Stability analysis of collisionless plasmas with specularly reflecting

boundary
arXiv:1112.4504v1 [math.AP] 19 Dec 2011

Toan Nguyen Walter A. Strauss

December 21, 2011

Abstract
In this paper we provide sharp criteria for linear stability or instability of equilibria of
collisionless plasmas in the presence of boundaries. Specifically, we consider the relativistic
Vlasov-Maxwell system with specular reflection at the boundary for the particles and with the
perfectly conducting boundary condition for the electromagnetic field. Here we initiate our
investigation in the simple geometry of radial and longitudinal symmetry.

1 Introduction
We consider a plasma at high temperature or of low density such that collisions can be ignored as
compared with the electromagnetic forces. Such a plasma is modeled by the relativistic Vlasov-
Maxwell system (RVM)
(
t f + + v x f + + (E + v B) v f + = 0,
(1.1)
t f + v x f (E + v B) v f = 0,

x E = , x B = 0, (1.2)
t E x B = j, t B + x E = 0, (1.3)
Z Z
= (f + f ) dv, j= v(f + f ) dv.
R3 R3

Here f (t, x, v) 0 is the density distribution for ions and electrons, respectively, x R3
is the p
particle position, is the region occupied by the plasma, v is the particle momentum,
hvi = 1 + |v|2 is the particle energy, v = v/hvi the particle velocity, the charge density, j the
current density, E the electric field, B the magnetic field and (E + v B) the electromagnetic
force. We assume that the particle molecules interact with each other only through their own

Division of Applied Mathematics, Brown University, 182 George Street, Providence, RI 02912, USA. Email:
Toan Nguyen@Brown.edu. Research of T.N. was partially supported under NSF grant no. DMS-1108821.

Department of Mathematics and Lefschetz Center for Dynamical Systems, Brown University, Providence, RI
02912, USA. Email: wstrauss@math.brown.edu.

1
electromagnetic forces. For simplicity, we have taken all physical constants such as the speed of
light and the mass of the electrons and ions equal to 1. This whole paper can be easily modified to
apply with the true physical constants.
Stability analysis for a Vlasov-Maxwell system of the type that we present in this paper has
so far appeared only in the absence of spatial boundaries, that is, either in all space or in a
periodic setting like the torus. In this paper we present the first systematic stability analysis in a
domain with a boundary. It is an unresolved problem to determine which boundary conditions
an actual plasma may satisfy under various physical conditions. Several boundary conditions are
mathematically valid and some of them are more physically justified than others. Stability analysis
is a central issue in the theory of plasmas. In a tokamak and other nuclear fusion reactors, for
instance, the plasma is confined by a strong magnetic field. This paper is a first, rather primitive,
step in the direction of mathematically understanding a confined plasma. We take the case of a
fixed boundary with specular and perfect conductor boundary conditions in a longitudinal and
radial setting.
The specular condition is
f (t, x, v) = f (t, x, v 2(v n(x))n(x)), n(x) v < 0, x , (1.4)
where n(x) denotes the outward normal vector of at x. The perfect conductor boundary
condition is
E n(x) = 0, B n(x) = 0, x . (1.5)
Under these conditions it is straightforward to see that the total energy
1 1
Z Z Z  
E(t) = v 2 (f + + f ) dvdx + |E|2 + |B|2 dx (1.6)
2 D R3 2 D
is conserved in time, and also that the system admits infinitely many equilibria. The main focus
of the present paper is to investigate stability properties of the equilibria.
Our analysis closely follows the spectral analysis approach in [15, 16, 17] which tackled the
stability problem in domains without spatial boundaries. Roughly speaking, that approach provided
the sharp stability criterion L0 0, where L0 is a certain nonlocal self-adjoint operator acting on
scalar functions that depend only on the spatial variables. The positivity condition was verified
explicitly for various interesting examples. It may also be amenable to numerical verification. In
our case with a boundary, every integration by parts brings in boundary terms. This leads to some
significant complications.
In the present paper, we restrict ourself to the stability problem in the simple setting of longi-
tudinal and radial symmetry. Thus the problem becomes spatially two-dimensional. Indeed, using
standard cylindrical coordinates (r, , z), the symmetry means that there is no dependence on z
and and that the domain is a cylinder = D R where D is a disk. We may as well assume that
D is the unit disk in the (x1 , x2 ) plane. It follows that x = (x1 , x2 , 0) D R, v = (v1 , v2 , 0) R3 ,
E = (E1 , E2 , 0), and B = (0, 0, B). In the sequel we will drop the zero coordinates so that x D and
v R2 . In terms of the polar coordinates (r, ), we denote er = (cos , sin ), e = ( sin , cos ).
It follows that the field has the form
1
E = r er t e , B = r (r)), (1.7)
r

2
where the scalar potentials (t, r) and (t, r) satisfy a reduced form of the Maxwell equations. See
the next section for details.

1.1 Equilibria
We will denote an equilibrium by (f 0, , E0 , B 0 ). Its field has the form
1
E0 = r 0 er , B0 = r (r 0 ). (1.8)
r
Then the particle energy and angular momentum

e (x, v) := hvi 0 (r), p (x, v) := r(v 0 (r)), (1.9)

are invariant along the particle trajectories. It is straightforward to check that (e , p ) solve
the Vlasov equations for any smooth functions (e, p). So we consider equilibria of the form

f 0,+ (x, v) = + (e+ (x, v), p+ (x, v)), f 0, (x, v) = (e (x, v), p (x, v)). (1.10)

The potentials still have to satisfy the Maxwell equations, which take the form
Z h i
0 = + (e+ , p+ ) (e , p ) dv
Z h i (1.11)
0 + + +
r = v (e , p ) (e , p ) dv

with r = r12 . Again, see the next section for details. It is clear that the boundary conditions
(1.4) and (1.5) are automatically satisfied for the equilibria since e and p are even in vr , and E0
is parallel to er . In the appendix we will prove that plenty of such equilibria do exist.
Let (f 0, , E0 , B0 ) be an equilibrium as just described with f 0, = (e , p ). We assume that
(e, p) are nonnegative, C 1 smooth, and satisfy

|
p (e, p)|
2
C

e (e, p) < 0, |
p (e, p)| + |e (e, p)| + (1.12)
|
e (e, p)| 1 + |e|

for some constant C and some > 2, where the subscripts e and p denote the partial derivatives.
In addition, we also assume that 0 , 0 are continuous in D. It follows that E0 , B0 C 1 (D), as
proven in the appendix.
We consider the Vlasov-Maxwell system linearized around the equilibrium. The linearization is

t f + D f = (E + v B) v f 0, , (1.13)

together with the Maxwell equations and the specular and perfect conductor boundary conditions.
Here D denotes the first-order linear differential operator: D := v x (E0 + v B0 ) v . See
the next section for details.

3
1.2 Spaces and operators
In order to state precise results, we have to define certain spaces and operators. We denote by
H = L2| | (D R2 ) the weighted L2 space consisting of functions f (x, v) which are radially
e
symmetric in x such that Z Z
| 2
e ||f | dvdx < +.
D
The main purpose of the weight function is to control the growth of f as |v| . Note that the
weight |e | never vanishes and it decays like a power of v as |v| . When there is no danger of
confusion, we will often write H = H .
For k 0 we denote by Hrk (D) the usual H k space on D that consists of functions that are
radially symmetric. If k = 0 we write L2r (D). By H k (D) we denote the space of functions = (r)
in Hrk (D) such that (r)ei belongs to the usual H k (D) space. The motivation for this space is to
get rid of the apparent singularity 1/r 2 at the origin in the operator r , thanks to the identity
 1
r = + 2 = ei (ei ).
r
By V we denote the space consisting of the functions in Hr2 (D) which satisfy the Neumann boundary
condition and which have zero average over D. Also, let V be the space consisting of the functions
in H 2 (D) which satisfy the Dirichlet boundary condition. The spaces V and V incorporate the
boundary conditions (2.9) for electric and magnetic potentials, respectively.
Denote by P the orthogonal projection on the kernel of D in the weighted space H . In the
spirit of [15, 17], our main results involve the three linear operators on L2 (D), two of which are
unbounded,
Z Z
A1 h = h e (1 P )h dv
0 + +
e (1 P )h dv,
Z Z  
A2 h = r h rv (p + p ) dvh v +
0 +
e P +
(v h) +
e P (v h) dv, (1.14)
Z   Z  
B0h = r +
p +
p dvh + + +
e P (v h) +
e P (v h) dv.

Here denotes (e , p ) = (hvi 0 (x), rv r 0 (x)). These three operators are naturally
derived from the Maxwell equations when f + and f are written in integral form by integrating
the Vlasov equations along the trajectories. See Section 3.2 for their properties. In the next section
we will show that both A01 with domain V and A02 with domain V are self-adjoint operators on
L2r (D). Furthermore, the inverse of A01 is well-defined on the range of B 0 , and so we are able to
introduce our key operator
L0 = A02 + (B 0 ) (A01 )1 B 0 . (1.15)
The operator L0 will then be self-adjoint on L2r (D) with its domain V . As the next theorem states,
L0 0 [which means that (L0 , )L2 0 for all V ] is the condition for stability.
Finally, by a growing mode we mean a solution of the linearized system (including the boundary
conditions) of the form (et f , et E, et B) with e > 0 such that f H and E, B L2 (D).

4
The derivatives and the boundary conditions are considered in the weak sense, which will be justified
in Lemma 2.2. In particular, the weak meaning of the specular condition on f will be given by
(2.10).

1.3 Main results

The first main result of our paper gives a necessary and sufficient condition for stability in the
spectral sense.

Theorem 1.1. Let (f 0, , E0 , B0 ) be an equilibrium of the Vlasov-Maxwell system satisfying (1.12).

Consider the linearization of the Vlasov-Maxwell system (1.13) for radially symmetric perturbations
together with the specular and perfect conductor boundary conditions. Then
(i) if L0 0, there exists no growing mode of the linearized system;
(ii) any growing mode, if it does exist, must be purely growing; that is, the unstable exponent
must be real;
(iii) if L0 6 0, there exists a growing mode.

Our second main result provides explicit examples for which the stability condition does or does
not hold. For more precise statements of this result, see Section 5.

Theorem 1.2. Let ( , E0 , B0 ) be an equilibrium as above.

(i) The condition p 0 0
p (e, p) 0 for all (e, p) implies L 0, provided that is bounded and
0
is sufficiently small. (So such an equilibrium is stable.)

(ii) The condition | 0
p (e, p)| 1+|e| for some > 2 and for sufficiently small implies L 0,
provided that 0 = 0. Here 0 is not necessarily small. (So such an equilibrium is stable.)
(iii) The conditions + (e, p) = (e, p) and p 2
p (e, p) c0 p (e), for some nontrivial non-
negative function (e), imply that for a suitably scaled version of ( , 0, B0 ), L0 0 is violated.
(So such an equilibrium is unstable.)

The instabilities in a plasma are due to the collective behavior of all the particles. For a ho-
mogeneous equilibrium (without an electromagnetic field) Penrose [19] found a beautiful necessary
and sufficient condition for stability of the Vlasov-Poisson system (VP). For a BGK mode, the equi-
librium has an electric field. In that case, proofs of instability for VP (electric perturbations) were
first given in [7, 8, 9] and especially in [13, 14] for non-perturbative electric fields. Once magnetic
effects are included, even for a homogeneous plasma, the situation becomes much more complicated:
see for instance [6, 10, 11]. In a series of three papers [15, 16, 17] a more general approach was
taken that treated fully electromagnetic equilibria with electromagnetic perturbations. The linear
stability theory was addressed in [15, 17], while in [16] fully nonlinear stability and instability was
proven in some cases. In all of the work just mentioned, boundary behavior was not addressed; the
spatial domains were either all space or periodic.
In this paper we do not address the question of well-posedness of the initial-value problem.
For VP in R3 well-posedness was proven in [20] and [18]. For RVM in all space R3 it is a famous
open problem. The relativistic setting seems to be required, for otherwise the Vlasov and Maxwell

5
characteristics would collide. However, for RVM in the whole plane R2 , which is the case most
relevant to this paper, well-posedness and regularity were proven in [4]. The same is true even in
the 2.5 dimensional case [3]. Although global weak solutions exist in R3 , they are not known to be
unique [2]. Furthermore, in a spatial domain with a boundary on which one assumes specular and
perfect conductor conditions, global weak solutions also exist [5]. Well-posedness and regularity of
VP in a convex bounded domain with the specular condition was recently proven in [12].
A delicate part of our analysis is how to deal with the specular boundary condition within the
context of weak solutions. This is discussed in subsection 2.3. Properly formulated, the operators
D then are skew-adjoint. In this paper, as distinguished from [15], we entirely deal with the weak
formulation. Our regularity assumptions are essentially optimal. In subsection 2.4 we prove that
the densities f of a growing mode of the linearized system decay at a certain rate as |v| .
As in [15], the stability part of Theorem 1.1 is based on realizing the temporal invariants of the
linearized system. They have to be delicately calculated due to the weak form of the boundary
conditions. This is done in subsection 3.3. The invariants are the generalized energy I and the
casimirs Kg , which are a consequence of the assumed symmetry of the system. A key part of the
stability proof is to minimize the energy with the magnetic potential being held fixed (see subsection
3.4). The purity of any growing mode, in part (ii) of Theorem 1.1, is a consequence of splitting
the densities into even and odd parts relative to the variable vr (see subsection 3.5). The proof of
stability is completed in subsection 3.6.
The proof of instability in Theorem 1.1(iii) requires the introduction of a family of linear
operators L which formally reduce to L0 as 0. The technique was first introduced by
Lin in [13] for the BGK modes. These operators explicitly use the particle paths (trajectories)
(X (s; x, v), V (s; x, v)). The trajectories reflect specularly a countable number of times at the
boundary. We use them to represent the densities f in integral form, like a Duhamel representa-
tion. This representation together with the Maxwell equations leads to the family of operators in
subsection 4.3. Self-adjointness requires careful consideration of the trajectories. It is then shown
in subsection 4.4 that L is a positive operator for large , while it has a negative eigenvalue for
small because of the hypothesis L0 6 0. Therefore L has a nontrivial kernel for some > 0
If is in the kernel, it is the magnetic potential of the growing mode. From we construct the
corresponding electric potential and densities f .
In Section 5 we prove Theorem 1.2. The stability examples are relatively easy. As we see, the
basic stabilizing condition is p p 0. To construct the unstable examples we make the simplifying
assumption that the equilibrium has no electric field so that L0 = A02 . In the expression (L0 , )L2
the term that has to dominate negatively is the one with pp . In order to make it dominate, we
scale the equilibrium appropriately. We first treat the homogeneous case (Theorem 5.3) and then
the purely magnetic case (Theorem 5.4).

6
2 The symmetric system
2.1 The potentials
It is convenient when dealing with the Maxwell equations to introduce the electric scalar potential
and magnetic vector potential A through

E = t A, B = A, (2.1)

in which without loss of generality, we impose the Coulomb gauge constraint A = 0. Note that
with these forms, there automatically hold the two Maxwell equations:

t B + E = 0, B = 0,

whereas the remaining two Maxwell equations become

= , t2 A A + t = j.

Under the assumption of radial and longitudinal symmetry, there is no z or dependence. We

use the polar coordinates x = (r cos , r sin ) on the unit disk D. A radial function f (x) is one
that depends only on r and in this case we often abuse notation by writing f (r). We also denote
the unit vectors by er = (cos , sin ) and e = ( sin , cos ). Thus er (x) = n(x) is the outward
normal vector at x D. Although the functions do not depend on , the unit vectors er and e
do. Then we may write f = f (t, r, v), where v = vr er + v e and A = Ar er + A e .
Now the Coulomb gauge in this symmetric setting reduces to 1r r (rAr ) = 0, so that Ar = h(t)/r.
We require the field to have finite energy, meaning that E, B L2 (D). Thus Er = r t Ar
L2 (D) only if h(t) is a constant. But if h(t) is a constant, we may as well choose it to be zero
because it will not contribute to either E or B. For notational convenience let us write in place
of A . The fields defined through (2.1) then take the form
1
E = r er t e , B = Bez , B= r (r)). (2.2)
r
We note that
1 1 n o
v x f = vr r f + v f = vr r f + v v vr f vr v f .
r r
In these coordinates the RVM system takes the form
   
t f + + vr r f + + Er + v B + 1r v v vr f + + E vr B 1r vr v v f + = 0,
    (2.3)
t f + vr r f Er + v B 1r v v vr f E vr B + 1r vr v v f = 0,
Z
= = (f + f )(t, r, v) dv,
R2
Z
t r = jr = vr (f + f )(t, r, v) dv, (2.4)
R2
Z
(t2 r ) = j = v (f + f )(t, r, v) dv,
R2

7
where r = r12 . The system (2.3) - (2.4) is accompanied by the specular boundary condition
on f , which is now equivalent to the evenness of f in vr at r = 1. In particular, jr = 0 on D.
The condition B n = 0 is automatic. The boundary conditions on and are
r (t, 1) = const., (t, 1) = const. (2.5)
The Neumann condition on comes naturally from the second Maxwell equation in (2.4) with
jr = 0. The Dirichlet condition on comes naturally from 0 = E n = (0, 0, t ).

2.2 Linearization
We linearize the Vlasov-Maxwell near the equilibrium (f 0, , E0 , B 0 ). From (2.3), the linearized
Vlasov equations are
t f + + D+ f + = (Er + v B)vr f 0,+ (E vr B)v f 0,+ ,
(2.6)
t f + D f = (Er + v B)vr f 0, + (E vr B)v f 0, .
The first-order differential operators D := v x (E0 + v B0 ) v now take the form
 1   1 
D+ := vr r + Er0 + v B 0 + v v vr vr B 0 + vr v v ,
r r (2.7)


0 0 1  
0 1 
D := vr r Er + v B v v vr + vr B vr v v .
r r
In order to compute the right-hand sides of (2.6) more explicitly, we differentiate the definition
f 0, = (e , p ) to get
vr f 0, =
e vr , v f 0, =
e v + rp .

Thus, together with the forms of E and B in (2.2), we calculate

(Er + v B)vr f 0,+ + (E vr B)v f 0,+
= (Er + v B)+ + +
e vr + (E vr B)(e v + rp )
= + + + +
e vr r p vr r (r) e v t rp t
= + + + + + +
e D p D (r) (e v + rp )t ,

where the last line is due to the fact that D+ = vr r for radial functions. Of course a similar
calculation holds for f 0, . Thus the linearization (2.6) becomes
t f + + D+ f + = + + + + +
e D + p D (r) + v [ ]t
= + + + +
e vr r + p vr r (r) + (e v + rp )t
(2.8)
t f + D f =
e D p D (r) v [ ]t
=
e vr r p vr r (r) (e v + rp )t

Of course, linearization does not alter the Maxwell equations (2.4). As for boundary conditions,
we naturally take the specular condition on f and
r (t, 1) = 0, (t, 1) = 0. (2.9)

8
2.3 The Vlasov operators
The Vlasov operators D are formally given by (2.7). Their relationship to the boundary condition
is given in the next lemma.

Lemma 2.1. Let g(x, v) = g(r, vr , v ) be a C 1 radial function on D R2 . Then g satisfies the
specular boundary condition if and only if
Z Z Z Z
g D h dvdx = D g h dvdx
D R2 D R2

(either + or -) for all radial C 1 functions h with v-compact support that satisfy the specular condi-
tion.

Proof. Integrating by parts in x and v, we have

Z Z Z 

{g D h + D g h} dvdx = 2 gh v er dv  .

D R2 r=1

If g satisfies the specular condition, thenR g and h are even functions ofR vr = v er on D, so that
the last integral vanishes. Conversely, if gh v er dv = 0 on D, then g(1, vr , v ) k(vr , v )dv = 0
for all test functions that are odd in vr , so it follows that g(1, vr , v ) is an even function of vr .

Therefore we define the domain of D to be

n  o

dom(D ) = g H  D g H, hD g, hiH = hg, D hiH , h C , (2.10)


where C denotes the set of radial C 1 functions h with v-compact support that satisfy the specular
condition. We say that a function g H with D g H satisfies the specular boundary condition
in the weak sense if g dom(D ). Clearly, dom(D ) is dense in H since by Lemma 2.1 it contains
the space C of test functions, which is of course dense in H.
It follows that
hD g, hiH = hg, D hiH (2.11)
for all g, h dom(D ). Indeed, given h dom(D ), we just approximate it in H by a sequence of
test functions in C, and so (2.11) holds thanks to Lemma 2.1.
Furthermore, with these domains, D are skew-adjoint operators on H. Indeed, the skew-
symmetry has just been stated. To prove the skew-adjointness of D+ , suppose that f, g H and
hf, hiH = hg, D+ hiH for all h dom(D ). Taking h C to be a test function, we see that
D+ g = f in the sense of distributions. Therefore (2.11) is valid for all such h, which means that
g dom(D + ).

2.4 Growing modes

Now we can state some necessary properties of any growing mode. Recall that by definition a
growing mode satisfies f H and E, B L2 (D).

9
Lemma 2.2. Let (et f , et E, et B) with e > 0 be any growing mode. Then E, B H 1 (D) and
dvdx
ZZ
(|f |2 + |D f |2 ) < .
DR2 |
e|

Proof. The fields are given by (2.2) where , satisfy the elliptic system (2.4) with Dirichlet or
Neumann boundary conditions, expressed weakly. The densities f satisfy (2.8). Explicitly,
f + D f =
e vr r p vr r (r) (e v + rp ). (2.12)
This equation implies that D f H. The specular boundary condition on f is expressed weakly
by saying that f dom(D ). Dividing by |
e | and defining g = f /|e |, we write the equation
in the form g + D g = h , where the right side h belongs to H = L2| | (D R2 ) thanks to

e
the decay assumption (1.12) on .
Letting w = |
e |/( + |e |) for > 0 and g = w g , we have hg + D g , g iH = hw h , g iH .

It easily follows that g H. In fact, g dom(D ), which means that the specular boundary
condition holds in the weak sense, so that (2.11) is valid for it. In (2.11) we take both functions to
be g and therefore hD g , g iH = 0. It follows that
||kg k2H = |hw h , g iH | kh kH kg kH .
Letting 0, we infer that g H, which means that
RR 2
|f | /|e |dvdx < .
Now the elliptic system for the field is
Z Z
= (f + f )dv, (2 r ) = v (f + f )dv

together with RR the boundary conditions r (t, 1) = 0, (t, 1) = 0, which are expressed weakly.
Because of |f |2 /|
e |dvdx < , the right sides of this system are now known to be finite
a.e. and to belong to L (D). So it follows that , H 2 (D) and E, B H 1 (D). This is the
2

first assertion of the lemma. Nevertheless, we emphasize that D f doesRR not satisfy the specular
boundary condition. However, directly from (2.12) it is now clear that |D f |2 /|
e |dvdx < .
This is the last assertion.

3 Linear stability
3.1 Formal argument
Before presenting the stability proof, let us sketch a formal proof. We consider the linearized RVM
system (2.8). For sake of presentation, let us consider the case with one particle f = f + , and thus
drop all the superscripts +. We have the linearized equation:
t f + Df = e D + p D(r) + (e v + rp )t .
The key ingredient for stability is the fact that the functional
1
Z Z h i Z h i
2 2
I(f, , ) := |f rp | rp v || dvdx + |E|2 + |B|2 dx
D |e | D

10
is time-invariant, which can be found by formally expanding the usual nonlinear energy-Casimir
functional around the equilibria. Next, we then write the linearized equation in the form

t (f e v rp ) + D(f e rp ) = 0. (3.1)

We observe that (f e v rp ) stays orthogonal to ker D for all time, and that in case of
stability we would expect that f e rp asymptotically belongs to ker D. That is, if P is the
projection onto the kernel, one would have at large times

f e rp = P(f e rp ), P(f e v rp ) = 0.

Adding up these identities, we obtain

f = e (1 P) + rp + e P(v ).

This asymptotic description of f is essential to the proof of stability, which we will prove rigorously
in subsection 3.4; see (3.14). Next, by plugging the identity into the functional I(f, , ), we will
obtain Z
0 0
I(f, , ) = (A1 , )L2 + (A2 , )L2 + |t |2 dx,
D

in which the operators A0j

are defined as in (1.14). Using the identity A01 = B 0 , which is obtained
from the Poisson equation, to eliminate , we can formally write
Z
I(f, , ) = (L0 , )L2 + |t |2 dx, (3.2)
D

with L0 := A02 + (B 0 ) (A01 )1 B 0 . (In fact, we are only able to prove a reverse inequality (), which
however suffices for stability). Now, if we assumed that there were a growing mode of the linearized
system, then the time-invariant functional I(f, , ) must be zero. Thus (3.2) shows that L0 6 0.
That is, L0 0 formally implies the stability.

3.2 Key operators

In this subsection, we shall derive the basic properties of the operators A0j and B 0 defined in (1.14).
Let us recall that P are the orthogonal projections of H onto the kernels
n  o
ker(D ) = f dom(D )  D f = 0 ,


and the key operators:

Z Z
A01 = +
e (1
+

P ) dv
e (1 P ) dv
Z Z  
A2 = r rv (p + p ) dv v +
0 + +
e P (v ) + e P (v ) dv (3.3)
Z   Z  
0 +
B =r p + p dv + + +
e P (v ) + e P (v ) dv.

11
We also recall that the functional space V consists of functions in Hr2 (D) that satisfy the Neumann
boundary condition and have zero average over D, whereas V consists of functions in H 2 (D) that
satisfy the Dirichlet condition on D.

Lemma 3.1.
(i) A01 is self-adjoint and positive definite on L2r (D) with domain V. A01 is a one-to-one map
from V onto the set {g L2r | D g dx = 0}. In particular, (A01 )1 is well-defined on the range of
R

B0.
(ii) B 0 is a bounded operator on L2r (D).
(ii) A02 and L0 are self-adjoint on L2r (D) with common domain V .

Proof. First, since the projections P preserve the radial symmetry, the operators A0j and B 0 also
preserve the symmetry. Next we observe that all the integral terms in (3.3) are bounded operators
in L2r (D). For example, we have
Z Z  Z 
+ +
e P dvdx sup |+ +
e | dv kkL2 kP kL2 C0 kkL2 kkL2 ,
 

D D

for some constant C0 that depends on the decay assumption (1.12) on + e . The other integrals
are similar since v is bounded by one. This proves (ii) and also proves that the integral terms
in A01 and A02 are relatively compact with respect to and r , which have domains V and
V , respectively. Thus, A01 and A02 are well-defined operators on L2r (D) with domains V and V ,
respectively.

Since P are self-adjoint on H, it is clear that all three operators A01 , A02 , and L0 are self-adjoint
on L2r (D). Note that the function 1 belongs to the kernel of D . To prove the positivity of A01 , we
use the orthogonality of P and 1 P in H to write
Z Z Z Z

e (1 P
) dvdx = 2
e |(1 P )| dvdx 0
D D

since 0 0
e < 0. Thus A1 is a nonnegative operator, and A1 = 0 if and only if is a constant.
0
Since A1 Rhas discrete spectrum, it is invertible on the orthogonal set to its kernel, that is, on
{g V | D g dx = 0}. In order to prove the invertibility of A01 on the range of B 0 , we note that
by the self-adjointness of P in H, we have
Z Z Z Z Z Z

e P
(v ) dvdx =
v
e P
(1) dvdx =
e v dvdx.
D D D

Thus Z XZ Z
0
B dx = (r
p + e v ) dvdx,
D D

which is identically zero by using the fact that v [ ] = v 0

e + rp . That is, B has zero
0 1
average, and so (A1 ) is well-defined on the range of B .0

12
3.3 Invariants
First we consider the linearized energy.

Lemma 3.2. Suppose that (f , , ) is a solution of our linearized system (2.8), (2.9), and (1.4)
such that f C 1 (R, L21/|e | (D R2 )) and , C 1 (R; H 2 (D)). Then the linearized energy
functional

I(f , , )
XZ Z h 1 i Z h
1 i
2 2 2 2 2
:= |f r p | r p v || dvdx + |t | + || + |r (r)| dx
D |
e| D r2

is independent of time.

Proof. For convenience of calculation, we denote the three terms as

1
Z Z h i
2 2
IV (f , ) := |f r p | r p v || dvdx
D |
e|
1
Z h i
IM (, ) := ||2 + |t |2 + 2 |r (r)|2 dx,
D r

so that
I(f , , ) = IV+ (f + , ) + IV (f , ) + IM (, ).
Now taking the time derivative of IV+ (f + , ) and then using the linearized Vlasov equation (2.8)
for f + , we get
1 d + +
I (f , )
2 dtZ VZ  
1 + + + + +
= (f rp )(t f rp t ) rp v t dvdx
D |+
e|
Z Z  
1 + +
n
+ + + + + + +
o
+
= (f r p ) D f + e D + p D (r) + v
e t r v
p t dvdx
D |+
e|
1 h
Z Z
= f + D+ f + + + + + + + + + + +
e f D + p f D (r) + e f v t + rp D f
+ +
D |+e|
i
r+ + + + 2 +
p e D (p ) rD (r) dvdx.

Among the nine terms, we have used the fact that two terms with t exactly cancel because
e < 0. Some of the remaining seven terms cancel, as follows. First, we observe that the sixth
term vanishes, upon writing D+ = vr r and noting that is even in vr . Next, by using the
skewsymmetric property (2.11) of D+ , which is the weak form of the specular boundary condition,
the first and seventh terms are also zero. Now the third and fifth terms can be combined to get
+ + +
p D (rf ), whose integral again vanishes due to the boundary condition = 0. We have used

13
the fact that both + + +
e and p belong to the kernel of D . Only the second and fourth terms survive,
so we have
1d + +
Z Z h i Z Z
+ + +
I (f , ) = f D f v t dvdx = E v f + dvdx,
2 dt V D D
in which we have used the fact that D+ = vr r together with the definition E = r er t e .
Entirely similar calculations hold for f . We therefore obtain
1 d + +  Z Z Z
+
I (f , ) + IV (f , ) = E v(f f ) dvdx = E j dx. (3.4)
2 dt V D D
R h i
Next, by (2.1) we may write IM (, ) = D |E|2 + |B|2 dx. Taking its time derivative and
using the Maxwell equations, we obtain
1 d
Z h i
IM (, ) = E t E + B t B dx
2 dt
ZD h i
= E j + E ( B) B ( E) dx.
DZ Z Z
= E j dx + (E B) n(x) dSx = E j dx
D D D
Here we have used the fact that (EB)n = (En)B, which vanishes due to the perfect conductor
boundary condition E n = 0. Together with (3.4), this yields invariance of I(f , , ).

Furthermore, we also obtain the following.

Lemma 3.3. Suppose that (f , , ) is a solution of our linearized system (2.8), (2.9), and (1.4)
such that f C 1 (R, L21/|e | (D R2 )) and , C 1 (R; H 2 (D)). Then the functional
Z Z  
Kg (f , ) = f
e v r
p g dvdx (3.5)
D
are independent
R Rin time, for all g ker D and for both + and . In particular, for g = 1 in (3.5),

the integrals D R2 f (t, x, v) dvdx are time-invariant; that is, the total masses are conserved.
Proof. Parenthetically, we remark that such invariants Kg (f , ) can easily be discovered by writ-
ing the Vlasov equations as in (3.1). Indeed, writing the Vlasov equations in the form (3.1) and
using the skew-symmetry property of D as in (2.11), we have
d
Z Z  
Kg (f , ) = t f v
e r
p g dvdx
dt D
Z Z  

= D f e rp g dvdx
DZ Z
 
= f
e rp D g dvdx = 0
D
due to the specular conditions on f
and g and the evennessRinRvr of . Now, if we take g = 1
in (3.5) and note that v = e v + r

p , the integrals D R2 f (t, x, v) dvdx are therefore
time-invariant.

14
3.4 Minimization
In this subsection we prove an identity that will be fundamental to the proof of stability. It involves
the functional
1 1
Z Z Z Z Z
+ + + 2 2
J (f , f ) := + |f rp | dvdx + |f + rp | dvdx + ||2 dx,
D | e | D | e | D

where = (r) V satisfies the Poisson equation

Z Z
+
= (f f ) dv, dx = 0, r (1) = 0. (3.6)
D

For each fixed L2r (D), let F be the space consisting of all pairs of measurable functions
(f + , f ) depending on (r, v) which satisfy the constraints
1
Z Z
2
|f | dvdx < +, (3.7)
D |e |
and
Z Z   Z Z  
+
f +
e v r+
p
+
g dvdx = 0, f +
e v + rp g dvdx = 0, (3.8)
D D

for all g ker D . Similarly, let F0 be the space of pairs (f + , f ) satisfying (3.7) and
Z Z Z Z
+ +
f g dvdx = 0, f g dvdx = 0, g ker D . (3.9)
D D

Note that the constraints in (3.8) with g = 1 imply that for such a pair of functions f , there is
a unique solution V of the Poisson problem (3.6). In particular, is radially symmetric since
f are radially symmetric. Thus the functional J is well-defined and nonnegative on F , and its
infimum over F is finite. We next show that it indeed admits a minimizer on F .
Lemma 3.4. For each fixed L2r (D), there exists a pair of functions f that minimizes the
functional J on F . Furthermore,
Z Z Z Z
J (f+ , f ) = ((B 0 ) (A01 )1 B 0 , )L2 +
e |P +
(v )|2
dvdx 2
e |P (v )| dvdx.
D D
(3.10)
Proof. Take a minimizing sequence fn in F such that J (fn+ , fn ) converges to the infimum of J .
Since {fn } are bounded sequences in L21/| | , the weighted L2 space associated with the constraint
e
(3.7), there are subsequences with weak limits in L21/| | , which we denote by f . It is clear that
e
f satisfy the constraints (3.8), and so they belong to F . That is, (f+ , f ) must be a minimizer.
In order to derive the identity (3.10), let the pair (f , f ) F be the minimizer and let
Hr2 (D) be the associated solution of problem (3.6) with f = f . For each (f + , f ) F ,
we denote
h+ := f + + +
e v rp , h := f +
e v + rp . (3.11)

15
In particular, h + +
:= f e v rp . It is clear that (f , f ) F if and only if (h , h ) F0 .
Since v [ ] =
e v + rp , we have
Z Z
(f f ) dv = (h+ h ) dv.
+

Thus if we let be the solution of the Poisson equation (3.6), is independent of the change of
variables in (3.11). Consequently, (f+ , f ) is a minimizer of J on F if and only if (h+
, h ) is a
minimizer of the functional
1 1
Z Z Z Z Z
J0 (h+ , h ) = + |h+
+ +
v
e |2
dvdx + |h

v
e |2
dvdx + ||2 dx,
D |e | D |e | D

on F0 . By minimization, the first variation is

1 + 1
Z Z Z Z Z
+ +
(h + v
e )h dvdx + (h v
e )h dvdx + dx = 0, (3.12)
D +
e D e D

for all (h+ , h ) F0 where solves (3.6). By the Neumann boundary condition on , we have
Z Z Z Z
dx = dx = (h+ h ) dvdx.
D D D

Adding this to the identity (3.12), we obtain

1 + 1
Z Z Z Z
+ + +
+ (h + v
e
e ) h dvdx + (h e v + e ) h dvdx = 0 (3.13)
D e D e

for all (h+ , h ) F0 . In particular, we can take h = 0 in (3.13) to get

1 +
Z Z
+ + +
+ (h + e v e ) h dvdx = 0
D e

for all h+ L21/|+ | satisfying D h+ g+ dvdx = 0, for all g + ker D+ .

R R
e
We claim that this identity implies h+ + + +
+ e v e ker D . Indeed, let

k = |+ 1 + + +
e | (h + e v e ), = |+ 1 +
e| h .

Using the inner product in H = L2|+ | , we have

e

hk , iH = 0 (ker D+ ) .

Because D+ (with the specular condition) is a skew-adjoint operator on H, we have k (ker D+ ) =

ker D+ . Thus

D+ {f+ r+ + + + + + + +
p e } = D {h + e v e } = e D k = 0.

16
This proves the claim. Similarly D {f + r
p + e } = 0. Equivalently,

f+ r+ + + + + +
p e = P (f rp e ),
f + r
p + e = P (f + rp + e ).

On the other hand, the constraints (3.8) can be written as P + (f+ + +

e v rp ) = 0 and

P (f + e v + rp ) = 0. Combining these identities, we have

f+ r+ + + + +
p = e (1 P ) + e P (v ),
(3.14)
f + r
p = e (1 P ) e P (v ).

Thus, using the orthogonality of P and (1 P ) in H, we compute

1 +
Z Z Z Z Z Z
2 2
|f
r p | dvdx = e |(1 P ) | dvdx 2
e |P (v )| dvdx
D e
ZD Z ZD Z
=
e (1 P ) dvdx 2
e |P (v )| dvdx.
D D

Inserting these identities into the definition of J (f+ , f ), together with the fact from the boundary
conditions that D | |2 dx = D dx, we obtain
R R

Z h Z Z i
J (f+ , f ) = + e (1 P +
) dv
e (1 P
) dv dx
D Z Z Z Z
+ + 2
e |P (v )| dvdx 2
e |P (v )| dvdx.
D D

By the definition (3.3) of A01 , the first group of integrals simply equals (A01 , )L2 .
Thus it remains to prove that A01 = B 0 , because A01 is invertible on the range of B 0 so
that = (A01 )1 B 0 . Indeed, we plug the identities (3.14) into the Poisson equation (3.6) for ,
resulting in the equation
Z Z
= e (1 P ) dv +
+ +
e (1 P ) dv
Z Z   (3.15)
+ + +
+ r (p + p ) dv + e P (v ) + e P (v ) dv.

In view of the definitions of A01 and B 0 , this identity (3.15) is equivalent to A01 = B 0 , as
desired.

3.5 Growing modes are pure

In this subsection, we show that if (et f , et E, et B) with e > 0 is a complex growing mode,
then must be real. See subsection 2.4 for the properties of a growing mode. We now follow the
splitting method in [15] to show that is real. Let fev and fod be the even and odd parts of f

17
with respect to the variable vr . That is, we have the splitting: f = fev + fod , and furthermore, by
inspection from the definition (2.7), the operators D map even to odd functions and vice versa.
We therefore obtain, from the Vlasov equations (2.8), the split equations
+
( +
fev + D+ fod = (+ + +
e v + rp ) = v [ ],
+
fod + D+ fev
+
= + + + +
e D + p D (r).
RR + 2 + , so
By Lemma 2.2, we know that |f | /|eRR
|dvdx < . It follows that the same is true for fev
+ 2
from the first split equation we also have |D + fod | /|e |dvdx < . The split equations imply
that
2 +
(2 D+ )fod = + + + +
e D e D (v ). (3.16)
+
Let f be the complex conjugate of f + . By the specular boundary condition on f + in its weak form
+ +
(2.11), it follows that fod satisfies the specular condition. (Formally, fod vanishes on the boundary
+ + + +
D.) However, since D fod is even in the variable vr , D fod also satisfies the specular condition.
+
Thus when we multiply equation (3.16) by f od /|+ 2
e | and integrate the result over D R , we may
apply the skew-symmetry property (2.11) of D . We obtain +

1 1
Z Z Z Z Z Z 
+ + +

2 + 2 + + 2 +
|f od | dvdx + |D f od | dvdx = D f od + v D f od dvdx.
D |+e| D |+
e | D

Similarly for f we obtain

1 1
Z Z Z Z Z Z 


2 2 2
|f od | dvdx + |D f od | dvdx = D f od + v D f od dvdx.
D |
e| D |
e| D

Adding up these identities and examining the imaginary part of the resulting identity, we get
1 1
Z Z  
+ 2 2
2em |f | + |f | dvdx
|+ | od |
e|
od
D
Z Z e Z Z (3.17)
+ + + +
= m (D f od D f od ) dvdx m v (D f od D f od ) dvdx.
D D

Let us now use the Maxwell equations to compute the terms on the right side of (3.17). First,
we recall that the second equation in (2.4) is
Z Z
r = vr (f + f ) dv = vr (fod
+
fod ) dv.

Together with the definition of D , this yields

Z Z Z Z Z
+ +

+ 2
(D f od D f od ) dvdx = r vr (f od f od ) dv dx = || |r |2 dx,
D D D

whose imaginary part is identically zero.

18
 Secondly, using the Vlasov equation for fev + , we estimate

Z Z
+
v (D+ f od D f od ) dvdx
D Z Z
+
= v (f ev + v [+ ] + f ev + v [ ]) dvdx
ZD  Z Z Z
+

2 2
= || v (f ev f ev dv dx || ||2 v (v [+ ] + v [ ]) dvdx.
D D

By (2.4) the first term on the right equals

1
Z Z Z
2 2 2 2
|| (r + )dx = || |r (r)|2 dx + ||2 ||2 dx.
D D r2 D

Here we integrated by parts and used the Dirichlet boundary condition on .

Putting these estimates back into (3.17), we obtain
1 1
Z Z   Z
+ 2 2 2
2em |fod | + |fod | dvdx = 2em|| ||2 dx.
D |+
e| |e | D

The opposite signs of the integrals imply that must be real.

3.6 Proof of stability

With the above preparations, we are ready to prove the following stability result, which is half of
Theorem 1.1.
Lemma 3.5. If L0 0, then there exists no growing mode (et f , et E, et B), with e > 0.
Proof. Assume that there were a growing mode (et f , et E, et B). For the basic properties of
any growing mode, see Lemma 2.2. By the result in the previous subsection, it is a purely growing
mode, and thus we can assume that (f , E, B) are real-valued functions. By the time-invariance in
Lemma 3.2, the functional I(f , , ) must be identically equal to zero, where and are defined
as usual through the relations (2.1). That is,
1
Z Z Z h i
I(f , , ) = J (f + , f ) rv (+
p +
p )||2
dvdx + ||2
||2
+ |r (r)|2
dx = 0.
D D r2
Furthermore, all the expressions Kg (f , ) defined in (3.5) must be zero. The vanishing of the
latter integrals is equivalent to the constraints in (3.8) and therefore the pair (f + , f ) belongs to
the function space F . We then apply the Lemma 3.4 to assert that J (f + , f ) J (f+ , f ).
Thus we have
Z Z Z Z
0 0 1 0 + + 2
I(f , , ) ((B ) (A1 ) B , )L2 e |P (v )| dvdx 2
e |P (v )| dvdx
D D
1
Z Z Z h i
+ 2
rv (p + p )|| dvdx + ||2 ||2 + 2 |r (r)|2 dx.
D D r
(3.18)

19
In addition, from (3.3) an integration by parts together with the Dirichlet boundary condition on
yields
1
Z   Z Z
(A02 , )L2 = |r |2 + 2 ||2 dx rv (+ 2
p + p )|| dvdx
D Z Z r D Z Z
+ + 2
e |P (v )| dvdx 2
e |P (v )| dvdx.
D D

Putting this calculation into (3.18) and using the definition of L0 , we then get
Z
0 = I(f , , ) (L0 , )L2 + ||2 ||2 dx.
D

This is obviously a contradiction since we are assuming L0 0. Thus there exists no growing mode
for the linearized system.

4 Linear instability
We now turn to the instability part of Theorem 1.1. It is based on a spectral analysis of the
relevant operators. We plug the simple form (et f , et E, et B), with > 0, into the linearized
RVM system (2.8) to obtain the Vlasov equations

( + D+ )f + = + + + + + +
(
e D + p D (r) + (e v + rp )
(4.1)
( + D )f =
e D p D (r) (e v + rp )

and the Maxwell equations

Z


= (f + f ) dv,
Z (4.2)
(r + 2 ) = +
v (f f ) dv.

As before, we impose the specular boundary condition on f , the Neumann boundary condition
on , and the Dirichlet boundary condition on . Recall that (f , , ) is a perturbation of the
equilibrium.

4.1 Particle trajectories

We begin with the + case (ions). For each (x, v) D R2 , we introduce the particle trajectories
(X + (s; x, v), V + (s; x, v)) defined by

X + = V + , V + = E0 (X + ) + V + B0 (X + ), (4.3)

with initial values (X + (0; x, v), V + (0; x, v)) = (x, v). Because of the C 1 regularity of E0 and B0 in
D, each trajectory can be continued for at least a certain fixed time. Thus each particle trajec-
tory exists and preserves cylindrical symmetry up to the first point where it meets the boundary.

20
The trajectories reflect at the boundary D at most a countable number of times as s .
Let s0 be any point at which the trajectory X + (s0 ; x, v) belongs to D. In general, we write
h(s) to mean the limit from the right (left). By the specular boundary condition, the trajectory
(X + (s; x, v), V + (s; x, v)) can be continued by the rule
(X + (s0 +; x, v), V + (s0 +; x, v)) = (X + (s0 ; x, v), V + (s0 ; x, v)), (4.4)
with V = (Vr , V ). Thus X + is continuous and V + has a jump at s0 . Whenever the trajectory
meets the boundary, it is reflected in the same way and then continued via the ODE (4.3). Such
a continuation is guaranteed for some short time s1 past s0 by the standard ODE theory. Since
E0 and B0 are C 1 smooth in D, the additional time |s1 s0 | is bounded below by some fixed
positive time T0 independent of s0 . This shows that the trajectories can bounce at the boundary
at most a countable number of times as |s| . When there is no possible confusion, we will
simply write (X + (s), V + (s)) or (R+ (s), Vr+ (s), V+ (s)) for the particle trajectories. The trajectories
(X (s), V (s)) for the case (electrons) are defined similarly.
Lemma 4.1. For each (x, v) D R2 , the particle trajectories (X (s; x, v), V (s; x, v)) are piece-
wise C 1 smooth in s R, and for each s R, the map (x, v) 7 (X (s; x, v), V (s; x, v)) is
one-to-one and differentiable with Jacobian equal to one at all points (x, v) such that x 6 D and
X (s; x, v) 6 D. In addition, the standard change-of-variables formula
Z Z Z Z
g(x, v) dxdv = g(X (s; y, w), V (s; y, w)) dwdy (4.5)
D R2 D R2
is valid for each s R and for measurable functions g for which the integrals are finite.
Proof. For each (x, v), the particle trajectory (X (s; x, v), V (s; x, v)) is smooth except when it
hits the boundary D, which happens countably many times. So the first assertion is clear. Given s,
let S be the set (x, v) in DR2 such that X (s; x, v) 6 D. Clearly, S is open and its complement in
DR2 has Lebesgue measure zero. For each s, the trajectory map is one-to-one on S since the ODE
(4.3) and (4.4) are time-reversible and well-defined. In addition, a direct calculation shows that the
Jacobian determinant is time-independent and is therefore equal to one. The change-of-variable
formula (4.5) holds on the open set S and so on D R2 , as claimed.
Lemma 4.2. Let g(x, v) be a C 1 radial function on D R2 . If g is specular on D, then for all s,
g(X (s; x, v), V (s; x, v)) is continuous and also specular on D. That is,
g(X (s; x, v), V (s; x, v)) = g(X (s; x, v), V (s; x, v)), (x, v) D R2 ,
where v = (vr , v ) for all v = (vr , v ).
Proof. It follows directly by definition (4.3) and (4.4) that for all x D, the trajectory is unaffected
by whether we start with v or v. So for all s we have
X (s; x, v) = X (s; x, v), V (s; x, v) = V (s; x, v), |Vr (s; x, v)| = |Vr (s; x, v)|. (4.6)
In fact we have Vr (s; x, v) = Vr (s; x, v) for any s at which X (s; x, v) 6 D, while Vr (s+; x, v) =
Vr (s; x, v) for s at which X (s; x, v) D. Because g is specular on the boundary, it takes the
same value at vr and vr . Therefore g(X (s), V (s)) is a continuous function of s at the points
of reflection. It is specular because of the rule (4.4).

21
4.2 Representation of the particle densities
We can now invert the operator ( + D ) to obtain an integral representation of f . By definition
of the operator D+ from (2.7) and the trajectories (X + (s), V + (s)) from (4.3) and (4.4), we have
Z 0 Z 0
s + + + d
e D g(X (s), V (s)) ds = es g(X + (s), V + (s)) ds
ds
Z 0
= g(x) es g(X + (s), V + (s)) ds,

for C 1 functions g = g(x, v) which belong to dom(D+ ). Multiplying the Vlasov equations (4.1) by
es and then integrating along the particle trajectories from s = to zero, we readily obtain
Z 0 Z 0
+ + + + s + +
f (x, v) = e + rp e e (R (s)) ds + es + +
e V (s)(R (s)) ds.

A similar derivation holds for the case. For convenience we denote

Z 0

Q (g)(x, v) := es g(X (s; x, v), V (s; x, v)) ds.

In particular, by Lemma 4.2, Q (g) is specular on D if g is. Thus we have derived the integral
representation for the particle densities:

f (x, v) =
e (1 Q ) + rp + e Q (v ) (4.7)

4.3 Operators
We next substitute (4.7) into the Maxwell equations (4.2). We introduce the operators
Z Z
+
+
A1 : = e (1 Q ) dv e (1 Q ) dv,
Z Z  
A2 : = (r + ) rv (p + p )dv v +
2 + +
e Q (v ) + e Q (v ) dv,
Z Z (4.8)
+
B : = e (1 Q )(v ) dv
+
e (1 Q )(v ) dv,
Z Z
+
(B ) : = v e (1 Q ) dv v
+
e (1 Q ) dv.

We also introduce
L := A2 + (B ) (A1 )1 B . (4.9)
We then have
Lemma 4.3. The Maxwell equations (4.2) are equivalent to the equations

A1 = B , A2 + (B ) = 0. (4.10)

22
Proof. Using (4.7), we write the Poisson equation for as
Z
= (f + f )(r, v) dv
Z Z Z
= e (1 Q ) dv + e (1 Q ) dv + r(+
+ +
p + p )dv
Z Z
+ e Q (v ) dv +
+ +
e Q (v ) dv.

Note that v [ ] = r
p + v e so that
Z Z
r(p + p )dv = v (+
+
e + e ) dv.

This gives the first equation in (4.10) by definition. Similarly we write

Z
(r + ) = v (f + f )(r, v) dv
2

Z Z Z
+
= v e (1 Q ) dv + v e (1 Q ) dv + rv (+
+
p + p )dv
Z Z
+ +
+ v e Q (v ) dv + v e Q (v ) dv,

which is equivalent to the second equation in (4.10).

As in Lemma 3.1, we now state some properties of these operators. We recall that the spaces
V and V , which are defined in Section 1.2, incorporate the boundary conditions.

Lemma 4.4. For any > 0,

(i) A1 is self-adjoint and positive definiteR on L2r (D) with domain V. Moreover, A1 maps from
V one-to-one onto the set 1 := {g L2r : D g dx = 0}.
(ii) B is a bounded operator on L2r (D) with its adjoint operator (B ) defined in (4.8). The
range of B is contained in 1 .
(iii) A2 and L are self-adjoint on L2r (D) with their common domain V .

Proof. We first check the self-adjointness of Aj and the formula for (B ) . Since + is constant on
trajectories and in view of (4.8), it clearly suffices to prove, for smooth functions g and h specular
on the boundary, that
Z Z Z Z
+ +
+
e h(x, v)Q (g(x, v)) dvdx = +
e g(x, v)Q (h(x, v)) dvdx, (4.11)
D D

where we denote v = (vr , v ). In order to prove (4.11), we recall the definition of Q+

and use the
change of variables

(y, w) := (X + (s; x, v), V + (s; x, v)), (x, v) := (X + (s; y, w), V + (s; y, w)),

23
which has Jacobian one where it is defined (see Lemma 4.1). So we can write the left side of (4.11)
as Z 0 Z Z
es + + +
e h(x, v) g(X (s; x, v), V (s; x, v)) dvdxds
D
Z 0 Z Z
= es + + +
e h(X (s; y, w), V (s; y, w)) g(y, w) dwdyds.
D
Observe that the characteristics defined by (4.3) and (4.4) are invariant under the time-reversal
transformation s 7 s, r 7 r, vr 7 vr , and v 7 v . Thus
X + (s; x, v) = X + (s; x, v), V + (s; x, v) = V + (s; x, v),
at least if we avoid the boundary. Using this invariance, we obtain
Z 0 Z Z
es + + +
e h(x, v) g(X (s; x, v), V (s; x, v)) dvdxds
D
Z 0 Z Z
= es + + +
e h(X (s; y, w), V (s; y, w)) g(y, w) dwdyds
D
Z 0 Z Z
= es + + +
e h(X (s; x, v), V (s; x, v)) g(x, v) dvdxds,
D

in which the last identity comes from the change of notation (x, v) := (y, w). By definition of Q+,
this result is precisely the identity (4.11). A similar calculation holds for the case. This proves
the adjoint properties claimed in the lemma.

Next we show that all the integral terms in (4.8) are bounded operators on L2r (D). For instance,
we have
Z Z  Z 0 Z Z 
+ + s + +
Q dvdx = e (X (s)) dvdxds
   
 e   e 
D D
Z 0 Z Z 1/2  Z 0 Z Z 1/2
es |+
e |||2
dvdxds e s
| +
e ||(X +
(s))|2
dvdxds
D D
 Z 
sup |+
e | dv kkL2 kkL2 . D D
D
(4.12)
In the last step we made the change of variables (x, v) = (X + (s; x, v), V + (s; x, v)) in the integral
for , which is possible thanks to (4.5). Similar estimates hold for the other integrals since v is
bounded by one. This proves that B is bounded on L2r (D) and also that the integral terms in A1
and A2 are relatively compact with respect to and r , respectively. Therefore A1 and A2
are well-defined operators on L2r (D) with domains V and V , which are the same as the domains of
and r , respectively.

Taking = in the previous estimate, we have

Z Z  Z 0 Z Z Z Z
+ + s + 2
Q dvdx e | ||| dvdx = + 2
e || dvdx
 
 e  e
D D D

24
so that Z Z
+
+
e (1 Q ) dv 0.
D

Thus A1 0, and A1
= 0 if and only if is a constant. Since A1 has discrete spectrum, it is

R
invertible on the set orthogonal to the kernel of A1 . That is, it is invertible on {g V : D g dx =
0}. For the invertibility of A1 on the range of B , we note by (4.11) and Q (1) = 1 that
Z Z Z Z
e (1 Q
)(v ) dvdx =
e v (1 Q )(1) dvdx = 0.
D D

This shows that B {g V : D g dx = 0} for all . Thus (A1 )1 is well-defined on the range
R

of B , and so L is well-defined. The self-adjoint property of L is clear from that of A1 .

Part (i) of Lemma 4.4 in particular shows that for each L2r (D) there exists a unique radial
function Hr2 (D) that solves
Z

A1 = B , dx = 0, r (1) = 0. (4.13)
D

4.4 Construction of a growing mode

Lemma 4.5. If L0 6 0, then there exist a > 0 and a nonzero function H 2 (D) such that
L = 0 and satisfies the Dirichlet condition on the boundary.

Proof. The proof is similar to the one and a half dimensional case given in [15] so that we merely
outline the main steps as follows.
(i) L 0 for large .
(ii) For all L2r , L converges strongly to L0 in L2 as 0, and thus L 6 0 when is
small.
(iii) The smallest eigenvalue := inf hL , iL2 of L is continuous in > 0, where the infimum
is taken over V with kkL2 = 1.
These three steps imply that must be zero for some > 0, from which the lemma follows.
To prove (i), it is easy to see that L is nonnegative for large , since (B ) (A1 )1 B 0 and
hA2 , iL2 is sufficiently large when is large. As for (ii), to show the convergence of L to L0 as
0, we use the remarkable fact, proved in [15, Lemma 2.6], that for all g H the strong limit

lim Q
g = P g (4.14)
0+

is valid in the L2| | = H norm. Here the P are the orthogonal projections of L2| | onto the
e e
kernels of D . However, it should be noted that the convergence is not true in the operator norm.
For all L2r , we use (4.14) and write
XZ h i
A2 A02 = 2 v
e Q
(v
) P
(v ) dv,

25
thereby obtaining the convergence of A2 to A02 in L2 . Similarly, A1 and B also converge
strongly in L2 to A01 and B 0 , respectively, and so does L to L0 . Finally, estimates similar to (4.12)
yield
Z Z 
+
v + +
e (Q (v ) P (v )) dvdx
 

D
Z 0 Z Z   
= es es v + V (s)(X +
(s)) dvdxds
 
e 
D
Z 0 Z Z 1/2  Z Z 1/2
|es es | |+
e ||(x)| 2
dvdx | +
e ||(X +
(s))|2
dvdx ds
D D
Z 0 
C0 |es es | ds kk2L2 C0 | log log |kk2L2 ,
D D

and thus  
hA2 A2 , i C0 | | + | log log | kk2L2
D

for all , > 0 and L2r .

Similarly, we obtain the same estimate for L , which proves the
continuity of the lowest eigenvalue of L .

Using , we can now construct the growing mode.

Lemma 4.6. Let , be as in Lemma 4.5, let be as in (4.13), and let f be defined by (4.7).
Then (et f , et , et ) is a growing mode of the linearized Vlasov-Maxwell system.
Proof. Because L = 0 and due to the definition of , both parts of (4.10) are satisfied. Therefore
(4.2) is satisfied. These are the first and third Maxwell equations in (2.4) together with the boundary
conditions for and . Next, the specular boundary condition for f follows directly by definition
(4.7) and the fact that Q (g) is specular on the boundary if g is. It remains to check the Vlasov
equations and the middle Maxwell equation in (2.4), namely r = jr .
We begin with the equation for f + . Recall that (X + (t; x, v), V + (t; x, v)) is the particle trajectory
initiating from (x, v). Evaluating (4.7) along the trajectory, we have
Z 0
+ + + + + + + + +
f (X (t), V (t)) = e (X (t)) + p R (t)(X (t)) e es (X + (s; X + (t), V + (t))) ds

Z 0
+ +
e es V+ (s; X + (t), V + (t))(X + (s; X + (t), V + (t))) ds.

By the group property (X + (s; X + (t), V + (t)) = X + (s + t) and V + (s; X + (t), V + (t)) = V + (s + t),
together with integration by parts in s, we have for each t
f + (X + (t), V + (t)) = + + + + +
e (X (t)) + p R (t)(X (t))
Z 0 h i
+
e es (X + (s + t)) V+ (s + t)(X + (s + t)) ds

Z t h i
= + + + + t
p R (t)(X (t)) + e e es s (X + (s)) + V+ (s)(X + (s)) ds.

26
Differentiation of this identity yields
d  t + +  d  t +  h i
e f (X (t), V + (t)) = +
p e R (t)(X + (t)) + +
ee
t
t (X + (t)) + V+ (t)(X + (t)) .
dt dt
We evaluate the above identity for t (0, ) and let 0. Note that by Lemma 4.1 the functions
f + (X + (t), V + (t)), (X + (t)) and (X + (t)) are piecewise C 1 smooth. Using the evolution (4.3)
and (4.4), we obtain

f + + D+ f + = + + + + +
e vr r + rp vr r + p vr + (e v + rp ).

This is the Vlasov equation (2.8) for f + . A similar verification can be done for f .

Finally, we verify the remaining Maxwell equation r = jr . Indeed, by performing the

integration in v of the Vlasov equations (2.6), we easily obtain + j = 0. Together with the
Poisson equation in (4.2), this yields
 1  1
r + (r ) = = = r + jr
r r
Thus r(r jr ) must be a constant. However, at the boundary r = 1 we have r = 0 and
jr = 0 by the specular boundary condition on f . So r jr = 0.

This completes the proof of Theorem 1.1.

5 Examples
The purpose of this section is to exhibit some explicit examples of stable and unstable equilibria,
and thereby prove Theorem 1.2.

5.1 Stable examples

By Theorem 1.1 the condition for spectral stability is

L0 = A02 + (B 0 ) (A01 )1 B 0 0. (5.1)

For each in the domain of L0 (thus in particular satisfying the Dirichlet boundary condition), we
have
hL0 , iL2 = hA02 , iL2 + hA01 , iL2 ,
where solves A01 = B 0 with the Neumann boundary condition on . Recall that
Z Z
A1 = e (1 P ) dv
0 + +
e (1 P ) dv,

1
 Z Z  
A2 = + 2 rv (p + p ) dv v +
0 +
e P +
(v ) +
e P (v ) dv,
r

27
in which denote (e , p ) = (hvi 0 , r(v 0 )). Integration by parts, together with the
boundary conditions for and , and the orthogonality of P and 1 P (separately for + and
) lead to the expressions
Z Z Z Z Z
0 2 + + 2
hA1 , iL2 = |r | dx e |(1 P )()| dvdx 2
e |(1 P )()| dvdx,
ZD  D D
1  Z Z 
hA02 , iL2 = |r |2 + 2 ||2 dx rv (+ 2
p + p ) dv || dx (5.2)
D r D
Z Z h i
+e |P +
(v )|2
+
e |P (v )| 2
dvdx.
D

Due to the assumption 0

e < 0, it is clear that hA1 , iL2 0 and so are the first and last terms
in hA2 , iL2 . We now exhibit two explicit sufficient conditions for hA02 , iL2 to be nonnegative.
0

This is Theorem 1.2 (i) and (ii).

Theorem 5.1. Let ( , 0 , 0 ) be an inhomogenous equilibrium.
(i) If
p
p (e, p) 0, e, p, (5.3)
then the equilibrium is spectrally stable provided that 0
L and 0 is sufficiently small in L .
(ii) If

|
p (e, p)| , (5.4)
1 + |e|
for some > 2, with sufficiently small and 0 = 0 but 0 not necessarily small, then the
equilibrium is spectrally stable.
Proof. First consider case (i). We only need to show that A02 0. Let us look at the second
integral of hA02 , iL2 in (5.2). By the definition (1.9) of p , we may write
Z Z Z
1 0
rv p (e , p ) dv = hvi p p (e , p ) dv r hvi1
p (e , p ) dv,

in which the first term on the right is nonpositive due to (5.3). Therefore we have
Z Z  Z Z
+ 2
rv (p + p ) dv || dx hvi1 (+ 0 2
p p ) dv r || dx
D D Z
 Z
0 1 +
sup | | sup hvi (|p | + |p |) dv r||2 dx.
r r D
Now by the Poincare inequality,
1
Z Z  
r||2 dx c0 |r |2 + 2 ||2 dx,
D D r
for
R some constant c0 . In addition, thanks to assumption (1.12), the supremum over r [0, 1] of
hvi1 (|+ 0 0
p | + |p |) dv is finite if is bounded. Thus if the sup norm of is sufficiently small,
0
or more precisely if satisfies
 Z 
0
c0 sup | | sup hvi1 (|+
p | + |p |) dv 1, (5.5)
r r

28
then the second term in hA02 , iL2 is smaller than the first, and so the operator A02 is nonnegative.
Case (ii) is even easier. As above, we only have to bound the second term in hA02 , iL2 . Using
|rv | 1 and e = hvi together with (5.4), we have
Z Z  ZZ
 Z
+ 2
rv (p + p ) dv || dx dv|| dx C ||2 dx.
2
 

1 + |v|
D

If is sufficiently small, the second term is smaller than the positive terms.

5.2 Unstable examples

For instability, it suffices to find a single function in the domain of L0 such that hL0 , iL2 < 0.
We shall construct some examples where this is the case. We limit ourselves to a purely magnetic
equilibrium ( , E0 , B 0 ) with E0 = 0 and B 0 = 1r r (r 0 ). Thus e = hvi and p = r(v 0 ). In
this subsection, we shall also make the assumption that

+ (e, p) = (e, p), e, p. (5.6)

This assumption holds for example if + = is an even function of p. It greatly simplifies the
verification of the spectral condition on L0 .
We now show that assumption (5.6) implies that the operator B 0 vanishes and so L0 simply
reduces to A02 . Indeed, let us recall that
Z   Z  
0 + + + + +
B =r p (e, p ) + p (e, p ) dv + e (e, p )P (v ) + e (e, p )P (v ) dv,

in which e = hvi and p = r(v 0 ). For the first term in B 0 , we again note by (5.6) that the
function
+ + 0 0
p (e, p ) + p (e, p ) = p (hvi, r(v + )) + p (hvi, r(v ))

is odd in v . Thus, the first integral in B 0 vanishes. As for the second integral, we note that
 1   1 
D = vr r + (v )B 0 + (v )(v ) vr vr B 0 + vr (v ) v .
r r
That is, D acting on functions f (vr , v ) is the same as D+ acting on f (vr , v ) (ignoring the
dependence on x). As a consequence we have

P + (f (vr , v ))(vr , v ) = P (f (vr , v ))(vr , v ).

Using this identity together with the fact that P (f ) = P (f ), we have

+ 0 + 0
e (e, r(v + )P (v )(v ) + e (e, r(v )P (v )(v )
= 0 + 0 +
e (e, r(v ))P (v )(v ) e (e, r(v + ))P (v )(v )
= + 0 + 0
e (e, r(v + ))P (v )(v ) e (e, r(v ))P (v )(v ).

29
Thus the function + + 0
e P (v )+e P (v ) is odd in v , so that the second integral in B vanishes.
Similarly, we easily obtain
Z Z
A2 = r 2 rv p (e, p ) dv 2 v
0
e (e, p )P (v ) dv. (5.7)

We summarize the above considerations in the following lemma.

Lemma 5.2. If ( , 0, 0 ) be an equilibrium satisfying (5.6), then B 0 = 0 and

L0 = A02 ,

for A02 as in (5.7). In particular, A02 6 0 implies the spectral instability of ( , 0, 0 ).

5.2.1 Homogeneous equilibria

We start with the homogenous case E0 = 0 and B 0 = 0, in which case the linearized Vlasov
operator reduces to D = v x = vr r . The projection P = P is simply the average
Z 1
1
Z
P() = (r) dx = 2 (r) r dr
D 0

for any radial function = (r). In addition, noting that D(rv ) = 0, we have
1
1
Z Z
P(v ) = rv P( ) = rv r 1 (r) dx = 2rv R , R := (r) dr.
r D 0

Thus, as in (5.2), we obtain the basic identity

1
Z   Z Z  Z Z
0 2 2 2
hA2 , iL2 = |r | + 2 || dx 2 rv p dv || dx 2 2
e |P(v )| dvdx
r
ZD  ZD  Z DZ Z
2 1 2

1

2
= |r | + 2 || dx 2 hvi pp dv || dx 8 r 2 v2
e dvdx|R |
2
D r D D
= I + II + III,

in which p = rv .

Theorem 5.3. Let = (e, p) be an homogenous equilibrium satisfying (5.6). Assume also that

p 2
p (e, p) c0 p (e), e, p, (5.8)

for some positive constant c0 and some nonnegative continuous function (e) such that 6 0.
Then there exists a positive number K0 such that both of the rescaled homogenous equilibria (i)
(K), (e, p) := K (e, p) and (ii) (K), (e, p) := (e, Kp) are spectrally unstable, for all K K0 .

30
Proof. Note that terms I and III are nonnegative since e < 0. So by Lemma 5.2, it suffices for
instability that the middle integral II be negative and dominate the other two. We begin with
case (i) of the theorem. Observe that since (K), (e, p) satisfy (5.6), the pair ((K),+ , (K), ) is
indeed an homogeneous equilibrium. It suffices to construct a function H 2 (D) such that
hA02 , iL2 < 0. We choose
1

1 (r),
0r
(r) := 2 (5.9)
1 (1 r), 1
r 1,

2
where 1 (r) = 0 r( 12 r), for some normalizing constant 0 . Clearly ( )R = 0 so that III = 0.
Moreover,
Z 1/2 
1 1
Z   
2 2
I= |r | + 2 | | dx = 2 |r 1 |2 + |1 |2 dr = 1
D r 0 r(1 r)
by choice of the constant 0 . So it remains to show that II < 1.
Using the assumption (5.8) and the first scaling (i), we have
Z Z  Z 1Z
1 (K), 2
II = 2 hvi pp dv | | dx 4c0 K hvi1 p2 (e) dv| |2 rdr
D 0
Z Z 1
1 2
4c0 K hvi v (e) dv r 3 | |2 dr.
0

The integral in v is a finite positive constant thanks to the decay assumption on . So we can
choose K large enough that II > 1. This settles case (i).
Similarly, for case (ii) we have
Z Z 
II = 2 hvi1 (Kp) 2
p (e, Kp) dv | | dx
D
Z 1Z Z Z 1
4c0 K 2 hvi1 p2 (e) dv| |2 rdr = 4c0 K 2 hvi1 v2 (e) dv r 3 | |2 dr,
0 0

which is again greater than one for sufficiently large K. This settles case (ii).

We remark that the constant K0 in Lemma 5.3 is certainly not optimal. For instance, we could
take to be the ground state of the operator r , which is a Bessel function.

5.2.2 Inhomogeneous equilibria

For spatially dependent equilibria we will prove a similar result. We first observe as in the homo-
geneous case that the projection P satisfies
1
Z

P () = (r) dx = 2(r)R , (5.10)
D

31
 R1
for functions = (r), where ()R := 0 (r) dr. We recall from (5.2) that

1
Z   Z Z  Z Z
0 2 2 2
hA2 , iL2 = |r | + 2 || 2 rv p dv || dx 2 2
e |P (v )| dvdx
D r D D

= I + II + III, where clearly I 0 and III 0. Of course, both e = hvi and p = r(v 0 )
belong to the kernel of D . Thus we have
   
P (v ) = hvi1 P (v ) = hvi1 P r(v 0 ) + 0 = hvi1 p P + hvi1 P ( 0 ).
r r
Since 0 and are functions depending only on r, we apply (5.10) to give

P (v ) = 2hvi1 p R + 2hvi1 (r 0 )R . (5.11)

By definition of p , we can write

Z Z Z
1 0
rv p dv = hvi p p dv + r hvi1
p dv.

Using the inequality (a + b)2 2a2 + 2b2 , we then obtain

1
Z   Z Z 
0 2 2
hA2 , iL2 |r | + 2 || dx 2 hvi1 p
p dv ||2 dx
D Z Z r D
Z  p
16 hvi (p ) e dvdx|R |2 + 2 sup
2 2
hvi1 | p | dv k r| 0 |k2L2 (D)
D r[0,1]
Z 
+ 16 sup hvi2 | 0 2
e | dv (r )R = I + IIA + IIIA + IIB + IIIB.
r[0,1]
(5.12)
We now scale in the variable p to get the following result.

Theorem 5.4. Assume that satisfy (5.6) and that

p 2
p (e, p) c0 p (e), e, p, (5.13)

for some positive constant c0 and some nonnegative function (e) such that 6 0. Define
(K), (e, p) := (e, Kp) and let (K),0 be the solution of the equation
Z h i
(K),0
r = v (K),+ (hvi, r(v + (K),0 )) (K), (hvi, r(v (K),0 )) dv, (5.14)

with (K),0 = 0 on the boundary D. Then there exists a positive number K0 such that the inho-
mogenous purely magnetic equilibria ((K), , 0, B (K),0 ), with B (K),0 = 1r r (r (K),0 ), are spectrally
unstable for all K K0 .

32
Proof. As before, we will check the instability by showing hA02 , iL2 < 0 for some . As in
Lemma 5.3, we make the simple choice of given by (5.9) so that ( )R = 0, whence IIIA = 0,
with the constant chosen so that the first integral term I in (5.12) equals 1. Thus we obtain
Z Z 
0
hA2 , iL2 1 2 hvi1 Kp 2
p (hvi, Kp ) dv | | dx
D
Z 
(K),0
+ C0 Kk kL sup hvi1 |
p (hvi, Kp )| dv (5.15)
r[0,1]
Z 
+ C0 k (K),0 k2L sup hvi2 |
e (hvi, Kp )| dv ,
r[0,1]

for some constant C0 that depends only on the L2 norm of . We shall show that the second integral
IIA in this estimate dominates all the other terms if K is large. From the decay assumption (1.12)
on
e and p , we have

1
Z Z
1
hvi |p (hvi, Kp )| dv C dv C ,
hvi(1 + hvi )
with > 2 and for some constant C independent of K. A similar estimate holds for the last
integral in (5.15). Now by using the assumption (5.13) and the fact that (hvi) is even in v , we
have
Z Z 
IIA = 2 hvi1 Kp
p (hvi, Kp ) dv | | dx
2
D
Z Z 
2
2c0 K r 2 hvi1 (v (K),0 )2 (hvi) dv | |2 dx
ZD  Z  Z Z
2 1 2 2 2 2 1
= 2c0 K hvi v (hvi) dv r | | dx 2c0 K hvi (hvi) dv r 2 | (K),0 |2 | |2 dx
D D
Z 
2 1 2 2 2 2
2c0 K hvi v (hvi) dv kr kL2 (D) = c1 K kr kL2 (D) ,

where c1 > 0 is independent of K. Combining these estimates, we have therefore obtained

hA02 , iL2 1 c1 K 2 kr k2L2 (D) + C0 C k (K),0 kL (K + k (K),0 kL ).

Furthermore, the L2 norm of r is nonzero. We claim that (K),0 is uniformly bounded indepen-
dently of K. Indeed, recalling that (K),0 satisfies the simple elliptic equation (5.14) and using the
decay assumption (1.12) on , we have
1
Z
(K),0
|r | C dv C
1 + hvi
for some constant C independent of K. Thus letting u(K) (r) := (K),0 (r) + C r 2 /3, we observe
that r u(K) 0 in D and u(K) = C /3 on the boundary D. By the standard maximum principle,
u(K) is bounded above and consequently so is (K),0 . In the same way they are bounded below.
This proves the claim. Summarizing, we conclude that hA02 , iL2 is dominated for large K by
IIA and it is therefore strictly negative.

33
A Equilibria
This appendix contains (i) the proof of regularity of our equilibria and (ii) the construction of some
simple examples of equilibria. We recall that E0 = r 0 er and B 0 = 1r r (r 0 ) where (0 , 0 )
depends only on r and satisfies the elliptic ODE system (1.11), which we rewrite as

0 = h(r, 0 , 0 ), r 0 = g(r, 0 , 0 ), (A.1)

Z h    i
0 0
h(r, , ) : = + hvi + 0 , r(v + 0 ) hvi 0 , r(v 0 ) dv
Z h    i
g(r, , ) : = v + hvi + 0 , r(v + 0 ) hvi 0 , r(v 0 ) dv.
0 0

For the regularity (i), we will verify that 0 , 0 C(D) implies that E0 , B 0 C 1 (D). Observe
that 01 = 1 and 02 = log r are two independent solutions of the homogeneous ODE 0 = 0
with wronskian 1/r. Similarly, 0 = r and 0 = 1/r are two solutions of the homogeneous ODE
r 0 = 0 with wronskian 2/r. Thus all the solutions (0 , 0 ) to (A.1) satisfy the integral
equations Z r
0 (r) = + s(log s log r)h(, 0 , 0 )(s) ds + log r,
0
Z r (A.2)
0 1
(r) = r + (s2 r 2 )g(, 0 , 0 )(s) ds + ,
2r 0 r
with arbitrary constants , , , . Since 0 and 0 are assumed to be continuous at the origin,
we require = = 0. Clearly h is continuous in D, so that 0 C 2 (D) by (A.2) and E0 =
0 C 1 (D). As for g, we note that limr0+ g(r, 0 (r), 0 (r)) = 0, which follows from the fact
the integrand is odd in v . So g is also continuous in all of D. Hence, B 0 = 1r r (r 0 ) C 1 (D), as
can be seen from (A.2).

As for (ii), the construction of some equilibria, for simplicity we only consider the case + (e, p) =
(e, p) = (e, p), which we take to be an arbitrary function subject to the conditions in (1.12).
Of course, E0 = 0, B0 = 0 is automatically an equilibrium for any . However, let us consider
the inhomogeneous case. Clearly, h(r, 0, 0) = g(r, 0, 0) = 0. No boundary condition is required on
0 , 0 . It is easy to choose so that the functions h(r, , ) and g(r, , ) are uniformly bounded, and
so that for all 1 , 1 , 2 , 2 they satisfy

|h(r, 1 , 1 ) h(r, 2 , 2 )| + |g(r, 1 , 1 ) g(r, 2 , 2 )| (|1 2 | + |1 2 |) (A.3)

for some < 1. Such an assumption is satisfied for instance if is uniformly Lipschitz in its
variables and replaced by for sufficiently small .
Now we denote by T (0 , 0 ) the right sides of the integral equations in (A.2) with = = 0.
It is clear from assumption (A.3) that T is well-defined from C([0, 1]) C([0, 1]) into itself. In
addition, T is a contraction map on a small ball B in this space if and are sufficiently small.
So for each small and there exists a unique solution (0 , 0 ) in B to (A.2) and thus to (A.1).

34
References
[1] J. Batt and K. Fabian, Stationary solutions of the relativistic Vlasov-Maxwell system of
plasma physics. Chinese Ann. Math. Ser. B 14 (1993), no. 3, 253278.

[2] R. J. DiPerna and P.-L. Lions, Global weak solutions of Vlasov-Maxwell systems. Comm.
Pure Appl. Math. 42 (1989), no. 6, 729757.

[3] R. T. Glassey and J. Schaeffer, The two and one-half-dimensional relativistic Vlasov
Maxwell system. Comm. Math. Phys. 185 (1997), no. 2, 257284.

[4] R. T. Glassey and J. Schaeffer, The relativistic Vlasov-Maxwell system in two space
dimensions. I, II. Arch. Rational Mech. Anal. 141 (1998), no. 4, 331354, 355374.

[5] Y. Guo, Global weak solutions of the Vlasov-Maxwell system with boundary conditions. Comm.
Math. Phys. 154 (1993), no. 2, 245263.

[6] Y. Guo, Stable magnetic equilibria in collisionless plasmas. Comm. Pure Appl. Math. 50
(1997), no. 9, 891933.

[7] Y. Guo and W. A. Strauss, Instability of periodic BGK equilibria. Comm. Pure Appl.
Math. 48 (1995), no. 8, 861894.

[8] Y. Guo and W. A. Strauss, Nonlinear instability of double-humped equilibria. Ann. Inst.
H. Poincare Anal. Non Linaire 12 (1995), no. 3, 339352.

[9] Y. Guo and W. A. Strauss, Unstable BGK solitary waves and collisionless shocks. Comm.
Math. Phys. 195 (1998), no. 2, 267293.

[10] Y. Guo and W. A. Strauss, Relativistic unstable periodic BGK waves. Comput. Appl.
Math. 18 (1999), no. 1, 87122.

[11] Y. Guo and W. A. Strauss, Magnetically created instability in a collisionless plasma. J.

Math. Pures Appl. (9) 79 (2000), no. 10, 9751009.

[12] H. J. Hwang and J. Velazquez, Global existence for the Vlasov-Poisson system in bounded
domains. Arch. Ration. Mech. Anal. 195 (2010), no. 3, 763796.

[13] Z. Lin, Instability of periodic BGK waves. Math. Res. Lett. 8 (2001), no. 4, 521534.

[14] Z. Lin, Nonlinear instability of periodic BGK waves for the Vlasov-Poisson system. Comm.
Pure Appl. Math. 58 (2005), 505-528.

[15] Z. Lin and W. A. Strauss, Linear Stability and instability of relativistic Vlasov-Maxwell
systems, Comm. Pure. Appl. Math., 60 (2007), no. 5, 724787.

[16] Z. Lin and W. A. Strauss, Nonlinear stability and instability of relativistic VlasovMaxwell
systems. Commun. Pure. Appl. Math. 60, 789837 (2007)

35
[17] Z. Lin and W. A. Strauss, A sharp stability criterion for the Vlasov-Maxwell system, Invent.
Math. 173 (2008), no. 3, 497546

[18] P.L. Lions and B. Perthame, Propagation of moments and regularity for the 3-dimensional
Vlasov-Poisson system, Invent. Math., 105:415430, 1991.

[19] O. Penrose, Electrostatic instability of a non-Maxwellian plasma, Phys. Fluids 3, 1960, pp.
258265.

[20] K. Pfaffelmoser, Global classical solutions of the Vlasov-Poisson system in three dimensions
for general initial data, J. Diff. Eqns., 95:281303, 1992.

36