If $X \sim \mathcal{N}(0,\sigma^2)$, then $X^2$ is distributed according to a scaled chi-square distribution.

If $X \sim \mathcal{N}(\mu,1)$, then $X^2$ is distributed according to a noncentral chi-square distribution.

But what about the case when $X \sim \mathcal{N}(\mu,\sigma^2)$, is the distribution of $X^2$ known in this situation?

  • 4
    $\begingroup$ Must be a scaled noncentral $\chi^2(1)$. Also, did you mean $\sigma$ or $\sigma^2$ in the last line? $\endgroup$ Nov 17, 2020 at 12:36
  • 7
    $\begingroup$ Observe that $X/\sigma\sim\mathcal{N}(\mu/\sigma,1)$ and (immediately) draw your conclusion. $\endgroup$
    – whuber
    Nov 17, 2020 at 12:39
  • 2
    $\begingroup$ Does this answer your question? Distribution of a quadratic form, non-central chi-squared distribution $\endgroup$
    – Xi'an
    Nov 17, 2020 at 13:30

3 Answers 3


Graphs in R per @whuber's Comment:

enter image description here


w = rnorm(10^6, 150, 15)
summary(w); sd(x)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  77.08  139.86  150.01  150.02  160.15  221.71 
[1] 1.000814

hdr1 = "W ~ NORM(150, 15)"
hist(w, prob=T, br=30, col="skyblue2", main=hdr1)
 curve(dnorm(x, 150, 15), add=T, col="red")


x = w/15
summary(x); sd(x)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  5.138   9.324  10.001  10.001  10.676  14.781 
[1] 1.000814

hdr2 = "X = W/15 ~ NORM(10, 1)"
hist(x, prob=T, br=40, col="skyblue2", main=hdr2)
 curve(dnorm(x, 10, 1), add=T, col="red")

See Wikipedia on non-central chi-squared distribution. Notice that the mean of $m = 10^6$ observations from $Y \sim \mathsf{Chisq}(\nu=1,\lambda=10^2)$ is consistent with $E(Y) = \nu+\lambda=101.$

y = x^2
summary(y); sd(y)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  26.40   86.94  100.01  101.02  113.99  218.48 
[1] 20.07186

hdr3 = "Y ~ CHISQ(DF=1, NCP=100)"
hist(y, prob=T, br=30, col="skyblue2", main=hdr3)
 curve(dchisq(x,1,100), add=T, col="red")


There is a really good proof of this on https://online.stat.psu.edu/stat414/lesson/16/16.5. To prove the theorem, it is needed to show that the p.d.f. of the random variable is the same as the p.d.f. of a chi-square random variable with 1 degree of freedom

  • 5
    $\begingroup$ If the link goes dead this will not prove very helpful. Can you abstract the main contents of the link and edit them into your question? $\endgroup$
    – mdewey
    Nov 17, 2020 at 14:36

Ok so I think I can answer this now based on the suggestions in the comments.

We have $X \sim \mathcal{N}(\mu, \sigma^2) = \sigma \mathcal{N}(\frac{\mu}{\sigma}, 1)$.

So $X/\sigma \sim \mathcal{N}(\frac{\mu}{\sigma},1)$ and thus $X^2/\sigma^2$ is is distributed according to $\chi_{1,\frac{\mu^2}{\sigma^2}}^2$, a noncentral chi-square distribution where the subscripts indicate we have $1$ degree of freedom and a noncentrality parameter $\frac{\mu^2}{\sigma^2}$, respectively.

Then $X^2 \sim \sigma^2 \chi_{1,\frac{\mu^2}{\sigma^2}}^2$. The pdf of $\sigma^2 \chi_{1,\frac{\mu^2}{\sigma^2}}^2$ is $$ f_{\sigma^2 \chi_{1,\frac{\mu^2}{\sigma^2}}^2}(x) = \frac{1}{\sigma^2}f_{\chi_{1,\frac{\mu^2}{\sigma^2}}^2}\bigg(\frac{x}{\sigma^2}\bigg), $$ where $f_{\sigma^2 \chi_{1,\lambda}^2}(x)$ is defined as $$ f_{\sigma^2 \chi_{1,\lambda}^2}(x) = \frac{1}{2}e^{-(x+\lambda)}\bigg(\frac{x}{\lambda}\bigg)^{-1/4}I_{-1/2}(\sqrt{\lambda x}). $$

  • 1
    $\begingroup$ When you divide by $\sigma$ you need to consider what that does to the mean $\endgroup$ Nov 17, 2020 at 22:45
  • $\begingroup$ Thanks, I've edited it to properly account for the division by $\sigma$. $\endgroup$
    – Bertus101
    Nov 18, 2020 at 9:54
  • 1
    $\begingroup$ en.wikipedia.org/wiki/… $\endgroup$
    – quester
    Nov 18, 2020 at 10:27
  • $\begingroup$ Thanks, I overlooked the fact that the normality parameter was based on the sum of means squared (I mistakenly thought it was just sum of means). $\endgroup$
    – Bertus101
    Nov 18, 2020 at 15:33

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.