Markov operators, positive semigroups and approximation processes (Q461179): Difference between revisions

The classical Weierstrass Approximation Theorem states that every continuous function on \([0,1]\) can be approximated in supremum norm by a sequence of polynomials. However, its standard proof does not reveal how to construct an approximating sequence and, in particular, whether this sequence can be chosen in such a way that it shares relevant properties of the limiting function. It was Sergei Bernstein who first showed how to explicitly construct an approximating sequence consisting of what are nowadays referred to as ``Bernstein polynomials''. An interesting feature of the Bernstein polynomials \(B_n(f)\) is that they are positive functions on \([0,1]\) if so is the limiting function \(f\). This observation paves the way for the investigation of positive approximation in different contexts: the first and last named authors have been world leading experts in this research field for decades now. This monograph is the compendium of recent developments obtained by them, many in cooperation with the two remaining authors. As the authors themselves stress in the introduction of Chapter 4, the interplay between positive approximation schemes and the classical theory of Markov operators is probably the most remarkable and interesting topic of this book. The leading idea of their whole program is that Bernstein's construction of the linear operators \(f\mapsto B_n(f)\) can be generalized in a tricky but elegant way, thus introducing so-called ``Bernstein-Schnabl operators'': as the authors convincingly show in Section 3.1, numerous operators can be discussed in this framework as they share a fundamental feature, viz.\ the same abstract construction based on some Markov operator \(T\). In the case of classical Bernstein operators, \(Tf\) is simply the projection onto the affine function interpolating the boundary values of \(f\); more generally, the special case of Markov operators \(T\) that are in fact projections was the topic of another monograph by one of the authors [\textit{F. Altomare} and \textit{M. Campiti}, Korovkin-type approximation theory and its applications. Berlin: Walter de Gruyter (1994; Zbl 0924.41001)]. Extending this theory to the case of Bernstein-Schnabl operators based upon general Markov operators is the main goal of this book. Recall that a bounded linear operator \(T\) acting on the space \(C(K)\) of continuous functions over a compact Hausdorff space \(K\) is said to be a Markov operator if it maps positive functions into positive functions and \(T\mathbf{1}=\mathbf{1}\) (or equivalently, if it is positive and its operator norm is 1). Chapter 1 is of introductory nature; it is devoted to the general theory of positive operators on spaces \(C(K)\) and related subjects, like Radon measures, positive operators on Lebesgue spaces and associated invariant measures, Choquet boundaries and Bauer simplices. Likewise, Chapter 2 summarizes some well-known facts from the theory of strongly continuous semigroups of bounded linear operators and, in particular, of those semigroups consisting of positivity preserving operators. The interesting Section 2.3 is devoted to the theory of Feller semigroups and their interplay with Markov processes. The focus is on spaces of continuous functions, whereas invariant measures and \(L^p\)-theory is hardly touched upon and almost no results are proved. The topics in the first two chapters are mostly classical, although some interesting, less well-known facts are also presented, like an approach to the issue of convergence of (continuous or discrete) semigroups of Markov operators whose central assumption is that Lipschitz continuous functions over \(X\) are left invariant and that their Lipschitz seminorm converges to 0 or Feller's characterization of boundary conditions for an elliptic operator on \([0,1]\). As anticipated above, then main topic of this book is the treatment of a certain class of generalizations of the classical Bernstein operators: the theory of such \textit{Bernstein-Schnabl operators} is discussed in Chapter 3, which is introduced in a nice axiomatic way based on the notion of Markov operator; special cases are then considered and classical examples recovered, including the classical Bernstein polynomials that approximate continuous functions on a bounded interval. The approximating properties of Bernstein-Schnabl operators are thoroughly investigated and convergence rates with respect to different topologies as well as the behavior of powers of these operators is studied. Just like the classical Bernstein operators, the Bernstein-Schnabl operators are positive, i.e., the Bernstein-Schnabl schemes approximate positive functions by sequences of positive functions; several further interesting properties are shown in the remainder of the chapter, including the fact that \(B_n(f)\) is a Hölder continuous (resp., convex) function if so is \(f\). The authors discuss in detail two different classes of related operators in Chapter 4: given a Markov operator \(T\), \(L_T\) is an operator defined as the limit of an operator sequence based on the Bernstein-Schnabl operators associated with \(T\), while \(W_T\) is a second-order differential operator in non-divergence form whose coefficients depend on the operator \(T\). Interestingly, both operators coincide whenever restricted to the space spanned by products of arbitrarily many affine functions. Under certain conditions, the closure of \(W_T\) generates a Markov semigroup that can be approximated by powers of the Bernstein-Schnabl operators associated with \(T\). In view of the results of Chapter 3 about heritage of smoothness properties, this suggests a smart method to show that a semigroup preserves Hölder continuity, monotonicity or convexity, provided one is able to show that its generator is of the form \(W_T\) for some Markov operator \(T\): this is in particular the case for Fleming-Viot-type operators on simplices and further differential operators with degenerate coefficients: these operators typically appear in differential equations that model systems with stochastic features, especially in genetics and finance. This idea is thoroughly developed in Sections 4.4 and 4.5. This is particularly remarkable because, unlike other qualitative properties (like positivity), it is currently not known whether preservation of smoothness or convexity can be discussed by variational methods in a Hilbert space setting, see [\textit{A. Bátkai} and \textit{A. Bobrowski}, Arch. Math. 98, No. 1, 37--48 (2012; Zbl 1242.47033)] for a discussion of these aspects. The main topic of Chapter 5 are perturbation results of different nature. An interesting class of perturbations consists of convex combinations of a Markov operator \(T\) on \(C(K)\) and the identity operators: this results in so-called ``Lotosky-Schnabl operators''. Further generalizations of Bernstein-Schnabl operators are considered throughout the chapter: they allow for investigations of more general differential operators, including general linear or even quasilinear elliptic operators. This book is a welcome contribution to the topic of approximation theory for solutions of partial differential equations with degenerate coefficients. It is written as -- and meant to be -- a standard reference for researchers working in this field, and it will certainly become one. Reading is made occasionally difficult by the somewhat heavy notation; but a 4-page-long symbol index, subdivided in five sections, will help the casual reader. On the other hand the dedicated reader will discover several little known pages of operator theory along with quite a few interesting challenges and possible directions of future research. The focus on applications to spaces of continuous functions makes this book complementary to existing classical monographs in semigroup theory, like [\textit{K.-J. Engel} and \textit{R. Nagel}, One-parameter semigroups for linear evolution equations. Berlin: Springer (2000; Zbl 0952.47036)] or [\textit{E. M. Ouhabaz}, Analysis of heat equations on domains. Princeton, NJ: Princeton University Press (2005; Zbl 1082.35003)].