\documentclass[10pt,reqno]{amsart} \usepackage[margin=0.87in]{geometry} \usepackage{mathtools,amssymb,booktabs,microtype,tikz} \usepackage[colorlinks=true,linkcolor=blue!45!black,citecolor=blue!45!black,urlcolor=blue!55!black]{hyperref} \hypersetup{pdftitle={Holes and islands of visible points on the triangular lattice: midpoint bond graphs and prime-sieve bounds},pdfauthor={Sabah Noori Jihad},pdfsubject={AI-assisted mathematical preprint; independent researcher, Iraq},pdfkeywords={arithmetic topology, visible lattice points, number theory, Betti numbers}} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition}\newtheorem{definition}[theorem]{Definition} \theoremstyle{remark}\newtheorem{remark}[theorem]{Remark} \newcommand{\Z}{\mathbb Z}\newcommand{\N}{\mathbb N}\newcommand{\dens}{\operatorname{dens}}\newcommand{\rk}{\operatorname{rank}}\newcommand{\e}{\mathbf e} \newcommand{\Splus}{\mathcal S_+} \title[Holes and islands of visible lattice points]{Holes and islands of visible points on the triangular lattice: midpoint bond graphs and prime-sieve bounds} \author{Sabah Noori Jihad} \address{Independent Researcher, Iraq} \email{sabahnoorij@gmail.com} \date{21 November 2026} \subjclass[2020]{Primary 11H06, 11N25; Secondary 05C10, 55N10, 52C07} \keywords{Visible lattice points, triangular lattice, Betti numbers, midpoint bond graph, prime sieves, Euler products, periodic approximation} \begin{document} \begin{abstract} For the induced complex on primitive integer pairs in a triangular lattice, we identify first homology with interior components of a half-scale midpoint bond graph. Its complementary planar dual has the visible-site graph as its line graph, giving a bijection between finite visible islands and bounded nonsingleton dual faces. In particular the hole density $h$ is a local Euler-product term plus the island density $b_0$: $h=3J_2-2J_3-J_1+b_0$, with $J_j=\prod_p(1-j/p^2)$. Each odd prime produces a matching, two primes produce a small forest, and diameter exclusion bounds cycle birth. We refine the bounds by retaining the actual earlier prime sieve at each step. A computer-assisted certificate, including a nonnegative coverage recurrence for islands, yields $0.109179780.10914952$, still below the island certificate, and an exact decomposition distinguishes old-component chords from cycles born among new cross-component bonds. Assuming the random-coprime nonpercolation theorem of Le Fourn--Liu--Martineau, the relative cycle increment of a new prime converges to a conditional connection probability. The planar correspondences, density existence, and numerical bounds do not require this probabilistic input. \end{abstract} \maketitle \section{Introduction} The visibility of an integer vector encodes the absence of a common prime divisor of its coordinates. Visible lattice sets have been studied through diffraction, pattern frequencies and arithmetic percolation \cite{BMP,PH,LLM,Cellarosi}. The density of visible points itself is classical; it does not directly describe the topology of an induced simplicial complex built on these points. Here we investigate the number of holes in a particular planar complex and connect them to the midpoints of edges in a coarser lattice. The broader motivation is quantitative topology of structured point sets: random simplicial complexes provide a contrasting framework (Kahle \cite[Sections 1--2]{Kahle}), while arithmetic percolation studies connectivity in highly dependent visibility patterns. Our complexes are deterministic induced subcomplexes, not Linial--Meshulam complexes or independent percolation. The purpose is to identify which global topological counts admit exact local arithmetic descriptions. \subsection*{Origin in the generative arithmetic fabric} The motivation comes from the author's \emph{Generative Arithmetic Fabric} \cite{GAF}, a geometric workbench based on cumulative Cayley growth in $D_3=\{(x,y,z)\in\Z^3:x+y+z\equiv0\pmod2\}$ with the twelve minimal root steps. Starting at the origin, successive applications of these steps give a graded lattice; arithmetic labels are read from the generated points rather than built into the growth rule. One graded triangular section is $(x,y)\mapsto(x,y,x+y)$ with $x,y\ge0$ and shell index $x+y$: the minimal steps along the section are $(1,0,1),(0,1,1),(1,-1,0)$ and their negatives. The coordinate gcd in this section is $\gcd(x,y)$. The earlier program motivated examining collisions of neighboring generated points at arithmetic midpoints. In this paper that idea becomes a precise two-scale model: the midpoint of an edge between two even points is tested for a common prime divisor, creating a bond if the midpoint is nonprimitive. Figure~\ref{fig:origin} illustrates the passage from graded growth to the planar readout; Figure~\ref{fig:reduction} gives the midpoint rule. This citation records the provenance of the research question, not an assumption of any theorem from the earlier work. \begin{figure}[ht] \centering \begin{tikzpicture}[x={(.78cm,0cm)},y={(.39cm,.6755cm)}] \foreach \i in {0,...,4}{\foreach \j in {0,...,4}{\ifnum\numexpr\i+\j\relax<5 \fill[blue!55!black] (\i,\j) circle (1.7pt);\fi}} \foreach \g in {1,...,4}{\draw[gray!65] (\g,0)--(0,\g);} \draw[blue!55!black] (0,0)--(4,0) (0,0)--(0,4); \node[below,font=\small] at (2,-.25) {Graded section of $D_3$}; \draw[->,thick] (4.3,1)--(5.3,1); \begin{scope}[shift={(6,0)}] \draw[gray!70] (0,0)--(4,0)--(0,4)--cycle; \foreach \x/\y in {0/0,2/0,4/0,0/2,2/2,0/4}{\fill[blue!55!black] (\x,\y) circle (2.5pt);} \draw[orange!80!black,very thick] (2,0)--(4,0) (0,2)--(0,4); \fill[orange!80!black] (3,0) circle (2.2pt);\fill[orange!80!black] (0,3) circle (2.2pt); \node[below,font=\small] at (2,-.25) {Midpoint bonds}; \end{scope} \end{tikzpicture} \caption{The generative origin of the planar construction. Successive shells give the triangular section; the even sublattice and selected nonprimitive midpoints give the coarse bond graph. Bonds are illustrated schematically, not exhaustively.}\label{fig:origin} \end{figure} The question arising in the author's earlier investigation is stated here independently: for the full primitive-vertex subcomplex $K_k$ of the triangle $x,y\ge0$, $x+y\le k$ with triangular-lattice adjacency, does the limit $\lim_{k\to\infty}\beta_1(K_k)/k^2$ exist, and can it be bounded explicitly? The present paper supplies a proof and effective enclosures using an exact geometric reduction rather than assuming independence of nearby visible points. The main additional ingredients are the decomposition of the bond graph into prime matchings and a quantitative control of the resulting cycle rank. The construction should be distinguished from the theory of the size and existence of infinite clusters: the density of \emph{components} counts each finite component once, not according to its cardinality. No hypothesis on infinite components is needed for the existence and quantitative enclosures; the later prime-increment law uses a known random-model nonpercolation theorem. \subsection*{Contributions and scope} The mathematical thesis is that global arithmetic absence (holes) equals a local Euler characteristic term plus finite islands of presence. The specific contributions for priority assessment are the exact boundary-preserving midpoint reduction and pointwise visible-island correspondence; the two Euler identities; prime matchings and two-prime forests; the diameter-sensitive and prime-stratified cycle-tail bounds and conditional asymptotic prime increment; and the certified numerical enclosure by complementary computations. The exact chord/nonchord decomposition and finite-path arithmetic certificate give an additional check of the sieve mechanism. The planar graph duality and inclusion--exclusion used in the expansion are standard tools. The face probabilities and cluster weights are explicit specializations of classical finite-pattern products; their use to bound this Betti density is specific to the present graph. Existence of a component density via a bounded-size approximation is already a standard consequence of pattern frequencies, and Cellarosi's Proposition~2.4 proves such an approximation for representatives of deep finite components on the cubical nearest-neighbor graph. We do not claim originality for that technique, Alexander duality, CRT, or Euler's formula. See Section~\ref{sec:literature} for the adjacency and observable distinctions. No claim of exhaustive priority is made. \section{The planar model and midpoint reduction}\label{sec:model} Work in $\Lambda=\Z^2$, endowed with the triangular-lattice adjacency given by \[ \Splus=\{(1,0),(0,1),(1,-1)\},\qquad \mathcal S=\Splus\cup(-\Splus). \] For $k\ge1$, put $T_k=\{(x,y)\in\Z_{\ge0}^2:x+y\le k\}$, $n_k=|T_k|=(k+1)(k+2)/2$. Let $\mathcal T_k$ be its usual triangulation, in which each edge of difference in $\mathcal S$ is present and every elementary triangle is filled. Define \[ D_k=\{(x,y)\in T_k:\gcd(x,y)\ne1\},\qquad K_k=\mathcal T_k[T_k\setminus D_k], \] with $\gcd(0,0)=0$. Use integral simplicial homology throughout. A component of a graph contained in $T_k$ is called interior if it has no vertex on $x=0$, $y=0$ or $x+y=k$. Denote its interior-component count by $c_{\rm int}$ and its total component count by $c$. \begin{lemma}[Planar component formula]\label{lem:dual} For any deleted set $D\subset T_k$, the full induced complex on $T_k\setminus D$ has free first homology of rank $c_{\rm int}(D)$, where $D$ carries the induced lattice adjacency. \end{lemma} \begin{proof} Let $X=|\mathcal T_k|$ be the closed triangular disc, $A=|\mathcal T_k[T_k\setminus D]|$ and $B_D=|\mathcal T_k[D]|$. Work in the one-point compactification $S^2$ of the ambient plane. The empty retained complex has $H_1=0$ and all deleted vertices form one boundary component, so suppose $A\ne\varnothing$. \emph{Step 1 (deformation retraction).} If $z\in X\setminus A$ has barycentric coordinates $\lambda_v$ in a simplex, the sum $b(z)=\sum_{v\in D}\lambda_v$ is positive because $A$ is full. Define \[ R(z)=\sum_{v\in D}\frac{\lambda_v}{b(z)}v,\qquad H(z,t)=(1-t)z+tR(z). \] All expressions use the vertices of the containing simplex. The restrictions agree on common faces, and the homotopy stays in that simplex with positive deleted weight; it fixes $B_D$ pointwise. This gives a strong deformation retraction onto $B_D$ and also preserves the boundary when restricted to $\partial X\setminus A$. Its components are exactly the graph components of the deleted vertices: every simplex connects its vertices by edges. This includes the case $D=\varnothing$, when $X\setminus A$ is empty. \emph{Step 2 (exterior and boundary).} A component of $X\setminus A$ meets $\partial X$ exactly when its retracted deleted component contains a boundary vertex. Each such component connects to the single exterior region $S^2\setminus X$: any point of $\partial X\setminus A$ has a small neighborhood disjoint from the closed set $A$, containing exterior points. An interior component cannot connect to the exterior, since a path leaving $X$ must first meet $\partial X$ without meeting $A$. No two interior components can join via the exterior for the same reason. These arguments apply to path components (the open complement in $S^2$ is locally path connected). Thus $S^2\setminus A$ has precisely $1+c_{\rm int}(D)$ components. All three sides, including the corners, are counted as boundary. \emph{Step 3 (duality and integral homology).} The nonempty finite polyhedron $A$ is compact and locally contractible, a proper subset of $S^2$. Alexander duality, Hatcher \cite[Corollary 3.45]{Hatcher}, gives $\widetilde H^1(A;\Z)\cong\widetilde H_0(S^2\setminus A;\Z)\cong\Z^{c_{\rm int}(D)}$. To see that $H_1(A;\Z)$ has no torsion, repeatedly remove a triangle together with a free edge: whenever a triangle remains, the union of remaining closed triangles is a finite planar union with an exposed side incident to exactly one triangle. Such a side exists, for example by taking a generic ray from an interior point to infinity and the last triangle side crossed. Each removal is an elementary simplicial collapse. The resulting finite graph has free integral first homology. By the universal coefficient theorem (with $H_0$ free), the rank of $H_1$ is the rank of $H^1$. No connectedness of $A$ was required. This proves the claim. \end{proof} \begin{definition}[Midpoint bond graph] On the coarse lattice $\Z^2$, let $G$ have the edge $\{u,u+s\}$, $s\in\Splus$, if and only if \begin{equation} \gcd(2u_1+s_1,2u_2+s_2)>1. \label{eq:bond} \end{equation} The point $2u+s$ is the midpoint of the edge joining $2u$ to $2(u+s)$ in the original lattice. Write $G_m=G[T_m]$ for the induced coarse graph. Isolated coarse vertices are included. We also call this the collision graph, in the terminology of the motivating fabric; this name describes midpoint incidence, not a physical process. \end{definition} \begin{theorem}[Exact midpoint reduction]\label{thm:reduction} For each $m\ge1$, the connected components of the deleted graph on $D_{2m}$ correspond bijectively to those of $G_m$. Under the correspondence, touching the boundary is preserved. In particular, \begin{equation}\label{eq:reduction} H_1(K_{2m};\Z)\cong\Z^{c_{\rm int}(G_m)},\qquad \beta_1(K_{2m})=c_{\rm int}(G_m). \end{equation} \end{theorem} \begin{proof} All vertices of $2\Z^2\cap T_{2m}$ are deleted, including the origin. We give the geometric details of the correspondence. \emph{Midpoint classification.} The triangulation obtained by scaling $\mathcal T_m$ by two has vertex set $2T_m$. Subdivide each of its triangles by joining the midpoints of its three sides. The resulting triangulation is exactly $\mathcal T_{2m}$: the four subtriangles are elementary lattice triangles of the original orientation or its reverse. For the three nonzero parity classes, representatives $(2a+1,2b)$, $(2a,2b+1)$, $(2a+1,2b+1)$ have unique pairs of coarse endpoints, respectively \[ \begin{array}{ll} (2a,2b),&(2a+2,2b),\\ (2a,2b),&(2a,2b+2),\\ (2a+2,2b),&(2a,2b+2). \end{array} \] If the midpoint lies in $T_{2m}$, these endpoints lie there too: for the first two classes its sum of coordinates is odd and at most $2m-1$; for the last class both endpoint sums equal the midpoint sum. These are its only even neighbors, since the six lattice directions contain exactly two directions of the relevant parity. There are no fine edges between distinct even vertices. The subdivision also classifies each fine edge between two midpoints: it is an interior side of one subdivided coarse triangle. The two corresponding coarse sides share a coarse vertex. \emph{Preservation of components.} Whenever two deleted midpoints are adjacent, their common even neighbor is deleted as well. Remove all midpoint--midpoint edges simultaneously; each removed edge has an alternative path of length two consisting entirely of retained edges of the deleted graph. Replacing every removed edge on any original path gives a walk in the reduced graph, and the reverse implication holds by inclusion. Every remaining deleted midpoint has degree two and occurs precisely when its coarse edge is active by \eqref{eq:bond}. Thus the remaining graph is exactly the edge subdivision of $G_m$, with isolated coarse vertices included. Suppressing each degree-two midpoint preserves the components in both directions. \emph{Preservation of the boundary.} For each of the three defining affine inequalities of the triangle, an average of two nonnegative slacks is zero exactly when both slacks vanish. Hence a deleted midpoint lies on a given side if and only if both coarse endpoints lie on that side. An even vertex on the boundary corresponds to a coarse boundary vertex. The component bijection therefore preserves meeting the boundary, including the corners. Lemma~\ref{lem:dual} now gives \eqref{eq:reduction}. This is a correspondence of components; it does not assert that the full deleted graph and $G_m$ have equal cycle ranks. The same local midpoint classification and path replacement hold on the entire plane whenever all even sites are deleted, with arbitrary remaining deletions. The subdivision associates finitely many midpoints to each coarse vertex and every coarse vertex is itself a deleted site. Thus corresponding components are finite simultaneously. This extension will be used only in the auxiliary random-residue model. \end{proof} \begin{figure}[ht] \centering \begin{tikzpicture}[scale=1.12, x={(1cm,0cm)}, y={(.5cm,.866025cm)}] \definecolor{coarse}{RGB}{35,73,119}\definecolor{active}{RGB}{175,73,24} \draw[gray!45,very thin] (0,0)--(2,0)--(0,2)--cycle; \draw[gray!45,very thin] (1,0)--(1,1)--(0,1)--cycle; \draw[gray!65,dashed,thick] (0,0)--(0,2)--(2,0); \draw[active,very thick] (0,0)--(2,0); \foreach \x/\y in {0/0,2/0,0/2}{\fill[coarse] (\x,\y) circle (2.5pt);} \fill[active] (1,0) circle (2.5pt); \foreach \x/\y in {0/1,1/1}{\draw[gray!65,fill=white] (\x,\y) circle (2.5pt);} \node[below,font=\small] at (1,-.12) {Fine lattice}; \begin{scope}[shift={(3.6,0)}] \draw[gray!65,dashed,thick] (0,0)--(0,2)--(2,0); \draw[active,very thick] (0,0)--(2,0); \foreach \x/\y in {0/0,2/0,0/2}{\fill[coarse] (\x,\y) circle (2.5pt);} \node[below,font=\small] at (1,-.12) {Collision graph}; \end{scope} \draw[->,thick] (2.5,.5)--(3.1,.5); \end{tikzpicture} \caption{Schematic reduction of a coarse triangle. Blue vertices are always deleted even points. An orange midpoint is deleted and activates its coarse edge; unfilled midpoints do not. For an actual example, add $(2,6)$ to the three displayed fine vertices: the horizontal midpoint $(3,6)$ is nonprimitive, whereas $(2,7)$ and $(3,7)$ are primitive.}\label{fig:reduction} \end{figure} \begin{lemma}[Odd--even equality]\label{lem:parity} For every $m\ge1$, $\beta_1(K_{2m-1})=\beta_1(K_{2m})$. \end{lemma} \begin{proof} A new primitive vertex on $x+y=2m$ has both coordinates odd. Its neighbors on that outer line are even and deleted. Its only possible retained neighbors are the two vertices one step inward, which are adjacent if both are retained. Hence each new vertex attaches along an empty, singleton or one-edge link; in the latter two cases its cone collapses back to the link. In the empty case it is an isolated new vertex. These operations do not change first homology. The two corner vertices on the new side are nonprimitive (their nonzero coordinate is $2m$). The new primitive vertices are pairwise nonadjacent, so their links are already in $K_{2m-1}$ and the collapses can be performed independently, including when $m=1$. \end{proof} \section{Periodic prime sieves and the existence of the limit}\label{sec:sieve} The main limiting quantities below are normalized per fine vertex: $h$ counts holes, $h_P$ coarse components at cutoff $P$, $\rho,\rho_P$ cycle ranks, $e,e_P$ bonds, and $\delta_P=e-e_P$ omitted bonds. The island density $b_0$ counts visible components, whereas $B_L$ bounds the conditional fraction of cycle-eligible new bonds. For real $P\ge3$ define $\mathcal P_P=\{p:3\le p\le P,\ p\text{ prime}\}$, $M_P=\prod_{p\in\mathcal P_P}p$. Let $G^{(P)}$ have all coarse vertices and only bonds in \eqref{eq:bond} witnessed by some $p\in\mathcal P_P$. No midpoint has both coordinates even, so the union over all odd primes is $G$. The truncated graph has period $M_P\Z^2$. Write $E_P(m)$ and $E(m)$ for the edge counts of $G^{(P)}[T_m]$ and $G_m$. Our densities are normalized by $n_{2m}$ (fine-lattice points), not $n_m$ (coarse-lattice points). In particular $n_m/n_{2m}\to1/4$. Put \begin{equation}\label{eq:edges} e_P=\frac34\left(1-\prod_{3\le p\le P}(1-p^{-2})\right),\qquad e=\frac34-\frac6{\pi^2},\qquad \delta_P=e-e_P. \end{equation} In all products, $p$ denotes a prime unless otherwise stated. The CRT gives $\lim_m E_P(m)/n_{2m}=e_P$: for each of the three orientations, $2u+s\equiv0\pmod p$ is a single residue class of $u$ modulo $p$, independent across primes. For the full sieve, the tail is controlled by counting points divisible by $p>P$. Since midpoint coordinates in $T_{2m}$ are between $0$ and $2m$, and the midpoint is never the origin, all its prime divisors are at most $2m$. For such $p$, the number of multiples of $p$ in the triangle is $O(m^2/p^2+m/p+1)$ uniformly. Summing over integers $p>P$ yields an upper bound $O(m^2/P+m\log m+m)$ for the omitted set. Periodic densities therefore converge to the full density. By $\prod_p(1-p^{-2})=6/\pi^2$ we obtain the second formula of \eqref{eq:edges} and, more precisely, \begin{equation}\label{eq:tail} \lim_m\frac{E(m)-E_P(m)}{n_{2m}}=\delta_P\le \frac34\sum_{p>P}p^{-2}. \end{equation} The same argument proves existence of every finite local pattern density of bonds: the difference between full and truncated pattern events is bounded by the union of the finitely many translated omitted-bond events. For fixed offsets the counting box expands by only a fixed amount; a nonzero midpoint has maximum absolute coordinate $O(m)$ and the midpoint cannot be zero because its parity is nonzero. Residue counts in a fixed-period triangle are $n_m/M^2+O_M(m)$ per class, including axes and corners. Thus the stated CRT probabilities apply to triangle averages, not merely box averages. This is an elementary instance of the established pattern-frequency theory \cite{PH}. \begin{lemma}[Periodic component density]\label{lem:periodic} In the quotient graph of $G^{(P)}$ by $M_P\Z^2$, let $F_P$ be the number of components whose lifts to $\Z^2$ are finite. Then \[ h_P:=\lim_{m\to\infty}\frac{c(G^{(P)}[T_m])}{n_{2m}}=\frac{F_P}{4M_P^2}. \] \end{lemma} \begin{proof} A component in the finite quotient lifts to finite components precisely when all closed walks in it have zero displacement in $M_P\Z^2$. In that case each lift contains one copy of every quotient vertex of that component, and exactly one translate per period cell of any chosen representative. Each finite lifted component accordingly contributes one per $M_P^2$ coarse vertices, up to a boundary error $O_P(m)$ in $T_m$. There are finitely many quotient components, hence their finite lifts have uniformly bounded sizes for fixed $P$. A restriction of an infinite lifted component to $T_m$ can only have components meeting the boundary, since a path to infinity must leave $T_m$ by a nearest-neighbor edge. Their number is $O(m)$ in total. Dividing by $n_{2m}$ proves the formula. The displacement criterion also provides an exact finite algorithm: traverse each quotient component with integer lifted coordinates; inconsistency along a previously visited vertex detects nonzero displacement. \end{proof} \begin{theorem}[Existence and effective approximation of the hole density]\label{thm:limit} The limit \begin{equation}\label{eq:h} h=\lim_{k\to\infty}\frac{\beta_1(K_k)}{n_k} \end{equation} exists. The sequence $h_P$ decreases to $h$ and obeys \begin{equation}\label{eq:sandwich} h_P-\delta_P\le h\le h_P, \qquad \lim_{k\to\infty}\frac{\beta_1(K_k)}{k^2}=\frac h2. \end{equation} In particular the constant can be approximated to arbitrary prescribed absolute accuracy with finite exact quotient counts and a prime-sieve tail bound. \end{theorem} \begin{proof} Adding a single bond decreases the number of components by either zero or one. For every $m$, \[ c(G^{(P)}[T_m])-(E(m)-E_P(m))\le c(G_m)\le c(G^{(P)}[T_m]). \] The difference between $c(G_m)$ and $c_{\rm int}(G_m)$ is at most the number of boundary vertices, $O(m)$. Divide by $n_{2m}$ and apply Theorem~\ref{thm:reduction}, Lemma~\ref{lem:periodic} and \eqref{eq:tail}. The limsup and liminf differ by at most $\delta_P$ for each $P$, and $\delta_P\to0$. Monotonicity of the truncated component counts gives monotonicity of $h_P$; the bounds give its limit. Lemma~\ref{lem:parity} extends the even subsequence limit to all $k$. Finally $n_k/k^2\to1/2$. \end{proof} \begin{corollary}[Cycle-density identity]\label{cor:euler} Let $r(H)=|E(H)|-|V(H)|+c(H)$ be the cycle rank of a finite graph. The limits $\rho=\lim_m r(G_m)/n_{2m}$ and $\rho_P=\lim_m r(G^{(P)}[T_m])/n_{2m}$ exist and \begin{equation}\label{eq:euler} h=\frac6{\pi^2}-\frac12+\rho,\qquad \rho_P=e_P-\frac14+h_P,\qquad 0\le\rho-\rho_P\le\delta_P. \end{equation} In particular $h>0$. \end{corollary} \begin{proof} Use the graph identity for cycle rank, $n_m/n_{2m}\to1/4$, and \eqref{eq:edges}--\eqref{eq:h}. Adding an edge increases cycle rank by at most one and never decreases it. The lower bound $6/\pi^2-1/2>0$ follows from $\rho\ge0$. \end{proof} \section{Planar duality and elementary faces}\label{sec:faces} Put $Q_j=\prod_{p\ge3}(1-j/p^2)$ for $0\le j\le5$. The complementary dual graph has one vertex per elementary triangle and joins adjacent triangles exactly when their shared coarse bond is absent. In $T_m$ include an exterior dual vertex on $S^2$ and connect across absent boundary sides. The components of this dual graph are the faces of $G_m$: deleting a primal edge merges its incident regions if distinct, and otherwise leaves them unchanged. Thus its component count minus one is the cycle rank, by Euler's planar formula. This construction works for disconnected graphs, isolated primal vertices, bridges and non-simple boundary walks. The infinite locally finite construction has the same face incidence; a bounded face corresponds to a finite dual component.\label{sec:dualface} \begin{proposition}[Elementary face probabilities]\label{thm:faces} The probability of an elementary triangular bounded face is $t=1-3Q_1+3Q_2-Q_3$. That of a two-triangle rhombic bounded face is $b=Q_1-4Q_2+6Q_3-4Q_4+Q_5$. Hence $\rho\ge t/2+3b/4$. \end{proposition} \begin{proof} The three midpoint offsets of the triangle are $\e_1,\e_2,\e_1+\e_2$, with $\e_1=(1,0),\e_2=(0,1)$. Require all three active. For the rhombus with vertices $0,\e_1,\e_2,\e_1+\e_2$, require its four outer midpoints $\e_1,\e_2,2\e_1+\e_2,\e_1+2\e_2$ active and its diagonal midpoint $\e_1+\e_2$ absent. These offsets are distinct modulo every odd prime in each test; CRT and inclusion--exclusion give the formulas. They count actual faces, hence are disjoint as face interiors; there are two triangles and three candidate rhombi per coarse vertex asymptotically. Fine normalization gives the result. \end{proof} For general finite triangle sets $A$, the event that $A$ is exactly a dual component requires active boundary sides and a connected complementary dual graph on $A$. Summing exact internal-edge configurations over normalized shapes of size at most $K$ gives a lower estimate $\Phi_K$ for $\rho$; larger dual components consume at least $K+1$ triangles each, so $0\le\rho-\Phi_K\le1/[2(K+1)]$. The supplement gives the full finite-pattern formula and certificate. This count permits dangling active edges inside a face, rather than assuming all its interior sides are absent. \section{Holes and islands: the complementary correspondence}\label{sec:islands} Call a finite component of the primitive-site graph an island. In an even finite triangle we call a visible component interior if it avoids the three boundary sides. The deleted-site and visible-site components are different objects. We now show why the nontriangular bounded faces of the midpoint graph encode the latter. \begin{theorem}[Pointwise island--face correspondence]\label{thm:islands} Every bounded face of the infinite midpoint graph other than the interior of a single elementary triangle corresponds to exactly one finite visible-site component, and conversely. For $G_m$, bounded faces consisting of at least two dual triangle vertices correspond bijectively to interior components of the visible graph on $T_{2m}$. In both cases the component consists of the visible midpoints of absent coarse sides between the triangles in that dual component. \end{theorem} \begin{proof} Use the mid-edge subdivision of Theorem~\ref{thm:reduction}. All even vertices are deleted, so every visible site is the midpoint of exactly one absent coarse side. A fine edge joining two such midpoints lies in a single subdivided coarse triangle, and conversely any two distinct sides of a coarse triangle have adjacent midpoints. Thus visible-site adjacency agrees precisely with edge incidence in the dual subgraph on triangles: the visible graph in the full plane is the line graph of the complementary dual graph (isolated dual vertices contribute no line-graph vertices). For a connected graph containing an edge, its line graph is connected; a finite component with at least one edge has a nonempty finite connected line graph and an infinite component of this locally finite dual graph has infinitely many edges. The dual component with no edges is an isolated triangle whose three sides are all active. Planar duality from Section~\ref{sec:dualface} identifies the finite dual components with bounded faces. For a finite triangle the exterior vertex may identify boundary edges that are not mutually adjacent in the visible graph, so apply the line-graph statement only to dual components not containing the exterior. A visible component meets the boundary exactly when it contains a midpoint of an absent boundary side. Such a side is incident to the exterior dual vertex. Conversely any chain of dual edges from an interior triangle to the exterior gives a chain of midpoint adjacencies until the first boundary side. This proves the finite assertion and handles isolated visible sites, bridges, and annular faces without requiring simple cycles. \end{proof} Write $J_j=\prod_p(1-j/p^2)$ for $j=1,2,3$, and let \[ b_0=\lim_{k\to\infty}\frac{\beta_0(K_k)}{n_k},\qquad \bar\chi=J_1-3J_2+2J_3. \] The limit is justified in the next proof; the same density is obtained by counting only interior visible components. The normalized density $b_0$ counts islands, not their vertices. \begin{theorem}[Two Euler identities]\label{thm:twoeuler} The island density exists and \begin{equation}\label{eq:twoeuler} h=-\bar\chi+b_0=3J_2-2J_3-J_1+b_0, \qquad \rho=\tfrac12(1-3Q_1+3Q_2-Q_3)+b_0. \end{equation} \end{theorem} \begin{proof} The local number of retained vertices, edges and filled elementary triangles per fine vertex tends respectively to $J_1,3J_2,2J_3$. Indeed each edge or triangle has distinct vertex residues modulo every prime: differences are minimal directions with a coordinate $\pm1$. CRT and the prime-tail estimate of Section~\ref{sec:sieve} give the three products, treating the origin separately as one point. Boundary counts of the potential edges and triangles are $O(k)$. The finite planar complex has $H_2=0$ by the collapse in Lemma~\ref{lem:dual}; hence $\beta_1=\beta_0-\chi$ even if disconnected. Since $h$ exists, the limit for $\beta_0/n_k$ is $h+\bar\chi$. Removing components meeting the boundary changes at most $3k$ counts. Every interior component is a finite component of the infinite visible graph, and a finite component fully contained in the interior is counted. Its density follows equivalently by fixed-size truncation, with components larger than $L$ contributing at most $1/(L+1)$. This proves the first formula. Since $J_1=3Q_1/4$, $J_2=Q_2/2$, $J_3=Q_3/4$, comparison with \eqref{eq:euler} proves the second. Alternatively Theorem~\ref{thm:islands} proves it directly: singleton dual components have density $(1-3Q_1+3Q_2-Q_3)/2$ in fine normalization and all remaining bounded faces count islands. \end{proof} \begin{remark}[Short-range exclusion and the hexagon] An elementary isolated dual triangle requires three active sides; a two-triangle face requires four active sides and one absent diagonal. Compare their probabilities $t$ and $b$ from Proposition~\ref{thm:faces} to the artificial independent-edge reference $p_e^3$ and $p_e^4(1-p_e)$, where $p_e=1-8/\pi^2$. The reference is not a model of the arithmetic dependencies. An instructive opposite phenomenon occurs for the six-triangle hexagon centered at a coarse vertex: the six boundary midpoint offsets, sums of consecutive nearest-neighbor directions, fall into just two classes $(1,1)$ and $(-1,-1)$ modulo $3$. The six spoke offsets are the six distinct nonzero root residues, disjoint from these two classes. Thus a single residue choice for $p=3$ can activate three alternating boundary bonds at once without activating any spoke. The site-cluster expansion below also captures this geometry as a six-midpoint ring. The finite cutoff $R=10000$ gives approximate ratios to the full-density independent reference of $0.316$ (triangle), $0.127$ (rhombus), and $3.922$ (isolated-center hexagon), calculated from the same finite pattern products. These ratios are descriptive, not certified limiting constants. The enhanced size-six contribution is a local arithmetic correlation; no conclusion about the existence of infinite clusters follows from comparison with independent bonds. \end{remark} \section{Prime matchings and an improved tail estimate}\label{sec:matching} The decomposition into prime sieves has a special property not visible in the one-point visibility density. \begin{proposition}[Matching and two-prime forest]\label{prop:matching} For any odd prime $p$, the edges witnessed by $p$ form a matching. For any two distinct odd primes $p,q$, their union is a forest, all of whose components have at most four vertices. If $p,q\ge5$, each component has at most three vertices; a component with four vertices can occur only if one of the primes is 3. \end{proposition} \begin{proof} At a vertex $u$, the six incident midpoint residues are $2u+s$, $s\in\mathcal S$. They are distinct modulo any odd prime, because no difference of distinct directions has all coordinates divisible by such a prime. Thus at most one incident edge is witnessed by $p$, establishing the matching property. Assume $q\ge5$ after possibly relabelling. Along a simple path, label each edge by any one of its witnesses in $\{p,q\}$. Consecutive labels differ. A subpath of three edges with labels $q,p,q$ is impossible. If its successive coarse directions are $d_1,d_2,d_3\in\mathcal S$, the difference of the first and third fine midpoints is $d_1+2d_2+d_3$. Each coordinate has magnitude at most 4, and this difference is nonzero because distinct coarse edges have distinct midpoints. It cannot belong to $q\Z^2$. An alternating path of length four would contain $q,p,q$; an alternating cycle likewise would contain it (a cycle is necessarily even because labels alternate). Thus there are no cycles or paths of length four. Since the degree is at most two, the components are paths of at most three edges, including isolated vertices. If both primes are at least 5, interchange their roles: the pattern $p,q,p$ is also impossible by the same nonzero displacement bound. Every three-edge alternating path has one of these two patterns. Hence all paths have at most two edges, giving the asserted sharper bound. An edge witnessed by both primes is an isolated two-vertex component of the two-prime graph, since each prime already occupies both endpoints. A multiply witnessed edge cannot create an exception: at a vertex shared with another edge, it could not share either prime, by the matching property. In particular any choice of witnesses on a simple path is alternating. \end{proof} At the first level the following conditional probability has a closed form: \begin{equation}\label{eq:B} B=1-2\prod_{p\ge3}(1-5/(p^2-1))<1.\end{equation} Indeed, given a central absent bond, the five other incident offsets at either endpoint are distinct modulo every odd prime and avoid the central residue. The conditional isolation probability at one endpoint is the product in \eqref{eq:B}. Both endpoints cannot be isolated: the eleven incident midpoint offsets together cover all nine classes modulo $3$. In the full sieve, the probability that both endpoints are nonisolated is therefore $B$. The product is positive by convergence of $\sum p^{-2}$. The size-sensitive theorem below includes this first-level bound. \subsection{Diameter exclusion and a hierarchy of cycle bounds} For an integer $L\ge1$, let $B_L$ denote the conditional spatial probability of the following event for the potential coarse edge $a=\{u,v\}$, given that $a$ is absent from $G$: \begin{equation}\label{eq:BL} \mathcal E_L(a)=\{C(u)=C(v)\}\ \cup\ \{|C(u)|>L,\ |C(v)|>L\}, \end{equation} where $C(w)$ denotes the component of $w$ in the graph, with infinite size allowed. Equivalently, the event fails if an endpoint belongs to a component of at most $L$ vertices which does not contain the other endpoint. This equivalent form is a finite local test: explore each endpoint out to graph distance $L-1$ in the underlying lattice and inspect all edges incident to the explored neighborhood. It detects every component of at most $L$ vertices together with its closed boundary. Thus the conditional probability exists by the finite-pattern argument in Section~\ref{sec:sieve}. The event is increasing under edge addition with $a$ held absent, and $B_{L+1}\le B_L$. Symmetry makes it independent of edge orientation. In particular $B_1=B$ of \eqref{eq:B}. \begin{lemma}[Diameter exclusion]\label{lem:diameter} Let $C$ be a component before a new prime $q$ is added. If $C$ has $\ell^\infty$ diameter at most $D$ and $q>2D+2$, at most one distinct $q$-witnessed edge can touch $C$. \end{lemma} \begin{proof} Choose endpoints $u_1,u_2$ in $C$ of two such edges and write their fine midpoints as $2u_i+s_i$, $s_i\in\mathcal S$. Their difference has infinity norm at most $2D+2$, is nonzero by uniqueness of edge midpoints, and belongs to $q\Z^2$. This is impossible under the stated strict inequality. \end{proof} \begin{theorem}[Size-sensitive cycle tail]\label{thm:hierarchy} For every positive integer $L$ and real $P\ge\max(3,2L)$, \[ 0\le\rho-\rho_P\le B_L\delta_P,\qquad h\le\frac6{\pi^2}-\frac12+\rho_P+B_L\delta_P. \] \end{theorem} \begin{proof} First work in the finite induced graph on $T_m$ and add a prime $q>P$ to an old finite prime sieve $S$. Contract each connected component of the old graph to one vertex and keep the genuinely new edges, including loops and parallel edges. The cycle-rank increment is the cycle rank of this multigraph: the identity $r=E-V+c$ gives this directly before and after contraction. If an old component has at most $L$ vertices, its diameter is at most $L-1$. Lemma~\ref{lem:diameter} shows that at most one new edge touches it. If that edge is not a loop, it is a bridge of the contracted graph and contributes no cycle rank. If it is a loop, its endpoints were in the same old component; this possibility must be counted. Removing all such bridges leaves only edges joining two large components or edges whose endpoints were in one old component. The rank increment is at most their number, so every eligible edge satisfies $\mathcal E_L$ in the old graph. For $L=2$ a genuinely new loop inside a small component is impossible, since a connected two-vertex component already contains its only potential edge. For $L\ge3$ this exception cannot be omitted. For fixed $S,q,L$, finite-boundary effects in the local test are $O_L(m)$ vertices. Conditional on the candidate edge being absent in $S$, the residue of the coarse translation modulo each $p\in S$ is uniform over the $p^2-1$ classes avoiding that edge. Additional prime residues conditioned to avoid the edge have independent CRT distributions. They can only increase $\mathcal E_L$. Consequently the conditional probability for $S$ is bounded by $B_L$, obtained as the limit of the full conditional finite-pattern frequencies. Activation by $q$ is independent of the old residues. Taking the spatial limit gives the increment bound $B_L$ times the genuinely new edge density. Sum this inequality over $PP$, and a fixed $L$ with $2L2L$, $\Delta\rho_q/\Delta e_q\le B_L$. For the reverse inequality, fix a finite window and a finite prime cutoff $R$. Every genuinely new bond with endpoints already connected by a path in that old window contributes one loop after contracting old components. The intervening primes between $R$ and $q$ only add bonds and cannot destroy the path. Adding each loop raises $E$ by one without changing $V$ or $c$, so these loops add individually to rank, even if they meet other new edges. For $q>R$, conditional on absence from all primes less than $q$, the old primes up to $R$ have exactly their distributions conditioned on central absence, independently of activation by $q$. Thus the ratio is at least the probability of that finite connection test. Boundary loss of these fixed-window tests is $O_{R,\mathrm{window}}(m)$ for each fixed window and cutoff. Increase both the cutoff and the window to exhaust all finite paths: continuity from below in the product measure gives lower limit at least $\theta$. Now let $L$ increase for the upper limit. This proves the first limit. CRT gives $\Delta e_q=(3/4)q^{-2}\prod_{3\le pP}p^{-2}$. By the prime number theorem and partial summation, $\sum_{p>P}p^{-2}=-\pi(P)/P^2+2\int_P^\infty \pi(t)t^{-3}\,dt\sim1/(P\log P)$, with $\pi(t)$ the prime counting function. \end{proof} The conditional theorem uses a named external preprint. The deterministic conclusions arise from finite spatial-pattern frequencies, not from asserting a typical origin or independent bonds. Neither existence of $h$ nor the bounds in the next section rely on it. The companion technical supplement contains the finite-pattern details and numerical certificates for the present planar model only. The odd--even equality also rules out a single linear second-order term $ck$ in $\beta_1(K_k)=hk^2/2+ck+o(k)$: consecutive even--odd differences would be $hk+o(k)$ instead of zero. \section{Exact computations and certified bounds}\label{sec:computations} The accompanying C++17 programs compare direct deleted-site counts, the reduced graph, and Euler counts. A lifted quotient traversal detects torus winding; independent union--find counts check the quotient component and edge totals. Tables~\ref{tab:finite} and \ref{tab:periodic} report the retained exact data from the earlier computations. The numerical certificates use these integer data, not a fit to the finite observations. \begin{table}[ht] \centering\small \caption{Direct and reduced counts at selected even levels. Here $r(G_m)$ is the cycle rank of the coarse graph, including boundary components.}\label{tab:finite} \begin{tabular}{rrrrr}\toprule $k=2m$ & $n_k$ & $\beta_1(K_k)$ & $r(G_m)$ & $\beta_1/n_k$ \\\midrule 50 & 1,326 & 131 & 2 & .098793\\ 100 & 5,151 & 517 & 10 & .100369\\ 200 & 20,301 & 2,192 & 40 & .107975\\ 500 & 125,751 & 13,604 & 188 & .108182\\ 1,000 & 501,501 & 54,417 & 616 & .108508\\ 2,000 & 2,003,001 & 218,256 & 2,468 & .108964\\ 4,000 & 8,006,001 & 872,199 & 10,136 & .108943\\ 8,000 & 32,012,001 & 3,492,812 & 40,226 & .109109\\ 22,000 & 242,033,001 & 26,416,327 & 303,174 & .109143\\\bottomrule \end{tabular} \end{table} \begin{table}[ht] \centering\small \caption{Periodic sieve counts. $M=M_P$, $E$ counts quotient edges, $F$ counts components with finite lifts. The listed cases have no winding components.}\label{tab:periodic} \begin{tabular}{rrrrrr}\toprule $P$ & $M$ & $M^2$ & $E$ & $F$ & $\rho_P$ \\\midrule 3 & 3 & 9 & 3 & 6 & 0\\ 5 & 15 & 225 & 99 & 126 & 0\\ 7 & 105 & 11,025 & 5,427 & 5,610 & .000272109\\ 11 & 1,155 & 1,334,025 & 684,315 & 652,206 & .000467757\\ 13 & 15,015 & 225,450,225 & 118,966,995 & 107,057,562 & .000636872\\\bottomrule \end{tabular} \end{table} At $P=13$, Lemma~\ref{lem:periodic} and Corollary~\ref{cor:euler} give the exact rational value \[ \rho_{13}=\frac{118966995-225450225+107057562}{4\cdot15015^2} =\frac{4351}{6831825}. \] In particular the cycle rank of the quotient (with all lifts finite) is $574332$. The inequality of Theorem~\ref{thm:hierarchy} further sharpens the matching bound without requiring any larger periodic enumeration. The $k=22000$ finite calculation gives $\beta_1/k^2\approx0.0545791880$; it is not itself a bound on the limit. \begin{corollary}[Computer-assisted numerical enclosure]\label{cor:certificate} The finite enumerations and rational certificates specified below give \begin{equation}\label{eq:interval} \boxed{0.109179780}=0$. All terms are nonnegative for feasible states. Outward integer rounding and a union bound for survival of $A$ under omitted primes give $b_0>0.00017854739$. The local Euler term has the certified lower bound $-\bar\chi>0.10900123267$, from Euler products with explicit tails and Machin arctangent bounds. Their exact rational sum exceeds $0.10917978$. The positive recurrence in \texttt{island\_positive.py} is an alternative to the signed symmetric inclusion--exclusion in \texttt{island\_certificate.py}; the older dual-face certificate through eight triangles gives $h>0.10917927$ by a separate geometric enumeration. The site procedures share shape generation and local arithmetic but not the large-prime probability recurrence. \emph{Upper endpoint.} At each near prime $13R}(p^2-1)^{-1}$ (see supplement). This yields $h<0.10923686$ without a larger quotient than $15015$. At $q=17$, \texttt{prefix\_diameter.cpp} counts exactly 557383680 absent bonds in the period $M=15015$, of which 29121696 are chords. Its winding-aware traversal verifies that no absent nonchord joins two lifted components both of diameter greater than 7. Proposition~\ref{prop:chords} therefore gives the exact increment \[ \Delta\rho_{17}=\frac{29121696}{4\cdot15015^2\cdot17^2}. \] Use this instead of the corresponding near-prime upper term. For $z=6/\pi^2$, the remaining tail expression $z-1/2+\rho_{13}+\Delta\rho_{17}+\sum_{1713$, let $S_{0.10914952.} \] This is strictly stronger than the earlier length-three chord certificate but weaker than the island-based lower endpoint of Corollary~\ref{cor:certificate}. \end{proposition} \begin{proof} Fix the coarse bond between $(0,0)$ and $(1,0)$. All simple paths avoiding that bond and of length at most six form a finite list of 144 paths. We deliberately use only a reproducible \emph{subset} of 23 of these paths, supported on 18 candidate bonds including the target. The subset is generated deterministically from paths of length at most three, then enlarged by all paths already supported and by a lexicographically selected sequence of paths subject to a 19-bond cap. Every selected path has length at most six. This is a sufficient event for connection, not the full event $\alpha_6$. For each odd prime $p$, the randomly translated forbidden residue modulo $p$ activates the selected bonds with the corresponding midpoint residue. Conditional on the target bond being absent, the residue excludes exactly one class out of $p^2$; the integer transition counts of activated-bond masks therefore have denominator $p^2-1$. Midpoint collisions modulo small primes are retained without an independence assumption. Let $\widehat\alpha_6(S)$ be the resulting exact conditional probability that at least one of the 23 selected paths is active, using primes in $S$. Then $0\le\widehat\alpha_6(S)\le\alpha_6(S)\le\vartheta(S)$, and $\widehat\alpha_6$ is monotone under adding primes. At $q=17$ we retain the exact independently counted increment $\Delta\rho_{17}$ of Proposition~\ref{prop:chords}. For $19\le q\le97$ replace $\alpha_3(S_{17}N_q$ in the statements of this article. The chord series alone is not asserted to equal the complete cycle density. \end{remark} \begin{remark}[Order-free prime interactions] For any \emph{finite} set $S$ of odd primes, Boolean-lattice M\"obius inversion gives the canonical coefficients \[ I(T)=\sum_{U\subseteq T}(-1)^{|T|-|U|}\rho(U),\quad \rho(S)=\sum_{T\subseteq S}I(T). \] Since the one- and two-prime graphs are forests, $I(T)=0$ for $|T|\le2$. The first possibly nonzero interactions have order three. For example, $I(\{3,5,7\})=\rho(\{3,5,7\})=12/44100=1/3675$; this value can be obtained from the period-$105$ quotient and the one- and two-prime forest theorem. This is an exact finite combinatorial transform, not a claim of convergence of an infinite Dirichlet series or of a new general M\"obius-inversion theorem. It is a natural starting point for studying order-independent prime interactions. \end{remark} \section{Relation to the literature and further questions}\label{sec:literature} Baake, Moody and Pleasants \cite[Proposition 4 and Theorem 3]{BMP} prove positive-density arbitrarily large vacant regions and determine diffraction of visible lattice points. Pleasants and Huck \cite[Section 3, Theorem 1]{PH} give the mixed finite-pattern frequency formula. The Euler products used here, including $Z(L)$, are applications of that formula restricted to the even translation class (the odd-prime CRT calculation gives the restriction directly). They are not proposed as new general arithmetic frequency identities. Le Fourn, Liu and Martineau \cite[Theorems 1.6--1.7, Sections 5.2--5.3 and 5.6.1]{LLM} explicitly treat the triangular and $D_d$ lattices with minimal-vector Cayley adjacency. In the random coprime coloring they prove an almost surely unique infinite visible cluster and, for these lattices, no infinite nonvisible cluster. Their result concerns a stationary random translate model; we do not infer the corresponding deterministic assertion for the fixed origin solely from an almost-sure statement. Our normalized counts require no such assertion and concern finite components, not the number of infinite clusters. Cellarosi \cite[Definitions 1.1--1.3, Proposition 2.4]{Cellarosi} studies non-$r$-free points on the cubical nearest-neighbor graph (four neighbors in dimension two). His component representatives belong to finite components containing an $R$-deep point; the size truncation in Proposition~2.4 has remainder at most $1/(L+1)$. That technique applies with our different adjacency and without a depth condition. His Theorem~1.7 and Corollary~1.8 concern large-$R$ logarithmic density asymptotics, not Betti densities or midpoint cycle rank. We claim no originality for the general finite-pattern or bounded-size approximation machinery. Our specific structural results are the midpoint and island correspondences, prime matchings and forests, diameter exclusion and stratified cycle bounds. Proposition~\ref{prop:decomp} makes explicit the positive chord/nonchord decomposition, and Proposition~\ref{prop:chordcertificate} provides a structurally separate, albeit numerically weaker, length-six lower certificate. Full global priority is not established. Using only full-graph $B_L$ has an intrinsic floor $B_L\ge\theta$ at fixed cutoff; an observed decimal plateau is not itself a certified barrier. The stratified theorem avoids that loss. Remaining questions include exponential island tails, exact evaluation of $h$, quantitative prime-increment convergence, and parity-specific second-order expansions. Extensions to higher-dimensional lattices require separate constructions and are outside the scope of this planar preprint. \section*{Reproducibility and status} The companion technical supplement specifies the enumeration, finite-pattern probabilities, rational rounding, and boundary conventions. The selected length-six chord-only certificate is not independent numerical corroboration of the sharp island endpoint: its rigorous value $0.10914952$ remains lower than the island bound $0.10917978$. Improvements at component-size cutoffs $L=6,7,8$ for the earliest primes remain uncompleted and are not included in the certified enclosure. The code and recorded data are supplied with this preprint. Run the full certificates, not just the saved-number checks, to reproduce the numerical enclosure. The numerical statements are computer-assisted, and their implementation is not formally verified in a proof assistant. The other main results are analytic; the prime-increment law and its rate corollary are expressly dependent on the stated external nonpercolation result. All finite-window measurements are descriptive, not statistical confidence intervals. A permanent code link can be added to the public repository record when the companion files are deposited. The corrected saved-certificate checker reads but never writes its input data; a complete replay of the expensive signature generator is recorded in the accompanying reproducibility archive. \textbf{AI-assistance disclosure.} This manuscript was developed with substantial generative-AI assistance from ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), and Copilot (Microsoft) in conjecture generation, proof formulation and development, computational design and execution, criticism, literature exploration and writing. The text and code are substantially AI-assisted; the systems are not authors or human referees. The author is responsible for reviewing, understanding, verifying and approving the entire work before dissemination. No human peer review, formal verification, or exhaustive priority determination is claimed. \begin{thebibliography}{99} \bibitem{GAF} S. N. Jihad, \emph{The Generative Arithmetic Fabric: A Unified Geometric Workbench for Arithmetic and Generator Deformation}, preprint v2.1, October 2026, Zenodo. \url{https://doi.org/10.5281/zenodo.23126296}. \bibitem{Kahle} M. Kahle, \emph{Topology of random simplicial complexes: a survey}, Contemporary Mathematics \textbf{620} (2014), 201--222. \url{https://arxiv.org/abs/1301.7165v2}. \bibitem{BMP} M. Baake, R. V. Moody and P. A. B. Pleasants, \emph{Diffraction from visible lattice points and $k$-th power free integers}, Discrete Math. \textbf{221} (2000), 3--42. \url{https://arxiv.org/abs/math/9906132v2}. \bibitem{PH} P. A. B. Pleasants and C. Huck, \emph{Entropy and diffraction of the $k$-free points in $n$-dimensional lattices}, Discrete Comput. Geom. \textbf{50} (2013), 39--68. \url{https://arxiv.org/abs/1112.1629v3}. \bibitem{LLM} S. Le Fourn, M. Liu and S. Martineau, \emph{Percolative properties of the random coprime colouring}, preprint (2025), arXiv:2509.08452v1. \url{https://arxiv.org/abs/2509.08452v1}. \bibitem{Cellarosi} F. Cellarosi, \emph{Density of large holes among power-free lattice points}, preprint (2026), arXiv:2609.25457v2. \url{https://arxiv.org/abs/2609.25457v2}. \bibitem{Hatcher} A. Hatcher, \emph{Algebraic Topology}, Cambridge University Press, 2002, Section 3.3, Corollary 3.45, p. 255. \url{https://pi.math.cornell.edu/~hatcher/AT/ATpage.html}. \end{thebibliography} \end{document}