Does path connectedness simplify the proof of the LES in reduced homology?

79 Views Asked by At

I'm reading Weintraub's Fundamentals of Algebraic Topology, in which there is an exercise (3.4.6 in the first edition) that wants us to show that for every path connected subspace $A$ of a path connected space $X$ there is a long exact sequence in reduced homology:

$$\cdots\to \tilde{H}_{n}(A) \to \tilde{H}_{n}(X) \to H_{n}(X,A) \to \tilde{H}_{n-1}(A) \to \cdots. (1)$$

Here $H_*$ is an arbitrary homology theory and the reduced homology groups are defined as

$$\tilde{H}_{n}(Y):=\ker(H_{n}(Y) \to H_{n}(\bullet)).$$

From the chapter that contains the exercise, we know that $\tilde{H}_{n}(Y) \cong H_{n}(Y,y_0)$ for any $y_0 \in Y$ and about the long exact homology sequence for triplets $B\subseteqq A \subseteqq X$ that is

$$\cdots\to H_{n}(A,B) \to H_{n}(X,B) \to H_{n}(X,A) \to H_{n-1}(A,B) \to \cdots. (2)$$

From what I can tell we get the solution immediately by applying the last result to some $B=x_0\in A$. It seems like the author wants us to re-derive this special case of the long exact sequence for triplets using the path-connectedness hypotheses about $A$ and $X$, but I am skeptical that this would work as it is a purely algebraic result.

My question is whether or not if there is a simpler argument to show that (1) exists in this situation, or if I've made a mistake in identifying (1) and (2).

1

There are 1 best solutions below

2
On

I think the reason for the path-connectedness hypothesis is related to the definition of reduced homology, and not the long exact sequence you are asked to establish: the relative homology groups $H_n(Y,y_0)$ for different $y_0 \in Y$ will be naturally isomorphic if $Y$ is path connected, but if $Y$ has more than one path-component, $H_n(Y,y_0)$ may vary for $y_0$ in different path-components. Thus if you want $\tilde{H}_n(X)$ and $\tilde{H}_n(A)$ to be well-defined, you need to assume that $A$ and $X$ are path-connected.

Update

Apologies -- I misunderstood the OP to be defining $\tilde{H}_*(X) = H_*(A,y_0)$, and Martin's comment is exactly right: taking $\tilde{H}_*(X) = \ker((a_X)_*)$, where $(a_X)_*$ is the map on homology induced by the canonical map $a_X \colon X \to \star$ from $X$ to a one-point set $\star$, the issue of path-connectedness arises when defining $\partial_{\text{red}} \colon H_n(X,A) \to \tilde{H}_{n-1}(A)$: The long exact sequence of a triple gives you the map $\partial_{\text{triple}} \colon H_n(X,A) \to H_{n-1}(A,B)$ hence taking $B = \{y_0\}$ for $y_0\in A$, we obtain $\partial_{\text{red}} \colon H_n(X,A) \to \tilde{H}_{n-1}(A)$ by composing $\partial_{\text{triple}}$ with the isomorphism $r_{y_0}\colon H_{n-1}(A,y_0) \to \tilde{H}_{n-1}(A)$. Since $r_{y_0}$ depends on the path-component of $y_0$ in $A$, $\partial_{\text{red}}$ will thus not be unique unless $A$ is path-connected.

Non-uniqueness: It might be useful to see what happens in a simple example: If $X = X_1\cup X_2\subset\mathbb R^2$ where $$ X_1 = \{(x,\sin(x): 0\leq x\leq 2\pi\}, \quad X_2 \{(x,0): 0\leq x \leq 2\pi\}, $$ and we let $A = X_1\cap X_2= \{(k\pi,0): k\in\{0,1,2\}\}$ and for convenience, let $a_k = (k\pi,0)\in A, (k\in \{0,1,2\})$.

Then $H_n(X)=0$ for $n \geq 2$, $H_1(X)=\mathbb Z^2$ and $H_0(X)=\mathbb Z$, while $H_n(A)=0$ for $n \geq 1$ and $H_0(A) = \mathbb Z^3$ with a basis given by the homology classes of the elements of $A$. The relative homology groups of $(X,A)$ all vanish apart from $H_1(X,A)$, which is isomorphic to $\mathbb Z^2$ (with a basis given by chains $\{[s_0],[s_1]\}$ corresponding to the sine curve between $[0,\pi]$ and $[\pi, 2\pi]$), while the relative homology groups of $(A,\{y_0\}) = (A,\{a_i\})$ (some $i \in \{0,1,2\}$) all vanish apart from $H_0(A,\{y_0\})= H_0(A)/\mathbb Z[a_i]$. The reduced homology of $A$ is $\{\sum_{i=1}^3 c_i[a_i]: \sum_{i=0}^3 c_i =0\}$, and we have isomorphisms $q_i\colon \tilde{H}_0(A) \to H_0(A,\{a_i\})$.

The boundary map $\partial_{\text{red}}\colon H_1(X,A) \to \tilde{H}_0(A)$ should take a chain with boundary in $A$ to a reduced chain in $A$, but the boundary of a chain may not lie in $\ker((a_A)_*)$: we can chose $s_0,s_1$ are as above so that $\partial[s_0] = [a_0]-[a_1]$, $\partial_{\text{pair}} [s_1] = [a_1]-[a_2]$, hence $\partial_{\text{pair}}([s_0]-[s_1]) = -2[a_1] \notin \ker((a_X)_*)$. However if we pick $y_0 = a_2$, the image of $\partial_{\text{triple}}([s_0]-[s_1]) = -2[a_1]+\mathbb Z.[a_2]$, and $q_i^{-1}(-2[a_1]+\mathbb Z.[a_2]) = 2[a_1]-2[a_2]$, giving $\partial_{\text{red}}([s_0]-[s_1]) = 2[a_1]-2[a_2]$. On the other hand, if we take $B=\{a_1\}$ and identify $\tilde{H}_0(A)$ with $H_0(A,\{a_1\})$ then we obtain a boundary map $\partial_{\text{red}}$ for which $\partial_{\text{red}}([s_0]-[s_1]) = 2[a_1] =0$.