Chapter VI. Measure Theory for Euclidean Space

Abstract

This chapter mines some of the powerful consequences of the basic measure theory in Chapter V.

Sections 1–3 establish properties of Lebesgue measure and other Borel measures on Euclidean space and on open subsets of Euclidean space. The main general property is the regularity of all such measures—that the measure of any Borel set can be approximated by the measure of compact sets from within and open sets from without. Lebesgue measure in all of Euclidean space has an additional property, translation invariance, which allows for the notion of the convolution of two functions. Convolution gives a kind of moving average of the translates of one function weighted by the other function. Convolution with the dilates of a fixed integrable function provides a handy kind of approximate identity.

Section 4 gives the final form of the comparison of the Riemann and Lebesgue integrals, a preliminary form having been given in Chapter III.

Section 5 gives the final form of the change-of-variables theorem for integration, starting from the preliminary form of the theorem in Chapter III and taking advantage of the ease with which limits can be handled by the Lebesgue integral. Sard's Theorem allows one to disregard sets of lower dimension in establishing such changes of variables, thereby giving results in their expected form rather than in a form dictated by technicalities.

Section 6 concerns the Hardy–Littlewood Maximal Theorem in $N$ dimensions. In dimension 1, this theorem implies that the derivative of a 1-dimensional Lebesgue integral with respect to Lebesgue measure recovers the integrand almost everywhere. The theorem in the general case implies that certain averages of a function over small sets about a point tend to the function almost everywhere. But the theorem can be regarded as saying also that a particular approximate identity formed by dilations applies to problems of almost-everywhere convergence, as well as to problems of norm convergence and uniform convergence. A corollary of the theorem is that many approximate identities formed by dilations yield almost-everywhere convergence theorems.

Section 7 redevelops the beginnings of the subject of Fourier series using the Lebesgue integral, the theory having been developed with the Riemann integral in Section I.10. With the Lebesgue integral and its accompanying tools, Fourier series are meaningful for more functions than before, Dini's test applies even to a wider class of Riemann integrable functions than before, and Fejér's Theorem and Parseval's Theorem become easier and more general than before. A completely new result with the Lebesgue integral is the Riesz–Fischer Theorem, which characterizes the trigonometric series that are Fourier series of square-integrable functions.

Sections 8–10 deal with Stieltjes measures, which are Borel measures on the line, and their application to Fourier series. Such measures are characterized in terms of a class of monotone functions on the line, and they lead to a handy generalization of the integration-by-parts formula. This formula allows one to bound the size of the Fourier coefficients of functions of bounded variation, which are differences of monotone functions. In combination with earlier results, this bound yields the Dirichlet–Jordan Theorem, which says that the Fourier series of a function of bounded variation converges pointwise everywhere, the convergence being uniform on any compact set on which the function is continuous. Section 10 is a short section on computation of integrals.