2
$\begingroup$

Let $0<a<d/2$, let $B$ be the unit ball in $\mathbb{R}^{d}$ centered at the origin, and let $f:B \to [0,\infty[$ be a a smooth function such that

(1) $f(x)\geq f(0).$

(2) $\nabla f(x)\neq 0,\quad \forall x\neq 0.$

(3) The Hessian matrix $D^2 f(0)$ is positive definite.

I am trying to prove $f^{-a}$ is integrable on $B$. I am obviously interested in the case $f(0)=0$.

Since $0$ is a minimum of $f$ we necessarily have $\nabla f(0)=0$. So there exists $\xi \in B$ such that $$f(x)=f(0)+x^T D^2 f(0) x+ \sum_{|\alpha|=3}\frac{x^{\alpha}}{\alpha !} D^{\alpha}f(\xi),\quad x \in B.$$

Since $D^2f(0)$ is positive definite, there exists $C>0$ such that $x^T D^2f(0)x\geq C|x|^2$. And, for $x \in B^{\prime}$, a small enough ball centered at $0$, we have $$|\sum_{|\alpha|=3}\frac{x^{\alpha}}{\alpha !} D^{\alpha}f(\xi)|<\frac{1}{2} C|x|^2. \qquad (1)$$ So, for $x \in B^{\prime}$, we have $$f(x)\geq \frac{1}{2} C|x|^2.$$

So we have the integral

$$\int_{B^{\prime}}f^{-a} \lesssim \int_{B^{\prime}}|x|^{-2 a}dx.$$ We are done with this part by the assumption $2a<d$.

What about the easier integral $$\int_{B\setminus B^{\prime}}f^{-a}$$

Edit: I think the answer to my question is simple. This is also inspired by the helpful comments of @Fedor Petrov and @Bazin

Since $D^{2}f(0)$ is positive definite then $f$ is strictly convex in some neighborhood of the origin, say $B(\delta_{0})\subseteq B^{\prime}$. We claim that if $|x|>\delta_{0}$ then $f(x)>\frac{1}{4} C\delta_{0}^2$. (Recall that we have shown $f(y)\geq \frac{1}{2}C|y|^2$ for all $y\in B^{\prime}$). Otherwise $f(x)\leq f(y)$ for some $y \in B(\delta_{0})$. This can not occur since $f$ has no critical points outside $B(\delta_{0})$, by the condition $(2)$.

$\endgroup$
3
  • $\begingroup$ This is brilliant. But the inequality $f(x)>\delta^{-1} f(\delta x)$ applies to $x$ that lies in convex set that contains $0$ on which $f$ is convex. The question remains : Is the claim ''$f$ is convex '' true and what is the best way to argue that? $\endgroup$
    – Medo
    Commented Apr 27, 2023 at 8:10
  • $\begingroup$ Would you like to elaborate a bit? Condition (3) says the Hessian is positive definite at the origin, so by smoothness, the Hessian is positive definite in some neighborhood of the origin. On that neighborhood the function is strictly convex. What do we infer outside that neighborhood given conditions (1)-(3) ? $\endgroup$
    – Medo
    Commented Apr 27, 2023 at 20:35
  • $\begingroup$ Ah, only at 0. Sorry, I misread $\endgroup$ Commented Apr 27, 2023 at 21:02

1 Answer 1

1
$\begingroup$

By the Morse Lemma, using your assumption (3), you can find a smooth change of variables $x\mapsto y$ such that $$ f(x)=\frac12 f''(0) y^2 $$ in a neighborhood of 0. This entails the integrability condition.

$\endgroup$
2
  • $\begingroup$ I edited the question. Do you find my answer correct ? Thanks $\endgroup$
    – Medo
    Commented Apr 27, 2023 at 7:24
  • $\begingroup$ It seems correct, but the Morse lemma is truly the adequate answer (in my opinion), since you do have a normal form for your function near $0$; also the Morse lemma is not difficult to prove and its proof is shorter than you specific argument. $\endgroup$
    – Bazin
    Commented Apr 27, 2023 at 8:26

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .