Let $S^{n-1}$ be the unit sphere in $\mathbb R^n$ and $\Gamma_n$ the collection of great circles on it. Assume $n\geq3$. The Funk transform of a function $f:S^{n-1}\to\mathbb R$ is a map $Ff:\Gamma_n\to\mathbb R$ defined by $$ Ff(\gamma)=\int_\gamma f \,ds. $$ It is clear that $F$ annihilates all odd functions. If we restrict our attention to smooth functions $f$, then this is the whole kernel of $F$.

It is a classical result (due to Funk) that if $n=3$, the Funk transform $Ff$ indeed determines the even part of $f\in C^\infty(S^2)$. If $n\geq3$, we can reduce the problem to this result by looking at $S^2$ — the sphere $S^{n-1}$ is a union of totally geodesic copies of $S^2$ in an obvious way, and we can recover the even part separately on each of these $S^2$s.

Assuming smoothness looks like a technical convenience rather than a necessary assumption, and I believe the result to remain true with much less regularity. Where could I find a (citable) proof of injectivity of the Funk transform for even $L^1$ functions? If $L^1$ is to much, other spaces are welcome, too, but I would like the space to include piecewise continuous functions.

If $f\in L^1$, then $Ff(\gamma)$ is not pointwise well defined; changing $f$ in a null set can change $Ff(\gamma)$. Therefore the Funk transform has to understood in a weaker sense. Changing $f$ in a null set should only affect $Ff(\gamma)$ for a null set of $\gamma$s (for a suitable measure on $\Gamma_n$), so that $Ff$ is well defined almost everywhere. The Radon and X-ray transforms (where one integrates over hyperplanes or lines in $\mathbb R^n$) have been studied for very low regularity (see eg. these lecture notes), so one should be able to study the Funk transform of $L^1$ functions as well.

It is possible that there is a nice duality argument showing that if $F$ is injective on even smooth functions, then it is injective on even distributions, but I don't recall seeing one. I know such an argument for Lie groups, but $S^{n-1}$ is only a symmetric space (for most $n$).

  • 1
    $\begingroup$ Is it even clear $Ff$ is well defined for $f\in L^1$? (What happens if I change $f$ on a null set?) $\endgroup$ – Christian Remling Jul 8 '15 at 22:29
  • $\begingroup$ @ChristianRemling, it should be enough to have $Ff(\gamma)$ well defined for almost every $\gamma$. Changing $f$ in a null set may change some values of $Ff$, but only in a null set of great circles. Analyzing the Radon or X-ray transform (where one integrates over lines or hyperplanes of $\mathbb R^n$) of $L^p$ functions or even distributions has been done. Of course, the proof should come with a suitable weak definition of the Funk transform, but I don't think this is a big issue. $\endgroup$ – Joonas Ilmavirta Jul 9 '15 at 8:55
  • $\begingroup$ @JoonasIlmavirta Do you have a recommended reference for the injectivity of the Funk transform on the space of smooth even functions where $n=3$? (There is always the original paper I guess, but perhaps you know a better presentation). Thanks. $\endgroup$ – Asaf Shachar Oct 31 '17 at 6:42
  • $\begingroup$ @AsafShachar See Helgason's The Radon Transform, Theorem 1.7, Section 1B, Chapter III. I don't know if anyone has published a more readable proof. The transform is also discussed in some detail in section 4 of this paper (also on arXiv). $\endgroup$ – Joonas Ilmavirta Oct 31 '17 at 16:37

One can define the Radon transform $F$ on the space of generalized functions $F\colon C^{-\infty}(S^{n-1})\to C^{-\infty}(Gr_{n,2})$ by duality to the space of smooth functions, where $Gr_{n,2}$ denotes the Grassmannian of 2-planes in $\mathbb{R}^n$. (In particular $F$ is defined on $L^1$.) (Here I am using the identity $$\int_{S^{n-1}}fG(g)=\int_{Gr_{n,2}}F(f)g,$$ where $G$ is the Radon transform from $Gr_{n,2}$ to $S^{n-1}$ and $f,g$ are infinitely smooth functions.)

The claim is that the kernel of $F$ on generalized functions (resp. $L^1$ functions) is equal precisely to odd generalized functions (resp. odd $L^1$ functions).

To prove this, notice that the group $SO(n)$ intertwines $F$. Hence the kernel of $F$ of $SO(n)$-invariant closed subspace (in both weak* and strong topologies). It is well known that

1)for such a subspace the space of $SO(n)$-finite functions is dense (a vector in a representation is called $SO(n)$-finite if it is contained in a finite dimensional $SO(n)$-invariant subspace). (see Helgason's book "Groups and geometric analysis", Ch. IV, Lemma 1.9.)

2) All $SO(n)$-finite functions in $C^{-\infty}(S^{n-1})$ are smooth. (This follows e.g. from Ex. 1(i) on p. 481 in the above book by Helgason.)

(Actually both statements (1) and (2) are true for any homogeneous space of a compact Lie group).

But we know that smooth functions from the kernel of $F$ are odd. Hence their closure in the generalized functions coincides with the odd functions.

  • $\begingroup$ If I understood correctly, for smooth functions $\int_{S^{n-1}}fFg=\int_{S^{n-1}}gFf$ and it is this identity that gives an easy definition by duality, and $F$ is continuous for smooth functions and distributions. Then a distribution in the kernel can be approximated by smooth functions in the kernel, whence the distribution has to be odd. Do you have a reference for the tools you used (about $SO(n)$-finite functions)? (Side note: Your $F$ agrees with my $F$ only for $n=3$, but your result is enough to give what I want for $L^1$ functions, so that's fine.) $\endgroup$ – Joonas Ilmavirta Jul 10 '15 at 16:51
  • $\begingroup$ 1) Your equality is correct. 2)I will try to find a reference. 3) I think your $F$ and my $F$ agree for any $n$. $\endgroup$ – makt Jul 10 '15 at 19:11
  • $\begingroup$ It seems to me that you are integrating over a sphere of codimension one but I'm integrating over a sphere of dimension one (a circle). That is, your $F$ is the Radon transform and mine is the X-ray transform. If you are using circles instead, how do you parametrize them with $S^{n-1}$? $\endgroup$ – Joonas Ilmavirta Jul 10 '15 at 19:20
  • $\begingroup$ You are right, sorry. I have made appropriate corrections in the answer. In fact the whole argument works even in greater generality if you integrate over great spheres of any dimension $0<k<n-1$. $\endgroup$ – makt Jul 11 '15 at 6:44
  • $\begingroup$ Excellent, thank you! I will check the book as soon as I can lay my hands on it. $\endgroup$ – Joonas Ilmavirta Jul 11 '15 at 12:53

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.