See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257831223

On Small Solutions of Nonlinear Equations with Vector Parameter in Sectorial

Neighborhoods

Article  in  Mathematical Notes · February 2012

DOI: 10.1134/S0001434612010105

CITATIONS READS
14 21

3 authors, including:

Nikolai A. Sidorov Aliona Dreglea

Irkutsk State University Irkutsk State Technical University
109 PUBLICATIONS   756 CITATIONS    41 PUBLICATIONS   302 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Solving Urban Infrastructure Problems Using Smart City Technologies View project

Skeleton decomposition of linear operators in the theory of degenerate differential equations View project

All content following this page was uploaded by Aliona Dreglea on 16 August 2018.

The user has requested enhancement of the downloaded file.

ISSN 0001-4346, Mathematical Notes, 2012, Vol. 91, No. 1, pp. 90–104. © Pleiades Publishing, Ltd., 2012.
Original Russian Text © N. A. Sidorov, R. Yu. Leont’ev, A. I. Dreglya, 2012, published in Matematicheskie Zametki, 2012, Vol. 91, No. 1, pp. 93–108.

On Small Solutions of Nonlinear Equations

with Vector Parameter in Sectorial Neighborhoods
N. A. Sidorov* , R. Yu. Leont’ev, and A. I. Dreglya**
Irkutsk State University
Received March 5, 2010

Abstract—We consider the nonlinear operator equation B(λ)x + R(x, λ) = 0 with parameter λ,
which is an element of a linear normed space Λ. The linear operator B(λ) has no bounded inverse
for λ = 0. The range of the operator B(0) can be nonclosed. The nonlinear operator R(x, λ)
is continuous in a neighborhood of zero and R(0, 0) = 0. We obtain sufficient conditions for the
existence of a continuous solution x(λ) → 0 as λ → 0 with maximal order of smallness in an open
set S of the space Λ. The zero of the space Λ belongs to the boundary of the set S. The solutions are
constructed by the method of successive approximations.

DOI: 10.1134/S0001434612010105
Keywords: nonlinear operator equation, Banach space, sectorial neighborhood, Fredholm
operator, bifurcation, Schmidt’s lemma, regularizer for a nonlinear operator.

1. INTRODUCTION
Suppose that X and Y are Banach spaces and Λ is a linear normed space. Consider the nonlinear
operator equation
B(λ)x + R(x, λ) = 0, (1.1)
where B(λ) is a closed linear operator with a dense (in X) domain independent of the parameter
λ ∈ Λ. The nonlinear operator R : X × Λ → Y is continuous in x and λ, in a neighborhood of zero
and R(0, 0) = 0.
It is required to construct the solution x(λ),
x(λ) → 0 as S  λ → 0, S∈Λ
(further, briefly, as λ → 0) of the maximal order of smallness (or the “minimal branch” of the small
solution) by the method of successive approximations in the nonregular case where the range of the
operator B(0) can be nonclosed and the dimension satisfies the inequality dim N (B(0)) ≥ 1. The
theory of branching of solutions of nonlinear equations was described in the well-known monograph [1].
Some recent results in this area were surveyed in [2]. In the present paper, we consider the problem of
constructing the “minimal branch” of the solution of Eq. (1.1), assuming that the operator B(λ) has an
inverse for λ ∈ S and
 
1
B −1 (λ) = O for S  λ → 0, (1.2)
a(λ)
where S is an open set in the space Λ whose boundary contains the point λ = 0 and
a(λ) : S ⊂ Λ → (0, +∞)
is a positive continuous functional, a(0) = 0. The set S is called a sectorial neighborhood of zero.
Estimates of the form (1.2) are valid in a number of applications. For example, suppose that X = Y = H.
*
E-mail: [email protected]
**
E-mail: [email protected]

90
 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 91

Consider the operator B(λ) = B0 + a(λ)B1 , where B0 is a self-adjoint nonnegative operator and B1 is
a self-adjoint positive operator, i.e.,
(B0 x, x) ≥ 0 and (B1 x, x) ≥ γ(x, x) for all x ∈ H.
Then we have
1
(B0 + a(λ)B1 )−1  ≤ ,
(γa(λ))
and estimate (1.2) holds.
If B(0) is a Fredholm operator, then in order to construct asymptotically small solutions of Eq. (1.1),
we can use classical results from the theory of branching of solutions of nonlinear equations [1]. General
existence theorems for bifurcation points in the Fredholm case and their applications can be found in
the monographs [2, pp. 5–49], [3, Chaps. 2, 5, 7], [4], etc. Approximate iterative methods and their
regularization in the neighborhood of branch points were studied in [3], [5]–[7], etc. However, in these
papers, it was assumed that Λ = R1 , the range of the operator B(0) is closed, and the method used in
them for choosing the initial approximations was complicated.
Let us introduce the set
Ω = {(x, λ) ∈ X × Λ, x ≤ a(λ)r, λ ∈ S}. (1.3)

The purpose of this paper is to prove constructive existence and uniqueness theorems for the minimal
branch of the solutions of Eq. (1.1) in domains of the form (1.3) with vector parameter λ in a sectorial
neighborhood S in the general case where the operator B(0) can fail to be normally solvable and
dim N (B(0)) ≥ 1.
The results of the paper are obtained by using methods from [1]–[3], [5]–[7], and they further
develop the theory of approximate methods in a neighborhood of branch points of solutions of nonlinear
equations. On the basis of the theorems proved in this paper, the required branch of the solution is
constructed by the method of successive approximations converging in the domain S for the zero initial
approximation. Thus, the problem of choosing the initial approximation does not arise in our method.
In Sec. 2, we study in detail the case in which B(0) is a Fredholm operator and the regularization of
the corresponding method of successive approximations is based on the application of the generalized
Schmidt lemma ([1, Sec. 21, Sec. 21.4] and [8, p. 221]). It is shown that, in the Fredholm case, we can
construct the method of successive approximations for the solution of Eq. (1.1), applying “asymptotic
regularizers” used earlier in [9] and in the monograph [3].
In Sec. 3, we present existence theorems and the method of successive approximations for the
minimal branches of the solutions of Eq. (1.1), provided that estimate (1.2) holds. The range of
the operator B(0) is not necessarily closed. The corresponding iterative formulas in Sec. 3 involve
the operator B −1 (λ). Therefore, in view of inevitable calculation errors, this method of successive
approximations requires a regularization in the sense of the theory of ill-posed problems [10]. Such a
regularization can be carried out in a neighborhood of the branch point of the solution by shifting the
parameter λ [11]. In some concrete examples of an explicitly constructed operator B −1 (λ) for which the
point λ = 0 turns out to be a removable singular point, the regularization is achieved by canceling the
admissible power of the parameter λ in the reduced equation.
The general existence theorems and the method of successive approximations are illustrated by
examples of the solution of nonlinear integral equations of the II and I kind and of boundary-value
problems often encountered in applications.

2. CONSTRUCTION OF THE MINIMAL BRANCH OF SOLUTIONS OF EQ. (1.1)

WITH FREDHOLM OPERATOR B(0)
Suppose that Eq. (1.1) is of the form
B(λ)x = R(x, λ) + b(λ). (2.1)

MATHEMATICAL NOTES Vol. 91 No. 1 2012

92 SIDOROV et al.

The linear operator B(λ) satisfies the conditions given in the Introduction; further, B(0) is a Fredholm
operator, {ϕi }n1 is a basis in N (B(0)), and {ψi }n1 is a basis in N ∗ (B(0)). The nonlinear operator
R : X × Λ → Y is continuous in x and λ in a neighborhood of zero, R(0, 0) = 0,
R(x1 , λ) − R(x2 , λ) ≤ L(r)x1 − x2 ,
for |λ| ≤ ρ and x1 , x2 from the ball S(0, r), and L(r) = O(r). The function b(λ) : Λ → Y is defined and
continuous in a neighborhood of the point λ = 0 and b(0) = 0. Since B(0) is a Fredholm operator, it is
easy to verify condition (1.2) by constructing the inverse operator B −1 (λ). Indeed, using the generalized
Schmidt lemma [1, Sec. 21, Sec. 21.4], we introduce the bounded Trenogin operator
 n −1
Γ = B(0) +
· , γi zi .
i=1

Here
ϕi , γk  = δik and
zi , ψk  = δik .
def
We introduce the notation A(λ) = B(λ) − B(0). Suppose that, for x ∈ D(B(0)),

I. A(λ)x ≤ c(λ)(x + bB(0)x), where c(λ) : S ⊂ Λ → (0, +∞) is a positive continuous func-
tional defined in a sectorial neighborhood S of zero, c(0) = 0, and b ≥ 0.

Note that
lim A(λ)Γ = 0
Sλ→0

and the operator I − A(λ)Γ has a bounded inverse λ ∈ S0 ⊂ S, where A(λ)Γ ≤ q < 1.
Suppose that

II. |det
(I − A(λ)Γ)−1 A(λ)ϕi , ψk i,k=1,...,n | = Δ(λ), where Δ(λ) ∼ a(λ) as S  λ → 0 and a(λ) is
a positive continuous functional, a(0) = 0.
Remark 1. If
  p−1  
 
det (A(λ)Γ)j
A(λ)ϕ , ψ  ∼ cλα as λ → 0
 i k 
j=0 i,k=1,...,n

and α ≤ p, then condition II holds for a(λ) = cλα .

Lemma 1. Suppose that B(0) is a Fredholm operator and conditions I and II hold. Then the
operator B(λ) has a bounded inverse λ ∈ S and estimate (1.2) holds.

Proof. Let us search for the solution of the linear equation B(λ)x = f for λ ∈ S in the form

n
x = Γu + ci ϕi ,
i=1

where
u, ψk  = 0, k = 1, . . . , n. Then
 
n 
−1
u = (I − A(λ)Γ) f + A(λ) ci ϕi .
i=1

)
Here, for λ ∈ S, the vector c = (c1 , . . . , cn is determined from the system of linear algebraic equations

n

(I − A(λ)Γ)−1 A(λ)ϕi , ψk ci =
(I − A(λ)Γ)−1 f, ψk , k = 1, . . . , n,
i=1

with nondegenerate (for λ ∈ S) matrix by condition II. In view of condition II, the estimate
c = O(1/a(λ)) holds, and so Lemma 1 is proved.

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 93

Further, suppose that B(λ) = B0 − c(λ)B1 , and, for x ∈ D(B(0)),

B1 x ≤ a(x + bB0 x), a > 0, b ≥ 0.
Suppose that [1, Chap. 9]

(k)
III. the elements {ϕi }, i = 1, . . . , n, k = 1, . . . , pi , constitute a complete B1 -Jordan block of the
(k)
Fredholm operator B0 , and the functionals {ψi }, i = 1, . . . , n, k = 1, . . . , pi , constitute the
complete B1∗ -Jordan block of the operator B0∗ .

Let us recall [1, Chap. 9] that, in condition III, we can assume without loss of generality that
= (Γ∗ B1∗ )k−1 ψi ,
(k) (1) (k) (1)
ϕi = (ΓB1 )k−1 ϕi , ψi k = 2, . . . , pi , i = 1, . . . , n,

B1∗ ψi i , ϕk 
(p ) (1) (p ) (1)

B1 ϕi i , ψk  = δik , = δik ,
 n −1
γi = B1∗ ψi
(pi ) (pi )
Γ= B0 +
· , γi zi , , zi = B1 ϕi .
i=1

Definition. A linear operator Γ(λ) is called a left asymptotic regularizer of the operator function B(λ)
if
lim Γ(λ)B(λ)x = x for all x ∈ D(B).
Sλ→0

In a similar way, we introduce the right asymptotic regularizer of the operator B(λ). If λ = 0 is an
isolated singular Fredholm point of the operator function B(λ), then the asymptotic regularizers can be
constructed in explicit form.

Lemma 2. Suppose that B(λ) = B0 − c(λ)B1 and condition III holds. Then, in a neighborhood
of 0 < |c(λ)| < ε, there exists a bounded inverse operator B −1 (λ), as well as left and right
regularizers Γl (λ) and Γr (λ) of the operator B(λ) defined by the formulas

n 
pi
c−s (λ)
· , ψi
(pi +1−s) (1)
Γl (λ) = Γ − ϕi ,
i=1 s=1

n 
pi
c−s (λ)
· , ψi ϕi
(1) (pi +1−s)
Γr (λ) = Γ − .
i=1 s=1

Here
B −1 (λ) = Γr (λ)(I − c(λ)B1 Γ)−1 ,
where


−1 1
B (λ) = O p , p = max(p1 , . . . , pn ).
c (λ)

If B1 is a bounded operator and

1
0 < |c(λ)| < ,
Γ B1 
then the following operator identities hold:
B −1 (λ) = (I − c(λ)ΓB1 )−1 Γl (λ) = Γr (λ)(I − c(λ)B1 Γ)−1 .

MATHEMATICAL NOTES Vol. 91 No. 1 2012

94 SIDOROV et al.

Proof. Under condition III, the existence of the inverse operator B −1 (λ) in a neighborhood of
0 < |c(λ)| < ε is well known [1, p. 428 (Russian edition)]. The validity of the operator identities
Γl (λ)(B0 − c(λ)B1 ) = I − c(λ)ΓB1 , (B0 − c(λ)B1 )Γr (λ) = I − c(λ)B1 Γ
can be verified by simple calculations, using the definition of the operator
 n −1
def
Γ = B0 +
· , γi zi
i=1
(j) (j)
and the construction of the elements {ϕi }, {ψi }, {γi }, {zi }.

Remark 2. Using the definition of the operator Γ and the identities

(k)

zl , ψi  = 0
for i = l, k = 1, it is easy to verify that the regularizers Γr (λ) and Γl (λ) are epimorphisms.

Remark 3. Lemma 3 can be generalized to the case of the bounded operators B(λ) = B0 − c(λ)M B1
if the linear operator M ∈ L , (X → X) commutes with the operators B1 and B0 and has a bounded
inverse. If condition III holds, then the left regularizer for such an operator is of the form

n 
pi
c−s (λ)
M −s ·, ψi
(pi +1−s) (1)
Γl (λ) = Γ − ϕi
i=1 s=1

and we have the identity

Γl (λ)(B0 − c(λ)M B1 ) = I − c(λ)ΓM B1 .

Left and right regularizers for the operator L(∂/∂t)B0 − B1 with the differential operator multiplying
the Fredholm operator B0 were introduced in [3, Chap. 1]. Similar regularizers were used in the
numerical solution of algebro-differential equations [12].
Left and right regularizers can be used for constructing the solutions of Eq. (2.1).
Theorem 1. Suppose that B(λ) = B0 − c(λ)B1 , condition III holds, max(p1 , . . . , pn ) = p, and B1
is a bounded operator. Suppose that
p·l
b(λ) = o(c(λ)), R(x, λ) = O(xl ), l ≥ 2, m≥ .
l−1
Then, for λ ∈ S0 ⊂ S, Eq. (2.1) has a unique minimal branch x(λ) → 0 as λ → 0. The branch x(λ)
satisfies the estimate
x(λ) = o(c(λ)p/(l−1) ),
and the sequence xn (λ), where
xn = c(λ)ΓB1 xn−1 + Γl (λ)(R(xn−1 , λ) + b(λ)), x0 = 0, n = 1, 2, . . . , (2.2)
converges to this branch for λ ∈ S0 .

Proof. Applying the operator Γl (λ), which is an epimorphism for λ ∈ S, to both sides of Eq. (2.1), we
obtain the equation
x − c(λ)ΓB1 x = Γl (λ)(R(x, λ) + b(λ)). (2.3)
Replacing
x = c(λ)p/(l−1) V (2.4)
we reduce Eq. (2.3) to the form
V − c(λ)ΓB1 V = c(λ)−p/(l−1) Γl (λ)(R(c(λ)p/(l−1) V, λ) + b(λ)). (2.5)

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 95

In view of the assumptions of Theorem 1 in a neighborhood of the points λ = 0, V = 0, Eq. (2.5) satisfies
all the assumptions of the implicit operator theorem [8, p. 409]. Therefore, Eq. (2.5) has a unique solution
V (λ) → 0 as λ → 0. The sequence
Vn = c(λ)ΓB1 Vn−1 + c(λ)−p/(l−1) Γl (λ)(R(c(λ)p/(l−1) Vn−1 , λ) + b(λ)), V0 = 0,
converges to this solution. Taking the replacement (2.4) into account, we see that the successive
approximations (2.2) also converge to the required branch.
Remark 4. Under the assumptions of Theorem 1, the system of ramification of the small solutions of
Eq. (2.1) (see [8, p. 429]) is of the form

n
aik (λ)ξk = Γi (ξ, λ) + bi (λ), i = 1, . . . , n,
k=1

where
aik (λ) = δik c(λ)pk + o(c(λ)pk ), i, k = 1, . . . , n,
max(p1 , . . . , pn ) · l
|Γi (ξ, λ)| = O(|ξ|l ), |bi (λ)| = o(|c(λ)|m ), m ≥ .
l−1
In the domain |ξ| ≤ rc(λ)p/(l−1) , this system has a unique minimal small solution with the estimate
|ξ| = o(c(λ)p/(l−1) ). Therefore, the existence and uniqueness of the minimal branch of the solution of
Eq. (2.1) are a consequence of the structure of the ramification system. Note that, in the case of the
problem of the bifurcation point when bi (λ) = 0, the minimal branch is the trivial solution.
Corollary 1. Suppose that, under the assumptions of Theorem 1,

B1 ϕi , ψk  = δik , i, k = 1, . . . , n.
Then the sequence xn , where xn is uniquely defined by the linear equation
    
1 
n n
B0 +
· , γi zi xn = I −
· , ψi zi (R(xn−1 , λ) + b(λ)) + c(λ)B1 xn−1 , (2.6)
c(λ)
i=1 i=1

x0 = 0, γi = B1∗ ψi , zi = B1 ϕi , converges to the minimal branch of the solution of Eq. (2.1).

Proof. Since
B1 ϕi , ψk  = δik , the left regularizer in formula (2.2) is of the form
1 
n
Γl (λ) = Γ −
· , ψi ϕi .
c(λ)
i=1

Therefore, by Theorem 1, the sequence xn can be written as follows:

 
1 
n
xn = c(λ)ΓB1 xn−1 + Γ −
· , ψi ϕi (R(xn−1 , λ) + b(λ)),
c(λ)
i=1

x0 = 0. Multiplying both sides of this expression by the inverse operator Γ−1 , we obtain the required
result (2.6).
Example 1. The integral equation
ˆ
1 2π
u(x) − cos x cos s[(1 + λ)u(s) − su3 (s)] ds = λ3 cos x
π 0
has exactly three small solutions

λ
u = −λ cos x + O(|λ| ),
1 2 3
u 2,3
= ±2
cos x + O(|λ|)
3π
√
and satisfies the assumptions of Corollary 3. Here n = 1, ϕ = ψ = γ = z = cos x/ π Γ = I.

MATHEMATICAL NOTES Vol. 91 No. 1 2012

96 SIDOROV et al.

Successive approximations converging to the minimal branch u1 (x, λ), can be constructed by using
Corollary 1 and the formula
ˆ  
1 2π s 3
um (x) = cos x cos s λum−1 (s) − sum−1 (s) + um−1 (s) ds − λ2 cos x + λ3 cos x,
3
π 0 λ
u0 (x) = 0.

3. SUFFICIENT CONDITIONS FOR THE EXISTENCE OF A MINIMAL BRANCH

OF THE SOLUTION OF EQ. (1.1) AND THE METHOD OF SUCCESSIVE
APPROXIMATIONS IN THE GENERAL CASE: THE OPERATOR B(0)
NEED NOT BE NORMALLY SOLVABLE
Let us present sufficient conditions for the existence of a small solution of Eq. (1.1) in a domain
Ω0 ⊆ Ω and describe a method of constructing the solution and its asymptotics for the general case in
which the range of the operator B(0) is not necessarily closed.

Theorem 2. Suppose that condition (1.2) holds in the domain Ω and also

1) the following inequality holds:

R(x1 , λ) − R(x2 , λ) ≤ L(λ)rx1 − x2 , (3.1)
where L(λ) = O(a(λ)) as λ → 0;

2) the following estimate holds:

R(0, λ) = o(a2 (λ)) as λ → 0.

Then there exists a domain

Ω0 = {(x, λ) ∈ X × Λ, x ≤ a(λ)r0 , λ ∈ S0 ⊆ S, 0 < r0 ≤ r}
in which there is a unique solution of Eq. (1.1) x(λ) → 0 as λ → 0. The sequence {xn }, x0 = 0,
where xn is constructed by the method of successive approximations (see formula (3.3) in the
proof), converges to this solution.

Proof. Replacing x(λ) = a(λ)V (λ), we can write Eq. (1.1) in the equivalent form
1
V =− B −1 (λ)R(a(λ)V, λ) ≡ Φ(V, λ). (3.2)
a(λ)
Here, in view of estimate (3.1),
Φ(V1 , λ) − Φ(V2 , λ) ≤ B −1 (λ)L(λ)r V1 − V2 
for λ ∈ S, Vi  ≤ r.
By estimate (1.2), we have
B −1 (λ)L(λ) ≤ C for λ ∈ S,
where C > 0 is a fixed constant. But, in that case,
Φ(V1 , λ) − Φ(V2 , λ) ≤ CrV1 − V2 .
Let 0 < q < 1, and choose r0 ≤ min{q/C, r}. Then the operator Φ(V, λ) is a contraction operator in the
ball V  ≤ r0 for all λ ∈ S.
Further,
1
Φ(V, λ) ≤ Φ(V, λ) − Φ(0, λ) + Φ(0, λ) ≤ qr0 + B −1 (λ)R(0, λ).
a(λ)

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 97

In view of estimate (1.2) and condition 2) of Theorem 2, there exists a sectorial neighborhood of zero
S0 ⊆ S such that, for all λ ∈ S, the following inequality holds:
1
B −1 (λ) R(0, λ) ≤ (1 − q)r0
a(λ)
and, therefore,
Φ(V, λ) ≤ r0 for λ ∈ S0 , V  ≤ r0 .
Thus, by the contraction mapping principle, Eq. (3.2) has a unique solution, the sequence {xn },
where
xn = −B −1 (λ)R(xn−1 , λ), x0 = 0, (3.3)
converges for all λ ∈ S0 . Theorem 2 is proved.
Remark 5. Under the assumptions of Theorem 2, there exists a unique small solution in the ball
x ≤ a(λ)r0 for λ ∈ S0 . Note that the order of this solution, although infinitely small at zero, can turn
out to be higher than that of the infinitely small quantity a(λ). In the domain x ≥ a(λ)r0 , Eq. (1.1) can
also have other small continuous solutions, obviously, with a lower order of smallness than that of the
solution x(λ), unique in the ball S(0, a(λ)r0 ), to which the sequence (3.3) converges. To find solutions of
lower order of smallness, one must construct more refined approximate methods; see [1]–[3] and [5]–[7].

The following Theorems 3 and 4 allow us to weaken condition 2) of Theorem 2 by replacing it by the
assumption of the existence of a continuous function b(λ) : S ⊂ Λ → Y , b(0) = 0, such that
R(0, λ) − b(λ) = o(a2 (λ)) for λ → 0. (3.4)
In Theorems 3 and 4, we assume that
B(λ) = B + a(λ)A + ω(λ), (3.5)
where B and A are closed linear operators with dense domains, D(B) ⊆ D(A) in the Banach space X,
while ω(λ) is a linear continuous operator function and ω(λ) = o(a(λ)) as λ → 0.
In the proofs of Theorems 3 and 4, just as in Theorem 2, we use the contraction mapping principle
and the equivalent equation (3.2). Note that the part of the proof dealing with the contraction property
of the corresponding operator will be omitted, because it is similar to the one in Theorem 2.
Theorem 3. Suppose that estimates (1.2), (3.1), and (3.4) hold and the operator B(λ) is of the
form (3.5). Suppose that, for λ ∈ S, the linear equation Bx = b(λ) has a solution x∗ (λ) such that
x∗ (λ) = o(a(λ)), Ax∗ (λ) = o(a(λ)).
Then, in the domain Ω0 , there exists a unique small solution of Eq. (1.1).

Proof. Just as for Theorem 2, by replacing x(λ) = a(λ)V (λ), we reduce Eq. (1.1) to the form
V = Φ(V, λ), (3.6)
where the operator Φ(V, λ) is defined by (3.2) and, in view of the proof of Theorem 2, it is a contraction
operator in the ball V  ≤ r0 for all λ ∈ S.
Let us show that there exists a domain S0 ⊆ S such that, for λ ∈ S0 , we have Φ : S(0, r0 ) →
S(0, r0 ). Indeed, using the fact that Φ(V, λ) is a contraction operator, the representation (3.5), and
estimates (1.2), (3.1), and (3.4), we obtain the chain of inequalities
1
Φ(V, λ) ≤ qr0 + B −1 (λ)R(0, λ)
a(λ)
1 1
≤ qr0 + B −1 (λ)R(0, λ) − b(λ) + B −1 (λ)Bx∗ (λ)
a(λ) a(λ)
1 1
≤ qr0 + B −1 (λ)R(0, λ) − b(λ) + B −1 (λ)(B(λ) − a(λ)A − ω(λ))x∗ (λ)
a(λ) a(λ)

MATHEMATICAL NOTES Vol. 91 No. 1 2012

98 SIDOROV et al.
1 1
≤ qr0 + B −1 (λ)R(0, λ) − b(λ) + x∗ (λ)
a(λ) a(λ)
1
+ B −1 (λ)Ax∗ (λ) + B −1 (λ)ω(λ)x∗ (λ).
a(λ)
Here, under the assumptions of Theorem 3, the last four summands tend to zero as λ → 0. Therefore,
by choosing a domain S0 ⊆ S, the sum of these four summands can be made less than (1 − q)r0 for all
λ ∈ S0 . Therefore, there exists a sectorial neighborhood of zero S0 ⊆ S such that, for all λ ∈ S0 , we have
Φ(V, λ) ≤ r0 in the ball V  ≤ r0 . Theorem 3 is proved.
Remark 6. If B is a Fredholm operator and the equation Bx = b(λ) is solvable, then we can set
x∗ (λ) = Γb(λ), where Γ is the Trenogin operator (see [1, p. 340 (Russian edition)] and Sec. 2 in this
paper). Further, if b(λ) = o(a(λ)), then, obviously, the following estimates hold:
x∗ (λ) = o(a(λ)) and Ax∗ (λ) = o(a(λ)).
In Theorems 2 and 3, the method of successive approximations is used to construct small solutions
of the maximal order of smallness. Note that the order of smallness of the constructed solutions may
turn out to be higher than that of the infinitely small quantity a(λ), because the principal term of the
asymptotics of the solution was not written out in Theorems 2 and 3.
Theorem 4 answers the question about the principal term of the asymptotics of the minimal branch.
In Theorem 4, we give sufficient conditions of the existence and uniqueness of the small continuous
solution in a ball x − a(λ)V0  ≤ a(λ)r0 , where V0 is a certain nonzero element from the space X. For
brevity, we introduce the notation
def
F (x, λ) = B(λ)x + R(x, λ).
Theorem 4. Suppose that estimates (1.2) and (3.4) hold and the operator B(λ) is of the form (3.5).
Suppose that there exists an element V0 ∈ X such that the linear equation
B(λ)x = F (a(λ)V0 , λ) (3.7)
has a solution x∗ (λ), with x∗ (λ) = o(a(λ)) as λ → 0. Suppose that condition 1) of Theorem 2
holds in the ball x − a(λ)V0  ≤ a(λ)r. Then there exists a sectorial neighborhood of zero S0 ⊆ S
and 0 < r0 ≤ r such that, for all λ ∈ S0 , there exists in the ball x − a(λ)V0  ≤ a(λ)r0 a unique
continuous solution of Eq. (1.1) of the form
x(λ) = a(λ)V (λ), where V (λ) → V0 as λ → 0.

Proof. Consider the equivalent equation (3.6) with the operator Φ(V, λ), λ ∈ S, which is a contraction
operator in view of the proof of Theorem 2, in some ball V − V0  ≤ r0 , where 0 < r0 ≤ r.
Let us show that there exists a sectorial neighborhood of zero S0 ⊆ S such that, for all λ ∈ S0 , the
mapping Φ(V, λ) : S(V0 , r0 ) → S(V0 , r0 ) holds.
Indeed, using the fact that Φ(V, λ) is a contraction operator, we obtain the inequality
Φ(V, λ) − V0  ≤ Φ(V, λ) − Φ(V0 , λ) + Φ(V0 , λ) − V0 
1 x∗ (λ)
≤ qr0 + B −1 (λ)(R(a(λ)V0 , λ) + B(λ)a(λ)V0 ) ≤ qr0 + .
a(λ) a(λ)
In view of the estimate x∗ (λ) = o(a(λ)), there exists a sectorial neighborhood of zero S0 ⊆ S such
that, for all λ ∈ S0 , the following inequality holds:
x∗ (λ)
≤ (1 − q)r0 ,
a(λ)
and, therefore, Φ(V, λ) − V0  ≤ r0 . Theorem 4 is proved.

In Theorem 4, the question of how to construct the element V0 appearing on the right-hand side of
the linear equation (3.7) remained open. The following lemma shows that this element can be chosen
from the subspace N (B(0)).

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 99

Lemma 3. Suppose that

1) R(0, λ) = a2 (λ)b + o(a2 (λ));

2) there exists an element ϕ ∈ N (B) such that the linear equation

Bx = b + Aϕ (3.8)
has a solution x0 ;

3) R(a(λ)ϕ, λ) − R(0, λ) = o(a2 (λ));

4) for the operator B(λ) of the form (3.5) estimate (1.2) holds.

Then, for V0 = ϕ, Eq. (3.7) has a small solution x∗ (λ) such that x∗ (λ) = o(a(λ)).

Proof. For V0 = ϕ, Eq. (3.7) takes the form

B(λ)x = a2 (λ)Aϕ + ω(λ)a(λ)ϕ + R(a(λ)ϕ, λ).
Taking into account the fact that Aϕ = Bx0 − b, we obtain the following unique solution of Eq. (3.7):
x∗ (λ) = B −1 (λ){a2 (λ)(Bx0 − b) + ω(λ)a(λ)ϕ + R(a(λ)ϕ, λ)}.
Let us now show that x∗ (λ) = o(a(λ)). To do this, we use the identity
B = B(λ) − a(λ)A − ω(λ),
and rewrite the solution as
x∗ (λ) = B −1 (λ){a2 (λ)(−a(λ)A − ω(λ))x0 + R(a(λ)ϕ, λ) − R(0, λ) + R(0, λ) − a2 (λ)b}
+ a2 (λ)x0 .
Hence we obviously have the estimate x∗ (λ) = o(a(λ)).

Remark 7. If the operator B is Fredholm, {ϕi }, {ψi }, i = 1, . . . , n, are bases, respectively, in N (B)
and N ∗ (B), and det
Aϕ
i , ψk |i,k=1,...,n = 0, then the required element ϕ in condition 2) of Lemma 3 can
be constructed as ni=1 ci ϕi , where the vector (c1 , . . . , cn ) is uniquely determined from the solvability
condition for Eq. (3.8).

As a consequence Theorem 4 and Lemma 3, we obtain the following result.

Corollary 2. Suppose that the assumptions of Lemma 3 hold. Suppose that condition 1) of
Theorem 2 holds in the ball
x − a(λ)ϕ ≤ a(λ)r.
Then there exists a sectorial neighborhood S0 ⊆ S and 0 < r0 ≤ r such that, for all λ ∈ S0 ,
there exists a unique small solution x(λ) of Eq. (1.1) in the ball x − a(λ)ϕ ≤ a(λ)r0 such that
x(λ) ∼ a(λ)ϕ as λ → 0.

If the range of the operator B is not closed, then, in N (B), there may be no vector ϕ, for which
Eq. (3.8) is solvable. If, however, in N (B), there exists a vector ϕ, for which b + Aϕ ∈ R(B), then the
following result can be obtained.

Corollary 3. Let conditions 1) and 3) of Lemma 3 hold; in condition 3), let the element ϕ ∈ N (B)
satisfy the inclusion b + Aϕ ∈ R(B). Suppose that condition 1) of Theorem 2 holds in the ball
x − a(λ)ϕ ≤ a(λ)r. Then the result of Corollary 2 remains valid.

MATHEMATICAL NOTES Vol. 91 No. 1 2012

100 SIDOROV et al.

Proof. Replacing x(λ) = a(λ)V (λ), we introduce Eq. (3.2). equivalent to (1.1) Just as in Theorem 2,
we can show that the corresponding operator Φ(V, λ) is a contraction one for all λ ∈ S in some ball
x − a(λ)ϕ ≤ a(λ)r0 , where 0 < r0 ≤ r.
It remains to show that there exists a sectorial neighborhood of zero S0 ⊆ S such that, for all λ ∈ S0
and all V ∈ S(ϕ, r0 ), the inequality Φ(V, λ) − ϕ ≤ r0 holds. To this end, using the fact that Φ(V, λ) is
a contraction operator, we obtain the estimate
Φ(V, λ) − ϕ ≤ Φ(V, λ) − Φ(ϕ, λ) + Φ(ϕ, λ) − ϕ
1
≤ qr0 + B −1 (λ)(R(a(λ)ϕ, λ) + B(λ)a(λ)ϕ).
a(λ)
Since ϕ ∈ N (B), we have
B(λ)ϕ = a(λ)Aϕ + ω(λ)ϕ.
Therefore,
1
B −1 (λ)(R(a(λ)ϕ, λ) + B(λ)a(λ)ϕ)
a(λ)
1
≤ B −1 (λ)(R(a(λ)ϕ, λ) − a2 (λ)b + ω(λ)a(λ)ϕ) + a(λ)B −1 (λ)(b + Aϕ).
a(λ)
The first summand in the last expression is an infinitely small quantity as λ → 0 in view of the
assumptions of Corollary 3. Let us show that
lim a(λ)B −1 (λ)(b + Aϕ) = 0,
λ→0

using the fact that b + Aϕ ∈ R(B). Consider the operator a(λ)B −1 (λ) for λ ∈ S. In view of es-
timate (1.2) valid for any sequence {λn } in S, the operators a(λn )B −1 (λn ) are uniformly bounded.
Suppose that the element g belongs to the range of the operator B, i.e., g = Bx, where x ∈ D(B).
Then
a(λ)B −1 (λ)g = a(λ)B −1 (λ)Bx = a(λ)B −1 (λ)(B(λ) − a(λ)A − ω(λ))x
= a(λ)x − a2 (λ)B −1 (λ)Ax − a(λ)B −1 (λ)ω(λ)x,
where the norms of all summands on the right are infinitely small as λ → 0. Thus,
lim a(λ)B −1 (λ)g = 0 for all g ∈ R(B).
λ→0

Therefore, by the Banach–Steinhaus theorem, for all g ∈ R(B) we have

lim a(λ)B −1 (λ)g = 0.
λ→0

Corollary 3 is proved.
Example 2. Consider the nonlinear integral equation
ˆ 1 ˆ 1
tsx(s) ds + λx(t) − x2 (s) ds − tλ3 = 0, x(t) ∈ C[0,1] . (3.9)
0 0
Here
ˆ 1 ˆ 1
B(λ)x = tsx(s) ds + λx(t), R(x, λ) = − x2 (s) ds − tλ3 .
0 0

The inverse operator B −1 (λ) is of the form

ˆ 1
−1 1 3t
B (λ)f = f (t) − sf (s) ds (3.10)
λ (3λ + 1)λ 0

and satisfies estimate (1.2) for a(λ) = |λ|.

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 101

Further, we can easily see that conditions 1) and 2) of Theorem 2 with the coefficient L(λ) = 2|λ|
hold in the ball x ≤ |λ|r.
Thus, by Theorem 2, Eq. (3.9) has a unique small solution in the ball x(t) ≤ |λ|r0 ≤ |λ|r for all λ,
0 < |λ| < ρ.
Let us point out a regular method for constructing the required solution by the method of successive
approximations. Using the replacement x(t) = λV (t) and the operator (3.10), we reduce Eq. (3.9) to
the regular form
 ˆ 1
3t 3λ2 t
V (t) = 1 − V 2 (s) ds + .
6λ + 2 0 3λ + 1
The unique small solution of this equation can be constructed by the method of successive approxima-
tions:
 ˆ 1
3t 3λ2 t
Vn+1 (t) = 1 − Vn2 (s) ds + ,
6λ + 2 0 3λ + 1
choosing V0 = 0 for the initial approximation or by the method of the indeterminate coefficients in the
form of the series
∞
V (t) = Vi (t)λi .
i=2

The solution thus constructed of Eq. (3.9) will have the form
45t + 6 5 81t + 36 6
x(t) = 3tλ3 − 9tλ4 + λ − λ + ··· .
2 2
Example 3. Consider the equation
ˆ
2 π
sin t sin sx(s) ds + (a1 λ1 + a2 λ2 )x(t) − x4 (t) − f (t, λ1 , λ2 ) = 0, x(t) ∈ C[0,π] , (3.11)
π 0
where λ ∈ R2 and ai = 0. The function f (t, λ1 , λ2 ) is continuous in t and is analytic in λ1 and λ2 . The
sectorial neighborhood S in which the solution is sought is defined by the condition sign λi = sign ai .
Here
ˆ
2 π
B(λ)x = sin t sin sx(s) ds + (a1 λ1 + a2 λ2 )x(t).
π 0

The operator B −1 (λ) is of the form

ˆ π
1 2 sin t
B −1 (λ)h = h(t) − sin sh(s) ds
a1 λ1 + a2 λ2 π(a1 λ1 + a2 λ2 + 1)(a1 λ1 + a2 λ2 ) 0

and satisfies estimate (1.2) for a(λ) = |a1 λ1 + a2 λ2 |.

If
f (t, λ1 , λ2 ) = o(|a1 λ1 + a2 λ2 |2 ),
then all the assumptions of Theorem 2 hold.
If

f (t, λ1 , λ2 ) = sin t · aik λi1 λk2 + o(|a1 λ1 + a2 λ2 |2 ),
i+k=2

then all the assumptions of Theorem 2 hold.

In both cases, the integral equation has the solution x(λ) → 0 as λ → 0.

MATHEMATICAL NOTES Vol. 91 No. 1 2012

102 SIDOROV et al.

Example 4. Consider the system of differential equations with boundary conditions

ü + u + R(u, x, λ) = 0, u(0) = u(π) = 0, (3.12)
where u = (u1 , . . . , un ) , λ = (λ1 , . . . , λm ) , R = (R1 , . . . , Rn ) , and λ ∈ S ⊂ Rm .
Suppose that

n
Ri = aik (λ)uk + fi (u, x, λ) + bi (x, λ),
k=1

where aik (λ) − a(λ)bik = o(a(λ)) as S  λ → 0 and det[bik ] = 0. Here

a(λ) : S ⊂ Λ → (0, +∞)
is a positive continuous function and a(0) = 0.
The functions fi are continuous and differentiable with respect to u for
u < r, x ∈ [0, π], λ ∈ S, |fi | = O(u3 ), |bi | = O(a2 (λ)).
(2)
We introduce the Banach space X = Ċ[0,π] of twice differentiable vector functions u satisfying the
boundary conditions in (3.12) and also the space Y = C[0,π] of continuous vector functions. Then the
boundary-value problem (3.12) will satisfy the assumptions of Theorem 2 and have the unique minimal
solution u → 0 as S  λ → 0. The sequence {un }, where un is the solution of the linear boundary-value
problem
ün + un + a(λ)Bun + R(un−1 , x, λ) − a(λ)Bun−1 = 0,
un (0) = un (π) = 0, u0 = 0, B = [bik ]ni,k=1 ,
converges for λ ∈ S0 ⊂ S to the minimal branch of the solution of problem (3.12).
Example 5. Theorem 2 can be used for the solution of differential equations on the semiaxis. Indeed,
consider the equation
d2 u
= f (u, λ), (3.13)
dx2
where λ is a small positive parameter and 0 < x < ∞. Suppose that f (u, λ) is a continuous function in
the domain Ω = {|u| < r, 0 < λ < ρ} satisfying the estimates

a) sup0<λ<ρ |fu (u, λ) − α2 | = O(|u|);

b) |f (0, λ)| = O(λ4 );

c) |f (u, λ) − α2 u − f (0, λ) = O(|u|2 ), where α is a constant.

It is required to construct a small solution u → 0 as λ → +0 on an arbitrarily large interval [0, 1/λ]

satisfying the boundary conditions
u|x=0 = 0, u|x=1/λ = 0. (3.14)
Similar boundary-value problems arise in some models of the theory of polymer modeling [13]. Since,
for λ = 0, Eq. (3.13) has a trivial solution u = 0, it can easily be verified that problem (3.13), (3.14) has a
solution. Indeed, by replacing x = t/ε, we can reduce the boundary-value problem (3.13), (3.14) to the
form
d2 u
λ2 2 = f (u, λ), 0 < t < 1,
dt
u|t=0 = 0, u|t=1 = 0
and consider it as the nonlinear operator equation
F (u, λ) = 0. (3.15)

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 SOLUTIONS OF NONLINEAR EQUATIONS IN SECTORIAL NEIGHBORHOODS 103

Here the operator

2

2d u
F (u, λ) = λ − f (u, λ), u(0), u(1)
dt2
(2)
acts from the Banach space X = C[0,1] to the Banach space

Y = C[0,1]  R2 .
The operator F is continuous in u, λ and has a Fréchet derivative with respect to u corresponding to the
linear boundary-value problem
d2 u
λ2 − α2 u = h(t),
dt2
u(0) = u(1) = 0.
Since the corresponding Green function is of the form
⎧
⎪
⎪ sinh k(t − 1) sinh ks
⎨ , s ≤ t,
G(t, s) = k
⎪
⎪ sinh kt sinh k(s − 1)
⎩ , s ≥ t,
k
where k = α/λ, using direct calculations, we can verify the estimate
 
−1 1
Fu (0, λ)L (X→X) = O 2 .
λ
Condition b) implies the estimate F (0, λ) = O(λ4 ). If we rewrite Eq. (3.15) as (1.1), where now
d2 u
B(λ)u = λ2 − α2 u, R(u, λ) = −f (u, λ) + α2 u,
dt2
then, in view of estimates a), b), c), all the assumptions of Theorem 2 become obvious. Therefore, the
sequence {un }, where un is the solution of the linear boundary-value problem
λ2 ün − α2 un = f (un−1 , λ) − α2 un−1 , (3.16)
un (0) = un (1) = 0, u0 = 0, (3.17)
converges to the solution of the boundary-value problem (3.13), (3.14).

In the numerical simulation of the singular boundary-value problems (3.16), (3.17), the process can
be regularized by using Shishkin’s adaptive grids [14].
In conclusion, note that, in the problem of the bifurcation point [2, pp. 5–50], the minimal branch is
the trivial solution. In that problem, iterative methods from [5]–[7] can be used to construct the nontrivial
branch.

ACKNOWLEDGMENTS

The author A. I. Dreglya wishes to express gratitude to Professor Daphne Gilbert for her attention
and support during the author’s work in Dublin Institute of Technology.
This work was supported by the Federal Program “Scientific and Pedagogical Personnel of Russian
Innovation for the years 2009–2013” (grant no. P696). The work of the third author was supported by
Dublin Institute of Technology (Ireland) and by Klüber Lubrication KG (Germany).

MATHEMATICAL NOTES Vol. 91 No. 1 2012

 104 SIDOROV et al.

REFERENCES
1. M. M. Vainberg and V. A. Trenogin, The Theory of Branching of Solutions of Nonlinear Equations
(Nauka, Moscow, 1969; Noordhoff, Leyden, 1974).
2. Nonlinear Analysis and Nonlinear Differential Equations, Ed. by V. A. Trenogin and A. F. Filippov
(Fizmatlit, Moscow, 2003) [in Russian].
3. N. Sidorov, B. Loginov, A. Sinitsyn, and M. Falaleev, Lyapunov–Schmidt Methods in Nonlinear Analysis
and Applications, in Math. Appl. (Kluwer Acad. Publ., Dordrecht, 2002), Vol. 550 [in Russian].
4. N. A. Sidorov and A. V. Sinitsin, “Analysis of bifurcation points and nontrivial branches of solutions to the
stationary Vlasov-Maxwell system,” Mat. Zametki 62 (2), 268–292 (1997) [Math. Notes 62 (2), 223–243
(1997)].
5. N. A. Sidorov, “Explicit and implicit parametrizations in the construction of branching solutions by iterative
methods,” Mat. Sb. 186 (2), 129–141 (1995) [Russian Acad. Sci. Sb. Math. 186 (2), 297–310 (1995)].
6. B. V. Loginov and N. A. Sidorov, “Group symmetry of the Lyapunov-Schmidt branching equation and
iterative methods in the problem of a bifurcation point,” Mat. Sb. 182 (5), 681–691 (1991) [Math. USSR-Sb.
73 (1), 67–77 (1991)].
7. N. A. Sidorov and V. R. Abdullin, “Interlaced branching equations in the theory of nonlinear equations,” Mat.
Sb. 192 (7), 107–124 (2001) [Russian Acad. Sci. Sb. Math. 192 (7), 1035–1052 (2001)].
8. V. A. Trenogin, Functional Analysis (Fizmatlit, Moscow, 2002) [in Russian].
9. N. A. Sidorov, “Minimal branches of solutions of nonlinear equations and asymptotic regularizers,” in
Nonlinear Boundary-Value Problems (Institute of Applied Mathematics and Mechanics, Donetsk, 2004),
Vol. 14, pp. 161–164 [in Russian].
10. A. N. Tikhonov, V. K. Ivanov, and M. M. Lavrent’ev, “Ill-posed problems,” in Partial Differential Equations,
Proceedings of the Symposium Dedicated to the 60th Anniversary of Academician S. L. Sobolev (Nauka,
Moscow, 1970), pp. 224–238 [in Russian].
11. V. A. Trenogin and N. A. Sidorov, “Regularization of simple solutions of nonlinear equations in the neigh-
borhood of a branching point,” Sibirsk. Mat. Zh. 19 (1), 180–185 (1978) [Siberian Math. J. 19, 128–132
(1978)].
12. V. F. Chistyakov, Algebro-Differential Operators with Finite-Dimensional Kernel (Nauka, Novosibirsk,
1996) [in Russian].
13. A. I. Dreglya, “Some analytical and exact solutions of systems of equations in the theory of modeling
polymers,” Sib. Zh. Ind. Mat. 11 (3), 61–70 (2008).
14. A. I. Dreglea [Dreglya] and G. I. Shishkin, “Robust numerical method for a singularly perturbed equation
with unboundedly growing convective term at infinity,” in Proceedings of the International Conference on
Computational Mathematics, ICCM-2004 (Novosibirsk, 2004), pp. 835–838 [in Russian].

MATHEMATICAL NOTES Vol. 91 No. 1 2012

View publication stats