Finding shortest path equation in product of metric spaces

72 Views Asked by At

Suppose I have two metric spaces $(X, d)$ and $(Y, d')$. Let $x_0, x_1 \in X$ and $y_0, y_1 \in Y$ arbitrarily. Suppose $s(t) : I \to X$ parameterizes the shortest possible path from $x_0$ to $x_1$ in $X$ and $j(t) : I \to Y$ likewise parameterizes the shortest possible path from $y_0$ to $y_1$ in $Y$.

We get a natural metric $D : (X \times Y)^2 \to \mathbb{R}_{\geq 0}$ defined by $$D : ((x, y), (x', y')) \mapsto \sqrt{ d(x, x')^2 + d'(y, y')^2 }$$ Intuitively, this makes me think we should also get a natural parameterization of the shortest possible path under $D$ from $(x_0, y_0)$ to $(x_1, y_1)$ in $X \times Y$ defined by $$k : t \mapsto (s(t), j(t))$$ ... but I don't really know how I would go about proving that this path is or is not optimal. What does a proof sketch look like for showing something like this? (I am happy to do the proof myself; I just don't know where to start and would like a point in the right direction.) Thanks!


EDIT: Here is my progress toward a proof:

I think am correct, and can prove it as follows.

A line from $u$ to $v$ in a metric space $(U, d)$ is a set $\{ a \in U \mid d(u, a) + d(a, v) = d(u, v) \}$.

Since we assume $s$ is optimal, it follows that $s$ follows a segment of a line from $x_0$ to $x_1$. Were this not the case, we could use the triangle inequality to find an at-least-as-good solution which does follow a segment of a line from $x_0$ to $x_1$. So, assume $s$ follows a segment of a line from $x_0$ to $x_1$. Apply this same logic to $j$ which goes from $y_0$ to $y_1$.

Let $k$ be as in the problem statement. Then $k$ follows a line-segment from $(x_0, y_0)$ to $(x_1, y_1)$:

$$ \iff D((x_0, y_0), k(t)) + D(k(t), (x_1, y_1)) = D((x_0, y_0), (x_1, y_1))$$

$$\iff D((x_0, y_0), (s(t), j(t))) + D((s(t), j(t)), (x_1, y_1)) = D((x_0, y_0), (x_1, y_1))$$

$$\iff \sqrt{ d(x_0, s(t))^2 + d'(y_0, j(t))^2 } + \sqrt{ d(s(t), x_1)^2 + d'(j(t), y_1)^2 } = \sqrt{ d(x_0, x_1)^2 + d'(y_0, y_1)^2 }$$

When I try to untangle this, the algebra is grotesque. Is this actually the proof structure? Just brute-forcing this despicable mess of algebra? Or is there a better way to do it?

1

There are 1 best solutions below

1
On BEST ANSWER

If you clean up the vagueness of the phrase "parameterizes the shortest possible path", then things become clearer.

The hypothesis you want is simply that the path $s(t) : [0,1] \to X$ is a reparameterized geodesic, meaning that $d(s(u),s(v))=|u-v| \cdot d(x_0,x_1)$ for all $u,v \in [0,1]$, and similarly for $j(t)$.

With that, it's then simple algebra to check that $k(t)$ is also a reparameterized geodesic.

By the way, n actual geodesic in this context would be a path of the form $s(t) : [0,d(x_0,x_1)] \to X$ such that $d(s(u),s(v)) = |u-v|$ for all $(u,v) \in [0,d(x_0,x_1)]$, and similarly for $j(t)$. But then the formula for $k$ would not make sense.