A well-known result is that the only density that is both spherical and independent error is normal: more precisely

Let $e_i$ be errors,

If the joint probability density satisfies $$f_n(e_1,e_2, ..., e_n) = f_1(e_1)f_1(e_2)...f_1(e_n)$$ (independence) and if it is spherical: $$f_n(e_1,e_2, ..., e_n) = g_n(\Sigma_{i=1}^{n} e_i^2)$$

then, the only densities that satisfy both conditions are the normal densities:

$$f_1(e_i)={1 \over \sqrt{2 \pi}\sigma} \, \exp(-{1\over2\sigma^2} e_i^2)$$

How could I prove it?

I guess taking the derivative of $e_i$ of $f(e_1,e_2, ..., e_n) = f(e_1)f(e_2)...f(e_n)= f(\Sigma_{i=1}^{n} e_i^2)$ would be a first move..

Similar result I just found..

"Theorem [Herschel-Maxwell]: Let Z∈Rn be a random vector for which (i) projections into orthogonal subspaces are independent and (ii) the distribution of Z depends only on the length ||Z||. Then Z is normally distributed.

Cited by George Cobb in Teaching statistics: Some important tensions (Chilean J. Statistics Vol. 2, No. 1, April 2011) at p. 54."

from What is the most surprising characterization of the Gaussian (normal) distribution?


1 Answer 1


This is a standard calculus derivation: spherical symmetry tells you that $f_1(x)$ is a function of $x^2$, i.e. $$f_1(x)=g_1(x^2).$$ Independence plus spherical symmetry tell you that $$g_1(u)g_1(0)=g_2(u) \quad\text{and}\quad g_1(u)g_1(v)=g_2(u+v)\propto g_1(u+v)$$ Therefore, rescaling $g_1$ into $h_1$ so that the above become an equality, we derive the identity $$h_1(u)h_1(v)=h_1(u+v)$$ for which the only solution is of the form $$ h_1(u) = \exp \{\alpha u\},\qquad \alpha\in\mathbb{R} $$ Thus, $$f_1(x) = \dfrac{1}{\sqrt{2\pi\sigma^2}} \exp\{-x^2/2\sigma^2\},\qquad \sigma\in\mathbb{R}_+,$$ since only negative factors $\alpha$ lead to densities.

  • $\begingroup$ thank you! obvious typos : $g_1(u)g_1(v)=g_2(u+v)=g_1(0)g_1(u+v)$ $\endgroup$
    – KH Kim
    Jun 6, 2012 at 10:05
  • $\begingroup$ @KHKim: thanks, but this is not a typo, only the use of a proportionality sign (and the proportionality coefficient is indeed $g_1(0)$) for the next step! $\endgroup$
    – Xi'an
    Jun 6, 2012 at 10:11
  • $\begingroup$ @KHKim: thanks for the $u+1$ instead of $u+v$ typo, indeed! $\endgroup$
    – Xi'an
    Jun 6, 2012 at 11:01

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.