No label defined (Q4947174)

A boundary value problem for an elliptic equation \[ \varepsilon u_{xx}+\varepsilon u_{yy}-a(x,y)u_x-b(x,y)u=f(x,y) \tag{1} \] with a small parameter \(\varepsilon\) at the highest derivatives is considered in a semi-infinite half-strip \(D=\left\{0 \leq x \leq \infty , 0 \leq y \leq 1 \right\}\). At infinity, the following condition is imposed: \[ \lim_{x\to \infty}u(x,y)=0. \tag{2} \] The domain \(D\) is replaced by the bounded rectangular region \(D_L=\left\{0 \leq x \leq L, \;0 \leq y \leq 1 \right \}\) and, on the boundary \( x = L \), a condition is imposed asserting that equation (1) degenerates on the coordinate \( x \): \[ a(x,y)u_x+b(x,y)u-\varepsilon u_{yy}\Big {|}_{x=L}=-f(x,y). \tag{3} \] It is proven that such a translation of the boundary condition from infinity preserves the boundary condition (2) as \(L \rightarrow \infty\). Estimates for the replacement error are obtained. For the problem in the rectangular domain \(D_L\), the difference scheme is constructed on the special nonuniform grid condensing near the boundaries \( y = 0 \) and \( y = 1 \), where the solution can possess boundary layers. It is proven that the finite-difference method under consideration is \(\varepsilon\)-uniform. This part of the article is somewhat related to the article by \textit{G. I. Shishkin} [U.S.S.R. Comput. Math. Math. Phys. 26, No. 4, 38-46 (1986; Zbl 0635.65110); translation from Zh. Vychisl. Mat. Mat. Fiz. 26, No. 7, 1019-1031 (1986; Zbl 0622.65078)]. Numerical results for the equation (1) with \(a = 1, \;b =2 \) are presented.