Hallo, I have the following question: Let $(X,d)$ be a complete metric space. Is then $(X,\operatorname{dist})$ also complete? Here by $\operatorname{dist}$ I mean the metric induced by $d$ by: $\operatorname{dist}(x,y)=\inf(L(\gamma))$, where the infimum is taken over all paths joining $x,y$ and $L(\gamma)=\sup(\sum_{i=1}^{n-1}d(\gamma(t_{i}),\gamma(t_{i+1})))$, where the supremum is taken over all partitions $P=(0=t_{1}< ... < t_{n}=1)$ of $[0,1]$, where $\gamma$ is defined. Intuitively it should be but I dont know how to prove it. If its true, how can one prove it?

cheers denis

  • $\begingroup$ en.wikipedia.org/wiki/Intrinsic_metric $\endgroup$ – Matthias Ludewig May 28 '13 at 6:48
  • $\begingroup$ sorry I do not see where this is answered. where exactly? $\endgroup$ – denis May 28 '13 at 6:51
  • $\begingroup$ can you point it out, please. $\endgroup$ – denis May 28 '13 at 6:55
  • $\begingroup$ is the statement in the question above true or not? If yes, how could one prove it ? $\endgroup$ – denis May 28 '13 at 7:10
  • $\begingroup$ What is left unsaid in the answer and the wikipedia link is that the original metric and the "intrinsic metric" can be pathologically different. For example, if $X$ is the Cantor set or any other totally disconnected metric space then dist completely degenerates to the unique $0,\infty$ extended metric. $\endgroup$ – Lee Mosher May 28 '13 at 13:57

The intrinsic metric induced by a complete metric space is always a complete extended metric. To prove this, we shall write $d^{\sharp}$ for the extended metric $dist(x,y)=\inf(L(\gamma))$.

Also, I want to remark that $d^{\sharp}$ is in general not a metric since we may have $d^{\sharp}(x,y)=\infty$ even when $X$ is path-connected. For instance, if we take some fractal $X$ such the boundary of the Koch snowflake, then $dist(x,y)=\infty$ for every pair of points $x,y\in X$.

However, $d^{\sharp}$ is always an extended metric (By an extended metric I mean that $d^{\sharp}$ is a metric in every way except for the fact that we may have $d(x,y)=\infty$). Notions such as Cauchy sequences and completeness still hold if we use extended metrics instead of metrics. Furthermore, if we let $d'(x,y)=Min(d^{\sharp}(x,y),1)$, then $d'$ is a metric with the same uniform structure as $d^{\sharp}$(and hence the same Cauchy sequences, topology,...).

We shall now prove that $d^{\sharp}$ is complete using standard arguments. To prove that $d^{\sharp}$ is complete assume that $(x_{n})_n$ is a Cauchy sequence with respect to the extended metric $d^{\sharp}$. Then since $d(x,y)\leq d^{\sharp}(x,y)$ for each $x,y\in X$, the sequence $(x_{n})_{n}$ is Cauchy with respect to the metric $d$. Since $(X,d)$ is a complete metric space the sequence $(x_{n})_{n}$ converges to some point $x\in X$ with respect the metric $d$. Since $(x_{n})_{n}$ is Cauchy with respect to $d^{\sharp}$, we may take a subsequence $(y_{n})_{n}$ of $(x_{n})_{n}$ such that $d^{\sharp}(y_{n},y_{n+1})<\frac{1}{2^{n}}$ for all $n$. In this case, there is some path $\gamma_{n}:[0,1]\rightarrow X$ such that $\gamma_{n}(0)=y_{n},\gamma_{n+1}(1)=y_{n+1}$ and $L(\gamma_{n})<\frac{1}{2^{n}}$. If we let $\gamma:[0,\infty]\rightarrow X$ be the function where $\gamma(r+n)=\gamma_{n}(r)$ for $n\geq 0$ and $r\in[0,1]$ and $\gamma(\infty)=x$, then $\gamma$ is a continuous function. However, the restricted function $\gamma|_{[k,\infty]}$ is a path from the point $y_{k}$ to $x$ with $L(\gamma|_{[k,\infty]})<\sum_{n=k}^{\infty}\frac{1}{2^{k}}=\frac{1}{2^{k-1}}$. Therefore $d^{\sharp}(y_{k},x)<\frac{1}{2^{k-1}}$, so the sequence $(y_{n})_{n}$ also converges to $x$ with respect to the generalized metric $d^{\sharp}$.

  • $\begingroup$ You say $\mathrm{dist}(x,y) = \infty$ for every pair of distinct points on the Koch snowflake. Surely this can't be true? Suppose $x$ was introduced by the $n^\mathrm{th}$ iteration of the construction, and $y$ by the $m^\mathrm{th}$, with $m < n$. Then both are in the $n^\mathrm{th}$ iteration, which is not fractal at all, so there should be a reasonable path between them. Adding more points can only provide more (shorter) paths (although I suspect the shortest path is actually already in stage $n$). $\endgroup$ – Ben Millwood Jun 25 '13 at 15:52
  • $\begingroup$ Ben. I don't think I understand your objection. I am talking about the limit of the Koch snowflake. $\endgroup$ – Joseph Van Name Jun 25 '13 at 18:25
  • $\begingroup$ Of course. But the Koch snowflake is only badly-behaved at its perimeter: the interior surely has perfectly ordinary paths with perfectly ordinary lengths in it. (Just to make sure we are talking about the same thing, here is a picture commons.wikimedia.org/wiki/… ) $\endgroup$ – Ben Millwood Jun 27 '13 at 13:39
  • $\begingroup$ I was talking about the boundary of the Koch snowflake. I edited the answer, so hopefully it is now more clear. $\endgroup$ – Joseph Van Name Jun 27 '13 at 22:14
  • $\begingroup$ Aha, okay. I find that much more believable. $\endgroup$ – Ben Millwood Jun 28 '13 at 16:08

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.