3
$\begingroup$

A sequence of random variables $X_n$ converges in distribution to $X$, if there is pointwise convergence of its characteristic functions, i.e. $\lim_{n\rightarrow\infty}\phi_{X_n}(\lambda) = \phi_X(\lambda)$ for all $\lambda \in \mathbb R$.

Imagine we do not only have pointwise, but uniform convergence of the characteristic functions ($\lim_{n\rightarrow\infty}\|\phi_{X_n}-\phi_X\|_\infty = 0$). Can we estimate the error in the distributions by knowing $\|\phi_{X_n}-\phi_X\|_\infty$, i.e. do we have $$ |P(X_n \le z ) - P(X < z)| \le \alpha\left(\|\phi_{X_n}-\phi_X\|_\infty\right)$$ for a function $\alpha(\cdot)$ and any $z \in \mathbb R$? What is the form of $\alpha(\cdot)$?

$\endgroup$

migrated from math.stackexchange.com Feb 9 '15 at 6:35

This question came from our site for people studying math at any level and professionals in related fields.

3
$\begingroup$

Assuming that you want $\alpha(0+)=0$, there is no such inequality. In fact, by Proposition 11.7.6 in Dudley's Real Analysis and Probability, the optimal $\alpha$ in your question, if taken literally, is identical to $1$ for strictly positive arguments.

As Dudley states in the end-of-chapter notes, the result of his Proposition 11.7.6 is due to Dyson (1953). If I recall it correctly, Dyson's original example is more complicated than Dudley's.

$\endgroup$

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.