Introduction
A singular foliation and its coefficients
Let
M = R 2 , F = ⟨ x ∂ x , y ∂ x ⟩ . M = \mathbb{R}^{2}, \qquad\mathcal{F} = \langle x\partial_x,y\partial_x\rangle. M = R 2 , F = ⟨ x ∂ x , y ∂ x ⟩ .
The foliation has rank one away from the origin and rank zero at the origin. Its module of generators is not locally free, but has the global resolution
0 ⟶ C ∞ ( M ) → ( − y , x ) C ∞ ( M ) 2 → ( x ∂ x , y ∂ x ) F ( M ) ⟶ 0. 0 \longrightarrow C^{\infty}(M) \xrightarrow{(-y,x)} C^{\infty}(M)^{2} \xrightarrow{(x\partial_x,y\partial_x)} \mathcal{F}(M) \longrightarrow0. 0 ⟶ C ∞ ( M ) ( − y , x ) C ∞ ( M ) 2 ( x ∂ x , y ∂ x ) F ( M ) ⟶ 0.
The associated geometric Lie 2-algebroid E a f f E_{\mathrm{aff}} E aff has CE algebra
B = C ∞ ( M ) [ θ 1 , θ 2 , η ] , ∣ θ i ∣ = 1 , ∣ η ∣ = 2 , B = C^{\infty}(M)[\theta_{1},\theta_{2},\eta], \qquad|\theta_i| = 1,\qquad|\eta| = 2, B = C ∞ ( M ) [ θ 1 , θ 2 , η ] , ∣ θ i ∣ = 1 , ∣ η ∣ = 2 ,
with differential
Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Qf = f_x(x\theta_1+y\theta_2), \qquad Q\theta_1 = -y\eta, Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η ,
Q θ 2 = θ 1 θ 2 + x η , Q η = 0. (1) Q\theta_2 = \theta_1\theta_2+x\eta,\qquad Q\eta= 0. \tag*{(1)} Q θ 2 = θ 1 θ 2 + x η , Q η = 0. (1)
The degree-two generator records a relation between vector-field generators. At the origin it contributes genuine higher isotropy. This paper proves coefficient Riemann–Hilbert on every ordinary open O ⊆ M O\subseteq M O ⊆ M , including opens containing the origin. The coefficient bodies are globally bounded and degreewise finite projective over C ∞ ( O ) C^{\infty}(O) C ∞ ( O ) ; their cohomology ranks may jump. We then determine how higher isotropy is retained by integration and which regular two-term coefficients can pass to ordinary quotients.
Geometric resolutions and their universal Lie infinity-algebroid structures provide a general way to organize such relations [5 ] . Their homotopy theory is not the same question as constructing the ordinary holonomy groupoid [AS09]. Here we retain full Maurer–Cartan simplices and work on a fixed smooth base. For a geometric algebroid E E E , full smooth integration is
Int ( E ) n ( U ) = Map N Q ( U × T [ 1 ] Δ n , E [ 1 ] ) , (2) \operatorname{Int}(E)_n(U) = \operatorname{Map}_{NQ}(U\times T[1]\Delta^n,E[1]), \tag*{(2)} Int ( E ) n ( U ) = Map N Q ( U × T [ 1 ] Δ n , E [ 1 ]) , (2)
with differential in the simplex direction and joint smoothness through all faces. A simplex includes its base map and anchor equations, not just pointwise Lie-algebra-valued forms.
The finite model and the Riemann–Hilbert theorem
Let G = Aff + ( 1 ) G=\operatorname{Aff}_{+}(1) G = Aff + ( 1 ) act by
( a , b ) ( x , y ) = ( a x + b y , y ) . (a,b)(x,y)=(ax+by,y). ( a , b ) ( x , y ) = ( a x + b y , y ) .
The crossed module defined in Section 3 has a strict Lie 2-groupoid nerve
K n = M × G n × R ( 2 ) . K_n=M\times G^n\times\mathbb{R}^{(2)}. K n = M × G n × R ( 2 ) .
This model appeared in the companion transport construction [13 ] . We recall its crossed-module structure and horn properties. The comparison with full MC integration and the coefficient results are the subject of the present paper. For an open O ⊆ M O\subseteq M O ⊆ M , put E O = E a f f ∣ O E_O=E_{\mathrm{aff}}|_O E O = E aff ∣ O and B O = B E O B_O=B_{E_O} B O = B E O . The appropriate finite-model restriction is
K [ O ] n = { k ∈ K n : [ min i x i , max i x i ] × { y } ⊂ O } , (3) K[O]_n=\left\{k\in K_n:\left[\min_i x_i,\max_i x_i\right]\times\{y\}\subset O\right\}, \tag*{(3)} K [ O ] n = { k ∈ K n : [ i min x i , i max x i ] × { y } ⊂ O } , (3)
where ( x i , y ) (x_i,y) ( x i , y ) are its vertices. The horizontal-hull condition keeps the vertices in a single horizontal component of O O O ; vertex membership alone does not ensure this. Likewise Int ( E O ) \operatorname{Int}(E_O) Int ( E O ) uses full MC simplices whose entire base image lies in O O O .
Here y y y denotes the represented smooth sheaf, and c ( y O ) c(yO) c ( y O ) is the constant simplicial subsheaf of unit-generated simplices. Write Coh ( B O ) \operatorname{Coh}(B_O) Coh ( B O ) for cohesive coefficients with real, globally bounded, degreewise finite-projective bodies over C ∞ ( O ) C^\infty(O) C ∞ ( O ) . Regularity means locally constant ranks of internal cohomology. The representation notation refers to the concrete normalized smooth cochain category of § 4 , with the same coefficient-body conditions. Throughout, representations retain their specified vertex complexes and normalized cochain operations.
Theorem I (Ordinary-open full affine comparison). For every ordinary open O ⊆ R 2 O \subseteq\mathbb{R}^2 O ⊆ R 2 , K [ O ] K[O] K [ O ] is a geometric Lie 2-groupoid and there is an actual vertex-preserving comparison
Φ O : Int ( E O ) ⟶ y K [ O ] . \Phi_O:\operatorname{Int}(E_O)\longrightarrow yK[O]. Φ O : Int ( E O ) ⟶ y K [ O ] . It is a trivial Kan fibration on every smooth parameter manifold. It admits a smooth simplicial section Ψ O \Psi_O Ψ O , with Φ O Ψ O = 1 \Phi_O\Psi_O=1 Φ O Ψ O = 1 , and a homotopy from Ψ O Φ O \Psi_O\Phi_O Ψ O Φ O to the identity, over y K [ O ] yK[O] y K [ O ] and fixed on the section image. In particular the homotopy is relative to the whole c ( y O ) c(yO) c ( y O ) . Actual pullback gives a dg quasi-equivalence
Φ O ∗ : Rep s m p e r f ( y K [ O ] ) ⟶ Rep s m p e r f ( Int ( E O ) ) . \Phi_O^*:\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(yK[O])\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(\operatorname{Int}(E_O)). Φ O ∗ : Rep sm perf ( y K [ O ]) ⟶ Rep sm perf ( Int ( E O )) . This also restricts to the regular full subcategories.
The map records edge holonomies and h i j k = − ∫ [ i j k ] β h_{ijk}=-\int_{[ijk]}\beta h ij k = − ∫ [ ij k ] β , where β \beta β is the degree-two MC field. Boundary extension, a Whitney–radial identity and a rank-uniform flat correction prove the comparison. The smooth simplicial retraction supplies more than an abstract weak equivalence: it justifies a change of the actual raw coefficient target. For a general O O O , smooth charts on the convex admissible arrow domains keep the whole base simplex in O O O . This range-preserving construction is distinct from the whole-plane section. It neither divides by a rank coordinate nor removes the origin. For different opens the sections are chosen independently; strict restriction-naturality is an additional compatibility condition.
Theorem II (Ordinary-open coefficient RH). For every ordinary open O ⊆ R 2 O \subseteq\mathbb{R}^2 O ⊆ R 2 , the actual functors
I O = Ψ O ∗ I N E O : Coh ( B O ) ⟶ Rep s m p e r f ( y K [ O ] ) , I_O=\Psi_O^*I_{NE_O}:\operatorname{Coh}(B_O)\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(yK[O]), I O = Ψ O ∗ I N E O : Coh ( B O ) ⟶ Rep sm perf ( y K [ O ]) , I N E O : Coh ( B O ) ⟶ Rep s m p e r f ( Int ( E O ) ) I_{NE_O}:\operatorname{Coh}(B_O)\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(\operatorname{Int}(E_O)) I N E O : Coh ( B O ) ⟶ Rep sm perf ( Int ( E O )) are smooth strictly unital A ∞ A_\infty A ∞ quasi-equivalences, without a regularity assumption on internal cohomology. Every integer Hom degree is covered. On the finite model, actual differentiation F F F supplies a preimage F R FR F R on the same body as R R R and an actual closed invertible morphism
q R : I O ( F R ) ⟶ R , ( q R ) 0 = 1 , F ( q R ) = 1. q_R:I_O(FR)\longrightarrow R,\qquad(q_R)_0=1,\qquad F(q_R)=1. q R : I O ( F R ) ⟶ R , ( q R ) 0 = 1 , F ( q R ) = 1. The full-MC comparison likewise has actual invertible identity-body counits. The preimages and counits are chosen objectwise.
The proof works directly on the body complex, including where its cohomology ranks jump. Strict differentiation is constructed on the actual normalized cochains and on globally finite-projective bundles. After a closed rectification of edge transport, two prescribed upper-decalage Hom models give a comparison T T T whose internal diagonal is the identity. If [ ℓ , u ] [\ell,u] [ ℓ , u ] bounds the internal Hom degrees, then
T = 1 + N , N u − ℓ + 1 = 0 , T − 1 = ∑ j = 0 u − ℓ ( − N ) j . T=1+N,\qquad N^{u-\ell+1}=0,\qquad T^{-1}=\sum_{j=0}^{u-\ell}(-N)^j. T = 1 + N , N u − ℓ + 1 = 0 , T − 1 = j = 0 ∑ u − ℓ ( − N ) j .
A weighted normal-ideal estimate and an all-degree scalar calibration prove the diagonal statement, including every power of the even generator η \eta η . This is a real-linear chain inverse. Internal degree is not a filtration by subcomplexes when the body differential is nonzero. The inverse is constructed in real vector spaces, and the two Hom differentials are related by conjugation.
Actual rectification extends the Hom comparison to all pairs. Global graded surjectivity then gives acyclicity of the actual smooth differentiation kernels. Solving one equation in that kernel constructs q R q_R q R ; its inverse is finite because its positive-arity correction lowers internal degree. The whole-unit-relative prism passes the conclusion to full MC. Infinitely flat and nongeneric rank-jumping coefficients make the absence of a cohomology-regularity hypothesis concrete.
The whole-plane scalar formula H ∙ ( B ) = C ∞ ( R y ) [ η ] / ( y η ) H^\bullet(B)=C^\infty(\mathbb{R}_y)[\eta]/(y\eta) H ∙ ( B ) = C ∞ ( R y ) [ η ] / ( y η ) does not extend by restriction to arbitrary nonrectangular opens. Regular minimalization and finite MC lifting instead provide the tools for the regular classification and vertical-family results; they are distinct from the direct body-complex argument of Theorem II.
A concrete higher-isotropy application
The two-term coefficient sector makes the significance of the higher class explicit. Here “two-term” means internal cohomology in degrees 0, 1, allowing acyclic terms in the original complex.
Theorem III (Two-term coefficients and ordinary quotients). The H 0 H^0 H 0 -isomorphism classes of regular two-term cohesive coefficients, and of their corresponding raw representations on y K yK y K and full MC, are classified by
( d 0 , d 1 , r ) , 0 ≤ r ≤ min ( d 0 , d 1 ) , (d_0,d_1,r),\qquad0\le r\le\min(d_0,d_1), ( d 0 , d 1 , r ) , 0 ≤ r ≤ min ( d 0 , d 1 ) , where d i d_i d i are the internal cohomology dimensions and r r r is the rank of a residue map N 0 : V 1 → V 0 N_0:V^1\to V^0 N 0 : V 1 → V 0 . A representative has D ( v 1 ) = N 0 ( v 1 ) η D(v_1)=N_0(v_1)\eta D ( v 1 ) = N 0 ( v 1 ) η and integrates on K K K to r 2 = − h N 0 r_2=-hN_0 r 2 = − h N 0 .
There is an ordinary groupoid G \mathcal{G} G in smooth sheaves, with objects y M yM y M , and a strict map q : y K → N ( G ) q:yK\to N(\mathcal{G}) q : y K → N ( G ) fixing objects. Such a representation descends along q q q exactly when r = 0 r=0 r = 0 . For r > 0 r>0 r > 0 it cannot be a pullback, up to H 0 H^0 H 0 -isomorphism, through any strict map to an ordinary groupoid nerve with represented objects, using the bounded finite-projective raw coefficient theory. The same obstruction holds on full MC.
Section 12 proves the normal forms by a global smooth gauge and an exact residue calculation. The residue matrix is considered up to graded frame changes, not as a canonical numeric matrix. The obstruction restricts to the exact B 2 Z B^2\mathbb{Z} B 2 Z at the origin. Ordinary groupoid representations can have higher components, but normalization makes their pullbacks constant on this particular simplicial subobject. The quotient is constructed in smooth sheaves, without a representability hypothesis on its arrow sheaf. Comparison with Androulidakis–Skandalis (AS) holonomy or refined higher holonomy requires additional geometric input. The obstruction allows nonregular and nonminimal coefficients on the ordinary target.
Vertical families and nontrivial coefficient bundles
We also study regular coefficients for the vertical algebroid E S E_S E S on S × R 2 S \times\mathbb{R}^2 S × R 2 , for every finite-dimensional Hausdorff second-countable smooth manifold S S S , including noncompact and disconnected bases. Its anchor has no S S S -component: this is not an inverse-image algebroid adding T S TS T S . Theorem 13.4 proves actual regular RH for globally bounded finite-projective coefficient bundles, without requiring those bundles to be trivial. Here regularity is a standing hypothesis on the fiber cohomology of the body complex over S × R 2 S \times\mathbb{R}^2 S × R 2 . Smooth parameter dependence of the simplices in Theorem I and regularity of a coefficient bundle are distinct conditions. Corollary 13.5 gives compatibility with all smooth parameter pullbacks and a formal equivalence of the corresponding C a t ∞ \mathrm{Cat}_{\infty} Cat ∞ -valued parameter diagrams.
This requires more than tensoring the scalar comparison with an arbitrary function space. A normalized global projector dictionary first makes raw finite perturbation valid for nontrivial vertex bundles. The reduction then produces a bundle over S S S , and finite-projective layers over C ∞ ( S ) C^{\infty}(S) C ∞ ( S ) permit global Hom comparison and MC lifting. The actual counit is given in (13.11) and is chosen objectwise.
The two-term family invariant is a bundle arrow N : H 1 → H 0 N : H^1 \to H^0 N : H 1 → H 0 over S S S , up to bundle isomorphism (Theorem 13.6 ). The Möbius-line example detects coefficient topology, while t t t and t 3 t^3 t 3 detect smooth vanishing order despite identical rank profiles. The rank of N N N may vary. Corollary 13.9 proves the exact ordinary-descent criterion N ≡ 0 N \equiv0 N ≡ 0 . The family theorem varies the parameter space and coefficient bundles while keeping the affine algebroid fixed.
Relation to earlier work and organization
The general fixed-base theory is developed in the companion Global L ∞ L_{\infty} L ∞ -Algebroid Resolutions: Homotopy Invariance and Smooth Coefficient Integration [12 ] . We use its pointed polynomial comparison, coherent global-resolution results, coefficient invariance and smooth coefficient-integration construction. Section 2 records the hypotheses needed here. The global-resolution machinery compares other admissible global resolutions with the selected affine model. The explicit RH proof for that model is given here independently of those comparisons.
Nuiten supplies the fixed-base semi-model and polynomial mapping-space inputs [7 ] . For general weak maps, we use the connective mapping-space comparison. Its connective form is essential at the singular foliation, where the anchor is not transitive. Chuang–Holstein–Lazarev establish the general smooth and normalized-cochain homotopy-invariance principles [3 Theorem 4.2, Corollary 3.1 and Remark 4.1, Theorem 5.2, Proposition 6.2 and Corollary 6.1] . The companion gives the precise jointly smooth, finite-projective and vertex-relative forms used here.
The universal integration kernel is due to Arias Abad–Schätz [AS13]. The smooth strictly unital functor used here is constructed in [12 Theorem 9.2] . Section 6 retains its units, vertex behavior, matrix signs, actual Taylor formulas and fixed-base naturality. The affine task is to prove that this functor is a quasi-equivalence and to compute its residue-sensitive representations.
An actual higher Riemann–Hilbert correspondence for full de Rham geometry on compact closed manifolds was established by Block–Smith [BS14, Theorem 4.1, p. 16]. Ordinary-groupoid loss of higher coefficient information also has a direct antecedent: Arias Abad–Schätz give a two-term T S 2 TS^2 T S 2 coefficient whose integration does not descend to the pair groupoid [AS13, Section 5.6 and Proposition 5.4, pp. 39–41]. These references use the preprint versions specified in the bibliography. Our contribution here is the actual-map reconstruction for this singular affine model, its explicit residue and bundle-arrow classifications, and the obstruction for strict maps to ordinary groupoid nerves with represented objects.
A complementary regular-foliation correspondence is developed in [10 ] . On a compact base it compares globally bounded finite-rank regular cohesive modules with global plot-smooth objects of the same regularity, using sheafified raw Hom presheaves. The affine equivalence here instead uses actual raw Hom complexes and permits cohomology ranks to jump. These different targets must be kept distinct when comparing the respective descent statements.
Higher Van Est theory provides an important antecedent [AC25; CH26]. The last-horn criterion of [2 Lemma 4.3.3 and Theorem 4.3.4, pp. 28–29] applies to this K K K in every scalar degree. That theorem concerns a differentiation map into its intrinsic CE quotient. Here we identify the affine quotient, construct actual coefficient differentiation, and calibrate the prescribed relative comparison in every internal degree. The reverse functor is the specified Ψ I N ∗ \Psi^*_{\mathrm{IN}} Ψ IN ∗ , rather than an arbitrarily chosen inverse to a scalar Van Est map.
After the conventions and cited interfaces, § 3 introduces the affine crossed module and the full-MC comparison map; its longer global lifting and retraction proofs are in Appendix § 15 . § 4 –§ 6 fix the raw coefficient target, prove its actual affine pullback comparison, and record the integration formulas. § 7 constructs the range-preserving ordinary-open comparison. The central argument in § 8 –§ 10 is strict differentiation, the prescribed relative Hom comparison, and Theorem 10.4 . § 11 illustrates its new scope with rank-jumping coefficients and a nonrectangular open. § 12 gives the point-parameter residue classification and ordinary non-descent theorem. § 13 proves the vertical-family extension and its bundle-valued examples. § 14 formulates questions for further work. Appendices B and C retain the regular reduction, scalar-parameter and finite MC-lifting tools used by the classification and family arguments, separately from the nonregular proof.
Conventions and comparison results
Manifolds are finite-dimensional, Hausdorff and second-countable. Write A = C ∞ ( M ) A = C^\infty(M) A = C ∞ ( M ) . Algebroid complexes are homological, with ∂ : E i → E i − 1 \partial: E_i \to E_{i-1} ∂ : E i → E i − 1 , and we use Nuiten’s convention l 1 = − ∂ l_1 = -\partial l 1 = − ∂ , ∣ l n ∣ = n − 2 |l_n| = n - 2 ∣ l n ∣ = n − 2 . The geometric CE generator dual to E i E_i E i has cohomological degree i + 1 i + 1 i + 1 . Coefficient complexes are cohomological.
Geometric objects and supplied global resolutions
Let G M \mathcal{G}_M G M be the category of split geometric L ∞ L_\infty L ∞ -algebroids over M M M whose homological section complexes are connective and degreewise finitely generated projective over A A A . Such complexes are underlying-A A A -cofibrant. Morphisms are fixed-base nonlinear infinity-morphisms, and W 1 W_1 W 1 consists of those whose first Taylor component is a quasi-isomorphism. Their Taylor components factor A A A -linearly through the corresponding symmetric powers and satisfy the anchor condition separately. The CE differential and anchored brackets obey their respective Leibniz rules over A A A . The geometric dictionary and underlying-cofibrancy statement are given in [12 Section 3, Lemmas 3.1 and 3.6] .
For E ∈ G M E \in\mathcal{G}_M E ∈ G M , write B E B_E B E for the graded algebra of global geometric CE sections with its differential Q Q Q . Positive grading forces Q f Qf Q f for f ∈ A f \in A f ∈ A to be linear in the degree-one generators. Thus the ordinary-body anchor is fiber-linear, also at singular rank. Infinity-morphisms may nevertheless be nonlinear in positive coordinates. Throughout, the base is an ordinary smooth manifold.
Write T A = Der R ( A ) T_A = \operatorname{Der}_{\mathbb{R}}(A) T A = Der R ( A ) . A geometric resolution of a foliation module f ↪ T A \mathfrak{f} \hookrightarrow T_A f ↪ T A is a strict augmentation p E : E ⟶ ∼ f p_E : E \overset{\sim}{\longrightarrow} \mathfrak{f} p E : E ⟶ ∼ f from such an E E E whose underlying global section complex resolves f \mathfrak{f} f in degree zero. Injectivity here is injectivity of the module anchor; it need not survive passage to a residue-field fiber. All resolution-comparison statements concern global resolutions with these hypotheses. The module f \mathfrak{f} f may be nonprojective. Global finite projectivity of the geometric terms is an explicit hypothesis; local finite rank or exactness of residue-field fibers is not a substitute. Resolution comparisons take place among these supplied global objects.
Smooth sheaves and full Maurer–Cartan simplices
Use the ordinary site of Euclidean opens with smooth maps and open covers, including the empty cover of the empty object. Write E \mathcal{E} E for its sheaf topos and y M ( U ) = C ∞ ( U , M ) yM(U) = C^\infty(U,M) y M ( U ) = C ∞ ( U , M ) . Euclidean opens are dense, so these sheaves can also be evaluated on smooth manifolds. Simplicial smooth sheaves are objects of s E s\mathcal{E} s E . Their tensor with a simplicial set is the sheafified copower; simplex labels may be locally constant on disconnected tests. A finite simplicial set has finitely many nondegenerate simplices, and Λ 0 [ 1 ] = { 0 } \Lambda^0[1] = \{0\} Λ 0 [ 1 ] = { 0 } .
An N Q NQ N Q -map is a map of the graded manifolds that commutes with their cohomological vector fields. Full smooth integration is
Int ( E ) n ( U ) = Map N Q ( U × T [ 1 ] Δ n , E [ 1 ] ) . \operatorname{Int}(E)_n(U) = \operatorname{Map}_{NQ}\left(U \times T[1]\Delta^n, E[1]\right). Int ( E ) n ( U ) = Map N Q ( U × T [ 1 ] Δ n , E [ 1 ] ) .
The differential acts only in the simplex direction. For an infinite positive grading this means degreewise geometric CE maps. A simplex contains its entire base map, anchor equation and higher flatness equations. All data are jointly smooth through the faces, not an algebraic tensor with C ∞ ( U ) C^\infty(U) C ∞ ( U ) . Gluing in smooth parameters for a fixed algebroid gives a simplicial sheaf [12 Lemma 4.1] .
For S 0 = y M S_0 = yM S 0 = y M , write c ( y M ) ⊆ S c(yM) \subseteq S c ( y M ) ⊆ S for the whole unit-generated subsheaf of totally degenerate simplices. A homotopy relative to c ( y M ) c(yM) c ( y M ) is constant on c ( y M ) × Δ [ 1 ] c(yM) \times\Delta[1] c ( y M ) × Δ [ 1 ] , not merely at vertex endpoints. For integration these are constant base maps with zero positive-degree fields.
Write ∣ S ∣ |S| ∣ S ∣ for open-cover infinity-sheafification of the objectwise realization, and Mon s h ( E ) = ∣ Int ( E ) ∣ \operatorname{Mon}^{\mathrm{sh}}(E)=|\operatorname{Int}(E)| Mon sh ( E ) = ∣ Int ( E ) ∣ . We also consider the hypercompletion of this smooth infinity-sheaf. The associated derived fixed-object Čech groupoid has levels y M × ∣ S ∣ ⋯ × ∣ S ∣ y M yM\times_{|S|}\cdots\times_{|S|}yM y M × ∣ S ∣ ⋯ × ∣ S ∣ y M . The raw coefficient category of § 4 is defined on S S S itself, with its specified vertices and simplices.
Cohesive coefficients
Let B = ⨁ r ≥ 0 B r \mathcal{B}=\bigoplus_{r\geq0}\mathcal{B}^r B = ⨁ r ≥ 0 B r be a sheaf of commutative cochain dg real algebras with B 0 = C M ∞ \mathcal{B}^0=C^\infty_M B 0 = C M ∞ . Each B r \mathcal{B}^r B r is a finite-rank bundle sheaf, multiplication is smooth, and d B d_{\mathcal{B}} d B is a local differential operator commuting with differentiation and compact integration in external smooth parameters. The geometric CE sheaves above satisfy these hypotheses. Put B = ⨁ r Γ ( M , B r ) B=\bigoplus_r\Gamma(M,\mathcal{B}^r) B = ⨁ r Γ ( M , B r ) .
Following [Blo10], an object of Coh ( B ) \operatorname{Coh}(B) Coh ( B ) is a globally bounded graded finite-projective A A A -module V V V with a flat degree-one right superconnection on V ⊗ A B V\otimes_A B V ⊗ A B :
D ( v b ) = D ( v ) b + ( − 1 ) ∣ v ∣ v d B b , D 2 = 0. D(vb)=D(v)b+(-1)^{|v|}v\,d_Bb,\qquad D^2=0. D ( v b ) = D ( v ) b + ( − 1 ) ∣ v ∣ v d B b , D 2 = 0.
A homogeneous degree-m m m morphism is signlessly right-linear, ϕ ( v b ) = ϕ ( v ) b \phi(vb)=\phi(v)b ϕ ( v b ) = ϕ ( v ) b , and has differential and composition
d ϕ = D W ϕ − ( − 1 ) m ϕ D V , ( ψ ϕ ) ( v ) = ψ ( ϕ ( v ) ) . d\phi=D_W\phi-(-1)^m\phi D_V,\qquad(\psi\phi)(v)=\psi(\phi(v)). d ϕ = D W ϕ − ( − 1 ) m ϕ D V , ( ψ ϕ ) ( v ) = ψ ( ϕ ( v )) .
The internal differential is D [ 0 ] D^{[0]} D [ 0 ] . The regular full subcategory Coh r e g ( B ) \operatorname{Coh}^{\mathrm{reg}}(B) Coh reg ( B ) requires locally constant fiber-cohomology ranks. It is shift-closed; a cone belongs to it when its fiber-cohomology ranks are locally constant. All coefficient ranks and amplitudes are globally bounded. In a fixed Hom degree m m m , only finitely many terms with j + r = i + m j+r=i+m j + r = i + m occur. The degrees of B B B may be unbounded. These are the conventions of [12 Section 6] . In interval formulas d t dt d t is on the left.
The notation Rep s m p e r f \operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}} Rep sm perf refers to the actual bounded finite-rank vertex-complex category of § 4 . Its definition retains the specified vertex complexes and normalized cochains, rather than passing to a derived-perfect replacement or localizing over stack presentations.
Fixed-base and global-resolution comparisons
The following results compare other admissible global resolutions with the selected affine resolution. The explicit affine comparison and its RH proof proceed independently of these results. All categories are taken in compatible universes.
Let ϵ E : B E → A \epsilon_E:B_E\to A ϵ E : B E → A be the graded zero-section augmentation. A polynomial parameter map H : B E ′ → B E ⊗ R Ω p o l y ∙ ( Δ n ) H:B_{E'}\to B_E\otimes_{\mathbb{R}}\Omega^\bullet_{\mathrm{poly}}(\Delta^n) H : B E ′ → B E ⊗ R Ω poly ∙ ( Δ n ) is pointed when
( ϵ E ⊗ Id ) H = ι ϵ E ′ , (\epsilon_E\otimes\operatorname{Id})H=\iota\epsilon_{E'}, ( ϵ E ⊗ Id ) H = ι ϵ E ′ ,
where ι \iota ι is constant extension. Fixing base functions alone does not impose this condition: positive generators must have no pure parameter-form terms. This is [12 Definition 3.5 and Lemma 3.6] .
For general weak maps, we use the connective mapping-space comparison in [7 Remarks 5.27–5.28] , as specified in [12 Section 3] . It identifies the polynomial model with the localization of connective strict-anchor algebroids and strict maps. Its target-fibrancy condition is positive-degree surjectivity, automatic over the discrete base. In contrast, the unbounded fibrancy condition requires a surjective anchor, which a singular resolution need not have.
The next statement collects [12 Theorems 3.9, 4.5 and 6.4, Corollary 8.4] .
Theorem 2.1 (Fixed-base comparison). Under the preceding geometric hypotheses and connective mapping-space input, a given F : E ⇝ E ′ F : E \rightsquigarrow E' F : E ⇝ E ′ in W 1 \mathcal{W}_1 W 1 has a geometric inverse up to finite chains of pointed polynomial homotopies. Its cohesive pullback is a dg quasi-equivalence, including on regular coefficients. The actual map Int ( F ) \operatorname{Int}(F) Int ( F ) is an objectwise simplicial homotopy equivalence, and its raw representation pullback is a dg quasi-equivalence, also on the regular full subcategories.
The inverse homotopies can be made jointly smooth. They preserve the entire base simplex and are relative to c ( y M ) c(yM) c ( y M ) . This is stronger than an abstract equivalence of smooth stacks.
The following resolution-specific route does not require that general connective localization input. For a degree-zero algebroid with injective module anchor f ↪ T A f \hookrightarrow T_A f ↪ T A , let R f \mathcal{R}_f R f be the nonempty simplicial category of its supplied strict global geometric resolutions, with all actual reduced pointed polynomial mapping simplices.
The resolution mapping spaces satisfy the following uniqueness result [12 Theorem 5.2] .
Theorem 2.2 (Coherent resolution uniqueness). For E , E ′ ∈ R f E,E' \in\mathcal{R}_f E , E ′ ∈ R f , the mapping space is a weakly contractible Kan complex. The homotopy-coherent nerve N h c R f N_{\mathrm{hc}}\mathcal{R}_f N hc R f is contractible. Every mapping simplex automatically commutes with the resolution augmentations.
Integration extends to these mapping spaces by [12 Theorem 5.3 and equation (5.2)] .
Theorem 2.3 (Enriched full-MC integration). Full smooth integration sends pointed polynomial mapping simplices to smooth simplicial mapping simplices, compatibly with identities and composition. For a mapping n n n -simplex H H H , an MC m m m -simplex σ \sigma σ and τ : [ m ] → [ n ] \tau: [m] \to[n] τ : [ m ] → [ n ] , its value is
H N Q ∘ ( σ , T [ 1 ] λ τ ) , ( λ τ ( t ) ) j = ∑ τ ( i ) = j t i . H_{NQ} \circ\left(\sigma,T[1]\lambda_\tau\right), \qquad\left(\lambda_\tau(t)\right)_j = \sum_{\tau(i)=j} t_i. H N Q ∘ ( σ , T [ 1 ] λ τ ) , ( λ τ ( t ) ) j = τ ( i ) = j ∑ t i . It preserves the actual base simplex and fixes the whole c ( y M ) c(yM) c ( y M ) throughout every mapping simplex.
Consequently, one obtains the following framed comparison [12 Corollary 5.5] .
Corollary 2.4 (Framed monodromy comparison). For either smooth infinity-target, the integration diagram on N h c R f N_{\mathrm{hc}}\mathcal{R}_f N hc R f is equivalent to a constant diagram. After selecting a resolution and framing the component there by the identity, the space of coherent comparisons is contractible.
For actual comparisons between these global resolutions, the cohesive and raw representation pullbacks are dg quasi-equivalences, including on regular coefficients, by [12 Corollaries 6.3 and 8.5] . Thus the supplied global resolutions give equivalent cohesive and raw representation categories.
Citation versions. Numbered citations to [7 ] follow arXiv:1712.03441v2; those to [3 ] use the published article. Page references to [6 ] use arXiv:1207.0249v1. For [AS13] the numbered source is arXiv:1011.4693v2. The higher Van Est references [2 AC25] use their respective version 2 preprints. Citations to [12 ] use its October 30, 2026 revision 8.
A finite Lie 2-groupoid model for the affine resolution
We now compare full smooth MC integration with a finite-dimensional model, without truncating the source simplicial sheaf. The example is the resolution of
M = R 2 , F = ⟨ x ∂ x , y ∂ x ⟩ M = \mathbb{R}^{2}, \qquad\mathcal{F} = \langle x\partial_x, y\partial_x\rangle M = R 2 , F = ⟨ x ∂ x , y ∂ x ⟩
used in the transport construction of [13 ] . The point-base Morita comparison of [9 arXiv v4, Theorem 2.15, p. 11] concerns the 2-truncated integration of a finite-dimensional strict Lie 2-algebra. Here the comparison allows nonconstant base maps in the MC simplices, arbitrary smooth parameter families, and all simplicial degrees.
The resolution and the strict target
Put A = C ∞ ( R 2 ) A = C^{\infty}(\mathbb{R}^{2}) A = C ∞ ( R 2 ) . The resolution has
0 ⟶ A ε → d 2 A e 1 ⊕ A e 2 → ρ F ( M ) ⟶ 0 , d 2 ε = − y e 1 + x e 2 , 0 \longrightarrow A\varepsilon\xrightarrow{d_{2}} Ae_{1} \oplus Ae_{2} \xrightarrow{\rho} \mathcal{F}(M) \longrightarrow0,\qquad d_{2}\varepsilon= -ye_{1} + xe_{2}, 0 ⟶ A ε d 2 A e 1 ⊕ A e 2 ρ F ( M ) ⟶ 0 , d 2 ε = − y e 1 + x e 2 ,
with ρ ( e 1 ) = x ∂ x \rho(e_{1}) = x\partial_x ρ ( e 1 ) = x ∂ x and ρ ( e 2 ) = y ∂ x \rho(e_{2}) = y\partial_x ρ ( e 2 ) = y ∂ x . Its CE differential is
Q f = ( x f x ) θ 1 + ( y f x ) θ 2 , Q θ 1 = − y η , Qf = (xf_x)\theta_{1} + (yf_x)\theta_{2}, \qquad Q\theta_{1} = -y\eta, Q f = ( x f x ) θ 1 + ( y f x ) θ 2 , Q θ 1 = − y η ,
Q θ 2 = θ 1 θ 2 + x η , Q η = 0 , ∣ θ i ∣ = 1 , ∣ η ∣ = 2. (4) Q\theta_{2} = \theta_{1}\theta_{2} + x\eta, \qquad Q\eta= 0, \qquad|\theta_i| = 1,\qquad|\eta| = 2. \tag*{(4)} Q θ 2 = θ 1 θ 2 + x η , Q η = 0 , ∣ θ i ∣ = 1 , ∣ η ∣ = 2. (4)
We denote this geometric algebroid by E a f f E_{\mathrm{aff}} E aff .
Let
G = Aff + ( 1 ) = { ( a b 0 1 ) : a > 0 } , G = \operatorname{Aff}_{+}(1) = \left\{ \begin{pmatrix} a & b \\ 0 & 1 \end{pmatrix} : a > 0 \right\}, G = Aff + ( 1 ) = { ( a 0 b 1 ) : a > 0 } ,
( a , b ) ( x , y ) = ( a x + b y , y ) , Z m = ( − y m x m 0 0 ) . (a,b)(x,y) = (ax + by,y), \qquad Z_m = \begin{pmatrix} -y_m & x_m \\ 0 & 0 \end{pmatrix}. ( a , b ) ( x , y ) = ( a x + b y , y ) , Z m = ( − y m 0 x m 0 ) .
The additive group bundle M × R M \times\mathbb{R} M × R maps to G ⋉ M G \ltimes M G ⋉ M by
∂ ( m , h ) = ( exp ( h Z m ) , m ) . \partial(m,h) = (\exp(hZ_m),m). ∂ ( m , h ) = ( exp ( h Z m ) , m ) .
The G G G -action transports m m m and leaves the scalar h h h unchanged.
Proposition 3.1 (Affine resolution and crossed module). The displayed complex is exact, and (3.1) squares to zero. The displayed group-bundle map is a crossed module. Its strict Lie 2-groupoid has a Duskin nerve
K n ≅ M × G n × R ( n 2 ) . K_n \cong M \times G^n \times\mathbb{R}^{\binom{n}{2}}. K n ≅ M × G n × R ( 2 n ) . The horn maps are surjective submersions, with unique fillers above degree two. Its smooth second isotropy is R \mathbb{R} R at the origin and zero elsewhere.
Proof. If x f + y g = 0 xf+yg=0 x f + y g = 0 , restriction to x = 0 x=0 x = 0 and continuity give g ( 0 , y ) = 0 g(0,y)=0 g ( 0 , y ) = 0 . Hadamard division gives g = x h g=xh g = x h ; density of x ≠ 0 x\ne0 x = 0 gives f = − y h f=-yh f = − y h and also proves injectivity of d 2 d_2 d 2 . The same integral division works globally on R 2 \mathbb{R}^2 R 2 . The graded Leibniz rule verifies Q 2 = 0 Q^2=0 Q 2 = 0 on the smooth coordinate functions and the three positive generators; for example
Q 2 θ 2 = − y η θ 2 − x θ 1 η + ( x θ 1 + y θ 2 ) η = 0. Q^2\theta_2=-y\eta\theta_2-x\theta_1\eta+(x\theta_1+y\theta_2)\eta=0. Q 2 θ 2 = − y η θ 2 − x θ 1 η + ( x θ 1 + y θ 2 ) η = 0.
Matrix multiplication gives Z m m = 0 Z_m m=0 Z m m = 0 and g Z m g − 1 = Z g m gZ_mg^{-1}=Z_{gm} g Z m g − 1 = Z g m . Moreover
exp ( h Z m ) = ( e − h y m x m h φ ( h y m ) 0 1 ) , φ ( t ) = ∫ 0 1 e − s t d s . (5) \exp(hZ_m)=\begin{pmatrix}e^{-hy_m}&x_mh\varphi(hy_m)\\0&1\end{pmatrix},\qquad\varphi(t)=\int_0^1e^{-st}\,\mathrm{d}s. \tag*{(5)} exp ( h Z m ) = ( e − h y m 0 x m h φ ( h y m ) 1 ) , φ ( t ) = ∫ 0 1 e − s t d s . (5)
Thus the boundary is smooth, fixes m m m , and is equivariant. The Peiffer identity follows from the trivial action on the scalar and the commutativity of the coefficient group.
Fix the triangle convention
g i k = g j k g i j exp ( h i j k Z m i ) . (6) g_{ik}=g_{jk}g_{ij}\exp(h_{ijk}Z_{m_i}). \tag*{(6)} g ik = g j k g ij exp ( h ij k Z m i ) . (6)
The root edges a i = g 0 i a_i=g_{0i} a i = g 0 i and rooted triangle parameters h 0 i j h_{0ij} h 0 ij determine every other coordinate:
m i = a i m 0 , g i j = a j exp ( − h 0 i j Z m 0 ) a i − 1 , h i j k = h 0 i j + h 0 j k − h 0 i k . (7) m_i=a_i m_0,\qquad g_{ij}=a_j\exp(-h_{0ij}Z_{m_0})a_i^{-1},\qquad h_{ijk}=h_{0ij}+h_{0jk}-h_{0ik}. \tag*{(7)} m i = a i m 0 , g ij = a j exp ( − h 0 ij Z m 0 ) a i − 1 , h ij k = h 0 ij + h 0 j k − h 0 ik . (7)
The tetrahedral identity is δ h = 0 \delta h=0 δ h = 0 . Degree-one horns have the source and target submersions of the action groupoid. In degree two, each missing edge is solved by multiplication and inverse for every chosen h h h ; these coordinate changes identify the horn map with a projection with fiber R \mathbb{R} R . The tetrahedral identity determines a missing degree-three filler, and its cocycle identity gives unique higher fillers. All formulas are smooth, also after applying C ∞ ( U , − ) C^\infty(U,-) C ∞ ( U , − ) . Finally, exp ( h Z m ) = 1 \exp(hZ_m)=1 exp ( h Z m ) = 1 forces h = 0 h=0 h = 0 when y m ≠ 0 y_m\ne0 y m = 0 , or when y m = 0 y_m=0 y m = 0 , x m ≠ 0 x_m\ne0 x m = 0 . At m = 0 m=0 m = 0 it imposes no condition on h h h .
The nerve has all degrees; an n n n -simplex is determined by its compatible 2-skeleton, but not every 2-skeleton extends. The scalar isotropy just described is a smooth group of 2-automorphisms, not the ordinary homotopy group of the topological manifold R \mathbb{R} R .
The full MC equations and comparison map
For a smooth parameter manifold U U U , a full MC simplex has a base map σ = ( x , y ) : U × Δ n → M \sigma=(x,y):U\times\Delta^n\to M σ = ( x , y ) : U × Δ n → M and relative forms α 1 , α 2 , β \alpha_1,\alpha_2,\beta α 1 , α 2 , β of degrees 1, 1, 2. All are jointly smooth through the faces, and
d y = 0 , d x = x α 1 + y α 2 , d α 1 = − y β , d α 2 = α 1 ∧ α 2 + x β , d β = 0. (8) \begin{aligned}
\mathrm{d}y&=0, & \mathrm{d}x&=x\alpha_1+y\alpha_2,\\
\mathrm{d}\alpha_1&=-y\beta, & \mathrm{d}\alpha_2&=\alpha_1\wedge\alpha_2+x\beta, & \mathrm{d}\beta&=0. \tag*{(8)}
\end{aligned} d y d α 1 = 0 , = − y β , d x d α 2 = x α 1 + y α 2 , = α 1 ∧ α 2 + x β , d β = 0. (8)
Here and below d d d differentiates only in Δ n \Delta^n Δ n . Write
A = ( α 1 α 2 0 0 ) , Z σ = ( − y x 0 0 ) . A=\begin{pmatrix}\alpha_1&\alpha_2\\0&0\end{pmatrix},\qquad Z_\sigma=\begin{pmatrix}-y&x\\0&0\end{pmatrix}. A = ( α 1 0 α 2 0 ) , Z σ = ( − y 0 x 0 ) .
The equations are equivalently
d σ = A σ , d A − A ∧ A = Z σ β , d β = 0 , (9) \mathrm{d}\sigma=A\sigma,\qquad\mathrm{d}A-A\wedge A=Z_\sigma\beta,\qquad\mathrm{d}\beta=0, \tag*{(9)} d σ = A σ , d A − A ∧ A = Z σ β , d β = 0 , (9)
and imply d Z σ = A Z σ − Z σ A \mathrm{d}Z_\sigma=AZ_\sigma-Z_\sigma A d Z σ = A Z σ − Z σ A .
For an oriented edge [ i j ] [ij] [ ij ] , let g i j g_{ij} g ij be the holonomy defined by the left matrix equation
P ′ ( t ) = A ( ∂ t ) P ( t ) , P ( 0 ) = 1. P'(t)=A(\partial_t)P(t), \qquad P(0)=1. P ′ ( t ) = A ( ∂ t ) P ( t ) , P ( 0 ) = 1.
For every oriented triangle set
h i j k = − ∫ [ i j k ] β . (10) h_{ijk}=-\int_{[ijk]}\beta. \tag*{(10)} h ij k = − ∫ [ ij k ] β . (10)
Theorem 3.2 (Full affine integration comparison). These edge holonomies and triangle coordinates define a natural vertex-preserving map of simplicial smooth sheaves
Φ : Int ( E a f f ) ⟶ y K . \Phi:\operatorname{Int}(E_{\mathrm{aff}})\longrightarrow yK. Φ : Int ( E aff ) ⟶ y K . For every smooth parameter manifold U U U , Φ ( U ) \Phi(U) Φ ( U ) is a trivial Kan fibration of simplicial sets. In particular, it is an objectwise weak equivalence of full simplicial sheaves, including families passing through the origin and the locus y = 0 y=0 y = 0 .
Appendix A proves the theorem by constructing lifts against every boundary inclusion and then supplies an actual smooth simplicial section and whole-unit-relative homotopy. The comparison itself is intrinsic; the auxiliary choices in a lift need not commute with all simplex operators. Section 7 gives the range-preserving ordinary-open construction.
A concrete coefficient target on a simplicial smooth sheaf
Smooth cochains with two endpoints
Smooth cochains with two endpoints. We use a raw dg category of representations on a simplicial smooth sheaf. The construction is a standard twisting-cochain form of homotopy-coherent representations, with explicit totalization signs; compare [AS13, Definition 2.2 and Remark 2.4]. The plot-smooth family condition and the uniform boundedness imposed here are additional requirements. We give the definitions and signs below; the dg identities follow from [12 Section 7] .
Let S S S be a simplicial smooth sheaf with S 0 = y M S_0=yM S 0 = y M . Let V , W V,W V , W be globally bounded finite-rank graded bundles on M M M , with finite-projective global section modules. Write
C s m p , q ( S ; V , W ) C_{\mathrm{sm}}^{p,q}(S;V,W) C sm p , q ( S ; V , W )
for plot-smooth sections of Hom q ( v p ∗ V , v 0 ∗ W ) \operatorname{Hom}^{q}(v_p^*V,v_0^*W) Hom q ( v p ∗ V , v 0 ∗ W ) on S p S_p S p , for p ≥ 0 p\geq0 p ≥ 0 . Precisely, such a section assigns to each σ ∈ S p ( T ) \sigma\in S_p(T) σ ∈ S p ( T ) a smooth degree-q q q bundle map from ( v p σ ) ∗ V (v_p\sigma)^*V ( v p σ ) ∗ V to ( v 0 σ ) ∗ W (v_0\sigma)^*W ( v 0 σ ) ∗ W on the test manifold T T T , naturally under smooth pullback. Transport acts from the last vertex to the first.
The total degree is p + q p+q p + q . Set
C s m m ( S ; V , W ) = ∏ p ≥ 0 C s m p , m − p ( S ; V , W ) . C_{\mathrm{sm}}^{m}(S;V,W)=\prod_{p\geq0}C_{\mathrm{sm}}^{p,m-p}(S;V,W). C sm m ( S ; V , W ) = p ≥ 0 ∏ C sm p , m − p ( S ; V , W ) .
This product is finite in each total degree: the internal degrees of Hom ( V , W ) \operatorname{Hom}(V,W) Hom ( V , W ) range over a fixed finite interval.
For α \alpha α of bidegree ( p , q ) (p,q) ( p , q ) and β \beta β of bidegree ( r , s ) (r,s) ( r , s ) with composable coefficient bundles, define
( α ⋆ β ) ( σ ) = ( − 1 ) q r α ( σ [ 0 , … , p ] ) β ( σ [ p , … , p + r ] ) . (11) (\alpha\star\beta)(\sigma)=(-1)^{qr}\alpha(\sigma[0,\ldots,p])\beta(\sigma[p,\ldots,p+r]). \tag*{(11)} ( α ⋆ β ) ( σ ) = ( − 1 ) q r α ( σ [ 0 , … , p ]) β ( σ [ p , … , p + r ]) . (11)
Here juxtaposition is composition, with the right factor applied first. Define the inner-face differential by
( δ α ) ( σ ) = ∑ i = 1 p ( − 1 ) i α ˙ ( d i σ ) , σ ∈ S p + 1 . (12) (\delta\alpha)(\sigma)=\sum_{i=1}^{p}(-1)^i\dot{\alpha}(d_i\sigma),\qquad\sigma\in S_{p+1}. \tag*{(12)} ( δ α ) ( σ ) = i = 1 ∑ p ( − 1 ) i α ˙ ( d i σ ) , σ ∈ S p + 1 . (12)
Inner faces retain the two endpoints, so every term has the correct bundle type. The product and inner differential satisfy the following identities [12 Lemma 7.1] .
Lemma 4.1. The product is associative, δ 2 = 0 \delta^2=0 δ 2 = 0 , and
δ ( α ⋆ β ) = ( δ α ) ⋆ β + ( − 1 ) p + q α ⋆ ( δ β ) . \delta(\alpha\star\beta)=(\delta\alpha)\star\beta+(-1)^{p+q}\alpha\star(\delta\beta). δ ( α ⋆ β ) = ( δ α ) ⋆ β + ( − 1 ) p + q α ⋆ ( δ β ) .
Representations and their Hom differential
Definition 4.2. An object of Rep s m p e r f ( S ) \operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(S) Rep sm perf ( S ) is ( V , r V ) (V,r_V) ( V , r V ) , where
r V = ∑ p ≥ 0 r V , p , r V , p ∈ C s m p , 1 − p ( S ; V , V ) , r_V=\sum_{p\geq0}r_{V,p},\qquad r_{V,p}\in C_{\mathrm{sm}}^{p,1-p}(S;V,V), r V = p ≥ 0 ∑ r V , p , r V , p ∈ C sm p , 1 − p ( S ; V , V ) , satisfies
δ r V + r V ⋆ r V = 0. \delta r_V+r_V\star r_V=0. δ r V + r V ⋆ r V = 0. We impose strict unital normalization
r V , 1 ( s 0 x ) = Id V x , r V , p ( s i σ ) = 0 ( p ≥ 2 ) , r_{V,1}(s_0x)=\operatorname{Id}_{V_x},\qquad r_{V,p}(s_i\sigma)=0\quad(p\geq2), r V , 1 ( s 0 x ) = Id V x , r V , p ( s i σ ) = 0 ( p ≥ 2 ) , and require every edge transport to be a fiberwise quasi-isomorphism. A degree-m m m morphism Φ \Phi Φ has components Φ p ∈ C s m p , m − p ( S ; V , W ) \Phi_p\in C_{\mathrm{sm}}^{p,m-p}(S;V,W) Φ p ∈ C sm p , m − p ( S ; V , W ) , vanishing on degeneracies for p ≥ 1 p\geq1 p ≥ 1 . Its differential is
D Φ = δ Φ + r W ⋆ Φ − ( − 1 ) m Φ ⋆ r V . (13) D\Phi=\delta\Phi+r_W\star\Phi-(-1)^m\Phi\star r_V. \tag*{(13)} D Φ = δ Φ + r W ⋆ Φ − ( − 1 ) m Φ ⋆ r V . (13) Composition is ⋆ \star ⋆ .
Here r V , 0 r_{V,0} r V , 0 is the internal differential. The degree-one Maurer–Cartan equation says that r V , 1 r_{V,1} r V , 1 is a chain map. In degree two it reads
− r V , 1 ( d 1 σ ) + r V , 0 r V , 2 + r V , 1 ( σ [ 0 , 1 ] ) r V , 1 ( σ [ 1 , 2 ] ) + r V , 2 r V , 0 = 0. -r_{V,1}(d_1\sigma)+r_{V,0}r_{V,2}+r_{V,1}(\sigma[0,1])r_{V,1}(\sigma[1,2])+r_{V,2}r_{V,0}=0. − r V , 1 ( d 1 σ ) + r V , 0 r V , 2 + r V , 1 ( σ [ 0 , 1 ]) r V , 1 ( σ [ 1 , 2 ]) + r V , 2 r V , 0 = 0. Thus r V , 2 r_{V,2} r V , 2 supplies the indicated chain homotopy. More precisely, set ε p = ( − 1 ) p ( p − 1 ) / 2 \varepsilon_p=(-1)^{p(p-1)/2} ε p = ( − 1 ) p ( p − 1 ) /2 and F p = ε p r V , p F_p=\varepsilon_pr_{V,p} F p = ε p r V , p . For k ≥ 1 k\geq1 k ≥ 1 the identity
( − 1 ) ( 1 − p ) ( k − p ) ε k − 1 ε p ε k − p = ( − 1 ) p + 1 (-1)^{(1-p)(k-p)}\varepsilon_{k-1}\varepsilon_p\varepsilon_{k-p}=(-1)^{p+1} ( − 1 ) ( 1 − p ) ( k − p ) ε k − 1 ε p ε k − p = ( − 1 ) p + 1 converts the degree-k k k structure equation to [AS13, Definition 2.2, equation (1)], whose front/back cup has no internal Koszul sign. For k = 0 k=0 k = 0 both equations say r V , 0 2 = 0 r_{V,0}^2=0 r V , 0 2 = 0 , and ε 1 = 1 \varepsilon_1=1 ε 1 = 1 preserves the strict unit. This object rescaling is not the morphism dictionary. Its all-degree extension, including the differential and composition, is Lemma 6.2.
If S S S is Kan on point-valued simplices, inverse edges and two-simplices give chain-homotopy inverses of the edge maps, hence the required fiberwise quasi-isomorphisms. They do not imply that the original edge maps are degreewise invertible. The edge quasi-isomorphism condition is included explicitly so that the definition does not assume raw Kan fibrancy for every integration used below.
These operations give the following dg structure [12 Proposition 7.3] .
Proposition 4.3. Definition 4.2 defines a dg category. The identity has only its degree-zero component Id V \operatorname{Id}_V Id V .
The inner differential δ \delta δ alone need not preserve normalization. It is the full twisted differential (4.3), with its unit terms, that does so.
At simplicial degree zero, ( D Φ ) 0 = r W , 0 Φ 0 − ( − 1 ) m Φ 0 r V , 0 (D\Phi)_0 = r_{W,0}\Phi_0 - (-1)^m\Phi_0r_{V,0} ( D Φ ) 0 = r W , 0 Φ 0 − ( − 1 ) m Φ 0 r V , 0 . When S S S is the nerve of an ordinary groupoid and the coefficient bundles are concentrated in degree zero, these objects are ordinary contravariant groupoid representations.
Let Rep s m p e r f , r e g ( S ) \operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf,reg}}(S) Rep sm perf , reg ( S ) be the full subcategory for which r V , 0 r_{V,0} r V , 0 has locally constant fiber-cohomology ranks. A cone belongs to this subcategory when its fiber-cohomology ranks are locally constant.
Strict simplicial maps induce actual raw dg pullback functors by precomposition and pullback of the vertex bundles. The inverse-homotopy hypotheses of Section 5 ensure that the pullback is a quasi-equivalence. For an open O ⊆ M O \subseteq M O ⊆ M , Int ( E ∣ O ) \operatorname{Int}(E|_O) Int ( E ∣ O ) uses simplices whose entire base image lies in O O O , not merely their vertices. The affine RH theorem uses this raw category before Hom-sheafification. The latter construction, described in [12 Proposition 7.4] , sheafifies the morphism complexes while retaining the existing global objects.
Representation invariance and the affine coefficient comparison
We use the actual raw dg category of Section 4 on the simplicial smooth sheaf itself. Homotopy invariance for normalized-cochain Maurer–Cartan and twisted-module categories is established in [3 Theorem 5.2, Proposition 6.2 and Corollary 6.1] . These results give dg quasi-equivalences. Their scalar coefficient models and strong interval conventions differ from the plot-smooth bundle category used here. The general plot-smooth prism proof is [12 Section 8] . We recall its relative-homotopy hypotheses and prove the consequence for the specified affine Φ \Phi Φ , Ψ \Psi Ψ , H H H .
The normalized prism
Let X 0 = Y 0 = y M X_0 = Y_0 = yM X 0 = Y 0 = y M and let H : X × Δ [ 1 ] → Y H : X \times\Delta[1] \to Y H : X × Δ [ 1 ] → Y have endpoints f = H ( − , 0 ) f = H(-, 0) f = H ( − , 0 ) and g = H ( − , 1 ) g = H(-, 1) g = H ( − , 1 ) , both the identity on the vertex sheaf. A homotopy relative to c ( y M ) c(yM) c ( y M ) is constant on the whole subsheaf c ( y M ) × Δ [ 1 ] c(yM) \times\Delta[1] c ( y M ) × Δ [ 1 ] , not merely at its degree-zero endpoints. For σ ∈ X n \sigma\in X_n σ ∈ X n put
h i ( σ ) = H n + 1 ( s i σ , τ i ) , 0 ≤ i ≤ n , h_i(\sigma) = H_{n+1}(s_i\sigma,\tau_i), \qquad0 \leq i \leq n, h i ( σ ) = H n + 1 ( s i σ , τ i ) , 0 ≤ i ≤ n ,
where τ i : [ n + 1 ] → [ 1 ] \tau_i : [n+1] \to[1] τ i : [ n + 1 ] → [ 1 ] is zero through i i i and one thereafter. For α ∈ C s m p , q ( Y ; V , W ) \alpha\in C_{\mathrm{sm}}^{p,q}(Y;V,W) α ∈ C sm p , q ( Y ; V , W ) with p ≥ 1 p \geq1 p ≥ 1 define
( P H α ) ( σ ) = ∑ i = 0 p − 1 ( − 1 ) i α ( h i σ ) , σ ∈ X p − 1 . (14) (P_H\alpha)(\sigma) = \sum_{i=0}^{p-1}(-1)^i\alpha(h_i\sigma), \qquad\sigma\in X_{p-1}. \tag*{(14)} ( P H α ) ( σ ) = i = 0 ∑ p − 1 ( − 1 ) i α ( h i σ ) , σ ∈ X p − 1 . (14)
Set P H = 0 P_H = 0 P H = 0 on simplicial degree zero. It lowers total degree by one, without an additional internal-degree sign. The first endpoint of every prism simplex is at time zero and the last at time one. Since f 0 = g 0 = Id M f_0 = g_0 = \operatorname{Id}_M f 0 = g 0 = Id M , the bundle types in the sum coincide.
The prism satisfies the following identities [12 Lemma 8.1] .
Lemma 5.1 (Normalized prism identities). The operator is natural on smooth plots, preserves positive-degree normalization, and satisfies
δ P H + P H δ = 0 , (15) \delta P_H + P_H\delta= 0, \tag*{(15)} δ P H + P H δ = 0 , (15)
P H ( α ⋆ β ) = ( P H α ) ⋆ g ∗ β + ( − 1 ) ∣ α ∣ f ∗ α ⋆ P H β . (16) P_H(\alpha\star\beta) = (P_H\alpha) \star g^*\beta+ (-1)^{|\alpha|}f^*\alpha\star P_H\beta. \tag*{(16)} P H ( α ⋆ β ) = ( P H α ) ⋆ g ∗ β + ( − 1 ) ∣ α ∣ f ∗ α ⋆ P H β . (16)
Finite gauges and all-degree naturality
For a whole-unit-relative homotopy, the prism also gives the following gauges [12 Lemma 8.2 and equation (8.4)] .
Lemma 5.2 (Pointed gauges). Assume now that H H H is relative to c ( y M ) c(yM) c ( y M ) . For a representation R = ( V , r ) R=(V,r) R = ( V , r ) set U R = P H r U_R=P_Hr U R = P H r . This is a normalized, closed degree-zero morphism
U R : g ∗ R ⟶ f ∗ R , ( U R ) 0 = Id V , U_R:g^*R\longrightarrow f^*R,\qquad(U_R)_0=\operatorname{Id}_V, U R : g ∗ R ⟶ f ∗ R , ( U R ) 0 = Id V ,
with an actual closed inverse. For a homogeneous morphism α : R V → R W \alpha:R_V\to R_W α : R V → R W of any total degree m ∈ Z m\in\mathbb{Z} m ∈ Z , one has
D P H α + P H D α = U W ⋆ g ∗ α − f ∗ α ⋆ U V . (17) DP_H\alpha+P_HD\alpha=U_W\star g^*\alpha-f^*\alpha\star U_V. \tag*{(17)} D P H α + P H D α = U W ⋆ g ∗ α − f ∗ α ⋆ U V . (17)
Here the first differential is on Hom X ( g ∗ R V , f ∗ R W ) \operatorname{Hom}_X(g^*R_V,f^*R_W) Hom X ( g ∗ R V , f ∗ R W ) ; the differential inside P H D P_HD P H D is on the original Hom \operatorname{Hom} Hom complex over Y Y Y .
The hypothesis on the whole c ( y M ) c(yM) c ( y M ) makes the vertical edge H ( s 0 x , τ 0 ) H(s_0x,\tau_0) H ( s 0 x , τ 0 ) degenerate, so ( U R ) 0 = 1 (U_R)_0=1 ( U R ) 0 = 1 . Equality of endpoint object maps alone does not justify it. For V V V supported in [ a , b ] [a,b] [ a , b ] , write U R = 1 + u U_R=1+u U R = 1 + u . The inverse is the finite formula [12 equation (8.5)]
U R − 1 = ∑ j = 0 b − a ( − u ) ⋆ j (18) U_R^{-1}=\sum_{j=0}^{b-a}(-u)^{\star j} \tag*{(18)} U R − 1 = j = 0 ∑ b − a ( − u ) ⋆ j (18)
with the reverse representation type. Positive components u p ∈ C p , − p u_p\in C^{p,-p} u p ∈ C p , − p lower internal degree, so u ⋆ N = 0 u^{\star N}=0 u ⋆ N = 0 for N > b − a N>b-a N > b − a . The twisted identity applies to nonclosed morphisms and negative total degrees as well.
Raw homotopy invariance
These identities yield homotopy invariance [12 Theorem 8.3] .
Theorem 5.3 (Raw representation invariance). Let f : X → Y f:X\to Y f : X → Y and g : Y → X g:Y\to X g : Y → X fix the common vertex sheaf y M yM y M . Suppose there are simplicial homotopies
H Y : f g ⟹ Id Y , H X : g f ⟹ Id X H_Y:f g\Longrightarrow\operatorname{Id}_Y,\qquad H_X:g f\Longrightarrow\operatorname{Id}_X H Y : f g ⟹ Id Y , H X : g f ⟹ Id X relative to c ( y M ) c(yM) c ( y M ) on both sides. Then the actual pullback dg functor
f ∗ : Rep s m p e r f ( Y ) ⟶ Rep s m p e r f ( X ) f^*:\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(Y)\longrightarrow\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(X) f ∗ : Rep sm perf ( Y ) ⟶ Rep sm perf ( X ) is a dg quasi-equivalence. Its maps on all Hom \operatorname{Hom} Hom complexes are chain homotopy equivalences. The result restricts to the regular full subcategories and requires no Kan hypothesis on X X X or Y Y Y .
The actual affine pullback
Corollary 5.4 (Global affine coefficient comparison). For the affine comparison of Theorem A.10 , pullback gives a dg quasi-equivalence
Φ ∗ : Rep s m p e r f ( y K ) ⟶ Rep s m p e r f ( Int ( E a f f ) ) , \Phi^*:\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(yK)\longrightarrow\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(\operatorname{Int}(E_{\mathrm{aff}})), Φ ∗ : Rep sm perf ( y K ) ⟶ Rep sm perf ( Int ( E aff )) , also on the regular full subcategories. For objects R , Q R,Q R , Q on y K yK y K , write
i = Φ ∗ , p = Ψ ∗ i=\Phi^*,\qquad p=\Psi^* i = Φ ∗ , p = Ψ ∗ on their Hom complexes. Then
p i = Id , D P H + P H D = Id − i p . (19) pi=\operatorname{Id},\qquad DP_H+P_HD=\operatorname{Id}-ip. \tag*{(19)} p i = Id , D P H + P H D = Id − i p . (19) Thus the Hom complex on full MC integration retracts onto the image of the Hom complex on the finite nerve.
Proof. Because H H H is over T = y K T=yK T = y K , its prism simplices satisfy
Φ ( h i σ ) = s i Φ ( σ ) . \Phi(h_i\sigma)=s_i\Phi(\sigma). Φ ( h i σ ) = s i Φ ( σ ) .
Strict units and normalization give U Φ ∗ R = Id U_{\Phi^*R}=\operatorname{Id} U Φ ∗ R = Id : the zero component is the unit, and every positive one uses a degenerate structure component of degree at least two. Also P H ( Φ ∗ γ ) = 0 P_H(\Phi^*\gamma)=0 P H ( Φ ∗ γ ) = 0 for a normalized morphism γ \gamma γ .
For an arbitrary cochain α \alpha α between Φ ∗ R \Phi^*R Φ ∗ R and Φ ∗ Q \Phi^*Q Φ ∗ Q , the twisted prism identity therefore becomes
D P H α + P H D α = α − ( Ψ Φ ) ∗ α . DP_H\alpha+P_HD\alpha=\alpha-(\Psi\Phi)^*\alpha. D P H α + P H D α = α − ( ΨΦ ) ∗ α .
The objects are fixed by ( Ψ Φ ) ∗ (\Psi\Phi)^* ( ΨΦ ) ∗ , since Φ Ψ = Id T \Phi\Psi=\operatorname{Id}_T ΦΨ = Id T , so this is an identity on one Hom complex, not only on cochains already in the image of i i i . The equality Ψ ∗ Φ ∗ = Id \Psi^*\Phi^*=\operatorname{Id} Ψ ∗ Φ ∗ = Id gives (5.6). For any object R ′ R' R ′ on Int ( E a f f ) \operatorname{Int}(E_{\mathrm{aff}}) Int ( E aff ) , the gauge U R ′ : R ′ → Φ ∗ Ψ ∗ R ′ U_{R'}:R'\to\Phi^*\Psi^*R' U R ′ : R ′ → Φ ∗ Ψ ∗ R ′ is closed and invertible. This proves essential surjectivity and completes the comparison. Regularity follows as in the theorem. □
This proof uses specified inverse data, not weak equivalence alone. The whole-plane Ψ , H \Psi,H Ψ , H need not restrict to an arbitrary open. For each O O O , the range-preserving Ψ O , H O \Psi_O,H_O Ψ O , H O satisfy the same whole-unit-relative hypotheses, so the same prism argument applies to Φ O \Phi_O Φ O . Thus the comparison acts on the raw coefficient categories. The subsequent differentiation and reconstruction argument identifies them with cohesive coefficients.
For fixed-base weak maps, use [12 Corollary 8.4] with the connective input of Theorem 2.1. For the supplied global resolutions, the direct route is [12 Corollary 8.5] . Those inverse homotopies preserve the entire base simplex, so their restriction and Hom-sheafification assertions apply directly. This differs from the open-by-open choices of affine Ψ O , H O \Psi_O,H_O Ψ O , H O . Hom-sheafification changes the morphism complexes while retaining the existing global objects. Together with Corollary 5.4 , these results connect the raw representation categories of the admissible global affine resolutions to that of K K K . Effective descent instead concerns reconstructing objects from local data; § 14 discusses this and the related coherence and holonomy questions.
Smooth coefficient integration
The general smooth construction is [12 Theorem 9.2] , using the universal higher de Rham kernel of Arias Abad–Schätz [AS13]. We recall the actual Taylor formulas and signs needed below: these specify the functor whose affine quasi-equivalence we will prove. The cited companion establishes convergence, gauge equivariance and smooth parameter dependence.
Theorem 6.1 (Coefficient integration). For every geometric fixed-base L ∞ L_\infty L ∞ -algebroid E E E in Section 2.1, there is a smooth strictly unital A ∞ A_\infty A ∞ -functor
IN E : Coh ( B E ) ⟶ Rep s m p e r f ( Int ( E ) ) . \operatorname{IN}_{E}:\operatorname{Coh}(B_{E}) \longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(\operatorname{Int}(E)). IN E : Coh ( B E ) ⟶ Rep sm perf ( Int ( E )) . It retains the underlying graded bundle and its internal differential, preserves the regular full subcategories, and commutes with open restriction and fixed-base infinity-morphisms.
This is [12 Theorem 9.2] . It uses globally bounded finite-projective coefficients and allows unbounded degrees of E E E . The construction depends on the CE algebra and full MC simplices, independently of Nuiten’s connective mapping-space comparison. For the affine algebroid, Theorem 10.4 will establish quasi-equivalence.
Full scalar extension and totalization
For σ ∈ Int ( E ) p ( U ) \sigma\in\operatorname{Int}(E)_{p}(U) σ ∈ Int ( E ) p ( U ) , let x σ : U × Δ p → M x_{\sigma}:U \times\Delta^{p} \to M x σ : U × Δ p → M be its base map and put Ω σ = Ω ∙ ( U × Δ p / U ) \Omega_{\sigma}=\Omega^\bullet(U \times\Delta^{p}/U) Ω σ = Ω ∙ ( U × Δ p / U ) . The full MC equations give a dg algebra map σ ∗ : B E → Ω σ \sigma^{*}:B_{E}\to\Omega_{\sigma} σ ∗ : B E → Ω σ . Scalar extension of M V = V ⊗ A B E M_{V}=V\otimes_{A}B_{E} M V = V ⊗ A B E uses
D σ ( m ⊗ ω ) = D V m ⊗ ω + ( − 1 ) ∣ m ∣ m ⊗ d Δ ω . D_{\sigma}(m\otimes\omega)=D_{V}m\otimes\omega+(-1)^{|m|}m\otimes d_{\Delta}\omega. D σ ( m ⊗ ω ) = D V m ⊗ ω + ( − 1 ) ∣ m ∣ m ⊗ d Δ ω .
It is a flat relative superconnection on the actual x σ ∗ V x_{\sigma}^{*}V x σ ∗ V and preserves Hom differentials and composition [12 Lemma 9.3] . Thus scalar extension retains the base map and anchor derivatives.
We use the forms-first braiding
b ( v ⊗ ω ) = ( − 1 ) ∣ v ∣ ∣ ω ∣ ω ⊗ v . b(v\otimes\omega)=(-1)^{|v||\omega|}\omega\otimes v. b ( v ⊗ ω ) = ( − 1 ) ∣ v ∣∣ ω ∣ ω ⊗ v .
It gives
D L ( ω v ) = d Δ ω v + ( − 1 ) ∣ ω ∣ ω D L ( v ) , ϕ L ( ω v ) = ( − 1 ) m ∣ ω ∣ ω ϕ L ( v ) D^{L}(\omega v)=d_{\Delta}\omega\,v+(-1)^{|\omega|}\omega D^{L}(v),\qquad\phi^{L}(\omega v)=(-1)^{m|\omega|}\omega\phi^{L}(v) D L ( ω v ) = d Δ ω v + ( − 1 ) ∣ ω ∣ ω D L ( v ) , ϕ L ( ω v ) = ( − 1 ) m ∣ ω ∣ ω ϕ L ( v )
for a degree-m m m homogeneous morphism. The right-module convention remains the one in Section 2.3.
This is [12 Lemma 9.4 and equation (9.1)] .
Lemma 6.2 (All-degree target dictionary). Let α \alpha α have total degree m m m in the raw target. The transformation
( T m α ) p = ( − 1 ) p m + p ( p + 1 ) / 2 α p (20) (T_{m}\alpha)_{p}=(-1)^{pm+p(p+1)/2}\alpha_{p} \tag*{(20)} ( T m α ) p = ( − 1 ) p m + p ( p + 1 ) /2 α p (20) identifies its dg category with the convention of [AS13, Definition 2.2 and Remark 2.4, pp. 6–7], where object structures transform by F p = ( − 1 ) p ( p − 1 ) / 2 r p F_{p}=(-1)^{p(p-1)/2}r_{p} F p = ( − 1 ) p ( p − 1 ) /2 r p . It preserves units, normalization and edge quasi-isomorphisms.
The kernel, matrix signs and twisting formulas
Use ∣ s a ∣ = ∣ a ∣ − 1 |sa|=|a|-1 ∣ s a ∣ = ∣ a ∣ − 1 , with suspended dg coderivation
b 1 ( s a ) = − s ( d a ) , b 2 ( s a ⊗ s b ) = ( − 1 ) ∣ a ∣ s ( a b ) . (21) b_1(sa)=-s(da),\qquad b_2(sa\otimes sb)=(-1)^{|a|}s(ab). \tag*{(21)} b 1 ( s a ) = − s ( d a ) , b 2 ( s a ⊗ s b ) = ( − 1 ) ∣ a ∣ s ( ab ) . (21)
The unsuspended Taylor convention is F ˉ k = s F k ( s − 1 ) ⊗ k \bar{F}_k=sF_k(s^{-1})^{\otimes k} F ˉ k = s F k ( s − 1 ) ⊗ k , including graded tensor signs. These are [12 equation (9.2)] . Both coderivation terms are the negatives of [AS13, Remark 3.1]; negating both source and target coderivations leaves the bar-map equation unchanged.
For a smooth manifold N N N , write Ω = Ω ∙ ( N ) \Omega=\Omega^\bullet(N) Ω = Ω ∙ ( N ) for its de Rham algebra and C n o r m = C n o r m ∙ ( S s m ( N ) ) C_{\mathrm{norm}}=C^\bullet_{\mathrm{norm}}(S_{\mathrm{sm}}(N)) C norm = C norm ∙ ( S sm ( N )) for normalized smooth singular cochains. The scalar kernel below is the cited natural A ∞ A_\infty A ∞ -map between these algebras. In the present application it is evaluated on the simplex-direction forms in Ω σ \Omega_\sigma Ω σ , with the external parameter held fixed. Jointly smooth dependence on that parameter is supplied by the stated all-derivative convergence and gluing results. No differential in the external parameter is included in d Δ d_\Delta d Δ .
Let ψ \psi ψ be the kernel of [AS13, Definition 3.23 and Theorem 3.25] from ( Ω , − d , ∧ ) (\Omega,-d,\wedge) ( Ω , − d , ∧ ) to ordinary smooth singular cochains. For scalar forms put τ ( a ) = ( − 1 ) ∣ a ∣ a \tau(a)=(-1)^{|a|}a τ ( a ) = ( − 1 ) ∣ a ∣ a and fix the uncurved A ∞ A_\infty A ∞ -morphism
Γ = ψ τ : ( Ω , d , ∧ ) ⟶ ( C n o r m , d f u l l , ∪ ) . \Gamma=\psi\tau:(\Omega,d,\wedge)\longrightarrow(C_{\mathrm{norm}},d_{\mathrm{full}},\cup). Γ = ψ τ : ( Ω , d , ∧ ) ⟶ ( C norm , d full , ∪ ) .
Thus Γ ˉ N ( s a 1 , … , s a N ) = ( − 1 ) ∑ i ∣ a i ∣ ψ N ( s a 1 , … , s a N ) \bar{\Gamma}_N(sa_1,\ldots,sa_N)=(-1)^{\sum_i|a_i|}\psi_N(sa_1,\ldots,sa_N) Γ ˉ N ( s a 1 , … , s a N ) = ( − 1 ) ∑ i ∣ a i ∣ ψ N ( s a 1 , … , s a N ) , with Γ ˉ N = s Γ N ( s − 1 ) ⊗ N \bar{\Gamma}_N=s\Gamma_N(s^{-1})^{\otimes N} Γ ˉ N = s Γ N ( s − 1 ) ⊗ N . The completed kernel is natural under smooth pullback. For a p p p -form its unary term is
Γ 1 ( a ) ( σ ) = ∫ Δ p σ ∗ a . \Gamma_1(a)(\sigma)=\int_{\Delta^p}\sigma^*a. Γ 1 ( a ) ( σ ) = ∫ Δ p σ ∗ a .
In form degree zero this is vertex evaluation. Every higher Γ N \Gamma_N Γ N , N > 1 N>1 N > 1 , vanishes if any input has ordinary form degree zero. Its outputs are normalized [AS13, Proposition 3.26].
For matrices identify s ( Ω ⊗ End V ) = ( s Ω ) ⊗ End V s(\Omega\otimes\operatorname{End}V)=(s\Omega)\otimes\operatorname{End}V s ( Ω ⊗ End V ) = ( s Ω ) ⊗ End V . On elementary inputs s ω a ⊗ T a s\omega_a\otimes T_a s ω a ⊗ T a with constant graded matrices, the suspended extension collects the scalar forms with sign
( − 1 ) ∑ a < b ∣ T a ∣ ( ∣ ω b ∣ − 1 ) , (22) (-1)^{\sum_{a<b}|T_a|(|\omega_b|-1)}, \tag*{(22)} ( − 1 ) ∑ a < b ∣ T a ∣ ( ∣ ω b ∣ − 1 ) , (22)
and multiplies the matrices in their original order. The corresponding unsuspended formula is
Γ ˉ N V ( ω 1 ⊗ T 1 , … , ω N ⊗ T N ) = ( − 1 ) ∑ a < b ∣ T a ∣ ∣ ω b ∣ Γ N ( ω 1 , … , ω N ) ⊗ T 1 ⋯ T N . (23) \bar{\Gamma}^{V}_N(\omega_1\otimes T_1,\ldots,\omega_N\otimes T_N)=(-1)^{\sum_{a<b}|T_a||\omega_b|}\Gamma_N(\omega_1,\ldots,\omega_N)\otimes T_1\cdots T_N. \tag*{(23)} Γ ˉ N V ( ω 1 ⊗ T 1 , … , ω N ⊗ T N ) = ( − 1 ) ∑ a < b ∣ T a ∣∣ ω b ∣ Γ N ( ω 1 , … , ω N ) ⊗ T 1 ⋯ T N . (23)
These are exactly [12 equations (9.3)–(9.4)] , the forms-first matrix extension of [AS13, Proposition A.17]. For composable morphisms use the corresponding block of the finite direct sum of coefficient bundles. This matrix extension is understood in every formula below.
In a simplexwise frame write D i L = d + A i D_i^L=d+A_i D i L = d + A i . For ϕ a : V a − 1 → V a \phi_a:V_{a-1}\to V_a ϕ a : V a − 1 → V a of degree m a m_a m a , the actual twisted Taylor expression is [12 equation (9.5)]
I k ( ϕ k , … , ϕ 1 ) = ∑ j 0 , … , j k ≥ 0 ( − 1 ) χ Γ k + ∑ j i ( A k ⊗ j k , ϕ k , A k − 1 ⊗ j k − 1 , … , ϕ 1 , A 0 ⊗ j 0 ) , (24) \mathcal{I}_k(\phi_k,\ldots,\phi_1)=\sum_{j_0,\ldots,j_k\geq0}(-1)^{\chi\Gamma_k+\sum j_i}(A_k^{\otimes j_k},\phi_k,A_{k-1}^{\otimes j_{k-1}},\ldots,\phi_1,A_0^{\otimes j_0}), \tag*{(24)} I k ( ϕ k , … , ϕ 1 ) = j 0 , … , j k ≥ 0 ∑ ( − 1 ) χ Γ k + ∑ j i ( A k ⊗ j k , ϕ k , A k − 1 ⊗ j k − 1 , … , ϕ 1 , A 0 ⊗ j 0 ) , (24)
χ = ∑ a = 1 k ( m a − 1 ) ∑ ℓ < a j ℓ , k ≥ 1. \chi=\sum_{a=1}^{k}(m_a-1)\sum_{\ell<a}j_\ell,\qquad k\geq1. χ = a = 1 ∑ k ( m a − 1 ) ℓ < a ∑ j ℓ , k ≥ 1.
The suspended expression inserts ( s A i ) ⊗ j i (sA_i)^{\otimes j_i} ( s A i ) ⊗ j i around s ϕ k , … , s ϕ 1 s\phi_k,\ldots,s\phi_1 s ϕ k , … , s ϕ 1 without an additional sign. Only the diagonal A i A_i A i are MC elements: the off-diagonal morphism inputs need not be closed or MC.
For an object the formula is [12 equation (9.6)]
a V = ∑ N ≥ 1 Γ N ( A , … , A ) , r V = e + a V , e 1 = Id , e p = 0 ( p ≠ 1 ) . (25) a_V=\sum_{N\geq1}\Gamma_N(A,\ldots,A),\qquad r_V=e+a_V,\qquad e_1=\operatorname{Id},\quad e_p=0\ (p\neq1). \tag*{(25)} a V = N ≥ 1 ∑ Γ N ( A , … , A ) , r V = e + a V , e 1 = Id , e p = 0 ( p = 1 ) . (25)
Here e e e is a one-cochain in the temporary constant frame, not a globally specified map between distinct base fibers. The unit shift is governed by [12 equation (9.7)] :
d f u l l α = δ α + e W ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e V . (26) d_{\mathrm{full}}\alpha=\delta\alpha+e_W\star\alpha-(-1)^{|\alpha|}\alpha\star e_V. \tag*{(26)} d full α = δ α + e W ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e V . (26)
The endpoint terms complete the inner differential to the full coboundary, with final-face sign ( − 1 ) p + 1 (-1)^{p+1} ( − 1 ) p + 1 on a p p p -cochain. The displayed outputs are already in our ordinary-cochain totalization. No further factor T T T is applied to them.
Smoothness, degree control and naturality
For fixed output simplex degree, Taylor arity, homogeneous inputs and a compact subset of a local smooth family-frame chart, the object and Taylor series converge locally uniformly with every external-parameter derivative [12 Lemma 9.5] . In a nonzero higher kernel word every input has positive ordinary form degree and
∑ i n p u t s ( deg f o r m − 1 ) = p − 1. (27) \sum_{\mathrm{inputs}}(\deg_{\mathrm{form}}-1)=p-1. \tag*{(27)} inputs ∑ ( deg form − 1 ) = p − 1. (27)
This is [12 equation (9.8)] . One-form connection insertions can be arbitrarily numerous; the general twisting series may therefore be infinite. Structure inputs of form degree at least two lower internal degree, whereas the distinguished morphism slots have their specified arbitrary internal degrees. Matrix products retain the input order in (6.4); this is the filtration control used in finite affine reconstruction.
The outputs are intrinsic smooth two-ended bundle maps, independent of temporary family frames [12 Lemma 9.6] . With endpoint identifications understood, their evaluation is [12 equation (9.10)] :
( I N E , k ( ϕ k , … , ϕ 1 ) ) p ( σ ) = I k ( ϕ k , σ , … , ϕ 1 , σ ) ( Id Δ p ) . (28) (\mathrm{IN}_{E,k}(\phi_k,\ldots,\phi_1))_p(\sigma)=\mathcal{I}_k(\phi_{k,\sigma},\ldots,\phi_{1,\sigma})(\operatorname{Id}_{\Delta^p}). \tag*{(28)} ( IN E , k ( ϕ k , … , ϕ 1 ) ) p ( σ ) = I k ( ϕ k , σ , … , ϕ 1 , σ ) ( Id Δ p ) . (28)
The k k k th Taylor map has degree 1 − k 1-k 1 − k ; this p p p -component has internal degree ∑ a ∣ ϕ a ∣ + 1 − k − p \sum_a|\phi_a|+1-k-p ∑ a ∣ ϕ a ∣ + 1 − k − p . Bounded coefficient amplitudes give finite cochain support in each total degree, without changing the normal-convergence assertion for the kernel words.
At vertices, r 0 = D [ 0 ] r_0=D^{[0]} r 0 = D [ 0 ] , the unary morphism is ϕ [ 0 ] \phi^{[0]} ϕ [ 0 ] , and higher Taylor maps vanish. Strict units mean I N E , 1 ( 1 ) = 1 \mathrm{IN}_{E,1}(1)=1 IN E , 1 ( 1 ) = 1 and I N E , k ( … , 1 , … ) = 0 \mathrm{IN}_{E,k}(\ldots,1,\ldots)=0 IN E , k ( … , 1 , … ) = 0 for k > 1 k>1 k > 1 . Edges are chain isomorphisms of the vertex complexes. For a fixed-base infinity-morphism F : E ⇝ E ′ F:E\rightsquigarrow E' F : E ⇝ E ′ ,
I N E F ∗ = Int ( F ) ∗ I N E ′ \mathrm{IN}_{E}F^*=\operatorname{Int}(F)^*\mathrm{IN}_{E'} IN E F ∗ = Int ( F ) ∗ IN E ′
in every arity, under the canonical scalar-extension identifications. Open restriction uses simplices whose entire base image lies in the open. These are properties of the actual functor in [12 Theorem 9.2] .
Postcomposition and the affine higher-isotropy test
Corollary 6.3. For the affine resolution the actual section in Theorem A.10 gives a global functor
I N E a f f K = Ψ ∗ I N E a f f : Coh ( B E a f f ) ⟶ Rep s m p e r f ( y K ) . \mathrm{IN}^{K}_{E_{\mathrm{aff}}}=\Psi^*\mathrm{IN}_{E_{\mathrm{aff}}}:\operatorname{Coh}(BE_{\mathrm{aff}})\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(y_K). IN E aff K = Ψ ∗ IN E aff : Coh ( B E aff ) ⟶ Rep sm perf ( y K ) . It is a smooth strictly unital A ∞ A_\infty A ∞ -functor and preserves regular coefficients.
Proof. Postcomposition uses the already constructed dg functor Ψ ∗ \Psi^* Ψ ∗ , with unchanged vertex complexes. The global raw target change is justified by Corollary 5.4 .
For each ordinary open O O O , postcomposition with the separately constructed Ψ O \Psi_O Ψ O likewise gives I O = Ψ O ∗ I N E O \mathrm{I}_O=\Psi_O^*\mathrm{IN}_{E_O} I O = Ψ O ∗ IN E O on y K [ O ] y_K[O] y K [ O ] . The sections are chosen separately for each open; strict restriction-naturality is a property of the intrinsic forward functor I N \mathrm{IN} IN itself.
Proposition 6.4 (A coefficient detecting the affine two-form). Let V = R v 0 ⊕ R v 1 V=\mathbb{R}v_0\oplus\mathbb{R}v_1 V = R v 0 ⊕ R v 1 over R 2 \mathbb{R}^2 R 2 with ∣ v 0 ∣ = 0 |v_0|=0 ∣ v 0 ∣ = 0 , ∣ v 1 ∣ = 1 |v_1|=1 ∣ v 1 ∣ = 1 , and
D ( v 0 ) = 0 , D ( v 1 ) = v 0 η . D(v_0)=0,\qquad D(v_1)=v_0\eta. D ( v 0 ) = 0 , D ( v 1 ) = v 0 η . Set N ( v 1 ) = v 0 N(v_1)=v_0 N ( v 1 ) = v 0 , N ( v 0 ) = 0 N(v_0)=0 N ( v 0 ) = 0 . Its integrated representation has
r 0 = 0 , r 1 = 1 , r 2 ( σ ) = ( ∫ [ 012 ] β ) N , r p = 0 ( p ≥ 3 ) . r_0=0,\qquad r_1=1,\qquad r_2(\sigma)=\left(\int_{[012]}\beta\right)N,\qquad r_p=0\quad(p\geq3). r 0 = 0 , r 1 = 1 , r 2 ( σ ) = ( ∫ [ 012 ] β ) N , r p = 0 ( p ≥ 3 ) . On K K K the two-component is exactly − h 012 N -h_{012}N − h 012 N , including at the singular origin.
Proof. The coefficient is flat because Q η = 0 Q\eta=0 Q η = 0 and N 2 = 0 N^2=0 N 2 = 0 . After full scalar extension and braiding its form is β ⊗ N \beta\otimes N β ⊗ N . Every word with two or more such inputs vanishes by N 2 = 0 N^2=0 N 2 = 0 . The actual linear kernel Γ 1 = ∫ \Gamma_1=\int Γ 1 = ∫ therefore fixes the displayed sign and normalization; the identity one-cochain supplies r 1 r_1 r 1 . The degree-three object equation is
( − h 123 + h 023 − h 013 + h 012 ) N = 0. (-h_{123}+h_{023}-h_{013}+h_{012})N=0. ( − h 123 + h 023 − h 013 + h 012 ) N = 0. where the first and last terms come from the two unit products, not the inner differential alone. Degree four uses N 2 = 0 N^2=0 N 2 = 0 . Since Φ Ψ = 1 \Phi\Psi=1 ΦΨ = 1 and h = − ∫ β h=-\int\beta h = − ∫ β , the formula on K K K follows. At the origin the exponential boundary e h Z 0 e^{hZ_0} e h Z 0 is the identity, but this two-component still varies linearly with the retained scalar h h h .
MC alone would also allow r 2 = − c h N r_2=-chN r 2 = − c h N for any scalar c c c , including zero. It is the linear integration kernel, not that equation alone, which determines c = 1 c=1 c = 1 . For these affine functors, the all-open Hom comparison and nonregular reconstruction establish quasi-equivalence in Theorem 10.4 . General-algebroid reconstruction and effective descent are discussed in § 14 .
Range-preserving integration on arbitrary opens
Let O ⊆ R 2 O \subseteq\mathbb{R}^{2} O ⊆ R 2 be an ordinary open set and write E O = E a f f ∣ O E_{O}=E_{\mathrm{aff}}|_{O} E O = E aff ∣ O . We construct the finite model and the inverse-homotopy data separately on each O O O . The interior base maps of the global section in Theorem A.10 need not stay in a prescribed open set, so the construction must preserve the range in O O O .
When the raw coefficient notation is applied to a simplicial manifold, we suppress its levelwise Yoneda embedding: Rep s m p e r f ( K [ O ] ) \operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(K[O]) Rep sm perf ( K [ O ]) means Rep s m p e r f ( y K [ O ] ) \operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(yK[O]) Rep sm perf ( y K [ O ]) .
Write G = Aff + ( 1 ) G=\operatorname{Aff}_{+}(1) G = Aff + ( 1 ) , with identity e = ( 1 , 0 ) e=(1,0) e = ( 1 , 0 ) , and use the root coordinates of (7) . For a simplex of K n K_{n} K n , denote its vertices by m i = ( x i , y i ) m_{i}=(x_{i},y_{i}) m i = ( x i , y i ) . All y i y_{i} y i agree. Define
K [ O ] n = { k ∈ K n : [ min i x i , max i x i ] × { y 0 } ⊆ O } . (29) K[O]_{n}=\left\{k\in K_{n}:\left[\min_{i}x_{i},\max_{i}x_{i}\right]\times\{y_{0}\}\subseteq O\right\}. \tag*{(29)} K [ O ] n = { k ∈ K n : [ i min x i , i max x i ] × { y 0 } ⊆ O } . (29)
The notation refers to the whole closed horizontal hull, not merely to membership of the vertices in O O O .
Proposition 7.1 (The open finite model). The spaces K [ O ] n K[O]_{n} K [ O ] n form a simplicial open submanifold of K K K . They constitute a Lie 2-groupoid with the same triangle and tetrahedral formulas as K K K . For m = ( x , y ) ∈ O m=(x,y)\in O m = ( x , y ) ∈ O , its source fibre is
D m = { ( a , b ) : a > 0 , [ min { x , a x + b y } , max { x , a x + b y } ] × { y } ⊆ O } . (30) D_{m}=\left\{(a,b):a>0,\ [\min\{x,ax+by\},\max\{x,ax+by\}]\times\{y\}\subseteq O\right\}. \tag*{(30)} D m = { ( a , b ) : a > 0 , [ min { x , a x + b y } , max { x , a x + b y }] × { y } ⊆ O } . (30) The total space D = { ( m , g ) : m ∈ O , g ∈ D m } D=\{(m,g):m\in O,\ g\in D_{m}\} D = {( m , g ) : m ∈ O , g ∈ D m } is open, and every D m D_{m} D m is convex and contains e e e .
Proof. A compact horizontal segment contained in O O O remains contained in O O O after a sufficiently small perturbation of its endpoints and its height. Thus (7.1) defines an open subset of K n K_{n} K n . Faces delete vertices and degeneracies repeat them, so these operations preserve the hull condition.
For fixed m = ( x , y ) m=(x,y) m = ( x , y ) , let I y ( m ) I_{y}(m) I y ( m ) be the component of O ∩ ( R × { y } ) O\cap(\mathbb{R}\times\{y\}) O ∩ ( R × { y }) containing m m m , identified with an open interval in the x x x -coordinate. Then
D m = { ( a , b ) : a > 0 , a x + b y ∈ I y ( m ) } . D_{m}=\{(a,b):a>0,\ ax+by\in I_{y}(m)\}. D m = {( a , b ) : a > 0 , a x + b y ∈ I y ( m )} .
This is an intersection of an open half-plane with the inverse image of an interval under an affine function. It is convex and contains ( 1 , 0 ) (1,0) ( 1 , 0 ) . Openness of the total space follows from compactness of the parameter interval in the map
( m , g , t ) ⟼ ( 1 − t ) m + t ( g m ) , 0 ≤ t ≤ 1. (m,g,t)\longmapsto(1-t)m+t(gm),\qquad0\leq t\leq1. ( m , g , t ) ⟼ ( 1 − t ) m + t ( g m ) , 0 ≤ t ≤ 1.
The degree-one source and target maps are restrictions of submersions to an open arrow manifold; they remain surjective submersions because units are present. For a horn of degree at least two, its one-skeleton connects all its vertices. The retained edges put those vertices in one horizontal component of O O O . Every filler in K K K has the same vertices, and hence belongs to K [ O ] K[O] K [ O ] . The original degree-two horn fibres are R \mathbb{R} R , and the higher horn fillers are unique by Proposition 3.1 . Restriction to the open hull condition preserves these assertions. ▫
A jointly smooth complete chart of the source fibres
The following chart is used both in the open integration section and in the relative de Rham contraction below. Its dependence on O O O is fixed once and for all.
Lemma 7.2 (Complete based fibre charts). There are jointly smooth diffeomorphisms
ψ m : D m ⟶ R 2 , ψ m ( e ) = 0. \psi_m:D_m\longrightarrow\mathbb{R}^{2},\qquad\psi_m(e)=0. ψ m : D m ⟶ R 2 , ψ m ( e ) = 0. They are defined for every m ∈ O m\in O m ∈ O , including the origin when it belongs to O O O .
Proof. Translate the group coordinates by writing z = ( a − 1 , b ) z=(a-1,b) z = ( a − 1 , b ) . Then D D D is an open subset of R m , z 4 \mathbb{R}^{4}_{m,z} R m , z 4 containing the zero section over O O O .
Choose a countable family of nonnegative functions f j ∈ C ∞ ( D ) f_j\in C^\infty(D) f j ∈ C ∞ ( D ) whose positive sets cover D D D , and extend them by zero to R 4 \mathbb{R}^{4} R 4 . Choose c j > 0 c_j>0 c j > 0 such that
c j max ∣ α ∣ ≤ j ∥ ∂ α f j ∥ ∞ ≤ 2 − j . c_j\max_{|\alpha|\leq j}\|\partial^\alpha f_j\|_\infty\leq2^{-j}. c j ∣ α ∣ ≤ j max ∥ ∂ α f j ∥ ∞ ≤ 2 − j .
For every fixed multiindex α \alpha α , the differentiated tail of ∑ j c j f j \sum_j c_jf_j ∑ j c j f j converges uniformly. Thus
F = ∑ j = 1 ∞ c j f j F=\sum_{j=1}^{\infty}c_jf_j F = j = 1 ∑ ∞ c j f j
is smooth with all derivatives, is positive exactly on D D D , and vanishes outside D D D . Set
ρ ~ = F 1 + F . \widetilde{\rho}=\frac{F}{1+F}. ρ = 1 + F F .
Then 0 ≤ ρ ~ < 1 0\leq\widetilde{\rho}<1 0 ≤ ρ < 1 and its positive locus is D D D .
Choose a locally finite partition of unity { λ i } \{\lambda_i\} { λ i } on O O O and constants ε i > 0 \varepsilon_i>0 ε i > 0 such that
{ m } × B ‾ 4 ε i ( 0 ) ⊆ D for m ∈ supp λ i . \{m\}\times\overline{B}_{4\varepsilon_i}(0)\subseteq D\quad\text{for }m\in\operatorname{supp}\lambda_i. { m } × B 4 ε i ( 0 ) ⊆ D for m ∈ supp λ i .
Put δ ( m ) = ∑ i λ i ( m ) ε i \delta(m)=\sum_i\lambda_i(m)\varepsilon_i δ ( m ) = ∑ i λ i ( m ) ε i . At each m m m , only finitely many terms are nonzero. If ε i \varepsilon_i ε i is maximal among the active terms, then 3 δ ( m ) ≤ 3 ε i < 4 ε i 3\delta(m)\leq3\varepsilon_i<4\varepsilon_i 3 δ ( m ) ≤ 3 ε i < 4 ε i ; consequently
{ m } × B ‾ 3 δ ( m ) ( 0 ) ⊆ D . \{m\}\times\overline{B}_{3\delta(m)}(0)\subseteq D. { m } × B 3 δ ( m ) ( 0 ) ⊆ D .
The function δ \delta δ is smooth and strictly positive.
Choose a smooth cutoff χ ( m , z ) \chi(m,z) χ ( m , z ) , depending on ∥ z ∥ 2 / δ ( m ) 2 \lVert z\rVert^2/\delta(m)^2 ∥ z ∥ 2 / δ ( m ) 2 , equal to one for ∥ z ∥ ≤ 2 δ ( m ) \lVert z\rVert\leq2\delta(m) ∥ z ∥ ≤ 2 δ ( m ) and zero for ∥ z ∥ ≥ 3 δ ( m ) \lVert z\rVert\geq3\delta(m) ∥ z ∥ ≥ 3 δ ( m ) . Define
ρ ( m , z ) = χ ( m , z ) + ( 1 − χ ( m , z ) ) ρ ~ ( m , z ) , X ( m , z ) = ρ ( m , z ) z . \rho(m,z)=\chi(m,z)+(1-\chi(m,z))\widetilde{\rho}(m,z),\qquad X(m,z)=\rho(m,z)z. ρ ( m , z ) = χ ( m , z ) + ( 1 − χ ( m , z )) ρ ( m , z ) , X ( m , z ) = ρ ( m , z ) z .
The vector field is vertical, smooth on O × R 2 O\times\mathbb{R}^{2} O × R 2 , positive radially on D ∖ ( O × { 0 } ) D\setminus(O\times\{0\}) D ∖ ( O × { 0 }) , zero outside D D D , and equal to z z z near the zero section. The estimate ∥ X ( m , z ) ∥ ≤ ∥ z ∥ \lVert X(m,z)\rVert\leq\lVert z\rVert ∥ X ( m , z )∥ ≤ ∥ z ∥ makes its flow ϕ t \phi_t ϕ t complete for positive and negative time.
Each ray meets D m D_m D m in an interval starting at zero. On that interval the radial equation is
r ˙ = ρ ( m , r ω ) r . \dot r=\rho(m,r\omega)r. r ˙ = ρ ( m , r ω ) r .
Backward flow tends to zero. Forward flow tends to the endpoint of the interval, finite or infinite: a positive interior limit would contradict positivity of ρ \rho ρ . A boundary point cannot be reached in finite time, since it is stationary and uniqueness would then force the original trajectory to be stationary.
For v ≠ 0 v \ne0 v = 0 , define
Ξ ( m , v ) = φ log ( ∥ v ∥ / δ ( m ) ) ( m , δ ( m ) v ∥ v ∥ ) , Ξ ( m , 0 ) = ( m , 0 ) . \Xi(m,v)=\varphi_{\log(\lVert v\rVert/\delta(m))}\left(m,\delta(m)\frac{v}{\lVert v\rVert}\right),\qquad\Xi(m,0)=(m,0). Ξ ( m , v ) = φ l o g (∥ v ∥ / δ ( m )) ( m , δ ( m ) ∥ v ∥ v ) , Ξ ( m , 0 ) = ( m , 0 ) .
When ∥ v ∥ ≤ 2 δ ( m ) \lVert v\rVert\le2\delta(m) ∥ v ∥ ≤ 2 δ ( m ) , this formula is exactly Ξ ( m , v ) = ( m , v ) \Xi(m,v)=(m,v) Ξ ( m , v ) = ( m , v ) , because the relevant trajectory remains where ρ = 1 \rho=1 ρ = 1 . Thus it is smooth at the zero section. On each ray it is strictly increasing and maps ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) onto the entire radial interval of D m D_m D m . Its radial derivative is positive, and its angular derivative has nonzero tangential part. Hence it is a bijective local diffeomorphism over O O O , with smooth inverse. Translating back from z z z to ( a , b ) (a,b) ( a , b ) gives ψ m \psi_m ψ m .
The chart determines the selected edge path
γ m , g ( t ) = ψ m − 1 ( t ψ m ( g ) ) , 0 ≤ t ≤ 1. (31) \gamma_{m,g}(t)=\psi_m^{-1}(t\psi_m(g)),\qquad0\le t\le1. \tag*{(31)} γ m , g ( t ) = ψ m − 1 ( t ψ m ( g )) , 0 ≤ t ≤ 1. (31)
It stays in D m D_m D m , depends jointly smoothly on ( m , g , t ) (m,g,t) ( m , g , t ) , and extends smoothly past both endpoints. The chart is chosen separately for each open.
Lifting with the current root
We use the smooth face extension, Whitney–radial primitive, and ordered split-locus extension of Lemmas A.2, A.3, and A.7. These are linear operations on the displayed form or function residuals. The following argument specifies how they interact with the moving root and the open condition.
For a parameter sheaf P P P , a smooth family of simplex q q q -forms means a morphism from P P P to the sheaf
U ⟼ Ω q ( U × Δ n / U ) , U\longmapsto\Omega^q(U\times\Delta^n/U), U ⟼ Ω q ( U × Δ n / U ) ,
defined on plots. The sheaf P P P may be nonrepresentable. The ordered extension of such vector-valued residuals is the construction of Lemma A.7.
Lemma 7.3 (Range-preserving split-locus lifting). Let P P P be a smooth parameter sheaf and let k : P → K [ O ] n k:P\to K[O]_n k : P → K [ O ] n have compatible prescribed full-MC faces over its faces. Suppose compatible full lifts are also prescribed on finitely many split loci P 1 , … , P N ⊆ P P_1,\ldots,P_N\subseteq P P 1 , … , P N ⊆ P , with retractions r j r_j r j satisfying
r j ( P i ) ⊆ P i ∩ P j ( i < j ) . r_j(P_i)\subseteq P_i\cap P_j\qquad(i<j). r j ( P i ) ⊆ P i ∩ P j ( i < j ) . Then the data admit a jointly smooth full-MC lift over k k k whose entire base simplex lies in O O O . The parameter retractions need not preserve k k k or its root.
Proof. For n = 0 n=0 n = 0 , the lift is forced. Suppose first that n ≥ 2 n\ge2 n ≥ 2 , and let m 0 = m 0 ( k ) m_0=m_0(k) m 0 = m 0 ( k ) . Put c = − h ( k ) c=-h(k) c = − h ( k ) . The tetrahedral identity says δ c = 0 \delta c=0 δ c = 0 . Let β ∂ \beta_{\partial} β ∂ be the prescribed boundary two-form. The baseline
β 0 = W ( c ) + d Ext n ( D ∂ ( β ∂ − W ( c ) ∣ ∂ ) ) \beta^0=W(c)+d\,\operatorname{Ext}_n\left(D_{\partial}\left(\beta_{\partial}-W(c)|_{\partial}\right)\right) β 0 = W ( c ) + d Ext n ( D ∂ ( β ∂ − W ( c ) ∣ ∂ ) )
is closed, has the prescribed boundary, and has triangle periods c c c , by Lemma A.4. On P j P_j P j , let β j \beta_j β j be the prescribed two-form and set
γ j = β j − β 0 ∣ P j , η j = D n γ j . \gamma_j=\beta_j-\beta^0|_{P_j},\qquad\eta_j=D_n\gamma_j. γ j = β j − β 0 ∣ P j , η j = D n γ j .
The form γ j \gamma_j γ j is closed and has zero triangle periods, so d η j = γ j d\eta_j=\gamma_j d η j = γ j . Face naturality of D n D_n D n makes η j \eta_j η j vanish on the whole boundary. Extend the compatible η j \eta_j η j by the ordered formula of Lemma A.7, within the linear space of one-forms vanishing on the boundary, and denote the extension by η \eta η . Then
β = β 0 + d η \beta=\beta^0+d\eta β = β 0 + d η
recovers all prescribed two-forms and faces. Its triangle periods remain c c c : every edge is contained in the boundary, so the exact correction has zero triangle integrals.
Set u = H v 0 β u=H_{v_0}\beta u = H v 0 β . Then d u = β du=\beta d u = β and u u u vanishes on every root edge. Flatten each prescribed face by
A i ′ = A i − Z v i u ∣ F i . A_i'=A_i-Z_{v_i}u|_{F_i}. A i ′ = A i − Z v i u ∣ F i .
Lemma A.1 gives compatible flat potentials on faces containing v 0 v_0 v 0 , normalized to one there. On the opposite face use the potential with value a 1 = g 01 ( k ) a_1=g_{01}(k) a 1 = g 01 ( k ) at v 1 v_1 v 1 . Its compatibility at v 2 v_2 v 2 is the exact calculation
∫ [ 12 ] u = − h 012 , g 0 ( v 2 ) = g 12 a 1 exp ( h 012 Z m 0 ) = a 2 . \int_{[12]}u=-h_{012},\qquad g_0(v_2)=g_{12}a_1\exp(h_{012}Z_{m_0})=a_2. ∫ [ 12 ] u = − h 012 , g 0 ( v 2 ) = g 12 a 1 exp ( h 012 Z m 0 ) = a 2 .
Uniqueness of the flat potential then gives agreement on all face intersections, including the opposite-face intersection in degree two.
The resulting boundary potential g ∂ g_{\partial} g ∂ satisfies σ ∂ = g ∂ m 0 \sigma_{\partial}=g_{\partial}m_0 σ ∂ = g ∂ m 0 . Every prescribed face has connected base image in a horizontal component of O O O containing its vertices. Those vertices lie in the component of m 0 m_0 m 0 by the hull condition. Therefore g ∂ g_{\partial} g ∂ takes values in D m 0 D_{m_0} D m 0 .
Extend the chart-valued boundary function, using the current root:
g 0 = ψ m 0 − 1 ( Ext n ( ψ m 0 ( g ∂ ) ) ) . g^0=\psi_{m_0}^{-1}\bigl(\operatorname{Ext}_n(\psi_{m_0}(g_{\partial}))\bigr). g 0 = ψ m 0 − 1 ( Ext n ( ψ m 0 ( g ∂ )) ) .
On P j P_j P j , flatten the prescribed lift with its recovered u u u and take its normalized potential g j g_j g j . It agrees with g ∂ g_{\partial} g ∂ on the boundary and lies in D m 0 D_{m_0} D m 0 . Thus
w j = ψ m 0 ( g j ) − ψ m 0 ( g 0 ) ∣ P j w_j=\psi_{m_0}(g_j)-\psi_{m_0}(g^0)|_{P_j} w j = ψ m 0 ( g j ) − ψ m 0 ( g 0 ) ∣ P j
is a compatible vector-valued residual vanishing on the boundary. Extend these residuals by the ordered split-locus formula, obtaining w w w , and set
g = ψ m 0 − 1 ( ψ m 0 ( g 0 ) + w ) . g=\psi_{m_0}^{-1}\bigl(\psi_{m_0}(g^0)+w\bigr). g = ψ m 0 − 1 ( ψ m 0 ( g 0 ) + w ) .
The retractions act on the vector residuals only. The inverse chart in this formula always uses the current m 0 ( k ) m_0(k) m 0 ( k ) , even if a retraction changes the target simplex. Since the chart has range all of R 2 \mathbb{R}^2 R 2 , g g g remains in D m 0 D_{m_0} D m 0 .
Define
σ = g m 0 , A = d g g − 1 + Z σ u . \sigma=gm_0,\qquad A=dg\,g^{-1}+Z_{\sigma}u. σ = g m 0 , A = d g g − 1 + Z σ u .
Reverse flat correction gives all the full-MC equations (8) . The base lies in O O O because g ∈ D m 0 g\in D_{m_0} g ∈ D m 0 . Root-edge holonomies are the prescribed a i a_i a i , and triangle periods are − h i j k -h_{ijk} − h ij k . The root coordinate formulas therefore give Φ ( σ , A , β ) = k \Phi(\sigma,A,\beta)=k Φ ( σ , A , β ) = k exactly. The construction recovers every face and split-locus lift.
For n = 1 n=1 n = 1 , use β = u = 0 \beta=u=0 β = u = 0 and the baseline g 0 ( t ) = ψ m 0 − 1 ( t ψ m 0 ( a 1 ) ) g^0(t)=\psi_{m_0}^{-1}(t\psi_{m_0}(a_1)) g 0 ( t ) = ψ m 0 − 1 ( t ψ m 0 ( a 1 )) . Each prescribed interval lift has a normalized flat potential. Extend its chart-valued difference from $g^0 within the functions vanishing at the two endpoints, and reconstruct by σ = g m 0 \sigma= gm_0 σ = g m 0 , A = d g g − 1 A = \mathrm{d}g g^{-1} A = d g g − 1 . This gives the same conclusion.
All constructions use smooth pullbacks, finite linear operations, the fixed smooth charts, and compact-interval integration. They commute with pullback of external plots and preserve joint smoothness through the closed faces.
Theorem 7.4 (The actual open-wise comparison). For every ordinary open O O O , the holonomy and period map restricts to
Φ O : Int ( E O ) ⟶ y K [ O ] . \Phi_O : \operatorname{Int}(E_O) \longrightarrow yK[O]. Φ O : Int ( E O ) ⟶ y K [ O ] . For every smooth parameter manifold U U U , Φ O ( U ) \Phi_O(U) Φ O ( U ) is a trivial Kan fibration. There are a smooth simplicial section Ψ O \Psi_O Ψ O and a smooth simplicial homotopy
H O : Int ( E O ) × Δ [ 1 ] ⟶ Int ( E O ) , H O ( − , 0 ) = Ψ O Φ O , H O ( − , 1 ) = Id , H_O : \operatorname{Int}(E_O) \times\Delta[1] \longrightarrow\operatorname{Int}(E_O), \qquad H_O(-,0) = \Psi_O\Phi_O, \qquad H_O(-,1) = \operatorname{Id}, H O : Int ( E O ) × Δ [ 1 ] ⟶ Int ( E O ) , H O ( − , 0 ) = Ψ O Φ O , H O ( − , 1 ) = Id , such that
Φ O Ψ O = 1 , Φ O H O = Φ O pr , H O ∣ Ψ O ( K [ O ] ) = pr . \Phi_O\Psi_O = 1, \qquad\Phi_OH_O = \Phi_O\operatorname{pr}, \qquad H_O|_{\Psi_O(K[O])} = \operatorname{pr}. Φ O Ψ O = 1 , Φ O H O = Φ O pr , H O ∣ Ψ O ( K [ O ]) = pr . In particular, H O H_O H O is relative to the whole c ( y O ) c(yO) c ( y O ) .
Proof. For a full-MC simplex in O O O , the equation d y = 0 \mathrm{d}y = 0 d y = 0 puts its connected base image in one horizontal component of O O O . Its vertices therefore satisfy the hull condition, so Φ \Phi Φ restricts to Φ O \Phi_O Φ O .
Lemma 7.3, without additional split loci, solves every relative boundary problem. In degree zero the map is the identity. Thus each Φ O ( U ) \Phi_O(U) Φ O ( U ) has the right lifting property against all boundary inclusions and is a trivial Kan fibration.
For the smooth section, induct on simplicial degree. At degree n n n , the prescribed faces are Ψ O , n − 1 d i \Psi_{O,n-1}d_i Ψ O , n − 1 d i . On the degeneracy locus B j = im s j B_j = \operatorname{im}s_j B j = im s j , prescribe
s j Ψ O , n − 1 d j + 1 . s_j\Psi_{O,n-1}d_{j+1}. s j Ψ O , n − 1 d j + 1 .
The retractions s j d j + 1 s_jd_{j+1} s j d j + 1 preserve earlier degeneracy loci. On B i ∩ B j B_i \cap B_j B i ∩ B j , i < j i < j i < j , both prescriptions are the common double degeneracy
s i s j − 1 Ψ O , n − 2 = s j s i Ψ O , n − 2 s_i s_{j-1}\Psi_{O,n-2} = s_j s_i\Psi_{O,n-2} s i s j − 1 Ψ O , n − 2 = s j s i Ψ O , n − 2
of the uniquely determined lower simplex. These are the identities of Lemma A.9. The split-locus lifting lemma therefore supplies Ψ O , n \Psi_{O,n} Ψ O , n with all required faces and degeneracies.
Construct the homotopy in the same way, degreewise for each label τ : [ n ] → [ 1 ] \tau: [n] \to[1] τ : [ n ] → [ 1 ] . Constant labels give the two endpoints. For a nonconstant label prescribe the already constructed faces and first impose the fixed locus Ψ O ( K [ O ] n ) \Psi_O(K[O]_n) Ψ O ( K [ O ] n ) , retracted onto by Ψ O Φ O \Psi_O\Phi_O Ψ O Φ O . For every j j j with τ ( j ) = τ ( j + 1 ) \tau(j) = \tau(j+1) τ ( j ) = τ ( j + 1 ) , impose the degeneracy locus with prescription
s j H O , n − 1 ( d j + 1 σ , d j + 1 τ ) . s_jH_{O,n-1}(d_{j+1}\sigma,d_{j+1}\tau). s j H O , n − 1 ( d j + 1 σ , d j + 1 τ ) .
On a double-degeneracy intersection, the two reduced labels agree because they factor through the same surjection ς j − 1 ς i = ς i ς j \varsigma_{j-1}\varsigma_i = \varsigma_i\varsigma_j ς j − 1 ς i = ς i ς j . The two prescriptions are consequently equal by the simplicial identities. On the fixed-locus intersections, they agree by the lower-degree fixedness condition. Each s j d j + 1 s_jd_{j+1} s j d j + 1 preserves the fixed locus, so the fixed locus can be placed first in the ordered split-locus construction. Lemma 7.3 completes the induction.
Every lift is taken over the current target and remains inside O O O . Labels on disconnected test manifolds are locally constant, so the constructions glue on such tests. Finally, Ψ O \Psi_O Ψ O is the identity in degree zero and preserves degeneracies; hence its image contains the whole unit-generated subsheaf, which H O H_O H O fixes. □
Corollary 7.5 (Raw coefficients on the two actual sources). The actual pullback dg functors along Φ O \Phi_O Φ O and Ψ O \Psi_O Ψ O are quasi-inverse dg quasi-equivalences on globally bounded finite-projective raw coefficients. Their maps on all Hom \operatorname{Hom} Hom complexes are chain-homotopy equivalences. The body cohomology ranks may vary.
Proof. Apply Theorem 5.3 to the inverse-homotopy data of Theorem 7.4 . Its hypothesis is the homotopy relative to the entire c ( y O ) c(y_O) c ( y O ) , not only agreement of the endpoint vertex maps. □
Remark 7.6. The canonical maps Φ O \Phi_O Φ O are compatible with restriction. The charts ψ m \psi_m ψ m , sections Ψ O \Psi_O Ψ O , and homotopies H O H_O H O are chosen separately on each O O O ; strict compatibility with inclusions of opens is an additional condition on such choices. The empty-open assertions are vacuous.
Actual differentiation, its integration retraction, and edge rectification
Fix an ordinary open O O O , and put
A = C ∞ ( O ) , G = K [ O ] , B O = A [ θ 1 , θ 2 , η ] . A = C^\infty(O), \qquad G = K[O], \qquad B_O = A[\theta_1,\theta_2,\eta]. A = C ∞ ( O ) , G = K [ O ] , B O = A [ θ 1 , θ 2 , η ] .
The differential on B O B_O B O is the restriction of (1) . The coefficient categories retain the globally bounded finite-projective hypotheses of § 2.3 and § 4 . Write
I = Ψ O ∗ I N E O . I = \Psi_O^*\mathop{\mathrm{IN}}\nolimits_{E_O}. I = Ψ O ∗ IN E O .
We first construct differentiation on the actual raw category, including every Hom \operatorname{Hom} Hom degree.
The scalar normalized quotient
For n > 0 n>0 n > 0 , let
S k = s k − 1 ( G n − 1 ) ⊆ G n , 1 ≤ k ≤ n , S_k=s_{k-1}(G_{n-1})\subseteq G_n,\qquad1\leq k\leq n, S k = s k − 1 ( G n − 1 ) ⊆ G n , 1 ≤ k ≤ n ,
and write I ( S k ) I(S_k) I ( S k ) for the ideal of smooth functions on G n G_n G n vanishing on S k S_k S k . Define
C N n = ⋂ k = 1 n I ( S k ) , J n = ∑ k = 1 n ( C N n ∩ I ( S k ) 2 ) , J ^ = J + δ f u l l J . (32) C_N^n=\bigcap_{k=1}^{n}I(S_k),\qquad J_n=\sum_{k=1}^{n}\left(C_N^n\cap I(S_k)^2\right),\qquad\widehat{J}=J+\delta_{\mathrm{full}}J. \tag*{(32)} C N n = k = 1 ⋂ n I ( S k ) , J n = k = 1 ∑ n ( C N n ∩ I ( S k ) 2 ) , J = J + δ full J . (32)
Here J = ⨁ n ≥ 0 J n J=\bigoplus_{n\geq0}J_n J = ⨁ n ≥ 0 J n , and δ f u l l \delta_{\mathrm{full}} δ full is the full simplicial coboundary. In degree zero, C N 0 = A C_N^0=A C N 0 = A and J 0 = 0 J_0=0 J 0 = 0 . The product is the ordinary front/back cup product.
Proposition 8.1 (The affine scalar differentiation). Write a = a 01 a=a_{01} a = a 01 and b = b 01 b=b_{01} b = b 01 for the dilation and translation coordinates of an edge g 01 g_{01} g 01 , and h g e o m = h 012 h_{\mathrm{geom}}=h_{012} h geom = h 012 for the triangle coordinate of (3.7). The normalized DGA quotient of [2 Theorem 3.3.5 and Definition 3.3.6] gives a unital DGA epimorphism
v e : C N ∙ ( G ) ⟶ B O \mathop{\mathrm{ve}}\nolimits:C_N^\bullet(G)\longrightarrow B_O ve : C N ∙ ( G ) ⟶ B O
with
ve ( log a ) = θ 1 , ve ( b ) = θ 2 , ve ( h g e o m ) = − η . (33) \operatorname{ve}(\log a)=\theta_1,\qquad\operatorname{ve}(b)=\theta_2,\qquad\operatorname{ve}(h_{\mathrm{geom}})=-\eta. \tag*{(33)} ve ( log a ) = θ 1 , ve ( b ) = θ 2 , ve ( h geom ) = − η . (33)
Its target differential is exactly
Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Q θ 2 = θ 1 θ 2 + x η , Q η = 0. Qf=f_x(x\theta_1+y\theta_2),\qquad Q\theta_1=-y\eta,\qquad Q\theta_2=\theta_1\theta_2+x\eta,\qquad Q\eta=0. Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Q θ 2 = θ 1 θ 2 + x η , Q η = 0.
Proof. The normalized infinitesimal bundles have ranks two in degree one and one in degree two. In degree two, fixing the two positive faces fixes both root edges, leaving the h h h -coordinate. Above degree two the positive horn fillers are unique. The quotient theorem therefore makes the graded algebra semi-free on the three generators in (8.2).
For f ∈ A f\in A f ∈ A ,
δ f u l l f ( g , m ) = f ( g m ) − f ( m ) . \delta_{\mathrm{full}}f(g,m)=f(gm)-f(m). δ full f ( g , m ) = f ( g m ) − f ( m ) .
Its first normal derivative is f x ( x log a + y b ) f_x(x\log a+yb) f x ( x log a + y b ) , giving the formula for Q f Qf Q f .
Put ℓ = log a \ell=\log a ℓ = log a . The first matrix entry of
g 02 = g 12 g 01 exp ( h Z m 0 ) g_{02}=g_{12}g_{01}\exp(hZ_{m_0}) g 02 = g 12 g 01 exp ( h Z m 0 )
gives
ℓ 02 = ℓ 12 + ℓ 01 − y h , δ f u l l ℓ = y h . \ell_{02}=\ell_{12}+\ell_{01}-yh,\qquad\delta_{\mathrm{full}}\ell=yh. ℓ 02 = ℓ 12 + ℓ 01 − y h , δ full ℓ = y h .
Hence Q θ 1 = − y η Q\theta_1=-y\eta Q θ 1 = − y η .
For the second entry, with φ ( t ) = ∫ 0 1 e − s t d s \varphi(t)=\int_0^1 e^{-st}\,ds φ ( t ) = ∫ 0 1 e − s t d s , one obtains
δ f u l l b = ( 1 − a 12 ) b 01 − a 12 a 01 x h φ ( h y ) . \delta_{\mathrm{full}}b=(1-a_{12})b_{01}-a_{12}a_{01}xh\varphi(hy). δ full b = ( 1 − a 12 ) b 01 − a 12 a 01 x h φ ( h y ) .
Modulo J 2 J_2 J 2 , the first term is − b 01 ℓ 12 -b_{01}\ell_{12} − b 01 ℓ 12 : the difference contains ℓ 12 2 b 01 \ell_{12}^{2}b_{01} ℓ 12 2 b 01 times a smooth function. For the second term use
a 12 a 01 φ ( h y ) − 1 = ( a 12 − 1 ) a 01 φ ( h y ) + ( a 01 − 1 ) φ ( h y ) + φ ( h y ) − 1. a_{12}a_{01}\varphi(hy)-1=(a_{12}-1)a_{01}\varphi(hy)+(a_{01}-1)\varphi(hy)+\varphi(hy)-1. a 12 a 01 φ ( h y ) − 1 = ( a 12 − 1 ) a 01 φ ( h y ) + ( a 01 − 1 ) φ ( h y ) + φ ( h y ) − 1.
After multiplication by h h h , each summand is normalized and lies in one of the required squared degeneracy ideals. The last assertion follows from φ ( t ) − 1 = t ψ ( t ) \varphi(t)-1=t\psi(t) φ ( t ) − 1 = t ψ ( t ) with ψ \psi ψ smooth; it is not division by y y y . Thus
δ f u l l b ≡ − b 01 ℓ 12 − x h ( m o d J 2 ) . \delta_{\mathrm{full}}b\equiv-b_{01}\ell_{12}-xh\pmod{J_2}. δ full b ≡ − b 01 ℓ 12 − x h ( mod J 2 ) .
Graded commutativity of the full quotient by J ^ \widehat{J} J gives
Q θ 2 = − θ 2 θ 1 + x η = θ 1 θ 2 + x η . Q\theta_2=-\theta_2\theta_1+x\eta=\theta_1\theta_2+x\eta. Q θ 2 = − θ 2 θ 1 + x η = θ 1 θ 2 + x η .
Finally, the tetrahedral equation is δ f u l l h = 0 \delta_{\mathrm{full}}h=0 δ full h = 0 , so Q η = 0 Q\eta=0 Q η = 0 . All calculations are local near units and hence apply on every O O O by the germ-locality of the quotient [2 Lemma 3.1.5] . □
The quotient is by J + δ f u l l J J+\delta_{\mathrm{full}}J J + δ full J , not merely by adjacent-degeneracy squares. The additional allowed-transposition relations are part of its DGA structure.
Proposition 8.2 (Scalar Van Est on every open). The map ve \operatorname{ve} ve is a quasi-isomorphism in every degree.
Proof. The degree-one last-horn fibres are target fibres, which inversion identifies with the convex domains D m D_m D m of (30) . A degree-two last horn specifies g 12 g_{12} g 12 and g 02 g_{02} g 02 . For every h ∈ R h \in\mathbb{R} h ∈ R , the unique remaining edge is
g 01 = g 12 − 1 g 02 exp ( − h Z m 0 ) . g_{01}=g_{12}^{-1}g_{02}\exp(-hZm_0). g 01 = g 12 − 1 g 02 exp ( − h Z m 0 ) .
The vertices already lie in one horizontal component, so this filler remains in K [ O ] K[O] K [ O ] . The fibre is therefore R \mathbb{R} R , including at the origin. Higher last-horn fibres are points. The last-horn criterion and higher Van Est theorem of [2 Lemma 4.3.3 and Theorem 4.3.4] apply in every finite degree. □
Matrices, arbitrary Hom degrees, and projective bodies
In a local homogeneous frame let e e e be the temporary identity one-cochain, with e 1 = 1 e_1=1 e 1 = 1 and all other components zero. It is not normalized. Directly from the raw product,
δ i n e + e ⋆ e = 0. \delta_{\mathrm{in}}e+e\star e=0. δ in e + e ⋆ e = 0.
and, for a cochain of bidegree ( p , q ) (p,q) ( p , q ) ,
δ f u l l ϕ = δ i n ϕ + e ⋆ ϕ − ( − 1 ) p + q ϕ ⋆ e . (34) \delta_{\mathrm{full}}\phi=\delta_{\mathrm{in}}\phi+e\star\phi-(-1)^{p+q}\phi\star e. \tag*{(34)} δ full ϕ = δ in ϕ + e ⋆ ϕ − ( − 1 ) p + q ϕ ⋆ e . (34)
The final endpoint coefficient is − ( − 1 ) p + q ( − 1 ) q = ( − 1 ) p + 1 -(-1)^{p+q}(-1)^q=(-1)^{p+1} − ( − 1 ) p + q ( − 1 ) q = ( − 1 ) p + 1 .
Theorem 8.3 (Actual strict dg differentiation). There is a restriction-natural strict dg functor
F = Diff O : Rep s m p e r f ( G ) ⟶ Coh ( B O ) . F=\operatorname{Diff}_{O}:\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf}}(G)\longrightarrow\operatorname{Coh}(BO). F = Diff O : Rep sm perf ( G ) ⟶ Coh ( B O ) . It retains the original graded bundle and its body differential. In a local homogeneous frame its formulas are
a R = r R − e , F R = Q + μ R , μ R = ve ( a R ) , F ( ϕ ) = ve ( ϕ ) . (35) a_R=r_R-e,\qquad FR=Q+\mu_R,\qquad\mu_R=\operatorname{ve}(a_R),\qquad F(\phi)=\operatorname{ve}(\phi). \tag*{(35)} a R = r R − e , F R = Q + μ R , μ R = ve ( a R ) , F ( ϕ ) = ve ( ϕ ) . (35) where differentiation is entrywise. These formulas apply to nonclosed morphisms in every integer degree and retain every mixed and terminal Maurer–Cartan equation.
Proof. The matrix product on normalized cochains is
( c X ) ( d Y ) = ( − 1 ) ∣ X ∣ ∣ d ∣ ( c ⌣ d ) X Y . (cX)(dY)=(-1)^{|X||d|}(c\smile d)_{XY}. ( c X ) ( d Y ) = ( − 1 ) ∣ X ∣∣ d ∣ ( c ⌣ d ) X Y .
It is precisely the forms-first version of the raw product. Writing r R = e + a R r_R=e+a_R r R = e + a R , equation (34) turns the raw MC equation into
δ f u l l a R + a R ⋆ a R = 0. \delta_{\mathrm{full}}a_R+a_R\star a_R=0. δ full a R + a R ⋆ a R = 0.
For a homogeneous morphism of total degree m m m ,
D ϕ = δ f u l l ϕ + a S ⋆ ϕ − ( − 1 ) m ϕ ⋆ a R . D\phi=\delta_{\mathrm{full}}\phi+a_S\star\phi-(-1)^m\phi\star a_R. D ϕ = δ full ϕ + a S ⋆ ϕ − ( − 1 ) m ϕ ⋆ a R .
Apply the scalar DGA map entrywise. It gives
Q μ R + μ R 2 = 0 Q\mu_R+\mu_R^2=0 Q μ R + μ R 2 = 0
and
F ( D ϕ ) = Q F ( ϕ ) + μ S F ( ϕ ) − ( − 1 ) m F ( ϕ ) μ R . F(D\phi)=QF(\phi)+\mu_SF(\phi)-(-1)^mF(\phi)\mu_R. F ( D ϕ ) = QF ( ϕ ) + μ S F ( ϕ ) − ( − 1 ) m F ( ϕ ) μ R .
This is the cohesive Hom differential in forms-first order. The fixed braiding of § 6 returns the signlessly right-linear convention. Products and identity zero-cochains are preserved strictly.
If old coordinates equal g g g times new coordinates, the actual endpoint frame transformation is
a R g = g − 1 a R g + g − 1 δ f u l l g . a^{g}_{R}=g^{-1}a_{R}g+g^{-1}\delta_{\mathrm{full}}g. a R g = g − 1 a R g + g − 1 δ full g .
Differentiation gives
μ R g = g − 1 μ R g + g − 1 Q g . (36) \mu^{g}_{R}=g^{-1}\mu_{R}g+g^{-1}Qg. \tag*{(36)} μ R g = g − 1 μ R g + g − 1 Q g . (36)
Thus the local operators glue to superconnections on the original bundle. For finite homogeneous presentations V R = im Π R V_R=\operatorname{im}\Pi_R V R = im Π R and V S = im Π S V_S=\operatorname{im}\Pi_S V S = im Π S , raw Hom cochains are endpoint corners. Degree-zero multiplication is cup multiplication at the corresponding endpoint, so
F ( Π S ( v 0 ) ϕ Π R ( v p ) ) = Π S F ( ϕ ) Π R . (37) F\bigl(\Pi_S(v_0)\phi\Pi_R(v_p)\bigr)=\Pi_SF(\phi)\Pi_R. \tag*{(37)} F ( Π S ( v 0 ) ϕ Π R ( v p ) ) = Π S F ( ϕ ) Π R . (37)
This proves global finite-projective compatibility directly from the endpoint projectors. The degree-zero component of the scalar quotient is the identity, so body differentials are unchanged. The single identity Q μ R + μ R 2 = 0 Q\mu_R+\mu_R^2=0 Q μ R + μ R 2 = 0 includes all curvature coefficients, including the last possible one allowed by the body amplitude. The construction therefore retains the full superconnection on the given body. □
Lemma 8.4 (Global graded Hom lifting). For every ordered pair of actual raw objects,
F R , S : Hom r a w ∙ ( R , S ) ⟶ Hom C o h ( B O ) ∙ ( F R , F S ) F_{R,S}:\operatorname{Hom}^{\bullet}_{\mathrm{raw}}(R,S)\longrightarrow\operatorname{Hom}^{\bullet}_{\mathrm{Coh}(B_O)}(FR,FS) F R , S : Hom raw ∙ ( R , S ) ⟶ Hom Coh ( B O ) ∙ ( F R , F S ) is degreewise surjective.
Proof. Lift a scalar ordered monomial by
f θ 1 ε 1 θ 2 ε 2 η k ⟼ f ⌣ ℓ ⌣ ε 1 ⌣ b ⌣ ε 2 ⌣ ( − h g e o m ) ⌣ k , ε i ∈ { 0 , 1 } . f\theta_1^{\varepsilon_1}\theta_2^{\varepsilon_2}\eta^k\longmapsto f\smile\ell^{\smile\varepsilon_1}\smile b^{\smile\varepsilon_2}\smile(-h_{\mathrm{geom}})^{\smile k},\qquad\varepsilon_i\in\{0,1\}. f θ 1 ε 1 θ 2 ε 2 η k ⟼ f ⌣ ℓ ⌣ ε 1 ⌣ b ⌣ ε 2 ⌣ ( − h geom ) ⌣ k , ε i ∈ { 0 , 1 } .
Proposition 8.1 shows that this is a graded lift. In finite homogeneous presentations, lift every ambient matrix entry and then apply both endpoint projectors. Equation (8.6) gives the prescribed cohesive corner. The sums are finite in each total degree because the internal Hom degrees are bounded. The lifts are global, smooth, and normalized. This construction provides a section at the level of graded vector spaces. □
Literal differentiation of the actual integration
The geometric weighted function estimate of Lemma 9.7 will annihilate higher normal-weight terms in the retraction calculation.
Lemma 8.5 (The two-form of the selected open section). For the section constructed in Theorem 7.4,
β Ψ O ( k ) = W ( − h ( k ) ) \beta_{\Psi_O(k)}=W(-h(k)) β Ψ O ( k ) = W ( − h ( k )) in every simplicial degree. Its degree-one potential is the path (7.3).
Proof. Whitney forms commute with every ordinal simplex map. Indeed, if λ \lambda λ is such a map, then λ ∗ t j = ∑ λ ( i ) = j t i \lambda^*t_j=\sum_{\lambda(i)=j}t_i λ ∗ t j = ∑ λ ( i ) = j t i ; substitution in the alternating formula for the Whitney form gives
λ ∗ ω j 0 … j r = ∑ i 0 < ⋯ < i r λ ( i a ) = j a ω i 0 … i r . \lambda^*\omega_{j_0\ldots j_r}=\sum_{\substack{i_0<\cdots<i_r\\\lambda(i_a)=j_a}}\omega_{i_0\ldots i_r}. λ ∗ ω j 0 … j r = i 0 < ⋯ < i r λ ( i a ) = j a ∑ ω i 0 … i r .
Repeated image indices have zero alternating coefficient. This proves naturality for degeneracies as well as faces.
Induct on the section degree. In degrees zero and one both two-forms vanish. If the assertion holds on the boundary of an n n n -simplex, the two-form baseline in Lemma 7.3 is already W ( − h ) \mathrm{W}(-h) W ( − h ) : its boundary residual is zero. On every prescribed degeneracy locus, the two-form is also W ( − h ) \mathrm{W}(-h) W ( − h ) by naturality. Thus all split-locus two-form residuals are zero, and the linear ordered extension leaves them zero. The subsequent potential extension does not change β \beta β .
In degree one, the sole degeneracy prescription is the identity edge. Its potential and the radial baseline are both constantly e e e , so its residual is zero. The selected potential is therefore exactly (7.3). □
Theorem 8.6 (Literal integration retraction). For every ordinary open O O O ,
F I = Id FI=\operatorname{Id} F I = Id as strictly unital A ∞ A_\infty A ∞ -functors. This is literal on objects and unary morphisms, on the original bundles; all higher Taylor maps of FI vanish.
Proof. Use root group coordinates u i = ( log a i , b i ) u_i=(\log a_i,b_i) u i = ( log a i , b i ) of weight one and rooted triangle coordinates H i j H_{ij} H ij of weight two, with the equivalent adjacent coordinates of Lemma 9.7. Weights concern smooth normal Taylor germs at the whole unit section; they do not require analyticity.
By Lemma 8.5, β = W ( − h ) \beta=\mathrm{W}(-h) β = W ( − h ) has weight at least two. Its simplex-radial primitive u = H v 0 β u=H_{v_0}\beta u = H v 0 β has the same bound. At a totally degenerate unit, the selected MC simplex is constant. Its normalized flat potential is therefore identically one. Smoothness gives g − 1 , d Δ g g-1,d_{\Delta}g g − 1 , d Δ g weight at least one. The reconstruction formula
A = d Δ g g − 1 + Z σ u A=d_{\Delta}g\,g^{-1}+Z_{\sigma}u A = d Δ g g − 1 + Z σ u
then shows that both MC one-forms have weight at least one. Consequently the pullback of a homogeneous scalar CE element of degree d d d has normal weight at least d d d .
Write Γ N Ψ \Gamma^\Psi_N Γ N Ψ for the actual scalar kernel pulled back along the chosen full-MC section Ψ O \Psi_O Ψ O . For this kernel, an output in simplicial degree n n n satisfies
n = ∑ i d i + 1 − N . n=\sum_i d_i+1-N. n = i ∑ d i + 1 − N .
where d i d_i d i are the ordinary input form degrees. For N > 1 N>1 N > 1 , every term with a zero-form input vanishes. In every remaining term with N > 1 N>1 N > 1 , all d i d_i d i are positive; fixed simplex pullbacks, products, and compact integrations then give normal weight at least
∑ i d i = n + N − 1 ≥ n + 1. \sum_i d_i=n+N-1\ge n+1. i ∑ d i = n + N − 1 ≥ n + 1.
The output is normalized. Lemma 9.7 therefore gives
v e Γ N Ψ = 0 ( N > 1 ) . \mathrm{ve}\,\Gamma_N^{\Psi}=0 \qquad(N>1). ve Γ N Ψ = 0 ( N > 1 ) .
There is no nonzero higher output in degree zero.
Let F N = v e Γ N Ψ F_N=\mathrm{ve}\,\Gamma_N^{\Psi} F N = ve Γ N Ψ . Vertex evaluation gives F 1 ( f ) = f F_1(f)=f F 1 ( f ) = f . On the selected edge,
∫ α 1 = log a , ∫ α 2 = b − ∫ b ( t ) d log a ( t ) . \int\alpha_1=\log a,\qquad\int\alpha_2=b-\int b(t)\,\mathrm{d}\log a(t). ∫ α 1 = log a , ∫ α 2 = b − ∫ b ( t ) d log a ( t ) .
The last integral has weight at least two and vanishes under degree-one differentiation. Thus F 1 ( θ i ) = θ i F_1(\theta_i)=\theta_i F 1 ( θ i ) = θ i . The exact period ∫ β = − h \int\beta=-h ∫ β = − h gives F 1 ( η ) = η F_1(\eta)=\eta F 1 ( η ) = η . Since all higher F N F_N F N vanish, the A ∞ A_\infty A ∞ identities make F 1 F_1 F 1 a DGA map. Its values on A , θ 1 , θ 2 , η A,\theta_1,\theta_2,\eta A , θ 1 , θ 2 , η give F 1 = 1 F_1=1 F 1 = 1 , including all powers of the even generator.
Now use the actual matrix extension (23) and the twisted formula (24) . In every word of length greater than one, either a zero-form slot kills the word or the preceding normal-weight argument kills its differentiated scalar entries. For an object, only the bare unary term of r − e r-e r − e survives, recovering the entire original superconnection, including its body differential. For a unary morphism, only the uninserted unary word survives. For a higher Taylor map, every word has length at least two and vanishes after differentiation. These formulas use the original target totalization throughout.
The general kernel convergence statement in § 6 is local uniform convergence with every smooth parameter derivative. The quotient in degree n n n uses a finite normal jet, so these derivatives can be taken termwise in the actual infinite connection series. Equivalently, the ordered-time estimates have a factorial denominator, with only polynomial growth from the finitely many distinguished slots and derivative allocations. Thus the vanishing applies to the actual sums, not only to individual formal words.
Finally, choose a homogeneous frame near a unit. Compactness of the simplex and joint smoothness of the chosen Ψ O \Psi_O Ψ O allow the normal neighborhood to be shrunk so that every selected base simplex lies in that frame patch. This uses the restriction of Ψ O \Psi_O Ψ O , not a separately chosen section for the smaller patch. The endpoint frame laws and (37) glue the identities on the original finite-projective bundles. □
Actual invertible-edge rectification
The original raw edges need only be fibrewise quasi-isomorphisms. The following replacement makes them actual chain isomorphisms by a closed raw gauge.
Theorem 8.7 (Closed edge rectification). Every raw object R R R on G G G admits a normalized closed isomorphism
E R : R e ⟶ R , ( E R ) 0 = 1 , E_R:R^{\mathrm{e}}\longrightarrow R,\qquad(E_R)_0=1, E R : R e ⟶ R , ( E R ) 0 = 1 , where the edges of R e R^{\mathrm{e}} R e are the invertible last-to-first ODE transports of the connection differentiated from r R , 1 r_{R,1} r R , 1 along (31) . The inverse of E R E_R E R is given by a finite series. No regularity assumption is required.
Proof. Fix an edge g : m → g m g:m\to gm g : m → g m and put γ ( t ) = ψ m − 1 ( t ψ m ( g ) ) \gamma(t)=\psi_m^{-1}(t\psi_m(g)) γ ( t ) = ψ m − 1 ( t ψ m ( g )) , m t = γ ( t ) m m_t=\gamma(t)m m t = γ ( t ) m . For small ε \varepsilon ε , the increment
c t , ε = γ ( t + ε ) γ ( t ) − 1 c_{t,\varepsilon}=\gamma(t+\varepsilon)\gamma(t)^{-1} c t , ε = γ ( t + ε ) γ ( t ) − 1
and the h = 0 h=0 h = 0 triangle formed with γ ( t ) \gamma(t) γ ( t ) and γ ( t + ε ) \gamma(t+\varepsilon) γ ( t + ε ) are admissible: their vertices lie in one horizontal component of O O O .
In local frames along the path, set
A t = ∂ ε ∣ 0 r R , 1 ( c t , ε , m t ) . A_t=\left.\partial_\varepsilon\right|_0 r_{R,1}(c_{t,\varepsilon},m_t). A t = ∂ ε ∣ 0 r R , 1 ( c t , ε , m t ) .
Under a frame change u ( t ) u(t) u ( t ) it transforms as u − 1 A t u + u − 1 u ˙ u^{-1}A_tu+u^{-1}\dot{u} u − 1 A t u + u − 1 u ˙ . Thus the equation
∂ t T s , t = T s , t A t , T s , s = 1 \partial_tT_{s,t}=T_{s,t}A_t,\qquad T_{s,s}=1 ∂ t T s , t = T s , t A t , T s , s = 1
defines intrinsic smooth invertible transport T s , t : V m t → V m s T_{s,t}:V_{m_t}\to V_{m_s} T s , t : V m t → V m s . Differentiating the edge chain-map equation gives
d ˙ t = d t A t − A t d t , d s T s , t = T s , t d t , d t = r R , 0 ( m t ) . \dot{d}_t=d_tA_t-A_td_t,\qquad d_sT_{s,t}=T_{s,t}d_t,\qquad d_t=r_{R,0}(m_t). d ˙ t = d t A t − A t d t , d s T s , t = T s , t d t , d t = r R , 0 ( m t ) .
Write R t = r R , 1 ( γ ( t ) , m ) R_t=r_{R,1}(\gamma(t),m) R t = r R , 1 ( γ ( t ) , m ) and define
K t = ∂ ε ∣ 0 r R , 2 ( γ ( t ) , c t , ε , γ ( t + ε ) ; 0 ) . K_t=\left.\partial_\varepsilon\right|_0 r_{R,2}\bigl(\gamma(t),c_{t,\varepsilon},\gamma(t+\varepsilon);0\bigr). K t = ∂ ε ∣ 0 r R , 2 ( γ ( t ) , c t , ε , γ ( t + ε ) ; 0 ) .
At ε = 0 \varepsilon=0 ε = 0 the triangle is degenerate and its r 2 r_2 r 2 is zero. The derivative is therefore independent of a temporary identification of the moving endpoint. The triangle MC equation yields
R ˙ t − R t A t = d m K t + K t d t . \dot{R}_t-R_tA_t=d_mK_t+K_td_t. R ˙ t − R t A t = d m K t + K t d t .
The bundle-valued integral
H t = ∫ 0 t K s T s , t d s : V m t → V m H_t=\int_0^t K_sT_{s,t}\,ds:V_{m_t}\to V_m H t = ∫ 0 t K s T s , t d s : V m t → V m
satisfies H ˙ t = K t + H t A t \dot{H}_t=K_t+H_tA_t H ˙ t = K t + H t A t . Using the formula for d ˙ t \dot{d}_t d ˙ t , both R t − T 0 , t R_t-T_{0,t} R t − T 0 , t and d m H t + H t d t d_mH_t+H_td_t d m H t + H t d t solve the same inhomogeneous ODE with zero initial value. Hence
R t − T 0 , t = d m H t + H t d t . R_t-T_{0,t}=d_mH_t+H_td_t. R t − T 0 , t = d m H t + H t d t .
Put H ( g , m ) = H 1 H(g,m)=H_1 H ( g , m ) = H 1 and define E = 1 + H E=1+H E = 1 + H , with H H H only in simplicial degree one and internal degree minus one. At a unit, H = 0 H=0 H = 0 . If the body width is w w w , then H ⋆ ( w + 1 ) = 0 H^{\star(w+1)}=0 H ⋆ ( w + 1 ) = 0 and
E − 1 = ∑ j = 0 w ( − H ) ⋆ j . E^{-1}=\sum_{j=0}^{w}(-H)^{\star j}. E − 1 = j = 0 ∑ w ( − H ) ⋆ j .
Define
r e = E − 1 ⋆ ( δ i n E + r R ⋆ E ) . r^e=E^{-1}\star(\delta_{\mathrm{in}}E+r_R\star E). r e = E − 1 ⋆ ( δ in E + r R ⋆ E ) .
The raw derivation identities prove its MC equation and the closedness of E : R e → R E:R^e\to R E : R e → R . Normalization follows from
r e = r R + E − 1 ⋆ D R E : r^e=r_R+E^{-1}\star D_RE: r e = r R + E − 1 ⋆ D R E :
Here D R D_R D R is the differential on Hom r a w ( R , R ) \operatorname{Hom}_{\mathrm{raw}}(R,R) Hom raw ( R , R ) . It preserves normalized cochains, which are closed under the required products. The body is unchanged. In degree one,
r 1 e = r R , 1 − d m H − H d g m = T 0 , 1 . r^e_1=r_{R,1}-d_mH-Hd_{g m}=T_{0,1}. r 1 e = r R , 1 − d m H − H d g m = T 0 , 1 .
Thus the new edges are actual chain isomorphisms. The finite inverse is closed by the Hom Leibniz rule. All operations are intrinsic, smooth in parameters, and compatible with the original projective bundles. □
Remark 8.8. Strict dg differentiation sends E R E_R E R to a closed invertible cohesive map, with F ( E R − 1 ) = F ( E R ) − 1 F(E_R^{-1}) = F(E_R)^{-1} F ( E R − 1 ) = F ( E R ) − 1 . Its body is the identity, but F ( E R ) F(E_R) F ( E R ) itself need not be the identity. These actual differentiated gauges will be retained when comparison is transferred back from rectified objects.
Relative Hom transfer and normal-symbol calibration
Fix an ordinary open O O O , put A = C ∞ ( O ) A = C^\infty(O) A = C ∞ ( O ) and G = K [ O ] G = K[O] G = K [ O ] , and let R , S R,S R , S be raw coefficients. Write ∂ R = r R , 0 \partial_R = r_{R,0} ∂ R = r R , 0 and ∂ S = r S , 0 \partial_S = r_{S,0} ∂ S = r S , 0 for their body differentials. If their bodies occupy [ a R , b R ] [a_R,b_R] [ a R , b R ] and [ a S , b S ] [a_S,b_S] [ a S , b S ] , set
ℓ = a S − b R , u = b S − a R . \ell= a_S - b_R,\qquad u = b_S - a_R. ℓ = a S − b R , u = b S − a R .
The construction below retains the body differential. Internal degree will control triangular operators, not a filtration by subcomplexes.
A finite perturbation lemma
Lemma 9.1 (Finite perturbation with a retained small differential). Suppose
( C 0 , b ) ⇄ p i ( C 1 , d ) (C_0,b)\mathrel{\mathop{\rightleftarrows}^{i}_{p}}(C_1,d) ( C 0 , b ) ⇄ p i ( C 1 , d ) has a degree-minus-one map h h h satisfying
p i = 1 , d i = i b , p d = b p , d h + h d = 1 − i p , h 2 = h i = p h = 0. pi = 1,\qquad di = ib,\qquad pd = bp,\qquad dh + hd = 1 - ip,\qquad h^2 = hi = ph = 0. p i = 1 , d i = ib , p d = b p , d h + h d = 1 − i p , h 2 = hi = p h = 0. Let ( d + Δ ) 2 = 0 (d+\Delta)^2 = 0 ( d + Δ ) 2 = 0 , and suppose h Δ h\Delta h Δ and Δ h \Delta h Δ h are locally nilpotent with finite bounds in each graded component. Put
Σ = Δ ( 1 + h Δ ) − 1 = ( 1 + Δ h ) − 1 Δ , \Sigma= \Delta(1+h\Delta)^{-1} = (1+\Delta h)^{-1}\Delta, Σ = Δ ( 1 + h Δ ) − 1 = ( 1 + Δ h ) − 1 Δ , i ′ = i − h Σ i , p ′ = p − p Σ h , h ′ = h − h Σ h , b ′ = b + p Σ i . i' = i-h\Sigma i,\qquad p' = p-p\Sigma h,\qquad h' = h-h\Sigma h,\qquad b' = b+p\Sigma i. i ′ = i − h Σ i , p ′ = p − p Σ h , h ′ = h − h Σ h , b ′ = b + p Σ i . Then these formulas define a strong deformation retract (SDR) of cochain complexes, with the same side conditions, between ( C 0 , b ′ ) (C_0,b') ( C 0 , b ′ ) and ( C 1 , d + Δ ) (C_1,d+\Delta) ( C 1 , d + Δ ) .
Proof. All inverses are finite sums. Set X = 1 + h Δ X = 1+h\Delta X = 1 + h Δ and Y = 1 + Δ h Y = 1+\Delta h Y = 1 + Δ h . The formulas give i ′ = X − 1 i i' = X^{-1}i i ′ = X − 1 i , p ′ = p Y − 1 p' = pY^{-1} p ′ = p Y − 1 , and h ′ = X − 1 h = h Y − 1 h' = X^{-1}h = hY^{-1} h ′ = X − 1 h = h Y − 1 . Using d Δ + Δ d + Δ 2 = 0 d\Delta+\Delta d+\Delta^2 = 0 d Δ + Δ d + Δ 2 = 0 and d h + h d = 1 − i p dh+hd = 1-ip d h + h d = 1 − i p , direct multiplication gives
X ( d + Δ ) i ′ = i b ′ , p ′ ( d + Δ ) Y = b ′ p . X(d+\Delta)i' = ib',\qquad p'(d+\Delta)Y = b'p. X ( d + Δ ) i ′ = i b ′ , p ′ ( d + Δ ) Y = b ′ p . The side conditions give p ′ i ′ = 1 p'i' = 1 p ′ i ′ = 1 . Moreover, both
X ( ( d + Δ ) h ′ + h ′ ( d + Δ ) ) Y and X ( 1 − i ′ p ′ ) Y X((d+\Delta)h' + h'(d+\Delta))Y\qquad\text{and}\qquad X(1-i'p')Y X (( d + Δ ) h ′ + h ′ ( d + Δ )) Y and X ( 1 − i ′ p ′ ) Y expand to
1 − i p + Δ h + h Δ + h Δ 2 h . 1-ip+\Delta h+h\Delta+h\Delta^2h. 1 − i p + Δ h + h Δ + h Δ 2 h . Thus the chain and homotopy identities hold. From ( d + Δ ) i ′ = i ′ b ′ (d+\Delta)i' = i'b' ( d + Δ ) i ′ = i ′ b ′ and p ′ i ′ = 1 p'i' = 1 p ′ i ′ = 1 one obtains b ′ 2 = 0 b'^2 = 0 b ′2 = 0 . Expanding the finite formulas and using h 2 = h i = p h = 0 h^2 = hi = ph = 0 h 2 = hi = p h = 0 gives h ′ 2 = h ′ i ′ = p ′ h ′ = 0 h'^2 = h'i' = p'h' = 0 h ′2 = h ′ i ′ = p ′ h ′ = 0 . No vanishing of b b b , and no module-linearity of Δ \Delta Δ , is required. □
The pullback-coefficient relative complex
Let
P p = G p + 1 , π p = d p + 1 : P p → G p , t p = v p + 1 : P p → O . P_p = G_{p+1}, \qquad\pi_p = d_{p+1}: P_p \to G_p, \qquad t_p = v_{p+1}: P_p \to O. P p = G p + 1 , π p = d p + 1 : P p → G p , t p = v p + 1 : P p → O .
Give P P P the upper-decalage faces and degeneracies, indexed 0 , … , p 0,\ldots,p 0 , … , p in degree p p p . The last degeneracy of G G G is not one of these operators. The map π \pi π forgets the appended terminal vertex, whereas t t t records it.
Define
D π p , r , q ( R , S ) = Ω r ( P p / G p ; π p ∗ Hom q ( v p ∗ V R , v 0 ∗ V S ) ) , ∣ Φ ∣ = p + r + q . \mathcal{D}_{\pi}^{p,r,q}(R,S) = \Omega^r\left(P_p/G_p;\ \pi_p^*\operatorname{Hom}^q(v_p^*V_R,v_0^*V_S)\right), \qquad|\Phi|=p+r+q. D π p , r , q ( R , S ) = Ω r ( P p / G p ; π p ∗ Hom q ( v p ∗ V R , v 0 ∗ V S ) ) , ∣Φ∣ = p + r + q .
All cochains are normalized with respect to the genuine degeneracies of P P P . The front and back maps retain the appended terminal vertex. For composable tridegrees, define
Φ ⋆ π Ψ = ( − 1 ) ( r + q ) p ′ + q r ′ Φ f r o n t ∧ Ψ b a c k . (38) \Phi\star_\pi\Psi=(-1)^{(r+q)p'+qr'}\Phi_{\mathrm{front}}\wedge\Psi_{\mathrm{back}}. \tag*{(38)} Φ ⋆ π Ψ = ( − 1 ) ( r + q ) p ′ + q r ′ Φ front ∧ Ψ back . (38)
The wedge includes ordinary matrix composition in the displayed order; the exponent supplies its forms-first Koszul sign.
Proposition 9.2 (The actual relative Hom differential). With d v = ( − 1 ) p d π d_v=(-1)^p d_\pi d v = ( − 1 ) p d π , the operator
D π Φ = δ i n Φ + d v Φ + π ∗ r S ⋆ π Φ − ( − 1 ) ∣ Φ ∣ Φ ⋆ π π ∗ r R (39) D_\pi\Phi=\delta_{\mathrm{in}}\Phi+d_v\Phi+\pi^*r_S\star_\pi\Phi-(-1)^{|\Phi|}\Phi\star_\pi\pi^*r_R \tag*{(39)} D π Φ = δ in Φ + d v Φ + π ∗ r S ⋆ π Φ − ( − 1 ) ∣Φ∣ Φ ⋆ π π ∗ r R (39) squares to zero and preserves normalization. Its body-plus-relative part is
D π , 0 = ( − 1 ) p ( d π + ( − 1 ) r d H ) , d H X = ∂ S X − ( − 1 ) q X ∂ R . (40) D_{\pi,0}=(-1)^p(d_\pi+(-1)^r d_H), \qquad d_HX=\partial_SX-(-1)^qX\partial_R. \tag*{(40)} D π , 0 = ( − 1 ) p ( d π + ( − 1 ) r d H ) , d H X = ∂ S X − ( − 1 ) q X ∂ R . (40) The perturbation Δ π = D π − D π , 0 \Delta_\pi=D_\pi-D_{\pi,0} Δ π = D π − D π , 0 preserves relative degree, raises horizontal degree, and does not increase internal degree.
Proof. The coefficient bundle is pulled from G p G_p G p , so its relative derivative is canonical. In a projective presentation the two endpoint projectors have zero d π d_\pi d π -derivative. This assertion does not apply to terminal projectors pulled back by t t t .
The product phase in (9.1) is bilinear in the two degree triples; its two associativity exponents therefore agree. The usual inner-face cancellation gives the total Leibniz rule for δ i n \delta_{\mathrm{in}} δ in . The ordinary relative Leibniz rule gives the same total rule for d v d_v d v : differentiating the front form changes the product exponent by p ′ p' p ′ , and differentiating the back form changes it by q q q . Genuine face pullbacks commute with d π d_\pi d π , so δ i n d v + d v δ i n = 0 \delta_{\mathrm{in}}d_v+d_v\delta_{\mathrm{in}}=0 δ in d v + d v δ in = 0 .
The coefficients π ∗ r R , π ∗ r S \pi^*r_R,\pi^*r_S π ∗ r R , π ∗ r S have zero relative derivative. The square of (9.2) is therefore the left and right action of their pulled-back MC curvatures, and is zero. At a repeated endpoint, the unit terms complete the inner differential to the full normalized cancellation. In particular the last unit coefficient is
− ( − 1 ) p + r + q ( − 1 ) r + q = ( − 1 ) p + 1 . -(-1)^{p+r+q}(-1)^{r+q}=(-1)^{p+1}. − ( − 1 ) p + r + q ( − 1 ) r + q = ( − 1 ) p + 1 .
Higher structure components vanish on their degeneracies. Thus the full operator preserves normalization.
The body signs are ( − 1 ) p + r (-1)^{p+r} ( − 1 ) p + r at the target and − ( − 1 ) p + r + q -(-1)^{p+r+q} − ( − 1 ) p + r + q at the source. They give (9.3). The body coefficients are constant along each π p \pi_p π p -fibre, so this operator squares to zero. The remaining terms have degree ( 1 , 0 , 0 ) (1,0,0) ( 1 , 0 , 0 ) or ( k , 0 , 1 − k ) (k,0,1-k) ( k , 0 , 1 − k ) with k ≥ 1 k\geq1 k ≥ 1 , proving the last assertion. □ \square □
The prescribed normalized column SDR
In root coordinates the fibre of π p \pi_p π p is
D m 0 × R p , g = a p + 1 , z i = h 0 , i , p + 1 . D_{m_0} \times\mathbb{R}^p,\qquad g=a_{p+1},\quad z_i=h_{0,i,p+1}. D m 0 × R p , g = a p + 1 , z i = h 0 , i , p + 1 .
Use the charts of Lemma 7.2 and the fixed contraction
ϱ s ( g , z ) = ( ψ m 0 − 1 ( s ψ m 0 ( g ) ) , s z ) . (41) \varrho_s(g,z)=\left(\psi_{m_0}^{-1}\left(s\psi_{m_0}(g)\right),sz\right). \tag*{(41)} ϱ s ( g , z ) = ( ψ m 0 − 1 ( s ψ m 0 ( g ) ) , sz ) . (41)
It contracts to
κ p ( k ) = ( k , e , 0 , … , 0 ) , \kappa_p(k)=(k,e,0,\ldots,0), κ p ( k ) = ( k , e , 0 , … , 0 ) ,
which appends the root. This is not the last-degeneracy section.
Let k p k_p k p be its relative Poincare operator. In the based fibre coordinates w = ( ψ m 0 ( g ) , z ) w=(\psi_{m_0}(g),z) w = ( ψ m 0 ( g ) , z ) , for an r r r -form with r ≥ 1 r\geq1 r ≥ 1 ,
( k p ω ) w ( v 1 , … , v r − 1 ) = ∫ 0 1 s r − 1 ω s w ( w , v 1 , … , v r − 1 ) d s , (k_p\omega)_w(v_1,\ldots,v_{r-1})=\int_0^1 s^{r-1}\omega_{sw}(w,v_1,\ldots,v_{r-1})\,ds, ( k p ω ) w ( v 1 , … , v r − 1 ) = ∫ 0 1 s r − 1 ω s w ( w , v 1 , … , v r − 1 ) d s ,
and k p = 0 k_p=0 k p = 0 on relative zero-forms.
Proposition 9.3 (The column SDR). Set
J π , 0 = π ∗ , e π , 0 = κ ∗ , H π , 0 = ( − 1 ) p k p , J_{\pi,0}=\pi^*,\qquad e_{\pi,0}=\kappa^*,\qquad H_{\pi,0}=(-1)^p k_p, J π , 0 = π ∗ , e π , 0 = κ ∗ , H π , 0 = ( − 1 ) p k p , where e π , 0 e_{\pi,0} e π , 0 kills positive relative degree. Put
Σ π = Δ π ( 1 + H π , 0 Δ π ) − 1 , \Sigma_\pi=\Delta_\pi(1+H_{\pi,0}\Delta_\pi)^{-1}, Σ π = Δ π ( 1 + H π , 0 Δ π ) − 1 , J π = π ∗ , J_\pi=\pi^*, J π = π ∗ , r π = e π , 0 − e π , 0 Σ π H π , 0 , r_\pi=e_{\pi,0}-e_{\pi,0}\Sigma_\pi H_{\pi,0}, r π = e π , 0 − e π , 0 Σ π H π , 0 , H π = H π , 0 − H π , 0 Σ π H π , 0 . (42) H_\pi=H_{\pi,0}-H_{\pi,0}\Sigma_\pi H_{\pi,0}. \tag*{(42)} H π = H π , 0 − H π , 0 Σ π H π , 0 . (42) Then
r π J π = 1 , J π r π = 1 − D π H π − H π D π , r_\pi J_\pi=1,\qquad J_\pi r_\pi=1-D_\pi H_\pi-H_\pi D_\pi, r π J π = 1 , J π r π = 1 − D π H π − H π D π , with all SDR side conditions. The small differential is literally the original raw Hom differential. The construction applies to the original, possibly noninvertible edge maps.
Proof. Cartan’s homotopy identity gives
d π k p + k p d π = 1 − π ∗ κ ∗ . d_\pi k_p+k_p d_\pi=1-\pi^*\kappa^*. d π k p + k p d π = 1 − π ∗ κ ∗ .
The operator kills pullbacks and satisfies κ ∗ k p = 0 \kappa^*k_p=0 κ ∗ k p = 0 . Its square is zero: the iterated integral inserts two collinear radial vectors into an alternating form.
A genuine P P P -degeneracy repeats a prefix vertex. It preserves m 0 m_0 m 0 and g g g and either inserts an initial zero in the z z z -list or repeats one of its entries. It therefore commutes with (9.4). Thus all column maps preserve normalization. They act as scalars on the pulled-back coefficient bundle and commute with the plain d H d_H d H . The body cross terms in the homotopy identity are
( − 1 ) r − 1 d H k p + ( − 1 ) r k p d H = 0. (-1)^{r-1}d_Hk_p+(-1)^rk_pd_H=0. ( − 1 ) r − 1 d H k p + ( − 1 ) r k p d H = 0.
Consequently these maps give an SDR for D π , 0 D_{\pi,0} D π , 0 over the raw body-only differential ( − 1 ) p d H (-1)^p d_H ( − 1 ) p d H .
The perturbation preserves relative degree and the homotopy lowers it. On a relative r r r -form,
( H π , 0 Δ π ) r + 1 = ( Δ π H π , 0 ) r + 1 = 0. (H_{\pi,0}\Delta_\pi)^{r+1}=(\Delta_\pi H_{\pi,0})^{r+1}=0. ( H π , 0 Δ π ) r + 1 = ( Δ π H π , 0 ) r + 1 = 0.
Lemma 9.1 applies. Moreover H π , 0 Δ π J π , 0 = 0 H_{\pi,0}\Delta_\pi J_{\pi,0}=0 H π , 0 Δ π J π , 0 = 0 , since its argument has relative degree zero. Thus the perturbed inclusion is still π ∗ \pi^* π ∗ . The simplicial map π \pi π preserves the original products and twisting data, so
Δ π J π , 0 = J π , 0 ( D r a w − ( − 1 ) p d H ) . \Delta_\pi J_{\pi,0}=J_{\pi,0}(D_{\mathrm{raw}}-(-1)^p d_H). Δ π J π , 0 = J π , 0 ( D raw − ( − 1 ) p d H ) .
The transferred differential is exactly D r a w D_{\mathrm{raw}} D raw . The formulas are finite in every relative degree. Smoothness follows from compact integration in the jointly smooth charts. They act intrinsically on the pullback bundles, so they glue on finite-projective bodies. □ \square □
Full terminal conjugation
Assume now that the edges of R , S R,S R , S are actual smooth chain isomorphisms. For a raw object R R R , let B R B_R B R be the terminal-body representation on P P P : its fibre is V R ∣ m ∗ V_R|_{m_*} V R ∣ m ∗ , its body is ∂ R ( m ∗ ) \partial_R(m_*) ∂ R ( m ∗ ) , its edges are identities, and its higher structures vanish.
Lemma 9.4 (The upper-decalage gauge). The maps
U p R = ( − 1 ) p r R , p + 1 : B R ⟶ π ∗ R U_p^R=(-1)^p r_{R,p+1}:B_R\longrightarrow\pi^*R U p R = ( − 1 ) p r R , p + 1 : B R ⟶ π ∗ R form a normalized closed degree-zero morphism. Their closed inverse is given by a finite geometric series.
Proof. The component has type V m p + 1 → V m 0 V_{m_{p+1}}\to V_{m_0} V m p + 1 → V m 0 and internal degree − p -p − p . For p ≥ 1 p\geq1 p ≥ 1 , normalization follows from that of r p + 1 r_{p+1} r p + 1 . Expanding its Hom differential gives
( D U ) p = ∑ i = 1 p − 1 ( − 1 ) p + i − 1 r p ( d i σ ) + ∑ a = 0 p ( − 1 ) a ( p − a ) r a ( σ [ 0 , … , a ] ) r p − a + 1 ( σ [ a , … , p + 1 ] ) − r p ( d p σ ) + ( − 1 ) p + 1 r p + 1 ( σ ) ∂ R ( m p + 1 ) . (DU)_p=\sum_{i=1}^{p-1}(-1)^{p+i-1}r_p(d_i\sigma)
+\sum_{a=0}^{p}(-1)^{a(p-a)}r_a(\sigma[0,\ldots,a])r_{p-a+1}(\sigma[a,\ldots,p+1])
-r_p(d_p\sigma)+(-1)^{p+1}r_{p+1}(\sigma)\partial_R(m_{p+1}). ( D U ) p = i = 1 ∑ p − 1 ( − 1 ) p + i − 1 r p ( d i σ ) + a = 0 ∑ p ( − 1 ) a ( p − a ) r a ( σ [ 0 , … , a ]) r p − a + 1 ( σ [ a , … , p + 1 ]) − r p ( d p σ ) + ( − 1 ) p + 1 r p + 1 ( σ ) ∂ R ( m p + 1 ) . The source identity edge supplies the term − r p ( d p σ ) -r_p(d_p\sigma) − r p ( d p σ ) . This is ( − 1 ) p + 1 (-1)^{p+1} ( − 1 ) p + 1 times the original MC equation in degree p + 1 p+1 p + 1 , and vanishes. At p = 0 p=0 p = 0 the assertion is the edge chain-map equation.
Let X = U 0 X=U_0 X = U 0 and v = X − 1 ⋆ U − 1 v=X^{-1}\star U-1 v = X − 1 ⋆ U − 1 . If the body width is w w w , then v v v has only positive horizontal components and v ⋆ ( w + 1 ) = 0 v^{\star(w+1)}=0 v ⋆ ( w + 1 ) = 0 . Therefore
U − 1 = ( ∑ j = 0 w ( − v ) ⋆ j ) ⋆ X − 1 . U^{-1}=\left(\sum_{j=0}^{w}(-v)^{\star j}\right)\star X^{-1}. U − 1 = ( j = 0 ∑ w ( − v ) ⋆ j ) ⋆ X − 1 . The inverse is two-sided in the graded cochain algebra, has the reverse endpoint type, and is closed by the Hom Leibniz rule. The zero-cochain X X X alone need not be closed for the full raw differential. □ \square □
Define
A ( χ ) = ( U S ) − 1 ⋆ π χ ⋆ π U R , A − 1 ( ζ ) = U S ⋆ π ζ ⋆ π ( U R ) − 1 . \mathcal{A}(\chi)=(U^{S})^{-1}\star_{\pi}\chi\star_{\pi}U^{R},\qquad\mathcal{A}^{-1}(\zeta)=U^{S}\star_{\pi}\zeta\star_{\pi}(U^{R})^{-1}. A ( χ ) = ( U S ) − 1 ⋆ π χ ⋆ π U R , A − 1 ( ζ ) = U S ⋆ π ζ ⋆ π ( U R ) − 1 .
The terminal complex has coefficients t p ∗ Hom ( V R , V S ) t^{*}_{p}\operatorname{Hom}(V_{R},V_{S}) t p ∗ Hom ( V R , V S ) . Define its differential by full conjugation:
D t = A D π A − 1 = D t , 0 + ∇ U , (43) \mathbb{D}_{t}=\mathcal{A}\mathbb{D}_{\pi}\mathcal{A}^{-1}=\mathbb{D}_{t,0}+\nabla_{U}, \tag*{(43)} D t = A D π A − 1 = D t , 0 + ∇ U , (43)
where
D t , 0 = δ f u l l + ( − 1 ) p + r d H , t , ∇ U = A d v A − 1 . \mathbb{D}_{t,0}=\delta_{\mathrm{full}}+(-1)^{p+r}d_{H,t},\qquad\nabla_{U}=\mathcal{A}d_{v}\mathcal{A}^{-1}. D t , 0 = δ full + ( − 1 ) p + r d H , t , ∇ U = A d v A − 1 .
Closedness of the full U U U proves this split. It retains all body and higher-curvature terms. The derivative is applied only after returning by A − 1 \mathcal{A}^{-1} A − 1 to the bundle with its canonical relative derivative.
Each component of A \mathcal{A} A or A − 1 \mathcal{A}^{-1} A − 1 adds horizontal degree a ≥ 0 a\geq0 a ≥ 0 and lowers internal degree by a a a . Thus ∇ U \nabla_{U} ∇ U raises relative degree by one and does not increase internal degree. Conjugation also gives
D t , 0 2 = ∇ U 2 = 0 , D t , 0 ∇ U + ∇ U D t , 0 = 0. \mathbb{D}_{t,0}^{2}=\nabla_{U}^{2}=0,\qquad\mathbb{D}_{t,0}\nabla_{U}+\nabla_{U}\mathbb{D}_{t,0}=0. D t , 0 2 = ∇ U 2 = 0 , D t , 0 ∇ U + ∇ U D t , 0 = 0.
These identities involve the full operators: the higher structure components participate in the curvature cancellations.
The actual horizontal SDR
Let A p = C ∞ ( P p ) A^{p}=C^{\infty}(P_{p}) A p = C ∞ ( P p ) , with A A A acting through the terminal point m ∗ m_{*} m ∗ . The relative frame is
α = d π g g − 1 , d z 1 , … , d z p . \alpha=d_{\pi}g\,g^{-1},\qquad dz_{1},\ldots,dz_{p}. α = d π g g − 1 , d z 1 , … , d z p .
For the root-changing face,
d 0 ∗ α = α − Z m ∗ d z 1 , d 0 ∗ d z j = d z j + 1 − d z 1 . (44) d_{0}^{*}\alpha=\alpha-Z_{m_{*}}dz_{1},\qquad d_{0}^{*}dz_{j}=dz_{j+1}-dz_{1}. \tag*{(44)} d 0 ∗ α = α − Z m ∗ d z 1 , d 0 ∗ d z j = d z j + 1 − d z 1 . (44)
Indeed,
g ′ = g exp ( − z 1 Z m 0 ) a 1 − 1 , z j ′ = H 1 , j + 1 + z j + 1 − z 1 , g'=g\exp(-z_{1}Z_{m_{0}})a_{1}^{-1},\qquad z'_{j}=H_{1,j+1}+z_{j+1}-z_{1}, g ′ = g exp ( − z 1 Z m 0 ) a 1 − 1 , z j ′ = H 1 , j + 1 + z j + 1 − z 1 ,
and the prefix quantities have zero relative derivative. Other faces delete the appropriate z z z , while the initial degeneracy inserts zero and the remaining degeneracies repeat an entry.
Let V p \mathcal{V}^{p} V p be free over A A A on e 1 , e 2 , v 1 , … , v p \mathbf{e}_{1},\mathbf{e}_{2},v_{1},\ldots,v_{p} e 1 , e 2 , v 1 , … , v p , with the operators prescribed by (9.7). Put
H q = Γ ( O , Hom q ( V R , V S ) ) , W p , r , q = Λ A r V p ⊗ A H q . \mathcal{H}^{q}=\Gamma(O,\operatorname{Hom}^{q}(V_{R},V_{S})),\qquad W^{p,r,q}=\Lambda_{A}^{r}\mathcal{V}^{p}\otimes_{A}\mathcal{H}^{q}. H q = Γ ( O , Hom q ( V R , V S )) , W p , r , q = Λ A r V p ⊗ A H q .
The actual terminal relative forms are A p ⊗ A W p , r , q A^{p}\otimes_{A}W^{p,r,q} A p ⊗ A W p , r , q . This is a finite-frame and finite-projective bundle identity; arbitrary jointly smooth dependence remains in A p A^{p} A p .
Let j j j pull coefficient functions back from m ∗ m_{*} m ∗ , and let ϵ \epsilon ϵ evaluate coefficient functions on the constant simplex, retaining the formal W W W -tensor. These are cosimplicial A A A -linear maps with ϵ j = 1 \epsilon j=1 ϵ j = 1 .
For p ≥ 1 p\geq1 p ≥ 1 , define
f j p : [ p + 1 ] ⟶ [ p ] , f j p ( s ) = { s , s ≤ j , p , s > j , 0 ≤ j ≤ p − 1 , f_{j}^{p}:[p+1]\longrightarrow[p],\qquad f_{j}^{p}(s)=
\begin{cases}
s,&s\leq j,\\
p,&s>j,
\end{cases}
\qquad0\leq j\leq p-1, f j p : [ p + 1 ] ⟶ [ p ] , f j p ( s ) = { s , p , s ≤ j , s > j , 0 ≤ j ≤ p − 1 ,
and h j p = G ( f j p ) : P p − 1 → P p h_j^p=G(f_j^p):P_{p-1}\to P_p h j p = G ( f j p ) : P p − 1 → P p . Set
L p = ∑ j = 0 p − 1 ( − 1 ) j + 1 ( h j p ) ∗ ⊗ σ W j , L 0 = 0. (45) L_p=\sum_{j=0}^{p-1}(-1)^{j+1}(h_j^p)^*\otimes\sigma_W^j,\qquad L_0=0. \tag*{(45)} L p = j = 0 ∑ p − 1 ( − 1 ) j + 1 ( h j p ) ∗ ⊗ σ W j , L 0 = 0. (45)
The first factor acts on coefficient functions, and the second is the formal-frame codegeneracy.
Lemma 9.5 (Horizontal coefficient contraction). The maps above satisfy
δ f u l l L + L δ f u l l = 1 − j ϵ . \delta_{\mathrm{full}}L+L\delta_{\mathrm{full}}=1-j\epsilon. δ full L + L δ full = 1 − j ϵ .
After genuine cosimplicial normalization, there are maps i 0 i_0 i 0 , p 0 p_0 p 0 , h 0 h_0 h 0 with
p 0 i 0 = 1 , δ f u l l h 0 + h 0 δ f u l l = 1 − i 0 p 0 , h 0 2 = h 0 i 0 = p 0 h 0 = 0. p_0i_0=1,\qquad\delta_{\mathrm{full}}h_0+h_0\delta_{\mathrm{full}}=1-i_0p_0,\qquad h_0^2=h_0i_0=p_0h_0=0. p 0 i 0 = 1 , δ full h 0 + h 0 δ full = 1 − i 0 p 0 , h 0 2 = h 0 i 0 = p 0 h 0 = 0.
They preserve relative and internal degree, and h 0 h_0 h 0 lowers horizontal degree by one.
Proof. For the first identity, write δ = δ f u l l \delta=\delta_{\mathrm{full}} δ = δ full within this proof and expand δ L + L δ \delta L+L\delta δ L + L δ . If A i j A_{ij} A ij denotes an unsigned term from δ L \delta L δ L and B i j B_{ij} B ij one from L δ L\delta L δ , the order-map identities give
B i j = { A i , j − 1 , i < j , A i − 1 , j , i > j + 1. B_{ij}=
\begin{cases}
A_{i,j-1}, & i<j,\\
A_{i-1,j}, & i>j+1.
\end{cases} B ij = { A i , j − 1 , A i − 1 , j , i < j , i > j + 1.
Their signs are opposite. These pairs exhaust the off-diagonal terms. For i = j , j + 1 i=j,j+1 i = j , j + 1 , the W W W -map is the identity, and the coefficient terms telescope: successively more vertices are retained instead of being sent to the terminal vertex. The sum is the identity minus constant-terminal evaluation. This also proves the identity in horizontal degree zero.
Use
N p T = ⋂ i = 0 p − 1 ker s i , Π N p = ( 1 − d p s p − 1 ) ⋯ ( 1 − d 1 s 0 ) . N^pT=\bigcap_{i=0}^{p-1}\ker s^i,\qquad\Pi_N^p=(1-d^ps^{p-1})\cdots(1-d^1s^0). N p T = i = 0 ⋂ p − 1 ker s i , Π N p = ( 1 − d p s p − 1 ) ⋯ ( 1 − d 1 s 0 ) .
The cosimplicial identities split the full complex into its normalized part and the positive-coface degenerate subcomplex. Thus Π N \Pi_N Π N and the normalized inclusion are chain maps.
Put L N = Π N L ι N L_N=\Pi_NL\iota_N L N = Π N L ι N .
The codegeneracy kernels on V p \mathcal V^p V p are generated by
v 1 , v 2 − v 1 , … , v p − v p − 1 . v_1,\qquad v_2-v_1,\qquad\ldots,\qquad v_p-v_{p-1}. v 1 , v 2 − v 1 , … , v p − v p − 1 .
Together with e 1 , e 2 e_1,e_2 e 1 , e 2 , these form a free basis. On the exterior algebra the corresponding kernels are the ideals generated by these vectors. Their intersection consists exactly of basis monomials containing all p p p vectors. Consequently
N p Λ A r V p = Λ A r − p ( A e 1 ⊕ A e 2 ) ∧ v 1 ∧ ⋯ ∧ v p . (46) N^p\Lambda_A^r\mathcal V^p=\Lambda_A^{r-p}(Ae_1\oplus Ae_2)\wedge v_1\wedge\cdots\wedge v_p. \tag*{(46)} N p Λ A r V p = Λ A r − p ( A e 1 ⊕ A e 2 ) ∧ v 1 ∧ ⋯ ∧ v p . (46)
It vanishes unless 0 ≤ r − p ≤ 2 0\le r-p\le2 0 ≤ r − p ≤ 2 . Tensoring this split calculation with H q \mathcal H^q H q is legitimate.
Identify the normalized small module with
B R , S = B O ⊗ A H , \mathbb B_{R,S}=\mathbb B_O\otimes_A\mathcal H, B R , S = B O ⊗ A H ,
by fixing
i 0 ( f η p θ I X ) = c p , j f ( m ∗ ) α I ∧ d z 1 ∧ ⋯ ∧ d z p X , c p , j = ( − 1 ) p j + p ( p + 1 ) / 2 , j = ∣ I ∣ . (47) i_0(f\eta^p\theta_I X)=c_{p,j}f(m_*)\alpha_I\wedge dz_1\wedge\cdots\wedge dz_pX,\qquad c_{p,j}=(-1)^{pj+p(p+1)/2},\qquad j=|I|. \tag*{(47)} i 0 ( f η p θ I X ) = c p , j f ( m ∗ ) α I ∧ d z 1 ∧ ⋯ ∧ d z p X , c p , j = ( − 1 ) p j + p ( p + 1 ) /2 , j = ∣ I ∣. (47)
The map p 0 p_0 p 0 is coefficient evaluation followed by the inverse of this identification. Set
P 0 = 1 − i 0 p 0 , a = P 0 L N P 0 , h 0 = a δ f u l l a . P_0 = 1 - i_0p_0,\qquad a = P_0L_NP_0,\qquad h_0 = a\delta_{\mathrm{full}}a. P 0 = 1 − i 0 p 0 , a = P 0 L N P 0 , h 0 = a δ full a .
Then
δ f u l l a + a δ f u l l = P 0 , a i 0 = p 0 a = 0 , P 0 a = a P 0 = a . \delta_{\mathrm{full}}a + a\delta_{\mathrm{full}} = P_0,\qquad ai_0 = p_0a = 0,\qquad P_0a = aP_0 = a. δ full a + a δ full = P 0 , a i 0 = p 0 a = 0 , P 0 a = a P 0 = a .
It follows that ( δ f u l l a ) 2 = δ f u l l a (\delta_{\mathrm{full}}a)^2 = \delta_{\mathrm{full}}a ( δ full a ) 2 = δ full a , ( a δ f u l l ) 2 = a δ f u l l (a\delta_{\mathrm{full}})^2 = a\delta_{\mathrm{full}} ( a δ full ) 2 = a δ full , and δ f u l l a 2 δ f u l l = 0 \delta_{\mathrm{full}}a^2\delta_{\mathrm{full}} = 0 δ full a 2 δ full = 0 . These identities give the homotopy and all three side conditions for h 0 h_0 h 0 . Every factor preserves r , q r,q r , q , so h 0 h_0 h 0 lowers p p p by one.
The homotopy is normalized using only the horizontal differential. It acts as the identity on the body-Hom tensor factor and is A A A -linear through m ∗ m_* m ∗ . Thus it commutes with the plain body-Hom differential, and the two total body signs cancel because it lowers p p p . The same maps therefore give an SDR for D t , 0 \mathbb{D}_{t,0} D t , 0 , with small differential
D B , 0 ( ω X ) = Q h ω X + ( − 1 ) ∣ ω ∣ ω d H X , Q h = η ι ( − y , x ) . (48) D_{\mathrm{B},0}(\omega_X) = Q_h\omega_X + (-1)^{|\omega|}\omega\,d_HX,\qquad Q_h = \eta\iota_{(-y,x)}. \tag*{(48)} D B , 0 ( ω X ) = Q h ω X + ( − 1 ) ∣ ω ∣ ω d H X , Q h = η ι ( − y , x ) . (48)
The small complex retains the full body and Koszul differentials.
Apply Lemma 9.1 to ∇ U \nabla_U ∇ U :
Σ U = ∇ U ( 1 + h 0 ∇ U ) − 1 , \Sigma_U = \nabla_U(1+h_0\nabla_U)^{-1}, Σ U = ∇ U ( 1 + h 0 ∇ U ) − 1 ,
D t r = D B , 0 + p 0 Σ U i 0 , i h = i 0 − h 0 Σ U i 0 , p h = p 0 − p 0 Σ U h 0 , H h = h 0 − h 0 Σ U h 0 . (49) \begin{aligned}
D_{\mathrm{tr}} &= D_{\mathrm{B},0}+p_0\Sigma_Ui_0,\\
i_h &= i_0-h_0\Sigma_Ui_0,\qquad p_h = p_0-p_0\Sigma_Uh_0,\\
H_h &= h_0-h_0\Sigma_Uh_0. \tag*{(49)}
\end{aligned} D tr i h H h = D B , 0 + p 0 Σ U i 0 , = i 0 − h 0 Σ U i 0 , p h = p 0 − p 0 Σ U h 0 , = h 0 − h 0 Σ U h 0 . (49)
For an input of total degree m m m and relative degree r r r , both h 0 ∇ U h_0\nabla_U h 0 ∇ U and ∇ U h 0 \nabla_Uh_0 ∇ U h 0 have total degree zero and raise relative degree by one. A nonzero output after k k k iterations satisfies r + k ≤ m − ℓ r+k\le m-\ell r + k ≤ m − ℓ . Hence
( h 0 ∇ U ) k = ( ∇ U h 0 ) k = 0 when k > m − ℓ − r . (50) (h_0\nabla_U)^k=(\nabla_Uh_0)^k=0\qquad\text{when }k>m-\ell-r. \tag*{(50)} ( h 0 ∇ U ) k = ( ∇ U h 0 ) k = 0 when k > m − ℓ − r . (50)
Thus all displayed maps are finite in each total degree, and
p h i h = 1 , i h p h = 1 − D t H h − H h D t , H h 2 = H h i h = p h H h = 0. p_hi_h=1,\qquad i_hp_h=1-D_tH_h-H_hD_t,\qquad H_h^2=H_hi_h=p_hH_h=0. p h i h = 1 , i h p h = 1 − D t H h − H h D t , H h 2 = H h i h = p h H h = 0.
Proposition 9.6 (The fixed comparison maps). Define
V m o d = r π A − 1 i h , S m o d = p h A J π , T = F V m o d . V_{\mathrm{mod}}=r_\pi\mathcal{A}^{-1}i_h,\qquad S_{\mathrm{mod}}=p_h\mathcal{A}J_\pi,\qquad T=FV_{\mathrm{mod}}. V mod = r π A − 1 i h , S mod = p h A J π , T = F V mod . Then V m o d V_{\mathrm{mod}} V mod and S m o d S_{\mathrm{mod}} S mod are chain-homotopy inverse maps between the actual raw Hom complex and ( B R , S , D t r ) (\mathbb{B}_R,S,D_{\mathrm{tr}}) ( B R , S , D tr ) . The map T T T is a chain map to the actual cohesive Hom complex, and
T S m o d − F = − d C o h K F − K F D r a w , K F = F r π A − 1 H h A J π . (51) TS_{\mathrm{mod}}-F=-d_{\mathrm{Coh}}\mathcal{K}_F-\mathcal{K}_FD_{\mathrm{raw}},\qquad\mathcal{K}_F=Fr_\pi\mathcal{A}^{-1}H_h\mathcal{A}J_\pi. \tag*{(51)} T S mod − F = − d Coh K F − K F D raw , K F = F r π A − 1 H h A J π . (51) Every displayed comparison map is internally nonincreasing.
Proof. The two model homotopies are
K C = r π A − 1 H h A J π , K B = p h A H π A − 1 i h . K_C=r_\pi\mathcal{A}^{-1}H_h\mathcal{A}J_\pi,\qquad K_B=p_h\mathcal{A}H_\pi\mathcal{A}^{-1}i_h. K C = r π A − 1 H h A J π , K B = p h A H π A − 1 i h .
The two SDR identities give
V m o d S m o d = 1 − D r a w K C − K C D r a w , V_{\mathrm{mod}}S_{\mathrm{mod}}=1-D_{\mathrm{raw}}K_C-K_CD_{\mathrm{raw}}, V mod S mod = 1 − D raw K C − K C D raw ,
e n d s m i d = 1 {ends_mid=1} e n d s m i d = 1
S m o d V m o d = 1 − D t r K B − K B D t r . S_{\mathrm{mod}}V_{\mathrm{mod}}=1-D_{\mathrm{tr}}K_{\mathrm{B}}-K_{\mathrm{B}}D_{\mathrm{tr}}. S mod V mod = 1 − D tr K B − K B D tr .
Applying F F F to the first identity proves (9.14), including its minus signs. The elementary column maps preserve internal degree; Δ π \Delta_{\pi} Δ π does not increase it. Positive components of U , U − 1 U,U^{-1} U , U − 1 lower it. The relative derivative preserves it, and the horizontal maps preserve it. Therefore all finite transfer expressions are internally nonincreasing. The map F F F preserves the internal matrix block by its entrywise quotient formula. Thus internal degree orders the comparison operators, while the body differential itself can raise internal degree.
The scalar inclusion and its actual differential
For scalar unit coefficients, U = 1 U=1 U = 1 and ∇ U = d v \nabla_U=d_v ∇ U = d v . Denote (9.10) by ι \iota ι . Split Q = Q h + Q v Q=Q_h+Q_v Q = Q h + Q v , with
Q v f = f x ( x θ 1 + y θ 2 ) , Q v θ 1 = 0 , Q v θ 2 = θ 1 θ 2 , Q v η = 0. Q_vf=f_x(x\theta_1+y\theta_2),\qquad Q_v\theta_1=0,\qquad Q_v\theta_2=\theta_1\theta_2,\qquad Q_v\eta=0. Q v f = f x ( x θ 1 + y θ 2 ) , Q v θ 1 = 0 , Q v θ 2 = θ 1 θ 2 , Q v η = 0.
The exact frame formulas give
δ f u l l ( α I ∧ d z 1 ∧ ⋯ ∧ d z p ) = ( − 1 ) j ( ι Z α I ) ∧ d z 1 ∧ ⋯ ∧ d z p + 1 . \delta_{\mathrm{full}}(\alpha_I\wedge dz_1\wedge\cdots\wedge dz_p)=(-1)^j(\iota_Z\alpha_I)\wedge dz_1\wedge\cdots\wedge dz_{p+1}. δ full ( α I ∧ d z 1 ∧ ⋯ ∧ d z p ) = ( − 1 ) j ( ι Z α I ) ∧ d z 1 ∧ ⋯ ∧ d z p + 1 .
The terms without a root-face change cancel by expanding ⋀ i = 2 p + 1 ( d z i − d z 1 ) \bigwedge_{i=2}^{p+1}(dz_i-dz_1) ⋀ i = 2 p + 1 ( d z i − d z 1 ) . Since
c p + 1 , j − 1 = ( − 1 ) j c p , j , c_{p+1,j-1}=(-1)^jc_{p,j}, c p + 1 , j − 1 = ( − 1 ) j c p , j ,
δ f u l l ι = ι Q h . \delta_{\mathrm{full}}\iota=\iota Q_h. δ full ι = ι Q h .
Also
d π f ( m ∗ ) = f x ( m ∗ ) ( x ∗ α 1 + y ∗ α 2 ) , d π α 1 = 0 , d π α 2 = α 1 ∧ α 2 , d π d z i = 0. d_{\pi}f(m_*)=f_x(m_*)(x_*\alpha_1+y_*\alpha_2),\qquad d_{\pi}\alpha_1=0,\qquad d_{\pi}\alpha_2=\alpha_1\wedge\alpha_2,\qquad d_{\pi}dz_i=0. d π f ( m ∗ ) = f x ( m ∗ ) ( x ∗ α 1 + y ∗ α 2 ) , d π α 1 = 0 , d π α 2 = α 1 ∧ α 2 , d π d z i = 0.
The identity c p , j + 1 = ( − 1 ) p c p , j c_{p,j+1}=(-1)^pc_{p,j} c p , j + 1 = ( − 1 ) p c p , j gives
d v ι = ι Q v (52) d_v\iota=\iota Q_v \tag*{(52)} d v ι = ι Q v (52)
Since h 0 ι = 0 h_0\iota=0 h 0 ι = 0 , also h 0 d v ι = 0 h_0d_v\iota=0 h 0 d v ι = 0 . Thus
i h s c = ι , D t r s c = Q . (53) i_h^{\mathrm{sc}}=\iota,\qquad D_{\mathrm{tr}}^{\mathrm{sc}}=Q. \tag*{(53)} i h sc = ι , D tr sc = Q . (53)
It remains to compute the composite ver π s c ι \operatorname{ver}_{\pi}^{\mathrm{sc}}\iota ver π sc ι .
Weighted smooth functions at the unit section
Use root coordinates
u i = ( log a i , b i ) , u 0 = 0 , H i j = h 0 i j , H 0 j = H i i = 0. u_i=(\log a_i,b_i),\qquad u_0=0,\qquad H_{ij}=h_{0ij},\qquad H_{0j}=H_{ii}=0. u i = ( log a i , b i ) , u 0 = 0 , H ij = h 0 ij , H 0 j = H ii = 0.
Set
e i = u i − u i − 1 , c i j = H i j − H i − 1 , j − H i , j − 1 + H i − 1 , j − 1 . (54) e_i=u_i-u_{i-1},\qquad c_{ij}=H_{ij}-H_{i-1,j}-H_{i,j-1}+H_{i-1,j-1}. \tag*{(54)} e i = u i − u i − 1 , c ij = H ij − H i − 1 , j − H i , j − 1 + H i − 1 , j − 1 . (54)
The inverse is
u i = ∑ a ≤ i e a , H i j = ∑ a ≤ i < b ≤ j c a b . u_i=\sum_{a\leq i}e_a,\qquad H_{ij}=\sum_{a\leq i<b\leq j}c_{ab}. u i = a ≤ i ∑ e a , H ij = a ≤ i < b ≤ j ∑ c ab .
The degeneracy repeating vertices k − 1 , k k-1,k k − 1 , k is exactly
S k = { e k = 0 , c i k = 0 ( i < k ) , c k j = 0 ( k < j ) } . (55) S_k=\{e_k=0,\ c_{ik}=0\ (i<k),\ c_{kj}=0\ (k<j)\}. \tag*{(55)} S k = { e k = 0 , c ik = 0 ( i < k ) , c k j = 0 ( k < j )} . (55)
Indeed the equations recover repeated cumulative u u u -entries, repeated rows and columns of H H H , and H k − 1 , k = 0 H_{k-1,k}=0 H k − 1 , k = 0 ; deleting that slot gives the lower simplex.
Assign each component of e i e_i e i weight one and label i i i , and assign c i j c_{ij} c ij weight two and labels i , j i,j i , j . Let W ≥ d \mathcal{W}^{\ge d} W ≥ d be the smooth unit-normal germs whose Taylor coefficients of weighted degree less than d d d vanish. Base functions have weight zero.
Lemma 9.7 (The weighted function bound). For every n ≥ 1 n \ge1 n ≥ 1 ,
C N n ( G ) ∩ W ≥ n + 1 ⊆ J n \mathcal{C}^{n}_{N}(G) \cap\mathcal{W}^{\ge n+1} \subseteq J_n C N n ( G ) ∩ W ≥ n + 1 ⊆ J n at unit-germ level and, after germ localization, globally. In particular,
ve n ( C N n ( G ) ∩ W ≥ n + 1 ) = 0. \operatorname{ve}_{n}\left(\mathcal{C}^{n}_{N}(G) \cap\mathcal{W}^{\ge n+1}\right)=0. ve n ( C N n ( G ) ∩ W ≥ n + 1 ) = 0. Every normalized degree-n n n function has normal weight at least n n n .
Proof. Choose a small normal-coordinate polydisc, with base parameters, inside G n G_n G n , stable under setting any subset of the coordinates in (54) to zero. Let P k P_k P k zero all coordinates carrying label k k k . The maps commute and retract onto (55) . Each P j ∗ P_j^* P j ∗ sends a generator of I ( S k ) I(S_k) I ( S k ) to itself or to zero, so preserves I ( S k ) 2 I(S_k)^2 I ( S k ) 2 . The operator
N n = ∏ k = 1 n ( 1 − P k ∗ ) \mathcal{N}_n=\prod_{k=1}^{n}(1-P_k^*) N n = k = 1 ∏ n ( 1 − P k ∗ )
normalizes functions and fixes normalized functions.
For a normalized f f f of weight at least n + 1 n+1 n + 1 , ordinary Taylor expansion gives
f ( ξ ) = T n f ( ξ ) + ∑ ∣ γ ∣ = n + 1 ξ γ R γ ( ξ ) , f(\xi)=T_n f(\xi)+\sum_{|\gamma|=n+1}\xi^\gamma R_\gamma(\xi), f ( ξ ) = T n f ( ξ ) + ∣ γ ∣ = n + 1 ∑ ξ γ R γ ( ξ ) ,
R γ ( ξ ) = n + 1 γ ! ∫ 0 1 ( 1 − s ) n ∂ γ f ( s ξ ) d s . R_\gamma(\xi)=\frac{n+1}{\gamma!}\int_0^1(1-s)^n\partial^\gamma f(s\xi)\,\mathrm{d}s. R γ ( ξ ) = γ ! n + 1 ∫ 0 1 ( 1 − s ) n ∂ γ f ( s ξ ) d s .
The remainders are smooth with all base and passive parameter derivatives. Restriction to a coordinate plane commutes with Taylor truncation. Hence T n f T_n f T n f is normalized, and every surviving coordinate monomial covers all n n n labels. Its weight is its number of label occurrences. Weight greater than n n n forces one label to occur twice, placing the monomial in a selected I ( S k ) 2 I(S_k)^2 I ( S k ) 2 .
Each remainder monomial has n + 1 n+1 n + 1 coordinate factors, so it also repeats a label. Its smooth multiple need not be normalized separately. Applying N n \mathcal{N}_n N n gives
N n ( ξ γ R γ ) ∈ C N n ∩ I ( S k ( γ ) ) 2 . \mathcal{N}_n(\xi^\gamma R_\gamma)\in\mathcal{C}^{n}_{N}\cap I(S_{k(\gamma)})^2. N n ( ξ γ R γ ) ∈ C N n ∩ I ( S k ( γ ) ) 2 .
because every factor of N n \mathcal{N}_n N n preserves the selected squared ideal. Applying the normalizer to the Taylor formula therefore gives exactly a sum of separately normalized squared-ideal summands.
Patch these local expressions near the unit section using a locally finite family of cutoffs supported inside their coordinate domains. The coordinate generators are finite in each arity. Grouping the locally finite coefficients by their finitely many repeated-label generator pairs gives global smooth squared-ideal summands, still normalized. The remaining normalized difference vanishes near the whole unit section and belongs to J n J_n J n by [2 Lemma 3.1.5] .
Finally, a coordinate monomial of weight less than n n n cannot cover every label. Applying the same restriction argument to the low Taylor polynomial shows that such terms do not occur in a normalized function. □ \square □
We apply the lemma entrywise after the column retraction, when the outputs are normalized functions. The full differentiating ideal remains J + δ f u l l J J+\delta_{\mathrm{full}}J J + δ full J , with its allowed-transposition relations.
Filteredness of the prescribed operators
Give relative coordinate differentials the weights of their coordinates. Equivalently, α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 have weight one and d z i dz_i d z i weight two. The change between these frames is filtered:
α 1 = d π u 1 , α 2 = d π u 2 − b d π u 1 . \alpha_1=d_{\pi}u^1,\qquad\alpha_2=d_{\pi}u^2-b\,d_{\pi}u^1. α 1 = d π u 1 , α 2 = d π u 2 − b d π u 1 .
The term b d π u 1 b\,d_{\pi}u^1 b d π u 1 has weight two.
Lemma 9.8 (Filtered operators). All maps used in (42) and (49) , as well as A , A − 1 \mathcal{A},\mathcal{A}^{-1} A , A − 1 , preserve normal weight. The assertion uses the fixed charts ψ m \psi_m ψ m and the fixed horizontal homotopy h 0 h_0 h 0 .
Proof. Group multiplication and inversion have additive linear part in ( log a , b ) (\log a,b) ( log a , b ) and higher terms of weight at least two. The H H H -coordinates transform linearly through triangle identities. Root-changing group formulas contain exp ( − H Z ) \exp(-HZ) exp ( − H Z ) , whose nonconstant terms have weight at least two. Base movement satisfies f ( a i m 0 ) = f ( m 0 ) + W ≥ 1 f(a_i m_0)=f(m_0)+W^{\geq1} f ( a i m 0 ) = f ( m 0 ) + W ≥ 1 . Thus every genuine face and degeneracy is filtered. Equation (44) is compatible: Z m ∗ d z 1 Z_{m_*}dz_1 Z m ∗ d z 1 has weight two, not one.
Write χ m ( u ) = ψ m ( g ( u ) ) \chi_m(u)=\psi_m(g(u)) χ m ( u ) = ψ m ( g ( u )) . The actual radial group-coordinate map is χ m − 1 ( s χ m ( u ) ) \chi_m^{-1}(s\chi_m(u)) χ m − 1 ( s χ m ( u )) . Its value at u = 0 u=0 u = 0 is zero and its derivative there is s 1 s\,1 s 1 . Smooth Taylor expansion, uniformly for s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] locally in m m m , gives
χ m − 1 ( s χ m ( u ) ) = s u + W ≥ 2 . \chi_m^{-1}(s\chi_m(u))=su+W^{\geq2}. χ m − 1 ( s χ m ( u )) = s u + W ≥ 2 .
The d s ds d s -coefficient and relative-differential coefficients of the pulled-back group coordinates have weight at least one. For a central coordinate,
d ( s z i ) = s d z i + z i d s d(s z_i)=s\,dz_i+z_i\,ds d ( s z i ) = s d z i + z i d s
has weight two in both terms. Contraction with ∂ s \partial_s ∂ s and compact integration therefore preserve weight. This proves filteredness of the k p k_p k p determined by the fixed charts.
The root-appending evaluation is also filtered. It retains prefix data: for example, e p + 1 = − u p e_{p+1}=-u_p e p + 1 = − u p there, and c i , p + 1 = − H i , p + H i − 1 , p c_{i,p+1}=-H_{i,p}+H_{i-1,p} c i , p + 1 = − H i , p + H i − 1 , p . These terms have their original weights.
For (45) , the order map sends a rooted group coordinate to a rooted group coordinate or zero, and an H H H -coordinate to an H H H -coordinate or zero. On the formal frame, codegeneracies preserve the weight-one e e e -vectors and the weight-two v v v -vectors or send them to zero. Thus L L L is filtered. Genuine normalization gives the same property for L N L_N L N .
The map p 0 p_0 p 0 evaluates coefficient functions at the constant simplex while retaining their formal frame tensor. Positive-weight coefficients vanish; surviving frame weights remain unchanged. The filtered frame isomorphism ensures that this conclusion does not depend on a chosen expression of the form. The map i 0 i_0 i 0 reevaluates coefficients at m ∗ m_\ast m ∗ and multiplies by the prescribed frame monomial of exactly the CE weight. Hence i 0 i_0 i 0 , p 0 p_0 p 0 , P 0 P_0 P 0 , a a a , h 0 h_0 h 0 are filtered.
Normalization and the preceding function lemma give r k ∈ W ≥ k r_k \in W^{\geq k} r k ∈ W ≥ k for k ≥ 2 k \geq2 k ≥ 2 , and strict units give r 1 − 1 ∈ W ≥ 1 r_1 - 1 \in W^{\geq1} r 1 − 1 ∈ W ≥ 1 . The same estimates apply to U 0 − 1 U_0 - 1 U 0 − 1 and its inverse, using the exact identity U 0 − 1 − 1 = − U 0 − 1 ( U 0 − 1 ) U_0^{-1} - 1 = -U_0^{-1}(U_0 - 1) U 0 − 1 − 1 = − U 0 − 1 ( U 0 − 1 ) . Positive components of U − 1 U^{-1} U − 1 are finite products. The relative derivative is filtered because a coordinate and its differential have the same weight. It follows that Δ π \Delta_\pi Δ π , both conjugations, and ∇ U \nabla_U ∇ U are filtered. Every transfer expression is finite in its established relative-degree bound, so every resulting operator is filtered.
For clarity, the associated symbol complexes consist of fixed-weight polynomial normal germs, with smooth coefficients in A A A . The relative symbol uses the genuine P P P -normalization. Finite normalization supplies polynomial representatives. This is a weighted associated graded, not an ordinary tangent linearization or a completed power series construction. The scalar CE component B O n B_O^n B O n is concentrated in weight n n n , so its weight-preserving differential is zero. The raw and relative symbol differentials generally remain nonzero.
The prescribed scalar retraction is the identity
Theorem 9.9 (Scalar normal-symbol calibration). The actual map
τ O = ver π s c ι : ( B O , Q ) ⟶ ( B O , Q ) \tau_O = \operatorname{ver}_{\pi}^{\mathrm{sc}} \iota: (B_O,Q) \longrightarrow(B_O,Q) τ O = ver π sc ι : ( B O , Q ) ⟶ ( B O , Q ) is literally the identity in every degree.
Proof. The column SDR identities of Proposition 9.3 descend to the weighted polynomial symbol:
r π , l i n π l i n ∗ = 1 , π l i n ∗ r π , l i n = 1 − D l i n H π , l i n − H π , l i n D l i n . r_{\pi,\mathrm{lin}}\pi_{\mathrm{lin}}^{*}=1,\qquad
\pi_{\mathrm{lin}}^{*}r_{\pi,\mathrm{lin}}=1-D_{\mathrm{lin}}H_{\pi,\mathrm{lin}}-H_{\pi,\mathrm{lin}}D_{\mathrm{lin}}. r π , lin π lin ∗ = 1 , π lin ∗ r π , lin = 1 − D lin H π , lin − H π , lin D lin . The map π l i n ∗ \pi_{\mathrm{lin}}^{*} π lin ∗ is a DGA map.
We check that the leading inclusion is a DGA map with the phase in (9.10). Take f η p θ I f\eta^p\theta_I f η p θ I and g η p ′ θ J g\eta^{p'}\theta_J g η p ′ θ J , with j = ∣ I ∣ j=|I| j = ∣ I ∣ , j ′ = ∣ J ∣ j'=|J| j ′ = ∣ J ∣ . For p > 0 p>0 p > 0 , the back central differentials pull to d z p + a − d z p dz_{p+a}-dz_p d z p + a − d z p . Their terms containing d z p dz_p d z p vanish against the front central block. The leading back α \alpha α -symbol is the same α \alpha α ; its Z d z p Z\,dz_p Z d z p correction has higher weight. The relative product contributes ( − 1 ) ( p + j ) p ′ (-1)^{(p+j)p'} ( − 1 ) ( p + j ) p ′ , and moving the back α J \alpha_J α J past the front p p p central differentials contributes ( − 1 ) p j ′ (-1)^{pj'} ( − 1 ) p j ′ . The exact identity
c p , j c p ′ , j ′ ( − 1 ) ( p + j ) p ′ + p j ′ = c p + p ′ , j + j ′ c_{p,j}c_{p',j'}(-1)^{(p+j)p'+pj'}=c_{p+p',j+j'} c p , j c p ′ , j ′ ( − 1 ) ( p + j ) p ′ + p j ′ = c p + p ′ , j + j ′ follows from
p ( p + 1 ) 2 + p ′ ( p ′ + 1 ) 2 + p p ′ = ( p + p ′ ) ( p + p ′ + 1 ) 2 . \frac{p(p+1)}{2}+\frac{p'(p'+1)}{2}+pp'
=\frac{(p+p')(p+p'+1)}{2}. 2 p ( p + 1 ) + 2 p ′ ( p ′ + 1 ) + p p ′ = 2 ( p + p ′ ) ( p + p ′ + 1 ) . The remaining signs are exactly those of exterior multiplication. Overlapping odd generators give zero. If p = 0 p=0 p = 0 , the back map has no preceding root shift and no d z 0 dz_0 d z 0 occurs. If p ′ = 0 p'=0 p ′ = 0 , the back central block is empty. If both are zero, this is ordinary exterior multiplication. Thus ι l i n \iota_{\mathrm{lin}} ι lin is a unital DGA map in every degree.
The induced ve l i n \operatorname{ve}_{\mathrm{lin}} ve lin is the DGA quotient furnished by the actual differentiation, including δ J \delta J δ J and its transposition relations. On cohomology, H ( r π , l i n ) = H ( π l i n ∗ ) − 1 H(r_{\pi,\mathrm{lin}})=H(\pi^{*}_{\mathrm{lin}})^{-1} H ( r π , lin ) = H ( π lin ∗ ) − 1 is an algebra map. Both scalar CE symbol endpoints have zero differential. Consequently their actual composite
τ l i n = ve l i n r π , l i n ι l i n \tau_{\mathrm{lin}}=\operatorname{ve}_{\mathrm{lin}}r_{\pi,\mathrm{lin}}\iota_{\mathrm{lin}} τ lin = ve lin r π , lin ι lin
is an algebra map. Multiplicativity here follows from the induced cohomology map; the column retraction r π r_{\pi} r π is used as a chain map.
In the following calculation, δ W \delta_{W} δ W is the weighted associated-graded map induced by the full horizontal differential on normalized scalar relative forms on P p / G p P_{p}/G_{p} P p / G p :
δ f u l l = ∑ i = 0 p + 1 ( − 1 ) i d i ∗ . \delta_{\mathrm{full}}=\sum_{i=0}^{p+1}(-1)^{i}d_{i}^{*}. δ full = i = 0 ∑ p + 1 ( − 1 ) i d i ∗ .
The original G G G -face d p + 2 d_{p+2} d p + 2 is omitted. It sends bidegree ( p , r ) (p,r) ( p , r ) to ( p + 1 , r ) (p+1,r) ( p + 1 , r ) , with no additional relative-form sign. Unlike the scalar CE symbol differential, this horizontal differential need not vanish. The index a ∈ { 1 , 2 } a\in\{1,2\} a ∈ { 1 , 2 } in u a u_{a} u a is a group-coordinate index, and u 1 , a u_{1,a} u 1 , a is that coordinate of the first prefix root arrow.
It remains to calibrate generators. Functions are sent to themselves by root-appending evaluation. For a group coordinate, at horizontal degree zero,
H π , 0 ( d u a ) = u a , δ W u a = − u 1 , a H_{\pi,0}(du_{a})=u_{a},\qquad\delta_{W}u_{a}=-u_{1,a} H π , 0 ( d u a ) = u a , δ W u a = − u 1 , a
in the symbol model. The next homotopy vanishes on a relative zero-form, so
r π , l i n ( d u a ) = u 1 , a . r_{\pi,\mathrm{lin}}(du_{a})=u_{1,a}. r π , lin ( d u a ) = u 1 , a .
(33) gives τ l i n ( θ i ) = θ i \tau_{\mathrm{lin}}(\theta_{i})=\theta_{i} τ lin ( θ i ) = θ i .
For η \eta η , the inclusion is − d z 1 -dz_{1} − d z 1 . Since H π , 0 = − k 1 H_{\pi,0}=-k_{1} H π , 0 = − k 1 in horizontal degree one,
H π , 0 ( − d z 1 ) = z 1 . H_{\pi,0}(-dz_{1})=z_{1}. H π , 0 ( − d z 1 ) = z 1 .
On P 2 = G 3 P_{2}=G_{3} P 2 = G 3 , put H 12 = h 012 H_{12}=h_{012} H 12 = h 012 , z 1 = h 013 z_{1}=h_{013} z 1 = h 013 , z 2 = h 023 z_{2}=h_{023} z 2 = h 023 . The tetrahedral equation gives h 123 = H 12 + z 2 − z 1 h_{123}=H_{12}+z_{2}-z_{1} h 123 = H 12 + z 2 − z 1 , hence
δ W z 1 = h 123 − h 023 + h 013 = H 12 . \delta_{W}z_{1}=h_{123}-h_{023}+h_{013}=H_{12}. δ W z 1 = h 123 − h 023 + h 013 = H 12 .
The column projection correction has a minus sign, so r π , l i n ( − d z 1 ) = − H 12 r_{\pi,\mathrm{lin}}(-dz_{1})=-H_{12} r π , lin ( − d z 1 ) = − H 12 . Its differentiated image is + η +\eta + η . Thus τ l i n \tau_{\mathrm{lin}} τ lin fixes A , θ 1 , θ 2 , η A,\theta_{1},\theta_{2},\eta A , θ 1 , θ 2 , η , and is the identity on the entire polynomial CE algebra, including every power of η \eta η .
Degree zero was already settled by the identity on A A A . For n ≥ 1 n\geq1 n ≥ 1 and x ∈ B O n x\in B_{O}^{n} x ∈ B O n , the actual r π s c ι ( x ) r_{\pi}^{\mathrm{sc}}\iota(x) r π sc ι ( x ) is a normalized raw degree-n n n cochain of weight at least n n n . Its weight-n n n raw symbol maps to x x x under ve l i n \operatorname{ve}_{\mathrm{lin}} ve lin . Lemma 9.7 annihilates the smooth remainder of weight n + 1 n+1 n + 1 . Therefore ve l i n r π s c ι ( x ) = x \operatorname{ve}_{\mathrm{lin}}r_{\pi}^{\mathrm{sc}}\iota(x)=x ve lin r π sc ι ( x ) = x . Only a finite normal symbol is used. □
The genuine coefficient diagonal
Theorem 9.10 (Identity internal diagonal). For every invertible-edge pair R , S R,S R , S , the fixed chain map
T = F r π A − 1 i h T=Fr_{\pi}\mathcal{A}^{-1}i_{h} T = F r π A − 1 i h
has identity internal-degree-preserving diagonal under the graded identification
B R , S = B O ⊗ A H . \mathbb{B}_{R,S}=\mathbb{B}_{O}\otimes_{A}\mathcal{H}. B R , S = B O ⊗ A H .
Proof. Every factor is internally nonincreasing by Proposition 9.6. Thus the diagonal of a composite is the composite of its diagonals. This is a statement about operators, not about an associated-graded differential.
Work in local homogeneous frames and take ω X \omega X ω X , where ω ∈ B O n \omega\in B_{O}^{n} ω ∈ B O n and X X X is an arbitrary constant homogeneous rectangular matrix of internal degree q q q . All smooth coefficients are included in ω \omega ω .
On σ ∈ P p \sigma\in P_{p} σ ∈ P p , the two diagonal endpoint transports are
L S = r S , 1 ( σ [ 0 , p + 1 ] ) , L R = r R , 1 ( σ [ p , p + 1 ] ) . L_{S}=r_{S,1}\bigl(\sigma[0,p+1]\bigr),\qquad L_{R}=r_{R,1}\bigl(\sigma[p,p+1]\bigr). L S = r S , 1 ( σ [ 0 , p + 1 ] ) , L R = r R , 1 ( σ [ p , p + 1 ] ) .
They are distinct pullbacks. Diagonal conjugation is X 0 − 1 ( χ ) = L S χ L R − 1 \mathcal{X}_{0}^{-1}(\chi)=L_{S}\chi L_{R}^{-1} X 0 − 1 ( χ ) = L S χ L R − 1 . Put
κ S = L S − 1 d π L S , κ R = L R − 1 d π L R . \kappa_{S}=L_{S}^{-1}d_{\pi}L_{S},\qquad\kappa_{R}=L_{R}^{-1}d_{\pi}L_{R}. κ S = L S − 1 d π L S , κ R = L R − 1 d π L R .
For tridegree ( p , r , q ) (p,r,q) ( p , r , q ) , differentiating this expression gives
∇ U , 0 χ = d v χ + κ S ⋆ π χ − ( − 1 ) p + r + q χ ⋆ π κ R . (56) \nabla_{U,0}\chi=d_{v}\chi+\kappa_{S}\star_{\pi}\chi-(-1)^{p+r+q}\chi\star_{\pi}\kappa_{R}. \tag*{(56)} ∇ U , 0 χ = d v χ + κ S ⋆ π χ − ( − 1 ) p + r + q χ ⋆ π κ R . (56)
The left product has phase ( − 1 ) p (-1)^{p} ( − 1 ) p and the right product phase ( − 1 ) q (-1)^{q} ( − 1 ) q . Thus in bare matrix-valued wedge notation the same formula is
( − 1 ) p ( d π χ + κ S ∧ b a r e χ − ( − 1 ) r χ ∧ b a r e κ R ) . (-1)^{p}\left(d_{\pi}\chi+\kappa_{S}\wedge_{\mathrm{bare}}\chi-(-1)^{r}\chi\wedge_{\mathrm{bare}}\kappa_{R}\right). ( − 1 ) p ( d π χ + κ S ∧ bare χ − ( − 1 ) r χ ∧ bare κ R ) .
This verifies both endpoints for every integer q q q . Strict units give L R , L S = 1 + W ≥ 1 L_{R},L_{S}=1+\mathcal{W}^{\geq1} L R , L S = 1 + W ≥ 1 , so κ R , κ S \kappa_{R},\kappa_{S} κ R , κ S have weight at least one. Hence C = ∇ U , 0 − d v C=\nabla_{U,0}-d_{v} C = ∇ U , 0 − d v raises weight.
Using i h = ( 1 + h 0 ∇ U ) − 1 i 0 i_{h}=(1+h_{0}\nabla_{U})^{-1}i_{0} i h = ( 1 + h 0 ∇ U ) − 1 i 0 , consider its diagonal finite expansion. An all-d v d_{v} d v positive correction vanishes at its rightmost factor because
h 0 d v i 0 ( ω X ) = h 0 i 0 ( Q v ω X ) = 0. h_{0}d_{v}i_{0}(\omega X)=h_{0}i_{0}(Q_{v}\omega X)=0. h 0 d v i 0 ( ω X ) = h 0 i 0 ( Q v ω X ) = 0.
Every remaining term contains a C C C , so
diag q i h ( ω X ) = ι ( ω ) X + E 1 , E 1 ∈ W ≥ n + 1 . \operatorname{diag}_{q}i_{h}(\omega X)=\iota(\omega)X+E_{1},\qquad E_{1}\in\mathcal{W}^{\geq n+1}. diag q i h ( ω X ) = ι ( ω ) X + E 1 , E 1 ∈ W ≥ n + 1 .
The body differential remains in (48) . It is absent from these perturbation words because h 0 h_{0} h 0 was normalized using only δ f u l l \delta_{\mathrm{full}} δ full .
Next, the exact identity
L S χ L R − 1 − χ = ( L S − 1 ) χ L R − 1 + χ ( L R − 1 − 1 ) L_{S}\chi L_{R}^{-1}-\chi=(L_{S}-1)\chi L_{R}^{-1}+\chi(L_{R}^{-1}-1) L S χ L R − 1 − χ = ( L S − 1 ) χ L R − 1 + χ ( L R − 1 − 1 )
shows that diagonal conjugation differs from the scalar identity by one additional normal weight.
Finally, write r 1 = 1 + ρ r_{1}=1+\rho r 1 = 1 + ρ in these local frames. The diagonal of the column perturbation is
Δ π , 0 = δ f u l l + E , \Delta_{\pi,0}=\delta_{\mathrm{full}}+\mathcal{E}, Δ π , 0 = δ full + E ,
E χ = ρ S π ⋆ π χ − ( − 1 ) p + r + q χ ⋆ π ρ R π . \mathcal{E}\chi=\rho_{S}^{\pi}\star_{\pi}\chi-(-1)^{p+r+q}\chi\star_{\pi}\rho_{R}^{\pi}. E χ = ρ S π ⋆ π χ − ( − 1 ) p + r + q χ ⋆ π ρ R π .
The right unit has coefficient − ( − 1 ) p + r + q ( − 1 ) r + q = ( − 1 ) p + 1 -(-1)^{p+r+q}(-1)^{r+q}=(-1)^{p+1} − ( − 1 ) p + r + q ( − 1 ) r + q = ( − 1 ) p + 1 , exactly the scalar last-face sign. Since ρ \rho ρ has weight at least one, E \mathcal{E} E raises weight. All r k r_{k} r k , k ≥ 2 k\geq2 k ≥ 2 , lower internal degree.
Compare the finite column formulas with Δ π , 0 = δ f u l l + E \Delta_{\pi,0}=\delta_{\mathrm{full}}+\mathcal{E} Δ π , 0 = δ full + E and with δ f u l l \delta_{\mathrm{full}} δ full alone. Every difference word contains an E \mathcal{E} E ; all other factors preserve weight. Consequently
diag q ( r π I i h − 1 ) ( ω X ) = r π s c ι ( ω ) X + E , E ∈ W ≥ n + 1 . \operatorname{diag}_{q}\left(r_{\pi}\mathcal{I}^{-1}_{ih}\right)(\omega X)=r_{\pi}^{\mathrm{sc}}\iota(\omega)X+E,\qquad E\in W^{\geq n+1}. diag q ( r π I ih − 1 ) ( ω X ) = r π sc ι ( ω ) X + E , E ∈ W ≥ n + 1 .
This is an entrywise comparison of Hom operators.
The final difference is a normalized function-valued raw degree-n n n cochain: both complete outputs are normalized. Individual intermediate errors need not be normalized, nor preserve every squared ideal. For n ≥ 1 n\geq1 n ≥ 1 , Lemma 9.7 annihilates E E E . For n = 0 n=0 n = 0 , the whole space G 0 = O G_0=O G 0 = O is the unit section, so the weight-one function E E E is identically zero. In either case, Theorem 9.9 gives
diag q T ( ω X ) = τ O ( ω ) X = ω X . \operatorname{diag}_{q}T(\omega X)=\tau_O(\omega)X=\omega X. diag q T ( ω X ) = τ O ( ω ) X = ω X .
To globalize, fix a point of the unit section. There are finitely many operations in each total degree. All their geometric maps take units to units, uniformly over the compact radial parameter. Shrink the normal neighborhood so that every intermediate vertex and radial image lies in the chosen source and target frame patches. The radial chart is the restriction of the fixed ψ O \psi_O ψ O , not a new chart for the patch. Testing all smooth scalar coefficients times the constant homogeneous matrix units tests all local sections. Such sections can be extended near the point using cutoffs, and the differentiated output depends only on their germs.
Frame agreement at a unit is per bundle, not an identification of source and target transition matrices. Frame and projector variation can enter the connection, conjugation, edge, and coefficient terms; all are included in the estimates above. (37) then gives the identity on the original global projective summands. □
Remark 9.11. The transferred and cohesive differentials act on the same graded module. The comparison between them is constructed in real vector spaces, and its identity diagonal will give a real-linear chain inverse in Proposition 10.1. Thus the differentials are related by conjugation with this particular T T T , rather than identified in advance.
All operators preserve jointly smooth passive finite-dimensional parameters under fixed global body bounds. The later choice of kernel primitives and counits is made objectwise.
The all-open affine Riemann–Hilbert theorem
We now pass from the prescribed transfer to the actual raw Hom complexes and then to the actual full-MC integration functor. Throughout, the notation Rep s m p e r f \operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}} Rep sm perf means the globally bounded finite-projective raw category defined in Section 4. The body cohomology ranks may jump.
Proposition 10.1 (Finite chain inverse of the fixed transfer). For an invertible-edge pair R , S R,S R , S , put
[ ℓ , u ] = [ a S − b R , b S − a R ] . [\ell,u]=[a_S-b_R,b_S-a_R]. [ ℓ , u ] = [ a S − b R , b S − a R ] . Under the fixed graded Hom identification,
T = 1 + N , N u − ℓ + 1 = 0 , T=1+N,\qquad N^{u-\ell+1}=0, T = 1 + N , N u − ℓ + 1 = 0 ,
and
T − 1 = ∑ j = 0 u − ℓ ( − N ) j (57) T^{-1}=\sum_{j=0}^{u-\ell}(-N)^j \tag*{(57)} T − 1 = j = 0 ∑ u − ℓ ( − N ) j (57)
is a two-sided real-linear chain inverse.
Proof. Proposition 9.6 makes T T T internally nonincreasing, and Theorem 9.10 makes its diagonal the identity. Thus N = T − 1 N=T-1 N = T − 1 strictly lowers internal degree. After u − ℓ + 1 u-\ell+1 u − ℓ + 1 iterations its value lies below the available interval, so it is zero. The finite geometric sum is therefore a two-sided graded inverse.
The already proved chain-map identity is d C o h T = T D t r d_{\mathrm{Coh}}T=TD_{\mathrm{tr}} d Coh T = T D tr . Multiplication by the graded inverse gives T − 1 d C o h = D t r T − 1 T^{-1}d_{\mathrm{Coh}}=D_{\mathrm{tr}}T^{-1} T − 1 d Coh = D tr T − 1 . Thus (57) is a chain inverse. The argument does not require either differential to preserve the internal-degree filtration. □
Theorem 10.2 (All-pair actual Hom comparison). For every ordinary open O ⊆ R 2 O\subseteq\mathbb{R}^2 O ⊆ R 2 and every ordered pair of actual raw coefficients R , S R,S R , S ,
F R , S : Hom r a w ∙ ( R , S ) ⟶ Hom C o h ( B O ) ∙ ( F R , F S ) F_{R,S}\colon\operatorname{Hom}_{\mathrm{raw}}^{\bullet}(R,S)\longrightarrow\operatorname{Hom}_{\mathrm{Coh}(B_O)}^{\bullet}(FR,FS) F R , S : Hom raw ∙ ( R , S ) ⟶ Hom Coh ( B O ) ∙ ( F R , F S ) is a quasi-isomorphism in every integer degree. Its actual global smooth kernel is acyclic.
Proof. First suppose the edges are actual chain isomorphisms. The map S m o d S_{\mathrm{mod}} S mod of Proposition 9.6 is a chain-homotopy equivalence, and T T T is a chain isomorphism by Proposition 10.1 . Therefore T S m o d TS_{\mathrm{mod}} T S mod is a quasi-isomorphism. Equation (51) identifies its map on cohomology with that of the actual F R , S F_{R,S} F R , S .
For arbitrary original edges, take the closed rectification isomorphisms
E R : R e ⟶ R , E S : S e ⟶ S E_R\colon R^{\mathrm e}\longrightarrow R,\qquad E_S\colon S^{\mathrm e}\longrightarrow S E R : R e ⟶ R , E S : S e ⟶ S of Theorem 8.7 . They give chain isomorphisms
ϕ ⟼ E S − 1 ⋆ ϕ ⋆ E R \phi\longmapsto E_S^{-1}\star\phi\star E_R ϕ ⟼ E S − 1 ⋆ ϕ ⋆ E R on raw Homs, and
Φ ⟼ F ( E S ) − 1 Φ F ( E R ) \Phi\longmapsto F(E_S)^{-1}\Phi F(E_R) Φ ⟼ F ( E S ) − 1 Φ F ( E R ) on cohesive Homs. Strict dg differentiation gives the commuting square
Hom r a w ( R , S ) → F R , S Hom C o h ( F R , F S ) ≅ ↓ ↓ ≅ Hom r a w ( R e , S e ) → F R e , S e Hom C o h ( F R e , F S e ) . \begin{CD}
\operatorname{Hom}_{\mathrm{raw}}(R,S) @>{F_{R,S}}>> \operatorname{Hom}_{\mathrm{Coh}}(FR,FS) \\
@V{\cong}VV @VV{\cong}V \\
\operatorname{Hom}_{\mathrm{raw}}(R^{\mathrm e},S^{\mathrm e}) @>{F_{R^{\mathrm e},S^{\mathrm e}}}>> \operatorname{Hom}_{\mathrm{Coh}}(FR^{\mathrm e},FS^{\mathrm e}).
\end{CD} Hom raw ( R , S ) ≅ ↓ ⏐ Hom raw ( R e , S e ) F R , S F R e , S e Hom Coh ( F R , F S ) ↓ ⏐ ≅ Hom Coh ( F R e , F S e ) . The differentiated gauges are retained; they need not be identities. The bottom map is a quasi-isomorphism by the invertible-edge case, and hence so is the top map.
Lemma 8.4 supplies a short exact sequence of actual global complexes
0 ⟶ K R , S ⟶ Hom r a w ( R , S ) → F R , S Hom C o h ( F R , F S ) ⟶ 0. 0\longrightarrow K_{R,S}\longrightarrow\operatorname{Hom}_{\mathrm{raw}}(R,S)\xrightarrow{F_{R,S}}\operatorname{Hom}_{\mathrm{Coh}}(FR,FS)\longrightarrow0. 0 ⟶ K R , S ⟶ Hom raw ( R , S ) F R , S Hom Coh ( F R , F S ) ⟶ 0. The long exact cohomology sequence now gives H m ( K R , S ) = 0 H^m(K_{R,S})=0 H m ( K R , S ) = 0 for every integer m m m . The acyclicity therefore holds in the global smooth kernel itself. □
Corollary 10.3 (Calibrated identity-body counits). Every actual raw object R R R admits a global smooth normalized closed isomorphism
q R : I ( F R ) ⟶ R , ( q R ) 0 = 1 , F q R = 1. q_R:I(FR)\longrightarrow R,\qquad(q_R)_0=1,\qquad Fq_R=1. q R : I ( F R ) ⟶ R , ( q R ) 0 = 1 , F q R = 1. The map q R q_R q R is chosen separately for each R R R .
Proof. Put R ′ = I ( F R ) R'=I(FR) R ′ = I ( F R ) . Theorem 8.6 gives F R ′ = F R FR'=FR F R ′ = F R , and the two objects have the same underlying graded bundle and body differential. On that common body the identity zero-cochain has
D R ′ , R 1 = r R − r R ′ ∈ K R ′ , R 1 , D ( D 1 ) = 0. D_{R',R}1=r_R-r_{R'}\in K^1_{R',R},\qquad D(D1)=0. D R ′ , R 1 = r R − r R ′ ∈ K R ′ , R 1 , D ( D 1 ) = 0.
By Theorem 10.2 , choose c ∈ K R ′ , R 0 c\in K^0_{R',R} c ∈ K R ′ , R 0 with D c = D 1 Dc=D1 D c = D 1 . Then q R = 1 − c q_R=1-c q R = 1 − c is closed and F q R = 1 Fq_R=1 F q R = 1 . Since F F F is the identity on body components, c 0 = 0 c_0=0 c 0 = 0 and ( q R ) 0 = 1 (q_R)_0=1 ( q R ) 0 = 1 .
Every positive component of c c c lowers internal degree. If the body’s width is w R = b R − a R w_R=b_R-a_R w R = b R − a R , then c ⋆ ( w R + 1 ) = 0 c^{\star(w_R+1)}=0 c ⋆ ( w R + 1 ) = 0 in the common underlying graded cochain algebra. Form there
q R − 1 = ∑ j = 0 w R c ⋆ j , q_R^{-1}=\sum_{j=0}^{w_R}c^{\star j}, q R − 1 = j = 0 ∑ w R c ⋆ j ,
and assign the resulting cochain the reverse object type R ⟶ R ′ R\longrightarrow R' R ⟶ R ′ . The two inverse identities hold in that graded algebra; the Hom Leibniz rule then makes the inverse closed. The powers of c : R ′ ⟶ R c:R'\longrightarrow R c : R ′ ⟶ R are taken in the common underlying graded cochain algebra before assigning the reverse object type.
Theorem 10.4 (All-open affine Riemann–Hilbert). For every ordinary open O ⊆ R 2 O\subseteq\mathbb{R}^2 O ⊆ R 2 , the actual functors
I = Ψ O ∗ IN E O : Coh ( B O ) ⟶ Rep s m p e r f ( K [ O ] ) I=\Psi_O^*\operatorname{IN}_{E_O}:\operatorname{Coh}(B_O)\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(K[O]) I = Ψ O ∗ IN E O : Coh ( B O ) ⟶ Rep sm perf ( K [ O ]) and
IN E O : Coh ( B O ) ⟶ Rep s m p e r f ( Int ( E O ) ) \operatorname{IN}_{E_O}:\operatorname{Coh}(B_O)\longrightarrow\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(\operatorname{Int}(E_O)) IN E O : Coh ( B O ) ⟶ Rep sm perf ( Int ( E O )) are smooth strictly unital A ∞ A_\infty A ∞ quasi-equivalences. The bodies are globally bounded and degreewise finite projective, with no regularity assumption on their fibre cohomology. Every finite-model raw object has the calibrated actual isomorphism of Corollary 10.3 . Every full-MC raw object has a global smooth normalized closed actual isomorphism from the integration of a cohesive object, with identity body.
Proof. For cohesive objects P , Q P,Q P , Q , literal retraction gives
F I P , I Q I 1 = 1. F_{IP,IQ}I_1=1. F I P , I Q I 1 = 1.
The first map is a quasi-isomorphism by Theorem 10.2 , so I 1 I_1 I 1 is a quasi-isomorphism in every Hom degree. Corollary 10.3 supplies essential surjectivity on H 0 H^0 H 0 . Strict unitality and smoothness are properties of the actual integration functor. Here I 1 I_1 I 1 is used as a chain map; the higher Taylor maps govern compatibility with composition.
For the full-MC target, write
i = Φ O ∗ , p = Ψ O ∗ , p i = 1. i=\Phi_O^*,\qquad p=\Psi_O^*,\qquad pi=1. i = Φ O ∗ , p = Ψ O ∗ , p i = 1.
The range-preserving homotopy of Theorem 7.4 gives, by the pointed prism, closed invertible maps
u X : X ⟶ i p X , ( u X ) 0 = 1. u_X:X\longrightarrow ipX,\qquad(u_X)_0=1. u X : X ⟶ i pX , ( u X ) 0 = 1.
For a homogeneous cochain α : X ⟶ Y \alpha:X\longrightarrow Y α : X ⟶ Y , its all-degree identity is
D P H O α + P H O D α = U Y ⋆ α − i p ( α ) ⋆ U X . D P_{H_O}\alpha+P_{H_O}D\alpha=U_Y\star\alpha-ip(\alpha)\star U_X. D P H O α + P H O D α = U Y ⋆ α − i p ( α ) ⋆ U X .
For objects pulled back from K [ O ] K[O] K [ O ] , the homotopy being over K [ O ] K[O] K [ O ] makes U i R = 1 U_{iR}=1 U i R = 1 . The identity therefore gives the Hom retraction, and arbitrary objects are reduced to these pullbacks by U X U_X U X . This is precisely the all-degree dg quasi-equivalence of Corollary 7.5. Since
p ( I N E O ) 1 = I 1 , p(\mathrm{IN}_{E_O})_1=I_1, p ( IN E O ) 1 = I 1 ,
the actual full-MC unary maps are quasi-isomorphisms.
For essential surjectivity, let X X X be a full-MC raw object, put R = p X R=pX R = pX , and take P = F R P=FR P = F R . Let q K : I ( P ) ⟶ R q_K:I(P)\longrightarrow R q K : I ( P ) ⟶ R be its finite-model counit. The required map is
U X − 1 ⋆ i ( q K ) ⋆ U I N E O ( P ) : I N E O ( P ) ⟶ X (58) U_X^{-1}\star i(q_K)\star U_{\mathrm{IN}_{E_O}(P)}:\mathrm{IN}_{E_O}(P)\longrightarrow X \tag*{(58)} U X − 1 ⋆ i ( q K ) ⋆ U IN E O ( P ) : IN E O ( P ) ⟶ X (58)
Its types are
I N E O ( P ) ⟶ i p I N E O ( P ) = i I ( P ) ⟶ i R = i p X ⟶ X . \mathrm{IN}_{E_O}(P)\longrightarrow ip\mathrm{IN}_{E_O}(P)=iI(P)\longrightarrow iR=ipX\longrightarrow X. IN E O ( P ) ⟶ i p IN E O ( P ) = i I ( P ) ⟶ i R = i pX ⟶ X .
All three factors are closed, normalized, invertible, and have identity body. This proves the stated full-MC reconstruction.
Corollary 10.5. For literal bodies in degrees 0, 1, the actual differentiation kernel vanishes in cohomology in every degree. In particular, this holds separately for pairs whose body differentials are injective on an open dense locus and for pairs whose body differentials are surjective on an open dense locus. The all-pair theorem applies to arbitrary body differentials.
Proof. This is the specialization of Theorem 10.2 .
Remark 10.6 (Coefficient categories and choices). The theorem concerns the global smooth Hom complexes of globally bounded, degreewise finite-projective real cohesive and raw coefficients on each ordinary open. Full-MC simplices retain their whole base image. These are the concrete coefficient categories, rather than Perf ( B O ) \operatorname{Perf}(B_O) Perf ( B O ) or a Hom-sheafified replacement.
The comparison T T T and its inverse conjugate the two Hom differentials as real-linear chain maps. Integration is an A ∞ A_\infty A ∞ -functor, with compatibility with composition governed by its higher Taylor maps. Counits are constructed objectwise, and the sections are chosen separately on each open. Naturality over coefficient moduli and strict compatibility with open restriction are additional conditions on these choices. The kernel argument chooses a primitive in the actual global complex for each closed kernel cochain.
The classification and vertical-family results below assume regular body cohomology and use the minimalization and finite MC-lifting arguments of Appendices B and C. Questions of descent and broader reconstruction are discussed in § 14 .
Nonregular coefficients and a nonrectangular open
Theorem 10.4 permits body cohomology to jump, including at an infinitely flat rank drop. The following examples show why neither regular minimalization nor a generic injectivity or surjectivity hypothesis suffices for the coefficient theorem.
An infinitely flat coefficient with a higher term
On R 2 \mathbb{R}^2 R 2 , define
s ( y ) = { e − 1 / y 2 , y ≠ 0 , 0 , y = 0 , a ( y ) = { e − 1 / y 2 / y , y ≠ 0 , 0 , y = 0. s(y)=
\begin{cases}
e^{-1/y^2}, & y\ne0,\\
0, & y=0,
\end{cases}
\qquad
a(y)=
\begin{cases}
e^{-1/y^2}/y, & y\ne0,\\
0, & y=0.
\end{cases} s ( y ) = { e − 1/ y 2 , 0 , y = 0 , y = 0 , a ( y ) = { e − 1/ y 2 / y , 0 , y = 0 , y = 0.
Then s = y a s=ya s = y a everywhere.
Example 11.1 (Infinitely flat higher-MC coefficient). Let V 0 , V 1 V^0,V^1 V 0 , V 1 be trivial real lines. In homogeneous frames ordered by degrees zero and one, put
d = ( 0 0 1 0 ) , h = ( 0 1 0 0 ) . d=\begin{pmatrix}0&0\\1&0\end{pmatrix},
\qquad
h=\begin{pmatrix}0&1\\0&0\end{pmatrix}. d = ( 0 1 0 0 ) , h = ( 0 0 1 0 ) . In forms-first notation define
D = Q + s ( y ) d + a ( y ) θ 1 1 + η h . (59) \mathcal{D}=Q+s(y)d+a(y)\theta_1 1+\eta h. \tag*{(59)} D = Q + s ( y ) d + a ( y ) θ 1 1 + η h . (59) This is a flat cohesive coefficient. Its body cohomology is zero off y = 0 y=0 y = 0 and has dimension one in each of degrees zero and one on y = 0 y=0 y = 0 .
Proof. Every derivative of s s s or a a a away from zero is a finite sum of terms y − N e − 1 / y 2 y^{-N}e^{-1/y^2} y − N e − 1/ y 2 . These tend to zero faster than every power of y y y . Successively extending each derivative by zero therefore proves that s s s and a a a are smooth and infinitely flat at zero.
The matrices have degrees ∣ d ∣ = 1 |d|=1 ∣ d ∣ = 1 , ∣ h ∣ = − 1 |h|=-1 ∣ h ∣ = − 1 and satisfy
d 2 = h 2 = 0 , d h + h d = 1. d^2=h^2=0,\qquad dh+hd=1. d 2 = h 2 = 0 , d h + h d = 1.
Thus each of the three displayed terms of D − Q \mathcal{D}-Q D − Q has total degree one. Since s , a s,a s , a depend only on y y y , Q s = Q a = 0 Qs=Qa=0 Q s = Q a = 0 , and
Q ( a θ 1 1 ) = − a y η 1. Q(a\theta_1 1)=-ay\eta1. Q ( a θ 1 1 ) = − a y η 1.
The three squares vanish:
( s d ) 2 = 0 , ( a θ 1 1 ) 2 = 0 , ( η h ) 2 = 0. (sd)^2=0,\qquad(a\theta_1 1)^2=0,\qquad(\eta h)^2=0. ( s d ) 2 = 0 , ( a θ 1 1 ) 2 = 0 , ( η h ) 2 = 0.
The mixed products with the one-form term are
( s d ) ( a θ 1 1 ) = − s a θ 1 d , ( a θ 1 1 ) ( s d ) = a s θ 1 d , (sd)(a\theta_1 1)=-sa\theta_1d,\qquad
(a\theta_1 1)(sd)=as\theta_1d, ( s d ) ( a θ 1 1 ) = − s a θ 1 d , ( a θ 1 1 ) ( s d ) = a s θ 1 d ,
and
( a θ 1 1 ) ( η h ) = a θ 1 η h , ( η h ) ( a θ 1 1 ) = − a η θ 1 h . (a\theta_1 1)(\eta h)=a\theta_1\eta h,\qquad
(\eta h)(a\theta_1 1)=-a\eta\theta_1h. ( a θ 1 1 ) ( η h ) = a θ 1 η h , ( η h ) ( a θ 1 1 ) = − a η θ 1 h .
They cancel in pairs. The remaining two products are
( s d ) ( η h ) + ( η h ) ( s d ) = s η ( d h + h d ) = s η 1. (sd)(\eta h)+(\eta h)(sd)=s\eta(dh+hd)=s\eta1. ( s d ) ( η h ) + ( η h ) ( s d ) = sη ( d h + h d ) = sη 1.
Consequently the full curvature is
D 2 = η ( s − a y ) 1 = 0. \mathcal{D}^2=\eta(s-ay)1=0. D 2 = η ( s − a y ) 1 = 0.
The body complex is R → s ( y ) R \mathbb{R} \xrightarrow{s(y)} \mathbb{R} R s ( y ) R in degrees zero and one. For y ≠ 0 y \ne0 y = 0 , this is an isomorphism. For y = 0 y = 0 y = 0 , its differential is zero. The stated fibre-cohomology dimensions follow. These concern the complexes at individual points; cohomology of the global-sections complex is a different calculation. □
The term η h \eta h η h is nonzero and cannot be deleted while leaving the other displayed components fixed: the resulting curvature would be − s η 1 -s\eta1 − sη 1 . The actual integration of (11.1) is covered by Theorem 10.4 , as are its restrictions to arbitrary ordinary opens.
Example 11.2 (A nonregular body which is nowhere injective or surjective). Take the direct sum of Example 11.1 with the trivial coefficient on one additional line in degree zero and one additional line in degree one. Its body differential, in degree-ordered bases, is
( s ( y ) 0 0 0 ) : R 2 → R 2 . \begin{pmatrix} s(y) & 0 \\ 0 & 0 \end{pmatrix} : \mathbb{R}^{2} \to\mathbb{R}^{2}. ( s ( y ) 0 0 0 ) : R 2 → R 2 . This body map is nowhere injective and nowhere surjective. Its kernel and cokernel each have dimension one off y = 0 y = 0 y = 0 and dimension two on y = 0 y = 0 y = 0 . The coefficient is flat and satisfies the all-open Riemann–Hilbert theorem.
Proof. The superconnection is block diagonal, so its curvature is the direct sum of the zero curvature computed above and the zero curvature of the trivial coefficient. The body matrix has rank one for y ≠ 0 y \ne0 y = 0 and rank zero for y = 0 y = 0 y = 0 . The kernel and cokernel dimensions follow immediately. In particular, neither the injectivity locus nor the surjectivity locus is dense; both are empty. The graded body is nevertheless bounded and finite free, so Theorem 10.4 applies. □
These examples lie beyond the regular normal-form classification of § 12 . Together they exhibit rank jumps and a body map with neither an injective nor a surjective locus. The all-pair theorem instead uses a comparison that retains the body differential.
A finite coefficient invisible at the origin
Example 11.3 (Nonconservativity of origin observation). On the whole plane let V 0 = A e 0 V^{0}=Ae_{0} V 0 = A e 0 , V 1 = A e 1 V^{1}=Ae_{1} V 1 = A e 1 , let d ( e 0 ) = e 1 d(e_{0})=e_{1} d ( e 0 ) = e 1 , d ( e 1 ) = 0 d(e_{1})=0 d ( e 1 ) = 0 , and set
P = ( V ⊗ A B , D ) , D = Q + ( y − 1 ) d . P=(V\otimes_{A}B,D), \qquad D=Q+(y-1)d. P = ( V ⊗ A B , D ) , D = Q + ( y − 1 ) d . This is a bounded finite-free cohesive coefficient. Its body complex at the origin is contractible, but P P P is not contractible in Coh ( B ) \operatorname{Coh}(B) Coh ( B ) . Thus the functor which observes only the body complex at the origin does not reflect isomorphisms on the H 0 H^{0} H 0 categories, even on these finite coefficients.
Proof. The curvature is zero because Q ( y − 1 ) = 0 Q(y-1)=0 Q ( y − 1 ) = 0 and d 2 = 0 d^{2}=0 d 2 = 0 . At the origin the body differential is − d -d − d ; the map e 1 ↦ − e 0 e_{1}\mapsto-e_{0} e 1 ↦ − e 0 , e 0 ↦ 0 e_{0}\mapsto0 e 0 ↦ 0 contracts it. At ( 0 , 1 ) (0,1) ( 0 , 1 ) the body differential vanishes on a nonzero two-term graded vector space. A cohesive contraction of P P P would induce a contraction of this body complex, which is impossible. Consequently the map 0 → P 0\to P 0 → P is not a homotopy equivalence, although its origin-body image is one. □
Integration preserves the body, while Theorem 10.4 reflects the nonzero object class of P P P . Thus origin observation still has a nontrivial kernel after integration. A reconstruction through that observer must account for this kernel.
A scalar class on a nonrectangular open
The scalar cohomology of an arbitrary open is not obtained by restricting the full-plane answer. In degree one the difference can already be seen directly.
Proposition 11.4 (Degree-one scalar cohomology). For every ordinary open O ⊆ R 2 O \subseteq\mathbb{R}^{2} O ⊆ R 2 , put ξ = x θ 1 + y θ 2 \xi=x\theta_{1}+y\theta_{2} ξ = x θ 1 + y θ 2 . Then
H 0 ( B O , Q ) = ker ∂ x , H 1 ( B O , Q ) ≅ C ∞ ( O ) / ∂ x C ∞ ( O ) , H^{0}(B_{O},Q)=\ker\partial_{x},\qquad H^{1}(B_{O},Q)\cong C^{\infty}(O)/\partial_{x}C^{\infty}(O), H 0 ( B O , Q ) = ker ∂ x , H 1 ( B O , Q ) ≅ C ∞ ( O ) / ∂ x C ∞ ( O ) , where the degree-one class of b b b is represented by b ξ b\xi b ξ . The latter is a quotient of real vector spaces, not an ideal quotient.
Proof. Direct calculation gives
Q ( P θ 1 + R θ 2 ) = ( − y P + x R ) η + ( − y P x + x R x + R ) θ 1 θ 2 . Q(P\theta_{1}+R\theta_{2})=(-yP+xR)\eta+(-yP_{x}+xR_{x}+R)\theta_{1}\theta_{2}. Q ( P θ 1 + R θ 2 ) = ( − y P + x R ) η + ( − y P x + x R x + R ) θ 1 θ 2 . If − y P + x R = 0 -yP+xR=0 − y P + x R = 0 , then away from the origin there is a unique smooth
b = x P + y R x 2 + y 2 b=\frac{xP+yR}{x^{2}+y^{2}} b = x 2 + y 2 x P + y R with P = x b P=xb P = x b and R = y b R=yb R = y b . If the origin belongs to O O O , choose a rectangle around it. The relation implies P ( 0 , y ) = 0 P(0,y)=0 P ( 0 , y ) = 0 , first for y ≠ 0 y\ne0 y = 0 and then by continuity. Hadamard division on that rectangle gives P = x b P=xb P = x b , and x ( R − y b ) = 0 x(R-yb)=0 x ( R − y b ) = 0 gives R = y b R=yb R = y b everywhere there. Uniqueness glues this local expression to the one away from the origin. For these values the second displayed curvature coefficient vanishes as well. Thus the degree-one cycles are exactly b ξ b\xi b ξ .
Multiplication by ξ \xi ξ is injective on smooth functions: a function killed by both x x x and y y y vanishes away from the origin and then everywhere by continuity. Since Q f = f x ξ Qf=f_{x}\xi Q f = f x ξ , the boundaries are exactly ( ∂ x f ) ξ (\partial_{x}f)\xi ( ∂ x f ) ξ . The same injectivity gives the degree-zero formula.
Example 11.5 (A nonzero class on a punctured open). Let
O = R 2 ∖ { ( 1 , 0 ) } , b ( x , y ) = 1 ( x − 1 ) 2 + y 2 . O=\mathbb{R}^{2}\setminus\{(1,0)\},\qquad b(x,y)=\frac{1}{(x-1)^{2}+y^{2}}. O = R 2 ∖ {( 1 , 0 )} , b ( x , y ) = ( x − 1 ) 2 + y 2 1 . Then [ b ξ ] ≠ 0 [b\xi]\ne0 [ b ξ ] = 0 in H 1 ( B O , Q ) H^{1}(B_{O},Q) H 1 ( B O , Q ) . For an admissible edge from ( x , y ) (x,y) ( x , y ) to ( x ′ , y ) (x',y) ( x ′ , y ) , its actual scalar integration is
I 1 ( b ξ ) ( g , m ) = ∫ x x ′ b ( u , y ) d u = ( x ′ − x ) ∫ 0 1 b ( x + t ( x ′ − x ) , y ) d t . (60) I_{1}(b\xi)(g,m)=\int_{x}^{x'}b(u,y)\,\mathrm{d}u=(x'-x)\int_{0}^{1}b\bigl(x+t(x'-x),y\bigr)\,\mathrm{d}t. \tag*{(60)} I 1 ( b ξ ) ( g , m ) = ∫ x x ′ b ( u , y ) d u = ( x ′ − x ) ∫ 0 1 b ( x + t ( x ′ − x ) , y ) d t . (60) Proof. The function b b b is smooth on O O O . If b = ∂ x f b=\partial_{x}f b = ∂ x f with f ∈ C ∞ ( O ) f\in C^{\infty}(O) f ∈ C ∞ ( O ) , then for 0 < ∣ y ∣ < 1 / 2 0<|y|<1/2 0 < ∣ y ∣ < 1/2 ,
f ( 3 / 2 , y ) − f ( 1 / 2 , y ) = ∫ 1 / 2 3 / 2 d x ( x − 1 ) 2 + y 2 ≥ ∫ 1 − ∣ y ∣ 1 + ∣ y ∣ d x ( x − 1 ) 2 + y 2 = π 2 ∣ y ∣ . f(3/2,y)-f(1/2,y)=\int_{1/2}^{3/2}\frac{\mathrm{d}x}{(x-1)^{2}+y^{2}}
\ge\int_{1-|y|}^{1+|y|}\frac{\mathrm{d}x}{(x-1)^{2}+y^{2}}=\frac{\pi}{2|y|}. f ( 3/2 , y ) − f ( 1/2 , y ) = ∫ 1/2 3/2 ( x − 1 ) 2 + y 2 d x ≥ ∫ 1 − ∣ y ∣ 1 + ∣ y ∣ ( x − 1 ) 2 + y 2 d x = 2∣ y ∣ π .
The two endpoint functions are smooth near y = 0 y = 0 y = 0 , so their difference is locally bounded, a contradiction. Proposition 11.4 therefore gives the nonzero class.
On a full-MC edge σ \sigma σ , the base equation is σ ∗ ζ = d Δ x σ \sigma^*\zeta= \mathrm{d}_{\Delta}x_{\sigma} σ ∗ ζ = d Δ x σ . The unary scalar kernel is integration, and the selected base path lies in one horizontal interval component of O O O . A one-form b ( x , y ) d x b(x,y)\,\mathrm{d}x b ( x , y ) d x on that interval has an ordinary primitive, so its integral depends only on the two endpoints. This gives (11.2). The hull condition ensures that the displayed straight segment lies in O O O , making the last formula jointly smooth on the admissible arrow manifold. □
Thus scalar cohomology can change on a nonrectangular open: degree-one cohomology vanishes on the full plane, whereas this open has the displayed nonzero class. The all-open theorem compares the actual cohomology on each open.
Two-term normal forms and ordinary non-descent
Throughout this section M = R 2 M = \mathbb{R}^2 M = R 2 , B = B E a f f B = B_{E_{\mathrm{aff}}} B = B E aff and K K K is the actual crossed-module nerve. All equivalences of coefficient objects mean isomorphisms in the H 0 H^0 H 0 category of the indicated dg category.
Normal forms for cohesive coefficients
12.1. Normal forms for cohesive coefficients. Let Coh [ 0 , 1 ] r e g ( B ) \operatorname{Coh}^{\mathrm{reg}}_{[0,1]}(B) Coh [ 0 , 1 ] reg ( B ) be the full subcategory whose internal cohomology is supported in degrees 0 , 1 0, 1 0 , 1 . The original coefficient complex may have acyclic terms outside these degrees. For raw representations, the same amplitude condition concerns vertex cohomology, not the support of the original graded bundle. For finite-dimensional V 0 V^0 V 0 , V 1 V^1 V 1 and a linear map N : V 1 → V 0 N : V^1 \to V^0 N : V 1 → V 0 , let P N P_N P N be the cohesive module on the constant graded bundle V = V 0 ⊕ V 1 V = V^0 \oplus V^1 V = V 0 ⊕ V 1 , defined on constant vectors by
D N ( v 0 ) = 0 , D N ( v 1 ) = N ( v 1 ) η . (61) D_N(v_0) = 0,\qquad D_N(v_1) = N(v_1)\eta. \tag*{(61)} D N ( v 0 ) = 0 , D N ( v 1 ) = N ( v 1 ) η . (61)
It is extended by the right-module Leibniz rule. Extend N N N by zero on V 0 V^0 V 0 ; then N 2 = 0 N^2 = 0 N 2 = 0 , and Q η = 0 Q\eta= 0 Q η = 0 proves flatness. Write 1 j \mathbf{1}_j 1 j for the trivial coefficient with one generator of internal degree j j j and zero differential on that generator. The elementary eta-block E η E_\eta E η has one generator in each of degrees 0 , 1 0,1 0 , 1 and D ( v 1 ) = v 0 η D(v_1) = v_0\eta D ( v 1 ) = v 0 η .
Taking CE degree zero and then evaluating at m ∈ M m \in M m ∈ M gives a dg functor from cohesive modules to finite complexes. The analogous vertex functor on raw representations is dg by (4.3). Thus an H 0 H^0 H 0 -equivalence induces a chain homotopy equivalence of vertex complexes. This uses actual homotopies, not exactness of point evaluation.
Theorem 12.1 (Two-term normal forms). Every object of Coh [ 0 , 1 ] r e g ( B ) \operatorname{Coh}^{\mathrm{reg}}_{[0,1]}(B) Coh [ 0 , 1 ] reg ( B ) is H 0 H^0 H 0 -isomorphic to some P N 0 P_{N_0} P N 0 with constant real matrix N 0 N_0 N 0 . Two such objects are isomorphic exactly when their triples
( d 0 , d 1 , r ) = ( dim V 0 , dim V 1 , rank N 0 ) (d_0,d_1,r) = \left(\dim V^0,\dim V^1,\operatorname{rank}N_0\right) ( d 0 , d 1 , r ) = ( dim V 0 , dim V 1 , rank N 0 ) agree. The possible triples are the nonnegative integers with r ≤ min ( d 0 , d 1 ) r \le\min(d_0,d_1) r ≤ min ( d 0 , d 1 ) , and the corresponding normal form is
E η ⊕ r ⊕ 1 0 ⊕ ( d 0 − r ) ⊕ 1 1 ⊕ ( d 1 − r ) . (62) E_\eta^{\oplus r} \oplus\mathbf{1}_0^{\oplus(d_0-r)} \oplus\mathbf{1}_1^{\oplus(d_1-r)}. \tag*{(62)} E η ⊕ r ⊕ 1 0 ⊕ ( d 0 − r ) ⊕ 1 1 ⊕ ( d 1 − r ) . (62) :::`
Proof. Apply Lemmas B.2 and B.3. The resulting cohomology bundle is concentrated in degrees 0, 1 and is trivial on R 2 \mathbb{R}^2 R 2 . Its internal differential and CE-degree-one connection vanish. A remaining connection component of CE degree p p p has internal degree 1 − p 1-p 1 − p . Hence the only possible component has p = 2 p=2 p = 2 and maps V 1 V^1 V 1 to V 0 V^0 V 0 .
In graded frames, write its coefficient matrix as N ( x , y ) η + T ( x , y ) θ 1 θ 2 N(x,y)\eta+T(x,y)\theta_1\theta_2 N ( x , y ) η + T ( x , y ) θ 1 θ 2 . Products of these negative internal endomorphisms vanish. The scalar differential calculation (B.5) therefore gives
( N x − T ) ( x θ 1 + y θ 2 ) η = 0. (N_x-T)(x\theta_1+y\theta_2)\eta=0. ( N x − T ) ( x θ 1 + y θ 2 ) η = 0.
Entrywise, multiplication by both x x x and y y y kills N x − T N_x-T N x − T . It is zero away from the origin and then at the origin by continuity. Thus the remaining coefficient is
b N = N η + N x θ 1 θ 2 . b_N=N\eta+N_x\theta_1\theta_2. b N = N η + N x θ 1 θ 2 .
Set N 0 = N ( 0 , 0 ) N_0=N(0,0) N 0 = N ( 0 , 0 ) and define smooth matrices
P = − ∫ 0 1 N y ( t x , t y ) d t , R = ∫ 0 1 N x ( t x , t y ) d t . P=-\int_0^1 N_y(tx,ty)\,\mathrm{d}t,\qquad R=\int_0^1 N_x(tx,ty)\,\mathrm{d}t. P = − ∫ 0 1 N y ( t x , t y ) d t , R = ∫ 0 1 N x ( t x , t y ) d t .
The fundamental theorem of calculus gives N − N 0 = − y P + x R N-N_0=-yP+xR N − N 0 = − y P + x R . These compact-interval integrals are globally smooth; no division by x x x or y y y is involved. For c = P θ 1 + R θ 2 c=P\theta_1+R\theta_2 c = P θ 1 + R θ 2 , equation (B.4) gives
Q c = ( N − N 0 ) η + N x θ 1 θ 2 = b N − N 0 η . (63) Qc=(N-N_0)\eta+N_x\theta_1\theta_2=b_N-N_0\eta. \tag*{(63)} Q c = ( N − N 0 ) η + N x θ 1 θ 2 = b N − N 0 η . (63)
Denote the reduced differential by D b N D_{b_N} D b N and regard v j v^j v j as the row of degree-j j j basis vectors. The right-linear degree-zero map from the b N b_N b N module to P N 0 P_{N_0} P N 0 ,
g ( v 0 ) = v 0 , g ( v 1 ) = v 1 + v 0 c , g(v^0)=v^0,\qquad g(v^1)=v^1+v^0c, g ( v 0 ) = v 0 , g ( v 1 ) = v 1 + v 0 c ,
has inverse with the negative correction, since its off-diagonal part squares to zero. Its orientation is fixed by
D N 0 g ( v 1 ) = v 0 ( N 0 η + Q c ) = v 0 b N = g D b N ( v 1 ) . D_{N_0}g(v^1)=v^0(N_0\eta+Qc)=v^0b_N=gD_{b_N}(v^1). D N 0 g ( v 1 ) = v 0 ( N 0 η + Q c ) = v 0 b N = g D b N ( v 1 ) .
The equality on v 0 v^0 v 0 is immediate, and the Leibniz rule extends these identities to the whole module. This proves existence.
For uniqueness, a general closed degree-zero map φ : P N ⟶ P N ′ \varphi:P_N\longrightarrow P_{N'} φ : P N ⟶ P N ′ between constant-residue models has the form
φ ( v 0 ) = w 0 S 0 , φ ( v 1 ) = w 1 S 1 + w 0 C , C = P θ 1 + R θ 2 , \varphi(v^0)=w^0S_0,\qquad\varphi(v^1)=w^1S_1+w^0C,\qquad C=P\theta_1+R\theta_2, φ ( v 0 ) = w 0 S 0 , φ ( v 1 ) = w 1 S 1 + w 0 C , C = P θ 1 + R θ 2 ,
where S 0 , S 1 S_0,S_1 S 0 , S 1 are smooth function matrices. Nonnegative CE degree excludes any other component. Closedness gives
Q S 0 = Q S 1 = 0 , N ′ η S 1 + Q C = S 0 N η . (64) QS_0=QS_1=0,\qquad N'\eta S_1+QC=S_0N\eta. \tag*{(64)} Q S 0 = Q S 1 = 0 , N ′ η S 1 + QC = S 0 N η . (64)
The eta coefficient of Q C QC QC is − y P + x R -yP+xR − y P + x R , which vanishes at the origin. Consequently
N ′ S 1 ( 0 ) = S 0 ( 0 ) N . N'S_1(0)=S_0(0)N. N ′ S 1 ( 0 ) = S 0 ( 0 ) N .
If φ \varphi φ is invertible in H 0 H^0 H 0 , its vertex map is a homotopy equivalence between zero-differential complexes. Thus S 0 ( 0 ) S_0(0) S 0 ( 0 ) and S 1 ( 0 ) S_1(0) S 1 ( 0 ) are actual isomorphisms. The dimensions and ranks agree. Conversely, equal dimensions and ranks give constant invertible S 0 , S 1 S_0,S_1 S 0 , S 1 intertwining N , N ′ N,N' N , N ′ . Setting C = 0 C=0 C = 0 in (64) gives a strict cohesive isomorphism. Putting N 0 N_0 N 0 in rank normal form proves (62) .
Zero matrices cover r = 0 r=0 r = 0 , and empty graded pieces cause no change in the argument. If both cohomology bundles vanish, the initial finite minimalization contracts the object to zero. The same argument removes all acyclic tails in the original bounded complex. □ \square □
The matrix N 0 N_0 N 0 depends on frames and reductions. Its GL ( V 0 ) × GL ( V 1 ) \operatorname{GL}(V^0)\times\operatorname{GL}(V^1) GL ( V 0 ) × GL ( V 1 ) orbit, equivalently its rank together with the two dimensions, is the object invariant. The classification concerns objects up to H 0 H^0 H 0 -isomorphism; their morphism complexes retain the dg structure.
The actual integrated normal forms
Corollary 12.2. The actual affine RH functors restrict to A ∞ A_\infty A ∞ quasi-equivalences on the regular full subcategories of internal cohomology amplitude [ 0 , 1 ] [0,1] [ 0 , 1 ] . On y K \mathsf{yK} yK , their normal forms are
R N 0 : r 0 = 0 , r 1 = 1 , r 2 = − h 012 N 0 , r p = 0 ( p ≥ 3 ) . (65) R_{N_0}:\quad r_0=0,\quad r_1=1,\quad r_2=-h_{012}N_0,\quad r_p=0\quad(p\geq3). \tag*{(65)} R N 0 : r 0 = 0 , r 1 = 1 , r 2 = − h 012 N 0 , r p = 0 ( p ≥ 3 ) . (65) The same triples ( d 0 , d 1 , r ) (d_0,d_1,r) ( d 0 , d 1 , r ) classify H 0 H^0 H 0 -isomorphism classes on y K \mathsf{yK} yK and on full MC integration.
Proof. The actual functors preserve vertex complexes. If a regular target has internal cohomology in degrees 0 , 1 0,1 0 , 1 , the preimage and counit of Theorem C.4 have homotopy equivalent vertex complexes. That preimage therefore already belongs to Coh [ 0 , 1 ] r e g ( B ) \operatorname{Coh}^{\mathrm{reg}}_{[0,1]}(B) Coh [ 0 , 1 ] reg ( B ) . Together with the all-degree Hom comparison, this proves the quasi-equivalence on these full subcategories.
For P N 0 P_{N_0} P N 0 , the argument of Proposition 6.4 applies to the whole matrix. Every connection word with two occurrences of N 0 N_0 N 0 vanishes because N 0 2 = 0 N_0^2=0 N 0 2 = 0 . The actual unary kernel is
Γ 1 = ∫ , \Gamma_1=\int, Γ 1 = ∫ ,
so on full MC the two-component is ( ∫ [ 012 ] β ) N 0 (\int_{[012]}\beta)N_0 ( ∫ [ 012 ] β ) N 0 . Pullback by Ψ \Psi Ψ and h = − ∫ β h=-\int\beta h = − ∫ β give (65) . Theorem 12.1 and the restricted RH equivalences give the classification. □ \square □
An actual ordinary quotient
Proposition 12.3. There is an internal ordinary groupoid G ‾ \overline{G} G in smooth sheaves with object sheaf y M \mathsf{yM} yM , and a strict simplicial sheaf map
q : y K ⟶ N ( G ‾ ) , q 0 = Id y M , q:\mathsf{yK}\longrightarrow N(\overline{G}),\qquad q_0=\operatorname{Id}_{\mathsf{yM}}, q : yK ⟶ N ( G ) , q 0 = Id yM , obtained by identifying parallel 2-isomorphic 1-arrows. Its arrow object is the sheafification of the presheaf of these classes.
Proof. For each test U U U , evaluate the strict affine 2-groupoid and take 2-isomorphism classes of its parallel 1-arrows. Horizontal composition and inverses preserve this relation, so the classes form an ordinary groupoid. In source coordinates, ( g , m ) (g,m) ( g , m ) is related to ( g exp ( h Z m ) , m ) (g\exp(hZm),m) ( g exp ( h Z m ) , m ) ; equivalently the multiplier on the left is exp ( h Z g m ) \exp(hZ_{gm}) exp ( h Z g m ) . Both arrows have target g m gm g m . Restriction pulls back the actual 2-isomorphisms, giving a presheaf of groupoids with object presheaf y M yM y M . Let P 1 P_1 P 1 be its arrow presheaf and a a a ordinary sheafification. Left exactness gives
a ( P 1 × y M P 1 ) ≅ a P 1 × y M a P 1 . a(P_1 \mathbin{\times_{yM}} P_1) \cong aP_1 \mathbin{\times_{yM}} aP_1. a ( P 1 × y M P 1 ) ≅ a P 1 × y M a P 1 .
Thus all groupoid structure maps sheafify, with the object sheaf unchanged, to G ‾ \overline{G} G .
Send a Duskin simplex to the consecutive classes [ g 01 ] , … , [ g n − 1 , n ] [g_{01}],\ldots,[g_{n-1,n}] [ g 01 ] , … , [ g n − 1 , n ] . Its triangle 2-isomorphisms imply [ g i k ] = [ g j k ] [ g i j ] [g_{ik}]=[g_{jk}][g_{ij}] [ g ik ] = [ g j k ] [ g ij ] , so inner faces become composition, outer faces omit end arrows, and degeneracies insert identities. This is a simplicial presheaf map to the ordinary nerve. Sheafification commutes with its finite nerve products and therefore yields the stated q q q . □
This is an ordinary groupoid object in the same smooth sheaf topos used for raw coefficients. In particular its object sheaf is represented, as required by Definition 4.2. It records ordinary 2-isomorphism classes of 1-arrows. Comparison with AS holonomy, refined higher holonomy, a geometric Morita quotient or an infinity-topos truncation requires additional geometric input.
The higher-isotropy obstruction
12.4. The higher-isotropy obstruction. There is a strict inclusion
ι : T = B 2 Z ⟶ y K \iota:T=B^2\mathbb{Z}\longrightarrow yK ι : T = B 2 Z ⟶ y K
at the origin: all root edges are identities and all rooted triangle coordinates are integers. Equation (3.4), with Z 0 = 0 Z_0=0 Z 0 = 0 , shows that every edge stays an identity and every other triangle coordinate stays integral under the simplex operators. Here T n T_n T n is the locally constant sheaf Z ( 2 ) \mathbb{Z}^{(2)} Z ( 2 ) , not a constant presheaf on disconnected tests. In particular T 0 = T 1 = ∗ T_0=T_1=\ast T 0 = T 1 = ∗ and T 2 T_2 T 2 contains the simplex h 012 = 1 h_{012}=1 h 012 = 1 . The simplicial object has all degrees.
Theorem 12.4 (Ordinary-groupoid non-descent). Let G G G be any internal ordinary groupoid in smooth sheaves with represented object sheaf G 0 = y M ′ G_0=yM' G 0 = y M ′ . For any strict simplicial sheaf map f : y K → N ( G ) f:yK\to N(G) f : y K → N ( G ) and any W ∈ Rep s m p e r f ( N ( G ) ) W\in\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(N(G)) W ∈ Rep sm perf ( N ( G )) , the normal form R N R_N R N with N ≠ 0 N\ne0 N = 0 is not H 0 H^0 H 0 -isomorphic to f ∗ W f^*W f ∗ W . The same assertion holds for INE a f f ( P N ) \operatorname{INE}_{\mathrm{aff}}(P_N) INE aff ( P N ) and any strict map Int ( E a f f ) → N ( G ) \operatorname{Int}(E_{\mathrm{aff}})\to N(G) Int ( E aff ) → N ( G ) . The coefficient W W W may be nonregular or nonminimal.
Proof. An ordinary nerve is determined by its consecutive 1-arrows. Since the unique 1-simplex of T T T is degenerate, f ι f\iota f ι sends every simplex to the totally degenerate simplex at p = f 0 ( 0 ) p=f_0(0) p = f 0 ( 0 ) . Consequently ( f ι ) ∗ W (f\iota)^*W ( f ι ) ∗ W is the constant representation of the finite vertex complex C = W p C=W_p C = W p : its internal differential is d C d_C d C , its edge transport is the identity, and its higher components vanish by normalization on their degenerate images.
Suppose an H 0 H^0 H 0 -equivalence existed. After restriction, represent it by a closed degree-zero morphism F : ι ∗ R N → C F:\iota^*R_N\to C F : ι ∗ R N → C . The vertex map F 0 : ( V , 0 ) → ( C , d C ) F_0:(V,0)\to(C,d_C) F 0 : ( V , 0 ) → ( C , d C ) is a chain homotopy equivalence. Normalization gives F 1 = 0 F_1=0 F 1 = 0 because every 1-simplex of T T T is degenerate. At simplicial degree two, the actual differential (4.3) becomes
d C F 2 + h F 0 N = 0. (66) d_C F_2+hF_0N=0. \tag*{(66)} d C F 2 + h F 0 N = 0. (66)
Indeed the sign of r C , 0 ⋆ F 2 r_{C,0}\star F_2 r C , 0 ⋆ F 2 is ( − 1 ) 1 ⋅ 2 = 1 (-1)^{1\cdot2}=1 ( − 1 ) 1 ⋅ 2 = 1 , and − F 0 ⋆ r N , 2 = h F 0 N -F_0\star r_{N,2}=hF_0N − F 0 ⋆ r N , 2 = h F 0 N . All other terms contain F 1 F_1 F 1 , the zero source internal differential or a zero target higher component.
Evaluate at the specified simplex h = 1 h=1 h = 1 and apply to v ∈ V v\in V v ∈ V . The term F 2 ( v ) F_2(v) F 2 ( v ) can lie in C − 1 C^{-1} C − 1 , and its differential contributes a boundary. Equation (12.6) therefore says that F 0 N v F_0Nv F 0 N v is a boundary. Thus H 0 ( F 0 ) ( N v ) = 0 H^0(F_0)(Nv)=0 H 0 ( F 0 ) ( N v ) = 0 in H 0 ( C ) H^0(C) H 0 ( C ) . Injectivity of H 0 ( F 0 ) : V 0 → H 0 ( C ) H^0(F_0):V^0\to H^0(C) H 0 ( F 0 ) : V 0 → H 0 ( C ) forces N v = 0 Nv=0 N v = 0 for every v v v , a contradiction. The argument evaluates the cochain identity on the specified simplex of B 2 Z B^2\mathbb{Z} B 2 Z .
For a strict map f : Int ( E a f f ) → N ( G ) f:\operatorname{Int}(E_{\mathrm{aff}})\to N(\mathcal{G}) f : Int ( E aff ) → N ( G ) , pull a hypothetical descent back along the actual global section Ψ \Psi Ψ . The equality Ψ ∗ IN E a f f ( P N ) = IN K ( P N ) = R N \Psi^*\operatorname{IN}_{E_{\mathrm{aff}}}(P_N)=\operatorname{IN}^K(P_N)=R_N Ψ ∗ IN E aff ( P N ) = IN K ( P N ) = R N would give descent along f Ψ f\Psi f Ψ , contradicting the first part. □
Corollary 12.5 (Exact two-term descent criterion). A regular two-term raw representation of y K y_K y K lies in the essential image of H 0 ( q ∗ ) H^0(q^*) H 0 ( q ∗ ) exactly when its residue rank is zero. For full MC the same criterion holds for q M C = q Φ q_{\mathrm{MC}}=q\Phi q MC = q Φ .
Proof. Corollary 12.2 puts every such object in a unique normal-form class. Nonzero rank is excluded by Theorem 12.4. At rank zero the normal form is a sum of the trivial degree-zero and degree-one coefficients. It is pulled back from the constant graded representation of G ‾ \overline{G} G with those dimensions. The same constant representation works for q Φ q\Phi q Φ on full MC; also q Φ Ψ = q q\Phi\Psi=q q ΦΨ = q . □
Here descent means membership in the specified essential image. The classification is global and regular, and the obstruction concerns strict maps to ordinary nerves in the raw coefficient theory. Ordinary-open RH follows from the direct comparison in Theorem 10.4; regular minimalization enters here to classify the resulting objects. Effective descent and comparison with refined higher holonomy are discussed in Section 14.
Vertical families and coefficient bundles
We extend the scalar parameter comparison of Corollary B.12 to regular bundle-valued coefficients. A frame domain on the object manifold need not contain every vertex of a simplex. Normalized projective cochain modules give a global description, and reconstruction then takes place over the central algebra of parameter functions.
Throughout this section, S S S is a finite-dimensional Hausdorff second-countable smooth manifold. It need not be compact or connected. Put
M S = S × R 2 , π : M S ⟶ S , A S = C ∞ ( M S ) , C S = C ∞ ( S ) . M_S=S\times\mathbb{R}^2,\qquad\pi:M_S\longrightarrow S,\qquad A_S=C^\infty(M_S),\qquad C_S=C^\infty(S). M S = S × R 2 , π : M S ⟶ S , A S = C ∞ ( M S ) , C S = C ∞ ( S ) .
All coefficient complexes are globally bounded and have finite-projective global section modules. Regularity continues to mean locally constant internal cohomology ranks. The empty parameter manifold gives zero coefficient modules; the assertions below include that case.
The vertical resolution and product integration
Let E S E_S E S be the graded-bundle extension of E a f f E_{\mathrm{aff}} E aff with zero anchor in the S S S directions. Thus
∂ ε = − y e 1 + x e 2 , ρ ( e 1 ) = x ∂ x , ρ ( e 2 ) = y ∂ x . \partial\varepsilon=-y e_1+x e_2,\qquad\rho(e_1)=x\partial_x,\qquad\rho(e_2)=y\partial_x. ∂ ε = − y e 1 + x e 2 , ρ ( e 1 ) = x ∂ x , ρ ( e 2 ) = y ∂ x .
Its CE algebra is
B S = A S [ θ 1 , θ 2 , η ] , ∣ θ i ∣ = 1 , ∣ η ∣ = 2 , Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Q θ 2 = θ 1 θ 2 + x η , Q η = 0. (67) B_S = A_S[\theta_1,\theta_2,\eta], \qquad|\theta_i| = 1,\qquad|\eta| = 2,
\qquad Qf = f_x(x\theta_1+y\theta_2),\qquad Q\theta_1=-y\eta,\qquad Q\theta_2=\theta_1\theta_2+x\eta,\qquad Q\eta=0. \tag*{(67)} B S = A S [ θ 1 , θ 2 , η ] , ∣ θ i ∣ = 1 , ∣ η ∣ = 2 , Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Q θ 2 = θ 1 θ 2 + x η , Q η = 0. (67)
This is not the inverse-image algebroid that adds T S TS T S . Write K S , n = S × K n K_{S,n}=S\times K_n K S , n = S × K n , so y K S = c ( y S ) × y K \mathsf{y}K_S=c(\mathsf{y}S)\times\mathsf{y}K y K S = c ( y S ) × y K , where c c c denotes the constant simplicial object, not the smooth singular complex.
Lemma 13.1. The vertical extension is a global geometric resolution of ⟨ x ∂ x , y ∂ x ⟩ \langle x\partial_x,y\partial_x\rangle ⟨ x ∂ x , y ∂ x ⟩ on M S M_S M S . There is a natural identification
Int ( E S ) = c ( y S ) × Int ( E a f f ) . \operatorname{Int}(E_S)=c(\mathsf{y}S)\times\operatorname{Int}(E_{\mathrm{aff}}). Int ( E S ) = c ( y S ) × Int ( E aff ) . Taking the product of Φ \Phi Φ , Ψ \Psi Ψ , H H H with c ( y S ) c(\mathsf{y}S) c ( y S ) gives maps Φ S \Phi_S Φ S , Ψ S \Psi_S Ψ S , H S H_S H S , respectively. The homotopy H S H_S H S remains relative to the whole c ( y M S ) c(\mathsf{y}M_S) c ( y M S ) . The resulting actual raw pullbacks are dg quasi-equivalences, also on regular full subcategories.
Proof. Suppose x f + y g = 0 xf+yg=0 x f + y g = 0 in A S A_S A S . At x = 0 x=0 x = 0 , the function g g g vanishes for y ≠ 0 y\ne0 y = 0 , and then for y = 0 y=0 y = 0 by continuity. The globally smooth parameterized Hadamard formula
h ( s , x , y ) = ∫ 0 1 g x ( s , t x , y ) d t h(s,x,y)=\int_0^1 g_x(s,tx,y)\,\mathrm{d}t h ( s , x , y ) = ∫ 0 1 g x ( s , t x , y ) d t
gives g = x h g=xh g = x h . The original relation now gives f = − y h f=-yh f = − y h on x ≠ 0 x\ne0 x = 0 , and hence everywhere. The map h ↦ ( − y h , x h ) h\mapsto(-yh,xh) h ↦ ( − y h , x h ) is injective by the same continuity argument. This proves exactness over A S A_S A S .
For a full MC simplex, the zero S S S -anchor makes its S S S -coordinate constant in the connected simplex variables. On a test manifold U U U that coordinate is precisely a smooth map U → S U\to S U → S ; all remaining equations are those of E a f f E_{\mathrm{aff}} E aff . This proves the displayed identification in every degree, naturally in tests and simplex operators. Products of the actual affine comparison maps and homotopy give the product retraction. The product homotopy remains relative to the whole c ( y M S ) c(\mathsf{y}M_S) c ( y M S ) , not merely to the endpoint vertices. Theorem 5.3 therefore applies. The vertex maps are identities, so regularity is preserved. □
A global normalized projective raw dictionary
Put C = C n o r m , s m ( y K S ) C=C_{\mathrm{norm},\mathrm{sm}}(\mathsf{y}K_S) C = C norm , sm ( y K S ) , with its full cochain differential d f u l l d_{\mathrm{full}} d full and Alexander–Whitney product. Distinguish this differential from the inner-face operator δ i n \delta_{\mathrm{in}} δ in of (12) . For a finite bounded graded real vector space W W W , the constant raw representation gives
F W = Hom r a w ( 1 , W ) = C ⊗ R W , ( c ⊗ w ) a = ( − 1 ) ∣ w ∣ ∣ a ∣ ( c ⌣ a ) ⊗ w . F_W=\operatorname{Hom}_{\mathrm{raw}}(1,W)=C\otimes_{\mathbb{R}}W,\qquad(c\otimes w)a=(-1)^{|w||a|}(c\smile a)\otimes w. F W = Hom raw ( 1 , W ) = C ⊗ R W , ( c ⊗ w ) a = ( − 1 ) ∣ w ∣∣ a ∣ ( c ⌣ a ) ⊗ w .
Homogeneous module maps use the signless right-linearity convention. A finite graded idempotent p ∈ A S ⊗ End R 0 ( W ) p\in A_S\otimes\operatorname{End}_{\mathbb{R}}^0(W) p ∈ A S ⊗ End R 0 ( W ) presents a vertex bundle V = im p V=\operatorname{im}p V = im p . Define F V = p F W F_V=pF_W F V = p F W . Its components are
F V k , j = Γ n o r m ( K S , k , v 0 ∗ V j ) . (68) F_V^{k,j}=\Gamma_{\mathrm{norm}}(K_{S,k},v_0^*V^j). \tag*{(68)} F V k , j = Γ norm ( K S , k , v 0 ∗ V j ) . (68)
For k > 0 k>0 k > 0 , the subscript means vanishing on all degeneracies. These are the graded modules induced from V V V over A S A_S A S through the displayed finite projectors.
Lemma 13.2 (Global projective raw dictionary). Raw representations on globally bounded finite-projective vertex bundles are dg equivalent to flat right superconnections on the modules F V F_V F V . The equivalence identifies their actual normalized Hom complexes and preserves the regular full subcategories.
Proof. Multiplication by p ( v 0 ) p(v_0) p ( v 0 ) commutes with degeneracies, giving (13.2). For presentations V = im p ⊂ W V=\operatorname{im}p\subset W V = im p ⊂ W and V ′ = im p ′ ⊂ W ′ V'=\operatorname{im}p'\subset W' V ′ = im p ′ ⊂ W ′ , finite free-module linearity gives
Hom C ( F V , F V ′ ) = p ′ Hom C ( F W , F W ′ ) p . (69) \operatorname{Hom}_{C}(F_V,F_{V'})=p'\operatorname{Hom}_{C}(F_W,F_{W'})p. \tag*{(69)} Hom C ( F V , F V ′ ) = p ′ Hom C ( F W , F W ′ ) p . (69)
In cochain degree k k k , the corner condition is exactly
α ( σ ) = p ′ ( v 0 σ ) α ( σ ) p ( v k σ ) . \alpha(\sigma)=p'(v_0\sigma)\alpha(\sigma)p(v_k\sigma). α ( σ ) = p ′ ( v 0 σ ) α ( σ ) p ( v k σ ) .
Indeed, postcomposition multiplies at the first vertex, whereas precomposition uses the last vertex in the cochain product. This supplies the bundle type Hom ( v k ∗ V , v 0 ∗ V ′ ) \operatorname{Hom}(v_k^*V,v_0^*V') Hom ( v k ∗ V , v 0 ∗ V ′ ) without requiring a common object-frame domain for the endpoints. The free-module calculation and the braided action give precisely the sign in (11) . In particular, a degree-j j j zero-cochain bundle map T T T extends as
( T ^ ( c ⊗ w ) ) ( σ ) = ( − 1 ) j k c ( σ ) T ( v 0 σ ) w , c ∈ C k . (70) \bigl(\widehat{T}(c\otimes w)\bigr)(\sigma)=(-1)^{jk}c(\sigma)T(v_0\sigma)w,\qquad c\in C^k. \tag*{(70)} ( T ( c ⊗ w ) ) ( σ ) = ( − 1 ) j k c ( σ ) T ( v 0 σ ) w , c ∈ C k . (70)
The reference connection on F V F_V F V is ∇ p = p d f u l l \nabla_p=p\,d_{\mathrm{full}} ∇ p = p d full . The Leibniz rule follows from p α = α p\alpha=\alpha p α = α , but this connection need not be flat. For an arbitrary right superconnection D D D on F V F_V F V , its difference A = D − ∇ p A=D-\nabla_p A = D − ∇ p is a normalized degree-one right-linear corner. Let e W e_W e W be the ambient identity one-cochain, which is not normalized, and set
r = p ⋆ e W ⋆ p + A . (71) r=p\star e_W\star p+A. \tag*{(71)} r = p ⋆ e W ⋆ p + A . (71)
Here δ i n p = 0 \delta_{\mathrm{in}}p=0 δ in p = 0 , whereas d f u l l p d_{\mathrm{full}}p d full p need not vanish. Applying (6.7) to α ∈ F V \alpha\in F_V α ∈ F V therefore gives the exact formula
D α = δ i n α + r ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e 1 . (72) D\alpha=\delta_{\mathrm{in}}\alpha+r\star\alpha-(-1)^{|\alpha|}\alpha\star e_1. \tag*{(72)} D α = δ in α + r ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e 1 . (72)
Projector variation contributes to curvature. For example the reference structure p ⋆ e W ⋆ p p\star e_W\star p p ⋆ e W ⋆ p has two-curvature
κ 2 p ( 012 ) = p 0 ( p 1 − 1 ) p 2 , p i = p ( v i ) , \kappa^p_2(012)=p_0(p_1-1)p_2,\qquad p_i=p(v_i), κ 2 p ( 012 ) = p 0 ( p 1 − 1 ) p 2 , p i = p ( v i ) ,
which can be nonzero.
The derivation identity of Lemma 4.1 and flatness of the source unit yield
D 2 α = ( δ i n r + r ⋆ r ) ⋆ α . D^2\alpha=(\delta_{\mathrm{in}}r+r\star r)\star\alpha. D 2 α = ( δ in r + r ⋆ r ) ⋆ α .
The curvature κ = δ i n r + r ⋆ r \kappa=\delta_{\mathrm{in}}r+r\star r κ = δ in r + r ⋆ r is itself an endpoint corner: κ = p ⋆ κ ⋆ p \kappa=p\star\kappa\star p κ = p ⋆ κ ⋆ p . If D 2 = 0 D^2=0 D 2 = 0 , test this identity on the finitely many zero-cochain generators p e j pe_j p e j coming from a homogeneous basis of W W W . On a k k k -simplex it says κ k ( σ ) p ( v k σ ) e j = 0 \kappa_k(\sigma)p(v_k\sigma)e_j=0 κ k ( σ ) p ( v k σ ) e j = 0 for every j j j . These vectors span the last-vertex fiber, so κ k = 0 \kappa_k=0 κ k = 0 . Thus flatness implies the raw MC equation globally. The reverse implication follows from the same identity.
Since A A A is normalized, (13.5) has r 1 ( s 0 m ) = p ( m ) ∣ V m = 1 V m r_1(s_0m)=p(m)|_{V_m}=1_{V_m} r 1 ( s 0 m ) = p ( m ) ∣ V m = 1 V m and normalized higher components. The degree-two MC identities on the h = 0 h=0 h = 0 inverse-arrow triangles in K S K_S K S give chain-homotopy inverses for its edge maps. Hence it is a raw representation, including the edge quasi-isomorphism requirement. Conversely, (13.6) is a flat right superconnection for a raw representation. The constructions are inverse in projector presentations. Finally, substitute (13.6) in the superconnection commutator on (13.3). The source-unit terms cancel, leaving
D H o m ϕ = δ i n ϕ + r V ′ ⋆ ϕ − ( − 1 ) ∣ ϕ ∣ ϕ ⋆ r V . D_{\mathrm{Hom}}\phi=\delta_{\mathrm{in}}\phi+r_{V'}\star\phi-(-1)^{|\phi|}\phi\star r_V. D Hom ϕ = δ in ϕ + r V ′ ⋆ ϕ − ( − 1 ) ∣ ϕ ∣ ϕ ⋆ r V .
This is the actual raw Hom differential, with its normalization, and composition is the same star product. Changing a projector presentation gives the corresponding module isomorphism and does not change this dg equivalence. □
The lemma gives a concrete dg description on induced projective modules equipped with flat superconnections, before any passage to a derived-perfect localization.
Regular reduction to bundles over the parameter base
Lemma 13.3 (Basic minimal coefficients). Every regular cohesive B S B_S B S -module is homotopy equivalent to one on π ∗ H \pi^*H π ∗ H with zero internal differential and zero CE-degree-one connection. Every regular raw representation of y K S \mathfrak{y}_{KS} y K S is homotopy equivalent to one on π ∗ H \pi^*H π ∗ H with r 0 = 0 r_0=0 r 0 = 0 and r 1 = 1 H r_1=1_H r 1 = 1 H . In both statements H H H is globally bounded and Γ ( S , H ) \Gamma(S,H) Γ ( S , H ) is finite projective over C S C_S C S .
Proof. Regularity and boundedness give a smooth contraction ( i , p H , h ) (i,p_H,h) ( i , p H , h ) of the vertex complex onto its cohomology bundle H M H_M H M , as in Lemma B.2. For the raw category, first use Lemma 13.2. Extend the contraction by (13.4): on k k k -cochains both d 0 d_0 d 0 and h h h carry ( − 1 ) k (-1)^k ( − 1 ) k . These factors cancel in d 0 h + h d 0 = 1 − i p H d_0h+hd_0=1-ip_H d 0 h + h d 0 = 1 − i p H , and all side conditions are preserved. The finite transfer formulas (B.1) now apply to the actual normalized projective modules. For Δ = D − d 0 \Delta=D-d_0 Δ = D − d 0 , every component has positive cochain degree, so h Δ h\Delta h Δ and Δ h \Delta h Δ h strictly lower internal degree. The global amplitude bound makes their inverse series uniformly finite. The right-linearity proof uses the contraction side conditions and the Leibniz rule, not centrality of A S A_S A S . The perturbation starts with the flat superconnection D D D ; the dictionary then returns raw maps and homotopies.
In particular the transferred edge is the cohomology transport. The degree-one part of the closed inclusion i ′ i' i ′ is
− r 0 ( v 0 ) i 1 ′ + r 1 i 1 ′ − i ( v 0 ) r H , 1 = 0. -r_0(v_0)i'_1+r_1i'_1-i(v_0)r_{H,1}=0. − r 0 ( v 0 ) i 1 ′ + r 1 i 1 ′ − i ( v 0 ) r H , 1 = 0.
The minus sign is ( − 1 ) 1 ⋅ 1 (-1)^{1\cdot1} ( − 1 ) 1 ⋅ 1 in the first star product. Multiplication by p H ( v 0 ) p_H(v_0) p H ( v 0 ) gives r H , 1 = p H ( v 0 ) r 1 i ( v 1 ) r_{H,1}=p_H(v_0)r_1i(v_1) r H , 1 = p H ( v 0 ) r 1 i ( v 1 ) . It is an isomorphism on the zero-differential fibers.
The bundle H M H_M H M is a smooth direct summand of the original finite-projective bundle. Restrict its finite idempotent to S × { ( 0 , 0 ) } S\times\{(0,0)\} S × {( 0 , 0 )} to obtain a bundle H H H with finite-projective C S C_S C S -module of sections. Choose a grading-preserving ordinary connection on H M H_M H M . Its transport along ( s , t x , t y ) (s,tx,ty) ( s , t x , t y ) , 0 ≤ t ≤ 1 0\le t\le1 0 ≤ t ≤ 1 , gives a smooth graded isomorphism
τ : π ∗ H ⟶ H M . \tau:\pi^*H\longrightarrow H_M. τ : π ∗ H ⟶ H M .
Local smooth dependence for this finite-interval transport requires no compactness of S S S . This is an ordinary bundle identification, not an algebroid path from the origin. Pull the minimal structure back by τ \tau τ .
For a minimal cohesive object write its one-form part as θ 1 L 1 + θ 2 L 2 \theta_{1}L_{1}+\theta_{2}L_{2} θ 1 L 1 + θ 2 L 2 , now with values in π ∗ End 0 ( H ) \pi^{*}\operatorname{End}^{0}(H) π ∗ End 0 ( H ) . The η \eta η -coefficient of flatness is − y L 1 + x L 2 = 0 -yL_{1}+xL_{2}=0 − y L 1 + x L 2 = 0 . The global Hadamard argument of Lemma B.3, in this fixed bundle over each s s s , gives
L 1 = x L , L 2 = y L , L ( s , x , y ) = ∫ 0 1 ∂ x L 1 ( s , t x , y ) d t . L_{1}=xL,\qquad L_{2}=yL,\qquad L(s,x,y)=\int_{0}^{1}\partial_{x}L_{1}(s,tx,y)\,\mathrm{d}t. L 1 = xL , L 2 = y L , L ( s , x , y ) = ∫ 0 1 ∂ x L 1 ( s , t x , y ) d t .
Solve the intrinsic endomorphism ODE
u x = − L u , u ( s , 0 , y ) = 1 H s . (73) u_{x}=-Lu,\qquad u(s,0,y)=1_{H_{s}}. \tag*{(73)} u x = − Lu , u ( s , 0 , y ) = 1 H s . (73)
It has an invertible solution for every finite x x x , jointly smooth in ( s , x , y ) (s,x,y) ( s , x , y ) . The local parameter estimates are on compact charts, not on all of S S S ; uniqueness makes solutions in parameter frames agree on overlaps. Since Q u = ( x θ 1 + y θ 2 ) u x Qu=(x\theta_{1}+y\theta_{2})u_{x} Q u = ( x θ 1 + y θ 2 ) u x , the transformed one-form part vanishes.
For a minimal raw object on π ∗ H \pi^{*}H π ∗ H , its triangle equation is (C.1), parameterwise. At h = 0 h=0 h = 0 it is a contravariant Aff + ( 1 ) \operatorname{Aff}_{+}(1) Aff + ( 1 ) -action, and its boundary identity is R ( exp ( h Z m ) , m ) = 1 R(\exp(hZ_{m}),m)=1 R ( exp ( h Z m ) , m ) = 1 . Differentiation gives the same relation − y L 1 + x L 2 = 0 -yL_{1}+xL_{2}=0 − y L 1 + x L 2 = 0 . The ODE (73) kills these infinitesimal matrices. The path-and-cocycle computation in Lemma C.2 then kills the whole edge representation, since Aff + ( 1 ) \operatorname{Aff}_{+}(1) Aff + ( 1 ) is connected. Thus r 1 = 1 H r_{1}=1_{H} r 1 = 1 H after this gauge, not merely to first order.
For clarity, let R H M R_{H_{M}} R H M denote the minimal raw object on H M H_{M} H M . Let R min R_{\min} R m i n be its transport to π ∗ H \pi^{*}H π ∗ H through τ \tau τ , and let R t r i v R_{\mathrm{triv}} R triv be the result of the ODE gauge. Their structures satisfy
r t r i v = u − 1 ⋆ r min ⋆ u . r_{\mathrm{triv}}=u^{-1}\star r_{\min}\star u. r triv = u − 1 ⋆ r m i n ⋆ u .
Since δ i n u = 0 \delta_{\mathrm{in}}u=0 δ in u = 0 for a zero-cochain, u : R t r i v → R min u:R_{\mathrm{triv}}\to R_{\min} u : R triv → R m i n is closed. The map τ : R min → R H M \tau:R_{\min}\to R_{H_{M}} τ : R m i n → R H M is the preceding bundle-transport isomorphism. Thus
g = τ u : R t r i v ⟶ R H M . (74) g=\tau u:R_{\mathrm{triv}}\longrightarrow R_{H_{M}}. \tag*{(74)} g = τ u : R triv ⟶ R H M . (74)
Composing with the perturbed inclusion i ′ i' i ′ gives the required homotopy equivalence. If H = 0 H=0 H = 0 , the finite transfer instead contracts the object to zero. Empty graded pieces, acyclic tails and disconnected bases cause no change. □
The actual family Riemann–Hilbert theorem
Theorem 13.4 (Global regular vertical-family RH). For every S S S as above, the actual functors
IN S K = Ψ S ∗ IN E S : Coh r e g ( B S ) ⟶ Rep s m p e r f , r e g ( y K S ) , \operatorname{IN}_{S}^{K}=\Psi_{S}^{*}\operatorname{IN}_{E_{S}}:\operatorname{Coh}^{\mathrm{reg}}(B_{S})\longrightarrow\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf,reg}}(yK_{S}), IN S K = Ψ S ∗ IN E S : Coh reg ( B S ) ⟶ Rep sm perf , reg ( y K S ) ,
IN E S : Coh r e g ( B S ) ⟶ Rep s m p e r f , r e g ( Int ( E S ) ) \operatorname{IN}_{E_{S}}:\operatorname{Coh}^{\mathrm{reg}}(B_{S})\longrightarrow\operatorname{Rep}_{\mathrm{sm}}^{\mathrm{perf,reg}}(\operatorname{Int}(E_{S})) IN E S : Coh reg ( B S ) ⟶ Rep sm perf , reg ( Int ( E S ))
are smooth strictly unital A ∞ A_{\infty} A ∞ quasi-equivalences. Their unary Hom maps are quasi-isomorphisms in every integer degree. Every regular target has a global regular cohesive preimage and an actual closed degree-zero homotopy equivalence from its integration.
Proof. The actual functors, their smoothness and strict units are those of Theorem 6.1 . We prove the additional Hom and reconstruction assertions.
Finite-projective scalar extensions. By Lemma 13.3 , source objects reduce to basic minimal coefficients on π ∗ H , π ∗ H ′ \pi^*H,\pi^*H' π ∗ H , π ∗ H ′ . Put
P q = Γ ( S , Hom q ( H , H ′ ) ) . P^q=\Gamma(S,\operatorname{Hom}^q(H,H')). P q = Γ ( S , Hom q ( H , H ′ )) .
This is a finite-projective C S C_S C S -module. If P q = e C S N P^q=eC_S^N P q = e C S N , extension to forms of CE degree r r r gives e ( B S r ) N e(B_S^r)^N e ( B S r ) N , and extension to normalized k k k -cochains gives e ( C k ) N e(C^k)^N e ( C k ) N . Consequently the actual source and target coefficient spaces are respectively
B S r ⊗ C S P q , C k ⊗ C S P q . (75) B_S^r\otimes_{C_S}P^q,\qquad C^k\otimes_{C_S}P^q. \tag*{(75)} B S r ⊗ C S P q , C k ⊗ C S P q . (75)
The idempotent depends only on s s s and commutes with normalization. These identities retain all jointly smooth functions; they do not identify smooth functions on a product with an algebraic tensor product of two function spaces.
We use C S C_S C S , rather than A S A_S A S , as the central scalar algebra annihilated by the differentials. Every vertex of a simplex has the same s s s , so C S C_S C S is central in C C C and d f u l l ( C S ) = Q ( C S ) = 0 d_{\mathrm{full}}(C_S)=Q(C_S)=0 d full ( C S ) = Q ( C S ) = 0 . Each actual integration kernel is separately C S C_S C S -multilinear: an input factor c ( s ) c(s) c ( s ) is constant throughout that simplex and its integrals. Thus the graded matrix extension descends to (13.9). Its suspended and unsuspended signs remain the distinct formulas (6.3) and (6.4). Parameter frames are legitimate here because an entire simplex lies over one parameter, unlike an arbitrary object-frame domain.
All Hom degrees. Filter the Hom complexes increasingly by internal coefficient degree. The filtration is finite. The source internal differential and one-form connection vanish, and the integrated target has r 0 = 0 r_0=0 r 0 = 0 , r 1 = 1 H r_1=1_H r 1 = 1 H . The associated-graded differentials are therefore Q Q Q and d f u l l d_{\mathrm{full}} d full , tensored with 1 P q 1_{P^q} 1 P q . Every non-bare word in the twisted unary map contains a strictly negative internal-degree connection component. Its only filtration-preserving term is consequently
gr q ( IN 1 K S ) = I S ⊗ C S 1 P q . \operatorname{gr}^q(\operatorname{IN}_1^{K_S})=I_S\otimes_{C_S}1_{P^q}. gr q ( IN 1 K S ) = I S ⊗ C S 1 P q .
where I S I_S I S is the actual scalar comparison of Corollary B.12 . Since P q P^q P q is finite projective, this is a direct summand of a finite sum of quasi-isomorphisms I S I_S I S . Induction through the finite filtration proves every integer Hom degree. As in Lemma B.4 , the actual second Taylor identity and strict unit transfer the conclusion back through the source homotopy equivalences. This transfer uses homotopy equivalences within the regular subcategories.
Global reconstruction. For a regular raw target R R R , apply Lemma 13.3 and write its edge-trivialized replacement as e H + b e_H+b e H + b on π ∗ H \pi^*H π ∗ H . Let
n H = ⨁ q < 0 Γ ( S , End q H ) , F ℓ n H = ⨁ q ≤ − ℓ Γ ( S , End q H ) . \mathfrak{n}_H=\bigoplus_{q<0}\Gamma(S,\operatorname{End}^q H),\qquad F^\ell\mathfrak{n}_H=\bigoplus_{q\leq-\ell}\Gamma(S,\operatorname{End}^q H). n H = q < 0 ⨁ Γ ( S , End q H ) , F ℓ n H = q ≤ − ℓ ⨁ Γ ( S , End q H ) .
This is a finite multiplicative nilpotent filtration. Its layers are finite projective over C S C_S C S , regardless of the ranks of individual endomorphism sections. Restrict the actual functor to the endomorphisms of the scalar unit and use its just-established parameter-linear matrix extension. It gives an A ∞ A_\infty A ∞ map on global DGAs
F H : B S ⊗ C S n H ⇝ C ⊗ C S n H . F^H:B_S\otimes_{C_S}\mathfrak{n}_H\rightsquigarrow C\otimes_{C_S}\mathfrak{n}_H. F H : B S ⊗ C S n H ⇝ C ⊗ C S n H .
Matrix composition gives the arity-additive filtration bound of (C.2). In layer ℓ \ell ℓ the unary map is IS ⊗ 1 Γ ( S , End − ℓ H ) \operatorname{IS} \otimes1_{\Gamma(S,\operatorname{End}^{-\ell} H)} IS ⊗ 1 Γ ( S , End − ℓ H ) , a quasi-isomorphism. Lemma C.3 applies to these global complexes over R \mathbb{R} R ; it does not require them to be finite-dimensional real vector spaces. It gives a global MC element μ \mu μ and an actual finite gauge q q q with b q = F ∗ H ( μ ) b^{q}=F_{*}^{H}(\mu) b q = F ∗ H ( μ ) . In particular the primitives are global smooth sections, not unrelated choices on parameter fibers.
The source object
P = ( Γ ( S , H ) ⊗ C S B S , Q + μ ) P = \left(\Gamma(S,H) \otimes_{C_S} B_S,Q+\mu\right) P = ( Γ ( S , H ) ⊗ C S B S , Q + μ )
uses the basic scalar connection Q Q Q and the forms-first matrix action of μ \mu μ . It is bounded finite projective and regular. Its internal differential and one-form part vanish, because a total degree-one element with negative internal degree has CE degree at least two. The actual object formula is literally
IN K S ( P ) = e H + F ∗ H ( μ ) . (76) \operatorname{IN}_{K_S}(P)=e_H+F_{*}^{H}(\mu). \tag*{(76)} IN K S ( P ) = e H + F ∗ H ( μ ) . (76)
Indeed every MC input has total degree one, so the outer unsuspension tensor sign is zero, as in the proof of Theorem C.4. The matrix sign remains (6.4), and (6.6) needs no second target conversion. All these object words are finite by negative internal degree.
The gauge q q q is a normalized closed degree-zero raw morphism from (13.10) to e H + b e_H+b e H + b , with direction fixed by (C.3). It is strictly invertible by a finite series and has q 0 = 1 H q_0=1_H q 0 = 1 H ; it can have higher cochain components. With the preceding τ , u , i ′ \tau,u,i' τ , u , i ′ , the actual counit is
ϵ R = i ′ τ u q : IN K S ( P ) ⟶ R . (77) \epsilon_R=i'\tau uq:\operatorname{IN}_{K_S}(P)\longrightarrow R. \tag*{(77)} ϵ R = i ′ τ u q : IN K S ( P ) ⟶ R . (77)
Here τ u \tau u τ u is a vertex-only invertible map, whereas i ′ i' i ′ is a homotopy equivalence, not necessarily invertible. This proves essential surjectivity on H 0 H^0 H 0 .
Full MC. Lemma 13.1 gives the actual raw dg quasi-equivalence Ψ S ∗ \Psi_S^* Ψ S ∗ and its unit-relative invertible gauges U Z : Z → Φ S ∗ Ψ S ∗ Z U_Z:Z\to\Phi_S^*\Psi_S^*Z U Z : Z → Φ S ∗ Ψ S ∗ Z . For a reconstructed map v : Ψ S ∗ IN E S ( P ) → Ψ S ∗ R v:\Psi_S^*\operatorname{IN}_{E_S}(P)\to\Psi_S^*R v : Ψ S ∗ IN E S ( P ) → Ψ S ∗ R , the map
U R − 1 Φ S ∗ ( v ) U IN E S ( P ) : IN E S ( P ) ⟶ R (78) U_R^{-1}\Phi_S^*(v)U_{\operatorname{IN}_{E_S}(P)}:\operatorname{IN}_{E_S}(P)\longrightarrow R \tag*{(78)} U R − 1 Φ S ∗ ( v ) U IN E S ( P ) : IN E S ( P ) ⟶ R (78)
is an actual closed degree-zero homotopy equivalence. The Hom comparison transfers through the same Ψ S ∗ \Psi_S^* Ψ S ∗ . This proves the theorem for the intrinsic full-MC functor. □
Corollary 13.5 (Parameter pullback). For every smooth map f : S ′ → S f:S'\to S f : S ′ → S , the actual family integration functors commute with parameter pullback in every Taylor arity, with the canonical identifications compatible with identities and composition. They give an equivalence of contravariant Cat ∞ \operatorname{Cat}_{\infty} Cat ∞ -valued diagrams on the ordinary category of smooth parameter manifolds and maps.
Proof. Parameter pullback uses the scalar map C ∞ ( S ) → C ∞ ( S ′ ) C^\infty(S)\to C^\infty(S') C ∞ ( S ) → C ∞ ( S ′ ) , c ↦ c ∘ f c\mapsto c\circ f c ↦ c ∘ f . Finite-projective coefficients pull back through finite idempotents. Regularity is preserved by pulling back the actual smooth splitting of the internal complex; no flatness of this ring map is required. The product descriptions of E S , K S , Φ S , Ψ S E_S,K_S,\Phi_S,\Psi_S E S , K S , Φ S , Ψ S and the absence of parameter differentiation give, in every Taylor arity, the actual identities
( f × K ) ∗ IN K S ≅ IN K S ′ ( f × 1 R 2 ) ∗ , (f\times K)^*\operatorname{IN}_{K_S}\cong\operatorname{IN}_{K_{S'}}(f\times1_{\mathbb{R}^2})^*, ( f × K ) ∗ IN K S ≅ IN K S ′ ( f × 1 R 2 ) ∗ ,
and the corresponding identities for full MC integration. For bundles, these use the canonical composition identifications of pullback. Equivalently, in idempotent models pullback is entrywise composition; the identities then respect composition directly. Passing to the dg/A ∞ A_\infty A ∞ nerves gives the map of contravariant diagrams. Theorem 13.4 makes it pointwise an equivalence, hence an equivalence in the functor infinity-category. □
The coherent inverse exists in the functor infinity-category. Metrics, contractions, ODE gauges and MC primitives in the proof are chosen objectwise; parameter naturality comes from the forward integration functors. The parameter is passive: the coefficient structure uses functions on S S S and differentiates only in the affine directions. A flat connection in the S S S directions would be additional coefficient data. The indexing category is the ordinary category of smooth parameter manifolds and maps, before localization at parameter homotopies.
Two-term bundle arrows and their examples
Theorem 13.6 (Two-term family normal forms). Regular objects of internal cohomological amplitude [ 0 , 1 ] [0,1] [ 0 , 1 ] in the three categories of Theorem 13.4 are classified up to H 0 H^0 H 0 -isomorphism by isomorphism classes of smooth bundle maps
N : H 1 ⟶ H 0 N:H^1\longrightarrow H^0 N : H 1 ⟶ H 0 over S S S , with Γ ( S , H 0 ) \Gamma(S,H^0) Γ ( S , H 0 ) and Γ ( S , H 1 ) \Gamma(S,H^1) Γ ( S , H 1 ) finite projective. The rank of N N N may vary. The cohesive representative on π ∗ ( H 0 ⊕ H 1 ) \pi^*(H^0\oplus H^1) π ∗ ( H 0 ⊕ H 1 ) is
D N ( v 0 ) = 0 , D N ( v 1 ) = N ( s ) ( v 1 ) η (79) D_N(v_0)=0,\qquad D_N(v_1)=N(s)(v_1)\eta \tag*{(79)} D N ( v 0 ) = 0 , D N ( v 1 ) = N ( s ) ( v 1 ) η (79) on basic sections, extended by the right Leibniz rule. Denote this object by P N P_N P N . Its actual integrated representative R N R_N R N on K S K_S K S is
r 0 = 0 , r 1 = 1 H , r 2 = − h 012 N ( s ) , r p = 0 ( p ≥ 3 ) . (80) r_0=0,\qquad r_1=1_H,\qquad r_2=-h_{012}N(s),\qquad r_p=0\quad(p\ge3). \tag*{(80)} r 0 = 0 , r 1 = 1 H , r 2 = − h 012 N ( s ) , r p = 0 ( p ≥ 3 ) . (80)
Proof. Finite minimalization removes acyclic tails and leaves π ∗ ( H 0 ⊕ H 1 ) \pi^*(H^0\oplus H^1) π ∗ ( H 0 ⊕ H 1 ) . After the one-form gauge, degree counting permits only a two-form component from H 1 H^1 H 1 to H 0 H^0 H 0 . Write it as N ( s , x , y ) η + T ( s , x , y ) θ 1 θ 2 N(s,x,y)\eta+T(s,x,y)\theta_1\theta_2 N ( s , x , y ) η + T ( s , x , y ) θ 1 θ 2 . Products of these negative-degree endomorphisms vanish. Equation (B.5), with parameters, gives T = N x T=N_x T = N x : the difference is killed by both x x x and y y y , and continuity covers the origin. Thus the remaining coefficient is b N = N η + N x θ 1 θ 2 b_N=N\eta+N_x\theta_1\theta_2 b N = N η + N x θ 1 θ 2 .
All its values at fixed s s s lie in Hom ( H s 1 , H s 0 ) \operatorname{Hom}(H^1_s,H^0_s) Hom ( H s 1 , H s 0 ) . The following integrals therefore define global bundle-valued functions:
N 0 ( s ) = N ( s , 0 , 0 ) , P = − ∫ 0 1 N y ( s , t x , t y ) d t , R = ∫ 0 1 N x ( s , t x , t y ) d t , N_0(s)=N(s,0,0),\qquad P=-\int_0^1N_y(s,tx,ty)\,\mathrm{d}t,\qquad R=\int_0^1N_x(s,tx,ty)\,\mathrm{d}t, N 0 ( s ) = N ( s , 0 , 0 ) , P = − ∫ 0 1 N y ( s , t x , t y ) d t , R = ∫ 0 1 N x ( s , t x , t y ) d t ,
c = P θ 1 + R θ 2 , Q c = ( N − N 0 ) η + N x θ 1 θ 2 = b N − N 0 η . c=P\theta_1+R\theta_2,\qquad Qc=(N-N_0)\eta+N_x\theta_1\theta_2=b_N-N_0\eta. c = P θ 1 + R θ 2 , Q c = ( N − N 0 ) η + N x θ 1 θ 2 = b N − N 0 η .
The last equality follows from the fundamental theorem of calculus and (B.4). The off-diagonal right-linear map 1 + c 1+c 1 + c , with inverse 1 − c 1-c 1 − c , points from the unreduced b N b_N b N object to P N 0 P_{N_0} P N 0 . Indeed on a degree-one basic section, D N 0 ( 1 + c ) = ( 1 + c ) D b N D_{N_0}(1+c)=(1+c)D_{b_N} D N 0 ( 1 + c ) = ( 1 + c ) D b N is precisely the displayed primitive identity. This proves existence globally.
For uniqueness, a closed degree-zero morphism between normal forms has diagonal function maps T 0 , T 1 T_0,T_1 T 0 , T 1 and an off-diagonal one-form C = P θ 1 + R θ 2 C=P\theta_1+R\theta_2 C = P θ 1 + R θ 2 . The closedness calculation (64) is valid in these bundles. At the zero section, the η \eta η -coefficient of Q C QC QC is − y P + x R = 0 -yP+xR=0 − y P + x R = 0 , so
N ′ T 1 ( s , 0 , 0 ) = T 0 ( s , 0 , 0 ) N . (81) N'T_1(s,0,0)=T_0(s,0,0)N. \tag*{(81)} N ′ T 1 ( s , 0 , 0 ) = T 0 ( s , 0 , 0 ) N . (81)
An actual inverse modulo Hom differentials restricts, together with its homotopies, to a vertex homotopy inverse there. The internal differentials of the normal forms are zero. Consequently the restricted T i T_i T i are actual smooth bundle isomorphisms. Here evaluation acts on the inverse maps and their specified homotopies. Conversely any bundle isomorphisms intertwining N , N ′ N,N' N , N ′ are basic closed degree-zero maps and give a strict cohesive isomorphism. This proves the bundle-arrow classification.
The actual functors and counits preserve vertex complexes up to the specified homotopies, so Theorem 13.4 restricts to the indicated amplitude full subcategories. For (13.13), every object word with two occurrences of N N N vanishes. The actual unary kernel integrates η \eta η to − h -h − h on K S K_S K S , giving (13.14) and transferring the same classification to both raw targets. Regularity concerns the zero internal differential, not the possibly varying rank of N N N . □
The invariant is a global bundle-arrow isomorphism class, not a pointwise rank profile. This classifies objects up to H 0 H^0 H 0 -isomorphism; their Hom complexes remain those of the respective dg categories.
Example 13.7 (Nontrivial coefficient topology). On S 1 S^1 S 1 , the periodic idempotent
p ( θ ) = 1 2 ( 1 + cos θ sin θ sin θ 1 − cos θ ) p(\theta)=\frac{1}{2}\begin{pmatrix}1+\cos\theta& \sin\theta\\ \sin\theta& 1-\cos\theta\end{pmatrix} p ( θ ) = 2 1 ( 1 + cos θ sin θ sin θ 1 − cos θ ) has image the real Möbius line L L L . It explicitly makes Γ ( S 1 , L ) \Gamma(S^1,L) Γ ( S 1 , L ) finite projective. A unit spanning vector on [ 0 , 2 π ] [0,2\pi] [ 0 , 2 π ] is v ( θ ) = ( cos ( θ / 2 ) , sin ( θ / 2 ) ) v(\theta)=(\cos(\theta/2),\sin(\theta/2)) v ( θ ) = ( cos ( θ /2 ) , sin ( θ /2 )) , with v ( 2 π ) = − v ( 0 ) v(2\pi)=-v(0) v ( 2 π ) = − v ( 0 ) . A nowhere-zero section would require a continuous nowhere-zero real coefficient a ( θ ) a(\theta) a ( θ ) with a ( 2 π ) = − a ( 0 ) a(2\pi)=-a(0) a ( 2 π ) = − a ( 0 ) , which is impossible. Thus L L L is not trivial.
The arrows Id L : L → L \operatorname{Id}_L:L\to L Id L : L → L and Id R : R → R \operatorname{Id}_{\mathbb{R}}:\mathbb{R}\to\mathbb{R} Id R : R → R have the same fiberwise dimension/rank triple ( 1 , 1 , 1 ) (1,1,1) ( 1 , 1 , 1 ) , but are not isomorphic bundle arrows. Their cohesive and integrated family coefficients are therefore inequivalent. The nontriviality here is in the coefficient bundles; the vertical affine resolution itself has free terms.
Example 13.8 (Smooth vanishing order beyond rank). For S = R S=\mathbb{R} S = R and trivial coefficient lines, the residues N ( t ) = t N(t)=t N ( t ) = t and N ′ ( t ) = t 3 N'(t)=t^3 N ′ ( t ) = t 3 have the same rank at every parameter. An isomorphism of bundle arrows would give nowhere-zero smooth functions a , b a,b a , b with t 3 b = a t t^3b=at t 3 b = a t . For t ≠ 0 t\ne0 t = 0 this forces a = t 2 b a=t^2b a = t 2 b , and continuity gives a ( 0 ) = 0 a(0)=0 a ( 0 ) = 0 , a contradiction. The compact example sin θ \sin\theta sin θ versus sin 3 θ \sin^3\theta sin 3 θ on S 1 S^1 S 1 has the same local obstruction at a simple zero. Both are regular coefficient families: changing residue rank is allowed.
The family ordinary-descent obstruction
Let G ‾ S \overline{\mathcal{G}}_S G S be the product of the ordinary quotient of Proposition 12.3 with the unit groupoid on S S S . It is an ordinary groupoid in smooth sheaves, with represented objects y M S yM_S y M S . The strict product map is
q S : y K S ⟶ N ( G ‾ S ) , q_S:yK_S\longrightarrow N(\overline{\mathcal{G}}_S), q S : y K S ⟶ N ( G S ) ,
and on full MC use q S Φ S q_S\Phi_S q S Φ S .
Corollary 13.9 (Exact family descent criterion). A regular two-term family lies in the essential image on H 0 H^0 H 0 of q S ∗ q_S^* q S ∗ exactly when its bundle-map residue N N N is identically zero. The same criterion holds for ( q S Φ S ) ∗ (q_S\Phi_S)_* ( q S Φ S ) ∗ on full MC. If N ≠ 0 N \ne0 N = 0 , descent is impossible through every strict map to an ordinary groupoid nerve with represented objects, even when the bounded finite-projective target coefficient is not regular or minimal.
Proof. If N = 0 N=0 N = 0 , the basic graded bundle over S S S has identity transport along every arrow of G S \mathcal{G}_S G S . Its pullback is (13.14) with zero higher components. The normal-form equivalences give the assertion for any representative of that class.
If N ≠ 0 N \ne0 N = 0 , choose s ∈ S s \in S s ∈ S with N s ≠ 0 N_s \ne0 N s = 0 . Strict restriction to { s } × y K \{s\} \times yK { s } × y K preserves the actual inverse maps and homotopies in any proposed H 0 H^0 H 0 -equivalence. A hypothetical pullback through any strict ordinary-nerve map would therefore give the forbidden point-parameter descent of Theorem 12.4 . That theorem permits these ordinary targets, including nonregular target coefficients. For full MC, first pull back along the actual Ψ S \Psi_S Ψ S , then restrict the parameter. The literal identity Ψ S ∗ I N E S ( P N ) = R N \Psi_S^* \mathrm{IN}_{E_S}(P_N)=R_N Ψ S ∗ IN E S ( P N ) = R N gives the same contradiction.
The quotient is used as a groupoid in smooth sheaves, without a representability hypothesis on its arrow sheaf. Comparison with AS or refined higher holonomy is a further geometric question. These regular families vary the parameter manifold and coefficient bundles while keeping the affine algebroid fixed.
Further directions
Theorem 10.4 covers every ordinary open O ⊆ R 2 O \subseteq\mathbb{R}^2 O ⊆ R 2 , including nonrectangular, disconnected, noncompact and origin-containing opens, on globally bounded degreewise finite-projective real coefficient bodies whose cohomology ranks may jump. The whole base simplex, not merely its vertices, must remain in O O O . The even generator η \eta η retains every polynomial power. The theorem compares the concrete cohesive and raw categories, rather than the ordinary category Perf ( B O ) \operatorname{Perf}(BO) Perf ( B O ) .
Actual differentiation and global smooth kernel acyclicity give invertible identity-body counits, chosen objectwise. Internal triangularity supplies a real-linear chain inverse that conjugates the transferred and cohesive differentials. Choosing counits coherently over coefficient moduli, and making the inverse sections compatible with restriction between opens, are further compatibility problems.
Parameters and effective descent. The regular vertical-family theorem of § 13 assumes regular body cohomology. Its globally bounded finite-projective bundles need not be trivial, and the parameter manifold need not be compact or connected. The anchor is vertical, rather than that of an inverse-image algebroid adding T S TS T S . Corollary 13.5 gives natural forward functors and an equivalence of parameter diagrams. The proof chooses metrics, contractions and MC primitives objectwise, and the parameter S S S remains passive.
Beyond fixed-product families, one can ask for a coefficient comparison when the affine domain O s ⊆ R 2 O_s \subseteq\mathbb{R}^2 O s ⊆ R 2 varies with s ∈ S s \in S s ∈ S , or when the parameter directions carry connection or anchor data. Such extensions require a jointly smooth global comparison compatible with the new geometry. Effective descent asks a different question: when do compatible local coefficients assemble into globally bounded finite-projective objects? Such an argument must control the amplitude and projectivity bounds together with the higher gluing data.
Holonomy and reconstruction. Monodromy and derived-transverse refined higher holonomy [13 ] are distinct constructions. The ordinary quotient used here is a groupoid in smooth sheaves, obtained from parallel 2-isomorphism classes. Theorem 12.4 concerns strict maps to ordinary nerves in the actual raw coefficient theory. Corollary 13.9 retains its exact zero-residue criterion for regular vertical families. An intrinsic marked affine image and base change along surjective submersions with connected fibres for closed saturated marks are constructed in [13 Theorems C–D] . Comparing our coefficient functor with these images requires compatibility of the entire source maps, not only agreement on the central degree-two class. An equivalence of coefficient categories would additionally need a suitable descent theorem.
Example 11.3 shows that coefficient RH does not make origin observation reflect coefficient homotopy equivalences, even on finite coefficients. A reconstruction statement through a chosen observer must specify whether it concerns the full coefficient category or a quotient detected by that observer. A monoidal reconstruction would additionally require compatible monoidal structures, and a geometric reconstruction would require the corresponding geometric interpretation. For derived-complete enlargements, the corresponding question must also specify the completion and its coefficient hypotheses.
A complementary infinitesimal perspective comes from CE filtrations. For finite-length universal resolutions, the fibre-weight spectral sequence is intrinsic from E 1 E_1 E 1 [11 Theorem 3.2] . Finite-type minimal isotropy carries a separate pointwise sequence [11 Theorem 4.1] ; in the holomorphic cubic Koszul example, a pointwise weight-one class has no unary-closed stalk lift [11 Proposition 6.2] . These filtrations suggest finer tests for what origin observation retains from coefficient transport.
Convolution and the affine relation. The same affine action appears in the convolution-algebra construction of [4 Theorem 6.3] . Its origin-germ AS relation module is the image of an injective right-module operator D = − a y ∂ a + a x ∂ b − y D = -ay \partial_a + ax \partial_b - y D = − a y ∂ a + a x ∂ b − y in the total-density frame. The vector-field part differentiates along the stabilizer relation, while the zeroth-order term comes from its density action. A comparison with the coefficient residue studied here would have to relate these distinct module categories and account for this density twist.
Beyond the affine model. For general Lie- and L ∞ L_\infty L ∞ -algebroids, one can ask when the integration functor of [12 ] is a quasi-equivalence. Related questions concern local-resolution effectivity and coherence at the level of coefficient categories. A source-valued cohesive factorization would require an enriched construction compatible with the integration and invariance results used here. Further comparison problems concern varying-base Morita maps and extensions to derived, Banach or noncommutative geometries.
Global affine lifting and deformation retraction
Throughout this appendix, M = R 2 M = \mathbb{R}^{2} M = R 2 , G = Aff + ( 1 ) G = \operatorname{Aff}_{+}(1) G = Aff + ( 1 ) , and K K K is the crossed-module nerve of § 3 . All differentials and forms are in the simplex direction, with jointly smooth external parameters. We prove the lifting and retraction results stated in that section; the extension operators are also used in Section 7.
Holonomy and flat correction
For an edge matrix A ( ∂ t ) = ( q 1 ( t ) q 2 ( t ) 0 0 ) A(\partial_t)=\begin{pmatrix}q_1(t)&q_2(t)\\0&0\end{pmatrix} A ( ∂ t ) = ( q 1 ( t ) 0 q 2 ( t ) 0 ) , the solution has entries
r ( t ) = exp ( ∫ 0 t q 1 ( s ) d s ) , s ( t ) = r ( t ) ∫ 0 t r ( v ) − 1 q 2 ( v ) d v . r(t)=\exp\left(\int_0^t q_1(s)\,ds\right),\qquad s(t)=r(t)\int_0^t r(v)^{-1}q_2(v)\,dv. r ( t ) = exp ( ∫ 0 t q 1 ( s ) d s ) , s ( t ) = r ( t ) ∫ 0 t r ( v ) − 1 q 2 ( v ) d v .
Thus it exists on the whole edge and stays in G G G . The formulas preserve arbitrary joint smooth U U U -dependence: the integrations have compact fixed fibers, and their parameter derivatives are computed locally in U U U . The base equation gives σ ( v j ) = g i j σ ( v i ) \sigma(v_j)=g_{ij}\sigma(v_i) σ ( v j ) = g ij σ ( v i ) .
Let H v H_v H v be the radial Poincare operator on a simplex based at a vertex v v v . It integrates the d s ds d s component of pullback by ( s , t ) ↦ v + s ( t − v ) (s,t)\mapsto v+s(t-v) ( s , t ) ↦ v + s ( t − v ) , so
d H v + H v d = Id − v ∗ . (82) dH_v+H_vd=\operatorname{Id}-v^*. \tag*{(82)} d H v + H v d = Id − v ∗ . (82)
For a closed 2-form, u = H v β u=H_v\beta u = H v β satisfies d u = β du=\beta d u = β and restricts to zero on all edges issuing from v v v .
Lemma A.1 (Rank-uniform flat correction). Suppose ( σ , A , β ) (\sigma,A,\beta) ( σ , A , β ) satisfies (3.5), and let u u u be a one-form with d u = β du=\beta d u = β . Then
A ′ = A − Z σ u A'=A-Z_\sigma u A ′ = A − Z σ u is flat and satisfies d σ = A ′ σ d\sigma=A'\sigma d σ = A ′ σ . A flat matrix form A ′ = ( ξ ζ 0 0 ) A'=\begin{pmatrix}\xi&\zeta\\0&0\end{pmatrix} A ′ = ( ξ 0 ζ 0 ) on a simplex has a unique potential g g g with d g g − 1 = A ′ dg\,g^{-1}=A' d g g − 1 = A ′ and any prescribed initial value in G G G at one vertex.
Proof. The base equation is unchanged because Z σ σ = 0 Z_\sigma\sigma=0 Z σ σ = 0 . Expanding curvature, the terms Z σ β − Z σ d u Z_\sigma\beta-Z_\sigma du Z σ β − Z σ d u cancel. The mixed terms cancel by d Z σ = A Z σ − Z σ A dZ_\sigma=AZ_\sigma-Z_\sigma A d Z σ = A Z σ − Z σ A and u ∧ A = − A ∧ u u\wedge A=-A\wedge u u ∧ A = − A ∧ u . The remaining quadratic term vanishes because u ∧ u = 0 u\wedge u=0 u ∧ u = 0 .
For the second assertion, flatness means d ξ = 0 d\xi=0 d ξ = 0 and d ζ = ξ ∧ ζ d\zeta=\xi\wedge\zeta d ζ = ξ ∧ ζ . Put
p = H v ξ , q = H v ( e − p ζ ) , g n o r m = ( e p e p q 0 1 ) . p=H_v\xi,\qquad q=H_v(e^{-p}\zeta),\qquad g_{\mathrm{norm}}=\begin{pmatrix}e^p&e^p q\\0&1\end{pmatrix}. p = H v ξ , q = H v ( e − p ζ ) , g norm = ( e p 0 e p q 1 ) . Equation (A.1) gives d p = ξ dp=\xi d p = ξ and d q = e − p ζ dq=e^{-p}\zeta d q = e − p ζ . Hence d g n o r m g n o r m − 1 = A ′ dg_{\mathrm{norm}}g_{\mathrm{norm}}^{-1}=A' d g norm g norm − 1 = A ′ and g n o r m ( v ) = 1 g_{\mathrm{norm}}(v)=1 g norm ( v ) = 1 . Right multiplication by the prescribed initial value gives g g g . If g 1 , g 2 g_1,g_2 g 1 , g 2 are two such potentials with the same initial value, g 1 − 1 g 2 g_1^{-1}g_2 g 1 − 1 g 2 has zero differential on the connected simplex and is the identity. All these operations preserve joint smooth parameters. □
For a potential based at v 0 v_0 v 0 with g ( v 0 ) = 1 g(v_0)=1 g ( v 0 ) = 1 , the base equation gives σ = g m 0 \sigma=gm_0 σ = g m 0 : d ( g − 1 σ ) = 0 d(g^{-1}\sigma)=0 d ( g − 1 σ ) = 0 and m 0 = σ ( v 0 ) m_0=\sigma(v_0) m 0 = σ ( v 0 ) . Equivariance gives Z σ = g Z m 0 g − 1 Z_\sigma=gZ_{m_0}g^{-1} Z σ = g Z m 0 g − 1 . Solving the edge equation in this gauge yields
g i j ( A ) = g ( v j ) exp ( Z m 0 ∫ [ i j ] u ) g ( v i ) − 1 . (83) g_{ij}(A)=g(v_j)\exp\left(Z_{m_0}\int_{[ij]}u\right)g(v_i)^{-1}. \tag*{(83)} g ij ( A ) = g ( v j ) exp ( Z m 0 ∫ [ ij ] u ) g ( v i ) − 1 . (83)
This can be verified directly by differentiating g ( t ) exp ( Z m 0 ∫ 0 t u ) g ( v i ) − 1 g(t)\exp\left(Z_{m_0}\int_0^t u\right)g(v_i)^{-1} g ( t ) exp ( Z m 0 ∫ 0 t u ) g ( v i ) − 1 . On a triangle based at v i v_i v i , take u = H v i β u=H_{v_i}\beta u = H v i β . Its two root-edge restrictions vanish, so Stokes gives ∫ [ j k ] u = ∫ [ i j k ] β = − h i j k \int_{[jk]}u=\int_{[ijk]}\beta=-h_{ijk} ∫ [ j k ] u = ∫ [ ij k ] β = − h ij k . Equation (A.2) then gives
g j k = g i k exp ( − h i j k Z σ ( v i ) ) g i j − 1 . g_{jk}=g_{ik}\exp\left(-h_{ijk}Z_{\sigma(v_i)}\right)g_{ij}^{-1}. g j k = g ik exp ( − h ij k Z σ ( v i ) ) g ij − 1 .
which is exactly (3.3). Stokes on a tetrahedron gives δ h = 0 \delta h=0 δ h = 0 . Thus the actual edges and triangles form a K K K -simplex. They commute with faces and degeneracies: affine simplex maps carry an edge to the corresponding edge or a constant edge, and a triangle to the corresponding triangle or a map of rank at most one. Degenerate coordinates are consequently the identity and zero. This proves that Φ \Phi Φ is a natural simplicial sheaf map, independently of choices in the subsequent lifting argument.
Write Ω r ( ∂ Δ n ) \Omega^r(\partial\Delta^n) Ω r ( ∂ Δ n ) for compatible tangential forms on the codimension-one faces. Only their pullbacks to intersections are identified; normal derivatives are not specified.
Lemma A.2 (Smooth face extension). For n ≥ 1 n\geq1 n ≥ 1 there is a linear extension operator
Ext n : Ω r ( ∂ Δ n ) ⟶ Ω r ( Δ n ) \operatorname{Ext}_n:\Omega^r(\partial\Delta^n)\longrightarrow\Omega^r(\Delta^n) Ext n : Ω r ( ∂ Δ n ) ⟶ Ω r ( Δ n ) right inverse to restriction, including for functions when r = 0 r=0 r = 0 . It preserves joint smooth parameter families and commutes with pullback in the external parameter.
Proof. For F i = { t i = 0 } F_i=\{t_i=0\} F i = { t i = 0 } use the retraction
( p i ( t ) ) i = 0 , ( p i ( t ) ) j = t j 1 − t i ( j ≠ i ) . (p_i(t))_i=0,\qquad(p_i(t))_j=\frac{t_j}{1-t_i}\quad(j\ne i). ( p i ( t ) ) i = 0 , ( p i ( t ) ) j = 1 − t i t j ( j = i ) .
on t i < 1 t_i<1 t i < 1 . Fix a smooth cutoff χ \chi χ equal to one near zero and zero on a neighborhood of [ 1 / 2 , 1 ] [1/2,1] [ 1/2 , 1 ] . Extend a face form ω \omega ω by χ ( t i ) p i ∗ ω \chi(t_i)p_i^*\omega χ ( t i ) p i ∗ ω , with zero continuation away from that collar. This is smooth through all corners; the denominator does not degenerate on the support. Process the faces in order. On F i F_i F i subtract the restriction of the extension already made, and extend this residual by the displayed operator. Compatibility makes the residual vanish after pullback to earlier face intersections. The map p i p_i p i sends F j F_j F j into F i ∩ F j F_i\cap F_j F i ∩ F j where defined, so the new term restricts to zero on every earlier F j F_j F j . After finitely many steps all face restrictions are correct. The construction uses only fixed smooth pullbacks, multiplication and addition, so it has the stated parameter properties. □
For an ordered p p p -face I = ( i 0 , … , i p ) I=(i_0,\ldots,i_p) I = ( i 0 , … , i p ) put
ω I = p ! ∑ r = 0 p ( − 1 ) r t i r d t i 0 ∧ ⋯ ∧ d t i r ^ ∧ ⋯ ∧ d t i p . \omega_I=p!\sum_{r=0}^{p}(-1)^r t_{i_r}\,dt_{i_0}\wedge\cdots\wedge\widehat{dt_{i_r}}\wedge\cdots\wedge dt_{i_p}. ω I = p ! r = 0 ∑ p ( − 1 ) r t i r d t i 0 ∧ ⋯ ∧ d t i r ∧ ⋯ ∧ d t i p .
Extend this notation alternatingly, with repeated indices giving zero. Barycentric identities give
d ω I = ∑ j ω j I , ∫ J ω I = δ I J . (84) d\omega_I=\sum_j\omega_{jI},\qquad\int_J\omega_I=\delta_{IJ}. \tag*{(84)} d ω I = j ∑ ω j I , ∫ J ω I = δ I J . (84)
Indeed the first identity follows from ∑ t i = 1 \sum t_i=1 ∑ t i = 1 and ∑ d t i = 0 \sum dt_i=0 ∑ d t i = 0 ; on its own face ω I \omega_I ω I is p ! p! p ! times the oriented coordinate volume form, and on another p p p -face it vanishes. For a simplicial cochain c c c , write W ( c ) = ∑ I ω I c I W(c)=\sum_I\omega_Ic_I W ( c ) = ∑ I ω I c I . Equation (A.3) gives d W ( c ) = W ( δ c ) dW(c)=W(\delta c) d W ( c ) = W ( δ c ) .
Lemma A.3 (A smooth degree-two radial identity). For a closed smooth 2-form γ \gamma γ , put
α i = H v i γ , f i j = H v j H v i γ , c i j k = ∫ [ i j k ] γ . \alpha_i=H_{v_i}\gamma,\qquad f_{ij}=H_{v_j}H_{v_i}\gamma,\qquad c_{ijk}=\int_{[ijk]}\gamma. α i = H v i γ , f ij = H v j H v i γ , c ij k = ∫ [ ij k ] γ . The operator
D n γ = ∑ i t i α i − ∑ i < j ω i j f i j D_n\gamma=\sum_i t_i\alpha_i-\sum_{i<j}\omega_{ij}f_{ij} D n γ = i ∑ t i α i − i < j ∑ ω ij f ij satisfies
d D n γ = γ − W ( c ) . dD_n\gamma=\gamma-W(c). d D n γ = γ − W ( c ) . It commutes with face restrictions, vanishes after restriction to any edge, and preserves arbitrary joint smooth parameter dependence.
Proof. The radial identity gives d α i = γ d\alpha_i=\gamma d α i = γ and d f i j = α i − α j df_{ij}=\alpha_i-\alpha_j d f ij = α i − α j . The function f i j f_{ij} f ij vanishes at v i v_i v i and v j v_j v j : one endpoint is its radial base, and α i \alpha_i α i vanishes on the radial edge [ i j ] [ij] [ ij ] . Moreover d ( f i j + f j i ) = 0 d(f_{ij}+f_{ji})=0 d ( f ij + f j i ) = 0 , and their common zero endpoint makes this constant zero. Thus f j i = − f i j f_{ji}=-f_{ij} f j i = − f ij . The function f i j + f j k − f i k f_{ij}+f_{jk}-f_{ik} f ij + f j k − f ik is constant. At v k v_k v k it equals ∫ [ j k ] α i \int_{[jk]}\alpha_i ∫ [ j k ] α i , which is c i j k c_{ijk} c ij k by Stokes on [ i j k ] [ijk] [ ij k ] .
Differentiate the formula for D n D_n D n . Since
∑ i < j ω i j ∧ ( α i − α j ) = ∑ i , j t i d t j ∧ ( α i − α j ) = − ∑ i d t i ∧ α i , \sum_{i<j}\omega_{ij}\wedge(\alpha_i-\alpha_j)=\sum_{i,j}t_i\,dt_j\wedge(\alpha_i-\alpha_j)=-\sum_i dt_i\wedge\alpha_i, i < j ∑ ω ij ∧ ( α i − α j ) = i , j ∑ t i d t j ∧ ( α i − α j ) = − i ∑ d t i ∧ α i , the mixed terms cancel, leaving
d D n γ = γ − 2 ∑ i < j d t i ∧ d t j f i j . dD_n\gamma=\gamma-2\sum_{i<j}dt_i\wedge dt_jf_{ij}. d D n γ = γ − 2 i < j ∑ d t i ∧ d t j f ij . Pointwise in the coefficient functions, (A.3) gives
2 ∑ i < j d t i ∧ d t j f i j = ∑ i < j < k ω i j k ( f i j + f j k − f i k ) = W ( c ) . 2\sum_{i<j}dt_i\wedge dt_jf_{ij}=\sum_{i<j<k}\omega_{ijk}(f_{ij}+f_{jk}-f_{ik})=W(c). 2 i < j ∑ d t i ∧ d t j f ij = i < j < k ∑ ω ij k ( f ij + f j k − f ik ) = W ( c ) . This proves the identity for smooth forms directly, not by polynomial approximation. On a face, terms involving an absent vertex vanish because its barycentric coordinate and differential restrict to zero. All remaining radial operators stay within the face. This proves naturality. A 2-form restricts to zero on an edge, so its D D D -primitive does too. Finite compact-interval integrations preserve the stated joint smoothness. □ \square □
Lemma A.4 (Closed extension with fixed triangle integrals). Let n ≥ 2 n\ge2 n ≥ 2 , let c c c be a smooth parameter family of simplicial 2-cocycles on Δ n \Delta^n Δ n , and let β ∂ \beta_{\partial} β ∂ be compatible closed boundary 2-forms with integrals c c c on the boundary triangles. There is a jointly smooth closed 2-form β \beta β on the simplex with this boundary and all triangle integrals equal to $c.
Proof. On each face apply Lemma A.3 to γ ∂ = β ∂ − W ( c ) ∣ ∂ \gamma_{\partial}=\beta_{\partial}-W(c)|_{\partial} γ ∂ = β ∂ − W ( c ) ∣ ∂ . Its triangle integrals vanish, so
η ∂ = D γ ∂ , d η ∂ = γ ∂ . \eta_{\partial}=D\gamma_{\partial},\qquad d\eta_{\partial}=\gamma_{\partial}. η ∂ = D γ ∂ , d η ∂ = γ ∂ .
The resulting 1-forms agree on intersections and vanish on every edge. Set
β = W ( c ) + d Ext n ( η ∂ ) . \beta=W(c)+d\operatorname{Ext}_{n}(\eta_{\partial}). β = W ( c ) + d Ext n ( η ∂ ) .
It is closed because δ c = 0 \delta c=0 δ c = 0 , and its restriction to each face is β ∂ \beta_{\partial} β ∂ . Every edge lies in the boundary when n ≥ 2 n\geq2 n ≥ 2 , so Stokes makes the exact correction have zero integral on every triangle. When n = 2 n=2 n = 2 , the boundary 2-forms and η ∂ \eta_{\partial} η ∂ are zero, and β = W ( c ) \beta=W(c) β = W ( c ) has the prescribed unrestricted top-triangle integral. For n = 3 n=3 n = 3 , δ c = 0 \delta c=0 δ c = 0 is exactly the tetrahedral compatibility. All operations have the parameter properties already proved. □
All relative boundary lifts
Proof of Theorem 3.2. Fix U U U and a relative boundary problem over k ∈ C ∞ ( U , K n ) k\in C^{\infty}(U,K_n) k ∈ C ∞ ( U , K n ) . Thus compatible full MC fields are prescribed on all faces, and their images are the faces of k k k . In degree zero Φ 0 \Phi_0 Φ 0 is the identity. In degree one the target specifies m 0 , a 1 m_0,a_1 m 0 , a 1 and m 1 = a 1 m 0 m_1=a_1m_0 m 1 = a 1 m 0 . Using ψ ( a , b ) = ( log a , b ) \psi(a,b)=(\log a,b) ψ ( a , b ) = ( log a , b ) , take
g ( t ) = ψ − 1 ( t ψ ( a 1 ) ) , σ = g m 0 , A = d g g − 1 , β = 0. g(t)=\psi^{-1}\bigl(t\psi(a_1)\bigr),\qquad\sigma=gm_0,\qquad A=dg\,g^{-1},\qquad\beta=0. g ( t ) = ψ − 1 ( t ψ ( a 1 ) ) , σ = g m 0 , A = d g g − 1 , β = 0.
This has the required endpoints and holonomy.
Suppose n ≥ 2 n\geq2 n ≥ 2 . Set c i j k = − h i j k c_{ijk}=-h_{ijk} c ij k = − h ij k from the target. Its tetrahedral equations say δ c = 0 \delta c=0 δ c = 0 , and boundary compatibility gives the required integrals of β ∂ \beta_{\partial} β ∂ . Lemma A.4 supplies β \beta β . Put u = H v 0 β u=H_{v_0}\beta u = H v 0 β ; then d u = β du=\beta d u = β and u ∣ [ 0 i ] = 0 u|_{[0i]}=0 u ∣ [ 0 i ] = 0 .
On a face F i F_i F i , the prescribed fields give the flat connection
A i ′ = A i − Z σ i u ∣ F i A'_i=A_i-Z_{\sigma_i}u|_{F_i} A i ′ = A i − Z σ i u ∣ F i
with the same base equation. These connections agree on intersections. For i ≠ 0 i\ne0 i = 0 , use the potential g i g_i g i based at v 0 v_0 v 0 with value 1, from Lemma A.1. Then σ i = g i m 0 \sigma_i=g_i m_0 σ i = g i m 0 , and g i ( v r ) = a r g_i(v_r)=a_r g i ( v r ) = a r whenever that vertex lies in F i F_i F i , since the root-edge correction is zero. The potentials agree on intersections containing v 0 v_0 v 0 .
On the opposite face F 0 F_0 F 0 , take the potential g 0 g_0 g 0 with initial value a 1 a_1 a 1 at v 1 v_1 v 1 . It satisfies σ 0 = g 0 m 0 \sigma_0=g_0m_0 σ 0 = g 0 m 0 . It agrees with every g i g_i g i , i ≠ 0 , 1 i\ne0,1 i = 0 , 1 , on their intersection, by their common value at v 1 v_1 v 1 and uniqueness of a flat potential. For F 0 ∩ F 1 F_0\cap F_1 F 0 ∩ F 1 , which contains v 2 v_2 v 2 even when n = 2 n=2 n = 2 , the holonomy formula gives
g 12 = g 0 ( v 2 ) exp ( Z m 0 ∫ [ 12 ] u ) a 1 − 1 . g_{12}=g_0(v_2)\exp\left(Z_{m_0}\int_{[12]}u\right)a_1^{-1}. g 12 = g 0 ( v 2 ) exp ( Z m 0 ∫ [ 12 ] u ) a 1 − 1 .
Stokes and the zero root-edge restrictions give ∫ [ 12 ] u = − h 012 \int_{[12]}u=-h_{012} ∫ [ 12 ] u = − h 012 . Hence
g 0 ( v 2 ) = g 12 a 1 exp ( h 012 Z m 0 ) = a 2 g_0(v_2)=g_{12}a_1\exp(h_{012}Z_{m_0})=a_2 g 0 ( v 2 ) = g 12 a 1 exp ( h 012 Z m 0 ) = a 2
by the target triangle convention. Uniqueness on the connected intersection now gives g 0 = g 1 g_0=g_1 g 0 = g 1 there. This constructs a compatible jointly smooth boundary potential g ∂ g_{\partial} g ∂ . Extend ψ ( g ∂ ) \psi(g_{\partial}) ψ ( g ∂ ) componentwise by Lemma A.2, and apply ψ − 1 \psi^{-1} ψ − 1 to obtain g g g on the full simplex. This step uses the global chart G ≅ R 2 G \cong\mathbb{R}^{2} G ≅ R 2 , not an extension assertion for an arbitrary Lie group. Define
σ = g m 0 , A ′ = d g g − 1 , A = A ′ + Z σ u . \sigma= gm_{0}, \qquad A^{\prime} = dg\,g^{-1}, \qquad A = A^{\prime} + Z_{\sigma}u. σ = g m 0 , A ′ = d g g − 1 , A = A ′ + Z σ u .
The form A ′ A^{\prime} A ′ is flat, d σ = A ′ σ d\sigma= A^{\prime}\sigma d σ = A ′ σ , and equivariance gives d Z σ = A ′ Z σ − Z σ A ′ dZ_{\sigma} = A^{\prime}Z_{\sigma} - Z_{\sigma}A^{\prime} d Z σ = A ′ Z σ − Z σ A ′ . Since Z σ σ = 0 Z_{\sigma}\sigma= 0 Z σ σ = 0 , the reverse flat-correction calculation yields
d σ = A σ , d A − A ∧ A = Z σ β . d\sigma= A\sigma, \qquad dA - A \wedge A = Z_{\sigma}\beta. d σ = A σ , d A − A ∧ A = Z σ β .
Together with d β = 0 d\beta= 0 d β = 0 , these are all of (8) . On each face the potential, base map, matrix form and 2-form recover the prescribed data exactly.
The root holonomies are a i a_i a i , since u u u vanishes on the root edges and g ( v 0 ) = 1 g(v_0) = 1 g ( v 0 ) = 1 . All triangle integrals are − h i j k -h_{ijk} − h ij k . The root coordinates (7) therefore show that the image is exactly k k k . We have solved every relative boundary problem, naturally in the external parameter. The right lifting property against ∂ Δ [ n ] ↪ Δ [ n ] \partial\Delta[n] \hookrightarrow\Delta[n] ∂ Δ [ n ] ↪ Δ [ n ] for all n ≥ 0 n \ge0 n ≥ 0 is precisely the trivial Kan fibration criterion in simplicial sets. The same smooth formulas apply across the rank-drop locus. □
Monodromy and the fixed-object source
Corollary A.5. The comparison gives an equivalence under y M yM y M
Mon s h ( E a f f ) ≃ ∣ y K ∣ \operatorname{Mon}^{\mathrm{sh}}(E_{\mathrm{aff}}) \simeq|yK| Mon sh ( E aff ) ≃ ∣ y K ∣ and the corresponding hypercomplete equivalence. These affine monodromy objects are 2-truncated smooth infinity-sheaves and hence hypercomplete. Their associated derived fixed-object Čech groupoid objects are equivalent.
Proof. Objectwise weak equivalence gives equivalence of presheaves of spaces and therefore of their open-cover sheafifications. Vertex preservation makes this an equivalence under y M yM y M . For every U U U , K ( U ) K(U) K ( U ) is the nerve of a strict 2-groupoid. Its realization is a homotopy 2-type, so the presheaf of spaces is 2-truncated. Left-exact sheafification preserves that property, and truncated infinity-sheaves are hypercomplete. Finally, the Čech levels are finite derived fiber products over these equivalent realizations with the common object y M yM y M . Their comparison is compatible with every simplicial operator. □
The derived Čech groupoid in the corollary is not the raw Duskin nerve K K K . The result compares monodromy sources. A refined transverse-holonomy comparison would also have to specify its target, slice and transport. Likewise, 2-truncation here is a statement in the smooth infinity-topos, not about the ordinary topological shape of a smooth stack.
Corollary A.6. Every other global geometric resolution of this same affine foliation satisfying the hypotheses of § 2.1 has equivalent full smooth integration and the same finite monodromy model K K K . In either smooth infinity-target these comparisons form a coherent diagram, with a contractible space of choices after framing at the specified affine comparison.
Proof. Apply Theorem 2.2 to F ( M ) \mathcal{F}(M) F ( M ) and use Corollary 2.4, with E a f f E_{\mathrm{aff}} E aff as the selected resolution. Compose its constant value with the equivalence induced by Φ \Phi Φ . At the presheaf level, the contractible Kan comparison spaces supply inverse vertices and polynomial Hom paths for each actual comparison. Theorem 2.3 turns them into simplicial homotopies. Thus these maps are also objectwise homotopy equivalences, independently of the general derived-source identification. □
An objectwise trivial Kan fibration does not by itself supply smooth simplicial inverse data. We now construct such data for the actual map Φ \Phi Φ . Local smooth simplicial deformation retractions are already available in [ŠŠ20, Theorems 8.1–8.2]. The assertion here concerns the global affine comparison and its specified crossed-module target.
For fixed n n n , jointly smooth simplex-direction forms constitute a sheaf of real vector spaces
U ⟼ Ω q ( U × Δ n / U ) . U \longmapsto\Omega^q(U \times\Delta^n/U). U ⟼ Ω q ( U × Δ n / U ) .
Data parametrized by an arbitrary smooth sheaf P P P mean morphisms from P P P to this vector sheaf, defined on plots. The sheaf P P P may be nonrepresentable; the data are sheaf morphisms rather than elements of a function algebra on P P P .
Lemma A.7 (Ordered split-locus extension). Let P 1 , … , P N ⊆ P P_1,\ldots,P_N \subseteq P P 1 , … , P N ⊆ P be split subsheaves with retractions r j : P → P j r_j:P \to P_j r j : P → P j . Suppose
r j ( P i ) ⊆ P i ∩ P j ( i < j ) . r_j(P_i) \subseteq P_i \cap P_j \qquad(i<j). r j ( P i ) ⊆ P i ∩ P j ( i < j ) . Compatible morphisms v j : P j → V v_j:P_j \to V v j : P j → V to a sheaf of real vector spaces extend to v : P → V v:P \to V v : P → V . The same holds inside any specified linear subsheaf of V V V .
Proof. Starting with v ~ 0 = 0 \widetilde v_0=0 v 0 = 0 , set
v ~ j = v ~ j − 1 + ( v j − v ~ j − 1 ∣ P j ) r j . \widetilde v_j=\widetilde v_{j-1}+\left(v_j-\widetilde v_{j-1}|_{P_j}\right)r_j. v j = v j − 1 + ( v j − v j − 1 ∣ P j ) r j .
The residual in parentheses vanishes on each earlier intersection P i ∩ P j P_i\cap P_j P i ∩ P j . Its pullback therefore vanishes on P i P_i P i , while r j ∣ P j = Id r_j|_{P_j}=\operatorname{Id} r j ∣ P j = Id gives the desired new restriction. All operations preserve a linear subsheaf. This also covers N = 0 N=0 N = 0 , and does not require commuting retractions. □
Lemma A.8 (Affine lifting with split parameter loci). Put S = Int ( E a f f ) S=\operatorname{Int}(E_{\mathrm{aff}}) S = Int ( E aff ) and T = y K T=yK T = y K . Let k : P → T n k:P\to T_n k : P → T n have compatible prescribed faces b i : P → S n − 1 b_i:P\to S_{n-1} b i : P → S n − 1 over its faces, when n > 0 n>0 n > 0 . Suppose lifts ℓ j : P j → S n \ell_j:P_j\to S_n ℓ j : P j → S n are given on finitely many split loci as in Lemma A.7, agreeing with these faces and with one another on intersections. Then there is a lift ℓ : P → S n \ell:P\to S_n ℓ : P → S n with
Φ n ℓ = k , d i ℓ = b i , ℓ ∣ P j = ℓ j . \Phi_n\ell=k,\qquad d_i\ell=b_i,\qquad\ell|_{P_j}=\ell_j. Φ n ℓ = k , d i ℓ = b i , ℓ ∣ P j = ℓ j . The retractions need not preserve k k k .
Proof. The face condition means d a b i = d i − 1 b a d_ab_i=d_{i-1}b_a d a b i = d i − 1 b a for a < i a<i a < i and Φ n − 1 b i = d i k \Phi_{n-1}b_i=d_i k Φ n − 1 b i = d i k . Thus all data are morphisms of sheaves of the indicated degrees. We extend only vector-valued residuals, using baselines constructed from the current target k k k .
For n ≥ 2 n\geq2 n ≥ 2 , put c = − h ( k ) c=-h(k) c = − h ( k ) and let β ∂ \beta_{\partial} β ∂ be the prescribed boundary 2 2 2 -forms. Lemma A.4 gives the baseline
β 0 = W ( c ) + d Ext n ( D ∂ ( β ∂ − W ( c ) ∣ ∂ ) ) . \beta^0=W(c)+d\operatorname{Ext}_n\left(D_{\partial}\left(\beta_{\partial}-W(c)|_{\partial}\right)\right). β 0 = W ( c ) + d Ext n ( D ∂ ( β ∂ − W ( c ) ∣ ∂ ) ) .
It has the prescribed boundary and triangle periods. If β j \beta_j β j is the 2-form of ℓ j \ell_j ℓ j , then
γ j = β j − β 0 ∣ P j , η j = D n γ j \gamma_j=\beta_j-\beta^0|_{P_j},\qquad\eta_j=D_n\gamma_j γ j = β j − β 0 ∣ P j , η j = D n γ j
satisfy d η j = γ j d\eta_j=\gamma_j d η j = γ j : the closed form γ j \gamma_j γ j has zero triangle periods. Face naturality of D n D_n D n makes η j \eta_j η j vanish on the whole boundary, since γ j ∣ ∂ = 0 \gamma_j|_{\partial}=0 γ j ∣ ∂ = 0 . The η j \eta_j η j agree on parameter-locus intersections. Apply Lemma A.7 in the linear sheaf of 1-forms vanishing on ∂ Δ n \partial\Delta^n ∂ Δ n , and set
β = β 0 + d η . \beta=\beta^0+d\eta. β = β 0 + d η .
This recovers every β j \beta_j β j and the full boundary. Every edge lies in the boundary for n ≥ 2 n\ge2 n ≥ 2 , so Stokes shows that the correction has zero triangle periods. In particular the current periods c ( k ) c(k) c ( k ) have not been pulled back along any r j r_j r j .
Put u = H v 0 β u=H_{v_0}\beta u = H v 0 β . Build the compatible boundary potential g ∂ g_{\partial} g ∂ exactly as in the preceding all-boundary proof, using this u u u and the current root values a i ( k ) a_i(k) a i ( k ) . The opposite face uses the potential with value a 1 ( k ) a_1(k) a 1 ( k ) at v 1 v_1 v 1 ; its compatibility is the same plus-h h h triangle calculation. Let g 0 g^0 g 0 be the extension of g ∂ g_{\partial} g ∂ through the global chart ψ ( a , b ) = ( log a , b ) \psi(a,b)=(\log a,b) ψ ( a , b ) = ( log a , b ) .
On P j P_j P j , flatten the prescribed simplex with u j = H v 0 β j = u ∣ P j u_j=H_{v_0}\beta_j=u|_{P_j} u j = H v 0 β j = u ∣ P j , and let g j g_j g j be its potential normalized by g j ( v 0 ) = 1 g_j(v_0)=1 g j ( v 0 ) = 1 . It agrees with g ∂ g_{\partial} g ∂ on all faces: on a face containing v 0 v_0 v 0 the flat equation and initial value agree; on the opposite face its value at v 1 v_1 v 1 is a 1 ( k ) a_1(k) a 1 ( k ) , because u j u_j u j vanishes on the root edge. Uniqueness of the flat potential gives the required agreement. Consequently
w j = ψ ( g j ) − ψ ( g 0 ) ∣ P j w_j=\psi(g_j)-\psi(g^0)|_{P_j} w j = ψ ( g j ) − ψ ( g 0 ) ∣ P j
is a compatible family of R 2 \mathbb{R}^2 R 2 -valued functions vanishing on the simplex boundary. Extend it within that linear subsheaf and put
g = ψ − 1 ( ψ ( g 0 ) + w ) , σ = g m 0 ( k ) , A = d g g − 1 + Z σ u . g=\psi^{-1}\bigl(\psi(g^0)+w\bigr),\qquad\sigma=gm_0(k),\qquad A=dg\,g^{-1}+Z_\sigma u. g = ψ − 1 ( ψ ( g 0 ) + w ) , σ = g m 0 ( k ) , A = d g g − 1 + Z σ u .
There is no range constraint on the chart-valued sum because ψ \psi ψ is global.
For completeness, the modified lift still solves the entire problem. The form A ′ = d g g − 1 A'=dg\,g^{-1} A ′ = d g g − 1 is flat, d σ = A ′ σ d\sigma=A'\sigma d σ = A ′ σ , and d Z σ = [ A ′ , Z σ ] dZ_\sigma=[A',Z_\sigma] d Z σ = [ A ′ , Z σ ] . With d u = β du=\beta d u = β , d β = 0 d\beta=0 d β = 0 and Z σ σ = 0 Z_\sigma\sigma=0 Z σ σ = 0 , reverse flat correction gives all of (8) . The forms and potential recover every prescribed face. Since w ∣ ∂ = 0 w|_{\partial}=0 w ∣ ∂ = 0 , g ( v 0 ) = 1 g(v_0)=1 g ( v 0 ) = 1 and g ( v i ) = a i ( k ) g(v_i)=a_i(k) g ( v i ) = a i ( k ) . The root-edge restrictions of u u u vanish, so (A.2) gives the current root holonomies a i ( k ) a_i(k) a i ( k ) . The triangle periods are − h i j k ( k ) -h_{ijk}(k) − h ij k ( k ) . The root coordinates (7) therefore give Φ n ℓ = k \Phi_n\ell=k Φ n ℓ = k exactly. On P j P_j P j we have β = β j \beta=\beta_j β = β j , u = u j u=u_j u = u j , g = g j g=g_j g = g j , and σ = g j m 0 ( k ∣ P j ) = σ j \sigma=g_jm_0(k|_{P_j})=\sigma_j σ = g j m 0 ( k ∣ P j ) = σ j . The matrix form is then A j A_j A j , so the whole prescribed simplex is recovered.
For n = 1 n=1 n = 1 , β = u = 0 \beta=u=0 β = u = 0 . Take
g 0 ( t ) = ψ − 1 ( t ψ ( a 1 ( k ) ) ) . g^0(t)=\psi^{-1}\bigl(t\psi(a_1(k))\bigr). g 0 ( t ) = ψ − 1 ( t ψ ( a 1 ( k )) ) .
Here ψ ( 1 ) = 0 \psi(1)=0 ψ ( 1 ) = 0 , so the endpoints are 1 and a 1 ( k ) a_1(k) a 1 ( k ) . Each prescribed simplex has a normalized flat potential g j g_j g j . Extend ψ ( g j ) − ψ ( g 0 ) ∣ P j \psi(g_j)-\psi(g^0)|_{P_j} ψ ( g j ) − ψ ( g 0 ) ∣ P j in the vector sheaf of functions vanishing at both endpoints, and reconstruct as above. For n = 0 n=0 n = 0 the lift is forced by Φ 0 = Id \Phi_0=\operatorname{Id} Φ 0 = Id .
All operations are natural on external plots: fixed pullbacks, finite linear operations, compact-fiber integrations, and the displayed smooth group formulas. These formulas are defined uniformly across the rank-drop locus.
Lemma A.9 (Ordered degeneracy loci). For a simplicial sheaf Z Z Z , put B i = im ( s i : Z n − 1 → Z n ) B_i = \operatorname{im}(s_i : Z_{n-1} \to Z_n) B i = im ( s i : Z n − 1 → Z n ) and r i = s i d i + 1 r_i = s_i d_{i+1} r i = s i d i + 1 . These are split loci satisfying r i ( B j ) ⊆ B i ∩ B j r_i(B_j) \subseteq B_i \cap B_j r i ( B j ) ⊆ B i ∩ B j for j < i j < i j < i . If i < j i < j i < j and σ ∈ B i ∩ B j \sigma\in B_i \cap B_j σ ∈ B i ∩ B j , then
σ = s i s j − 1 z = s j s i z , z = d i d j σ . \sigma= s_i s_{j-1}z = s_j s_i z,\qquad z = d_i d_j \sigma. σ = s i s j − 1 z = s j s i z , z = d i d j σ . Proof. The identity d i + 1 s i = Id d_{i+1}s_i = \operatorname{Id} d i + 1 s i = Id gives the splitting. For j < i j < i j < i ,
r i s j = s i d i + 1 s j = s i s j d i = s j s i − 1 d i . r_i s_j = s_i d_{i+1}s_j = s_i s_j d_i = s_j s_{i-1}d_i. r i s j = s i d i + 1 s j = s i s j d i = s j s i − 1 d i . This proves preservation of the earlier locus. For the intersection formula write σ = s j b \sigma= s_jb σ = s j b with b = d j σ b = d_j\sigma b = d j σ . Then d i σ = s j − 1 d i b d_i\sigma= s_{j-1}d_i b d i σ = s j − 1 d i b , and σ = s i d i σ \sigma= s_i d_i\sigma σ = s i d i σ because σ ∈ B i \sigma\in B_i σ ∈ B i . The two displayed double degeneracies agree by the simplicial identity. □
Theorem A.10 (Global affine simplicial retraction). The comparison Φ : S → T \Phi: S \to T Φ : S → T admits a smooth simplicial section Ψ : T → S \Psi: T \to S Ψ : T → S and a simplicial homotopy
H : S × Δ [ 1 ] ⟶ S , H ( − , 0 ) = Ψ Φ , H ( − , 1 ) = Id S . H : S \times\Delta[1] \longrightarrow S,\qquad H(-,0) = \Psi\Phi,\qquad H(-,1) = \operatorname{Id}_S. H : S × Δ [ 1 ] ⟶ S , H ( − , 0 ) = ΨΦ , H ( − , 1 ) = Id S . Moreover Φ H = Φ pr S \Phi H = \Phi\operatorname{pr}_S Φ H = Φ pr S and H H H is fixed on Ψ ( T ) \Psi(T) Ψ ( T ) . In particular it is relative to c ( y M ) c(y_M) c ( y M ) . Thus the specified affine map is a global smooth simplicial strong deformation retraction.
Proof. First construct the section by induction. In degree zero set Ψ 0 = Id M \Psi_0 = \operatorname{Id}_M Ψ 0 = Id M . At degree n > 0 n > 0 n > 0 , take parameter sheaf T n T_n T n , current target k = Id T n k = \operatorname{Id}_{T_n} k = Id T n , and boundary b a = Ψ n − 1 d a b_a = \Psi_{n-1}d_a b a = Ψ n − 1 d a . On B i = im ( s i ) ⊂ T n B_i = \operatorname{im}(s_i) \subset T_n B i = im ( s i ) ⊂ T n prescribe
ℓ i ( k ) = s i Ψ n − 1 ( d i + 1 k ) . \ell_i(k) = s_i\Psi_{n-1}(d_{i+1}k). ℓ i ( k ) = s i Ψ n − 1 ( d i + 1 k ) .
The face identities d a s i = s i − 1 d a d_a s_i = s_{i-1}d_a d a s i = s i − 1 d a for a < i a < i a < i , d i s i = d i + 1 s i = Id d_i s_i = d_{i+1}s_i = \operatorname{Id} d i s i = d i + 1 s i = Id , and d a s i = s i d a − 1 d_a s_i = s_i d_{a-1} d a s i = s i d a − 1 for a > i + 1 a > i+1 a > i + 1 show that these prescriptions have boundary b a ∣ B i b_a|_{B_i} b a ∣ B i . Their target is k ∣ B i k|_{B_i} k ∣ B i by the lower-degree section identities. On B i ∩ B j B_i \cap B_j B i ∩ B j , i < j i < j i < j , use Lemma A.9: the two values are
s i s j − 1 Ψ n − 2 ( z ) = s j s i Ψ n − 2 ( z ) . s_i s_{j-1}\Psi_{n-2}(z) = s_j s_i\Psi_{n-2}(z). s i s j − 1 Ψ n − 2 ( z ) = s j s i Ψ n − 2 ( z ) .
The same lemma orders the split loci. Lemma A.8 gives Ψ n \Psi_n Ψ n , with all faces and degeneracies prescribed. This constructs a simplicial section, not just levelwise sections.
Now construct H n ( − , τ ) H_n(-,\tau) H n ( − , τ ) inductively for each label τ ∈ Δ [ 1 ] n \tau\in\Delta[1]_n τ ∈ Δ [ 1 ] n . Constant labels give the two stated endpoint maps. For a nonconstant label take P = S n P = S_n P = S n , k = Φ n k = \Phi_n k = Φ n , and boundary
b a ( σ ) = H n − 1 ( d a σ , d a τ ) . b_a(\sigma) = H_{n-1}(d_a\sigma,d_a\tau). b a ( σ ) = H n − 1 ( d a σ , d a τ ) .
The lower homotopy identities make this boundary compatible and over k k k . First impose the fixed locus B ∗ = Ψ ( T n ) B_* = \Psi(T_n) B ∗ = Ψ ( T n ) , with retraction Ψ Φ \Psi\Phi ΨΦ and prescribed value σ \sigma σ there. For every i i i with τ ( i ) = τ ( i + 1 ) \tau(i) = \tau(i+1) τ ( i ) = τ ( i + 1 ) , also impose B i = im ( s i ) B_i = \operatorname{im}(s_i) B i = im ( s i ) and prescribe
ℓ i ( σ ) = s i H n − 1 ( d i + 1 σ , d i + 1 τ ) . \ell_i(\sigma) = s_iH_{n-1}(d_{i+1}\sigma,d_{i+1}\tau). ℓ i ( σ ) = s i H n − 1 ( d i + 1 σ , d i + 1 τ ) .
The three face–degeneracy identities above, applied to both σ \sigma σ and its label, give the stated boundary. The target is s i d i + 1 k = k s_i d_{i+1}k = k s i d i + 1 k = k on B i B_i B i .
Here is the intersection check, including the labels. Write ς i \varsigma_i ς i for the order-preserving codegeneracy and, on B i ∩ B j B_i \cap B_j B i ∩ B j with i < j i < j i < j , write σ = s i s j − 1 z = s j s i z \sigma= s_i s_{j-1}z = s_j s_i z σ = s i s j − 1 z = s j s i z . The equal adjacent values of τ \tau τ let it factor through the common surjection
ς j − 1 ς i = ς i ς j . \varsigma_{j-1}\varsigma_i = \varsigma_i\varsigma_j. ς j − 1 ς i = ς i ς j .
The two reduced labels coincide: their pullbacks along this surjection are τ \tau τ , and the surjection is onto. Calling this label τ ^ \widehat{\tau} τ , the two prescribed values are
s i s j − 1 H n − 2 ( z , τ ^ ) = s j s i H n − 2 ( z , τ ^ ) . s_i s_{j-1}H_{n-2}(z,\widehat{\tau})=s_j s_iH_{n-2}(z,\widehat{\tau}). s i s j − 1 H n − 2 ( z , τ ) = s j s i H n − 2 ( z , τ ) .
On B ∗ ∩ B i B_* \cap B_i B ∗ ∩ B i , if Ψ ( k ) = s i σ ′ \Psi(k)=s_i\sigma' Ψ ( k ) = s i σ ′ , then k = s i Φ ( σ ′ ) k=s_i\Phi(\sigma') k = s i Φ ( σ ′ ) and simpliciality of Ψ \Psi Ψ , followed by the splitting of s i s_i s i , gives σ ′ = Ψ Φ ( σ ′ ) \sigma'=\Psi\Phi(\sigma') σ ′ = ΨΦ ( σ ′ ) . The lower homotopy fixes this lower simplex, so both prescriptions return Ψ ( k ) \Psi(k) Ψ ( k ) . Boundary compatibility on B ∗ B_* B ∗ follows from the same lower fixed-locus condition.
Every r i = s i d i + 1 r_i=s_i d_{i+1} r i = s i d i + 1 preserves B ∗ B_* B ∗ because Ψ \Psi Ψ and Φ \Phi Φ are simplicial. It also preserves all earlier degeneracy loci. Thus order B ∗ B_* B ∗ first, followed by the relevant B i B_i B i increasingly, and apply Lemma A.8. The resulting H n H_n H n has all required faces and degeneracies of the product. Labels on a test manifold are locally constant; the constructions on constant labels glue, including on disconnected test manifolds.
In degree one the sole nonconstant label is ( 0 , 1 ) (0,1) ( 0 , 1 ) , so only the fixed locus is imposed. In degree two the nonconstant labels ( 0 , 0 , 1 ) (0,0,1) ( 0 , 0 , 1 ) and ( 0 , 1 , 1 ) (0,1,1) ( 0 , 1 , 1 ) impose B 0 B_0 B 0 and B 1 B_1 B 1 , respectively. Their face conditions are precisely the degree-one conditions already constructed; the opposite-face calculation is included in the lifting lemma. All formulas include the origin and y = 0 y=0 y = 0 . Finally, Ψ \Psi Ψ preserves degeneracies and is the identity in degree zero, so c ( y M ) ⊆ Ψ ( T ) c(yM)\subseteq\Psi(T) c ( y M ) ⊆ Ψ ( T ) and H H H fixes it. □
§ 5 extracts the raw coefficient consequence from these actual inverse data, not from weak equivalence alone. This particular whole-plane section and homotopy can move the interior base simplex and cannot simply be restricted to arbitrary opens. The construction of § 7 instead uses range-preserving charts on admissible arrow domains, chosen separately for each open. The global chart G ≅ R 2 G\cong\mathbb{R}^2 G ≅ R 2 is essential to this proof. For other simply connected Lie groups, an analogous construction would require a replacement for this chart-based extension.
Appendix B. Regular reductions and scalar parameters
This appendix proves the regular-reduction and scalar-parameter results used by the normal forms and the vertical-family theorem. It also gives the regular Hom comparison by that route. The nonregular proof in Sections 9 and 10 does not use the cohomology splitting made here. Throughout this appendix, E = E a f f E=E_{\mathrm{aff}} E = E aff , B = B E B=B_E B = B E , G = Aff + ( 1 ) G=\operatorname{Aff}_+(1) G = Aff + ( 1 ) , and K K K is the crossed-module nerve of Section 3. Set
R K = Rep s m p e r f ( y K ) , IN K = Ψ ∗ IN E . \mathcal{R}_K=\operatorname{Rep}^{\mathrm{perf}}_{\mathrm{sm}}(yK),\qquad\operatorname{IN}_K=\Psi^*\operatorname{IN}_E. R K = Rep sm perf ( y K ) , IN K = Ψ ∗ IN E .
The base is the whole R 2 \mathbb{R}^2 R 2 , including the origin. The section Ψ \Psi Ψ is the whole-plane choice; ordinary opens use the range-preserving sections of Section 7.
Theorem B.1 (Global affine Hom comparison). For every P , Q ∈ Coh r e g ( B ) P,Q\in\operatorname{Coh}^{\mathrm{reg}}(B) P , Q ∈ Coh reg ( B ) , the actual first Taylor map
IN 1 K : Hom Coh ( B ) ( P , Q ) ⟶ Hom R K ( IN K P , IN K Q ) \operatorname{IN}^{K}_1:\operatorname{Hom}_{\operatorname{Coh}(B)}(P,Q)\longrightarrow\operatorname{Hom}_{\mathcal{R}_K}(\operatorname{IN}^{K}P,\operatorname{IN}^{K}Q) IN 1 K : Hom Coh ( B ) ( P , Q ) ⟶ Hom R K ( IN K P , IN K Q ) is a quasi-isomorphism in every integer degree. The same Hom assertion holds for IN E \operatorname{IN}_E IN E on the full MC target.
The proof has two parts. A finite regular coefficient reduction makes a scalar comparison sufficient. An explicit relative de Rham calculation then proves that scalar comparison for the actual nerve. The coefficient reduction is what makes the scalar comparison determine all Hom degrees.
Finite regular minimalization
The following elementary perturbation statement also applies to an associative coefficient DGA whose degree-zero algebra is not central. This distinction will be useful for normalized simplicial cochains.
Lemma B.2 (Finite superconnection perturbation). Let D \mathcal{D} D be a nonnegative cochain DGA with D 0 = A = C ∞ ( M ) \mathcal{D}^{0}=A=C^{\infty}(M) D 0 = A = C ∞ ( M ) , and use the right-module convention of Section 2.3. Let ( V ⊗ A D , D ) (V\otimes_A\mathcal{D},D) ( V ⊗ A D , D ) have a flat right superconnection D D D , bounded finite-projective coefficient module V V V , and regular internal differential d 0 = D [ 0 ] d_0=D^{[0]} d 0 = D [ 0 ] . It is cohesively homotopy equivalent to a module ( H ⊗ A D , D H ) (H\otimes_A\mathcal{D},D_H) ( H ⊗ A D , D H ) with D H [ 0 ] = 0 D_H^{[0]}=0 D H [ 0 ] = 0 , where H = H ∙ ( V , d 0 ) H=H^\bullet(V,d_0) H = H ∙ ( V , d 0 ) is a bounded finite-projective graded module. The displayed formulas are finite and define smooth maps and homotopies on the resulting globally bounded finite-projective bundles. If the differential and the chosen contraction data vary jointly smoothly with auxiliary finite-dimensional parameters and with a fixed global degree bound, the same is true of the transferred formulas. For a smaller coefficient class, this conclusion requires preservation by the chosen contraction maps and by all operations used in the formulas.
Proof. Boundedness and locally constant cohomology ranks imply locally constant ranks of the differentials, by induction from either end of the finite complex. Thus kernels and images are smooth subbundles. A smooth fiber metric gives a decomposition V k = B k ⊕ H k ⊕ C k V^k=B^k\oplus H^k\oplus C^k V k = B k ⊕ H k ⊕ C k , where B k = im d 0 B^k=\operatorname{im}d_0 B k = im d 0 and d 0 : C k → B k + 1 d_0:C^k\to B^{k+1} d 0 : C k → B k + 1 is an isomorphism. Take i , p i,p i , p to be inclusion and projection of H H H , and let h h h be this inverse on B B B , zero on H ⊕ C H\oplus C H ⊕ C . This gives
( H , 0 ) ⇄ p i ( V , d 0 ) , p i = 1 , d 0 h + h d 0 = 1 − i p , h 2 = h i = p h = 0. (H,0)\mathrel{\mathop{\rightleftarrows}^{i}_{p}}(V,d_0),\qquad pi=1,\qquad d_0h+hd_0=1-ip,\qquad h^2=hi=ph=0. ( H , 0 ) ⇄ p i ( V , d 0 ) , p i = 1 , d 0 h + h d 0 = 1 − i p , h 2 = hi = p h = 0.
In particular, H H H is a direct summand of V V V and is finite projective. Extend i , p , h i,p,h i , p , h right D \mathcal{D} D -linearly.
Put Δ = D − d 0 \Delta=D-d_0 Δ = D − d 0 . A component of positive D \mathcal{D} D -degree r r r changes internal coefficient degree by 1 − r 1-r 1 − r . Consequently both h Δ h\Delta h Δ and Δ h \Delta h Δ h strictly lower internal degree. This also holds for the scalar-differential term in the Leibniz rule. Their inverse series therefore terminate. Define
Σ = Δ ( 1 + h Δ ) − 1 = ( 1 + Δ h ) − 1 Δ , \Sigma=\Delta(1+h\Delta)^{-1}=(1+\Delta h)^{-1}\Delta, Σ = Δ ( 1 + h Δ ) − 1 = ( 1 + Δ h ) − 1 Δ ,
i ′ = i − h Σ i = ( 1 + h Δ ) − 1 i , p ′ = p − p Σ h = p ( 1 + Δ h ) − 1 , h ′ = h − h Σ h = h ( 1 + Δ h ) − 1 , D H = p Σ i = p Δ i . (85) \begin{aligned}
i'&=i-h\Sigma i=(1+h\Delta)^{-1}i, &
p'&=p-p\Sigma h=p(1+\Delta h)^{-1},\\
h'&=h-h\Sigma h=h(1+\Delta h)^{-1}, &
D_H&=p\Sigma i=p\Delta i. \tag*{(85)}
\end{aligned} i ′ h ′ = i − h Σ i = ( 1 + h Δ ) − 1 i , = h − h Σ h = h ( 1 + Δ h ) − 1 , p ′ D H = p − p Σ h = p ( 1 + Δ h ) − 1 , = p Σ i = p Δ i . (85)
The perturbation Δ \Delta Δ is not assumed D \mathcal{D} D -linear. To check right-linearity, observe that the maps ( h Δ ) n i (h\Delta)^n i ( h Δ ) n i are right-linear and are killed by h h h : at the induction step the extra Leibniz term is ( − 1 ) ∣ z ∣ h z d D b = 0 (-1)^{|z|}hz\,d_{\mathcal{D}}b=0 ( − 1 ) ∣ z ∣ h z d D b = 0 . For the other two series, put
f n = p ( Δ h ) n , k n = h ( Δ h ) n . f_n=p(\Delta h)^n,\qquad k_n=h(\Delta h)^n. f n = p ( Δ h ) n , k n = h ( Δ h ) n .
Inductively each is right-linear and f n h = k n h = 0 f_nh=k_nh=0 f n h = k n h = 0 . Indeed, if f n f_n f n has these properties, then
f n Δ h ( z b ) = f n Δ h ( z ) b + ( − 1 ) ∣ z ∣ − 1 f n h ( z ) d D b = f n Δ h ( z ) b , f_n\Delta h(zb)=f_n\Delta h(z)b+(-1)^{|z|-1}f_nh(z)d_{\mathcal{D}}b=f_n\Delta h(z)b, f n Δ h ( z b ) = f n Δ h ( z ) b + ( − 1 ) ∣ z ∣ − 1 f n h ( z ) d D b = f n Δ h ( z ) b ,
and f n Δ h h = 0 f_n\Delta h\,h=0 f n Δ h h = 0 ; the argument for k n k_n k n is identical. Thus i ′ i' i ′ , p ′ p' p ′ , h ′ h' h ′ are right-linear. This proof uses neither commutativity nor centrality of D 0 \mathcal{D}^0 D 0 .
Since p ′ i ′ = p i = 1 p'i'=pi=1 p ′ i ′ = p i = 1 ,
D H ( z b ) = D H ( z ) b + ( − 1 ) ∣ z ∣ z d D b . D_H(zb)=D_H(z)b+(-1)^{|z|}z\,d_{\mathcal{D}}b. D H ( z b ) = D H ( z ) b + ( − 1 ) ∣ z ∣ z d D b .
For completeness, the finite operator identities give
D i ′ = i ′ D H , p ′ D = D H p ′ , p ′ i ′ = 1 , D h ′ + h ′ D = 1 − i ′ p ′ , D H 2 = 0. (86) \begin{aligned}
Di'&=i'D_H, & p'D&=D_Hp', & p'i'&=1,\\
Dh'+h'D&=1-i'p', & D_H^2&=0. \tag*{(86)}
\end{aligned} D i ′ D h ′ + h ′ D = i ′ D H , = 1 − i ′ p ′ , p ′ D D H 2 = D H p ′ , = 0. p ′ i ′ = 1 , (86)
For example, i ′ = i − h Δ i ′ i'=i-h\Delta i' i ′ = i − h Δ i ′ and d 0 Δ = − Δ D d_0\Delta=-\Delta D d 0 Δ = − Δ D imply ( 1 + h Δ ) D i ′ = i D H (1+h\Delta)Di'=iD_H ( 1 + h Δ ) D i ′ = i D H . Similarly p ′ = p − p ′ Δ h p'=p-p'\Delta h p ′ = p − p ′ Δ h and Δ d 0 = − D Δ \Delta d_0=-D\Delta Δ d 0 = − D Δ give p ′ D ( 1 + Δ h ) = D H p p'D(1+\Delta h)=D_Hp p ′ D ( 1 + Δ h ) = D H p ; here D H = p Δ i ′ = p ′ Δ i D_H=p\Delta i'=p'\Delta i D H = p Δ i ′ = p ′ Δ i . These identities prove the two chain-map assertions. The side conditions give p ′ i ′ = 1 p'i'=1 p ′ i ′ = 1 directly. To check the homotopy identity, multiply it on the left by 1 + h Δ 1+h\Delta 1 + h Δ and on the right by 1 + Δ h 1+\Delta h 1 + Δ h . Both sides become
1 − i p + Δ h + h Δ + h Δ 2 h , 1-ip+\Delta h+h\Delta+h\Delta^2h, 1 − i p + Δ h + h Δ + h Δ 2 h ,
using D 2 = d 0 2 = 0 D^2=d_0^2=0 D 2 = d 0 2 = 0 . Finally D i ′ = i ′ D H Di'=i'D_H D i ′ = i ′ D H and p ′ i ′ = 1 p'i'=1 p ′ i ′ = 1 give D H 2 = 0 D_H^2=0 D H 2 = 0 . Every term of D H D_H D H has positive D \mathcal{D} D -degree. All expressions are finite in the global coefficient amplitude. The chosen contraction data and the differential are smooth, so the transferred maps, homotopies and superconnection are smooth on the stated finite-projective bundles. □
Recall the actual CE differential:
B = C ∞ ( R 2 ) [ θ 1 , θ 2 , η ] , ∣ θ i ∣ = 1 , ∣ η ∣ = 2 , Q f = f x ( x θ 1 + y θ 2 ) , Q θ 1 = − y η , Q θ 2 = θ 1 θ 2 + x η , Q η = 0. (87) \begin{aligned}
B&=C^\infty(\mathbb{R}^2)[\theta_1,\theta_2,\eta], & |\theta_i|&=1, & |\eta|&=2,\\
Qf&=f_x(x\theta_1+y\theta_2), & Q\theta_1&=-y\eta,\\
Q\theta_2&=\theta_1\theta_2+x\eta, & Q\eta&=0. \tag*{(87)}
\end{aligned} B Q f Q θ 2 = C ∞ ( R 2 ) [ θ 1 , θ 2 , η ] , = f x ( x θ 1 + y θ 2 ) , = θ 1 θ 2 + x η , ∣ θ i ∣ Q θ 1 Q η = 1 , = − y η , = 0. ∣ η ∣ = 2 , (87)
Write ξ = x θ 1 + y θ 2 = Q x \xi=x\theta_1+y\theta_2=Qx ξ = x θ 1 + y θ 2 = Q x .
Lemma B.3. Every regular cohesive B B B -module is cohesively homotopy equivalent to one whose internal differential and CE-degree-one connection term both vanish.
Proof. Apply Lemma B.2. The resulting graded cohomology bundle is globally trivial on R 2 \mathbb{R}^2 R 2 . In a graded frame, write its minimal superconnection as Q + A Q+A Q + A , with
A 1 = θ 1 L 1 + θ 2 L 2 . A_1=\theta_1L_1+\theta_2L_2. A 1 = θ 1 L 1 + θ 2 L 2 . The η \eta η coefficient of the CE-degree-two flatness equation is − y L 1 + x L 2 = 0 -yL_1+xL_2=0 − y L 1 + x L 2 = 0 . Higher connection terms do not contribute to this equation, since the internal differential is zero. The relation gives L 1 ( 0 , y ) = 0 L_1(0,y)=0 L 1 ( 0 , y ) = 0 for y ≠ 0 y\ne0 y = 0 , and then at y = 0 y=0 y = 0 by continuity. The global Hadamard formula
L ( x , y ) = ∫ 0 1 ∂ x L 1 ( t x , y ) d t L(x,y)=\int_0^1\partial_xL_1(tx,y)\,dt L ( x , y ) = ∫ 0 1 ∂ x L 1 ( t x , y ) d t gives L 1 = x L L_1=xL L 1 = xL . The same relation and continuity give L 2 = y L L_2=yL L 2 = y L . Thus A 1 = ξ L A_1=\xi L A 1 = ξ L on the whole plane.
In the forms-first convention, a degree-zero frame change gives A g = g − 1 A g + g − 1 Q g A^g = g^{-1}Ag + g^{-1}Qg A g = g − 1 A g + g − 1 Q g .
Solve
g x = − L g , g ( 0 , y ) = 1. g_x = -Lg,\qquad g(0,y)=1. g x = − Lg , g ( 0 , y ) = 1.
This linear matrix ODE has an invertible solution for every finite x x x , smoothly in y y y . The usual local estimates on compact parameter sets give joint smoothness; no compactness of R 2 \mathbb{R}^2 R 2 is assumed. Since Q g = ξ g x Q_g = \xi g_x Q g = ξ g x , the transformed one-form part vanishes. All remaining connection components strictly lower internal degree. □
Lemma B.4 (Reduction of the actual Hom map). Let
I : ( B , Q ) ⟶ C K ∙ : = C n o r m , s m ∙ ( K ) I:(B,Q)\longrightarrow C_K^\bullet:=C_{\mathrm{norm,sm}}^\bullet(K) I : ( B , Q ) ⟶ C K ∙ := C norm , sm ∙ ( K ) be the scalar first Taylor map, and write δ \delta δ for the full simplicial coboundary on C K ∙ C_K^\bullet C K ∙ throughout this appendix. If I I I is a quasi-isomorphism, then the first assertion of Theorem B.1 holds.
Proof. First replace the two objects using Lemma B.3. An A ∞ A_\infty A ∞ -functor induces a functor on graded cohomology categories: its second Taylor identity identifies the two compositions on cohomology. For the actual functor these are the identities of Theorem 6.1, and strict unitality gives I N 1 ( 1 ) = 1 I N_1(1)=1 I N 1 ( 1 ) = 1 . Thus source homotopy equivalences become target homotopy equivalences, and the Hom assertion can be transferred back after it is proved for these replacements. This transfer uses homotopy equivalences within the regular category.
Choose constant graded frames W P , W Q W_P,W_Q W P , W Q for the replacements and put
T q = ⨁ i Hom R ( W P i , W Q i + q ) . T^q=\bigoplus_i \operatorname{Hom}_{\mathbb{R}}(W_P^i,W_Q^{i+q}). T q = i ⨁ Hom R ( W P i , W Q i + q ) . This is a finite-dimensional constant matrix space, placed in internal degree q q q ; it is not the full cohesive Hom module. Filter both Hom complexes increasingly by matrix degree q ≤ s q\le s q ≤ s . The filtration has finite length, independent of total degree. Minimalization removes the degree-raising internal differential; the subsequent one-form gauge makes the degree-preserving part scalar. These are different roles for the two reductions. In forms-first order its associated-graded complexes are
( B , Q ) ⊗ R T q , ( C K ∙ , δ ) ⊗ R T q . (B,Q)\otimes_{\mathbb{R}}T^q,\qquad(C_K^\bullet,\delta)\otimes_{\mathbb{R}}T^q. ( B , Q ) ⊗ R T q , ( C K ∙ , δ ) ⊗ R T q . The leading differentials are Q ⊗ 1 Q\otimes1 Q ⊗ 1 and δ ⊗ 1 \delta\otimes1 δ ⊗ 1 , with T q T^q T q in the indicated degree. For the target, r 0 = 0 r_0=0 r 0 = 0 and r 1 = 1 r_1=1 r 1 = 1 , so the two endpoint terms complete the inner differential to the full simplicial coboundary. In particular the final endpoint sign is − ( − 1 ) p + q ( − 1 ) q = ( − 1 ) p + 1 -(-1)^{p+q}(-1)^q=(-1)^{p+1} − ( − 1 ) p + q ( − 1 ) q = ( − 1 ) p + 1 .
Every higher connection component has negative internal degree. Every non-bare word in the twisted first Taylor map contains such a component. Hence the only filtration-preserving term is the ordinary unary kernel Γ 1 \Gamma_1 Γ 1 , and
gr q ( I N 1 K ) = I ⊗ 1 T q . \operatorname{gr}^q(I N_1^K)=I\otimes1_{T^q}. gr q ( I N 1 K ) = I ⊗ 1 T q . The unary matrix-sign sum in (6.4) is empty. This uses the actual kernel, with no second target-sign conversion. Finite-filtration induction, using the successive short exact sequences, proves the Hom quasi-isomorphism in every integer degree. □
Scalar source cohomology and detection
Proposition B.5. As a graded ring,
H ∙ ( B , Q ) = C ∞ ( R y ) [ u ] / ( y u ) , ∣ u ∣ = 2 , u = [ η ] . H^\bullet(B,Q)=C^\infty(\mathbb{R}_y)[u]/(yu),\qquad|u|=2,\qquad u=[\eta]. H ∙ ( B , Q ) = C ∞ ( R y ) [ u ] / ( y u ) , ∣ u ∣ = 2 , u = [ η ] . Proof. Directly from (B.3),
Q ( a θ 1 + b θ 2 ) = ( − y a + x b ) η + ( − y a x + x b x + b ) θ 1 θ 2 (88) Q(a\theta_1+b\theta_2)=(-ya+xb)\eta+(-ya_x+xb_x+b)\theta_1\theta_2 \tag*{(88)} Q ( a θ 1 + b θ 2 ) = ( − y a + x b ) η + ( − y a x + x b x + b ) θ 1 θ 2 (88) Q ( v η n + w θ 1 θ 2 η n − 1 ) = ( v x − w ) ξ η n , n ≥ 1. (89) Q(v\eta^n+w\theta_1\theta_2\eta^{n-1})=(v_x-w)\xi\eta^n,\qquad n\geq1. \tag*{(89)} Q ( v η n + w θ 1 θ 2 η n − 1 ) = ( v x − w ) ξ η n , n ≥ 1. (89) If a smooth scalar satisfies g ξ = 0 g\xi=0 g ξ = 0 , it vanishes away from the origin and then everywhere by density. In degree zero, Q f = 0 Qf=0 Q f = 0 therefore means f x = 0 f_x=0 f x = 0 , hence f = f ( y ) f=f(y) f = f ( y ) .
The same smooth syzygy division as above shows that a closed odd element is c ξ η n c\xi\eta^n c ξ η n . For n = 0 n=0 n = 0 it has primitive ∫ 0 x c ( s , y ) d s \int_0^x c(s,y)\,\mathrm{d}s ∫ 0 x c ( s , y ) d s ; for n ≥ 1 n\geq1 n ≥ 1 it has primitive − c θ 1 θ 2 η n − 1 -c\theta_1\theta_2\eta^{n-1} − c θ 1 θ 2 η n − 1 . Thus all odd cohomology vanishes.
In degree 2 n > 0 2n>0 2 n > 0 the cycles are exactly
z v ( n ) = v η n + v x θ 1 θ 2 η n − 1 . z_v^{(n)}=v\eta^n+v_x\theta_1\theta_2\eta^{n-1}. z v ( n ) = v η n + v x θ 1 θ 2 η n − 1 . Equation (B.4) shows that boundaries correspond exactly to v ∈ ( x , y ) v\in(x,y) v ∈ ( x , y ) . Global radial Hadamard division identifies this ideal with the functions vanishing at the origin. Consequently H 2 n ( B ) = A / ( x , y ) = R H^{2n}(B)=A/(x,y)=\mathbb{R} H 2 n ( B ) = A / ( x , y ) = R . Finally z v ( n ) z w ( m ) = z v w ( n + m ) z_v^{(n)}z_w^{(m)}=z_{vw}^{(n+m)} z v ( n ) z w ( m ) = z v w ( n + m ) . The degree-zero action on [ η ] [\eta] [ η ] is evaluation at y = 0 y=0 y = 0 , which proves the ring formula. □
Lemma B.6. The actual scalar integration map I I I is injective on cohomology.
Proof. At the origin, restrict K K K to the simplices with identity G G G -edges and integer triangle labels. Since Z 0 = 0 Z_0=0 Z 0 = 0 , these form the exact simplicial subset B 2 Z = K ( Z , 2 ) B^2\mathbb{Z}=K(\mathbb{Z},2) B 2 Z = K ( Z , 2 ) . The restriction of h h h is its universal integral two-cocycle. Thus its powers are nonzero in H ∙ ( K ( Z , 2 ) ; R ) = R [ c ] H^\bullet(K(\mathbb{Z},2);\mathbb{R})=\mathbb{R}[c] H ∙ ( K ( Z , 2 ) ; R ) = R [ c ] .
The actual kernel gives I ( η ) = − h I(\eta)=-h I ( η ) = − h and is multiplicative on cohomology by the second Taylor identity in Theorem 6.1 , applied to the unit coefficient object and pulled back by Ψ \Psi Ψ . Every positive even source generator is therefore detected. In degree zero I I I is literally f ( y ) ↦ f ( y ) f(y)\mapsto f(y) f ( y ) ↦ f ( y ) on vertices. Together with Proposition B.5, this proves injectivity. The argument uses the entire exact simplicial subset, not sampling at finitely many integer labels. □
The upper-decalage relative de Rham complex
Let
P p = ( Dec u p K ) p = K p + 1 , π p = d p + 1 : P p ⟶ K p . P_p=(\operatorname{Dec}^{\mathrm{up}}K)_p=K_{p+1},\qquad\pi_p=d_{p+1}:P_p\longrightarrow K_p. P p = ( Dec up K ) p = K p + 1 , π p = d p + 1 : P p ⟶ K p .
The faces and degeneracies of P p P_p P p have indices 0 , … , p 0,\ldots,p 0 , … , p ; the last degeneracy of K K K is an extra degeneracy, not one of these operators. The maps π p \pi_p π p form a simplicial map. In the root coordinates of § 3 ,
P p = K p × G × R p , g = a p + 1 , z i = h 0 , i , p + 1 . (90) P_p=K_p\times G\times\mathbb{R}^p,\qquad g=a_{p+1},\quad z_i=h_{0,i,p+1}. \tag*{(90)} P p = K p × G × R p , g = a p + 1 , z i = h 0 , i , p + 1 . (90)
In particular π p \pi_p π p is globally a product submersion, even at the singular base point.
Define the first-quadrant bicomplex
C p , q = Ω q ( P p / K p ) , δ = ∑ i = 0 p + 1 ( − 1 ) i d i ∗ , D T o t = δ + ( − 1 ) p d r e l . (91) C^{p,q}=\Omega^q(P_p/K_p),\qquad\delta=\sum_{i=0}^{p+1}(-1)^i d_i^*,\qquad D_{\mathrm{Tot}}=\delta+(-1)^p d_{\mathrm{rel}}. \tag*{(91)} C p , q = Ω q ( P p / K p ) , δ = i = 0 ∑ p + 1 ( − 1 ) i d i ∗ , D Tot = δ + ( − 1 ) p d rel . (91)
These are actual jointly smooth relative forms. The simplicial square for π \pi π makes every face pullback well-defined on vertical forms and commuting with d r e l d_{\mathrm{rel}} d rel ; hence D T o t 2 = 0 D_{\mathrm{Tot}}^2=0 D Tot 2 = 0 .
Lemma B.7 (The relative frame). Let m ∗ = g m 0 m_* = gm_0 m ∗ = g m 0 be the last vertex. The forms
α = d r e l g g − 1 = ( α 1 α 2 0 0 ) , d z 1 , … , d z p \alpha=d_{\mathrm{rel}}g\,g^{-1}=
\begin{pmatrix}
\alpha_1 & \alpha_2\\
0 & 0
\end{pmatrix},
\qquad dz_1,\ldots,dz_p α = d rel g g − 1 = ( α 1 0 α 2 0 ) , d z 1 , … , d z p give a global frame of the relative cotangent bundle. For the root-changing face, in degrees p ≥ 1 p\geq1 p ≥ 1 ,
d 0 ∗ α = α − Z m ∗ d z 1 , d 0 ∗ d z j = d z j + 1 − d z 1 ( 1 ≤ j ≤ p − 1 ) . (92) d_0^*\alpha=\alpha-Z_{m_*}\,dz_1,\qquad d_0^*dz_j=dz_{j+1}-dz_1\quad(1\leq j\leq p-1). \tag*{(92)} d 0 ∗ α = α − Z m ∗ d z 1 , d 0 ∗ d z j = d z j + 1 − d z 1 ( 1 ≤ j ≤ p − 1 ) . (92) Other faces delete the corresponding z i z_i z i and leave α \alpha α unchanged. Degeneracy zero inserts z 1 = 0 z_1=0 z 1 = 0 ; other degeneracies repeat a z z z -coordinate and leave α \alpha α unchanged.
Proof. The frame assertion follows from (B.6) and the global right Maurer–Cartan frame on G G G . When the root vertex is deleted, the actual nerve formulas give
g ′ = g exp ( − z 1 Z m 0 ) a 1 − 1 , z j ′ = h 0 , 1 , j + 1 + z j + 1 − z 1 . g'=g\exp(-z_1Z_{m_0})a_1^{-1},\qquad z'_j=h_{0,1,j+1}+z_{j+1}-z_1. g ′ = g exp ( − z 1 Z m 0 ) a 1 − 1 , z j ′ = h 0 , 1 , j + 1 + z j + 1 − z 1 .
The additive tetrahedron identity is part of the crossed-module nerve, including at Z m 0 = 0 Z_{m_0}=0 Z m 0 = 0 ; it is not inferred by cancelling an exponential there. The quantities a 1 , m 0 , h 0 , 1 , j + 1 a_1,m_0,h_{0,1,j+1} a 1 , m 0 , h 0 , 1 , j + 1 are base coordinates for π p \pi_p π p , so their relative differentials vanish. Since g Z m 0 g − 1 = Z m ∗ gZ_{m_0}g^{-1}=Z_{m_*} g Z m 0 g − 1 = Z m ∗ , differentiation gives (B.8). All formulas also hold when m ∗ = 0 m_*=0 m ∗ = 0 . The remaining assertions follow by deleting or repeating the indicated vertex in the rooted triangle labels. □
Put A p = C ∞ ( P p ) A^p=C^\infty(P_p) A p = C ∞ ( P p ) and A = C ∞ ( M ) A=C^\infty(M) A = C ∞ ( M ) , acting through the last vertex. Let V p V^p V p be free over A A A on e 1 , e 2 , v 1 , … , v p e_1,e_2,v_1,\ldots,v_p e 1 , e 2 , v 1 , … , v p , with cofaces and codegeneracies given by Lemma B.7. All these operators preserve the last vertex and are A A A -linear. The frame gives an isomorphism of horizontal cosimplicial modules
C p , q = A p ⊗ A Λ A q V p . (93) C^{p,q}=A^p\otimes_A\Lambda_A^qV^p. \tag*{(93)} C p , q = A p ⊗ A Λ A q V p . (93)
The second factor is finite free. Arbitrary smooth dependence on all fiber coordinates remains in A p A^p A p ; no algebraic tensor description of a smooth function space on a product is assumed.
An explicit horizontal coefficient contraction
Lemma B.8. For any cosimplicial A A A -module W ∙ W^\bullet W ∙ , last-vertex pullback and constant-simplex evaluation induce inverse horizontal chain-homotopy equivalences
W ∙ ⇄ j r A ∙ ⊗ A W ∙ . W^\bullet\ \underset{r}{\overset{j}{\rightleftarrows}}\ A^\bullet\otimes_A W^\bullet. W ∙ r ⇄ j A ∙ ⊗ A W ∙ . The assertion holds after normalization and with arbitrary external jointly smooth parameters. In particular, W ∙ W^\bullet W ∙ may be nonflat over A A A .
Proof. Let i p : A → A p i^p:A\to A^p i p : A → A p be last-vertex pullback, and let ϵ p : A p → A \epsilon^p:A^p\to A ϵ p : A p → A evaluate on the completely constant ( p + 1 ) (p+1) ( p + 1 ) -simplex. They are cosimplicial A A A -algebra maps. Set j = i ⊗ 1 j=i\otimes1 j = i ⊗ 1 , r = ϵ ⊗ 1 r=\epsilon\otimes1 r = ϵ ⊗ 1 ; then r j = 1 rj=1 r j = 1 .
There is a geometric contraction behind the formulas. For λ : [ p ] → [ 1 ] \lambda:[p]\to[1] λ : [ p ] → [ 1 ] , define an order map a λ : [ p + 1 ] → [ p + 1 ] a_\lambda:[p+1]\to[p+1] a λ : [ p + 1 ] → [ p + 1 ] by retaining s s s if s ≤ p s\le p s ≤ p and λ ( s ) = 0 \lambda(s)=0 λ ( s ) = 0 , and sending it to p + 1 p+1 p + 1 otherwise. It fixes the last vertex. If β : [ t ] → [ p ] \beta:[t]\to[p] β : [ t ] → [ p ] is extended to β + \beta^+ β + by adding a last vertex, then
a λ β + = β + a λ β . a_\lambda\beta^+=\beta^+a_\lambda\beta. a λ β + = β + a λ β .
Both sides send a vertex either to β ( s ) \beta(s) β ( s ) or to p + 1 p+1 p + 1 . Thus K ( a λ ) K(a_\lambda) K ( a λ ) is a simplicial homotopy, over the last vertex, from the identity of P P P to the constant-last-vertex map. It fixes constant simplices. Pullback gives a cosimplicial right homotopy of A ∙ A^\bullet A ∙ . Tensor its degree-p p p operators with 1 W p 1_{W^p} 1 W p . The displayed order-map identity and functoriality of W W W give their cosimplicial naturality and the endpoints 1 , j r 1,jr 1 , j r . Preservation of the last vertex makes all these operators A A A -linear. This is tensoring an actual cosimplicial homotopy, not tensoring a chain homotopy degreewise.
We give the resulting cochain homotopy explicitly. For p ≥ 1 p\ge1 p ≥ 1 and 0 ≤ j ≤ p − 1 0\le j\le p-1 0 ≤ j ≤ p − 1 , let
f j p : [ p + 1 ] ⟶ [ p ] , f j p ( s ) = { s , s ≤ j , p , s > j . f_j^p:[p+1]\longrightarrow[p],\qquad f_j^p(s)=
\begin{cases}
s,&s\le j,\\
p,&s>j.
\end{cases} f j p : [ p + 1 ] ⟶ [ p ] , f j p ( s ) = { s , p , s ≤ j , s > j .
and put h j p = K ( f j p ) : P p − 1 → P p h_j^p=K(f_j^p):P_{p-1}\to P_p h j p = K ( f j p ) : P p − 1 → P p . Let σ W j : W p → W p − 1 \sigma_W^j:W^p\to W^{p-1} σ W j : W p → W p − 1 be the codegeneracy identifying j j j and j + 1 j+1 j + 1 . Define
L p = ∑ j = 0 p − 1 ( − 1 ) j + 1 ( h j p ) ∗ ⊗ σ W j , L 0 = 0. (94) L_p=\sum_{j=0}^{p-1}(-1)^{j+1}(h_j^p)^*\otimes\sigma_W^j,\qquad L_0=0. \tag*{(94)} L p = j = 0 ∑ p − 1 ( − 1 ) j + 1 ( h j p ) ∗ ⊗ σ W j , L 0 = 0. (94)
Then
δ L + L δ = 1 − j r . (95) \delta L+L\delta=1-jr. \tag*{(95)} δ L + L δ = 1 − j r . (95)
To verify the identity in every degree, write A i j A_{ij} A ij for the unsigned tensor operator in δ p − 1 L p \delta_{p-1}L_p δ p − 1 L p , and B i j B_{ij} B ij for that in L p + 1 δ p L_{p+1}\delta_p L p + 1 δ p . Their signs are ( − 1 ) i + j + 1 (-1)^{i+j+1} ( − 1 ) i + j + 1 . As pairs of order maps, the off-diagonal terms satisfy
B i j = { A i , j − 1 , i < j , A i − 1 , j , i > j + 1. B_{ij}=
\begin{cases}
A_{i,j-1},&i<j,\\
A_{i-1,j},&i>j+1.
\end{cases} B ij = { A i , j − 1 , A i − 1 , j , i < j , i > j + 1.
The matched signs are opposite, and these cases exhaust all A \mathbf{A} A terms. In the remaining cases i = j , j + 1 i=j,j+1 i = j , j + 1 , the W W W -map is the identity. On the coefficient factor their signed sum is c j − c j − 1 c_j-c_{j-1} c j − c j − 1 , where c j c_j c j retains vertices 0 , … , j 0,\ldots,j 0 , … , j and sends all others to p + 1 p+1 p + 1 . Summing over j = 0 , … , p j=0,\ldots,p j = 0 , … , p gives c p − c − 1 c_p-c_{-1} c p − c − 1 , namely the identity minus constant-last-vertex evaluation. This proves (B.11), including p = 0 p=0 p = 0 .
Use the normalization N p T = ⋂ i = 0 p − 1 ker ( s i : T p → T p − 1 ) N^pT=\bigcap_{i=0}^{p-1}\ker(s^i:T^p\to T^{p-1}) N p T = ⋂ i = 0 p − 1 ker ( s i : T p → T p − 1 ) . A compatible finite projection is
Π N p = ( 1 − d p s p − 1 ) ⋯ ( 1 − d 1 s 0 ) . \Pi_N^p=(1-d^ps^{p-1})\cdots(1-d^1s^0). Π N p = ( 1 − d p s p − 1 ) ⋯ ( 1 − d 1 s 0 ) .
Its complement is the sum of the positive-coface images; the cosimplicial identities show that both summands are subcomplexes. Thus Π N \Pi_N Π N and the normalized inclusion ι N \iota_N ι N are cochain maps. Since j , r j,r j , r are cosimplicial, L N = Π N L ι N L_N=\Pi_NL\iota_N L N = Π N L ι N satisfies the same homotopy identity. For T = C ∙ , q T=C^{\bullet,q} T = C ∙ , q the projection also commutes with d r e l d_{\mathrm{rel}} d rel : it uses only genuine P P P -operators. All formulas are finite smooth pullbacks and remain valid with jointly smooth external parameters. □
Remark B.9. The retraction and homotopy in the lemma are horizontal only. For example, at p = 0 p=0 p = 0 the coefficient F ( g , m ) = log a F(g,m)=\log a F ( g , m ) = log a satisfies r ( F ) = 0 r(F)=0 r ( F ) = 0 , whereas r ( d r e l F ) = e 1 r(d_{\mathrm{rel}}F)=e_1 r ( d rel F ) = e 1 . Thus this horizontal contraction does not commute with the relative differential. The total comparison instead uses the total chain map constructed below and a first-quadrant filtration.
Lemma B.10 (Exact exterior normalization). Over A A A itself,
N A p Λ A q V ∙ = Λ A q − p ( A e 1 ⊕ A e 2 ) ∧ v 1 ∧ ⋯ ∧ v p . N_A^p\Lambda_A^q V^\bullet=\Lambda_A^{q-p}(Ae_1\oplus Ae_2)\wedge v_1\wedge\cdots\wedge v_p. N A p Λ A q V ∙ = Λ A q − p ( A e 1 ⊕ A e 2 ) ∧ v 1 ∧ ⋯ ∧ v p . It vanishes unless 0 ≤ q − p ≤ 2 0\le q-p\le2 0 ≤ q − p ≤ 2 .
Proof. The codegeneracy kernels on V p V^p V p are the independent lines generated by w 1 = v 1 w_1=v_1 w 1 = v 1 , w 2 = v 2 − v 1 , … , w p = v p − v p − 1 w_2=v_2-v_1,\ldots,w_p=v_p-v_{p-1} w 2 = v 2 − v 1 , … , w p = v p − v p − 1 . Together with e 1 , e 2 e_1,e_2 e 1 , e 2 these form a free A A A -basis. On the exterior algebra, each kernel is the ideal generated by the corresponding w i w_i w i . A basis monomial belongs to all these ideals precisely when it contains all p p p symbols. Their wedge is v 1 ∧ ⋯ ∧ v p v_1\wedge\cdots\wedge v_p v 1 ∧ ⋯ ∧ v p . For p = 0 p=0 p = 0 the intersection is empty and the formula is Λ q ( A e 1 ⊕ A e 2 ) \Lambda^q(Ae_1\oplus Ae_2) Λ q ( A e 1 ⊕ A e 2 ) . The argument uses a free basis over A A A , not a fiberwise rank calculation at regular points. □
The actual scalar comparison
Theorem B.11 (Scalar affine comparison). The actual map I : ( B , Q ) ⟶ ( C K ∙ , δ ) I:(B,Q)\longrightarrow(C_K^\bullet,\delta) I : ( B , Q ) ⟶ ( C K ∙ , δ ) is a quasi-isomorphism. Consequently
H ∙ ( C K ∙ ) = C ∞ ( R y ) [ u ] / ( y u ) , u = [ − h ] , ∣ u ∣ = 2 , H^\bullet(C_K^\bullet)=C^\infty(\mathbb{R}_y)[u]/(yu),\qquad u=[-h],\quad|u|=2, H ∙ ( C K ∙ ) = C ∞ ( R y ) [ u ] / ( y u ) , u = [ − h ] , ∣ u ∣ = 2 , with this ring identification induced by I I I .
Proof. There are two different filtrations to use. First, π p ∗ \pi_p^* π p ∗ defines a cochain map from unnormalized smooth cochains of K K K to Tot C \operatorname{Tot} C Tot C . The fiber in (B.6) has global coordinates ( log a , b , z ) (\log a,b,z) ( log a , b , z ) and contracts radially to ( 0 , 0 , 0 ) (0,0,0) ( 0 , 0 , 0 ) . The relative Poincare operator integrates over a compact interval and is smooth in all base parameters. Each column therefore has cohomology C ∞ ( K p ) C^\infty(K_p) C ∞ ( K p ) in relative degree zero and zero otherwise. Filtering by the horizontal degree p p p gives a quasi-isomorphism
J : C s m ∙ ( K ) ⟶ Tot C . J:C_{\mathrm{sm}}^\bullet(K)\longrightarrow\operatorname{Tot} C. J : C sm ∙ ( K ) ⟶ Tot C . No compatibility of the chosen Poincare operators with faces is needed, since J J J itself is a cochain map. Ordinary cosimplicial normalization is a natural chain-homotopy equivalence, constructed from the genuine face and degeneracy maps. These commute with d r e l d_{\mathrm{rel}} d rel , as does J J J . Thus the same conclusion holds for C K ∙ C_K^\bullet C K ∙ and Tot N C \operatorname{Tot} NC Tot N C . This is full cosimplicial normalization, not normalization of the inner-chain complex used for higher prisms.
For the second comparison give B B B the bigrading
B p , q = A η p Λ q − p ( θ 1 , θ 2 ) , q − p ∈ { 0 , 1 , 2 } . B^{p,q}=A\eta^{p}\Lambda^{q-p}(\theta_1,\theta_2),\qquad q-p\in\{0,1,2\}. B p , q = A η p Λ q − p ( θ 1 , θ 2 ) , q − p ∈ { 0 , 1 , 2 } .
The total degree is p + q p+q p + q . Split Q = Q h + Q v Q=Q_h+Q_v Q = Q h + Q v , where Q h = η ι Z Q_h=\eta\iota_Z Q h = η ι Z for Z = ( − y , x ) Z=(-y,x) Z = ( − y , x ) , and
Q v f = f x ξ , Q v θ 1 = 0 , Q v θ 2 = θ 1 θ 2 , Q v η = 0. Q_vf=f_x\xi,\qquad Q_v\theta_1=0,\qquad Q_v\theta_2=\theta_1\theta_2,\qquad Q_v\eta=0. Q v f = f x ξ , Q v θ 1 = 0 , Q v θ 2 = θ 1 θ 2 , Q v η = 0.
These components square to zero and anticommute. For the mixed identity on θ 2 \theta_2 θ 2 the two terms are − ξ η -\xi\eta − ξ η and + ξ η +\xi\eta + ξ η ; on functions use Q h ξ = 0 Q_h\xi=0 Q h ξ = 0 . The other generators give zero directly. Thus this is a genuine first-quadrant double complex. For a theta monomial θ I \theta_I θ I of exterior degree r r r , put
ι ( f η p θ I ) = c p , r f ( m ∗ ) α I ∧ d z 1 ∧ ⋯ ∧ d z p , c p , r = ( − 1 ) p r + p ( p + 1 ) / 2 . (96) \iota(f\eta^p\theta_I)=c_{p,r}f(m_*)\alpha_I\wedge dz_1\wedge\cdots\wedge dz_p,\qquad c_{p,r}=(-1)^{pr+p(p+1)/2}. \tag*{(96)} ι ( f η p θ I ) = c p , r f ( m ∗ ) α I ∧ d z 1 ∧ ⋯ ∧ d z p , c p , r = ( − 1 ) p r + p ( p + 1 ) /2 . (96)
Lemma B.10 identifies its finite free part exactly with N p Λ q V N^p\Lambda^qV N p Λ q V .
We verify the total chain-map identity directly. In the horizontal differential the terms without the root-face change of α \alpha α cancel:
δ ( d z 1 ∧ ⋯ ∧ d z p ) = 0. \delta(dz_1\wedge\cdots\wedge dz_p)=0. δ ( d z 1 ∧ ⋯ ∧ d z p ) = 0.
Expanding ∏ j = 2 p + 1 ( d z j − d z 1 ) \prod_{j=2}^{p+1}(dz_j-dz_1) ∏ j = 2 p + 1 ( d z j − d z 1 ) gives the signed sum of the wedges omitting one d z i dz_i d z i , which is exactly this cancellation. The remaining root-face term is
δ ( α I ∧ d z 1 ∧ ⋯ ∧ d z p ) = ( − 1 ) r ( ι Z α I ) ∧ d z 1 ∧ ⋯ ∧ d z p + 1 . \delta(\alpha_I\wedge dz_1\wedge\cdots\wedge dz_p)=(-1)^r(\iota_Z\alpha_I)\wedge dz_1\wedge\cdots\wedge dz_{p+1}. δ ( α I ∧ d z 1 ∧ ⋯ ∧ d z p ) = ( − 1 ) r ( ι Z α I ) ∧ d z 1 ∧ ⋯ ∧ d z p + 1 .
Since c p + 1 , r − 1 = ( − 1 ) r c p , r c_{p+1,r-1}=(-1)^rc_{p,r} c p + 1 , r − 1 = ( − 1 ) r c p , r , it gives δ ι = ι Q h \delta\iota=\iota Q_h δ ι = ι Q h . For the vertical differential the right Maurer–Cartan equation gives
d r e l f ( m ∗ ) = f x ( m ∗ ) ( x ∗ α 1 + y ∗ α 2 ) , d r e l α 1 = 0 , d r e l α 2 = α 1 ∧ α 2 . d_{\mathrm{rel}}f(m_*)=f_x(m_*)(x_*\alpha_1+y_*\alpha_2),\qquad d_{\mathrm{rel}}\alpha_1=0,\qquad d_{\mathrm{rel}}\alpha_2=\alpha_1\wedge\alpha_2. d rel f ( m ∗ ) = f x ( m ∗ ) ( x ∗ α 1 + y ∗ α 2 ) , d rel α 1 = 0 , d rel α 2 = α 1 ∧ α 2 .
All d z i dz_i d z i are closed. The identity c p , r + 1 = ( − 1 ) p c p , r c_{p,r+1}=(-1)^pc_{p,r} c p , r + 1 = ( − 1 ) p c p , r therefore gives ( − 1 ) p d r e l ι = ι Q v (-1)^pd_{\mathrm{rel}}\iota=\iota Q_v ( − 1 ) p d rel ι = ι Q v . Thus (B.12) is an actual total chain map.
For every fixed q q q , its horizontal part is the normalized coaugmentation of Lemma B.8 , after the exact exterior identification. It is a rowwise quasi-isomorphism. Filtering now by the relative degree q q q shows that ι : B → Tot N C \iota:B\to\operatorname{Tot}NC ι : B → Tot N C is a quasi-isomorphism. Both comparisons are first quadrant with only finitely many summands in each total degree. We have therefore constructed a chain zigzag
C K ∙ → ∼ J Tot N C ← ∼ ι ( B , Q ) . (97) C^\bullet_K\xrightarrow[\sim]{J}\operatorname{Tot}NC\xleftarrow[\sim]{\iota}(B,Q). \tag*{(97)} C K ∙ J ∼ Tot N C ι ∼ ( B , Q ) . (97)
The zigzag computes the graded vector spaces. To identify the product for the actual map I I I , we use its cohomological multiplicativity and the injectivity proved above. Positive even target cohomology has dimension one and odd target cohomology vanishes. Lemma B.6 therefore makes H ( I ) H(I) H ( I ) an isomorphism in positive degrees. For degree zero argue directly: a smooth G G G -invariant function on M M M is independent of x x x for y ≠ 0 y\ne0 y = 0 , since the translation parameter moves x x x arbitrarily. Smoothness extends this independence to y = 0 y=0 y = 0 . Thus target H 0 = C ∞ ( R y ) H^0=C^\infty(\mathbb{R}_y) H 0 = C ∞ ( R y ) , and I I I is literally the identity on it. Negative scalar cochain degrees vanish. This proves quasi-isomorphism of the actual I I I . Its A ∞ A_\infty A ∞ multiplicativity on cohomology, together with I ( η ) = − h I(\eta)=-h I ( η ) = − h , gives the stated ring identification.
Corollary B.12 (External scalar parameters). The same actual scalar comparison is a quasi-isomorphism for jointly smooth coefficients on U × R 2 U \times\mathbb{R}^{2} U × R 2 and cochains on U × K p U \times K_{p} U × K p , with the differential only in the affine variables, for every external smooth parameter manifold U U U .
Proof. Put A U = C ∞ ( U × M ) A_{U}=C^{\infty}(U \times M) A U = C ∞ ( U × M ) and A U p = C ∞ ( U × P p ) A_{U}^{p}=C^{\infty}(U \times P_{p}) A U p = C ∞ ( U × P p ) , with the latter an A U A_{U} A U -algebra through ( u , m ∗ ) (u,m_{\ast}) ( u , m ∗ ) . The actual relative complex is
C U p , q = Ω q ( ( U × P p ) / ( U × K p ) ) = A U p ⊗ A U Λ q V U p , V U p = A U p + 2 . C_{U}^{p,q}=\Omega^{q}((U \times P_{p})/(U \times K_{p}))=A_{U}^{p}\otimes_{A_{U}}\Lambda^{q}V_{U}^{p}, \qquad V_{U}^{p}=A_{U}^{p+2}. C U p , q = Ω q (( U × P p ) / ( U × K p )) = A U p ⊗ A U Λ q V U p , V U p = A U p + 2 .
in the same frame α 1 , α 2 , d z 1 , … , d z p \alpha_{1},\alpha_{2},dz_{1},\ldots,dz_{p} α 1 , α 2 , d z 1 , … , d z p . The tensor description uses this finite free frame and retains all joint smooth dependence in A U p A_{U}^{p} A U p . All parameterized operators are obtained by taking the product with 1 U 1_{U} 1 U , so they preserve the external parameter u u u . The genuine upper-decalage simplicial operators and the horizontal order maps used in L L L preserve the terminal vertex m ∗ m_{\ast} m ∗ . Their coefficient pullback and evaluation maps are therefore A U A_{U} A U -linear through ( u , m ∗ ) (u,m_{\ast}) ( u , m ∗ ) , and the horizontal identity δ L + L δ = 1 − j r \delta L+L\delta=1-jr δ L + L δ = 1 − j r and the exterior normalization hold over A U A_{U} A U .
The column radial homotopy is instead a homotopy over U × K p U \times K_{p} U × K p . It generally moves m ∗ = g m 0 m_{\ast}=gm_{0} m ∗ = g m 0 and need not be A U A_{U} A U -linear for the terminal action. Its relative Poincare operator is C ∞ ( U × K p ) C^{\infty}(U \times K_{p}) C ∞ ( U × K p ) -linear, which is the property required for the column cohomology calculation. Both constructions are C ∞ ( U ) C^{\infty}(U) C ∞ ( U ) -linear and commute with smooth pullback of the external parameter.
The column Poincare operator still integrates only the radial fiber variables ( log a , b , z ) (\log a,b,z) ( log a , b , z ) over [ 0 , 1 ] [0,1] [ 0 , 1 ] . It is smooth in ( u , K p ) (u,K_{p}) ( u , K p ) , and commutes with every smooth change of external parameters. The displayed formula for ι \iota ι also remains a total chain map, since no differential acts in the U U U directions. The same two first-quadrant filtrations therefore give actual quasi-isomorphisms
C n o r m , s m ∙ ( U × K ) → ∼ J U Tot N C U ← ∼ ι U ( B U , Q ) , B U = A U [ θ 1 , θ 2 , η ] . C_{\mathrm{norm,sm}}^{\bullet}(U \times K)\xrightarrow[\sim]{J_{U}}\operatorname{Tot}NC_{U}\xleftarrow[\sim]{\iota_{U}}(B_{U},Q), \qquad B_{U}=A_{U}[\theta_{1},\theta_{2},\eta]. C norm , sm ∙ ( U × K ) J U ∼ Tot N C U ι U ∼ ( B U , Q ) , B U = A U [ θ 1 , θ 2 , η ] .
All three complexes and both arrows commute with pullback along U ′ → U U'\to U U ′ → U through the explicit parameterized operators above.
The primitives and radial Hadamard formula in Proposition B.5 integrate only the x , y x,y x , y variables and remain jointly smooth in u u u . The source degree-zero cohomology is C ∞ ( U × R y ) C^{\infty}(U \times\mathbb{R}_{y}) C ∞ ( U × R y ) , the positive even cohomology is C ∞ ( U ) C^{\infty}(U) C ∞ ( U ) , and the odd cohomology vanishes. Let S U = H ( ι U ) − 1 H ( J U ) S_{U}=H(\iota_{U})^{-1}H(J_{U}) S U = H ( ι U ) − 1 H ( J U ) be the resulting natural C ∞ ( U ) C^{\infty}(U) C ∞ ( U ) -linear cohomology isomorphism. The actual I U I_{U} I U is also parameter-natural: its simplex integrals keep u u u fixed, and hence commute with such pullbacks. For π : U → ∗ \pi:U\to\ast π : U → ∗ and n ≥ 1 n\geq1 n ≥ 1 this gives
S U H ( I U ) [ η n ] = π ∗ ( S ∗ H ( I ∗ ) [ η n ] ) = λ n [ η n ] , λ n ∈ R × . S_{U}H(I_{U})[\eta^{n}]=\pi^{\ast}(S_{\ast}H(I_{\ast})[\eta^{n}])=\lambda_{n}[\eta^{n}], \qquad\lambda_{n}\in\mathbb{R}^{\times}. S U H ( I U ) [ η n ] = π ∗ ( S ∗ H ( I ∗ ) [ η n ]) = λ n [ η n ] , λ n ∈ R × .
Nonvanishing follows from the unparameterized theorem. Multiplication by this fixed real constant is invertible. The fixed nonzero scalar gives an inverse on the rank-one C ∞ ( U ) C^{\infty}(U) C ∞ ( U ) -module in each positive even degree. The degree-zero invariant argument works for each u u u and gives literally f ( u , y ) ↦ f ( u , y ) f(u,y)\mapsto f(u,y) f ( u , y ) ↦ f ( u , y ) . Joint smooth dependence is carried by the displayed parameterized complexes on the whole affine plane.
Proof of Theorem B.1 . Apply Theorem B.11 to Lemma B.4 . For the full MC target, the actual functor Ψ ∗ \Psi^* Ψ ∗ is the dg quasi-equivalence inverse to Φ ∗ \Phi^* Φ ∗ from Corollary 5.4 . Since I N K = Ψ ∗ I N E \mathrm{IN}^{K}=\Psi^*\mathrm{IN}_{E} IN K = Ψ ∗ IN E , its Hom quasi-isomorphisms give the corresponding assertion for I N E \mathrm{IN}_{E} IN E through the constructed raw comparison. □ \square □
Appendix C. Regular reconstruction and finite MC lifting
We use the whole-plane notation E E E , B B B , K K K , R K \mathcal{R}_{K} R K and I N K \mathrm{IN}^{K} IN K of Appendix B. We complement the regular Hom comparison with a finite reconstruction proving essential surjectivity in the regular raw category. The finite MC-lifting lemma below is also used in the vertical-family argument of Section 13.
Lemma C.1 (The raw cochain-module dictionary). In global graded frames, raw representations of K K K are the finite free graded right C K ∙ C_{K}^{\bullet} C K ∙ -modules with superconnections encoded by
r = e + b , d f u l l b + b ⋆ b = 0 , e 1 = 1 , e p = 0 ( p ≠ 1 ) . r=e+b,\qquad d_{\mathrm{full}}b+b\star b=0,\qquad e_{1}=1,\qquad e_{p}=0\ (p\ne1). r = e + b , d full b + b ⋆ b = 0 , e 1 = 1 , e p = 0 ( p = 1 ) . Their Hom differential is exactly the raw representation differential. The degree-zero algebra C K 0 = C ∞ ( M ) C_{K}^{0}=C^{\infty}(M) C K 0 = C ∞ ( M ) acts at the two endpoints.
Proof. Every bounded graded vertex bundle is trivial on M = R 2 M=\mathbb{R}^{2} M = R 2 ; choose its constant graded vector space W W W . In forms-first order, the endomorphism algebra of the corresponding right module is C K ∙ ⊗ End R ( W ) C_{K}^{\bullet}\otimes\operatorname{End}_{\mathbb{R}}(W) C K ∙ ⊗ End R ( W ) , with product
( ω ⊗ T ) ( ν ⊗ S ) = ( − 1 ) ∣ T ∣ ∣ ν ∣ ( ω ⌣ ν ) ⊗ T S . (\omega\otimes T)(\nu\otimes S)=(-1)^{|T||\nu|}(\omega\smallsmile\nu)\otimes TS. ( ω ⊗ T ) ( ν ⊗ S ) = ( − 1 ) ∣ T ∣∣ ν ∣ ( ω ⌣ ν ) ⊗ T S .
Indeed its action on C K ∙ ⊗ W C_{K}^{\bullet}\otimes W C K ∙ ⊗ W is Φ ω , T ( c ⊗ v ) = ( − 1 ) ∣ T ∣ ∣ c ∣ ω c ⊗ T v \Phi_{\omega,T}(c\otimes v)=(-1)^{|T||c|}\omega c\otimes Tv Φ ω , T ( c ⊗ v ) = ( − 1 ) ∣ T ∣∣ c ∣ ω c ⊗ T v . Here the right module structure is the braided one, ( c ⊗ v ) a = ( − 1 ) ∣ v ∣ ∣ a ∣ ( c a ) ⊗ v (c\otimes v)a=(-1)^{|v||a|}(ca)\otimes v ( c ⊗ v ) a = ( − 1 ) ∣ v ∣∣ a ∣ ( c a ) ⊗ v . This makes the displayed endomorphisms right-linear with the signless convention for homogeneous morphisms. The formula uses associativity, not commutativity of C K C_{K} C K .
The unit representation satisfies δ i n e + e ⋆ e = 0 \delta_{\mathrm{in}}e+e\star e=0 δ in e + e ⋆ e = 0 , and
d f u l l α = δ i n α + e ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e . d_{\mathrm{full}}\alpha=\delta_{\mathrm{in}}\alpha+e\star\alpha-(-1)^{|\alpha|}\alpha\star e. d full α = δ in α + e ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ e .
Substituting r = e + b r=e+b r = e + b in the raw MC equation therefore gives the displayed ordinary DGA MC equation. For a morphism from b P b_{P} b P to b Q b_{Q} b Q , the differential becomes
d f u l l α + b Q ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ b P , d_{\mathrm{full}}\alpha+b_{Q}\star\alpha-(-1)^{|\alpha|}\alpha\star b_{P}, d full α + b Q ⋆ α − ( − 1 ) ∣ α ∣ α ⋆ b P ,
the superconnection Hom differential. Degree-zero functions multiply a cochain at its first vertex on the left and its last vertex on the right. Their noncentrality is exactly what the cochain product records. Normalization gives the strict units. The degree-two MC identities on the h = 0 h=0 h = 0 inverse-arrow triangles exhibit the reverse edge as a chain-homotopy inverse, so the edge quasi-isomorphism condition also holds. Changing a vertex frame by g g g gives b g = g − 1 b g + g − 1 d f u l l g b^{g}=g^{-1}bg+g^{-1}d_{\mathrm{full}}g b g = g − 1 b g + g − 1 d full g , so the dictionary is independent of the chosen frames. □ \square □
Lemma C.2 (Minimal target and trivial edge transport). Every regular object of R K \mathcal{R}_{K} R K is homotopy equivalent inside that dg category to one with r 0 = 0 r_{0}=0 r 0 = 0 and r 1 = 1 r_{1}=1 r 1 = 1 .
Proof. Apply Lemma B.2 to the generally noncommutative DGA C K C_K C K using Lemma C.1. In the forms-first frames, extend the contraction by left ⋆ \star ⋆ -multiplication by its zero-cochains. For a degree-j j j matrix T T T and a scalar p p p -cochain c c c , the actual extension is
( T ^ ( c ⊗ v ) ) ( σ ) = ( − 1 ) j p c ( σ ) T ( v 0 σ ) v . (\widehat{T}(c\otimes v))(\sigma)=(-1)^{jp}c(\sigma)T(v_{0\sigma})v. ( T ( c ⊗ v )) ( σ ) = ( − 1 ) j p c ( σ ) T ( v 0 σ ) v .
Thus h h h carries ( − 1 ) p (-1)^p ( − 1 ) p , not unsigned pointwise extension. Associativity gives right C K C_K C K -linearity. The signs from d 0 d_0 d 0 and h h h cancel in d 0 h + h d 0 d_0h+hd_0 d 0 h + h d 0 ; i i i , p p p have degree zero, so all contraction side conditions remain valid. The proof of right-linearity there uses only the Leibniz rule and the contraction side conditions. Its perturbation has positive cochain degree, so the internal-degree series is finite. All operations preserve actual smooth normalized cochains. The resulting vertex cohomology bundle H H H is again globally trivial, and its internal differential is zero.
The transferred edge map is its genuine cohomology transport. Write p j = p ( v j ) p_j=p(v_j) p j = p ( v j ) and i j = i ( v j ) i_j=i(v_j) i j = i ( v j ) for the contraction matrices at the endpoints. The first positive-degree term is
b 1 ′ = p 0 [ ( i 1 − i 0 ) + ( r 1 − 1 ) i 1 ] = p 0 r 1 i 1 − 1. b'_1=p_0\left[(i_1-i_0)+(r_1-1)i_1\right]=p_0r_1i_1-1. b 1 ′ = p 0 [ ( i 1 − i 0 ) + ( r 1 − 1 ) i 1 ] = p 0 r 1 i 1 − 1.
The first bracketed term is d f u l l i d_{\mathrm{full}}i d full i ; it cannot be omitted. Thus r 1 ′ = p 0 r 1 i 1 r'_1=p_0r_1i_1 r 1 ′ = p 0 r 1 i 1 , an isomorphism on the zero-differential cohomology fibers, equal to 1 1 1 on a unit. Moreover D H − d f u l l D_H-d_{\mathrm{full}} D H − d full is right-linear, so it is left multiplication by a normalized cochain b ′ b' b ′ . Its degree-zero part vanishes. These facts give an actual minimal object of the raw representation category.
Now assume r 0 = 0 r_0=0 r 0 = 0 and denote the edge map by R ( g , m ) R(g,m) R ( g , m ) , from the fiber at g m gm g m to the fiber at m m m . The degree-two MC equation says
R ( g 12 g 01 exp ( h Z m ) , m ) = R ( g 01 , m ) R ( g 12 , g 01 m ) . (98) R(g_{12}g_{01}\exp(hZ_m),m)=R(g_{01},m)R(g_{12},g_{01}m). \tag*{(98)} R ( g 12 g 01 exp ( h Z m ) , m ) = R ( g 01 , m ) R ( g 12 , g 01 m ) . (98)
At h = 0 h=0 h = 0 this is a strict contravariant action-groupoid representation. Units and inverse arrows make its edge maps invertible. Setting both displayed g i j g_{ij} g ij to the identity gives R ( exp ( h Z m ) , m ) = 1 R(\exp(hZ_m),m)=1 R ( exp ( h Z m ) , m ) = 1 . Here ρ ( Z m ) ∣ m = 0 \rho(Z_m)|_m=0 ρ ( Z m ) ∣ m = 0 , so exp ( h Z m ) \exp(hZ_m) exp ( h Z m ) fixes m m m . The triangle and the last-to-first transport are therefore typed over the stated vertices, including the origin.
Let E 1 , E 2 E_1,E_2 E 1 , E 2 be the matrix generators whose fundamental fields are x ∂ x , y ∂ x x\partial_x,y\partial_x x ∂ x , y ∂ x , and set
L i ( m ) = d d t ∣ t = 0 R ( exp ( t E i ) , m ) . L_i(m)=\left.\frac{d}{dt}\right|_{t=0}R(\exp(tE_i),m). L i ( m ) = d t d t = 0 R ( exp ( t E i ) , m ) .
Differentiating the boundary identity gives − y L 1 + x L 2 = 0 -yL_1+xL_2=0 − y L 1 + x L 2 = 0 . Global Hadamard division yields L 1 = x L , L 2 = y L L_1=xL,L_2=yL L 1 = xL , L 2 = y L as before. The graded vertex gauge
g x = − L g , g ( 0 , y ) = 1 g_x=-Lg,\qquad g(0,y)=1 g x = − Lg , g ( 0 , y ) = 1
is globally smooth and invertible. It changes the infinitesimal matrices to g − 1 L i g + g − 1 ρ i ( g ) = 0 g^{-1}L_i g+g^{-1}\rho_i(g)=0 g − 1 L i g + g − 1 ρ i ( g ) = 0 .
This kills the whole edge representation, not merely its first derivative. Along a smooth path a ( t ) a(t) a ( t ) in G G G starting at the identity, write a ( t + ϵ ) = c t ( ϵ ) a ( t ) a(t+\epsilon)=c_t(\epsilon)a(t) a ( t + ϵ ) = c t ( ϵ ) a ( t ) . Then c t ( 0 ) = 1 c_t(0)=1 c t ( 0 ) = 1 and c t ′ ( 0 ) = a ˙ ( t ) a ( t ) − 1 c'_t(0)=\dot a(t)a(t)^{-1} c t ′ ( 0 ) = a ˙ ( t ) a ( t ) − 1 . The contravariant cocycle equation gives
d d t R δ ( a ( t ) , m ) = R δ ( a ( t ) , m ) L a ˙ ( t ) a ( t ) − 1 δ ( a ( t ) m ) = 0. \frac{d}{dt}R^\delta(a(t),m)=R^\delta(a(t),m)L^\delta_{\dot a(t)a(t)^{-1}}(a(t)m)=0. d t d R δ ( a ( t ) , m ) = R δ ( a ( t ) , m ) L a ˙ ( t ) a ( t ) − 1 δ ( a ( t ) m ) = 0.
The group G = Aff + ( 1 ) G=\operatorname{Aff}_{+}(1) G = Aff + ( 1 ) is connected, so R g ( a , m ) = 1 R^{g}(a,m)=1 R g ( a , m ) = 1 for every arrow. Here the ODE gives an ordinary change of frame, rather than algebroid transport from the origin. The ordinary gauge is a strict closed isomorphism from the edge-trivialized representation to the preceding minimal one. □ \square □
Finite filtered Maurer–Cartan lifting
We give the finite obstruction argument needed for reconstruction. The finite filtration makes every sum and gauge below terminate.
Lemma C.3 (Finite MC lifting). Let L , M L,M L , M be nonunital DGAs over R \mathbb{R} R with finite descending filtrations F 1 L = L F^{1}L=L F 1 L = L , F ν + 1 L = 0 F^{\nu+1}L=0 F ν + 1 L = 0 , and similarly for M M M . Assume the differential preserves filtration and F i F j ⊂ F i + j F^{i}F^{j}\subset F^{i+j} F i F j ⊂ F i + j . Let F : L ⇝ M F:L\rightsquigarrow M F : L ⇝ M be an uncurved A ∞ A_{\infty} A ∞ -map with the arity-additive bounds
F n ( F i 1 L , … , F i n L ) ⊂ F i 1 + ⋯ + i n M . (99) F_{n}(F^{i_{1}}L,\ldots,F^{i_{n}}L)\subset F^{i_{1}+\cdots+i_{n}}M. \tag*{(99)} F n ( F i 1 L , … , F i n L ) ⊂ F i 1 + ⋯ + i n M . (99) Here gauges are elements g ∈ 1 + M 0 g\in1+M^{0} g ∈ 1 + M 0 in the unitization, with
b g = g − 1 b g + g − 1 d g . b^{g}=g^{-1}bg+g^{-1}dg. b g = g − 1 b g + g − 1 d g . Their inverses are finite by the filtration. If gr ℓ F 1 \operatorname{gr}^{\ell}F_{1} gr ℓ F 1 is a quasi-isomorphism for every ℓ \ell ℓ , every target MC element is gauge-equivalent to F ∗ ( a ) F_{*}(a) F ∗ ( a ) for a source MC element a a a .
Proof. Use the suspended bar convention of Section 6. If F ‾ n : ( s L ) ⊗ n → s M \overline{F}_{n}:(sL)^{\otimes n}\to sM F n : ( s L ) ⊗ n → s M are its bar components, define
F ∗ ( a ) = s − 1 ∑ n = 1 ν F ‾ n ( ( s a ) ⊗ n ) . F_{*}(a)=s^{-1}\sum_{n=1}^{\nu}\overline{F}_{n}\bigl((sa)^{\otimes n}\bigr). F ∗ ( a ) = s − 1 n = 1 ∑ ν F n ( ( s a ) ⊗ n ) .
Higher terms vanish by (C.2). This is the primitive part of the formal group-like bar expression; only its finite filtered components are used. The suspension and tensor signs follow the convention fixed in Section 6.
Put Ω ( a ) = d a + a 2 \Omega(a)=da+a^{2} Ω ( a ) = d a + a 2 . Since ∣ s a ∣ = 0 |sa|=0 ∣ s a ∣ = 0 , the bar identities b 1 ( s a ) = − s ( d a ) b_{1}(sa)=-s(da) b 1 ( s a ) = − s ( d a ) and b 2 ( s a , s a ) = − s ( a 2 ) b_{2}(sa,sa)=-s(a^{2}) b 2 ( s a , s a ) = − s ( a 2 ) give the curvature identity
s Ω ( F ∗ ( a ) ) = ∑ r , t ≥ 0 F ‾ r + t + 1 ( ( s a ) ⊗ r , s Ω ( a ) , ( s a ) ⊗ t ) . s\Omega(F_{*}(a))=\sum_{r,t\geq0}\overline{F}_{r+t+1}\bigl((sa)^{\otimes r},s\Omega(a),(sa)^{\otimes t}\bigr). s Ω ( F ∗ ( a )) = r , t ≥ 0 ∑ F r + t + 1 ( ( s a ) ⊗ r , s Ω ( a ) , ( s a ) ⊗ t ) .
All terms in this primitive expression are finite by the same filtration bound.
Suppose, at stage ℓ \ell ℓ , that Ω ( a ) ∈ F ℓ L 2 \Omega(a)\in F^{\ell}L^{2} Ω ( a ) ∈ F ℓ L 2 and that the current target MC element b b b , after the accumulated target gauge, satisfies c = b − F ∗ ( a ) ∈ F ℓ M 1 c=b-F_{*}(a)\in F^{\ell}M^{1} c = b − F ∗ ( a ) ∈ F ℓ M 1 . This holds initially with ℓ = 1 \ell=1 ℓ = 1 and a = 0 a=0 a = 0 . Work modulo F ℓ + 1 F^{\ell+1} F ℓ + 1 . The Bianchi identity makes Ω ( a ) \Omega(a) Ω ( a ) closed in gr ℓ L \operatorname{gr}^{\ell}L gr ℓ L . In the curvature identity, every additional insertion of a ∈ F 1 a\in F^{1} a ∈ F 1 raises filtration, so
0 = Ω ( b ) ≡ F 1 ( Ω ( a ) ) + d c . 0=\Omega(b)\equiv F_{1}(\Omega(a))+dc. 0 = Ω ( b ) ≡ F 1 ( Ω ( a )) + d c .
Indeed the omitted products F ∗ ( a ) c , c F ∗ ( a ) F_{*}(a)c,cF_{*}(a) F ∗ ( a ) c , c F ∗ ( a ) have filtration at least ℓ + 1 \ell+1 ℓ + 1 , and c 2 c^{2} c 2 has filtration 2 ℓ ≥ ℓ + 1 2\ell\geq\ell+1 2 ℓ ≥ ℓ + 1 . Injectivity on H 2 H^{2} H 2 gives u ∈ F ℓ L 1 u\in F^{\ell}L^{1} u ∈ F ℓ L 1 with d u = − Ω ( a ) du=-\Omega(a) d u = − Ω ( a ) . Replacing a a a by a + u a+u a + u kills its curvature in this layer. Recompute c = b − F ∗ ( a ) c=b-F_{*}(a) c = b − F ∗ ( a ) after this correction; it changes by − F 1 ( u ) -F_{1}(u) − F 1 ( u ) modulo F ℓ + 1 F^{\ell+1} F ℓ + 1 .
The new discrepancy c c c is closed in the target layer. Surjectivity on H 1 H^{1} H 1 gives a source representative t ∈ F ℓ L 1 t\in F^{\ell}L^{1} t ∈ F ℓ L 1 , closed modulo F ℓ + 1 F^{\ell+1} F ℓ + 1 , whose image has the same class. Replacing a a a by a + t a+t a + t preserves the improved curvature condition. Again recompute the discrepancy, now subtracting F 1 ( t ) F_1(t) F 1 ( t ) in this layer. The remaining discrepancy is d z dz d z modulo F ℓ + 1 F^{\ell+1} F ℓ + 1 , for z ∈ F ℓ M 0 z \in F^\ell M^0 z ∈ F ℓ M 0 .
In the unitization take g = 1 − z g=1-z g = 1 − z ; its inverse is a finite geometric series. Since products with b ∈ F 1 b \in F^1 b ∈ F 1 have higher filtration,
b g = g − 1 b g + g − 1 d g = b − d z = F ∗ ( a ) ( m o d F ℓ + 1 ) . b^g=g^{-1}bg+g^{-1}dg=b-dz=F_*(a)\pmod{F^{\ell+1}}. b g = g − 1 b g + g − 1 d g = b − d z = F ∗ ( a ) ( mod F ℓ + 1 ) .
This completes the induction. After finitely many stages, Ω ( a ) = 0 \Omega(a)=0 Ω ( a ) = 0 and b q = F ∗ ( a ) b^q=F_*(a) b q = F ∗ ( a ) for the accumulated gauge q = g 1 g 2 ⋯ g ν q=g_1g_2\cdots g_\nu q = g 1 g 2 ⋯ g ν in induction order: b g 1 g 2 = ( b g 1 ) g 2 b^{g_1g_2}=(b^{g_1})^{g_2} b g 1 g 2 = ( b g 1 ) g 2 . Its orientation is fixed by
( d + b ) q = q ( d + b q ) . (100) (d+b)q=q(d+b^q). \tag*{(100)} ( d + b ) q = q ( d + b q ) . (100)
Thus q q q is a closed invertible map from F ∗ ( a ) F_*(a) F ∗ ( a ) to b b b , giving an actual morphism between coefficient objects by finite induction. □
The global affine Riemann–Hilbert equivalence
Theorem C.4 (Global regular affine RH). The actual integration functors
I N K : Coh r e g ( B ) ⟶ R K r e g , I N E : Coh r e g ( B ) ⟶ Rep s m p e r f , r e g ( Int ( E ) ) \mathrm{IN}^{K}:\operatorname{Coh}^{\mathrm{reg}}(B)\longrightarrow\mathcal{R}^{\mathrm{reg}}_{K},\qquad
\mathrm{IN}_{E}:\operatorname{Coh}^{\mathrm{reg}}(B)\longrightarrow\operatorname{Rep}^{\mathrm{perf,reg}}_{\mathrm{sm}}(\operatorname{Int}(E)) IN K : Coh reg ( B ) ⟶ R K reg , IN E : Coh reg ( B ) ⟶ Rep sm perf , reg ( Int ( E )) are strictly unital A ∞ A_\infty A ∞ quasi-equivalences. For each target object there is a bounded finite-projective regular cohesive preimage and an actual closed degree-zero homotopy equivalence from its integration to that object.
Proof. Quasi-full faithfulness is Theorem B.1. It remains to construct a preimage of an arbitrary regular raw representation R R R of K K K . Lemma C.2 gives a minimal, edge-trivialized representation on a fixed finite graded space H H H . Write it as e + b e+b e + b .
Let
N = ⨁ q < 0 End R q ( H ) , F ℓ N = ⨁ q ≤ − ℓ End R q ( H ) . N=\bigoplus_{q<0}\operatorname{End}^{q}_{\mathbb{R}}(H),\qquad
F^\ell N=\bigoplus_{q\leq-\ell}\operatorname{End}^{q}_{\mathbb{R}}(H). N = q < 0 ⨁ End R q ( H ) , F ℓ N = q ≤ − ℓ ⨁ End R q ( H ) . This is a finite multiplicative nilpotent filtration. Because r 0 = 0 r_0=0 r 0 = 0 , r 1 = 1 r_1=1 r 1 = 1 , every remaining structure cochain lies in C K ∙ ⊗ N C^\bullet_K\otimes N C K ∙ ⊗ N , in total degree one. On the source use B ⊗ N B\otimes N B ⊗ N . A total-degree-one element there has CE degree at least two.
Restrict the actual I N K \mathrm{IN}^{K} IN K to the endomorphism algebra of the trivial rank-one coefficient and take its graded matrix extension (6.3)–(6.4). Its A ∞ A_\infty A ∞ property is the forms-first matrix extension already fixed in Theorem 6.1, using [AS13, Proposition A.17]. It gives the A ∞ A_\infty A ∞ map
F N : B ⊗ N ⇝ C K ∙ ⊗ N . F^{N}:B\otimes N\rightsquigarrow C^\bullet_K\otimes N. F N : B ⊗ N ⇝ C K ∙ ⊗ N . The matrix products make its Taylor filtration arity-additive. In layer ℓ \ell ℓ , its first component is
I ⊗ 1 End − ℓ ( H ) , I\otimes1_{\operatorname{End}^{-\ell}(H)}, I ⊗ 1 End − ℓ ( H ) , with that matrix space placed in internal degree − ℓ -\ell − ℓ . It is a quasi-isomorphism by Theorem B.11. Only tensoring with a finite-dimensional real vector space occurs here.
Lemma C.3 produces an MC element a ∈ ( B ⊗ N ) 1 a \in(B \otimes N)^1 a ∈ ( B ⊗ N ) 1 and a finite gauge q q q with b q = F ∗ N ( a ) b^q = F_*^N(a) b q = F ∗ N ( a ) . Let P = ( H ⊗ B , Q + a ) P = (H \otimes B,Q+a) P = ( H ⊗ B , Q + a ) . It is bounded finite projective and regular, with zero internal differential and zero one-form part.
To identify the actual object, use Γ ‾ n = s Γ n ( s − 1 ) ⊗ n \overline{\Gamma}_n=s\Gamma_n(s^{-1})^{\otimes n} Γ n = s Γ n ( s − 1 ) ⊗ n . The tensor sign is
( s − 1 ) ⊗ n ( s a 1 ⊗ ⋯ ⊗ s a n ) = ( − 1 ) ∑ i ( n − i ) ( ∣ a i ∣ − 1 ) a 1 ⊗ ⋯ ⊗ a n . (s^{-1})^{\otimes n}(sa_1\otimes\cdots\otimes sa_n)=(-1)^{\sum_i(n-i)(|a_i|-1)}a_1\otimes\cdots\otimes a_n. ( s − 1 ) ⊗ n ( s a 1 ⊗ ⋯ ⊗ s a n ) = ( − 1 ) ∑ i ( n − i ) ( ∣ a i ∣ − 1 ) a 1 ⊗ ⋯ ⊗ a n .
Every MC input has total degree ∣ a i ∣ = 1 |a_i|=1 ∣ a i ∣ = 1 , including its matrix degree, so this exponent is zero. Therefore F ∗ N ( a ) = ∑ n Ψ ∗ Γ ‾ n ( a , … , a ) F_*^N(a)=\sum_n\Psi^*\overline{\Gamma}_n(a,\ldots,a) F ∗ N ( a ) = ∑ n Ψ ∗ Γ n ( a , … , a ) literally, with the unsuspended sign (6.4). No extra ( − 1 ) n ( n − 1 ) / 2 (-1)^{n(n-1)/2} ( − 1 ) n ( n − 1 ) /2 occurs. Since Ψ ∗ \Psi^* Ψ ∗ is a strict map of normalized cochain DGAs, (6.6) gives
IN K ( P ) = e + F ∗ N ( a ) . \operatorname{IN}^{K}(P)=e+F_*^N(a). IN K ( P ) = e + F ∗ N ( a ) .
No second target conversion T T T is applied. In particular the scalar component still sends η \eta η to − h -h − h , as fixed by (3.7). Every object word has strictly decreasing internal degree, so these particular twisting sums are finite.
The gauge q q q is an actual normalized raw representation isomorphism toward the edge-trivialized target, by (C.3). Every degree-zero correction in F ℓ ( C K ⊗ N ) F^\ell(C_K\otimes N) F ℓ ( C K ⊗ N ) has positive cochain degree, so q 0 = 1 q_0=1 q 0 = 1 . Let g g g denote the ordinary vertex gauge from that object to the minimal representation, and let i ′ i' i ′ be the perturbed inclusion from the minimal representation to R R R . Then
ϵ R = i ′ g q : IN K ( P ) ⟶ R \epsilon_R=i'gq:\operatorname{IN}^{K}(P)\longrightarrow R ϵ R = i ′ g q : IN K ( P ) ⟶ R
is closed, degree zero, and a homotopy equivalence. The two gauge factors are strictly invertible; i ′ i' i ′ is the homotopy equivalence supplied by finite transfer. This proves essential surjectivity on H 0 H^0 H 0 .
Finally Ψ ∗ \Psi^* Ψ ∗ is the actual raw dg quasi-equivalence of Corollary 5.4, and IN K = Ψ ∗ IN E \operatorname{IN}^{K}=\Psi^*\operatorname{IN}_{E} IN K = Ψ ∗ IN E . More explicitly, that corollary supplies closed invertible gauges U Z : Z → Φ ∗ Ψ ∗ Z U_Z:Z\to\Phi^*\Psi^*Z U Z : Z → Φ ∗ Ψ ∗ Z on the full MC target. For a reconstructed map v : Ψ ∗ IN E ( P ) → Ψ ∗ R v:\Psi^*\operatorname{IN}_{E}(P)\to\Psi^*R v : Ψ ∗ IN E ( P ) → Ψ ∗ R , the actual full-MC map is
U R − 1 Φ ∗ ( v ) U IN E ( P ) : IN E ( P ) ⟶ R . U_R^{-1}\Phi^*(v)U_{\operatorname{IN}_{E}(P)}:\operatorname{IN}_{E}(P)\longrightarrow R. U R − 1 Φ ∗ ( v ) U IN E ( P ) : IN E ( P ) ⟶ R .
It is a closed degree-zero homotopy equivalence. The Hom quasi-isomorphisms likewise transfer through Ψ ∗ \Psi^* Ψ ∗ , proving both assertions for the intrinsic functor. □
Corollary C.5 (Other global affine resolutions). Let E ′ E' E ′ be another supplied strict global geometric resolution of f = ⟨ x ∂ x , y ∂ x ⟩ \mathfrak{f}=\langle x\partial_x,y\partial_x\rangle f = ⟨ x ∂ x , y ∂ x ⟩ , with E , E ′ ∈ R f E,E'\in\mathcal{R}_{\mathfrak{f}} E , E ′ ∈ R f , under the fixed-base comparison hypotheses of Theorem 2.1 . Then IN E ′ \operatorname{IN}_{E'} IN E ′ is a quasi-equivalence on the regular cohesive and raw perfect representation categories.
Proof. By Theorem 2.2 , choose a comparison F : E ⇝ E ′ F:E\rightsquigarrow E' F : E ⇝ E ′ commuting with the resolution augmentations. Both complexes resolve f \mathfrak{f} f , so F ∈ W 1 F\in W_1 F ∈ W 1 . Its cohesive pullback and actual full-MC representation pullback are dg quasi-equivalences on the regular classes. The naturality of IN \operatorname{IN} IN identifies the resulting square with the affine integration square. Quasi-full faithfulness and essential surjectivity transfer through these two equivalences. □
its stated global resolutions, using the whole-plane section Ψ \Psi Ψ . Theorem 10.4 separately proves RH for nonregular coefficients on every ordinary open by a different comparison mechanism and separately chosen range-preserving data. The regular tools here remain inputs to the vertical-family theorem. Effective descent through a refined holonomy quotient is a further question, as discussed in § 14 .
Acknowledgments
This paper grew out of part of the author’s Ph.D. thesis work. The author is grateful to Jonathan Block for his advice and support, to David White for discussions on the homotopy theory of algebras over operads, and to Sylvain Lavau and Jim Stasheff for discussions on L ∞ L_\infty L ∞ -algebroids and singular foliations.
AI assistance
This manuscript was developed with substantial assistance from GitHub Copilot CLI and Azure OpenAI GPT-5.6 Sol, including mathematical proof exploration, drafting and revision of the exposition, and automated consistency checks. Preparation of the present revision also used Azure OpenAI GPT-6 Astra.
References