Let $L_t(s):S^1 \rightarrow \mathbb{R}^2(t\in [0,1])$ be a Jordan curve, $O(t)$ be its interior and $H(t,s)=L_t(s)$. If $H$ is a homotopy from $L_0(s)$ to$L_1(s)$, does there exist a continuous function $f:[0,1]\rightarrow \mathbb{R}^2$ s.t. $\forall t \in[0,1],f(t)\in O(t)$?
2 Answers
By compactness one can replace the Jordan curves by a family of polygonal ones. The cut locus $C(t)$ of the Jordan curve $L_t$ is a rectifiable tree and one can choose the midpoint $m(t)$ of $C(t)$ is the value of $f$ at $t$, giving a continuous function with image in the interior of the Jordan curve for every $t$.
-
$\begingroup$ I believe the "tree" part but not the "rectifiable" part (imagine a narrow tentacle around $\sqrt{x}\sin(1/x)$) "Locally rectifiable" it is, but that's not enough to define the midpoint unless I misunderstand what the midpoint is. $\endgroup$– fedjaCommented Aug 11, 2016 at 0:05
-
$\begingroup$ @fedja I did not say the Jordan curve is rectifiable. rather the cut locus is rectifiable as proved in the literature around Marcel Berger. At any rate this is utterly irrelevant because for the purposes of this problem everything can be replaced by polygonal curves by compactness. $\endgroup$ Commented Aug 11, 2016 at 7:48
-
$\begingroup$ The cut locus of such a tentacle will go along the curve all the way to the origin, so it will have infinite length. What am I missing? As to the polygons, you'll have to show that the midpoint of the cut locus of the approximation is not just inside but deep inside the region, which is, probably, true, but not quite obvious. $\endgroup$– fedjaCommented Aug 11, 2016 at 12:41
-
$\begingroup$ @fedja, when you take a tubular neighborhood of a curve and then look for the cut locus of the boundary of the tubular epsilon-neighborhood, you won't necessarily get the original curve back except special cases when you have control on curvature and epsilon sufficiently small. $\endgroup$ Commented Aug 11, 2016 at 12:45
-
$\begingroup$ the "deep inside" is just a compactness argument. $\endgroup$ Commented Aug 11, 2016 at 12:46
Here's a simple construction of the function $f$. It makes use of the Hausdorff metric on compact subsets of $\mathbb{R}^2$, defined by $$d_H(A,B) = \inf\bigl\{r \in [0,\infty) \,\bigm|\, A \subset N(B,r), \, B \subset N(A,r)\bigr\} $$ The image $L_t(S^1)$ is compact for each $t$. Continuity of the homotopy $H(t,s)$ implies continuity of the function $t \mapsto L_t(S^1)$ in the Hausdorff metric space on compact sets.
For each $t \in [0,1]$ make the following choices:
- pick $x_t \in O(t)$,
- then pick $r_t > 0$ so that $d(x_t,L_t(S^1)) > r_t$,
- then pick an open subset $I_t \subset [0,1]$ containing $t$ such that for all $s \in I_t$ we have $d_H(L_t(S^1),L_s(S^1)) < r_t$.
It follows from these choices that for each $t$ and each $s \in I_t$, the open ball $B(x_t,r_t)$ is a subset of $O(s)$. The proof is that $L_t(S^1)$ has winding number $1$ around $x_t$; but winding number of $L_s(S^1)$ around $x_t$ is continuous as a function $s \in I_t$, and so also equals $1$ for all $s \in I_t$.
The sets $I_t$ form an open covering of $[0,1]$. Choose a Lebesgue number $\lambda$ for this open covering. Let $$0 = t(0) < t(1) < \cdots < t(K) =1 $$ be a sequence of intervals such that $t(k) - t(k-1) < \lambda$ for $k=1,...,K$.
From the choices, it follows for each $k=1,...,K$ and each $s \in [t(k-1),t(k)]$ that the straight line segment $\overline{x_{t(k-1)},x_{t(k)}}$ is disjoint from $L_s(S^1)$. Thus by concatenating these segments from $k=1$ to $K$ we obtain a piecewise straight path $f(t)$ such that $f(t) \in O(t)$ for each $t \in [0,1]$.