0
$\begingroup$

Suppose $f(x)$ is the probability density function of a random variable $X$, which means:

$$\int_{a}^{b} f(x) dx = 1$$

Also suppose $f$ is continuous and differentiable.

Provide a non-trivial condition under which $\int_{a}^{b} |f^\prime(x)| dx$ exists.

$[a,b]$ maybe a compact interval (regular integral) or $[-\infty, \infty]$ (improper integral).

EDIT: a non-trivial condition is a condition that is satisfied by many well-known distributional families.

$\endgroup$
10
  • $\begingroup$ can you give an example where this integral does not exist? $\endgroup$ Commented Mar 22, 2023 at 9:23
  • $\begingroup$ See this post. $\endgroup$ Commented Mar 22, 2023 at 11:08
  • 5
    $\begingroup$ Provide a non-trivial condition Define "non-trivial". $\endgroup$
    – fedja
    Commented Mar 22, 2023 at 11:30
  • 1
    $\begingroup$ Actually $|f'|$ is measurable and non-negative so the integral does exist. Maybe you mean "is finite" ? $\endgroup$ Commented Mar 22, 2023 at 18:41
  • 1
    $\begingroup$ If f is absolutely continuous then f' [exists a.e.] and is Lebesgue integrable $\endgroup$ Commented Mar 22, 2023 at 18:57

1 Answer 1

3
$\begingroup$

A sufficient condition is unimodality of $f$, namely the existence of some $c \in~]a,b[$, such that $f$ non-decreasing on $[a,c]$ and $f$ non-increasing on $[a,c]$. If this property holds, then by Fatou's lemma \begin{eqnarray*} \int_a^c |f'(x)|dx &=& \int_a^c \lim_{h \to 0+} \frac{f(x+h)-f(x)}{h}dx \\ &\le& \liminf_{h \to 0+} \int_a^c \frac{f(x+h)-f(x)}{h}dx \\ &=& \liminf_{h \to 0+} \frac{1}{h}\int_c^{c+h} f(x)dx - \frac{1}{h}\int_a^{a+h} f(x)dx \\ &=& f(c)-f(a). \end{eqnarray*} In a same way, $$\int_a^c |f'(x)|dx \le f(c)-f(b).$$ The conclusion follows.

The same argument applies when $f$ has only finitely many critical points, since, by Darboux Theorem, the sign of $f'$ remains constant between two critical points.

$\endgroup$
2
  • $\begingroup$ How about cases when the integral is an improper one? i.e., both $a$ and $b$ are infinity. $\endgroup$ Commented Mar 23, 2023 at 3:37
  • $\begingroup$ The argument given in my post still apply in this case. $\endgroup$ Commented Mar 23, 2023 at 7:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .