Abstract
So far, we have studied Fourier transforms or Fourier coefficients of functions defined on \(\mathbb {R}\) or \(\mathbb {T}\). Inverse procedures to recover original functions from given Fourier transforms or Fourier coefficients were also discussed (spectral synthesis). However, there are many functions to which the methods of classical Fourier analysis can not be applied. In this chapter, we develop the theory of Fourier transforms of measures as a similar but new method to overcome such difficulties.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
A function f is said to vanish at infinity if there exists, for each ε > 0, some compact set K ⊂ X such that |f(x)| < ε for all x ∈ K c.
- 2.
- 3.
The interval of integration should be interpreted as [−π, π).
- 4.
See p. 83 for dual operators.
- 5.
- 6.
The Fourier coefficients of P(x) are given by
$$\displaystyle \begin{aligned} \frac{1}{\sqrt{2\pi}}\int_{-\pi}^\pi P(x)e^{-ijx}dx=\sqrt{2\pi}\alpha_j\end{aligned}$$for \(|j|\leqq n\), and 0 for \(|j|\geqq n+1\). Hence the Fourier series of P(x) is computed as
$$\displaystyle \begin{aligned} P(x)=\sum_{j=-n}^n \sqrt{2\pi}\alpha_j \frac{1}{\sqrt{2\pi}}e^{ijx}=\sum_{j=-n}^n \alpha_j e^{ijx}. \end{aligned}$$ - 7.
G may be, more generally, a commutative group.
- 8.
See textbooks on linear algebra for Hermitian forms and their signs.
- 9.
- 10.
- 11.
- 12.
- 13.
See Sect. 6.7 in this chapter for the convolution of two measures.
- 14.
“\(\hat {\mu }(\xi )=0\) for a.e. ξ” is sufficient to get μ = 0. However, \(\hat {\mu }(\xi )\) is eventually equal to 0 for every ξ, since we know that \(\hat {\mu }(\xi )\) is continuous.
- 15.
- 16.
\(g(\xi )\hat {\mu }(-\xi )\rightarrow g(\xi )\varphi (-\xi )\). And {μ n} is bounded (with respect to the operator norm), since μ n → μ in w ∗-topology. Furthermore, since ∥g∥∞ < ∞, there exists some \(F\in \mathfrak {L}^1\) such that
$$\displaystyle \begin{aligned} |g(\xi)\hat{\mu}_n(-\xi)|\leqq |g(\xi)| \cdot C\leqq F(\xi). \end{aligned}$$Then apply the dominated convergence theorem.
- 17.
Assume, on the contrary, that \(\hat {\mu }(-\bar {\xi })-\varphi (-\bar {\xi })\neq 0\) (say, > 0) for some \(\bar {\xi }\). Then there exists a neighborhood U of \(\bar {\xi }\) such that \(\hat {\mu }(-\xi )-\varphi (-\xi )>0\) on U. There exists some \(g\in \mathfrak {S}\) such that supp g ⊂ U and g(x) > 0 in the interior of its support. Then clearly
$$\displaystyle \begin{aligned} \int_{-\infty}^\infty g(\xi)(\hat{\mu}(-\xi)-\varphi(-\xi))d\xi>0. \end{aligned}$$Contradiction.
- 18.
For a 2π-periodic function f, we have
$$\displaystyle \begin{aligned} \int_{-\infty}^\infty f(x)d\mu =\int_{-\pi}^\pi f(x)d\nu. \end{aligned}$$ - 19.
- 20.
\(\displaystyle {\left ( \frac {x+2\pi m}{\lambda }\right )^2 \geqq \left (\frac {\pi -2\pi |m|}{\lambda }\right )^2 =\frac {\pi ^2}{\lambda ^2}(1-2|m|)^2 \geqq \frac {\pi ^2}{\lambda ^2}m^2, }\) where \(|m|\geqq 1.\)
- 21.
- 22.
- 23.
- 24.
If \(f\in \mathfrak {A}\) and \(f\geqq 0\), there exists a sequence {g n} in \(\mathfrak {C}_0^\infty \) which converges to \(\hat {f}\) in \(\mathfrak {L}^1\). Let f n be the inverse Fourier transform of g n; i.e. \(\hat {f}_n=g_n\). Then {f n} is a sequence in \(\mathfrak {S}_c\). If
$$\displaystyle \begin{aligned} \int_{-\infty}^\infty \hat{f}_n(\xi)\varphi(-\xi)d\xi =\int_{-\infty}^\infty g_n(\xi)\varphi(-\xi)d\xi \geqq 0 \end{aligned}$$for each n, we obtain, by passing to the limit,
$$\displaystyle \begin{aligned} \int_{-\infty}^\infty \hat{f}(\xi)\varphi(-\xi)d\xi \geqq 0. \end{aligned}$$ - 25.
- 26.
By Lemma 6.4, a positive semi-definite function is uniformly continuous on \(\mathbb {R}\) if it is continuous at 0.
- 27.
- 28.
Here φ denotes the tempered distribution defined by the bounded function φ. Hence
$$\displaystyle \begin{aligned} \varphi(s)=\int_{\mathbb{R}}\varphi(x)s(x)dx, \quad s \in \mathfrak{S}(\mathbb{R}). \end{aligned}$$ - 29.
We note that there exists a unique function in \(\mathfrak {S}(\mathbb {R})\), the Fourier transform of which is just \(\sqrt {p}\).
- 30.
I repeat again the exposition of the concept of the torus. A binary relation ∼ on \(\mathbb {R}\) is defined by
$$\displaystyle \begin{aligned} x \sim y \Longleftrightarrow x-y\in 2\pi\mathbb{Z}. \end{aligned}$$∼ is an equivalence relation, and any two numbers are equivalent if and only if the difference between them is a multiple of 2π. Any real number has its equivalent number in [−π, π).
Given a function \(f:[-\pi , \pi ) \rightarrow \mathbb {C}\), we may not be able to define f(x + y), since x + y may not be in [−π, π). However, there exists some z ∈ [−π, π) which is equivalent to x + y. We understand f(x + y) as f(z). A similar convention applies to the Dirac measures on [−π, π). That is, we understand δ x+y as δ z. More generally, we can consider f(x 1 + x 2 + ⋯ + x n) or \(\delta _{x_1+x_2+\cdots +x_n}\) in the same way.
When we consider a function on \(\mathbb {R}\) rather than [−π, π), such a convention is not necessary.
Exactly speaking, we consider a function defined on the quotient group \(\mathbb {R}/2\pi \mathbb {Z}\) modulo \(2\pi \mathbb {Z}\). This is called torus and denoted by \(\mathbb {T}\). cf. Appendix A. We are indebted to Malliavin [7] Chap. III, §1 for the expositions in this section.
- 31.
- 32.
Any bounded Borel measure μ on G can be represented as μ = μ a + μ s + μ d, where μ a ≪ m (Lebesque measure) μ s is continuous and μ s ⊥ m, and μ d is discrete. Such a decomposition is uniquely determined. cf. Igari [4] p. 137.
- 33.
\(\widehat {\mu ^\sharp }=\bar {\hat {\mu }}\).
- 34.
We owe this to Katznelson [5] Chap. I, §7 and Chap. VI, §2.
- 35.
- 36.
Assume t ≠ 0. Then
$$\displaystyle \begin{aligned} \varphi_n(t)=\frac{1}{it}e^{i\xi t}\bigg|{}_{-n}^n =\frac{1}{it}(e^{int}-e^{-int}) =\frac{\, 2\,}{t} \sin nt.\end{aligned} $$Hence
$$\displaystyle \begin{aligned} \frac{1}{2n}\varphi_n(t)=\frac{1}{nt}\sin t.\end{aligned} $$We obtain (6.88) as n →∞.
- 37.
The upper and lower ends of the integration can be \(\displaystyle \int _{\theta _n}^{\theta _n+n}\) for any sequence {θ n}.
References
Bochner, S.: Monotone Funktionen, Stieltjessche Integrale und harmonische Analyse. Math. Ann. 108, 378–410 (1933)
Carathéodory, C.: Über den Variabilitätsbereich der Koefficienten von Potenzreihen die gegebene Werte nicht annemen. Math. Ann. 54, 95–115 (1907)
Herglotz, G.: Über Potenzreihen mit positiven reellen Teil in Einheitskreise. Berichte Verh Säcks. Akad. Wiss. Leipzig. Math.-phys. Kl. 63, 501–511 (1911)
Igari, S.: Jitsu-kaiseki Nyumon (Introduction to Real Analysis). Iwanami Shoten, Tokyo (1996) (Originally published in Japanese)
Katznelson, Y.: An Introduction to Harmonic Analysis, 3rd edn. Cambridge University Press, Cambridge (2004)
Lax, P.D.: Functional Analysis. Wiley, New York (2002)
Malliavin, P.: Integration and Probability. Springer, New York (1995)
Maruyama, T.: Kansu Kaisekigaku (Functional Analysis). Keio Tsushin, Tokyo (1980) (Originally published in Japanese)
Maruyama, T.: Sekibun to Kansu-kaiseki (Integration and Functional Analysis). Springer, Tokyo (2006) (Originally published in Japanese)
Maruyama, T.: Herglotz-Bochner representation theorem via theory of distributions. J. Oper. Res. Soc. Japan 60, 122–135 (2017)
Naimark, M.A.: Normed Algebras. Wolters Noordhoff, Groningen (1972)
Schwartz, L.: Théorìe des distributions. Nouvelle édition, entièrement corrigée, refondue et augmentée, Hermann, Paris (1973)
Toeplitz, O.: Über die Fouriersche Entwickelung positive Funktionen. Rend. di Cinc. Mat. di Palermo 32, 191–192 (1911)
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer Nature Singapore Pte Ltd.
About this chapter
Cite this chapter
Maruyama, T. (2018). Fourier Transforms of Measures. In: Fourier Analysis of Economic Phenomena. Monographs in Mathematical Economics, vol 2. Springer, Singapore. https://doi.org/10.1007/978-981-13-2730-8_6
Download citation
DOI: https://doi.org/10.1007/978-981-13-2730-8_6
Published:
Publisher Name: Springer, Singapore
Print ISBN: 978-981-13-2729-2
Online ISBN: 978-981-13-2730-8
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)