OctoInformBeta

OpenAI preprint, September 2026 ยท 4. Partial cylinders and elementary expansions

ยง4 (proof of Theorem 1.1): descending induction and removal of the cutoff: At sufficiently high cutoffs, reflected elementary expansions satisfy exact boundary tests that descend through the unit

Expansion contraction holds at every threshold.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

An elementary expansion adjoins a cell along its source face. The boundary test asks whether a cell in the target, with a prescribed boundary from the source, can be joined to the image of a source cell having exactly that boundary.โœฆ

The globe category ๐”พ\mathbb G has one object in each natural-number dimension, identity maps, and source and target faces from dimension mm to dimension nn whenever m<nm<n. A globular presheaf XX on ๐”พ\mathbb G is a contravariant set-valued functor on ๐”พ\mathbb G.โœฆ The representable disks in the category of globular presheaves on ๐”พ\mathbb G are the Yoneda images of the globes, and their disk faces are the Yoneda images of the globe face maps. Write GjG_j for the globe arity in dimension jj.โœฆ

A table of dimensions specifies a finite globular sum SS, with a distinguished last globe of dimension โ„“\ell. The realization of a finite globular sum SS with a distinguished last globe is the iterated pushout of its representable disks, with the distinguished inclusion of the last globe.โœฆ An attachment to a finite globular sum SS whose last globe has dimension โ„“\ell is a gluing that adds a globe of dimension hh along its source bb-face and the target bb-face of the last globe of SS, where b<โ„“b<\ell and b<hb<h.โœฆ The elementary gluing is the pushout in (1); its maps from SS and GhG_h are the previous-sum inclusion and the attached-globe inclusion, and the map Gbโ†’SG_b\to S is the seam inclusion. SโˆGbGh.S\mathbin{\amalg}_{G_b}G_h.(1) The height of SS is the greatest dimension among its globes.โœฆ The globe inclusion into finite globular sums sends each globe of dimension jj to its globe arity GjG_j and each face to the corresponding representable disk face.

A globular theory CC is a category with the finite globular sums as objects and an arity functor from their category that is the identity on objects and preserves the elementary pushouts.โœฆ A morphism F:Cโ†’Cโ€ฒF:C\to C' of globular theories is a functor that fixes arities and preserves operations under composition, identities, and the structural arity maps.โœฆ Operations Gjโ†’SG_j\to S are parallel when their source and target restrictions agree; all pairs are parallel in dimension zero.โœฆ An admissible pair Gjโ‡‰SG_j\rightrightarrows S in a globular theory is a parallel pair of operations with height⁡(S)โ‰คj+1\operatorname{height}(S)\le j+1.โœฆ For a morphism F:Cโ†’Cโ€ฒF:C\to C' of globular theories and an admissible pair s,t:Gjโ‡‰Ss,t:G_j\rightrightarrows S in CC, a filler in Cโ€ฒC' is an operation z:Gj+1โ†’Sz:G_{j+1}\to S whose source and target restrictions are F(s)F(s) and F(t)F(t).

A cellular presentation PP of a globular theory CC is a sequence beginning at an initial theory, with each transition freely adjoining fillers for a specified family of admissible pairs, and with CC as its colimit.โœฆ Freely adjoining fillers means that a morphism out of the previous theory, together with chosen fillers of those pairs, extends uniquely across the transition preserving those choices.โœฆ A coherator CC is a globular theory that admits a cellular presentation and has a filler for every admissible pair.โœฆ

A model XX of CC is a presheaf on CC carrying the elementary globular sums to pullbacks.โœฆ The globe inclusion into CC sends globes and faces to their structural arity images.โœฆ Restriction along this inclusion gives the underlying globular presheaf of XX.

A globular set XX is a family of sets XjX_j with source and target maps s,t:Xj+1โ†’Xjs,t:X_{j+1}\to X_j satisfying ss=stss=st and ts=ttts=tt.โœฆ A globular map f:Xโ†’Yf:X\to Y between globular sets is a family of cell maps commuting with source and target; composition of globular maps is dimensionwise. A model morphism f:Xโ†’Yf:X\to Y induces such a map through its components fjf_j on globes.โœฆ The underlying globular set of a globular presheaf PP has jj-cells P(j)P(j), with source and target given by the images of the source and target faces. A morphism f:Pโ†’Qf:P\to Q of globular presheaves induces a globular map between their underlying globular sets whose map on jj-cells is the component of ff at the jj-globe.

Two cells are parallel when they have the same source and target, with every pair of zero-dimensional cells parallel.โœฆ The boundary sets and boundary maps are given by (2), with โˆ‚0\partial_0 the unique map to the singleton. A globular map ff induces a boundary map f*f_* by applying ff to both entries, and by the unique map in dimension zero.โœฆ B0(X)={*},Bj+1(X)={(u,v)โˆˆXjร—Xjโˆฃuโˆฅv},โˆ‚j+1x=(sx,tx).B_0(X)=\{*\},\qquad B_{j+1}(X)=\{(u,v)\in X_j\times X_j\mid u\parallel v\},\qquad \partial_{j+1}x=(sx,tx).(2) A globular set XX has the boundary filler property when every โˆ‚j:Xjโ†’Bj(X)\partial_j:X_j\to B_j(X) is surjective. Cells u,vโˆˆXju,v\in X_j are joined, written uโคณXvu\leadsto_X v, when there is an (j+1)(j+1)-cell with source uu and target vv.โœฆ A globular map f:Xโ†’Yf:X\to Y satisfies the exact boundary test, written EB⁡(f)\operatorname{EB}(f), when it satisfies (3). For every j,ฮฒโˆˆBj(X),yโˆˆYj with โˆ‚jy=f*ฮฒ,there is xโˆˆXj with โˆ‚jx=ฮฒandfjxโคณYy.\begin{gathered} \text{For every }j,\ \beta\in B_j(X),\ y\in Y_j \text{ with }\partial_jy=f_*\beta,\\ \text{there is }x\in X_j\text{ with } \partial_jx=\beta\quad\text{and}\quad f_jx\leadsto_Y y. \end{gathered}(3)โœฆ

Disks, Cutoffs, and Lifting Classes

The free model C(โˆ’,S)C(-,S) on a finite globular sum SS is the corresponding representable model.โœฆ The free disk DjD_j is the free model on GjG_j.โœฆ The canonical boundary โˆ‚Dj\partial D_j of the free disk DjD_j, with inclusion Ij:โˆ‚Djโ†’DjI_j:\partial D_j\to D_j, represents boundaries; the canonical boundary construction starts at the empty boundary in dimension zero and proceeds by successor boundaries. The disk Yoneda correspondence and the boundary representation are the natural bijections in (4), under which restriction along IjI_j corresponds to taking a cellโ€™s boundary. The source inclusion Jj:Djโ†’Dj+1J_j:D_j\to D_{j+1} is induced by the source face Gjโ†’Gj+1G_j\to G_{j+1}.โœฆ Hom⁡(Dj,X)โ‰…Xj,Hom⁡(โˆ‚Dj,X)โ‰…Bj(X),Ij*โ†”โˆ‚j.\operatorname{Hom}(D_j,X)\cong X_j,\qquad \operatorname{Hom}(\partial D_j,X)\cong B_j(X),\qquad I_j^*\longleftrightarrow\partial_j.(4)

A boundary attachment Xโ†’YX\to Y is a model morphism that freely adjoins a family of cells along specified boundaries in XX: maps from YY to any model ZZ correspond bijectively to maps Xโ†’ZX\to Z together with chosen images of the attached cells having the prescribed image boundaries.โœฆ A cellular model XX of CC is a model admitting a continuous transfinite presentation from the initial model, with boundary attachments at successor stages and XX as colimit.โœฆ

A qq-coskeletal model XX of CC is a model for which โˆ‚j\partial_j is bijective for every j>qj>q.โœฆ Write ๐’œq\mathcal A_q for the full category of these models.โœฆ A coskeletal reflection LqXL_qX of a model XX is an object of ๐’œq\mathcal A_q with a unit ฮทX:Xโ†’LqX\eta_X:X\to L_qX for which precomposition gives the bijection (5) for every Yโˆˆ๐’œqY\in\mathcal A_q. ฮฆX,Y:Hom⁡๐’œq(LqX,Y)โ†’โ‰…Hom⁡C-Mod(X,Y),hโ†ฆhโˆ˜ฮทX.\Phi_{X,Y}:\operatorname{Hom}_{\mathcal A_q}(L_qX,Y) \xrightarrow{\ \cong\ }\operatorname{Hom}_{C\text{-}\mathrm{Mod}}(X,Y), \qquad h\longmapsto h\circ\eta_X.(5)โœฆ The reflected disks and boundaries are Djq=LqDjD_j^q=L_qD_j and โˆ‚Djq=Lq(โˆ‚Dj)\partial D_j^q=L_q(\partial D_j), with reflected inclusions Ijq=LqIjI_j^q=L_qI_j and Jjq=LqJjJ_j^q=L_qJ_j. For a coherator CC, the coskeletal reflection functor Lq:C-Modโ†’๐’œqL_q:C\text{-}\mathrm{Mod}\to\mathcal A_q is the left adjoint to the inclusion of ๐’œq\mathcal A_q into the category of CC-models, with the hom-set bijection in (5).โœฆ

For a class ๐’ฆ\mathcal K of maps, ๐’ฆโŸ‚\mathcal K^\perp denotes its right lifting class, and โŸ‚๐’ฆ\mathord{}^\perp\mathcal K its left lifting class. The generating boundary and source classes at threshold rr, and the resulting left lifting class, are those of (6). โ„q,r={Ijqโˆฃjโ‰ฅr},๐’ฅq,r={Jjqโˆฃjโ‰ฅr},โ„’q,r=โŸ‚(๐’ฅq,rโŸ‚).\mathcal I_{q,r}=\{I_j^q\mid j\ge r\},\qquad \mathcal J_{q,r}=\{J_j^q\mid j\ge r\},\qquad \mathcal L_{q,r}={}^\perp(\mathcal J_{q,r}^{\perp}).(6) A cutoff model Xโˆˆ๐’œqX\in\mathcal A_q is cofibrant when every map Xโ†’QX\to Q lifts along every map Pโ†’QP\to Q in โ„q,0โŸ‚\mathcal I_{q,0}^{\perp}.โœฆ The expansion contraction principle E(q,r)E(q,r) asserts that if a map i:Xโ†’Yi:X\to Y belongs to โ„’q,r\mathcal L_{q,r} and XX is cofibrant and has the boundary filler property, then YY has the boundary filler property.โœฆ

Cutoff path data for Yโˆˆ๐’œqY\in\mathcal A_q at threshold rr consist of an object Pโˆˆ๐’œqP\in\mathcal A_q and maps p0,p1:Pโ†’Yp_0,p_1:P\to Y in โ„q,0โŸ‚\mathcal I_{q,0}^{\perp} with the following joint extension property. For kโ‰ฅrk\ge r, maps t:Dkqโ†’Pt:D_k^q\to P and u0,u1:Dk+1qโ†’Yu_0,u_1:D_{k+1}^q\to Y satisfying pฯตโˆ˜t=uฯตโˆ˜Jkqp_\epsilon\circ t=u_\epsilon\circ J_k^q for ฯต=0,1\epsilon=0,1 admit an extension as in (7). โˆƒl:Dk+1qโ†’P,lโˆ˜Jkq=t,p0โˆ˜l=u0,p1โˆ˜l=u1.\exists\,l:D_{k+1}^q\to P,\qquad l\circ J_k^q=t,\qquad p_0\circ l=u_0,\qquad p_1\circ l=u_1.(7)

The Claim

Theorem 1. Let CC be a coherator. For every cutoff qq and source threshold rr, E(q,r).E(q,r).(8)โœฆ

Suppose i:Xโ†’Yi:X\to Y belongs to โ„’q,r\mathcal L_{q,r} in ๐’œq\mathcal A_q, XX is cofibrant, and there is a retraction z:Yโ†’Xz:Y\to X with zโˆ˜i=1Xz\circ i=1_X. If YY admits cutoff path data at threshold rr, then EB⁡(i).\operatorname{EB}(i).(9)โœฆ

Let XX be a cellular CC-model and nโ‰คqn\le q. Suppose the maps a:Dnโ†’Xa:D_n\to X, Jn:Dnโ†’Dn+1J_n:D_n\to D_{n+1}, i:Xโ†’Yi:X\to Y, and b:Dn+1โ†’Yb:D_{n+1}\to Y form a pushout square. The reflected map satisfies EB⁡(Lqi).\operatorname{EB}(L_qi).(10)โœฆ

Consider globular sets U,V,R,TU,V,R,T and maps i:Uโ†’Vi:U\to V, j:Rโ†’Tj:R\to T, a:Uโ†’Ra:U\to R, and b:Vโ†’Tb:V\to T with bโˆ˜i=jโˆ˜ab\circ i=j\circ a. Suppose jj satisfies the exact boundary test. Fix a dimension dd such that aka_k is bijective for every kโ‰คdk\le d, bdb_d is injective, and bd+1b_{d+1} is surjective. For every boundary ฮฒโˆˆBd(U)\beta\in B_d(U) and cell yโˆˆVdy\in V_d with โˆ‚dy=i*ฮฒ\partial_dy=i_*\beta, โˆƒxโˆˆUd,โˆ‚dx=ฮฒandidxโคณVy.\exists\,x\in U_d,\qquad \partial_dx=\beta\quad\text{and}\quad i_dx\leadsto_V y.(11)โœฆ

In the paper, the result reads: the path construction establishes E(q,n)E(q,n) by descending induction, and for an elementary expansion of ordinary models in dimension nn and a fixed-boundary test in dimension dd, choosing qโ‰ฅmax⁡{n+1,d+1}q\ge\max\{n+1,d+1\}, reflecting the actual pushout, applying the test at the cutoff, and using the low-dimensional bijections of the unit recovers the source dd-cell and the target (d+1)(d+1)-cell before reflection.

What Is Checked

The assertions (8), (9), and (10) are stated over a coherator CC, with lifting cofibrancy in ๐’œq\mathcal A_q and transfinite cellularity for the ordinary expansion; (11) concerns globular sets and the specified low-dimensional comparison maps.

Every model XX admits a coskeletal reflection in ๐’œq\mathcal A_q whose unit (ฮทX)k(\eta_X)_k is bijective for kโ‰คqk\le q.โœฆ For Y,Yโ€ฒโˆˆ๐’œqY,Y'\in\mathcal A_q, a map g:Yโ†’Yโ€ฒg:Y\to Y', and a map h:LqXโ†’Yh:L_qX\to Y, the adjunction bijection is natural as in (12). ฮฆX,Yโ€ฒ(gโˆ˜h)=gโˆ˜ฮฆX,Y(h).\Phi_{X,Y'}(g\circ h)=g\circ\Phi_{X,Y}(h).(12)โœฆ