4
$\begingroup$

I came across an interesting problem but unable to see how to approach it. How do I use the dominated convergence theorem (LDCT), to show that first derivative of characteristic function of probability distribution at t = 0, $\phi^′(0)=iE[X]$? Any ideas?

References:

http://en.wikipedia.org/wiki/Characteristic_function_(probability_theory) http://en.wikipedia.org/wiki/Dominated_convergence_theorem

$\endgroup$
5
$\begingroup$

First, one has $$e^{ir}-1=O(|r|) $$ for all $r\in \mathbb{R}.$ This means that there exist constant (independent of $r$) $A>0$ such that $$|e^{ir}-1|\le A|r| \mbox{ for all } r\in \mathbb{R} $$ (In fact, you can take $A=1$, as $|e^{ir}-1|$ is the length of the chord from 1 to $e^{ir}$ and $r$ is the corresponding arc length on the unit circle.)

Let us prove your question in the case $X $ is contiuous. (The discrete case can be proved similarly.) Suppose $f(x)$ is the probability density function of $X$. Then $$ \phi(t)=E(e^{itX})=\int_{-\infty}^\infty f(x)e^{itx} dx$$ and $\phi(0)=\int_{-\infty}^\infty f(x) dx =1$. So by linearity of integral $$\phi'(0)=\lim_{t\to 0}\frac{\phi(t)-\phi(0)}{t}=\lim_{t\to 0}\int_{-\infty}^\infty f(x)\frac{e^{itx}-1}{t} dx.$$ Since $$\left|f(x)\frac{e^{itx}-1}{t}\right|\le f(x) A \frac{|tx|}{|t|}=A f(x)|x|,$$ and $$\lim_{t\to 0}\frac{e^{itx}-1}{t}=\frac{d}{dt} e^{itx}|_{t=0}=ix e^0=ix,$$ it follows from DCT that if $E(|X|)<\infty$, then $$\phi'(0)= \int_{-\infty}^\infty f(x)ix dx=i E(X).$$

$\endgroup$
1
$\begingroup$

You need to show that if $(a_n)$ is a sequence of nonzero numbers which tends to zero, that $(\phi(a_n)-\phi(0))/a_n\to iE[X]$. Now $$\frac{\phi(a)-\phi(0)}{a}=E\left(\frac{e^{iaX}-1}{a}\right).$$ You need some hypothesis on $X$ for the result to work, for instance that $|X|$ has finite expectation. Certainly $(e^{iaX}-1)/a\to iX$ so to apply dominated convergence we need a function $f(X)$ with finite expectation and with $|(e^{iaX}-1)/a|\le f(X)$ at least for $a$ in a deleted neighbourhood of $0$. Does $f(X)=|X|$ work?

$\endgroup$

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.