You are on page 1of 17

Branching points for a class of

variational operators 
Antonio Ambrosetti

1 Introduction and Main Results


Let E be a Hilbert space and consider the equation
Lu + H (u) = u; u 2 E; (1)
where L : E ! E is linear and H 2 C 1(E; E ) is such that H (0) = 0,
H 0(0) = 0. Let  denote the closure of the set of non-trivial solutions
(; u) 2 IR  E of (1). A point  2 IR is a bifurcation point (of (1)) if
(; 0) 2 . If  contains a connected set S such that (; 0) 2 S and S n
f(; 0)g 6= ;, we will say that  is a branching point. A rst bifurcation
result goes back essentially to Lyapunov and Schmidt and states that
any simple eigenvalue of L is a branching point: indeed, near (; 0)
 is a C 1 curve. For an introduction to bifurcation, see for example
[1] and references therein. When L and H are compact operators, any
eigenvalue of odd multiplicity is a bifurcation point, see [7], Chapter IV,
Theorem 2.1 . This result has been improved, see [11], by showing that
 contains a global continuum which either is unbounded in IR  E ,
or meets another eigenvalue 0 6=  of L of odd multiplicity. Finally,
when L + H is a variational operator, namely when L is symmetric
and H is the gradient of a functional h : E ! IR, it is possible to
show that any eigenvalue  of L with nite multiplicity is a bifurcation
point, see [7] Chapter VI, Theorem 2.2, see also [2, 3, 6, 8, 9, 13]. In
all these latter papers no results concerning branching points are given,
apart from a Theorem of [2] where it is shown that if h is real analytic
then there is an analytic curve bifurcating from (; 0). Furthermore, a
 Supported by M.U.R.S.T. and by E.E.C. contract n. ERBCHRXCT940494.

1
counterexample is given where h is not real analytic and  contains
only isolated points.
In the present paper we deal with the variational case and prove
that for a large class of operators, possibly not analytic, every eigenvalue
of nite multiplicity of L is a branching point.
Let ( j ) denote the scalar product in E . We shall suppose:
(A1) L 2 L(E; E ) is a symmetric Fredholm operator with index zero.
(A2) There exists a functional h 2 C k (E; IR), for some k  3, such that
H (u) = h0(u). Moreover h(0) = h0(0) = h00(0) = 0.
Here if h is a functional on E , h0 denotes its gradient. Let us explicitely
point out that (A1;2 ) imply that the linearization of (1) at u = 0 is
given by v Lv = 0. Let us de ne f 2 C k (E; IR) by setting
f (u) = 21 kuk2 12 (Lu j u) h(u); (2)
so that f 0(u) = u Lu H (u) and  is the closure of the set of the
critical points u of f on E such that u 6= 0. Let  2 IR be an eigenvalue
of nite multiplicity of L and set Z = Ker[I L], where I denotes the
identity map in E . We shall also assume
(A3) 9 an integer k  3 and z~ 2 Z such that Dj h(0) = 0, 8 j =
1; : : : ; k 1, and Dk h(0)[~z ]k 6= 0.
For z 2 Z let us set
k (z) = k1! Dk h(0)[z]k :
Then k : Z ! IR is homogeneous of degree k and there results
h(z) = k (z) + o(kzkk ) as kzk ! 0, z 2 Z: (3)
Moreover, one also has
kH (u)k  a1 kukk 1; as kuk ' 0, u 2 E: (4)
Our next assumption is concerned with k . Let T denote the bound-
ary of the unit ball in Z . Let M := maxT k , m := minT k and let
 2 T , resp.  2 T , be such that k ( ) = M , resp. k () = m. We
assume
2
(A4) kM and km are not eigenvalues of the matrix D2 k ( ), resp.
D2 k ().
Theorem 1 Suppose that (A1), (A2 ), (A3), (A4) hold and let  be an
isolated eigenvalue of nite multiplicity of L. Then  is a branching
point of (1).
Remarks 2 (a) Assumption (A3 ) rules out the counterexample of [2].
Actually, in such counterexample, the nonlinearity is a C 1 functional
h 6 0 which has all the derivatives at u = 0 equal to zero. (A4 ) rules
out k such that
k (z)  const. on Z: (5)
jzjk
We also point out that the proofs will make it clear that (A4 ) can be
substituted by the weaker requirement that the points where maxT k
and minT k are achieved are isolated.
(b) It is easy to check that in Theorem 1 as well as in all the following
results, one can assume that E is a Banach space and assume that H is
de ned on a neighbourhood U of the origin in E . Furthermore, in some
cases it is possible to weaken the regularity assumptions on h, which
can be merely C 2 on Z . See Remark 11 in Section 3.
The following theorem contains a more precise description of  and can
be considered as a sharpening of the results of [12]. First, some more
notation is in order. If A   we set (A) = f 2 IR : (; u) 2 g
and A = fu 2 X : (; u) 2 Ag. Let S denote the maximal conneceted
component of  that contains (; 0).
Theorem 3 Suppose that (A1), (A2 ), (A3), (A4) hold and let  be an
isolated eigenvalue of nite multiplicity of L. Then the following state-
ments hold:
(i) If k is odd then (S ) contains an interval [a; b] such that a <  < b.
(ii) Let k be even. Then (S ) contains a one-sided neighbourhood  of
 such that for all  2  n fg (1) has at least two distinct nontrivial
solutions on S .
(iii) Let k be even, let d = dim(Z )  2 and suppose k (z) > 0 (resp.
< 0) for all z 2 Z n f0g. Then for every  =  + , with  > 0 (resp.
3
 < 0) suciently small, (1) possesses at least two pairs of distinct
solutions on S .
The proofs of Theorems 1 and 3 rely on a nite dimensional re-
duction that is carried out in Section 2. Roughly, this reduction leads
to studying the critical points of a functional  on Z . Furthermore,
assumption (A3 ) allows us to nd a dominating part of  which, for
 6= , has the mountain-pass geometry.
Although in applications assumption (A3) is veri ed by a quite
broad class of nonlinearities (see Section 4), it is still possible in some
cases when (A3) does not hold to control the behaviour of  on Z by
studying the further terms in the Taylor expansion of . See Theorem
12 in Section 3 below.
An application of the preceding abstract theorems to nonlinear elliptic
eigenvalue problems is discussed in section 4.
We end this introduction by an example suggested by Prof. E. N.
Dancer. It shows that if (5) is violated there could be no branching.
Example 4 Let B 2 C 1(IR2 ; IR) denote the map introduced by Bohme
in his counterexample and consider (subscripts denote partial deriva-
tives) (
x = x + x(x2 + y2) + Bx(x; y)
y = y + y(x2 + y2) + By (x; y):
Multiplying the rst equation by y and the second by x and subtracting
we nd yBx(x; y) xBy (x; y) = 0. As in [2] this implies that  = 1 is
not a branching point.

2 The nite dimensional reduction


In order to prove Theorem 1 we rst perform, like for the case of the
simple eigenvalue, a nite dimensional reduction.
Let W denote the orthogonal complement of Z in E : E = Z  W ,
and let P denote the orthogonal projection on W , parallel to Z . Setting
u = z + w, z 2 Z , w 2 W and  =  + , equation (1) becomes
F (; z; w) := (I L)w + z + w H (z + w) = 0: (6)
Lemma 5 There exists w = w(; z) de ned in a neighbourhood O of
(0; 0) in IR  Z such that PF (; z; w) = 0. Moreover w 2 C (O; W )
k

4
and one has that w(; 0)  0, Dzj w(0; 0) = 0 8 j = 1; : : : ; k 2. In
particular, 9 a > 0 such that
kw(; z)k  akzk2 ; (7)
uniformly for jj small.
Proof. One has that PF (0; 0; 0) = 0 as well as
PDw F (0; 0; 0)[v] = v Lv; (v 2 W ):
Then PDw F (0; 0; 0) is injective and hence invertible, because L is Fred-
holm. Then the result follows from the Implicit Function Theorem.

Let us de ne  : Z ! IR by setting
(z) = f (z + w(; z)):
Lemma 6 If z 2 Z is a critical point of  then u = z + w(; z ) is
    
a solution of (1) with  =  + . Furthermore, if z = 6 0 and kz k ! 0
 
as jj ! 0, then u = 6 0 and ku k ! 0.
 

Proof. If z 2 Z is a critical point of  there results


 

(f 0(u ) j  + D w(; z )[ ]) = 0; 8  2 Z:


 z 

By the de nition of w(; z) one has that Pf 0(z + w) = 0. Thus


(f 0(u) j Dz w(; z)[ ]) = 0; 8  2 Z
and hence (f 0(u) j  ) = 0; 8  2 Z . Using again the fact that
Pf 0(u) = 0 we conclude that f 0(u) = 0.

We now use assumption (A3) to evaluate 


Lemma 7 If (A3 ) holds then, for jj and kzk small, there results
(z) = 2 kzk2 k (z) + R(; z);
where
R(; z) = o(jj + kzkk ):

5
Proof. One has (for brevity we write w instead of w(; z ))

(z) = 2 kzk2 + 12 ( + )kwk2 12 (Lw j w) h(z + w):


>From (6) it follows that
( + )kwk2 (Lw j w) = (H (z + w) j w)
and thus
(z) = 2 kzk2 + 12 (H (z + w) j w) h(z + w): (8)
Using (3) we infer for s 2 (0; 1)
h(z + w) = h(z)+(H (z + sw) j w) = k (z)+(H (z + sw) j w)+ o(kzkk ):
(9)
Next, from (4) it follows
 
j(H (z + w) j w)j  a kzkk 1 + kwkk 1  kwk
and hence (7) implies, for jj small, that (H (z + w) j w) = o(kzkk ).
This, together with (8) and (9) proves the Lemma.

When (A3) does not hold and k  0 Lemma 7 needs to be replaced.


Setting
vk 1 = (k 1 1)! Dzk 1w(0; 0)[z]k 1 (2 W );
there results
w(; z) = vk 1 + o(jj + kzkk 1 ): (10)
Furthermore we set
k (z) = ((I L)vk 1 j vk 1) :
We anticipate that in Theorem 12 we will give conditions implying that
k (z) is not identically zero.
Lemma 8 Let k  3 be an integer such that Dj h(0) = 0, 8 j =
1; : : : ; k 1 and suppose that k  0 while k (z) 6 0. Then, for jj
and kz k small, there results
(z) =  kzk2 1 k (z) + o(jj + kzk2k 2 ):
2 2
6
Proof. Since w = w(; z ) satis es (6) then we infer that w1 =
Dzk 1w(0; 0)[z]k 1 satis es
(I L)w1 = P Dk 1H (0)[z]k 1: (11)
Recall that there results, see (8):
(z) = 2 kzk2 + 12 (H (z + w) j w) h(z + w):
Let us evaluate separately the two terms h(z + w) and (H (z + w) j w).
Since k (z)  0 we have
h(z + w) = k1! Dk h(0)[z + w]k + o(jj + kz + wkk )
= (k 1 1)! Dk h(0)[z]k 1 [w] + o(jj + kzkk ):

Using (10) and (11) it follows


h(z + w) = (k 1 1)! Dk h(0)[z]k 1 [vk 1] + o(jj + kzk2k 2 )
= ((I L)vk 1 j vk 1) + o(jj + kzk2k 2)
= k (z) + o(jj + kzk2k 2): (12)
Similarly, one also have
H (z + w) = (k 1 1)! Dk 1H (0)[z]k 1 + o(jj + kzkk 1)
and hence
(H (z + w) j w) = (k 1 1)! (Dk 1H (0)[z]k 1 j vk 1) + o(jj + kzk2k 2)
= ((I L)vk 1 j vk 1) + o(jj + kzk2k 2 )
= k (z) + o(jj + kzk2k 2 ): (13)
Putting together (12) and (13) we nd
1 (H (z + w) j w) h(z + w) = 1 (z) + o(jj + kzk2k 2); (14)
2 2 k
and the Lemma follows.

7
3 Proofs of the main Theorems
In the rst part of this section we will carry out the proofs Theorems
1 and 3. Next we will discuss a result (see Theorem 12 below) in which
(A3;4) is replaced by other conditions.
First some more notation is in order. If p 2 Z let B (p; r) denote a ball
in Z centered in p with radius r and set T (r) = @B (0; r), T = T (1).
Let
1  kzk2 (z):
 (z ) =
2 k

Extending, if necessary, k by homogeneity, we can assume that  is


de ned on all of Z . If g : Z ! Z is continuous, and N  Z is open
bounded set such that g(z) 6= 0; 8 z 2 @N , deg(g; N; 0) denotes the
topological degree of g with respect to N and 0. Recall that if p is
an isolated solution of g = 0 then deg(g; B (p; r); 0) is well de ned and
constant for all r > 0 small and this common value is denoted by i(g; p),
the index of g at p. For properties of the degree, see for example [5].
Proof of Theorem 1. Since k 6 0, either M := maxT k > 0 or
minT k < 0. The proof will be carried out assuming the former: in
the other case it suces to consider . Let us point out that for
 > 0 the functional  has the Mountain-Pass geometry because
k (z) = o(kzk2 ), k 6= 0 somewhere and k  3. To nd the Mountain-
Pass critical point let  2 T be a point where M is achieved. By homo-
geneity it immediately follows that k0 ( ) = kM and that p = t is
a critical point of  whenever t satis es the equation
 ( > 0):
tk 2 = kM (15)
It is easy to check that p is the Mountain-Pass critical point of  we
were seeking. Let us explicitely point out that one has
p ! 0 as  ! 0+: (16)
Lemma 9 p is a non-degenerate mountain-pass critical point of 
and there results
i( 0; p) = 1: (17)
Proof. Let IZ denote the identity in Z . There results
D2 (p) = IZ D2 k (p):
8
Since p = t one nds
D2 (p) = IZ tk 2D2 k ( ) = IZ kM  D2 ( ):
k

Using (A4 ) it follows that D2 (p) is invertible. Then  is non degen-


erate and hence p is also a non degenerate critical point of . As an
non degenerate mountain-pass critical point, it is well known that (17)
holds.
Referring to Remark 2 (a), we point out that in the present speci c
case the proof of (17) can be carried out even if suppose that  is (pos-
sibly degenerate, but) isolated. Let us give a sketch of the arguments.
Consider the hypersurface M = fz 2 Z : kzk2 = k k (z)g  Z . One
readily checks that the p is the (possibly degenerate) global minimum
of  on M. Letting T M denote the tangent space to M at p, it follows
that for the restriction f0 of 0 to T M there results
i(f0; p) = 1:
For t  0 let (t) := (t ) and remark that t = t is a (non-degenerate)
maximum of . Then the multiplicative property of the degree yields
(note that T M is transversal to the ray IR+ )
i( 0; p) = i(f0; p)  i( ; t) = 1;
proving (17).

From the preceding Lemma it follows that, for  > 0 suciently small
one also has
deg(0; B (p; r); 0) = 1; r > 0 small: (18)
In particular  has a critical point z 2 E near p.
Lemma 10 There exists  > 0 and a closed connected set Y  IR  Z
such that
1) (; 0) 2 Y and (Y ) contains the interval [;  + ].
2) If ( + ; z) 2 Y with jj  , then z is a non-trivial critical point
of  .
Proof. First, let us x  > 0 and r such that (18) holds with  =  .
For each positive integer n such that 1=n < , (18) and the properties
of the degree imply, in the usual way, there exists a connected set Yn 
[ + 1=n;  + ]  Z with the following properties:
9
(a) if ( + ; z) 2 Yn then z is a critical point of  for all 1=n    ;
(b) ( + 1=n; z) 2 Yn ) z 2 B (p1=n ; r1=n);
(c) ( + ; z) 2 Yn ) z 2 B (p ; r ).
(d) [ + 1=n;  + ]  (Yn).
We now apply to Yn a topological theoretic result, Theorem 9:1 of [14].
According to that: if (i) Sn Yn is precompact; and (ii) lim infn(Yn) is
not empty then Y := lim sup Yn is not empty closed and connected. In
the present case (i) is trivial. As for (ii), take any zn such that ( +
1=n; zn) 2 Yn. Then (b) and (16) imply that zn ! 0 as n ! 1 proving
that (; 0) 2 lim inf Yn. To prove that the set Y satis es property 1)
let zn be such that ( + ; zn ) 2 Yn. Then by (c) zn 2 B (p ; r ) and
hence, up to a subsequence, zn converges to some z 2 B (p ; r ) with
( + ; z) 2 Y . Since Y is connected, 1) follows. As for property 2), let
( + ; z) 2 Y . Then there exists a sequence ( + n; zn) 2 Yn such that
n ! , zn ! z and 0n (zn) = 0. Passing to the limit we infer that z is
a critical point of . In addition, since 0 is an isolated critical point of
, 2) follows (taking  > 0 possibly smaller), proving the Lemma.
Proof of Theorem 1 completed. Let X = f(; u) : u = z+w(; z ); (; z ) 2
Y g. From Lemma 10 we deduce:
- X is is closed and connected because Y is and w is continuous.
- If (; u) 2 X then u is a solution of (1), see also Lemma 6.
- (; 0) 2 X and (X ) contains the interval [;  + ].
This shows that (; 0) is a branching point and completes the proof of
Theorem 1.
Remarks 11 (a) If (A1 2 3 ) hold,  has a mountain-pass critical point
; ; 
z~ ! 0 as  ! 0. This suces to prove that  is a bifurcation point of

(1).
(b) Assumption (A3) implies that hjZ is C k . However, it is possible
to weaken such regularity condition. For example, all the preceding
proof remains true if in (A3) we suppose that h(z) = kzk k (z) with
0 <  < 1 and k is homogeneous of degree k  2.

10
(c) In all the preceding arguments we have used the variational structure
only to study the functional . As a consequence, we can extend the
above results to the case that (1) is replaced by
Lu + H (u) + K (u) = u;
where L and H satisfy (A1;2;3;4) and K is a possibly non-potential op-
erator which is higher order than H at u = 0.
Proof of Theorem 3. The proof is mainly based on a sharpening of the
preceding arguments and thus we will be sketchy.
(i) If k is odd then (15) has solutions both for  > 0 and  < 0. Repeating
the above arguments, we nd at least two families of critical points
u+ ' 1=(k 2) ; ( > 0) and u ' 1=(k 2) ; ( < 0)
and (i) follows.
(ii) We deal again with the case that k  0, otherwise we consider .
Since k is even then (15) has two solutions for  > 0 in correspondence
of which  has a pair of critical points
 1=(k 2)
p =  kM
  ;
which are isolated provided (A4) holds. We can now repeat the argu-
ments of Lemma 10 to yield two disinct sequences of connected sets
Yn+; Yn satis ying the properties (a b c) above with p replaced by
p+ , esp. p . Then letting Y  : lim sup(Yn) and
X  = f(; u) : u = z + w(; z); (; z) 2 Y g
we nd two distinct connected solution sets branching to the right hand
side of (; 0).
(iii) Let k be even, d  2. If k > 0 on T , then 0 < m < M . Let  6= 
of k on T such that k () = m. It follows that  has another pair of
isolated critical points q 6= p,
 1=(k 2)
q =  km
  :
Since q are isolated maxima of  we have
i( 0; q) = ( 1)d:
11
By repeating the arguments carried out before, the proof of statement
(iii) follows.
Our next result deals with a case when (A3;4) do not hold. We use
the notation introduced in Section 2, after Lemma 7. Let us assume
that
(A5) there is z 2 Z such that k (z ) 6= 0.
Let M  := maxT k , m := minT k and let   2 T , resp.  2 T ,
be such that k ( ) = M  , resp. k () = m . We further assume
(A6) kM  and km are not eigenvalues of the matrix D2 k ( ), resp.
D2 k ().
Theorem 12 Suppose that (A1), (A2), (A5) and (A6) hold and let
(z)  0 (otherwise we use Theorems 1 and 3). Then any isolated
k
eigenvalue  of nite multiplicity of L is a branching point of (1). Fur-
thermore (S ) has the properties stated in Theorem 3 -(ii).
In particular, if (A5 ) is substituted by
(A7) there exists an integer k  3, z 2 Z and w 2 W such that
Dj h(0) = 0 for all 1  j  k 1 and Dk h(0)[z]k 1 [w] 6= 0,
then the greatest eigenvalue  of L is a branching point and (S ) con-
tains a right neighbourhood  of , such that for all  2  n fg, (1)
has at least two distinct nontrivial solutions on S .
Proof. It suces to repeat the arguments used in the Proof of The-
orems 1 and 3, taking into account Lemma 8 and the fact that k is
homogeneous of degree 2k 2.
To prove the last claim let us recall that we have set vk 1 :=
k 1 6= 0. Fur-
1 w where w satis es (11). Then (A ) implies that v
(k 1)! 1 1 7
thermore, if  is the greatest eigenvalue of L, its variational character-
ization implies that k (z) = ((I L)vk 1 j vk 1) > 0.

We end this section with some remarks.


Remarks 13 (a) Unlike the results on bifurcation cited in Section 1,
apart from the one dealing with the simple eigenvalue case, we can

12
describe in a rather precise way the (local) behaviour of S . Precisely, if
k is odd:
u ' ( )1=(k 2) 
for some  2 Z , k k = 1. While if k is even and k > 0, resp. < 0,
u '  j j1=(k 2)  with  > , resp  < :
(b) It is worth pointing out that the sharp description of S , in particular
statements (i) (ii) of Theorem 3, depends on the behaviour of h on Z ,
in particular on the assumption (A3 ). For example, when Theorem 12
applies k is odd and k  0, the behaviour of S is not like that found
in Theorem 3 (i).
(c) As a consequence of the preceding results it follows that for any
r > 0 small enough, (1) has at least two solutions (; u) such that
kuk = r. This has to be compared with the results of [2, 7, 8].
(d) If k is even and k is a Morse function, then we can nd much
stronger multiplicity results. If, for example, k > 0 on Z nf0g then for
all  > 0 small (1) possesses d pairs of solutions. Let us point out that
we do not need to assume that h is even.
(e) Theorems 1 and 3 can also be seen as the extension of some results
of [7], Chapt. IV, in the variational case. Under the assumption that
the operator H is of the form H (u) = C (u) + R(u), where C is k-
homogeneous, R is a higher order term and (I P )C (u) does not vanish
on T , it is shown that (; 0) is a bifurcation point. Neither branching nor
multiplicity results are proved in [7]. Moreover, it is worth mentioning
that the condition that (I P )C (u) 6= 0 on T is stronger than our
assumption (A3).

4 Examples
Theorems 1 and 3 are prompted for b.v.p. like
(
 u = u + G0(u); in
; (19)
u = 0; on @
;
where
is a bounded domain in IRn with smooth boundary @
and G
satis es, for some integer k  3,
(G1) G 2 C k (IR),
(G2) G(t) = k1 tk + o(jtjk ), as t ! 0.
13
Actually, it is well known that solutions of (19) are the critical points
of (2) on E = H01(
), the usual Sobolev space endowed with scalar
product Z
(u j v) = ru  rv dx;

L and h are de ned by


R
(Lu j v) =
u(x)v (x) dx;
R
h(u) =
G(u(x)) dx:
Let us point out that the bifurcating solutions of (19) have norm which
is small in E and, by regularity, in C (
). Thus, without loss of gener-
ality, we can assume that G is, say, quadratic at in nity so that h is
well de ned and smooth. The eigenvalues  of L are nothing but the
characteristic value of  on H01(
). It is immediate to verify that the
assumptions (A1 ) and (A2) hold true. Furthermore, xed one of these
, let '1; : : : ; 'd 2 E denote a basis of orthonormal
Pd
eigenfunctions of
' = ' so that any z 2 Z has the form z = i=1 zi 'i. Then we
nd !k
Z X
1
k (z) = k
d
zi'i(x) dx:

i=1
It follows that (A3 ) holds when k is even while, if k is odd, then (A3)
is satis ed provided
R Pd k
()
i=1 z i ' i ( x ) dx 6 0.
Remark 14 When k is odd it could happen that () does Rnot hold.
For example this is the case if G(u) = u3, Z = spanf'g and
'3 = 0.

In the following example we discuss conditions (A4) and ().

Example 15 Let us consider (19) with G(t) = 1 t , k  3, and assume


k
k

that  is a characteristic value of  in E . Let us assume that Z is


2 dimensional and spanned by '1 ; '2.
If k = 3 condition () holds provided at least one of the following
integrals Z Z Z Z
'31 ; '21'2; '1 '22; '32 (20)



14
is di erent from zero.
As for (A4 ), a straight calculation shows:
1) let k = 3 and let
Z Z Z Z
'3 = '3 = 1; '2 '2 = '1'22 = 0:

1
2
1

Then (A4) holds.


2) let k = 4 and let
Z Z Z Z Z
'41 = '42 = 1; '21 '22 = a; '31'2 = '1 '32 = 0:




Then (A4) holds for all a but a = 1.


Remark 16 The abstract results also apply to b.v.p. on unbounded
domains. For example, any isolated eigenvalue  of the Schroedinger
operator  + V (x) on H 1(IRn) is a branching point of
u + V (x)u = u + G0(u); u 2 H 1(IRn)
provided G satis es (i ii) and () holds.
The last example is concerned with an application of Theorem 12 to
a class of eigenvalue problem involving the bi-harmonic operator 2 =
.
Example 17 Consider the nonlinear eigenvalue problem
(
 2 u = u + G0(u); in
; (21)
u(x) = @u=@n = 0; on @
;
where @=@n denotes the outer normal derivative. Problem (21) can be
written in the abstract form (1). This can be carried out in a standard
way, similar to that discussed before for (19), with obvious changes. We
take k = 3 and
to be the annulus fx 2 IR2 : r < jxj < 1g. Let  be
the greatest characteristic value of the linear problem
(
 2 u = u; in
; (22)
u = @u=@n = 0; on @
:
When the internal radius r is suciently small then  is double and the
corresponding eigenfunctions '1; '2 change sign, see, e.g. [4]. Moreover
the preceding characteristic value ~ <  turns out to be simple and the
15
corresponding eigenfunctions do not change sign in
. In this case all
the integrals in (20) could be zero and hence () could not be satis ed.
On the other side one nds
Z
D3h(0)['1]2 [ ] = 2 '21 6= 0:

This shows that (A7) holds with z = '1 and w = . Hence Theorem
12 applies and  is a branching point for (21).

Acknowledgments
The author wishes to thank F. Bernis for indicating references on the bi-
harmonic operator including [4], J. Toland for many valuable comments
and E. N. Dancer for pointing out a gap in the preceding version of the
paper as well as for indicating Example 4.

References
[1] A. Ambrosetti - G. Prodi, A Primer of Nonlinear Analy-
sis, Cambridge Studies in Advanced Math. 34 (1993), Cambridge
Univ. Press.
[2] R. Bohme, Die Losung der Verzweigungsgleichungen fur nichtlin-
eare Eigenvertprobleme, Math. Zeit. 128 (1972), 105-126.
[3] S.-N. Chow - R. Lauterbach, A bifurcation theorem for crit-
ical points of variational problems, Nonlinear Analysis TMA 12
(1988), 51-61.
[4] C.V. Coffman - R.J. Duffin - H. Shaffer, The fundamental
mode of vibration of a clamped annular plate is not of one sign, in
Constructive Approaches to Mathematical Models, Ed. C.V. Co -
man and G.J. Fix, Academic Press, 1979.
[5] I. Fonseca - W. Gangbo, Degree Theory in Analysis and Ap-
plications, Oxford Univ. Press, New York, 1995.
[6] H. Kielhofer, A bifurcation theorem forpotential operators, J.
Funct. Anal. 77 (1988), 1-8.

16
[7] M.A. Krasnoselski, Topological Methods in the Theory of Non-
linear Integral Equations, Macmillan, New York, 1964
[8] A. Marino, La biforcazione nel caso variazionale, Conf. Sem.
Mat. Univ. Bari 132 (1973).
[9] A. Marino - G. Prodi, La teoria di Morse per gli spazi di
Hilbert, Rend. sem. Mat. Univ. Padova 41 (1968), 43-68.
[10] G. Prodi, Problemi di diramazione per equazioni funzionali, Boll.
U.M.I. 22 (1967), 413-433.
[11] P.H. Rabinowitz, Some global results on nonlinear eigenvalue
problems, J. Funct. Anal. 7 (1971), 487-513.
[12] P.H. Rabinowitz, A bifurcation theorem for potential operators,
J. Funct. Anal. 25 (1977), 412-424.
[13] C. Stuart, An introduction to bifurcation theory based on dif-
ferential calculus, Research Notes in Math. no. 39, Pitman (R.J.
Knops, ed.), 1979, 76-132.
[14] G.T. Wyburn, Topological Analysis, Princeton, 1955.

17

You might also like