6
$\begingroup$

Prove that $E[X\mid X^3-3X]=0$, with $f(x)=\frac{|x^2-1|}{4}$ being the density function of $X$ in the interval $[-2,2]$.

My attempt:

Let $Y=X^3-3X$ and $x_1, x_2, x_3$ the roots of $x^3-3x=y$. \begin{gather*} P_{X\circ Y}(A\times B)=P_\Omega(X\in A\cap Y\in B)=\int_{y\in B}I_A(x)f(x)dx=\\ =\int_B\sum_{i=1}^3I_A(x_i)f(x_i)/|3x_i^2-3|dy=\int_B\frac{\sum I_A(x_i)f(x_i)/|3x_i^2-3|}{\sum f(x_i)/|3x_i^2-3|}dP_Y \end{gather*} From where: $$P_{X/Y}(y)(A)=P_\Omega[X\in A|Y=y]=\frac{1}{3}\sum I_A(x_i)$$ Therefore: $$E[X|Y=y]=\frac{1}{3}\sum x_i=0$$

$\endgroup$
8
  • $\begingroup$ What have you tried so far? $\endgroup$
    – Galen
    Commented Dec 22, 2023 at 15:40
  • 1
    $\begingroup$ Hint: Start with the definition of the conditional expectation of a continuous random variable and work back to what you know. $\endgroup$
    – Galen
    Commented Dec 22, 2023 at 19:43
  • $\begingroup$ @Galen Meta-hint: there's a much simpler solution requiring no calculation at all. $\endgroup$
    – whuber
    Commented Dec 22, 2023 at 19:53
  • 1
    $\begingroup$ @whuber I am not suggesting calculation. I am suggesting going back to definitions. Often when I cannot solve a problem like this it is because I have not paid enough attention to the elementary definitions or results. 🙂 $\endgroup$
    – Galen
    Commented Dec 22, 2023 at 19:57
  • 1
    $\begingroup$ @Galen Fair enough -- but applying the definition devolves to computing something. And even a definition isn't needed here: merely the truly basic idea that the conditional expectation is a property of the joint distribution alone, not of the random variables. $\endgroup$
    – whuber
    Commented Dec 22, 2023 at 20:06

1 Answer 1

5
$\begingroup$

This is a great question (+1, where does it come from?), and your final result $0$ is correct. However, your notation and reasoning are hard to follow/comment because they are largely non-standard. So I would like to post a new answer that follows the conventional measure-theoretic conditional expectation framework.

Intuition

To begin with, it is helpful to study patterns of the function $f(x) = \frac{|x^2 - 1|}{4}$ and the function $g(x) = x^3 - 3x$ on the domain $x \in [-2, 2]$. The plots of $f$ and $g$ are shown as follows:

images of functions at stake

From the chart above, it is clear that

  • $\max\limits_{x \in [-2, 2]} |g(x)| = 2$;
  • For any $y \in (-2, 2)$, the equation $g(x) = y$ always has three real roots in $[-2, 2]$. In addition, if denote the three roots from left to right by $x_1, x_2, x_3$ respectively, then $x_1 \in [-2, -1], x_2 \in [-1, 1], x_3 \in [1, 2]$.

In addition, since the quadratic, linear and constant coefficients of the cubic polynomial $x^3 - 3x - y$ are $0, -3$ and $-y$ respectively, it follows by Vieta's formula that
\begin{align*} & x_1 + x_2 + x_3 = 0, \tag{1}\label{1} \\ & x_1x_2 + x_1x_3 + x_2x_3 = -3, \tag{2}\label{2} \\ & x_1x_2x_3 = y. \tag{3}\label{3} \end{align*}

Now for a given $y \in (-2, 2)$, suppose that we have observed $X^3 - 3X = y$, then $X$ has to be one of the three roots $x_1, x_2, x_3$. In view of the distribution pattern of $x_i$'s (i.e., the second bullet point above) and $P(-2 \leq X \leq -1) = P(-1 \leq X \leq 1) = P(1 \leq X \leq 2) = \frac{1}{3}$, it follows by $\eqref{1}$ that \begin{align*} E[X\mid X^3 - 3X = y] = \frac{1}{3}x_1 + \frac{1}{3}x_2 + \frac{1}{3}x_3 = 0. \tag{4}\label{4} \end{align*} Since $\eqref{4}$ holds for all $y$, we can conclude that \begin{align*} E[X\mid X^3 - 3X] = 0. \tag{5}\label{5} \end{align*}

Rigorous Proof

To validate the intuitive answer $\eqref{5}$, we need to examine that $0$ satisfies the two defining properties of the conditional expectation:

  1. $0$ is $\sigma(X^3 - 3X)$-measurable;
  2. For any $G$ in $\sigma(X^3 - 3X)$, the functional equation holds: \begin{align*} \int_G X ~\mathrm dP = \int_G 0 ~\mathrm dP = 0. \end{align*}

Bullet point 1 holds trivially. To verify bullet point 2, note that since the sub-$\sigma$-field $\sigma(X^3 - 3X)$ is generated by sets $\{\{X^3 - 3X \leq y\}: y \in \mathbb{R}\}$, it suffices to show that for any $y \in \mathbb{R}$, \begin{align*} \int_{X^3 - 3X ~\leq~ y} X ~\mathrm dP = \int_{\{x \in [-2, 2]:\; x^3 - 3x~ \leq ~y\}}xf(x)~\mathrm dx = 0, \tag{6}\label{6} \end{align*} where the first equality in $\eqref{6}$ follows from the change-of-variable formula.

When $y \geq 2$, \begin{align*} \int_{\{x \in [-2, 2]:\; x^3 - 3x ~\leq~ y\}}xf(x)~\mathrm dx = \int_{-2}^2 xf(x)~\mathrm dx = 0, \end{align*} hence $\eqref{6}$ holds.

When $y \leq -2$, \begin{align*} \int_{\{x \in [-2, 2]:\; x^3 - 3x ~\leq~ y\}}xf(x)~\mathrm dx = \int_\varnothing xf(x)~\mathrm dx = 0, \end{align*} hence $\eqref{6}$ holds.

When $y \in (-2, 2)$, as in the Intuition block, let $x_1 < x_2 < x_3$ be three roots of the equation $x^3 - 3x = y$. It then follows by $x_1 \in [-2, -1], x_2 \in [-1, 1]$ and $x_3 \in [1, 2]$ that \begin{align*} & \int_{\{x \in [-2, 2]:\; x^3 - 3x~ \leq~ y\}}xf(x)~\mathrm dx \\ =& \int_{-2}^{x_1}xf(x)~\mathrm dx + \int_{x_2}^{x_3}xf(x)~\mathrm dx \\ =& \frac{1}{4}\int_{-2}^{x_1}x(x^2 - 1)~\mathrm dx + \frac{1}{4}\int_{x_2}^1x(1 - x^2)~\mathrm dx + \frac{1}{4}\int_{1}^{x_3}x(x^2 - 1)~\mathrm dx \\ =& \frac{1}{16}\left[(x_1^4 + x_2^4 + x_3^4) - 2(x_1^2 + x_2^2 + x_3^2)\right] - \frac{3}{8} \\ =& \frac{1}{16}\left[(x_1^2 + x_2^2 + x_3^2) + y(x_1 + x_2 + x_3)\right] - \frac{3}{8} \tag{7}\label{7} \\ =& \frac{1}{16}(x_1^2 + x_2^2 + x_3^2) - \frac{3}{8} \tag{8}\label{8} \\ =& \frac{1}{16}[(x_1 + x_2 + x_3)^2 - 2(x_1x_2 + x_1x_3 + x_2x_3)] - \frac{3}{8} \\ =& \frac{1}{16}[0 - 2 \times (-3)] - \frac{3}{8} \tag{9}\label{9} \\ =&~ 0. \end{align*} This shows $\eqref{6}$ holds for $y \in (-2, 2)$. In the calculation above, $\eqref{8}$ used identity $\eqref{1}$, $\eqref{9}$ used identities $\eqref{1}$ and $\eqref{2}$. $\eqref{7}$ is a consequence of \begin{align*} & x_1^4 + x_2^4 + x_3^4 - 3(x_1^2 + x_2^2 + x_3^2) \\ =& x_1(x_1^3 - 3x_1) + x_2(x_2^3 - 3x_2) + x_3(x_3^3 - 3x_3) \\ =& x_1y + x_2y + x_3y = y(x_1 + x_2 + x_3). \end{align*}

This completes the proof.

$\endgroup$
10
  • 1
    $\begingroup$ Much simpler rigorous proof: because the distribution is symmetric about $0$ and the function $x\to x^3-3x$ is odd, its expectation is $0,$ QED. $\endgroup$
    – whuber
    Commented May 20 at 12:44
  • 2
    $\begingroup$ @whuber Sorry, by my standard, I would not call it "rigorous". For example, how does this argument can check for general $y$, $\int_{X^3 - 3X \leq y} XdP = 0$? To put it in another way, $x \mapsto x^3 - x$ is also odd, but does $E[X|X^3 - X] = 0$? The coefficient $-3$ is crucial! $\endgroup$
    – Zhanxiong
    Commented May 20 at 13:04
  • $\begingroup$ @Zhanxiong And how would you prove $P_{X/Y}(y)(A)=\frac{1}{3}\sum I_A(x_i)$? $\endgroup$
    – Speltzu
    Commented May 20 at 14:07
  • $\begingroup$ @Speltzu I am afraid I cannot answer that because as I said at the very beginning of the answer, I don't have any knowledge of your notation system (e.g., I have no idea what $P_{X/Y}$ is)... $\endgroup$
    – Zhanxiong
    Commented May 20 at 14:10
  • $\begingroup$ @Zhanxiong $P_{X/Y}(y)(A)=P_\Omega[X\in A|Y=y]$. $\endgroup$
    – Speltzu
    Commented May 20 at 14:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.