Is it true that every bounded sequence $\phi: \mathbb{Z} \rightarrow \mathbb{C}$ is the image of some Borel measure $\mu$ under the Fourier transform $\mathcal{F}$, i.e. $$ \forall\, k \in \mathbb{Z}, \, \phi(k) = \int_{(-1, 1]} e^{- 2 \pi i k \cdot \omega} d\mu(\omega) \, \, ? $$ If not, what would be a counterexample?

If $\phi \in l^{\infty}(\mathbb{Z})$ is positive definite, then the answer is yes and the Fourier preimage $\mu$ is a positive measure by Bochner's Theorem.

Writing a Borel measure $\mu$ as a linear combination of positive measures, the question is then equivalent to asking whether the positive definite functions span $l^{\infty}(\mathbb{Z})$.

In turn, this is the same as asking whether a bi-infinite Toeplitz matrix $$ M = [M_{ij}]_{i, j \in \mathbb{Z}} \,\, ( = \phi(i-j) ) $$ is in general a linear combination of (formally) positive semidefinite bi-infinite Toeplitz matrices.

(The matrix formulation seems to suggest the answer is "no". The positive and negative parts of an $n \times n$ symmetric Toeplitz matrix need not be Toepiliz.)

  • $\begingroup$ The function should definitely be bounded... $\endgroup$
    – PhoemueX
    Jul 5 '20 at 6:00

No, it's not true that every bounded sequence is the image of a Borel measure under the Fourier transform.

One counterexample is the following: $$\phi(k) = \text{p.v.} \int_{-1/2}^{1/2} \frac{e^{-2 \pi i k x}}{x} \, dx.$$

In other words, this sequence is the Fourier transform of the periodic distribution $F$ given by $$\langle F, \psi \rangle = \text{p.v.} \int_{-1/2}^{1/2} \frac{\psi(x)}{x} \, dx$$.

There are now two points to sort out.

(1) The sequence $\phi$ so defined is bounded.

To see this, first note that the real part is identically zero, since the real part of each integrand is an odd function.

The imaginary part of $\phi(k)$ can be written as $$-2 \int_{0}^{1/2} \frac{\sin(2 \pi k x)}{x} \, dx = -2 \int_{0}^{k/2} \frac{\sin(2 \pi y)}{y} \, dy,$$ and using integration by parts you can show that this remains bounded as $|k| \to \infty$.

(2) This sequence does not arise from a Borel measure.

The distribution $F$ pretty clearly doesn't come from a (finite) Borel measure because of the positive and negative divergences at 0. Of course it's like the difference between two non-finite positive measures, but I imagine that's outside the bounds of what you were going for.

One could also justify why there isn't a finite Borel measure whose Fourier transform coincides with that of $F$. I'd view this as a general property of distributions and the Fourier transform, i.e. the Fourier transform is 1-to-1. As a reference, for periodic distributions this is briefly touched on in Folland's Real Analysis at the top of p. 298.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.