I was trying to show that if two variables have diagonal covariance, this does not necessarily guarantee their independence. For this, I was using an example where $x \sim U(-1,1)$ and $y=X^{2}$ to show this.

Here is my try:

$$ p\left(y | x\right)=\delta\left(y-x^{2}\right) $$

Now we would like to show that although the two variables are dependent, the covariance matrix between them is diagonal. $$ \operatorname{cov}[X, Y]=\mathrm{E}[X Y]-\mathrm{E}[X] \mathrm{E}[Y] $$

$$ p\left(x, y\right)=p\left(x\right) p\left(y | x\right) $$

\begin{align*} E[X Y]=& \iint x y p(x, y) d y d x \\ &= \iint x y p\left(x\right) p\left(y | x\right) d y d x \\ &= \int x p\left(x\right) \int y \delta\left(y-x^{2}\right) d y d x \\ % &= \int y p(y) d y \cdot \int x p(x) d x \\ % &= E[X] E[Y] \end{align*} Here is the deal: I was trying to show that all off-diagonal elements are in fact zero, but I'm stuck at that integral. Any suggestion on how to proceed? Thanks


$E(XY)=E(X^3)=\frac{1}{2}\int_{-1}^1x^3dx=0$. Therefore the covariance $=0$, since $E(X)=0$.

  • $\begingroup$ What about the probability term then? Also, we are explicitly asked to use $p(x, y)=p(x) p(y | x)$ $\endgroup$ – Blade Apr 7 at 22:00
  • 1
    $\begingroup$ Your definition of $p(y|x)$ is unfamiliar to me. Therefore I just went back to basics. $\endgroup$ – herb steinberg Apr 7 at 22:14
  • 1
    $\begingroup$ @Blade Your integral looks straightforward. $\int y\delta (y-x^2)dy=x^2$, so your final integral is $\int x^3p(x)dx$, which is identical to that I wrote. $\endgroup$ – herb steinberg Apr 8 at 3:32

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.