nLab delooping via torsors

Redirected from "torsor delooping".

Context

Group Theory

Type theory

natural deduction metalanguage, practical foundations

  1. type formation rule
  2. term introduction rule
  3. term elimination rule
  4. computation rule

type theory (dependent, intensional, observational type theory, homotopy type theory)

syntax object language

computational trinitarianism =
propositions as types +programs as proofs +relation type theory/category theory

logicset theory (internal logic of)category theorytype theory
propositionsetobjecttype
predicatefamily of setsdisplay morphismdependent type
proofelementgeneralized elementterm/program
cut rulecomposition of classifying morphisms / pullback of display mapssubstitution
introduction rule for implicationcounit for hom-tensor adjunctionlambda
elimination rule for implicationunit for hom-tensor adjunctionapplication
cut elimination for implicationone of the zigzag identities for hom-tensor adjunctionbeta reduction
identity elimination for implicationthe other zigzag identity for hom-tensor adjunctioneta conversion
truesingletonterminal object/(-2)-truncated objecth-level 0-type/unit type
falseempty setinitial objectempty type
proposition, truth valuesubsingletonsubterminal object/(-1)-truncated objecth-proposition, mere proposition
logical conjunctioncartesian productproductproduct type
disjunctiondisjoint union (support of)coproduct ((-1)-truncation of)sum type (bracket type of)
implicationfunction set (into subsingleton)internal hom (into subterminal object)function type (into h-proposition)
negationfunction set into empty setinternal hom into initial objectfunction type into empty type
universal quantificationindexed cartesian product (of family of subsingletons)dependent product (of family of subterminal objects)dependent product type (of family of h-propositions)
existential quantificationindexed disjoint union (support of)dependent sum ((-1)-truncation of)dependent sum type (bracket type of)
logical equivalencebijection setobject of isomorphismsequivalence type
support setsupport object/(-1)-truncationpropositional truncation/bracket type
n-image of morphism into terminal object/n-truncationn-truncation modality
propositional equalitydiagonal function/diagonal subset/diagonal relationpath space objectidentity type/path type
completely presented setsetdiscrete object/0-truncated objecth-level 2-type/set/h-set
setset with equivalence relationinternal 0-groupoidBishop set/setoid with its pseudo-equivalence relation an actual equivalence relation
equivalence class/quotient setquotientquotient type
inductioncolimitinductive type, W-type, M-type
higher inductionhigher colimithigher inductive type
-0-truncated higher colimitquotient inductive type
coinductionlimitcoinductive type
presettype without identity types
set of truth valuessubobject classifiertype of propositions
domain of discourseuniverseobject classifiertype universe
modalityclosure operator, (idempotent) monadmodal type theory, monad (in computer science)
linear logic(symmetric, closed) monoidal categorylinear type theory/quantum computation
proof netstring diagramquantum circuit
(absence of) contraction rule(absence of) diagonalno-cloning theorem
synthetic mathematicsdomain specific embedded programming language

homotopy levels

semantics

Contents

Idea

In homotopy type theory, a fundamental characterization of delooping is that the delooping of a group is equivalent to the type of torsors for that group:

For GG a group internal to Set (0-truncated types) there is (up to appropriate universe-level adjustment) an equivalence

BG≃GTors L(G). \mathbf{B}G \simeq G\mathrm{Tors}_L(G).

between:

  1. the delooping groupoid (BG,pt)(\mathbf{B} G,\mathrm{pt}) (a pointed connected 1-type equipped with a group-compatible identification of its loop space/identity with GG, Id pt,ptBG≃GId_{pt,pt} \mathbf{B}G \simeq G

  2. the moduli type of left GG-torsors.

We make this explicit below.

The key observation is that for every term t:BGt \colon \mathbf{B}G, the path type

P t≔(pt=t) P_t \coloneqq (\mathrm{pt} = t)

naturally carries a GG-torsor structure, where the group acts by path concatenation. This gives a function

BG ⟶ GTors L(G) t ↦ P t. \begin{array}{ccc} \mathbf{B}G &\longrightarrow& G\mathrm{Tors}_L(G) \\ t & \mapsto& P_t \mathrlap{\,.} \end{array}

and the theorem asserts that this function exhibits an equivalence.

This characterization reveals a relationship between algebraic symmetries and identity types: the group structure of GG is represented by loops in BG\mathbf{B}G, while the points of BG\mathbf{B}G can be recovered as torsors for those symmetries.

The following sections explain the construction, its mathematical proof, and its implementation in Cubical Agda, with explicit references to the machine-checked declarations.

Construction

Left torsors

Let GG be a group whose underlying type is a set (meaning: an h-set). A left GG-action on a set XX is an operation

G×X ⟶ X (g,x) ↦ g⋅x \begin{array}{ccc} G\times X &\longrightarrow& X \\ (g,x) &\mapsto& g\cdot x \end{array}

satisfying the usual unit and associativity laws.

For x,y:Xx,y:X, define the type of transporters by

Trans G(x,y)≔∑ g:G(g⋅x=y). \mathrm{Trans}_G(x,y) \coloneqq \sum_{g:G}(g\cdot x=y).

A left GG-torsor is a left action on a set XX equipped with a term of type ‖X‖ −1\|X\|_{-1} and a proof of

∏ x:X∏ y:XisContr(Trans G(x,y)). \prod_{x:X}\prod_{y:X} \operatorname{isContr}(\mathrm{Trans}_G(x,y)).

The first condition ‖X‖ −1\|X\|_{-1} asserts mere inhabitation without selecting a distinguished element. The second says that every pair of points admits a unique transporter; one should verify that this is indeed the HoTT presentation of free and transitive.

For any chosen x:Xx:X, the orbit map

G ⟶ X g ↦ g⋅x \begin{array}{ccc} G &\longrightarrow& X \\ g &\mapsto& g\cdot x \end{array}

is an equivalence. Given y:Xy:X, consider the fiber of the orbit map over yy:

fib orb x(y) =∑ g:G(orb x(g)=y) =∑ g:G(g⋅x=y) =Trans G(x,y). \begin{aligned} \mathrm{fib}_{\mathrm{orb}_x}(y) &= \sum_{g:G}(\mathrm{orb}_x(g)=y) \\ &= \sum_{g:G}(g\cdot x=y) \\ &= \mathrm{Trans}_G(x,y). \end{aligned}

In homotopy type theory, a function is an equivalence if and only if every fiber is contractible. Therefore we have

(∏ y:XisContr(Trans G(x,y)))→isEquiv(orb x). \left( \prod_{y:X}\operatorname{isContr}(\mathrm{Trans}_G(x,y)) \right) \to \operatorname{isEquiv}(\mathrm{orb}_x).

Thus a choice of origin identifies a torsor with the group, but this identification is not part of the torsor structure.

The regular left torsor is GG itself, equipped with left multiplication.

The moduli type

Fix a univalent universe of small sets. Let

GTors L≔∑ X:hSet ℓLeftTorsorStructure G(X) G Tors_L \coloneqq \sum_{X:\mathrm{hSet}_{\ell}} \operatorname{LeftTorsorStructure}_G(X)

denote the moduli type of small left GG-torsors.

Its elements consist of carrier sets equipped with group actions and proofs of the torsor conditions. Since the relevant law fields are propositions, the structure identity principle applies.

For two torsors TT and UU, it identifies their path type with the type of equivariant equivalences:

(T=U)≃Equiv G(T,U). (T=U)\simeq\operatorname{Equiv}_G(T,U).

Hence identity paths in the moduli type encode the symmetries of torsor structures.

The regular left torsor determines a distinguished point

T reg:GTors L. T_{\mathrm{reg}}:G Tors_L.

Delooping properties

Connectedness and truncation

The moduli type GTors LG Tors_L is connected.

Indeed, any torsor TT is merely inhabited. Given a temporary choice x:Tx:T, the orbit map gives a GG-equivariant equivalence between the regular torsor and TT. The structure identity principle turns this into a path

T reg=T. T_{\mathrm{reg}}=T.

Since connectedness requires only the propositional truncation of this path, the choice of xx can be eliminated.

Moreover, GTors LG Tors_L is a 1-type. By the structure identity principle, its identity types are equivalent to types of GG-equivariant equivalences between sets, which are themselves sets.

Thus the moduli type of torsors is a pointed connected 11-type.

Recovering the group from loops

At the regular basepoint, the loop type is

Ω(GTors L,T reg)≃Aut G(T reg). \Omega(G Tors_L,T_{\mathrm{reg}}) \simeq \operatorname{Aut}_G(T_{\mathrm{reg}}).

An equivariant automorphism f:G→Gf:G\to G of the regular left torsor satisfies

f(gx)=gf(x). f(gx)=gf(x).

Taking x=1x=1 gives

f(g)=gf(1). f(g)=gf(1).

Consequently, every such automorphism is a right translation

R h(x)≔xh R_h(x)\coloneqq xh

for a unique h:Gh:G. Evaluation at the group unit therefore gives an equivalence

Aut G(T reg)≃G. \operatorname{Aut}_G(T_{\mathrm{reg}})\simeq G.

Combining these equivalences yields

Ω(GTors L,T reg)≃G. \Omega(G Tors_L,T_{\mathrm{reg}}) \simeq G.

This is an equivalence of underlying types. Compatibility with the group structure requires additional coherence between composition of equivariant automorphisms and path concatenation.

In particular, if p(g)p(g) is the loop corresponding to the right translation R gR_g under the structure identity principle, the composition convention used here gives

p(gh)=p(g)⋅p(h), p(gh)=p(g)\cdot p(h),

where the right-hand operation is path concatenation.

This explains how the multiplication of GG is represented by composition of identity paths in the moduli type of torsors.

Path-torsor characterization

The central result of the torsor delooping construction is that any delooping of a group is equivalent to the type of torsors for that group.

This is the content of Myers, Sati & Schreiber 2024, Lem 5.8.

Statement

Lemma (Any delooping is equivalent to the type of torsors).

Let GG be a group, and let (BG,pt)(\mathbf{B}G,\mathrm{pt}) be a delooping of GG: a pointed connected 11-type equipped with a looping equivalence

ℓ:Ω ptBG≃G \ell:\Omega_{\mathrm{pt}}\mathbf{B}G\simeq G

compatible with the group structures.

Then the assignment

BG →≃ GTors L t ↦ Id BG(pt,t) \begin{array}{ccc} \mathbf{B}G &\xrightarrow{\simeq}& G Tors_L \\ t &\mapsto& \operatorname{Id}_{\mathbf{B}G}(\mathrm{pt},t) \end{array}

defines an equivalence between BG\mathbf{B}G and the type of left GG-torsors.

The group action on the identity type is induced by concatenation of identifications, using the inverse of the given looping equivalence.

Construction of the group action

Fix t:BGt:\mathbf{B}G and define the path torsor

P t≔Id BG(pt,t). P_t\coloneqq\operatorname{Id}_{\mathbf{B}G}(\mathrm{pt},t).

For g:Gg:G, the inverse looping equivalence determines a loop ℓ −1(g):Id BG(pt,pt)\ell^{-1}(g):\operatorname{Id}_{\mathbf{B}G}(\mathrm{pt},\mathrm{pt}).

Concatenation with this loop defines a left GG-action:

G×P t ⟶ P t (g,p) ↦ ℓ −1(g)⋅p \begin{array}{ccc} G\times P_t &\longrightarrow& P_t \\ (g,p) &\mapsto& \ell^{-1}(g)\cdot p \end{array}

Here the operation on the right is path concatenation. The compatibility of the looping equivalence with the group structure ensures the unit and associativity laws.

This is the construction described in equation (160) of Myers, Sati & Schreiber 2024.

Proof that the path type is a torsor

We first verify that P tP_t is a left GG-torsor.

Mere inhabitation.

Since BG\mathbf{B}G is connected, for every t:BGt:\mathbf{B}G there merely exists an identification from pt\mathrm{pt} to tt, expressed by ‖P t‖ −1\|P_t\|_{-1}.

This establishes mere inhabitation without choosing a distinguished path.

Existence of a transporter.

Let p,q:Id BG(pt,t)p,q:\operatorname{Id}_{\mathbf{B}G}(\mathrm{pt},t). We wish to construct a group element g:Gg:G satisfying g⋅q=pg\cdot q=p.

By path inversion and concatenation of identifications, the composite p⋅q −1p\cdot q^{-1} is a loop at pt\mathrm{pt}. Applying the looping equivalence gives

g≔ℓ(p⋅q −1):G. g\coloneqq\ell(p\cdot q^{-1}):G.

By construction, ℓ −1(g)⋅q=p\ell^{-1}(g)\cdot q=p, up to the canonical identification supplied by the equivalence and the groupoid laws for paths. Hence gg transports qq to pp.

This corresponds to the explicit construction in equation (161) of Myers, Sati & Schreiber 2024.

Uniqueness of the transporter.

Suppose another group element g′:Gg':G also satisfies g′⋅q=pg'\cdot q=p. By definition of the action, we have ℓ −1(g)⋅q=p=ℓ −1(g′)⋅q\ell^{-1}(g)\cdot q=p=\ell^{-1}(g')\cdot q.

Concatenating with q −1q^{-1} and using the groupoid laws for paths gives ℓ −1(g)=ℓ −1(g′)\ell^{-1}(g)=\ell^{-1}(g'). Since ℓ\ell is an equivalence, it follows that g=g′g=g'.

More precisely, the type of transporters is

Trans G(q,p)≔∑ g:G(ℓ −1(g)⋅q=p). \operatorname{Trans}_G(q,p) \coloneqq \sum_{g:G}(\ell^{-1}(g)\cdot q=p).

This is the fiber over pp of the orbit map g↦ℓ −1(g)⋅qg\mapsto\ell^{-1}(g)\cdot q. Since this map is the composite of the looping equivalence ℓ −1\ell^{-1} with concatenation by qq, both of which are equivalences, its fibers are contractible.

Thus P tP_t satisfies the unique-transporter condition and, together with mere inhabitation, is a left GG-torsor.

This is the uniqueness argument corresponding to equation (162) of Myers, Sati & Schreiber 2024.

Proof that the construction is an equivalence

We have shown that the assignment t↦P tt\mapsto P_t defines a map F:BG→GTors LF:\mathbf{B}G\to G Tors_L. It remains to prove that FF is an equivalence.

The source BG\mathbf{B}G is a pointed connected 11-type by assumption. As established above, the moduli type GTors LG Tors_L is also a pointed connected 11-type, with the regular torsor as its distinguished point.

At t=ptt=\mathrm{pt}, the associated torsor is P pt=Id BG(pt,pt)P_{\mathrm{pt}}=\operatorname{Id}_{\mathbf{B}G}(\mathrm{pt},\mathrm{pt}). The looping equivalence identifies this torsor with the regular left GG-torsor, giving the required basepoint identification.

Moreover, a path r:s=tr:s=t induces a GG-equivariant equivalence by concatenation:

P s →≃ P t q ↦ q⋅r \begin{array}{ccc} P_s &\xrightarrow{\simeq}& P_t \\ q &\mapsto& q\cdot r \end{array}

Its inverse is given by concatenation with r −1r^{-1}. Equivariance follows from associativity of path concatenation.

By the structure identity principle, this equivariant equivalence corresponds to a path F(s)=F(t)F(s)=F(t) in the moduli type.

At the basepoint, this construction induces an equivalence on loop types. Indeed, equivariant automorphisms of P ptP_{\mathrm{pt}} are determined by their value at the identity path, and the induced action on loops agrees with this identification.

We now apply the fundamental lemma for pointed connected types: a pointed map between connected types that induces an equivalence on loop types is itself an equivalence, in the present 11-truncated setting.

Consequently,

BG≃GTors L. \boxed{\mathbf{B}G\simeq G\mathrm{Tors}_L.}

This completes the proof of the torsor characterization of deloopings.

Formalization in Cubical Agda

The constructions described above have been formalized in Cubical Agda, using existing library interfaces for univalence, the structure identity principle, and Eilenberg–Mac Lane types.

The verified development consists of twelve modules, all checked with Agda’s --safe option. The source code and mathematical exposition are available in the nLab materials index:

The following declarations are local to this development; the underlying Cubical library provides the general-purpose type-theoretic interfaces.

Torsor structures

  • LeftAction.agda defines LeftAction and Equivariant.

  • LeftTorsor.agda defines IsLeftTorsor and LeftTorsor, using mere inhabitation and contractible transporters.

  • Regular.agda constructs the regular left torsor regularLeftTorsor.

  • Morphisms.agda establishes that equivariant maps between left torsors are equivalences.

Moduli and loops

Comparison and classification

The last module introduces PointedLoopGroupData, consisting of connectedness, 11-truncation, a loop equivalence, and explicit unit and multiplication coherences.

It constructs pathTorsorAction, pathTorsorOrbitEquiv, pathTorsor, and pathTorsorModuli, together with the basepoint comparison and the equivalence induced on loop types.

Its main result, liftedPathTorsorModuliEquiv, establishes the universe-lifted equivalence

Lift ℓ+1(BG)≃GTors L. \operatorname{Lift}_{\ell+1}(\mathbf{B}G) \simeq G Tors_L.

Here the formalization uses a chosen loop equivalence G≃(pt=pt)G\simeq(\mathrm{pt}=\mathrm{pt}), inverse in direction to the equivalence used in the mathematical exposition above.

The universe lifting accounts for the universe level of the torsor moduli type. The final declaration is an equivalence of underlying types, with basepoint comparison paths established separately; no additional pointed-equivalence or group-isomorphism package is claimed.

All twelve modules pass Agda’s --safe type-checking mode with the pinned Cubical library.

The development concerns set-level groups and their homotopy-theoretic deloopings. The construction of particular groups, such as the multiplicative group of invertible complex numbers, and their subsequent applications to topological quantum gates remain future work.

For the complete mathematical explanation, statement-to-declaration index, and reproducibility information, see the accompanying mathematical companion paper.

References

Last revised on November 15, 2026 at 11:44:36. See the history of this page for a list of all contributions to it.