Could someone give me an example of a local diffeomorphism from $\mathbb{R}^p$ to $\mathbb{R}^p$ (function of class say $C^k$ with an invertible differential map in each point) that is not a diffeomorphism..

in the real line (1 dim case) that would mean a function with a continuous non null derivative on an open $V$ of $\mathbb{R}$ that is not bijective which does not make sense thus any local diffeomorphism on the real line is a diffeo..

Could one give me a counterexample in a higher dimension?

  • 4
    $\begingroup$ Take the complex exponential $e^z$ as a function from $R^2$ to $R^2$. For every horizontal strip [iy+ i(y+2Pi) ) (i.e., including iy, but not i(y+2Pi)) there is an inverse--a logz --, but there is no global inverse. $\endgroup$ – gary May 15 '11 at 1:37
  • 1
    $\begingroup$ there is no inverse around zero $\endgroup$ – yoyo May 15 '11 at 2:27
  • 1
    $\begingroup$ Correction: that should be the strip [iy, i(y+2Pi)) $\endgroup$ – gary May 15 '11 at 2:38
  • 1
    $\begingroup$ @yoyo: I assume you mean the target zero , right? Otherwise, the inverse function guarantees the existence of a local diffeomorphism at each point, including 0 in the domain, since d/dz($e^z$)=1 at z=0; same for all other points' or using the $R^2$ version of the inverse function theorem. $\endgroup$ – gary May 15 '11 at 22:36
  • 3
    $\begingroup$ You are wrong about the one dimensional case: the exponential function $\mathbb R\to \mathbb R: x\mapsto e^x$ is a non surjective local diffeomorphism. $\endgroup$ – Georges Elencwajg Sep 1 '14 at 20:09

Consider $f:\mathbb{R}^2\rightarrow \mathbb{R}^2$ defined by $f(x,y)=(e^x \cos y,e^x \sin y)$

$Df(x,y)$ is always invertible because $\det Df(x,y)=e^{2x}$ but clearly $f$ is not one to one. It is periodic with period $2\pi$.

In higher dimensions, one can use $f(x)=(e^{x_1}\cos x_2, e^{x_1}\sin x_2, x_3, \dots, x_n)$.

  • $\begingroup$ To draw the link to complex analysis, treating x and y as the usual coordinates on the complex plane, f = e^z, z = x + iy. The inverse, ln(f(z)) is multi-valued, and hence a branch cut must be taken, e.g. 0 < arg(f(z)) = y < 2pi. $\endgroup$ – Rascalniikov May 13 '17 at 6:28

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.