I have been trying to understand the proof that the correlation between two random variables $X$ and $Y$ is between $-1$ and $1$. For simplicity, suppose $X$ and $Y$ have mean zero. Then


I know that there's a proof that doesn't use the Cauchy-Schwarz inequality but I'd like to understand the one that does.

To show that correlation cannot exceed $1$, I want to show that:

$$\mathbb E[XY] \le \sqrt{\mathbb E[X^2] \mathbb E[Y^2]}.$$

Now this does look exactly like one half of the CS inequality ($\rho \ge -1$ is the other half) but we have the $\mathbb E[\cdot]$ operator wrapped around things and instead of inner products of vectors, $\langle x,y \rangle $, we have products of random variables. The proofs I've seen stop here and say "true by the CS inequality". Apparently there is a probabilistic version of it.

My question is, starting from one side of the CS inequality for vectors I know from linear algebra $$\langle x,y \rangle \le \sqrt{\langle x,x \rangle}\sqrt{\langle y,y \rangle},$$ are there steps that actually take me to the above equation?

Edit: The answer and comments helped, although I simply don't know enough math to really follow. If someone else is in the same boat, I found a link that explicitly goes into why and how random variables can be thought of as vectors in a vector space: http://www.randomservices.org/random/expect/Spaces.html

  • 1
    $\begingroup$ The wikipedia article of the Cauchy-Schwarz inequality seems to cover it okay. $\endgroup$
    – Glen_b
    Oct 10, 2014 at 22:56
  • 2
    $\begingroup$ You only have to prove that the sum of cross-products of the vector's coordinates are these vectors' dot product. Covariance is thus a dot product (when variables are centered). Then from the definition of "dot product" and your first formula it follows that correlation is cosine part (of that definition). Cosine has bounds [-1,1]. $\endgroup$
    – ttnphns
    Oct 11, 2014 at 0:32

1 Answer 1


The Cauchy-Schwarz inequality is the same. Only the inner product is a different one. You might have seen $E(X)$ written as an integral? $$<X,Y>:=\int_{\Omega} XY\mathrm{d}P$$ for real random variables on a probability space $\Omega$ with probability measure is the solution. Now you have to check the axioms, in particular definiteness. To get the idea there, remember that "$=$" is to strict in probability, since we don't have to care about what happens on sets with measure $0$, i.e. for events that never occur.


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.