Given a metric space $(X,d)$ we define the path metric $\rho(x,y)$ as the infimum of all paths lengths from $x$ to $y$. The length of a path $\gamma$ means the supremum of all $\sum_i d \big (\gamma(x_i),\gamma(x_{i+1}) \big )$ for all choices of points $0 \le x_1 < x_2 <\ldots < x_n \le 1$.

It is known $\rho$ induces a finer topology than $d$. Sometimes strictly finer. For example consider the harmonic fan: the union of all line segments from points $(1,1/n)$ and $(1,0)$ to $(0,0)$. Then $(X,d)$ is compact but $(X,\rho)$ is noncompact.

The harmonic fan is not locally arc-connected. I wonder what happens if we assume the space is arc-connected.

Suppose the metric space $X$ is arc-connected, locally compact and locally arc-connected. Is the metric on $X$ equivalent to the path metric?

I believe this is equivalent to the following (which I cannot prove either).

Suppose $X$ is above and $x \in X$. Let $U_1 \supset U_2 \supset \ldots $ be a chain of open neighborhoods of $x$ with diameter decreasing to zero. Then the path-diameter of $U_i$ also decreases to zero.

Any ideas?

If it helps we can also assume each point has a basis of closed arc-connected neighborhoods.

  • $\begingroup$ You definition of the path length is not correct. It is the supremum of such sums where $x_i = \gamma(t_i)$ and $t_1,...,t_n$ is a partition of the interval on which $\gamma$ is defined. See for example encyclopediaofmath.org/index.php/Rectifiable_curve. $\endgroup$ – Paul Frost Aug 2 '18 at 11:49
  • $\begingroup$ How do you treat the case that there exist paths with infinite length? $\endgroup$ – Paul Frost Aug 2 '18 at 11:57
  • $\begingroup$ Correct. There was an error in the definition of path length. $\endgroup$ – Daron Aug 2 '18 at 12:17
  • $\begingroup$ We're assuming local arc-connectedness. Arcs in $X$ have finite length. For let $\Gamma \subset X$ be an arc. By compactness the homeomorphism $\Gamma \to [0,1]$ is uniformly continuous and the modulus lets us put a bound on the length of $\Gamma$ in $X$. $\endgroup$ – Daron Aug 2 '18 at 12:21
  • $\begingroup$ Error not yet corrected. If you have a path $\gamma : [0,1] \to X$ and allow all choices $x_i \in \gamma([0,1])$, the supremum will always be infinite. You have to take the supremum over all sequences $x_1,..., x_n$ having the form $x_i = \gamma(t_i)$ with $0 = t_1 < t_2 < ... < t_n = 1$. $\endgroup$ – Paul Frost Aug 2 '18 at 12:34

Here are two old results. The first:

Theorem (Bing: A convex metric for a locally-connected continuum, 1928)

Each locally connected metric continuum admits an equivalent convex metrix.

The second:

Theorem (Goebel & Kirk: Topics in Metric Fixed Point Theory p 24)

Each complete and convex metric space has the path metric equivalent to the original metric.

Now suppose $(X,d)$ is as described in your question. The path metric topology is always finer than the original topology. We need to show it is coarser too. So suppose $x_n \to x$ with respect to $d$ and let $K$ be a locally-connected continuum neighborhood of $x$.

Some tail $y_m$ of $x_n$ is contained in the metric subspace $(K,d_K)$. The above theorems say $d_K$ is equivalent to the $d_K$ path metric $\rho_K$. Let $\epsilon>0$ be arbitrary. Some tail $y_N,y_{N+1}, \ldots$ has all $\rho(y_i,x)<\epsilon$. That means there are paths $\gamma_i \subset K$ from $y_i$ to $x$ of $d_K$ length less than $\epsilon$.

But since $d_K$ is the restriction of $K$ these paths $\gamma_i$ are also paths of $d$-length length less than $\epsilon$. Since $\epsilon$ is arbitrary we get $\rho(y_m,x) \to 0$ as required.


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.