Let $M$ be a differential manifold and let $f:M\to \mathbb{R}$ be a $C^{\infty}$ function such that there are exactly two points $x$ which satisfy $d_xf=0$. Let $p$ and $q$ those two points, and suppose that $f(p)=f(q)$. Prove that there is another $C^{\infty}$ function $g:M\to \mathbb{R}$ such that $p$ and $q$ are the only critical points but $g(p)\neq g(q)$.

I have been trying to solve this problem but I cannot come up with a solution. Clearly we can suppose $f(p)\neq 0$, so I intended to separate (by the $T2$ property) $p$ and $q$ with two open subsets $U$ and $V$, and letting $W$ an open set such that $p\in W\subseteq \overline{W}\subset U$ and $\overline{W}$ compact. Then I intended to make use either of a bump function or partition of unity, but I could not succeed.

How would you solve the problem?


You're on the right track. Take $q\in W\subset\overline W\subset U$, and create a smooth function $\phi$ on $M$ that is $1$ on $\overline W$ of $q$ and $0$ outside $U$. We're going to consider the function $g=f+\phi$. Obviously, $p$ and $q$ are critical points of $g$ and $g(p)\ne g(q)$. But we may introduce other critical points, which necessarily must lie in $\overline U-W$.

So here's the trick. Let $C = \inf\limits_{x\in \overline U-W} \|df_x\|$ and let $C' = \sup\limits_{x\in\overline U-W} \|d\phi_x\|$. Now replace $\phi$ with $c\phi$, where $0<c<C/C'$. (You can compute these norms by working in a single chart centered at $q$. Just choose $U$ inside that chart.)

  • $\begingroup$ But if the function is $0$ at a neighborhood of $p$, it would have an infinite number of critical points, and I want only two. $\endgroup$ – solomeo paredes Jul 19 '18 at 15:21
  • $\begingroup$ Ah, I intended you to add this function to the original function $f$. But we need to take some care to make sure that we don't introduce another critical point far away from $p$ and $q$. I'll edit slightly. $\endgroup$ – Ted Shifrin Jul 19 '18 at 15:26
  • $\begingroup$ @solomeoparedes: Do you have any further questions? If not, please accept the answer so the question no longer appears as "unanswered." $\endgroup$ – Ted Shifrin Jul 21 '18 at 16:57

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.