Family Tree Decomposition

• Family tree decomposition is a graph-theoretic method that organizes k-blocks using nested vertex separations to yield canonical, Aut(G)-invariant tree-decompositions.
• It employs tight and nested separation systems to resolve crossing separations, ensuring efficient distinction and robust identification of k-blocks.
• The technique merges fixed‑k decompositions into a unified global tree-decomposition, providing a coherent framework for analyzing connectivity in finite graphs.

Family tree decomposition is a canonical graph-theoretic organization of the families of $k$-blocks of a finite graph by means of nested vertex separations and the tree-decompositions they induce. For each fixed $k$, it produces an $\operatorname{Aut}(G)$-invariant tree-decomposition of adhesion at most $k$ in which distinct $k$-blocks lie in distinct parts and are separated efficiently; under additional robustness assumptions that are both mild and necessary, these fixed-$k$ decompositions refine one another and can be merged into a single overall decomposition that distinguishes all robust blocks simultaneously across all $k$ (Carmesin et al., 2011).

1. Separation systems as the organizing language

The formal basis of family tree decomposition is a system of vertex separations. For a finite graph $G=(V,E)$, a separation is an ordered pair $(A,B)$ with $A,B\subseteq V$ and $k$0. Its separator is $k$1, and its order is

$k$2

A separation is proper if both $k$3 and $k$4 are nonempty. Separations are partially ordered by

$k$5

Flipping a separation reverses this order: $k$6 iff $k$7 (Carmesin et al., 2011).

The central structural dichotomy is between nested and crossing separations. Two separations $k$8 and $k$9 are nested, written $\operatorname{Aut}(G)$0, if one is comparable with the other or with its inverse under $\operatorname{Aut}(G)$1; otherwise they cross, written $\operatorname{Aut}(G)$2. Crossing is analyzed by the four corners $\operatorname{Aut}(G)$3, $\operatorname{Aut}(G)$4, $\operatorname{Aut}(G)$5, and $\operatorname{Aut}(G)$6. The corresponding corner separations, such as $\operatorname{Aut}(G)$7, remain nested with the original separations and are the key tools for resolving conflicts among crossing candidates.

A separation system $\operatorname{Aut}(G)$8 is useful for decomposition only when it can resolve such conflicts without losing distinguishing power. This is formalized by the condition that $\operatorname{Aut}(G)$9 “separates well” a family $k$0 of $k$1-inseparable vertex sets: whenever $k$2 and $k$3 in $k$4 cross and $k$5 contains $k$6 and $k$7, the system must contain a corner separation $k$8 that still places $k$9 on its $k$0-side and satisfies $k$1. This condition is what allows a nested subsystem to be extracted canonically.

Object	Definition	Function
Separation $k$2	$k$3	Encodes a vertex cut
Nested system	Pairwise nested separations	Supports a structure tree
Corner separation	$k$4, etc.	Resolves crossings
Tight separation	Every separator vertex meets both sides	Simplifies comparability

A tight separation is one in which every vertex of $k$5 has a neighbor in $k$6 and another in $k$7. For tight separations, the single inclusion $k$8 already suffices to infer $k$9, which simplifies the inductive construction of nested systems (Carmesin et al., 2011).

2. $k$0-inseparability, $k$1-blocks, and robustness

The objects localized by family tree decomposition are the $k$2-blocks of a graph. A set $k$3 is $k$4-inseparable if $k$5 and no set $k$6 of at most $k$7 vertices separates two vertices in $k$8. A $k$9-block is a maximal $k$0-inseparable set; equivalently,

$k$1

$k$2

The rank $k$3 of a block $k$4 is the smallest $k$5 for which $k$6 is a $k$7-block (Carmesin et al., 2011).

For a fixed $k$8, the decomposition theory separates all $k$9-blocks. To combine decompositions across all $G=(V,E)$0, however, an additional hypothesis is necessary. A $G=(V,E)$1-inseparable set $G=(V,E)$2 is $G=(V,E)$3-robust if whenever $G=(V,E)$4 is a $G=(V,E)$5-separation with $G=(V,E)$6 and $G=(V,E)$7 is a separation of order at most $G=(V,E)$8 crossing $G=(V,E)$9 with

$(A,B)$0

then $(A,B)$1 or $(A,B)$2. Intuitively, robustness excludes the unique crossing obstruction that prevents a unified nested system.

Several classes of blocks are automatically robust. Large $(A,B)$3-blocks of size at least $(A,B)$4 are robust, complete graphs are robust, and robustness is preserved upward in the sense that containing a robust block makes a larger block robust. These facts explain why the all-$(A,B)$5 theory applies broadly even though it does not encompass every possible $(A,B)$6-block (Carmesin et al., 2011).

The necessity of robustness is not merely technical. The obstruction described in the theory consists of a configuration in which exactly one horizontal separation of order $(A,B)$7 distinguishes two $(A,B)$8-blocks and exactly one vertical separation of order $(A,B)$9 distinguishes two $A,B\subseteq V$0-blocks; these separations cross, so no unified nested system can contain both. Robustness rules out precisely this phenomenon.

3. From nested separations to a structure tree

A nested symmetric separation system $A,B\subseteq V$1 canonically yields a tree, called the structure tree $A,B\subseteq V$2. The construction uses an equivalence relation $A,B\subseteq V$3 on $A,B\subseteq V$4: $A,B\subseteq V$5 if either they are equal or $A,B\subseteq V$6 is a predecessor of $A,B\subseteq V$7 in the poset $A,B\subseteq V$8, meaning $A,B\subseteq V$9 and there is no $k$00 with $k$01. Distinct equivalent separations are incomparable under $k$02. The nodes of $k$03 are the $k$04-classes $k$05, and its edges are the unordered pairs $k$06. The resulting multigraph is a tree (Carmesin et al., 2011).

The associated tree-decomposition is obtained by assigning to each node $k$07 the part

$k$08

Each $k$09 is $k$10-inseparable. The parts come in two types. If $k$11 is a maximal $k$12-inseparable set, then $k$13 is a block node. If $k$14 equals a separator $k$15 for some $k$16, then $k$17 is a hub node. Every $k$18-block appears as some $k$19, and the edges of $k$20 induce exactly the separations in $k$21.

This construction translates nested separation data into a standard tree-decomposition

$k$22

satisfying the usual axioms: vertex coverage, edge coverage, and the running-intersection property. The adhesion is

$k$23

and if the separations in $k$24 all have order at most $k$25, then the adhesion is at most $k$26. Since automorphisms of $k$27 preserving $k$28 act naturally on $k$29, canonicity at the level of separations becomes canonicity of the decomposition tree itself (Carmesin et al., 2011).

4. Canonical extraction and the fixed-$k$30 decomposition theorem

The decisive existence statement is the extraction theorem for nested subsystems. If $k$31 is a separation system, $k$32 is a family of $k$33-inseparable vertex sets, $k$34 is $k$35-relevant, and $k$36 separates $k$37 well, then there exists a nested $k$38-relevant subsystem $k$39 that weakly distinguishes all weakly $k$40-distinguishable pairs in $k$41. Moreover,

$k$42

so the construction is canonical and $k$43-invariant whenever $k$44 and $k$45 are (Carmesin et al., 2011).

The extraction operates recursively through extremal separations. A separation $k$46 is extremal in $k$47 if for every $k$48, either $k$49 or $k$50. Under the “separates well” hypothesis, every $k$51-minimal relevant separation is extremal; extremal separations are nested with all of $k$52, and their $k$53-side is a $k$54-block. One collects all extremal separations $k$55, removes from consideration those vertex sets already hit by $k$56, and recurses on the remainder. The union of all stages is $k$57.

Applied to $k$58-blocks, this yields the fixed-$k$59 theorem: for every finite graph $k$60 and every integer $k$61, there exists an $k$62-invariant tree-decomposition of adhesion at most $k$63 that efficiently distinguishes all $k$64-blocks. If $k$65 and $k$66 are distinct $k$67-blocks, then they lie in different parts, and along the unique path between those parts there is an edge $k$68 whose adhesion satisfies

$k$69

where $k$70 is the minimum order of a separation separating $k$71 from $k$72. The construction is explicitly canonical: it uses no arbitrary tie-breaking and depends only on the structure of $k$73 (Carmesin et al., 2011).

5. The family tree across all $k$74

The expression “family tree decomposition” refers most directly to the passage from the individual decompositions for fixed $k$75 to a coherent decompositional family indexed by $k$76, and then to a single overall decomposition for all robust blocks. The construction proceeds inductively. Starting from $k$77, one builds tight, nested, $k$78-invariant systems $k$79 of separations of order at most $k$80. At stage $k$81, the new separations in $k$82 are order-$k$83 separations that distinguish robust $k$84-blocks not already separated by $k$85. For each previous block $k$86 that still contains multiple relevant $k$87-blocks, one forms the subsystem $k$88 of order-$k$89 separations nested with $k$90, verifies that $k$91 separates the family inside $k$92 well, extracts the canonical nested subsystem $k$93, and sets

$k$94

Passing to structure trees yields the tree-decompositions $k$95 (Carmesin et al., 2011).

The all-$k$96 theorem states that every finite graph $k$97 admits a sequence $k$98 such that: each decomposition has adhesion at most $k$99 and distinguishes all robust $\operatorname{Aut}(G)$00-blocks; each $\operatorname{Aut}(G)$01 is a minor-refinement of $\operatorname{Aut}(G)$02,

$\operatorname{Aut}(G)$03

and each is $\operatorname{Aut}(G)$04-invariant. The refinement means that $\operatorname{Aut}(G)$05 is obtained from $\operatorname{Aut}(G)$06 by contracting subtrees, and parts at level $\operatorname{Aut}(G)$07 are unions of parts at level $\operatorname{Aut}(G)$08.

Under the robustness assumptions, these decompositions can be merged into a single overall tree-decomposition that simultaneously distinguishes all robust blocks. Equivalently, one may take $\operatorname{Aut}(G)$09 and consider the decomposition associated with this union. From that overall decomposition, the fixed-$\operatorname{Aut}(G)$10 view can be recovered by contracting the subtree supporting each robust $\operatorname{Aut}(G)$11-block into a single node. In this sense, the family tree is simultaneously a sequence of canonical refinements and a single canonical global object (Carmesin et al., 2011).

6. Classical antecedents, examples, and scope

Family tree decomposition sits within a line of results on organizing connectivity by trees. For $\operatorname{Aut}(G)$12, Tutte’s theorem yields a tree-decomposition of adhesion $\operatorname{Aut}(G)$13 whose torsos are either $\operatorname{Aut}(G)$14-connected or cycles. The present framework rephrases this through $\operatorname{Aut}(G)$15-inseparable sets and canonical nested separations. It also extends work of Dunwoody and Krön on the canonical separation of $\operatorname{Aut}(G)$16-inseparable sets, while eliminating dependence on arbitrary tie-breaking such as vertex enumerations (Carmesin et al., 2011).

The simplest special case is $\operatorname{Aut}(G)$17. If $\operatorname{Aut}(G)$18 is the set of all proper $\operatorname{Aut}(G)$19-separations $\operatorname{Aut}(G)$20 with $\operatorname{Aut}(G)$21 connected, then the structure tree $\operatorname{Aut}(G)$22 is essentially the block-cut tree: block nodes correspond to maximal $\operatorname{Aut}(G)$23-connected subgraphs or bridges, hub nodes correspond to cut vertices lying in at least three blocks, and adhesion is $\operatorname{Aut}(G)$24. For $\operatorname{Aut}(G)$25, the decomposition parallels SPQR-type separation by minimal $\operatorname{Aut}(G)$26-separators, but is formulated directly through nested separation systems and is explicitly canonical under $\operatorname{Aut}(G)$27 (Carmesin et al., 2011).

A standard illustrative graph consists of two copies of $\operatorname{Aut}(G)$28 joined by a long path $\operatorname{Aut}(G)$29. For $\operatorname{Aut}(G)$30, the only $\operatorname{Aut}(G)$31-blocks are the two cliques. One canonical decomposition is the three-part path decomposition with parts $\operatorname{Aut}(G)$32, $\operatorname{Aut}(G)$33, and $\operatorname{Aut}(G)$34, of adhesion at most $\operatorname{Aut}(G)$35; another canonical choice is a finer decomposition with all $\operatorname{Aut}(G)$36 bags along the path. Both are $\operatorname{Aut}(G)$37-invariant, and the family tree point of view further refines the picture so as to separate lower-$\operatorname{Aut}(G)$38 blocks as well.

The scope is explicitly finite graphs. No extension to infinite graphs or directed graphs is claimed. The adhesion bound is exactly the relevant $\operatorname{Aut}(G)$39, but stronger guarantees about torsos or finer part structure are outside the stated scope. Likewise, the constructions are algorithmic in outline, yet algorithmic complexity is not optimized and computational questions are deferred (Carmesin et al., 2011).

A later development broadens the canonical framework from $\operatorname{Aut}(G)$40-blocks alone to $\operatorname{Aut}(G)$41-blocks together with tangles of order $\operatorname{Aut}(G)$42, using profiles as the unifying language and again producing canonical tree-decompositions invariant under graph automorphisms (Carmesin et al., 2013). That later profile-based theory clarifies the wider landscape, but the family tree decomposition remains the specific mechanism by which all robust block families of a finite graph are organized into one coherent hierarchy of canonical nested separations and tree-decompositions.