1
$\begingroup$

For a random variable $X$ and numbers $0<r<s<\infty$

$$E(|X|^r)^\frac{1}{r}\le E(|X|^s)^\frac{1}{s}$$

This almost immediately follows from Holder's inequality for random variables $Y,Z$

$$E(|ZY|)\le E(|Z|^p)^\frac{1}{p}E(|Y|^q)^\frac{1}{q}$$ Setting $Y=1$ we end up with

$$E(|Z|)\le E(|Z|^p)^\frac{1}{p}.$$

With $|Z|=|X|^r$ we get

$$E(|X|^r)\le E(|X|^{rp})^\frac{1}{p}.$$

Now, since $p \ge1$ we have $rp \ge r$ and setting $rp=s$ we get

$$E(|X|^r)\le E(|X|^{s})^\frac{r}{s}$$

or

$$E(|X|^r)^\frac{1}{r}\le E(|X|^{s})^\frac{1}{s}.$$

Is there a way to expand this inequality to $-\infty<r<s<\infty$?

$\endgroup$
1
  • $\begingroup$ The original result you state at the beginning is a certain Lyapunov inequality discussed here previously. $\endgroup$ Oct 14 '18 at 14:17
1
$\begingroup$

Yes, assuming the expectations exist (for $r=0$, it is $\lim_{r\to 0}(\mathbb{E}[|X|^r])^{1/r}=\exp(\mathbb{E}[\log |X|])$). This is just the power mean inequality in the discrete case. So the lazy way to prove it is simply take limit of discrete case. A more honest way to prove this is to establish comparing $r=0$ with $s>0$ (Jensen's inequality with $\varphi(x)=-\log x$ for the random variable $|X|^s$)), and then use $Y=X^{-1}$ for first the case $r=-1$, $s=0$, and finally for general $r,s$.

$\endgroup$

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.