If X is a Gaussian, N($\mu_x, \sigma_x^2$), then I came across the following properties:

1) Y = X + k, k is a constant. Then Y is normally distributed as N($\mu_y, \sigma_y^2$) such that:

$\mu_y = \mu_x + k$

$\sigma_y = \sigma_x$

2) Y = kX, k is a constant. hen Y is normally distributed as N($\mu_y, \sigma_y^2$) such that:

$\mu_y = k*\mu_x$

$\sigma_y = |k|*\sigma_x$

Is there a proof for the above two statements? Thanks in advance.

{ Edit: I have modified the original question.

Original question: Deriving Mean and Variance of (constant * Gaussian Distribution).

Due to some misunderstandings, I came to know that the above is not a feasible PDF and that what I actually want is: Deriving Mean and Variance of (constant * Gaussian Random Variable) and (constant + Gaussian Random Variable). }

Edit 2:

If X is Gaussian, prove kX is Gaussian where k is a constant such that k<0.

$P(kX < x)$

$ = P(X > \frac{x}{k})$ (since k<0)

$ = \int_{\frac xk}^\infty \frac{1}{\sqrt{2\pi}\sigma}\exp \left(-\frac{(t-\mu)^2}{2\sigma^2}\right)dt$

I am not able to resolve the limits to get the resulting equation into the form for Cumulative Density Function. How do I proceed?

  • $\begingroup$ If $f(x)$ denotes a distribution's PDF, $cf(x)$ is only a PDF if $c=1$. Perhaps you meant $cf(cx)$. $\endgroup$ – J.G. Oct 21 '19 at 14:18
  • $\begingroup$ Thank you. I see. No. I do mean c*f(x). So the definition of E(x) probably does not hold for this distribution. My reason for this question is, while learning Neural Nets I came across this idea that if f(x) is a standard normal distribution, multiplying it by square_root(2/n) where n is the number of nodes in the layer transforms the distribution to one with variance 2/n. I am looking for more intuition into this. Any help pointing me in the right direction appreciated. Will add this to the original question too. $\endgroup$ – Nitin Oct 21 '19 at 14:25
  • $\begingroup$ It should be $\mu_y=\mu_x+k$ in the first one. $\endgroup$ – Milten Oct 23 '19 at 9:28
  • $\begingroup$ And $\sigma_y=|k|\sigma_x$ in the second one, in case $k$ can be negative. $\endgroup$ – Milten Oct 23 '19 at 9:30
  • $\begingroup$ Sorry. Yeah, I've updated it. $\endgroup$ – Nitin Oct 23 '19 at 9:43

The expressions for the mean and standard deviation are true for any random variable (EDIT: for which $EX$ and $VX$ are defined): $$ E(X+k) = EX+Ek = EX + k, \quad V(X+k)=VX+Vk=VX, $$ $$ E(kX) = kEX, \quad V(kX) = k^2VX $$ I am sure you can find proofs for these easily. But we still need to show that the transformed variables have normal distributions. We can do that as follows. Assume $k>0$. Then: $$ P(kX<x) = P(X<\frac xk) = \int_{-\infty}^\frac xk \frac{1}{\sqrt{2\pi}\sigma}\exp \left(-\frac{(t-\mu)^2}{2\sigma^2}\right)dt $$ $$ = \int_{-\infty}^x \frac{1}{\sqrt{2\pi}k\sigma}\exp \left(-\frac{(u -k\mu)^2}{2(k\sigma)^2}\right)du $$ where we substituted $u=kt$. This is the CDF of $\mathcal N(k\mu, k^2\sigma^2)$, so $kX$ has that distribution. (Note that we rediscovered the mean and variance). The cases $kX$ for $k<0$, and $X+k$ can be shown similarly.

EDIT: The case of $kX$, $k<0$. You showed $$ P(kX<x) = \int_{\frac xk}^{\infty} \frac{1}{\sqrt{2\pi}\sigma}\exp \left(-\frac{(t-\mu)^2}{2\sigma^2}\right)dt $$ We substitute $u=kt$, so $du=kdt$. The lower limit becomes $k\frac xk = x$, and the upper limit becomes $k\infty = -\infty$ (you can rephrase this as a limit, if you want). Plugging in: $$ \int_x^{-\infty} \frac 1k \cdot \frac{1}{\sqrt{2\pi}\sigma}\exp \left(-\frac{(\frac uk -\mu)^2}{2\sigma^2}\right)du $$ $$ = -\int_x^{-\infty} \frac 1{|k|} \cdot \frac{1}{\sqrt{2\pi}\sigma}\exp \left(-\frac{(u -k\mu)^2}{2k^2\sigma^2}\right)du $$ $$ = \int_{-\infty}^x \frac{1}{\sqrt{2\pi}|k|\sigma}\exp \left(-\frac{(u - k\mu)^2}{2(|k|\sigma)^2}\right)du $$ which is the CDF for $\mathcal N(k\mu,(|k|\sigma)^2)$.

| cite | improve this answer | |
  • $\begingroup$ Thank you. Extending for $ X + k $ is pretty simple. However, I am not able to extend for $kX$ when k<0. I have edited the question to add my workings. Could you help me please? $\endgroup$ – Nitin Oct 24 '19 at 16:21
  • 2
    $\begingroup$ Correction: these columns of statements are true respectively if $E(X),\,V(X)$ are defined. $\endgroup$ – J.G. Oct 24 '19 at 16:29
  • 1
    $\begingroup$ @Nitin You are close. You just need to make the substitution $u=kt$. I will edit my answer tomorrow to include it. $\endgroup$ – Milten Oct 24 '19 at 19:12
  • 1
    $\begingroup$ @Nitin I made the edit :) $\endgroup$ – Milten Oct 25 '19 at 7:12

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.