2
$\begingroup$

I tried to ask this question on MSE (link), but got no comments or answers. So, I hope someone on MO would advise.

Given a set of functions $f_{mv}(r,\phi)=J_{v}(k_{mv}r)\cos(v \phi)$ in polar coordinates, where $J_{v}$ denotes Bessel's function of the first kind of $v$th order and $k_{mv}$ denotes the $m$th root of $J'_{v}(k_{mv}a)=0$ at boundary $r=a$, how do we prove that these functions are orthogonal iff $v$ is integer?

[Note that they are tested for orthogonality over $v$ and $m$, as they will be later used to express functions by expansions like $\sum\limits_{v}\sum\limits_{m}c_{mv}f_{mv}$].

And then, for a case where we have rational $v$ (ratio of two integers, $v=M/N$), is there a way to construct a new orthogonal set that is of similar form to the one above (or linearly re-adjust/combine the terms in the original one above to make it orthogonal under such condition)?

$\endgroup$
  • $\begingroup$ @paul It is intended: if and only if (=iff). Apologies. $\endgroup$ – user135626 Apr 3 '17 at 11:35
  • $\begingroup$ @paul Question was updated to fix this. $\endgroup$ – user135626 Apr 3 '17 at 11:41
  • $\begingroup$ @Francois Done. $\endgroup$ – user135626 Apr 3 '17 at 16:20
2
$\begingroup$

Your functions $r\mapsto J_\nu(kr)$ are orthogonal because they are eigenfunctions of a self-adjoint Sturm-Liouville problem. This follows from Lagrange's (a.k.a. Green's) identity, or most explicitly from the relation (e.g. Lebedev (5.14.3, 5.14.9)) $$ \int_0^a J_\nu(\alpha r)J_\nu(\beta r)\,r\,dr = \frac{a\beta J_\nu(\alpha a)J_\nu'(\beta a)-a\alpha J_\nu(\beta a)J_\nu'(\alpha a)}{\alpha^2-\beta^2}. $$ For non-integral $\nu$, this still works but the product with $\cos(\nu\phi)$ won't be single-valued in polar coordinates...

$\endgroup$
  • $\begingroup$ If we could ignore or avoid the angular multi-valuedness for a moment (e.g. by limiting the chosen domains of $\phi$ in practice), would the product itself be orthogonal for different values of $v$ when $v=M/N$? $\endgroup$ – user135626 Apr 3 '17 at 11:57
  • $\begingroup$ @user135626 Rather the opposite: if $\mu, \nu$ have common denominator $N>1$ (e.g. $\mu=0$, $\smash{\nu=\frac12}$), then $\smash{\int_0^b e^{i(\nu-\mu)\phi}d\phi=0}$ generally requires $b$ to be a multiple of $2N\pi$, not just $2\pi$, i.e. you must expand the domain to an $N$-fold cover of the circle. $\endgroup$ – Francois Ziegler Apr 3 '17 at 14:35

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.