Rony Keppens & Jon Sundqvist

Computational Methods for

Astrophysical Applications
–Hands-on simulating with MPI-AMRVAC–

October 12, 2020

Springer Nature
Contents

5 Transonic stellar winds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

5.1 Governing PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
5.2 Isothermal and polytropic stellar winds: theory . . . . . . . . . . . . . . . . . . 2
5.2.1 Parker’s isothermal wind solution . . . . . . . . . . . . . . . . . . . . . . . 4
5.2.2 Generalizations to polytropic winds . . . . . . . . . . . . . . . . . . . . . 10
5.2.2.1 Spherically symmetric, polytropic winds . . . . . . . . . 10
5.2.2.2 Equatorial, polytropic winds for rotating stars . . . . 12
5.2.2.3 2D winds for rotating stars . . . . . . . . . . . . . . . . . . . . 15
5.3 Handling orthogonal coordinate systems∗∗ . . . . . . . . . . . . . . . . . . . . . . 18
5.3.1 Geometrical source terms∗∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3.2 Boundary conditions at a symmetry axis∗∗ . . . . . . . . . . . . . . . 21
5.4 Inertial versus non-inertial reference frames∗∗ . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

v
Chapter 5
Transonic stellar winds

Abstract In this chapter, we discuss theoretical and numerical aspects related to

transonic stellar winds, where the isothermal, spherically symmetric Parker wind
represents a reference stationary solution. It is applicable to the solar wind, which is
thermally driven and smoothly varies from subsonic to supersonic flow speeds. We
discuss extensions to rotating, axisymmetric transonic winds, that are all described
by the polytropic hydrodynamic equations from the previous chapter, extended with a
central gravitational source term. In computing steady-state, or even time-dependent,
transonic wind solutions numerically, we usually work in orthogonal curvilinear
coordinates, such as cylindrical or spherical coordinate systems. We discuss the
resulting geometrical source terms, as well as the boundary conditions to be imposed
at symmetry or polar axes. We end with a note on the equivalent equations to solve,
involving fictitious forces, in co-rotating, non-inertial reference frames.

5.1 Governing PDEs

In this chapter, we augment the isothermal/polytropic hydro Eqns. (??) from the
previous chapter, with the influence of an external and central gravitational field, due
to a star of total mass M∗ . This generally introduces an external gravitational acceler-
ation g = −∇Φgr derived from a gravitational potential Φgr . Here, we only consider
the simplest form for the potential, namely Φgr = −GM∗ /r where r denotes the radial
distance to the star, which acts as a point mass. G = 6.6743 × 10−11 m3 kg−1 s−2 is
the gravitational constant (in SI units). The set of governing PDEs then features a
gravitational source term in the momentum equation, and we get

∂t ρ + ∇ · ( ρ v) = 0,
(
(5.1)
∂t m + ∇ · (m v + pI) = Sg ,

where
GM∗
Sg = ρg = −ρ êr . (5.2)
r2

1
2 5 Transonic winds

As before, the closure relation specifying the pressure p( ρ) for the isothermal case
is p = ci2 ρ, while a polytropic case writes p = cad ργ . These equations allow for
solutions that represent stellar outflows, i.e. winds, or inflows, i.e. accretion processes.
That such stellar winds can go transonic without necessarily involving shocks, hence
may vary smoothly from subsonic to supersonic flow regimes, is best illustrated with
the basic Parker wind solution [1]. This analytic solution is possible by reducing
the above PDEs to ODEs that describe a 1D, spherically symmetric, steady-state
outflow. We will discuss the Parker wind solution in detail, and gradually generalize
it to discuss multi-dimensional, stationary stellar outflows for rotating stars.

Eugene Parker
Eugene Parker (born 1927) is a unique theoretical astro-
physicist who made numerous contributions to our un-
derstanding of solar physics and astrophysics in general.
He predicted in the 1950s that the solar atmosphere is
in a state of continuous transonic expansion, making the
Earth and all other solar system planets engulfed by a su-
personic outflow known as the ‘solar wind’. This was later
on verified by satellite measurements, and his model now
serves as the basic model for all stellar outflows, which are
omnipresent. In august 2018, NASA launched the Parker
solar probe, making Eugene Parker the first living scien-
tist with a spacecraft named after him. Parker is professor
emeritus at the University of Chicago, and received the
Crafoord prize in Astronomy in 2020. The citation reads:
"for pioneering and fundamental studies of the solar wind and magnetic fields from stellar
to galactic scales"

The (annual) Crafoord prize was first awarded in 1982, and its prize sum of 6 million
swedish crowns makes it one of the largest scientific prizes to receive.
Parker’s name is meanwhile linked to many astrophysical processes, such as the
Parker wind, the Parker spiral, the Parker instability, Parker dynamo, Sweet-Parker
reconnection, . . .

5.2 Isothermal and polytropic stellar winds: theory

The system of PDEs from Eq. (5.1) actually represents a multitude of models.
They range from those governing one-dimensional, purely radial flows in the central
gravitational stellar field, to models that try to incorporate flows that deviate from the
purely radial direction – such as resulting from stellar rotation, to those that govern
truly multidimensional flows in multidimensional settings. To that end, we should
note that the system (5.1) is written in vector notation, expressing the conservation
5.2 Isothermal and polytropic stellar winds: theory 3

of mass and governing the evolution of vectorial momentum. When written out for
a Cartesian (x, y, z) coordinate system, and restricting to 1D such that v = v êx ,
this system (without the central gravity) indeed becomes the one given in Eq. (??),
which in turn was a specific 1D reformulation of the conservations laws expressed
in equations (??).
Due to the central gravitational field, we may prefer to look at the system in ways
that closely reflect underlying symmetries. Indeed, we can look for solutions that
are perfectly spherically symmetric, which in essence reduce the set of unknowns
to ρ(r, t) and m = mr êr , where mr (r, t) = ρ(r, t)vr (r, t). There are then only two
equations left, but the divergence terms in the mass and momentum evolutionary
laws need to be written out for spherical geometry. A more complete description
may only assume axisymmetry, where the flow may have three components, but
where an axis through the central star is chosen as special – a logical choice being
the rotation axis when the star is assumed to spin as a rigid body. For axisymmetric
conditions, we may consider so-called 2.5D settings, where we have ρ(r, θ, t) and
similarly for velocity v = vr êr + vθ êθ + vϕ êϕ in a spherical coordinate (r, θ, ϕ)
system. Alternatively, we may in axisymmetry adopt a cylindrical coordinate system
(R, Z, ϕ), where then the unknowns are ρ(R, Z, t) and v = vR êR + v Z ê Z + vϕ êϕ .
When no symmetry is assumed to prevail at all, it may still be beneficial to solve
the governing equations in these latter coordinate systems, using the corresponding
vector components.
This all means that we will need to write out the divergence of a vector, and
the divergence of a tensor quantity (namely the symmetric tensor ρvv + pI in the
momentum equation) in these coordinate systems. To that end, we note that any
textbook on tensor calculus reveals
1 1
∇·w= ∂R (Rw R ) + ∂Z w Z + ∂ϕ wϕ
R R
1 1 1
= 2 ∂r (r wr ) +
2
∂θ (sin θ wθ ) + ∂ϕ wϕ , (5.3)
r r sin θ r sin θ
for a vector w, while for a second rank tensor T we have
4 5 Transonic winds
" #
1 1 Tϕϕ
∇ · T = êR ∂R (RTRR ) + ∂Z TZ R + ∂ϕ TϕR −
R R R
" #
1 1
+ê Z ∂R (RTRZ ) + ∂Z TZ Z + ∂ϕ TϕZ
R R
" #
1 1 TϕR
+êϕ ∂R (RTRϕ ) + ∂Z TZϕ + ∂ϕ Tϕϕ + (5.4)
R R R
Tθθ + Tϕϕ
" #
1 1 1
= êr 2 ∂r (r 2Trr ) + ∂θ (sin θ Tθr ) + ∂ϕ Tϕr −
r r sin θ r sin θ r
Tθr cot θ Tϕϕ
" #
1 1 1
+êθ 2 ∂r (r 2Tr θ ) + ∂θ (sin θ Tθθ ) + ∂ϕ Tϕθ + −
r r sin θ r sin θ r r
θ Tϕθ
" #
1 1 1 T ϕr cot
+êϕ 2 ∂r (r 2Trϕ ) + ∂θ (sin θ Tθϕ ) + ∂ϕ Tϕϕ + + .
r r sin θ r sin θ r r

This is all that is needed to write out the governing equations (5.1) for any choice of
dimensionality and symmetry combination that is used in the rest of this chapter.

Exercise
5.1 Write out the governing equations for an axisymmetric 2.5D instance of
Eqns. (5.1), when using (a) cylindrical and (b) spherical coordinates and vector
components.

5.2.1 Parker’s isothermal wind solution

In his pioneering paper [1], Parker reduced the problem of seeking solutions to
Eqns. (5.1) to solutions that are time-independent or stationary. By further assuming
perfect spherical symmetry, making v = vr (r)êr and neglecting all angular variation
∂θ = 0, ∂ϕ = 0 in a spherical coordinate system, we are left with ODEs, as all that
remains from Eqns. (5.1) is
1 d  
∇ · ( ρv) = 0 ⇒ r 2 ρvr = 0 (5.5)
r2 dr
GM∗ 1 d   2p GM∗
∇ · ( ρvv + pI) = −ρ 2 êr ⇒ 2 r 2 ρvr2 + r 2 p − = −ρ 2 .
r r dr r r
This is in reality only valid for non-rotating (and also unmagnetized) stars, with only
a radial variation of density and radial velocity, and where only the r-component of
the momentum equation remains. For isothermal conditions, the pressure is directly
proportionate to density p = ci2 ρ, and we can rewrite the latter equation to
5.2 Isothermal and polytropic stellar winds: theory 5

dvr dρ GM∗
ρvr + ci2 + ρ 2 = 0. (5.6)
dr dr r
In case we have a radially decreasing density profile ρ(r) as expected for a stellar
atmosphere, we can use this equation to study the possibility where a continuously
accelerating wind solution dv dr > 0 becomes feasible, where thermal acceleration
r

counteracts the gravitational pull. We can in fact directly integrate both governing
equations, since we obtain from mass conservation that the mass flux ρr 2 vr must be
constant, and that Eq. (5.6) integrates to

vr2 GM∗
+ ci2 ln ρ − = C, (5.7)
2 r
where C is an integration constant. This constant also has a physical meaning, as
it actually expresses how along a streamline energy is conserved1. Another useful
form reorganizes Eq. (5.6) to the expression

dvr vr * 2ci2 − GM∗ /r +

= . (5.8)
dr r , vr2 − ci2 -

In this form, it is clear that at any specific location where the radial velocity happens
to be identical to the isothermal sound speed ci , the denominator of the expression in
brackets must also vanish locally to have finite values for the radial velocity derivative,
hence this can only occur at the critical point r s = G2cM2∗ . For the isothermal hydro
i
case, the radial position of this sonic point where vr = ci can be determined a priori,
as we know the stellar mass M∗ and we note√that a given coronal temperature T0
actually sets the isothermal sound speed from RT0 = ci . At this critical r s location,
it will be possible to uniquely match a subsonic solution to a supersonic solution
part, realizing a continuously accelerating, transonic wind.
To demonstrate that possibility, let us first introduce dimensionless quantities, by
measuring all radii in base radii r 0 by setting r̄ = r/r 0 : this r 0 could be the stellar
radius itself r 0 = r ∗ , but may be taken differently too. We further express all densities
in terms of the base density value ρ0 , and measure all velocities in terms of the Mach
number M = vr /ci . Simply rewriting the equation (5.7) in these new, dimensionless
quantities gives
M2 v̄ 2
!
1
+ ln 2 − esc = Ē (5.9)
2 r̄ M 2r̄
where Ē is a constant,√and where we introduced the ratio of the escape speed to the
sound speed, v̄esc = 2GM∗ /r 0 /ci = 3.3015. The numeric value given is typical
for solar conditions. Note that the dimensionless quantification for the critical point
r̄ s = r s /r 0 = v̄esc
2 /4 is just a change of notation, and for solar conditions, we find

r s = 2.72498r 0 .

1 Actually, energy may be the wrong word in this context: the isothermal equations do not express
energy conservation, but ensure that the temperature is uniform throughout, requiring unspecified
energy loss/gain terms.
6 5 Transonic winds

Exercise
5.2 Go through the details of rewriting the equations for the Parker wind in
its various forms. In particular, familiarize yourself with the way in which we
can write the governing equations only in terms of dimensionless quantities,
which is an essential part for doing numerical simulations relating to a specific
astrophysical problem. The freedom to choose a unit of length, mass and time
means that in general, a hydro simulation will be free to measure distances in
a preferred reference length, velocities in a preferred velocity magnitude, and
densities in a relative fashion to densities at a prechosen position. A proper
choice of such code units is very important to tailor to the simulation at hand,
given the fact that rounding and truncation errors are unavoidable: one should
always try to ensure that in the scaled dimensionless code units, some (or all)
of the values encountered in the computation (both the domain setting and all
the initial conditions) are of order unity.

Fig. 5.1 The function Ē (r̄, M ) plotted with level contours in the (r̄, M ) plane. The escape speed
was taken in accord with solar coronal conditions.

If we actually plot the 2-dimensional function Ē(r̄, M), we note that solutions
to our original problem will correspond to contour lines of Ē(r̄, M) in the (r̄, M)
plane. Using the solar value for the v̄esc , the result is shown in Fig. 5.1, where we
used a logarithmic scaling on both the horizontal (scaled radius) and vertical (Mach
number) axis. Any solution to our original problem of stellar outflow is a contour
5.2 Isothermal and polytropic stellar winds: theory 7

Fig. 5.2 The function Ē (r̄, M ) plotted with level contours in the (r̄, M ) plane, this time using a
linear scale on both axes, and for a case with a higher than solar value of the escape speed, such
that the sonic point is at 5 stellar radii.

Fig. 5.3 The function Ē (r̄, M ) from Fig. 5.1, plotted as a surface plot, demonstrating the saddle
point topology near the critical point.

line in this plot, which should connect r̄ = 1 to r̄ = ∞ in a single-valued curve: at

any radius, only one velocity should prevail. The same function Ē(r̄, M), this time
8 5 Transonic winds

for a case where r s = 5r 0 (i.e. for a higher escape speed value than the solar case),
using a linear scale on both horizontal and vertical axis is shown in Fig. 5.2.
The topology of Ē(r̄, M), for the solar case additionally visualized as a surface
plot in Fig. 5.3, shows that 4 classes of solution exist, of which double valued curves
must be discarded. There are then a multitude of solutions that are purely supersonic
and have M > 1 throughout, as well as a multitude of solutions that always remain
subsonic M < 1. The latter are known as stellar breeze solutions, while the former
are inconsistent with the solar case: no supersonic speeds occur near the solar
surface. The two remaining solid curves in Fig. 5.1 show the distinct possibility of
transonic solutions: one starting subsonically and increasing to supersonic speeds by
passing through the critical point, and one which would correspond to an accretion
flow which starts subsonic at far distances, and turns supersonic near the accreting
object (this is known as the Bondi solution [2], relevant for accretion inflow where
vr < 0 throughout). The unique transonic wind solution was pointed out by Parker
in 1958 [1], before satellites confirmed its existence! The closed from solution for it
is given by
M2
!
r̄ r¯s 3
− ln M − 2 ln −2 + =0 (5.10)
2 r̄ s r̄ 2
where we used the fact for that unique solution, we can obtain the value of the
constant Ē by requiring the curve to pass through (r̄, M) = (r̄ s, 1). Manipulating this
equation, we find that
M2
 r̄  2 3 r̄
s s
M e− 2 = e 2 −2 r̄ (5.11)
r̄
Analyzing this governing transcendental equation for the velocity (Mach number) as
function of radius r̄, we conclude that at large radii, the velocity diverges to infinity
since p
M∞ ∝ ln(r) (5.12)
The density (and hence pressure) for this solution behave physically ok, since mass
conservation dictates ρ(r) = p(r)/ci2 = ρ0 M 0 1
M r̄ 2 so that density and pressure drop to
zero for r → ∞. The fact that in an isothermal corona (like valid near the sun whose
coronal temperature is near constant for several solar radii and about 106 K) the
pressure gradient can drive a continuous outflow led Parker to predict the existence
of a transonic wind, such that the earth’s magnetosphere is embedded in supersonic
solar wind outflow. This was later confirmed by satellite measurements, and would
rightfully deserve Parker to receive the Nobel prize in physics.

Exercise
5.3 Go through the derivations to obtain the transcendental equation (5.11) for
the transonic solution branch. Find a means to plot this solution, showing the
variation of density, pressure and radial velocity with radius.
5.2 Isothermal and polytropic stellar winds: theory 9

Note that the supersonic solar wind is eventually also terminated by a shock
transition, demarkating the heliosphere from the interstellar medium. The location
of this termination shock can be estimated from the transonic solar wind solution, as
its pressure is at large distance dominated by the ram pressure pram ≡ ρvr2 alone. The
termination shock can then be expected to be where the (assumed known) pressure
of the interstellar medium equals the local ram pressure.
Since the unique isothermal transonic wind solution is eventually governed by the
single transcendental equation (5.11) for the Mach number M (r̄), one can directly
solve this transcendental equation and use standard root-finding procedures to obtain
the run of Mach number (hence velocity) at any radial distance. This is best done
from the known location r̄ s of the critical point where M = 1, on both an inwards
(towards smaller radii, where the subsonic branch is followed) and an outwards (to
larger radii where the supersonic regime prevails) iteration. One can use a standard
fixed point iteration (also referred to as Picard iteration) which writes the governing
nonlinear expression as M = f (M), collecting all nonlinear/transcendental behavior
in the right hand side function f (M). Then, one subsequently iterates to a fixed
point using Mi+1 = f (Mi ) from a starting guess M0 , and verifies whether there is
convergence in the sequence M0, M1, M2, M3, . . . to a prechosen desired accuracy.
For the isothermal Parker wind solution, we note that for the subsonic branch, it is
most convenient to use the expression
 r̄  2 M2
s + 32 −2 r̄r̄s
M= e 2 (5.13)
r̄
in this fixed point iteration procedure, while the supersonic branch better exploits
s
M r̄ 2
!
r̄ s
M = 2 ln 2
+4 −3. (5.14)
r̄ s r̄

Hence, this solution can be computed to any desired accuracy, and this ‘exact’ so-
lution can then be used e.g. as initial condition. We can then check any numerical
solver for the isothermal hydro system in spherical symmetry, allowing for an exter-
nal central gravitational field, by prescribing this particular transonic wind solution
at t = 0, and checking whether the employed discretization can maintain this solution
as a (numerical) steady state. In particular, since the TVDLF scheme is second order
accurate for smooth solutions, and can handle both sub- and supersonic regimes well,
it can be demonstrated that this (as well as many other) shock-capturing scheme rec-
ognizes such ‘exact’ transonic solutions without serious problems (a small transient
dynamical phase may occur, where there is some adjustment due to specific details
in the boundary treatment).

Exercise
5.4 Set up a Picard iteration scheme to build the unique transonic Parker wind
solution. Use this to initialize a 1D, time-dependent simulation in spherical
symmetry using MPI-AMRVAC, and verify that (after some relaxation), the code
10 5 Transonic winds

can recognize this as a smooth, steady-state solution to the governing PDEs.

You must now work in code units, and assign meaningful boundary conditions
at r = r 0 and at the end of your simulation domain.

Exercise
5.5 Modify all the above to consider the unique transonic Bondi solution,
showing steady-state, spherically symmetric accretion onto a star.

5.2.2 Generalizations to polytropic winds

The 1D isothermal transonic wind solution can be generalized in many ways. In

what follows, we restrict ourselves to those generalizations that are relatively easy
to accomplish. We first discuss how to obtain stationary, spherically symmetric
solutions for polytropic, instead of isothermal, winds. Next, we allow for rotation
where we at first focus on the wind solution in the equatorial plane alone, to finally
generalize this to axisymmetric polytropic hydrodynamic winds for rotating stars.

5.2.2.1 Spherically symmetric, polytropic winds

The isothermal hydrodynamic system wrote the pressure p = ci2 ρ, where ci is the
constant isothermal sound speed, in essence set by the assumed constant temperature.
A direct generalization writes the prevailing pressure-density relation, or equation
of state, as p = cad ργ , where cad as well as γ are given constants. This includes
the isothermal case when cad = ci2 and γ = 1, but generalizes this to a polytropic
pressure-density relation which is characterized by a globally constant ‘entropy’
pρ−γ . One can again make analytic progress for the time-independent, 1D spherically
symmetric case, and find out that mass conservation and radial momentum equations
still dictate a constant mass flux ρr 2 vr as well as a constant ‘energy’

vr2 cs2
!
GM∗
E= + − /cs0
2
(5.15)
2 (γ − 1) r

2 = γc ρ γ−1
where the base value of the squared sound speed is cs0 ad 0 . Further algebraic
manipulations of the governing equations then reveal that the transonic spherically
symmetric stellar wind has the location for the sonic point r s satisfying
5.2 Isothermal and polytropic stellar winds: theory 11

! 2(γ + 1) ! 2(γ − 1)
rs vesc 5 − 3γ cs0 5 − 3γ
= (5.16)
r0 2cs0 vr0

This equation instantly demonstrates a key point for transonic polytropic winds:
only the isothermal γ = 1 case is such that this critical point location is determined
a priori, and computable from given parameters (stellar mass, bottom density and
temperature). As soon as the polytropic index γ differs from unity, the location of
the critical point is only known when the full solution is known, since it depends on
the base velocity vr0 . Therefore, computing polytropic transonic wind solutions can
not be done in the same way as explained for the isothermal Parker wind, where we
mentioned the use of a fixed point iteration on the governing transcendental equation,
since one would at least need to augment it with an outer iteration which gradually
shifts the location of the critical point as well.

Fig. 5.4 The radial variation of Mach number for three different polytropic wind solutions, for
increasing values of the polytropic index γ, showing the outward shift of the sonic point. From [3].

However, when we instead use a (pseudo) time stepping scheme, such as the con-
servative, shock-capturing TVDLF scheme, we can initialize the numerical solution
with the γ = 1 Parker wind, while solving the equations for a fixed γ > 1 value2. The
solution will then readjust, and after a transient phase relax to the ‘exact’ transonic
solution for the given polytropic index γ > 1. In this relaxation, it is essential to use
boundary conditions that are physically motivated: e.g. at the supersonic end, one
can use a mere extrapolation of all quantities in a von Neumann boundary condition
where the radial gradient of all quantities vanishes. At the subsonic end, things are
more subtle, as we must allow for sufficient freedom to allow the solution to read-
just to the correct local base velocity and acceleration value, and this can be done

2 This is the approach taken in [3], and several of the following results are discussed in that paper.
12 5 Transonic winds

using a Dirichlet prescription setting ρ(r = r 0 ) = ρ0 combined with a fixing of the

initial gradient of ρvr (this in essence sets the mass flux at the base). Using this
strategy, Fig. 5.4 shows three polytropic wind solutions, for three different values of
the polytropic index γ (including the isothermal γ = 1 case). Note that the effect of
increasing γ translates in a gradual outwards shift of the sonic point. This polytropic
wind solution is mathematically also better behaved for large radii than the isother-
mal solution, in the sense that one may deduce the following asymptotic, constant
radial speed value s
2
2cs0 GM∗
vr∞ = −2 . (5.17)
γ−1 r0
This expression has been obtained assuming that the Mach number at the base is
much less than unity, and also shows that for a given stellar mass and radius (hence,
for a given escape speed vesc ) for which the base (coronal) density and pressure are
known, there is a restriction for the polytropic index to end up with a positive value
underneath the square root. In particular, for solar values where vesc = 3.3015cs0 , the
allowed range is 1 < γ ≤ 1.183. For polytropic indices within this range, the radial
velocity at infinity for a transonic polytropic wind becomes the constant value (5.17),
which is to be contrasted with the fact that the isothermal wind for γ = 1 diverges to
infinity as expressed by (5.12).

Exercise
5.6 Verify the asymptotic behaviour of the radial velocity for the isothermal
(Eq. (5.12)) as well as the polytropic (Eq. (5.17)) wind solutions.

5.2.2.2 Equatorial, polytropic winds for rotating stars

The spherically symmetric, transonic wind solutions discussed thus far were purely
1D problems involving a density (hence pressure, due to the isothermal/polytropic
relation) and radial velocity stratification. More realistic models for stellar winds
will need to account for the fact that many stars are rotating (even differentially, i.e.
with different angular velocity at different latitudes, as observed on our Sun), and
harbor magnetic fields (which play a central role in the creation of the hot stellar
coronae, which ultimately drive thermally accelerated outflows). Both rotation and
magnetic fields necessitate the study of multi-dimensional stellar winds, where we
may restrict to cases where the stellar rotation axis and the magnetic pole coincide,
such that we can still assume axisymmetry. If the stellar rotation and magnetic axis
do not coincide (i.e., in essentially all realistic cases), the fully 3D situation must be
considered.
Both rotation as well as magnetic fields add new ways in which the outflow can
be accelerated: besides the thermal drive from the radially decreasing pressure, cen-
5.2 Isothermal and polytropic stellar winds: theory 13

trifugal as well as Lorentz forces introduce (possibly combined) magneto-centrifugal

driving. For fast rotating stars, and stars with strong magnetic fields, magneto-
centrifugal effects may dominate thermal acceleration. In fact, it is important to
note that even for our (slowly rotating, relatively low magnetized) Sun, magneto-
centrifugal effects matter! Indeed, while the solar wind is in effect a thermally driven
outflow (controlling the mass loss rate of the Sun), the magnetic influence on the so-
lar wind is quite important for quantifying long timescale effects related to the Sun’s
angular momentum balance: how fast the solar wind carries off angular momentum,
thereby causing solar rotational spindown [4].
If we momentarily ignore magnetic fields, a first, still essentially 1D generalization
allows to study the effect of additional centrifugal forces on equatorial winds. Indeed
the stationary (∂t = 0) solutions for the equatorial plane θ = π/2, where vθ = 0, and
where we assume axisymmetry ∂ϕ = 0, rewrite the governing equations (5.1) to
 
1 d
r 2 ρv =0


 2
r  dr r

vϕ2
 
 ρ vr dr − r
dvr
= −ρ GrM − dp

dr .
 ∗
 2 (5.18)
  
dvϕ vr vϕ
 ρ vr dr + r =0




In addition to a density and radial velocity, we now need to determine also the
azimuthal velocity introduced by stellar rotation. For these stationary, equatorial
winds, the last equation can be solved directly to give

rvϕ = L , (5.19)

expressing the fact that angular momentum is conserved, introducing the constant
specific angular momentum L. In total, the 3 unknown functions of radius ρ, vr ,
vϕ obey three constants of motion, the specific angular momentum L, the mass flux
r 2 ρvr , and we can once more integrate the r-component of the momentum equation
to the conservation of ‘energy’ expressed by

[vr2 + vϕ2 ] cs2 GM∗ + 2

E=* + − /c . (5.20)
, 2 (γ − 1) r - s0

The same r-component of the momentum equation can be manipulated to

 
vr 2cs − r + vϕ
2 G M∗ 2
dvr
=   , (5.21)
dr r vr2 − cs2

which directly allows to deduce the equation governing the location of the critical
points r s where vr = cs . If we introduce a dimensionless rotational parameter
ζ = Ω∗ r 0 /cs0 which uses the stellar angular rotation rate Ω∗ , this equation is
14 5 Transonic winds

!2 2 r ! 2  (γ − 1)
r0 vesc vr0 r 0 

0
ζ 2
− 2 +2 = 0. (5.22)
rs 2cs0 r s  vr s r s 

As can be seen: for the case where γ = 1, which strictly speaking needs a different
expression for the constant of motion (5.20), this is a second degree polynomial for
critical point radii r s , so that one may have multiple ‘critical’ points.

Fig. 5.5 The radial variation of Mach number for three different rotating polytropic wind solutions,
for increasing rotation rates as quantified by ζ. From [3].

Fig. 5.6 The radial variation of the radial velocity for the fastest rotating wind from Fig. 5.5,
demonstrating the transition from deceleration to acceleration near the base. From [3].
5.2 Isothermal and polytropic stellar winds: theory 15

In a numerical approach to obtaining polytropic wind solutions for rotating stars,

we can simply add the known toroidal velocity profile vϕ (r) = Ω∗ r ∗ rr∗ . The cen-
trifugal force is now entering the radial momentum equation (this comes about from
the inertial term v · ∇v), and the same technique where we relax a solution without
rotation to one where rotation is included can be followed. The result of this, for
three different rotation speeds, is shown in Fig. 5.5. We took a polytropic index
γ = 1.13, and relaxed the spherically symmetric polytropic wind solution to an
equatorial wind for rotation corresponding to ζ = 0.0156 (essentially near-identical
to the non-rotating case, and corresponding to a solar rotation value), ζ = 1 and
ζ = 1.9. Note how the wind achieves a gradually higher radial velocity at the base,
due to the additional centrifugal acceleration. At fixed polytropic index, the sonic
point where M = 1 now moves inwards for increasing ζ (faster rotation). According
to the governing equation (5.22) for the location of the critical points, the ζ = 1.9
solution has a second critical point in the domain at r = 1.087r 0 . To represent the
behavior of the wind accurately, this translates in a need for extreme grid resolu-
tion at the base. This second ‘critical’ point is not a sonic point where M = 1, but
a point where the solution changes from deceleration to acceleration. This is also
shown clearly in Fig. 5.6, this time showing the radial velocity instead of the Mach
number, for the fastest rotating case. In the topological interpretation as a contour
line of the constant E function, this critical point represents an O-type point (local
miminum/maximum), while the sonic point itself is of an X-type topology (saddle
point). The approach used here allows to study thermo-centrifugally driven stellar
winds, by varying the relative influence of rotation to pressure gradient.

5.2.2.3 2D winds for rotating stars

The inclusion of rotation in the previous section was reduced to 1D, by restricting
attention to the equatorial plane. It is of interest to demonstrate the essentially 2D
nature of the stellar wind of a rotating star, and in essence, our numerical approach
easily allows for that. To arrive at a crude model for the coronal expansion of a
rigidly rotating star, where the expansion is due to a combination of thermal and
centrifugal effects, we set forth to construct an axisymmetric, steady-state, polytropic
wind solution valid throughout a poloidal cross-section (i.e. a plane of constant ϕ).
With the polar axis as rotation and symmetry axis, we need to generalize the rotating,
polytropic Parker wind which succesfully modeled the equatorial regions. Whereas
this generalized solution had at least one critical point, its 2D extension is expected
to give rise to critical curves in the poloidal plane. The degree of rotation determines
the deviation from perfect circles for those critical curves. They are perfect circles
(actually spheres) in the non-rotating, spherically symmetric case.
To initialize a 2D time-stepping procedure to arrive at a steady-state wind, we
can use the 1D Parker solution with identical escape speed vesc , polytropic index
γ, and rotational parameter ζ. We may use a spherical (r, θ) grid in the poloidal
plane, where the grid spacing is equidistant in θ, but may be accumulated towards
the base in the radial direction, to capture especially the strong gradients expected
16 5 Transonic winds

near the base. We can take a 300 × 20 grid and only model a quarter of a full poloidal
cross-section, i.e. let θ ∈ [0, π/2].
The density is initialized such that for all angles θ, the radial variation equals
the 1D Parker wind appropriate for the equator. Writing the equatorial Parker
solution as ρ P (r), vrP (r), vϕP (r), we set ρ(r, θ; t = 0) = ρ P (r), and similarly,
we set vr (r, θ; t = 0) = vrP (r). However, for the azimuthal velocity we adopt
vϕ (r, θ; t = 0) = vϕP (r) sin(θ) such that it vanishes at the pole θ = 0. The final
third vθ velocity component is initialized to vθ (t = 0) = 0 everywhere.
Boundary conditions then impose symmetry conditions at the pole (θ = 0) and
the equator (θ = π/2). The radial coordinate covers r ∈ [1, 50]r ∗ , as in the earlier 1D
calculations. Since the solutions are supersonic at r = 50r ∗ , the boundary conditions
there merely extrapolate the density and all three momentum components linearly
in the ghost cells. The stellar rotation enters as a boundary condition in the toroidal
momentum component, which enforces vϕ = Ω∗ R∗ = Ω∗ r ∗ sin θ, where (R, Z )
are the cylindrical coordinates in the poloidal (r, θ) plane. Note that the toroidal
momentum may still change in the process, since we can no longer fix the density at
the stellar surface to a θ-independent constant value. This is because in steady-state,
the density profile ρ(r ∗, θ) should establish a gradient in the θ direction to balance
the component of the centrifugal force in that direction. In the purely radial direction,
the inwards pointing gravity must be balanced by the combination of the pressure
gradient and the radial component of the centrifugal force. We therefore extrapolate
the density linearly at the base. To enforce the total mass flux as in the equatorial
Parker solution, we determine the constant mass flux f mass = ρ P r 2 vrP from the 1D
calculation, and fix ρvR = f mass R/r 3 and ρv Z = f mass Z/r 3 at the stellar surface
for its 2D extension.

Fig. 5.7 2D Polytropic HD winds. We show streamlines and contours (dotted for values below
unity) of the poloidal Mach number M p in the poloidal plane. For low (left panel) and high (right
panel) rotation rates. From [3].

An analytic treatment for a 2D polytropic steady-state wind solution proceeds by

noting that mass conservation is ensured when the poloidal momentum is derived
5.2 Isothermal and polytropic stellar winds: theory 17

from an arbitrary stream function χ(R, Z ) such that ρv p = (1/R)êϕ × ∇ χ. It is then

easily shown that the toroidal momentum equation is equivalent with the existence
of a second arbitrary function L( χ) = Rvϕ , corresponding to the conservation of
specific angular momentum along a poloidal streamline. Note that this is similar to
the 1D results from before, but as will become obvious from our solutions, these
streamlines are no longer purely radially oriented (except of course along the equator,
and along the pole, due to symmetry constraints).
Similarly, energy conservation along a streamline introduces

E( χ) = vR2 + v Z2 + vϕ2 /2 + ργ−1 /(γ − 1) − GM∗ /r.

f g
(5.23)

Across the poloidal streamlines in the poloidal cross-section, all forces must balance
out. We show streamlines and the contours of constant poloidal Mach number Mp =
q
(vR2 + v Z2 )/cs2 for two hydrodynamic wind solutions for vesc = 3.3015cs∗ , γ =
1.13, and with ζ = 0.0156 (left panel) and ζ = 0.3 (right panel) in Fig. 5.7. We
restricted the plotting region to about 10r ∗ . For the imposed mass flux parameter,
we used the values f mass = 0.01377 for the slow rotator and f mass = 0.01553 for
the faster rotator, as found from the equatorial Parker solution for the same γ and ζ.
Note how for low rotation rates, the wind solution is almost spherically symmetric
with nearly radial streamlines and circular Mach curves. For higher rotation rates, the
critical Mach curve where Mp = 1 moves inwards at the equator and outwards at the
pole when compared to a non-rotating case. The streamlines show the equatorward
deflection when material is released from the stellar surface due to the centrifugal
force.

central gravity

centrifugal force

Fig. 5.8 A sketch illustrating the physical reason for equatorward streamline bending: gravity is a
central force, while the centrifugal force is oriented perpendicular to the rotation axis.

The physical reason for this equatorial streamline bending is obvious: while the
gravitational force is central, the centrifugal force is always oriented perpendicular
to the rotation axis (see Fig. 5.8). Along latitudes different from pole or equator, the
component of the centrifugal force in the êθ direction is unbalanced by gravity and the
density/pressure distribution must adjust to realize a true 2D stationary equilibrium.
Above the pole, we still have a situation where only gravity and pressure gradient
18 5 Transonic winds

compete (as in the 1D spherically symmetric case for non-rotating star), while at the
equator, the three forces are all radial and represent a solution where gravity, centrifu-
gal and pressure gradient balance (as in the equatorial wind discussed previously).
Throughout the poloidal cross-section, density readjusts to realize equilibrium, and
we find a wind-induced density enhancement near the equatorial regions, forming
a kind of wind-fed circumstellar disk. For these numerical steady-state solutions,
we can then verify that the specific angular momentum L, mass flux, as well as the
total energy E, are conserved along the streamlines, and thus in that way we can
quantify any remaining numerical errors. This can be used to decide when to stop
the (pseudo) time stepping and call the endresult a converged steady-state solution.

5.3 Handling orthogonal coordinate systems∗∗

As became obvious in this chapter on stellar winds, when a central force field like
gravity is involved, it makes sense to solve the governing equations in other than
Cartesian coordinates. In particular, we mentioned the use of two orthogonal curvi-
linear coordinate systems: cylindrical (R, Z, ϕ) or spherical (r, θ, ϕ). The analytic
form of the equations in those coordinates, when exploiting the corresponding vector
components, must be aware of the divergence expressions given by Eqns. (5.3)-(5.4).
For those special cases of orthogonal coordinates, we will now discuss some aspects
that enter a numerical treatment of the governing equations, when handling (near or
exact) conservation laws in a finite volume treatment. These ideas can be generalized
to any orthogonal curvilinear coordinate system (e.g. bipolar cylindrical or bipolar
spherical coordinate systems, which may be beneficial to solve stellar wind aspects
in binary systems).

5.3.1 Geometrical source terms∗∗

We discussed in chapter ?? the finite volume method, and that a generic system
of conservation laws, extended with source terms, works with volume-averaged
conserved quantities Ūi as its unknowns, and uses the rewritten form (??), repeated
here for convenience:
Z
1 X 1
∂t Ūi + Fl · nl dS l = Sdx . (5.24)
Vi l Vi Vi

In practice, Eq. (5.24) implies that we must evaluate physical fluxes at cell edges
(Fl ), where we have normals nl and surfaces dSl , while the cell itself has a volume
Vi . The right hand side gives the contribution from a physical source term in the
equations, such as the one resulting from an external central gravitational field.
5.3 Handling orthogonal coordinate systems∗∗ 19

In a 3D cylindrical (R, Z, ϕ) coordinate system, let us consider a volume element

that is centered on a cell location where R = Ri , Z = Z j and ϕ = ϕk , where i, j, k are
the indices that identify an individual cell center point in a regular grid, exploiting
∆R, ∆Z and ∆ϕ spacings. Since the analytical volume element for a cylindrical
coordinate system is given by R dR dZ dϕ, the volumes we need to use in our finite
volume approach are
Z Ri +∆R/2 Z Z j +∆Z/2 Z ϕ k +∆ϕ/2
Vi jk = R dR dZ dϕ
Ri −∆R/2 Z j −∆Z/2 ϕ k −∆ϕ/2

R2 Ri +∆R/2
" #
= ∆ϕ ∆Z |
2 Ri −∆R/2
= Ri ∆R ∆Z ∆ϕ . (5.25)

Hence, this is indeed directly corresponding with the analytical R dR dZ dϕ. The
normals on the grid cell faces are, for this orthogonal cylindrical system, just given
by ±êR , ±ê Z and ±êϕ . The surfaces dSl can also be found easily, and they are
!
∆R
radial direction êR : dSR = Ri ± ∆Z∆ϕ
2
vertical direction ê Z : dSZ = Ri ∆R∆ϕ
azimuthal direction êϕ : dSϕ = ∆R∆Z . (5.26)

This all means that when we do the replacement

1 X
∇·F⇒ Fl · nl dS l (5.27)
Vi l

and the flux F is a second rank tensor as is the case for the momentum equation in
Eq. (5.1), we actually obtain for the divergence as
(" #
1 i+ 1 ∆R i− 21 ∆R
F∗R 2 (Ri + )∆ϕ∆Z − F∗R (Ri − )∆ϕ∆Z
Ri ∆R ∆Z ∆ϕ 2 2
j+ 1 j− 1
 
+ F∗Z 2 Ri ∆R∆ϕ − F∗Z 2 Ri ∆R∆ϕ
k+ 1 k− 1
 
+ F∗ϕ 2 ∆R∆Z − F∗ϕ 2 ∆R∆Z . (5.28)

In this expression, the ∗ would select either the ∗ = R, ∗ = Z or ∗ = ϕ component

for the momentum equation. It is easy to see that the above expression is a numerical
proxy for
1 1
≈ ∂R (RF∗R ) + ∂Z F∗Z + ∂ϕ F∗ϕ , (5.29)
R R
which can be recognized to correspond exactly to a divergence for a vector as in
Eq. (5.3). However, for the momentum equation, the flux is actually a second rank
tensor, so the correct formula to use should approximate Eq. (5.4). In practice, we
20 5 Transonic winds

conclude that so-called geometrical source terms must be added to two components
of the momentum equation, namely one for the R-component taking account of the
−Fϕϕ /R term, and one for the azimuthal component for the FϕR /R term, in accord
with Eq. (5.4). These geometric source terms are then for our purposes added during
a timestep ∆t as follows

n + ∆t ρvϕ +p
 2 
m n+1
R = m R R
f ρvR vϕ g
mϕn+1 = mϕn + ∆t − R . (5.30)

Note that these geometrical source terms are to be distinguished from the physical
source terms (like gravity), as they arise purely due to our choice of a non-Cartesian,
orthogonal coordinate system.

Exercise
5.7 As an exercise, derive the expressions for the volumes and surface elements
for a spherical coordinate system. You should get

(∆r) 2
" #
∆θ
Vi jk = ∆ϕ∆r r i +
2
2 sin θ j sin (5.31)
12 2

which is seen to approximate ≈ r 2 sin θ dr dθ dϕ, and

!2
∆r
radial direction êr : dSr = r i ± sin θ j ∆θ∆ϕ
2
∆θ
polar direction êθ : dSθ = r i sin(θ j ± )∆r∆ϕ
2
azimuthal direction êϕ : dSϕ = r i ∆θ∆r . (5.32)

Exercise
5.8 Using the result of the previous exercise, identify the following geometrical
source terms for solving system (5.1) in a spherical coordinate system.
ρvθ +ρvϕ +2p
 2 2 
mrn+1 = mrn + ∆t r
(ρv 2 +p) cot θ
 
mθn+1 = mθn + ∆t − ρvrr vθ + ϕ r
f ρv v ρv v cot θ g
mϕn+1 = mϕn + ∆t − rr ϕ − θ ϕr . (5.33)
5.3 Handling orthogonal coordinate systems∗∗ 21

Exercise
5.9 Using the result of the previous exercise, and realizing that a 2D polar grid
(r, ϕ) is a special case of the spherical grid, verify the source terms mentioned
in section ?? for the shallow water equations.

5.3.2 Boundary conditions at a symmetry axis∗∗

When we employ cylindrical (R, Z, ϕ), spherical (r, θ, ϕ), or 2D polar (r, ϕ) coordi-
nates, our computational domain may extend to include a symmetry or polar axis. In
that case, the boundary conditions on some or all of the variables are to be chosen
in accord with this symmetry. Quite a few possibilities can occur, depending on the
dimensionality of the problem (i.e. are we solving a 2D, 2.5D or 3D variant) and
the symmetry encountered, with axisymmetry where ∂ϕ = 0 for a cylindrical or
spherical setup being the most relevant case.
In the discussion of stationary, transonic stellar wind solutions, we encountered
spherically symmetric 1D to axisymmetric 2.5D solutions. There, we worked in
spherical coordinates and used corresponding vector components vr (r, θ), vθ (r, θ)
and vϕ (r, θ), which reduces to vr (r) in a spherically symmetric 1D wind. In the
2.5D case, when θ ranges from [0, π] both θ = 0 and θ = π correspond to a polar
symmetry axis when ∂ϕ = 0 is assumed. If we only simulate a quarter of the
meridional plane, i.e. when θ ∈ [0, π/2], the θ = 0 is still a polar symmetry axis,
while the equator at θ = π/2 can represent an equatorial symmetry where we may
want to impose an up-down symmetry in the meridional cross-section. The same
polar symmetry and/or equatorial symmetry conditions can be encountered in a
2.5D cylindrical setup where we have axisymmetry ∂ϕ = 0, and where we work with
vR (R, Z ), v Z (R, Z ) and vϕ (R, Z ) components. This will be the case when the domain
extends from R ∈ [0, Rmax ] with a polar symmetry at R = 0, while Z ∈ [Zmin, Zmax ].
An equatorial symmetry in that 2.5D cylindrical case may conveniently place the
equator at Z = 0, and then impose the up-down symmetry there. In Table 5.1,
the various boundary conditions to be imposed on the vector components at either
polar or equator symmetry axes are listed, where we distinguish symmetric versus
asymmetric choices. A symmetric boundary condition employs a mirror symmetry
to fill the ghost cells, while an asymmetric one combines this mirror symmetry
with a change of sign. These two types, along with a Neumann boundary where a
continuous extrapolation (cont) is used into the ghost cells, are illustrated in Fig. 5.9.
If however full 3D solutions are to be computed on cylindrical or spherical grids
(or 2D planar problems on a 2D polar grid), we should not impose the same kind
of symmetry conditions at the polar axis as we do for the axisymmetric cases. Note
that in contrast to this polar axis, we may still insist on up-down symmetry across
an equatorial plane, which does remain the same. The pole at R = 0 or θ = 0, π
22 5 Transonic winds

Table 5.1 Boundary conditions in axisymmetry ∂ϕ = 0, in spherical or cylindrical setups.

2.5D spherical
pole ρ vr vθ vϕ
symmetric symmetric asymmetric asymmetric
equator ρ vr vθ vϕ
symmetric symmetric asymmetric symmetric
2.5D cylindrical
pole ρ vR vZ vϕ
symmetric asymmetric symmetric asymmetric
equator ρ vR vZ vϕ
symmetric symmetric asymmetric symmetric

Fig. 5.9 The filling of ghost cell values (in red) from the first two inner grid cell values (black
circles), taken to contain the values 1 and 3. The ghost cells values differ according to (1) a
continuous extrapolation (i.e. a Nuemann boundary condition with zero gradient) at left; (2) an
asymmetric prescription in the middle panel; and a (3) symmetric prescription at right.

is special in the full 3D case, since it is a kind of singular axis where some of
the geometric factors become ill-defined, due to divisions by R or the occurence
of sin θ factors in the divergence expressions (5.3). Of course, the actual volumes
Vi jk and surface elements dSl we use never really become ill-defined, since the
volumes will be based on cell-centered coordinates, while the surface elements may
become zero, hence naturally nullifying the flux contribution over this pole in our
finite volume evaluations. In fact, the polar axis is then to be treated with a special
type of periodicity, called π-periodicity [5], which just recognizes the fact that it
should be allowed to displace gas across the pole, but only in a manner which knows
about the local singular grid structure there. The same π-periodic periodicity is to
be imposed in a 2D planar problem that is computed on a polar (r, ϕ) grid, when the
r = 0 pole is included in the domain. On a complete circle where r ∈ [0, r max ] while
ϕ = [0, 2π], the use of polar components vr (r, ϕ) and vϕ (r, ϕ) means that at r = 0, the
5.3 Handling orthogonal coordinate systems∗∗ 23

π-periodicity must be imposed, while the ϕ-direction has the usual 2π-periodicity.
This is illustrated in Fig. 5.10 for a polar grid that employs 12 angular and 5 radial
grid points to cover the full circle: the red cells illustrate a π-periodic relation across
the pole, while the blue grid cells must use a 2π-periodic relation across the ϕ = 0
direction. Note that the angular direction then always needs an even amount of grid
cells. A summary of the various possibilities, both 2D and 3D, that encounter a
pole-periodic aspect is given in Table 5.2. The outer (or innermost when r min , 0)
radial boundary conditions are to be chosen in accord with the problem at hand, and
could use a Neumann boundary condition (indicated with cont), or one where all
flow across the boundary is either outgoing or vanishes (indicated with noinflow),
or any other kind of condition that follows from the problem under consideration
(such as fixing the mass flux at a spherical inner boundary r = r min , 0).

Table 5.2 Boundary conditions for 2D polar or 3D spherical or cylindrical setups.

2D polar r =0 r = rmax φ ∈ [0, 2π] –
π-periodic noinflow or cont periodic –
3D cylindrical R=0 R = Rmax Z ∈ [Zmin, Zmax ] ϕ ∈ [0, 2π]
π-periodic noinflow or cont noinflow or cont periodic
3D spherical r ∈ [rmin, rmax ] θ=0 θ=π ϕ ∈ [0, 2π]
fixing mass flux, cont, . . . π-periodic π-periodic periodic

Fig. 5.10 When a planar 2D polar grid uses 5 radial and 12 angular cells, a finite volume compu-
tation will store all values at the cell centers (indicated by black dots) of the 5 × 12 grid in (r, ϕ)
shown by the grey lines. The polar axis at r = 0 needs to employ a π-periodicity across the pole,
shown by the red-colored cells, and a 2π-periodicity across the ϕ = 0 direction, as shown by the
blue-colored cells. The number of cells in the ϕ direction must necessarily be even.
24 5 Transonic winds

5.4 Inertial versus non-inertial reference frames∗∗

When discussing extensions of the isothermal Parker wind, we ended up showing

stationary and transonic hydrodynamic stellar winds for rotating stars, assuming a
polytropic closure relation. These wind solutions were solving Eqns. (5.1) for the
region outside the stellar surface r = r ∗ , and the fact that the star was assumed to
rotate rigidly with angular frequency Ω∗ never really entered the governing equations,
except for prescribing the initial conditions on the vϕ profile and when imposing the
boundary conditions at the stellar surface. This is because we chose to solve the
equations in an inertial frame centered on the star, in which the star itself is rotating.
That means that the base velocity at (or just above) the stellar surface inherits the
rigid rotation from the star as matter is released into the atmosphere. The initial
profile for the azimuthal flow vϕ (r) at the equator is then ∝ 1/r due to the constancy
of angular momentum, while we have a fixed base value of vϕ (r ∗ ) = Ω∗ r ∗ . In a 2D
setting, we similarly enforced vϕ (r ∗, θ) = Ω∗ R∗ = Ω∗ r ∗ sin θ.
Now, it is also possible to analyze and solve the governing equations in the frame
rotating along with the star, where then the obvious boundary prescription for the
azimuthal velocity component is much simpler, namely vϕrot = 0. As we then are
no longer in an inertial reference frame, getting the physically correct solution then
entails to account for the fictitious forces introduced by the uniformly rotating frame
of reference. It is well-known that the vector relations for the velocity and acceleration
in both frames are related by

v = vrot + Ω × r ,
a = arot + 2Ω }+Ω
× vrot
| {z × (Ω × r) , (5.34)
| {z }
Coriolis centrifugal

where the additional acceleration terms introduce the fictitious Coriolis and cen-
trifugal accelerations3. The vector Ω has a magnitude corresponding to the stellar
(uniform) rotation frequency Ω∗ , and is oriented along the rotation axis. The mo-
mentum equation, written in non-conservative form where it is showing the direct
correspondence with Newton’s law ma = F, is then to be modified for the rotating
frame to yield

ρ [∂t v + (v · ∇)v] = −∇p + ρg − ρΩ × (Ω × r) − 2ρΩ × v . (5.35)

The mass conservation equation in the rotating reference frame keeps its original
shape, namely
∂t ρ + ∇ · ( ρv) = 0 . (5.36)
This means that when we combine these into the conservation form for ρ and m = ρv,
switching our viewpoint from an inertial frame to a rotating refernce frame just needs
to add the two fictitious forces as source terms, treated in the same fashion as the
gravity term. In a cylindrical coordinate system, where Ω = Ω∗ ê Z and r = RêR + Z ê Z

3 When the rotation were to change in time, an additional Euler force Ω̇ × r would enter as well.
5.4 Inertial versus non-inertial reference frames∗∗ 25

we can find easily that

f g
−ρΩ × (Ω × r) − 2ρΩ × v = êR ρΩ2∗ R + 2ρΩ∗ vϕ + êϕ −2ρΩ∗ vR .
 
(5.37)

These are then the source terms that appear explicitly in the equations, when we opt
for this non-inertial reference frame. In that frame, we must then initialize the az-
imuthal velocity with zero values, and similarly for its stellar boundary prescription.

Exercise
5.10 Derive the corresponding expressions for the fictitious forces, when pre-
ferring to use a spherical coordinate system with corresponding vector com-
ponents. Use the fact that r = r êr , that Ω × r = Ω∗ Rêϕ = Ω∗ r sin θ êϕ and
v = vr êr + vθ êθ + vϕ êϕ while still Ω = Ωê Z . You could also directly use
the known relationships between unit vectors êR, ê Z , êϕ and êr , êθ , êϕ and the
corresponding relations between vR, v Z , vϕ and vr , vθ , vϕ .
26 5 Transonic winds

References

1. E.N. Parker, ApJ 128, 664 (1958). DOI 10.1086/146579

2. H. Bondi, MNRAS 112, 195 (1952). DOI 10.1093/mnras/112.2.195
3. R. Keppens, J.P. Goedbloed, Astronomy & Astrophysics 343, 251 (1999)
4. R. Keppens, K.B. MacGregor, P. Charbonneau, Astronomy & Astrophysics 294, 469 (1995)
5. B. van der Holst, R. Keppens, Journal of Computational Physics 226(1), 925 (2007). DOI
10.1016/j.jcp.2007.05.007
Index

axisymmetric 2.5D, 3 geometrical source terms, 20

gravitational source term, 1
boundary conditions
π-periodic, 22 Parker wind, 2, 4
cont, 21 Picard iteration, 9
noinflow, 23
rotating reference frame, 24
code units, 6
critical or sonic point, 5 solar breeze, 8
curvilinear coordinate systems, 18 solar wind, 8
stream function, 17
divergence of tensor, 3 symmetry boundary conditions, 21
divergence of vector, 3
termination shock, 9
Eugene Parker, 2 thermo-centrifugally driven winds, 15

fictitious forces, 24 wind-fed disks, 18

27