Let $O\subset\mathbb{R}^n$ be a open set which is star shaped with respect to the origin. How does one prove that there exists an increasing sequence of star shaped (w.r.t the origin) domains $O_i$ such that :

  • $O_i$ has a smooth boundary
  • $O_i\subset O$
  • $\lambda (O\backslash O_i)\to 0$ when $i\to\infty$, or, even better for my purpose, $O\backslash O_i$ is contained in an $\varepsilon_i$-neighborhood of $\partial O$.

For my purpose, it wouldn't hurt to assume that $O$ is bounded if that helps. This kind of approximation is handy for estimating the distributional laplacian of the distance function on a Riemannian manifold.

That seems like a pretty standard fact but I haven't been able to locate or derive a clean proof of this. I apologize if the question is not up to the standards of MO.

  • 1
    $\begingroup$ How about taking domains given by the set of points $p$ such that $dist(p,\partial O)>1/i$? (In case $O$ is bounded,but otherwise just look at the problem on $\mathbb S^n$). I don't know if those domains are star-shaped, but they look like it on the examples I've (quickly) considered. $\endgroup$ Mar 2, 2016 at 12:10
  • $\begingroup$ Oops, I forgot one crucial requirement: I want the $O_i$ to have smooth boundary. $\endgroup$ Mar 2, 2016 at 12:27
  • $\begingroup$ You could choose a radial diffeomorphism to euclidean space and take preimages of balls . $\endgroup$ Mar 2, 2016 at 21:06
  • $\begingroup$ I am not sure this works without any additional knowledge on the diffeomorphism. $\endgroup$ Mar 4, 2016 at 11:33
  • $\begingroup$ @MohanRamachandran Please be so kind and post the links. $\endgroup$ Mar 7, 2016 at 9:58

1 Answer 1


This is not really an answer, but an equivalent and hopefully easier reformulation of the question that does not fit in one comment.

You can describe the starshaped domain $O$ by a function $\rho\colon S^{n-1}\to(0,\infty]$ such that $$O\setminus\{0\}=\{\,x\in\mathbb R^n\setminus\{0\}\mid |x|<\rho(x/|x|)\,\}\;.$$ Then $O$ is open if and only if $\rho$ is lower smicontinuous. You want to approximate $\rho$ by smooth functions $\rho_i$ from below. This turns out to be impossible (both with respect to Lebesgue measure and with respect to distance) if $\rho$ jumps from $\infty$ down to a finite number.

If $\rho$ is bounded by $C$, the "$\varepsilon_i$" condition is equivalent to the following: all $\rho_i<\rho_{i+1}<\rho$ for all $i$, and for each $\varepsilon>0$, there exists an $i$ such that for each $p\in S^{n-1}$, there is $q\in S^{n-1}$ with $|q-p|<\varepsilon$ such that $\rho_i(p)>\rho(q)-\varepsilon$. The idea is that there exists $r\ge \rho(q)$ such that $|rq-\rho_i(p)\,p|<(1+C)\varepsilon$, and there is a boundary point of $O$ somewhere on the line segment from $\rho_i(p)\,p$ to $rq$. I am sure one can show that smooth approximations exist that satisfy this.

If $\rho$ is unbounded, one has to demand instead that for each $\varepsilon>0$, there exists an $i$ such that for each $p\in S^{n-1}$, there is $q\in S^{n-1}$ with $|q-p|<\varepsilon/\rho_i(p)$ such that $\rho_i(p)>\rho(q)-\varepsilon$. Just adapt the argument above.

  • $\begingroup$ Thanks. I had come to this reduction at some point, but failed to get a clean proof from there. $\endgroup$ Mar 2, 2016 at 13:26
  • 1
    $\begingroup$ At least I see how to get a sequence of Lipschitz functions from here. If someone knows exactly how good one can smooth those, then you are done. $\endgroup$ Mar 2, 2016 at 16:35

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.