3
$\begingroup$

I am thinking about the statement if the characteristic function of a random variable $X$, $\Phi_X$, is always differentiable.

By definition, $$\Phi_X(t)=\int_{\Bbb{R}^d}e^{i\langle t,y \rangle}P_X(dy)$$ Hence, I think it has something to do with changing the integral and the derivative right? But my intuition tells me that there is a counterexample but I can't find one.

$\endgroup$
11
  • $\begingroup$ If $E|X| < \infty$, then $\Phi_X^\prime(t) = E(iXe^{itX})$ by LDCT, so you will have to look amongst RVs without that moment condition. $\endgroup$
    – Jose Avilez
    May 9 at 18:16
  • $\begingroup$ @JoseAvilez So would it work if I consider $X=1$ $\endgroup$
    – Wave
    May 9 at 18:20
  • $\begingroup$ No. $X=1$ has finite mean. $\endgroup$
    – Jose Avilez
    May 9 at 18:23
  • $\begingroup$ @JoseAvilez ah sure that's because of the properties of the probability measure right? But what if I take $X=x$? $\endgroup$
    – Wave
    May 9 at 18:25
  • $\begingroup$ Same. All constants have finite mean. $\endgroup$
    – Jose Avilez
    May 9 at 18:30

1 Answer 1

Reset to default
8
$\begingroup$

If $E|X| < \infty$ then the derivative of the characteristic function is given by: $$\Phi_X^\prime (t) = E(iX e^{itX})$$ as the Lebesgue Dominated Convergence theorem would allow us to exchange the order of differentiation and integration. Thus, we must look for a random variable $X$ with infinite mean.

Let $X$ be a Cauchy random variable, so that its pdf is given by $$f_X(x) = \frac{1}{\pi (1 + x^2)} \qquad x \in \mathbb{R}$$ and its characteristic function can be computed to be $$\Phi_X(t) = \exp (-|t|)$$ $\Phi_X(t)$ is readily seen not to be differentiable at $t = 0$, as desired.

$\endgroup$

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.