Benci CriticalPointTheory 1982

On Critical Point Theory for Indefinite Functionals in The Presence of Symmetries
Author(s): Vieri Benci

Source: Transactions of the American Mathematical Society , Dec., 1982, Vol. 274, No. 2
(Dec., 1982), pp. 533-572
Published by: American Mathematical Society
Stable URL: https://www.jstor.org/stable/1999120
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend
access to Transactions of the American Mathematical Society
This content downloaded from

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
All use subject to https://about.jstor.org/terms
TRANSACTIONS OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 274, Number 2, December 1982
ON CRITICAL POINT THEORY FOR

INDEFINITE FUNCTIONALS IN
THE PRESENCE OF SYMMETRIES
To the memory of Guido Stampacchia
BY
VIERI BENCI
ABSTRACT. We consider functionals which are not bounded from above or from
below even modulo compact perturbations, and which exhibit certain symmetries
with respect to the action of a compact Lie group.
We develop a method which permits us to prove the existence of multiple critical
points for such functionals. The proofs are carried out directly in an infinite
dimensional Hilbert space, and they are based on minimax arguments.
The applications given here are to Hamiltonian systems of ordinary differential
equations where the existence of multiple time-periodic solutions is established for
several classes of Hamiltonians. Symmetry properties of these Hamiltonians such as
time translation invariancy or evenness are exploited.
Introduction. Many variational problems which arise from physics or mathemati-

cal physics are indefinite in the sense that the functionals involved are not bounded
from below or from above. However some of these functionals, defined in an
appropriate function space, are "semidefinite" in the sense that they are bounded
from below (or from above) if perturbed with a weakly continuous functional.
This paper deals with functionals which are not semidefinite. Usually problems
involving indefinite functionals are more difficult to handle and only very recently a
method has been developed which permits us to treat such problems directly in an
infinite dimensional function space [7]. In [7] some theorems have been proved
which establish the existence of at least one nontrivial critical point for such
functionals.
In many physical situations there are problems which have symmetries with
respect to the action of some Lie group. In this case we expect the existence of many
critical points; this has been established for semidefinite functionals (cf. e.g. [8 and
3] for even functionals, [9] for a Zp-action with p prime numbers [15, 16, 10 and 6]
for an S'-action).
In this paper we develop a method which allows us to estimate the number of
critical points of indefinite functionals in the presence of symmetries. More precisely
Received by the editors July 16, 1980 and, in revised form, June 26, 1981.
1980 Mathematics Subject Classification. Primary 58E05, 70H05; Secondary 34C25.
Key words and phrases. Critical point, symmetry, minimax methods, Hamiltonian systems, periodic
solutions.
? 1982 American Mathematical Society

0002-9947/82/0000-0677/$ 12.50
533

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
534 VIERI BENCI
we consider a class of functionals defined on a real Hilbert space H having the

following form:
(0-1) f(u) = 2<Lu, u) + (I(u)

where L is a bounded selfadjoint operator and 1 E C'(H, R) is a functional whose
Frechet derivative T = 1" E C(H, H) is a compact operator.

We assume also that both L and q9 are equivariant with respect to the action of a
linear representation of a Lie group. Before going on in our discussion we shall
expose one of the theorems to be proved later on. It is not in the most general form,
but it clarifies the nature of our results.
THEOREM 0.1. Let co < coo < 0 be two constants and let f E C'(H, R) be a functional
which satisfies the following assumptions:
(fI) f has the form (0-1).

(f2) Any sequence {U,U C H such that f(un) -' c E [co, co] and 11 fU'(Ju)I - 0 as
n +oo has a convergent subsequence.
(f3) ?'(u) = ?(-u)
(f4) There are two closed subspaces of H, H? and H-, and a constant p > 0 such
that
(a)f(u) > cofor u E H+,
(b)f(u) < co, < 0 for u E H-FnS. (Sp {u E H lull p}).
Then the number of pairs of nontrivial critical points of f is greater or equal to
dim(H+ n) H-)-cod(H-+H+). Moreover the corresponding critical values belong
to [co, co"].
(f2) is a slightly weakened version of the well-known condition of Palais and

Smale. (f3) expresses the equivariancy of the functional f under the action of the
antipodal map. (The action of more general Lie groups is considered in the paper.)
(f4) is a geometrical condition which permits us to give a lower bound to the number
of solutions.
The most interesting case occurs when both H+ and H- are infinite dimensional;
otherwise the functional f would be semidefinite according to our definition and we
would obtain a variation of known results. For example, if H+ is finite dimensional,
Theorem 0.1 is just a variant of Theorem 2.19 of Ambrosetti and Rabinowitz [3]. If
H- is finite dimensional, Theorem 0.1 is a variant of Theorem 12 of [8] (cf. also [6] in
the case in which f is invariant under an S' -action).
Also note that Theorem 0.1 is similar to Theorem 0.1 of [7]. In [7] the form of the
functional and the "geometry" are very similar, but no equivariancy property such
as (f3) is assumed. Therefore a weaker result is obtained, i.e. the existence of only
one nontrivial solution. The proofs of our theorems are based on minimax argu-
ments and they are carried out directly in an infinite dimensional Hilbert space.
This paper is organized as follows: In ?1, we describe the abstract framework in
which the theory will be constructed. This is done in order to clarify the main steps
of the construction of the theory. For this purpose we introduce the concept of a
" pseudoindex theory" which gives an axiomatic description of the properties required
for the multiplicity results.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
CRITICAL POINT THEORY FOR INDEFINITE FUNCTIONALS 535
When an "index theory" (in the sense of [10, 6, 13, 15] and Definition 1.1) is
available, it is always possible to construct some pseudoindex theories. A pseudo-
index theory is simply a tool which permits us to get rid of the indefiniteness. In this
framework many of the known results for semidefinite functionals can be simplified
and generalized (cf. e.g. [4]).
In ?2 we introduce two concrete "pseudoindex theories" which can both be
applied to the study of indefinite functionals. In a sense which will become clear
later on, they are duals of each other, and they give different characterizations of the
critical values.
In order that the minimax principle be applicable, we need a suitable "deforma-
tion theorem" consistent with our pseudoindex theories. The proof of this theorem
(Theorem 3.4) is the topic of ?3. It is the central and most delicate part of this paper.
Actually, the particular choice of our pseudoindex theories can be understood only
in relation to the deformation theorem.
In ?4 we can finally establish some abstract multiplicity results on the existence of
critical points of (0-1) and in particular we shall prove Theorem 0.1.
In the last two sections we deal with applications.
In ?5 we look for periodic solutions of fixed period of asymptotically linear
Hamiltonian systems. We suppose that the Hamiltonian function satisfies the
following assumptions:
aH
az(Z)=AZ?o(IZI) aslzl-*+x
and
aH (z) = Bz + o(I z l) as I z 0.
z C R2n and A, B: R2n R2 are linear symmetric operators.
We define a symplectic invariant O(TB/27T, TA/27T) (T is the period we are
interested in) which assumes only even values.
Under various assumptions on the Hamiltonian function, we show that the
corresponding Hamiltonian system has at least O(TI2B/7T, TAA/2T) nonconstant
i-periodic solutions.
In the case in which 4(TB/2T, TA/2T) > 0, our results are an improvement of
those of Amann and Zehnder [1] since they required the Hamiltonian function to be
strictly convex. If (TBIB/2T, TA/2T) < 0, we need a further symmetry property for
the Hamiltonian function such as evenness. Then we get different results from theirs.
In the last sections we make some remarks on periodic solutions for super-
quadratic Hamiltonian functions and indicate how to apply the theory of ?4 to such
situations.
I thank P. H. Rabinowitz and E. Zehnder for helpful conversations.
1. The abstract framework. Let X be a Riemannian manifold modelled on a

Hilbert space H. For A C X, Sk(A) denotes the space of k times Frechet differentia-
ble maps from A to R. For A and B C X, Ck(A, B) denotes the sets of k times
Frechet differentiable maps from A to B. Id denotes the identity map. Moreover we
set N&(A) = {x C XI dist(x, A) < 8? .

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
536 VIERI BENCI
DEFINITION 1.1. An index theory I on X is a triplet {E, 9T, i} which fulfills the
following properties:
(l-1) Y is a family of closed subsets of X such that A U B, A n B, A \B E z
whenever A, B G E.
(1-2) 9T is a set of continuous mappings containing the identity and closed under
composition.
(1-3)VA E , andVh 6X, h(A) E E.
(1-4) i: 2 -- N U { + oo } is a mapping which satisfies the following properties.
(i-l)i(A) O0ifandonlyif A = 0.
(i-2) (Monotonicity) If A C B then i(A) < i(B) VA, B E E.
(i-3) (Subadditivity) i(A U B) < i(A) + i(B) VA, B E E.
(i-4) (Continuity) If A E . is a compact set, then 38 > 0 such that i(N (A)) =
i(A).
(i-5) (Supervariancy) i(A) < i(h(A)) VA Cz Vh C 9t.
If 2 is the family of all closed subsets of X, 9Th is the family of the contin
mappings of X homotopic to a constant map, and catx is the Ljiusternik-Schnirel-
mann category, then it is immediate to check that {E, 9T, cat } defines an index
theory on X.
In ?2, we shall give other examples of index theories.
When we deal with indefinite functionals (i.e. functionals unbounded from below)
the existence of an index theory may not be sufficient to guarantee the existence of
critical points if we simply use the Ljiusternik-Schnirelmann theory in a direct way.
Then the concept of pseudoindex theory turns out to be useful.
DEFINITION 1.2. Let X and I = {E, 9Th, i} be as in Definition 1.1. A pseudoindex
theory I* is a couple { 9T*, i*} which satisfies the following assumptions
(1-6) 9Th* C '9T is a group of homeomorphisms of X onto X.
(1-7) i*: I -- N U {+ } is a map with the following properties
(i*-l) i*(A) < i(A) VA E E.
(i*-2) If A C B, then i*(A) < i*(B) VA, B E E.
(i*-3) i*(A \B) > i*(A) - i(B) VA, B E E.

(i*-4) i*(h(A)) = i*(A) Vh E 9* VA C E.
We shall show how the concept of pseudoindex can be applied in the search for
critical points of a functionalf E C1(H).
For each c E R we set
(1-8) (&C= {x e XIf(x) < c},

Kc = {x E XIf(x) = c andf'(x)-O}.
DEFINITION 1.3. If f C C'(X) and co, cOO E R (with co < cx),
triplet {f, co, c. } satisfies the property (P) with respect to the cou
(a) (T, Kc C E and K. is compact for every c E [Ico, c'

(1-9) 1(b) Vc C [co, c.,], VN = Nd(KC), 3e > Oand
such that
-q N) CU 9, C-

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
In concrete cases, the property (P) is strictly related to the assum

and Smale (cf. (f2), Theorem 3.4 and Corollary 3.5). The following theorem holds:
THEOREM 1.4. Let X be a Riemannian manifold with an index theory I ={, 9'T, i}
and a pseudoindex theory I* 6{'AR*, i*}. Suppose that f C1( X) is a functional such
that
(a) there exist constants co, coo E R such that

{ f, co, c OO satisfies the property (P)
with respect to {y, 9Th*}.
(1-10) (b) i*(A) = O VA EE C such that A C d.
(c) there exists A E 2 such that A C 6co and
i*(A) > k 1.
Then the real numbers
( 1 -11) ck inf sup f(x), k =1, .., k,

i*(A)'-'k x&=A
are critical values of f and co ? cl < C2 * * Ck-? co. Moreover if c ck =

ck+r with k > 1 and k + r < k then i(Kc) > r + 1.
The proof of Theorem 1.4 follows standard arguments and it will not be given
here.
REMARK 1.5. If in the assumption (c) we know that i*(A) = +oo, then clearly
(1-11) defines critical values for each k E N+ .
In order to apply Theorem 1.4 to concrete situations, it is necessary to construct
an appropriate pseudoindex theory, which of course, depends on the functional f,
whose critical points we seek.
In the following section we shall use a method to construct pseudoindex theories
which is described by the following proposition.
PROPOSITION 1.6. Let I = 1{, 9T, i} be an index theory on the Riemannian

manifold X.
Let 6Th* C 'Th be a group of homeomorphisms on X.
Given Q E Y, we set i*(A) = minhC,9 i(h(A) n Q) for each A E E.
Then I* = {Th*, i*} is a pseudoindex theory.
PROOF. We have to show that the properties (1-7) are verified. (i-1) (i-2) (i-4) are
trivial. Let us prove (i-3). For each h E 'AR*, and A, B E 2, we have
h(A\B) n Q= (h(A)\h(B)) n Q= (h(A) n Q)\h(B)

then applying (i-3) and (i-5), we get
i(h(A \B) n Q) = i( (h(A) n Q)\h(B))

> i(h(A) n Q) - i(h(B)) = i(h(A) n Q)) - i(B)

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
538 VIERI BENCI
therefore
i*(~A\W)= min it
min
h &z DR,*
[i(h(A) n Q)-i(B)] = i*(A)-i(B). I
In applications, when the appropriate index theory is already known, the main
difficulty is to find an appropriate pseudoindex theory. Essentially the problem is to
determine the right class of homomorphisms 6DTh*. This class should be "big" enough
in order to contain a function 'q such that (1-9)(b) is satisfied. But if 'DTh* is "too
big", it may happen that i*(A) = 0 or 1 for each A E . (cf. the construction of the
pseudoindex theory of Proposition 1.6). Therefore Theorem 1.4 may not be applica-
ble or gives the existence of only one critical value (cf. (1-1O)(c)). In the next section
we shall construct pseudoindex theories which will be useful in the search for critical
values of functionals defined on a Hilbert space which are indefinite in the sense
given in the introduction and which are symmetric with respect to the action of some
Lie group.
However the abstract framework presented in this section can be applied to
various situations and many known results about semidefinite functionals can be
simplified and generalized. In [4] there is a detailed analysis of some of those
situations.
2. Index and pseudoindex theories. From now on we shall consider only index and
pseudoindex theories on a real Hilbert space H on which the unitary representation
Tg of a compact Lie group G acts. Some notation is now necessary.
K< , ) denotes the scalar product on H and the symbol (, ) will be left for the
scalar product in RW, Bp(u) denotes the closed ball of center u and radius p. Also we
set Bp = Bp(O) and Sp = aBp.
A functionalf E C1(H) is said to be Tg-invariant if
f(Tgu) = f(u) Vu C H, Vg C G.
A map h C C(H, H) is said to be Tg-equivariant if
h(Tgu) = Tgh(u) Vu E H, Vg E G.
If f C C'(H), then f' E 60(H, H) since we identify H with its dual, and if f is
Tg-invariant f' is Tg-equivariant. A subset A C H is said to be Tg-invariant if
TgA =A VgEG.
F = {u E H I Tgu = u Vg E G} will denote the linear space of the Tg-inv

points of H.
We recall (cf. e.g. [18]) that, by virtue of the Peter-Weyl theorem, a Hilbert space
on which the representation of a compact Lie group acts may be decomposed in the
following way:
00
(2-1) H = @ Hj where Hj is finite dimensional and invariant Vj C N.

=l o
V k will denote a k-dimensional Tg-invariant sub sp ace of H.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Moreover we set
(Tg) {A C H I A is closed and T-invariant},
6DTh(Tg) {h E C(H, H) I h is T-equivariant}.

We shall say that an index theory {E, 6Th, i} is related to the representation T if
2 = Y(Tg) and 9 = 6D(Tg).
In this paper we shall consider only index theories related to some group
representations.
In particular we shall consider only index theories which have an important
property described by the following definition.
DEFINITION 2.1. An index theory is said to satisfy the dimension property if there
is a positive integer d such that
(2-2) i(Vdk n SI) = k for all dk-dimensional subspaces Vdk E Y such that Vdk n
F = {O}.
If an index theory has the dimension property, it is not difficult to prove t
e.g. [4])
(2-2') i(A) < k if A C Vdk andA n F= Oand
(2-2") i(K) < +ox if K is compact and K n F = 0.
It is not difficult to realize that any index theory with the dimension property has
the further property that
(2-3) i(A) +oo whenever A n F # 0.
In fact, if ui E F, the constant map 4i1: H (u) belongs to 6R. Then, if
(i-5) and (2-2), we have that
i({U}) > i(4I(H n SI)) > i(H n SI) > d-1 dim H.
Since, in general, dim H = +oo, (2-3) follows.
The property (2-3) causes some problems when multiplicity results need to be
proved. This difficulty sometimes may be overcome by constructing appropriate
pseudoindex theories and using some further tricks. All these remarks will become
clear in the next sections.
We shall give three examples of index theories related to the representations of
some Lie group which satisfy the dimension property with d = 1, 2 and 4 respec-
tively. The first two examples will be used in the applications in the last sections.
I EXAMPLE. Consider the group Z2 = {O, 1) and the unitary representation of this
group on a real Hilbert space H defined as follows
Tou=u, T1u=-u, uCH.
If A C Y.(Tg) = {closed subsets of H symmetric with respect to th

y(A) = k
if k is the smallest integer such that there exists a continuous odd
D: A Rk\ {O}.
If such a map does not exist we set i(A) +oo and we set y(0) = 0. This set
function called "genus" has been introduced by Krasnoselskii in an equivalent form

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
540 VIERI BENCI
(see [11]) and used by many authors (cf. e.g. [3, 4, 8, 13]). In [11] and in [13] it has
been proved that {Y(Tg), ?1T(Tg), y} is an index theory and that it satisfies the
dimension property with d = 1.
Actually in the papers mentioned above the genus has been defined only for those
sets A e , such that 0 ( A, but it can be extended to all E. In fact if 0 C A, (2-3)
shows i(A) = +cc and that all the required properties are satisfied.
II EXAMPLE. Consider the group S' { z E C I I z I = 1) and a unitary represen-
tation Tz of this group on a real Hilbert space H. To simplify the notation we shall
write Ts instead of Tz if z eis (s E [0, 27T)). If A E 2(Ts) ={closed Ts-invariant
subsets of H) we set T(A) = k if k is the smallest integer for which there exist a
number n E N+ and a continuous map
D: A Ck\{0}
such that
1(Tsu) = einsD(u) Vu EA, Vs E [0,2X].
If such a map does not exist we set T(A) = +oo; moreover we set T(0) = 0. In [6] it
has been proved that {f(Ts), t(Ts), i} is an index theory.
If Vk' E l is an invariant subspace of H, and Vk' n F = (0) it is not difficult to
see that its dimension is even i.e. k' = 2k, k E N. In [6] it has been proved that the
above index theory satisfies the dimension property with d = 2. Since all the
invariant spaces are even dimensional this makes sense.
III EXAMPLE. Let H be the set of all sequences of quaternions {ajl},N (a. E H)
such that
00
E laj 12 <+00
j=1
Clearly H has the natural structure of Hilbert space on the real field if we
with R4.
A unitary representation of
s3= {w ECHJIw= 1)
defined in the natural way acts on H:
Tw({a]}aCN) ={Wajh},N-
In [10] it has been proved that there exists an index theory related to the above
representation with d = 4. The index theories y and T will be used in the applications
of ??5 and 6. In the literature there are many other ones, perhaps with different
names. We mention only the index of Fadell and Rabinowitz [10] and we refer to
their paper for further information on this topic.
Our program now is to construct two pseudoindex theories I* and I2 related to
any index theory which satisfies the dimension property. As has already been

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
observed these pseudoindex theories will permit us to deal with indefinite function-
als. To begin this program, we define the following families of maps. For X is a
Banach space, we set
10(X) = {h E C (X, X) Ih = U+ Twhere Uis linear andTis compact),

I,( X) = {h = U + q E rO I h and U are homeomorphisms}.
LEMMA 2.2. Let h = U + q9 E FI(X). Then
=h1 - U-1
is compact.
PROOF. Set v = h(u), so that
u = h-'(v) = U-1(v) + +(v) = U-1 o h(u) + +(v)
= U-'(Uu + p(u)) + A(v) = u + U- o p(u) + A(v).
Then
(v) = -U- 0 p(u) = -U-' 0 0 ? h-'(v)
and since q9 is compact, it follows that 4 is compact. L

The pseudoindex theories we are going to construct are related to a linear
Tg-invariant subspace H+ C H, H+ E E. The most interesting case occurs wh
both the dimension and the codimension of H+ are infinite. Otherwise, our
pseudoindex theories would give results for semidefinite functionals, but these
functionals could be treated in a simpler and more general way (cf. [4, 13]).
Let 6L be a group of linear homeomorphisms such that
f (a) TgU= UTU VU Et,

(2-4) (b H=9V~{
(b) UH+ = H+ VdU E O
In our applications to the functional (0-1), we shall suppose that H+ is L-invariant
and we shall set
(2-5) te {tL I t E R}.

However, in the construction of our pseu
satisfies (2-4) works as well.
Let 9DTh* denote a class of mappings h such that
(a) h C D 'T i.e. h is Tg-equivariant,
(2-6) (b) h E IF,(H) i.e. h is a homeomorphism of the

form U + g where q is compact, and
(c) U E t.
By Lemma 2.2, the following corollary is straightforward.
COROLLARY 2.3. 6DTh* is a group of homeomorphisms.
If we suppose that F C H+ we set
(2-7) I*= {6D*, i*} where i*(A) = min i(h(A) n H+).

I I I ~ ~~~~h c6 *

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
542 VIERI BENCI
Then by virtue of Corollary 2.3 and proof 1.6, I* is a pseudoindex theory. In order
that this pseudoindex theory be useful it is necessary to show that it is not trivial, i.e.
that there are sets of any pseudoindex.
We shall prove the following theorenm.
THEOREM 2.4. Let H-, H+ E I be linear subspaces of H such that cod(H- + H+)
< +x, dim(H+ nH-) < +0 and F C H+ . Then
i*(H- n Sp) = d [dim(H+ n H-) - cod(H+ +H-)].
Before proving Theorem 2.4, we shall introduce another pseudoindex theory which
will be useful in different situations.
If we suppose that H+ n F = {O}, we fix a constant p > 0, and we set
(2-8) I2* = T i2*} where i* = min i(h(A) n S, n H+).

2 2 ~~h &z 6X
Also in this case, Corollary 2.3 and Proposition 1.6 show that (2-8) defines a
pseudoindex theory. The following nontriviality theorem holds.
THEOREM 2.5. Let H-, H+ E . be a linear subspace of H. Suppose that F n H+ =

{O}, F C H-, dim(H+ nH-) < +x; cod(H+ +H-) < +? and that H- is invariant
for every U E 6L. Then
i2*(H-) = [dim(H+ nH-) - cod(H+ +H-)].
In order to prove Theorem 2.4 and Theorem 2.5 some work is necessary.
LEMMA 2.6. Let V, W E 2 be two invariant subspaces of H such that
FCVCW and codwV<+cc.

If A E 2 is a bounded subset of W, then
i(A n V) 2 i(A) -I codw V.
PROOF. In order to simplify the notation we set k = i(A); k1 - d- dim V, where

V, is the orthogonal complement of V in W.
Now we argue indirectly and we suppose that
i(A n V) < k - k- 1.
By (i-4) there exists a neighborhood N = N;(A n V) such that
i(N) < k - k- 1.
We set
Al = A n N, A2 = A \N.
Then we have
dist(A2, V) 2 a > O.
If P denotes the orthogonal projector on V,, by (i-5), (2.2'), and the above
inequality,
i(A2) < i(PA2) < Idim V1 = k

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Then using (i-3), (i-2) and the above inequalities, we get
i(A) < i(Al) + i(A2) < i(N) + i(A2) < (k - k- 1) + k = k - 1

and this contradicts our assumptions. Thus the lemma is proved. D
LEMMA 2.7. Suppose that H = ED I I Hi where Hi (i = 1,... ,4) are four mutually
orthogonal subspaces of H. Moreover suppose that dim H, < +x for i 2,4 and that
F C H3. Let
{ = D F 9c I T D = Id + g where g has finite dimensional range }

then
(2.9) i(JD(A,) n (H2 E H3)) 2 [dim H2 - dim H4]
where
AP = sp n (H1 E H2)
PROOF. Given D = Id + g, let Wo be a finite dimensional space such that

WO D g(AP) and we set
v = WO nH
and
W = V1 E H2 E H3 ED H4
Then we have
(2-10) (D(AP) n (H2 E H3) D D(Ap n w) n (H2 E H3)

[(Ap n w) n w n (H2 E H3).
By the definition of W
(D(AP n w) c w
so by the above formula, (i-5) and (2-2)
i((D(AP n w) n w) = i((D(AP n w)) 2i(Ap n w)

(2-11)1 +1
)= i(Sp n (v ED H
On the other hand we have that
(2-12) codw(H2 E H3) = dim V1 + dim H4.

Using Lemma 2.6, (2-10), (2-11) and (2-12) we get
i(D(AP) n (H2 ED H3)) i(D(Ap n w) n w) - Icodw(H2 E H3)
2 ( ddim V, + ddim H2 -( dim V, + Idim H4)
(dim H2-dim H4). D

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
544 VIERI BENCI
LEMMA 2.8. Let H, H1 (i =

= D c F1 I - = Id + q' wh
then
i( (DAP) n (H2 ED H3)) d[dim H2 - dim H4].

PROOF. First of all we shall prove that the set
(2-13) K= ?D(AP) n (H2 ED H3) = {[sp n (H, ED H2)] + (D(Ap)) n (H2 ED H3)
is compact. Consider the sequence Un C K. Then we can write
(2-14) Un = Vn + Wn + Zn
with vn E HI, wn & H2, Zn = qg(wn + vn) and vn +

sional and wn is bounded then we can suppose th
ing a subsequence). Since qg is a compact map we c
P, be the orthogonal projector on Hi. Applying
Thus, since P1 Zn converges, we deduce that vn co
and K is compact.
By virtue of (i-4) there is a constant 8, > 0 such that
(2-15) i((D(Ap) n (H2 ED H3))

where
6a = Na(?D(Ap) n (H2 ED H3)).
We set B = (1(DAL/4 ). Clearly we have that

(2-16) (D(A A\B) n (H2EH3) = 0.
We claim that there exists a constant 82 > 0 such that
(2-17) 1P1,4UH 2VuC (A \B) where P1,4 = P1 + P4.

In order to prove (2-17) we argue indirectly and we suppose that there exists a
sequence {Un} C (D(Ap \B) such that P1i4Un -O 0. As in (2-14) we set un = Vn + W, +

zn with v,, C H1, wn E H2 and Z,n = (Un + Wn). Then we have
P1,4 un Vn + P1,4Zn-
Since P1 4un and P1 4Zn are convergent sequences (possibly considering subsequences)
then also vn converges, and since wn converges, un converges to a point u- C ID(Ap \ B
(we recall that (D(Ap \ B) is a closed set since 1D is a homeomorphism). Sin
P1i4U = limn00 P1i4un = 0, then u- e H2 ED H3. Therefore
u E? (DAp \ B) n ( H2 ffl H3 )
and this fact contradicts (2-16). Then (2-17) is proved. Now let g be a finite
dimensional equivariant map such that
1g(u)- (u)l 8 Vu C A where 8 = S min(,

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Such a map can be easily constructed by virtue of (2-1). If we set 4 = Id + g, then

D E J(5is defined in Lemma 2.7) and
(2-18) II 'D(u) - 4(u)fI < 8 Vu F Ap.

It is easy to see that
(2-19) ?(A \B) n(H2DH3)= 0.

In fact, if z FAP \B, by (2-17) and (2-18), we get
IP1 I,4D(Z)II I IP1 ,4D(Z)II - 11 P,4(D(z) - (z

62 > ?
82 - I 'D(Z)-(Z)l I I;z: 2
Using (2-19), we have that
(2-20) (AP) n (H2 E H3) = j(Ap n B) n (H2 E H3).

Also by (2-18) we have that
D(B) c %^,.
By the above formula and (2-20), we get
D(Ap) n (H2 E H3) c 9t1.

Therefore using (2-15), (i-2) and Lemma 2.7 we have that
i(D(Ap) n (H2 E H3)) = i(GX,) > i(D(Ap) n (H2 E H3))

'd-'(dimH2 - dimH4). D
PROOF OF THEOREM 2.4. If H -n F # {0} then, since F is h-invariant
h(H-Snp) f F# 0 Vhc 9Th.

Therefore, by (i-2), i(h(H-nSs) n H+ ) >i(h(H-nSs) n F)= +x and Theo-
rem 2.4 is proved. If Hf-Fn {0}, then we set
H2 = H+ nH-,
H1 = orthogonal complement of H2 in H-,
H3 = orthogonal complement of H2 in H+,
H4 = orthogonal complement of H1 E H2 D H3 in H.
Because of our assumptions F C H3. Then the assumptions of Lemma 2.8 are
satisfied. For every h = U + T E 9*, we have
i(h(H- n Sp) n H+) i(U-' o h(H- n Sp) n U-H+) by (i-5)
i i((Id + U 1 o p)(H- n Sp) n H+ ) by (2-4)(b)
i((Id + u- o p)((HI D H2) n Sp) n (H2 D H3)).
Since Id + U-1 o p F 3, by Lemma 2.8, we have that
i(h(H-nsp) nH + ) >I [dimH2-dimH4] VhFE9t*.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
546 VIERI BENCI
Since dim H2 = dim H+ nH- and dim H4 = cod H+ +H-, by the definition of i*
the conclusion follows. D
PROOF OF THEOREM 2.5. We set
H2= H+ nH-,
H1 = orthogonal complement of H2 in H+,
H3 = orthogonal complement of H2 in H-,
H4 = orthogonal complement of H1 E H2 E H3 in H.
Since F n H+ = {O} and F C H-, then F C H3. Then the assumptions of Lemma
2.8 are satisfied.
For every h = U + qg F 91*, we have
i( h(H-) n Sp n H+ ) = i( H- nh-( (Sp n H+ )) by ( i-5).

By Lemma 2.2, h-1 = U-1 + 4 with 4 compact. Since UH-= H- we have
i(h(H-) n Spn H+) i(H-n (U-1 + f)(sp H+ ))
= i(UH-n (Id + U o f)(sp nH-)) by (i-5)

= i(H2 E H3 n (Id + U o )(Spn H, ED H2)).
Since Id + U o qg C C, by Lemma 2.8, we have that
i(h(H-) n Sp H+) I [dim

d
H2 - dim H4]
- [dim(H+ nH-) - cod(H+ +H-)] Vh E 9Th*.
Since for h = Id, we get the equality, the conclusion follows from the definition of
i2*. F-
3. A deformation theorem. In ?2 we have constructed two pseudoindex theories. If

we want to apply Theorem 1.4, we need to characterize the functions f E C1(H)
which satisfy the property (P) with respect to such pseudoindex theories. This is the
aim of this section. Clearly the main difficulty is to find a function 'q as in (1-9)(b).
As usual this function will be a deformation, i.e. 71(u) = (El(to, u) where (1~ is a flow
on H and to a fixed real number. This flow will be related to a vector field which is a
suitable approximation of the vector field -f'. In our case, the main difficulty in
carrying out this program is the fact that q must belong to O1*; therefore the
construction of the flow GqL(t, u) will involve some technicalities which will be
treated in this section. We shall start by proving some lemmas to be used later in the
construction of 1t1.
LEMMA 3.1. Let X and Y be two Banach spaces and let f: X -- Y be a locally
Lipschitz continuous function. Then any compact set K C X has a neighborhood DL
that f 1K is (uniform) Lipschitz continuous.
PROOF. For each u C K, there exist an open ball B(u, 8(u)) and a constant l(u)
such that
11 f(v) -f(w)l ? l(u)l v - w1, Vv, w C B(u,8(u)).

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
The family of sets {B(u, 2O(u))}UeK is an open covering of K. The

extract a finite covering
{ B( ui, 2 8( u))i} where I is a finite set of indices.

We now set
6L U B(ui, 'S(ui)) and M= sup Ilull.

iC! u Ef(OX)
M is less than infinity b
I=max 4M, max

We claim that
11 f(v) -f(w)l s Ilv - w for each v, w C L.

In order to prove the above inequality we distinguish two cases
First case. II v - w II : 8/2. Then we have
Hf(v) - f(w) f(v) + f(w)Hl <2M? 4M < llv - wll.

6 2
Second case. I v - w I < 8/2. Suppose that v C B(ul, I8(ul)). Then w and v E
B(ui, 8(ui)). Therefore we have
II f(v) -f(w)l I(ui)vII v - w v - w
LEMMA 3.2. Let w: H -* H be a compact operator. Then, given y >
operator 9: H -- H which satisfies the following assumptions:
f (a) j is compact,
(3-1) (b) 45 is locally Lipschitz continuous,
(c) 11Hp(u) - 4f(u)ll < y.

Moreover, if qg is Tg-equivariant, zq can be chosen to be equivar
PROOF. For each w C H, we set LW = {u C H I Iq(u)-(w)ll < y and u-w1

< 11. {DL,W}WeH is an open covering of H. Therefore there exist a locally finite
refinement Ri and points w1 such that Ri C Dw Let pi(u) denote the distance of u
to the complement of Ri. Then pi is Lipschitz continuous and vanishes outside of
'x?. Let
pi(U) = Pi(a)/ I pj(u).

jcl
Since (DLiij is a locally finite covering, for each u F H the deno

a finite sum and the above formula makes sense. Moreover the
continuous and
E/3(u)- 1.
By the definition of the y 's and the RL's, we have that
(KT(u) - q4(w), v)< y1v1, Vu C l Vv F H.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
548 VIERI BENCI
Thus, since the Pi3's va
/,3(u)(K(u) - q(w
If we sum the above expression over the i's, we get
(u) - EP,3(u)q(wj), <yjv, Vu, v E H.
If we set (p(u) = 2,/,(u)q(wj), by the above formula we get

lq(u)-(p u)Hy < y, Vu E H.
Thus (p satisfies (3-1)(c). To prove that (p satisfies (3-1)(a) and (3-1)(b), observe that
for each bounded ball BR C H
q (BR) C {convex hull of T(BR+1)}

Indeed, for u E BR, (~u) is a convex combination of elements w, E BR+1. Since
T(pBR+1) is relatively compact, so is its convex hull. Therefore 5( BR) is relatively
compact and this proves (3-1)(a). Moreover (p is locally Lipschitz continuous since in
each point it is the finite sum of locally Lipschitz continuous functions. Thus also
(3-1)(b) holds.
It remains to prove that (p may be chosen Tg-equivariant whenever (p is Tg-equi-
variant. Suppose we have constructed , which satisfies the (3-1) but it is not
necessarily equivariant. We set
9(u) fTg-'T I(Tgu)dIdL

G
where y is the Haar measure on the group G. Clearly (p is Tg-equivariant. We claim

that it satisfies the (3-1). First of all, let us prove the compactness. For any R > 0 we
set K = (p1(TgBR) =T(BR), Vg F G (TgBR BR because Tg is unitary.) Clearly
9(BR) fTg 'Kd={vf= Tg udlLIuEK}.
Since the map u " JG Tg-'u dpt is continuous, q (BR) is compact and this proves
(3-1)(a). Now let us prove (3-1)(b). Take u F H; the set {Tgu}geG is compact since G
is a compact group. Therefore by Lemma 3.1, there exist a neighborhood 9L
Nd({TgU}geG) and a constant 1 > 0 such that
(3-2) Hp(v)- p(w)H ? tHy - wl lu, v F .

For each u E 6X there exists a neighborhood 6% such that Tgv and Tgw E '
whenever v and w E %0. Then for each v and w F 6% we have
H( (v) - ( (w) 11 fTgj1 1p(Tgv) )- 1(Tgw)] d1i
j 1q(Tg(v)) - T(Tg(W)) d1i
< 1| | Tg(v-w ) |diL by (3-2)

= llv - wll.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Finally we prove (3- 1)(c): using the fact that p is equivariant
11 (u) - U - )2p(U ) T I T1TI(TU) d1- fT (p((Tgu)dI
|I ICTS ( - Tgu)] d1i < f H19p(Tgu) - T(u) ? y.

This completes the proof. D
LEMMA 3.3. Suppose that L + (p ? FO(H) and that the following Cauchy problem
(3-4) {1(t) (t) + (t)),
has a unique solution 1lt (t, u) (1lt (t,*) ? C1(R, C0(H, H))). We suppose that for every
t, Glt(t,) maps bounded sets into bounded sets. Then for every t ? R, Gtl(t, u) has the
following form
6?t(t, u) = eLu - g(t, u)
where g(t,*) is a compact operator for each t C R.
PROOF. Without loss of generality, we suppose t > 0. The operator qLt(t, u)

satisfies the following identity which is the Volterra form of the equation (3-4):
G?t(t, u) etLU + te(tes)L9p(G?I(s, u)) ds.

0
We have to show that the operator
g(t, u) =-te(t-s)L p(6?1,s, u)) ds

0
is compact. By our assumptions, for every R > 0, there exists R > R such that
?(s, BR) C BR Vs C [O, t]

then
p(G?L(s, BR)) C Rp(Bff)

where (p(BR) is a precompact subset of H. Now we set
K U[ e(t-s)L(p(B-)]
SepO, t]
K is a compact subset of H. In fact let {vn} be a sequence in K; then there exist

two sequences {Snj C [0, t] and {wn} C cp(Bff) such that
vn = eUt-Sn) L(p( Wn)
Now let {Sn,k} and {wnfk} be two subsequences such that
{Sn)k} -S & and Tp(wn) -*z i T(BR)

then
Vn- e <i z F K.
Therefore K is compact. Now set
K1 = {convex hull of K)

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
550 VIERI BENCI
By a well-known theorem K1 is compact. Since for every u C BR,
fe(t-s)L T(91(s, u)) ds F K1
-g(t, u) C K. Therefore g(t, ) is compact. D
THEOREM 3.4 (DEFORMATION THEOREM). Let f C C'(H) be a function which

satisfies the following assumptions
def
(f1) f(u) = 2(Lu, u) + 1D(u) where L is a bounded selfadjoint operator and Tp-='
is compact.
(f2) f satisfies P.S. in [co, cO ] i.e. any sequence {um} C H such that f(um) -* c C
[co, c.] andf'(um) -O 0 as m -- +x, has a convergent subsequence.
Then Vc C [co, coj, VV = Nd(KC), 3c > 0 and 3rq = J,6L + 4 (where 8 > 0 is a
constant and 4 a compact operator) such that
( 3-6 ) C1(C OC'-?

where K., (A c are defined by (1.8).
Moreover, if
(f3) f is Tg-invariant
then q is Tg-equivariant, i.e. 71 F 1*
PROOF. Because of (f2), Kc is a compac

constant such that DL C BP . We set
d
4(1 + 211LIIpi)
Because of (f2), there exists c > 0 such that
(3-7) lf'(u)l I foru F ZC,E\6d/2
where Zc ? = f -'([c - , c + -]) and (Ld/2 = Nd/2(K

In fact suppose that (3-7) does not hold. Then th
and a sequence c,7 O such that un C Zc ? \ d/2 a
have that f(un) c and 11 f'(UJ)II - 0, and by (f2),
Then we have that f(u-) = c and f'(u-) = 0 and this
Thus (3-7) holds. Also we can suppose that
(3-8) < 8/5.

We now set
(3-9) -y = min( e/28; 21 / V 8) -

Now, by virtue of Lemma 3.2, there exists a compact, locally Lipschitz continuous
operator (p such that
(3-10) II(pH(u)-(u)11 y, Vu E H.
By the above inequality we get
11 P(U)H 2 - 2KT(u), q (u)) + H (U)H 2 ? 72

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Then
(3-11) KT(u), (p (u)) 2( Up(u)H 2 + q(U)H 2) - y2/2

I(p(u)II _j(H (u)jj - y2 /2.
Now set x(t) = min(l, I /t) and
V(u) (U (u)
It is easy to check that V is a compact, locally Lipschitz continuous operator and
that
(3-12) 11 V(u)H1 < 1 + 11 Lu 11, Vu F H.

We now claim that
(3-13) (f'(u), Lu + V(u)>2 R Vu E ZC',? \%d/2'

In order to prove (3-13) we distinguish two cases.
First case. Suppose that I I ((u)H I 1 + IH L(u)IH. Then we have
(f'(u), Lu + V(u))= (f'(u), Lu + g (u))
= (f'(u), Lu + c(u))- (f'(u), cp(u) -(u))
I If'(U)112 -yIf'(u)II by (3-10)
2 3c/8 by (3-7) and (3-9).
Second case. Suppose that
(3-14) l~(u)jj > 1 + IIL(u)HI.

By the definition of V we get
(f'(u), Lu + V(u)> = (u + 9(u), Lu + I +
= 11 Lu 112 +I + I Lu ) (Lu, (u)) +

+~~~~1 T e) ((U), (u)
I+ I Lull II * (u*
II+Lull
( 1 )2(UL
- (111+ )1
I Lull)
(U)1- -()IlLull
Y y(- - IlLull 1q(u)ll
1 + ll Lull y 2
+ (1 + IlLuII)IIqp(u)II - ll~)l --by (3-1 1)
II Lull12 _ II LullI - II Lul112 - II LulIIl 11 q(u)ll1
+ 11q(u)ll II Lull + 11q(u)ll - Y- by (3-14)

2
llp(u)ll - IlLull- Y2
211 (u)ll - -y -11 Lu 11- y2/2 by (3-10)
21-y-y2/2 by(3-14)
25/83-c /28-1/4 4c/ 2 3 3R/8 by (3-8) an

Then (3-13) is proved.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
552 VIERI BENCI
Now consider the following Cauchy problem in the Hilbert space H
(3-15) dt -Lt(t)-
t(O) = u.
Since V is locally Lipschitz continuous, (3-15) has a unique local solution. Moreover,
by (3-12), we have that
(3-16) 11Lu + V(u)11 < 1 + 211H11 11u11.

Then by standard results on O.D.E.'s it follows that ((t) is defined for all t C R.
We shall denote by Q(tl(, u) the flow relative to the Cauchy problem (3-15). Now
we shall prove that
(3-17) 6l(t, u) C H\ Ld/2 for each u C H\ Ld and for each t c [O, 8].
In order to prove this we argue indirectly.
Suppose that there are points to ti C [0, 8] (to < t,) such that
6(to, u) a3d; 6tll(, U) F ad2 and 6(t, u) F d \d/2
for all t C [to t1].Then we have
d/2 < G?1t(t , u) - 6?t(to, u)1 d| f 'Ii?I(t, u) dt
f|{-LQI(t, u) - V(9I(t, u))} dt
s L 6LQ((t, u) + V(9L(t, u))1 dt
? f'(1 + 211 L 11 1 (t, u)H) dt by (3-16)

to
s (1 + 211 L 11 p )(t1 - to) (1 + 211 L 1)&1 )S.
The last inequalities have been possible since (Ld C BP,. Then using the definition of
8 we get
dl 2 *(I+21ILHIIp1) =d14.
4(1 + 2ILlpl) p1)
This is a contradiction, then (3-17) is proved.
Next we shall prove that
(3-18) t(6, u) e/ C-E_ VU E i9C+F \ (d

First of all, we remark that, by virtue of (3-13), we have
(31) dOt (q(t, U)) ( t (t, u)), -(t ?(t, u)) < -3e/8.
In order to prove (3-18), we have to show that
(3-20) there exists to E [0, 8] such that f(<l(to, u)) < c -

vu c?d(F \Xd

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
and
(3-21) if 3t1 E R such that f(tl(t , u)) = c-, then f(tl(t,

-c-E V t tI.
First we prove (3-20). We argue indirectly and we suppose that for each t F [0, 8],
f(9L(t, u)) > c - E then we have
(3-22) 2? > -[Af(9?(, u)) -f(G?(0, u))] -f tf(9i(t, u)) dt.
By (3-17), QEt(t, u) X (%d/2. Moreover, by (3-19), the function t H-4f((t(t, u)) is

decreasing. Then we have that 1Et4t, u) E ZC? \ 6xd/2* Therefore, using again (3-19)
and by (3-22), it follows that
2c | (3/8) dt = 3e.
This is a contradiction, thus (3-20) is proved. (3-21) follows directly from (3-19).
Then (3-18) is proved. By Lemma 3.3, we have
6?t(t, u) = e-tL + g(t, u)

where g(t, ) is a compact operator for each t F R.
If we set
q(u) = t(S, u)
we see that q has the desired form with A(u) = g(8, u). Moreover, by (3-18), (3-6
follows. Then the first part of the theorem is proved.
If f is Tg-invariant, then L and (p are Tg-equivariant. Then by the last part of
Lemma 3.2, (p can be chosen equivariant. Since Tg is a unitary representation, the
functional
x 149(U)lI
1 + IILu
is Tg-invariant. In fact
H I9 (Tgu)H A IiTTg9(u)H 1A = __9(_ A

XV I + IILTguH ) X ( + I1TgLu ) ( I + IILu2 J
Then V is a Tg-equivariant operator. Therefore the operator 6(E(t,) and conse-
quently , are Tg-equivariant. D
By the deformation theorem the following corollary follows.
COROLLARY 3.5. Iff satisfies (f1), (f2) and (f3) then { f, co, cO } satisfy the property
(P) with respect to {E, O1U*}.
PROOF. By (f3) it follows that LK and Kc E E. By (f2), it follows that Kc is

compact. Then (1-9)(a) is satisfied. (1-9)(b) follows directly by Theorem 3.4. D
4. Some abstract multiplicity theorems. In this section we shall use the theory
developed in the previous sections to give a lower bound for the number of critical
values of the functional (0.1) in several different situations.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
554 VIERI BENCI
THEOREM 4.1. Let H be a Hilbert space with an index theory I related to a

representation Tg which satisfies the dimension property.
Let f be a functional which satisfies (fI) (f2) (f3) of Theorem 3.4. Moreover suppose
that there are two invariant spaces H+ , H- C E and a constant p > 0 such that
p(a) F C H+,~ LH+ = H+,'

(f4) (b) f(u) >co Vu E H+,
l(c) f(u) < cO, Vu E H-nsp.
Then, if I* is the pseudoindex theory (2-7) and if the integer
(4-1) k =Jj[dim(H+ nH-) - cod(H+ +H-)],
is well defined and positive, the numbers
Ck- inf supf(u) fork = I,...,k

l*(A)-k uCA
are critical values off and co < cl < < c1 - c,,. Moreover if
fc = Ck ck+r
(4-2) q then
ti(Kc) r + 1.
PROOF. We apply Theorem 1.4. By (fl) (f2) (f3) and Corollary 3.5, the triplet
{f, co, cO,} satisfies the property (P). Now take A C L with A C eco; then by (f4)(b),
A n H+ = 0. Therefore i*(A) = i(A n H+) 0 and (1-10)(b) is satisfied. Setting
A =S, nH-, by (f3)(c) we have A C &c, then by Theorem 2.4 i1(A) =

shows that (1-10)(c) holds and the conclusion follows. O
We now give a theorem analogous to Theorem 4.1 which makes use of the
pseudoindex I*.
THEOREM 4.2. Suppose that all the assumptions of Theorem 4.1, but (f4) are
satisfied. Instead suppose that
(a) FnHH+ {0}; FCH-,LH-=H-,
(f4) {(b) f(u)>co VuEH+ nSp,

1(c) f(u) <C. Vu E H-.
Then, if I* is the pseudoindex theory (2-8), and k is well defined by (4-1) and positive,
the numbers
(4-3) Ck - inf sup f(u)

k 1*(A)-k UCA
are critical values of f and c < cl - < ck < c,,.

PROOF. We apply again Theorem 1.4. By (f1), (f2) and (f3) and Corollary 3.5, the
triplet { f, co, cC,,} satisfies the property (P).
Now take A e L such that A C 6B ; then by (f*)(b),A nH+ nsp = 0. There-
fore minh Ti(h(A) n H+ n Sp) = 0 and (1-O0)(b) is satisfied. Moreover, by (f*)(a),
the assumptions of Theorem 2.5 are satisfied. Then i*(H-) = k.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Therefore setting A = H-, (1- 10)(c) follows. O

REMARK 4.3. If f satisfies the assumptions of Theorem 4.1, thus -f essentially
satisfies the assumptions of Theorem 4.2 if we interchange H' and H-, and co
coo.
Those two theorems are related to Theorem 2.19 of [3] and Theorem 12 of [8] for
indefinite functionals. Actually in [3] and in [8] only even functionals have been
considered, and of course (f ) is not required since the functionals are semidefinite.
Even if those theorems sound different, in a deeper analysis the similarity is evident
(also cf. Theorem 2.19 of [3] with Corollary 4.5). While Theorems 2.19 of [3] and 12
of [8] apply to two completely different classes of problems we see that in Theorems
4.1 and 4.2, those two different situations are unified.
REMARK 4.4. Theorems 4.1 and 4.2 are not sufficient to guarantee the existence of
at least k (orbits of) critical points. In fact it might happen that Kc n F #+ 0. Then
i(K,) = +oo but it might contain only one fixed point (cf. (2.3) and the following
remarks). Thus to have a multiplicity result we must add some assumptions
depending on the nature of F and f. For example in Theorem 4. 1, we could assume
inf f(u) = v > cOO

uCF
and this, of course, implies Kc n F= 0 and the multiplicity result. A similar

argument is used in the proof of Theorem 0.1 given below and Theorems 5.1 and 5.3.
More subtle arguments for proving Kc n F = 0 are given in the applications to
Hamiltonian systems in ?5.
PROOF OF THEOREM 0.1. We shall use Theorem 4.1 (but we could use as well
Theorem 4.2).
(f 1 ) is equal to (f 1 ) and (f2 ) implies (f2)

(f3) implies (f3) where the group representation is the one described in Example I
of ?2.
For such representation F {0}, thus (f4)(a) is satisfied.
(f4)(b) and (c) follow by (f4).

Thus Theorem 4.1 can be applied.
Then in order to get the conclusion of Theorem 0.1, it is necessary to prove that
KC n F = 0 but this follows from Remark 4.4. O
By Theorem 4.2, the following corollary follows.
COROLLARY 4.5. Suppose that f satisfies all the assumptions of Theorem 4.2, except
(f4 )(c). Moreover suppose that
(a) dim(H )' - +xo,

(4-3') (b) f(u) is upper bounded on (H?) E Vdk
I for every Vdk E E, Vdk C H+.
Then (4-3) defines a critical value for each k E N, Ck ck+ 1, and (4-2) holds.
PROOF. It is enough to choose k E N arbitrarily and to show that the ck's are
critical values for k < k. In order to do this, we set H- (H+ E) Vdk and apply
Theorem 4.2. O

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
556 VIERI BENCI
From Theorem 4.1, we shall deduce another theorem which is less general, but it
can be applied more directly to some concrete problems we shall consider later. First
of all we make the following assumption on L:
0 4 Ue(L) where U,e(L) denotes the essential spectrum of L

(f5) (this means that 0 is either an isolated eigenvalue of finite
multiplicity or it belongs to the resolvent).
Then we can decompose H in the following way
H = V-ED V ED V+
where
V? = kerL;
(4-4) KLu, u)< allull2 Vu C V-,

KLu, u) > /uH112 Vu C V+,
where a < 0 < / are suitable constants.
If Q is a compact operator, by well-known theorems, (see e.g. [17]) Ue(L)=
Ue(L + Q). Therefore H has the following decomposition:
H= W-EDW ED W+
where
W? = ker(L + Q),
(4-4') ((L + Q)u, u)< 'llullH2 Vu E W-,

((L + Q)u, u)> /,'11u2 Vu C W+,
where a' < 0 < /' are suitable constants.
The following theorem holds.
THEOREM 4.6. Suppose that f satisfies (f1), (f3), (f5) and
(a) D (0) = 0.
(b) every sequence {Un} such that f(Un) -) c < 0 and
II f'(Un)II -*asnx-- oo is bounded.
(4-5) (c) there exists a compact operator Q such that
(p4(u) = Qu + o(llul) whereT = V.
(d) lim (u) = -.
(e) FC V+.
If we set H+ V+ and H- W -, then the same conclusion of Theorem 4.1 holds with
co and co, suitable negative constants. In particular we have
k = dim(V+ n w-) - -cod(V+ + W-)
- dim(V+ n w-) -_dim[(V-EDv ) n (w+ ED w)]

where the above spaces are defined by (4-4) and (4-4').

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
REMARK 4.7. If L has discrete spectrum and it commutes with Q, then k is easy to
compute.
In fact, let {ek} keN be a common set of eigenvectors of L and Q corresponding to
the eigenvalues {Xk} of L and {qk} of Q.
Then
v n w =Sp{ek Xk > 0; Xk + qk < 01
- Sp{ek 0 < Xk < -qkl
and
(v-D vo) n (w+ G w0) Sp{ek I Xk ? 0; Xk + qk 0)

= Sp{ek I 0 Ak > -qkl
Therefore
k #k E Z I ? < Xk < -qk - #{k EZ 0 2 Xk -qk

Before proving the Theorem 4.6 we need the following lemma.
LEMMA 4.8. If f satisfies (fl), (f5) and (4-5)(b) then it satisfies (f2) in [co, co]
(co < coo).
PROOF. Let {un} be a sequence such that f(un) -- c and

(4-6) f (Un) -->o.
By (4-5)(b) it is bounded.
By (f5) there is a compact operator Ko such that
0 4 a(L + Ko).
By (4-6) we have that the sequence
def
(4-7) n- (L + KO)Un - KOUn + (P(Un)

is converging to 0.
Since {un} is bounded, -KOun + (p(un) is pre-compact, then the set
D = {Koun - (P(Un) + vn I n E NJ
is compact.
By (4-7), (L + KO)un E D, then Un e (L + Ko)-D. Since (L + Ko)-1D is a

compact set, { un } has a converging subsequence. O
PROOF OF THEOREM 4.6. We shall check all the assumptions of Theorem 4.1 with
H+ = V+, H-= W-, coo = -y,/2 and c0 = 72-1 where Y2 < 71 < 0 are constants
to be determined later. Because of Lemma 4.8, (f2) holds. Then (fI) (f2) and (f3) are
satisfied. Let us prove (f4). (f4)(a) follows by (4.5)(e). Before proving (f4)(b), we
shall prove (f4)(c). We use the Taylor formula forf at 0:
f(u) = 4KLu, u)?+ IQu, u)? o(H1uH12) for lull - 0.

Then for u e W- we have f(u) s a'lHuI2 + o(2HuH12) where a' is the constant
appearing in (4-4'). Since a' < 0, there exists a constant p > 0 and y1 < 0 such that
f(u)< yl VuFEw-nsp.
Setting cO =y1/2, (f4)(c) is satisfied.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
558 VIERI BENCI
In order to prove (f4)(b) we first observe that, by virtue of (4-5)(d), there is r > 0
such that
II(p(u)II < (//2)u11 u 112 whenever II u II > r

where /3 is the constant appearing in (4-4).
Moreover, since (p is compact, there is M > 0 such that
(p(u)ll < M if 11 u 11 < r

thus
(p(u)H1 < M + //211 u 11 Vu E H.

Using the above formula we get
D(u) '=| (p(tu), u) dt|
?| f{M _u11u + Pt 11lu2}dt
<MIIKUII + 3HIUH112 Vu e H.
Then by (4-4) and the above inequality, for every u E V+, we get
f(u) =KLu, u) + ?(u)

?(1/2)11U112- (U)(1) (/4)11U112-M IUI.
This implies that f is bounded from below on V+ and we can set
72 = inf f( u) and co = y2-w with w > 0 such that Y2-w < cW .

ue V+
Then by Theorem 4.1 the conclusion follows. O

REMARK 4.9. Under the assumptions of Theorem 4.6, we could equally well have
used the pseudoindex theory I2* and Theorem 4.2 in order to get similar results.
REMARK 4.10. If we strengthen (f5) in the following way
(fs ) 0 E a(L)
then, by (4-5)(d), we can easily deduce (4-5)(b).
(ft ) is a nonresonance condition at ox.
Using this remark we have the following corollary.
COROLLARY 4.11. If f satisfies (f1), (f3), (f ), (4-5)(a) (c) (d) (e), then the same
conclusion of Theorem 4.6 holds.
REMARK 4.12. If (f4)(a) (or equivalently (f *)(a) or (4-5)(c)) does not hold, then the
pseudoindex theories studied in ?2 cannot be applied directly, and a general way to
deal with such problems has not yet been developed. However, in [6], a method has
been introduced which permits us to treat semidefinite functionals even when (f4)(a)
does not hold (cf. Remark 6.9).
5. Applications to asymptotic linear Hamiltonian systems. Let H E C2(R2n, R) and

consider the Hamiltonian system of ordinary differential equations

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
where p and q are n-tuples and denotes d/dt. Setting z = (p, q) and J = (? -Id)
where Id denotes the identity matrix in RW, (6-1) becomes
(5-2) 2= JHz(z)
where Hz = aH/az. We are co

We suppose that H satisfies the following properties
(H1) HZ(z) = Az + o(| Z I) for z +xo,

(H2) Hz(z) = Bz + o(| z l) for z 0,
where A, B: R2n R2n are two linear symmetric operators. Our aim is to give a
lower bound for the number of nontrivial T-periodic solutions of (5-2) by ap-
propriately comparing the operators A and B and the period T.
More precisely we are going to define an even integer number 9(TB/27, TA/2T)
which will give this lower bound. Given two hermitian operators A, B: C2n C2n
we set
N(A) = number of negative eigenvalues of A),
N(A) = number of nonpositive eigenvalues of A),
Uj(B, A) = N(iJ + B) - N(iJ + A),

+0o
4(B, A) = E 4J(B, A).

j=-oo)
We observe that i(B, A) is a finite number. In fact, forj big enough N(iJ
N(iJ + A) = n and this implies that Uj(B, A) = 0 except for a finite numb
Now we can state the main theorems of this section.
THEOREM 5.1. Suppose that H satisfies (H1), (H2) and

(H3) (Az, z) > O Vz E R2n \ tO}
(H4) H(z) > O Vz E R2n
(H5) (TJA/27) n iz 0.
Then equation (5-2) has at least 2 &(TA/27, TB/27) nonconstant T-periodic solutions
whenever i(TA/27, TB/27) > 0.
(H3) is a technical assumption which will be discussed in Remark 5.9.

(H4) is a condition which assures us that the periodic solutions we shall find are
not constant (cf. the proof of Lemma 5.8).
If (H4) is dropped, then we conclude that there are nonzero periodic solutions but
they could be constant; therefore we lose information about their number (cf.
Remark 4.4).
(H) is a "nonresonance condition at ox" and it expresses the fact that the
linearized equation at ox, i.e. z = JAz has no T-periodic solutions. If (H5) does not
hold, Theorem 5.1 is not any longer true unless we add some other conditions on H.
For example we can add a condition which corresponds, roughly speaking, to the
Landezman-Lazer condition for elliptic equations as is shown by the following
theorem:

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
560 VIERI BENCI
THEOREM 5.2. Suppose that H satisfies (H1), (H2), (H3), (H4) and
(H) V4(z) is bounded where V(z) = H(z) - I(Az, z).
(H7)V(z) ---+oo asI z +-o.
Then the conclusion of Theorem 5.1 holds.
Other assumptions can replace (H5) as the theorem below shows.
THEOREM 5.3. Suppose that H satisfies (H1), (H2), (H3), (H4) and
(H8) there is R > 0 and p E (1, 2) such that pH(z) > (z, Hz(z)) > O for I Z I R.
( ) i lzl 00|Z I- I HAz() I c -Y C 2-
(H1O) there are constants a, > 0 and a2 > 0 such that H(z) > a I z [P - a2.
Then the same conclusion of Theorem 5.1 holds.
REMARK 5.4. In order to give a feeling for the meaning of ,(.,-) we shall indicate
some of its properties even if they will not be used in the proofs of the theorems.
First of all, we observe that & corresponds to the number 4k defined by (4.1) as it
will be shown in Lemma 6.6.
Moreover it satisfies the following properties:
(#1) if &(XB, XA) > 0, then &(XA, XB) s 0.
(p2) &(XA, XA) s 0.
( 93) if B and A satisfy the nonresonance condition i.e.
u(XJA) n iz = u(XJB) n iZ = 0 (cf. (H5))
then
,&(XB, XA) = -&(XA, IXB).

(4) if a(JA) n iR = a(JB) n iR = Othen &(XB, XA) = &(XA, XB) 0.
(95) *(.,) is a symplectic invariant in both its arguments i.e.
'&(XSTAS1, XS2TBS2) = &(XA, XB) VS1, S2 e SP(2n).

Amann and Zehnder in [1] have introduced a symplectic invariant Ind(A, B, a) with
a real number. We have
(p6) if XB and XA satisfy (H5) then
Ind(B, A, A-) = (XB, XA)

and in general we have
I Ind(B, A, X-1) i5 (XB, XA).

PROOF. ( &l) and (p2) follow directly from the definition.
Let us prove ( I3 ). SinceJ2 -Id we have that
ker(ijJ + XA) = {v E C2n I XJAv jv}.
Therefore if (H5) holds,
ker(ijJ + XA) { tO} and N(ijJ + XA) = N(ijJ + XA).

For the same reason N(ijJ + XB) = N(ijJ + XB).
Then (#93) follows from the definition.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
If a(JA) n iR = 0, then ker(iJ + XA) = {0} VX E R. Then, since the operator

ijJ + XA is not singular for any X e R, the function
X i-N(iJ + XA)= N(J + XA)

is constant. In particular it equals n since for X = 0 we have that N(iJ) = n. Arg
in the same manner we get that N(iJ + XB) = N(iJ + XB) = n. Then (t$4) fo
Let us prove (15). It is known that if Q: C2, __ C2, is any hermitian operator
S is any nonsingular operator
N(Q) = N(S*QS) and N(Q) = N(S*QS)

where S* is the adjoint of S.
If S E Sp(2n), then ST = S* (the transpose is equal to the adjoint since S has real
coefficients). Then using the above formula with Q = ijJ + XA and Q = iJ + XB,
we get
N(jJ + XA) - N(11JS[ + XSTASI) = N(iJJ+ XSTAS,) VS, E Sp(2n)

and
N(ijJ + XB) = N(J + XS2TBS2) VS2 E Sp(2n).

By the definition of t$(-, -) the conclusion follows.
(p6) follows from the characterization (5-16) of the eigenvalues of iJ + XA and
easy computations.
REMARK 5.4'. If A and B commute with J the number t$(XB, XA) is expressible in
a form which is easy to compute. In fact let p I ... M2n be the set of eigenvalues of A
corresponding to the eigenvectors vI,... , v2 n. Then the vectors Vk + iJvk are eigen-
vectors of the operator iJ + XA and they correspond to the eigenvalues j + Ayk
(k = 1,...,2n). Then setting = {1,. . . ,2n} we get
N (ijJ + XA) = #{k E Jq 1j + XMLk < 0}.

If the k5s are the eigenvalues of B, arguing in the same manner we get
N(ijJ + XB) = {k e ' Ij + Xvk < O0

Then, for X > 0, we have
{k E q I Xpvk < j= {k e @ I Pk < Pk; XMk ?< j}

U {k e @ I Pk < Pk; Xvk < -j -< Xuk}
U {k ( @ I Pk -> Mk; XVk < j}
and
{k E XM I X,tk k -j} ={ k |>k; Xk k < -1

U { k |> Itk Xk X -jI t kk}
U {k E Pk < Mk; XMk ? -11
Thus
ij (XB, XA) = # {ke PVk < M k and Xvk < -Xk}

# {k e M I Ik < Pk and Xk t -j< XVk}.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
562 VIERI BENCI
Therefore we have that
i&(XB,XA)= #{(j,k) EZX IVk<k andXvk<jX<Ak}

-?{(j, k) e Z X @ I Pk > [k and XMk ij < vk}
We remark that the number #(XA, XB) does not depend on the order of the Mk
and the vk'S. If it is possible to order them in such a way that Pk < [k for eac
k E @, then
4(XB, XA) > 0 and lim (XB, XA) = +o.

x-A +oo
This fact has an obvious interpretation in terms of the number of periodic solutions
of (5-2).
In order to prove Theorems 5.1, 5.2, 5.3 we shall use Theorem 4.6 (or Corollary
4.1 1). Before doing this some work is necessary.
First of all, making the change of variable
(5-3) t H -t = X- t, (5-2) transforms to 2 = XJH,(z)
and we seek 27-periodic solutions of (5-3) which, of course, correspond to the

T-periodic solutions of (5-2). We shall treat (5-3) in the Hilbert space H=
H1/2(S1, R2n), i.e. the space of 2n-ples of 27"-periodic functions which possess
square integrable "derivative of order ". Perhaps the simplest way to introduce this
space is as follows. Let (C??(S1, R2n) be the space of 27-periodic n-ples of (C?-func-
tions. If z E C??(S1, R2n) it has the following Fourier expansion
a I 4o
(5-4) z(t) = 1 + (aj cos jt + - j3s
22 7 T j=O
which in complex notation becomes
1 +00
(5-5) z(t) a e'j' with a = a- C2.
H112(S1, R2n) is the closure of C((S', R2n) with respect to the
+00 ~~~~~~~~- 1/2

(5-6) llz ( I)( 2 12)
j=O
+oo 1/2
= E (I +Ij 1) I
j=-oo
H1(S, R2n) and L2(S1, R2n). From now on HI/2(S',R2n) will be denoted by H. On
H1/2(S1, R2n) can also be obtained by interpolation from the Sobolev spaces
H a "canonical" representation of the group S1 acts. If s E Sl (s is thought of as a
real number in [ 0, 27T)) we have the representation
T,z(t) = z(t + s)
here the" +-" is the operation of the group Sl; i.e. the sum modulo 2 7.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
In terms of the Fourier expansion (5-5), the representation T7 has the following
form:
+-00
Tsz(t) = E (aje J)e'j

j= -00
i.e. the Fourier expansion is a decomposition of T7 in its irreducible components (see

e.g. Weyl [18]).
We shall denote by I ={E, 6X, i} the index theory relative to the representation
Ts described in II Example of ?2. The 2T-periodic solutions of (5-3) correspond to
the critical points in H of the functional of the action:
f(z) = _f2N(p, 4)Rw - XH(z)} dt

(5-7) 01
(5 f)2W{2 I(J?, z)R2n
(We have put the minus sign in front of the functional in order to make the notation
consistent with the abstract theorems of ?4.) If we set
(5-8) V(z) = H(z)-'(Az,z) and ?(z) =fA| V(z)dt
then we can write
(5-9) f(z) = 2 (Jz + XAz, z)R2ndt + ID(z).
Let L, qp: H -- H be the operators defined by the following formulas:
<Lz, v>= (=Jz2 + XAz, v)R2n dt Vv E H,
<Qp(z), v>= Xf (Vz(z), v) dt Vv E H,
then we have
f(u) = {KLu, u)>+ ??(u) and f'(u) = Lu + p(u).

The following lemma holds.
LEMMA 5.5. If the Hamiltonian function H satisfies (H1) and (H2) then f (defined by
(5-7)) satisfies (fI), (f3), (f5) (4-5) (a), (d) and (4-5)(c) where Q is defined by the
formula
(5-9') (Qz, v)>= Xf (Bz - Az, v) dt.
PROOF. By the Sobolev inequalities and standard arguments, it follows that

P = ' is compact (see e.g. [7]). Then f satisfies (f ) and since L and (p are
equivariant with respect to T, f satisfies (f3).
In order to prove (f5), it is enough to observe that the essential spectrum of the
operator defined by the bilinear form
(z, v) 1 ( JdtZ v dt
J\dt'

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
564 VIERI BENCI
is just { + 1, -1}. Since L is a compact perturbation of such operator, by well-known

theorems (see e.g. [17]), it follows that
ao(L) = f {1,-1}.
(4-5)(a) follows by the definition of (D.
Let us prove (4-5)(d). By (HI), for every - > 0, 3M > 0 such that
IHz(z) - Az e z I +M.
Then, by (5-8), using the H6lder inequality we get
f2|T( vz( z ), v ) dt| =f| | (Hz(Z) - Az, v) dt
<? e f2 z I I v I dt| + 27TMH1v11
< -IIzHl Ivll + 27TM1vH

thus
lim 1(P()1 - lim sup <P( ) 0

ll zll+- oo llZll lzl'-fI llv,,=, llZll
= lim sup -| (Vz(z), v) dt

ll z II llvll=lV llZll
( 27XM
? lim (tX? + )X= .
Ilzll- -I-oo llZH
By the arbitrariness of c, (4-5)(d) follows.
It remains to prove (4-5)(c). By (H1) we have that Ve > 0, 381 e (0, 1) such that
I Hz(z) -Bz I < -- I z I if z 1< -

On the other hand, by (H2), 382 > 1, such that
I H(z) -Bz < (I A I + IBI + 1) Iz for I Z I 82

and by the continuity of Hz(z) there is a constant a1 > 0 such that
IHz(z)-BzI<a1 for3 S 1I ?2.

By the above inequality we get
IHz(z)-BzI<eIzI +a2Iz12 witha2 = max(a 132, 1AA - -BI +1)

for all z E R2n.
Then we have
|7(Hz(z) -Bz, v) dt <?f Hz(z) - Bz v I dt
?t| lz I j dt + a2f IZ VI dt
<?c-ZII * llvl + a2(f Izv4dt) 1/2
< H H II || II ||lV a2 || Z || L4 - ||VII

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
where 1 1 L4 iS the norm in L4(SI, R2W). By the Sobolev ine

Then we get
f 2(Hz(z) - Bz, v) dt < ell z 11 -11 v I + a2a 2 11 z 112 11 v
Then using the above formula we have
I <p(u) - Qu, v)j= X f2 (Vz(z) - (B - A)z, v) dt
< X 2r(Hz(z) - Bz, v) dt by (6-8)
? 11 z 11 -I11 v 11 + Xa2a3211z 12 l v II.
Then
ir 11q(z) -Qz1 = I n sup (qp(z) - Qz, v)

llzll-o ZI llzll-o llvll=l l
By the arbitrariness of c, (4-5)(c) follows. O

In order to apply Theorem 4.6 or Corollary 4.11, we need the following lemma.
LEMMA 5.6. If H satisfies (HI) and (H2), using the notation (5-8), (5-9), (5-9'), (4-4),
and (4-4') we have
dim(V+ n W-) - cod(V+ + W-) = i(XB, XA).
In order to prove Lemma 5.6, we shall study the structure of the spaces defined by
(4-4) and (4-4').
We start this program by studying the spectrum of the differential operators
E =(f~Z
Ji d?+
dt j XA
and
Ez= (J d XBz
regarded as (unbounded) selfadjoint operators in L2(S1, R2n) sin
(a) (Lz, v> =f2 (&z, v) dt Vz & D(E), Vv

(5-10) 0,
(b) ((L + Q)z, v)= f (&z, v) dt Vz E D(e), Vv E H.
0
Let us consider the following eigenvalue problem
(5-11) [ C+ 2 A)u = u, i = ,j E Z.
The operator ijJ is a Hermitian operator, so ther

(k = 2... . ,2n; j E Z) of eigenvectors correspondin
is an eigenvector, then U-k is also an eigenvector corresponding to the same
eigenvalue.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
566 VIERI BENCI
We claim that U]k cannot have all real components forj =# 0. In fact we have
ij(jUjk, Ujk) + (AUjk, Ujk) = P'k( Ujk, Ujk)-

In the above equality, the second term and the right-hand side term are real;
therefore ij(JUjkk U1k) is real. Therefore, for j 7# 0, (JUjk, Ujk) must be

imaginary number, and this contradicts the fact that all the components of U1k are
real. Then it follows that Ujk and U-k are linearly independent. Therefore every
eigenvalue has even multiplicity (for j 0# O). Also we can order the Ujk'S in such a
way that U.k corresponds to the same eigenvalue U.k as Uj k+n for k = ... . . n.
Also we can suppose that
(5-12) u1k = uj,k+n for]j # 0.
Fork=I,...,2nandjEZweset
(5-13) Ujk = Vjk + lWjk

Then by (5-12),
(5-14) (a) Vjn+k =_ Wk for j :# 0.

(b) Wl,,n+k =Vk
Forj = 0, we choose all the UO,k real, then we have

(5-15) Uo,k = VO,k, WO,k ?0
Using elementary linear algebra we have
LEMMA 5.7. The V. k'S are an orthonormal basis for R2n for each j E Z. Similarly the
W, k's are an orthonormal basis for R2n for each j E Z \ {0}.
PROOF. For j =# 0 take a vector a E R2n. In particular a E C2n, so there are
numbers ak =fk ? 1Yk E C (I3kI Yk E R) such that
2n 2n 2n
a E akUjk = E (hkV]k - YkWjk) ? i E YkVjk + fkWjk

k=1 k=1 k=1
Since we have assumed a real, the last term vanishes and using (5-14) we have
n n
a = kV
k=1 k=1
This proves that th

The same argumen
the conclusion follows by (5-15). 0
Using (5- 11) and (5-13) we can write a formula which involves only real compo-
nents:
(5-16) | jJWjk + XAvjk = IjkVjk,

(-JjV,k + XAWjk = [_kWjk-
The above formula could give an alternative definition of the V, k'S and the W, k
now set
(5-17) e,k= (VjkCOS

.2 ?T
jt + w1ksin jt), k = I,...,2n andj E Z.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
It is easy to check that the ej,k's are eigenfunctions of C, i.e.

f&ej, k =- jkejk for k = 1,. . ., 2n andj E Z.
It is easy to see that the eigenspace
HI,k =Sp{e,k, eJ,k+n}, k = l ...,n,j Z \ {0},

is T7-invariant.
Forj = 0, we get that
Sp{eo,1, .e0,2n = F.
Since the U1,k's have been chosen normalized in C2n, all the ejlk's are normalized in
L2(S1, R2n). Using the Fourier expansion (5-5) and the fact that the vk's and the wk's
form a basis in R2n (Lemma 5.7), it is not difficult to show that the ej k'S form an
orthonormal basis in L2(S1, R2n).
Using (5-10) and (5-6') it is easy to see that the functions
(1 + ljl)IIe,k, ji Z, k = 1,.. .,2n,
are the normalized eigenvectors of L corresponding to the eigenvalues (1 + ]j 1)-ltjk

and that they form an orthonormal basis in H = HI/2(S', R2 n).
Next we consider the eigenvalue problem
(5-18) {(ijJ + XB)u = Au.

Using the same arguments used for the eigenvalue problem (5-11) we can find a
basis of normalized eigenfunctions of e
ei k = cos jt + WJk sinjt

where vjk + WVjk = Ujk are the eigenvectors of the problem (5-18). These eigenfunc-
tions correspond to the eigenvalues Ajk of the problem (5-18).
Then the functions (1 + Il l)-ej,k are normalized in H. They are the eigenvectors
of L + Q and they correspond to the eigenvalues (1 + 1 IY 0-'jk.
Finally we are able to characterize the spaces of the formulas (4-4) and (4-4').
In fact we have
V+ = Sp{ej,k ej n+kIjZk E Zk,. .k ,n, and Ujk ; O},

V0=Sp{ej,k, ej?n+kIjEZ, k=l,...,n, and [jk=0}I
w Sp{ejk,e?n+kIj E Z,k I ,...,n,and AjkO0},
W = p{ jk ejk Ij E Z, k = ,.. .,n, and Ajk ! ?}
(the closures are taken in H).

Now we are able to prove Lemma 5.6.
PROOF OF LEMMA 5.6. We set
Hj = Sp{ejk ej,k+n k l,... ,n}, j
V + = Hj n v+ and WJ+ = HJ n W .

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
568 VIERI BENCI
By our construction we get
(a) dim Wj-= N(ijJ + XB),

(b) dim Vj+ = 2 n-N(ijJ + IXA).
By elementary linear algebra, we have
(5-20) dim(Vj+ + Wj-) = dim VJ+ + dim Wj--dim(Vj+ n Wj-)

and, since Vj+ and Wj- C Hj
(5-21) dim(Vj, + Wy) = dim H+ - cod(j W-)

= 2n - codH(/j+ ? w-).
Combining (5-20) and (5-21) and using (5-19) we get:
dim(Vj+ n Wj-) - codH,(Vj+ + Wj-) = dim VJ+ +dim Wj--2n
= N(ijJ + XB) - N(ijJ + XA) = Oj(XB; XA).

SinceH= I]jEZHjweget
dim(V+ n w-)- codH(V+ + W) = 2jX(B; XA) = O (XB, XA).
Thus Lemma 5.6 is proved. O
LEMMA 5.8. If the Hamiltonian function satisfies (HI), (H2), (H3), H4) and
functional (5-7) satisfies (4-5)(b), then the equation (5-3) has at least ji(XB, X
nontrivial 2 ?r-periodic solutions.
PROOF. By our assumptions and Lemma 5.6, f satisfies all the hypotheses of
Theorem 4.6 except (4-5)(e). F in our case is the space of the constants. By (H3),
0, k > 0, SO F C V+ . Now we can apply Theorem 4.6 and we can define k critical
values with k = 'i?(XB, XA) via Lemma 5.6. It remains to prove that they do not
correspond to constant functions (cf. Remark 4.4). We argue indirectly and we
suppose that Zk E F is a critical point of f. Then, since all the critical values defined
by Theorem 4.6 are negative we have
f(Zk) = Ck < 0
By the definition of f (6-7), we get
f(Zk) = 27TXH(Zk)
and
H(zk) = (2rX)'Ck <0
which contradicts (H4). Thus the lemma is proved. O

PROOF OF THEOREM 5.1. We claim that if (H5) holds, then 0 4 a(L).
We have proved that
a(L) {(l ? II) IYj,kII E Zandk = I,...,2n}.

Thus it is sufficient to show that
0E a(ijJ + XA) with X y

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
We argue indirectly and we suppose there is a v E C2n such that
ijJv + XAv = 0.
Then, applying J, we get
XJAv = ijv.
Thus ij E a(XJA) = a(TJA/27r) which contradicts (H5). Then, from easy computa-
tions, (4-5)(b) follows (cf. Remark 4.10). Therefore we get the conclusion from
Lemma 5.8. O
PROOF OF THEOREM 5.2. Because of Lemma 5.8, it is enough to prove that the
functional (5-7) satisfies (4-5)(b). We let MI, M2,... be positive constants. Let Zm b
a sequence such that
(5-22) f'(Zm) 0.
Weshallwritezm = Zm?+ z + Zm + Z - V V ? DV0 V.

By (5-22), for m large enough, we get
(LZm,ZZ+ )J+ A( V2z(z), ZZ+) dt < 1IzmIl
By (H6), 1 Vz(z) I s MI. Then by the above inequality and (4-4)
fiz+ 11 2 ,- K(Lzm, z+)I 2)TIXMIIzZ + ? Izm II (/ > 0).

This gives a uniform bound for Z+. In the same manner, we can get a uniform
bound for zm. Moreover, sincef(zm) is supposed to converge, it is bounded and
M2 ?f(Zm) > Xf V(Zm) dt - (Lzm, zm)
? Af 7TV(z ;) dt + X 2,(V(Zm) - V(zo)) dt - M3

o o
? 2 fV
o o
> Af|
Therefo
there is
(a) X(0) = 0; lim X(t)= ?c+;X'(t) >0,

(5-23) b)Vz --X(e(b) +t-"
+0 M60
Since V0 is a finite dimensional space, if II zo I + ?, then
2|7 X (I z m) dt --.+x .
0
On the other hand, by (5-23)(b), we have
f| x(i zmI) dt ? f| V(zmA) dt + 27 < M5 + 2X6.

0herefore 0
Therefore also the zo are uniformly bound

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
570 VIERI BENCI
PROOF OF THEOREM 5.3. Because of Lemma 5.8, it is enough to show that the
functional (5-7) satisfies (4-5)(b). This fact is shown in [7] in the proof of Corollary
4.11. O
REMARK 5.9. Amann and Zehnder [2] have estimated the number of T-periodic
solutions for uniformly convex Hamiltonians which satisfy H1, H2 and other techni-
cal assumptions. Under (H5) (cf. (03)), this number is greater or equal to
max{Ind(B, A, X-1); Ind(A, B, VI)} (x 2 T
which, by (06) and (p3), equals I t(XB, XA) I .

Their proof reduces the problem to a finite dimensional one with no fixed points.
Our proof works directly in infinite dimensional setting. Moreover we have treated
more general Hamiltonians, at the expense of the condition
(5-24) #(XB, XA) > 0
which we needed to control F and get meaningful results in Theorems 5.1, 5.2 and
5.3.
Adding the further symmetry property H(z) = H(-z) to the Hamiltonian the
space of fixed points F reduces to {0} and the previous theorem can be improved as
is shown in Theorem 5.10.
In addition to the result of Amann and Zehnder and Theorem 5.10, other results
are available when O(XB, XA) < 0. If the Hamiltonian function has the form
H(p,q) p2 + V(q),
an estimate of the number of T-periodic solution given in [6] usin

corresponding variational problem was semidefinite. For example, under (H1), (H2),
(H4) and V(q) s 0 for every q E Rn such that 8V(q)/aq 0 o it was shown (Theo-
rem 4.4) that (5.2) possesses at least I O(XA, XB) nonconstant periodic solutions
(actually the number of solutions was estimated by comparing the eigenvalues of
a2V/aq2 at 0 and at so; but it is not difficult to show that this number equals
2#(XA, XB)).
We conclude this section with
THEOREM 5.10. Suppose that H satisfies (HI), (H2), (H5) and

(HI,) H(z) = H(-z).
Then the equation (5-2) has at least k nonze
k= 2mX#2vz 2<TA; 2
PROOF. We consider the space
H' 12(sl, RWn)

with the action of the antipodal mapping described in the examples of ?2.
If 4(XB, XA) t> 4(XA, XB) (X = /2,r) then we argue in the same manner we
have done for proving Lemma 5.8 and Theorem 5.1. We do not need (H3) since it
was used only to prove that F C V+ while here F = {0}. Also we do not need (H4)
since it has been used only for proving that the critical values do not correspond to
constant functions.

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
Therefore we get the same conclusion as Theorem 4.6 and since in this case d = 1,
we get #(XB, XA) critical values. However, if they correspond to nonconstant
periodic solutions, they are degenerate, i.e. i(K,) = 2 (or a multiple of 2). If we w
to count the independent periodic solutions (i.e. if we identify the solutions obtained
from each other by a time translation) we have to divide this number by 2.
If O(XA, XB) > i(XB, XA) we argue in the same way replacing the functional
(6-7) with its negative. O
REMARK 5.1 1. We have not used (H4), which was used in the proof of Lemma 5.8
to show that we did not get constant solutions. In fact the conclusion of Theorem
5.10 is about nonzero periodic solutions, but it might happen that there are some
constant solutions.
REMARK 5.12. Theorem 5.10 could have been applied also to the nonautonomous
case for a T-periodic Hamiltonian function since we have exploited only the evenness
and not the time translation invariancy. Then, instead of getting k independent
families of T-periodic solutions we would just get 2k pairs of periodic solutions.
6. Remarks on superquadratic Hamiltonian systems. The Hamiltonian system (5-2)

is called superquadratic if
(6-1) { there is a constant p E (2, +?) and R > 0 such that 0 <
( pH(z) (Hz(z), z) when I z I> R.
In [16] the existence of at least one periodic solution for any T > 0 was established
for superquadratic Hamiltonian systems.
This fact of course implies that (5-2) possesses infinitely many T-periodic solutions
since a T/I-periodic solution is also a T-periodic solution for any I E N+ . Thus if we
apply our abstract theorems, we do not improve such known results for autonomous
superquadratic Hamiltonian systems.
However, when H is even and time dependent Theorem 4.6 can be applied to get a
new result:
THEOREM 6.1. Suppose H(t, z) = '(Az, z) + H(t, z) with A: R2n R2n is a

symmetric operator and H satisfies
(a) H(t, z) Vt E R,z R2z

(b) H(t,z) o(z12) asIzl- 0.
(c) there is p E (2, +?) and R > 0 such that
0 < pH(t, z) < (Hz(z), z), Vt E R and z 1 R.

(6-2) (d) there is T > O such that H(t + , z)
=H(t, z) Vt E R, Vz E R2n.
(e) there are constants a, R, > 0 such that

IfHz(t, z) 1< a(z, HJ(t, z)), Vt E R, I z I> R,.
(f) H(t, z) A (t,-z).
Moreover suppose that the eigenvalue problem
(6-3) At = XJt

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00
572 VIERI BENCI
has 2n purely imaginary
(6-4) i = JH,(t, z)
has infinitely many T-periodic solutions.
In [14] Rabinowitz proved that the system (6-4) has at least one -periodic solution
without requiring (6-2)(f). Adding the symmetry property (6-2)(f) to H, we can prove
that (6-4) has infinitely many periodic solutions. The proof is a straightforward
application of Corollary 6.5, Lemma 4.8 and the estimates of [14]; so we shall not
carry out the details.
REFERENCES
1. H. Amann and E. Zehnder, Nontrivial solutions for a class of nonresonance problems and applications
to nonlinear differential equations, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 9 (1980), 539-603.
2. _ , Multiple periodic solutions of asymptotically linear Hamiltonian equations, Manuscripta
Math. 32 (1980), 149-189.
3. A. Ambrosetti and P. H. Rabinowitz, Dual variational methods in critical point theory and
applications, J. Funct. Anal. 14 (1973), 349-381.
4. P. Bartolo, V. Benci and D. Fortunato, Abstract critical point theorems and applications to some
nonlinear problems with strong resonance at infinity (to appear).
5. V. Benci, Some criticalpoint theorems and applications, Comm. Pure Appl. Math. 33 (1980).
6. __, A geometrical index for the group S1 and some applications to the study of periodic solutions of
ordinary differential equations, Comm. Pure Appl. Math. 34 (1981), 393-432.
7. V. Benci and P. M. Rabinowitz, Critical point theorems for indefinite functionals, Invent. Math. 52
(1979), 241-273.
8. D. C. Clark, A variant of Ljiusternik Schnirelmann theory, Indiana Univ. Math. J. 22 (1972), 65-74.
9. I. Eckeland and J. M. Lasry, Sur le nombre de points critiques de fonctions invariantes par des
groupes, C. R. Acad. Sci. Paris Ser. A 282 (1976), 559-562.
10. E. Fadell and P. H. Rabinowitz, Generalized cohomological index theories for Lie group action with an
application to bifurcation questions for Hamiltonian systems, Invent. Math. 45 (1978), 134-174.
11. M. A. Krasnoselskii, Topological methods in the theory of nonlinear integral equations, Macmillan,
New York, 1964.
12. L. A. Ljusternik and L. Schnirelmann, Topological methods in the calculus of variations, Hermann,
Paris, 1934.
13. P. H. Rabinowitz, A variational method for nonlinear eigenvalue problems, Nonlinear Eigenvalue
Problems (G. Prodi, editor), Edizione Cremonese, Roma, 1974, pp. 141-195.
14. _ , On subharmonic solutions of Hamiltonian systems, Comm. Pure. Appl. Math. 33 (1980),
609-633.
15. , Periodic solutions of Hamiltonian systems, Comm. Pure Appl. Math. 31 (1978), 225-251.
16. , Variational methods for finding periodic solutions of differential equations, Nonlinear
Evolution Equations (M. Crandall, editor), Academic Press, New York, 1978, pp. 225-251.
17. M. Schechter, Spectra of partial differential operators, North-Holland, Amsterdam, 1971.
18. H. Weyl, The theory of groups and quantum mechanics, Dover, New York, 1950.
ISTITUTO DI MATEMATICA APPLICATA, UNIVERSITA DI BARI, BARI, ITALY
Current address: Mathematics Research Center-University of Wisconsin, 610 Walnut Street, Madison,
Wisconsin 53706

157.193.164.114 on Mon, 19 Feb 2024 12:08:45 +00:00

Benci CriticalPointTheory 1982

Uploaded by

Copyright:

Available Formats

You might also like

Benci CriticalPointTheory 1982

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Benci CriticalPointTheory 1982

Uploaded by

Copyright:

Available Formats

On Critical Point Theory for Indefinite Functionals in The Presence of Symmetries

Author(s): Vieri Benci

Stable URL: https://www.jstor.org/stable/1999120

This content downloaded from

ON CRITICAL POINT THEORY FOR

To the memory of Guido Stampacchia

Introduction. Many variational problems which arise from physics or mathemati-

? 1982 American Mathematical Society

This content downloaded from

we consider a class of functionals defined on a real Hilbert space H having the

(0-1) f(u) = 2<Lu, u) + (I(u)

Frechet derivative T = 1" E C(H, H) is a compact operator.

(fI) f has the form (0-1).

(f2) is a slightly weakened version of the well-known condition of Palais and

This content downloaded from

1. The abstract framework. Let X be a Riemannian manifold modelled on a

This content downloaded from

(i*-3) i*(A \B) > i*(A) - i(B) VA, B E E.

(1-8) (&C= {x e XIf(x) < c},

(a) (T, Kc C E and K. is compact for every c E [Ico, c'

This content downloaded from

In concrete cases, the property (P) is strictly related to the assum

(a) there exist constants co, coo E R such that

(c) there exists A E 2 such that A C 6co and

Then the real numbers

( 1 -11) ck inf sup f(x), k =1, .., k,

are critical values of f and co ? cl < C2 * * Ck-? co. Moreover if c ck =

PROPOSITION 1.6. Let I = 1{, 9T, i} be an index theory on the Riemannian

h(A\B) n Q= (h(A)\h(B)) n Q= (h(A) n Q)\h(B)

i(h(A \B) n Q) = i( (h(A) n Q)\h(B))

This content downloaded from

A map h C C(H, H) is said to be Tg-equivariant if

F = {u E H I Tgu = u Vg E G} will denote the linear space of the Tg-inv

(2-1) H = @ Hj where Hj is finite dimensional and invariant Vj C N.

V k will denote a k-dimensional Tg-invariant sub sp ace of H.

This content downloaded from

(Tg) {A C H I A is closed and T-invariant},

6DTh(Tg) {h E C(H, H) I h is T-equivariant}.

If A C Y.(Tg) = {closed subsets of H symmetric with respect to th

This content downloaded from

1(Tsu) = einsD(u) Vu EA, Vs E [0,2X].

This content downloaded from

10(X) = {h E C (X, X) Ih = U+ Twhere Uis linear andTis compact),

PROOF. Set v = h(u), so that

u = h-'(v) = U-1(v) + +(v) = U-1 o h(u) + +(v)

= U-'(Uu + p(u)) + A(v) = u + U- o p(u) + A(v).

(v) = -U- 0 p(u) = -U-' 0 0 ? h-'(v)

and since q9 is compact, it follows that 4 is compact. L

f (a) TgU= UTU VU Et,

(2-5) te {tL I t E R}.

(a) h C D 'T i.e. h is Tg-equivariant,

(2-6) (b) h E IF,(H) i.e. h is a homeomorphism of the

COROLLARY 2.3. 6DTh* is a group of homeomorphisms.

If we suppose that F C H+ we set

(2-7) I*= {6D*, i*} where i*(A) = min i(h(A) n H+).

This content downloaded from

i*(H- n Sp) = d [dim(H+ n H-) - cod(H+ +H-)].

(i-3) i(A \B) > i*(A) - i(B) VA, B E E.

(2-7) I= {6D, i} where i(A) = min i(h(A) n H+).

(2-8) I2* = T i2} where i = min i(h(A) n S, n H+).