Suppose the topological space $X$ is the union of two open pathwise connected subspaces $U_1$ and $U_2$ whose intersection is also pathwise connected. Let $x\in U_1\cap U_2$ and let $f_k:\pi_1(U_k,x)\to G$ ($k=1,2$) be a homomorphism, where $G$ is a group. Assume $f_{U_1}\circ (i_{U_1})_\ast=f_{U_2}\circ (i_{U_2})_\ast$, where the $i_{U_k}$ are the inclusions. Consider a loop $\gamma:I\to X$ based at $x$; assume $I$ is divided into subintervals $I_j=[x_j,x_{j+1}]$ ($j\in \Lambda$) so that the image of $\gamma\restriction_{I_j}$ lies entirely within $U_1$ or $U_2$. Let $\beta_i: I\to U_1\cap U_2$ be a path with $\beta_i(0)=x,\beta_i(1)=\gamma(x_i)$. Then the concatenation $\Gamma_i=\beta_{i-1} \cdot \gamma\restriction_{I_{i+1}}\cdot \beta_i^{-1}$ is a path in $U_k$ for some $k\in\{1,2\}$. And we can consider the product of $f_{U_i}([\Gamma_i])$ over all $i\in \Lambda$. (If $\Gamma_i$ lies in $U_i$, then the image under $f_{U_i}$ (of $[\Gamma_i]$) is taken.)

I'm trying to understand why the value of the product is independent of the choice of the partition of $I$ into the $I_j$ as well as of the choice of $\beta_i$. I have no idea about the former, and the latter doesn't even look plausible for me. If $\delta_i$ is a path which enjoys the properties of $\beta_i$, then I guess I have to show that $\delta_i$ is fixed-end-point homotopic to $\beta_i$ (or what else can I show?), but the intersection needn't be simply connected.. So the independence of both choices looks quite mysterious for me.

  • $\begingroup$ The point about the $\beta_i$ is that in $\Gamma_i\cdot \Gamma_{i+1}$, they cancel. Actually, that same argument (with a "common refinement") is also what shows independence of the partition $\endgroup$ – Hagen von Eitzen Nov 26 '18 at 4:51
  • $\begingroup$ @HagenvonEitzen But to collect $\Gamma_i$ and $\Gamma_{i+1}$ in one place, one has to apply the homomorphism property (I believe); however it may be the case that the product is of the form $f_{U_1}(..)f_{U_2}(..)f_{U_1}(..)...$, so the homomorphism property cannot be applied. (Also, the group isn't commutative.) And even if it did, I don't quite see how this shows independence. $\endgroup$ – user531587 Nov 26 '18 at 5:05

We need some more precision.

1) Let $I$ be divided into $n$ subintervals $I_j = [x_{j-1},x_j]$, $j = 1,\dots,n$, such that $\gamma(I_j)$ is contained in $U_1$ or $U_2$.

2) For $j = 1,\dots,n-1$ let $\beta_j : I \to U_1 \cap U_2$ be a path such that $\beta_j(0) = x, \beta_j(1) = \gamma(x_j)$. For $j = 0, n$ let $\beta_j$ be the constant path $\beta_j(t) \equiv x$.

3) For each $j = 1\dots,n$ choose $k(j) \in \{ 1,2 \}$ such that $\gamma(I_j) \subset U_{k(j)}$. Note that $\gamma(I_j) \subset U_1 \cap U_2$ is possible. For $j = 1,\dots,n$ define a closed path in $U_{k(j)}$ by $$\Gamma_j = \Gamma_j(\beta_{j-1},\beta_j) = \beta_{j-1} \cdot \gamma \mid_{I_j} \cdot \beta^{-1}_j .$$ Then define $$f(\gamma) = f_{k(1)}([\Gamma_1]) \dots f_{k(n)}([\Gamma_n]) \in G.$$

The claim is that $f(\gamma)$ does not depend on the above choices.

a) Independence of the choice of $\beta_j$.

Recall that a choice was made only for $j = 1,\dots,n-1$. Thus it suffices to show that for $j = 1,\dots,n-1$ $$f_{k(j)}([\Gamma_j(\beta_{j-1},\beta_j)]) f_{k(j+1)}([\Gamma_{j+1}(\beta_j,\beta_{j+1})]) = f_{k(j)}([\Gamma_j(\beta_{j-1},\beta'_j)]) f_{k(j+1)}([\Gamma_{j+1}(\beta'_j,\beta_{j+1})]).$$ The path $\delta_j = \beta'_j \cdot \beta^{-1}_j$ is a closed path in $U_1 \cap U_2$ which begins and ends at $x$. Let $i_k : U_1 \cap U_2 \to U_k$ denote inclusion. We write $$\phi = f_1 \circ (i_1)_* = f_2 \circ (i_2)_* .$$ We have $$[\Gamma_j(\beta_{j-1},\beta'_j)] \cdot [i_{k(j)}\delta_j] = [\Gamma_j(\beta_{j-1},\beta_j)]$$ and therefore $$f_{k(j)}([\Gamma_j(\beta_{j-1},\beta_j)]) = f_{k(j)}([\Gamma_j(\beta_{j-1},\beta'_j)]) f_{k(j)}([i_{k(j)}\delta_j]) = f_{k(j)}([\Gamma_j(\beta_{j-1},\beta'_j)]) f_{k(j)}(i_{k(j)})_*([\delta_j])$$ $$= f_{k(j)}([\Gamma_j(\beta_{j-1},\beta'_j)]) \phi([\delta_j]) .$$ Similarly we see that $$ f_{k(j+1)}([\Gamma_{j+1}(\beta_j,\beta_{j+1})]) = \phi([\delta_j])^{-1} f_{k(j+1)}([\Gamma_{j+1}(\beta'_j,\beta_{j+1})])$$ which completes the proof of a).

b) Independence of the choice of $k(j)$.

A choice is only possible when $\gamma(I_j) \subset U_1 \cap U_2$. In that case we use $f_1 \circ (i_1)_* = f_2 \circ (i_2)_*$.

c) Independence of the choice of the partition of $I$.

Let us first consider a refinement of $\{ I_j \}$ by inserting one additional partition point between $x_j$ and $x_{j+1}$ for some $j$. By considerations similar as in a) we see that this does not change $f(\gamma)$. Procceeding inductively we see that the same is true for any refinement of $\{ I_j \}$. Now any two partitions have a common refinement which proves c).

Note that the proof has to be continued by showing that $f(\gamma)$ depends only on $[\gamma]$. But that is another story.

  • $\begingroup$ Thank you! I like the style of your proof very much - you are very precise. There is one thing I'm not sure I understand though, namely why is it sufficient to show what you are claiming (in the proof of (a))? Also, as a bonus question, the discussion in the comments above alludes to the fact that the map that assigns to $\gamma$ the corresponding product behaves nicely under products of paths (has the "homomorphism property"). Is that indeed the case? (Please let me know if this deserves a separate question, which I will then ask.) $\endgroup$ – user531587 Nov 27 '18 at 3:29
  • $\begingroup$ @user531587 Concerning a): We know that $f_{k(j)}([\Gamma_j(\beta_{j-1},\beta_j)]) f_{k(j+1)}([\Gamma_{j+1}(\beta_j,\beta_{j+1})]) = f_{k(j)}([\Gamma_j(\beta_{j-1},\beta'_j)]) f_{k(j+1)}([\Gamma_{j+1}(\beta'_j,\beta_{j+1})])$. This means that in $f_{k(1)}([\Gamma_1]) \dots f_{k(n)}([\Gamma_n])$ which is based on $\beta_1,\dots,\beta_{n-1}$ (recall that $\beta_0,\beta_n$ are constant!) we can replace any $\beta_j$ by $\beta'_j$ without changing the product. $\endgroup$ – Paul Frost Nov 27 '18 at 9:53
  • $\begingroup$ Concerning your bonus question: Given closed paths $\gamma_1, \gamma_2$, the path $\gamma =\gamma_1 \cdot \gamma_2$ is defined by $\gamma(t) =\gamma_1(2t)$ for $t \le 1/2$ and $\gamma(t) =\gamma_2(2t-1)$ for $t \ge 1/2$. Now choose a partition of $I$ such that $1/2$ is a partition point. Then you will easily see that $f(\gamma) = f(\gamma_1) f(\gamma_2)$. $\endgroup$ – Paul Frost Nov 27 '18 at 9:53

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.