Nonlinear Mathematical Physics 1996, V.3, N 1–2, 40–50.

On Asymptotic Methods in the Theory

of Differential Equations of Mathematical
Physics
Mykola SHKIL’
Drahomanov Pedagogical University,
9 Pyrogov Street, Kyı̈v, Ukraı̈na

In 1807 Fourier suggested a original method of solving partial differential equations. The
method is known to lead to ordinary differential equations containing some arbitrary
parameter. Making use of the method, two main problems appear:
1. Finding the solutions (eigenfunctions) of the differential equations obtained.
2. Expansion of a function into the series in terms of eigenfunctions. These two
problems in the particular cases were already solved by Fourier.
A great progress in this trend was made by Liouville’s works (1838), where an expansion
of an arbitrary function into the series in terms of eigenfunctions of the equation
d2 y
+ (λg(x) − r(x))y = 0, (1)
dx2
(λ is a parameter) with certain boundary conditions was considered.
Eigenfunctions obtained by Liouville for Eq.(1) have the property of orthogonality. So,
he solved the problems 1 and 2 directly using some Sturm’s results.
After appearing works of Fourier, Sturm and Liouville, the theory of representation of
the solutions in the form of asymptotic formulae in the parameter λ, when λ → ∞, or,
that is the same, the asymptotic integration theory, became to develop very quickly.
An important place in the theory is taken by the Poincaré work ”New methods in
celestial mechanics” (1888), where the practical application of formal series utilized in the
asymptotic representation of solutions to differential equations was first demonstrated.
Let us illustrate that by means of an example of the second-order differential equation
introduced in [1].
Let us consider the differential equation
d2 y a
 
2
+ 1 − 2 y = 0, (2)
dt t
where a is a complex number, t > 0 is a real variable.
Eq.(2) contains a varying coefficient and so it can not be integrated. Its solution will
be sought in the form of a series
 
y(t) = eit C0 + C−1 t−1 + · · · + C−n t−n + · · · . (3)

Copyright
c 1996 by Mathematical Ukraina Publisher.
All rights of reproduction in any form reserved.
ON ASYMPTOTIC METHODS IN THE THEORY OF DIFFERENTIAL EQUATIONS 41

Then by a certain choice of the series coefficients in (3), the function y(t) can formally
satisfy Eq.(2) in the sense that when substituting it in Eq.(2), we obtain the identity, in
which coefficients of t−m (m = 0, 1, 2, . . .) are equal to zero. But series (3) is a divergent
one. And simultaneously, as Poincaré proved, for large values of t (t → ∞) an exact
solution to Eq.(2) is represented in the form of the asymptotic formula
  
y(t) = eit Sn (t) + O t−n−1 , (4)

where Sn (t) is the n-th partial sum of series (3).

Formal solutions in the form (3) which have the property (4) is said to be asymptotic
solutions of the differential equations under consideration and the methods by means of
which one can obtain solutions is said to be asymptotic ones.
Note that in Eq.(2) the parameter is missing and asymptotic formula (4) is constructed
in the independent variable t (t → ∞). As to the previous investigations, their methods
make possible to built asymptotic solutions in the parameter λ (λ → ∞) for all x ∈ [a, b],
although it is naturally to ask a question on building asymptotic solutions with two vari-
ables: the independent variable and the parameter. Such investigations have been carried
out by mathematicians. For instance, original results were obtained by M.V. Fedoryuk
and elucidated in his doctor theses [2]. But we do not concern these investigations here.
The main attention is given to the methods of constructing asymptotic solutions for both
large and small parameters.
A great contribution to developing asymptotic methods has been made by the Russian
mathematician V.A. Steklov. Proving the ”completeness” theorem for eigenfunctions of
the Sturm–Liouville equation (Eq.(1)), the scientist solved two problems 1) and 2) on
expansion into a series in terms of eigenfunctions of the equation with the same degree of
accuracy as for ordinary trigonometric Fourier series.
But all investigations mentioned concern mostly a generalization of Liouville’s results,
and, therefore, yield selfadjoint differential equations and systems of them. Such quite es-
sential restrictions were removed in the works of L. Schlesinger, G. Birkhoff, Ya. Tamarkin.
Thus, Birkhoff derived asymptotic formulae for solving a differential equation of the type
y (n) + ρan−1 (x, ρ)y (n−1) + · · · + ρn a0 (x, ρ)y = 0, (5)
where ai (x, ρ) (i = 0, 1, . . . , n − 1) are analytical functions of the complex parameter ρ
(|ρ| → ∞), which are infinitely differentiable with respect to the real variable x ∈ [a, b].
As distinct from Schlesinger [3] who has proved the asymptotic properties of solutions
only for some fixed ray arg ρ = α with |ρ| → ∞, Bikhoff [4] has proved the asymptotic
properties of solutions for some sector of the complex plane θ ≤ arg ρ ≤ η.
Schlesinger’s and Birkhoff’s results have been generalized by Tamarkin [5] for the
systems of differential equations of the form
n
dyi X
= aij (x, ρ)yj , (6)
dx j=1

where aij (x, ρ), i, j = 1, n are analytical functions of the complex parameter ρ in a
neighborhood of the point ρ = ∞, which have a singularity (the pole of order r ≥ 1)
at this point and are infinitely differentiable with respect to the real variable x on the
segment [a, b].
42 M. SHKIL’

Ya.D. Tamarkin [5] has obtained asymptotic formulae for the solutions of system (6)
with the parameter ρ in the case, where the roots of the characteristic equation
(0)
det (aij (x) − λδij ) = 0, i, j = 1, n (7)

(0)
are simple for all x ∈ [a, b]. Here aij (x) are free terms in the expansions in the parameter
ρ; δij are Kronecker symbols.
Only for the second-order equation

y 00 + ρa1 (x, ρ)y 0 + ρ2 a2 (x, ρ)y = 0, (8)

he has succeeded in obtaining two formal solutions, when characteristic equation (7) has
two equal roots on the segment [a, b]. For differential equations of higher order and also for
differential systems of type the (6), the case with multiple roots of the characteristic equa-
tion has remained not investigated for a long time, and only in the 60–70s in M.I. Shkil’s
works (who is the author of this review) and, in particular, in his doctor theses [6], the
cases with multiple roots of the characteristic equation have been studied in detail. These
results and also those obtained by followers of M.I. Shkil are discussed below.
Note that to represent formal solutions of the systems of differential equations contain-
ing a large parameter λ by analogy with the well-known Euler method of solving a linear
system of differential equations with constant coefficients, one makes use of the expression
Zx r−1
 
∞
−s
ρ−s ys (x),
X X
exp ρr ρ λs (τ )dτ  (9)
a s=o s=0

from which one obtains asymptotic formulae for the solution of the form
Zx r−1
 
m
!
−s −s −m
X X
r
y(x, ρ) = exp ρ ρ λs (τ )dτ  ρ ys (x) + O(ρ ) . (10)
a s=o s=0

Asymptotic representation of the solutions in the more general form

 
n m
yij (x)ρ−j + O(ρ−m ) ,
X X
yi (x, ρ) = zk (x, ρ)  (11)
k=1 j=1

where zk (x, ρ) (k = 1, n) are functions satisfying some system of linear differential equa-
tions, is adduced in the works of V.S. Pugachov [7].
Pugachov’s results have been used by I.M. Rapoport to build the linear transformati-
ons, by means of which one can reduce some class of differential equations to a differential
system of the L-diagonal type. For solutions of these systems, the asymptotic formulae
with respect to the independent variable t in the neighborhood of the point t = +∞
were obtained (1954). Pugachov’s ideas have been making use of also by S.F. Feshchenko
[8] in asymptotic splitting the system of linear differential equations with slowly varying
coefficients.
The works of F. Hukuhara (1934–1936) and G. Territin (1957) are devoted to asymp-
totic representations of solutions, where the asymptotic splitting of the systems of linear
ON ASYMPTOTIC METHODS IN THE THEORY OF DIFFERENTIAL EQUATIONS 43

differential equations with a large parameter into subsystems of smaller order is adduced
[9]. Essential results of asymptotic integration of differential equations have been obtained
in works of V. Vazov (1968), L. Chezary (1964), R. Langer (1931), where, in particular,
an extraordinary complicated case related to furning points is investigated [10–12].
Our short review on asymptotic representation of solutions of differential equations con-
taining a parameter would not be completed enough if we omit the Krylov–Bogolyubov–
Mitropolskii methods in nonlinear mechanics [13]. The methods were developed at first by
N. Krylov and M. Bogolyubov (1937) for approximate integrating the nonlinear differential
equations of the form
d2 x dx
 
2
+ ω x = εf x, , ε, t , (110 )
dt2 dt
where ε is a small parameter.
Note that equation (110 ) is of a quite different form than the equations considered
previously. Really when ε = 0, (110 ) becomes a second-order linear differential equation
with the constant coefficient ω 2 , which is elementary integrable. So, the Poincaré method
[14] can be used for equation (11), by which the solution is found in the form of series
in powers of the small parameter ε. But in this way, approximate solutions are obtained,
which have so–called secular terms (terms in which the independent variable is in front of
sine and cosine). This fact constricts considerably the temporal interval t during which
one can observe an ossillatory process.
Approximate formulae obtained by the Krylov–Bogolyubov method do not contain
secular terms and it makes possible to investigate the ossillatory process on the finite but
large enough interval of varying the time t (of order O( 1ε )).
Making use of the Krylov–Bogolyubov method, Yu.A. Mitropolskii studied a nonsta-
tionary ossillatory processes in systems with one and a few degrees of freedom. He first
investigated the rather complicated phenomenon of passing the system through inter-
nal and external resonances and published his fundamental results in the monograph [15].
Mitropolskii’s results have been generalized by his disciples, in particular, A.M. Samoilenko,
to other classes of differential equations.
Under influence of the Krylov–Bogolyubov–Mytropolskii ideas, the theory of asymp-
totic representation of solutions to linear differential equations with slowly varying co-
efficients became to intensively develop. And the subject of our investigations is the
differential equations with slowly varying coefficients and asymptotic methods of their
solution.
Function f (t) is said to be a slowly varying one on the segment (a, b) if its derivative
0
f (t) is commensurable with some small parameter ε ∈ (0, ε0 ) (ε0 is a real small enough
number, ε0  1).
It is clear that any function differentiated on the segment (a, b) is a slowly varying one
of t if it depends on variable τ = εt. Really, in this case we have
df (τ ) df (τ )
=ε . (12)
dt dτ
The variable τ is said to be a ”slow time” as well.
Differential equations in which the coefficients are slowly varying functions are said to
be those with slowly varying coefficients. Many differential equations of different types,
in particular, the equations considered by Liouville, Schlesinger, Birkhoff and Tamarkin,
44 M. SHKIL’

can be reduced to such differential equations. We show this by the example of the Sturm–
Liouville equation (1). For this purpose we put

1
x = τ = εt, ε= √ . (13)
λ

Then we obtain the equation

d2 y
+ (g(τ ) − ε2 r(τ ))y = 0 (14)
dt2
with slowly varying coefficients.
Many mathematical works are devoted to investigating differential equations, in which
the coefficient of the higher derivative term is a small parameter. These equations can be
also reduced to differential equations with slowly varying coefficients. Let us consider, for
example, the differential equation

d2 y dy
ε + p(x) + g(x)y = 0, (15)
dx2 dx
where ε > 0 is a small parameter; p(x), g(x) are functions of the variable x ∈ [a, b].
The equation is not integrable exactly. But if we put ε = 0, then we have a so-called
singular equation that is integrable in quadratures.
Eq.(15) is said to be nonsingular. It is natural to ask a question: how are solutions of
the nonsingular equation with small values of the parameter ε connected with solutions
of a singular one, in particular, does the solution yε of Eq.(15) with ε → 0 approach the
solution y0 of the singular one? Such problems were investigated by a great number of
mathematicians. Fundamental results were first obtained for linear differential equations
by I.S. Gradshtein (1950) and for nonlinear ones by A.M. Tikhonov (1948). They deter-
mined that yε → y0 not always with ε → 0. For this statement to be true, it is necessary to
impose some constraints on coefficients of the nonsingular equation and on the initial con-
ditions. If the constraints are not fulfilled then it is necessary to integrate the nonsingular
equation at least approximately.
Let us show that Eq.(15) can be reduced to an equation with slowly varying coefficients.
For this matter it is sufficient to put

x = τ = εt

and we arrive at the equation

d2 y dy
2
+ p(τ ) + εg(τ )y = 0 (16)
dt dt
with slowly varying coefficients.
We give our attention only to two types of differential equations which are reduced to
differential equations with slowly varying coefficients by means of changing the indepen-
dent variable. It is possible to point out a number of types (see, e.g., [16]) of differential
equations which can be reduced to those with slowly varying coefficients.
ON ASYMPTOTIC METHODS IN THE THEORY OF DIFFERENTIAL EQUATIONS 45

The systematic investigation of linear differential equations with slowly varying coef-
ficients has begun from 50s by works of Feshchenko S.F. [17]. He proposed an asymptotic
method of representating solutions to the differential equation of the type

d2 y dy
2
+ εp(τ, ε) + q(τ, ε)y = εf (τ, ε)eiθ(t,ε) , (17)
dt dt
dθ(t, ε)
where p(τ, ε), q(τ, ε), f (τ, ε), k(τ ) = dt are slowly varying functions of the variable
τ = εt; ε > 0 is a small parameter.
There was considered the case where the function k(τ ) (an external part) for some
τ ∈ [0, L] becomes such that is equal to one of simple roots (fundamental frequencies) of
the characteristic equation corresponding to Eq.(17). This case often occurs in problems of
mathematical physics. In mechanics, it is called a resonance. Feshchenko proved an asymp-
totic character of the solutions obtained in the same sense as in the Krylov–Bogolyubov–
Mitropolskii methods, namely, the m–partial sums of these solutions approach the exact
solution not when m increases, but when m ≥ 1 is fixed and ε → 0.
Feshchenko carried out analogous investigations for a system of differential equations
of the type
d2 x dx
A(τ, ε) 2 + εC(τ, ε) + B(τ, ε)x = f (τ, ε)eiθ(t,ε) , (18)
dt dt
where matrices A(τ, ε), C(τ, ε), B(τ, ε) and vector f (τ, ε) are represented by power series
in the small parameter ε. Here it was supposed that matrices A0 (τ ), B0 (τ ), C0 (τ ) (free
terms in general expansions) are symmetric. This strong enough requirement has been
removed later in the works of Shkil and his disciples.
Feshchenko has proved a number of extraordinary important theorems on splitting a
system of differential equations of the type

dx
= A(τ, ε)x (19)
dt
into subsystems of a lower order.
Feshchenko’ theorems include, as a particular case, the results of Schlesinger, Birkhoff
and Tamarkin on asymptotic representation of solutions to linear differential systems in
the case, where roots of the characteristic equation are simple.
But with the help of the asymptotic splitting theorems one can reduce the order of
a system only approximately. In the general case, for example, where some roots of the
characteristic equation are multiple, while using these theorems, we can’t obtain solutions
of the system of differential equations in question. But this case occurs often enough
both in investigating theoretical questions and solving practical problems. So, already in
studying a simple question, that is the Sturm–Liouville equation, one has to deal with
multiple roots. They occur in the investigation of differential equations with a small pa-
rameter in the part of derivative terms, in optimal control problems and so on. Let us
note that the case of multiple roots, especially where multiple roots correspond to multiple
elementary divisors, is extraordinary complicate. It is explained by that, generally speak-
ing, the initial system of differential equations has no solutions admitting an expansion
in integer powers of the parameter ε. Such solutions, as opposed to the case of somple
roots of the characteristic equation, are represented by formal series in different fractional
46 M. SHKIL’

powers of the parameter with exponents depending not only on multiplicity of the roots
of the characteristic equation but also on miltiplicity of elementary divisors corresponding
to them.
The case of multiple roots of the characteristic equation has been studied comprehen-
sively by M.I. Shkil. In 60s he obtained essentially new mathematical results on asymp-
totic representation of solutions to systems of differential equations with slowly varying
coefficients. We focus our attention to his some results.
Let the characteristic equation

det (A0 (τ ) − λE) = 0 (20)

(E is an identity matrix) have at least one root λ = λ0 of constant multiplicity k (2 ≤

k < n), corresponding to one elementary divisor with the same multiplicity. Then the
following theorem is true.

Theorem 1 If A(τ, ε) has on the segment [0, L] all derivatives with respect to τ and the
matrix
dT (τ )
 
−1
C(τ ) = T (τ ) − A1 (τ )T (τ ) , (21)
dτ
where T (τ ) is the matrix that reduces A0 (τ ) to the Jordan form, is such that its element

Ck,1 (τ ) 6= 0, ∀τ ∈ [0, L], (22)

then the system of differential equations (19) has a formal solution of the form
 t 
Z
x(t, ε) = U (τ, µ) exp  λ(τ, µ)dt , (23)
0

where the n-dimensional vector U (τ, µ) and the scalar function λ(τ, µ) have the expansion
∞
X ∞
X
U (τ, µ) = µs Us (τ ), λ(τ, µ) = λ0 (τ ) + µs λs (τ ), µ = ε1/k . (24)
s=0 s=1

The case, where Ck,1 = 0, has been also studied but

Ck−1,1 (τ ) + Ck,2 6= 0, ∀τ ∈ [0; L]. (25)

Then system (19) has a formal solution of the type (23) where U (τ, µ), λ(τ, µ) are
represented by formal series with µ = ε1/(1−k) , and one formal solution is expanded in
integer powers of the parameter ε.
Let us adduce the more general result obtained by Shkil for system (19) in the case of
multiple roots of the characteristic equation.
Let the following conditions be valid:
1) matrix A(τ, ε) has all derivatives with respect to τ on the segment [0, L];
2) characteristic equation (20) has one root λ = λ0 (τ ) of a constant multiplicity n;
3) the root λ0 (τ ) corresponds to r ≥ 1 elementary divisors of the type

(λ − λ0 (τ ))k1 , . . . , (λ − λ0 (τ ))kr ; (26)

ON ASYMPTOTIC METHODS IN THE THEORY OF DIFFERENTIAL EQUATIONS 47

4) one of the formulae

a) k1 = k2 = . . . = kr = k,
(27)
b) k1 > k2 > . . . > kr .
is satisfied. Then the following theorem is valid.

Theorem 2 Let the conditions 1–4a be satisfied. Then for the vector
 t 
Z
x = U (τ, µ) exp  λ(τ, µ)dt , (28)
0

where the n–dimensional vector U (τ, µ) and the scalar function λ(τ, µ) are represented by
the formal series
∞
X ∞
X
U (τ, µ) = µs Us (τ ), λ(τ, µ) = µs λs (τ ), (29)
s=0 s=0

1
with µ = ε k , to be a formal vector–solution of system (19) it is necessary and sufficient
that the function (λ1 (τ ))k for all τ ∈ [0, L] be a root of the equation


ρ + Ck,1 (τ )Ck,k+1 (τ ) . . . Ck,lr−1 +1 (τ )

C2k,1 (τ )ρ + C2k,k+1 (τ ) . . . C2k,lr−1 +1 (τ )
det = 0, (30)

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ··

Cn,1 (τ )Cn,k+1 (τ ) . . . ρ + Cn,lr−1 +1 (τ )

where Ck,1 (τ ),. . .,Cn,lr−1 +1 (τ ); lr−1 = (r − 1)k are elements of matrix (21).

The proof of the sufficient condition of the theorem gives also an algorithm of obtaining
the solution to be found.
An analogous theorem is valid for the case 4b. For the both cases, formal solutions have
been proved to be asymptotical expansions in the parameter ε of the exact solutions of
system (19). These results have been generalized on the case, where multiple roots of the
characteristic equation correspond to the both multiple and simple elementary divisors.
The mathematical results obtained by Feshchenko and Shkil were published as the
common monograph, with L.D. Nikolenko being its co-author [17].
The asymptotic expansion theory developed for finite–dimensional spaces has been
generalized for infinite-dimensional spaces by Yu.L. Daletsky and S.G. Krein [18].
It is worth to recall the classical works of Kh.L. Territin [19], who built the linear trans-
formations, with the help of which the case of multiple roots of the characteristic equation
can be reduced to that of simple roots, and the investigations of V.Vasov, E. Koddington
and N. Levinson that simplified maximally the problems with multiple roots as well.
Essentially new methods of asymptotic integration of differential equations with a
small parameter in the higher derivative term have been also developed by M.I. Vyshyk
and L.A. Lyusternik [20], A.N. Tikhonov, A.B. Vasyl’eva. In particular, Vyshyk and
Lyusternik established that the solutions of linear, singularly disturbed problems contain
functions of the so-called boundary-layer type and worked out the algorithm of building
asymptotic solutions. Developing their ideas, Vasyl’eva worked out the method of building
48 M. SHKIL’

asymptotic solutions of singularly disturbed problems which is especially effective concern-

ing nonlinear equations [22].
A separate method of the asymptotic integrating of singularly disturbed equations on
the finite interval of varying the independent variable was developed by S.A. Lomov in
70s, who proposed the so-called regularization method [23]. Its essence is to replace the
singular problem by a regular one via transition to the space of larger dimension. Making
use of the method, Lomov and his disciples obtained a set of new results on the asymptotic
integration of linear systems.
Concentrating on no details, note that in the last years are intensively developed
asymptotic methods of integraton of linear differential systems of the type

dx
εB(t, ε) = A(t, ε)x. (31)
dt

Some results of these investigations and also ones related to the presence of turning
points in the system are adduced in the monograph of M.I. Shkil, I.I. Starun, V.P. Yakovets
[24].
It should be noted that the mentioned above results concerning the asymptotic inte-
gration of systems of differential equations have been obtained under the certain rather
essential condition: roots of the characteristic equation have either to be simple or keep a
constant multiplicity on the given interval of the independent variable. If these conditions
are violated (turning points appear) then the results obtained before become incorrect. A
construction of even formal solutions of the differential systems in question is considerably
complicated. More or less complete results under appearance of turning points have been
obtained only for one second-order differential equation and systems of two first-order
differential equations.
M.I. Shkil has found sufficient conditions of existing formal solutions to systems of
differential equations when turning points are presented. But in contract to the previous
results, where formal series are pure expansions in some powers of the parameter ε, when
turning points are present, coefficients of the formal series depend in turn on the parameter.
Let us formulate M.I. Shkil’s result about the system of first-order differential equations
of the type
dx
ε1/r = A(t, ε)x, (32)
dt
where A(t, ε) is a matrix of (n × n)-order, r ≥ 1 is a positive integer.

Theorem 3 If the following conditions are true:

1) A(t, ε) has the expansion
∞
X
A(t, ε) = εs As (t);
s=0

2) matrices As (t) are infinitely differentiable on the segment (−∞, +∞);

3) there exists an integer k ≥ 0 such that roots of the equation

det (A0 (t) + εA1 (t) + · · · + εk Ak (t) − λE) = 0 (33)

ON ASYMPTOTIC METHODS IN THE THEORY OF DIFFERENTIAL EQUATIONS 49

are simple for ∀t ∈ (−∞, +∞) and ε ∈ (0, ε0 ];

then there exists a formal vector-solution of system (32) of the form

x(t, ε, h) = U (t, ε, h)ψ(t, ε, h), (34)

where U (t, ε, h) is an (n × n)-matrix and ψ(t, ε, h) is an n-dimensional vector determed by

the system of differential equations
dψ(t, ε, h)
h = Λ(t, ε, h)ψ(t, ε, h), (35)
dt
in which Λ(t, ε, h) is a diagonal (n×n)-matrix represented similarly to the matrix U (t, ε, h)
by the formal expansions:
∞
hs Us (t, ε),
P
U (t, ε, h) =
s=0
∞
(36)
hs Λs (t, ε), h = ε1/r .
P
Λ(t, ε, h) =
s=0

References
[1] Koddigton E. A., Levinson N., Theory of Ordinary differential equations, 1958, Moscow, Foreign lit.
printing, 474p. (in Russian)

[2] Fedoryuk M. V., Asymptotic methods in the theory of linear ordinary differential equations, Moscow,
1966, 24p.

[3] Schlesinger L., Vorlesungen über Differntialgleichungen, Tenbner. Leipzig, Berlin, 1908, V.12, S.46.

[4] Birkhoff G.D., On the asymptotic character of the solutions of certain linear differential equations
containing a parameter, Trans. Amer. Math. Soc., 1908, V.9, 219–231.

[5] Tamarkin Yu. D., On some general questions of the theory of differential equations and on expansion
of a function in series, 1917, 308p.

[6] Shkil M. I., On some asymptotic methods in theory of differential equations with slowly varying
coefficients: doctor dissertation, Kyiv, 1968, 420p. (in Russian)

[7] Pugachov V. S., On asymptotic representations of the integral of a linear system of differential
equations containing a parameter, Mat. sborn. N.S., 1944, V.15, N 1, 13–46.

[8] Feshchenko S. F., On asymptotic splitting system of linear differential equations, Ukr. Math. J., 1955,
V.7, N 2, 167–179; N 4, 252–263.

[9] Hukuhara, Sur les pants singuliers des equation differentielles linealres, T. Fac. Scl., Univ. Hokkaido
Math, 1934–1936, V.2, 13–81.

[10] Vasov V., Asymptotic expansions of solutions to ordinary differential equations, Mir, Moscow, 1968,
464p.

[11] Chesary A., Asymptotic behavior and stability of solutions, Mir, Moscow, 1964, 477p.

[12] Langer R. E., Turning points in the linear asymptotic theory, Bol. Soc. Math. Mech., 1960, V.2,
N 5, 1–12.

[13] Bogolyubov N. N., On some statistical methods in mathematical physics, Kyiv, A. S. Ukraine prin-
ting, 1945, 137p.
50 M. SHKIL’

[14] Poincaré H., Sur les equations de la physique et mathematique, Rend. Pal., 1894, 10–11, 57–156.

[15] Mitropolskii Yu. A., Problems of asymptotic theory of nonstable fluctuations, Nauka, Moscow, 1964,
432p.

[16] Shkil M. I., Asymptotic methods in differential equations, Vyshcha shkola, Kyiv, 1971, 228p.

[17] Feshchenko S. F., Shkil M. I., Nikolenko L. D., Asymptotic methods in the theory of linear differential
equations, American Elsvier Publ. Comp., New York, 1967, 270p.

[18] Daletski Yu.L. and Krein S.G., On differential equations in Hilbert space, Ukr. Math. J., 1950, V.2,
N 4, 71–91.

[19] Territin Kh. L., Asymptotic expansion of solutions to system of ordinary differential equations con-
taining parameter, Matematika: Sb. per., 1957, V.1, N 2, 29–59.

[20] Vyshchyk M. I., Lyusternik L. A., Regular singularization and a boundary layer for linear differential
equations with a small parameter, Uspekhi mat. nauk, 1957, V.12, N 15, 3–122.

[21] Tikhonov A. I., On dependence of solutions of differential equations on small parameter, Mat. sb.,
1948, V.28, N 2, 193–204.

[22] Vasyl’eva A. B., On differential equations containing small parameters in derivatives terms, Mat, sb.,
1951, V.31, N 3, 587–644.

[23] Lomov S. A., Introduction to the general theory of singular disturbance, Nauka, Moscow, 1981, 398p.

[24] Shkil M. I., Starun I. I., Yakovets V. P., Asymptotic integrating linear systems of differential equations
with singularity, Vyshcha shkola, Kyiv, 1991, p.207.