Recovering spaces from paths and all their higher homotopies

← All digests

Family 312 · Topology
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-22

1. The question

A space can be explored by recording its points, paths between points, deformations between paths, and deformations between those deformations. Continuing indefinitely produces a hierarchy of geometric information. Can this hierarchy be described algebraically without losing the space’s homotopy type?

Grothendieck’s homotopy hypothesis proposes that suitably weak infinity-groupoids do exactly this. “Weak” means that composition laws hold through specified higher homotopies, rather than through rigid equalities at every level. The challenge is to make a particular algebraic definition recover the same homotopy theory as spaces.

Family 312 claims this for every Grothendieck coherator in the Ara–Henry convention. Its central assertion is deceptively concrete: adding a new cell together with a higher cell identifying it with an old one does not change any homotopy group. The difficulty is that the algebraic construction also generates many composites and coherence operations.

2. Background and essential definitions

Two continuous maps are homotopic if one can be continuously deformed into the other. The homotopy groups $\pi_j(X,x)$ record based maps from the $j$-sphere into a space $X$, up to homotopy. The set $\pi_0(X)$ records path components. A weak homotopy equivalence induces a bijection on components and isomorphisms on all these based groups.

A groupoid is a category in which every arrow is invertible. A space gives a familiar example: points are objects and homotopy classes of paths are arrows. This fundamental groupoid captures components and fundamental groups, but it forgets higher-dimensional information. A sphere such as $S^2$, for instance, has important two-dimensional homotopy despite being simply connected.

An infinity-groupoid adds cells in all dimensions. A 1-cell joins objects; a 2-cell joins parallel 1-cells; a 3-cell joins parallel 2-cells; and so on. “Parallel” means having the same source and target. The globular organization treats each cell as a disk with a source hemisphere and a target hemisphere.

Composition introduces choices. Three paths may be composed with either parenthesization. In a weak structure, the two composites are connected by a specified 2-cell. Relations among these 2-cells require 3-cells, and the process continues.

A coherator is an algebraic theory specifying enough such operations freely. Its models are globular sets equipped with interpretations of those operations. In the convention used here, a generating operation with output dimension $j$ has a pasting diagram of dimension at most $j$. This restriction concerns the generators, not every composite operation.

Write $D^j$ for the free algebraic model on one $j$-globe and $\partial D^j$ for its boundary. A model is cellular if it is built by successively attaching these globes along boundaries, allowing a transfinite sequence. “Free” is consequential: adjoining one generator also produces all composites demanded by the theory.

3. Development history and previous partial results

Algebraic models of spaces existed well before this conjecture. Simplicial sets, for example, recover the homotopy theory of spaces through geometric realization and the singular-complex construction. Grothendieck’s proposal in Pursuing Stacks asked for a model organized directly around paths and higher invertible cells.

Maltsiniotis made the coherator formulation precise. Berger’s work organized higher operations by globular pasting diagrams, while Ara developed homotopy groups and weak-equivalence tests for Grothendieck infinity-groupoids. Bourke established structural properties of freely generated theories, including faithfulness results needed to control their stages. These developments supplied a well-defined algebraic setting.

The word “weak” cannot simply be removed. Ara showed that, under a realization preserving homotopy groups, the simply connected types represented by strict globular infinity-groupoids are precisely products of Eilenberg–Mac Lane spaces. Such a space has homotopy concentrated in one specified degree before products are taken. General spaces contain interactions that this rigid format misses.

Henry isolated a particular pushout conjecture and proved that it would imply both the canonical semi-model structure and the desired comparison with spaces. Lanari constructed substantial parts of a globular path object, but the compatible interpretation of every higher operation remained an issue. In dimension three, Lanari and then Henry–Lanari obtained the corresponding semi-model and comparison results for the relevant groupoids.

M1 presents its advance as the all-dimensional elementary-expansion theorem for an arbitrary fixed coherator in the stated convention. The final comparison uses Henry’s existing framework. It is therefore a resolution of a specific remaining algebraic obstruction, rather than the first discovery that spaces admit algebraic descriptions. These historical steps are documented in the primary references listed in Section 8 and M1.

4. What the new manuscripts claim

M1, dated 24 September 2026, fixes any such coherator $C$. Let $X$ be a cellular $C$-infinity-groupoid, and choose any $n$-cell $a$, for $n\geq0$. Form

$$X^+=X\mathbin{\amalg}_{D^n}D^{n+1},$$

where $D^n\to D^{n+1}$ selects the source face and $D^n\to X$ selects $a$. The pushout is formed inside models of the fixed theory.

Concretely, this adds a parallel cell $a'$ and a higher cell from $a$ to $a'$. The theorem claims that

$$X\longrightarrow X^+$$

is a weak equivalence: components and all based homotopy groups are unchanged. The chosen $a$ need not be a generating cell; it can already be a composite or coherence cell. When $n=0$, the operation adds an object and an arrow joining it to an existing object.

The paper then states that the category $\operatorname{Mod}(C)$ has the canonical left semi-model structure, with weak equivalences defined by those intrinsic homotopy groups. A semi-model structure provides the homotopical lifting and factorization machinery with specified cofibrancy restrictions; it should not silently be called a full model structure.

Finally, geometric realization and the fundamental-infinity-groupoid construction give a Quillen equivalence with spaces. In particular,

$$\operatorname{Mod}(C)[W^{-1}]\simeq\operatorname{Ho}(\mathrm{Spaces}),$$

where $W$ is the class of weak equivalences and the notation on the left means that those maps are formally inverted. Different permitted coherators thus describe the same homotopy theory, even though their categories of algebraic models differ.

5. Why the advance matters

Importance: 5/5, provisionally. If correct, the claim completes a central version of Grothendieck’s program connecting geometry with weak higher algebra. It establishes independence from the particular freely chosen coherences within the stated convention.

This independence is conceptually valuable. Choosing composition operations and their higher compatibility data should be a way to describe a homotopy type, rather than an extra geometric invariant. The result says that these algebraic choices disappear after passing to homotopy theory.

The elementary-expansion formulation also identifies a useful local principle. A construction that visibly adds a redundant geometric cell remains redundant after freely completing the entire algebraic structure. That principle connects manageable cell operations to a global equivalence of theories.

6. The main idea

Why a retraction is insufficient

There is an evident map back from $X^+$ to $X$: send the new cell to the old one and the connecting cell to an identity. But a retraction alone does not prove that an inclusion preserves homotopy type. A space can be a retract of a much more complicated space.

The missing statement is that all the additional composites and coherences can be contracted in a way compatible with the original theory. It is not enough to contract the newly named generators separately.

Work below a cutoff, then descend

The manuscript first replaces the model by a controlled version in which every boundary above a chosen dimension $q$ has a unique filler. Cells through dimension $q$ are retained. This supplies a top dimension at which the expansion argument becomes immediate.

At fixed $q$, the proof descends in the expansion dimension $n$. The inductive hypothesis therefore controls expansions of higher dimension. Those already controlled expansions are used to duplicate generators and promote individual homotopies into compatible statements about whole free models.

A geometric intermediate step compares a finite tree-shaped pasting diagram with a string on the same vertices. The comparison is homotopic to the identity on generating edges. Higher expansions then extend that information to all coherences, yielding contractibility of the required tree models.

Build enough paths for the current step

These tree models provide partial cylinders: an $n$-cell moves through an $(n+1)$-cell while its lower boundary stays fixed. Higher cylinders describe compatible movements along shared faces. Their representing objects are shown to be contractible using a skeletal gluing calculation.

The paper then interprets the coherator’s operations in their original free stage order. This matters because that order is not the same as ordering operations by output dimension. An operation can depend on choices made at earlier stages across several dimensions.

The resulting partial path construction has two endpoint maps with the lifting properties needed for the current expansion. A retract argument yields the weak-equivalence claim. Finally, each fixed-boundary test involves only finitely many dimensions, so a sufficiently high cutoff retains both the test and its homotopy witness. Removing the cutoff recovers the all-dimensional result.

Henry’s established comparison then converts elementary expansions into the semi-model structure and the equivalence with spaces. The manuscript’s new work is concentrated in making the expansion and coherence control available for every permitted coherator.

7. Scope, limitations, and open questions

The theorem uses the Ara–Henry coherator convention, including its bound on generating arities and its free construction. It is not a statement about every definition ever called a weak infinity-category or infinity-groupoid.

The elementary pushout theorem is stated for cellular objects; the subsequent machinery treats the relevant cofibrant retracts. One should not discard these conditions and assert an arbitrary pushout theorem for every model.

Equivalence of homotopy theories does not mean that a space and its algebraic description are literally equal. Nor does it give a short finite presentation for every homotopy type, an algorithm for computing all homotopy groups, or a strictification to strict globular groupoids.

The conclusion leaves substantial practical questions: which coherators make particular computations convenient, how efficiently geometric constructions can be translated, and how this formulation compares explicitly with other higher-categorical models. Those questions concern the use of the equivalence, beyond the existence claim reported here.

8. Sources and further reading

M1 supplies the claims and the cutoff, tree, and partial-cylinder strategy. For the precise predecessors, see Ara’s homotopy theory of Grothendieck infinity-groupoids and strict-groupoid limitation; Henry’s algebraic models and pushout conjecture; and Lanari’s globular path construction. Maltsiniotis’s coherator formulation gives further foundational context.

Source manuscripts

• M1. OpenAI, The Grothendieck homotopy hypothesis via elementary expansions (September 24, 2026).

Sources consulted on 22 October 2026 at repository revision 1581d4c. Download the manuscript citations.