If I collapse a (say, closed) set in a length space, I obtain a length space: is there some literature on this?

We consider length spaces as defined by Gromov and others. [However the case of a Riemannian distance already leads to interesting examples of what I am writing]. If $X$ is one such space, it has a distance $d$ which can be recovered by the length of curves. Suppose $X$ is path connected. Let $c:[a,b]\to X$ be a curve, define $|c'(t)|=\lim_{\epsilon\to0}\sup_{|u-t|,|v-t|\le\epsilon}\frac{d(c(u),c(v))}{|u-v|}$ and $$ L(c)=\int_a^b|c'(t)|dt, $$ and set $$ D(x,y)=\inf\{L(c) \text{ $c$ is a curve having endpoints $x$ and $y$}\}. $$ We are interested in cases where $d=D$ and where the distance $D=d$ is realized by lengths of geodesics (i.e. the hard work has been already done).

Now, let $E\subseteq X$ be closed and define a distance $D_X$ on $X\setminus E\cup\{E\}$ ($X$ with $E$ collapsed to a point): $$ L_E(c)=\int_a^b|c'(t)|\chi_{X\setminus E}(c(t))dt, $$ and set $$ D_E(x,y)=\inf\{L_E(c) \text{ $c$ is a curve having endpoints $x$ and $y$}\}, $$ if $x,y\in X\setminus E$, and $D_E(x,E)=D(x,E)=\inf_{y\in E}D(x,y)$.

The space $X\setminus E\cup\{E\}$ is clearly a length space w.r.t. the distance $D_E$.

The reason I find these objects interesting is that, if $E$ is the smooth boundary of an open subset of $X$, a nice Riemannian manifold, then the points of $E$ play the role of the unit vectors in the tangent space of a point; this meaning that they can parametrize those geodesics leaving $E$ which, at least locally, minimize the distance from $E$. Even in the case of the Euclidean plane one obtains interesting pictures.

I would be very surprised if no one had developed this viewpoint in the past.

  • $\begingroup$ Post Scriptum: one may view the definition of $L_E$ and $D_E$ as a limiting case in optics, with the light having infinite speed inside $E$. $\endgroup$ Jun 11, 2019 at 11:57
  • 1
    $\begingroup$ you mean $\chi_{X\setminus E}$ instead of $\chi_E$? $\endgroup$ Jun 11, 2019 at 13:36
  • $\begingroup$ Yes! Thanks Emanuele... I edit. $\endgroup$ Jun 11, 2019 at 15:20
  • 1
    $\begingroup$ On a side note: for a continuous, non-constant curve of finite length $c:[a,b]\to X$ one may have $|c'(t)|=0$ a.e., thus $\int_a^b|c'(t)dt=0$ (e.g. for the Cantor function). In fact, I think this is always true up to reparametrizations, so you would have $D(x,y)=0$ for all $x,y$! However, the equality "integral of |c'(t)| = length" is true for Lipschitz curves, which is ok for many purposes since continuous curves, finite length curves are Lipschitz up to reparametrization. $\endgroup$ Jun 12, 2019 at 7:42
  • $\begingroup$ (more precisely: if $c:[0,1]\to X$ is any curve and $C:[0,1]\to[0,1]$ is the Cantor function and ${\bf C}\subset[0,1]$ is the Cantor set, then $|(c\circ C)'|=0$ a.e., because $c\circ C$ is locally constant in ${\bf C^c}$) $\endgroup$ Jun 12, 2019 at 8:14

1 Answer 1


If you take $X=\mathbb R$ and $E=[1,2]\cup[3,4]$ you see that $D_E(0,5) = 3$ while the geodesic distance between $0$ and $5$ in $X\setminus E \cup\{E\}$ is $2$ because you can join $0$ to $E$ with a curve of length $1$ and $E$ to $5$ with a curve of length $1$. I think you need to require $E$ to be geodesic-convex if you want the resulting space to be a length space.

Maybe what you really mean is $$ d_E(x,y) = \min\{d(x,y), d(x,E) + d(y,E)\}. $$ This is the particular case of a quotient space. It is understood that the quotient space of a length space is a length space if the quotient has good properties (in your case $E$ being closed). See, for example, https://people.math.ethz.ch/~lang/LengthSpaces.pdf Proposition 3.2

  • $\begingroup$ You are right Emanuele: with the definition I gave I am collapsing each connected component of $E$ to a distinct point (which is what I had in mind in this morning's mental picture), while you definition collapses the whole set to a point (which is my usual mental picture and which gives your definition). I have read the Proposition in Lang's notes. Both mental pictures correspond to his assumptions, with different equivalence relations. $\endgroup$ Jun 11, 2019 at 15:33

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.