2
$\begingroup$

Let $Y$ be an integrable nonnegative random random variable on some probability space $X$, and Let $F:[0,\infty) \to \mathbb R$ be a continuous function.

Suppose that $F(Y) \in L^1(X)$. (I am fine with assuming $Y$ is bounded as well).

Do there always exist simple functions $Y_n \ge 0$ on $X$ such that

$$ E(Y)=\lim_{n \to \infty} E(Y_n), E(F(Y))=\lim_{n \to \infty} E(F(Y_n)) $$ both hold simultaneously?

Taking $Y_n$to be increasing, we get the first equality, due to the monotone convergence theorem. But if $F$ is not increasing, then $F(Y_n)$ won't necessarily be increasing.

$\endgroup$
  • $\begingroup$ Are both $Y$ and $F(Y)$ integrable? $\endgroup$ – zhoraster Jun 3 at 17:05
  • $\begingroup$ Yes, I have updated the question to mention this explicitly. $\endgroup$ – Asaf Shachar Jun 3 at 19:06
1
$\begingroup$

Under the assumption $Y$ is bounded: There exists a positive real $M$ such that $0\le Y \le M$ on $X.$ Let $Y_n$ be the usual simple functions approximating $Y.$ We then have $0\le Y_n\le M$ for all $X.$ Furthermore, $Y_n\to Y$ uniformly on $X;$ this you will recall holds because $Y$ is bounded. Now $F$ is uniformly continuus on $[0,M],$ and from this it follows that $F\circ Y_n\to F\circ Y$ uniformly on $X.$ The measure on $X$ is finite and the result follows.

| cite | improve this answer | |
$\endgroup$
1
+50
$\begingroup$

You don't need $Y$ to be bounded. First, we can assume WLOG that $F$ is non-negative and $F(0)=0$. Let $(\varepsilon_n)$ be a sequence decreasing to $0$ and $A_n := \{Y \le n \}$. Since $F$ is uniformly continuous on $[0,n]$ for every $n$, there exists $\delta_n$ such that if $|x-y| < \delta_n$ and $0 \le x,y \le n$ we have $|F(x)-F(y)| < \varepsilon_n$.

Take $Y_n \ge 0$ to be a simple function such that $Y_n \le Y$, $Y_n = 0$ on $A_n^c$, and $|Y-Y_n| < \delta_n$ on $A_n$. Then we have

$$ 0 \le \mathbb{E}[Y-Y_n] = \mathbb{E}[(Y - Y_n)1_{A_n}] + \mathbb{E}[Y1_{A_n^c}] < \delta_n + \mathbb{E}[Y1_{A_n^c}]. $$

The first term goes to $0$ by our choice of $\delta_n$, and the second term goes to $0$ by the monotone convergence theorem, so $(Y_n) \rightarrow Y$ in $L^1$. Similarly, we compute

$$\mathbb{E}[|F(Y)-F(Y_n)|] = \mathbb{E}[|F(Y)-F(Y_n)|1_{A_n}] + \mathbb{E}[|F(Y)-F(0)|1_{A_n^c}] < \varepsilon_n + \mathbb{E}[|F(Y)|1_{A_n^c}]$$

and again the first term goes to $0$ by our choice of $\varepsilon_n$ and the second term goes to $0$ by the monotone convergence theorem.

| cite | improve this answer | |
$\endgroup$

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.