Global bran

hes of multi bump periodi solutions of the

Swift-Hohenberg equation

G.J.B. VAN DEN B ERG & L.A. P ELETIER

Mathemati al Institute, Leiden University, Leiden, The Netherlands

W.C. T ROY

Department of Mathemati s, University of Pittsburgh, Pittsburgh, PA 15260, USA

1. Introdu tion
In this paper we present new families of global bran hes of single and multi
bump periodi solutions of the fourth order equation
d4 u d2 u 3
+ q + u u = 0; q 2 R: (1:1)
dx4 dx2
This equation arises in a variety of problems in mathemati al physi s and me han-
i s. As an important example we mention that (1.1) des ribes stationary solutions
of the Swift- Hohenberg (SH) equation:
U   2 2
= 1 + 2 U +  U U 3;  > 0: (1:2)
t x
Equation (1.2) was rst introdu ed by Swift & Hohenberg [SH℄ in studies of
Rayleigh-Benard onve tion, and was proposed by Pomeau & Manneville [PM℄
as a good des ription of ellular ows just past the onset of instability. Of par-
ti ular interest in these studies was the formation of stationary periodi patterns,
and the sele tion of their wave-lengths. For further referen es about the SH equa-
tion we refer to the book by Collet & E kmann [CE℄ and the survey by Cross &
Hohenberg [CH℄. If  > 1, then stationary solutions U of equation (1.2), when
suitably s aled, are readily seen to be solutions u of equation (1.1). Spe i ally,
U and u are related through
1 2
u(x) = p U (( 1) 1=4 x) and q=p : (1:3)
 1  1
It is lear from (1.3) that in this example, q only takes positive values. An example
where it takes negative values of q is the Extended Fisher-Kolmogorov (EFK)
equation [CER,DS℄,
U  4U  2U
= 4 + 2 +U U 3; > 0; (1:4)
t x x
whi h yields (1.1) if we set
u(x) = U ( 1=4
x) and q= 1=2
: (1:5)
1
Both, the Swift-Hohenberg equation and the Extended Fisher-Kolmogorov are
gradient systems involving fun tionals I (u) of the form
Z n o
1 q 02
I (u) = (u00 )2 (u ) + F (u) dx;
2 2
and equation (1.1) is the orresponding Euler-Lagrange equation. Su h as furn -
tionals also arise in variational problems arising in the study of layering phenomena
in se ond order materials [LM,CMM,MPT℄. We mention in parti lar in this on-
text the governing equation of a strut [HB, TG℄ with stiness EI under an axial
ompression P and subje ted to a load Q(y ):
EIy iv + P y 00 + Q(y ) = 0:
Here, y denotes the de e tion of the strut in a dire tion perpendi ular to its axis.
The investigation in this paper is part of a study of omplex patterns in
physi s and me hani s in whose des ription equation (1.1) plays an important
role. A typi al example of su h a pattern is the phenomenon of lo alized bu kling
in me hani s. In this type of bu kling, the de e tions are onned to a small
portion of the otherwise unperturbed material. In Figure 1.1 we give an example
of su h a pattern due to Ramsay [R℄. It shows of the ee t of ompression on
a layered material in whi h the layers have dierent stiness. Be ause the stier
layer (bla k, in the enter) will not ontra t so easily as the more du tible material
that surrounds it, the sti layer de e ts sideways and produ es folds.

Fig. 1.1. Folding of a sti layer in a du tible material [R℄

Patterns are des ribed by bounded solutions of equation (1.1) on the real line.
Thus, mathemati ally this study amounts to an investigation of the dierent types
of bounded solutions equation (1.1) possesses.
In re ent years a great deal has been learnt about the stru ture of the set of
bounded solutions of equation (1.1) on the real line. It turns out to depend very
p of thepparameter q. In parti ular, one an identify two riti al
mu h on the value
values of q : + 8 and 8. At these values the linearisation around the onstant
2
solutions u = 1, i.e. the points P = (1; 0; 0; 0) in (u; u0 ; u00 ; u000 ) phase spa e,
hanges hara ter, as indi ated in Figure 1.2.
Im Im Im

o
o

o
o
Re Re Re
o

o
o

o
p p p p
(a) q  8 (b) q 2 ( 8; 8) ( ) q  8
Fig. 1.2. The spe trum of the linearisation around P+ and P
p
For q  8, the set of bounded solutions is very limited, and onsists {
modulo translations { of a one parameter family of single bump periodi solutions,
whi h are even with respe t to their extrema and odd with respe t to their zeros,
and two hetero lini orbits, or kinks, onne ting P+ and P [Be2,3, PT1,2℄: both
are odd, one is stri tly in reasing and one is stri tly de reasing. Be ause
p we shall
often need to refer to it, we denote the odd in reasing kink for q = 8 by '(x).
p
As q in reases beyond p 8 the set of bounded solutions be omes mu h ri her.
It has been proved that if 8 < q  0 it in ludes a great variety of multi bump
periodi solutions, hetero lini orbits and homo lini orbits leading to P+ and P ,
as well as haoti solutions. For detailed results we refer to [KV,KKV,PT2-4℄. For
q > 0 the results are more tentative and in omplete: although numeri al studies
for equation (1.1) [Be1℄ and related equations [BCT,CT℄ suggest an abundan e
of bounded solutions in this parameter range as well, mu h of this still remains
unproved.
The obje t of this paper is to investigate the existen e and qualitative prop-
erties of multi bump periodi solutions of equation (1.1). By this we mean here
solutions whi h have more than p one riti al point in ea h period. When q lies in
a right neighbourhood of 8, then all lo al extrema of the periodi solutions lie
near the onstant solutions u = 1, and solutions have transitions between these
uniform states. In this regime the term `multi bump' orresponds to the way it
is p ommonly used in dynami al systems theory. However, as q moves away from
8, lo al extrema are no longer tied to u = 1, and it is not easy to identify
the transitions. However, we shall still des ribe su h solutions as multi bump peri-
odi solutions. Thus, this work extends previous results [PT2,MPT℄ in whi h the
properties of single bump periodi solutions were studied. In parti ular, we will
investigate the existen e and global behaviour of families of odd and even, single
and multi bumppperiodi solutions whi h bifur ate from the stri tly in reasing
kink ' at q = 8. Odd solutions may also be even with respe t to some of their
riti al points. On the other hand, by even solutions we mean solutions, whi h are
not odd with respe t to any of their zeros. Below we indi ate some of our ndings:
(I) We
p obtain a family of periodi solutions, bifur ating from the kink p ' at
q = 8 and extending to innity, i.e, these solutions exist for all q > 8. The
3
family onsists of a ountable innity of distin t periodi solutions. The simplest
examples of these are shown in Figures 1.3-5. In the bifur ation diagram we graph
the supremum norm M = jjujj1 against q .
(II) In addition, another pair of families, both onsisting pof a ountable innity
of distin t periodi solutions, are proven to exist for q 2 ( 8; 0℄. These solutions
ontinue to exist for some, but not all, positive values of q . Numeri al eviden e
suggests that the solutions from both families pairwise lie on p loops in the (q; M )
plane of whi h the proje tion on the q -axis is of the form ( 8; q  ℄ (See Figure
1.6), and one solution lies on the top of the loop whilst the other solution lies on
the bottom. At q  > 0 the two solutions oales e.
(III) Finally, we nd a third kind of periodi solutions. These again ome as
a family of ountable
p many distin t periodi solutions whi h bifur ate from the
kink ' at q = 8. However, this family does not extend to innity nor do they
lie on loops. Instead, our numeri al results indi ate that these periodi solutions
bifur ate from the onstant solution u = 1 as q tends to a riti al value qn whi h
is of the form p  1
qn = 2 n + ; n = 1; 2; : : : : (1:6)
n
p
Note that q1 = 8, that qn+1 > qn for every n  1, and that qn ! 1 as n ! 1.
For n  2 these solutions ome in pairs. Graphs of some of them are shown
in Figures 1.7, 1.8 and 1.10. The riti al values qn arise when the moduli of the
eigenvalues of the linearisation around u = 1 are a multiple of one another. Further
details of the derivation of (1.6) are given in Se tion 5. (see also [CTh℄). Samples
of the bifur ation urves in the (q; M ) plane of these types of solutions are shown
in Figure 1.2 below.
M M M
1.50
|

4
|

1.30
|

Γ+
1.25
|

1.15
|

2
|

q q
| | | | |
8 8 8 q1 q2 q3 q4
0
Γ−
|

8 5 q
|

Type (I) Type (II) Type (III)

Fig. 1.2. The three types of bifur ation graphs
Equation (1.1) admits a rst integral, often referred to as the Energy,
1 00 2 q 0 2
= u0 u000
E (u) def (u ) + (u ) + F (u); (1:7a)
2 2
where
1
F (u) = (u2 1)2 ; (1:7b)
4
whi h is onstant if u is a solution. For the solutions u(x) = '(x) and u(x) = 1
it is lear that E (u) = 0. In this paper we fo us on bran hes of periodi solutions
4
whi h bifur ate from either u = ' or u = 1 or both. This motivates us to onsider
solutions whi h have zero energy, that is for whi h
E (u) = 0: (1:8)
In onstru ting periodi solutions, we make extensive use of symmetry prop-
erties of solutions: if a 2 R is a point where u0 = 0 as well as u000 = 0, then thanks
to the reversibility of equation (1.1), it is easily veried that u is even with respe t
to a:
u(a y ) = u(a + y ) for all y 2 R:
Also, sin e the fun tion F is even, it follows that if b 2 R is a point where u = 0
as well as u00 = 0, then u is odd with respe t to b:
u(b y ) = u(b + y ) for all y 2 R:

We begin with a brief summary of previous results [PT2,MPT℄ in whi h the

existen e of two families of single-bump periodi solutions u+ and u was proved.
p
Theorem A. For every q > 8 there exist two periodi solutions u+ and u
of equation (1.1) su h that E (u ) = 0. Both u+ and u are odd with respe t to
their zeros and even with respe t to their riti al points, and

M+ = maxfu+ (x) : x 2 Rg > 1 and M = maxfu (x) : x 2 Rg < 1:

We denote these two families of solutions by

p
 = fu (
; q ) : q > 8g:
Numeri al omputations of these bran hes made with AUTO97 [D℄ are shown
in Figure 1.3. As will be done throughout this paper when depi ting solution
bran hes, we set q along the horizontal axis and the supremum norm of the solu-
tion,
= kuk1 ;
M def
along the verti al axis.
M

4
|

Γ+

2
|

0
Γ−
|

8 5 q
|

Fig. 1.3. The bran hes + and

5
 Graphs of u+ and u made with Phaseplan [E℄ are given in Figure 1.4.
u u
1 1

| | | |
x | | | |
x
−3.75 −1.25 1.25 3.75 −7 −3 3 7

−1 −1

Fig. 1.4. Small and large single bump periodi solutions (q = 1)

The two families + and will be used to onstru t families of multi bump
periodi solutions of greater omplexity. We shall hara terise these solutions by
the riti al points and the riti al values of their graphs. We label the positive
riti al points orresponding to lo al maxima by fk g and the positive riti al
points orresponding to lo al minima by fk g. For odd solutions with u0 (0) > 0
these points satisfy
0 < 1  1  2  : : : :
In fa t, sin e u(x) is a solution whenever u(x) is, we will assume throughout that
u0 (0) > 0 for odd solutions. For even, non onstant solutions su h that E (u) = 0,
we nd that u(0) 2 R n f 1; 1g. In this ase the fun tion u0 also has innitely
many positive zeros and these satisfy

0 < 1  1  2  : : : if u00 (0) > 0;

0 < 1  2  2  : : : if u00 (0) < 0:

Starting from the solutions u+ and u , we use a shooting te hnique to obtain

a ountable familypof odd multi bump periodi solutions whi h also exists on the
entire q -interval 8 < q < 1. The solutions of this family obey the rule that all
their lo al maxima lie above u = +1 and all their lo al minima lie below u = 1.
However, the rst point of symmetry  is an ex eption: at su h a point u( ) lies
below u = 1 if it is a maximum and above u = +1 if it is a minimum:
(
< 1 if  = k for some k  1;
u( )
> +1 if  = k for some k  1:

We denote by TN the bran h of odd periodi solutions of this family, of whi h the
N th riti al point N is the rst point of symmetry:
p
TN = fu(
; q ) : q > 8; u(
; q ) is symmetri with respe t to N g:

In Figure 1.5 we give the numeri ally omputed bran hes T2 and T3 , as well as
spe i solutions whi h lie on T2 and T3 at q = 2.
6
 u u
M 1 1
3 T3
|

T2 | | | |
x | |
2 x
|

−10.5 −3.5 3.5 10.5 −7.6 7.6

q
| | |
8 2 4 −1 −1
|

Fig. 1.5. The bran hes T2 and T3 , and orresponding solutions at q = 1:5
In Se tion 3 the pre ise result for this family is formulated in Theorem 3.2.
p
In addition to these bran hes whi h exist on the entire q -interval ( p8; 1),
we prove the existen e of a se ond family of odd periodi solutions on ( 8; 0℄.
They exist in pairs, and our numeri al experiments indi ate that they sit on loop
shaped bran hes, whi h extend well into the regime q > 0. An example of su h
a loop, together with the orresponding two solutions, is given in Figure 1.6.
For this parti ular family, the solutions are symmetri with respe t to 1 with
1 < u(1 ) < 1. The pre ise des ription of these solutions is given in Theorems
3.3 and 3.4.
M u u
1 1
1.30
|

| | | |
x | |
x
1.15 −21 −7 7 21 −20 20
|

q
8 −1 −1
8

Fig. 1.6. Loop shaped bran h and odd periodi solutions at q = 1

p
10

The te hniques used to prove the existen e of these solutions for 8 <
q  0 have been developed in [PT2,3,4℄. In the present paper we show how to
obtain several families of single and multi bump periodi solutions via this method.
However, we do not aim at ompleteness, and there are many more pbran hes of
periodi solutions that an be established using this te hnique for 8<q0
(see also [PT3,4℄) than those presented here. We are not able to extend this
existen e proof to the regime q > 0. An essential diÆ ulty seems to be that the
solutions ease to exist at a oales en e point q  , whi h is dierent for every bran h
of solutions.
A dierent method for p proving the existen e of periodi solutions with energy
E (u) = 0 in the regime 8 < q < 0 has been presented in [KKV℄. There, a
minimisation pro edure is used to obtain periodi solutions both with and without
symmetry with respe t to a zero or an extremum. Sin e only the minimisers of
7
an asso iated fun tional are onsidered, the variational method establishes (for
example) the existen e of only one of the two solutions in Figure 1.7.
In Se tions 4, 5 and 6 we turn to even periodi solutions of equation (1.1).
Here we nd a third typepof bran hing phenomenon: solutions existing on nite
q -intervals of the form ( 8; qn ), still bifur ating from the kink ' at the lower end,
and a ording to numeri al eviden e, also bifur ating from the onstant solution
u = 1 at the top end. Our results here extend those obtained in our analysis [PT5℄
of the equation
uiv + qu00 + eu 1 = 0; q = 2 ; (1:9)
proposed in [LM K℄ in onne tion with the study of travelling waves (with speed
) in suspension bridges. Con erning even single bump periodi solutions of (1.1)
we prove the following existen e theorem:
p p
Theorem B. For every q 2 ( 8; 8) there exists a periodi solution u of equa-
tion (1.1), su h that E (u) = 0, whi h is even with respe t to all its riti al points,
with the properties:
1 < minfu(x) : x 2 Rg < +1 < maxfu(x) : x 2 Rg:

Two su h solutions, at q = 2 and q = +2, are shown in Figures 1.7b and 1.7 ,
and the bran h of solutions on whi h they lie is presented in Figure 1.7a.
M u u
1 1
1.30
|

Γ1

|
x x
1.15 | | | | | | |
|

−39 −13 13 39 −15 −5 5 15

q
8 8 −1 −1

(a) Bran h (b) q = 2 ( ) q = +2

Fig. 1.7. Bran h of even single bump even periodi solutions
In order to formulate our results about even multi bump periodi solutions we
need to introdu e the notion of an n-lap solution. If u is an even periodi solution
for whi h all the riti al points are lo al maxima or minima, we say that u is an
n-lap solution if it is symmetri with respe t to its nth riti al point, so that its
graph will have 2n monotone segments in one period.
Theorem C. For ea h p n  2 there exist two families of even periodi n-lap
solutions when q 2 ( 8; qn ). At the points of symmetry, n , we have
u(n ) > 1 for every n  1:
p
For n = 2 and n = 3, and q 2 ( 8; qn ) it is possible to show in addition
that the riti al values of the solutions all lie in the interval ( 1; 1) with the
8
ex eption of the point of symmetry, and if the solution is symmetri with respe t
to a minimum, its neighbouring maxima.
Note that the ase n = 1 is dis ussed in Theorem B, and orresponding 1-lap
solutions are shown in Figure 1.7.
In Figure 1.8 we present two 2-lap solutions and in Figure 1.9 we show the bran hes
of the two solutions, as well as a blowup near the point (q; M ) = (q2 ; 1).
u u
1 1

| | | | | | | |
x
−21 −7 7 21 x −18 −6 6 18

−1 −1

(a) On the upper bran h 2a (b) On the lower bran h 2b

Fig. 1.8. Two even periodi 2-lap solutions at q = 2

M M
1.50
|

Γ2b
Γ2a

1.25 1.25
|
|

Γ2b
Γ2a
q q
|

8 8 q2 q2

Fig. 1.9. Bran hes of even periodi 2-lap solutions, and a blowup at q2
In Figure 1.9, the solutions on the upper bran h 2a satisfy u00 (0) < 0 and are
symmetri with respe t to 2 . Along the lower bran h 2b the solutions satisfy
u00 (0) > 0 and are symmetri with respe t to 1 . Thus, both bran hes onsist of
2-lap solutions. The existen e of these solutions is proved in Theorem 6.2.
Corresponding results for 3-lap solutions are presented in Figures 1.10 and
1.11.
u u

1 1

| |
x | |
x
−7 7 −6.5 6.5

−1 −1

(a) On the upper bran h 3a (b) On the lower bran h 3b

Fig. 1.10. Two even periodi 3-lap solutions at q = 2

9
 M
1.50 Γ3b Γ3a

|
Γ3a

1.25 Γ3b
|

q q
| | | | |
1 3 q3 q3
8

Fig. 1.11. Bran hes of even periodi 3-lap solutions, and a blowup at q3
Solutions on the upper bran h 3a satisfy u00 (0) > 0 and are symmetri with
respe t to 2 . On the lower bran h 3b the solutions satisfy u00 (0) < 0 and are
symmetri with respe t to 2 . Thus both bran hes onsist of 3-lap solutions. The
existen e of these solutions is proved in Theorem 6.4.
It is interesting to note the dieren e in the lo al behaviour of the solution
bran hes near the points (q2 ; 1) and (q3 ; 1) in the (q; M )-plane (see Figures 1.9 and
1.11). Although a detailed analysis of this lo al behaviour is beyond the s ope of
this paper, we do present a lo al analysis of the bran hes 2a and 2b near (q2 ; 1).
This yields the angles a and b between the bran hes 2a and 2b and the q axis
at (q2 ; 1). They are given by
p p
2 2 2
tan a = and tan b = :
3 3
p
the bran hes of n-lap solutions extending over ( 8; qn ), whi h
In addition to p
bifur ate at q = 8 and at q = qn , it is possible
p to onstru t bran hes of even
periodi solutions whi h bifur ate at q = 8 and at q = qm;n where
p  n m
qm;n = 2 + ; m; n  1: (1:10)
m n
In this paper we do not study these solution exhaustively, but merely prove
p the
existen e of two ban hes of 3-lap periodi solutions whi h onne t q = 8 and
q = q2;3 . This is done in Theorem 6.5.
As an example of the general phenomenon we present in Figures 1.12 and 1.13
solutions whi h lie on the bran hes that bifur ate at q = qm;5 for m = 1; 2; 3; 4.
In ea h ase only solutions on one of the two bran hes bifur ating from ea h
bifur ation point are shown. All depi ted solutions are at q = 2. Without going
into details, we observe that n is the number of monotone laps, while m is the
number of laps that ross the onstant solution u = 1 between two points of
symmetry.
10
 gamma_(5,N) bifn. curves. 6/7/2000
gamma51.ps in /berg/

M
1.50 Γ5,1

|
Γ5,2
Γ5,3
1.25

|
Γ5,4
| | | | |

8 q5,4 q5,3 q5,2 q5,1

Fig. 1.12. Four bran hes of periodi 5-lap solutions.

/berg/sol51c.ps 6/8/2000 /berg/sol52c.ps 6/8/2000
gamma_(5,1) solution at q=2 gamma_(5,2) solution at q=2
u u
1
1

| |
x |
0 |
x
−11.1 11.1 −11.5 11.5

−1 −1

(a) m = 1
/berg/sol53c.ps 6/8/2000
(b) m = 2
/berg/sol54c.ps 6/8/2000
gamma_(5,3) solution at q=2 gamma_(5,4) solution at q=2
u u
1 1

|
0 | |
0 |
x
−12 12 x −13.2 13.2

−1 −1

( ) m = 3 (d) m = 4
Fig. 1.13. Periodi 5-lap solutions on bran hes that bifur ate at q = qm;5
for m = 1; 2; 3; 4. All solutions are at q = 2.
The organisation of the paper is the following. In Se tion 2 we introdu e
some notation and re all the important properties of riti al points obtained in
earlier papers [PT1-5℄. In Se tion 3 we establish the existen e
p of several families
of odd periodi solutions, some of whi h exist for all q > 8 and some on nite
q -intervals only. In Se tion 4 we begin
p our analysis
p of even periodi solutions
with a dis ussion of the interval 8 < q < 8. After a brief se tion on the
p u = 1, this study is ontinued in Se tion 6 for
lo al behaviour of solutions near
values of q in the interval ( 8; q3 ). In this interval it is possible to obtain
information about the lo ation of the riti al values of the solution graphs thanks
11
 p
to a Comparison Lemma whi h is valid for 8 < q < q3 . In Se tion 7, we study
the lo al behaviour of solution bran hes near q2 , and in Se tion 8, we establish
of families of periodi solutions with an arbitrary large number of laps (Theorem
C). In Se tion 9, we on lude with the rather te hni al proof of the Comparison
Lemma used in Se tion 6.
2. Criti al points
To establish the existen e of new families of periodi solutions we shall further
develop the topologi al shooting method established in [PT2-5℄. For this, we begin
with a summary of the key properties of riti al points in Lemmas 2.1, 2.2 and
2.3. Then, in Lemma 2.4, we prove a new global result. Finally, Lemmas 2.5 and
2.6 summarize two previously obtained global results. These properties will allow
us to extend our shooting method and enables us to obtain further families of
solutions and more detailed information about their qualitative properties.
The solutions we dis uss in this paper will be either even or odd, and thus we
shall study equation (1.1), whi h we restate here for onvenien e:
uiv + qu00 + u3 u = 0 (2:1)
and supply appropriate initial onditions. When looking for odd solutions we
impose the onditions
u(0) = 0; u0 (0) = ; u00 (0) = 0; u000 (0) = ; (2:2a)
and when looking for even solutions we set
u(0) = ; u0 (0) = 0; u00 (0) = ; u000 (0) = 0: (2:2b)
In both ases we shall assume that the rst integral is zero, i.e.
1 00 2 q 0 2 1 2
= u0 u000
E (u) def (u ) + (u ) + (u 1)2 = 0: (2:3)
2 2 4
This means that the onstants  and  are related by
8 q 1
>
< ( 6= 0) for odd solutions; (2.4a)
 =  () =
def 2 4
>
:  p1 j2 1j for even solutions: (2.4b)
2
For brevity we denote Problem (2.1), (2.2a), (2.3), (2.4a) by Problem A and Prob-
lem (2.1), (2.2b), (2.3), (2.4b) by Problem B. We observe that if u is a solution of
equation (2.1), then so is u. Therefore, when dis ussing odd solutions of Problem
A, we restri t our attention to solutions with a positive initial slope, i.e.  > 0.
For any given  2 R+ there exists a unique lo al solution of Problem A
and for any given  2 R there exists a unique lo al solution of Problem B. In
12
both ases we denote it by u(x; ). The riti al points of the solution graphs of
u(x; ), that is the zeros of u0 (x; ), will play a pivotal role in the onstru tion and
lassiation of the dierent families of periodi solutions. Below we summarize the
most important properties of these points. They were derived in [PT1-5℄.
We begin with a preliminary lemma whi h implies that all riti al points are
isolated.
Lemma 2.1. [PT2℄ Suppose that u is a non onstant solution of (1.1) su h that
E (u) = 0, and that u0 (x0 ) = 0 at some x0 2 R.
(a) If u00 (x0 ) = 0 then u(x0 ) = 1 and u000 (x0 ) 6= 0;
(b) If u(x0 ) = 1 then u00 (x0 ) = 0 and u000 (x0 ) 6= 0:

Lemma 2.1 implies that, unless u is a onstant solution, we an number the

riti al points of the graph of u(x; ). We denote the positive lo al maxima by k
and the minima by k with k = 1; 2; : : :. At in e tion points these points oin ide.
To start the sequen es, we need to distinguish two ases:
(i) u0 > 0 in (0; Æ ) and (ii) u0 < 0 in (0; Æ );
for some small Æ > 0.
Case (i). When u0 > 0 in a right-neighbourhood of the origin, we dene
1 = supfx > 0 : u0 > 0 on [0; x)g: (2:5a)
If u00 (1 ) < 0 we set
1 = supfx > 1 : u0 < 0 on (1 ; x)g: (2:5b)
When u00 (1 ) = 0, and so u(1 ) = 1 by Lemma 2.1, we set
1 = 1 : (2:5 )

Case (ii). When u0 < 0 in a right-neighbourhood of the origin, we skip 1 and

dene
1 = supfx > 0 : u0 < 0 on [0; x)g: (2:5d)
In (2.5a- ) we have dened the rst terms in the sequen es fk g and fk g in
both ases. We an now ontinue formally to larger values of k. For k  2 we
dene
(
supfx > k 1 : u0 > 0 on (k 1 ; x)g; if u0 > 0 in (k 1 ; k 1 + Æ1 );
k =
k 1 otherwise;
(2:6a)
13
where Æ1 is some small positive number. Similarly, we set
(
supfx > k : u0 < 0 on (k ; x)g; if u0 < 0 in (k ; k + Æ2 );
k = (2:6b)
k otherwise;
in whi h Æ2 is some small positive number. It is readily seen that
k  k  k+1 ; k  1: (2:7)
We will make extensive use of the following observation:
Remark. If u is a non onstant solution, then one of the inequalities in (2.7) must
be stri t.
To see this, suppose that
k = k :
Then, be ause the zeros of u0 are isolated by Lemma 2.1, it follows from (2.6b)
that u0 > 0 in a right-neighbourhood of k , so that u has an in e tion point at k ,
where u000 > 0. Hen e, by (2.6a),
k < k+1 :
On the other hand, if
k = k+1 ;
then by Lemma 2.1 and (2.6a), u0 < 0 in a right-neighbourhood of k and u has
an in e tion point at k , where u000 < 0. Therefore, by (2.6b),
k < k :
In parti ular, this implies that
k < k+1 and k < k+1 for every k  1: (2:8)
In the following lemma we present the important ontinuity properties of the
riti al points. In parti ular, we emphasize that, as  hanges, riti al points are
preserved and annot disappear by oales ing with one another.
p
Lemma 2.2. [PT2,3℄ Suppose that q > 8. For every  2 I , where I = R+ in
Problem A, and I = R n f 1; +1g in Problem B, and for every k  1,
(a) k () < 1 and k () < 1;
(b) u0 (k (); ) = 0 and u0 (k (); ) = 0;
( ) k 2 C (I ) and k 2 C (I ).
Remarks. (1) In [PT2,3℄, Lemma 2.2 has been proved for solutions of Problem
A. For solutions of Problem B the proof is similar (see also [PT5℄).
14
 p
(2) For odd solutions we have the following result: If q  8, then there exists
a unique value 0 > 0 for whi h the orresponding solution u(x; 0) tends mono-
toni ally to 1 (the Kink), so that 1 (0 ) = 1 and the sequen e fk g is not well
dened [Be2,3,PT1℄.
In order to pro eed with the onstru tion of new families of periodi solutions
with omplex stru ture, we need to determine the pre ise lo al behaviour of u(k )
and u(k ) when they ross the level u = 1 or u = 1 as  hanges. This is the
subje t of the next lemma.
Lemma 2.3. [PT3,5℄ Let q 2 R. Suppose that for some k  1,

u(k ) = 1 and u00 (k ) = 0 at  =  ; (2:9)

and for some Æ > 0,

u(k (); ) > 1 for  <  <  + Æ: (2:10)

(a) If u000 (k ) > 0 at  , then there exists an " > 0 su h that
u(k (); ) > u(k (); ) > 1 for  <  <  + ": (2:11a)

(b) If u000 (k ) < 0 at  , then there exists an " > 0 su h that
u(k (); ) > u(k 1 (); ) > 1 for  <  <  + ": (2:11b)

Remarks. (1) In [PT3℄, Part (a) of Lemma 2.3 was rst proved for q  0, and
in [PT5℄ this restri tion on q was subsequently removed. The proof of Part (b) is
ompletely analogous to that of Part (a).
(2) A similar result applies when u(k ) and u(k ) ross the line u = +1 from
below, or when u(k ) and u(k ) ross the line u = 1 from above or below.
(3) It is lear from Lemma 2.3 that riti al values ross the lines u = 1 in pairs.
This is a property whi h we shall very mu h exploit in Se tion 8.
The next three lemmas give important global properties of solutions of equa-
tion (1.1). The rst one applies to solutions whi h have a riti al point on the line
u = 1 or on u = 1. Thus, let u be a solution of equation (1.1), and let a 2 R be
a riti al point where u has the following properties:
u(a) = 1; u0 (a) = 0; u00 (a) = 0 and u000 (a) > 0: (2:12)
Then u0 > 0 in a right-neighbourhood of a so that the point
b = supfx > a : u0 > 0 on (0; x)g (2:13)
15
is well dened. By Lemma 2.2, it is also nite. We now derive some properties of
u and its derivatives at b.
Lemma 2.4. Suppose that
p p  1
8 < q < q3 = 2 3 + :
3
Let u be a solution of equation (1.1) whi h at a point a 2 R has the properties
listed in (2.12). Then
u(b) > 1; u0 (b) = 0; u00 (b) < 0 and u000 (b) < 0: (2:14)

The proof of Lemma 2.4 is given in Se tion 9. This result will play an important
role in the analysis of n-lap periodi solutions given in Se tion 6.
The se ond lemma applies to solutions for whi h 0  u  p13 at a riti al
point, and yields properties of the subsequent maxima and minima. We emphasize
that it is only valid for nonpositive values of q .
p
Lemma 2.5. [PT4℄ Suppose that 8 < q  0. Let u be a solution of equation
(1.1) on R su h that E (u) = 0, and that for some a 2 R

0  u(a)  p1 ; u0 (a) = 0; u00 (a) > 0 and u000 (a)  0:

3
Then p
juj > 2 whenever u0 = 0 on (a; ! ):
Here [a; ! ) is the maximal interval in [a; 1) on whi h u exists.
We on lude with a universal bound for bounded solutions.
p
Lemma 2.6. [Be2,3,PT4℄ Suppose that 8 < q  0. Let u be a solution of
equation (1.1), whi h is uniformly bounded on R. Then
p
kuk1 < 2:

3. Odd periodi solutions

In this se tion we investigate the existen e and qualitative properties of odd
periodi solutions u of equation (1.1) for whi h E (u)p= 0. In previous studies
( f. [Be℄ and [PT2℄) is was shown that for q  8 there are no su h
p odd
zero energy periodi solutions for whi h E (u) = 0. However, for q > 8 odd
p solutions do exist. In fa t, it was proved in [PT2℄ that as q
zero energy periodi
in reases from 8, two families of odd, single bump periodi solutions emerge
as thepresult of a bifur ation from the unique in reasing kink ' whi h exists at
q = 8, and it was shown in [PT2℄ and [MPT℄ that they ontinue to exist for all
16
 p
q > 8. A pre ise des ription of these results is given below in Theorem 3.1. As
we shall see, these families of single bump periodi solutions will form a basis for
our topologi al shooting arguments, whi h lead to the onstru tion of multi bump
periodi solutions with a more ompli ated stru ture.
p
Theorem 3.1. For every q > 8 there exist two odd periodi solutions u+ and
u of equation (1.1), su h that E (u ) = 0, with the following properties:
(a) ku+ k1 > 1 and ku k1 < 1:
(b) If u (a) = 0 for some a 2 R, then
u (a y ) = u (a + y ) for y 2 R:
( ) If u0 (a) = 0 for some a 2 R, then
u (a y ) = u (a + y ) for y 2 R:

By way of onvention we hoose the origin su h that u0 (0) > 0.

p
It was also shown in [PT2℄ that, as q de reases to 8, bothpfamilies of
periodi solutions tend to the unique odd in reasing kink ' at q = 8:
p
u (
; q ) ! ' as q ! 8 (3:1)
uniformly on ompa t sets. On the other hand, as q tends to innity, the small
amplitude solutions u tend to zero uniformly on R, whilst the amplitude of the
large solutions u+ tends uniformly to innity. More spe i ally,

u (x; q )  p1 sin(xpq) as q ! 1; (3:2a)

q 2
and p
u+ (x; q )  q V (x q ) as q ! 1; (3:2b)
where V is an odd solution of the equation
v iv + v 00 + v 3 = 0; (3:3)
whi h possesses the symmetry properties listed in Theorem 3.1, and

p1 :


maxfjV (t)j : t 2 Rg 2 0;
2 2
At present it is not known whether the solution V is unique. Thus, the onvergen e
in (3.2b) is along sequen es, and the fun tion V may possibly depend on the hoi e
of sequen e.
17
 A numeri ally obtained plot of the two bran hes + and of odd single
bump periodi solutions:
p
 = fu (
; q ) : q > 8g;
is presented in Figure 3.1. Along the verti al axis we set M = kuk1 .
M

|
Γ+

|
0
Γ−
|

8 5 q
|

Fig. 3.1. The bran hes + and of odd single bump periodi solutions
In the gure below we give graphs of solutions on the bran hes + and at
q = 1.
u u
1 1

| | | |
x | | | |
x
−3.75 −1.25 1.25 3.75 −7 −3 3 7

−1 −1

Fig. 3.2. Small and large single bump periodi solutions (q = 1)

As mentioned earlier, the two families of single bump periodi solutions will
be used to onstru t further families of periodi solutions. In the
p following theorem
we present a family whi h also exists on the entire interval 8 < q < 1. These
solutions look like u+ in that their lo al maxima lie above u = +1 and all the lo al
minima lie below u = 1, with the ex eption of the rst point of symmetry x =  .
At that point these extrema lie on the wrong side of the onstant solution u = +1,
in ase of a maximum, or u = 1 in ase of a minimum:
n < 1 if  = k for some k  2;
u( )
> +1 if  = k for some k  1:
We denote by TN (N  2) the bran h of odd periodi solutions of this family, of
whi h the N th riti al point is the rst point of symmetry:
p
TN = fu(
; q ) : q > 8; u0 (N ; q ) = 0; u000 (N ; q ) = 0g:
The bran hes T2 and T3 , as well as solutions on these bran hes at q = 1:5 are
presented in Figure 1.5.
18
 The existen e of this family is the ontent of the next theorem.
p
Theorem 3.2. Let q > 8.
(a) For ea h N  1 there exists an odd periodi solution u of equation (1.1) su h
that E (u) = 0 and u0 (0) > 0, whi h is symmetri with respe t to N , and has the
properties
u(k ) > 1 for 1  k  N
and u(N ) > 1: (3:4a)
u(k ) < 1 for 1  k  N 1 (N  2)
(b) For ea h N  2 there exists an odd periodi solution u of equation (1.1), su h
that u0 (0) > 0, whi h is symmetri with respe t to N , and has the properties
u(k ) > 1 for 1  k  N 1
and u(N ) < 1: (3:4b)
u(k ) < 1 for 1  k  N 1
Proof. We use an iterative type of argument, and begin by proving the ex-
isten e of a periodi solution whi h is symmetri with respe t to 1 . This is the
ase N = 1 of Part (a). Su h a solution is illustrated in Figure 1.5.
We denote the initial slopes of the zero energy odd single bump periodi
solutions u+ and u , onstru ted in Theorem 3.1, by + and  respe tively,
that is  = u0 (0). From the onstru tion in [PT2℄ we know that 0 <  < + .
Plainly (see Figure 3.2),
u(k ) > 1 and u(k ) < 1 for k  1 when  = + ; (3.5a)
0 < u(k ) < 1 and 1 < u(k ) < 0 for k  1 when  =  ; (3.5b)
where u(k ) = u(k (); ) and u(k ) = u(k (); ). In view of (3.5a) we an dene
a1 = inf f > 0 : u(1 ) > 1 on (; + )g;
and it follows from (3.5b) that a1 2 ( ; + ). By Lemma 2.2, the lo ation of
the rst riti al point, 1 (), depends ontinuously on , and by standard theory,
the solution u(x; ) of Problem (2.1), (2.2) depends ontinuously on  for x in
ompa t sets. Sin e E (u) = 0, it follows from Lemma 2.1 that
u(1 ) = 1; 1 = 1 and u000 (1 ) > 0 if  = a1 : (3:6)
By Lemma 2.3 this implies that u(1 (); ) > 1 for  2 (a1 ; a1 + Æ ), where Æ > 0
is a small positive onstant. Hen e, we an dene
a+1 = supf > a1 : u(1 ) > 1 on (a1 ; )g;
As we saw in (3.5b), u(1 ) < 1 at + , so that a+1 2 (a1; +). Invoking the
ontinuity of 1 () and u(x; ), we dedu e that
u(1 ) = 1; 1 = 2 and u000 (1 ) < 0 if  = a+1 : (3:7)
19
Using the ontinuity of 1 () and of u and its derivatives we see that (3.6) and
(3.7) imply that there must exist a point 1 2 (a1 ; a+1 ) where u000 (1 ) vanishes,
and so
u0 (1 (1 ); 1 ) = 0 and u000 (1 (1 ); 1 ) = 0:
This means that the solution u(x; 1 ) is symmetri with respe t to 1 (1 ). Sin e
it is also odd with respe t to the origin, we on lude that u(x; 1 ) is a periodi
solution with period 41 . It is readily veried that it has the desired properties.
We ontinue with the onstru tion of the periodi solution whi h is symmetri
with respe t to 2 . This is the ase N = 2 of Part (b), and su h a solution is
illustrated in Figure 1.5.
Be ause u(1 ) < 1 at + , we an dene
b1 = inf f < + : u(1 ) < 1 on (; + )g;
and it follows from (3.6) that b1 2 (a+1 ; + ). Like at a+1 , we on e again invoke the
ontinuity of 1 and u and the fa t that E (u) = 0 to on lude from Lemma 2.1
that
u(1 ) = 1; 1 = 2 and u000 (1 ) < 0 at b1 : (3:8)
By a result similar to Lemma 2.3 we nd that u(2 ) < 1 in an interval (b1 ; b1 + Æ ),
where Æ > 0 is suÆ iently small. Thus, we an dene
b+1 = supf > b1 : u(2 ) < 1 on (b1 ; )g:
Remembering (3.5a), we see that b+1 2 (b1 ; + ), and using the ontinuity properties
of 2 and u, we on lude that
u(2 ) = 1; 2 = 2 and u000 (2 ) > 0 at b+1 : (3:9)
Another appli ation of the ontinuity of 2 and u and its derivatives implies the
existen e of a point b1 2 (b1 ; b+1 ) su h that
u(2 ) < 1; u0 (2 ) = 0 and u000 (2 ) = 0 at b1 :
As in the previous ase, this means that u(x; b1 ) is a periodi solution with period
42 . Re all that b1 > a1 , so that u(1 ) > 1. Thus, this solution has the desired
properties.
Continuing in this manner, we su essively prove the existen e of all the pe-
riodi solutions listed in Theorem 3.2.
p
In addition to these bran hes of solutions, whi h exist for allpq > 8, there
exists a multitude of odd zero energy periodi solutions for 8 < q  0. In
Theorems 3.3 and 3.4 we present a few of these families. They exist in pairs.
Those orresponding to Theorem 3.3 are shown in Figures 3.3 and 3.4, and those
obtained in Theorem 3.4 in Figure 3.6.
20
 u u
1 1

| | | |
x | |
x
−21 −7 7 21 −20 20

−1 −1

Fig. 3.3. Solutions symmetri with respe t to 1 from Theorem 3.3(a)

(N = 1; q = 101 )

u u
1 1

| | | | | |
x −18 −9 9 18 x
−14 14

−1 −1

Fig. 3.4. Solutions symmetri with respe t to 2 from Theorem 3.3(b)

(N = 2; q = 101 )
A numeri al study shows that the solutions obtained in Theorem 3.3 lie on bran hes
whi h are loop shaped. The bran h for Part (a) is shown in Figure 3.5.
M

1.30
|

1.15
|

q
8 8

Fig. 3.5. Bran h of solutions as onstru ted in Theorem 3.3(a) for N = 1

p
Theorem 3.3. Let 8 < q  0.
(a) For ea h N  1 there exist two odd periodi solutions u1 and u2 of equation
(1.1) su h that E (ui ) = 0 and u0i (0) > 0 (i = 1; 2), whi h are symmetri with
respe t to N , and have the properties
u1 (k ) > 1; u2 (k ) > 1 for 1kN
u1 (k ) < 1; u2 (k ) < 1 for 1kN 1 (if N  2) (3:10a)
1 < u1 (N ) < 0 < u2 (N ) < 1:
21
(b) For ea h N  2 there exist two odd periodi solutions u1 and u2 of equation
(1.1), su h that E (ui ) = 0 and u0i (0) > 0 (i = 1; 2), whi h are symmetri with
respe t to N , and have the properties
u1 (k ) > 1; u2 (k ) > 1 for 1kN 1
u1 (k ) < 1; u2 (k ) < 1 for 1kN 1 (3:10b)
1 < u1 (N ) < 0 < u2 (N ) < 1:

Proof. We pi k up the line of argument in the proof of Theorem 3.2, and

onsider the interval [a1 ; b1 ℄. We re all that
u(1 ) = 1 at a1 and u(1 ) = 1 at b1 :
By ontinuity this implies that u(1 ) has a zero on (a1 ; b1 ). Let 1 be the smallest
zero of u(1 ) on (a1 ; b1), and let
1 = inf f < 1 : u(1 ) < 1 on (; 1 )g: (3:11a)
Similarly, let d1 be the largest zero of u(1 ) on (a1 ; b1 ), and let
d+1 = supf > d1 : u(1 ) > 1 on (d1 ; )g: (3:11b)
Plainly, 1 2 [a+1 ; 1) and d+1 2 (d1 ; b1℄, and
u(1 ) = 1 at 1 and u(1 ) = 1 at d+1 :
Sin e u(1 ) > 1 on (a1 ; b1℄, it follows that u000 (1 ) < 0 at 1 as well as at d+1 .
However, we dedu e from Lemma 2.5 that u000 (1 ) > 0 at 1 and at d1 . Therefore,
u000 (1 ) hanges sign on ( 1 ; 1 ) and on (d1 ; d+1 ), so that there exist a point 1 2
( 1 ; 1 ) and a point d1 2 (d1 ; d+1 ) su h that
u000 (1 ) = 0 at 1 and d1 :
Writing u1 (x) = u(x; d1 ) and u2 (x) = u(x; 1 ), we on lude that u1 and u2 are
odd periodi solutions whi h are symmetri with respe t to 1 , and that they have
the properties
u1 (1 ) > 1; u2 (1 ) > 1 and 1 < u1 (1 ) < 0 < u2 (1 ) < 1:
This ompletes the proof of the ase N = 2 of Part (a).
Next, we onstru t a pair of odd periodi solutions whi h are symmetri with
respe t to 2 . This orresponds to the ase N = 2 of Part (b). To this end, we
dene
a2 = inf f < + : u(2 ) > 1 on (; + )g:
22
Plainly, b1 < a2 < + . Be ause u(2 ) = 1 at b1 and u(2 ) = +1 at a2 , u(2 )
has a zero on (b1 ; a2 ). Let e1 be the smallest zero of u(2 ) on (b1 ; a2 ) and f1 the
largest. Then, pro eeding as in the previous ase we show that u000 (2 ) has two
zeros, e1 and f1 , on (b+1 ; a2 ) su h that u1 (x) = u(x; e1 ) and u2 (x) = u(x; f1) are
periodi solutions with the following properties:
ui (1 ) > 1; ui (1 ) < 1 (i = 1; 2);
and
1 < u1 (2 ) < 0 < u2 (2 ) < 1:
This ompletes the proof of the ase N = 2 of Part (b).
For the next step we dene
b2 = inf f < + : u(2 ) < 1 on (; + )g:
Sin e u(2 ) = 1 at a2 , it follows that b2 > a2 . By repeating the arguments applied
to the interval [a1 ; b1℄ to [a2 ; b2 ℄, we prove Part (a) for N = 2. We an ontinue
this pro ess indenitely, and so su essively prove all the ases in Parts (a) and
(b) of Theorem 3.3.
Remark. Using a ontinuity argument, we an show that the families of periodi
solutions des ribed in Theorem 3.3 ontinue to exist for small values of q > 0. This
is onsistent with the numeri ally omputed bifur ation bran h shown in Figure
3.5 for the ase N = 1 of Part (a).
In the next theorem we obtain a dierent family of periodi solutions. To
explain the dieren e, let u be a periodi solution whi h is symmetri with respe t
to its nth riti al point n . Then solutions of Theorem 3.3 have the property that
ju(k )j > 1 for 1  k  n 1 and ju(n)j < 1:
In ontrast, the solutions of Theorem 3.4 have the property that
ju(k )j > 1 for 1  k  n 2 and ju(n 1)j < 1; ju(n)j < 1:
Thus, whereas in the rst family, the point of symmetry is the only riti al point
at whi h u 2 ( 1; 1), in the se ond family the value of u at the point of symmetry,
as well as at the two adja ent riti al points, lie in the interval ( 1; 1). A pair of
su h solutions is shown in Figure 3.6. Sin e the hara teristi s of these solutions
are not very lear when q  0, we have set q = 1:5.
M u u
1 1
1.50
|

x | |
x
1.25 | | | |
|

−20 −7 7 20 −22.5 7.5

q
8 8 −1 −1

23
 Fig. 3.6. Solutions symmetri with respe t to 2 from Theorem 3.4(a)
(N = 2; q = 1:5)
p
Theorem 3.4. Let 8 < q  0.
(a) For ea h N  2 there exist two odd periodi solutions u1 and u2 of equation
(1.1) su h that E (ui ) = 0 and u0i (0) > 0 (i = 1; 2), whi h are symmetri with
respe t to N , and have the following properties:
)
ui (k ) > 1 for 1kN 1
for i = 1; 2; (3:12a)
ui (k ) < 1 for 1kN 2 (if N  3)
and
0 < ui (N ) < 1 for i = 1; 2;
(3:12b)
1 < u1 (N 1 ) < 0 < u2 (N 1 ) < 1:
(b) For ea h N  2 there exist two odd periodi solutions u1 and u2 of equation
(1.1) su h that E (ui ) = 0 and u0i (0) > 0 (i = 1; 2), whi h are symmetri with
respe t to N , and have the following properties:
)
ui (k ) > 1 for 1kN 2 (if N  3) for i = 1; 2; (3:12 )
ui (k ) < 1 for 1kN 1

and
0 < ui (N ) < 1 for i = 1; 2;
(3:12d)
1 < u1 (N ) < 0 < u2 (N ) < 1:
Proof. (a) We begin with the proof for N = 2. To that end, we return to
the interval [a1 ; b1℄ dened in the proof of Theorem 3.3, and onsider the two
subintervals:
I1 = [a1 ; 1 ℄ and I1+ = [d1 ; b1℄;
where 1 and d1 are, respe tively, the smallest and the largest zero of u(1 ) on
(a1 ; b1 ). and we observe that by Lemma 2.5,
u(k ) > 1 and u(k ) < 1 for k  2 at 1 and d1 : (3:13)
We rst onsider the interval I1 . As in (3.11a) we set
1 = inf f < 1 : u(1 ) < 1 on (; 1 )g:
Then a+1  1 < 1 . By Lemma 2.3, u(2 ) < 1 in a right-neighbourhood of 1 , so
that we an dene
+1 = supf > 1 : u(2 ) < 1 on ( 1 ; )g:
24
It is lear that +1 2 ( 1 ; 1 ) and that
u(2 ) = u(2 ) = 1 and u000 (2 ) > 0 at +1 :
Sin e u000 (2 ) < 0 at 1 , it follows that u000 (2 ) has a zero 1 2 ( 1 ; +1 ). Thus
u2 (x) = u(x; 1 ) is an even periodi solution with the properties
u2 (1 ) > 1; 0 < u2 (2 ) < 1; 0 < u2 (1 ) < 1 and u000 2 (2 ) = 0: (3:14a)

For the se ond solution we onsider the interval I1+ = [d1 ; b1 ℄. Be ause u(2 ) =
1 at b1 we an dene
b1 = inf f < b1 : u(2 ) < 1 on (; b1)g;
and in view of (3.13) it follows that b1 2 (d1 ; b1 ). Sin e u(1 )  0 on I1+ and
u(2 ) = 1 at b1 , we on lude that u000 (2 ) > 0 at b1 . Let
~b1 = supf > b1 : u(2 ) > 0 on (b1 ; )g:

Plainly, b1 < ~b1 < b1 . We dedu e from Lemma 2.5 that u000 (2 ) < 0 at ~b1 .
Therefore, u000 (2 ) hanges sign on (b1 ; ~b1 ), and hen e there exists a point b1 2
(b1 ; ~b1 ) where u000 (2 ) vanishes. This means that u1 (x) = u(x; b1 ) is a periodi
solution endowed with the properties
u1 (1 ) > 1; 0 < u1 (2 ) < 1; 1 < u1 (1 ) < 0 and u000 (2 ) = 0: (3:14b)
1

This ompletes the proof of Theorem 3.4(a) for N = 2.

To prove Part (a) for N = 3, we repeat the above arguments for the interval
[a2 ; b2 ℄ dened in the proof of Theorem 3.3. For N = 4 we onsider the orre-
sponding interval [a3 ; b3 ℄ and generally, we onsider the interval [aN 1 ; bN 1 ℄ for
arbitrary N  2.
For the proof of Part (b), say for N = 2, we onsider the interval [b1 ; a2 ℄.
Pro eeding as in the proof of Part (a) (N = 2) we now nd solutions whi h are
symmetri with respe t to 2 for values of  on (b1 ; e1 ) and (f1 ; a2 ), where e1
and f1 have been dened in the proof of Theorem 3.3. The argument, and its
generalisation to higher values of N is very similar to the arguments involved in
the proof of Part (a), and we shall therefore omit the details.
p p
4. Even periodi solutions: 8<q< 8
In this se tion we establish the existen e of an innite
p p sequen e of ountable families of
even periodi solutions of equation (1.1) for q 2 ( 8; 8), distinguished by the number
and lo ation of lo al maxima and minima of their graphs. Thus, whereas some of the
results obtained for for odd solutions were only validpfor q  0, the results proved in this
se tion are also valid for positive values of q up to 8. In the next ve se tions we go
even beyond this number, and show how the bran hes of multi bump periodi solutions
25
obtained in this se tion extend to higher values of q . As we stated in Se tion 2, we shall
use a shooting te hnique to establish the existen e of these solutions, and hen e, thanks
to symmetry with respe t to x = 0, we will study the initial value problem
(
uiv + qu00 + u3 u = 0 for x > 0 (4.1a)
u(0) = ; u (0) = 0; u (0) = ; u000 (0) = 0:
0 00 (4.1b)
Again, we only dis uss solutions for whi h the rst integral is zero, i.e.
1 00 2 q 0 2
= u0 u000
E (u) def (u ) + (u ) + F (u) = 0; (4:2a)
2 2
where
1
F (s) = (s2 1)2 and F 0 (s) = f (s) = s3 s: (4:2b)
4
This means that
1
=  () =  p j2 1j:
 def (4:3)
2
The ases u00 (0) > 0 and u00 (0) < 0 will be dealt with in su ession. We refer to them by,
respe tively, Case I and Case II:
Case I : u00 (0) > 0 and Case II : u00 (0) < 0:
In both ases we shall denote the solution of problem (4.1) by u(x; ).
Note that the single bump periodi solutions of Se tion 3 be ome even solutions after
a shift over a quarter of a period. Thus if the period of u is 4` , and M = ku k1 , then
u(x; M ) = u (x ` ) in Case I;
(4:4)
u(x; +M ) = u (x ` ) in Case II:
These solutions provide the point of departure for the shooting arguments whi h will yield
new families of even periodi solutions with more ompli ated stru ture. For onvenien e
we provide the graphs of u(x; M ) in Figure 4.1.
u u
1 1

| |
x | |
x
−2.5 2.5 −4.7 4.7
−M−
−1
−M+

Fig. 4.1. The small and the large single bump periodi solutions for Case I
We begin by establishing the existen e of a new family of even, single bump periodi
solutions whose maxima lie above the line u = +1, and whose minima lie between the lines
26
u = 1 and u = +1. In Figure 4.2 we give examples of two su h solutions omputed at
q = 2 and q = 2.
u u
1 1

| | | |
x
| | | |
x
−39 −13 13 39 −10 −5 5 10

−1 −1

q= 2 q=2
Fig. 4.2. Even single bump periodi solutions for q = 2 and q = 2
It is readily apparent that these solutions are qualitatively dierent from those shown
in Figure 4.1, be ause ju(x; M )j < 1 for every x 2 R and ju(x; M+ )j > 1 at every
riti al point. Like and + , this new familypof solutions forms a bran h 1 whi h
bifur ates from the unique odd kink '(x) at q = 8. However, as the bifur ation diagram
in Figure 4.3 below shows, in ontrast to the bran hes and + , whi h extend all the
way to q = +1 (see p pFig. 1.3), our omputations pindi ate that 1 only extends over the
nite q -interval ( 8; 8), and bifur ates at q = 8 from the onstant solution p up= +1.
In Theorem 4.1 we prove that the new solutions indeed exist for every q 2 ( 8; 8).
1

1.30
|

Γ1

1.15
|

q
8 8

Fig. 4.3. Bran h of even single bump periodi solutions

p p
Theorem 4.1. Let q 2 ( 8; 8). Then there exists an even, single bump periodi
solution u su h that E (u) = 0, and
1 < minfu(x) : x 2 Rg < +1 < maxfu(x) : x 2 Rg:
Proof. The proof uses ideas developed in [PT5℄. Sin e we seek a single bump periodi
solution, it follows that u00 (0) > 0, so that we are in Case I. We follow u(1 ) as  varies,
and seek a value of  2 ( 1; 1) su h that
u(1 ) > 1 and u000 (1 ) = 0: (4:5)
Then, by symmetry, u is a periodi solution with half-period L = 1 whose maxima and
minima are lo ated as indi ated above:
minfu(x) : x 2 Rg =  2 ( 1; 1) and maxfu(x) : x 2 Rg = u(1 ) > 1:
27
To nd su h a value of , we use the auxiliary fun tional

H(u) = 12 (u00 )2 + 2q (u0 )2 + F (u); (4:6)

where F has been dened in (4.2b). Let u(x) be a smooth fun tion. Then we write

= H(u(x)):
H (x) def
Dierentiation yields
H 0 = u00 u000 + qu0 u00 + f (u)u0 ; (4:7)
and if, in addition, u is a solution of equation (1.1), then
H 00 = (u000 )2 + qu0 u000 + f 0 (u)(u0 )2 : (4:8)
The right hand side of (4.8) is a se ond order polynomial in u000 , with dis riminant
D = fq 2 4f 0 (u)g(u0 )2 : (4:9)
Thus, H 00 will be nonnegative whenever
q2 + 4
q 2 < 4(3u2 1) or u2 > : (4:10)
12
Dene r
q2 + 4
0 = : (4:11)
12
Then 0 2 (0; 1) as long as q 2 < 8.
Lemma 4.2. Let q 2 < 8, and let  2 [0 ; 1). Then

u(1 ) > 1 and u000 (1 ) < 0:

Proof of Lemma 4.2. Let  2 [0; 1). Then u00 (0) > 0, so that u0 (x; ) > 0 and
u(x; ) > 0 for 0 < x  1. This implies, by (4.8) and (4.10), that H 00 (x) > 0 for
0 < x  1 , and so
H 0 (x) > H 0 (0) = 0 for x 2 (0; 1℄: (4:12)
Hen e
H (1 ) > H (0): (4:13)
A ording to the identity (4.2),
(u00 )2 = 2F (u) if u0 = 0: (4:14)
Therefore, at any riti al point  of u we have
H ( ) = 2F (u( )): (4:15)
28
Combining (4.13) and (4.15), we on lude that
F (u(1 )) > F (0 ):
Be ause F 0 = f < 0 on (0; 1) and 0 2 (0; 1), this implies that u(1 ) > 1. From (4.14)
we dedu e that u00 (1 ) < 0. Be ause H 0 (1 ) > 0 by (4.12), we on lude from (4.7) that
u000 (1 ) < 0, and the proof of Lemma 4.2 is omplete.
We now ontinue with the proof of Theorem 4.1. By Lemma 4.2, u(1 ) > 1 at  = 0 .
Thus, remembering that u(1 ) = M < 1 when  = M , we an introdu e the point
1 = inf f < 0 : u(1 ) > 1 on (; 0 )g;
and on lude that 1 2 ( M ; 0 ). It follows from the ontinuity of 1 () and u(1 (); ),
established in Lemma 2.1 and Lemma 2.2, whi h we an apply be ause E (u) = 0, that
u(1 ) = 1; u000 (1 ) > 0 and 1 = 1 at 1 : (4:16)
The fa t that u000 (1 ) is positive at 1 follows from Lemma 2.1 and the denition of 1 .
Thus u000 (1 ) has hanged sign on (1 ; 0). Sin e u000 (1 (); ) depends ontinuously on 
by Lemma 2.2, there must be a point 1 2 (1 ; 0) where u000 (1 ) vanishes, i.e.
u000 (1 (1 ); 1 ) = 0: (4:17)
Remembering that u0 (1 ) = 0 as well, it follows by symmetry that the fun tion u(x; 1 ) is
an even, single bump periodi solution, whi h is symmetri with respe t to 1 , su h that
u(1 ) > 1 and the period is 21 . This ompletes the proof of Theorem 4.1.
We ontinue with the onstru tion of an even, 1-bump periodi solution, whi h is
symmetri with respe t to 2 . We have seen in (4.16) that when  = 1 , then u = 1 and
u000 > 0 at 1 . This means that u(2 ) > 1 at 1 . We now dene the point
~ 1 = supf > M : u(1 ) < 1 on ( M ; )g:
Then ~ 1 2 ( M ; 0), and (4.16) holds, but now at ~1. Thus, u(2) > 1 at ~1 , and we
dene
2 = inf f < ~ 1 : u(2 ) > 1 on (; ~ 1 )g;
Sin e u(2 ; M ) = M < 1, we an on lude again that 2 2 ( M ; ~ 1).
As before, u000 (2 ) > 0 at 2 . Thus, it remains to determine the sign of u000 (2 ) when
 = ~ 1 . We have
H 0 (1 ) = 0 at ~ 1 :
Be ause u(x) > 1 for x 2 (1 ; 2℄, it follows that H 00 > 0 on (1 ; 2 ), and
H 0 (2 ) = u00 (2 ) u000 (2 ) > 0 at ~ 1 :
29
Plainly, u(2 ) > 1, and hen e, by the rst integral, u00 (2 ) < 0. Thus, u000 (2 ) < 0 at ~ 1 ,
and u000 (2 ) has hanged sign on (2 ; ~ 1 ), and therefore has a zero 2 in this interval:

u000 (1 (2 ); 2 ) = 0: (4:18)

By symmetry this yields an even periodi solution u(x; 2 ) whi h is symmetri with respe t
to 2 , so that u(2 ) > 1, u(1 ) < 1 and the period is 22 ( f. Fig 4.4(a)).
We an now onstru t an N -bump periodi solution for any N  2 by ontinuing the
above pro ess in an iterative manner. This yields a de reasing sequen e of numbers fk g
su h that the solutions uk (x) = u(x; k ) are even and periodi with period 2k . They have
the properties

uk (j ) < 1 for j = 1; : : : ; k 1 and uk (k ) > 1: (4:19)

Thus we have proved:

p p
Theorem 4.3. Let 8 < q < 8. Then for any N  2 there exists an even periodi
solution u of equation (1.1) su h that E (u) = 0 and u00 (0) > 0, whi h is symmetri with
respe t to N and has the properties:
(a) u has N 1 lo al maxima on the interval (0; N ).
(b) All the lo al maxima of u, ex ept the one at the point of symmetry, N , lie below the
line u = 1, but u(N ) > 1, i.e.

u(k ) < 1 for 1  k  N 1 and u(N ) > 1:

For N = 2 and N = 3 su h solutions are shown below in Figure 4.4.

u u

1 1

| |
x | |
x
−7 7 −11 11

−1 −1

(a) N = 2 (b) N = 3
Fig. 4.4. Multi bump periodi solutions of Theorem 4.3; u00 (0) > 0 and q = 2
In addition to the family of periodi solutions uN des ribed in Theorem 4.1 and
Theorem 4.3, whi h are symmetri with respe t to N for some N  1, there exists a
orresponding family of periodi solutions whi h are similar to uN , but they are symmetri
with respe t to N . Su h solutions are shown in Figure 4.5. The existen e of this family
of solutions is established in Theorem 4.4.
30
 u u
1 1

| |
x | |
x
−8.5 8.5 −12.5 12.5

−1 −1

(a) N = 2 (b) N = 3
Fig. 4.5. Multi bump periodi solutions of Theorem 4.4; u00 (0) > 0 and q = 2
p p
Theorem 4.4. Let 8 < q < 8. Then for every N  1 there exists an even periodi
solution u of equation (1.1) su h that E (u) = 0 and u00 (0) > 0, whi h is symmetri with
respe t to N and has the properties:
(a) u has N lo al maxima on the interval (0; N ).
(b) u(k ) < 1 for 1  k  N 1 if N > 2;
( ) u(N ) > 1 and u(N ) > 1:

Proof. We give the proof for N = 2. For N = 1 and for N  3 it is similar. For
further details we refer to [PT5℄. We re all the point 2 dened in the proof of Theorem
4.3, and in parti ular that
u(2 ) = 1; u000 (2 ) > 0 and 2 = 2 at 2 : (4:20)
Therefore, by Lemma 2.3, u(2 ) > 1 for  2 (2 ; a2 + Æ ) for some small Æ > 0. At the point
2 > 2 { also dened in the proof of Theorem 4.3 { the solution u(x; 2 ) is symmetri
with respe t to 2 , and hen e
u(2 ) = u(1 ) < u(1 ) < 1 at 2 :
Thus
2 = supf > 2 : u(2 ) > 1 on (2 ; )g
is well dened, and 2 2 (2 ; 2 ). We have
u(2 ) > 1; u(2 ) = 1 and u000 (2 ) < 0 at 2 :
Remembering from (4.20) that u000 (2 ) > 0 at 2 , we on lude that there must be a point
 000 
2 2 (2 ; 2 ) where u (2 ) vanishes, so that u(x; 2 ) is a periodi solution of equation
(1.1) with the properties listed in Theorem 4.4 for N = 2.
31
 p p
For any q 2 ( 8; 8), there exists yet another family fun g of even periodi solutions.
They are hara terised by the properties:
(P1) For every n  1, un is symmetri with respe t to the nth riti al point n ;
)
un (k ) > 1 if k is a maximum
(P2) if k = 1; : : : ; n 1:
un (k ) < 1 if k is a minimum
and these inequalities are reversed at k = n. This family was rst investigated in some
detail in [PT5℄, where the proof of their existen e an be found. We show three solutions
of this family in Figure 4.6.
u u u
1 1 1

| |
x | |
x | |
x
−4.5 4.5 −7 7 −9.5 9.5

−1 −1 −1

Fig. 4.6. Multi bump periodi solutions at q = 2:5

These solutions are interesting be ause of the following onje ture:
Conje ture. Let un be a sequen e of even periodi solutions with the properties P1 and
P2, and let un (0) = n . Then n !  as n ! 1, and u(x;  ) is an even homo lini
solution of equation (1.1) with u(0;  ) < 1. In a similar manner one an onstru t a
homo lini solution u(x) of equation (1.1) with u(0) > 1.

We now investigate Case II, where we assume that u00 (0) < 0, and establish results
similar to those obtained in Theorems 4.1, 4.3 and 4.4 for Case I. We emphasize that sin e
the rst riti al point is now a minimum, whi h is denoted 1 , we skip 1 and number the
riti al points as follows:
0 < 1 < 2 < 2 < : : : :
We begin our analysis of Case II by proving a result analogous to Lemmap4.2. For this p
re all the denition of 0 given in (4.11), and the fa t that 0 < 0 < 1 for 8 < q < 8.
Lemma 4.5. Let q 2 < 8, and u00 (0) < 0. Then there exists a point  ~ 2 (0 ; 1) su h that
if  2 [~
; 1), then
u(2 ) > 1 and u000 (2 ) < 0 at :
Proof. From a linear analysis at u = 1, of whi h the details an be found in the
Appendix, we see that
3 a  a 
2 () ! p p and u(1 (); )  1 (1 ) sinh as  ! 1;
q+ 8 b 2b
32
where a and b are positive onstants whi h are independent of , and given in equation
(A.4) of the Appendix. Thus, there exists an  2 (0 ; 1) su h that if  2 (; 1), then
u(x; ) > 0 if 0 < x  2 :
Hen e, by (4.10),
H 00 (x) > 0 if 0 < x  2 :
Be ause H 0 (0) = u00 (0)u000 (0) = 0, this implies that
H 0 (x) > 0 if 0 < x  2 : (4:21)
Thus H (0) < H (1 ) < H (2), and hen e, by (4.15), F (u(0)) < F (u(1 )) < F (u(2 )). This
means that u(2 ) > 1. We also dedu e from (4.21) that H 0 = u00 u000 > 0 at 2 . Sin e
u00 (2 ) < 0, we on lude that u000 (2 ) < 0. This ompletes the proof.
Thus, for  2 (0; 1) suÆ iently lose to 1, we have u(2 ) > 1 and u000 (2 ) < 0. On the
other hand, when  = M , we have u(2 ) < 1. Therefore, we an dene the number
2 = inf f < 1 : u(2 ) > 1 on (; 1)g
and 2 2 (M ; ~ ). Plainly, u(2 ) = 1 and u000 (2 ) > 0 at 2 , so that u000 (2 ) must have a
zero for some 2 2 (M ; 1), whi h yields the rst of a family of even periodi solutions
with u00 (0) < 0. As in the proof of Theorems 4.3 and 4.4, we an ontinue indu tively and
prove the existen e of two families of periodi solutions:
p p
Theorem 4.6. Let 8 < q < 8. Then for any N  2 there exists an even periodi
solution u of equation (1.1) su h that E (u) = 0 and u00 (0) < 0, whi h is symmetri with
respe t to N and has the properties:
(a) u has N 2 lo al maxima on the interval (0; N ).
(b) u(k ) < 1 for 2  k  N 1 if N  3;
( ) u(N ) > 1 if N  2:
p p
Theorem 4.7. Let 8 < q < 8. Then for every N  2 there exists an even periodi
solution u of equation (1.1) su h that E (u) = 0 and u00 (0) < 0, whi h is symmetri with
respe t to N and has the the properties:
(a) u has N 1 lo al maxima on the interval (0; N ).
(b) u(k ) < 1 for 2  k  N 1 if N  3
( ) u(N ) > 1 and u(N ) > 1 if N  2:

u u

1 1

| | | |
x | |
x
−21 −7 7 21 −6.5 6.5

−1 −1

33
 (a) See Theorem 4.6: N = 2 (b) See Theorem 4.7: N = 2
Fig. 4.7. Multi bump periodi solutions: u00 (0) < 0 and q = 2

5. Lo al analysis near u=1

p
In order to extend the existen e results of the previous se tion to the range q  8,
we need to develop
p further analyti al te hniques. This is be ause Lemma 4.2 no longer
holds for q  8. These te hniques will rely on a detailed analysis of the lo al behaviour of
solutions near u = 1. Thus, we substitute u = 1+ "v into (1.1) and require that u(0) = 1 "
After omitting the higher order terms in ", we then obtain the linear equation
v iv + qv 00 + 2v = 0 (5:1a)
and at the origin, the initial onditions be ome
v (0) = 1; v 0 (0) = 0; (v 00 )2 (0) = 2; and v 000 (0) = 0: (5:1b)
The fa t that v should be even implies that v 0 and v 000 vanish at the origin, and the
assumption that the energy E is zero leads to the ondition on (v 00 )2 . A detailed analysis
of this problem is given in the Appendix of this paper and in Appendix B of [PT5℄. For
easy referen e, we give here the main results of this analysis. As in Se tion 4, it is ne essary
to distinguish two ases:
p p
Case I : v 00 (0) = 2 and Case II : v 00 (0) = 2;
00
p we denote the solution of Problem (5.1) in these two ases by v (x), so that v (0) =
and
 2.
The roots  and  of the orresponding hara teristi equation are dened by
 = ia and  = ib; (5:2a)
in whi h a > 0 and b > 0 are dened by
1 p 1 p
a2 = (q + q 2 8) and b2 = (q q 2 8): (5:2b)
2 2
In what follows, the values of q at whi h resonan e o urs ( f. [GH℄, p. 397), i.e.
a n
= ; m; n 2 N (n  m) (5:3)
b m
will play a spe ial role. These values are readily omputed to be
p  n m
qm;n = 2 + : (5:4a)
m n
34
For onvenien e we write qn = q1;n , i.e.
p  1
qn = 2 n + : (5:4b)
n
In the following two lemmas we state the main results about the solutions v (x) of Problem
(5.1). In the rst one we present the expli it expressions of these solutions.
Lemma 5.1. The solutions v (x) of Problem (5.1) are given by

v (x) = A os(ax) + B os(bx); (5:5a)

where p p
b2  2 a2  2
A = 2 2 and B = : (5:5b)
a b a2 b2
In the se ond lemma we fo us on the riti al points of the solutions v of Problem
(5.1). In parti ular, it will be important for our shooting arguments in Se tion 6 that we
know the lo ation of these point with respe t to the v = 0 axis { above or below { and the
sign of the third derivative v000 at these points.
Lemma 5.2. Let  be a riti al point of the solution v of Problem (5.1), i.e. v 0 ( ) = 0.
Then
sin(a ) + sin(b ) = 0 in Case I ; (5.6a)
sin(a ) sin(b ) = 0 in Case II ; (5.6b)
and p
a2  2 n b o
v ( ) = 2 2  os(a ) os(b ) (5.6 )
a bp a
000
v ( ) = b(a2  2) sin(b ): (5.6d)
In both ases,

sgn v ( ) = sgn os(b ) (5:6e)
and

sgn v 000 ( ) = sgn sin(b ) : (5:6f)
The proof of Lemmap5.2 is elementary, and makes use of the observation that A =B =
 b=a and that ab = 2.
p
6. Even periodi solutions: 8  q < q3
In this se tion, and in Se tion 8, we investigate the existen e of even periodi solutions
u for whi h E (u) = 0. As was explained in the Introdu tion, we nd it onvenient to label
the solutions a ording to the number of monotone segments, or Laps, that go in a half-
period. Thus, the solutions of Theorem 4.1 are alled 1-lap solutions.
p p In the subsequent
theorems of Se tion 4 we have shown that for every q 2 ( 8; 8) there exist n-lap
solutions for any n  1.
35
 p
p no longer exist for q > 8, but that
Numeri al eviden e suggest that 1-lap solutions
2-lap solutions still exist for some values of q > 8, and that n-lap solutions exist for n
large enough. Spe i ally, we prove the following result. Let
p  1
qn = 2 n + ; n  1: (6:1)
n
Theorem 6.1. p For ea h n  2 there exist two families of even periodi n-lap solutions
when q 2 ( 8; qn ). At the points of symmetry, n , we have
u(n ) > 1 for every n  1:

p Whereas in Se tion 4 we ould use the properties of the fun tional H(u), when q >
8 this is no longer possible. The main ingredients used in the proof will now be the
linear analysis
p given in Se tion 5, a powerful Comparison Lemma (Lemma 2.4) valid for
q 2 ( 8; q3 ) and, in Se tion 8, a ounting argument whi h is reminis ent of a topologi al
degree argument.
The Comparison Lemma enables us to obtain information about the lo ation of all
p of the solutions with respe t to the line u = 1. In the present se tion we
the riti al values
assume that 8 < q < q3 , so that the Comparison Lemma holds. In Se tion 8 we allow
q to be arbitrary large, and develop the ounting argument.
p
In the rst result of this se tion we show that when 8 < q < q2 , then there exist
two families of n-lap solutions with n  2. We begin with a pair of 2-lap solutions.
p
Theorem 6.2. Let 8 < q < q2 . Then there exist two even 2-lap periodi solutions
u2a and u2b of equation (1.1), su h that E (u2a ) = 0 and E (u2b ) = 0, with the following
properties:
(a) The solution u2a is even with respe t to 2 ,
u002a (0) < 0 and u2a (1 ) > 1:
(b) The solution u2b is even with respe t to 1 ,
u00 (0) > 0
2b and u2b (1 ) > 1:
We denote the bran hes of these solutions by, respe tively, 2a and 2b . These
bran hes, as well as graphs of two spe i solutions u2a and u2b , are presented in Fig-
ure 6.1. Re all that M = kuk1 .
2

M M
1.50
|

Γ2a
Γ2a

1.25 1.25
|
|

Γ2b
Γ2b
q q
| |

8 8 q2 q2 3.6

36
 (a) The bran hes 2a and 2b and a blowup at q2
u u
1 1

| | | | | | | |
x
−21 −7 7 21 x −18 −6 6 18

−1 −1

(b) Graph of u2a at q = 2 ( ) Graph of u2b at q = 3

Fig. 6.1. Solutions of Theorem 6.2
Proof. We rst onsider Part (b) and prove the existen e of the solution u2b for whi h
00
u (0) > 0 (Case I). For onvenien e we have dropped the subs ript 2b. We distinguish
three ases:
p
(i) 8  q < q3;5 (ii) q = q3;5 ; (iii) q3;5 < q < q2 :
A ording to (5.3) and (5.4a), these ases orrespond to
a 5 a 5 5 a
(i) 1  < ; (ii) = ; (iii) < < 2:
b 3 b 3 3 b
Case (i): Re all from Se tion 5 that v denotes the solution of equation (5.1a), the lineari-
sation of equation (1.1) around u = 1. It follows from Lemma 5.2 that in this ase
v (1) > 0 and v (1 ) < 0:
Hen e, it follows from ontinuity that there exists a Æ > 0 su h that
u(1 ) > 1 and u(1 ) < 1 for 1 Æ <  < 1: (6:2)
Re all that u(1 ) = M < 1 when  = M , and dene
a1 = inf f < 1 : u(1 ) > 1 on (; 1)g:
Then, by Lemma 2.3, u(1 ) > 1 for a1 <  < a1 +  for some small  > 0. However,
u(1 ) < 1 for  lose to 1. Therefore
b1 = supf > a1 : u(1 ) > 1 on (a1 ; )g 2 (a1 ; 1):
Plainly, at b1 we have u(1 ) = 1 and by Lemma 2.1, u000 (1 ) < 0. Sin e u000 (1 ) > 0 at
a1 and u000 (1 ) < 0 at b1 , it follows that u000 (1 ) has a zero for some  2 (a1 ; b1 ), and
again we use symmetry to on lude that u(x;  ) is a periodi solution with the desired
properties.
37
Case (iii): From Lemma 5.2 we see that in this ase
v (1 ) > v (1 ) > 0 and v 000 (1 ) < 0:
Hen e, there exists a Æ > 0 su h that
u(1 ) > u(1 ) > 1 and u000 (1 ) < 0 for 1 Æ <  < 1: (6:3)
As before, we dene
a1 = inf f < 1 : u(1 ) > 1 on (; 1℄g:
and
b1 = supf > a1 : u(1 ) > 1 on (a1 ; )g 2 (a1 ; 1):
Re all that u000 (1 ) > 0 at a1 , so that b1 > a1 by Lemma 2.3. If b1 = 1, then (6.3) implies
that u000 (1 ) < 0 for  near b1 . On the other hand, if b1 < 1, then u000 (1 ) < 0 at b1 .
Thus, in both ases u000 (1 ) hanges sign on (a1 ; b1). On e again the existen e of a periodi
solution of the desired type follows.
Case (ii): From Lemma 5.2 we see that in this ase,
v (1 ) > v (1 ) = 0 and v 000 (1 ) < 0:
Thus, there exists a Æ > 0 su h that
u(1 ) > 1 and u000 (1 ) < 0 for 1 Æ <  < 1: (6:4)
Fix  2 (1 Æ; 1). If, for this value of , u(1 ) < 1 we an omplete the proof as in Case
(i), and if u(1 )  1, we an omplete it as in Case (iii). This nishes the proof of Part
(b).
Next, we prove Part (a). Re all that 2 denotes the rst positive lo al maximum of u
sin e u00 (0) < 0. From Lemma 5.2 we on lude that
v (1 ) < 0 and v (2 ) > 0; v 000 (2 ) < 0
for the entire interval q1  q < q2 . Hen e, there exists a Æ > 0 su h that
u(1 ) < 1 and u(2 ) > 1 u000 (2 ) < 0 for 1 Æ <  < 1: (6:5)
Set
a2 = inf f < 1 : u(2 ) > 1 on (; 1)g:
Then M < a2 < 1. Plainly, u000 (2 ) > 0 at a2 . Sin e by (6.5), u000 (2 ) < 0 for  lose
to 1, it follows that there exists a point 2 2 (a2 ; 1) su h that u000 (2 ) > 0 at 2 , and
hen e u(x; 2 ) is an even periodi solution with period 22 (2 ). Finally, u(2 ; 2 ) > 1
sin e a2 < 2 < 1. This ompletes the proof of Theorem 6.2.
38
 p
Remark. We have now shown that the solutions u2a and u2b exist over the entire interval
( 8; q2 ). Our numeri al investigation (Figure 6.1(a)) indi ates that u2b does not exist
past q2 although it appears that u2a does exist on a small q -interval beyond q2 . Our
experiments also indi ate that the bran hes 2a and 2b , on whi h these solutions sit,
disappear at q2 as a result of a bifur ation from the onstant solution u = 1. In Se tion 7
we shall give an explanation for the behaviour of the two bran hes 2a and 2b near the
point (q; M ) = (q2 ; 1).
Theorem 6.2, together with Lemma 2.4, enables us to establish the existen e of two
families of even n-lap periodi solutions, for any n  2.
p
Theorem 6.3. Let 8 < q < q2 , and let n  2. Then there exist two families of even
n-lap periodi solutions u of equation (1.1), su h that E (u) = 0, one for whi h u00 (0) > 0
and one for whi h u00 (0) < 0. They are symmetri with respe t to the nth riti al point
n . The riti al values have the properties
u(k ) < 1 if k  n 2 (n  3) and u(n ) > 1;
and
u(n 1 ) < 1 if n is a maximum;
u(n 1 ) > 1 if n is a minimum:
Proof. The proof pro eeds very mu h along the lines of the proofs of Theorems 4.3
and 4.4. To show how Lemma 2.4 is used, we give the proof for the 3-lap solution when
u00 (0) > 0. This solution is symmetri with respe t to 2 . Otherwise, we leave the proof
as an exer ise.
Let
a1 = supf > M : u(1 ) < 1 on ( M )g:
It follows from Theorem 6.2 that a1 2 ( M ; 1). In addition,
u(1 ) = 1 and u000 (1 ) > 0 at a : 1

Thus, in view of Lemma 2.4,

u(2 ) > 1 and u000 (2 ) < 0 at a1 :
Next, let
~ 2 = inf f < a1 : u(2 ) > 1 on (; a1 )g:
Then ~ 2 2 ( M ; ~ 2), and
u(2 ) = 1 and u000 (1 ) > 0 at ~ 2 :
Therefore, u000 (2 ) hanges sign, and thus has a zero 2 , on (~2 ; a1 ). It is lear from the
onstru tion that the solution u(x; 2 ) has the required properties.
p
Next, we turn our attention to the interval ( 8; q3 ), and show that n-lap solutions
ontinue to exist up to q3 for n  3. We rst fo us on 3-lap solutions.
39
 p
Theorem 6.4. Let 8 < q < q3 . Then there exist two even 3-lap periodi solutions
u3a and u3b of equation (1.1), su h that E (u3a ) = 0 and E (u3b ) = 0, with the following
properties:
(a) The solution u3a is even with respe t to 2 ,
u003a (0) > 0 and u3a (1 ) < 1; u3a (2 ) > 1:

(b) The solution u3b is even with respe t to 2 ,

u003b (0) < 0 and u3b (2 ) > 1:

The bran hes 3a and 3b of these solutions, as well as graphs of the solutions u3a
and u3b at a spe i value of q are presented in Figure 6.2.

M
1.50 Γ3b Γ3a
|

Γ3a

1.25 Γ3b
|

q q
| | | | |
1 3 q3 q3
8

(a) The bran hes 3a and 3b and a blowup at q3

u u

1 1

| |
x | |
x
−7 7 −6.5 6.5

−1 −1

(b) Graph of u3a ( ) Graph of u3b

Fig. 6.2. Solutions of Theorem 6.3 at q = 2
p
Proof. For 8 < q < q2 , the existen e of solutions su h as u3a and u3b has been
established in Theorems 4.3, 4.7 and 6.3. Thus, it suÆ es to prove Theorem 6.4 for
q2  q < q3 .
(a) Set
a1 = supf > M : u(1 ) < 1 on ( M ; )g:
From Lemma 5.2 we know that when q2  q < q3 , then
v (1 ) > 1 and v (2 ) < 1:
40
Hen e, by ontinuity, there exists a onstant Æ > 0 su h that
u(1 ) > 1 and u(2 ) < 1 for 1 Æ <  < 1: (6:6)
Therefore a1 2 ( M ; 1), and
u(1 ) = 1; u00 (1 ) = 0 and u000 (1 ) > 0 at  = a1 :
This implies that u(2 ) > 1 and, by Lemma 2.4, that u000 (2 ) < 0. Let
a2 = inf f < a1 : u(2 ) > 1 on (; a1 )g:
Then
u(2 ) = 1 and u000 (2 ) > 0 at  = a2 ;
be ause u(1 ) < u(1 ) < 1. Therefore, u000 (2 ) hanges sign on (a2 ; a1), so that there exists
a point a2 2 (a2 ; a1) su h that u000 (2 ) = 0 at a2 , and u(x; a2 ) is a periodi solution whi h is
symmetri with respe t to 2 . By onstru tion u(1 ; a2 ) < 1 and u(2 ; a2 ) > 1, as required.
(b) We analyse the following three ases separately:
(i) q2  q < q3;7 (ii) q = q3;7 ; (iii) q3;7 < q < q3 :
Again, a ording to (5.3) and (5.4a), these ases orrespond to
a 7 a 7 7 a
(i) 2  < ; (ii) = ; (iii) < < 3:
b 3 b 3 3 b

Case (i): It follows from Lemma 5.2 that in this ase

v (2 ) > 0; v (2 ) < 0 and v 000 (2 ) < 0;
so that for some small Æ > 0,
u(2 ) > 1 and u(2 ) < 1 for 1 Æ <  < 1: (6:7)
Dene
a1 = inf f < 1 : u(2 ) > 1 on (; 1)g:
Be ause u(2 ) = M < 1 when  = M , it follows that M < a1 < 1, and hen e
u(2 ) = 1 and u000 (2 ) > 0 at a1 ;
sin e u(1 ) < M < 1 at a1 . This means, a ording to Lemma 2.3, that u(2 ) > 1 for
a1 <  < a1 +  where  > 0 is some small onstant. Dene
b1 = supf > a1 : u(2 ) > 1 on (a1 ; )g: (6:8)
41
As we have seen in (6.7), u(2 ) < 1 for  lose to 1. Therefore b1 2 (a1 ; 1). Sin e u(2 ) > 1
at b1 it follows that u000 (2 ) < 0 at b1 . Hen e, in view of (6.8), u000 (2 ) hanges sign, and
therefore has a zero, at a point 1 2 (a1 ; b1). Thus, u(x; 1 ) is an even periodi solution
whi h, by onstru tion, is symmetri with respe t to 2 , and u(2 ; 1 ) > 1, as required.
Case (iii): In this ase
v (2 ) > 0; v (2 ) > 0 and v 000 (2 ) < 0;
so that
u(2 ) > 1; u(2 ) > 1 and u000 (2 ) < 0 for (1 Æ; 1) (6:9)
when Æ > 0 is suÆ iently small. As before, we dene
a2 = inf f < 1 : u(2 ) > 1 on (; 1)g;
and sin e u(2 ) = M when  = M , it follows that a2 2 (M ; 1). Be ause in this ase,
u(1 ) <  < 1 we dedu e that
u(2 ) = u(2 ) = 1 and u000 (2 ) > 0 if  = a2 : (6:10)
By Lemma 2.3, u(2 ) > 1 in a right-neighbourhood of a2 . Let
b2 = supf > a2 : u(2 ) > 1 on (a2 ; )g:
We laim that u000 (2 ) hanges sign on (a2 ; b2 ).
{ If b2 = 1, then this assertion follows at on e from (6.9), (6.10), and ontinuity.
{ If b2 < 1, then u(2 ) = 1 at b2 , and be ause u(2 ) > 1 at b2 , it follows that u000 (2 ) < 0,
so that in view of (6.10), u000 (2 ) also hanges sign on (a2 ; b2 ).
Thus, there exists a point 2 2 (a2 ; b2 ) su h that u(x; a2 ) is an even periodi solution
whi h is symmetri with respe t to 2 , su h that
u(1 ) < 1; u(2 ) > 1 and u(2 ) > 1:

Case (ii): By Lemma 5.2,

v (2 ) > 0; v (2 ) = 0 and v 000 (2 ) < 0;
so that for some small Æ > 0,
u(2 ) > 1 and u000 (2 ) < 0 for 1 Æ <  < 1: (6:11)
If u(2 ) < 1 we ontinue as in the proof of Case (i) and if u(2 )  1 we ontinue the proof
as in Case (iii). This ompletes the proof of Theorem 6.4.
About the families of n-lap solutions
p we an repeat the laim of Theorem 6.3 for the
larger range of q -values: q 2 ( 8; q3 ). However, the minimum number of laps is now
raised from 2 to 3.
42
 p bran hes of even multibump periodi solutions with E = 0
Up till now we have studied
on intervals of the form ( 8; qn ) for n  1. These bran hes appear to bifur ate from
the points (qn ; 1). In addition to these solutions p there exist families of even periodi
solutions whi h exist on intervals of the form ( 8; qm;n ), where qm;n is given in (1.10),
and n > m  2. The orresponding bran hes appear to bifur ate from the points (qm;n ; 1).
We shall not go into a general analysis of su h solutions. Instead, we provide the details
for one example, and hoose m = 2 and n = 3. We denote the two bran hes of even
periodi solutions by 2;3a and 2;3b , and the solutions that sit on these bran hes by u2;3a
and u2;3b , respe tively. Two su h solutions are presented in Figure 6.3b. , at q = 2.
p
We note that q2;3 = 136 2 2 (q1 ; q2 ). Therefore, we may again use Lemma 2.4 to
prove the existen e of these two new families of solutions.
p
Theorem 6.5. Let 8 < q < q2;3 . Then there exist two even 3-lap periodi solutions
u2;3a and u2;3b of equation (1.1) su h that E (u2;3a ) = 0 and E (u2;3b ) = 0 with the following
properties:
(a) The solution u2;3a is symmetri with respe t to 2 , and
u2;3a (0) < 1; u002;3a (0) < 0; u2;3a (2 ) > 1; u2;3a (2 ) < 1:
(b) The solution u2;3b is symmetri with respe t to 2 , and
u2;3b (0) > 1; u002;3b (0) > 0; u2;3b (1 ) < 1; u2;3a (2 ) > 1:

M
1.30 Γ2,3a
|

Γ2,3b
1.15
|

q
|
q2,3
8

(a) The bran hes 2;3a and 2;3b and a blowup at q2;3
u u

1
1

| |
x | |
x
−7.35 7.35 −7.5 7.5

−1 −1

(b) Graph of u2;3a ( ) Graph of u2;3b

Fig. 6.3. Solutions of Theorem 6.5 at q = 2
43
 Proof. (a) If q1 < q < q2;3 , then, a ording to Lemma 5.2,

v (0) = 1; v (1 ) < 0; v (2) > 0; v (2 ) < 0 and v 000 (2 ) > 0: (6:11)
Hen e, for any q 2 (q1 ; q2;3 ) there exists by ontinuity a Æ > 0 su h that for  2 (1 Æ; 1)
u(1 ) < 1; u(2 ) > 1; u(2 ) < 1; and u000 (2 ) > 0: (6:12)
Let
a1 = inf f < 1 : u(2 ) > 1 on (; 1)g:
Be ause u(2 ) < 1 when  = M , where M is the amplitude of the small single bump
periodi solution u ( f Fig. 4.1), it follows that a1 2 (M ; 1), and
u(2 ) = 1 and u(2 ) = 1 at a1 :
Next, let
a2 = inf f < 1 : u(2 ) < 1 on (; 1)g:
Then a2 > a1 by Lemma 2.3 and hen e
u(2 ) = 1; u(2 ) > 1 and u000 (2 ) < 0 at a2 : (6:13)
Remembering (6.12), we on lude that u000 (2 ) hanges sign on (a2 ; 1). Thus, by ontinuity
there exists an a2 2 (a2 ; 1) su h that u000 (2 ) = 0 at a2 . Therefore, u (x) = u(x; a2 ) is a
periodi solution whi h is symmetri with respe t to 2 and has the properties
u (0) < 1; u (1 ) < 1; u (2 ) > 1; u (2 ) < 1;
as required.
(b) If q1 < q < q2;3 , then, a ording to Lemma 5.2, or (6.11) when we multiply by a fa tor
1 and relabel the riti al points,
v (0) = 1; v (1 ) > 0; v (1 ) < 0; v (2 ) > 0 and v 000 (2 ) < 0: (6:14)
We now pro eed exa tly as in Case (a), ex ept that instead of the small single bump peri-
odi solution u , we now use that large single bump periodi solution u+ with amplitude
M+ ( f Fig. 4.1) as a referen e. We omit the details.
p
Proving that the solutions u2;3a and u2;3b exist for 8 < q  q1 as well is also left
as an exer ise.
7. Lo al behaviour near q2
In Figure 1.9 (or 6.1) we saw that the numeri ally omputed bran hes 2a and 2b
of even 2-lap periodi solutions approa h the bifur ation point (q2 ; 1) in the (q; M )-plane
from dierent dire tions. In this se tion we present a lo al analysis at the point (q2 ; 1),
44
and ompute the angles a and b whi h these bran hes make with the positive q -axis at
the bifur ation point. Spe i ally we obtain the following result.
Proposition 7.1. Let a and b be the angles with whi h the bran hes 2a and 2b
approa h (q2 ; 1). Then
p p
2 2 2
tan a = and tan b = :
3 3
We shall dis uss the two bran hes in su ession.
(1) The bran h 2a . The solutions that sit on 2a have the properties

u00 (0) < 0 and u000 (2 ) = 0;

and 2 is the rst point of symmetry. We use " = 1 u(0) > 0 as a small parameter, and
we make the Ansatz:
u(") = 1 + "v + "2 w + O("3 ); (7.1a)
q (") = q0 + "q1 + O("2 ); (7.1b)
 (") = 0 + "1 + O(" ):
2
(7.1 )
To keep the notation simple, we have denoted the zeroth order terms in the expansion of
q and 2 by, respe tively, q0 and 0 . Thus, in this notation, q0 = q2 and 0 = 2 j"=0 . We
re all that

q0 = p5 and hen e a = 23=4 and b = 2 1=4

(a = 2b); (7:2a)
2
and we obtain from Lemma 5.1 that

0 = = 21=4 : (7:2b)
b
In this notation, we need to ompute
v (0)
tan a = :
q1
When we substitute the Ansatz (7.1) into equation (1.1), use the initial onditions

u(0) = 1 "; u0 (0) = 0; u00 (0) = p1 (1 u2 ); u000 (0) = 0;

2
and equate terms of equal order in ", we obtain
(
v iv + q0 v 00 + 2v = 0 (7.3a)
p
v (0) = 1; v 0 (0) = 0; v 00 (0) = 2; v 000 (0) = 0; (7.3b)
45
and 8 iv
<w+ q0 w00 + 2w = q1 v 00 (x) 3v 2 (x) (7.4a)
1
: w (0) = 0; w 0 (0) = 0; w 00 (0) = + p ; w 000 (0) = 0: (7.4b)
2
The problem for v has been dis ussed in Se tion 5, where we found that
v (x) = os(ax) 2 os(bx): (7:5)
Note that v does not depend on the value of q1 . For later use we note that
v (0 ) = 3; v 0 (0 ) = 0; v 00 (0 ) = 6b2 ; v 000 (0 ) = 0; v iv (0 ) = 18b4 : (7:6)
Thus, we nd that
u(; "; q ) = 1 + "v (0 + "1 ) + "2 w(0 + "1 ; q1 ) + : : :
= 1 + "v (0 ) + O("2 )
= 1 + 3" + O("2 ): (7:7)
In order to ompute 1 we write
u0 (; "; q ) = "v 0 (0 + "1 ) + "2 w0 (0 + "1 ; q1 )
= "v 0 (0 ) + "2 1 v 00 (0 ) + "2 w0 (0 ; q1 ) = 0:
Hen e
w0 (0 ; q1 )
1 = : (7:8)
v 00 (0 )
To ompute q1 we observe that we an write
w(x; q1 ) = w0 (x) + q1 w1 (x); (7:9)
and an easy omputation shows that
w0 (x) = f (x) 3g (x); (7:10a)
where f is given by

f (x) = p 21 f os(ax) + os(bx)g: (7:10b)

2(a b2 )
and g is given by Z
1 x
g (x) = K (x t)v 2 (t) dt; (7:10 )
a2 b2 0

in whi h
sin(bx) sin(ax)
K (x) = : (7:10d)
b a
46
For w1 (x) we nd Z
1 x
w1 (x) = K (x t)v 00 (t) dt: (7:11)
a2 b2 0
We wish to hoose q1 in su h a way that u000 (; "; q ) = 0. Dierentiating u we obtain
u000 (; "; q ) = "v 000 (0 ) + "2 1 v iv (0 ) + "2 fw0000 (0 ) + q1 w1000 (0 )g + O("3 )

= "2 1 v iv (0 ) + w0000 (0 ) + q1 w1000 (0 ) + O("3 ); (7:12)
be ause v 000 (0 ) = 0 by (7.6). Remembering from (7.8) the expression for 1 we write
v iv (0 ) 0
1 v iv (0 ) = fw ( ) + q1 w10 (0)g:
v 00 (0 ) 0 0
(7:13)

We saw in (7.6) that v iv (0 )=v 00 (0 ) = p32 . When we use this in (7.13) and substitute
the result into (7.12), we nd that
u000 (; "; q ) = "2 X (q1 ) + O("3 );
where n o
3 3
X (q1 ) = w0000 (0 ) + p w00 (0 ) + q1 w1000 (0 ) + p w10 (0 ) : (7:14)
2 2
Thus we need to hoose q1 so that X (q1 ) = 0. Using the expression for w0 given in
(7.10a) and for w1 given in (7.11), we nd that
3
w00 (0 ) = 0 and w0000 (0 ) =
b
 7 
w10 (0 ) = and w1000 (0 ) = b:
3b 3
Therefore  7 
3
X (q1 ) =
b
+ q1
3
b p3 3b = 3b + q1 43 b;
2
so that X (q1 ) = 0 if
9 9
= p : q1 =
4b 2
2 2
Remembering (7.7), we on lude that the bran h 2a leaves the point (q; M ) = (q2 ; 1)
under an angle a given by p
3 2 2
tan a = = :
q1 3
(2) The bran h 2b . The solutions that sit on 2b have the properties
u00 (0) > 0 and u000 (1 ) = 0:
47
In the notation introdu ed in (7.1), we now need to ompute
v (0)
tan b = ;
q1
where now 0 denotes the lo ation of the rst maximum of v when q = q0 .
For solutions on this bran h, 1 is the rst point of symmetry. We expand u and q as
in (7.1) drop the subs ript 1 from 1 , and write
 (") = 0 + "1 + O("2 ): (7:15)
We emphasize that in the remainder of this se tion, 1 will denote the rst order term in
the expansion for  ("). From Lemma 5.1 we nd that
1
v (x) =
3
f os(ax) + 2 os(bx)g;
and
p
  1 2 7
0 = ; 0 = ; v (0 ) = ; v 00 (0 ) = ; v iv (0 ) = ;
3b b 2 3 3
so that
v iv (0 ) 7
00 = p :
v (0 ) 2
We now pro eed exa tly as in the previous ase, and we nd that
u000 (; "; q ) = "2 Y (q1 ) + O("3 );
where n o
7 7
Y (q1 ) = w0000 (0 ) + p w00 (0 ) + q1 w1000 (0 ) + p w10 (0 ) : (7:16)
2 2
For w0 and w1 and their derivatives we obtain at 0 :
2 5
w00 (0 ) = 3 and w0000 (0 ) = ;
9b 9b
0 
w1 (0 ) = and w1000 (0 ) = b:
3b
Therefore
 4
Y (q1 ) = + q1 b;
b 3
and Y (q1 ) = 0 if
q1 = p3 :
2 2
48
Sin e v (0 ) = 12 , it follows that
p
11 2
tan b = = :
2 q1 3
Near q = q2 and  = 1 (" = 0), we may use (7.1a) and the information about v and
w obtained in this se tion to make the following observations.
(i) There exists a Æ > 0 su h that in the ase that u00 (0) < 0 we have for q = q2
u(2 ) > 1 and u000 (2 ) < 0 for 1 Æ <  < 1:
Therefore, the proof of the existen e of u2a in Theorem 6.1 also holds for q = q2 .
(ii) Again in the ase that u00 (0) < 0, for every Æ > 0 we have
u000 (2 ) < 0 for q = q2 + "(q1 + Æ )
u000 (2 ) > 0 for q = q2 + "(q1 Æ )
for " > 0 suÆ iently small. Hen e, xing Æ > 0, there exists an "0 > 0 su h that for
q2 < q < q2 + "0 there are two periodi solutions u2a and u~2a whi h are symmetri with
respe t to 2 and su h that u(2 ) > 1. The rst one has 1 qq1 qÆ2 <  < 1 qq1 +qÆ2 , and it
is found by varying  between these values and sear hing for an  su h that u000 (2 ) = 0.
The se ond one has  < 1 qq1 qÆ2 and an be found as in the proof of Theorem 6.1. A
similar statement holds for the solutions of type u2b , but there is only one solution of this
type. Continuity of the bran hes near q = q2 and  = 1 an be proved using the Impli it
Fun tion Theorem, but we will not going into that here.

8. Even periodi solutions: q  q3

In Se tionp4 we have exhibited
p the existen e of an even single bump, or 1-lap periodi
solution for 8 < q < q1 = 8. In Se tion 6, we p extended this result and showed that
p even 2-lap periodi solutions for 8 < q < q2 and
there exist two two even 3-lap periodi
solutions 8 < q < q3 . One solution is onvex at the origin (u00 (0) > 0) and one solution
is on ave at the origin (u00 (0) < 0). The goal of this se tion is to extend these results to
the parameter regime q  q3 . It is important to note that the sharp qualitative results
obtained in Se tions 4 and 6 were proved by means of Lemma 2.4. However, this lemma
no longer applies in the range q  q3 . Thus, to determine the existen e and qualitative
properties of periodi solutions, we shall develop further analyti al te hniques.
We re all that an even periodi solution u is alled an n-lap periodi solution if it is
even, and symmetri with respe t to its nth positive riti al point n (and not symmetri
with respe t to any of the riti al points in between 0 and n ). In order to determine if an
n-lap periodi solution u attains a relative maximum or minimum at the point of symmetry,
one needs to know whether n is odd or even, as well as whether u00 (0) is positive or negative.
For onvenien e we list the orresponden e in the following table:
)
n = (n+1)=2 when n is odd
if u00 (0) > 0;
n = n=2 when n is even
49
 )
n = (n+2)=2 when n is even
if u00 (0) < 0:
n = (n+1)=2 when n is odd
In the present se tion we extend the above existen e theorems to n-lap periodi solu-
tions for arbitrary n  2.
Theorem 8.1. Let n  2, and let
p
8 < q < qn :
Then there exist two even n-lap periodi solutions, ua and ub , su h that

ua;b (n ) > 1 and u000

a;b (n ) = 0; (8:1)
whilst u00a (0) > 0 and u00b (0) < 0.

We prove Theorem 8.1 in two steps: rst, we establish the existen e of n-lap solutions
p (q3;2n 1 ; qn ), and then we show that these n-lap
whi h satisfy (8.1) in intervals of the form
solutions exist on the whole interval ( 8; qn ).
Lemma 8.2. Let n  2. Then for all q3;2n 1 < q < qn there exist two even periodi
solution ua and ub whi h are symmetri with respe t to their nth positive riti al point n ,
su h that
ua;b (n ) > 1 and u00a (0) > 0; u00b (0) < 0:
Proof. For the ases n = 2 and n = 3 we refer to the stronger results of Se tion 6.
Thus, in the remainder of this proof we assume that n  4.
We only onsider the ase where u00 (0) > 0. The ase u00 (0) < 0 is analogous. Let
u(x;  ) be a small amplitude, even, single bump periodi solution for whi h u00 (0) > 0.
Then  = M , where M = ku(
;  )k1 < 1. In parti ular, u(k ( );  ) < 1 and
u(k ( );  ) > 1 for all k  1 (see Figure 4.1).
Fix n  4 and dene m = (n + 1)=2 if n is odd, and m = n=2 if n is even. We set
a = supf >  : u(i (); )  1 for i = 1; : : : ; mg: (8:2)
We stress that this denition is ompletely dierent from the ones used so far. For all
 > a at least one of the maxima i with i = 1; 2; : : : m lies in the region fu > 1g.
We assert that a < 1. To see this we study the behaviour of u(x; ) when  is lose
to 1. Let v be the solution of the problem obtained by linearising around u = 1, whi h
was introdu ed in Se tion 5. Then, a ording to Lemma 5.2,
v (m) > 0 and v 000 (m ) < 0 if q 2 (q3;2n 1 ; qn ):
Therefore, there exists a onstant Æ > 0 su h that
u(m (); ) > 1 and u000 (m (); ) < 0 for 1 Æ <  < 1: (8:3)
50
>From the rst inequality in (8.3) we dedu e that a < 1, as laimed.
It follows from the denition of a and Lemma 2.1 that
u(k (a); a) < 1 for k = 1; : : : ; m 1 and u(m (a); a) = 1: (8:4)
>From (8.4) and the energy identity (1.8) it follows that u000 (m ) =
6 0 at a. Be ause
u(m 1 ) < u(m 1 ) < 1 it follows that
u000 (m (a); a) = u000 (m (a); a) = u000 (n (a); a) > 0; (8:5)
where we remark that n = m if n is odd, and n = m if n is even. We now dene
b = supf 2 (a; 1) : u(n ) > 1 on (a; )g; (8:6)
whi h is well-dened be ause of the denition of a, Equation (8.4) and Lemma 2.3. Besides,
we dene (in view of (8.5))
= supf 2 (a; 1) : u000 (n ) > 1 on (a; )g:
We now rst onsider the ase that n = m (i.e. n odd). If b = 1, then it follows
from (8.3) that < 1, thus u(x; ) is an even n-lap periodi solution whi h is symmetri
with respe t to m and u(m ) > 1.
If b < 1 we use the following result to obtain a solution.
Lemma 8.3. Let a and b be dened as in (8.2) and (8.6). If b < 1, then u000 (n ) < 0 at b.

We postpone the proof of Lemma 8.3 for a moment and rst nish the proof of
Lemma 8.2. We on lude from Lemma 8.3 that < b, and as we saw before this implies
that there exists an even periodi solution whi h is symmetri with respe t to m su h
that u(m ) > 1. The ase that n = m (i.e. n even) is dealt with in a similar manner,
but we have to distinguish three ases (the situation is similar to the proof of Part (b) of
Theorem 6.2). A ording to Lemma 5.2 we have
v 000 (m ) < 0 for all q 2 (q3;2n 1 ; qn );
and
v (m ) > 0 if q3;2n+1 < q < qn ;
v (m ) = 0 if q = q3;2n+1 ;
v (m ) < 0 if q3;2n 1 < q < q3;2n+1 :
If q3;2n+1 < q < qn then the proof is nished in the same way as above. If q3;2n 1 <
q < q3;2n+1 then b < 1 and the proof is nished with the help of Lemma 8.3. Finally, if
q = q3;2n+1 then we have, for Æ > 0 small enough,
u(m ) > 1 and u000 (m ) < 0 for 1 Æ <  < 1:
51
We now hoose an  2 (1 Æ; 1). If u(m ) > 1 then we nish the proof as in the ase
where q3;2n+1 < q < qn , whereas if u(m )  1 then we nish the proof as in the ase where
q3;2n 1 < q < q3;2n+1 . This ompletes the proof of Lemma 8.2.
Before we give the proof of Lemma 8.3 we introdu e some notation. We dene the
sets of maxima and minima in the region fu > 1g by
C+ = f1  k  m 1 : u(k ) > 1g;
C = f1  k  m 1 : u(k ) > 1g:
The following Proposition shows that for all  2 [a; b℄ we have u(k ) > 1 if and only if
u(k ) > 1. Besides, if u(` ) > 1 for some 1  `  m 1 then u( ) > 1 for all riti al
points  in between ` and m .
Proposition 8.4. For all  2 [a; b℄ we have

C+ = C = ; or C+ = C = f`; ` + 1; : : : ; m 1g for some ` 2 f1; 2; : : : m 1g: (8:7)

Proof. We rst noti e that u(0) < 1 and u(m ) > 1 for all  2 (a; b) (note that
u(m ) > 1 implies that u(m ) > 1). For  = a Equation (8.7) holds sin e C+ = C = ;.
It follows from the denition of a and Lemma 2.3 that C+ = C = ; for  2 (a; a + Æ )
with Æ > 0 suÆ iently small. We now use a ontinuation argument in  to show that
Equation (8.7) holds for all  2 [a; b). We dene
 = supf 2 (a; b) : Equation (8:7) holds on (a; )g;
and we suppose, for ontradi tion, that  < b. Then at  we have u(n ) = 1 for some
1  n  m 1. There are now two possibilities: either u(n 1 ) = 1 or u(n ) = 1.
Con erning the rst ase, it follows from Lemma 2.1 that at  = 
u(n 1 ) > 1 and u(n ) < 1; (8:8)
and by ontinuity (8.8) holds for  2 ( Æ;  ) with Æ > 0 suÆ iently small. However,
this ontradi ts the fa t that Equation (8.7) holds for all  2 (a; ). Thus at  there
there is no k 2 f1; 2; : : : ; m 1g su h that u(k ) = u(k 1 ) = 1.
Therefore, we must have u(n ) = u(n ) = 1 at  for some 1  n  m 1. As before,
it follows that
u(n+1 ) > 1 and u(n 1 ) < 1:
We assert that this implies that u(n 1 ) < 1. Namely, the possibility u(n 1 ) > 1 is
ex luded by the denition of  and the fa t that u(n 1 ) < 1. Besides, u(n 1 ) = 1 would
imply that u(n 2 ) = 0 whi h has already been ex luded above. Hen e u(n 1 ) < 1, and
thus also u(n 2 ) < 1. A repeated argument shows that u(k ) < 1 for all 1  k  n 1.
Analogously it is proved that u(k ) > 1 for all n + 1  k  m 1. By ontinuity this also
holds for  lose to  .
52
 Finally, by the denition of  there must exist a sequen e i #  su h that
u(n (i ); i ) > 1 and u(n (i ); i ) < 1:
The existen e of su h a sequen e is ex luded by the proof of Lemma 2.3, whi h an be
found in [PT5, Lemma 2.5℄. Hen e, having obtained a ontradi tion, we have proved that
Equation (8.7) holds on the entire interval  2 [a; b). The ase  = b follows by ontinuity.
Remark. It follows from the previous Proposition that u(k ) and u(k ) an only enter
and leave the region fu > 1g together.
Proof of Lemma 8.3. To prove Lemma 8.3 we argue by ontradi tion. Thus, suppose
that u000 (n ) > 0 at b. Then n = m = m , and it follows that u(m 1 ) < 1 at  = b.
Therefore, m 1 62 C . By Propostion 8.4, this implies that u(k ) < 1 for all 1  k  m 1,
so that C = ;. Besides, sin e C = C+ by Proposition 8.4, this also implies that C+ = ;.
Thus, we on lude that u(k )  1 for all 1  k  m 1 and u(m ) = u(m ) = 1 at  = b.
Sin e a was dened as the largest value of  for whi h this situation o urs, this situation
is ex luded and we have rea hed a ontradi tion.
We re all the denition
= supf > a : u000 (n ) > 0 on (a; )g:
Clearly < b  1 by the proof of Lemma 8.2.
Proposition 8.5. For all  2 [a; ℄ we have that

u000 (k ) > 0 and u000 (k ) > 0 for all k 2 C+ = C :

Proof. We rst noti e that sin e u(k ) and u(k ) only enter fu > 1g together, we must
have u000 (k ) > 0 and u000 (k ) > 0 at the point of entry. Suppose now, for ontradi tion,
that there exists a smallest  2 [a; ℄, for whi h Proposition 8.5 does not hold: let
d = supf > a : u000 (k ) > 0; u000 (k ) > 0 for all k 2 C+ g;
and suppose that d  . Then at  = d there is a riti al point  2 (0; n) su h that
u( ) > 1 and u000 ( ) = 0, i.e.,  is a point of symmetry.
Sin e by Proposition 8.4 all the riti al values between  and n lie above u = 1, it
follows that that u(x) > 1 for x 2 [; n ℄. In fa t, sin e  is a point of symmetry, we have
that u(x) > 1 for all x 2 [2 n ; n ℄. Sin e u(0) < 1, this means that 2 n > 0. By
symmetry, 2 n is a riti al point, and from the denition of d we see that u000 (2 n )  0.
Therefore, again by symmetry, u000 (n )  0, so that from denition of it follows that
 d. Sin e by assumption, d  , we on lude that d = , and u000 (n ) = 0. But then u is
symmetri with respe t to both  and n . This means that u(x) > 1 for all x 2 R, whi h
ontradi ts the fa t that u(0) < 1.
Remark. Another way of obtaining the above ontradi tion is via the observation that

(u000 + qu0 )0 = u(1 u2 ) < 0 on (; n ):

53
Upon integrating over (; n) we obtain that u000 (n ) < 0 at  = d, ontradi ting the
assumption that d  .
Lemma 8.6. Let n  4. For every q 2 [qn 1 ; qn ) there exist two even n-lap solutions ua
and ub su h that ua;b (n ) > 1 and u000 00 00
a;b (n ) = 0, whilst ua (0) > 0 and ub (0) < 0.

Proof. Sin e q3;2n 1 < qn 1 for n  4, it follows immediately from Lemma 8.2 that
there exist two periodi solutions whi h are symmetri with respe t to n , and su h that
ua;b (n ) > 1. We assert that n is the rst point of symmetry. Proposition 8.5 ensures that
n is the rst point of symmetry in the region fu > 1g. Besides, there is no riti al point
0 <  < n with u000 ( ) = 0 in the region fu  1g, sin e that would imply (by Proposition
8.4 and the symmetry with respe t to 0) that u(x)  1 for all x 2 R, ontradi ting the
fa t that u(n ) > 1.
p 8.1. It shows that the n-lap
The following Lemma is a reformulation of Theorem
solutions obtained in Lemma 8.6 exists for all q 2 ( 8; qn ).
p
Lemma 8.7. Let n  2 and let 8  q < qn . For any N  n there exist two even
N -lap solutions ua and ub su h that ua;b (N ) > 1 and u000 00
a;b (N ) = 0, whilst ua (0) > 0 and
00
ub (0) < 0.
p
Proof. The range q 2( 8; q3 ) was already overed in Se tions 4 and 6. For q 2
[qn 1 ; qn ) with n  4 we see from Lemma 8.6 that there exists two n-lap solutions. For
N > n we again restri t our attention to the ase u00 (0) > 0, the other ase being ompletely
analogous. We an dene a, b and as before for both n and N . We have an < n < bn  1
and aN < N  bN  1. If N < 1 then learly we have an N -lap solution. Arguing by
ontradi tion we assume that N = 1. It follows dire tly from the denition of a and (8.4)
that aN < an . Proposition 8.5 shows that u000 (n ) > 0 if u(n ) > 1 for all  2 (aN ; 1℄.
However, u000 (n ) = 0 for  = n > an > aN , a ontradi tion.

9. Proof of Lemma 2.4

In this se tion we prove Lemma 2.4. We re all the setting: we assume that u is a
solution of equation (1.1), and that a 2 R is a riti al point where u has the following
properties:
u(a) = 1; u0 (a) = 0; u00 (a) = 0 and u000 (a) > 0: (2:13)
Then u0 > 0 in a right-neighbourhood of a so that the point

b = supfx > a : u0 > 0 on (0; x)g (2:14)

is well dened. By Lemma 2.2, it is also nite.

Lemma 2.4. Suppose that
p p  1
8 < q < q3 = 2 3 + :
3
54
Let u be a solution of equation (1.1) whi h at a point a 2 R has the properties listed in
(2.13). Then, at its next riti al point b dened by (2.14), we have
u(b) > 1; u0 (b) = 0; u00 (b) < 0 and u000 (b) < 0: (2:15)
p p
Proof. If 8 < q  8 we use the fun tion H introdu ed in Se tion 4. Sin e u > 1
on (a; b), it follows that H 00 > 0 on (a; b). At riti al points of u, we have by (4.7)
H 0 = u00 u000 :
Hen e H 0 = 0 at a, and therefore H 0 > 0 at b. Sin e u00 (b)  0, and even u00 (b) < 0 by the
rst integral, it follows that u000 (b) < 0, as asserted.
p
If q > 8 we an no longer prove that H 00 > 0 on (a; b) and we have to pro eed
dierently.
Without loss of generality we may assume that a = 0 and onsider the initial value
problem (
uiv + qu00 + u3 u = 0; (9.1a)
0 00 000
u(0) = 1; u (0) = 0; u (0) = 0 u (0) = ; (9.1b)
where the initial values are derived from (2.13) and  is a positive number. We denote
the solution of Problem (9.1) by u(x; ), and its rst riti al point, orresponding to b, by
 ():
 () = supfx > 0 : u0 (
; ) > 0 on (0; x)g:
We need to show that
u000 ( (); ) < 0 for every  > 0: (9:2)
As a rst step we show that (9.2) is satised for  large enough. This an be proved
by means of a simple s aling argument. With the variables
x= 1=5
t and u(x; ) = 2=5 w(t; );  > 0; (9:3)
the initial value problem (9.1) an be written as
(
wiv + q 2=5
w00 + w3  4=5 w = 0;
w(0) =  2=5
; w0 (0) = 0; w00 (0) = 0; w000 (0) = 1:
By standard ODE arguments, w(t; ) ! W (t) on ompa t intervals, as  ! 1, where W
is the solution of the limit problem
(
W iv + W 3 = 0;
W (0) = 0; W 0 (0) = 0; W 00 (0) = 0; W 000 (0) = 1:
Plainly,
T = supft > 0 : W 0 > 0 on (0; t)g < 1
55
and W 000 (T ) < 0. Hen e, by ontinuity, for  large enough, w000 (
; ) < 0 at the rst zero
of w0 (
; ). Thus, u000 ( (); ) < 0 for  large enough.
Pro eeding with the proof of Lemma 2.4, we suppose that (9.2) is false and that
u ( (0 ); 0 )  0 for some 0 > 0. Then there exist a onstant  2 [0 ; 1) su h that
000
u000 ( ( );  ) = 0. At   =  ( ) we then have
u0 > 0 on (0;  ); u0 (  ) = 0 and u000 (  ) = 0: (9:4a)
This means that u(
;  ) is symmetri with respe t to   and that
u0 < 0 on (  ; 2  ); u(2  ) = 1 and u0 (2  ) = 0: (9:4b)
For further referen e, we introdu e the notation

p1 1 ()2 :


u(  ) =  and hen e u00 (  ) = (9:4 )
2
In the remainder of the pproof we show that if 1 < a=b < 3 , where a and b have been
dened in (5.2b) (i.e. if 8 < q < q3 ), then a solution u(x;  ) with the properties listed
in (9.4) annot exist. We shift the origin to x =   and u = 1 and write
x =   + x~ and u(x) = 1 + w(~x): (9:5)
Then w is the solution of the problem
8 iv
<w + qw00 + 2w = w2 (3 + w); (9.6a)
1

: w (0) =  1; w 0 (0) = 0; w 00 (0) = p 1 ( )2 ; w 000 (0) = 0; (9.6b)
2
where we used (9.4) and have dropped the tilde. We shall ompare the solution w =
w(x;  ) of Problem (9.6) with the solution v of the linear problem
8 iv
<v + qv 00 + 2v = 0; (9.7a)
1
: v (0) =  1; v 0 (0) = 0; v 00 (0) = p (1 2 ); v 000 (0) = 0; (9.7b)
2
where  > 1 is an arbitrary number, whi h eventually will be hosen equal to  . We
denote the solution of Problem (9.7) by v (x) or v (x; ). An elementary omputation
shows that v (x) an be written as
v (x) = A os(ax) + B os(bx); (9:8)
where we re all that a and b are the positive roots of the equation
k4 qk2 + 2 = 0;
56
given by
1 p 1 p
a2 = (q + q 2 8 ) and b2 = (q q 2 8 ): (9:9)
2 2
The oeÆ ients A and B are given by
b2 ( 1) +  a2 ( 1) + 
p1 (2 1):
A= and B = ; = (9:10)
a2 b2 a2 b2 2
Let
x0 () = supfx > 0 : v (
; ) > 0 on [0; x)g:
In order to pro eed with our omparison argument we need the following result.
Lemma 9.1. Suppose that
p a
8 < q < q3 or ; equivalently ; 1< < 3:
b
Then for any  > 1 we have x0 () < 1 and v 0 (
; ) < 0 on (0; x0 ()℄.
We postpone the proof of this lemma until after the proof of Lemma 2.4 has been ompleted.
We now ontinue with the proof of Lemma 2.4. Let
y0 () = supfx > 0 : w(
; ) > 0 on [0; x)g:
Our goal is to prove that w0 (y0 (); ) < 0 for any  > 0. Translating this result ba k to
the variables x and u, we nd in parti luar that u0 (2  ) < 0, whi h ontradi ts (9.4b) and
ompletes the proof.
By the variation of onstants formula, we nd that
Z xn o
1 1 1
w(x) = v (x) sin(bt) sin(at) h(x t) dt; (9:11)
a2 b2 0 b a
where h(s) = w2 (s)f3 + w(s)g  0 as long as w  0, i.e. on [0; y0℄. Note that
Z
1 x
w0 (x) = v 0 (x) f os(bt) os(at)gh(x t) dt: (9:12)
a2 b2 0

At the rst zero x0 of v we have

Z xn o
1 1 1
w(x0 ) = sin(bt) sin(at) h(x0 t) dt:
a2 b2 0 b a
Hen e, if
1 1
K (t) def
= sin(bt) sin(at) > 0 for 0 < t < x0 ; (9:13)
b a
57
then w < v on (0; x0 ℄, and therefore y0 < x0 . In addition, if
K 0 (t) = os(bt) os(at) > 0 for 0 < t < x0 ; (9:14)
then w0 < v 0 < 0 on (0; y0℄ by (9.12). In parti ular, w0 (y0 ) < 0, as asserted.
The on lusion of the proof onsists of a detailed analysis of the fun tion K (t) to show
that (9.13) and (9.14) hold on (0; y0℄. An elementary omputation shows that
K (0) = 0; K 0 (0) = 0; K 00 (0) = 0; K 000 (0) = a2 b2 > 0:
Hen e K > 0 and K 0 > 0 in a right-neighbourhood of the origin. We set
t1 = supft > 0 : K > 0 on (0; t)g;
and
t0 = supft > 0 : K 0 > 0 on (0; t)g:
Plainly 0 < t0 < t1 and K 0 (t1 )  0. We re all the assumption that
a   3 3
1<
b
<3 () < < < :
2a 2b 2a 2b
(9:15)

Be ause a=b > 1, it follows from (9.13) and (9.15) that t1 2 ( 2b ; 32b ), and hen e, that
t0 < 32b . On the other hand, sin e 2a < 2b it is lear that K 0 > 0 on (p; 2a ℄ so that
t0 > 2a . Observe that
 3 
os(at) < 0 on ( ; ) and os(bt) > 0 on (0; ):
2a 2a 2b
Be ause a=b < 3 and hen e 3
2a
>  , it follows that K 0 > 0 on [  ;  ℄, and we on lude
2b 2a 2b
that
 3
< t0 < : (9:16)
2b 2b
The value of v at t0 an easily be omputed by using the formula (9.14) for K 0 in the
expression (9.8) for v . We obtain
 1
v (t0 ) = os(bt0 );
a2 b2
whi h, in view of (9.16), shows that v (t0 ) < 0. We on lude that x0 < t0 < t1 , and hen e
that the properties (9.13) and (9.14) of respe tively K and K 0 are true. This ompletes
the proof of Lemma 2.4.
Proof of Lemma 9.1. We rst show that the assertion is true for
p
 > q 2 1: (9:17)
58
Note that
(v 000 + qv 0 )0 = 2v < 0 as long as v > 0; (9:18)
and hen e, still as long as v > 0,
1 +  
v 00 (x) + qv (x)  v 00 (0) + qv (0) = (1 ) p q :
2
Therefore, if (9.17) holds then
v 00 < qv < 0 as long as v > 0;
so that x0 () < 1 and v 0 (
; ) < 0 for 0 < x  x0 ().
Let
 = inf f^ > 1 : x0 () < 1 and v 0 (
; ) < 0 for ^ <  < 1g:
If  = 1, then the assertion is proved. Therefore, we suppose that  > 1. We distinguish
two ases:
(i) x0 ( ) = 1 and v 0 (
; )  0 for 0  x < 1, or
(ii) x0 ( ) < 1 and v 0 (x0 ;  ) = 0, for some x1  x0 ( ).
To keep the notation as simple as possible, we shall hen eforth omit the asterisk when
referring to  .
Case (i): Be ause v 0 = 0 and v 000 = 0 at the origin, integration of (9.18) over (0; x) shows
that
v 000 (x) + qv 0 (x) < 0 for x > 0:
Hen e, integrating (9.18) again, but now over (1; x) we on lude that
(v 00 (x) + qv (x))0 < v 000 (1) + qv (1) = Æ for x > 1;
where Æ is a postive onstant. Hen e
v 00 (x) + qv (x) < C Æ (x 1) and therefore v 00 (x) < C Æ (x 1) for x > 1;
where C = v 00 (1) + qv (1) is a onstant. Thus v 00 (x) ! 1 as x ! 1, whi h implies that
x0 () < 1, a ontradi tion.
Case (ii): Suppose that x1 < x0 (). Then
v (x1 ) > 0; v 0 (x1 ) = 0 and v 00 (x1 ) = 0: (9:19)
When we multiply equation (9.7a) by v 0 , integrate over (0; x) and use the initial onditions,
we obtain
1 00 2 q 0 2 2 n 1 o
v 0 v 000 (v ) + (v ) + v = ( 1)2 1 (1 + )2 for x 2 [0; x0℄;
2 2 4
59
Evaluating the left hand side at x1 , using the properties of v at x1 listed in (9.19), we
obtain n o
1
v 2 (x1 ) = ( 1)2 1 (1 + )2 : (9:20)
4
Sin e  > 1, the right hand side of (9.20) is negative, whilst the left hand side is nonneg-
ative, a ontradi tion.
It remains to onsider the ase that x1 = x0 , so that
v (x0 ) = 0 and v 0 (x0 ) = 0: (9:21)
We begin with a preliminary result.
Lemma 9.2. Suppose that 1 < a=b < 3. Then, if the solution v of Problem (9.7) has the
properties (9.21), the point x0 must lie in the interval (0; 2b ).
Proof. We re all the formula for v :

v (x) = A os(ax) + B os(bx); (9:22)

where a and b are given in (9.9) and A and B are given in (9.10). We dis uss in su ession
the three ases:
(i) A = 0; (ii) A < 0 and (iii) A > 0:
Case (i) A = 0: Sin e A + B =  1, it follows that in this ase B =  1 and hen e
v (x) = ( 1) os(bx):
Thus, x0 = 2b . However, v 0 (x0 ) < 0, whi h ontradi ts (9.21), so that A annot be zero.
Case (ii) A < 0: Observe that in this ase
B os(bx0 ) = jAj os(ax0 );
and B =  1 A =  1 + jAj > jAj. Remembering that a > b, we on lude that

B os(bx) > jAj os(ax)  0 if 0  x  :
2a
By assumption
 a  3
< < : (9:23)
2 b2 2
and hen e  
B os(bx)  0 > jAj os(ax) if <x :
2a 2b
Therefore
 3
< x0 < : (9:24)
2b 2b
60
This implies that


v 00 + a2 v = (a2 b2 ) os(bx) < 0 on ;x ;
2b 0
and hen e, be ause v 0 < 0 on (0; x0),


f(v0 )2 + a2v2 g0 = 2(v00 + a2v)v0 > 0 on ;x :
2b 0
When we integrate this inequality over ( 2b ; x0 ), we nd that

(v 0 )2 (x0 ) > f(v 0 )2 + a2 v 2 g > 0;
=2b

whi h means that v 0 (x0 ) < 0. This ontradi ts (9.21), so that the ase A < 0 annot o ur
either.
Case (iii) A > 0: We have

a 

v = A os :
2b b2



But, in view of (9.23), os ab 2 < 0. This means that v 2b < 0, so that x0 < 2b .
Summarising, we have found that the onstant A in (9.22) must be positive, and
hen e, that x0 < 2b .
Proof of Lemma 9.1 { Con lusion. Using the expli it expression (9.22) for v , we dedu e
from (9.21) that the onstants A and B must satisfy the equations
A os(ax0 ) + B os(bx0 ) = 0 and aA sin(ax0 ) + bB sin(bx0 ) = 0: (9:25)
Sin e A + B =  1, they annot both be equal to zero. Therefore, the determinant of
the system (9.25) of equations must be zero, and hen e
os(bx0 ) sin(ax0 ) = os(ax0 ) sin(bx0 ): (9:26)
By Lemma 9.2, os(bx0 ) > 0. Hen e, if os(ax0 ) = 0, then sin(ax0 ) = 0 as well, and this
is impossible. Thus, os(ax0 ) 6= 0 and we may divide (9.26) by os(ax0 ) os(bx0 ). We thus
nd that x0 must be a solution of the equation
a 
 

tan(bx) = tan(ax)
b
in 0;
2a
[ 2a ; 2b ; (9:27)

be ause 3
2a
> 
2b
a ording to (9.23). With  = a=b and bx = t, we an write (9.27) as

 

(t) def
=  tan(t) tan(t) = 0 in I def
= 0;
2
[ ; :
2 2
(9:28)

Thus, we seek a root  = bx0 of equation (9.28).

61
 Lemma 9.3. If 1 <  < 3, then equation (t) = 0 has no roots in the set I .
Proof. Observe that

 tan(t) > tan(t) if 0 < t < (9:29a)
2
and  
 tan(t) < 0 < tan(t) if <t< : (9:29b)
2 
If   2, then =2  =2, so that it is immediately lear from (9.29) that  annot have
a zero in I .
If 2 <  < 3, then = < =2 < 3=2 and it follows that if (t) has a zero in I , then it
must lie in the interval (  ; 2 ), where both terms in (t) are positive. Plainly,
 
( ) = tan
 
< 0 and (t) ! 1 as t ! 2 : (9:30)

Suppose that (t) has a zero, and that t0 is the largest zero on (  ; 2 ). Then (9.30) implies
that 0 (t0 )  0. However, an easy omputation shows that
0 (t) = 2 1 > 0 when (t) = 0:
Thus, we have a ontradi tion, and (t) annot have a zero in I . This ompletes the proof
of Lemma 9.1.
Referen es
[Be1℄ Berg, G.J.B. van den, MS Thesis, Mathemati al Institute, Leiden University, 1996.
[Be2℄ Berg, G.J.B. van den, Uniqueness of solutions for the extended Fisher-Kolmogorov
equation, C. R. A ad. S i. Paris, 326 Serie I (1998)447-452.
[Be3℄ Berg, G.J.B. van den, The phase plane pi ture for a lass of fourth-order onserva-
tive dierential equations, Report Mathemati al Institute, Leiden University W98-23,
1998.
[BCT℄ Buoni, B., Champneys, A.R. & J.F. Toland, Bifur ation and oales en e of a plethora
of homo lini orbits for autonimous Hamiltonian systems, J. Dynami s and Dieren-
tial Equations 8 (1996) 221-279.
[CH℄ Cross, M.C. & Hohenberg, P.C., Pattern formation outside of equilibrium, Rev. Mod.
Phys. 65 (1993) 851-1112.
[CE℄ Collet, P. & J.P. E kmann, Instabilities and fronts in extended systems, Prin eton
Series in Physi s, Prin eton University Press, 1990.
[CER℄ Coullet, P., C. Elphi k, & D. Repaux, Nature of spatial haos, Phys. Rev. Lett. 58
(1987) 431-434.
[CT℄ Champneys, A.R. & J.F. Toland Bifur ation of a plethora of multi-modal homo lini
orbits for autonimous Hamiltonian systems, Nonlinearity 6 (1993) 665-722.

62
 [CTh℄ Champneys, A.R. & J.M.T. Thompson, A multipli ity of lo alised bu kling modes for
twisted rod equations, Pro . Royal So . of London A 452 (1996) 2467-2491.
[CMM℄ Coleman, B.D., M. Mar us & V.J. Mizel, On the thermodynami s of periodi phases,
Ar h. Rational Me h. Anal. 117 (1992) 321-347.
[D℄ Doedel, E.J. et :al :, AUTO97: Continuation and Bifur ation Software for Ordinary
Dierential Equations, available by ftp from ftp. s. on ordia. a
[DB℄ Danani , V. & A. Bjelis, Domain patterns in in ommensurate systems with the uni-
axial real order parameter, Phys. Rev. E 50 (1994) 3900-3910.
[DS℄ Dee, G.T. & van Saarloos, W., Bistable systems with propagating fronts leading to
pattern formation, Phys. Rev. Lett. 60 (1988) 2641-2644.
[E℄ Ermentraut, G.B., Phaseplane, omputer software for solving ODE's.
[GH℄ Guggenheimer, J. & P.J. Holmes, Nonlinear os illations, dynami al systems, and bi-
fur ations of ve tor elds, Applied Math. S i., Springer Verlag, New York, 1983.
[KKV℄ Kalies, W.D., J. Kwapisz & R.C.A.M van der Vorst, Homotopy lasses for stable
onne tions between Hamiltonial saddle-fo us equilibria, Comm. Math. Phys. 193
(1998) 337-371.
[KV℄ Kalies, W.D. & R.C.A.M van der Vorst, Multitransition homo lini and hetero lini
solutions to the Extended Fisher-Kolmogorov equation, J.Di. Equ. 131 (1996) 209-
228.
[LM K℄ Lazer, A. C, & P.J. M Kenna, Large-amplitude os illations in suspension bridges:
some new onne tions with nonlinear analysis, SIAM Rev. 32 (1990) 537-578.
[LM℄ Leizarowitz, A. & V.J. Mizel, One dimensional innite-horizon variational problems
arising in ontinuum me hani s, Ar h. Rational Me h. Anal. 106 (1989) 161-194.
[MPT℄ Mizel, V.J., L.A. Peletier & W.C. Troy, Periodi phases in se ond order materials, to
appear in Ar h. Rational Me h. Anal. 145 (1998) 343-382.
[PT1℄ Peletier, L.A. & W.C. Troy, Spatial patterns des ribed by the Extended Fisher-
Kolmogorov (EFK) equation: Kinks, Di. Int. Eq. 8 (1995) 1279-1304.
[PT2℄ Peletier, L.A. & W.C. Troy, Spatial patterns des ribed by the Extended Fisher-
Kolmogorov (EFK) equation: Periodi solutions, SIAM J. Math. Anal. 28 (1997)
1317-1353.
[PT3℄ Peletier, L.A. & W.C. Troy, A topologi al shooting method and the existen e of kinks
of the Extended Fisher-Kolmogorov equation, Topol. Methods Nonlinear Analysis 6
(1996) 331-355.
[PT4℄ Peletier, L.A. & W.C. Troy, Chaoti spatial patterns des ribed by the Extended
Fisher-Kolmogorov equation, J. Di. Eqns 126 (1996) 458-508.
[PT5℄ Peletier, L.A. & W.C. Troy, Multi bump periodi travelling waves in suspension
bridges, Pro . Roy. So . Edinb. 128A (1998) 631-659.
[PM℄ Pomeau, Y., & P. Manneville, Wavelength sele tion in ellular ows, Phys. Letters
75A (1980) 296-298.
Appendix

63
 We linearize around the onstant solution u = 1 of the equation
uiv + qu00 + u3 u = 0: (A:1)
Thus, we write u = 1 + "v and substitute into the initial value problem for u. Omitting
higher order terms, we obtain
(
v iv + qv 00 + 2v = 0: (A.2a)
p
v (0) = 1; v 0 (0) = 0; v 00 (0) =  2; v 000 (0) = 0: (A.2b)
We denote the orresponding solutions by v (x).
Substitution of v (x) = ex yields the hara teristi equation
4 + q2 + 2 = 0: (A:3)
We distinguish two ases:
p p p
Case A : 8<q< 8 and Case B : 8  q < 1:

Case A: Equation (A.3) has roots

 = a  ib; (A:4a)
in whi h a > 0 and b > 0 are given by
1
qp 1
qp
a= 8 q and b = 8 + q: (A:4b)
2 2
An elementary omputation shows that the solutions v+ (x) and v (x) of Problem (A.2)
are given by
v (x) = osh(ax) os(bx) + K sinh(ax) sin(bx); (A:5a)
where s p
a
K+ = =
b
p8 q and K =
b
a
: (A:5b)
8+q
Thus,
v+0 (x) = (K+ a + b) osh(ax) sin(bx)
v 0 (x) = (K b a) sinh(ax) os(bx);
and we see that for v (x), we obtain
 3
1 () ! p p and 2 () ! p p as  ! 1: (A:6a)
8+q 8+q
64
and, sin e " = 1 ,
b  a 
u (1 (); )  1 (1 ) sinh as  ! 1: (A:6b)
a 2b

Case B: Equation (A.3) has roots

 = ia and  = ib; (A:7a)
where
1 p 1 p
a2 = (q + q 2 8 ) and b2 = (q q 2 8 ): (A:7b)
2 2
For the solutions v (x) of Problem (A.2) we nd
v (x) = A os(ax) + B os(bx); (A:8a)
in whi h p p
b2  2 a2  2
A = 2 2 and B = : (A:8b)
a b a2 b2
Properties of the solutions v (x) and their riti al points and values are given in Lemma
5.2.

65