---
title: 'Coarse Grid Theorem: Structures & Applications'
url: https://www.emergentmind.com/topics/coarse-grid-theorem
type: topic
---

# Coarse Grid Theorem: Structures & Applications

Searching arXiv for recent papers on "coarse grid theorem", "counterexample coarse grid theorem", and related graph theory / numerical linear algebra usages.
“Coarse Grid Theorem” appears in the papers considered here in several distinct settings. In large-scale graph theory it names a conjectured coarse analogue of the Robertson–Seymour grid theorem, formulated using fat minors and quasi-isometry, and it is now known to be false in that form [2508.15342]. In the same broad area, positive coarse grid phenomena survive under different hypotheses, notably in the coarse Halin grid theorem for asymptotic minors [2507.12973]. In numerical linear algebra, the phrase refers to a coarse-grid approximation theorem for a Galerkin coarse problem constructed from a discontinuous, subdomainwise polynomial basis [1504.00907]. This suggests that the term is best understood as a family of theorem-schemas about grid-like structure or coarse-level approximation, rather than as a single canonical result.

## 1. Graph-theoretic formulation: fat grids, quasi-isometry, and bounded tree-width

The conjectured coarse grid theorem of Georgakopoulos and Papasoglu was formulated as follows: for every \(K,n \in \mathbb{N}\), there exist \(M,A,g \in \mathbb{N}\) such that every graph with no \(K\)-fat \((n\times n)\)-grid minor is \((M,A)\)-quasi-isometric to a graph of tree-width at most \(g\) [2508.15342]. The intended meaning was that the absence of a sufficiently coarse grid minor should force bounded tree-width after passage to a quasi-isometric model.

The coarse notion of minor used here is a \(K\)-fat minor model. A model of a graph \(X\) in a graph \(G\) consists of disjoint connected branch sets \(U_x\) for \(x\in V(X)\) and branch paths \(E_e\) for \(e\in E(X)\), joining the appropriate branch sets and internally disjoint from all other branch sets. The model is \(K\)-fat if distinct objects in the model are pairwise at distance at least \(K\), except that a branch path may meet its two incident branch sets at endpoints [2508.15342]. This is the large-scale replacement for ordinary minors in the conjecture.

The quasi-isometric side of the statement is equally central. For \(M \ge 1\) and \(A \ge 0\), an \((M,A)\)-quasi-isometry \(f:V(G)\to V(H)\) satisfies
\[
M^{-1} d_G(u,v) - A \le d_H(f(u),f(v)) \le M d_G(u,v)+A,
\]
and every vertex of \(H\) lies within distance \(A\) of some \(f(v)\) [2508.15342]. The conjecture therefore mixed two kinds of coarse structure: exclusion of a fat grid minor on one side, and quasi-isometric reduction to bounded-tree-width geometry on the other.

A useful technical consequence recorded in the counterexample paper is that connected sets remain coarsely connected under quasi-isometry, and coarse separators are preserved after an explicit thickening. If \(X\) separates \(Y\) and \(Z\) in \(G\), then
\[
B_H(f(X),r),\qquad r=M(M(3A+1)+2M^{-1}A+1),
\]
separates \(f(Y)\) and \(f(Z)\) in \(H\) [2508.15342]. This separator-transport principle is one of the mechanisms through which the conjecture was tested.

## 2. Refutation of the conjectured coarse grid theorem

The conjecture is refuted by the theorem that for every \(M,A,n \in \mathbb{N}\) with \(M \ge 1\), there exists a graph \(G\) such that the \((154\times 154)\)-grid is not a \(3\)-fat minor of \(G\), and \(G\) is not \((M,A)\)-quasi-isometric to any graph with no \(K_n\) minor [2508.15342]. Since quasi-isometry to bounded-tree-width graphs would in particular imply quasi-isometry to graphs excluding large clique minors, this directly contradicts the conjectured conclusion.

The construction is a recursively defined family \(G_{h,d,m}\), described as a slight modification of the recent counterexample to the weak coarse Menger conjecture from Nguyen, Scott and Seymour [2508.15342]. At the base level \(G_{h,d,2}\) consists of a binary tree of height \(h\), a long base path, and spine paths from leaves to the base path. For \(m>2\), the graph is built from a chain of copies of \(G_{h,d,m-1}\), a new binary tree, and additional spines. The crucial modification is that the construction does not include the dotted shortcut paths between the \(S\)-vertices or the \(T\)-vertices, and likewise not between the \(V_i^{m-1}\). The paper notes that with those shortcuts the construction would contain every graph as a fat minor; deleting them preserves the intended bottlenecks [2508.15342].

Two structural lemmas drive the argument. First, if \(P,Q\) are paths between \(\SSS(G)\) and \(\TTT(G)\), then either one of them passes through the root \(\RR(G)\), or \(d_G(P,Q)\le 2\) [2508.15342]. Second, for suitable \(\ell,h,d,m\), every set \(X\) of fewer than \(m\) vertices leaves an \(\SSS(G)\)-\(\TTT(G)\) path \(P\) with
\[
d_G(P, X\cup\{\RR(G)\})>\ell
\]
[2508.15342]. Together these statements show that the graph retains strong separator obstructions while refusing the sort of large-scale linkedness that a fat grid would force.

The exclusion of the \(3\)-fat \((154\times154)\)-grid is proved via a local obstruction and a tree-decomposition argument. Locally, no set \(V_i^j\) inside \(G_{h,d,m}\) contains a \(3\)-fat \(7\)-path-connected set [2508.15342]. Globally, a supposed \(3\)-fat grid model would produce a row yielding a \(3\)-fat \(77\)-path-connected set, and the tree decomposition forces a large subset of that row into one of the forbidden local pieces. The contradiction shows that the graph avoids the prescribed fat grid.

The second half of the theorem shows that this grid avoidance does not permit quasi-isometric simplification to any \(K_n\)-minor-free graph. Assuming an \((M,A)\)-quasi-isometry \(f:G\to H\), the separator transport lemma and the recursive structure of \(G_{h,d,m}\) are used to produce many disjoint \(\SSS(H)\)-\(\TTT(H)\) paths in \(H\), then to force them through a family of coarse separators, and finally to assemble a \(K_n\) minor in \(H\) [2508.15342]. The broader conclusion stated in the paper is that absence of a large fat grid minor does not force a graph to be quasi-isometric to bounded-tree-width graphs.

## 3. Positive coarse grid phenomena: the coarse Halin grid theorem

A different large-scale grid theorem survives in the setting of asymptotic minors. The paper “A coarse Halin Grid Theorem with applications to quasi-transitive, locally finite graphs” proves that every one-ended, locally finite graph that contains the disjoint union of infinitely many rays as an asymptotic minor also contains the half-grid as an asymptotic minor [2507.12973]. More precisely, if \(G\) is one-ended and locally finite and \(\omega\cdot R \prec_\infty G\), then \(HG \prec_\infty G\).

The underlying minor relation is again coarse. A graph \(X\) is an asymptotic minor of \(G\), written \(X\prec_\infty G\), if \(X\) is a \(K\)-fat minor of \(G\) for every \(K\) [2507.12973]. The paper also works with end-specific models \(X\prec^\varepsilon_K G\), diverging minors, and ultra-fat minors. The quantitative form of the main theorem states that if
\[
\omega\cdot R \prec^\varepsilon_{2K-1} G,
\]
then
\[
HG \prec^\varepsilon_K G,
\]
and if \(G\) contains \(\varepsilon\)-rays \(R_0,R_1,\dots\) such that
\[
d_G(R_i,R_j)\ge \max\{i,j\}\qquad (i\neq j),
\]
then
\[
HG \prec_{UF}^\varepsilon G
\]
[2507.12973].

The paper emphasizes that the correct coarse analogue of Halin’s theorem is not phrased directly in terms of thick ends. In the coarse setting, having a thick end alone is not enough in general: there are graphs with a thick end that do not contain the half-grid as an asymptotic minor [2507.12973]. The right hypothesis is the presence of \(\omega\cdot R\) as an asymptotic minor, together with geometric control on the ray-models. In the general, non-locally-finite case, two additional assumptions are imposed: the rays must be equivalent, and no two rays may be separated by deleting finitely many balls of finite radius [2507.12973].

The proof strategy is explicitly structural. Infinitely many coarse-disjoint rays are upgraded to a subdivision of the hexagonal half-grid, then contracted to a \(K\)-fat half-grid minor while preserving the fatness constraints [2507.12973]. For one-ended quasi-transitive locally finite graphs, the paper proves a stronger application:
\[
\text{Every one-ended, quasi-transitive, locally finite graph contains the half-grid as an ultra-fat minor.}
\]
Consequently such graphs contain the half-grid as an asymptotic minor and as a diverging minor; this includes all locally finite Cayley graphs of one-ended finitely generated groups [2507.12973].

## 4. Related exact and grid-type theorems in graph structure theory

The coarse graph-theoretic discussion sits beside several exact or non-coarse grid theorems. The “Directed Grid Theorem” proves the full directed analogue of the Robertson–Seymour grid theorem: there is a function \(f\) such that every digraph of directed tree-width at least \(f(k)\) contains a cylindrical grid of order \(k\) as a butterfly minor [1411.5681]. Equivalently, large bramble order forces a cylindrical grid as a butterfly minor. The theorem uses the modified version of directed tree-width with possibly empty bags, and this directed tree-width is closed under butterfly minors:
\[
dtw(H)\le dtw(G)\quad\text{whenever }H\text{ is a butterfly minor of }G
\]
[1411.5681].

The directed theorem is not formally called a coarse grid theorem. The paper explicitly states that it does not use that phrase as a formal named theorem, and that the result is the full structural analogue of the undirected grid theorem rather than a weaker approximation [1411.5681]. At the same time, its proof architecture passes through intermediate grid-like objects—webs, fences, pseudo-fences, and acyclic grids—before producing a cylindrical grid. The structural backbone is summarized as:
\[
\text{large directed tree-width} \Rightarrow \text{large bramble} \Rightarrow \text{web or cylindrical grid} \Rightarrow \text{fence} \Rightarrow \text{cylindrical grid}
\]
[1411.5681]. This suggests that “coarse grid” intuition may arise inside proofs even when the final theorem is exact.

A different nearby development is the induced-subgraph grid theorem for perforated graphs. For all \(c,o,t\in\mathbb N\) and \(s\in\mathbb N\cup\{0\}\), there exists
\[
\tau=\tau(c,o,s,t)\in\mathbb N
\]
such that every \((K_t,K_{t,t})\)-free \((c,o)\)-perforated graph \(G\) with \(\tw(G)>\tau\) contains a full \((s,o)\)-occultation [2305.15615]. In the \(c\)-perforated case, the resulting obstruction theory says that a graph has large treewidth if and only if it contains, as an induced subgraph, either a large complete graph, or a large complete bipartite graph, or a large full occultation [2305.15615]. The paper emphasizes that this is the first grid theorem for a hereditary induced-subgraph class that requires an obstruction other than subdivided walls and their line graphs.

Taken together, these results show that grid theorems in modern graph structure theory may involve ordinary minors, butterfly minors, asymptotic minors, or induced subgraphs, and that the “grid” itself may be a cylindrical grid, a half-grid, or a full occultation [1411.5681] [2507.12973] [2305.15615].

## 5. Numerical linear algebra usage: coarse-grid approximation in domain decomposition

In domain decomposition, “coarse grid theorem” refers to a different theorem entirely. The paper “The Discretely-Discontinuous Galerkin Coarse Grid for Domain Decomposition” starts from a symmetric positive definite linear system
\[
Au=f,
\]
partitions the discrete domain into non-overlapping subdomains
\[
\Omega = \bigcup_i \Omega_i,
\]
and constructs a coarse basis from the restriction of user-supplied generating vectors \(F\) to each subdomain [1504.00907]. For subdomain \(i\), the basis functions are
\[
\phi_i = R_i^T P_i F,
\]
the global coarse space is
\[
V_0 = \operatorname{range}(R_0^T),
\]
and the Galerkin coarse operator is
\[
A_0 = R_0 A R_0^T.
\]

The central theorem of Section 4, “Coarse Grid Analysis,” states that under moderate regularity assumptions the coarse Galerkin solution satisfies
\[
\|R_0^T u_0 - u\|_A \le c\, H^{1+p-q}\bigl(1+H^{q-\frac12}\bigr)\, |u|_{\widetilde{W}_\infty^{1+p}(\Omega)}
\]
for PDEs of degree \(2q\) [1504.00907]. The paper interprets this as showing that the coarse space is not just stable, but a high-order rediscretization of the fine PDE.

The proof uses Galerkin best approximation and a domain split into an interior region \(\Omega_+\) and a thin interface region \(\Omega_-\). Inside each subdomain, the coarse basis reproduces degree-\(p\) polynomials, yielding
\[
|v-u|_{A_+} \le c H^{1+p-q}|u|_{\widetilde{W}_\infty^{1+p}(\Omega)}.
\]
Near interfaces, the basis is discontinuous, but the bad region has small measure, and the interface contribution is controlled by
\[
|v-u|_{A_-} \le c\, H^{1+p-q}H^{q-\frac12}|u|_{\widetilde{W}_\infty^{1+p}(\Omega)}
\]
[1504.00907]. The same argument also yields an algebraic version in terms of \(\|u\|_2\).

This coarse-grid theorem is then tied to a two-level symmetric multiplicative overlapping Schwarz preconditioner. The paper reports optimal scaling, with convergence requiring a constant number of iterations, independent of fine problem size, on a range of scalar and vector-valued second-order and fourth-order PDEs [1504.00907]. Here “coarse grid” does not mean a graph-theoretic minor; it means a Galerkin coarse correction designed to approximate the error left after local Schwarz smoothing.

## 6. Conceptual status of the term

The current state of the graph-theoretic story is sharply split. The clean conjectured coarse grid theorem based on excluding a fat grid minor and demanding quasi-isometric bounded tree-width is false [2508.15342]. By contrast, strong positive statements do hold for asymptotic minors, especially when one works with \(\omega\cdot R\), one-ended locally finite graphs, and ultra-fat or diverging models of the half-grid [2507.12973].

This division matters for interpretation. The counterexample shows that a straightforward coarse translation of the classical Robertson–Seymour paradigm is too optimistic [2508.15342]. The coarse Halin theorem shows that robust large-scale grid forcing remains possible, but only after replacing the input and output notions by more carefully controlled asymptotic minors [2507.12973]. The nearby directed and perforated-graph theorems indicate that exact structural theorems often require changing the minor notion, the grid object, or the obstruction family itself [1411.5681] [2305.15615].

Outside graph theory, the same phrase designates a mathematically unrelated result about coarse-level approximation quality in multilevel solvers [1504.00907]. This suggests that “Coarse Grid Theorem” is not a single theorem with a fixed statement, but a context-dependent label for results in which a coarse-scale object—a fat grid, a half-grid, a cylindrical grid, an occultation, or a Galerkin coarse space—controls the large-scale structure of the underlying problem.

Source: https://www.emergentmind.com/topics/coarse-grid-theorem