Hurewicz theorem

Intuition [Beginner]

The Hurewicz theorem answers a basic question about how the two earliest invariants of a space — its homotopy groups and its homology groups — relate to each other. Both record information about loops, surfaces, and higher-dimensional gadgets sitting inside a space. Homotopy is the more refined record (it remembers exactly how a sphere maps in, up to deformation); homology is the cruder record (it remembers only the algebraic shadow). The theorem says that in the lowest dimension where a space carries any structure at all, the two records agree up to a small abelianisation correction.

The intuition is that an $n$-dimensional sphere mapped into a space carries an integer count, and this count is what homology in degree $n$ measures. If your space is connected and has nothing happening in dimensions below $n$, then a sphere mapping into it is rigid enough that the homotopy class is the integer count. In higher dimensions or in the presence of lower features, homotopy outpaces homology and remembers things homology forgets.

The reason this matters: homology is computable through linear algebra, but homotopy is famously hard. The theorem gives you a free transfer of information from the easy side to the hard side, in exactly the range where homotopy is still simple enough to read off.

Visual [Beginner]

A schematic with a connected blob of space on the left, a sphere mapping into it labelled by its homotopy class, and the corresponding homology cycle on the right with the integer count attached. An arrow between them is labelled "Hurewicz map" and a small caption indicates that this arrow is an isomorphism in the lowest non-vanishing dimension.

The picture captures the essential message: a sphere mapping into a space gives both a homotopy class and a homology class, and the two coincide once the lower-dimensional clutter is cleared away. Above the lowest non-vanishing dimension, the two records can diverge — homotopy keeps remembering things that homology forgets.

Worked example [Beginner]

Compute the Hurewicz map for the circle $S^1$ in dimension one and watch it become the abelianisation.

Step 1. The fundamental group of the circle is ${\pi }_1(S^1)=\mathbb{Z}$, generated by the loop that wraps around once. The integer assigned to a loop is its winding number — how many times it goes around.

Step 2. The first homology of the circle is $H_1(S^1)=\mathbb{Z}$, generated by the same loop interpreted now as a one-dimensional cycle. The integer assigned to a one-cycle is again the winding number.

Step 3. The Hurewicz map $h_1:{\pi }_1(S^1)\to H_1(S^1)$ sends winding number to winding number. It is the identity map $\mathbb{Z}\to \mathbb{Z}$, which is an isomorphism.

Step 4. Now consider a figure-eight $X=S^1\vee S^1$. The fundamental group is the free group on two generators, ${\pi }_1(X)=F_2=⟨a,b⟩$. The first homology is ${\mathbb{Z}}^2$, generated by the two loops independently. The Hurewicz map sends a word in $a$ and $b$ to its abelianisation: the count of $a$s minus $a^{-1}$s and the count of $b$s minus $b^{-1}$s. The kernel is exactly the commutator subgroup — words like $ab{a}^{-1}{b}^{-1}$ that have zero net count of each generator.

What this tells us: in dimension one, the Hurewicz map measures the abelianisation of the fundamental group. Loops that bracket-cancel each other in homology are exactly those whose group-theoretic word collapses after the generators are forced to commute.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a topological space with basepoint $x_0$ and let $n\ge 1$. The Hurewicz map is the homomorphism $$ h_n : \pi_n(X, x_0) \to H_n(X) $$ defined as follows. Fix a generator $[S^n]\in H_n(S^n)=\mathbb{Z}$ (the fundamental class of the sphere, oriented in a chosen way). For a homotopy class $[\sigma ]\in {\pi }_n(X,x_0)$ represented by a map $\sigma :S^n\to X$, set $$ h_n([\sigma]) = \sigma_*([S^n]) \in H_n(X), $$ where ${\sigma }_{\ast }:H_n(S^n)\to H_n(X)$ is the induced map on singular homology. The definition is independent of the representative because homotopic maps induce equal homology maps, and independent of the chosen sphere model up to sign by the orientation choice.

The Hurewicz map is natural: for a continuous map $f:X\to Y$ with $f(x_0)=y_0$, the diagram $$ \begin{array}{ccc} \pi_n(X, x_0) & \xrightarrow{h_n} & H_n(X) \ \downarrow f_* & & \downarrow f_* \ \pi_n(Y, y_0) & \xrightarrow{h_n} & H_n(Y) \end{array} $$ commutes. This follows directly from naturality of the singular-homology pushforward.

The map is a group homomorphism. For $n=1$ this requires checking that the concatenation of loops corresponds to the sum of one-cycles up to a boundary; for $n\ge 2$ it follows from the suspension structure on $S^n$ and the fact that the codomain $H_n(X)$ is abelian, which is automatic since singular homology is always abelian.

A space $X$ is $(n-1)$-connected for $n\ge 1$ if it is path-connected and ${\pi }_k(X,x_0)=0$ for all $1\le k\le n-1$. Equivalently, every map $S^k\to X$ extends to $D^{k+1}\to X$ for $k<n$. The convention is that $(-1)$-connected means non-empty and $0$-connected means path-connected. A space is $n$-connected if ${\pi }_k=0$ for $0\le k\le n$, equivalently $(n+1)$-connected in the indexing above.

A pair $(X,A)$ with $A\subseteq X$ is $n$-connected if ${\pi }_k(X,A)=0$ for $1\le k\le n$, equivalently every map of pairs $(D^k,S^{k-1})\to (X,A)$ for $k\le n$ is homotopic rel $S^{k-1}$ to a map landing in $A$.

Counterexamples to common slips

• The Hurewicz map is not injective in general beyond the lowest non-vanishing dimension. The standard example is $S^2$: in dimension three, $h_3:{\pi }_3(S^2)\to H_3(S^2)=0$ kills the Hopf map, which generates ${\pi }_3(S^2)=\mathbb{Z}$. Above the Hurewicz range, homotopy genuinely outpaces homology.
• The kernel of $h_1$ is the commutator subgroup of ${\pi }_1(X)$, not an arbitrary normal subgroup of finite index. Spaces with non-abelian fundamental group are abundant — every closed surface of genus $\ge 1$ has non-abelian ${\pi }_1$ — and the commutator kernel is what the theorem identifies.
• The higher-dimensional theorem requires the connectivity hypothesis ${\pi }_k=0$ for $k<n$. Without it, $h_n$ need be neither injective nor surjective. The torus $T^2=S^1\times S^1$ has ${\pi }_2(T^2)=0$ but $H_2(T^2)=\mathbb{Z}$, so $h_2$ fails to be surjective — the failure is because $T^2$ is not simply-connected.
• The relative Hurewicz theorem requires $A$ simply-connected, not merely path-connected. The simplest counter-example is $(D^2,S^1)$ in dimension two: ${\pi }_2(D^2,S^1)={\pi }_1(S^1)=\mathbb{Z}$ via the boundary map, but $H_2(D^2,S^1)=\mathbb{Z}$ also; the Hurewicz map happens to be an isomorphism here, but for richer pairs the failure when ${\pi }_1(A)$ acts with non-zero action on ${\pi }_n(X,A)$ becomes visible.

Key theorem with proof [Intermediate+]

Theorem (Hurewicz, low-dimensional form; Hatcher Theorem 2A.1). Let $X$ be a path-connected topological space. The Hurewicz map $h_1:{\pi }_1(X,x_0)\to H_1(X)$ is surjective, and its kernel is the commutator subgroup $[{\pi }_1(X),{\pi }_1(X)]$. Equivalently, $h_1$ induces an isomorphism $$ \pi_1(X, x_0)^{\mathrm{ab}} \xrightarrow{\sim} H_1(X). $$

Theorem (Hurewicz, higher-dimensional form; Hatcher Theorem 4.32). Let $X$ be path-connected and $(n-1)$-connected for some $n\ge 2$. Then $H_k(X)=0$ for $1\le k<n$, and the Hurewicz map $h_n:{\pi }_n(X,x_0)\to H_n(X)$ is an isomorphism.

Proof of the higher-dimensional form. The argument uses cellular models and the comparison of cellular and singular homology.

By CW approximation, replace $X$ up to weak equivalence by a CW complex with a single $0$-cell and no cells in dimensions $1,\dots ,n-1$. (This uses the connectivity hypothesis: an $(n-1)$-connected space has a CW model whose $(n-1)$-skeleton is a point.) Call this CW model $Y$. The cellular chain complex of $Y$ is $$ \cdots \to C_{n+1}^{\mathrm{cell}}(Y) \xrightarrow{d_{n+1}} C_n^{\mathrm{cell}}(Y) \xrightarrow{d_n} C_{n-1}^{\mathrm{cell}}(Y) = 0, $$ where $C_n^{\mathrm{cell}}(Y)={⨁}_{\alpha }\mathbb{Z}$ is the free abelian group on the $n$-cells and $d_n=0$ since the target is zero. Cellular and singular homology agree, so $$ H_n(Y) = C_n^{\mathrm{cell}}(Y) / \mathrm{im}(d_{n+1}). $$ For $1\le k<n$ the cellular chain group $C_k^{\mathrm{cell}}(Y)=0$, so $H_k(Y)=0$.

The $n$-skeleton $Y^{(n)}$ is a wedge of $n$-spheres, $Y^{(n)}={\bigvee }_{\alpha }{S}_{\alpha }^n$, one for each $n$-cell. By the basis-of-attaching-maps fact for CW complexes, $$ \pi_n(Y^{(n)}) = \bigoplus_\alpha \mathbb{Z}, $$ generated by the inclusions $S_{\alpha }^n\hookrightarrow Y^{(n)}$. The Hurewicz map at this stage is the identity ${⨁}_{\alpha }\mathbb{Z}\to {⨁}_{\alpha }\mathbb{Z}$ since each $S_{\alpha }^n$ contributes its fundamental class.

Attaching the $(n+1)$-cells via maps ${\phi }_{\beta }:S^n\to Y^{(n)}$ produces $Y^{(n+1)}$. The long exact sequence of the pair $(Y^{(n+1)},Y^{(n)})$ in homotopy and in homology, combined with the fact that the relative homotopy and homology groups in dimension $n+1$ are both free abelian on the $(n+1)$-cells, identifies the kernels of the maps to ${\pi }_n$ and $H_n$ as both being the image of the same boundary $d_{n+1}$. Concretely: an element of ${\pi }_n(Y^{(n)})$ becomes null-homotopic in $Y^{(n+1)}$ if and only if it is a $\mathbb{Z}$-linear combination of the homotopy classes of the attaching maps ${\phi }_{\beta }$, and the same element vanishes in $H_n(Y^{(n+1)})$ if and only if it is the same $\mathbb{Z}$-linear combination of $d_{n+1}([e_{\beta }^{n+1}])={\phi }_{\beta \ast }([S^n])$. Thus $$ \pi_n(Y^{(n+1)}) = \pi_n(Y^{(n)}) / \langle [\varphi_\beta] \rangle, \qquad H_n(Y^{(n+1)}) = C_n^{\mathrm{cell}} / \mathrm{im}(d_{n+1}), $$ and the Hurewicz map is the identification of these two presentations. Cells in dimension $\ge n+2$ do not affect ${\pi }_n$ or $H_n$ by general position, so ${\pi }_n(Y)={\pi }_n(Y^{(n+1)})$ and $H_n(Y)=H_n(Y^{(n+1)})$, and the Hurewicz map between them is an isomorphism.

The CW approximation $Y\to X$ induces an isomorphism on all homotopy and homology groups, and naturality of $h_n$ transfers the conclusion to $X$. $\square$

Proof of the low-dimensional form. The argument is essentially the same, with the abelianisation appearing because ${\pi }_1$ need not be abelian.

By CW approximation, replace $X$ by a CW model $Y$ with a single $0$-cell. The $1$-skeleton $Y^{(1)}$ is a wedge of circles ${\bigvee }_{\alpha }{S}_{\alpha }^1$, with ${\pi }_1(Y^{(1)})=F=$ free group on the loops $\alpha$ and $H_1(Y^{(1)})=F^{\mathrm{ab}}={⨁}_{\alpha }\mathbb{Z}$. The Hurewicz map $h_1$ at this stage sends a word in the free group to its exponent vector, so $h_1$ is exactly abelianisation, surjective with kernel the commutator subgroup.

Attaching $2$-cells via maps ${\phi }_{\beta }:S^1\to Y^{(1)}$ imposes relations: ${\pi }_1(Y^{(2)})=F/⟨⟨[{\phi }_{\beta }]⟩⟩$, the quotient of $F$ by the normal closure of the attaching maps. On the homology side, $H_1(Y^{(2)})=F^{\mathrm{ab}}/⟨d_2(e_{\beta }^2)⟩$, the quotient of $F^{\mathrm{ab}}$ by the image of the cellular boundary $d_2$, which is the $\mathbb{Z}$-linear span of the abelianised attaching words.

The abelianisation of $F/⟨⟨r_{\beta }⟩⟩$ is $F^{\mathrm{ab}}/⟨r_{\beta }^{\mathrm{ab}}⟩$ — this is the standard fact that abelianisation commutes with quotient by a normal subgroup. So the abelianisation of ${\pi }_1(Y^{(2)})$ matches $H_1(Y^{(2)})$, with the Hurewicz map realising the identification. Cells in dimension $\ge 3$ do not affect ${\pi }_1$ or $H_1$, so the conclusion transfers to $Y$, hence to $X$. $\square$

Bridge. The Hurewicz theorem builds toward the entire infrastructure relating homotopy and homology in algebraic topology. The foundational reason it holds is that in the lowest non-vanishing dimension of a CW complex, the cells themselves form a free basis for both the homotopy group (as homotopy classes of attaching spheres) and the homology group (as cellular chains), and the Hurewicz map is exactly the identity on this common basis. This is exactly the same matching of cells with generators that drives the cellular chain complex computation. The central insight is that homotopy and homology agree precisely where they are forced to: in the bottom non-vanishing dimension the homotopy group has no room to be non-abelian or to carry torsion beyond what the cells dictate, and the Hurewicz map identifies these forced agreements.

Putting these together, the theorem reduces to a calculation on the cellular chain complex of an $(n-1)$-connected CW model. The bridge is the recognition that homology is the abelianised shadow of homotopy in the lowest dimension where structure exists, and this same abelianisation pattern appears again in 03.12.20 (Whitehead's theorem) where a map of CW complexes inducing isomorphisms on all ${\pi }_n$ is forced to be a homotopy equivalence — the integer-form Hurewicz theorem is the dimension-by-dimension input to that conclusion. The bridge is also the recognition that the failure of $h_n$ to be injective above the Hurewicz range, exemplified by the Hopf map $\eta :S^3\to S^2$ generating ${\pi }_3(S^2)=\mathbb{Z}$ that maps to zero in $H_3(S^2)=0$, is what forces the rest of stable homotopy theory into existence — every higher homotopy class invisible to homology is a measurement that homology cannot make.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib has the singular chain complex and singular homology as a functor on TopCat, and the homotopy groups ${\pi }_n$ as types via Topology.Homotopy, but does not yet ship the Hurewicz homomorphism nor the Hurewicz theorem. The intended formalisation would read schematically:

[object Promise]

The proof gap is substantive. Mathlib needs the fundamental class $[S^n]$ as a chosen generator (this requires a coherent orientation convention), the Hurewicz map as a group homomorphism (which for $n\ge 2$ uses the suspension structure), the CW approximation theorem identifying every space with a CW model up to weak equivalence, the cellular chain complex of a CW complex, the cellular approximation theorem, and the comparison of cellular and singular homology. Each piece is formalisable from existing infrastructure but has not been packaged. The kernel-as-commutator-subgroup statement at $n=1$ additionally requires the abelianisation functor on groups, which Mathlib has, and the fact that abelianisation commutes with quotient by a normal subgroup, which is a one-line lemma.

Advanced results [Master]

Theorem (relative Hurewicz; Hatcher Theorem 4.37). Let $(X,A)$ be a topological pair with $A$ simply-connected and $(X,A)$ being $n$-connected for some $n\ge 2$. Then $H_k(X,A)=0$ for $1\le k<n$, and the relative Hurewicz map $$ h_n : \pi_n(X, A) \to H_n(X, A) $$ is an isomorphism.

The relative form is the engine for Whitehead's theorem and for the Hurewicz comparison of homotopy and homology fibres of a map. The hypothesis that $A$ is simply-connected ensures the action of ${\pi }_1(A)$ on ${\pi }_n(X,A)$ is the identity action, which is what allows the relative homotopy group to be abelian and to receive a homomorphism to the relative homology.

Theorem (consequences for CW basis). For a CW complex $X$ with one $0$-cell and no cells in dimensions $1,\dots ,n-1$ for some $n\ge 2$, the $n$-skeleton $X^{(n)}$ is a wedge of $n$-spheres, $X^{(n)}={\bigvee }_{\alpha }{S}_{\alpha }^n$, and $$ \pi_n(X^{(n)}) = \bigoplus_\alpha \mathbb{Z}, \qquad H_n(X^{(n)}) = \bigoplus_\alpha \mathbb{Z}, $$ with the Hurewicz map being the identity on this common basis.

This is the basis-of-attaching-maps fact: the $n$-cells of an $(n-1)$-connected CW complex form simultaneously a $\mathbb{Z}$-basis for ${\pi }_n$ at the $n$-skeleton and for $H_n$ at the $n$-skeleton, and the Hurewicz map is the identity on this basis. Attaching higher cells then quotients both groups by the same image of the cellular boundary $d_{n+1}$, preserving the Hurewicz isomorphism.

Theorem (Hopf's example; failure of injectivity beyond Hurewicz range). The Hurewicz map $h_3:{\pi }_3(S^2)\to H_3(S^2)$ is the zero map. Since ${\pi }_3(S^2)=\mathbb{Z}$ is generated by the Hopf map $\eta :S^3\to S^2$ and $H_3(S^2)=0$, the map $h_3$ kills the generator. Beyond the Hurewicz range, homotopy carries information that homology does not see.

The Hopf invariant $H:{\pi }_{4n-1}(S^{2n})\to \mathbb{Z}$ is a homotopy-theoretic measurement that survives precisely because homology cannot detect it: the cup-product structure on $S^{2n}$ vanishes in degree $4n$ for ordinary cohomology, but the Hopf invariant detects the linking number of the inverse images of two regular values of a map $S^{4n-1}\to S^{2n}$, which is genuine new information.

Theorem (Hurewicz and rational homotopy). Let $X$ be a simply-connected space. The rational Hurewicz map $$ h_n \otimes \mathbb{Q} : \pi_n(X) \otimes \mathbb{Q} \to H_n(X; \mathbb{Q}) $$ is an isomorphism in the lowest non-vanishing rational dimension. Equivalently, for $X$ rationally $(n-1)$-connected (i.e., ${\pi }_k(X)\otimes \mathbb{Q}=0$ for $k<n$), ${\pi }_n(X)\otimes \mathbb{Q}\cong H_n(X;\mathbb{Q})$.

The rational Hurewicz theorem is what powers Sullivan's theory of minimal models 03.12.06 and Quillen's theory of rational homotopy via differential graded Lie algebras: the rational homotopy groups of a simply-connected space are computable from the rational cohomology ring and the higher coproducts, with the lowest rational Hurewicz isomorphism as the entry point.

Theorem (Hurewicz and Postnikov towers). For a Postnikov section $X\to X_n$ killing all homotopy above degree $n$, the Hurewicz map ${\pi }_n(X_n)={\pi }_n(X)\to H_n(X_n)$ is an isomorphism, since $X_n$ is the Eilenberg-MacLane space $K({\pi }_n(X),n)$ if $X$ is $(n-1)$-connected. The Hurewicz theorem identifies the bottom $k$-invariant of a Postnikov tower as a cohomology class with coefficients in the bottom homotopy group.

The Postnikov-tower interpretation makes the Hurewicz theorem the foundation of obstruction theory: the $k$-invariants $k^{n+1}\in H^{n+1}(X_{n-1};{\pi }_n(X))$ are cohomology classes whose bottom factor is recognised through the Hurewicz isomorphism.

Theorem (Whitehead's theorem in the Hurewicz form). Let $f:X\to Y$ be a map between simply-connected CW complexes. Then $f$ is a weak homotopy equivalence if and only if $f_ : H_*(X) \to H_*(Y)$ is an isomorphism. Equivalently, between simply-connected CW complexes, integer homology and homotopy detect weak equivalence interchangeably.*

Whitehead's theorem in this form 03.12.20 reduces to the relative Hurewicz theorem applied inductively to the pair $(M_f,X)$, where $M_f$ is the mapping cylinder of $f$ — exactly the argument sketched in Exercise 8 above.

Synthesis. The Hurewicz theorem is the foundational reason that homotopy and homology agree in the lowest non-vanishing dimension. The central insight is that a CW complex with no cells below dimension $n$ has its $n$-cells serving as a simultaneous $\mathbb{Z}$-basis for both ${\pi }_n$ at the $n$-skeleton and $H_n$ at the $n$-skeleton, and the Hurewicz map is the identity on this common basis. Putting these together, the cellular argument identifies the bottom homotopy and homology groups as the same quotient of the same free abelian group by the same image of the cellular boundary, with the abelianisation appearing only at $n=1$ where ${\pi }_1$ can be non-abelian. Above the Hurewicz range, homotopy genuinely outpaces homology — the Hopf map is the canonical witness — and the gap between ${\pi }_n$ and $H_n$ becomes the substance of stable homotopy theory.

This is exactly the same matching of cells with generators that drives the cellular chain complex computation, and it generalises to the relative form for pairs $(X,A)$ with $A$ simply-connected. The bridge to Whitehead's theorem 03.12.20 is the recognition that homology detects weak equivalence between simply-connected CW complexes, with the relative Hurewicz theorem as the dimension-by-dimension input. The bridge to rational homotopy theory 03.12.07 is the rational Hurewicz isomorphism, which seeds the Sullivan minimal-model construction — every simply-connected rational space is reconstructed from its rational cohomology ring and higher coproducts, with the bottom rational Hurewicz isomorphism identifying the first-non-vanishing rational homotopy with the first-non-vanishing rational cohomology. The bridge to obstruction theory and Postnikov towers 03.12.05 is the identification of the $k$-invariants as cohomology classes of Eilenberg-MacLane spaces, with the bottom factor recognised through the Hurewicz isomorphism. Putting these together, the Hurewicz theorem is the dimensional anchor of the entire homotopy-homology comparison: where the two agree, they agree because of cells; where they diverge, the divergence is the content of higher homotopy theory.

Full proof set [Master]

Theorem (Hurewicz, low-dimensional form), proof. Let $X$ be path-connected and $Y$ a CW model of $X$ with a single $0$-cell. Write $Y^{(1)}={\bigvee }_{\alpha }{S}_{\alpha }^1$, the $1$-skeleton. The fundamental group of a wedge of circles is the free group $F$ on the loops, and the first homology is the abelianisation $F^{\mathrm{ab}}={⨁}_{\alpha }\mathbb{Z}$. The Hurewicz map $h_1:F\to F^{\mathrm{ab}}$ is the abelianisation, surjective with kernel the commutator subgroup $[F,F]$.

Attach the $2$-cells via maps ${\phi }_{\beta }:S^1\to Y^{(1)}$. Each attaching map represents a homotopy class $[{\phi }_{\beta }]\in {\pi }_1(Y^{(1)})=F$, namely a word in the generators. By van Kampen's theorem, $$ \pi_1(Y^{(2)}) = F / N, $$ where $N=⟨⟨[{\phi }_{\beta }]⟩⟩$ is the normal closure of the attaching words. On the homology side, the cellular boundary $d_2:C_2^{\mathrm{cell}}\to C_1^{\mathrm{cell}}=F^{\mathrm{ab}}$ sends each generator $e_{\beta }^2$ to the abelianisation $[{\phi }_{\beta }{]}^{\mathrm{ab}}$ of the attaching word, and $$ H_1(Y^{(2)}) = F^{\mathrm{ab}} / \mathrm{im}(d_2). $$ The standard commutative-algebra identity says abelianisation commutes with quotient by a normal subgroup: $$ (F / N)^{\mathrm{ab}} = F^{\mathrm{ab}} / N^{\mathrm{ab}}, $$ where $N^{\mathrm{ab}}$ is the image of $N$ in $F^{\mathrm{ab}}$, which is exactly $\mathrm{im}(d_2)$ since $N$ is normally generated by the words $[{\phi }_{\beta }]$ and abelianising kills the conjugation. So $$ \pi_1(Y^{(2)})^{\mathrm{ab}} = F^{\mathrm{ab}} / \mathrm{im}(d_2) = H_1(Y^{(2)}), $$ and the Hurewicz map realises this identification.

Cells in dimension $\ge 3$ do not affect ${\pi }_1$ (general position, or the cellular approximation theorem) nor $H_1$ (the cellular boundary $d_2$ is the only differential into $C_1^{\mathrm{cell}}$, and $d_3$ has codomain $C_2^{\mathrm{cell}}$ which does not affect $H_1$). So the conclusion transfers from $Y^{(2)}$ to $Y$, and from $Y$ to $X$ via the weak equivalence. $\square$

Theorem (Hurewicz, higher-dimensional form), proof. Let $X$ be $(n-1)$-connected for some $n\ge 2$ and $Y$ a CW model of $X$ with a single $0$-cell and no cells in dimensions $1,\dots ,n-1$. Such a CW model exists by the standard CW approximation argument: starting from any CW model, kill the lower homotopy by attaching cells, which the connectivity hypothesis allows without changing the homotopy type.

The cellular chain complex of $Y$ has $C_k^{\mathrm{cell}}(Y)=0$ for $1\le k\le n-1$, so $H_k(Y)=0$ in this range. The $n$-skeleton $Y^{(n)}$ is a wedge of $n$-spheres, $Y^{(n)}={\bigvee }_{\alpha }{S}_{\alpha }^n$, with one wedge factor per $n$-cell.

Compute ${\pi }_n(Y^{(n)})$. A wedge of $n$-spheres for $n\ge 2$ is $(n-1)$-connected (a map $S^k\to \bigvee S^n$ for $k<n$ misses the wedge point generically and lifts to one factor). The Hurewicz map at the wedge of spheres is the identity ${⨁}_{\alpha }\mathbb{Z}\to {⨁}_{\alpha }\mathbb{Z}$ — this is the base case of the absolute Hurewicz theorem on a wedge of spheres, which can be checked directly using the suspension isomorphism and induction on the number of wedge factors, or as a consequence of the general higher-dimensional Hurewicz theorem applied to a single $n$-sphere where it is immediate.

Attach the $(n+1)$-cells via maps ${\phi }_{\beta }:S^n\to Y^{(n)}$. Each attaching map represents a homotopy class $[{\phi }_{\beta }]\in {\pi }_n(Y^{(n)})={⨁}_{\alpha }\mathbb{Z}$. By the cellular approximation theorem and the long exact sequence of the pair $(Y^{(n+1)},Y^{(n)})$, $$ \pi_n(Y^{(n+1)}) = \pi_n(Y^{(n)}) / \langle [\varphi_\beta] \rangle, $$ where $⟨[{\phi }_{\beta }]⟩$ is the subgroup generated by the attaching classes. On the homology side, the cellular boundary $d_{n+1}:C_{n+1}^{\mathrm{cell}}\to C_n^{\mathrm{cell}}$ sends each $e_{\beta }^{n+1}$ to ${\phi }_{\beta \ast }([S^n])$, which under the identification ${\pi }_n(Y^{(n)})=H_n(Y^{(n)})$ via the wedge-of-spheres Hurewicz isomorphism is exactly $[{\phi }_{\beta }]$. So $$ H_n(Y^{(n+1)}) = C_n^{\mathrm{cell}} / \mathrm{im}(d_{n+1}) = \pi_n(Y^{(n)}) / \langle [\varphi_\beta] \rangle = \pi_n(Y^{(n+1)}), $$ and the Hurewicz map at $Y^{(n+1)}$ is an isomorphism.

Cells in dimension $\ge n+2$ do not affect ${\pi }_n$ (cellular approximation) nor $H_n$ (the differentials $d_k$ for $k\ge n+2$ do not touch $C_n^{\mathrm{cell}}$). So ${\pi }_n(Y)={\pi }_n(Y^{(n+1)})$ and $H_n(Y)=H_n(Y^{(n+1)})$, with the Hurewicz map an isomorphism. The CW approximation transfers the conclusion from $Y$ to $X$. $\square$

Theorem (relative Hurewicz), proof. Let $(X,A)$ be a pair with $A$ simply-connected and $(X,A)$ being $n$-connected for $n\ge 2$. Replace $(X,A)$ by a CW pair $(Y,B)$ weakly equivalent, with $B$ simply-connected and no relative cells in dimensions $1,\dots ,n-1$. The relative cellular chain complex $$ C_k^{\mathrm{cell}}(Y, B) = \bigoplus_{\text{relative } k\text{-cells}} \mathbb{Z} $$ vanishes for $1\le k\le n-1$, so $H_k(Y,B)=0$ in this range.

The relative $n$-skeleton modulo $B$ is a wedge of $n$-spheres, one per relative $n$-cell, and the relative homotopy group ${\pi }_n(Y^{(n)},B)$ is free abelian on these spheres. The free-abelian-ness uses two facts: $B$ is simply-connected, so ${\pi }_1(B)$ acts as the identity on ${\pi }_n(Y,B)$, and $n\ge 2$, so ${\pi }_n(Y,B)$ is abelian. The Hurewicz map at the relative $n$-skeleton is the identity on the common basis.

Attaching $(n+1)$-cells imposes the same relations on ${\pi }_n(Y^{(n+1)},B)$ as on $H_n(Y^{(n+1)},B)$, by the relative version of the cellular argument. Higher cells do not affect ${\pi }_n(Y,B)$ or $H_n(Y,B)$. The Hurewicz map is an isomorphism throughout. $\square$

Theorem (consequences for CW basis), proof. A CW complex with one $0$-cell and no cells in dimensions $1,\dots ,n-1$ for $n\ge 2$ has $n$-skeleton $X^{(n)}$ obtained by attaching $n$-cells to the $0$-cell along the constant map $S^{n-1}\to \mathrm{pt}$. Each attached $n$-cell becomes a wedge factor $S^n$, giving $X^{(n)}={\bigvee }_{\alpha }{S}_{\alpha }^n$. The wedge is $(n-1)$-connected (general-position lifting), so the Hurewicz theorem applies and gives ${\pi }_n(X^{(n)})=H_n(X^{(n)})={⨁}_{\alpha }\mathbb{Z}$. The Hurewicz map sends each generator to itself by definition of the fundamental class on $S^n$. $\square$

Theorem (Hopf's example), proof. The third homology group of the $2$-sphere is computed from the cellular chain complex of the standard CW structure with one $0$-cell and one $2$-cell: $$ C_3^{\mathrm{cell}}(S^2) = 0, \qquad C_2^{\mathrm{cell}}(S^2) = \mathbb{Z}, \qquad C_1^{\mathrm{cell}}(S^2) = 0, \qquad C_0^{\mathrm{cell}}(S^2) = \mathbb{Z}. $$ The chain group in degree three is zero, so $H_3(S^2)=0$. Hence $h_3:{\pi }_3(S^2)\to H_3(S^2)$ is the zero map. The Hopf map $\eta :S^3\to S^2$ generates ${\pi }_3(S^2)=\mathbb{Z}$ (a separate computation, using the Hopf fibration $S^1\to S^3\to S^2$ and the long exact sequence in homotopy), and $h_3(\eta )=0$. $\square$

Theorem (Hurewicz and rational homotopy), proof. Tensoring the integer Hurewicz map $h_n:{\pi }_n(X)\to H_n(X)$ with $\mathbb{Q}$ produces $h_n\otimes \mathbb{Q}:{\pi }_n(X)\otimes \mathbb{Q}\to H_n(X)\otimes \mathbb{Q}=H_n(X;\mathbb{Q})$ (the equality uses the universal coefficient theorem with field coefficients 03.12.18). For $X$ rationally $(n-1)$-connected, the Serre $\mathbb{Q}$-class argument allows replacing $X$ by a $\mathbb{Q}$-equivalent space that is genuinely $(n-1)$-connected, and the integer Hurewicz theorem applies. Tensoring with $\mathbb{Q}$ then gives the rational Hurewicz isomorphism. $\square$

Theorem (Hurewicz and Postnikov towers), proof. Let $X$ be $(n-1)$-connected. The Postnikov section $X_n$ has ${\pi }_k(X_n)={\pi }_k(X)$ for $k\le n$ and ${\pi }_k(X_n)=0$ for $k>n$. By the higher Hurewicz theorem, $X_n$ is the Eilenberg-MacLane space $K({\pi }_n(X),n)$, and $H_n(X_n)={\pi }_n(X)$. The $k$-invariant of the next stage of the Postnikov tower is a cohomology class $k^{n+2}\in H^{n+2}(X_n;{\pi }_{n+1}(X))=H^{n+2}(K({\pi }_n(X),n);{\pi }_{n+1}(X))$, computed via the universal coefficient theorem and the cohomology of Eilenberg-MacLane spaces. $\square$

Theorem (Whitehead's theorem in the Hurewicz form), proof. This is Exercise 8 above, expanded: replace $f:X\to Y$ by an inclusion via the mapping cylinder, then induct on $n$ using the relative Hurewicz theorem. The base case is ${\pi }_1(Y,X)=0$ from the simply-connectedness hypothesis. The inductive step uses the relative Hurewicz isomorphism ${\pi }_n(Y,X)\cong H_n(Y,X)=0$ at each $n\ge 2$, then the long exact sequence in homotopy gives ${\pi }_n(X)\stackrel{\sim }{\to }{\pi }_n(Y)$. $\square$

Connections [Master]

• Homotopy and homotopy group 03.12.01. The Hurewicz theorem connects the homotopy groups defined in 03.12.01 to singular homology. The dimension-one form is the abelianisation of ${\pi }_1$; the higher-dimensional form is the isomorphism in the bottom non-vanishing dimension. Without the Hurewicz theorem, the homotopy groups would remain a separate world from homology, computable only through fibrations and explicit homotopy lifting; with it, the homotopy groups in the lowest non-vanishing dimension become accessible through the much easier homology computation.

• Singular homology 03.12.11. The Hurewicz theorem identifies the lowest non-vanishing singular homology group with the abelianisation of the lowest non-vanishing homotopy group. The proof uses the cellular comparison theorem (cellular and singular homology agree on CW complexes), the cellular approximation theorem, and CW approximation. The combination realises the Hurewicz isomorphism explicitly on the cellular chain complex of an $(n-1)$-connected CW model.

• Cellular homology 03.12.13. The proof of the higher-dimensional Hurewicz theorem is fundamentally a cellular argument: an $(n-1)$-connected CW complex has $n$-cells forming a basis for both ${\pi }_n$ at the $n$-skeleton and $H_n$ at the $n$-skeleton, and attaching $(n+1)$-cells imposes the same relations on both sides. The cellular boundary $d_{n+1}$ encodes the attaching maps as elements of ${\pi }_n$, and the Hurewicz map identifies these elements with the corresponding elements of $H_n$.

• Whitehead's theorem 03.12.20. Whitehead's theorem, in the simply-connected form, says that a map between simply-connected CW complexes inducing isomorphisms on integer homology is a weak homotopy equivalence. The proof inducts on dimension using the relative Hurewicz theorem: at each $n$, the pair $(Y,X)$ is shown to be $n$-connected by combining the relative Hurewicz isomorphism ${\pi }_n(Y,X)=H_n(Y,X)=0$ with the long exact sequence in homotopy. The Hurewicz theorem is the dimension-by-dimension input.

• Eilenberg-MacLane space 03.12.05. An Eilenberg-MacLane space $K(\pi ,n)$ for $\pi$ abelian and $n\ge 1$ is characterised by ${\pi }_n=\pi$ and ${\pi }_k=0$ for $k\ne n$. The Hurewicz theorem identifies $H_n(K(\pi ,n))=\pi$, which is the bottom homology group; higher homology of $K(\pi ,n)$ encodes the Steenrod operations and is computed by the Serre spectral sequence of the path-space fibration. The representability of cohomology by Eilenberg-MacLane spaces — the Brown representability theorem — uses the Hurewicz isomorphism as the bottom case.

• Whitehead tower and rational homotopy 03.12.07. The rational Hurewicz theorem (the rational version of the higher-dimensional Hurewicz isomorphism) is the entry point for Sullivan's theory of minimal models. A simply-connected rational space is determined by its rational cohomology ring and higher coproducts, with the bottom rational Hurewicz isomorphism identifying the first-non-vanishing rational homotopy with the first-non-vanishing rational cohomology. The Whitehead tower kills homotopy from the bottom up using fibrations, and the Hurewicz theorem identifies the bottom layer at each stage.

• Universal coefficient theorem 03.12.18. The rational Hurewicz theorem combines the integer Hurewicz isomorphism with the universal coefficient theorem applied to $\mathbb{Q}$ coefficients: tensoring with $\mathbb{Q}$ kills torsion, and the universal coefficient theorem gives $H_n(X)\otimes \mathbb{Q}=H_n(X;\mathbb{Q})$. The integer Hurewicz isomorphism plus this identification gives the rational Hurewicz isomorphism.

Historical & philosophical context [Master]

The Hurewicz theorem is due to Witold Hurewicz, in a four-paper series titled Beiträge zur Topologie der Deformationen published in Proceedings of the Koninklijke Akademie van Wetenschappen te Amsterdam during 1935 and 1936 ^[pending]. The first two papers introduced the higher homotopy groups ${\pi }_n$ for $n\ge 2$, established their abelianness, and developed the long exact sequence of a fibration. The third paper (vol. 38, 1935) contains the integer-form Hurewicz theorem in its higher-dimensional incarnation: for an $(n-1)$-connected space, ${\pi }_n$ and $H_n$ coincide. The fourth paper (vol. 39, 1936) extended the framework to cover applications including the Hopf invariant.

Hurewicz's introduction of the higher homotopy groups was a conceptual breakthrough comparable to Poincaré's introduction of the fundamental group in 1895. Higher homotopy groups had been considered earlier in special cases — Čech had studied ${\pi }_n(S^n)$ and Hopf had computed ${\pi }_3(S^2)$ — but Hurewicz gave the systematic definition, proved abelianness for $n\ge 2$, and identified the comparison with homology that bears his name. The four-paper series is the foundation of homotopy theory as a separate discipline from homology theory.

The dimension-one form of the theorem — the identification $H_1={\pi }_1^{\mathrm{ab}}$ — has earlier roots. Poincaré had observed in his 1895 Analysis Situs that the first Betti number depends only on the abelianisation of the fundamental group, and explicit computations for surfaces and lens spaces confirmed the pattern. The clean modern statement of the abelianisation isomorphism is due to Hurewicz, who packaged it as the $n=1$ case of his general theorem.

The relative Hurewicz theorem was developed in the late 1940s and early 1950s in the work of J. H. C. Whitehead on CW complexes (Combinatorial homotopy I, II, Bull. AMS 55, 1949) ^[pending] and in the systematic textbook treatments by Steenrod and Spanier in the 1950s and 1960s. The relative form is the engine for Whitehead's theorem on weak equivalence between simply-connected CW complexes (Whitehead 1949) and for the systematic comparison of homotopy and homology in obstruction theory (Eilenberg-MacLane 1947, Cohomology theory in abstract groups III).

The rational Hurewicz theorem, identifying ${\pi }_n(X)\otimes \mathbb{Q}$ with the lowest non-vanishing rational cohomology of a simply-connected space, was developed in the 1950s and 1960s by Serre (using his $\mathbb{Q}$-class formalism in 1953) and brought to its current form in the rational homotopy theories of Quillen (1969) and Sullivan (1977, Infinitesimal computations in topology, Pub. IHES 47).

Bibliography [Master]

[object Promise]