0
$\begingroup$

In D. Mackay's book on Information Theory he states that we can make a change of variables for a probability density $P(x)$ by for example $l=\ln(x)$. Then, the density of $x$ is transformed to $$P(l)=P(x(l))\left\vert \frac{\delta x}{\delta l}\right\vert = P(x(l))x(l)$$

I'm confused about the notation of $x$ as a function of $l$: $x(l)$. What exactly is $x(l)$? If I got this right, then we have $l=\ln(x) \rightarrow x = \exp(l) $, so $$P(x(l))\left\vert \frac{\delta x}{\delta l}\right\vert = P(x(l))\left\vert \frac{\delta \exp(l)}{\delta l}\right\vert = P(x(l))\exp(l)=P(x(l))x$$ So are we essentially saying $x(l)=\exp(l)$? Is it that trivial and just a notational trick?

$\endgroup$
1
$\begingroup$

When doing change of variables one typically uses a smooth invertible transformation. So if $l = f(x)$ we can also think of $x = f^{-1}(l)$. For notational convenience, many authors simply write $x(l)$ instead of $f^{-1}(l)$. That is, yes it is trivial and just a notational trick.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

$x(l)$ is "$x$ as a function of $l$".   In this case, $x(l)=\exp(l)$

$P(x(l))\lvert \dfrac{\partial x(l)}{\partial l}\rvert = P(\exp(l))\left\lvert\dfrac{\partial \exp(l)}{\partial l}\right\lvert = P(\exp(l)) \cdot\exp(l)$

| cite | improve this answer | |
$\endgroup$

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.