Electronic Journal of Qualitative Theory of Differential Equations 2011, No. 38, 1-17; http://www.math.u-szeged.hu/ejqtde/
Asymptoti stability of two dimensional
systems of linear dierene equations and of
seond order half-linear dierential equations
with step funtion oeients
L. Hatvani
1 ∗
and L. Székely
2
1
Corresponding author
University of Szeged, BolyaiInstitute
Aradivértanúk tere 1, Szeged, H-6720 Hungary
E-mail: hatvanimath.u-szeged.hu
2
Szent István University
Institute of Mathematis and Informatis
E-mail: szekely.laszlogek.szie.hu
Abstrat
We give a suient ondition guaranteeing asymptoti stability
with respet to
x
for the zero solution of the half-linear dierential equationx ′′ | x ′ | n− 1 + q(t) | x | n− 1 x = 0, 1 ≤ n ∈ R ,
withstepfuntionoeient
q
. Thegeometrimethodoftheproofanbe applied alsoto two dimensional systems oflinear non-autonomous
dierene equations. The appliation gives a new simple proof for a
sharpenedversionofÁ.Elbert'sasymptotistabilitytheoremsforsuh
diereneequationsandlinearseondorderdierentialequationswith
step funtionoeients.
∗
Supported by the Hungarian National Foundation for Sienti Researh (OTKA
K75517), Analysis and Stohastis Researh Group of the Hungarian Aademy of Si-
enes, andTÁMOP4.2.2.
funtion oeients; half-lineardierentialequation.
2000 Mathematis Subjet Classiation: Primary 34D20, 39A30.
1 Introdution
Consider the diereneequation
x n+1 = M n x n , n = 0, 1, 2, . . . ,
(1)where
x n ∈ R 2 and M n ∈ R 2 × 2. We do not onsider the trivial ase when
all the entries of M n are equal to 0
for some n
. Let k M k
be the spetral
M n are equal to 0
for some n
. Let k M k
be the spetral
norm, i.e.,
k M k
isthe square rootof thelargest eigenvalue of the symmetripositive semi-denite matrix
M T M
. It is well-known [3, p. 232℄ that ifQ ∞
n=0 k M n k = 0, then all solutions of equation (1)tend to zero as n → ∞
,
i.e., thezerosolutionisasymptotiallystable. Á.Elbert[10℄gaveasuient
ondition for the asymptoti stability under the assumptions
(i) Q ∞
n=0 max {k M n k , 1 } < ∞,
(ii) 0 < Q ∞
n=0 k M n k,
(iii) Q ∞
n=0 max {| det M n | , 1 } < ∞.
His proofwasbased onestimation ofthe normof some speial matriesand
a triky deomposition of matries
M n. He applied this result to dedue
anArmellini-Tonelli-Sansone-typetheorem(abbreviated asA-T-S theorem),
i.e., a theorem guaranteeing asymptoti stability with respet to
x
for thezero solution of the linear seondorder dierentialequation
x ′′ + a(t)x = 0 (a(t) ր ∞ , t → ∞ )
(2)with step funtion oeient
a
[11, 12℄.I.Bihari[5℄ andElbert [9℄introduedthehalf-lineardierentialequation
x ′′ | x ′ | m−1 + q(t) | x | m−1 x = 0, m ∈ R + ,
(3)whihhasattratedattention,andithasanextensiveliterature(see,e.g.,[7℄,
[8℄and thereferenes therein). Bihari[6℄has generalizedthe A-T-S theorem
tothisequationintheaseofsmoothoeient
q
,requiringregulargrowthof
q
. Roughly speaking, this ondition means that the growth ofq
annotbe loated to a set with small measure (see Setion 3). Of ourse, a step
funtion
q
doesnotsatisfythisondition. Elbert's method,usingawideanddeep mahinery from linear analysis,does not apply tothe half-linear ase.
InthispaperweestablishanA-T-Stheoremforthehalf-lineardierential
equationwithstepfuntionoeient
q
. Theproofisbaseduponageometrimethod. This method applies also to the linear ase, so we an give a new
simple proof for Elbert's result,assuming only
lim sup n→∞ Q n
k=0 k M k k < ∞
instead of
(i)
(iii)
.2 Dierene equation
To investigateequation(1), wewilldeneadiereneequationonthe plane
whih has the same stability properties as equation (1). Let us introdue
the following notations for the matriesof the reetion with respet to the
x
-axis, and ofthe rotationaroundthe origin ounterlokwise withϕ
inR 2:
R =
1 0 0 − 1
, E(ϕ) =
cos ϕ − sin ϕ sin ϕ cos ϕ
.
(4)Obviously,
E(ϕ 1 )E(ϕ 2 ) = E(ϕ 1 + ϕ 2 ), E(ϕ)R = RE( − ϕ).
(5)We willneed the following theorem (see, e.g., [16, p. 188℄):
Theorem (polar fatorization). Every
M ∈ R n×n an be represented as a
produt M = SQ
where S
is symmetri, positive semi-denite, and Q
is
orthogonal.
S
is uniquely determined whileQ
is unique if and only ifM
isnon-singular.
In this theorem
S
is the square root of the symmetri positive semi-denitematrix
M T M
. IfM ∈ R n × nisnon-singular,thentheprodutM T M
ispositivedenite,thusitan bediagonalized:
M T M = PD 2 P − 1,whereD 2
isthediagonalmatrixontainingtheeigenvaluesof
M T M
andtheorthogonalmatrix
P
has the propereigenvetorsinitsolumns. ThenS = PDP − 1 and
M = PDP − 1 Q.
(6)
Denote by
Λ
andλ
the eigenvalues ofM T M ( k M k = Λ ≥ λ > 0)
. Supposethat the diagonalelements in
D
are in dereasing order. Ifdet M = 0
, thenS
is positive semi-denite and the symmetri matrixS ˜ := k M k − 1 S
an berepresented as
S ˜ = P DP ˜ −1, whereP
is orthogonaland
D ˜ =
1 0 0 0
.
Applying the aboveargument tothe oeientmatries of (1), we have
M n = k M n k P n D ˆ n P − n 1 Q n ,
(7)where
D ˆ n :=
1 0 0 d n
, d n :=
( q λ n
Λ n > 0, if det M n 6 = 0
;
0,
if det M n = 0
.
(8)
Letusexaminetheow
F n := Q n
k=0 M k
ofequation(1). Usingthefat,thatthe produt of orthogonal matriesare alsoorthogonal,
F n has the form
F n =
n
Y
k=0
P k D ˆ k P − k 1 Q k =
n
Y
k=0
k M k k
! P n
n
Y
k=0
D ˆ k O k
!
,
(9)where the orthogonal matries
O k (k = 0, . . . , n + 1)
are dened byO 0 := P −1 0 Q 0 , O k = P −1 k Q k P k− 1 , k = 1, . . . , n,
(10)and the produt
Q n
k=0 N k
ismeantin the orderN n · · · N 0. It isknown from
the elementary geometry that in the plane every orthogonal transformation
is a rotation or a produt of a rotation and a reetion with respet to the
x
-axis. Thus, ifO k is not a rotation, then let O k = E(ϑ k )R
for some ϑ k.
Sine
R
is ommutable with every diagonalmatries, from (5)we obtainF n =
n
Y
k=0
k M k k
!
R m E(α n )
n
Y
k=0
D ˆ k E(ω k )
!
(11)
for some
m ∈ N 0 (m ≤ n + 1
)andsome ω k's,whereα k,ω k an bealulated
α k,ω k an bealulated
from
M 0 , . . . , M k.
x n+1 = k M n k
1 0 0 d n
cos ω n − sin ω n
sin ω n cos ω n
x n ,
0 ≤ d n ≤ 1, n = 0, 1, 2, . . .
(12)
The equilibrium
(0, 0)
of (1) is stable (asymptotiallystable) if and only if the equilibrium(0, 0)
of (12) is stable (asymptotiallystable). Now, we an state the main theorem of this setion:Theorem 1. Suppose that
lim sup n→∞ Q n
k=0 k M k k < ∞. If
∞
X
n=0
min { 1 − d n , 1 − d n+1 } sin 2 ω n+1 = ∞ ,
(13)then the zero solution of dierene equation (12) is asymptotially stable.
Proof. Obviously, itisenoughto dealwith the ase
k M k k = 1
(k = 0, 1, . . .
)and to showthat
Q ∞
n=0 D ˆ n E(ω n )
=0. Geometrially,the dynamis of (12) is omposed of onseutive rotationsand ontrations along they
-axis. Letus introdue polaroordinates
r
,ϕ
so thatx :=
x y
, x = r sin ϕ, y = r cos ϕ.
Inthese oordinatesthe phasespaeforsystem(12) is
r ≥ 0
,−∞ < ϕ < ∞
.Using the notations
˜
x n = E(ω n )x n , κ n := ϕ n+1 − (ϕ n + ω n ), ∆r n := r n+1 − r n , n = 0, 1, . . .
we have
p x 2 n + y n 2 = p
˜
x 2 n + ˜ y n 2 , x n+1 = ˜ x n , y n+1 = d n y ˜ n
ϕ n+1 = ϕ 0 +
n
X
i=0
(ω i + κ i ), r n+1 = r 0 +
n
X
i=0
∆r i ,
and
∆r i ≤ 0
beause of the ontration. Therefore, the sequene{ r n } ∞ n=0 is
monotonously dereasing.
Supposethatthestatementofthetheoremisnottrue,i.e.,
¯ r := lim n→∞ r n
> 0
. Then− ∆r i = r i − r i+1 = q
x 2 i + y i 2 − q
x 2 i+1 + y i+1 2
= q
˜
x 2 i + ˜ y 2 i − q
˜
x 2 i + d 2 i y ˜ 2 i = (1 − d 2 i )˜ y i 2 p x ˜ 2 i + ˜ y i 2 + p
˜
x 2 i + d 2 i y ˜ 2 i
≥ (1 − d 2 i )r 2 i cos 2 (ϕ i + ω i ) 2r i ≥ ¯ r
2 (1 − d i ) cos 2 (ϕ i + ω i ).
(14)
Wewanttogetthe ontraditionthat thesum ofthe lowerestimatingterms
in(14)diverges. Theproblemisthatthesetermsontain
ϕ i's,whihdepend
on solutions,so they are unknown; we haveto get rid of them. Obviously,
| cos(ϕ i + ω i ) | = | cos ϕ i cos ω i − sin ϕ i sin ω i |
≥ | sin ϕ i || sin ω i | − | cos ϕ i || cos ω i | .
(15)For arbitrarily xed
0 < γ < ε < 1
, deneµ(ε, γ) := p
1 − γ 2 − εγ
. Sinelim ε→0,γ→0 µ(ε, γ) = 1
, we may assume thatµ(ε, γ) ≥ 1/2
. We distinguish three ases:a)
γ| sin ω i | ≥ | cos ϕ i |
and| cos ω i | ≥ ε.
Then| sin ϕ i | ≥ | cos ω i |
, andfrom (15) weget
| cos(ϕ i + ω i ) | ≥ | sin ω i || cos ω i | (1 − γ) ≥ | sin ω i | (1 − γ)ε.
(16)In this ase, estimate (14) isontinued as
− ∆r i ≥ r ¯
2 (1 − d i ) cos 2 (ϕ i + ω i ) ≥ r ¯
2 (1 − γ) 2 ε 2 (1 − d i ) sin 2 ω i .
(17)b)
γ| sin ω i | ≥ | cos ϕ i |
and| cos ω i | < ε
. Then| sin ϕ i | ≥ q
1 − γ 2 sin 2 ω i ≥ p
1 − γ 2 ,
(18)and
| cos(ϕ i + ω i ) | ≥ ( p
1 − γ 2 − εγ) | sin ω i | = µ(ε, γ ) | sin ω i | ≥ 1
2 | sin ω i | .
Then
− ∆r i ≥ r ¯
2 (1 − d i ) cos 2 (ϕ i + ω i ) ≥ r ¯
8 (1 − d i ) sin 2 ω i .
(19))
γ| sin ω i | < | cos ϕ i |
. In this ase we an estimate− ∆r i− 1 (instead
of
− ∆r i )
from below by| sin ω i |
. In fat, usingalso the inequality| cos ϕ i | = | y i |
p x 2 i + y 2 i = d i−1 | y ˜ i−1 | p x ˜ 2 i− 1 + d 2 i− 1 y ˜ i− 2 1
≤ | y ˜ i − 1 |
p x ˜ 2 i− 1 + ˜ y i− 2 1 = | cos(ϕ i−1 + ω i−1 ) | ,
(20)
from (14) weobtain
− ∆r i−1 ≥ ¯ r
2 (1 − d i−1 ) cos 2 (ϕ i−1 + ω i−1 ) ≥ r ¯
2 (1 − d i−1 ) cos 2 ϕ i
≥ ¯ r
2 γ 2 (1 − d i− 1 ) sin 2 ω i ≥ r ¯
2 γ 2 min { 1 − d i− 1 , 1 − d i } sin 2 ω i .
(21)
Setting
c := r ¯
2 min { (1 − γ) 2 ε 2 ; 1
4 ; γ 2 } > 0,
for every
i
we havec min { 1 − d i − 1 ; 1 − d i } sin 2 ω i ≤ − ∆r i − 1 − ∆r i = r i − 1 − r i+1 .
Summarizing these inequalitieswe obtain
c
∞
X
i=1
min { 1 − d i− 1 ; 1 − d i } sin 2 ω i ≤ r 0 − r < ¯ ∞ ,
whih ontradits assumption (13).
3 The half-linear equation
In this setion we onsider the half-linearseond orderdierentialequation
x ′′ | x ′ | n− 1 + q(t) | x | n− 1 x = 0, n ∈ R + ,
(22)whihwas introdued by Bihari[5℄ and Elbert [9℄. They alled it half-linear
beauseitssolutionsetishomogeneous, butitisnotadditive. Thisequation
is a generalizationof the seond order lineardierentialequation
x ′′ + q(t)x = 0
(23)500℄, we alla non-trivialsolution
x 0 (t)
of (22) small ift lim →∞ x 0 (t) = 0. (24)
H. Milloux [18℄ proved, that if
q
is dierentiable, monotonously inreasing andtends toinnityast → ∞
,thenthe linearequation(23) hasatleastonesmallsolution. Healsoonstrutedanequationwithsuhaoeient
q
hav-ingnotsmallsolutions,too. The famousArmellini-Tonelli-SansoneTheorem
(see, e.g., [17℄) gave a suient ondition guaranteeing that all solutions of
(23) were small. Many papers examined andsharpened theabove theorems,
even for nonlinear dierential equations or dierene equations (see, e.g.,
[15,17℄ and the referenes therein).
F. V. Atkinson and Elbert [4℄ extended the theorem of H. Milloux to
the half-lineardierentialequation(22). Anextensionof theA-T-S theorem
to (22) was given by Bihari with the following onept. A nondereasing
funtion
f : [0, ∞ ) → (0, ∞ )
withlim t→∞ f(t) = ∞
is alled to grow in-termittently if for every
ε > 0
there is a sequene{ (a i , b i ) } ∞ i=0 of disjoint
intervals suh that
a i → ∞
asi → ∞
,andlim sup
i →∞
i
X
k=1
b k − a k
b i ≤ ε,
∞
X
i=1
(f (a i+1 ) − f (b i )) < ∞
are satised. If suh a sequene does not exist, then
f
is alled to growregularly.
Theorem B (Bihari[6℄). If
q
is ontinuously dierentiable and it grows to innity regularly ast → ∞
, then all non-trivial solutions of equation (22) are small.Thesimplestaseoftheintermittentgrowthiswhen
q
isamonotonously inreasing step funtion. In this setion we will examine this ase, i.e., theequation
x ′′ | x ′ | n− 1 + q k | x | n− 1 x = 0 (t k ≤ t < t k+1 , k = 0, 1, . . .),
(25)where
t 0 = 0, lim
k→∞ t k = ∞ , 0 < q 0 ≤ q 1 ≤ . . . ≤ q k ≤ q k+1 ≤ . . . , lim
k→∞ q k = ∞ .
equation (25)has a smallsolution. Elbert [11, 12℄proved anA-T-S theorem
for the linear (
n = 1
) ase of equation (25) as a diret appliation of histheorem onthe asymptotistability of the trivialsolution of (1).
Theorem C (Elbert [11℄). Let
n = 1
. If∞
X
k=0
min
1 − q k
q k+1
, 1 − q k+1
q k+2
sin 2 ( √ q k+1 (t k+2 − t k+1 )) = ∞ ,
(26)then allnon-trivial solutions of equation (25) are small.
Our main goal is to extend Theorem C to the ase
n > 1
of half-linear equation(25). To thisend, weneed the so-alledgeneralizedsineand osinefuntions introdued by Elbert [9℄. Consider the solution
S = S n (Φ)
of theinitialvalue problem
( S ′′ | S ′ | n− 1 + S | S | n− 1 = 0
S(0) = 0, S ′ (0) = 1.
(27)Multiplying the dierential equation by
S ′ and integrating it over [0, Φ]
we
obtain the relation
| S ′ | n+1 + | S | n+1 = 1 ( −∞ < Φ < ∞ ),
(28)whihan be onsideredas ageneralizationof the lassialidentity
cos 2 ϕ + sin 2 ϕ = 1
(the asen = 1
).S
andS ′ are periodi funtionswith period 2ˆ π
,
where
π ˆ
is dened asˆ
π = 2 n+1 π sin n+1 π ,
whih gives bak
π
in the ordinary asen = 1
(see [9℄). Furthermore,S
isodd and
S ′ is even. The generalized tangent funtion an be introdued as well:
T (Φ) = S(Φ) S ′ (Φ) .
Now wean state our main theorem.
Theorem 2. Let
n > 1
. If∞
X
k=0
min
1 − q k
q k+1
, 1 − q k+1
q k+2
S q
1 n+1
k+1 (t k+2 − t k+1 )
!
n+1
= ∞ ,
(29)then allnon-trivial solutions of equation (25) are small.
Proof. First, using the notation
q(t) := q k (t k ≤ t < t k+1 , k = 0, 1, 2 . . .)
weintroduea new time variable
τ = ϕ(t) = Z t
0
q(s) n+1 1 ds, τ k := ϕ(t k ).
(30)Let
x(t) = x(ϕ − 1 (τ )) =: y(τ )
, whereϕ − 1 is the inverse funtionof ϕ
. Then
x ′ (t) = ˙ y(τ)q n+1 1 (t), x ′′ (t) = ¨ y(τ )q n+1 2 (t) (t 6 = t k , k = 0, 1, 2, . . .),
where
( · ) · = d( · )/dτ
. Thus, equation (25) is transformed intothe form¨
y(τ) | y(τ ˙ ) | n − 1 + | y(τ) | n − 1 y(τ ) = 0, (τ 6 = τ k k = 0, 1, . . .).
(31)Sineany solution
x
ofequation(25) hastobeontinuouslydierentiableon(0, ∞ )
,x ′ (t k+1 − 0) = x ′ (t k+1 + 0) = x ′ (t k+1 )
must hold foreveryk ∈ N
, i.e.,˙
y(τ k+1 ) = ˙ y(τ k+1 + 0) = q k
q k+1 n+1 1
˙
y(τ k+1 − 0),
where
f (t − 0)
andf(t + 0)
denotes the left-hand side and the right-handside limitofafuntion
f
att
, respetively. Weobtainthat(25) isequivalent to the followingdierentialequation with impulses:
¨
y(τ) | y(τ ˙ ) | n− 1 + | y(τ ) | n− 1 y(τ ) = 0, τ 6 = τ k
˙
y(τ k+1 ) = q k
q k+1
n+1 1
˙
y(τ k+1 − 0), k = 0, 1, 2, . . .
(32)
Let us introdue the generalized polar oordinates
y ˙ = ρS ′ (Φ)
,y = ρS(Φ)
,where
ρ = ( | y ˙ | n+1 + | y | n+1 ) n+1 1 , T (Φ) = y
˙
y , −∞ < Φ < ∞ .
Thisistheso-alledgeneralizedPrüfertransformation. Withtheaidofthese
variableswe an rewrite equation(31) into
˙Φ = 1, ρ ˙ = 0, (τ k ≤ τ < τ k+1 , k = 0, 1, . . .).
(33)So the dynamis of system (32) on the Minkowski plane [19℄
( ˙ y, y)
is thefollowing. It turns any point
( ˙ y 0 , y 0 )
around the origin on the Minkowskiirle with radius
ρ 0 := ( | y ˙ 0 | n+1 + | y 0 | n+1 ) n+1 1 on[τ 0 , τ 1 )
, and at τ 1 the point
( ˙ y(τ 1 − 0), y (τ 1 − 0))
jumps tothe point( ˙ y(τ 1 ), y(τ 1 )) :=
q 0
q 1
n+1 1
˙
y(τ 1 − 0), y(τ 1 − 0)
! .
This proess is repeated onseutively for
[τ 1 , τ 2 )
,[τ 2 , τ 3 ), . . .
. Deneρ k := | y(τ ˙ k ) | n+1 + | y(τ k ) | n+1 n+1 1
, Φ k := Φ(τ k ), Ω k := τ k+1 − τ k ,
∆ρ k := ρ k+1 − ρ k , κ k := Φ k+1 − (Φ k + Ω k ), k = 0, 1, . . .
Obviously,
Φ k+1 = Φ 0 +
k
X
i=0
(Ω i + κ i ), ρ k+1 = ρ 0 +
k
X
i=0
∆ρ i , k = 0, 1 . . .
Sine
∆ρ i ≤ 0
,thesequene{ ρ k } ∞ k=0 ismonotonouslydereasing,thereforeit
has alimitρ ¯ := lim k→∞ ρ k. Ifthe statementof thetheorem is nottrue, then
there exists a solution
(ρ, Φ)
suh thatρ > ¯ 0
. Let us onsider this solutionand estimate
− ∆ρ i:
− ∆ρ i = ρ i − ρ i+1
= ( | y(τ ˙ i ) | n+1 + | y(τ i ) | n+1 ) n+1 1 − ( | y(τ ˙ i+1 ) | n+1 + | y(τ i+1 ) | n+1 ) n+1 1
= ( | y(τ ˙ i+1 − 0) | n+1 + | y(τ i+1 − 0) | n+1 ) n+1 1
− ( | y(τ ˙ i+1 ) | n+1 + | y(τ i+1 ) | n+1 ) n+1 1
= ( | y(τ ˙ i+1 − 0) | n+1 + | y(τ i+1 − 0) | n+1 ) n+1 1
− q i
q i+1 | y(τ ˙ i+1 − 0) | n+1 + | y(τ i+1 − 0) | n+1 n+1 1
= 1
n + 1 ρ n+1 i+1 + η i ρ n+1 i − ρ n+1 i+1 − n+1 n
×
1 − q i
q i+1
| y(τ ˙ i+1 − 0) | n+1
≥ 1
n + 1 (¯ ρ) n+1 − n+1 n
1 − q i
q i+1
ρ n+1 i | S ′ (Φ i + Ω i ) | n+1
≥ ρ ¯ n + 1
1 − q i
q i+1
| S ′ (Φ i + Ω i ) | n+1
(34)
with some
η i ∈ (0, 1)
for alli ∈ N 0. Now we need to estimate | S ′ (φ i + Ω i ) |
from belowby either
| S(Ω i ) |
or| S(Ω i+1 ) |
, similarlytothe proof of Theorem1, where weused the addititonal formulae for the osine funtion. However,
to our best knowledge,the problemof nding exat additionformulaefor
S
and
S ′ is not ompletely solved, although there are some papers about this
topi (see, e.g., [1℄, [2℄). Therefore, to omplete the proof we need a new
method dierent from one we used in the proof of Theorem 1 after formula
(14).
Funtions
| S ′ (Φ + Ω) |
and| S(Ω) |
areπ ˆ
-periodi with respet to bothvariables
Φ, Ω
,henewemayrestritourselvestothequadrant[ − π/2, ˆ ˆ π/2] × [ − π/2, ˆ π/2] ˆ
on the(Φ, Ω)
plane. Thanks to the symmetry properties ofS
and
S ′,it isenough tomakethe estimate onQ := [0, π/2] ˆ × [0, ˆ π/2]
.
At rst, letus handlethe set
Q ε := { (Φ, Ω) ∈ Q : | S ′ (Φ) | < ε } ,
where
ε > 0
is small enough. The omplementer set ofQ ε with respet to
Q
will be treated in another way. The same way will be used also for theomplementer set of
Q γ := { (Φ, Ω) ∈ Q : | S ′ (Φ) | ≤ γ | S(Ω) |} (0 < γ < 1),
so nowweonsider the set
Q γ ε := Q ε ∩ Q γ (see the gure).
A part of the boundary of this set is a piee of the urve dened by the
equation
Γ : | S ′ (Φ) | = γ | S(Ω) | .
We show that the tangent to
Γ
at(ˆ π/2, 0)
is the lineΦ = ˆ π/2
, i.e.,Φ → lim π 2 − 0 f ′ (Φ) = −∞ ; f(Φ) := S −1 1
γ S ′ (Φ)
,
(35)provided
n > 1
. The statementof the theorem for the linearasen = 1
wasproved in Theorem 1, so proving (35) we an restrit ourselves to the ase
n > 1
.It iseasy tosee that
(S − 1 ) ′ (W ) = 1
(1 − W n+1 ) n+1 1 (0 ≤ W ≤ 1).
S ′′ (Φ) = −| S ′ (Φ) | −n+1 | S(Φ) | n−1 S(Φ).
(36)Therefore,
d
d
Φ f(Φ) = f ′ (Φ) = − γ 1 (S ′ (Φ)) −n+1 S n (Φ) 1 − γ n+1 1 (S ′ (Φ)) n+1 n+1 1 ,
onsequently, (35) holds, independently of
γ.
(35) implies the existene of aδ > 0
suh thatf ′ (Φ) < − 2
(S ′ ) −1 (ε) < π ˆ
2 − δ < Φ < π ˆ 2
,
whene we get
f (Φ) ≥ − 2
Φ − π ˆ 2
,
whihmeansthat
Γ
isloatedonthe right-handside ofthe lineΩ = − 2(Φ − ˆ
π/2)
near the point(ˆ π/2, 0)
(see the gure). To estimate| S ′ (Φ i + Ω i ) |
frombelowby
| S(Ω i ) |
in(34)wehavetoestimatethequotient| S ′ (Φ + Ω) | / | S(Ω) |
frombelow. In
Q γ ε wederease this quotientexhangingpoint(Φ, Ω)
for the
horizontally orresponding point
(ˆ π/2 − Ω/2, Ω)
of the lineΦ = ˆ π/2 − Ω/2
(see the gureagain). Therefore, by the L'Hospital Rule and (36) we get
lim Φ → ˆ π
2 − 0,Ω → 0+0,(Φ,Ω) ∈ Q γ ε
| S ′ (Φ + Ω) |
| S(Ω) | ≥ lim
Ω → 0+0
− S ′ ˆ
π
2 − 1 2 Ω + Ω S(Ω)
= lim
Ω → 0+0
− S ′ ˆ
π
2 + 1 2 Ω
S(Ω) = lim
Ω → 0+0
− S ′′
ˆ π
2 + 1 2 Ω
1 2
S ′ (Ω)
= lim
Ω → 0+0
S ′
ˆ π
2 + Ω 2
− n+1 S
ˆ π
2 + Ω 2
n − 1
S ˆ
π 2 + Ω 2
2S ′ (Ω) = ∞ .
This means that there exists a
κ > 0
suh that| S ′ (Φ + Ω) | ≥ κ | S(Ω) | ((Φ, Ω) ∈ Q γ ε ).
(37)Nowwe are readyto ompleteestimate(34). Wedistinguish threeases:
A)
(Φ i , Ω i ) ∈ Q γ ε .
Then by (34) and (37) we have− ∆ρ i ≥ ρ n + 1
1 − q i
q i+1
κ n+1 | S(Ω i ) | n+1 .
(38)In the remaining ases we estimate
− ∆ρ i− 1 .
By the analogue of (20) it isalwaystrue that
− ∆ρ i − 1 ≥ ρ n + 1
1 − q i − 1
q i
| S ′ (Φ i − 1 + Ω i − 1 ) | n+1
≥ ρ n + 1
1 − q i − 1
q i
| S ′ (Φ i ) | n+1 .
B)
(Φ i , Ω i ) ∈ Q ε \Q γ ε. Then | S ′ (Φ i ) | ≥ γ | S(Ω i ) |
, and
− ∆ρ i−1 ≥ γ n+1 ρ n + 1
1 − q i − 1
q i
| S(Ω i ) | n+1 .
(39)C)
(Φ i , Ω i ) ∈ Q \ Q ε. Then | S ′ (Φ i ) | ≥ ε | S(Ω i ) |
and
− ∆ρ i− 1 ≥ ε n+1 ρ n + 1
1 − q i−1
q i
| S(Ω i ) | n+1 .
(40)Setting
C := ρ
n + 1 min { κ n+1 ; γ n+1 ; ε n+1 } > 0,
and taking intoaount (38), (39), (40), for every
i
we haveC min
1 − q i − 1
q i
; 1 − q i
q i+1
| S(Ω i ) | n+1 ≤ ∆ρ i−1 − ∆ρ i = ρ i−1 − ρ i+1 .
Summarizing these inequalitieswe obtain
C
∞
X
n=1
min
1 − q i − 1
q i
; 1 − q i
q i+1
| S(Ω i ) | n+1 ≤ ρ 0 − ρ < ∞ ,
whih ontradits the assumption of the theorem.
Theorem2extends Elbert'sTheorem Ctohalf-linearequationsprovided
n > 1
. It would bee interesting to nd anextension tothe asen < 1
,too.Referenes
[1℄ I. Adamaszek, On generalized sine and osine funtions,Demonstratio
Math. 28(1995),263270.
[2℄ I. Adamaszek-Fidytek, On generalized sine and osine funtions, II.
Demonstratio Math.37(2004), 571578.
[3℄ R.P.Agarwal,Dierene EquationsandInequalities,Monographsand
TextbooksinPureandAppliedMathematis,vol.115,MarelDekker,
In., New York, 1992.
linear dierential equations, Pro. 6'th Colloq. Qual. Theory Dier.
Equ., No.8, Eletron. J. Qual. Theory Dier. Equ., Szeged, 2000.
[5℄ I.Bihari,Ausdehnung derSturmshenOsillationsundVergleihssätze
aufdieLösungen gewissernihtlinearenDierentialgleihungenzweiter
Ordnung, Publ. Math. Inst. Hungar. Aad. Si.,2(1957), 154165.
[6℄ I. Bihari, Asymptoti result onerning equation
x ′′ | x ′ | n − 1 + a(t)x ∗ n.
ExtensionofatheorembyArmellini-Tonelli-Sansone,StudiaSi.Math.
Hungar. 19(1984), 151157.
[7℄ O. Do²lý, Half-linear dierential equations, Handbook of Dierential
Equations, 161357, Elsevier/North-Holland,Amsterdam, 2004.
[8℄ O.Do²lý,P.ehák, Half-linearDierentialEquations, Elsevier/North-
Holland, Amsterdam, 2005.
[9℄ Á. Elbert, A half-linear seondorder dierentialequation, Qualitative
Theory of Dierential Equations, Vol. I, II (Szeged, 1979), 153180,
Colloq.Math. So.János Bolyai,30, North-Holland,Amsterdam-New
York,1981.
[10℄ Á. Elbert, Stability of some differene equations, Advanes in Die-
rene Equations: Proeedings of the Seond International Conferene
on Dierene Equations and Appliations, Veszprém, Hungary, 7-11
August 1995,Gordon and Breah Siene Publishers, 1997, 155178.
[11℄ Á.Elbert,On asymptotistabilityof someSturm-Liouvilledierential
equations,General Seminar in Mathematis, University ofPatras,22-
23(1996/97), 5766.
[12℄ Á. Elbert,On dampingoflinearosillators,Studia Si.Math. Hungar.
38(2001), 191208.
[13℄ P. Hartman, Ordinary Dierential Equations, Birkhäuser, Boston,
1982.
[14℄ L. Hatvani, On stability properties of solutionsof seond order dier-
entialequations, Pro.6'th Colloq.Qual. Theory Dier. Equ.,No. 11,
Eletron. J. Qual. Theory Dier. Equ., Szeged,2000.
near osillatorwithaninreasing elastiityoeient,Georgian Math.
J. 14(2007),no. 2,269278.
[16℄ N.Jaobson, Letures inAbstrat Algebra,II. Linear Algebra,Springer
Verlag (New York- Heidelberg - Berlin),1953.
[17℄ J. W. Maki, Regular growth and zero tending solutions, Ordinary
Dierential Equations and Operators (Dundee, 1982), Leture Notes
in Math., vol. 1032, 358399.
[18℄ H.Milloux,Surl'equationdierentielle
x ′′ +A(t)x = 0
,PraeMat.-Fiz.,41(1934), 3954.
[19℄ H.Rund,TheDierentialGeometryofFinslerSpaes,Springer,Berlin,
1959.
(Reeived April 2,2011)