OctoInformBeta

OpenAI preprint, September 2026 · §4. Filtered local systems on surfaces

§4.1 (setting): surface diagrams, seams and solutions: Ordered side occurrences and identified flags give matrix equations with a coordinate algebra

Surface diagrams and their algebraic solutions.

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

Matrix transports on facets must respect filtrations along seams, including the maps to specified graded modules. At a vertex, the relevant identified flags are compared after the incident fiber coordinates have been identified. The library defines these conditions over a commutative ring RR and a commutative RR-algebra AA.

For a finite coordinate set II and a grading a:I→ℕa:I\to\mathbb N, let AIA^I have coordinate prefixes PkP_k and coordinate projections πk\pi_k to the named modules GkG_k. A frame e:AI→∼Me:A^I\xrightarrow{\sim}M gives a prefix Fk(e)F_k(e) and a grade map γe,k\gamma_{e,k}, as in (1): Pk={u∈AI:ui=0 whenever a(i)≥k},Gk=A{i:a(i)=k},πk(u)=(ui)a(i)=k,Fk(e)=e(Pk),γe,k=πk∘e−1.\begin{aligned} P_k&=\{u\in A^I:u_i=0\text{ whenever }a(i)\geq k\}, & G_k&=A^{\{i:a(i)=k\}},\\ \pi_k(u)&=(u_i)_{a(i)=k}, & F_k(e)&=e(P_k),\qquad \gamma_{e,k}=\pi_k\circ e^{-1}. \end{aligned}(1) Two frames e,f:AI→∼Me,f:A^I\xrightarrow{\sim}M determine the same identified flag when (2) holds: FlagEq⁡a(e,f)⇔{Fk(e)=Fk(f)for every k,γe,k(v)=γf,k(v)for every k and v∈Fk+1(e).\operatorname{FlagEq}_a(e,f)\quad\Longleftrightarrow\quad \left\{ \begin{aligned} F_k(e)&=F_k(f)&&\text{for every }k,\\ \gamma_{e,k}(v)&=\gamma_{f,k}(v)&& \text{for every }k\text{ and }v\in F_{k+1}(e). \end{aligned} \right.(2)✦ The grade map on Fk+1(e)F_{k+1}(e) identifies its quotient by Fk(e)F_k(e) with GkG_k. For a finite coordinate set JJ, an invertible coordinate matrix from AIA^I to AJA^J is a pair of matrices X,YX,Y with XY=1XY=1 and YX=1YX=1; its linear isomorphism sends uu to XuXu. Identity matrices, composition and direct sums give the corresponding isomorphisms, and bijections of coordinate sets relabel their rows and columns.

The Definition

Definition 1. Let F,S,VF,S,V be sets of facets, seams and vertices, and let ms∈ℕm_s\in\mathbb N be the number of ordered children of a seam s∈Ss\in S. The set of side occurrences is (3): Side⁡(S,m)=∐s∈S({*}⊔{0,…,ms−1}).\operatorname{Side}(S,m) =\coprod_{s\in S}\bigl(\{*\}\sqcup\{0,\ldots,m_s-1\}\bigr).(3)✦ Here (s,*)(s,*) is the parent occurrence and (s,j)(s,j) is the jjth child occurrence. A facet label does not identify distinct elements of (3). A side germ is a pair (a,ε)(a,\varepsilon) of an occurrence and an endpoint in {0,1}\{0,1\}.✦ Set η(s,*)=1\eta(s,*)=1 and η(s,j)=0\eta(s,j)=0; the initial and terminal germs of aa are (a,1−η(a))(a,1-\eta(a)) and (a,η(a))(a,\eta(a)). The positive boundary orientation follows the seam parameter on the parent occurrence and opposes it on every child occurrence.✦

The allowed vertex types form the set in (4): 𝒦={cont⁡m:m≥0}⊔{int⁡m,i:0≤i<m−1}⊔{split⁡a,b,c,merge⁡a,b,c:a,b,c≥0}.\mathcal K= \{\operatorname{cont}_m:m\geq0\} \sqcup\{\operatorname{int}_{m,i}:0\leq i<m-1\} \sqcup\{\operatorname{split}_{a,b,c},\operatorname{merge}_{a,b,c}:a,b,c\geq0\}.(4)✦ Each type in (4) has a local vertex table: ports pp with ordered child sets CpC_p, endpoint labels tpt_p, and an involution τ\tau on pairs (p,c)(p,c), where cc is either ** or a child of pp. The involution pairs incoming and outgoing oriented sides. A continuation has two ports with endpoint labels 1,01,0 and pairs corresponding parents and children. An interchange uses the same table with the child pairing given by the transposition of ii and i+1i+1.

A split has ports called before, after and branch, with endpoint labels 1,0,11,0,1. Their child lists have the respective forms L,*,NL,*,N, L,M,NL,M,N and MM, where the ordered lists L,M,NL,M,N have lengths a,b,ca,b,c.✦ The corner pairing joins the before and after parents, joins corresponding children in LL and NN, joins the distinguished before child to the branch parent, and joins the after children in MM to the branch children.✦ A merge reverses every endpoint label in this table.✦

A local port is a pair (v,p)(v,p) with pp a port of the type kvk_v assigned to vv; a local end is a triple (v,p,c)(v,p,c). Write Port⁡(kv)\operatorname{Port}(k_v) for the ports at vv, let xx range over local ends, write ρ\rho for endpoint realization and side\operatorname{side} for the map forgetting a germ’s endpoint; an ordered port assembly 𝒜\mathcal A on the facet, seam and vertex sets F,S,VF,S,V consists of the data in (5), with the indicated conditions: 𝒜=((kv)v∈V,α,ϕ),α:∐v∈VPort⁡(kv)→∼S×{0,1},ϕ:Side⁡(S,m)→F,α(v,p)=(s,tp),|Cp|=ms,ϕ(side⁡(ρτx))=ϕ(side⁡(ρx)).\begin{gathered} \mathcal A=\bigl((k_v)_{v\in V},\alpha,\phi\bigr),\qquad \alpha:\coprod_{v\in V}\operatorname{Port}(k_v) \xrightarrow{\sim}S\times\{0,1\},\qquad \phi:\operatorname{Side}(S,m)\to F,\\ \alpha(v,p)=(s,t_p),\qquad |C_p|=m_s,\qquad \phi\bigl(\operatorname{side}(\rho\tau x)\bigr) =\phi\bigl(\operatorname{side}(\rho x)\bigr). \end{gathered}(5)✦ In (5), the endpoint realization ρ\rho sends (v,p,c)(v,p,c) to the germ at the attached seam endpoint, using ** for the parent and the prescribed enumeration of CpC_p for its children; τ\tau acts within each vertex. The induced vertex assembly has corner pairing μ=ρτρ−1\mu=\rho\tau\rho^{-1}.✦ The boundary successor ν\nu sends an occurrence to the side of the germ paired by μ\mu with its terminal germ.✦

A surface incidence diagram DD is an ordered port assembly together with facet ranks rfr_f, genera gfg_f, boundary counts hfh_f and positive boundary lengths ℓf,b\ell_{f,b}, subject to (6): D=(𝒜,r,g,h,ℓ,β),β:∐f∈F∐b=0hf−1{0,…,ℓf,b−1}→∼Side⁡(S,m),rϕ(s,*)=∑j=0ms−1rϕ(s,j),ϕ(β(f,b,i))=f,ν(β(f,b,i))=β(f,b,(i+1)mod⁡ℓf,b).\begin{gathered} D=(\mathcal A,r,g,h,\ell,\beta),\qquad \beta:\coprod_{f\in F}\coprod_{b=0}^{h_f-1} \{0,\ldots,\ell_{f,b}-1\}\xrightarrow{\sim}\operatorname{Side}(S,m),\\ r_{\phi(s,*)}=\sum_{j=0}^{m_s-1}r_{\phi(s,j)},\qquad \phi\bigl(\beta(f,b,i)\bigr)=f,\\ \nu\bigl(\beta(f,b,i)\bigr) =\beta\bigl(f,b,(i+1)\bmod\ell_{f,b}\bigr). \end{gathered}(6)✦ All ranks, genera and counts in (6) are nonnegative integers. A scalar puncture datum PP on DD is the choice in (7): P=((pf)f∈F,(λf,j)f∈F,0≤j<pf),pf∈ℕ,λf,j∈R×.P=\bigl((p_f)_{f\in F},(\lambda_{f,j})_{f\in F,\,0\leq j<p_f}\bigr), \qquad p_f\in\mathbb N,\quad \lambda_{f,j}\in R^\times.(7)✦ The units in (7) specify scalar puncture monodromies.

The transport generators are a side matrix TaT_a of size rϕ(a)r_{\phi(a)}, a seam frame Hs,εH_{s,\varepsilon} at each endpoint, and two handle matrices Xf,k,Yf,kX_{f,k},Y_{f,k} of size rfr_f for 0≤k<gf0\leq k<g_f. Write 𝒢\mathcal G for this generator set and nen_e for the size of a generator ee. Seam frame sizes and their coordinate gradings are given by (8): ds,j=rϕ(s,j),ns=∑j=0ms−1ds,j,Ans≃⨁j=0ms−1Ads,j,as(j,t)=j.d_{s,j}=r_{\phi(s,j)},\qquad n_s=\sum_{j=0}^{m_s-1}d_{s,j},\qquad A^{n_s}\simeq\bigoplus_{j=0}^{m_s-1}A^{d_{s,j}},\qquad a_s(j,t)=j.(8) A transport word ww in the generators 𝒢\mathcal G over RR is a matrix expression formed from those generators and constant invertible matrices by products, inverses and block sums. Its evaluation substitutes the generator matrices and applies R→AR\to A to constants; an empty product is the identity.

The boundary monodromy word Bf,bB_{f,b} of boundary bb of facet ff, the surface word Ωf\Omega_f, scalar target Λf\Lambda_f, and the left and right seam words Ls,MsL_s,M_s are defined in (9); products are written in increasing index order: Bf,b=Tβ(f,b,ℓf,b−1)⋯Tβ(f,b,0),Ωf=(∏k=0gf−1Xf,kYf,kXf,k−1Yf,k−1)(∏b=0hf−1Bf,b),Λf=(∏j=0pf−1λf,j)−1Irf,Ls=T(s,*)Hs,0,Ms=Hs,1diag⁡(T(s,0)−1,…,T(s,ms−1)−1).\begin{aligned} B_{f,b}&=T_{\beta(f,b,\ell_{f,b}-1)}\cdots T_{\beta(f,b,0)},\\ \Omega_f&=\left(\prod_{k=0}^{g_f-1} X_{f,k}Y_{f,k}X_{f,k}^{-1}Y_{f,k}^{-1}\right) \left(\prod_{b=0}^{h_f-1}B_{f,b}\right),\\ \Lambda_f&=\left(\prod_{j=0}^{p_f-1}\lambda_{f,j}\right)^{-1}I_{r_f},\\ L_s&=T_{(s,*)}H_{s,0},\qquad M_s=H_{s,1}\operatorname{diag}\bigl(T_{(s,0)}^{-1},\ldots, T_{(s,m_s-1)}^{-1}\bigr). \end{aligned}(9) The edge flag circuit has one comparison for every facet and seam, using constant grading 00 on a facet and asa_s on a seam. Its compatibility condition ℰ(g)\mathscr E(g) for an assignment g∈∏e∈𝒢GL⁡ne(A)g\in\prod_{e\in\mathcal G}\operatorname{GL}_{n_e}(A) is (10): ℰ(g)⇔{FlagEq⁡0(Ωf(g),Λf)(f∈F),FlagEq⁡as(Ls(g),Ms(g))(s∈S).\mathscr E(g)\quad\Longleftrightarrow\quad \left\{ \begin{aligned} \operatorname{FlagEq}_0\bigl(\Omega_f(g),\Lambda_f\bigr)&\quad(f\in F),\\ \operatorname{FlagEq}_{a_s}\bigl(L_s(g),M_s(g)\bigr)&\quad(s\in S). \end{aligned} \right.(10)✦ Matrices in (10) act as frames of their coordinate modules. A flag incidence circuit 𝒞\mathscr C consists of transport generators and constraints, each with two transport words and a specified grading; a solution of the circuit over AA is an assignment of invertible matrices to its transport generators for which the two words at every constraint determine the same identified flag for its specified grading.

The boundary monodromy word Bf,bB_{f,b} of boundary bb of facet ff is the product of its boundary side transports in reverse boundary order, as in (9).✦ The handle commutator word at handle kk of facet ff is Xf,kYf,kXf,k−1Yf,k−1X_{f,k}Y_{f,k}X_{f,k}^{-1}Y_{f,k}^{-1}.✦

The local occurrence rank at (v,p,c)(v,p,c) is the rank of its facet.✦ Corner identifications identify the corresponding coordinate sets. The port frame is Hα(v,p)H_{\alpha(v,p)}, with columns indexed by the sum of its child coordinate sets and rows by its parent coordinate set.✦ A passage is a continuation or an interchange; its flag comparison uses the parent identification and the child permutation, and its quotient grading is jj for a continuation, while an interchange at ii assigns grade ii also to child i+1i+1.

For a split, write U0⊕(W⊕Z0)U_0\oplus(W\oplus Z_0) for the before child module and U1⊕(B⊕Z1)U_1\oplus(B\oplus Z_1) for the after child module. Here Uε,ZεU_\varepsilon,Z_\varepsilon are the sums of the prefix and suffix coordinate modules, and BB is the sum of the branch child modules. Corners identify U1U_1 with U0U_0, Z1Z_1 with Z0Z_0, and WW with the branch parent module. The splitting refinement isomorphism from the after child module to the before child module uses the corner identifications on the prefix and suffix and inserts the branch frame from BB to WW into the splitting block, with the prescribed child coordinate order.✦ The splitting comparison is the framed flag comparison of the refined frame with the unrefined frame after inserting the branch frame into the splitting block and identifying the parent fibers.✦ It retains each named graded module and its quotient identification in prefix, subblock, suffix order. For a merge, the comparison uses the same ranks with the endpoint orientations reversed.

Write cv(g)c_v(g) for the resulting vertex flag comparison, with coordinate grading ava_v and frames cvL(g),cvR(g)c_v^L(g),c_v^R(g). Define vertex compatibility by 𝒱(g)⇔FlagEq⁡av(cvL(g),cvR(g))\mathscr V(g)\Longleftrightarrow \operatorname{FlagEq}_{a_v}(c_v^L(g),c_v^R(g)) for every v∈Vv\in V. A solution gg of (D,P)(D,P) over AA is an element of (11): Sol⁡D(P;A)={g∈∏e∈𝒢GL⁡ne(A):ℰ(g) and 𝒱(g)}.\operatorname{Sol}_D(P;A)= \left\{g\in\prod_{e\in\mathcal G}\operatorname{GL}_{n_e}(A): \mathscr E(g)\text{ and }\mathscr V(g)\right\}.(11)✦

Introduce two matrices of indeterminates Ze0,Ze1Z_e^0,Z_e^1 for each e∈𝒢e\in\mathcal G in the polynomial ring B=R[ze,ijε]B=R[z_{e,ij}^\varepsilon]. The polynomial form of a word replaces a generator and its inverse by these two matrices; inversion reverses the order of a product and inverts each block of a block sum.✦ For an edge comparison t∈F⊔St\in F\sqcup S, let QtQ_t be the polynomial form of its right word inverse times its left word, and let ata_t be its grading. The edge coordinate algebra BD(P)B_D(P) of (D,P)(D,P) is the quotient in (12): BD(P)=B/((ZeεZe1−ε−I)ij,(Qt−I)ijwith at(j)≤at(i)).B_D(P)=B\Big/ \left( (Z_e^\varepsilon Z_e^{1-\varepsilon}-I)_{ij},\quad (Q_t-I)_{ij}\ \text{with }a_t(j)\leq a_t(i) \right).(12) The parentheses in (12) denote the ideal generated by the displayed elements.✦ The images of Ze0Z_e^0 are the universal edge transports, and the images of Ze1Z_e^1 are their inverse matrices. Evaluate every vertex comparison on these transports; let vv,ijv_{v,ij} be the (i,j)(i,j) entry of (cvR)−1cvL−I(c_v^R)^{-1}c_v^L-I whenever av(j)≤av(i)a_v(j)\leq a_v(i).✦ The surface coordinate algebra CD(P)C_D(P) of (D,P)(D,P) is the quotient algebra in (13): CD(P)=BD(P)/(vv,ij:v∈V,av(j)≤av(i)).C_D(P)=B_D(P)\Big/\bigl(v_{v,ij}:v\in V,\ a_v(j)\leq a_v(i)\bigr).(13)✦ For every commutative RR-algebra AA, evaluating polynomial coordinates at a solution and evaluating the universal transports at an algebra map give the correspondence in (14): Sol⁡D(P;A)→∼Hom⁡R-alg(CD(P),A).\operatorname{Sol}_D(P;A)\xrightarrow{\sim} \operatorname{Hom}_{R\text{-alg}}\bigl(C_D(P),A\bigr).(14)✦ For a matrix presentation ℰ\mathcal E, the matrix coordinate algebra CℰC_{\mathcal E} is the polynomial ring in a forward and an inverse matrix for each occurrence, modulo the ideal generated by both inverse identities and all its incidence equations. The matrix point correspondence is the bijection between matrix incidence solutions over AA and RR-algebra maps Cℰ→AC_{\mathcal E}\to A, obtained by evaluating polynomial coordinates at a solution and evaluating universal transport matrices at an algebra map.

The coordinate algebra C𝒞C_{\mathscr C} of a flag incidence circuit 𝒞\mathscr C is the matrix coordinate algebra for its flag equations: the entries of the relative transport minus the identity whose column grade is at most their row grade.✦

For an equation family (ej)(e_j) in an RR-algebra BB, passage to the quotient identifies the RR-algebra maps B→AB\to A annihilating every eje_j with the RR-algebra maps B/(ej)→AB/(e_j)\to A.✦ The equation quotient algebra is the quotient of BB by the equation ideal.✦

In the paper, the result reads: a surface diagram DD consists of finitely many compact oriented facets of ranks between 00 and a fixed rr, with finitely many interior punctures carrying prescribed scalar roots of unity in a field KK; boundary pieces are grouped into seams with a parent EE and ordered children W1,…,WmW_1,\ldots,W_m satisfying rank⁡(E)=∑jrank⁡(Wj)\operatorname{rank}(E)=\sum_j\operatorname{rank}(W_j) and child boundary orientations opposite to the parent’s, with continuation, adjacent interchange, split and merge vertices and distinct side occurrences even when their facets coincide; a solution assigns local systems and parallel filtrations 0=E0⊂⋯⊂Em=E0=E_0\subset\cdots\subset E_m=E with identified quotients Ej/Ej−1≃WjE_j/E_{j-1}\simeq W_j, satisfying all vertex compatibilities, including, for the two adjacent graded modules AA and BB at an interchange, preservation of the coarsened identified quotient A⊕BA\oplus B and complementarity of the intermediate summands with their natural projection maps; choosing frames and paths gives the algebraic solution space S(D)S(D), and integral solutions use free RR-fibers and the specified free graded modules.

What Is Checked

The constructions are stated for arbitrary commutative coefficient rings and their commutative algebras, with the occurrence incidence data of (6) and unit puncture data of (7); the finite type setting takes the facet and seam sets to be finite.

For a matrix presentation, evaluation on a solution annihilates the relation ideal.✦ In a flag circuit, agreement of the identified flags is equivalent to the vanishing of the entries of the relative transport minus the identity at the positions whose column grade is at most their row grade.✦ The edge point correspondence identifies solutions of (10) with RR-algebra maps from BD(P)B_D(P).✦ For every flag incidence circuit 𝒞\mathscr C, its solutions over AA are in bijection with the RR-algebra maps C𝒞→AC_{\mathscr C}\to A.