6
$\begingroup$

Is there a simple way to calculate/estimate the volume of Minkowski sum of an n-dimensional unit ball and an n-dimensional ellipsoid? Even a simple ellipsoid like $\frac{x_1^2}{a^2} + x_2^2 + \ldots + x_n^2 = 1$ will do.

$\endgroup$
1
  • 3
    $\begingroup$ One remark is that the Minkowski sum of an ellipse and a very small round circle gives you the perimeter of the ellipse, which is of course the original source of elliptic integrals. $\endgroup$ Jun 19, 2011 at 21:26

3 Answers 3

6
$\begingroup$

I guess you know that the result can be written as a polynomial. $$w_0+w_1{\cdot}r+\dots+w_n{\cdot}r^n.$$ So your question is how to estmate the coefficients $w_i$; this is so called "cross-sectional measures" and they can be defined for any convex body $K$.

  • $w_0$ is the volume of $K$,
  • $w_n$ is the volume of unit ball.
  • $w_1$ is the area of the boundary of $K$, I think there is no good formula for ellipsoid, but estimates are known and you could write it as an integral.
  • I would be surprised if some remaining $w_i$ can be expressed by simple formula.

If you want to write $w_i$ as an integral, check for example Burago--Zalgaller, Geometric inequalities. For example $$w_i=\mathrm{MayBeAConst}\cdot\int\limits_{\partial K} \sigma_{i-1}(k_1,k_2,\dots,k_{n-1})\\, d\mathrm{area}.$$ where $\sigma_i$ is the $i$-th elementary symmetric polynomial and $k_i$ are principle curvatures. In your case it is easy to find $k_i$...

$\endgroup$
3
$\begingroup$

Because your ellipsoid has all axes but one of equal length we don't need the "full theory":

Let $x_1=:x$ and $(x_2, \ldots, x_n)=:y\in{\mathbb R}^{n-1}$. The Minkowski sum $M$ of the unit ball and the given ellipsoid is rotationally symmetric with respect to the $x$-axis and consists of all points $z=(x'+x'', y'+y'') \in {\mathbb R}^n$ with $${x'^2\over a^2}+|y'|^2\leq 1\ ,\qquad x''^2 +|y''|^2\leq 1\ .$$ A hyperplane $x={\rm const.}$ intersects $M$ in an $(n-1)$-dimensional ball whose radius can be found by maximizing $$|y'+y''|\leq |y'|+|y''|\leq \sqrt{1-x'^2/a^2} + \sqrt{1-x''^2}$$ under the constraint $x'+x''=x$.

$\endgroup$
3
$\begingroup$

For more on this, check out "Surface area and other measures of ellipsoids", by one I. Rivin:

[math/0403375] Surface area and other measures of ellipsoids by I Rivin ► Surface area and other measures of ellipsoids - Elsevier by I Rivin - 2007 Advances in Applied Mathematics 39 (2007)

$\endgroup$

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.