510 WARGELIN & DRAKE Vol. 578

tively set n n ¼ 1:1n H i . Few experimental measurements of
CX cross sections are available, and theoretical calculations
have large uncertainties, but a good compilation of avail­
able data on C and O ions colliding with H, He, and H 2 is
that of Phaneuf, Janev, & Pindzola (1987). Cross sections
for C, N, and O, along with estimated uncertainties and the
dominant CX capture energy levels, are listed in Table 3,
assuming a wind velocity of 550 km s #1 . For comparison,
the slow solar wind has a velocity of roughly 400 km s #1
(800 km s #1 for the fast wind), but Proxima Cen's wind is
probably somewhat faster, given its higher coronal tempera­
ture. (The significance of slow vs. fast winds is discussed in
x 4.2.) The actual velocity is not very important, as the rele­
vant CX cross sections vary by no more than about #10%
between 400 and 700 km s #1 .
To obtain an average #CX for use in our spatial emission
simulations, we applied a 10:1 weighting for H:He cross sec­
tions, with further weighting for each ion species according
to its emission contribution in the energy range of interest
(see x 4.2). We thus obtain an average cross section of
4:6 # 10 #15 cm 2 . With n n ¼ 0:055 cm #3 inside the astro­
sphere, the path length for CX around Proxima Cen is then
#240 AU (60 AU in the hydrogen wall and 120 AU in the
undisturbed ISM).
Proxima Cen's CX emission is thousands of times weaker
than its coronal emission, so the best place to look for CX
photons is in the wings of the PSF, but not so far out that
the background per unit radius (in annuli around the star),
which increases linearly with radius, becomes too large. For
Proxima Cen, our analysis shows that a good choice is an
annulus with inner and outer radii of 20 and 80 pixels, corre­
sponding to 10 00 and 39 00 , or 13 and 51 AU. The fraction of
primary CX emission, f CX1 , within this annulus lies between
0.16 and 0.28, depending upon the hydrogen wall location
(higher fractions for smaller H walls). Those same limits
apply to the cases in which the H wall is at infinity or does
not exist at all (so the astrospheric neutral density is the
same as in the undisturbed medium), respectively. We there­
fore adopt f CX1 ¼ 0:22 as the most likely value. The fraction
for secondary CX, f CX2 , is about 30% as large as f CX1 .
In addition to studying the e#ect of the hydrogen wall, we
also examined the sensitivities of f CX1 and f CX2 to variations
in the values of # n and n n #CX . Varying # n by a factor of 3,
corresponding to roughly a factor of 10 in mass­loss rate,
had little e#ect (less than 15% for f CX1 ). The overall uncer­
tainty in #CX , keeping in mind the dominance of O CX, is
20%--30%. Linsky et al. (2000) do not list a formal uncer­
tainty on n H i for the G cloud, but it appears to be compar­
able or smaller. A 50% error in n n #CX , which is therefore
rather pessimistic, leads to roughly a 40% change in f CX1
(33% higher f CX1 with larger n n # CX , 42% lower with smaller
n n # CX ). Variation in f CX2 is somewhat larger, but it is less
important because of its smaller contribution to total emis­
sion within the chosen annulus. Including uncertainties such
as those in the various hydrogen wall parameters and sim­
plistic spatial models yields an overall uncertainty of
roughly a factor of 2 in the value of f CX .
4.2. Spectrum
To further increase the contrast of CX emission against
coronal X­rays and the background, we focus on a narrow
energy range around He­like OK# for several reasons: oxy­
gen is the most abundant metal species in stellar winds; a
large fraction of oxygen ions in Proxima Cen's wind are
highly charged (bare and H­like) and therefore emit CX X­
rays; nearly all the CX X­ray emission from He­like ions is
emitted as K# photons (n ¼ 2 ! 1 transitions; 561--574
eV), because selection rules inhibit direct radiative decay to
ground from higher n levels; coronal emission is relatively
brighter at higher energies; and the ACIS detection e#­
ciency decreases rapidly at lower energies, while the back­
ground rises steeply (see Fig. 3, bottom left panel).
To decide how wide to make the energy range, and to
determine the expected photon detection rate, we con­
structed a simple model of the expected CX spectrum,
including emission from C and N as well as from O. Other
ions either have insignificant abundances or do not emit X­
rays near 570 eV (with the minor exceptions of Li­like Si
and S, which we discuss later). As noted above, H­like ions
charge­exchange to create He­like ions, which emit K# pho­
tons almost exclusively. Bare ions charge­exchange into H­
like ions, preferentially populating the n # q 0:7 level (see
Table 3), where q is the ion charge. Modeling the resulting
cascade process is di#cult and depends on, among other
things, knowing the initial sublevel population distribution
immediately following electron transfer, but roughly half of
the excited electrons will decay directly to ground and the
other half will cascade down to the n ¼ 2 level and then emit
a Ly# photon (Kharchenko & Dalgarno 2001; Beiersdorfer
et al. 2000). The ACIS­S FWHM energy resolution is #130
eV at the energies of interest here, compared to the tens of
eV separating high­n levels in H­like C, N, and O, so the
exact distribution of high­n Lyman series emission is unim­
portant in our analysis. Model line energies and Chandra's
e#ective area at each energy are listed in Table 4.
We assume that elemental abundances in Proxima Cen's
wind are the same as for the Sun, and we model its ionic
composition by analogy to the `` slow '' solar wind, with
adjustments to reflect the higher temperature of Proxima
Cen's corona. The solar abundances, ion fractions, and
`` freezing­in '' temperatures (T freeze ) listed in Table 5 are
TABLE 3
Charge Exchange Cross Sections
Ion
Neutral
Species
#CX
(10 #15 cm 2 )
Uncertainty
(%)
Capture
Levels
C 5+ ...... H 2.2 a 20 n = 4
He 1.5 a 50 n = 3
C 6+ ...... H 3.9 a 20 4f, 4d
He 0.9 a 40 3p, 3d
N 7+ ...... H 5.0 b 30 n = 5
He 3.0 c 40 n = 4
O 7+ ...... H 5.1 a 20 n = 4, 5
He 1.5 a 30 n = 4
O 8+ ...... H 5.0 a 20 5f, 5g, 5d
He 3.0 a 30 n = 4
a Data from Phaneuf et al. 1987, assuming a collision velocity of
550 km s #1 . Cross sections vary by less than 10% from the listed
value between 400 and at least 700 km s #1 .
b The theoretical cross sections of Harel, Jouin, & Pons 1998 for
O 8+ and N 7+ +H are virtually identical (5:66 # 10 #15 cm 2 at 440 km
s #1 ), so we apply the Phaneuf et al. 1987 value for O 8+ to N 7+ .
c By analogy, we use the same cross section for N 7+ +He as for
O 8+ +He. Cross sections for He­like Si, He­like and Li­like S, and H­
like N 6+ are assumed to be 4 # 10 #15 cm 2 ; these are minor contribu­
tors, and errors in their cross sections are unimportant.
No. 1, 2002 CONSTRAINTS ON MASS LOSS FROM PROXIMA CEN 511

taken from the tabulations of von Steiger et al. (2000) and
Schwadron & Cravens (2000), based on data collected by
the Ulysses Solar Wind Ion Composition Spectrometer at a
distance of roughly 1 AU from the Sun. Note that the
T freeze ­values, which characterize the ionization state of ele­
ments within the wind, are much higher than the kinetic
temperature of the wind, which is only of order 10 4 K.
The justification for basing our model on solar slow­wind
parameters has two parts. The first is the much greater sur­
face coverage of active areas on Proxima Cen, compared to
the Sun. On the Sun, it is these active regions that are the
source of the slow wind. The observed coronal X­ray flux
from Proxima Cen in quiescence, after correction for pile­
up, is about 3 # 10 #12 ergs cm 2 s #1 , corresponding to a total
coronal luminosity of somewhat more than 1 # 10 27 ergs
s #1 (including a factor of 2 to account for the unseen side of
the stellar disk). This level is comparable to the typical value
for the total solar coronal X­ray luminosity at solar maxi­
mum, derived by Peres et al. (2000), of 4:7 # 10 27 ergs s #1 .
Based on the radius of 0.17 R # for Proxima Cen (Reid &
Gilmore 1984), the surface X­ray flux of Proxima Cen dur­
ing the Chandra observation is a factor of #10 higher than
that of the Sun during solar maximum. If the coronal
plasma density of Proxima Cen is comparable to that of
solar active regions, as seems likely based on Chandra meas­
urements for other active M dwarfs (see, e.g., Gu ˜ del et al.
2001; Maggio et al. 2002), then the surface coverage of emit­
ting regions on Proxima Cen is significantly greater than
that on the Sun during its most active time. Indeed, Lim et
al. (1996b) deduced from radio observations that up to 88%
of Proxima Cen's surface might be covered by nonflaring
active regions and that up to 13% might be covered by even
hotter flaring regions. Our second argument is that any
`` fast '' wind, although it will be somewhat less ionized than
the slow wind (D log T freeze ¼ #0:16 for C and O in the solar
wind), will still be `` hot '' enough that nearly all C and N
and most O ions will be either fully ionized or H­like.
To estimate the fraction of bare and H­like C, N, and O
ions in Proxima Cen's wind, we use the ion balance tabula­
tions of Mazzotta et al. (1998), with freezing­in tempera­
tures based on solar values, but scaled up to reflect Proxima
Cen's hotter corona. At solar maximum, log T corona ¼ 6:30,
while for Proxima Cen, log T corona # 6:55, excluding its fre­
quent flares, which are even hotter. We therefore expect
T freeze ­values for Proxima Cen's wind to be at least
D log T ¼ 0:25 higher than for the solar wind. As shown in
Table 5, von Steiger et al. (2000) state that log T freeze ¼ 6:20
for O in the slow solar wind, but they note that a single­tem­
perature characterization becomes increasingly suspect for
higher Z elements. Based upon the actual ion fractions listed
by Schwadron & Cravens (2000) and comparisons to the
Mazzotta et al. (1998) tables, we find that log T freeze ¼ 6:20
best matches the H­like O ion fraction, while
log T freeze ¼ 6:30 corresponds to the observed bare O frac­
tion. We therefore assume that the best value of log T freeze
for O ions in Proxima Cen's wind is 6:25 × 0:25 ¼ 6:50 and
for C and N roughly 6.40 (see Table 5). The exact value for
C and N is unimportant, because C is more than 90% fully
ionized above log T ¼ 6:30 and N contributes only a few
percent to the total emission within the energy range of
interest.
To allow for uncertainty in the actual ion fractions, we
created model spectra for log T freeze ¼ 6:4, 6.5, and 6.6. The
intensity of each line was weighted by elemental abundance
relative to O, ion fraction, line yield (1 for He­like K# and
0.5 for Ly# and the high­n Lyman lines), fraction of emis­
sion within the chosen annulus (f CX1 or f CX2 , as appropri­
ate), and Chandra/ACIS­S e#ective area at the line energy.
Results, assuming a detector energy resolution of 130 eV,
are shown in Figure 8.
Based on those spectra, and with the aim of maximizing
the contrast of CX emission versus the corona and the
detector background, we chose the energy range 453--701
eV. This range contains 97% of the ACIS pulse­height distri­
bution for the OK# line, 80% of O Ly#, and 54% of C Ly#.
After we assigned these last weighting factors to each line,
the net e#ective area for detecting CX photons, normalized
to O abundance (A relO in cm 2 ), was then calculated for each
of the three models. The results, using freezing­in tempera­
tures of log T ¼ 6:4, 6.5, and 6.6, are A relO ¼ 73, 82, and 83
cm 2 , respectively. The fraction of bare or H­like O ions
ranges from 77% to 97%, and oxygen emission contributes
very nearly 76% of the total A relO in each case. For compari­
son, A relO ¼ 28 cm 2 for the slow solar wind, and A relO ¼ 137
cm 2 if all ions are fully stripped.
He­like Si ions have significant abundance at the T freeze ­
values we assume and emit Li­like 5; 6 ! 3 CX photons
around 460 eV. Li­like and Be­like S also contribute some
emission, adding another #6 cm 2 to A relO , assuming
#CX # 4 # 10 #15 cm 2 . We adopt a total A relO ¼ 85 cm 2 , with
TABLE 4
Modeled CX Emission Lines
Line
Energy
(eV)
E#ective Area
(cm 2 )
CK#........ 304 55
CLy# ...... 368 200
CLy# ....... 459 322
NK# ....... 426 274
N Ly# ...... 500 371
N Ly# ...... 625 416
OK#........ 570 359
OLy# ...... 654 443
OLy# ....... 836 561
Note.---H­like Lyman emission is
assumed to be divided equally between
Ly# and transitions from the CX capture
level directly to ground. E#ective areas
have an uncertainty of approximately
10%.
TABLE 5
Parameters for the Slow Solar Wind
Element
Relative
Abundance
H­like
Fraction
Bare
Fraction
log T freeze
(K)
C ................. 0.670 0.31 0.48 6.13
N ................ 0.0785 0.74 0.08 6.15
O................. 1 0.20 0.07 6.20
Note.---Data are taken from von Steiger et al. 2000 and Schwadron
& Cravens 2000. The value of T freeze for N ions was not listed, and it is
our estimate. As noted by von Steiger et al., a single­temperature char­
acterization becomes problematic with increasing Z; based on the tabu­
lation of Mazzotta et al. 1998, we estimate that the best log T freeze value
for O lies between 6.2 and 6.3. For the fast solar wind, log T freeze is
roughly 0.16 lower for each element.
512 WARGELIN & DRAKE Vol. 578

an uncertainty of 10--15 cm 2 to reflect the spread among
results from the di#erent ionization models and to allow for
uncertainties in the applicability of the slow­wind scenario.
We allow another 10--15 cm 2 for uncertainties in radiative
branching ratios (primarily for the Lyman lines). Errors
caused by uncertainties in the abundances of C, N, Si, and S
will be small, because O dominates the spectrum. Adding
the errors in quadrature, we estimate that the net uncer­
tainty in A relO , exclusive of errors in f CX1 and f CX2 , which
were already discussed, is no more than 25 cm 2 .
4.3. Determination of Mass­Loss Limit by Comparison of
Quiescent and Flare PSFs
Very accurate knowledge of Chandra's PSF in the specific
energy range of interest (453--701 eV) is required in order to
look for an excess arising from CX emission. Rather than
relying on existing PSF models, which are well calibrated at
only a few widely spaced energies and even then appear to
underestimate the amount of power in the wings, we use the
observed PSF during the bright flare (F1), which will have
negligible contamination by any CX emission. The PSF is
roughly independent of photon energy below #1 keV, so we
use a slightly larger energy range (307--1007 eV) for the flare
to improve the statistics. The fraction of flare events in the
chosen energy range contained within the annulus of radii
10 00 and 39 00 is then compared against the fraction in the
same annulus during the time when the coronal emission is
weakest (Q1+Q2) and CX photons will be easiest to see.
We find that the fractions are the same to within statisti­
cal uncertainty. Based upon the fraction of flare events
(0:0096 # 0:0019), 57 # 12 events would be expected during
the quiescent period in the absence of any CX counts, while
39 # 11 were observed. We can, however, put an upper limit
on the mass­loss rate by computing how many events
beyond the expected 57 would be required to detect a statis­
tically significant CX signature. We find that 17 excess
events would be required for a 1 # detection, 36 for 2 #, and
61 for 3 #.
We convert those detection limits to mass­loss rates by
comparing the number of CX counts expected if Proxima
Cen's wind has the same element abundances as the solar
wind. The number of detected events is given by
N det ¼
ROA relO t
4#d 2 ; Ï1÷
where R O is the rate at which O ions are injected into the
wind, A relO is the net e#ective area (normalized to O abun­
dance), taking into account all the spatial and spectral
detection details described in xx 4.1 and 4.2, t is the exposure
time, and d is the distance to Proxima Cen (1.30 pc).
The solar mass­loss rate, _
MM # , is 2 # 10 #14 M # yr #1 , where
M # is 2 # 10 33 g. With abundance ratios of 0.044 for He rel­
ative to H (Schwadron & Cravens 2000) and 0.002 for heav­
ier elements, the average wind ion has a mass of 1.155 amu,
which yields a total rate of 6:57 # 10 35 ions s #1 injected into
the wind, of which 6:27 # 10 35 ions s #1 are hydrogen.
Schwadron & Cravens list the O abundance relative to H as
5:6 # 10 #4 , so RO ¼ 3:51 # 10 32 O ions s #1 . The total
Q1+Q2 exposure time t is 28,688 s, and A relO ¼ 85 cm 2 , so
N det ¼ 4:2 counts are expected for a 1 _
MM# mass­loss rate.
Comparing this with the statistical upper limits on observed
counts derived earlier, we see that the 1 # upper limit on
Proxima Cen's mass­loss rate is 4.0 _
MM# , with 8.6 _
MM# and 14
_
MM # for the 2 # and 3 # limits, respectively.
As discussed at the end of x 4.1, systematic uncertainties
in our model of the spatial distribution of CX emission lead
to roughly a factor of 2 uncertainty in our results. Errors in
spectral modeling are considerably smaller, so the last major
source of uncertainty is the abundance of O relative to H in
the wind. We assumed that this last parameter is the same as
for the Sun, but it could easily di#er by 50%. Our estimate of
overall model uncertainty is therefore roughly a factor of 3.
Both the systematic and statistical uncertainties should
be easy to improve in the near future. The question of Prox­
ima Cen's coronal, and therefore wind, metallicity should
be answered by high­resolution spectra recently obtained by
Chandra and XMM­Newton. Experimental and theoretical
studies of CX cross sections and the relevant radiative path­
ways should likewise permit more precise spectral models in
the next few years, further reducing systematic uncertain­
ties. As for statistical limitations, ongoing improvements in
the calibration of Chandra's PSF in the wings will allow the
derivation of tighter limits on the mass­loss rate, as would,
of course, more Chandra observations.
Note added in manuscript.---Mass­loss rates for an addi­
tional four stars have recently been measured using the Ly#
technique, as reported by Wood et al. (2002).
We gratefully acknowledge very helpful discussions with
John Raymond and Vasili Kharchenko. The authors were
supported by NASA contract NAS8­39073 to the Chandra
X­Ray Center during the course of this research. B. W. was
also supported by NASA's Space Astrophysics and Analy­
sis program under grant NAG5­10443.
Fig. 8.---Model spectra of CX emission from Proxima Cen's wind within
an annulus of radii 10 00 and 39 00 , with ACIS resolution of 130 eV FWHM.
Ion balances were derived from Mazzotta et al. 1998 for three freezing­in
temperatures, assuming slow solar wind element abundances (von Steiger
et al. 2000). The integrated area under each curve and between the dotted
vertical lines (marking the energy range of interest) is equal to the e#ective
area for detection of photons within that energy range and annulus by
Chandra. Major emission lines at logT freeze = 6.4 are also shown, with 10
eV resolution. The minor contributions of Si and S ions are not included.
[See the electronic edition of the Journal for a color version of this figure.]
No. 1, 2002 CONSTRAINTS ON MASS LOSS FROM PROXIMA CEN 513

REFERENCES
Agrawal, P. C., Rao, A. R., & Sreekantan, B. V. 1986, MNRAS, 219, 225
Allende Prieto, C., Lambert, D. L., &Asplund, M. 2001, ApJ, 556, L63
Anders, E., &Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53, 197
Anosova, J., Orlov, V. V., & Pavlova, N. A. 1994, A&A, 292, 115
Badalyan, O. G., &Livshits, M. A. 1992, Soviet Astron., 36, 70
Beiersdorfer, P., et al. 2000, Phys. Rev. Lett., 85, 5090
Benedict, G. F., et al. 1998, AJ, 116, 429
---------. 1999, AJ, 118, 1086
Bond, H. E., Mullan, D. J., O'Brien, M. S., & Sion, E. M. 2001, ApJ, 560,
919
Bowyer, S., Drake, J. J., &Vennes, S. 2000, ARA&A, 38, 231
Brinkman, A. C., et al. 2001, A&A, 365, L324
Cash, W. 1979, ApJ, 228, 939
Cox, D. P. 1998, in The Local Bubble and Beyond, ed. D. Breitschwerdt,
M. J. Freyberg, & J. Tru ˜ mper (Berlin: Springer), 121
Cravens, T. E. 1997, Geophys. Res. Lett., 24, 105
---------. 2000, ApJ, 532, L153
Cravens, T. E., Robertson, I. P., & Snowden, S. L. 2001, J. Geophys. Res.,
106, 24883
Doyle, J. G., &Mathioudakis, M. 1991, A&A, 241, L41
Drake, J. J. 1999, ApJS, 122, 269
---------. 2002, in ASP Conf. Ser. 277, Stellar Coronae in the Chandra and
XMM­Newton Era, ed. F. Favata & J. J. Drake (San Francisco: ASP), in
press
Drake, J. J., Brickhouse, N. S., Kashyap, V., Laming, J. M.,
Huenemoerder, D. P., Smith, R., &Wargelin, B. J. 2001, ApJ, 548, L81
Feldman, U., &Laming, J. M. 2000, Phys. Scr., 61, 222
Fleming, T. A., Giampapa, M. S., Schmitt, J. H. M. M., & Bookbinder,
J. A. 1993, ApJ, 410, 387
Freeman, P., Doe, S., & Siemiginowska, A. 2001, Proc. SPIE, 4477, 76
Gayley, K. G., Zank, G. P., Pauls, H. L., Frisch, P. C., & Welty, D. E.
1997, ApJ, 487, 259
Gehrels, N. 1986, ApJ, 303, 336
Greenwood, J. B., Williams, I. D., Smith, S. J., & Chutjian, A. 2000, ApJ,
533, L175
Grevesse, N., & Sauval, A. J. 1998, Space Sci. Rev., 85, 161
Gu ˜ del, M., Audard, M., Magee, H., Franciosini, E., Grosso, N., Cordova,
F. A., Pallavicini, R., &Mewe, R. 2001, A&A, 365, L344
Haisch, B., Antunes, A., & Schmitt, J. H. M. M. 1995, Science, 268, 1327
Haisch, B., Kashyap, V., Drake, J. J., &Freeman, P. 1998, A&A, 335, L101
Haisch, B. M., Linsky, J. L., Bornmann, P. L., Stencel, R. E., Antiochos,
S. K., Golub, L., &Vaiana, G. S. 1983, ApJ, 267, 280
Haisch, B. M., Linsky, J. L., Harnden, F. R., Jr., Rosner, R., Seward,
F. D., &Vaiana, G. S. 1980, ApJ, 242, L99
Harel, C., Jouin, J., & Pons, B. 1998, At. Data Nucl. Data Tables, 68, 279
Houdebine, E. R., Foing, B. H., &Rodono ‘ , M. 1990, A&A, 238, 249
Izmodenov, V. V., Lallement, R., &Malama, Y. G. 1999, A&A, 342, L13
Jay, J. E., Guinan, E. F., Morgan, N. D., Messina, S., & Jassour, D. 1996,
BAAS, 189, 120.04
Kaastra, J. S. 1992, An X­Ray Spectral Code for Optically Thin Plasmas
(updated version 2.0; Leiden: Internal SRON­Leiden Report)
Kharchenko, V., &Dalgarno, A. 2001, ApJ, 554, L99
Kundu, M. R., & Shevgaonkar, R. K. 1988, ApJ, 334, 1001
Lallement, R., & Bertin, P. 1992, A&A, 266, 479
Liedahl, D. A., Osterheld, A. L., &Goldstein, W. H. 1995, ApJ, 438, L115
Lim, J., &White, S. M. 1996, ApJ, 462, L91
Lim, J., White, S. M., &Cully, S. L. 1996a, ApJ, 461, 1009
Lim, J., White, S. M., & Slee, O. B. 1996b, ApJ, 460, 976
Linsky, J. L., Redfield, S., Wood, B. E., & Piskunov, N. 2000, ApJ, 528,
756
Maggio, A., Drake, J. J., Kashyap, V., Peres, G., & Sciortino, S. 2002, in
ASP Conf. Ser. 277, Stellar Coronae in the Chandra and XMM­Newton
Era, ed. F. Favata & J. J. Drake (San Francisco: ASP), in press
Matthews, R., &Gilmore, G. 1993, MNRAS, 261, L5
Mazzotta, P., Mazzitelli, G., Colafrancesco, S., & Vittorio, N. 1998,
A&AS, 133, 403
Mullan, D. J., Doyle, J. G., Redman, R. O., & Mathioudakis, M. 1992,
ApJ, 397, 225
Mullan, D. J., Sion, E. M., Bruhweiler, F. C., & Carpenter, K. G. 1989,
ApJ, 339, L33
Mu ˜ ller, H.­R., Zank, G. P., &Wood,B. E. 2001, ApJ, 551, 495
Peres, G., Orlando, S., Reale, F., Rosner, R., & Hudson, H. 2000, ApJ,
528, 537
Phaneuf, R. A., Janev, R. K., & Pindzola, M. S. 1987, Atomic Data for
Fusion, Vol. 5: Collisions of Carbon and Oxygen Ions with Electrons, H,
H 2 , and He (ORNL­6090; Oak Ridge: Oak Ridge Natl. Lab.)
Phillips, K. J. H., Mathioudakis, M., Huenemoerder, D. P., Williams,
D. R., Phillips, M. E., &Keenan, F. P. 2001, MNRAS, 325, 1500
Reid, N., &Gilmore, G. 1984, MNRAS, 206, 19
Schwadron, N. A., &Cravens, T. E. 2000, ApJ, 544, 558
Snowden, S. L., et al. 1995, ApJ, 454, 643
van den Oord, G. H. J., &Doyle, J. G. 1997, A&A, 319, 578
Vidal­Madjar, A., &Ferlet, R. 2002, ApJ, 571, L169
von Steiger, R., et al. 2000, J. Geophys. Res., 105, 27217
Wargelin, B. J., &Drake, J. J. 2001, ApJ, 546, L57
White, N. E. 1996, in ASP Conf. Ser. 109, Cool Stars, Stellar Systems and
the Sun, ed. R. Pallavicini &A.Dupree (San Francisco: ASP), 193
Wood, B. E., Alexander, W. R., &Linsky, J. L. 1996, ApJ, 470, 1157
Wood, B. E., Linsky, J. L., Mu ˜ ller, H.­R., & Zank, G. P. 2001, ApJ, 547,
L49
Wood, B. E., Mu ˜ ller, H.­R., Zank, G. P., & Linsky, J. L. 2002, ApJ, 574,
412
514 WARGELIN & DRAKE