Accepted for publication in The Astrophysical Journal

Preprint typeset using LATEX style emulateapj v. 12/16/11

ORBITAL AND PHYSICAL PROPERTIES OF THE PLEIADES BINARY 27 TAU (ATLAS)

Guillermo Torres1 , Andrew Tkachenko2 , Krešimir Pavlovski3 , Seth Gossage4 , Gail H. Schaefer5 , Carl Melis6 ,
Michael Ireland7 , John D. Monnier8 , Narsireddy Anugu5 , Stefan Kraus9 , Cyprien Lanthermann5 , Kathryn
Gordon10 , Robert Klement11 , Simon J. Murphy12 , and Rachael M. Roettenbacher8
Accepted for publication in The Astrophysical Journal
arXiv:2507.15933v1 [[Link]] 21 Jul 2025

ABSTRACT
We report new spectroscopic and interferometric observations of the Pleiades binary star Atlas, which
played an important role nearly three decades ago in settling the debate over the distance to the cluster
from ground-based and space-based determinations. We use the new measurements, together with
other published and archival astrometric observations, to improve the determination of the 291-day
orbit and the distance to Atlas (136.2±1.4 pc). We also derive the main properties of the components,
including their absolute masses (5.04 ± 0.17 M⊙ and 3.64 ± 0.12 M⊙ ), sizes, effective temperatures,
projected rotational velocities, and chemical composition. We find that the more evolved primary star
is rotationally distorted, and are able to estimate its oblateness and the approximate orientation of
its spin axis from the interferometric observations. The spin axis may well be aligned with the orbital
axis. Models of stellar evolution from MESA that account for rotation provide a good match to all
of the primary’s global properties, and point to an initial angular rotation rate on the zero-age main
sequence of about 55% of the breakup velocity. The current location of the star in the H-R diagram
is near the very end of the hydrogen-burning main sequence, at an age of about 105 Myr, according
to these models. Our spectroscopic analysis of the more slowly-rotating secondary indicates that it is
a helium-weak star, with other chemical anomalies.

1. INTRODUCTION Binary stars in the cluster featured prominently among

After the publication of the Hipparcos catalog (ESA the approaches to this problem, as they permit inde-
1997), a heated debate ensued in the literature about the pendent estimates of the distance that can be very pre-
mean distance to the Pleiades cluster. Significant tension cise. One of the objects used for this purpose was the
was found between the result from the satellite obser- A-type eclipsing binary HD 23642 (V1229 Tau; Miles
vations, ∼118 pc (van Leeuwen 1999), and the classical 1999; Torres 2003). It was first analyzed by Munari et
ground-based determinations, which consistently pointed al. (2004) to argue convincingly in favor of the ground-
to larger distances of ∼130 pc (e.g., Meynet et al. 1993; based distance determinations, and was later reanalyzed
van Leeuwen 2009a). For reviews on this subject, the by others (Southworth et al. 2005; Groenewegen et al.
reader may consult, e.g., Stello & Nissen (2001), Perci- 2007; David et al. 2016; Southworth et al. 2023). An-
val et al. (2005), or van Leeuwen (2007, 2009b). other Pleiades binary that played a central role is Atlas
(27 Tau, HD 23850, V = 3.63, B8 V), the second bright-
est object in the cluster, and the subject of this paper.
1 Center for Astrophysics | Harvard & Smithsonian, 60 Gar-
Atlas is a 291 day binary whose duplicity was discov-
den St., Cambridge, MA 02138, USA; gtorres@[Link]
2 Institute of Astronomy, KU Leuven, Celestijnenlaan 200D, ered in 1968 by the lunar occultation technique (Nather
3001 Leuven, Belgium & Evans 1970; Evans 1971).13 It was also resolved much
3 Department of Physics, Faculty of Science, University of Za-
later by Pan et al. (2004), using long-baseline interferom-
greb, 10000 Zagreb, Croatia
4 Center for Interdisciplinary Exploration and Research in As- etry. These authors combined their well constrained as-
trophysics (CIERA), Evanston, IL 60201, USA trometric orbit with a mass-luminosity relation, and ob-
5 The CHARA Array of Georgia State University, Mount Wil- tained an estimate of the distance that was again consis-
son Observatory, Mount Wilson, CA 91023, USA
6 Department of Astronomy & Astrophysics, University of Cal-
tent with the traditional ground-based values. Zwahlen
et al. (2004) improved on this analysis by providing
ifornia, La Jolla, CA 92093-0424, USA
7 Research School of Astronomy and Astrophysics, Australian further interferometric observations, but more impor-
National University, Canberra ACT 2601, Australia tantly, by obtaining the first double-lined spectroscopic
8 Astronomy Department, University of Michigan, Ann Arbor,
orbit for Atlas. This permitted a completely model-
MI 48109, USA
9 Astrophysics Group, Department of Physics & Astronomy,
University of Exeter, Exeter EX4 4QL, UK 13 These authors initially expressed reservations about their de-
10 Mitchell Community College, Statesville, NC 28677, USA
tection. However, in a subsequent paper with others (McGraw et
11 European Organisation for Astronomical Research in the
al. 1974), the observation was re-reduced and found to be sup-
Southern Hemisphere (ESO), Casilla 19001, Santiago 19, Chile ported by additional occultation events, confirming it as a genuine
12 Centre for Astrophysics, University of Southern Queens- detection. See below for a much earlier but spurious claim of du-
land, Toowoomba, QLD 4350, Australia plicity.
2 Torres et al.

independent determination of the distance via the orbital 6.8 km s−1 . Reductions were performed with a dedicated
parallax. These two binaries, along with other more di- pipeline (see Buchhave et al. 2012), and wavelength solu-
rect methods of measuring the distance that included tions relied on exposures of a thorium-argon lamp taken
space-based trigonometric parallax determinations with before and after each science exposure. The determi-
HST (Soderblom et al. 2005) and VLBI parallax mea- nation of RVs from this material is described later in
surements from radiointerferometry (Melis et al. 2014), Section 6.
finally settled the distance debate and demonstrated con- 3. ASTROMETRIC OBSERVATIONS
vincingly that Hipparcos was in error.
Along with the interferometric orbit of Atlas, the 3.1. CHARA Array
Zwahlen et al. (2004) study found that both B-type com- The Center for High Angular Resolution Astronomy
ponents are rapid rotators, with v sin i values they es- (CHARA) Array is a long-baseline optical interferome-
timated to be about 240 km s−1 for the primary and ter operated by Georgia State University, and located
60 km s−1 for the secondary. Their masses were inferred at Mount Wilson Observatory in southern California
to be 4.74 and 3.42 M⊙ , with relative uncertainties of (USA). The CHARA Array combines the light from up
5.3% and 7.3%, respectively. Additionally, the angular to six 1 m telescopes (S1, S2, E1, E2, W1, W2) with
diameter of the primary has been measured more recently baselines ranging from 34 to 331 m (ten Brummelaar
by Gordon et al. (2019), which enables its absolute radius et al. 2005, 2016). We used new and archival observa-
to be inferred, once a distance is adopted. All of these tions of Atlas recorded with the MIRC-X (Anugu et al.
properties make Atlas potentially useful for an indepen- 2020), MYSTIC (Setterholm et al. 2023), CLIMB (ten
dent age determination for the Pleiades, as its location Brummelaar et al. 2013), and PAVO (Ireland et al. 2008)
at the very tip of the main sequence of the cluster means beam combiners. Table 1 presents a log of CHARA ob-
it is among the most sensitive and favorable objects for servations that lists the UT date, instrument, bandpass
that purpose. (λ), telescopes, number of data sets recorded on Atlas
Because of this, and in view of the historical impor- (N ), and calibrator stars. To calibrate the interferomet-
tance of Atlas in the context of the Hipparcos distance ric transfer function, observations of either unresolved or
controversy, we have gathered independent spectroscopic small angular diameter calibrator stars were interspersed
observations to revisit the determination of its proper- before and after the observations of Atlas. Table 2 lists
ties, and to compare them for the first time against cur- the uniform disk diameters (UDD) adopted for the cali-
rent models of stellar evolution. Additional interferomet- brators (Bourgés et al. 2014).
ric observations have been obtained as well, strengthen- MIRC-X combines the light from six telescopes and
ing the constraint on the size of the primary. operates in the near-infrared H band. In 2022, we also
The paper is structured as follows. The new spectro- recorded data with the six telescope K-band combiner
scopic observations are described in Section 2. We re- MYSTIC that operates simultaneously with MIRC-X.
port our interferometric observations in Section 3, where The MIRC-X and MYSTIC data were recorded in the
we also mention other existing astrometric observations low spectral resolution mode (R = λ/∆λ = 50 prism)
for Atlas from the literature, including lunar occultation and reduced using the standard pipeline for MIRC-
measurements, and the Hipparcos observations. In Sec- X/MYSTIC14 (version 1.3.5 and 1.4.0; Anugu et al.
tion 4 we explain our approach to separating the spec- 2020). The pipeline produces squared visibilities and clo-
tra of the individual components with the technique of sure phases for each baseline over 8–10 spectral channels.
spectral disentangling, which we then use in the follow- We used an integration time of 150 sec.
ing section to determine the spectroscopic properties of CLIMB combines the light from three telescopes and
both stars (effective temperature, surface gravity, pro- operates in the near-infrared H and K bands. The
jected rotational velocity, metallicity, etc.). The radial CLIMB data were originally published by Gordon et
velocities (RVs) we infer from our spectra are described al. (2019). We re-reduced the CLIMB data using the
and reported in Section 6. The details of our orbital pipeline developed by J. D. Monnier, with the general
analysis combining the velocities and the astrometry are method described by Monnier et al. (2011) extended to
presented in Section 7, along with the results for Atlas. three beams (e.g., Kluska et al. 2018), producing broad-
The physical properties of the two components are then band squared visibilities for each baseline and closure
discussed and compared with current models of stellar phases for each closed triangle.
evolution in Section 8, and in Section 9 we draw our PAVO is a visible light combiner that operates in the R
conclusions. band (0.65–0.79 µm). The data on Atlas were collected
2. SPECTROSCOPIC OBSERVATIONS in two-telescope mode, and reduced with the standard
Atlas was observed spectroscopically at the Center for PAVO reduction pipeline15 , producing squared visibili-
Astrophysics (CfA) beginning in October of 2017, using ties over ∼23 spectral channels.
The calibrated data sets from PAVO, CLIMB, MIRC-
the Tillinghast Reflector Échelle Spectrograph (TRES; X, and MYSTIC will be made available in the OIFITS
Fűrész 2008; Szentgyorgyi & Fűrész 2007) on the 1.5 m format through the Jean-Marie Mariotti Center (JMMC)
Tillinghast reflector at the Fred L. Whipple Obser- Optical Interferometry Database16 (OIDB).
vatory on Mount Hopkins (Arizona, USA). This is a
bench-mounted, fiber-fed instrument providing a resolv- 3.2. Other Astrometric Observations
ing power of R ≈ 44,000, with a CCD detector record-
ing 51 échelle orders between 3800–9100 Å. We collected 14 [Link]
15
48 spectra through March of 2022, with signal-to-noise [Link]
16 [Link]
ratios ranging from 190 to 860 per resolution element of
 Atlas 3

Table 1
Log of Observations at the CHARA Array

UT Date Instrument λ Telescopes N Calibrators

2012Nov15 PAVO R S2E2 1 HD 23753
2014Sep22 CLIMB H S1E1W1 1 HD 23338
2014Sep23 CLIMB H S1E1W1 2 HD 23338, HD 23753
2014Sep24 CLIMB H S1E1W1 2 HD 23338, HD 23753
2015Sep03 PAVO R E2W2 3 HD 21050, HD 25175, HD 22860
2015Sep05 PAVO R S1W1 3 HD 21050, HD 25175
2015Sep06 PAVO R E2S1 4 HD 21050, HD 25175, HD 24368
2015Sep08 PAVO R W2E1 2 HD 21050, HD 25175
2015Sep09 PAVO R S2W2 1 HD 22860
2015Sep10 PAVO R S2W2 1 HD 21050, HD 22860
2015Sep14 CLIMB H S1E1W1 3 HD 23338, HD 23753
2015Nov08 PAVO R W2S2 3 HD 21050, HD 25175
2017Nov19a MIRC-X H E1W2W1S2S1E2 2
2017Nov20b MIRC-X H E1W2W1S2S1E2 1 HD 32630
2018Dec10 MIRC-X H E1W2W1S2S1E2 1 HD 16730, HD 23232, HD 24398, HD 23183
2022Nov15c MIRC-X H E1W2W1E2 1 HD 218235, HD 20150, HD 23288, HD 27627, HD 27808, HD 28406, HD 36667
2022Nov15c MYSTIC K E1W2W1E2 1 HD 218235, HD 20150, HD 23288, HD 27627, HD 27808, HD 28406, HD 36667
Note. — Column N is the number of data sets recorded on Atlas on each night.
a The observing log for UT 2017 Nov 19 indicates that Atlas was observed as a calibrator. Atlas was observed with all 6 telescopes, while
the next nearest calibrators were more than 60◦ away, observed more than 4 hours later, and obtained fringes on only 3 or 4 telescopes.
The Atlas data were not calibrated, so only the closure phases were used in the binary fit.
b The observing log for UT 2017 Nov 20 indicates that Atlas was observed as a calibrator. The closest 6-telescope calibrator was located
24◦ away and observed 4 hours later using a different pop configuration. This resulted in poor calibration of the visibilities (V 2 > 1 on
several baselines). Only the closure phases were used in the binary fit.
c On UT 2022 Nov 15, the S1 and S2 telescopes were offline because of a problem with the metrology signal on the S1 delay line cart,
and a mechanical problem with the drive bearings on the S2 telescope.
servation.17 Zwahlen et al. (2004) reported further in-
Table 2 terferometric observations from the Mark III instrument
Adopted Calibrator Uniform Disk Diameters
between 1989 and 1992, and additional ones gathered
Calibrator λ UDD σUDD Calibrator λ UDD σUDD
with the Navy Prototype Optical Interferometer (NPOI,
(mas) (mas) (mas) (mas) Flagstaff, Arizona) in 1991–2000. These were provided
in polar coordinates (ρ, θ), with their corresponding er-
HD 16730 H 0.656 0.015 HD 24368 R 0.216 0.006
ror ellipses. We incorporated all of these measurements
HD 20150 H 0.350 0.013 HD 24398b H 0.636 0.026
HD 20150 K 0.350 0.013 HD 25175 R 0.193 0.005 into our analysis as published.
HD 21050 R 0.187 0.005 HD 27627 H 0.273 0.006 Lunar occultation detections of the companion to At-
HD 22860a R 0.141 0.008 HD 27627 K 0.273 0.006 las have been made by a number of authors between 1968
HD 23183 H 0.829 0.064 HD 27808 H 0.275 0.007 and 1991. While such measurements only yield the pro-
HD 23232 H 0.653 0.016 HD 27808 K 0.275 0.007
HD 23288 H 0.228 0.007 HD 28406 H 0.277 0.007 jection of the true separation of the binary along the
HD 23288 K 0.228 0.007 HD 28406 K 0.277 0.007 direction of the Moon’s motion (ρocc ), they can still pro-
HD 23324 R 0.198 0.006 HD 32630 H 0.388 0.040 vide constraints that can be used to strengthen the orbit,
HD 23338 H 0.320 0.030 HD 36667 H 0.284 0.007
HD 23441 R 0.159 0.005 HD 36667 K 0.284 0.007
as we do here. The usefulness of these observations was
HD 23753 R 0.222 0.006 HD 218235 H 0.390 0.009 also mentioned by Pan et al. (2004), although they only
HD 23753 H 0.226 0.006 HD 218235 K 0.390 0.009 employed them as a consistency check on their interfero-
Note. — Except where noted, the calibrator uniform disk di- metrically derived orbit. A complete list of these obser-
ameters (UDD) were adopted from the JMMC Stellar Diameters vations is given in Table 3, to correct several misprints
Catalogue (Bourgés et al. 2014). or mistakes found in the tabulation of Pan et al. (2004).
a We adopted an angular diameter for HD 22860 based on the
The Hipparcos catalog (ESA 1997) reported the first
Gaia parallax of 5.82 ± 0.11 mas and an estimated radius of 2.60 ± reasonably accurate orbital solution for Atlas (source
0.13 R⊙ from Kervella et al. (2019).
b We adopted the limb-darkened angular diameter of HD 24398 identifier HIP 17847), with a period of 290.7 ± 8.6 d.18
from Schaefer et al. (2016), and converted it to a uniform disk It was based on the motion of the center of light of the
diameter in the H band using a limb-darkening coefficient of 0.1792 binary on the plane of the sky, as the system was not
(Claret & Bloemen 2011), based on an effective temperature of spatially resolved. It also assumed that the orbit is cir-
21,950 K and log g = 3.061 (Huang & Gies 2008).
cular, due to a lack of constraints. The Hipparcos in-
termediate astrometric data (“abscissa residuals”) used
for that determination are publicly available, and have
typical individual uncertainties for Atlas of 1–2 mas that
Additional long-baseline interferometric observations make them useful in our analysis for constraining the
of Atlas were reported by Pan et al. (2004). They
were collected between 1989 and 1992 with the Mark III 17 The published signs of both coordinates turn out to be re-
stellar interferometer (Mount Wilson, California), and versed. This comes from the fact that the interferometric squared
in 1996–1999 with the Palomar Testbed Interferometer visibilities have an inherent 180◦ ambiguity in the position angles.
18 An earlier claim of a spectroscopic orbit by Abt et al. (1965)
(PTI, Mount Palomar, California). They were presented
in Cartesian coordinates (∆α, ∆δ) for the epoch of ob- had a period of 1255 d, but proved to be erroneous.
4 Torres et al.

Table 3
Lunar Occultation Measurements for Atlas

JD Year θocc ρocc Phase Source

(2,400,000+) (◦ ) (mas)
40221.5734 1968.998 236* 6.2 ± 0.4 0.3962 1,2
41314.6075* 1971.991 128.7 2.2 ± 0.5 0.1523 3
41396.3724 1972.215 273.7 2.5 ± 0.4 0.4333 4
41396.3844 1972.215 124.0 7.4 ± 0.4 0.4333 4
41724.7154 1973.114 228.6* 4.0 ± 0.4 0.5616 2
47106.5562 1987.848 21.09 6.8 ± 0.2 0.0556 5
47106.5562 1987.848 21.09 6.1 ± 0.4 0.0556 5
48309.2844 1991.141 297.45 6.2 ± 0.3 0.1887 5
48336.5221* 1991.216 264 13.1 ± 0.4 0.2823 6
Note. — JD epochs marked with an asterisk indicate observa-
tions flagged as poor by the authors, which we have not used. We
note, however, that the last one may simply suffer from a misprint:
a vector separation of ρocc = 3.1 mas, instead of 13.1 mas, would
fit our final model almost perfectly. Vector angles θocc marked
with an asterisk have been adjusted here by 180◦ . All angles are
for the equinox of the date of observation. Separation vectors
ρocc with no reported uncertainties are assigned an initial error
of 0.4 mas. Orbital phases are based on the ephemeris in Table 6.
Source codes in the last column are: (1) Evans (1971); (2) Mc-
Graw et al. (1974); (3) de Vegt & Gehlich (1976); (4) Bartholdi
(1975); (5) Meyer et al. (1995); (6) Hill & Osborn (1996).
Figure 1. Top: Two of our observed spectra of Atlas (black)
astrometric orbit. We will therefore make use of these at opposite quadratures (maximum velocity separation). Bottom:
Disentangled spectra of the binary components in the spectral re-
observations below in Section 7. gion 4430–4530 Å, centered on the He I 4471 Å and Mg II 4481 Å
We note, finally, that the existence of a wider compan- lines. The secondary component is shifted upward by 0.04, for clar-
ion to Atlas (ρ ≤ 0.′′ 8), roughly 3 mag fainter than the ity.
combined light of the close pair, has been claimed since
at least 1827 (as reported by Struve 1837). The Wash-
of SPD (Simon & Sturm 1994; Hadrava 1995), the or-
ington Double Star Catalog (WDS; Worley & Douglass
bital elements of the binary system and the components’
1997; Mason et al. 2001) contains several claimed detec-
spectra are calculated simultaneously, in a self-consistent
tions through 1929, and many more instances when it was
manner. The RVs are by-passed, and the orbital param-
not detected. Many of the most prominent visual binary
eters are optimized directly. Here, however, the orbital
observers of the 19th and 20th centuries never saw it dou-
motion is also strongly constrained by the astrometric
ble. Similar attempts at detection since the 1970s by the
observations described previously. In order to take ad-
speckle technique, as well as more recently with adap-
vantage of that information, we have therefore chosen
tive optics, have also failed, even though the companion
to use SPD in pure separation mode (e.g., Pavlovski &
should have been a relatively easy target. Examination
Hensberge 2010). In this approach, the orbital elements
of the claimed detections, kindly provided by R. Mat-
are not varied, but are instead held fixed from a separate
son (U.S. Naval Observatory), shows that the position
orbital analysis that in this case combines the astrom-
angles and separations scatter considerably, as indicated
etry with RVs, as we describe later in Section 7. This
also in the notes to the WDS. Although occasional ref-
avoids the inconsistency that would arise from the use of
erences to the existence of this companion, and even the
a different set of spectroscopic orbital elements for the
use of its claimed properties, are still found in the recent
reconstructed spectra than those adopted for the rest of
literature (e.g., Neiner et al. 2015; Gordon et al. 2019),
the paper. Here we fixed the orbital elements to those
at the present time we do not regard the evidence to be
presented below in Section 7.
sufficiently convincing to support its reality.
The disentangling code applied in this work, FDBi-
4. SPECTRAL DISENTANGLING nary (Ilijić et al. 2004), uses the Fast Fourier Transform
and makes the selection of spectral segments for SPD
As is common in double-lined binaries, the spectra of very flexible, also allowing one to closely preserve the
Atlas display a complex pattern of overlapping lines from original resolution of the observed spectra. This is par-
the two components that changes with orbital motion. ticularly important in order to keep as much wavelength
In Atlas, this is exacerbated by the rapid rotation of the space as possible at the edges of the échelle orders (a
stars, particularly the primary, such that the lines of the prerequisite of SPD in the Fourier domain), which is al-
two objects are never completely separated at any orbital ready limited due to heavy line blending given the broad
phase. Two examples at opposite quadratures are shown spectral features of the primary.
in black at the top of Figure 1. The broader features of
We applied SPD over the spectral range 4000–5600 Å,
the primary star dominate, and have the smaller RV am-
plitude of the two, while the secondary can be identified in segments of 50 to 100 Å. For regions containing the
by the superimposed weaker and narrower lines. broad Balmer lines, the segments were 150–200 Å wide.
For this work, we made use of the method of spec- The bottom section of Figure 1 shows a ∼100 Å portion
tral disentangling (SPD) to isolate the individual spectra of the disentangled spectra centered on the He I 4471 Å
of the components for further analysis. In typical uses and Mg II 4481 Å lines, which are very prominent in the
 Atlas 5

primary. A significant difference is seen in the projected The latter result is unexpected for stars born in the
rotational velocities of the two stars, as well as in their same binary system from a common natal cloud, prompt-
relative flux contributions. ing a more detailed investigation. Upon closer inspection
In eclipsing systems in which the flux ratio between of the spectral fits, we found that the high metallicity
the components changes appreciably over the course of estimate for the secondary was largely driven by a sys-
the orbital cycle, it is possible to infer the proper nor- tematic increase in the depths of iron, chromium, and ti-
malization factors giving the correct continuum flux of tanium lines. Additionally, the secondary’s helium lines
one component relative to the other (see, e.g., Pavlovski appeared significantly weaker than expected for a star
et al. 2009, 2022, 2023). That is not the case for Atlas, with its effective temperature. This is consistent with an
however, so there is an inherent ambiguity in reconstruct- earlier classification of Atlas as a He-weak object (Ren-
ing the component spectra. They are separated correctly, son & Manfroid 2009). We therefore fixed the global
i.e., the shapes and relative strengths of the spectral lines metallicity of the secondary to the solar value, consis-
are as they should be for each star, but their fractional tent with the primary, and re-optimized its atmospheric
light contribution cannot be determined without exter- parameters alongside the individual abundances of He,
nal information. We discuss this further in the following Fe, Cr, and Ti. The resulting atmospheric parameters
section. and elemental abundances from this analysis are listed
in Table 4 (columns 2 and 3).
5. ATMOSPHERIC PARAMETERS With the above results in hand, we employed the
For the analysis of the disentangled spectra, we em- gssp binary module to analyze the disentangled spec-
ployed the method of spectral synthesis as implemented tra of the two components simultaneously, replacing the
in the gssp software package (Tkachenko 2015). This wavelength-independent light factors with the radius ra-
package is designed for the analysis of spectra of sin- tio between the stars. It is well known that strong abun-
gle stars and binary systems, and consists of three mod- dance anomalies for helium and/or metals, such as those
ules. Two of them were used in this work: gssp single found in the secondary, lead to non-negligible changes
and gssp binary. The former allows for the analysis in the atmospheric structure — which can, in turn, af-
of the disentangled spectra of double-lined binaries, but fect the determination of atmospheric parameters, if ig-
treats each component as if it were an isolated single nored (e.g., Lehmann et al. 2007; Shulyak et al. 2009).
star, and treats light dilution in the disentangled spectra To avoid this bias, we employed a custom grid of model
as a wavelength-independent effect. This implies that the atmospheres for the secondary, computed by fixing the
parallel analysis of the binary components’ disentangled abundances of He, Fe, Cr, and Ti to the values obtained
spectra may result in a total light factor that either ex- from the unconstrained mode analysis.
ceeds or falls short of unity. The gssp binary module, The results are presented in Table 4 (columns 4 and
on the other hand, couples the components’ spectra and 5), and are considered the final set of parameters in this
computes the light contribution of each star per wave- study. While both stars are now found to be slightly hot-
length bin. In this case, the light factors are replaced ter than suggested by the previous analysis, the elemen-
by a single free parameter — the squared radius ratio tal abundances of He, Fe, Cr, and Ti in the atmosphere
between the two stars — as shown by Tkachenko (2015, of the secondary remained largely unchanged. Figure 2
his Eqs. 3 and 4). shows a comparison between the disentangled and best-
Because the gssp single module is generally faster, fit model spectra for both components (top row for the
it is a good choice for obtaining initial estimates of the primary, bottom row for the secondary), focusing on the
atmospheric properties of the binary components, albeit Hγ line profile (left column) and a set of metal lines (right
under the assumption of a wavelength-independent light column).
dilution factor. Once adequate initial guesses are deter- Our findings leave no doubt that the secondary com-
mined for both components, the more computationally ponent of Atlas is a chemically peculiar star. Numerous
demanding but physically more accurate gssp binary spectroscopic studies of such objects in the literature re-
module can be used. The gssp package employs the Syn- port a wide range of surface chemical abundance distri-
thV radiative transfer code (Tsymbal 1996) to compute butions. Those distributions vary from simple to com-
synthetic spectra over arbitrary wavelength ranges and plex and, in some cases, they appear to correlate with
for arbitrary surface compositions. To generate theoret- magnetic field topologies (e.g., Kochukhov et al. 2014,
ical spectra, SynthV relies on atmosphere models — in 2017; Alecian 2015). Observationally, such distributions
our case, a pre-computed grid of LLmodels (Shulyak et manifest as temporal variability in the line profiles of
al. 2004) as published by Tkachenko et al. (2012). the affected elements. In fact, upon close inspection of
Our initial analysis of the disentangled spectra using the average spectral lines of the secondary, we detected
the gssp single module revealed the following key fea- changes in the shapes of line profiles with time that are
tures: (i) although the two stars have similar effective unrelated to the orbital motion of the star. We describe
temperatures, their log g values differ significantly, indi- and illustrate this in more detail in the next section. On
cating that the primary component is more evolved; (ii) the other hand, the method of spectral disentangling em-
the primary displays substantially broader spectral lines ployed in Section 4 assumes the absence of variability in
than the secondary, suggesting a large difference in rota- the observed composite spectra, beyond that caused by
tional velocities — unless the secondary is seen at a low orbital motion. As a result, the 1σ uncertainties listed
inclination relative to the line of sight; (iii) a striking mis- for the atmospheric parameters of both stars in Table 4
match in metallicity is found between the components, should be regarded as purely statistical errors.
with the secondary exhibiting [M/H] ≈ +0.7 dex, while A sanity check on the spectroscopic parameters for At-
the primary shows a nearly solar composition. las may be obtained by comparing the predicted mag-
6 Torres et al.

1.0 1.000
0.9 0.995
Normalized Flux

0.8 0.990
0.985
0.7 Primary Primary
0.980
Hγ Metal lines
0.6 0.975
0.5 0.970
4300 4320 4340 4360 4380 4400 5300 5350 5400 5450 5500 5550
1.00 1.000
0.98 0.995
Normalized Flux

0.96
0.990
0.94
0.985
0.92 Secondary Secondary
Hγ 0.980 Metal lines
0.90
0.975
4300 4320 4340 4360 4380 4400 5380 5390 5400 5410 5420 5430 5440
Wavelength (Å) Wavelength (Å)
Figure 2. Comparison between the disentangled spectra (solid black line) and best-fit model (dashed red line) for the primary and
secondary components of the Atlas system. Two different wavelength regions are shown, featuring Hγ and a section rich in metal lines.
Note the changing vertical scales in the four panels.

Table 4
Spectroscopic Properties of Atlas

Unconstrained Mode Constrained Mode

Parameter Primary Secondary Primary Secondary
Teff (K) 12100 ± 200 12150 ± 450 12525 ± 200 12835 ± 450
log g (dex) 3.40 ± 0.05 4.15 ± 0.11 3.38 ± 0.07 4.20 ± 0.15
ξ (km s−1 ) 0.0+1.2 1.0 ± 1.5 0.5+0.9
−0.5 1.0+1.5
−1.0
v sin i (km s−1 ) 212 ± 10 48 ± 5 217 ± 9 47 ± 7
[M/H] −0.02 ± 0.06 0.0 (fixed) −0.03 ± 0.06 0.0 (fixed)
Light factor 0.83 ± 0.03 0.15 ± 0.03 ··· ···
R1 /R2 ··· 2.33 ± 0.07
[He/Ntot ] ··· −1.98 ± 0.07 ··· −2.00 ± 0.09
[Fe/Ntot ] ··· −3.53 ± 0.08 ··· −3.53 ± 0.08
[Ti/Ntot ] ··· −6.25 ± 0.40 ··· −6.35 ± 0.40
[Cr/Ntot ] ··· −5.43 ± 0.25 ··· −5.45 ± 0.40
Note. — ξ represents the microturbulent velocity. The element abundances in the constrained mode
above, expressed relative to the Sun, are as follows: He (−0.90 dex), Fe (+1.05 dex), Ti (+0.80 dex),
and Cr (+0.95 dex). Results from the constrained mode are adopted for the remainder of this work.

nitude difference ∆m between the components against servations by Pan et al. (2004), our CHARA observa-
independent measurements from astrometry. To predict tions reported in Section 7 at three wavelengths from the
∆m as a function of wavelength, we used synthetic spec- PAVO, CLIMB, MIRC-X, and MYSTIC beam combin-
tra based on Kurucz model atmospheres (Castelli & Ku- ers, and a value in the Hipparcos bandpass, also reported
rucz 2003) computed for solar metallicity and the compo- later. The model predictions are compared against these
nents’ derived properties (Teff , log g), scaling the fluxes measurements in Figure 3. The agreement is generally
by the derived radius ratio. Empirical measurements of very good, with the exception of a few of the early lu-
the magnitude difference are available from the lunar oc- nar occultation measurements, which have large formal
cultation observations of Atlas, the interferometric ob- uncertainties. The two that deviate the most come from
 Atlas 7

1.01

1.00

Normalized flux
0.99

0.98

0.97

0.96

400 300 200 100 0 100 200 300 400

RV (km s 1)
Figure 4. An observed composite LSD profile (black solid line)
fitted with a superposition of two model LSD profiles (red solid
line). The black and blue dashed lines show the respective model
LSD profiles of the primary and secondary components. Residu-
Figure 3. Measured magnitude differences between the compo- als, obtained by subtracting the composite model profile from the
nents of Atlas, together with the prediction from our spectroscopic observations, are shown with a black dotted line at the top, shifted
analysis in Section 5. The shaded area around the curve is the 1σ upward by 0.01, for clarity.
uncertainty range obtained by propagating the errors in the stellar
temperatures, surface gravities, and radius ratio. The horizontal
error bars on the measurements represent the width of the band-
Table 5
pass.
CfA Radial Velocities of Atlas
observations that were reported as problematic in the
BJD RV1 RV2 S/N Phase
original publications (poor seeing, etc.). We view the (2,400,000+) (km s−1 ) (km s−1 )
good agreement as broadly supportive of the accuracy of
our spectroscopic determinations. 58050.8325 33.4 ± 3.9 −20.6 ± 1.6 686 0.6531
58079.8815 23.8 ± 3.0 −12.5 ± 1.7 484 0.7529
6. RADIAL VELOCITIES 58109.8188 10.6 ± 4.6 7.1 ± 1.4 542 0.8558
58127.5616 −3.6 ± 3.5 22.7 ± 1.7 641 0.9168
Inferring RVs with high precision benefits from data 58142.6446 −16.4 ± 5.1 40.5 ± 1.7 463 0.9686
with high signal-to-noise ratios (S/N). Since all spectral Note. — S/N represents the signal-to-noise ratio per res-
lines in binary star spectra are affected by the Doppler ef- olution element. Orbital phases are based on the ephemeris
fect in the same way, it is advantageous to apply spectral of Table 6. (This table is available in its entirety in machine-
readable form.)
line averaging techniques to maximally enhance the S/N
of the resulting mean profile used for the RV analysis. In
this work, we employed the method of Least-Squares De- Each LSD profile from the time series, computed in this
convolution (LSD, Donati et al. 1997), as implemented way, was analyzed to infer the RVs of both components.
by Tkachenko et al. (2013). In practice, the LSD method A separate grid of synthetic LSD profiles was computed
performs a “deconvolution” of all spectral lines that are for each star based on the values of Teff , log g, [M/H],
common between the observed spectrum of the target and ξ as listed in Table 4. The grids span a range of pro-
and a pre-computed line list, and then computes a mean jected rotational velocity of v sin i = [180, 260, 2] km s−1
profile from all these “deconvolved” spectral features. for the primary and v sin i = [25, 75, 1] km s−1 for the
There are two key limitations to consider when pre- secondary. Each observed LSD profile was then fitted
computing the line list: (i) all spectral lines should have with a superposition of two synthetic LSD profiles —
similar shapes, which means that hydrogen and helium one for each binary component — with a total of six ad-
lines should be excluded due to their pressure-dominated justable parameters: their positions (represented by the
broadening mechanisms; (ii) the weakest lines are also RVs), widths (represented by v sin i), and depths (rep-
best excluded to avoid introducing significant noise into resented by line-depth scaling factors). Our fitting pro-
the resulting mean profile. cedure is conceptually similar to fitting a superposition
Given that the components of the Atlas system have of two Gaussians with variable positions, widths, and
similar spectral types, and that the primary contributes amplitudes, but different in that our model is based on
a larger share of the total light, we prepared a single line detailed radiative transfer computations rather than an
list by extracting relevant information from the Vienna arbitrary analytical function. An example of a fit to one
Atomic Line Database (VALD, Piskunov et al. 1995; of the LSD profiles is shown in Figure 4. RVs derived in
Kupka et al. 1999), using the atmospheric parameters this way for both binary components are listed in Table 5.
of the primary derived in Section 5 and listed in Table 4 As mentioned in the previous section, the anomalous
(column 4). In our selection, we discarded all spectral abundances of He, Fe, Cr, and Ti found in the secondary
lines with predicted line strengths weaker than 0.03 (i.e., of Atlas manifest themselves in the spectrum through
3% depth in continuum units). This threshold is dictated distortions in the line profiles that change with time. To
by the high rotational velocity of the primary component, visualize these distortions, we subtracted the best-fit av-
which causes weaker lines to be buried in the noise, con- erage profiles of the primary — obtained in the previous
tributing no useful signal to the calculation of the LSD step — from the observed composite LSD profiles of the
profiles. system at each epoch. The resulting average profiles of
8 Torres et al.

which represent the velocity semiamplitudes and center-

of-mass velocity.
Although the Hipparcos mission did not spatially re-
solve the Atlas binary, it did detect the astrometric wob-
ble due to the companion, and was able to follow the
path of the center of light of the system, with the sim-
plifying assumption that the orbit is circular. That path
(once proper motion and the parallactic motion are re-
moved) is simply a scaled-down version of the relative
orbit, reflected around the center of mass, but otherwise
with the same shape and orientation. Here we have in-
corporated the one-dimensional intermediate astrometric
data from the satellite (abscissa residuals) directly into
our modeling of the system, which allows us to remove
the assumption of circularity, but introduces several ad-
ditional free parameters. One is the angular semimajor
axis of the photocenter (a′′phot ). Another five new param-
eters are corrections to the catalog values of the position
and proper motion components, and parallax (∆α∗ , ∆δ,
∆µ∗α , ∆µδ , ∆πHip ), where the notation ∆α∗ = ∆α cos δ
and ∆µ∗α = ∆µα cos δ. For details of the formalism for
incorporating these observations into an orbital solution,
see, e.g., Torres (2007), and references therein.
The observations for Atlas included in our analysis are
the 19 pairs of (∆α∗ , ∆δ) observations from Pan et al.
(2004), the 12 pairs of (ρ, θ) measurements by Zwahlen
Figure 5. LSD profiles of the secondary component, shifted to et al. (2004), 7 lunar occultation measurements, our 48
zero velocity, obtained after subtracting the primary’s contribu- pairs of RVs, and 36 Hipparcos abscissa residuals. Not all
tion from the composite average profiles. The red dots are the
measurements, and the solid blue lines represent spline fits to the
of the interferometric measurements from CHARA allow
profiles. The spectra are numbered by date, in the same order as for the determination of relative positions on a nightly
they appear in Table 5. basis, due to insufficient uv coverage in some cases. For
this reason, we incorporated all the (squared) visibilities
the secondary are shown in Figure 5. The changes in from PAVO, CLIMB, MIRC-X, and MYSTIC directly
shape are clearly visible to the eye, and are suggestive of into our analysis, along with closure phases where avail-
rotational modulation caused by surface inhomogeneities able. For MIRC-X and MYSTIC, the recorded wave-
on the secondary star. Inverting these variations to in- lengths were adjusted by dividing by the recommended
fer detailed surface abundance maps is beyond the scope correction factors 1.0054 ± 0.0006 and 1.0067 ± 0.0007,
of this study, and would be challenging with the sparse respectively, following Gardner et al. (2022) and Gard-
sampling we have. Nevertheless, it is reasonable to ex- ner (2022, priv. comm.). Bandwidth smearing was ac-
pect that the distortions will increase the scatter in the counted for following Gardner et al. (2021).
velocities of the secondary (see the next section), and it is The use of these data requires us to solve for the
therefore likely that the formal RV uncertainties for that flux ratio F2 /F1 in the three different CHARA wave-
star reported in Table 5 are somewhat underestimated. length bands (PAVO, H, and K), as well as for the an-
gular diameter of the primary (ϕ1 ), which is resolved
7. ORBITAL ANALYSIS in these observations. Gordon et al. (2019) measured
7.1. Procedure and Treatment of Errors ϕ1 = 0.464 ± 0.043 mas using CLIMB. The smaller sec-
ondary is unresolved, so we held its angular diameter
A joint orbital analysis of the RVs and the astrom- fixed at a value of ϕ2 = 0.20 mas, based on its estimated
etry for Atlas was carried out within a Markov chain radius and the distance (see below). To reduce the num-
Monte Carlo framework, using the emcee package19 of ber of free parameters, the primary diameter that we
Foreman-Mackey et al. (2013). For consistency, all po- solved for is a limb-darkened value (ϕ1,LD ) common to all
sition angles were precessed to the year J2000.0, where three bandpasses, using the appropriate visibility func-
needed. The orbital elements describing the relative mo- tion for a linear limb-darkened stellar disk model from
tion on the plane of the sky are the period (P ), the an- Hanbury Brown et al. (1974). Limb-darkening coeffi-
gular semimajor axis (a′′ ), the eccentricity (e) and argu- cients were taken from the tabulation of Claret & Bloe-
ment of periastron for the secondary (ω2 ), the orbital in- men (2011), for the properties of the stars as determined
clination angle (iorb ), the position angle of the ascending
earlier.20
node for J2000.0 (Ω), and a reference time of periastron
The line profile distortions that are obvious in the sec-
passage (Tperi ). For numerical efficiency, we expressed
√ ondary of Atlas can potentially introduce a bias in the ve-
the eccentricity
√ and argument of periastron as e cos ω2 locities, such that the RV zeropoints for the primary and
and e sin ω2 , and the inclination angle as cos iorb . The
three spectroscopic orbital elements are K1 , K2 , and γ, 20 The limb-darkening coefficients in the PAVO, H, and K bands
are 0.312, 0.164, and 0.138 respectively for the primary, and 0.295,
19 [Link] 0.149, and 0.126 for the secondary.
 Atlas 9

secondary may not be the same. To prevent this from af-

fecting the solution, we introduced an additional free pa-
rameter (∆RV ) to represent a possible difference between
the zeropoints. Additionally, as the formal errors for the
observations may not always be accurate, and to ensure
proper weighting of the different datasets, we solved for
additional parameters in the form of multiplicative scale
factors for the uncertainties, separately for each type of
observation: two for the RVs (fRV1 , fRV2 ), two others
(fP , fZ ) for the interferometric measurements of Pan et
al. (2004) and Zwahlen et al. (2004), respectively, one
more for the lunar occultation measurements (focc ), and
another for the Hipparcos abscissa residuals (fHip ). Vis-
ibility calibration uncertainties from the MIRC-X and
MYSTIC beam combiners have been estimated to be at
the level of about 5%, and the formal V 2 errors already
incorporate this contribution. For PAVO, they are es-
timated to be roughly 3% (J. Jones, priv. comm.), but Figure 6. Limb-darkened diameter of the primary of Atlas as a
function of the baseline orientation (PA). The points represent in-
are not factored into the formal errors. Initial checks dependent diameter estimates over separate PA intervals, with the
for PAVO and CLIMB suggested uncertainties somewhat horizontal error bars indicating the range in each set. The measure-
larger than those produced by the reduction pipeline for ment in parentheses was based on a small number of visibilities,
these instruments. To account for this, we solved for a and is very uncertain. Formal diameter error bars are typically
smaller than the point size, and are not shown. The curve is not a
contribution σV 2 added in quadrature to the internal er- fit to these measurements, but is instead a fit to all CHARA vis-
rors. For the closure phases, we solved for similar jitter ibilities and closure phases with an elliptical model for the stellar
terms σCP for CLIMB, MIRC-X, and MYSTIC. In all, disk (see the text). The inset shows the projection of the stellar
shape on the plane of the sky, along with the measurements (du-
we solved for 34 adjustable parameters. plicated, given that interferometric visibilities are invariant under
Our MCMC analysis used 100 random walkers with a 180◦ rotation around the center).
20,000 links each, after burn-in. We checked for conver-
gence by examining the chains visually, and by requiring In addition to having the highest spatial resolution
a Gelman-Rubin statistic of 1.05 or smaller (Gelman & among the CHARA measurements of Atlas, the PAVO
Rubin 1992). The priors for all parameters were uniform data were taken over different and relatively small ranges
over suitable ranges, except for those of the error scaling of baseline orientations on each night. While this limited
factors, which were log-uniform. uv coverage prevents us from measuring nightly relative
We point out that we did not enforce equality in our positions, it makes these data ideally suited for exploring
analysis between the final Hipparcos parallax (i.e., the possible differences in ϕ1 . An MCMC solution was car-
catalog value plus ∆πHip ) and the orbital parallax, be- ried out in which we used all other non-CHARA observa-
cause of the known systematic error in the Hipparcos tions in order to constrain the orbit, and solved for a sep-
determinations for the Pleiades (Section 1). By allow- arate primary diameter from the PAVO measurements
ing the Hipparcos parallax to be different, the goal was on each of the eight nights in which that instrument was
to avoid biasing the constraint that the abscissa residu- used. A suggestive trend was seen, with the diameters
als contribute to the orbital motion, which is our main reaching a minimum near the middle of the range of po-
reason for incorporating those measurements. sition angles (i.e., baseline orientations). To improve the
sampling, we then merged together the visibilities from
7.2. Model for the Primary Disk all of the beam combiners, split them into seven 10–30◦
After an initial fit as described above, additional evi- intervals of baseline orientation, and repeated the exer-
dence prompted us to consider a model for the brightness cise. The resulting limb-darkened diameters as a function
distribution of the primary star on the plane of the sky of the mean position angle of the baseline are shown in
that is different from a circular model assumed initially. Figure 6. They support the results from PAVO alone,
There were two motivations for this. The first was that and lead us to conclude that the non-circular shape of
earlier fits to some of the CHARA data alone, intended the primary of Atlas is indeed detectable.
to solve for the relative position, flux ratio, and primary To first order, a star distorted by rapid rotation will
angular diameter on nights with sufficient uv coverage have a projected shape on the plane of the sky that is
(MIRC-X, MYSTIC), displayed a larger scatter in the similar to an ellipse. We therefore changed our initial
ϕ1 values than we expected. Secondly, our spectroscop- model of a circular limb-darkened disk for the primary
ically derived v sin i value of 217 km s−1 is large enough to one featuring a limb-darkened elliptical disk. For con-
that one might expect the star to be rotationally dis- venience, we used the polar form of the equation for an
torted, to a degree that may even be detectable with our ellipse to express the diameter measured along a given
interferometric measurements. This has previously been orientation as
achieved in other similarly rapidly rotating stars such as ϕmin
α Aql (Altair), α Oph, α Cep, and others (see, e.g., van ϕ1 (PA) = q , (1)
Belle 2012, and references therein). The effect would pro- 1 − e2ϕ cos2 (PA − θϕ )
vide an explanation for the scatter mentioned above, as
cross-sections of the apparent disk at different baseline where ϕmin represents the minor axis of the ellipse, θϕ
orientations would lead to slightly different values for ϕ1 . is the orientation angle of the major axis relative to the
10 Torres et al.

north, and PA is the position angle of the projected base-

line (east of north). Thepeccentricity of the ellipse, eϕ ,
has the expression eϕ = 1 − (ϕmin /ϕmaj )2 in terms of
the major and minor axes.
For our final MCMC solution, we solved for the three
parameters of the apparent elliptical disk together with
the rest of the elements mentioned earlier, incorporating
all CHARA visibilities and closure phases directly. We
obtained ϕmin = 0.4523±0.0026 mas, eϕ = 0.534±0.012,
and θϕ = −9.◦ 9 ± 1.◦ 5. The corresponding axial ratio is
ϕmin /ϕmaj = 0.8456±0.0076. An illustration of the shape
and size of the ellipse on the plane of the sky is shown in
the inset in Figure 6.
The above model is only an approximation to the true
stellar shape, and does not account for effects such as
gravity darkening. This may not be negligible for At-
las, and it introduces an asymmetry. Because of this,
the formal errors reported above may not reflect the true
uncertainties of the primary’s disk parameters. Never-
theless, the results serve to show for the first time that
the primary in the Atlas system has a detectable oblate-
ness caused by its rapid rotation. More realistic models
where all physical effects are properly considered are cer-
tainly possible, and have been applied to other fast rota- Figure 7. RV measurements for Atlas with our model. The
tors (see, e.g., van Belle et al. 2001; Domiciano de Souza center-of-mass velocity is indicated with the dotted line. Resid-
et al. 2003; Aufdenberg et al. 2006; Monnier et al. 2007; uals are shown at the bottom.
Zhao et al. 2009). Such an analysis for Atlas is beyond
the scope of the present work, and would benefit from
additional interferometric observations.

7.3. Results
Our orbital elements for Atlas from our final MCMC
solution that uses the elliptical model for the shape of
the primary are presented in Table 6, along with derived
properties for the Atlas system, including the compo-
nent masses and the orbital parallax.21 The RVs and our
model for the spectroscopic orbit are shown in Figure 7,
and a representation of the astrometric observations and
visual orbit model may be seen in Figure 8.
As anticipated, we find a systematic velocity zeropoint
offset of ∆RV = 2.15 ± 0.53 km s−1 between the primary
and secondary, which may be related to the peculiarities
in the secondary. We also find that this star’s RV un-
certainties require an inflation factor of fRV2 ≈ 1.7 in
order to achieve a reduced χ2 value near unity, which is
likely caused by the line profile variations illustrated pre-
viously. On the other hand, the formal RV errors for the
primary reported in Table 5 appear to be overestimated
by about a factor of two.
Despite the approximation of a circular orbit for Atlas Figure 8. Interferometric observations of Atlas from Pan et al.
used in the Hipparcos catalog, the semimajor axis of the (2004) (shown in red) and Zwahlen et al. (2004) (blue). Error
photocenter reported there (4.23±0.97 mas) is consistent ellipses represent the uncertainty on each axis, and short line seg-
ments connect the measurement with the predicted position on the
with our more precise value in Table 6. Similarly for the orbit. The one-dimensional lunar occultation measurements can-
Hipparcos inclination angle (108◦ ± 25◦ ). As expected, not be shown here, and are represented with triangles at their pre-
our revised Hipparcos parallax for Atlas (πHIP = 8.08 ± dicted location from the model, merely to illustrate their phase cov-
erage. Similarly with the CHARA visibilities and closure phases,
0.46 mas) is about 0.7 mas larger than the more precise and with the one-dimensional Hipparcos observations. The pre-
orbital parallax, reflecting the known overestimate in the dicted locations of the Hipparcos measurements are shown here on
satellite results for the Pleiades. the relative orbit rather than on the scaled-down photocenter or-
bit, as the phase coverage is of course the same. Collectively, the
21 While our value for Ω follows the usual convention, and repre- phase coverage of the astrometric observations is near complete.
sents the node where the secondary is receding from the observer, The dotted line indicates the line of nodes, with the ascending
the angle reported by Pan et al. (2004) is flipped by 180◦ (see foot- node marked as “Ω”. The square labeled “P” indicates periastron,
note 17). The angles ω and Ω by Zwahlen et al. (2004) both need coincidentally very near the descending node.
to be changed by 180◦ to be in the right quadrant.
 Atlas 11

8. DISCUSSION
Table 6
Orbital Parameters for Atlas 8.1. Properties of the Primary of Atlas
Our discovery that the apparent disk of the primary of
Parameter Value Prior
Atlas is reasonably well represented by an ellipse, rather
P (day) 290.9919 ± 0.0028 [250, 300] than a circle, enables other properties of the rapidly ro-
a′′ 12.9896 ± 0.0036
√ (mas) [5, 20] tating star to be inferred, with proper consideration of
√e cos ω2 +0.44243 ± 0.00032 [−1, 1]
e sin ω2 −0.19974 ± 0.00063 [−1, 1]
projection effects. While the measured apparent major
cos iorb −0.30675 ± 0.00054 [−1, 1] axis of the ellipse, ϕmaj , coincides with the equatorial
Ω (deg) 334.202 ± 0.025 [0, 360] diameter of the star, the true polar diameter can only
Tperi (BJD) 50585.988 ± 0.096 [50550, 50600] be determined if we know the inclination angle i of the
K1 (km s−1 ) 27.09 ± 0.40 [15, 50] rotation axis relative to the line of sight. The apparent
K2 (km s−1 ) 37.63 ± 0.53 [15, 50]
γ (km s−1 ) +8.41 ± 0.32 [0, 15]
minor axis as measured, ϕmin , will generally be larger
∆RV (km s−1 ) +2.15 ± 0.53 [−20, 20] than the true polar diameter ϕpol due to projection, and
a′′
phot (mas) 3.43 ± 0.40 [0, 20] for an oblate ellipsoidal object, the two are related by
∆α∗ (mas) +0.03 ± 0.41 [−50, 50] q
∆δ (mas) −0.85 ± 0.28 [−50, 50] ϕmin = ϕ2pol sin2 i + ϕ2maj cos2 i . (2)
∆µ∗α (mas yr−1 ) +0.44 ± 0.44 [−50, 50]
∆µδ (mas yr−1 ) +0.07 ± 0.35 [−50, 50] Then, for a given inclination angle, the linear polar ra-
∆πHip (mas) −0.49 ± 0.46 [−50, 50]
(F2 /F1 )PAVO 0.1509 ± 0.0051 [0.01, 1]
dius of the star, Rpol , follows directly from ϕpol and the
(F2 /F1 )H 0.17984 ± 0.00064 [0.01, 1] known distance. The inclination angle may be estimated
(F2 /F1 )K 0.18308 ± 0.00054 [0.01, 1] by using our measured projected rotational velocity of
ϕmin (mas) 0.4523 ± 0.0026 [0.1, 1.5] the primary, as follows. Under the assumption of hydro-
eϕ 0.534 ± 0.012 [0, 1]
θϕ (deg) −9.9 ± 1.5 [−90, 90]
static equilibrium, uniform rotation, and a point mass
gravitational potential, the equatorial rotational velocity
Error Adjustment Parameters of a star is given by
fRV1 0.518 ± 0.054 [−5, 5] s  
fRV2 1.67 ± 0.18 [−5, 5] 2GM Rpol
fP 0.679 ± 0.082 [−5, 5] veq = 1− (3)
fZ 0.86 ± 0.14 [−5, 5] Rpol Req
focc 1.90 ± 0.60 [−5, 5]
fHip 0.83 ± 0.11 [−5, 5] (e.g., Jones et al. 2015), in which G is the gravitational
σV 2 for PAVO 0.0469 ± 0.0022 [0, 1] constant, M is the stellar mass, and Rpol and Req are the
σV 2 for CLIMB 0.0778 ± 0.0088 [0, 1] polar and equatorial radii, respectively. Multiplying the
σCP for CLIMB (deg) 6.6 ± 2.8 [0, 30]
σCP for MIRC-X (deg) 1.882 ± 0.045 [0, 30]
expression above by sin i, and equating the right-hand
σCP for MYSTIC (deg) 0.495 ± 0.053 [0, 30] side to our measured v sin i value of 217 km s−1 , allows
us to solve Eq.[2] and Eq.[3] jointly for the inclination an-
Derived Properties gle and Rpol . We obtained i = 64◦ ± 20◦ (or 116◦ ± 20◦ )
iorb (deg) 107.863 ± 0.032 ··· and Rpol = 6.48 ± 0.50 R⊙ . Within its admittedly large
e 0.23565 ± 0.00011 ··· uncertainty, the larger of the two values of i is consis-
ω2 (deg) 335.697 ± 0.082 ···
a (au) 1.768 ± 0.018 ··· tent with being the same as the orbital inclination angle
M1 (M⊙ ) 5.04 ± 0.17 ··· (iorb = 107.◦ 9). Furthermore, the nominal orientation of
M2 (M⊙ ) 3.64 ± 0.12 ··· the primary’s sky-projected disk (θϕ = −9.◦ 9, or 350.◦ 1)
q ≡ M2 /M1 0.721 ± 0.014 ··· is also rather similar to the position angle of the line of
πorb (mas) 7.340 ± 0.076 ···
Distance (pc) 136.2 ± 1.4 ···
nodes for the orbit (Ω = 334.◦ 2), although we note that θϕ
πHip (mas) 8.08 ± 0.46 ··· has an inherent 180◦ ambiguity that we cannot resolve,
(F2 /F1 )Hip 0.179 ± 0.044 ··· stemming from the cosine squared term in Eq.[1]. Nev-
µ∗α (mas yr−1 ) +18.21 ± 0.44 ··· ertheless, with the appropriate choices for both i and θϕ ,
µδ (mas yr−1 ) −44.63 ± 0.35 ··· we conclude that the orbital and spin axes may well be
ϕmaj (mas) 0.5340 ± 0.0030 ···
ϕmin /ϕmaj 0.8456 ± 0.0076 ··· close to alignment. Indeed, proceeding with those choices
and with the formal uncertainties for the primary’s disk
Note. — The values listed correspond to the mode of the properties, and keeping in mind the caveats mentioned
posterior distributions, with uncertainties representing the 68.3%
credible intervals. Priors in square brackets are uniform over the earlier, we obtain a true relative angle between those axes
ranges specified, except those for the error inflation factors f , of ψ = 21◦ ± 12◦ .22
which are log-uniform. The time of periastron passage, Tperi , is With the equatorial radius Req = 7.81 ± 0.18 R⊙
referenced to BJD 2,400,000.
(from ϕmaj and the distance), the true oblateness is then
Rpol /Req = 0.828 ± 0.057. This is a slightly smaller
With our measurement of the semimajor axis of number (greater rotational flattening) than the appar-
the photocenter from the Hipparcos observations, it is ent oblateness of 0.846 reported above in Table 6. The
straightforward to compute the magnitude difference be- estimate of i implies a rotational velocity at the equa-
tween the components in the Hipparcos bandpass (∆Hp). tor of veq = 233 ± 45 km s−1 . Given the simple nature of
It follows from the relation a′′phot = a′′ (B − β), in which 22 This estimate of the true relative inclination follows from the
B = M2 /(M1 +M2 ) is the mass fraction of the secondary, expression cos ψ = cos iorb cos i + sin iorb sin i cos(Ω − θϕ ). We re-
and β is its fractional light contribution, expressed also as port the median and standard deviation from a Monte Carlo exer-
β = 1/(1+100.4∆Hp ). We obtain ∆Hp = 1.87±0.27 mag. cise.
12 Torres et al.

our toy model for the primary, the uncertainties reported

above have been conservatively increased by a factor of
two over their formal values, which we consider to be too
optimistic.

8.2. The Distance

Trigonometric parallax measurements for Atlas have
been reported in the original and revised editions of the
Hipparcos catalog, as well as by the Gaia mission, in both
its second and third data releases (DR2 and DR3). How-
ever, neither of the Gaia values accounted for the orbital
motion of the binary. Our orbital parallax therefore pro-
vides the most accurate measure of the distance to date.
As a check, additional estimates may be obtained by the
classical moving cluster method, which assumes a com-
mon space motion for all members of the Pleiades. These
“kinematic” distances rely only on the known space mo-
tion of the cluster, the sky position of the convergent
point and of Atlas itself, and the measured proper mo-
tion of the object. Figure 9 shows several of these es- Figure 9. Kinematic determinations of the distance to Atlas
timates, which we calculated from proper motion mea- (moving cluster method), based on its proper motion from various
sources, as labeled on the x axis. Also indicated are the few direct
surements compiled from the literature, selected to be as determinations from the trigonometric parallax, as well as our or-
independent as possible. We adopted the space motion bital parallax result. The dashed line represents the mean cluster
and convergent point coordinates for the cluster based on distance from Gaia DR2 (Gaia Collaboration et al. 2018a), after a
Gaia DR2 (Gaia Collaboration et al. 2018b). Also shown slight zeropoint adjustment following Lindegren et al. (2018), giv-
ing 135.3 pc. Source numbers on the x axis correspond to: (1)
in the figure are the distances inferred from the available Nikoloff et al. (1982); (2) Fricke et al. (1988); (3) ESA (1997); (4)
direct trigonometric parallax measurements (Hipparcos, Høg et al. (2000); (5) van Leeuwen (2007); (6) Vondrák & Štefka
Gaia), which are all too small.23 On the other hand, the (2007); (7) Röser et al. (2008); (8) Frouard et al. (2015); (9) Gaia
kinematic distances are all seen to be quite close to the Collaboration et al. (2018b); (10) Gaia Collaboration et al. (2023);
(11) Brandt (2021); (12) This work: the distance from the proper
mean value for the Pleiades (135.3 pc; Gaia DR2). The motion and from the direct parallax measurement both rely on our
distance inferred from the orbital parallax in the present reanalysis of the Hipparcos intermediate data (Section 7).
work, 136.2 ± 1.4 pc, is only 0.7% larger. Given that the
position of Atlas on the sky is only about half a degree Here we have used the Modules for Experiments in
from the cluster center, we conclude that its location in Stellar Astrophysics code (MESA; Paxton et al. 2019,
space must also be very near the center of the Pleiades. and references therein) to generate suitable evolutionary
This is not unexpected given the mass segregation known tracks. This provides us with greater flexibility for tuning
to exist in the Pleiades (e.g., Raboud & Mermilliod 1998; various parameters of interest, particularly the initial ro-
Converse & Stahler 2008; Alfonso et al. 2024), as Atlas tation rate (see also Gossage et al. 2019)24 . Specifically,
is one of the heavier members of the cluster. the version of MESA we used here is release r22.11.1
(2022). We adopted the solar metallicity, which is essen-
8.3. Comparison with Stellar Evolution Models tially the composition we derived spectroscopically (see
The physical properties of Atlas derived in this work Table 4). Recent versions of MESA that follow Paxton et
allow for a meaningful comparison against current mod- al. (2019) utilize a new method of accounting for centrifu-
els of stellar evolution. However, standard models are gal distortion in stellar structure. The prior implemen-
inadequate for this case, because of the large rotational tation followed the method of Endal & Sofia (1976), with
velocity of the primary. Centrifugal forces in rapid ro- numerically stable calculations valid for rotation rates up
tators cause significant departures from spherical sym- to roughly ω/ωcrit = 0.6. Here, ω is the angular rotation
metry, and give rise to gravity darkening. The struc- frequency, and ωcrit is the angular frequency at which
ture, global properties, and evolution of stars change sig- the centrifugal force would match gravity at the stellar
nificantly, in ways that standard, spherically symmetric equator (“critical” rotation). The new implementation
models cannot account for (see, e.g., Meynet & Maeder of centrifugal distortion improves on these calculations,
2000; Maeder & Meynet 2010). While a small number with validity up to ω/ωcrit ≈ 0.9. This new implementa-
of grids of rotating models have been published, they tion derives from analytical fits to the Roche potential of
typically offer only a few fixed values for the mass and a single star as it becomes distorted via centrifugal forces
rotation rate, and many other parameters are also held (Paxton et al. 2019). When comparing evolutionary cal-
fixed. culations, it is important to bear in mind that other mod-
els such as PARSEC (Nguyen et al. 2022), mentioned
23 As a demonstration of the bias for Gaia DR3, we carried out below, do not use the recent implementation brought by
numerical simulations following Perryman et al. (2014), and ver- Paxton et al. (2019) based on the Roche potential, and
ified that not accounting for orbital motion causes a systematic instead use an implementation of the method of Endal
error in the derived trigonometric parallax of Atlas (source identi-
fier 66526127137440128) toward larger values, i.e., toward shorter & Sofia (1976). Overall, stellar behavior near critical
distances, just as observed, by about the amount we see in this
star. 24 See [Link]
 Atlas 13

rotation rates remains an unresolved matter in need of empirical estimates for slightly less massive stars in
observational constraint. eclipsing binaries, as reported by Claret & Torres (2019).
In addition to the individual masses, several other em- The mixing length parameter was set to αML = 1.82, as
pirical properties of both stars in Atlas are available for in the MIST calculations.
the comparison with models. They include the spectro- Extensive tests indicated that, within the uncertain-
scopic effective temperatures and surface gravities, as ties, a model for a star spinning at an initial rate of about
well as the absolute radii, R1 = 7.56 ± 0.63 R⊙ and 55% of the breakup value (ω0 /ωcrit ≈ 0.55) reaches a sat-
R2 = 3.25 ± 0.29 R⊙ . The radius of the rotationally dis- isfactory agreement with the spectroscopically measured
torted primary star, calculated here from its mass and Teff and our two estimates of log g, for the nominal pri-
the spectroscopic log g, represents a mean value inher- mary mass of 5.04 M⊙ (see Figure 10, top left panel). For
ited from the nature of log g, which is itself an average comparison, we show also an evolutionary track from the
over the visible disk of the star. The radius for the sec- PARSEC v2.0 series (Nguyen et al. 2022), for a mass of
ondary follows from the value for the primary and the 5.00 M⊙ and ω0 /ωcrit = 0.60, which are the nearest avail-
spectroscopic radius ratio (Table 4). able values to those adopted for our MESA calculations.
Another mean value of R1 that is independent of log g, As seen in the figure, the PARSEC model is some 500 K
and is more directly connected to the interferometry and hotter than the one from MESA, and therefore does not
our model for the shape of the primary, may be esti- agree as well with the measurements for Atlas.
mated from the polar and equatorial radii reported pre- In Figure 10, the evolutionary tracks suggest that the
viously. These rely on a different spectroscopic prop- primary of Atlas is at or near the so-called blue hook,
erty (v sin i) and on the distance. With the same as- referring to the brief blueward “hook” exhibited by the
sumption as before that the star is reasonably well rep- models. This evolutionary phase typically follows core
resented by an oblate ellipsoid, the volume-equivalent ra- hydrogen exhaustion, and is due to a relatively brief halt
2
dius can be calculated as R1,vol = (Req Rpol )1/3 . We ob- in nuclear burning and subsequent contraction of the
tain R1,vol = 7.32±0.26 R⊙ . The corresponding volume- star, until hydrogen reignites in a shell around the in-
equivalent surface gravity is log g1,vol = 3.414 ± 0.025. ert helium core. Our MESA model displays some slight
Both of these estimates are consistent with those more wiggles in its evolutionary track around this time that are
closely related to the spectroscopy through log g. Fur- associated with this contraction. As the model contracts,
ther constraints inferred from the distorted shape of the its rotation rate increases and the model approaches a
primary are its true oblateness, and the expected surface critical rotation rate at this point in the evolution. Prior
rotational velocity at the equator (see Section 8.1). to this contraction, the point at which the model most
In the presence of rapid rotation and the resulting grav- closely matches the temperature of the primary of Atlas
ity darkening, observed global properties such as the tem- corresponds to a rapid rotation rate of ω/ωcrit ≈ 0.77.
perature or luminosity become dependent on the inclina- Consequently, theory predicts that the primary of Atlas
tion angle of the spin axis relative to the line of sight. may have a significantly distorted stellar structure under
For example, a star viewed close to pole-on (i ≈ 0◦ ) the effects of rapid rotation, as we actually observe.
would present a hotter disk-averaged temperature and a The radius vs. temperature plot is shown in the top
higher luminosity than one viewed at higher inclination middle panel, where the match is similar to the previous
angles, because the polar temperature is hotter. MESA panel, as expected given that R1 depends on log g. The
and other codes typically report directional averages of age of Atlas, as inferred from the MESA model, is be-
those properties over the surface of the star. In the case tween 102 and 106 Myr (top right panel), which is not far
of the primary of Atlas, our knowledge of the inclination from other age estimates for the Pleiades using a variety
angle of its rotation axis allows us to infer “projected” of methods: 112 Myr (Dahm 2015), 104–117 Myr (Naylor
properties from the models, by applying adjustments to 2009), 118 Myr (Frasca et al. 2025), 130 Myr (Barrado
the MESA predictions that depend on i and the rate of y Navascués et al. 2004), and 110–160 Myr (Gossage et
rotation, ω/ωcrit . These projected properties will then al. 2018), among many others.
more closely correspond to those we actually measured The MESA and PARSEC models provide predictions
for the star. Here we adopted corrections based on the for the change in the equatorial rotational velocity (veq )
gravity-darkening model of Espinosa Lara & Rieutord as a function of age. Within the uncertainties, the expec-
(2011), as implemented in the GDit25 code written by tation from MESA is consistent with our estimate of that
Aaron Dotter. quantity described in Section 8.1 (veq ∼ 233 km s−1 ),
Comparisons of the observations for the more evolved while the PARSEC model underestimates it. The mid-
primary star were made against evolutionary tracks from dle panel at the bottom of Figure 10 displays our oblate-
MESA, adjusted as described above, for a wide range ness estimate for the primary, along with the evolution of
of initial angular rotation rates relative to the breakup this quantity expected from both models. Again, there
rate, ω0 /ωcrit . The prescription we used for convective is good consistency between theory and the observation
core overshooting is the diffusive approximation (Frey- for MESA, assuming the star is near the point of central
tag et al. 1996; Herwig et al. 1997), with an overshoot- hydrogen exhaustion, as suggested by the other compar-
ing parameter fov = 0.016, identical to that adopted in isons. On the other hand, the PARSEC model would
other sets of MESA-based models, such as the MESA predict the star to have a more spherical shape than we
Isochrones and Stellar Tracks series (MIST; Choi et al. measure. In order to match our estimates of veq and the
2016). This fov value is also consistent with semi- oblateness, the PARSEC model would require a much
higher initial rotation rate of ω0 /ωcrit ∼ 0.80, but we
find that such a model would overestimate the radius,
25 [Link] and underestimate log g.
14 Torres et al.

Figure 10. Comparison of the measured properties of the primary of Atlas against theory. The blue line corresponds to our MESA
model for the nominal mass of M1 = 5.04 M⊙ determined in this work, Solar metallicity was assumed. The initial rotation rate was
set to ω0 /ωcrit = 0.55, and the convective overshooting parameter was fixed at fov = 0.016. The dotted magenta line is a model from
the PARSEC v2.0 series, with properties similar to those above. Both evolutionary tracks have been adjusted for the effects of gravity
darkening, as described in the text. Arrows indicate the direction of evolution. The measured temperature is represented by the shaded
area, and R1 , log g, veq , and the oblateness are shown with their corresponding error boxes. The two values shown for the radius and log g
(‘spectroscopic’ and ‘interferometric’) were derived in different ways (see the text), but are consistent.

Convective core overshooting and rotation have some- It is quite possible that some of this disagreement
what similar effects on the models, in the sense that they has to do with the chemical anomalies in the secondary.
both favor mixing of fresh hydrogen fuel into the core. While we did account for the underabundance of He and
This typically leads to an extension of the evolutionary the overabundance of Fe, Cr, and Ti in the calculation of
tracks toward cooler temperatures and higher luminosi- model atmospheres for the star, potential anomalies in
ties, and results in longer main-sequence lifetimes. One the abundances of other elements were not investigated
may therefore expect it might be possible for the models here. Moreover, chemically peculiar stars are also known
to match the observations with a range of different com- to exhibit vertical stratification of elements in their atmo-
binations of the strength of overshooting (fov ) and initial spheres, which influences the profiles of physical quanti-
rotation (ω0 /ωcrit ). However, for Atlas we find that our ties such as temperature and pressure — ultimately lead-
estimates of the current shape and equatorial rotation ing to further perturbations in the atmospheric structure
of the primary tend to lift that degeneracy. They con- (e.g., Shulyak et al. 2009; Pandey et al. 2011; Makaganiuk
strain the initial rotation to be near the value we report, et al. 2012). These perturbations were not taken into ac-
given the measured mass and effective temperature of the count in our spectroscopic analysis. Additionally, the
star. Consequently, overshooting strengths much differ- method of spectral disentangling used in this study as-
ent from what we assumed do not improve the fit to the sumes that the line profiles do not change shape as a
measurements. function of time — an assumption that is violated in the
Finally, in the bottom right panel of Figure 10 we in- case of the secondary. Given these issues, we cannot rule
clude the predictions from MESA for the difference be- out potential biases in the atmospheric properties for the
tween the polar and equatorial temperature caused by secondary that are not reflected in the purely statistical
gravity darkening, which our observations do not con- errors reported in this work.
strain. It is expected to be roughly 650 K at the current Because the secondary component is relatively un-
evolutionary state of the star. evolved, the age is not well constrained by the observa-
Figure 11 presents an analogous comparison of the ob- tions (i.e., the tracks are nearly vertical in the bottom left
servations for the secondary of Atlas against MESA mod- panel of Figure 11), and our estimate is highly sensitive
els, for the nominal mass of M2 = 3.64 M⊙ . Two differ- to the temperature. With these models, the secondary
ent initial rotation rates are shown (ω0 /ωcrit = 0.10 and appears slightly older than the primary. This may be
0.20), although their impact on the surface gravity and related as well to the anomalous composition of the star,
radius as a function of effective temperature is minimal. which is not taken into account in the MESA models.
We find only marginal agreement with the model in the A somewhat better match to log g can be achieved by
log g vs. Teff plane, but good agreement for the radius. adopting a model mass 1σ lower than the nominal value
 Atlas 15

(M2 = 3.52 M⊙ ; see Figure 11, dotted line). This also field strength on different nights, they concluded that
yields better agreement with the age inferred for the pri- the magnetic axis of the secondary is not aligned with
mary. its axis of rotation, which is not uncommon. Abundance
The rotational velocity comparison is shown in the top anomalies in these magnetic, early-type stars have also
right panel of Figure 11. As the inclination angle of the been associated with the presence of chemical patches
secondary’s spin axis is unknown (but see below), here or “spots” on the surface (e.g., David-Uraz et al. 2019),
we have compared the predicted true equatorial velocity which naturally lead to line profile variations as the star
from theory against our measured projected equatorial rotates or the spots change. We observe such variations
velocity (v sin i = 47 km s−1 ). There is formal agreement very clearly in the secondary.
with the model for ω0 /ωcrit = 0.10, within the uncer- Several authors have established that Atlas is photo-
tainty, but any inclination angle different from 90◦ would metrically variable (e.g., McNamara 1985, 1987; Wraight
convert the empirical measurement to a higher equato- et al. 2012; Zwintz et al. 2024). The detailed study by
rial value, and imply a true initial angular rotation rate White et al. (2017) identified several frequencies corre-
larger than this. sponding to periods of ∼2.5 days or less, some of which
Finally, the lower right panel of Figure 11 displays the are the same as seen by others. The frequency with the
primary and secondary of Atlas together in the H-R di- largest amplitude (2.1 mmag) corresponds to a period of
agram, to illustrate their relative states of evolution. 2.428 d. White et al. (2017) concluded that these fre-
quencies, if they all originate on the same star, are most
9. FINAL REMARKS likely due to pulsations, rather than rotation, and argued
With the present analysis, Atlas now ranks among the that they probably come from the primary, which is more
better characterized binary systems in the Pleiades clus- than five times brighter than the secondary. Neverthe-
ter, both in terms of its improved (3D) orbit and the less, they cautioned that further investigation is required
physical properties of the individual components. Our for confirmation. However, if the 2.428 d periodicity were
spectroscopic and interferometric observations have re- in fact due to rotation, it could not be caused by the pri-
vealed two salient features of the stars in the system. mary, as our measured radius and v sin i for that star
One is that the primary is a rotationally distorted ob- would imply veq < v sin i. On the other hand, it could
ject, for which we have been able to measure the oblate- well come from the secondary, and in that case, the in-
ness (Rpol /Req ≈ 0.83) directly from the interferometric ferred inclination angle of its spin axis would be 44◦ ±10◦ .
observations. We have also obtained a rough estimate of The distance to Atlas is now known to 1%, and the
the orientation of its spin axis, which is possibly aligned masses of the components to about 3%, currently lim-
with the axis of the orbit. Only about a dozen rapidly ro- ited by the spectroscopy. Other relevant properties de-
tating, early-type stars have had their distended shapes termined here for both stars include the effective tem-
determined in this way, as instrumental limitations of peratures, radii, log g, and the projected rotational ve-
the interferometric technique typically require them to be locities. Atlas is one of only four binaries in the Pleiades
bright and nearby. Atlas is by far the most distant exam- cluster that have dynamical mass determinations to date,
ple, all others being closer than 50 pc. Additional inter- and is the most massive. The other three are HII 1431
ferometric observations to support a more sophisticated (HD 23642; Munari et al. 2004; Southworth et al. 2005;
analysis that accounts for gravity darkening and other Groenewegen et al. 2007; David et al. 2016; Southworth
effects should be able to improve upon our estimates. A et al. 2023), HCG 76 (V612 Tau; David et al. 2016), and
second result of interest is that while the chemical com- HII 2147 (Torres et al. 2020). The measured properties
position of the primary appears to be essentially solar, of the primary of Atlas are found to be in good agreement
consistent with the known metallicity of the cluster, the with stellar evolution models from MESA that account
secondary is a helium-weak star with significant enhance- for the rapid rotation of the star, and suggest the initial
ment of Fe, Cr, and Ti, and perhaps other elements. The rotation rate on the zero-age main sequence was about
latter three are about 10 times more abundant than in 55% of the breakup velocity. The current rotation rate is
the Sun, on average, while He is about 8 times weaker. estimated to be ω/ωcrit ≈ 77%, according to these mod-
Atlas had previously been classified as a He-weak ob- els. The comparison with theory places the star at the
ject in the catalog of Renson & Manfroid (2009), but very end of the main sequence, prior to core hydrogen
until now it had not been established which of the two exhaustion (the beginning of the blue hook), at an age
components has the anomaly, or whether both do. Our between 102 and 106 Myr that is similar to other esti-
spectroscopic analysis has now revealed the secondary mates for the cluster. The secondary component rotates
to be the culprit. Together with its other abundance more slowly, and the predictions from MESA models are
abnormalities, which are found with similar patterns in also generally consistent with its other properties at its
about 10% of all A- and B-type main-sequence stars, this measured mass.
places it in the “ApBp” class of chemically peculiar ob-
jects. Stars in this class are often found to also have
strong magnetic fields (e.g., Smith 1996; Briquet et al. The spectroscopic observations of Atlas were gathered
2007; Petit & Oksala 2025), which are considered to be of with the assistance of P. Berlind, M. Calkins, G. Es-
fossil origin, i.e., descended from the fields present in the querdo, and A. Medina. J. Mink (CfA) is thanked for
natal molecular clouds. In fact, Neiner et al. (2015) used maintaining the database of échelle observations at the
the Zeeman signature to determine that the secondary CfA. We are also grateful to R. Matson (USNO) for pro-
of Atlas has a dipolar magnetic field with a strength of viding a record of the astrometric observations of Atlas
several hundred Gauss. The primary, on the other hand, from the Washington Double Star Catalog, maintained
shows no indication of having one. From changes in the at the U.S. Naval Observatory (USNO), and to J. Jones
16 Torres et al.

Figure 11. Similar to Figure 10, now comparing the spectroscopic properties for the secondary of Atlas against theory, at its nominal
mass of M2 = 3.64 M⊙ . In this case, MESA models are shown for two different rotation rates, as labeled. An additional model for
M2 = 3.52 M⊙ (1σ lower than the nominal mass; dotted magenta line) provides a marginally better match to the measured log g. The
panel on the lower right illustrates the relative evolutionary states of the two components of Atlas. Stellar luminosities for this illustration
were calculated from the effective temperatures and radii. The error bars are tilted because of the correlation between L and Teff .
(Georgia State Univ.) for information about systematic numbers GBMF8477 and GBMF12341. This research
uncertainties for the PAVO beam combiner. S. Chhabra, has made use of the Jean-Marie Mariotti Center (JMMC)
I. Codron, C. L. Davies, J. Ennis, T. Gardner, M. Gutier- Aspro and SearchCal services.
rez, D. Huber, N. Ibrahim, J.-B. Le Bouquin, V. Maestro, This work has also made use of the SIMBAD and
B. R. Setterholm, T. ten Brummelaar, and P. Tuthill con- VizieR databases, operated at the CDS, Strasbourg,
tributed to the CHARA observations of Atlas in other France, and of NASA’s Astrophysics Data System
ways, either as instrument PIs, members of the MIRC- Abstract Service. The computational resources used
X/MYSTIC development and commissioning teams, or for this research include the Smithsonian High Per-
as observers. We thank them all. We also acknowledge formance Cluster (SI/HPC), Smithsonian Institution
helpful comments by the anonymous referee. ([Link]
This work is based upon observations obtained with
the Georgia State University Center for High Angular REFERENCES
Resolution Astronomy Array at Mount Wilson Observa-
tory. The CHARA Array is supported by the National Abt, H. A., Barnes, R. C., Biggs, E. S., et al. 1965, ApJ, 142, 1604
Science Foundation under Grant No. AST-1636624, Alecian, G. 2015, MNRAS, 454, 3, 3143
AST-2034336, and AST-2407956. Institutional support Alfonso, J., Vieira, K., & Garcı́a-Varela, A. 2024, A&A, in press
(arXiv:2410:23527)
has been provided from the GSU College of Arts and Sci- Anugu, N., Le Bouquin, J.-B., Monnier, J. D., et al. 2020, AJ,
ences and the GSU Office of the Vice President for Re- 160, 158
search and Economic Development. Time at the CHARA Aufdenberg, J. P., Mérand, A., Coudé du Foresto, V., et al. 2006,
Array was granted internally and through the NOIRLab ApJ, 645, 664
community access program (NOIRLab PropID: 2018B- Barrado y Navascués, D., Stauffer, J. R., & Jayawardhana, R.
2004, ApJ, 614, 386
0326; PI: C. Melis; NOIRLab PropID: 2022B-235883; Bartholdi, P. 1975, AJ, 80, 445
PI: G. Torres). SK acknowledges funding for MIRC-X re- Bourgés, L., Lafrasse, S., Mella, G, et al. 2014, ASPC, 485, 223
ceived from the European Research Council (ERC) under Brandt, T. D. 2021, ApJS, 254, 42
the European Union’s Horizon 2020 research and inno- Briquet, M., Hubrig, S., De Cat, P., et al. 2007, A&A, 466, 1, 269
Buchhave, L. A., Latham, D. W., Johansen, A., et al. 2012,
vation program (Starting Grant No. 639889 and Con- Nature, 486, 375
solidated Grant No. 101003096). JDM acknowledges Castelli, F. & Kurucz, R. L. 2003, in Modelling of Stellar
funding for the development of MIRC-X (NASA-XRP Atmospheres, IAU Symp. 210, eds. N. Piskunov, W. W. Weiss
NNX16AD43G, NSF-AST 2009489) and MYSTIC (NSF- & D. F. Gray (San Francisco, CA: ASP), p. A20
ATI 1506540, NSF-AST 1909165). SJM was supported Choi, J., Dotter, A., Conroy, C., et al. 2016, ApJ, 823, 102
Claret, A. & Bloemen, S. 2011, A&A, 529, A75
by the Australian Research Council through Future Fel- Claret, A. & Torres, G. 2019, ApJ, 876, 134
lowship FT210100485. SG acknowledges support from Converse, J. M. & Stahler, S. W. 2008, ApJ, 678, 431
the Gordon and Betty Moore Foundation under project Dahm, S. E. 2015, ApJ, 813, 108
 Atlas 17

David, T. J., Conroy, K. E., Hillenbrand, L. A., et al. 2016, AJ, McNamara, B. J. 1985, ApJ, 289, 213
151, 112 McNamara, B. J. 1987, ApJ, 312, 778
David-Uraz, A., Neiner, C., Sikora, J., et al. 2019, MNRAS, 487, Meynet, G. & Maeder, A. 2000, A&A, 361, 101
304 Melis, C., Reid, M. J., Mioduszewski, A. J., et al. 2014, Science,
de Vegt, C., Gehlich, U. K. 1976, A&A, 48, 245 345, 1029
Domiciano de Souza, A., Kervella, P., Jankov, S., et al. 2003, Meyer, C., Rabbia, Y., Froeschle, M., et al. 1995, A&AS, 110, 107
A&A, 407, L47 Meynet, G., Mermilliod, J.-C., & Maeder, A. 1993, A&AS, 98, 477
Donati, J.-F., Semel, M., Carter, B. D., et al. 1997, MNRAS, 291, Miles, R. 1999, JBAA, 109, 106
4, 658 Monnier, J. D., Zhao, M., Pedretti, E., et al. 2007, Science, 317,
Endal, A. S. & Sofia, S. 1976, ApJ, 210, 184 342
ESA 1997, The HIPPARCOS and Tycho Catalogues, Vol. 1200 Monnier, J. D., Zhao, M., Pedretti, E., et al. 2011, ApJ, 742, L1
(Noordwijk: ESA) Munari, U., Dallaporta, S., Siviero, A., et al. 2004, A&A, 418, L31
Evans, D. S. 1971, AJ, 76, 1107 Nather, R. E. & Evans, D. S. 1970, AJ, 75, 575
Espinosa Lara, F. & Rieutord, M. 2011, A&A, 533, A43 Naylor, T. 2009, MNRAS, 399, 432
Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. Neiner, C., Buysschaert, B., Oksala, M. E., et al. 2015, MNRAS,
2013, PASP, 125, 306 454, L56
Frasca, A., Zhang, J. Y., Alonso-Santiago, J. et al. 2025, A&A, Nguyen, C.-T., Costa, G., Girardi, L. et al. 2022, A&A, 665, A126
698, A7 Nikoloff, I., Høg, E., Ayers, C., et al. 1982, Perth 75. A catalogue
Freytag, B., Ludwig, H.-G., & Steffen, M. 1996, A&A, 313, 497 of positions of 2589 FK4 and FK4S stars., by Nikoloff, I.; Høg,
Fricke, W., Schwan, H., Lederle, T., et al. 1988, E.; Ayers, C.; Spencer, C.; Sultana, M.; Verveer, A.; Høyer, P.
Veroeffentlichungen des Astronomischen Rechen-Instituts J.. Bickley (Western Australia): Perth Observatory, 71 p.
Heidelberg, 32, 1 Pan, X., Shao, M., & Kulkarni, S. R. 2004, Nature, 427, 326
Frouard, J., Dorland, B. N., Makarov, V. V., et al. 2015, AJ, 150, Pandey, C. P., Shulyak, D. V., Ryabchikova, T., et al. 2011,
141 MNRAS, 417, 1, 444
Fűrész, G. 2008, PhD thesis, Univ. Szeged, Hungary Pavlovski, K. & Hensberge, H. 2010, Binaries - Key to
Gaia Collaboration, Babusiaux, C., van Leeuwen, F., et al. 2018a, Comprehension of the Universe, ASP Conf. Ser., 435, eds. A.
A&A, 616, A10 Prša & M. Zejda (San Francisco, CA: ASP), 207
Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2018b, Pavlovski, K., Hummel, C. A., Tkachenko, A., et al. 2022, A&A,
A&A, 616, A1 658, A92
Gaia Collaboration, Vallenari, A., Brown, A. G. A., et al. 2023, Pavlovski, K., Southworth, J., Tkachenko, A., et al. 2023, A&A,
A&A, 674, A1 671, A139
Gardner, T., Monnier, J. D., Fekel, F. C., et al. 2021, ApJ, 921, 41 Pavlovski, K., Tamajo, E., Koubský, P., et al. 2009, MNRAS,
Gardner, T., Monnier, J. D., Fekel, F. C., et al. 2022, AJ, 164, 184 400, 791
Gelman, A. & Rubin, D. B. 1992, Statistical Science, 7, 457 Paxton, B., Smolec, R., Schwab, J., et al. 2019, ApJS, 243, 10
Gordon, K. D., Gies, D. R., Schaefer, G. H., et al. 2019, ApJ, 873, Percival, S. M., Salaris, M., & Groenewegen, M. A. T. 2005,
91 A&A, 429, 887
Gossage, S., Conroy, C., Dotter, A., et al. 2018, ApJ, 863, 67 Perryman, M., Hartman, J., Bakos, G. Á., et al. 2014, ApJ, 797,
Gossage, S., Conroy, C., Dotter, A., et al. 2019, ApJ, 887, 199 14
Groenewegen, M. A. T., Decin, L., Salaris, M., et al. 2007, A&A, Petit, V. & Oksala, M. E. 2025, Stellar Magnetism, Encyclopedia
463, 579 of Astrophysics eds. I. Mandel and F. R. N. Schneider
Hadrava, P., 1995, A&AS, 498, 5730 (Amsterdam: Elsevier), (arXiv:2504.00179)
Hanbury Brown, R., Davis, J., Lake, R. J. W., et al. 1974, Piskunov, N. E., Kupka, F., Ryabchikova, T. A., et al. 1995,
MNRAS, 167, 475 A&AS, 112, 525
Herwig, F., Bloecker, T., Schoenberner, D., et al. 1997, A&A, Raboud, D. & Mermilliod, J.-C. 1998, A&A, 329, 101
324, L81 Renson, P. & Manfroid, J. 2009, A&A, 498, 961
Hill, J., & Osborn, W. 1996, International Amateur-Professional Röser, S., Schilbach, E., Schwan, H., et al. 2008, A&A, 488, 401
Photoelectric Photometry Comm., 64, 54 Setterholm, B. R., Monnier, J. D., Le Bouquin, J.-B., et al. 2023,
Høg, E., Fabricius, C., Makarov, V. V., et al. 2000, A&A, 355, L27 JATIS, 9, 025006
Huang, W. & Gies, D. R. 2008, ApJ, 683, 1045 Schaefer, G. H., Gies, D. R., Monnier, J. D., et al. 2010, AJ, 140,
Ilijić, S., Hensberge, H., Pavlovski, K., et al. 2004, 1838
Spectroscopically and Spatially Resolving the Components of Schaefer, G. H., Hummel, C. A., Gies, D. R., et al. 2016, AJ, 152,
the Close Binary Stars, ASP Conf. Ser., 318, eds. R. W. 213
Hilditch, H. Hensberge & K. Pavlovski (San Francisco, CA: Shulyak, D., Tsymbal, V., Ryabchikova, T., et al. 2004, A&A,
ASP), 111 428, 993
Ireland, M. J., Mérand, A., ten Brummelaar, T. A., et al. 2008, Shulyak, D., Ryabchikova, T., Mashonkina, L., et al. 2009, A&A,
Proc. SPIE, 7013, 701324 499, 3, 879
Jones, J., White, R. J., Boyajian, T., et al. 2015, ApJ, 813, 58 Simon, K. P., & Sturm, E., 1994, A&A, 281, 286
Kervella, P., Arenou, F., Mignard, F., & Th’evenin, F. 2019, Smith, K. C. 1996, Ap&SS, 237, 77
A&A, 623, 72 Soderblom, D. R., Nelan, E., Benedict, G. F., et al. 2005, AJ,
Kluska, J., Kraus, S., Davies, C. L., et al. 2018, ApJ, 855, 44 129, 1616
Kochukhov, O., Lüftinger, T., Neiner, C., et al. 2014, A&A, 565, Southworth, J., Maxted, P. F. L., & Smalley, B. 2005, A&A, 429,
A83 645
Kochukhov, O., Silvester, J., Bailey, J. D., et al. 2017, A&A, 605, Southworth, J., Murphy, S. J., & Pavlovski, K. 2023, MNRAS,
A13 520, L53
Kupka, F., Piskunov, N., Ryabchikova, T. A., et al. 1999, A&AS, Stello, D. & Nissen, P. E. 2001, A&A, 374, 105
138, 119 Struve, F. G. W. 1837, Astronomische Nachrichten, 14, 249
Lehmann, H., Tkachenko, A., Fraga, L., et al. 2007, A&A, 471, 3, Szentgyorgyi, A. H., & Fűrész, G. 2007, RMxAC, 28, 129
941 ten Brummelaar, T. A., Gies, D. G., McAlister, H. A., et al. 2016,
Lindegren, L., Hernández, J., Bombrun, A., et al. 2018, A&A, Proc. SPIE, 9907, 03
616, A2 ten Brummelaar, T. A., McAlister, H. A., Ridgway, S. T., et al.
Maeder, A. & Meynet, G. 2010, New Astronomy Reviews, 54, 32 2005, ApJ, 628, 453
Makaganiuk, V., Kochukhov, O., Piskunov, N., et al. 2012, A&A, ten Brummelaar, T. A., Sturmann, J., Ridgway, S. T., et al. 2013,
539, A142 JAI, 2, 1340004
Mason, B. D., Wycoff, G. L., Hartkopf, W. I., et al. 2001, AJ, Tkachenko, A., Lehmann, H., Smalley, B., et al. 2012, MNRAS,
122, 3466 422, 4, 2960
McGraw, J. T., Dunham, D. W., Evans, D. S., et al. 1974, AJ, 79,
1299
18 Torres et al.

Tkachenko, A., Van Reeth, T., Tsymbal, V., et al. 2013, A&A, van Leeuwen, F. 2009b, A&A, 500, 505
560, A37 Vondrák, J. & Štefka, V. 2007, A&A, 463, 783
Tkachenko, A. 2015, A&A, 581, A129 White, T. R., Pope, B. J. S., Antoci, V., et al. 2017, MNRAS,
Torres, G. 2003, Information Bulletin on Variable Stars, 5402, 1 471, 2882
Torres, G. 2007, AJ, 133, 2684 Worley, C. E. & Douglass, G. G. 1997, A&AS, 125, 523
Torres, G., Melis, C., Kraus, A. L., et al. 2020, ApJ, 898, 2 Wraight, K. T., Fossati, L., Netopil, M., et al. 2012, MNRAS,
Tsymbal, V. 1996, M.A.S.S., 108, 198. 420, 757
van Belle, G. T. 2012, A&A Rev., 20, 51 Zhao, M., Monnier, J. D., Pedretti, E., et al. 2009, ApJ, 701, 209
van Belle, G. T., Ciardi, D. R., Thompson, R. R., et al. 2001, Zucker, S., & Mazeh, T. 1994, ApJ, 420, 806
ApJ, 559, 1155 Zwahlen, N., North, P., Debernardi, Y., et al. 2004, A&A, 425,
van Leeuwen, F. 1999, A&A, 341, L71 L45
van Leeuwen, F. 2007, Astrophysics Space Science Library, Vol. Zwintz, K., Pigulski, A., Kuschnig, R., et al. 2024, A&A, 683, A49
350, HIPPARCOS, the New Reduction of the Raw Data
(Berlin: Springer)
van Leeuwen, F. 2009a, A&A, 497, 209