I understand the general idea of a homotopy, but I'm a little lost on how to create them myself. For example if I wanted to show

$$ \text{Let} \: \alpha, \beta, \text{and} \: \gamma \: \text{be paths} \: I \to X, \: \text{from} \: x_{0} \: \text{to} \: y_{0}, y_{0} \: \text{to} \: z_{0}, \: \text{and} \: z_{0} \: \text{to} \: u_{0}. \: \text{Then} \: \\ (\alpha \cdot \beta) \cdot \gamma \sim \alpha \cdot (\beta \cdot \gamma) $$

A possible homotopy is $F: I \times I \to X$, given by $$\\ F(t,s) = \begin{cases} \alpha(\frac{4t}{1+s}) & 0 \leq t \leq \frac{s+1}{4} \\ \beta(4t-1-s) & \frac{s+1}{4} \leq t \leq \frac{s+2}{4} \\ \gamma(\frac{4t - 2 - s}{2-s}) & \frac{s+2}{4} \leq t \leq 1 \\ \end{cases}$$

What I don't understand is where this comes from. What is the intuition here and how can I form explicit homotopies like this?


See the picture below.

enter image description here

The idea is that, for any choice of $s$, the loop $t \mapsto F(t, s)$ consists of walking along $\alpha$, $\beta$ and $\gamma$, in that order. The difference between the various choices of $s$ is in the "time schedule": each choice of $s$ allocates different amounts of time to $\alpha$, $\beta$ and $\gamma$.

  • If $s = 0$, you walk path $\alpha$ in time $[0, \tfrac 1 4]$, then walk $\beta$ in time $[\tfrac 1 4, \tfrac 1 2 ]$, then walk $\gamma$ in time $[\tfrac 1 2 , 1]$.

  • If $s = 1$, you walk path $\alpha$ in time $[0, \tfrac 1 2]$, then walk $\beta$ in time $[\tfrac 1 2, \tfrac 3 4]$, then walk $\gamma$ in time $[\tfrac 3 4 , 1]$.

For intermediate choices of $s$, the schedule is given by interpolation.

So for example, if $s = \tfrac 1 2$, you walk $\alpha$ in time $[0, \tfrac 3 8]$, then walk $\beta$ in time $[\tfrac 3 8, \tfrac 5 8]$, then walk $\gamma$ in time $[\tfrac 5 8, 1]$. And so on.


This is basically a reparametrization of the curve, but the parametrization is changing continuously. We may come up with the homotopy in two steps.

Let $u:I\to X$ be a curve, let $\phi:I\to I$ be a continuous map with $\phi(0)=0$ and $\phi(1)=1$, then $u$ is homotopic to $u\circ\phi$.

Proof: define $H(s, t)=u((1-s)t+s\phi(t))$, then $H(0,t)=u(t)$, $H(1,t)=u(\phi(t))$ and $H(s, 0)=u(0)$, $H(s, 1)=u(1)$. Essentially this is just mapping a homotopy between $\operatorname{Id}$ and $\phi$ in $I$ to the homotopy in $X$ by $u$.

We could construct $\phi:I\to I$ by rescaling the speed according to the time of travel through each curve. More explicitly, we have $[\alpha\cdot(\beta\cdot\gamma)] (t) = [(\alpha\cdot\beta) \cdot\gamma] (\phi(t)) $ if we define: $$\phi(t) =\begin{cases}\frac 12 t, & t\in[0,\frac 12]\\ t-\frac14, & t\in [\frac12, \frac34] \\ 2t-1, & t\in [\frac34, 1]\end{cases} $$

However if you apply the formula above, while we still get a homotopy, but the formula is not as clean (as in we need to divide into 5 cases). To obtain the formula you stated, let's look at the graphs of the intermediate reparametrizations, for each fixed $s_0$, the graph of $\phi_{s_0}(t) =1-s_0)t + s_0 \phi(t) $ would look like a "sheared up" version of $\phi(t)$. So image of the points at which $\phi_{s_0}$ are piecewisely defined do not match with the points at which $(\alpha\cdot\beta) \cdot\gamma$ is piecewisely defined. To make them match we need to "shear" $\phi(t)$ to the left.

So we should take $\tilde\phi_s(t)$ the inverse function of $(1-s)t+s\phi^{-1}(t)$, i.e. $\tilde\phi_s(t)$ is the inverse of $$\begin{cases}(1+s)t, & t\in[0,\frac14] \\ t+\frac s4, & t\in[\frac 14,\frac 12]\\ \frac 12(2-s)t +\frac s2, & t\in[\frac 12,1]\end{cases}$$

Which is given by: $$\tilde\phi_s(t) =\begin{cases} \frac t{1+s}, & t\in[0,\frac {s+1}4]\\ t-\frac s4, & t\in[\frac{s+1}4,\frac{s+2}4]\\ \frac 2{2-s}t - \frac s{2-s}, & t\in[\frac {s+2}4,1]\end{cases} $$

The homotopy you provided in the post is then $H(s, t) =[(\alpha\cdot \beta)\cdot \gamma](\tilde\phi_s(t)) $.


I have checked the definition and all axioms of the first homotopy group in this note and in this I also explain the exact homotopy you describe. I hope it helps you...


There is an easier way to do this by using the notion of a "Moore path" as a pair $(f,r)$ where $r \in \mathbb R, r \geqslant 0$ and $f:[0, \infty) \to X$ is constant $[r, \infty)$. This definition is used in the quite old book on Knot Theory by Crowell and Fox. Alternatively, as in Topology and Groupoids, one considers a path of length $r \geqslant 0$ to be a map $f:[0, r] \to X$. The composite of a path of length $r$ with a path of length $s$ is then, when defined, of length $r+s$. This corresponds intuitively to the idea of path as a "journey". Composition is then associative. We also have paths of length $0$ and in both definitions the composition of paths in $X$ gives a category $PX$.

We also write $s$ for a constant path of length $s$ at $y \in X$ and say two paths $f,g$ from $x$ to $y$ in $X$ are equivalent if there are real numbers $s,t \geqslant 0$ such that $f+s, g+t$ are homotopic rel end points. This gives the fundamental groupoid $\pi_1(X)$.

One easily proves that any path is equivalent to a path of length $1$. This is called normalisation, which is not a process one uses for journeys.

The formulae needed for all this work out much easier than the usual ones, which seems to me a good thing. (I've said all this somewhere else on this site.)


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.