Well-posedness of parabolic difference equations. Transl. from the Russian by A. Iacob (Q1324804)

This is a well written research monograph. The main theme of the book is the construction of highly accurate difference schemes for parabolic initial-boundary value problems. The basic tool used to do this is the so-called Padé approximation. Moreover the authors abundantly use the results of a modern theory of interpolation of linear operators . As for Padé approximation, it uses rational functions \(R_{j,l} (z) = P_{j,l} (z)/a_{j,l} (z)\) which, under a normalization assumption for the polynomial \(Q_{j,l}\) contain \(j + \ell + 1\) independent parameters to be determined in a given situation. From the numerical point of view (but, of course, not only) convergence properties and especially the rate of convergence are of paramount importance. The most interesting for applications is the so-called coercive stability, because it allows to obtain sharp, two-sided estimates of the rate of convergence. It is worth to emphasize that the existence of coercivity inequalities is equivalent to the natural notion of well-posedness of the difference problems. In Chapter 1 the abstract Cauchy problem (1) \(v'(t) + Av(t) = f(t)\), \(0 \leq t \leq 1\), \(v(0) = v_ 0\) is treated in an arbitrary Banach space \(E\). This means: \(v_ 0\) is a given element of \(E\), \(A\) is a linear operator acting in \(E\) with domain \(D(A) \subset E\) and a solution \(v\) is sought in the space \(C(E) = C([0,1],E)\) of continuous functions \([0,1] \to E\). A definition of solution is natural: \(v(t)\) must be continuously differentiable, \(v(t)\) belongs to \(D(A)\) for \(t \in [0,1]\) and the function \(Av(t)\) is continuous in \([0,1]\) (but generally \(A\) may be not). If \(A \in L(E)\), then one can easily derive that \((*)\) \(\| v(\cdot, f, v_ 0) \|_{C(E)} \leq M (\| f \|_{C(E)} + \| v_ 0 \|_ E)\), where: \(\| \varphi \|_{C(E)} : = \sup \{\| \varphi (t) \|_ E : t \in [0,1]\}\), \(v (\cdot, f, v_ 0)\) is a solution (the only one) of (1) with right hand \(f \in C(E)\) and initial condition \(v_ 0 \in D (A)\) (here \(D(A) = E)\). These considerations give rise to a natural definition of well-posedness for the Cauchy problem (1) in \(C(E)\): (i) \(v(\cdot, f, v_ 0)\) exists and is unique for all \(f \in C(E)\), \(v_ 0 \in D(A)\), (ii) \(v(t, \cdot, \cdot)\) is continuous on \(C(E) \times E\), which is obviously equivalent to \((*)\). The further analysis shows that such stated well-posedness (for the abstract Cauchy problem in \(C(E))\) yields the following coercivity inequality: (2) \(\| v' \|_{C(E)} + \| Av \|_{C(E)} \leq M(\| f \|_{C(E)} + \| Av_ 0 \|_ E)\), where \(M\) does not depend on \(f\) and \(v_ 0\). But it turns out that if the coercivity inequality (2) holds for arbitrary solutions \(v\) of problem (1) with given \(f \in C(E)\), \(v_ 0 \in E\), then the problem (1) is really well- posed in \(C(E)\). Analogous facts are valid for other problems in various suitable fractional spaces or in \(L^ p\). The treatment is by aids of semigroup theory. The notion of well-posedness is discussed in great details. The second chapter deals with discretization in the time variable \(t\) via the fully implicit Euler scheme (called ``Rothe difference scheme'' by the authors), again in the context of various function spaces. In Chapter 3 a nice introduction to the theory of Padé approximation of the function \(\exp (-z)\) near \(z=0\) is presented and a thorough investigation of a Padé difference scheme is carried out. In Chapter 4 the abstract methods and results are applied to the treatment of linear parabolic equations with constant coefficients in the elliptic differential operator. The effect of additional discretization in the spatial variables is analyzed by the method of symbols (based on the Fourier transform). The book can be recommended to the researcher in the field who has already good knowledge of the more concrete aspects of theory and techniques of discretization.