Whitehead's crossed module of a pair

Intuition Beginner

A group records symmetries that you can perform and undo. Sometimes two layers of symmetry sit on top of each other: a small group $M$ of "fine" moves, and a larger group $P$ of "coarse" moves that can rearrange the fine ones. A crossed module is the precise bookkeeping for this two-layer situation. It is a rule that turns each fine move into a coarse move, together with a way for coarse moves to act on fine ones, so that the two descriptions never contradict each other.

The picture that makes this concrete comes from topology. Take a space $X$ and a chosen subspace $A$ inside it, with a basepoint. Loops in $A$ form a group, the fundamental group. Two-dimensional "disks in $X$ whose rim sits in $A$" form another group. Pushing the rim of such a disk around records a loop in $A$, and that is the boundary rule. Loops in $A$ can also drag a disk around, which is the action. Whitehead's discovery was that these two structures fit together exactly as a crossed module.

Why bother? Because the fundamental group alone cannot see how two-dimensional pieces are glued in. The crossed module remembers both the one-dimensional loops and the two-dimensional fillings, and the compatibility between them. That extra memory is what lets you compute genuinely two-dimensional homotopy information that one-dimensional tools throw away.

Visual Beginner

A disk drawn so that its outer rim lies entirely inside the subspace $A$, while the disk itself bulges out into the larger space $X$. An arrow labelled "boundary" runs from the disk to the loop traced by its rim inside $A$. A second arrow shows a loop in $A$ sweeping the whole disk around, illustrating the action of loops on disks.

The two arrows are the two halves of a crossed module: the boundary rule sending a fine object to a coarse one, and the action of coarse objects on fine ones.

Worked example Beginner

Take the simplest non-empty case where the two layers interact: let $P$ be any group acting on a group $M$, and ask when the data form a crossed module with the boundary rule sending everything to the identity.

Step 1. Set the boundary rule to be the rule that sends every element of $M$ to the identity element of $P$. Call this the zero boundary.

Step 2. Check the first axiom. It asks that the boundary of "$p$ acting on $m$" equals "$p$ times the boundary of $m$ times $p$ inverse". Both sides are the identity of $P$, since every boundary is the identity. So the first axiom holds for any action.

Step 3. Check the second axiom, the Peiffer rule. It asks that "the boundary of $m$, acting on $m^{\prime }$" equals "$m$ times $m^{\prime }$ times $m$ inverse". The left side uses the boundary of $m$, which is the identity, and the identity acts without doing anything, so the left side is just $m^{\prime }$. The right side is $m$ times $m^{\prime }$ times $m$ inverse.

Step 4. So the Peiffer rule forces $m^{\prime }=m{m}^{\prime }{m}^{-1}$ for all $m,m^{\prime }$ in $M$. That is exactly the statement that $M$ is commutative.

What this tells us: a crossed module with zero boundary is the same thing as a commutative group $M$ carrying an action of $P$. The boundary rule is what allows $M$ to be non-commutative; the Peiffer rule ties the failure of commuting directly to the boundary.

Check your understanding Beginner

Formal definition Intermediate+

A crossed module is a group homomorphism $\partial :M\to P$ together with a (left) action of $P$ on $M$ by group automorphisms, written $(p,m)\mapsto p\cdot m$, subject to two axioms:

CM1 (equivariance). $\partial (p\cdot m)=p\partial (m)p^{-1}$ for all $p\in P$, $m\in M$.

CM2 (Peiffer identity). $\partial (m)\cdot m^{\prime }=m{m}^{\prime }{m}^{-1}$ for all $m,m^{\prime }\in M$.

The action being by automorphisms means $p\cdot (m{m}^{\prime })=(p\cdot m)(p\cdot m^{\prime })$ and $(p{p}^{\prime })\cdot m=p\cdot (p^{\prime }\cdot m)$ with $1\cdot m=m$. CM1 says $\partial$ is $P$-equivariant for the conjugation action of $P$ on itself; CM2 says the action of $M$ on itself induced through $\partial$ is the inner (conjugation) action.

Two elementary consequences follow directly and are recorded here because they organise every later computation. First, $\mathrm{im}\partial$ is a normal subgroup of $P$: by CM1, $p\partial (m)p^{-1}=\partial (p\cdot m)\in \mathrm{im}\partial$. Second, $\ker \partial$ is central in $M$: for $k\in \ker \partial$ and $m\in M$, CM2 gives $mk{m}^{-1}=\partial (m)\cdot k$; applying CM2 the other way, $km{k}^{-1}=\partial (k)\cdot m=1\cdot m=m$, so $k$ commutes with every $m$. Moreover $\ker \partial$ is acted on by $P$ and the action factors through $\mathrm{coker}\partial =P/\mathrm{im}\partial$, making $\ker \partial$ a module over $\mathrm{coker}\partial$.

A morphism of crossed modules $(\partial :M\to P)\to ({\partial }^{\prime }:M^{\prime }\to P^{\prime })$ is a pair of homomorphisms $f:M\to M^{\prime }$, $g:P\to P^{\prime }$ with ${\partial }^{\prime }f=g\partial$ and $f(p\cdot m)=g(p)\cdot f(m)$. Crossed modules and their morphisms form a category, written $\mathbf{XMod}$.

Counterexamples to common slips

• The Peiffer identity is not automatic from CM1. Given any $\partial$ and action satisfying CM1, the data become a crossed module only if the induced action of $M$ on itself is conjugation; a pair $(M,P,\partial )$ can satisfy CM1 and fail CM2.
• The boundary need not be injective and need not be surjective. $\ker \partial$ measures the central "two-dimensional" part invisible to $P$; $\mathrm{coker}\partial$ measures the part of $P$ not realised by boundaries.
• An commutative $M$ with $\partial =0$ is a crossed module, but a non-commutative $M$ with $\partial =0$ is never one: CM2 forces $\partial =0$ to make $M$ abelian.

Notation

Write ${\pi }_n(X,A,x_0)$ for the relative homotopy group of the pair as a set-with-group-structure (a group for $n\ge 2$). When the crossed-module structure on the $n=2$ boundary is what is being invoked, write the boundary explicitly as ${\Pi }_2(X,A,x_0):{\pi }_2(X,A,x_0)\stackrel{\partial }{\to }{\pi }_1(A,x_0)$, following the convention of Brown-Higgins-Sivera ^{[Brown-Higgins-Sivera 2011]}. Thus ${\pi }_2(X,A,x_0)$ names the group, while ${\Pi }_2(X,A,x_0)$ names the homomorphism-with-action — the crossed module itself.

Key theorem with proof Intermediate+

Theorem (Whitehead 1949). Let $(X,A,x_0)$ be a pointed pair of spaces. The boundary homomorphism $\partial :{\pi }_2(X,A,x_0)\to {\pi }_1(A,x_0)$ of the long exact sequence of the pair, together with the standard action of ${\pi }_1(A,x_0)$ on ${\pi }_2(X,A,x_0)$, is a crossed module ${\Pi }_2(X,A,x_0)$.

The formal definition follows Brown-Higgins-Sivera ^{[Brown-Higgins-Sivera 2011]} and Whitehead's original ^{[Whitehead 1949]}.

Proof. Set $M={\pi }_2(X,A,x_0)$ and $P={\pi }_1(A,x_0)$. An element of $M$ is represented by a map of triples $\alpha :(D^2,S^1,s_0)\to (X,A,x_0)$, and $\partial [\alpha ]=[\alpha {|}_{S^1}]\in P$ is the class of the boundary loop. The group operation on $M$ stacks two such maps along the $s_0$-collar, and $\partial$ is a homomorphism because the boundary of a stacked map is the concatenation of boundary loops. The action $p\cdot \alpha$ is defined by transporting $\alpha$ along a loop $p$: drag the basepoint $s_0$ of the disk around the loop $p$ in $A$, dragging the whole map with it; on homotopy classes this gives a well-defined automorphism of $M$ because homotopic loops induce the same transport (this is the relative analogue of basepoint conjugation in the fundamental groupoid 03.12.08).

Verification of CM1. Let $\alpha$ represent $m\in M$ with boundary loop $\partial m=[\ell ]$, and let $p\in P$. Transporting $\alpha$ along $p$ drags both the disk and its rim around $p$. The rim of $p\cdot \alpha$ is therefore the loop $p\cdot \ell \cdot p^{-1}$: travel out along $p$, run around the old rim $\ell$, travel back along $p^{-1}$. Passing to classes, $$ \partial(p \cdot m) = [p \cdot \ell \cdot p^{-1}] = p,[\ell],p^{-1} = p,\partial(m),p^{-1}. $$ This is CM1.

Verification of CM2. This is the substantive axiom. Let $m,m^{\prime }\in M$ be represented by disk-maps $\alpha ,\beta :(D^2,S^1,s_0)\to (X,A,x_0)$, with $\partial m=[\ell ]$ the boundary loop of $\alpha$. We must show $\partial (m)\cdot m^{\prime }=m{m}^{\prime }{m}^{-1}$ in $M$. Both sides are classes of maps $(D^2,S^1,s_0)\to (X,A,x_0)$; we exhibit an explicit homotopy of triples between representatives.

On the left, $\partial (m)\cdot m^{\prime }=[\ell ]\cdot \beta$ is $\beta$ transported around the loop $\ell$. Concretely, take a small sub-disk $D^{\prime }$ in the interior of the parameter disk $D^2$, place a copy of $\beta$ on $D^{\prime }$, and on the annular collar $D^2\setminus D^{\prime }$ run the transport: the collar maps into $A$ by sweeping the basepoint along $\ell$ from the outer rim inward to the rim of $D^{\prime }$. The outer rim of the whole disk traces $\ell \cdot {\ell }^{-1}$, which is null-homotopic in $A$, so the representative does lie in the class $[\ell ]\cdot m^{\prime }$ of $M$, with boundary $\partial ([\ell ]\cdot m^{\prime })=\partial (m)\partial (m^{\prime })\partial (m{)}^{-1}$ by CM1.

On the right, $m{m}^{\prime }{m}^{-1}$ is the disk-map obtained by stacking $\alpha$, then $\beta$, then a reverse copy $\bar{\alpha }$ of $\alpha$ along the collar of $D^2$. Subdivide $D^2$ into three nested annular bands carrying $\alpha$, $\beta$, $\bar{\alpha }$ respectively, with $\alpha$ on the outermost band.

Now construct the homotopy. Because $\alpha$ and $\bar{\alpha }$ are mutually inverse disk-maps, the outer band carrying $\alpha$ and the inner band carrying $\bar{\alpha }$ admit a homotopy, rel the $\beta$-band and rel $s_0$, that slides $\bar{\alpha }$ outward to cancel $\alpha$ against it across the $\beta$-band; what survives the cancellation is precisely the boundary rim $\ell$ of $\alpha$, swept around the $\beta$-band as a loop in $A$. That surviving data is exactly the transport of $\beta$ along $\ell$ — the left-hand representative $[\ell ]\cdot \beta$. The homotopy is a map of triples throughout, since the cancelling bands map into $X$ with their interface rims in $A$, and $s_0$ is fixed. Therefore $$ m m' m^{-1} = [\ell] \cdot m' = \partial(m) \cdot m' $$ in $M={\pi }_2(X,A,x_0)$, which is CM2. $\square$

Bridge. The Peiffer verification is the foundational reason crossed modules are the right algebra for pairs: the boundary loop of a 2-cell acts the way conjugation acts, so that two-dimensional fillings carry exactly the conjugation data that one-dimensional loops impose. This builds toward the free crossed module description of ${\pi }_2$ of a 2-complex, where the Peiffer identity is what cuts a free group of formal fillings down to the actual relative homotopy group, and it appears again in the higher Seifert-van Kampen theorem of Brown-Higgins, where ${\Pi }_2$ is the value of a functor that preserves certain colimits. This is exactly the structure that the fundamental groupoid 03.12.08 is to dimension one: putting these together, ${\Pi }_2$ generalises ${\pi }_1$ to a two-dimensional invariant, and the central insight is that the action plus the Peiffer law is precisely the data lost when one passes from the pair $(X,A)$ to the bare group ${\pi }_2(X,A)$.

Exercises Intermediate+

Advanced results Master

The structural payoff of ${\Pi }_2$ is the theory of free crossed modules, which describes ${\pi }_2$ of a space built from a 1-complex by attaching 2-cells, and which separates the simply-connected from the non-simply-connected case.

Free crossed modules. Let $P$ be a group and $R$ a set with a function $w:R\to P$. The free crossed $P$-module on $w$ is a crossed module $\partial :C(w)\to P$ with a function $R\to C(w)$ lifting $w$, universal among crossed $P$-modules receiving $R$ compatibly. Concretely $C(w)$ is the quotient of the free group $F$ on the set $P\times R$ (symbols $(p,r)$, thought of as $p\cdot \tilde{r}$) by the Peiffer relations, with $\partial (p,r)=pw(r)p^{-1}$ and the evident $P$-action. The quotient of $F$ by the Peiffer subgroup is exactly what enforces CM2; before quotienting one has only a precrossed module.

Theorem (Whitehead 1949). Let $A$ be a connected CW-complex and let $X=A{\cup }_{\{{\phi }_r\}}\{e_r^2{\}}_{r\in R}$ be obtained by attaching 2-cells along maps ${\phi }_r:S^1\to A$ with attaching classes $w(r)=[{\phi }_r]\in {\pi }_1(A,x_0)$. Then $$ \Pi_2(X, A, x_0) = \big(\partial : \pi_2(X, A, x_0) \to \pi_1(A, x_0)\big) $$ is the free crossed ${\pi }_1(A,x_0)$-module on the attaching classes $w:R\to {\pi }_1(A,x_0)$.

This is the two-dimensional analogue of the statement that ${\pi }_1$ of a graph is free: the 2-cells contribute free generators of a crossed module exactly as 1-cells contribute free generators of a group. It is the engine behind the identification of ${\pi }_2$ of a presentation complex with the module of identities among relations in combinatorial group theory.

The mapping cone, two cases. Let $f:S^1\to A$ be a map with class $w=[f]\in {\pi }_1(A,x_0)$, and let $X=A{\cup }_f{e}^2$ be the mapping cone (attach one 2-cell along $f$). Then ${\Pi }_2(X,A)$ is the free crossed ${\pi }_1(A)$-module on the single generator $w$.

Case 1: $A$ simply-connected. If ${\pi }_1(A)=1$ then $w=1$ and the ${\pi }_1(A)$-action is the identity action, so the free crossed module on one generator is just the infinite cyclic group $\mathbb{Z}$ with zero boundary: ${\Pi }_2(X,A)=(\mathbb{Z}\stackrel{0}{\to }1)$. The long exact sequence then gives ${\pi }_2(X)\cong {\pi }_2(X,A)=\mathbb{Z}$ when ${\pi }_2(A)=0$ — for instance $A=S^2$, $X=S^2{\cup }_f{e}^3$ in the analogous one-dimension-up statement, or the cleanest model $A=\ast$, $X=S^2$, recovering ${\pi }_2(S^2)=\mathbb{Z}$ by Hurewicz 03.12.19. Here classical commutative methods suffice: ${\pi }_2$ is the free commutative group on the 2-cells.

Case 2: $A$ not simply-connected. Take $A=S^1$, so ${\pi }_1(A)=\mathbb{Z}=⟨t⟩$, and attach a 2-cell along $f$ of degree $n$, i.e. $w=t^n$. Then $X=S^1{\cup }_{t^n}{e}^2$ is the Moore-type space (presentation complex of $⟨t\mid t^n⟩$), with ${\pi }_1(X)=\mathbb{Z}/n$. The free crossed $\mathbb{Z}$-module on one generator $r$ with $w(r)=t^n$ has underlying group the free $\mathbb{Z}[\mathbb{Z}/n]$-module on one generator (after abelianisation forced by the Peiffer relations, since $\partial r=t^n$ is central modulo $\mathrm{im}\partial$). Concretely ${\pi }_2(X)=\ker \partial$ is the module of identities: it is the $\mathbb{Z}[\mathbb{Z}/n]$-module generated by the single identity $t^n=1$, which works out to ${\pi }_2(X)\cong \mathbb{Z}$ as a group but with a non-identity $\mathbb{Z}/n={\pi }_1(X)$-action permuting the $n$ lifts of the 2-cell to the universal cover. Equivalently, the universal cover $\tilde{X}$ is a wedge-like 2-complex whose second homology is the cyclic $\mathbb{Z}[\mathbb{Z}/n]$-module of identities, and ${\pi }_2(X)=H_2(\tilde{X})$ as ${\pi }_1(X)$-modules. Plain commutative methods see only $H_2(X)=0$; they miss the ${\pi }_1$-module structure entirely, which is exactly what the crossed module retains.

Brown-Spencer equivalence. Crossed modules are not an isolated gadget. Brown-Spencer 1976 ^{[Brown-Spencer 1976]} proved that the category $\mathbf{XMod}$ is equivalent to the category $\mathbf{Cat}(\mathbf{Grp})$ of internal categories in groups (equivalently, group objects in categories, i.e. strict 2-groups). Under this equivalence $\partial :M\to P$ corresponds to the internal category with object group $P$ and morphism group $M⋊P$, source and target encoding $\partial$.

Synthesis. Whitehead's crossed module is the foundational reason the second relative homotopy group is more than a commutative group: the central insight is that the ${\pi }_1(A)$-action plus the Peiffer law upgrade ${\pi }_2(X,A)$ from a module to a genuinely non-commutative two-dimensional invariant, and this is exactly the structure that survives the passage to free crossed modules on attached 2-cells. The mapping-cone computation shows where classical methods stall: in the simply-connected case ${\pi }_2$ is the free commutative group on the cells and Hurewicz suffices, whereas in the non-simply-connected case ${\pi }_2$ becomes the $\mathbb{Z}[{\pi }_1]$-module of identities among relations, which abelian homology cannot see. Putting these together, ${\Pi }_2$ generalises ${\pi }_1$ from dimension one to dimension two, is dual to nothing simpler — it is irreducibly two-layered — and builds toward the higher van Kampen theorem of Brown-Higgins, where the crossed module is the value of a colimit-preserving functor. The bridge is the Brown-Spencer equivalence: crossed modules are strict 2-groups, so the homotopy 2-type of a pair is exactly an algebraic object, and the entire programme of nonabelian algebraic topology appears again as the statement that $n$-types are modelled by crossed $n$-fold complexes.

Full proof set Master

Proposition (the free crossed module on one generator over $\mathbb{Z}$). Let $P=\mathbb{Z}=⟨t⟩$ and let $w:\{r\}\to P$ send $r\mapsto t^n$ with $n\ge 1$. The free crossed $P$-module $\partial :C(w)\to P$ has $\mathrm{im}\partial =⟨t^n⟩$, $\mathrm{coker}\partial =\mathbb{Z}/n$, and $\ker \partial \cong \mathbb{Z}$ with $P$ acting through $\mathrm{coker}\partial =\mathbb{Z}/n$ by cyclic permutation of a basis of the regular representation's invariants; concretely $C(w)$ is the free $\mathbb{Z}[\mathbb{Z}/n]$-module on one generator and $\partial$ is the $\mathbb{Z}[\mathbb{Z}/n]$-linear augmentation-twisted map sending the generator to $t^n$.

Proof. Start from the precrossed module: the free group $F$ on the set $P\times \{r\}=\{(t^k,r):k\in \mathbb{Z}\}$, with $P$-action $t^j\cdot (t^k,r)=(t^{j+k},r)$ and $\partial (t^k,r)=t^{k}w(r)t^{-k}=t^k{t}^n{t}^{-k}=t^n$, using commutativity of $P=\mathbb{Z}$. The crossed module $C(w)$ is $F$ modulo the Peiffer subgroup $⟨⟨F,F⟩⟩$ generated by the Peiffer commutators $$ \langle a, b \rangle = (a b a^{-1}),(\partial(a) \cdot b)^{-1}, \qquad a, b \in F. $$ For generators $a=(t^i,r)$, $b=(t^k,r)$ one has $\partial (a)=t^n$, so $\partial (a)\cdot b=t^n\cdot (t^k,r)=(t^{n+k},r)$. Hence the Peiffer relation forces $$ (t^i, r),(t^k, r),(t^i, r)^{-1} = (t^{n+k}, r) $$ in $C(w)$. Setting $i=0$ shows $(t^k,r)$ and $(t^{n+k},r)$ are conjugate by $(1,r)$; more importantly, taking the relation for varying $i$ shows conjugation by any generator shifts the second index by $n$. Modulo the Peiffer subgroup the group $C(w)$ becomes commutative on the cosets of the index modulo $n$: write $x_j=(t^j,r)$ for $j\in \mathbb{Z}/n$ as the image classes. The Peiffer relations make distinct $x_j$ commute (their Peiffer commutator is a relator) and identify $\partial$-values, so $C(w)$ is the free commutative group ${⨁}_{j\in \mathbb{Z}/n}\mathbb{Z}{x}_j$, i.e. the free $\mathbb{Z}[\mathbb{Z}/n]$-module on one generator $x_0$, with $P=\mathbb{Z}$ acting via $t\cdot x_j=x_{j+1}$ (indices mod $n$), i.e. through $\mathrm{coker}\partial =\mathbb{Z}/n$.

Compute $\partial$. On the module generator, $\partial (x_j)=t^n$ for every $j$, so $\partial \left({\sum }_j{a}_j{x}_j\right)=t^{n{\sum }_j{a}_j}$, whence $\mathrm{im}\partial =⟨t^n⟩$ and $\mathrm{coker}\partial =\mathbb{Z}/⟨t^n⟩=\mathbb{Z}/n$. The kernel is $\{{\sum }_j{a}_j{x}_j:{\sum }_j{a}_j=0\}$, the augmentation kernel of the rank-$n$ free commutative group, which is free commutative of rank $n-1$ — but as a $\mathbb{Z}[\mathbb{Z}/n]$-module it is the augmentation ideal of $\mathbb{Z}[\mathbb{Z}/n]$. For the topological mapping cone $X=S^1{\cup }_{t^n}{e}^2$, the group ${\pi }_2(X)=\ker \partial$ matches $H_2(\tilde{X})$ of the universal cover, the cyclic 2-complex with $n$ lifted cells; the module of identities is generated by the single identity $t^n=1$ and is the augmentation ideal as a $\mathbb{Z}/n$-module, with underlying group ${\mathbb{Z}}^{n-1}$ when $n\ge 1$. The crossed-module axioms hold by construction: CM1 holds because $\partial$ is constant on the $P$-orbit up to conjugation in the commutative $P$, and CM2 holds because the Peiffer quotient is exactly the imposition of CM2. $\square$

Proposition (Peiffer subgroup is normal and $P$-invariant). In any precrossed module $\partial :M\to P$ satisfying CM1, the Peiffer subgroup $⟨⟨M,M⟩⟩$ is a normal $P$-invariant subgroup, and the quotient $M/⟨⟨M,M⟩⟩\to P$ is the universal crossed module under $M$.

Proof. Write $⟨m,m^{\prime }⟩=(m{m}^{\prime }{m}^{-1})(\partial (m)\cdot m^{\prime }{)}^{-1}$. For $P$-invariance, apply $p\in P$ and use CM1 ($\partial (p\cdot m)=p\partial (m)p^{-1}$) together with the automorphism property of the action: $$ p \cdot \langle m, m'\rangle = \big((p\cdot m)(p\cdot m')(p\cdot m)^{-1}\big)\big(\partial(p\cdot m)\cdot(p\cdot m')\big)^{-1} = \langle p \cdot m,, p \cdot m'\rangle, $$ since $\partial (p\cdot m)\cdot (p\cdot m^{\prime })=(p\partial (m)p^{-1})\cdot (p\cdot m^{\prime })=p\cdot (\partial (m)\cdot m^{\prime })$. So the generating set of Peiffer commutators is permuted by $P$, hence the subgroup they generate is $P$-invariant. For normality in $M$, a direct expansion shows $m_0⟨m,m^{\prime }⟩m_0^{-1}$ is a product of Peiffer commutators (the Peiffer commutators satisfy a cocycle-type identity making the subgroup they generate normal); this is the standard verification that the Peiffer subgroup is the smallest normal $P$-invariant subgroup whose quotient satisfies CM2. The induced $\bar{\partial }:M/⟨⟨M,M⟩⟩\to P$ then satisfies CM1 (inherited) and CM2 (imposed), and any crossed-module morphism out of $M$ factors uniquely through the quotient by universality of quotients. $\square$

Connections Master

• The relative homotopy group 03.12.52 is the underlying group ${\pi }_2(X,A,x_0)$ on which the crossed module is built; this unit equips that group with its ${\pi }_1(A)$-action and boundary, turning a bare second relative group into the homomorphism-with-action ${\Pi }_2(X,A,x_0)$. Everything here is the relative homotopy group plus exactly one extra layer of structure.

• The crossed complex 03.12.55 extends a single crossed module to a whole sequence $\cdots \to C_3\to C_2\stackrel{\partial }{\to }{C}_1\rightrightarrows C_0$, the dimension-$n$ analogue in which ${\Pi }_2$ sits as the bottom non-degenerate two stages; the free crossed modules computed here are the degree-2 part of the free crossed complex of a CW-filtration.

• The higher homotopy van Kampen theorem 03.12.58 is the colimit theorem for which crossed modules are the receiving category: ${\Pi }_2$ of a union is computed as a pushout of crossed modules, generalising the groupoid Seifert-van Kampen of 03.12.09 from dimension one to dimension two, and the free crossed module on attached 2-cells is its prototypical output.

• The 2-group / internal-category picture 03.12.56 is the Brown-Spencer repackaging: a crossed module is the same data as a strict 2-group, so ${\Pi }_2(X,A)$ is the algebraic model of the homotopy 2-type of the pair, linking this material to higher-categorical homotopy theory.

Historical & philosophical context Master

J.H.C. Whitehead introduced the crossed module of a pair in his 1949 paper Combinatorial homotopy II (Bull. Amer. Math. Soc. 55, 453-496) ^{[Whitehead 1949]}, as part of his programme to compute ${\pi }_2$ of a 2-complex algebraically. The Peiffer identity is named for Renée Peiffer, who studied the relevant identities among relations in the late 1940s in work connected with Karl Reidemeister's school on combinatorial group theory; the relation between identities among relations and second homotopy groups was a central theme of that circle. Whitehead's free-crossed-module theorem is the precise sense in which the 2-cells of a presentation complex generate ${\pi }_2$ freely, modulo the Peiffer relations.

The abstract notion of crossed module was isolated from its topological origin and connected to other structures over the following decades. Brown and Spencer in 1976 (Proc. Konink. Nederl. Akad. Wetensch. 79, 296-302) ^{[Brown-Spencer 1976]} established the equivalence between crossed modules and internal categories in the category of groups, identifying crossed modules with strict 2-groups and thereby placing ${\Pi }_2$ inside higher category theory. The systematic development of crossed modules and crossed complexes as the working algebra of a nonabelian, higher-dimensional algebraic topology is the subject of Brown, Higgins, and Sivera's Nonabelian Algebraic Topology (2011) ^{[Brown-Higgins-Sivera 2011]}, whose §2.1 is the anchor for this unit.

Bibliography Master

@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}
}

@article{BrownSpencer1976,
  author  = {Brown, Ronald and Spencer, Christopher B.},
  title   = {{G}-groupoids, crossed modules and the fundamental groupoid of a topological group},
  journal = {Proc. Konink. Nederl. Akad. Wetensch. Ser. A},
  volume  = {79},
  year    = {1976},
  pages   = {296--302}
}

@book{BrownHigginsSivera2011,
  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}
}

@book{BrownTopologyAndGroupoids,
  author    = {Brown, Ronald},
  title     = {Topology and Groupoids},
  publisher = {BookSurge},
  year      = {2006}
}

@article{Peiffer1949,
  author  = {Peiffer, Ren\'ee},
  title   = {\"Uber Identit\"aten zwischen Relationen},
  journal = {Mathematische Annalen},
  volume  = {121},
  year    = {1949},
  pages   = {67--99}
}