I would like to compute the average distance travelled by particle points at constant speed $v>0$ with uniformly-distributed directions and placed uniformly at random inside a hollow sphere of radius R until it hits the sphere.

My guess would be to initiate the computation as

$$\frac{1}{V}\int_0^R f(r)4\pi r^2dr$$

where $f(r)$ would be the mean distance to the sphere, with $r \leq R$ (as this function is obviously radial).

My problem comes from this question :

enter image description here

To get $f(a)$, should I integrate over $\color{blue}{\theta}$ or $\color{green}{\phi}$?

My first instinct was to average over $\color{blue}{\theta}$, can somebody confirm that I am doing well? The instinct comes from my assumption that the direction of the particle is uniformly distributed around itself, not around the origin O, but perhaps due to integration/infinitesimal trickery the correct answer is actually $\color{green}{\phi}$? I am befuddled as the answers seem to point at the fact that averaging over an inner sphere amounts to integrating over $\color{green}{\phi}$, which I guess is what I want to do if I were to place this before $4\pi r^2 dr$, but this contradicts my basest intuition which points at the uniformity of directions of $\vec v$. So basically, what should prevail here, uniformity of directions ($\color{blue}{\theta}$) or averageness over an inner sphere ($\color{green}{\phi}$)?

Edit: actually it seems like $\color{green}{\phi}$ provides an isotropic averaging over the sphere, while $\color{blue}{\theta}$ provides a uniform averaging over directions. But uniform averaging over directions does not translate into isotropic averaging over the sphere because of asymmetry, so $\color{blue}{\theta}$ still seems to be the way to go. However, I would like some feedback anyway please.

Sorry if this sounds elementary, I've not practiced physics in a long time, so I am full of self-doubt. If the quantitative answer matters to passive viewers, both answers are to be found in the question linked above. In particular, I am not asking for the computation as it is already available and would be redundant.

Thank you for your answers.

  • $\begingroup$ Note that blue-green color blindness is a thing, and about 1 in 500 people have it, which makes it likely that possible answerers here might have it. Consider changing one of the $\theta$'s to $\phi$ in the diagram and the text rather than distinguishing them by color. $\endgroup$
    – march
    Commented Apr 29 at 23:13
  • 1
    $\begingroup$ @march Done, thanks for the tip! $\endgroup$
    – Evariste
    Commented Apr 29 at 23:20

1 Answer 1


This one, you just gotta break down to all the degrees-of-freedom, and slowly reject the trivial ones. There's too many to wing it.

Also: when on/in the sphere, don't deal with polar angle, deal with its cosine. Also, bring azimuth into the mix with:

$$ d^2\Omega=\sin\theta d\theta d\phi = d(\cos\theta)d\phi = dz d\phi $$

where the "dz" thing is not common, but I am going to use it--it's not unheard of. What it does, as you can tell, is: it turns your uniform sphere into a uniform rectangle, and that is convenient.


$$ d^3\vec r = r^2drd^2\Omega \equiv d^3V $$

OK, for colors, though they are nice, I don't have the skills. Unprimed is spacial angle, and primed are in velocity space

$$ d^3\vec v = v^2dv d\Omega'$$

(save that part for Maxwell's Distribution).

Though, I think, you have $|v| =v $ fixed? If so, we need to deal only with $d\Omega'$.

Now consider the distance as a function of all the coordinates--hang on, using "d" is dumb, since it's also the differential...I will call it $x$, no, $l$..no, looks like $1$...$L$:

$$ L(\vec r, \Omega') = {\rm ?} $$

(It would have been nice had you supplied the $?$", but well go with that abstract form and figure it out):

$$ L(\vec r, \Omega') = {\rm ?} $$

So, assuming:

$$ \int_V\int_{\Omega'} d^3Vd^2\Omega'= N$$

(if not, adjust accordingly), we can write the full integral for the expectation value of $L$:

$$ \langle L \rangle = \frac 1 N \int_V\int_{\Omega'}L(\vec r, \Omega')d^3Vd^2\Omega'$$

Now we can expand that into coordinates:

$$ \bar{L} = \frac 1 N\int_{r=0}^R\int_{z=1}^1 dz\int_{\phi=0}^{2\pi}d\phi \int_{z'=1}^{-1}dz'\int_{\phi'=-}^{2\pi} L(r, z,\phi,z',\phi')d\phi' $$

Now we can start chucking coordinates:

$$ L(r, z,\phi,z',\phi') = L(r, z') $$

(by symmetry), so we can do the constant integrals:

$$ \bar{L} = \frac 1 N 8\pi^2\int_{r=0}^Rr^2dr\int_{z'=1}^{-1} dz'L(r, z') $$

So now you need to use the formula for $L(r, z')$ and verify the $N=\langle 1\rangle$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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