Project-Connex Tree Decompositions

• Project-Connex Tree Decompositions are connectivity-aware frameworks that derive global structure from local separator, projection, or intersection constraints.
• They encompass methods from orthogonal tree-decompositions in graph theory to tree projections in hypergraphs and project-connex decompositions in database query optimization.
• Enhanced techniques like spaghetti tree-decompositions illustrate how added directionality can overcome traditional chromatic and width obstructions in structural decompositions.

Project-Connex Tree Decompositions, as the term is used across the literature considered here, denotes a family of connectivity-aware decomposition frameworks rather than a single invariant. In graph theory it includes orthogonal tree-decompositions, where several tree-like decompositions interact through bounded bag intersections; in hypergraph and database theory it includes tree projections and project-connex decompositions with nested witness subtrees for projected variables; and in adjacent structural programs it touches canonical decompositions, clique-tree constructions, and edge-decompositions into copies of trees (Felsner et al., 2017, Greco et al., 2012, Figueira et al., 18 Jul 2025). A common theme is the attempt to derive global structural control from local separator, projection, or intersection constraints. The literature also identifies sharp obstructions: bounded bag intersections do not, by themselves, force $\chi$-boundedness in the tree–tree setting, and stronger “project-connex” connectivity constraints need not preserve optimal width (Felsner et al., 2017, Greco et al., 2012).

1. Core notions and formal scope

In the graph-theoretic setting, a tree-decomposition of a graph $G$ is a pair $(T,\beta)$ where $T$ is a tree and each node $x\in V(T)$ carries a bag $\beta(x)\subseteq V(G)$ such that every edge of $G$ is contained in some bag and, for every vertex $v\in V(G)$, the set of tree nodes whose bags contain $v$ induces a non-empty connected subtree of $T$. Its width is $G$0. A path-decomposition is the special case in which $G$1 is a path, with minimum width $G$2; the corresponding tree-width is $G$3 (Felsner et al., 2017).

For several decompositions at once, the relevant invariant is the intersection profile of bags. Given $G$4 tree-decompositions $G$5 of $G$6, their $G$7-width is

$G$8

For $G$9, two decompositions are $(T,\beta)$0-orthogonal when

$(T,\beta)$1

The minimum such value over all $(T,\beta)$2-tuples of tree-decompositions is the $(T,\beta)$3-tree-width, also called $(T,\beta)$4-medianwidth, and the path analogue is the $(T,\beta)$5-path-width, also called $(T,\beta)$6-latticewidth (Felsner et al., 2017).

In the hypergraph setting, tree projections generalize structural decomposition methods. For hypergraphs $(T,\beta)$7, one writes $(T,\beta)$8 when every edge of $(T,\beta)$9 is contained in some edge of $T$0. A hypergraph $T$1 is a tree projection of $T$2 with respect to $T$3 if $T$4 is $T$5-acyclic and

$T$6

Minimal tree projections are those without useless redundancies under the paper’s containment order, and they exhibit normal-form connectivity properties closely analogous to minimal tree-decompositions (Greco et al., 2012).

In the database setting, project-connex decompositions are defined for aggregate queries of the form

$T$7

A tree decomposition $T$8 of the query hypergraph is project-connex if, for every $T$9, there is a witness subtree $x\in V(T)$0 such that

$x\in V(T)$1

The associated width parameter is the project-connex generalized hyperwidth $x\in V(T)$2 (Figueira et al., 18 Jul 2025).

These notions are related but not interchangeable. Orthogonality concerns intersections of bags across multiple decompositions; tree projections concern an acyclic intermediary hypergraph between $x\in V(T)$3 and $x\in V(T)$4; project-connex decompositions for queries impose nested witness-subtree conditions on projected variable sets. This suggests a shared “projected connectivity” viewpoint, but the precise constraints and the phenomena they control differ substantially.

2. Orthogonal tree-decompositions and the chromatic-number obstruction

A classical starting point is the Asplund–Grünbaum theorem, which states that intersection graphs of axis-aligned rectangles in the plane are $x\in V(T)$5-bounded, in fact with $x\in V(T)$6. In decomposition language, this is equivalent to the existence of a function $x\in V(T)$7 such that whenever a graph $x\in V(T)$8 admits two path-decompositions $x\in V(T)$9 and $\beta(x)\subseteq V(G)$0 with

$\beta(x)\subseteq V(G)$1

then $\beta(x)\subseteq V(G)$2. The equivalence uses the fact that the $\beta(x)\subseteq V(G)$3-path-width of $\beta(x)\subseteq V(G)$4 is the minimum $\beta(x)\subseteq V(G)$5 such that $\beta(x)\subseteq V(G)$6 is a subgraph of an intersection graph of axis-aligned boxes in $\beta(x)\subseteq V(G)$7 with clique number at most $\beta(x)\subseteq V(G)$8 (Felsner et al., 2017).

Dujmović, Joret, Morin, Norin, and Wood asked whether the same phenomenon survives if the two path-decompositions are replaced by two tree-decompositions. Felsner, Joret, Micek, Trotter, and Wiechert gave a negative answer. They proved that there are graphs with arbitrarily large chromatic number admitting two tree-decompositions whose pairwise bag intersections all have size at most $\beta(x)\subseteq V(G)$9, and that this remains true even if one of the two decompositions is required to be a path-decomposition (Felsner et al., 2017).

The obstruction is furnished by Burling graphs. These graphs $G$0 are defined inductively together with a distinguished family of stable sets $G$1. One starts from $G$2, a single vertex, and at stage $G$3 takes a master copy $G$4, a copy $G$5 for each stable set $G$6, and for each $G$7 adds a vertex $G$8 adjacent to all vertices of $G$9 and to no other vertices. Burling’s theorem states that every $v\in V(G)$0 is triangle-free and satisfies $v\in V(G)$1 (Felsner et al., 2017).

The constructive theorem relevant to orthogonality is precise: for every $v\in V(G)$2, the Burling graph $v\in V(G)$3 has a tree-decomposition $v\in V(G)$4 and a path-decomposition $v\in V(G)$5 such that

$v\in V(G)$6

The induction maintains three properties: the orthogonality bound itself; the existence, for every distinguished stable set $v\in V(G)$7, of a tree bag equal to $v\in V(G)$8; and the sparsity condition $v\in V(G)$9 for every such stable set and every path bag (Felsner et al., 2017).

The significance of this result is mainly negative. It shows that bounded pairwise bag intersections alone do not control chromatic number when at least one decomposition axis is a tree. The path–path case is special because it can be interpreted through rectangle intersection graphs, and that geometric representation carries $v$0-boundedness. Once paths are replaced by general trees, that geometric leverage disappears. The paper further notes boundary phenomena: some classes, including planar graphs and more generally $v$1-minor-free graphs, have bounded $v$2-tree-width; all bipartite graphs even have $v$3-tree-width and $v$4-path-width at most $v$5; yet bounded $v$6-tree-width does not imply $v$7-boundedness in general (Felsner et al., 2017).

The main proposed repair is additional directionality. A spaghetti tree-decomposition is a rooted tree-decomposition in which, after orienting edges away from the root, the bags containing any fixed vertex form a directed path. The authors conjecture that if a graph admits a spaghetti tree-decomposition and a path-decomposition with

$v$8

then $v$9 for some function $T$0. They further note that the conjecture might even hold for two spaghetti tree-decompositions. A plausible implication is that orthogonality becomes informative only when it is coupled with path-like monotonicity of vertex footprints (Felsner et al., 2017).

3. Tree projections, minimality, and the stronger project-connex condition

Tree projections were introduced as a common framework for structural decomposition methods on hypergraphs. Within this framework, minimality is the decisive normal-form condition. If a tree projection exists, then a minimal tree projection exists; every minimal tree projection is reduced; and for a minimal tree projection $T$1 of $T$2 with respect to $T$3, one has $T$4 (Greco et al., 2012).

The central structural theorem is that minimal tree projections preserve component structure. For every hyperedge $T$5 and every component notion defined modulo $T$6, a set $T$7 is an $T$8-component of $T$9 if and only if it is an $G$00-component of $G$01. This eliminates artificial connectivity created solely by enlarging hyperedges in the acyclic intermediary (Greco et al., 2012).

This preservation yields a strong normal form for join trees. If $G$02 is a minimal tree projection of $G$03, then any join tree of $G$04 is $G$05-connected, and for any root hyperedge $G$06 there exists a join tree rooted at $G$07 that is simultaneously $G$08-connected and an $G$09-component tree. Consequently, $G$10 has a tree projection if and only if it has a connected tree projection in this sense (Greco et al., 2012).

The same paper isolates a stricter notion of connectedness, attributed to Subbarayan and Andersen, that is often associated with “projected connectedness” or “project-connex.” For a rooted (generalized) hypertree decomposition $G$11, the condition is:

• the root $G$12 satisfies $G$13;
• for every parent–child edge $G$14 of $G$15 and every $G$16,

$G$17

This requirement is stronger than ordinary connectedness: every resource edge used at a child must project non-trivially to the parent–child separator (Greco et al., 2012).

A key negative result is that this stronger property is not a harmless normal form. The paper constructs a graph $G$18 for which

$G$19

Thus a connected generalized hypertree decomposition in the stronger SH07 sense may require strictly larger width than an unrestricted hypertree decomposition, even on graphs (Greco et al., 2012).

The same work supplies a game-theoretic characterization. In the Captain and Robber game on $G$20, the Captain chooses subsets of resource edges of $G$21 and the Robber moves within components of $G$22 not blocked by the current position. The paper proves that a winning strategy exists if and only if a tree projection exists, and that every winning strategy can be converted into a monotone winning strategy. This yields a constructive interpretation of tree projections and explains why the $G$23-connected normal form is robust, whereas the stricter project-connex condition is not (Greco et al., 2012).

The resulting distinction is fundamental. In this literature, “connected” in the sense of minimal tree projections is a theorem; “project-connex” in the stronger SH07 sense is an additional restriction that may destroy optimality.

4. Project-connex width for aggregate and group-by conjunctive queries

In database theory, project-connex decompositions arise for aggregate conjunctive queries over commutative semirings. The query language considered is generated from joins and group-by projections:

• $G$24 multiplies annotations of atoms in a conjunction $G$25 using the semiring product $G$26;
• $G$27 groups answers of $G$28 by their restriction to $G$29 and aggregates within each group using the semiring sum $G$30 (Figueira et al., 18 Jul 2025).

For a nested aggregate query

$G$31

a tree decomposition of the query hypergraph is project-connex if it contains witness subtrees $G$32 with

$G$33

For a single projection, this is exactly the usual free-connex condition (Figueira et al., 18 Jul 2025).

The main structural theorem identifies project-connex width with ordinary generalized hypertree width on an augmented Boolean query. The augmented query $G$34 is obtained by adding a fresh binary atom $G$35 for every pair $G$36 joined by an $G$37-frontier-path for some $G$38. Then

$G$39

and decompositions can be transformed in linear time. This reduces project-connex recognition and construction to classical decomposition algorithms (Figueira et al., 18 Jul 2025).

The algorithmic consequences are direct. If $G$40 has a project-connex decomposition of width $G$41, then evaluation admits constant-delay enumeration after polynomial preprocessing of order $G$42. When the outermost projected set is empty, the scalar value is computable in time

$G$43

The width-$G$44 case is especially explicit: $G$45 has a width-$G$46 project-connex decomposition if and only if each query

$G$47

has a join tree, where $G$48; such a decomposition and its witness subtrees can be built in $G$49 time (Figueira et al., 18 Jul 2025).

The framework unifies several earlier tractability criteria. For counting conjunctive queries, evaluating $G$50 is expressed as

$G$51

Hence a free-connex decomposition of width $G$52 yields evaluation time $G$53. The paper also states that, assuming $G$54, a recursively enumerable bounded-arity class $G$55 of conjunctive queries has polynomial-time counting exactly when $G$56 is bounded, equivalently when both $G$57 and $G$58 are bounded (Figueira et al., 18 Jul 2025).

In this setting, unlike the hypertree setting above, project-connex is not a problematic strengthening but the target structural notion itself. Its usefulness comes from the theorem $G$59, which makes the constraint algorithmically reducible to standard width computation.

5. Canonical, parameter-controlled, and infinite clique-tree decompositions

A neighboring branch of the literature studies decompositions that are canonical, bounded in auxiliary parameters, or extended to infinite graphs. Carmesin, Diestel, Hamann, and Hundertmark proved that for every finite graph $G$60 and integer $G$61, there exists a canonical tree-decomposition of adhesion $G$62 that distinguishes all $G$63-blocks and all tangles of order $G$64. The construction is based on nested systems of proper separations and strategy families such as $G$65, $G$66, and $G$67; the resulting decompositions commute with graph isomorphisms, are invariant under automorphisms, and have parts that are either $G$68-blocks or hubs (Carmesin et al., 2013).

A different optimization problem concerns simultaneous control of width, spread, order, and decomposition-tree degree. Wood proved that every graph $G$69 with $G$70 has a tree-decomposition with width at most $G$71 in which each vertex $G$72 appears in at most $G$73 bags, and also a tree-decomposition with width at most $G$74 and degree at most $G$75 satisfying the same spread bound. The paper’s joint optimization theorem states that for any graph $G$76 and integer $G$77, there is a tree-decomposition of width at most $G$78, degree at most $G$79, order at most

$G$80

with

$G$81

for every vertex $G$82. The proof uses slick decompositions, balanced separators, weak tree-decompositions, and a conversion lemma that turns weak decompositions of width $G$83 into standard decompositions of width $G$84 while preserving the spread control (Wood, 1 Sep 2025).

In infinite graph theory, a related clique-tree program has been developed for chordal graphs. Every chordal graph without a strict comb of cliques admits a tree-decomposition into maximal cliques. More sharply, a connected graph admits a tree-decomposition into finite cliques if and only if it is chordal, admits a normal spanning tree, and does not contain the graph

$G$85

with both $G$86 and $G$87 adjacent to every vertex of $G$88, as an induced minor. The proofs proceed through greedy constructions of length $G$89 and, for the finite-clique theorem, an Extension Lemma based on Halin’s finiteness theorem for minimal separators (Pitz et al., 25 Mar 2026).

These results do not define “project-connex” in the technical sense of the query literature, but they extend the same structural agenda: separators should be canonical, bag occurrences should be sparse and controllable, and clique-based decompositions should be characterized even in infinite settings. This suggests that connex-style decomposition theory is as much about auxiliary regularity conditions as it is about width alone.

6. Edge-decompositions into trees: a distinct but adjacent decomposition program

A separate body of work uses “tree decomposition” in the sense of decomposing the edge set of a graph into copies of a fixed tree. For a tree $G$90 on $G$91 edges, the Barát–Thomassen conjecture asserts that there exists $G$92 such that every $G$93-edge-connected simple graph $G$94 with $G$95 has a $G$96-decomposition. Bensmail, Harutyunyan, Le, Merker, and Thomassé had established the conjecture with very large bounds; the paper “Decomposing highly edge-connected graphs into homomorphic copies of a fixed tree” proved a homomorphic-copy version for every tree, showing that there exists $G$97 such that every $G$98-edge-connected graph with $G$99 has a $(T,\beta)$00-decomposition into homomorphic copies of $(T,\beta)$01. Moreover, if $(T,\beta)$02, then every such homomorphic copy is isomorphic to $(T,\beta)$03, yielding a genuine $(T,\beta)$04-decomposition. The same paper verifies the Barát–Thomassen conjecture for all trees of diameter at most $(T,\beta)$05 (Merker, 2016).

The proof architecture is highly structural. It reduces to bipartite graphs, constructs $(T,\beta)$06-equitable edge-colorings, and combines Mader’s lifting theorem, the Nash–Williams–Tutte theorem on edge-disjoint spanning trees, Ellingham’s small-degree spanning trees, and an orientation modulo $(T,\beta)$07 theorem. A central technical theorem states that if a bipartite graph is sufficiently edge-connected and one side has degrees divisible by $(T,\beta)$08, then it can be decomposed into spanning subgraphs with prescribed proportional degrees on one side and prescribed divisibility on the other (Merker, 2016).

A later improvement by Hasanvand lowered the connectivity dependence to linear in $(T,\beta)$09. The paper proves that every simple graph $(T,\beta)$10 with

$(T,\beta)$11

admits an edge-decomposition into isomorphic copies of any fixed tree $(T,\beta)$12 of size $(T,\beta)$13. In the high-girth regime, an explicit theorem gives decomposition under

$(T,\beta)$14

and the abstract together with Remark 5.4 states that the minimum-degree condition can be dropped while maintaining the same linear edge-connectivity threshold (Hasanvand, 2022).

An algebraic culmination of the exact-decomposition direction is the theorem that every tree on $(T,\beta)$15 edges decomposes $(T,\beta)$16 and $(T,\beta)$17 for every positive integer $(T,\beta)$18. The method introduces oriented $(T,\beta)$19-labelings, proves that every tree admits such a labeling using the polynomial method, and derives as immediate consequences both the graceful tree conjecture and decompositions of complete and complete bipartite hosts by cyclic constructions (Chalise et al., 2024).

This edge-partition program is conceptually adjacent to tree-decomposition theory rather than identical to it. Its central invariants are edge-connectivity, girth, and labeling structure, not bags and adhesions. Nevertheless, within the Project-Connex summaries it plays a parallel role: high connectivity plus carefully organized local structure can force global decomposition into tree-shaped pieces.

7. Conceptual synthesis, misconceptions, and open directions

Several misconceptions are explicitly ruled out by the literature. First, small pairwise bag intersections across two tree-like decompositions do not imply chromatic control: Burling graphs show that even triangle-free graphs with arbitrarily large chromatic number can admit a tree-decomposition and a path-decomposition with all intersections of size at most $(T,\beta)$20 (Felsner et al., 2017). Second, stronger “project-connex” connectivity constraints are not automatically a normal form: in hypertree theory, the SH07 condition can increase width, as witnessed by $(T,\beta)$21 (Greco et al., 2012). Third, the query-theoretic notion of project-connex should not be conflated with either of the previous two; there it is a positive structural criterion equivalent to ordinary generalized hypertree width on an augmented query (Figueira et al., 18 Jul 2025).

A recurring corrective pattern is the addition of stronger but more targeted structure. In orthogonal decomposition theory, the conjectured repair is the spaghetti condition, which enforces directed-path behavior for every vertex footprint (Felsner et al., 2017). In canonical decomposition theory, the corrective ingredient is nested separation systems that distinguish profiles while commuting with automorphisms (Carmesin et al., 2013). In low-spread tree-decomposition theory, it is slickness, which ties the number of bags containing a vertex to $(T,\beta)$22 (Wood, 1 Sep 2025). In infinite chordal graphs, the needed extra conditions are the exclusion of strict combs of cliques, or, for finite-clique bags, chordality plus a normal spanning tree plus exclusion of $(T,\beta)$23 as an induced minor (Pitz et al., 25 Mar 2026). In the edge-decomposition program, the extra hypotheses are high edge-connectivity, degree divisibility, equitable colorings, or girth conditions that prevent homomorphic collisions (Merker, 2016, Hasanvand, 2022).

This suggests a broad unifying interpretation. “Connex” decomposition methods are most effective not when they merely bound width or intersections, but when they also regulate how local pieces attach, propagate, or project across the decomposition. That interpretation is consistent with the database results, where nested witness subtrees yield tractability, and with the graph-theoretic obstructions, where the failure of monotonicity or separator discipline is precisely what defeats naive bounded-intersection heuristics (Figueira et al., 18 Jul 2025, Felsner et al., 2017).

Open directions remain explicit. The spaghetti tree-decomposition conjecture asks whether path-like directionality restores $(T,\beta)$24-boundedness in the orthogonal setting (Felsner et al., 2017). The parameter-controlled decomposition program leaves significant room to improve constants such as $(T,\beta)$25, $(T,\beta)$26, and $(T,\beta)$27 (Wood, 1 Sep 2025). The Barát–Thomassen program still seeks sharper bounds and a full elimination of supplementary girth or degree assumptions in general (Merker, 2016, Hasanvand, 2022). In query evaluation, project-connex width has already been linked to free-connex width, counting, and semiring aggregates, but the supplied account also points toward extensions to monotone width measures and more output-sensitive algorithmic bounds (Figueira et al., 18 Jul 2025).

Taken together, Project-Connex Tree Decompositions form a technically diverse but structurally coherent research area. Its central lesson is not that one decomposition parameter dominates all others, but that carefully chosen connectivity constraints can be decisive—provided one knows which constraints are genuinely structural and which are too weak, or too strong, for the problem at hand.