Bounded Baker-treewidth Decompositions

• Bounded Baker-treewidth is a framework that refines tree decompositions by ensuring each bag induces a subgraph of bounded treewidth.
• It combines optimal global width control with local structural simplicity, drawing on Baker’s layering method for algorithmic advantages.
• The approach extends to coarse variants using unions of bounded-diameter balls, linking quasi-isometry concepts with efficient dynamic programming.

Searching arXiv for papers on bounded Baker-treewidth and closely related bounded-treewidth decompositions. arXiv search: "bounded Baker-treewidth" Bounded Baker-treewidth denotes a strengthened tree-decomposition regime in which width control is combined with local structural simplicity inside the bags. In the formulation centered on induced subgraphs, the object of study is a tree-decomposition of a graph $G$ whose width is optimal or near-optimal—typically $\operatorname{tw}(G)$—and in which each bag induces a subgraph of bounded treewidth. This perspective is explicitly motivated by connections with Baker’s layering method and with dynamic-programming algorithms on sparse graph classes (Hendrey et al., 27 Nov 2025). Closely related work also studies a coarse variant in which each bag is required to lie in the union of a bounded number of bounded-diameter balls; that variant turns out to be equivalent to quasi-isometry to a graph of bounded treewidth (Nguyen et al., 16 Jan 2025, Hickingbotham, 18 Jan 2025).

1. Definition and terminological scope

A tree-decomposition of a graph $G$ is a tree $T$ whose nodes carry bags of vertices, subject to the standard covering and connectivity conditions: every vertex appears in some bag, every edge is contained in some bag, and the bags containing any fixed vertex form a connected subtree. The treewidth $\operatorname{tw}(G)$ is the minimum, over all tree-decompositions, of the maximum bag size minus $1$.

The bounded Baker-treewidth viewpoint adds a second structural requirement. One may define $\operatorname{ttw}(G)$ as the minimum $k$ such that $G$ has a tree-decomposition in which every bag induces a subgraph of treewidth at most $k$; in this language, bounded Baker-treewidth concerns classes admitting decompositions of width $\operatorname{tw}(G)$0, or close to it, with $\operatorname{tw}(G)$1 bounded independently of the graph size (Hendrey et al., 27 Nov 2025). This is stronger than ordinary bounded treewidth, because the width of the decomposition and the induced complexity of the bags are treated as separate parameters.

A related but distinct usage appears in coarse graph theory and in recent structural work around Baker’s technique. There, one considers decompositions in which each bag is the union of at most $\operatorname{tw}(G)$2 sets of bounded diameter, or equivalently a bounded number of balls of bounded radius. In some recent literature, a “bounded Baker-treewidth” graph is one in which, after removing a bounded number of balls of bounded diameter, the remaining graph has bounded treewidth; the centered-bag and bounded-ball formulations generalize and refine that intuition (Hickingbotham, 18 Jan 2025). The terminology therefore covers two nearby themes: bounded induced treewidth in bags, and bounded coarse complexity of bags.

2. Optimal-width decompositions with bounded-treewidth bags

The most explicit optimal-width results are presently known for proper minor-closed classes. Every planar graph $\operatorname{tw}(G)$3 has a tree-decomposition of width $\operatorname{tw}(G)$4 such that each bag induces a subgraph of treewidth at most $\operatorname{tw}(G)$5, and this bound is best possible; $\operatorname{tw}(G)$6 witnesses optimality because any tree-decomposition must contain the clique in some bag (Hendrey et al., 27 Nov 2025). More generally, every graph of Euler genus $\operatorname{tw}(G)$7 has a tree-decomposition of width $\operatorname{tw}(G)$8 in which each bag has treewidth at most $\operatorname{tw}(G)$9, and this bound is best possible up to constant factor. For each fixed $G$0, every $G$1-minor-free graph has a tree-decomposition of width $G$2 such that every bag induces a subgraph of treewidth at most a polynomial function $G$3 (Hendrey et al., 27 Nov 2025).

These results separate two notions that are usually conflated in ordinary treewidth theory. A decomposition may have optimal width without its bags being easy objects in their own right; bounded Baker-treewidth asserts that both can be achieved simultaneously on large minor-closed classes. The proofs for the general $G$4-minor-free case use the notion of irreducible sets, extending unbreakability-style arguments from planar and surface settings to handle apex phenomena and other complications of the Robertson–Seymour structure theorem (Hendrey et al., 27 Nov 2025).

Graph class	Decomposition width	Bag guarantee
Planar graphs	$G$5	bag treewidth $G$6
Euler genus $G$7	$G$8	bag treewidth $G$9
$T$0-minor-free graphs	$T$1	bag treewidth $T$2, for polynomial $T$3

The significance of these theorems is structural as much as algorithmic. They show that for minor-closed families, optimal global decompositions can be refined so that the local geometry of each bag remains controlled by a bounded-treewidth template.

3. Failure at optimal width outside the minor-closed setting

The same phenomenon does not persist in simple non-minor-closed monotone classes. For every $T$4 and $T$5, there is a $T$6-planar graph $T$7 such that every tree-decomposition of width at most $T$8 has a bag of treewidth greater than $T$9 (Hendrey et al., 27 Nov 2025). In particular, bounded Baker-treewidth fails for $\operatorname{tw}(G)$0-planar graphs at optimal width and even within an additive constant of optimal width.

This negative result is important because $\operatorname{tw}(G)$1-planar graphs are among the simplest sparse classes that are not minor-closed. It shows that the existence of sparse drawings, bounded expansion-like behavior, or low crossing complexity does not by itself force optimal-width decompositions with bounded-treewidth bags. The excluded-minor hypothesis in the positive theorems is therefore not merely technical.

At the same time, the obstruction is specifically about optimal or near-optimal width. The same work shows that every $\operatorname{tw}(G)$2-planar $\operatorname{tw}(G)$3-vertex graph has a tree-decomposition of width $\operatorname{tw}(G)$4 with bounded-treewidth bags, and that this $\operatorname{tw}(G)$5 bound is asymptotically tight (Hendrey et al., 27 Nov 2025). A plausible implication is that bounded Baker-treewidth in the strongest sense is best viewed as an optimal-width phenomenon for minor-closed classes, while more general sparse classes often admit only bicriteria versions.

4. Layered and bicriteria forms

The bicriteria viewpoint is formalized through bounded layered treewidth. If a class has layered treewidth at most $\operatorname{tw}(G)$6, then any $\operatorname{tw}(G)$7-vertex graph in the class admits a tree-decomposition of width at most $\operatorname{tw}(G)$8 such that the union of any $\operatorname{tw}(G)$9 bags induces a subgraph of treewidth at most $1$0 (Hendrey et al., 27 Nov 2025). This statement extends the bagwise condition to bounded unions of bags and provides a controlled tradeoff between overall width and local induced complexity.

This is still recognizably “Baker-style.” Instead of requiring a decomposition of width $1$1, it permits a larger width while keeping the induced subgraphs arising from boundedly many bags structurally tame. Many graph classes, including planar and bounded-genus classes, have bounded layered treewidth, so the statement supplies a general framework for asymptotically tight width bounds together with bounded local treewidth behavior (Hendrey et al., 27 Nov 2025).

The distinction between optimal-width and bicriteria bounded Baker-treewidth is substantial. In proper minor-closed classes one can keep the width exactly at $1$2 and bound bag treewidth. In broader settings such as $1$3-planar graphs, one may still retain bounded local bag structure, but only after accepting a decomposition whose width scales as $1$4 rather than $1$5.

5. Coarse tree-width and quasi-isometric characterizations

A second line of work recasts the subject in coarse graph-theoretic terms. A tree-decomposition is called $1$6-centred if each bag is the union of at most $1$7 sets of weak diameter at most $1$8, equivalently a bounded number of balls of bounded diameter. A graph is quasi-isometric to a graph with bounded treewidth if and only if it has such a centred tree-decomposition (Hickingbotham, 18 Jan 2025).

More precisely, there is a function $1$9 such that every graph with a $\operatorname{ttw}(G)$0-centred tree-decomposition is $\operatorname{ttw}(G)$1-quasi-isometric to a graph with treewidth at most $\operatorname{ttw}(G)$2, and every graph that is $\operatorname{ttw}(G)$3-quasi-isometric to a graph with treewidth at most $\operatorname{ttw}(G)$4 has a $\operatorname{ttw}(G)$5-centred tree-decomposition (Hickingbotham, 18 Jan 2025). Closely related results state that if a connected graph admits a tree-decomposition in which every bag is the union of at most $\operatorname{ttw}(G)$6 sets each of diameter at most $\operatorname{ttw}(G)$7, then it admits an $\operatorname{ttw}(G)$8-quasi-isometry to a graph with tree-width at most $\operatorname{ttw}(G)$9; conversely, an $k$0-quasi-isometry to a graph of treewidth at most $k$1 yields a decomposition whose bags are unions of at most $k$2 sets each of diameter at most $k$3 (Nguyen et al., 16 Jan 2025).

This coarse characterization generalizes earlier work of Berger and Seymour on bags of bounded radius and quasi-isometry to trees. It also clarifies the relation to Baker’s technique. Rather than removing layers or balls to simplify a graph, one can demand that the entire decomposition be assembled from boundedly many bounded-diameter pieces per bag. In that sense, coarse bounded Baker-treewidth is an intrinsic, quasi-isometry-invariant analogue of the classical layering intuition.

6. Baker-type analogues, algorithmic significance, and caveats

The Baker perspective is not confined to planar or minor-closed classes. For even-hole-free graphs of bounded degree, and more generally for $k$4-free odd-signable graphs of bounded degree, bounded treewidth is obtained by partitioning relevant cutsets into a bounded number of laminar collections and recursively processing central bags. The direct Baker-style layering used in planar graphs does not extend globally to these hereditary classes, but the resulting cutset hierarchy is explicitly described as conceptually parallel: instead of spatial layers, one works with laminarized star cutsets, and the class enjoys “a form of Baker-treewidth” in which the treewidth bound is a function of the maximum degree (Abrishami et al., 2020).

The algorithmic motivation for bounded Baker-treewidth is the general power of bounded-treewidth decompositions. In the CONGEST model, for any MSO formula $k$5 and any constant $k$6, there is an algorithm deciding whether a graph of treewidth at most $k$7 satisfies $k$8 in $k$9 rounds, and there is also a distributed procedure that either concludes $G$0 or computes a tree decomposition of width $G$1 and depth $G$2 in $G$3 rounds (Jauregui et al., 2018). In nonmonotonic reasoning, Reiter’s propositional default logic admits a dynamic-programming algorithm on tree decompositions of the semi-primal graph that decides extension existence in linear time in the input and triple exponential time in the treewidth, and enumerates generating defaults after the same preprocessing with linear delay (Fichte et al., 2017). These results explain why decompositions with bounded-treewidth bags are structurally attractive: they promise not only a small global decomposition width, but also bags that remain compatible with further dynamic programming.

A common misconception is that stronger decomposition structure automatically implies fixed-parameter tractability for every natural problem under treewidth parameterization. This is false. Defensive Alliance is W[1]-hard when parameterized by treewidth, even though it is FPT by solution size and even though bounded-treewidth graphs support many classical dynamic-programming techniques (Bliem et al., 2017). This suggests that bounded Baker-treewidth should be viewed as a structural resource whose algorithmic value depends on the problem’s interaction with treewidth, bag interfaces, and any auxiliary constraints, rather than as a universal tractability guarantee.