Total Twin-Width in Graph Contractions

• Total twin-width is defined as the minimum, over all contraction sequences of a graph, of the maximum number of red edges in any intermediate trigraph.
• It establishes a tight quadratic relationship with linear clique-width, replacing previous exponential bounds with precise, algorithmically useful estimates.
• This parameter underpins efficient algorithms for graph colorings and FO model checking by leveraging global red-edge budgets in contraction sequences.

Searching arXiv for the cited papers to ground the article in current preprints. Total twin-width is a contraction-sequence parameter of graphs defined on intermediate trigraphs obtained while repeatedly contracting vertices. For a graph $G$, $\mathbf{ttww}(G)$ is the minimum, over all contraction sequences of $G$, of the maximum number of red-edges appearing in any intermediate trigraph. In "Improved Bounds for Twin-Width Parameter Variants with Algorithmic Applications to Counting Graph Colorings" (Baril et al., 5 Sep 2025), total twin-width is studied together with total vertex twin-width and linear clique-width, yielding a tight quadratic relationship that replaces previously known exponential and double-exponential bounds. Within the broader twin-width framework, red edges record non-homogeneity created by contractions, and contraction sequences already underpin fixed-parameter tractability results such as FO model checking for bounded twin-width when a witness sequence is given (Bonnet et al., 2020).

1. Definition and contraction-sequence framework

The ambient framework is that of trigraphs and contractions. In the standard twin-width formalism, a contraction sequence is a sequence $G_n, G_{n-1}, \ldots, G_1$ of trigraphs with $G_n=G$, where each step contracts two vertices and uses red edges to encode ambiguous adjacencies created by the merge (Bonnet et al., 2020). Standard twin-width measures the maximum red degree along such a sequence.

Total twin-width replaces the red-degree measure by a global edge count. The definition stated in (Baril et al., 5 Sep 2025) is the following: if $(G_n,\dots,G_1)$ is a contraction sequence of $G$, then its width is the maximum number of red-edges in any $G_k$ of the sequence, and $\mathbf{ttww}(G)$ is the minimum such width over all contraction sequences. Thus, whereas ordinary twin-width is vertex-local through red degree, total twin-width is sequence-global through the total number of red edges.

The same paper introduces total vertex twin-width, denoted $\mathbf{tvtww}(G)$, defined as the maximum, over the sequence, of the number of vertices incident to at least one red edge, including red loops (Baril et al., 5 Sep 2025). The basic comparison between the two total variants is

$\mathbf{ttww}(G)$0

This inequality places $\mathbf{ttww}(G)$1 and $\mathbf{ttww}(G)$2 in the same quantitative regime, up to a quadratic distortion.

2. Quantitative relation to linear clique-width

The central structural contribution of (Baril et al., 5 Sep 2025) is a tight quadratic comparison between total twin-width and linear clique-width $\mathbf{ttww}(G)$3. Before this result, the known relationships were much weaker: $\mathbf{ttww}(G)$4 and

$\mathbf{ttww}(G)$5

These were respectively double exponential and exponential.

The new theorem is

$\mathbf{ttww}(G)$6

for every graph $\mathbf{ttww}(G)$7 (Baril et al., 5 Sep 2025). The same work proves the sharper vertex-variant comparison

$\mathbf{ttww}(G)$8

Together with the bound between $\mathbf{ttww}(G)$9 and $G$0, these inequalities show that total twin-width and linear clique-width are quadratically related.

Parameter comparison	Bound
$G$1 vs. $G$2	$G$3
$G$4 vs. $G$5	$G$6
$G$7 vs. $G$8	$G$9

This establishes a much tighter correspondence between a contraction-based parameter family and an expression-based one than had previously been available.

3. Proof architecture and tightness of the quadratic bound

The lower and upper inequalities in the quadratic theorem are obtained through distinct constructions. The lower bound follows from

$G_n, G_{n-1}, \ldots, G_1$0

combined with

$G_n, G_{n-1}, \ldots, G_1$1

(Baril et al., 5 Sep 2025). This passes through the vertex-incident variant and shows that a graph of small total twin-width cannot have substantially larger linear clique-width.

The upper bound is constructive. The paper derives contraction sequences from linear clique-width expressions and shows that the number of red-edges in these sequences cannot exceed $G_n, G_{n-1}, \ldots, G_1$2; combined with

$G_n, G_{n-1}, \ldots, G_1$3

this yields the quadratic upper estimate (Baril et al., 5 Sep 2025). The result is therefore not only existential but algorithmically meaningful, because it translates one representation of a graph into another.

The same source states that the quadratic relationship is tight: there exist graph classes where the quadratic, rather than linear or exponential, dependence is achieved (Baril et al., 5 Sep 2025). Accordingly, the theorem is not merely an improvement in constants or proof technique; it identifies the correct asymptotic scale of the correspondence.

4. Algorithmic consequences

A principal consequence of the quadratic equivalence is transferability of algorithms. The paper explicitly states that any complexity result or algorithm parameterized by linear clique-width transfers to the total twin-width regime with quadratic overhead at most (Baril et al., 5 Sep 2025). This is significant because the contraction-sequence viewpoint can be operationally simpler than linear clique-width expressions: only vertex contractions are needed, rather than the four basic operations customary for clique-width formalisms.

The paper also derives an approximation consequence. Using algorithms for linear clique-width, one can, for a graph $G_n, G_{n-1}, \ldots, G_1$4 and parameter $G_n, G_{n-1}, \ldots, G_1$5, find a contraction sequence with total vertex twin-width at most $G_n, G_{n-1}, \ldots, G_1$6, or certify that the relevant value is larger, in time $G_n, G_{n-1}, \ldots, G_1$7 for computable $G_n, G_{n-1}, \ldots, G_1$8 (Baril et al., 5 Sep 2025). Through the proven inequalities, this yields fixed-parameter approximability for total twin-width and its vertex variant.

The motivating algorithmic application in (Baril et al., 5 Sep 2025) is counting homomorphisms, especially $G_n, G_{n-1}, \ldots, G_1$9-Coloring. The paper states that because total twin-width is functionally equivalent to linear clique-width up to quadratic overhead, FPT algorithms for counting $G_n=G$0-colorings parameterized by linear clique-width apply analogously to total twin-width. It further notes that dynamic programming for $G_n=G$1-Coloring can be implemented over contraction sequences parameterized by total twin-width or total vertex twin-width (Baril et al., 5 Sep 2025). This places total twin-width inside a transferable algorithmic toolkit rather than as an isolated structural invariant.

5. Position within the twin-width program

Total twin-width belongs to the broader family of contraction-sequence parameters initiated by twin-width. In the original twin-width framework, bounded red degree along a contraction sequence yields algorithmic leverage: given a $G_n=G$2-contraction sequence, FO model checking on bounded twin-width structures is fixed-parameter tractable in time $G_n=G$3, with $G_n=G$4 computable and non-elementary (Bonnet et al., 2020). The importance of explicit witnesses in that theorem provides useful context for total twin-width, whose definitions and applications are likewise sequence-based.

At the same time, the general problem of finding good contraction sequences is nontrivial. For standard twin-width, deciding whether a graph has twin-width at most $G_n=G$5 is NP-complete, and no $G_n=G$6-time algorithm exists unless ETH fails (Bergé et al., 2021). This does not state the same hardness for total twin-width, but it indicates that constructive comparisons with more accessible parameters, such as linear clique-width, have substantial methodological value.

Within this landscape, total twin-width can be viewed as a red-edge-count analogue of twin-width’s red-degree control. A plausible implication is that it is especially well suited when global red-edge budgets, rather than local red-degree budgets, align more naturally with a dynamic program or a representation transfer. The main contribution of (Baril et al., 5 Sep 2025) is to show that this variant is not merely definitional: it sits in a tight and constructive relationship with linear clique-width.

6. Scope, interpretation, and common points of confusion

Total twin-width should be distinguished from ordinary twin-width. Standard twin-width controls the maximum red degree in every intermediate trigraph (Bonnet et al., 2020), whereas total twin-width controls the maximum number of red-edges appearing anywhere in the sequence (Baril et al., 5 Sep 2025). The two notions are therefore related but not identical, and the paper does not claim equality between them.

It should also be distinguished from total vertex twin-width. The latter counts vertices incident to red edges, while total twin-width counts red edges themselves. The inequalities

$G_n=G$7

make the distinction precise and show that the two parameters coincide only up to quadratic distortion in general (Baril et al., 5 Sep 2025).

Finally, the quadratic relation to linear clique-width does not mean that the parameters are interchangeable on the nose. What the theorem provides is a tight asymptotic correspondence: $G_n=G$8 The significance of this statement is structural and algorithmic. Structurally, it replaces earlier exponential and double-exponential translations by the correct quadratic scale. Algorithmically, it means that contraction-sequence methods based on total twin-width inherit the parameterized relevance of linear clique-width while retaining the operational simplicity of vertex-contraction descriptions (Baril et al., 5 Sep 2025). The constructive proof methodology further suggests pathways for relating other contraction-sequence parameters to expression-based width measures, although that broader extrapolation remains an implication rather than a theorem stated beyond the cases proved in the paper.