0
$\begingroup$

I'm reading a derivation that confidently and without further comment asserts $$ \int_{x_1=-\infty}^\infty \cdots \int_{x_n=-\infty}^\infty e^{-\sum_i x_i^2}\,dx_n\cdots dx_1 = \int_{r=0}^\infty e^{-r^2}\,\frac{d\text{Volume}(S_r^n)}{dr}\,dr\,. $$ (Where $S_r^n$ is the $n$-dimensional hypersphere with radius $r$.)

Why is this obvious? The only way I know how to do a change of variables is by finding an explicit transformation whose Jacobian I can then derive and stick into the integral together with the new variables. And I'm not even sure how exactly polar coordinates should generalize to $n$ dimensions. Clearly, the author of this derivation was able to sidestep all that with some geometric intuition that I'm not sharing.

$\endgroup$
  • $\begingroup$ The intuition is that the function is spherically symmetric, i.e., depends only on the distance $r=\sqrt{\sum x_i^2}$ from the origin. $\endgroup$ – Hagen von Eitzen May 12 '18 at 20:08
  • $\begingroup$ I see that that's true, but that's not giving me an epiphany. $\endgroup$ – Sebastian Oberhoff May 12 '18 at 20:11
  • $\begingroup$ What are you struggling with in terms of understanding? $\endgroup$ – rubikscube09 May 12 '18 at 20:13
  • $\begingroup$ In 2d there would be a $r\,d\theta$ in place of the $d\text{Volume}/dr$. What would the $n$-dimensional analogue of $r\,d\theta$ be and why is that equal to $d\text{Volume}/dr$? $\endgroup$ – Sebastian Oberhoff May 12 '18 at 20:23
  • $\begingroup$ On the left, you are integrating over all of $\mathbb R^n$. Imagine that space is carved up into a bunch of thin concentric shells centered at the origin. Compute the contribution of each shell, and add up all these individual contributions to obtain the integral on the right. $\endgroup$ – littleO May 12 '18 at 21:15
2
$\begingroup$

Consider the two-dimensional case, using polar coordinates: $$ I = \int_{x=-\infty}^{x=\infty} \int_{y=-\infty}^{y=\infty} e^{-(x^2+y^2)}\,dy\,dx = \int_{r=0}^{r=\infty} \int_{\theta=0}^{\theta=2\pi} e^{-r^2}r\,d\theta\,dr. $$ We can take a constant factor out of the inner integral: $$ I = \int_{r=0}^{r=\infty} \left(e^{-r^2}\int_{\theta=0}^{\theta=2\pi} r\,d\theta\right)\,dr. $$ Now, $r\,d\theta$ is just an element of arc length, that is, $\int_{\theta=0}^{\theta=2\pi} r\,d\theta$ measures the circumference of a circle of radius $r,$ which is the boundary of a disk of radius $r,$ which has the same measure as the rate of change of the area of the disk with respect to $r.$ Putting this in the notation used in the question, $$ \int_{\theta=0}^{\theta=2\pi} r\,d\theta = \frac{d\text{Volume}(S_r^2)}{dr} = 2\pi r. $$ In the three-dimensional case, the inner integral would be $$ \int_{\phi=0}^{\phi=\pi} \int_{\theta=0}^{\theta=2\pi} r^2 \sin\phi\,d\theta\,d\phi = \frac{d\text{Volume}(S_r^3)}{dr} = 4\pi r^2. $$ We can explicitly do higher-dimensional integrals the same way, but the effort we have to make in order to correctly represent the volume element of the inner integral in $n$ dimensions merely obscures the fact that in every case the inner integral is just integrating $1$ over the boundary of an $n$-dimensional ball, and the integral is just the measure of the boundary, which is $\frac{\text{Volume}(S_r^n)}{dr}.$

A possible source of confusion is that this author uses the notation $\text{Volume}(S_r^n)$ where the volume measure of an $n$-dimensional ball is required, but calls $S_r^n$ an "$n$-dimensional hypersphere," which seems inconsistent with the way other authors use that term. We have to resolve that ambiguity in the intended way to see what the factor $\frac{\text{Volume}(S_r^n)}{dr}$ means in this context.

$\endgroup$
0
$\begingroup$

Change of variables works in a quite general way. If we write $B_r = \left\{ x \in \mathbb{R}^n : |x| \leq r \right\}$, then

\begin{align*} \int_{\mathbb{R}^n} e^{-|x|^2} \, \operatorname{Vol}(dx) &= \sum_{i=1}^{\infty} \int\limits_{\frac{i-1}{N} < |x| \leq \frac{i}{N}} e^{-|x|^2} \, dx\\ &\approx \sum_{i=1}^{\infty} e^{-(i/N)^2} \operatorname{Vol}\left(\left\{ x \in \mathbb{R}^n : \tfrac{i-1}{N} < |x| \leq \tfrac{i}{N}\right\}\right) \\ &\approx \int_{0}^{\infty} e^{-r^2} \, d\operatorname{Vol}\left(B_r\right) \\ &= \int_{0}^{\infty} e^{-r^2}\frac{d\operatorname{Vol}\left(B_r\right)}{dr} \, dr, \end{align*}

and this becomes exact as $N\to\infty$. You may think of this as a generalization of the shell method.

$\endgroup$

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.