Let $X\subseteq \mathbb{R}^n$, $f: X \rightarrow \mathbb{R}$: what further conditions on $X$ and $f$ do we need to ensure that $\text{Int}\{x|f(x)\leq 0\}=\{x|f(x)<0\}$?

Context: I've encountered this kind of question when dealing with open bounded $X$ with $C^1$ boundary, in partial differential equations, but I'm sure that one could benefit from such a fact in optimization, for instance.

Clearly, continuity is not enough, since $f:[0,1]\rightarrow \mathbb{R} , f(x)=0$ is such that $[0,1]=\{f\leq 0\}$, but the interior of this set is not the empty set.


If $f$ is continuous, it is easy to see that $$\tag {1}\{x:f(x)<0\}\subseteq \text{Int}\{x:f(x)\le 0\}$$

This holds because the LHT is open and the RHT is, by definition, the biggest open subset of $\{x:f(x)\le 0\}$.

Let us first suppose $X=\mathbb{R}^n$:

One necessary and sufficient condition for equality, provided that $f$ is differentiable, is that $\nabla f\neq 0$ on $f^{-1}(0)$.

To prove it, let $x$ be a point inside $\text{Int}\{x:f(x)\le 0\}\backslash \{x:f(x)<0\}$. This means that $x$ is locally a maximum, and thus $\nabla f=0$

This clearly generalizes to open subsets of $\mathbb{R}^n$. It is now easy to see that, for general $X$ and continuous $f$, a necessary and sufficient condition is that $f^{-1}(0)$ does not contain local maxima (although this is not a very useful condition)

Note that some kind of regularity hypotesis is necessary: as Whitney proved, given a closed subset $K$ of $\mathbb{R}^n$, there's always a $\mathcal{C}^\infty(\mathbb{R}^n)$ function with $f^{-1}(0)=K$. Given such an $f$, considering $f^2$ gives you a class of smooth examples with strict (or trivial) inclusion.

  • $\begingroup$ Thank you. Why is there such a neighborhood? $\endgroup$
    – warm_fish
    Nov 17 '20 at 13:14
  • $\begingroup$ @warm_fish My argument was faulty, I modified it. Indeed, there does not need to be such a neighbourhood as the one I was describing, as $f(x,y)=-y^2$ shows $\endgroup$
    – Caffeine
    Nov 17 '20 at 13:24
  • 1
    $\begingroup$ You only need $\nabla f$ to be nonzero at the points where $f(x) = 0$. $\endgroup$ Nov 17 '20 at 13:28
  • 1
    $\begingroup$ From a slightly different perspective, this is also essentially the condition that $f(x)=0$ defines a submanifold. $\endgroup$
    – Toffomat
    Nov 18 '20 at 13:01

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.