0
$\begingroup$

This is a super soft question in a way, but I was wondering why the central limit theorem says that the distribution of means of converges to the normal distribution. I understand why we might want a symmetric distribution, and I understand why we might want the majority of mass centered at 0 (these are both elements of unbiased means). What I don't get is why the specific pdf we converge to is the normal distribution, and not some other "almost-gaussian" (symmetric, centered at zero) distribution?


This is mainly motivated by the fact that Lagrange's proof of the standard Gaussian integral seems kind of miraculous, but the general strategy seems to be creating a radially symmetric product measure from a symmetric measure in one dimension by taking the measure product with itself. The problem is that, on a calculus level it seems like the normal distribution is kinda the only one Lagrange's proof will apply to, but on the measure theory level I don't really see why this is true. It seems that, a priori, there might be any number of symmetric distributions who's product measure with itself would be radially symmetric. My point is just that... the normal distribution doesn't seem that special, except for the CLT, which makes me think there is something special about it. I just don't get why it has to be the normal and not an almost-normal distribution, except, of course, for the fact that it is the case.


I guess my question is pretty simple, why not just an almost normal?

$\endgroup$
  • $\begingroup$ The derivation of the normal distribution appears in thousands of books and innumerable sites online. Consult these for this very very well and extensively treated topic. $\endgroup$ – David G. Stork Mar 26 at 22:49
  • $\begingroup$ @DavidG.Stork Almost everything I have seen starts at the point of "here is the limit in measure we want, lets find the right inequalities...", similarly, most derivations are like "here is the normal pdf, cdf, et cetera", but if I didn't know the statement of the CLT and wanted to find a distribution for the convergence of sample means, the normal isn't necessarily my first guess, although I would guess something symmetric and centered at zero, but why guess the normal? $\endgroup$ – Juan Sebastian Lozano Mar 26 at 22:56
  • $\begingroup$ My motivation points to a possible reason: historical, but the thing the properties used in the historical proofs don't seem unique either, so that's not really convincing. $\endgroup$ – Juan Sebastian Lozano Mar 26 at 22:58
  • 1
    $\begingroup$ The derivations of the normal from the central limit theorem don't start or assume the form of the final distribution, instead they derive it. See Proof of CLT (en.wikipedia.org/wiki/Central_limit_theorem). What more could you want? $\endgroup$ – David G. Stork Mar 26 at 23:03
  • $\begingroup$ Actually, that's exactly what I want, I guess every proof I've seen has been weird in that way? $\endgroup$ – Juan Sebastian Lozano Mar 26 at 23:06
3
$\begingroup$

The derivations of the normal from the central limit theorem don't start or assume the form of the final distribution, instead they derive it. See Proof of CLT.

$\endgroup$
  • $\begingroup$ Thanks, I feel silly having missed that on the wiki. $\endgroup$ – Juan Sebastian Lozano Mar 27 at 1:32

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.