03.12.57 · modern-geometry / homotopy

Cubical $ω$ -groupoid $ρ (X_{*})$

shipped3 tiersLean: none

Anchor (Master): Brown-Higgins-Sivera Part III §13-§14 (cubical $\omega$-groupoids with connections and the equivalence $\gamma$); Brown-Higgins 1981 *On the algebra of cubes* (J. Pure Appl. Algebra 21); Brown-Higgins 1981 *The equivalence of $\infty$-groupoids and crossed complexes* (Cahiers Top. Géom. Diff. 22)

Intuition Beginner

A square is a more cooperative shape than a disk or a triangle. Two squares that share an edge fit together into a longer rectangle, and you can keep gluing left to right, or top to bottom, building a whole grid of squares like floor tiles. The boundary of a square has four flat sides, and each side is itself a smaller cube — an edge. This self-similar, grid-friendly shape is the reason the cubical picture of a space is easier to compute with than the loop-and-disk picture.

The cubical record of a filtered space collects all the ways of mapping a cube of each dimension into the space, respecting the floors of the filtration: vertices land on the bottom floor, edges on the next, faces above that, and so on. A square's four edges must sit one floor down. After we glue and slide these cubes around — counting two cubes as the same when one can be deformed into the other through the floors — we get a single algebraic object that knows how every dimension of the space fits together.

Two features make this object more than a pile of cubes. First, in an $n$ -dimensional cube there are $n$ independent directions, so there are $n$ separate ways to glue: along direction one, direction two, and so on. Second, there are special filler cubes called connections that let you "turn a corner" — fold one direction onto another. Connections are degenerate cubes that are not merely constant; they bend. Together the directional gluings and the connections give the grid enough flexibility to model the full homotopy content of the space.

Visual Beginner

Picture a flat grid of square tiles. Each tile is a map of a square into the space. Two tiles in the same row share a vertical edge and glue side by side; two tiles in the same column share a horizontal edge and glue top to bottom. The grid shows the two independent gluing directions of squares. Off to the side, draw one special tile bent like a folded napkin: that is a connection, a filler that turns the horizontal direction into the vertical one.

The two things to keep in view are the two gluing directions in the grid and the bent connection tile. The grid gives the compositions; the connection gives the folding that an ordinary constant filler cannot. Both are needed before the grid of cubes becomes a faithful model of the space.

Worked example Beginner

Compose two squares of paths along a shared edge, in the simplest filtered space: a space filtered so that the bottom floor is a single point and everything else is free.

Step 1. Take a square $a$ whose four edges are paths, with its right edge equal to a path $p$ . Take a second square $b$ whose left edge is the same path $p$ .

Step 2. Because the right edge of $a$ matches the left edge of $b$ , the two squares share that edge exactly. Glue them along it. The result is a wider rectangle, which we reparametrise back to a single square. Call it $a +_{1} b$ , the composite in direction one (the horizontal direction).

Step 3. Read off the edges of the composite. Its top edge is the top of $a$ followed by the top of $b$ . Its bottom edge is the bottom of $a$ followed by the bottom of $b$ . Its left edge is the left edge of $a$ . Its right edge is the right edge of $b$ . The shared path $p$ has disappeared into the interior.

Step 4. Check the bookkeeping in the other direction. If instead $a$ and $b$ had matched along a horizontal edge — bottom of one equal to top of the other — we would glue vertically and write $a +_{2} b$ . The two gluings $+_{1}$ and $+_{2}$ are independent operations on squares.

What this tells us: squares carry two separate compositions, one per direction, and each is just edge-matching plus regluing. There is no need to choose a basepoint or to break the symmetry between the directions. This even-handed, grid-like algebra is exactly what makes the cubical model easier to subdivide and reassemble than the loop picture.

Check your understanding Beginner

Formal definition Intermediate+

Let $I = [0, 1]$ and let $I^{n}$ be the $n$ -cube with its skeletal filtration $I_{*}^{n}$ : $(I^{n})_{q}$ is the union of all faces of dimension $\leq q$ , so $(I^{n})_{0}$ is the set of $2^{n}$ vertices, $(I^{n})_{1}$ the 1-skeleton, and so on. A filtered space $X_{*}$ is as in 03.12.54: a nested sequence $X_{0} \subseteq X_{1} \subseteq \dots$ with $X = ⋃_{n} X_{n}$ .

The cubical singular complex. The cubical singular complex $R (X_{*}) = R_{*} (X_{*})$ has, in degree $n$ , the set $$ R_n(X_*) = \mathrm{FTop}\big(I^n_*, X_*\big) $$ of filtered maps $f : I_{*}^{n} \to X_{*}$ , that is, continuous maps $f : I^{n} \to X$ with $f ((I^{n})_{q}) \subseteq X_{q}$ for all $q$ ^{[Brown-Higgins-Sivera Part III]}. For each $i \in {1, \dots, n}$ and $α \in {-, +}$ there are face maps $\partial_{i}^{α} : R_{n} \to R_{n - 1}$ , $(\partial_{i}^{α} f) (t_{1}, \dots, t_{n - 1}) = f (t_{1}, \dots, t_{i - 1}, α, t_{i}, \dots, t_{n - 1})$ (with $α$ read as $0$ or $1$ ), and degeneracy maps $ε_{i} : R_{n - 1} \to R_{n}$ , $(ε_{i} g) (t_{1}, \dots, t_{n}) = g (t_{1}, \dots, t_{i}, \dots, t_{n})$ , dropping the $i$ -th coordinate. These satisfy the standard cubical identities.

$R (X_{*})$ also carries connections $Γ_{i}^{α} : R_{n - 1} \to R_{n}$ for $1 \leq i \leq n - 1$ , induced by precomposition with the connection maps $γ_{i}^{α} : I^{n} \to I^{n - 1}$ , $$ \gamma_i^+(t_1, \dots, t_n) = (t_1, \dots, t_{i-1}, \max(t_i, t_{i+1}), t_{i+2}, \dots, t_n), $$ and similarly $γ_{i}^{-}$ with $min$ in place of $max$ . Thus $Γ_{i}^{α} g = g \circ γ_{i}^{α}$ . A connection is a degeneracy that folds the $i$ and $i + 1$ directions together by a maximum or minimum; it is not the constant-extension degeneracy $ε_{i}$ , and it is the extra cubical operation that has no simplicial analogue.

The fundamental cubical $ω$ -groupoid. Define $ρ (X_{*}) = ρ_{*} (X_{*})$ to be the quotient $$ \rho_n(X_*) = R_n(X_*) / \simeq, $$ where $f ≃ f^{'}$ means $f$ and $f^{'}$ are filtered-homotopic rel vertices: there is a filtered homotopy $H : I^{n} \times I \to X$ through filtered maps, constant on the vertices $(I^{n})_{0}$ ^{[Brown-Higgins-Sivera Part III]}. Write $⟨ f ⟩$ for the class of $f$ . The faces, degeneracies, and connections descend to $ρ$ .

On $ρ$ there are $n$ partial compositions $+_{i}$ ( $1 \leq i \leq n$ ): for $⟨ a ⟩, ⟨ b ⟩ \in ρ_{n}$ with $\partial_{i}^{+} a = \partial_{i}^{-} b$ (the relevant faces agree on the nose, which can always be arranged within a class), the composite $a +_{i} b$ is the cube obtained by juxtaposing $a$ and $b$ in the $i$ -direction and reparametrising $[0, 2] \to [0, 1]$ : $$ (a +i b)(t_1, \dots, t_n) = \begin{cases} a(t_1, \dots, 2 t_i, \dots, t_n) & t_i \le \tfrac12, \ b(t_1, \dots, 2 t_i - 1, \dots, t_n) & t_i \ge \tfrac12. \end{cases} $$ These compositions, together with the faces, degeneracies, and connections, make $\rho(X*) $a * * c u bi c a l$ \omega $- g r o u p o i d w i t h co nn ec t i o n s * * : a c u bi c a l se tw i t h co nn ec t i o n s in w hi c h e a c h$ +_i $i s a p a r t ia l l y - d e f in e d, a ssoc ia t i v eco m p os i t i o n w i t h t h e d e g e n er a t ec u b es$ \varepsilon_i $a s i d e n t i t i es, e v er y e l e m e n t ha s anin v er se$ -_i $f or e a c h$ i$, and the interchange law $$ (a +_i b) +_j (c +_i d) = (a +_j c) +_i (b +j d) $$ holds whenever both sides are defined ( $i \neq = j$ ). The connections satisfy the transport laws coupling $Γ_{i}^{α}$ with the compositions, e.g. $\Gamma_i^+(a +{i} b)$ and the connection of an identity is an identity. The axioms in full are CG1-CG6 of ^{[Brown-Higgins 1981 cubes]}.

Counterexamples to common slips

A connection $Γ_{i}^{α}$ is not the degeneracy $ε_{i}$ . The degeneracy $ε_{i} g$ is constant in the $i$ -direction; the connection $Γ_{i}^{α} g$ varies, folding two directions by $max$ or $min$ . Dropping the connections leaves a cubical $ω$ -groupoid that is not equivalent to crossed complexes — the connections are exactly what supply the "second degeneracy" the equivalence needs.
The compositions $+_{i}$ are partial, not total: $a +_{i} b$ is defined only when $\partial_{i}^{+} a = \partial_{i}^{-} b$ . Writing $a +_{i} b$ for non-matching faces is meaningless, just as composing arrows requires matching endpoints in an ordinary groupoid.
Filtered homotopy is rel vertices and through filtered maps: a homotopy that leaves the filtration is not allowed, or the quotient would collapse the filtration data and lose the higher homotopy groups.

Key theorem with proof Intermediate+

Theorem (interchange holds in $ρ$ ). Let $X_ $b e a f i l t er e d s p a ce an d l e t$ 1 \le i < j \le n $. F or$ \langle a \rangle, \langle b \rangle, \langle c \rangle, \langle d \rangle \in \rho_n(X_*)$ such that all four faces match so that both sides below are defined, the interchange law* $$ (a +_i b) +_j (c +_i d) = (a +_j c) +_i (b +j d) $$ *holds in $\rho_n(X) $. M or eo v er t h er e l a t i o n$ \simeq $i s a co n g r u e n ce : e a c h$ +_i$ is well-defined on classes.

The construction follows Brown-Higgins ^{[Brown-Higgins 1981 cubes]} and the Part III treatment of ^{[Brown-Higgins-Sivera Part III]}.

Proof. First, the congruence claim. Suppose $a ≃ a^{'}$ and $b ≃ b^{'}$ with $\partial_{i}^{+} a = \partial_{i}^{-} b$ and $\partial_{i}^{+} a^{'} = \partial_{i}^{-} b^{'}$ . Let $H : a ≃ a^{'}$ and $K : b ≃ b^{'}$ be filtered homotopies rel vertices. Because the homotopies are rel vertices, in particular they agree on the shared face throughout (the matching $(n - 1)$ -cube has its own vertices fixed), so $H$ and $K$ can be juxtaposed in the $i$ -direction at every time $s \in I$ to give a filtered homotopy $H +_{i} K : (a +_{i} b) ≃ (a^{'} +_{i} b^{'})$ , again rel vertices. Hence $⟨ a +_{i} b ⟩$ depends only on $⟨ a ⟩$ and $⟨ b ⟩$ , so $+_{i}$ is well-defined on $ρ_{n}$ .

Now interchange. Consider the four cubes $a, b, c, d$ arranged in a $2 \times 2$ block in the $i$ - and $j$ -directions: $a$ at position $(-, -)$ , $b$ at $(+, -)$ in the $i$ -direction, $c$ at $(-, +)$ in the $j$ -direction, $d$ at $(+, +)$ . The matching hypotheses are $$ \partial_i^+ a = \partial_i^- b, \quad \partial_i^+ c = \partial_i^- d, \quad \partial_j^+ a = \partial_j^- c, \quad \partial_j^+ b = \partial_j^- d, $$ and at the centre the four agree on the corner $(n - 2)$ -cube $\partial_{i}^{+} \partial_{j}^{+} a = \partial_{i}^{-} \partial_{j}^{+} b = \partial_{i}^{+} \partial_{j}^{-} c = \partial_{i}^{-} \partial_{j}^{-} d$ . Subdivide the cube $I^{n}$ in the $i$ - and $j$ -coordinates at $\frac{1}{2}$ , producing four sub-blocks $B_{--}, B_{+-}, B_{-+}, B_{++}$ . Define a single map $Φ : I^{n} \to X$ by setting it equal to (a reparametrised) $a$ on $B_{--}$ , $b$ on $B_{+-}$ , $c$ on $B_{-+}$ , $d$ on $B_{++}$ .

This $Φ$ is well-defined and continuous: on the internal wall between $B_{--}$ and $B_{+-}$ (where $t_{i} = \frac{1}{2}$ ) the two pieces restrict to $\partial_{i}^{+} a$ and $\partial_{i}^{-} b$ , which agree; on the wall between $B_{--}$ and $B_{-+}$ (where $t_{j} = \frac{1}{2}$ ) they restrict to $\partial_{j}^{+} a$ and $\partial_{j}^{-} c$ , which agree; the remaining two internal walls match by the other two hypotheses; and at the central edge $t_{i} = t_{j} = \frac{1}{2}$ all four agree by the corner condition. By the pasting lemma $Φ$ is a continuous filtered map. Now $Φ$ reparametrised by first cutting in the $i$ -direction then the $j$ -direction is exactly $(a +_{i} b) +_{j} (c +_{i} d)$ , while cutting first in the $j$ -direction then the $i$ -direction is $(a +_{j} c) +_{i} (b +_{j} d)$ . These two reparametrisations of $I^{n}$ differ only by an order-preserving self-homeomorphism of $I^{n}$ fixing the vertices, namely the one that linearly identifies the two ways of cutting $[0, 1]^{2}$ into four equal sub-squares. Such a self-homeomorphism is a filtered homotopy rel vertices from one reparametrisation to the other (slide the cut lines continuously). Hence the two iterated composites are filtered-homotopic rel vertices, so equal in $ρ_{n} (X_{*})$ . $□$

Bridge. The interchange law is the foundational reason the cubical compositions assemble into a coherent algebra rather than $n$ unrelated monoid structures: it is exactly the statement that the two ways of reading a $2 \times 2$ grid — rows-then-columns or columns-then-rows — give the same answer, which is what a faithful model of a filtered homotopy type must satisfy. The pasting argument is dual to the way a double groupoid composes squares, and it generalises the Eckmann-Hilton interchange that forces $π_{n}$ to be abelian for $n \geq 2$ : here the interchange is kept with its directional information rather than collapsed, which is why $ρ$ remembers more than homology. Putting these together, the cubical $ω$ -groupoid builds toward the equivalence with crossed complexes, where the rows-and-columns algebra of $ρ$ is matched against the boundary algebra of $Π$ , and the same subdivision-and-paste mechanism appears again in the proof of the higher homotopy van Kampen theorem, where small cubes are composed and then deformed into the gluing subspaces.

Exercises Intermediate+

Exercise 3 (medium, short-answer).

Show that the degenerate cube $ε_{i} g$ is a two-sided identity for $+_{i}$ : for any $a$ with $\partial_{i}^{-} a = g$ , the composite $(ε_{i} g) +_{i} a = a$ in $ρ$ , and similarly on the right.

Hint

$ε_{i} g$ is constant in the $i$ -direction; juxtaposing a constant block then reparametrising is a filtered homotopy rel vertices to the original.

Answer

The composite $(ε_{i} g) +_{i} a$ juxtaposes a block constant in the $i$ -direction (with value $g$ on the matching face) in front of $a$ and rescales $[0, 2] \to [0, 1]$ . The constant block can be shrunk to zero width by a filtered homotopy rel vertices: slide the cut line $t_{i} = \frac{1}{2}$ to $t_{i} = 0$ , which is an order-preserving self-homeomorphism of $I^{n}$ fixing vertices, hence a filtered homotopy. In the limit the composite becomes $a$ reparametrised by the identity, so $(ε_{i} g) +_{i} a = a$ in $ρ$ . The right identity is symmetric. Rubric: full credit for the constant-block-shrinking homotopy and the rel-vertices justification.

Exercise 4 (medium, short-answer).

Give the inverse $-_{i} a$ of a class $⟨ a ⟩ \in ρ_{n}$ under $+_{i}$ , and explain why $a +_{i} (-_{i} a)$ is the degenerate identity in $ρ$ but not on the nose in $R$ .

Hint

Reverse the $i$ -th coordinate: $(-_{i} a) (\dots, t_{i}, \dots) = a (\dots, 1 - t_{i}, \dots)$ . The composite is an "out-and-back" that contracts.

Answer

Define $(-_{i} a) (t_{1}, \dots, t_{n}) = a (t_{1}, \dots, 1 - t_{i}, \dots, t_{n})$ , the reflection of $a$ in the $i$ -direction; its faces are arranged so that $\partial_{i}^{+} a = \partial_{i}^{-} (-_{i} a)$ , making $a +_{i} (-_{i} a)$ defined. On the nose $a +_{i} (-_{i} a)$ is the cube that runs $a$ then runs $a$ backwards in the $i$ -direction; this is not literally constant, so it is not the identity in $R$ . But it is filtered-homotopic rel vertices to $ε_{i} (\partial_{i}^{-} a)$ by the standard "out-and-back contracts" homotopy (push the turning point in from $t_{i} = 1$ to $t_{i} = 0$ ). Hence in $ρ$ , where we have quotiented by $≃$ , the composite equals the degenerate identity. Rubric: full credit for the reflection inverse and the contraction homotopy distinguishing $ρ$ from $R$ .

Exercise 5 (medium, short-answer).

Explain why the connection $Γ_{1}^{+} g$ (for $g$ a 1-cube, i.e. a path) is a square that is genuinely two-dimensional in content, and write down its four edges.

Hint

$Γ_{1}^{+} g = g \circ γ_{1}^{+}$ with $γ_{1}^{+} (s, t) = max (s, t)$ , a path of the variable $max (s, t)$ .

Answer

$Γ_{1}^{+} g (s, t) = g (max (s, t))$ . Its edges: $\partial_{1}^{-} (Γ_{1}^{+} g) (t) = g (max (0, t)) = g (t)$ and $\partial_{1}^{+} (Γ_{1}^{+} g) (t) = g (max (1, t)) = g (1)$ (constant); $\partial_{2}^{-} (Γ_{1}^{+} g) (s) = g (max (s, 0)) = g (s)$ and $\partial_{2}^{+} (Γ_{1}^{+} g) (s) = g (max (s, 1)) = g (1)$ (constant). So two opposite edges are the path $g$ and the other two are constant at the endpoint $g (1)$ : the square folds the path against the corner. It is not the constant-extension degeneracy $ε_{1} g$ (whose edges would all be $g$ or constant in the constant pattern), because $max (s, t)$ varies in both coordinates. Rubric: full credit for the four edges and the contrast with $ε_{1} g$ .

Exercise 6 (hard, short-answer).

Prove the special case of interchange known as the middle-four exchange for $2$ -cubes: for squares $a, b, c, d$ filling a $2 \times 2$ grid, $(a +_{1} b) +_{2} (c +_{1} d) = (a +_{2} c) +_{1} (b +_{2} d)$ in $ρ_{2}$ .

Hint

This is the $n = 2$ , $i = 1$ , $j = 2$ case of the Key theorem. Reuse the single map $Φ$ on the four sub-squares.

Answer

Tile the square $I^{2}$ into four sub-squares by cutting at $s = \frac{1}{2}$ and $t = \frac{1}{2}$ . Put $a$ bottom-left, $b$ bottom-right, $c$ top-left, $d$ top-right (after reparametrisation). The matching conditions $\partial_{1}^{+} a = \partial_{1}^{-} b$ , $\partial_{1}^{+} c = \partial_{1}^{-} d$ , $\partial_{2}^{+} a = \partial_{2}^{-} c$ , $\partial_{2}^{+} b = \partial_{2}^{-} d$ , plus agreement at the centre vertex, make the four pieces agree on every internal wall, so by the pasting lemma they define one continuous filtered square $Φ$ . Reading $Φ$ as columns-then-rows gives $(a +_{1} b) +_{2} (c +_{1} d)$ ; reading it as rows-then-columns gives $(a +_{2} c) +_{1} (b +_{2} d)$ . The two readings differ by a vertex-fixing self-homeomorphism of $I^{2}$ , a filtered homotopy rel vertices, so the classes are equal in $ρ_{2}$ . Rubric: full credit for the single-square pasting and the reparametrisation homotopy.

Exercise 7 (hard, symbolic).

The connection laws include $Γ_{i}^{+} ε_{i} = ε_{i} ε_{i}$ and the transport law. Verify symbolically that $\partial_{i}^{-} Γ_{i}^{+} = id$ and $\partial_{i}^{+} Γ_{i}^{+} = ε_{i} \partial$ on the appropriate face, using $γ_{i}^{+} = max$ .

Hint

Set the $i$ -th coordinate to $0$ or $1$ inside $max (t_{i}, t_{i + 1})$ and simplify.

Answer

Write $Γ_{i}^{+} g = g \circ γ_{i}^{+}$ with $γ_{i}^{+} (\dots, t_{i}, t_{i + 1}, \dots) = (\dots, max (t_{i}, t_{i + 1}), \dots)$ . Then $\partial_{i}^{-} Γ_{i}^{+} g$ sets $t_{i} = 0$ : $max (0, t_{i + 1}) = t_{i + 1}$ , so the $i$ -coordinate becomes $t_{i + 1}$ and the result is $g$ with its arguments shifted back, i.e. $\partial_{i}^{-} Γ_{i}^{+} = id$ . Next $\partial_{i}^{+} Γ_{i}^{+} g$ sets $t_{i} = 1$ : $max (1, t_{i + 1}) = 1$ , so the folded coordinate is constant at $1$ and the cube is constant in the (former) $t_{i + 1}$ direction — this is $ε_{i}$ applied to the face $\partial_{i}^{+} g$ , i.e. $\partial_{i}^{+} Γ_{i}^{+} = ε_{i} \partial_{i}^{+}$ . These are two of the connection-face identities CG; the $min$ connection $Γ_{i}^{-}$ gives the mirror pair with $-$ and $+$ exchanged. Rubric: full credit for both $max$ evaluations and the identification of the second as a degeneracy.

Advanced results Master

The cubical $ω$ -groupoid $ρ (X_{*})$ and the crossed complex $Π X_{*}$ of 03.12.55 are two presentations of one homotopical object, and the theorem that binds them is the Brown-Higgins equivalence of categories. Let $ω -Gpd$ denote the category of cubical $ω$ -groupoids with connections, and $Crs$ the category of crossed complexes.

The equivalence $γ$ . There is an equivalence of categories $$ \gamma \colon \omega\text{-Gpd} \xrightarrow{\ \sim\ } \mathsf{Crs}, \qquad \lambda \colon \mathsf{Crs} \xrightarrow{\ \sim\ } \omega\text{-Gpd}, $$ with $γ$ and $λ$ mutually inverse up to natural isomorphism ^{[Brown-Higgins 1981 equivalence]}. The functor $γ$ extracts from a cubical $ω$ -groupoid $G$ its crossed complex of homotopy-addition data: $γ (G)_{n}$ is the subset of $G_{n}$ of cubes all of whose faces except one are degenerate (the cubes that look like an $n$ -cell with a single non-degenerate boundary face), and the boundary $δ_{n}$ is the homotopy-addition map reading the one remaining face. The action of $γ (G)_{1}$ and the crossed-module structure at $δ_{2}$ come from the compositions and connections of $G$ . The inverse $λ$ rebuilds all the cubes of $G$ from the crossed complex by a cubical analogue of the Dold-Kan correspondence, using the connections to manufacture the higher degeneracies that an ordinary cubical set lacks. Connections are essential here: without them $λ$ could not produce the folding cubes, and the equivalence fails ^{[Brown-Higgins 1981 cubes]}.

Compatibility with the geometry. Applied to a filtered space, the two functors agree: $$ \gamma,\rho(X_*) \cong \Pi X_* $$ naturally in $X_{*}$ ^{[Brown-Higgins-Sivera Part III]}. The crossed complex of a filtered space is the $γ$ -image of its cubical $ω$ -groupoid; equivalently $ρ ≅ λ Π$ . This is the precise sense in which the cube and the globe record the same data.

Why the cube proves the van Kampen theorem. The decisive use of $ρ$ is the proof of the higher homotopy van Kampen theorem 03.12.59. Suppose $X_{*}$ is covered by filtered subspaces whose interiors cover $X$ , with connected filtered intersections. To show $Π X_{*}$ is the colimit (pushout) of the $Π$ of the pieces, one argues in $ρ$ . A cube $f : I_{*}^{n} \to X_{*}$ is subdivided finely enough — by the Lebesgue covering lemma — that each small subcube lands in one of the covering subspaces. The small subcubes are elements of $ρ$ of the pieces, and the original cube is their multiple composition $\sum_{(k_{1}, \dots, k_{n})} f_{(k_{1}, \dots, k_{n})}$ under the $n$ partial compositions $+_{1}, \dots, +_{n}$ . The interchange law guarantees this multiple composition is unambiguous regardless of the order of assembly. A class in $ρ X_{*}$ is thus represented by a composite of classes coming from the pieces, which is exactly a colimit cocone; and filtered homotopies between such composites are again subdividable, giving the universal property. The colimit and subdivision steps are short and geometric on the cubical side. Transporting the conclusion across $γ$ yields the colimit statement for $Π$ in $Crs$ , where the invariant is computable. This division of labour — easy colimits in $ω -Gpd$ , computation in $Crs$ — is the engineering reason both models are kept ^{[Brown-Higgins-Sivera Part III]}.

Monoidal structure and the classifying space. The category $ω -Gpd$ carries a tensor product matching the crossed-complex tensor of 03.12.55 under $γ$ , so $γ$ is monoidal, and $ρ (X_{*} \otimes Y_{*}) ≅ ρ X_{*} \otimes ρ Y_{*}$ on CW filtrations. The cubical model makes the tensor especially transparent, since the product of cubes is a cube: $I^{m} \times I^{n} = I^{m + n}$ , so the geometric origin of the exponential law is the identity $I^{m + n} = I^{m} \times I^{n}$ rather than a formal construction. This is the cubical reason the homotopy classification of maps into a crossed complex, packaged by the internal hom $CRS$ , has a clean cubical description as well.

Synthesis. The cubical $ω$ -groupoid is the foundational reason the higher homotopy van Kampen theorem is provable at all: the central insight is that subdivision of a cube and reassembly by the $n$ partial compositions turns the colimit statement into a geometric fact about small cubes filling a covered space, and the interchange law is exactly what makes that reassembly well-defined independent of order. This is dual to the crossed-complex side, where the same homotopy type is computable but the colimit is hard to see directly; the equivalence $γ$ is the bridge that lets each model do what it does best, and putting these together gives both the proof and the computation in one package.

The connections are the technical device that makes $γ$ an equivalence rather than a mere comparison — they supply the extra degeneracies that the inverse $λ$ needs, and they generalise the single degeneracy of an ordinary cubical set to the folding operations a strict higher groupoid requires. The exponential law $ρ (X_{*} \otimes Y_{*}) ≅ ρ X_{*} \otimes ρ Y_{*}$ is the cubical face of the monoidal closed structure on $Crs$ , and it appears again in the classifying-space functor that realises a crossed complex as a space; the bridge is the elementary identity $I^{m + n} = I^{m} \times I^{n}$ , which is exactly why cubes, not simplices or disks, are the right shapes for this theory. The cubical $ω$ -groupoid is therefore not an alternative bookkeeping but the engine room: the model in which the theorems are proved before being read off in crossed-complex language.

Full proof set Master

Proposition (filtered homotopy is a congruence for every $+_{i}$ ). For each $i$ the partial composition $+_{i}$ on $R(X_) $d esce n d s t o a w e l l - d e f in e d p a r t ia l co m p os i t i o n o n$ \rho(X_*) = R(X_*)/\simeq $, w h er e$ \simeq$ is filtered homotopy rel vertices.*

Proof. Fix $i$ and suppose $a ≃ a^{'}$ , $b ≃ b^{'}$ with the face-matching conditions $\partial_{i}^{+} a = \partial_{i}^{-} b$ and $\partial_{i}^{+} a^{'} = \partial_{i}^{-} b^{'}$ holding so both composites are defined. Let $H : I^{n} \times I \to X$ be a filtered homotopy rel vertices from $a$ to $a^{'}$ , and $K$ from $b$ to $b^{'}$ . For each time $s \in I$ , the slices $H_{s} = H (-, s)$ and $K_{s} = K (-, s)$ are filtered maps. Track the matching face $\partial_{i}^{+} H_{s}$ against $\partial_{i}^{-} K_{s}$ : at $s = 0$ these are $\partial_{i}^{+} a = \partial_{i}^{-} b$ , and at $s = 1$ they are $\partial_{i}^{+} a^{'} = \partial_{i}^{-} b^{'}$ ; in between, because $H$ and $K$ are rel vertices and the shared face is an $(n - 1)$ -cube whose own vertices are fixed, the homotopies restricted to that face are filtered homotopies of $\partial_{i}^{+} H_{s}$ and $\partial_{i}^{-} K_{s}$ that we may assume coincide after composing $K$ with the invertible vertex-fixing homotopy realigning its left face to $H$ 's right face.

Granting this alignment, define $L (t, s) = (H_{s} +_{i} K_{s}) (t)$ , the $i$ -juxtaposition performed slice by slice. $L$ is continuous because $H$ and $K$ are, and the juxtaposition formula is continuous in all variables; it is filtered because each slice is a filtered map and the filtration is checked pointwise in $s$ ; and it is rel vertices because the vertices of $a +_{i} b$ come from vertices of $a$ and $b$ , all fixed by $H$ and $K$ . Thus $L : (a +_{i} b) ≃ (a^{'} +_{i} b^{'})$ is a filtered homotopy rel vertices, so $⟨ a +_{i} b ⟩ = ⟨ a^{'} +_{i} b^{'} ⟩$ in $ρ_{n}$ . The composition is therefore well-defined on classes. $□$

Proposition (associativity of $+_{i}$ in $ρ$ ). For each $i$ , $+_{i}$ is associative on $\rho(X_) $:$ (a +_i b) +_i c = a +_i (b +_i c)$ whenever the faces match.*

Proof. On the nose in $R (X_{*})$ the two triple composites are different reparametrisations of the same juxtaposition of $a, b, c$ along the $i$ -direction: $(a +_{i} b) +_{i} c$ cuts $[0, 1]$ into $[0, \frac{1}{2}], [\frac{1}{2}, \frac{3}{4}], [\frac{3}{4}, 1]$ in the $i$ -coordinate, while $a +_{i} (b +_{i} c)$ cuts it into $[0, \frac{1}{4}], [\frac{1}{4}, \frac{1}{2}], [\frac{1}{2}, 1]$ . The map from one subdivision to the other is the unique piecewise-linear order-preserving self-homeomorphism $ϕ : I \to I$ fixing $0$ and $1$ and carrying the first cut points to the second. Applying $ϕ$ only in the $i$ -coordinate is a vertex-fixing self-homeomorphism $Φ : I^{n} \to I^{n}$ , and $Φ$ is filtered (it preserves each skeleton, being a product of the identity with $ϕ$ ). Sliding the interior cut points linearly from the first triple to the second along $s \in I$ gives a filtered homotopy rel vertices from $(a +_{i} b) +_{i} c$ to $a +_{i} (b +_{i} c)$ . Hence the two are equal in $ρ_{n} (X_{*})$ . $□$

Proposition (the equivalence is compatible with the geometry). There is a natural isomorphism $\gamma,\rho(X_) \cong \Pi X_* $o f cr osse d co m pl e x es, f u n c t or ia l in t h e f i l t er e d s p a ce$ X_*$.*

Stated with proof-sketch; full proof in ^{[Brown-Higgins-Sivera Part III]} and ^{[Brown-Higgins 1981 equivalence]}. The crossed complex $γ ρ (X_{*})$ has in degree $n$ the classes of cubes in $ρ_{n} (X_{*})$ all of whose faces are degenerate except $\partial_{1}^{-}$ . Such a cube is, up to filtered homotopy rel vertices, a map of the pair $(I^{n}, \partial I^{n})$ that collapses all but one face into $X_{n - 1}$ — precisely a representative of a relative homotopy element $π_{n} (X_{n}, X_{n - 1}, *)$ . This assignment is a bijection on each degree (the cubical and the relative-homotopy descriptions of an $n$ -cell agree), it carries the homotopy-addition boundary $δ_{n}$ of $γ ρ$ to the relative-homotopy boundary $\partial$ of $Π$ , and it intertwines the $γ ρ_{1}$ -action with the $π_{1}$ -action because both are basepoint transport. The crossed-module law at $δ_{2}$ matches on both sides because the connection-and-composition derivation of the Peiffer identity in $γ ρ$ is the cubical form of the Whitehead boundary. Naturality in $X_{*}$ holds because every constituent map (faces, compositions, connections, the relative-homotopy boundary) is natural for filtered maps. Hence $γ ρ ≅ Π$ as functors $FTop \to Crs$ . $□$

Stated without proof — see Brown-Higgins-Sivera Part III §14 ^{[Brown-Higgins-Sivera Part III]}. The functors $γ : ω -Gpd \to Crs$ and $λ : Crs \to ω -Gpd$ are mutually inverse equivalences of categories, both monoidal for the respective tensor products, and the unit and counit are natural isomorphisms; $λ$ is constructed by a cubical Dold-Kan procedure that uses the connections to generate the higher degeneracies.

Connections Master

The crossed complex of a filtered space 03.12.55 is the $γ$ -image of the cubical $ω$ -groupoid built here: $γ ρ (X_{*}) ≅ Π X_{*}$ naturally, and the equivalence $γ : ω -Gpd ≃ Crs$ identifies the two models of one filtered homotopy type. The tensor product and exponential law developed in that unit's Advanced results are matched by the cubical tensor $ρ (X_{*} \otimes Y_{*}) ≅ ρ X_{*} \otimes ρ Y_{*}$ , whose geometric origin is the identity $I^{m + n} = I^{m} \times I^{n}$ . The crossed complex supplies the computability that the cube supplies the proofs for.
The filtered space 03.12.54 is the geometric input on which both $R (X_{*})$ and $ρ (X_{*})$ are constructed: the cubical singular complex is the set of filtered maps $I_{*}^{n} \to X_{*}$ , and the connectivity condition $ϕ$ established there is exactly what makes $ρ (X_{*})$ model the filtered homotopy type faithfully and makes the subdivision argument of the van Kampen proof valid. The skeletal filtration of the cube $I_{*}^{n}$ is itself an object of $FTop$ .
The 2-group / internal-category picture 03.12.56 is the bottom-degree truncation of the same data: truncating $ρ (X_{*})$ to dimensions one and two gives a double groupoid with connections, which under the Brown-Spencer correspondence is the strict 2-group / crossed module $τ_{\leq 2} Π X_{*}$ . The cubical interchange law proved here is the double-groupoid interchange that defines that structure, and the connections are what upgrade a double category to a double groupoid with the folding operations.
The higher homotopy van Kampen theorem for crossed complexes 03.12.59 is proved by the cubical subdivision-and-multiple-composition argument summarised in the Advanced results: the colimit statement is established in $ρ$ , where small cubes filling a covered space compose under $+_{1}, \dots, +_{n}$ with interchange ensuring well-definedness, and then transported across $γ$ to the computable crossed-complex statement. The cubical $ω$ -groupoid is the object on which that theorem's proof actually runs.

Historical & philosophical context Master

The cubical $ω$ -groupoid emerged from a sustained search by Ronald Brown for a higher-dimensional version of the fundamental group that would satisfy a van Kampen theorem. The decisive technical step was the recognition that cubes, unlike simplices, carry $n$ commuting partial compositions and that an extra family of degeneracies — the connections — is needed to make a cubical set into a faithful algebraic model. The connections and the algebra of multiple compositions were set out by Ronald Brown and Philip J. Higgins in On the algebra of cubes (J. Pure Appl. Algebra 21, 1981, 233-260) ^{[Brown-Higgins 1981 cubes]}, which isolated the axioms CG of a cubical $ω$ -groupoid with connections and proved that the partial compositions satisfy the interchange laws.

In the companion paper The equivalence of $\infty$ -groupoids and crossed complexes (Cahiers de Topologie et Géométrie Différentielle 22, 1981, 371-386) ^{[Brown-Higgins 1981 equivalence]}, Brown and Higgins constructed the equivalence of categories $γ$ between cubical $ω$ -groupoids with connections and crossed complexes, with inverse $λ$ built by a cubical Dold-Kan procedure. The connections are exactly the ingredient that makes $λ$ well-defined; their earlier double-groupoid work with C.B. Spencer had shown the same phenomenon in dimension two. The systematic account, including the fundamental cubical $ω$ -groupoid $ρ (X_{*})$ of a filtered space, the natural isomorphism $γ ρ ≅ Π$ , and the cubical proof of the higher homotopy van Kampen theorem, was consolidated in the 2011 monograph of Brown, Higgins, and Sivera, Nonabelian Algebraic Topology (EMS Tracts in Mathematics 15) ^{[Brown-Higgins-Sivera Part III]}, whose Part III develops the cubical theory. This places the cubical $ω$ -groupoid in the strict-higher-groupoid lineage as the computational engine of the Brown-Higgins programme, the model in which colimit and subdivision arguments are geometric before being transported to the crossed-complex side for calculation.

Bibliography Master

@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{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}
}

@article{BrownHiggins1981Equivalence,
  author  = {Brown, Ronald and Higgins, Philip J.},
  title   = {The equivalence of {$\infty$}-groupoids and crossed complexes},
  journal = {Cahiers de Topologie et G\'eom\'etrie Diff\'erentielle},
  volume  = {22},
  year    = {1981},
  pages   = {371--386}
}

@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}
}

Prerequisites

03.12.55
03.12.54

Tier anchors

beginner: Brown-Higgins-Sivera *Nonabelian Algebraic Topology* Part III informal (cubes glue along faces; connections turn a corner)
intermediate: Brown-Higgins-Sivera §13 (the cubical singular complex $R(X_*)$, the partial compositions $+_i$, the connections $\Gamma_i$, and the quotient $\rho(X_*) = R(X_*)/\simeq$); Brown-Higgins 1981 *On the algebra of cubes*
master: Brown-Higgins-Sivera Part III §13-§14 (cubical $\omega$-groupoids with connections and the equivalence $\gamma$); Brown-Higgins 1981 *On the algebra of cubes* (J. Pure Appl. Algebra 21); Brown-Higgins 1981 *The equivalence of $\infty$-groupoids and crossed complexes* (Cahiers Top. Géom. Diff. 22)

References

Brown, Higgins, Sivera — Nonabelian Algebraic Topology · Part III §13-§14 (the cubical singular complex $R(X_*)$, the cubical homotopy $\omega$-groupoid $\rho(X_*)$ with connections, and the equivalence $\gamma \colon \omega\text{-Gpd} \simeq \mathsf{Crs}$); free at https://groupoids.org.uk/pdffiles/NAT-book.pdf
Brown, Higgins — On the algebra of cubes · J. Pure Appl. Algebra 21 (1981) 233-260; cubical complexes with connections, the partial compositions and interchange laws, and the algebra of the connection operators $\Gamma_i$
Brown, Higgins — The equivalence of $\infty$-groupoids and crossed complexes · Cahiers Top. Géom. Diff. 22 (1981) 371-386; the equivalence of categories $\gamma$ between cubical $\omega$-groupoids with connections and crossed complexes, with inverse $\lambda$

Estimated time

beginner: 17m
intermediate: 45m
master: 85m