OctoInformBeta

OpenAI preprint, September 2026 · §6. From separating filtrations to finite chains

Component labels and the averaged chain map: Uniform vertex vectors label the faces of a separating filtration

Component labels define finite averaged chains.

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 construction replaces subdivided simplicial chains by exterior chains whose vertex vectors are constant on components of a separating filtration. The parameter of the filtration then supplies an average of these chains.

Let YY be the vertex set of a connected graph Γ\Gamma, with its shortest-path metric, and let KK be a simplicial complex on YY. Write V=ℝ(Y)V=\mathbb R^{(Y)} for the finitely supported real vectors and δy\delta_y for the unit vector at yy. Give VV the Euclidean norm from its coordinate inclusion in ℓ2(Y)\ell^2(Y).✦ The standard realization 𝒦\mathcal K of KK has simplex vertices {δy:y∈s}\{\delta_y:y\in s\} for each face ss of KK.✦

A finitely supported probability measure pp on YY is a vector in VV with nonnegative coefficients summing to one.✦ For finitely supported probability measures p,qp,q on YY, a real function a:Y→ℝa:Y\to\mathbb R, and n>0n>0, the expectation 𝔼p[a]\mathbb E_p[a] and expectation difference Δp,q(a)\Delta_{p,q}(a) are the real numbers in (1), and the normalized dual transportation distance ρ(p,q)\rho(p,q) is the supremum in that display, with the Lipschitz condition taken in the graph metric. 𝔼p[a]=∑y∈Yp(y)a(y),Δp,q(a)=𝔼p[a]−𝔼q[a],ρ(p,q)=1nsupLip⁡(a)≤1Δp,q(a).\mathbb E_p[a]=\sum_{y\in Y}p(y)a(y),\qquad \Delta_{p,q}(a)=\mathbb E_p[a]-\mathbb E_q[a],\qquad \rho(p,q)=\frac{1}{\sqrt n} \sup_{\operatorname{Lip}(a)\leq 1}\Delta_{p,q}(a).(1) Each realization point has a barycentric probability measure with those same coordinates and support a face of KK.✦ The realization distance uses (1) on |𝒦||\mathcal K| and is zero when either vector is outside |𝒦||\mathcal K|.✦ The unnormalized supremum in (1) is the Wasserstein distance W1(p,q)W_1(p,q).

For a geometric complex BB and a subset PP of its ambient vector space, the vertex restriction of BB to PP and the geometric restriction of BB to PP are the complexes with the respective face sets in (2). {s∈B:s⊆P},{s∈B:conv⁡(s)⊆P}.\{s\in B:s\subseteq P\},\qquad \{s\in B:\operatorname{conv}(s)\subseteq P\}.(2) A subdivision TT of 𝒦\mathcal K has the same underlying space, and each simplex of TT lies in an original simplex. It respects each original simplex PP: for every face ss of TT, the intersection conv⁡(s)∩P\operatorname{conv}(s)\cap P is conv⁡(s∩P)\operatorname{conv}(s\cap P).✦ Its vertex restriction to each original simplex has finitely many faces. A refinement BB of a geometric complex CC is a subdivision of CC.✦

A marked subset ZZ of |T||T| is a union of convex hulls of faces of TT.✦ A locally constant subdivision family (Tx)x∈X(T_x)_{x\in X} of 𝒦\mathcal K is a family of subdivisions of 𝒦\mathcal K, indexed by a space XX, whose face sets are locally constant after vertex restriction to each original simplex.✦ A marked family (Zx)x∈X(Z_x)_{x\in X} for a subdivision family (Tx)x∈X(T_x)_{x\in X} is a family of marked subsets Zx⊆|Tx|Z_x\subseteq|T_x| whose face sets are locally constant after geometric restriction to ZxZ_x and vertex restriction to each original simplex, in the senses of (2).✦

Let a group HH act on YY and XX. Its action on VV is the linear reindexing hδy=δhyh\delta_y=\delta_{hy}.✦ An equivariant marked subdivision pair (Tx,Zx)(T_x,Z_x) satisfies hTx=ThxhT_x=T_{hx} on faces and hZx=ZhxhZ_x=Z_{hx}.✦ A separating filtration QQ of length nn and radius R≥0R\geq0 is a sequence (Ti,x,Zi,x)(T_{i,x},Z_{i,x}) of these pairs with marked subsets, satisfying (3); the refinement and nesting conditions there apply for i<ni<n. Tn,x=𝒦,Zn,x=|𝒦|,Ti,x refines Ti+1,x,Zi,x⊆Zi+1,x.T_{n,x}=\mathcal K,\qquad Z_{n,x}=|\mathcal K|,\qquad T_{i,x}\text{ refines }T_{i+1,x},\qquad Z_{i,x}\subseteq Z_{i+1,x}.(3) It also satisfies dimension drop: if FF is a face of Ti+1,xT_{i+1,x} contained geometrically in Zi+1,xZ_{i+1,x} and EE is a face of Ti,xT_{i,x} with conv⁡(E)⊆conv⁡(F)∩Zi,x\operatorname{conv}(E)\subseteq\operatorname{conv}(F)\cap Z_{i,x}, then |E|<|F||E|<|F|.✦ Every connected subset of Zi+1,x\Zi,xZ_{i+1,x}\setminus Z_{i,x} has ρ\rho-diameter at most 2R2R.✦ Fixing xx gives the pointwise marked filtration QxQ_x.✦

A final triangulation face FF is a face of T0,xT_{0,x}.✦ The vertex mean mFm_F of a final triangulation face FF is the vector mF=|F|−1∑v∈Fvm_F=|F|^{-1}\sum_{v\in F}v, with the mean of the empty set taken to be zero.✦ Put Z−1,x=⌀Z_{-1,x}=\varnothing and write Ai,x=Zi,x\⋃j<iZj,xA_{i,x}=Z_{i,x}\setminus\bigcup_{j<i}Z_{j,x} for the filtration strata.✦ The vertex support vsupp⁡(U)\operatorname{vsupp}(U) of a subset U⊆VU\subseteq V is the subset of YY in (4), and the uniform vertex vector 𝒖(S)\mathbf u(S) of a finite set S⊆YS\subseteq Y is the vector in ℂ(Y)\mathbb C^{(Y)} in that display, taken to be zero when SS is empty. vsupp⁡(U)=⋃p∈Usupp⁡(p),𝒖(S)=∑y∈S|S|−1δy∈ℂ(Y).\operatorname{vsupp}(U)=\bigcup_{p\in U}\operatorname{supp}(p),\qquad \mathbf u(S)=\sum_{y\in S}|S|^{-1}\delta_y\ \in\mathbb C^{(Y)}.(4)

Exterior Chains and Subdivision

The exterior chain space Ek−1(Y)=⋀ℂkℂ(Y)E_{k-1}(Y)=\bigwedge^k_{\mathbb C}\mathbb C^{(Y)} has kk wedge factors; E−1(Y)=ℂE_{-1}(Y)=\mathbb C is the augmented degree.✦ Its oriented simplex basis consists of the wedges ese_s of unit vectors, ordered by a fixed ordering of the vertices of each kk-element set ss.✦ An alternating function a:Yk→ℂa:Y^k\to\mathbb C is a function on ordered kk-tuples of vertices that changes by the sign of a permutation and vanishes on repeated vertices.✦ Its linear extension sends the basis wedge ese_s to its value on the ordered vertices of ss.✦ A vertex map induces the linearized map on unit vectors and its exterior powers.

For a finite geometric complex BB, let ℱk(B)\mathcal F_k(B) be its faces with kk vertices.✦ The fundamental chain [B]χ,k[B]_{\chi,k} of a finite geometric complex BB, for k≥1k\geq1 and a coordinate map χ\chi to ℝk\mathbb R^k, is the exterior chain given by (5), where w1,…,wkw_1,\ldots,w_k are the ordered vertices of ss. [B]χ,k=∑s∈ℱk(B)sign⁡det⁡(χ(w1),…,χ(wk))es.[B]_{\chi,k}=\sum_{s\in\mathcal F_k(B)} \operatorname{sign}\det\bigl(\chi(w_1),\ldots,\chi(w_k)\bigr)e_s.(5) For a subdivision TT and an ordered tuple v=(v1,…,vk)v=(v_1,\ldots,v_k) of vertices of KK, its restricted simplex triangulation is the vertex restriction of TT to conv⁡{δv1,…,δvk}\operatorname{conv}\{\delta_{v_1},\ldots,\delta_{v_k}\}.✦ The simplex coordinate chart χv\chi_v associated with an ordered tuple v=(v1,…,vk)v=(v_1,\ldots,v_k) of vertices of KK is the map from VV to ℝk\mathbb R^k given by χv(p)=(p(v1),…,p(vk))\chi_v(p)=(p(v_1),\ldots,p(v_k)).✦ The ordered simplex subdivision value is the fundamental chain (5) of this triangulation in that chart when the tuple’s vertex set is a face of KK, and is zero otherwise.✦ The subdivision map ST,k:Ek−1(Y)→Ek−1(V)S_{T,k}:E_{k-1}(Y)\to E_{k-1}(V) is the linear extension of these values; for k=0k=0 it is the exterior power with zero factors of the vertex map y↦δyy\mapsto\delta_y.✦

For face labels b(F)∈ℂ(Y)b(F)\in\mathbb C^{(Y)} assigned to finite subsets FF of VV, the barycentric label map Lb,k:Ek−1(V)→Ek−1(Y)L_{b,k}:E_{k-1}(V)\to E_{k-1}(Y) is the complex linear map defined recursively by coning the labeled boundary of each simplex with its face label.✦ Its map in augmented degree is the identity.

Let XX be compact and equipped with a finite Borel measure μ\mu. A locally constant function g:X→Wg:X\to W, for a complex vector space WW, has finite image and measurable fibers; its associated simple function has the same values.✦ The finite average Av⁡μ(g)\operatorname{Av}_{\mu}(g) of a locally constant function g:X→Wg:X\to W is the vector in WW given by (6). For a family of complex linear maps whose value on each vector is locally constant in xx, its averaged linear map sends each vector to the finite average of those values. Av⁡μ(g)=∑z∈g(X)μ(g−1{z})z.\operatorname{Av}_{\mu}(g)=\sum_{z\in g(X)}\mu(g^{-1}\{z\})\,z.(6)✦

The Definition

Definition 1 (Component labels and averaged chains). Fix the graph and separating filtration above, and choose y0∈Yy_0\in Y. For x∈Xx\in X and a final face FF of T0,xT_{0,x}, the face level ix(F)i_x(F) is the least filtration index specified in (7). ix(F)=min⁡{i≤n:mF∈Zi,x}.i_x(F)=\min\{i\leq n:m_F\in Z_{i,x}\}.(7)✦ The face component Ux(F)U_x(F) of a final face FF of T0,xT_{0,x} is the connected component of Aix(F),xA_{i_x(F),x} containing mFm_F, as in (8). Ux(F)=Comp⁡Aix(F),x(mF).U_x(F)=\operatorname{Comp}_{A_{i_x(F),x}}(m_F).(8)✦ For a final face FF of T0,xT_{0,x}, the component vertex set Jx(F)J_x(F) is the vertex support of Ux(F)U_x(F) in (9), and the component weight ux(F)u_x(F) is its uniform vertex vector in (10), using (4). Jx(F)=vsupp⁡(Ux(F)).J_x(F)=\operatorname{vsupp}(U_x(F)).(9)✦ ux(F)=𝒖(Jx(F)).u_x(F)=\mathbf u(J_x(F)).(10)✦ The face weight bx(F)b_x(F) of a finite subset F⊆VF\subseteq V is the vector in ℂ(Y)\mathbb C^{(Y)} specified by (11). bx(F)={ux(F),F is a face of T0,x,δy0,otherwise.b_x(F)= \begin{cases} u_x(F),&F\text{ is a face of }T_{0,x},\\ \delta_{y_0},&\text{otherwise}. \end{cases}(11)✦ For k≥0k\geq0, the sparse chain family (fx,k)x∈X(f_{x,k})_{x\in X} consists of the complex linear maps in (12), and the averaged sparse chain map FkF_k is the complex linear map in (13), with the average of (6). fx,k=Lbx,k∘ST0,x,k:Ek−1(Y)→Ek−1(Y).f_{x,k}=L_{b_x,k}\circ S_{T_{0,x},k}: E_{k-1}(Y)\longrightarrow E_{k-1}(Y).(12)✦ Fk(c)=Av⁡μ(x↦fx,k(c)).F_k(c)=\operatorname{Av}_{\mu}\bigl(x\mapsto f_{x,k}(c)\bigr).(13)✦

In the paper, the result reads: for fixed n,Rn,R and the filtration supplied by Proposition 5.1 for ε>0\varepsilon>0, a barycentric vertex mFm_F receives its least level ii and its component UFU_F in Zi\Zi−1Z_i\setminus Z_{i-1}, a component UU has labels JU=⋃p∈Usupp⁡(p)J_U=\bigcup_{p\in U}\operatorname{supp}(p) and uU=|JU|−1∑y∈JUδyu_U=|J_U|^{-1}\sum_{y\in J_U}\delta_y, replacement by the wedge of component vectors gives, after subdivision, the map fx:C*(K)→E*f_x:C_*(K)\to E_* from the simplicial chain complex of KK to the exterior chain complex, and Fε(c)=∫Xfx(c)dμ(x)F^{\varepsilon}(c)=\int_X f_x(c)\,d\mu(x).

What Is Checked

The definitions are stated over the graph realization with its Euclidean vector topology and normalized transportation distance, an equivariant separating filtration with n>0n>0 and R≥0R\geq0, and the measured compact parameter space. Every final face has its vertex mean in some Zi,xZ_{i,x} with i≤ni\leq n✦, and the vertex support of its face component is finite.✦ Each function x↦fx,k(c)x\mapsto f_{x,k}(c) is locally constant.✦ Ordered simplex subdivision changes by the sign of a permutation of its vertices✦ and vanishes when vertices repeat.✦ The coordinate inclusion of VV in ℓ2(Y)\ell^2(Y) is injective.✦