I recently learned about characteristic functions and in particular the Bochner Theorem which helps us ascertain when a given function is a characteristic function for some probability distribution.

The standard version of Bochner's theorem from Wikipedia states that: For any normalized continuous positive definite function $f$ on $G$ (normalization here means $f$ is $1$ at the unit of $G$), there exists a unique probability measure on ${\displaystyle {\widehat {G}}}$ such that

$${\displaystyle f(g)=\int _{\widehat {G}}\xi (g)d\mu (\xi )}$$

i.e. $f$ is the Fourier transform of a unique probability measure $\mu$ on ${\displaystyle {\widehat {G}}}$. In other words, any continuous positive-definite function on the real line is the Fourier transform (characteristic function) of a (positive) finite measure.

Another version, which combines Khinchine’s theorem is stated as follows:

Let $\phi: \mathbb{R} \to \mathbb{C}$ be a continuous function with $\phi(0) =1$. Then $\phi$ is a characteristic function $\iff$ $\forall n \in \mathbb{N}, t_i \in \mathbb{R}$ and $\lambda_i \in \mathbb{C}$ for $i = 1, \dots, n$, we have $$\sum_{i, j =1}^n \phi(t_i - t_j) \lambda_i \overline{\lambda_k} \geq 0$$

Wikipedia notes that the trouble with these theorems is that computational verification of the positive-definiteness is not easy. I'm curious to learn whether there is an effective/efficient way to check for positive definiteness of a function (or the condition in the second version provided below) because based on the definition of positive-definiteness for functions, it seems like we would have to prove the square matrices for all sizes and possible permutations are Hermitian.

I am also aware of Polya's criterion as an alternative to deal with the problem motivating Bochner's theorem but are there any easy ways to verify that there exists a distribution such that a given function corresponds to the characteristic function for it. Until now, I have defined appropriate random variables to solve relatively direct problems (convex combinations of characteristic functions and square of the absolute value of a characteristic function also define characteristic functions).

  • $\begingroup$ As far as I know there is no simple way of determining if a given function is positive definite. If such a method existed it would have made the job determining which distributions are infinitely divisible an easy one, but this job is not easy. $\endgroup$ – Kavi Rama Murthy Mar 18 '18 at 12:09

I think that the answer to your question will be heavily dependent on the group $G$. In the case of finite cyclic groups $G = \mathbb{Z}/n\mathbb{Z}$ you can apply the Fast Fourier Transform to get the Fourier coefficients of $\phi$ with a computational cost of $O(n \log n)$ and then you only have to check the positivity of $\hat\phi$. The same method will work for finite Abelian groups.

I do not think there is short answer for $G$ infinite.


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.