Flexible List Coloring Advances

• Flexible list coloring is a graph coloring variant where each vertex has a prescribed list and local color requests must be satisfied for a positive fraction of vertices.
• It employs weighted formulations and probabilistic distributions to ensure uniform satisfaction bounds dictated by the Hall ratio and the graph's structure.
• Research explores its applications in degenerate, planar, and DP-coloring contexts, using algebraic, probabilistic, and algorithmic methods to verify flexibility.

Flexible list coloring is a refinement of list coloring in which one asks not only for a proper coloring from prescribed lists, but also for uniform guarantees on how well additional local preferences can be respected. In the standard modern formulation, a graph $G$ with list assignment $L$ and request $r$ on a subset of vertices is flexibly list-colorable if every such request admits a proper $L$-coloring satisfying a positive fraction of the requested colors; the theory also interacts with broader notions of flexibility such as non-uniform list sizes, bounded palettes, and local list constraints (Dvořák et al., 2016, Kaul et al., 2022, Dvořák, 2023, Bonamy et al., 2015).

1. Core definitions and scope

Let $G$ be a graph and $L$ a list assignment, where each vertex $v$ receives a set $L(v)$ of admissible colors. A request is a function $r$ with domain $D \subseteq V(G)$ such that $L$0 for each $L$1. The triple $L$2 is $L$3-satisfiable if there exists a proper $L$4-coloring $L$5 such that

$L$6

A graph $L$7 is $L$8-flexible if this holds for every $L$9-assignment $r$0 and every request $r$1 of $r$2 (Dvořák et al., 2016, Kaul et al., 2022).

A weighted version replaces requests on vertices by nonnegative weights on pairs $r$3 with $r$4. Weighted $r$5-flexibility asks for a proper $r$6-coloring whose satisfied weight is at least an $r$7-fraction of the total weight. This formulation is technically convenient because it converts flexibility into a distributional condition: if there exists a probability distribution on $r$8-colorings such that every admissible pair $r$9 occurs with probability at least $L$0, then $L$1 is weighted $L$2-flexible (Dvořák et al., 2016).

The dominant recent invariant is the list flexibility number, denoted $L$3 or $L$4 depending on notation. Writing the Hall ratio as

$L$5

one defines

$L$6

This is the flexibility analogue of the list chromatic number and places request satisfaction at the extremal threshold permitted by the independence structure of $L$7 (Kaul et al., 2022, Bowdoin et al., 28 Sep 2025).

The term “flexible list coloring” is not entirely uniform across the literature. In some computational and algebraic work, “flexible” refers to non-uniform list sizes $L$8 rather than requests; in bounded-palette and local list-coloring papers, it refers to restrictions on the admissible color universe or to local intersections of lists. These are distinct definitions, but they share the principle that the classical choosability model is being made more locally adaptive (Dvořák, 2023, Bonamy et al., 2015, Dhawan, 2023).

2. Extremal parameters and fundamental limitations

The Hall ratio gives the sharp universal barrier for request satisfaction. For any graph $L$9, flexibility with parameter $G$0 is impossible for every $G$1 whenever $G$2, and conversely there exists $G$3 for which $G$4 is $G$5-flexible exactly when $G$6; one formulation states that $G$7 already suffices (Kaul et al., 2022, Bowdoin et al., 28 Sep 2025). Thus $G$8 is the maximal asymptotic satisfaction ratio allowed by induced subgraphs.

This yields the basic inequality chain

$G$9

with the intermediate list packing number $L$0 providing another comparison point. In particular, every graph is $L$1-flexible, so packing arguments give one route to upper bounds on $L$2 (Kaul et al., 2022, Bowdoin et al., 28 Sep 2025).

The relationship between ordinary choosability and flexible choosability can be strict. One result constructs graphs with $L$3, and the same source states that the gap can be unbounded in general (Kaul et al., 2022). This rules out any naive expectation that request satisfaction at the Hall-ratio threshold should follow automatically from classical list coloring.

Weighted and unweighted flexibility also differ in a quantitatively nontrivial way. If every request is $L$4-satisfiable in the unweighted sense, then every weighted request is only guaranteed to be $L$5-satisfiable, and this logarithmic loss is tight (Dvořák et al., 2016). The weighted theory is therefore not merely a cosmetic reformulation; it is a stricter requirement and often the more stable one under inductive and probabilistic arguments.

3. Sparse graphs, degeneracy, and structural classes

The first general existence theorem states that for every $L$6 there exists $L$7 such that every $L$8-degenerate graph with lists of size at least $L$9 is $v$0-flexible, and in fact weighted $v$1-flexible (Dvořák et al., 2016). The same work proved that if $v$2 is prime, then every singleton request in a $v$3-degenerate graph is fully satisfiable from lists of size $v$4, linking flexible coloring to Combinatorial Nullstellensatz and Alon–Tarsi methods (Dvořák et al., 2016).

Later work sharpened these bounds. One explicit improvement states that every $v$5-degenerate graph is $v$6-flexible, and for bipartite $v$7-degenerate graphs every singleton request is $v$8-satisfiable already from $v$9-lists for all $L(v)$0, using Alon–Tarsi rather than the earlier prime-based argument (Kaul et al., 2022). The same paper elevates the Hall ratio to the canonical limiting parameter and formalizes the list flexibility number as a graph invariant (Kaul et al., 2022).

Flexibility at the degree threshold is subtler. For connected graphs of maximum degree $L(v)$1, every graph not isomorphic to $L(v)$2 is $L(v)$3-flexibly $L(v)$4-choosable, and the weighted version has explicit constant $L(v)$5 (Bradshaw et al., 2020). This is a Brooks-type phenomenon for flexible list coloring: the complete obstruction to $L(v)$6-choosability remains the complete graph, but the guarantee is now a uniform positive request-satisfaction ratio rather than mere existence of a coloring.

The sparse regime around average degree $L(v)$7 has become a focal point. A theorem for graphs with $L(v)$8 shows that such graphs are weighted $L(v)$9-flexibly $r$0-choosable, implying in particular that every planar graph of girth at least $r$1 has this property (Bi et al., 2023). This result is notable both for its explicit constant and for its proof architecture: it extends the reducible-subgraph framework by allowing reducible subgraphs of arbitrarily large order (Bi et al., 2023).

Several finer structural classes are also understood. Graphs of treewidth $r$2 are weighted $r$3-flexibly $r$4-choosable; graphs of treedepth $r$5 are $r$6-flexibly $r$7-choosable and weighted $r$8-flexible; and for $r$9-trees with suitable $D \subseteq V(G)$0-assignments, a $D \subseteq V(G)$1-flexibility bound is available (Bradshaw et al., 2020). These results show that low-width decompositions can support flexibility constants matching clique obstructions up to the expected order of magnitude.

4. Exact results, multipartite graphs, and join phenomena

Complete multipartite graphs furnish one of the cleanest exact theories. For $D \subseteq V(G)$2 with $D \subseteq V(G)$3,

$D \subseteq V(G)$4

so within this family the list flexibility number never exceeds the coloring number (Bennett et al., 2024). This gives a broad positive answer to a fundamental question asking whether $D \subseteq V(G)$5 can exceed the coloring number.

The same study determines the list epsilon-flexibility function for several complete bipartite graphs. For every star $D \subseteq V(G)$6, one has $D \subseteq V(G)$7 for all $D \subseteq V(G)$8. For $D \subseteq V(G)$9, the function is also completely determined: it equals $L$00 except in the cases $L$01 with $L$02, where it is $L$03 (Bennett et al., 2024). These exact formulas make visible how Hall-ratio optimality can emerge immediately once choosability crosses the relevant threshold.

The flexibility analogue of chromatic-choosability under joins behaves differently from classical Ohba theory. For classical list coloring, large clique joins eventually satisfy $L$04, but replacing $L$05 by $L$06 fails in general (Bowdoin et al., 28 Sep 2025). The precise characterization is that there exists $L$07 such that

$L$08

if and only if every subset $L$09 of size $L$10 contains two vertices that can receive the same color in some proper $L$11-coloring of $L$12 (Bowdoin et al., 28 Sep 2025).

This criterion separates positive and negative families sharply. Uniquely $L$13-colorable graphs with $L$14 are counterexamples, while perfect graphs, bipartite graphs, complete multipartite graphs, and odd cycles satisfy the required condition (Bowdoin et al., 28 Sep 2025). A plausible implication is that flexible list coloring is controlled not only by clique and Hall-ratio data, but also by the internal geometry of optimal colorings.

5. Proof methods and computational techniques

A recurring method is the construction of distributions on proper list colorings. In the foundational probabilistic framework, flexibility follows once every admissible pair $L$15 occurs with probability at least $L$16 under some distribution on $L$17-colorings (Dvořák et al., 2016). This viewpoint underlies weighted flexibility, facilitates inductive decompositions, and reappears in later probabilistic greedy arguments for degenerate graphs (Kaul et al., 2022).

Algebraic methods enter through the Alon–Tarsi polynomial. For an orientation $L$18 of $L$19, the graph polynomial is

$L$20

Nonvanishing coefficients of $L$21 certify choosability, and an efficient implementation based on truncated polynomial multiplication makes this approach practical for graphs with around $L$22 edges (Dvořák, 2023). The same work extends the classical certificate: when the standard Alon–Tarsi criterion fails, additional coefficients yield linear constraints on the characteristic vectors of non-colorable list assignments, often reducing the problem to a small family of assignments checked by brute force or SAT/CSP/ILP tools (Dvořák, 2023). This is particularly effective for non-uniform $L$23-choosability, which that paper treats as a flexible setting relevant to reducibility arguments (Dvořák, 2023).

Sparse-graph flexibility proofs often use reducibility and discharging. The $L$24 theorem relies on a generalized reducible-subgraph framework in which reducible subgraphs need not have bounded order, together with weighted distributions satisfying both lower marginal bounds and forbiddance properties (Bi et al., 2023). In the DP-coloring extension discussed below, the corresponding argument is phrased in terms of a potential

$L$25

combined with structural analysis of minimal counterexamples (Bradshaw et al., 14 Oct 2025).

From a pure decision-complexity perspective, general list coloring admits a broad dynamic-programming upper bound: if $L$26 fits an $L$27-vertex graph $L$28, then list-colorability can be decided in $L$29 time, where fitting means that every $L$30-vertex induced subgraph has at most $L$31 inclusion-maximal independent sets (Hujter et al., 2021). This result is not a flexibility theorem in the request-satisfaction sense, but it provides a general algorithmic template for ordinary list-coloring subproblems that flexible methods frequently reduce to.

6. Variants, extensions, and boundaries

The closest extension is DP-coloring. In this setting, every loopless multigraph with maximum average degree less than $L$32 is $L$33-flexibly DP $L$34-colorable except for an explicit infinite family $L$35, and the constant $L$36 is best possible in the weighted setting (Bradshaw et al., 14 Oct 2025). The same paper gives a negative answer to the DP analogue of the question whether every $L$37-degenerate graph should be flexibly $L$38-choosable: even some $L$39-degenerate graphs fail to be flexibly DP $L$40-colorable (Bradshaw et al., 14 Oct 2025). This sharply separates DP-coloring from ordinary list coloring.

Other nearby notions of flexibility modify the list model itself. In bounded-palette list coloring, a graph is $L$41-choosable when every $L$42-list is drawn from a fixed palette of size $L$43; here the gap between palette-restricted and unrestricted choosability can be exponential, with

$L$44

and more generally super-polynomial growth when $L$45 (Bonamy et al., 2015). In local list edge-coloring of multigraphs, the crucial parameter is the local intersection

$L$46

and sufficient conditions such as $L$47, $L$48, or $L$49 in the bipartite case recover local analogues of Shannon, Vizing, and König (Dhawan, 2023). In list supermodular coloring, elementwise list sizes $L$50 suffice, simultaneously extending Galvin, Borodin–Kostochka–Woodall, and Iwata–Yokoi (Yokoi, 2017).

The boundary between graph-theoretic and matroidal behavior is also striking. For matroids, the on-line list chromatic number equals the ordinary chromatic number, and a weighted variable-list version gives a necessary and sufficient condition for on-line $L$51-colorability (Lasoń et al., 2013). This suggests that the obstructions responsible for nontrivial flexible list coloring in graphs are substantially graph-specific rather than inherent to all independence systems.

Algorithmic variants continue to broaden the subject. In distributed computing, degree$L$52 list coloring admits randomized $L$53-round algorithms and palette sparsification theorems (Halldórsson et al., 2021); two-round LOCAL list-color reduction is possible under explicit list-size conditions in oriented graphs (Maus et al., 2020); and recursive deterministic distributed list coloring yields $L$54-round algorithms for $L$55-list coloring and related problems (Kuhn, 2019). At the opposite computational extreme, a Grover-search-based quantum algorithm solves general list coloring with query complexity

$L$56

providing quadratic speedup over exhaustive search for arbitrary list instances (Mukherjee, 2021).

Taken together, these developments show that flexible list coloring is not a single isolated theorem but a research program. Its core questions concern how much local preference can be enforced uniformly, how this threshold depends on Hall ratio and graph structure, and how algebraic, probabilistic, structural, and algorithmic methods can certify that threshold in concrete classes of graphs.