The case
\(n = 1\) is well-known (
\(\pi_i(S^1) = 0\) for
\(i \geq 2\)). Assume
\(n \geq 3\) odd. Let
\(f \colon S^n \to K(\mathbb{Z}, n)\) generate
\(\pi_n K(\mathbb{Z}, n) \cong \mathbb{Z}\text{,}\) and replace
\(f\) by an equivalent fibration
\(p \colon E_f \to K(\mathbb{Z}, n)\) with
\(S^n \xrightarrow{\simeq} E_f\text{.}\) Let
\(F\) be the fibre. The long exact sequence in homotopy gives:
-
\(F\) is
\(n\)-connected;
-
\(\pi_i(F) \cong \pi_i(S^n)\) for
\(i > n\text{.}\)
So
\(F \simeq S^n_{> n}\) (the
\(n\)-connected cover of
\(S^n\)), and it suffices to show
\(\pi_q(F)\) is finite for all
\(q > n\text{.}\) By repeated application of Hurewicz + Whitehead-tower arguments (
Theorem 23.5 and its proof), it is enough to show
\(H_q(F; \mathbb{Z})\) is finite for all
\(q > n\text{.}\)
Claim: \(H_q(F; \mathbb{Z})\) is finite for all \(q > n\text{.}\)
Suppose for contradiction that there is some
\(q > n\) with
\(H_q(F)\) infinite, and let
\(v > n\) be the minimal such. Then
\(H_v(F)\) has positive rank, while
\(H_q(F)\) is finite for all
\(q\) with
\(n < q < v\text{.}\)
For odd
\(n\text{,}\) the cohomology
\(H^*(K(\mathbb{Z}, n); \mathbb{Q})\) is an exterior algebra on a single generator in degree
\(n\text{;}\) in particular
\(H^q(K(\mathbb{Z}, n); \mathbb{Z})\) is finite for
\(q > n\) except in degree
\(0\text{.}\) By UCT,
\(H_q(K(\mathbb{Z}, n); \mathbb{Z})\) is finite for
\(q > 0\) with
\(q \neq n\text{,}\) and
\(H_n(K(\mathbb{Z}, n); \mathbb{Z}) = \mathbb{Z}\text{.}\)
Run the homological Serre spectral sequence of \(F \to E_f \simeq S^n \to K(\mathbb{Z}, n)\text{:}\)
\begin{equation*}
E^2_{p,q} = H_p\bigl(K(\mathbb{Z}, n);\, H_q(F)\bigr) \;\Longrightarrow\; H_{p+q}(S^n).
\end{equation*}
The bottom row \(E^2_{p,0} = H_p(K(\mathbb{Z}, n))\) is finite in positive degrees except for the entry \(E^2_{n,0} = \mathbb{Z}\text{;}\) the column \(E^2_{0,q}\) is finite for \(n < q < v\) and infinite at \(q = v\text{.}\) By the universal coefficient theorem, every entry \(E^2_{p,q}\) with \(q < v\) is finite, with the two exceptions \(E^2_{0,0} = \mathbb{Z}\) and \(E^2_{n,0} = \mathbb{Z}\text{:}\) indeed, for \(0 < q \leq n\) we have \(H_q(F) = 0\) since \(F\) is \(n\)-connected, and for \(n < q < v\) the group \(H_q(F)\) is finite by minimality of \(v\text{,}\) so \(H_p(K(\mathbb{Z},n); H_q(F))\) is finite for every \(p\text{.}\)
Look at the bidegree
\((0, v)\text{:}\) it contains
\(E^2_{0,v} = H_v(F)\text{,}\) which is infinite. Every differential
out of \(E^r_{0,v}\) vanishes, since its target
\(E^r_{-r,\, v+r-1}\) lies outside the first quadrant. Every differential
into \(E^r_{0,v}\) originates at
\(E^r_{r,\, v-r+1}\text{,}\) and this source is always finite: if
\(v - r + 1 > 0\) it is a subquotient of a finite
\(E^2\)-entry by the previous paragraph, while if
\(v - r + 1 = 0\) (that is,
\(r = v + 1\)) it is a subquotient of
\(E^2_{v+1,\,0} = H_{v+1}(K(\mathbb{Z},n))\text{,}\) which is finite because
\(v + 1 > n\text{.}\) Hence the image of every incoming differential is a finite subgroup, and consequently each
\(E^{r+1}_{0,v} = E^r_{0,v} / \mathrm{im}(d^r)\) has the same rank as
\(E^r_{0,v}\) — infinite. By stabilisation,
\(E^\infty_{0,v}\) is infinite.
But by convergence,
\(E^\infty_{0,v} = F_0 H_v(S^n) \subseteq H_v(S^n) = 0\) (since
\(v > n\)). Contradiction.
Hence
\(H_q(F)\) is finite for all
\(q > n\text{,}\) and therefore
\(\pi_q(F) \cong \pi_q(S^n)\) is finite for all
\(q > n\text{.}\)