Stellar Stability and Asteroseismology: R. Scuflaire, A. Thoul

Belgian Asteroseismology Group
STELLAR STABILITY
AND ASTEROSEISMOLOGY
R. Scuflaire, A. Thoul
Institute of Astrophysics and Geophysics

University of Liège
September 2002
i
Contents
1 Introduction 1
2 Characteristic timescales 4
2.1 The dynamical timescale . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Pulsation timescale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 The Kelvin-Helmholtz timescale . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 The nuclear timescale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 General equations 8
3.1 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Momentum equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Poisson equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4 Energy conservation equation . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.5 Transport equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.6 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.7 Material equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4 Equilibrium configuration 12
5 Small perturbations method 14

5.1 Lagrangian and eulerian perturbations . . . . . . . . . . . . . . . . . . . . 15
5.2 Perturbation of the differential equations . . . . . . . . . . . . . . . . . . . 16
5.3 Perturbation of the material equations . . . . . . . . . . . . . . . . . . . . 16
6 Adiabatic perturbations 19
7 Radial oscillations 23
7.1 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.2 Integral equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.3 Dynamical modes and secular modes . . . . . . . . . . . . . . . . . . . . . 27
ii
8 Adiabatic radial oscillations 29

8.1 Energy of a mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8.2 Behavior of the eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . 34
8.3 Some unstable cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
9 Asymptotic expression for the radial frequencies 38

9.1 Central singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.2 Surface singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.3 Joining of the two solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.4 Brief reminder on Bessel functions . . . . . . . . . . . . . . . . . . . . . . . 40
10 Vibrational stability 42
10.1 The quasi-adiabatic approximation . . . . . . . . . . . . . . . . . . . . . . 43
10.2 The nuclear excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
10.3 The influence of the transport terms . . . . . . . . . . . . . . . . . . . . . 46
10.4 Strange modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
11 The pulsation mechanism in the instability strip and the light phase lag 49
11.1 The existence of an instability . . . . . . . . . . . . . . . . . . . . . . . . . 49
11.2 The light phase lag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
12 Non radial oscillations 55

12.1 Spherical coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
12.2 Perturbation equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
12.3 The Cowling approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 60
13 Non radial modes 62

13.1 Orthogonality of the eigenfunctions . . . . . . . . . . . . . . . . . . . . . . 62
13.2 Components of the displacement . . . . . . . . . . . . . . . . . . . . . . . . 62
13.3 p, g and f modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
13.4 Spheroidal and toroidal modes . . . . . . . . . . . . . . . . . . . . . . . . . 65
13.5 Asymptotic expression for the frequencies . . . . . . . . . . . . . . . . . . . 69
14 Influence of rotation 72
14.1 Non radial oscillations in variable stars . . . . . . . . . . . . . . . . . . . . 74
iii
15 Helio- and asteroseismology 76

15.1 Determination of Ω(r) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
15.2 Determination of the solar structure . . . . . . . . . . . . . . . . . . . . . . 78
15.3 Non adiabatic asteroseismology . . . . . . . . . . . . . . . . . . . . . . . . 79
1
Chapter 1
Introduction
It is usual, when studying a physical system, to first look at its equilibrium configurations.
This approach is obviously justified in the case of stars. Stellar evolution studies teach us
that during the major part of their lives stars are equilibrium structures which evolve very
slowly under the effects of the changes of their chemical composition. It is then important
to find out whether those equilibrium structures are stable or not. The study of stellar
stability is therefore obviously complementary to the study of stellar structure.
When the equilibrium structure is unstable, we will study the development of the insta-
bility: time needed for its growth, and how it will appear. Very often this instability will
appear as some kind of variability occuring on a timescale much smaller than the charac-
teristic evolution timescale. The fraction of variable stars is small, but they are very much
studied. The role played by RR Lyr stars and Cepeids to estimate astronomical distances
is well known. In addition, the study of stellar oscillations provides information about
internal stellar structure features which are not directly observable and allows therefore a
test of the theory of stellar evolution.
The increase in the precision and in the time resolution of the observational instruments
allows the observation of an increasing number of oscillations. Thousands of modes with
periods close to 5 minutes (frequencies close to 3 mHz) have been identified on the Sun
and their frequencies measured with an accuracy of the order of one µHz. The study of
these data to determine the internal structure of the Sun is called helioseismology. The
same technique applied to other stars is called asteroseismology.
Here we will only study gaseous stars. We will exclude the cases where the star or parts
of it are in a state comparable to a solid state (white dwarfs under certain conditions,
neutron stars). We will place ourselves in the context of non relativistic mechanics and the
newtonian theory of gravitation, excluding the cases where it is justified to use a relativistic
treatment (very condensed white dwarfs, neutron stars, black holes, supermassive stars).
It is well known that a mechanical system is stable if its equilibrium configuration cor-
responds to a minimum of the potential. This property can be extended to the case of
a stellar configuration as long as only adiabatic transformations are considered. The ex-
tention of the method is not straightforward if non adiabatic terms are included. This
method, based on the energy, has been rarely used and will not be presented in these
2
Figure 1.1: HR diagram with some types of variable stars.
lectures. In fact, a large fraction of these lectures will be dedicated to the small pertur-
bations method. This one is easy to use, the properties of the solutions are quite well
understood, and it is sufficient to explain most of the stellar variability linked to stellar
stability. However, it is necessary to use a nonlinear theory to explain some behaviors
(oscillations amplitudes, chaos, . . . ).
We have approximately placed in an HR diagram some types of variable stars which will
be refered to in these lectures (figure 1.1). We refer the reader to a textbook on variable
stars for a detailed description of these different types of stars. We stress here that the
δ Sct, RR lyr, δ Cep and W Vir variables lie in a region of the HR diagram called the
instability strip.
References
Papers by Ledoux and Walraven (1958), Ledoux (1969), Cox (1974), Cox (1980), Unno et
al. (1989) are classic references on this topic. A recent and short survey of the theory can
be found in the papers by Gautschy and Saio (1995 and 1996). The energy-based method
is descibed in a paper by Ledoux (1958).
Cox J.P., 1974. Pulsating stars. Rep Prog Phys, 37, 563–698.
Cox J.P., 1980. Theory of stellar pulsation. Princeton University Press.
Gautschy A., Saio H., 1995. Stellar pulsations across the HR diagram: part 1. Ann Rev
Astron Astrophys, 33, 75–113.
3
Gautschy A., Saio H., 1996. Stellar pulsations across the HR diagram: part 2. Ann Rev
Ledoux P., 1958. Stellar stability. In Handbuch der Phys, vol 51, 605–688, edit. Flügge
S., Springer.
Ledoux P., 1969. Oscillations et stabilité stellaires. In La structure interne des étoiles,
11ème cours de perfectionnement de l’association vaudoise des chercheurs en physique,
45–211, edit. Joseph C., Janin G., Maeder A., Mayor M., Saas-Fee.
Ledoux P., Walraven T., 1958. Variable stars. In Handbuch der Phys, vol 51, 353–604,
edit. Flügge S., Springer.
Unno W., Osaki Y., Ando H., Saio H., Shibahashi H., 1989. Nonradial oscillations of
stars, 2nd edit. University of Tokyo press.
4
Chapter 2
Characteristic timescales
The different physical mechanisms operating in a star have extremely different timescales.
It is important to know these timescales in order to model these mechanisms and use
adequate approximations, or to choose correctly the integration timesteps in a numerical
approach. We will give rough estimates for these timescales.
2.1 The dynamical timescale

It is well known that the hydrostatic equilibrium of a star results from the competition
between gravity and pressure gradient forces. If the equilibrium between those two forces
was broken, the star would respond on a timescale called the dynamical timescale. To
estimate this timescale, assume that the pressure forces suddenly disappear and cease to
oppose the gravity forces. Consider a unit mass element of matter. It would be subjected
to the gravity force only and would undergo a freefall. Its motion would obey the equation
Gm
r̈ = − .
r2
Let τf f be the characteristic timescale of this freefall. Using very rough estimates in the
last equation, we have
R GM
2
≈ 2 ,
τf f R
which gives q
τf f ≈ R3 /GM .
It is also possible to estimate the timescale of the star reaction in response to the loss of
hydrostatic equilibrium by assuming now that the gravity forces suddenly cease to exist.
The pressure forces by themselves would dislocate the star. An element of matter would
be accelerated outwards following the equation
1 dP
r̈ = − .
ρ dr
5
Let τexpl be a characteristic timescale of this explosion. We can get a rough estimate of it
by writing
R P c2
2
≈ ≈
τexpl ρR R
q
where c = Γ1 P/ρ is a characteristic sound speed in the star. We get
R
τexpl ≈ .
c
Because of the hydrostatic equilibrium, these two timescales are of the same order and
define the dynamical timescale τdyn
s
R3 1
τdyn ≈ ≈√
GM Gρ
where ρ is the mean stellar density. By comparing these two expressions we get an estimate
of the sound speed in the star s
GM
c≈ .
R
The table below gives a few typical values of the dynamical timescales in some types of
stars. √
Star ρ (g cm−3 ) τdyn = 1/ Gρ
neutron stars 1015 0.12 ms
white dwarfs 106 3.9 s
Sun 1.41 54 min
red supergiant 10−9 3.9 yrs
Exercise
Show that the dynamical timescale also characterizes the circling motion of a satellite
in a low orbit, as well as the rapid rotation of a star at the limit of disruption due to
centrifugal forces.
2.2 Pulsation timescale

It is necessary to be careful when defining a characteristic pulsation timescale. We will
see that a star has an infinity of pulsation modes, at all timescales. This should not be
surprising: a simple vibrating string has a fundamental mode and an infinity of harmonics,
whose periods tend to zero. For the most characteristic variables (cepheids, RR Lyr) the
observed pulsation is in fact a radial acoustic mode of low order (fundamental or first
harmonic). The estimate we will establish concerns this type of pulsation. Since it is a
pressure mode (acoustic wave) we get an estimate for the pulsation characteristic timescale
τpuls ≈ R/c.
6
The characteristic timescale obtained like that is again the dynamical timescale
q
τpuls ≈ τdyn ≈ 1/ Gρ.
√
The product Period × Gρ is therefore a dimensionless number of order unity, relatively
independent of the stellar model. For the Sun (period of the fundamental mode: 63
minutes), it is 1.6. It is common to use the pulsation constant Q which is proportional to
it and defined by s
ρ
Q = Period × .
ρ¯
Q has the dimensions of a time and detailed calculations show that as a rule we have
0.03 days ≤ Q ≤ 0.08 days.
In some unusual circumstances the estimate of the fundamental mode made above can be
completely wrong (this is the case for a model close to dynamical instability).
2.3 The Kelvin-Helmholtz timescale

When a star slowly burns a nuclear fuel, the thermal equilibrium is the result of the
competition between the rate of energy production and the rate of radiative energy loss.
We get an estimate of the characteristic timescale of the thermal processes in the star
by assuming that it suddenly loses its source of nuclear energy. It should then take the
energy it radiates from its total energy E. The characteristic timescale of this process is
called the Kelvin-Helmholz timescale. It is given by
τKH ≈ |E|/L.
In usual stellar conditions, the virial theorem provides a relationship between the total
energy of the star and its potential energy Ω:
1
E = Ω.
2
On the other hand Ω is given by

GM 2
Ω = −q
R
where q is a factor close to unity, as shown in the following table
Model q
homegeneous model 0.6
polytrope of index n 3/(5 − n)
main sequence model 1.5
7
Thus we can write

GM 2
τKH ≈ .
LR
For the Sun, this expression gives 3.1 × 107 yrs.
It is interesting to compare the dynamical timescale to the thermal timescale
τdyn LR5/2
≈ 3/2 5/2 .
τKH G M
This ratio is much smaller than unity. For the Sun it is equal to 1.6 × 10−12 .
The comparison of these two timescales show that the thermal processes are very slow
compared to the dynamical processes. In first approximation, it is therefore justified to
ignore them when studying dynamical processes. These are global estimates, however, and
locally thermal processes can have much smaller characteristic timescales, comparable to
the dynamical timescale (e.g. in the external layers of the star).
2.4 The nuclear timescale

Consider a star on the main sequence, centrally burning its hydrogen, and try to estimate
the time needed to significantly change its chemical composition. One gram of hydrogen
liberates an energy of the order of 0.007c2 ≈ 6 × 1018 ergs. If about 10 % of the stellar
mass undergoes fusion, we get a nuclear lifetime of
M
τnuc ≈ 6 × 1017 (CGS).
L
For the Sun, this gives 9.8 × 109 years. The ratio of the Kelvin-Helmholtz timescale to
the characteristic nuclear timescale is given by
τKH M
≈ 1.11 × 10−25 (CGS).
τnuc R
This ratio is usually small. For the Sun it is equal to 3.2 × 10−3 . For a main sequence
star, it is therefore justified to neglect the variations in the chemical composition when
studying thermal processes.
8
Chapter 3
General equations
The study of stellar stability relies on the same physics as the construction of equilibrium
models: hydrodynamics, radiative transport theory, thermodynamics, nuclear reactions
theory, etc.
Two methods are commonly used to describe the motion of a fluid: the lagrangian and
the eulerian methods. In the lagrangian method, each fluid particle is assigned a label and
followed in its motion as in classical mechanics. The particle label could be for example
its initial position ~r0 . We will more generally use some vector ~a as label. The fluid is then
described by the functions ~r(~a, t), ρ(~a, t), P (~a, t), . . ..
In the eulerian description, particles are not followed one by one. Rather, at each position
~r the fluid is described by the functions ~v (~r, t), ρ(~r, t), P (~r, t), . . .
The risk of confusion between the two descriptions comes from the fact that the same
symbol, for example ρ, is used for two different functions, ρ(~a, t) and ρ(~r, t).
In particular, the time derivatives in these two formalisms do not have the same meaning:
in the lagrangian formalism ∂/∂t is the time derivative following the motion of the fluid,
while in the eulerian formalism this symbol represents the time derivative at a given point.
Therefore we have
∂~r
= ~v in the Lagrange formalism,
∂t
∂~r
= 0 in the Euler formalism.
∂t
In the eulerian formalism we introduce a differential operator called the Stokes derivative,
or derivative with respect to t following the fluid: D/Dt or d/dt. It is defined as follows:
dX ∂X
= + ~v · grad X.
dt ∂t
It is clear that Ã !
∂X dX
= .
∂t Lagrange
dt
9
In practice, the two formalisms are often used simultaneously. To prevent any confusion
between (∂/∂t)Euler and (∂/∂t)Lagrange we only use the notation ∂/∂t for (∂/∂t)Euler and
we use d/dt for the operator (∂/∂t)Lagrange . This convention is in agreement with the
relationship between the Stokes derivative and the lagrangian time derivative, as seen
above.
3.1 Continuity equation

The continuity equation represents the mass conservation. It can be written as
∂ρ dρ
+ div(ρ~v ) = 0 or + ρ div ~v = 0.
∂t dt
In the lagrangian formalism it can also be written as
ρX = ρ0 X0 ,
where X is the jacobian determinant

¯ ¯
¯ ∂(x) ¯
¯ ¯
X=¯ ¯.
¯ ∂(a) ¯
We easily recover the first form of the equation if we note that

dX
= X div ~v .
dt
3.2 Momentum equation

The momentum equation is given by
∂~v 1
+ (~v · grad)~v = − grad Φ − grad P
∂t ρ
d~v 1
or = − grad Φ − grad P.
dt ρ
We did not include viscosity terms. Indeed in a star the effects of molecular viscosity are
totally negligible and the fluid can be considered as perfect. In the convective zones, the
velocity in the momentum equation is the bulk motion velocity, obtained by averaging over
regions larger than the convective cells. The presence of convective motions is described by
additional pressure and turbulent viscosity terms in the momentum equation. Contrary
to molecular viscosity terms, they are usually not negligible. Unfortunately there is no
good theory of non stationary convection. In this course, we will not consider the problem
of non stationary convection. We will use the equations for the radiative zones.
10
3.3 Poisson equation

The gravitational potential field Φ obeys the Poisson equation
∆Φ = 4πGρ.
The physical solution is regular at infinity and is given by
Z
ρQ dVQ
Φ(P ) = −G .
|P Q|
3.4 Energy conservation equation

The energy conservation equation is written as
Ã !
∂S 1
T + ~v · grad S = ²− div F~
∂t ρ
dS 1
or T = ²− div F~ ,
dt ρ
where S is the entropy per unit mass, ² is the rate of nuclear energy generation per
unit mass and F~ is the energy flux density. Remember that the amount of energy going
−
→ −
→
through a surface dS per unit time is given by F~ · dS where dS = ~ndS, ~n being the unit
vector normal to dS. In general F~ is the sum of a radiative term and a convective term
F~ = F~R + F~C .
3.5 Transport equation

In stellar interiors, the photon mean free path is extremely short compared to other
characteristic length scales. The diffusion equation is therefore an excellent approximation
of the radiative transport equation,
4acT 3
F~R = −λ grad T with λ= .
3κρ
We note that the conduction flux in white dwarfs is given by a similar equation (with a
different conduction coefficient λ). We will not give here the expression for the convective
flux, for reasons given above.
3.6 Boundary conditions

Adequate boundary conditions must be imposed at the stellar surface. To zeroth order
we can impose zero pressure and zero temperature at the surface of the star. To the
next order, we consider that the photosphere (optical depth τ = 2/3) defines the limit
between the star interior and its atmosphere. We impose continuity at this limit. This
separation between interior modelling and atmosphere modelling is useful because they
deal differently with the transport equation.
11
3.7 Material equations

In addition to the partial differential equations and the associated boundary conditions
described above, we also need the equations describing the behavior of the matter as a
function of its chemical composition and the thermodynamic variables. These equations
are sometimes called constitutive equations or material equations. They include the equa-
tion of state, the opacity κ and the nuclear energy generation rate ². We will use the
symbol χ to describe the chemical composition, and we will use ρ and T as independent
thermodynamic variables. The properties of the matter can be decribed by relations of
the form
P = P (ρ, T, χ), U = U (ρ, T, χ), κ = κ(ρ, T, χ), ² = ²(ρ, T, χ).
Note that the nuclear energy generation rate is usually the result of many nuclear reactions
whose rates depend on elements with very small abundances, very short lifetimes (on
stellar evolution timescales) and not described by χ. It is only when these elements reach
their equilibrium abundance values that ² can be considered a function of ρ, T et χ.
12
Chapter 4
Equilibrium configuration
A spherically symmetric configuration obeys the following equations, obtained from the
general equations.
1 dP dΦ
+ = 0,
ρ dr dr
Ã !
1 d 2 dΦ
r = 4πGρ,
r2 dr dr
1 d ³ 2 ´
²− 2 r F = 0,
ρr dr
4acT 3 dT
F =− (radiative zone).
3κρ dr
The condition for stability of a radiative zone against convection is given by the Schwarzschild
criterion
d ln ρ 1 d ln P
A≡ − < 0.
dr Γ1 dr
Let m(r) be the mass of the sphere of radius r and L(r) its luminosity.
Z r
m= 4πr2 ρ dr ,
0
L = 4πr2 F .
The above equations can then be written under their usual form:
dP Gmρ
= − ,
dr r2
dm
= 4πr2 ρ ,
dr
dL
= 4πr2 ρ² ,
dr
dT 3κρL
= − (radiative zone).
dr 16πr2 acT 3
13
We note that in some evolution phases, the models are not in thermal equilibrium and
there is a term involving the derivative of the entropy in the energy conservation equation.
In what follows we will not consider out-of-thermal-equilibrium models.
We must add the boundary conditions to these equations. We must impose natural bound-
ary conditions at the surface of the star and artificial boundary conditions at the center
(because the singularity comes from the spherical coordinates themselves). At the center,
m and L must be zero. At the surface (photosphere) the boundary conditions come from
a detailed model of the outer layers. If the atmosphere is decribed by a temperature law
of the form µ ¶
4 3 4 2
T (τ ) = Te τ + ,
4 3
we can adopt the following boundary conditions
2 GM
P = ,
3 κ̄r2
T = Te or L = 4πr2 σT 4 .
References
For more details on equilibrium configurations, we refer to a course on Stellar Evolution,
or the book by Kippenhahn and Weigert (1990).
Kippenhahn R., Weigert A., 1990. Stellar structure and evolution. Springer.
14
Chapter 5
Small perturbations method
We cannot solve exactly the equations describing a star. However, if the hydrodynamical
variables remain close to their equilibrium values (or close to a known solution), it is
possible to write any variable X as the sum of its equilibrium or unperturbed value X0 ,
and a small perturbation δX, so that
X = X0 + δX.
Substituting these expressions into the hydrodynamical equations and keeping only terms
up to the first order in the perturbed quantities, we get linear equations. These equations
are much easier to study than the original equations, and their solutions are very useful
approximations.
The small perturbations method makes it possible to study the stability of the stellar
models against sufficiently small perturbations. It will not, however, give us any infor-
mation on the stability against finite amplitude perturbations, on metastable states, or
on limited cycles close to nonperturbed solutions. It will also not give us the pulsation
amplitude of a variable star.
If the unperturbed configuration is stationary (as will be considered throughout this
course), the coefficients of the linearized equations are time-independent. We can then
write the general solution as a linear combination of simple solutions which depend ex-
ponentially on the time as est (s can be complex) and which are called normal modes.
In the case of a mechanical system with a finite number of degrees of freedom, there is
also a finite number of normal modes. Here, we have an infinite number of degrees of
freedom and there exists an infinite number of normal modes of oscillation (as in the case
of a vibrating string). A given mode is stable if <s < 0, unstable if <s > 0. A stellar
configuration is stable if all its normal modes are stable, but it is unstable as soon as one
mode is unstable. In the exceptional case where the stability of one or several modes is
marginal (<s = 0), the other modes being stable, the linear analysis does not give the
information on whether the considered model is stable or not.
Exercise
How do you write the perturbations of a sum, a product, a division ?
15
5.1 Lagrangian and eulerian perturbations

To the two descriptions of hydrodynamics correspond two types of perturbations. The
lagrangian perturbation δX of X is described by
X(~a, t) = X0 (~a, t) + δX(~a, t)
and the eulerian perturbation X 0 by
X(~r, t) = X0 (~r, t) + X 0 (~r, t).
The relation between the lagrangian and the eulerian perturbations of a variable X is
given by
−
→
δX = X 0 + δr · grad X0 .
Since the perturbed variable X does not enter into the perturbation equations, but only
the quantities X0 , δX and X 0 , it is convenient to omit the subscript 0 in the unperturbed
value of the variable. The above relation will then be written as
−
→
δX = X 0 + δr · grad X.
Let xi be some system of coordinates. The following relations are straightforward:

Ã !0
∂X 0 ∂X
= ,
∂t ∂t
Ã !0
∂X 0 ∂X
= ,
∂xi ∂xi
d δX dX
= δ .
dt dt
It is important to be very careful when using lagrangian perturbations in the eulerian
formalism (or vice-versa). Indeed, the relation between ∂ δX/∂xi and δ (∂X/∂xi ) is not
straightforward. Using the relations given above, we have
∂ δX ∂X X ∂ δxj ∂X
=δ + .
∂xi ∂xi j ∂xi ∂xj
In general it is easier to write the equations in the eulerian formalism. However, the
lagrangian formalism is better to describe the radial oscillations.
Notes
The following remarks are often useful.
1) If the unperturbed configuration is static, (~v = 0 everywhere), then
d ∂
= .
dt ∂t
2) If the quantity X is independent of the coordinates in a given region of the unperturbed
configuration, then we have
δX = X 0
in that region.
16
5.2 Perturbation of the differential equations

Continuity equation:

0 →
−
∂ρ ∂ δr  −
→
+ div ρ =0 or ρ0 + div(ρ δr) = 0.
∂t ∂t
Momentum equation:
−
→
∂ 2 δr 0 ρ0 1
2
= − grad Φ + 2
grad P − grad P 0 .
∂t ρ ρ
Poisson equation:
∆Φ0 = 4πGρ0 .
Energy conservation equation:
Ã !
∂S 0 ρ0 1
T + ~v · grad S = ²0 + 2 div F~ − div F~ 0 .
∂t ρ ρ
Radiative transfer equation:
F~ 0 = −λ0 grad T − λ grad T 0 .
The boundary conditions will be described for each particular case considered below.
In non-stationary conditions, the convective flux cannot be calculated by simply perturb-
ing the expression giving the flux in the stationary case. Indeed, the convective cells have
a finite lifetime and given their inertia (mechanical and thermal) they do not instanta-
neously respond to the changing conditions. It is only in the case where the lifetime of the
perturbation is much larger then the lifetime of the convective cells that the convection
can be considered as responding instantaneously to the changes; the usual expression for
the convective flux can then be used.
5.3 Perturbation of the material equations

During the pulsation the relations P = P (ρ, T ) and κ = κ(ρ, T ), valid in the stationary
case, continue to be satisfied at each moment. Therefore we have
δP δρ δT
= Pρ + PT ,
P ρ T
δκ δρ δT
= κρ + κT ,
κ ρ T
where
Ã !
∂ ln P
Pρ = ,
∂ ln ρ T
17
Ã !
∂ ln P
PT = ,
∂ ln T ρ
Ã !
∂ ln κ
κρ = ,
∂ ln ρ T
Ã !
∂ ln κ
κT = .
∂ ln T ρ
These quantities can be calculated or deduced from the tables used for the calculation
of the stellar models. The coefficients Pρ and PT can be calculated as functions of the
coefficients Γ , which we will often use.
Ã ! Ã ! Ã !
∂ ln P Γ2 − 1 ∂ ln T ∂ ln T
Γ1 = , = , Γ3 − 1 = .
∂ ln ρ S
Γ2 ∂ ln P S
∂ ln ρ S
Only two of these three coefficients are independent. Indeed, it is easy to show that
Γ2 − 1 Γ3 − 1
= .
Γ2 Γ1
Let U be the internal energy per unit mass. We know that
δU = T δS − P δV,
i.e., Ã ! Ã !
∂U ∂U
T = and P = − .
∂S V
∂V S
Using
∂ 2U ∂ 2U
= ,
∂S∂V ∂V ∂S
we have Ã ! Ã !
∂P ∂T
− = .
∂S V ∂V S
Using ρ and S as independent variables, this last relation can be rewritten as
Ã !
∂ ln P (Γ3 − 1)ρT
= .
∂S ρ
P
We can of course write
Ã ! Ã !
δP ∂ ln P δρ ∂ ln P
= + δS,
P ∂ ln ρ S
ρ ∂S ρ
Ã ! Ã !
δT ∂ ln T δρ ∂ ln T
= + δS.
T ∂ ln ρ S
ρ ∂S ρ
The coefficient (∂ ln T /∂S)ρ is in fact the inverse of cv , the specific heat at constant volume
per unit mass. The above relations can therefore be written as
δP δρ (Γ3 − 1)cv ρT δS
= Γ1 + ,
P ρ P cv
δT δρ δS
= (Γ3 − 1) + .
T ρ cv
18
Eliminating δS we get the linearized equation of state

" #
δP (Γ3 − 1)2 cv ρT δρ (Γ3 − 1)cv ρT δT
= Γ1 − + ,
P P ρ P T
i.e.,
(Γ3 − 1)2 cv ρT
P ρ = Γ1 − ,
P
(Γ3 − 1)cv ρT
PT = .
P
The case of ², the rate of nuclear energy generation, is completely different. If the pulsation
time is long compared to the reactants lifetimes, the abundances or the different nuclei
participating to the nuclear reactions assume their instantaneous equilibrium values and
we can write
δ² δρ δT
= ²ρ + ²T ,
² ρ T
with Ã ! Ã !
∂ ln ² ∂ ln ²
²ρ = and ²T = .
∂ ln ρ T
ln T ρ
However, in general, this will not be the case and it will be necessary to take into account
the details of the nuclear reactions and to study the perturbed solutions to the equations
describing their kinetics. We then get
δ² δρ δT
= ²ρ (σ) + ²T (σ) ,
² ρ T
where the coefficients ²ρ and ²T are functions of the pulsation frequency σ. These coeffi-
cients are usually complex numbers (phase shift). They are often designated by µef f and
νef f in the literature.
Exercises
1. Derive the thermodynamic coefficients of a perfect gas, of a mixture of gas and radia-
tion, of partially ionized gas.
2. Derive the coefficients ²ρ (σ) and ²T (σ) for the proton-proton chain and for the carbon
cycle.
19
Chapter 6
Adiabatic perturbations
We saw that the characteristic timescale for energy transfer in a star is much larger
than the characteristic dynamical timescale (except in the external layers of the star).
It is therefore natural, as a first approximation, to neglect transport phenomena and
energy production for perturbations which evolve on a dynamical timescale. The energy
conservation equation then becomes
δS = 0.
This equation describes an adiabatic perturbation. As a consequence, there is a simple

relation between the variations of density and pressure:
δP δρ
= Γ1 or δP = c2 δρ,
P ρ
q
where c = Γ1 P/ρ is the sound speed.
Of course, in the adiabatic approximation, the equation of energy transport is no longer
needed. We will then keep the adiabatic equation, the continuity equation, the momentum
equation, and the Poisson equation. We will show that this simplified problem can be
written as a self-adjoint problem.
→
−
All the perturbed quantities can be easily written as functions of the displacement δr.
The continuity equation can be written as
−
→ →
−
ρ0 + div(ρ δr) = 0 or δρ + ρ div δr = 0
which gives
−
→ −
→
ρ0 = − div(ρδr) or δρ = −ρ div δr.
The adiabatic equation then gives δP or P 0
−
→ −
→ − →
δP = −Γ1 P div δr or P 0 = −Γ1 P div δr − δr · grad P.
Finally the Poisson equation

∆Φ0 = 4πGρ0
20
has the following solution

Z Z −
→
0 ρ0Q dVQ div(ρ δr)Q dVQ
Φ (P ) = −G =G .
|P Q| |P Q|
The momentum equation

−
→
d2 δr 0 ρ0 1
2
= − grad Φ + 2
grad P − grad P 0
dt ρ ρ
can then be written as −
→
d2 δr −
→
2
= L δr,
dt
where L is the linear operator defined by
Z →
−
−
→ div(ρ δr)Q dVQ 1 −
→
Lδr = −G gradP − 2 div(ρ δr) grad P
|P Q| ρ
1 −
→ 1 −
→
+ grad(Γ1 P div δr) + grad( δr · grad P ).
ρ ρ
Let us consider the functional space of vector fields continuously derivable and define the
scalar product of two such vector fields by
Z
(~u, ~v ) = ρ~u · ~v dV = (~v , ~u).
V
In the integral, V stands for a volume which is large enough to include the whole stellar
configuration. We will use S to designate the surface delimiting this volume.
We will show that L is a symmetrical operator. As it is obviously real, it will therefore
be hermitian (i.e. self-adjoined). For two real fields ~u and ~v , we will show that we have
(L~u, ~v ) = (~u, L~v ).
To prove this property we will transform some integrals in the following way:
Z Z
~a · grad α dV = [div(α~a) − α div ~a ] dV
V
IV Z Z
−
→
= α~a · dS − α div ~a dV = − α div ~a dV.
S V V
This last equality is justified only if the surface integral is zero, which must be checked
for each case.
Z Z Z
div(ρ~u)Q 1
(L~u, ~v ) = −G dVP ρ~v · gradP dVQ − (~v · grad P ) div(ρ~u) dV
|P Q| ρ
Z Z
+ ~v · grad(Γ1 P div ~u) dV + ~v · grad(~u · grad P ) dV
= A + B + C + D.
21
We have
ZZ
div(ρ~u)Q div(ρ~v )P
A = G dVP dVQ ,
|P Q|
Z Z
1
B = − (div ~u)~v · grad P dV − (~u · grad ρ)(~v · grad P ) dV = B1 + B2 ,
ρ
Z
C = − Γ1 P (div ~u)(div ~v ) dV ,
Z
D = − (~u · grad P ) div ~v dV .
The expressions A, B2 and C are obviously symmetrical (note for B2 that grad ρ k grad P ).
The same can be said for B1 + D.
Let us come back to the momentum equation. The interesting solutions for us must also
satisfy the linear boundary conditions, which are homogeneous and time-independent.
~ r)est . These solutions are the
The problem therefore has simple solutions of the form ξ(~
~ r) obey the following equation (from now
normal modes of oscillation of the star. The ξ(~
on we will omit the arrow above ξ)
Lξ = s2 ξ.
The function ξ is called the eigenfunction, and s the eigenvalue. In fact, it is really
λ = s2 which should be called the eigenvalue. The problem admits an infinite number
of eigenvalues. The analogy with a mechanical system with a finite number of degrees
of freedom is obvious. We will accept without proof that these eigenfunctions form a
complete set, i.e. that they constitute a basis to write any perturbation.
The fact that L is hermitian has several interesting consequences.
1◦ ) The eigenvalues λ are real. Indeed, if ξ is the eigenvalue associated to λ
Lξ = λξ.
We take the product with ξ

(Lξ, ξ) = λ(ξ, ξ).
But
(Lξ, ξ) = (ξ, Lξ) = (Lξ, ξ),
so that λ = (Lξ, ξ)/(ξ, ξ) is real. Therefore, the eigenfunctions are real up to a constant
multiplying factor.
√ We will assume hereafter that we have chosen real eigenfunctions.
If λ > 0, s = ± λ and one of the two √ modes is unstable. This instability is called a
dynamical instability. If λ < 0, s = ±i −λ and the linear analysis does not establish a
dynamical instability. A model in which all the modes have λ < 0 is dynamically stable.
2◦ ) The eigenfunctions associated to different eigenvalues are orthogonal. Indeed,
Lξi = λi ξi and Lξj = λj ξj ,

λi (ξi , ξj ) = (Lξi , ξj ) = (ξi , Lξj ) = λj (ξi , ξj ) .
22
If λi and λj are distinct, we have (ξi , ξj ) = 0. On the other hand, a family of independent
eigenfunctions associated to the same eigenvalue can always be orthogonalized. We can
therefore consider that all the eigenfunctions are orthogonal to each other.
3◦ ) The eigenvalues and eigenfunctions obey to a variational principle. Consider the
functional
(Lξ, ξ)
Λ(ξ) =
(ξ, ξ)
and its first variation
1
δΛ = {[(Lδξ, ξ) + (Lξ, δξ)](ξ, ξ) − (Lξ, ξ)[(δξ, ξ) + (ξ, δξ)]} .
(ξ, ξ)2
The condition δΛ = 0 is equivalent to
(ξ, ξ)(Lξ, δξ) − (Lξ, ξ)(ξ, δξ) + (ξ, ξ)(Lξ, δξ) − (Lξ, ξ)(ξ, δξ) = 0 .
Since δξ is arbitrary, we have
(Lξ, ξ)ξ
Lξ = = Λ(ξ)ξ .
(ξ, ξ)
The functional Λ(ξ) is therefore stationary when ξ is an eigenfunction, and its corre-
sponding eigenvalue is Λ(ξ). From this variational principle, it is possible to deduce a
method to calculate the eigenmodes. It has been used, although not very frequently. It
is also possible to deduce a method to improve the calculation of an eigenvalue: a gross
approximation of ξ will give a good approximation Λ(ξ) for the eigenvalue λ.
For a stable mode, we will write s = −iσ. Note that the average kinetic energy of a mode
with an angular frequency σ is given by
σ2
T = (ξ, ξ) .
4
References
The operator L is just a particular case of the operator used by Lynden-Bell and Ostriker
(1967). We follow rather closely Cox (1980), chapter 5.
Cox J.P., 1980. Theory of stellar pulsation. Princeton University Press.
Lynden-Bell D., Ostriker J.P., 1967. On the stability of differentially rotating bodies.
MNRAS, 136, 293–310.
23
Chapter 7
Radial oscillations
7.1 Differential equations

In the case of radial oscillations, it is particularly useful to use lagrangian perturbations
because it is possible to replace the spatial coordinate r by a lagrangian coordinate m.
The advantage of this change of coordinates is that the operator ∂/∂m commutes with
the operator of lagrangian perturbation δ.
We have a relation between ∂/∂m and ∂/∂r
∂ 1 ∂
= .
∂m 4πr2 ρ ∂r
We now write the general equations for a spherically symmetric fluid. We have already
shown that the Poisson equation can be integrated once to give
∂Φ Gm
= 2 .
∂r r
The continuity, momentum, energy conservation, and radiative transfer equations are
given by
1 dρ 1 ∂ 1 dρ ∂ 2
+ 2 (r2 v) = 0 or + 4πρ (r v) = 0,
ρ dt r ∂r ρ dt ∂m
dv Gm 1 ∂P dv Gm ∂P
=− 2 − or = − 2 − 4πr2 ,
dt r ρ ∂r dt r ∂m
dS 1 ∂ dS ∂L
T = ² − 2 (r2 F ) or T =²− ,
dt ρr ∂r dt ∂m
4acT 3 ∂T 64π 2 r4 acT 3 ∂T
F =− or L=− .
3κρ ∂r 3κ ∂m
The lagrangian perturbation operator δ commutes with the two operators d/dt and ∂/∂m
which appear in those equations. It is therefore easy to write the perturbed equations
δρ ∂ 2
+ 4πρ (r δr) = 0,
ρ ∂m
24
d2 δr Gm δr ∂P δr ∂ δP
2
=2 2 − 8πr2 − 4πr2 ,
dt r r ∂m r ∂m
d δS ∂ δL
T = δ² − ,
(dt Ã ∂m ! )
64π 2 r4 acT 3 ∂T δr δT δκ ∂ δT
δL = − 4 +3 − + .
3κ ∂m r T κ ∂m
We now introduce again the operator ∂/∂r and we get
Ã ! Ã !
∂ δr 1 δr δρ
= − 3 + ,
∂r r r r ρ
Ã ! ( Ã !)
∂ δP 1 dP δP δr r3 d2 δr
= − +4 − ,
∂r P P dr P r Gm dt2 r
Ã ! Ã !
∂ δL 1 dL δL δ² 4πr2 ρT d δS
= − − − ,
∂r L L dr L ² L dt
Ã ! Ã !
∂ δT 1 dT δr δT δL δκ
= − 4 +4 − − .
∂r T T dr r T L κ
These equations must be completed by the perturbed boundary conditions and material
equations.
The coefficients of these linear equations are time-independent. We will therefore look for
simple solutions of the form
δX δX
(r, t) = (r)est .
X X
Note that for simplicity we will continue to write δX/X for the r-dependent function.
The partial differential equations can then be reduced to ordinary differential equations
depending on the parameter s.
Ã ! Ã !
d δr 1 δr δρ
= − 3 + ,
dr r r r ρ
Ã ! ( Ã ! )
d δP 1 dP δP r3 s2 δr
= − + 4− ,
dr P P dr P Gm r
Ã ! Ã !
d δL 1 dL δL δ² 4πr2 cv ρT δS
= − − − s ,
dr L L dr L ² L cv
Ã ! Ã !
d δT 1 dT δr δT δL δκ
= − 4 +4 − − .
dr T T dr r T L κ
The choice of a coordinate system presenting a singularity at the star center introduces a
regular singularity at this point. We will impose the condition that the physical quantities
δr, δP , . . . remain regular. A simple power series expansion gives the regularity conditions
at the center:
δr δρ
3 + = 0,
r ρ
Ã !
δL δ²
² − + sT δS = 0.
L ²
25
At the surface, the simplest models (polytropes) impose vanishing pressure and temper-
ature conditions. In this case, the system of differential equations presents a surface
singularity. By requiring that the physical quantities remain regular at the surface we get
the following boundary conditions
Ã !
δP R3 s2 δr
+ 4− = 0,
P GM r
δr δT δL δκ
4 +4 − − = 0.
r T L κ
If the equilibrium model was joined to a model atmosphere, we would impose that the per-
turbations of the interior are continuously joined to the perturbations of the atmosphere.
If the above mechanical condition is relatively satisfactory, the radiative condition can be
easily improved. Assume that the thermal structure of the atmosphere can be described
by the Eddington approximation at all times
3 2 3L 2
T 4 = Te4 (τ + ) ≈ 2
(τ + ).
4 3 16πr σ 3
In addition Z ∞
κ ∆m
τ= κρ dr ≈ .
r 4πr2
We then get Ã !
δT δr δL τ δκ δr
4 +2 − − −2 = 0.
T r L τ + 2/3 κ r
During the pulsation, the photosphere level moves through the fluid. It is therefore
incorrect to directly perturb the equation
L = 4πR2 σTe4
and to apply it to the level corresponding to τ = 2/3 of the equilibrium model.

The search for radial oscillation modes comes down to solving an eigenvalue-boundary
conditions system of fourth order differential equations.
7.2 Integral equations

We can derive integral equations from the above equations. These integral equations will
prove very useful for the remainder of this course. Let us write the continuity, momentum,
and energy equations as
d ³ 2 ´ δρ
r δr = −r2 ,
dr ρ
Ã !
d δP 4 dP
+ ρs2 + δr = 0,
dr r dr
δS d δL
scv T = δ² − .
cv dm
26
We take the product of the momentum equation by 4πr2 δr and integrate it over the
stellar radius
Z Z Ã !
d δP 4 dP
s2 4πr2 ρ|δr|2 dr + 4πr2 δr + δr dr = 0.
dr r dr
Using the continuity equation and integrating by parts, this equation can be written as
Z Z ½ ¯ ¯2 ¾
δP δρ 4r dP ¯ δr ¯
¯ ¯
s2 |δr|2 dm + + ¯ ¯ dm = 0.
ρ ρ ρ dr ¯ r ¯
This equation will be used to study vibrational stability. Let’s write δP as function of δρ
and of δS, then write δS using the energy conservation equation.
Z Z ½ ¯ ¯2 ¯ ¯2 ¾ Ã !
2 2 Γ1 P ¯ δρ ¯
¯ ¯ 4r dP ¯ δr ¯
¯ ¯ 1Z δρ d δL
s |δr| dm + ¯ ¯ + ¯ ¯ dm + (Γ3 − 1) δ² − dm = 0.
ρ ¯ ρ ¯ ρ dr ¯ r ¯ s ρ dm
The s-independent term can be written in a different, interesting form. Its first term can
be developed as follows
Z ¯ ¯2 Z ¯ ¯2
Γ1 P ¯ δρ ¯ ¯ d
4πΓ1 P ¯
¯ ¯ ¯ 2 ¯
¯ ¯ dm = ¯ (r δr)¯ dr = · · ·
ρ ¯ ρ ¯ ¯ dr
r2 ¯
Z ½ ¯ Ã !¯2 ¯ ¯2 ¾
¯ δr ¯¯ ¯ ¯
4¯ d 3 ¯ δr ¯ d
= 4πΓ1 P r ¯ ¯ − 12πr ¯ ¯ (Γ1 P ) dr
¯ dr r ¯ ¯ r ¯ dr
Z ½ ¯ Ã !¯ 2 ¯ ¯2 ¾
Γ1 P r2 ¯¯ d δr ¯¯ r d ¯ δr ¯
¯ ¯
= ¯ ¯ − (3Γ 1 P ) ¯ ¯ dm.
ρ ¯ dr r ¯ ρ dr ¯ r ¯
The s-independent term can then be written as

Z ½ ¯ Ã !¯2 ¯ ¯2 ¾
Γ1 P r2 ¯ d δr ¯¯ r d ¯ δr ¯
¯ ¯ ¯
¯ ¯ − [(3Γ1 − 4)P ] ¯ ¯ dm.
ρ ¯ dr r ¯ ρ dr ¯ r ¯
We have obtained the cubic equation
s3 + As + B = 0,
with
Z ½ ¯ ¯2 ¾ Z
¯ δρ ¯ Gm
¯ ¯
A = c ¯ ¯ − 4 3 |δr| dm/ |δr|2 dm
2 2
¯ ρ ¯ r
Z ½ ¯ Ã !¯2 ¯ ¯2 ¾ Z
¯ δr ¯¯ ¯ δr ¯
2 2¯ d r d ¯ ¯
= c r ¯¯ ¯ − [(3Γ1 − 4)P ] ¯¯ ¯¯ dm/ |δr|2 dm,
dr r ¯ ρ dr r
Z Ã ! Z
δρ d δL
B = (Γ3 − 1) δ² − dm/ |δr|2 dm.
ρ dm
27
7.3 Dynamical modes and secular modes

We notice that the eigenvalue appears twice in the differential equations coefficients. If
we write these in their dimensionless form, the first one occurs in the momentum equation
as r3 s2 /Gm, and the second one in the energy conservation equation as 4πr3 cv ρT s/L.
To each of these is associated a family of modes: the dynamical modes and the secular
modes. The characteristic times of these modes are given by the corresponding coefficients
of s. For the dynamical modes the characteristic time is the dynamical time, and for the
secular modes it is the Kelvin-Helmholtz time
s
r3 4πr3 cv ρT
≈ τdyn , ≈ τKH .
Gm L
These statements can be justified by numerous calculations. We will try to make them
plausible by evaluating the order of magnitude of the roots of the cubic equation we have
established above.
We will assume that the eigenfunctions are of the same order of magnitude to get order-of-
magnitude estimates for the coefficients A and B (this approximation relies on the results
of numerous numerical integrations). We get
2 1
A ≈ 1/τdyn et B≈ 2
.
τdyn τKH
We will therefore write A = A0 /τdyn

2
and B = B 0 /τdyn
2
τKH , where A0 and B 0 are dimen-
sionless numbers of order unity. The cubic equation can then be written as
s03 + A0 s0 + αB 0 = 0,
where s = s0 /τdyn and α = τdyn /τHK ¿ 1. It is easy to see that this equation has two
roots of order unity, which can be approximated by setting α = 0,
√
s0 = ± −A0
and one root of order α. We get an approximation of the latter by neglecting the cubic
term or by noticing that the product of the roots is equal to −αB 0 ,
s0 = −αB 0 /A0 .
We therefore get two roots of order 1/τdyn and one root of order 1/τKH for s. The first two
roots are associated to perturbations whose characteristic evolution time is the dynamical
time τdyn . These modes are called dynamical modes. The third root is associated to
perturbations evolving on a much longer time-scale, the Kelvin-Helmholtz time-scale.
These modes are called secular modes. Their existence is linked to the B coefficient which
contains all the non adiabatic effects.
The preceding argument has however a weakness. The coefficients of the cubic equation
are calculated starting from unknown eigenfunctions which will of course be different for
28
dynamical and for secular modes. But the existence of two types of radial modes can also
be justified through other methods: local analysis of the system of differential equations,
use of the simple model such as the one-layer model of Baker.
The distinction between the two families of modes can generally be made without any
ambiguity. There are however cases where the modes cannot be attributed to one or the
other of these families. This happens when the thermal timescale becomes comparable to
the dynamical timescale (very luminous stars, models close to dynamical instability).
References
The surface radiative boundary condition used above was obtained Baker and Kippenhahn
(1965).
We have used a diffusion equation to describe the radiative transport. In the external
layers of the star, this approximation is sometimes unsatisfactory. It is then possible to
use the Eddington approximation. More details on this topic can be found in Unno and
Spiegel (1966) and in Christensen-Dalsgaard and Frandsen (1983).
The cubic equation in s was obtained by Ledoux (1963 and 1969).
Baker N., Kippenhahn R., 1965. The pulsations of models of delta Cephei stars II.
Astrophys J, 142, 868–889.
Christensen-Dalsgaard J., Frandsen S., 1983. Radiative transfer and stellar oscillations.
Solar Phys, 82, 165–204.
Ledoux P., 1963. Stellar stability and stellar evolution. In Gratton L. (edit.), Evoluzione
delle stelle, Academic Press, 394–445.
Ledoux P., 1969. Oscillations et stabilité stellaires. In Joseph C., Janin G., Maeder A.,
Mayor M. (edit.), La structure interne des étoiles, XIème cours de perfectionnement de
l’association vaudoise des chercheurs en physique, Saas-Fee, 44–211.
Unno W., Spiegel E.A., 1966. The Eddington approximation in the radiative heat equa-
tion. Pub Astron Soc Japan, 18, 85–95.
29
Chapter 8
Adiabatic radial oscillations
The periods of the dynamical modes of a relatively low order are of the order of the
dynamical timescale and are much shorter than the characteristic timescale of thermal
energy transport. This justifies, as a first order approximation, the use of the adiabatic
approximation to study these modes. If this study shows an instability, the star is said
to be dynamically unstable. In the opposite case, it is said to be dynamically stable.
If the stability of the star is not marginal, the adiabatic approximation will give a very
good approximation of the eigenvalues of the normal modes. The adiabatic eigenfunctions
will also constitute a very good approximation, except in the exterior layers where the
adiabatic approximation is no longer justified (the local thermal characteristic timescale
is no longer large compared to the dynamical timescale).
As we have shown, the adiabatic approximation generally leads to a self-adjoint problem
where the eigenvalues s2 are real. In the case of dynamical stability, s is purely imaginary.
The adiabatic analysis therefore cannot predict whether the corresponding mode is excited
or damped. The answer to that question can only be found by considering the non
adiabatic terms and constitutes the problem of the vibrational stability, which will be
discussed in the next chapter.
In the adiabatic approximation, δP/P and δρ/ρ are related through
δP δρ
= Γ1 .
P ρ
The adiabatic radial problem can therefore be described by the differential equations
Ã ! Ã !
d δr 1 δr 1 δP
= − 3 + ,
dr r r r Γ1 P
Ã ! ( Ã ! )
d δP 1 dP δP r3 s2 δr
= − + 4− .
dr P P dr P Gm r
and the boundary conditions
δr 1 δP
3 + = 0 at r = 0 ,
r Γ1 P
Ã !
δP R3 s2 δr
+ 4− = 0 at r = R .
P GM r
30
It is useful to use dimensionless variables. We define

s
r m δr δP GM
x= , q= , ξ= , η= and s = −iσ with σ = ω.
R M r P R3
We have then
µ ¶
dξ 1 η
= − 3ξ + ,
dx x Γ1
( Ã ! )
dη d ln P x3 ω 2
= − η+ 4+ ξ .
dx dx q
with the boundary conditions

η
3ξ + = 0 at x = 0 ,
Γ1
η + (4 + ω 2 )ξ = 0 at x = 1 .
Consider now two homologous stellar configurations. Let α = R0 /R and β = M 0 /M . We

can establish between the points of these two models a correspondence such that
r0 = αr, m0 = βm, ρ0 = α−3 βρ, P 0 = α−4 β 2 P, ...
In such a homologous transformation, the coefficients of the differential system and the
boundary conditions are invariant. The two models are characterized by the same values
of the dimensionless frequencies
ω0 = ω and σ 0 = β 1/2 α−3/2 σ .
The pulsation periods of the two stars are therefore in the same ratio as their dynamical
timescales. Indeed,
0
τdyn = β −1/2 α3/2 τdyn .
Eliminating η, we get the differential equation

Ã ! ( )
d dξ d
Γ1 P r4 + r 3
[(3Γ1 − 4)P ] + σ 2 ρr4 ξ = 0
dr dr dr
and the boundary conditions

dξ
= 0 at r = 0 ,
dr Ã !
dξ R3 σ 2
Γ1 R + 3Γ1 − 4 − ξ = 0 at r = R .
dr GM
This problem is of the Sturm-Liouville type. Therefore it admits an infinity of eigenvalues

which can be labelled such that
σ02 < σ12 < · · · < σk2 < · · · with lim σk2 = +∞ .
k→∞
31
The eigenfunction ξk associated to σk has k zeros in the ]0, R[ interval. It is also possible
to show that the ξk form a complete set, i.e., a basis in the space of the ξ functions
describing a radial perturbation.
We write the differential equation as
Lξ = σ 2 ξ
with Ã !
1 d dξ 1 d
Lξ = − 4 Γ1 P r4 − [(3Γ1 − 4)P ] ξ .
ρr dr dr ρr dr
This operator is essentially the same as the one we have introduced earlier (the sign has
−
→
changed, and it acts on δr/r instead of δr). As exercise, one can verify that it is hermitian
for the scalar product Z
(u, v) = ρr4 uv dr .
The σk2 are the stationary values for the functional
(u, Lu)
Λ(u) = .
(u, u)
In particular
(u, Lu)
σ02 = min .
u (u, u)
If the model is dynamically stable, the operator L is positive definite and vice-versa.
Integrating by parts, one can write
Z ½ ¯ ¯ ¾
¯ du ¯2
4¯ ¯ 3 2 d
(u, Lu) = Γ1 P r ¯¯ ¯¯ − r |u| [(3Γ1 − 4)P ] dr .
dr dr
Assume that Γ1 is constant in the whole star. It is obvious that if Γ1 > 4/3, for all u 6= 0
we have (u, Lu) > 0 and dynamical stability is assured. On the other hand if Γ1 < 4/3,
it is sufficient to consider u = constant to show that L is not positive definite and that
consequently there is dynamical instability. In the case where Γ1 = 4/3, the dynamical
stability is marginal. The system then admits the solution σ = 0, ξ = 1 and η = −4,
which describes a homologous transformation of the star.
Assume Γ1 > 4/3. We have
(u, Lu)
σ02 ≤ .
(u, u)
Taking u = 1, we have
Z Z Z
3 d Gm q dq
r [(3Γ1 − 4)P ] dr dm GM
σ02 ≤ − dr Z = (3Γ1 − 4) Z r = (3Γ1 − 4) 3 Z x .
4
ρr dr 2
r dm R x2 dq
32
We can integrate the second term by parts to get

Z
3Γ − 4 c2 dm 3Γ1 − 4 c2
1
σ02 ≤ 3 Z =3 .
Γ1 r2 dm Γ1 r2
We can find a lower limit for σ02 as follows.

Z ¯ ¯ Z
¯ du ¯2
¯ ¯
(u, Lu) = Γ1 P r ¯ ¯ dr + (3Γ1 − 4) Gmρr|u|2 dr
4
¯ dr ¯
Z
GM q 4 2 GM Z
≥ (3Γ1 − 4) 3 3
ρr |u| dr ≥ (3Γ1 − 4) 3 ρr4 |u|2 dr ,
R x R
because q/x3 = ρ̄(r)/ρ̄ > 1. We then have
(u, Lu) GM
σ02 = min ≥ (3Γ1 − 4) 3 .
u (u, u) R
If Γ1 is not constant, it is no longer possible to take it out of the integration sign, unless
we use some average value Γ1 . Γ1 can then be smaller than 4/3 in a limited region of the
star without leading to radial dynamical instability.
8.1 Energy of a mode

We write the momentum equation as
d2 δr 4Gmρ ∂ δP
ρ 2
= 3
δr − ,
dt r ∂r
and we multiply it by d δr/dt
d 1 2 2Gm 2 ∂ δP d δr
ρ ( v − 3 δr ) = − .
dt 2 r ∂r dt
We transform the right-hand side of this equation
∂ δP d δr
− = −~v · grad δP = −div(~v δP ) + δP div ~v ,
δr dt
then Ã !2
Γ1 P d δρ 1 d δρ
δP div ~v = − δρ = − ρc2
ρ dt ρ 2 dt ρ
and finally
d n h 1 2 1 2 ³ δρ ´2 Gm io
ρ v + c − 2 3 (δr)2 = − div(δP ~v )
dt 2 2 ρ r
or
d −
→
(ρE) = − div F .
dt
33
1 1 ³ δρ ´2 Gm ³ δr ´2
E = v 2 + c2 −2
2 2 ρ r r
is the mechanical energy of the pulsation per unit mass. The three terms correspond to
the kinetic energy, acoustic potential energy, and gravific potential energy respectively.
The vector
−
→
F = δP ~v
is the energy flux density.
We group together the two potential energy terms
EP = EA + EG .
We then have
E = EK + EP .
If we isolate the time-dependence as
δr(r, t) = δr(r) cos σt ,
we see that
EK (r, t) = EK (r) sin2 σt and EP (r, t) = EP (r) cos2 σt .
Integrating these expressions over the whole stellar mass we have

Z
EK (t) = EK (r, t) dm = EK sin2 σt ,
Z
EP (t) = EP (r, t) dm = EP cos2 σt ,
Z
E(t) = E(r, t) dm = EK sin2 σt + EP cos2 σt .
Integrating the equation of conservation of the mechanical energy of the pulsation over
the whole volume of the star we get
dE
=0 i.e. E = constant .
dt
It follows that
E = EK = EP
and
1
EK (t) = EP (t) = E.
2
The total pulsation energy can be written as
σ2 Z 2
E= δr dm
2
where the δr under the integration sign is the time-independent amplitude δr(r).
34
Figure 8.1: Fundamental radial oscillation mode of the standard model (polytrope of
index 3).
8.2 Behavior of the eigenfunctions

It is important to have an idea of the behavior of the eigenfunctions ξk for a dynamically
stable model. The figures 8.1 to 8.3 show some radial oscillations eigenfunctions for the
standard model (polytrope of index 3). The arbitrary factors in their definition were
chosen such that
(ξ0 , ξ0 ) = (ξ1 , ξ1 ) = · · ·
in order to allow a comparison between modes.
We note that the amplitude increases in magnitude as we get near the surface, and it does
it faster as the mode order is higher. This is in complete agreement with the fact that at
the surface we have
d ln |ξ| ω 2 + 4 − 3Γ1
= > 0.
dx Γ1
We define the average of a physical quantity X weighted by the energy of the k-th mode
as Z
Xr2 ξk2 dm
hXi = Z .
r2 ξk2 dm
The value of hxi (x = r/R) allows then to measure the relative importance of the central
regions and the external layers for the considered mode. The following table gives a few
indications for some radial modes of the standard model.
35
Figure 8.2: First harmonic of the radial oscillation of the standard model (polytrope of
index 3).
Figure 8.3: Second harmonic of the radial oscillation of the standard model (polytrope of
index 3).
36
Mode ω ξs /ξc hxi

0 3.04215 22.44 0.633
1 4.12123 -58.88 0.681
2 5.33690 135.3 0.715
··· ··· ··· ···
9 14.16770 -1681 0.735
We see from this table that the external layers play a more important role for the har-
monics than for the fundamental mode, and this role increases as the order of the mode
increases. This is true not only for the standard model, but is in fact completely general.
Let us also point out that the amplitude ratio between the surface and the center is higher
if the model concentration (ρc /ρ) is higher.
8.3 Some unstable cases

At the border between dynamical stability and instability, the characteristic timescale
of the unstable mode stops being small compared to the characteristic timescales of the
secular modes. In that case, it is artificial to distinguish between dynamical and secular
instability
1) In supermassive stars (M > 105 M¯ ), the radiation pressure is larger than gas pressure.
Indeed the ratio β of the gas pressure to the total pressure is approximately given by
s
M¯
βµ = 4.28 .
M
However when β is small, Γ1 is close to 4/3 and is given approximately by
4 β
Γ1 ≈ + .
3 6
For such stars it is necessary to consider general relativity effects, and the criterion for
dynamical stability can be written as
4 GM
Γ1 > + 2.25 2 (c = speed of light) .
3 Rc
As a result, the supermassive stars whose mass is larger than some critical mass between
105 and 106 M¯ are dynamically unstable.
2) Several effects tend to dynamically destabilize very condensed white dwarfs. The
relativistic degeneracy of the electrons lowers Γ1 around 4/3. On the other hand, the
equilibrium between β disintegration and electron capture contributes also to lower Γ1 .
Finally the general relativity effects make the stability criterion harder to satisfy. The con-
figurations with a central density larger than a critical density of the order of 1010 g cm−3
are dynamically unstable. The only stable high density stars are the neutron stars.
37
3) In the initial contraction phases of a proto-star, there is no dynamically stable equilib-

rium model due to the H2 dissociation and the ionization of H and He. In these phases
the proto-star therefore evolves on the dynamical timescale.
4) In the final phases of evolution of a sufficiently massive star, the establishment of
a nuclear equilibrium (iron-peak) lowers Γ1 and creates a dynamical instability. This
instability will define the initial phase of a supernova.
5) The enormous mass losses of the S Dor (or LBV) variable stars could be due to a
dynamical instability in their envelope. In these massive and very luminous stars, the
radiation pressure is very high (β small) and the existence of a hydrogen and helium
partial ionization zone is enough to lower the average value of Γ1 below the critical value.
References
The reader is refered to the book by Ince (1956), chapter 10, for more details on the
Sturm-Liouville problem, and to the article by Beyer (1995) for a rigorous proof of the
properties of the spectrum of adiabatic radial oscillations.
Beyer H.R., 1995. The spectrum of radial adiabatic stellar oscillations. J Math Phys, 36,
4815–4825.
Ince E.L., 1956. Ordinary differential equations. Dover.
38
Chapter 9
Asymptotic expression for the radial

frequencies
We have seen above that the differential equation describing the adiabatic radial oscilla-
tions can be written as
Ã ! ½ ¾
d dξ d
Γ1 P r4 + r 3
[(3Γ1 − 4)P ] + σ 2 ρr4 ξ = 0.
dr dr dr
For high order modes, the σ 2 term is the dominant term in the coefficient of ξ. This
will allow the use of asymptotic methods to obtain approximate values for the oscillation
frequencies. There are different such methods, more or less complicated, and more or less
rigorous. We will restrict ourselves to a simple approach.
We use the change of variables
Z r
dr
τ= and w = r2 (Γ1 P ρ)1/4 ξ.
0 c
We note that τ is the time needed to travel, at the sound speed, from the center of the
star to the point under consideration. The differential equation reduces to
½ ¾
d2 w 2 1 d 1 d2 2
+ σ + [(3Γ 1 − 4)P ] − [r (Γ1 P ρ)1/4 ] w = 0.
dτ 2 r(Γ1 P ρ)1/2 dτ r2 (Γ1 P ρ)1/4 dτ 2
The solution must satisfy the boundary conditions w(0) = w(τR ) = 0, where τR = τ (R).
For high order modes, σ 2 is large and it is tempting to ignore the other terms in the
coefficient w. However, these terms are singular at the center and at the surface, and they
are therefore not negligible compared to σ 2 near those points. Without these singularities
we could simplify the equation to
d2 w
+ σ 2 w = 0.
dτ 2
The solutions satisfying the boundary conditions have the form
kπ
wk ∝ sin σk τ with σk = for k = 1, 2, 3, . . .
τR
39
wk has k − 1 nodes in the interval ]0, τR [. If we number the modes like this, we must give
the value k = 1 to the fundamental mode, k = 2 to the first harmonic, . . . We will justify
this numbering of the modes when we will study the non radial modes.
To take into effect the singularities, we will develop two approximations. The first one
will take into account the central singularity, and the second the surface singularity. We’ll
join the two solutions at a point situated far from both the center and the surface.
9.1 Central singularity

The singularity comes from the term with a second derivative with respect to τ . We can
write the differential equation as
· ¸
d2 w 2 2
+ σ − + f (τ ) w = 0,
dτ 2 τ2
where f (τ ) is a regular function of τ at τ = 0. We can now neglect f (τ ) compared to the
term in σ 2 and the singular
√ term. The equation can be reduced to a Bessel equation by
writing z = στ and w = zu(z).
µ ¶
d2 u 1 du 9
2
+ + 1 − 2 u = 0.
dz z dz 4z
The regular solution at z = 0 is given by the Bessel function of the first kind of order 3/2,
u(z) = J3/2 (z). Let w0 (τ ) be the approximative solution thus obtained. It is valid as long
as we are not too close to the surface, whose singularity has not been taken into account.
We use the asymptotic approximation of J3/2 (z) for large z to write w0 (τ ), not too close
to the center, as
π
w0 (τ ) ∝ sin(στ − ).
2
9.2 Surface singularity

We can roughly describe the structure of the superficial layers with an effective polytropic
index ne such that ρ ∝ (R − r)ne and P ∝ (R − r)ne +1 . If these layers are convective, ne
is equal to the polytropic index associated to the value of Γ1 , i.e. ne = 1/(Γ1 − 1). If the
superficial layers are radiative and if the opacity is given by κ ∝ ρr T −s , ne is given by
ne = (r + 3)/(s + 1). We easily get the following relations
τR − τ ∝ (R − r)1/2 , ρ ∝ (τR − τ )2ne , P ∝ (τR − τ )2ne +1 , c ∝ (τR − τ ).
We can then write the differential equation as

" #
d2 w 2 n2e − 14
+ σ − + f (τ ) w = 0,
dτ 2 (τR − τ )2
40
where√f (τ ) is a regular function in τR . Omitting this term and using z = σ(τR − τ ) and
w = zu(z), we get Ã !
d2 u 1 du n2e
+ + 1 − 2 u = 0,
dz 2 z dz z
whose regular solution is given by Jne (z). We thus obtain an approximation w00 (τ ) valid
away from the center. Using the asymptotic expression of the Bessel function for large z,
we can write for w00 (τ ), not too close to the surface,
π ne π
w00 (τ ) ∝ sin(στ − στR − + ).
4 2
9.3 Joining of the two solutions

We join the two solutions w0 (τ ) and w00 (τ ) at any point where both approximations are
valid. It is enough to ask that the phase difference of the sine arguments be a multiple of
π and we then have
π ne 1 π
wk (τ ) ∝ sin(σk τ − ) with σk = (k + + ) for k = 1, 2, 3, . . .
2 2 4 τR
We check that wk (τ ) has indeed k − 1 zeros in the interval ]0, τR [. Let τ ∗ be one zero of
wk . We’ll assume that it is the `-th from the center and the m-th from the surface; wk
then has ` + m − 1 zeros. Using the asymptotic expression for the zeros of Jν (z), we get
1 ne 1
σk τ ∗ = (` + )π and σk (τR − τ ∗ ) = (m + − )π.
2 2 4
Adding these two relations and using the expression for σk , we get k = ` + m, which
confirms that wk indeed has k − 1 zeros. The modes are therefore numbered by the index
k as above, the index 1 being attributed to the fundamental mode, 2 to the first harmonic,
...
9.4 Brief reminder on Bessel functions

The regular solution at z = 0 of the Bessel equation
Ã !
d2 u 1 du ν2
+ + 1 − u=0
dz 2 z dz z2
is the Bessel function of the first kind of order ν, written Jν (z). For large positive values
of z, it can be approximated by
s µ ¶
2 π νπ
Jν (z) ≈ sin z + −
πz 4 2
and its k-th positive zero jν,k , for large values of k, is approximately given by
µ ¶
ν 1
jν,k ≈ k+ − π.
2 4
41
References
The asymptotic behavior of the adiabatic radial oscillations has been studied by Ledoux
(1962, 1963) using the Langer (1935) method. We have essentially followed that method
here. Tassoul and Tassoul (1968) have obtained a more precise approximation (of the sec-
ond order) using Olver’s (1956) method. Through a different choice of the large parameter,
Ruymaekers and Smeyers (1991) have improved the above approximations.
Langer R.E., 1935. On the asymptotic solutions of ordinary differential equations, with
reference to the Stokes phenomenon about a singular point. Trans Am Math Soc, 37,
397–416.
Ledoux P., 1962. Sur la forme asymptotique des pulsations radiales adiabatiques d’une
étoile. I. Bull Acad Roy Belg, Cl des Sci, 5e série, 48, 240–254.
Ledoux P., 1963. Sur la forme asymptotique des pulsations radiales adiabatiques d’une
étoile. II. Comportement asymptotique des amplitudes. Bull Acad Roy Belg, Cl des
Sci, 5e série, 49, 286–302.
Olver F.W.J., 1956. The asymptotic solution of linear differential equations of the second
order in a domain containing one transition point. Phil Trans Roy Soc London, A, 249,
65–97.
Ruymaekers E., Smeyers P., 1991. Second-order asymptotic approximation for radial
oscillations of a gaseous star. Astron Astrophys, 241, 142–452.
Tassoul M., Tassoul J.-L., 1968. Asymptotic approximations for stellar pulsations. Ap J,
153, 127–133.
42
Chapter 10
Vibrational stability
Consider a dynamically stable stellar model. We can calculate the frequencies of the
normal modes of oscillation with the adiabatic approximation, but we cannot determine
whether the modes are excited or damped by thermal processes. We would also like to
know what are the excitation mechanisms of a mode observed in a variable star. It is
necessary to take into account the non adiabatic terms and to solve the fourth order
differential system to get this kind of information.
We have already established the relation
Z Z ½ ¯ ¯2 ¾
δP δρ 4r dP ¯ δr ¯
¯ ¯
s2 |δr|2 dm + + ¯ ¯ dm = 0.
¯ r ¯
ρ ρ ρ dr
We can take the imaginary part of it
Z
δP δρ
= dm
ρ ρ
2 <s =s = − Z .
|δr|2 dm
We have also shown that the expression in the denominator is related to the mechanical
energy of the oscillation. There is also a simple physical interpretation for the numerator.
Consider a gram of matter, undergoing the thermodynamical cycle of period τ described
by the equations
P (t) = P0 + a cos(φ − σt),
ρ(t) = ρ0 + b cos(ψ − σt).
Using the usual conventions, this can also be written as
δP (t) = δP e−iσt (δP = aeiφ ),
δρ(t) = δρ e−iσt (δρ = beiψ ).
The work done by this system during one cycle can be written as
I Z τ Z τ
dV P dρ
T = P dV = P dt = − dt
0 dt 0 ρ2 dt
πab ³ δP δρ ´
= sin(φ − ψ) = π= .
ρ2 ρ ρ
43
The average power of the thermodynamical cycle can then be written as
T σ ³ δP δρ ´
W= = = .
τ 2 ρ ρ
R
Let W = W dm be the average power developed by the whole star . We have
Z
δP δρ 2
= dm = W.
ρ ρ σ
We can transform the numerator as done above to obtain the cubic equation in s. More
simply, we can use the cubic equation
1
s2 + A + B = 0.
s
We take its imaginary part, and we get
1Z δρ ³ d δL ´
= (Γ3 − 1) δ² − dm
s ρ dm
2 <s =s = − Z
|δr|2 dm
10.1 The quasi-adiabatic approximation

The adiabatic approximation is usually excellent in the internal layers of the star and
fails only in its external layers. Therefore we consider the non conservative terms as
small perturbations to the adiabatic case. We can then obtain some corrections to the
eigenfunction and to the eigenvalue s using a perturbation method. Since we are mostly
interested in knowing whether a given mode is excited or damped by the non conservative
terms, we are interested in the real part of s. It can be obtained simply from the above
expression, by substituting the adiabatic solutions in the right-hand side.
We write s = σ 0 − iσ, where σ 0 will be called the growth coefficient (or −σ 0 the damping
coefficient). We have
Z
δT ³ d δL ´
1 δ² − dm
σ0 = 2 TZ dm .
2σ |δr|2 dm
There is a simple physical explanation for this expression. We recall that
σ Z δP δρ 1 Z δT ³ d δL ´
W = = dm = δ² − dm ,
2 ρ ρ 2 T dm
σ2 Z
E= |δr|2 dm ,
2
so that we get
W
σ0 = .
2E
44
This result can be justified as follows. If the amplitude of the oscillation grows exponen-
0 0
tially as eσ t , its energy grows as e2σ t . We then have
1 dE W
2σ 0 = = .
E dt E
An obvious benefit of the integral expression of the growth coefficient is the interpretation
we have made for its numerator. On one hand, the excitation or damping role of each
stellar layer can be assessed. If it brings a positive (negative) contribution to the integral,
it has an exciting (damping) role for the oscillation. On the other hand, this expression
allows us to determine the mechanism responsible for the excitation or the damping (i.e.
whether it is due to the transport or to the energy generation term).
There is however a serious problem with the quasi-adiabatic approximation. The adiabatic
eigenfunctions used to calculate the integral are not a valid approximation in the external
layers. A transition zone is often defined by the relation
cv T ∆m ≈ Lτ
where τ is the period and ∆m the mass above the point under consideration. cv T ∆m
is of the order of the internal energy of the layers above the considered level. Lτ is the
energy radiated during one period. We can then separate the star into three regions
1◦ ) an internal adiabatic zone where cv T ∆m À Lτ . The adiabatic approximation is very
good here.
◦
2 ) the transition zone
3◦ ) an external zone, strongly non adiabatic, where cv T ∆m ¿ Lτ . Here, we will write
the thermal energy conservation equation using the following approximate relation
∆ δL δS
= −scv T .
∆m cv
The variation of δL/L in this zone can be written as
δL scv T ∆m δS
∆ =− .
L L cv
The coefficient in the right-hand side is very small. We conclude that the heat capacity
of these superficial layers is too small to influence δL, which is more or less constant
through the whole zone. One must be careful however when an abundant element
is partially ionized, because the ionization process can still absorb or liberate large
quantities of energy.
Due to its small heat capacity, the strongly non adiabatic zone cannot strongly affect the
excitation or the damping of the oscillation. However, the δL calculated using the adia-
batic approximation grows in the external layers of the star. The disagreement between
the true behavior of δL and the one deduced from the adiabatic approximation can badly
influence the result of the quasi-adiabatic calculation. The problem is often solved by
excluding the strongly non adiabatic zone from the integration domain.
Z M Z m∗
δT ³ d δL ´ δT ³ d δL ´
δ² − dm −→ δ² − dm .
0 T dm 0 T dm
45
The following equality is often used as a cutoff criterion for the integration
¯ δT ¯ ¯ δT ¯
¯ ¯
¯ ¯ = ¯¯ ¯
¯
T ad T nonad
where ¯ δT ¯ ¯ δρ ¯ ¯ δT ¯ ¯ δS ¯
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ 1 ¯¯ d δL ¯¯
¯ ¯ = (Γ3 − 1)¯ ¯ and ¯ ¯ =¯ ¯= ¯ ¯.
T ad ρ T nonad cv σcv T dm
This process is quite crude. When the excitation mechanism lies in the central regions
of the star, it is legitimate to hope that the result will be meaningful. The situation is
more delicate when the excitation mechanism lies in the external layers of the star (for
example in the transition zone). It is then more prudent to integrate the whole system of
equations rather than to use the quasi-adiabatic approximation.
When we integrate directly the non adiabatic equations the coefficient of δS/cv in the
thermal energy equation is very large in the internal layers of the star, and this can create
some problems for the numerical integration. They can however be solved. Another
difficulty comes from the fact that σ 0 is small compared to σ, which makes it hard to
evaluate precisely the growth coefficient. In this case it is possible to improve the precision
of σ 0 by using the integral expression with the eigenfunctions of the non adiabatic case.
10.2 The nuclear excitation

The nuclear term always has an exciting influence on the pulsation. The nuclear energy
generation generally happens in the internal layers of the star, where the adiabatic approx-
imation is excellent. The nuclear contribution to the numerator of the growth coefficient
can be written as
Z Z ³ δρ ´2
δT
δ² dm = (Γ3 − 1)[²ρ + (Γ3 − 1)²T ] ² dm > 0 .
T ρ
For main sequence stars, ² decreases very rapidly from the center. Only the central
regions will participate in this destabilizing effect. We have seen that the eigenfunctions
are usually small in the central regions. This is unfavorable to the exciting influence of
the nuclear reactions, which is in competition with the generally damping effect of the
transport terms (see later).
The excitation mechanism relying on the nuclear energy generation is called the ² mech-
anism.
For massive main sequence stars, the radiation pressure is larger and Γ1 is close to 4/3.
As a result, the eigenfunctions grow at a slower rate towards the exterior (for Γ1 =
4/3, we have δr/r = constant). This case is more favorable to the development of a
vibrational instability of nuclear origin. Calculations indeed show that main sequence
stars with a mass higher than some critical mass are vibrationnally unstable. The exact
value of the critical mass depends on the chemical composition and on the opacity and
the energy generation laws used. It goes from 90 M¯ for metal-poor stars to 120–150 M¯
for population I stars (Stothers, 1992). These stars however suffer from a more violent
instability associated to the strange modes (see later).
46
10.3 The influence of the transport terms

The contribution of the transport term to the numerator of the expression for σ 0 is given
by Z Z
δT d δL δT d δL
− dm = − dr .
T dm T dr
δT d δL
− is positive (destabilizing effect) if δL decreases towards the exterior when
T dm
δT > 0, i.e. if the matter absorbs the heat at high temperature and gives it back at
low temperature. This is the usual mechanism of a thermodynamical engine.
Given the rapid growth of the eigenfunctions towards the exterior, it is the exterior layers
which will contribute the most to the integral and more exactly the transition zone. We
recall however that δL tends to be constant in the exterior layers (unless an abundant
element is partially ionized).
Let’s try to estimate the sign of the contribution of the transport term, where the adiabatic
approximation is still valid. We have
δT δρ
≈ (Γ3 − 1)
T ρ
and we get from the transport equation
d(δT /T )
δL dr δr δT δκ
= +4 +4 − .
L d ln T r T κ
dr
For low order modes we can neglect the δT /T derivative term and the δr/r term to get
δL δρ
≈ [(4 − κT )(Γ3 − 1) − κρ ] .
L ρ
In usual circumstances (e.g., Kramers opacity law) we have
Γ3 ≈ 5/3 , κρ ≈ 1 , κT ≈ −3.5 .
Therefore
[(4 − κT )(Γ3 − 1) − κρ ] ≈ 4 > 0 .
δL/L has thus the same sign than δρ/ρ and δT /T and grows in magnitude. As a result
δT d δL
− < 0,
T dr
which shows that the transport terms generally have a stabilizing effect.
To have an exciting effect, it is neccessary to change the sign of the coefficient for the
transport terms. This can happen if one of the following conditions is satisfied.
1◦ ) If Γ3 − 1 is small enough. This can happen in a region where an abundant element is
partially ionized.
47
Table 10.1: Ratio between the growth time and the period for some variables.
Type of variable τ 0 /τ
Classical Cepheids and RR Lyrae 102 to 103
δ Sct 104 to 106
W Vir 10 to 20
long period variables (Mira) 1 to 10
2◦ ) If κT is positive, which happens in the external layers, due to the presence of the ion
H− .
This excitation mechanism which relies on the increase of the opacity during an adiabatic
compression is called the κ mechanism. It is also sometimes called the γ mechanism, when
one wants to insist on the role played by the decrease of Γ3 − 1 in the exciting region.
The variability of some intrinsically variable stars can be explained by a vibrational insta-
bility due to the transport terms and resulting from the partial ionization of an abundant
element. The mechanism lies in the partial second ionization zone of helium (He+ * )He++ )
for variables of the instability strip: RR Lyr, δ Cep, W Vir, RV Tau, δ Sct. In the case
of Mira-type variables , the partial ionization of hydrogen (H* )H+ ) could be responsible
for the instability, and in the case of β Cep type variables, the cause could be an increase
of the opacity due to iron around 200000 K.
Table 1 gives the order of magnitude of the ratio between the growth time τ 0 = 1/σ 0 and
the period τ .
10.4 Strange modes

This label designates dynamical modes initially discovered in very luminous stellar models
and whose behavior was very puzzling. Thus, in very massive (50 to 150 M¯ ) main
sequence models, the frequencies of the strange modes evolve differently from those of
the other modes when the stellar mass is changed. At the points where the strange
modes frequencies should cross the frequencies of the regular modes, we see either avoided
crossing or the development of an instability (the two modes whose frequencies are crossing
acquire <s of opposite signs). In many cases, the strange modes are strongly non adiabatic
and their existence seems to be linked to a much larger radiation pressure than gas pressure
and to a density inversion in an external convective zone. The pulsation energy is confined
in a small region including and above the density inversion (trapped mode). In massive
main sequence stars these modes are responsible for the instability for masses larger than
a critical mass going from 60 M¯ for Z = 0.03 to more than 120 M¯ for metal poor stars.
These strange modes could also be responsible for violent instabilities in LBV variables.
They were also found in hydrogen deficient stars, in low-mass supergiants, in the stars at
the center of planetary nebulae, in Wolf-Rayet stars, and even in cepheids.
Buchler et al. (1997) showed that strange modes exist in weakly non adiabatic stars
(cepheids) and that their behavior can be explained in simple mechanical terms. The
48
equations for the pulsation can be written under a form similar to the Schrödinger equa-
tion. The stars with strange modes are characterized by the existence of a potential
barrier (potential meaning here a term which plays the same role as a potential in the
Schrödinger equation) which allows the trapping of the modes (the strange modes) in the
external stellar layers. The trapping of the strange modes explains the particularities of
their behavior. Another point of view and a detailed bibliography can be found in the
paper by Saio et al. (1998).
References
We assumed a model in thermal equilibrium in our discussion. However, thermal equilib-
rium is not satisfied in stars during some phases of their evolution, and a large fraction
of the radiated energy is provided by the gravific contraction. More information on stel-
lar vibrational stability during those evolution phases can be found in Demaret (1974ab,
1975ab, 1976) and Demaret and Perdang (1977).
Buchler J.R., P.A. Yecko, Z. Kolláth, 1997. The nature of strange modes in classical
variable stars. Astron Astrophys, 326, 669–681.
Demaret J., 1974a. Vibrational stability of stars in thermal imbalance: a solution in terms
of asymptotic expansions. Bull Acad Roy de Belgique, Cl des Sciences, 5e série, 60,
183–190.
Demaret J., 1974b. Vibrational stability of stars in thermal imbalance: a solution in terms
of asymptotic expansions. I. Isentropic oscillations. Astrophys Space Sci, 31, 305–331.
Demaret J., 1975a. Vibrational stability of stars in thermal imbalance: a solution in terms
of asymptotic expansions. II. The general non-isentropic oscillations. Astrophys Space
Sci, 33, 189–213.
Demaret J., 1975b. Vibrational stability of stars in thermal imbalance. In Phénomènes
hydrodynamiques dans les étoiles, 19e colloque international d’astrophysique de Liège,
Mém Soc Roy Sci de Liège, 6e série, 8, 161–171.
Demaret J., 1976. Vibrational stability of stars in thermal imbalance. III. A general
discussion of energy methods. Astrophys Space Sci, 45, 31–45.
Demaret J., Perdang J., 1977. Vibrational stability of stars in thermal imbalance. IV.
Towards a definition of vibrational stability. Astrophys Space Sci, 52, 137–167.
Saio H., Baker N.H., Gautschy A., 1998. On the properties of strange modes. Mon Not
Roy Astr Soc, 294, 622–634.
Stothers R.B., 1992. Upper limit of the mass of pulsationally stable stars with uniform
chemical composition. Astrophys J, 392, 706–709.
49
Chapter 11
The pulsation mechanism in the

instability strip and the light phase
lag
Stability calculations of stellar models in the instability strip give results which are more
or less in agreement with the observations. They show that, in the instability strip, the
fundamental radial mode or the first harmonic are vibrationally unstable. This vibrational
instability lies in the second helium ionization zone of the envelope. The hydrogen ioniza-
tion zone, which roughly coincides with the first helium ionization zone, also contributes,
although to a smaller extent, to the instability.
The calculations also predict correctly the position of the blue limit (i.e. the left limit in
the HR diagram, on the high temperature side) of the instability strip, but they fail for the
red limit. This is due to the fact that when the effective temperature is high, convection
transports little energy whereas at lower effective temperatures the transport of energy
by convection plays an important role. The lack of a satisfactory theory of convection
in the presence of pulsations is probably enough to explain the failure to determine the
instability limit on the right side of the HR diagram.
The theory also reproduces the period-luminosity relation obeyed by the cepheid stars
and used to estimate stellar distances. This relation is even obeyed by the δ Sct variables
as shown in figure 11.1.
11.1 The existence of an instability

Cox (1967) gave a simple interpretation of the instability strip.
The effect of the ionization of an abundant element is to lower Γ3 −1 (figure 11.2) and this
is favorable to a vibrational instability because of the transport term. We examine this
argument in more details. We have established that in the quasi-adiabatic approximation,
δL δρ
≈ [(Γ3 − 1)(4 − κT ) − κρ ] .
L ρ
50
Figure 11.1: Period-luminosity diagram for cepheids and δ Sct variables pulsating in
fundamental mode (Fernie, 1992).
Figure 11.2: Behavior of Γ3 − 1 in the ionization region of He+ in a stellar envelope model
(Cox, 1967).
51
Figure 11.3: Behavior of δL/L in the superficial layers of a stellar model (Cox, 1967).
If the ionization zone is within the adiabatic zone of the star, δL/L will behave as described
in figure 11.3 during compression. In the internal part of the ionization zone δL decreases
during compression. This zone therefore absorbs energy at high temperature and its work
is positive. This destabilizing effect is balanced in the external part of the ionization zone
in which we can carry the opposite reasoning. An ionization zone situated in the adiabatic
zone of the star is therefore unable to create a vibrational instability.
An ionization zone in the strongly non-adiabatic zone is also unable to create a vibrational
instability because δL is basically constant in these external layers of low heat capacity
(figure 11.4).
The most favorable case for the development of the vibrational instability happens when
the ionization region coincides with the transition region. If the pulsation ceases to be
adiabatic in the external part of the ionization region, δL will tend to become independent
of r and the positive work done in the internal part of the ionization region will no longer
be compensated. (figure 11.5).
Detailed calculations confirm this interpretation. The instability strip is the region in the
HR diagram where the transition region and the helium second ionization region coincide.
To the left of this instability strip, for higher effective temperatures, this ionization region
lies in the strongly non-adiabatic region. To the right of the instability strip, for lower
effective temperatures, this ionization zone lies in the adiabatic region. The hydrogen
ionization region now coincides whith the transition zone, but the important role of the
convection in the energy transport complicates the process.
11.2 The light phase lag

According to the adiabatic theory, the maximum of the luminosity should correspond
to the minimum of the radius. However, for the variables of the instability strip, the
52
53
Figure 11.6: The pulsation of δ Cep: (a) light curve, (b) temperature, (c) radius, (d)
radial velocity (Petit, 1987).
54
maximum of the luminosity corresponds to the maximum of the expansion velocity (fig-
ure 11.6). For a sine oscillation, the phase lag over the adiabatic prediction should then
be of a quarter of the period. It is slightly smaller, because of the asymmetry of the light
curves and of the radial velocity. Detailed calculations reproduce nicely this phase lag
and show that it is due to the hydrogen ionization zone. It can be explained with a simple
linear theory (Castor 1968 et 1971, Cox 1980). We will only describe it briefly here.
Even though the hydrogen ionization zone corresponds to temperatures between 8000 K
and 15000 K, it is very narrow and corresponds to a small fraction (of the order a twen-
tieth) of the pressure scale height. This ionization zone can therefore be considered as a
discontinuity (as in a phase transition). During the pulsation, this discontinuity moves
through the stellar mass. Let assume that below the ionization front, δL and δr have
opposite phases. At the minimum of the radius, δL > 0 below the ionization front. The
latter absorbs energy and therefore moves through the stellar mass towards the exterior.
It is only a quarter of a period later, when δL goes through 0 below the ionization front,
that this one will reach its most exterior position. But the stellar layers above the ion-
ization front have a very simple structure which depends mostly on the position of the
ionization front. This is why the effective temperature is in phase with the position of
the ionization front and reaches its maximum when the latter reaches its most exterior
position.
This mechanism, responsible for the light phase lag, cannot exist in stars whose effective
temperature is higher than 104 K. This is in agreement with the observations: β Cep
variables do not exhibit any phase lag.
Note
The ionization potential of H is 13.6 eV. For helium, it is 24.6 eV for the first ionization
and 54.4 eV for the second. The second helium ionization zone is around a temperature
of 40000 K, while the hydrogen ionization zone is around 10000 K.
References
Castor J.I., 1968. A simplified picture of the cepheid phase lag. Ap J, 154, 793–798.
Castor J.I., 1971. On the calculation of linear, non adiabatic pulsations of stellar models.
Ap J, 166, 109–129.
Cox J.P., 1967. The linear theory: initiation of pulsational instability in stars. In Thomas
R.N. (edit.), Aerodynamic phenomena in stellar atmospheres, IAU Symp n◦ 28. Aca-
demic Press, 3–72.
Cox J.P., 1980. Theory of stellar pulsation. Princeton University.
Fernie J.D., 1992. A new approach to the cepheid period-luminosity law: δ Scuti stars as
small cepheids. Astron J, 103, 1647–1651.
Petit M., 1987. Variable stars. Wiley and Sons.
55
Chapter 12
Non radial oscillations
12.1 Spherical coordinates

The spherical coordinates and the cartesian coordinates are related through:
x = r sin θ cos φ ,
y = r sin θ sin φ ,
z = r cos θ .
We will use the local cartesian basis ~er , ~eθ , ~eφ . The differential expressions for these
vectors are:
d~er = ~eθ dθ + sin θ ~eφ dφ ,

d~eθ = −~er dθ + cos θ ~eφ dφ ,
d~eφ = −(sin θ ~er + cos θ ~eθ ) dφ .
The expressions for the differential operators applied to the coordinates or the basis vectors
can be easily deduced from the preceding relations.
grad r = ~er , div ~er = 2r , curl ~er = 0 ,

1
grad θ = r ~eθ , div ~eθ = r cotg θ , curl ~eθ = 1r ~eφ ,
1
grad φ = r sin1 ~e , div ~e = 0 , curl ~eφ = 1r cotg θ ~er − 1r ~eθ .

θ φ φ
Let α be a scalar field and ~a be a vector field. We will write ~a as
~a = ar~er + aθ~eθ + aφ~eφ .
We easily get
∂α 1 ∂α 1 ∂α
grad α = ~er + ~eθ + ~eφ ,
∂r r ∂θ r sin θ ∂φ
1 ∂ ³ 2 ´ 1 ∂ 1 ∂aφ
div ~a = 2 r ar + (aθ sin θ) + ,
r ∂r r sin θ ∂θ r sin θ ∂φ
56
Figure 12.1: The spherical coordinates.

" # " #
1 ∂ 1 ∂aθ 1 ∂ar 1 ∂
curl ~a = (aφ sin θ) − ~er + − (raφ ) ~eθ
r sin θ ∂θ r sin θ ∂φ r sin θ ∂φ r ∂r
" #
1 ∂ 1 ∂ar
+ (raθ ) − ~eφ ,
r ∂r r ∂θ
Ã ! Ã !
1 ∂ 2 ∂α 1 ∂ ∂α 1 ∂ 2α
∆α = 2 r + 2 sin θ + 2 2 .
r ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂φ2
Let Ã !
2 1 ∂ ∂ 1 ∂2
L =− sin θ − ,
sin θ ∂θ ∂θ sin2 θ ∂φ2
where we recognize the square angular momentum operator used in quantum theory. It
is, in a way, the angular part of the laplacian.
Ã !
1 ∂ ∂α 1
∆α = 2 r2 − 2 L2 α .
r ∂r ∂r r
The spherical functions Ylm (θ, φ) are the eignefunctions of L2 :
L2 Ylm (θ, φ) = `(` + 1)Ylm (θ, φ) .
12.2 Perturbation equations

Consider a perturbation with a est time dependence. We write the displacement vector as
−
→
δr = δr ~er + r δθ ~eθ + r δφ sin θ ~eφ .
57
→
−
Note that here δr is not the magnitude of δr, but rather its radial component.
Using the eulerian perturbations, the differential equations of the problem can be written
as follows.
Continuity equation:
( )
0 dρ 1 ∂ 1 ∂ ∂ δφ
ρ + δr + ρ 2 (r2 δr) + (sin θ δθ) + = 0.
dr r ∂r sin θ ∂θ ∂φ
Momentum equations:
∂Φ0 ρ0 dP 1 ∂P 0
s2 δr = − + 2 − ,
∂r ρ dr ρ ∂r
1 ∂Φ0 1 ∂P 0
s2 r δθ = − − ,
r ∂θ ρr ∂θ
1 ∂Φ0 1 ∂P 0
s2 r sin θ δφ = − − .
r sin θ ∂φ ρr sin θ ∂φ
Poisson equation: Ã !
0
1 ∂ 2 ∂Φ 1 2 0
r − L Φ = 4πGρ0 .
r2 ∂r ∂r r2
Energy equation:
Ã !
0 dS ρ0 1 d 2
sT S + δr = ²0 + (r F )
dr ρ2 r2 dr
( )
1 1 ∂ 2 0 1 ∂ 0 1 ∂Fφ0
− (r Fr ) + (sin θ Fθ ) + .
ρ r2 ∂r r sin θ ∂θ r sin θ ∂φ
Transport equations:
dT ∂T 0
Fr0 = −λ0 −λ ,
dr ∂r
λ ∂T 0
Fθ0 = − ,
r ∂θ
λ ∂T 0
Fφ0 = − .
r sin θ ∂φ
We can write P 0 and T 0 in terms of ρ0 , S 0 and the components of the displacement. There
remains 9 unknowns: ρ0 , S 0 , δr, δθ, δφ, Fr0 , Fθ0 , Fφ0 and Φ0 , which must satisfy the 9 partial
differential equations.
This problem is quite complicated. In particular, it is not easy to write the surface
boundary conditions for the components of the flux. We will restrict ourselves to the
study of the non-radial, adiabatic oscillations. The energy equation is replaced by δS = 0.
−
→
It is no longer necessary to determine F 0 and the transport equations become unnecessary
as well.
58
δθ and δφ can be determined from the momentum equations:

1 ∂χ
δθ = − ,
∂θs2 r2
1 ∂χ
δφ = − 2 2 2 ,
s r sin θ ∂φ
where we have defined for simplicity χ = Φ0 + P 0 /ρ. We substitute these expressions in
the continuity equation:
dρ ρ ∂ ρ
ρ0 + δr + 2 (r2 δr) + 2 2 L2 χ = 0 .
dr r ∂r sr
We develop δr, Φ0 , ρ0 and P 0 as series of spherical functions:
∞ X̀
X
δr(r, θ, φ, t) = δr`m (r)Y`m (θ, φ)est ,
`=0 m=−`
···
In these differential equations, the derivatives with respect to θ and φ only appear in the
L2 operator. As the spherical functions are eigenfunctions of this operator, the equations
can be separated. We get ordinary differential equations for the radial functions δr`m (r),
. . . We therefore get, for each couple (`, m), a fourth order differential system of the form
(we omit the indices ` and m):
Ã !
dρ ρ d ρ`(` + 1) P0
ρ + δr + 2 (r2 δr) +
0
Φ0
+ = 0,
dr r dr s2 r2 ρ
0 0 0
dΦ ρ dP 1 dP
s2 δr = − + 2 − ,
dr ρ dr ρ dr
Ã !
0
1 d 2 dΦ `(` + 1) 0
2
r − Φ = 4πGρ0 .
r dr dr r2
We still have to specify the boundary conditions which must be satisfied by the solutions
of this system. At r = 0, some coefficients of the differential system are singular. We’ll
impose that the solutions remain regular. A series development shows that we must
impose two boundary conditions at the center, and that in its neighborhood we must
have
δr ∝ r`−1 , P 0 and Φ0 ∝ r` .
At the stellar surface, we’ll impose that δP = 0. It is unnecessary to refine this condition
when using the adiabatic approximation.We’ll also impose the continuity of the gravity
potential and of its gradient. To write it, remember that outside the star Φ0e (we will use
the index e to designate the exterior) satisfies the Laplace equation
Ã !
0
1 d 2 dΦ `(` + 1)Φ0
r − = 0,
r2 dr dr r2
whose regular solution (i.e. the one which goes to zero at infinity) can be written as
A
Φ0e =
r`+1
59
where A is a constant.
The continuity of the potential and its derivative with respect to r at the stellar surface
can be written as:
δΦ = δΦe ,
dΦ dΦe
δ =δ ,
dr dr
or
dΦ dΦe
Φ0 + δr = Φ0e + δr ,
dr dr
dΦ0 d2 Φ dΦ0e d2 Φe
+ δr 2 = + δr 2 .
dr dr dr dr
We note that the first derivatives of the potential are equal
dΦ dΦe
= .
dr dr
The second derivatives are equal only if the density goes to zero at the surface. Indeed
we have
d2 Φe 2 dΦe
+ = 0,
dr2 r dr
d2 Φ 2 dΦ
+ = 4πGρ .
dr2 r dr
Substracting these equations we get
d2 Φe d2 Φ
− 2 = −4πGρ .
dr2 dr
The conditions of continuity then give
A
Φ0 = ,
r`+1
dΦ0 (` + 1)A
=− − 4πGρ δr .
dr r`+2
We eliminate the constant A to get the required condition
dΦ0 ` + 1 0
+ Φ + 4πGρ δr = 0 .
dr r
We must now solve, for each value of (`, m), a homogeneous system of ordinary differential
equations with boundary conditions. For an arbitrary value of s, the only solution is zero.
It is only for some particular values of s, called eigenvalues, that non-zero solutions will
exist. For each value of (`, m), there is an infinity of solutions, which we will write as
sk`m .
60
The m index does not appear in the differential equations, nor in the boundary conditions.
Therefore, we have
sk`m = sk`m0 .
The eigenvalues can therefore be written using two indices only sk` . Each eigenvalue thus
corresponds to 2` + 1 different values of m, and therefore to 2` + 1 different non radial
oscillations. One says that it is 2` + 1 times degenerate. The eigenfunctions describing
these 2`+1 modes have the same radial factor, and differ only through their angular factor.
This degeneracy is due to the spherical symmetry of the equilibrium configuration. It also
appears in the theory of the hydrogen atom in quantum theory. It is possible to show the
existence of this degeneracy through the theory of groups. This degeneracy can be lifted
by something which breaks the spherical symmetry, such as rotation.
12.3 The Cowling approximation

In non radial oscillations, the perturbations of the gravific potential are generally rather
small. This is especially true for the high order modes (k or ` high). If we neglect them,
we get a second order system of differential equations, which is easier to solve. This is the
Cowling approximation.
dρ ρ d `(` + 1)P 0
ρ0 + δr + 2 (r2 δr) + = 0,
dr r dr s2 r2
ρ0 dP 1 dP 0
s2 δr = 2 − .
ρ dr ρ dr
We assume that Γ1 is constant and make the following change of variables
v = r2 δr P 1/Γ1 , w = P 0 /P 1/Γ1 .
We use s = −iσ, and after some calculations we get
dv ³ L2 ´ r 2 P 2/Γ1
`
= −1 w,
dr σ2 ρc2
dw ρ
= (σ 2 − n2 ) 2 2/Γ1 v .
dr r P
At the boundaries r = 0 and r = R we impose v = 0. The parameters L` and n have the

dimensions of frequencies. They are the Lamb frequency and the Brunt-Väisälä frequency,
respectively; they are defined by the relations
`(` + 1)c2
L2` = and n2 = −Ag ,
r2
with
d ln ρ 1 d ln P
A= − .
dr Γ1 dr
We recall that the criterion of stability towards convection (Schwarzschild criterion) is
given by A < 0 or n2 > 0.
If Γ1 is not constant, we can get a similar system through a more complicated change of
variables.
61
References
The reader will find in Dupret (2001) and Dupret et al. (2002) a detailed treatment of the
outer boundary conditions in the non radial non adiabatic case and a detailed treatment
of the pulsation in the very external layers of the star.
It is possible to show through the theory of groups that the frequencies degeneracy results
from the spherical symmetry of the unperturbed configuration (Perdang, 1968).
The change of variables which must be done in the case of the Cowling approximation
when Γ1 is not constant, is given by Gabriel and Scuflaire (1979).
Dupret M.-A., 2001. Nonradial nonadiabatic stellar pulsations: A numerical method and
its application to a β Cephei model. Astron Astrophys, 366, 166–173.
Dupret M.-A., De Ridder J., Neuforge C., Aerts C., Scuflaire R., 2002. Influence of
non-adiabatic temperature variations on line profile variations of slowly rotating β Cep
star and SPBs. I. Non-adiabatic eigenfunctions in the atmosphere of a pulsating star.
Gabriel M., Scuflaire R., 1979. Properties of non-radial stellar oscillations. Acta Astron,
29, 135–149.
Perdang J., 1968. On some group-theoretical aspects of the study of non-radial oscilla-
tions. Astrophys Space Sci, 1, 355–371.
62
Chapter 13
Non radial modes
13.1 Orthogonality of the eigenfunctions

We have seen previously that the momentum equation of an adiabatic perturbation can
be written as
−
→ −
→
σ 2 δr = L δr ,
where L is a self-adjoint operator relative to the scalar product
Z
(~u, ~v ) = ρ~u · ~v dV .
The properties we established previously are valid for non radial adiabatic oscillations. In
particular the eigenfunctions of this problem can be chosen to be orthogonal. Note that
the eigenfunctions corresponding to the same frequency but with different ` or m indices
are orthogonal because of the orthogonality of the spherical functions Y`m .
13.2 Components of the displacement

For a given mode we have
−
→
δr = δr ~er + r δθ ~eθ + r sin θ δφ ~eφ
Ã !
1 ∂χ 1 ∂χ
= δr ~er + 2 ~eθ + ~eφ ,
rσ ∂θ sin θ ∂φ
δr = δr(r) Y`m (θ, φ) e−iσt ,
χ = χ(r) Y`m (θ, φ) e−iσt .
We write
→
−
δr = (a(r)~² + b(r)~η ) e−iσt ,
with
a(r) = δr(r) ,
63
χ(r)
b(r) = ,
rσ 2
~² = Y`m (θ, φ) ~er ,
∂Y`m 1 ∂Y`m ∂Y`,m imY`,m
~η = ~eθ + ~eφ = ~eθ + ~eφ .
∂θ sin θ ∂φ ∂θ sin θ
We note that
Z
|²|2 dΩ = 1 ,
Z
|η|2 dΩ = `(` + 1) .
We justify the last equality as follows

Z Z 2π Z π h¯ ¯ i
¯ ∂Y`m ¯2 m2 2
|η|2 dΩ = dφ ¯ ¯ + 2 |Y `m | sin θ dθ
0 0 ∂θ sin θ
Z 2π nh ∂Y`m iπ
= dφ sin θ Y `m
0 ∂θ 0
Z πh
1 ∂³ ∂Y`m ´ m2 i o
+ − sin θ + Y `m Y `m sin θ dθ .
0 sin θ ∂θ ∂θ sin2 θ
The integrated term gives zero and
h 1 ∂³ ∂Y`m ´ m2 i
2
− sin θ + Y`m = L Y`m = `(` + 1)Y`m .
sin θ ∂θ ∂θ sin2 θ
Therefore Z Z 2π Z π
2
|η| dΩ = `(` + 1) dφ |Y`m |2 sin θ dθ = `(` + 1)
0 0
and finally Z Z h i
−
→
| δr|2 dm = ρr2 a2 + `(` + 1)b2 dr .
13.3 p, g and f modes

The differential system describing the non radial oscillations of a star can only be solved
analytically in the very unrealistic case of a homogeneous model. This case will neverthe-
less reveal the different types of non radial modes (see figure 13.1).
For each couple of indices (`,m), we get an infinity of unstable modes (σ 2 = −s2 < 0). The
values of σ 2 present an accumulation point in 0. The buoyancy force plays a preponderant
role in the dynamics of these modes. They are similar to the internal gravity modes. In
fact they describe the convective instability of the homogeneous model. They are called
g modes (g = gravity).
There is also an infinity of stable modes (σ 2 = −s2 > 0). The values of σ 2 do not have
an accumulation point at finite distance. The pressure forces play a dominant role in the
dynamics of these modes. They are similar to acoustic modes. They are called p modes
(p = pressure).
64
Figure 13.1: Frequencies of non radial modes of the homogeneous model.
Figure 13.2: Frequencies of the non radial modes of a physical model.
Finally when ` > 1, there is a stable mode, whose frequency is lower than those of the p
modes. It is called the f mode (f = fundamental).
For more realistic models, the differential system is too complicated and cannot be solved
analytically. Numerous numerical integrations show that the non radial modes of physi-
cally realistic models can be classified as those of the homogeneous model. The result is
exactly the same if the model is entirely convective. If the model is entirely radiative, the
g modes are stable. Their frequencies are lower than the f mode and p modes frequencies
and have an accumulation point in 0. If the model has both radiative and convective zones,
there are two spectra of g modes, one stable, the other unstable, as shown in figure 13.2.
The stable g modes are labelled g + and the unstable g modes are labelled g − .
In the Cowling approximation it is possible to rigourously demonstrate the existence of
these different types of modes. Less rigorously, we can see that for high values of σ 2 ,
neglecting the 1/σ 2 term, we have a Sturm-Liouville problem with λ = σ 2 as parameter.
65
Table 13.1: Some characteristics of non radial modes of degree ` = 2 for the standard
model.
mode ω ξs /ξc hxi
g10 0.567887 3.977(−3) 0.299
··· ··· ··· ···
g3 1.34992 −2.518(−2) 0.280
g2 1.68171 5.521(−2) 0.278
g1 2.21688 −0.2399 0.292
f 2.85926 3.6763 0.493
p1 3.90687 −57.34 0.702
p2 5.16947 213.0 0.735
p3 6.43999 −453.2 0.742
··· ··· ··· ···
p10 15.2849 4204 0.739
The corresponding solutions are the p modes. For small values of σ 2 , neglecting the σ 2
term, we have a Sturm-Liouville problem with λ = 1/σ 2 as parameter. The corresponding
solutions are the g modes.
The non radial modes can be physically described as follows. The p modes are acoustic
modes. The g − modes describe the convective instability. The g + modes are internal
gravity waves. In very concentrated models, the low k order p and g + modes can present
a mixed character and behave as gravity waves in the central regions of the star and as
acoustic waves in the external layers.
Table 13.1 gives ω, ξs /ξc and hxi for a few non radial modes of the standard model
(polytrope of index 3 with Γ1 = 5/3), with
Z
→
−
δr X| δr|2 dm
ξ = x1−` and hXi = Z −→2 .
R | δr| dm
The figures 13.3 to 13.7 show ξ for a few non radial modes of the standard model. We
note that for the p modes, as for the radial modes, ξ grows in the external layers of the
star. On the contrary, for the g modes, ξ is larger in the central regions.
13.4 Spheroidal and toroidal modes

Do the radial and non radial modes studied so far form a complete set ? We will show
that they do not, and how it can be completed.
Any vector field can be written as
~u = grad φ + curl ~v .
66
Figure 13.3: Standard model, non radial oscillation mode ` = 2, p1 .
Figure 13.4: Standard model, non radial oscillation mode ` = 2, p2 .

67
Figure 13.5: Standard model, non radial oscillation mode ` = 2, g1 .
Figure 13.6: Standard model, non radial oscillation mode ` = 2, g2 .

68
Figure 13.7: Standard model, non radial oscillation mode ` = 2, f .
Under some conditions (behavior of the fields at infinity), this decomposition is unique.
The grad φ term is called the longitudinal component of the field and curl ~v is its transver-
sal component. The transversal component (also called solenoidal component) can be
decomposed as
curl ~v = curl(χ~er ) + curl curl(ψ~er ) .
The curl(χ~er ) term is a toroidal field and curl curl(ψ~er ) is a poloidal field.
A vector field ~u can thus be described by three scalar potentials φ, χ and ψ.
~u = grad φ + curl(χ~er ) + curl curl(ψ~er ) .
The poloidal term can be developed, and we get
~u = α~er + grad β + curl(χ~er ) .
We note that in these two expressions the toroidal term is uniquely determined by the
vector field ~u.
−
→
We now write δr using the above expressions. The momentum equation can be written
as
−
→ 1 ρ0
s2 δr = − grad Φ0 − grad P 0 + 2 grad P .
ρ ρ
Using the adiabatic relation and the continuity equation we get
−
→ ~ div −
→
s2 δr = − grad χ + c2 A δr ,
where
P0
χ = φ0 + ,
ρ
~ = 1 grad ρ − 1 grad P .
A
ρ Γ1 P
69
−
→
We have thus written δr as
→
−
δr = α~er + grad β ,
with
→
−
α = c2 A div δr/s2 ,
β = −χ/s2 .
The non radial modes we have studied so far do not have any toroidal component. It is
therefore obvious that they do not form a complete set. To obtain a complete set we must
also consider the 0 frequency modes, which were so far neglected. We will not develop
this point in details.
The 0 frequency modes are divergenceless. We will distinguish two classes.
1. There are three spheroidal modes ` = 1, m = −1, 0, 1. The radial component is a

non-zero constant. They describe a solid translation of the star.
→
− n ∂Y`m 1 ∂Y`m o
δr = a Y`m~er + ~eθ + ~eφ = constant vector.
∂θ sin θ ∂φ
Formally we could consider them as f modes of degree ` = 1.
2. There is an infinity of toroidal modes. They are characterized by the absence of

radial displacement. They are of the form
−
→ n 1 ∂Y ∂Y`m o
`m
δr = a(r) ~eθ − ~eφ .
sin θ ∂φ ∂θ
The horizontal and divergenceless displacements do not of course perturb the hy-
drostatic equilibrium of the star and we have
ρ0 = 0, P 0 = 0, Φ0 = 0 .
−
→
The eigenfunctions δr of the radial and of the non radial problems (zero and non zero
−
→
frequency modes) form a complete set. Any displacement field δr can be written as a
series in terms of elements of this set.
13.5 Asymptotic expression for the frequencies

The study of the asymptotic behavior of the non radial modes is more complicated than
that of the radial modes, even in the Cowling approximation. From the equations for v
and w obtained previously we can write a second order equation in v or in w. In addition
to the singularities at the center and at the surface there is also a moving singularity at
the position where σ 2 = σa2 in the first case, and one or several moving singularities where
σ 2 = n2 in the second case. As for the asymptotic study of the radial oscillations, the
model is divided into several zones, each with one single singularity. The approximate
70
solutions obtained in each zone are then joined continuously. We will give without proof
the lowest order approximation for the frequencies.
For the p modes we have Ã !
` ne 1
k+ + + π
2 2 4
σkl ≈ Z R .
dr
0 c
For ` = 0, this expression reduces to the one obtained above for radial modes, numbered
as above. We note the frequency equidistance and approximate superpositions given by
σk+1,l − σk,l ≈ const , σkl ≈ σk−1,l+2 and σk,l+1 ≈ (σk,l + σk+1,l )/2.
For the g ± modes, the asymptotic frequencies are given by

Ã !
q `
k+ +C π
`(` + 1) 2
≈ Z .
|σkl | |n|
dr
r
Here, the integral is computed over the radiative zone for a g + mode, and over the con-
vective zone for a g − mode. The constant C depends on the position of the integration
zone in the model (central regions or superficial layer). It is even more complicated if
there are several radiative or convective zones.
References
The paper by Smeyers (1984) on non radial modes is particularly interesting.
For more information on the physical meaning of the different types of modes, we suggest
the papers by Scuflaire (1974ab). Tolstoy (1963) gives a remarquable account of the
theory of plane waves in a simple geometric context.
For a rigorous study of the non radial oscillations in the Cowling approximation, with
variable Γ1 , we recommend the paper by Gabriel and Scuflaire (1979). It also contains
the proof of existence of the p, g ± and f modes. Christensen-Dalsgaard and Gough (2001)
have developed an interesting reflexion about the classification of non radial modes and
the status of the ` = 1 f mode.
−
→
We find a detailed discussion of the decomposition of δr in its spheroidal and toroidal
components in the paper by Aizenman and Smeyers (1977).
We recommend the papers by Kaniel and Kovetz (1967) and Eisenfeld (1969) for further
information about the completeness of the eigenfunctions of the non radial problem.
The asymptotic behavior of the non radial modes in the Cowling approximation is very
well described in Tassoul (1980). There are numerous references and comments on pre-
vious studies in this paper. In later papers, it became possible to go further than the
71
Cowling approximation: Tassoul (1990) and Smeyers et al. (1996) for p modes ; Smeyers
et al. (1995) and Willems et al. (1997) for g modes.
Aizenman M.L., Smeyers P., 1977. An analysis of the linear adiabatic oscillations of a
star in terms of potential fields. Astrophys Space Sci, 48, 123–136.
Christensen-Dalsgaard J., Gough D.O., 2001. On the dipolar f mode of stellar oscillation.
MNRAS, 326, 1115–1121.
Eisenfeld J., 1969. A completeness theorem for an integro-differential operator. J Math
Anal Appl, 26, 357–375.
Gabriel M., Scuflaire R., 1979. Properties of non-radial stellar oscillations. Acta Astron,
29, 135–149.
Kaniel S., Kovetz A., 1967. Schwarzschild’s criterion for instability. Phys Fluids, 10,
1186–1193.
Scuflaire R., 1974a. Space oscillations of stellar non radial eigen-functions. Astron Astro-
phys, 34, 449–451.
Scuflaire R., 1974b. The non radial oscillations of condensed polytropes. Astron Astro-
phys, 36, 107–111.
Smeyers P., 1984. Non-radial oscillations. In Noels A., Gabriel M. (edit.), Theoreti-
cal problems in stability and oscillations, Proceedings of the 25th Liège international
astrophysical colloquium, Université de Liège, Institut d’Astrophysique, 68–91.
Smeyers P., de Boeck I., van Hoolst T., Decock L., 1995. Asymptotic representation of
linear, isentropic g-modes of stars. Astron Astrophys, 301, 105–122.
Smeyers P., Vansimpsen T., de Boeck I., van Hoolst T., 1996. Asymptotic representation
of high-frequency, low-degree p-modes in stars and in the Sun. Astron Astrophys, 307,
105–120.
Tassoul M., 1980. Asymptotic approximations for stellar nonradial pulsations. Astrophys
J Suppl, 43, 469–490.
Tassoul M., 1990. Second order asymptotic approximation for stellar nonradial acoustic
modes. Astrophys J, 358, 313–327.
Tolstoy I., 1963. The theory of waves in stratified fluids including the effects of gravity
and rotation. Rev Mod Phys, 35, 207–230.
Willems B., Van Hoolst T., Smeyers P., 1997. Asymptotic representation of non radial
g + -modes in stars with a convective core. Astron Astrophys, 318, 99–107.
72
Chapter 14
Influence of rotation
We study the effect of a slow rotation of the star on the non radial adiabatic pulsations.
The model undergoes differential rotation around the z-axis at angular velocity Ω(r, θ)
and the rotation is sufficiently slow to ignore the Ω2 terms (if they had to be accounted
for, the model could no longer be considered spherical).
Compared to the case without rotation, the difference appears in the development of the
momentum equation
−
→
d2 δr −
→
2
= L δr,
dt
which now becomes Ã !2
∂ →
− −
→
+ ~v · grad δr = L δr,
∂t
where ~v is the rotation velocity in the unperturbed model,
~v = Ωr sin θ−
→
eφ .
We look for a solution with a time-dependence e−iσt and we neglect the second-order terms
in v (or in Ω), and we get
−
→ →
− −
→
σ 2 δr + 2σM δr + L δr = 0,
where
−
→ →
−
M δr = i(~v · grad) δr.
M is a purely imaginary linear operator (M = −M) and it is easy to show that it is
hermitian. Indeed
Z h i
~ ~η ) =
(Mξ, iρ (~v · grad)ξ~ · ~η dV
Z h i Z Z h i
= i div ρ(ξ~ · ~η )~v dV − i(ξ~ · ~η ) div(ρ~v ) dV − iρ (~v · grad)~η · ξ~ dV.
The first term can be transformed into a surface integral which vanishes, and the second
~ M~η ).
one contains the zero term div(ρ~v ). We then obtain (ξ,
~ it is possible to write
For any function ξ,
J = (ξ, ξ) , M = (ξ, Mξ) and L = (ξ, Lξ)

73
and to solve the equation

JΣ2 + 2M Σ + L = 0
with respect to Σ. We thus define a functional Σ(ξ) which has an interesting variational
property. If ξ is a solution of the momentum equation for a real value of σ then ξ makes
the functional Σ stationary and σ = Σ(ξ).
We now look on an oscillation mode. In the absence of rotation it is described by an
eigenfunction ξ0 and its frequency σ0 obeys the equation
J0 σ02 + L0 = 0 with J0 = (ξ0 , ξ0 ) and L0 = (ξ0 , Lξ0 ).
In the presence of slow rotation its eigenfunction and its eigenfrequency are written as
ξ = ξ0 + ξ1 and σ = σ0 + σ1 ,
where ξ1 and σ1 are small corrections due to the rotation. The frequency obeys the
equation
Jσ 2 + 2M σ + L = 0.
We develop this equation neglecting the terms higher than first order terms in the cor-
rection, noting that M0 = (ξ0 , Mξ0 ) must also be considered as a correction since this
expression contains the rotation velocity of the star. We get
σ1 = −M0 /J0 .
It is easy to show that

Mξ~ = −mΩξ~ + iΩ ~
~ × ξ,
and we get Z · ¸
ρ mΩ|ξ| − i(Ω, ~ ξ~ ) dV
~ ξ,
2
σ1 = Z .
ρ|ξ|2 dV
This expression is zero for a radial mode. For a non radial mode ξ~ can be written as
Ã !
∂Y`m im
ξ~ = aY`m~er + b ~eθ + Y`m~eφ
∂θ sin θ
and we get
2
~ ξ~ ) = 2mΩab|Y`m |2 + mΩb2 ∂|Y`m | cotg θ.
~ ξ,
i(Ω,
∂θ
Finally,
 ¯ ¯ 
Z  ¯ ∂Y ¯2 2 2 
¯ `m ¯ m ∂|Y `m |
m ρΩ (a2 − 2ab)|Y`m |2 + b2 ¯ ¯ + |Y `m | 2
− cotg θ  dV
 ¯ ∂θ ¯ sin2 θ ∂θ 
σ1 = Z .
ρr2 [a2 + `(` + 1)b2 ] dr
74
This expression can be simplified if we assume that Ω = Ω(r). We then get

Z
m ρr2 Ω[a2 + `(` + 1)b2 − 2ab − b2 ] dr
σ1 = Z .
ρr2 [a2 + `(` + 1)b2 ] dr
In the particular case of uniform rotation, this expression can be further simplified to get
0
σ1 = mβΩ or σk`m = σk` + mβk` Ω,
0
where σk` is the eigenfrequency in the absence of rotation and the constant βk` is calculated
from the eigenfunctions of the mode (k, `) without rotation,
Z
ρr2 [a2 + `(` + 1)b2 − 2ab − b2 ] dr
βk` = Z .
ρr2 [a2 + `(` + 1)b2 ] dr
The rotation removes completely the degeneracy.

The rotation of the star also affects the toroidal modes. In the presence of rotation these
modes acquire non zero frequencies and the displacement ceases to be purely horizontal
and toroidal. These modes have low frequency and are similar to the Rossby or planetary
waves. Their dynamics is dominated by the Coriolis force. We will not study them
here. Note that in the case of uniform rotation, their frequencies are given, as first order
approximation, by the relation
2mΩ
σ = mΩ − .
`(` + 1)
14.1 Non radial oscillations in variable stars

The evidence for non radial modes in variable stars lies on several observations: character-
istic deformation of spectral lines, pulsation frequency lower than that of the fundamental
radial mode, frequency ratios incompatible with radial pulsations, frequency multiplets
produced by the frequency splitting due to the rotation.
Some β Cep variables and a number of δ Sct variables exhibit non radial pulsations in
addition to their radial modes.
The variability of the spectral lines in some variable B stars (SPB or slowly pulsating B
stars) comes from the existence of non radial modes in the presence of rotation.
Variable white dwarfs are usually multiperiodic and their periods go from a few hun-
dreds to many thousands of seconds. The observed modes are g modes of low ` degree.
Frequency splitting due to rotation is sometimes observed.
The variability of Ap stars is due to surface inhomogeneities and to rotation. In some of
these stars, small amplitude light variations with periods between 4 and 15 minutes are
superimposed to those. Their study yields regularly spaced frequencies, illustrating the
rotational splitting.
75
The best studied case is the 5 minutes solar oscillation. It is made up of thousands of
modes of all values of the ` index of the spherical functions between 0 and 3000, whose
frequencies are around 3 mHz. For low values of `, these are p modes of order between
10 and 30. For a given value of ` the frequency spacing ∆ν = ∆σ/2π of two consecutive
modes is close to 136 µHz. The relative error on the frequencies is smaller than 10−4 for
most of these modes and of the order 10−5 for some of them, so that the fine structures
due to the rotation can be put forward (the rotational splitting in a multiplet is of the
order of 0.4 µHz).
References
The expression for the frequencies in the presence of rotation was obtained by Ledoux
(1949 and 1951) and by Cowling and Newing (1949) in the case of uniform rotation. The
theory for stars with differential rotation was established by Lynden-Bell and Ostriker
(1967), Aizenman and Cox (1975), Hansen, Cox and Van Horn (1977) and Gough (1981).
Toroidal modes are described in Aizenman and Smeyers (1977). The influence of the
rotation on these modes has been studied by Papaloizou and Pringle (1978), Saio (1982)
and Lee and Saio (1986).
Aizenman M.L., Cox J.P., 1975. Vibrational stability of differentially rotating stars. Ap
J, 202, 137–147.
Aizenman M.L., Smeyers P., 1977. An analysis of the linear adiabatic oscillations of a
star in terms of potential fields. Astrophys Space Sci, 48, 123–136.
Cowling T.G., Newing R.A., 1949. The oscillations of a rotating star. Ap J, 109, 149–158.
Gough D.O., 1981. A new measure of the solar rotation. MNRAS, 196, 731–745.
Hansen C.J., Cox J.P., Van Horn H.M., 1977. The effects of differential rotation on the
splitting of nonradial modes of stellar oscillation. Ap J, 217, 151–159.
Ledoux P., 1949. Contributions à l’étude de la structure interne des étoiles et leur stabilité.
Mém Soc Roy Sci Liège, 4e sér, 9.
Ledoux P., 1951. The nonradial oscillations of gaseous stars and the problem of Beta
Canis Majoris. Ap J, 114, 373–384.
Lee U., Saio H., 1986. Overstable convective modes in uniformly rotating massive main-
sequence stars. MNRAS, 221, 365–376.
Lynden-Bell D., Ostriker J.P., 1967. On the stability of differentially rotating bodies.
MNRAS, 136, 293–310.
Papaloizou J., Pringle J.E., 1978. Non-radial oscillations of rotating stars and their rele-
vance to the short-period oscillations of cataclysmic variables. MNRAS, 182, 423–442.
Saio H., 1982. R-mode oscillations in uniformly rotating stars. Ap J, 256, 717–735.
76
Chapter 15
Helio- and asteroseismology
The principle goal of helioseismology or asteroseismology is to infer solar or stellar struc-

ture from the observed properties of the solar or stellar oscillations. This is often called
the inverse problem. Much more observational data is available for the Sun than for other
stars. We will therefore mostly look at helioseismology.
The simplest inversion method consists in having a family of models with few parameters
and in adjusting their values, for example using the least mean square method, to best
reproduce the observed frequencies. It is thus possible to adjust the helium abundance
and the mixing length parameter in the convective zone. This method does not use
sophisticated calculation techniques but it also does not extract a maximum of information
from the available helioseismic data.
Another method uses the asymptotic expressions of the oscillation frequencies. These
expressions have the advantage of putting forward the most important factors determining
the frequencies, but their validity for not too high order modes is doubtful.
We will describe below some aspects of the numerical inversion methods. The first step is
always to solve the direct problem, which consists in calculating the oscillation frequencies
of a reference solar model. As we have seen, this means solving a system of differential
eigenvalue equations with boundary conditions whose coefficients depend on the equilib-
rium model. The second step consists in getting corrections to those coefficients using
the observed solar values. Of course helioseismology cannot provide a unique solution for
the solar model. There certainly are solar properties that cannot be deduced from the
oscillation frequencies, and in addition there is a finite amount of data, with observational
errors.
15.1 Determination of Ω(r)

The rotational splitting gives information on the rotation angular velocity Ω inside the
Sun. For simplicity we will assume that it only depends on r.
Z
σk`m − σk`0 = m Kk` (r)Ω(r) dr ,
77
where the kernel Kk` (r) is constructed from the eigenfunctions of the (k, `) mode in the
absence of rotation. Using a unique index i for the couple (k, `), the seismic data impose
linear conditions on Ω
Z
Ki (r)Ω(r) dr = wi , i = 1, . . . , N.
It is clear that this finite number of equations does not uniquely determine Ω(r). To each
solution of these equations one can add a function Ω⊥ (r) orthogonal to all the kernels
Ki (r) and obtain a new solution.
Z
Ki (r)Ω⊥ (r) dr = 0 , i = 1, . . . , N.
In addition there are errors on the data wi , which reinforces the uncertainty on Ω(r). The
problem is therefore not only to find an approximate solution of the equations above, but
to find, amongst an infinity of solutions, the one that would best describe the real angular
velocity distribution inside the Sun. In order to do that we would need more information
on Ω(r), coming from a non-helioseismic source. Without such information, it is necessary
to impose arbitrary conditions on Ω(r). We will outline two inversion methods.
Spectral development
Since the component of Ω(r) orthogonal to the Ki (r) is not accessible to observations,
it seems natural to determine only the part of Ω(r) which can be written as a linear
combination of the Ki (r):
N
X
e
Ω(r) = Ωj Kj (r) .
j=1
The coefficients Ωj must satisfy the following equations

N
X Z
Aij Ωj = wi , i = 1, . . . , N with Aij = Aji = Ki (r)Kj (r) dr .
j=1
The problem is that the matrix A is almost singular. Small errors on the wi lead to large
errors on the Ωj . The solution that would be obtained through the direct solving of this
system of equations would be dominated by the errors on the data and would be totally
wrong. This is called an ill-posed problem.
To show where the problem lies, we write the symmetric positive-definite matrix A as
A = U diag(λ1 , . . . , λN ) Ue ,
where U is orthogonal and the eigenvalues are ordered by decreasing values λ1 ≥ λ2 ≥
. . . ≥ λN ≥ 0 with λN ¿ λ1 . If Ω is the vector of components Ωj , we get
µ ¶
1 1 e
Ω = U diag ,... Uw .
λ1 λN
We see that the amplification of the data errors comes from the small eigenvalues of the
matrix A. The reason is that the number of independent data is much smaller than the
number of measured frequencies. In this case, the singular value decomposition technique
can bring some help to obtain a reasonable solution.
78
Least square fit method

e
We choose a basis of functions φj (r) (j = 1, . . . , M ) to write an approximation Ω(r) of Ω:
M
X
e =
Ω Ωj φj (r) .
j=1
Given the redundance of the data compared to the information they provide we choose
M < N and we determine the Ωj using the least square fit method. We generally impose
e
additional conditions on Ω(r), e.g., that it does not vary too fast. We must then usually
minimize an expression such as
N ·
X Z ¸2 Z Ã 2 e !2
e dΩ
S= wi − Ki (r)Ω(r) dr +µ dr ,
i=1 dr2
where µ is an arbitrarily chosen positive parameter. The last term softens the variations
e
of Ω(r), but other expressions can be used. We then determine the Ωj which minimize
the value of S by solving the linear equations
∂S
= 0, j = 1, . . . , M .
∂Ωj
Results
The superficial rotation of the Sun has been known for a long time: the equatorial regions
rotate faster than the polar regions. We cannot however get an agreement to better than
2% between the different observations. Thanks to helioseismology, it has been possible to
obtain the rotational angular velocity as a function of depth and latitude with a precision
of a few percents in the convective zone and the upper part of the radiative core (down
to 0.4 R¯ ). In the convective zone, the rotation seems to be very similar to what it is at
the surface, faster at the equator than in the polar regions. Under the convective zone,
that rotation appears to be uniform (as in a solid) with an angular velocity intermediate
between the equatorial and the polar values. We do not know much about the rotation
below 0.4 R¯ because of the uncertainty on the splitting data for low ` modes. It seems
however that the most internal layers rotate faster.
15.2 Determination of the solar structure

The adiabatic pulsation frequencies are determined by the distribution of mass and of
Γ1 in the model, i.e. by the functions ρ(r) and Γ1 (r). It is easy to see that the other
coefficients in the pulsation equations can be deduced from these two functions. We could
choose two other independent functions instead ρ(r) and Γ1 (r), such as for example ρ(r)
and c(r) (the sound speed). The asymptotic expression of the p modes clearly shows the
essential role played by the sound speed in the determination of the frequencies of these
modes. In what follows, we will show how the seismic data give information on the sound
speed. The process can be generalized to get simultaneously information on c(r) and ρ(r).
79
The pulsation frequencies depend on c(r) in a complicated non linear way. The process
used to determine Ω(r) can only be applied after linearization of the problem around a
reference model. We examine how the frequency of a given mode (for simplicity we will
omit the indices k, `) is changed by a small change in the sound speed δc(r). We will
calculate δσ neglecting the terms higher than first order in δc(r). From
σ 2 ξ = −Lξ ,
we easily get
δσ (ξ, δL ξ)
=− 2 ,
σ 2σ (ξ, ξ)
where δL is the correction to the operator L, linear in δc(r). It is easy to get
( )
−
→ 1 d h 2 i `(` + 1)
div ξ = 2
r a(r) − b(r) Y`m (θ, φ) ,
r dr r
Z ¯ →¯2 δc
−
(ξ, δL ξ) = −2 ρc2 ¯¯div ξ ¯¯ dV
( c )2
Z
2 1 d 2 δc
= −2 ρc (r a) − `(` + 1)b dr ,
Z r dr c
h i
(ξ, ξ) = ρr2 a2 + `(` + 1)b2 dr .
We therefore have
δσ Z δc
= K(r) dr ,
σ c
with ( )2
2 1 d 2
ρc (r a) − `(` + 1)b
r dr
K(r) = Z h i .
σ2 ρr2 a2 + `(` + 1)b2 dr
Finally we must solve the system

Z
δc σi obs − σi calc
Ki (r) dr = ≡ wi , i = 1, . . . , N .
c σi calc
We can use the linear inversion techniques mentionned earlier. If necessary, several cor-
rection cycles can be done.
The thickness of the convective zone and the sound speed inside the Sun have been
obtained by the analysis of helioseismic data.
For stars other than the Sun, it has been possible to use seismic data from rapidly oscil-
lating Ap stars, δ Sct variables and variable white dwarfs.
15.3 Non adiabatic asteroseismology

The techniques described above rely on the comparison between theoretical and observed
frequencies and are almost independent of non adiabatic effects. But multicolor photome-
try, through amplitude ratios and phase differences, is able to bring to the fore the strongly
80
non adiabatic behavior of the pulsation in the atmosphere of the star. The fitting of the
theoretical predictions to this type of observation produces constraints on the structure
of the outer layers of the star (convective zone, metallicity). The expression non adiabatic
asteroseismology has been coined to designate this type of investigation.
References
The singular value decomposition technique can also be applied to non square matri-
ces. Examples can be found in the papers by Korzennik and Ulrich (1989), Christensen-
Dalsgaard et al. (1990), Gough and Thompson (1991), Gu (1993) and Christensen-
Dalsgaard and Thompson (1993).
For more information on inversion methods, we recommend Gough (1985), Christen-
sen-Dalsgaard et al. (1990), Gough and Thompson (1991), Christensen-Dalsgaard and
Thompson (1993), Antia and Basu (1994). Sekii (1991)’s inversions determined Ω as
function of r and θ.
More information on solar rotation can be found in the paper by Libbrecht and Morrow
(1991). Information on asterosismology in stars other than the sun can be found in the
paper by Brown and Gilliland (1994).
Dupret et al. (2002) have used non adiabatic asteroseismology to obtain information on
the metallicity of a β Cep variable.
Antia H.M., Basu S., 1994. Nonasymptotic helioseismic inversion for solar structure.
Astron Astrophys Suppl Ser, 107, 421–444.
Brown T.M., Gilliland R.L., 1994. Asteroseismology. Ann Rev Astron Astrophys, 32,
37–82.
Christensen-Dalsgaard J., Schou J., Thompson M.J., 1990. A comparison of methods for
inverting helioseismic data. MNRAS, 242, 353–369.
Christensen-Dalsgaard J., Thompson M.J., 1993. A preprocessing strategy for helioseismic
inversions. Astron Astrophys, 272, L1–L4.
Dupret M.-A., De Ridder J., De Cat P., Aerts C., Scuflaire R., Noels A., Thoul A.,
2002. A photometric mode identification method, including an improved non-adiabatic
treatment of the atmosphere. Submitted to Astron Astrophys.
Gough D., 1985. Inverting helioseismic data. Solar Phys, 100, 65–99.
Gough D.O., Thompson M.J., 1991. The inversion problem. In Cox A.N., Livingston
W.C., Matthews M.S. (edit.), Solar interior and atmosphere, the University of Arizona
Press, Tucson, 519–561.
Gu Y., 1993. Nonlinear inversion for solar oscillation frequencies. Ap J, 413, 422–434.
Korzennik S.G., Ulrich R.K., 1989. Seismic analysis of the solar interior. I. Can opacity
changes improve the theoretical frequencies. Ap J, 339, 1144–1155.
Libbrecht K.G., Morrow C.A., 1991, The solar rotation. In Cox A.N., Livingston W.C.,
Matthews M.S. (edit.), Solar interior and atmosphere, the University of Arizona Press,
Tucson, 479–500.
Sekii T., 1991. Two-dimensional inversion for solar internal rotation. Publ Astron Soc
Japan, 43, 381–411.

Stellar Stability and Asteroseismology: R. Scuflaire, A. Thoul

Uploaded by

Copyright:

Available Formats

Stellar Stability and Asteroseismology: R. Scuflaire, A. Thoul

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stellar Stability and Asteroseismology: R. Scuflaire, A. Thoul

Uploaded by

Copyright:

Available Formats

Belgian Asteroseismology Group

Institute of Astrophysics and Geophysics

5 Small perturbations method 14

8 Adiabatic radial oscillations 29

9 Asymptotic expression for the radial frequencies 38

12 Non radial oscillations 55

13 Non radial modes 62

15 Helio- and asteroseismology 76

Figure 1.1: HR diagram with some types of variable stars.

2.1 The dynamical timescale

2.2 Pulsation timescale

0.03 days ≤ Q ≤ 0.08 days.

2.3 The Kelvin-Helmholtz timescale

On the other hand Ω is given by

Thus we can write

2.4 The nuclear timescale

3.1 Continuity equation

where X is the jacobian determinant

We easily recover the first form of the equation if we note that

3.2 Momentum equation

3.3 Poisson equation

3.4 Energy conservation equation

3.5 Transport equation

3.6 Boundary conditions

3.7 Material equations

P = P (ρ, T, χ), U = U (ρ, T, χ), κ = κ(ρ, T, χ), ² = ²(ρ, T, χ).

Small perturbations method

5.1 Lagrangian and eulerian perturbations

Let xi be some system of coordinates. The following relations are straightforward:

5.2 Perturbation of the differential equations

Radiative transfer equation:

F~ 0 = −λ0 grad T − λ grad T 0 .

5.3 Perturbation of the material equations

Eliminating δS we get the linearized equation of state

This equation describes an adiabatic perturbation. As a consequence, there is a simple

Finally the Poisson equation

has the following solution

The momentum equation

(L~u, ~v ) = (~u, L~v ).

We take the product with ξ

Lξi = λi ξi and Lξj = λj ξj ,

The condition δΛ = 0 is equivalent to

Since δξ is arbitrary, we have

7.1 Differential equations

and to apply it to the level corresponding to τ = 2/3 of the equilibrium model.

7.2 Integral equations

The s-independent term can then be written as

We have obtained the cubic equation

7.3 Dynamical modes and secular modes

We will therefore write A = A0 /τdyn

Adiabatic radial oscillations

It is useful to use dimensionless variables. We define

with the boundary conditions

Consider now two homologous stellar configurations. Let α = R0 /R and β = M 0 /M . We

r0 = αr, m0 = βm, ρ0 = α−3 βρ, P 0 = α−4 β 2 P, ...

ω0 = ω and σ 0 = β 1/2 α−3/2 σ .

Eliminating η, we get the differential equation

and the boundary conditions