2
$\begingroup$

I wonder if it is possible to derive an analytical form for the Fourier transform of the indicator function of a semi-ball: $$g_1(\underline{\xi}) = \int_{\mathcal{B}_+(R)} e^{i \underline{\xi} \cdot \underline{X} } \mathrm{d}\underline{X} $$ with $\underline{\xi}$ a vector of $\mathbb{R}^3$ and $\mathcal{B}_+(R) = \{ \underline{X} \in \mathbb{R}^3 : \lVert\underline{X}\rVert_2 \leq R, X_3 \geq 0\}$ a demi-ball. The Fourier transform of the unit-ball is itself well-known (as well as the unit-sphere): $$g_0(\underline{\xi}) = \int_{\mathcal{B}(R)} e^{i \underline{\xi} \cdot \underline{X} } \mathrm{d}\underline{X} = \frac{4\pi}{\xi^3}\left(\sin(R\xi)-R\xi \cos(R\xi)\right) $$
with $\xi = \lVert\underline{\xi}\rVert_2 $. Unfortunately, the symmetries applicable for these two calculations do not hold for a semi-ball. Note that $g_1(\underline{\xi}) + \overline{g_1(\underline{\xi})} = g_1(\underline{\xi}) + g_1(-\underline{\xi}) = g_0(\underline{\xi})$ so the only unknown lies in the imaginary part of $g_1$.

One month ago, I asked this question on math.stackexchange but it seems that this issue was not covered in existing Fourier transform literature and is therefore a research question more suitable to this site; sorry for the cross-posting. @ShannonStarr did suggest a number of leads, but I could not conclude.

A one-dimensional integral has been obtained, through the following manipulation, but it is probably a dead end: \begin{align}g_1(\underline{\xi}) &= \int_0^1 \left( \int_0^{r} J_0\big(\xi_\rho w\big) \exp\left(i \xi_3 \sqrt{r^2-w^2}\right) \frac{w/r}{\sqrt{r^2-w^2}}\, dw\right)\, dr \\ &= \int_0^{1} \left( \int_w^{1} \exp\left(i \xi_3 \sqrt{r^2-w^2}\right) \frac{w/r}{\sqrt{r^2-w^2}}\, dr\right) dw \\ &= \int_0^{1} \frac{w}{\sqrt{1-w^2}}\, \left(\int_0^1 \frac{e^{i a t}}{b^2+t^2}\, dt \right)dw \end{align} with $\xi_{\rho} = \sqrt{\xi_1^2+\xi_2^2}$, $a = \xi_3 \sqrt{1-w^2}$, $b = w/\sqrt{1-w^2}$ and the following Mathematica integral $$\int_0^1 \frac{e^{i a t}}{b^2+t^2}\, dt = \frac{i}{2b}[e^{-ab}(-\Gamma(0,-ab)+\Gamma(0,-a(b + i))-\ln(1-ib)+\ln(-ib)-\ln(-ab)+\ln(-a(b+i)) + e^{ab}(\Gamma(0,ab)+\Gamma(0,a(b -i))-\ln(1+ib)+\ln(ib)-\ln(ab)+\ln(a(b-i))].$$

Other methods have been discussed, with even less success. Any suggestions/answers?

$\endgroup$

1 Answer 1

3
$\begingroup$

Without loss of generality I can rotate the $x_1$ and $x_2$ axes so that the $x_2$ component of the vector $\xi$ is zero, $\xi=(\xi_1,0,\xi_3)$. For the integral over $x_1,x_2$ I use polar coordinates, $$g_1(\xi_1,\xi_3)=\int_0^1 dx_3\, e^{i\xi_3 x_3}\int_0^{\sqrt{1-x_3^2}}rdr\int_0^{2\pi}d\phi\, e^{i\xi_1r\cos{\phi}}$$ $$\qquad=\frac{2\pi }{\xi_1}\int_0^1 dx_3\, e^{i\xi_3 x_3}\sqrt{1-x_3^2} \,J_1\left(\xi_1 \sqrt{1-x_3^2}\right).$$ I don't think this integral over a Bessel function can be reduced any further in terms of known functions — but at least you have a compact expression.

$\endgroup$
1
  • 1
    $\begingroup$ Nice expression. I have the feeling that a series enables more efficiency for numerical computation so I would express that as follows: \begin{align}g_1(\xi_\rho,\xi_3) - g_0(\xi) &= \frac{2\pi}{\xi_\rho} \Im\left(\int_0^1 e^{i\xi_3\sqrt{1-t^2}}\frac{t^2}{\sqrt{1-t^2}}J_1(\xi_\rho t)\mathrm{d}t\right) \\ &= i\frac{2\pi}{\xi_\rho} \sum_{n=0}^{+\infty} (-1)^n\frac{(\xi_3)^{2n+1}}{(2n+1)!}\int_0^1 t^2(1-t^2)^nJ_1(\xi_\rho t)\mathrm{d}t \\ & = i\frac{2\pi\xi_3}{(\xi_\rho)^2} \sum_{n=0}^{+\infty} \left(-\frac{2(\xi_3)^2}{\xi_\rho}\right)^n\frac{n!}{(2n+1)!}J_{n+2}(\xi_\rho) \\ \end{align} $\endgroup$
    – CNS
    Feb 20 at 13:55

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.