Skip to main content

Section 6.4 The Long Exact Sequence of a Pair

We now have three natural maps at our disposal. The inclusion of pairs \(i \colon (A, x_0) \hookrightarrow (X, x_0)\) and the inclusion of triples \(j \colon (X, \{x_0\}, x_0) \hookrightarrow (X, A, x_0)\) induce
\begin{align*} i_* &\colon \pi_n(A, x_0) \longrightarrow \pi_n(X, x_0),\\ j_* &\colon \pi_n(X, x_0) = \pi_n(X, \{x_0\}, x_0) \longrightarrow \pi_n(X, A, x_0), \end{align*}
where we used Example 6.4(i). Concretely, \(j_*\) regards a map \(f \colon (I^n, \partial I^n) \to (X, x_0)\) as a map of triples, which is allowed since \(f(\partial I^n) = \{x_0\} \subseteq A\) and \(f(J^{n-1}) = \{x_0\}\text{.}\) The third map is the boundary map, defined by restriction to the distinguished face:
\begin{equation*} \partial \colon \pi_n(X, A, x_0) \longrightarrow \pi_{n-1}(A, x_0), \qquad [f] \mapsto \bigl[f|_{I^{n-1}}\bigr]. \end{equation*}
As observed above, \(f|_{I^{n-1}}\) is a based map \((I^{n-1}, \partial I^{n-1}) \to (A, x_0)\text{,}\) and a homotopy of triples restricts on \(I^{n-1} \times I\) to a based homotopy in \(A\text{,}\) so \(\partial\) is well defined. In the disc model, \(\partial\) sends \(f \colon (D^n, S^{n-1}, s_0) \to (X, A, x_0)\) to its restriction \(f|_{S^{n-1}} \colon (S^{n-1}, s_0) \to (A, x_0)\text{.}\)
For \(n \geq 2\) all three maps are group homomorphisms: for \(i_*\) and \(j_*\) this is functoriality, and for \(\partial\) it holds because the sum \(f + g\) is formed by concatenation in the \(s_1\)-direction, which restricts to concatenation in the \(s_1\)-direction on the distinguished face. For \(n = 1\) they are maps of pointed sets.
Recall that a sequence of pointed sets \((L, l_0) \xrightarrow{\alpha} (M, m_0) \xrightarrow{\beta} (N, n_0)\) is exact at \(M\) if \(\operatorname{im}(\alpha) = \beta^{-1}(n_0)\text{.}\) For groups and homomorphisms this is the usual notion of exactness.

Proof.

We verify exactness at each of the three types of position.
Exactness at \(\pi_n(X, x_0)\text{.}\) First, \(j_* \circ i_* = 0\) (i.e., \(\operatorname{im}(i_*) \subseteq \ker(j_*))\text{:}\) if \(f \colon (I^n, \partial I^n) \to (A, x_0)\text{,}\) then \(j_* i_* [f]\) is represented by \(i \circ f\text{,}\) a map of triples whose image lies in \(A\text{,}\) so it is trivial by the compression criterion. Conversely, suppose \([f] \in \pi_n(X, x_0)\) with \(j_*[f] = 0\text{,}\) i.e., \([f] \in \ker(j_*)\text{.}\) By the compression criterion, \(f\) is homotopic rel \(\partial I^n\) to a map \(g\) with \(g(S^{n-1}) \subseteq A\text{.}\) Since the homotopy fixes \(\partial I^n\text{,}\) it is a based homotopy, so \([f] = [i \circ g] = i_*[g]\) where \([g] \in \pi_n(A, x_0)\text{.}\)
Exactness at \(\pi_n(X, A, x_0)\text{.}\) First, \(\partial \circ j_* = 0\text{:}\) if \(f \colon (I^n, \partial I^n) \to (X, x_0)\) then \(\partial j_* [f] = [f|_{I^{n-1}}] = [c_{x_0}]\text{,}\) because \(I^{n-1} \subseteq \partial I^n\) is mapped to \(x_0\text{.}\) Conversely, suppose \([f] \in \pi_n(X, A, x_0)\) satisfies \(\partial[f] = 0\text{.}\) Then \(f|_{I^{n-1}}\) is null-homotopic in \(A\) through based maps: there is \(H \colon I^{n-1} \times I \to A\) with \(H(-,0) = f|_{I^{n-1}}\text{,}\) \(H(-,1) = c_{x_0}\) and \(H(\partial I^{n-1} \times I) = \{x_0\}\text{.}\) Define \(g \colon I^n \to X\) by
\begin{equation*} g(s_1, \ldots, s_n) = \begin{cases} H(s_1, \ldots, s_{n-1},\; 1 - 2s_n), & 0 \leq s_n \leq \tfrac{1}{2}, \\[4pt] f(s_1, \ldots, s_{n-1},\; 2s_n - 1), & \tfrac{1}{2} \leq s_n \leq 1. \end{cases} \end{equation*}
This is continuous, since both formulas give \(f|_{I^{n-1}}\) at \(s_n = \tfrac12\text{.}\) We claim that \(g(\partial I^n) = \{x_0\}\text{:}\) on the bottom face \(s_n = 0\) we get \(H(-,1) = c_{x_0}\text{;}\) on the top face \(s_n = 1\) we get \(f(-,1) = c_{x_0}\text{,}\) as the top face lies in \(J^{n-1}\text{;}\) and on the side faces \(s_i \in \{0, 1\}\) with \(i < n\text{,}\) both \(H\) and \(f\) take the value \(x_0\text{,}\) since \(\partial I^{n-1} \times I \subseteq J^{n-1}\text{.}\) Thus \([g] \in \pi_n(X, x_0)\text{.}\) Finally, \(j_*[g] = [f]\text{:}\) the map
\begin{equation*} G(s_1, \ldots, s_n, t) = g\Bigl(s_1, \ldots, s_{n-1},\; \frac{s_n + t}{1 + t}\Bigr) \end{equation*}
is a homotopy from \(G(-,0) = g\) to \(G(-,1) = g(-, (s_n+1)/2) = f\text{.}\) It is a homotopy of triples: on the bottom face it takes values \(H(-, 1 - 2t/(1+t)) \in A\text{,}\) and on \(J^{n-1}\) it is constant at \(x_0\) by the computation of \(g\) on \(\partial I^n\) above. Here we used that the reparametrisation \(s_n \mapsto (s_n + t)/(1+t)\) is a self-map of \(I\) fixing \(1\text{.}\)
Exactness at \(\pi_{n-1}(A, x_0)\text{.}\) First, \(i_* \circ \partial = 0\text{:}\) if \(f \colon (I^n, \partial I^n, J^{n-1}) \to (X, A, x_0)\text{,}\) then \(i_* \partial [f]\) is represented by \(f|_{I^{n-1}}\text{,}\) viewed as a map into \(X\text{.}\) Regarding \(f\) as a map \(I^{n-1} \times I \to X\) with \(s_n\) as the homotopy parameter, \(f\) itself is a based null-homotopy of \(f|_{I^{n-1}}\) in \(X\text{:}\) it is based because \(\partial I^{n-1} \times I \subseteq J^{n-1}\text{,}\) and it ends at \(f(-,1) = c_{x_0}\) because the top face lies in \(J^{n-1}\text{.}\) Conversely, let \([g] \in \pi_{n-1}(A, x_0)\) with \(i_*[g] = 0\text{,}\) and let \(H \colon I^{n-1} \times I \to X\) be a based null-homotopy of \(i \circ g\text{,}\) so \(H(-,0) = g\text{,}\) \(H(-,1) = c_{x_0}\) and \(H(\partial I^{n-1} \times I) = \{x_0\}\text{.}\) Under the identification \(I^{n-1} \times I = I^n\text{,}\) the map \(H\) sends the distinguished face \(I^{n-1} \times \{0\}\) to \(g(I^{n-1}) \subseteq A\text{,}\) and sends \(J^{n-1} = (I^{n-1} \times \{1\}) \cup (\partial I^{n-1} \times I)\) to \(x_0\text{.}\) Hence \(H\) is a map of triples, \([H] \in \pi_n(X, A, x_0)\text{,}\) and \(\partial[H] = [H|_{I^{n-1} \times \{0\}}] = [g]\text{.}\)

Remark 6.15. Naturality.

The long exact sequence is natural: a map of triples \(\varphi \colon (X, A, x_0) \to (Y, B, y_0)\) induces a map of long exact sequences, i.e. all squares in the ladder
commute. For the squares involving \(i_*\) and \(j_*\) this is functoriality, and for the square involving \(\partial\) it holds because \((\varphi \circ f)|_{I^{n-1}} = \varphi \circ (f|_{I^{n-1}})\text{.}\)

Remark 6.16.

The change-of-basepoint bijections are natural with respect to maps of triples, so the group \(\pi_1(A, x_0)\) acts on every term of the long exact sequence: on \(\pi_n(A, x_0)\) and \(\pi_n(X, A, x_0)\) by the constructions above, and on \(\pi_n(X, x_0)\) via \(i_* \colon \pi_1(A, x_0) \to \pi_1(X, x_0)\) and the action of \(\pi_1(X, x_0)\text{.}\) All three maps \(i_*\text{,}\) \(j_*\) and \(\partial\) are equivariant for these actions; for \(\partial\) this is Remark 6.12. This extra structure will be important when we discuss the Hurewicz theorem in Chapter 14.