Arch. Rational Mech. Anal.

211 (2014) 359388

Digital Object Identifier (DOI) 10.1007/s00205-013-0664-2

Real Solutions to the Nonlinear Helmholtz

Equation with Local Nonlinearity
Gilles Evquoz & Tobias Weth
Communicated by P. Rabinowitz

Abstract
In this paper, we study real solutions of the nonlinear Helmholtz equation
u k 2 u = f (x, u), x R N
satisfying the asymptotic conditions
 1N 
 1N 
2u
2
and
as r = |x| .
u(x) = O |x| 2
(x)
+
k
u(x)
=
o
|x| 2
r 2
We develop the variational framework to prove the existence of nontrivial solutions
for compactly supported nonlinearities without any symmetry assumptions. In addition, we consider the radial case in which, for a larger class of nonlinearities,
infinitely many solutions are shown to exist. Our results give rise to the existence of
standing wave solutions of corresponding nonlinear KleinGordon equations with
arbitrarily large frequency.

1. Introduction
The study of the existence and qualitative properties of solutions to nonlinear
wave equations
2
(t, x) (t, x) + V (x)(t, x) = f (x, (t, x)), (t, x) R R N , (1)
t 2
goes back to the sixties, see for example the classical paper by Jrgens [13].
Since then, many authors have been investigating various aspects of this problem,
including the question of existence and orbital stability of standing wave (or solitary
wave) solutions. An ansatz for solutions of this type is given by
(t, x) = ei(t+) u(x), , R,

(2)

360

Gilles Evquoz & Tobias Weth

with a real-valued function u on R N . Taking, for example, a nonlinearity of the

form f (x, ) = g(x, ||2 ), where g is a real-valued function, we see that such
a solves (1) if and only if u solves the reduced wave equation
u + V (x)u 2 u = f (x, u), x R N .

(3)

The requirement that u be real-valued guarantees that the corresponding energy

density

2

1 
 + |(t, x)|2 + V (x)|(t, x)|2
(t,
x)
E(t, x) =


2
t

 |(t,x)|2

g(x, ) d
0

is constant in t at every point x R N , a property which is characteristic of standing

wave solutions of (1). Note that, due to the presence of the linear potential V , it is
natural to consider nonlinearities f satisfying u f (x, 0) = 0 on R N . In this case,
in almost all of the available literature, it is assumed that 2 is not contained in the
essential spectrum of the Schrdinger operator  + V . We refer the reader to the
surveys and monographs [3,8,19,24,25,29] and the references therein for results in
this case. In the special case where V V0 is a constant, this restriction amounts to
assuming V0 > 2  0. Some authors have also considered the limiting case where
2 coincides with the infimum or another boundary point of the essential spectrum
of V , see, for example, [14] for a survey of classical results and [13,6,9,17,21,30]
for more recent work in this case. On the contrary, very little seems to be known if
2 is contained in the interior of the essential spectrum of V . In the present paper,
we consider this situation in the special case V V0 and 2 > V0 , so by setting
k 2 = 2 V0 we arrive at the nonlinear Helmholtz equation
u k 2 u = f (x, u), x R N ,

(4)

with k > 0. It seems unclear a priori in which space one should approach this
problem and whether variational methods can be used. In the present paper, we
provide first results in the case where f is supported in a bounded subset of R N ,
that is, f vanishes in [R N \ B R ] R for R > 0 sufficiently large. Here, B R R N
denotes the open ball centered at 0 with radius R. Note that in this case no nontrivial
solution of (4) exists in the space L 2 (R N ), as follows immediately from a classical
result of Rellich, see [20, Satz 1]. We will focus instead on solutions satisfying the
asymptotic conditions
 1N 
 1N 
2u
and
as r = |x| . (5)
u(x) = O |x| 2
(x)+k 2 u(x) = o |x| 2
2
r
Thus, the solutions decay to zero as |x| if N  2. We emphasize that, in
general, not all solutions of (4) satisfy (5) if N  2. In particular, in case N  2
and f 0, (5) is not satisfied by one-dimensional standing wave solutions of (4)
given by
x  sin(kx + ) with a unit vector R N

and R.

(6)

Real Solutions to the Nonlinear Helmholtz Equation

361

The above restriction on f allows us to work with the Dirichlet to Neumann map
associated with the exterior problem for the linear Helmholtz equation u + k 2 u =
0 on R N \ B R , together with a suitable asymptotic condition on u. To explain this in
more detail, let us suppose for a moment that the nonlinearity f (x, u) is replaced by
an inhomogeneous source term f (x) supported in B R . In this case, a well-studied
problem is to analyze the far field expansion of the (unique) complex solution of
(4) satisfying the Sommerfeld (outgoing) radiation condition




 u

 (x) iku(x) = o r 1N
2
as r = |x| .
(7)
 r

This condition has been introduced in Sommerfelds classical work [22], and it
corresponds to the study of outgoing waves excited by the source term f (x). Moreover, by a well-known result (going back to Rellich [20, p.58]), for given and
sufficiently regular Dirichlet boundary data on S R := B R , there exists a unique
complex solution of u + k 2 u = 0 in R N \ B R satisfying (7). Furthermore, the corresponding Dirichlet to Neumann map TR on S R , also called the capacity operator
(see [18]), is well understood and can be computed explicitly in terms of spherical
harmonics, see Section 6 below. This operator assigns, to a given boundary datum
on S R , the normal derivative of the corresponding unique solution u + k 2 u = 0
in R N \ B R satisfying (7). However, condition (7) rules out (real-valued) standing
wave solutions u which are the subject of the present paper. Nevertheless, a standing
wave solution uas explained in [22, pp. 328329]can be realized as a superposition of an outgoing wave and an incoming wave having opposite frequency
and thus as real part of a function satisfying (7). In particular, it still satisfies the
weaker asymptotic condition (5). Technically, this amounts to working with the
real part K R of the operator TR restricted to a Sobolev space of real functions on
S R , see Section 2 below. With the help of the operator K R , we will be able reduce
the problem of finding real solutions of (4) satisfying (5) to a variational problem
in H 1 (B R ). While such a reduction clearly requires that f vanishes outside of B R ,
we shall also see that the vanishing of f leads to difficulties in the proof of Ceramis condition, which is needed to show the existence of critical points of the
corresponding functional.
To state our results, we need to introduce further notation and to state our
assumptions. Let 2 denote the critical Sobolev exponent, that is, 2 := 2N /(N 2)
if N  3 and 2 := if N = 1, 2. For our main result, we shall suppose that the
nonlinearity f : R N R R is continuous, and that there exists a bounded set
R N of positive measure such that the following holds:
( f 0 ) f (x, u) = 0 for x R N \ , u R.
( f 1 ) There exists a > 0, p (2, 2 ) such that | f (x, u)|  a(1 + |u| p1 ) for
every x R N , u R.
( f 2 ) f (x, u) = o(|u|) uniformly in x as u 0.
( f 3 ) F(x, u)  0 for every x R N , u R, and F(x, u)/u 2 as |u|
for every x .
( f 4 ) There exists s0 > 0 such that for every x we have f (x, s0 )  0 
f (x, s0 ), and the map u  f (x, u)/|u| is nondecreasing on (s0 , ) and on
(, s0 ).

362

Gilles Evquoz & Tobias Weth

Here, we set F(x, u) =

nonlinearities of the type

u
0

f (x, s) ds for x R N , u R. We point out that

f (x, u) = q(x)|u| p2 u or f (x, u) = q(x)u log(1 + |u|s ), s > 0,

satisfy these assumptions if q is continuous and q > 0 on , q 0 on R N \ .
Moreover, if f satisfies ( f 0 )( f 4 ) and if g : R N R [0, ) is continuous,
vanishes outside of a bounded subset of [0, ) and satisfies ( f 2 ), then the sum
f + g also satisfies ( f 0 )( f 4 ).
We need one more definition related to the asymptotics of the solutions we
1 (R N , R) such
consider. For R > 0, we let R R denote the set of all functions u Hloc
N
N

that u = Re(w) in R \ B R for a function w C (R \ B R , C)C (R N \ B R , C),

satisfying
(i) w u on S R ,
(ii) w + k 2 w= 0 in R N \ B R ,
 1N 


1N

2
(iii) w(x) = O r 2 and  w
r (x) ikw(x) = o r

as r = |x| .

Note that every function u R R satisfies the asymptotic conditions (5). Our main
result is the following.
Theorem 1. There exists a countable set D (0, ) such that, for R (0, )\D
with B R and every nonlinearity f satisfying assumptions ( f 0 )( f 4 ), we have:
(i) Equation (4) admits a solution in R R .
(ii) If, in addition, f (x, t) = f (x, t) for all (x, t) R N R, then equation (4)
admits a sequence of pairs of solutions {u n }, n N in R R with the property
that
u n  H 1 (B R ) as n .

(8)

Some remarks seem in order. Elliptic regularity theory implies that the solutions
2,q
given by Theorem 1 belong to Wloc (R N ) for all 1  q < , so there are strong
1,
solutions of (4) in Cloc
(R N ) for all 0 < < 1. The set D in Theorem 1 is defined
by the property that an associated R-dependent linear eigenvalue problem admits
the eigenvalue zero iff R D, see Lemma 2(iv) below for details. The restriction
R (0, ) \ D in Theorem 1 is not necessary, but the proof is more complicated
in the case where R D. Since the choice of R has no impact on the validity of the
asymptotic conditions (5), we decided not to consider the case R D in the present
paper. Nevertheless, it is natural to ask whether different choices of R in Theorem 1
give rise to different solutions of problem (4), (5). The following partial answer to
this question can be derived from a careful study of the explicit representation of
the capacity operator (and its real part) discussed in Section 6 below.
Theorem 2. For fixed R > 0 and u R R with u| S R  0, there exist at most
countably many R > R such that u R R .
As a consequence, the problem (4), (5) admits infinitely many solutions if the
nonlinearity f satisfies assumptions ( f 0 )( f 4 ).

Real Solutions to the Nonlinear Helmholtz Equation

363

It might be somewhat surprising at first glance that Theorem 2 also gives rise
to infinitely many solutions u n , n N of (4), (5) in the case where f is not odd in
its second variable. However, a local unboundedness property as in (8) should not
be expected for these solutions. It is natural to ask whether it is possible to relax
the condition that f vanishes outside a compact set. We have no general answer
to this question, yet, but in the case where f is radial in x, a shooting argument
yields radial solutions of (4), (5) under much less restrictive assumptions on f . For
a precise result, see Theorem 4 below. We do not even need to assume that f tends
to 0 as r = |x| .
The paper is organized as follows. In Section 2, we set up the variational framework used to prove Theorem 1. Here we also state key properties of the capacity
operator TR and its real part K R in the case N  2, but we postpone the derivation
of these properties to Section 6 since the underlying computationsrelying on special properties of Hankel functionsare somewhat technical. In Section 3 we then
complete the proof of Theorem 1 in the case N  2. As already remarked above, a
key difficulty in the proof is the validity of Ceramis condition (see Proposition 1),
and the proof of this property is rather long. In Section 3, we also establishunder
stronger assumptions on the nonlinearitya rigid minimax principle with respect to
families of half-spaces for the solution which minimizes the corresponding energy
functional among all critical points, see Theorem 3. We believe that this minimax
principle gives rise to additional properties of the corresponding (ground state) solutions. Section 4 contains a sketch of the proof of Theorem 1 in the one-dimensional
case. This case is much easier than the case N  2, but must be treated slightly
differently. Section 5 is devoted to the radial case. As pointed out already, we apply
a shooting argument to prove the existence of infinitely many radial solutions of
(4), (5) under much less restrictive assumptions on f , see Theorem 4. Finally, as
noted above, in Section 6 we derive key properties of the capacity operator TR and
its real part K R . We note that some of these properties are well known (see, for
example, [18] for the case N = 3), but we could not find an appropriate reference
for general N  2. Moreover, it seems that the operator K R has not been studied
in the degree of detail that we need for our purposes. The final part of Section 6 is
devoted to the proof of Theorem 2.
We close this introduction with a remark on some open questions which we
plan to address in future work. First, one may ask for which compactly supported
nonlinearities f satisfying ( f 0 ) there exist solutions of (4) given as superposition of a function of type (6) and a function satisfying (5). Second, one may try
to prove the existence of standing wave solutions without assumption ( f 0 ). This
would require a completely different (variational) approach. Another natural question, more closely related to the scattering theory of (1), is the existence of complex
solutions of (4) satisfying the radiation condition (7). It is not clear whether solutions can be expected in the case where f (, 0) 0 as assumed in ( f 2 ). This
problem is non-variational due to the lack of symmetry of the Dirichlet to Neumann map TR associated with (7), so a completely different approach has to be
used.

364

Gilles Evquoz & Tobias Weth

2. The Variational Framework

We assume that N  2 in the next two sections, referring to Section 4 for the
case N = 1. In this section we will introduce the capacity operator and develop a
variational framework for problem (4), (5). We start by fixing some notation. Let
R > 0 be such that B R , where B R := B R (0) is the open ball with radius R
centered at zero. We also set E R := R N \ B R and consider the space

u
C
1
(E R , C) :
L 2 (E R , C),
H R := u Hloc
(1 + r 2 )1/2

u
2
N u
2
iku L (E R , C) ,
L (E R , C ),
(9)
(1 + r 2 )1/2
r
wherehere and in the followingr always denotes the radial variable, that is,
r = |x|. It is known (see [18] for the case N = 3 and Section 6 below for general
1
N  2) that for every u H 2 (S R , C), S R := B R , there exists a unique weak
C
solution w H R of the problem

w + k 2 w = 0 in E R ,
(10)
w=u
on S R .
Here, weak solution means that trace(w) = u and
 

w k 2 w dx = 0 for all Cc1 (E R ),
ER

where Cc1 (E R ) denotes the space of C 1 -functions with compact support in E R . We

then define the capacity operator (or Dirichlet to Neumann map)
1

TR : H 2 (S R , C) H 2 (S R , C), TR u =

1
w
H 2 (S R , C),

(11)

where w H RC is the unique solution of (10) corresponding to u H 2 (S R , C)

and denotes the normal unit vector field on S R pointing outside B R and inside
ER.
As shown in [18] for the case N = 3 and detailed in Section 6 below for
1
1
general N  2, the operator TR : H 2 (S R , C) H 2 (S R , C) is continuous,
and suitably normalized spherical harmonics (when considered as functions of
spherical angles) form an orthonormal basis of eigenfunctions of TR . The operator TR is not symmetric and, therefore, does not give rise to a variational
formulation of our nonlinear problem given by (4) and (5). Therefore, we set
1
1
1
1
H 2 (S R ) = H 2 (S R , R), H 2 (S R ) = H 2 (S R , R), and we let
1

K R : H 2 (S R ) H 2 (S R ),

K R u = Re[TR u]
1

(12)

denote the real part of the restriction of TR to H 2 (S R ). This operator can be seen
as the Dirichlet to Neumann map corresponding to the problem (10), with real data
u on S R and solutions w given as a real part of a function in H RC . The explicit

Real Solutions to the Nonlinear Helmholtz Equation

365

calculations of w in terms of u in Section 6 below immediately imply that w

satisfies the asymptotic conditions (5). Moreover, the operator K R has the following
properties.
Lemma 1. The operator K R is bounded, symmetric and negative definite. More
precisely, there are constants R > 0 and  R > 0 such that


K R u 1
  R u 1
and
u K R u d  R
u 2 d (13)
H

2 (S R )

H 2 (S R )

SR

SR

1
2

for all u H (S R ). Moreover,



v K R u d =
SR

u K R v d

for u, v H 2 (S R ).

(14)

SR

We postpone the proof of this lemma to Section 6, below. Setting X := H 1 (B R ),

we now recall the standard estimate



u 2 dx  c
|u|2 dx + c
u 2 d for u X
(15)
BR

BR

SR

with some constant c = c(R) > 0, see, for example, [24, Theorem A.9]. Moreover,
we consider the bilinear form
 


2
Bk : X X R, Bk (u, v) =
u v k uv dx
v K R u d. (16)
BR

SR

From (13) and (15), we easily deduce the following

Corollary 1. B0 defines a scalar product on X = H 1 (B R ) that is equivalent to
the standard scalar product, that is, the corresponding norms are equivalent.
We also set Bk (u) := Bk (u, u) for u X in the following. In the next lemma,
we collect key facts concerning Bk and the (nonlocal) eigenvalue problem

in B R ,
u = u,
(17)
u

= K Ru
on S R .

Lemma 2. (i) The eigenvalue problem (17) admits an unbounded sequence of

eigenvalues 0 < 1  2  . . . and a corresponding system of eigenfunctions
e j , j N, which is complete in X . Moreover, each eigenfunction e j , j N is
analytic in B R .
(ii) There exists a scalar product , , equivalent to the standard scalar product
on X = H 1 (B R ), and an orthogonal splitting X = X X 0 X + , such that
Bk (u) = u + 2 u 2

for all u X,

1
2

where  = ,  is the induced norm, and u , u 0 denote the corresponding

orthogonal projections of u onto X , X 0 , respectively. More precisely,
X = span{e j : j  j }, X 0 = span{e j : j < j < j } and
X + = span{e j : j  j },

366

Gilles Evquoz & Tobias Weth

where j := max{ j N : j < k 2 } and j := min{ j N : j > k 2 }. In

particular, X and X 0 are finite dimensional.
(iii) The family e j , j N is orthogonal with respect to the scalar products
, , B0 and the scalar product of L 2 (B R ).
(iv) There exists a countable set D (0, ) such that X 0 = {0} if and only if
R D.
Proof. (i) We first consider H 1 (B R ) with the equivalent scalar product B0 . Since
K R is negative definite, all eigenvalues of (17) must be positive. Hence, u
H 1 (B R ) is an eigenfunction of (17) corresponding to R if and only if
> 0 and u is an eigenfunction of the operator K L (X ) defined by

uv dx for u, v X,
(18)
B0 (K u, v) =
BR

corresponding to the eigenvalue 1 . The operator K is bounded, symmetric with

respect to B0 , nonnegative and compact, since the embedding X  L 2 (B R )
is compact. Moreover, 0 is not an eigenvalue of K . Hence, K admits a sequence
of positive eigenvalues 1  2  . . . such that j 0 as j , and a
corresponding complete system of B0 -orthogonal eigenfunctions e j , j N.
Assertion (i) now follows with j = 1j , j N and the family {e j : j N}
thus obtained.
(ii) For u X , let u , u 0 denote the L 2 (B R )-orthogonal projections of u onto the
subspaces X , X 0 , respectively, as defined in the assertion. For u, v X , we
define

+ +

u 0 v 0 dx.
u, v = Bk (u , v ) Bk (u , v ) +
BR

It is easy to see that this scalar product has the desired properties and, by
construction, the splitting X = X X 0 X + is orthogonal with respect to
this scalar product.
(iii) The B0 -orthogonality of the family {e j : j N} has already been shown
above, and the L 2 (B R )-orthogonality then follows from (18). From this, the
orthogonality with respect to ,  immediately follows by definition.
(iv) This part, which relies on special properties of the capacity operator and
Hankel functions, will be proved in Section 6, below. 

We now consider the functional

1 + 2
1
Bk (u) (u) =
u  u 2 (u),
(19)
2
2

t
where (u) = B R F(x, u(x)) dx for u X and F(x, t) = 0 f (x, s) ds for t R.
It is well known that C 1 (X, R) as a consequence of assumption ( f 1 ), and
: X R, (u) =

is nonnegative on X

(20)

Real Solutions to the Nonlinear Helmholtz Equation

367

by ( f 3 ). Moreover, the critical points of correspond to restrictions to B R of

solutions of (4). Indeed, if u X is a critical point of , then
 


0=
u w k 2 uw f (x, u)w dx
wK R u d
BR

SR

for every w X , hence u is a weak solution of the problem

u k 2 u = f (x, u) in B R ,
u
(21)
= K Ru
on S R .

As explained in Section 6, extending u by the real part of the unique solution of

(10) in H RC then yields a solution of (4), (5).

3. Proof of Theorem 1
We fix R (0, ) \ D from now on, so that X 0 = {0} by Lemma 2. As
remarked in the introduction, this restriction is made only in order to simplify the
proofs, whereas all results can still be provedwith additional effortin the case
where X 0 has positive dimension.
We first collect useful facts about the functional defined in (19). For this, we
recall the following well-known consequence of ( f 1 ) and ( f 2 ):
> 0, C > 0 such that | f (x, u)|  |u| + C |u| p1 , (x, u) R N R.
(22)
Lemma 3. (i) There exists 0 > 0 such that inf  > 0 for (0, 0 ), where
 := {u X + : u = }.
(ii) Let Z be a closed cone contained in a finite-dimensional subspace W of X and
such that
{x : w(x) = 0} has positive measure for
every w Z \ {0} with w+   w .

(23)

Then there exists = (Z ) > 0 such that (u)  0 for all u Z satisfying
u  .
Here and in the following, a set Z X is called a cone if x Z for every
x Z ,  0. In particular, (ii) applies to Z = W if W is a finite-dimensional
subspace of X .
Proof. (i) For u X + we have (u) = 21 u2 (u) and (u) = o(u2 ) as
u 0 by (22) and Sobolev embeddings. Hence the conclusion follows.
(ii) Suppose by contradiction that a sequence (u n )n Z exists with (u n ) > 0
for all n and u n  as n . Setting wn = uu nn  , we may pass to a
subsequence such that wn w W , since W is finite-dimensional. Since

368

Gilles Evquoz & Tobias Weth

wn Z and wn  = 1 for all n, we have w Z and w = 1. Moreover,

by (20) we have

 1

(u n )
1
w + 2 w 2 ,

0  lim inf
lim wn+ 2 wn 2 =
2
n u n 
2 n
2
and therefore w+   w . Hence, (23) implies that w := {x :
w(x) = 0} has positive measure. Passing to a subsequence, we may also
assume that wn w pointwise almost everywhere in B R , which implies that
|u n (x)| = u n |wn (x)|

as n for almost every x w .

By ( f 3 ) and Fatous Lemma, we therefore deduce that

 
F(x, u n ) 2
1 + 2
(u n )
2
w

w

wn dx
0
n
n
2
u n 
2
u 2n
w
as n . This contradiction proves the claim.




To proceed with the proof of Theorem 1, we shall decompose the nonlinearity

f as follows. We write f = f 1 + f 2 , where f i : R N R R, i = 1, 2 are
defined by

|u|  s0 ,

f (x, u),

u f (x, s ),
0  u  s0 ,
0
f 1 (x, u) = s02

u2

2 f (x, s0 ), s0  u  0,
s0
u
and we put f 2 = f f 1 . Setting Fi (x, u) = 0 f i (x, s) ds, i = 1, 2 for x
R N , u R, we see that
f 2 (x, u) = 0 if x R N \ or |u|  s0 ;
F2 (x, u) = 0 if x R N \ , u R;
f 2 and F2 are bounded on R N R.

(24)

Moreover, f 1 satisfies condition ( f 2 ) and the following stronger version of condition

( f 4 ). For every x R N ,
u 

f 1 (x, u)
is nondecreasing on R \ {0}.
|u|

(25)

We decompose the functional : X R accordingly and write = 1 + 2 ,

with


Fi (x, u(x)) dx =
Fi (x, u(x)) dx, i = 1, 2.
i (u) =
BR

We note that 2 is bounded on X by (24). The following proposition will be a main

technical step in the proof of Theorem 1.

Real Solutions to the Nonlinear Helmholtz Equation

369

Proposition 1. satisfies the Cerami condition in X , that is, every sequence

(u n )n X such that (u n ) c for some c R and (1 + u n ) (u n ) 0
as n has a subsequence which converges in X .
The proof of this proposition is quite long and requires subtle estimates. Parts of
the proof are inspired by [15] and [16], but we need new arguments to deal with the
difficulty that the nonlinearity may vanish on a subset of B R of positive measure.
A key role in the proof is played by the useful inequality


 s
+ 1 u + (1 + s)v + F1 (x, u) F1 (x, [1 + s]u + v)  0
f 1 (x, u) s
2
for x R N , u, v R and s  1,
(26)
which follows from (25). Indeed, as noted in [16], this inequality is a weak version
of [26, Lemma 2.2]. As a consequence of (26) and of the properties of f 2 , we may
derive the following
Lemma 4. For every K > 0 there is a constant C = C(K ) > 0 with the following
property. If u, s, v R are numbers with 1  s  K and |v|  K , then
 s


f (x, u) s
+ 1 u + (1 + s)v + F(x, u) F (x, [1 + s]u + v)
2
 C for all x R N .
Proof. By (24) there exists a constant C1 > 0 (depending on K ) such that

 
s


+ 1 u   C1
| f 2 (x, u)|  C1 , |F2 (x, u)|  C1 and  f 2 (x, u)s
2
for u R, x R N , |s|  K + 1. Consequently, we have


 s
+ 1 u + (1 + s)v + F2 (x, u) F2 (x, [1 + s]u + v)
f 2 (x, u) s
2
 C1 [1 + K (K + 1)] + 2C1
for x R N and u, s, v R with 1  s  K and |v|  K . Since f =
f 1 + f 2 , F = F1 + F2 and (26) holds, the claim follows with C := C1 [1 + K (K +

1)] + 2C1 . 
The next step in the proof of Proposition 1 is the following relative energy
estimate. In the following, a sequence (u n )n in X is called a Cerami sequence for
if (u n ) c for some c R and (1 + u n ) (u n ) 0 as n .

Lemma 5. For every > 0 there exists C = C()

> 0 with the following property.
If (u n )n is a Cerami sequence for , and rn  0, vn X , n N are given with
rn  and vn   for all n N, then
(rn u n + vn )  (u n ) + C + o(1) as n .

370

Gilles Evquoz & Tobias Weth

Proof. We first note that, by a standard bootstrap argument using elliptic regularity
theory, there exists a constant K = K () > such that
for every v X with v  .

v L (B R )  K

We write rn = 1 + sn with 1  sn  1  K and set wn = rn u n + vn =

(1 + sn )u n + vn for n N. Then

1
(wn ) (u n ) = [Bk (wn ) Bk (u n )] +
(F(x, u n ) F(x, wn )) dx
2
BR


1 
(1 + sn )2 1 Bk (u n ) + 2(1 + sn )Bk (u n , vn ) + Bk (vn )
=
2
(F(x, u n ) F(x, wn )) dx
+
BR



s

vn 2
n
+ Bk u n , sn
+ 1 u n + (1 + sn )vn
=
2
 2
(F(x, u n ) F(x, wn )) dx

+
BR



s
 
n
+ 1 u n + (1 + sn )vn +
 Bk u n , sn
(F(x, u n ) F(x, wn )) dx.
2
BR
Since (u n )n is a Cerami sequence, vn   and |sn |  K + 1 for all n, we have


s
n

+ 1 u n + (1 + sn )vn )
Bk (u n , sn
2


 
 s
n
+ 1 u n + (1 + sn )vn dx 
f (x, u n ) sn

N
2
 R  s


n


=  (u n ) sn
+ 1 u n + (1 + sn )vn 
2
 c1  (u n )u n  + c2  (u n )vn  = o(1),
as n with constants c1 , c2 > 0 (depending on K ). Consequently,

(wn ) (u n ) 



 s
n
+ 1 u n + (1 + sn )vn
f (x, u n ) sn
2
BR

+F(x, u n ) F(x, wn ) dx + o(1).


Choosing C = C(K ) as in Lemma 4, we conclude that

(wn ) (u n )  |B R |C + o(1)
Hence the assertion follows with C = |B R |C + 1.

as



n .

Real Solutions to the Nonlinear Helmholtz Equation

371

We may now complete the

Proof of Proposition 1. Let (u n )n be a sequence with the assumed properties. We
first show that (u n )n is bounded in X . Assuming by contradiction that this is false,
we may pass to a subsequencestill denoted by (u n )n such that u n  as
n . Setting wn := uu nn  , we may assume, passing again to a subsequence, that
wn  w weakly in X for some w X, wn w in L q (B R ) for all 1  q < 2 ,
and wn w pointwise almost everywhere on B R as n . Moreover, we have
wn+  w + weakly in X and wn w strongly in X as n , since X
is finite-dimensional. Passing to a further subsequence, we may also assume that
wn+  s  0 as n for some s  w+ . Since

(u n )
1 + 2
2
w
,
o(1) =



w

n
n
u n 2
2
by (20), we find that
w   s.

(27)

Hence, 1 = wn 2 = wn+ 2 + wn 2 s 2 + w 2  2s 2 as n ,
indicating that s > 0. Next, we suppose by contradiction that
w 0 almost everywhere on ,

(28)

which implies that

(twn ) 0

as n

We claim that there exist t > 0 and vn

X ,

for all t > 0.

(29)

n N with

vn   1 and (twn + vn ) > c + C + 1

for

n sufficiently large , (30)

where C = C(1)
is chosen as in Lemma 5 corresponding to = 1. To prove this,
we have to distinguish different cases. We first note that

t2  2
s w 2 + o(1) as n
(twn ) =
2
for every t > 0 by (29). Hence, if w  < s, we can find t > 0 such that
(twn ) > c + C + 1

for

n sufficiently large ,

and therefore (30) follows with vn = 0 for every n.

Next, we consider the remaining case w  = s, and we note that for every
t > 0 we have twn + wn+ tw + w + in L q (B R ) for all 1  q < 2 , and also
pointwise almost everywhere on B R . Hence,
(twn + wn+ ) (tw + w + ) = (w + ) as n
for all t > 0 by (28). Consequently,



1 2
s (t + 1)2 t 2 s 2 (w + ) + o(1)
twn + wn+ =
2 

1
(w + ) + o(1),
= s2 t +
2

(31)

372

Gilles Evquoz & Tobias Weth

so that there exists t > 0 such that



twn + wn+ > c + C + 1 for n sufficiently large.
Again, (30) follows with t + 1 in place of t and vn = wn , since twn + wn+ =
(t + 1)wn wn for every n. Next, fixing t > 0 and vn , n N such that (30) holds,
we write twn + vn = sn u n + vn with sn = utn  for every n, so that 0 < sn  1 for
large n and vn   wn  = 1. By Lemma 5, we therefore have
(twn + vn ) = (sn u n + vn )  (u n ) + C + o(1),
as n , which contradicts (30). The contradiction shows that (28) is false, and
therefore the set w := {x : w(x) = 0} has positive measure. Moreover,
|u n (x)| = u n  |wn (x)| + as n

for almost every x w .

Hence, Fatous Lemma, the L 2 -convergence wn w in B R and ( f 3 ) imply


F(x, u n )
1
(u n )

|wn |2 dx ,
o(1) =

2
u n 
2
|u n |2
w
as n , a contradiction. The contradiction shows that (u n )n is bounded in X .
Therefore, we can find a subsequencestill denoted by (u n )n and some u X
such that u n  u (weakly) in X, u n u in L q (B R ) for all 1  q < 2 and
u n (x) u(x) for almost every x B R . As a consequence of (22), there holds

( f (x, u n ) f (x, u))(u n u) dx 0 as n .
BR

Hence,

B0 (u n u, u n u) = ( (u n ) (u))(u n u) + k
(u n u)2 dx
BR

( f (x, u n ) f (x, u))(u n u) dx 0,
+
2

BR

as n . It follows from Corollary 1 that u n u strongly in X . The proof is

finished. 

We may now complete the
Proof of Theorem 1 (Case N  2). (i) The existence of a nontrivial solution follows from a variant of the classical linking theorem, where the PalaisSmale
condition is replaced by the Cerami condition [4, Theorem 2.3]. To see this,
we proceed as follows. Considering the sequence of eigenfunctions (e j ) jN of
(17) given by Lemma 2, we set u = e j X + and put
Q := {tu + v : v X , v  , 0  t  } for > 0.
Note that the sets Q are contained in the finite dimensional subspace W =
X Ru X . Since every function in W is analytic in B R , Z := W satisfies

Real Solutions to the Nonlinear Helmholtz Equation

373

condition (23). Using Lemma 3 (ii) and the fact that is nonpositive on X
by (20), we thus find that
sup = 0 for > 0 sufficiently large .

According to Lemma 3 (i), we may further choose > 0 sufficiently small such
that the sets  and Q link and inf  > 0 (see for example [24] or [4,
Section 2] for the notion of linking of sets). Finally, we have sup Q < +
by the compactness of Q . Taking Proposition 1 into account, we can apply the
linking theorem and obtain that has a nontrivial critical point u X , such
that
0 < inf  (u)
 sup .


(ii) Let us now assume that f (x, t) = f (x, t) for all (x, t) R N R. In
this case, a variant of the Fountain Theorem (see [5,7] and [29, Theorem 3.6])
yields the existence of a sequence of pairs {u n }, n N of critical points of
, such that
(u n ) as n .

(32)

More precisely, we use a version of the Fountain Theorem where the Cerami
condition is used in place of the PalaisSmale condition. To see that such a
variant exists, it suffices to note the validity of a deformation lemma giving rise
to Cerami sequences instead of PalaisSmale sequences. Such a deformation
lemma has already been established in [4, Theorem 1.3]. In order tocheck the
other assumptions of the Fountain Theorem, we remark that X =
jN Re j
where (e j ) jN is given by Lemma 2 (i). We set
X j = Re j , Y j =

j


X  and Z j =

=1

X

= j

for j N. Since every function in Y j is analytic, we see from Lemma 3

(ii), applied to Z = Y j , that for every j N there exists j > 0 such that
(u)  0 for u Y j with u  j . It only remains to check that for some
sequence (r j ) j (0, )
inf{(u) : u Z j , u = r j }

as

j .

(33)

This will be proved by similar arguments as in [29, Theorem 3.7]. Indeed, if

j  j , then Z j X + and therefore, by (22),

1

C
1
p
F(x, u) dx  u2 u2L 2 (B )
(u) = u2
u L p (B R )
R
2
2
2
p
BR
1
C p
2
u p for u Z j ,
 u
(34)
4
p j

374

Gilles Evquoz & Tobias Weth

where
=

u2
1
inf
2 uX \{0} u2 2

> 0 and j := sup{u L p (B R ) : u Z j , u = 1}.

L (B R )

Since ( j ) j [0, ) is a decreasing sequence, := lim j j exists.

Moreover, for each j we can find some u j Z j such that u j  = 1 and

j  u j  L p (B R ) > 2j . From the definition of Z j we obtain that u j  0

(weakly) in X and the compact Sobolev embedding X  L p (B R ) then gives
1

u j 0 in L p (B R ). Thus, j 0 as j . Choosing r j = (2C j ) 2 p ,

we obtain from (34) that


1
1
for all u Z j with u = r j .
(u)  r 2j

4 2p
Since r j as j , the assertion follows. Moreover, since is bounded
on bounded subsets of X = H 1 (B R ), (32) implies that u n  H 1 (B R ) as
n , as claimed in (8).
To conclude the proof of Theorem 1, we remark that if u is a critical point of ,
1
then the restriction of u to S R belongs to H 2 (S R ) and there exists a unique weak
solution w of (10). Therefore, extending u on R N by setting u Re(w) on E R ,

we see that u R R is a solution of (4). 
We close this section with an observation on a rigid minimax characterization
of the ground state energy level (that is, the least energy of a nontrivial critical point
of ) in the case where ( f 4 ) is replaced by a stronger condition.
Theorem 3. Suppose that the nonlinearity f satisfies ( f 1 )( f 3 ) and the following
stronger version of ( f 4 ): For every x R N ,
u 

f (x, u)
is nondecreasing on R \ {0}.
|u|

(35)

Then the ground state energy level

c := inf{(u) : u nontrivial critical point of }
is positive and equivalently given by
c=

inf

sup

uX + \{0} t 0,vX

(tu + v).

Moreover, c is attained within the set of nontrivial critical points of , that is,
within the set of nontrivial solutions of (21).
Proof. Let u be a nontrivial critical point of . Then u  X , since otherwise

f (x, u)u dx = Bk (u) = u 2 < 0,
BR

Real Solutions to the Nonlinear Helmholtz Equation

375

contrary to (35) and ( f 2 ). Next we claim that

(u)  (w)

for every

w X + R+ u = X + R+ u + .

(36)

Indeed, similar to the proof of Lemma 5 we have, for w = (1+s)u +v with s  1

and v X ,

 

s
+ 1 u + (1 + s)v +
(w) (u)  Bk u, s
(F(x, u) F(x, w)) dx
2
BR
 
 s


f (x, u) s( +1)u +(1 + s)v + F(x, u) F(x, w) dx.
=
2
BR
Moreover, the last integral is nonpositive since, as a consequence of (35), the
inequality (26) holds for f in place of f 1 . Hence, (36) is true. Putting
c =

inf

sup

uX + \{0} t 0,vX

(tu + v),

we thus infer that c  c , whereas Lemma 3 (i) implies that c  inf  > 0 for
some > 0, where  := {u X + : u = }.
Furthermore, c is attained among nontrivial critical points of . Indeed, if (u n )n
is a sequence of critical points of with (u n ) c, then by Proposition 1 we
have u n u 0 in X for a subsequence, where u 0 is a critical point of with
(u 0 ) = c > 0, and therefore u 0 = 0.
It remains to show that c  c . For this, let u X + \ {0} be such that
sup
t 0,vX

(tu + v) < .

(37)

Let Z := {tu + v : v X , t  0}. We claim that condition (23) holds for this set
Z . Suppose by contradiction that this is false, that is, there exists w Z \ {0} with
w +   w  and such that w 0 almost everywhere on . Since w+  > 0,
we then find

1
(t + 1)2 w + 2 t 2 w 2 (w + )
(tw + w + ) =
2


1
w + 2 (w + ) ,
 t+
2
as t . This contradicts (37), since tw + w + = (t + 1)w + + tw Z for
every t > 0. The contradiction shows that condition (23) holds for Z . Since Z
is a closed cone contained in the finite dimensional space X Ru, Lemma 3 (ii)
implies that there exists > 0 with
sup = 0

where

Q := {tu + v : v X , v  , 0  t  }. (38)

Applying the linking theorem [4, Theorem 2.3] exactly as in the proof of Theorem 1
with the present choice of Q , we obtain a nontrivial critical point u of with
(u)
 max 
Q

sup
t 0,vX

(tu + v).

We thus conclude that c  c , so equality holds and the proof is finished.




376

Gilles Evquoz & Tobias Weth

4. The One-Dimensional Case

In this section we sketch the proof of Theorem 1 in the case N = 1, which must
be treated slightly differently. Note that for R > 0 we have B R = (R, R), S R :=
{R, R} and E R = (, R) (R, +), and consider the space H RC as defined
in (9). For a given function u : S R C, it is easy to see that the unique solution
w H RC of the problem

w + k 2 w = 0
w=u

in E R ,
on S R

(39)

is then given by

w(x) =

u(R)eik(x+R)
u(R)eik(xR)

for
for

x  R,
x  R.

Hence, the capacity operator TR is simply given by [TR u](R) = iku(R) and
[TR u](R) = iku(R) for any function u : S R C, and thus its real part
K R , as defined in (12), is zero. In particular, the second inequality in (13) is not
true if N = 1, and also Corollary 1 does not hold in this case. On the other hand,
the quadratic form Bk defined in (16) is simply given by

Bk (u, v) =

R
R

(u v k 2 uv) dx, u, v H 1 (R, R),

and (17) reduces to the homogeneous Neumann eigenvalue problem

u = u

in (R, R), u (R) = 0.

(40)

As a consequence, the properties listed in Lemma 2 are well known and easy to
check in the case N = 1. Furthermore, the inner problem (21) reduces to

u k 2 u = f (x, u)
u (R) = 0,

in (R, R),

(41)

and solutions of (41) correspond to critical points of the functional

: H 1 (R, R) R, (u) =

1
2




(u )2 k 2 u 2 dx

F(x, u) dx.

Since the structure of is the same as in the multidimensional case, the proof
of Theorem 1 can be finished exactly as detailed in Section 3. In particular, the
one-dimensional analogues of Lemma 3 and Proposition 1 are valid.
In addition, we remark that, by exactly the same proof, Theorem 3 is also true
in the one-dimensional case.

Real Solutions to the Nonlinear Helmholtz Equation

377

5. Existence of Radial Solutions

For N  2, we look for radially symmetric solutions of
u k 2 u = f (|x|, u), x R N ,

(42)

where k > 0, and f satisfies the following assumptions:

(g1 ) f C ([0, ) R) with f (, 0) = 0, and f is locally Lipschitz continuous
with respect to u R.
(g2 ) 0  F(r, u)  21 f (r, u)u for all (r, u) [0, ) R, where F(r, u) =
u
0 f (r, s) ds.
(g3 ) For every u R, F(, u) C 1 (0, ) with r F(r, u)  0 for all r > 0.
Setting u(x) = u(r ) with r = |x|, we can rewrite (42) as
u

N 1
u k 2 u = f (r, u), r > 0,
r

(43)

and we shall look for solutions of class C 2 of this equation which satisfy u (0) = 0.
The following local existence and uniqueness result (in a singular setting) is
well known, see, for example, [28, Lemma 3.1] for a proof.
Lemma 6. For every R, there exists > 0 such that the initial value problem associated with (43), u(0) = and u (0) = 0 has a unique solution u
C 2 ([0, ), R).
Using properties (g2 ) and (g3 ), we now continue the local solution given by
Lemma 6 to a global solution on R+ and analyze its asymptotic behavior.
Theorem 4. Let f satisfy (g1 ) to (g3 ). Then for every R, there exists a unique
radially symmetric solution of (42), u = u(|x|), u C 2 ([0, ), R) satisfying
u(0) = and u (0) = 0. Moreover,



(44)
sup r N 1 u (r )2 + u(r )2 < ,
r 1

so thatidentifying u(x) with u(|x|) for x R N we have

u
,
1+|x|2

|u|

1+|x|2

L 2 (R N ). If, in addition,
(g4 ) f (r, u) = o(|u|) as u 0, uniformly in r > 0,
then u satisfies the asymptotic condition (5).
Proof. Let > 0. By Lemma 6, there exists > 0 and a unique solution u
C 2 ([0, 2), R) satisfying (43) and u(0) = , u (0) = 0. Let [0, r0 ), r0 [2, ]
denote the maximal existence interval of this solution. We claim that


(45)
sup r N 1 u (r )2 + u(r )2 < ,
r [,r0 )

378

Gilles Evquoz & Tobias Weth

To see this, we use the change of variables

v(r ) = r

N 1
2

u(r ), r (0, r0 ),

so that v solves



N 1
N 1
(N 1)(N 3)
v = r 2 f r, r 2 v , r (0, r0 ). (46)
v k 2
2
4r
We consider the C 1 -function r  (r ) := v (r )2 + k 2 v(r )2 on (0, r0 ) which
satisfies
N 1
N 1
1
(N 1)(N 3)
v(r )v (r ) r 2 f (r, r 2 v(r ))v (r ).
(r ) =
2
2
4r

For r [, r0 ), we thus obtain

 r
(N 1)(N 3)
(r ) = ()+2
v(s)v (s) ds
2
4s
 r


N 1
N 1
2
s 2 f s, s 2 v(s) v (s) ds

 r
r
(N 1)(N 3)


N 1
N 21
v(s)v
(s)
ds
2s
F(s,
s
v(s))
= ()+2

2

4s


 



r
N 1
N 1
N 1
(N 1)
s N 2 f s, s 2 v(s) s 2 v(s)2F s, s 2 v(s) ds

 r
N 1
N 1
+2
s
r F(s, s 2 v(s)) ds

 r
1N
(N 1)(N 3)
N 1
2
 () + 2
F(,
v()) + 2
v(s)v (s) ds
4s 2


  r (N 1)|N 3|
1N
 () + 2 N 1 F , 2 v() +
(s) ds,
4ks 2

using (g2 ) and (g3 ). Hence, Gronwalls inequality (see [12, Theorem III.1.1]) gives
(r )  [() + 2 N 1 F(,
From this we derive (45), since


r N 1 u (r )2 + u(r )2  C (r )

1N
2

for

v())]e

(N 1)|N 3|
4k2

for

r [, r0 ).

r [, r0 ) with some constant C > 0.

As a consequence of (45), we find that r0 = and that (44) holds.

We finally assume that (g4 ) holds, and we check that the radiation condition in
(5) is fulfilled. For this we let r  1 and write
r

N 1
2





(N 1)




f r, r 2 v(r )


 (N 1) (N 1)

2


u (r ) + k u(r ) = 
v(r
)
v(r
)

v
(r
)



(N 21)
r
2r


r
v(r )

 

 f r, r (N 21) v(r ) 


(N 1)
(N 1)2

 v .
v

v
+

+



(N 21)
2r 2
r


r
v(r )

Real Solutions to the Nonlinear Helmholtz Equation

(N 1)

Since lim r r 2 v(r ) = 0, the assumption (g4 ) gives limr r


+k 2 u(r ) = 0, that is, condition (5) holds. 

379
N 1
2


u (r )

We point out that if f satisfies (g1 )(g4 ), every solution of (42) given by
Theorem 4 is oscillatory. This follows from the fact that for every such solution
u, 0 lies in the essential spectrum of any self-adjoint realization of the Sturm
Liouville differential expression


(N 1)(N 3)
f (r, u(r ))
w
w = w k 2
+
r2
u(r )
associated with (46) (see [10, Corollary XIII.7.14 and Theorem XIII.7.40]).

6. Properties of the Capacity Operator and Its Real Part

In this section we derive key properties of the operator TR introduced in Section 2 and its real part K R in the case N  2 (see Section 4 for the case N = 1).
In particular, we will give the proofs of Lemma 1 and Lemma 2 (iv). Part of the
material presented in this section should be well known to experts and appears in
the standard literature in the special case N = 3 (see, for example, [18]). However,
we could not find the corresponding formulas for the general N -dimensional case.
Moreover, as already remarked in the introduction, it seems that the real part of the
capacity operator has not been studied in the degree of detail that we need for our
purposes.
Let N  2, and consider the exterior Dirichlet problem for the Helmholtz
equation

w + k 2 w = 0 in E R := R N \B R ,
(47)
w=u
on S R ,
where k > 0, R > 0 and u H 2 (S R , C) are given. Recall the space H RC defined
in (9) which enforces a weak form of Sommerfelds radiation condition. We wish to
construct the unique solution w H RC explicitly in terms of a Fourier representation
of the boundary datum u. For this we let  S denote the LaplaceBeltrami operator
on the unit sphere S1 R N , so that
 N 1 w 

1
w = r N11 r
r
(48)
r + r 2  S w.
1

We recall (see, for example, [23, Corollary 2.3]) that the linear span of the spherical
harmonics (that is, restrictions to S1 of harmonic polynomials on R N with complex
coefficients) is dense in L 2 (S1 , C). More precisely, denoting by HN the space of
+22)  N +2
spherical harmonics of degree  N0 , then dN := dim HN = (N

(N
+2)

N is a dense subspace of
if   1, d0N := dim H0N = 1, and span
H
=0

L 2 (S1 , C). Furthermore, according to (48), every nonzero Y HN is an eigenfunction of  S :
 S Y + (N +  2)Y = 0 in

S1 ,

(49)

380

Gilles Evquoz & Tobias Weth

see [11, Theorem 3.2.11]. Moreover, starting from the orthogonal bases {1} of H02
and {ei , ei } of H2 ,   1, ( = (x1 , x2 ) = sgn(x2 ) arccos x1 ) we can
inductively construct, according to [11, Lemma 3.5.3], orthogonal bases of HN
{Y1 , . . . , Y N },  N0 for all N  2, with the property that, for each element Y
d

in such a basis, the element Y also belongs to this basis.


dN
1
Proposition 2. For u H 2 (S R ) given by the expansion u(R ) =
=0
m=1
u m Ym ( ), S1 , the problem (47) has a unique solution w H RC given by
N

d

(1)
 r  N 2 
H (kr ) 
2
u m Ym ( ) for r  R,
w(x) =
(1)
R
H (k R)

S1

=0

m=1

and x = r ,

(50)
(1)

(2)

where, here and in the following,  =  + (N 22) for  N0 and H , H are

the two Hankel functions (or Bessel functions of the third kind) of order (see for
example [27, 3.6]).
Proof. We start by rescaling the space variable x to Rx . Setting w(x) = v( Rx ), the
problem (47) becomes

v + (k R)2 v = 0 in E 1 ,
(51)
on S1 ,
v = uR
where u R ( ) = u(R ), S1 . We first remark that w H RC if and only if v
belongs to the space

u
C
1
H1,k R := u Hloc
(E 1 , C) :
L 2 (E 1 , C),
(1 + r 2 )1/2

u
2
N u
2
ik Ru L (E 1 , C) . (52)
L (E 1 , C ),
(1 + r 2 )1/2
r
 (x) = f (r )Y  ( )
We will therefore seek solutions in this space of the form vm

m
N
for some  N0 , 1  m  dl . Using (48) and (49), we may rewrite (51) for
functions of this form to obtain




(N 1)
(N +  2)
f
f  (r ) + (k R)2
0 = f  (r ) +
(r
)
Ym ( ).

r
r2

Setting g (s) = ( ksR )

N 2
2

f  ( ksR ), we find


2 
 + N 22
1

g (s) + g (s) + 1
g (s) = 0, s > 0,
s
s2

which is Bessels equation with parameter =  +

given by

N 2
2 .

g(s) = AH(1) (s) + B H(2) (s),

Its general solution is

Real Solutions to the Nonlinear Helmholtz Equation

(1)

381

(2)

where A, B C, and H , H are the two Hankel functions of order . Now,

we observe that the Sommerfeld radiation condition, given as the third condition
C in (52), can be rewritten in the form
in the definition of the space H1,k
R



s 1 2 As[(H(1) ) (s) i H(1) (s)] A(N 2)H(1) (s)
kR

2

+2Bs[(H(2) ) (s) i H(2) (s)] B(N 2)H(2) (s) ds < .

( p)
Using the recurrence formula (H( p) ) (s) = H( p) (s) H+1 (s), p = 1, 2, we
s
obtain



(1)
s 1 2 As[H+1 (s) + i H(1) (s)] + A(N 2 2)H(1) (s)
kR

2

(2)
+2Bs[H+1 (s) + i H(2) (s)] + B(N 2 2)H(2) (s) ds < .

(53)
( p)

According to [27, Formulas 7.2 (1) and (2)], the asymptotic behavior of H (s),
p = 1, 2, is given by




2 i s 2+1
(1)
4
e
1 + O(s 1 ) ,
H (s) =
s




2 i s 2+1
(2)
4
H (s) =
1 + O(s 1 ) .
e
s
As a consequence,


1
2
|H( p) (s)| = s 2
+ O(s 1 )
for p = 1, 2,

(1)

s|H+1 (s) + i H(1) (s)| = O(s 2 ) and

 
 
1
2
(2)
(2)
+ O s 1 .
s|H+1 (s) + i H (s)| = s 2 2

C of
Thus, (53) can be satisfied only if B = 0. We conclude that a function in H1,k
R
 (x) = f (r )Y  ( ) is a solution of the differential equation in (51) if
the form vm

m
and only if it can be written as

vm
(x) = Ar

(N 2)
2

H(1)
(k Rr ) Ym ( ),


dN  

for some A C with  =  + N 22 . Since u R ( ) =
=0
m=1 u m Ym ( ), a
solution of the boundary value problem (51) is thus given by
N

v(x) =r

(N 22)

d



H(1)
 (k Rr )
=0

(1)

H (k R)

u m Ym ( )

for

r  1, S1

and x =r .

m=1

Rescaling back, we thus obtain the formula (50) for the (unique) solution w H RC
of (47). 


382

Gilles Evquoz & Tobias Weth

The formula (50) gives rise to the following expression for the capacity operator
1
1
TR : H 2 (S R , C) H 2 (S R , C) (see [18] for the case N = 3):
N

d


1 
[TR u](R ) =
z  (k R)
u m Ym ( ), S1
R
=0

for u H 2 (S R , C) given by u(R ) =

 dN
=0

(1)

z  (r ) =

(54)

m=1

r drd H (r )
(1)
H (r )



m=1 u m Ym ( ),

S1 , where

N 2
, r > 0.
2

We need some estimates for the coefficients z  (r ). For this we recall that
H(1) = J + iY , and H(2) = J iY = H(1)

for

R,

where J and Y denote the Bessel functions of the first and second kind. Setting
G (r ) =

(1)

d
r dr
H (r )
(1)
H (r )

, for r > 0, R, we see that



(1)
(1)
d
H (r )
r Re H (r ) dr



(1)
d
r drd |H (r )|2 r dr J2 (r )+Y2 (r )
=
(1)
2 |H(1) (r )|2
2 J2 (r )+Y2 (r )
|H (r )|2


(1)
(1)
r Im H (r ) drd H (r )
r (J (r )Y (r ) J (r )Y (r ))
Im G (r ) =
=
.
(1)
(1)
|H (r )|2
|H (r )|2

Re G (r ) =

(55)

(56)

We need the following estimates (see [18] for the case N = 3):
"
!
we have
Lemma 7. For r > 0 and   max 0, 3N
2
(N 1)
 Re z  (r )   + N 2
2
0 < Im z  (r )  r.
In the case N = 2, we also have 0 < Re z 0 (r ) 
r > 0.

1
2

(ii)

(58)

and Im z 0 (r ) > 0 for all

Proof. We first prove (57). According to [27, 13.74], we have

(i)

(57)

d 2
(J (r ) + Y2 (r )) < 0 for r > 0, R,
dr

 0 if  21
d
2
2
r (J (r ) + Y (r ))
for r > 0.

dr
 0 if  21 ,

Real Solutions to the Nonlinear Helmholtz Equation

383

From (i) we obtain Re G (r ) < 0 for all r > 0, R, and (ii) gives Re G (r ) 
21 for r > 0,  21 . Moreover, (ii) implies ReG 1 (r ) = 21 for r > 0. Now,
2

(1)
H(1) (r ) = H1
(r ) r H(1) (r ), we find that


d
|H(1) (r )|2 (Re G (r ) + ) = Re r H(1) (r ) H(1) (r ) + |H(1) (r )|2
dr




(1)
(1)
(1)
= r Re H (r )H1 (r ) = r Re H1 (r )H(1) (r )

using the recurrence formula

d
dr

(1)

= |H1 (r )|2 (Re G 1 (r ) ( 1)).

Since Re G (r ) < 0 holds for all , the previous formula gives Re G (r ) 
for all  1. Summarizing, we have shown
1
 Re G (r )  , r > 0
2

for

This shows (57), since z  (r ) = G + N 2 (r )

2

21

N 2
2 .

!1"
2

[1, ).

In the case = 0, we have

 Re G 0 (r ) < 0 for all r > 0. Hence, when N = 2, we obtain 21  z 0 (r ) < 0

for all r > 0.
We now turn to the proof of (58). Using (56) and the fact that
W (J , Y ; r ) := J (r )Y (r ) J (r )Y (r ) =

2
r

for

R, r > 0,

(59)

(see [27, 3.63 (1)]), we find that

Im G (r ) =

r W (J , Y ; r )
(1)
|H (r )|2

>0

(1)
|H (r )|2

for

R, r > 0.

Furthermore, using (ii) above, we see that

r |H(1) (r )|2  lim t|H(1) (t)|2 =
t

for

r > 0, 

1
.
2

We thus obtain Im G (r )  r for r > 0,  21 . As before, (58) follows from the

identity z  (r ) = G + N 2 (r ) N 22 . 

2

Recall that in Section 2 we have also introduced the operator

1

K R : H 2 (S R ) H 2 (S R ),
1

K R u = Re[TR u],
1

(60)
1

where H 2 (S R ) = H 2 (S R , R), H 2 (S R ) = H 2 (S R , R). For u H 2 (S R ) given

dN  

by u(R ) =
=0
m=1 u m Ym ( ), we then have
N

d


1 
Re(z  (k R))
u m Ym ,
K Ru =
R
=0

(61)

m=1

dN  
since m=1
u m Ym is a real-valued function for every  N0 . Using Lemma 7 we
may therefore easily complete the

384

Gilles Evquoz & Tobias Weth

Proof of Lemma 1. We first note that, for s R, one may define an equivalent
norm on H s (S R , C) by
N

u2s

d



=
(1 + 2 )s
|u m |2
=0

for u H 2 (S R , C) given by u(R ) =

1
2

m=1

 dN
=0



m=1 u m Ym ( ). Hence, by (57) and
1
H 2 (S R , C) is bounded, and thus

(58), the capacity operator TR : H (S R , C)

K R defined in (60) is bounded as well. In particular, the integrals on both sides of
(14) exist, and by (61) and the orthogonality of spherical harmonics (corresponding
to different values of ), they coincide. Finally, the second inequality in (13) also
follows, by orthogonality, from (61) and the estimates in Lemma 7. 

Next, we study the eigenspace X 0 of the eigenvalue problem (17) corresponding
to the eigenvalue k 2 , and we complete the
Proof of Lemma 2(iv). Recall that u X 0 if and only if u solves

u + k 2 u = 0 in B R ,
u
=v
on S R

(62)

with v = K R u on SR . Put, as before,  =  + N 22 for  N0 . In the case where

 2N


d
2 J (kr ) 
= 0 for all  N0 , the inhomogeneous Neumann problem

dr r
r =R

(62) has, for given v H 2 (S R ), the unique solution

u(r ) =

 r  N 2 
2

J (kr )

d



vm
Ym ( ),
(N 2)

=0 k J (k R) 2R J (k R) m=1

 are determined by v(R ) =

where the coefficients vm
particular, the restriction of u to S R satisfies
N

u(R ) =

d


=0 m=1

u m Ym ( ) with u m =

0 < r < R,

 dN
=0



m=1 vm Ym ( ).

J (k R)
k J  (k R)


.
vm
(N 2)
J
(k
R)

2R

If, in addition, v = K R u on S R , then by (61) we have


vm



(1)
k R drd H (k R)
N 2
1

= Re
u m
(1)
R
2
H (k R)


J  (k R)J (k R) + Y  (k R)Y (k R)
N 2
=k

u m ,
J2 (k R) + Y2 (k R)
2k R

In

Real Solutions to the Nonlinear Helmholtz Equation

385

and therefore
u m =

J (k R)

J (k R) (N2k2)
R J (k R)
 

J (k R)J (k R) + Y  (k R)Y (k R)
J2 (k R) + Y2 (k R)


N 2

u m
2k R

for  N0 , 1  m  dN . As a consequence of (59), this gives for each  N0 the

alternative
u m = 0 for all 1  m  dN

or

Y (k R) = 0.

We may now finish the proof by setting


d  2N

2
r
J (kr ) 
= 0 or Y (k R) = 0 for some  N0 .
D := R > 0 :
r =R
dr

Indeed, since J and Y are analytic functions on (0, ) for every  N0 , the set
D is countable. Moreover, for R (0, ) \ D and u X 0 we conclude u m = 0

for all  N0 , 1  m  dN and therefore u 0. 
We close this section with the
Proof of Theorem 2. Let R > 0 be fixed, and let u R R with u| S R  0. Furthermore, let R > R > 0 be given such that u R R . Then we have:
(i) In E R , u coincides with the real part of a solution w1 of the linear Helmholtz
equation satisfying (7);
(ii) In E R , u coincides with the real part of a solution w2 of the linear Helmholtz
equation satisfying (7);
(iii) u| S R w1 | S R and u| S R w2 | S R .
Since w := w1 w2 also satisfies the linear Helmholtz equation in E R , together
1N
with (7) and Re(w) 0 in E R , it follows that w(x) = o(r 2 ) as r = |x| ,
and hence w 0 in E R by a classical result of Rellich [20, Satz 1]. Hence,
w1 = w2 in E R , and by (iii) this implies that Im(w1 | S R ) Im(w2 | S R ) 0.
Writing, again,
u(R ) = w1 (R ) =

u  ( ), S1 ,

=0

with u  ( ) =

dN



m=1 u m Ym ( ),

w1 (R ) =

R
R

we have, by Proposition 2,

 (N 22) 

(1)
H (k R )
(1)
=0 H (k R)

u  ( ), S1 ,

(63)

386

Gilles Evquoz & Tobias Weth

where the functions  u  ( ),  N0 are real-valued as a consequence of (63)

and the fact that u is real-valued. Moreover, the assumption u| S R  0 forces that
u   0 on S1 for at least one  N0 , and therefore

Im

(1)

H (k R )
H(1)
 (k R)


= 0, that is, J (k R )Y (k R) = J (k R)Y (k R ).

Since the functions J and Y are real analytic, the latter can only happen for
countably many R > R. We thus conclude that u R R for, at most, countably
many R > R.
It remains to be shown that problem (4), (5) admits infinitely many solutions
if the nonlinearity f satisfies ( f 0)( f 4). For this, we fix R1 (0, ) \ D such
that B R1 ; then Theorem 1 yields a nontrivial solution u 1 R R1 of (4), (5).
The nontriviality of u implies that u 1 | S R1  0, since otherwise u 0 on E R1 ,
and therefore on all of R N , as a consequence of unique continuation. Hence, there
exists R2 (R1 , ) \ D such that u 1  R R2 , whereas Theorem 1 yields a solution
u 2 R R2 of (4), (5). Moreover, there exists R3 (R2 , ) \ D such that u 1 , u 2 
R R2 , whereas Theorem 1 yields a solution u 3 R R3 of (4), (5). Inductively, we
now obtain an infinite sequence of pairwise different solutions of (4), (5). 

Acknowledgments. The authors wish to thank the referee for his/her very valuable and
detailed remarks and suggestions.

References
1. Alves, C.O., Souto, M.A.S., Montenegro, M.: Existence of solution for two classes
of elliptic problems in R N with zero mass. J. Differ. Equ. 252, 57355750 (2012)
2. Ambrosetti, A., Felli, V., Malchiodi, A.: Ground states of nonlinear Schrdinger
equations with potentials vanishing at infinity. J. Eur. Math. Soc. (JEMS) 7, 117144
(2005)
3. Ambrosetti, A., Malchiodi, A.: Perturbation Methods and Semilinear Elliptic Problems on Rn . Progress in Mathematics, Vol. 240. Birkhuser, Basel, 2006
4. Bartolo, P., Benci, V., Fortunato, D.: Abstract critical point theorems and applications to some nonlinear problems with strong resonance at infinity. Nonlinear Anal.
7, 9811012 (1983)
5. Bartsch, T.: Infinitely many solutions of a symmetric Dirichlet problem. Nonlinear
Anal. 20, 12051216 (1993)
6. Bartsch, T., Ding, Y.: On a nonlinear Schrdinger equation with periodic potential.
Math. Ann. 313, 1537 (1999)
7. Bartsch, T., Willem, M.: Infinitely many nonradial solutions of a Euclidean scalar
field equation. J. Funct. Anal. 177, 447460 (1993)
8. Cerami, G.: Some nonlinear elliptic problems in unbounded domains. Milan J. Math.
74, 4777 (2006)
9. Costa, D.G., Tehrani, H.: Existence of positive solutions for a class of indefinite
elliptic problems in R N . Calc. Var. Partial Differ. Equ. 13, 159189 (2001)
10. Dunford, N., Schwartz, J.T.: Linear Operators. Part II: Spectral Theory. Wiley, New
York, 1963

Real Solutions to the Nonlinear Helmholtz Equation

387

11. Groemer, H.: Geometric Applications of Fourier Series and Spherical Harmonics.
Encyclopedia of Mathematics and its Applications, Vol. 61. Cambridge University Press,
Cambridge, 1996
12. Hartman, P.: Ordinary Differential Equations, 2nd edn. Classics in Applied Mathematics, Vol. 38. Society for Industrial and Applied Mathematics (SIAM), Philadelphia,
2002
13. Jrgens, K.: Das Anfangswertproblem im Groen fr eine Klasse nichtlinearer Wellengleichungen. Math. Z. 77, 295308 (1961)
14. Kuzin, I., Pohozaev, S.: Entire Solutions of Semilinear Elliptic Equations. Progress
in Nonlinear Differential Equations and their Applications, Vol. 33. Birkhuser, Basel,
1997
15. Li, G., Wang, C.: The existence of a nontrivial solution to a nonlinear elliptic problem
of linking type without the Ambrosetti-Rabinowitz condition. Ann. Acad. Sci. Fenn.
Math. 36, 461480 (2011)
16. Liu, S.: On superlinear Schrdinger equations with periodic potential. Calc. Var. Partial
Differ. Equ. 45, 19 (2012)
17. Moroz, V., Van Schaftingen, J.: Semiclassical stationary states for nonlinear
Schrdinger equations with fast decaying potentials. Calc. Var. Partial Differ. Equ. 37,
127 (2010)
18. Ndlec, J.C.: Acoustic and Electromagnetic Equations. Integral Representations for
Harmonic Problems. Applied Mathematical Sciences, Vol. 144. Springer, New-York,
2001
19. Rabinowitz, P.H.: A Note on A Semilinear Elliptic Equation on Rn , pp. 307317.
Quaderni. Scuola Norm. Sup. Pisa, 1991
20. Rellich, F.: ber das asymptotische Verhalten der Lsungen von u + u = 0 in
unendlichen Gebieten. Jber. Deutsch. Math. Verein. 53, 5765 (1943)
21. Schneider, M.: Existence and nonexistence of positive solutions of indefinite elliptic
problems in R N . Adv. Nonlinear Stud. 3, 231259 (2003)
22. Sommerfeld, A.: Die Greensche Funktion der Schwingungsgleichung. Jber. Deutsch.
Math. Verein. 21, 309353 (1912)
23. Stein, E.M., Weiss, G.: Introduction to Fourier Analysis on Euclidean Spaces. Princeton
Mathematical Series, Vol. 32. Princeton University Press, Princeton, 1971
24. Struwe, M.: Variational methods: Applications to Nonlinear Partial Differential Equations and Hamiltonian Systems, 4th edn. Springer, Berlin, 2008
25. Stuart, C.A.: Bifurcation from the essential spectrum. Topological Nonlinear Analysis,
II (Frascati, 1995). Progress in Nonlinear Differential Equations and their Applications,
Vol. 27, pp. 397443. Birkhuser, Boston, 1997
26. Szulkin, A., Weth, T.: Ground-state solutions for some indefinite variational problems.
J. Funct. Anal. 257, 38023822 (2009)
27. Watson, G.N.: A Treatise on the Theory of Bessel Functions. Reprint of the second
(1944) edition. Cambridge Mathematical Library, Cambridge University Press, Cambridge, 1996
28. Weth, T.: Global bifurcation branches for radially symmetric Schrdinger equations.
Adv. Differ. Equ. 10, 721746 (2005)
29. Willem, M.: Minimax Theorems. Progress in Nonlinear Differential Equations and
Their Application, Vol. 24. Birkhuser, Basel, 1996
30. Willem, M., Zou, W.: On a Schrdinger equation with periodic potential and spectrum
point zero. Indiana Univ. Math. J. 52, 109132 (2003)

388

Gilles Evquoz & Tobias Weth

Institut fr Mathematik,
Johann Wolfgang Goethe-Universitt,
Robert-Mayer-Str. 10,
60054 Frankfurt am Main, Germany.
e-mail: evequoz@math.uni-frankfurt.de
and
e-mail: weth@math.uni-frankfurt.de
(Received December 7, 2012 / Accepted June 20, 2013)
Published online September 3, 2013 Springer-Verlag Berlin Heidelberg (2013)