Higher Homotopy Seifert-van Kampen theorem

Intuition Beginner

The ordinary van Kampen theorem is a gluing rule for loops. Cover a space by two open pieces, know the loops in each piece and in their overlap, and you can read off the loops of the whole space. The Higher Homotopy Seifert-van Kampen theorem does the same thing one dimension up: it glues two-dimensional fillings, not just one-dimensional loops, so it computes higher homotopy by cover-and-glue.

Why is this a surprise? The second homotopy group counts spheres mapped into a space, and the obvious tools for it — homology — throw away the action of the loops. The information lost is exactly the part that distinguishes a sphere sitting over a complicated fundamental group from one sitting over a simple one. Brown and Higgins found that if you keep the loops and the fillings together in one algebraic object, the gluing rule survives into dimension two. Their 1981 theorem is the precise form of that rule.

The headline payoff is a computation classical homology cannot do. Take a circle and glue on a disk by wrapping its rim twice around — the recipe that builds the real projective plane's two-dimensional skeleton. The theorem hands you the exact second homotopy group, a copy of the whole numbers, carried as a module over the group with two elements. You compute by gluing, in dimension two, what no abelian invariant alone can see.

Visual Beginner

Two filtered pieces of a space, drawn as overlapping regions, each carrying its own tower of algebraic floors (loops on the ground floor, two-dimensional fillings above). The overlap carries a smaller tower. Three arrows feed the two piece-towers and the overlap-tower into a single combined tower on the right, which is the tower of the whole space. The combining rule on each floor is "glue the piece-towers along the overlap-tower," one floor at a time.

The single idea to carry away: the theorem takes a pushout of spaces and turns it into a pushout of these towers. The bottom floor reproduces the old van Kampen gluing of loops; the floors above are the new content, where the fillings glue too.

Worked example Beginner

Build the space $S^1{\cup }_2{D}^2$: start with a circle, take a disk, and attach the disk so that its rim wraps twice around the circle. Filter it in two stages — the circle is the first floor, the whole space with its disk is the second.

Cover this by two filtered pieces. Let one piece be a thickened neighbourhood of the circle together with most of the disk; let the other be a small open disk around the centre of the attached disk. The first piece carries the circle's loop, counted by winding number, so its ground floor is the whole numbers. The second piece is a plain disk, so it has no loops and no rim content. Their overlap is a thin annulus, an open ring, whose loop is again counted by winding number.

The theorem glues these along the overlap. On the ground floor it does the old van Kampen computation: the loop of the circle survives, but the rim of the attached disk forces the relation "twice around equals nothing," giving the group with two elements. On the second floor a new generator appears, the attached disk itself read as a two-dimensional filling.

What this tells us: the second homotopy group of $S^1{\cup }_2{D}^2$ is not zero. It is a copy of the whole numbers, and the group with two elements acts on it by sign. Homology alone reports zero here; the gluing theorem, by keeping the loops and the filling together, reports the true answer.

Check your understanding Beginner

Formal definition Intermediate+

Let $\mathsf{FTop}$ be the category of filtered spaces and $\Pi :\mathsf{FTop}\to \mathsf{Crs}$ the fundamental crossed complex functor of 03.12.55, so $\Pi X_{\ast }$ has ${\pi }_1(X_1,X_0)$ in degree one and ${\pi }_n(X_n,X_{n-1},\ast )$ in degree $n\ge 2$ ^{[Brown-Higgins-Sivera §8.1]}. The Higher Homotopy Seifert-van Kampen theorem (HvKT) asserts that $\Pi$ carries a class of colimits of filtered spaces to colimits of crossed complexes.

The covering data. Let $X_{\ast }$ be a filtered space and $\mathcal{U}=\{U^{\lambda }{\}}_{\lambda \in \Lambda }$ an open cover of $X$ such that the interiors $\{\mathrm{int}{U}^{\lambda }\}$ still cover $X$. Filter each $U^{\lambda }$ and each finite intersection $U^{{\lambda }_0\cdots {\lambda }_p}=U^{{\lambda }_0}\cap \cdots \cap U^{{\lambda }_p}$ by intersection with the filtration of $X_{\ast }$, writing $U_{\ast }^{\lambda }$ and $U_{\ast }^{{\lambda }_0\cdots {\lambda }_p}$. Form the diagram in $\mathsf{FTop}$

$$\underset{\lambda ,\mu }{⨆}{U}_{\ast }^{\lambda \mu }\rightrightarrows \underset{\lambda }{⨆}{U}_{\ast }^{\lambda }\to X_{\ast }$$

whose two left maps are the two families of inclusions $U_{\ast }^{\lambda \mu }\hookrightarrow U_{\ast }^{\lambda }$ and $U_{\ast }^{\lambda \mu }\hookrightarrow U_{\ast }^{\mu }$.

The connectivity condition. Each filtered space $Y_{\ast }$ appearing in the cover — each $U_{\ast }^{\lambda }$ and each finite intersection $U_{\ast }^{{\lambda }_0\cdots {\lambda }_p}$ — is required to be connected in the sense of the filtration: the inclusion $Y_0\hookrightarrow Y_r$ induces a surjection on ${\pi }_0$ for $r\ge 0$ (every component of every stage is reached from a base point in $Y_0$), and ${\pi }_r(Y_r,Y_{r-1},v)=0$ for all $r\ge 1$ and all $v\in Y_0$. A filtered space all of whose covering pieces and their intersections satisfy this is said to satisfy condition $\varphi$. For the skeletal filtration of a CW complex, condition $\varphi$ holds whenever the cover is by subcomplexes (or open neighbourhoods of subcomplexes that deformation-retract onto them).

The pushout statement. Under condition $\varphi$, the diagram obtained by applying $\Pi$,

$$\underset{\lambda ,\mu }{⨆}\Pi U_{\ast }^{\lambda \mu }\rightrightarrows \underset{\lambda }{⨆}\Pi U_{\ast }^{\lambda }\to \Pi X_{\ast },$$

is a coequaliser (a colimit) in $\mathsf{Crs}$. In the two-set case $X=U\cup V$ with $\mathrm{int}U\cup \mathrm{int}V=X$, this reads: $\Pi X_{\ast }$ is the pushout in $\mathsf{Crs}$ of $\Pi U_{\ast }\leftarrow \Pi (U_{\ast }\cap V_{\ast })\to \Pi V_{\ast }$. The notation $\Pi (U_{\ast }\cap V_{\ast })$ means the fundamental crossed complex of the intersection with its induced filtration, and the comparison morphism out of the colimit is the canonical one induced by the inclusions.

Key theorem with proof Intermediate+

Theorem (Higher Homotopy Seifert-van Kampen, two-set form). Let $X_$beafilteredspace,$X = U \cup V$with$\operatorname{int} U \cup \operatorname{int} V = X$,andsuppose$U_*$,$V_*$,and$U_* \cap V_*$eachsatisfythefiltered-connectivitycondition$\phi$. Then the square*

$$\begin{array}{ccc}\Pi (U_{\ast }\cap V_{\ast })& ⟶& \Pi U_{\ast }\\ \downarrow & & \downarrow \\ \Pi V_{\ast }& ⟶& \Pi X_{\ast }\end{array}$$

is a pushout in $\mathsf{Crs}$ ^{[Brown-Higgins 1981]}.

The proof runs through the cubical model, not the crossed complex directly, and that detour is the whole point of the method: a colimit argument needs an algebra with enough room to subdivide and recompose, and the globular crossed complex does not have it while the cubical $\omega$-groupoid does.

Proof structure via the cubical functor $\rho$. To a filtered space $X_{\ast }$ one assigns the cubical set $R{X}_{\ast }$ with $(R{X}_{\ast }{)}_n$ the filtered maps $I_{\ast }^n\to X_{\ast }$ from the $n$-cube with its skeletal filtration; quotienting by filtered homotopy rel vertices gives $\rho X_{\ast }$, the homotopy $\omega$-groupoid of 03.12.57, whose cells carry $n$ partial compositions ${+}_1,\dots ,{+}_n$ and connection operators ^{[Brown-Higgins "On the algebra of cubes"]}. The first structural input is the equivalence of categories $\gamma :\rho \stackrel{\simeq }{\to }\Pi$ between cubical $\omega$-groupoids and crossed complexes; under condition $\varphi$ it is an isomorphism on the spaces at hand, so a colimit statement for $\rho$ transports to one for $\Pi$. It therefore suffices to prove that $\rho X_{\ast }$ is the colimit of $\rho U_{\ast }$ and $\rho V_{\ast }$ over $\rho (U_{\ast }\cap V_{\ast })$.

The colimit statement for $\rho$ is verified by checking the universal property directly, and here the cubical compositions do the work. Given a cubical $\omega$-groupoid morphism on the pieces — that is, $\omega$-groupoid maps $f_U:\rho U_{\ast }\to G$ and $f_V:\rho V_{\ast }\to G$ agreeing on $\rho (U_{\ast }\cap V_{\ast })$ — one must produce a unique $f:\rho X_{\ast }\to G$ restricting to them. Surjectivity-type step (existence of $f$): take a filtered cube $\alpha :I_{\ast }^n\to X_{\ast }$. By the Lebesgue-number lemma applied to the open cover $\{{\alpha }^{-1}\mathrm{int}U,{\alpha }^{-1}\mathrm{int}V\}$ of the compact $I^n$, subdivide $I^n$ into a grid of small subcubes each mapping into $U$ or into $V$. The subdivided cube is a multiple composition $\alpha ={\sum }_{i_1,\dots ,i_n}{\alpha }_{(i_1,\dots ,i_n)}$ in $\rho X_{\ast }$, each factor a cube in $U$ or $V$; define $f(\alpha )$ by applying $f_U$ or $f_V$ to each factor and composing the images in $G$ by the same multiple composition.

Well-definedness step (uniqueness and independence of choices): two subdivisions have a common refinement, and the agreement of $f_U,f_V$ on the overlap together with the interchange laws among the ${+}_j$ shows the two assembled values in $G$ coincide; condition $\varphi$ is exactly what lets a cube straddling the overlap be pushed by a filtered homotopy into $U\cap V$, where the two definitions match. This makes $f$ well-defined on $\rho X_{\ast }$ and forces it to be the unique extension, so $\rho X_{\ast }$ is the colimit. Transporting along $\gamma$, $\Pi X_{\ast }$ is the pushout of $\Pi U_{\ast },\Pi V_{\ast }$ over $\Pi (U_{\ast }\cap V_{\ast })$ in $\mathsf{Crs}$.

As the rigorous core, take the special case that makes the theorem concrete. Let $X=S^1{\cup }_2{D}^2$ with $X_0=X_1=\{\ast \}\cup$ the circle (skeletally, $X_1=S^1$) and $X_2=X$. Choose $U_{\ast }$ a filtered open neighbourhood of $S^1$ together with the punctured attached disk (deformation-retracting onto $S^1$), and $V_{\ast }$ a filtered open disk around the cone point (contractible, with $V_1=\varnothing$ promoted to a point on retraction). Then $\Pi U_{\ast }=\Pi (S^1{)}_{\ast }=(\mathbb{Z}$ in degree one$)$, $\Pi V_{\ast }=(1$ in degree one, $0$ above$)$, and $U_{\ast }\cap V_{\ast }$ retracts to an annulus, $\Pi (U_{\ast }\cap V_{\ast })=(\mathbb{Z}$ in degree one, free on the boundary loop$)$. The pushout in $\mathsf{Crs}$ glues the degree-one groupoids to ${\pi }_1=\mathbb{Z}/2$ (the boundary loop maps to $t^2$ on the $U$ side and to the identity on the $V$ side, after the disk is filled), and in degree two it produces the free crossed $\mathbb{Z}$-module on one generator $r$ with ${\delta }_{2}r=t^2$. Reading off ${\pi }_2=\ker {\delta }_2$ gives a copy of $\mathbb{Z}$ as a $\mathbb{Z}[\mathbb{Z}/2]$-module — the computation completed in full in the Master proof set below. $\square$

Bridge. The cubical detour is the foundational reason the theorem can be proved at all: a colimit is a statement about subdividing and reassembling, and that is exactly the operation the multiple compositions ${+}_1,\dots ,{+}_n$ of $\rho X_{\ast }$ provide and the globular $\Pi X_{\ast }$ lacks. This is exactly the Lebesgue-number subdivision that proves the 1-dimensional van Kampen theorem of 03.12.09, lifted from paths to $n$-cubes, so the higher theorem generalises the classical one rather than replacing it. Putting these together, the equivalence $\gamma :\rho \simeq \Pi$ is the bridge that carries the cubical colimit to a crossed-complex pushout, and the central insight is that condition $\varphi$ is precisely what makes an overlap cube deformable into $U\cap V$. The connectivity hypothesis builds toward the connectivity condition on filtered spaces studied in 03.12.55, and the cubical machinery appears again in 03.12.57, where $\rho X_{\ast }$ and the equivalence $\gamma$ are constructed in full.

Exercises Intermediate+

Advanced results Master

The theorem above is the two-set instance of a general colimit-preservation statement: $\Pi$ carries the colimit of any open cover satisfying condition $\varphi$ to the colimit of crossed complexes. The single hypothesis is connectivity of the pieces and their finite intersections; no path-connectedness of overlaps, no simple-connectivity, no dimension restriction. Three consequences locate the theorem in the surrounding homotopy theory.

Recovery of the ordinary van Kampen theorem. The truncation functor ${\tau }_{\le 1}:\mathsf{Crs}\to \mathsf{Grpd}$ that retains degree one and discards the modules above is a left adjoint, hence preserves colimits, so it carries the HvKT pushout to a pushout of fundamental groupoids:

$${\pi }_1(X,X_0)=\mathrm{colim}({\pi }_1(U,A)\leftarrow {\pi }_1(U\cap V,A)\to {\pi }_1(V,A)).$$

This is Brown's 1967 groupoid Seifert-van Kampen theorem ^{[Brown 1967]}, and choosing $A=\{x_0\}$ with $U\cap V$ path-connected collapses it to the classical amalgamated free product of 03.12.09. The higher theorem is therefore not a different theorem in dimension one — it restricts on the nose to the old one.

The relative Hurewicz theorem as a corollary. For a pair $(X,A)$ with $A$ connected and $(X,A)$ being $(n-1)$-connected, the HvKT applied to a cellular cover computes ${\pi }_n(X,A)$ as a colimit of crossed complexes whose bottom interesting degree is $n$. When $n=2$ this gives ${\Pi }_2(X,A)$ as a free crossed ${\pi }_{1}A$-module on the relative $2$-cells, and abelianising — factoring out the Peiffer action of ${\pi }_{1}A$ — yields $H_2(X,A)$ together with the relative Hurewicz comparison map. The same mechanism in higher dimensions reproduces the relative Hurewicz isomorphism ${\pi }_n(X,A)\cong H_n(X,A)$ for $(n-1)$-connected pairs (with the module structure over ${\pi }_{1}A$ kept), now derived from a gluing principle rather than from a spectral-sequence or chain-level argument ^{[Brown-Higgins-Sivera §8.1]}.

Relation to Blakers-Massey. The Blakers-Massey excision theorem 03.12.21 controls the failure of ${\pi }_n(X,A)\to {\pi }_n(X/B,A/B)$ to be an isomorphism for an excisive triad, in a range of dimensions. HvKT gives the bottom of that range exactly and without a range restriction: in the lowest nonvanishing dimension the comparison is an isomorphism, computed as a colimit of crossed modules, and the first obstruction beyond it is the generalised Whitehead product that Blakers-Massey measures. The two theorems are complementary — HvKT computes the colimit precisely where excision is exact, and Blakers-Massey quantifies the deviation just above that. The HvKT proof, being a direct universal-property verification, gives the isomorphism in the exact range as an equality rather than as the edge of a long exact sequence.

Synthesis. The Higher Homotopy Seifert-van Kampen theorem is the foundational reason higher homotopy can be computed by gluing at all: it states that the fundamental crossed complex functor $\Pi$ preserves the colimits of connected filtered spaces, and this is exactly the property that turns a cover-and-glue decomposition of a space into a presentation of its homotopy. The central insight is that the proof cannot be run in the globular crossed complex and must pass through the cubical $\omega$-groupoid $\rho X_{\ast }$, whose multiple compositions provide the subdivision algebra; the equivalence $\gamma :\rho \simeq \Pi$ is the bridge that returns the result to $\mathsf{Crs}$. The theorem generalises the 1-dimensional van Kampen of 03.12.09 — which it recovers on truncation, not by analogy but as a literal special case — and is dual in spirit to the Mayer-Vietoris principle, replacing the long exact sequence of an excisive cover by a colimit that retains the nonabelian degree-two data an abelian sequence destroys. Putting these together, the relative Hurewicz theorem and the exact range of Blakers-Massey both fall out as corollaries: each is the HvKT colimit read in a regime where it simplifies, the first by abelianising the crossed module and the second by sitting at the bottom of the excision range. The pattern recurs throughout the Brown-Higgins programme, where every classical computation is recovered as a colimit that the crossed complex makes nonabelian and the cube makes provable.

Full proof set Master

Proposition (the canonical computation: ${\pi }_2(S^1{\cup }_2{D}^2)$). Let $X=S^1{\cup }_2{D}^2$ be the mapping cone of the degree-$2$ map $S^1\stackrel{2}{\to }{S}^1$, equivalently the $2$-skeleton of the real projective plane, filtered by $X_0 = {}$,$X_1 = S^1$,$X_2 = X$.ThenHvKTcomputesthebottomof$\Pi X_*$tobethefreecrossed$\pi_1(S^1) = \mathbb{Z}$-module* $$ \delta_2 \colon C_2 \to C_1 = \mathbb{Z} = \langle t \rangle, \qquad \delta_2(r) = t^2, $$ on a single generator $r$; consequently ${\pi }_1(X)=\mathbb{Z}/2$, and $$ \pi_2(X) = \ker\delta_2 \cong \mathbb{Z}, \qquad \text{a } \mathbb{Z}[\mathbb{Z}/2]\text{-module on which the generator of } \mathbb{Z}/2 \text{ acts by } -1. $$

Proof. Choose the cover $U=$ an open filtered neighbourhood of $S^1\subseteq X$ together with the attached disk minus its centre, and $V=$ an open disk about the cone point of the attached $2$-cell. Then $\mathrm{int}U\cup \mathrm{int}V=X$. The piece $U$ deformation-retracts (as a filtered space) onto $S^1$, so $\Pi U_{\ast }$ is concentrated in degree one with $C_1(U)={\pi }_1(S^1)=\mathbb{Z}=⟨t⟩$ and identity higher degrees. The piece $V$ is a contractible $2$-cell, so $\Pi V_{\ast }$ has identity degree one and a single free generator in degree two coming from the cell: $C_2(V)=\mathbb{Z}⟨r⟩$ with ${\delta }_2^V=0$ into the identity group $C_1(V)$. The intersection $U\cap V$ is an annulus retracting onto the attaching circle, so $\Pi (U_{\ast }\cap V_{\ast })$ is concentrated in degree one, $C_1(U\cap V)=\mathbb{Z}=⟨s⟩$, free on the boundary loop $s$ of the disk; condition $\varphi$ holds since each of $U_{\ast },V_{\ast },U_{\ast }\cap V_{\ast }$ is filtered-connected with vanishing relative groups in the relevant degrees.

The two inclusions send the boundary loop $s$ to its image in each piece. In $U$, the boundary of the attached $2$-cell wraps twice around $S^1$, so $s\mapsto t^2\in C_1(U)$. In $V$, the boundary loop bounds the disk $V$, so $s\mapsto 1\in C_1(V)$, and moreover in $V$ the loop $s$ is the boundary ${\delta }_2(r)$ of the cell generator $r$. The pushout in $\mathsf{Crs}$ is computed degree by degree (the truncation functors are left adjoints).

Degree one. The pushout of groupoids $\mathbb{Z}\stackrel{s\mapsto t^2}{\leftarrow }\mathbb{Z}\stackrel{s\mapsto 1}{\to }1$ is $\mathbb{Z}/⟨t^2⟩=\mathbb{Z}/2=⟨t\mid t^2⟩$. Hence ${\pi }_1(X)=\mathbb{Z}/2$.

Degree two. The colimit in $\mathsf{Crs}$ glues the degree-two data along the boundary identification. The generator $r\in C_2(V)$ acquires, in the pushout, the boundary ${\delta }_2(r)=$ (image of $s$ in the glued $C_1$) $=t^2$. The crossed-module axioms force $C_2={\pi }_2(X,S^1)$ to be the free crossed $⟨t⟩=\mathbb{Z}$-module on the one generator $r$ with ${\delta }_{2}r=t^2$. By the structure theory of free crossed modules (the Peiffer identity makes the underlying group abelian modulo the action), the underlying abelian group of $C_2$ is the free $\mathbb{Z}[{\pi }_{1}X]=\mathbb{Z}[\mathbb{Z}/2]$-module on $r$: $$ C_2 \cong \mathbb{Z}[\mathbb{Z}/2],r = \mathbb{Z}\langle r \rangle \oplus \mathbb{Z}\langle t\cdot r \rangle, $$ with ${\pi }_{1}X=\mathbb{Z}/2=\{1,t\}$ acting by permuting the two basis elements (and $t^2=1$ acting as the identity).

Reading off ${\pi }_2$. The boundary ${\delta }_2:C_2\to C_1$ on the abelianised module sends ${\sum }_j{a}_j{t}^{j}r\mapsto t^{2{\sum }_j{a}_j}=t^0=1$ in $C_1=\mathbb{Z}$, because each application of ${\delta }_2$ to a ${\pi }_1$-translate $t^{j}r$ lands on the conjugate $t^j(t^2)t^{-j}=t^2$, and in the universal cover the boundary is read additively as $\epsilon :\mathbb{Z}[\mathbb{Z}/2]\to \mathbb{Z}[\mathbb{Z}/2]$ given by multiplication by $(1+t)$ followed by the degree count. Concretely, identifying $C_2=\mathbb{Z}[\mathbb{Z}/2]$ and $C_1^{\mathrm{ab}}=\mathbb{Z}[\mathbb{Z}/2]/(t-1)=\mathbb{Z}$ (the ${\pi }_1$-coinvariants, since the attaching word is a single relator), the cellular boundary ${\delta }_2$ is the augmentation $\epsilon (t)=1$ scaled by the relation exponent. The kernel is the augmentation ideal $$ \pi_2(X) = \ker\delta_2 = I(\mathbb{Z}/2) = \langle 1 - t \rangle \subseteq \mathbb{Z}[\mathbb{Z}/2], $$ which is the free abelian group of rank one generated by $(1-t)r$. The action of $t$ on $(1-t)r=r-tr$ gives $tr-t^{2}r=tr-r=-(1-t)r$, so the generator of $\mathbb{Z}/2$ acts by $-1$. Therefore ${\pi }_2(X)\cong \mathbb{Z}$ as an abelian group, carrying the sign action of $\mathbb{Z}/2$ — the nonidentity $\mathbb{Z}[\mathbb{Z}/2]$-module structure. $\square$

Remark (what abelian homology misses). The integral homology of $X=S^1{\cup }_2{D}^2=\mathbb{R}{P}^2$-skeleton is $H_0=\mathbb{Z}$, $H_1=\mathbb{Z}/2$, $H_2=0$. The vanishing $H_2=0$ reports nothing in dimension two, yet ${\pi }_2(X)=\mathbb{Z}\ne 0$. The discrepancy is the action: $H_2$ is the ${\pi }_1$-coinvariants of ${\pi }_2$ of the universal cover, and the coinvariants of the sign module $\mathbb{Z}$ under $\mathbb{Z}/2$ are $\mathbb{Z}/(1-(-1))=\mathbb{Z}/2$-torsion that lands in $H_1$ via the Bockstein, leaving $H_2=0$. HvKT keeps the module before passing to coinvariants and so retains the full ${\pi }_2$.

Proposition (functorial 1-truncation recovers Brown 1967). The functor ${\tau }_{\le 1}:\mathsf{Crs}\to \mathsf{Grpd}$ preserves the HvKT pushout, and on the spaces of the theorem it yields the groupoid Seifert-van Kampen pushout.

Proof. ${\tau }_{\le 1}$ is left adjoint to the inclusion $\mathsf{Grpd}\hookrightarrow \mathsf{Crs}$ that views a groupoid as a crossed complex concentrated in degree one (the unit and counit are the identity on degree one and the zero map above), so ${\tau }_{\le 1}$ preserves all colimits, in particular the pushout. On degree one $\Pi X_{\ast }$ is ${\pi }_1(X_1,X_0)$, and the pieces likewise, so the truncated square is the pushout ${\pi }_1(X,A)=\mathrm{colim}({\pi }_1(U,A)\leftarrow {\pi }_1(U\cap V,A)\to {\pi }_1(V,A))$ in $\mathsf{Grpd}$. This is the statement of Brown's 1967 theorem ^{[Brown 1967]}; specialising $A=\{x_0\}$ and assuming $U\cap V$ path-connected recovers the amalgamated free product of 03.12.09. $\square$

Connections Master

• The crossed complex of a filtered space 03.12.55 is the receiving object of this theorem: HvKT is exactly the statement that the functor $\Pi :\mathsf{FTop}\to \mathsf{Crs}$ built there preserves the colimits of connected filtered spaces. The connectivity condition $\varphi$ and the free-crossed-resolution-on-skeleta structure established in that unit are what make the pushout computable, and the crossed-module bottom ${\delta }_2$ proved there is where the nonabelian content of the computation lives.

• The Seifert-van Kampen theorem 03.12.09 is the 1-dimensional case recovered on truncation: applying ${\tau }_{\le 1}$ to the HvKT pushout gives back Brown's 1967 groupoid van Kampen pushout, and a further single-base-point specialisation gives the classical amalgamated free product. This is a literal restriction, so the higher theorem is the genuine generalisation that the 1981 Brown-Higgins paper was named for.

• The relative homotopy group 03.12.52 supplies the objects the theorem computes: the output ${\pi }_n(X_n,X_{n-1},\ast )$ in each degree is a relative homotopy group, and the crossed-module structure on ${\pi }_2(X,A)$ — its action by ${\pi }_{1}A$ and its boundary $\partial$ — is exactly the data HvKT glues. The $S^1{\cup }_2{D}^2$ computation here is the headline instance of the free-crossed-module description introduced in that unit.

• The cubical $\omega$-groupoid of a filtered space 03.12.57 is the machine that proves the theorem: $\rho X_{\ast }$ carries the multiple compositions that make the subdivision argument expressible, and the equivalence $\gamma :\rho \simeq \Pi$ transports the cubical colimit to the crossed-complex pushout. HvKT cannot be proved inside $\mathsf{Crs}$ directly; it is proved in the cubical model and exported.

• The Blakers-Massey theorem 03.12.21 is complementary: HvKT computes the colimit exactly in the dimension where excision is an isomorphism, and Blakers-Massey quantifies the first deviation just above that range as a generalised Whitehead product. Read together they bracket the homotopy excision behaviour of an excisive triad from below (HvKT, exact) and above (Blakers-Massey, with obstruction).

• The Hurewicz theorem 03.12.19 is a corollary in the connected-pair regime: abelianising the crossed module that HvKT produces — factoring out the Peiffer action of ${\pi }_{1}A$ — turns the colimit of crossed complexes into the colimit of chain complexes that computes $H_n(X,A)$, and the comparison morphism is the relative Hurewicz map, an isomorphism for $(n-1)$-connected pairs.

Historical & philosophical context Master

The theorem originates with Ronald Brown and Philip J. Higgins. Their 1981 paper Colimit theorems for relative homotopy groups (J. Pure Appl. Algebra 22, 11-41) ^{[Brown-Higgins 1981]} proved that the fundamental crossed complex functor $\Pi$ preserves the colimits of connected filtered spaces, making relative second homotopy groups computable in cases beyond classical homology. The companion paper of the same year, On the algebra of cubes (J. Pure Appl. Algebra 21, 233-260) ^{[Brown-Higgins "On the algebra of cubes"]}, built the cubical homotopy $\omega$-groupoid $\rho X_{\ast }$ with its multiple compositions and homotopy addition lemma — the algebraic apparatus without which the colimit argument has no place to run. The two papers are a single result split across the geometry (the crossed complex carries the answer) and the algebra (the cube proves it).

A higher-dimensional, nonabelian van Kampen theorem was long thought impossible. The fundamental group is computed by van Kampen because ${\pi }_1$ is a group and gluing is an amalgamation of groups; but ${\pi }_2$ of a space is abelian (the Eckmann-Hilton argument forces commutativity), and an abelian invariant cannot remember the nonabelian gluing data, so it was expected that no gluing theorem for ${\pi }_2$ could carry more than homology already does. Brown and Higgins located the error in this expectation: the relevant object is not the absolute abelian ${\pi }_2(X)$ but the relative ${\pi }_2(X,A)$ as a crossed module over ${\pi }_{1}A$, which is nonabelian in the only way that matters and does glue. Recognising the crossed module (Whitehead 1949) as the right receiving structure, and the cubical $\omega$-groupoid as the right proof structure, is what made the higher theorem both true and provable. The systematic account, including the connectivity condition, the tensor product, and the equivalence of the cubical and crossed-complex models, is the 2011 monograph of Brown, Higgins, and Sivera, Nonabelian Algebraic Topology (EMS Tracts in Mathematics 15) ^{[Brown-Higgins-Sivera §8.1]}, whose §8 is the anchor for this unit.

Bibliography Master

@article{BrownHiggins1981Colimit,
  author  = {Brown, Ronald and Higgins, Philip J.},
  title   = {Colimit theorems for relative homotopy groups},
  journal = {J. Pure Appl. Algebra},
  volume  = {22},
  year    = {1981},
  pages   = {11--41}
}

@article{BrownHiggins1981Cubes,
  author  = {Brown, Ronald and Higgins, Philip J.},
  title   = {On the algebra of cubes},
  journal = {J. Pure Appl. Algebra},
  volume  = {21},
  year    = {1981},
  pages   = {233--260}
}

@book{BrownHigginsSivera2011NAT,
  author    = {Brown, Ronald and Higgins, Philip J. and Sivera, Rafael},
  title     = {Nonabelian Algebraic Topology: Filtered Spaces, Crossed Complexes, Cubical Homotopy Groupoids},
  series    = {EMS Tracts in Mathematics},
  volume    = {15},
  publisher = {European Mathematical Society},
  year      = {2011}
}

@article{Brown1967Groupoids,
  author  = {Brown, Ronald},
  title   = {Groupoids and van {K}ampen's theorem},
  journal = {Proc. London Math. Soc.},
  volume  = {17},
  year    = {1967},
  pages   = {385--401}
}

@article{Whitehead1949CombinatorialII,
  author  = {Whitehead, J. H. C.},
  title   = {Combinatorial homotopy {II}},
  journal = {Bulletin of the American Mathematical Society},
  volume  = {55},
  year    = {1949},
  pages   = {453--496}
}