On a general discrete boundary value problem
Extracted source text ยท arXiv 2304.03926
\documentclass[12pt]{amsart}
\usepackage[cp1251]{inputenc}
\usepackage[english]{babel}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{amsfonts}
\begin{document}
\title[On a general discrete boundary value problem]
{On a General Discrete Boundary Value Problem }
\author[V. Vasilyev]{Vladimir Vasilyev}
\address{Chair of Applied Mathematics and Computer Modeling\\
Belgorod State National Research University\\
Pobedy street 85, Belgorod 308015, Russia}
\email{vbv57@inbox.ru}
\author[A. Vasilyev]{Alexander Vasilyev}
\address{Chair of Applied Mathematics and Computer Modeling\\
Belgorod State National Research University\\
Pobedy street 85, Belgorod 308015, Russia}
\email{alexvassel@gmail.com}
\author[A. Mashinets]{Anastasia Mashinets}
\address{Chair of Applied Mathematics and Computer Modeling\\
Belgorod State National Research University\\
Pobedy street 85, Belgorod 308015, Russia}
\email{anastasia.kho@yandex.ru}
%\urladdr{http://math....}
%\thanks{Received , accepted .}
%\thanks{Research supported under grant number...}
\keywords{elliptic symbol, invertibility, digital pseudo-differential operator, discrete equation, periodic wave factorization, system of integral equation}
\subjclass[2010]{Primary: 35S15; Secondary: 65T50}
%\date{February 26, 2017}
\begin{abstract}
We study a general discrete boundary value problem in Sobolev--Slobodetskii spaces in a plane quadrant and reduce it to a system of integral equations. We show a solvability of the system for a small size of discreteness starting from a solvability of its continuous analogue.
\end{abstract}
\maketitle
%\tableofcontents
\section{Introduction}
The theory of pseudo-differential operators \cite{Ta,Tr} and related equations and boundary value problems \cite{E} exists more than a half-century, and up to now it takes attention of scholars. Also there are some discrete theories of boundary value problems for partial differential equations \cite{S,R}, but these studies are not applicable for pseudo-differential equations. According to this fact the first author has initiated a studying discrete theory of pseudo-differential equations \cite{VV1,V6,V3,V4} having in mind forthcoming studies their approximation properties and applications to computational algorithms \cite{VVT,MVV}.
Since model equations in \cite{E} were studied in a half-space, it is a canonical domain for manifold with smooth boundary, next step was done with a cone, it is a canonical domain for manifold with conical points at boundary \cite{V1}. At this step one needs a special factorization for an elliptic symbol, it was transferred to the discrete case \cite{V5}using analogies with Fourier series \cite{Ed}.
This paper is devoted to a model discrete pseudo-differential equation and discrete boundary value problem in a quadrant in the plane and its solvability starting from solvability their continuous analogues under small parameter of discreteness.
\section{Digital pseudo-differential operators and discrete equations}
Here we introduce some starting concepts and results which will help us moving to statement of a general boundary value problem.
Let $\mathbb Z^2$ be the integer lattice in a plane, $K=\{x\in\mathbb R^2: x=(x_1,x_2), x_1>0, x_2>0\}$ be the first quadrant, $K_d=h\mathbb Z^2\cap K, h>0$, $\mathbb T^2=[-\pi,\pi]^2, \hbar=h^{-1}$. We consider functions of a discrete variable $u_d(\tilde x), \tilde x=(\tilde x_1,\tilde x_2)\in h\mathbb Z^2$.
We use also notations $\zeta^2=h^{-2}((e^{ih\cdot\xi_1}-1)^2+(e^{ih\cdot\xi_2}-1)^2)$ and $S(h\mathbb Z^2)$ for the discrete analogue of the Schwartz space of infinitely differentiable rapidly decreasing at infinity functions.
{\bf Definition 1.} {\it
The space $H^s(h\mathbb Z^2)$ consists of discrete functions and it is a closure of the space $S(h\mathbb Z^2)$ with respect to the norm
\begin{equation}\label{1}
||u_d||_s=\left(\int\limits_{\hbar\mathbb T^2}(1+|\zeta^2|)^s|\tilde u_d(\xi)|^2d\xi\right)^{1/2},
\end{equation}
where $\tilde u_d(\xi)$ denotes the discrete Fourier transform
$$
(F_du_d)(\xi)\equiv\tilde u_d(\xi)=\sum\limits_{\tilde x\in h\mathbb Z^2}e^{i\tilde x\cdot\xi}u_d(\tilde x)h^2,~~~\xi\in\hbar\mathbb T^2.
$$
}
Let $A_d(\xi)$ be a measurable periodic function defined in $\mathbb R^2$ with the basic cube of periods $\hbar\mathbb T^2$.
{\bf Definition 2.} {\it
A digital pseudo-differential operator $A_d$ with the symbol $A_d(\xi)$ in discrete quadrant $K_d$ is called the following operator
\[
(A_du_d)(\tilde x)=\sum\limits_{\tilde y\in h\mathbb Z^2}h^2\int\limits_{\hbar\mathbb T^2} A_d(\xi)e^{i(\tilde y-\tilde x)\cdot\xi}\tilde u_d(\xi)d\xi,~~~\tilde x\in K_d,
\]
}
Here we will consider symbols satisfying the condition
\[
c_1(1+|\zeta^2|)^{\alpha/2}\leq|A_d(\xi)|\leq c_2(1+|\zeta^2|)^{\alpha/2}
\]
with positive constants $c_1, c_2$ non-depending on $h$. This class of symbols satisfying this condition will be denoted by $E_{\alpha}$. The number
$\alpha\in\mathbb R$ is called an order of the digital pseudo-differential operator $A_d$.
We study solvability of the discrete equation
\begin{equation}\label{2}
(A_du_d)(\tilde x)=0,~~~\tilde x\in K_d,
\end{equation}
in the space $H^s(K_d)$, and for this purpose we need certain specific domains of two-dimensional complex space $\mathbb C^2.$ A domain of the type ${\mathcal T}_h(K)=\hbar\mathbb T^2+iK$ is called a tube domain over the quadrant $K$. We will work with holomorphic functions $f(x+i\tau)$ in such domains ${\mathcal T}_h(K)$.
{\bf Definition 3.} {\it
Periodic wave factorization of the symbol
$A_d(\xi)\in E_{\alpha}$ is called its representation in the form
\[
A_d(\xi)=A_{d,\neq}(\xi)A_{d,=}(\xi),
\]
where the factors $A_{d,\neq}(\xi), A_{d,=}(\xi)$ admit holomorphic continuation into tube domains ${\mathcal T}_h(K), {\mathcal T}_h(-K)$ respectively satisfying the estimates
\[
c_1(1+|\hat\zeta^2|)^{\frac{\ae}{2}}\leq|A_{d,\neq}(\xi+i\tau)|\leq c'_1(1+|\hat\zeta^2|)^{\frac{\ae}{2}},
\]
\[
c_2(1+|\hat\zeta^2|)^{\frac{{\alpha-\ae}}{2}}\leq|A_{d,=}(\xi-i\tau)|\leq c'_2(1+|\hat\zeta^2|)^{\frac{{\alpha-\ae}}{2}},
\]
with positive constants $c_1, c'_1, c_2, c'_2$ non-depending on $h$;
\[
\hat\zeta^2\equiv\hbar^2\left((e^{ih(\xi_1+i\tau_1)}-1)^2+(e^{ih(\xi_2+i\tau_2)}-1)^2\right),~~~\xi=(\xi_1,\xi_2)\in\hbar\mathbb T^2,
\]
\[
\tau=(\tau_1,\tau_2)\in K.
\]
The number $\ae\in{\mathbb R}$ is called an index of periodic wave factorization.
}
{\it Everywhere below we assume that we have this periodic wave factorization of the symbol $A_d(\xi)$ with the index $\ae$.}
Using methods developed in \cite{VV1} we can prove the following result.
{\bf Theorem 1.} {\it
Let $\ae-s=n+\delta, n\in\mathbb N, |\delta|<1/2$. Then a general solution of the equation \eqref{2} has the following form
\begin{equation}\label{3}
\tilde u_d(\xi)= A^{-1}_{d,\neq}(\xi)\left(\sum\limits_{k=0}^{n-1}\left(\tilde c_k(\xi_1)\zeta_2^k+\tilde d_k(\xi_2)\zeta_1^k\right)\right),
\end{equation}
where $\tilde c_k(\xi_1), \tilde d_k(\xi_2), k=0,1,\cdots,n-1,$ are arbitrary functions from-- $\widetilde H^{s_k}(h\mathbb T), s_k=s-\ae+k-1/2.$
The a priori estimate
\[
||u_d||_s\leq const\sum\limits_{k=0}^{n-1}([c_k]_{s_k}+[d_k]_{s_k}),
\]
holds, where $[\cdot]_{s_k}$ denotes a norm in the space $H^{s_k}(h\mathbb T)$, and
$const$ doesn't depend on $h$.
}
\section{Discrete boundary value problem}
\subsection{Statement and solvability}
Starting from Theorem 1 we introduce the following boundary conditions
\begin{equation}\label{4}
\begin{aligned}
&(B_{d,j}u_d)(\tilde x_1,0)=b_{d,j}(\tilde x_1),
\\[3pt]
&(G_{d,j}u_d)(0,\tilde x_2)=g_{d,j}(\tilde x_2),~~~j=0,1.\cdots,n-1,
\end{aligned}
\end{equation}
where $B_{d,j}, G_{d,j}$ are digital pseudo-differential operators of order $\beta_j,\gamma_j\in{\mathbb R}$ with symbols $\widetilde B_{d,j}(\xi)\in E_{\beta_j}, \widetilde G_{d,j}(\xi)\in E_{\gamma_j}$
\[
\begin{aligned}
&(B_{d,j}u_d)(\tilde x)=\frac{1}{(2\pi)^2}\int\limits_{\hbar{\bf T}^2}\sum\limits_{\tilde y\in h{\bf Z}^2}e^{i\xi\cdot(\tilde x-\tilde y)}\widetilde B_{d,j}(\xi)\tilde u_d(\xi)d\xi,
\\[3pt]
&(G_{d,j}u_d)(\tilde x)=\frac{1}{(2\pi)^2}\int\limits_{\hbar{\bf T}^2}\sum\limits_{\tilde y\in h{\bf Z}^2}e^{i\xi\cdot(\tilde x-\tilde y)}\widetilde G_{d,j}(\xi)\tilde u_d(\xi)d\xi.
\end{aligned}
\]
One can rewrite boundary conditions
\eqref{4}
in Fourier images
\begin{equation}\label{5}
\begin{aligned}
&\int\limits_{-\hbar\pi}^{\hbar\pi}\widetilde B_{d,j}(\xi_1,\xi_2)\tilde u_d(\xi_1,\xi_2)d\xi_2=\tilde b_{d,j}(\xi_1),
\\[3pt]
&\int\limits_{-\hbar\pi}^{\hbar\pi}\widetilde G_{d,j}(\xi_1,\xi_2)\tilde u_d(\xi_1,\xi_2)d\xi_1=\tilde g_{d,j}(\xi_2),~~~j=0,1.\cdots,n-1,
\end{aligned}
\end{equation}
so that according to properties of digital pseudo-differential operators and trace properties we need to require $b_{d,j}(\tilde x_1)\in H^{s-\beta_j-1/2}(h{\mathbb Z}), g_{d,j}(\tilde x_2)\in H^{s-\gamma_j-1/2}(h{\mathbb Z})$.
Multiplying the equality \eqref{3} by $\widetilde B_{d,j}(\xi_1,\xi_2)$ and $\widetilde G_{d,j}(\xi_1,\xi_2)$, integrating over $[-\hbar\pi,\hbar\pi]$ on $\xi_2$ and $\xi_1$, taking into account the conditions \eqref{5} we obtain the following $(2n\times 2n)$-system of linear integral equations
\begin{equation}\label{6}
\begin{aligned}
&\sum\limits_{k=0}^{n-1}\left(r_{jk}(\xi_1)\tilde c_k(\xi_1)+\int\limits_{-\hbar\pi}^{\hbar\pi}l_{jk}(\xi_1,\xi_2)\tilde d_k(\xi_2)d\xi_2\right)=\tilde b_{d,j}(\xi_1)
\\[3pt]
&\sum\limits_{k=0}^{n-1}\left(\int\limits_{-\hbar\pi}^{\hbar\pi}m_{jk}(\xi_1,\xi_2)\tilde c_k(\xi_1)d\xi_1+p_{jk}(\xi_2)\tilde d_k(\xi_2)\right)=\tilde g_{d,j}(\xi_2),
\\[3pt]
&j=0,1,\dots,n-1,
\end{aligned}
\end{equation}
with unknowns functions $\tilde c_k, \tilde d_k, k=0,1,\dots,n-1$. We have used the following notations
\[
r_{jk}(\xi_1)=\int\limits_{-\hbar\pi}^{\hbar\pi}\tilde B_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)\zeta_2^kd\xi_2,~~~p_{jk}(\xi_2)=\int\limits_{-\hbar\pi}^{\hbar\pi}\tilde G_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)\zeta_1^kd\xi_1,
\]
\[
l_{jk}(\xi_1,\xi_2)=\tilde B_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)\zeta_1^k,~~~m_{jk}(\xi_1,\xi_2)=\tilde G_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)\zeta_2^k,
\]
$j,k=0,1,\dots,n-1$.
Thus, we can formulate the following assertion.
{\bf Theorem 2.} {\it
The boundary value problem \eqref{2},\eqref{4} is uniquely solvable in the space $H^s(K_d)$ with data $b_{d,j}\in H^{s-\beta_j-1/2}(h\mathbb Z_+, g_{d,j}\in H^{s-\gamma_j-1/2}(h\mathbb Z_+$ if and only if the system \eqref{6} has the unique solution $\tilde c_k,\tilde d_k\in\tilde H^{s_k}(\hbar\mathbb T)$, $j,k=0,1,\dots,n-1$.
}
\subsection{Continuous case}
Here we will describe continuous boundary value problem which is related to considered discrete boundary value problem \eqref{2},\eqref{4}.
Let $A$ be a pseudo-differential operator
\[
(Au)(x)=\int\limits_{\mathbb R^2}\int\limits_{\mathbb R^2}\tilde A(\xi)e^{i\xi(y-x)}u(y)dyd\xi
\]
with symbol $\tilde A(\xi)$ satisfying the condition
\begin{equation}\label{7}
|\tilde A(\xi)|\sim(1+|\xi|)^{\alpha}
\end{equation}
and admitting the wave factorization with respect to $K$
\[
\tilde A(\xi)=A_{\neq}(\xi)\cdot A_=(\xi).
\]
with index $\ae$ such that $\ae-s=n+\delta, n\in\mathbb N, |\delta|<1/2$.
Further, let $B_j,G_j, j=0,1,\dots,n-1$ be pseudo-differential operators with symbols $\tilde B_j(\xi),\tilde G_j(\xi)$ satisfying the condition \eqref{7} with $\beta_j,\gamma_j$ instead of $\alpha$.
The following boundary value problem
\begin{equation}\label{8}
\begin{aligned}
&(Au)(x)=0,~~~x\in K,
\\[3pt]
&(B_ju)(x_1,0)=b_j(x_1),
\\[3pt]
&(G_ju)(0,x_2)=g_j(x_2),~~~j=0,1,\dots,n-1
\end{aligned}
\end{equation}
is a continuous analogue of the discrete boundary value problem \eqref{2},\eqref{4}. It was shown in \cite{V1} the problem \eqref{8} is equivalent to the following system of integral equations
\begin{equation}\label{9}
\begin{aligned}
&\sum\limits_{k=0}^{n-1}\left(R_{jk}(\xi_1)\tilde C_k(\xi_1)+\int\limits_{-\infty}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2\right)=\tilde b_{j}(\xi_1)
\\[3pt]
&\sum\limits_{k=0}^{n-1}\left(\int\limits_{-\infty}^{+\infty}M_{jk}(\xi_1,\xi_2)\tilde C_k(\xi_1)d\xi_1+P_{jk}(\xi_2)\tilde D_k(\xi_2)\right)=\tilde g_{j}(\xi_2),
\\[3pt]
&j=0,1,\dots,n-1,
\end{aligned}
\end{equation}
with unknowns functions $\tilde C_k, \tilde D_k, k=0,1,\dots,n-1$. The following notations
are used
\[
R_{jk}(\xi_1)=\int\limits_{-\infty}^{+\infty}\tilde B_{j}(\xi)A_{\neq}^{-1}(\xi)(i\xi_2)^kd\xi_2,~~~P_{jk}(\xi_2)=\int\limits_{-\infty}^{+\infty}\tilde G_{j}(\xi)A_{\neq}^{-1}(\xi)(i\xi_1)^kd\xi_1,
\]
\[
L_{jk}(\xi_1,\xi_2)=\tilde B_{j}(\xi)A_{\neq}^{-1}(\xi)(i\xi_1)^k,~~~M_{jk}(\xi_1,\xi_2)=\tilde G_{j}(\xi)A_{\neq}^{-1}(\xi)(i\xi_2)^k,
\]
$j,k=0,1,\dots,n-1$. If we can solve the system \eqref{9} and find $\tilde C_k, \tilde D_k, k=0,1,\dots,n-1$ the the solution of the boundary value problem \eqref{9} can be constructed by the formula \cite{V1}
\begin{equation}\label{10}
\tilde u(\xi)= A^{-1}_{\neq}(\xi)\left(\sum\limits_{k=0}^{n-1}\left(\tilde C_k(\xi_1)(i\xi_2)^k+\tilde D_k(\xi_2)(i\xi_1)^k\right)\right),
\end{equation}
where $\tilde C_k(\xi_1), \tilde D_k(\xi_2), k=0,1,\cdots,n-1,$ are arbitrary functions from $\widetilde H^{s_k}(\mathbb R), s_k=s-\ae+k-1/2$.
Our next problems are the following. Given operator $A$ and boundary operators $B_j,G_j$ how to choice the digital operators $A_d$ and $B_{d,j},G_{d,j}$ to obtain the implication: the unique solvability of the system \eqref{9} gives the unique solvability of the system \eqref{6} for enough small $h$. This question will be discussed in the next section.
\section{Comparison theorems}
\subsection{Projection method}
Let us introduce the following space of vector-functions:
\[
\tilde{\bf H}^{\Lambda}(\mathbb R)=\tilde{\bf H}^S(\mathbb R)\oplus\tilde{\bf H}^S(\mathbb R),~~~\tilde{\bf H}^S(\mathbb R)=\oplus\sum\limits_{k=0}^{n-1}\tilde{H}^{s_k}(\mathbb R),
\]
Norms in these spaces will be defined in the following way. For $f\in\tilde{\bf H}^{S}(\mathbb R), f=(f_0,\dots,f_{n-1}), f_k\in\tilde{H}^{s_k}(\mathbb R), g\in\tilde{\bf H}^{S}(\mathbb R), g=(g_0,\dots,g_{n-1}), g_k\in\tilde{H}^{s_k}(\mathbb R)$ we put
\[
||f||_{S}=\sum\limits_{k=0}^{n-1}||f_k||_{s_k}.~~~||g||_{S}=\sum\limits_{k=0}^{n-1}||g_k||_{s_k},
\]
and if $F\in\tilde{\bf H}^{\Lambda}(\mathbb R), F=(f,g), f\in\tilde{\bf H}^{S}(\mathbb R), g\in\tilde{\bf H}^{S}(\mathbb R)$ we put
\[
||F||_{\Lambda}=||f||_{S}+||g||_{S}.
\]
%A norm in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$ is defined in the same way.
Let us introduce the following notations. We denote the system \eqref{9} in the following way
\[
\begin{pmatrix}
R&L\\
M&P
\end{pmatrix}
\begin{pmatrix}
C\\
D
\end{pmatrix}=\begin{pmatrix}
B\\
G
\end{pmatrix},
\]
where $C=(\tilde c_0,\dots,\tilde c_{n-1})^T, D=(\tilde d_0,\dots,\tilde d_{n-1})^T, B=(\tilde b_0,\dots,\tilde b_{n-1})^T, G=(\tilde g_0,\dots,\tilde g_{n-1})^T$; operators $R,L,M,P$ acting in the space $\tilde{\bf H}^S(\mathbb R)$ are the following: $R$ is multiplier by the matrix-function $(r_{jk})_{j,k=0}^{n-1}$, $P$ is multiplier by the matrix-function $(p_{jk})_{j,k=0}^{n-1}$, $L$, $M$ are matrix integral operators with kernels $L_{jk}, M_{jk}$ respectively.
Further, we will denote $\Xi_h$ the restriction operator on the segment $\hbar\mathbb T$ so that for $f\in\tilde{\bf H}^S(\mathbb R), f=(f_0,\dots,f_{n-1})$ the notation $\Xi_h f$ means the following
\[
\Xi_h f=(\chi_h f_0,\dots,\chi_h f_{n-1}),
\]
where $\chi_h$ is an indicator of $\hbar\mathbb T$.
We denote by $Q$ the operator
\[
Q=\begin{pmatrix}
R&L\\
M&P
\end{pmatrix}
\]
{\bf Theorem 3.} {\it Let $s-\beta_j>1, s-\gamma_j>2, j=0,1,\dots,n-1$.
We have the estimate
\[
||\Xi_hQ-Q\Xi_h||_{\tilde{\bf H}^{\Lambda}(\mathbb R)\rightarrow\tilde{\bf H}^{\Lambda}(\mathbb R)}\leq~const~ h^{\varepsilon},
\]
where
$$
\varepsilon =\min\limits_{0\leq j\leq n-1}\{s-\beta_j-1,s-\gamma_j-1\},
$$
const does not depend on $h, s_k=s-\ae+k-1/2, k=0,1,\dots,n-1$.
}
\begin{proof}
Obviously, the matrices $R,P$ give vanishing result in the norm, and we need to work with integral operators only. Let us consider the operator $L$, and extract one its component $L_{jk}$,
\[
\int\limits_{-\infty}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2,~~~L_{jk}(\xi_1,\xi_2)=\tilde B_{j}(\xi)A_{\neq}^{-1}(\xi)\xi_1^k.
\]
We have
\[
\chi_h(\xi_1)\int\limits_{-\infty}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2-
\int\limits_{-\bar\pi}^{+\hbar\pi}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2
\]
\[
=\begin{cases}
\left(\int\limits_{-\infty}^{-\hbar\pi}+\int\limits_{\hbar\pi}^{+\infty}\right)L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2,~~~\xi_1\in\hbar\mathbb T,\\
-\int\limits_{-\hbar\pi}^{\hbar\pi}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2,~~~\xi_1\notin\hbar\mathbb T.
\end{cases}
\]
Let us consider the first case and estimate
\[
\left|\int\limits_{\hbar\pi}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2\right|\leq~const\int\limits_{\hbar\pi}^{+\infty}(1+|\xi|)^{\beta_j-\ae}|\xi_1|^k|\tilde D_k(\xi_2)|d\xi_2
\]
\[
\leq~const\int\limits_{\hbar\pi}^{+\infty}(1+|\xi|)^{\beta_j-s+1/2}|\tilde D_k(\xi_2)|(1+|\xi_2|)^{s_k}d\xi_2
\]
(we have taken into account $s_k=s-\ae+k-1/2$ and now we apply the Cauchy--Schwartz inequality)
\[
\leq~const(1+|\xi_1|+\hbar)^{\beta_j-s+1}||\tilde D_k||_{s_k}\leq~const~h^{s-\beta_j-1}||D_k||_{s_k}.
\]
Squaring the latter inequality, multiplying by $(1+|\xi_|)^{2s_k}$ and integrating over $\hbar\mathbb T$ we obtain
\[
\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi_|)^{2s_k}\left|\int\limits_{\hbar\pi}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2\right|^2d\xi_1
\]
\[
\leq~const~h^{2(s-\beta_j-1)}||D_k||^2_{s_k}\int\limits_{0}^{+\infty}(1+|\xi_|)^{2s_k}d\xi_1\leq~const~h^{2(s-\beta_j-1)}||D_k||^2_{s_k}
\]
For the second case ($|\xi_1|>\hbar\pi$) we obtain
\[
\left|\int\limits_{-\hbar\pi}^{+\hbar\pi}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2\right|\leq~const\int\limits_{-\hbar\pi}^{+\hbar\pi}(1+|\xi|)^{\beta_j-\ae}|\xi_1|^k|\tilde D_k(\xi_2)|d\xi_2
\]
\[
\leq~const\int\limits_{-\hbar\pi}^{+\hbar\pi}(1+|\xi|)^{\beta_j-\ae}|\xi_1|^{n-1}(1+|\xi_2|)^{-s_k}|\tilde D_k(\xi_2)|(1+|\xi_2|)^{s_k}d\xi_2
\]
\[
\leq~const~|\xi_1|^{n-1}(1+|\xi_1|)^{-s_k}\int\limits_{-\hbar\pi}^{+\hbar\pi}(1+|\xi|)^{\beta_j-\ae}|\tilde D_k(\xi_2)|(1+|\xi_2|)^{s_k}d\xi_2
\]
(we apply the Cauchy--Schwartz inequality in the integral)
\[
\leq~const~(1+|\xi_1|)^{n-s_k-1}(1+|\xi_1|)^{\beta_j-\ae+1/2k}||D_k||_{s_k}
\]
Squaring the latter inequality, multiplying by $(1+|\xi_|)^{2s_k}$ and integrating over $\mathbb R\setminus\hbar\mathbb T$ we obtain
\[
\left(\int\limits_{-\infty}^{-\hbar\pi}+\int\limits_{\hbar\pi}^{+\infty}\right)(1+|\xi_|)^{2s_k}\left|\int\limits_{\hbar\pi}^{+\infty}L_{jk}(\xi_1,\xi_2)\tilde D_k(\xi_2)d\xi_2\right|^2d\xi_1
\]
\[
\leq~const||D_k||^2_{s_k}\int\limits_{\hbar\pi}^{+\infty}(1+\xi_1)^{2n-2+2\beta_j+1-2\ae}d\xi_1\leq~const||D_k||^2_{s_k}h^{2s-2\beta_j+2\delta},
\]
since $2n+2\beta_j-2\ae=2n+2\beta_j-2(s+n+\delta)=2\beta_j-2s-2\delta<0$.
Thus, we have proved that
\[
||\chi_hL_{jk}-L_{jk}\chi_h||_{H^{s_k}(\mathbb R)\rightarrow H^{s_k}(\mathbb R)}\leq~const~h^{s-\beta_j-1},
\]
since $s-\beta_j-1<s-\beta_j+\delta$.
Almost the same inequality can be obtained for $M_{jk}$
\[
||\chi_hM_{jk}-M_{jk}\chi_h||_{H^{s_k}(\mathbb R)\rightarrow H^{s_k}(\mathbb R)}\leq~const~h^{s-\gamma_j-1},
\]
These estimates complete the proof.
\end{proof}
{\bf Corollary 1.} {\it Under conditions of Theorem 3 the invertibility of the operator $Q$ in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$ implies the invertibility of the operator $\Xi_hQ\Xi_h$ in the space $\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)$ for enough small $h$.
}
\begin{proof}
We apply the results of the paper \cite{K} which imply the following. If
\[
||\Xi_hQ-Q\Xi_h||_{\tilde{\bf H}^{\Lambda}(\mathbb R)\rightarrow\tilde{\bf H}^{\Lambda}(\mathbb R)}\rightarrow 0,~~~h\to 0
\]
then the equation in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$
\begin{equation}\label{11}
Qu=v
\end{equation}
admits applying so called {\it projection method}. In other words it means that unique solvability of the equation \eqref{11} in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$ implies unique solvability of the equation
\begin{equation}\label{12}
\Xi_hQ\Xi_hu=\Xi_hv
\end{equation}
in the space $\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)$ for enough small $h$. Moreover, if there is bounded operator $Q^{-1}$ in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$ then there is bounded operator $(\Xi_hQ\Xi_h)^{-1}$ for enough small $h$ and
\[
||(\Xi_hQ\Xi_h)^{-1}||_{\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)\rightarrow\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)}\leq~const,
\]
where const does not depend on $h$.
Indeed, a reader can easily verify that
\[
||(\Xi_hQ\Xi_h)^{-1}-\Xi_hQ^{-1}\Xi_h||_{\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)\rightarrow\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)}\rightarrow 0,~~~h\to 0.
\]
\end{proof}
\subsection{Discrete and continuous}
To compare discrete and continuous operators we need a special choice of discrete operators. We will do it in the following way.
The symbol $A_d(\xi)$ of the discrete operator $A_d$ will be constructed as follows. Given wave factorization for $\tilde A(\xi)$
\[
\tilde A(\xi)=A_{\neq}(\xi)\cdot A_=(\xi)
\]
we take restrictions of factors $A_{\neq}(\xi), A_=(\xi)$ on $\hbar\mathbb T^2$ and periodically continue them into $\mathbb R^2$. We denote these elements by $A_{d,\neq}(\xi), A_{d,=}(\xi)$ and construct the periodic symbol $A_d(\xi)$ which admits periodic wave factorization with respect to $K$
\[
A_d(\xi)=A_{d,\neq}(\xi)\cdot A_{d,=}(\xi)
\]
with the same index $\ae$. We construct discrete pseudo-differential operators $B_{d,j}, G_{d,j}$ taking their symbol as restrictions of symbols $\tilde B_j(\xi), \tilde G_j(\xi)$ on $\hbar\mathbb T^2$ with periodical continuations into $\mathbb R^2,, j=0,1,\dots,n-1$. The discrete boundary functions $b_{d,j}$, $g_{d,j}$ are constructed in the same way.
Thus, we have the corresponding discrete boundary value problem \eqref{2},\eqref{4}.
{\bf Lemma 1.} {\it The estimate
\[
|(i\xi_m)^k-\zeta_m^k|\leq~const~h|\xi_m|^{k+1}
\]
holds for $\xi_m\in\hbar\mathbb T, m=1,2$ const does not depend on $h$.
}
\begin{proof}
First, we estimate
\[
|\zeta_1|=\left|\sum\limits_{\nu=1}^{\infty}\frac{(i\xi_1)^{\nu+1}h^{\nu}}{(\nu+1)!}\right|=|\xi_1|\left|\sum\limits_{\nu=0}^{\infty}\frac{(i\xi_1)^{\nu}h^{m}}{(\nu+1)!}\right|
\leq|\xi_1|\sum\limits_{\nu=0}^{\infty}\frac{(|\xi_1|h)^{\nu}}{\nu!}
\]
\[
=|\xi_1|e^{|\xi_1|h}\leq|\xi_1|e^{\pi}.
\]
Second,
\[
|\zeta_1-i\xi_1|=|\hbar(e^{i\xi_1h}-1)-i\xi_1|=\left|\sum\limits_{\nu=1}^{\infty}\frac{(i\xi_1)^{\nu+1}h^{\nu}}{(\nu+1)!}\right|
\]
\[
\leq|\xi_1|^2h\sum\limits_{\nu=0}^{\infty}\frac{|\xi_1|^{\nu}h^{\nu}}{\nu!}=|\xi_1|^2he^{|\xi_1|h}\leq|\xi_1|^2he^{\pi}.
\]
We have
\[
\zeta_1^k-(i\xi_1)^k=(\zeta_1-i\xi_1)\left(\sum\limits_{\nu=0}^{k-1}\zeta_1^\nu(i\xi_1)^{k-1-\nu}\right),
\]
and thus
\[
|\zeta_1^k-(i\xi_1)^k|\leq|\zeta_1-i\xi_1|\sum\limits_{\nu=0}^{k-1}|\zeta_1|^\nu|\xi_1|^{k-1-\nu}
\]
Applying above estimates we obtain required inequality.
\end{proof}
{\bf Lemma 2.} {\it Let $s-\beta_j>2, s-\gamma_j>2, j=0,1,\dots,n-1$.The following estimates
\[
\begin{aligned}
&|L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2)|\leq~const~h(1+|\xi|)^{\beta_j-\ae+k+1},
\\[3pt]
&|M_{jk}(\xi_1,\xi_2)-m_{jk}(\xi_1,\xi_2)|\leq~const~h(1+|\xi|)^{\gamma_j-\ae+k+1},
\\[3pt]
&|R_{jk}(\xi_1)-r_{jk}(\xi_1)|\leq~const~h(1+|\xi_1|)^{\beta_j-\ae+k+2},
\\[3pt]
&|P_{jk}(\xi_2)-p_{jk}(\xi_2)|\leq~const~h(1+|\xi_1|)^{\gamma_j-\ae+k+2}
\end{aligned}
\]
hold for $\xi_1,\xi_2\in\hbar\mathbb T$.
}
\begin{proof}
According to above conventions we have for $\xi\in\hbar\mathbb T^2$ and using Lemma 1 we have
\[
|L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2)|=|\tilde B_j(\xi)A_{\neq}^{-1}(\xi)-\tilde B_{d,j}\xi)A_{d,\neq}^{-1}(\xi)||B_j(\xi)||\xi_1^k-\zeta_1^k|
\]
\[
\leq~const~(1+|\xi|)^{\beta_j-\ae}h|\xi_1|^{k+1}\leq~const~h(1+|\xi|)^{\beta_j-\ae+k+1}.
\]
Further,
\[
|R_{jk}(\xi_1)-r_{jk}(\xi_1)|=\left|\int\limits_{-\infty}^{+\infty}\tilde B_j(\xi)A_{\neq}^{-1}(\xi)\xi_2^kd\xi_2-\int\limits_{-\hbar\pi}^{+\hbar\pi}\tilde B_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)\zeta_2^kd\xi_2\right|
\]
\[
\leq\int\limits_{-\hbar\pi}^{+\hbar\pi}|\tilde B_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)||\xi_2^k-\zeta_2^k|d\xi_2+\left(\int\limits_{-\infty}^{-\hbar\pi}+\int\limits_{\hbar\pi}^{+\infty}\right)
|\tilde B_j(\xi)A_{\neq}^{-1}(\xi)\xi_2^k|d\xi_2.
\]
For the first integral we have
\[
\int\limits_{-\hbar\pi}^{+\hbar\pi}|\tilde B_{d,j}(\xi)A_{d,\neq}^{-1}(\xi)||\xi_2^k-\zeta_2^k|d\xi_2\leq~const~h\int\limits_{-\hbar\pi}^{+\hbar\pi}(1+|\xi|)^{\beta_j-\ae+k+1}d\xi_2
\]
\[
\leq~const~h(1+|\xi_1|)^{\beta_j-\ae+k+2},
\]
since $\beta_j-\ae+k+2<0, s-\beta_j>2.$
The second summand
\[
\left|\left(\int\limits_{-\infty}^{-\hbar\pi}+\int\limits_{\hbar\pi}^{+\infty}\right)
|\tilde B_j(\xi)A_{\neq}^{-1}(\xi)\xi_2^k|d\xi_2\right|\leq~const\int\limits_{\hbar\pi}^{+\infty}(1+|\xi_1|+|\xi_2|)^{\beta_j-\ae+k}d\xi_2
\]
\[
\leq~const(1+|\xi_1|+\hbar)^{\beta_j-\ae+k+1}\leq~const~h(1+|\xi_1|)^{\beta_j-\ae+k+2}.
\]
The same estimates are valid for $M_{jk}-m_{jk}$ and $P_{jk}$ with $\gamma_j$ instead of $\beta_j$.
\end{proof}
We introduce similar notations for the system \eqref{6} so that this system takes the form
\[
\begin{pmatrix}
r&l\\
m&p
\end{pmatrix}
\begin{pmatrix}
c\\
d
\end{pmatrix}=\begin{pmatrix}
B_d\\
G_d
\end{pmatrix},
\]
where
\[
q=\begin{pmatrix}
r&l\\
m&p
\end{pmatrix}
\]
is linear bounded operator acting in the space $\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)$.
{\bf Theorem 4.} {\it Let $s-\beta_j>3, s-\gamma_j>3, j=0,1,\dots,n-1$. A comparison between operators $Q$ and $q$ is given by the estimate
\[
||\Xi_hQ\Xi_h-q||_{\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)\rightarrow\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)}\leq~const!h,
\]
where const does not depend on $h$.
}
\begin{proof}
We need to estimate $H^{s_k}(\hbar\mathbb T$-norms of the following elements
\[
(R_{jk}(\xi_1)-r_{jk}(\xi_1))f(\xi_1),~~~(R_{jk}(\xi_2)-p_{jk}(\xi_2))f(\xi_2),
\]
\[
\int\limits_{-\hbar\pi}^{\hbar\pi}(L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2))f(\xi_2)d\xi_2,
\]
\[
\int\limits_{-\hbar\pi}^{\hbar\pi}(M_{jk}(\xi_1,\xi_2)-m_{jk}(\xi_1,\xi_2))f(\xi_1d\xi_1.
\]
We have according to Lemma 2
\[
|(R_{jk}(\xi_1)-r_{jk}(\xi_1))f(\xi_1)|\leq~const~h(1+|\xi_1|)^{\beta_j-\ae+k+2}|f(\xi_1)|.
\]
%\[
%=const~h(1+|\xi_1|)^{\beta_j-s+5/2}|f(\xi_1)|(1+|\xi_1|)^{s_k}
%\]
%since $\beta_j-\ae+k+2-s_k=-\beta_j-s+5/2$.
Multiplying the latter inequality by $(1+|\xi_1|)^{s_k}$, squaring, integrating over $\hbar\mathbb T$ and applying the Cauchy--Schwartz inequality we obtain
\[
\int\limits_{-\hbar\pi}^{+\hbar\pi}(1+|\xi_1|)^{2s_k}|R_{jk}(\xi_1)-r_{jk}(\xi_1)|^2|f(\xi_1)|^2d\xi_1
\leq~const~h^2||f||^2_{s_k}
\]
since $\beta_j-\ae+k+2<0$.
Let us consider
\[
\int\limits_{-\hbar\pi}^{\hbar\pi}(L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2))f(\xi_2)d\xi_2.
\]
Using Lemma 2 we have
\[
\left|\int\limits_{-\hbar\pi}^{\hbar\pi}(L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2))f(\xi_2)d\xi_2\right|
\]
\[
\leq~const~h\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi|)^{\beta_j-\ae+k+1}|f(\xi_2)|d\xi_2
\]
\[
\leq~const~h\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi|)^{\beta_j-s+5/2}|f(\xi_2)|(1+|\xi_2|)^{s_k}d\xi_2
\]
since $\beta_j-\ae+k+2-s_k=-\beta_j-s+5/2$. Now applying Cauchy--Schwartz inequality we find
\[
\left|\int\limits_{-\hbar\pi}^{\hbar\pi}(L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2))f(\xi_2)d\xi_2\right|
\]
\[
\leq~const~h||f||_{s_k}\left(\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi_1|+|\xi_2|)^{2\beta_j-2s+5}d\xi_2\right)^{1/2}
\]
\[
\leq~const~h||f||_{s_k}(1+|\xi_1|)^{\beta_j-s+3}
\]
according to the condition $s-\beta_j>3$. Squaring, multiplying by $(1+|\xi_1|)^{2s_k}$ and integrating over $\hbar\mathbb T$ we conclude
\[
\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi_1|)^{2s_k}\left|\int\limits_{-\hbar\pi}^{\hbar\pi}
(L_{jk}(\xi_1,\xi_2)-l_{jk}(\xi_1,\xi_2))f(\xi_2)d\xi_2\right|^2d\xi_1
\]
\[
\leq~const~h^2||f||^2_{s_k}\int\limits_{-\hbar\pi}^{\hbar\pi}(1+|\xi_1|)^{2(\beta_j-s+3+s_k)}d\xi_1\leq~const~h^2||f||^2_{s_k}
\]
since $2(\beta_j-s+3+s_k)<-1$. Indeed, $2(\beta_j-s+3+s_k)=2(\beta_j-s+3+s-\ae+k-1/2)=2(\beta_j-s-\delta).$ Obviously, the inequality $2(\beta_j-s-\delta)<-1$ is equivalent to $s-\beta_j>-1-\delta$.
\end{proof}
{\bf Corollary 2.} {\it Under conditions of Theorem 4 the invertibility of the operator $Q$ in the space $\tilde{\bf H}^{\Lambda}(\mathbb R)$ implies the invertibility of the operator $q$ in the space $\tilde{\bf H}^{\Lambda}(\hbar\mathbb T)$ for enough small $h$.
}
\begin{proof}
Indeed, we have the invertibility of $\Xi_hQ\Xi_h$ by Corollary 1 and the invertibility of $q$ is obtained by Theorem 4.
\end{proof}
\section*{Conclusion}
Main goal of the paper was to prove unique solvability of discrete boundary value problem having in main unique solvability of its continuous analogue. It was done by a special choice od discrete operator and discrete boundary conditions. We hope that estimates of Theorem 3,4 will help us to obtain some estimates for discrete and continuous solutions.
\begin{thebibliography}{99}
\bibitem{Ta} M Taylor, Pseudodifferential operators, Princeton University Press, Princeton,
1981.
\bibitem{Tr} F Treves, Introduction to pseudodifferential operators and Fourier integral op-
erators, Springer, New York, 1980.
\bibitem{E} G Eskin, Boundary value problems for elliptic pseudodifferential equations,
AMS, Providence, 1981.
\bibitem{V6} V.B. Vasilyev, Operators and equations: discrete and continuous, J. Math. Sci. 257 (2021), no. 1, 17--26.
\bibitem{Ed} R Edwards, Fourier series. A modern introduction, Springer--Verlag, Heidelberg,
1982.
\bibitem{S} A Samarskii, The theory of difference schemes, CRC Press, Boca Raton, 2001.
\bibitem{R} V Ryaben'kii, Method of difference potentials and its applications, Springer--
Verlag, Berlin--Heidelberg, 2002.
\bibitem{VVT} A.V. Vasilyev, V.B. Vasilyev, O.A. Tarasova, Discrete boundary value problems as approximate condtructions, Lobachevskii J. Math. 43 (2022), no.6, 1446--1457.
\bibitem{K} A.V. Kozak,
On a projection method for solving operator equations in Banach space, .
Sov. Math. Dokl. 14 (1973), 1159--1162.
\bibitem{V1} V. B Vasil'ev, Wave factorization of elliptic symbols: Theory and applications,
Kluwer Academic Publishers, Dordrecht--Boston--London, 2000.
\bibitem{MVV} A.A. Mashinets, A.V. Vasilyev, V. B. Vasilyev, On discrete Neumann problem in a quadrant, Lobachevskii J. Math, 44 (2023), no. 3, 1011--1021.
\bibitem{VV1} A. V. Vasilyev and V. B Vasilyev, Pseudo-differential operators and equations
in a discrete half-space, Math. Model. Anal. 23 (2018), no. 3, 492--506.
\bibitem{V3} V. B. Vasilyev, Discreteness, periodicity, holomorphy, and factorization, Integral
Methods in Science and Engineering (New York) (C. Constanda, M. Dalla
Riva, P.D. Lamberti, and P. Musolino, eds.), Theoretical Technique, vol. 1,
Birkhauser, 2017, pp. 315--324.
\bibitem{V4} V. B. Vasilyev, On discrete boundary value problems, Proceedings of the International
Conference on Functional Analysis in Interdisciplinary Applications" (FAIA2017)
(Melville) (T. Kal'menov and M. Sadybekov, eds.), AIP Conf. Proc., vol. 1880,
AIP Publishing, 2017, p. 050010.
\bibitem{V5} V. B. Vasilyev The periodic Cauchy kernel, the periodic Bochner kernel, discrete
pseudo-differential operators, Proceedings of the International Conference on
Numerical Analysis and Applications (ICNAAM-2016) (Melville) (T. Simos
and C. Tsitouras, eds.), AIP Conf. Proc., vol. 1863, AIP Publishing, 2017,
p. 140014.
\end{thebibliography}
%\vspace{5cm}
\end{document}