If $X \sim \mathrm{Normal}(\mu,\sigma^2)$ and $Y \sim \mathrm{Normal}(\mu,\sigma^2)$ are independent random variables, how do I prove that $X+Y$ and $X-Y$ are also independent?

What happens with the independence between $X+Y$ and $X-Y$ when $X \sim \mathrm{Normal}(\mu_x,\sigma_x^2)$ and $Y \sim \mathrm{Normal}(\mu_y,\sigma_y^2) $

Thank you

  • 1
    $\begingroup$ Given that $X$ and $Y$ are independent normal, $X+Y$ and $X-Y$ are independent if and only if $\sigma_X^2=\sigma_Y^2$. $\endgroup$ – Did Oct 3 '12 at 6:21
  • $\begingroup$ You only need to prove their covariance is 0 since uncorrelation implies independence for normal distribution. $\endgroup$ – JACKY Li Oct 3 '12 at 6:21
  • 4
    $\begingroup$ @Patrick: that is not true. If $X$ has a standard normal distribution and $Y=X$ when $|X| \gt k \approx 1.538$ and $Y=-X$ when $|X|\le k$ then correlation is $0$ and $Y$ also has a standard normal distribution but they are not independent. $\endgroup$ – Henry Oct 3 '12 at 6:42
  • $\begingroup$ @Henry Thank you for pointing it out. They need to be jointly normal distributed. $\endgroup$ – JACKY Li Oct 3 '12 at 11:23
  • $\begingroup$ math.stackexchange.com/q/3831387/321264 $\endgroup$ – StubbornAtom Sep 18 '20 at 18:13


  1. Independent Gussian random variables make a Gaussian random vector.
  2. Affine transform $Y=A X + b$ of Gaussian random vector $X$ is Gaussian.
  3. Distribution of Gaussian random vector is determined by its mean vector, and covariance matrix.
  4. If components $X_i$ and $X_j$ of the Gaussian random vector are independent, then $\mathbb{Cov}(X_i, X_j) = 0$.

Combining facts given above, it follows that evaluation of $\mathbb{Cov}(X+Y,X-Y)$ will help establish the result needed.

  • 1
    $\begingroup$ No, you've got your "if" and "then" in the wrong order to get the conclusion you want. You want to prove independence. You need to show that IF the covariance is $0$ THEN they're independent, NOT that IF they're independent THEN the covariance is $0$. $\endgroup$ – Michael Hardy Oct 3 '12 at 4:58
  • $\begingroup$ @MichaelHardy Yes, I agree. I was saying that since the mean vector and covariance matrix determine the distribution of a Gaussian vector, zero cross-covariance $\mathbb{Cov}(X_i,X_j)=0$ implies independence of $X_i$ and $X_j$. $\endgroup$ – Sasha Oct 3 '12 at 5:29
  • $\begingroup$ Hence one should add 0. Independent Gaussian random variables make Gaussian vectors. $\endgroup$ – Did Oct 3 '12 at 5:34
  • $\begingroup$ @did Thanks. Added. $\endgroup$ – Sasha Oct 3 '12 at 5:38

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.