Advertisement

Advances in Difference Equations

, 2013:220 | Cite as

On power and non-power asymptotic behavior of positive solutions to Emden-Fowler type higher-order equations

  • Irina AstashovaEmail author
Open Access
Research
Part of the following topical collections:
  1. Progress in Functional Differential and Difference Equations

Abstract

For the equation

y ( n ) = y k , k > 1 , n = 12 , 13 , 14 , Open image in new window

the existence of positive solutions with non-power asymptotic behavior is proved, namely

y = ( x x ) α h ( log ( x x ) ) , α = n k 1 , x < x , Open image in new window

where x Open image in new window is an arbitrary point, h is a positive periodic non-constant function on R.

To prove this result, the Hopf bifurcation theorem is used.

Keywords

asymptotic behavior Emden-Fowler higher-order equations 

Introduction

For the equation
y ( n ) = p ( x , y , y , , y ( n 1 ) ) | y | k sgn y , n 2 , k > 1 , Open image in new window
(1)
Kiguradze posed the problem on the asymptotic behavior of its positive solutions such that
lim x x 0 y ( x ) = . Open image in new window
(2)
He found an asymptotic formula for these solutions to (1) with n = 2 Open image in new window (see [1]) and supposed all such solutions to have power asymptotic behavior for other n, too. The problem was solved for n = 3 Open image in new window and n = 4 Open image in new window [2]. For these n, it was proved that all such solutions behave as
y ( x ) = C ( x x ) α ( 1 + o ( 1 ) ) , x x 0 , Open image in new window
(3)
with
α = n k 1 , C = ( α ( α + 1 ) ( α + n 1 ) p 0 ) 1 k 1 , Open image in new window
(4)

p 0 = const > 0 Open image in new window - is a limit of p ( x , y 0 , , y n 1 ) Open image in new window as x x 0 Open image in new window, y 0 , , y n 1 Open image in new window.

So, the hypothesis of Kiguradze was confirmed in this case.

The existence of solutions satisfying (3) was proved for arbitrary n 2 Open image in new window. For 2 n 11 Open image in new window, an ( n 1 ) Open image in new window-parametric family of such solutions to equation (1) was proved to exist (see [2], [3], Ch.I(5.1)).

For the equation
y ( n ) = y k , k > 1 , Open image in new window
(5)
a negative answer to the conjecture of Kiguradze for large n was obtained. It was proved [4] that for any N and K > 1 Open image in new window, there exist an integer n > N Open image in new window and k R Open image in new window, 1 < k < K Open image in new window, such that equation (5) has a solution
y = ( x x ) α h ( log ( x x ) ) , Open image in new window
(6)

where α is defined by (4), h is a positive periodic non-constant function on R.

Still, it was not clear how large n should be for the existence of that type of solutions.

Preliminary results

Suppose the following conditions hold:
  1. (A)
    The continuous positive function p ( x , y 0 , , y n 1 ) Open image in new window has a limit p 0 = const > 0 Open image in new window as x x 0 Open image in new window, y 0 , , y n 1 Open image in new window, and for some γ > 0 Open image in new window, it holds
    p ( x , y 0 , , y n 1 ) p 0 = O ( ( x x ) γ + j = 0 n 1 y j γ ) . Open image in new window
    (7)
     
  2. (B)
    For some K 1 > 0 Open image in new window and μ > 0 Open image in new window in a neighborhood of x Open image in new window for sufficiently large y 0 , , y n 1 Open image in new window, z 0 , , z n 1 Open image in new window, it holds
    | p ( x , y 0 , , y n 1 ) p ( x , z 0 , , z n 1 ) | K 1 max j | y j μ z j μ | . Open image in new window
    (8)
     
Then equation (1) can be transformed (see [2] or [3], Ch.I(5.1)) by using the substitution
x x = e t , y = ( C + v ) e α t , Open image in new window
(9)
where C and α are defined by (4). The derivatives y ( j ) Open image in new window, j = 0 , 1 , , n 1 Open image in new window, become
e ( α + j ) t L j ( v , v , , v ( j ) ) , Open image in new window
where v ( j ) = d j v d t j Open image in new window and L j Open image in new window is a linear function with
L j ( 0 , 0 , , 0 ) = C α ( α + 1 ) ( α + j 1 ) 0 Open image in new window

and the coefficient of v ( j ) Open image in new window equal to 1.

Thus (1) is transformed into
e ( α + n ) t L n ( v , v , , v ( n ) ) = ( C + v ) k e α k t p ˜ ( t , v , v , , v ( n 1 ) ) , Open image in new window
(10)

where the function p ˜ ( t , v 0 , , v n 1 ) Open image in new window is obtained from p ( x , y 0 , , y n 1 ) Open image in new window with x , y 0 , , y n 1 Open image in new window properly expressed in terms of t , v 0 , , v n 1 Open image in new window. This function tends to p 0 Open image in new window as t Open image in new window, v 0 , , v ( n 1 ) 0 Open image in new window.

Due to condition (8) for the function p ( x , y 0 , , y n 1 ) Open image in new window, we obtain the following inequalities for sufficiently large t and sufficiently small v 0 , , v n 1 Open image in new window, w 0 , , w n 1 Open image in new window:
| p ˜ ( t , v 0 , , v n 1 ) p ˜ ( t , w 0 , , w n 1 ) | K 1 max j e μ ( α + j ) t | L j μ ( v 0 , , v n 1 ) L j μ ( w 0 , , w n 1 ) | . Open image in new window
Since L j ( 0 , 0 , , 0 ) 0 Open image in new window, the function L j μ Open image in new window is a C Open image in new window one in a neighborhood of 0 and
| p ˜ ( t , v 0 , , v n 1 ) p ˜ ( t , w 0 , , w n 1 ) | K 2 e μ α t max j | v j w j | Open image in new window

for some K 2 > 0 Open image in new window.

Solving (10) for v ( n ) Open image in new window and using formulae (4), we obtain the equation
v ( n ) = ( C + v ) k p ˜ ( t , v , v , , v ( n 1 ) ) p 0 C k j = 0 n 1 a j v ( j ) , Open image in new window
(11)
where a j Open image in new window are the coefficients of the linear function L n Open image in new window. Equation (11) can be written as
v ( n ) = k C k 1 p 0 v j = 0 n 1 a j v ( j ) + f ( v ) + g ( t , v , v , , v ( n 1 ) ) , Open image in new window
(12)
where
f ( v ) = p o ( ( C + v ) k C k k C k 1 v ) = O ( v 2 ) as  v 0 , f ( v ) = O ( v ) as  v 0 , g ( t , v 0 , , v n 1 ) = ( C + v 0 ) k ( p ˜ ( t , v 0 , , v n 1 ) p 0 ) = O ( e γ t + j = 0 n 1 e γ ( α + j ) t ) = O ( exp ( γ min ( α , 1 ) t ) ) as  t , v 0 0 , , v n 1 0 . Open image in new window
Besides, for sufficiently large t and sufficiently small v 0 , , v n 1 Open image in new window, w 0 , , w n 1 Open image in new window, it holds
| g ( t , v 0 , , v n 1 ) g ( t , w 0 , , w n 1 ) | | ( C + v 0 ) k ( C + w 0 ) k | | p ˜ ( t , v 0 , , v n 1 ) p 0 | + ( C + w 0 ) k | p ˜ ( t , v 0 , , v n 1 ) p ˜ ( t , w 0 , , w n 1 ) | K 3 max j | w j v j | e min ( γ , μ ) min ( α , 1 ) t . Open image in new window
Suppose that V is the vector with coordinates V j = v ( j ) Open image in new window, j = 0 , , n 1 Open image in new window. Then equation (12) can be written as
d V d t = A V + F ( V ) + G ( t , V ) , Open image in new window
(13)
where A is a constant n × n Open image in new window matrix
A = ( 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 a ˜ 0 a 1 a 2 a 3 a n 1 ) , Open image in new window
with
a ˜ 0 = a 0 k c k 1 p 0 = a 0 k α ( α + 1 ) ( α + n 1 ) = a 0 ( α + 1 ) ( α + n 1 ) ( α + n ) Open image in new window
and eigenvalues satisfying the equation
0 = det ( A λ E ) = ( 1 ) n + 1 ( a ˜ 0 a 1 λ a n 1 λ n 1 λ n ) = ( 1 ) n + 1 ( ( α + 1 ) ( α + 2 ) ( α + n ) ( λ + α ) ( λ + α + n 1 ) ) , Open image in new window
which is equivalent to
j = 0 n 1 ( λ + α + j ) = j = 0 n 1 ( 1 + α + j ) . Open image in new window
(14)
The mappings F : R n R n Open image in new window and G : R × R n R n Open image in new window satisfy the following estimates as t Open image in new window:
{ F ( V ) = O ( V 2 ) , F V ( V ) = O ( V ) , G ( t , V ) = O ( e 2 β t ) , G ( t , V ) G ( t , W ) K V W e 2 β t Open image in new window
(15)

with some constants β > 0 Open image in new window, K > 0 Open image in new window.

Lemma 1 [3]

Suppose that (15) holds and A is an arbitrary constant n × n Open image in new window matrix. Then there exists a solution V ( t ) Open image in new window to equation (13) tending to zero as t Open image in new window.

Lemma 2 [3]

Let the conditions of Lemma  1 hold. If equation (14) has m roots with negative real part, then there exists an m-parametric family of solutions V ( t ) Open image in new window to equation (13) tending to zero as t Open image in new window.

If equation (13) has a solution V ( t ) Open image in new window tending to 0 as t Open image in new window and V 0 ( t ) Open image in new window is its first coordinate, then the function
y ( x ) = ( V 0 ( log ( x x ) ) + C ) ( x x ) α Open image in new window

with C and α defined by (4) is a solution to (1) such that (2) and (3) hold.

Theorem 1 [2, 3]

Suppose that conditions (A) and (B) are satisfied. Then for such x Open image in new window there exists a solution to (1) with power asymptotic behavior (3).

Investigating the signs of the real parts of the roots of equation (14), by the Routh-Hurwitz criterion, we can prove the following theorem.

Theorem 2 [2, 3]

Suppose that 3 n 11 Open image in new window and conditions (A) and (B) are satisfied. Then there exists an ( n 1 ) Open image in new window-parametric family of solutions to equation (1) with power asymptotic behavior (3).

Theorem 3 [2, 3, 5]

Suppose that n = 3 Open image in new window or n = 4 Open image in new window in equation (1), the continuous positive function p ( x , y 0 , , y n 1 ) Open image in new window is Lipschitz continuous in y 0 , , y n 1 Open image in new window and has a limit p 0 > 0 Open image in new window as x x 0 Open image in new window, y 0 , , y n 1 Open image in new window. Then any positive solution to this equation with a vertical asymptote x = x Open image in new window has asymptotic behavior (3).

To prove the main results of this article, we use the Hopf bifurcation theorem [6].

Theorem (Hopf)

Consider the α-parameterized dynamical system x ˙ = L α x + Q α ( x ) Open image in new window in a neighborhood of 0 R n Open image in new window with linear operators L α Open image in new window and smooth enough functions Q α ( x ) = O ( | x | 2 ) Open image in new window as x 0 Open image in new window. Let λ α Open image in new window and λ ¯ α Open image in new window be simple complex conjugated eigenvalues of the operators  L α Open image in new window. Suppose that Re λ α ˜ = Re λ ¯ α ˜ = 0 Open image in new window for some α ˜ Open image in new window and the operator L α ˜ Open image in new window has no other eigenvalues with zero real part.

If Re d λ α d α ( α ˜ ) 0 Open image in new window, then there exist continuous mappings ε α ( ε ) R Open image in new window, ε T ( ε ) R Open image in new window, and ε b ( ε ) R n Open image in new window defined in a neighborhood of 0 and such that α ( 0 ) = α ˜ Open image in new window, T ( 0 ) = 2 π / Im λ α ˜ Open image in new window, b ( 0 ) = 0 Open image in new window, b ( ε ) 0 Open image in new window for ε 0 Open image in new window, and the solutions to the problems
x ˙ = L α ( ε ) x + Q α ( ε ) ( x ) , x ( 0 ) = b ( ε ) Open image in new window

are T ( ε ) Open image in new window-periodic and non-constant.

Main results

In this section, the result about the existence of solutions with non-power asymptotic behavior is proved for equation (5) with n = 12 , 13 , 14 Open image in new window.

Theorem 4 For n = 12 , 13 , 14 Open image in new window, there exists k > 1 Open image in new window such that equation (5) has a solution y ( x ) Open image in new window with
y ( j ) ( x ) = ( x x ) α j h j ( log ( x x ) ) , j = 0 , 1 , , n 1 , Open image in new window

where α is defined by (4) and h j Open image in new window are periodic positive non-constant functions on R.

Proof To apply the Hopf bifurcation theorem, we investigate equation (13) with G ( t , V ) 0 Open image in new window corresponding to the case of the constant function p and the roots of the algebraic equation (14). F is a vector function with all zero components F ( V ) = ( 0 , , 0 , F n 1 ( V ) ) Open image in new window, V = ( V 0 , , V n 1 ) Open image in new window, and
F n 1 ( V ) = ( ( C + V 0 ) k C k k C k 1 V 0 ) = O ( V 0 2 ) , V 0 0 , d d V F n 1 ( V ) = O ( | V 0 | ) , V 0 0 . Open image in new window

If equation (14) has a pair of pure imaginary roots, we have to check other conditions of this theorem and then apply it.

Proposition 1 For any integer n > 11 Open image in new window, there exist α > 0 Open image in new window and q > 0 Open image in new window such that
j = 0 n 1 ( q i + α + j ) = j = 0 n 1 ( 1 + α + j ) Open image in new window
(16)

with i 2 = 1 Open image in new window.

Remark 1 In the particular case n = 12 Open image in new window, this result was obtained by Vyun [7].

Proof Consider the positive functions ρ n ( α ) Open image in new window and σ n ( α ) Open image in new window defined for all α > 0 Open image in new window via the equations
j = 0 n 1 ( ρ n ( α ) 2 + ( α + j ) 2 ) = j = 0 n 1 ( 1 + α + j ) 2 Open image in new window
(17)
and
j = 0 n 1 arg ( σ n ( α ) i + α + j ) = 2 π Open image in new window
(18)

supposing arg z [ 0 , 2 π ) Open image in new window for all z C { 0 } Open image in new window.

First, we prove the functions to be well defined for all α > 0 Open image in new window.

The product j = 0 n 1 ( q 2 + ( α + j ) 2 ) Open image in new window is continuous and strictly increasing as a function of q > 0 Open image in new window.

It tends to j = 0 n 1 ( α + j ) 2 < j = 0 n 1 ( 1 + α + j ) 2 Open image in new window as q 0 Open image in new window and to +∞ as q + Open image in new window. Hence, for any α > 0 Open image in new window, there exists a unique q > 0 Open image in new window such that j = 0 n 1 ( q 2 + ( α + j ) 2 ) = j = 0 n 1 ( 1 + α + j ) 2 Open image in new window.

In the same way, for any α > 0 Open image in new window, the sum j = 0 n 1 arg ( q i + α + j ) Open image in new window is a continuous function of q > 0 Open image in new window strictly increasing from 0 to π n 2 > 2 π Open image in new window. So, there exists a unique q > 0 Open image in new window such that the sum is equal to 2π.

Since both the product and the sum considered are C 1 Open image in new window-functions with positive partial derivative in q > 0 Open image in new window, the implicit function theorem provides both ρ n ( α ) Open image in new window and σ n ( α ) Open image in new window to be C 1 Open image in new window-functions, too.

Now it is sufficient to prove the existence of α > 0 Open image in new window such that ρ n ( α ) Open image in new window and σ n ( α ) Open image in new window are equal to the same value q, which makes the two sides of (16) be equal.

Compare the functions ρ n ( α ) Open image in new window and σ n ( α ) Open image in new window near the boundaries of their common domain.

Equation (17) defining the function ρ n ( α ) Open image in new window may be written as
j = 0 n 1 ( 1 + 2 j α + j 2 α 2 + ( ρ n ( α ) α ) 2 ) = j = 0 n 1 ( 1 + j + 1 α ) 2 . Open image in new window

This shows that ρ n ( α ) α 0 Open image in new window as α + Open image in new window.

Equation (18) defining the function σ n ( α ) Open image in new window may be written as
j = 0 n 1 arctan σ n ( α ) α 1 + j α = 2 π . Open image in new window

This shows that σ n ( α ) α tan 2 π n > 0 Open image in new window as α + Open image in new window. Thus, ρ n ( α ) < σ n ( α ) Open image in new window for sufficiently large α.

Now, to prove Proposition 1, it is sufficient to show that ρ n ( α ) > σ n ( α ) Open image in new window for sufficiently small α. To compare the functions ρ n ( α ) Open image in new window and σ n ( α ) Open image in new window for small α > 0 Open image in new window, we need some lemmas.

Lemma 3 For all α > 0 Open image in new window, it holds ρ n ( α ) 2 < 2 ( α + n ) 1 Open image in new window.

Proof Suppose that ρ n ( α ) 2 2 ( α + n ) 1 Open image in new window for some α > 0 Open image in new window. Then
j = 0 n 1 ( ρ n ( α ) 2 + ( α + j ) 2 ) j = 0 n 1 ( 2 ( α + n ) 1 + ( α + j ) 2 ) > j = 0 n 1 ( 2 ( α + j + 1 ) 1 + ( α + j ) 2 ) = j = 0 n 1 ( 1 + ( α + j ) ) 2 . Open image in new window

This contradiction with the definition of ρ n ( α ) Open image in new window completes the proof of Lemma 3. □

Lemma 4 For all α > 0 Open image in new window, it holds ρ n + 1 ( α ) > ρ n ( α ) Open image in new window.

Proof According to the definition of ρ n ( α ) Open image in new window by (17) and Lemma 3, we have
j = 0 n ( ρ n ( α ) 2 + ( α + j ) 2 ) = j = 0 n 1 ( 1 + α + j ) 2 ( ρ n ( α ) 2 + ( α + n ) 2 ) < j = 0 n 1 ( 1 + α + j ) 2 ( 2 ( α + n ) 1 + ( α + n ) 2 ) < j = 0 n ( 1 + α + j ) 2 . Open image in new window

In order to make the first and the last products be equal, we have to replace ρ n ( α ) Open image in new window in the first one by a greater value. This means that ρ n + 1 ( α ) > ρ n ( α ) Open image in new window and Lemma 4 is proved. □

Lemma 5 For all α > 0 Open image in new window, it holds σ n + 1 ( α ) < σ n ( α ) Open image in new window.

Proof According to the definition of σ n ( α ) Open image in new window by (18), we have
j = 0 n arg ( σ n ( α ) i + α + j ) = 2 π + arg ( σ n ( α ) i + α + n ) > 2 π . Open image in new window

In order to make the sum equal 2π, we have to replace σ n ( α ) Open image in new window by a smaller value. So, σ n + 1 ( α ) < σ n ( α ) Open image in new window and Lemma 5 is proved. □

Due to Lemmas 3, 4, 5 proved, it is sufficient now for the proof of Proposition 1 to show that ρ 12 ( α ) > σ 12 ( α ) Open image in new window for sufficiently small α > 0 Open image in new window.

Lemma 6 It holds ρ 12 ( α ) > 2 Open image in new window for all sufficiently small α > 0 Open image in new window.

Proof Straightforward exact calculations show that
lim α 0 j = 0 11 ( 2 2 + ( α + j ) 2 ) = j = 0 11 ( 4 + j 2 ) = 192 , 175 , 659 , 520 , 000 , 000 < 2 10 17 Open image in new window
and
lim α 0 j = 0 11 ( 1 + α + j ) 2 = ( 12 ! ) 2 = 229 , 442 , 532 , 802 , 560 , 000 > 2 10 17 . Open image in new window
So, for sufficiently small α > 0 Open image in new window, we have
j = 0 11 ( 2 2 + ( α + j ) 2 ) < 2 10 17 < j = 0 11 ( 1 + α + j ) 2 . Open image in new window

Hence, for these α, in order to avoid contradiction with the definition of ρ 12 ( α ) Open image in new window, the inequality ρ 12 ( α ) 2 > 2 2 Open image in new window is necessary. Lemma 6 is proved. □

Lemma 7 It holds σ 12 ( α ) < 2 Open image in new window for sufficiently small α > 0 Open image in new window.

Proof Consider the limit
lim α 0 j = 0 11 arg ( 2 i + α + j ) = arg 2 i + arctan 2 + arctan 1 + arctan 2 3 + arctan 1 2 + j = 5 11 arctan 2 j = 5 π 4 + arctan 2 3 + j = 5 11 arctan 2 j = 5 π 4 + arctan 2 3 + 2 5 1 2 3 2 5 + arctan 2 6 + 2 7 1 2 6 2 7 + arctan 2 8 + 2 9 1 2 8 2 9 + arctan 2 10 + 2 11 1 2 10 2 11 = 5 π 4 + arctan 16 11 + arctan 13 19 + arctan 1 2 + arctan 21 53 = 5 π 4 + arctan 16 11 + 13 19 1 16 11 13 19 + arctan 1 2 + 21 53 1 1 2 21 53 = 5 π 4 + arctan 447 + arctan 19 17 . Open image in new window
Note that
tan ( arctan 447 + arctan 19 17 ) = 447 + 19 17 1 447 19 17 = 3 , 809 4 , 238 . Open image in new window

Hence, arctan 447 + arctan 19 17 > 3 π 4 Open image in new window and j = 0 11 arg ( 2 i + α + j ) > 2 π Open image in new window for sufficiently small α > 0 Open image in new window. Thus, for these α, we have σ 12 ( α ) < 2 Open image in new window, which completes the proof of Lemma 7. □

Now Proposition 1 is also proved.  □

Proposition 2 For any α > 0 Open image in new window and any integer n > 1 Open image in new window, all roots λ C Open image in new window to equation (14) are simple.

Proof Since we consider a polynomial equation of degree n, it is sufficient to prove the existence of n different roots to (14). We will show that for any integer m such that n < m n Open image in new window, there exists μ m C Open image in new window satisfying
j = 0 n 1 | μ m + j | = j = 0 n 1 ( 1 + α + j ) Open image in new window
(19)
and
j = 0 n 1 arg ( μ m + j ) = m π Open image in new window
(20)

with argz denoting the principal value of the argument lying in the open-closed interval ( π , π ] Open image in new window. Surely, all these 2n complex numbers μ m Open image in new window are different. Those with even m generate, via the relation λ m + α = μ m Open image in new window, just n different roots λ m Open image in new window to (14).

We begin to accomplish this plan by noting that the set of μ satisfying equation (20) with m = 0 Open image in new window is the real semi-axis ( 0 , + ) Open image in new window containing a single point satisfying (19), namely μ 0 = 1 + α Open image in new window.

Similarly, the set of μ satisfying equation (20) with m = n Open image in new window is the real unbounded interval ( , 1 n ) Open image in new window containing a single point satisfying (19), namely μ n = α n Open image in new window.

Now consider the cases 0 < m < n Open image in new window and the upper complex half-plane. For any ω > 0 Open image in new window, the smooth function
ϕ ω ( r ) = j = 0 n 1 arg ( r + ω i + j ) = j = 0 n 1 arccot r + j ω Open image in new window

monotonically decreases from to 0 as r increases from −∞ to +∞. So, for any ω > 0 Open image in new window and b ( 0 , n π ) Open image in new window, there exists a unique value r such that ϕ ω ( r ) = b Open image in new window. Due to the inequality d ϕ ω d r ( r ) < 0 Open image in new window, the implicit function theorem provides the existence of the smooth functions r m ( ω ) Open image in new window satisfying ϕ ω ( r m ( ω ) ) = m π Open image in new window.

Note that if r m Open image in new window, then r + j < 0 Open image in new window for all j < m Open image in new window and r + m 0 Open image in new window. Hence,
lim _ ω + 0 j = 0 n 1 arccot r + j ω lim _ ω + 0 j = 0 m 1 arccot r + j ω + lim _ ω + 0 arccot 0 ω = m π + π 2 > m π Open image in new window

and such r cannot be the value of r m ( ω ) Open image in new window for sufficiently small ω > 0 Open image in new window.

Similarly, if r 1 m Open image in new window, then r + j > 0 Open image in new window for all j > m 1 Open image in new window and r + m 1 0 Open image in new window. Hence,
lim ω + 0 j = 0 n 1 arccot r + j ω lim ω + 0 j = 0 m 2 arccot r + j ω + π 2 + lim ω + 0 j = m n 1 arccot r + j ω ( m 1 ) π + π 2 + 0 < m π Open image in new window

and such r cannot be the value of r m ( ω ) Open image in new window for sufficiently small ω > 0 Open image in new window.

So, if ω > 0 Open image in new window is sufficiently small, then r m ( ω ) Open image in new window satisfies the inequality m < r m ( ω ) < 1 m Open image in new window and thereby is negative.

Consider the product j = 0 n 1 | r m ( ω ) + ω i + j | Open image in new window with 0 < m < n Open image in new window and investigate its behavior for small ω > 0 Open image in new window.

If j m Open image in new window, then for sufficiently small ω > 0 Open image in new window, we have | r m ( ω ) + j | = r m ( ω ) + j < j Open image in new window and
j = m n 1 | r m ( ω ) + j | j = m n 1 j < j = m n 1 ( 1 + j ) . Open image in new window
(21)
If j m 1 Open image in new window, then for sufficiently small ω > 0 Open image in new window, we have | r m ( ω ) + j | = r m ( ω ) j < m j = 1 + ( m 1 j ) Open image in new window
j = 0 m 1 | r m ( ω ) + j | j = 0 m 1 | 1 + ( m 1 j ) | = J = 0 m 1 ( 1 + J ) , J = m 1 j . Open image in new window
(22)
Combining (21) and (22), we obtain, for sufficiently small ω > 0 Open image in new window,
j = 0 n 1 | r m ( ω ) + j | < j = 0 n 1 ( 1 + j ) , Open image in new window
and
j = 0 n 1 | r m ( ω ) + ω i + j | < j = 0 n 1 ( 1 + α + j ) . Open image in new window
As for large ω, the left-hand side of the above inequality evidently tends to +∞ as ω + Open image in new window and hence is greater than its right-hand side for sufficiently large ω. By continuity there exists ω m > 0 Open image in new window such that
j = 0 n 1 | r m ( ω m ) + ω m i + j | = j = 0 n 1 ( 1 + α + j ) . Open image in new window

Thus, we can take μ m = r m ( ω m ) + ω m i C Open image in new window to satisfy (19) and (20) for 0 < m < n Open image in new window. For n < m < 0 Open image in new window, we can take the conjugates μ m = μ m ¯ Open image in new window. Thus, the existence of all μ m Open image in new window needed is proved. This completes the proof of Proposition 2. □

Lemma 8 If 12 n 14 Open image in new window, α > 0 Open image in new window, and q > 0 Open image in new window satisfy the polynomial equation
j = 0 n 1 ( ( α + j ) 2 + q 2 ) = j = 0 n 1 ( α + j + 1 ) 2 , Open image in new window

then 2 α + 4 < q 2 < 3 α + 5 Open image in new window.

Proof It can be proved in the same way for all n mentioned. We show this for n = 12 Open image in new window.

First, compute the right-hand side of the equation:
j = 0 11 ( α + j + 1 ) 2 = α 24 + 156 α 23 + 11 , 518 α 22 + 535 , 392 α 21 + 17 , 581 , 135 α 20 + 433 , 823 , 676 α 19 + 8 , 353 , 410 , 208 α 18 + 128 , 665 , 048 , 512 α 17 + 1 , 612 , 229 , 817 , 055 α 16 + 16 , 625 , 859 , 652 , 116 α 15 + 142 , 196 , 061 , 481 , 318 α 14 + 1 , 013 , 438 , 536 , 648 , 512 α 13 + 6 , 032 , 418 , 472 , 347 , 265 α 12 + 29 , 989 , 851 , 619 , 249 , 236 α 11 + 124 , 253 , 074 , 219 , 885 , 468 α 10 + 427 , 135 , 043 , 298 , 835 , 872 α 9 + 1 , 209 , 806 , 045 , 835 , 003 , 760 α 8 + 2 , 795 , 060 , 589 , 044 , 133 , 696 α 7 + 5 , 194 , 030 , 186 , 679 , 450 , 688 α 6 + 7 , 613 , 724 , 634 , 416 , 755 , 712 α 5 + 8 , 564 , 233 , 279 , 835 , 510 , 784 α 4 + 7 , 096 , 936 , 674 , 284 , 421 , 120 α 3 + 4 , 059 , 952 , 667 , 309 , 260 , 800 α 2 + 1 , 424 , 017 , 035 , 657 , 216 , 000 α + 229 , 442 , 532 , 802 , 560 , 000 . Open image in new window
Now, estimate the left-hand side supposing q 2 3 α + 5 > 0 Open image in new window:
j = 0 11 ( ( α + j ) 2 + q 2 ) j = 0 11 ( ( α + j ) 2 + 3 α + 5 ) α 24 + 168 α 23 + 13 , 216 α 22 + 647 , 658 α 21 + 22 , 191 , 136 α 20 + 565 , 650 , 624 α 19 + 11 , 143 , 609 , 279 α 18 + 174 , 022 , 752 , 156 α 17 + 2 , 192 , 303 , 359 , 180 α 16 + 22 , 557 , 120 , 652 , 044 α 15 + 191 , 221 , 185 , 335 , 728 α 14 + 1 , 343 , 463 , 278 , 373 , 840 α 13 + 7 , 851 , 135 , 965 , 424 , 751 α 12 + 38 , 226 , 775 , 470 , 470 , 448 α 11 + 155 , 030 , 143 , 411 , 290 , 136 α 10 + 522 , 520 , 458 , 095 , 057 , 994 α 9 + 1 , 457 , 064 , 439 , 886 , 002 , 624 α 8 + 3 , 337 , 255 , 633 , 900 , 992 , 816 α 7 + 6 , 209 , 925 , 089 , 367 , 687 , 345 α 6 + 9 , 237 , 499 , 888 , 429 , 090 , 764 α 5 + 10 , 723 , 421 , 856 , 201 , 549 , 372 α 4 + 9 , 360 , 016 , 963 , 404 , 522 , 912 α 3 + 5 , 777 , 193 , 048 , 791 , 013 , 360 α 2 + 2 , 247 , 088 , 906 , 508 , 241 , 600 α + 413 , 920 , 896 , 501 , 672 , 000 . Open image in new window
The difference of this polynomial and the previous one is equal to
j = 0 11 ( ( α + j ) 2 + 3 α + 5 ) j = 0 11 ( α + j + 1 ) 2 = 12 α 23 + 1 , 698 α 22 + 112 , 266 α 21 + 4 , 610 , 001 α 20 + 131 , 826 , 948 α 19 + 2 , 790 , 199 , 071 α 18 + 45 , 357 , 703 , 644 α 17 + 580 , 073 , 542 , 125 α 16 + 5 , 931 , 260 , 999 , 928 α 15 + 49 , 025 , 123 , 854 , 410 α 14 + 330 , 024 , 741 , 725 , 328 α 13 + 1 , 818 , 717 , 493 , 077 , 486 α 12 + 8 , 236 , 923 , 851 , 221 , 212 α 11 + 30 , 777 , 069 , 191 , 404 , 668 α 10 + 95 , 385 , 414 , 796 , 222 , 122 α 9 + 247 , 258 , 394 , 050 , 998 , 864 α 8 + 542 , 195 , 044 , 856 , 859 , 120 α 7 + 1 , 015 , 894 , 902 , 688 , 236 , 657 α 6 + 1 , 623 , 775 , 254 , 012 , 335 , 052 α 5 + 2 , 159 , 188 , 576 , 366 , 038 , 588 α 4 + 2 , 263 , 080 , 289 , 120 , 101 , 792 α 3 + 1 , 717 , 240 , 381 , 481 , 752 , 560 α 2 + 823 , 071 , 870 , 851 , 025 , 600 α + 184 , 478 , 363 , 699 , 112 , 000 , Open image in new window

which is positive for any α 0 Open image in new window. This shows that the polynomial equation cannot be satisfied by α > 0 Open image in new window and q > 0 Open image in new window with q 2 3 α + 5 Open image in new window.

In the same way, compute
j = 0 11 ( α + j + 1 ) 2 j = 0 11 ( ( α + j ) 2 + 2 α + 4 ) = 96 α 22 + 13 , 156 α 21 + 844 , 624 α 20 + 33 , 778 , 316 α 19 + 943 , 838 , 852 α 18 + 19 , 590 , 096 , 240 α 17 + 313 , 464 , 915 , 984 α 16 + 3 , 960 , 996 , 926 , 744 α 15 + 40 , 162 , 617 , 066 , 616 α 14 + 330 , 203 , 929 , 721 , 796 α 13 + 2 , 215 , 299 , 128 , 334 , 800 α 12 + 12 , 163 , 303 , 361 , 220 , 828 α 11 + 54 , 651 , 209 , 110 , 677 , 476 α 10 + 200 , 323 , 721 , 839 , 107 , 240 α 9 + 595 , 229 , 721 , 350 , 941 , 648 α 8 + 1 , 419 , 051 , 246 , 703 , 474 , 880 α 7 + 2 , 673 , 079 , 829 , 956 , 829 , 568 α 6 + 3 , 889 , 993 , 689 , 940 , 050 , 432 α 5 + 4 , 228 , 750 , 706 , 659 , 177 , 984 α 4 + 3 , 257 , 831 , 645 , 648 , 401 , 920 α 3 + 1 , 625 , 109 , 784 , 526 , 284 , 800 α 2 + 437 , 271 , 322 , 981 , 376 , 000 α + 37 , 266 , 873 , 282 , 560 , 000 . Open image in new window

Hence, j = 0 11 ( α + j + 1 ) 2 > j = 0 11 ( ( α + j ) 2 + q 2 ) Open image in new window if 2 α + 4 q 2 Open image in new window.

This contradiction yields 2 α + 4 < q 2 < 3 α + 5 Open image in new window. So, Lemma 8 is proved. □

The condition Re d λ α d α ( α ˜ ) 0 Open image in new window needed for the Hopf theorem, expressed explicitly by means of the implicit function theorem, looks like
[ j = 0 n 1 α + j q 2 + ( α + j ) 2 ] 2 + [ j = 0 n 1 q q 2 + ( α + j ) 2 ] 2 j = 0 n 1 α + j q 2 + ( α + j ) 2 j = 0 n 1 1 1 + α + j . Open image in new window
Lemma 9 If 12 n 14 Open image in new window, α > 0 Open image in new window and 0 < q 2 < 3 α + 5 Open image in new window, then
[ j = 0 n 1 α + j q 2 + ( α + j ) 2 ] 2 + [ j = 0 n 1 q q 2 + ( α + j ) 2 ] 2 > j = 0 n 1 α + j q 2 + ( α + j ) 2 j = 0 n 1 1 1 + α + j . Open image in new window
(23)

Proof Hereafter all sums and products with no limits indicated are over j = 0 , 1 , , n 1 Open image in new window.

Multiplying inequality (23) by U = ( 1 + α + j ) Open image in new window and then twice by V = [ q 2 + ( α + j ) 2 ] Open image in new window, we obtain the following equivalent inequality provided α > 0 Open image in new window:
U [ ( ( α + j ) V j ) 2 + q 2 ( V j ) 2 ] > V ( α + j ) V j U j Open image in new window
(24)

with the polynomials U j = U 1 + α + j Open image in new window and V j = V q 2 + ( α + j ) 2 Open image in new window.

Put q 2 = 3 α + 5 1 + w Open image in new window, w > 0 Open image in new window. Substituting this into inequality (24) and multiplying the result by ( 1 + w ) 2 n 1 Open image in new window, we obtain another equivalent one:
U [ ( 1 + w ) ( ( α + j ) P j ) 2 + ( 3 α + 5 ) ( P j ) 2 ] > P ( α + j ) P j U j Open image in new window
(25)

with P = [ 3 α + 5 + ( 1 + w ) ( α + j ) 2 ] Open image in new window and P j = P 3 α + 5 + ( 1 + w ) ( α + j ) 2 Open image in new window.

Both sides of inequality (25) are polynomials of α and w with non-negative integer coefficients. So, they can be computed exactly, with no rounding. This rather cumbersome computation gives the following result for the difference of the left- and right-hand sides of (25) expressed as
U [ ( 1 + w ) ( ( α + j ) P j ) 2 + ( 3 α + 5 ) ( P j ) 2 ] P ( α + j ) P j U j = j = 0 5 n 2 Δ j α j Open image in new window
(26)

with polynomials Δ j R [ w ] Open image in new window. Straightforward though very cumbersome calculations show that Δ 5 n 2 = 0 Open image in new window, and all other Δ j Open image in new window in (26) are polynomials with positive coefficients.

This completes the proof of Lemma 9. □

To apply the Hopf bifurcation theorem, we need to check that equation (14) cannot have more than a single pair of imaginary conjugated roots. It can be easily obtained by considering equation (16).

Now, the Hopf bifurcation theorem and the lemmas proved provide, for n = 12 , 13 , 14 Open image in new window, the existence of a family α ε > 0 Open image in new window such that equation (14) with α = α 0 Open image in new window has imaginary roots λ = ± q i Open image in new window and for sufficiently small ε, system (13) with α = α ε Open image in new window has a periodic solution V ε ( t ) Open image in new window with period T ε T = 2 π q Open image in new window as ε 0 Open image in new window. In particular, the coordinate V ε , 0 ( t ) = v ( t ) Open image in new window of the vector V ε ( t ) Open image in new window is also a periodic function with the same period. Then, taking into account (9), we obtain
y ( x ) = ( C + v ( ln ( x x ) ) ) ( x x ) α . Open image in new window
Put h ( s ) = C + v ( s ) Open image in new window, which is a non-constant continuous periodic and positive for sufficiently small ε function and obtain the required equality
y ( x ) = ( x x ) α h ( ln ( x x ) ) . Open image in new window

In the similar way, we obtain the related expressions for y ( j ) ( x ) Open image in new window, j = 0 , 1 , , n 1 Open image in new window.

Theorem 4 is proved. □

Conclusions, concluding remarks and open problems

  1. 1.

    Computer calculations give approximate values of α providing equation (14) to have a pure imaginary root λ. They are, with corresponding values of k, as follows:

     

if n = 12 Open image in new window, then α 0.56 Open image in new window, k 22.4 Open image in new window;

if n = 13 Open image in new window, then α 1.44 Open image in new window, k 10.0 Open image in new window;

if n = 14 Open image in new window, then α 2.37 Open image in new window, k 6.9 Open image in new window.
  1. 2.

    Note that equation (14) has no pure imaginary roots if n 11 Open image in new window. So, the Hopf bifurcation theorem cannot be applied, but it does not follow that Theorem 4 cannot be proved for some n < 12 Open image in new window.

     
  2. 3.

    Equation (5) with n = 3 Open image in new window has solutions of type (6) with oscillatory h (see [3, 5]).

     
  3. 4.

    If n 15 Open image in new window, then the inequality needed for the Hopf bifurcation theorem Re d λ α d α ( α ˜ ) 0 Open image in new window cannot be proved in the same way because the estimate q 2 < 3 α + 5 Open image in new window does not hold.

     

Notes

Acknowledgements

The research was supported by RFBR (grant 11-01-00989).

References

  1. 1.
    Kiguradze IT, Chanturia TA: Asymptotic Properties of Solutions of Nonautonomous Ordinary Differential Equations. Kluwer Academic, Dordrecht; 1993.CrossRefGoogle Scholar
  2. 2.
    Astashova IV: Asymptotic behavior of solutions of certain nonlinear differential equations. 1(3). In Reports of Extended Session of a Seminar of the I. N. Vekua Institute of Applied Mathematics. Tbilis. Gos. Univ., Tbilisi; 1985:9-11. (Russian)Google Scholar
  3. 3.
    Astashova IV: Qualitative properties of solutions to quasilinear ordinary differential equations. In Qualitative Properties of Solutions to Differential Equations and Related Topics of Spectral Analysis. Edited by: Astashova IV. UNITY-DANA, Moscow; 2012:22-290. (Russian)Google Scholar
  4. 4.
    Kozlov VA: On Kneser solutions of higher order nonlinear ordinary differential equations. Ark. Mat. 1999, 37(2):305-322. 10.1007/BF02412217MathSciNetCrossRefGoogle Scholar
  5. 5.
    Astashova IV: Application of dynamical systems to the study of asymptotic properties of solutions to nonlinear higher-order differential equations. J. Math. Sci. 2005, 126(5):1361-1391.MathSciNetCrossRefGoogle Scholar
  6. 6.
    Marsden JE, McCracken M: The Hopf Bifurcation and Its Applications. Springer, Berlin; 1976. XIIICrossRefGoogle Scholar
  7. 7.
    Astashova IV, Vyun SA: On positive solutions with non-power asymptotic behavior to Emden-Fowler type twelfth order differential equation. Differ. Equ. 2012, 48(11):1568-1569. (Russian)Google Scholar

Copyright information

© Astashova; licensee Springer 2013

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Authors and Affiliations

  1. 1.Department of Mechanics and MathematicsLomonosov Moscow State UniversityMoscowRussia
  2. 2.Department of Higher MathematicsMoscow State University of Economics, Statistics and InformaticsMoscowRussia

Personalised recommendations