Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top
Your link unfortunately returns a "file expired" error for me. I assume you know that $d\sigma/\sigma$ is the same as $d ln(\sigma)$, so that a change of variable may give a useful simplification. – hardmath Dec 15 '10 at 18:02
up vote 3 down vote accepted

don't know exactly what you want to know about $d\sigma/\sigma$.

mathematically, it is like lebesgue measure $dx$ on $\mathbb{R}$ - in the following sense:

$dx$ does not change under a simple additive change [or shift] of a variable: $y= x+a$, where $a$ is a constant. that is, $dy=dx$.

similarly, $d\sigma/\sigma$ does not change under a multiplicative change [or rescaling] of a variable: $y=c\sigma$, where $c>0$. that is, $dy/y=d\sigma/\sigma$.

both $dx$ on $\mathbb{R}$ and $d\sigma/\sigma$ on $\mathbb{R}^+$ are examples of haar measures [i.e., invariant measures] on $\mathbb{R}$ and $\mathbb{R}^+$, respecctively, considered as groups acting on themselves, as indicated.

as for jeffries: many [non-orthodox] bayesians consider $dx$ to be like a uniform [prior] distribution on $\mathbb{R}$. [it is the limit - in some sense - of the uniform distribution on [-A, A], as A $\to\infty$.] it is considered to be an 'objective prior', in that everyone [supposedly] agrees that it is THE uniform distribution on $\mathbb{R}$.

that $dx$ is improper - it is not actually a probability distribution on $\mathbb{R}$ - doesn't seem to faze these folks; as long as one comes up with a proper posterior distribution when using it [as is often the case], it can be taken as a 'prior'. an example is inference about the unknown mean $\mu$ of a normal distribution with $\sigma = 1$. if $X$ is one observation from this distribution, the conditional pdf of $X|\mu$ is

$$f_{X|\mu}(x|\mu) = \frac{1}{\sqrt{2\pi}}e^{\frac{1}{2}(x-\mu)^2},\quad -\infty < x < \infty.$$

multiplying by the 'uniform' prior 'pdf' of $\mu$, the [improper] joint pdf of $(X,\mu)$ is

$$f_{X,\mu}(x,\mu) = f_{X|\mu}(x|\mu){\mathbb{1}}_{(-\infty.\infty)}(\mu) = \frac{1}{\sqrt{2\pi}}e^{\frac{1}{2}(x-\mu)^2},\quad -\infty < x, \mu <\infty.$$

note that the [improper] prior 'pdf' for $\mu$ is $f_\mu(\mu)={\mathbb{1}}_{(-\infty,\infty)}(\mu)$, in that

$$\mathrm{'prob'}(a< \mu< b) = \int_a^b f_\mu(\mu) d\mu = \int_a^b d\mu = b-a.$$

then, integrating out the $\mu$ in the joint pdf, one gets that the marginal 'pdf' of $X$ is

$$f_X(x) = \int_{-\infty}^{\infty} f_{X|\mu}(x|\mu)d\mu = \int_{-\infty}^{\infty} \frac{1}{\sqrt{2\pi}}e^{\frac{1}{2}(x-\mu)^2}d\mu = {\mathbb{1}}_{(-\infty.\infty)}(x),$$

so that $X$ marginally has an improper distribution as well. finally, the posterior pdf for $\mu$ is

$$f_{\mu|X}(\mu|x)= f_{X,\mu}(x,\mu)/f_{X}(x) = \frac{1}{\sqrt{2\pi}}e^{\frac{1}{2}(x-\mu)^2},\quad -\infty < \mu < \infty,$$

so that $\mu|X$ is normal with mean $X$ and SD = 1.

with this example in mind, perhaps you can persuade yourself that $d\sigma/\sigma$ is some sort of analog of a uniform distribution on $\mathbb{{R}^+}$. [if one considers the parameter $\nu=\log\sigma $, this turns into the [lebesgue] uniform distribution for $\nu$, since [as pointed out by @hardmath], $d\nu=d(\log\sigma)=d\sigma/\sigma$.

whether or not you buy into this depends, i guess, on your propensity for being a 'true believer'.

share|cite|improve this answer
Thank you very much for this answer. I guess I am a 'true believer.' – Neil G Dec 17 '10 at 7:43

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.