OctoInformBeta

OpenAI preprint, September 2026 · §5. Local replacement and the number of zero strata

Proposition 5.1: Expected counts in original simplices for invariant finite measures and near-minimal equivariant filtrations

Large dimension permits sparse zero strata.

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

The geometric setting of the Bass trace conjecture includes equivariant filtrations of a simplicial realization. The quantity at issue is the expected number of points of the lowest marked stratum in each original top-dimensional simplex.

Let HH act on a vertex set YY and a topological parameter space XX. Let Γ\Gamma be a connected graph on YY, let KK be an HH-invariant abstract simplicial complex on YY, and fix a nonnegative integer nn as the dimension parameter. The graph metric dΓd_\Gamma is shortest-path distance.✦ Write V=ℝ(Y)V=\mathbb R^{(Y)} for the finitely supported real vertex vectors and eye_y for the unit vector at yy. Equip VV with the norm induced by its coordinate inclusion in ℓ2(Y)\ell^2(Y) and with the corresponding Borel structure. Write 𝒦\mathcal K for the standard geometric realization complex and |K||K| for its underlying space.

A finitely supported probability measure pp on YY is a finitely supported coefficient function with nonnegative values and total mass one.✦ For a real-valued function ff on YY, the expectation of ff under pp and its expectation difference from another finitely supported probability measure qq on YY are 𝔼p[f]=∑y∈Ypyf(y),Δp,qf=𝔼p[f]−𝔼q[f].\mathbb E_p[f]=\sum_{y\in Y}p_y f(y),\qquad \Delta_{p,q}f=\mathbb E_p[f]-\mathbb E_q[f].(1) The dual transportation distance uses the expectation difference in (1). For the graph metric, its normalized version is W1(p,q)=supLip⁡dΓ(f)≤1Δp,qf,ρn(p,q)={W1(p,q)/n,n>0,0,n=0.W_1(p,q)=\sup_{\operatorname{Lip}_{d_\Gamma}(f)\leq1}\Delta_{p,q}f, \qquad \rho_n(p,q)= \begin{cases} W_1(p,q)/\sqrt n,&n>0,\\ 0,&n=0. \end{cases}(2) For v∈|K|v\in|K|, the barycentric probability measure pvp_v has coefficient vector vv and support a face of KK.✦ The realization distance on VV is ρ(v,w)={ρn(pv,pw),v,w∈|K|,0,otherwise,\rho(v,w)= \begin{cases} \rho_n(p_v,p_w),&v,w\in|K|,\\ 0,&\text{otherwise}, \end{cases}(3)✦ where the normalization is that of (2).

Marked Subdivisions and Separating Filtrations

For a finite set ss of geometric vertices, write [s]=conv⁡(s)[s]=\operatorname{conv}(s). The vertex restriction B|AB|_A of a geometric complex BB to a subset A⊆VA\subseteq V is the geometric complex consisting of the faces of BB whose vertices lie in AA.✦ The geometric restriction B[A]B[A] of a geometric complex BB to a subset A⊆VA\subseteq V is the geometric complex consisting of the faces of BB whose convex hull lies in AA.✦ Simplicial compatibility of BB with AA means [s]∩A=[s∩A](s a face of B).[s]\cap A=[s\cap A]\qquad(s\text{ a face of }B).(4)✦ A simplicial subdivision TT of 𝒦\mathcal K is a geometric complex with underlying space |K||K|, each of whose faces lies in an original simplex, which satisfies (4) for each original simplex and has finitely many faces in the vertex restriction to that simplex.✦ A complex CC refines a complex BB if CC is the complex of a simplicial subdivision of BB.✦

A locally constant subdivision family TT over XX is a family of simplicial subdivisions T(x)T(x) of 𝒦\mathcal K, indexed by x∈Xx\in X, with the face set of the vertex restriction of T(x)T(x) to each original simplex locally constant in xx.✦ A marked subset ZZ of a geometric complex BB is a union of convex hulls of faces of BB.✦ A marked subdivision family for TT is a family Z(x)Z(x) marked in T(x)T(x) such that the face set of T(x)[Z(x)]|[s]T(x)[Z(x)]|_{[s]} is locally constant for each original geometric face ss.✦

The vertex action extends to the linear reindexing h*h_* on VV. An equivariant marked subdivision pair 𝒫=(T,Z)\mathcal P=(T,Z) consists of a subdivision family and a marked family satisfying (h*v)y=vh−1y,T(hx)=h*T(x),Z(hx)=h*Z(x)(h∈H,x∈X).(h_*v)_y=v_{h^{-1}y},\qquad T(hx)=h_*T(x),\qquad Z(hx)=h_*Z(x) \quad(h\in H,\ x\in X).(5) Here the middle equality in (5) is equality of face sets. Subdivision equivariance of TT means that h*h_* sends every face of T(x)T(x) to a face of T(hx)T(hx) for every h∈Hh\in H and x∈Xx\in X.✦ Marked subset equivariance of ZZ means that h*vh_*v belongs to Z(hx)Z(hx) whenever v∈Z(x)v\in Z(x), for every h∈Hh\in H and x∈Xx\in X.✦

A subset P⊆VP\subseteq V has component diameter at most bb for ρ\rho if ρ(v,w)≤b\rho(v,w)\leq b for every pair of points in every connected subset of PP.✦ For geometric complexes BB and CC and subsets P,Z⊆VP,Z\subseteq V, dimension drop from (B,P)(B,P) to (C,Z)(C,Z) means that every face FF of BB with [F]⊆P[F]\subseteq P and every face EE of CC with [E]⊆[F]∩Z[E]\subseteq[F]\cap Z satisfy #E<#F\#E<\#F.✦

A separating filtration QQ of length nn and radius RR is a sequence of equivariant marked pairs (Ti,Zi)(T_i,Z_i), indexed by nonnegative integers, with Tn(x)=𝒦,Zn(x)=|K|,Ti(x) refines Ti+1(x),Zi(x)⊆Zi+1(x)(i<n).T_n(x)=\mathcal K,\qquad Z_n(x)=|K|,\qquad T_i(x)\text{ refines }T_{i+1}(x),\qquad Z_i(x)\subseteq Z_{i+1}(x)\quad(i<n).(6) In addition to (6), there is dimension drop from (Ti+1(x),Zi+1(x))(T_{i+1}(x),Z_{i+1}(x)) to (Ti(x),Zi(x))(T_i(x),Z_i(x)), and Zi+1(x)\Zi(x)Z_{i+1}(x)\setminus Z_i(x) has component diameter at most 2R2R.

An admissible step from a marked pair (T,Z)(T,Z) is a refinement T′T' with a marked family Z′⊆ZZ'\subseteq Z, in which ZZ remains marked, with dimension drop from (T,Z)(T,Z) to (T′,Z′)(T',Z') and component diameter at most 2R2R for Z\Z′Z\setminus Z'.✦ An equivariant admissible step AA from a marked pair (T,Z)(T,Z) is an admissible step whose new subdivision and marking satisfy (5); its resulting marked pair is denoted AoutA_{\mathrm{out}}. For real-valued costs cic_i on marked pairs, a separating filtration Q=(Ti,Zi)Q=(T_i,Z_i) is ε\varepsilon-near-minimal, written NM⁡ε(Q;c)\operatorname{NM}_\varepsilon(Q;c), if ci(Ti,Zi)≤ci(Aout)+εfor every i<n and every equivariant admissible step A from (Ti+1,Zi+1).c_i(T_i,Z_i)\leq c_i(A_{\mathrm{out}})+\varepsilon \quad\text{for every }i<n\text{ and every equivariant admissible step }A \text{ from }(T_{i+1},Z_{i+1}).(7)✦

Simplex Areas and Expected Counts

An original top-dimensional simplex σ∈Kn\sigma\in K_n is a face of KK with n+1n+1 vertices.✦ The action on these simplices is hσ={hy:y∈σ}h\sigma=\{hy:y\in\sigma\}.✦ Choose an enumeration y0,…,yny_0,\ldots,y_n of the vertices of σ\sigma. For an original top-dimensional simplex σ\sigma with enumerated vertices y0,…,yny_0,\ldots,y_n, the simplex graph Γσ\Gamma_\sigma is the graph induced by Γ\Gamma in this enumeration✦, the coordinate map is ησ(v)=(vy0,…,vyn)\eta_\sigma(v)=(v_{y_0},\ldots,v_{y_n})✦, and the original geometric simplex and its intrinsic interior are |σ|=conv⁡{ey:y∈σ},Wσ=intrint⁡|σ|.|\sigma|=\operatorname{conv}\{e_y:y\in\sigma\},\qquad W_\sigma=\operatorname{intrint}|\sigma|.(8)✦ The standard simplex vertices associated with σ\sigma are the unit coordinate vectors ey∈Ve_y\in V for y∈σy\in\sigma.✦

A spanning tree TT of a graph Λ\Lambda on {0,…,n}\{0,\ldots,n\} is a tree on the same vertices whose edges belong to Λ\Lambda; write 𝒯(Λ)\mathcal T(\Lambda) for the set of spanning trees of Λ\Lambda.✦ Let eie_i denote the iith unit vector of ℤn\mathbb Z^n, and put z0=0z_0=0 and zi=ei∈ℤnz_i=e_i\in\mathbb Z^n for i>0i>0. The integer reduced incidence matrix BTB_T of a spanning tree TT with enumerated edges and chosen endpoint orders (ak,bk)(a_k,b_k) is the integer matrix with columns zak−zbkz_{a_k}-z_{b_k}.✦ Regard this matrix BTB_T over ℝ\mathbb R for the tree parametrization β(t)=(1−∑i=1nti,t1,…,tn),ϕT(u)=β(BTu)(u∈ℝn).\beta(t)=\left(1-\sum_{i=1}^nt_i,t_1,\ldots,t_n\right),\qquad \phi_T(u)=\beta(B_Tu)\quad(u\in\mathbb R^n).(9) For a nonnegative integer jj and a spanning tree TT, the tree area aj,T(B)a_{j,T}(B) of a set B⊆ℝn+1B\subseteq\mathbb R^{n+1} is the jj-dimensional Euclidean Hausdorff measure of its inverse image under the map ϕT\phi_T in (9).✦ The averaged simplex area ajΛ(B)a_j^\Lambda(B) of a set B⊆ℝn+1B\subseteq\mathbb R^{n+1} is the average of its tree areas over the spanning trees of Λ\Lambda: aj,T(B)=ℋEj(ϕT−1(B)),ajΛ(B)=1#𝒯(Λ)∑T∈𝒯(Λ)aj,T(B).a_{j,T}(B)=\mathcal H_E^j(\phi_T^{-1}(B)),\qquad a_j^\Lambda(B)=\frac{1}{\#\mathcal T(\Lambda)} \sum_{T\in\mathcal T(\Lambda)}a_{j,T}(B).(10) These measures take values in [0,∞][0,\infty]. For a marked subdivision family ZZ, a linear coordinate map η:V→ℝn+1\eta:V\longrightarrow\mathbb R^{n+1}, a region W⊆VW\subseteq V, and a degree jj, the local marked simplex area at x∈Xx\in X is the averaged simplex area in degree jj of η(Z(x)∩W)\eta(Z(x)\cap W).✦ The original simplex marked area uses the particular interior in (8): ajσ(𝒫,x)=ajΓσ(ησ(Z(x)∩Wσ)).a_j^\sigma(\mathcal P,x)= a_j^{\Gamma_\sigma}\bigl(\eta_\sigma(Z(x)\cap W_\sigma)\bigr).(11)✦ The rooted vertex coordinate of vertex ii is the vector zi∈ℤnz_i\in\mathbb Z^n: the root has coordinate zero, and each nonroot vertex has its corresponding unit coordinate vector.✦ An edge enumeration of a spanning tree TT on {0,…,n}\{0,\ldots,n\} is a bijection from {1,…,n}\{1,\ldots,n\} onto its edge set.✦ The barycentric parametrization β:ℝn→ℝn+1\beta:\mathbb R^n\longrightarrow\mathbb R^{n+1} in (9) adjoins a zeroth coordinate equal to one minus the sum of the given coordinates.✦ The real reduced incidence matrix BTB_T of a spanning tree TT is obtained from its integer reduced incidence matrix by replacing each integer entry by the corresponding real number.✦

Let μ\mu be a measure on XX, and let ℛ⊆Kn\mathcal R\subseteq K_n be finite. Write mσm_\sigma for the cardinality of the stabilizer of σ\sigma, assigned value zero if that stabilizer is infinite. For t∈[0,∞]t\in[0,\infty], let tℝt_{\mathbb R} be its real value when finite and zero when t=∞t=\infty. The stabilizer-normalized expected area is 𝒜jℛ(𝒫)=∑σ∈ℛmσ−1∫X∞ajσ(𝒫,x)dμ(x),Jjℛ(𝒫)=(𝒜jℛ(𝒫))ℝ.\mathcal A_j^{\mathcal R}(\mathcal P)= \sum_{\sigma\in\mathcal R}m_\sigma^{-1} \int_X^{\infty}a_j^\sigma(\mathcal P,x)\,d\mu(x),\qquad J_j^{\mathcal R}(\mathcal P)= \bigl(\mathcal A_j^{\mathcal R}(\mathcal P)\bigr)_{\mathbb R}.(12)✦ Write ℛ0\mathcal R_0 for the finite set formed by choosing one representative from each top-simplex orbit.✦ The costs in (7) will be those of (12).

The zero-stratum count and its two expected values are Nσ(Q,x)=#(Z0(x)∩|σ|)∈[0,∞],Iσ(Q)=∫X(Nσ(Q,x))ℝdμ(x),Iˆσ(Q)=(∫X∞Nσ(Q,x)dμ(x))ℝ.\begin{aligned} N_\sigma(Q,x)&=\#\bigl(Z_0(x)\cap|\sigma|\bigr)\in[0,\infty],\\ I_\sigma(Q)&=\int_X\bigl(N_\sigma(Q,x)\bigr)_{\mathbb R}\,d\mu(x),\\ \widehat I_\sigma(Q)&= \left(\int_X^{\infty}N_\sigma(Q,x)\,d\mu(x)\right)_{\mathbb R}. \end{aligned}(13) The dimension threshold and the radius associated with a degree bound DD are ν(D)=(4D3+D+3)2,RD=4D2.\nu(D)=(4D^3+D+3)^2,\qquad R_D=4D^2.(14) A finite orbit-covering face family is a finite collection FF of faces of KK such that every face of KK is hτh\tau for some h∈Hh\in H and τ∈F\tau\in F.

The Claim

Theorem 1. Suppose KK has a finite orbit-covering face family. For any measure μ\mu, any finite ℛ⊆Kn\mathcal R\subseteq K_n, and positive RR and ε\varepsilon, there is a separating filtration QQ of radius RR such that NM⁡ε(Q;Jℛ).\operatorname{NM}_\varepsilon(Q;J^{\mathcal R}).(15)✦

Let μ\mu be finite and invariant under the action of HH. Suppose D≥2D\geq2, every vertex of Γ\Gamma has at most DD neighbors, and n≥ν(D)n\geq\nu(D), with ν\nu and RDR_D as in (14). For each σ∈Kn\sigma\in K_n there is a constant Cσ≥0C_\sigma\geq0 such that every ε≥0\varepsilon\geq0 and every separating filtration QQ of radius RDR_D satisfying NM⁡ε(Q;Jℛ0)\operatorname{NM}_\varepsilon(Q;J^{\mathcal R_0}) obey Iˆσ(Q)≤Cσε.\widehat I_\sigma(Q)\leq C_\sigma\varepsilon.(16)✦

Under these measure, degree and dimension assumptions, suppose also that KK has a finite orbit-covering face family. There are constants Cσ≥0C_\sigma\geq0, chosen for all σ∈Kn\sigma\in K_n, such that for every ε>0\varepsilon>0 there is a separating filtration QQ of radius RDR_D satisfying Iσ(Q)≤Cσε(σ∈Kn),I_\sigma(Q)\leq C_\sigma\varepsilon\qquad(\sigma\in K_n),(17)✦ where the expected values are those of (13).

In the paper, the result reads: for the fixed graph Γ\Gamma, complex KK, free probability parameter (X,μ)(X,\mu), dimension nn and radii r<Rr<R of Sections 3 and 4, with volume VR<(R−r)n/n!V_R<(R-r)^n/n!, every ε>0\varepsilon>0 admits an admissible equivariant locally constant filtration (Zj,Tj)j=0n(Z_j,T_j)_{j=0}^n satisfying ∫X#(Z0(x)∩|σ|)dμ(x)≤Cσε\int_X\#(Z_0(x)\cap|\sigma|)\,d\mu(x)\leq C_\sigma\varepsilon for every original top simplex σ\sigma, where CσC_\sigma depends only on the fixed geometric data, independently of ε\varepsilon and the complexity of the chosen subdivisions.

What Is Checked

The assertions (15), (16) and (17) concern locally constant equivariant marked subdivisions of 𝒦\mathcal K, the distance (3) and the costs (12), with finite face-orbit data for the existence assertions and an invariant finite measure under the degree and dimension bounds for the count estimates. The coordinate inclusion V→ℓ2(Y)V\longrightarrow\ell^2(Y) is injective.✦