MNRAS 000, 1–12 (2024) Preprint 15 April 2025 Compiled using MNRAS LATEX style file v3.

The role of wind asphericity in dynamical friction

Jesús Carrillo-Santamaría1★ , Diego López-Cámara1 , Fabio De Colle1 , Enrique Moreno Méndez2 ,
F. J. Sánchez-Salcedo3
1 Instituto de Ciencias Nucleares, Universidad Nacional Autónoma de México, A. P. 70-543 04510 CDMX. Mexico
2 Facultad de Ciencias, Universidad Nacional Autónoma de Mexico, A. P. 70-543 04510 CDMX Mexico
3 Instituto de Astronomía, Universidad Nacional Autónoma de Mexico, A. P. 70-264, 04510. CDMX, Mexico
arXiv:2504.08884v1 [astro-ph.HE] 11 Apr 2025

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT
Dynamical friction (DF) may affect the dynamics of stars moving through dense media. This is the case for stars and compact
objects (COs) crossing active galactic nuclei (AGN) discs, stellar clusters, and common envelopes (CE), driving stellar migration.
DF may decelerate the moving stellar object and may also, under certain conditions, produce an acceleration. In this paper,
we study the DF and its effects in the interaction between a star and the ambient gaseous medium through a set of two-
dimensional, hydrodynamical numerical simulations using a wind tunnel configuration. Three different stellar wind configurations
are considered: isotropic, polar, and equatorial. We confirm that the DF can decelerate and accelerate the star and find the critical
value of the normalized velocity (𝑢 𝑐 ) that marks the transition between these regimes, for the three wind profiles. The value of
𝑢 𝑐 for the isotropic wind differs slightly from that obtained in the thin shell approximation; for an aspherical wind, it may either
be larger or smaller. Aspherical winds with small 𝑢 values produce larger accelerations than isotropic winds, while at high 𝑢
values, they lead to greater deceleration than the isotropic case. The timescale for DF to substantially affect the velocity of a
stellar object is calculated. It is shown to be relevant in AGN discs and CEs.
Key words: (stars:) binaries: general – stars: evolution – stars: winds, outflows – stars: jets – hydrodynamics

1 INTRODUCTION enhancement in front of the object and a density reduction behind it.
Interestingly, if the wind velocity is supersonic, the body experiences
The gravitational coupling between a massive object and the sur-
not a DF but a pushing force (negative DF, NDF).
rounding medium can lead to momentum transfer. In the case of
an object moving through a homogeneous gaseous medium, if the Wilkin (1996, hereafter W96) derived an analytical solution for
object is modeled as a point-like mass (perfect accretor) or as an the structure of a thin shell bow shock resulting from the interaction
extended, non-accreting perturber, it always experiences a retarding between a star — emitting an isotropic wind - and moving at constant
force (Ostriker 1999; Sánchez-Salcedo & Brandenburg 1999; Cantó velocity through a uniform-density interstellar medium. Gruzinov
et al. 2011). In fact, both the shocked ambient medium (wake) and et al. (2020, hereafter G20) computed the DF on a star with an
the accreted gas onto an accretor lead to a drag force (also known isotropic stellar wind with velocity 𝑣 𝑤 , moving with velocity 𝑣 𝑎
as dynamical friction, DF). The morphology of the wake can be sig- through a homogeneous medium, using the analytical solution of
nificantly modified by magnetic fields (e.g., Sánchez-Salcedo 2012), W96. They found that the gravitational DF is opposite to its velocity
heating feedback (e.g., Masset & Velasco Romero 2017; Park et al. for 𝑢 ≡ 𝑣 𝑎 /𝑣 𝑤 ≳ 1.71. If 𝑢 ≲ 1.71, the gravitational force pointed,
2017; Li et al. 2020; Toyouchi et al. 2020) or mechanical feedback instead, in the direction of the velocity of the object, that is, NDF.
from outflows (Shima et al. 1986; Inaguchi et al. 1986; Gruzinov They argued that the effect of NDF was negligible for windy stars
et al. 2020). In this paper, we focus on the effect of outflows on the moving through the interstellar medium. However, we note that it may
DF experienced by a moving object. be relevant for black holes (BHs) in dense environments because they
Rephaeli & Salpeter (1980) estimated the DF when a moving can drive strong outflows if they accrete mass at a rate well above
object emits a spherically symmetric wind. They considered the limit the Eddington limit. The solution of G20 has been applied to the
in which the ram pressure is sufficiently strong to strip the gas from evolution of a binary system where the two stellar components have
the outflow. They found the mass outflow results in an increased DF. isotropic winds (Wang & Li 2022) and to study the interactions of
Their work was analytic, considered only the influence of the wind multiple stars with isotropic outflows inside open clusters (Liu et al.
behind the star, and derived a DF that opposed the motion of the star. 2025).
Shima et al. (1986) and Inaguchi et al. (1986) conducted numerical Li et al. (2020, hereafter L20) investigated, through hydrodynam-
simulations and found that the DF with mass loss is reduced compared ical simulations, the effect of outflows on the DF experienced by a
to the scenario without mass loss, as the outflow leads to a density compact object (CO) in a homogeneous medium. For isotropic out-
flows with 0.1 < 𝑢 < 0.5, they found that the strength of the DF was
in good agreement with analytical calculations. They also considered
★ E-mail: jesus.carrillo@correo.nucleares.unam.mx accretion-powered jets. In these models, the mass loss rate represents

© 2024 The Authors

2 J. D. Carrillo-Santamaria et al.
a fraction of the accretion rate. Their simulations demonstrated that
the gravitational DF was reduced compared to models without out-
flows. However, NDF was never achieved for jets aligned with the 𝝆𝒂 𝒄𝒔 𝑴𝒂 𝝆𝒂 𝒄𝒔 𝑴𝒂
𝑹𝟎
object’s motion, even for 𝑢 as small as 0.03. For jets perpendicu- 𝜽𝒘
𝑹𝟎 𝜽𝒘 𝒗𝒘
lar to the velocity, NDF was possible but only in highly powerful 𝒓𝟎
jets (𝑢 ≃ 0.03), and only when the mass within several Bondi radii 𝒓𝟎
𝑴̇𝒘
was taken into account. On the other hand, Toyouchi et al. (2020) Star
investigated the effect of radiation on DF, considering a 104 𝑀⊙ BH Star
embedded in a dusty medium. They found that DF becomes negative 𝒗
𝑴̇𝒘 𝒘
if the medium has a density ≲ 106 cm −3 and 𝑣 𝑎 ≲ 60 km s −1 . Ogata
et al. (2021) studied, through numerical simulations, the accretion z z
onto a black hole accretion disc which may be emitting a wind.
DF is relevant in various astrophysical contexts (see Szölgyén R R
et al. 2022, and references therein). Among the astrophysical ob-
jects potentially influenced by DF are: a) Wolf-Rayet stars, which
come from massive stars (𝑀★,𝑍 𝐴𝑀𝑆 ≳ 20 𝑀⊙ ) that evolve over
timescales of a few million years, and exhibit high mass loss rates Figure 1. Scheme of the system (not to scale) in a polar (left panel) and
(Conti et al. 1983; Anastasopoulou et al. 2024); b) B[e] stars (mas- equatorial (right panel) configuration. The stellar wind is injected at a distance
𝑟0 from the star, with an opening angle 𝜃𝑤 , a mass loss rate 𝑀¤ 𝑤 , and a velocity
sive late B type stars characterized by rapid rotation and mass loss
𝑣𝑤 . The ambient wind has density 𝜌𝑎 , velocity 𝑣𝑎 , and sound speed 𝑐𝑠 . The
rates comparable to those of Wolf-Rayet stars) which, due to their
green line indicates the bow shock. 𝑅0 is the radius where the stellar and
high rotational velocities, emit strong equatorial winds and form a
ambient winds balance out.
circumstellar decretion disc (Martin et al. 2025); c) “immortal stars”,
stars that are embedded in extremely dense environments, where the
mass gained from the accretion compensates for the wind mass loss,
√︁
speed 𝑐 𝑠 = 𝛾𝑃 𝑎 /𝜌 𝑎 , where 𝑃 𝑎 is the ambient pressure. The Mach
resulting in sustained powerful outflows (Dittmann et al. 2021); d) number of the ambient medium is 𝑀𝑎 = 𝑣 𝑎 /𝑐 𝑠 .
“runaway stars”, stellar objects that exhibit anomalously high ve- The stellar wind is launched from the injection radius 𝑟 0 at constant
locities that have abandoned their birth cluster (Bhat et al. 2022); velocity 𝑣 𝑤 and constant density 𝜌 𝑤 . Thus, the stellar wind has
e) young massive clusters (YMCs), which are dense aggregates of constant mass loss 𝑀¤ 𝑤 . The stellar wind can either be isotropic or
young stars (∼100 Myr) formed within galaxies (their mass exceeds aspherical. In the aspherical case, we consider two configurations:
104 𝑀⊙ and may have ambient density of 𝑛 ∼ 104 cm −3 (Portegies polar (see left panel of Figure 1) or equatorial (right panel of the
Zwart et al. 2010)); and f) common envelopes (CE), which are an same figure). The opening angle 𝜃 𝑤 is measured from the 𝑍-axis for
evolutionary phase of close binary systems in which the orbit decays the polar configuration or from the 𝑅-axis for the equatorial case.
to the point where the secondary star enters the envelope of the pri- The interaction between the stellar wind and the ambient gas forms
mary, subsequently spiraling inward toward the primary’s core (e.g., a bow shock located (on the 𝑧 axis) at a distance 𝑅0 from the centre
Chamandy et al. 2020). We investigate whether the DF acceleration of the star. The stellar wind density 𝜌 𝑤 required for the stellar wind
can contribute to these phenomena. ram pressure to balance that of the ambient medium at 𝑅0 is:
In this work, we present a comprehensive study of the DF exerted
on a star with a wind, as it moves through a uniform medium. Our 𝑀¤ 𝑤
𝜌𝑤 = , (1)
study is based on a series of two-dimensional hydrodynamical simu- 4𝜋 (cos 𝜃 1 − cos 𝜃 2 ) 𝑟 02 𝑣 𝑤
lations, in which the effects produced by the DF on the star’s velocity
are analyzed inside a wind tunnel configuration. In Section 2 we where 𝑀¤ 𝑤 = 4𝜋𝑅02 𝜌 𝑎 𝑣 2𝑎 /𝑣 𝑤 (W96), and 𝜃 1 , 𝜃 2 are the polar angles
describe the setup of our simulations. The results are presented in limiting the stellar wind injection region, that is: 𝜃 1,2 = (0, 𝜋/2),
Section 3 and 4. We discuss our findings and conclude in Section 5. (0, 𝜃 𝑤 ), (𝜋/2 − 𝜃 𝑤 , 𝜋/2) for the isotropic, polar, and equatorial
cases, respectively. A smoothing function was applied to ensure a
gradual transition at 𝜃 𝑤 . In terms of 𝜌 𝑎 , 𝑣 𝑤 , 𝑅0 , 𝑟 0 , and 𝑀¤ 𝑤 , Equa-
tion (1) can also be written as 𝜌 𝑤 = (𝑅0 /𝑟 0 ) 2 𝜌 𝑎 𝑢 2 /(cos 𝜃 1 −cos 𝜃 2 ).
2 SETUP AND UNITS For the isotropic case 𝜌 𝑤 = 𝑀¤ 𝑤 /(4𝜋𝑟 02 𝑣 𝑤 ) or, equivalently, 𝜌 𝑤 =
(𝑅0 /𝑟 0 ) 2 𝑢 2 𝜌 𝑎 .
2.1 Numerical setup and models
At injection, the stellar wind pressure is assumed to be two orders
In this work, we aim to determine the amount of DF produced by the of magnitude lower than its ram pressure: 𝑃 𝑤 = 𝜌 𝑤 𝑣 2𝑤 /100. Under
interaction between a stellar wind and the environment over which the these conditions, 𝑃 𝑎 ≲ 𝑃 𝑤 ≪ 𝜌 𝑤 𝑣 2𝑤 . Therefore, in our simulations,
star moves through. For this purpose, we run a set of two-dimensional both 𝑃 𝑤 and 𝑃 𝑎 are negligible.
(2D) simulations using the hydrodynamical code Mezcal (De Colle We run a set of 2D, hydrodynamic (HD) simulations (31 models)
et al. 2012) in cylindrical coordinates. The adaptive-mesh-refinement in which we explore a range of values for the velocity ratio between
code integrates the hydrodynamic equations and is parallelized using the ambient medium and the stellar wind (𝑢 = 𝑣 𝑎 /𝑣 𝑤 ), varying
the “Message Passing Interface” library. from 𝑢 = 0.07 to 𝑢 = 5.0. For the opening angle, we consider
Figure 1 shows the setups used in our simulations. The star moves 𝜃 𝑤 = 30◦ , 45◦ and 60◦ , as well as the isotropic case. In all our
at constant velocity through the medium and emits a stellar wind. simulations, the Mach number 𝑀𝑎 is set to 5, 𝑐 𝑠 = 1 and 𝑅0 = 1.
The coordinate system is set in the co-moving frame of the star (in The total integration time for all models is 𝑡 𝑓 = 200. The parameters
a “wind tunnel” configuration). The star is fixed at the origin of the of each model are given in Table 1.
domain and the ambient medium has a velocity −𝑣 𝑎 ẑ. The ambient The computational domain extends from 𝑅min = 0 to 𝑅max = 25𝑅0
medium, an ideal gas with 𝛾 = 5/3, has density 𝜌 𝑎 , and sound and from 𝑍min = −40𝑅0 to 𝑍max = 10𝑅0 . The ambient wind is

MNRAS 000, 1–12 (2024)

 Wind asphericity effects in DF 3

Table 1. Model parameters of the 31 simulations. The first column indicates Table 2. Normalization factors (primed values indicate cgs units).
the value of 𝑢 = 𝑣𝑎 /𝑣𝑤 , the second column the orientation of the wind:
isotropic (Iso), polar (P) or equatorial (E), and the third column the opening Variable Normalization factor
angle of the wind 𝜃𝑤 . For the isotropic case, we ran simulations with 𝑢 =
0.07, 0.2, 0.6, 0.8, 1.3, 1.7, 2.0, 2.6, 3.0, 4.0, 5.0. Length 𝑅0′
Density 𝜌𝑎′

𝑢 Orientation 𝜃 𝑤 (◦ ) Velocity 𝑐𝑠′ = 𝑣𝑎′ /5

Mass 𝑀★′
0.07 - 5.0 Iso - Pressure 𝜌𝑎 𝑐𝑠2
0.2 P, E 30, 45, 60 Time 𝑐𝑠′ /(𝑅0′ 𝐺 𝜌𝑎 ′ )
0.6 P 45, 60 Mass transfer rate ′ 𝑐′ 𝑢
𝑅0′2 𝜌𝑎 𝑠
2.0 P 45, 60 DF 𝐺 𝑀★′ 𝜌𝑎 ′ 𝑅′
0
2.6 P, E 45, 60 Ostriker DF 𝐺 2 𝑀★′2 𝜌𝑎 ′ /𝑐 2
𝑠
5.0 P, E 30, 45, 60 DF ratio 𝑅0 𝑐𝑠 /(𝐺 𝑀★′ )
′ ′2

3 ISOTROPIC AND ASPHERICAL WIND EVOLUTION

injected at the 𝑍 = 𝑍max boundary while the stellar wind is imposed
Figure 2 shows the temporal evolution of models with isotropic stellar
from the injection radius 𝑟 0 = 0.2𝑅0 (which is centred at the origin
winds for 𝑢 = 5.0 and 0.2. Specifically, we present density maps
of the domain). All other boundaries are set with outflow boundary
and velocity fields at three different times. The first time, 𝑡1 = 1,
conditions. At most six levels of refinement are used with (250 ×
corresponds to an early stage; 𝑡𝑖 = 20 corresponds to an intermediate
500) cells at the coarsest level, along the (𝑅, 𝑍), corresponding to a
stage; and 𝑡 𝑓 = 200 corresponds to the maximum integration time
maximum resolution of 3×10 −3 𝑅0 . The stellar wind injection region
where a stationary state has been reached. For 𝑢 = 5.0, the wake has
is resolved with ∼ 200 cells for the anisotropic cases and ∼ 1000 cells
high-density and slow material (𝜌 ∼ 10, and 𝑣 ∼ 1) compared to
for the isotropic case. Six resolution levels are used since there is just
the ambient (𝜌 𝑎 = 1,𝑣 𝑎 = 5). On the contrary, for 𝑢 = 0.2 the wake
∼ 5% difference in the DF with respect to using higher resolution
has low-density and fast material (𝜌 ∼ 10 −2 , and 𝑣 ∼ 20). At 𝑡 = 𝑡 𝑖 ,
levels. Table A1 presents the DF for different resolutions for the
the model with 𝑢 = 5.0 presents a more filamentary tail along the
isotropic wind case, with 𝑢 = 5.0 once steady state has been reached.
𝑧-axis compared to 𝑢 = 0.2. This occurs because the injected stellar
wind velocity is 25 times higher in the 𝑢 = 0.2 case. As a result,
the stellar material fills the computational domain more rapidly. By
𝑡 = 𝑡 𝑓 , steady state is reached, and the shocked wind forms an
2.2 Dynamical friction extended and nearly cylindrical structure. The interaction between
the stellar wind and the ambient medium forms a bow shock that
The DF will be calculated by (Kim 2010):
encloses a region that contains both shocked ambient material and
∫
𝐺 𝑀★ (𝜌 − 𝜌 𝑎 )𝑧 shocked stellar material. The general structure differs at early and
𝐹𝐷 = 𝑑𝑉 , (2) intermediate times and is basically the same and nearly independent
(𝑅 2 + 𝑧 2 ) 3/2
of 𝑢 at 𝑡 = 𝑡 𝑓 .
where 𝐺 is the gravitational constant, 𝑀★ is the mass of the star, In the thin shell approximation, the position of the shock front as
𝜌(𝑅, 𝑍) is the density of the environment, shaped by the stellar wind, a function of the polar angle is given by (W96):
𝜌 𝑎 is the ambient density, and 𝑑𝑉 = 2𝜋 𝑅 𝑑𝑅 𝑑𝑍 is the volume of √︁
𝑅(𝜃) = 3(1 − 𝜃 cot 𝜃) , (4)
each cell. To calculate the DF at a certain time in our simulations, the
 √︁ 
𝑍 domain is divided into 350 equal-sized sections. For each vertical where 𝜃 = arccos 𝑧/ 𝑅 2 + 𝑧 2 and cot 𝜃 = 𝑧/𝑅. This bow shock
subdivision, the DF is integrated over its total domain 𝑅. We take
solution is overlaid in Figure 2 and in Figure 3. Our simulations quali-
the DF to be positive when its direction is parallel to the direction
tatively reproduce the solution of W96, but show notable differences.
of motion of the star, and negative when it is antiparallel (this is
While W96 assumed a thin shell with efficient cooling, the latter is
opposed to the reference frame proposed by G20).
neglected in our simulations, resulting in a wide shock structure. The
The obtained DF is compared to that given by Ostriker (1999)
𝑂𝑠 ), which assumed that the star had no wind, and was given by: position of the bow shock derived from the ram pressure equilibrium
(𝐹𝐷
between the stellar and ambient winds (𝑅0 = 1) roughly corresponds
"  1# in our simulations to the location of the contact discontinuity. How-
𝑂𝑠 4𝜋𝐺 2 𝑀★2 𝜌 𝑎 1 2 ever, the bow shock itself is located at 𝑍 (𝑅 = 0) ∼ 1.8.
𝐹𝐷 = ln Λ 1 − 2 , (3)
𝑣 2𝑎 𝑀𝑎 Figure 3 shows that the thin shell approximation does not apply to
our models. The simulations reveal four clearly distinct regions: the
where Λ = 𝑏 max /𝑏 min is the Coulomb impact factor ratio with 𝑏 max injected stellar wind, the shocked stellar wind, the shocked ambient
and 𝑏 min being the maximum and minimum impact parameter values, medium, and the ambient medium. The shocked stellar wind and the
respectively. For the extension of our computational domain and shocked ambient material are separated by a contact discontinuity.
injection radius, we have 𝑏 max = 25 𝑅0 and 𝑏 min = 𝑟 0 . This discontinuity is not present in the analytical model of W96. The
We use dimensionless quantities, which can be converted to real density of the ambient medium jumps by a factor ∼ 4 after the main
physical units using the appropriate normalization factors. The nor- shock (as expected in the adiabatic case), and the velocity of the
malization factors of the variables used in the simulations are the unshocked medium and wind change by a factor of ∼ 5. Meanwhile,
ambient density (𝜌 ′𝑎 ) for the density, the bow shock radius (𝑅0′ ) for the shocked stellar wind density depends strongly on the value of
the length, and the sound speed (𝑐 ′𝑠 ) for the velocity; the rest are 𝑢. The normalizations employed imply that 𝜌 𝑤 = 25𝜌 𝑎 𝑢 2 for the
listed in Table 2 and their details are shown in Appendix A. isotropic case (see Section 2.1).For the 𝑢 = 0.2 model, the stellar

MNRAS 000, 1–12 (2024)

4 J. D. Carrillo-Santamaria et al.

Figure 3. Close-up of the launching region for the isotropic wind models.
The 𝑢 = 0.2 case (left panel) and the 𝑢 = 5.0 case (right panel) are shown.
The density, velocity field, and axes are the same as in Figure 1. The white
dotted line represents the analytical bow shock solution from W96. The time
shown is 𝑡 = 𝑡 𝑓 .

stellar wind material may have a higher or lower density than for the
isotropic wind depending on the value of 𝑢. For 𝑢 = 0.2 in the equa-
torial case, the shocked wind has a density an order of magnitude
higher than the isotropic case (𝜌 ∼ 10 −1 ), and, for the polar case, it
has one order of magnitude lower than the isotropic case (𝜌 ∼ 10 −3 );
meanwhile, for 𝑢 = 5.0 the equatorial shocked wind may have a
density that is very similar to that of the isotropical wind (𝜌 ∼ 10),
and the polar shocked wind has a density that is around two orders of
magnitude lower than the isotropic case (𝜌 ∼ 10 −1 ). Independently
of the orientation and the velocity, and as for the isotropic case, the
low density material has high velocities and the high density material
Figure 2. Density maps and velocity fields for isotropic wind models with has low velocities.
𝑢 = 𝑣𝑎 /𝑣𝑤 = 5 (upper panels) and 𝑢 = 0.2 (lower panels). Three different-
time snapshots are shown: the initial time (𝑡1 = 1), an intermediate time
(𝑡𝑖 = 20), and the final time (𝑡 𝑓 = 200), at which a steady state has been 4 DYNAMICAL FRICTION EFFECTS
reached. The black solid lines represent the analytical bow shock solution
from W96. The DF is calculated using Equation (2) and the methodology de-
scribed in Section 2.2. In all figures, the DF is normalized relative to
the DF computed by Ostriker (1999) (𝐹𝐷 𝑂𝑠 , see Equation (3)).
wind is injected with low density and fast material (𝜌 ≃ 1, 𝑣 𝑤 ≈ 25).
Meanwhile, the shocked stellar wind has a lower density with faster
material (𝜌 ∼ 10 −1 , 𝑣 ∼ 12) and the shocked ambient medium has a 4.1 Distance and velocity dependence
density of 𝜌 ∼ 4 and a velocity of 𝑣 ∼ 4. For the model with 𝑢 = 5.0,
Figure 5 shows the 𝐹𝐷 /𝐹𝐷 𝑂𝑠 profiles, integrated from the upper edge
the injected stellar wind is denser and slower (𝜌 ∼ 625, 𝑣 ≲ 1),
the shocked stellar wind has a density of 𝜌 ∼ 500 and a velocity of of the computational box, up to the distance 𝑍 from the star along the
𝑣 ≲ 1, and the shocked ambient medium has a density of 𝜌 ∼ 4 and symmetry axis (note that the star is located at 𝑅 = 𝑍 = 0). The figure
a velocity of 𝑣 ∼ 4. includes isotropic and aspherical winds (polar and equatorial) with
Figure 4 presents a comparison between simulations with aspher- different opening angles and velocity ratios 𝑢 = 5.0 and 𝑢 = 0.2.
ical winds. Specifically, we show density maps and velocity fields We first describe the behavior of the DF in the spherical case. The
for two polar and two equatorial stellar wind models (with 𝜃 𝑤 = 30◦ contributions to the DF can be divided into two regions: the area
and 𝑢 = 5.0 or 𝑢 = 0.2), once they have reached steady state. Wind around the star and the region well behind it. Regardless of the value
orientation strongly affects the morphology of the shock. For a polar- of 𝑢, the shocked region in front of the star has a higher density than
oriented wind, the radius at which the ram pressure of the stellar wind the region behind it (see Figure 3). As a result, the upwind shocked
balances with that of the ambient medium is ∼ 3 times farther away material exerts a positive DF, which tends to accelerate the star. The
from the star compared to the isotropic wind, while for equatorial integrated DF increases and peaks at 𝑍 ≈ 0 (the location of the star)
winds, it is ∼ 0.7 times farther (also relative to the isotropic wind). for 𝑢 = 5.0 and at 𝑍 ≳ 0 for 𝑢 = 0.2.
Regardless of the 𝑢 value and wind orientation, the shocked am- Notably, the contribution to the DF of material behind the star
bient medium has densities and velocities similar to those in the strongly depends on 𝑢, as 𝑢 dictates the density structure behind
isotropic case (𝜌 ∼ 1, 𝑣 ∼ 4). However, the shocked stellar wind ma- the star. For 𝑢 = 5, the density in the tail behind the star is higher
terial close to the star (that is, located in the region −5 ≤ 𝑍 ≤ 0) may than that of the unshocked medium, whereas it is lower for 𝑢 = 0.2.
Consequently, for 𝑢 = 5.0 the DF on the star is negative (𝐹𝐷 /𝐹𝐷 𝑂𝑠 ∼
reach densities up to one or two orders of magnitude lower than the
isotropic case (𝜌 ∼ 10 −2 and 𝜌 ∼ 10 −3 for the 𝑢 = 5.0 and 𝑢 = 0.2 𝑂𝑠
−35), and positive (𝐹𝐷 /𝐹𝐷 ∼ 7) for 𝑢 = 0.2, leading to deceleration
cases, respectively). Far from the star (−40 ≤ 𝑍 ≤ −30) the shocked and acceleration of the star, respectively. The possibility of a pushing

MNRAS 000, 1–12 (2024)

 Wind asphericity effects in DF 5
625.0 500
𝜽𝒘 = 𝟑𝟎° 𝑷 𝜽𝒘 = 𝟑𝟎° 𝑬 w = Iso
𝟎 𝒖 = 𝟓. 𝟎 𝒖 = 𝟓. 𝟎 400 w = 60 P
u = 5.0
w = 60 E
25.0 300 w = 45 P
w = 45 E
−𝟏𝟎 200 w = 30 P
1.0
100 w = 30 E
𝑍

𝝆
−𝟐𝟎 10−1 0

FD /FDOs
10−2 100
10
−𝟑𝟎 u = 0.2
10−3 8
10−4
6
−𝟒𝟎
𝜽𝒘 = 𝟑𝟎° 𝑷 𝜽𝒘 = 𝟑𝟎° 𝑬
25.0
4
𝟎 𝒖 = 𝟎. 𝟐 𝒖 = 𝟎. 𝟐
2
20.0
0
−𝟏𝟎 40 30 20 10 0 10
15.0 Z
𝒗
𝑍

−𝟐𝟎 Figure 5. 𝐹𝐷 /𝐹𝐷𝑂𝑠 profile for isotropic and aspherical winds as a function of
10.0 𝑍. The top panel shows models for 𝑢 = 5.0 and the bottom panel for 𝑢 = 0.2.
The black line represents the isotropic case. The green, blue, and red lines
indicate 𝜃𝑤 = 30◦ , 45◦ , 60◦ , respectively. Solid lines indicate polar-oriented
−𝟑𝟎 5.0 winds, and dotted lines represent the equatorial ones.

−𝟒𝟎 0.0 ranges between 8 and 9, while equatorial models have the highest DF
−𝟐𝟎 −𝟏𝟓 −𝟏𝟎 −𝟓 𝟎 𝟓 𝟏𝟎 𝟏𝟓 𝟐𝟎 values, i.e. 𝐹𝐷 /𝐹𝐷𝑂𝑠 ≃ 9.5 − 11. As the opening angle narrows, the
𝑅 DF increases.
Figure 5 also shows that the DF value converges to a limiting value
as 𝑍 becomes increasingly negative. To estimate the DF at infinity,
Figure 4. Density maps and velocity fields for asymmetric stellar winds at ∞ ≡ 𝐹 (𝑍 → −∞), we analytically solve Equation (2),
that is, 𝐹𝐷 𝐷
𝑡 = 𝑡 𝑓 . Polar winds (left panels) and equatorial winds (right panels) are shown
for 𝑢 = 5.0 (top) and 𝑢 = 0.2 (bottom). In all cases, 𝜃𝑤 = 30◦ . The density,
assuming that the shocked-wind density is constant far from the star
velocity, and axes are the same as in Figure 2. and scales as ∝ (𝑅 2 + 𝑧 2 ) −1 close to it. The solution is the following
(see Appendix B1 for further details):
∞
𝐹𝐷 (𝑧) = 𝐴 · 𝐻1 (𝑧) + 𝐵 · 𝐻2 (𝑧) + 𝐹𝐷 , (5)
force was also discussed by G20, who showed that in the regime 1 1
where 𝑢 ≪ 1, a low-density bubble forms downstream of the star, where 𝐴, 𝐵 are constants, 𝐻1 (𝑧) = (𝑟 02 + 𝑧 2 ) − 2 − (𝑟 12 + 𝑧 2 ) − 2 and
1 1
resulting in a forward-directed force that pushes the star. 𝐻2 (𝑧) = (𝑟 02 + 𝑧 2 ) 2 − (𝑟 12 + 𝑧 2 ) 2 . Here, 𝑟 0 is again the injection
Close to the star, at 𝑍 = 0, the behavior of the 𝑢 = 5.0 and 𝑢 = 0.2 radius of the wind, and 𝑟 1 is the corresponding radius to 𝐹𝐷 ∞ . For
models differs significantly in the asymmetric cases. For 𝑢 = 5.0, all ∞
each model, the 𝐹𝐷 value and statistical error were obtained and
the curves reach a maximum, while for 𝑢 = 0.2, the curves reach a were always below 4%. The best fit values of 𝐴, 𝐵 and 𝑟 1 (using
minimum. The peak preceding the minimum in the 𝑢 = 0.2 models Equation (5)) for representative models (two isotropic wind models
at 𝑍 > 0 is attributed to the shocked region located in front of the and two aspherical wind models) are provided in Table A2. The
star, where the density is higher than in the region behind it, resulting comparison between the fit and the data for the representative models
in a net positive DF (see Figure 3). is shown in Figure B1 (isotropic wind with 𝑢 = 5.0 and 𝑢 = 0.2) and
The asymmetric cases exhibit the same general trends as the spher- in Figure B2 (polar wind with 𝑢 = 5.0 and 𝜃 𝑤 = 45◦ and equatorial
ical case, with the following differences. In the case of 𝑢 = 5, the wind with 𝑢 = 0.2 and 𝜃 𝑤 = 60◦ ).
peak at 𝑍 = 0 increases for more collimated polar winds (as more Figure 6 shows the dependence of 𝐹𝐷 ∞ /𝐹 𝑂𝑠 for different winds as
𝐷
material is launched in the polar direction). For polar winds, the DF a function of 𝑢. Specifically, we show the isotropic wind model and
the polar and equatorial oriented wind models with 𝜃 𝑤 = 45◦ and
is closer to zero (𝐹𝐷 /𝐹𝐷 𝑂𝑠 from −15 to 0). For equatorial winds, the

result is almost independent of the value of 𝜃 𝑤 (𝐹𝐷 /𝐹𝐷 𝑂𝑠 ≃ −80), 𝜃 𝑤 = 60◦ . The error bars for each model are included in the figure
suggesting a weaker dependence on the opening angle compared to but are smaller than the size of the data points. For each case, the
the polar models. For the 𝑢 = 0.2 case, 𝐹𝐷 /𝐹𝐷 𝑂𝑠 always has positive critical value of 𝑢 = 𝑢 𝑐 for which 𝐹𝐷 ∞ is exactly zero (by fitting the
values, with a minimum at 𝑍 = 0, and increases as the distance from data with a parabola) is obtained. For the isotropic wind, we obtain
the star increases. The aspherical winds tend to yield larger values 𝑢 𝑐 = 1.95. For the isotropic wind case, G20 found a slightly lower
with respect to the spherical case. In polar winds, the normalized DF value of 𝑢 𝑐 = 1.71 for the same set of parameters that we used.

MNRAS 000, 1–12 (2024)

 6 J. D. Carrillo-Santamaria et al.

10 0
Polar
0 20 Equatorial
10
40
20
FD /FDOs

30 60
40 w = Iso 80 u = 5.0

FD /FDOs
50 w = 60 P 11
w = 60 E

uc = 1.15
uc = 1.43
uc = 1.95
uc = 2.78
uc = 3.21
w = 45 P
u = 0.35

60
w = 45 E u = 1.0 10
70
10 1 100
u 9

Figure 6. 𝐹𝐷 ∞ /𝐹 𝑂𝑠 as a function of 𝑢 for different wind geometries. The 8

𝐷
isotropic wind case is indicated by the solid black line, polar-oriented winds u = 0.2
are indicated by solid colored lines, and equatorial-oriented winds by dotted 730 40 50 60 70 80 90
lines. 𝜃𝑤 = 60◦ is in red and 𝜃𝑤 = 45◦ in blue. The vertical lines indicate
the values 𝑢 = 0.35, 1.0, 1.15, 1.43, 1.95, 2.78, 3.21 for which the DF is w
null. ∞ /𝐹 𝑂𝑠 as a function of 𝜃 for different winds. The upper and
Figure 7. 𝐹𝐷 𝐷 𝑤
lower panels show the 𝑢 = 5.0 and 𝑢 = 0.2 cases, respectively. Blue and red
The difference between the critical values is that we do not assume dots and lines represent equatorial and polar winds, respectively.
a thin shell approximation. We expect our value to drop slightly as
a function of cooling, approaching 𝑢 = 1.71 for a nearly isothermal 𝑢 = 0.2 to 5.0). In both the 𝑢 ≫ 1 and 𝑢 ≪ 1 limits, the DF
post-shock region, which is the regime well described by the thin increases as the opening angle narrows. This result is consistent with
shell approximation. the findings of L20, who followed the DF produced by a polar jet
Figure 6 shows that the asymptotic value of 𝐹𝐷 ∞ strongly depends
with 𝜃 𝑤 = 45◦ and found that the positive DF would dominate.
on the value of 𝑢. For 𝑢 < 1, the density in the tail behind the star Meanwhile, the DF for the equatorial wind follows a quadratic
is lower than the unperturbed ambient density, leading to a positive profile. The DF is always negative for 𝑢 = 5, 0 (with the isotropic
DF, whereas the opposite occurs for 𝑢 > 1. As a result, for 𝑢 < 1, case being the maximum) and positive for 𝑢 = 0.2 (with the isotropic
the DF is positive and increases with distance from the star. For model being the minimum). For smaller values of 𝑢, 𝐹𝐷 ∞ shifts toward
𝑢 = 1, in particular, the asymptotic DF is 𝐹𝐷 ∞ /𝐹 𝑂𝑠 ≈ 5.37. Also,
𝐷 larger positive values, while it shifts to a more negative value for
for 1 < 𝑢 < 1.95, the DF remains positive but decreases with the larger 𝑢’s. Determining the exact value of the new 𝑢 𝑐 in this case
distance from the star. For 𝑢 > 1.95, the DF becomes increasingly requires numerical calculations. The magnitude of the 𝐹𝐷 ∞ for the
negative (i.e., it decelerates the star). equatorial winds is larger than that for the polar and isotropic cases.
For small 𝑢 values, aspherical winds have larger accelerations than
that for an isotropic wind; larger 𝑢 values, may produce different
outcomes depending on the geometry of the wind. For example, 4.3 Timescale
for 𝑢 ∼ 1.5 values, polar winds accelerate while equatorial winds
To obtain 𝑢 as a function of time, we integrate the normalized DF
decelerate; for large 𝑢 values (𝑢 ≳ 3) equatorial winds produce
and its quadratic distribution shown in Figure 6:
larger decelerations than the polar case. Independently of the wind
∞
orientation, narrower aspherical winds have lower critical 𝑢 values. 𝑑𝑢 𝐹𝐷
= − 𝑂𝑠 = −𝑃0 𝑢 2 − 𝑃1 𝑢 − 𝑃2 , (6)
For the polar (and equatorial) models with 𝜃 𝑤 = 45◦ the critical 𝑑𝜏 𝐹
value is 𝑢 𝑐 = 2.78 (and 𝑢 𝑐 = 1.15) and for 𝜃 𝑤 = 60◦ the critical
𝐷

where we have defined the dimensionless time 𝜏 = 𝑡𝐹𝐷 𝑂𝑠 /(𝑀𝑣 )

value is 𝑢 𝑐 = 3.71 (and 𝑢 𝑐 = 1.43). 𝑤
(see Appendix B3 for further details).
For the isotropic wind the best-fit values are 𝑃0 = −1.90, 𝑃1 =
4.2 Opening angle dependence −0.02, and 𝑃2 = 7.29. Integrating from 𝑢 min = 𝑢 𝑚 to 𝑢 max = 𝑢 𝑐 for
∞ /𝐹 𝑂𝑠 as a function of the opening angle of the the case where 𝑢 < 𝑢 𝑐 , and from 𝑢 min = 𝑢 𝑐 to 𝑢 max = ∞ for 𝑢 > 𝑢 𝑐
Figure 7 shows 𝐹𝐷 𝐷 we obtain:
wind for polar and equatorial winds (with 𝑢 = 5.0 and 𝑢 = 0.2).  
The error bars for each model are included (and are smaller than the  tanh 𝜏𝜏𝑛 for 𝑢 0 < 𝑢 𝑐



data points for the 𝑢 = 5 case). Independently of 𝑢, the DF for polar 𝑢(𝜏) = 𝑢 𝑚 + (𝑢 𝑐 − 𝑢 𝑚 ) ·   , (7)
𝜏
 coth 𝜏𝑛
 for 𝑢 0 > 𝑢 𝑐
models follows a linear fit, with the minimum value being that for 
the isotropic case. For the 𝑢 = 5.0 case, the critical opening angle at
where 𝑢 𝑚 ≈ −0.03, 𝑢 𝑐 ≈ 1.95 and 𝜏𝑛−1 = −𝑃0 (𝑢 𝑐 − 𝑢 𝑚 ) ∼ 41 . For
which the asymptotic DF will be zero is 𝜃 𝑤,𝑐 = 10.16◦ .
further details, see Appendix B2. In Figure 8, we plot 𝑢 as a function
In order to analyze how the angular dependence of the DF for the
polar wind varies as a function of 𝑢, we run an extra set of polar
models (with 𝜃 𝑤 = 45◦ and 60◦ ) for different velocity values (from 1 For an isotropic wind 𝜏 = 0.26, and for a polar wind with 𝜃𝑤 = 45◦ or

MNRAS 000, 1–12 (2024)

 Wind asphericity effects in DF 7
6 the critical velocity in a shorter time (𝑡 𝑐 ∼ (1 − 10) Gyr ≲ 𝑡 𝐻 ), but
u > uc the DF will still remain negligible.
5 u = uc For a Sun-like star moving in an elliptical orbit (hence, with a
u < uc high Mach number) through the inner part of a disc around an ac-
4 tive galactic nucleus (AGN) (specifically, the inner most region of a
Seyfert galaxy with 𝑛 𝑎 ≳ 1015 cm −3 , Jiang et al. 2019), and assum-
ing the same parameters as above, we obtain 𝑡 𝑐 ∼ 10 Myr. Thus, the
3
u

DF is negligible since the timescale is much larger than the trans-

lation of the star around a 108 M ⊙ supermassive black hole (which
2 is ∼ 90 days, considering an AGN disc size of 1 lightday, Jha et al.
2022). In contrast, the DF acting on a Wolf-Rayet star located within
1 the inner part of an AGN disc is significant (since its critical time is
𝑡 𝑐 ∼ (100 − 1000) yr). If the Wolf-Rayet star is located in the outer
00.00 0.05 0.10 0.15 0.20
part of the AGN disc, the DF is unimportant (𝑡 𝑐 ∼ (0.4 − 4) Myr).
/ n
B[e] supergiants are an example of a star with an equatorial wind
(Martin et al. 2025). They are supergiants in the post-main sequence
evolutionary stage of massive stars, similar to Wolf-Rayets (charac-
Figure 8. Velocity ratio 𝑢 = 𝑣𝑎 /𝑣𝑤 as a function of time 𝜏/𝜏𝑛 for the terized by having a P Cygni profile, Balmer emission lines, forbidden
isotropic case. The regimes where 𝑢 > 𝑢𝑐 (blue dotted line), 𝑢 = 𝑢𝑐 (black oxygen and iron lines, and strong infrared excess, see, e.g., Zickgraf
dot-dashed line), and 𝑢 < 𝑢𝑐 (red dashed line) are shown. et al. 1986; Curé et al. 2005). These stars have slow high-density
equatorial winds, which form a decretion disc around the star (Kraus
& Lamers 2003; Shokry et al. 2018; Rubio et al. 2023). The decre-
of 𝜏/𝜏𝑛 for an isotropic wind and highlight two regimes (𝑢 < 𝑢 𝑐 and tion disc may produce a DF that alters the stellar migration within
𝑢 > 𝑢 𝑐 ). its YMC . B[e] stars with equatorial discs have 𝑀¤ 𝑤 ∼ 10 −6 𝑀⊙ yr −1
We estimate the time scale 𝜏𝑐 ≡ 𝜏 (𝑢 = 𝑢 𝑐 ) required for the star and 𝑣 𝑤 ≈ 103 km s −1 (Curé et al. 2005). For the case when the
with an isotropic stellar wind to converge to the critical velocity (due star is located in a YMC with 𝑛 𝑎 ∼ 104 cm −3 (Portegies Zwart
to the DF exerted by the ambient medium through which it moves). et al. 2010), the critical time is 𝑡 𝑐 ∼ 300 Myr. Due to the similar
Regardless of whether the star initially has 𝑢 > 𝑢 𝑐 and a negative DF, 𝑀¤ 𝑤 and 𝑣 𝑤 , is expected that a Wolf-Rayet star will have a similar
or 𝑢 < 𝑢 𝑐 and a positive DF, its velocity converges to the critical value 𝑡 𝑐 inside a cluster. Also, a B[e] star inside the accretion disc of an
over a timescale of 𝜏𝑐 ∼ 0.20 𝜏𝑛 (see Figure 7) for the isotropical and AGN is expected to have a similar 𝑡 𝑐 to a Wolf-Rayet for the same
polar oriented winds, and 𝜏𝑐 ∼ 0.15 𝜏𝑛 for the equatorial oriented reason. To explore whether the DF can play a role in accelerating
winds. The critical time in cgs units (𝑡 𝑐 ), for an isotropic wind, is a star out of a YMC and potentially produce a runaway star (for
given by (for further details see Appendix B4): more details, see Bhat et al. 2022), we integrate Equation (7) with
 1   − 12 respect to time 𝜏 for the case 𝑢 0 < 𝑢 𝑐 . For B[e] and Wolf-Rayet stars
𝑛𝑎 − 2 𝑀¤ 𝑤

6 (with 𝑀 ∼ (18 − 40) 𝑀⊙ and 𝑣 𝑤 = 1000 km s −1 , Massey 1981;
𝑡 𝑐 = 1.07 × 10
cm −3 10 −14 𝑀⊙ yr −1 Martin et al. 2025) within a massive cluster (with 𝑛 𝑎 ∼ 104 cm −3
 1   and a diameter of ∼ 20 pc, Piskunov et al. 2007). Assuming that the
𝑣𝑤 2 𝑣𝑎
× Gyr , (8) Wolf-Rayet or B[e] star starts crossing the cluster with a velocity of
100 km s −1 100 km s −1
𝑣 𝑎 = 10 km s −1 , the velocity when the end of the cluster is reached
where 𝑛 𝑎 is the ambient medium number density. This equation is estimated to be within (10.95 − 11.99) km s −1 (the details of
adopts typical values for the density of the interstellar medium, and this estimation are shown in Appendix B5). This velocity is below
wind velocity, wind mass loss rate, and displacement velocity of the escape velocities of stellar clusters (∼ 17 km s −1 , Weatherford
Sun-like stars. et al. 2023). Thus, the DF induced acceleration is not a significant
mechanism for producing runaway stars.
It has been proposed that some of the stars embedded within the
external regions of accretion discs of AGNs (with 𝑛 𝑎 ≃ 108 cm −3 )
5 ASTROPHYSICAL APPLICATIONS
could become “immortal”, as the accretion may counterbalance mass
For a star moving through the interstellar medium with 𝑛 𝑎 , 𝑀¤ 𝑤 , loss (Dittmann et al. 2021). If the outflow winds have velocities equal
𝑣 𝑤 , and 𝑣 𝑎 equal to the referential values in Equation (8) we obtain to a fraction 𝜆 of the escape velocity 𝑣 𝑒 , then 𝑢 = 𝑀𝑎 𝑐 𝑠 /(𝜆𝑣 𝑒 ). For
a critical time 𝑡 𝑐 ∼ 105 𝑡 𝐻 , with 𝑡 𝐻 the Hubble timescale. Also, 𝑀𝑎 = 5, 𝑐 𝑠 = 10 km s −1 and 𝑣 𝑒 ≃ 103 km s −1 , then 𝑢 ≤ 0.15
for G20 (which used 𝑛 𝑎 = 10 cm −3 , 𝑀¤ 𝑤 = 7.24 × 10 −11 𝑀⊙ yr −1 , if 𝜆 ≥ 1/3. Taking 𝜆 = 1/3, the outflow is sufficiently powerful
𝑣 𝑤 = 1000 km s −1 , and 𝑣 𝑎 = 10 km s −1 ), the critical time to reach to inflate a bubble (i.e. 𝑅0 is much larger than the Bondi-Hoyle-
steady state is well above the Hubble timescale (𝑡 𝑐 ≳ 103 Gyr ≫ 𝑡 𝐻 ). Lyttleton, BHL, radius, Hoyle & Lyttleton 1939; Bondi & Hoyle
Therefore, the DF is negligible in this case. A higher environment 1944) if 𝑀¤ 𝑤 ≫ 10 −5 𝑀⊙ yr −1 . Thus, only during phases of ex-
density or a greater stellar mass loss rate is required to reduce the tremely fast and massive outflows does the immortal star experience
timescale over which the DF becomes important. For example, a a push rather than a pull.
Wolf-Rayet star with 𝑀¤ ∼ (10 −6 − 10 −4 ) 𝑀⊙ yr −1 (Barlow et al. Another relevant case is that of a CO -either a neutron star or a
1981; Korb et al. 2024) moving through the same medium will reach stellar mass black hole (with mass 𝑀𝑐𝑜 )- moving through an AGN
disc. In this scenario, the CO could accrete at a super-Eddington rate,
ejecting part of the accreted material in a collimated outflow. Thus,
𝜃𝑤 = 60◦ , 𝜏𝑛 = 0.27 and 𝜏𝑛 = 0.29, respectively. For an equatorial wind the outflow mass loss rate is expected to be close to the BHL rate.
with 𝜃𝑤 = 45◦ or 𝜃𝑤 = 60◦ , 𝜏𝑛 = 0.16 and 𝜏𝑛 = 0.18, respectively. Assuming 𝑣 𝑎 = 10 km s −1 and 𝑣 𝑤 ∼ 𝑐/3 (this is, 𝑢 ∼ 10 −4 ), we

MNRAS 000, 1–12 (2024)

8 J. D. Carrillo-Santamaria et al.
have: ejection power with low accretion and low ejection power with high
accretion.
4𝜋𝜌 𝑎 𝐺 2 𝑀𝑐𝑜
2 𝑀𝑐𝑜 2
  
−15 𝑛𝑎 Our simulations do not include the gravitational force exerted by
𝑀¤ BHL = ≃ 5.86×10 𝑀⊙ yr −1 . the star on the gas. This simplification is justified if the BHL radius
𝑣 3𝑎 1 cm −3 𝑀⊙
(9) (𝑅BHL = 2𝐺 𝑀∗ /𝑣 2𝑎 , Hoyle & Lyttleton 1939; Bondi & Hoyle 1944)
is smaller than ∼ 0.25𝑅0 (where 𝑅0 is the stand-off radius; for more
For the case where the CO is located at the innermost region details see Shima et al. 1986). In terms of 𝑢, this condition implies
of the AGN disc (𝑛 𝑎 ∼ 1015 cm −3 ), the critical time is 𝑡 𝑐 ∼ that
5 𝑓 −1/2 (𝑀𝑐𝑜 /𝑀⊙ ) −1 yrs (where 𝑓 < 1 is the fraction of the BHL 𝑀¤ 𝑤
mass accretion rate that feeds the outflow, 𝑀¤ 𝑤 = 𝑓 𝑀¤ 𝐵𝐻 𝐿 ). Thus, 𝑢≲ , (11)
60 𝑀¤ BHL
the DF is important in the inner AGN disc for a CO. Meanwhile, the
DF in the outer part of an AGN disk (where 𝑛 𝑎 ∼ 108 cm −3 ) is not where 𝑀¤ 𝑤 is the wind mass loss rate of the stellar wind and 𝑀¤ BHL
of great importance since 𝑡 𝑐 ∼ 50 𝑓 −1/2 (𝑀𝑐𝑜 /𝑀⊙ ) −1 Myrs. Since is the BHL accretion rate. If the latter condition is satisfied, the
𝑢 = 𝑣 𝑎 /𝑣 𝑤 ≪ 1, the DF acts by accelerating the CO, causing it to density enhancement in the gravitational wake induced by the star at
migrate to a wider orbit. As the object moves outwards, the DF de- distances larger than ∼ 𝑅0 can be treated in linear theory. If so, the DF
−1/2 is 𝐹𝐷 + 𝐹𝐷 𝑂𝑠 , where 𝐹 𝑂𝑠 is given by Equation (3) with 𝑏
min ≃ 𝑅0 .
creases, since the timescale for this process scales as 𝑡 𝑐 ∝ 𝑛 −1 𝑣 𝑎 . 𝐷
Given that the density in a disc scales as 𝑛 ∝ 𝑟 −3/2 (see G20 and ref- The discussed astrophysical cases where the DF plays an important
erences therein), and that the velocity follows 𝑣 ∝ 𝑟 −1/2 (assuming a role satisfy this condition. The 𝑢 value for a Wolf-Rayet star in the
keplerian disc), we obtain 𝑡 𝑐 ∝ 𝑟 7/4 . This implies that although the inner part of an AGN disc and a B[e] star within a YMC is 𝑢 ≈ 10 −2 ,
DF causes the CO to migrate outwards, its effect gradually drops as while 𝑢 ∼ 10 −4 for the CO inside a CE. Thus, the condition from
the object moves to larger orbital radii. Equation (11) for the Wolf-Rayet case, 𝑢 ≲ 0.015, is satisfied; for
A CO moving through a CE also accretes at a fraction of the BHL the B[e] case, 𝑢 ≲ 0.18, is satisfied; and for the CO inside a CE,
rate. The high accretion rate can lead to the ejection of a jet or outflow 𝑢 ≲ 10 −3 , is also satisfied.
with 𝑣 𝑤 ≳ 𝑐/3 (see, e.g. Moreno Méndez et al. 2017; López-Cámara Our simulations have inherent limitations, particularly due to their
et al. 2019, 2020, 2022; Dori et al. 2023; Soker 2023, 2025). In partic- 2D nature and the chosen coordinate system. Because they are re-
ular, for a 16 𝑀⊙ Red Giant (RG) star with a density profile given by stricted to 2D, they lack turbulence and instabilities which may be
𝜌 = 0.68(𝑎/𝑅 ⊙ ) −2.7 g√︁cm −3 (Papish et al. 2015) and an orbital Ke- present in three-dimensional (3D) simulations. Also, by imposing
plerian velocity 𝑣 𝑎 = 𝐺 𝑀/𝑎 ≈ 100 (𝑀RG /16𝑀⊙ ) 1/2 (𝑎/𝑅 ⊙ ) −1/2 axis-symmetry, the jet only moves vertically along the polar axis. In
km s −1 (where 𝑀𝑅𝐺 is the mass of the RG), the critical timescale is contrast, 3D simulations have shown that the jet may wobble around
its axis of motion (López-Cámara et al. 2013) and that some degree
𝑎 2.2 𝑀𝑐𝑜 −1 𝑀RG 1/2 of asymmetry may be present (e.g., DuPont et al. 2024), though this
     
1
𝑡 𝑐 = 60 𝑓 − 2 s. (10) asymmetry is unlikely to significantly impact the DF. Future work
𝑅⊙ 𝑀⊙ 16𝑀⊙
will explore 3D models to capture a broader range of jet configura-
For 𝑓 = 0.1, a 1 𝑀⊙ CO, and an orbital separation of 𝑎 = 𝑅 ⊙ , tions. Other limits are that we neglect curvature effects in the motion
𝑎 = 10 𝑅 ⊙ and 𝑎 = 100 𝑅 ⊙ , the steady-state configuration is reached of the star, which can be potentially important in the context of CE
at 𝑡 𝑐 ∼ 190 s, 8 hours and 55 days respectively. At large orbital or AGN discs (Kim 2010). A detailed numerical study of a star or a
separations, the acceleration due to DF is negligible and becomes CO moving through a realistic CE is left for future work.
significant as the CO moves inward. This effect opposes to the 𝛼𝜆
mechanism (van den Heuvel 1976; Eggleton et al. 1976; Webbink
1984; de Kool et al. 1987). The fact that the DF depends on the
6 CONCLUSIONS
outflow geometry illustrates the relevance that outflows and jets can
have on the dynamical evolution of CEs (Shiber et al. 2019). In this paper, we study through a set of 2D, HD simulations the
Our results confirm the role of stellar winds in modifying the DF dynamical friction produced by the interaction between a stellar wind
and accelerating the star, as predicted by the analytical estimates of and its environment. We use a “wind tunnel” configuration, where the
G20. Additionally, we find that this effect also occurs for aspherical ambient medium has a constant velocity and the star remains fixed.
winds. A potential caveat is that if the star accretes, the accretion Three different stellar wind configurations are considered: isotropic,
process could suppress the wind ejection, implying that the DF will polar, and equatorial. Additionally, the wind opening angle and the
always act by decelerating the star (see L20 for further details). We velocity ratio between the stellar wind and ambient wind (𝑢 = 𝑣 𝑎 /𝑣 𝑤 )
argue that this is not necessarily the case, as accretion can suppress are varied. The integration time is such that steady state is obtained
wind ejection under the assumptions of Bondi accretion, particularly for all models.
if the process is spherically symmetric. In a more general scenario of Previous studies focused on how the dynamical friction decelerates
asymmetric accretion, accretion and ejection can occur at the same a moving stellar object. Recent studies found that acceleration may
time, with accretion taking place on the equatorial plane and ejection also be produced by the dynamical friction (see L96 and G20). We
happening along the polar directions. This mechanism may operate in confirm the latter and find the value of 𝑢 for which each regime
many astrophysical phenomena where jets are present, for example: takes place in different wind geometries. The critical value of 𝑢 that
Herbig-Haro (Bally et al. 1996), AGN (Blandford et al. 2019), tidal separates these two regimes is found for the isotropic wind, and for
disruption events (De Colle & Lu 2020), and CEs (Shiber et al. 2019; a subset of polar and equatorial wind models. The 𝑢 𝑐 obtained for
López-Cámara et al. 2019). In the case of a jet launched during the isotropic wind differs slightly (∼ 15%) from that of G20 where a
the CE phase, López-Cámara et al. (2019) computed this accretion- thin shell was assumed. For the aspherical wind models, the 𝑢 𝑐 value
ejection process self-consistently by assuming that a fraction of the may be lower or larger (∼ 65%) than that of the isotropic wind.
accreted material could power the jet. In this case, the ejection process The interaction between the stellar wind and the ambient medium
is expected to be intermittent, fluctuating between phases of high generates four distinct regions: stellar wind, shocked stellar wind,

MNRAS 000, 1–12 (2024)

 Wind asphericity effects in DF 9
shocked ambient medium, and ambient medium (the latter two are Gruzinov A., Levin Y., Matzner C. D., 2020, MNRAS, 492, 2755
separated by a bow-shock). For an isotropic wind, the position of Hoyle F., Lyttleton R. A., 1939, Proc. Camb. Philol. Soc., 35, 592
the bow-shock is found to differ from the thin shell solution of W96 Inaguchi T., Matsuda T., Shima E., 1986, MNRAS, 223, 129
(nearly double the distance value). Jha V. K., Joshi R., Chand H., Wu X.-B., Ho L. C., Rastogi S., Ma Q., 2022,
Aspherical winds produce different dynamical friction values com- MNRAS, 511, 3005
Jiang J., et al., 2019, MNRAS, 489, 3436
pared to isotropic winds. Aspherical winds with small 𝑢 values pro-
Kim W.-T., 2010, ApJ, 725, 1069
duce larger accelerations than those for an isotropic wind. Mean-
Korb E., Mapelli M., Iorio G., Costa G., 2024, arXiv e-prints, p.
while, aspherical winds with large 𝑢 values decelerate more than the arXiv:2410.08965
isotropic case (especially equatorial winds). Narrower winds produce Kraus M., Lamers H. J. G. L. M., 2003, A&A, 405, 165
an asymptotic-DF absolute value that is always greater than that for Li X., Chang P., Levin Y., Matzner C. D., Armitage P. J., 2020, MNRAS,
the isotropic case. 494, 2327
The DF effects eventually vanish for every model as the star reaches Liu M., Wang L., Fu X., Ho L. C., 2025, ApJ, 978, 87
a critical velocity; then they move at a constant velocity. The accel- López-Cámara D., De Colle F., Moreno Méndez E., 2019, MNRAS, 482,
eration of the star is well described by a quadratic function of the 3646
velocity of the star, and the critical time is similar for all wind ge- López-Cámara D., Moreno Méndez E., De Colle F., 2020, MNRAS, 497,
ometries. 2057
López-Cámara D., De Colle F., Moreno Méndez E., Shiber S., Iaconi R.,
We apply our results to various astrophysical phenomena. We find
2022, MNRAS, 513, 3634
that the dynamical friction is important for stars and CO in AGN
López-Cámara D., Morsony B. J., Begelman M. C., Lazzati D., 2013, ApJ,
discs and within CEs (and the migration of the star or CO may 767, 19
substantially change). For low density media, like those in YMCs, Martin R. G., Lubow S. H., Vallet D., Overton M., Lepp S., Zhu Z., 2025,
the DF timescale is so large that its effect is practically negligible MNRAS: Letters, 539, L31
compared to other relevant timescales of the system (e.g. the stellar Masset F. S., Velasco Romero D. A., 2017, MNRAS, 465, 3175
lifetime timescale). Massey P., 1981, ApJ, 246, 153
Moreno Méndez E., López-Cámara D., Colle F. D., 2017, MNRAS, 470,
2929
Ogata E., Ohsuga K., Yajima H., 2021, PASJ, 73, 929
ACKNOWLEDGEMENTS
Ostriker E. C., 1999, ApJ, 513, 252
JDCS acknowledges support from a SECIHTI fellowship. We ac- Papish O., Soker N., Bukay I., 2015, MNRAS, 449, 288
knowledge support from the DGAPA/PAPIIT grant IN113424. We Park J., et al., 2017, MNRAS, 472, 1995
gratefully acknowledge the computing time granted by DGTIC Piskunov A. E., Schilbach E., Kharchenko N. V., Röser S., Scholz R. D.,
2007, A&A, 468, 151
UNAM on the Miztli supercomputer (projects LANCAD-UNAM-
Portegies Zwart S. F., McMillan S. L. W., Gieles M., 2010, ARA&A, 48, 431
DGTIC-281 and LANCAD-UNAM-DGTIC-321). Rephaeli Y., Salpeter E. E., 1980, ApJ, 240, 20
Rubio A. C., Carciofi A. C., Bjorkman J. E., 2023, in Two in a Mil-
lion - The Interplay Between Binaries and Star Clusters. p. 42,
DATA AVAILABILITY doi:10.5281/zenodo.8345003
Shiber S., Iaconi R., De Marco O., Soker N., 2019, MNRAS, 488, 5615
The data underlying this article will be shared on reasonable request Shima E., Matsuda T., Inaguchi T., 1986, MNRAS, 221, 687
to the corresponding author. Shokry A., et al., 2018, A&A, 609, A108
Soker N., 2023, RAA, 23, 095002
Soker N., 2025, RAA, 25, 025023
REFERENCES Szölgyén A., MacLeod M., Loeb A., 2022, MNRAS, 513, 5465
Sánchez-Salcedo F. J., 2012, ApJ, 745, 135
Anastasopoulou K., et al., 2024, A&A, 690, A25 Sánchez-Salcedo F. J., Brandenburg A., 1999, ApJ, 522, L35
Bally J., Morse J., Reipurth B., 1996, in Benvenuti P., Macchetto F. D., Toyouchi D., Hosokawa T., Sugimura K., Kuiper R., 2020, MNRAS, 496,
Schreier E. J., eds, Science with the Hubble Space Telescope - II. p. 491 1909
Barlow M. J., Smith L. J., Willis A. J., 1981, MNRAS, 196, 101 Wang L., Li X., 2022, ApJ, 932, 108
Bhat A., Irrgang A., Heber U., 2022, A&A, 663, A39 Weatherford N. C., Kıroğlu F., Fragione G., Chatterjee S., Kremer K., Rasio
Blandford R., Meier D., Readhead A., 2019, ARA&A, 57, 467 F. A., 2023, ApJ, 946, 104
Bondi H., Hoyle F., 1944, MNRAS, 104, 273 Webbink R. F., 1984, ApJ, 277, 355
Cantó J., Raga A. C., Esquivel A., Sánchez-Salcedo F. J., 2011, MNRAS, Wilkin F. P., 1996, ApJ, 459, L31
418, 1238 Zickgraf F. J., Wolf B., Stahl O., Leitherer C., Appenzeller I., 1986, A&A,
Chamandy L., Blackman E. G., Frank A., Carroll-Nellenback J., Tu Y., 2020, 163, 119
MNRAS, 495, 4028 de Kool M., van den Heuvel E. P. J., Pylyser E., 1987, A&A, 183, 47
Conti P. S., Garmany C. D., De Loore C., Vanbeveren D., 1983, ApJ, 274, van den Heuvel E. P. J., 1976, in Eggleton P., Mitton S., Whelan J., eds, IAU
302 Symposium Vol. 73, Structure and Evolution of Close Binary Systems.
Curé M., Rial D. F., Cidale L., 2005, A&A, 437, 929 p. 35
De Colle F., Lu W., 2020, New Astron. Rev., 89, 101538
De Colle F., Granot J., López-Cámara D., Ramirez-Ruiz E., 2012, ApJ, 746,
122
Dittmann A. J., Cantiello M., Jermyn A. S., 2021, ApJ, 916, 48
Dori N., Bear E., Soker N., 2023, ApJ, 954, 143 APPENDIX A: UNIT CONVERSION
DuPont M., Gruzinov A., MacFadyen A., 2024, ApJ, 971, 34
Eggleton P., Mitton S., Whelan J., eds, 1976, Structure and evolution of close We use dimensionless units that can be rescaled to real physical
binary systems : symposium no. 73 held in Cambridge, England, 28 units using the correspondent normalization factor (see Table 2).
July-1 August, 1975 IAU Symposium Vol. 73 The variables in cgs units are indicated by a prime (e.g., the distance

MNRAS 000, 1–12 (2024)

10 J. D. Carrillo-Santamaria et al.

Table A1. Normalized DF at infinity (𝐹𝐷 ∞ /𝐹 𝑂𝑠 ) for different levels of re- of the stellar wind and 𝑟 1 the final radius of the stellar wind, 𝑧 0 is
𝐷
finement. The model used in all cases is an isotropic wind with 𝑢 = 5.0 and the location of the bow-shock (along the 𝑧-axis), and 𝑧 is an arbitrary
the values reported are those obtained once steady state has been reached. point away from the star. We consider only the DF produced by the
stellar wind, neglecting the effect of the shocked ambient medium,
∞ /𝐹 𝑂𝑠 which density differs from that of the environment at most by a factor
Levels 𝐹𝐷 𝐷
of four. With these hypotheses, Equation (2) reduces to:
3 −42.63 (𝜌(𝑟) − 𝜌0 )𝑅𝑑𝑅𝑧 ′ 𝑑𝑧 ′
∫ 𝑧 ∫ 𝑟1
𝐹𝐷 = 4𝜋𝐺 𝑀★ 3
4 −31.01 𝑧0 𝑟0 (𝑅 2 + 𝑧 ′2 ) 2
∫ 𝑧 ∫ 𝑟1 " 𝜌 𝑟 2 #
5 −31.36 1 0 𝜌0
= 4𝜋𝐺 𝑀★ 5
− 3
𝑅𝑑𝑅𝑧 ′ 𝑑𝑧 ′
6 −35.12 𝑧0 𝑟0 (𝑅 2 + 𝑧 2 ) 2 (𝑅 2 + 𝑧 2 ) 2
∫ 𝑧 " 𝜌 𝑟2 
−33.86 1 0

7 3 3
= 4𝜋𝐺 𝑀★ − (𝑟 12 + 𝑧 ′2 ) − 2 − (𝑟 02 + 𝑧 ′2 ) − 2
𝑧0 3
8 −33.10
 1 1
i
9 −33.42 −𝜌0 (𝑟 12 + 𝑧 ′2 ) − 2 − (𝑟 02 + 𝑧 ′2 ) − 2 𝑧 ′ 𝑑𝑧 ′ =

𝜌1 𝑟 02 h 2
"
1 1
= 4𝜋𝐺 𝑀★ − (𝑟 1 + 𝑧 2 ) − 2 − (𝑟 02 + 𝑧 2 ) − 2
∞ and 𝑟 (each one normalized to 3
Table A2. Best-fit parameters 𝐴, 𝐵, 𝐹𝐷 𝑖
𝑂𝑠 1 1
i h 1
𝐹𝐷 ) for the representative models shown in Figures B1 and B2. The 𝑢 −(𝑟 12 + 𝑧 20 ) − 2 + (𝑟 02 + 𝑧 20 ) − 2 − 𝜌0 (𝑟 12 + 𝑧 2 ) 2
parameter, 𝜃𝑤 , and the wind orientation for each model are indicated. I
1 1 1
ii
stands for isotropic, P for polar, and E for equatorial.
−(𝑟 02 + 𝑧 2 ) 2 . − (𝑟 12 + 𝑧 20 ) 2 + (𝑟 02 + 𝑧 20 ) 2 ,

Model 𝑂𝑠
𝐴/𝐹𝐷 𝑂𝑠
𝐵/𝐹𝐷 ∞ /𝐹 𝑂𝑠
𝐹𝐷 𝑟1
𝐷
(𝑢, 𝜃𝑤 , orientation) which can be re-written as:
h 1 1
i
(0.2, 90◦ ,𝐼) 0.30 3.16 7.33 2.53 𝐹𝐷 (𝑧) = 𝐴 (𝑟 02 + 𝑧 2 ) − 2 − (𝑟 12 + 𝑧 2 ) − 2
(5.0, 90◦ , 𝐼 ) 40.36 −3.58 −40.82 10.69 h 1 1
i
∞
(5.0, 45◦ , 𝑃) 85.09 −0.47 −18.41 34.85 +𝐵 (𝑟 02 + 𝑧 2 ) 2 − (𝑟 12 + 𝑧 2 ) 2 + 𝐹𝐷 ,
(0.2, 60◦ , 𝐸 ) 0.78 4.81 9.89 2.09
from where we obtain Equation (5), this is:
∞
𝐹𝐷 (𝑧) = 𝐴 · 𝐻1 (𝑧) + 𝐵 · 𝐻2 (𝑧) + 𝐹𝐷 ,
in cgs units is 𝑟 ′ ), and those in code units are indicated without a
prime (e.g., the normalized distance is 𝑟). where
1 1
The normalization factors for distance, density, velocity, and mass 𝐻1 (𝑧) = (𝑟 02 + 𝑧2 ) − 2 − (𝑟 12 + 𝑧 2 ) − 2 ,
are 𝑅0′ , 𝜌 ′𝑎 , 𝑐 ′𝑠 , and 𝑀★′ , respectively. This is 𝑟 ′ = 𝑟 · 𝑅0′ , 𝜌 ′ = 𝜌 · 𝜌 ′𝑎 , 1 1
𝑣 ′ = 𝑣 · 𝑐 ′𝑠 , and 𝑀 ′ = 𝑀 · 𝑀★′ . Since the Mach number is fixed at 5, 𝐻2 (𝑧) = (𝑟 02 + 𝑧2 ) 2 − (𝑟 12 + 𝑧 2 ) 2 ,
the normalization factor for the sound speed is 𝑐 ′𝑠 = 𝑣 ′𝑎 /𝑀𝑎 = 𝑣 ′𝑎 /5, ∞ are:
and the constants 𝐴, 𝐵, 𝐹𝐷
thus 𝑣 ′ = 𝑣 · 𝑣 ′𝑎 /5.
The normalization factor for the pressure is the ram pressure, hence 4𝜋𝐺 𝑀★ 𝜌1 𝑟 02
𝐴 = ,
𝑃 ′ = 𝑃 · 𝜌 ′𝑎 𝑐 ′2 ¤′
𝑠 . For the mass loss rate, we have 𝑀 = 4𝜋𝑟 𝜌 𝑣 =
′2 ′ ′ 3
¤ ′2 ′ ′
𝑀 · (𝑅0 𝜌 𝑎 𝑐 𝑠 ). We set 𝑅0 = 1, 𝜌 𝑎 = 1, 𝑐 𝑠 = 1, and 𝑀𝑎 = 5 in the 𝐵 = 4𝜋𝜌0 𝐺 𝑀★ ,
h 1 1
i
code. For the time normalization factor, see Appendix B4. ∞
𝐹𝐷 = 𝐴 (𝑟 12 + 𝑧 20 ) − 2 − (𝑟 02 + 𝑧 20 ) − 2
The conversion for the DF in code units and in physical units is: h i
1 1
′ +𝐵 (𝑟 12 + 𝑧 20 ) 2 − (𝑟 02 + 𝑧 20 ) 2 .
𝐹𝐷 = 𝐹𝐷 · 𝐺 𝑀★′ 𝜌 ′𝑎 𝑅0′ , (A1)
since the gravitational constant in the code is set to 𝐺 = 1. Mean- The sign change of the constant 𝐵 shown in Table A2 happens because
𝑂𝑠 , the conversion is:
while, for the case of 𝐹𝐷 in the 𝑢 = 5.0 models, the shocked stellar wind has higher density
than the ambient, while in the 𝑢 = 0.2 models the shocked stellar wind
𝑂𝑠 ′
𝐹𝐷 𝑂𝑠
= 𝐹𝐷 · 𝐺 2 𝑀★′2 𝜌 ′𝑎 𝑐 ′−2
𝑠 . (A2) has lower density than the ambient, causing the quantity 𝜌(𝑟) − 𝜌0
to be positive or negative depending on the fitted model.
Consequently, the normalization factor for the DF ratio is: For 𝑧 ≫ 1 and using a second-order Taylor approximation, we get:
𝑅0′ 𝑐 ′2 ′
𝑠 /(𝐺 𝑀★). 𝐴 h 2 i 𝐵 h i
∞
𝐹𝐷 (𝑧) ≃ 𝑟 1 − 𝑟 02 + 𝑟 02 − 𝑟 12 + 𝐹𝐷 .
2𝑧 2𝑧
For the asymptotic case 𝑧 → ∞:
APPENDIX B: ANALYTICAL MODELS ∞
𝐹𝐷 (𝑧 → ∞) ≡ 𝐹𝐷
B1 DF equation
Figures B1 and B2 show examples of fits obtained using this simple
To compute analytically the DF, we consider a wind density profile analytical description. The analytical model accurately reproduces
with two components. Close to the star, the wind density scales as both the behavior of the DF near the star, where the increase and
𝜌(𝑟) = 𝜌1 𝑟 02 /(𝑅 2 + 𝑧 2 ), where 𝑧 and 𝑅 are the vertical and radial drop in DF are due to the (approximately spherical) stellar wind,
cylindrical coordinates. Far from the star, the density is taken as and at 𝑧 ≪ 0, where the DF is determined by the approximately
constant inside a cylinder. In the following, 𝑟 0 is the injection radius cylindrical structure formed behind the star.

MNRAS 000, 1–12 (2024)

 Wind asphericity effects in DF 11

200 Simulation 400 Simulation

Theoretical fit Theoretical fit
150 300
100
200
50
100
0 u = 5.0, Iso u = 5.0, w = 45 ,P
FD /FDOs

FD /FDOs
0
7 10
6 8
5
6
4
3 4
2
2
1 u = 0.2, Iso u = 0.2, w = 60 ,E
0 0
40 30 20 10 0 10 40 30 20 10 0 10
Z Z

Figure B1. Comparison between the simulation (red solid line) and the fitted Figure B2. Comparison between the simulation (red solid line) and the fitted
theoretical model (blue dotted line) for the isotropic case. The top panel shows theoretical model (blue dotted line) for different winds. The top panel shows
the case for 𝑢 = 5.0 and the bottom panel for 𝑢 = 0.2. the case for a polar wind with 𝜃𝑤 = 45◦ and 𝑢 = 5.0. The bottom panel
shows the case for an equatorial wind with 𝜃𝑤 = 60◦ and 𝑢 = 0.2.

B2 Differential equation
In the case 𝑢 0 > 𝑢 𝑐 :
We solve Equation (6)    
𝜏 𝑢 − 𝑢𝑚 𝑢 − 𝑢𝑚
𝑑𝑢
= −𝑃0 𝑢 2 − 𝑃1 𝑢 − 𝑃2 , = arccoth − arccoth 0 ,
𝑑𝜏 𝜏𝑛 𝑢𝑐 − 𝑢𝑚 𝑢𝑐 − 𝑢𝑚
by defining 𝑢 𝑚 = −𝑃1 /(2𝑃0 ) ≈ −0.03 as the value of 𝑢 corre- For the limiting case 𝑢 0 → ∞:
sponding to the maximum of the parabola, and 𝑢 𝑐 as the physically  
𝜏 𝑢 − 𝑢𝑚
plausible root of the equation, that is = arccoth ,
√︃ 𝜏𝑛 𝑢𝑐 − 𝑢𝑚
𝑃12 − 4𝑃0 𝑃2 thus we obtain:
𝑢𝑐 = 𝑢𝑚 − ≈ 1.95 ,
2𝑃0 
𝜏

𝑢 = 𝑢 𝑚 + (𝑢 𝑐 − 𝑢 𝑚 ) coth .
Then, we can rewrite the differential equation as 𝜏𝑛
𝑃12
" !#
𝑃1 2

𝑑𝑢 𝑃2
= −𝑃0 𝑢 + − −
𝑑𝜏 2𝑃0 4𝑃02 𝑃0 B3 Time normalization in code units
h i
= −𝑃0 (𝑢 − 𝑢 𝑚 ) 2 − (𝑢 𝑐 − 𝑢 𝑚 ) 2 . The DF acceleration (𝑑𝑢/𝑑𝑡) is:
∞
𝐹𝐷
Thus, the time 𝜏 is: 𝑑𝑢
= − ,
𝑑𝑢 ′ 𝑑𝑡 𝑀★ 𝑣 𝑤
∫ 𝜏 ∫ 𝑢
1
𝜏= 𝑑𝜏 ′ = − ′
. ∞ is the DF at infinity, 𝑀 is the mass of the star, and 𝑣 is
0 𝑃0 𝑢0 (𝑢 − 𝑢 𝑚 ) − (𝑢 𝑐 − 𝑢 𝑚 ) 2
2 where 𝐹𝐷 ★ 𝑤
The integral admits two solutions, corresponding to the cases 𝑢 < 𝑢 0 the stellar wind velocity.
Normalizing by 𝐹𝐷 𝑂𝑠 (which is a constant), one obtains:
and 𝑢 > 𝑢 0 .
In the case 𝑢 0 < 𝑢 𝑐 , we have: 𝑑𝑢
∞
𝐹𝐷 1
𝜏

𝑢 − 𝑢𝑚
 
𝑢 − 𝑢𝑚
   = − 𝑂𝑠 .
= arctanh − arctanh 0 , (B1) 𝑂𝑠
𝑑 𝑡 𝐹𝐷 𝐹𝐷 𝑀★ 𝑣 𝑤
𝜏𝑛 𝑢𝑐 − 𝑢𝑚 𝑢𝑐 − 𝑢𝑚
where we have defined 𝜏𝑛 = 1/[−𝑃0 (𝑢 𝑐 − 𝑢 𝑚 )]. For the limiting Rearranging terms and assuming that the mass and stellar wind do
case 𝑢 0 = 𝑢 𝑚 : not change much, then:
  ∞
𝜏 𝑢 − 𝑢𝑚 𝑑𝑢 𝐹𝐷
= arctanh , = − 𝑂𝑠 .
𝑢𝑐 − 𝑢𝑚
 
𝜏𝑛 𝑂𝑠 /(𝑀 𝑣 ) 𝐹𝐷
𝑑 𝑡 𝐹𝐷 ★ 𝑤
thus we obtain:
  𝑂𝑠 /(𝑀 𝑣 ) with
𝜏 Thus, the dimensionless time parameter is 𝜏 = 𝑡 𝐹𝐷 ★ 𝑤
𝑢 = 𝑢 𝑚 + (𝑢 𝑐 − 𝑢 𝑚 ) tanh
𝜏𝑛 which we obtain Equation (6).

MNRAS 000, 1–12 (2024)

12 J. D. Carrillo-Santamaria et al.
B4 Critical time For 𝑣 𝑤 = 1000 km s −3 , 𝑛 𝑎 = 104 cm −3 , 𝑥 = 20 pc, and the initial
stellar velocity 𝑣 𝑎 = 10 km s −1 (i.e., 𝑢 0 = 0.01), we have:
Variables with the prime (’) symbol represent quantities in cgs units  
as explained in Appendix A. The law of motion for the velocity of 𝑀★
𝑟 = 5.52 × 10 −18 .
the star, 𝑣 𝑎 , is: 𝑀⊙
∞ ′
𝑑𝑣 ′𝑎 𝐹𝐷 Thus, for 𝑀★ = 18 𝑀⊙ we get 𝑟 = 9.94 × 10 −7 and for 𝑀★ = 40 𝑀⊙
= − .
𝑑𝑡 ′ 𝑀★′ we get 𝑟 = 2.21 × 10 −6 .
Integrating 𝑑𝑟/𝑑𝜏 = 𝑢, and considering Equation (B1) from Ap-
Since 𝑢 ′ = 𝑢 and dividing by 𝑣 𝑤 (assuming that the wind velocity
𝑂𝑠 ′ (which is also taken as constant), pendix B2, we get:
is constant in time) and by 𝐹𝐷 ∫ 𝑟 ∫ 𝜏  ′ 
then: 𝜏
′
𝑟= 𝑑𝑟 ′ = 𝑢 𝑚 + (𝑢 𝑐 − 𝑢 𝑚 )tanh + 𝐶0 𝑑𝜏 ′
∞
𝑀★′ 𝑣 ′𝑤 𝑑𝑢 𝐹𝐷 0 0 𝜏𝑛
′ = − 𝑂𝑠 ′ .
   
𝐹 𝑂𝑠 𝑑𝑡 ′
𝜏
𝐷 𝐹𝐷 = 𝑢 𝑚 𝜏 + (𝑢 𝑐 − 𝑢 𝑚 )𝜏𝑛 ln cosh + 𝐶0 sech (𝐶0 ) ,
𝜏𝑛
Using the conversion factor for the DF ratio shown in Table 2 (and
derived in Appendix A), and rearranging terms, we have: where 𝐶0 = arctanh[(𝑢 0 − 𝑢 𝑚 )/(𝑢 𝑐 − 𝑢 𝑚 )].
∞
Solving for 𝑟 ≈ 9.94 × 10 −7 , 𝜏𝑛 = 0.25, 𝑢 𝑐 = 1.95, 𝑢 𝑚 = −0.03,
𝑑𝑢
=
𝑑𝑢 𝐹𝐷
= − 𝑂𝑠 , and 𝐶0 = 0.02, we get 𝜏 ≈ 1.29 × 10 −4 , and for 𝑟 ≈ 2.21 × 10 −6 we
𝑂𝑠 ′
get 𝜏 ≈ 2.71 × 10 −4 .
 
𝑅0′ 𝑐𝑠′2 𝐹𝐷 𝑑𝜏 𝐹
𝑑 𝑡′ 𝐺 𝑀★′ 𝑀★′ 𝑣𝑤
′ 𝐷
Then the final velocity (𝑣 𝑎 ) for each 𝜏 is:
′   
𝑅0′ 𝑐𝑠′2 𝐹𝐷 𝑂𝑠 𝜏
where 𝜏 = 𝑡 ′ ′2 ′ is the dimensionless time parameter. 𝑣 𝑎 = 𝑣 𝑤 𝑢 𝑚 + (𝑢 𝑐 − 𝑢 𝑚 )tanh + 𝐶0 ,
𝐺 𝑀★ 𝑣𝑤 𝜏𝑛
Thus, using Equation (3) with Λ = 125, 𝑀𝑎 = 𝑣 ′𝑎 /𝑐 ′𝑠 = 5, 𝑐 ′𝑠 = 𝑣 ′𝑎 /5,
and replacing 𝑅0 as defined in W96, we get: where we have considered 𝑣 𝑎 = 𝑣 𝑤 𝑢. Thus, for the 𝑀★ = 18 𝑀⊙ case
(in which 𝜏 ≈ 1.29 × 10 −4 ) we get 𝑣 𝑎 ≈ 10.95 km s −1 , and for the
′1/2
25𝑣 𝑤 𝑣 ′𝑎 𝑀★ = 40 𝑀⊙ case (with 𝜏 ≈ 2.71×10 −4 ) we get 𝑣 𝑎 ≈ 11.99 km s −1 .
𝑡′ = 𝜏√ ′ 1/2  .
′
4𝜋𝐺 𝑀¤ 1/2 𝜌 𝑎 ln 25 × 241/2

This paper has been typeset from a TEX/LATEX file prepared by the author.
Finally, using 𝜏𝑐 ≈ 0.2𝜏𝑛 = 0.05 and the number density, the critical
time in cgs units is:
−1/2 ′−1/2 ¤ ′−1/2 ′1/2 ′
𝑡 𝑐′ 0.07 𝐺 −1 𝑚 𝐻 𝑛 𝑎
= 𝑀 𝑣𝑤 𝑣𝑎 .
Using typical values for the ambient medium and stellar wind, we
recover Equation (8).

B5 Equation of movement
The star velocity is:
𝑑𝑥
= 𝑢𝑣 𝑤 ,
𝑑𝑡
where 𝑥 is the displacement of the star, 𝑣 𝑤 is the stellar wind velocity,
and 𝑢 = 𝑣 𝑎 /𝑣 𝑤 is the velocity of the star to wind ratio. Replacing the
time by the dimensionless time parameter 𝜏 = 𝑡 𝐹𝐷 𝑂𝑠 /(𝑀 𝑣 ), we
★ 𝑤
get
𝑂𝑠
𝐹𝐷 𝑑𝑥
= 𝑢.
𝑀★ 𝑣 2𝑤 𝑑𝜏
Rearranging terms and assuming that the stellar mass and wind are
constant in time, we get:
 
𝑑 𝑥 𝐹𝐷 𝑂𝑠 /(𝑀 𝑣 2 )
★ 𝑤
= 𝑢.
𝑑𝜏
Thus, the dimensionless displacement parameter is 𝑟 =
𝑂𝑠 /(𝑀 𝑣 2 ). For the parameters in cgs units we get:
𝑥 𝐹𝐷 ★ 𝑤

𝐺 2 𝑀★ 𝜌 𝑎
𝑟 = 60.42 𝑥
𝑣 2𝑎 𝑣 2𝑤
  
−13 𝑀★ 𝑛𝑎
= 2.76 × 10
𝑀⊙ cm −3
  −2   −2  
𝑣𝑎 𝑣𝑤 𝑥
× .
100 km s −1 100 km s −1 pc

MNRAS 000, 1–12 (2024)