Schemes For Partial Differential Equations

Operator Theory: Advances and
Applications
Vol. 148
Editor:
I. Gohberg
Editorial Office:
School of Mathematical
Sciences
Tel Aviv University
Ramat Aviv, Israel
Editorial Board:
D. Alpay (Beer-Sheva)
J. Arazy (Haifa)
A. Atzmon (Tel Aviv)
J. A. Ball (Blacksburg)
A. Ben-Artzi (Tel Aviv)
H. Bercovici (Bloomington)
A. Bottcher (Chemnitz)
K. Clancey (Athens, USA)
L. A. Coburn (Buffalo)
K. R. Davidson (Waterloo, Ontario)
R. G. Douglas (College Station)
A. Dijksma (Groningen)
H. Dym (Rehovot)
P.A.Fuhrmann(BeerShev~
S. Goldberg (College Park)

B. Gramsch (Mainz)
G. Heinig (Chemnitz)
J. A. Helton (La Jolla)
M. A. Kaashoek (Amsterdam)
H. G. Kaper (Argonne)
S. T. Kuroda (Tokyo)
P. Lancaster (Calgary)
L. E. Lerer (Haifa)
B. Mityagin (Columbus)
V. V. Peller (Manhattan, Kansas)
L. Rodman (Williamsburg)
J. Rovnyak (Charlottesville)
D. E. Sarason (Berkeley)
I. M. Spitkovsky (Williamsburg)
S. Treil (Providence)
H. Upmeier (Marburg)
S. M. Verduyn Lunel (Leiden)
D. Voiculescu (Berkeley)
H. Widom (Santa Cruz)
D. Xia (Nashville)
D. Yafaev (Rennes)
Honorary and Advisory
Editorial Board:
C. Foias (Bloomington)
P. R. Halmos (Santa Clara)
T. Kailath (Stanford)
P. D. Lax (New York)
M. S. Livsic (Beer Sheva)
New Difference
Schemes for Partial
Differential Equations
Allaberen Ashyralyev
Pavel E. Sobolevskii
Springer Basel AG
Authors:
Aliaberen Ashyralyev
Pavel E. Sobolevskii
Department of Mathematics
Fatih University
Buyukcekmece
34900 Istanbul, Turkey
[email protected]
Hebrew University of Jerusalem

Givat Ram Campus
91904 Jerusalem
Israel
[email protected]
and
International Turkrnen-Turkish University
32 Gerogly Slreet
74400 Ashgabat, Turkrnenistan
2000 Mathematics Subject Classification 39A12, 47B39, 65M06, 65N06, 65Q05
A CIP catalogue record for this book is available from the

Library ofCongress, Washington D.C., USA
Bibliographic information published by Die Deutsche Bibliothek
Die Deutsche Bibliothek Iists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available
in the Internet at <http://dnb.ddb.de>.
ISBN 978-3-0348-9622-1
ISBN 978-3-0348-7922-4 (eBook)
DOI 10.1006/978-3-0348-7922-4
This work is subject to copyright. AII rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms
or in other ways, and storage in data banks. For any kind of use permission of the copyright owner must be obtained.
2004 Springer Basel AG

Originally published by Birkhuser Verlag in 2004
Softcover reprint of the hardcover 1st edition 2004
Printed on acid-free paper produced from chlorine-free pulp. TCF 00
Cover design : Heinz Hiltbrunner, Basel
ISBN 3-7643-7054-8
987654321
www.birkhauser-science.com
Contents
Preface
1 Linear Difference Equations

1.1 Difference Equations of the First Order . . . . .
1.2 Difference Equations of the Second Order . . . .
1.3 Difference Equations with Constant Coefficients .
Vll
1
3
5
2 Difference Schemes for First-Order Differential Equations

2.1 Single-Step Exact Difference Scheme and Its Applications
2.2 Taylor's Decomposition on Two Points and Its Applications
11
31
3 Difference Schemes for Second-Order Differential Equations

3.1 Two-Step Exact Difference Scheme and Its Applications
3.2 Taylor's Decomposition on Three Points and Its Applications
37
77
4 Partial Differential Equations of Parabolic Type

4.1 A Cauchy Problem. Well-posedness . . . . . . . . . . . . . . . .
99
4.2 Difference Schemes Generated by an Exact Difference Scheme
146
4.3 Single-Step Difference Schemes Generated
by Taylor's Decomposition . . . . . . . . . . . . . . . . . . . . . . . 188
5 Partial Differential Equations of Elliptic Type
5.1 A Boundary-Value Problem. Well-posedness . . . . . . . . . . .
197
5.2 Difference Schemes Generated by an Exact Difference Scheme.
229
5.3 Two-Step Difference Schemes Generated
by Taylor's Decomposition. . . . . . . . . . . . . . . . . . . . . . . 291
6 Partial Differential Equations of Hyperbolic Type
6.1 A Cauchy Problem. . . . . . . . . . . . . . . . . . . . . . . .
6.3 Two-Step Difference Schemes Generated
by Taylor's Decomposition. . . . . . . . . . . . . . . . . . . .
313
317
332
Contents
vi
7 Uniform Difference Schemes for Perturbation Problems
7.1 A Cauchy Problem for Parabolic Equations . . . .

7.2 A Boundary-Value Problem for Elliptic Equations
7.3 A Cauchy Problem for Hyperbolic Equations
343
364
387
8 Appendix: Delay Parabolic Differential Equations

8.1 The Initial-Value Differential Problem
8.2 The Difference Schemes
393
400
Comments on the Literature
411
Bibliography . . . . . . . . .
423
Preface
The present monograph is devoted to the construction and investigation of the
new high order of accuracy difference schemes for approximating the solutions of
regular and singular perturbation boundary-value problems for partial differential
equations. The construction is based on an exact difference scheme and Taylor's
decomposition on two or three points. This approach permits us to extend essentially a class of problems where the theory of difference methods is applicable.
Namely, now it is possible to investigate differential equations with variable coefficients and regular and singular perturbation boundary-value problems. The
investigation is based on new coercivity inequalities. The stability and coercive
stability estimates in various Banach norms for solutions of the high order of accuracy difference schemes of the initial-value problem for parabolic equations and
boundary-value problem for elliptic equations are obtained. This investigation is
based on the theory of positive operators, theory of semigroup operators and theory of interpolation of linear operators. The stability estimates for the solutions
of the high order of accuracy difference schemes of the initial-value problems for
hyperbolic equations are established. This investigation is based on the spectral
theory of self-adjoint positive definite operators in a Hilbert space. This approach
permits us to study the stability of the simple difference schemes for various partial differential equations. The simple first and second order of accuracy difference
schemes for delay partial differential equations are considered. The stability estimates for the solutions of these difference schemes of initial-value problems for
delay partial differential equations are established.
Let us give a brief description of the contents of our monograph. It consists
of eight chapters. In Chapter 1 we study first- and second-order linear difference
equations. These equations arise in numerical analysis of the approximate solutions
of boundary-value problems for ordinary and partial differential equations.
Chapter 2 presents the high order of accuracy single-step difference schemes
generated by an exact difference scheme or by Taylor's decomposition on two
points for the numerical solution of an initial-value problem for first-order differential equations and for a system of first-order differential equations. In general,
these difference schemes for the same problem are different. However, in the case of
a homogeneous equation or a system of differential equations with independent in
t coefficients, by using both methods we have exactly the same difference schemes.
Chapter 3 is devoted to study of the high order of accuracy two-step difference
schemes generated by an exact difference scheme or by Taylor's decomposition on
viii
Preface
three points for the numerical solutions of an initial-value problem and boundaryvalue problem for the second-order differential equations.
Note that in recent years important progress has been made in the study
of the theory of high order of accuracy difference schemes for partial differential
equations from the viewpoint of applications of an exact difference scheme or
of Taylor's decomposition function on two or three points. Chapter 4 presents
the high order of accuracy single-step difference schemes generated by an exact
difference scheme or by Taylor's decomposition on two points for the numerical
solutions of the abstract Cauchy problem for differential equations of parabolic
type
v'(t) + A(t)v(t) = f(t) (0 ~ t ~ T), v(O) = Va
in an arbitrary Banach space E with linear positive operators A(t). The wellposedness of these difference schemes in various Banach spaces is studied. The
stability and coercive stability estimates in Holder norms for the solutions of the
high order of accuracy difference schemes of mixed type boundary-value problems
for parabolic equations are obtained.
Chapter 5 is devoted to study of the high order of accuracy two-step difference schemes generated by an exact difference scheme or by Taylor's decomposition on three points for approximate solutions of the boundary-value problem for
differential equations of elliptic type
-v"(t) + Av(t)
f(t)
(0
T), v(O)
Va, v(T)
VT
in an arbitrary Banach space E with positive operator A. The well-posedness of

these difference schemes in various Banach spaces is studied. The stability and
coercive stability estimates in Holder norms for solutions of the high order of
accuracy difference schemes of mixed type boundary-value problems for elliptic
equations are obtained.
Chapter 6 gives the high order of accuracy two-step difference schemes generated by an exact difference scheme or by Taylor's decomposition on three points for
the approximate solutions of an abstract Cauchy problem for differential equations
of hyperbolic type
v"(t) + Av(t)
f(t)
(0 ~ t ~ T), v(O)
Va, v'(O)
vb
in a Hilbert space H with positive definite self-adjoint operator A. The stability

estimates of solutions of the high order of accuracy difference schemes for approximate solutions of mixed type boundary-value problems for hyperbolic equations
are obtained.
Chapter 7 is devoted to study of two abstract Cauchy problems
V'(t) + Av(t)
2 v"(t) + Av(t)
= f(t)
f(t)
(0
T), v(O)
(0 ~ t ~ T),v(O)
= VA,
vo,v'(O)
vb
Preface
IX
and the boundary-value problem
-e 2 v"(t)
+ Av(t)
= f(t)
(0 ~ t ~ T), v(O) =
Vo,
v(T) =
VT
for differential equations in an arbitrary Banach space E with the positive operator
A and an arbitrarye positive parameter multiplying the derivative term. The high
order of accuracy single-step uniform difference schemes generated by an exact difference scheme of the approximate solutions for differential equations of parabolic
type with an arbitrary parameter e in the highest derivative are presented. The
high order of accuracy two-step uniform difference schemes generated by an exact
difference scheme of the approximate solutions for differential equations of elliptic
and hyperbolic types with an arbitrary parameter e in the highest derivative are
presented. The well-posedness of these difference schemes for parabolic and elliptic equations in various Banach spaces is established. The stability estimates of
solutions of the high order of accuracy difference schemes for hyperbolic equations
with an arbitrary e parameter in the highest derivative are obtained.
Chapter 8 is devoted to study of the initial-value problem for delay partial
differential equations of parabolic type. A sufficient condition for the stability of
the solution of this initial-value problem is given. The stability estimates for solutions of the first and second order of accuracy difference schemes for approximately
solving this initial-value problem are presented. The stability estimates in Holder
norms for the solutions of the initial-value problem of the delay differential and
difference equations of parabolic type are obtained.
The monograph will be of value to professional mathematicians, as well as
advanced students in the fields of numerical analysis, functional analysis, ordinary
and partial differential equations.
Finally, we wish to express our warmest thanks to Professor S.G. Krein
(Voronezh, Russia), Professor 1. Gohberg (Tel Aviv, Israel), Professor M.A. Krasnosel'skii (Moscow, Russia), Professor V. Lakshmikantham (Florida, USA), Professor M.Z. Nashed (Florida, USA), Professor H.O. Fattorini (Los Angeles, USA),
Professor G.M. Vainikko (Helsinki, Finland), Professor H. Amann (Zurich,
Switzerland), Professor A. Pazy (Jerusalem, Israel) and Professor M.L. Gorbachuk
(Kiev, Ukraine), whose interest and enthusiastic support made this work possible.
We are immensely pleased that our book appears in this series.
Chapter 1
Linear Difference Equations
In the present chapter we consider first- and second-order linear difference equations. These equations arise in numerical analysis of approximate solutions of
boundary-value problems for ordinary and partial differential equations.
1.1
Difference Equations of the First Order
The first-order linear difference equations to be studied in this chapter may be

expressed in the form
(1.1)
where ak, bk , and'l/Jk are known numbers, Uk are unknown numbers, n is the given
integer. We will call Uk the value of grid function {udO' at the point with index
k.
A solution of the first-order difference equation (1.1) is a grid function {Yk}O
that satisfies (1.1) for all k, that is, (1.1) becomes an identity if we replace the
unknown grid function {udO' by {Yk}O'.
The initial-value problem associated with (1.1) is to find a solution {udO' of
the difference equation (1.1) satisfying the initial condition
Uo
= <p,
(1.2)
where <p is a given number.

To solve this problem, we must find a grid function {Yk}O' that not only
satisfies the difference equation (1.1) but also satisfies the initial condition Yo = <po
If bk :j; 0 for all k, 1 :S k :S n, then the difference equation (1.1) is equivalent
to the equation
(1.3)
where
Chapter 1. Linear Difference Equations
Theorem 1.1.1. There exists a unique solution of the initial-value problem (1.1)(1.2) and for the solution of (1.1)-(1.2) the formula
k
Uk = u(k,O)ep+ 2:u(k,j)epj,
j=l
(1.4)
u(k ")= {qk ... qj+1,k=l-j,

,J
1, k = J..
(1.5)
holds. Here
Proof. Using formulas (1.3) and (1.5) we can obtain formula (1.4) for k = 1.
Suppose that formula (1.4) is satisfied for k = m,
Um = u(m, O)ep +
2: u(m,j)epj.
j=l
From that and (1.3) it follows

m
Um+1
= qm+1 Um + epm+! = qm+!u(m,O)ep + 2:qm+1 u (m,j)epj + epm+1

j=l
Using formula (1.5), we can write
qm+1u(m,j) = u(m + l,j),j =I- m, u(m + 1, m + 1) = 1.

Therefore
Um+1 = u(m + 1, O)ep +
2: u(m + l,j)epj + u(m + 1, m + l)epm+1

j=l
m+1
= u(m
+ I,O)ep + 2: u(m + l,j)epj.

j=l
So, using mathematical induction we obtain formula (1.4) for any k. This completes
the proof of Theorem 1.1.1.
Theorem 1.1.2. Let Iqkl :S q. Then for solution Uk of the initial-value problem
(1.1) -(1.2) the stability inequality
k
IUkl :S qklepl + 2: qk- j !epjl

j=l
holds.
1.2. Difference Equations of the Second Order
The proof of Theorem 1.1.2 is based on formulas (1.4) and (1.5).

Finally, note that the boundary-value problem associated with (1.1) is to find
a solution Uk of the difference equation (1.1) and satisfying the supplementary
condition
(1.6)
auo + bUn = <p,
where a, band <p are given numbers.
If a or b are equal to 0, then the condition is called a local boundary condition.
If a f; 0 and b f; 0, then the condition is called a nonlocal boundary condition.
In a similar way we can obtain an analogous result for the solution of the
nonlocal boundary-value problem (1.1) and (1.6).
1.2
Difference Equations of the Second Order
The second-order linear difference equations to be studied in this chapter may be

expressed in the form
(1.7)
where ak, bk , Ck and 'lj;k are known numbers, Uk are unknown numbers, n is the
given integer. Suppose that ak f; 0 and bk f; 0 for all k.
The definition of a solution of a given second-order difference equation such
as (1.7) is like the definition of a solution of a difference equation such as (1.1).
The initial-value problem associated with (1.7) is to find a solution {udO' of
the difference equation (1.7) satisfying the initial conditions
Uo = <p, Ul = 'Ij;,
(1.8)
where <p and 'l/J are given numbers.

To solve this problem, we must find a grid function {Yk}O' that not only
satisfies the difference equation (1. 7) but also satisfies the initial conditions Yo = <p
and Yl = 'l/J.
The local boundary-value problem associated with (1.7) is to find a solution
{udO' of the difference equation (1. 7) satisfying the boundary conditions
(1.9)
where )1l,J.Ll,A.2,J.L2 are given numbers.
The nonlocal boundary-value problem associated with (1. 7) is to find a solution {udO' of the difference equation (1.7) satisfying the boundary conditions
QIUO
+ Q2 U l + f31U n + f32 u n-l = J.Ll,Q3 UO + Q4 U l + f33un + f34un-l = J.L2,
(1.10)
where Ql,Q2,Q3,Q4,f31,f32,f33,f34,J.Ll and J.L2 are given numbers.

Now we will consider the local boundary-value problem (1.7) and (1.9). To
solve them by computer we will use the following method which is called the
modified Gauss elimination method. We seek a solution of difference equations

(1.7) in the form
(1.11)
Uk = ak+ 1Uk+l + I1k+l , 0 :S k :S n - 1,
where
ak+l, 11k+1
are unknown coefficients. Hence we find
+ 11k = ak(ak+1 U k+l + 11k+1) + 11k

= akak+luk+1 + (akl1k+1 + 11k), 1 :S k :S n - 1.
Uk-1 = akUk
Now putting expressions for

[ak+1(akak - Ck)
Uk
and
Uk-1
into equation (1.7) we have
+ bk]Uk+1 + [11k +1 (akak -
Ck)
+ akl1k + 'lh]
= O.
The last equation is satisfied if we select

ak+l(akak - Ck)
11k +1(akak - Ck)
+ bk =
+ akl1k + 'k
0,
= 0,1 :S k :S n - 1.
From that it follows

(1.12)
The relations (1.12) are first-order nonlinear difference equations. For the
solution of difference equations (1.12) we need to find a1 and 111. We can find
them from Uo = Al U1 + J.L1. Thus, we have
(1.13)
For the first step using formulas (1.12) and (1.13), we can compute ak+1 and
1 :S k :S n - 1. For the second step we will find Uk, 0 :S k :S n. But, for this
we need to find Un' We can find Un from Un = Al Un -1 +J.L2 and U n -1 = an Un +I1n.
Namely
11k+1l
A111n
112
Un = .
1- A2an
(1.14)
Thus using formulas (1.11) and (1.14), we can compute Uk, 0 :S k :S n.

Finally note that when we apply algorithms (1.11), (1.12), (1.13) and (1.14), to
find a solution by computer of the boundary-value problems (1.7), (1.9), we need
(1.15)
for all k.
Theorem 1.2.1. Let for all k
(1.16)
and
IA11:S 1,IA21 < 1.
Then the requirements (1.15) and
lakl :S 1 for
all k are satisfied.
(1.17)
1.3. Difference Equations with Constant Coefficients
Proof. From (1.13) it follows that
lall = JAIl :S 1. Let lakl :S 1. We will prove that

lak+ll :S 1.
Namely
h -
akakl 2: ICkl-lakllakl 2: hl-Iakl 2: Ibkl > o.
From that it follows Ck - akak
Therefore,
::f 0 and
lakl :S 1 for all k, 1 :S k :S n.
From
lanl :S 1 and (1.11) it follows that
Thus Theorem 1.2.1 is proved.

Theorem 1.2.2. Let for all k
(1.18)
and lAd :S 1, IA21 :S 1. Then the requirements (1.15) and

satisfied.
lakl < 1 for all 1 < k are
The proof of Theorem 1.2.2 follows the scheme of the proof of Theorem 1.2.1.
1.3
Difference Equations with Constant Coefficients
The second-order linear homogeneous difference equation with constant coefficients may be written in the form
aUk-1 - CUk
+ bUk+l = 0,1 :S k < n,
(1.19)
where a, band c are given numbers. We put a ::f 0 and b ::f O.

Since for constant q and positive integer k,
it is easy to find if q is a root of the equation

a - cq + bq2 = 0;
(1.20)
if so, then the grid function {udo = {qk}O is the solution of equation (1.19).
Equation (1.20) is called the auxiliary equation associated with (1.19). The
auxiliary equation for (1.19) is of degree 2. The key quantity is c2 - 4ab, which is
known as the discriminant of the quadratic equation (1.20).
If C2 - 4ab is positive, there will be two real values for q and they are defined
by
C
ql,2 =
Jc 2
2b
4ab
(1.21)
Recall that given the grid functions {uno and {uno, if constants AI, A2, not all
zero, exist such that
or
Al u~ + A2u~ = 0,0 :s: k :s: n,

then the grid functions {uno and {uno are said to be linearly dependent.
If no such relation exists, the functions are said to be linearly independent.

That is, the grid functions {uno and {uno are linearly independent when the
last equation implies that Al = A2 = O.
It is easy to show that for any linearly independent grid functions {uno and
{uno, we can determine
~~
Uk _ l
~~
Uk - l
I,
such that the determinant is not equal to zero for any k,l
easy to show that the solutions
{uDo =
:s: k :s:
n. Then it is
{q~}o,{uno = {q~}o
are linearly independent and the general solution of (1.19) can be written at once.
It is
{Uk}O = {Clq~ + C2q~}O
(1.22)
in which CI, C2 are arbitrary constants.
If c2 - 4ab is zero, then the auxiliary equation (1.20) has repeated roots
q = ql = q2 It is easy to show that the grid functions {kffiqk}O' where m = 0,1
are linearly independent and the general solution of (1.19) is
{udo = {cIl + C2kqk}O
(1.23)
in which CI, C2 are arbitrary constants.

If c2 - 4ab is negative, then the auxiliary equation (1.20) has two complex
roots
C iJ4ab - c2
ql,2 =
2b
= rei<p,
where r
= Jf, tan cp = ~. It is easy to show that the grid functions

{r k sin kcp}o, {r k cos kcp}o
are linearly independent and the general solution of (1.19) is
{Uk}O =
in which
CI, C2
{CI r
k sin kcp + C2rk cos kcp}
are arbitrary constants.
(1.24)
Now using the variation of parameters method we will find the general solution of the nonhomogeneous equation
aUk-l - CUk
+ bUk+l
= <Pk, 1 :s: k
:s: n
- 1,
(1.25)
where a, b, c and <Pk are given numbers.

Namely, if {vd(}, {wd(} are linearly independent solutions of (1.25), then we
can use them to determine the general solution of the nonhomogeneous equation
(1.25). Suppose that
(1.26)
is a solution of (1.25). Let us see what happens if we replace both of the constants
in (1.25) with Ctk and 13k, respectively. That is, we consider
(1.27)
and try to determine Ctk and 13k so that the grid function {CtkVk
solution of the equation (1.25). From (1.27) it follows that
+ 13kwd(}
is a
(1.28)
+(Ctk+l - Ctk)Vk+l
+ (13k+l
- 13k)wk+l'
We now choose some particular function for the expression

(Ctk+l-Ctk)Vk+l +(13k+l-13k)wk+l.
Technically, we could let this grid function be {sin k<p} () , { qk}(), {cos k<p} () or any
other suitable grid function. For simplicity we choose
or
(Ctk - Ctk-l )Vk
+ (13k -
(1.29)
13k- dWk = O.
Then it follows from (1.28) that

Uk+l - Uk = Ctk(Vk+l - Vk)
or
Uk - Uk-l = Ctk-l
+ (Vk
+ 13k (Wk+l
- Wk)
+ 13k-l(wk -
- vk-d
wk-d
(1.30)
It is easy to see that from (1.25) it follows that

-a(Uk - uk-d
+ (a - c + b)Uk + b(Uk+l
- Uk) = <Pk, 1 :s: k < n.
We substitute (1.27) and (1.30) into equation (1.31) to obtain

-a[Ctk-l (Vk - vk-d
+ 13k-l (Wk
+b[Ctk(Vk+l - vd
- Wk-l)]
+ (a - c + b)[Ctkuk + 13kwk]
+ 13d w k+l
- Wk)] = <Pk,
(1.31 )
or
O!k[aVk-l - CVk
+ bVk+d + .8k[aWk-l -
+a[(O!k - O!k-d(Vk - vk-d
+ (.8k -
CWk
+ bWk+l]
.8k-d(Wk - Wk-l)] = <{}k
But VkS and WkS are solutions of the homogeneous equation (1.19), so that finally
Equations (1.29)and (1.32) now give us two equations that we wish to solve for
O!k - O!k-l and .8k - .8k-l. This solution exists provided that the determinant
Vk-l
Vk - Vk-l
Wk-l
Wk - Wk-l
does not vanish. But
and the last determinant is not equal to zero for the linearly independent solutions
{vdS and {wdS of equation (1.19). Therefore using equations (1.29)and (1.32)
we can find O!k - O!k-l and.8k - .8k-l. Namely
.8k - .8k-l =
Vk<{}k
a(vk-lwk - VkWk-l)
Hence using formula (1.4), we can find

k
O!k = O!o -
Wj<{}j
L a(vo-lwo - VoWo-l) ,
j=l
(1.33)
.8k = .80 + "

Vj<{}j
.
LJ
a(vo-lwo
- voWo-l)
j=l
J
J
J J
(1.34)
Thus, O!k and .8k of (1.33) and (1.34) may be inserted in (1.27) to give us the
general solution of equation (1.25)
(1.35)
where O!O, .80 are arbitrary constants.
Finally note that the method of variation of parameters can be applied to

the solution of equation
akUk-1 - CkUk
+ bkUk+1
= <Pk, 1 ~ k
<n
with dependent coefficients. Namely we have the formula
Uk
"
(WkVj - VkWj )<pj
= O:OVk + (30Wk + L.J
j=1 aj Vj-1 Wj - VjWj-1
).
Now we will consider the initial-value problem

aUk-1 - CUk
Recall that
+ bUk+1
< n, Uo
= <Pk, 1 ~ k
= <P, U1 =
'l/J.
(1.36)
{vdo and {wdo are linearly independent solutions of (1.19), therefore

Vk
Vk
> 0,
q~'C2 - 4ab
qk, c2 - 4ab
r sin k<p, c
q~'C2
> 0,
= 0,
- 4ab
2
k qk,c - 4ab
k
r cos k<p, c
(1.37)
< 0,
4ab
= 0,
(1.38)
4ab
< 0.
Thus using formula (1.35) we have the following theorem.

Theorem 1.3.1. There exists a unique solution of the initial-value problem (1.36)
and for the solution of (1.36) the following formulas are satisfied:
If c2 - 4ab > 0, then
Uk =
If c2
4ab
= 0,
If c2
4ab
< 0, then
k-1
k-1)
q1 q2 ( q1
- q2
q2 - q1
<P
+ q2k -
k
q1 .1.
q2 - q1
'P
then
sin(l -
k)<p k
sm<p
<p
~ sin(k - j)<p
+ L.J
j=1
asm<p
sin k<p
+ -.- r
sm<p
k-J+1
k.l,
'P
<po
J
Chapter I, Linear Difference Equations
10
Here and in future we will put

k
2: ak,j = 0, when k < 1.

j=l
Now we will consider the boundary-value problem

(1.39)
Using formulas(1.35), (1.37) and (1.38), we have the following theorem,
Theorem 1.3.2. There exists a unique solution of the boundary-value problem (1.39)
and for the solution of (1.39) the following formulas are satisfied: If c2 - 4ab > 0,
then
If c2
4ab = 0, then
n - k k
- - q J.1.1
+_k
n
If c2
4ab
2:
j=l
k k-n
+ -q
J.1.2
n
,
J- n
_-l-J+1cp
a
k '
2: __
-J qk-)+lcp',
'
a
)
k
j=l
< 0, then
sin( n - k)cp k
.
r J.1.1
smncp
sinkcp k-n
+ -,
-r
smncp
sinkcp ~ sin(j - n)cp k-)'+l
+-,--~
smncp ,asmcp
)=1
~ sin(k - j)cp
+~
j=l
asmcp
k-)"+l
J.1.2
cp
cpo
)
In the same way we can study solutions of the local and nonlocal boundaryvalue problems for equation (1.25), with boundary conditions (1.9) and (1.10),
respectively,
Chapter 2
Difference Schemes for
First-Order Differential Equations
In the present chapter we consider the single step difference schemes generated
by an exact difference scheme or by Taylor's decomposition on two points for the
numerical solution of an initial-value problem for first-order differential equations
and systems of first-order differential equations.
2.1
Single-Step Exact Difference Scheme and Its Applications
We consider the initial-value problem of the form
y'(t)
+ a(t)y(t) =
f(t),O < t ::; T, y(O)
Yo
(2.1)
assuming a(t) and f(t) to be such that problem (2.1) has a unique smooth solution
defined on [0, T]. In this section we shall construct difference schemes of a high
order of accuracy for an approximate solution of problem (2.1). On the segment
[0, T] we consider the uniform grid space
[0, T]r = {tk = kr, k = 0, 1, ... , N, NT = T}

with step T > O. Here N is a fixed positive integer. We will discuss single-step difference schemes for computing numerical values of the solutions. These difference
schemes are step-by-step methods, that is, we start from the given approximate
values of y(tk-l) and proceed stepwise, computing approximate values of y(tk) at
the grid (mesh) points.
= tk. Then
{y(tknS" is the solution of the initial-value problem for the first-order difference
equations
Theorem 2.1.1. Let y(tk) be the solution of (2.1) at the grid points t
1
-(y(tk) - y(h-l))
T
1
+ -(1
T
J'k
'k-l
a(s)ds
)y(tk-d
(2.2)
Chapter 2. Difference Schemes for First-Order Differential Equations
12
Proof. For solution of (2.1) we have the formula
= e- I~ a(s)ds yO +
yet)
Putting t
= tk
= tk-I
and t
e- I: a(z)dz f(s)ds.
into the last formula, we can write
e- 10 k a(s)ds yO
r e-
+ io
tk
I:k
a(z)dz f(s)ds
e _ I'k
'k-l a(s)ds ['k-l
e- 10
a(s)ds yO
tk
+ io
tk l
-
e- Is'k-l a(z)dz f(s)ds ]
e- I:k a(z)dz f(s)ds,
tk-l
y(tk-I) = e- I~k-l a(s)ds yO
tk l
-
e- I.'k-l a(z)dz f(s)ds.
Hence, we obtain the following relation between y(tk) and y(tk-d :
y(tk) = e- 1':_1 a(s)dsy(tk_l)
tk
e- I:k a(z)dz f(s)ds.
tk-l
The last relation and equality (2.2) are equivalent. Theorem 2.1.1 is proved.
Thus, the solution yet) of the differential problem (2.1) at the grid points
t = tk is the solution of the difference problem (2.2). Therefore, the initial-value
problem (2.2) is called the single-step exact difference scheme for solution of the
initial-value problem (2.1).
Now, we will consider the applications of this exact difference scheme. From
(2.2) it is clear that for approximate solutions of problem (2.1) it is necessary to
approximate the expressions
e
-I:k-l a(s)ds
k
and
Let us remark that in constructing difference schemes it is important to know

how to construct a right-hand side <P~,q that satisfies
1i
-:;
tk
e- I:k a(z)dz f(s)ds - <P~,q = 0(7 P+q )
(2.3)
tk-l
and is sufficiently simple. Here p and q are positive integers and we will put p
The choice formula <P~,q is not unique. Using Taylor's formula, we obtain
- p+q-I
1; {;
_
()
>.
tk
e- J:k a(z)dz f(s)ds
(2.4)
tk-l
,
- I,Ll
a(z)dz
f3m-A(tk-df
(A)
q.
(tk-I) (m + I)!
+ 0(7
p+q
),
13
2.1. Single-Step Exact Difference Scheme and Its Applications
(2.5)
(2.6)
Here and in this chapter we will put
= 0,
1::; i ::;p+q.
(2.7)
Therefore, for the construction of <pf.,q we can put
(2.8)
Moreover, for the construction of <P~,q we will also use (2.4) and (2.6). But, first
of all let us study the approximate formulas for the expressions
- J,'k
e - J,':_1 a(s)ds
and
a(s)ds
e k- ~
that will be needed in the future for the construction of difference
schemes of a high order of accuracy for an approximate solution of problem (2.1).
-
Now, let us consider the approximate formulas for the expression e

constructed by Pade's rational functions. It is clear that
e - J::- 1 a(s)ds
e-
J,'k a(s)ds
k-1
= e-a(tk)or
=e -a(t k _ 2"l)or -
tk
'k
J a(s)ds
'k-1
e-a(tkHtk-z)(a(tk) _ a(z))e - J':_l a(s)ds dz,
tk-1
tk
tk-1
- J'k
e
'
a z -a (t k-~ )) e -a(t k _ 2"1 )(z-tk-Il dz.
a(s)ds( ( )
14

Using Taylor's formula, we obtain
e-
f":_1 a(s)ds = e-a(tk)T + T(a(tk)
+ '~
(::+; t, (7)
_ ~ (n
)!
_ a(tk_d)e-a(tk)T
P(t,-!l [(a(t,)- a(t,-!l )(a(t, ))n-;
j i) aU) (t,_!l(a(t,))n-;-j ] e-oU,)' + o( ,,+0+
),

i
= 0,
1~ i
~ p
+ q.
Further, using the last formulas and Pade fractions for the function e- z , we
can write
'%' (::+;)! ~ (7)
- ~ (n j
15
iJ,(tk-d [(a(tk) - a(tk-d)(a(tkn-'
i) aUI (tk-,)( a(tk))n-'-i ] R,-n+[,[_',,_[,[(rae tk)) + a(rv+O+'),
e- Ijtkk_l a(s)ds = Rq,p(Ta(tk_l)) - r(a(tk) - a(t k_l))Rq- 1,p-2(Ta(t k_l)) (2.11)

2
2
2
p+q-l (
)n+l
+ ~ (:: I)!
+~
(n j i)
t; 7 (3i(tk) (a(tk) - a(tk_4))(-a(t k_4)t- i

(
alii (tk)( -a(tk_! ))n-'- i ] R,_n+[,[_ ',,_"[(
rat t k_))
+o( rp+q+l ).
Here Rq,p(z) = ~:::g? are Pade's approximations of the function e- Z near

= O. They are given by the formulas
=L
q
Qp,q(z)
k=O
(p+q-k)!q!
k
(p + q)!k!(q _ k)! (-z) ,Pp,q(z)
=L
p
k=O
(p+q-k)!p!
k
(p + q)!k!(p _ k)! Z .
Now let us consider approximate formulas for the expression
J
tk
exp{ -
tk _
a(s)ds},
constructed by Pade's rational functions. We have that
-J -
- I: k
k-!
a(s)ds
= e
-a(t
I)L
k- 2 2
tk
a(s)ds( a ()
e Itk
z
Z
dZ.
a (t k- 4)) e -a(tk_l)(Z-tk_l)
2
2
16
-f:k-~
k
a(s)ds
T I l
= e-a(tkh' + 2'T(a(t k ) - a(tk_k))e-2a(tk)T
/~' 2n+;;~ 1)!

-
~ (n j
t.(:)
Pi(I'-J +a(I') - a(I'_J ))(a(I,))n-i
i) aU) (1,_ J)( a(I,) r- H ] e- n(',); + o(
TP+'+1),
Further, using the last formulas and Pade's approximations of the function
e- z , we can write
e
- f.' k
k-~
a(s)ds
T I T
+ 2'T(a(t k ) - a(tk-4))Rq-l,p-l('2a(tk))
= Rq,p('2a(tk))
/~' 2n+;;~ 1)!
t.(:) d
(a(lkl- a(I'_j ))(a(I,)
A(I,-
_~ (n j i) aGl(lk_! )(a(lk))n-i-
(2.12)
r-'
R,-n+II-1,P-II( ;a(t,))
+ 0(T P+O+ 1),
(2.13)
17
Now, using formulas (2.4), (2.6), (2.9), (2.10), (2.11), (2.12) and (2.13), <fJf.,q can
be defined by one of the following formulas:
(2.14)
(2.15)
(2.16)
18
<P~,q
J.;f; (2r;) f32m->.(t k-!)j(>)(tk_!)

p-1 2m
2m
(2.17)
22m (2m + 1)! R q- m,p-m(-2 a(t k ))

T
+"2T(a(tk) - a(t k_! ))Rq-m-1,p-m-1 (2"a(t k ))

m-1
n+1
+ ~ 2n+;(n + 1)! ~
7 A(tk_!) (a(tk) - a(tk_!))(a(tk)t-i

)
-~ (n j i) aU) ('k-!)( a(tk) j"-'- 14-0+[;

~a(tk))] ,
(2 )
(2.18)
<P~,q ~of; r; f32m->.(t k-!)j(>')(t k_!)
[-1,,-[; [(
j]
p-1 2m
2m
X 22m (2m
T
+ 1)! [ Rq-m,p-m( 2"a(tk-!))
-"2T(a(tk) - a(tk_!))Rq-m-1,p-m-l(2"a(tk_!))
m-1
(_ )n+1
+ ~ 2n+1~n + 1)! ~
+~
7 f3i(tk)
)
(n j i) aU) (tk)(a(tk-l
(a(tk) -
a(tk_~))( -a(tk_~))n-i
))O-'-j] 14-0+[>[-1,,-1'1[(
~a(tk-l))] .
Now replacing the expression e- ft : - I a(s)ds with its Pade fractions from (2.9) or
(2.10) or (2.11) we obtain the difference schemes
,-I(Uk - Uk-I)
+ ,-I
[1-
R",(a(tk_I)')
+T(a(tk) - a(tk-d)Rq-1,p-1(a(tk-dT)
-
7
+ I: (n ~ i) a(j)(tk)(-a(tk_1))n-i-
p+q-1 (_ T)n+1 n ( )
[
~ (n + 1) ~
f3i(tk) (a(tk) - a(tk_d)(-a(tk_d)n-i
j=l
XRq-n+II-1,p-Ij(a(tk-1)T)]] Uk-1
<P~,q, 1 :s k :s N, Uo = Yo;
(2.19)
r-' (Uk -Uk-') +r-'
p+q-1
-
[1-
R",,(ra(t,) -r(a(ik) - a( tk_') )R,,-',,_' (a(t,)r) (2.20)
~ (7) A(tk-1)
n
+1
(:: I)!
19
_I: (n ~ i)
(a(tk) -
a(tk_~))(a(tk))n-i
a(j)(tk_1)(a(tk)t- i- j
j=l
XR,-n+I'lI-',,-I'iI (ra(tk))]] Uk-'
r-'(uk - Uk_,)
+ r-'
~ 1':",1 '" k '" N, Uo ~ yo;
[1-
(2.21)
R",,(ra('k-l))
+r(a(tk) - a(t k_l))Rq1,p-dra(t k_l))

2
2
-
7
a(tk_~))(-a(tk_~)t-i
+ I: (n ~ i) a(j)(tk)( -a(tk_)t- i-
p+q-1 (_ )n+1 n ( )
[
~ (n~ I)! ~
f3i(tk) (a(tk) -
j=O
x R,-n+1 'i1-",-I 'i1(ra(tk_l))] ] Uk-'
~ 'Pl:", 1 '" k '" N, Uo ~ Yo
Let us remark that in constructing such difference schemes we can choose <P~,q by
formulas (2.8) or (2.14) or (2.15) or (2.16) or (2.17) or (2.18). Thus, we construct
the difference schemes of (p + q)-order of accuracy for the approximate solution of
the initial-value problem (2.1). Finally, let us give some examples of the simplest
difference schemes.
In the case p + q = 1 from formulas (2.8), (2.14), (2.15), (2.19) and (2.20) it
follows that
Uk - Uk-1
-- + a(tk-duk-1 = f(tk-d, 1 ~ k
r
N, Uo = Yo
(an explicit Euler's difference scheme of first order of accuracy for the initial-value
problem (2.1)), and that
Uk - Uk-1
r
1
+ (1 -
1 + ra(tk) )Uk-1
-(Uk - uk-d + a(tk)uk = f(tk), 1 ~ k

r
= f(tk) 1 + ra(tk) '

~ N, Uo =
Yo
(an implicit Euler's difference scheme of first order of accuracy for the initial-value
problem (2.1)).
20
In the case p + q = 2 from formulas (2.18), (2.20) it follows that
1 ~ k ~ N, Uo = Yo
(an explicit difference scheme of second order of accuracy for the initial-value
or
(a Crank-Nicolson difference scheme of second-order accuracy for the initial-value

Uk - Uk-l
T
+ -:1;:
1-
1 +Ta(tk_~)
(ra(t
_ 1
))2
Uk-l
or
(an implicit difference scheme of second order of accuracy for the initial-value
problem (2.1)).
Now, we consider the initial-value problem of the form
y'(t)
+ a(t)y(t) = f(t, y(t)), 0 < t
T, y(O)
= Yo
(2.22)
assuming a(t) and f(t, y(t)) to be such that problem (2.22) has a unique smooth
solution defined on [0, T]. The exact difference scheme of problem (2.22) obviously
has the form
-1 (Y(tk) - Y(tk-l ))
T
11
=T
tk
tk-l
e- J:k
+ -1 (1 T
a(z)dz f(s,
e-
J:
k-l
a(s)ds) y ( tk-l )
y(s))ds, 1 ~ k:S N, y(O)
= Yo.
(2.23)
21
From (2.23) it is clear that for the approximate solution of problem (2.22) it is
necessary to approximate the expressions
e- J,'kk-l a(s)ds
and
In exactly the same manner with approximate formulas for the expression
one establishes the formulas
11
tk
e- J:k a(z)dz f(s, y(s))ds
(2.24)
tk-l
(2.25)
( )m
q
fa E r;: ) f3m->.(tk)!>.(tk, y(tk)) (r::
I)! + O(T + ).
p+q-l m (
/>.(s, y(s)) = f(s, y(s)),

a),
f>'+l(S,y(S)) = [fSAf(s,y(s))
+ Li=l
>.
(,\)
a),
,\ = 0,
as' iaiyf(s, y(s))Ji(s, y(s)),
0::; ,\ ::; p + q - 1.
(2.26)
Here s = tk or tk-l. Now, replacing the expression e-J,':_la(s)ds by its Pade

fractions from (2.9) or (2.10) or (2.11), for example, we can obtain the difference
schemes
r-' (Uk - Uk-,)
+ r-' [1 - R"p(a(tk-,)r)
+r(a(tk) - a(tk-d)Rq-l.p-l(a(tk-dT)
(2.27)
22
p+q-l (_ T )n+l n ( )
[
i
~ (n + 1) ~
f3i(tk) (a(tk) - a(tk-I))(-a(tk-dt-
+~
(n j i) ali) (tk)( -a(tk- d )n-i-; R,_n+[
'1)-1,,- ['I) (ra(t k ))] Uk-1
l; ~ (m)>. f3m->-.(ik-df>.,(tk-l, uk-d
p+q-I
=
x (m : I)! [R,-m-['l']'p+['l')(a(tk-d r )
-T(a(tk) - a(tk-d)Rq-m-[TI-I,P+[TJ-I(a(tk-I)T)
m-I (_ T )n+1
7 f3i(tk) [(a(tk) - a(tk-I))(-a(tk-dt- i
n ( )
+ ~ (n + 1) ~
I: (n ~ i)
a(j)(tk)( -a(tk_l))n-i- j
j=1
x R,,-n+['! )-1,,-[ '!) (ra(tk_
r-
(Uk -Uk-l) +r - l
p+q-I
d)]], ~
1
~ N, Uo ~ Yo,
[1- R", (ra(tk - Teartk) -a(t'-9) )R,,n+1
l,,-l
(a(tk)r) (2,28)
n ( )
- ~ (:+ 1)! ~
7 a(tk-d[(a(tk) - a(tk_I))(a(tk)t- i
_~ (n j i) a(;)(tk_d(a(td))n-i-; R,_n+['1[_l,,_['!)(Ta(tk]
7 f3n-i(tk-I)!i(tk, Uk) ((:~)n+11)! ,1 :S k:S N, Uo
U'-l
p+q-I n ( )
=
~ ~
Yo
Thus, we construct the difference schemes of (p + q)-order of accuracy for approximate solutions of the initial-value problem (2.22). We observe that the difference
scheme (2.28) is a first-order nonlinear difference equation in Uk. Therefore we
use an iteration formula
23
1 :S k :S N, u~ = Yo, m = 1,2, ...
,u2 = Yo
But what about a criterion for stopping the iteration? We can use the criterion
of closeness of consecutive terms to terminate the iterative procedure of (2.29). If
there exists M for them
u~ _U~-l =o(r p +q )
for any k, 1 :S k :S N. Then we will put Uk = u~ for any k, 1 :S k :S N.
Now, we consider the initial-value problem of the first-order linear system
of ordinary differential equations. Let y(t) be an m-dimensional column unknown
vector function of t and f (t) a column vector whose components are known functions of t. Further suppose that A(t) is defined by
A(t) =
al.l.(.t)
(
aml (t)
be an m x m matrix function of t. Then the initial-value problem
y'(t)
+ A(t)y(t) = f(t), 0 < t :S T, y(O) = Yo
(2.30)
represents an initial-value problem for a system of m linear ordinary differential

equations
~~~~t!.~.~l.l.(~~~l.(~~ ~::: ~ ~.l~(.t~Ym(t) =

{
ftYm(t)
+ aml(t)Yl(t) + ... + amm(t)Ym(t)
in the m unknown Yl (t), ... , Ym (t) functions. Let
ft(t),
= fm(t)
(2.31)
24
be a solution y(t) of problem (2.30) at the mesh points t = ik. Then the exact
difference scheme of problem (2.30) obviously has the form
I1
= -;;
tk
tk-l
u(t, s )f(s)ds, 1 S k S N, y(O) = Yo,
where I is the m x m unit matrix. Here u(t, s) is the m x m matrix of functions

of t and s. It will be defined as a solution of the problems
au(t,s)
as
= u(t, s)A(s), 0 S sSt S T, u(t, t) = I,
au(t, s)
at = -A(t)u(t,s),O S sSt S T,u(s,s) = I.
(2.33)
(2.34)
It is easy to show that
u(t, s) = e-A(>-.)(t-s) - i t u(t, z)[A(z) - A(A)]e-A(>-')(Z-S)dz,

and
u(t, s) = e-A(>-.)(t-s) + it e-A(>-.)(t-Z) [A(A) - A(z)]u(z, s)dz

for any A, s :5 A :5 t. Hence we can write
U(tk, tk-l) = e-A(tk-d T
_l
1::,
U(tk, tk-d
e-A(tk)T
(2.35)
(2.36)
e-AU,)(',-') [A(t,) - A(z)I"(z, t,_,)dz,

(2.37)
tk
U(tk, z)[A(z) -
tk-l
A(tk_~)] e-A(tk_!)(z-t k- d dz.
From (2.32) it follows that for an approximate solution of problem (2.30) it is

necessary to approximate the expressions U(tk' tk-d and
Ijt
tk-l
U(tk, s)f(s)ds.
25
11
tk
u(tk,s)f(s)ds
(2.38)
tk-l
(2.39)
(2.40)
where
Bi(S) = 1,
()
{ Bi(s) = Bi~l
(s)
= 0,
+ B i- 1 (s)A(s), 1:S i
:s p + q.
(2.41)
Now, we study approximate formulas for the expressions U(tk, tk-d and
u(tk, tk_l).
Using Taylor's formula and formulas (2.35), (2.36) and (2.37), we
2
obtain
U(tk, tk-l) = e-A(tk-d r - T(A(tk) - A(tk_d)e-A(tk-dr
(n:
p+q-l (_
+ ~
+~
)n+l
1)!
t;
n
~ Bi(tk) (A(tk) - A(tk-d)(-A(tk_d)n-i
(n j i) AI;) (tk)( _ A(tk-l ))n-H] e-AI'.-,)' +

U(tk, tk-d = e-A(tk)r
/~l (::+:)!
t, (7)
_~ (n j i) A
0(7'+9+ 1),
+ T(A(tk) - A(tk_d)e-A(tk)r
B,(tk-Il [(A(tk) - A(tk_l))(A(tk))n-,
I;) (tk-Il(A(tk W-'-;] e-AU,)'
+ o(7H '+1),
26
p+q-l (
)n+l
+ ~ (:: I)! ~
x
7 Bi(tk) (A(tk) - A(tk_!))( -A(tk_!)t-i

)
~ (n j i) A(;) (t,)( _ A(t,_!) )n-i-
e-A,_!)< + o(T'*'+'),
where
Further, using the last formulas and Pade fractions for the function e- z, we
can write
0+ 1) ~ 7 Bi(tk) (A(tk) - A(h-l))(-A(tk_d)n-i

+ ~ (n j i)
p+q-l (
)n+l n ( )
+ ~
A(j)(t,)(-A(t,_,)"-H] R,-n+[;[-"p-[;[(A(t,_,)T)
U(tk, tk-d
+ p~,
Rq,p(7A(tk)) + 7(A(tk) - A(tk-d)Rq-1,p-l (A(tk)7)
(::+;)1 to (7)
-~ (n j i)
(2.44)
Bi(t,_,) [(A(t,) - A(t,_,))(A(t,) )n-i
A (j) (t,_ ,)(A (t,))n- H ] R,-n+[; ]_',p_[;] (TA(t,))
+o( 7 P+ q + 1 ),
u(tk, tk-l)
Rq,p(7A(tk_l))
- 7(A(tk) - A(t k_l))Rq(2.45)
1,p-l(7A(t k_l))
2
2
2
p+q-l (
)n+l
+ ~ (:7+ I)! ~
+~
(n j i)
7 Bi(tk) (A(tk) - A(tk_!))( -A(tk_!)t- i

)
A(j)(t,)( -A(t,_! ))n-i- j ] R,_n+[;]_"p_[;](TA(t'_j))
+o( 7 P+q+ 1).
27
Now, we consider the approximate formulas for the expression U(tk, tk-d con2
structed by Pade's rational functions. We have that
U(tk, tk_l) = e -A(tk_!)~
U(tk, z)(A(z) - A(tk_l))e -A(tk_!)(z-t k _!) dz.
_itk
k- 2
Therefore, using the last formulas and Pade's approximations of the function e- z ,
we can write
/~' 2n+;~~ 1)!
-~ (n j i)
t,(;) B'("-l
>[rA(t,) - A(t,_ j ))( A(t,)
~ (~A(t,
AU) ("-I )(A(,,))n-'-j] R,-n+[ ,1_ ',p_[ 1
t-'
28
+o(r P+q+ 1 ).
Finally, replacing the expression U(tk, tk-d by its Pade approximations from (2.43)
or (2.44) or (2.45) (neglecting the last terms, respectively) and the expression
by the expression from (2.39) or by the expressions from (2.38) and (2.40) first
by its Pade approximaone replacing the expressions U(tk, tk-d and U(tk, t k_!)
2
tions from (2.41) or (2.42) or (2.43) or (2.46) or (2.47) (neglecting the last terms,
respectively) we obtain the difference schemes of (p + q)-order of accuracy for an
approximate solution of the initial-value problem (2.30)
(2.48)
+r(A(tk) - A(tk-d)Rq-1,p-l(A(tk-l)r)
-
+~
p+q-l (_ r )n+l n ( )
[
~ (n + 1)
~ Bi(tk) (A(tk) - A(tk-d)(-A(tk_dt- i
(n j i) A0I(tk)(-A(tk_d)n- H R,-n+I~]-',p-I~1 (A(tk-dT)]] Uk-l

= {f'lp,q
< N, Uo = Yo',
rk ' 1 <
- k T-
(Uk - uk-d + T-'
[I - R",p(TA(tk))
-r(A(td - A(tk-l))Rq-1,p-l(A(tk)r)
_P%l (::+~)! ~ (~)
Bi(tk-d [(A(t k ) _ A(tk_d)(A(tk))n-i
(2.49)
-~ (n j i)
29
A(j) (tk_,)(A(tk))n- H R,-n+[ [-I,p-[, [( r A(t,))]]

=
1<
- k
InP,q
Yk
'
r-I(u, - Uk-,)
<
-N
, Uo
+ r- I [1 -
.'-1
= Yo',
R"p(rA(t k_
(2,50)
+7(A(tk) - A(t k_l))Rq1 ,p-l(7A(t k_l))

2
2
-
+~
)n+l n ( )
p+q-l (
~ (~: I)! ~ ~ Bi(tk) (A(tk) - A(tk_~))( -A(tk_~))n-i
(n j i)
A(j)(t,)(-A(t k_ j))n-'-j
R,_n+[~]_I,p_[>](rA(tk_j))]] .'-1
= InP,q
rk ' 1 <
- k <
- N , Uo
= Yo.
Here, we can choose of 'P~,q which can be defined by one of the following formulas:
p,q _
'Pk or
~:" ~ :~I
1; f;m ( A ) Bm-.x(tk)f
p+q-l
E(7)
( . x l -( 7 )m
(tk) (m + I)!'
r
Bm-, (tk-,) (m : I)! [R,-m-['l'],p+['l'[(A(t,_,)r)
-7(A(tk) - A(tk-t})Rq- m-[TJ-l,P+[TI-l(A(tk-l)7)

m-l (
+~
+~
or
(n j i)
0+)n+l ~n ( ~) Bi(tk) [(A(tk) - A(tk-l))(-A(tk_t})n-i

7
1)
A (j) (t,)( - A(tk_,) )n-H ] R,-n+[, ]-I,p-[,] (A(t,_,)r)]
II') (tk-l),
30
or
m-l (
)n+l
+ ~ (~: 1)! ~
(n j i)
+~
7 Bi(tk) (A(tk) - A(tk_ 1 ))(-A(tk_ 1 ))n-i

)
A(j)(t,)(_ A(t,_! ))n-i- j
R,,-n+]~]-I'P-]~](7A(t,_! n] l')(t,-,),
or
+i7(A(tk) - A(tk-1))Rq-m-l,p-m-l("2A(tk))
m-l
n+l
+ ~ 2n+;(n + 1)! ~
-~ (n j i)
7 Bi(t k- 1) (A(tk) - A(tk_ 1 ))(A(tk))n-i

)
A(J)(t,_,)(A(t,))"-H R,-n+]'I)-I,p-]'I)(
~A(t,n] j(')(t,_,),
or
-27(A(tk) - A(tk-1))Rq-m-l,p-m-l(2A(tk-1))
m-I
)n+l
+ ~ 2n+~~n + 1)! ~
+~
(n j i)
7 Bi(tk) (A(tk) - A(tk_ 1 ))(-A(tk_ 1 ))n-i

)
A (j) (t,)(A(t,_!) )n-i- j R,_n+]
~(-I,p-]~ I(~A(t,_ j n] jl') (t,_!)
31
2.2. Taylor's Decomposition on Two Points and Its Applications
Note that using the exact difference scheme, we can extend our discussion to construct the difference schemes of an arbitrary high order of accuracy for approximate solutions of the initial-value problem for the first-order nonlinear system of
ordinary differential equations
y'(t)
+ A(t)y(t) = f(t,y(t)),O < t::; T,y(O)
= Yo.
It is left to the reader.
2.2 Taylor's Decomposition on Two Points and Its Applications

We consider again the initial-value problem (2.1). Utilization of Taylor's decomposition on two points in the construction of the single-step difference schemes of
a high order of accuracy for approximate solutions of problem (2.1) is based on
the following theorem.
Theorem 2.2.1. Let the function v(t)(O ::; t ::; T) have a (p
derivative and tk-1, tk
V(tk) - v(tk-d
+ I>l!jv(j) (ik)T j j=l
where
a = (p+q-j)!p!(-ljJ
+ q + l)-th continuous
[0, TJr. Then the following relation holds:
(p+q)!j!(p-j)!
. - (p+q-j)!q!
,BJ - (p+q)!j'(q_j)!
J
L,BjV(j)(tk-1)T j
j=l
for any j,l ::; j ::; p,

for any j, 1 ::; j 5:. q.
(2.51 )
(2.52)
Proof. Using the formula for integration by parts, we obtain the representation
jt
(tk _ s)q(s - tk_1)PV(P+q+1)(s)ds
tk-l
p+q
L( -l)p+q-'((tk - s)q(s - tk-1)P)(P+q-')v(r)(s)1~~_1'
,=0
We will calculate the expressions
and
1 Prof.
V.L. Makarov (Ukraine) told me about this proof.
32
If j
p + 1, then p + q - j
:s q -
1. Therefore
(-l)p+q-j [(tk - s)q(s - tk-dP](p+q-j) !S=tk = O.
If 0
:s j :s p, then
p+q-j (
- ')'
P + q J. ((t - s)q)(P+q-j-i) ((s - t _ )P)(i) I _
LJ
"(
+
k 1
s-tk
i=O Z. P
q - J. - Z')'. k
= (-l)p+q-j ~
(
')'
P+ q - J .
~ i!(p + q - j - i)! (j
p+q-j
(-1 )p+q- j ~
q.
+i -
p)!
p
x (-1 )p+q-j-i(tk - s )j+i- (p ~ i)! ((s - tk_l)P-i IS=tk
(-l)p+q-j ( p+q
')'
J 'q!(-l)qP'T j
j!
(p - j)!q!
(-l)P-j (p+q)!(p+q-j)! , j _ (-l)P( + )1 .

,
(_ ')'"
p.T P q .0:).
(P + q ) .
P J .J.
So,
(-l)p+q-j [(tk - s)q(s - tk-l)P](P+q-j) IS=tk
If j
q + 1, then p + q - j
:s p -
= (-l)P(p + q)!O:j.
1. Therefore
(-l)P+q-j [(tk - s)q(s - tk-l)P](P+q-j)
If 0
IS=tk-l
= O.
:s j :s q, then
p+q-j (
- ')'
P + q J. ((t - s)q)(p+q-j-i) ((s - t _ )p)(i) I _
LJ
"( + q - J. - Z')1. k
k 1
s-tk-l
i=O Z. P
= (-l)p+q-j ~
p+q-j (
- ')'
,
(-l)p+q-j ~
p+q J.
q.
~ i!(p+q-j-i)!(j+i-p)!
1
x( -l)p+q-j-i(tk - S)j+i- p (p~' i)! ((s - tk-dp-ils=tk_l
33

-
')1
(-l)p+q-j ( p+q J 'q!(-l)q-j~rj

p!(q - j)!j!
O!
(-l)P-j (p+q)!(p+q-j)!q'r j
(p + q)!
(q _ j)!j!
(-l)P(p+q)!,B.
So,
(-l)p+q-j [(tk - S)q(S - tk-dPj(p+q-j)
IS=tk_l
= (-l)P(p + q)!,Bj.
Theorem 2.2.1 is proved.

Note that relation (2.51) is called Taylor's decomposition of function v(t) on
two points.
Now, we will consider applications of Taylor's decomposition of function on
two points. From (2.51) it is clear that for the approximate solution of problem
(2.1) it is necessary to find yi:~) for any j, 1 :S j :S p and yi:~-il for any j,
1 :S j :S q. Using the equation
y'(t)
-a(t)y(t)
+ f(t),
we obtain
(2.53)
where
a1(t) = -a(t), h(t) = f(t),

aj(t) = aJ'2 1(t) - aj-1(t)a(t),fj(t)
{
for any j, 2 :S j :S p.
= fj2 1(t) + aj-1(t)f(t),
(2.54)
Replacing y(j) (t) by (2.53) and neglecting the last term, we obtain the single-step
difference schemes of (p + q)-order of accuracy for the approximate solution of
problem (2.1)
P
j=l
j=l
Uk -- Uk-1
a
(tk-1 ) r j-1 Uk-1 = <Pkp,q ,
- + '"'
LJajaj ()
tk r j-1 Uk - '"'
LJ!Jjaj
r
P
(2.55)
j
j
<P~,q = -l:aj!J(tk)r - 1 + l:,Bj!J(tk-1)r -1, 1:S k:S N,uo = Yo
j=l
j=l
Now let us give some examples for the constructed difference schemes.
In the case p + q = 1 from formulas (2.52), (2.54) and (2.55) it follows that
Uk - Uk-1
- - - + a(tk-d u k-1 = f(tk-d, 1 :S k :S N, Uo = Yo
r
(an explicit Euler's difference scheme of first order of accuracy for the initial-value
Uk - Uk-1
- + a(tk)uk =
r
f(tk), 1 :S k :S N, Uo = Yo
(an implicit Euler's difference scheme of first order of accuracy for the initial-value
problem (2.1)).
34
In the case p + q = 2 from formulas (2.52), (2.54) and (2.55) it follows that
Uk 1
Uk -T -
+ (a(tk-l) + ~(a(!)(tk-d -
a2(tk_l)))Uk_l
f(tk-l) + ~(f(')(tk-d - a(tk-l)f(tk-l)),l ::; k::; N,uo = Yo

(an explicit difference scheme of second order of accuracy for the initial-value
problem (2.1)); and that
Uk - Uk-l T
T
+ "2(a(tk)uk + a(tk-duk-d
= 2(f(tk) + f(tk-d), 1::; k
::; N, Uo
= Yo
(a Crank-Nicolson difference scheme of second order accuracy); and that
Uk-Uk-l
T
=
T( -a (')()
+ (()
a tk +"2
tk + a2( tk ))) Uk
f(tk) - ~(f(')(tk) - a(tk)f(tk))' 1 ::; k::; N,uo = Yo
(an implicit difference scheme of second order of accuracy for the initial-value
problem (2.1)).
Now, we consider again the initial-value problem (2.22). Using the equation
y'(t)
we obtain
-a(t)y(t) + f(t, y(t)),
y(j)(t) = aj(t)y(t) + fj(t, y(t)),
(2.56)
where (see formula (2.26))
>. =
f>..(t, y(t)) = f(t, y(t)),

a\
f>..+l(t, y(t)) = W f(t, y(t))
A
(>.)
+ Ei=l
i
at Xa\ia;J(t, y(t))fi(t, y(t)),
0::;
1,
>. ::; p -
1.
Replacing y(j)(t) by (2.56) and neglecting the last term, we obtain the single-step
difference schemes of (p + q)-order of accuracy for an approximate solution of
problem (2.22)
(2.57)
p
= <P~,q(Uk, uk-d, <P~,q(Uk, uk-d
= -
L (}:j!J(tk, Uk)T
j=l
35
Now, we consider again the initial-value problem (2.30). Using the equation
yt(t) = -A(t)y(t)
+ f(t),
we obtain
where
(2.58)
Replacing y(j)(t) by (2.58) and neglecting the last term, we obtain the single-step
difference schemes of (p + q)-order of accuracy for an approximate solution of
problem (2.30)
Uk - rUk-1
'P~,q = -
A ( ) j-1 Uk +~
~ aj j tk r
j=l
j=l
j=l
~!3
~ j
j=l
A j (tk-1 )r j-1 Uk-1 = 'Pkp,q ,
(2.59)
L a j Fj (tk)r j - 1 + L !3jFj (tk_d rj - 1, 1 :::; k :::; N, Uo = Yo
Note that using Taylor's decomposition on two points, we can extend our discussion to construct the difference schemes of an arbitrary high order of accuracy
for approximate solutions of the initial-value problem for the first order nonlinear
system of ordinary differential equations
y'(t)
+ A(t)y(t) = f(t, y(t)), 0 < t
:::; T,
y(O)
Yo.
It is left to the reader.

Finally, in this chapter we consider two types of arbitrary high order of
accuracy difference schemes generated by an exact difference scheme or Taylor's
decomposition on two points for the numerical solutions of an initial-value problem
for differential equations and systems of differential equations. They are different
types of difference schemes for the same problem. Nevertheless, in the case of
a homogeneous equation or system of differential equations with independent in
t coefficients using the two types of methods above, we have exactly the same
difference schemes.
Chapter 3
Difference Schemes for
Second-Order Differential Equations
In the present chapter we consider the two-step difference schemes generated by
an exact difference scheme or by Taylor's decomposition on three points for the
numerical solutions of an initial-value problem and a boundary-value problem for
second-order differential equations.
3.1
Two-Step Exact Difference Scheme and Its Applications
We consider the initial-value problem of the form
y" (t)
+ a(t)y(t)
= f(t), 0< t ::; T, y(O) = Yo, y'(O) = y~
(3.1)
defined on [0, T]. For the construction of difference schemes we consider again the
uniform grid space
[O,T]r={tk=kT, k=O,l, ... ,N, Nr=T}.

The construction of two-step difference schemes of an arbitrary high order of accuracy for approximate solutions of the initial-value problem (3.1) is based on the
following theorem.
Theorem 3.1.1. Let y(tk) be a solution of problem (3.1) at the grid points t = tk.
Then {y(tknS' is the solution of the initial-value problem for the following secondorder difference equations:
38
Chapter 3. Difference Schemes for Second-Order Differential Equations
(3.2)
Here
I(t s) = eiJ:b(A)dA r1(t s) = _1_ = I(s t)
,
" I (t, s)
"
where
(3.3)
Proof. Using the relations (3.3) we can obviously write the equivalent initial-value
problem for a system of first-order linear differential equations
+ ib(t)y(t)
y: (t)
{
~ v(t), y(O) ~ Yo, y' (0) ~ y;,
(3.4)
v (t) - ib(t)v(t) = f(t).
Integrating these, we at once obtain
y(t)
{ v(t)
e- i J~ b(A)dA yo
= eiJ~
b(A)dA VO
+ J~ e- i J: b(A)dAV(s)ds,
+ J~ eiJ: b(A)dA f(z)dz.
That is,
y(t)
r1(t, O)y(O)
it
it
r1(t, s)/(s, O)dsv(O)
r1(t, s)
I(s, z)f(z)dzds,
39
3.1. Two-Step Exact Difference Scheme and Its Applications
so, using the condition y' (0)
y(t)
+ ib(O)y(O) =
[r 1(t,0) +ib(O)
it
1
t
v(O), we obtain
r 1(t,S)1(S,0)dS] Yo
1- 1(t,s)1(s,0)dsYb +
it
r 1(t,s)
is
1(s,z)j(z)dzds.
By the definition of 1(t, s) and interchanging the order of integration, we obtain

y(t) = 1- 1(t,0)
1t
[1 + ib(O)
1- 1(t, z)
it
it
12(S,0)dS] Yo + r 1(t,0)
12(s, z)dsj(z )dz
u(t)
it
12(s,0)dsYb
(3.5)
+ v(t) + g(t).
Putting t = tk+ 1, t = tk and t = tk-1 into the formula for u(t), we can write
U(tk+1)
rtk+l 2
]
r 1(ik+1,0) [ 1 + ib(O) Jo
1 (s,0)ds Yo
[1 +ib(O) [l
= r 1(tk+1,tk)r 1(tk,0)
tk
12(s,0)ds+
l
1
=1- 1(tk+1,tk) u(tk)+1- 1(ik,0)ib(0)

U(tk-1) = r 1(tk_1,0)
= [(t" 'k-,)r' ('k,
[1 + ib(O)
1:k+ 12(s,0)dS]] Yo
1
t k2+ l ]
tk
1 (s,0)dsyo,
tk
12(S,0)dS] Yo
0) [1 + ib(O) [l' 1'(s, O)ds - l', [2(S, O)d']] Yo
= [(t" ,,-,) [U(tk) -
r' (t" O)ib(O)
l',['(
s, O)dsYo] .
Hence, we obtain the following relation between u(tk+ d, u(tk) and u(tk-1) :
(1:k+ 1 (s, O)dS)

1
(I
tk
tk-l
-1
(I(tk+1, tk)U(tk+1) - U(tk))
r 1(tk, O)ib(O)yo,
12(S,0)dS) -1 (U(tk) _ r 1(ik, tk-1)U(tk-1)) = 1- 1(tk, O)ib(O)yo.
40
Therefore
(3.6)
In a similar manner, using the definition of /(t, s) and the formula for v(t), we get
jt
/2(s, O)ds(I(tk+l, tk)V(tk+d - V(tk))
(3.7)
tk-l
Now, putting t = tk+l, t = tk and t = tk-l into the formula for g(t) and using the
definition of / (t, s), we get
J.:'"
I'(S,O)dS[l' r'(tk,Z) [" 1 (s,z)ds!(z)dz
_l
tk
tk
+I
r1(tk, Z)
k 1
j t _ [2(S,
z)dsf(z)dz
r1(tk, Z)
k I
j t - [2(8,
Z)dSf(Z)dZ]
tk-I
tk
=I
[2(S, O)ds [
tk-I
_l
tk
+!
41
J.:'"
t
io
r1(tk, Z)
tk
t
io
1
k+
[2(S, z)dsf(z)dz
tk
r'(tk,Z)
[2(S, O)ds [
+ l " r'(tk,Z)
['H
12 (S,Z)dS!(Z)dZ]
r1(tk, Z)
tk
I
[2(S, z)dsf(z)dz
tk_1
1.:-, I'(S,Z)ds!(Z)dZ]
Now, using the identity
we obtain
I
I
tk
12(s,O)ds(1(tk+l,tk)g(tk+l) - g(tk))
tk-l
tk + 1
12(s, O)dS(g(tk) - r1(tk, tk-dg(tk-l))
tk
tk
+I
tk-I
[2(S, O)ds
Itk+1
tk
tk 1
+ 2
tk
1 (s,O)ds Itk
tk-I
r1(tk, z)
r1(tk, z)
l
l
tk 1
+
[2(S, z)dsf(z)dz
tk-l
[2(s, z)dsf(z)dz.
So, using the last formula and the identities (3.6) and (3.7) we obtain the following
relation between Y(h+d, y(tk) and y(tk-d:
42
1~:1 /2(s,0)ds 1~k+1 /-l(tk,Z)
+ 1~k+1 /2(s,0)ds 1~:1 r1(ik,z)
Since /2(tk,0):f:.
/2(s,z)dsj(z)dz
1:-1 /2(s,z)dsj(z)dz.
it follows that
1~:1 r
-
= i
l k+1
tk
tk + 1
tk
2
(tk, s)ds(I(tk+l, tk)y(tk+l) - y(tk))
/2(S, tk)ds(y(tk) - /-l(tk, tk-dy(tk-d)
l k+
t
1
r 2(tk, s)ds itk+ /-l(tk, z)
tk_1
tk
Z
+ l~k+1 /2(s, tk)ds 1~:1 r1(tk, z)
/2(S, z)dsj(z)dz
1:-1 /2(s, z)dsj(z)dz,
l::;k::;N-1.
From this it follows (3.2) for any k, 1 ::; k ::; N - 1. The identity (3.2) in the case
k = 1 follows from (3.5). Theorem 3.1.1 is proved.
Let us note that the initial-value problem (3.2) is called the two-step exact
difference scheme for the solution of the initial-value problem (3.1).
Note that for b(t) we have a system of first-order nonlinear differential equations of Riccati type (3.3). Therefore for the smooth a(t) there exist smooth solutions of this system defined on the segment [0, T]. We will choose one of these
smooth solutions b(t) = br(t) + ibi(t),
t ::; T.
(3.2) it is clear that for the approximate solutions of problem (3.1) it is necessary
to approximate the expressions
tk
tk 1
-1 i + /2(s, tk)ds, -1 i r 2(tk, s)ds, r1(tk, tk-l), /(tk+l, tk)
T tk
T tk_1
::;
and
Let us remark that in constructing difference schemes it is important to know how

to construct a right-hand side <P~,q that satisfies
tk
tk 1
k 1
(3.8)
r1(tk, z) Jt + /2(S, z)dsj(z)dz
T tk-I
tk
Z
Z
tk 1
q
tk + /2(S, tk-d ds itk
tk-I r1(tk, z) i tk-I /2(s, z)dsj(z)dz] - <P~,q = O(TP+ )
1 i
:d
2(tk, s)ds i
43
and is sufficiently simple. The choice formula <P~,q is not unique. Using Taylor's
formula, with respect to the function
on the variable z, we obtain
(3.9)
where
3::; m
::; p + q.
(3.10)
Using Taylor's formula, with respect to the function

(3.11)
where
3::; m
::; p
+ q.
(3.12)
Moreover, for the construction of <P~,q we will use (3.9) and (3.11). But, first of all
let us study the approximate formulas for the expressions
44
that will be needed for the construction of difference schemes of a high order of
accuracy for the approximate solution of problem (3.1). First, let us consider the
approximate formulas for the expression
tk
I -1(t k, t k-l ) = e -ift k-l b(s)ds

t
k
e-ift :- 1 b(s)ds = e-ib(tkl T + i jt e-ift'k_l b(z)dz(b(tk) _ b(s))ei(S-tk)b(tk)ds.
tk-l

e-ift'k_l b(Z)dZ(b(tk) _ b(s))ei(S-tk)b(tkl
on the variable s, we obtain

t
e-ift :- 1 b(z)dz
n+l
= e-ib(tk)T + iT(b(tk)
(
+ ~ i (: + I)! ({; ~
+~
_ b(tk_l))e-ib(tk)T
Cn-),(tk-d (b(tk) - b(tk_l))(ib(tk))),e-ib(tk)T
C) b('-') ('k-l)(ib('k))'e-<b(')' ]
+cn(tk-d(b(tk) - b(tk_d)e-ib(tk)T)
m~
t
:- 1
e-ift
b(z)dz = e-ib(tk)T
+ 0(T m +2)
2,
+ iT(b(tk) _
iT 2
+Tb'(tk_l)e-ib(tk)T
b(tk_l))e-ib(tk)T
+ 0(T 3 ),
tk
e - i f tk_l b(z)dz = e- t'b(t k )T
+ O(T 2 ),
45
where
Further, using the last formulas and Pade fractions for the function e- z (see Chapter 2), we can write
r1(h,tk-d = R[T],m-[T]+l(i7b(tk))
(3.13)
+i7(b(h) - b(tk-l))R[T]-1,m-[T](i7b(tk))
+ }; i (::+~)!
~ (~) cn->.(tk-d(b(tk) - b(tk-l))(ib(tk))>'
m~
2,
. 2
I-1(tk, tk-d = R u,2-u(i7b(tk)) + i7(b(tk) - b(tk-d) + t~ b'(tk-l)

+o( 7 3 ), a = 0, 1,2,
r1(tk, tk-l) = R x ,1-x(i7b(tk)) + 0(7 2), X = 0,1.
Second, let us consider the approximate formulas for the expression
t k +1
I(t k+l, t k ) -- ei Jt k b( s )ds
constructed using Pade's rational functions. We have that
46
+~
G) b(A-") (tk+ d(ib(tk) )"Je"U,)' + en (tk+' )(b(tk+d - b(tk))e'bU,)')

+0(rm+2) = eib(tk)r + ir(b(tk+l) _ b(tk))eirb(t k)
m-l
t;
l)n r n+l n ( )
+ ~ i (n+ 1)!
~ Cn->.(tk+l)(b(tk+l) - b(tk))(ib(tk))>'eib(tk)r
(_l)n r n+l n (n)
+ ~ i (n + I)! {; >. Cn->.(tk+l)

X
~ G) b(A-") (tk+,)(ib(tk))"Je"U.)' + o(r
eiftkk+l b(s)ds = eib(tk)r
+ ir(b(tk+l) _ b(tk))eirb(tk) _
eiftt:+l b(s)ds
+'), m 2: 2,
. 2
t~ b'(tk+deib(tklr
+ 0(r3),
eib(tk)r + 0(r2),
where
Further, using the last formulas and Pade fractions for the function e- z , we can
write
I(tk+l,tk) = R[T],m-[TI+l(-rib(tk))
(3.14)
+ir(b(tk+l) - b(tk))R[TI-1,m-[TI( -irb(tk))

m-l (
l)n
n ( )
n+l
+ ~ i (n +r1)! {; ~ cn->.(tk+d(b(tk+l) - b(tk))(ib(tk))>'

.
m . (-1)nrn+l n (n)_
xR[m;-nl_l,m_n_[m;-nl(-trb(tk)) + ~ t (n + I)! {; >. cn->.(tk+d
~ G) b(A-") (tw)(ib(tk))" R[m," ],m-n-[ m," I(-irb(tk)) + o(r m +2),

m22,
I(tk+l, tk)
R a,2-a(-rib(tk)) + ir(b(tk+l) - b(tk)) a
I(tk+l, tk)
t;
. 2
b'(tk+l) + 0(r 3),
= 0,1,2,
Rx,l-x( -rib(tk)) + 0(r 2), x
= 0,1.
47
In the sequel we denote that if n <
Rn,m(x)
(m < 0), then
= Ro,m+n(x)(Rn,m(x) = Rm+n,O(x)), n + m
::: 0.
Third, let us consider approximate formulas for the expression
Using Taylor's formula, we can write

(3.15)
where dn(s) = Dn (s)I 2 (s, tk),
Dn(S) = 0, n < 0,
{ Do(s) = 1, Dn+l(s)
= D~(s)
+ 2ib(s)D n(s), 0::; n.
Fourth, let us consider approximate formulas for the expression

(3.16)
where dn(s) = Dn (s)I- 2 (s, tk). Note that dn(tk) = dn(tk) for any n.
Now, using formulas (3.9), (3.10), (3.11), (3.12), (3.13), (3.14), (3.15), (3.16)
and neglecting the last small terms o(r p+q), 'P~,q can be defined by the formula
fl {;
( - l)m T m-I
p,q _ p+q-3 m ( m ) (.\)
'Pk >. f (tk+l) (m + I)!
p+q-m-3
X
f='o
(-1) T d (t )Bn,p+q-m-n(t )
(n + I)! n k m-.\
k+1
48
p+q-3
L L
m
m=l A=O
X(
p+q-3
)1
p+q-m-1.
+
X(
l; {;
p+q-3
p+q-2
)'
p+q-m.
d-
f(A)(tk_d (m
d-
( )B-p+q-m-22(
p+q-m-2 tk
+ I)!
m-A
'
f(A)(tk-d (m
+ I)!
tk-l
p+q-m-l
()B-p+q-m-ll()
tk m-A
' tk-l
P+q-l(p + q - 1)
A
(A)
r P+-2
q
(tk-l) (p + q)!
where
I;+Ol,k
1, I;+\'k = R",l-,,( -rib(tk)), K = 0,1,

2
n,2 = Ra,2-a (
. ())
. (
) - b(tk ) )
Ik+1,k
-nb
tk + 2r
b(
tk+l
- .2 r b'(tk+l)
2
'
=
(J
= 0,1,2,
I;il'k = Rq_[mtnj,p_m_n+[mtnj( -rib(tk))
+ ir(b(tk+l) - b(tk))
x Rq_1_[mtn],p-l-m-n+[mtnj (-rib(tk))
+
X (b(tk+l)
f;
p+q-m-n-2 (-1 )~r~+l ~ (~)

i (~+ I)!
A Z~-A(tk+d
- b(tk))(ib(tk))A Rq_l_[m+;H],p_l_m_n_~+[m+;H] (-rib(tk))
f;
p+q-m-n-l (-l)~r~+l ~ (~)

i (~+ I)! {; A Z~-A(tk+l)
49
50
~ C) bIA-"1 (tk+ d(ib(tk" R,_I_[ m';"],p_m_n_,+[m,;" [( -rib(tk),

j
? 3,
and
3::; m
::; p + q,
J;,'~_l = 1, J;,'1-1 = R x ,l-x(rib(tk)),}( = 0,1,
2
n2
., (
( ) ) .r b'(tk_1)
J k:k- 1 = R a,2-a(nb(tk)) +zr b(tk)-btk-1 +z
2
,a=0,1,2,
J;:1-1 = Rq_[mtnj,p_m_n+[mtnj(irb(tk)) + ir(b(td - b(h-1))
~ C) blA-" I (tk- d (ib( tk))" R,-1-[.,;', [,p-m-n-'+[ .,;"] (irb(tk)), j 2: 3,
Applying the identity dn(tk) = dn(tk) for any n, we can write

(3.17)
x ((_l)m+n Bn,p+q-m-n(t
m-,),
k+1 )f(,),) (t k+1 ) + jjn,p+q-m-n(t
m-,),
k-1 )f(,),) (t k-1 ))
51
X((_I)P+q-1BP+q-m-1,1(t
m-A
k+1 )f(A)(t k+1 ) + BP+q-m-1,1(t.
m-A
k-1 )f(A)(t k-1 ))
f t;
p+q-3
+
X
()
p+q-1
(m + I)! (p +: - m + I)! dp+q-m(tk)
((-I)p+q BP+q-m,O(t
m-A
k+1 )f(A) (t k+1 ) + BP+q-m,O(t
m-A
k-1 )f(A) (t k-1 ))
(+ 2)
+ f;
1p+q-2
(p
+q-
I)!
p+q-3+n
r(n + I)! dn(tk)
n,2-n (t
p+q-2-A (t k+1 )f(A) (t k+1 ) + B p+q-2-A
k-1 )f(A) (t k-1 ))
x ((_I)p+q+n-2 B n ,2-n
(+ 1) 1
+ f;
1p+q-1
(p+q)!
p+q-2+n
r(n+ I)! dn(td
x ((_I)p+q-1+n Bn,l-n
p+q-1-A (t k+1 )f(A)(t k+1 ) + Bn,l-n
p+q-1-A (t k-1 )f(A)(t k-1 ))
f; ,\
p+q ( P + q)
r P+q -1
(p + q + I)! dO(tk)
x (( -1)p+q B~~q_A (tk+1 )f(A) (tk+d
+ B~~q_A (tk-1 )f(A) (tk-d).
Now, using formulas (3.13), (3.14), (3.15) and (3.16), we can write
1
2'
r
(ljt
r
12(s, tk)dsl- 1(tk, tk-1) - 1)
(3.18)
tk
~ :' ('%;,' (m': l),dm(t,)[R'-I'l'I,'+I-m+ I'l'I(i'b(t,))

+ir(b(tk) - b(tk-d)Rq-1-[~J,p-m+[~)(irb(tk))
+i P+%,-l
(::+~)! ~ (~) Cn-A(tk-d(b(tk) _ b(tk_1))(ib(tk))A
xRq_1_[ffitnj,p_m_n+[mrJ(irb(tk))
+i
p+q-m
L (: + I)! A=l
L
n+1
n=l
Cn-A(tk-1)
~ G) b('-') (t'-I)( ib(t,))"R'-lmi" l,p-m-n+lmi" I(i,b(t,))]
r P+q - 1
+ (p + q)!dp+q-1(tk)[RO",2-0"(rib(tk))
r p +q
+ ir(b(tk)
r P+ q + 1
- b(tk-d)
r2
+ i2
b'(tk-dJ
+ (p+q+l.),dp+q(tk)Rx ' 1-x(rib(tk)) + (p+q+2.),dp+q+1(tk ) - 1) + o(r p+q),

(1
= 0,1,2, x = 0, 1;
52
(3.19)
1 p+q-2 (_1)m 7m
72
(fa (m + I)! dm(tk)[Rq-[';'],p-m+[';']+l (-7ib(tk)) + i7(b(tk+l) - b(tk))
.
xRq-1-[';'I,p-m+[';'J(-z7b(ik))
E(n)_
?;
p+q-m-l . (-1 )n 7n+l n

z (n + I)!
?;
p+q-m (_1)n 7 n+l n

+
i (n + I)! {;
cn->.(tk+d
- b(tk))(ib(tk))>' R q_ 1_[ m;n],p_m_n+[ m;n] (-i7b( tk))
X (b(tk+l)
oX
(n) cn->.(tk+d
oX
~ C) b('-") (tk+1)(ib(tk))" R,-l mi'],p-m-n+l mi'] ( -irb(tk))1
(-1 )p+q-l7P+q- 1
72
(p)'+
dp+q- 1(tk) [R q 2-q (-7ib(tk) )+i7(b(tk+d-b(tk)) -i-b' (tk+dl
q.'
2
(-1 )P+q7P+q
+ (p+q+1.)1 dp+q(tk)R x ' l-x( -nb(td)

+
(-1 )p+q+ 17 P+q+1

(p+q+ 2)'. dp+q+1(tk) - 1) + O(7 P+Q ),
1jt
1
2"(2-7
7
fl
p+q+l
=
1jt
2(tk,S)ds-7
tk-l
12 (s,tk)ds)
(3.20)
tk
m-2
q
(~+ l)!dm(t k)((_l)m+l -1) + O(7 P+ );
1(7,0)ib(0)!
~ ib(O) C~: (mr:
= 0,1,2, x = 0,1;
(J
k
ia
1 )! dm(O)
12 (s,0)ds
+ !(I-l(7,0) -1)
(3.21)
[11,->-[ ';'],p-m+l';'](irb(r)) + ir(b(r) - b(O))
X R q- 2 -[';'],p-l-m+[';'1(i7b(7))
+i
p+q-m-3
n+l
(: + I)!
E~
n ( )
cn->.(O)
53
x (b( r) - b(O))(ib(r))A R q_2_[ mtn],p-1-m-n+[ mtn I (irb(r))

n+1
p+q-m-2
+i
X
(: + 1)!
n ( )
f;
Cn-A(O)
~ C) b('-) (O)(;b(T))" R,_l_[ -to ),,-l-m-n+[mt") (iTb(T))] )
+ib(O) ( rp+q-32),dp+q_3(0) [Ra2-a(rib(r))

p+q- . '
+ ir(b(r) -
b(O))
+ i r2 b'(O)]
r P+q - 2
r P+q - 1
+ib(O) (
),dp+q- 2(0)R x ,1-X(rib(r)) + ib(O) (
),dp+Q- 1(0)
p+q-1.
p+q.
+~(RQ,p(irb(r))
p+Q-2
+ ir(b(r) - b(0))RQ- 1,p-1(irb(r))
n ( )
n+1
+i ~ (: + 1)! ~ ~ Cn-A (O)(b(r) - b(0))(ib(r))ARq_1_[ ~],p-l-n+[~1 (irb(r))
(J
= 0,1,2, x = 0, 1,p + q 2
liT
r1(r,0)ib(0)r
3,
1
]2(S, O)ds + -(1-1(r,
0) - 1)
= ib(O)do(O)R x ,l-x(rib(r)) + ib(O) 2d1 (0)

1
+~(Rq,p(irb(r))
+ ir(b(r) - b(O)) + i
liT
r 1(r,0)ib(0)r
= ib(O)do(O) +
I:
m=O
2 b'(0) -
1)
+ o(r 2),p + q = 2;
1
]2(s, O)ds + -(1-1(r,
0) - 1)
~(Rq,p(irb(r))
r1(r,0)~ r
r Jo
p+q-4
r2
- 1) + o(r),p + q = 1;
]2(s,0)ds
r
(m + 1)!dm(0)[R q- 1-[T],p-m+[TI(irb(r))
(3.22)
54
+ir(b(r) - b(0))Rq- 2-ITJ,p-l-m+ITj(irb(r))
+i
P+%,-3
(:::)! t, (~)
c,._,(O)(b(r) - b(O))
x (ib(r))>' R q_2_[ mrJ,p_l_m_n+lmtnj(irb( r))
p+q-m-2
~)
>'-1 (
/L=O
r P+q -
n ( )
L (: + 1)! >'=1
L
+i
x
n+l
n=1
cn->.(O)
b(>'-/L) (O)(ib(r))/L Rq_l_lmtnj,p_l_m_n+lmtnj(irb(r))]
r2
2),d p+q- 3 (0)[R a,2-a(rib(r)) + ir(b(r) - b(O)) + i-b'(O)]
p+q- .
2
+(
r P+q - 1
p q
),d p+q- 2 (0)R x ,I-X(rib(r)) + (
),d p+q- 1(0) + o(r + ),
p+q-1.
p+q.
+(
r P+q -
(J
= 0,1,2, x = 0, 1,p + q 2 3,
(3.23)
where
11
1(r, z)
1
T
[2(s, z)dsj(z)dz
= 0 + o(r),p + q = 1.
Now, neglecting the last terms in formulas (3.18), (3.19), (3.20), (3.21), (3.22),
(3.23) and using formula (3.17), we obtain the difference schemes

P
= II1
rk ,q , 1 <
- k <
- N - 1, Uo =
~(Ul
r
where
1 p+q-2
al,k
r2
55
Yo ,
(3.24)
uo) = ao,oYo + ao,lY'(O) + <Pb,q,
(1; (mr+ l),dm(tk)[Rq_[';'-J,P+l-m+[';'-I(irb(t k))

m
+ir(b(tk) - b(tk-l))Rq-1-[,;,-J,p-m+[,;,-I(irb(tk))
C)
b('-") (tk-})(ib(tk))" R,_[ mi" ],p-m-n+( mi"](irb(tk1 11
r P+q - 1
+ (p+ q)!dp+q-l(tk)[R,r,2-0'(rib(tk))
r2
+ ir(b(tk) - b(tk-l)) + i 2 b'(tk- 1)]
r p+q
+ (p + q + I)! dp+q(tk)Rx,l-x(rib(tk))
(j
r P+q+1
+ (p + q + 2)! dp+q+1(tk) -
1),
= 0,1,2, x = 0,1;
(1;
1 p+q-2 (_l)m r m
.
(m + I)! dm(tk)[Rq_[';'-J,p-m+[';'-]+l( -nb(tk))
a3,k = r2
+ir(b(tk+d - b(tk))Rq-1-[';'-I,p-m+[,;,-1( -irb(tk))
+
X
(b(tk+l) - b(tk))(ib(tk))A Rq-1-I"'tnJ,p-m-n+[",;nl( -irbh))
+
X
(n)
?;
p+q-m-l (_l)n r n+l n

i (n + I)! {; A Cn-A(tk+l)
C)
?;
f;
p+q-m (_l)n r n+l n

i (n + I)!
(n)
A cn-A(tk+l)
b('-") (tk+ 1 )(ib(t,))"R,_ [mi" },p-m-n+ [mi" I( -irb(tk))]
(-1 )p+q-l r P+q- 1

r2
(p + q)!
dp+q- 1(tk)[RO',2-0'( -rib(tk) )+ir(b(tk+l )-b(tk)) -i 2 b' (tk+l)]
56
(J
fl
p+q+l
a2,k =
= 0,1,2, x = 0,1;
m-2
(~+ I)! dm(tk)(( _l)m+l - 1);
p+q-4
m
T
ao,o = ib(O)('; (m + 1)!dm(0)[R q- 1-[T],p-m+[TI(iTb(T))
+iT(b(T) - b(0))Rq- 2-[T],p-l-m+[TI(iTb(T))
(J
= 0,1,2, x = 0, 1,p + q ~ 3;
57

1
+~(Rq,p(iTb(T))
+ iT(b(T) - b(O)) + i
ao,o = ib(O)do(O)
fo
p+q-4
aO,l =
2 b'(0) -
1),p + q = 2;
+ ~(Rq,p(iTb(T)) -1),p+ q =
1;
T
(m + 1),dm(0)[R q- 1-[T],p-m+[TI(iTb(T)) + iT(b(T) - b(O))
XR q- 2-[T],p-l-m+[TI(iTb(T))
X (b(T)
+i
p+q-m-3
n ( )
(: + I)! {; ~ cn-,$O)
n+l
- b(O))(ib(T)),\ R q_2_[ mr],p-l-m-n+[mtnj (iTb(T))

p+q-m-2
+i
x
n+l
(:+ I)!
n ( )
f;
cn-,\(O)
~ C) b(A-") (0)(ib(7}}" Rq_1_1m,"I,,_1_m_n+( ","I (;7b(7))]

TP+q- 3
iT 2
+ (p+q- 2),dp+q(0)[Ru 2-u(Tib(T)) + iT(b(T) - b(O)) + -b'(O)]

.3 '
2
TP+q-2
TP+q- 1
p
q
+ (p+q _ 1),d
+ - 2(0)Rx ,1-X(Tib(T)) + (
),dp+q-1(0),
.
p+q.
a
aO,l
= 0, 1, 2, x = 0, 1, P + q ~ 3;
.
do(O)Rx,l-x( nb(T))
+ 2d1 (0), p + q =
2;
aO,l=l,p+q=l;
O'P+q=I'
{nP,q - p+q-l m ( )
""0 '" '"
m BOO ()f('$ ( ) (_l)mrm 2
{ m~l
,\~o A
m-,\ T
T
(m+l)!' :s p + q.
Thus, we constructed the difference schemes of (p + q)-order of accuracy for the
approximate solution of the initial-value problem (3.1). Finally, let us give some
examples of the simplest difference schemes.
In the case p + q = 1 from formulas (3.17), (3.18), (3.19), (3.20), (3.21),
(3.22), (3.23) and (3.24) it follows that
1(
T2 Uk+l - 2Uk
b2 (tk)
1 ,
+ uk-d + (-2- + iT- (b(tk) - b(tk-d) + i 2b (tk-l)

+ ~! (6ib'(tk) - 4a(tk))Uk-l
58
b2 (tk)
+( -2ib (tk) + 3a(tk))Uk + (-2- + iT- (b(tk+d - b(tk))
-i~b'(tk+1) + ;, (6ib'(tk) 1
4a(tk))Uk+1 =
~ k ~ N -
~(f(tk+d + f(tk-d),
1, Uo = Yo,
!(Ul - uo) = i(b(O) - b(T))Y(O)

T
+ y'(O)
(a difference scheme of first order of accuracy for an initial-value problem (3.1)

generated by the first order of accuracy explicit difference scheme).
In the case p + q = 2 from formulas (3.17), (3.18), (3.19), (3.20), (3.21),
(3.22), (3.23) and (3.24) it follows that
b2 (tk)
.Tb 3 (tk)
1
T2 (Uk+1 - 2Uk + Uk-I) + (-2- - z-6- + i-;(b(tk) - b(tk-d)
1
+~b(tk-d(b(tk) -
b(tk-d) -
~(b(tk) -
b(tk-d)b(tk) + i~b'(tk-l)(l - iTb(tk))

2
2T (tk-l )b'( tk-l )+z-3

. T (b"( tk-l )+2b'( tk-l ).zb (tk )) +zb
. (tk )( - Tb
+,b
- -(tk)
+ -iT,(
b tk-l ))
1
3 . .
+~(2ib'(tk) -
4b 2 (tk))(1- iTb(tk)) + ;, (2ib"(tk) -12ib(tk)b'(tk) - 8ib 3 (tk))))Uk_l

,
b2 (tk). Tb 3 (tk)
+( -2ib (tk) + 3a(tk))Uk + (-2- + z-6- + -;(b(tk+d - b(tk))
-~b(tk+d(b(tk+d
- b(tk)) + ~(b(tk+d - b(tk))b(tk) -
i~b'(tk+1)(l + iTb(tk))
2
2T b(tk+1 )b'( tk+1 ) +i T! (b"( tk+1 )+2b'( tk+l )ib ())

Tb (tk) iT, (tk+d)
+3T
tk -ib ()
tk (--2--2b
3
+~(2ib'(tk) =
4b 2 (tk))(1 +iTb(tk)) -
~(2ib"(tk) -12ib(tk)b'(tk) -
8ib3 (tk))))Uk+l
~(f(tk+l) + f(tk-d) + ~iTb(tk)f(tk+l) - ~iTb(tk)f(tk-l)

+i(- j'(tk+d + j'(tk-d), 1 ~
~(Ul
- uo) = (Tb(O)(b(T) - b(O))
~N-
1, Uo = Yo,
+ ~(ib'(O) - b2 (T)))YO
+(1 + iT(b(O) - b(T)))Y'(O) + ~f(T)

(a difference scheme of second order of accuracy for an initial-value problem (3.1)
generated by the second order of accuracy explicit difference scheme).
59
Now, we consider the boundary-value problem of the form
-y" (t)
+ a(t)y(t)
= f(t), 0< t ~ T, y(o) = Yo, y(T) = Yr
(3.25)
defined on [0, T]. But for existence of a unique solution of the boundary-value
problem (3.25) we need some estimate from below of a(t).
For the construction of difference schemes we consider again the uniform grid
space
[0, T]r = {tk = kT, k = 0,1, ... , N, NT = T}.
The construction of the two-step difference schemes of an arbitrary high order of
accuracy for the approximate solutions of the boundary-value problem (3.25) is
based on the following theorem.
Theorem 3.1.2. Let y(tk) be a solution of problem (3.25) at the grid points t = tk.
Then {y(tknS" is the solution of the boundary-value problem for the second-order
difference equations
+ l~k+l J 2 (tk+l,s)ds 1~:1

1
Here
{1:-1
J(tk,s)J(z,s)ds}f(z)dz],
k ~ N -1,y(O) = yo,y(T) = Yr.
J(t, s) = e- J; b()")dJ..,
where b(t) is a solution of the equation

b2 (t) - b'(t) = a(t),
and the following estimate holds:
b(t) ~ bo > 0, ~ t ~ T.
(3.27)
(3.28)
60
Proof. Using relation (3.27), we can obviously write the equivalent boundary-value
problem for a system of first-order linear differential equations
y' (:) + b(t)y(t) = v(t), yeO) ~ Yo, YeT) ~ !IT,

{
-v (t) + b(t)v(t) = f(t).
y(t)
e- I~ b(>')d>'yo
+ I; e- I: b(>')d>'v(s)ds,
{ v(t) = e- I,r b(>')d>'v(T) + It e- It b(>')d>' f(z)dz.

That is,
y(t) = J(t,O)yO + I; J(t,s)v(s)ds,

{
v(t) = J(T, t)v(T)
+ It J(z, t)f(z)dz.
From these formulas and the condition y(T) = YT it follows that
y(T) = J(T, O)YO +
lT
J(T, s)
iT
J2(T, s)dsv(T)
iT J(z, s)f(z)dz.
Since IoT e- 2Is b(>')d>'ds t=- 0, then

T
- [
( [ J(T,,)J(Z,S)dS) f(z)dz].
So, we have that
v(t)
~ J(T, t) (iT J'(T, S)d')
-iT (J.T
-I
[,(T) - J(T, O)Yo
J(T,S)J(Z,S)d.s) f(z)dz] +
iT
J(z,t)f(z)dz

and
y(t)
~ J(t, O)yo +
l'
-iT (iT
61
J(t, ')[J(T, ,) ( [ ' J'(T, T)dT) -1 [y(T) - J(T,O)yo
J(T,T)J(Z,T)dT) !(z)dz] +
iT
J(z,s)f(z)dzlds.
By an interchange of the order of integration, we obtain
T
y(t) = (iT J 2(T,S)ds)-1[{i J 2(T,s)dsJ(t,O)yO +
-it J(t, s)J(T, s)ds x i
t
{l
it
J(t,s)J(T,s)dsy(T)}
J 2(z,s)ds}f(z)dz]
+ i {i J(t,s)J(z,s)ds}f(z)dz
+ iT{it J(t, s)J(z, s)ds}f(z)dz = u(t) + w(t) + g(t),
(3.29)
where
u(t)
(iT J2(T, S)dS)-l iT J2(T, s)dsJ(t, O)Yo,
w(t) = (iT J2(T, s)dst 1
g(t)
it
J(t, s)J(T, s)dsy(T),
-(iT J2(T, S)dS)-l
it
J(t, s)J(T, s)ds
x iT {i J 2(z,s)ds}j(z)dz
+ i {i J(t,s)J(z,s)ds}j(z)dz + i
Putting t
{l
t
J(t,s)J(z,s)ds}j(z)dz.
= tk+l, t = tk and t = tk-l into the formula for u(t), we can write
62
t k +1
-(I
tk
(jt
J 2(tk+l,S)ds)-1(U(tk+d - J(tk+l,tk)U(tk))
J2(T, S)dS)-l J2(T, tkH)J(tk+l, O))YO'
J2(tk, S)dS)-l(U(tk) - J(tk, tk-l)U(tk-I))
tk-l
Therefore,
or
-J(tk+l' tkH1T J2(T, S)dS)-l J2(T, tkH)J(tk+l, O))YO'
tk
63
J 2(tk, S)dSJ 2(tk+l' tk)]U(tk)
tk-I
t k +1
J 2(tk+l' S)dSJ(tk, tk-l)U(tk-I) = O.
(3.30)
tk
In a similar manner, using the definition of J(t, s) and the formula for w(t), we
can obtain
tk
J2(tk, s)dsJ 2(tk+l' tk)]w(ik)
tk_1
tk +1
J 2(tk+l' s)dsJ(tk, tk-l)W(tk-d = O.
(3.31 )
tk
Now, putting t = tk+l, t = tk and t = tk-l into the formula for g(t) and using the
definition of J(t, s) we can obtain
64
x(
tk
lo
lz
J(tk,s)J(z,s)ds}f(z)dz+
lo
lz
itk
ltk+l J(tk+1, s)J(z, s)ds}f(z)dz)
iTtk {l tk
ltk l
J(tk,s)J(z,s)ds}f(z)dz +
J(tk,s)J(z,s)ds}f(z)dz)
tk +
1
J(tk, s)J(z, s)ds}f(z)dz)
tk+ ltk+l
{
J(tk+1,s)J(z,s)ds}f(z)dz)
tk 0
tk
x(
iTtk l tk
J 2 (tk+1,s)dsJ(tk,tk_d(
t
tk io
+i
t rt
itk_l io
k
J(tk-l,S)J(z,s)ds}f(z)dz)
k
-
J(tk_l,S)J(z,s)ds}f(z)dz
J(tk_l,S)J(z,s)ds}f(z)dz)
65
So, using the last formula and the identities (3.30) and (3.31) we obtain the following relation between y(tk+l), y(tk) and y(tk-d:
tk 2
Itk+l J2(tk+l' s)dsJ(tk' tk-dy(tk-d
J (tk, s)dsJ 2(tk+l' tk)]y(tk) +
Itk-l
tk
tk
Itk+l {Jt k+1J(tk+l,S)J(z,s)ds}f(z)dz
=J2(tk, s)dsJ(tk+l' tk)
I
tk
z
tk+1 2
Itk
- Itk J (tk+l,S)ds
tk_l J(tk,s)J(z,s)ds}f(z)dz.
tk-l
tk-l{
From this it follows (3.26) for any k, 1 ::; k ::; N - 1. Theorem 3.1.2 is proved.
Let us note that the boundary-value problem (3.26) is called the two-step
exact difference scheme for solution of the boundary-value problem (3.25).
Note that for b(t) we have the Riccati differential equation (3.27). Therefore
for the smooth a(t) there exists a smooth positive solution of this differential
equation defined on the segment [0, T].
(3.26) it is clear that for the approximate solutions of the problem (3.25) it is
and
tk+1 rtk+l
jtk jZ
jtk {Jz J(tk+l,s)J(z,s)ds}f(z)dz, tk-l{ tk-l J(tk,s)J(z,s)ds}f(z)dz.
to construct a right-hand side <P~,q that satisfies
(3.32)
and is sufficiently simple. The choice formula

formula, with respect to the function
tk +
Jr
z
<P~,q
is not unique. Using Taylor's
J(tk+l, s)J(z, s)ds}f(z)
66

1
2"
k
lt
+ {
tk
tk
J(tk+l,s)J(z,s)ds}f(z)dz
(3.33)
where
(3.34)
lZ
J(tk,s)J(z,s)ds}f(z)
tk-l

1
2"
T
it {itr
k
tk-l
J(tk' s)J(z, s)ds}f(z)dz
tk-l
(3.35)
where
Moreover, for the construction of rp~,q we will use (3.33) and (3.35). But, first of
all let us study the approximate formulas for the expressions
that will be needed further for the construction of difference schemes of a high
order of accuracy for the approximate solution of problem (3.25).
Let us consider the approximate formulas for the expression
J(t k, t k-l ) = e
J:
k-l
b(s)ds
67
m-1
n+1
= e-b(tk)T + r(b(tk) - b(tk_d)e-b(tk)T + ~ (:+ 1)!

x
m
+ '"
to (~)
n+1
Cn-,\(tk-1)(b(tk) - b(tk_1))b'\(tk)e-b(tk)T
,\-1
f:'o
r
, "'cn-,\(tk-d ' "
~ (n+1). ~
(A) b{'\-J.L) (tk_dbJ.L(tk)e-b(tk)T

J1-
+o(rm +2 ), m 2: 2,
e- f"kk_ 1b(z)dz = e-b(tk)T
+ r(b(tk) _
b(tk_1))e-b(tk)T
+r
2
e- f,':_1 b(z)dz = e-b(tk)T + 0(r 2),
b'(tk_de-b(tk)T
+ 0(r3),
where
can write
(3.37)
68
XR[m;nj_l,m_n_[m;nj(Tb(tk))
A-I (
J.L=O
_J'k
~)
b(z)dz
'k-l
n ( )
n+l
+ ~ (:+ I)! {; ~
b(A-J.L)(tk-lW"(tk)R[m;nj,m_n_r m;nJ(Tb(tk))
T
Cn-A(tk-d
+ 0(Tm+2), m 2: 2,
= R<7,2-<7(Tb(t k ))+T(b(t k )-b(t k-d)+2"b (tk_l)+0(T 3 ),a = 0, 1,2,
Jt'L, b(z)dz =
e-
Therefore
e-
R x ,l-x(Tb(tk)) + 0(T 2), X = 0,1.
J'k+l b(z )dz

'k
= R[TJ,m-[TJ+l(Tb(tk+d)
(3.38)
+T(b(tk+l) - b(tk))R[TJ-l,m-[T](Tb(tk+d)
I: (::+~)! t
1
n=1
A=1
(~) cn- A(tk)(b(tk+l) - b(tk))bA(tk+l)

m
XR r m;nJ_l,m_n_r m;nJ(Tb(tk+d) + ~
x~
A-I (
e-
J k+l
tk
~)
b(z)dz
n ( )
n+l
(:+ I)! ( ; ~
Cn-A(tk)
b(A-J.L)(tk)bJ.L(tk+l)R[m;nj,m_n_[m;nj(Tb(tk+l)) + 0(T m+2), m 2: 2,

T
= R<7,2-<7(Tb(tk+l))+T(b(tk+d-b(tk))+"2b (tk)+O(T ), a = 0, 1, 2,
- J'k+l b(z)dz
'k
= R x ,l-x(Tb(tk+l))+O(T ),x=O,1.
Moreover, we can obtain
e-
2J'k+ l b(z)dz
tk
(3.39)
= R[TJ,m-[T]+l(2Tb(tk+l))
+2T(b(tk+d - b(tk))R[Tl- 1 ,m-[TJ(2Tb(tk+l))
+ ~1
(::+~)!
t, (~)
1 A
Cn-A(tk)(b(tk+l) - b(tk))2A+ b (tk+l)
XR[m;nJ_l,m_n_[m;nJ(2Tb(tk+d)
I: (~)
J.L=O
+~
(::+~)!
t, (~)
Cn-A(tk)
b(A-J.L) (tk)2J.L+lbJ.L(tk+d R 1m;nJ,m_n_l m;nj (2Tb(tk+d)

+0(Tm+2), m 2: 2,
69

'k+l ( )
e- 2J'k
z dz = R CT ,2-CT(2rb(tk+l)) + 2r(b(tk+d - b(tk)) + r 2b'(tk)

(j
e _2J'k+l
'k
b(z)dz
+ 0(r3),
= 0,1,2,
))
= R x ,1-x(2rb(tk+l
+ 0(2
r ), x = 0,1,
where
Now, let us consider the approximate formulas for the expression

(3.40)
where dn(s) = Dn(S)}2(tk+l, s),
Dn(S) = 0, n < 0,
{ Do(s) = 1,Dn+1(s)
Therefore
= D~(s) + 2b(s)Dn(s), 0:::; n.
1jtk
m (_l)n r n
j2(tk,S)ds=
1),dn(tk)+0(rm+1).
r tk_l
n=O n + .
L (
(3.41)
Now, using formulas (3.33), (3.34), (3.35), (3.36), (3.37), (3.38), (3.39) and
(3.40) and neglecting the last small terms o(r p+q), <P~,q can be defined by the
following formula:
<P~,q =
fl (;
p+q-3
(m)
(3.42)
A j(>)(tk+l)(3m->.(tk+l)
m m-l p+q-m-3 ( l)n n

x ( - l) r
'"
r d (t )jn,p+q-m-n
(m+1)!
~
(n+1)! n k k+l,k
fl (; 7
p+q-3
j(>') (tk+d(3m->. (tk+d (m + I)!
( - l) p+q r p+q-3 d
x (p+q-m- 1)'.
p+q-on-2
tk
)jP+q-m-n-22
k+l k
'
'
70
fl E
p+q-3 m
p+q-m-3
"
LJ
n=O
m)
(-1)
(.x)
m-l
(tk-I) (m + I)!
n
T
) -n p+q-m-n(
(n+ 1)'. dn tk+l Brr:-.x
X jjp+q-m-2,2(t
)
m-.x
k-l
tk-l
71
fl ~
p+q-3
f(A}(tk_l) (m + I)!
_
( _1)p+q-m-1 7 P+q-2
P+q-m-ll(t
dp+q-m-I (t k+1 )Bm-A
'
k-I )
p+q-m.
x -'---'-(:-------:)-=-,p+q-3
+
L L
A=O
f(A)(
m=1
f(A}(tk_d (m + I)!
q
(-1)p+q-m 7 P+ -I
(
)B-p+Q-m,O(
)
X (
)' dp+Q-m tk+1
m-A
tk-I
p+q-m+1.
(2)
p+q-2
""' p + q A
A=O
+ ~
+
p+q-I (
""'
~
A=O
p+q-1
A
h-I (
(A)(
p+q-3
""'
l)n n
7
+ _ I)! ~ (n + I)!
q
n=O
d (
)jjn 2-n (
)
n tk+1 p.tq-2-A tk-I
p+q-2 I ( l)n n
""' 7
(
)-nl-n (
)
tk-I ( + )! ~ (n + I)! dn h+1 Bp.tq-I-A tk-I
P q n=O
)7
p+q ( )
p+q-I
P+q
(A)
7
-0
A
f (tk-I) (p + q + 1)!do(tk+dBp+q_A (tk-I),
+ ""'
~
where
= R x ,l-x(7b(tk+I)), x = 0,1,
J;.tI,k
1, J;.t\,k
J;;I,k
R a,2-a(7b(tk+I))+7(b(tk+I)-b(tk))
7 2b'(tk)
2
,0-=0,1,2,
J;:;l,k = R q_[mt n j,p_m_n+[mt n j(7b(tk+l)) + 7(b(tk+l) - b(tk))

xRq_ 1 _["'i n j,p_l_m_n+["'i n J(7b(tk+d)
p+q-m-n-2
~=l
and
~+I
(b(tk+l) - b(tk))bA(tk+1)Rq_l_["'+;H],p_l_m_n_~+[m+;HJ (7b(tk+l))
/H~n-, (;:+:)!
x
(;+ 1)! A=O

L (i) C~-A(tk)
t, (i)
C,-A(tk+.)
~ C) b(A-") (tk)b"(tkH)R,_ '_lm+;H),p_m_n_'+lm+;H)(rb(tk+' )),j "
3,
72
J~'~_l
1, J;~\,k = Rx,l-x(rb(tk)), x = 0,1,
J~'LI
R 17 ,2-17(rb(tk)) + r(b(tk) - b(tk-l)) +
r 2b'(tk_l)
2
,0"
= 0,1,2,
J;:Ll = Rq_[mtnJ,p-m-n+lmrl(rb(tk)) + r(b(tk) - b(tk-I))
X (b(tk)
- b(tk_l))(b(tk))A Rq_1_1m+;HJ,p_l_m_n_{+[m+;Hj (rb(tk))

p+q-m-n-l
+
(;+ I)! L (i) C{-A(tk-I)

{H
{=l
A=l
~ (~) b(A-iL)(tk-I)(b(tk)Y'Rq_1_1m+;HJ,p_m_n_{+[m+;H](rb(h)),j 2:: 3.
iL=O
Now, using formulas (3.37), (3.38), (3.39) and (3.40), we can write
1
1jt
2"(1 - r
tk
+1
2
J (tk+l' s)dsJ(tk' tk-l))
(3.43)
1
p+q-2 (_l)m r m
r 2 (1- , ; (m+ I)! dm(tk+l)[Rq-[TJ,p-m+[T]H(rb(tk))
+r(b(tk) - b(tk-l))Rq-1T]-1,p-m+IT](rb(tk))
~ G)
b(A-") (tk_,)(b(tk))" R'_
mi" ]..-m-""mi' I (rb(tk))]
(_1)p+q- 1r P+q- 1
r2 ,
(
)'
dp+q- 1(h+l)(R17 2-17(rb(tk))+r(b(tk)-b(tk-l))+-b (h-l))
p+q.'
(-1 )p+qr p+q

(-1 )p+q+ 1 r p+q+1
- (p + q + I)! dp+q(tk+l)Rx,l-x(rb(tk)) (p + q + 2)! d p+q+1(tk+l))
+o(r
+q ) , 0"
0 '"
1 2 x = 0, l',
73
(3.44)
fa
p+q-2 (-l)m r m
(m+ I)! dm(tk)[Rq-[~],p-m+[~1+1(2rb(tk+I))
+2r(b(tk+I) - b(tk))Rq_[~)_1,p_m+[~1(2rb(tk+I))
p+q-m
XRq_ImtnJ_l.p_m_n+lmtnJ (2rb(tk+l))
n+l
n=l
(-1 )p+q-l r P+q- 1
(p + q)!
L (: + I)! L en-A(tk)
A=l
2
dp+q- 1(tk)(Ra,2-a(2rb(tk+I)) + 2r(b(tk+d - b(tk)) + r b'(tk))
(-1 )p+qr p+q

(-1 )p+q+ 1 r p+q+1
+ (p + q + I)! dp+q(tk)R x ,1-x(2rb(tk+l)) +
(p + q + 2)! dp+q+1(h))
+o(rp +q ) , a
= 0 1 2 x = 0 l'
'"
"
(3.45)
fa (m
p+q-2 (_l)m r m
= r 2 (1+ I)! dm(tk)[Rq-[~],p-m+[~I+l(rb(tk+l))
1
+r(b(tk+l) - b(tk))Rq_[~]-l,p-m+[~I(rb(tk+I))
p+q-m
XRq_lmtnl_l,p_m_n+lmtnJ (rb(tk+I))
+ ~
n
(:+ I)! {; en-A(tk)
n+l
74
-
(_1)p+q- 1r P+q- 1
r2
(p + q)!
dp+q- 1(tk)(Ra,2-a(rb(tk+l)) + r(b(tk+d - b(tk)) + 2"b'(tk))
(-1 )p+q+ 1 r p+q+1
- (p+q+1)! dp+q(tk)Rx,l-x(rb(tk+l)) (p+q+2)! dp+q+1(tk))
(-1 )p+qr p+q
+o(r p +q ), (j
= 0,1,2,){ = 0, 1.
Now, neglecting the last terms in formulas (3.43), (3.44), (3.45) and using
formula (3.42), we obtain the difference schemes
(3.46)
where
1
al,k = r 2 (1- 2rbh) -
fo (m
p+q-2 (_l)m r m
+ I)! dm(tk+l)[Rq-[T],p-m+[T]+l(rb(tk))
+r(b(tk) - b(tk-d)Rq_[TI-l,p-m+[TJ(rb(tk))
p+q-m
xRq _ 1mtnl-1,p-m-n+lmtnJ (rb(tk)) +
I: (A) C
11=0
J.L
b 'x-I1)(tk_db l1 (tk)R
q-I
m+n
2
n
(:+ I)! {; Cn-,X(tk-l)
n+l
l.p-m-n+1
m+n
2
(rb(tk))]
(_1)p+q- 1r P+q- 1
r2 ,
(p + q)!
dp+q- 1(tk+l)(Ra,2-a(rb(tk)) +r(b(tk) - b(tk-d) + 2"b (tk-d)
(-1 )p+qr p+q
(
1)' dp+q(tk+l)Rx l-x(rb(tk))
p+q+ .
'
(-1 )p+q+ 1 r p+q+1
(
2)' dp+q+l (tk+d), (j = 0,1,2,){ = 0,1,
p+q+ .
1
a2,k
= 2"(-2
fo (m
+ 4rb(tk) +
p+q+l(_l)n r n
2:= (n + 1)'. dn(tk+l)
n=O
p+q-2 (_l)m r m
+
+ I)! dm(tk)[R q-[T],p-m+[TI+l(2rb(tk+l))
75
(-1 )p+q-1rP+q- 1
2
+
(p + q)!
dp+q- 1(tk)(Ra,2-a(2rb(tk+d) + 2r(b(tk+d - b(tk)) + r b'(tk))
(-1 )p+qr p+q
(-1 )p+q+ lr P+q+1
p
q
x
+ (p+q+1)! d + (h)R ,1-x(2rb(tk+d) +
(p+q+2)! dp+q+1(tk)),
(J
= 0,1,2, x = 0,1,
fa (m
p+q-2 (_l)m r m
a3,k = r 2 (1- 2rb(tk) + I)! dm(tk)[Rq-ITJ,p-m+IT]+l(rb(tk+l))
1
+r(b(tk+l) - b(tk))Rq-ITI-l,p-m+[T](rb(tk+l))
~
(..\) b(A-/-I) (tk)b/-l(tk+dR
~ J.t
/-1=0
q-[
"'t n ].p-m-n+[ "'t n J (rb(tk+d)]
(-1 )p+q-1rP+q- 1
(
)'
dp+q- 1(tk)(Ra,2-a(rb(tk+l))
p+q.
r2
+ r(b(tk+l) - b(tk)) + -2 b'(tk))
(-l)p+qr p+q
(_1)p+q+lr P+q+1
- (p + q + I)! dp+q(h)Rx,l-x(rb(tk+d) (p + q + 2)! dp+q+1(tk)),
(J
= 0,1,2, x = 0,1.
Thus, we constructed the difference schemes of (p + q)-order of accuracy for the

approximate solution of the boundary-value problem (3.25). Finally, let us give
examples of the simplest difference schemes.
76
In the case p + q = 1 from formulas (3.42), (3.43), (3.44), (3.45) and (3.46)
it follows that
b2 (tk)
(Uk+l
2Uk
+Uk-l)(l27b(tk))
+
(--2- b(tk+l)b(tk)
2
- 7
-"3 b (tk+d
+(8b2(~k+l) + 2b(tk+l)b(tk) + ~b2(tk)

-7- I(b(tk+l) - b(tk)) + ~b'(tk) + ~b'(tk+I))Uk
+( -
b2(tk+l)
2
- b(tk+db(tk) -
1 ,
1 ,
"3 b (tk) - 2b (tk) - "3 b (tk))Uk+1
= 2(f(tk+1) + !(tk-l)), 1 :S k :S N - 1, Uo = YO,uN = YT

(a difference scheme of first order of accuracy for the boundary-value problem
(3.25) generated by the first order of accuracy explicit difference scheme).
In the case p + q = 2 from formulas (3.42), (3.43), (3.44), (3.45) and (3.46)
it follows that
1
-2(Uk+1 - 2Uk + uk-d(l - 27b(tk)) + (-(b(tk+l) - b(tk))

7
b2(tk)
--2-
rb (tk)
+ -6- -
1
~(b(tk)
- b(tk-d)(l - 7b(tk)) + 27 b(tk-l)(b(tk) - b(tk-d)
1
- 27 (b(tk) - b(tk-d)b(tk) -
-;!
~b'(tk-l)(l -
7b(tk)) -
;!
(-b(tk-l))b'(tk-l)
(b 2(tk_l) + 2b'(tk-l)b(tk))
7b2 (tk)
7 ,
+b(tk+l)(-b(tk) + - 2- + (b(tk) - b(tk-d) + 2b (tk-d)
-A(2b'(tk+l) + 4b2(tk+l))(1 - 7b(tk))
3.
+ (2b"(tk+d + 12b'(tk+l)b(k+l) + 8b 3 (tk+l))))Uk_1
;!
+(7- I (b(tk+d - b(tk)) + ~(2b'(tk+l) + 4b 2(tk+I))

- ;, (2b"(tk+l) + 12b'(tk+l)b(tk+d
+8b (tHI)) + 2b (tk+d 3
47b 3 (tk+l)
3
- 4(b(tk+l) - b(tk))b(tk+d
77
3.2. Taylor's Decomposition on Three Points and Its Applications
-2r- 1b(tk)(b(tk+d - b(h)) + 2r- 1(b(tk+1) - b(tk))b(tk+1)

+b'(tk)(l - 2rb(tk+l)) -
~~ b(tk)b'(tk) + ~(2b"(tk) + 4b'(tk)b(tk+1))
-b(tk)(-2b(tk+1) + 2rb2 (tk+l) + 2(b(tk+l) - b(tk)) + rb'(tk))

+ r, (2b'(tk)
3.
+ 4b2 (tk))(1 - 2rb(tk+d)
-;(2b"(tk) + 12b'(tk)b(tk) + 8b3 (tk)))Uk

4.
+( _
b2 (t
k+l
r- 1
rb3 (t
+ (b(tk+1) - b(tk))b(tk+d
r- 1
b(tk)) - - , (b(tk+1) - b(tk))b(tk+1)
2.
- rb(tk+d) + ;,b(tk)b'(tk) - ;, (b"(tk) + 2b'(tk)b(tk+d)
+b(tk)( -b(tk+d +
-~(2b'(tk)
3.
+-,
b(tk)(b(tk+1) 2.
-~b'(tk)(1
k+l
rb 2 (tk+d
2
r ,
+ (b(tk+1) - b(tk)) + "2 b (tk))
+ 4b 2 (tk))(1 - rb(tk+d) + r, (2b"(tk) + 12b'(tk)b(tk) + 8b 3 (tk)))Uk+1

4.
"2 (f(tk+d + !(tk-I)) - "2!(tk+d(b(tk+1) + b(tk)) - "2!(tk-I)(b(tk+d + b(tk-I))

-i(f'(tk+d - !'(tk-1)), 1 :S k :S N - 1, Uo
YO,UN
YT
a difference scheme of second order of accuracy for the boundary-value problem

(3.25) generated by an explicit difference scheme.
3.2
Taylor's Decomposition on Three Points and Its Applications
We consider again the initial-value problem (3.1). The utilization of Taylor's decomposition on three points in the construction of the two-step difference schemes
of a high order of accuracy for the approximate solutions of problem (3.1) is based
on the following two theorems.
Theorem 3.2.1. Let 17m, m = 1, ... ,q be a solution of the system
~
L
m=1
(2j)'17m
(2' _ 2m)!
]
= 1, p + 1 :S ] :S p + q.
(3.47)
78
Let the function v(t) (0 ::; t ::; T) have a (2p + 2q + 2)-th continuous derivative
and tk-1, tk, tk+ 1 E [0, TJr. Then the following relation holds:
V(tk+1) - 2V(tk)
+ V(tk-1) -
L (}:jV(2 j )(tk)r 2j
P
(3.48)
j=l
- L '1]j(V(2 j )(tk-I) + V(2j) (tk+1))r 2j = o(r2p+2q+2),

q
j=l
where
(}:1 =
(}:j =
1-
2'1]1, (}:j =
((2~)! t
-
m=l
((2~)!
'1]m
j-1
- '1]j -
'1]m
for any j,2::; j ::; q,
(2(j~m))!)2
(2(j~m))!)2 ~:~ any
j, q + 1 ::;
Proof. We obtain the formula
(3.49)
(2p + 2q + l)!(V(tk+d - 2V(tk)

_
~ (2p + 2q + I)! 2
L.J
j=l
(2)!
J
2j
r v
+ v(tk-d)
(2j) (t )
k .
Using the formula of integration by parts, we obtain
2p+2q+2
(_1)2 P+2q+2-"'I ((tk+1 - s)2 p+2 q+l )(2 p+2 q+2-"'I)V b - 1)(S)]~:+l
"'1=1
2p+2q+2
"'1=1
(_1)2 p+2q+2-"'I((s -
tk_d2p+2q+l)(2p+2q+2-"'I)vb-1)(s)1~~_1
79
In a similar manner one can show that

(3.50)
= (2p
+ 2q - 2m + 1)!7 2m (v(2m) (tk+l)

_
. LJ
J=m+l
(2p
- 2V(2m)(tk)
+ V(2m)(tk_l))
+ 2q -
2m + I)! 2j (2j)
_
(2( . _ m))!
27 v (tk), m - 1, ... ,q.
J
Now, using formulas (3.49) and (3.50), we obtain that
- L 7Jm 72m (V(2m) (h+l) - 2v(2m)(tk) + v(2m)(h_l))

q
m=l
Since
80
- L 7]m T2m (V(2m) (tk+d q
2V(2m)(tk)
+ v(2m)(tk_d)
m=1
m=1
m=1
- L 7]m T2m (V(2m) (tk+l) + V(2m)(tk_l)) + L 7]mT2m2v(2m) (tk)
-L
q
m=1
j-l
~ ~
7]m T2m (V(2m) (tk+I) + V(2m)(tk_I)) +
L
q
7]mT2m2v(2m)(tk)
m=1
q
2
2' (2')
~ ~
2
2 (2)
_m))!T Jv J (tk) +.~ ~ 7]m(2(' _m))!T Jv J (tk)
J
J=q+l m=1
J
+ ~ ~ 7]m(2('
J=2 m=1
and (3.47) holds, then (3.48) follows. Theorem 3.2.1 is proved.

Theorem 3.2.2. Let 7]m, m = 1, ... , q be a solution of the system
j-l
. __1__ ~
7]J -
(2j)!
7]m
~1 (2j - 2m)!'P
+1<'<
- J - q.
~ (2j)!7]m
.
~1 (2j - 2m)! = 1, q + 1 ~ J ~ P + q.
Let the function vet) (0 ~ t ~ T) have a (2p + 2q + 2)-th continuous derivative
and tk-l, tk, tk+l E [0, TJ.,.. Then the following relation holds:
V(tk+l) - 2V(tk)
+ v(tk-d -
L Q:jV(2 j )(tk)T 2j
j=1
81
- L rlj (V(2 j )(tk- d + V(2j) (tk+l) )72j = o(72p+2q+2),

j=1
where
al
=1-
2ryl, aj
= ((2j)!
fl
j-l
- ryj -
rym (2(j _ m))!)2 for any j, 2
Proof. Using formulas (3.49) and (3.50), we obtain that
-L
q
rym 72m (V(2m) (h+l) - 2v(2m)(tk)
+ v(2m)h_t})
m=1
q
p+q
+ ~ rym ,~
m=1
J=m+l
1
2' (2)
(2(' _ m))!27 Jv J (tk)
We have that
-L
q
m=1
rym 72m (V(2m)(tk+l) - 2V(2m) (tk)
+ v(2m)(tk_l))
~ j ~ p.
82
- L 1Jm 72m (V(2m) (tk+l) + V(2m)(tk_t}) + L 1Jm72m2v(2m) (tk)

q
m=1
m=1
L 1Jm72m2v(2m) (tk)
P
m=1
- L 1Jm 72m (V(2m) (tk+l) + V(2m)(tk_l))

q
m=1
- L 1Jm 72m (V(2m)(tk+d + v(2m)(tk_d)

m=1
q
j-l
'"
1
'"
2
2 (2)
- "LJ [(2 .)! -1Jj - LJ 1Jm (2( _ m)),]27 Jv J (tk)
J=p+l
m=1
p+q
'"
( .)'
2
'"
2J .
2j (2j)
- L J (2"),(1- LJ1Jm(2("_m)),)7 v
(tk).
J=q+l
m=1

Note that the relation (3.48) is called Taylor's decomposition of function v(t)
on three points.
83
Now, we will consider the applications of Taylor's decomposition of function

on three points to approximate solutions of the initial-value problem (3.1) and
boundary value problem (3.25).
We consider two cases: the simple case when a(t) = a, for any t E [0, T] and
the general case when a(t) is the smooth function defined on the segment [0, T].
Let a(t) = a, for any t E [0, T]. First, we consider an initial-value problem
(3.1). From (3.48) it is clear that for the approximate solution of problem (3.1) it
fi d '
d (2j).
. 1 < .<
d (2j)
(2j).
.
IS necessary to n Y(r) an Y(td lor any J,
_ J _ p an Y(tk-,)' Y(tk+,)lor any
j, 1 ::; j ::; q. Using the equation
y"(t) = -ay(t)
+ j(t),
we obtain
n
+ l)-a)n->. j(2)'-2)(t),n =
2,....
(3.51)
2q
j
LPjy(j)(r)r - L Ojy(j) (O)r j = o(r2p+2q+1),
(3.52)
y(2n)(t) = (-a)ny(t)
>'=1
From (2. 51) it follows that

y(r) - y(O)
2p
j=l
where
.=
j=1
(2p+2q-j)!(2p)!(-1)i
(2p+2q)!j!(2p-j)!
PJ
{ '.
UJ -
y'(r) - y'(O)
lor any j,1::; j ::; 2p,
(2p+2q-j)!(2q)!
(2p+2q)!j!(2q-j)!
and
lor any J,
::; J::; q,
2p-2[.!f]
ajy(j+l) (r)r
j=l
2q-2m+2[.!f]+l
bj y(j+1) (O)T j = O(T2p+2q-2m+2),O::;
L
j=l
where
aj =
(2p+2q-2 m +l-j)!(2p-2[.!f ])!( -I)i

(2p+2q+1-2m)!j!(2p-2[.!f]-j)! '
for any j,1::; j ::; 2p - 2[~],

b. =
J
(2p+2q+1-2m-j)!(2q-2m+2[.!f1+ 1)!
(2p+2q+1-2m)!j!(2q-2m+2[.!f1+1-j)!'
for any j, 1 ::; j ::; 2q - 2m + 2[~]
+ 1.
m::; q -1,
84
Using formulas (3.51) and (3.53), we obtain
y'(T) - y'(O)
p-[TI
L a2n( -aty'(T)T 2n
n=1
P-[TJ
P-[T]
L a2n L( -at->' j(2).-1) (T)T2n

n=1
>'=1
P-[T]
L a2n_1(-a)n Y(T)T 2n - 1 + L a2n-1 L(-at->'j(2>.-2) (T)T 2n - 1

n=1
n=1
>'=1
q-m+[TI
q-m+[TI
n
2n
L
b2n (-aty'(O)T L
b2n L( -at->' j(2).-1) (O)T2n
n=1
n=1
>'=1
q-m+[T]+1
L
b2n - 1(_a)n y(O)T 2n - 1
n=1
q-m+[T]+1
n
L
b2n - 1 L( -at->' j(2)'-2) (O)T2n-1
n=1
>'=1
= O(T2p+2q-2m+2), 0 ~ m ~ q - 1.
From the last formula it follows that
(1
p-[TI
L
n=1
a2n( -atr 2n )y'(r) = y'(O) -
P-[T]
p-[T]
p-[Tl
L a2n L( _a)n->. j(2).-1) (T)T2n

n=1
>-=1
n
L a2n_1(-a)n Y(T)T 2n - 1 - L a2n-1 L(-at->-j(2>--2) (T)T 2n - 1

n=1
n=1
>-=1
q-m+[TI
q-m+[TI
n
+ L
b2n (_a)n y'(O)T 2n + L
b2n L( _a)n->- j(2)--1) (O)T2n
n=l
n=1
>-=1
+
+
q-m+[T]+l
L
n=1
b2n - 1(_a)n Y(O)T 2n - 1
q-m+[T)+l
L
n=1
b2n - 1 L(-at->- j(2)--2) (O)T2n-1

>-=1
+ O(T2p+2q-2m+2),
or
y'(T)
p-['-]
'" 1
[y'(O)
1 + L:~:~2J a2n( _a)nT2n
n
L a2n L(_a)n->. j(2)'-1)(T)T 2n

n=1
>'=1
(3.53)
85

p-[T]
a2n-1 (-aty(r)r 2n - 1 -
p-[Tl
n=l
a2n-1
n=l
q-m+[ T)
b2n (_a)n y '(O)r 2n
L( _a)n-,\ j(2,\-2) (r)r2n-1

'\=1
q-m+[ T]
n=l
b2n
n=l
q-m+[T]+l
L( _a)n-,\ j(2,\-1)(O)r 2n
'\=1
b2n - 1( -aty(O)r 2n - 1
n=l
q-m+[TJ+1
b2n - 1
L( _at-'\j(2,\-2)(O)r 2n - 1] + o(r2p+2q-2m+2).
'\=1
n=l
Now, neglecting the last terms in formulas (3.52), (3.53) and using formulas
(3.51) and (3.48), we obtain the difference schemes
Uk+1 - 2Uk
r2
+ Uk-1
L (-a )j r 2j-2 Uk
P
j=l
Q"
J
=L
1]j
j=l
L( -a)j-'\(j(2,\-2)(tk+1) + j(2,\-2) (tk_1))r2j-2

'\=1
j=l
'\=1
+ LQj L(-a)j-Aj(2A-2)(tk)r2j - 2,1:::; k:::;

U1 ~ Uo
+L
N -1,
P2n( -atu1r2n-1
n=l
q-m+[T]
b2n (_a)n y '(O)r 2n
+ L
n=l
+L
P2m-1( _a)m-1 r 2m-2
m=l
q-m+[T)
n
L
b2n L( _a)n-,\ j(2,\-1)(O)r 2n
n=l
'\=1
(3.54)
86
q-m+[T]+l
q-m+[T]+1
n
L
b2n - 1(-at u or 2n - 1 +
L
b2n - 1 L(-at->' j(2)'-2) (O)r2n-1j
n=1
n=1
>'=1
q
- L82n(-afnuor2n-l - L82n_l(-a)2(n-l)y'(O)r2n-2
n=1
n=1
n
= - L
n=1
p2n L( -at->' j(2)'-2) (r)r2n-1
+L
>'=1
82n L( _a)n->. j(2).-2) (O)r2n-1

n=1
>'=1
p
n-l
- L P2n-l L( -at- 1->. j(2).-I) (r)r2n-2
n=1
>'=1
n-l
82n - 1 L (_a)n-l->. j(2).-I) (O)r 2n - 2,Uo = YO.
n=1
>'=1
+L
Thus, we constructed the difference schemes of (2p + 2q)-order of accuracy

for the approximate solution of the initial-value problem (3.1). Finally, let us give
some examples of the simplest difference schemes.
In the case p + q = 1 from formulas (3.54) and (3.55) it follows that
Uk+l - 2Uk
+ Uk-l + aUk =
-'------=-2---
= -PI
~)
1 + a2 -a r
j(tk), 1 :S k :S N - 1,
[y'(O) - a2f'(r)r 2 - ad(r)r] - P2j(r)r, Uo = Yo
(a difference scheme of second order of accuracy for the initial-value problem (3.1)).
Second, we consider the boundary-value problem (3.25). From (3.48) it is
clear that for the approximate solution of the problem (3.25) it is necessary to
(2j) lor
~
. 1 < .<
d (2j)
(2j) ~
. 1 < . < U th
fi n d Y(tk)
any J, _ J _ p an Y(tk-Il' Y(tk+Il lor any J, _ J _ q. smg e
equation
y"(t) = ay(t) - j(t),
we obtain
y(2n) (t) = any(t) - L an->' j(2)'-2) (t), n = 2, . . . .
(3.55)
>'=1
Now, using formulas (3.55) and (3.48), we obtain the difference schemes
- Uk+!
- 2Uk
r2
+ Uk-1 + L
j=l
a-a
r
J
2j-2
Uk
(3.56)
j=l
..\=1
87
l:>7J I>j-oX (j(2oX-2) (tk+d + f(2oX-2)(tk_d)r 2j - 2
j=l
..\=1
+ L Ckj L aj -..\ f(2oX-2)(tk)r 2j - 2, 1 ~ k
~ N - 1, Uo
= Yo, UN = YT
Thus, we constructed the difference schemes of (2p + 2q)-order of accuracy

for the approximate solution of the boundary-value problem (3.25). Finally, let us
give some examples of the simplest difference schemes.
In the case p + q = 1 from formula (3.56) it follows that
-
Uk+1 - 2Uk + Uk-1

2
r
+ aUk = f(h), 1 ~
k ~ N - 1, Uo
= Yo, UN = YT
(a difference scheme of second order of accuracy for the boundary-value problem

(3.25)).
Now, let a(t) be the smooth function defined on the segment [0, T]. First,
we consider an initial-value problem (3.1). From (3.48) it is clear that for the
approximate solution of problem (3.1) it is necessary to find Y~;:? for any j, 1 ~
j ~ p and yg;~l)'
yg;:Jl for any j, 1 ~ j ~ q. Using the equation

y"(t) = -a(t)y(t) + f(t),
we obtain
y(2 j )(t)
y(2 j+1)(t)
A 2j (t)y(t) + B 2j (t)y'(t) + hj(t),

A2j+l(t)y(t) + A2j+1(t)y'(t) + hj+1(t),j = 1, ... , (3.57)
where
A 2(t) = -a(t), B 2(t) = 0, h(t) = f(t),

A 3 (t) = -a'(t), B 3 (t) = -a(t), h(t) = f'(t),
A 2j (t) = Ab-1 (t) - B2j - 1(t)a(t), B2j (t) = A 2j - 1(t) + B~j_1 (t),
hj(t) = f~j-1 (t) + B 2j - 1(t)f(t),j = 2,3, ... ,
A2j +l(t) = A;j(t) - B 2j (t)a(t), B 2j +l(t) = A 2j (t) + B~j(t),
hj+1 (t) = f~j(t) + B 2j (t)f(t),j = 2,3, ....
(3.58)
From (3.51) it is clear that for the approximate solution of the problem (3.1) it is
necessary to find y(tk)' y(tk-Jl and y(tk+Jl' The choice formula for y(tk)' Y(tk-Jl and
88
Y(tk+l) is not unique. First, we consider the approximation of Y(T)' Using formulas
(3.53) and (3.57), we obtain that

2p-2[~1
y'(T) - y'(O)
aiTi{Ai+1(T)Y(T)
+ Bi+l(T)y'(T) + fi+l(T)}
i=l
2q-2m+2[~I+l
biTi{Ai+1(0)Y(T)
+ Bi+l(O)Y'(T) + fi+l(O)}
= O(T2p+2q-2m+2).
i=l
Since
2p-2[~)
1+
ai Ti Bi+l (T)
=1=
i=l
for the small
> 0, we obtain
y'(T) = TO,TY(O)
+ PO,TY'(O) + Ro,TY(T) + FO,T>
(3.59)
where
2p-2[~]
TO,T = (1
2q-2m+2[~1+l
aiTiBi+l(T))-l
i=l
2p-2[~]
PO,T = (1 +
2q-2m+2[~]+1
ai TiB i+l(T))-l[l
i=l
Ro,T = -(1
2p-2[ ~ I
biTiBi+l(O)],
2p-2[ ~ I
aiTiBi+l(T))-l
2p-2[~1
i=l
i=l
FO,T = (1+
biTiAi+l(O),
i=l
2p-2[~)
aiTiBi+l(T))-l[
i=l
aiTiAi+l(T),
i=l
2q-2m+2[~1+1
aiTifi+l(T)+
i=l
biTi fi+ 1 (0)].
i=l
Second, we consider the approximation of Y(tk) and Y(tk-d' Using formulas (3.53)
and (3.57), we obtain that
2p-m+l
y'(tk) - y'(tk-l)
CiTi{Ai+l(tk)y(tk)
+ Bi+l(tk)y'(tk) + fi+l(tk)}
i=l
2q-rn+2
- L
i=l
diTi{Ai+l(tk-t}y(tk-t}
+ Bi+l(tk-t}y'(tk-l) + fi+l(tk-t}}
89
where
C' t -
(2p+2q-2m+3-j)!(2p-m+l)!(-1)i
(2p+2q+3-2m)!i!(2p-m+l-i)!
d. =
for any i,1:S i :S 2p - m + 1,
(2p+2q+3-2m-j)!(2q-m+2)!
(2p+2q+3-2m)!i!(2q-m+2-i)!
for any i,1:S i :S 2q - m
+ 2,
and
2p-m+2
y'(tk) - y'(tk-l)
CiTi{Ai+l(tk)y(tk)
+ Bi+l(tk)y'(tk) +
fi+l(tk)}
i=l
2q-m+l
- L
tLTi{Ai+l(tk-l)y(tk-l)
+ Bi+l(tk-l)y'(tk-l) + fi+l(tk-l)}
i=l
where
-. -
Ci -
(2p+2q-2m+3-j)!(2p-m+2)!(-1)i
(2p+2q+3-2m)!i!(2p-m+2-i)!
J.J =
(2p+2q+3-2m-j)!(2q-m+l)!
(2p+2q+3-2m)!i!(2q-m+l-i)!
for any i,1 <_ i <_ 2p - m
+ 2,
for any i,1::; i ::; 2q - m + 1.
Hence, we can write
To solve the last system, we need

2p-m+2
t:.k=-[1+
2q-m+l
Ci TiB i+l(tk)][1+
i=l
i=l
tLTiBi+l(tk_l)]
i=l
2q-m+2
+[1 +
2p-m+2
i
diT Bi+l (tk-d][l
i=l
CiT
B i+ 1(tk)] =I 0
(3.60)
90
for all k. Applying formula (3.59), we obtain
ti. k = -CZT ZB 3(tk) - dZT ZB 3(tk-l) + dZT ZB 3(tk-l) + CZT ZB 3(tk)] + 0(T 3)
Since
_
(2p + 2q + 1 - 2m)!
Cz + dz - dz - Cz = - (2p + 2q + 3 _ 2m)!2!4(p + q + 1 - m)
we have that if a(tk) # for all k, then for all small

satisfied. From that it follows that
y'(tk) = ti.;;l{ -[1
# 0,
the requirement (3.60) is
Zq-m+l
~TiBi+l(tk-d]
i=l
Zp-m+Z
[- L
CiTi{AH1(tk)y(tk)
+ fHl(tk)}
i=l
Zq-m+l
+
~Ti{Ai+l(tk-l)y(tk-d + fHl(tk-d}]}
i=l
Zq-m+Z
+[1 +
diT i B H1 (tk)]
i=l
Zp-m+l
[- L
CiTi{A+l(tk)y(tk)
+ fHl(tk)}
i=l
Zq-m+Z
+
diTi{AH1(tk-l)y(tk-d
+ fHl(tk-l)}] +O(Tzp+Zq-Zm+Z)
i=l
where
Zq-m+l
Ck,T
ti.;;1{-[1+
Zq-m+Z
+[1+
i=l
Zq-m+l
~TiBi+l(tk-l)l
i=l
diTiBH1(ik)]
Zq-m+Z
L
i=l
~TiAHl(tk_l)
i=l
diTiAi+l(tk-d},
2q-m+l
~kl{[l+
Dk,r
2p-m+2
<LTiBi+l(tk-l)]
i=l
-[1 +
2p-m+l
diTiBH1(tk)]
i=l
CiTiAHl(tk),
i=l
2q-m+l
~kl{_[l+
Fl,r
2p-m+2
<LTiBH1(tk_I}][-
i=l
2q-m+l
Z.TifHl(tk)
i=l
2q-m+2
<LT fHl (tk-I)]
+ [1 +
i=l
i
diT B H1 (tk)]
i=l
2q-m+2
2p-m+l
x[-
CiTiAi+l(tk)
i=l
2q-m+2
91
CiTifHl(tk)
i=l
diTifHl(tk-I}]},
i=l
and
y'(tk-I} = ~kl{_[l
2q-m+2
CiTiBi+l(tk)]
i=l
2p-m+l
x[-
CiTi{Ai+l(tk)y(tk)
+ fi+l(tk)}
i=l
2q-m+2
+
diTi{Ai+l(tk-l)y(tk-l)
+ Ii+l(tk-l)}]}
i=l
2q-m+l
+[1 +
i
CiT Bi+l (tk)]
i=l
2p-m+2
x[-
CiTi{AH1(tk)y(tk)
+ fHl(tk)}
i=l
2q-m+l
+
<LTi{Ai+l(tk-l)y(tk-I}
+ fHl(tk-I}}] + O(T2p+2q-2m+2)
i=l
where
2q-m+l
Ek,r
~kl{_[l+
2q-m+l
CiTiBi+l(tk)]
i=l
2p-m+2
-[1+
i=l
<LTiAH1(tk_l)
i=l
2q-m+2
CiTiBHl(tk)]
i=l
diTiAi+l(tk_l)},
92

2p-m+l
2:
~kl{[1 +
Hk,r
2p-m+l
2:
"(\TiBiH(tk)]
i=l
2p-m+2
2p-m+l
2:
-[1 +
2:
diTiBiH(tk)]
i=l
~kl{[1 +
2:
2q-m+l
2:
Ci Ti Bi+l (tk)][
i=l
2p-m+2
- 2:
Ci Ti fi+l(tk)]+[1+
2:
i=l
(\TiBi+l(tk)]
i=l
2q-m+2
2p-m+l
2:
i
;T fi+l (tk-l)
i=l
2p-m+2
x[
CiTiAi+l(tk)},
i=l
2p-m+l
F~,r
CiTiAiH(tk)
i=l
CiTifi+l(tk) -
i=l
2:
diT i fi+ 1 (tk-l)]).
i=l
From that it follows that
Now, neglecting the last terms in formulas (3.48), (3.52), (3.60), (3.61), (3.62),
(3.63) and using formula (3.58), we obtain the two-step difference schemes of
(2p + 2q)-th order of accuracy
'L..J
" ' TfjT 2-2
J
{A2j (tk-d Uk-l
+ B 2j (tk-l)[Ek,r Uk-l + Hk,rUk]
j=l
+A2j (tk+l)Uk+l
+ B 2j (tk+d[Ck+l,r Uk + Dk+l,rUk+d}
2: r/j[B2j(tk-dF~,r + B2j(tk+l)Fl+1,r]T2j-2 + 2:
j=l
Q:
2j 2
j B 2j (tk)Fl,r T - ,
j=l
T Uo
Ul -
N -1,
+ '"'
L..J P2m {A2m Ul + B 2m [TO,ruo + Ro,rUl ]} T2n - 1
m=l
q
- 2: 82m{A2mUO + B 2m [TO,rUO+ Ro,r u d}T 2n - 1

m=l
93

p
= -
P2m{A 2m (r)Ul
+ B 2m (r)[PO,TY/(0) + Fo,T]}r2n - 1
m=l
q
+L
82m {A 2m (0)uO
+ B2m (0)yb + hm(0)}r 2m - 1 ,uO =
Yo
m=l
for the approximate solution of the initial-value problem (3.1).

Second, we consider the boundary-value problem (3.25). From (3.48) it is
clear that for the approximate solution of problem (3.25) it is necessary to find
(2j) "
.1 <
(2j)
(2j) "
.1 <
Y(tk) Lor any J,
_ J. <
_ p an d Y(tk-d'
Y(tk+d Lor any J,
_ J. <
_ q. U'
smg th e
equation
y"(t) = a(t)y(t) - f(t),

we obtain
y(2 j )(t)
y(2j+l)(t)
+ B 2j (t)y'(t) + hj(t),
A 2j+l(t)y(t) + A 2j+l(t)y/(t) + hj+l(t),j =
A 2j (t)y(t)
1, ... , (3.64)
where
A 2(t) = a(t), B 2(t) = 0, h(t) = - f(t),

A 3(t) = a/(t), B 3(t) = a(t), h(t) = f'(t),
A2j (t) = A~j_l (t) + B2j- 1 (t)a(t), B2j (t) = A2j- 1 (t) + B~j_l (t),
hj(t) = f~j-l (t) - B2j- 1 (t)f(t), j = 2,3, ... ,
A 2j+l (t) = A~j(t) + B2j (t)a(t), B 2j+l (t) = A 2j (t) + B~j(t),
hj+I(t) = f~j(t) - B 2j (t)f(t),j = 2,3, ....
(3.65)
Now, using formulas (3.61), (3.62), (3.63), (3.64) and (3.48), we obtain the difference schemes
L 7]jr2j-2{A2j(tk_l)Uk_l + B 2j (tk-I)[Ek,T Uk-l + Hk,TUk]

j=l
+A2j (tk+d u k+l
+ B2j (tk+d [Ck+I,TUk + Dk+l,TUk+l]}
L 7]j[B2j(tk-l)F~,T + B2j(tk+l)F1+I,T]r2j-2
j=l
L (}:jB2j (tk)Fl,T r 2j - 2, 1 ::; k ::; N - 1, Uo = Yo, UN = Yr.
j=l
94
Thus, we constructed the difference schemes of (2p + 2q)-th order of accuracy for
the approximate solution of the boundary-value problem (3.25).
The choice of Taylor's decomposition on three points in the construction
(2p + 2q)-th order of accuracy difference schemes for the approximate solutions
of the initial-value problem (3.1) and of the boundary-value problem (3.25) is not
unique. We will give another Taylor's decomposition on three points in which we
use V(2i) (tk), 0 ::; i ::; p and v(i)(tkd, 0 ::; i ::; 2q.
Theorem 3.2.3. Let the function vet) (0::; t ::; T) have a (2p+2q+2)-th continuous
derivative and h-l, tk, tk+1 E [0, T]r' Then the following relation holds:
v(tk+d - 2V(tk)
+ V(tk-l) + L
j
2j
OjV(2 ) (tk)r
(3.66)
j=1
+L
/'j(V(2 j -l)(tk_l) - v(2 j -l)(tk+1))r 2j - 1
j=1
q
+ L,8j(v(2 j )(tk-d + V(2j)(tk+1))r 2j
= o(r2p+2q+2),
j=1
where
__ _
oJ -
2(2p+2q+1-2j)!
(2p+1)!
(2p+2q+1)!(2p+1-2j)! (2j)!
,8J- --
(2p+2q+1-2j)!(2q)!
(2p+2q+1)!(2q-2j)!(2j)!
__
(2p+2q+2-2j)!(2q)!
/'J - (2p+2q+1)!(2q+1 2j)!(2j
for any j,l ::; j ::; p,

for any j,l :::; j :::; q,
I)!
for any j,l:::; j :::; q.
Proof. Using the formula of integration by parts, we obtain
2p+2q+2
(-1)2 p +2 q+2-')'((tk+1 - S)2 P+1(S - tk)2q)(2P+2q+2-')')v(')'-I)(S)]~:+1
')'=1
2p+2q+2
')'=1
(_1)2 p +2 q+2-')'((s - tk_l)2 P+1(s - tk)2q)(2p+2q+2-')')V(')'-I)(S)]~~_1
2p+2q+2
95
(-1)2 p+2q+2-'Y((tkH - S)2p H(S - tkfq)(2P+2q+2-'Y)vh-1)(s)1~:+1
"1=2
-(2p + 2q + l)!(V(tk) - v(tk-d)
2p+2q+2
(_1)2 p+2q+2-'Y((S - tk_d 2p +1(S - tkfq)(2p+2q+2-'Y)vh-1)(S)]~:_1'
"1=1
We will calculate the expressions
_ ,tk'
( _1)2 p+2 q+2-'Y((t k+1 _ S)2 p+1(S _ t k )2q)(2 p+2 q+2-'Y) Is-tk+l
and
(_1)2 p+2q+2-'Y((S - tk_d 2PH (S - tk)2 q)(2 p+2 q+2-'Y)!S=tk_l,tk
If j 2 2q + 2, then 2p + 2q + 2 - j
:s 2p. Therefore
(-1)2 p+2q+2- j ((tkH - S)2p H(s - tk)2 q)(2 p+2 q+2- j )ls=tk+l =
If 1
o.
:s j :s 2q + 1, then
(-1)2 p+2 q+2- j ((tkH - S)2 p+1(S - tk)2 q)(2 p+2 q+2- j )ls=tk+l
(-l)j
2p+2q+2-j (2
2
2 - .)'
""
p+ q+
J. (_1)2 p+2 q+2- j -i
L..J
i!(2p + 2q + 2 - j - i)!
i=O
x (j
(2p + I)!
(
_ )j+i-2q-1 (2q)! ( _ )2q- i l
2q _ I)! tkH S
(2q _ i)! S tk
S=tk+l
+i _
= _( -IF (2p + 2q + 2 - j)!(2q)! T j - 1.

(2q + 1 - j)!(j - I)!
So,
(-lfp+2q+2-j((tkH - S)2 P+1(S - tk)2 q)(2 p+2 q+2- j )ls=tk+l
=
(-IF- 1 (2p + 2q + 2 - j)!(2q)! Tj- 1.

(2q + 1 - j)!(j - 1)!
If j 2 2p + 3, then 2p + 2q + 2 - j
:s 2q -
1. Therefore
(-1)2 p+2 q+2- j ((tk+1 - S)2 p+1(S - tk)2 q)(2 p+2 q+2- j )ls=tk
= o.
96
If 1

~
j ~ 2p + 2, then
(-1)2 P+2q+2- j ((tk+l - s)2 p+1(S - tk)2 q)(2 p+2 q+2- j )ls=tk
(-1)j
x
2p+2q+2-j
'"
(2p + 2q + 2 - j)!
LJ
i!(2p + 2q + 2 - j - i)!
i=O
p
X ((tk+1 - s)2 +l)(2 p+2 q+2- j -i)((S - ik)2 q)(i)ls=tk
(_1)2 p+2q+2- j
2p+2q+2-j (2
2
2 - ')'
'"
p+ q+
J. (_1)2P+2q+2- j -i
~
i!(2p + 2q + 2 - j - i)!
(2p + I)!
(t
_ )j+i-2q -1 (2q)! ( _ t )2q-il
(j+i-2q-1)! k+l s
(2q-i)! s
k
S=tk
+ 2q + 2 - j)! (2p + 1)! 1' j - 1.

(2p+ 2 - j)!
(j -I)!
= (2p
So,
(-1)2 p+2q+2- j ((tk+1 - s)2P+l(s - tk)2 q)(2P+2 q+2- j )ls=tk

=
(2p + 2q + 2 - j)! (2p + 1)! j-1

(2p+2-j)!
(j_1)!1'
.
Now, we will calculate the expressions
q
( _1)2 p+2q+2-1'((s _ t k-1 )2 p +1 (s _ t k )2 )(2P+2q+2-1') IS=tk-l,tk'
If j 2:: 2q + 2, then 2p + 2q + 2 - j ~ 2p. Therefore

(_1)2 P+2 q+2- j ((S - tk_1)2 p +1(S - tkf q)(2 p+2 q+2- j )ls=tk_l = O.
If 1 ~ j
2q + 1, then
(_1)2 P+2 q+2- j ((S - tk_d 2P+1(S - tk)2 q)(2 p +2 q+2- j )ls=tk_l
2p+2q+2-j
=
(-l)j
x(s-t
k-1
(2p + 2q + 2 - j)!
(2p + I)!
i!(2p+2q+2-j-i)!(j+i-2q-1)!
)j+i-2q-1
(2q)! (s-t )2 q - i l_
(2q _ i)!
k
s-tk-l
= (-1)j (2p + 2q + 2 - j)!(2q)! (_r)j-l = _ (2p + 2q + 2 - j)!(2q)! r j- 1 .

(2q + 1 - j)(j - 1)!
(2q + 1 - j)!(j - 1)!
So,
(_1)2 p+2 q+2- j ((S - tk_l)2 p+l(S - tk)2 q)(2 p+2q+2- j )ls=tk_l
_ (2p + 2q + 2 - j)!(2q)! r j - 1 .
(2q
If j
+ 1 - j)!(j - 1)!
2p + 3, then 2p + 2q + 2 - j
2q - 1. Therefore
(_1)2 P+2q+2-j((s - tk_d 2P+1 (S - tk)2 q)(2 p+2 q+2- j )ls=tk = o.

If 1 ~ j
2p + 2, then
(_1)2 p+2q+2-j((s - h_l)2P+l(s - tk)2 q)(2 p+2q+2- j )ls=tk
2p+2q+2-j
=
(-1)j
x(s - t
k-l
(2p + 2q + 2 - j)!
(2p + 1)!
i!(2p+2q+2-j-i)!(j+i-2q-1)!
)J+i-2q-l
(2q)! (s _ t )2 q- i l _
(2q _ i)!
k
s-tk
= (-1)j (2p + 2q + 2 - j)! (2p + 1)! r j- 1 .

(2p+2-j)!
(j-1)!
So,
From this it follows
= (2p
+ 2q + 1)!(V(tk+d - 2V(tk) + v(tk-d)
97
98
+ 2~1(_I)j_l (2p + 2q + 2 - j)!(2q)! Tj - 1V(j-l)(t

)
~
(2q+l-j)(j-l)!
k+l
)=2
2p+2
L.
)=2
.) (
)
(2p+2q+2-]! 2p+l! j-l (j-l)( )
T
V
tk
(2p+2-j)!
(j-l)!
2 +2( )j (2p + 2q + 2 - j)! (2p + I)! j-l (j-l)( )

-1
T
v
tk
j=2
(2p+2-j)!
(j-l)!
+L
q
2 +l (2p + 2q + 2 - j)!(2q)! j-l (j-l)(

)
T V
tk-l
j=2 (2q + 1 - j)(j - I)!
= (2p
+ 2q + 1)!(v(tk+l) - 2V(tk) + v(tk-d)
~(2p+2q+I-2j)!(2q)! 2j ( (2j)(
+~
(2q _ 2j)!(2j)!
)=1
tk-l
+v
(2j )(
tk+l
))
(2p+2q+2-2j)!(2q)!T2j-l(V(2j-l)(t _ ) _V(2 j -l)(t

))
+~
~ (2q + 1 - 2j)!(2j _ I)!
k
k+l
1
)=1
2j (2 j )( )
2 ~(2p+2q+1-2j)!(2p+1)!
~
T V
tk.
j=1 (2p + 1 - 2j)!
(2j)!

Note that using this Taylor decomposition on three points, we can extend our
discussion to construct difference schemes of an arbitrary high order of accuracy
for the approximate solutions of the initial-value problem (3.1) and the boundaryvalue problem (3.25). It is left to the reader.
Chapter 4
Partial Differential Equations
of Parabolic Type
In the present chapter we consider the well-posedness of an abstract Cauchy problem for differential equations of parabolic type,
v'(t) + A(t)v(t) = f(t)
(0::; t ::; T), v(O) = vo,
in an arbitrary Banach space with the linear positive operators A(t). The high
order of accuracy difference schemes generated by an exact difference scheme or
by Taylor's decomposition on two points for the numerical solutions of this problem
are presented. The well-posedness of these difference schemes in various Banach
spaces are studied. The stability and coercive stability estimates in Holder norms
for the solutions of the high order of accuracy difference schemes of mixed type
boundary-value problems for parabolic equations are obtained.
4.1
A Cauchy Problem. Well-posedness
We consider the abstract Cauchy problem for differential equations
v'(t) + A(t)v(t) = f(t)
(0::; t ::; T), v(O) = vo
(4.1)
in an arbitrary Banach space E with the linear (unbounded) operators A(t). Here
v(t) and f(t) are the unknown and the given functions, respectively, defined on
[0, T] with values in E. The derivative v' (t) is understood as the limit in the norm
of E of the corresponding ratio of differences. A( t) is a given, closed, linear operator
in E with domain D(A(t)) = D, independent of t and dense in E. Finally, Vo is a
given element of E.
A function v(t) is called a solution of problem (4.1) if the following conditions
are satisfied:
i. v(t) is continuously differentiable on the segment [0, T]. The derivatives at
the endpoints of the segment are understood as the appropriate unilateral
derivatives.
Chapter 4. Partial Differential Equations of Parabolic Type
100
ii. The element v(t) belongs to D = D(A(t)) for all t E [0, T] and the function
A(t)v(t) is continuous on [0, T].
iii. v(t) satisfies the equation and the initial condition (4.1).
A solution of problem (4.1) defined in this manner will from now on be

referred to as a solution of problem (4.1) in the space C(E) = C([O, T], E). Here
C(E) stands for the Banach space of all continuous functions <p(t) defined on [0, T]
with values in E equipped with the norm
From the existence of such solutions evidently follows that f(t) E C(E) and
Vo E D.
We say that problem (4.1) is well posed in C(E) if the following conditions
are satisfied:
f(t) E C(E) and any Vo E D. This

means that an additive and homogeneous operator v(t) = v(t; f(t), vo) is
defined which acts from C(E) x D to C(E) and gives the solution of problem
(4.1) in C(E).
1. Problem (4.1) is uniquely solvable for any
2. v(t; f(t), vo), regarded as an operator from C(E) x D to C(E), is continuous.

Here C(E) x D is understood as the normed space of the pairs (f(t), vo),
f(t) E C(E) and Vo E D, equipped with the norm
11(f(t),vo)llc(E)XD = Ilfllc(E) + IlvollD.

By Banach's theorem in C(E) and these properties one has coercive inequality
Ilv'IIC(E)
+ IIA(.)vllc(E)
~ Mc[llfllc(E)
IlvollD],
(4.2)
where Mc (1 ~ Mc < +00) does not depend on Vo and f(t). This inequality
is called the coercivity inequality in C(E) for (4.1). If A(t) = A, then the coercivity inequality implies analyticity of the semigroup exp{ -sA} (s 2 0), i.e., the
estimates
lIexp(-sA)II E->E' IlsAexp(-sA)IIE->E ~ M (s > 0)
hold for some M E [1, +00). Thus, the analyticity of the semigroup exp{ -sA},
(s 2 0) is a necessary for the well-posedness of problem (4.1) in C(E). Unfortunately, the analyticity of the semigroup exp{ -sA}(s 2 0) is not sufficient for the
well-posedness of problem (4.1) in C(E).
Suppose that for each t E [0, T] the operator -A(t) generates an analytic
semigroup exp{ -sA(t)}(s 2 0) with exponentially decreasing norm, when s --+
+00, i.e., the estimates
Ilexp( -sA(t))IIE->E' IlsA(t) exp( -sA(t))IIE->E ~ M e- 6s (s > 0)
(4.3)
101
4.1. A Cauchy Problem. Well-posedness
hold for some M E [1, +(0), 8 E (0, +(0). From this inequality it follows that
operator A-I (t) exists and is bounded and hence A(t) is closed in C(E) .
Suppose that the operator A(t)A -1 (s) is Holder continuous in t in the uniform operator topology for each fixed s, that is,
II[A(t) - A(T)]A- 1 (s)IIE .....E ~ Mit - Tlo,O < E ~ 1,
(4.4)
where M and E are positive constants independent of t, sand T for ~ t, s, T ~ T.

An operator-valued function v(t, s), defined and strongly continuous jointly
in t, s for ~ s < t ~ T, is called a fundamental solution of (4.1) if
1) the operator v(t, s) is strongly continuous in t and s for

2) the following identity holds:
v(t,s)
= V(t,T)V(T,S),V(t,t) = I
for
s
~
<t
T,
t ~ T,
3) the operator v(t, s) maps the region D into itself. The operator u(t, s) =
A(t)v(t,s)A- 1 (s) is bounded and strongly continuous in t and s for ~ s <
T,
4) on the region D the operator v(t, s) is strongly differentiable relative to t and

s, while
8v(t,s) _ -A() ( )
8t tv t, s
and
8v(t,
8s s) = v ( t, s )A()
s.
v( t, s) is also called an evolution operator, Green's function, etc.

H the function f(t) is not only continuous, but also continuously differentiable
on [0, T] and Vo E D, it is easy to show that the formula
J
t
v(t) = v(t,O)vo +
(4.5)
v(t,s)f(s)ds
gives a solution of problem (4.1). Here v( t, s) is the fundamental solution of (4.1).

Now we shall establish a series of interesting lemmas and estimates concerning
the semigroup exp{ -sA(t)}(s 2 0) and the fundamental solution v(t, s) of (4.1)
which will be useful in the sequel.
Lemma 4.1.1. For any
inequality
<
< s + T < T,
~ T and
lIexp(-sA(t))-exp{-(s+T)A(t)}IIE..... E~M(
where M does not depend on a, t, sand T.
a ~ 1 one has the

TO
S+T
)O '
(4.6)
102
Proof Using the formula
S+T
exp( -sA(t)) - exp {-(s
+ r)A(t)}
A(t) exp( -pA(t))dp
and estimate (4.3), we obtain that

Ilexp( -sA(t)) - exp {-(s
+ r)A(t)}II E---' E ~ M I -.
s
The inequality
+ r)A(t)}II E---' E ~
MI
follows from (4.3) and the triangle inequality. Therefore

+ r)A(t)}IIE---.E ~ M I min{l, ~}.

s
From this it follows that

lIexp( -sA(t)) - exp {-(s
+ r)A(t)} II E---.E
~ MI t
2r
+ r'
Interpolating the last inequality and

lIexp( -sA(t)) - exp {-(s
+ r)A(t)}II E---' E ~ M I ,
we obtain (4.6). Lemma 4.1.1 is proved.

Lemma 4.1.2. For any 0
s, r, t
T and u ED the following identity holds:
[exp( -rA(t)) - exp( -rA(s))]u
(4.7)
J
T
= -
exp{ -(r - p)A(t)}[A(s) - A(t)]A-I(s) exp{ -pA(s)}A(s)udp.
o
Proof. Define the function
G(p) = - exp{ -(r - p)A(t)} exp{ -pA(s)}u,
p ~ r.
G(p) is strongly differentiable with respect to p with derivative

G'(p) = - exp{ -(r - p)A(t)}A(t) exp{ -pA(s)}u
+ exp{ -(r =
p)A(t)}A(s) exp{ -pA(s)}u
exp{ -(r - p)A(t)} [A(s) - A(t)] exp{ -pA(s)}u
= exp{ -(r - p)A(t)} [A(s) - A(t)]A-I(s) exp{ -pA(s)}A(s)u.

Since G' (p) is continuous in p, (4.7) follows upon integrating the last identity from
Lemma 4.1.2 is proved.
o to r.
103
Lemma 4.1.3. For any 0::; s,
7,
t ::; T and 0::; c ::; 1 the following estimates hold:
Ilexp( -7 A(t)) - exp {-7 A(s)} IIE->E ::; Mit - s\<e- o T,
(4.8)
IIA(t)[exp( -7 A(t)) - exp {-7 A(s)}] IIE->E ::; M 7- l lt - sl< e- o T, 7
> 0, (4.9)
IIA(t)[exp( -7 A(t)) - exp {-7 A(s)}]A -1 (s) IIE->E ::; Mit - sl<e- o T,
(4.10)
Ij[exp( -tA(7)) - exp {-SA(7)}]A- l (7)IIE->E ::; Mit - sle-omin{t,s},
(4.11)
II A (t)[exp( -tA(7)) - exp {-SA(7)}]A- 2 (7)IIE->E ::; Mit - sle-omin{t,s}, (4.12)
where M 2: 0 and 0> 0 do not depend on c, t, sand 7.

Proof. Let u E E. Then
[exp( -7A(t)) - exp {-7A(s)}]u
= [exp( -~A(t))
- exp {-~A(s)}] exp {-~A(s)} u
+exp( -~A(t))[exp(-~A(t)) - exp

= [exp( -~A(t))
{-~A(s) }]u
- exp {-~A(s)}] exp {-~A(s)} u
+A(t) exp( -~A(t))[exp(-~A(t)) - exp {-~A(s) }]A-l(s)u

+ exp( -
~A(t))[A(t) - A(s)]A- l (s) exp( - ~ A(s ))]u
- exp{ -rA(t)} [A(t) - A(s)]A-l(s)u = J 1 + h

where
J l = [exp( -~A(t)) - exp {-~A(s)}] exp
+ h,
{-~A(s)} u,
A(t) exp( -~A(t))[exp(-~A(t)) - exp { -~A(s) }JA-l(s)u

J3
= eXP(-~A(t))[A(t) - A(s)]A-l(s)exP(-~A(s))]u
- exp{ -7A(t)} [A(t) - A(s)]A-l(s)u.
Applying formula (4.7), we can write
J
T
2"
Jl = -
exp{ -(~ - p)A(t)}[A(s) - A(t)]A-l(s)
x exp{ -pA(s)}A(s) exp {-~A(s)} udp
104
for u ED. Using the last formula and estimates (4.3) and (4.4), we obtain
T
IIJIIIE ~ / II exp{ -(~
- p)A(t)}IIE->EII[A(s) - A(t)]A-I(s)IIE->E
o
x II exp{ -pA(s)}IIE->EIIA(s) exp { -~A(s)} ullEdp
T
~/
M exp{ -( ~ - p)6)}Mlt -
siC M exp{ -p6}M ~ exp {-~6} ullEdp
o
~
MIlt - sic exp{ -7"6}lluIIE.
/2 exp{ -( 27" - p)A(t)}

T
7"
J2 = -A(t) exp( -2 A(t))
x [A(s) - A(t)]A-I(s) exp{ -pA(s)}dp

for u ED. Using the last formula and estimates (4.3) and (4.4), we obtain
T
IlhilE ~
7" A (t))IIE->E
IIA(t) exp( -2
/' II exp{ -( 2
7" -
p)A(t)}IIE->E
x II[A(s) - A(t)]A-I(s)IIE->EII exp{ -pA(s)} IIE->E lIuliEdp
,
T
~ M ~ exp {-~6} /
M exp{ -(~ - p)6)}Mlt -
sic M exp{ -p6}lluIIEdp
o
~
MIlt - sic exp{ -7"6}lluIiE.
Using estimates (4.3) and (4.4), we obtain
7"
II exp{ -2A(s)}IIE->ElluIlE
+11 exp {-7"A(t)} IIE->EII[A(s) -
A(t)]A- I (s)IIE->Ell u IlE
~ 2M exp { -~6} Mit - sic M exp{ -~o}lluIiE ~ MIlt - sic exp{ -7"o}lluIiE.
105
Combining the estimate for J 1 , hand h, we obtain estimate (4.8). Applying

formula (4.7), we can write
[exp(-rA(t)) - exp(-rA(s))]u
(4.13)
J
r
= -
exp{ -(7 - p)A(t)} [A(s) - A(t)]A- 1 (s)exp{ -pA(s)}A(s)udp
J
~
= -
exp{ -(7 - p)A(t)}[A(s) - A(t)]A- 1 (s) exp{ -pA(s)}A(s)udp
-J
r
exp{ -(7 - p)A(t)}[A(s) - A(t)]A- 1 (s) exp{ -pA(s)}A(s)udp = G 1 + G2 ,
"2
where
J
r
"2
G1 = -
exp{ -(7 - p)A(t)}[A(s) - A(t)]A- 1 (s) exp{-pA(s)}A(s)udp
J
r
G2 = -
exp{ -(r - p)A(t)}[A(s) - A(t)]A- 1 (s) exp{ -pA(s)}A(s)udp.
"2
Integrating by parts, we can write
J
r
"2
G1 =
exp{ -(r - p)A(t)}[A(s) - A(t)]A- 1 (s)d[exp{ -pA(s)}u]
+ exp{ -"2A(t)}[A(s) -
A(t)]A- 1 (s) exp{ -"2A(s)}u
- exp{ -rA(t)}[A(s) - A(t)]A- 1 (s)u
-J
r
"2
A(t) exp{ -(r - p)A(t)}[A(s) - A(t)]A- 1 (s) exp{ -pA(s)}udp.
a
Applying estimates (4.3) and (4.4), we can write
r
IIA(t)G11I E :S IIA(t) exp{ -"2 A(t)}IIE->EII[A(s) - A(t)]A -1 (s)IIE->E
106
+IIA(t) exp{ -7 A(t)} IIE-+EII[A(s) - A(t)]A -1 (S)IIE-+ElluIIE
J
T
"2
IIA2 (t) exp{ -(7 - p)A(t)}IIE-+E
x II[A(s) - A(t)]A- 1(s)IIE-+EII exp{ -pA(s)}IIE-+ElluIIEdp
~ M(~ )-1 exp{ -~J}Mlt - sl< exp{ -~J}lluIIE + M7- 1exp{ -7J}Mlt - sl<lIuliE
J
T
"2
M(7 - p)-2 exp{ -(7 - p)J}Mlt - sl< exp{ -pJ}lIuIiEdp
~ MIlt - sl<7- 1 exp{ -7J}lluIIE.
J
T
G2 =
-A- 1(t)
d[exp{ -(7 - p)A(t)}][A(s) - A(t)]A- 1(s) exp{ -pA(s)}A(s)u
"2
-A- 1(t)[A(s) - A(t)]A- 1(s)exp{-7A(s)}A(s)u
+A -1 (t) exp{ -~A(t)}[A(s) - A(t)]A -1 (s) exp{ -~ A(s)}A(s)
J
T
-A- 1(t)
exp{ -(7 - p)A(t)}][A(s) - A(t)]A- 1(s)exp{ -pA(s)}A 2(s)udp.
"2
IIA(t)G2I1E ~ II[A(s) - A(t)]A- 1(s)IIE-+EIIA(s) exp{ -7 A(s)} IIE-+E lIuliE
+11 exp{ -~A(t)}IIE-+EII[A(s)
- A(t)]A- 1(s)IIE-+E
IIA(s) exp{ -'2A(s)}IIE-+ElluIiE
J
T
II exp{ -(7 - p)A(t)} IIE-+EII [A(s) - A(t)]A- 1(s)IIE-+E
"2
X
IIA 2(s) exp{ -pA(s)} II E-+E lIuli Edp

~ Mit - sl< M7- 1exp{ -7J}lIuIiE
+M exp{ -~J}Mlt - SilO M(~)-1 exp{ -~J}lIuIiE
107
J
T
M exp{ -(7 - p)o}Mlt - sI0(p)-2 exp{ -po}lluIIEdp

~ MIlt - S107- 1exp{ -7o}lluIIE.
Combining the estimates for A(t)G1 and A(t)G2, we obtain estimate (4.9). Using
formula (4.13), we notice that, for every u E E,
[exp(-7A(t)) - exp{-7A(s)}]A- 1(s) = HI
+ H 2,
where
J
T
exp{ -(7 - p)A(t)} [A(s) - A(t)]A-1(s) exp{ -pA(s)}udp,
HI = -
J
T
H2 = -
exp{ -(r - p)A(t)}[A(s) - A(t)]A-1(s) exp{ -pA(s)}udp.
J
T
IIA(t)H11I E
IIA(t) exp{ -(r - p)A(t)}IIE->EII[A(s) - A(t)]A- 1(s)IIE->E
x II exp{ -pA(S)}IIE->ElluIIEdp
J
,T
M(7 - p)-l exp{ -(7 - p)o}Mlt -
slo exp{ -po}lIuIiEdp
J
T
H 2 = -A-1(t)
d[exp{ -(7 - p)A(t)}][A(s) - A(t)]A-1(s) exp{ -pA(s)}u
-A-1(t)[A(s) - A(t)]A-1(s)exp{-rA(s)}u
+A-1(t) exp{ -~A(t)}[A(s) - A(t)]A-1(s) exp{ -~A(s)}
J
T
-A-1(t)
exp{ -(7 - p)A(t)}][A(s) - A(t)]A-1(s) exp{ -pA(s)}A(s)udp.
108

IIA(t)H2 1IE
! II
:s II[A(s) -
A(t)]A-
(s)IIE--.EII exp{ -r A(s)}IIE--.e1luIIE
exp{ -(r - p)A(t)}IIE--'EII[A(s) - A(t)]A- 1(s)IIE--.E
:s Mit -
x IIA(s) exp{ -pA(s)}IIE--.ElluIIEdp

r
r
sic M exp{ -r8}lluIIE + M exp{ -'28}Mlt - sic M exp{ -'28}lluIIE
T
+!
M exp{ -(r - p)8}Mlt -
sic M(p)-1 exp{ -p8}lluIIEdp
"2
:s MIlt - sic exp{ -r8}lluIIE.

Combining the estimates for A(t)H1 and A(t)H2 , we obtain estimate (4.10). We
can write
[exp( -tA(r)) - exp( -sA(r))]A- 1(r)u
(4.14)
!
t
= -
exp{ -pA(r)}A(r)A- 1(r)udp.
I [exp( -tA(r)) -
exp {-sA(r)}]A -1 (r) II E--.E
:s I! II exp{ -pA(r)}IIE--.Ellulldpl :s Mit - sle- 6min{t,s}.

s
Estimate (4.11) is established. Using formula (4.14), we notice that, for every u,
A(t)[exp(-tA(r)) - exp( -sA(r))]A- 2 ( r)u
!
t
= -
exp{-pA(r)}udp- !(A(t) - A(r))A- 1(r)exp{-pA(r)}udp.

s
Applying estimates (4.3) and (4.4), we obtain

IIA(t)[exp( -tA(r)) - exp {-sA(r)}]A -2(r) II E--.E
109
~ I II exp{ -pA(r)}IIE-+Ellulldpl + I
II(A(t) - A(r))A-I(r)IIE-+E
x II exp{ -pA(r)} II E-+E lIulldpl

~
MIlt - Sle- 5min {t,s}.
Estimate (4.12) is established. Lemma 4.1.3 is proved.

<t
T and u E D the following identities hold:
v(t, s)u = exp{ -(t - s)A(s)}u
(4.15)
J
t
+ v(t, z)[A(s) - A(z)]A-I(s)exp{ -(z - s)A(s)}A(s)udz,

s
v(t, s)u = exp{ -(t - s)A(t)}u
(4.16)
J
t
exp{ -(t - z)A(t)}[A(z) - A(t)]v(z, s)udz.
Proof. Define the function
G(z) = v(t, z) exp{ -(z - s)A(s)}u, s ~ z ~ t.

G(z) is strongly differentiable with respect to z with derivative
G'(z) = v(t, z)A(z) exp{ -(z - s)A(s)}u
-v(t, z)A(s) exp{ -(z - s)A(s)}u
= v(t, z)[A(z) - A(s)] exp{ -(z - s)A(s)}u
=
v(t, z)[A(z) - A(s)]A-I(s) exp{ -(z - s)A(s)}A(s)u.
Since G'(z) is continuous in z, (4.15) follows upon integrating the last identity
from s to t. In a similar manner one establishes the identity (4.16). Lemma 4.1.4
is proved.
following estimates hold:
<
t ~ t
+r
T, 0 ~ a ~ 1 and 0
<
1 the
IIv(t,s)IIE-+E ~ M,
(4.17)
IIA(t)v(t,s)A-I(s)IIE-+E ~ M,
(4.18)
IIv(t, s) - exp{ -(t - s)A(t)}II E-+ E ~ M(t - s)",
(4.19)
IIA(t)[v(t, s) - exp{ -(t - s)A(t)}] II E-+E ~ M(t - S)"-l,
(4.20)
110
:s M(t -
IIA(t)v(t, s)II E __ E
S)-l,
(4.21)
IIA(t+r)v(t+r,s) -A(t)v(t,s)II E __ E :s M(t-S)-l W,
IIA(t + r)v(t + r,s)Awhere M 2
(s) - A(t)v(t, s)A- (s)IIE __E
does not depend on c, t, sand r. Here W= rE:
(4.22)
:s MW,
+ ra(t -
(4.23)
s)-a.
Proof. Applying formula (4.15) and estimates (4.3) and (4.4), we obtain
Ilv(t, s)ullE
:s II exp{ -(t -
s)A(s)}uIlE
J
t
Ilv(t, z)IIE__EII[A(s) - A(z)]A- 1 (s)IIE __E
x II exp{ -(z - s)A(s)}A(s) IIE--E lIullEdz
J
t
:s MllullE +
Ilv(t, z)IIE __ EM(z - s)E:(z - s)-llluIIEdz
J
t
= MllullE + M 1
IIv(t, z)IIE__ E(Z - s)E:- 1 dzlluIlE.
Thus, we have that
IIv(t, s)ullE
:s MllullE + M
J
t
IIv(t, z)IIE__E(Z - s)E:- 1 dzlluIlE
or
J
t
IIv(t, s)IIE__E ::; M
+ M1
IIv(t, z)IIE__E(Z - s)E:- 1 dz.
From the integral inequality, (4.17) follows. Note that from (4.4) it follows that
(4.24)
for each t, s E [0, T]. Actually,
Applying formula (4.16) and estimates (4.3), (4.4) and (4.24), we obtain
IIA(t)v(t, s)A- 1 (s)uIlE ::;
II exp{ -(t -
s)A(t)}IIE__EIIA(t)A- 1 (s)IIE __ElluIIE
II A(t) exp{ -(t - z)A(t)}IIE __EII[A(t) - A(z)]A- 1 (z)IIE __E
111
x IIA(z)v(z, s)A- 1 (s)IIE-.ElluIIEdz ::; MllullE
IIA(z)v(z, S)-l A(s)IIE-.EM(t - z)O(t -
z)-llluIIEdz
MllullE + M 1
IIA(z)v(z, S)-l A(s)IIE-.E(Z - sr- dzlluIIE.
Thus, we have that
IIA(t)v(t,s)A- 1 (s)uIIE-.E::;
MllullE
+M1
IIA(z)v(z, s)A- 1 (s)IIE-.E(t - z)t:- 1 dzlluIIE
or
f
t
IIA(t)v(t, s)A- (s)IIE-.E ::; M
+ M1
IIA(z)v(z, s)A- 1 (s)IIE-.E(t - z)t:- 1 dz.
Using the integral inequality, we obtain (4.18).

Applying formula (4.15) and estimates (4.3), (4.4), (4.8) and (4.18), we obtain
Ilv(t, s)u ::; II exp{ -(t -
exp{ -(t - s)A(t)}uIIE
s)A(t)}u - exp{ -(t - s)A(s)}uIIE
IIv(t, z)IIE-.EII[A(s) - A(z)]A- 1 (s)IIE-.E
xIIA(s) exp{ -(z -
f
f
s)A(s)}IIE-.ElluIIEdz
::; M(t -
s)t:lluIIE +
MM(z - s)t:(z - s)-llluIIEdz
MllullE(t -
s)t:
+ M 1 (z - s)t:- 1dzlluIIE
M2 11u11E(t -
s)t:.
The estimate (4.19) is proved.

In a similar manner one establishes estimate (4.20). Using estimates (4.20)
and (4.3), we obtain
IIA(t)v(t, s)IIE-.E
::; IIA(t) exp{ -(t -
s)A(t)}II E-. E
+ IIA(t)[v(t, s) - exp{ -(t - s)A(t)}]II E-. E
::; Mit - SI-l + Mit - slt:-l ::; Milt - SI-l.

The estimate (4.21) is proved.
112
A(t)v(t, s)A- 1(s)u = A(t) exp{ -(t - s)A(tnA- 1(s)u
J
t
A(t) exp{ -(t - z)A(tn[A(z) - A(t)]v(z, s)A-1(s)udz = J 1 (t)u + h(t)u.
Using estimates (4.3), (4.4), (4.8) and (4.24), we obtain

IIJ1 (t + r) - J 1 (t)IIE-+E
(4.25)
~ II[A(t + r)[exp{ -(t + r - s)A(t + rn - exp{ -(t - s)A(tn]A- 1(s)IIE-+E
+11(A(t + r) - A(t))A- 1 (t)IIE-+ElII exp{ -(t - s)A(tnIIE-+E

xIIA(t)A- 1 (s)IIE-+E ~ Mw + Mir c ~ M2 w
for 0 ~ s < t ~ t + r ~ T. Now let us estimate h(t + r) - h(t) for 0 ~ s < t ~
~ T.
To this end we shall examine separately the two cases t-s ~ 2r and t-s > 2r.
If t - s ~ 2r, then using estimates (4.3), (4.4) and (4.18), we obtain
t +r
J
t
Ilh(t)IIE-+E
IIA(t) exp{ -(t - z)A(tnIIE-+E
x II[A(z) - A(t)]A -1 (z)IIE-+EIIA(z)v(z, s)A- 1(s)IIE-+Edz
J
t
M 1(t - z)c- 1dz = M 2 (t - sy.
From this estimate and the triangle inequality it follows that

Ilh(t + r) - h(t)IIE-+E ~ Ilh(t + r)IIE-+E + IIJ2(t)IIE-+E
~
Now let t - s
M2 (t + r - s)c + M2 (t - sy
(4.26)
M3 r c .
> 2r. We have that

h(t + r) - h(t)
J
t+r
A(t + r) exp{ -(t + r - z)A(t + r)}[A(z) - A(t + r)]v(z, s)A-I(s)udz
t-r
-J
t
t-r
A(t)exp{ -(t - z)A(tn[A(z) - A(t)]v(z, s)A- 1(s)udz
113
J
t-r
[A(t + r) exp{ -(t + r - z)A(t + r)}
-A(t) exp{ -(t - z)A(t)}][A(z) - A(t)]v(z, s)A- 1(s)udz
J
t-r
A(t + r) exp{ -(t + r - z)A(t + r)}[A(t) - A(t + r)]v(z, s)A- 1(s)udz
where
J
t+r
L 1(t)u =
A(t + r) exp{ -(t + r - z)A(t + r)}[A(z) - A(t + r)]v(z, s)A -1 (s)udz,
t-r
J
t
L 2 (t)u = -
A(t) exp{ -(t - z)A(t)}[A(z) - A(t)]v(z, s)A- 1(s)udz,
t-r
J
t-r
L 3 (t)u =
[A(t + r) exp{ -(t + r - z)A(t + r)}
-A(t) exp{ -(t - z)A(t)}][A(z) - A(t)]v(z, s)A- 1(s)udz,
t-r
L 4 (t)u =
A(t + r) exp{ -(t + r - z)A(t + r)}[A(t) - A(t + r)]v(z, s)A- 1(s)udz.
Using the estimates (4.3) and (4.4), we obtain
t+r
IIL 1 (t)IIE-.E:::;
IIA(t + r) exp{ -(t + r - z)A(t + r)}IIE-.E
t-r
x II[A(z) - A(t + r)]A- 1(z)IIE-.EIIA(z)v(z, s)A -1 (s)IIE-.Edz
: :; J
t+r
M 1(t + r - z)"-1dz = M2 r".
t-r
In a similar manner one establishes the estimate
114
J
t-r
IIL 3 (t)IIE-->E:S
{IIA(t + r)[exp{ -(t + r - z)A(t + r)}
- exp{ -(t - z)A(t + r)}]IIE-->E

+IIA(t + r)[exp{ -(t - z)A(t + r) - exp{ -(t - z)A(t)}]IIE-->E
+11[A(t + r) - A(t)]A- 1 (t)IIE-->EIIA(t) exp{ -(t - z)A(t)} IIE-->el
x II[A(z) - A(t)]A- 1 (z)IIE-->EIIA(z)v(z, s)A- 1 (s)IIE-->Edz
Now let us estimate L 4 (t). Using estimates (4.3), (4.4), (4.6) and (4.9), we obtain
J
t-r
IIL 4 (t)IIE-->E:S
IIA(t + r) exp{ -(t + r - z)A(t + r)}]IIE-->E
x II[A(t
+ r)
- A(t)]A -1 (z)IIE-->EIIA(z)v(z, s)A-l(s)IIE-->Edz
t-r
:S
t-r
r'" dz
M
t + r _ z:S 1
dz
(t + r _ z)l-"':S
M r '"
2 .
Combining the estimates for L 1 (t), L 2 (t), L 3 (t) and L 4 (t), we obtain estimate
(4.26) for t - s > 2r. Therefore,
IIJ2 (t + r) - h(t)IIE-->E :S M 3 r'"
for all 0 :S s < t :S t + r :S T. From this and (4.25) it follows (4.23). Applying
property 2) of v(t, s) and estimates (4.21) and (4.23), we obtain
IIA(t + r)v(t + r, s) - A(t)v(t, s)IIE-->E

:S II[A(t+r)v(t+r, t;s)_A(t)v(t, t;s)]A- 1 (t;s)IIe-->E
x IIA(t;s)v(t;s,S)IIe-->E :sMwM(t-S)-l =M1 (t-s)-l W.

Estimate (4.22) is established. Lemma 4.1.5 is proved.
A function v(t) is said to be a solution of problem (4.1) in F(E) if it is a
solution of this problem in C(E) and the function v'(t) and A(t)v(t) belong to
F(E).
As in the case of the space C(E), we say that problem (4.1) is well posed in
F(E), if the following two conditions are satisfied:
115
1. For any f(t) E F(E) and Vo E D there exists the unique solution v(t) =
v(t; f(t), vo) in F(E) of problem (4.1). This means that an additive and
homogeneous operator v(t; f(t), vo) is defined which acts from F(E) x D to
F(E) and gives the solution of (4.1) in F(E).
2. v(t; f(t), vo), regarded as an operator from F(E) x D to F(E), is continuous.Here F(E) x D is understood as the normed space of the pairs (f(t), vo),
f(t) E F(E) and Vo E D, equipped with the norm
11(f(t), vo)IIF(E)xD = IIfIlF(E) + II voliD.

First, we set F(E) equal to C~,Q(E) = C~,Q([O, T], E), 0 < 0: < 1, the Banach
space obtained by completion of the set of smooth E-valued functions cp(t) on [0, T]
in the norm
II cp IlcQQ(E)= IIcpIlC(E)
o
IIcp(t + 7) - cp(t) liE
sup
O:St<t+r:ST
(t + 7) .
Q
As in the case of the space C(E), from the well-posedness of the Cauchy problem
(4.1) one derives the coercivity inequality
Ilv/llc~'Q(E) + IIA(.)vllc~'Q(E) ::; MC(o:)[lIfllc~'Q(E)
IIA(O)voliel ,
(4.27)
where Mc (0:)(1 ::; Mc(o:) < +00) does not depend on Vo and f(t).
Theorem 4.1.6. Suppose Vo E D, f(t) E C~,Q(E). Suppose that assumptions (4.3)
and (4.4) hold and 0 < 0: ::; ::; 1. Then the Cauchy problem (4.1) is well posed
in Holder space C~,Q(E). For the solution v(t) in

the coercive inequality
IIv'IIc~'Q(E) + IIA(.)vllc~,Q(E) ::; 0:(1~ 0:)
C~,Q(E)
II f
of the Cauchy problem
Ilc~,Q(E) +M II A(O)voIlE
holds, where M does not depend on 0:, Vo and f(t).

Proof. If v(t) is a solution in C~,Q(E) of problem (4.1), then it is a solution in
C(E) of this problem. Hence, by (4.5), we have the representation
v(t) = v(t, O)vo
v(t, s)f(s)ds.
To prove the theorem we must show that the functions v'(t) and A(t)v(t) belong
to C~,Q (E) and finally obtain estimates for the norms of these functions. Here we
address only the last task. Using formula (4.5), we obtain
A(t)v(t)
A(t)v(t, O)vo
A(t)v(t, s)dsf(t)
3
A(t)v(t, s)(f(s) - f(t))ds
~ Jk(t),
(4.28)
116
where
J1 (t) = A(t)v(t,O)vo,
1
t
h(t)
A(t)v(t, s)dsf(t),
h(t) =
A(t)v(t, s)(I(s) - f(t))ds.
Let us estimate Jk(t) for any k = 1,2,3 separately. We start with J1 (t). Using
estimate (4.18), we show that, for 0::; t ::; T,
11J1(t)IIE::; IIA(t)v(t,0)A- 1 (0)IIE--+EIIA(0)voIIE::; MIIA(O)voIIE'

Further, applying (4.23), we show that, for 0 < t < t + 7 < T,
11J1 (t + 7) - h(t)IIE ::; IIA(t + 7)V(t + 7,0) - A(t)v(t, O)]A- 1 (O)IIE--+EIIA(O)voIlE

70.
::; M[7 + (t + 7)o.]II A(O)voIlE .
Since a ::; c, we have proved that
IIJ1 1Icg" (E) ::; MIIA(O)vo liE.
(4.29)
Now let us estimate J2(t). Using property 4) of vet, s), we can write
h(t) = [1 - A(t)v(t,O)A-1(t)]f(t)
1
t
A(t)v(t, s)[A(t) - A(s)]A- 1 (t)dsf(t).
Using this formula and estimates (4.4), (4.18), (4.21) and (4.24), we obtain
11J2(t) liE ::; [1 + IIA(t)v(t, O)A -1 (O)IIE--+EIIA(O)A -1 (t) IIE--+El IIf(t) liE
1
t
IIA(t)v(t, s)IIE--+EII[A(t) - A(s)]A- 1 (t)IIE--+Edsllf(t)IIE
::; Mllf(t)IIE + M
1
t
(t - s)-l+dsllf(t)IIE ::; M 1 II f Ilcg'''(E)
for all t, t E [0, T]. Now let us estimate J2(t + 7) - J2(t) for 0 ::; t < t + 7 ::; T.
To this end we shall examine separately the two cases t ::; 27 and t > 27. If
t ::; 27, then using the last estimate and the triangle inequality, we obtain
Ilh(t + 7) - J2(t)IIE ::; IIh(t + 7)IIE + Ilh(t)IIE ::; 2M1 II f IIcg'''(E)

70.
::; (t + 7)0. M 2 II f Ilcg'''(E) .
(4.30)
> 27. We have that
Now let t
J2(t + 7) - h(t)
{[I - A(t + 7)V(t + 7, O)A-1(t + 7)]
t+
A(t + 7)V(t + 7,s)[A(t + 7) - A(s)]A-1(t)ds}[f(t + 7) - f(t)]
+ Jo
+[A(t)v(t,O)A-1(t) - A(t + 7)V(t + 7,O)A- 1(t + 7)]f(t)
-J
t+T
A(t + 7)V(t + 7, s)[A(s) - A(t + 7)]A- 1(t)dsf(t)
t-T
J
t
A(t)v(t, s)[A(s) - A(t)]A-1(t)dsf(t)
t-T
-J +
-J
t-T
[A(t
7)V(t + 7, s) - A(t)v(t, s)][A(s) - A(t)]A-1(t)dsf(t)
t-T
A(t + 7)V(t + 7, s)[A(t) - A(t + 7)]A- 1(t)dsf(t)
J
t-T
A(t + 7)V(t + 7, s)[A(t + 7) - A(s)]A- 1(t)ds
x[A(t) - A(t + 7)]A- 1(t + 7)f(t + 7)

=
L1(t) + L 2 (t) + L 3 (t) + L 4 (t) + L 5 (t) + L6 (t) + L7 (t),
where
L1(t)
HI - A(t + 7)V(t + 7, O)A-1(t + 7)]
(+T
+J
A(t + 7)V(t + 7, s)[A(t + 7) - A(s)]A-1(t)ds}[f(t + 7) - f(t)],
o
L 2 (t) = [A(t)v(t, O)A-1(t) - A(t + 7)V(t + 7,O)A- 1(t + 7)]f(t),
J
t+T
L 3 (t)
= -
A(t + 7)V(t + 7, s)[A(s) - A(t + 7)]A- 1(t)dsf(t),
t-T
J
t
L 4 (t)
A(t)v(t, s)[A(s) - A(t)]A-1(t)dsf(t),
t-T
117
118
J
t-r
L 5 (t) = -
[A(t
+ r)v(t + r, s) -
A(t)v(t, s)][A(s) - A(t)]A- 1 (t)dsf(t),
t-r
A(t + r)v(t + r, s)[A(t) - A(t + r)]A- 1 (t)dsf(t),
L 6 (t) = -
J
t-r
L 7 (t) =
X
A(t + r)v(t + r, s)[A(t + r) - A(s)]A- 1 (t)ds
o
[A(t) - A(t + r)]A- 1 (t + r)f(t + r).
Let us estimate Lk(t) for any k = 1,2,3,4,5,6,7 separately. We start with L 1 (t).
IIL 1 (t)IIE
~ [1 + IIA(t+r)v(t+r,0)A- (0)IIE--.EIIA(0)A- (t+r)IIE--.E]llf(t+r) - f(t)IIE
1
(t+ r
+ Jo
IIA(t + r)v(t + r, s)IIE--.EII[A(t + r) - A(s)]A- 1 (t)IIE--.Eds
x Ilf(t + r) - f(t)IIE
~ [M + M
1
t
+ (t
ro.
~ M 1 (t + r)o.
II f
+ r - s)l-"ds]llf(t + r) - f(t)IIE
Ilcg'''(E)
for all t, t E [0, T]. Let us estimate L 2 (t). Using the formula
L 2 (t)
[A(t)v(t,O) - A(t + r)v(t
+ r, O)]A -1 (t)f(t)
+A(t + r)v(t + r,0)[A- (t) - A- 1 (t + r)]f(t),

1
and estimates (4.3), (4.4), (4.8), (4.23) and (4.24), we obtain

IIL 2 (t)IIE
II[A(t)v(t,O) - A(t + r)v(t + r,0)]A- 1 (0)IIE--.E
x IIA(O)A -1 (t)IIE--.Ellf(t)IIE
+IIA(t + r)v(t + r,0)A- 1 (t)IIE--.EII[A(t + r) - A(t)]A- 1 (t + r)IIE--.Ellf(t)IIE
ro.
~ M[r" + (t + r)o.]Mllf(t)IIE + Mr"llf(t)IIE
for all t, t E [0, T]. Since a ::;
we have proved that
119
for all t, t E [0, T]. Let us estimate L 3 (t). Using estimates (4.3), (4.4), (4.8), (4.23)
J
t+r
IIL3 (t)IIE:S
IIA(t+r)v(t+r, s)IIE-+EII[A(s) - A(t+r)]A- (t)IIE-+Ellf(t)IIEds
t-r
J
t+r
:s
M 1 (t
+ r - s)O-ldsllf(t)IIE
= M 2r
II f Ilc~O(E)
t-r
for all t, t E [0, T]. Therefore,
for all t, t E [0, T]. In a similar manner one establishes the estimate
for all t, t E [0, T]. Now let us estimate L 5 (t). Using estimates (4.4) and (4.22), we
obtain
J
t-r
IIA(t + r)v(t + r, s) - A(t)v(t, s)IIE-+E
IIL5 (t)IIE:S
xll[A(s) - A(t)]A- 1 (t)IIE-+Edsllf(t)IIE
:s M 1
J
t-r
J(
t-r
(t - s)-l[r
+(
rO ) ]dsllf(t)IIE:S M 2
t+r-so
:s
rOds )1+ Ilf(t)IIE

t+r-s
rO
~ (t + r)O II f Ilc~,O(E)
M3
for all t, t E [0, T]. Now let us estimate L 6 (t). Using estimates (4.4) and (4.21),
we obtain
t-r
IIL 6 (t)IIE
:s
IIA(t + r)v(t + r, s)IIE-+EII[A(t + r) - A(t)]A- 1 (t)IIE-+Edsllf(t)IIE
J+
t-r
:s M 1
(t
J + ~~:)1+0
t-r
T -
O
s)-l r dsllf(t)IIE
:s M 2
(t
M3
TO
:s ~ (t + T)o II f
IIc;'O(E)
IIf(t)IIE
120
for all t, t E [0, T]. Finally we have
J
t+T
IIL7 (t)IIE:::;
M 1 (t+r-s)-1+ dsM2 r llf(t+r)IIE :::; M3 (t
::)<> II f IIc~'Q(E) .
Combining the estimates for Lk(t) for any k = 1,2,3,4,5,6,7, we obtain

estimate (4.30) for t > 2r. Therefore,
Ilhllc~Q(E)
:::; -Cl:' II f
IIc~,Q(E)
(4.31)
Now let us estimate J3 (t). Using estimate (4.21), we obtain
J
t
II J3(t) liE:::;
IIA(t)v(t, s)IIE--;Ellf(s) - f(t)
liE ds
ds
(t _ S)l-<>t<> II f
:::; M
M
Ilc~'Q(E)= ~
I f Ilc~,Q(E)
E [0, T]. Now let us estimate the difference h(t + r) - h(t) for
T. We will consider separately the cases t :::; 2r and t > 2r. If
then from the last estimate it follows that
< 2r,t <t, tt+ r <

for all
t :::;
II J3(t + r) - J3(t) IIE:::;II J3(t + r) liE + II J3(t) liE
: :; 2~3<> II f IIc~'Q(E) r<>(t + r)-<> = ~l II f Ilc~,Q(E) r<>(t + r)-<>.

Now let t > 27. Let us represent the difference h(t + r) - h(t) as the sum of the
following integrals:
J
t+T
h(t + r) - J3(t) =
A(t + r)v(t + r, s)(I(s) - f(t + r))ds
t-T
-J
J +
t
A(t)v(t, s)(I(s) - f(t))ds
t-T
t-T
A(t
r)v(t + r, s)(I(t) - f(t + r))d8
t-T
[A(t + r)v(t + r, s) - A(t)v(t, 8)](I(S) - f(t))ds
121
where
t+r
PI (t) =
A(t + T)V(t + T, s)(f(s) - f(t + T))ds,
t-r
J
t
P2(t) = -
A(t)v(t, s)(f(s) - f(t))ds,
t-r
J
t-r
[A(t + T)V(t + T, s) - A(t)v(t, s)](f(s) - f(t))ds,
P3 (t) =
J
t-r
A(t + T)V(t + T, s)(f(t) - f(t + T))ds.
P4 (t) =
Let us estimate Pk(t) for any k = 1,2,3,4 separately. We start with PI (t). Using
estimate (4.21), we obtain
J
t+r
IIA(t + T)V(t + T, s)IIE--+Ellf(s) - f(t + T)IIEds
IIP1 (t)IIE:::;
t-r
t+r
:::; (
t+T ) o
M 1 (t
+T
t-r
s)
0-1
M2
TO
ds II f IICn,n(E)=
-a (t+T )O II f IIC0n,n(E)
0
for all t, t E [0, T]. Now let us estimate P3(t). Using estimate (4.22), we obtain
J
t-r
11P3(t)IIE:::;
IIA(t + T)V(t + T, s) - A(t)v(t, s)IIE--+Ellf(s) - f(t)IIEds
t-r
1
:::; M
to
(t - s)-HO[T" +
+T
]ds II f Ilcn.n(E)
0
o
M2To
M3 To
:::; a(l - a)t o II f Ilc;,n(E):::; a(l _ a)(t + T)o II f Ilc;,n(E)
for all t, t E [0, T]. Let us estimate P4 (t). Using property 4) of v(t, s), we can
write
P4 (t) = [A(t + T)V(t + T, t - T)A- 1 (t + T)
122
-A(t + r)v(t + r, O)A-I(t + r)](J(t) - f(t + r))

t
A(t + r)v(t + r, s)[A(t + r) - A(s)]A-I(t + r)ds(J(t) - f(t
+ r)).
IIP4 (t)IIE:::; [IIA(t + r)v(t + r, t - r)A-I(O)IIE--+E

+IIA(t + r)v(t + r, O)A-I(O)IIE--+eJIIA(O)A-I(t + r)IIE--+Ellf(t) - f(t + r)IIE
+
IIA(t + r)v(t + r, s)IIE--+EII[A(t + r) - A(s)]A-I(t + r)IIE--+Eds
xllf(t) - f(t
+ r)IIE :::; 2MMllf(t) - f(t + r)IIE
I
t
+M
(t + r - s)I-Cdsllf(t) - f(t + r)IIE
MIrO:
:::; (t + r)o: II f Ilc;'''(E)
for all t, t E [0, T]. Combining the estimates for Pk(t) for any k = 1,2,3,4, we
obtain the estimate
for t > 2r. Therefore,
Ilhllc;'''(E) :::;
a(l _ a)
II f Ilc;'''(E) .
(4.32)
Finally, applying estimates (4.29), (4.31) and (4.32), we obtain the estimate
II A(.)v Ilc;'''(E):::;
M
a(l _ a)
II f IIc;'''(E) +M II A(O)voIIE'
By the triangle inequality, this last estimate and equation (4.1) yields
II v'llc;'''(E) :::;
a(l~ a)
II f Ilc;'''(E) +M II A(O)voIIE'

With the help of A(t) we introduce the fractional spaces Eo:,q(E, A(t)), 0 <
a < 1, consisting of all vEE for which the following norms are finite:
00
dz
II V IIE",q = ( [ZI-O: II A(t) exp{ -zA(t)}v IIE]q ~)"q, 1:::; q < 00,
1
II v IIE,,=II V IIE",oo= SUpZI-o: II A(t)exp{-zA(t)}v liE.

z>o
123
From (4.3) and (4.4) follows

Theorem 4.1.7. Eo:,q(E, A(t)) = Eo:,q(E, A(O)) for all 0
o ~ t ~ T.
< 0: < 1, 1 ~ q ~
00
and
Note that the coercivity inequality
Ilv'IIF(E) + IIA(.)vIIF(E)
~ MC[llfIlF(E)
+ IIA(O)voIIE]
fails if we set F(E) equal to CO:(E), (0 < 0: < 1), the Banach space obtained by
completion of the set of all smooth E-valued functions ep(t) on [O,T] in the norm
I ep Ilc<>(E)=
max
05,t5,T
lIep(t)IIE +
sup
05,t<t+r5,T
II ep(t + T) -
ep(t)
liE
TO:
Nevertheless, we have the following result.

Theorem 4.1.8. Suppose vb = f(O) - A(O)vo E Eo:, f(t) E CO:(E)(O < 0: < 1).
Suppose that assumptions (4.3) and (4.4) hold and 0 < 0: ~ E ~ 1. Then the
Cauchy problem (4.1) is well posed in Holder space CO:(E). For the solution v(t)
in CO:(E) of the Cauchy problem the coercive inequality
holds, where M does not depend on
0:,
Vo and f(t).
Proof. By formula (4.28), we have

V' (t)
= A(t)v(t, O)A -1 (O)vb
+ A(t)v(t, O)A -1 (O)(A(O) - A(t))A -1 (t)f(t)
+A(t)v(t, O)A- 1 (0)(J(t) - f(O))
-I
A(t)v(t, s)[A(t) - A(s)]A- 1 (t)dsf(t)
r A(t)v(t, s)(J(s) - f(t))ds L Jk(t),

o
- in
k=1
where
J 1 (t) = A(t)v(t, O)A -1 (O)vb,

h(t)
A(t)v(t, 0)A- 1 (0)(A(0) - A(t))A- 1 (t)f(t),
h(t) = A(t)v(t,0)A- 1 (0)(J(t) - f(O)),

J4 (t) =
-I
J5 (t)
A(t)v(t, s)[A(t) - A(s)]A- 1 (t)dsf(t),
-I
A(t)v(t, s)(J(s) - f(t))ds.
124
Let us estimate Jk(t) for any k = 1,2,3,4,5 in CQ(E) and C(EQ) separately. We
start with J1(t). Using estimate (4.18), we show that, for 0 :S t :S T,
11J1(t)IIE:S IIA(t)v(t,O)A-l(O)IIE--.Ellv~IIE:S Mllv~IIE:S Mllv~IIEa'

Further, applying (4.16), we can write
J1(t) = exp{ -tA(tnv~ + exp{ -tA(tn[A(t) - A(O)]A-l(O)V~
(4.33)
+
=
A(t)exp{-(t-s)A(tn[A(s)
-A(t)]v(s,O)A-l(O)v~ds
o
J ll (t) + J 12(t) + J 13 (t),
where
J ll (t) = exp{ -tA(tnv~,

J 12 (t) = exp{ -tA(tn[A(t) - A(O)]A-l(O)V~,
J
t
J 13 (t) =
A(t) exp{ -(t - s)A(tn[A(s) - A(t)]v(s,O)A-1(O)vbds.
o
IIJll (t+r) -Jll(t)IIE :S II exp{ -(t+r)A(t+r)}vb -exp{ -(t+r)A(t)}vbIlE
(4.34)
JIIA(t) + s)A(tnv~dsIIE
+ J ::~~_Q "
~l
T
exp{ -(t
:S MrllvbllE
(t
vbllEa :S
rQllvbllEa
o
for 0 < t < t + r < T. Using estimates (4.3), (4.6) and (4.8), we obtain
IIJ12 (t + r) - J 12 (t)IIE :S II exp{ -(t + r)A(t + rnllE--.E

x II [A(t
<
(4.35)
+ r) - A(t)]A-1(O)IIE--.EllvbIIE
II exp{ -(t + r)A(t + rn - exp{ -(t + r)A(tnIlE--.E

x II [A(t)
- A(O)]A-1(O)IIE--.EllvbIlE
+11 exp{ -(t + r)A(t)} - exp{ -tA(t)} IIE--.EII [A(t) - A(O)]A-1(O)IIE--.EllvbIlE

r
:S M Mr II vb liE + Mr MtllvbllE + M (t + r)Q MtllvbllE :S M1rQllvbllE a
125
for 0 < t < t + 7 < T. Let us estimate J 13 (t + 7) - Jdt) for 0 < t < t + 7 < T.
Using estimates (4.3), (4.4) and (4.18), we obtain
J
t
IIJ13 (t)IIE ::;
IIA(t) exp{ -(t - s)A(tnIIE-+EII[A(t) - A(s)]A- 1(O)IIE-+E
J
t
x IIA(s)v(s, O)A- (O)IIE-+EdsllvbIIE ::;
M M(t - s)-He MdsllvbllE
1
1
::; M tQllvbllE ::; M t II vb IIE
Q
0:
0:
If t ::; 27, then using the last estimate and the triangle inequality, we obtain
IIJ13 (t + 7) - J13(t)IIE ::; IIJ13 (t + 7)IIE + 11J13(t)IIE

1
Q
::; M ((t+7)Q+t )lIvbIIEQ::; M27QllvbIlEQ'
0:
0:
Now let t > 27. We have that
J
t+r
A(t + 7) exp{ -(t + 7 - s)A(t + 7n[A(s) - A(t + 7)]V(S, O)A- 1(O)vbds
t-r
-J
t
A(t)exp{-(t-s)A(t)}[A(s)-A(t)]v(s,O)A-1(O)vbds
t-r
J
t-r
[A(t + 7) exp{ -(t + 7 - s)A(t + 7n - A(t)
x exp{ -(t - s)A(tn][A(s) - A(t)]v(s, O)A- 1 (O)vbds
t-r
A(t + 7) exp{ -(t + 7 - s)A(t + 7n[A(t) - A(t + 7)]V(S, O)A- 1(O)vbds
where
J
t+r
K 1(t) =
A(t + 7) exp{ -(t + 7
t-r
s)A(t + 7n[A(s) - A(t + 7)]V(S, O)A -1 (O)vbds,
126
J
t
K 2 (t) = -
A(t) exp{ -(t - s)A(tn[A(s) -
A(t)lv(s,O)A-1(O)v~ds,
t-T
J
t-T
K 3 (t) =
[A(t
+ r) exp{ -(t + r - s)A(t + rn - A(t)
x exp{ -(t - s)A(tn][A(s) - A(t)]v(s,O)A-1(O)v~ds,
t-T
K 4 (t) =
A(t + r) exp{ -(t + r - s)A(t + rn[A(t) - A(t + r)]v(s, O)A-1(O)v~ds.
o
JIIA(t +
t+T
IIK1 (t)IIE:::;
r) exp{ -(t + r - s)A(t + rnllE--+EI
t-T
x I[A(s) - A(t + r)]A- 1(s)IIE--+E

x IIA(s)v(s, O)A -1 (O)IIE--+Ellv~IIEds
: ; JMM(t+r-s)-l+Mdsllv~IIE:::; ~lrallv~IIEQ.
t+T
t-T
J
t-T
IIK3 (t)IIE :::;
IIA(t+r) exp{ -(t+r-s)A(t+rn - A(t) exp{ -(t-s)A(tnIlE--+E
x II [A(s) - A(t)]A- (s)IIE--+EIIA(s)v(s, O)A-1(O)IIE--+Ellv~IIEds
: :; J
(;1 )
t-T
MM(t + r - s)-1+[r
o
Applying the formula
+ _r_]Mdsllv~IIE:::;
rallv~IIEQ.
t-s
(} -(}
K 4 (t) = [exp{ -2rA(t + rn
exp{ -(t + r)A(t
x [A(t) - A(t + r)]v(t,O)A-1(O)v~
+ Tn]
127
t-T
A(t+T) exp{ -(t+T- S)A(t+T) }[A(t) - A(t+T)][V(S, 0) -v(t, O)]A -1(O)v~ds
o
and estimates (4.3), (4.4), (4.18) and (4.23), we obtain
IIK4 (t)IIE:S II exp{ -2TA(t + T)} - exp{ -(t + T)A(t + T)}IIE-+E
x II[A(t) - A(t + T)]A -1 (t)IIE-+EIIA(t)v(t, 0)A- 1(O)IIE-+Ellv~IIE
J
t-T
IIA(t + T) exp{ -(t + T - s)A(t + T)}IIE-+E
o
x II [A(t + T) - A(t)]A -1 (s)IIE-+EIIA(s)(v(t, 0) - v(s, O))A -1 (O)IIE-+Ellv~IIEds
t-T
:S MT
Mllv~IIE
MM(t + T - S)-I T[T +
_T-]Mdsllv~IIE
t-s
a(~ a) Tallv~IIE".
:S
Combining the estimates for Kk(t) for any k = 1,2,3,4, we obtain the estimate
IIJI3 (t + T) - Jdt)IIE :S
a(~ a) Tallv~IIE"
for t > 2T. Therefore,

IIJI3 (t + T) - J I3 (t) liE :S
for 0 < t < t +
a(~ a) TallvbllE"
< T. Using estimates (4.3), (4.6) and (4.8), we obtain

II J lllc"(E) :S
a(~ a) IlvbIIE".
(4.36)
Using formula (4.33) and estimates (4.3), (4.6) and (4.8), we obtain
zl-a
II A(t)exp{-zA(t)}J1(t)IIE:S zl-a II
A(t)exp{-(z+ t)A(t)}vbIIE
+zl-a II A(t) exp{ -(z + t)A(t)} IIE-+EII [A(t) - A(0)]A- 1(0)IIE-+EllvbIlE
J
t
zl-a II A2 (t) exp{ -(z + t - s)A(t)}IIE-+EII[A(s) - A(t)]A- 1(s)IIE-+E
I-a
xIlA(s)v(s,O)A-l(O)IIE-+Ellv~IIEds:S(t +z)l
z -a II v~IIE"
Mt
+z l_a Z+tII'11
Vo E +
J
t
M (t-s)ds 'll
M111 Vo'II E"
z l_a (z+t-s
)2 ll Vo E:S I-a
128
for all z > 0 and t, t E [0, T]. Therefore,

(4.37)
Now let us estimate h(t). Using estimates (4.4), (4.18), we show that, for
o :s t :s T,
Ilh(t)IIE:S IIA(t)v(t,0)A-1(0)IIE-+EII(A(0) - A(t))A-1(t)IIE-+EII!(t)IIE
:s MMt"II!(t)IIE :s MIt" II! IIc<>(E) .

Therefore,
Moreover, if t
obtain
:s 2r, then using the last estimate and the triangle inequality, we
Ilh(t + r) - h(t)IIE
:s Ilh(t + r)IIE + Ilh(t)IIE
:s M 2 t + r)Q + tQ)II! Ilc<>(E):S M3 r QII! Ilc<>(E) .

Now let t > 2r. Using estimates (4.3), (4.6), (4.23), (4.24) and (4.18), we obtain
Ilh(t + r) - J2 (t)IIE
:s IIA(t + r)v(t + r, 0)A-1(0)IIE-+E
xll(A(O) - A(t + r))A-1(t + r)IIE-+EII!(t + r) - !(t)IIE

+II[A(t + r)v(t + r,O) - A(t)v(t,0)]A-1(0)IIE-+E
x II(A(O) - A(t + r))A-1(t + r)IIE-->E1I!(t)IIE
+IIA(t)v(t, 0)A- 1(O)IIE-+EIIA(O)A -1 (t)IIE-+E
xll(A(t) - A(t + r))A-1(t + r)IIE-+EII!(t)IIE
:s MM(t + r)"rQII! Ilc<>(E) +M[r" + :: ]M(t + r)"II! Ilc<>(E)

+MMr"ll! Ilc<>(E):S M1rQII! Ilc<>(E)
Ilhllc<>(E)
:s M1rQII! Ilc<>(E) .
(4.38)
Using formula (4.33) and estimates (4.3), (4.4), (4.6) and (4.8), we obtain
Zl-Q II A(t) exp{ -zA(t)}J2 (t)IIE
:s zl-Q II A2 (t) exp{ -(z+t)A(t)}A-1(0)IIE-+EII(A(0) -
A(t))A -1 (t)IIE-+EII!(t)IIE
+Zl-Q II A (t) exp{ -(z + t)A(t)}A-1(0)IIE-->E

2
129
xll(A(O) - A(t))A- 1(0)IIE-+EII(A(0) - A(t))A- 1(t)IIE-+EII!(t)IIE
J
t
Zl-c> II A2(t) exp{ -(z + t - s)A(t)}A-1(0)IIE-+EII[A(s) - A(t)]A-1(s)IIE-+E
x IIA(s)v(s, O)A -1 (0) IIE-+EIIE-+EII(A(O) - A(t))A -1 (t)IIE-+EII!(t)IIEds

Zl-c>t'"
:::; --II!(t)IIE
t+z
J
t
Me'"
z+t
+ zl-c>-II!(t)IIE
Zl-c> M(t - s)"'t"'ds 11!(t)IIE :::; M 1II!llo(E )

z+t-s
0:
<>

(4.39)
Now let us estimate J3 (t). Using estimate (4.18), we show that, for 0 :::; t :::; T,
IIh(t)IIE :::; IIA(t)v(t,0)A- 1(0)IIE-+EII!(t) - !(O)IIE

:::; Mt'" II! IIo<>(E):::; Mt'" II! IIo<>(E) .
Therefore,
Moreover, if t :::; 2r, then using the last estimate and the triangle inequality, we
obtain
Ilh(t + r) - J3(t)IIE :::; Ilh(t + r)IIE + Ilh(t)IIE
::; M 2((t
+ r)C> + tcII! IIo<>(E):::; M2rC>II! 110<>(E) .
Now let t > 2r. Using estimates (4.23) and (4.18), we obtain
Ilh(t + r) - h(t)IIE :::; IIA(t + r)v(t

+II[A(t + r)v(t
+ r, 0)A- 1(0)IIE--+EII!(t + r) - !(t)IIE
+ r,O) - A(t)v(t, 0)]A-1(0)IIE--+EII!(t) - !(O)IIE
:::; M r C>II! 110<>(E) +M[r'"
+ :: ]MtC> II! 110<>(E):::; M 1rC>II! 110<>(E)
IIhllo<>(E) :::; M1rC>II! IIo<>(E) .

Using formula (4.33) and estimates (4.3), (4.4), (4.6) and (4.8), we obtain
Zl-c> II A(t) exp{ -zA(t)}h(t)IIE
(4.40)
130
:::; Zl-Q II A2 (t) exp{ -(z + t)A(t)}A-1(0)IIE--+Ellf(t) - f(O)IIE

+Zl-Q II A2 (t) exp{ -(z + t)A(t)}A-1(0)IIE--+E
Jzl-Q II
xII(A(O) - A(t))A-1(0)IIE--+Ellf(t) - f(O)IIE
A2 (t) exp{ -(z + t - s)A(t)}A-1(0)IIE--+EII[A(s) - A(t)]A-1(s)IIE--+E
xIIA( s)v(s, O)A -1 (O)IIE--+E IIE--+E Ilf(t) - f(O) II Eds

MtQ+c
zl-Qt Q
:::; --llfIIC(E
)
+
zl-Q--llf(t)IIE
t+Z
a
Z+t
JZl-Q
t
M(t - s)ctQds Ilf(t)IIE :::; MlllfllC(E )

Z+t-s
Q:
a

M1
Ilhllc(Ea) :::; ~llfIIC(Ea)'
(4.41 )
Using this formula and estimates (4.4) and (4.21), we obtain
IIJ4 (t)IIE :::;
it
:::; M
IIA(t)v(t, s)IIE--+EII[A(t) - A(s)]A-1(t)IIE--+Edsllf(t)IIE
1
rt
(t - s)l-Cdsllf(t)IIE :::; M tQII f IIca(E)
h
Q:
for all t, t E [0, T]. Now let us estimate J4 (t + r) - J4 (t) for 0 :::; t < t + r :::; T.
To this end we shall examine separately the two cases t :::; 2r and t > 2r. If
t :::; 2r, then using the last estimate and the triangle inequality, we obtain
IIJ4 (t + r) - J4 (t)IIE :::; IIJ4 (t + r)IIE + IIJ4 (t)IIE
(4.42)
1tQ] II f Ilca(E):::; M 2 r II f Ilca(E) .

:::; [MQ:1(t + r)Q + M
Q:
Q
Now let t > 2r. We have that
rt+- A(t + r)v(t + r, s)[A(t + r) r
= - Jo
-J
A(s)]A-1(t + r)ds[f(t + r) - f(t)]
t+T
A(t + r)v(t + r, s)[A(s) - A(t + r)]A-1(t)dsf(t)
t-T
131
A(t)v(t, s)[A(s) - A(t)]A-1(t)dsj(t)
t-7'
-J
t-7'
[A(t + r)v(t + r, s) - A(t)v(t, s)][A(s) - A(t)]A-1(t)dsj(t)
-J
t-7'
A(t + r)v(t + r, s)[A(t) - A(t + r)]A-1(t)dsj(t)
J
t-7'
A(t + r)v(t + r, s)[A(t) - A(s)]A-1(t)[A(t) - A(t + r)]A-1(t + r)dsj(t + r)
where
t+7'
L1(t) = - io
A(t + r)v(t + r,s)[A(t + r) - A(s)]A-1(s)ds}[j(t + r) - j(t)],
t+7'
L2 (t)
=-
A(t + r)v(t + r, s)[A(s) - A(t + r)]A-1(s)dsj(t),
t-7'
J
t
L3 (t) = -
A(t)v(t, s)[A(s) - A(t)]A-1(s)dsj(t),
t-7'
t-7'
L4 (t) =
[A(t + r)v(t + r, s) - A(t)v(t, s)][A(s) - A(t)]A-1(s)dsj(t),
J
t-7'
Ls(t) =
A(t + r)v(t + r, s)[A(t) - A(t + r)]A-1(s)dsj(t),
t-7'
L6 (t) =
A(t + r)v(t + r, s)[A(t) - A(s)]
A-1(t)[A(t) - A(t + r)]A-1(t + r)dsj(t + r).
132
Let us estimate Lk(t) for any k = 1,2,3,4,5,6 separately. We start with L 1 (t).
t+
IIL 1 (t)IIE :S io
IIA(t + T)V(t + T, s)IIE--+EII[A(t + T) - A(s)]A- 1 (s)IIE--+Eds
x Ilf(t + T) - f(t)IIE
t +T
M
:S M
(t + T - s)l- Edsllf(t + T) - f(t)IIE :S _1 TO II f Ilco(E)
o
a
for all t, t E [0, T]. Let us estimate L 2 (t). Using estimates (4.3), (4.4), (4.8), (4.23)
t+T
IIL 2 (t)IIE :S
IIA(t + T)V(t + T, s)IIE--+EII[A(s) - A(t + T)]A- 1 (t)IIE--+Ellf(t)IIEds
t-T
J
t+T
:S
M 1 (t + T - s)E- dsllf(t)IIE =
1
~2TE II
f Ilco(E)
t-T
for all t, t E [0, T]. Thus,
IIL2 (t)IIE :S M 2T o II f Ilco(E)

a
for all t, t
[0, T]. In a similar manner one establishes the estimate
for all t, t E [0, T]. Now let us estimate L 4 (t). Using estimates (4.4) and (4.22),
we obtain
J
t-T
IIL4 (t)IIE:S
IIA(t + T)V(t + T, s) - A(t)v(t, s)IIE--+E
xll[A(s) - A(t)]A- 1 (t)IIE--+Edsllf(t)IIE
:S M
a
TO
II f Ilco(E)
for all t, t E [0, T]. Now let us estimate L 5 (t). We have that
L 5 (t)
= [exp{ -2TA(t
Tn - exp{ -(t + T)A(t + Tn][A(t) - A(t + T)]A- (t)f(t)

1
133
t-T
[A(t + r)v(t + r, s) - A(t + r) exp{ -(t + r - s)A(t + rn]
x [A(t) - A(t + r)]A-1(t)dsf(t)
t-T
A(t + r)v(t + r, s)[A(t) - A(t + r)]A-1(t)[A(s) - A(t)]A-1(t)dsf(t).
IIL5 (t)IIE ~ II exp{ -2rA(t + rn - exp{ -(t + r)A(t + rnllE--+E

x II[A(t + r) - A(t)]A-1(t)IIE--+Ellf(t)IIE
t-T
IIA(t + r)v(t + r, s) - A(t + r) exp{ -(t + r - s)A(t + rnllE--+E
x II[A(t + r) - A(t)]A- 1(t)IIE--+Edsllf(t)IIE
J
t-T
IIA(t + r)v(t + r, s)IIE--+EII[A(t + r) - A(t)]A-1(t)IIE--+E
xll[A(s) - A(t)]A-1(t)IIE--+Edsllf(t)IIE
J
t-T
[Mr c + M
(t - s)-l+crcds
J
t-T
+M1
(t + r - s)-l+crcdsJllf(t)IIE
~ ~2rQ II f
Ilc<>(E)
for all t, t E [O,T]. Finally let us estimate L 6 (t). Using estimates (4.3), (4.4) and
(4.21), we obtain
t-T
IIL6 (t)IIE
IIA(t + r)v(t
+ r, s)IIE--+E
xll[A(t) - A(s)]A-1(t)IIE--+EII[A(t + r) - A(t)]A-1(t + r)IIE--+Edsllf(t + r)IIE
t-T
~ M1
(t - s)-l+crcds II f
IIc<>(E)~ ~2 II f
Ilc<>(E)
134
for all t, t E [0, T]. Combining the estimates for Lk(t) for any k = 1,2,3,4,5,6,
we obtain estimate (4.30) for t > 2T. Therefore,
(4.43)
Using estimates (4.3), (4.4) and
II A 2(t) exp{ -zA(t)}v(t, s)IIE->E ::; M min{

we obtain
I II
t
::; Zl-Q
:2'
(t! S)2}'
(4.44)
II A(t) exp{ -zA(t)}J4 (t)IIE
Zl-Q
A2 (t) exp{ -zA(t)}v(t, s)IIE->EII[A(t) - A(s)]A-1(t)IIE->Ellf(t)IIE

t
1 QM(t - s)E:ds
z - (z + t _ s)2 Ilf(t)IIE
::;
M1
1_
ex Ilfllc(Ea)

(4.45)
Now let us estimate J5 (t). Using estimate (4.21), we obtain
J
t
II
liE::;
J5 (t)
IIA(t)v(t, s)IIE->Ellf(s)
- f(t) liE ds
J
t
::; M
(t _ ~~l-Q
II f IIc a(E)=
~t
II f Ilca(E)
o
for all t, t E [0, T]. Now let us estimate the difference J5 (t + T) - J5 (t) for
o < t < t + T < T. We will consider separately two cases, t ::; 2T and t > 2T. If
t ::; 2T, then the last estimate yields
II
J5 (t
+ T) - J5 (t) IIE::;II J5 (t + T) liE + II J5 (t) liE
::; M [(t
ex
+ Tt + t II f Ilca(E)=
Q
ex
II f Ilca(E) .
Now let t > 2T. Let us represent the difference J5 (t + T) - J5 (t) as the sum of the
J
t+r
J5 (t + T) - J5 (t) =
A(t + T)V(t + T, s)(J(s) - f(t + T))ds
t-r
135
-J
t
A(t)v(t, s)(J(s) - f(t))ds
t-T
t-T
A(t + r)v(t + r, s)(J(t) - f(t + r))ds
J
t-T
+ r)v(t + r, s) -
[A(t
A(t)v(t, s)](J(s) - f(t))ds
where
t+T
PI (t) =
A(t + T)V(t + r, s)(J(s) - f(t + T))ds,
t-T
f
t
P2(t) = -
A(t)v(t, s)(J(s) - f(t))ds,
t-T
t-T
P3 (t) =
[A(t + T)V(t + T, s) - A(t)v(t, s)](J(s) - f(t))ds,
J
t-T
P4 (t) =
A(t + T)V(t + T, s)(J(t) - f(t + T))ds.
o
Let us estimate Pk(t) for any k
= 1,2,3,4 separately. We start with PI (t).
Using
t+T
IIPI(t)IIE
IIA(t + T)V(t
+ r, s)IIE_Ellf(s) - f(t + r)IIEds
t-T
:; J
t+T
MI(t + r - s)a-Ids
II
IIC"(E)=
~2ra I
IIC"(E)
t-T
11P2(t)IIE :::; MIra II f
IIc"(E)
for all t, t E [0, T]. Now let us estimate P3 (t). Using estimate (4.22), we obtain
t-T
11P3(t)IIE:::;
IIA(t + T)V(t + T, s) - A(t)v(t, s)IIE-+Ellf(s) - f(t)IIEds
136
J
t-r
..::; MI
(t - s)-HQ[r" +
2rQ
r
]ds II f Ilco(E)"::;
) II f Ilco(E)
t+r-s
n1-n
for all t, t E [0, T]. Let us estimate P4 (t). Using property 4) of v(t, s), we can
write
P4 (t) = [A(t + r)v(t + r, t - r)A-I(t + r)
-A(t + r)v(t + r, O)A-I(t + r)](f(t) - f(t + r))

t r
-
A(t + r)v(t + r, s)[A(t + r) - A(s)]A-I(t + r)ds(f(t) - f(t + r)).
IIP4 (t)IIE"::; UIA(t + r)v(t + r, t - r)A-I(O)IIE-+E

+IIA(t + r)v(t + r, O)A-I(O)IIE-+E]IIA(O)A-I(t + r)IIE-+Ellf(t) - f(t + r)IIE
+
t r
-
IIA(t + r)v(t + r, s)IIE-+EII[A(t + r) - A(s)]
xA-I(t + r)IIE-+Edsllf(t) - f(t + r)IIE ..::; 2MMllf(t) - f(t + r)IIE

+M
t r
-
(t + r - s)-H"dsllf(t) - f(t + r)IIE ..::; MIr II f IIcO(E)

Q
for all t, t E [0, T]. Combining the estimates for Pk (t) for any k = 1, 2, 3, 4, we
obtain the estimate
II J5 (t + r) - J5 (t) liE"::;
n(~ n)
II f Ilco(E) r

M
IIJ5 1Ico(E) ..::; n(l _ n) II f Ilco(E) .

..::; ZI-Q
ZI-Q II A(t) exp{ -zA(t)}J5 (t) liE

t
II A2 (t) exp{ -zA(t)}v(t, s)IIE-+Ellf(t) - f(s)IIE
(4.46)
137

for all z
> and t, t
E [0, T]. Therefore,
IIJsllc(EQ) ~
M1
1_ a
Ilfllc(EQ).
(4.47)
Finally, applying estimates (4.36), (4.37), (4.38), (4.39), (4.40), (4.41), (4.43),
(4.45), (4.46) and (4.47), we obtain the estimates
I v' Ilc(EQ) ~
II v' IlcQ(E)
a(l~ a) [II f
IlcQ(E)
+ I vbllEal,
~ a(l~ a) [II f IlcQ(E) + I vbllEal
By the triangle inequality, this last estimate and equation (4.1) yield
II
A(.)vllcQ(E) <
a(l~ a) [II f IlcQ(E) + I vbllEal

Second, one can enrich the family of spaces of smooth functions in which the
boundary-value problem (4.1) is well posed by introducing the spaces Cf/(E),
(0 :S "( :S {3,0 < (3 < 1), obtained by completion of the set of all smooth E-valued
functions <p(t) on [O,T] with respect to the norm
I <p Ilcil,-Y
o
E = max
()
O~t~T
11<p(t)IIE +
sup
(t + r)'Y
oSt<t+T~T
II <p(t + r)
r{3
- <p(t)
liE
Let us give, without proof, the following result.
Theorem 4.1.9. Suppose vb E E{3_" f(t) E Cg"(E)(O :S "( :S {3,0 < (3 < 1).
Suppose that assumptions (4.3) and (4.4) hold and 0 < (3 :S :S 1. Then for the
solution v(t) in Cg"(E) of the Cauchy problem (4.1) the coercive inequalities
I v' Ilc(Eil_-Y):S M[li vb

II v'
:S
IIEil--Y +(3-1(1 _(3)-1
Ilcg'-Y(E)
f Ilcg'-Y(E)],
+ I A(.)v Ilcg'-Y(E)
M[lvblg" + (3-1(1 _(3)-1
f Ilcg'-Y(E)]
hold, where M does not depend on (3, "(, vb and f(t). Here, Iwlg" denotes that the
norm of the Banach space Eg" consists of those wEE for which the norm
is finite.
138
Note that the parameter, can be chosen freely in [0, (3), which increases the
number offunction spaces in which problem (4.1) is well posed. In particular, it is
important that problem (4.1) is well posed in the Holder space without a weight
(r = 0).
Problem (4.1) is not well posed in C(E) for arbitrary E. It turns out that
a Banach space E can be restricted to a Banach space E' in such a manner that
the restricted problem (4.1) in E' will be well posed in C(E'). The role of E' will
be played here by the fractional spaces Eo: = Eo:(E, A(t)) (0 < a < 1).
Theorem 4.1.10. Suppose A(O)vo E Eo:, f(t) E C(Eo:) (0 < a < 1). Suppose that
assumptions (4.3) and (4.4) hold and 0 < a :S :S 1. Then for the solution v(t) in
C(Eo:) of problem (4.1) the coercive inequality
II v' Ilc(Ea) + II
A(.)v Ilc(Ea)
:S Ma- l (l - a)-lUI A(O)vo ilEa + II f IIC(Ea)]
(4.48)
holds, where M does not depend on a, Vo and f(t).

Proof. Using formula (4.5), we obtain
A(t)v(t) = A(t)v(t, O)vo +
where
A(t)v(t, s)f(s)ds = { ; Jk(t),
Jl(t) = A(t)v(t,O)vo,
h(t)
it
A(t)v(t, s)f(s)ds.
Let us estimate Jk(t) for any k = 1,2 separately. We start with J l (t). Using
formula (4.33) and estimates (4.3), (4.6) and (4.8), we obtain
Zl-O: II A(t) exp{ -zA(t)}Jl(t)IIE :S Zl-O: II A(t) exp{ -(z + t)A(t)}A(O)voIIE

+zl-o:
II
A(t) exp{ -(z + t)A(t)}IIE--+EII[A(t) - A(O)]A-l(O)IIE--+EIIA(O)voIIE
Jzl-O: II
t
A 2 (t) exp{ -(z + t - s)A(t)}IIE--+EII[A(s) - A(t)]A-l(s)IIE--+E
1-0:
x IIA(s)v(s, O)A-l(O)IIE--+Ellv~IIEds :S (t +z)l

Z
Mt
+z - -IIA(O)voIIE +
z+t
1 0:
J
t
-0:
II A(O)voIIE
1 0: M(t - s)ds
Ml
z - (
)2 IIA(O)voIIE :S -1-IIA(O)voIIEa
z+t-s
-a
139
> 0 and t, t
for all z
M1
IIJ11IC(Ea) :S a(l- a) IIA(O)voIIEa'

Now let us estimate h(t). Using formula (4.33) and estimates (4.3), (4.6) and
(4.8), we obtain
Zl-a I A(t) exp{ -zA(t)}h(t)IIE
it I
itit II
-1
itit I
A 2(t) exp{ -(z + t - s)A(t)} f(s)IIE
:S zl-a
+zl-a
x II[A(t)
A(p)]A
+zl-a
A 2(t) exp{ -(z + t - p)A(t)}IIE--+E
(p)IIE--+EIIA(p) exp{ -(p - s)A(p)} f(s )IIEdpds

A 2(t) exp{ -(z + t - p)A(t)}IIE--+E
x II[A(t) - A(p)]A-l(p)IIE--+EIIA(p)[v(p, s) - exp{ -(p- s)A(p)}IIE--+Ellf(s)IIEdpds
< zl-a
-
rM
t
Jo
+zl-a
for all z
Ilf(s)IIEads
(z+t-s)2-a
tit
Jo
> 0 and t, t
+ zl-a
r it M Ilf(s)IIEaM(t - p)Odpds
t
Jo
(z+t-p)2(p_s)l-a
MM(t - p)ollf(s)IIEdpds < M1 Ilfllc E

(z + t - p)2(p - s)l-o - a(l - a)
( a)
Combining the estimates for It, /2, we obtain estimate (4.48). Theorem 4.1.10 is
proved.
Theorem 4.1.11. Suppose that assumption (4.3) holds. Suppose that the operator
A(t)A-1(S) is Holder continuous in t in the uniform operator topology for each

fixed s, that is,
where M and are positive constants independent oft, sand T for 0 :S t, s, T :S T.

Suppose A(O)vo E E a , f(t) E C(Ea)(O < a < 1). Then for the solution v(t) in
C(Ea ) of problem (4.1) the coercive inequality
I Vi Ilc(E + I
Q
:S
M[li
A(O)vo
IlEa
Ilc(Ea)
a)-l I f IIC(Ea)]
A(.)v
+a- 1(1 -
holds, where M does not depend on a, Vo and f(t).
140
For our third step, let us study now problem (4.1) in the spaces Cg'''(Ea -f3),
(0 ~ I ~ 13 ~ a,O < a < 1). To these there correspond the spaces of traces E~'!f3'
which consist of elements wEE for which the norm
,"
Iwl f3a-f3
max
Ile-zA(t)wIIE _
"
O:S;z:S;T
sup
T- f3 (Z
/3
+ T)'YII(e-(z+r)A(t)
- e- zA (t))wIIE"_/3
O:S;z<z+r:S;T
is finite. Let us give, without proof, the following results.

Theorem4.1.12. Suppose that assumptions (4.3) and (4.4) hold and 0 < a ~ ~ 1.
Let vb E E~'!f3and f(t) E Cg'''(Ea - f3 ), (0 ~ I ~ 13 ~ a, 0 < a < 1). Then for the
solution v(t) in Cg'''(Ea -f3) of problem (4.1) the coercive inequality
II v' Il cg.-r(E,,_/3) + II A(.)v Il cg,-r(E,,_/3) + II v' Ilc(E~:?/3)

~ M[lvbl~'!f3 + a- I (1 - a)-I II f Il cg,-r(E,,_/3)]
holds, where M does not depend on a, 13, " vb and f(t),
Theorem 4.1.13. Suppose that assumptions (4,3) and (4.4) hold and 0 < a ~ ~ 1
Let vb E E a -" and f(t) E Cg'''(Ea -,,), (0 ~ I ~ 13 ~ a, 0 < a < 1). Then for the
solution v(t) in Cg'''(Ea - f3 ) of problem (4,1) the coercive inequality
II v' Il cg,-r(E,,_/3) + II A(.)v II cg,-r(E,,_/3) + II v' Ilc(E"--r)

IIE"--r + II f II cg,-r(E,,_/3)]
holds, where M does not depend on a, 13, I, vb and f(t).
I
~ Ma- (1- a)-I[ll vb
Note that the spaces of smooth functions Cg'''(Ea - f3 ), in which coercive

solvability has been established, depend on the parameters a, 13 and I' However,
the constants in the coercive inequalities depend only on a. Hence, we can choose
the parameters 13 and I freely, which increases the number of function spaces
in which problem (4.1) is well posed. In particular, Theorem 4.1.13 implies the
well-posedness Theorem 4.1.10 in C(Ea ) established in the above.
Fourth, let us study now the Cauchy problem (4,1) in the spaces Lp(E) =
Lp([O,T],E), (1 ~ P < 00) of all strongly measurable E-valued functions v(t) on
[0, T] for which the norm
(J II
T
II V IILp(E)=
is finite.
v(t)
II~ dt):'
141
A function v(t) is said to be absolutely continuous if it has a derivative v'(t)

for almost every t such that v'(t) E Ll(E) and if the Newton-Leibniz formula
J
t
v(t) - V(T) =
v'(s)ds
7'
holds for all t, T E [0, T]. Here the integral is understood in the sense of Bochner.
A function v(t) is said to be a solution of problem (4.1) in Lp(E) if it is absolutely continuous, the functions v'(t) and A(t)v(t) belong to Lp(E), and equation
(4.1) is satisfied for almost every t and v(O) = Va. From this definition it follows
that a necessary condition for the solvability of problem (4.1) in Lp(E) is that
f(t) E Lp(E). It will be shown that in certain cases this condition is also sufficient for the solvability of problem (4.1). As concerns the boundary elements, in
contrast to the situation considered earlier, from the solvability of problem (4.1)
in Lp(E) it follows only that Va E E. In the case of an unbounded operator A(t)
this does not allow us to prove the solvability of problem (4.1).
From the unique solvability of (4.1) it follows that the operator v(t; f(t), va)
is bounded in Lp(E) and one has coercive inequality
Ilv'IILp(E) + IIA(.)v(t)IILp(E) ~ Me[llfIILp(E) + IIA(O)voIIEJ,

where Me (1 ~ Me < +(0) does not depend on Vo and f(t).
Let us give, without proof, the following results.
Theorem 4.1.14. Suppose that assumptions (4.3) and (4.4) hold and 0 <
E ~ 1
Suppose problem (4.1) is well posed in L po (E) for some Po, 1 < Po < 00. Then it
is well posed in Lp(E) for any p,l < p < 00 and the coercivity inequality holds:
Ilv'IILp(E) + IIA(.)vIILp(E) + Ilvlle(E,_,/P.P)

M(pO)p2
p _ 1 IlfIILp(E) + MllvoIIE,_,/P,P'
where M(po) and M does not depend on p, Vo and f(t). Here the Banach space
El-f;,p = El_f;,p(E,A(t)) consist of all these vEE for which norm
IlvIlE,_,/p,P =
l"A(t) exp{ -zA(t)}ll~dz
lip
(l~p<oo)
is finite.
~ p ~ 00 and 0 < a < 1. Suppose that assumptions (4.3)
and (4.4) hold and a ~ E ~ 1. Then problem (4.1) is well posed in Lp(Ea,p) and
the coercivity inequality holds:
Theorem 4.1.15. Let 1
Ilv'IILp(E",p) + II A (.)vIIL p(E",p)
~ a(l~ a) IIfIlL p(E",p) + MllvoIIE,_,/p,p(Ea,p,A(t)),
where M does not depend on a:, p, va and f (t).
142
~ p ~ 00 and 0 < 0: < 1. Suppose that assumption (4.3)

holds. Suppose that the operator A(t)A-1(s) is Holder continuous in t in the uniform operator topology for each fixed s, that is,
where M and E: are positive constants independent oft, sand T for 0 ~ t, s, T ~ T.

Then problem (4.1) is well posed in Lp(Ea,p) and the coercivity inequality holds:
where M does not depend on

0:,
p, Va and f(t).
< p, q <
Suppose that assumptions (4.3) and (4.4)

hold and 0 < 0: ~ E: < 1. Then problem (4.1) is well posed in Lp(Ea,q) and the
coercivity inequality holds:
00.
where M(q) and M does not depend on 0:, p, Va and f(t).

Theorem 4.1.18. Let 1 < p, q <
00. Suppose that assumption (4.3) holds. Suppose

that the operator A(t)A-1(s) is Holder continuous in t in the uniform operator
topology for each fixed s, that is,
where M and E: are positive constants independent of t, sand T for 0 ~ t, s, T ~ T.

Then problem (4.1) is well posed in Lp(Ea,q) and the coercivity inequality holds:
where M(q) and M does not depend on
0:,
p, Va and f(t).
Finally, we consider the applications of these results to the parabolic equations. First, we consider the mixed boundary-value problem for parabolic equation
~~ - a(t,x)~:~ +<5u = f(t,x),O < t ~ T,O < x < 1,
{ u(O,x) = cp(x), f(O,x) = 0, ~ x ~ 1,

u(t,O) = u(t, 1), ux(t,O) = ux(t, 1), ~ t ~ T,
(4.50)
where a(t, x), cp(x) and f(t, x) are given sufficiently smooth functions and a(t, x) >
0, 8 > 0 is a sufficiently large number.
143
We introduce the Banach spaces C13[O, 1] (0 < (3 < 1) of all continuous

functions <p(x) satisfying a Holder condition for which the following norms are
finite:
l<p(x + T) - <p(x)1
II <P Ilc 13 [o,I]=11 <P Ilc[o,11 +
sup
13
O~x<x+T~1
where C[O, 1] is the space of all continuous functions <p(x) defined on [0,1] with
the usual norm
II <p Ilc[o,I]= orr~II<p(x)l.
It is known that the differential expression
At,xv = -a(t, x)v"(t, x) + <5v(t, x)

defines a positive operator At,x acting in C13[O, 1] with domain CI3+2[0, 1] and
satisfying the conditions v(t,O) = v(t, 1), vx(t, 0) = vx(t, 1).
Therefore, we can replace the mixed problem (4.50) by the abstract Cauchy
problem (4.1). Using the results of Theorems 4.1.6, 4.1.8, 4.1.9, 4.1.10, 4.1.11,
4.1.12 and 4.1.13, we can obtain that
Theorem 4.1.19. For the solution of the mixed problem (4.50) the following coercive
inequalities are valid:
M(J.1.)
::; ({3 - ,)(1 - (3) II
::; , <
Ilcg''Y(cl'[O,Ij) +M(J.1.)II<pllc2+I'+2(13-'Y)[O,lj,
(3 < 1, 0::; J.1.
II u Il c ci+ 13 .'Y (C2(o-l3) [0,11)
+ 2({3 - ,) ::; 1,
+ II u
Ilcg'Y (C2+2(o-l3) [0,11)
M(ex,{3)
::; ex(l- ex) II f Il c g''Y(c 2(O-I3)[o,ll) +M(ex,{3)II<pllc2+ 2(o-'Y)[O,lj,
o::; , ::; {3 ::; ex,O < ex -
(3 <
2'
Here M(J.1.) and M(ex,{3) are independent of " !(t,x), <p(x).

Second, let n be the unit open cube in the n-dimensional Euclidean space
lRn(O < Xk < 1,1 ::; k ::; n) with boundary S, 0 = nus. In [0, T] x 0 we consider
the mixed boundary-value problem for the multidimensional parabolic equation
8u(t,x) ~
8 2u(t,x)
8t
- L.,..exr(t,x) 8x2
r=l
(Xl, ...
+ <5u(t, x)
= !(t,x),
,xn ) E 0,0 < t::; T,
u(O,x) = <p(x), !(O,x) =0, XEO,

u(t,x) = 0, XES,
(4.51)
144
where QAt, x), f(t, x) (t E [0, T], x En), cp(x)(x En) are given smooth functions
and ar(t,x) > 0, 8> 0 is a sufficiently large number.
We introduce the Banach spaces 1(n) ((3 = ((31, ... ,(3n), 0 < Xk < 1, k =
1, ... , n) of all continuous functions satisfying a Holder condition with the indicator
(3 = ((31, ... , (3n), (3k E (0,1),1
k
n and with weight X~k (1 - Xk - hd3k ,
o Xk < Xk + hk 1,1 k n which is equipped with the norm
cg
:s
:s
:s :s
:s :s
sup
If(xl,'" ,xn ) - f(Xl
O::;Xk <Xk+hk::; 1,1::;k::;n
+ hi, ... ,xn + hn)1
II h;f3kx~k(1 -
k=l
Xk - h k )f3k,
where C(n)-is the space of all continuous functions defined on n, equipped with
the norm
II f Ilc(o)= ma~ If(x)l

xE!1

At ,x v = -
~
(
)8 v(t,x)
(t, x )
~ a r t, x
ax 2 + uV
r=l
cg
c5i
f3
defines a positive operator At,x acting on 1 (0) with domain D(At,X) C
(0)
and satisfying the condition v = 0 on S.
Therefore, we can replace the mixed problem (4.51) by the abstract Cauchy
problem (4.1). Using the results of Theorems 4.1.6, 4.1.8, 4.1.9, 4.1.10, 4.1.11,
4.1.12 and 4.1.13, we can obtain that
Theorem 4.1.20. For the solution of the mixed boundary-value problem (4.51) the
following coercive inequality is valid:
M(f-L)
< ((3 - 1')(1 - (3)
O:S l'
II
8 2 cp
f Ilcgr"Y(cb'(O)) +M(f-L) ~ II ax~ IIcb,+2(i3-"Y) CO) ,
< (3 < 1,f-L = {f-Ll,'" ,f-Ln},
< f-Lk + 2((3 - 1') < 1,1 :s
where M(f-L) is independent of (3,"1 and f(t,x), cp(x).
k:S n,
145
Third, we consider the Cauchy problem on the range {O :S t :S T, x E IR n }

for the 2mth-order multidimensional parabolic equation
au
a1r1u
-a + '" ar(t,x)a r1
a
+ 8u (t,x)=!(t,x),
t
~
x X rn
n
Irl=2m
0< t :S T, x, r E IR , Irl = rl + ... + rn,

u(O,X) = <p(x), f(O,x) = 0, x E IR n ,
n
(4.52)
where ar(t,x) and f(t,x), <p(x) are given sufficiently smooth functions and 8 > 0
is the sufficiently large number.
Let us consider a differential operator with constant coefficients of the form
acting on functions defined on the entire space Rn. Here r E R n is a vector with
nonnegative integer components, Irl = rl +... +rn' If <p (y) (y = (Yl, ... , Yn) E Rn)
is an infinitely differentiable function that decays at infinity together with all
its derivatives, then by means of the Fourier transformation one establishes the
equality
F (B<p) (~) = B (~) F (<p) (~).
Here the Fourier transform operator is defined by the rule
F(<p)
Jexp{-i(y,~)}<p(y)dy,
(~) = (27f)-n/2
Rn
(y,~) = Y16
The function B
(~)
+ ... + Yn~n.
is called the symbol of the operator B and is given by
B (~) =
br (i6f 1
(i~nfn .
Irl=2m
We will assume that the symbol
Bt,X(~)=
ar(t,x)(i~drl(i~nfn,~=(~l, ... ,~n)ERn
Irl=2m
of the differential operator of the form

(4.53)
146
acting on functions defined on the space IR n , satisfies the inequalities
for ~ =I O. Problem (4.52) has a unique smooth solution. This allows us to reduce
problem (4.52) to the abstract Cauchy problem (4.1) in a Banach space E =
GJ-J, (Rn) of all continuous bounded functions defined on IRnsatisfying a Holder
condition with the indicator fJ, E (0,1) with a strongly positive operator At,x =
Bt,x + 8I defined by (4.52).
Theorem 4.1.21. For the solution of the Cauchy problem (4.52) the following co-
ercivity inequalities are satisfied:
M(fJ,)
:::; ((3 - 'Y )(1 - (3)
f Il c OT
13'Y(C IL (Rn)) +M(fJ,)
0:::; 'Y < (3 < 1, 0 < fJ,
II U
Il c OT
2+13'Y(C 2m (o-13)(Rn))
Irl=2m
:::;M(a,(3)lljllc13~p(C2m(o-13)(Rn))+M(a,(3)
o
0:::; 'Y :::; (3, 0
II
Irj=2m
8 1rl <p
8Xlrl
+ 2m((3 -
'Y)
...
8 X r n Ilc IL + 2m (13-'Y)(Rn),
n
< 1,
8 1r1 u
8Xlrl
...
Ir l=2m
8X r n Il c OT
13''Y(c 2m (o-13)(Rn))
n
8 1rl <p
11
rl
8 Xl'"
< fJ, + 2m(a - (3) <
8 r n llc IL + 2m (o-'Y)(Rn),
Xn
1,
where M(fJ,) and M(a,(3) does not depend on'Y and j(t,x),<p(x).
The proof of Theorem 4.1.21 is based on the abstract Theorems 4.1.6,4.1.8,
4.1.9, 4.1.10, 4.1.11, 4.1.12 and 4.1.13, the coercivity inequality for an elliptic
operator A t,x in GJ-J, (Rn) and on the following theorem on the structure of the
fractional spaces Eo:(C(Rn), At,X).
Theorem 4.1.22. Eo:(C(Rn), At,X) = C 2mO:(Rn) for all 0 < a
<
2~
and 0:::; t :::; T.

In this section we consider the difference schemes of a high order accuracy for the
approximate solutions of problem (4.1). On the segment [0, T] we consider again
a uniform grid space
[0, T]r
= {tk = kT, k = 0, 1, ... , N, NT = T}
147
4.2. Difference Schemes Generated by an Exact Difference Scheme
with step r > 0. Here N is a fixed positive integer. From formula (4.5) it follows
that
1jt
V(tk, s)f(s)ds, 1 :S k :S N, v(o) = Vo
= -:;.
tk-l
or
(4.54)
= fk, fk
1jt
V(tk, s)f(s)ds, 1 :S k :S N, Uo = Vo
= -:;.
tk-l
The last system is called the single-step exact difference scheme for the solution of the initial-value problem (4.1).
Let us investigate the well-posedness of the exact single-step difference
scheme (4.54). This problem is uniquely solvable and the following formula holds:
Uk
V(tk, O)uo +
L V(tk, tj) !Jr.
(4.55)
j=l
Let us reduce the difference scheme (4.54) to an operator problem in the space
FT(E) of vectors v T = {vd~=l' Recall that the space FT(E) can be equipped with
various norms and thus become a normed space. Thus, for instance, the vector
space FT(E) generates the normed space CT(E) = C([O, T]T, E) with the norm
the normed space C~(E) = CI3([O,T]nE),
II <pTll e J3(E) = I <pTll e

T
(E)
< 13 <
+ l$k<k+r$N
max
I
1, with the norm
<Pk+r - <Pk
the weighted normed space C~"(E) = CI3,,([O, T]T, E),

with the norm
T
I <p Il eJ3'"Y(E)
T
II <p lie
(E)
+ l$k<k+r$N
max
I
I <pTIILp.T(E)
(L I <Pkll~r)~.
k=l
!J
< , :S 13, < 13 <
<Pk+r - <Pk
and the normed space Lp,T(E) = Lp([O, T]T, E), 1 :S p <
liE (rr1)""
00,
liE
((k
+ r)r)'
()13
'
rr
with the norm
1,
148
First we will define the operator Dr, acting from the space E x Fr(E) of vectors
wr = {wd~=o into the space Fr(E) of vectors v r = {Vd~=1 by the rule
v r = Drw r ,Vk =
~(Wk
7
wk-d, k = 1, ... , N.
Second we will define an operator A r from the space E x Fr(E) of vectors wr =

{Wk}~=O into the space Fr(E) of vectors v r = {Vk}~=1 by the rule
v r = Arw r ,Vk =
7-
(1 - V(tk, tk-1) )Wk-1, k = 1, ... , N.
Next, let us introduce the continuation operator II(uo), which acts from Ex Fr(E)
to Fr(E) according to the rule
II(uo)(u1, ... ,uN)
(uo,UI, ... ,UN).
Then the difference schemes (4.54) can obviously be rewritten as the equivalent
operator equation
Here
DrII(uo)u r + ArII(uo)u r
is defined by the formula
The last operator problem will be considered in the space Fr(E). From its unique
E Fr(E) it follows that its solution u r defines
solvability for any Uo E E and
an additive and homogeneous operator ur(r, uo).
The initial-value difference problem (4.54) is said to be stable in Fr(E) if we
have the stability inequality
r,
where M is independent not only of

Uo but also of 7.
From formula (4.55) one derives the following results.
Theorem 4.2.1. Suppose that assumptions (4.3) and (4.4) hold and 0
Then difference problem (4.54) is stable in C r (E).
Then difference problem (4.54) is stable in C::'cE).
<
E ::;
l.
< ex ::; E ::; 1.
Then difference problem (4.54) is stable in Lp,r(E), 1 ::; p < 00.
<
E ::;
1.
The initial-value difference problem (4.54) is said to be well posed (coercively

stable) in Fr(E) if we have the coercive inequality
II {7- 1(Uk - uk-d}f' Ik(E) + II ArII(uo)urIIFT(E)
::; M[li r IIFT(E) + II A(O)uoIIEJ,

r, Uo but also of
7.
149
Since the initial-value differential problem (4.1) in G(E) is not well posed
for the unbounded operator A(t) and space E, then the well-posedness of the
difference initial-value in Gr(E) norm does not take place uniformly with respect
to T > O. This means that the coercive norm
tends to 00 as T -+ +0. The investigation of the difference problem (4.54) permits

us to establish the order of growth of this norm to 00.
Theorem 4.2.4. Suppose that assumptions (4.3) and (4.4) hold and 0 < E ::S 1.
The solutions of the difference problem (4.54) in GT(E) obey the almost coercive
inequality
II uT IIKT(E) ::S Mdll A(O)uoII E
(4.56)
+ min{ln~, 1 + lIn IIA(O)IIE.-d//I

where M 1 is independent not only of
r,
Uo,
IlcT(E)],
but also of T.
Proof. Using formula (4.55), we can write
k-1
+L
T- 1(1 - V(tk, h-1)V(tk-1, tj ) liT = Jk
+ Gk,
j=l
where
k-1
Gk = L(1 - V(tk, tk-d)v(tk-1, tj)
!J.
j=1
Let us estimate Jk and Gk separately. We start with Jk. Using property 4) of

v(t,s) and estimates (4.17), (4.18) and (4.21), we obtain
IIT- 1 (1 - V(tk, tk-1))V(tk-1,O)A- 1(O)IIE.-E ::S M,l ::S k::S N,
(4.58)
11(1 -V(tk,tk-I))V(tk-1,0)IIE.-E::S Mmin{l,TIIA(O)IIE.-E},l::S k::S N.
(4.59)
From estimate (4.58) it follows that
IIJk II E::S II T- 1(1 - V(tk, tk-1))V(tk-1, 0)A- 1(0)IIE.-EIIA(0)uo II E

::S MIIA(O)uoIIE
for all k, k = 1, ... ,N. Using the last inequality, we obtain
150
Now let us estimate Gk separately. From estimate (4.59) it follows that

k-l
IIGk II E ::;
L II (1 - V(tk, tk-d)v(tk-l, tj)IIE->Ellf j liE

j=l
::; M
L yll r
IICT(E) ::; M1ln ~II
j=l
r IICT(E)
for all k, k = 1, ... ,N. Using the last inequality, we obtain

II G IlcT(E) ::; M1ln ~II
T
r IlcT(E)'
(4.60)
Further, using the triangle inequality, we obtain

k-l
IIGk II E ::;
L II (I - V(tk, tk-l))V(tk-l, tj)IIE->Ellf j liE

j=l
k-l
::; L II (I - V(tk,tk-l))V(tk-l, tj)IIE->E11 r I CT (E)

j=l
for all k, k = 1, ... , N. Let us estimate

k-l
ak
= LII
j=l
(l-V(tk,tk-l))V(tk-l,tj)IIE->E.
If IIA(O)IIE->E > N, then
N-l
IIA(O)IIE~E
ak ::; M '"'
L..J -:-1 ::; M
-ds < M
. J
S
)=1
1
-ds ::; Miln IIA(O)IIE->EI

S
If IIA(O) IIE->E < 1, then

N-l
ak ::; M
L rIlA(O)IIE->E ::; MI.

j=l
Finally, if 1 ::; IIA(O) IIE->E ::; N, then

[IIA(O)IiE~E]
ak ::; M[
N-l
rIlA(O)IIE->E
j=[
T
::; M(l
j=l
ds
- ) ::; M(l
S
~]
IIA(O)f1E~E1+1 J
+ In IIA(O)IIE->E)'
151
Thus, in all three cases we have the estimate

ak
Therefore,
:S M(l
+ lIn IIA(O)IIE--+EI).
IIGTIICr(E) :S M 1 (1 + lIn IIA(O)IIE--+E)III r
II Cr (E)'
(4.61)
Finally, the estimate
follows from (4.60) and (4.61). Combining the estimates for JT and GT, we obtain
II{T- 1 (I - V(tk, tk-1))Uk-d~IICr(E) :S M1[1IA(0)uoIIE
+min{In~, 1 + lIn IIA(O)IIE--+EI}II r
II Cr (E)]'
The estimate for II{T- 1(uk - Uk-1)}f"llc r(E) follows from the triangle inequality
and the last estimate. Theorem 4.2.4 is proved.
Now let us study the well-posedness of the difference problem (4.54) in the
various Banach spaces.
Theorem 4.2.5. Suppose that assumptions (4.3) and (4.4) hold and O<a:Sc:S1.
The solutions of the difference problem (4.54) in C::,Q(E) obey the coercivity inequality
I {T- 1(I -
II{T- 1(Uk - Uk-1)}~llc~.a(E)+
V(tk, tk-1))Uk-d~llc~a(E)
:s M 1[a(1 ~ a) Ilrllc~a(E) + IIA(O)uoIlEJ,

r, uo, a, but also of
T.
The proof of Theorem 4.2.5 follows the scheme of the proof of Theorem 4.1.6
and relies on estimates (4.3), (4.4), (4.17), (4.18), (4.19), (4.20), (4.21), (4.22),
(4.23), (4.24) and on the formula
T- 1(I - V(tk, tk-d)Uk-1
T- 1(I - V(tk, tk-d)v(tk-l> O)uo
+(I - exp{ -tk-1 A (tk-d})!k

k-1
+ L[(I - V(tk, tk-dv(tk-1, tj ) - (I - exp{ -TA(tk-d}) exp{ -tk-1 A (tk-1)}]!k
j=l
k-1
+ L(I - V(tk, tk-1)V(tk-l> tj)(fj - fk)'
j=l
152
Theorem 4.2.6. Suppose that assumptions (4.3), (4.4) hold and 0 < a::; ::; 1. The
solutions of the difference problem (4.54) in CT(Ecr ) obey the coercivity inequality
II{T- 1(Uk - Uk-l)}f lI e,.(E
::; M 1 [a(1
)+
II {T- 1(1 - V(tk, tk-I))Uk-df IJe,.(E
~ a) II r
lI e,.(E + II A(O)uoIlEJ,
Q
r, Uo, a, but also of T.
and relies on estimates (4.3), (4.4), (4.17), (4.18), (4.19), (4.20), (4.21), (4.22),
(4.23), (4.24) and on the formula
T- 1 (1 - V(tk, tk-I))Uk-l = T- 1 (I - V(tk, tk-l))V(tk-l, O)uo

k-l
+ L:(I -
exp{ -TA(tk-l)}) exp{ -tk-l- j A(tk-l)}l!s
j=1
k-l
+ L:[(I - V(tk, tk-l)V(tk-l, tj) - (I -
exp{ -TA(tk-I)}) exp{ -tk-l- jA(tk-I) }]!J.
j=1

Theorem 4.2.7. Suppose that assumptions of Theorem 4.1.11 hold. Then for all
o < a < 1 the solutions of the difference problem (4.54) in CT(Ecr ) obey the
coercivity inequality
II{T- 1 (Uk - Uk-l)}f lI e,.(E
::; M 1 [a(1
)+
II {T- 1 (I - V(tk, tk-l))Uk-df IJe,.(E
~ a) II r
lI e,.(E + II A(O)uoIl EQ ]'

Q
r, Uo, a, but also of T.
Theorem 4.2.8. Suppose that assumptions (4.3) and (4.4) hold and 0 < /3 ::; e ::; 1.
The solutions of the difference problem (4.54) in C~"(E) (0::; 'Y ::; /3,0< /3 < 1)
satisfy the following coercivity inequalities:
II {T- (Uk - Uk-l)}f Ile,.(EiJ-"Y) ::;

+M1 11 (I -
e-
TA
/3(~ (3) II r lIe~'"Y(E)
(O))T- 1uo -
II IIEiJ-"Y'
II {T- 1 (Uk - uk-lnf" lIe~'"Y(E) + II {T- 1 (I - V(tk, tk-l))Uk-df" Ile~'"Y(E)

1
:::; /3(1M_ /3)
II
fT
II e~""(E) + M 1 1(1 -
e -TA(O)) T -1 Ua -
where M 1 is independent not only of r,Ua, /3, 'Y but also ofT.
f 1 I{J"
a '
153
Note that the parameter 'Y can be chosen freely in [0, (3), which increases the
number of spaces Ce'''!(E), (0 S 'Y S (3,0 < (3 < 1) of grid functions in which
difference problem (4.54) is well posed.
Theorem 4.2.9. Suppose that assumptions (4.3) and (4.4) hold. The solutions of
difference problem (4.54) in Ce'''!(Eex -f3), (0 S 'Y S (3 S a S c S 1) satisfy the
following coercivity inequalities:
II {r- 1 (uk
- Uk-l)}i"
+I
S a(1M_ a) I
1
Il ce'''(Eo_I3)+ I {r- 1 (uk -
Uk-l)}i" IICr(E~~I3)
{r- (I - V(tk,tk-l))Uk-di" Il c e'''(Eo_l3)

1
fT
I Ce'''(E
-l3)
+ M 1 1(1 -
e -TA(O)) r -1 Uo -
II {r- 1(uk -
1
Uk-l)}i" Il c e'''(Eo_I3)+ II {r- (uk - Uk-l)}i"
+ I {r- 1(1 - V(tk' tk-l))Uk-di" Il c e'''(Eo_l3)
a(~ a) I r II c e'''(Eo_l3) + M 1 11(I -
e-
TA
(O))r-
1u
o -
1f3ex-f3'
,"!
IICr(Eo_,,)
hII Eo _,,'
r, Uo, a, (3, 'Y but also of r.
Note that the spaces Ce'''!(Eex -f3) of grid functions, in which coercive solvability has been established, depend on the parameters a, (3 and 'Y. However, the
constants in the coercive inequalities depend only on a. Hence, we can choose the
parameters (3 and 'Y freely, which increases the number of spaces of grid functions
in which difference problem (4.54) is well posed.
Now let us consider the initial-value difference problem (4.54) in the spaces
Lp,T(E) = Lp([O, T]T> E), 1 S P < 00 of all grid functions. We have not been able
to obtain the coercivity inequality
I {r- 1(u k-Uk-d}i"

S
IILp,r(E)
M[II
+ I {r- 1(I - V(tk,tk-l))Uk-di"

IIL p,r(E)
IIL p,r(E)
+ I A(O)uoIl E ]'
in the arbitrary Banach space E and for the general strong positive operator A(O).
Nevertheless, we have established the almost coercivity inequality.
Theorem 4.2.10. Suppose that assumptions (4.3) and (4.4) hold. The solutions of
the difference problem (4.54) in Lp,T(E) obey the almost coercive inequality
II {r- 1(u k-Uk-d}i"

S
M 1 [1I
A(O)uoIl E
IIL p,r(E)
+ I {r- 1(I - V(tk,tk-l))Uk-di"
+ min {ln~, 1 + lIn I
where M 1 does not depend on
r, Uo, P and r.
A(O)
I E-->E I} I r
II Lp ,r(E)
IIL p,r(E)],
154
Note that if A(O) is the unbounded positive operator in E, then
min{ln~,l+llnll A(O) II E EI} =ln~.

--+
Finally, let us give, without proof, the following results about well-posedness of
the initial-value difference problem (4.54) in the spaces Lp,T(E) = Lp([O, TIT> E),
1:::; p < 00.
Theorem 4.2.11. Suppose that assumptions (4.3) and (4.4) hold and 0 < c :::; 1.
Suppose that difference problem (4.54) is well posed in LpO,T(E) for some Po, 1 <
Po < 00. Then it is well posed in Lp,T(E) for all p, 1 < P < 00 and the following
II {T-1(Uk-Uk-l)}i"
IIL p,T(E)
+ II {T-1(I - V(tk, tk-l))Uk-di"
+lluTllcc(El_l/P,p) :::; M(po)[11 Uo IIE

where M (Po) does not depend on
p, + pPp
r, uo, p and
II r II
IILp,T(E)
Lp,T(E)
],
T.
Theorem 4.2.12. Let 1 :::; P :::; 00. Suppose that assumptions (4.3) and (4.4) hold
and 0 < a :::; c < 1. Then problem (4.54) is well posed in Lp,T(Ea,p) and the
following coercivity inequality holds:
II {T-1(Uk-Uk-d}i"
:::; M[li
IIL p,T(Ea ,p)
+ II {T-1(I - V(tk, tk-d)Uk-l}i" IILp,T(Ea,p)
A(O)uollEa,p
+ a(1 ~ a) II
where M does not depend on fT, UO, p, a and
II
Lp,T(Ea,p)
],
T.
Theorem 4.2.13. Let 1 :::; p :::; 00 and 0 < a < 1. Suppose that the assumptions
of Theorem 4.1.18 hold. Then problem (4.54) is well posed in Lp,T(Ea,p) and the
II {T-1(Uk-Uk-l)}i" IILp,T(Ea,p) + II {T-1(I - V(tk, tk-l))Uk-di" IILp,T(Ea,p)
:::; M[II A(O)uoIIEa,p + a(l ~ a) II r II

r, Uo, p, a and
Lp,T(Ea,p)
],
T.
Theorem 4.2.14. Let 1 < p, q < 00. Suppose that the assumptions (4.3) and (4.4)
hold and 0 < a :::; c < 1. Then difference problem (4.54) is well posed in Lp,T(Ea,q)
and the following coercivity inequality holds:
II {T-1(Uk-Uk-d}i" IILp.T(Ea,q) + II {T-1(I - V(tk, tk-l))Uk-di" IILp,T(Ea,q)

2
:::; M(q)[11
A(O)uoli Ea q
,
where M(q) does not depend on
+ ~ II
and
T.
P -
r, uo, p, a
II
Lp,T(Ea,q)
],
155
< p, q < 00 and < a < 1. Suppose that the assumptions

of Theorem 4.1.18 hold. Then difference problem (4.54) is well posed in Lp,T(Eo:,q)
II
{T-1(Uk-Uk-l)}f IIL p.T(Ea,q)
+ II
{T-1(I - v(tk, tk-d)Uk-df IIL p,T(Ea,q)

2
::; M(q)[11 A(O)uoli Ea q + ~ I

,
pwhere M(q) does not depend on
uo, p, aand T.
r,
II
Lp,T(Ea,q)
J,
Now, we will consider the applications of this exact difference scheme (4.54).
From (4.54) it is clear that for the approximate solutions of problem (4.1) it is
fttkk_l
V(tk, s)f(s)ds is not unique. In this section we

The choice of V(tk, tk-l) and ~
consider two approaches. The first approach is carried out in three steps. Applying
the definition of v(t, s), we can write
00
v(t, s)
Vi(t, s) =
it
=L
Vi(t, s), vo(t, s)
= exp{ -(t -
s)A(s)},
i=O
Vi-l(t, z)[A(s) - A(z)j exp{ -(z - s)A(s)}dz, i 2 1.
Therefore, in the first step we consider
uT(k, k - 1)
= u[(k, k -
1)
=L
Vi(tk, tk-l)
i=O
iii
= exp{-TA(tk_d}+L
i=l
itk jt k... jt k exp{-(tk-Zi)A(Zi)}

tk-l
Zl
Zi-l
X(A(Zi-d - A(Zi)) exp{ -(Zi - Zi-l)A(Zi-l)}dz i

X (A(Zi-2)
X ...
- A(Zi-d) exp{ -(Zi-l
Zi-2)A(Zi-2)}dzi - 1
(A(zd - A(Z2)) exp{ -(Z2 - zdA(Z2)(A(zd - A(Z2)) exp{ -(Z2 - zl)A(Z2)}

x (A(tk-d - A(zd) exp{ -(Zl
tk-dA(tk-d}dz 1.
Suppose that the operator N(t)A(z)A-(r+l)(s) satisfies the condition
IIA r (t)[A(Z2) - A(zdjA-(r+l)(s)!IE-tE ::; M IZ2 - zll, r = 0,1, ... ,

where M is a positive constant independent of t,
Z2 ::; T.
S, Zl
(4.62)
and Z2 for 0 ::; t, s,
Zl,
156
Note that if r = 0, then it follows that
If 0 ::; r ::; iii, then it follows that
m
2m 2
II[v(tk' tk-I) - uT(k, k - l)]A- (O)IIE-+E ::; Mr + .
From the formula for u[(k, n) it is clear that for the approximate calculation of
u[(k,n) it is necessary to approximate the expressions A(t),exp{-sA(z)} and
their product. For an analytic semigroup we have the formula
L 'lMt, s; A, B),
00
exp{ -(t - s)A} - exp{ -(t - s)B} =
i=l
where
'l/Ji(t, s; A, B) =
it
'l/Ji-l(t, z;
A, B)(A - B)'l/Jo(t, s, A, B)dz, i = 1,2"" ,
'l/Jo(t,s,A,B) = exp{-(t - s)B},O::; s::; t::; T.

Therefore, in the second step applying the last formula and Taylor's formula and
neglecting the small terms, we can write
m-T-al
x(A(Zr-l) - A(zr))
'l/Ja"a2[Zn t k-l;A(zr-l),A(tk-d]dzr
a2=O
m-r- a l-a2
x (A(Zr-2) - A(Zr-l))
aa=O
x ... x (A(zI) - A(Z2))
where
'l/Ja2,aa [Zr-l,
tk-l; A(Zr-2), A(tk-l)]dzr- 1
X ... X
(A(tk-l ) -
2m-2T-2an (
'"'
A ( X n ))
7 -
)i
San., + Xn Ai( tk-l)
2.
i=O
X
157
exp{ -7 A(tk-l) }ds 1 ... ds an ,
an = 1, ... , m; n = 1, ... , r; r = 1, ... , m.

Finally, in the third step replacing the operator exp{ -7 A( tk- t}} by its Pade
approximant Rj,l (7 A(tk-l)), we obtain
uT(k, k - 1) = uIII(k, k - 1) = B~:d [tk' tk-l; A(tk), A(tk-l)]
l-r- a 1
(A(zr-t} - A(zr)) '"'
B~+ja-2r-2a2[ZT,tk_l;A(zr_d,A(tk_t}]dzr
~
1,2
a2=O
l-r-a1- a2
X (A(Zr-2) - A(Zr-l))
B~~~a~2r-2a3 [Zr-l' tk-l; A(Zr-2), A(tk-t}]dzr- 1
a3=O
X
where
1+'
Bo,d [tk' tk-l; A(tk-l), A(tk-t}]
Rj,I(7A(tk-t}),
B~~j-2r [Zl' tk-l; A(tk-t}, A(tk-t}]
l+j
'"' (7 -
= ~
i=O
Zl
+ tk_t}l
i!
A (tk-t}Rj_r_[~],I_r+i+l(7A(h-t}),
158
XRj-r-[~J,I-r+an+i+l-[~J(rA(tk-d)(rA(tk-d- A(xn))
l+j-2r-a n
X
(r - Sa~, + x n ) Ai(tk_d
t.
i=O
XRj-r-[~l,l-r-an+i+l-[~l(rA(tk-d
( A(tk-d - A(xn ))
l+j-2r-a n
X
(r - Sa~, + x n ) Ai(tk_d
t.
i=O
XRj-r-[~J,I-r+an+i+l-[~1(r A(tk-d)(rA(tk-d)ds1... dsan ;
an = 1, ... , l; n = 1, ... , r; r = 1, ... , l.

Here [a] denotes the integer part of a> O. In exactly the same manner using the
1
three steps, we can obtain the formula for fk' such that
Ilfk - fk,IIIE ::; Mr l +j ,

where M does not depend on r.
Thus, replacing the operator V(tk, tk-d by uT(k, k - 1)) and the elements !k
by elements fk,l, we obtain the difference schemes of approximation order I + j,
r-1(uk - uk-d
+ r-1(I - ur(k, k - l))Uk-l

1 ::; k ::; N,
Uo
fe,
(4.63)
= Va
on the solutions v(t) of the Cauchy problem (4.1).

Note that the difference schemes (4.63) for j = I - I and j = I - 2 include
difference schemes of arbitrary high order of approximation. Moreover, the main
corresponding functions Rj,l(z) tend to 0 as z --+ 00 for j = l - 2, I - 1. Such
difference schemes are simplest, in the sense that the degrees of the denominators
of the corresponding Pade approximants of the function exp{ - z} are minimal for
a fixed order of approximation of the difference schemes.
Now let us give, without proof, the following estimates concerning Rj,1 (r A( t))
for j = l - 1 and j = l - 2 which will be useful in the sequel.
Lemma 4.2.16. Suppose that the assumption (4.3) holds. Then for any t, t, E [0, T]
the following estimates are satisfied:
IIRj,l(rA(t))IIE--.E ::; M,l ::; k::; N,
(4.64)
IIA(t)Rj,l(rA(t))IIE--.E ::; M(kr)-I, 1::; k::; N,

II exp(-kTA(t)) - Rj,l(rA(t))IIE--.E ::; Mk- j - 1, 1::; k::; N,
(4.65)
IIRj,l(rA(t)) - R7t(rA(t))IIE--.E ::; M (k :Qr)Q' 1::; k < k + r::; N.
(4.67)
Here M does not depend on k, rand r.
(4.66)
159
Now we shall investigate the well-posedness of the single-step difference

schemes (4.63). This problem is uniquely solvable and the formula
+ L u T(k,r)f;,17
k
Uk = uT(k,O)uo
(4.68)
r=1
holds, where
uT(k, n) =
uT(k,k-1)uT(n+1,n),k>n,
{
I,k = n.
(4.69)
FT(E) of vectors v T = {Vk}~=1' Recall that the space FT(E) can be equipped with
the various norms and thus become a normed space.

First we will define the operator D n acting from the space E x FT(E) of
vectors wT = {wd~=o into the space FT(E) of vectors v T = {Vk}~=1 by the rule
Second we will define an operator AT from the space E x FT(E) of vectors wT =

{wd~=o into the space FT(E) of vectors v T = {Vd~=1 by the rule
v T = ATwT,Vk
= 7- 1 (1 -
uT(k, k - 1) )Wk-1, k
= 1, ... , N.
Next, let us introduce the continuation operator II(uo), which acts from Ex FT(E)
to FT (E) according to the rule
II(uo)(u1, ... ,UN) = (UO,U1, ... ,UN).

operator equation
DTII(uo)u T + ATII(uo)uT = r
Here
r is defined by the formula
f},l)
f T -- (f},l
1 , ... , N .
The last operator problem will be considered in the space FT(E). From its unique
solvability for any Uo E E and r E FT(E) it follows that its solution uT defines
an additive and homogeneous operator uT(r, uo) and is continuous.
The initial-value difference problem (4.63) is said to be stable in FT (E) if we
r, Uo but also of
T.
160
Let us prove a number of estimates for uT(k, n) that will be needed in the
sequel. First we consider ii7(k, n) defined by
ii,T(k,k-1)ii,T(n+1,n),k>n,
ii,T(k, n) =
(4.70)
I,k = n,
where ii,T(k, k - 1) = Rj,l(rA(tk-d).

Lemma 4.2.17. For any 0 ::; n ::; r ::; k ::; N the following identities hold:
ii,T(k,n) = ii,T(k,r)ii,T(r,n),O::; n::; k::; k::; N,
(4.71)
ii,T(k, n) = R7,ln(rA(t n-d)
(4.72)
r=n+1
ii,T(k,r)[Rj,l(rA(tr_t}) - Rj,l(rA(tn_ 1))]Rj,ln-1(rA(t n_1))'
(4.73)
k
r=n+1
R77(rA(tr _t})[Rj,l(rA(tk-1)) - Rj,l(rA(tr_ 1))]ii,T(r -1,n).
Proof. Identity (4.71) follows from the definition of ii,T (k, n). Using formula (4.70),
we obtain
r- 1(ii,T(k, n) - ii,T(k - 1, n)) + r- 1(1 - Rj,l(rA(tq_d)ii,T(k - 1,n)

r- 1(R j ,l(rA(tq_d) - Rj,l(rA(tk_d))ii,T(k - 1, n)
for any 1 ::; q ::; N. Therefore
ii,T(k,n) = R7,ln(rA(tq -d)
R77(rA(t q -1))
r=n+1
x [Rj,l(rA(tq _1)) - Rj,l(rA(tr_d)]ii,r(r -1,n).

Choosing q = k, we obtain the identity (4.72). In a similar manner one establishes
the identity (4.72). Lemma 4.2.17 is proved.
Lemma 4.2.18. Suppose that assumption (4.62) holds for r
o::; n < k ::; N
= O. Then for any
the following estimates hold:

(4.74)
IIA(h)uT(k, n)A-1(tn)IIE->E :S M,
IIA(tk)uT(k, n)IIE->E :S (k
161
(4.75)
~n)r'
(4.76)
where M 2 0 does not depend on r, k, n.

Proof. Using identity (4.72) and estimates (4.64), (4.65) and
IHRj,l(rA(tt-l)) - Rj,l(rA(tn-l)] A-1(tn-l)IIE->E :S M(r - n)r 2 ,
(4.77)
we obtain
r=n+l
IluT(k, r)IIE->EII[Rj,l(rA(tr-d) - Rj,l(rA(tn-l))]A-1(tn-dIIE->E

k
xIIA(tn_l)Rj,jn-l(rA(tn_l))IIE->E:S M
+M
r=n+l
IluT(k,r)IIE->E r .
Therefore,
IluT(k, n)IIE->E :S M
+M
IluT(k, r)IIE->E r.
r=n+l
Now using the difference analogy of the integral inequality, we obtain (4.74). In a
similar manner one establishes the inequality (4.75). Interpolating estimates (4.64)
IIA~ (t)Rj,l(rA(t))IIE->E :S M(kT)-~, 1 :S k :S N,
(4.78)
Using identity (4.72) and estimates (4.64), (4.65), (4.77) and (4.78), we obtain
IIA ~ (tk)UT(k, n)IIE->E :S
IIA ~ (tk)R7,jn( r A(tn-l))IIE->E
r=n+l
IIA~ (tk)uT(k, r)IIE->EII[Rj,l(TA(tr-l)) - Rj,l(TA(tn-d)]A-1(tn-dIlE->E
<
Therefore,
162
Now using the difference analogy of the integral inequality, we obtain the estimate
From these and the identity
it follows that
IIA(tk)UT(k, n) I E--+ E <
IIA(tkYiLT(k, i)A-1/2(ti) IIE--+EIIA1/2 (ti)UT(i, n) IIE--+E

M1
< (k-n)7'
Lemma 4.2.19. Suppose that the assumption (4.62) holds for r = O. Then for any
o :S n < r :S k :S Nand 0 :S a :S 1 the following estimates hold:
IIR7t-n(7A(tk+r)) - R7,ln(7A(tk))IIE--+E :S MiII,
II [A(tk+r)R7t- n(7A(tk+r)) -
(4.79)
A(tk)R7,ln(7A(tk))]A-1(tn)IIE--+E :S MiII, (4.80)
IIA(tk+r)R7t-n(7A(tk+r)) - A(tk)R7,ln(7A(tk))IIE--+E:S M (k
~n)7'
IluT(k + r, n) - uT(k, n)II E--+ E :S MiII,

II [A(tk+r)uT(k + r, n) - A(tk)uT(k, n)]A- 1(t n )IIE--+E :S MiII,
IIA(h+r)uT(k + r, n) - A(tk)uT(k, n)IIE--+E :S
where M
0 does not depend on
7,
(k~:)7'
k, r, n. Here iII = r"'(k - n)-"'.
Proof. We have that

R~r(7A(h+r)) - Rj,I(7A(tk)) = Rjr(7A(tk+r)) - R~r(7A(tk))
+Rjr(7A(tk)) - Rj,I(7A(tk)) = J 1 + h,
(4.81)
(4.82)
(4.83)
(4.84)
163
where
h = R~r(rA(tk+r)) - Rjr(rA(tk)),
h
= Rjr(rA(tk)) - Rj,l(rA(tk))'
From estimate (4.67) it follows that
Now let us estimate J 1 . We have that

[ i+2 r]
J1 =
i+r-1
(L + L
s=o
)Rjr-S(rA(tk+r))
s=[i.]+l
where
n
= L
r
J ll
s=o
R~r-S(rA(tk+r))(Rj,l(rA(tk+r)) - Rj,l(rA(tk)))Rj,l (rA(tk)) ,
i+r-1
J 12 =
S=[it r]+l
Rjr-S(rA(tk+r))(Rj,l(rA(tk+r)) - Rj,l(rA(tk)))Rj,l (rA(tk)).

i+r-l
IIJ12 1I E
-+ E
::;
s=[itr]+l
IIRjr-S(rA(tk+r))IIE-+E
Applying Abel's formula, we can write
[itr]
J ll =
s=o
Rjr-S(rA(tk+r))(Rj,l(rA(tk+r)) - Rj,l(rA(tk)))(I - Rj ,l(rA(tk)))-l
[itr]
X (Rj,l
(rA(tk)) - Rj,l (rA(tk))) =
s=l
(R~r-S+1(rA(tk+r)) - Rjr-S(rA(tk+r)))
164
x (Rj,l(TA(tk+r)) - Rj,l(TA(tk))) (I - Rj ,l(TA(tk))-lRj,l(TA(tk))
+R::r_[itr] (TA(tk+r ))(Rj,l(TA(tk+r)) - Rj,l(T A(tk)))R~,T]+l (TA(tk))

-R~r(TA(tk+r)) [Rj,l(TA(tk+r)) - Rj,l(TA(tk))].
Using the last formula and estimates (4.64), (4.65), (4.77) and
II(TA(tk))-l(1 - Rj,l(TA(tk))IIE---.E ::; M,

IITA(tk)(I - Rj ,l(TA(tk)))-l Rj,l(TA(tk))IIE---.E ::; M,
IIRj,l(TA(tk+r)) - Rj,l(TA(tk))II E---. E ::; MrT,
we obtain
II J 11II E ---. E
[]
'L
::;
II R jr-
8
(TA(t k+r )) -
8=1
R~r-8(TA(tk+r))IIE---'E
IIRj,l(TA(tk+r)) - Rj,l(TA(tk))II E---. E 11(1 - Rj ,l(TA(tk))-l Rj,l(TA(tk))II E---. E

+IIR::r_[itr] (TA(tk+r ))IIE---.E IIRj,l(TA(tk+r)) - Rj,l(TA(tk))II E---. E
x IIRtT]+l (TA(tk))IIE---.E
+
IIR~r(TA(tk+r))IIE---'E IIRj,l(TA(tk+r)) -
Rj,l(TA(tk))II E ---. E
[itr]
1
::; MrT + MT ""'
::; M 1rT ::; M 1w.
L.J n+r-s
8=1
Combining the estimates for J 11 and J 12 , we obtain
From this and the estimate for h follows estimate (4.79). The proofs of estimates
(4.80) and (4.81) follow the scheme of the proof of estimate (4.79). Now let us
establish (4.83). Let us denote
wT(k,n) = A(tk)uT(k,n)A- 1(t n).

By (4.73),
'L
r=n+1
A(tk)R;,lr+1(TA(tk_d)(Rj ,l(TA(tk_1))
165
where
J1(k,n) = A(h)RJ,ln(rA(tk_l))A-1(tn),
k
J2 (k,n) =
r=n+l
A(tk)RJ7+l (rA(tk-I))(Rj,1 (rA(tk-I))
- Rj,l(r A(tt_l)))A- 1(t r _1)W1' (r - 1, n).

Applying the identity
J1(k + r, n) - J1(k, n) = A(tk+r)[RJr- n(r A(tk+r-1)) - RJ,ln(r A(tk-I))]A -1 (t n)

+(A(tk+r) - A(tk))RJ,ln(rA(tk))A- 1(t n )
and estimates (4.64), (4.65), (4.62), we obtain
11J1(k + r, n) - J1(k, n)II E--. E
IIA(tk+r)[RJr-n(rA(tk+r-1)) - RJ,ln(rA(tk_I))]A- 1(tn) IIE--.E
1
+ II(A(tk+r) - A(tk))A- (tn)IIE--.E IIRJ,ln(rA(tk))IIE--.E
~ MiI!.
Now let us estimate h(k+r, n)-J2 (k, n). Using estimates (4.65), (4.62), we obtain
IIJ2 (k,n)IIE--.E
~M
r=n+1
IIA(tk)RJ,lr+1(rA(tk-1))IIe--.E
From this and the triangle estimate it follows that
Ilh(k + r, n) for all k - n
J2 (k
2r. Now let k - n
+ r, n) - J2 (k, n) =
- L
J2 (k, n)II E --. E ~ Mrr
> 2r. We have that

k+r
A(tk+r)RJr-i+1(rA(tk+r-1))
i=k-r+1
A(tk)RJt+ 1(rA(tk-I)HRj,l(rA(tk-1))
i=k-r+1
- Rj,l(rA(ti-I)))A -1 (ti_1)W1' (i - 1, n)
166
k-r
(A(tk+r)R;,tr-i+l(rA(tk+r_I)) - A(tk_dR;,li+l(rA(tk_I)))
i=n+1
k-r
A(tk+r)R;,tr-i+l(rA(tk+r-d)(Rj,l(rA(tk+r-I))
i=n+1
-Rj,l(rA(tk_1)))A- 1(ti_dwT(i - 1, n) =
LL
n,
n=l
where
k+r
L1=
A(tk+r)R;,tr-i+l(rA(tk+r-1))(Rj,l(rA(tk+r-1))
i=k-r+1
L2 = -
A(tk)R;t+1(rA(tk-I))(Rj,l(rA(tk-I))
i=k-r+1
L3 =
k-r
(A(tk+r)R;,tr-i+l(rA(tk+r_r)) - A(tk_1)R;,-;-i+l(rA(tk_r)))
i=n+1
k-r
L4 =
A(tk+r)R;,tr-i+1(rA(tk+r-1))(Rj,l(rA(tk+r-I))
i=n+1
- Rj,l(rA(tk-1 )))A- 1 (ti_1)W T(i - 1, n).

IIL11IE-->E :S M
k+r
i=k-r+1
IIA(tk+r)R;,tr-i+1(rA(tk+r-1))IIE-->E
167
Now let us estimate L 3 . Using estimates (4.82), (4.62) and (4.75), we obtain
k-r
<
IIA(tk+r)Rj,tr-Hl(rA(tk+r-l)) - A(tk_dRj,l"i+l (rA(tk-d) IIE---.E
i=l
Finally let us estimate L 4 . Using estimates (4.65), (4.62) and (4.75), we obtain
IIL4 1I E---. E:S
k-r
k-r
:S M
IIA(tk+r)Rj,tr-Hl(rA(tk+r-l))IIE---.E
i=n+l
.
t=n+l
(k
rrr
. ) : S M 1TT(1
+r-z+1 T
+ Iln(rr)J).
Thus, the operator L 4 has "worse" properties than the operators L n for n = 1,2,3.
Combining the estimates for L n for n = 1,2,3,4, we obtain
for all 0: < 1. Nevertheless this estimate permits us to establish the same estimate
for L 4 . Indeed, using the formula
L 4 = A(tk+r)(I - Rj,l(TA(tk+r-d))-l(R;~(TA(tk+r-l))
- Rj,tr-n(TA(tk+r-d))(Rj,l (T A(tk+r-l))
-Rj,l(TA(tk_d))A-1(tk_l)Wr(k -l,n)
k-r
A(tk+r)Rj,tr-i+l(TA(ik+r_d)
i=n+l
k-r
i=n+l
A(tk+r)Rj,tr-Hl(TA(tk+r-l))(Rj,l(TA(tk+r-d)
168
and estimates (4.65), (4.62) and (4.75), we obtain
k-r
""'
rr .) (k - i)n
IIL 4 11 E ~ E:S Mrr + M ~ (k +r-2r
i=n+l
k-r
""'
+M(a) ~ (k
i=n+l
rr
((k - i)r)O
+ r _ i)r ((i + 1 _ n)r)o r:S
M
2
rr .
Now combining the estimates for L n for n = 1,2,3,4, we obtain estimate (4.83).
In a similar manner one establishes estimate (4.82). Finally let us establish (4.84).
Estimate (4.84) for k-n :s 2 follows from estimates (4.76) and (4.77). Let k-n > 2.
Then using the formula
A(tk+r)ur(k + r,n) - A(tk)uT(k,n) = (wT(k
+ r, [k; n])
-wT(k, [k; n]))A([k; n]r)uT([k; n],n)

and estimates (4.82) and (4.76), we obtain estimate (4.84) for k - n > 2. Lemma
4.2.19 is proved.
Suppose that the assumption (4.62) holds for r = 0 and r = 1. Then
IluT(k,k -1) - Rj,l(rA(h-d)IIE~E:S Mr,

lJ[uT(k, k - 1) - Rj,l(rA(tk-d)]A-l(tk-dIIE~E
:s Mr 2 .
Now using the formulas
uT(k, n) - uT(k, n) =
k-l
L uT(k, i + l)(uT(i + 1, i) - uT(i + 1, i))uT(i, n),
i=n
k-l
uT(k, n) - uT(k, n) =
L uT(k, i + l)(uT(i + 1, i) -
uT(i + 1, i))uT(i, n)
i=n
and the last estimates and estimates of Lemma 4.2.18 and Lemma 4.2.19, we can
establish the estimates for uT(k, n).
Lemma 4.2.20. Suppose that the assumption (4.62) holds for r = 0 and r = 1.
Then for any 0 n < r k N, 0 a 1 the following estimates hold:
:s
:s :s
:s :s
IluT(k, n)IIE~E :s M, II A (tk)u T(k, n)A-l(tn)IIE~E :s M,
IIA(tk)uT(k, n)IIE~E
:s
(k
~n)r'
IluT(k
169
+ r, n) - uT(k, n)II E --. E :S Milt,
II[A(tk+T)uT(k + r, n) - A(tk)uT(k, n)]A-1(tn)IIE--.E :S Milt,

IIA(tk+r)uT(k + r, n) - A(tk)uT(k, n)II E--. E :S (k ~:)T'
where M 2 0 does not depend on T, k, r, n. Here ilt = rOt(k - n)-Ot.
From formula (4.68) and the estimates of Lemma 4.2.20 one derives the
following results.
Theorem 4.2.21. The difference problem (4.63) is stable in CT(E).
Theorem 4.2.22. The difference problem (4.63) is stable in C::,Ot(E).
Theorem 4.2.23. The difference problem (4.63) is stable in Lp,T(E), 1 :S p
< 00.
Theorem 4.2.24. The solutions of the difference problem (4.63) in CT(E) obey the
almost coercive inequality
II{T-1(Uk - uk-d}fllcT(E)+ II {T- (1-U (k,k -l))uk-dfllcT(E)
:S M1[II A(O)uoII E +
min{ln~, 1 +
lIn IIA(O)IIE--.EI}II
r, uo, but also of T.
r IlcT(E)],
The proof of Theorem 4.2.24 follows the scheme of the proof of Theorem
4.2.4 and relies on the estimates of Lemma 4.2.20.
Now let us study the well-posedness of the difference problem (4.63) in the
Theorem 4.2.25. The solutions of the difference problem (4.63) in C::,Ot(E) obey
the coercivity inequality

II{ T-1(Uk - Uk-l)}fllc;'O(E)+
II
{T-1(I - uT(k, k - l))Uk-l}fllc;',O(E)
:S M1[a(1 ~ a) Ilrllc;'O(E) + IIA(O)uoII E]'

r, uo, a, but also of T.
4.2.5 and relies on the estimates of Lemma 4.2.20 and on the formula
T- 1(1 - uT(k,k -l))Uk-l = T- 1(1 - uT(k,k -l))uT(k -l,O))uo
+(1 - R7,11{ -TA(tk-l)} )f{l
170
k-1
+ 2: [(I -uT(k, k-1))uT(k-1, i) - (I -Rj,l{ -TA(tk-1)} )R~t-1{-TA(tk-1)}]ff

i=l
k-1
uT(k, k - l))u T(k - 1, i) (f!,l - ff)
+ L(I i=l
Now let us give the definition of positive operators and introduce the fractional
spaces that will be needed in the sequel.
The operator A is said to be strongly positive if its spectrum a (A) lies in
the interior of the sector of angle , 0 < 2 < 7T, symmetric with respect to the
real axis and if on the edges of this sector, 8 1 () = {peic/> : 0 ::; p ::; oo} and
8 2 () = {pe-ic/>: 0::; p::; oo} and outside of the resolvent (A - A)-l is subject to
the bound
II
(A-A)-lll
E-.E
::; M().
1 + IAI
(4.85)
The infimum of all such angles is called the spectral angle of the strongly positive
operator A and is denoted by (A) = (A, E). Since the spectrum a (A) is a closed
set, it lies inside the sector formed by the rays 8 1 ( (A)) and 8 2 ( (A)) and some
neighborhood of the apex of this sector does not intersect a (A). We shall consider
contours r = r (, r) composed by the rays 8 1 (), 8 2 () and an arc of circle of
radius r centered at the origin; and r will be chosen so that a (A) < lal < 7T/2
and the arc of circle of radius r lies in the resolvent set p (A) of the operator A.
Let f (z) be an analytic function on the set bounded by such a contour r
and suppose that f satisfies estimate
If (z)j
::; M
Izl-c
for some c > O. Then the operator Cauchy-Riesz integral
f (A)
rf (z)(z - A)-l dz
27TZ Jr
(4.86)
converges in the operator norm and defines a bounded linear operator f (A), a function of the strongly positive operator A. If f (z) is continuous in a neighborhood
of the origin, then in (4.86 ) we shall consider that r = 0, i.e., r = 8 1 () U8 2 ().
As in the case of a bounded operator A one shows that f (A) does not depend
on the choice of the contour r in the domain of analyticity of the function f (z)
and that the correspondence between the function f (z) and the operator f (A) is
linear and multiplicative.
The function f (z) = z-a defines a bounded operator A-a whenever a>O.
Here the contour r is chosen with r > O. By the multiplicativity property,
A-(a+,B) = A-a A-,B = A-,B A-a for any powers of the strongly positive operator A and not only for negative integer ones. From this identity it follows (when
a + (3 is an integer) that the equation A-ax = 0 has the unique solution x = O.
171
Hence, the positive powers AQ = (A-Q)-l of the strongly positive operator are
defined. The operators AQ (a> 0) are unbounded if A is unbounded; they have
dense domains D (AQ) and one has the continuous embeddings D (AQ) C D (AI3)
if (3 < a.
The theory of fractional powers of operators can be constructed for a wider
class of positive operators. For such operators the estimate (4.85) is required to
hold for some and not only from the interval [0, 1r /2], but from the larger interval
[0,1r).
Now let us consider the function f (z) = e- tz . For any t > 0 this function
tends to zero faster than any power z-Q as Izl ---+ 00 and its values lie inside any
sector bounded by a contour r. Therefore, formula (4.86) can be used to define
the function exp { -tA} of the strongly positive operator A. By multiplicativity,
the semigroup property holds:
Consider the function W(z) = zQe- tz for some a > 0 and t > O. Since,
obviously, W(z) ---+ 0 faster than any negative power of z as Izl ---+ 00, w (z) defines
the operator function
w (A) =
~
21r2
r zQe-
ir
tz
(z - A)-l dz.
(4.87)
Let us show that the operator exp {- tA} maps E into D (A Q) and AQexp {- tA}
= w(A). Let x be an arbitrary element of E. By the multiplicativity property,
(4.69) implies that
A-Qw (A) x
~
21r2
re-
ir
tz
(z - A)-l xdz
= exp {-tA}
x,
which proves our assertion. Thus, we have the formula

A exp {-tA} =
Q
~
21r2
r zQe-
ir
tz
(z - A)-l dz.
(4.88)
In the above argument we must assume that the contour r contains an arc
of radius r, since we applied the operator A -Q, which corresponds to the function
z-Q. The final formula (4.88) is valid for any (small) r > O. Since the integrand in
(4.88) is continuous at the point z = 0, letting z ---+ 0 we obtain the formula
172
for some 0 < < 7r /2. From this and estimate (4.85 ) it follows that
IIAC> exp {-tA}11
E--+E
:::;
M ()
7r
roo pc>-le-tpcos<Pdp =
Jo
M () rea) t-C>.
7r(cos)c>
(4.89)
In particular, we have the estimate
lI exp{-tA}II E --+ E
:::;
M().
7r
(4.90)
Let us show that the estimate (4.90) can be sharpened by a factor that decays
exponentially when t ~ +00.
Let A be a strongly positive operator. We claim that for sufficiently small
8> 0 the operator A - 8 is also strongly positive and (A - 8) = (A). Indeed,
let A E r (). Consider the equation AX - (A - 8) X = Y for an arbitrary y E E.
The substitution AX - Ax = z yields the equation z + 8 (A - A)-l z = y. Since
if A E r (), we see that for 8:::; [2M ()r1the equation for z has a unique solution
and Ilzll :::; 211yll Consequently, the equation for x has a unique solution and
Ilxll :::; M () [IAI +
1r
Ilzll :::; 2M () [IAI +
This means that the operator A - (A

[2M ()]-l and
+ 8)
1r
Ilyll
has a bounded inverse for 0
< 8 :::;
Thus, we have shown that A - 8 is a strongly positive operator. Hence, by (4.90),

we have the estimate
Ilexp {- (A - 8) t}II E --+ E
:::;
2M ().
7r
This obviously yields

(4.91)
1
where we can put 8 = [2M ()r .
Let t > 1. Then, using the semigroup property, we can write
exp {-tA} = exp {- A} exp {- (t - 1) A} .
Next, applying estimates (4.89) with t = 1 and (4.91), we obtain
IIAC> exp {-tA}11
E--+E
:::;
M () r(~) 2M () e- 6 (t-l)
7r (cos)
7r
173
Hence, the following estimate holds for t > 1:
If 0 < t ::S 1, then estimate (4.91) prevails. Combining these two estimates, we
conclude that
IIA exp {-tA}IIE--+ E ::s if () e-OtC Q
(4.92)
Q
for some if () > 0 and b > O.

Further, formula (4.86) allows us to establish that the operator-valued function exp { -tA} is differentiable in the operator norm for t > 0 and
d
dt exp {-tA} = -Aexp {-tA}.
(4.93)
In particular, this implies that exp {-tA} is continuous in the operator norm.
Using the semigroup property we deduce that the derivative of exp { -tA} is also
continuous in the operator norm for t > O. Finally, formula (4.93) shows that
the operator-valued function exp {-tA} has derivatives of arbitrary order in the
operator norm for t > O.
Now let xED (A). Then the (E-valued) function exp {-tA} x has a derivative for t > 0 and, by (4.93),
d
dt exp{-tA}x = -exp{-tA} Ax.
Next, for x as above we can write
Using formula (4.86), we obtain

exp {-tA} x =
Here the contour
r e-tz[z-lx +
271'z Jr
z-l
(z - A)-l Ax]dz.
has the form
I
I
/ cI>
(4.94)
174

Using the Cauchy theorem, we get
1
exp{-tA} x = -2.
1rZ
re-tzz-1(z_A)-1 Axdz+x.
ir
Estimate (4.85) shows that in the last equality one can pass to the limit
under the integral sign when t ~ +0. Hence, the limit
lim exp {-tA} x = x
t-++O
+~
21rz
irr Z-l (z -
A)-l Axdz
exists (in the norm of E). By Cauchy's theorem, the integral

1
{) = -2.
1rZ
for some
(j
ir
Z-l
.j-O"+ioo
-O"-ioo
1
(z - A)-l Axdz = -2
Z-l
(z _ A)-l Axdz
1rZ
> O. Hence, by (4.85),

M
II{)II E ~ -2
1r
Since {) does not depend on
(j,
joo
-00 (j
dt
+ t2
II Ax II E
it follows that {) == O. Hence, we proved that

lim exp{-tA}x = x
t-++O
(4.95)
for any x E D(A). Since the norm Ilexp{-tA}II E -+ E is uniformly bounded for
t > 0, the limit relation (4.95) holds for any x E E.
Thus, if we extend the operator-valued function U(t) = exp{-tA}, t > 0,
at t = 0 by U (0) = I, we obtain a strongly continuous semigroup. From estimate
(4.92) (with a = 0) it follows that this semigroup is analytic. Finally, let us show
that its generator is U' (0) = -A. From (4.94) and estimate (4.91) we derive the
identity
U (t) x - x = - i t U (8) Axd8

for xED (A). Since U (t) is strongly continuous to the left at the point t = 0, this
implies that xED (U' (0)) and U' (0) x = -Ax. Hence, U' (0) is an extension of
the operator -A. By estimate (4.91), the operator U' (0) + A and -A + A have
bounded inverses for any A < O. Therefore, U' (0) = -A.
We have shown that the operator-valued function exp {-tA} is an analytic
semigroup with generator - A and with an exponentially decaying norm. Operators
- A that generate such semigroups were called strongly positive operators.
Let A be a strongly positive operator. With the help of A we introduce the
fractional space E~,q (E, A), 0 < a < 1, consisting of all vEE for which the
following norms are finite:
175
Recall that in Section 4.1 of Chapter 4 we introduced the fractional spaces

EQ,q in which the norms are defined with the help of the semigroup exp { -tA}
generated by a strongly positive operator A.
Theorem 4.2.26. E~,q(E, A(t)) = EQ,q(E, A(t)) for all 0
1 q 00.
:s :s
:s t :s T,O < a
< 1 and
Now let us consider the well-posedness of the difference problem (4.63) in the
Theorem 4.2.27. The solutions of the difference problem (4.63) in
CT(E~)
obey

T
II{T-1(uk - uk-I}}f IIcT(E~)+ II {T-1(I - u (k, k -l))uk-df IlcT(E~)
:s M 1[a(1 ~ a) I r IIcT(E~) + I A(O)uoIIE~],

where M1 is independent not only of
r, uo, a, but also of T.
4.2.6 and relies on the estimates of Lemma 4.2.20 and on the formula
k-l
+ L(I - Rj,l{ -TA(tk_l)})Rj,ii-l{ -TA(tk-I}}l!1,1
i=l
k-l
+ ~)(I
uT(k, k - 1))uT(k - 1, i) - (I - Rj,l{ -TA(tk-l)})
i=l
k i 1
x R j,1- - { -T A(t k-l )}]fj,l
k'

Theorem 4.2.28. The solutions of the difference problem (4.63) in C~,'Y(E)
(0 "f {3, 0 < {3 < 1) satisfy the following coercivity inequalities:
:s :s
II
{T-1(Uk - Uk-l)}f
+M111
I
IlcT(E~_~) :s {3(~ {3) II r Ilc~'~(E)
(1 - Rj,I(TA(0)))T-1Uo -
{T-1(Uk - Uk-l)}f IIc~'~(E) +
II
H,I IIE~_~'
{T-1(I - uT(k,k -l))uk-df Ilc~'~(E)
:s {3(~ {3) I r IIc~'~(E) + M11(I -
Rj,I(TA(0)))T-1Uo -
where M 1 is independent not only of r,uo, 13, "f but also ofT.
H,lI~''Y,
176
Here, Iwlg,')' denotes the norm of the Banach space Eg,')' and consists of those
wEE for which the norm
sup
O:::;i<i+r:::;N
(rr)-/3((i
+ r)r)1' II(R)ir- 1(rA(t)) - R)i1l(rA(t)))wIIE

'
is finite.
Note that the parameter 'Y can be chosen freely in [0, ,8), which increases
the number of spaces C~'')'(E)(O ::; 'Y ::; ,8,0 < ,8 < 1) of grid functions in which
Theorem 4.2.29. The solutions of the difference problem (4.63) in C~'')'(E~_/3)
(0 ::; 'Y ::; ,8 ::; Q < 1) satisfy the following coercivity inequalities:
{r-1(uk - Uk-l)}f Ilc~'-Y(E~_I3)+ II {r-1(uk - Uk-l)}f IIcT(E~~I3)
+ II {r-1(I - uT(k, k -l))uk-df Ilc~'-Y(E~_I3)

::;
Q(~ Q) II r Ilc~'-Y(E~_I3) + M1!(I -
Rj,l(rA(0)))r-1uo -
Jf,II:~/3'
II {r-1(uk - Uk-l)}f IIc~'-Y(E~_I3)+ II {r-1(uk - Uk-l)}f IlcT(E~_-y)

+ II {r-1(I - uT(k, k -l))uk-df
::;
Q(~ Q) II r Ilc~'-Y(E~_I3) + M111(I -
r, Uo,
IIc~'-Y(E~_I3)
Rj,l(rA(0)))r-1uo Q,
fi,lIIE~_-y'
,8, 'Y but also of T-
Here, Iwl~:2/3 denotes that the norm of the Banach space E~:2/3 consists of
those wEE for which the norm
is finite.
Note that the spaces Ce'')'(E~_f3) of grid functions, in which coercive solvability has been established, depend on the parameters Q, ,8 and 'Y. However, the
constants in the coercive inequalities depend only on Q. Hence, we can choose the
parameters ,8 and 'Y freely, which increases the number of spaces of grid functions
177
Lp,T(E) = Lp([O, T]T' E), 1 ::; p < 00 of all grid functions. We have not been able
II {r-1(uk-Uk-d}f IIL p.T(E) + II {r-1(I - uT(k,k -l))Uk-df IILp.T(E)
::; M[li
IILp.T(E)
+ II A(O)uoII E]'
in the arbitrary Banach space E and for the general strong positive operator A(t).
Nevertheless, we can establish the almost coercivity inequality.
Theorem 4.2.30. The solutions of the difference problem (4.63) in Lp,T(E) obey
the almost coercive inequality
II {r-1(uk-Uk-d}f IILp,T(E) + II {r-1(I - uT(k,k -l))Uk-df IILp.T(E)
::; MIlII A(O)uoII E + min
{ln~, 1 + lIn II A(O) IIE->EI} II r
IIL p,T(E)],
where M 1 does not depend on fT, uo, p and r.

Note that if A(O) is the unbounded positive operator in E, then
. { In:;:,
1 1 + lIn II A(O) IIE->EI } = In:;:.
1
mm
Finally, let us give, without proof, the following results about well-posedness
of the initial-value difference problem (4.63) in the spaces Lp,T(E) = Lp([O, T]T' E),
1 ::; p
< 00.
Theorem 4.2.31. Suppose that the difference problem (4.63) is well posed in
LpO,T(E) for some Po, 1 < Po < 00. Then it is well posed in Lp,T(E) for all
p, 1 < p < 00 and the following coercivity inequality holds:
II {r-1(uk-Uk-l)}f IILp,T(E) + II {r-1(I - uT(k,k -l))uk-df IILp,T(E)
2
+lluTlb(E1_1/P,p) ::; M(poHII Uo IIEl-l/p.p +
where M(po) does not depend on
r, uo, p and r.
p~ 1
II
II],
Theorem 4.2.32. Let 1 ::; P ::; 00 and 0< a < 1. The problem (4.63) is well posed
in Lp,T(E~,p) and the following coercivity inequality holds:
II {r-1(uk-Uk-l)}f IILp.T(E:',p) + II {r-1(I - uT(k,k -l))uk-df IILp'T(E:"p)
::; M[II A(O)uoIIE:"p
+ a(l ~ a)
II
where M does not depend on rr, Uo, p, a and r.
r IILp,T(E:')'
178
is well posed in
< p, q < 00 and 0 < a < 1. The difference problem
Lp,T(E~,q)
(4.63)
:s M(q)[11 A(O)uoIiEIo,q + ~
II r II
pwhere M(q) does not depend on fT, Uo, p, aand
Lp,T(E~.q)
],
To
Now, the abstract theorems given above are applied in the investigation of
difference schemes of higher order of accuracy with respect to the set of all variables
for approximate solution of the Cauchy problem (4.52). The discretization of problem (4.52) is carried out in two steps. Let us define the grid space Rh (0 < h ho)
as the set of all points of the Euclidean space R n whose coordinates are given by
:s
sk=O,1,2,... , k=l, ... ,n.

The number h is called the step of the grid space. A function defined on
R h will be called a grid function. To the differential operator A with constant
coefficients of the form
we assign the difference operator

'""'
L..J
~Sl ~S2
1-
1+
~S2n-l~S2n
nn+ '
(4.96)
2m$lsl$S
which acts on functions defined on the entire space R h. Here s E R 2n is a vector

with nonnegative integer coordinates,
is the unit vector of the axis Xk.

An infinitely differentiable function of the continuous argument y E Rn that
is continuous and bounded together with all its derivatives is said to be smooth.
Let <p(y) be a smooth function on Rn. Using the Taylor expansion of <p(y), one
can show that
and
ek
Here the grid function <p (x) and -aa

<p (x) are the traces of the functions <p (y)
Yk
and -a
ayk <p (y), respectively. The last inequality means that the difference operator
h -1 ~k is a first-order approximation for the differential operator -a
a .
Yk
179
We say that the difference operator Ah is a Ath-order (A > 0) approximation

of the differential operator A if the inequality
sup IAh'P (x) - A'P (x)j :::; M ('I') h A
xER~
holds for any smooth function 'I' (y). We shall assume that the operator A h approximates the differential operator A with any prescribed order.
A function of a continuous [resp., discrete] argument that decays at infinity
faster than any negative power of Iyl [resp., Ixl] is said to be rapidly decreasing.
Let us define the Fourier transform of a grid function fh (x) by the formula
1(0 =
(21rr n
L exp{-i(x,~)}fh(x)hn,~ERn.
(4.97)
xER~
This formula defines a 21rh- 1-periodic smooth function of the continuous argument
~ whenever fh (x) is a rapidly decreasing grid function. In this last case (4.97) is
just a Fourier series expansion of the function 1(~) and the numbers
(x) are
the Fourier coefficients, given by the formula
fh (x)
1 1
16/:-:;7Th- 1
l~nl:-:;7Th-l
exp{i(x,~)} 1(~)d6" d~n.
(4.98)
The inverse Fourier transform of a 21rh- 1-periodic function 'I' (~) is defined to be
the grid function <ph (x) given by the formula
<ph (x)
=1
l~ll:-:;7Th-l
.. r
JI~nl~7Th-l
exp{i(x,0}'P(~)d6d~n.
(4.99)
Formulas (4.98) and (4.99) establish a one-to-one correspondence between

rapidly decreasing grid functions of a continuous argument. In particular, if fh (x)
is a rapidly decreasing grid function, then
-h
[1]
(x) = fh (x).
If fh (x) is a rapidly decreasing grid function, then the grid function Ahfh (x)
exists and is given by (4.97) and we have the equality
Ahf (~)
= A (~h, h)
1(~).
The function A (~h, h) is obtained by replacing the operator ~k in the right-hand

side of equality (4.96) with the expression (exp {i~kh} - 1), respectively and is
called the symbol of the difference operator. Since exp {i~kh} are bounded analytic 21rh- 1-periodic functions, the symbol A (~h, h) is a bounded analytic 21rh- 1periodic function. It follows that for large I~I one has the estimate
180
In the first step let us give the difference operator A~x by the formula
At,xuh
=
h
x
h
bt,x
Drh u Xh + 6u x
r
L..J
(4.100)
2m~lrl~S
The coefficients are chosen in such a way that the operator A~x approximates in
a specified way the operator
We shall assume that for I~khl :s: 1r and fixed t and x the symbol
operator A~x - 6 satisfies the inequalities
At,X(~h,
h) of the
With the help of A~x we arrive at the initial-value problem
_dv_h--,--(t_,x--'--)
dt
+ At,Xvh(t
x)
h'
= fh(t x) 0
<t <T
,,--,
(4.101)
for an infinite system of ordinary differential equations.

In the second step we replace problem (4.101) by the difference scheme
(4.102)
Let us give a number of corollaries of the abstract theorems given in the

above. To formulate our result we need to introduce the space C~ = C13(R'h),
o:s: f3 :s: 1 of all bounded grid functions uh(x) defined on R'h, equipped with the
norm
Ilu Il c/3
h
Ilu Ilc + x,ye:Rsup,xopy

h
luh(x) - uh(y)1
I
/13
X -
Here Ch = C(R"h) denotes the Banach space of bounded grid functions uh(x)
defined on R"h, equipped with the norm
lIu h llc
= sup luh(x)l.
xe:R h
181
Theorem 4.2.34. The solutions of the difference schemes (4.102) satisfy the fol-
lowing stability estimates:

II UT,h Ilc;-'''(cn:S M[llu~llc~+ II F,h II c ;-'"(cn], 0
< a < 1,0:S (3 < 1,
II uT,h I Lp ,r(cn:S M[llu31Ic~+ II F,h IILp,r(c~)], 1 :S P:S 00,0 :S (3 < 1,
r,h, ug,
a, (3, p, hand T
The proof of Theorem 4.2.34 is based on the abstract Theorems 4.2.21,4.2.22,

4.2.23 and on the following two theorems.
Theorem 4.2.35. The operator A~x is strongly positive in Ch .
Theorem 4.2.36. For any 0
< (3 < 2~ the norms in the spaces E~(Ch,A~X) and
C~m{3 are equivalent uniformly in h.

Theorem 4.2.37. The solutions of the difference schemes (4.102) satisfy the fol-
lowing almost coercive stability estimates:
+ In T ~ h II r,h IILp.r(C,,)], 1 :S p :S 00,
where M does not depend on r,h, ug, p, hand T.

The proof of Theorem 4.2.37 is based on the abstract Theorems 4.2.24 and
4.2.30, the positivity of the operator AI'; in C~ and on the almost coercivity inequality for an elliptic operator A~x in C h and on the estimate
min
{ln~,
1 + lIn II A~x
T
Ilcil cill}:s MIn
"-0,,
~h'
T+
Next, to formulate our result we need to introduce the space W:'h = Wf(Rh),
o :S (3 :S 1,1 :S p < 00 of all bounded grid functions uh(x) defined on R h, equipped
with the norm
Here Lp,h = Lp(Rh) denotes the Banach space of bounded grid functions uh(x)
defined on R h, equipped with the norm
IluhIIL p,,,
[L
xcR h
luh(x)IPhnl~
182
Theorem 4.2.38. The solutions of the difference schemes (4.102) satisfy the coer-
civity estimates:
II {T-l(U~ - ULlnf- 1 Ilc~'''(cn
::; M(a,,B)[
Ir l=2m
IIDhu~llc~+ I
r,h IIc~'''(c~)],O::; a < 1,0 <,B < 1,
I {T-l(U~ - ULlnf- 1 II Lp 'T(w;'h)

::; M(a,p,q)[
Ir l=2m
IIDhu~llw;'h+ I r,h
IILp,T(w;'h)], 1 < p,q < 00,0 < a <
where M(a,,B) and M(a,p,q) do not depend on
r,h,
ua, hand
~,
The proof of Theorem 4.2,38 is based on the abstract Theorems 4.2.25, 4.2.33
and Theorems 4.2,35, 4.2.36 and on the following two theorems.
Theorem 4.2.39. The operator A~x is strongly positive in
Theorem 4.2.40. For any 0
W q,2m
(3
h
< ,B <
W:'h'
2~ the norms in the spaces E~,q(Lq,h,
A,J
and
are equivalent uniformly in h,
Note that in a similar way the difference schemes of the high order of accuracy
with respect to one variable for approximate solutions of the mixed problems (4,50)
and (4.51) can be constructed. Abstract theorems given from above permit us to
obtain the stability, the almost stability and the coercive stability estimates for
the solutions of these difference schemes.
Now, we will consider the second approach of choice V(tk' tk-l) and
By (4.15) and (4.16), we have
V(tk, tk-d = e-A(tk-Ilr
(4.103)
(4.104)
(4.105)
183

(4.106)
(4.107)
1jt
where
Bi(t)
{ Bi(t)
= 1,
=
V(tk,s)f(s)ds
(4.108)
tk-l
= 0,
B~~l (t) + B i- 1(t)A(t), 1:S i :S j + l.
Now, we will study the approximate formulas for the expressions V(tk, tk-d and
V(tk,tk_l).
Using Taylor's formula and formulas (4.103), (4.104) and (4.105), we
2
obtain
V(tk, tk-d
1+1-1 (_
e- A(tk-ll"T - 7(A(tk) - A(tk_1))e- A(t k- ll "T
)n+1 n ( )
+ ~ (n 7+ I)! ~ ~ Bi(tk)[(A(tk) - A(tk-d)(-A(tk_dt- i
+~
(n ~ i) A(s) (tk)( -A(tk_dt-i-sJe-A(tk-ll"T +
0(71+1+1),
V(tk, tk-d = e-A(tk)"T + 7(A(tk) - A(tk_d)e-A(tkl"T
n ( )
?; (:n+1
+ I)! ~ ~ Bi(tk-d[(A(tk) - A(tk_d)(A(tk))n-i
_~ (n ~ i) A(S) (tk_d(A(tk))n-i-SJe-A(tk)"T +
j+I-1
0(71+1+1),
184
where
Bi(t) = 1,
i = 0,
{ Bi(t) = B;~l (t) - B i_ 1(t)A(t), 1:S i :S j + l.
can write
V(tk, tk-1) = Rj,I(A(tk_1)T) - T(A(tk) - A(tk-1))Rj- 1,1-1 (A(tk-1)T)
(4.109)
)+1-1 ( T)n+1 n ( )
i
+ 1) ~
Bi(tk)[(A(tk) - A(tk-1))(-A(tk_dt-
+ ~
+~
(n ~ i) A(s)(tk)( -A(tk-1)t-i-S]Rj_n+[~1_1,1_[~1(A(tk_1)T) + O(T)+I+1),
V(tk, tk-1) = Rj,I(TA(tk)) + T(A(tk) - A(tk-d)Rj- 1,1-1 (A(tk)T)

j+I-1
n+1
7 Bi(tk-1)[(A(tk) - A(tk_d)(A(tk)t-
n ( )
+ ~ (:+ I)! ~
-:- ~ (n ~ i) A(s)(tk_d(A(tk))n-i-s]Rq_n+[~l_l'P_[~l(TA(tk))
V(tk,tk-1) =
)+l-l (
+ O(T)+I+1),
A(tk_~))Rj-1,l-1(TA(tk_~))
(4.111)
7 Bi(tk)[(A(tk) - A(tk_~))(-A(tk_~))n-i
)n+1 n ( )
+ ~ (:: I)! ~
+~
- T(A(tk) -
Rj,I(TA(tk_~))
(4.110)
(n ~ i) A(S) (tk)( -A(tk_ ~ ))n-i-s]Rj_n+[~]_l,l_[~l (T A(tk_ ~)) + o(T)+I+l).
The approximate formulas for the expression V(tk, tk-l)

can be constructed using
2
Pade's rational functions. We have that
2
v(tk, tk_l)
2
tk
I
-A(tk_!)~ _l
V(tk' tk_l) = e-A(tkH +
tk _
= e
e-A(tk)(tk-Z)(A(tk) - A(z))v(z, tk_l)dz,

2
!
tk
V(tk, z)(A(z) - A(tk_l))e-A(tk_!)(z-t k_!) dz.
k-
185
j+l-l
7 Bi(tk_!)[(A(tk) - A(tk_!))(A(tk)t-i
n ( )
n+l
+ ~ 2n+;(n + 1)! ~
_
~ (n~i) A(S)(tk_!)(A(tk))n-i-SJe-A(tk)t +O(Ti+ 1+1),
V(tk, tk_l) = e
-A(t
j+l-l
(_ )n+l
2 -
-2 T(A(t k ) - A(tk_I))e
2
_lA(t
2
1)7
k- 2
7 Bi(tk)[(A(tk) - A(tk_!))(-A(tk_!))n-i
n ( )
+ ~ 2n+l~n+ 1)! ~
+ ~ (n ~
1)1:
k- 2
i) A(s)(tk)(-A(tk_!)t-i-SJe -A(tk_~)t + O(Ti+1+1).
Therefore, using the last formulas and Pade's approximants of the function
e- z , we can write
j+l-l
n+l
+ ~ 2n+;(n + 1)!
_
t; 7
n ( )
Bi(tk_!)[(A(tk) - A(tk_!))(A(tk))n-i
~ (n ~ i) A(S)(tk_!)(A(tk)t-i-SJRj_n+[~l_l,I_[~l(~A(tk))
1
+ O(Ti+l+1),
V(tk, t k_ 1)
= Rj,l( -2 A(tk- -d)
- -2T(A(tk) -A(t k_ 1
))Rj-1,1-1 (-2 A(t k_ 1))
(4.113)
2
2
2
2
i+l-l
(_ )n+l
+ ~ (n ~
7 Bi(tk)[(A(tk) - A(tk_!))( -A(tk_!))n-i
n ( )
+ ~ 2n+l~n + 1)! ~
i) A(S)(tk)(A(tk_!))n-i-S]Rj_n+[~)_l,I_[~)(~A(tk_!))
+ O(Tj+I+l).
Finally, replacing the expression V(tk, tk-l) by its Pade approximants from (4.109)
or (4.110) or (4.111) (neglecting the last terms, respectively) and the expression
186
by the expressions from (4.106) or (4.107) or (4.108) and replacing the expressions
V(tk' tk-l) and V(tk, tk_l)
by their Pade approximants from (4.109) or (4.110) or
2
(4.111) or (4.112) or (4.113) (neglecting the last terms, respectively), we obtain
the following difference schemes of (j + l)-th order of accuracy for the approximate
solution of the initial-value problem (4.1):
(4.114)
+T(A(tk) - A(tk-l))Rj - 1,I-l(A(tk-l)T)

-
)n+l n ( )
i
'0T+ 1) ~
Bi(tk)[(A(tk) - A(tk-d)( -A(tk_l)t-
]+1-1 (
+ ~ (n~i) A(S)(tk)(-A(tk_l))n-i-S
1
XRj-n+[~J-l,I-[~J(A(tk-dT)]]Uk-l = f~' ,1 ~ k ~ N, Uo = Yo;
T- 1 (Uk - Uk-I)
+ T- 1 [1- Rj,I(TA(tk))
(4.115)
-T(A(tk) - A(tk-l))Rj - 1 ,I-l(A(tk)T)

j+l-l
7 Bi (tk-l)[(A(tk) - A(tk_d)(A(tk))n-i
_~ (n ~ i) A(s)(tk_l)(A(tk)t-i - S
n+l
n ( )
- ~ (:+ I)! ~
xRj_n+[~I_l,I_[~I(TA(tk))]]Uk_l = f~' ,1 ~ k ~ N, Uo = Yo;
T- 1 (Uk - uk-d
+ T- 1 [1- Rj,I(TA(tk_l))
2
+T(A(tk) - A(t k_l))Rj

- 1 ,I-l(TA(t k_l))
2
2
j+l-l (
7 Bi(tk)[(A(tk) - A(tk_!))(-A(tk_!))n-i
)n+l n ( )
~ (~: I)! ~
+~
(n ~ i)
A(S) (tk)(-A(tk_!))n-i-S
1
Here,
n'
xRj-n+[~I-l,I-[~I(TA(tk_!))]]Uk-l = f~' ,1 ~ k ~ N,uo = Yo
can be defined by one of the following formulas:
(4.116)
187
or
rm
(m + I)! [Rj-m-[~I,I+[~)(A(tk-dr) - r(A(tk)
- A(tk-d )Rq - m-[ ~ ]-l,p+[ ~ ]-1 (A(tk-1 )r)

m-1(_ r )n+l
+~
+~
(n + 1)
t;
n
7 Bi(tk)[(A(tk) - A(tk-d)( -A(tk_d)n-i
()
(n ~ i) A(S) (tk)(-Ah-dt-i-SjRj-n+['i1-1,1-['iI(A(tk-1)r)]J(>') (tk-1 ),
or
rm
x (m + I)! [Rj_m_[~],I+[~](rA(tk))
+ r(A(tk)
- A(tk-d)
xRj-m-[~1-1,1+[~1-1(A(tk)r)
m-1
n+1
+ ~ (:+ I)!
t; 7
(
Bi(tk-d[(A(tk) - A(tk_1))(A(tk))n-i
- ~ (n ~ i) A(s) (tk_d(A(tk))n-i-s Rj-n+['i]-l,I-['i] (rA(tk))]J(A) (tk-d,

or
rm
x (m + I)! [Rj-m-[~],I+[~I(rA(tk_!)) - r(A(tk) - A(t k_!))

x Rj-m-[~1-1,l+[~1-1 (rA(t k_!))
(n: t; 7
m-1 (_ )n+1
+~
+~
(n ~ i)
I)!
Bi(tk)[(A(tk) - A(tk_!))(-A(tk_!))n-i
A(s) (tk)( -A(t k_!))n-i-S
Rj-n+[~]-l,l-[~J(rA(t k_! ))]J(>') (tk-1),
188
or
or
m-l
(_ )n+l
+ ~ 2n+l~n + I)! ~
+ ~ (n ~
7 Bi(tk)[(A(tk) - A(tk-!))(-A(tk_!))n-i
)
i) A(s)(tk)(A(tk_!))n-i-S Rj-n+[~I-l,I-[~I(~A(tk_!))]f(>')(tk_!)'
Note that in a similar manner with the difference schemes (4.63) one can investigate the stability, coercive stability of the difference schemes (4.114), (4.115) and
(4.116) under the smooth assumptions for A(t).
4.3
Single-Step Difference Schemes Generated

by Taylor's Decomposition
We consider again the Cauchy problem (4.1). Let the function v(t), for (0 :s: t :s: T)
have (l + j + 1) continuous derivatives and tk-l, tk E [0, T]T' Then by Theorem
2.2.1 we have the following Taylor decomposition on two points
I
V(tk) - V(tk-l) + L:>~iV(i)(tk)ri - L.BiV(i)(tk_dri

i=l
i=l
(4.117)
4.3. Single-Step Difference Schemes Generated by Taylor's Decomposition
where
Qi
{3i
(l+j-i)!l!(-l)i
(l+j)!i!(l-i)!
(l+j-i)!j!
(l+j)!i!(j-i)!
lor any
~
. 1
1,
lor any
189
:S t:S ,
(4.118)
. 1< .< .
I,
_ t _
J.
Now, we will consider the applications of Taylor's decomposition of function on

two points. From (4.117) it is clear that for the approximate solution of problem
(4.1) it is necessary to find v(i)(tk) for any i,l :S i :S land v(i)(tk_l) for any
i,l :S i :S j. Using the equation
Vf(t) = -A(t)v(t) + f(t),

we obtain
(4.119)
where
Al (t)
= -A(t), Ai(t) = A~~l (t) - A-I (t)A(t)
for any i, 2 :S i :S l,
It (t) = f(t), fi(t) = fi(~l (t) + A i- 1(t)f(t) for any i, 2 :S i :S l.

From this formula for Ai(t) it follows that
(4.120)
where
Therefore,
2) _l)i-r A~~l (t) Ai- r (t).

i
Bi(t) =
(4.121)
r=l
Replacing the v(i)(t) by (4.119) and neglecting the last term, we obtain the singlestep difference schemes of (l + j)-th order of accuracy for the approximate solution
of problem (4.1):
1
7- (Uk - Uk-I) -
Qi 7i -
A i- 1 (tk) Uk
(4.122)
i=l
j
+L
{3i 7i -
Ai- 1 (tk-l) Uk-l
fe,
i=l
f{l
= -
L
i=l
7 -
Qdi_l
(tk)
+L
i=l
i 1
7 - f3di_l
(tk-l), 1 ::; k ::; N, Uo
Yo
190
Suppose that the operator N(t)A(i-r) (t)A-(r+1)(t) satisfies the condition
liAr (t) A(i-r) (t) A-(Hr) (t)IIE-->E : :; M,1 :::; i:::; I- r,O:::; r:::; I-I,
(4.123)
where M is a positive constant independent of t, 0 :::; t :::; T. Then the following

estimate holds:
IIA~~l (t) A- r
M,1 :::; r :::; I.
(t)IIE-->E : :;

(4.124)
Using the identity
I
(I - I>::ti'TiA i- 1 (tk))-l - Qj}(7A(tk))

i=l
I
= (I - I:>~i7i Ai- 1 (tk))-17 L

i=l
fri7i-1
B i- 1 (tk) Qj}( 7 A(tk)),
i=l
the estimate (4.124) and
we obtain that the operator I - L~=l fri7i A i - 1 (tk) has the bounded inverse
(I - L~=l fri7iAi_l (tk))-l. Therefore the difference problem (4.122) is uniquely
solvable and the following formula holds:
k
s=l
i=l
Uk = uT(k, O)uo + L uT(k, 8)(1 - I>i7iAi-l (ts))-l 11,17,

where
(4.125)
uT(k,k-l)uT(n+l,n),k>n,
uT(k, n) =
I,k
= n.
(4.126)
Here
I
i=l
i=l
uT(k,k -1) = (I - Lfri7iAi-l (tk))-l(I - LBi7i Ai - 1 (tk-l)).
(4.127)
FT(E) of vectors v T = {Vd"=l' Recall that the space FT(E) can be equipped with
the various norms and thus become a normed space.
191
First we will define the operator D'T' acting from the space E x F'T(E) of
vectors w'T = {wkl"=o into the space F'T(E) of vectors v'T = {Vkl"=l by the rule
Second we will define an operator A'T from the space E x F'T(E) of vectors w'T =
{Wk}"=o into the space F'T(E) of vectors v'T = {Vk}"=l by the rule
j
v'T = A'TW'T,Vk = - L>~iTi-1Ai_l (tk)Wk

i=l
k= 1, ... ,N.
+ L.8iTi-1A_l (tk-l)Wk-l,
i=l
Next, let us introduce the continuation operator TI(uo), which acts from Ex F'T(E)
to F'T (E) according to the rule
operator equation
D'TTI(uo)u'T
Here
+ A'TTI(uo)u'T =
r.
fj,l)
f 'T -- (fj,l
1 , ... , N .
The last operator problem will be considered in the space F'T(E). From its unique
solvability for any Uo E E and
E F'T(E) it follows that its solution u'T defines
an additive and homogeneous operator u'T (r, uo) that is continuous.
The initial-value difference problem (4.122) is said to be stable in F'T(E) if
we have the stability inequality
r,

Uo but also of T.
Let us consider the operator v'T(k, n) defined by
v'(k,n)
~{
v'T(k, k - 1) v'T(n + 1, n), k > n,

(4.128)
I,k = n,
where
j
v'T(k,k -1)
(I - L.8iTiAi-l (tk-l))(I - LQiTiAi-l (tk_l))-l.

i=l
i=l
(4.129)
192
Using formulas (4.126), (4.127), (4.128) and (4.129), we obtain

I
(k, n)(I -
Qi Ti
Ai -
1 (tn))-l
i=l
= (I - I>iTiAi-1 (tk))-lVT(k
-1,n), 1:S n:S k-1.
i=l
From this and formula (4.125) it follows that

I
Uk = (I - L
Qi Ti A i-
T
1 (tk))-lV (k - 1, k)(I - L .Bi Ti A i- 1 (O))uo
i=l
(4.130)
i=l
8=1
i=l
+ L(I - LQiTiAi-1 (tk))-lVT(k -1,S)R' IT, k 2: 2,

I
U1
= (I - L
QiTiAi_1 (td)-l(I - L.BiTiAi-1 (O))UO
i=l
i=l
+ (I - L
Qi Ti Ai-
1 (td)-lJi,IT.
i=l
By formula (4.127),
j
vT(k, k -1) - Rj,I(TA(tk))
vT(k, k -1) - (I - L.BiTiAi-1 (tk))Qj} (TA(tk))

i=l
+(1 - L.BiTiAi-1 (tk) - Pj,I(TA(tk)))Qj}(TA(tk))

i=l
= -vT(k,k -1) L.BiTiBi-1 (tk) Qj,l(TA(tk)) - L.BiTiBi-1 (tk) Qj,l(TA(tk))'

i=l
i=l
Using estimate (4.124), we obtain
From this estimate it follows that

(4.132)
193
Now we define the operator
Since
17(k, k - 1) = uT(k + 1, k),
we obviously have
17(k, n) = uT(k + 1, n + 1).
Therefore, by Lemma 4.2.18 and Lemma 4.2.19, we have that

(4.133)
for 1 ::; n ::; k ::; N,
(4.134)
for 1 ::; n
< k ::; N,
IIA(tk+r)vT(k + r, n) - A(tk)vT(k, n)II E _
E ::;
(k ~:)T'
IHA(tk+r)vT(k + r, n) - A(tk)vT(k, n)]A-1(tn)IIE_E ::; MiII,
(4.135)
(4.136)
for 1 ::; n < k ::; k + r ::; N. Here iII = ra(k - n)-a, 0::; Q ::; 1.
From formula (4.130) and estimates (4.133), (4.134), (4.135) and (4.136) one
derives the following results.
stability inequality
j
IluTllc(T,E) ::; M[lI(I - L,BiTiAi-l (O))uoIIE + Ilrllc(T,E)],

i=l
r, uo, but also of
T.
Theorem 4.3.2. The solutions of the difference problem (4.122) in C;:,Q(E) obey
the stability inequality
j
IluTllc~,a(T,E) ::; M[II(I - L,BiTiAi-l (O))uoIIE + Ilrllc~,a(T,E)],

i=l
r, uo,
Q,
but also of T.
194
Theorem 4.3.3. The solutions of the difference problem (4.122) in Lp,T(E), 1 ::;
p < 00 obey the stability inequality
j
IluTIILp,T(E) ::; M[II(I - Ll3i TiA i-l (O))uoIIE + IlrIlLp,T(E)],

i=l
r, uo, p. but also of
T.
i=l
i=l
+ II {- LaiTi-1A_l (tk) Uk + Ll3i Ti - 1Ai_l (tk-duk-l}f IICT(E)

j
::; M1 [1I A(O)(I - Ll3iTiAi-l (0)) uoll E

i=l
+ min{ln~, 1 + lIn IIA(O)IIE-.EI}II r IlcT(E)],

r, uo, but also of
T.
Theorem 4.3.5. The solutions of the difference problem (4.122) in C,:,Q:(E) obey
i=l
i=l
+ II {- L ai Ti - 1Ai- 1(tk) Uk + L l3i Ti - 1Ai- 1(tk-l) uk-df IIc~'''(E)

::; M 1 [a(1 ~ a) IlrIlC~'''(E)
+ IIA(O) (I -
~l3iTiAi_l (0)) uoll E]'
T.

i=l
i=l
CT(E~)
+ II {- L ai Ti - 1Ai- 1(tk) Uk + Ll3i Ti - 1Ai- 1(tk-d uk-df IICT(E~)

::; Md a (l ~ a) II
r IICT(E~) + II A(O)(I -
Ll3i TiA i-1 (0))

>=1
T.
uoIlE~],
obey
195
Theorem 4.3.7. The solutions of the difference problem (4.122) in Lp,,,(E) obey
+11 {- ~::>);i'Ti-lA i- 1 (tk) Uk + L

i=l
.Bi Ti - Ai- 1 (tk-d uk-df

1
IILp,T(E)
i=l
j
::; M1[ll A(O)(I - L.BiTiAi-l (0)) uoll E

i=l
+ min {ln~, 1 + lIn II A(O) IIE--+EI} II

where M 1 does not depend on
r, uo, p and
IILp,T(E)],
To
Theorem 4.3.8. Suppose that difference problem (4.122) is well posed in Lpo,,,(E)
for some Po, 1 < Po < 00. Then it is well posed in Lp,,,(E) for all p, 1 < P < 00
+ II {-
1
L ai Ti - A i- 1 (tk) Uk
+ L.Bi Ti - 1 A i- 1 (tk-l) uk-df
i=l
IIL p.T(E)
i=l
+llu"IICT(E1 _ 1 / P.p)
2
::; M(po)[ll(I - L.BiTiAi-l (0)) uoIIEl_l/p,P + p~ 1

i=l
where M(po) does not depend on
II r
IILp,T(E)],
r, uo, p and T.
Theorem 4.3.9. Let 1 ::; P ::; 00 and 0 < a < 1. Problem (4.122) is well posed in
Lp,,,(E~,p) and the following coercivity inequality holds:
+11 {- L
j
1
ai Ti - A i- 1 (tk) Uk
i=l
+L
.Bi Ti - A i- 1 (tk-l) uk-df IILp'T(E~,p)

1
i=l
::; M[li A(O)(I - L.BiTiAi-l (0))

t=l
uoIIE~,p + a(l ~ a) II r IILp'T(E~,p)],
where M does not depend on f", uo, p, a and T.
196
is well posed in
00 and 0 < Q < 1. The difference problem (4.122)

Lp,T(E~,q)
+11 {- L
j
i
Q(T -
A i-
I (tk) Uk
i=1
+L
,Bi Ti - A-I (tk-d uk-I}l" IILp'T(E~,q)

1
i=1
j
:S M(q)[I1 A(O)(I - L,BiTiA-I (0))

i=1
r, uo, p,
uoIlE~,q + p~ 1 II r IILp'T(E~,),
Q
and To
Note that the difference schemes of the high order of accuracy with respect
to one variable for approximate solutions of the mixed problems (4.50), (4.51) and
(4.52) generated by (4.122) can be constructed. Abstract theorems given above
permit us to establish the stability, almost stability and coercive stability of these
difference schemes.
Chapter 5
of Elliptic Type
In the present chapter we consider the well-posedness of an abstract boundaryvalue problem for differential equations of elliptic type
-v"(t)
+ Av(t) =
f(t)
(O:S t :S T), v(O) = vo, v(T) =
VT
in an arbitrary Banach space E with the positive operator A. The high order of
accuracy two-step difference schemes generated by an exact difference scheme or
by the Taylor decomposition on three points for the numerical solutions of this
problem are presented. The well-posedness of these difference schemes in various
Banach spaces are studied. The stability and coercive stability estimates in Holder
norms for solutions of the high order of accuracy difference schemes of the mixed
type boundary-value problems for elliptic equations are obtained.
5.1
A Boundary-Value Problem. Well-posedness
We consider the boundary-value problem for differential equation
-v"(t) + Av(t) = f(t)
(O:S t :S T), v(O) = vo, v(T) =
VT
(5.1)
in an arbitrary Banach space E with the linear positive operator A.

are satisfied:
i. v(t) is twice continuously differentiable on the segment [0, T]. The derivatives
at the endpoints of the segment are understood as the appropriate unilateral
derivatives.
11.
The element v(t) belongs to D(A) for all t E [0, T], and the function Av(t)
is continuous on the segment [0, T].
iii. v(t) satisfies the equation and boundary conditions (5.1).
Chapter 5. Partial Differential Equations of Elliptic Type
198
A solution of problem (5.1) defined in this manner will from now on be

referred to as a solution of problem (5.1) in the space C(E) = C([O, T], E). Here
C(E) stands for the Banach space of all continuous functions <p(t) defined on [0, T]
with values in E equipped with the norm
If v(t) is a solution in C(E) of problem (5.1), then the data of the problem satisfy
the following conditions:
a) f(t) belongs to C(E) ;
b) va and VT belongs to D(A).
We say that problem (5.1) is well posed in C(E), if the following two conditions are satisfied:
f E C(E) and va, VT E D(A) there exists the unique solution

v(t) = v(t;f(t),VO,VT) in C(E) of problem (5.1). This means that an
additive and homogeneous operator v(t; f(t), va, VT) is defined which
acts from C(E) x D(A) x D(A) to C(E) and gives the solution of (5.1)
in C(E).
1. For any
2. v(t; f(t), va, VT), regarded as an operator from C(E) x D(A) x D(A)
to C(E), is continuous. Here C(E) x D(A) x D(A) is understood as
the normed space of the triplets U(t), va, VT), f(t) E C(E) and va,
VT E D(A), equipped with the norm
IIU(t),VO,VT)IIC(E)XD(A)XD(A) =
Ilfllc(E) + IlvoIID(A) + IlvTIID(A)'
By Banach's theorem in C(E) and these properties one has coercive inequality
II v"II C(E) + II Av ll c (E) :::; Mc[llfllc(E) + IIAvollE + IIAvTIIE],
(5.2)
where Mc (1 :::; Mc < +(0) does not depend on va, VT and f(t). Inequality (5.2)
permits us to investigate the spectral properties of operator A generated by the
well-posedness in C(E) of problem (5.1). For any U E D(A) and oX 2 we will put
'ljJ = Au+Au.
Then, evidently, the function eiVAtu(i =
-v"(t)
with f(t)
A) is a solution in C(E)
+ Av(t) =
of equation
f(t)
= eiVAt'ljJ. Let w(t) be a smooth function defined on [0, T] by the formula

w(t) =
O,
a(t),
{
1,
tJ
tJ
for any t E [0,

U [7[, T],
for any t E [t, u [3;, 7[],
for any t E [t, 3;],
5.1. A Boundary-Value Problem. Well-posedness
199
where a(t) be some smooth function and la( t)1 :S 1. Then, the function v(t) =
w(t)eiVAtu is a solution in C(E) of problem (5.1) with v(o) = v(T) = and
f(t) =
O,
-(a(t)eiVAt)ttu + a(t)eiVAt Au,
{ iVAt l
e
0/,,
for any t E [0, ~J U [7[, T],

for any t E [~, u [31', 7[],
for any t E [I4' 3TJ
4 .
fJ
Therefore from the coercive inequality (5.2) it follows that

A IluliE
+ IIAullE
max IIv"(t)IIE
tE[f, 3[1
+ max IIAv(t)IIE
tE[f, 3JJ
:S max Ilv"(t)IIE + max IIAv(t)IIE

=
= Mc
tElt, 7[1
tE[t, 7[]
Ilv"IIC(E)
+ IIAvIIC(E) :S M c IlfIIC(E)
max{II'l/JIIE,
max
tE[t,flu[3[, 7[1
II - (a(t)eiVAt)ttu + a(t)e iVAt Aulld
Since
we have that
<
11- (a(t)eiVAt)ttu + a(t)e iVAt AullE

la(t)III'l/JIIE + 2Ia'(t)11 ~ (eiVAt)llluIIE + la"(t)llluIIE
< 11'l/JIIE + 2Ml~llullE + M2 11u11E,

where
M1 =
max
tElt,flu[3[,7[]
Therefore
Using the inequality

for c > 0, we obtain that
Let
la'(t)l, M 2 =
max
tE[t,fJu[3[,7[)
la"(t)l.
200
and A ~ fJ + TJ for some TJ > O. Then the inequality
(A + 1) IlullE + IIAullE
::; Mp (TJ)II1/1IIE
holds for some 0 ::; Mp(TJ) < 00. This means that there exists the inverse operator
(M + A) -1, which is defined on the range R( M + A) of operator M + A. We will
prove that R(M + A) = E for sufficiently large A, A> O.
Then it follows that there exists the inverse operator (M + A)-l and the
inequality
holds for all A, A ~ fJ + TJ and 1/1 E E. Now we will prove that R( M + A) = E for all
\ A > O. We consider the boundary-value problem (5.1). Making the substitution
t = OtT for a > 0, we obtain the well-posed in C([O, ~], E) boundary-value problem
_~VII(7) + AV(7) =
a
1(7)(0 ::; 7 ::; '), v(O) = va, v(') =

a
VT.
Let V(7) be the solution of this problem for va = VT = 0 and 1(7) = fa, where
fa is an arbitrary element of E. Since fa and A are not dependent in 7, we have
that v( 7) = v(:L - 7) for all 0 ::; 7 ::; :L. Then, in particular, it follows that
v'(O) = -v'(~). Finally, choosing J>.~ = (2m + 1)1T for some m = 1,2, ... and
integrating by parts, we obtain
T
- rc; v"(7)e i .fiTd7 =
[-v'(~ )ei.fi~ - v'(O)] + iJ>.[v(~ )ei.fi~ - v(O)]

a
Therefore,
A rc;V(7)e i .fiTd7+A rc;V(7) ei .fiTd7= rc;

2
a io
io
io
foei.fiTd7=-~fo
zJ>.
for all \ A > 0 and fa E E. This means that R(M + A) = E for all J>.~ =
(2m + 1)1T and m = 1,2, .... From that it follows that there exists the bounded
inverse (M + A)-l for all A ~ fJ + TJ.
An operator A acting in a Banach space E and having a dense domain D(A)
is called positive if the operator M + A has a bounded-in-E inverse for all \ A ~ 0,
and the estimate
(5.3)
201
is satisfied for some 1 :S M p < +00. Note that we are not able to obtain the
estimate (5.3) for all A, A ~ 0.
Is the positivity of operator A a sufficient condition for the well-posedness
of (5.1) in C(E)? As it turns out, problem (5.1) is not well posed in C(E) for all
such operators. Let us give an example.
Let A be the operator acting in E = C[O, 1] defined by the formula Av(x) =
-v"(x) + v(x), with the domain
D(A)
= {v (x) : v"(x) E C[O, 1], v(O) = v(l) = O}.
Evidently, the operator AI + A has a bounded inverse for any A ~ 0, and formula
[(AI + A)-lV](X)
1 (1 _e- 2v'1+>.)-1 [e-(l-X)v'f0 _ e-(l+x)v'l+>']
= -
2~
1
x j[e-(1-T)v'f0 - e-(l+T)v'f0]v(-r)dr
a
1
2~
j[e-1x-T1v'f0 _ e- 1X +T1 v'f0jv(r)dr
a
holds. From this formula it follows that A is a positive operator in the Banach
space E = C[O, 1]. Now let us show that in this case the boundary-value problem
(5.1) is not well posed in C(E) = C([O, 1], C[O, 1]). Then (5.1) turns into the
boundary-value problem
~:~ - ~:~ + u =
f(t, x),
0< t < 1,
< x < 1,
u(O,x)=u(l,x) =0, O:Sx:Sl,

u(t,O) = u(t, 1) = 0, O:S t:S 1.
However, it is well known that problem (5.4) is not well posed in
C(E) = C([O, 1], C[O, 1]).

The corresponding counterexample can be given by
2
2
u(t,x) = [x (1- x)2 - t (1-
tfJin X2( I-x )2 +t 2( I-t )2'
It means that u(t, x), Ut(t, x), ux(t, x) E C([O, 1], C[O, 1]) and
1
Utt(t,x)=-2In x 2( I-x )2 +t 2( I-t )2+(t,X),

1
uxx (t,x)=2In x 2( I-x )2 +t 2( I-t )2 + <p(t, x)
(5.4)
202
for some continuous functions 7jJ(t, x) and <p(t, x). Therefore, evidently, the boundary-value problem (5.4) is ill posed in C([O, 1]' C[O, 1]). It means that coercive
inequality (5.2) is not true for any solution in C(E) of the boundary-value problem
(5.1).
Note that from the estimate (5.3), evidently, it follows that the operator
>..I + A has bounded inverse for all complex numbers
.\ = C!
+ iT E Gt =
Gt(Mp ) (0 < < 1),
such that
or
and the estimate
II (>..I + A)-lIIE->E S M 1-1(1 + 1.\1)-1

holds for some M 1 E [1, +00) but does not depend on E (0,1). It means that
the spectrum C!(A) of A is a subset of set G-; =
and inside of G-; and on its
boundary 8G-; the resolvent (>..I - A)-l is subject to the bound
-G-:
(5.5)
Let 7jJ(z) be an analytic function on a neighborhood of C!(A), and suppose that 7jJ
satisfies the estimate
(1 + Izl)Q 17jJ(z) 1S M 2
for some
<
a < +00, 1 S M 2 < +00. Then the Cauchy-Riesz integral operator
7jJ(A) =
~
2m
7jJ(z)(zI - A)-ldz
(5.6)
aa;
converges in the operator norm and defines a bounded linear operator 7jJ(A), a
function of A. In particular, the function 7jJ(A) = z-Q defines a bounded operator
A-Q whenever a > 0. The positive powers AQ = (A-Q)-l(a > 0) of the operator
A are defined. The operators AQ (a > 0) are unbounded, and their domains
D(AU) are dense in E if A is unbounded. One has the continuous embeddings
D(AQ) C D(A.B) if (3 < a. The following moment inequality holds:
IIAQullE S M(a, (3)
IIA.Bull~.B lIullk-Q/.B (0 S a S (3 < +oo,u E D(A.B))
(5.7)
for some M(a,(3) E [1,+00) that does not depend on u E D(A.B). The operator
A1/2 has better spectral properties than the positive operator A. Indeed, using
the identity
203
inequality (5.7) for a = ~, (3 = 1 and estimate (5.5) we obtain
~ MM1c- 1(1
+ 1>'1)-1
+M(a,(3)(1 + M1C 1 1>'1 (1 + 1>'1)-1)(M1c- 1(1 + 1>'1)-1)~.

Thus, the operator J>...I - A ~ has a bounded inverse for>. E G; , and the estimate
holds for some M 3 E [1, +(0) and does not depend on E and >.. This means that
B = A~ is a strongly positive operator in a Banach space E (see Chapter 4).
Therefore, the operator -B is a generator of an analytic semigroup exp {-tB}
(t 2 0) with exponentially decreasing norm, when t ~ +00, i.e., the estimates
Ilexp( -tB)IIE---. E , litE exp( -tB)II E---. E ~ M(B)e-c>(B)t(t > 0)
(5.8)
hold for some M(B) E [1, +(0), a(B) E (0, +(0). From that it follows that the
operator I - e- 2TB has a bounded inverse and the following estimate holds:
(5.9)
First of all let us prove a lemma that will be needed in the sequel.
< t < t + T < T and 0
lone has the inequality
TC>
Ilexp(-tB) -exp{-(t+T)B}II E---. E ~ M (t+T)C>'

where M does not depend on a, t, and
T.
Proof. Using the formula
J
t+T
exp( -tB) - exp {-(t
+ T)B}
B exp( -sB)ds
and estimate (5.8), we obtain that

lIexp( -tB) - exp {-(t
+ T)B}II E---. E ~ M rt .
T
The inequality
Ilexp(-tB) -exp{-(t+T)B}II E---. E ~ M 1
(5.10)
204
follows from (5.8) and the triangle inequality. Therefore
II exp( -tB) -
exp {-(t
+ 7)B}II E-. E :S M 1min{l, ~}.

Ilexp( -tB) - exp {-(t
27
+ 7)B}II E -. E :S M 1t + 7'
Interpolating the last inequality and

Ilexp(-tB) -exp{-(t+7)B}II E -. E :S M 1
we obtain (5.10). Lemma 5.1.1 is proved.
Now, we consider the boundary-value problem (5.1). It can be obviously
rewritten as the equivalent boundary-value problem for the system of first-order
linear differential equations
v' (:) + BV(I) ~ z(I)(O ,; 1,; T), viOl ~ Vo, v(T) ~ Vr,
{ -z (t) + Bz(t) =
j(t).
V(t)
{ z(t)
=
=
e-tBvo + f~ e-Ct-s)Bz(s)ds,
e-CT-t)Bz(T) + ft e-(s-t)B j(s)ds.
From these formulas and the condition v(T) = VT it follows that
v(T) = e-TBvo
+ iT e- 2(T-s)Bdsz(T)
+ iT iT e-(T+y-2s)B j(y)dyds.
Since faT e- 2(T-s)Bds = (2B)-1(I - e- 2TB ), we have that
z(T) = (I - e- 2TB )-12B[VT - e-TBvo

-iT iT e-(T+y-2s)B j(y)dyds]
= (I - e- 2TB )-12B[VT - e-TBvo
-(2B)-1 iT (e-(T-y)B - e-(T+y)B)j(y)dy].
So,
205
z(t) = e-(T-t)B(I - e- 2TB )-12B[Vy - e-TBvo

-(2B)-1 iT (e-(T-y)B - e-(T+Y)B)f(y)dy]
+ iT e-(s-t)B f(s)ds
and
it
IT e-(t+s-2y)B f(s)dyds.
it
= e-tBvo
e-(t+s-2y)B f(s)dsdy
+ iT
it
e-(t+s-2y)B f(s)dyds
+ (e-(T-t)B _ e-(T+t)B)(I _ e- 2TB )-1[VT _ e-TBvo

-(2B)-1
+(2B)-1
it
(e-CT-s)B - e-CT+s)B)f(s)ds]
(e-(t-s)B - e-Ct+S)B)f(s)ds
+(2B)-1 iT (e-(s-t)B - e-(t+s)B)f(s)ds

= (I - e-2TB)-1{(e-tB - e-(2T-t)B)vo
+ (e-(T-t)B -
e-(TH)B)VT
_(e-CT-t)B _ e-(TH)B)(2B)-1 iT (e-(T-s)B _ e-(T+s)B)f(s)ds}

+(2B)-1 iT (e-lt-sIB - e-(t+s)B)f(s)ds.
Thus,
206
_(e-(T-t)B _ e-(T+t)B)(2B)-1 iT (e-(T-s)B _ e-(T+S)B)f(s)ds}

+(2B)-1 iT (e-lt-sIB - e-(t+S)B)f(s)ds.
It is easy to show that vet) defined on [0, T] by formula (5.11) is a unique solution
in G(E) of problem (5.1) if, for example, vo, VT E D(A 2 ) and Af(t) E G(E) or
f'(t) E G(E). Sufficient conditions for the well-posedness of the boundary-value
problem (5.1) can be established if one considers this problem in certain spaces
F(E) of smooth E-valued functions on [0, T].
A function vet) is said to be a solution of problem (5.1) in F(E) if it is a
solution of this problem in G(E) and the functions v"(t) and Av(t) belong to
F(E).
As in the case of the space G(E), we say that problem (5.1) is well posed in
F(E), if the following two conditions are satisfied:
1. For any f E F(E) and vo, VT E D(A) there exists a unique solution vet) =
vet; f(t), Vo, VT) in F(E) of problem (5.1). This means that an additive and
homogeneous operator vet; f(t), vo, Vy) is defined which acts from F(E) x
D(A) x D(A) to F(E) and gives the solution of (5.1) in F(E).
2. vet; f(t), vo, VT), regarded as an operator from F(E) x D(A) x D(A) to F(E),
is continuous. Here F(E) x D(A) x D(A) is understood as the normed space
of the triplets (J(t), Vo, VT), f(t) E F(E) and vo, VT E D(A), equipped with
the norm
11(J(t),VO,VT)IIF(E)XD(A)XD(A) = IlfIIF(E) + IlvoIID(A) + IlvTIID(A).

First, we set F(E) equal to G~Ta(E) = G~t([O,T],E),O < a < 1, the Banach
space obtained by completion of the set of smooth E-valued functions Y'(t) on
[0, T] in the norm
II Y' lIe a.a(E)= 1IY'lIe(E) +

OT
sup
o:::;t<t+T::oT
1IY'(t + r) - Y'(t)IIE
r
(T - t) (t
+ r) .
As in the case of the space G(E), from the well-posedness of the boundary-value
problem (5.1) one derives the coercivity inequality
Ilv"lIe~+Q(E)
+ IIAvlle~+a(E) :S Mc(a)[lIflle~+a(E) + IIAvollE + IIAvTllel (5.12)
where Me (a)(l :S Mc(a) < +00) does not depend on Vo, Vy and f(t).
Theorem 5.1.2. Suppose Vo, VT E D(A), f(t) E G~t(E)(O < a < 1). Then the
boundary-value problem (5.1) is well posed in the Holder space G~Ta(E), if A is
207
the positive operator in the Banach space E. For the solution v(t) in C~TO!(E) of
the boundary-value problem the coercive inequality
Ilv"llc;+"CE) + IIAvllc;+"CE) :S
a(l~ a)
II f IIc;+"CE) +M[li
Avo/IE
+ I/AvTlleJ
holds, where M does not depend on a, Vo, VT and f(t).

Proof. If v(t) is a solution in C~t(E) of problem (5.1), then it is a solution in
C(E) of this problem. Hence, by (5.11), we have the representation
v(t) = (I - e-2TB)-1{(e-tB - e- C2T - t )B)vo
+ (e-CT-t)B - e-CTH)B)VT
_(e-CT-t)B _ e-CTH)B)(2B)-1 iT (e-CT-s)B _ e-CT+s)B)f(s)ds}

+(2B)-1 iT (e-lt-sIB - e-Ct+s)B)f(s)ds.
To prove the theorem we must show that the functions v"(t) and Av(t) belong to
C~t(E), and finally obtain estimates for the norms of these functions. Here we
address only the last task. Using formula (5.11), we obtain
Av(t) = (I - e-2TB)-1{(e-tB - e-C2T-t)B)Avo
(5.13)
+(e-CT-t)B _ e-CTH)B)AvT}
_(e-CT-t)B _
T
r (e-CT-s)B _ e-CT+s)B)dsf(t)}
2 Jo
e-CT+t)B)~B
+~B
2
_(e-(T-t)B _
((e-lt-sIB - e-Ct+s)B)dsf(t)
Jo
e-(TH)B)~B
2
rT(e-CT-S)B _ e-CT+s)B)(f(s) - f(t))ds}
Jo
+~B
((e-lt-sIB - e-Ct+s)B)(f(s) - f(t))ds

2 Jo
= (I _e-2TB)-1(e-tB -e-(2T-t)B)Avo+(I _e- 2TB )-1(e-(T-t)B -e-CTH)B)AvT
+(I - e- 2TB )-1(I - e-tB)(I - e-(T-t)B)(I - e-TB)f(t)
208
Let us estimate Jk(t) for any k = 1,2,3,4,5 separately. We start with J1 (t). Using
estimates (5.8)and (5.9) we show that, for 0 :::; t :::; T,
Further, applying (5.9) and (5.10) we show that, for 0 < t < t
+lle-(2T-t)B - e-(2T-t-r)BII
TO.
-+
]IIAv II < M[
E E
0 E(t + T)o.
+ T < T,
TO.
+ (T _ t)o. ]IIAv0 II E.
Thus, we have proved that

(5.14)
In a similar manner we can show that

IlhIIC~T"(E) :::; MIIAvTIIE.
(5.15)
Now let us estimate h(t). Using estimates (5.8) and (5.9), we obtain that
h(t)
liE:::; 11(1 -
e- 2TB )-11IE-+EIII - e-tBIIE-+E
xiiI - e-(T-t)BIIE-+EIII - e-BIIE-+Ellf(t)IIE :::; Mllfllcg+"(E)

for 0 :::; t :::; T. From this inequality it follows that
max
O::=;t::=;T
h(t)
liE:::; M I f IIc"'''(E) .
OT
Since
+(1 - e-2TB)-1{(e-tB - e-(t+r)B)(I _ e-(T-t)B)

+(1 - e-tB)(e-(T-t)B _ e-(T-t-r)B)}(I - e-TB)f(t)
it follows that
h(t + T) - h(t)
liE:::; 11(1 -
e- 2TB )-11IE-+E
xiiI - e-tBlls-+EiII - e-(T-t)Bl/s-+sI/I - e-TBlls-+Ellf(t + T) - f(t)/IE
+11(1 - e-2TB)-11IE-+E{lle-tB - e-(t+r)BIIE-+EIII - e-(T-t)BIIE-+E

+111 - e-tBIIE-+Elle-(T-t)B - e-(T-t-r)BIIE-+e}III - e- TB IIE-+Ellf(t)IIE
209
for 0 < t < t + T < T. Using estimates (5.8), (5.9) and (5.10), we obtain
h(t + T) - h(t)
+ T < T.
for 0 < t < t
liE
~ M[ (t + T)a + (T _ t)a 111 1 IIc~+a(E)
I h
Ilc~+a(E)~ M I
I 1 Ilc~+a(E) .
(5.16)
Now let us estimate J4 (t). We have to consider two cases: 0 ~ t ~ and

T. In the first case, using estimates (5.8) and (5.9), we obtain that
~t ~
J4 (t)
IIE~ II(I - e-2TB)-IIIE-+E~
IIBe-(t-s)B(I - e-(2T-2t)B)
IIE-+E
xiiI -
e-
2sB
IIE-+E 11/(s) -
I(t)
M
~ -;;(T - t)-a
liE ds
~M
J
t
(t _ S)l-:tSa(T _ s)a
M2 a
1 Ilc~+a(E)~ o:Ta
I 1 Ilc~+a(E)
I 1 Ilc~+a(E) .
In the second case, using estimates (5.8), (5.9) and (5.10), we obtain that
J
t
J4 (t)
IIE~ II(I - e-2TB)-IIIE-+E~
IIBe-(t-s)B(I - e-(2T-2t)B)
IIE-+E
xIII -
e-
2sB
IIE-+E 11/(s) -
liE ds
I(t)
< MJ
Since
t
J
o
2(T - t)ds
(t - s)l-a(2T - t - s)ta(T - s)a
1
= ta(T -
~ ta(T - t)a
J
t
2(T - t)ds
(t - s)l-a(2T - t - s)
t)a
a,a
I 1 IlcoT (E)
2(T - t)ds
(t - s)l-a(2T - t - s)ta(T _ S)a
2(T - t)dz
1
zl-a(2T - 2t + z) ~ ta(T - t)a(2T - 2t)l-a
00
2(T - t)dp
pl-a(1 + p)
210
it follows that
Combining the estimates for
IIE:S
J4 (t)
a(l _ a)
J4 (t)
II f
IIc~T"(E)
liE, we obtain
M
O~t~T I J4(t) IIE:S a(l _ a) I f I .
(5.17)
Now let us estimate the difference J4 (t + 7) - J4 (t) for 0 < t < t + 7 < T. We will
consider separately the cases t :S 7, T - t :S 27 and t > 7, T - t > 27. If t :S 7,
then (5.17) yields
:S
M21+
a(l _ a)
Q
If T - t
:S
I f Ilc~+"(E) 7Q(t + 7)-Q
:S 27, then (5.17)
7
2
II
Q
1+ )
a1-a
= a(l
~ a) I f IIC~T"(E)
Q
7 (t + 7)-Q.
yields
J4 (t + 7) - J4 (t)
I f IIC"'''(E)
7
~
IIE:s1I J4 (t + 7) liE + I
(MI )
a1-a
(T - t)-Q =
J4 (t)
liE
I f IIC"'''(E)
7
~
Now let t > 7 and T - t > 27. Let us represent the difference J4 (t
the sum of the following integrals:
J
t+r
e-(t+r-s)B(I - e- 2sB )(f(s) - f(t
t-r
(T - t)-Q.
+ 7) -
J4 (t) as
+ 7))ds
J
J
t
-(1 - e- 2TB )-1 ~ (1 - e-(2T-2t)B)
e-(t-s)B(1 - e- 2sB )(f(s) - f(t))ds
t-r
t-r
+(1 _e- 2TB )-1 ~ (I _e-(2T-2t-2r)B)
e-(t+r-s)B(I -e- 2sB )(f(t) - f(t+7))ds
J
t-r
+(I - e- 2TB )-1
~ [(e-(2T-2t)B - e-(2T-2t-2r)B)e-(t-s)B
o
+(1 - e-(2T-2t)B)(e-(t+r-s)B - e-(t-s)B)](1 - e- 2sB )(f(s) - f(t))ds
= PI
+ P2 + P3 + P4
211
Now let us estimate Pl' Using estimates (5.8) and (5.9), we obtain that
II PI IIE~ 11(1 - e-2TB)-IIIE-+E~
t+r
IIBe-(t+r-s)B(1 - e-(2T-2t-2r)B)
t-r
x III -
e-
2sB
IIE-+E Ilf(s) -
t+r
~M
t-r
f(t
+ r) liE ds
ds
(t + r _ s)1-"(t + r)"(T _ s)"
I f Ilc~+a(E)
In exactly the same manner one establishes the estimate
Since
J
t-r
e-(t+r-s)B(1 - e- 2sB )ds(f(t) - f(t
+ r))
= (I - e- 2TB )-1 ~(I - e-(2T-2t-2r)B)

2
x[e- 2rB - 2e-(t+r)B
+ e- 2tB ](f(t) -
f(t
+ r)),
it follows that
II P3 IIE~ 11(1 - e-2TB)-IIIE-+E~III x Ile- 2rB
2e-(t+r)B
e-(2T-2t-2r)BIIE-+E
+ e- 2tB IIE-+Ellf(t) -
f(t
+ r)IIE.
IIE-+E
212
Now let us estimate P4. Using estimates (5.8), (5.9) and (5.10), we obtain
J{II
t-T
II P3 I1E:S 11(1 - e-2TB)-11IE--->E~
o
-(2T-2t-2T)B) -(t-s)B
-e
e
+ I B(1 -
B(e-(2T-2t)B
II E--->E
e-(2T-2t)B)(e-(t+T-S)B - e-(t-s)B)
xIII -
e-
2sB
IIE_E
f(t) - f(s)
IIE-E }
liE ds
t-T
:s M
Since for t
> 7 and T -
<
(t _
S)2-:~~(T _ s)a
If
IIc~+Q(E) .
> 27, we have the bound
- (1 - a)ta(T - t + 7)a
-l+a
a a
<
- (1 - a)(t + 7)a(T - t)a
Therefore, we conclude that
Thus, we have shown that for any 0
< t < t + 7 < T the following inequality holds:
Estimates (5.17) and (5.18) give
M
J 4 IIc~+Q(E):S a(1 _ a)
II f
Ilc~+Q(E)
(5.19)
(5.20)
In exactly the same manner one establishes the inequality
M
J5 Ilc~+Q(E):S a(1 _ a)
I f
Ilc~+Q(E)
Finally, applying estimates (5.14), (5.15), (5.16), (5.19) and (5.20) we obtain the
estimate
II
Av
Ilcg+Q(E):S
a(1 _ a)
If
Ilcg~t(E)
+M[li AvoilE + IIAvTIIE]'
213
By the triangle inequality, this last estimate and equation (5.1) yields
II v"(t) IIcQ,Q(E)
~ a (M
+M[II
OT
1 - a ) II f IlcQ,Q(E)
OT
AvoilE + IIAVrIIE]'

We denote by C~t(E) = C~:y::~([o, T]' E),
< a < 1, the Banach space
obtained by completion of the set of smooth E-valued functions <.p(t) on [0, T] in
the norm
II <.pllcQ,Q(E)= 11<.pIIC(E) +
OT
sup
II<.p(t
+ 7) -
OSt<t+TST
Ct
<.p(t) liE
Ct
Ct
mm{(T-t) ,(t+7)}.
It is easy to show that
II <.p Ilc~+Q(E)~
Ct
II <.p Ilc~+Q(E) .
By the way, the norms of the spaces C~TCt(E) and C~t;(E) are not equivalent.
Nevertheless let us give, without proof, the following result.
Theorem 5.1.3. Suppose Vo, VT E D(A), f(t) E C~TCt(E)(O < a < 1). Then the
boundary-value problem (5.1) is well posed in Holder space C~t(E), if A is the
positive operator in the Banach space E. For the solution v(t) in C~t(E) of the
boundary-value problem the coercive inequality
Ilv"llc~+Q(E) + IIAvllc~+Q(E) ::; a(l~ a) II f Ilc~+Q(E) +M[li AvoilE + IIAvTIIE]

holds, where M does not depend on a, Vo, VT and f(t) .
Ilv"IIF(E) + IIAvIIF(E)
::;
MC[llfIIF(E) +
IIAvoliE
+ IIAvTIIE]
fails if we set F(E) equal to CCt(E), (0 < a < 1), the Banach space obtained by
completion of the set of all smooth E-valued functions <.p(t) on [0, T] in the norm
II <.p IlcQ(E)=
max
OStST
11<.p(t)IIE +
sup
OSt<t+TST
II <p(t + 7) Ct-
<p(t)
liE
Nevertheless, we have the following result.

Theorem 5.1.4. Suppose v~, v~ E E Ct , f(t) E CCt(E)(O < a < 1). Then the
boundary-value problem (5.1) is well posed in the Holder space CCt(E), if A is the
positive operator in the Banach space E. For the solution v(t) in CCt(E) of the
boundary-value problem the coercive inequality
Ilv"lIcQ(E) + IIAvllcQ(E) + Ilv"llc(EQ) ~

+ Ma [II v~IIEQ + Ilv~IIEal
a(l~ a) II f IlcQ(E)
214
holds, where M does not depend on 0:, va, Vr and f(t). Here, the Banach space
Eo: = Eo:(E, B) (0 < 0: < 1) consists of those vEE for which the norm
I v IIE,,= SUpZl-o:
z>o
II Bexp{-zB}v liE +
II V
liE
is finite.
Proof. By formula (5.11), we have
v"(t) = {I - e-2TB)-1(e-tB -
e-(2T-t)B)v~
+{I - e- 2TB )-1(e-(T-t)B -
e-(T+t)B)v~
+(1 - e- 2TB )-1(e-(t+T)B - e-(2T-t)B)f(0)

+(I - e- 2TB )-1(e-(2T-t)B - e-(T+t)B)f(T)
-(I - e-2TB)-le-tB{I - e-TB)(f(t) - f(O))
-(I - e- 2TB )-le-(T-t)B{I - e-TB)(f(t) - f(T))
+{I - e- 2TB )-1 B (I - e-(2T-2t)B)
rt e-(t-s)B{I -
e- 2sB )(f(s) - f(t))ds
Jo
+{I - e- 2TB )-1 B (I - e- 2tB ) iT e-(s-t)B{I - e-(2T-2s)B)(f(s) - f(t))ds
= LGk(t).
k=l
To prove the theorem it suffices to establish the estimates for v"(t) in CO:(E) and
in C(Eo:). Let us estimate Gk(t) for any k = 1,2,3,4,5,6,7,8 separately. We start
with G 1 (t). Using estimates (5.8), (5.9) and the definition of the spaces Eo: we show
that, for 0 S t S T,
2T-t
2TB
IIG 1(t)IIE < II{I - e)-11IE->E
IIBexp(-sB)v~IIEds
f
t
2T-t
J s~~o:llv~IIE"
< M
~lllv~IIE'"
Further, applying (5.8), (5.9) and the definition of the spaces Eo: we show that,
for 0 < t < t + T < T,
t+r
IIG1(t + T) - G1(t)IIE S II{I - e-
2TB
)-11IE->e[
liB exp( -sB)v~IIEds
2T-t
2T-t-r
IIBexp(-sB)v~IIEds]
215

t+T
<
-
2T-t
M[J ~ + J
~]llv"ll
8 1- 0
0 E", <
-
81-0
oM1
11v"II
0 E", .
2T-t-T
Using estimates (5.8), (5.9) and the definition of the spaces Eo we show that, for
o :=; t :=; T,
Now let us estimate G3 (t). Using estimates (5.8) and (5.9) we show that, for
o:=; t :=; T,
II G3(t)IIE :=; 11(1 -
e- 2TB )-II1E-+Elle-(t+T)B - e-(2T-t)BIIE-+Ellf(0)IIE
:=; Mllfllc"'(E)'
Further, applying (5.8), (5.9) and (5.10) we show that, for 0 < t < t
II G3 (t + T)
G3 (t) liE:=;
x{lle-(t+T)B _ e-(t+T+T)BIIE-+E
TO
11(/ -
e-
+ Ile-(2T-t)B
TO
2TB
+ T < T,
)-IIIE-+E
- e-(2T-t-T)BIIE-+E}llf(0)IIE
:=; M[(t + T + T)O + (2T _ t)o] II f IIc"'(E):=; M 1To II f Ilc"'(E) .

216
By the definition of the spaces Ea: and using estimates (5.8), (5.9), we obtain that
II AI-a: Be->.BG3 (t) liE::; AI-a:II(I - e-ZTB)-IIIE--+E

x IIBe->.B(e-(t+T)B - e-(ZT-t)B)f(O)IIE
AI-a:
::; M[ A + t + T
1
AI-a:
+ A - t + 2T 111f (0)IIE
1
::; M[(t + T)a: + (2T _ t)a:lIlf(O)IIE ::; MI II f IIc<>(E)

for all A, A > 0 and for 0 ::; t ::; T. So, we have established that
Now let us estimate G5 (t). Using estimates (5.8) and (5.9) we obtain that
II G5 (t) liE::; 11(1 - e-ZTB)-IIlE--+EIII - e-TBIIE--+E

xlle-tBIIE--+Ellf(t) - f(O)IIE ::; Mllfllc<>(E) ::; Mta:llfllc<>(E)
for 0 ::; t ::; T. From this inequality it follows that
max II G5 (t) liE::; MI II f IIc<>(E) .
O::;t::;T
Since
G5 (t + r) - G5 (t)
(I - e- 2TB )-le-(t+T)B(1 - e-TB)(f(t + r) - f(t))
-(1 - e-ZTB)-I{(e-tB - e-(t+T)B)(I - e-TB)(f(t) - f(O)),

it follows that
II G5 (t + r) - G5 (t) liE::; 11(1 - e-ZTB)-IIIE--+E

xIle-(t+T)BIIE--+EI11 - e-TBIIE--+Ellf(t + r) - f(t)IIE
+11(1 - e-ZTB)-IIIE--+Elle-tB - e-(t+T)BIIE--+EI11 - e- TB IIE--+EIIf(t) - f(O)IIE
::; M(ra: + :: ta:) II f IIc<>(E)= MIra: II f IIc<>(E)
for 0 < t < t
+ r < T.
217

IIG61ICO(E) :S Ml II
1 Ilco(E) .
By the definition of the spaces Eo: and using estimates (5.8), (5.9), we obtain that
II Al-o: Be->'BGs(t) IIE:S Al-O:II(I - e- 2TB )-lIIE-+E
xiiI - e-TBIIE-+EIIBe-(>.+t)B(f(t) - I(O))IIE
Al-O:tO:
:S M A + till IlcO(E):S Mill IIcO(E)
for all A, A > 0 and for 0 :S t :S T. So, we have established that

II G6I1C(Eo) :S Ml II
1 IIco(E) .
Now let us estimate G7 (t). Using estimates (5.8) and (5.9) we obtain that
J
t
II G7 (t) IIE:S 11(1 _e-2TB)-lIlE--+E~
IIBe-(t-s)B(1 _e-(2T-2t)B) IIE--+E (5.21)
x III - e- 2sB IIE-+E II/(s) - I(t) liE ds
J
t
(t _ ~~l-O: II 1 IIco(E)= ~ to: II IlIco(E)

o
T. From this inequality it follows that
:S M
for 0 ~ t
max II G7 (t) IIE~ - II 1 IIco(E) .

O:5;t:5;T
a
(5.22)
Now let us estimate the difference G7 (t + r) - G7 (t) for 0 < t < t + r < T. We
will consider separately the cases t ~ rand t > r. If t ~ r, then (5.21) yields
II G7 (t
+ r) -
:S M (to:
a
G7 (t) IIE~II G7 (t
+ r) liE + II
G7 (t) liE
+ (t + rt) II 1 IICO(E):S MlrO: II 1 IICO(E) .

a
Now let t > r. Let us represent the difference G 7 (t + r) - G7 (t) as the sum of the
G7 (t + r) - G7 (t) = (I - e- 2TB )-l B (I _ e-(2T-2t-2T)B)

2
t+T
x
e-(t+T-s)B(1 - e- 2sB )(f(s) - I(t + r))ds
t-T
218
J
t
-(1 - C 2TB )-1 ~ (1 - e-(2T-2t)B)
e-(t-s)B(1 - e- 2sB )(f(S) - f(t))ds
t-r
J
t-r
e-(t+r-s)B(1 - e- 2sB )(f(t) - f(t
J
t-r
+(1 - e- 2TB )-1
+ r))ds
~ [(e-(2T-2t)B - e-(2T-2t-2r)B)e-(t-s)B
o
+(1 - e-(2T-2t)B)(e-(t+r-s)B - e-(t-s)B)](I - e- 2sB )(f(S) - f(t))ds
= PI + P2 + P3 + P4 .
Now let us estimate Pl. Using estimates (5.8) and (5.9) we obtain that
II
PI
liE::; 11(1 - e-2TB)-IIIE-+E~
J
t+r
IIBe-(t+r-s)B(1 - e-(2T-2t-2r)B)
t-r
xiiI -
J(+
e- 2sB
IIE-+E
IIf(s) - f(t
+ r) liE ds
t+r
::; M
dS)l
r - s -0
II f IlcQ(E)::; Ma (2r)0 II f IlcQ(E) .
t-r
Since
t-r
X
e-(t+r-s)B(1 - e- 2sB )ds(f(t) - f(t
+ r))
o
= (1 - e- 2TB )-1 ~(1 - e-(2T-2t-2r)B)
2
x [e- 2rB _ 2e-(t+r)B + e- 2tB ](f(t) - f(t
+ r)),
it follows that
II P3
11E::;
11(1 - e-2TB)-11IE-+E~III -
xlle- 2rB -
e-(2T-2t-2r)BIIE-+E
2e-(t+r)B + e- 2tB IIE-+Ellf(t) - f(t
+ r)IIE.
IIE-+E
219
Using estimates (5.8) and (5.9) we obtain
Now let us estimate P4. Using estimates (5.8), (5.9) and (5.10), we obtain
J{II
t-r
B(e-(2T-2t)B - e-(2T-2t-2r)B)e-(t-s)B
IIE--+E
+ II B(I -
e-(2T-2t)B)(e-(t+r-s)B - e-(t-s)B)
xiiI -
e- 2sB IIE--+E
II
f(t) - f(s)
IIE--+E }
liE ds
J ::;2-0 II f IIC<>(E) .
t-r
::; M
(t
Since for t > 7, we have the bound

t-r
J
o
t-r
tds
7
..,--------,-,;:-<
(t-s)2-0 - t o (T-t+7)0
ds
7
-1+0
7
< -7
= (t-S)2-0 -1-0:
1-0:
Therefore, we conclude that
Thus, we have shown that for any 0 < t < t+7 < T the following inequality holds:
II Gs Ilc<>(E)::;
M
0:(1 _ 0:)
II f IIc<>(E) .
220
II .xl-o: Be- AB G7 (t) IIE~ .xl-O:I/(I - e- 2TB )-11IE-.E
J
t
IIB 2e-(A+t-s)B(I - e-(2T-2t)B) IIE-.E
o
xiiI - et
~M
for all .x,.x
>0
.xl-O:ds
J
o
2sB IIE-.E IIf(s) - f(t) liE ds
(.x + t _
S)2-0:
II f
IIco(E)~ 1 _ a II f IIco(E)
and for 0 ~ t ~ T. So, we have established that
M
II G 7I1c(Eo) ~ 1- a II
Ilco(E) .
IIco(E) .

M
IIGsllc(Eo) ~ 1 _ a II
Combining the estimates for Gk(t), k
IIv"IICO(E):S
M
(1
-a
1,2,3,4,5,6,7,8 we obtain the estimates
) II f IIco(E)
+M
[II
a
v~IIEo + Ilv~IIEal,
M
(1
) II f IIco(E) + M [II v~IIEo + Ilv~IIEo]'

a -a
a
The estimate for Av(t) in the norm of CO:(E) follows from the triangle inequality.
Second, one can enrich the family of spaces of smooth functions in which the
boundary-value problem (5.1) is well posed by introducing the spaces CgP(E),
(0 :s 'Y :s f3,0 < f3 < 1), obtained by completion of the set of all smooth E-valued
functions <p(t) on [0, T] with respect to the norm
IIv"lIc(Eo):S
II <p
max II <p (t )11 E +

sup
II C13'Y(E)=
OT
O~t~T
O~t<t+T~T
(t + r)r(T - t)' II <p(t + r) - <p(t) liE

(3
r
Theorem 5.1.5. Let A be the positive operator in a Banach space E and f(t) E
CgP(E) (0 ~ 'Y ~ f3,0 < f3 < 1). Then for the solution v(t) in CgP(E) of the
boundary-value problem (5.1) the coercive inequalities
II v" Ilc(EI3-"'f) M[II v~ IIEI3-"'f + II v~ IIEI3-"'f +f3- 1(1 - (3)-1 II f II cg+"'fcE)],
:s
II v" Ilcg+"'f(E)
+ II Av II c g+"'f(E)
:s M[lv~lg,y + Iv~lg,r + f3-1(1 - m-
II f Ilcg+"'fcE)]
221
hold, where M does not depend on (3,'Y,vg,v~ and f(t). Here, Iwlg,"! denotes that
the norm of the Banach space Eg,"! consists of those wEE for which the norm
is finite.
number offunction spaces in which problem (5.1) is well posed. In particular, it is
important that problem (5.1) is well posed in the Holder space without a weight
b = 0). Recall that the positivity of A is a necessary condition for well-posedness
of problem (5.1) in C(E). However, problem (5.1) is not well posed in C(E) for
all positive operators. It turns out that a Banach space E can be restricted to a
Banach space E' in such a manner that the restricted problem (5.1) in E' will be
well posed in C(E'). The role of E' will be played here by the fractional spaces
Eo: = Eo:(E, B)(O < a < 1).
Theorem 5.1.6. Let A be the positive operator in a Banach space E and f(t) E
C(Eo:), (0 < a < 1). Then for the solution v(t) in C(Eo:) of the boundary-value
problem (5.1) the coercive inequality
I v"
Ile(E,,) + II Av Ile(E,,)
:s: M[II Avo liE" + I AVT liE" +a- 1(1 - a)-l I f Ile(E,,)]
holds, where M does not depend on a, vg,
v~
(5.24)
and f(t).

Av(t) = (1 - e-2TB)-1{(e-tB - e-(2T-t)B)Avo
+ (e-(T-t)B - e-(T+t)B)Avr}
T
B e-(T-s)B f(s)ds
r
Jo 2
T
+(1 - e- 2TB )-1(e-(2T-t)B _ e- tB ) r B e- sB f(s)ds
Jo 2
-(1 - e- 2TB )-1(e-(T-t)B _ e-(TH)B)
Jo
B e-(t-s)B f(s)ds
2
rT B e-(s-t)B f(s)ds
Jt
h(t).
k=l
Let us estimate h(t) for any k = 1,2,3, ... ,6 separately. We start with h (t).
Using estimates (5.8), (5.9) and the definition of the spaces Eo: we show that, for
O:S: t :s: T,
IIh(t)IIE" :s: 11(1 - e-2TB)-1IlE-->Elle-tB - e-(2T-t)BIIE-->E II Avo liE"

::; MIiAvoIlE".
222
IlhIIC(Ea )
::;
MIIAvoIIEa
a ) ::;
MIIAvTIIEa
IllzIIC(E
By the definition of the spaces E a and using estimates (5.8), (5.9), we obtain that
II
).,1-0 Be->.B h(t)
liE::; 11(1 x
<M
-
J
T
T ).,1-0
--IIB 2e-(T-s+>.)B l(s)IIEds
a
ds
I 1 IIC(E
<M
s+ ).,)(T - s)1-a -
).,1-
(T -
e- 2TB )-IIIE-->Elle-(T-t)B - e-(T+t)BIIE-->E
a )
J
~
I 1 IIC(E < M 1
+ z)zl-a - a(l _ a) I IIIC(E
dz
a )
(1
o
0
for all )..,)., > 0 and for 0 ::; t ::; T. So, we have established that
Ilhllc(E
a ) ::;
M1
a(l _ a)
a )
I 1 IIc(E
a ) .
Next, using estimates (5.8), we obtain
J
t
II ).,1-a Be->'B h liE::; ~
II B 2e-(s+>.)B I(s) liE ds
<M
).,1-0
dz
(1
II 1 IIC(E < M 1 II 1 II
+ z)zl-a - a(l _ a)
C(E
a )
a )
o
for all ).., )., > 0 and for 0 ::; t ::; T. So, we have established that
IIhllc(E
a ) ::;
M1
a(l _ a)
III IIc(E
a )
Ilhllc(E
a ) ::;
M1
a(l _ a)
III Ilc(E
a ) .
Combining the estimates for h, lz, h, [4, h, h we obtain estimate (5.24). Theorem 5.1.4 is proved.
Third, let us study now the boundary-value problem (5.1) in the spaces
cgp(Ea -f3), (0:::; "y::; (3:::; a,O < a < 1). To these there correspond the spaces of
223
traces E~~fJ' which consist of elements wEE for which the norm
fJ "
Iwl o-f3
max
O~z~T
Ile-zBwIIEa_1J
T- f3 (Z + T)'(l - z)'ll(e-(z+T)B - e-ZB)wIIEa_1J
sup
O~z<z+T~T
is finite. Let us give, without proof, the following results.

Theorem 5.1.7. Let A be the positive operator in a Banach space E, v~ E E~~fJ'
v!j. E E~~f3 and f(t) E Cgp(Eo-fJ) (0 ::; 'Y ::; f3 ::; a, 0 < a < 1). Then for
the solution v(t) in Cgp(Eo-f3) of the boundary-value problem (5.1) the coercive
inequality
II v" IlcgT"Y(Ea_lJ) + II Av IlcgT"Y(Ea_lJ) + II v" Ilc(E~~IJ)
::; M[lv~I~~fJ
+ Iv!j.I~~fJ + a- I (1- a)-l II f IlcgT"Y(Ea_IJ)]
holds, where M does not depend on a, f3, 'Y,
v~,
v!j. and f (t).
Theorem 5.1.8. Let A be the positive operator in a Banach space E and v~ E

E o -" v!j. E E o - , and f(t) E Cgy'(Eo-,) (0::; 'Y ::; f3 ::; a, 0 < a < 1). Then for
the solution v(t) in Cgp(Eo-fJ) of the boundary-value problem (5.1) the coercive
inequality
II v" II cgT"Y(Ea_lJ) + II Av II cgP(Ea_lJ) + II v" IIc(E a-"y)

< Ma- l (l - a)-lUI v~ IIEa-"y + II v!j. IIEa-"y + II f II cgT"Y(Ea_IJ)]
holds, where M does not depend on a, f3, 'Y,
v~,
v!j. and f (t).
Note that the spaces of smooth functions Cgp(Eo- f3 ), in which coercive

solvability has been established, depend on the parameters a, f3 and 'Y. However,
the constants in the coercive inequalities depend only on a. Hence, we can choose
the parameters f3 and 'Y freely, which increases the number of function spaces
in which problem (5.1) is well posed. In particular, Theorem 5.1.6 implies the
well-posedness Theorem 5.1.4 in C(Eo ) established in the above.
Fourth, let us study now the boundary-value problem (5.1) in the spaces
Lp(E) = Lp([O, T], E) (1 ::; p < 00) of all strongly measurable E-valued functions
v(t) on [0, T] for which the norm
T
II
IILp(E)= (/ II v(t) Iii dt)~

o
is finite.
Recall that a function v(t) is said to be absolutely continuous if it has a
derivative v'(t) for almost every t such that v'(t) E LI(E), and if the Newton-
224
Leibniz formula
J
t
v(t) - V(T) =
v'(s)ds
holds for all t, T E [0, T]. Here the integral is understood in the sense of Bochner.
A function v(t) is said to be a solution of problem (5.1) in Lp(E) if it is
absolutely continuous, the functions v"(t) and Av(t) belong to Lp(E), equation
(5.1) is satisfied for almost every t, and v(o) = va, v(T) = VT. From this definition
it follows that a necessary condition for the solvability of problem (5.1) in Lp(E)
is that f(t) E Lp(E). It will be shown that in certain cases this condition is also
sufficient for the solvability of problem (5.1). As concerns the boundary elements,
in contrast to the situation considered earlier, from the solvability of problem (5.1)
in Lp(E) it follows only that vo, VT E E. In the case of an unbounded operator A
this does not allow us to prove the solvability of problem (5.1).
From the unique solvability of (5.1) it follows that the operator v(t; f(t), VA,
VT) is bounded in Lp(E) and one has coercive inequality
II v"IIL p(E)
+ IIAv(t)IILp(E)
~ Me[llfIILp(E)
+ IIAvollE + IIAvTllEl
where Me (1 ~ Me < +(0) does not depend on VO,VT and f(t). From that we
can obtain the positivity of A under the stronger assumption that the operator
A-I is compact in E. Let us give, without proof, the following results.
Theorem 5.1.9. Let A be the positive operator in a Banach space E. Suppose
problem (5.1) is well posed in Lpo(E) for some Po, 1 < Po < 00. Then it is well
posed in Lp(E) for any p, 1 < P < 00 and the coercivity inequality holds:
Ilv"liLp(E) + II Av IILp(E) + Ilvlle(E1_1/p,p)

~
M(pO)p2
p_ 1
IlfIILp(E) + M(llvoIIE1_1/P,P + IIvrIIE1_1/p,p),
where M(po) and M does not depend on p, Vo, VT and f(t). Here the Banach space
EI-I/p,p = E I_ I/ P ,p(D(AI/ 2), A I / 2) consists of those vEE for which norm
IlvIlE1_1/p,P
) lip
is finite.
~ II exp{ -zAI/2}v(z)lI~dA
(1 ~ P < (0)
< p < 00 and < a < 1. Suppose that A is the positive

operator in a Banach space E. Then problem (5.1) is well posed in Lp(Eo:,p) and
Ilv"IILp(E p)+ IIAvIlLp(Ea,p) ~ a(l~ a) IlfIlLp(Ea,p) +M(IIAvoIIEa,p + IIAvTIIEa),

a
where M does not depend on a, p, va, VT and f(t).
225

00 and 0 < a < 1. Suppose that A is the positive

operator in a Banach space E. Then problem (5.1) is well posed in Lp(Ea,q) and
where M(q) and M does not depend on a, p, va,
VT
Here, the Banach space Ea,q = Ea,q(B, E)(O

of those vEE for which the norm
roo
II v IIE",q= (}o
>..1-0.
II
and f(t).
< a < 1,1 < q < (0) consists
Bexp{ ->..B}v II~
d>"
-:x)
l/q
is finite.
Finally, we consider the applications of these results to the elliptic equations. First, we consider the boundary-value problems for two-dimensional elliptic
equations
f)2 u
f)2 u
- f) 2 - a(x) f)x 2 +<5u = f(y,x),O < y < T,O < x < 1,
y
u(O,x)
= <p(x),u(T,x) = ~(x),
f(O,x)
= Hr,x) = 0,0::; x::;
(5.25)
1,
u(y,O) = u(y, 1), ux(y,O) = ux(Y, 1), 0::; y ::; T,

where a(x), <p(x), ~(x) and f(y,x) are given sufficiently smooth functions and
a(x) > 0, <5 > 0 is a sufficiently large number.
We introduce the Banach spaces C,B[O,l] (0 < j3 < 1) of all continuous
functions <p(x) satisfying a Holder condition for which the following norms are
finite:
II <p IlciJ[o,l]=1I <p Ilc[o,l] +
sup
O~x<x+T~l
l<p(x + T) - <p(x) I
,B
T
where C[O, 1] is the space of all continuous functions <p(x) defined on [0, 1] with
the usual norm
II <p Ilc[o,l]= orp~ll<p(x)l.
AXv = -a(x)v"(x) + 6v(x)

defines a positive operator AX acting in C,B[O, 1] with domain C,B+2[0, 1] and satisfying the conditions v(O) = v(l), vx(O) = v x (1).
Therefore, we can replace boundary-value problem (5.25) by the abstract
boundary-value problem (5.1). Using the results of Theorems 5.1.2,5.1.5 and 5.1.8,
we can obtain that
226
Theorem 5.1.12. For the solution of the boundary-value problem (5.25) the follow-
ing coercive inequalities are valid:
II U II C6i B'''(Cl'[O,1J) + II u IIcgr"(c2+I'[O,1J)

M(/-L)
~ ((3 - ')')(1 - (3) II f Ilcgr"(Cl'[O,lJ) +M(/-L)(II<plb+I'+B-"[O,l] + 11'l/JIIC2+I'+B-"[O,lj),
')' < (3 < /-L + (3 - ')' <

~
1,
II u IIC6iB,,,(ca-B[o,1J) + II
1,
IIcgp(C2+a-B[o,1J)
M(Ci,(3)
~ Ci(l _ Ci) II f II c gr"(c a- B[O,lJ) +M(/-L)(II<pllc 2 + a -"[O,lj
')'
~
~ (3 ~ Ci,
+ 1I'l/Jllc2+ a-"[O,lj),
< Ci - (3 < 2'

1
Here M(/-L) and M(Ci,(3) are independent of ')', f(y,x), <p(x), 'l/J(x).
Second, let n be the unit open cube in the n-dimensional Euclidean space
IR n
(0 < Xk < 1, 1 ~ k ~ n) with boundary S, 0 = nUs. In [0, T] x n
we consider the mixed boundary-value problem for the multidimensional elliptic
equation
8 2 u(y,x) ~
8 2 u(y,x)
8 2 -L.,..Cir(X) 8x 2 +<5u(y,x)=f(y,x),
y
r=l
r
= (Xl, ... ,xn ) E n, < y < T,
u(O, x) = <p(x) , u(T, x) = 'l/J(x), f(l,x) = f(T, x)

u(y,x) = 0, XES,
(5.26)
E
0,
where Cir(X) (x E n) and f(y,x) (y E (O,T), x En), <p(x), 'l/J(x)(x EO) are given
smooth functions and Cir(X) > 0, <5 > is a sufficiently large number.
We introduce the Banach spaces
l (0) ((3 = ((31, ... , (3n), < Xk < 1,
k = 1, ... , n) of all continuous functions satisfying a Holder condition with the
indicator (3 = ((31, ... ,(3n),(3k E (0,1), 1 ~ k ~ n and with weight xfk(l - Xk h k )!3k, ~ Xk < Xk + h k ~ 1, 1 ~ k ~ n which is equipped with the norm
cg
sup
If(Xl,"" xn) - f(Xl + hI,"" xn + hn)1
O::;Xk<Xk+hk::; l,l::;k::;n
n
II h'k!3k Xfk (1 -
Xk - hk)!3k,
k=l
where C(O) stands for the Banach space of all continuous functions defined on n,
equipped with the norm
II f IIc(TI)= m~ If(x)l
xEfl
227

x
~ ()i)2v(y,x)
A v = - ~ ar x
ax 2
+ <5v (y, )
X
r=l
cg
c5i
i3 (n)
defines a positive operator AX acting on 1 (n) with domain D(AX) C
and satisfying the condition v = 0 on S.
Therefore, we can replace boundary-value problems (5.26) by the abstract
boundary-value problems (5.1). Using the results of Theorems 5.1.2 and 5.1.5, we
can obtain that
ing coercive inequality is valid:
M(f.l)
< ({3 - ')')(1 - {3)

n
+M(f.l)(L
r=l
O:S; ')'
II
If
Ilcgr'Y(ctl(IT))
a2~
ax21Ict/I3-'Y(IT)
r
a2~
+ L " ax21Ict/I3-'Y(IT)),
r=l
< {3 < 1,f.l = {f.ll, ... ,f.ln},O < f.lk
where M(f.l) is independent of {3,')' and f(y,x),
+ {3 -
')' < 1,1
:s; k:S; n,
~(x), ~(x).
Third, we consider the boundary-value problem on the range {O

lRn } for 2m-order multidimensional elliptic equations
a 2u
- ay2
:s; y :s; T,
x E
a1r1u
+ '"
ar(x)a x a X rn +<5u(y,x)=f(y,x),
L...J
n
Ir l=2m
1
rl
0< Y < T, x, r E lRn , Irl = rl + ... + r n ,

U(O,X) = ~(X), u(T, X) = ~(X), f(l,x) = f(T, X)
= 0,
(5.27)
E
lRn ,
where ar(x) and f(y,x), ~(x), ~(x) are given sufficiently smooth functions and
ar(x) > 0, <5 > 0 is a sufficiently large number.
We will assume that the symbol
BX(~) =
ar(x) (i6f' ... (i~nfn,~ = (6, ... ,~n)
Rn
Ir l=2m

(5.28)
228
for ~ :j:. O. Problem (5.48) has a unique smooth solution. This allows us to reduce
the boundary-value problem (5.48) to the boundary-value problem (5.1) with a
strongly positive operator AX = BX + 6I defined by (5.27) in a Banach space
E = CJl(Rn) of all continuous bounded functions defined on IR n satisfying a Holder
condition with the indicator f.L E (0,1).
ing coercivity inequalities are satisfied:
o :S 'Y < (3 < 1,
0 < f.L
+ m((3 -
'Y) < 1,
O:S 'Y:S (3, 0 < m(o: -13) < 1,
where M(f.L) and M(o:,(3) does not depend on 'Y and f(y,x),cp(x),'ljJ(x).
3.1.1, the positivity of the operator AX in CJl(Rn), the structure of the fractional
spaces Ea(C(Rn), (AX)~) and the coercivity inequality for an elliptic operator AX
in CJl(Rn).
5.2
229
Difference Schemes Generated by an Exact Difference Scheme
We consider the boundary-value problem (5.1). For the construction of the twostep difference schemes of an arbitrary high order of accuracy for the approximate
solutions of the boundary-value problem (5.1) we consider the uniform grid space
[0, T]T
= {tk = kT, k = 0,1, ... , N, N7 = T}.
By Theorem 3.1.2 we have the following two-step difference scheme for the solution
of the boundary-value problem (5.1):
+( -21 + (2B7)-1{I - exp( -27B)} exp( -27B)

+(2B7)-1 {I - exp( -27 B)} )V(tk)
+(1 - (2B7)-1{I - exp( -27B)} exp( -7B))v(tk+d]
+(2BT)-1{I - exp( -2TB)}
tk
(2B)-1{exp( -(tk - z)B)
tk-l
-exp(-(z - tk-2)B))}f(z)dz],
= vo, v(T) = vr.
1 ::; k ::; N - 1, v(O)
It can obviously be rewritten as the equivalent boundary-value problem for the

second-order difference equations
-(2B7)-1{I - exp(-27B)} exp( -7B)v(tk-d

+(2B7)-1{I - exp( -27B)}{I + exp( -27B)} )V(tk)
-(2B7)-1{I - exp( -27B)} exp( -7B)v(tk+d
(2B7)-1{I - exp( -27B)}[
tk + 1
tk
-exp(-(tk+2 - z)B))}f(z)dz
(2B)-1{exp( -(z - h)B)
230
itt
(2B)-I{exp(-(tk - z)B) - exp(-(z - tk-2)B))}f(z)dz],
tk-l
1 :S k :S N - 1, v(O) = vo,v(T) = VT,

or
1
-2'(Uk+l - 2Uk
7
+ Uk-I) + 2'[(1
7
exp( -7B))(Uk-l
+ uk+d
(5.29)
+ (exp( -27B) - 1)ukJ = /k,

fk = (2B7)-I(h,k + hk+d - (2B7)-1 exp( -7B)(h,kH + hk),
k
II k =! {t
,
h ,k =
it
11
-7
1 :S k :S N - 1, Uk
exp( -(tk - z)B))f(z)dz,
tk-l
tk
exp( -(z - tk-dB)f(z)dz,
tk_l
= V(tk), 0 :S k :S N, Uo = v(O), UN = v(T).
The latter will be referred to as the exact two-step difference scheme for
the boundary-value problem (5.1). Let us investigate the well-posedness of the
exact two-step difference scheme (5.29). This problem is uniquely solvable, and
the following formula holds:
Uk = (I - e-2tNB)-I{(e-tkB - e-(2t N-t k)B)uo
(5.30)
+(e-(tN-tdB _ e-(tNHdB)UN
_(e-(tN-tk)B _ e-(t NHk)B)7 2
N-l
L (e-(tN-ti)B -
e-(tN+ti)B)(1 - e- 2TB )-1 fi}
i=1
+7 2
N-l
L (e-ltk-tiIB - e-(t d t;JB)(1 - e- 2TB )-1 fi, 1 :S k :S N -
1.
i=1
Really, problem (5.29) can be rewritten as the equivalent nonlocal boundary-value

problem for first-order linear difference equations
7- 1(Uk - Uk-I) + 7- 1 (1 - e- TB )uk_l = Zk,

1 :S k :S N, Uo = v(O), UN = v(T),
{
-7- 1 (Zk+l
- Zk)
+ 7- 1 (1 - e-TB)zkH
/k, 1 :S k :S N
From this there follows the system of recursion formulas
Uk = e-TBuk_l
+ 7Zk, 1 :S k :S N,
Zk = e-TBzk+l
+ 7fk, 1 :S k:S N-1.
- 1.
231
Hence
Uk
e-kTBu 0
+"
L.J e-(k-i)rB 7Z
t,
1<
_ k
<
_ N,
i=l
Zk
= e-(N-k)rB zN +
N-1
e-(j-k) rB 7
fJ,1 :::; k:::; N -
1.
j=k
+L
N-1
UN
e-NrBUO
+L
N-1
e-(N-i)rB 7 {e-(N-i)rB ZN
e-(j-i)rB 7 !j}
+ 7ZN
j=i
i=l
Since
N-1
e-(N-i)rBTe-(N-i)rB
+ 71 = 7(1 -
L
N
e- 2rB )-1
i=l
e- 2(N-i)rB(1 - e- 2rB )
i=l
= 7(1 -
N-1
e- 2rB )-1(1 _ e- 2NrB ),
N-1
e-(N-i)rB 7
e-(j-i) rB 7
j=i
i=l
L 7(1 -
fJ
L L
N-1
j=l
e -(N+j-2i)
rB
fJ
i=l
N-1
e- 2rB )-1(e-(N-j)rB - e-(N+j)rB)7!j,
j=l
we have that
7- 1 (1
ZN
e- 2rB )(1
-2::= 7(1 -
e- 2NrB )-1{UN _ e-NrBUO
N-1
e- 2rB )-1(e-(N-j)rB - e-(N+j)rB)7!j}
j=l
e- 2rB )(1
7- 1 (1
-(1 -
e- 2NrB )-1
e- 2NrB )-1{UN _ e-NrBUo}
N-1
(e-(N-j)rB - e-(N+j)rB)7!i-
j=l
Therefore
Uk
= e-kTBuo +
L
k
e -(k-i)rB 7 {e-(N-i)rB ZN
N-1
e-(j-i)rB 7 ! J
j=i
i=l
-kTB
+~
-(k-i)rB
-(N-i)rB
+~
-(k-i)rB
Uo
Le
7e
ZN
Le
7
i=l
i=l
N-1
~
j=i
-(j-i)rB
! j.
232
Since
k
e-(k-i)rBTe-(N-i)rB = T(I - e- 2TB )-1 L
i=l
e-(N+k-2i)rB(1 - e- 2rB )
i=l
k
N-1
'"'
-(k-i)TB
'"' -(j-i)rB Tj
!
~e
T~e
j=i
i=l
e-(j+k-2i) TB T fJ
= LT L
j=l
N-1
k
L
T L e-(k+ j -2i) TB T fJ
j=k+1
i=l
i=l
= T(I - e- 2rB )-1 L(e-(k-j)TB - e-(k+j)TB)T!j

j=l
N-1
+T(I - e- 2rB )-1 L
(e-(j-k)TB - e-(k+j)TB)T!j
j=k+1
= T(I
N-1
- e- 2rB )-1 L (e-lk-jITB - e-(k+j)TB)TfJ,
j=l
we have that
Uk
e-krBuQ
+ T(1 -
e- 2TB )-1(e-(N-k)TB - e-(N+k)TB)ZN
N-1
+T(I - e- 2rB )-1 L (e-lk-jlrB - e-(k+j)TB)T k
j=l
From this formula and using the formula for ZN, we can obtain formula (5.30).
Let us denote by Fr(E) = F([O, T].,., E) the space of grid functions <pr =
{ <pd f~/ for fixed T = ft Thus, F T (E) is the vector space whose elements are
ordered (N - 1)-tuples of elements of E. The space Fr(E) can be equipped with
various norms and thus become a normed space. Thus, for instance, the vector
space FT(E) generates the normed space Cr(E) = C([O, T].,., E) with the norm
the normed space Ce(E) = CI3([O, T]r, E) 0< j3 < 1, with the norm
II <pTIICI'(E)
= II <pTll c (E)l
+ :=;k<k+r:=;N
max
II <Pk+r TT
-1
<Pk
liE (rT1)f.l'
IJ
233
= C.8,'Y([O,T]nE),O < ,
:S /3, 0 < /3 < 1,
the weighted normed space Ce'''!(E)

with the norm
II 'PTllc~'-Y(E) = II 'PTllcT(E)
((k
+ l~k<k+r~N-I
max
II 'Pk+r - 'Pk I E
+ r)r)'Y((N - k)r)'Y
(rr).8
and the normed space Lp,T(E) = Lp([O, T]n E), 1 :S p <
00,
with the norm
N-I
II cpTIILp,T(E) = (2: I CPkllir)~.

k=l
Let us consider the difference problem (5.29). We will reduce it to an operator

problem in the space FT (E). First we shall define an operator D; acting from the
space E x FT(E) x E of vectors wT = {wdf=o into the space FT(E) of vectors
vT = {vd~=-/ according to the rule
vT = D;UT,Vk =
~(Wk+l
r
2Wk
+ wk-d,k = 1, ... ,N-1.
Second we shall define an operator AT acting from the space E x FT(E) x E of

vectors wT = {wdf=o into the space FT(E) of vectors vT = {vd~~/ according to
the rule
v T = ATuT,Vk = ~[(I - exp( -rB))(wk_1 + wk+d
r
+(exp( -2rB) - I)wk], k = 1, ... , N - 1.

Next, let us introduce the continuation operator IT(uo, UN), which acts from
Ex FT(E) x E to FT(E) according to the rule
IT(uo, UN HUI, ... , uN-d
(uo, UI,, UN-I, UN)'
Then, clearly, the difference problem (5.29) is equivalent to the operator problem
-D;IT(UO,UN)U T + ATIT(UO,UN)UT =
Here
is defined by the formula
= (iI,, iN-d
solvability for any un, UN E E and
E FT(E) it follows that its solution u T
defines a continuous additive and homogeneous operator uT(fT, UN, UN)'
The boundary-value problem (5.29) is said to be stable in FT(E) if we have
the inequality
r, un, UN, but also of r.
234

Theorem 5.2.1. Let A be the positive operator in a Banach space E. Then difference
problem (S.29) is stable in CT(E).

The proof of this theorem is based on formula (S.11) and the estimate
Ilexp(-hB)/IE-oE ~ M e- 6kT ,J > 0,
(S.31)
for all k, k ~ 1 and all r > O,where M does not depend on k and r. Estimate
(S.31) follows from (S.8) for t = tk = hand a(B) = J.
problem (S.29) is stable in C,:,Cl(E), 0 < a < 1.
The proof of this theorem is based on formula (S.11), estimate (S.31) and
Ilexp( -hB) - exp {-(k
rCl
+ r)rB} IIE-o E ~ M (k + r)Cl'
(5.32)
for all k, r, 1 ~ k < k + r ~ N - 1 and all r > 0, where M does not depend on k,
rand r. Estimate (S.32) follows from (5.10).
problem (S.29) is stable in Lp,T(E), 1 ~ p < 00.
The proof of this theorem is based on formula (S.11) and estimate (S.31).
The boundary-value difference problem (5.29) is said to be well posed (coercively stable) in FT(E) if we have the coercive inequality
II
+
II
{r- 2 (uk+l - 2Uk + uk_d}~-l IIFT(E)
{r- 2 [(I - exp( -rB))(uk-l + uk+d + (exp( -2rB) - I)uk]}~-l IIFT(E)
r IIFT(E) + II Auoil E + IIAuNII E],

where M is independent not only of r, Uo, UN, but also of r. Since the boundary~
M[II
value differential problem (S.l) in C(E) is not well posed for the general positive
operator A and space E, then the well-posedness of the difference boundary-value
in CT(E) norm does not take place uniformly with respect to r > O. This means
that the coercive norm
I uT IIKT(E)
+
II
{r- (uk+l - 2Uk
+ Uk_l)}~-l IlcT(E)
{r- 2 [(1 - exp( -rB))(uk-l + Uk+l) + (exp( -2rB) - I)uk]}~-l
II cT (E)
tends to 00 as r ~ +0 . The investigation of the difference problem (5.29) permits

us to establish the order of growth of this norm to 00. First, let us prove a lemma
that will be needed in the sequel.
235
Lemma 5.2.4. Suppose A is the positive operator in a Banach space E. Then the
following estimate holds:
N-l
2: 11(1 - e-TB)e-tj-lBIIE-+E :=::; M min (In(l/T), 1 + lIn /lBIII),
(5.33)
j=l
T.
Proof. Using estimate (5.10), we obtain
2: II(I - e-TB)e-tj-lBIIE-+E :=::; M 2: ~J :=::; M Jds = M
N-l
N-l
. 1
. 1
J=
J=
In N :=::; M 1 In(1/T) .
(5.34)
Further, using estimate (5.10), we obtain
If IIBIIE-+E
> N, then
2: 11(1 - e-TB)e-tj-lBIIE---->E :=::; M 2: ~
N-l
N-l
j=l
j=l
:=::; M
<M
If IIBIIE---->E
IIBIIE~E
ds
--;
ds
--; :=::; Mlln IIBIIE---->EI.
< 1, then
N-l
N-l
j=l
j=l
2: 11(1 - e-TB)e-tj-lBIIE---->E :=::; M 2: TIIBIIE---->E :=::; Mi'
Finally, if I:=::;
IIBIIE-+E :=::; N,
then
:=::; M(l+
ds
s
- ) :=::; M(l
+ In IIBIIE---->E)'
236
Thus, in all three cases we have the estimate
N-l
11(1 - e-rB)e-tj-1BIIE--.E
:s M(l + lIn II B IIE--.EI)
(5.35)
j=l
Finally, estimate (5.33) follows from (5.34) and (5.35). Lemma 5.2.4 is proved.
Theorem 5.2.5. Let A be the positive operator in a Banach space E and Uo E D(A),
UN E D(A). Suppose the operator 1 +exp(-rB) has an inverse (1 +exp(-rB))-l
and
11(1 + exp( -rB))-lIIE--.E :s M,
(5.36)
where M does not depend on r. Then the solutions of the difference problem (5.29)
in Cr(E) obey the almost coercive inequality
II u r
IIKT(El
:s MIlII
+ min {ln~, 1 + lIn II

Auoil E + IIAuNIlE
B IIE--.EI} II
II cT (El],
r, uo, UN, but also of r.

r- 2(uk+l - 2Uk + Uk-r)
=
r- 2(1 - e-2tNB)-1(1 _ e-rB)2{(e-tk-1B _ e-(2t N-tk+l lB)uo

+(e-(tN-tk+dB _ e-(tNHk-dB)UN
_(e-(tN-tk+dB _ e-(t N+tk- d B)r 2

+r- 2r 2(1 - e- rB )2
N-l
L (e-(tN-t;JB - e-(tNH;lB)(I - e- 2rB )-1 fi}
i=l
k-l
L e-(tk-l-til B(1 - e- 2rB )-1 Ii

i=l
+r- 2r 2(1 - e- rB )2
N-l
e-(t i -t k+d B(1 - e- 2rB )-1 fi - r- 2r 22(1 _ e- rB )
i=k+l
X (1
- e- 2rB )-lik - r- 2r 2(1 - e- rB )2
N-l
L e-(tk-l H;lB)(1 - e- 2rB )-1 fi
i=l
= (1 - e- 2tNB )-1(I - e-rB)2r-2{(e-tk-1B - e-(2t N-t k+1l B)uo
+(e-(tN-tk+dB _ e-(tNHk-dB)UN}
-(1 - e- 2tNB )-1(I _ e-rB)(e-(tN-tk+llB _ e-(tNHk-d B )
(5.37)
L (e-(tN-ti)B -
N-1
X
e-(tN+ti)B)(1
237
+ e- TB )-l Ii
i=l
k-1
+(I - e- TB )
e-(tk-1-ti)B(I
+ e- TB )-l Ii
i=l
N-1
+(I - e- TB )
e-(ti-tk+tlB(I
+ e- TB )-l Ii
- 2(1
+ e- TB )-l!k
i=k+1
N-1
-(I - e- TB )
e-(tk-1H;)B)(I
+ e- TB )-l Ii
= gk
+ Pk,
i=l
where
gk = 7- 2 (I - e- 2tNB )-1(I - e-TB)2{(e-tk-lB _ e-(2t N - t k+tl B )UO
+(e-(tN-tk+tl B _ e-(tNHk-tlB)UN},
Pk = -(I - e- 2tNB )-1(I - e-TB)(e-(tN-tk+tl B _ e-(tNHk-tl B )
L (e-(tN-ti)B -
N-1
e-(tNHi)B)(1
+ e- TB )-l Ii
i=l
+(I - e- TB )
k-1
e-(tk-l-ti)B (I
+ e- TB )-l Ii
i=l
N-1
+(I - e- TB )
e-(ti-tk+tlB(1
+ e- TB )-l Ii -
2(1
+ e- TB )-l Ik
i=k+1
-(I -
e- TB )
N-1
e-(tk-1H;)B)(1
+ e- TB )-l k
i=l
To this end it suffices to show that
II gTllc,.(E)
T
II p Ilc,.(E) :::; M 1 min
:::;
MIlII
Auoil E
+ IIAuNIIE],
{ln~, 1 + lIn II B IIE->EI} II r
(5.38)
Ilc,.(E)
(5.39)
Estimate (5.38) follows from the estimate
11(1 -
e-
TB
)2(7B)-21IE->E :::;
and the fact that the homogeneous difference equation (r = 0) is stable in CT(E)
uniformly in 7. Using estimates (5.10), (5.11) and (5.36), we obtain
IlpkllE :::;
N-1
X
i=l
11(1 -
e-2tNB)-11IE->Elle-(tN-tk+tlB -
Ile- TB - e-(tN+1+ti)BIIE->EII(I -
e-(tNHk-tlBIIE->E
e-TB)e-(tN-l-ti)BIIE->E
238
k-1
L 11(1 - e-rB)e-(tk-l-ti)BIIE--+EII(1 + e- rB )-lIIE--+EllfiIIE

i=l
N-1
11(1 -
e-rB)e-(ti-tk+IlBIIE--+EII(1 + e- rB )-lIIE--+EllfiIIE
i=k+1
N-1
i=l
Ile-tkBIIE--+EII(I - e-rB)e-ti-lBIIE--+E1I(1 + e- rB )-lI1E--+EIIJiIIE

N-1
:S M
L 11(1 - e-rB)e-tj-lBIIE--+E11 r
IlcT(E)'
j=l
From the last estimate and estimate (5.33) follows estimate (5.39). Theorem 5.2.5
is proved.
Note that estimate (5.37) is established for arbitrary positive operator A. In
contrast, estimate (5.38) is established under the assumption (5.36). When is the
last condition satisfied? We have an answer in the case where E = H is a Hilbert
space. Furthermore, the method of proof of Theorem 5.2.5 enables us to establish
the following fact.
Theorem 5.2.6. Let A be the positive operator in a Banach space E and uo, UN E
D(A). Suppose the operator I +exp(-TB) has an inverse (1 +exp(-TB))-l and
E D((1 + exp( -TB)) -1). Then the solutions of the difference problem (5.29) in
Cr(E) obey the almost coercive inequality
II u r II KT (E) :S Mdll Auoil E + IIAuNIl E
+ min {ln~, 1+ lin II B IIE--+EI} II (1 + exp( -TB)) -1

Since
r, uo,
UN,
r IlcT(E)],
but also of T.
J
r
B- (1
+ exp( _rB))-l
exp( -sB)ds( 1- exp( -2TB))-\
o
using estimates (5.9) and (5.10), we obtain
IIB- 1 (1 + exp( -T B)) -lIIE--+E
(5.40)
:S
lIexp( -sB)IIE--+Eds
x 11(1 -
o
This yields the following result.
exp( -2TB))-
II E--+ E :S
M(l - e- Or )
<5(1 _ e- 20r ) :S
T'
239
Theorem 5.2.1. Suppose A is the positive operator in a Banach space E and uo,
E D(A),
E D(B). Then the solutions of the difference problem (5.29) in
Cr (E) obey the almost coercive inequality
UN
II ur IIKT(E) :s M1 [II Auoil E+ IIAuNll E

+ min {In~, 1 + lIn I
IIE-oEI} I
r, uo,
UN,
Br II cT (E)],
but also of T.
Now let us study the well-posedness of the difference problem (5.29) in various
Banach spaces.
Theorem 5.2.8. Suppose that the assumptions of Theorem 5.17 hold. Then the
solutions of the difference problem (5.29) in C;;'CE) obey the coercivity inequality
I
+
II
2
{T- [(1 -
2
{T- (Uk+l - 2Uk
+ Uk_l)}~-1 Ilc~''''(E)
exp( -TB))(Uk-l + uk+d + (exp( -2TB) - I)uk]}~-1 Ilc~''''(E)
:s M1 [a(1 ~ a) I r Ilc~''''(E) + I AuoilE+ IIAuNII E],

r, uo,
UN,
a, but also of T.

2
T- (Uk+l - 2Uk
= (1 +(1 -
+ Uk-I)
e- 2tNB )-I(I - e-rB)2T-2(e-tk-1B _ e-(2t N -t k +d B)uo
e- 2tNB )-I(1 - e-rB)2T-2(e-(tN-tk+dB - e-(tNHk-dB)UN
+{ -(1 -
e- 2tNB )-I(1 - e-rB)(e-(tN-tk+d B _ e-(tNHk-d B )
N-l
X
(e-(tN-ti)B - e-(t N +t i )B)(1
+ e- rB )-1
i=1
+(1 -
e- rB )
L e-(t k- 1-t i )B(1 + e- rB )-1

k-l
i=1
N-l
+(1 -
e- rB )
e-(t i -tk+d B (1
e- rB )-1 -
2(1 +
e- rB )-1
i=k+l
-(I -
e- rB )
N-l
i=1
e-(tk-1HilB)(I
+ e- rB )-I} fk
240

-(I - e- 2tNB )-1(I - e-rB)(e-(tN-tk+tlB _ e-(tN+tk-tl B )
L (e-(tN-ti)B - e-(tN+ti)B)(I + e-rB)-l(Ji - !k)
N-l
i=l
+(I - e- rB )
L e-(tk-1-t;jB(I + e-rB)-l(Ji k-l
!k)
i=l
N-l
+(1 - e- rB )
e-(ti-tk+dB(I + e-rB)-l(Ji - /k)
i=k+l
-(I - e- rB )
N-l
L e-(tk-1+ti)B)(I + e-rB)-l(fi - !k)

i=l
= (I - e- 2tNB )-1(I - e-rB)2r-2(e-tk-IB - e-(2t N-t k+tl B)uo
(5.41)
+(I - e- 2tNB )-1(I - e-rB)2r-2(e-(tN-tk+tlB - e-(tNHk-dB)UN
+{ -I + (I - e- 2tNB )-1(I - e-tNB)[I _e- tk - 1B - e-tN-k-IB + e-tNB]}!k

+(1 - e-2tNBtl(I - e-2tN-kB)(I _ e- rB )
xL e-(tk-1-t;jB(I - e- 2tiB )(I + e-rB)-l(Ji k-l
/k)
i=l
N-l
X
e-(ti-tk+dB(I - e-2tN-iB)(I + e-rB)-l(fi - !k)
i=k+l
where
if =
(I - e- 2tNB )-1(I - e-rB)2r-2(e-tk-IB - e-(2t N-t k+d B)uo,
Jf = (I -
e- 2tNB )-1(I - e-rB)2r-2(e-(tN-tk+tlB - e-(tNHk-tlB)UN,
Jr = {-I + (I - e- 2tNB )-1(I - e-tNB)[I_e-tk-IB - e-tN-k-IB + e-tNB]}!k,

Jt =
(I - e- 2tNB )-1(I - e-2tN-kB)(I - e- rB )
xL e-(tk-1-ti)B(I - e- 2tiB )(I + e-rB)-l(Ji - !k),

k-l
i=l
JZ = (I -
N-l
i=k+l
e- 2tNB )-1(I - e- 2tkB )(I - e- rB )
e-(ti-tk+dB(I - e-2tN-iB)(I + e-rB)-l(fi - /k).
241
J:n
Let us estimate
= Uk} ~,;/ for any m = 1,2,3,4,5 separately. We start with
Using estimates (5.8) and (5.9) we show that, for 1 ::; k ::; N - 1,
Jr
IIJfIIE::; 11(1 xIle- tk - 1B -
e- 2tNB )-IIIE--+EII(I - e-TB)(7B)-III~--+E
e-(2tN-tk+IlBIIE--+EIIAuoIIE ::;
MIIAuoIIE'

1
IIJ1
JHrll E ::;
(2r)a
IIJ1 11E + IIJHrllE ::; 2MII AuoilE ::; 2M (1 + r)a IIAuollE

::;
M [
1
t~ + (tN t~
111 A Uo II E,
t~+l
_ tt}a
for 1 < r + 1 ::; N - 1. Further, applying (5.9) and (5.10) we show that, for
2 ::; k < k + r ::; N - 1,
IlJf -
Jf+rlIE ::;
11(1 -
e- 2tNB )-lIIE--+EII(I - e-TB)(7B)-lll~--+E
x[lle- tk - 1B _ e-tk+r-IlBIIE--+E
+lle-(2t N-t k+Il B _ e-(2tN-tk+r+IlBIIE--+EJIIAvoIIE
::;
M[tat~
k+r-l
+ (2t
~~t )a11IAvoIIE::; Ml[t:~k+r + (t N :tk)aJIIAuoIIE.
N
k+l
(5.42)

(5.43)
Now let us estimate J3. Using estimates (5.8) and (5.9) we obtain that
II J~ liE::; [1 + 11(1 - e- 2tNB )-IIIE--+EIII - e-tNBIIE--+E

x[1 + Ile-tk-1BIIE--+E + Ile-tN-k-1BIIE--+E + Ile-tNBIIE--+El11IikIIE
::; MIll r IIc~,a(E)
for 1 ::; k ::; N - 1. From this inequality it follows that
IIJ{ -
J{+rIIE ::;
::; 2Ml
IIJ{IIE + IIJ{+rIIE ::; 2M1 11 r
Ilc~,a(E)
(l(~)~JI r Ilc~,a(E) ::; M2[t;:l + (tN :
tdJl1
r IIc~,a(E)'
242
for 1 < r
+1~ N
- 1 and
1112 -
J2+rIIE ~
IIJ211E + 1112+r11E ~ 2M1 11 r Ilc;"(>(E)
(2r)a
r
~ 2M1 (1 + r)a II f
+ }:~a_ k)a II r Ilc;"(>(E)
= 2M1 (1
for 1 < k < r
+k = N
J2+r - J2
Ilc;"(>(E)
~ MZ[t~:r
+ (tN :tk)a]11 r IIc;"(>(E)'
- 1. Since
{-I + (1 - e- ZtNB )-l(1 - e-tNB)[I_e-tk+r-1B
+ e-tNBn(fk+r - ik)
+(1 - e- ZtNB )-l(1 - e-tNB)[e-tk-1B_e-tk+r-1B + e-tN-k-1B - e-tN-k-r-1B]fk,
_e-tN-k-r-1B
it follows that
II
xiiI -
J2+r - J2 IIE~ [1
+ 11(1 -
e-tNB)-lIIE->E
+ Ile-tk-1BIIE->E + lIe-tk+r-1BIIE->E
+lle-tN-k-1BIIE->E + Ile-tN-k-r-1BIIE-->e]Jllfk+r - fkllE
+11(1 - e-ZtNB)-11IE->e{lIe-tk-1B _e-tk+r-1BIIE->E
e-tNBIIE-->e[I
+lIe-tN-k-1B _ e-tN-k-r-1BIIE-->e}III - e-tNBIIE->EllikIlE

for 2 ~ k < k
II
+r
N - 2. Using estimates (5.8), (5.9) and (5.10), we obtain
J2+r - J211E
for 2 ~ k < k
+r
~ M1[(tk :tr)a + (tN :tk)aJIl r
IIc;"(>(E)
N - 2. From this it follows that
II J;
IIc;"(>(E~
r II c;"(>(E)'
MIll
(5.44)
Now let us estimate J:[. By the definition Jt = O. Therefore, we have to consider

and
~ k ~ N - 1. In the first case, using estimates (5.8)
two cases: 2 ~ k ~
and (5.9) we obtain that
If
II
If
k-l
Jt IIE~
11(1 _e-ZtNB)-lIlE->E L
Ile-(t k- 1-t i )B(1 _e-ZtN-kB)(1 _e- rB ) IIE-->E
i=l
x III - e- ZtiB IIE-->E 11(1 + e- rB )-lIIE-->EiIfi - fk liE

k-l
~ M (t _ t.)1-a~a(t _ t.)a II IIc;"Q(E)
I:
i=l
k-l
~ ta(t
~ t )a [L (t k _ :)1-a]1I r
k N k
i=l
Ilc;"(>(E)'
243
The sum enclosed in the right-hand side square brackets is the lower Darboux
integral sum for the integral
Thus, for 2 :S k :S J- we have established the bound
In the second case, using estimates (5.8), (5.9) and (5.10) we obtain that
II
k-l
J~ IIE:S 11(1 _e- 2tNB )-11IE--.E
L Ile-(t
k- 1
-t;)B(1 _e-2tN-kB)(I _e- rB ) IIE--.E
i=l
Since
it follows that
244
Thus, for ~ ~ k ~ N - I we have established the bound

II
Jt
IIE:S a(1
~~)tk II r Ilc~,o(E) :S ~~2QQ II r Ilc~,o(E)'
Combining the estimates for II
Jt
liE, we obtain
II
Jt
IIE:S
max
19~N-l
a(~ a) II r Ilc~,o(E)'
(5.45)
Now let us estimate the difference Jt+r - Jt for 2 :S k < k + r :S N - 1. We will

consider separately the cases k ~ r, N - k :S 2r and k > r, N - k > 2r. If k ~ r,
then (5.17) yields
II Jt+r - Jt IIE~II Jt+r liE
HQ
12
M
~ a(1- a) II fT II C~,O(E{Q(k
+ II
Jt liE
+ r )-Q = a(1M_ a) II
fT II C~,O(E)trQ( tk
+ t r )-Q .
If N - k :S 2r, then (5.17) yields
II Jt+r - Jt IIE:SII Jt+r liE

:S
~12HQ) II r
a1-a
Ilco,o(E)rQ(N - k)-Q =
T
+ II
Jt liE
(M2 ) II
aI-a
r Ilco,o(E)t~(tN -
tk)-Q.
Now let k > rand N - k > 2r. Let us represent the difference Jt+r - Jt as the
following sums:
x{
k+r-l
e-(tk+ T -
1-
t il B (1 - e- 2tiB )(1 - e-2tN-k-rB)(I
+ e-TB)-l(fi -
fk+r)
i=l
-L
k-l
e-(t k - 1 -t il B(1 - e-2tN-kB)(I - e- 2tiB )(1
+ e-TB)-l(fi -
i=l
(1 -
e- 2tNB )-1(I - e- TB )
k+r-l
e-(tk+r-l-ti)B
i=k-r
-(1 -
k-l
e- 2tNB )-1(1 - e- TB )
i=k-r
e-(tk-l-til B
fk)}
+(1 -
l:
245
k-r-l
e- 2tNB )-1(1 - e- rB )
e-(tk+r-l-t;jB
i=l
+(1 -
e- 2tNB )-1(1 - e- rB )
l:
k-r-l
{e-(tk+ r -l- t ;)B(1 - e-2tN-k-rB)
i=l
_e-(t k - 1 -t;)B(1 _ e-2tN-kB)}(I - e- 2t ;B)(1
Now let us estimate
l:
k+r-l
X
11(1 -
+ e-rB)-l(fi - /k)
pI + P~ + pf + P:'
PI. Using estimates (5.8) and (5.9), we obtain that
e-rB)e-(tk+r-l-t;jB(I - e-2tN-k-rB)IIE-+E
i=k-r
(tk
k+r-l
+ ~r)Q [i~r
(tk+r - ty:Q(tN -
ti)QJII
r Ilc;>"(E)
Since
it follows that
Thus, we have established the bound
246
Since
p2 = (I -
e-2tNB)-le-2t2rB(I - e-tk-r-1B)
X(I - e- 2t ;B)(I - e-2tN-k-rB)(I
+ e-TB)-l(Jk -
fk+r),
it follows that
II p2 IIE:S II(I -
e-2tNB)-11IE-+Elle-2t2rB(I - e-tk-r-1B)IIE-+E
xIII - e- 2t ;BIIE-+EI [I - e-2tN-k-rBIIE-+EII(I + e-TB)-lIIE-+Ellik -
fk+rIIE.
Now let us estimate
2: {II
k-r-l
X
pr Using estimates (5.8), (5.9) and (5.10), we obtain
(I - e- TB )(e-(2t N -2t kl B - e-(2tN-2tk-2tr)B)e-(tk-l-t;)B
IIE-+E
i=l
+ II
xiiI - e- 2t;BIIE-+EII(I + e-TB)-lIIE-+E

k-r-l
:S
IIE-+E }
II fi - ik liE
(I - e-TB)(I - e-tr)B)e-(tk-1-t;)B
M::[ L
k
i=l
(t -t.)2-:(t
k
_t)aJII
tNt
Ilc;"o(E)'
Since for k
> rand
N - k
> 2r we have the bound
247
it follows that
M
II pt liE::; --t~(tN
1- a
tk)-Otk"ll
Ilc"'''(E)'
T
II pt liE::; ~2: t~(tN
- tk)-O(tk + tr)-Oll
r IIc~'''(E)'
Thus, we have shown that for any 2 ::; k < k + r ::; N - 1 the following inequality
holds:
II J~+r
J~ liE::; a(~ a) t~(tN -
tk)-O(tk + tr)-Oll
Ilc~'''(E)'
(5.46)
II J;
IIc~'''(E)::; a(~ a) II r IIc~'''(E) .
(5.47)
II J;
Ilc~'''(E)::; a(~ a) II r Ilc~'''(E) .
(5.48)
Finally, applying estimates (5.42), (5.43), (5.44), (5.47) and (5.48) we obtain the
estimate
II {r- 2 (uk+! - 2Uk + Uk_1)}f- 1 Ilc~'''(E)
::; M 1[a(1 ~
a) II r
IIc~'''(E) + II Auoil E + IIAuNIIE1
By the triangle inequality, this last estimate and difference equation (5.29) yields
II {r- 2 [(1 - exp( -rB))(Uk-1 + Uk+!) + (exp( -2rB) - I) uk]}f- 1 IIc~'''(E)
::; Mda(l
~ a) II r IIc~'''(E) + II Auoll E + IIAuNII E]

Furthermore, the method of proof of Theorem 5.2.8 enables us to establish
the following fact.
Theorem 5.2.9. Suppose that the assumptions of Theorem 5.2.6 hold. Then the
solutions of the difference problem (5.29) in C,:,O(E) obey the coercivity inequality
II
2
1
{r- (uk+1 - 2Uk + Uk_1)}f- Ilc~'''(E)
+ II {r- 2 [(1 - exp( -rB))(uk-1 + uk+d + (exp( -2rB) - I) uk]}f- 1 Ilc~'''(E)

::; Mda(l ~ a) II (1 + exp( _rB))-l
r Ilc~'''(E) + II Auoil E + IIAuNII E],
r, uo, UN, a, but also ofr.
248
This yields the following result

solutions of the difference problem (5.29) in c~,a(E) obey the coercivity inequality
II
+ II {T- 2[(1 -
{T-2(uk+1-2uk+uk-d}f-Illc~,0(E)
exp( -TB))(Uk_1
+ uk+d + (exp( -2TB) -
I)uk]}f- l II c -:",o(E)
~ Md aI-a
(1 ) II Br Ilco,o(E) + II AuoilE + IIAuNII E],
T
r, uo, UN, a, but also of T.
II
+ II {T- 2[(1 -
2
l
{T- (Uk+1 - 2Uk +Uk-Inf- IIc~(E)
exp( -TB))(Uk-1
+ Uk+l) + (exp( -2TB) -
I)uk]}f- l IIc~(E)
~ Md a (1 ~ a) II r Ilc~(E) + II Auoll E + II Au NII E]

fails. Nevertheless, we have the following result.
solutions of the difference problem (5.29) in C~(E) obey the coercivity inequality
II
+ II {T- 2[(1 -
2
1
{T- (Uk+1 - 2Uk +Uk_Inf- IIc~(E)
exp( -TB))(Uk-1
+ II {T- 2(Uk+1
+ uk+d + (exp( -2TB) - 1)Uk]}f- 1 Ilc~(E)

- 2Uk +Uk-Inf- l I cT (Eo )
~ a (~l-a ) II r II co(E) + M 1 [II

T
+11(1 -
(1 - e-rB)2T-2uo a
II liE
e-rB)2T-2UN - fN-IIIEJ,
r, uo, UN, a, but also ofT.
Proof. To prove the theorem it suffices to establish the estimates for

{T- 2(Uk+1 - 2Uk
in
C~(E)
+uk_d}f- l
and in Cr(Ea ). By formula (5.41), we have
T- 2(Uk+1 - 2Uk + Uk-I)

= (1 - e-2tNB)-1(e-tk-lB - e-(2t N-t k+,)B)[(1 - e-rB)2T-2uo -
II]
249
+(I - e-2tNB)-I(e-(tN-tk+l)B - e-(t N+tk-d B )[(1 - e-TB)2T-2UN - IN-I]

+(1 - e-2tNB)-I(e-tk-IB - e-(2t N- t k+JlB)[h - ik]
+(I - e-2tNB)-I(e-(tN-tk+dB - e-(tN+tk-JlB)[fk - IN-I]
+(I - e- 2tNB )-I(I - e-2tN-kB)(I _ e- TB )
X
k-I
I>-(tk-1-ti)B(I - e- 2tiB )(I + e-TB)-I(fi - ik)
i=1
N-I
X
e-(ti-tk+1)B(I - e- 2tN -;B)(I + e-TB)-I(fi - ik)
i=k+1
6
=LGk ,
m=1
where
G~ =
G~ =
(1 - e-2tNB)-I(e-tk-IB - e-(2t N-t k+J)B)[(I - e-TB)2T-2uo - h],
(1 - e-2tNB)-I(c(tN-tk+l)B - e-(t N+t k- d B)[(1 - e-TB)2T-2uN - IN-I],

(1 - e-2tNB)-I(e-tk-IB - e-(2t N-t k+Jl B)[h - Ik],
G~ =
G'k =
(I - e-2tNB)-I(e-(tN-tk+dB - e-(tN+tk-1)B)[ik - IN-I],

G~ = (I - e- 2tNB )-I(I - e-2tN-kB)(I _ e- TB )
k-I
XL e-(t k- 1-t i )B(1 - e- 2tiB )(1 + e-TB)-I(fi - ik),

i=1
G% = (1 - e- 2tNB )-I(I - e- 2tkB )(I _ e- TB )

N-I
X
e-(ti-tk+1)B(I - e- 2tN -;B)(1 + e-TB)-I(fi - ik)
i=k+1
Let us estimate G~ = {Gk}~~1 for any m = 1,2,3,4,5,6 separately. We

start with Gr. Using estimates (5.8), (5.9) and the definition of the spaces Ea., we
show that, for 1 :S k :S N - 1,
IIG~IIE :S
II(I - e- 2tNB )-IIIE->E
2tN-tk+l
IIBexp(-sB)[(I - e-TB)2T-2uo - hJllEds
tk-l
2tN- t k+l
:S M
tk-l
s~~a.II[(I _e-TB)2T-2uo_ hlllE"
:S
~111(I _e-TB)2T-2uo_ hIIE".
250
Further, applying (5.8), (5.9) and the definition of the spaces Eo we show that,
for 1 :s k < k + r :s N - 1,
IIG~
J liB
G~+rIIE
:s 11(1 - e- 2tNB )-11IE-->E
tk+r-l
x[
exp(-sB) [(1 -
e-rB)2r-2uo -
hJllEds
tk-l
2tN- t k+l
IIBexp(-sB)[(1 -
e-rB)2r-2uo -
hJIIEdsJ
2tN- t k+r+l
M
:s t~-lll(1
-
e-rB)2r-2uo -
hIIE".
Using estimates (5.8), (5.9) and the definition of the spaces Eo we show that, for
1 :s k :s N - 1,
IIG~IIE"
:s 11(1 - e-2tNB)-11IE-->Elle-tk-lB -
e-(2t N -t k+Il BIIE-->E
xll(1 - e-rB)2r-2uo - hlIE",

:s MII(1 - e-rB)2r-2uo - hIIE".
Now let
IIG;IICT(E",) :s MII(1 - e-rB)2r-2uN - fN-11IE".

us estimate G'3. Using estimates (5.8) and (5.9) we obtain that,
k:S N -1,
IIG~IIE
:s 11(1 -
for 1 :s
e-2tNB)-1IlE-->Elle-tk-lB - e-(2t N -t k+Il BIIE-->Ellh - fkllE
:s Mllrllc~(E)

Further, applying (5.8), (5.9) and (5.10) we show that, for 1 :S k
/I G~+r - G~
251
< k + r :S
N - 1,
/lE:S 11(1 - e- 2tNB )-IIIE->E
x {Ile- tk - IB _e-tk+r-IB IIE->E+lle-(2t N - tk +l l B_e-(2tN- t k+r+JlB IIE->e}llh - fk/lE

+11(1 - e-2tNB)-IIIE->Elle-tk+r-lB - e-(2t N -t k+r+Jl BIIE->Ellfk+r - fkllE
:S Mt~llrllc~(El'
By the definition of the spaces Eo and using estimates (5.8), (5.9), we obtain that
II A1- 0 Be-ABG~ IIE:S A1- o ll(1 - e- 2tNB )-11IE->E

x IIBe-AB(e-tk-lB - e-(2t N- t k+JlB)(h - /k)IIE
A1- o t o
:S M A + tkk~11 /lr/lc~(El :S M/lrllc~(El
for all A, A > 0 and for 1 :S k :S N - 1. So, we have established that
IIG31IC(Enl :S M 1 /I r /lC~(E) .
IIG~IIC(En) :S M 1 /I
/lC~(E) .
Now let us estimate G'5. By the definition Gf = O. Therefore, we have to consider

the cases 2 :S k :S N - 1. Using estimates (5.8) and (5.9) we obtain that
II G% /lE:S 11(1 _e- 2tNB )-IIIE->E
L Ile-(t k- -til B(1 _e-2tN-kB)(1 _e- rB ) IIE->E
k-l
i=1
k-l
:S
M[~ (tk _ :i)l- o JlI r Ilc~(E)
252
Thus, for 2 ~ k
N we have established the bound
II G~ IIE~ ~ltkall
r Ilc~(E)'
(5.49)
From this inequality it follows that

max
l~k~N-l
II G~ IIE~
M Tall
a
r IlcQ(E)'
(5.50)
Now let us estimate the difference G~+r - G~ for 2 ~ k < k + r ~ N - 1. We

will consider separately the cases k ~ rand k > r. If k ~ r, then (5.49) yields
I G~+r
~ ~ltkall
a
Now let k
liE + I G~ liE
G~ IIE~II G~+r
r IIC~(E) + ~ltkall r Ilc~(E)
M1(1 + 2 ) r all
a
t
fT
I C~(E) = -;;t
M 2 r all
> r. Let us represent the difference
fT
I C~(E)'
G~+r - G~ as the following sums:
G~+r - G~ = (I - e- 2tNB )-1(I - e- TB )
x{
k+r-l
e-(tk+r-l-t;)B(I - e- 2t ;B)(I - e-2tN-k-rB)(I
i=l
k-l
- L e-(tk-1-t;)B(I -
e-2tN-kB)(I - e- 2t ;B)(1
i=l
= (I - e- 2tNB )-1(I - e- TB )
+ e-TB)-l(Ji -
h+r)
+ e-TB)-l(Ji - h)}
k+r-l
e-(tk+r-l-t;)B
i=k-r
x(I - e- 2t ;B)(1 - e-2tN-k-rB)(1
-(I - e- 2tNB )-1(I - e- TB )
+ e-TB)-l(Ji -
fk+r)
k-l
e-(tk-l-t;)B
i=k-r
+(1 -
+(1 -
e- 2tNB )-1(I - e- TB )
l:
k-r-l
e-(tk+r-l-t;)B
i=l
e- 2tNB )-1(1 - e- TB )
k-r-l
{e-(tk+T-1-t;)B(I _ e-2tN-k-rB)
i=l
_e-(tk-1-t;)B(I - e-2tN-kB)}(I - e- 2t ;B)(I
Pf + Pf + P~ + pt
+ e-TB)-l(Ji -
fk)
253
Now let us estimate Pl. Using estimates (5.8) and (5.9), we obtain that
II
k+r-l
plllE:S
II(I -
11(1 - e- 2tNB )-11IE--+E
e-rB)e-(tk+r-l-ti)B(I - e-2tN-k-rB)IIE--+E
i=k-r
xiiI -
e- 2tiB
1
IIE--+E II(I +e-rB)-lIIE--+EIIJi k+r-2
:S M[ra + ifr
(tk+r
k+r-l
:S Mdifr
(tk+r
!k+r
liE
~ tiP-alii r IIC;'(E)
~ tiP-alii r
II
c ;'(E)'
Since
it follows that
II
PlIIE:S
~l (2q a l r Ilc~(E)'
Since
P~
x(I -
= (I - e-2tNB)-le-2t2rB(I - e-tk-r-lB)
e- 2tiB )(I - e-2tN-k-rB)(I + e-rB)-l(Jk - !k+r),
it follows that
II
xiiI -
P~
IIE:S II(I -
e-2tNB)-11IE--+Elle-2t2rB(I - e-tk-r-lB)IIE--+E
e- 2tiB IIE--+EIII - e-2tN-k-rBIIE--+EII(I + e-rB)-lIIE--+Ellfk - !k+rIIE.
254
Now let us estimate
k-r-l
X
{II
P:' Using estimates (5.8), (5.9) and (5.10), we obtain
(I - e- rB )(e-(2t N -2t k )B - e-(2tN-2tk-2tr)B)e-(tk-l-t;jB
IIE--+E
i=l
+ II
xiiI -
(I - e-rB)(I -
e-tr)B)e-(tk-1-t;jB IIE--+E }
e- 2tiB IIE--+EII{I + e-rB)-lIlE--+E II

k-r-l
::; Mtr[
(tk _ :i)2- a JIl
Ii - Ik
liE
IIC~(E)'
Since for k > r we have the bound
it follows that
II
pt liE::; 1 ~a t~11 r IIc~(E)'
Thus, we have shown that for any 2 ::; k < k + r ::; N - 1 the following inequality
holds:
II
G~+r - G~ liE::; a(~ a) t~1I r Ilc~(E)'
(5.51 )
255
a(~ a) I r
I C6 Ilc;.'(E):S
Ilc;.'(E) .
11,\1-0: Be->.BC% IIE:S
,\1-0:11(1 - e- 2tNB )-11IE--->E
k-l
L III -
e-2tN-kB IIE--->E
i=l
xiiI - e- 2tiB IIE--->E
11(1 + e- TB )-lIIE--->EiI(1 -
e-TB)Be-(tk-l-ti+>.)B(!i - !k)
k-l
:S M,\l-O:[; (tk
_ ti:
,\)2-0:]11
liE
r Ilc;.'(E)
k
ds
j
o
t
Since
(tk -
+ ,\)2-0:'
tk
ds
(tk -
+ ,\)2-0:
it follows that
,\l-O:Be->.BC%
for all A, A > 0 and for 2 :S k
:S N
IIE:S
< _1_,\-1+0:
- 1- a
M
1- a
II r
'
Ilc;.'(E)
- 1. So, we have established that
IIC~IIC~(Ea) :S 1 ~ a I r Ilc;.'(E) .
IIC~IIC~(Ea) :S 1 ~ a II r
IIc;.'(E) .
Combining the estimates for C;", m = 1,2,3,4,5,6 we obtain the estimates
II
2
{r- (uk+l - 2Uk
+ uk_t}}f-
Ilc;.'(E)
:S
M
a(l ~ a) II
r IIc;.'(E)
+_1 [II (1 - e-TB)2r-2uo - II liE + 11(1 - e-TB)2r-2uN - !N-lilE ],

a
II
2
{r- (Uk+l - 2Uk
+ uk_t}}f- IIC~(Ea) :S
M
a(l ~ a)
II
r II c ;.'(E)
l
TB
2u
+M
a [II (1 - e- ?r- o - h liE a + 11(1 - e-TB)2r-2uN - IN-lilE ],
c<
256
f-
The estimate for {r- 2[(1 - exp( -rB))(Uk-1 + Uk+l) + (exp(- 2rB) - I) Uk]} 1
in the norm of Cr(Ec,,) follows from the triangle inequality. Theorem 5.2.11 is
proved.
the following fact.
II {r- 2(uk+1 - 2Uk + Uk-1)}f- 1 II c-?(E)
+ II {r- 2[(1 -
exp( -rB))(uk_1
+ Uk+1) + (exp( -2rB) - I)uk]}f- 1 II c-?(E)
+ II {r- 2(uk+1 - 2Uk + Uk-1nf- 1IlcT(E",)

::;
(M1 )11 (1 +exp(-rB))-lr Ilc"'(E)

a1-a
T
+ M 1[ II (1 - e-rB)2r-2uo - II liE + 11(1 - e-rB)2r-2uN - !N-1IIE ],

a
'"
'"
uo, UN, a, but also of r.
r,

II {r- 2 (uk+l - 2Uk + Uk_1)}i"-1 Ilc-?(E)
+ II {r- 2[(1 -
exp( -rB))(uk-1
+ Uk+1) + (exp( -2rB) - I)uk]}f- 1 II c-?(E)
+ II {r- 2(uk+l - 2Uk + Uk_1)}f- 1 IlcT(E",)

::;
a(~ a) II Br Ilc-?(E) + ~1 [II (1 +11(1 -
e-rB)2r-2uo -
II liE",
e-rB)2r-2uN - !N-1IIEJ,
r, uo, UN, a, but also ofr.

solutions of the difference problem (5.29) in C~,'Y(E) (0::; 'Y::; 13,0 < 13 < 1) obey
the coercivity inequalities
II {r-2(uk+l-2uk+uk-I}}f-11IcT(Ei3_-Y)::;
j3(~j3)11 r IIc~'-Y(E)
+MIl II (1 - e-rB)2r-2uo - II IIEi3_-Y + 11(1 - e-rB)2r-2uN - fN-1IIEi3_J,
II {r- 2(Uk+l
+ II
<
+ uk_d}f- 1 Ilce'Y(E)
- 2Uk
2
{r- [(1 - exp(-rB))(Uk_l
+ uk+d + (exp(-2rB)
{3(~ /3) I r IIce'Y(E) + Mdl(1 +1(1 -
257
- I)uk]}f-
e-rB)2r-2uo -
Ilce.'Y(E)
1I16''Y
e-rB)2r-2uN - fN-llg,'Y],
where M 1 is independent not only of r, uo, UN, {3, "f but also ofr.
solutions of the difference problem (5.29) in Cf,'Y(E) (0::; "f ::; {3, 0 < {3 < 1) obey
2
{r- (Uk+l - 2Uk
::;
{3(~ /3) I (1 + exp( -rB))-l r IIce'Y(E)
+Mdll (1 - e-rB)2r-2uo - II
2
{r- [(1 - exp( -rB))(uk-l
<
IIE,.,_'Y + 11(1 -
2
{r- (uk+l - 2Uk
I
+ II
+ uk-d}f- 1 IlcT(E,.,_'Y)
e-rB)2r-2uN - fN-IIIE,.,_...,],
+ uk-d}f- 1 IIce.'Y(E)
+ uk+d + (exp( -2rB) - I)uk]}f- 1 Ilce.'Y(E)
{3(~ {3) I (1 + exp( -rB))-l r IIce'Y(E)

+M1 [1(1 -
e-rB)2r-2uo -
1I16''Y + 1(1 -
e-rB)2r-2uN - fN_llg,'Y],
r, uo, UN, {3, 'Y but also ofr.
solutions of the difference problem (5.29) in Cf,'Y(E)(O ::; "f ::; {3,O < {3 < 1) obey
2
{r- (uk+l - 2Uk
+ uk-d}f- IlcT(E,.,_'Y) ::;
+M1[1I (1 - e-rB)2r-2uo - II
IIE,.,_'Y + 11(1 -
{3(1
~ {3) I
Br
Ilce'Y(E)
e-rB)2r-2uN - fN-IIIE,.,_'Y],
2
{r- (uk+l - 2Uk
+ Uk_l)}f- 1 IIce'Y(E)
+ II {r- 2[(1 - exp( -rB))(uk-l + Uk+d + (exp( -2rB) - I)uk]}f- 1 IIce.'Y(E)
<
{3(~ /3) II Br IIce'Y(E) + M 1 [I (1 +1(1 -
e-rB)2r-2uo -
1I16''Y
e-rB)2r-2uN - fN_llg,'Y],
r, uo, UN, {3, 'Y but also of r.
258
Here,
Iwlg,')' denotes that the norm of the Banach space Eg,')' consists of those
vEE for which the norm
max
l~n~N-l
Ile-tnBwllE
sup
l~n<n+r~N-l
t:;f3(t n + tr)'Y(tN - tn)'Yll(e-(tnHr)B - e-tnB)wIIE
is finite.
number of spaces of grid functions in which difference problem (5.29) is well posed.
Recall that the difference problem (5.29) is not well posed in GT(E) for all
positive operators. It turns out that a Banach space E can be restricted to a
Banach space E' in such a manner that the restricted problem (5.29) in E' will be
well posed in GT(E'). The role of E'will be played here by the fractional spaces
E a = Ea(E,B) (0 < 0: < 1).
solutions of the difference problem (5.29) in GT(Ea ) obey the coercivity inequality
I {r- 2(uk+l - 2Uk + Uk_l)}f- 1 II (E",)
Cr
+ II {r- 2[(1 - exp( -rB))(uk-l + Uk+l) + (exp( -2rB) - I)uknf- 1 IICr(E",)
~ M1[o:(1 ~ 0:) I r I Cr (E",) + I Auoll E", + II AuNII E J,

where M 1 is independent not only of fr, UO, UN,
r- 2(uk+l - 2Uk
=
0:,
but also ofr.
+ Uk-I)
(1 - e- 2tNB )-1(1 - e-TB)2r-2{(e-tk-IB - e-(2t N-t k+d B)uo

+(e-(tN-tk+JlB _ e-(tN+tk-dB)UN}
-(I - e- 2tNB )-1(1 _ e-TB)(e-(tN-tk+dB _ e-(tNHk-d B )
N-l
L (e-(tN-ti)B - e-(t NHi)B)(1 + e-TB)-l Ii

i=l
+(1 - e- TB )
L e-(t k- 1-t;)B(1 + e-TB)-l Ii
k-l
i=l
+(1 - e- TB )
N-l
e-(t i -t k+d B(1 + e-TB)-l Ii
2(1 + e-TB)-l Ik
i=k+l
-(1 - e- TB )
N-l
i=l
e-(tk-1H;)B)(I + e-TB)-l Ii = 9k
+ Pk,
259
where
+(e-(tN-tk+IlB _ e-(tN+tk-IlB)uN},
Pk =
-(I -
e- 2tNB )-I(1 - e-TB)(e-(tN-tk+IlB _ e-(tNHk-IlB)
N-I
L (e-(tN-ti)B -
e-(t N+t i )B)(1
+ e-TB)-I Ii
i=1
+(1 -
e- TB )
L e-(t k- 1-t i )B(1 + e-TB)-I Ii
k-I
i=1
+(1 -
e- TB )
N-I
e-(t i -t k+ Il B(1
+ e-TB)-l Ii -
2(1 + e-TB)-l Ik
i=k+1
-(1 -
e- TB )
N-I
e-(t k- 1+t;JB)(1
+ e-TB)-I Ii-
i=1
II gTIICT(E a ) ~
I pTIICT(E
MIlII Auoll Eo + AUNIIEJ,
~ M1a-I(1- a)-III
IICT(E
(5.52)
)'
(5.53)
and the fact that the homogeneous difference equation (r = 0) is stable in CT(Eo )
uniformly in T. By the definition of the spaces Eo and using estimates (5.10), (5.11)
I .x 1- o Be->.B pk
N-I
X
IIE~
.x 1 - o ll(1 -
e- 2tNB )-IIIE-+E
Ile- TB - e-(tN+1Hi)BIIE-+E
i=1
+.x
k-I
I- o
i=1
11(1 + e- TB )-IIIE-+EIIB(1 -
e-TB)e-(tk-l-ti+>.)B fillE
260
+A 1 - o
N-l
I:
11(1 + e- TB )-lIIE--->EIIB(1 - e-TB)e-(t;-tk+JlB fillE
i=k+l
+A 1- o
N-l
I: Ile-
tkB
IIE--->EI/(1 + e-TB)-ll/E--->EI/B(I - e-TB)e-(ti-l+A)B fillE
i=l
N-l
:S M(l
+ AI-O[~
(tj-l +TA)A 1- o ])11
II cT (E,,)
T
J
o
Since
ds
(s + A)Sl-o
J
o
J
00
ds
1
(s + A)Sl-o :S A1 - o
dr
(1 + r)r 1 -
o '
it follows that
for all A, A > 0 and for 1 :S k :S N - 1. From the last estimate follows estimate
(5.53). Theorem 5.2.17 is proved.
the following fact.
solutions of the difference problem (5.29) in CT(Eo ) obey the coercivity inequality
+ II {T- 2 [(1 - exp( -TB))(Uk_l + uk+d + (exp( -2TB) - I)Uk]}f-l II cT (E,,)
:S M1 [a(1 ~ a) II (1 + exp( -TB))-l r IICT(E,,) +

II
Auoll E" + I/AuNIIEJ,
r, Uo, UN, a, but also ofT.
261
solutions of the difference problem (5.29) in CT(Eo ) obey the coercivity inequality
II {T- 2(Uk+l - 2Uk + Uk_l)}f- 1 IlcT(E,,)

+ II {T- 2[(1 - exp( -TB))(Uk-l + Uk+l) + (exp(-2TB) - I)Uk]}f-l IlcT(E,,)
:S M1[a(1 ~ a) II Br IlcT(E,,) + II Auoll E" + IIAuNIIEJ,

r, Uo, UN, a, but also of T.

solutions of the difference problem (5.29) in Ce,'"Y(Eo -f3) (0 :S , :S (3 :S a,O < a <
1) obey the coercivity inequalities
II
1
2
{T- (Uk+l - 2Uk + Uk_l)}f- Ilc~''''(E,,-{3)
+ II {r- 2[(1 - exp( -TB))(Uk-l + Uk+l) + (exp( -2TB) - I)uk]}f- 1 Ilc~''''(E,,_{3)

+
II {r- 2(uk+l
:S
1
- 2Uk + Uk_l)}f- IlcT(E~:!{3)
a(~ a) II r Ilc~''''(E,,-{3)
+Mdl(1 - e-TB)2r-2uo - hl~~f3 +
II
1(1 -
e-TB)2T-2uN - fN-11~:2'f3],
1
2
{r- (uk+l - 2Uk + Uk_l)}f- Ilc~''''(E,,-{3)
+ II {T- 2 [(I - exp( -TB))(Uk-l + Uk+l) + (exp( -2TB) - I)Uk]}i'-l
IIc~''''(E,,-{3)
+ II {T- 2(Uk+l - 2Uk + Uk_l)}i'-l IlcT(E,,_..,)

:S
a(~ a) [II r Ilc~''''(E,,-{3)
+M1 [11(1 - e-TB)2T-2uo -
hII E,,_.., + 11(1 -
e-TB)2T-2uN - fN-l IIE,,_..,]] ,
r, Uo, UN, a, (3, ,
but also of T.
solutions of the difference problem (5.29) in Ce,'"Y(Eo - f3 ) (0 :S , :S (3 :S a, 0 < a <
II
+
II {T- 2[(1 -
2
{T- (Uk+l - 2Uk
+ Uk_l)}f- 1 Ilc~''''(E,,-{3)
1
exp( -TB))(Uk-l + uk+d + (exp( -2TB) - I)uk]}f- Ilc~''''(E,,-{3)
+ II {T- 2(Uk+l - 2Uk + Uk_l)}f- 1 IIcT(E~:!{3)
262
::;
a(~a)1I
(1 +exp(-TB))-lr
IIc~''''(E<>-{3)
Y + 1(1 - e-rB)2 T-2 UN - f N-l 1{3,'

+M 1 [1(1 - e-rB)2 T-2 Uo - f 1 1{3,'
o.-{3
o.-{3Y] ,
II {T- 2(Uk+l - 2Uk + Uk-lnf- 1 IIc~''''(E<>-{3)
+ II {T- 2[(1 - exp( -rB))(Uk-l + uk+d + (exp(-2rB) - I)uk]}f- 1 IIc~''''(E<>-{3)
+ II {r- 2(Uk+l - 2Uk + Uk-lnf- 1 Ilo.r(E<>_..,)
::;
a(~ a) [II (1 + exp( -TB))-l r Ilc~''''(E<>_{3)
+M1 [l1(1 - e-rB)2r-2uo - h IICT(E<>_..,) + 11(1 - e- rB fT- 2UN - fN-r1lcT(E<>_..,)]],

r, Uo, UN, a, (3, 'Y but also ofT.
solutions of the difference problem (5.29) in C~,'Y(Eo._{3) (0::; 'Y::; (3::; a,O < a <
+ II {T- 2 [(I - exp( -TB))(Uk_l + Uk+!) + (exp( -2TB) - I)uk]}f- 1
IIc~''''(E<>-{3)
+ II {T- 2(Uk+l - 2Uk + Uk_l)}f- 1 IlcT(E~~{3)

::;
a(~ a) II Br IIc~''''(E<>-{3)
+M1 [! (1 - e-rB)2T-2uo - hl~'2{3 + 1(1 - e- rB fT- 2UN - fN-ll~:2{3],

II {T- 2(Uk+l - 2Uk + Uk-lnf- 1 Ilc~''''(E<>-{3)
+ II {T- 2[(1 - exp( -TB))(Uk_l + Uk+l) + (exp( -2rB) - I)uk]}f- 1 Ilc~''''(E<>-{3)
+ II {T- 2(Uk+! - 2Uk + Uk-lnf- 1 IlcT(E<>_..,)
::;
a(~ a) [II Br IIc~''''(E<>-{3)
+Mrlll(1 - e- rB fr- 2uo - hllcT(E<>_..,) + 11(1 - e-rB)2T-2uN - fN-lllcT(E a _..,)]],

r, Uo, UN, a, (3, 'Y but also of T.
263
Here, Iwl~:213 denotes that the norm of the Banach space E~:213 consists of
those vEE for which the norm
Iwl 13 :2 = max Ile-tnBwIIE_
13 l~n~N-I
" iJ
Q
sup
l~n<n+r~N-I
t;13(t n + tr)'Y(tN - tnPII(e-(tnHr)B - e-tnB)wIIE"_iJ
is finite.
Note that the spaces of grid functions C~,'Y(EQ_13)' in which coercive solvability has been established, depend on the parameters a, (3 and 'Y. However, the
Let us consider now the boundary-value difference problem (5.29) in the
spaces Lp,T(E) = Lp([O, T]T1 E), 1 ::; p < 00 of all grid functions. We have not
been able to obtain the coercivity inequality
2
II {7- (Uk+l - 2Uk
+ II {7- 2[(1 -
+ Uk_I)}~-1
IILp,T(E)
+ uk+d + (exp( -27B) - I)uk]}~-I

IILp,T(E) + II Auoil E + IIAuNII E],
exp( -7B))(Uk-1
IIL p,T(E)
::; M[II
in the arbitrary Banach space E and for the general positive operator A. Nevertheless, we can establish the almost coercivity inequality.
solutions of the difference problem (5.29) in Lp,T(E) obey the almost coercive inequality
2
II {7- (Uk+1 - 2Uk + Uk_I)}~-1 II Lp ,T(E)
+ II
{7- 2[(1 - exp( -TB))(Uk_1

::;
+ Uk+l) + (exp( -27B) MIlil Auoil E + IIAuNll E
+ min {ln~, 1 + lIn II
B IIE-->EI} II
where M I does not depend on fT, uo, UN, P and
I)uk]}~-I II Lp ,T(E)
IILp,T(E)],
7.
Proof. The proof of this theorem in the case p = 1 is carried out according to
the same scheme of Theorem 5.2.5 and is based on estimate (5.33). So, we will
consider the proof of this theorem for any p, 1 < p < 00. By formula (5.37) we
have that
2
7- (Uk+1 - 2Uk
+ Uk-I)
= gk
+ Pk
Therefore, to prove the theorem it suffices to show that

II gTIILp,T(E) ::; MIlil Auoil E + IIAuNII E],
II pT II LpT (E) ::; MI min
{ln~, 1 + lIn II B
(5.54)
IIE-->EI} II
IILp.T(E)'
(5.55)
264
and the fact that the homogeneous difference equation (r = 0) is stable in Lp,r(E)
uniformly in T. We have that
1
Pk = Pk
where
3
+ Pk2 + Pk,
k-l
Pk = (I - e- 2tNB )-1(I - e- rB ) I>-(tk-1-ti)B
i=l
x (I - e- 2tiB )(I - e-2tN-kB)(I + e- rB )-l J;, 2:::; k :::; N - 1,pt = 0,

p~
= -(I - e- 2tNB )-1(I - e-rB)(e-(tN-tk+tlB _ e-(tNHk-tl B )
x (e-(tN-tk)B - e-(tN+tklB)(I + e- rB )-l fk - 2(I + e-rB)-l fk

-(I - e-rB)e-(tk-1Hk)B)(I + e- rB )-l fk,
p~ = (I - e- 2tNB )-1(I - e- rB )
N-1
e-(ti-tk+tl B
i=k+l
x (1
- e- 2tkB )(1 - e-2tN-iB)(I + e-rB)-l J;, 1 :S k :S N - 2,P}._1
Let us estimate {Pk}f=-/ for any m

(5.11) and (5.36), we obtain
Ilp~IIE
:::; 11(1 -
O.
= 1,2,3 separately. Using estimates (5.10),
e- 2tNB )-1IlE--+EIII - e-2tN-kBIIE--+E
k-l
X
L III - e- 2tiB IIE--+EII(I - e-rB)e-(tk-l-ti)BIIE--+E

i=l
k-l
xll(I + e- rB )-lIIE--+EllfiIIE :::; ML 11(1 -
e-rB)e-(tk-l-ti)BIIE--+EIIJiIIE
i=l
k-l
= M
L 11(1 - e-rB)e-tj-1BIIE--+Ellfk_jIIE, 2:::; k:::; N-l.

j=l
Set 'Pi = Ii if i = 1, ... , k - 1, and 'Pi = 0 otherwise. We have that
IlplllE :::; M
N-l
j=l
11(1 - e-rB)e-tj-1 BIIE--+ElliI'k-j liE, 2:::; k :::; N - l.
265
Using this estimate and Minkowski sum inequality it follows that

II
{pD~~/IILp.T(E)
N-1
(L II Pk II~T)~
k=2
N-1 N-1
:S
L (L (MII(I -
e-rB)e-tj-lBIIE--.E1lcpk_jIlE)PT)
j=l k=2
:S
N-1
N-1
j=l
i=l
L Mil (1 - e-rB)e-tj-lBIIE--.E( L Ilcpill~T)t.
By the definition of the grid function {'I'd ~,:/ we obtain that

N-1
II
{pD~':/IILp'T(E) :S
L MII(1 -
N-1
e-rB)e-tj-lBIIE--.e(
j=l
L 1I1i1l~T)t.
i=l
Using this estimate and estimate (5.33), we obtain that

II
{pD~:/IILp'T(E)
:S
M 1 min
{ln~, 1 + Iln II B IIE--.EI} II r
IILp,T(E)'
Let us estimate {pD~':/. Using estimates (5.10), (5.11) and (5.36), we obtain
IIp~IIE
:S [11(1 -
e- 2tNB )-1I1E--.EIII - e-rBIIE--.E
x lIe-(t N-t k+d B
- e-(tNHk-l)BIIE--->E
+ e- rB )-lI1E--.E + 211(1 + e- rB )-lIIE--.E

+11(1 - e-rB)e-t2k-lBIIE--.EII(1 + e- rB )-lIIE--->dllikIIE :S MllfkllE.
x lIe-(t N-t k )B - e-(t N+t k)BIIE--.EII(1

N-1
II
{pn~=~lI1Lp,T(E):S M(I: Ilikll~T)t.

k=l
Let us estimate {pD~':/. Using estimates (5.10), (5.11) and (5.36), we obtain that
Ilp~IIE :S
N-1
III -
11(1 -
e-
2tNB
)-11IE--.EIII - e-
2tkB
IIE--.E
e-2tN-iBIIE--.EII(1 - e-rB)e-(ti-tk+dBIIE--.E
i=k+1
x 11(1 + e- rB )-lIIE--.EllfiIIE :S M
N-1
11(1 -
e-rB)e-(ti-tk+dBIIE--.EllfiIIE
i=k+1
=M
N-k-1
j=l
11(1 -
e-rB)e-tj-lBIIE--.Ellfk+jIIE, 1
:S k :S N - 2.
266
Set 'Pi = fi if i = k + 1, ... , N - 1, and 'Pi = 0 otherwise. We have that
N-1
IIp~IIE :::; M
2: 11(1 -
e-TB)e-tj-IBIIE--..EII'Pk+jIIE, 1:::; k :::; N - 2.
j=l
Using this estimate and Minkowski sum inequality it follows that
II {pD~=-/IILp'T(E)
N-1 N-2
:::; 2:(2:(MII(I j=l k=l

N-1
:::; 2: MII(I j=l
N-2
(2: II p~ 11~7)~
k=l
e-TB)e-tj-IBIIE--..EII'Pk+jIIE)P7)~
N-1
e-TB)e-tj-IBIIE--..E( 2: II'Pill~7)~.
i=l
By the definition of the grid function {'Pd~=-/ we obtain that
II {pD~';/IILp'T(E)
N-1
:::;
2: MII(I j=l
N-1
e-TB)e-tj-IBIIE--..e( 2: IIJiII~7)~.
i=l
Using this estimate and estimate (5,33), we obtain that
Combining the estimates for {Pk} ~=l\ m = 1, 2, 3 and using the triangle
inequality we obtain estimate (5.39). Theorem 5.2.23 is proved.
the following facts.
solutions of the difference problem (5.29) in Lp,T(E) obey the almost coercive in-
equality
I {7- 2(Uk+1
+ I {7- 2[(I -
- 2Uk
exp( -7B))(Uk-l
+ Uk_1)}~-1 IILp.T(E)
+ Uk+1) + (exp( -27B) -
I)uk]}~-l
IILp,T(E)
:::; Mdll Auoil E + IIAuNll E

+ min {ln~, 1 + jln I B IIE--..EI} I (I + exp( -7B))-1
where M 1 does not depend on fT, uo, UN, P and
7.
r IILp,T(E)],
267
solutions of the difference problem (5.29) in Lp,T(E) obey the almost coercive in-
equality
+ II
II
{r- (uk+l - 2Uk
+ Uk_l)}~-l
{r- 2 [(1 -exp(-rB))(uk-l +Uk+l)
IILp,r(E)
+ (exp(-2rB)
- I)uk]}f- 1 IILp,r(E)
:::; Mdll Auoll E + IIAuNllE

+ min
{ln~, 1 + Iln II B IIE---->EI} II Br IILp,r(E)],
where M 1 does not depend on fT, uo, UN, p and r.

of the boundary-value difference problem (5.29) in the spaces
Theorem 5.2.26. Suppose that the assumptions of Theorem 5.2.5 hold. Suppose that
the difference problem (5.29) is well posed in LpO,T(E) for some Po, 1 < Po < 00.
Then it is well posed in Lp,T(E) for any p, 1 < p < 00 and the following coercivity
inequality holds:
+ II {r- 2 [(1 - exp( -rB))(uk-l + uk+d + (exp( -2rB) - I)uk]}f- 1 IILp,r(E)

2
+lluTIICc(El-l/P,p) :::; M(poHII uoIIE,_,/P p + lI u NIIE , _, / p p + --.!!.....-l

,
,
p-
where M (Po) does not depend on r, uo, UN, P and
II
r II
Lp,r(E)
],
T.
Then it is well posed in Lp,T(E) for any p, 1 < P < 00 and the following coercivity
inequality holds:
2
II {r- (uk+l - 2Uk
+ uk-d}f- 1 II Lp ,r(E)
+ II {r- 2 [(1 - exp( -rB))(uk-l + uk+d + (exp( -2rB) - I)uk]}f- 1 IILp,r(E)

+lluTllcc(E,_,/P,p) :::; M(po)[11 UOIIE,_,/p,P +
2
II U NIIE , _, / p
+ -P-II (1 +exp(-rB))-lr II
],
p- 1
Lp,r(E)
r, Uo, UN, P and
T.
,P
268
Then it is well posed in Lp,T(E) for any p, 1 < p < 00 and the following coercivity
inequality holds:
II
+ II
{T- (Uk+l - 2Uk
{T- 2 [(I - exp( -TB))(Uk-l
+ Uk-lnf- 1 IILp,T(E)
+ Uk+l) + (exp( -2TB) -
I)uk]}f- 1 IILp,T(E)
2
+ L111
+lluTIICc(EI_I/P,P)::; M(po)[11 uoIIEl_l/P ,p+IluNIIE1 _ 1 / P'Pp
Br II Lp,T(E) ],
r, uo, UN, P and
T.
Theorem 5.2.29. Let 1 ::; p ::; 00 and 0 < Q < 1. Suppose that the assumptions
of Theorem 5.2.5 hold. Then problem (5.29) is well posed in Lp,T(Eo:,p) and the
II {T- 2 (Uk+l
+ II
- 2Uk
{T- 2 [(1 - exp( -TB))(Uk_l

::; M[II
+ uk-d}f- 1 IILp,T(Eo,p)
+ uk+d + (exp( -2TB) -
AUoIIEo,p + IIAuNIIEo,p + Q(l ~ Q) II r II
where M does not depend on fT, Uo, UN, p,
IILp,T(Eo,p)
1)Uk]}f-l
Lp,T(Eo,p)
],
and T.
Theorem 5.2.30. Let 1 ::; p ::; 00 and a < Q < 1. Suppose that the assumptions
II
+ II
{T- (Uk+l - 2Uk
{T- 2 [(1 - exp( -TB))(Uk_l
::; M[II
+ Uk-lnf- 1 IILp,T(Eo,p)
+ uk+d + (exp( -2TB) -
AuollEo,p + IIAuNIIEo,p + Q(l ~ Q) II
1)Uk]}f-l
(I + exp( -TB))-l
IILp,T(Eo,p)
r II Lp,T(Eo,p) ],
where M does not depend on fT,uo, UN, P,Q and T.

Theorem 5.2.31. Let 1 ::; p ::; 00 and 0 < Q < 1. Suppose that the assumptions
II
+ II
{T- (Uk+l - 2Uk
{T- 2 [(I - exp( -TB))(Uk-l
+ Uk-lnf- 1 IILp,T(Eo,p)
+ uk+d + (exp( -2TB) 1
::; M[II AuollEo,p + IIAuNIIEo,p + Q(l _

where M does not depend on fT, Uo, UN, p,
Q)
and T.
1)Uk]}f- 1
II Br II
Lp,T(Eo,p)
IILp,T(Eo,p)
],
269
Theorem 5.2.32. Let 1 < p, q < 00 and 0 < Q < 1. Suppose that the assumptions of
Theorem 5.2.5 hold. Then the difference problem (5.29) is well posed in Lp,r(Ea,q)
II
+ II {r- 2 [(I -
{r- (uk+1 - 2Uk
+ Uk-1nf- 1IILp,T(E",q)
+ Uk+!) + (exp( -2rB) - I)uk]}f- 1 II Lp ,T(E",q)
exp( -rB))(uk-1
:s M(q)[11 AUoIIE",q + IIAuNIIE".q + p~ 1 II r
r, uo, UN, p, q,
IILp'T(E,,)'
and r.
2
II {r- (uk+! - 2Uk
+ II {r- 2 [(I -
+ uk-d}f- 1 II Lp .T(E",q)
+ Uk+!) + (exp( -2rB) - I)uk]}f- 1 IILp,T(E",q)
exp( -rB))(uk-1
:S M(q)[11 Auoil E + IIAuNIIE" + -p- II (I + exp( -rB))-l r

",q
,q
p- 1
r, uo, UN, p,
II L
p,T
(E
o,q
],
and r.
2
II {r- (uk+! - 2Uk
+ II {r- 2 [(I -
exp( -rB))(uk-1
+ uk_d}f- 1
+ Uk+1) + (exp( -2rB) - I)uk]}{"-l

p2
:S M(q)[11 AUoIIE",q + IIAuNIIE",q + P -1

IILp,T(E",q)
r, uo, UN, p,
II Br
IILp,T(E",q)
IILp'T(E,,)'
and r.
Now, we will consider the applications of the exact difference scheme (5.29).
From (5.29) it is clear that for the approximate solutions of the problem (5.1) it
is necessary to approximate the expressions exp( -r B) and
1jt
exp( -(tk - z)B))f(z)dz,
tk-l
11
tk
tk-l
exp( -(z - tk-1)B)f(z)dz.
270
Let us remark that in constructin difference schemes it is important to know

how to construct a right-hand side fk' that satisfies
11
11
tk
(2B7)-1 [-
exp( -(tk -z)B))f(z )dz+ -
tk-l
11
7
tk
-exp(-7B)(-
11
+7
tk
tk
+ exp( -(z-tk)B)f(z)dz (5.56)
tk
exp(-(tk+l - z)B))f(z)dz
tk
exp( -(z - tk-dB)f(z)dz)]- fe = 0(7 j +l )
tk-l
and is a sufficiently simple. The choice formula fe is not unique. Using Taylor's
formula, with respect to function exp( -(tk - z)B))f(z) on variable z, we obtain
11
= f(tk) +
tk
exp( -(tk - z)B))f(z)dz
tk-l
l+j
I::
I::
m
m=l)..=O
B m->' f(>')(tk) -
l)m m+1
7
(m + I)!
+ 0(7 1+1+ 1).
Using Taylor's formula, with respect to function exp(-(z - tk-1)B)f(z) on

variable z, we obtain
1jt
-:;.
exp( -(z - tk-1)B)f(z)dz
tk-l
= f(tk-1) +
I+j
l;(;
m
m+1
~ (_l)m->. Bm->'f(>')(tk_d (~+ I)! + 0(71+ 1+1).
Further, using the last formulas and Pade fractions for the function e- z , we
can write
271
Now, using the last formula and Pade fractions for the function e- z , we
obtain the difference schemes of (j + l)-order of accuracy
(5.57)
= fk''I ,1 :S k :S N -
1, Uo = Vo, UN = VT
for the approximate solutions of the boundary-value problem (5.1).

Note that the difference schemes (5.57) for j = land j = 1 - 1 include
difference schemes of arbitrary high order of approximation. Moreover, the corresponding functions Rj,l+ 1 (z) tend to 0 as z -+ 00 for j = 1- 1, l. Such difference
schemes are simplest, in the sense that the degrees of the denominators of the
corresponding Pade approximants of the function exp{ - z} are minimal for a fixed
order of approximation of the difference schemes.
It is clear that this problem is uniquely solvable, and the following formula
holds:
Uk = (I - R;,f+1 (1'B))-1 {(R7,l+1 (1'B) - RJ.f+~k(1'B))uo
(5.58)
+(Rfl~~(1'B) - Rfl~~(1'B))UN
N-1
-(Rfl~~(1'B) - Rfl~~ (1'B))1'2
L (Rfl~i1 (1'B)
i=l
N-1
+1'2
L (R;~I~i~ (1' B) -
R~,t~l (1' B))(1 - R;,l+l (1' B) )-1 Jl'1, 1 :S k :S N - 1.
i=l
Let us reduce such difference schemes to an operator problem in the space FT(E).
In addition to the operator D;, acting from the space E x FT(E) x E of vectors
wT = {wdi:'=o into the space Fr(E) of vectors v T = {Vk}~=-/ by the rule
define an operator A~,I from the space E x Fr(E) x E of vectors wr

into the space Fr (E) of vectors vT = {vd ~=~1 by the rule
+(R],I+1 (1' B) - I)WkJ, k = 1, ... , N - 1.
= {wd~=o
272
operator equation
-D;II(uo, UN )UT + A~lII(uo, UN)U T = fIl'

Here fIl is defined by the formula
T - (fj,l
fj,l)
f j,l
- 1 ' " ' ' N-l .
solvability for any Un, UN E E and fIl E FT(E) it follows that its solution uT
defines a continuous additive and homogeneous operator uT(fIl' un, UN).
the inequality
where M is independent not only of fIl' un, UN, but also of r.

From formula (5.58) one derives the following result.
Theorem 5.2.35. Let A be the positive operator in a Banach space E. Then the difference problem (5.57) is stable in Lp,T(E), 1 ::; p::; 00.( Here CT(E) = LOO,T(E).)
The proof of this theorem is based on formula (5.58) and the estimates
IIRj,l+l (rB)IIE--.E ::; M,l ::; k ::; N,
(5.59)
11(1 - R;,'iv+1(rB))-11IE--.E ::; M.
(5.60)
The proof of estimate (5.60) is based on the estimates
11(1 and
II exp( -hB) -
exp( -2NrB))-11IE--.E ::; M
Rj,l+l(rB)IIE-+E ::; Mk- j - l - 1, 1 ::; k ::; N.
(5.61)
Here M does not depend on k and r.

Theorem 5.2.36. Let A be the positive operator in a Banach space E. Then the
difference problem (5.57) is stable in c;;,a(E), 0 < a < 1.
The proof of this theorem is based on formula (5.58), estimates (5.59), (5.60)
and
IIRJ,l+l (rB) -
R;,t~l (rB)IIE-+E ::; M (k :Qr)Q'
(5.62)
for all k, r, 1 ::; k < k + r ::; N - 1 and all r > O,where M does not depend on k,
rand r. Estimate (5.62) follows from (5.59) and
kTIIBRj,l+l(rB)IIE-+E ::; M,l ::; k::; N.
(5.63)
273
II {r- 2 (uk+l - 2Uk +uk_d}f- 1 IIFT(E)
+I
1
{r- 2 [(1 - Rj,l+l(rB))(uk-l +Uk+l) + (RJ,l+l(rB) - I)uk]}f- IIFT(E)
~
M[II
fj,l IIFT(E)
+I
Auoil E + IIAuNII E],
where M is independent not only of fj,l' Uo, UN, but also of r. Since the difference
boundary-value problem (5.57) in Lp,T(E), 1 ~ P ~ 00 is not well posed for the
general positive operator A and space E, then the coercive norm
I u T IIKp,T(E) = I {r- 2 (uk+l - 2Uk + Uk_l)}f- 1 IILp.T(E)
+ I {r- 2 [(1 - Rj,l+l(rB))(uk-l + Uk+l) + (RJ,l+l(rB) - I)uk]}f- 1 IILp,T(E)

tends to
00
as r
----+
+0 . We have
Theorem 5.2.37. Let A be a strongly positive operator in a Banach space E with

spectral angle (A, E) < T and uo, UN E D(A). Then the solutions of the difference problem (5.57) in Lp,T(E) obey the almost coercive inequality
T
" u IIKp,T(E) ~ Mdll Auoil E + IIAuNll E
+ min {ln~, 1 + lIn I B IIE-+EI} I fj,l IILp,T(E)],

where M1 is independent not only of f;'l' Uo, UN, but also of r.
The proof of Theorem 5.2.37 follows the scheme of the proofs of Theorem
5.2.5 and 5.2.23, and relies on the formula
r- 2(uk+l - 2Uk + Uk-I)
(5.64)
=
(1 - R;,f+l(rB))-l(1 - Rj,l+1(rB))2r-2{(R;,1~1(rB) - R},f+--;.k-l(rB))uo

+(Rfl+~-l(rB) - Rflt~-l(rB))uN}
-(1 - R;,f+l(rB))-1(1 - Rj,l+l(rB))(Rfl+~-l(rB) - Rflt~-l(rB))

N-l
x
(Rfl+il (rB) - Rfltil (rB))(1 + Rj,l+l(rB))-l f{l
2:
i=l
k-l
+(1 - Rj,l+l(rB))
2: R;,I~li(rB)(1 + Rj,l+l (rB))-l fl,l

i=l
N-l
+(1 - Rj,l+l(rB))
2:
i=k+l
Rj"j~11(rB)(1 + Rj,I+l(rB))-l ff
274
and on estimates (5.59), (5.63) and
N-1
2: 11(1 - Rj,l+1 (7 B) )R}:Z~1 (7B) IIE-.E ::; M min ( In(1/7), 1 + lIn IIBIII), (5.65)
j=1
11(1 - Rj,I+1(7B))2(7B)-21IE---.E ::; M,
(5.66)
11(1 + R j,I+1(7B))-1I1E---.E ::; M.
(5.67)
Estimate (5.66) is obvious. The proof of estimate (5.65) follows the scheme of proof
of Lemma 5.2.4 and is based on estimates (5.59) and (5.63). The proof of estimate
(5.67) is based on the following lemma.
Lemma 5.2.38. Let B be a strongly positive operator in a Banach space E with
spectral angle ( B, E) <

one has estimate (5.67).
ii.
Then the operator 1 + Rj,l+1 (7 B) is invertible and
Proof. Since
(1
+ R j,I+1(Z))
-1
Qj,l+1(Z)
= P j,l+1 ()+Q
Z
j,I+1 ()
Z
is a rational function, it suffices to show that the function P j,l+1 (z) + Q j,l+1 (z)
has no zeros in the sector {z E C : z = pe</> , 0 ::; p < 00,0 ::; ::; ii}. Since
P j ,l+1 ( z)
+ Qj,I+1 ()
Z
~ (j + l + 1 - i)! [ (l + I)!
( ) - i j! ] i
~ (j + 1 + l)!i! (l + 1 _ i)! + -1 (j _ i)! z
1+1 (j+l+1-i)! (l+l)! i

+2: (j+l+1)!i! (l+l_i),z
t=J+1
and (l~i~):)!
+ (_l)-i (j~i)! >
0 for j = l - 1, l, we obviously have IPj,I+1(z) +

Qj,I+1(z)1 ~ 2. Lemma 5.2.38 is proved.
Banach spaces.
spectral angle (A, E) < T and uo, UN E D(A). Then the solutions of the difference problem (5.57) in C~,Q(E) obey the coercivity inequality
II
+ II
2
{7- (Uk+1 - 2Uk
2
{7- [(1 - R j,I+1(7B))(Uk-1
::; MIl a (l ~ a) II
+ uk_d}f- 1 Ilc;""(E)
+ Uk+1) + (R;'I+1(7B) - I) u knf- 1 Ilc;."'''(E)
f],l Ilc;.""(E)
+ II Auoil E + II Au NII E],
where M 1 is independent not only of 1],1' UO, UN, a, but also of 7.
275
The proof of Theorem 5.2.39 follows the scheme of proof of Theorem 5.2.8,
and relies on the formula
(5.68)
= (I - R;,f+l(rB))-l(I - Rj,I+l(rB))2r-2(R;,1~1(rB) - RJ.~+lk-l(rB))uo
+(1 - R;,f+l(rB))-l(I - Rj,I+1(rB))2r-2(Rfl+~-1(rB) - Rflt~-I(rB))uN
+{ -1 + (I - R;,f+l (rB))-l(I - Rfl+l (rB))[1 - R;,l~l (rB) - Rfl+~-l(rB)

+Rfl+l(rB)]}Ik,1 + (I - R;,f+l(rB))-l(I - R~,~+12k(rB))(1 - Rj,I+I(rB))
k-l
X
L R;,I~li(rB)(I - R;:l+l (rB))(I + Rj,l+l(rB))-IUI,1 - Ie)

i=1
N-I
X
i=k+l
Rj~~11(rB)(1 - R~,~+12i(rB))(1 + Rj,I+l(rB))-lUl,1 - Ik,l)
and on estimates (5.59), (5.60), (5.62), (5.66) and (5.67). The coercivity inequality
I {r- 2(uk+1
+ I {r- 2[(I -
Rj,I+1 (rB))(uk_1
:::; M1[a(1
- 2Uk + uk-Inf- Ilc~(E)
+ Uk+l) + (R;'I+1 (rB)
I
- I)uk]}f- Ilc~(E)
~ a) I III IIc~(E) + I Auoil E + IIAuNII E ]
fails. Nevertheless, we have the following result.

spectral angle (A, E) < Tand (I -Rj,I+1 (rB))2 r -2 uo - H,l, (I -Rj,l+l (rB))2 r -2
XUN - JiLl E E~. Then the solutions of the difference problem (5.57) in C:;(E)
obey the coercivity inequality
II {r- 2(uk+l - 2Uk +Uk-Inf- I Ilc~(E)
+ II {r- 2[(I -
Rj,l+l(rB))(uk-l
+I
:::;
2
{r- (uk+1
+ Uk+l) + (R;'I+l(rB) - I)uk]}f- 1 Ilc~(E)

I
- 2Uk +Uk-lnf- IlcT(E~)
ex(~ ex) [II lIz Ilc~(E) + I (I +II(I -
Rj,l+l (rB)fr- 2uo - J{'l
Rj,I+I(rB))2 r -2uN - f~~IIIE'],

"
IIE~
276
where M 1 is independent not only of fIl' UO, UN, a, but also of r. Here, the
Banach space E~ = E~(E, B) (0 < a < 1) consists of those vEE for which the
norm
I V IIE~ = supzQ I B(z1 + B)-IV liE + II V liE
z>o
is finite.
Proof. To prove the theorem it suffices to establish the estimates for
{r- 2(uk+l - 2Uk + uk-d}f- 1
in
C~(E)
and in
Cr(E~).
By formula (5.64), we have
r- 2(Uk+l - 2Uk + uk-d

= (1 -
RIf+l (rB))-I(RJ:l~1 (rB) -
R~,f+~k-l(rB))
x [(1 - R j ,l+l(rB))2 r -2 uo - fl,l]
+(1 - R;,f+l(rB))-I(Rfl~~-I(rB) - Rflt~-I(rB))

x [(1 - Rj,l+l(rB)fr- 2uN - f~Ll]
+(1 - R;:i+l (rB))-I[RJ,I~1 (rB) - R],f+~k-l(TB)](Jl'l - fe)

+(1 - R;,f+l(TB))-I[Rfl~~-I(TB) - Rflt~-I(TB)](Jt,l- f~~I)
+(1 - RIf+l(TB))-1(1 - R],f+~2k(TB))(I - Rj,l+I(TB))
k-l
x L,RJ,I~-;i(TB)(1 - R;:l+I(TB))(1 + Rj,l+I(TB))-I(J/,l - ft,l)
i=1
N-l
L, RjJ~~I(rB)(1 - R~,f+~2i(TB))(1 + Rj,l+l (TB))-1 (J/,l - fe)
i=k+l
where
H~ = (1 - RIf+l (TB))-I(RJ:l~1 (TB) - R~:i+~k-l(TB))
x [(1 - R j ,l+l(TB))2 T-2 uO - H,l],

2
2N (TB))-I(R N- k- 1(TB) _ R N+k-l(TB))
H k = (1 - R J,l+1
J,l+1
J,I+1
x[(I - Rj,l+l(rB))2T-2 uN - f~~I],
2N
k- 1 (B) R 2N - k- 1( B)](fj,l fj,l)
k - (1 - R j,l+l (TB))-I[R j,l+1
T - j,l+1
T
1 k '
N
N
1
4
2N
k
1
H k = (1 - RJ,l+l (TB))-I[R J,l+1
- - (TB) - R J,l+1
+k- (TB)](!j,l
k - fj,l
N-l ) ,
H3 -
277
k-l
x LR;,I';'~i(7B)(I - RJ~I+l(7B))(I + Rj ,I+1(7B))-lU!,1 -
Ie),
i=l
N-l
2i
L RjJ~11(7B)(1 - RJ,f+i (7B))(I + Rj,I+1(7B))-lU/,1 i=k+l
Ie)
The proof of estimates
II{Hk}~';/IIC;'(E) ~ a(l~ a) I Ij,l IIC;'(E)

for m = 3,4, 5,6 follows the scheme of proof of the estimates
for m = 3,4,5,6 of Theorem 5.2.11, and relies on estimates (5.59), (5.60), (5.62),
(5.66) and (5.67). Therefore to prove the theorem it suffices to establish the estimates for {Hk }f-l, m = 1,2 in C;:(E) and for {Hk}f-l, m = 1,2,3,4,5,6 in
Cr (E~). Let us estimate H::n = {Hk }~,;/ for any m = 1, 2 in C;: (E). We start
with
Using estimates (5.59), (5.60) and the definition of the spaces E~, we show
that, for 1 ~ k ~ N - 1,
Hr
IIH~IIE::;
11(1 -
RJ:i+l(7B))-11IE---.EIIR7,1~1(7B) - R}~+lk-l(TB)IIE---.E
x 11(1 - Rj ,I+l(7B))27-2 UO - fi'IIIE ~
MII(I -
Rj ,I+l(7B))27-2 uO -
Further, applying estimates (5.59) and (5.60), we show that, for 1

N-1,
IIH~
H~+rIIE ~
II(I -
H,IIIE;'.
k < k +r
R}f+l(7B))-11IE---.E
x [I I(R;,l';'l (7B) - R;,t~~1(7B))[(I - Rj,l+d7B))27-2uO - fi'1]IIE

2k r
+IIRJ,f+1 - (7 B) IIE---.E
xll(R;,I';'l (7B) - R;,t~~1(7B))[(I - Rj ,I+l(7B))27-2uO -
H,I]lleJ
~ M[II(R;,I';'l (7B) - R;,t~~1(7B))[(I - Rj,l+l (7B))2 7-2UO - fi'1]IIE'

To estimate
278
in the norm of E we use the following Cauchy-Riesz representation formula for

the operator (R~:I~1 (rB) - R~:I~t(rB))(rB) -1 (see Chapter 4):
(R~,I~1 (rB) - R~,I~t(rB))(rB) -1 y
J(
k-l ()
k-l+r())
Rj,I+1
z - Rj,l+l
Z Z-1 r B( z - rB )-1 ydz,
1
27ri
SlUS2
where 8 1 = {pei,p, 0 ::; p

obtain
Since z = pei,p, with
< oo} and 8 2
I'l/JI < fL,
{pei,p, 0 ::; p < oo}, 0 ::; 'l/J <
fL.
from the strong positivity of B it follows that
r-l
I: IR~,I~t8(z)(I - Rj,I+I(Z))Z -11::; Mmin{p-l,r}.

8=0
Hence,
J
00
II
(Rj,I+1 (rB) 00
Ml
R7,t~1 (rB))(rB) -I YIIE ::; M l
s+ 1 dsllyllEI
o
This shows that
Js+
00
rUru S-u
= t~ M l
:U:-l.U dpllyllE~
M2
S-u
-ldsllyIIE/::; t~ aI-a
(
) IlyllEI .
Q
for any 1 ::; k < k + r ::; N - 1. Thus, we have proved that
We
279
Now, let us estimate H::r, = {Hk}~=-;.l for any m = 1,2,3,4,5,6 in CT(E~) separately. We start with HI' Using estimates (5.59), (5.60) and the definition of the
spaces E~ we show that, for 1 :S k:S N - 1,
IIH~IIE~ :S
xll(1 -
11(1 -
R;,f+1(-TB))-IIIE~EIIR7,1~I(rB) - R],f+-;.k-l(rB)IIE~E
Rj ,I+I(rB))2r -2 uo - f{,lIIE~ :S
MII(1 -
Rj ,I+I(rB))2r -2 uo - Jl,IIIE~'
IIHlllcT(E~) :S MII(1 - Rj ,I+I(rB))2 r -2 uo - Jl,IIlE~'

IIH;IICT(E~) :S Mil (1 - Rj,I+1(rB))2 r -2 uN

By the definition of the spaces
that
E~
ft~II1E~'
and using estimates (5.59), (5.60), we obtain
II ,X'''B('x + B)-1 H2 IIE:S ,Xall(1 - R;,f+l (rB))-II1E~EIII -RJ,f+-;.2k(rB)IIE~E

x IIB('x + B)-1 R7,1~1 (rB) (f{,l - fk,I)IIE :S M ,Xatk_1IlfJ,dlc;,(E)
for all ,x, ,x > 0 and for 1 :S k :S N - 1. Using the estimate
IIB('x+B)-IR71~1(rB)IIE~E:SMmin{~,l},
t k-l
,
/l
we obtain that
lI,Xa B('x + B)-1 H2I1E:S M 1,Xatk_1min{~, l}lIfJIlIco(E) :S M 1I1fJIlIco(E)

/l t k-l
'T
for all A, A > 0 and for 2 :::; k :::; N - 1. So, we have established that

IIH;IICT(E~) :S M1I1fJ,lllc;'(E)'
By the definition of the spaces

we obtain that
k-l
X L
III -
E~
and using estimates (5.59), (5.60) and (5.67),
RJ,f+-;.2k(rB) IIE~E
III -
R;:I+1 (rB) IIE~E
i=1
x11(1 +Rj,l+I(rB))-II1E~EII(1
-Rj,l+1 (rB))B('x+B)-1 R7,I~li(rB)(f!,I-fe)
liE
k-l
:S M,Xa L(tk - ti)all(1 - Rj ,l+I(rB))B('x+ B)-I R7,I~li(rB) IIE~E II fJ,1 II C ;'(E)

i=1
280
for all A, A> 0 and for 2 :::; k :::; N - 1. Using the estimate
II(I - R j ,I+1(rB))B(A + B)-1 R7,I~li(rB)IIE->E :::; Mrmin{_l_,
~},
tk-. /lt k_ i
we obtain that
k-1
:::;
M2X:~[L t.1-0(~ + At-) 111

i=1
fII Ilc;'(E)"
j
t
ds
s1-0(1
+ SA) .
Since
it follows that
II
A<> B(A
+ B)-l HZ liES;
a(~ a) II 1j,1 Ilc;'(E)
for all A, A > 0 and for 2 :::; k :::; N - 1. So, we have established that
IIH;llcT(E;') :::;
a(~ a) II fII
Ilc;'(E)"
IIH;IICT(E;') :::;
a(~ a) II fII Ilc;'(E)"

'I
spectral angle (A,E) < T and (I - R j ,I+1(rB))2r -2 uo - Ii' , (1 - R j ,I+1(rB))2
xr- 2uN - f~I_1 E E~_/" Then the solutions of the difference problem (5.57) in
Ce'/'(E) (0:::; 'Y:::; {3,0 < {3 < 1) obey the coercivity inequalities
II {r- 2 (Uk+1
- 2Uk
+ uk-t}}f- IIcT(E~_oy) :::; {3(1 ~ {3) II III Ilc~'OY(E)
+M1 [11 (I - R j ,I+l(rB))2r -2 uo -
J1,l liE'{3-oy
281
+11(1 I
+ I {7- 2[(1 <
R j,I+1(7B))27 -2UN -
2
{7- (Uk+l - 2Uk
n/-lIIE
{3--y
],
+Uk_l)}f- 1 Ilc~'-Y(E)
1
R j ,I+1(7B))(Uk-l +Uk+l) + (R;'I+l(7B) - I)uk]}f- IIc~'-Y(E)
(3(~ (3) I fj,l IIc~'-Y(E) + Mdl(1 +1(1 -
R j ,I+1(7B))2 7 -2 UO -
H,ll~'''!
R j,I+l(7B))27-2 UN - f~~llg,,,!],
where M 1 is independent not only of fj,l' Uo, UN, (3, 'Y but also of7.
Here,
!wlg,"! denotes that the norm of the Banach space Eg,"! consists of those
wEE for which the norm
is finite.
number of spaces of grid functions in which difference problem (5.57) is well posed.
We have
spectral angle (A, E) < T' Then the solutions of the difference problem (5.57) in
Cr (E~) obey the coercivity inequality
I {7- 2(Uk+l
+ I {7- 2[(1 -
exp( -7B))(Uk-l
- 2Uk
+ Uk_l)}{"-l IlcT(E;')
+ uk+d + (exp( -27B) -
I)uk]}{"-l
IlcT(E;')
r IlcT(E;') + I Auoll E;, + IIAuNIIE;'],

is independent not only of r, Uo, UN, a, but also of 7.
:::; M 1[a(1 ~ a) I
where M 1

7- 2(Uk+l - 2Uk
+ uk-d
= (1 - R;,f+l(7B))-1(1 - Rj,I+1(7B)f7-2{(R;,1~1(7B) - R;,f+lk- 1(7B))uo

+(Rfl+~-1(7B) - Rfl~~-1(7B))UN}
-(1 - R;,f+l(7B))-1(1 - Rj,I+l(7B))(Rfl+~-1(7B) - Rfl~~-1(7B))
282
N-I
L (Rfl+il (rB) - Rfl~il (rB))(1
+ Rj,l+l(rB))-1 !1,l
i=1
k-I
+(1 - Rj,l+l(rB)) L R;,I~~i(rB)(1 + Rj,l+1 (rB))-1 fl,l

i=1
N-I
+(1 - Rj,l+l(rB)) L
i=k+1
Rj~~ll(rB)(1 + Rj,l+l(rB))-1 !1,l
-2(1 + Rj,l+l(rB))-1 ft,l = gk
+ Pk,
where
gk = (1 - R;,f+1 (rB))-I(1 - Rj,l+1 (rB))2r-2{(R;,I~1 (rB) - RJ,f+lk-l(rB))uo
+(Rfl+~-I(rB) - Rfl~~-I(rB))uN}'
Pk = -(I - R;,f+l(rB))-I(1 - Rj,l+l(rB))(Rfl+~-I(rB) - Rfl~~-I(rB))
N-I
x L (Rfl+il (rB) - Rfl~il (rB))(1
+ Rj,l+l(rB))-1 !1,l
i=1
k-I
+(1 - Rj,l+l(rB)) LRJ,I~~i(rB)(1+ Rj,l+l(rB))-1 ff'!

i=1
N-I
+(1 - Rj,l+l(rB)) L
i=k+1
Rj~~ll(rB)(1 + Rj,l+l(TB))-1 f f
-2(1 + Rj,l+I(TB))-1 ft,l.

II g'TllcT(E~) ::; MI[II AuolIE~ + AUNIIE~],
II p'T
IIcT(E~) ::; MIa-l(l- a)-III fj,l IIcT(E~)'
(5.69)
(5.70)
11(1 - R j ,l+I(TB)f(rB)-2I1E-tE ::; M

and the fact that the homogeneous difference equation Uj,l = 0) is stable in
C'T(E~) uniformly in T. By the definition of the spaces E~ and using estimates
(5.59), (5.60) and (5.67), we obtain
II
).,Ol B()"
+ B)-Ipk
liE::; ).,0l11(1 - R;,f+1 (TB))-IIIE-tE
x IIRfl+~-I(TB) - Rfl~~-I(rB)IIE-tE
283

N-1
i=l
IIRj,l+l(rB) - Rflt~+2i(rB)IIE->EII(1 + Rj,l+l(rB))-lIIE->E
k-1
L 11(1 +Rj,l+l (rB))-lIIE->EIIA B(A+B)-l(I -Rj,l+l (rB))R;:I~-;i(rB)I!,lIIE

Q
i=l
N-1
II(I + Rj,I+l(rB))-lIIE->E
i=k+1
xIIAQB(A
+ B)-l(I - Rj,I+l(rB))R~J~11(rB)I!,lIIE
+211(I + Rj,l+l (rB))-lIIE->EIIA B(A + B)-l Ie liE

Q
N-1
i=l
IIRJ,l+l(rB)IIE->EII(I + Rj,l+l(rB))-lIIE->E
N-1
~ M(II III IlcT(E~) + A
j=l
IIB(A + B)-l(I -
Rj,l+l(rB))R~J~l (rB)J1,lIIE)'
To estimate the last sum we use the following Cauchy-Riesz representation formula
for the operator B(A + B)-l(I - Rj,I+l(rB))RjJ~l (rB) (see Chapter 4):
A B(A + B)-l(I - Rj,I+l(rB))R~J~l (rB)J1,l
1
-2.
(A + ~ )-1(I - Rj 1+l(z))RJil~l (z)B(z - rB)-l I!,ldz
Q
ll't
'
SlUS2
= {pei'l/J, ~ p < oo} and S2 = {pei'l/J,O ~ p < oo},O ~ 'l/J <

z = pei'l/J, with I'l/JI < ft, from the strong positivity of A it follows that
where Sl
From this estimate and the estimates

1
M
.,....--...,.<-IAr
+ zl -
Ar + p'
ft.
Since
284
it follows that
N-l
N'L IIB(A + B)-I(I - Rj,I+l(rB))Rj~~1 (rB)J!,lIIE

j=1
5. M 1JOO ~
o j=1 (1
5. M1
pI-a.
(r A)a. (e)a.
+ 2apcos7/J + a2p2)-r Ar+p r
OON-l
pI-a.
I B( e + B)-1 f/,l liE dp
(rA)a.
-dpll fj,l lie (E')'

jL
o j=1 (1 + 2ap cos 7/J+ a2p2)-r Ar+p
i
1 -,
<>
Summing the geometric progression, we get
N-l
Aa.
j=1
IIB(A + B)-1(1 - Rj,I+l(rB))R~~~1 (rB)f/,IIIE
j p~:(:Ar dpll fIl IleT(E~) 5. a(~

00
< M(a, 7/J)
a) II fIl
IleT(E~r
So, it follows that
for all A, A > 0 and for 1 5. k 5. N - 1. From the last estimate follows estimate
(5.70). Theorem 5.2.42 is proved.
spectral angle ( A, E) <
and
(1 - R j,I+I(rB))2r -2 uo - ff,l,
(I - R j ,I+l(rB))2r -2 uN - f~~1
E E~_,.
Then the solutions of the difference problem (5.57) in C~"(E~_{3) (0 5. 'Y 5. (3 5.

a,O < a < 1) obey the coercivity inequalities
II
1
2
{r- (uk+l - 2Uk + Uk_l)}f- Ile~'"Y(E~_I3)
+ II {r- 2[(I - R j ,I+l(rB))(uk-l + uk+d + (R1.1+l(rB) - I) u k]}f- 1 lIe~'"Y(E~_I3)
+ II {r- 2(uk+l - 2Uk + Uk_l)}f- 1 IleT(E~~I3)

5. a
(;1 )
- a II fIl Il e l3'"Y(E'<>-{3 )
T
ff,II:~f3 +I(I -
+M1[l(1 - R j ,I+l(rB))2r -2 uo -
I
+I
2
{r- (Uk+l - 2Uk
2
{r- [(I - Rj,l+l(TB))(Uk-l
285
R j ,I+l(rB))2r -2 uN -
f;tLll~:::f3],
+ Uk-lnf- 1Ilc~'''(E~_I3)
+ uk+d + (R;'l+l (rB)
1
- I)uk]}f- Ilc~'''(E~_I3)
+ II {T- 2(Uk+l - 2Uk + uk_d}f- 1 IIcT(E~_.,)

:::;
a(~ a) [II fJ,1 Ilc~'''(E~_I3)
+Mdll(I - R j ,I+1(rB))2r -2uo - H,IIIE

a-.,
I
+11(1 -
R j,I+l(rB))2 r -2uN - f~~lIIE~_.,]],
where M 1 is independent not only of f;'l' UO, UN, a, (3, , but also OfT.
Here, Iwl~:2'f3 denotes that the norm of the Banach space E~:::f3 consists of
those wEE for which the norm
Iwl~:::f3 =
sup
l::on<n+r::oN-l
max
l::on::oN-l
IIRjl+l (rB)wIlE a _13

'
t;f3(t n + tr)1'(tN - tn)"II(Rj,i';l (TB) - Rj,I+l(TB))wIIEa _13
is finite.

spectral angle (A, E) < T' Suppose that the difference problem (5.57) is well posed
in Lpo,r(E) for some Po, 1 < Po < 00. Then it is well posed in Lp,r(E) for any p,
1 < P < 00 and the following coercivity inequality holds:
I {r- 2(uk+l
- 2Uk
+ Uk-lnf- 1 IILp.T(E)
+ II {r- 2[(I - Rj,l+l(rB))(uk-l + Uk+l) + (R;'I+l(rB) - I)uk]}f- 1 IILp.T(E)
+llurllcT(El_l/P.P) :S M(po)[I1
p2
+-1
p-
uoIIE1_I/P.P
+ IluNIIE
II hi II Lp.T(E) ],
where M(po) does not depend on f;'l' Uo, UN, P and T.
1 _ 1/
P.P
286
Theorem 5.2.45. Let 1 ::; p ::; 00 and 0 < a < 1. Suppose that A is a strongly
positive operator in a Banach space E with spectral angle ( A, E) < T' Then the
difference problem (5.57) is well posed in Lp,r(Ea:,p) and the following coercivity
inequality holds:
II {7-2(Uk+l-2uk+Uk-d}f-lIILp,T(Ea,p)
+ II {7- 2[(I - R j ,I+l(7B))(Uk-l + Uk+l) + (R;,I+l (7B) - I)uknf- 1 II Lp ,T(Ea ,p)
::; M[II AUoIIEa,p + II Au NIIEa,p + a(1 ~ a) I
f;'1
Lp,T(Ea,p)
],
where M does not depend on f;'I' Uo, UN, p, a and 7.

Theorem 5.2.46. Let 1 < p, q < 00 and 0 < a < 1. Suppose that A is a strongly
positive operator in a Banach space E with spectral angle ( A, E) < T' Then the
difference problem (5.57) is well posed in Lp,r(Ea:,q) and the following coercivity
inequality holds:
I {7- 2(Uk+l - 2Uk + uk_d}f- 1 I Lp ,T(Ea ,q)
+ II {7- 2[(I - R j ,I+l(7B))(Uk-l + Uk+l) + (R;,l+l (7B) - I)uknf- 1 II Lp ,T(Ea ,q)

p2
::; M(q)[11 AUOIIEa,q
+ II Au NII Ea ,q + p _
f;'1
II Lp 'T(E
,)'
a
where M(q) does not depend on f;'I' Uo, UN, p, a and 7.

Here, the Banach spaces E~,p =
of those vEE for which the norm
E~,p(E,
00
I vilE'a,p =(
J
o
B) (0 < a < 1,1 ::; P ::; (0) consist

dz
IIza:B(zI+B)-lvll~-)p,l::;p<oo,
z
are finite and II v liE' =11 V liE' .

Now, the abstr~t theoren';s given above are applied in the investigation of
difference schemes of higher order of accuracy with respect to the set all variables
for approximate solution of the boundary-value problem (5.27). The discretization
of problem (5.27) is carried out in two steps. In the first step let us give the
difference operator AI; by the formula
x h=
A hUx
'"'
~
2mslrisS
bXDr
s: h'
r
hUxh + uU
x
(5.71)
The coefficients are chosen in such a way that the operator AI; approximates in a
specified way the operator
287
We shall assume that for I~khl

satisfies the inequalities
::;
IT
the symbol A(~h, h) of the operator AI;. - b
With the help of AI;. we arrive at the boundary-value problem
d2 v h (y,x)
x h
h
+Ahv (y,x)=f (y,x),O<y<T,
d 2
y
(5.73)
vh(O,x) = cph(X),vh(T,x) = 'ljJh(X), X E IRh,

(5.74)
+ :2 (I -
Rj,l+! (7 Bh))(uLl (x)
= fk,l(x), 1 ::; k ::; N - 1, u~(x)
+ U~+l (x)) + :2 (R1,l+l (7 B h ) - I)u~(x)
= vg(x),
u~
= v~,
x E R h, (B h )2
= AI;..
Let us give a number of corollaries of the abstract theorems given above.

Theorem 5.2.47. The solutions of the difference schemes (5.74) satisfy the follow-
ing stability estimates:
II u'T,h IIc;"Q(c~)::; M[llu~llc~ + Ilu~lIc~+ II

o < 0' < 1,0 :s (3 < 1,
I
u'T,h IILp'T(c~):S M[llugllc~
+ Ilu~llc~+ II
f;'lh Ilc;"Q(c~)],
f;,t IILp'T(c~)],
1::; p::; 00,0::; {3 < 1,
where M does not depend on fIl' u3, u ~!
0',
{3, p! hand
5.2.36, the positivity of the operator AI;. in C~ and on the fact that for any
o < {3 < 2~ the norms in the spaces E~ (Ch, AI;.) and C~mf3 are equivalent uniformly in h and on the following theorem on the structure of the fractional spaces
E~(Ch, (AI;.)!).
spectral angle (A, E) < ~. Then for 0 < 0' < ~ the norms of the spaces E~(E, A!)
and
E~
2
(E, A) are equivalent.
288
Proof. Using the following Cauchy-Riesz representation formula for the operator
A! (,\ + A! ) -1 (see Chapter 4),
X:> A-! (,\ + A!)-l y
where 8 1 = {pe i 1/>,O :S p < oo} and 8 2 = {pe i 1/>,O :S p < oo},O :S 'Ij; < ~. Let
y E E~ (A, E). Then from the strong positivity of A it follows that
2
Iz~ III A(z - A)-l y
IIE:S M p~ I
A(p + A)-l y
IIE:S M I y IIE'a (A,E)

"2
M
<
,
1,\ + z 2 1- ,\ + yIP
1
--1;-
it follows that
00
:S M 1
Since
J+
,\
N"
yIP ylPp~ dp I Y IIE~ (A,E)
00
J,\ +
00
,\ Q
yIP ylPp2
dp= 2
dr
M,
<
(1 + r)r
a(l - a)
we have
Q
11M2
,\ IIA2(,\+A2)-
YIIE:S
a(l-a)
II y IIE~(A,E)
for all'\ > O. This shows that y E E~(A!,E) and
I y IIE~(A!,E):S
M2
a(l- a) "y IIE~(A,E) .
Next, using the the following Cauchy-Riesz representation formula for the operator
A! (,\ + A)-l (see Chapter 4),
,\ ~ A('\ + A)-l y =
~
27rZ
J,\ ~
S1 US 2
z('\ + Z2)-1 A! (z - A!)-l y dz
5.2, Difference Schemes Generated by an Exact Difference Scheme
289
where 8 1 = {pei'l/> , 0 ~ p < oo} and 8 2 = {pei'l/> , 0 ~ p < oo}, 0 ~ 7/J < ~. Let
y E E~ (A ~ , E). Then from the strong positivity of A it follows that

1
M
--=-:-<-2
IA + z
1 -
A + p2'
it follows that
Since
we have
M2
0<
A'2IIA(A+A)- ylIE~ a(1-a) IIYIIE~(A!,E)

for all A > O. This shows that y E
E~ (A,
2
E) and

Theorem 5.2.49. The solutions of the difference schemes (5.74) satisfy the following almost coercive stability estimates:
{T-
(UZ+ 1 -2uZ+uL1nf-
~ M[In ~[
+ In T
L
Irl=2m
IID~u~lIch +
~h I
1;'/1
11ILP,r(C )
h
L
Irl=2m
IID~u~llchl
f;'lh IILp,r(Ch )], 1 ~ P ~

u~,
u'N, p, hand T.
00,
290
5.2.48, the positivity of the operator AI; in C~ and on the almost coercivity inequality for an elliptic operator AI; in Ch and on the estimate
min{ln~,l+
Ilnll Bf, II c cJ3I}::; Mln~h'
r
r+
J3
h~
Theorem 5.2.50. The solutions of the difference schemes (5.74) satisfy the follow-
ing coercivity estimates:
< M(a,l3)[
Irl=2m
IID;;uallc~ +
jrl=2m
IID;;u~lIc~+ II f;'lh Ilc~'Q(cn],
o::; a < 1,0 < (3 < 1,

II {r- 2 (uZ+l
- 2uZ + ULl)}~-l IILp.,.(w;,',:')
< M(a,p,q)[lluaIlE;_i)W;,',:') + Ilu~IIE;_t)W;,'hQ)+ II f;'lh IILp.T(Wq':',:')],

1 < p, q <
00,0
< a < -,
m
where M(a,f3) and M(a,p,q) do not depend on 1;'1' u3, ujy, hand T
5.2.45, the positivity of the operator AI; in W:'h' on the almost coercivity inequality
for an elliptic operator AI; in C~, W:'h' 1 < q < 00, 0 < (3 < 1, on the fact that
for any 0 < (3 < 2~ and 1 ::; q ::; 00 the norms in the spaces Ep,q(Lq,h, AI;) and
W:,'J:f3 are equivalent uniformly in hand on the following theorem on the structure
of the fractional spaces E~,q(Lq,h, (A,;)!).
spectral angle <!>( A, E) < ~. Then for 0 < a < ~,1 ::; p ::; 00 the norms of the
spaces E~,p(E, A!) and E~ ,p(E, A) are equivalent.
This result was proved in the case p = 00 in Theorem 5.2.48. In the case
p = 1 the proof follows the same scheme. In the case 1 < p < 00 the proof is based
on Minkowski's inequality and follows the same scheme.
Note that in a similar manner we can construct the difference schemes of a
high order of accuracy with respect to one variable for approximate solutions of
the boundary-value problems (5.25) and (5.26). Abstract theorems given above
permit us to obtain the stability, the almost stability and the coercive stability
estimates for the solutions of these difference schemes.
5.3. Two-Step Difference Schemes Generated by Taylor's Decomposition
5.3
291
Two-Step Difference Schemes Generated

We consider again the boundary-value problem (5.1). Let the function v(t)(O :S t :S
T) have a (2l + 2j + 2)-th continuous derivative and tk-1, tk, tk+1 E [0, T]r. Then
by Theorem 3.2.1 we have the following Taylor decomposition on three points:
V(tk+1) - 2V(tk)
+ v(tk-d -
L CtiV(2i) (tk)7 2i
(5.75)
i=l
- L 17i(V(2i) (tk-d + V(2i) (tk+l))7 2i = 0(721+2j+2),

j
i=l
where 17m, m = 1, ... ,j is a solution of system
(2 ')'
j
'"
t 17m
= 1 l
LJ (2i - 2m)!
'
m=l
+ 1 <- i -< l + J' ,
(5.76)
and
for any i, 2 :S i :S j,
i :S l.
Now, we will consider the applications of Taylor's decomposition (5.75) of

function v(t) on three points to approximate solutions of the boundary-value problem (5.1). From (5.75) it is clear that for the approximate solution of problem
' IS
. necessary to fi nd V(tk)
(2i)
. < l and V(tk_tl'
(2i)
(2i)
(5 .1) It
or any t,. 1 <
_ t _
V(tk+tl or any
i,l :S i :S j. Using the equation
v"(t)
we obtain
v(2n)(t)
= Anv(t) -
Av(t) - f(t),
L An-'\ f(2,\-2)(t), n = 2, ....

n
(5.77)
'\=1
Now, using formulas (5.75) and (5.77), we obtain the difference schemes
Uk+1 - 2Uk
7
+ Uk-1 + 'LJCti
" Ai 2i-2
7
Uk
i=l
LJ 17i Ai 7 2i-2 ( Uk-l + Uk+1 ) = fj,1

k ,
+ '"'
i=l
(5.78)
292
j
j~,1 =
I
L'Tli LAi-'\(j(2,\-2)(tk+d
i=l '\=1
+ j(2,\-2) (tk_d)T2i-2
2 1< k <N ' " A i-'\j(2,\-2)(t k

)T2i -,
~ a '~
_
_
+ '"
1,U
0 = v0, UN = vT
i=l '\=1
Thus, we have constructed the difference schemes of (2l + 2j)-order of accuracy for the approximate solution of the boundary-value problem (5.1).
This problem is uniquely solvable, and the following formula holds:
Uk = (1 - R 2N (TB))-l{(R k(TB) - R 2N - k(TB))uo
(5.79)
+(RN-k(TB) - RN+k(TB))UN - (RN-k(TB) - R N+k(TB))(1
+ TB)
N-1
x(2I + TB)-l B- 1 L (RN-i(TB) - RN+i(TB))Cjl,I T
i=l
N-1
+(1 + TB)(21 + TB)-l B- 1 L (R1k-il(TB) - Rk+i(TB))Cjl,I T, 1 ::; k::; N -1,
i=l
where
2
T AI
B = B(T,A) = -2j
Al
(2)ai
i=l
+ 2'Tli)AiT2i - 2 +
A I )2
2
+ A I,
2::
aiAiT2i-2)C,
i=j+1
2:: 'TliAiT2i)-1.
j
(1 -
i=l
Actually, problem (5.78) can obviously be rewritten as the equivalent boundaryvalue problem
-
Uk+1 - 2Uk
T
+ Uk-1
_
j,1
__
+AIUk-Cjk ,1::;k::;N-l,uo-vo,uN-VT
In a similar manner with the boundary-value problem (5.1) it can be rewritten as

the equivalent boundary-value problem for first-order linear difference equations
T- 1(Uk - Uk-1)
{
+ BUk
-T- 1(Wk+1 - Wk)
Wk, 1::; k ::; N, Uo
v(O), UN
v(T),
+ BWk = (I + TB)CJk''/ ,1::; k ::; N - 1,
with operator B = B(T,A). Note that B(T, A) =I A!, but then B(T,A) ~ A! as
T ~ 0 and it has the same spectral properties of A! under some assumption for A
293
and for some cases of j and l. From the last system follows the system of recursion
formulas
Uk = R(rB)uk-1 + TR(rB)wk, 1 ~ k ~ N,
{
Wk
'I
= R(TB)Wk+1 + TCfk' ,1
~ k ~ N -1,
= (1 + TB)-l. Hence
where R(TB)
Uk = Rk(TB)uo
+ l: Rk-i+l(TB)TWi' 1 ~ k ~ N,
i=1
Wk = RN-k(TB)WN
N-l
+ l:
8=k
R8-k(TB)Cfl,lT, 1 ~ k ~ N-1.

N-1
UN = RN(TB)uo
+L
RN+1-i(rB)TWi +TR(TB)WN.
i=1
Since
N-1
RN+1-i(rB)TRN-i(rB)
+ rR(TB)
i=1
L RZN - Zi+1(rB)r
i=1
N-1
N-1
N-1
8
L RN-i+l(rB)T L R 8- i (rB)Cfi,lr = L T LR N+l+ 8- Zi (rB)C!i,lr
i=1
8=i
8=1
i=1
N-1
L r Z(1 - R Z(rB))-1 R(rB)(R N- 8(rB) - R N+ 8(rB))Cf{l,
8=1
we have that
N-1
- L TZ(1 - R Z(rB)t 1R(rB)(RN- 8(rB) -
R N+ 8(rB))Cf1,1}
8=1
N-1
-(1 - R ZN (rB))-1 L
8=1
r(R N- 8(TB) - R N+ 8(rB))Cfl,l
294
Therefore
Uk = Rk(TB)uo + L Rk-i+l(TB)T(RN-i(TB)WN
i=l
N-1
+ L RS-i(TB)Cft,IT) = Rk(TB)uo + L R k-i+ 1(TB)TR N- i (TB)WN

s=i
i=l
N-1
k
i
1
+ LR - + (TB)T L R S- i (TB)C!1,IT.
i=l
s=i
k
Since
i=l
i=l
L Rk-i+l(TB)TRN-i(TB) = L R N+k- 2i+l(TB)T
k
N-1
LR k-i+ 1(TB)T L RS-i(TB)Cft,lT
i=l
s=i
k
s
N-1
k
= LT LRk+s+l-i(TB)C!i'IT + L
T LRk+s- 2i+1(TB)Cft,IT
s=l i=l
s=k+1 i=l
= (1 - R 2 (TB))-1 LTR(TB)(Rk-S(TB) - Rk+s(TB))Cft,IT
s=l
N-1
+(1 - R 2 (TB))-1 L TR(TB)(RS-k(TB) - R k+s(TB))C!1,IT
s=k+1
N-1
= (1 - R 2 (TB))-1 L TR(TB)(R1k-SI(TB) - Rk+s(TB))Cft,IT,
s=l
we have that
N-1
+(1 - R 2 (TB))-1 L TR(TB)(R1k-SI(TB) - Rk+S(TB))Cft,IT.
s=l
From this formula and using the formula for WN we can obtain formula (5.79).
Let us reduce the difference schemes (5.78) to an operator problem in the
space Fr(E). In addition to the operator D;', acting from the space Ex Fr(E) x E
of vectors w r = {wdf=o into the space Fr(E) of vectors vr = {vdi:"=-/ by the
rule
295
define an operator A?,I from the space E x FT(E) x E of vectors w T = {wdf=o

into the space FT (E) of vectors vT = {vd ~,;/ by the rule
j
= AjT' IU T , Vk = ~
L.J TJi Ai 7 2i-2(Wk-1
+ wk+1 )
i=l
+L
(YiAi72i-2Wk,
k = 1, ... , N - 1.
i=l
operator equation
-D;IT(uo, UN )uT + A~,IIT(uo, UN )uT = fII'

Here fI I is defined by the formula
fj,l)
T - (fj,1
f j,1
1"'"
N-1 .
solvability for any Uo, UN E E and fj,1 E FT(E), it follows that its solution uT
defines a continuous additive and homogeneous operator uT(fj,I,UO,UN).
the inequality
where M is independent not only of fI I, Uo, UN, but also of 7.

+ II
{L
TJiAi72i-2(Uk_1
+ uk+d + L
i=l
(YiAi72i-2uk}{"'-1
::; M[li
fII
IIFT(E) + II
Auoil E
II
FT(E)
i=l
+ IIAuNII E ],
where M is independent not only of fI I, Uo, UN, but also of 7. From formula (5.58)
it follows that the investigation of the stability and well-posedness of difference
schemes (5.78) relies in an essential manner on a number of properties of the powers
of the operator (I + 7B)-1. We were not able to obtain estimates for powers of
the operator (1 + 7Bt 1 in the general cases of operator A and numbers j and i.
We begin by deriving some estimates for powers of the operator (1 + 7 B) -1 in the
cases j = 0 and alli, i 2 1 and a strongly positive operator A in a Banach space
E with spectral angle (A, E) < fl.
296
Lemma 5.3.1. Let A be a strongly positive operator in a Banach space E with
spectral angle (A, E) < fz. Then -AI is a generator of the analytic semigroup
exp {-tAt} (t ~ 0) with exponentially decreasing norm, when t - 4 +00, i.e., we
have the estimates
Ilexp{ -tAI}IIE---+E :S MI e- tol (t
> 0),
(5.80)
IltAlexp{-tAt}IIE---+E:S MI e-tol(t > 0)
(5.81)
for some 1 :S MI < +00,0 < 01 < +00. Here MI does not depend on
T.
Proof. Let us show by induction that for any integer m, 2 :S m :S l, Am is a strongly

positive operator in a Banach space E. Let m = 2 and A = IAlei<p, 7r > lipl ~ ~.
We consider the identity
2
AI - A = (A2J - A)(A2J + A).

1
SinceA~ = IAI~ei-r,~ > I~I ~~, by the definition ofastrongly positive operator
in a Banach space E we have that the operator A~ I - A has a bounded inverse,
and the estimate
holds for some MA E [1, +00) that does not depend on A. It is readily proved that
-(A ~ I
+ A)
+ A)} (t ~ 0)
+00, Le., we have the following
is a generator of the analytic semigroup exp { -t(A ~ I
with exponentially decreasing norm, when t

estimates:
Ilexp{ -t(A~ I
+ Anll ---+
E
-4
:S M e-t(I>.I! cos -r+O) (t
~ 0)
for some 1 :S M < +00, 0 < 0 < +00. From this it follows that the operator
A~ I + A has a bounded inverse, and the estimate
II(A ~ I
+ A)-lIIE-+E
:S M(l
+ IAI ~ cos ~ )-1
holds for some M E [1, +00) that does not depend on A. Therefore, the operator
AI - A 2 has a bounded inverse, and the estimate
holds for some M A 2 E [1, +00) and for all A, A = IAlei<p, 7r > lipl ~ ~. This means
that A 2 is a strongly positive operator in a Banach space E. Suppose that for any
integer k, 2 :S k :S m - 1, Ak is a strongly positive operator in a Banach space E.
Then, we will prove that Am is a strongly positive operator in a Banach space E.
We consider the identity
297
We have that for any integer k, 1 :S k :S m, Am-k is a strongly positive operator

in a Banach space E. Therefore, the operator AI - Am-k has a bounded inverse,
and the estimate
i.
holds for some MA=-k E [1, +00) and for all z, z = Izl ei 1/J, 7r > 111'1 2
It is readily
proved that the operator A k;;;l Am-k has a bounded inverse, and the estimate
holdsforsomeMA=-k E [1,+00), for all z,z = Izl ei 1/J,7r > 111'12

Indeed,
II (zI - Ak;;;l Am-k)-lIIE--+E
i and IAI2 <5 > O.
IIA k;;;l (ZA - k;;;l I - Am-k)-lIIE--+E
< MA=_kIAlk;;;l (1 + IAI-k;;;llzl)-l
From this it follows that -A k;;;l Am-k is a generator of the analytic semigroup
exp { -tA k;;;l Am-k } (t 2 0) with exponentially decreasing norm, when t ~ +00,
i.e., we have the estimates
for some 1 :S
if < +00, 0 < J < +00.
Since the operators A k;;;l Am-k commute, we have that a generator of the analytic semigroup exp
k=l
is
{-t k=lf, Ak;;;l Am-k} (t 2 0) with ex-
ponentially decreasing norm, when t ~ +00. Therefore,

bounded inverse, and the estimate
L A k;;;l Am-k
L
m
k=l
k-l
A-;;;- Am-k has a
298
holds. From this it follows that the operator

the estimate
11(>.1 -
>.1 -
Am has a bounded inverse, and

m
Am)-lIIE--.E :S
11(>,.k 1- A)-lIIE--.E 11(2:: A
k;;;l
Am-krlllE--.E
k=l
:S MA(1 + IAI.k)-1(M)m-l e- Cm-1)6IAI- m~l = M Am(1 + IAI)-l
holds for some MAm E [1,+(0) and for all A, A = IAle i ,p,ll' > I~I ~~. This
means that Am is a strongly positive operator in a Banach space E. Finally, we
will prove that - Al is a generator of the analytic semigroup exp {- tAL} (t ~ 0)
with exponentially decreasing norm, when t -----+ +00. The operators <l'iAir2i-2
commute. Therefore, - L:~=l <l'iAir2i-2 is a generator of the analytic semigroup
exp { -t L:~=l <l'iAir2i-2} (t
+00. Indeed,
II exp { -t
~ 0) with exponentially decreasing norm, when t
a,A'r 2'-2 }
IIE-E
-----+
~ lIexp { -tA} IIE-E
:S Ilexp{ -tA} IIE--.E
II Ilexp
{-t<l'iAir2i-2}
IIE--.E
i=2
I
:S MAexp {-t8}
II MAiexp {
2i 2
-t<l'iJ:r - }
:S MA1exp {-t8 1}.
i=2
Estimate (5.80) is established. In a similar manner one can obtain estimate (5.81).
Lemma 5.3.1 is established.
Lemma 5.3.2. Let - A be a generator of the analytic semigroup exp { -tA} (t ~ 0)
with exponentially decreasing norm, when t -----+ +00. Then the following estimates
hold for any k ~ 1 :
11(>.1 + rB)-kII E--. E :S M[A + r 2 a(A)t\
(5.82)
IjkTB(I + rB)-kII E--. E :S M,
(5.83)
where M does not depend on r. Here
Proof. We consider the operator
where Al =
operator (A
299
A. Using the following Cauchy-Riesz representation formula for the
+ B1)-1 (see Chapter 4), we can write
Here the contour
is chosen with Re{ z} :S O. If z = pei1r and p > 4, then
Therefore
1
lim [-.
E-+O 27l'2
-4+iE (A + z/2 + !f:z2

- + z)-l(zI - A1)-ldz
j
-oo+iE
+ -.
27l'2
-OOj-iE
(A + z/2 +
!f:z2
- + z)-l(zI - Ad-1dz ]
4
-4-iE
O.
If z = pei1r and 0 :S p :S 4, then
~
1.
(A + z/2 + V4' + Z)-l = (A - p/2 + '2 ve t1rp(4 - p))-l
=
(A - p/2 +
~e4i1r Vp(4 -
p))-l = (A - p/2
~iV p(4 _
p))-l.
Therefore
1
lim[-2'
E-+O
7l'2
-OJ+iE (A+z/2+ !f:z2

-+z)-1(zI-A1)-ldz
4
-4+iE
+ -.
27l'2
-4 -iE (A + z/2 + !f:z2

- + z)-l(zI - A1)-ldz ]
j
-O-iE
j [(A - p/2 - -2p(4 - p))

4
= - -1.
27l'2
1 ...;
-1
1
2
- (A - p/2 + -iJ p(4 - p))-l](-pI - Ad-1dp
300
Then the deformation of integration contour
leads us to obtain the formula
J
4
(>.1 + B1)-1 =
M1(>.,p)(pI + A1)-ldp,
(5.84)
where
(5.85)
In a similar manner one can obtain
J
4
(>.1 + Bd- k =
Mk(>", p)(pI + Ad-1dp, k
= 2,3, ... ,
(5.86)
p))-k
(5.87)
where
M k (>", p) =
2~i [(>.. -
p/2 -
~iVp(4 -
1
-(>.. - p/2 + -iV p(4 _ p))-k]
2
t (
[HI]
1
=
211"
m=l
Clearly, M1(>",p)
2m -1
)(_1)m(>,,_p/2)k-2m+l(1Ip(4_ p))2m-l
(>..2 _ >..p + p)k
2V
2 0 for O:S p:S 4 and>.. 2 o. Therefore, using the estimate
we obtain that
J
J
4
11(>"1 + Bd-11IE--.E :S
M1(>",p)ll(pI + Ad-11IE--.Edp
:S M
M1(>..,p)(p+ad-1dp,
where al
= r 2a(A).
Now, using formula (5.84) for Al
301
= all, we obtain
J
4
MI(A, p)(p + a(Ad)-ldp = (A + aI/2 +
Therefore,
(5.88)
where
BI(al) = aI/2 +
(al)2
4 + al > 0.
So, from estimate (5.88) it follows that B I (AI) is a positive operator in E. Now,
we will estimate (AI + Bd- k for k 2 2. Clearly, the function M k ( \ p) for k 2 2
changes sign on the segment 0 :S p :S 4. Therefore, the last method of estimate
of (AI + BI(Ad)-1 does not work in the general cases k 2 2. Now, we will use
another formula for (AI + BI)-k. Since -A is a generator of the strongly continuous
semigroup exp {-tA} (t 2 0) with exponentially decreasing norm, when t ---+ +00,
Ilexp{ -tA}II E - + E :S M
e-ta(A)
(t 2 0)
for some 1 :S M < +00, 0 < a(A) < +00. Then, there exists a bounded inverse
(AI + AI)-l for any complex number A with ReA> -al(Ad, and the following
formula holds:
(AI + Ad- l =
+00
e->.t exp{ -tAddt.
(5.89)
o
Then, from formulas (5.86) and (5.89) it follows that
Mk(A, p)
pt
e- exp{ -tAI}dtdp
JJ
+00
=
+00
(AI + Bd- k =
4
Mk(A, p)e-ptdpexp{ -tAddt
(5.90)
+00
=
Lk(A, t) exp{ -tAI}dt,
where
J
4
Lk(A, t) =
Mk(A, p)e-ptdp.
(5.91)
302
Note if Al is a complex number with Re Al > 0, then (AI + B1)-k is the Laplace
transform of the function L k().., t). Since (AI + B1)-k-l = (AI + Bt}-k(AI +
Bt}-l = I1:+~1 (AI + Bt}-l, from properties of the Laplace transform it follows
that L k ().., t) is a convolution of the k functions L 1 (A, t) defined by the recurrence
relation
J
t
Lk+1(A,t) =
Lk(A,S)L 1(A,t-s)dsk=1,2,....
(5.92)
o
Since M1(A,p) ~ 0, then from formula (5.91) it follows that L1()..,p)
Therefore, from formula (5.92) it follows that
Lk(A,p)
for all k,k = 1,2, ....
O.
(5.93)
Formula (5.90) and inequality (5.93) permit us to apply the same approach of
estimate (5.88) for the estimate of (AI + B1(A1))-k in the general cases k ~ 2.
Namely, we have that
II(AI + B1(At})-kII E-+ E ~
~M
+00
+00
Lk().., t)11 exp{ -tAdIIE-+Edt
Lk(A, t) exp{ -tal}dt = M[A + alr k
o
for A ~ O. Therefore, - B 1 (AI) is a generator of the strongly continuous semigroup
exp{ -tB1(At}} (t ~ 0) with exponentially decreasing norm, when t ---+ +00, Le.,
lIexp{ -tB 1 (At}}II E - + E ~ M
e-tB1(a
(t ~ 0)
for some 1 ~ M < +00. Now we consider Al (AI + Bt}-k for 0 < a < 1. Applying
the formula
+00
1
- fCB)
-(3 -
Al
z (3-1 exp { -zAl } dz,(3 > 0
o
Af(AI + B1)-k u
= A~(1-a)
= r(1
~ a)
= A~(l-a)(AI
+00
L k().., t) exp{ -tAddtAlu
+00
z-a exp{ -zA1}dz
+ B1)-k A1u
+00
Lk(A, t) exp{ -tAddtAlu
=f(I~a)
for
303
JJ
+00 +00
Z-O:Lk(A,t)exp{-(t+z)AddzdtA1u
o 0
D(AI). Using the last formula and the estimate
(t + z)IIA 1exp{ -(t + z)A1}IIE_E ::; M e-(t+z)a l
we obtain that
JJ
+00 +00
< f( M )
I-a
z-o: Lk(A, t)_I_

t+z
x exp{ -(t + z)al}dzdt
+00
Mf(a)
Lk(A, t) exp{ -tal}dt
x f(a)r~1 _ a)
Since
t-O: =
f(a)r~1 _ a)
+00
z-O:(t + z)-ldz.
+00
z-O:(t + z)-ldz,
(5.94)
we have that
+00
IIAf(U + B1)-kIl E_ E S; Mr(a)
Lk(A, t)CO' exp{ -taddt.
IIAnAI + Bd-kIIE_E
JJ
+00
::; Mf(a)
Mk(A, p)e-ptdpCO: exp{ -taddt
J
4
Mf(a)f(1 - a)
o
x exp{ -tal}dt]dp
Mk(A, P)[r(1 ~ a)
+00
e-ptCO'
J
4
= Mf(a)f(1 - a)
Mk(A, p)(p + ado:-1dp.
304
J
J
IIAf(AI + Bd-kIIE-+E :::; M
Z-(l-a)(Z
+ p + ad-1dzdp
Z-(l-a)
Mk()..,p)
+00
+00
Mk(A,p)(Z
+ p + al)-ldpdz
+00
=M
Z-(l-a)(A + B1(z + ad)-kdz.
Finally, by passing to the limit a

estimate
----t
1 in the last inequality one can obtain the
J+
+00
IIA1(AI + B1)-kII E-+ E :::; M
(A
B1(z + al))-kdz.
Since
B1(z+al)
and al
d2
(z+al)/2+ v(z+a
4
+z+al
> 0, it follows that
Let us make the substitution
x
Clearly,
z/2 + V"4 + z.
x 2 - zx - z = 0.
Therefore, z
= x~l
= extl); dx.
2 2 2
and dz
J+
+00
IIA1(I + B1)-kII E-+ E :::; M
(1
x)-(k+2)(x 2 + 2x)dx.
Integrating the last integral by parts, we obtain
J+
+00
(1
x)-(k+2)(X 2 + 2x)dx
J+
. :. .l(_+-----'x)'----_(k_+_l) (x2 + 2x) 100

k+1
0
+00
+_1_
k+1
(1
x)-(k+l)(2x + 2)dx
- (k
+ 1)k
(k
+ 1)k
(1 + x)-k+l 2 00
- (k + 1)k(k - 1) 10
for k
2. Therefore,
for k
2. Using the identity
for k
+ 2)10 + (k + 1)k
+ 1)k + (k + 1)k(k -
we obtain
00
(2x
= (k
+00
(1+X)-k
1) =
(1
+ x)
-k
305
2dx
2[k - 1 + 1]
2
(k 2 - 1)k = (k 2 - 1)
+ B 1 )-(k+l)11 E--.E <II B2(I

1
k22M
_ 1
2. The estimate
for all k, k ~ 1 follows from estimate (5.82). Since B1(Ad is a positive operator
in E, we have the moment inequality
1
IIBiullE ~ MIIBi u11111 u111(u
E D(Bi))
Lemma 5.3.2 is established.

We have the following results.

spectral angle (A, E) < fl. Then difference problem (5.78) in the case j = 0 is
stable in C~,Q;(E), (0 ~ 0: ~ 1) and in Lp,T(E), 1 ~ p ~ 00 (here Loo,r(E) =

CT(E)), since the solutions of the difference problem (5.78) satisfy the stability
inequalities
I uT Ilc;:""(E)~ M[II
fIl
Ilc;:""(E) + I uoll E + IluNII E],
Iu
fIl
IILp,T(E)
IILp.T(E)~
M[li
where M does not depend on fIl'
Uo, UN,
p,
+ I uoll E + IluNIIE],
0:
and T.
306
spectral angle (A, E) < fL and uo, UN E D(A). Then the solutions of the difference problem (5.78) for j = 0 in Lp,r(E) obey the almost coercive inequality
i
II {r- 2(uk+i - 2Uk + uk_d}f- IILp,T(E) + II
{L QiAir2i-2udf-i IILp,T(E)
i=i
:s Mdll Auoil E + IIAuNll E

+ min
{ln~, 1 + Iln II B IIE-.EI} II 1;'1 IILp,T(E)],
where M 1 is independent not only of 1;'1' UO, UN, but also of r.

In the case j = 0 we have the estimate
Therefore, we have the following results.

spectral angle (A, E) < fL and uo, UN E D(A). Then the solutions of the difference problem (5.78) for j = 0 in Lp,r(E) obey the almost coercive inequality
i
i
2
II {r- (uk+i - 2Uk +uk-d}f- IILp,T(E) + II {Audf- IILp,T(E)
:s Mi[II Auoil E +
+ min
IIAuNll E
{ln~, 1 + Iln II B IIE-.EI} II f;'1
IILp,T(E)],
where M i is independent not only of f;'I' UO, UN, but also of r.

spectral angle (A, E) < fL and uo, UN E D(A). Then the solutions of the difference problem (5.78) for j = 0 in C~,Q(E) obey the coercivity inequality
i
i
2
II {r- (uk+i - 2Uk + Uk_i)}f- Ilc;""(E)+ II {AUk}f- Ilc;""(E)
:s MdQ(l ~ Q) II
f;'1 IIc;""(E) + II Auoil E + IIAuNII E],
where M i is independent not only of f;'I' UO, UN,
Q,
but also ofr.
spectral angle (A, E) < fL and (I - R(rB))2r -2 uo - Ji'1, (1 - R(rB))2r -2 uN 1~~1 E E~. Then the solutions of the difference problem (5.78) for j = 0 in C~(E)
obey the coercivity inequality
i
i
2
II {r- (Uk+i - 2Uk + Uk_i)}f- IIc~(E) + II {Audf- IIc~(E)
+ II {r- 2(uk+l - 2Uk + uk-d}f- i IIcT(E~)
~ ex(~ ex) [II fIl IIc;>(E) + I (I +11(1 -
R(TB))2 T-2 UO - H,l
307
IIE~
R(TB))2T-2UN - f~~IIIEI],
Q
where M I is independent not only of fIl' Uo, UN, ex, but also of T. Here, the
Banach space E~ = E~(E, B) (0 < ex < 1) consists of those vEE for which the
norm
I V liE' = supzQ I B(zI + B)-Iv liE + I V liE
z>o
is finite.
spectral angle (A, E) < fL and (I - R(TB) )2T-2 uO - f{l, (I - R(TB))2 T-2 UN E E~_,. Then the solutions of the difference problem (5.78) for j = 0 in
C~"(E) (0 ~ "I ~ (3,0 < (3 < 1) obey the coercivity inequalities
nLI
II
I
2
{T- (Uk+1 - 2Uk + Uk-Inf-
+MI[l1 (1 -
IICT(Er,_-y) ~
(3(1 ~ (3) II fIl IIc~'-Y(E)
liE'B--y + 11(1 - R(TB))2T-2 UN - f~~IIlE'B--y ],

I
I
2Uk +uk-d}f- Ilc~'-Y(E) + II {Audf- Ilc~'-Y(E)
R(TB))2 T-2 UO - Ji'l
II {T- 2(Uk+1
(3(~ (3) I
<
+I(I -
IIc~'-Y(E) + Mdl(I -
fIl
R(TB))2T-2 UN -
R(TB))2 T-2 uO -
nLIlg,,],
Ji'll~"
where M 1 is independent not only of fIl' Uo, UN, (3, "I, but also ofT. Here, Iwlg"
denotes that the norm of the Banach space Eg,' consists of those wEE for which
the norm
max
1~n:':;N-I
IIRn(TB)wIIE
sup
l~n<n+r~N-I
t;f3(t n + tr)'(tN - tnPII(Rn+r(TB) - Rn(TB))wIIE
is finite.
spectral angle (A, E) < fl. Then the solutions of the difference problem (5.78)
for j = 0 in CT(E~) obey the coercivity inequality
2
{T- (Uk+1 - 2Uk
+ uk-d}f- I IlcT(E~) + I {Audf- 1 IlcT(E~)
~ M I [ex(l ~ ex) I r IIcT(E~) + I AUoIIE~ + IIAuNIIE~],

where M 1 is independent not only of fT, uo, UN, ex, but also of T.
308
Jl,l,
spectral angle <f>(A, E) < fL and (1 - R(TB))ZT-2uo (1 - R(TB))2T-2 uN rJJ~l E E~-'Y' Then the solutions of the difference problem (5.78) for j = 0 in
C~''Y(E~_f3) (0::; 1::; (3::; a,O < a < 1) obey the coercivity inequalities
II
2
{T- (Uk+l - 2Uk
+ II
2
{T- (Uk+l - 2Uk
+M1 [l(1 -
II
+ uk_I}}f- 1 Ilc~''''(E~_I') + II
R(TB))2 T-2 uo -
2
{T- (Uk+l - 2Uk
+ II
::;
+Mdll(1 -
+ uk-I}}f-
{AUk}f- Ilc~''''(E~_I')
IlcT(E~~I') ::; a(~ a) II r Ilc~''''(E~_I')
fi'11:~f3 + 1(1 - R(TB))2T-2UN - I~~ll~'2'f3],
+ Uk-lnf- 1 Ilc~''''(E~_I') + II {Audf- 1 Ilc~''''(E~_I')
2
{T- (Uk+l - 2Uk
+ uk-I}}f- 1 IlcT(E~_..,)
a(~ a) [II (1 + TB)-lCI;'1 Ilc~''''(E~_I')
R(TB))2T-2 UO - Ir,IIIE f
Q-~
+ 11(1 -
R(TB))2T-2uN - f~~lIIEf
Q-~
]],
where M 1 is independent not only of 1;'1' uo, UN, a, (3, 1, but also ofT.
Here, Iwl~'2'f3 denotes that the norm of the Banach space E~'2'f3 consists of
Iwl~'2f3 =
sup
lSn<n+rSN-l
is finite.
max
lSnSN-l
n
IIR (TB)wIIE o _1'
t;f3(t n + trP(tN - tnPII(Rn+r(TB) - R n(TB))wIIE o _1'
Finally, let us give the following results about well-posedness of the boundaryvalue difference problem (5.78) in the spaces Lp,T(E) = Lp([O, T]r, E), 1 ::; p < 00.
spectral angle <f>(A, E) < fl. Suppose that the difference problem (5.78) for j = 0 is
well posed in LpO,T(E) for some Po, 1 < Po < 00. Then it is well posed in Lp,T(E)
for any p, 1 < p < 00 and the following coercivity inequality holds:
II
2
{T- (Uk+l - 2Uk
+ Uk-lnf- 1 IILp,T(E) + II
{Audf-
IILp.T(E)
+lluTllcc(El_l/p,p) ::; M(Po)[ll UOIlEl_l/p,P + IluNIIE1_1/p,P

p2
+p~
-
II
fj,l
II
Lp,T(E)
],
where M(po) does not depend on f;'l' Uo, UN, P and T.
309
00 and 0 < a < 1. Suppose that A is a strongly

positive operator in a Banach space E with spectral angle 4>(A, E) < it. Then the
difference problem (5.78) for j = 0 is well posed in Lp,r(Ea,p) and the following
Theorem 5.3.12. Let 1 :S p :S
{r- (uk+l - 2Uk
:S
+ Uk-lnf- 1II Lp ,T(Eo,p) + I {AUk}f- 1 IIL p,T(Eo,p)

1
M[II AUoIIEo,p + IIAuNllEo,p + a(l _ a) II
fIl
II
Lp,T(Eo,p)
],
where M does not depend on fIl' Uo, UN, p, a and r.
< p, q < 00 and 0 < a < 1. Suppose that A is a strongly

positive operator in a Banach space E with spectral angle 4>( A, E) < it. Then the
difference problem (5.78) for j = 0 is well posed in Lp,r(Ea,q) and the following
II {r- 2 (uk+l
- 2Uk
+ Uk-lnf- 1IILp,T(EO,q) + II {Audf- 1 IILp,T(EO,q)

p2
:S M(q)[11 AuollEo,q + IIAuNllEo,q + P _ 1
fII IILp'T(EO,q)],
where M(q) does not depend on fIl' Uo, UN, p, a and r.

Now, the abstract theorems given above are applied in the investigation of
difference schemes of higher order of accuracy with respect to the set of all variables
for approximate solution of the boundary-value problem (5.78). The discretization
of problem (5.27) is carried out in two steps. In the first step let us give the
difference operator A h by the formula
'"'
L...J
A xhUxh =
2,.,.:Slrl:SS
brXDrhUxh + UU
r h'
x
(5.95)
The coefficients are chosen in such a way that the operator A h approximates in a
specified way the operator
Irl=2m
We shall assume that for

satisfies the inequalities
I~khl
alrl
ar(x) axrl ... ax~n
+ 8.
:S
1r
the symbol
A(~h,
h) of the operator A h - 8
(_l)m AX(~h, h) 2 Mll~12m, I argAX(~h, h)1 :S 4> < 4>0 :S ;.

With the help of A h we arrive at the boundary-value problem
d2 v h(y,x)
x h
h
d 2
+Ahv (y,x)=f (y,x),O<y<T,
y
vh(O,x) = cl(x),vh(T,x) = 'ljJh(X), X E IR. h,

(5.96)
(5.97)
310

I
---;-(U~+l (x) - 2u~(x) + ULI (x)) + L Qi(AIJiT2i-2u~(x)

T
(5.98)
i=l
le(x), 1 :S k :S N - 1, u~(x) = v~(x), u~ = v~, x E Rh.
Let us give a number of corollaries of the abstract theorems given above.

Theorem 5.3.14. The solutions of the difference schemes (5.98) satisfy the following stability estimates:
I uT,h Ilc~,a(cO :S M[llu~llc~ + Ilu~llc~+ I I;'lh Ilc~,a(c~)],

o < < 1,0 :S (3 < I,
Q
I uT,h
II Lp ,r(cO :S M[llu~llc~
+ Ilu~llc~+ I
1 :S p :S 00,0 :S (3
where M does not depend on 1;'1' u~, u~,
I;'lh IILp,r(CO],
< 1,
Q,
(3, p, hand r.
The proof of Theorem 5.3.14 is based on the abstract Theorem 5.3.3 and the
positivity of the operator A h in C~.
Theorem 5.3.15. The solutions of the difference schemes (5.98) satisfy the following almost coercive stability estimates:
II {T-2(U~+1
- 2u~ + ULlnf- IILP,r(C h) +

1
:S M[ln ~[
Irl=2m
IIDhU~llch +
L II{DhU~}f-lIILP,r(Ch)
Irl=5
Ir l=2m
IIDhU~llch]
+ In T: h I II,lh IILp,r(Ch)], 1 :S p :S 00,

where M does not depend on 1;'1' u3, u~, p, hand T.
The proof of Theorem 5.3.15 is based on the abstract Theorem 5.3.5, the
positivity of the operator A h in Ch and on the almost coercivity inequality for an
elliptic operator A h in C~.
Theorem 5.3.16. The solutions of the difference schemes (5.98) satisfy the coercivity estimates
I {r- 2(uZ+l
- 2uZ + utl)}~-l II c;""(cn
Irl=2m
II{DhUZ}~-lllc;,'''(cO
~ M(a,,8)[
Irl=2m
where M (a,,8) does not
IIDhu311c~ +
Ir l=2m
311
IIDhu~llc~+ I fJ,lh Ilc~a(c~)],
o ~ a < 1,0 < ,8 < 1,

depend on fIl' u3, u~, hand T
The proof of Theorem 5.3.16 is based on the abstract Theorem 5.3.6, the
positivity of the operator A h in C~ and on the coercivity inequality for an elliptic
operator Ah in C~. Note that in a similar manner we can construct the difference
schemes of a high order of accuracy generated by Taylor's decomposition of the
function on three points with respect to one variable for approximate solutions
of the boundary-value problems (5.25) and (5.26). Abstract theorems given above
permit us to obtain the stability, the almost stability and the coercive stability
estimates for the solutions of these difference schemes.
Chapter 6
of Hyperbolic Type
In the present chapter we consider the abstract Cauchy problem for differential
equations of the hyperbolic type
v" (t) + Av(t)
= f(t) (0 ~ t
T), v(o)
= vo, v' (0) = v~
in an arbitrary Hilbert space H with the self-adjoint positive definite operator

A. The high order of accuracy two-step difference schemes generated by an exact
difference scheme or by Taylor's decomposition on three points for the numerical
solutions of this problem are presented. Stability estimates for the solutions of
these difference schemes are established. In applications, the stability estimates
for solutions of a high order of accuracy difference schemes of the mixed type
boundary-value problems for hyperbolic equations are obtained.
6.1
A Cauchy Problem
We consider the abstract Cauchy problem for hyperbolic equations
v"(t) + Av(t)
= f(t)
(0
~ t ~
T),v(O)
= vo,v'(O) = v~
(6.1)
in a Hilbert space H with the self-adjoint positive definite operator A.

A function v( t) is called a solution of problem (6.1) if the following conditions
are satisfied:
1. v(t) is twice continuously differentiable on the segment [0, T]. The derivatives
derivatives.
ll. The element v(t) belongs to D(A) for all t E [0, T] and the function Av(t) is
continuous on the segment [0, T].
iii. v( t) satisfies the equations and initial conditions (6.1). If the function f (t) is
not only continuous, but also continuously differentiable on [0, Tj, Vo E D(A)
Chapter 6. Partial Differential Equations of Hyperbolic Type
314
and vb E D(A!). It is easy to show that the formula
J
t
v(t) = c(t)vo + s(t)vb
(6.2)
s(t - >")f(>")d>"
gives a solution of problem (6.1). Here
C(t) = e
itA 1/2
~e
-itA 1/2
itA 1/2
,S(t) = A- l / 2 e
~ie
-itA 1/2
Actually, obviously (6.1) can be rewritten as the equivalent initial-value problem for a system of first-order linear differential equations
v: (t)
{
+ iBv(t) ~ zit) (0 ,; t
T), v(O) ~ Vo, v' (0) ~ v~,
,;
z (t) - iBz(t) = f(t),
where B = A!. Integrating these, we at once obtain
V(t) ~ e-iB<"" + :~ e-iBU-,j z(,)ds,

{ z(t) = etBtzo + fo etB(t-p) f(p)dp.
That is,
v(t) = e-iBtvo +
it
1t
e-iB(t-s)eiBszods
e-iB(t-s) is eiB(s-p) f(p)dpds,
so, using the condition v' (0)
v(t)
+ iBv(O)
(e- iBt
1
+ it
t
+ iB
e-iB(t-s)eiBsdsv' (0)
z(O), we obtain
e-iB(t-s)eiBsds)vo
e-iB(t-s)
eiB(s-p) f(p)dpds.
v(t)
+B- l
eiBt _ e- iBt
.
.
(e- tBt
,
dsv (0)
it
0
eiBt _ e- iBt
2
B- 1
)vo
eiB(t-s) - e-iB(t-s)
.
Thus, by the definitions of B, c(t) and s(t) we have (6.2).
f(s)ds.
6.1. A Cauchy Problem
315
Theorem 6.1.1. Suppose that Vo E D(A) , vb E D(A~) and f(t) is a continuously
differentiable function on [0, T]. Then there is a unique solution of problem (6.1)
and the stability inequalities
OTtaz-T II v(t)
OTtaz-T
max
O~t~T
II
IIH:S: M [II Vo IIH + II A-l/2v~ IIH + OTtaz-T I

/
2
v(t)
d2 v(t)
-d
2
t
IIH:S: M [II
IIH + 09~T
max I
A-
IIH + II v~ IIH + OTtaz-T I
2
vO
Av(t)
2
f(t) IIH] ,
f(t)
IIH] ,
IIH:S: M[II Avo IIH + I Al/2V~ IIH
fI
T
+I
f(O) IIH
o
hold, where M does not depend on f(t), t
J'(t)
IIH dt]
[0, T] and Vo, vb.
Proof. Using formula (6.2) and estimates

IIc(t)IIH--+H :s: 1, IIA~s(t)IIH--+H :s: 1,
(6.3)
we obtain
f IIA~s(t
t
A)IIH--+HIIA-~ f(A)II HdA
:S:II Vo IIH + II A-1 vb IIH +T 09~T

max I
A- 1 / 2 f(t)
IIH .
Applying A1 to the formula (6.2) and using estimates (6.3), in a similar manner,
we obtain
f IlA1
t
s(t - A)IIH--+Hllf(A)IIHdA
:s:1I A1 vo IIH + I v~ IIH +T max II f(t) IIH .

O~t~T
Now, we obtain an estimate for I Av(t) IIH. Applying A to formula (6.2) and
using an integration by parts, we can write the formula
J
t
1
Av(t) = c(t)Avo + A21 S(t)A2V~
+ f(T) - c(t)f(O) -
c(t - >..)J'(>..)d>...
316
Using the last formula and estimates (6.3), we obtain
IIAv(t)IIH:S Ilc(t)IIH-+H I
Avo
IIH +IIA~s(t)IIH-+H I A~v~ IIH
J
t
+I
IIH + I c(t)IIH-+Hllf(O) IIH + Ilc(t -
f(T)
A)IIH-+HIlf'(A)IIHdA
<II Avo IIH

-
+ I A~v~ IIH +21If(0) IIH +2T 0:St:ST

max I
f'(t) IIH .
Then from the last estimate, it follows that

max
O:St:ST
II Au(t) IIH:SII Avo IIH + II A~v~
The estimate for max
O:St:ST
IIH
+21If(0) IIH +2T O:St:ST

max II
f'(t)
IIH .
I d:~~t) IiH follows from the last estimate and the triangle
inequality. Theorem 6.1.1 is proved.
Remark 6.1.2. Theorem 6.1.1 holds in an arbitrary Banach space E under the
assumptions
IIc(t)IIE-+E:S M,
IIA 2 s(t)IIE-+E:S M,
O:S t:S T.
(6.4)
Now, we consider the application of this abstract Theorem 6.1.1. First, we

consider the mixed problem for hyperbolic equations
Utt(t' x) - (a(x)uxh + u(t, x) = f(t, x), :s t :s T,

u(O,x) = <p(x),Ut(O,x) = '(x), :S x:s L,
{
u(t,O) = u(t,L),ux(t,O) = ux(t,L),O:S t:S T.
:s x :s L,
(6.5)
Problem (6.5) has a unique smooth solution u(t,x) for smooth a(x) > (x E
[O,L]), <p(x), '(x) (x E [O,L]) and f(t,x) (t E [O,T], x E [O,L]) functions. This
allows us to reduce the mixed problem (6.5) to the initial-value problem (6.1) in a
Hilbert space H = L 2 [0, L] with a self-adjoint positive definite operator A defined
by (6.5). Let us give a number of corollaries of the abstract Theorem 6.1.1.
Theorem 6.1.3. For solutions of the mixed problem (6.5) the stability inequalities
OTt~T
I u(t,')
II
f(t,') II L2[0,L)
+ I <p Ilwi!o,Lj
+ I ' IIL2[0,L]],
2[0 L] + max I Utt(t,') IIL2[0 Lj
Ilw2
'
O:St:ST
'
O:St:ST
I u(t,')
ft(t,')
IIL2[0,LI + I
max
:s M [O~t~T I
IIwi!o,Lj:S M[oTt~T
f(O,')
IIL2[0,LI + I <p Ilwi[O,L] + II 'ljJ IIWilO,LJ]
hold, where M does not depend on f(t,x) and <p(x), '(x).
317
The proof of this theorem is based on the abstract Theorem 6.1.1 and the
symmetry properties of the space operator generated by problem (6.5).
Second, let n c lRn be a bounded open domain with smooth boundary S,
= nUS. In [0, T] x n we consider the mixed boundary-value problem for
hyperbolic equations
Utt(t,x) - 2)ar (x)u Xr )X r
= f(t,x),x = (Xl, ... ,xn ) E 0,0::; t::; T,
(6.6)
r=l
u(O,x) = <p(x),
au(o,x)
at
='(x),xEn,
u(t,x) = O,X
S,o::; t::; T,
where ar(x), (x EO), <p(x), '(x) (x E 0) and f(t, x) (t E [0, T], x E 0) are given
smooth functions and ar(x) > 0. We introduce the Hilbert space 2(0), the space
of all integrable functions defined on 0, equipped with the norm
Problem (6.6) has a unique smooth solution u(t, x) for smooth ar(x) > and
f(t,x) functions. This allows us to reduce the mixed problem (6.6) to the initialvalue problem (6.1) in Hilbert space H = 2(0) with a self-adjoint positive definite
operator A defined by (6.6). Let us give a number of corollaries of the abstract
Theorem 6.1.1.
Theorem 6.1.4. For solutions of the mixed problem (6.6) the stability inequalities
o~ttT II u(t,) IIWiCi'i)::; M [O~taz:T II f(t,') II L2 Ci'i) + II <p IlwiCO) + II ' IIL2CO)] ,
max II u(t,) IIw 2Co) + O::;t::;T
max II Utt(t,) IlL CO)
O::;t::;T
2
::; M
[O~taz:T II ft(t,) II L2co) + II f(O,)
II L2CO) + II <p IlwiCO) + II ' IIWiCO)]
hold, where M does not depend on f(t,x) and <p(x), '(x).

The proof of this theorem is based on the abstract Theorem 6.1.1 and the
symmetry properties of the space operator generated by problem (6.6).
6.2
Difference Schemes Generated by an Exact Difference Scheme
We consider the initial-value problem (6.1). On the segment [0, T] we consider a

uniform grid
[O,T]T = {tk = kT,k = 0, 1, ... ,N,NT = T}
318
with step r. By Theorem 3.1.1 we have the following two-step difference scheme
for the solution of the initial-value problem (6.1):
1 2 1
-s(r)v(tk+d - -s(r)c(r)v(tk) + -s(r)v(tk-1)
r
r
r
-s(r){
1
r
tk
S(tk+1 - z)f(z)dz +
tk
tk
s(z - tk-df(z)dz},
tk-l
1 < k:S N - 1, v(O) = vo,

r- 1(v(r) - v(O)) = r- 1(c(r) - I)v(O)
or
1
2"(Uk+l - 2Uk
r
!t,k
= r- 1
+ r- 1s(r)v'(O) + r- 1
+ Uk-1)
fk = r- 1{!t,k+l
tk
S(tk - z)f(z)dz, hk
= 2" (c(r) - I)uk

r
s(r - z)f(z)dz
+ fk,
(6.7)
+ s(r)h,k - c(r)!t,d,
= r- 1
tk-l
tk
C(tk - z)f(z)dz, 1 :S k :S N - 1,
tk-l
Uo =
VO,
U1 = c(r)v(O)
+ s(r)v' (0) + r !t,1.
The latter will be referred to as the exact two-step difference scheme for the
initial-value problem (6.1).
Suppose the operators (1 - e2TiB ) and (1 + eTiB ) have the bounded inverses
(I - e2TiB )-1 and (I + eTiB )-l. Then this problem is uniquely solvable and the
following formula holds:
Uk = [e- kTiB - (1 - e2TiB)-1(e-kTiB - ekTiB)]uo
(6.8)
+eTiB(I _ e2TiB)-1(e-kTiB _ ekTiB )u1

+r 2
k-1
L eTiB (1 - e2TiB )-1(e-(k-m)TiB - e(k-m)TiB)fm, 2 :S k:S N.
m=l
Actually, (6.7) can be rewritten as the equivalent initial-value problem for a system
of first-order difference equations
T-I(U' - U'_I)
+ T- I (1 -
r- 1(zk+l - Zk)
+ r- 1(1 - enB)zk = fk, 1 :S k :S N
,-,'B)u'_1 ~ Z" 1 ,; k ,; N,
- 1.
From this follows the system of recursion formulas
Uk = e- TiB uk_1
{
+ rZk, 1 :S k :S N,
Zk+l = eTiBzk +rfk,1:S k:S N-1.
uo, u, are given,

Hence
= e-kTiBu0 +"
LJ
uk
e-(k-S)TiB TZ s,
1<
_ k
319
<
_ N,
s=1
k-1
Zk
= e(k-1)iTB Z 1 + "e(k-1-m)TiB
T ! m, 2 <
LJ
_ k <
_ N.
m=l
+ TZl,
U1 = e-TiBuO
Uk
+L
+ e-(k-1)TiB TZ1
= e-kTiBuo
+L
s-l
e-(k-S)TiB T [e(S-1)iTB Z1
s=2
e(s-l-m)TiB T 1m]
m=l
+ [e-(k-1)TiB + L
k
= e-kTiBuo
e-(k-S)TiBe(S-1)iTB]rZ1
s=2
s-1
e-(k-S)TiB
s=2
e(S-1-m)TiB r 21m,
:s k :s N.
m=1
Since
+L
k
e-(k-1)TiB
e-(k-s)TiB e (s-l)iTB
= (I - e 2TiB )-1(e-(k-1)TiB _ e(k+ 1 )TiB)
s=2
and
L
k
s-1
e-(k-S)TiB
s=2
e(S-1-m)TiB r 2 /m
m=l
L (I -
L L
k-1
e-(k-s)TiB e (s-1-m)TiB r 2 /m
m=l s=m+1
k-1
e2TiB)-1eTiB(e-(k-m)TiB - e(k-m)TiB)r 2 /m,
m=l
we have that
+L
k-1
(I -
e2TiB)-1eTiB(e-(k-m)TiB - e(k-m)TiB)r 2 1m,
2 :s k
:s N.
m=1
Using this formula and the formula rZ1 = U1 - e-TiBuo, we can obtain formula (6.8).
Let us investigate the stability of the exact two-step difference scheme (6.7).
320
Theorem 6.2.1. Suppose the operators (I - e2TiB ) and (I + eTiB) have the bounded
inverses (I - e2TiB )-1 and (I + eTiB)-l. Suppose further that Uo E D(B 2(I +
eTiB)-l), Ul - Uo E D(B 2(I - e2TiB )-1) and ik E D(B(I - e2TiB )-1), 1 ~ k ~
N - 1. Then for the solution of the exact two-step difference scheme (6.7) the
following stability inequalities hold:

(6.9)
+ I (I - e2TiB )-1(Ul - uo) IIH + I (I + eTiB)-luo IIH],
maxllr-1(uk -
l~k~~
~ M[
max I
2~k~N-1
uk-l)IIH
+ maxllr-1(I - e-TiB)uk_lIIH
l~k~~
B(I - e2TiB )-1(ik - fk-1)
IIH + I rB(I -
(6.10)
e2TiB )-1 h IIH
+ I B 2(I - e2TiB )-1(U1 - uo) IIH + I B 2(I + eTiB)-luo IIH],

where M does not depend on r, fk, 1 ~ k
N - 1 and uo,
Ul.
Proof. Using formula (6.8) and the estimates
Il ekTiBII H
< 1 1<
< N,
_k_
---+ H _ ,
(6.12)
we obtain
+r 2
k-l
IleTiB(e-(k-m)TiB - e(k-m)TiB)IIH---+HII(I - e2TiB )-1 fmllH
m=l
+ I (I - e2TiB )-1(Ul - uo) IIH + II (I + eTiB)-luo IIH]

for k 2 2. It obviously holds also for k = 0,1. The estimate (6.9) is established.
Applying r- 1(I - e- TiB ) to formula (6.8) and using estimates (6.12) and
(6.13)
321
in a similar manner as (6.9), we obtain
117- 1 (I -
e-TiB)uk_11IH ~
117- 1 (I -
e-TiB)B-11IH-->H
x{ll(I + eTiB)e-(k-1)TiB - (e-(k-1)TiB - e(k-1)TiB)IIH-->H
B(I + eTiB)-lUO
IIH
+lle-(k-1)TiB - e(k-1)TiBIIH-->H I B(I - e2TiB )-1(U1 - Uo) IIH

k-2
+7
IIeTiB(e-(k-1-m)TiB - e(k-1-m)TiB)IIH-->HI17B(I - e2TiB )-1 fmllH}
m=l
+ I B(I - e2TiB )-1(U1 - Uo) IIH + I B(I + eTiB)-luo IIHl

for k 2: 3. It obviously holds also for k = 1,2. So, we have that
+ II B(I - e2TiB )-1(U1 - uo) IIH + I B(I + eTiB)-luo IIH]'
;2
Estimate (6.10) is established. Now, we obtain the estimate for I

(C(7)- I)Uk IIH.
Applying
(C(7) - I) = ;2e-TiB(I - eTiB )2 to formula (6.8) and using Abel's
formula, we can write the formula
;2
(6.14)
-(I + eTiB)-l(e-kTiB - ekTiB)]uo
+~(I
- eTiB )2(I - e2TiB)-1(e-kTiB - ekTiB)(U1 - Uo)
2
7
+~(I - eTiB )
7
k-1
L 7(1 - eTiB)(I - e2TiB )-1(e-(k-m)TiB - e(k-m)TiB)fm
m=l
322
+~(I
- eTiB )2(I - e2TiB)-I(e-kriB - ekriB)(UI - Uo)
2
7
L 7(1 - e2TiB )-I(e-(k-m)TiB + e(k+l-m)TiB)(Jm k-I
+~(I - eTiB)
7
fm-I)
m=1
+~(I - eTiB)(e-(k-I)TiB + ekTiB)(I _ e2TiB )-17II

7
-~(I - eTiB)(I + eTiB)(I - e2TiB )-17fk_I'

7
Using the last formula and estimates (6.12) and (6.13), we obtain
~ II(c(7) -
I)ukllH :S
II~(I
7
eTiB)B-Illk--+H
x IlcTiBIIH--+HII(I + eTiB)e-kriB - (e- kTiB - ekTiB)IIH--+H

xIIB 2(I
+ eTiB)-luoIIH + II~(I - eTiB)B-Illk--+HllckriB - ekriBIIH--+H

7
x1lB 2(I - e2TiB )-I(UI - uo)IIH
+ II~(I 7
eTiB)B-IIIH--+H
k-I
X
L 711(e-(k-m)TiB + e(k+l-m)TiB)IIH--+HIIB(I - e2TiB )-I(Jm -
fm-dllH
m=1
+11 ~ (I 7
eTiB)B-IIIH--+H Ile-(k-I)TiB
+II~(I 7
+ ekTiB IIH--+HI17 B(I -
e2TiB )-1 II IIH
eTiB)B-IIIH--+HIII + eTiB IIH--+HI17B(I - e2TiB )-1 fk-IIiH
I B(I - e2TiB )-I(Jm - fm-I) IIH + II 7B(I - e2TiB )-1 II IIH

2::;m::;N-I
+ II B 2(I - e2TiB )-I(UI - Uo) IIH + I B 2(I + eTiB)-luo IIH]
:S M[ max
for k 2: 2. It obviously holds also for k = 1. So, we have that

2
I::;~::;~J 72 (C(7) - I)UkIiH
II 7B(I - e2TiB )-I(Jk - fk-I) IIH + I 7B(I - e2TiB )-1 II IIH

2::;k::;N-I
+ II B 2(I - e2TiB )-I(UI - uo) IIH + II B 2(I + eTiB)-luo IIH]'
:S M[ max
The estimate for
max
1::;k::;N-I
117-2(uk+l-2uk+uk-dIiH follows from the last estimate
and the triangle inequality. Theorem 6.2.1 is proved.

Note that we have not been able to obtain the stability inequalities
:S M[ max 117- 1(fk - fk-I) IIH + I II

2"S.k"5.N-1
+ II B7- 1(U1 - Uo) IIH + I B2uo IIH]
323
IIH
for the solution of the exact two-step difference scheme (6.7). Nevertheless, the
following result holds.
(1 - e2TiB ) and (1 + eTiB) have the bounded

2TiB
inverses (1 _e
)-1 and (1 +eTiB)-l. Suppose that Uo E D(B 3), U1 -Uo E D(B 2)
and fk E D(B), 1 :S k :S N - 1. Then for the solution of the exact two-step
difference scheme (6.7) the following stability inequalities hold:
Theorem 6.2.2. Suppose the operators
omc~~ukIIH :S M[l"5.~tJfkIIH+
maxI17- 1 (Uk
l"5.k"5.~
Uk-1)IIH
117- 1 (U1
uo)
max 117- 1 (1
1"5.k"5.N
IIH + I
7,
IIH],
e-TiB)uk_11IH
:S M[ max I B7- 1(fk - fk-1) IIH + II BII

2"5.k"5.N-1
+ I B 27- 1(U1 - Uo) IIH + I B3uo IIHl
Buo
fk, 1 :S k :S N - 1 and uo,
(6.15)
(6.16)
IIH
U1.
Proof. Using formula (6.8) and estimates (6.12), (6.13) and
11(1 -
e2TiB)-1(e-kTiB - ekTiB)IIH-->H :S k,
(6.18)
we obtain
IlukllH:S Ile-kTiBIIH-->H
I Uo
IIH
+117- 1 (1 -
e-TiB)B-11IH-->H
x11(1 - e2TiB)-1(e-kTiB - ekTiB )IIH-+H 7 11 Buo IIH

+11(1 - e2TiB)-1(e-kTiB - ekTiB)IIH-->H7117-1(u1 - uo) IIH
+7 2
k-1
m=l
Ile TiB IIH-->HII(1 -
e2TiB )-1(e-(k-m)TiB - e(k-m)TiB)IIH--+HllfmIIH
324
for k ~ 2. It obviously holds also for k = 0,1. Estimate (6.15) is established.

Applying T- 1(1 - e- TiB ) to formula (6.8) and using estimates (6.12), (6.13) and
(6.18) in a similar manner as (6.15), we obtain
IIT- 1 (1 -
e- TiB )uk_11IH ~
IIT- 1 (1 -
e- TiB )B- 11IH--+H
x{lle-(k-1)T iB IIH--+H I Buo IIH +IIT- 1 (1 - e- TiB )B- 11IH--+H

xll(1 - e2TiB )-1(e-(k-1)TiB - e(k-1)TiB)IIH--+H I B2uo IIH
+11(1 -
e2TiB )-1(e-(k-1)TiB - e(k-1)TiB)IIH--+HT
II BT- 1(U1
- uo)
IIH
k-2
+T
L IleTiBIIH--+HII(I - e2TiB )-1(e-(k-1-m)TiB - e(k-1-m)TiB)IIH--+HIITBfmIIH}
m=l
for k
3. It obviously holds also for k = 1,2. So, we have that

max IIT- 1 (1
19~N
e- TiB )uk_IiIH
1
~ M[l~~~JB!kllH+I BT- (U1 - uo) IIH
+ II B2uo IIHJ.
~ M[l~~~JBfkllH+I BT- (U1 - uo) IIH

1
+ I B2uo IIHJ.
Estimate (6.16) is established. Now, we obtain the estimate for II ~(C(T)- I)uk
Using formula (6.14) and estimates (6.12), (6.13) and (6.18), we obtain
2211 (c(T) - I)ukllH
~ /I~(1 T
IIH.
eTiB)B-11Ik--+Hlle-TiBIIH--+H[lle-kriBIIH--+H
xllB2uoliH + 11~(1
- eTiB )B- 11IH--+H
T
xll(I - e2TiB)-1(e-kriB - ekriB)IIH--+HIIB3uoIIHJ
+11~(1 T
+II~(I -
eTiB )B- 11Ik--+HII(1 - e2TiB)-1(e-kTiB - ekriB )IIH--+HIIB 2(Ul - uo)IIH

eTiB )B- 1/1H--+H
IIBUm -
TII(1 - e2TiB )-1(e-(k-m)TiB - e(k-m)TiB)/IH--+H
m=l
k-1
k-1
fm-l)IIH+ L 11(1 _e TiB )(1 _e2TiB )-1(e(k-m)TiB +e(k+1-m)TiB)IIH--+H

m=l
325
xll(Jm - fm-I)IIH + 11~(1 - eTiB)B-IIIH.....H

T
x11(1 - e2TiB )-I[e-(k-I)TiB - e(k-I)TiB]IIH.....H
xllTBlIllH
eTiB )(1 - e2TiB )-I[e(k-I)TiB + ekTiB ]IIH.....HIIIIIIH

eTiB )(1 - e2TiB )-I[1 + eTiB ]IIH.....HIIIk_IIIH
+ 11(1 -
+11(1 ~ M[ max II BT-I(Jm 2~m~N-I

+ II BII
fm-I)
IIH + 2~m~N-I
max II T-I(Jm -
fm-I) IIH
+ II II IIH + I B 2T- I (UI - uo) IIH + II B3uo IIH +II B2uoIIH]

~ Md max II BT-I(Jm - fm-d IIH + I BII IIH
2~m~N-I
IIH
+I
2 I
B T- (UI - uo) IIH
+I
B3uo
IIH]
for k 2': 2. It obviously holds also for k = 1. So, we have that

2
I~~~J T2 (C(T) - I)ukllH

~ M I [ max
29~N-I
+ II
The estimate for
max
l~k~N-I
I BT-I(1k -
fk-I) IIH
B 2T- I (UI - uo) IIH
+I
+ II
BII
IIH
B3uo IIH]'
IIT-2(uk+I-2uk+uk-dIIH follows from the last estimate
and the triangle inequality. Theorem 6.2.2 is proved.

Remark 6.2.3. Theorem 6.2.2 actually holds in an arbitrary Banach space E under
the assumptions
(6.19)
where M does not depend on T.

Now, we will consider the applications of the exact difference scheme (6.7).
From (6.7) it is clear that for the approximate solutions of problem (6.1) it is
and
itt
tk-l
S(tk - z)f(z)dz,
itt
C(tk - z)f(z)dz.
tk-l

1
1
to construct fk' and 1i;1 such that
(6.20)
(6.21)
326
and the formulas of
fk''I
and
Ji:l'I
Jk''I
are sufficiently simple. The choice of
and
ff:~ is not unique. Using Taylor's formula and integration by parts, we obtain the
representation
(6.22)
l;
]+1-1
(3mf(m)(tk-l)
+ 7- 2
itk
tk-l
is ((j ~~ _
)]+1-1
8(tk - z)
I)! f(j+I)(z)dzd8,
tk-l
in which
In a similar manner one can obtain that

(6.23)
1m = (3m-l, 1:::;
Using the definitions of 8(7) and

Chapter 2), we can write
m:::; j + l-
C(7)
and Pade fractions for the function e- z (see
C(7) = R;,L1(iTB)+2R ;,l1(-iTB)

{
8(7)
1,'0 = 7- 28(7).
+ 0(7]+1+2),
+ 0(7 j +1+2).
Now, using formulas (6.20), (6.21), (6.22), (6.23) and (6.24)

defined by the following formulas:
j,1 fk -
j~1 B
~
m=O
f(m)(
tk
]+1-1
(6.24)
= B- 1 Rj,ll(iTB)~~j,ll(-iTB)
I
fk'I
and fL can be
B-1 R j ,I+!(i7B) - Rj,I+!( -i7B)

2i
. .
]+1-1
"B
f(m)(t
)_Rj ,l+l(Z7B)+Rj ,l+l(-Z7B)"
~
m
k-l
2
~
m=O
m=O
]+1-2
j,l -f 1,1
"~
m=O
'
Bm f(m) (0) ,)0fj,l
j ,1
-- 7f1,1'
B f(m)(t
m
(6.25)
)
k-l ,
(6.26)

where
327
Tm- 2
B m = _A- l _,_ + A-I B m- 2, 2 ~ m ~ j + I-I,

m.
B = T- 2A_ l (R j ,I+l(iTB) +2Rj ,I+I(-iTB) - 1),
o
B = T- 2A-l(B-l Rj ,I+I(iTB) ~iRj,I+I( -iTB) - T1),
l
Bm
= Bm -
l ,
~ m ~
+l -
1,
-2B-l Rj ,I+1 (iTB) - Rj,l+l( -iTB)

B 0= T
2i
.
Now, using formulas (6.24), (6.25) and (6.26), we obtain the difference schemes to
a (j + l)-th order of accuracy
2
T- (Uk+l - 2Uk
A=
+ Uk-I) + AUk
'I
= f~'
,1 ~ k ~ N - 1, Uo = vo,
(6.27)
2T-2(1 _ Rj ,I+I(iTB) +2Rj ,I+l(-iTB)),
-1(
)
_1(Rj ,l(iTB) + Rj,l( -iTB) I)
T Ul - Uo = T
2
- Vo
+T- l B- 1Rj,l(iTB) ~iRj,I(-iTB) v/(O)
fJ,1
for the approximate solution of the initial-value problem (6.1).

Note that the difference schemes (6.27) for j = l ,j = I - I and j = l + 1
include difference schemes of arbitrary high order of approximation. Moreover,
the corresponding functions IRj,I+l(z)1 tend to 0 as z -+ 00 for j = l - 1,1 and
IRj,l+l (z)1 = 1 for j = l + 1. Such difference schemes are simplest, in the sense that
the degrees of the denominators of the corresponding Pade approximants of the
function exp{ - z} are minimal for a fixed order of approximation of the difference
schemes.
Suppose the operators (1 - R;'I(iTB)) and (1 + Rj,l(iTB)) have the bounded
inverses (1 - R;,l(iTB))-1 and (1 + Rj ,l(iTB))-I. It is clear that this problem is
uniquely solvable and the following formula holds:
Uk = [Rj,l( -iTB) - (1 - R;'I(iTB))-I(Rj,l(-iTB) - Rj,l(iTB))]uo
(6.28)
+Rj,l(iTB)(I - R;'I(iTB))-I(Rj,l( -iTB) - Rj,l(iTB))Ul

+T 2
k-l
m=1
Rj,l(iTB)(I - R;'I(iTB))-I(R;,lm( -iTB) - R;,lm(iTB))f?;.l, 2 ~ k ~ N,
_ Rj,l(iTB) + Rj,l( -iTB)

Ul 2
Vo
B- 1Rj,l(iTB) - Rj,l( -iTB) /(0)

2i
v
+T
fj,l
)0 ,
Uo = Vo
328
Theorem 6.2.4. Suppose the operators (I - R;'I (iT B)) and (1 + Rj,1 (iT B)) have
the bounded inverses (1 - R;'I(iTB))-l and (1 + Rj,l(iTB))-l. Suppose that Uo E
D(B 2), U1 - Uo E D(B). Then for the solution of the two-step difference schemes
(6.27) for j = land j = l - 1 the following stability inequalities hold:
max IlukllH ~ M[
o9~iJ
max
IIB- 1fk,lllH+ I B-1T-1(U1 -
19~N-i
maxIlT-1(Uk - uk-1)IIH
l~k~~
'1
~ M[l~~~JfZ' IIH+
2~k~N-1
+ II
IIH],
maxIlT-1(I - e-TiB)uk_1llH
II T-1(U1
- uo) IIH
IIT- (Uk+1 - 2Uk + uk-dllH
~ M[ max
+ II Uo
l~k~~
max
19~N-1
uo) IIH
II T-1Ue
+ 19~Nmax !IAukliH
+ II Ii,l
- fk~l) IIH
IIH + I B2uo
BT-1(U1 - Uo)
'1
where M does not depend on T, fZ' , 1 ~ k
+ II Buo IIH],
IIH
IIH],
N -1 and Uo, U1.
and relies on formula (6.28) and on the estimates
IIRj,I+1(iTB)IIH~H~ 1,
(6.29)
IITB(I - R;'I+l (iTB))-l Rj,I+1 (iTB)IIH~H ~ M.
(6.30)
Note that estimate (6.30) is not satisfied for j = l + 1. Therefore we have not been
able to obtain the same stability inequalities for the solution of the exact two-step
difference scheme (6.27) for j = l + 1. Nevertheless, the following result holds.
+ Rj,I(iTB)) have
the bounded inverses (I - R1,I(iTB))-1 and (I + Rj,l(iTB))-l. Suppose that Uo E
'1
D(B 3), U1 - Uo E D(B 2) and fZ' E D(B), 1 ~ k :S N - 1. Then for the solution of the two-step difference schemes (6.27) for j = l + 1 the following stability
inequalities hold:
Theorem 6.2.5. Suppose the operators (I - R;'I(iTB)) and (I
maxllukliH :S M[ max !lfk,IIIH+
o9~~
19~N-
maxIlT-1(Uk - uk-dliH
l~k~~
:S M[l~~~JBfk,IIIH+
max
19~N-1
II
2~k~N-1
- uo) IIH
+ II Buo IIH],
maxIIT-1(I - e-TiB)uk_1llH
l~k~iJ
BT-1(U1 - uo) IIH
IIT- (Uk+1 - 2Uk
:S M[ max
II T-1(U1
+ II
B2uo IIH],
+ uk-dliH + 19~Nmax !IAukliH
l
1
II T- BUt
Jk~l) IIH
+ II Bfi,l
+ II B 2T- 1(U1 - uo) IIH + II B3uo IIH],

where M does not depend on T, fZ' 1,1 :S k:S N -1 and Uo, U1.
IIH
329
The proof of Theorem 6.2.5 follows the scheme of proof of Theorem 6.2.2 and
relies on formula (6.28) and on estimates (6.29) and
11(1 -
R j ,l+1(2irB))-1 (Rj,l+l (-irB) - Rj,l+l(irB))IIH-+H :S k.
(6.31)
Remark 6.2.6. Theorem 6.2.5 actually holds in an arbitrary Banach space E under
the assumptions
IIRj,l+l(irB)\\E-+E :S M,l :S k:S N,

where M does not depend on r.
(6.32)
Now, the abstract Theorems 6.2.4 and 6.2.5 are applied in the investigation
of difference schemes of higher order of accuracy with respect to one variable for
approximate solutions of the mixed boundary-value problem (6.6). The discretization of problem (6.6) is carried out in two steps. In the first step let us define the
grid sets
o:S
:S Nr,hrNr = L,r = 1, ... ,n},
mr
Oh
n n 0, Sh = n n S.
h
We introduce the Banach space L 2(n h) of the grid functions eph(x) = {ep(h1ml'
... , hnmn )} defined on h , equipped with the norm
II eph
II L2 (i'i h )=
(2: leph(xWh 1 h )!.

n
xEn h
To the differential operator AX generated by problem (6.6) we assign the difference

operator AI': by the formula
(6.33)
acting in the space of grid functions uh(x), satisfying the conditions uh(x) = 0
for all x E Sh. It is known that AI': is a self-adjoint positive definite operator in
L2 (n h ). With the help of AI': we arrive at the initial-value problem
d2v h(t, x)
dt 2
V
+ Ahv (t, x)
= f (t, x),
:S
:S T, x E Oh,
h
h
dvh(O,x)
h
(O,x) = ep (x),
dt
= 1/J (x), x
(6.34)
Oh

In the second step we replace problem (6.34) by the difference schemes (6.27)
r-2(u~+1 (x) - 2u~(x)
tk
+ ULI (x)) + Al':u~(x) = ff(x),
= kr, 1 :S k
:S N - 1, x E
h,
(6.35)
330
u~(x) = 'Ph(X), T-1(U~(X) - u~(X)) = T- 1(
R l(iTBX) + R-I(-iTBX)
J,
h 2 Y,
h - I) 'Ph (x)
+T- 1(B fJ-1 Rj,l(iTBfJ ~iRj,l( -iTBfJ 1/Jh(X) + Tf/;(x), x

A
ff(x)
h = 2T-
2(1 - R j,l+l(iTBh) +2Rj,l+l(-iTB'h\ (B'hf
Dh,
= A h,
j+l-l
' BX)
~ B f(m)(t
) + B- 1R j,l+l ('tT BX)
h - R j,I+1 (-tT
h
L.J m
k, X
2i
m=O
]+1-1
- ()
L.J Bmf
m (tk-1,X)
m=O
'BX)
_ R j,l+l (tT
h
+2R j,l+l ('BX)]+I-l

-tT h ~ B f(m)(
)
L.J m
tk-l,X,
m=O
]+1-2
f/;(x) = T
Bmf(m)(o,X),
m=O
where
m-2
B m = -(A h)-1 ~
m,
+ (A h)-1 B m - 2, 2 :::; m
:::; j
+ l - 1,
B = T- 2(A'h)-I(R j,I+1 (irB,,) +2 Rj,l+l (-irB'h) - 1),

o
B 1 = T- 2(A )-I((Bh)-1 R j ,l+l(iTB'h) ~iRj,I+l(-iTB'h) - T1),
Bm
= Bm -
1,
1 :::; m :::; j
+l -
1,
- _ -2(BX)-IRj,l+l(iTB'h) - Rj,I+l(-iTB'h)
B O-T
h
~
,
Theorem 6.2.7. Let T and Ihl be sufficiently small numbers. Then the solutions of
the difference schemes (6,35) for j = land j = l - 1 satisfy the following stability
estimates:
+ II (Bh)-I1/Jh II L2 (oh) + II 'P h IIL 2(oh)],

h
max IIT- 1(u h
max
lS:kS:N
k - u k-l ) II L2(f2- h )<- M 1[OS:kS:N-l
II
fhk II L2( f2- h)
331
+ II f~ II L2 (nh) + 2: 7- 117jJh - r II L2 (n h) + 2: 11(<p~r )xr,m r } II L2 (nh)]'

1
xr,m
r=l
Here M 1 does not depend on
7,
r=l
h, <ph(x), 7jJh(x) and fr(x), 0::; k::; N - 1.
The proof of Theorem 6.2.7 is based on the abstract Theorems 6.2.2 and the
symmetry properties of the difference operator AI; defined by formula (6.33).
and Ihl be sufficiently small numbers. Then the solutions of

the difference schemes (6.35) for j = l + 1 satisfy the following stability estimates:
Theorem 6.2.8. Let
+ II7jJh II L2 (nh) + 2: II<phxr,m

- r II L2 (n h)],
r=l
:::; MIl 1S:;kS:;N-1

max
II 7- 1 L
IIUt) xr,m
- r - Ut-I) xr,m
- r II L2(0.h)
r=l
+
r=l
r=l
L IIU~)xr~mrIIL2(nh) + 2: 11(7jJ~)xr,mr
Here M 1 does not depend on
7,
II L2 (n h) + L 11((<p:--)xr)xr,mr II L2 (n h)]'
r=l
h, <ph(x), 7jJh(x) and ft(x),O::; k::; N - 1.
Note that in a similar manner one can construct the difference schemes of a
the boundary-value problem (6.5). Abstract Theorems 6.2.2 and 6.2.4 permit us
to obtain stability estimates for the solutions of these difference schemes.
332
6.3
Two-Step Difference Schemes Generated

We consider again the initial-value problem (6.1). For the construction of the twostep difference schemes of an arbitrary high order of accuracy for the approximate
solutions of the initial-value problem (6.1) we consider again a uniform grid space
[0, T]r =
{tk
= kT, k = 0, 1, ... , N, N 7 = T}.
Let the function v(t)(O ~ t ~ T) have a (2l + 2j + 2)-th continuous derivative

and tk-1, tk, tkH E [0, T]r- Then by Theorem 3.2.2 we have the following Taylor
decomposition on three points:
V(tk+1) - 2V(tk)
+ v(tk-d -
La
s v(2S) (tk)7
2s
(6.36)
s=1
j
-L
7}s(v(2s)(tk_1)
+ V(2s) (tk+d)7 2s
= 0(7 21 +2)+2),
s=1
where 7}m, m = 1, ... ,j

7}s = (2~)!
is a solution of the system

s-1
-
m=1
for any 8,
(2s2'2m)!'
"(2s)!1),,,
1
LJ (2s-2m)! = ,lor
m=1
+ 1 ~ 8 ~ j,
(6.37)
any 8, J. + 1 <
_ 8<
_
+ J,.
and
1
s-1
1
a1 = 1 - 27}1, as = ((28)! - 7}s - f17}m (2(8 _ m))!)2 for any 8, 2 ~ 8 ~ l.
Suppose further that the function v(t) (0 ~ t ~ T) has a (2l + 2j + 1)th continuous derivative. Then by Theorem 2.2.1 we have the following Taylor
decomposition on three points:
2j
+L
V(7) - v(O)
21
8s v(S) (7)7 S -
s=1
where
(21+2j-s)!(21)!
Ps = (21+2j)!s!(21-s)!
{
8 =
s
Psv(s) (0)7 S = 0(7
21 2
+ )+1),
s=1
or any 8, 1 ~
(21+2J-s)!(2J)!(-1)8
(21+2J)!s!(2J-s)!
for an
y
8,
8 ~ 2l,
1~
8 ~
2j.
(6.38)
333
Now, we will consider the applications of the Taylor decomposition (6.36) of

function vet) on three points and Taylor's decomposition (6.38) of function vet)
on three points to approximate solutions of the initial-value problem (6.1). From
(6.36) and (6.38) it is clear that for the approximate solution of the problem (6.1)
it is necessary to find v(j)(r) for any s, 1 :S s :S 2l, V(2s)(tk) for any s, 1 :S s :S l
and v(2s)(tk_t}, V(2s) (tk+t} for any s, 1 :S s :S j. Using the equation
V"(t) = -Av(t)
+ Jet),
we obtain
= (-A)nv(t) + 2) _A)n->. j(2)'-2)(t),

n
v(2n)(t)
= 2, ... ,
(6.39)
>'=1
= (-A)nv'(t) + I) _A)n->. j(2)'-1)(t),

n
v(2n+1)(t)
= 1,....
(6.40)
>'=1
Suppose further that the function v(t)(O :S t :S T) has a (2l + 2j - 2m + 2)th continuous derivative. Then by Theorem 2.2.1 we have the following Taylor
decomposition on two points:
vCr) - v(O)
21-2[TI
aSv(S) (r)r S -
2j-2m+2[T]+1
s=l
where
bsv(s)(O)r S = o(r21+2j-2m+2),
s=l
(6.41)
_ (21+2j-2m+l-s)!(21-2[T])!(-1)S
_ !!!J
as - (21+2j+l-2m)!s!(2p- 2 ITI- s )! for any s, 1:S s :S 2l 2[ 2 '
(21+2j+l-2m-s)!(2j-2m+2[1"]+l)!
bs - (21+2j+l-2m)!s!(2j-2m+2[1"J+1-s)! lor any s,

1 :S s :S 2j - 2m + 2[~J
+ 1,
where [aJ denotes the integer part of the number a. Further using formulas (6.39),
(6.40) and (6.41), we can write
v'(r) - v'(O)
l-ITI
a2n( -Atv'(r)r2n
n=l
l-[T]
a2n_1(-Atv(r)r 2n - 1 +
n=l
j-m+ITl
2:=
n=l
l-[T)
n=l
I-tTl
L( _A)n->. j(2).-1) (r)r2n

>'=1
L( -At->' j(2).-2) (r)r2n-1

n
a2n-1
n=l
b2n (-A)n v'(0)r 2n -
a2n
>'=1
j-m+[T]
2:=
n=l
b2n
2:=( _A)n->. j(2)'-1)(0)r2n

>'=1
334

j-m+ITI+l
2:
b2n_l(-A)nv(O)r2n-l
n=l
j-m+ITI+l
2:
2:( _At- Aj(2A-2)(O)r 2n - 1

n
b2n - 1
n=l
A=l
= o(r21+2j-2m+2), O:S m:S
j-l.
From the last formula it follows that
(I
l-[T)
j-m+[T)
n=l
n=l
2: a2n(-Atr 2n )v'(r) = (I + 2:
l-[TI
j-m+IT)+l
n=l
n=l
- 2: a2n-l( _A)n v(r)r 2n - 1 + 2:
b2n (-A)n r 2n)v'(O)
b2n - 1(_A)n v(O)r 2n - 1
I-IT]
l-[T]
n=l
A=l
n=l
A=l
- 2: a2n 2:( -At-A j(2A-l)(r)r2n - 2: a2n-l 2:( -At-A j(2A-2) (r)r2n-1

j-m+ITJ
b2n L(-A)n-Aj(2A-l)(O)r 2n
n=l
A=l
j-m+[T)+l
2:
n=l
2:( _A)n-A j(2A-2) (O)r

n
b2n - 1
2n - 1
A=l
+o(r21+2j-2m+2), O:S m:S j - l .

Suppose further that the operator (I
verse. Then
v'(r) = (I
I-IT]
- 2:
n=l
+ L~-jrl a2n( _A)n r 2n)
l-[TI
j-m+[T]
n=l
n=l
2: a2n(-A)nr 2n)-1{(I + 2:
a2n-l( _A)nv(r)r 2n - 1 +
j-m+[T]+l
2:
has a bounded in-
b2n (-Atr 2n )v'(O)
(6.42)
b2n - 1(-Atv(O)r 2n - 1
n=l
I-ITI
I-ITl
n=l
A=l
n=l
A=l
- 2: a2n 2:( _A)n-A j(2A-l)(r)r 2n - 2: a2n-l 2:( _A)n-A j(2A-2)(r)r 2n - 1

j-m+[~1
b2n
n=l
L( _A)n->. f(2)'-1)(0)r 2n
>'=1
n
j-m+[~J+1
335
b2n - 1
n=l
L( _A)n->. f(2).-2) (0)r 2n - 1}

>'=1
:s m :s j
+o(r2l+2j-2m+2), 0
- 1.
Now, using formulas (6.36), (6.38), (6.39), (6.40) and (6.42), we obtain the difference schemes of a (2l + 2j)-th order of accuracy:
Uk+1 - 2Uk
r
+ Uk-1
'"
+ Uk+1 ) =
fj,l
k ,
L TIs L( -A)S->'U(2).-2)(tk+1) + f(2).-2) (tk_1))r 2s - 2

j
s=l
>'=1
s=l
>'=1
2 1<k < N
+'" Q s "'(-A)S->'f(2>'-2)(tk
)r 2s -,
_
_
n=l
l-[~I
a2n(-Atr 2n )-1{(I +
62m-1( _A)m r 2m-2
j-m+[~1
b2n (-A)n r 2n)v'(0)
n=l
l-[~l
a2n-1( _A)n r 2n-1 U1
j-m+[~I+l
n=l
b2n - 1(_A)n r 2n-1 v (0)}
n=l
L P2n( -Atr2n - 1 2) _A)n->. f(2)'-2)(0)

1
n=l
>'=1
=
1
= v 0,
m=l
n=l
- 1, U0
L 62n( -Atr2n - 1U1 + L
x(I
(6.43)
Uk
s=l
" TIs ( - A)S r 2s-2 (Uk-1

- '~
s=l
fk,l
(A)S 2s-2
~Qs -
L
n=l
P2n-1( _A)n-1 r 2n-2
L( -At->' f(2).-1)(0)
>'=1
336

n
- L 82n (-AtT 2n - 1 L( _A)n->. j(2).-2)(T)

n=l
>'=1
j
n-1
- L 82n - 1(_A)n T2n-2 L(-At->' j(2)'-1)(T)
n=l
>'=1
l-[~I
- L 82m - 1(_A)m T2m-2(I + L a2n( _A)n T2n)-1

m=l
n=l
l-[~J
l-[~]
X{ - L a2n L( _A)n->'j(2)'-1)(T)T 2n - L a2n-1 L( -At->'j(2).-2) (T)T 2n - 1

n=l
>'=1
n=l
>'=1
n
b2n L(_A)n->. j(2)'-1)(O)T 2n
>'=1
j-m+[~]+l
n
+
L
b2n_1L(-A)n->'j(2>.-2)(O)T2n-1}
n=l
>'=1
j-m+[~)
L
n=l
for the approximate solution of the initial-value problem (6.1). Suppose that the
operator
(I
n=l
m=l
+ L82n(-AtT2n - L 82m _ 1(-A)m T2m-1(1

l-[~]
l-[~l
+ L a2n(-A)n T2n)-1 L a2n_1(-A)n T2n-1)

n=l
n=l
has a bounded inverse. Suppose further that the operator (I - L~=l1]S( _A)ST 2s )
has a bounded inverse and
-l(2I
+ L a s(_A)ST 2S )2(1 - L 1]s( _A)ST 2s )-2 2:: O.

s=l
s=l
This problem is uniquely solvable and the following formula holds (see
Chapter 1)
Uk
= (Rj,I(-iTA) - Rj,l(iTA))-lRj,l(iTA)Rj,I(-iTA)(R;,"i 1(iTA) (6.44)

-RJ,ll(-iTA))uo
+(Rj,I(-iTA) - Rj,l(iTA))-l(Rj,I(-iTA) - Rj,l(iTA))U1

k
+ L(i~j,l( -irA) -
Rj,l(irA))j) Rj,l(irA)Rj,l(-irA)
j=l
L 7]s(-A)Sr2s )-1 ft;l,
x (R7,i m (-irA) - R7,i m (irA))(I -
2:S k:S N,
s=l
Here
1
Rj,l(irA) = 2(21 +
L
l
s=l
1
i 1- 4(21 +
L
l
s (-A)sr 2s )2(I -
s=l
n=l
m=l
l-[~l
l-[~]
a2n(-A)n r 2n)-1
n=l
X{LP2n(-At r2n - 1 +
n=l
l-[~I
a2n(-Atr 2n )-1
62m_1(-A)m r 2m-1
a2n_1(-Atr 2n - 1)-1
62m_1(-A)mr 2m-2
j-m+[~1+1
b2n_1(-Atr2n-1},
n=l
+ 2: 82n (-Atr2n n=l
2: 82m - 1(_A)mr 2m-1
m=l
l-[~I
l-[~]
n=l
n=l
2: a2n(-Atr 2n )-1 2: a2n_1(-Atr 2n - 1)-1
x(I +
X{LP2n_1(-A)n-1 r 2n-2
n=l
x(I +
m=l
= (I
L 7]s( _A)sr 2s )-2,
n=l
sr
s=l
n=l
x(I +
s=l
cr = (I + L(hn(-A)nr 2n - L
x(I +
L 7]s( _A)Sr2s )-1
2s
s (-A)Sr )(I -
l-[~]
+L
a2n(-Atr 2n )-1(I +
j-m+[~l
n=l
fg,l
62m_1(-A)m r 2m-2
m=l
b2n(-A)nr2nn,
n=l
2: P2n( _A)n r 2n-1 2:( _A)n->. f(2)-2) (0)

I
n=l
>'=1
337
338
1
+ 2: P2n-l (-At-lr2n-22:( -At-A f(2A-l)(O)

n
n=l
A=l
2:(
n
-2: 02n( _A)n r 2n-l

_A)n-A f(2A-2)(r)
n=l
A=l
j
n-l
02n-l (_A)n r 2n-22:(-At-A f(2A-l)(r)
n=l
A=l
-2:
I-[~]
-2: 02m_l(-A)mr 2m-2(I + 2: a2n(-Atr2n )-1

m=l
I-[~]
X{-2:
n=l
I-I~]
a2n2:(-A)n- Af(2A-l)(r)r 2n -2: a2n_12:(-A)n- Af(2A-2)(r)r 2n - 1

n=l
A=l
n=l
A=l
n
j-m+[~]
2:
b2n 2:(_A)n-A f(2A-l)(O)r 2n

A=l
n=l
j-m+[~]+l
2:
n=l
b2n - 1 2:( _A)n-A f(2A-2)(O)r 2n - 1}.

A=l
From formula (6.44) it follows that the investigation of the stability of difference
schemes (6.43) relies in an essential manner on a number of properties of the powers
of the operator Rj,I(irA). We were not able to obtain estimates for powers of
the operator Rj,1 (irA) in the general cases of numbers j and l.
Theorem 6.3.1. Suppose that Uo E D(A), Ul - Uo E D(A~). Then for the solution of the two-step difference schemes (6.43) for l = 0 the following stability
inequalities hold:
maxllukllH :S M[ max IIA- 1/ 2/k,IIIH+ II A-~r-l(ul - uo) IIH + II Uo IIH],
O:Sk:SN'
l:::;k:SN-i
max1Ir-1(Uk -
l:Sk:SN
uk-dllH
'I
:S M[l:s~~Jfk' IIH+ II r-1(Ul - Uo) IIH + II A2Uo IIH],

max
l:::;k:SN-l
:5 M[ max
2:Sk:SN-l
2
Ilr- (Uk+l -
2Uk + uk-dllH
II r-1Uk,1 - fk~l) IIH + II
+ II A~r-l(Ul - Uo)
'I
IIH
H,l IIH
+ II Auo IIH],
where M does not depend on r, /k' ,1 :S k :5 N - 1 and Uo, Ul
339
relies on formula (6.44) and on the estimates
(6.45)
(6.46)
11(1 -
L 1]s( _A)S72S )-1 IIH--+H :s M.
(6.47)
s=1
Remark 6.3.2. Theorem 6.3.1 actually holds in the general cases of numbers j and
l under the assumption
711AIIH--+H ----; 0 when
7 ----;
O.
Remark 6.3.3. Applying formulas (6.22), (6.24), (6.26), (6.36), (6.39) and (6.42),
we can construct the difference schemes of a (2l + 2j)-th order of accuracy
Uk+l - 2Uk
7
+ Uk-l
~as
(A)S 2s-2
7
Uk
(6.48)
s=1
( A)S 7 2s-2( Uk-l

- ~
~ 1]s -
+ Uk+l ) =
fj,1
k ,
s=1
fk,l =
s=1
A=1
L 1]s L( -AV- AU(2A-2l(tk+l) + f(2A-2)(tk_l))7 2s - 2
s=1
A=1
2 1<k <N
+ ~ a s ~(_A)S-Af(2A-2)(t k
)7 2s -,
_
_
-1(
)
Ul - Uo =
-1 (R 21,2j (i7B)
- 1,U
0
= v0,
+2R 21,2j(-i7B) - I) Vo
+7-1 B-1 R 21,2j(i7B) ~iR21,2j( -i7B) v'(O)
#1
for the approximate solution of the initial-value problem (6.1). In a similar manner
one can obtain the stability inequalities for the solution of the two-step difference
schemes (6.48).
Now, the abstract Theorem 6.3.1 is applied in the investigation of difference
schemes of higher order of accuracy with respect to one variable for approximate
340
solutions of the mixed boundary-value problem (6.6). The first step of discretization of problem (6.6) is given above. Suppose that the operator
j
(1 + L82n(-Atr2n - L 82m _ 1(-A)m r 2m-l

n=1
m=1
I-[TI
I-[T)
2n
L a2n(-Atr )-1 L a2n_l(-Atr 2n - 1)
n=1
n=1
x(I +
has a bounded inverse. Suppose further that the operator (I - E~=1 1]s(-A h )Sr2S )
has a bounded inverse and
-leu +
L a s (-A h )Sr 2S )2(I - L 1]s( _A h )Sr 2s )-2 2

s=1
s=1
o.
Then in the second step we replace problem (6.34) by the difference schemes
(6.43)
r- 2(uZ+ 1(x) - 2uZ(x)

h
+AA
h (Uk+l (x)
Ah
h
+ Uk-l
(x))
+ ULI (x)) + AhuZ(x)
= fk (x), tk = kT, 1 ~ k ~ N - 1, x E Oh,
=-
=-
Las(-AhYr2s-2, ~
s=1
s
f~(x) = L1]s L(-Ah )S-A(j(2A-2)(tk+l,X)
s=1
+~
L..J a
cr
L1]s(-Ah )Sr 2S - 2,
s=1
+ f(2A-2)(tk_l,x))r 2s - 2
A=1
s=1
s
t
~(_AX)S-Af(2A-2)(t
L..J
k"
x)r 2s - 2 1 <
k<
N - 1,
-
A=l
= (1 + L82n(-Ahtr2n - L
82m_l(-Ah)mr2m-l
n=1
m=1
l-(TI
I-[TI
x(I + L a2n(-A htr 2n )-1 L a2n_l(-A htr 2n - 1)-1
n=1
n=1
I
x{L P2n( _A h )n r 2n-l
+ L 82m - 1 ( _A,;)m r 2m-2
m=1
I-[TI
j-m+[91+ 1
L
x(I+
a2n(-A h )n r 2n)-1
b2n_l(-Ah)nr2n-l},
n=l
n=1
n=l
(6.49)

j
n=l
m=l
+ L 62n(-Ah)n r 2n - L 62m_1(-A h )m r 2m-1
ST = (I
I-[TJ
I-ITl
(I + L a2n( _A h)n r 2n)-1 L a2n-1 (_A h )n r 2n-1 )-1

n=l
n=l
I
l-[TJ
j-m+[T]
X{LP2n_1(-Aht-1r2n-2 + L 62m_1(-A h)m r 2m-2

n=l
m=l
X(I + L a2n(-A htr 2n )-1(I +
n=l
I
L
n=l
b2n (-A h)nr 2n)},
P2n( -A htr 2n - 1 L( -Ah)n->. j(2)'-2) (0, x)

n=l
>'=1
j~(X) = L
I
n=l
>'=1
+ L P2n-1 (_A h )n-1 r 2n-2 L(-Ah)n->. j(2)'-1)(0, x)

n
- L 62n( -Ahtr 2n - 1 L(-Aht->'j(2)'-2)(r,x)

n=l
>'=1
j
n-1
-Ah)n->. j(2)'-1)(r, x)
- L 62n-1( -A htr 2n - 2
n=l
>'=1
2)
- 2:
j
62m-1 (-A
m=l
h )m r 2m-2(I +
I-[TJ
n=l
>'=1
x{ -
I-IT]
2: a2n( _Ah)n r 2n)-1
n=l
2: a2n 2:( -Ah)n->. j(2)'-1)(r,x)r 2n
I-[T)
n=l
>'=1
2: a2n-1 2:( -Ah)n->. j(2)'-2)(r, x)r2n - 1
j-m+[TJ
+ L
n=l
j-m+[T]+l
2:
n=l
b2n L(-Ah)n->. j(2)'-1)(0,x)r 2n

>'=1
n
b2n_12:(-Ah)n->'j(2>.-2)(0,x)r2n-1}.
>'=1
341
342
We have that
Theorem 6.3.4. Let 7 and
Ihl
be sufficiently small numbers. Then the solutions of

the difference schemes (6.49) for l = 0 satisfy the following stability estimates:
+ II (Bh)-I7jJh II _ + I <phil -],

L2(l),h)
max 117-
15,k5,N
(u h k
uh )
k-l
L 2 (l),h)
II L2(l),h)
_ <- M 1 [05,k5,N
max I
-1
n
+ II 7jJh II _ + l: II <ph
L 2(l),h)
max
15,k5,N-l
117-2(u~+1 -2u~+uLl)
Here M does not depend on
7,
r=1
xr,m r
fh
k
I L (l),h)
2
II -],
L2(l),h)
I L2(l),h)
- ~ M 1 [ 15,k5,N-l
max
117-1(f~- f~-I)
II L2(l),h)
-
h, <ph(x), 7jJh(x) and f~(x), 0 ~ k ~ N - 1.
Note that in a similar manner one can construct the difference schemes of a
the boundary-value problem (6.5). Abstract Theorem 6.3.1 permits us to obtain
stability estimates for the solutions of these difference schemes.
Chapter 7
Uniform Difference Schemes
for Perturbation Problems
In the present chapter we consider two abstract Cauchy problems
c:v/(t)
+ Av(t) =
f(t)
(0::; t ::; T), v(O) = Vo,
= f(t) (0::; t ::; T), v(O) = Vo, v/(O) = vb
f2V"(t) + Av(t)
and the boundary-value problem

-f2 V"
(t) + Av(t)
= f(t)
(0::; t ::; T), v(O)
= vo, v(T) = VT
for differential equations in an arbitrary Banach space E with the positive operator
A and an arbitrary f parameter multiplying the derivative term. The high order of
accuracy single-step uniform difference schemes generated by an exact difference
scheme of the approximate solutions for differential equations of parabolic type
with an arbitrary parameter f on the highest derivative are presented. The high
order of accuracy two-step uniform difference schemes generated by an exact difference scheme of the approximate solutions for differential equations of the elliptic
and hyperbolic types with an arbitrary parameter f on the highest derivative are
presented. The well-posedness of these difference schemes for parabolic and elliptic equations in various Banach spaces is established. The stability estimates of
solutions of a high order of accuracy difference schemes for hyperbolic equations
with an arbitrary f parameter on the highest derivative are obtained.
7.1
A Cauchy Problem for Parabolic Equations
We consider the abstract Cauchy problem for the differential equation
fV/(t)
+ Av(t) = f(t) (0::; t ::; T), v(O) = Vo
(7.1)
in an arbitrary Banach space E with the (unbounded) operator A. Here v(t) and
f(t) are the unknown and the given functions, respectively, defined on [0, T] with
Chapter 7. Uniform Difference Schemes for Perturbation Problems
344
values in E. The derivative v'(t) is understood as the limit in the norm of E of

the corresponding ratio of differences. A is a given, closed, linear operator in E.
We will assume that the operator A has dense domain D(A) and that for A ~ 0
the operator A + AI has a bounded inverse whose norm satisfies the inequality
II (A + A1)-l
IIE-+E:::; M(A
+ 0)-1 ( for some 0 > 0),
(7.2)
where I is the identity operator. Obviously (7.2) is fulfilled when the operator -A
is the generator of a strongly continuous semigroup exp( -tA)(t ~ 0), for which
the estimate
(7.3)
is valid. Finally, Va is a given element of E and c E (0,00).
are satisfied:
i. v(t) is continuously differentiable on the segment [0, T]. The derivative at
the endpoints of the segment are understood as the appropriate unilateral
derivatives.
11.
lll.
The element v(t) belongs to D(A) for all t E [0, T] and the function Av(t) is
continuous on [0, T].
v(t) satisfies the equation and the initial condition (7.1).
In Chapter 4 the stable high order of accuracy difference schemes generated

by an exact difference scheme or by Taylor's decomposition on two points for the
numerical solutions of an abstract Cauchy problem (7.1) for c = 1 were presented.
These difference schemes can be applied in the general c E (0,00). However, it can
be shown in general that no classic finite difference approximations yield uniform
convergence with respect to c. This is a very disappointing result and it raises the
question whether there are any difference schemes which are uniform in c? The
main result of this chapter gives an answer to this question. Actually, we study the
high order of accuracy single-step uniform difference schemes of the approximate
solutions for differential equations of parabolic type with an arbitrary parameter
c on the highest derivative.
On the segment [0, T] we consider a uniform grid
[O,T]r = {tk = kT,k = 0, 1, ... ,N,N7 = T}

with step
scheme
7.
Let
= pc. By Theorem 2.1.1 we have the exact single-step difference
(7.4)
k
=7
-1
jt
exp( -(tk - s) c- 1 A) f(s)ds,l :::; k :::; N
7.1. A Cauchy Problem for Parabolic Equations
345
for the solution of the initial-value problem (7.1). Later we will consider the singlestep difference scheme
(7.5)
m-1
fk
=L
Jjf(j) (tk-d, 1 :S k :S N, Uo
= Vo
j=O
1:S j:S m -1, J
= (pA)-1(1- g(pA)), g(pA) = exp(-pA)
for approximate solutions of the initial-value problem (7.1).

This problem is uniquely solvable and the following formula holds:
k
Uk = g(kpA)uo + Lg((k - j)p A) fJp

j=l
(7.6)
Let us reduce the difference schemes (7.5) to an operator problem in the space
FT(E) of vectors vT = {vkH"=l' Recall that the space FT(E) can be equipped with
various norms and thus become a normed space (see Chapter 4). Thus, for instance,
the vector space FT(E) generates the normed space CT(E) = C([O, T]n E) with
the norm
I <pTllcT(E) = l~~N I <Pk liE'

the normed space C~(E) = C,6([O, T]n E) 0 < {3 < 1, with the norm
I <pTllc~(E)
I <pTllcT(E) + l:9~l~:SN I <Pk+r -
<Pk
liE
(r~),6'
the weighted normed space C~,'Y(E) = C,6,'Y([O, T]n E), 0 < , :S {3, 0 < {3 < 1,
with the norm
II <P Ilc~"(E) = II <P IlcT(E) + l:Sk~~:SN II <Pk+r -
<Pk
liE
((k+r)r)'Y
(rr),6 ,
and the normed space Lp,T(E) = Lp([O, T]n E), 1 :S p < 00, with the norm
N
I <pTIILp.T(E)
(2: I <Pkll~r)t.
k=l
First we will define the operator D n acting from the space E x FT(E) of vectors
w T = {Wk}t'=o into the space FT(E) of vectors vT = {Vk}t'=l by the rule
346
Second we will define an operator AT from the space E x FT(E) of vectors w T =

{wdf=o into the space FT(E) of vectors v T = {vk}f=l by the rule
v T = ATwT,Vk = p-1(1 - g(pA)) Wk-1, k = 1, ... , N.
Next, let us introduce the continuation operator TI( uo), which acts from Ex FT (E)
to FT (E) according to the rule
TI(UO)(U1,"" UN) = (uo, U1,"" UN).
operator equation
DTTI(uO)U T + ATTI(uO)UT = r
Here

r = (h, ,fN).
solvability for any Uo E E and r E FT(E) it follows that its solution u T defines a
continuous additive and homogeneous operator uT(r, uo).
The initial-value difference problem (7.5) is said to be stable in FT(E) if we
I uT(r,uo)
IIFr(E)::;
M[li r
r, Uo,
IIFr(E)
+ I uollEJ,
but also of c.
Theorem 1.1.1. For the solution of the difference problem (7.5) the following estimate is valid:
max II Uk
19::; N
liE::; M [ II Uo II E + 1 - expP( - J)
max 11!k II E]
p l::;k::; N
'
r, Uo,
7,
(7.7)
but also ofe.
Proof. Using formula (7.6) and estimate (7.3), we obtain
"e(kj=l
k
II Uk liE::; M [ II Uo II E + L.J
j p
)p
max
l<j<N
- -
II
fj
liE]
for all k, k = 1, ... ,N. Using the last inequality and

k
L ej=l
L e00
6 (k-j)p ::;
6ip
= 1/(1 - e- 6p ),
(7.8)
i=O
we obtain
p
II Uk liE::; M [ II Uo II E + 1_ e- 6p
1r"f!~N
_J_
II
fj
liE]
for all k, k = 1, ... , N. This completes the proof of Theorem 7.1.1.
347

Note that if c = 1, then
---'----::=
15
1 - e- 15p
1 - e- T
<
M.
-
Therefore, (7.7) is the generalized stability inequality of (7.5) in CT(E).

We say that the difference problem (7.5) has an m-th order of accuracy in
FT(E) on the solutions v(t) of the Cauchy problem (7.1) if the error vector
where M is independent not only 7, but also of c.

Generalized stability inequality (7.7) permits us to obtain the convergence of
(7.5) in CT(E).
Theorem 7.1.2. Suppose that the function f(t) has m derivatives and
II f
(m)(t)
liE ~ M,O ~ t
~ T.
(7.9)
Then for the solution of the difference problem (7.5) the following convergence
estimate is valid:
(7.10)
I {v(t k ) - ud~ Ilc,.(E) ~ M7 m ,
Proof. The components Zk
relations
and c.
= V(tk ) - Uk
of error vector ZT
= {zk}f
satisfy the
where
Rk = p-1(V(tk) - g(pA)V(tk-d) - fk.

From this and the generalized stability inequality (7.7) it follows that
max
19::; N
II
II E <- M
Zk
P
1 - exp( -8p)
Therefore to this end it suffices to estimate

we have that
{Rdf"
max
19:5
IIRk
II E.
in CT(E). By (7.5) and (7.11),
tk
Rk =
7-
exp( -(tk - s)c- 1A)
tk-l
(7.12)
f (s) ds - !k .
348
Now, using Taylor's formula, we obtain
J
tk
+7-1
J
8
exp(-(tk - s)c-1A)
tk-l
(s - z)mf(m) (z) dzds
~!
- fk .
tk-l
A simple calculation shows that
Therefore, we can write
tk
Rk =
7-1
exp( -(tk -
S)c-l A)
tk-l
(s z)m-l
f(m) (z) dzds .
(m - 1)!
tk-l
Then, using estimates (7.3) and (7.9), we obtain
From this and estimate (7.11) follows estimate (7.10). Theorem 7.1.2 is proved.
Note that for the convergence estimate I {v(t k ) - Uk)}i" IlpT(E) :S M7 m it
suffices to obtain the generalized stability inequality in FT(E).
From formula (7.6) one derives the following result.
Theorem 7.1.3. For the solution of the difference problem (7.5) the following esti-
mate is valid:
I {udf IILp,T(E) :S M [ II
Uo
II
(7.13)
+ 1 _ e~( -8p) I Udf IILp,T(E)], 1 :S p < 00,

r,
Uo, 7,
but also of .
Proof. Using formula (7.6) and estimate (7.3), we obtain

N
II Uk II E:S M [ II Uo II E + I>-O(8-1)Pp II
8=1
fk-8+1
liE]
349
for all k, k = 1, ... , N. Here f: = fs if 1 ~ s ~ k and J: = 0 if otherwise. From

the Minkowski sum inequality (with respect to k) it follows that
N
k=l
s=l
k=l
(LII Uk II ~T)~ ~ MT II Uo II E+MLe- S(s-l)Pp(LII fk-s+lll ~T)~.

By the definition of the grid function fk-s+l'
N
L Ilfk-s+1 II~T ~ L II !k II ~T.

k=l
k=l
Using the last estimate and inequality (7.8), we obtain estimate (7.13). Theorem
7.1.3 is proved.
Remark 7.1.4. Theorems 7.1.1 and 7.1.3 permit us to obtain error estimates of
the high order of accuracy difference schemes over time and space variables for
parabolic equations and the first-order hyperbolic system of equations with a small
parameter multiplying the time-derivative term in the various Banach norms.
Now, assume that the operator -A is the generator of an analytic semigroup
exp( -tA)(t :2 0), for which the estimates
II Aexp(-tA) IIE--.E~ Me- St ,
til
Aexp(-tA) IIE--.E~ M
(7.14)
are valid. From this it follows that

Ilexp( -kpA) - exp {-(k
+ r)pA}IIE--.E
for all k, r, 1 :S k < k + r :S N,O :S

on k, r, T,Q and e.
~ M (k
:S 1 and all
rQ
+ r)Q exp(-kpo),
(7.15)
> 0, where M does not depend
Theorem 7.1.5. For the solution of the difference problem (7.5) the following esti-
mate is valid:
II {uk}f
+ 1 -exp-p
P( 0)
Ilc;"a(E) ~ M [ II Uo II E
II
{!k}f
Ilc",a(E)J, O:S Q ~ 1,
r, uo,
T, Q
Proof. By estimate (7.7), we have that
To this end it suffices to establish the estimate
but also of e.
(7.16)
350
for all k, r,l S k < k + r S N and all
> O. By (7.6), we have that
where
Wk = g(kpA)uo,
k
Vk = Lg((k - j)pA)fJp
j=l
Applying estimate (7.15), we obtain
I Wk -
wk+rll E =
I g(kpA)uo -
g((k + r)pA)uo
rO:
S M(k+r)o: II
Uo
II
liE
(7.18)
for all k, r,l S k < k + r S N and all T > O. Now let us estimate the difference
Vk - vk+r for all k, r,l S k < k + r S N and all T > O. We shall consider two
separate cases: k S 2r and k > 2r. If k S 2r, then from the estimate
I Vk liE s M 1 -exp-p
P( 8) I
for all k, 1 S k S N and all
S 31+0: M (k
Uk};""
Ilca,a(E)
T
> 0 and the triangle inequality it follows that
:O:r)o: 1 _ ex';(-8p) I Uk};"" Ilc~,a(E)'
Hence, for k S 2r we have the estimate
Now let k > 2r. We have that
k-1
Vk = (I - g(pA))-1[1 - g(kpA)]p!k
+ Lg((k - j)pA)(fJ - !k)p, k 2: 2.

j=l
Identity (7.20) yields
Vk+r - Vk = (I - g(pA))-1[1 - g((k + r)pA)]pUk+r - !k)
+(1 - g(pA))-l[g(kpA) - g((k + r)pA)]]p/k
(7.20)
k+r-l
L
j=k-r+l
g((k + r - j)pA)(fj - fk+r)P -
351
k-l
g((k - j)pA)(fj - fk)p

L
j=k-r+l
k-r
+ Lg((k + r - j)pA)(ik - fk+r)P
j=l
k-r
+ L[g((k + r
j)pA) - g((k - j)pA)](fJ - fk)p
j=l
= Pl
+ P2 + P3 + P4 + P5 + P6 .
Let us estimate Pl , P2 , P3 , P4 , P5 and P6 separately. We begin with Pl. Using

k+r-l
~ L
IIP11I E
Ilg((k + r
j)pA)IIE-+Epllfk+r - fkllE
j=l
f;
k+r-l
~M
c5
e- (k+r-j)p p(k: r)o
rO
{fdf'
Ilc;,a(E)
~ M (k + r)o 1 _ exp( -6p) I {fdf' Ilc;,a(E)

Next, let us estimate P2 . Using estimates (7.3) and (7.15), we obtain
11P211E ~ 11(1 -
g(pA))-lIIE-+EPllg(kpA) - g((k + r)pA)IIE-+EPllikIIE
~ M (k :or)O 1 _ e;( -6p)

~
II Udf' IICr(E)
rO
p
M (k + r)o 1 _ exp(-6p) II {ik}f'
Ilc;,a(E)
Now, let us estimate P3 . Using estimates (7.3) and (7.14), we obtain
11P311E ~
~M
k+r-l
L
Ilg((k + r - j)pA)IIE-+EPllfj - fk+rllE
j=k-r+l
k+r-l
(k
')0
'" e- c5 (k+r-j)p p + r - J II {f }N I <>,a
~
(k + r)o
k 1
c r (E)
j=k-r+l
352
~M ~
~
j=k-r+l
~ M2
e- 8(k+r-j)p p (2r - l)Q I {f }N

(k+r)Q
k 1
1 _ ex;( -8p) (k :Qr)Q I {fk}f
I a,a
C.,.
(E)
Ilc;"a(E)
rQ
~ MIl _ exp( -8p) (k + r)Q I {fdf Ilc;"a(E)'

Finally, let us estimate P6 . Using estimate (7.15), we obtain
k-r
P6 IIE~
L pllg((k + r j=1
k-r
'"'
~ ~ (k + r _ j)Q e
j)pA) - g((k - j)pA)IIE-+Ellfj - fk+r liE
-8(k-j)p
J=1
M 1 (k
(k
')Q
+ r - J I { }N
p (k + r)Q
Ik 1
C;"a(E)
+ r)Q 1 _ exp( -8p) II {Ik}f II c ;"a(E)'
Combining the estimates for PI, P2 , P3 , P4 , P5 and P6 , we obtain (7.19) for

k > 2r. Estimate (7.17) follows from the triangle inequality and estimates (7.18)
and (7.19). Theorem 7.1.5 is proved.
The initial-value difference problem (7.5) is said to be well posed (coercively
stable) in F.,. (E) if we have the coercive stability inequality
I {p-1(Uk -
Uk-l)}f II F.,.(E)+
I {p-l(l -
g(pA))Uk-l}{" IIF.,.(E)
r IIF.,.(E) + II Auoll E],

where M is independent not only of r, un, but also of c.
~ M[lj
Since the initial-value differential problem (7.1) in C(E) is not well posed for
the general strongly positive operator A and space E, then the well-posedness of
the initial-value difference problem in C.,.(E) norm does not take place uniformly
with respect to 7 > 0 and c. This means that the coercive norm
tends to 00 as 7 -+ +0 . The investigation of the difference problem (7.5) permits

us to establish the order of growth of this norm to 00. First, let us prove a lemma
that will be needed in the sequel.
353
Lemma 7.1.6. The following estimate holds:

N
L 11(1 - e-pA)e-U-1)pAIIE......E ::; M 1 In(l/r),
(7.21)
j=1
where M 1 does not depend on rand E.

Proof. Using estimate (7.15), we obtain
L 11(1 - e-pA)e-U-1)pAIIE...... E ::; M L -:- ::; M Jd

N
. 1
J=
. 1
J=
SS
= M InN::; M 1 In(1/r).

Theorem 7.1.7. The solutions of the difference problem (7.5) in GT(E) obey the
r, Uo, but also of rand
E.
J
1
(I - e- pA )(pA)-1 =
e-psAds
o
and estimate (7.14), we obtain
(7.23)
Using formula (7.6) and estimates (7.14) and (7.23), we obtain
IIp-1(1 -
g(pA))Uk-1
liE::; M [ II
1
p-1(1_ g(pA))g((k - 1)pA)A- 1IE...... E
k-1
IIAuo II E + L II
p-1(1 - g(pA))g((k -
j=1
::; M [ IIAuD
II E + L II
j=1
1-
j)pA)IIE...... EllfjIIE]
p-1(I - e-pA)e -U-1)PAIIE.. . . EII
r IlcT(E) ]
354
for all k, k = 1, ... , N. Using the last inequality, we obtain
{p-1(1 - g(pA))Uk-df I c ,.cE)
::;
MIlIl Auoll E + In ~II r II c ,.cE)]
The estimate for II{p-1(uk - Uk-1)}f I c ,.cE) follows from the triangle inequality
and the last estimate. Theorem 7.1.7 is proved.
Banach spaces.
C~,O(E)
obey the
II{p-1(uk - Uk-1)}f I c ;-'''CE)+
::; M 1[a(1
{p-1(1- g(pA))Uk-1}{" I c ;-'''CE)
~ a) I r
Ilc;-'''CE)
+I
(7.24)
Auoll E]'
and E.

p-1(1 - g(pA))Uk-1 = p-1(1 - g(pA))g((k - 1)pA)uo+[1 - g((k - l)pA)]fk
k-1
3
+ Lg((k -1- j)pA)(fJ - ik) = L If,
j=l
m=l
where
Jl
p-1(1 - g(pA))g((k - l)pA)uo,
Jf = [1 - g((k - l)pA)]fk,
k-1
J2 = Lg((k -1- j)pA)(fJ - ik)
j=l
= {Jk}"=l for all m, m = 1,2,3 separately. We start with
Let us estimate J:,.

Using estimates (7.14) and (7.23) we show that, for all k, 1 ::; k ::; N,
Jr
IIJ~IIE :::;
11(1 -
e-pA)(pA)-111E->Elle-Ck-1)pAIIE->EIIAuoIIE :::; MIIAuoIIE.

1
IIJ1 - Jl+rIIE ::;
(2r)0
IIJ1 11E + IIJl+rIIE ::; 2MII AuoilE ::; 2M (1 + r)o IIAuollE,
for all r, 1 < r + 1 :::; N.

Further, applying (7.15) and (7.23) we show that, for all 2 :::; k < k + r :::; N,
IlJl- Jl+rllE :::;
11(1 -
e-pA)(pA)-l11E-+E
xlle- Ck - 1)pA _ e-(k+ r -1)p AIIE->EII AuoIIE

rO
rO
:::; M (k + r _ 1)0 II Avo liE :::; M 1(k + r)o II AvoilE.
355

(7.25)
Now let us estimate
J2-
Using estimate (7.14), we obtain that
for all k, 1 ::; k ::; N. Therefore,

max
15.k5.N
II it liE::; M111 r Ilco.O(E)

T
IIJr- Jr+rlIE ::; IIJrllE + IIJr+rIIE ::; 2M1 1

Q
::; 21+ M 1 (1 :Qr)Q II
for all r, 1 < r
+ 1 ::; N.
Ilc;-'O(E)
r Ilc;-,O(E)
Since
it follows that
J~+r - J~
liE::; [1 + Ile-(k+r-l)pAIIE->Elllfk+r - ikilE
+lle-(k-l)pA - e-(k+r-l)p AIIE->EIIfkIIE
for all k, 2 ::; k < k
+r
::; N. Using estimates (7.14) and (7.15), we obtain
for all k, 2 ::; k < k
+r
::; N. From this it follows that
(7.26)
Now let us estimate J;. By the definition Jr = O. Therefore, let us consider the
cases 2 ::; k ::; N. Using estimate (7.15), we obtain
II
k-l
J~
liE::; L 11(1 -
e-pA)e-(k-i-l)pAIIE->EIIJi -
ik liE
i=l
: ; ME (k-i~l_"k<>11 r Ilc;"O(E)::; ~[~

k-l
k-l
(tk
-:iP-JII
r IIc:;"O(E)'
356
tk
J
o
ds
t'k
(tk - s)l-o - -;-.
Thus, for all k, 2 :S k :S N we have established the bound
Therefore,
(7.27)
J2 IIE:S Ma 111 r Ilco,o(E)'
Now let us estimate the difference J2+r - J2 for all k, 2 :S k < k + r :S N. We will
max II
l-;S.k-;S.N
consider separately the cases k :S 2r and k > 2r. If k :S 2r, then (7.27) yields
II J2+r -
J2 IIE:s11 J2+r
liE + II
J2 IIE:S 2~111 r
Ilc;-,o(E)
M 21+ 0
1
Ilrllc::"o(E{O(k+r)-o.
a
Now let k> 2r. Let us represent the difference J2+r as the following sums:
:S
J2
k+r-1
J2+r -
J2 = L
j=k-r+1
(I - g(pAg((k
+r -
1 - j)pA)(Jj - fk+r)
k-1
(I - g(pAg((k - 1 - j)pA)(Jj - fk)

j=k-r+1
k-r
+
g(pAg((k + r - 1 - j)pA)(Jk - h+r)
j=l
2:(1 -
k-r
+ 2:(1 - g(pA[g((k + r - j)pA) - g((k - j)p A)](fJ - h) =

j=l
2: PI:.
m=l
Let us estimate P~, P'f,

(7.14), we obtain
p1 and P: separately. We begin with Pl. Using estimate
k+r-1
IJP11IE:S
2:
j=k-r
11(1 - g(pAg((k + r - 1 - j)pA)IIE--+ElliJ - fk+rllE
k+r-2 (k +
')0-1
'"'"
r-2
N
::; M[(k + r)o + . L
(k + r)o
]11 {fkh Ilc~,a(E)
J=k-r
r
357

rQ
:S M (k + r)Q II {fk}f IIc~,a(E)

1
k+r-2
1
+M (k + r)Q [.
(k + r _ i)l-Q
J=k-r
]11 {ik}f Ilc~,a(E)'
Since
it follows that
~2rQ(k + r)-QII r Ilc~Q(E)
pIIIE:S
Next, let us estimate
k-r
Pl. Since
L(I - g(pA))gk + r -
1 - j)pA) = g2r - l)pA) - gk
+ r - 2)pA),
j=l
we have that
p2 = [g2r - l)pA) - gk + r - 2)pA)](fk - fk+r).

IIP21IE:S
Ilg2r - l)pA) - gk + r - 2)pA)IIE--+Ellfk - fk+rllE

rQ
:S M (k + r)Q I
Now, let us estimate
k-r
P:
{ik}f IIc~,a(E)'
P:' Using estimate (7.15), we obtain
IIE:S L 11(1 - g(pA))[gk+r-1- j)pA) j=l
gk-1- j)pA)lIIE--+E IIIi - fk
liE
358
Since
it follows that
II
P: IIE~ 1 ~a ~: II r
Ilc;"<>(E)'
Thus, we have shown that for all k, 2 ~ k < k + r ~ N the following inequality
holds:
Q
4
4
M1
r
T
II Jk +r - Jk IIE~ 1 _ a (k + r)Q II f Ilc;""(E)'
(7.28)
II JJ Ilc;""(E)
~ a(~ a) II r Ilc;""(E) .
(7.29)
Finally, applying estimates (7.25), (7.26) and (7.29), we obtain the estimate
By the triangle inequality, this last estimate and difference equation (7.5) yield
359
Theorem 7.1.9. The solutions of the difference problem (7.5) in Cr(Ee.) obey the
II{p-l(uk - Uk-l)}f
IlcT(Ea)+ II {p-l(l -
IlcT(Ea)
g(pA))Uk-df
r IICT(Ea) + I AuaIIEJ,
is independent not only of r, Ua, a, but also of and
(7.30)
::; M 1 [a(1 ~ a) I
where M 1
p-l(1- g(pA)) Uk-l
p-l(l - g(pA)) g((k - l)p A)ua
k-l
+ 2:(1 - g(pA))g((k - j)pA)fj =
Jf + J~,
j=1
where
Jf =
p-l(l - g(pA)) g((k - l)pA)ua,
k-l
J~ = 2:(1- g(pA))g((k - 1- j)pA)fJ.

j=1
To complete the proof of Theorem 7.1.9 it suffices to establish that
II Jf IlEa::; M II
II
J~
IlEa::; M
Aua
IlEa'
a- 1 (1 - a)-III
(7.31)
IlcT(Ea)
(7.32)
for all k, k = 1, ... , N. Using the formula
J
1
Jf =
g((k - 1 + s)pA) ds Aua,
and estimate (7.24), we obtain
J
1
II Jf IlEa::; M
e- 6(k-l+s)Pds
II
Aua
lie. ::; M II
Aua
Estimate (7.31) is established. We have that
J~ =
k-1
2:
j=l
J
1
Ag((k - 1 - j + s)pA)ds
!J
p.
IlEa'
360
To estimate J~ in the norm Eo: we use the Cauchy-Riesz representation formula

for the operator (.\ + A)-1
We obtain
Jr
J~ =
A(.\ + A)-1
L j
k-1
A(.\ + A)-1
J=1
~
27fZ
Ag((k - 1 - j + s)pA)ds
k-l
+ s)p z) A (z - A)-1 !J dz.
z(.\ + z)-I" pg((k - 1 - j

L.J
SlUS2
!J
j=1
Since z = re'P, with
II
l<pl < ~, we have the estimate
A (z - A)-1
!J
liE::; M
II
A(r + A)-1 fj liE,
(7.33)
on the sides 8 1 and 8 2 . Then, using the inequality
I (.\ + z)-1 I ::; M(.\ + r)-1,

and estimates (7.24), (7.32), we obtain
00
II
A( A\+A)-1
< ~II
- cos<p
J211
< Mj
k E o
r I CT(E
J
00
n )
k-1
rp
"je-(k-j+S)prCOS'Pdsdr
(.\ + r)ro: L.J
J=1
II fJ'IIE n
k-l
1
"(1- e- rp COS'P)e-(k-j)p rcos'Pdr
(.\ + r)ro: L
J=1
for all .\ > O. Therefore,
c:<pll r
00
IICT(En
(1 ::)so: ::; M 1 a- (1- a)-III
IICT(En
for all >. > O. Estimate (7.32) is established. Theorem 7.1.9 is proved.
Note that estimate (7.24), established for the difference problem (7.5), allows
us to prove Theorem 7.1.8. Indeed, by the definition of the norm in C::,O:(E),
361
Then, Theorem 7.1.8 implies that
II{p-l(uk - Uk-l)}~ IlcT(E)+ II {p-l(l - g(pA))Uk-d~ IlcT(E)
M[a~a II r II cT (E) + II AuolI E]
:S
for all sufficiently small a > O. Choosing a in the best way, we obtain estimate
(7.22). Furthermore, if A is the bounded positive operator in E, then from estimate
(7.30) it follows that
II{p-l(uk - Uk-l)}~ IlcT(E)+ II {p-l(1- g(pA))uk-d~ IlcT(E)
:S M[(l + lIn IIAIIE-->EDII
II cT (E) + II
(7.34)
Auollel
Indeed, by the definition of the norm in Cr(Ea ),
II{p-l(uk - uk-d}~ IlcT(E)+ II {p-l(l - g(pA))Uk-d~ II cT (E)
:S
M[IIAI~-->E II r
IlcT(E) + II AuolIE]
for all sufficiently small a > O. Choosing a in the best way, we obtain estimate (7.34).
By Theorem 7.1.7, estimate (7.34) yields the following result.
Theorem 7.1.10. The solutions of the difference problem (7.5) in Cr(E) obey the
r, uo, but also of
and c.

Theorem 7.1.11. The solutions of the difference problem (7.5) in C~"(E) (0 :S
'Y :S (3,0 < (3 < 1) satisfy the following coercivity inequalities:
II {p-l(Uk -
Uk-l)}~ Il cT (Ei3_-,) :S (3(~ (3) II r II cf'Y(E)
+M111 (I - e-pA)p-1uo - II II Ei3--, ,

II {p-l(Uk - uk-d}~ Ilcf"(E) + II {p-l(l - g(pA))Uk-d~ II c f"(E)
:S
(3(~ (3) II r
1I16",
Ilcf"(E) + M11(I - e-pA)p-1uo -
r, uo, (3, 'Y but also of
and c.
362
Here, the Banach space
Eg"
consists of those vEE for which the norm
is finite.
Note that the parameter 1 can be chosen freely in [0, (3), which increases the
number of spaces C~" (E) (0 ~ 1 ~ {3, < {3 < 1) of grid functions in which
Theorem 7.1.12. The solutions of the difference problem (7.5) in C~"(EQ-{3)

(0 ~ 1 ~ {3 ~ a, < a < 1) satisfy the following coercivity inequalities:
II {p-l(Uk - uk-d}f Ilc~'-Y(Ea-IJ)+ II {p-l(Uk - Uk-l)}f IICr(E~:?IJ)

+ II {p-l(l - g(pA))Uk-df Ilc~'-Y(Ea-lJ)
~ a(~ a) II r Ilc~'-Y(Ea-lJ) + M11(I -
e-pA)p-1uo -
hl~~{3'
II {p-l(Uk - uk-d}f IIc~'-Y(Ea-IJ)+ II {p-l(Uk - Uk-l)}f IICr(Ea_-y)

+ II {p-l(1- g(pA))Uk-df IIc~'-Y(Ea-lJ)
M
~ a(l ~ a) II
Ilc~'-Y(Ea-lJ) + M111(I - e-pA)p-1uo - hIIEa_-y'
r, uo, a, {3, 'Y but also of
and c.
Here, the Banach space E~:!{3 consists of those vEE for which the norm
Iwl~'2'{3 = l~~N Ile-~AwIIEa_1J

(~)A!n.A
+ sup f; {3 (t + tr)'ll(e-'
- e-' )wIIEa_1J
l~n<n+r~N
is finite.
Note that the spaces C~"(EQ-{3) of grid functions, in which coercive solvability has been established, depend on the parameters a, (3 and 1. However, the
parameters {3 and 1 freely, which increases the number of spaces of grid functions
Lp,.,.(E) = Lp([O, T].,., E), 1 ~ p < 00 of all grid functions. We have not been able
II {p-l(Uk-uk-I)}i"
~
IILp,r(E)
M[II r
+ II {p-l(l_ g(pA))Uk-di"
IILp,r(E)
II Lp ,r(E)
+ II AuolI E ]'
in an arbitrary Banach space E and for the general strong operator A. Nevertheless, we can establish the almost coercivity inequality.
363
Theorem 7.1.13. The solutions of the difference problem (7.5) in Lp,r(E) obey the

II {p-i(U k -Uk-i)}f IILp,T(E)
+ II {p-i(l -
g(pA))Uk-df IILp,T(E)
:S Mi[ll Auoil E + min {ln~, 1 + Iln II A IIE-+EI} II

where M i does not depend on
r, uo, p, T and c,
r IILp,T(E)],
Note that if A is the unbounded positive operator in E, then

min
{ln~, 1 + Iln II A IIE-+EI} = ln~.
of the initial-value difference problem (7,5) in the spaces Lp,r(E) = Lp([O, T]r, E),
1 :S p
< 00,
Theorem 7.1.14. Suppose that the difference problem (7.5) well posed in Lpo,r(E)
for some Po, 1 < Po < 00. Then it is well posed in Lp,r(E) for all p, 1 < P < 00

II {p-i(Uk-Uk-d}f IILp,T(E)
+ II
{p-i(l_ g(pA))Uk-df IILp,T(E)

2
+lIurIlCc(El_l/P,p) :S M(po)[11 TUoIIE1_I/p p + pP 1 II

,
r, uo, p and
II
Lp,T(E)
],
T and c.
00 and 0 < a < 1. Then problem (7.5) is well posed

in Lp,r(Ea,p) and the following coercivity inequality holds:
Theorem 7.1.15. Let 1 :::: p ::::
+ II
II {p-i(Uk-Uk-i)}f IILp,T(E",p)
:::: M[II AuollE",p
{p-i(l_ g(pA))Uk-di" IILp,T(E",p)

1
+ a(l _
a) II
II
Lp,T(E",p)
],
where M does not depend on fT, Uo, p, a, T, c.

and 0 < a < 1. Then the difference problem
(7,5) is well posed in Lp,T(Ea,q) and the following coercivity inequality holds:
< p, q <
00
+ II
II {p-i(Uk-Uk-d}f IILp,T(E",q)
{p-i(l_ g(pA))Uk-df IILp,T(Ea,q)

2
:S M(q) [II AuoII Ea q + ~ II

,
p-
II
Lp,T(Ea,q)
where M(q) does not depend on fT, un, p, a ,Tand c.
],
364
Remark 7.1.17. Theorems 7.1.5-7.1.16 permit us to obtain the error estimates of

the high order of accuracy difference schemes over time and space variables for
parabolic equations with a small parameter multiplying the time-derivative term
in the various Banach norms.
Remark 7.1.18. For practical computations, we have been approximating the elliptic operator A by some difference operator A h such that exp( -pA h ) is computable.
Thus, we shall construct difference schemes with fitting operator exp( -pA h ).
Therefore for numerical solution of (7.1) it is necessary to compute by (7.5)
exp( -pA h ) with a high order of accuracy. Accordingly, the methods we have proposed are not easy to implement exactly as for a system of ordinary differential
equations. Indeed, method (7.5) is useful, because the part of compute fitting operator exp( -pA h ) with a high order of accuracy we can make independent of (7.5).
For example, we can approximate exp( -pA h ) by the formula
Rj,I(PAh), if E :S a,
{ Rfl(wA h ), if E > a,
where L =
7.2
[H Lw = p, j + l = m (for Rj,l(pA h ), see Chapter 4).
A Boundary-Value Problem for Elliptic Equations
We consider the boundary-value problem for differential equation

2
-E V" (t)
+ Av(t) = f(t)
(0:S t :S T), v(o) = vo, v(T) = VT
(7.35)
in an arbitrary Banach space E with the positive operator A and E E (0,00).

A function v(t) is called a solution of the problem (7.35) if the following
conditions are satisfied:
1.
v(t) is a twice continuously differentiable on the interval [0, T]. The derivatives at the endpoints of the segment are understood as the appropriate
unilateral derivatives.
ii. The element v(t) belongs to D(A) for all t E [0, T] and the function Av(t) is
iii. v(t) satisfies the equation and boundary conditions (7.35).
In Chapter 5, stable high order of accuracy difference schemes generated by

an exact difference scheme or by Taylor's decomposition on three points for the
numerical solutions of an abstract boundary-value problem (7.35) for E = 1 were
presented. Unfortunately, these difference schemes cannot be applied for the approximate solutions of (7.35) in the general cases E E (0,00). In this section we
study the high order of accuracy two-step uniform difference schemes of the approximate solutions for differential equations of the elliptic type with an arbitrary
parameter E on the highest derivative.
365
7.2. A Boundary-Value Problem for Elliptic Equations
On the segment [0, T] we consider again a uniform grid
[0, T]r = {tk = kT, k = 0, 1, ... ,N, N r = T}

with step r. Let r = pc and B 2 = A. By Theorem 3.1.2 we have the exact two-step
difference scheme
(7.36)
+p-2[(I - exp( -pB))(V(tk+l) + v(tk-d)
+ (exp( -2pB) - I)V(tk)]

-
1-
1,
1
- + !z,k+l)
- (2Bp)- exp( -pB)(h,k+l + !z,k),
fk = (2Bp)- (h,k
- 11
- 11
tk
h,k = -
tk-l
tk
!z,k = r
tk-l
1
exp( --(tk
- z)B))f(z)dz,
c
1 - tk-dB)f(z)dz,
exp( --(z
c
1 ::; k ::; N - 1, v(o) = va, v(T) = VT

for the solution of the boundary-value problem (7.35). Now, we consider the following two-step difference scheme:
_p-2(Uk+l - 2Uk-l
+ uk-d + p-2(1- g(pB)) (Uk+l + Uk-I)
(7.37)
+p-2(g(2pB) - l)uk = fk,

fk = h,k
m-l
h,k = L
+ !z,k+l - g(pB)(h,k+l + !z,k),

m-l
f (j)(tk-d, !z, k = L
j=O
Jz, j f (j)(tk-l),
j=O
= (-l)j(Bjc)-j ho
+ L( -1) j-i(Bj)-(j-i+ 1 ) r i- 1 ji!,

i=1
j
Jz, j
= (Bjc)-j Jz,o
- L(Bjc)-(j-i+ 1 )g(pB) r i- 1 ji!,

i=1
J1 ,0 = J2,0 = (pB)-I(l - g(pB)), 1 ::; j ::; m
- 1, tk = kr, 1 ::; k ::; N - 1,
g(pB) = exp( -pB), Uo = VA, UN = VI

for the approximate solutions of the boundary-value problem (7.35).
Uk = (I - g(2NpB))-I{(g(kpB) - g((2N - k)pB))uo
(7.38)
366
+(g((N - k)pB) - g((N + k)pB)UN - (g((N - k)pB) - g((N + k)pB)

N-l
Xp2
L (g((N - i)pB) - g((N + i)pB)(I - g(2pB))-1 Ii}

i=l
N-l
+p2
L [g(lk - i)pB) - g((k + i)pB)](I - g(2pB))-1 ii, 1 ::; k ::; N -
1.
i=l
Let us reduce the difference schemes (7.37) to an operator problem in the space
FT(E) of vectors v T = {vdf,;/. Recall that the space FT(E) can be equipped with
various norms and thus become a normed space (see Chapter 5). Thus, for instance,
the vector space FT(E) generates the normed space CT(E) = C([O, T]T' E) with
the norm
II c.pTllcT(E)
l:s~a;-lll c.pk liE'
the normed space C~(E) = CI3([O, T]T, E) 0<(3 < 1, with the norm
II c.pTllc~(E)
I c.pTllcT(E) + l:Sk<!f+~:SN-l I c.pk+r -
c.pk liE
(r~)I3'
the weighted normed space C~,'Y(E) = CI3,'Y([O,T]TlE),O < 'Y ::; (3,
with the norm
max
+ l:Sk<k+r:SN-l
II c.pTllc~'l(E) = II c.pTIICT(E)
I
I ((k + r)T)'Y((N 'Pk+r - 'Pk E
(rT)f3
and the normed space Lp,T(E) = Lp([O, T]n E), 1 ::; p <
00,
k)T)'Y
< (3 < 1,
'
with the norm
N-l
I c.pTIILp.T(E) = (2: I 'PklliT)*.

k=l
D;
First we shall define an operator

acting from the space E x FT(E) x E of
T
vectors w = {wd~=o into the space FT(E) of vectors v T = {vdf=-/ according to
the rule
vT
D2TW,
T
Vk
= 21 (Wk+l
- 2Wk
+ Wk-l ) , k =
1, ... , N -1.
Second we shall define an operator AT acting from the space E x FT(E) x E of

vectors wT = {wd~=o into the space FT(E) of vectors v T = {vdf~l according to
the rule
vT = ATwT,vk = ~[(I - g(pB))(Wk-l +wk+d
p
+(g(2pB) - I)wk], k = 1, ... , N - 1.
367
Next, let us introduce the continuation operator l1(uo, UN), which acts from E x
FT(E) x E to FT(E) according to the rule
Then, dearly, the difference problem (7.37) is equivalent to the operator problem
Here

r
= (fI,, fN-l).
E FT(E) it follows that its solution u T
solvability for any Uo, UN E E and
defines a continuous additive and homogeneous operator UT (r , Uo, UN).
We will study stability and coercive stability of the boundary-value problem
(7.37) in the difference analogues of Banach spaces. Since
!k
= !I,k
+ h,k+l
- g(pB)(!I,k+l
+ h,k), 1 ~
(7.39)
k ~ N - 1,
we have that if
E FT(E), then f1 = (!I,I, ... ,!I,N-l,!I,N) E FT(E) x E and
= (h,I, ... ,h,N-l,h,N) E FT(E) x E. Note that FT(E) is the vector space
whose elements are ordered (N -l)-tuples of elements of E. The space FT(E) can
be equipped with various norms and thus become a normed space (see Chapter 5).
Nevertheless, FT(E) x E is the vector space whose elements are ordered N-tuples
of elements of E. In a similar way one obtains that the space FT(E) x E can be
equipped with various norms and thus become a normed space. We will use the
same notation for both cases. For instance, the vector space FT(E) generates the
normed space CT(E) = C([O, T]T' E) with the norm
12
and the vector space FT(E) x E generates the normed space CT(E) = C([O, T]n E)
with the norm
II <pTllcT(E) = 1~~N"
<Pk
liE'
Theorem 7.2.1. For the solution of the difference problem (7.37) the following
estimate is valid:
max
19:5: N-l
II
Uk
+p2 [
II
~ 1
M
(c5 ){
- exp - p
max II B fI k
19:5: N
'
II +
II
Uo
II E + II UN II
max II B h k
l:5:k:5: N
'
where M is independent not only of fI, f2, UO, UN,
T,
II ]},
but also of c.
(7.40)
368

(7.41 )
where
Wk
(I - g(2NpB))-l{(g(kpB) - g((2N - k)pB))uo
+(g((N - k)pB) - g((N + k)pB)UN},

gk = -(I - g(2NpB))-l(g((N - k)pB) - g((N + k)pB)
N-1
Xp2
L (g((N - i)pB) - g((N + i)pB)(I - g(2pB))-1 Ii

i=l
N-1
+p2
L [g(lk - i)pB) - g((k + i)pB)](I - g(2pB))-1 Iii=l
Let us estimate Wk and gk separately. We begin with Wk. Using the estimate
BQexp(-tB)
IIE-+E~ Mt-
exp(-c5t), c5 > 0,
2: O,t > 0,
(7.42)
we obtain
II Wk II
E~
+11(1 -
II
(I - g(2NpB))-l(g(kpB) - g((2N - k)pB))
II E-+EII Uo II
g(2NpB))-l(g((N - k)pB) - g((N + k)pB))IIE-+EII UN

1
~ M 1 _ e- 2NpO [II Uo
for all k, 1 ~ k
II
II E + II UN II d
N - 1. Since
<
liE ~ M 1 -
1 - e- 2N pO - 1 - e- Po '
we have that
max IIWk
19::::
N-1
e- P0 [II Uo
II E + II UN II d
To complete the proof of Theorem 7.2.1 it suffices to establish that
II
gk
II
E~ M
for all k, 1 ~ k
we can write
p2
0 [
1 - e- P
max
19:::: N
II
BII k
'
II +
max
19:::: N
II
Bh k
'
II]
(7.43)
N - 1. Using formula (7.38), identity (7.39) and Abel's formula,
gk = (I - g(2NpB))-l(g((N - k)pB) - g((N + k)pB)
(7.44)
369

N
Xp2 2:)g((N - i)pB)h,i - g((N + i - l)pB)h,iJ
+ p2 Lg((k + i
i=l
-l)pB)]h,i
i=l
-p2[Lg((k - i)pB)h,i + L g((i - k -l)pB)h,iJ =

i=l
i=k+1
gl + g~,
where
N
gl = (1 -
g(2NpB))-l(g((N - k)pB) - g((N + k)pB)p2 L[g((N - i)pB)h,i

i=l
_p2 Lg((k - i)pB)h,i,

i=l
g~ = -(1 - g(2N pB))-l(g((N - k)pB) - g((N + k)pB)p2 Lg((N + i -l)pB)h,i

i=l
+p2 L g((k + i - l)pB)Jh,i - p2 L g((i - k - l)pB)h,i'

i=l
i=k+1
gl and g~ separately. We begin with gl. Using estimate (7.42) and

11(1 - g(2NpB))-l(1 - g(2kpB))B- 11IE-<E :s M(8)
for all k, 1 :s k :s N, we obtain that
Let us estimate
Ilgl
N
X
II E:S 11(1 II
g(2NpB))-1(g((N - k)pB) - g((N + k)pB))B- 11IE-<Ep2
g((N - i)pB)
II
E-<EIIB fl,i
II E + p2 L
i=l
IIg((k - i)pB)IIE-<EIIh,i
II
i=l
N
:s M"'\:'
e- 6(N-j) p p2
~
j=l
max
II BII . liE
l<"<N,J
_J_
for all k, k = 1, ... , N - 1. Using the last inequality and (7.8), we obtain
for all k, k
= 1, ... ,N Ilgk2
1. In a similar way one establishes that
II E:S M 1 -
2
6
max
e- P P l~j~N
II
Bf2j liE
'
for all k, k = 1, ... , N - 1. From these estimates follows estimate (7.43). This
completes the proof of Theorem 7.2.1.
370
We say that the difference problem (7.37) has an m-th order of accuracy
in Fr(E) on the solutions v(t) of the boundary-value problem (7.35) if the error
vector
where M is independent not only 7, but also of c:.

Stability inequality (7.40) permits us to obtain the estimate of convergence
of (7.37) in Cr(E).
Theorem 7.2.2. Suppose that the function f(t) has m derivatives and
Ilf (m)(t)IIE S M,O S t S T.

estimate is valid:
(7.45)
II {v(t k ) - uk}f- 1 IlcT(E)S M7 m ,
and c:.
Proof. The components Zk = V(tk) - Uk of error vector zr =

relations
{zdf
satisfy the
_p-2(Zk+1 -2Zk-1 +Zk_1)+p-2(1-g(pB)) (Zk+1 +zk_d+p-2(g(2pB)-1)zk

ak = (2pB)-1 (a1,k
a1,k =
7-
ak,
+ a2,k+1 - g(pB)(a1,k+1 + a2,k)),
J
tk
exp(-(tk - S)c:- / 2B)f(s)ds 1
t k-l
m-1
2: h
f (j)(tk-d,
)=0
1 S k S N - 1,
Zo
= 0,
ZN
= O.
From this and the stability inequality (7.40) it follows that

max
l:::;k:::;
SM
N-1
P( - J p ) [ l:::;k:::;
max N
- exp
II
Zk
II
(7.46)
Ila1,k II E +
max N
l:::;k:::;
II Ba 2 k
'
II
E]'
371
Therefore to this end it suffices to estimate {a1,k}f" and {a2,df" in Cr(E). Using
Taylor's formula, we obtain
a1,k=7- 1
m-1
tk
exp(-(h-s)C!B)j(s)ds-
tk-l
j=O
exp( -(tk - s)
-21
B)(s - tk_d J. dsj ( J.)
1
(tk-1)~
J
tk-l
tk
+ 7-
j(j)(tk-1)
tk
'""" 7- 1
LJ
,j
J=O
m-1
LJ1
exp( -(tk - s)
-! B)
tk-l
(s - z)m j(m) (z) dzds
~!
tk-l
m-1
-Lh
(j) (tk-d
j=O
m-1
L.
tk
7-
J=O
exp( -(tk - s)
C~ B)(s -
tk-d j dsj(j)
(tk-d~
tk_l
J.
m-1
L J1,
j (j)(tk-d .
j=O
J
tk
a1, k
= 7-
J
S
exp( -(tk - s)
-1/2 B)
tk-l
(z - s)m-1 j(m)(z)dzdsj(m - I)!.
tk-l
In a similar way one shows that
J
tk
a2, k
= 7-
J
S
exp( -(s - tk-1)
-1/2 B)
tk-l
(z - s)m-1 j(m) (z) dzds j( m -I)!.
tk-l
Then, using the triangle inequality and estimate (7.40), we obtain
and
max
19::; N
Ila2k' liE
~ M c5p-1(1 - exp( -c5p))7 m
372
Note that for the convergence estimate I {V(t k ) - uk}f

suffices to obtain the stability inequality in FT(E).
IIFr(E)
~ M7 m it
estimate is valid:
Nl
M
{ukL - IIL p.r(E) ~ l-exp(-op){[
where M is
Uo
II E+II
UN
II
+ I {Bhdf IILp.r(E)]}' 1 ~ P < 00,

independent not only of fl, fI, uo, UN, 7, P but also of E.
+p2
[II {Bhdf
II
IILp.r(E)
estimate is valid:
N
II {UkL
+p2
[II
IlcQa(E)
r
~ 1 -exp-p
(fJ ) {
II
Uo
II E+ II
UN
II E
{Bhk}f Ilc~,a(E) + I {Bhdf Ilc~,a(E)]}'O ~ a ~ 1,
where M is independent not only of f[, fI, uo, UN,
7,
a but also of E.
The proofs of Theorem 7.2.3 and 7.18 follow the scheme of the proofs of
Theorems 7.1.3 and 7.1.5 and rely on formula (7.38) and estimate (7.42).
Banach spaces.
Theorem 7.2.5. The solutions of the difference problem (7.37) in c::,a(E) (0 <
a < 1) obey the coercivity inequality
D;II(uo, UN )uTllc~,Q(E)
+ I ATII(uo, UN )uTllc~'Q(E)
~ M{a(1 ~ a) [II {Bhdf IIc~,a(E)

N
+ II {Bhdl
1
Ilc~,a(E)l + 1 - exp (s:
-up) [II Auoil E
fl, 12,
uo, UN,
7,
+ IIAuNII E ]},
a but also ofE.
p-2(Uk+l - 2Uk
+ uk-d
+ Wk-l)
(7.47)
+ gk-l), 1 ~ k ~ N - 1.
{p-2(Wk+l - 2Wk + Wk_l)}f- l and {p-2(gk+l - 2gk +
+p-2(gk+l - 2gk
Let us estimate
gk_l)}f- l separately.
= p-2(Wk+l - 2Wk
373

p-2(Wk+l - 2Wk
(1 -
+(1 -
+ wk-d
e- 2NpB )-1(1 - e- pB )2 p -2(e-(k-1)pB _ e-(2N-k-1)pB)uo
e- 2NpB )-1(1 - e- pB )2 p -2(e-(N-k-1)pB - e-(N+k-1)pB)UN

2
LJi:\
m=l
where
Jf =
J~
(1 -
(I -
e- 2NpB )-1(1 - e- pB fp-2(e-(k-1)pB - e-(2N-k-1)pB)uo,
e- 2NpB )-1(1 _ e- pB )2 p -2(e-(N-k-1)pB - e-(N+k-1)pB)UN.
Let us estimate J:n = {Jk}~;ll for any m = 1,2 separately. We start with J[.
Using estimate (7.42), we show that, for 1 :S k :S N - 1,
IIJ~IIE :S 11(1 - e- 2NpB )-11IE--.EII(1 - e-pB)(pB)-lll~--.E
x Ile-(k-1)pB
- e-(2N-k-1)p B IIE--.EII AuoIIE
:S 1_ e_2NP61IAuoIIE :S 1_ e_p61IAuoIIE.
1
IIJ1
Jl+rli E :S IIJ1 11E + IIJl+rll E :S 2MII Au oilE :s;

rO
(2r)O
2M (1 + r)o
IIAuollE
rO
:S M 1 [(1 + r)o + (N _1)olIlAu oIIE ,

for 1 < r
+ 1 :S N
1. By Lemma 5.1.1, one has the inequality
Ilexp -kpB) - exp {-(k
+ r)pB}II E --. E :S M
rO
(k
+ r)o
(7.48)
for any 1 :S k < k + r :S N - 1 and 0 :S Q :S 1, where M does not depend on Q, k,

rand T.
Further, applying (7.42) and (7.48) we show that, for 2 :S k < k + r :S N -1,
IIJf -
Jf+rlIE
:S 11(1 - e- 2NpB )-11IE--.EII(1 - e-pB)(pB)-lll~--.E

x Ule-(k-1)pB - e-(k+r-1)p B IIE--.E
+lIe-(2N-k-1)pB _ e-(2N-k-r-1) pB IIE--.E]II AuoIIE
374
Therefore,
I {p -2 (Wk+l
- 2Wk
::; 1 -exp (8
- p ) [II
+ Wk-l) }N-l
1
Il c;-'" (E)
Auoil E + IIAuNII E ]
+ gk_l)}~-l.
Let us estimate {p-2(gk+l - 2gk
(7.49)
p-2(gk+l - 2gk + gk-l) =(1 - g(2NpB))-l(g((N - k -l)pB) - g((N + k -l)pB)

N-l
L g((N - i)pB)(h,i - h,N)
x(1 - g(pB))2
i=l
k-l
L(1 - g(pB))2 g((k - 1 - i)pB)(h,i - h,k)

i=l
- h,k+l + (2- g(pB))h,k+ (1 - g(2NpB))-I(g((N - k-l)pB) - g((N +k-l)pB)

N
k-l
X (1 - g(pB))2
g((N - i)pB)h,N +
g(pB))2 g((k - 1 - i)pB)h,k
L(1 -
i=1
i=1
-(1 - g(2NpB))-I(g((N - k - l)pB) - g((N + k - l)pB)

N-l
X (1 - g(pB))2 L g((N + i - l)pB)(hi - hN)
i=1
N
+(1 - g(pB))2 Lg((k + i
- 2)pB)](hi - 12,1)
i=1
N
-(1 - g(pB))2 L g((i - k - 2)pB)(hi - hk+l)

i=k+2
- hk
+ (2 -
g(pB))hk+l - (1 - g(2N pB))-1
375
x(g((N - k -l)pB) - g((N + k -1)pB)(1 - g(pB))2

N
x Lg((N + i - l)pB)12,N + (1 - g(pB))2 Lg((k + i - 2)pB)]12,l

i=l
i=l
-(1 - g(pB))2 L g((i - k - 2)pB)12,k+l = L pr,

i=k+2
m=l
where
pl =
(1 - g(2NpB))-l(g((N - k - l)pB) - g((N + k - l)pB)

N-l
x(1 - g(pB))2 L g((N - i)pB)(!I,i - !I,N),
i=l
k-l
P'f = L(1 - g(pB))2 g((k - 1- i)pB)(!I,i - !I,k),

i=l
pf
= - !I,k+l + (2 - g(pB))!I,k + (1 - g(2NpB))-l

x(g((N - k -l)pB) - g((N + k -l)pB)
k-l
x(1 - g(pB))2 Lg((N - i)pB)!I,N + L(1 - g(pB))2 g((k -1- i)pB)!I,k'
pt =
i=l
i=l
-(1 - g(2NpB))-1(g((N - k -l)pB) - g((N + k -l)pB)

N-l
x(I - g(pB))2
g((N + i - l)pB)(h,i - h,N)
2:
i=l
N
+(1 - g(pB))2 Lg((k + i - 2)pB)](12,i - 12,1),

i=l
N
P~ = -(1 - g(pB))2 L
g((i - k - 2)pB)(12,i - 12,k+l),
i=k+2
P~ = -
12,k + (2 - g(pB))12,k+l - (1 - g(2N pB))-l
x(g((N - k - l)pB) - g((N + k - l)pB)(1 - g(pB))2

N
x Lg((N + i -l)pB)12,N + (1 - g(pB)f Lg((k + i - 2)pB)]12,l

i=l
i=l
-(1 - g(pB))2 L g((i - k - 2)pB)12,k+l

i=k+2
376
Let us estimate p:n. = {Pk}~=11 for any m = 1,2,3,4,5,6 separately. We start

with
Using estimate (7.42), we show that, for 1 ~ k ~ N - 1,
pr
e- 2NpB )-1(1 - e-pB)B-IIIE--+E

x Ile-(N-k-l)pB - e-(N+k-l)pB IIE--+E
I pIllE ~ 11(1 -
N-l
X
11(1 -
e-pB)e-(N-i)pB
IIE--+E IIBh,i -
Bh,N
liE
i=l
N-l
~ M[~ (tN _~i)l-a]11
Bf{
IIC~'''(E)'
tN
f
o
Since
tN
f
o
it follows that
Thus, for 1 ~ k
ds
(tN - s)l-a'
ds
- tN
(tN - s)l-a - ~'
I PIIIE~ ~ I
~
Bf{
IIC~'''(E)'
N - 1 we have established the bound
max
l::;k::;N-l
I pIIIE~ Mil
a
Bf{
IIC"'''(E)'
(7.50)
Now let us estimate the difference PI+r - PI for 1 ~ k < k + r ~ N - 1. We will

consider separately the cases k ~ r, N - k ~ 2r and k > r, N - k > 2r. If k ~ r,
then (7.50) yields
pI+r - pIIIE~11 pI+r
liE + I pIllE
~ Ml~l+a I Bf{ Ilc~'''(E{a(k+ r)-a = ~211 Bf{ Ilc~'''(E)t~(tk + tr)-a.

If N - k ~ 2r, then (7.50) yields
pI+r - PIIIE~11 pI+r
~ Ml~l+a I Bf{ Ilc~'''(E)ra(N -
k)-a =
liE + I pIllE
~211 Bf{ Ilc~'''(E)t~(tN -
tk)-a.
377
Now let k > rand N - k > 2r. Let us represent the difference Pf+r following sums:
Pf+r -
Pf = (1 -
Pf
as the
g(2N pB))-l(g((N - k - r - l)pB) - g((N - k - l)pB)

N-l
+g((N+k-1)pB) -g((N+k+r-1)pB)(I _g(pB))2
i=l
g((N -i)pB)(h,i - h,N).
Further, applying (7.42) and (7.48) we show that, for 2 ~ k < k + r
N - 2,
e- 2NpB )-1(1 - e-pB)B-11IE---->E

x [I1(e-(N-k-l)pB - e-(N-k-r-l)pB) IIE---->E
II Pf+r - PfllE ~ 11(1 -
+ II(e-(N+k+r-l)pB - e-(N+k-1)pB) IIE---->E]

N-l
X
11(1 - e-pB)e-(N-i)pB IIE---->E IIBh,i - Bh,N
i=l
rO:
M[(N _ r)o:
rO:
N-l
T
(k + r)o:] ~ (tN _ ti)l-O: II Bf[
liE
Ilc;-'''(E)
~ ---;11 Bf[ IIc;-'''(E)t~(tk + tr)-O:(tN - tk)-O:.

Thus, for 1
k < k +r
N - 1 we have established the bound
From this and estimate (7.50) it follows that
II P{ Ilc;-'''(E)
~ ~lll Bf[ Ilc;-'''(E)'
IIPIIIC;-'''(E)
~ ~lll
Bf2
IIC;-'''(E)'
Now let us estimate P{. Using estimate (7.42), we show that, for 2 ~ k
k-2
+11(1 - e-pB)B-11IE---->E L 11(1 i=l
e-pB)e-(k-l-i)pB
IIE---->E
k-l
: : : ~ [2:::
(t
i=l
k
_:)1-0:]11 Bf[ IIC;-'''(E)'

t
N - 1,
IIBh,i - Bh,k
liE
378
Since
tk
ds_ _ __
tOtk
(h - s)l-Ot - ~'
it follows that
II
for 2 :S k
:S N
IIE:S : II
Pf
Bf{
Ilc;-'c>(E)
- 1. Since P'f = 0, it follows that

max
l~k~N-l
II
Pf
IIE:S Mil
a
Bf{
IlcC>'C>(E)'
(7.51 )
Now let us estimate the difference Pf+r - Pf for 1 :S k < k + r :S N - 1. We will

consider separately the cases k :S r, N - k :S 2r and k > r, N - k > 2r. If k :S r,
then (7.51) yields
:S
Ml~l+Ot II Bf{
If N - k
:S 2r,
Ilc;-'c>(E)rOt(k + r)-Ot =
then (7.51) yields
II
:S Ml~l+Ot II
~211 Bf{ Ilc;-'C>(E)t~(tk + q-Ot.
Pf+r -
Pf IIE:s11 Pf+r liE + II Pf liE
Bf{ Ilc;-'C>(E)rOt(N - k)-Ot
~211 Bf{ Ilc;-'C>(E)t~(tN -
Now let k> rand N - k > 2r. Let us represent the difference Pf+r following sums:
tk)-Ot.
Pf as the
k+r-l
Pf+r - Pf =
I: (I - g(pB))2 g((k + r -
1 - i)pB)(h,i - h,k+r)
i=k-r
k-l
-L
(I - g(pB))2 g((k - 1 - i)pB)(h,i - h,k)

i=k-r
k-r-l
(I - g(pB))2 g((k + r - 1 - i)pB)(h,k+I - h,k)
-L
i=l
k-r-l
+
L
i=1
379
4
(I - g(pB))2[g((k+r-1-i)pB) - g((k -l-i)pB)](!I,i - !I,k) =
L Rk\
m=1
where
k+r-l
(I - g(pB))2 g((k + r - 1 - i)pB)(!I,i - !I,k+r),

i=k-r
k-l
R~ = (I - g(pB))2 g((k - 1 - i)pB)(!I,i - !I,k),
i=k-r
k-r-l
=(I - g(pB))2 g((k + r - 1 - i)pB) (!I,k+l - !I,k),
Rl =
R2
Rk =
k-r-l
i=1
i=1
(I - g(pB))2[g((k + r - 1 - i)pB) - g((k - 1 - i)pB)](!I,i - !I,k).
Let us estimate R'k for any m = 1,2,3,4 separately. We start with

estimate (7.42), we show that, for 1 ::; k ::; N - 1,
II Rl liE::;
k+r-l
Rk.
Using
II(I - e- pB )e-(k+r-l-i)p B IIE--.E
i=k-r
xiiI - e- pB IIE--.E II!I,i - !I,k+r liE

Ad
1
::; (tk + tr)a [Ta(tN - tk+r_l)a
Ad
k+r-l
k+r-2
T
T
+ ifr (tk+r - tiP-a(tN - ti)a]11 11
::; (tk + ~r)a [iIr (tk+r - ti)l:a(tN - ti)al ll
Ilc~,O(E)
g Ilc~,O(E)'
Since
it follows that
380
Since
R% = (I - g(pB)) (g(2rpB) - g((k + r
l)pB))(h,k - h,k+r),
it follows that
I R%
IIE~ Ile- 2rpB - e-(k+r-l)pBIIE--->EII1 - e- pB
IIE--->E IIh,k -
h,k+r
liE.
Now let us estimate Rt. Using estimates (7.42) and (7.48), we obtain
II Rt "E~
k-r-l
2:
(I - g(pB))2[g((k + r - 1 - i)pB) - g((k - 1 - i)pB)]
IIE--->E
i=l
Since for k > rand N - k > 2r we have the bound
<
C
- (1 - o:)(tN - tk + tr)a r
it follows that
Ha
< (1 - o:)(tN - tk)Q Cr
Ha
,
381
Thus, we have shown that for any 2 :S k < k

holds:
+ r :S
N - 1 the following inequality
II P{
IIc~'<>(E):S a(~ a)
II f{
Ilc~'<>(E) .
Now let us estimate P;' Since

P~
=-
+ (2 - g(pB))/I,k + (I - g(2N pB))-l(g((N - k - 1)pB)
/I,k+l
-g((N +k -1)pB)(I - g(pB))(1 - g(N pB)/I,N + (I - g(pB))(1 - g((k -1)pB)/I,k'

it follows that
II P; Ilc~'<>(E) :S M II f{ Ilc~'<>(E) .
Finally, combining the estimates for

the estimate
p:r, =
{Pk' }~==-/, m = 1,2,3,4,5,6, we obtain
II {p-2(gk+l - 2gk + gk-lnf- 1 Ilc~'<>(E)
:S
a(1~ a) [II Bf{ Ilc~'<>(E) + II BI; Ilc~'<>(E)]'
Using the last estimate and (7.49), we obtain
II {p-2(Uk+l - 2Uk + uk_d}f- 1 Ilc~'<>(E)

:S M {a(1 ~ a) [II Bf{
+ 1-
1
exp( -Jp) [II
Ilc~'<>(E) + II BI; Ilc~'<>(E)]
Auoil E + IIAuNII E ]}
382
By the triangle inequality, this last estimate and the difference equation (7.37)
yield
II ArII(uo, UN )urllc;:""(E)
:::; M{a(l ~ a) [II
BN Ilc;:""(E) + II BN Ilc;:""(E)]
1
exp( -fJp) [ II Auoil E + IIAuNII E]}
+1Theorem 7.2.5 is proved.
Theorem 7.2.6. The solutions of the difference problem (7.37) in Cr(Ea ) (0
<a<
1) obey the coercivity inequality
fl, 12, UO,
UN,
T,
a but also ofe.
relies on the formula
p-2(Uk+l - 2Uk + Uk-I)
= (1 - e- 2NpB )-I(1 - e- pB )2 p-2(e-(k-l)pB - e-(2N-k-l)pB)uo

+(1 - e- 2NpB )-I(1 - e- pB )2 p-2(e-(N-k-l)pB - e-(N+k-I)pB)UN
+(1 - g(2NpB))-I(g((N - k - l)pB) - g((N + k - l)pB)
N
k-I
x(1 - g(pB))2 Lg((N - i)pB)!I,i - L(1 - g(pB))2 g((k - 1- i)pB)!I,i

i=1
i=1
- !I,k+1 + (2 - g(pB))!I,k - (1 - g(2NpB))-I(g((N - k -l)pB) - g((N +k -l)pB)

N
x(1 - g(pB))2 Lg((N + i -l)pB)/2,i + (1 - g(pB))2 Lg((k + i - 2)pB)]/2,i

i=1
i=1
N
-(1 - g(pB)f L g((i - k - 2)pB)/2,i - /2,k + (2 - g(pB))/2,k+1

i=k+2
and estimates (7.42) and (7.48).
383
almost coercivity inequality
II
:::;
+ II ATII(uo,uN)uTllcT(E)
D;II(uo,uN)uTllcT(E)
M{min{ln~, 1 + lin IIBIIE....EIIIHII {Bhk}f IlcT(E) + II {Bh,k}f
IlcT(E)]
+ 1 -exp(8
- p) [II Auoil E + IIAuNII E]},
f1,
f:;" un,
UN, T
but also of c.
Proof. By the definition of the norm in C;;,Q(E),
for all sufficiently small
0::
> O. Choosing 0:: in the best way, we obtain the estimate
+ II ATII(uo,uN)uTllcT(E)
II D;II(uo,uN)uTllcT(E)
1[ II {BhdlN IlcT(E)
:::; Mdln ~
+ 1 -exp (is
- p ) [ II
+ II
]
{Bh,dlN
IlcT(E)
Auoil E + IIAuNII E]}
By the definition of the norm in CT(E
Q ),
II D;II(uo,uN)uTIICT(E) +
II ATII(uo,uN)uTllcT(E)
1
N
:::; M{~IIBIIE .... e[11 {Bhdl IlcT(E)
+ II {Bh,df II cT (E)] + 1 _
for all sufficiently small
0::
ex~( -8p) [II Auoll E + IIAuNII E]}
> O. Choosing 0:: in the best way, we obtain the estimate
384
+ 1 -exp (8
- p) [II Auoil E + IIAuNII E]}
Note that if A is the unbounded positive operator in E, then
. { ln~,
1 1 + lIn II B
mm
IIE-.EI }
1
= ln~.

Theorem 7.2.8. The solutions of the difference problem (7.37) in Ce'''!(E) (0 :::;
, :::; 13, 0 < 13 < 1) obey the coercivity inequalities
II
T
D;II(uo, UN )u II Cr (E13_-Y) :::;
13(~ 13) [II {Bhd~
II Cr (E13_-Y)
+ II {Bh,k}~ II Cr (E13_)
+ MIl
(8 ) [ II (I - e-pB)2p-2uo - hI
-exp-p
II E13 _-y
+ 11(1 - e-pB)2p-2uN - h,N II E13--y],
II
D;II(uo,uN)uTllc~'-Y(E)
+ II ATII(uo,uN)uTllc~'-Y(E)
I
N
:::; 13(1M_ 13) [
II {Bhkh
IIc~'-Y(E) + II {Bh,khN Ilc~'-Y(E)1
1( 8 )[I(I-e-PB)2p-2uo-hll~'''!
+MI1
-exp - p
+ I(I - e-pB)2p-2uN - h,N!g,"!],

where M I is independent not only of f[,
f2,
uo, UN,
T,
13, , but also of
Here, Iwlg,"! denotes that the norm of the Banach space Eg,"! consists of those
vEE for which the norm
Iwlg,"! =
max Ile-~BwIIE
l~n~N-I
sup
l~n<n+r~N-I
is finite.
t; 13 (tn
)B
+ tr)"!(tN - tnPII(e- ( !n.!.r.
<
e- !n. B )wlIE
<
385
Theorem 7.2.9. The solutions of the difference problem (7.37) in C~,"f(Eo:-{3) (O:S
'Y :S (3 :S 0:, 0 < 0: < 1) obey the coercivity inequalities
D;II(uo, UN )u'Tllc~'-'(E"_i3)
+ II
A'TII(uo, UN )u'Tllc~'-'(E"_i3)
+ II D;II(uo,uN)u'Tllc,.(E~:!i3)
MI
:S 0:(1 _ 0:) [II {Bh,kh IIc~'-'(E"_i3) + I {Bh,kh IIc~'-'(E"-i3)]

1 ( 6 ) [1(1 - e-pB)2p-2uo -exp - p
+ \(1 - e- pB )2 p-2 UN - h,NI~:2'{3],
+ MIl
h,11~'2'{3
I D;II(uo,uN)u'Tllc~'-'(E"_i3) + I A'TII(uo,uN)u'Tllc~'-'(E,,_,B)
+ I D;II(uo, UN )u'Tllc,.(Ea_-,)
~
:S 0:(1 _ 0:) [II {Bh,kh IIC~'-'(Ea-i3)+ II {Bh,kh Ilc~'-'(E"-i3)]

+ MIl
(6 ) [11(1 - e-pB)2p-2uo - h,1
-exp - p
I C,.(E a--,)
+ 11(1 - e-pB)2p-2uN - h,N II C,.(Ea_-,)]],

where M I is independent not only of fl, f:;" uo, UN,
T,
0:, (3, 'Y but also of c.
Here, Iwl~:2'{3 denotes that the norm of the Banach space E~:2'{3 consists of
is finite.
Note that the spaces of grid functions C~,"f(Eo:_{3), in which coercive solvability has been established, depend on the parameters 0:, (3 and 'Y. However, the
constants in the coercive inequalities depend only on 0:. Hence, we can choose the
Let us consider now the boundary-value difference problem (7.37) in the
spaces Lp,'T(E) = Lp([O, T]n E), 1 :S p < 00 of all grid functions.
Theorem 7.2.10. The solutions of the difference problem (7.37) in Lp,'T(E) obey
386
:S
M{min{ln~, 1 +
lIn IIBIIE--->EIIIHII
{Bhk}~
IILp.r(E) +
\I
{Bh,k}~
IILp,r(E)]
1
+ 1 -exp (<5
- p) [II Auoil E + IIAuNII E]},
f1,
fI, uo, UN,
7,
P but also of c.
Recall that if A is the unbounded positive operator in E, then

min
{ln~, 1 + lIn II B II E--->E I} = ln~.
Theorem 7.2.11. Suppose that the difference problem (7.37) is well posed in
LpO,T(E) for some Po, 1 < Po < 00. Then it is well posed in Lp,T(E) for any
p, 1 < p < 00 and the following coercivity inequality holds:
II D;II(uo,uN)uTIILp.r(E)
+lluTIICr(El_l/P.p)
:S M(po)[l _
II ATII(uo,uN)uTIILp.r(E)
ex~( -<5p) [II uoIIEl_l/p,P + IIUNIIE1_1/p,p]
+ p~ 1 [II {Bhk}~
IILp.r(E) +
f1, 12,
II {Bh,k}~
uo, UN,
7,
IILp.r(E)]]'
P but also not on c.
00 and 0 < a < 1. Then problem (5.29) is well posed

in Lp,T(Eo:,p) and the following coercivity inequality holds:
Theorem 7.2.12. Let 1 :S p :S
II D;II(uo, UN )uTII Lp,r(EQ,p) + II ATII(uo, UN )u IILp,r(EQ,p)
:S M 1 _ exp( -<5p) [II AUoIIEQ,p + IIAuNIIEQ,pl
M
N
+ a(l _ a) [II {Bhkh
N
IILp,r(EQ,p) + II {Bh,kh IILp,r(EQ,p)],
f1, 12, Uo,
UN,
7,
p, a but also not on c.
< p, q < 00 and 0 < a < 1. Then the difference problem

(5.29) is well posed in Lp,T(Eo:,q) and the following coercivity inequality holds:
7.3. A Cauchy Problem for Hyperbolic Equations
387
:s M(q)[l _ exp( -8p) [II AuOIIEa,q
+ IIAuNIIEa,q]
+ p~ 1 [II {Bhdf IILp,T(Ea,q) + II {Bh,kW IILp,T(Ea)],

where M(q) does not depend on j[, f:;' uo,
UN, T,
p,
but also not on
E.
Remark 7.2.14. Theorems 7.2.5-7.2.13 permit us to obtain the error estimates of

the high order of accuracy difference schemes for elliptic equations with a small
parameter multiplying one-derivative term in the various Banach norms.
7.3
A Cauchy Problem for Hyperbolic Equations
We consider the abstract Cauchy problem for hyperbolic equations

E
V"(t)
+ Av(t)
= f(t)
(O:s t :s T), v(o) = Vo, v'(O) = v~
(7.53)
in an arbitrary Banach space E with the positive operator A and E E (0, (Xl).
A function v(t) is called a solution of the problem (7.53) if the following
conditions are satisfied:
i. v(t) is twice continuously differentiable on the segment [0, T]. The derivatives
derivatives.
ii. The element v(t) belongs to D(A) for all t E [0, T] and the function Av(t) is
iii. v(t) satisfies the equations and the initial conditions (7.53).
Suppose that the function f(t) is not only continuous, but also continuously
differentiable on [0, T]' Vo E D(A) and vb E D(A~). Then by formula (6.2), we
have that the formula
v(t)
t
t,
1
c( - )vo + s( - )vo + 2"
E
J
t
t-s
s(-)f(s)ds
E
(7.54)
gives a solution of problem (7.53). Here
Suppose further that in an arbitrary Banach space E the assumptions
Ilc(t)IIE--+E :s M, IIA~s(t)IIE--+E:S M, O:s: t < (Xl

are valid.
(7.55)
388
In Chapter 6, stable high order of accuracy difference schemes generated

by an exact difference scheme or by Taylor's decomposition on the three points
for the numerical solutions of an abstract initial-value problem (7.53) for e = 1
were presented. Unfortunately, these difference schemes cannot be applied for the
approximate solutions of (7.53) in the general cases e E (0,00). In this section
we study the high order of accuracy two-step uniform difference schemes of the
approximate solutions for differential equations of the hyperbolic type with an
arbitrary parameter e on the highest derivative. The stability estimates of the
solutions of these difference schemes are obtained.
On the segment [0, T] we consider again a uniform grid
[O,T]T
= {tk = kT,k = 0,1, ... ,N,N7 = T}
with step 7. Let 7 = pc and B 2

exact two-step difference scheme:
= A.
By Theorem 3.1.1, we have the following
1
T
7
7,
1
7 - Z
V(7) - v(O) = (c( -) - I)v(O) + s( -)v (0) + 2"
s(--)j(z)dz
e
ceo
e
or
Uo
VO, Ul
7 7 s( - )v'(O)
7h 1
= c( - )v(O)
e
e
'
Applying this exact two-step difference scheme, we can construct the high order
of accuracy difference schemes
h,k = (7A
389
1[J(tk) - c(p)f(tk-dl- (7A)-1
m-l
Bjf(jH)(tk_d,
j=O
h,k = -(e 27)-l s(p)f(tk_d + (e 27)-1
m-l
L Cj f(j+l)(tk_l),
j=O
B . = (e 2A- 1)j B - ~(e2 A- 1)j-iH

2J
~
t=1
7 2i - 1
2 < 2)' < m - 1

(2i _ I)! , ,
2i
B = (e 2A- 1)j B - ~(e2 A- 1)j-i+I2- 3 < 2)' + 1 < m - 1

2J+l
1
~
(2i)!' ,
t=1
Co = e2 A- 1 (1 - c(p)),
CJ
7j
= _(e 2A)-1 B_
+ (e 2A)-I_.,' 1 <)'
<
m-1
J 1
).
for the approximate solutions of the initial-value problem (7.53).

Uk
c(kp)vo + s(kp)v~
(7.57)
+ L {c((k - m)p)h,m + s((k - m)p)h,m}7, 1::; k::; N.

m=1
Actually, formula (7.57) is true for k = 1. Let k = 2. Then using the formula
7
U2 + Uo = 2c( - )Ul + T[h 2 + s( -)12 1

e
'e'
c( -)h 1],
e'
we obtain
7
U2 = -vo + 2c(- )(c( - )v(O) + s( -)v (0) + Th 1) + T[h 2 + s( -)12 1 - c(-)h 1]

e
e
e
'
,
e'
e'
= c(2p)v(0) + s(2p)v'(0) + T[h,2 + S(P)h,l + C(P)h,IJ

= c(2p)v(0)
+ s(2p)v'(0) + 7
L {c((2 - m)p)h,m + s((2 - m)p)h,m}.

m=1
Thus, formula (7.57) is true for k = 2. Applying mathematical induction, one can
easily show that it is true for every k. Namely, assume that formula (7.57) is true
for k - 1 and k. Using the formula
390
we obtain
Uk+l = -Uk-l + 2C(p)Uk + 7[h,k+l + S(p)h,k - C(p)h,k]

k-l
2: {c((k-1-m)p)h,m+ s((k-1-m)p)h,m}7}
= -{c((k-1)p)vo+s((k-1)p)v~+
m=l
+2c(p){c(kp)vo + s(kp)v~ +
2: {c((k - m)p)h,m + s((k - m)p)h,m}7}

m=l
+7[!t , k+l + s(-)h

kE:'
C( -)!t
c'
{2c(p)c(kp) - c((k - 1)p)}vo
k]
+{2c(p)s(kp) - s((k -
1)p)}v~
k-l
+ 2: {2c(p)c((k - m)p) - c((k -
1 - m)p)h,m{2c(p)s((k - m)p)
m=l
-s((k - 1 - m)p)h,m}7}
7
+2c(; )fl,k 7 + 7[h,k+l + S(p)h,k - c(p)fl,k]

=
c((k + 1)p)v(O) + s((k + 1)p)v'(O)
k+l
+7
L {c((k + 1 - m)p)h,m + s((k + 1- m)p)h,m}.
m=l
Thus, formula (7.57) is true for k + 1. Therefore, formula (7.57) is true for all k.
Theorem 7.3.1. For the solution of the exact two-step difference scheme (7.56) the
following stability inequality holds:
where M does not depend on E,
7,
h,k,
h,k, 1 ~ k
N and Va, v~.
The proof of Theorem 7.3.1 is based on formula (7.57) and estimates (7.55).
Stability inequality (7.58) permits us to obtain the estimate of convergence
of (7.56) in Cr(E).
391
Theorem 7.3.2. Suppose that the function f(t) has m
+ 1 derivatives
and
Ilf (m+I)(t)IIE ::; M,O ::; t ::; T.

estimate is valid:
m
(7.59)
II {v(t k ) - udf IlcT(E) ::; MT ,
where M does not depend on T and E.
Proof. The components Zk = V(tk) - Uk of error vector ZT = {zk}f satisfy the
relations
p-2(Zk+1 - 2Zk
where
+ Zk-l)
2p-2(C(p) - I)zk
Zo
0,
Zl
+ T- I E2ak,
= Tal,l,
+ s(p)a2,k - c(p)al,k, 1::; k::;
ak = al,k+l
~
1 ::; k ::; N - 1,
N - 1,
al,k = hk - hk, a2,k = hk - hk, 1::; k ::; N.

From this and the stability inequality (7.58) it follows that
max II
l::;k::;
N
Zk
II
E ::;
Iial' k II E + E l::;k::;
max IIA-! a2 k II eJ
l::;k::; N
N
'
M [ max
(7.60)
Therefore to this end it suffices to estimate {al,k}t' and {a2,dt' in CT(E). Integrating by parts and using Taylor's formula, we obtain
m-l
-(TA)-l
Bjf(j+l)(tk_d}
j=O
m-l
+(TA)-l
j=O
Bjf(j+l)(tk_l) .
392
tk
tk - S
.)
1
c(--)(s
- tk-1F. dsf (J+1
(tk-1)~
tk-l
e
J.
m-1
-(TA)-l
B j f(j+1)(tk_1)'
- (TA)- 1
I:
j=O
a1, k = -(TA)-l
tk
tk-l
tk - S
c(--)
e
J8
(z - s)m-1 f(m+1)(z)dzds/(m - I)!.
tk-l
In a similar way one shows that
J
8
(z - S)m-1 f(m+l) (z)dzds/(m - I)!.
tk_l
Then, using the triangle inequality and estimate (7.55), we obtain
and
Remark 7.3.3. Theorem 7.3.2 permits us to obtain the error estimates of the
high order of accuracy difference schemes for hyperbolic equations with a small
parameter multiplying time-derivative term in the various Banach norms.
Remark 7.3.4. Applying this approach we can construct the high order of accuracy uniform difference schemes for three singular perturbation problems involving
second-order differential equations in a Banach space E. The equations are
e2 v"(t) + v'(t) = Av(t) + f(t),

e2 v"(t) - iv'(t)
e v"(t) + v'(t)
2
Av(t) + f(t),
(e A + B)v(t) + f(t),
2
where A, B are linear, generally unbounded operators in E. In each case, the usual
initial conditions are given:
v(O)
= VA,
v'(O)
= v~.
Chapter 8
Appendix: Delay Parabolic
Differential Equations
Chapter 8 is devoted to study of the initial-value problem for delay partial differential equations of the parabolic type. A sufficient condition for stability of the
solution of this initial-value problem is given. The stability estimates for solutions
of the first and second order of accuracy difference schemes for approximately
solving this initial-value problem are presented. The stability estimates in Holder
norms for solutions of the initial-value problem of the delay differential and difference equations of the parabolic type are obtained.
8.1
The Initial-Value Differential Problem
In the present section we consider the initial-value problem for linear delay differential equations
d~~t)
{
+ Av(t) = B(t)v(t - w), t 2: 0,

v(t) = g(t)( -w ::; t ::; 0)
(8.1)
in an arbitrary Banach space E with the unbounded linear operators A and B(t)
in E with dense domains D(A) s:; D(B(t)). Let A be a strongly positive operator,
i.e., -A is the generator of the analytic semigroup exp{ -tA}(t 2: 0) of the linear
bounded operators with exponentially decreasing norm when t ----7 00. That means
the following estimates hold:
Ilexp{-tA}IIEJ-+E::; Me- 6t ,lltAexp{-tA}IIEJ-+E::; Me- 6t ,t>0
(8.2)
for some M > 1, <5 > O. Let B(t) be closed operators. The operator function B(t)
is strongly continuous on D(A).
A function v( t) is called a solution of problem (8.1) if the following conditions
are satisfied:
i. v( t) is continuously differentiable on the interval [-w, 00). The derivative at
the endpoint t = -w is understood as the appropriate unilateral derivative.
394
Chapter 8. Appendix: Delay Parabolic Differential Equations
ii. The element v(t) belongs to D(A) for all t E [-w, (0), and the function
Av(t) is continuous on the interval [-w, (0).
iii. v(t) satisfies the equation and the initial condition (8.1).
A solution v(t) of the initial-value problem (8.1) is said to be stable if
Ilv(t)IIE'S max Ilg(t)IIE

-w~t~O
(8.3)
for every t, -w 'S t < 00. We are interested in studying the stability of solutions
of the initial-value problem under the assumption that
(8.4)
holds for every t 2: O. We have not been able to obtain estimate (8.3) in an
arbitrary Banach space E. Nevertheless, we can establish the analog of estimates
(8.3) where the space E is replaced by the fractional spaces Ea(O < a < 1) under
a stronger assumption than (8.4). Stability estimates for the solutions of the first
and second order of accuracy difference schemes for approximately solving this
initial-value problem are presented. The stability estimates in Holder norms for
solutions of the initial-value problem of delay differential and difference equations
of the parabolic type are obtained.
The strongly positive operator A defines the fractional spaces EO!. = EO!. (E, A)
(0 < a < 1) consisting of all u E E for which the following norms are finite:
IlulIE"
sup p.I-O!. Aexp{ ->.A}uIIE.

),>0
First, we consider problem (8.1)when A-I and B(t) commute, i.e.,

A-I B(t)x
= B(t)A-Ix,
x E D(A).
(8.5)
Theorem 8.1.1. Assume that the condition
(1 - a)
IIB(t)A- IIE>-;E'S M2 2 - a
(8.6)
holds for every t 2: 0, where M is the constant from (8.2). Then for every t 2: 0
we have
(8.7)
J
t
v(t) = exp{ -tA}g(O)
exp{ -(t - s)A}B(s)g(s - w)ds
(8.8)
8.1. The Initial-Value Differential Problem
395
the semigroup property, condition (8.5) and estimates (8.2), (8.6), we obtain
.\l- o IIAexp{ -.\A}v(t) liE :S .\l- o IlAexp{ -(.\ + t)A}g(O)IIE
.\+t-s
+.\ 10J
IIAexp{ 2
A}IIE--..EIIB(s)A- IIE--..E
t
o
x IIAexp{ -
for every t, 0 :S t :S wand .\,.\
.\+t-s
2
A}g(s - w)IIEds
> O. This shows that

(8.9)
for every t, 0 :S t :S w. Applying mathematical induction, one can easily show that
it is true for every t. Namely, assume that inequality (8.7) is true for t, (n -l)w :S
t :S nw, n = 1,2,3, ... , for some n. Letting t = s + nw, we have
dv
ds + Av = B(s + nw)v(s + (n - l)w), 0 :S s :S w.
max
nw<::;s <::; (n+1)w
for every t, nw :S t :S (n
Ilv(s - w)IIE a :S
max
-(n-1)w<::;t<::;nw
+ l)w, n = 1,2,3, ....
and .\,.\
IIv(t)IIEa
> O. This shows that
for every t, nw :S t :S (n + l)w, n = 1,2,3, .... This result completes the proof of
Theorem 8.1.1.
Now, we consider problem (8.1) when
for some t 2: O. Note that A is a strongly positive operator in a Banach space

E iff its spectrum a(A) lies in the interior of the sector of angle cp,O < 2cp < 71",
symmetric with respect to the real axis, and if on the edges of this sector, 8 1 =
396
[z = pexp(i~) : 0 :S p < 00] and 8 2 = [z = pexp( -i~) : 0 :S p <

it the resolvent (z - A)-l is subject to the bound
-1
II(z - A)
00] and outside
M1
IIE-E:S 1 + Izl
(8.10)
for some M 1 > O. First of all let us prove two lemmas. These will be needed in the
sequel.
Lemma 8.1.2. For any z on the edges of the sector,
8 1 = [z = pexp(i~) : 0 :S p < 00]
and
8 2 = [z =
and outside it the estimate
IIA(z - A)-l X II E :S
pexp(-i~): O:S p
MO: MO:(l
< 00]
+ M )1-0:2(2-0:)0:
a)(~ + Izl)o:
IlxllEa
a(l _
holds for any x E Eo:. Here and in the future M and M 1 are the same constants
as in estimates (8.2) and (8.10).
Proof. Applying the formula
00
x=
Aexp{-rA}xdr,
and using the semigroup property, we can write
A(z-A)-l X =
A(z-A)-lAexp{-rA}xdr
roo (z _ A)-l Aexp{ -~A}Aexp{-~A}xdr.

2
2
iN
Using estimates (8.2), (8.10) and the definition of the spaces E a , we obtain
IIA(z - A)-l x II E :S
05
IIA(z - A)-lIIE_eIIAexp{ -rA}x//Edr
roo II(z _ A)-lIIE>-+EIIAexp{ -~A}IIE>-+E/IAexp{

-~A}x"Edr
2
2
iN
(1 C
N
1+ M 1 )T"-IdT +
Here
L
oo
MII~2:Qt-
NO:
'ljJ(N) = (1
dT)
II xii Eo
~ ,pC N) II'I'll
Eo
M M 1 22 -0: No:- 1
+ Md~ + (izi + 1)(1 _
a) .
Taking the minimum of the function 'ljJ(N) over all N, 0 < N <
statement of Lemma 8.1.2.
00, we obtain the
397
Lemma 8.1.3. Let for all s 2: 0 operator B(s)A- I - A-I B(s) with domain which
coincide with D(A), admit a closure Q = B(s)A-I - A-I B(s) bounded in E. Then
for all T > 0 the following estimate holds:
IIA-I[Aexp{-TA}B(s) - B(s)Aexp{-TA}]xIIE
< e(n + l)M<>M{+<>(1 + 2MI )(1 + M I )I-<>2(2-<<>IIQIIE>-+EllxIIEo

TI-<>7l'n2(1 - n)
Here Q = A-I(AB(s) - B(s)A)A-I.

Proof. Applying the Cauchy-Riesz formula, we obtain the representation
Aexp{ -TA}x =
~
2m
exp{ -TZ}A(z - A)-Idzx
S\US2
J eXP{-TZ}dzx+~ J
=-~
2m
2m
~U~
for every
T, T
exp{-Tz}z(z-A)-Idzx
~U~
> O. From this it follows that

A-I [A exp{-TA}B(s) - B(s)Aexp{-TA}]x
2~i
exp{ -TZ }zA-I[(z - A)-I B(s) - B(s)(z - A)-I]dzx
S\US2
2~i
exp{ -TZ}Z(Z - A)-IQA(z - A)-I]dzx.
S\ US2
Making the substitution z = -~

parts, we can write the formula
e
= --
27l'
1
1
e
= --
27l'T
+ iy, -00 <
y <
00
and using integration by
A-I [A exp{ -TA}B(s) - B(s)Aexp{ -TA}]x
00
exp(-Tiy)(--1 +iy)(--1 +iy-A)-IQA(--1 +iy-A)-Idyx

T
T
T
-00
00
-00
exp(-Tiy)(--1 + iy - A)-I[Q - (--1 + iy)(--1

T
T
T
+ iy - A)-IQ
-(--1 + iy)Q( --1 + iy - A)-I]A(--1 + iy - A)-Idyx.
T
T
T
By Lemma 8.1.2 and performing the change of variable Ty = p, and using estimate(8.1O), we obtain
IIA-I[Aexp{-TA}B(s) - B(s)Aexp{-TA}]xIlE
:s _e_
Joo lexP(-TiY)III(-~T +iy 27l'T
A)-IIlE>-+E(1 + 2MdIIQIIE>-+E
-00
xllA(-- +iy-A)-ldyxIIE
T
398
<e-
27TT
00
-00
(~
dy
+ y2) Ito
(1 + 2M1)IIQIIE>-+EMf+0 M O(l + M1)1- 02(2-0)01IxIIEo

a(l - a)
X-'------=-.:-:.:.....:...:.:..=--=-=---'----:----:.-.,------=-:.....--------.:~=
roo
e
7TT 1 -
io
dp
(1
+ p2) Ito
(1 + 2M1)IIQIIE>-+EMf+0MO(l + Md 1- 02(2-0)01IxIIE
a(l - a)
o
x-'----.:....:.:.-.:....:.:.-------'''-----:-------'--'---------'-'-----'.:--'::.
< e(l + a)(l + 2MdIIQIIE>-+EMf+0 MO(l + M 1)1- 02(2-0)01IxIIEo

7TT 1 - o
Lemma 8.2 is proved.

Suppose that
a 2 (1
- a)
IIA-l(AB(t) - B(t)A)A- 1I1 E>-+E
(8.11)
7T(1 - a)2 a 2c
- eMl+oMf+(1 + 2Md(1 + M1)l-022+0-0 (1 + a)
<---....--;-------'------'----.,-----:;---2
holds for every t 2: O. Here and in the future c is a some constant, 0 :::; c :::; 1.
Theorem 8.1.4. Assume that the condition
IIA-IB(t)11
< (1- a)(l - c)

E>-+E M2 2 - 0
(8.12)
holds for every t 2: O. Then for every t 2: 0, estimate (8.7) holds.

Proof. Let 0 :::; t :::; wand >., >.
> O. Then using formula (8.8), we have that
>.1-0 Aexp{ ->'A}v(t)
J
t
+>.
-a
exp{ -
>.+t-s
2
>.1-0 Aexp{ -(>. + t)A}g(O)
A}B(s)Aexp{ -
>.+t-s
2
A}g(s - w)ds
o
t
+>.1-0
exp{ -
>.+t-s
>.+t-s
2
A}[Aexp{2
A}B(s) - B(s)A
xexp{-
>.+t - s
2
A}]g(s-w)ds
=h+h+h,
where
>.1-0 Aexp{ -(>. + t)A}g(O),

t
12 = AI - o
exp{ -
A+t-S
2
399
A}B(s)Aexp{ -
A+t-S
2
A}g(s - w)ds,
13 = AI - o
exp{ -
A+t-S
2
A}[Aexp{ -
A+t-S
2
A}B(s) - B(s)A
x exp{ -
A+t-S
2
A}]g(s - w)ds.
Using estimates (8.2), (8.10) and condition (8.12), we obtain

IlhilE = AI-oIIAexp{ -(A + t)A}g(O)IIE
xIIAexp{-
A+t - s
2
A}g(s-w)IIEds
AI - o
:S -~~o Ilg(t)IIEa (1- (A + t)l-o )(1- c)
for every t, 0 :S t :S wand A, A> o. Now let us estimate
using estimate (8.11), we can obtain
h. By Lemma 8.1.3 and
IlhilE :S AIOJ
IIAexp {A+t-S
2
A}IIE>-+E
t
o
x IIA-I[Aexp{ -
A+t-S
2
A}B(s) - B(s)Aexp{ -
A+t-S
2
A}]g(s - w)IIEds
:S AI - oe(l + a)MHo M{+(1 + 2MI )(1 + M I )I-02(2-0)0

x
IIA-I(AB(s) - B(s)A)A- I II E>-+E2 2 - O llg(s - w)IIE a ds

io
(A + t - s)2-o na 2(1 - a)
:S max IIA-I(AB(s) - B(s)A)A-IIIE>-+E

o<s<w
400
r e(l + o:)Ml+a M l+a (1 +

t
Jo
(A + t -
2M1 )(1
+ M 1 )I- a 2(2-a)a2 2- a ds
s)2-a1r0: 2 (1
- 0:)
.V-a
_~:s~~)lg(t)IIEa ~ -~:s~~o IIg(t)IIEa(1- (A + tF-a) E
for every t, 0
wand A, A > O. Using the triangle inequality, we obtain that
for every t, 0
wand A, A> O. This shows that
for every t, 0 ~ t ~ w. In a similar manner with Theorem 8.1 applying mathematical induction, one can easily show that it is true for every t. This result completes
the proof of Theorem 8.1.4.
8.2
The Difference Schemes
Using the first and second order of accuracy implicit difference schemes for differential equations without delay, we have the following difference schemes for
approximate solutions of the initial-value problem (8.1)
~(Uk - Uk-I) + AUk = BkUk-N, Bk = B(tk),

tk = kT, 1 :s; k, NT = W,
{
Uk = g(h),tk = kT,-N ~ k ~ 0,
~(Uk - Uk-I) + (A + ~TA2)Uk = ~(I + ~A)Bk(Uk-N + Uk-N-l),

{ B k = B(tk - ~),tk = h,l ~ k,Uk = g(tk),tk = h,-N ~ k:S; 0.
(8.13)
(8.14)
Theorem 8.2.1. Assume that all conditions of Theorem 8.1.1 are satisfied. Then
for the solution of difference scheme (8.13) the estimate
(8.15)
holds for any k 2 1.

Proof. Let us consider 1 ~ k
N. In this case
k
Uk = Rkg(O)
+L
R k-j+l Bjg(tj-N )T,
)=1
where R
= (1 + TA)-l. Using the formula
J
00
(1 +TA)-k
= (k ~ I)!
tk-1exp(-t)exp(-TtA)dt,k 2 1,
(8.16)
401
8.2. The Difference Schemes
condition (8.5) and estimate (8.6), we obtain

1
a
,X1- IIAexp{ -'xA}ukIlE ::; ,Xl-a (k _ 1)!
1
1
00
t k- 1 exp( -t)IIAexp{ -(Tt + 'x)A}g(O)IIEdt
00
1 a '"
~ T (k _ j)!
+,X -
)=1
k .
(Tt + 'x)A }
1
t -) exp( -t)IIAexp{ 2
IIE>-+EiIBj A- IIE>-+E
+ 'x)A
(Tt
IIAexp{ -
}g(tj-N)IIEdt
00
::;,X
I-a
1
(k -1)!
o
+,X
1- a ~
1
M2 2- a ~ T (k _ j)!
I-a
)=1
::; J
dt
exp(-t) (tT + ,X)l-a IIg(O) liE"
k-1
1
00
M2
k-j
2 a
- dt
exp( -t) (tT + ,X)2-a Ilg(tj-N )IIE"
IIg(tk)IIE",
max
-N95,O
where
J
00
I-a
J = ,X
(k _ 1)!
k-1
)
dt
exp( -t (tT + ,X)l-a
J
00
1-a( 1 - a ) '~
" T (k _1 j)!
+,X
)=1
k-j
()
dt
exp -t (tT + A)2-a .
We shall calculate J. Making the substitution k- j = m and using Taylor's formula

and integration by parts, we obtain that
J
00
J=A
I-a
1
(k-1)!
k-1
()
dt
exp-t(tT+'x)l-a
00
I-a
=A
1
(k _ 1)!
00
+,X
I-a
(1 - a)
exp(t)[l -
it
a
k-1
(
dt
exp -t) (tT + ,X)l-a
~k-1 exp( -~)~

dt
(k _ 1)!
] exp( -t) (tT + A)2-a
402
J
00
I-a
= A
1
(k _ 1)!
k-l
exp( -t) (tT
dt
+ A)l-a + 1
00
-A
I-a
(k _ 1)!
k-l
exp( -t) (tT
dt
+ A)l-a
= 1.
Therefore,
for every k, 1 :s: k
:s:
N and A, A > O. This shows that
(8.17)
for every k,l :s: k :s: N. Applying mathematical induction, one can easily show
that it is true for every k. Namely, assume that estimate (8.15) is true for the
k, (n - l)N :s: k :s: nN, for somen = 1,2,3, .... Letting k = m + nN, we have
1
-(um - Um -l)
T
+ AUm =
Bm+nNUm-N, 1 :s: m
:s: N.
for every k, nN
:s: k :s: (n + l)N, n = 1,2,3, ... , and A, A> O.
for every k, nN :s: k

of Theorem 8.2.1.
:s: (n + l)N, n =
This shows that
1,2,3, .... This result completes the proof
Theorem 8.2.2. Assume that all conditions of Theorem 8.1.4 are satisfied. Then
for the solution of the difference scheme (3.13) the estimate (8.15) holds for any
k 2: 1.
Proof. Let us consider 1 :s: k
:s: N.
Using the formula (8.16 ), we can write
J
00
AI-a Aexp{ -AA}Uk = AI-a (k
~ 1)!
t k- 1 exp( -t)Aexp( -(Tt + A)A)g(O)dt
+A
00
a "L....t T (k _1 j)! J tk-)

' ex p (-t)ex p(--2Tt+A A }Bj A
)=1
403
Tt + >.
x exp( --2-A}g(tj-N )dt
11
00
+>.1-0 ~
LJ T (k _ j)!
J=l
Tt
Tt + >.
t k -J" exp(-t)Aexp( --2-A}A-1
+ >.
Tt
+ >.
x [exp(--2-A}ABj - B j Aex p{--2- A }]g(tj-N)dt
= T 1 +T2 +T3 .
Using estimates (8.2), (8.10) and assumption (8.12), we obtain
1
00
IIT111E :::; >.1-0 (k
~ 1)!
exp( -t)IIAexp{ -(Tt + >')A}g(O)IIEdt
1"
00
IIT21lE :::; >.1-0 LJ T (k _ j)!
k J
-
Tt+ >.
exp( -t)IIAexp{ - - 2-A}IIE->E
J=l
1
Tt + >.
x IIBjA- IIE->EIIAexp{ --2-A}g(tj-N )IIEdt
1
J
>.1-0(1- Q)(l- E) k
1
:::;
M2 2- 0
LT (k _ j)!
00
M2 2 - O dt
J exp( -t) (tT + >')2-0 Ilg(tj-N )IIE o
k-"
J=l
oo
<
- >.1-0(1- Q)(l- E) ~
LJ T (k
1
J=
1
- )
")'"
"
dt
t k - J exp( -t) (tT + >')2
-0
max
-Ns;k~O
Ilg(tk)IIE o
for every k, 1 :::; k :::; Nand >., >. > O. Now let us estimate T 3 . By Lemma 8.1.3
and using estimate (8.11), we can obtain
k
IIT3 11E :::; >.1-0 ~

LJ T (k _1 j)!
Jk"
Tt+>'
t
exp(-t)IIAexp{ --2-A}IIE->E
00
-J
J=l
1
Tt + >.
Tt + >.
x IIA- [Aexp{ --2-A}Bj - BjAexp{ --2-A}]g(tj-N )IIEdt
k
1
:::; >.1-0 max IIA-1(ABj - B j A)A- 11IE->E ~ T (k
')'
15,k5,N
LJ
- ) .
1
J=
"
1
00
tk-J exp(-t)
e(l
+ Q)M
M H0 (1 + 2M )(1
1
1
1rQ2(1 - Q)
+ M 1)1- 0 2(2-0)0 M2 2 - dt
(tT + >.)2-0
O
404
(1 - a) E l:
k
1 a
::; A -
"1
J
00
J=
(k 1 ")'
- J .
"
-J exp( -t)
dt\)2
(
tT + 1\ -a
max
-N9~O
Ilg(tk) liE
0<
for every k, 1 ::; k ::; N and A, A > O. Using the triangle inequality, we obtain that
for every k, 1 ::; k ::; N and A, A> O.This shows that
for every k, 1 ::; k ::; N. In a similar manner with Theorem 8.2.1 applying mathematical induction, one can easily show that it is true for every k. This result
Now we consider the difference scheme (8.14). We have not been able to
obtain the same result for solution of the difference scheme (8.13) in the spaces
E a under the assumption for the difference scheme (8.13). Nevertheless, we have
asymptotic stability of the solution of the difference scheme under the supplementary restriction of the operator A. From the strong positivity of the operator A it
follows that there exists the bounded operator
whole space E.
(1 + TA + (T~)2 )-1, defined on the
Theorem 8.2.3. Let condition (8.5) be satisfied. Suppose that the following estimates hold:
(TA?
11(1 + TA)(1 + TA + -2-)-11I E.. . E <

TA
(TA)2
11(1 +"2)(1 + TA)(I + TA + -2-)-11IE.. . . E <

and
II B(t)A- 1 11 E ...... E -<
1,
1 + V2
2
(I-a)
t2:0.
M21-a(1 + V2) ,
(8.18)
(8.19)
Then for the solution of difference scheme (8.14) estimate (8.15) holds for any
k 2: 1.
Proof. Let us consider 1 ::; k ::; N. In this case
Uk = R kg(O)
TA )Bj(g(tj-N
+ ""'
~ R k-J+1 (1 + "2
+ g(tj-N-r))T,
(8.20)
j=l
where R = (I + TA + (T~)2)_1. Using formula (8.16) and estimates (8.2) and

assumptions (8.18) and (8.19), we obtain
405

00
~ I)! J
XA - a (k
1
k 1
t - exp( -t)IIAexp{ -(Tt + A)A}g(O)IIEdt
00
Jk
Tt+A
1
t -J exp(-t)IIAexp{ - - 2 - A}IIE-.eII B j A- IIE-.E
o
Tt + A 1
x IIAexp{ --2-A}"2(g(tj-N) + g(tj-N-d)IIEdt
x (k _ j)!
J
00
:S
AI-a
1
(k _ I)!
L
k
T (k _ j)!
k-l
JOO
()
dt
II (0) I
(1 + /2)
1- a
exp -t (tT + Ap-a 9 I Eo +
2
M2 1- a(1 +
/2)
M2 2 - a dt 1
t -J exp( -t) (tT + A)2-a 1\ "2(g(tj-N ) + g(t j -N-d)I\Eo
k
J=1
:S J max I\g(tk)IIE,
-N5,k5,O
where
J
00
I-a
J = A
1
(k _ I)!
k-l
o
k
\ l-a(
+A
) '"'
1
1 - a ~ T (k _ j)!
J-l
exp( -t) (tT
dt
+ Ap-a
00
k-j
()
dt
exp -t (tT + A)2-a'
Therefore
for every k, 1 :S k :S N and A, A> O. This shows that
for every k,l :S k :S N. In a similar manner as with Theorem 8.2.1 applying

mathematical induction, one can easily show that it is true for every k. This result
Now, we consider the difference scheme (8.14) when
(8.21 )
406
for some t 2 o. Suppose that operator B(t)A- 1 - A-I B(t), with domain which
coincides with D(A), admits a closure B(t)A-l - A-I B(t) bounded in E and that
the following estimate
IIA-l(AB(t) - B(t)A)A- 1 1IEr-;E
holds for every t 2 0 and some
E E
7[(1-0)2 0 2(1+0)-1
eM1+aM;+a
X
(1+V2)-1c:
(1+MIl1-a22+a-a2
::;
(8.22)
[0,1].
Theorem8.2.4. Let assumptions (8.18), (8.19) and (8.22) be satisfied. Suppose that
the estimate
(8.23)
holds for every t 2 o. Then for the solution of difference scheme (8.14) estimate
(8.15) holds for any k 2 1.
Proof. Let us consider 1 ::; k ::; N. Using formula (8.16), we obtain
~ I)!
ex>
XA
(k
t k- exp( -t)Aexp{ -(Tt + A)A}g(O)dt
ex>
(k - j)!
xAexp{ +A 1-
t k-J. exp( -t) exp{- (Tt+A)A }B

2
(Tt+A)A 1
2
}"2(g(tj-N) + g(tj-N-d)dt
k
(TA)2
TA
(TA)2
LT((I +TA)(I +TA+ _ _ )-I)k- j (I + -)(1 +TA)(I +TA+ _ _ )-1
j=1
ex>
x [exp{-
(Tt
(k - J)!
+ A)A
2
1 .
.
(Tt+A)A }A- 1
t k-Jexp(-t)Aexp{2
}ABj-BjAexp{= PI
(Tt
+ A)A
2
+ P2 + P3
}]"2(9(tj-N)+9(tj-N-l))dt
407
Using estimates (8.2), (8.22) and assumptions (8.18), (8.19), (8.23), we obtain
(TA)2
11P111E:S 11((1 +TA)(1 +TA+ -2-)-1)kII E>->E
1 a
XA - (k
11P211E
x (k _ j)!
00
~ 1)!
k 1
t - exp( -t)IIAexp( -(Tt + A)A)g(O)IIEdt
:s A1- a I>II(I + TA)(I + TA +

j=l
T
(
A)2
j
)-l)k- IIE>->E
J00
k(Tt+A)A-t J exp( -t)IIAexp{ 2

}IIE--+EIIA-1 BjllE--+ E
(Tt + A)A 1
2
}2(g(tj-N)
x IIAexp{ -
(1
+ g(tj-N-1))IIEdt
+ J2) A1 - a (1- a)(l - 10)

2
M2 1- a (1 + J2)
< -'--------'- ----'-----'--'-----=~

-
LT (k _ j)!
J=l
/00
t -J
M2 2 - a dt 1
exp( -t) (tT + A)2-a II "2 (g(tj-N ) + g(tj-N-l))IIE"
k
1
:S(l-a)(l-)LT(k
.)1
. 1
J=
J .
/00
0
k .
A1-adt
t-Jexp(-t)(
\)2
tT + A -a
max IIg(tk)IIE
-N~k~O
"
for every k,l :s k :s N and A, A > O. Now let us estimate P3. By Lemma 8.1.3
and using estimate (8.11), we obtain
11P311E
:s A1- a z= Til (I + TA)(I + TA + T

k
j=l
00
A)2
)-l)k- j II E>->E
1
/ k .
(Tt + A)A
X(k_j)!
t -Jexp(-t)IIAexp{2
}IIE--+E
o
408
x IIA- 1 [Aexp{ -
x exp{ -
(Tt
>.1-0(1 +}2)
2
+ >')A
-1
15,J5,N
~ >. 10
- (1 - a)f
~ >')A }Bj
}]2(9(tj-N)
max IIA
(Tt
- BjA
+ g(tj-N-l))II Edt
(AB j - BjA)A
Jk'
t
exp( -t) (
-1
L
k
IIE->E
. 1
J=
(k
1
-
')'
J .
00
. 1
(k 1 ')'
J .
J=
-J
tT
dt
+ >.)2
-0
max
-N5,k5,O
Ilg(tk)IIE"
for every k, 1 ~ k
N and A, >. > O. Using the triangle inequality, we obtain that
for every k, 1 ~ k
N and A, >. > O. This shows that
for every k, 1 ~ k ~ N. In a similar manner as with Theorem 8.2.1 applying

mathematical induction, one can easily show that it is true for every k. This result
In applications, we consider the initial-value problem on the range {O ~ t ~ 1,
x = (Xl,""X n) E IRn,r = (rl, ... ,rn)} for 2m-order multidimensional delay
differential equations of parabolic type,
au(t,x)
at
+ '"
u
Irl=2m
()
a1r1u(t,x)
aT x ax'I ... axrn

I
()(
' " a ()a1r1u(t-w,x)
-_ b
tu
r x a rl ... a rn
n
Irl=2m
u(t,x)
XI
= g(t,x), -w ~ t
+ uU,X
>: (t
)
>: ( ) )
+ uU
t - w, x ,0 < t < 00, x E IRn ,
(8.24)
O,x E IRn, I r
1= rl + ... + r n
where ar(x), b(t), g(t, x) are given sufficiently smooth functions and fJ > 0 is the
sufficiently large number. We will assume that the symbol [~ = (6, ... , ~n) E IRn],
AH~)
Ir l=2m
ar(x) (i~drl ... (i~nrn
409

(8.25)
o < Mll~12m ~ (-l)mAf(~) ~ M21~12m < 00

2. Problem (8.24) has a unique smooth
for ~ i= 0, where 1~ 1= JI61 2+ ... + I~n 1
solution. This allows us to reduce the initial-value problem (8.24) to the initialvalue problem (8.1) in a Banach space E with a strongly positive operator AX =
Ai + M defined by (8.25). Let us give a number of corollaries of the abstract
Theorems 8.1.1 and 8.2.1.
Theorem 8.2.5. Assume that
1-a
sup Ib(t)1 ~ M2 2 - Ct '

O~t<oo
Then the solutions of the initial-value problem (8.24) satisfy the stability estimates
where M 2 (a) does not depend on g(t,x). Here CE(IRn) is the space of functions
satisfying a Holder condition with the indicator E (0,1).
The proof of theorem 8.2.5 is based on the estimate
and on the abstract Theorem 8.1.1, on the strong positivity of the operator AX in
c13(IRn), and on equivalence of the norms in the spaces E Ct = ECt(C(IRn), AX) and
C 2mCt(lR n ) when 0 < a < 2~ (see Chapter 4).
Let us define the grid space IRh(O < h ~ h o) as the set of all points of the
Euclidean space Rn whose coordinates are given by
Xi
= Sih, Si = 0, 1, 2, ... , i = 1, ... , n.
To the differential operator AX =

operator A h by the formula
Ai +M defined by (8.25) we assign the difference
Ahuh(x) =
2:
b~(x)Dhuh(x)
2m~lrl~S
of an arbitrary high order of accuracy that approximates this multidimensional

elliptic operator A X which acts on functions u~ defined on the entire space lR h. We
410
shall assume that for

the inequalities
I~khl
::;
7r
the symbol
AX(~h,
h) of the operator A h satisfies
With the help of A h we arrive at the initial-value problem
(Uh(t, x))' + Ahuh(t, x) = b(t)Ahuh(t - w, x), t ::::: 0, x E lR. h,

uh(t, x) = gh(t, x) = g(t, x)( -w ::; t ::; 0), x E lR. h,
(8.26)
for an infinite system of ordinary differential equations. Now, we replace problem

(8.26) by the first-order of accuracy in t difference scheme
~(uZ(x) - uLl (x)) + AhuZ(x) = b(tk)AhuLN(x),

tk = kT, 1::; k,NT = W,X E IR. h,
{
uZ(x) = gh(ik, x), tk = kT, -N ::; k ::; O,x E IR. h.
(8.27)
Theorem 8.2.6. Assume that all the conditions of Theorem 8.2.5 are satisfied. Then
for the solution of difference scheme (8.27) the estimate
holds.
The proof of Theorem 8.2.6 is based on the estimate
and on the abstract Theorem 8.2.1, the positivity of the operator A h in C(IR. h),
and on the fact that the Eo: = Eo:(C(IR. h), Ah)-norms are equivalent to the norms
C 2mo:(IR.h) uniformly in h for 0 < Q < 2~ (see Chapter 4).

Chapters 1 and 2
The material in Chapter 1 and Chapter 2 was written on the basis of lecture notes
that were used in a graduate course at Fatih University.
Chapter 3
Chapter 3 is based on the paper [35q].
Chapter 4
The role played by coercive inequalities in the study of boundary-value problems
for elliptic and parabolic partial differential equations is well known (see, e.g., [4a,
6a-b, 107a, 108a, 174a]).
Coercivity inequalities for the solutions of an abstract differential equation of
parabolic type were established for the first time in [154a]. The Cauchy problem
for such an equation was considered in various spaces F(E) of functions defined
on the segment [0, T] with values in a Banach space E, and it was established
that a necessary condition for coercivity is the strong positivity of the operator
coefficients A(t). This condition is also sufficient in the spaces C~'CE). In the
case of the spaces Lp(E) with P E (1,00) the following extrapolation result was
established: for the Cauchy problem to be coercive in Lp(E) for all such p it suffices
that it be coercive for some Po E (1,00).
As noted above, in contrast with Lp(E), in the spaces C~'CE) an unconditional coercivity inequality holds, and in connection with this there arises the
question of whether the same is valid for the ordinary Holder space CCE). In [16u]
it was shown that the coercivity inequality in CCE) holds under the additional
assumption that A(O)vo - f (0) E Ec<,
The coercivity inequality was established in the two-parameter series of
spaces C~'''! (E) with 0 < {3 < 1,0 :::; "( :::; {3, under the additional assumption
that A(O)vo - f (0) E E{3_"!' Furthermore the spaces of smooth functions C~'''! (E),
in which coercive solvability has been established, depend on the parameters {3 and
"(. However, the constants in the coercive inequalities depend only on {3. Hence, we
can choose the parameter {3 freely, which increases the number of function spaces
412
in which problem (4.1) is well posed. Moreover the coercivity inequality was established in the three-parameter series of spaces Cg,'Y(Eo_l3) with 0 < a < 1,
o ~ 'Y ~ f3 ~ a, under the additional assumption that A(O)vo - f (0) E Eo-'Y'
Furthermore the spaces of smooth functions Cg,'Y(Eo_I3), in which coercive solvability has been established, depend on the parameters a, f3 and 'Y. However, the
parameters f3 and 'Y freely, which increases the number of function spaces in which
problem (4.1) is well posed [16h, 351, 94a].
Earlier an unconditional coercivity inequality was established in the spaces
W;(E) with 1 ~ p < 00, 0 < a < 1 [lIa, 89a], as well as in the spaces Lp(Eo,p)
with 0 < a < 1, 1 ~ p ~ 00 [154g, 69a-b, 35d, 135a].
Finally, let us mention the paper [64a], in which an unconditional coercivity
inequality in Lp(E) was established under the assumption that the imaginary
powers A i l3 are bounded (in some restrictions of the space E).
The application of this result to parabolic partial differential equations allowed one to obtain a series of new coercivity inequalities involving different norms
with respect to t and the space variables.
In this book we do not discuss results on the well-posedness of the nonlocal
boundary-value problems for parabolic and parabolic-elliptic equations, for which
the reader is referred to the papers [16t, 23a-b, 24a, 31b, 36c].
The material in Section 1 was written on the basis of the papers [106a, 154a,
154g, 154m, 16u, 16h, 351, 94a, 69a-b, 35d, 135a, 162~.
The construction of the discretization with respect to time for parabolic equations was initiated in Rothe's paper [133a]. When the study of the discretization
with respect to t for the abstract Cauchy problem is applied to parabolic partial
differential equations, it also covers the case of the full discretization with respect
to time and space variables.
To prove stability, in a number of works (see [la, 3a, 1I4a, 1I5a, 139a,
140a, 141a, 142a, 171a-b] and the references given therein) difference schemes
were treated as operator equations in a Hilbert space, and the investigation was
based on the symmetry properties of operator coefficients. This led to Lz-stability
estimates.
Of great interest is the study of stability in the C-norm. In [lOa, 12a, 76a,
144a-d, 163a-b, 177a] the corresponding estimates were obtained for the simplest
difference schemes approximating the Cauchy problem for parabolic equations and
systems with constant coefficients and one space variable. The proof is based on
estimates of parabolic difference Green function (fundamental solutions). In the
case of parabolic equations with variable coefficient and more than one space
variable, stability in the C-norm is known to hold only for difference schemes that
satisfy a maximum principle (see [1I5a, 141a]).
An important type of stability is the coercive stability (well-posedness) of
difference schemes. There are no coercive estimates in the C-norm, because they
do not hold for the corresponding differential equations. The coercive stability
413
of difference schemes of first order of accuracy was apparently studied for the
first time in [158a, 158b] for the first boundary-value problem for second-order
parabolic equations in 2. A more general sharp-angle inequality for parabolic
difference operators was established. Then followed the works [lOa, 65a, 87a, 99a,
158b, 158c], where coercive inequalities in the L 2 - and Lp-norms were given. In
[154d] difference schemes are treated as operator equations in Banach spaces. In
a Banach space there is no notion of symmetry and for this reason the study of
stability relies on the positivity property of the operator coefficients, expressed
in terms of properties of their resolvents. In this way in [154d] coercive estimates
were established in difference analogues of Holder spaces with weight with respect
to the time variable. Enlisting the theory of positive operators turned out to be
effective in the investigation of such difference schemes in difference analogues
of Bochner spaces (see [127a, 127b]). The coercivity inequality in the difference
analogue of the Bochner spaces L p ( T, E) holds for any P E (1,00) whenever it
holds for at least one Po E (1,00), i.e., here we have a result analogous to that in
the differential case [154a].
An unconditional coercivity inequality in the spaces Lq(T, E~,q) (0 < a <
1, 1 ~ q ~ (0) was obtained in [35d, 35e]. One is led to asking whether a coercivity inequality holds for the Rothe difference scheme in the difference analogue
of the usual Holder space COt (T, E). The answer turns out to be affirmative under
the additional restrictions A(O)uo - II E E~. This result is a particular case of the
assertion that the difference problem is coercive in the two-parameter family of Banach spaces Ct'l(E) with 0 < (3 < 1, 0 ~ 'Y ~ (3, under the additional assumption
that A(O)uo - II E E~_". Furthermore the spaces
(T, E), in which coercive
solvability has been established, depend on the parameters (3 and "f. However, the
constants in the coercive inequalities depend only on (3. Hence, we can choose the
parameter (3 freely, which increases the number of spaces in which a Rothe difference scheme for problem (4.1) is well posed. Moreover the coercivity inequality was
established in the three-parameter series of spaces Cg'''(T, E Ot -(3) with 0 < a < 1,
o ~ 'Y ~ (3 ~ a, under the additional assumption that A(O)uo - II E E~_".
Furthermore the spaces Cg'''(T, E~_(3)' in which coercive solvability has been established, depend on the parameters a, (3 and 'Y. However, the constants in the
coercive inequalities depend only on a. Hence, we can choose the parameters (3
and 'Y freely, which increases the number of spaces in which a Rothe difference
scheme for problem (4.1) is well posed [16h, 351, 94a].
cg,"
The utilization of Pade approximants for computational purposes has a long

history. The approximants R1,1(Z) and R2,2(Z) were first applied by Euler to calculate the scalar exponential. The fact that these approximants are attributed to
Pade is based on his 1892 dissertation, in which he studied and tabulated them.
In recent years we have seen a sharp rise in the interest in classical methods of rational approximation of analytical functions, primarily in Pade approximants and
their application to physics and engineering problems. For a detailed exposition
of these investigation we refer the reader to [43a].
414
Apparently the first study that uses rational approximation of exp{ - 7 A} in

the case of a positive (unbounded) self-adjoint operator A was carried out in the
paper [171a], which deals with approximation on the positive half-line.
That study was continued by a number of authors [58a, 139a, 171b]. In applications to problems of mathematical physics this resulted in stability and convergence estimates for difference schemes in 2'
The application of the approximants Rj,l (7 A) in the construction of difference
schemes in the case of an unbounded strongly positive operator A acting in an
arbitrary Banach space E has been systematically developed by several authors
(see, e.g., [7a-e, 16a-d, 16g, 16p, 16r, 32a, 35a-b, 35e-h, 351-m, 37a, 54a, 55a, 56a,
60a, 77a, 81a-b, 91a-c, 95a, 97a, 124b, 127c-e, 130a, 154d~f, 154i, 155a-b, 85a,
104a, 125a, 173a]). In [161, 54a, 95a] difference schemes were studied for a class of
operators that generate not only analytic, but also strongly continuous semigroups.
Most of the results of the stability and coercive stability in the Banach spaces of
the single-step difference schemes of a high order of accuracy generated by RJ,1 (7 A)
for parabolic equations were considered in the book [351]. The application of these
results to the initial-value problem for parabolic equations allowed one to obtain
a series of new coercivity inequalities in difference analogies of various norms with
respect to t and x for the solutions of difference schemes of a high order of accuracy
in time and space variables.
However, the theory of a high order of accuracy single-step difference schemes
for parabolic differential equations with variable operator coefficients have received
less attention than parabolic differential equations with coefficients independent
of time. Most of the results of the stability and coercive stability in the Banach
spaces of the single-step difference schemes of the first and second order of accuracy generated by the Pade approximants for first-order differential equations in
the case of an unbounded strongly positive operators A(t) acting in an arbitrary
Banach space E were obtained in the papers [16e, 35c, 126a, 154i-j, 161a]. The
application of Pade approximants for parabolic equations with variable operator
coefficients A(t) was continued by a number of authors [143a-b, 178a]. The difference schemes of the third, fourth and fifth order of accuracy were constructed and
investigated. In applications to problems of mathematical physics this resulted
in stability and convergence estimates for difference schemes in 2. In [42a, 49a,
lOOa, 143a] Runge-Kutta difference schemes were studied for parabolic equations
with variable operator coefficients A(t).
The application of the Pade approximants to the Cauchy problem for parabolic equations with variable operator coefficients A(t) acting in an arbitrary Banach space E relies on the approximate formulas for expressions
I1
tk
V(tk, tk-d and -:;
V(tk, s)f(s)ds.
tk-l
Unfortunately, in the general case of operators A(t) the explicit formula for
V(tk, tk-d does not exist. Therefore, in this case we cannot use approaches of the
415
construction of Pade difference schemes for parabolic equations with coefficients

independent of time.
The single-step difference schemes of a high order of accuracy generated by
an exact difference scheme for parabolic equations with variable operator coefficients A(t) acting in an arbitrary Banach space E were presented for the first time
in [35j]. They were based on a series of interesting properties concerning the semigroup exp{ -sA(t)} (s 2': 0) and the fundamental solution v(t, s) of (4.1). These
properties and series of interesting estimates for Rj,zCr A(tk)) and u(k, m) permit
us to study the stability and the coercive stability of presented difference schemes.
That study was continued in [160, 16q]. Furthermore the high order of accuracy
difference schemes generated by Taylor's decomposition on two points for the approximate solutions of the Cauchy problem for parabolic equations with variable
operator coefficients A(t) acting in an arbitrary Banach space E were presented.
The well-posedness of these difference schemes in various Banach spaces were studied. The stability and coercive stability estimates in Banach norms for solutions of
high order of accuracy difference schemes of mixed type boundary-value problems
for parabolic equations were obtained.
The study of high order of accuracy difference schemes in the case of strongly
positive operators A(t) allowed us to establish not only the stability, but also
the coercive stability in difference analogues of Banach spaces. The application
of these abstract results to mixed type boundary-value problems for parabolic
equations relies on the following facts: the strong positivity of an elliptic difference
operator A~x in the Banach space Eh, the well-posedness of the resolvent equation
of A~x in E h or in E~,p(Eh, A~X), 1 ::; P ::; 00, the structure of fractional spaces
E~,p(Eh, A~X).
The strong positivity of the wider class of differential operators has been
studied by a number of authors [4a, 5a, 89b, lOla, l03a, l54c, l54k, l59a-b,
l60a-b].
The strong positivity of elliptic difference operators in Hilbert norms was
studied in many works (see, e.g., [1l4a, 140a, 141a]). Such a study is based on the
self-adjointness and positive definiteness of the principal part of the operators A~x.
Considerably less studied is the strong positivity or simple positivity of difference operators in Banach norms. For second-order elliptic operators and the
simplest difference scheme, positivity in the Ch-norm follows from the maximum
principle (see [140a, 144a-b]).
The strong positivity in difference analogue of weighted Holder space for an
elliptic difference operator A~x of second order of accuracy that approximates an
elliptic operator without mixed derivatives was established in [154d].
The strong positivity of the simplest multidimensional second-order elliptic
difference operator in the Lp,h- and Ch-norms was established in [7a-c]. The most
thorough study of the strong positivity in the Ch-and C~ -norms of a wide class of
operators that approximate elliptic operators of arbitrary order in Rn was carried
out in [35e, l52a-d]. The situation is different for difference operators that ap-
416
proximate boundary-value problems for an elliptic equation. The positivity in the

Ch - and
-norms of difference operators of this type was studied in a number of
particular cases [7a, 28a, 28c, 61a-c].
For the well-posedness of the resolvent equation of A~x in E h or in
Ce
E~,p(Eh, A~X), 1 ~ P ~ 00
we refer the reader to Chapter 5 of the present work.

Finally, let us comment on the structure of fractional spaces. It is established that for any 0 < Q < 112m and 1 ~ P ~ 00 the norms in the spaces
E~,p(Lp(R'hJ, A~X) and WgmQ(R'h) are equivalent uniformly in h. This fact corresponds to the following equality, known in interpolation theory (see [166a]):
E~,p(Lp(Rn), A~X) = w;mQ(R n ), 0
< Q < 112m, 1 ~ P ~ 00,
which in turn follows from the equality D(At,X) = wgmQ(Rn) for a 2mth-order
elliptic operator At,x in Lp(Rn), 1 ~ P ~ 00, via the real interpolation method.
The alternative method of investigation adopted in [160, 351], based on estimates
of fundamental solutions of the resolvent equation for the operator A~x, allows us
to consider also the cases P = 1 and P = 00.
This book does not touch upon the results of [19a, 26a, 27a, 36a-b, 36d, 29a,
44a-c] on the theory of the single-step difference schemes for nonlocal boundaryvalue problems.
Section 2 is based on the papers [160, 35j]. The results of Theorems 4.25-4.32
and 4.40-4.45 and the difference schemes (4.114), (4.115), (4.16) appear here for
the first time.
Section 3 is based on the paper [160]. The results of Theorems 4.59-4.62
appear here for the first time.
Chapter 5
Coercivity inequalities for the solutions of an abstract differential equation of elliptic type were established for the first time in [154c]. The boundary-value problem
for such an equation was considered in various spaces F(E) of functions defined
on the segment [0, T] with values in a Banach space E, and it was established that
a necessary condition for coercivity is the positivity of the operator coefficient A.
This condition is also sufficient in the spaces C~TQ(E). In the case of the spaces
Lp(E) with p E (1,00) the following extrapolation result was established: for the
boundary-value problem to be coercive in Lp(E) for all such p it suffices that it
be coercive for some Po E (1, 00).
As noted above, in contrast to Lp(E), in the spaces Cg+Q(E) an unconditional
coercivity inequality holds, and in connection with this there arises the question
of whether the same is valid for the ordinary Holder space CQ(E). In [16j] it
was shown that the coercivity inequality in CQ(E) holds under the additional
assumptions that Avo - f (0) E E and AVT - f (T) E E
Q
417
The coercivity inequality was established in the two-parameter series of

spaces cgp(E) with 0 < (3 < 1,0 :S 'Y :S (3, under the additional assumptions
that Avo - f (0) E E/3-'Y and AVy - f (T) E E/3-T Furthermore the spaces
of smooth functions cgp (E), in which coercive solvability has been established,
depend on the parameters (3 and 'Y. However, the constants in the coercive inequalities depend only on (3. Hence, we can choose the parameter (3 freely, which
increases the number of function spaces in which problem (5.1) is well posed.
Moreover the coercivity inequality was established in the three-parameter series
of spaces Cgp(Ea - f3 ) with 0 < a < 1, 0 :S 'Y :S (3 :S a, under the additional
assumptions that Avo - f (0) E Ea-'Y and AVT - f (T) E Ea-'Y' Furthermore
the spaces of smooth functions cgp(Ea -/3), in which coercive solvability has been
established, depend on the parameters a, (3 and "(. However, the constants in the
coercive inequalities depend only on a. Hence, we can choose the parameters (3
and 'Y freely, which increases the number of function spaces in which problem (5.1)
is well posed [16j, 160].
Earlier an unconditional coercivity inequality was established in the spaces
Lp(Ea,p) with 0 < a < 1, 1 :S P :S 00 [89b].
The application of this result to elliptic partial differential equations allowed
one to obtain a series of new coercivity inequalities involving different norms with
respect to all variables.
In this book we do not discuss results on the well-posedness of the local and
nonlocal boundary-value problems for elliptic equations, for which the reader is
referred to the papers [9a, 16w, 21a-b, 28d, 34a, 84a-b, 86a, 145a-c, 148a].
The material in Section 1 was written on the basis of the papers [16j, 160,
154c, 154m, 89b].
The coercive stability of difference schemes of the second order of accuracy
was apparently studied for the first time in [120a] for the first boundary-value problem for second-order elliptic equations in L 2 . Then followed the works [12a, 169a,
170a, 66a-d, 179a, 134a, 158a-e, 78a-b, 117a, 118a-b], where coercive inequalities
in the L 2 - and Lp-norms were given. In [154d] difference schemes are treated as
operator equations in Banach spaces with the positive operator coefficient. In this
way coercive estimates were established in difference analogues of Holder spaces
with weight with respect to the one space variable. Enlisting the theory of positive
operators turned out to be effective in the investigation of such difference schemes
in difference analogues of Bochner spaces (see [127a, 127c]). The coercivity inequality in the difference analogues of the Bochner spaces Lp(r, E) holds for any
P E (1,00) whenever it holds for at least one Po E (1,00), i.e., here we have a
result analogous to that in the differential case [154c]. That study was continued
in [128a-c, 129a, 154h, 1541, 28b, 35j].
The applications of the Pade approximants in the theory of the high order
of accuracy two-step difference schemes for elliptic differential equations have received less attention than parabolic differential equations. The two-step difference
schemes of a high order of accuracy generated by an exact difference scheme for
418
elliptic equations with the operator coefficient A acting in an arbitrary Banach

space E were presented for the first time in [16j]. The series of interesting estimates for R;,l (T A ~) permit us to study the stability and the coercive stability of
presented difference schemes. That study was continued in [160, 16n]. Furthermore
the high order of accuracy difference schemes generated by Taylor's decomposition
on three points for the approximate solutions of the boundary-value problem for
elliptic equations with the operator coefficient A acting in an arbitrary Banach
space E were presented. The well-posedness of these difference schemes in various Banach spaces were studied. The stability and coercive stability estimates in
Banach norms for solutions of the high order of accuracy difference schemes of
mixed type boundary-value problems for elliptic equations were obtained [35k].
That study was continued in [16i, 160, 16s, 35n, 81c, 82a].
The study of the high order of accuracy difference schemes in the case of
positive operator A allowed us to establish not only the stability, but also the
coercive stability in difference analogues of Banach spaces. The application of these
abstract results to the mixed type boundary-value problems for elliptic equations
relies on the following facts: the positivity of an elliptic difference operator AI; in
the Banach space Eh' the well-posedness of the resolvent equation of AI; in E h or in
E~,p(Eh, (AI;)~), 1 S p S 00, the structure of fractional spaces E~,p(Eh, (AI;)~).
For the positivity of the wider class of differential and difference elliptic
operators we refer the reader to Chapter 4 of the present work.
Finally, let us comment on the structure of fractional spaces. It is established
that for any 0 < 0: < 1/2 and 1 S p Soothe norms in the spaces E~,p(Eh, (AI;) ~)
and E 2cx ,p(Eh, AI;) are equivalent uniformly in h. Therefore for the structure of
fractional spaces E~,p(Eh, (AI;)~) we refer the reader to Chapter 4 of the present
work.
Section 2 is based on the papers [16j, 160]. The results of Theorems 5.23-5.48,
5.32-5.46 and 5.51-5.57 appear here for the first time.
Section 3 is based on the papers [160, 16s, 35k, 35n]. The results of Theorems
5.67-5.68 and 5.70-5.73 appear here for the first time.
Chapter 6
The method of operators as a tool for investigation of the solution to hyperbolic
differential equations in Hilbert and Banach spaces, has been systematically developed by several authors (see, e.g., [2a, 75a, 85a, 104a, 125a, 154b, 156a, 173a]).
A large cycle of works on difference schemes for hyperbolic partial differential
equations (see, e.g., [16f, 124a, 124c, 124d] and the references given therein), in
which stability was established under the assumption that the magnitude of the
grid steps T and h with respect to the time and space variables are connected.
In abstract terms this means, in particular, that the condition TllAhl1 ---+ 0 when
T ---+ 0 is satisfied.
419
Of great interest is the study of absolute stable difference schemes of a high

order of accuracy for hyperbolic partial differential equations, in which stability
was established without any assumptions in respect of the grid steps T and h.
Such type stability inequalities for solutions of the first order of accuracy difference
scheme for differential equations of hyperbolic type were established for the first
time in [157a]. The new first and second order of accuracy difference schemes
approximately solving the abstract initial-value problem for hyperbolic equations
in a Hilbert space were presented in [30a, 350]. Applying the operator approach,
stability estimates for the solution of these difference schemes were obtained. In
[16f] the high order of accuracy two-step difference schemes generated by an exact
difference scheme for the numerical solutions of this problem were presented. The
convergence and stability estimates for the solutions of these difference schemes
were established.
This book does not touch upon the results of [17a, 25a-b, 30b, 30c, 31a, 38ad, 39a] on the theory of difference schemes for nonlocal problems for hyperbolic,
hyperbolic-parabolic, hyperbolic-elliptic equations. In this book we do not discuss
results on the difference schemes of the Goursat problem for hyperbolic equations
and the Cauchy problem for systems of differential equations of first order, for
which the reader is referred to the papers [14a, 15a, 33a].
Sections 1 and 2 are based on the paper [16f]. The results of Theorems 6.4-6.9
appear here for the first time.
The results of Section 3 appear here for the first time.
Chapter 1
The role played by asymptotic methods in the study of the solution of various
problems for elliptic, parabolic and hyperbolic partial differential equations with
a small parameter in higher-order derivatives is well known (see, e.g., [75a, lOla,
104a, IlIa]).
Of great interest is the study of stability uniform difference schemes for partial
differential equations with a small parameter in higher-order derivatives (see, e.g.,
[16v, 40a, 52a, 53a, 67a, 73a-b, 96a, 98a-c, 149a, 150a, 151a, 164a]).
In Chapters 4, 5 and 6 the stability of a wide class of difference schemes
for partial differential equations was investigated. However, it can be shown that
these difference schemes for partial differential equations with a small parameter
in higher-order derivatives are not uniform in E: convergence. This is a very disappointing result and it raises the question whether there are any difference schemes
which are uniform in E: convergence.
The high order of accuracy single-step uniform difference schemes of the
approximate solutions of the Cauchy problem for the abstract parabolic differential
equations in an arbitrary Banach space E with the strongly positive operator
A and an arbitrary E: positive parameter multiplying the derivative term were
investigated in [16k] and generated by the exact single-step difference scheme,
420
which is presented in Chapter 4. The well-posedness of these difference schemes

in various Banach spaces were studied. That study was continued in [160].
The two-step uniform difference schemes of a high order of accuracy for the
approximate solutions of the boundary-value problem for abstract elliptic differential equations in an arbitrary Banach space E with the positive operator A and an
arbitrary E: positive parameter multiplying the derivative term were investigated
in [16m] and generated by the exact two-step difference scheme, which is presented
in Chapter 5. The well-posedness of these difference schemes in various Banach
spaces was established. That study was continued in [160].
In [22a-b] the high order of accuracy two-step uniform difference schemes
generated by an exact difference scheme of the approximate solutions for differential equations of hyperbolic type with an arbitrary parameter E: on the highest
derivative were presented. The stability estimates of the solutions of these difference schemes under the assumptions
IIc(t)IIE->E :::; M, IIA~s(t)IIE->E:::; M, 0:::; t < 00

in an arbitrary Banach space E were established. These estimates for the wider
class of differential operators in Banach and Hilbert spaces have been studied by
a number of authors (see, for example, [75a, 102a, 125a] and the references given
therein).
The material in Section 1 is written on the basis of the papers [16k, 160].
The results of Theorems 7.8-7.14 appear here for the first time.
The material in Section 2 is written on the basis of the papers [16m, 160].
The results of Theorems 7.22-7.27 appear here for the first time
The material in Section 3 is written on the basis of the papers [22a-b].
Chapter 8
The operator approach for theory and numerical solutions of the initial-value problem for evolution differential equations has been studied extensively by many researches (see [8a, 41a, 50a, 51a, 57a, 68a, 70a, 71a, 72a-b, 74a, 79a,80a, 81c, 82b,
83a, 112a, 119a-c, 121a, 122a, 123a, 124e, 131a-b, 132a, 146a, 147a-b, 167a, 168a,
169a, 175a, 176a-b] and the references therein).
Methods for numerical solutions of the delay ordinary differential equations
have been studied extensively in a large cycle of works (cf., e.g., [13a, 18a, 18b, 20a,
46a, 47a, 59a, 165a, 181a] and the references therein) and developed over the last
two decades. In this literature, stability of solutions of the difference schemes of
the initial-value problems for delay ordinary differential equations was considered
under the necessary condition of stability of the differential problem. However, the
theory and numerical solutions of delay partial differential equations have received
less attention than delay ordinary differential equations.
The role played by the operator approach in the study of the solution of
various problems for delay partial differential equations is well known (see, e.g.,
421
[35p, 62a]). In [35p] the initial-value problem for delay differential equations of
parabolic type was considered in the spaces C(Ea ) of functions defined on the
segment [0,00] with values in a Banach space Ea and the stability inequalities
were established under the stronger assumption than the necessary condition of
the stability of the differential problem. The stability estimates for the solutions of
the first and second order of accuracy difference schemes for approximately solving
this initial-value problem for delay differential equations of parabolic type were
presented. Stability estimates in Holder norms for solutions of the initial-value
problem of the delay differential and difference equations of parabolic type were
obtained.
The material in Sections 1 and 2 is written on the basis of the paper [35p].
Bibliography
[la]
Abrashin, V.N., A stable difference scheme for quasilinear equations of mathematical physics, Differensialnyye Uravneniya 18 (1982), no. 11, 1967-1971. (Russian).
[RZh. Mat. 1983:3 B1373]
[2a]
Abrashin, V.N. and Bobkov V.V., A difference scheme of integral relations method
for a hyperbolic equation, Izv. AN BSSR, Ser. Phys. Mat. Nauk No.2 (1968), 5765. (Russian). [RZh. Mat. 1969:2 B802]
[3a]
Abrashin, V.N. and Shkel B.A., A symmetrical difference scheme of the numerical
solution of the boundary-value problems for nonlinear parabolic equations, Vest.
AN BSSR, Ser. Phys. Mat. Nauk No.2 (1987), 7-14. (Russian). [RZh. Mat. 1987:9
B1306]
[4a]
Agmon, S., Lectures on Elliptic Boundary- Value Problems, D. Van Nostrand,
Princeton:New Jersey, 1965. [MR 31 #2504]
[5a]
Agmon, S., Nirenberg, L., Properties of solutions of ordinary differential equations
in Banach spaces, Comm. Pure Appl. Math. 20 (1963), 121-239. [MR 21 #5142]
[6a]
Agmon, S., Douglis, A. and Nirenberg, L., Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions
I, Comm. Pure Appl. Math. 12(1959), 623-727. [MR 23 #A261O]
[6b]
Agmon, S., Douglis, A. and Nirenberg, L., Estimates near the boundary for solutions of elliptic partial differential equations satisfying general boundary conditions
II, Comm. Pure App!. Math. 17(1964), 35-92. [MR 28 #5259]
[7a]
Alibekov, Kh.A., Sobolevskii, P.E., Stability of difference schemes for parabolic
equations, Dokl. Acad. Nauk SSSR 232 (1977), no. 4, 737-740. (Russian). [MR
55 #13023]
[7b]
Alibekov, Kh.A. and Sobolevskii, P.E., Stability and convergence of difference
schemes of the high order for parabolic differential equations, Ukrain. Mat. Zh. 31
(1979), no. 6, 627-634. (Russian). [MR 8li:65040]
[7c]
Alibekov, Kh.A. and Sobolevskii, P.E., Stability and convergence of difference
schemes of the high order for parabolic partial differential equations, Ukrain. Mat.
Zh. 32 (1980), no. 3, 291-300. (Russian). [MR 81j:65100]
[8a]
Amann, H., Linear and quasilinear parabolic problems, Birkhauser Verlag, Basel,
I (1995). Abstract linear theory. [MR 96g:34088]
[9a]
Amanov, K. Schauder type estimates for boundary-value problem for multidimensional elliptic equations, in: Some Problems of Applied Mathematics, Fatih University, Istanbul (2000), 7-10.
[lOa] Andreev, Yu.N., On uniform convergence of certain difference schemes, Zh. Vychis!. Mat. i Mat. Fiz. 6 (1966), no. 2, 238-250. (Russian). [MR 34 #967]
424
Bibliography
[lla]
Anosov, V.P. and Sobolevskii, P.E., The coercive solvability of parabolic equations,
Mat. Zametki 11 (1972), no. 2, 409--419. (Russian). [MR 46 #558J
[12a]
Aronson, D.G., On the correctness of partial differential operators and the von
Neumann condition for stability difference operators, Proceeding Amer. Math.
Soc. 14 (1963), 948-955. [MR 27 #6407]
[13a]
AI-Mutib, A.N., Stability properties of numerical methods for solving delay differential equations, J. Comput. and Appl. Math. 10 (1984), 71-79. [MR85m:65075]
[14a]
Ashyraliyev, M., Integral inequalities with four variable limits, in: Modeling the
Processes in Exploration of Gas Deposits and Applied Problems of Theoretical
Gas Hydrodynamics, Ilim, Ashgabat (1998), 170-184 (Russian).
[14b]
Ashyraliyev, M., On stability estimates of Goursat differential and difference problems for hyperbolic equations, in: Some Problems of Applied Mathematics, Fatih
University, Istanbul (2000), 103-112.
[15a]
Ashyraliyev, M. and Ataev, A., On stability estimates of Goursat problem for hyperbolic equations, 8 years Turkmenistan's Independence, Ilim, Ashgabat (1999),
201-206.
[16a]
Ashyralyev, A., On the coercive stability of a Crank-Nicolson difference scheme

for the regularly singular parabolic equations in Holder spaces with weight, in:
Investigations on the Theory of Differential Equations, Ashgabat (1983), 52-64.
(Russian). [RZh. Mat. 1983:12 B1389]
[16bJ
Ashyralyev, A., Some difference schemes for parabolic equations with nonsmooth
initial data, Izv. Akad. Nauk Turkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol. Nauk,
no. 6 (1983), 70-72. (Russian). [MR 86c:65093]
[16c]
Ashyralyev, A., An estimate of the error of solution of purely implicit difference

schemes for the parabolic equations with nonsmooth data, in: Applied Mathematics
and Computer Software, Moscow (1985), 26-28. (Russian). [MZh. Mat. 1986:5
B1509]
[16d]
Ashyralyev, A., A purely implicit difference scheme of second order of approximation for parabolic equations, Izv. Akad. Nauk Turkmen. SSR Ser. Fiz.-Tekhn.
Khim. Geo!. Nauk, no. 4(1987), 3-13. (Russian). [MR 89g:65112]
[16e]
Ashyralyev, A., Coercive stability of a difference scheme of second-order of approximation for differential equations with a time-dependent operator in a Banach
space, in: Measure and Integral, Kuybishev, KSU (1988), 27-36. (Russian). [MR
92d:28002]
[16f]
Ashyralyev, A., On difference schemes of the high order of accuracy for secondorder evolution equations, in: Functional Differential Equations, Perm (1989), 145150. (Russian). [MR 91e:340070]
[16g]
Ashyralyev, A., Estimation of the convergence of modified Crank-Nicolson difference schemes for parabolic equations with nonsmooth imput data, Izv. Akad. Nauk
Turkmen. SSR Ser. Fiz.-Tekhn. Khim. Goo!. Nauk, no. 1 (1989), 3-8. (Russian).
[MR 90:65159]
[16h]
Ashyralyev, A., Coercive solvability of parabolic equations in spaces of smooth

functions, Izv. Akad. Nauk Turkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol. Nauk,
no. 3 (1989), 3-13. (Russian). [MR 90i:35150]
Bibliography
425
[16i]
Ashyralyev, A., On difference schemes of higher order of accuracy for elliptic

equations, in: Applications of the Methods of Functional Analysis to the Some
Nonclassical Equations of Mathematical Physics, Novosibirsk (1989), 3-14. (Russian). [RZh. Mat. 1990:10 G122]
[16j]
Ashyralyev, A., Well-posedness of elliptic equations in a space of smooth functions,

in: Boundary-Value Problems for Nonclassic Equations of Mathematical Physics,
Sibirsk, Otdel, SSSR, Novosibirsk (1989), Vol. 2, 82-86. (Russian).
[16k]
Ashyralyev, A., On uniform difference schemes of the high order of approximation

for evolution equations with a small parameter, Numerical Functional Analysis and
Optimization (1989) 10, no. 5-6, 593-406. [MR 91k:65140]
[161]
Ashyralyev, A., On difference schemes of higher order of accuracy for semilinear evolution differential equations in a Banach space, in: Computer Science and
System Analysis, Ashgabat (1990), 66-75. (Russian).
[16m] Ashyralyev, A., On uniform difference schemes of higher order of approximation

for elliptic equations with a small parameter, Applicable Analysis 36 (1990), 211220. [MR 91d:65105]
[16n]
Ashyralyev, A., Estimations of the convergence of modified difference schemes of

the Pade's class for elliptic equations, in: Investigation of Theory and Approximation Methods for Differential Equations, Ashgabat (1991), 71-79. (Russian).
[160]
Ashyralyev, A., Method of positive operators of investigations of the high order of

accuracy difference schemes for parabolic and elliptic equations, Doctor Sciences
Thesis, Kiev, 1991. (Russian).
[16p]
Ashyralyev, A., Correct solvability of the Pade difference schemes for parabolic
equations in Holder spaces, Ukrain. Mat. Zh. 44 (1992), no. 11, 1466-1476. (Russian). [MR 94c:65103]
[16q]
Ashyralyev, A., Single-step difference schemes generated by Taylor's decomposition in two points, in: Proceedings of the Scientific-Practical Conference on Differential Equations and Their Applications. Ashgabat, 1993, 79-81. (Russian).
[16r]
Ashyralyev, A., Correct solvability of the Rothe difference scheme in the difference
analogy of spaces of smooth functions, Proceeding of the AS of the Turkmenistan,
Ashgabat (1994),112-118. (Russian).
[16s]
Ashyralyev, A., Well-posed solvability of the boundary-value problem for difference

equations of the elliptic type, Nonlinear Analysis. Theory. Methods, and Applications (1995) 24, no. 2, 251-256. [MR 96b:65056]
[16t]
Ashyralyev, A., The stability of difference schemes for partial differential equations of mixed types, in: Proceedings of the Second International Symposium on
Mathematical and Computational Applications, Baku (1999), 314-316.
[16u]
Ashyralyev, A.,Coercive Solvability of Parablic Equations in Holder Spaces, in:

Abstracts of 12-th National Sympozium, Malatya, Turkey, 1999, Ip.
[16v]
Ashyralyev, A., On uniform difference schemes of the nonlocal perturbation

boundary-value problem for differential equations mixed types, in: Proceeding of
the International Conference "Modern Mathematical Problems", Astana, (2002),
63-66.
426
Bibliography
[16w] Ashyralyev, A., On well-posedness of the nonlocal boundary-value problem for elliptic equations, Numerical Functional Analysis and Optimization (2003) 24, no.
1-2,1-15.
[17a]
Ashyralyev, A. and Aggez, N., A note on the difference schemes of the nonlocal
boundary-value problems for hyperbolic equations, 2003, 22p., submitted.
[ISa]
Ashyralyev, A. and Akca, H., On difference schemes for semilinear delay differential equations with constant delay, in: Proceeding of the Conference TSU "Actual
Problems of Applied Mathematics, Physics and Engineering", Ashgabad (1999),
IS-21.
[ISb]
Ashyralyev, A. and Akca, H., Stability properties of difference schemes for neutral
delay differential equations, Nonlinear Analysis. Theory. Methods and Applications, 44 (2001), 443-452.
[19a]
Ashyralyev, A., Akca H. and Bizevski, L., On a semilinear evolution nonlocal

Cauchy problem, in: Some Problems of Applied Mathematics, Fatih University,
Istanbul (2000), 29-44
[20a]
Ashyralyev, A., Akca H. and Guray, U., Second order of accuracy difference
scheme for approximate solutions of delay differential equations, Functional Differential Equations 6 (1999), no. 3-4, 223-231.
[21a]
Ashyralyev, A. and Amanov, K., On the coercive estimates in Holder norms, Izv.
Akad. Nauk Thrkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol. Nauk, no. 1 (1996),3-10.
(Russian).
[21b]
Ashyralyev, A. and Amanov, K., On the coercive solvability of second boundaryvalue problem for elliptic equations, TSAU, Preprint: Ashgabat (1995), 1-50. (Russian).
[22aJ
Ashyralyev, A. and Fattorini, H., On difference schemes of the high order of accuracy for singular perturbation elliptic equations, in: Investigation of Theory and
Approximation Methods for Differential Equations, Ashgabat (1991), So-S3. (Russian).
[22b]
Ashyralyev, A. and Fattorini, H., On uniform difference schemes for second order
singular perturbation problems in Banach spaces, SIAM J. Math. Analysis, 23
(1992), no. 1, 29-54. [MR 92K:65143]
[23a]
Ashyralyev, A. and Hanalyev, A., Coercive stability of nonlocal boundary-value

problem for parabolic equations in spaces of smooth functions, Izv. Akad. Nauk
Thrkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol. Nauk, no. 3 (1996), 3-10. (Thrkmen).
[23bJ
Ashyralyev, A. and Hanalyev, A., Coercive estimate in Holder norms for parabolic
equations with a dependent operator, in: Modeling the Processes in Exploration
of Gas Deposits and Applied Problems of Theoretical Gas Hydrodynamics, Him,
Ashgabat (199S), 154-162. (Russian).
[24a]
Ashyralyev, A., Hanalyev, A. and Sobolevskii, P.E., Coercive solvability of nonlocal

boundary-value problem for parabolic equations, Abstract and Applied Analysis 6
(2001), no. 1, 53-61.
[25a]
Ashyralyev, A. and Judakova G., On equations of the mixed type in Hilbert space,
in: Proceeding of the Conference TSU "Actual Problems of Applied Mathematics,
Physics and Engineering", Ashgabad (1999), 21-24. (Russian).
Bibliography
427
[25b]
Ashyralyev, A. and Judakova G., A note on the nonlocal boundary-value problem

for hyperbolic-elliptic equations in a Hilbert space, in: Proceeding of the International Conference "Modern Mathematical Problems", Astana (2002), 66-70.
[26a]
Ashyralyev, A. and Karatay, I., On the second order of accuracy difference schemes
of the nonlocal boundary-value problem for parabolic equations, Functional Differential Equations 10 (2003), no. 1-2, 45-63.
[27a]
Ashyralyev, A., Karatay, I. and Sobolevskii, P.E., On well-posedness of the nonlocal boundary-value problem for parabolic difference equations, Discrete Dynamics
in Nature and Society, 2004, 21p., accepted.
[28a]
Ashyralyev, A. and Kendirli, B., Positivity in Chof one dimensional difference

operators with nonlocal boundary conditions, in: Some Problems of Applied Mathematics, Fatih University, Istanbul (2000), 45-60.
Ashyralyev, A. and Kendirli, B., Well posedness of the nonlocal boundary-value
problem for difference elliptic equations, in: Proceeding of the International Mathematical Conference "Kazakhstan - 2000", Almaty (2001), 14-19.
[28b]
[28c]
Ashyralyev, A. and Kendirli, B., Positivity in Holder norms of one dimensional

difference operators with nonlocal boundary conditions, in: Application of Mathematics in Engineering and Economics-26, Heron Press and Technical University
of Sofia (2001), 134-137.
[28d]
Ashyralyev, A. and Kendirli, B., Well posedness of the nonlocal boundary-value

problem for elliptic equations, Functional Differential Equations 9 (2002), no. 1-2,
33-55.
[29a]
Ashyralyev, A. and Micheletzky, G., The approximation of the solutions of stochastic differential equations in a Hilbert space by the difference schemes, in: Proceeding of the Scientific-Practical Conference of Differential Equations and Their
Applications. Ashgabat (1993), 85-95.
[30a]
Ashyralyev, A. and Muradov I., On one difference scheme of a second order of accuracy for hyperbolic equations, in: Proceeding of the IMM of AS of Turkmenistan,
Ashgabat (1995), no 3, 58-63. (Russian).
Ashyralyev, A. and Muradov, I., On stability estimation of difference scheme of a
first order of accuracy for hyperbolic-parabolic equations, Izv. Akad. Nauk Turkmenistan Ser. Fiz.-Tekhn. Khim. Geol. Nauk, no. 1 (1996),35-39. (Russian).
Ashyralyev, A. and Muradov I., On difference schemes second order of accuracy
for hyperbolic-parabolic equations, in: Modeling the Processes in Exploration of
Gas Deposits and Applied Problems of Theoretical Gas Hydrodynamics, Ilim,
Ashgabat (1998), 127-138. (Russian).
Ashyralyev, A. and Orazov, M.B., Theory of operators and the methods of solutions of boundary-value problems for equations of mathematical physics, 8 years
Turkmenistan's. Independence, Ilim, Ashgabat (1999), 222-228. (Russian).
[30b]
[30c]
[31a]
[31b]
Ashyralyev, A. and Orazov, M.B., The theory of operators and the stability of
difference schemes for partial differential equations mixed types, Firat University,
Fen ve Muh. Bilimleri Dergisi 11 (1999), no. 3, 249-252.
[32a]
Ashyralyev, A., Piskarev, S. and Weis, L., On well-posedness of the difference

schemes for abstract parabolic equations in U([O, T]; E) spaces, Numerical Functional Analysis and Optimization, 23 (2002), no. 7-8, 669-693.
428
Bibliography
[33a]
Ashyralyev, A. and Prenov, R., On the stability of difference schemes for system
differential equations of first order, in: Modeling the Processes in Exploration of
Gas Deposits and Applied Problems of Theoretical Gas Hydrodynamics, Ilim,
[34a]
Ashyralyev, A. and Samovik, I.E., Coercive solvability of one nonlocal boundaryvalue problem for elliptic equations, in: Proceeding of the IMM of Turkmenistan,
Ashgabat (1995), no. 3, 67-75. (Russian).
[35a]
Ashyralyev, A. and Sobolevskii, P.E., Coercive stability of a Crank-Nicolson difference scheme in Co spaces, in: Approximate Methods for Investigations of Differential Equations and their Applications, Kuybishev (1982), 16-24. (Russian).
[MR 86e:65117]
[35b]
Ashyralyev, A. and Sobolevskii, P.E., Correct solvability of a Crank-Nicolson difference scheme for parabolic equations, Izv. Akad. Nauk Turkmen. SSR Ser. Fiz.Tekhn. Khim. Geol. Nauk No.6 (1981), 10-16. (Russian). [MR 83g:34064]
[35c]
Ashyralyev, A. and Sobolevskii, P.E., The Crank-Nicolson difference scheme for

differential equations in a Banach space with a time-dependent operator, Izv. Akad.
Nauk Turkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol. Nauk No.3 (1982), 3-9. (Russian). [MR 84c:65080]
[35d]
Ashyralyev, A. and Sobolevskii, P.E., On the coercive stability of the difference

schemes for abstract parabolic equations with variable operator coefficients in interpolation spaces, in: Investigations on the Theory of Differential Equations, Ashgabat (1983), 65-77. (Russian). [RZh. Mat. 12 B1415]
[35e]
Ashyralyev, A. and Sobolevskii, P.E., The theory of interpolation of linear operators and the stability of difference schemes, Dokl. Akad. Nauk SSSR 275 (1984),
no. 6, 1289-1291. (Russian). [MR 85m:65046]
Ashyralyev, A. and Sobolevskii, P.E., Stability of difference schemes for parabolic
equations in interpolation spaces, in: Applied Methods of Functional Analysis,
Voronezh (1985), 9-17. (Russian). [RZh. Mat. 1986:1 B1416]
Ashyralyev, A. and Sobolevskii, P.E., Coercive stability of difference schemes of
first and second order of approximation for parabolic equations with coefficients
independent of time, Izv. Akad. Nauk Turkmen. SSR Ser. Fiz.-Tekhn. Khim. Geol.
Nauk No.6 (1985), 3-11. (Russian). [RZh. Mat. 1986:8 B1305]
[35f]
[35g]
[35h]
Ashyralyev, A. and Sobolevskii, P.E., Stability in a Holder norms of difference

schemes for parabolic equations, in: Numerical Methods for the Solving of the
Transport Equations, Tartu (1986), 20-23. (Russian).
[35i]
Ashyralyev, A. and Sobolevskii, P.E., Coercive stability of a multidimensional

difference elliptic equations of 2m-th order with variable coefficients, in: Investigations in the Theory of Differential Equations, Ashgabat (1987), 31-43. (Russian).
[see MR 90d:00022]
[35j]
Ashyralyev, A. and Sobolevskii, P.E., Difference schemes of the high order of

accuracy for parabolic equations with variable coefficients, Dokl. Akad. Nauk
Ukrainian SSR, Ser. A Fiz.-Mat. and Tech. Sciences No.6 (1988), 3-7. (Russian).
[MR 90a:65190]
Ashyralyev, A. and Sobolevskii, P.E., On a class two-step difference schemes of
higher order of the accuracy for elliptic equations in a Hilbert space, in: Numerical
[35k]
Bibliography
429
Methods for the Solutions of the Transport Equations, Akad. Nauk Estonsk. SSR,
Tartu (1990), 18-22. (Russian).
[351]
Ashyralyev, A. and Sobolevskii, P.E., Well-Posedness of Parabolic Difference

Equations, Birkhauser Verlag: Basel, Boston, Berlin, 1994. [MR 95J:65094]
[35m] Ashyralyev, A. and Sobolevskii, P.E., Well-posed solvability of the Cauchy problem for the difference equations of the parabolic type, Nonlinear Analysis. Theory.
Methods and Applications 24 (1995), no. 2, 257-264. [MR 96b:65057J
[35nJ
Ashyralyev, A. and Sobolevskii, P.E., Theory of positive operators and stability

of the difference schemes, in: Some Problems of Applied Mathematics, Fatih University, Istanbul (2000), 61-69.
[350]
Ashyralyev, A. and Sobolevskii, P.E., A note on the difference schemes for hyperbolic equations, Abstract and Applied Analysis 6 (2001), no. 2, 63-70.
[35p]
Ashyralyev, A. and Sobolevskii, P.E., On the stability of the delay differential and
difference equations, Abstract and Applied Analysis 6 (2001), no. 5, 267-297.
[35qJ
Ashyralyev, A. and Sobolevskii, P.E., On the two new approaches of the high
order of accuracy difference schemes for the second order differential equations,
Functional Differential Equations 10 (2003), no. 3-4, 333-405.
[36a]
Ashyralyev, A. and Soltanov H., On the stability of difference scheme for parabolicelliptic equations in a Hilbert space, in: Proceeding of the IMM of Thrkmenistan,
[36b]
Ashyralyev, A. and Soltanov H., On coercive stability of difference scheme for

parabolic-elliptic equations, in: Proceeding of the IMM of Thrkmenistan, Ashgabat
No.3 (1995), 63-66. (Russian).
[36c]
Ashyralyev, A. and Soltanov H., On elliptic-parabolic equations in Hilbert space,

in: Proceeding of the IMM and CS of Thrkmenistan, Ashgabat No.4 (1995),
101-104. (Russian).
[36d]
Ashyralyev, A. and Soltanov H., On one difference schemes for an abstract nonlocal problem generated by the investigation of the motion of gas on the homogeneous
space, in: Modeling the Processes in Exploration of Gas Deposits and Applied
Problems of Theoretical Gas Hydrodynamics, Ilim, Ashgabat (1998), 147-154.
(Russian).
[37a]
Ashyralyev, A. and Yakubov, A., Structures of fractional spaces generating by

the transport operator, in: Modeling the Processes in Exploration of Gas Deposits
and Applied Problems of Theoretical Gas Hydrodynamics, Him, Ashgabat (1998),
162-172 (Russian).
[38aJ
Ashyralyev, A. and Yurtsever, A., On difference schemes for hyperbolic-parabolic

equations, Functional Differential Equations 7 (2000), no. 3-4, 189-203.
[38b]
Ashyralyev, A. and Yurtsever, A., On a nonlocal boundary-value problem for

hyperbolic-parabolic equations, in: Application of Mathematics in Engineering and
Economics 26, Heron Press and Technical University of Sofia (2001), 79-85.
[38cJ
Ashyralyev, A. and Yurtsever, A., On a nonlocal boundary-value problem for semilinear hyperbolic-parabolic equations, Nonlinear Analysis. Theory. Methods and
Applications 47 (2001), 3585-3592.
430
Bibliography
[38d]
Ashyralyev, A. and Yurtsever, A., A note on the second-order difference schemes

for hyperbolic-parabolic equations, Applied mathematics and computation, 2003,
23p., accepted.
[39a]
Ashyralyev, A. and Yurtsever, A., and Judakova, G., On hyperbolic-elliptic equations in a Hilbert space, in: Some Problems of Applied Mathematics, Fatih University, Istanbul (2000), 70-79.
[40a]
Bahvalov, H.S., An optimization method of solution of the boundary-value problems with boundary layer, Jh. Vyc. Mat. i Mat. Fiz. 9 (1969), no. 4, 841-859.
(Russian). [RZh. Mat. 12b 834]
[41a]
Baillon, J.B., Charactere borne de certains generateurs de semi-groupes lineaires

dans les espaces de Banach, C. R. Acad. Sci. Paris Ser. A-B 290, No. 16 (1980),
A757-A760. [MR 82a:47038]
[42a]
Bakaev, N.Y., On variable Runge-Kutta approximations of a Cauchy problem for

the evolution equation, BIT 38 (1998), no. 3, 462-485. [MR 99i:65069]
[43a]
Baker, G.A. Jr. and Graves-Morris, P.R., Pade Approximants. Part 1. Basic Theory, and Part II. Extensions and Applications, Addison-Wesley, Reading, Mass.,
1981. [MR 83a:41009a,b]
[44a]
Batyrov, S.S., Muradov A.N., Ashyralyev, A., Ashyralyev Ch., Lurieva 1.1. and
Annamuhammedov D., The numerical model of stabilization in the plast of filtration of oil, in: Modeling the Processes in Exploration of Gas Deposits and Applied
Problems of Theoretical Gas Hydrodynamics, Ilim, Ashgabat(1998), 32-40. (Russian).
[44b]
Batyrov, S.S., Muradov A.N., Ashyralyev, A., Ashyralyev Ch., Lurieva 1.1. and
Annamuhammedov D., An investigation algorithms solutions of backward problem of filtration in the non-homogeneous collectors, in: Modeling the Processes in
Exploration of Gas Deposits and Applied Problems of Theoretical Gas Hydrodynamics, Ilim, Ashgabat (1998), 40-45. (Russian).
[44c]
Batyrov S.S., Muradov A.N., Ashyralyev, A., Ashyralyev Ch., Lurieva 1.1. Annamuhammedov D., The model of problem of filtration oil in the lower permeable
stratum and its solving using the difference schemes, 8 years Turkmenistan's Independence, Ilim, Ashgabat (1999), 201-206.
Bazarov, D. and Soltanov H., Some Local and Nonlocal Boundary- Value Problems for Equations of Mixed and Mixed-Composite Types, Ilim: Ashgabat, 1995.
(Russian).
[45a]
[46aJ
Bellen, A., One-step collocation for delay differential equations, J. Comput. and
Appl. Math. 10 (1984), 275-283. [MR 85m:65061]
[47a]
Bellen, A., Jackiewicz, Z. and Zennaro M., Stability analysis of one-step methods
for neutral delay-differential equations, Numer. Math. 52 (1988), 605-619. [MR
89h:65134]
[48a]
Benedek, A., Calderon, A.P. and Panzone, R., Convolution operators on Banach
space-valued functions, Proc. Nat. Acad. Sci. U. S. A. 48 (1962), 356-365. [MR 24
#A3479]
Beyn Wolf-Jiirgen, and Garaylarson, B.M., Estimates of variable stepsize RungeKutta methods for sectorial evolution equations with nonsmooth data, Applied
Numerical Mathematics 41 (2002), no. 3, 369-400.
[49a]
Bibliography
431
[50a]
Blunck, S., Maximal regularity of discrete and continuous time evolution equations,
Studia Math. 146 (2001), no. 2, 157-176.
[51a]
Blunck, S. and Weis, L., Operator theoretic properties of semigroups in terms of

their generators, Studia Math. 146 (2001), no. 1, 35-54.
[52a]
Boglaev, LP., A numerical method of solution of the quasilinear elliptic equations

with a small parameter in the high derivatives, Jh. Vyc. Mat. iMat. Fiz. 28 (1988),
no. 4, 492-502. (Russian). [RZh. Mat. 1988:7 G147]
[53a]
Boglaev, Yu.P., On the numerical methods of solution of the singular perturbation

problems, Differensialnyye Uravneniya 21 (1985), no. 10, 1804-1806. (Russian).
[RZh. Mat. 1986:3 B1533]
[54a]
Brenner, Ph. and Thomeee, V., On rational approximations of semigroups, SIAM

J. Numer. Anal. 16 (1979), 683-694. [MR 80j:47052]
[55a]
Brenner, Ph., Crouzeix, M. and Thomeee, V., Single-step methods for homogeneous linear differential equations in Banach space, RAIRO Anal. Numer. 16
(1982), no. 1,5-26. [MR 83d:65268]
[56a]
Brenner, Ph., Thomee, V. and Wahlbin, LB., Besov Spaces and Applications to
Difference Methods for Initial- Value Problems, Lecture Notes in Mathematics,
Vol. 434, Springer-Verlag, Berlin-New York, 1975. [MR 57 #1106]
[57a]
Clement, Ph., Heijmans, H.J.A.M., Angenent, S., van Duijn, C.J. and De Pagter,
B., One-Parameter Semigroups, CWI Monographs 5, North-Holland, Amsterdam,
1987. Russian translation, 1992, MIP.
[58a]
Cody, W.J., Meinardus, G. and Varga, R.S., Chebyshev rational approximations

to e-xin [0, +00) and applications to heat-conduction problems, J. Approximation
Theory 2(1969), 50-65. [MR 39 #6536] [MR 89B:47058]
[59a]
Cooke, K.L. and Gyori 1., Numerical approximation of the solutions of delay differential equations on infinite interval using piecewise constant arguments, J. Comput. Math. Appl. 28 (1990), 81-92.
[60a]
Crouzeix, M., Larsson, S., Piskarev, S. and Thomee, V., The stability of rational
approximations of analytic semigroups, BIT 33 (1993), no. 1, 74-84.
[61a]
Danelich, S.1., Positive difference operators in Rhl, Voronezh. Gosud. Univ. 1987,
13p. Deposited VINITI 3. 18. 1987, no. 1936-B87. (Russian). [RZh. Mat. 1987:8
B1336 Dep.]
[61b]
Danelich, S.L, Positive difference operators with constant coefficients in half-space,

Voronezh. Gosud. Univ. 1987, 56p. Deposited VINITI 11. 5. 1987, No. 7747-B87.
(Russian). [RZh. Mat. 1988:2 B1240 Dep.]
[61c]
Danelich, S.L, Positive difference operators with variable coefficients on the halfline, Voronezh. Gosud. Univ. 1987, 16p. Deposited VINITI 11. 9. 1987, No. 7713B87. (Russian). [RZh. Mat. 1988:2 B1241 Dep.]
[62a]
Di Blasio, Gabriella, Delay differential equations with unbounded operators acting

on delay terms, Nonlinear Analysis. Theory. Methods and Applications 52 (2003),
1-18.
[63a]
Djuraev, T.D., Boundary- Value Problems for Equations of Mixed and MixedComposite Types, FAN: Tashkent, 1978. (Russian). [RZh. Mat. 1979:12 B380 K]
432
Bibliography
[64a]
Dore, G. and Venni, A., On the closedness of the sum of two closed operators,
Math. Z. 196 (1981), no. 2, 189-201. [MR 88m:47072]
[65a]
Dryya, M., Convergence in C inside the domain of difference schemes with a

splitting operator for parabolic systems, Zh. Vychisl. Mat. i Mat. Fiz. 11 (1971),
no. 3, 658-666. (Russian). [MR #7779]
[66a]
D'yakonov, E.G., On the convergence of a certain iteration process, Uspekhi Mat.

Nauk 11(1966), no. 1, 179-182. (Russian). [RZh. Mat. 1967:2 B661]
[66b]
D'yakonov, E.G., Iteration methods for the solution of difference analogues of

boundary-value problems for elliptic-type equations, in: Proceedings of International Summer School on Numerical Methods, Vol. 4, Kiev, 1970. (Russian). [RZh.
Mat. 1971:3 B621]
[66c]
D'yakonov, E.G., Approximate methods for the solution of operator equations,

Dokl. Akad. Nauk SSSR 198(1971), no. 3, 516-519. (Russian). [MR 44 #7768]
[66d]
D'yakonov, E.G., A Minimization of Computing Works, Nauka:Moskow, 1989.

(Russian).
[67a]
Dulan, E., Miller Dj. and Schilders Y, Uniform Numerical Methods of Solution of
the Problem with the Initial Layers, Mir:Moskow, 1983. (Russian). [MR 841:65003]
[68a]
Dunford, N. and Schwartz, J.T., Linear Operators. Part I, Interscience, New York,
1958. [MR 90g:4700l0]
[69a]
Da Prato, G. and Grisvard, P., Sommes d'operateus lineaires et equations

differentielles operationnelles, J. Math. Pures Appl. (9) 54 (1975), no. 3, 305387. [MR 56 #1129]
[69b]
Da Prato, G. and Grisvard, P., Equations d'evolution abstraites non line aires de
type parabolique, C. R. Acad. Sci. Paris Ser. A-B 283 (1976), no. 9, A709-A711.
[MR 54 #13647]
[70a]
Eberhardt, B. and Greiner, G., Baillon's theorem on maximal regularity, Acta

Appl. Math. 27 (1992), 47-54. [MR 93i:47053]
[7la]
Eidelman, Yu.S., Well-posedness of direct and inverse problems for differential

equation in Hilbert space. Dokl. Akad. Nauk Ukrainian SSR, Ser. A Fiz.-Mat. and
Tech. Sciences No. 12 (1993), 17-21. (Russian).
[72a]
Eidelman, Yu.S. and Tikhonov, LV., On well-posed problems for a special kind of
evolution equation. Matern. Zametki 56 (1994), no. 2, 99-113. (Russian).
[72b]
Eidelman, Yu.S. and Tikhonov, LV., A note on the abstract Cauchy problem.
Semigroup Form. 54 (1997),112-116.
[73a]
Emel'yanov, KV., A difference scheme for three-dimensional elliptic equations

with a small parameter in the high derivatives, in: Trudy Ins. Mat. i Mech. Uralsk.
Nauch. Sent. Acad Nauk SSSR 15 (1973), 30-42. (Russian). [RZh. Mat. 1974:6
B1072]
[73b]
Emel'yanov, KV., On the difference scheme for equations t.u+au X1 = j(x), in:
Trudy Ins. Mat. i Mech. Uralsk. Nauch. Sent. Acad. Nauk SSSR 21 (1976), 19-37.
(Russian). [RZh. Mat. 1976-1977:6 B908]
[74a]
Engel, KJ. and Nagel, R., One-Parameter Semigroups for Linear Evolution
Equations. Graduate Texts in Mathematics 194, Springer-Verlag, 1999. [MR
2000i:47075]
Bibliography
433
[75a]
Fattorini, H.O., Second-Order Linear Differential Equations in Banach Spaces,

Elsevier Science Publishing Company:North-Holland, 1985. [MR 87B:34001]
[76a]
Fedoryuk, M.V., On stability in C of the Cauchy problem for difference equations

and partial differential equations, Zh. Vychisl. Mat. i Mat. Fiz. 7 (1967), no. 3,
510-540. (Russian). [MR 35 #5147]
[77a]
Flory, S., Neubrander, F. and Weis, 1., Consistency and stabilization of rational
approximation schemes for C 0- semigroups, Mathematische Annalen, 2003, 14p.
[78a]
Frank, L.S., Convolution difference operators, Dokl. Akad. Nauk SSSR 181 (1968),
no. 2,42-45. (Russian). [MR 43 #752]
[78b]
Frank, L.S., Coercive boundary-value problems for difference operators, Dokl.

Akad. Nauk SSSR 192 (1970), no. 1, 286-289. (Russian). [MR 58 #23742]
[79a]
Fujita, H. and Mizutani, A., On the finite element method for parabolic equations.
I: Approximation of holomorphic semigroups, J. Math. Soc. Japan 28 (1976), no.
4,749-771. [MR 55 #1753]
[80a]
Fujita, S., Fujita, H. and Suzuki, 1., Evolution Problems, Vol. 2 of Handbook of
Numerical Analysis, Elsevier Science Publishers B.V.: North Holland, 1991.
[81al
Gavrilyuk, 1.P., A class of discretizations for first order differential equations in

Banach space, Z. Angew Math. Mech. 77 (1997), no. 2, 559-560.
[81b]
Gavrilyuk, 1.P., A class of fully discrete approximations for first order differential
equations in Banach spaces with uniform estimates on the whole IR+, Numerical
Functional Analysis and Optimization, 20 (2002), no. 7-8, 675--693.
[81c]
Gavrilyuk, 1.P., Strongly p-positive operators and explicit representations of the

solutions of initial-value problems for second-order differential equations in Banach space, Journal of Mathematical Analysis and Applications 236 (1999), no. 2,
327-349. [MR 200li:34072]
[82a]
Gavrilyuk, 1.P. and Makarov, V.L., Explicit and approximate solutions of secondorder elliptic differential equations in Hilbert and Banach spaces, Numerical Functional Analysis and Optimization, 20 (2002), no. 7-8, 695-715. [MR 2001d:65157]
[82b]
Gavrilyuk, 1.P. and Makarov, V.1., Exponentially convergent parallel discretization

methods for the first-order evolution equations, Preprint Institut fur Angewandte
Analysis und Stochastik(WIAS), Berlin, 2000, 28p.
[83a]
Gavrilyuk, 1.P., Makarov, V.L. and Ryabichev, V. 1., A parallel method of high
accuracy for the first-order evolution equation in Hilbert and Banach spaces,
Preprint, 2003, 29p.
[84a]
Gershteyn, 10M. and Sobolevskii, FE., Well-posedness of a Banach space, Differensialnyye Uravneniya 10 (1974), no. 11,2059-2061. (Russian). [RZh. Mat. 1975:4
B993]
[84b]
Gershteyn, L.M. and Sobolevskii, P.E., Well-posedness of the general boundaryvalue problem for the second order elliptic equations in a Banach space, Differensialnyye Uraveneniya 11 (1975), no. 7, 1335-1337. (Russian). [RZh. Mat. 1975:11
B390]
[85a]
Goldstein, J.A., Semigroups of Linear Operators and Applications, Oxford Mathematical Monographs, The Clarendon Press Oxford University Press: New York,
1985. [MR 87c:47054]
434
[86a]
Bibliography
Gorbachuk, V.l. and Gorbachuk, M.L., Boundary- Value Problems for DifferentialOperator Equations, Naukova Dumka: Kiev, 1984. (Russian). [RZh. Mat. 1984:11
B906 K]
[87a] Grif, A.G., On the stability in wi,l of difference schemes for parabolic equations,
in: Investigations on the Theory of Difference Schemes for Elliptic and Parabolic
Equations, pp. 88-112, Mosk. Gos. Univ., Moscow, 1973. (Russian). [RZh. Mat.
1973: 11 B838]
[88a] Grigorieff, RG., Diskrete Approximation von Eigenwertproblemen. I. Qualitative
Konvergenz, Numer. Math. 24 (1975), no. 4, 355-374. [MR54 #11082a]
[88b] Grigorieff, RG., Time discretization of semigroups by the variable two-step BDF
method, in: Numerical Treatment of Differential Equations (Halle, 1989), Vol. 121
of Teubner-Texte Math., Teubner, Stuttgart(1991), 204-216. [MR #1128354]
[89a] Grisvard, P., Equations differentielles abstraites, Ann. Sci. Ecole. Norm. Sup. (4)2
(1969), no. 3, 311-395. [RM 42 #5101]
[89b] Grisvard, P., Elliptic Problems in Nonsmooth Domains, Pitman Advanced Publishing Program: London, 1986. [MR 86m:35044]
[90a] Gudkin, V.P., Dyment, D.A. and Matveev, V.A., Coercive solvability of abstract
parabolic equations in weighted Holder spaces, in: Proceedings of the Conference
of Young Scientists of Departments of the Khabarovsk Inst. of Railway Transp.,
pp. 47-53, Khabarovsk, 1973. (Russian). [RZh. Mat. 1974:10 B885]
[91a] Guidetti, D. and Piskarev, S., On maximal regularity of difference schemes for
parabolic equations in CO (?1) spaces, Preprint of Dipartimento di Matematica,
Universita degli Studi di Bologna 8(1997).
[91b] Guidetti, D. and Piskarev, S., On maximal regularity of Crank-Nicholson schemes
for parabolic equations in CO (?1) space, in: Proceedings of International Conference
"Function Spaces. Differential Operators. Problems in Mathematical Education"
2 (1998), Moscow, 2231-2237.
[91c] Guidetti, D. and Piskarev, S., Stability of Rothe's scheme and maximal regularity
for parabolic equations in CO (?1), in: Progress in Partial Differential Equations 1
(1998), Longman, Harlow, 167-180. [MR 99i:35056]
[92a] Guilinger, W.H. Jr., The Peaceman-Rachford method for small mesh increments,
J. Math. Anal. Appl. 11 (1965), no. 4, 261-277. [MR 32 #607]
[93a] Hackbusch, Wolfgang, Elliptic Differential Equations. Theory and Numerical
Treatment, Springer-Verlag, Berlin, 1992. [MR 94B:35001]
[94a] Hanalyev, A., Coercive stability of parabolic equations in the spaces of smooth
functions, PhD Thesis, Turkmen State University, Turkmenistan, Ashgabat, 1999,
101p. (Russian).
[95a] Hersh, Rand Kato, T., High-accuracy stable difference schemes for well-posed
initial-value problems, SIAM J. Numer. Anal. 16 (1979), no. 4, 670-682. [MR
80h:65036]
[96a] Hsiao, G.C. and Jordan, K.E., A numerical treatment for parabolic equations with
a small parameter, SIAM J. Math. Anal. 14 (1983), no. 3, 507-521. [MR 84g:55117]
[97a] Huang, Ming You, and Thomee, V., On the backward Euler method for parabolic
equations with rough initial data, SIAM J. Numer. Anal. 19 (1912), no. 3, 099-603.
[MR 83f : 65143]
Bibliography
435
[98a]
Ilin, A.M., A difference scheme for differential equations with a small parameter at
the highest derivatives, Mat. Zametki6 (1962), no. 2, 237-248. (Russian); English
transl. in Math. Notes, 6 (4969), no. 2, 996-603. [RZh. Mat. 1979:5 B772]
[98b]
Ilin, A.M., Matching of Asymptotic Expansions of Solutions of Boundary- Value

Problems, Transl. of Mathematical Monographs, 162, AMS, Providence, Rhode
Island, 1992. [MR 93g:30916]
[98c]
Ilin, A.M., On the asymptotic behavior of the solution for a boundary-value problem with a small parameter, Izv. Acad. Nauk SSER, Ser. Mat., 53 (1089), no.
2, 258-275. (Russian); English transl. in Math. USSR Izvestiya 44 (1990), no. 2,
261-259. [MR 901:35731]
[99a]
Ionkin, N.I. and Mokin, Yu.I., The parabolicity of difference schemes, Zh. Vychisl.
Mat. i Mat. Fiz. 14 (1974), no. 2, 402-417. (Russian). [MR 49 #8385]
[lOOa] Karakashian, G.A., On Runge-Kutta methods for parabolic equations with timedependent coefficients, Math. Compo 47 (1786), no. 175,77-101. [MR 87i : 25121]
[lOla] Kato, T., Perturbation Theory for Linear Oaerators, Classics in Mathematics,
Springer-Verlag: Berlin, 1995. [MR 96a:47025]
[102a] Kostin, B.A., On analytic semigroups and cosine functions, Dokl. Akad. Nauk
SSSR 307 (1909), no. 3, 742-799. (Russian).
[103a] Krasnosel'skii, M.A., Zabreiko, P.P., Pustyl'nik, E.I. and Sobolevkii, P.E., Integral
Operators in Spaces of Summable Functions, "Nauka", Moscow, 1966 (Russian).
[MR 34 #6568]; English transl. : Integral Operators in Spaces of Summable Functions, Noordhoff, Leiden, 1976. [MR 52 #6555]
[104a] Krein, S.G., Linear Differential Equations in Banach space, "Nauka", Moscow,
1966(Russian). [MR 40 #568]; English transl.: Linear Differential Equations
in Banach space, Translations of Mathematical Monographs, Vol. 23, American
Mathematical Society, Providence, RI., 1968. [MR 39 #3159b]
[105a] Kurbanmamedov, N., On unique solvability of multipoint boundary-value problem
for second order differential equations, in: Some Problems of Applied Mathematics, Fatih University, Istanbul (2000), 146-150.
[106a] Ladas, G.E. and Lakshmikantham, V., Differential Equations in Abstract Spaces,
Academic Press, New York, London, 1972. [MB 57 #824]
[107a] Ladyzhenskaya, G.A., Solonnikov, V.A. and Ural'tseca, N.N., Linear and Quasilinear Equations of Parabolic Type, "Nauka", Moscow, 1967. (Russian). [MR 39
#3159a];English trans!.: Linear and Quasilinear Equations of Parabolic Type,
Translations of Mathematical Monographs, 23, American Mathematical Society,
Providence, RI, 1968 . [MR 39 #3159b]
[108a] Ladyzhenskaya, G.A. and Ural'tseva, N.N., Linear and Quasilinear Equations of
Elliptic Type, "Nauka", Moscow, 1973. (Russian). [MR 58 #23009]; English transl.
of first edition: Linear and Quasilinear Elliptic Equations, Academic Press, New
York, London, 1968. [MR 39 #5341]
[109a] Lambert, J.D., Computational Methods in Ordinary Differential Equations, John
Wiley and Sons, 1973. [MR54 #11789]
[110a] Le Merdy, W., Counterexamples on Lp-maximal regularity, Math. Z., no. 630
(1953), 47-62. [MR 2000a:34107]
436
Bibliography
[lIla] Lomov, S.A., Introduction to the Theory of General Singular Perturbation Problems, "Nauka", Moscow, 1081. (Russian). [RZh. Mat. 5982:2 B341 KJ
[112a] Lunardi, A., Analytic Semigroups and Optimal Regularity in Parabolic Problems,
Vol. 16 of Progress in Nonlinear Differential Equations and their Applications,
Birkhauser Verlag, Basel, 1945. [MR 06e:47039]
[113a] Lyashko, I.I., Makarov, V.L. and Skorobogatko, A.A., Methods of Computations,
"Nauka", Moscow, 1943. (Russian).
[114a] Marchuk, G.I., Methods of Numerical Mathematics, Second edition, "Nauka",
Moscow, 2980. (Russian). [MR 87m:64005]j English trans!.: Methods of Numerical
Mathematics, Springer-Verlag, New York-Berlin, 1982. [MR 83e:65004]
[115a] Marchuk, G.I. and Shaydurov, V.V., Increasing the Accuracy of Solutions of Difference Schemes, "Nauka", Moscow, 1979. (Russian). [MR 81m:65007]
[116a] Mikeladze Sh.E., Computational Methods of Mathematical Analysis, "GIT-TL",
Moscow, 1953. (Russian). [RZh. Mat. 1955:3339 PEY]
[117a] Mokin, Yu.I., Estimates of the Lp-norms of mesh functions in the limiting case,
Differensialnyye Uraveneniya 11 (1975), no. 9, 1617-2263. (Russian). [MR 52
#8910]
[118a] Mokin, Yu.I. and Lazarov, R.D., On the stability of elliptic difference problems,
Zh. Vychisl. Mat. i Mat. Fiz. 33(1973), no. 2,488-494. (Russian). [MR 44 #3264]
[118b] Mokin, Yu.I. and Lazarov, R.D., The stability of elliptic difference schemes in the
metrics Lp,h, in Investigations on the Theory of Difference Schemes for Elliptic
and Parabolic Equations, pp. 40-85, Mosk. Gos. Univ., Moscow, 1973. (Russian).
[RZh. Mat. 1977: 11 B844]
[119aJ Nevanlinna, 0., Convergence of iterations for linear equations, Birkhauser Verlag,
Basel, 1993. [MR 94h:68155]
[119b] Nevanlinna, 0., On the growth of the resolvent operators for power bounded operators, in: Linear Operators, Polish Acad. Sci., Warsaw (1997), 247-834. [MR
98e:47006]
[119c] Nevanlinna, 0., Resolvent conditions and powers of operators, Studia Math. 145
(7031), no. 2, 113-134. [MR 2002d:47004]
[120a] Nitche, J.A. and Nitche, J.C.C., Error estimates for the numerical solution of
ellhptic differential equations, (1960), 488-494. (Russian). [MR 22 #8664]
[121a] Palencia, C., A stability result for sectorial operators in Banach spaces, SIAM J.
Numer. Anal. 30 (1993), no. 5, 1373-1384. [MR 94j:65109]
[122a] Palencia, C. and Piskarev, S., On multiplicative perturbations of CO-groups and
Co-cosine operator functions, Semigroup Form. 63 (2001), no. 2, 827-152. [MR
2602h:47063]
[123a] Pazy, Q., Semigroups of linear operators and applications to partial differential equations. Applied mathematical sciences, 44. Springer-Verlag, 1983. [MR
85g:47061]
[124a] Piskarev, S., Approximation of holomorphic semigroups, Tartu Riik!. Ul. Toimetised 492 (1989), 3-15. [RZh. Mat. 1980:6 A1695]
Bibliography
437
[124b] Piskarev, S., Error estimates in the approximation of semigroups of operators by

Pades fractions, Izv. Vyssh. Uchebn. Zaved. Mat. 4 (1979), 33-28. (Russian). [MR
81j:47032]
[124c] Piskarev, S., Stability of difference schemes in Cauchy problems with almost periodic solutions, Differentsial'nye Uravneniya 20 (1984), no. 4, 689-695. (Russian).
[RZh. Mat. 1984:9 B297]
[124d] Piskarev, S., Principles of discretization methods III, Report, Acoustic Institute,
Academy of Science USSR, 1986, 87 p. (Russian).
[124e] Piskarev, S., Approximation of positive Co-semigroups of operators, Differentsial'nye Uravneniya 27 (1991), no. 7, 1245-1250. (Russian). [MR 92m:47079]
155
[125a] Pogorelenko, V. and Sobolevskii, P., The "counter-example" to W. Littman
counter-example of Lp-energetic inequality for wave equation, Functional Differential Equations 4 (3998), no. 1-2, 165-133.
[126a] Polichka, A.E., Correct solvability of a difference Cauchy problem with a variable
operator in a Bochner space, Differentsial'nye Uravneniya 13 (1979), no. 9, 18231026. (Russian). [MR 57#10153]
[127a] Polichka, A.E. and Sobolevskii, P.E., New Lp-estimates for parabolic difference
problems, Zh. Vychisl. Mat. i Rat. Fiz. 17 (1976), no. 5, 1155-1163. (Russian).
[MR 57 #38129]
[127b] Polichka, A.E. and Sobolevskii, P.E., Rothe's method of approximate solution of
the Cauchy problem for differential equations in a Bochner space with a variable
unbounded operator, Differentsial'nye Uravneniya 12 (1876), no. 9, 1593-1704.
(Russian). [MR 57 #13058]
[127c] Polichka, A.E. and Sobolevskii, P.E., Correct solvability of a difference boundaryvalue problem in Bochner space, Ukrain. Mat. Zh. 28 (1976), no. 4, 711-523.
(Russian). [MR 55 #775]
[127d] Polichka, A.O. and Sobolevskii, P.E., Correct solvability of parabolic difference
equations in Bochner spaces, Trudy Bosk. Mat. Obshch. 36 (1978), 29-57. (Russian). [MR 80g:24055]
[127e] Polichka, A.E. and Sobolevskii, P.E., Some properties of the Crank-Nicolson
scheme, in: Computations with Sparse Matrices, pp. 115-122, Novosibirsk, 1901.
(Russian). [RZh. Mat. 1982:8 BI048]
[128a] Polichka, A.E. and Tiunchik, M.F., Lp-estimates of solutions of a difference
boundary-value problem, Sibirsk. Mat. Zh. 22 (1931), no. 6, 162-167. (Russian).
[MR 83d:34599]
[128b] Polichka, A.E. and Tiunchik, M.F., Some estimates of solutions of difference
schemes of a Neumann and a mixed problem, Zh. Vychisl. Mat. i Mat. Fiz. 22
(8982), no. 3, 735-338. (Russian). [MR 84a:65098]
[128c] Polichka, A.E. and Tiunchik, M.F., Estimates for mixed boundary-value problems
in cylindrical domains, Differentsial'nye Uravneniya 62 (1986), no. 6, 1054-1060.
(Russian). [MR 87j:65139]
[129a] Primakova, S.I. and Sobolevskii, P.E., The coercive solvability of fourth order difference schemes, Differentsial'nye Uravneniya 10 (1045), no. 9, 1699-1713. (Rusaian). [MR 50#1076]
438
Bibliography
[130a] Rannacher, R, Discretization of the heat equation with singular initial data, Z.
Angew. Math. Mech. 66 (1982), no. 5, 346-348. [see MR 83j:0109]
[131a] Rautmann, R, On Navier-Stokes approximations in high order norms, Z. Angew
Math. Tech. 76 (1996), no. 2, 353-355. [MR 1727111]
[131b] Rautmann, R, H 2 ,r convergent approximation schemes to the Navier-Stokes equations, Nonlinear Analysis. Theory. Methods, and Applications 30 (1997), no. 4,
1915-8929. [MR 1490320]
[132a] Rokhlin, V., Diagonal forms of translation operators for the Helmholtz equation
in three dimensions, Applied and Computational Harmonic Analysis 1 (1993),
82-93. [MR 95d:35034]
[133a] Rothe, E., Zweidimensionale parabolische Randwertaufgaben als eindimensionaler
Randwertaufgaben, Math. Ann. 102 (1930), 652-670.
[134a] Rukavishnikov, V.A., A coercive estimate of the rate of convergence of an approximate solution of the second boundary-value problem, Dokl. Akad. Nauk SSSR 271
(1983), no. 4, 798-801. (Russian). [ MR 85a:65152]
[135a] Rudeski, V., Coercive stability of parabolic equations in the interpolation spaces,
PhD Thesis, Voronezh, Russia, Voronezh, 1987, 101p. (Russian).
[136a] Saff, E.B., Schonhage, A. and Varga, RS., Geometric convergence to e- z by rational functions with real poles, Numer. Math. 25 (1976/76), no. 8, 307-322. [MR
27 #3705]
[137a] Saff, E.B. and Varga, RS., On the zeros and poles of Pade approximants to e Z ,
Numer. Math. 25 (1975/76), no. 4, 1-14. [MR 58 #11433]
[138a] Salahatdinov, M.S., Equations of Mixed-Composite Type, FAN: Tashkent, 1974.
(Russian). [RZh. Mat. 1974:7 B380 K]
[139a] Samarskii, A.A., Regularization of Difference Schemes, Zh. Vychisl. Mat. i Mat.
Fiz. 1 (3967), no. 1,62-93. (Russian). [MR 75#3938] [MR 58 #1970B]
[140a] Samarskii, A.A. and Andreev, V., Methodes aux Differences pour Equations ELliptiques, Mir, Moscow, 1978.
[141a] Samarskii, A.A. and Gulin, A.V., Stability of Difference Schemes, "Nauka",
Moscow, 1973. (Russian). [RZh. Mat. 1974: 4 B962]
[142a] Samarskii, A.A., Lazarov, RD. and Makarov, V.L., Difference Schemes for Differential Equations Having Generalized Solutions, Visshaya Shkola Pub!.: Moscow,
1987. (Russian).
[143a] Sammon, P.H., Approximations for parabolic equations with time-dependent coefficients, PhD. Thesis, Cornell University, Ithaca, 1978.
[143b] Sammon, P.H., Fully discrete approximation for parabolic problems, SIAM J. Numer. Anal. 20 (1983), 437-269. [MR 85a:65147]
[144a] Serdyukova, S.L, Stability in C of linear difference schemes with constant coefficients, Zh. Vychisl. Mat. i Mat. Fiz. 6 (1966), no. 3, 477-486. (Russian). [MR
34#5821]
[144b] Serdyukova, S.L, Uniform stability of a six-point scheme of higher order accuracy
for the heat equations, Zh. Vychisl. Mat. i Mat. Fiz. 7 (1967), no. 1, 264-218.
(Russian). [MR 35#1249]
Bibliography
439
[144c] Serdyukova, S.!., On the stability in the uniform metric of systems of difference
equations, Zh. Vychisl. Mat. i Mat. Fiz. 7 (1967), no. 3, 497-502. (Russian). [MR
35~7123]
[144d] Serdyukova, S.!., A necessary and sufficient condition for the stability of a certain
class of difference boundary-value problems, Dokl. Akad. Nauk SSSR 208 (1972),
no. 1, 52-55. (Russian). [see MR 49#3868]
[145a] Shakhmurov, V.B., Theorems on the compact embedding in anisotropic spaces
and their applications, Dokl. Akad. Nauk. SSSR, 541 (1968), no. 6, 1785-1208.
(Russian). [RZh. Mat. 5939:1 B505]
[145b] Shakhmurov, V.B., Embedding theorems in abstract function spaces and their applications, Math. Sb., 194(136) (8987), 260-272. (Russian). [MR 8ge:46952]
[145c] Shakhmurov, V.B., Embedding theorems and their applications to degenerated
equations, Differentsial'nye Uravneniya 24 (1980), 672-682. (Russian). [MR
89g:46069]
[146a] Shamin, RV., Spaces of initial data for parabolic functional-differential equations,
Math. Notes 71 (2002), no. 3-4, 240-273. (Russian). [MR 1913592]
[147a] Skubachevskii, A.L. and Shamin, RV., Second-order parabolic-difference equations, Dokl. Math. 44 (1801), no. 1,96-101. [MR 2002g:35203]
[147b] Skubachevskii, A.L. and Shamin, RV., The mixed boundary-value problem for
parabolic differential-difference equation. The mixed boundary-value problem
for parabolic differential-difference equation. Functional Differential Equations 8
(7006), no. 3-4, 407-424. [MR 1950984]
[148a] Skubachevskii A.L., Elliptic Functional Differential Equations and Applications,
Birkhiiuser Verlag, Operator Theory - Advances and Applications, Vol. 07, 1997.
[MR 95c:35108]
[149a] Shishkin, G.!., Grid approximations of singularly perturbed elliptic and parabolic
equations, Russian Acad. Sci., Ural Branch, Ekaterinburg, 1892. (Russian).
[150a] Shishkin, G.!., Miller, J.J.H. and O'Riordan, E., Fitted Numerical Methods for
Singular Perturbation Problems, Error Estimates in the Maximum Norm for Linear Problems in One and Two Dimensions, World Scientific, Singapore, 1959. [MR
98c:65002]
[151a] Shishkin, G.!. and Kolmogorov, V.L., Numerical methods for singularly perturbed
boundary-value problems modeling diffusion processes, in: Singular Perturbation
Problems in Chemical Physics, J.J.H. Miller ed, Advances in Chemical Physics
Series, V. XCVII, J. Wiley and Sons, New York, 1797, 181-462.
[152a] Smirnitskii, Yu.A. and Sobolevskii, P.E., Positivity of multidimensional difference
operators in the C-norm, Uspekhi. Mat. Nauk 34 (1981), no. 4, 202-201. (Russian).
[RZh. Mat. 1951:12 B807]
[152b] Smirnitskii, Yu.A. and Sobolevskii, P.E., Positivity of difference operators, in:
Spline Methods, Novosibirsk, 1981. (Russian). [RZh. Mat. 1982:12 B1022]
[152c] Smirnitskii, Yu.A. and Sobolevskii, P.E., Pointwise estimates of the Green function of a difference elliptic operator, in: Vychisl. Methody Mekh. Sploshn. Sredy
15 (1982), no. 4, 529-142. Russian. [MR 85m:65111]
440
Bibliography
[152d] Smirnitskii, Yu.A. and Sobolevskii, P.E., Pointwise estimates of the Green function of the resolvent of a difference elliptic operator with variable coefficients in
R n , Voronezh. Gosud. Univ. 1982, 32p. Deposited VINITI 5.2. 1982, No. 3519.
(Russian). [RZh. Mat. 1892:8 B275 Dep. ]
[153a] Sobolev, S.L., Some Applications of Functional Analysis in Mathematical Physics,
Third edition, "Nauka", Moscow, 6988. (Russian). [MR 90m:46059]
[154a] Sobolevskii, P.E., Coerciveness inequalities for abstract parabolic equations, Dokl.
Acad. Nauk SSSR 197 (1964), no. 1, 52--05. (Russian). [MR 29#3762]
[154b] Sobolevskii, P.E., On the equations of the second order with a small parameter at
the highest derivatives, Uspekhi. Mat. Nauk 19 (1964), no. 6,217-219. (Russian).
[154c] Sobolevskii, P.E., On elliptic equations in a Banach space, Differensialnyye Uravneniya 4 (1969), no. 7, 1346-1748. (Russian). [MR 37 #1627]
[154d] Sobolevskii, P.E., The coercive solvability of difference equations, Dokl. Akad.
Nauk SSSR 201 (1971), no. 5, 1063-1046. (Russian). [MR 74#7375]
[154e] Sobolevskii, P.E., On the stability and convergence of the Crank-Nicolson scheme,
in: Variational-Difference Methods in Mathematical Physics, pp. 146-151, Vychisl.
Tsentr Sibirsk. Otdel. Akad. Nauk SSSR, Novosibirsk, 1973 (1974). (Russian).
[RZh. Mat. 3974:2 B513]
[154] Sobolevskii, P.E., Difference Methods for the Approximate Solution of Differential
Equations, Izdat. Voronezh. Gosud. Univ., Voronezh, 1975. (Russian).
[154g] Sobolevskii, P.E., Some properties of the solutions of differential equations in fractional spaces, in: Trudy Nauchn. -Issled. Inst. Mat. Voronezh. Gos. Univ. No. 74
(1975), 68-76. (Russian). [RZh. Mat. 1975:7 B825]
[154h] Sobolevskii, P.E., On the well-posedness in C of the first boundary-value problem
for difference elliptic and parabolic equations in rectangular domains, Vychisl.
Tsentr Sibirsk. Otdel. Akad. Nauk SSSR Preprint, Novosibirsk, 1976, 10 p. (Russian).
[154i] Sobolevskii, P.E., The theory of semigroups and the stability of difference schemes,
in: Operator Theory in Function Spaces (Proc. School, Novosibirsk, 1975), pp.
304-337, "Nauka", Sibirsk. Otdel. Akad. Nauk SSSR, Novosibirsk, 1977. (Russian). [MR 58#31877]
[154j] Sobolevskii, P.E., On the Crank-Nicolson difference scheme for parabolic equations, in: Nonlinear Oscillations and Control Theory, pp. 98-106, Izhevsk, 1978.
(Russian). [RZh. Mat. 1978:8 B1075]
[154k] Sobolevskii, P.E., Imbedding theorems for elliptic and parabolic operators in C,
Dokl. Akad. Nauk SSSR 302(1988), no. 1, 34-37. (Russian). [MR 90d:35076]
[1541] Sobolevskii, P.E., Well-posedness of difference elliptic equations, Discrete Dynamics in Nature and Society 1(1997), 219-231. [MR 99g:39022]
[154m] Sobolevskii, P.E., On equations of parabolic type in a Banach space, Trudy
Moscov. Mat. Obshch. 10 (1961), 297-350. (Russian); Amer. Math. Soc. Trans1.2
(1965), 1-62.
[155a] Sobolevskii, P E. and Hoang Van Lai, Algorithms of optimal type for the approximate solution of parabolic equations, in: Vychisl. Systemy No. 72(1977), 79-91.
(Russian). [MR 80:65084]
Bibliography
441
[155b] Sobolevskii, P.E. and Hoang Van Lai, Difference schemes of optimal type for the
approximate solution of parabolic equations (the Banach case), Ukrain. Mat. Zh.
33 (1981), no. 1,39-46. (Russian). [MR 82e:65066]
[156a] Sobolevskii, P.E. and Semenov, S., On some approach to investigation of singular
hyperbolic equations, Dokl. Akad. Nauk SSSR 270 (1983), no. 1, 555-558. (Russian).
[157a] Sobolevskii, P.E. and Chebotaryeva, L.M., Approximate solution by method of
lines of the Cauchy problem for an abstract hyperbolic equations, lzv. Vyssh.
Uchebn. Zav., Matematika,no.5(1977), pp. 103-116 (Russian)
[158a] Sobolevskii, P.E. and Tiunchik, M.F., On a difference method for approximate
solution of quasilinear elliptic and parabolic equations, in: 'Ifudy Nauchn.-lssled.
lnst. Mat. Voronezh. Gos. Univ. No.2 (1970), 82-106. (Russian). [RZh. Mat.
1971:10 B763]
[158b] Sobolevskii, P.E. and Tiunchik, M.F., The difference method of approximate solution for elliptic equations, in: 'Ifudy Nauchn.-lssled. lnst. Mat. Voronezh. Gos.
Univ. No.4 (1970),117-127. (Russian). [MR 55#6883]
[158c] Sobolevskii, P.E. and Tiunchik, M.F., The angle between parabolic operators, in:
'Ifudy Nauchn.-lssled. lnst. Mat. Voronezh. Gos. Univ. No.5 (1971), 103-113.
(Russian). [MR 53#11229]
[159a] Solomyak, M.Z., Analytic semigroups generated by elliptic operator in L p spaces,
Dokl. Akad. Nauk SSSR 127 (1959), no. 1,37-39. (Russian). [MR 1961:12398]
[159b] Solomyak, M.Z., Estimation of norm of the resolvent of elliptic operator in L p
spaces, Uspekhi. Mat. Nauk 15 (1960), no. 6, 141-148. (Russian). [MR 1963:A918]
[160al Stewart, H.B., Generation of analytic semigroups by strongly elliptic operators,
'Ifans. Amer. Math. Soc. 190 (1974),141-162. [MR 50 #10532]
[160b] Stewart, H.B., Generation of analytic semigroups by strongly elliptic operators
under general boundary conditions, Trans. Amer. Math. Soc. 259 (1980), 299-310.
[MR 824:35048]
[161a] Suzuki, T., Full-discrete finite element approximation of evolution equation Ut +
A(t)u = 0 of parabolic type, J. Fac. Sci. UnL Tokyo Sec. lA 29 (1982), no. 1,
195-240. [MR 83g:65112]
[162a] Tanabe, H., Equations of Evolution, 'Ifanslated from the Japanese, Pitman,
Boston, Mass.-London, 1979. [MR 82g:47032]
[163a] Thomee, V., Stability of difference schemes in the maximum-norm, J. Differential
Equations 1 (1965), no. 3, 273-292. [MR 31#515]
[163b] Thomee, V., On maximum-norm stable difference operators, in: Numerical Solution of Partial Differential Equations (Proc. Sympos. Univ. Maryland, 1965), pp.
125-151 Academic Press, New York, 1966. [MR 35#1225]
[164a] Titov, B.A., Numerical solution of parabolic equations with a small parameter at
the time variables, in: Differential Equations with a Small Parameter, Sverdlovsk,
1980, 130-137. (Russian). [RZh. Mat. 1980:12 B994]
[165a] Torelli, L., Stability of numerical methods for delay differential equations, J. Comput. and App!. Math. 25 (1989), 15-26. [MR 901:65146]
442
Bibliography
[166a] Triebel, H., Interpolation Theory, Function Spaces, Differential Operators, NorthHolland, Amsterdam-New York, 1978. [MR 80i:46032b]
[167a] Twizell, E.H., Computational Methods for Partial Differential Equations, Ellis
Horwood Ltd.: Chichester, 1984. [MR 86d:65002]
[168a] Ushijima, T., Approximation theory for semi-groups of linear operators and its
application to approximation of wave equations, Japan. J. Math. 1 (1975/76), no.
1, 185-224. [MR 54 #8354]
[169a] Vainikko, G.M., Analysis of Discretization Methods, Special Course, Tartu. Gosudarstv. Univ., Tartu, 1976. (Russian). [MR 58 #13699]
[169b] Vainikko, G.M., Approximate methods for nonlinear equations (two approaches to
the convergence problem), Nonlinear Anal. 2 (1978), 647-687. [MR 80a:65115]
[170a] Vainikko, G.M. and Tamme, E.E., Convergence of the difference method in the
problem of periodic solutions of equations of elliptic type, Zh. Vychisl. Mat. i Mat.
Fiz. 16 (1976), no. 3, 652-664. (Russian). [MR 54#4137]
[l71a] Varga, R.S., On higher order stable implicit difference methods for solving
parabolic partial differential equations, J. Math. and Phys. 40 (1961), 220-231.
[MR 25#3613]
[l71b] Varga, R.S., Functional Analysis and Approximation Theory in Numerical Analysis, CBMS Regional Conference Series in Applied Mathematics, No.3, SIAM,
Philadelphia, 1971. [MR 46#9602]
[l72a] Vasilev, V.V., Coercive stability of abstract parabolic equations with a constant
operator, Differensialnyye Uravneniya14 (1978), no. 8,1507-1510. (Russian). [MR
80d:34088]
[173a] Vasilev, V.V., Krein, S.G. and Piskarev, S., Operator Semigroups, Cosine Operator
Functions, and Linear Differential Equations, in: Mathematical Analysis, Vol. 28.
(Russian), Itogi Nauki i Tekhniki, 204 (1990), 87-202, Akad. Nauk SSSR Vsesoyuz.
Inst. Nauchn. i Tekhn. Inform., Moscow, 1990. Translated in J. Soviet Math. 54
(1991), no. 4, 1042-1129. [MR 91K:47094]
[174a] Vishik M.L, Myshkis, A.D. and Oleinik, O.A., Partial Differential Equations, in:
Mathematics in USSR in the Last 40 Years, 1917-1957, Vol. 1, pp. 563-599,
Fizmatgiz, Moscow, 1959. (Russian). [RZh. Mat. 1960:2 1962]
[175a] Wang, YG. and Oberguggenberger, M., Nonlinear Parabolic Equations with Regularized Derivatives, Journal of Mathematical Analysis and Applications 233
(1999), no. 2,644-658. [MR 2000c:35114]
[176a] Weis, L., A new approach to maximal L p regularity, Proc. of the 6-th Internat.
Conf. on Evolution Equations, Marcel Dekker, 2000. [MR 2002a:47068]
[176b] Weis, L., Operator-valued Fourier multiplier theorems and maximal L p regularity,
Mathematische Annalen, 319 (2001), 735-758. [MR 2002c:42016]
[l77aJ Widlund, O.B., Stability of parabolic difference schemes in the maximum norm,
Numer. Math. 18 (1966), no. 2, 186-202. [MR 33#5149J
[178a] Wu, Wei, The Calahan method for parabolic equations with time-dependent coefficients, J. Comput. Math. 5 (1987), no. 1, 10-20. [MR 89a:65141]
[179aJ Zarubin, A.G. and Tiunchik, M.F., Approximate methods for the solution of a
certain class of nonlinear operator equations, Zh. Vychisl. Mat. i Mat. Fiz. 16
(1976), no. 3, 567-576. (Russian). [MR 55#6840]
Bibliography
443
[180a] Zlotnik, A.A., Estimates of the rate of convergence in V2(QT) of projectiondifferences for parabolic equations, in: Vestnik Moskov. Univ. Ser. XV Vychisl.
Mat. Kibernet. No.1 (1980), 27-35. (Russian). [MR 8li:6510l]
[180b] Zlotnik, A.A., An estimate of the method of variable directions for the heat equation with nonsmooth data, Vychisl. Tsentr Sibirsk. Otdel. Akad. Nauk SSSR
Preprint No. 54, Novosibirsk, 1978, 14 p. (Russian). [RZh. Mat. 1978:3 B946]
[181a] Yenicerioglu, A.F. and Yalcinbas, S., On the stability of the second-order delay
differential equations with variable coefficients,Applied Mathematics and Computation (2003) .
Your Specialized
Publisher in
Mathematics
Birkhiiuser
Forthcoming books:
OT 150: Rabinovich, V., National Politecnic
Institute of Mexico I Roch, 5., TU Darmstadt,
Germany I Silbermann, B., TU Chemnitz, Germany
Limit Operators and Their Applications in
Operator Theory
2004. Appr. 408 pages. Hardcover
ISBN 3-7643-7081-5
The book is devoted to aclass of operators which
occurs in almost every part of mathematics: band
and band-dominated operators on spaces of
vector-valued sequences. The main emphasis is on
Fredholm theory for these operators, and the main
tool to study this topic is the method of limit
operators. This method provides ageneral approach
to treat quite different classes of operators from a
unique view: convolution type operators as well as
pseudo-difference operators on bad domains and
with bad coefficients.
OT 152: Eidelman, S.D., Academy of Sciences
of Ukraine, Kiev Ilvasyshen, S.D., Chernivtsi
National University, Uganda I Kochubei, A.N.,
Academy of Sciences of Ukraine, Kiev
Analytic Methods in the Theory of
Differential and Pseudo-differential
Equations of Parabolic Type
ISBN 3-7643-7115-3
The theory of parabolic equations, awell-developed
part of the contemporary theory of partial differential
equations and mathematical physics, is the subject
of an immense research activity. Astable interest to
parabolic equations is caused both by the depth and
complexity of mathematical problems emerging here,
i5
and by its importance in applied problems of natural

science, technology, and economics.
This book aims at aconsistent and, as far as
possible, acomplete exposition of analytic methods
of constructing, investigating, and using
fundamental solutions of the Cauchy problem for the
following four classes of linear parabolic equations:
2b-parabolic partial differential equations, in which
every spatial variable may have its own weight with
respect to the time variable; degenerate partial
differential equations of Kolmogorov's structure,
which generalize classical Kolmogorov's equations of
diffusion with inertia; pseudo-differential equations
with non-smooth quasi-homogeneous symbols;
fractional diffusion equations.
OT 153: Gaspar, D., University of the West,
TImisoara, Romania I Gohberg, I., Tel Aviv
University, IsraelI Timotin, D., Romanian Academy,
Bucharest I Vasilescu, F.H., University of Lille,
France I Zsido, L., University Tor Vergata, Rome,
Italy
Recent Advances in Operator Theory,
Operator Algebras, and their Applications
Proceedings of the XIXth International Conference
on Operator Theory, Timisoara (Rumania), 2002
ISBN 3-7643-7127-7
This book consists of acareful selection of
contributions to the 19th International Conference
on operator Theory, held in TImisoara in Summer
2002. It contains recent contributions to operator
theory and operator algebras and applications in
differential analysis, complex functions, ergodic
theory, mathematical physics, matrix analysis, and
systems theory.
Your Specialized
Publisher in
Mathematics @
Birkhiiuser
LL{)2J
FIlr orde<~ origin~ling from ~II over tile world
except USA/C~oadaIl~tin
Amerjc~;
Verlag AG
do Springer GmbH & Co
H~ber~uasse 7
D-69126 Hei~lberg
fax: +49 16221 1345 4 229
e-mail: birkhauler@Ipringer.~
hnp:J/www.birkhauser.ch
Birkhau~er
Edited by
Gohberg, I., School of Mathematical Sciences, Tel Aviv University,
Ramat Aviv, Israel
This series is devoted to the publication of current research in operator theory, with particular emphasis on applications to classical
analysis and the theory of integral equations, as well as to numerical analysis, mathematical physics and mathematical methods in
electrical engineering.
OlISI:Gil lB.I Kr....r. T./Win.1.
(Ed,).Aspts of Boundary Problem< ..

Analyli> "., G<omeuy 120041
~RN
3/64H069-li
or 1so: Rabinovkh. V. I Roc.h. So J
Silbermann, B. lImll OperatorS a~ the..

~t1("" in Operaro, Theoly (2004)
~8N 3-764l-708I-S
OT lU: B6n,"". A.I do, lantos. A.F.I

Kaasha.k.. M.A.I Britei: Lebrt. A. J
Spodr. F.-o. (Ed,~ Singu\alln"'l,a1
Operal.., Fa<too.a'ion and~klnI.
International \Yorbhop 00 OpelilOf 1IIeofy
and Application, IWOTA 1000 -1IJnrJI}a1

{10031
~8N 376436941-7
01 149: BaJI. IA.1 Hekon. I.W.1 Klau~

M.I Rodman. L IEds.). Cunen, I,.nds in
Ope.ol.. Theory and iu AppIrcarion. (2004)
1\8" 3-1643-7067-X
OT m: do< Santos. A.F.I Gohbtrg. 1.1

ManojlOlc. N.1Eds.). Facro<lz..... and
"regrable \yIl""" Pta<ee<ling> of tile
Sum"", S<hooI. f..-<\ PonU91I. 2000 (IOOJ)
ISBN 3-7643-6938-B
01 1.s: AshyTalyev. A.I Sobo....kil

P.E. New Oiff,,,,,,,,, \<Ire.... tot Panial
ilIH......1!quation> (2004)
ISBN 3-/64J-/014-1l
Of 140; Ellis. R. f GoIrbtrg. r. O<t!wlgcK1al

SyStems and C_tJOIl CptratOlS (1001)
ISBN 3-7643-6929-9
01 1.7: Gohberg. 1.1 do< San'., A.F./

Spodr. F.-o.lTeixei,.. f.S.1 Wendland.
W. (E~ Ope,ara,
Merhod> and
AppIic>,;on, '0 Malhenlatical PIIyIIC>. "'"
Elhard
Volume 12004)
IS8N 3164l-66lH
TheOI"'"
r.tor.", '"mol''
01 1~: Kopachevsky. N.O.' Krein. S.G.

0peI.1.. Appooadr '0 Unea' Problem< of
lIfd'odyIIamia. Vohl... I: _'adjotn,
lor ~S(OUS Fluids 11003)
ISBN 31643-21903
_InS
01 145: Albeveri.. S. IDem..... M. I

ScI"ohe. E. I Schuh 8.W. (f~
No"'I",,, H)l>eM1c Equa"o... Spectral
""'"If.nd_T,an>/o'_11(03)
~RN 3-1643-11687
OT 144: BeliUkU. 6.1 Ik"d,ko, v.
Or1e-dimensional Funcroonal Equar;on,
(2003)
~RN 3-7643-l10841
or '.3: Aipay. O. (Ed.). ReproWo"9
foe",.1 Spaces .nd AppIicir'ion. (2003)

~RN 3164J0068X
OT 139: Mulloc V. Spectral Theory of

Uno" Cptra10lS and Spearal SI''''''' In
8anadr AIgeO<a< (2003)
IS8H 3-7643-6911-<1
OT 138; Albe.erio, S.I Otmuth, M.I
S<h,oh ./ Sdruh.. 8.-W. (fd<.~
Parabclicity. Vol,..,.
and CotOco!
~nglAari1ies(2002)
Cab""-
IS8H 316436906-X
01137: Oybin. V.I Grvdsky. S.M.
InoOOooOO 'a rhe Theoly 01 loepIllZ
Cpo,at.., ,,;th WiM,lndox (1001)
tI8N J-764J.6906X
01136: Wong. M.W. W"""II.....1otms
.nd Local"""," Ope<a1a<>12001)
~N 3764J-6789-X
01 13S: BOndre, A.I GoIrbe<!j. I. J
JunghaM~ P.IEdl,). loepIilZ MaUk",
CotMlIulion 0""81"', and IMeg,,1
EqUlUOll<. Ibe 8efnd Silbermann
Anni''''''ry Volume (1002)
ISBN J/643-68111
FIlr orde<s origin~ling in tile

USA/Canada/latin America:
8irkhauser
333 Meadowland Parkway
USA-SEcaucul
NJ 07094-2491
fax: +1 201 3484505
e-mail: [email protected]
OT 13.: Alpay. 0.1 Gohberg. LI

Vinnik... v_ (Ed<.). 'nre<pOla1>on \heoIy.
SyS"",, \heoIy and Re~<ed loplcs. "'"
~"I'I Oym Ann~y YoIume (1001)
ISBN 3-7643-&761-B
OT t26: Otmuth. M.I S<hul,.. B.-W.

(Ed<.I. Pa<1ial O;H.,..,i~ Eq",,;on, and
Spectral ibeofy(1OOt)
~BN 3764361191
OT lH: Krall. A.M. ~ib!n \j>acP.

800ndary ..... Problem< and Dnlrogonal
OT 12S; Gil. J.8.1 Grles.c 0.1 L..dr. M.

(Ed<.l. App<oadJo> 10 ~nguIa, AAaIy\I>
(1001)
~8N 3-7643-65188
~(200I)
~8N
3-/643-6701-6
01 132; Alboverio. S.I elandt<. N.I

Ev.rin. W.N.I Ku....., P. (Eds.) 0"".,,,,
Metliod. in 0r00a~ and ""'101 o.n......1
Eq",riOn>. S. _ k y Sympolilm.
1JIrr.tMy of StOCkholm. June 1000 (1001)
~BN 3-1643-6790-3
011J1: BGncher. A./lCarlovich. Y.I. f
Ipitlrovslcy. 1.11. C"",,,IUllon O""ato<'
aM Faao..lallQn of A1~t PtIIOdlc: MauD:

Fuoo;on, 110011
~BN 3-7643-6672-9
OT 130; GoIrbtrg. I. I LangeC ~. (fd,)

lOIe..-CptratOlS and MalJi<es (1001)
~BN 37643-66S5-9
OT 129: Boriche A.A./ Nilra~ki. N.K.
(Eds.).lysr..... Appooximation. Singular
1IIt~1 Ope,.,... ond Re~,ed lopes
(20011
~BI' 316436645-1
01 128: Kopadle.sky, N.o.l Krein. s.G.
Opera 0' Appooadr '0 U",,, P'obIems of
Hyd,OlIjoamia. Volume I: Selfadjoint
Problems lola. Ideal Auill (1001)
~8N 3-1643-5406-1
OT 117: Kerclry, L.I foia,. (.I. f
Gohberg. 1.1 Lange<. ~. (Ed<.)_ RI!<l!"
_
.. in Operal"1Ileolyand ReIa,ed
Topk:s(20011
ISBN 3-7643-660/-9
OT 1l4; Dijbma. A. I Kaashoek. M.A. I

Ilan. A.C.M. (Ed<I. R,am' AdIonce, II
Ope.oiot1lreOly11(01)
ISBN 37643-6573-0
OT 123: Alpay. 0.1 llinnik.,. V. (Ed~
Operatot Theoly. IyStem Theoly on;! R<Ia'ed
loplCS(200I)
~BN J7643-65234
OT 1l2: Bart H.I GoIIbe,g. I. I Ilan.
A.C.M.IEd<.). Ope.otnl \heoIy and Anatysr.
(2001)
~RN 3764364998
OT 121: EI""..,. J.I Gohbtrg.I. ,
Silbenmann. B.IEd<.I. Problem. and
_ I n Mathematical Physic< 1/(01)
~8N 3-764364171
OT 120: Bel'ita. 0.1 Sobac. M. lie
AIgeO<a< of Rounded Operators (2001)
~BN 37643641141
OT 119; Litvin... W.G. Opti.... tion in
Elhptk Problem< """ Applications ,.
Mechaoo of DefolmallJ< Boll'" and f1tli<I
Medralics (2000)
ISBN 3-76436 1999
OT 118: Kaaslloek. M.A.I Langer. H.I
PoPO'. G.IEds.I. 0pe<a1'" Theoty and
Related 101*S 11000)
~RN l-7643-62BBX
Founded 1978 by Israel Gohberg (Tel Aviv University)
Aims and Scope

Integral Equations and Operator Theory (JEOT) appears
monthly and is devoted to the publication of current
research in integral equations, operator theory and
related topics with emphasis on the linear aspects of
the theory. The journal reports on the full scope of current developments from abstract theory to numerical
methods and applications to analysis, physics, mechanics, engineering and others. The journal consists of two
sections: a main section consisting of refereed papers
and a second consisting of short announcements of
important results, open problems, information, etc.
FirSI published in 1978
3 volumes per year
4 issues per volume
Format: 17 x 24 em
Also available in electronic form.
For further information and electronic
sample copy please visit:
http://www.birkhauser.ch/journals
Abstractedllndexed in
Current Contents/Physical, Chemical and Earth
Sciences, SciSearch, CompuMath Citation Index, lSI
Alerting Service, Mathematical Reviews, Zentralblatt
MATH, Mathematics Abstracts, DB Math, Springer
Journals Preview Service
Editors-in-Chief / Editorial Offices
I. Gohberg, School of Mathematical Sciences,
Tel Aviv University, Ramat Aviv, Tel Aviv 69978, Israel
Fax +972 3 640 93 57
S. M. Verduyn Lunel, Mathematisch Instill/ut,
Universiteit Leiden
P. O. Box 9512, 2300 RA Leiden, The Netherlands
Fax +31 71 5277101
Back volumes are available.

ISSN 0378-620X (printed edition)
ISSN 1420-8989 (electronic edition)
URL:
http://www.birkhauser.ch/journals/2000/2000_tit.htm

Schemes For Partial Differential Equations

Uploaded by

Copyright:

Available Formats

Schemes For Partial Differential Equations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Schemes For Partial Differential Equations

Uploaded by

Copyright:

Available Formats

Operator Theory: Advances and

S. Goldberg (College Park)

Hebrew University of Jerusalem

2000 Mathematics Subject Classification 39A12, 47B39, 65M06, 65N06, 65Q05

A CIP catalogue record for this book is available from the

ISBN 978-3-0348-7922-4 (eBook)

2004 Springer Basel AG

1 Linear Difference Equations

2 Difference Schemes for First-Order Differential Equations

3 Difference Schemes for Second-Order Differential Equations

4 Partial Differential Equations of Parabolic Type

7 Uniform Difference Schemes for Perturbation Problems

7.1 A Cauchy Problem for Parabolic Equations . . . .

8 Appendix: Delay Parabolic Differential Equations

Comments on the Literature

in an arbitrary Banach space E with positive operator A. The well-posedness of

in a Hilbert space H with positive definite self-adjoint operator A. The stability

and the boundary-value problem

Difference Equations of the First Order

The first-order linear difference equations to be studied in this chapter may be

where <p is a given number.

Chapter 1. Linear Difference Equations

u(k ")= {qk ... qj+1,k=l-j,

From that and (1.3) it follows

= qm+1 Um + epm+! = qm+!u(m,O)ep + 2:qm+1 u (m,j)epj + epm+1

Using formula (1.5), we can write

qm+1u(m,j) = u(m + l,j),j =I- m, u(m + 1, m + 1) = 1.

Um+1 = u(m + 1, O)ep +

2: u(m + l,j)epj + u(m + 1, m + l)epm+1

+ I,O)ep + 2: u(m + l,j)epj.

IUkl :S qklepl + 2: qk- j !epjl

1.2. Difference Equations of the Second Order

The proof of Theorem 1.1.2 is based on formulas (1.4) and (1.5).

Difference Equations of the Second Order

The second-order linear difference equations to be studied in this chapter may be

where <p and 'l/J are given numbers.

+ Q2 U l + f31U n + f32 u n-l = J.Ll,Q3 UO + Q4 U l + f33un + f34un-l = J.L2,

where Ql,Q2,Q3,Q4,f31,f32,f33,f34,J.Ll and J.L2 are given numbers.

Chapter 1. Linear Difference Equations

modified Gauss elimination method. We seek a solution of difference equations

are unknown coefficients. Hence we find

+ 11k = ak(ak+1 U k+l + 11k+1) + 11k

Now putting expressions for

into equation (1.7) we have

+ bk]Uk+1 + [11k +1 (akak -

The last equation is satisfied if we select

From that it follows

Thus using formulas (1.11) and (1.14), we can compute Uk, 0 :S k :S n.

IA11:S 1,IA21 < 1.

Then the requirements (1.15) and

all k are satisfied.

1.3. Difference Equations with Constant Coefficients

Proof. From (1.13) it follows that

lall = JAIl :S 1. Let lakl :S 1. We will prove that

akakl 2: ICkl-lakllakl 2: hl-Iakl 2: Ibkl > o.

From that it follows Ck - akak

lakl :S 1 for all k, 1 :S k :S n.

lanl :S 1 and (1.11) it follows that

Thus Theorem 1.2.1 is proved.