Edge-Coloring and Forbidden Patterns

• Edge-coloring problems with forbidden patterns are defined as constraints on colored graphs to exclude specific color configurations through homomorphism or local transition models.
• The topic addresses decision, counting, and extremal problems by analyzing thresholds, forbidden substructures, and complexity classifications in various graph settings.
• Applications include designing graph algorithms for proper edge-coloring, avoiding rainbow structures, and optimizing color distributions in extremal and Ramsey-type contexts.

Edge-coloring problems with forbidden patterns study constraints on colored graphs in which specified color configurations are excluded, either as exact colored subgraphs, as homomorphic images of colored obstructions, or as prescribed color distributions on a target graph. In one general formulation, the input is a graph $X$, the colors are $[r]$, and the question is whether $E(X)$ can be colored so that no member of a finite obstruction family $\mathcal F$ admits a color-preserving homomorphism into the resulting colored graph (Barsukov et al., 25 Jul 2025). The subject therefore encompasses decision problems, extremal thresholds, counting questions, and structural colorings such as rainbow-avoiding, parity-constrained, and locally proper colorings.

1. Formal models for forbidden patterns

A general obstruction-based language is given by edge-colored graphs $(G,\gamma)$, with $\gamma:E(G)\to[r]$, and a finite family $\mathcal F$ of finite edge-colored graphs over the same color set. The coloring problem $\Col(\mathcal F)$ asks whether an uncolored input graph $X$ has an edge-coloring $\gamma:E(X)\to[r]$ such that $[r]$0 is $[r]$1-free in the homomorphism sense, while the extension problem $[r]$2 asks for an $[r]$3-free extension of a partial edge-coloring $[r]$4 on $[r]$5 (Barsukov et al., 25 Jul 2025). This homomorphism-based formulation is broad enough to subsume Ramsey-type avoidance, monochromatic clique avoidance, and mixed cycle/clique restrictions.

A complementary formulation constrains the color classes inside a forced copy of a graph $[r]$6. For a family $[r]$7 of graphs, the parameter

$[r]$8

is the smallest $[r]$9 such that every edge-coloring of $E(X)$0 with at least $E(X)$1 colors contains a copy of $E(X)$2 in which each color class induces a member of $E(X)$3 (Caro et al., 2024). Anti-Ramsey theory appears as the special case $E(X)$4, local rainbow colorings correspond to $E(X)$5, strong odd colorings correspond to the family of all odd graphs, and class-parity colorings correspond to the family of all odd graphs together with all even graphs (Caro et al., 2024).

A third formalization treats forbidden patterns as local transition rules on consecutive edges. In a forbidden-transition graph $E(X)$6, a transition is an unordered pair of incident edges, $E(X)$7 specifies the permitted transitions, and a walk is compatible if every consecutive pair of edges belongs to $E(X)$8. Properly colored walks in edge-colored graphs are a special case obtained by declaring $E(X)$9 exactly when the colors of $\mathcal F$0 and $\mathcal F$1 differ (Bellitto et al., 2020). This suggests a common viewpoint in which forbidden patterns may be encoded as local constraints on edge pairs, global colored obstructions, or admissible color-class families.

2. Local pattern avoidance in proper edge-colorings

One important branch of the subject imposes forbidden patterns on proper edge-colorings. A proper edge-coloring of $\mathcal F$2 is a map $\mathcal F$3 such that adjacent edges receive distinct colors. Within this setting, Gyárfás and Sárközy introduced A-, B-, and C-colorings: an A-coloring is a proper edge-coloring such that the union of any two color classes contains no $\mathcal F$4 and no $\mathcal F$5; a B-coloring is a proper edge-coloring such that every $\mathcal F$6 is rainbow; and a C-coloring satisfies both conditions (Wang, 5 Jun 2026). Strong edge-coloring is the more restrictive condition that all distance-2 edges must receive distinct colors.

A recent addition to this hierarchy is the D-coloring. For a graph $\mathcal F$7, a D-coloring is a proper edge-coloring such that every diamond subgraph, meaning an induced $\mathcal F$8, is rainbow (Wang, 5 Jun 2026). Equivalently, if two edges $\mathcal F$9 are at distance $(G,\gamma)$0 in the line graph and their four endpoints span at least three of the four possible connecting edges, then $(G,\gamma)$1 and $(G,\gamma)$2 must receive distinct colors (Wang, 5 Jun 2026). The associated parameter is the D-chromatic index $(G,\gamma)$3, and the basic inequalities

$(G,\gamma)$4

place D-coloring strictly between ordinary proper edge-coloring and stronger rainbow-type conditions (Wang, 5 Jun 2026).

A different proper-coloring constraint forbids rainbow cycles globally. A proper rainbow-cycle-forbidding edge-coloring is a proper edge-coloring in which every cycle has at least two edges with the same color; a graph is PRCF-good if it admits such a coloring and PRCF-bad otherwise (Noble, 2021). This condition is not expressed by forbidding one fixed small subgraph, but by excluding the entire class of rainbow cycles. It therefore sits naturally beside D-coloring and B-coloring as a proper edge-coloring problem with a forbidden pattern family rather than a single forbidden adjacency relation.

3. Unavoidability, omnitonality, and density thresholds

A large extremal literature asks when density in each color forces the appearance of prescribed colored patterns. For each $(G,\gamma)$5 and $(G,\gamma)$6, the finite family $(G,\gamma)$7 consists of certain $(G,\gamma)$8-edge-colored complete graphs $(G,\gamma)$9 whose vertex set is partitioned into parts $\gamma:E(G)\to[r]$0 of size $\gamma:E(G)\to[r]$1, each $\gamma:E(G)\to[r]$2 and each $\gamma:E(G)\to[r]$3 is monochromatic, all $\gamma:E(G)\to[r]$4 colors appear, and every part is essential in the sense that removing it makes some color disappear. For sufficiently large $\gamma:E(G)\to[r]$5, any $\gamma:E(G)\to[r]$6-edge-colored $\gamma:E(G)\to[r]$7 with at least $\gamma:E(G)\to[r]$8 edges in each color contains a member of $\gamma:E(G)\to[r]$9, and it is conjectured that the exponent $\mathcal F$0 is sufficient (Bowen et al., 2019). In the case $\mathcal F$1, Girão and Narayanan proved the $\mathcal F$2 exponent, and the same paper gives a short proof by dependent random choice (Bowen et al., 2019).

In the bipartite host $\mathcal F$3, the threshold picture becomes more explicit. For every fixed $\mathcal F$4, there exists $\mathcal F$5 such that any 2-edge-coloring of $\mathcal F$6 with at least $\mathcal F$7 edges in each color contains a colored copy of $\mathcal F$8 in which one color induces a $\mathcal F$9 (Hansberg et al., 2024). This leads to the notions of bipartite $\Col(\mathcal F)$0-tonality and bipartite omnitonality: a bipartite graph $\Col(\mathcal F)$1 is bipartite $\Col(\mathcal F)$2-tonal iff there exists $\Col(\mathcal F)$3 contained in one bipartition class with $\Col(\mathcal F)$4, and $\Col(\mathcal F)$5 is bipartite omnitonal iff such a set exists for every $\Col(\mathcal F)$6; every tree is bipartite omnitonal (Hansberg et al., 2024).

For 2-colorings of the complete graph, omnitonality is characterized differently. A graph $\Col(\mathcal F)$7 is omnitonal if, for every $\Col(\mathcal F)$8, there exists both a partition $\Col(\mathcal F)$9 with $X$0 and a subset $X$1 with $X$2 (Caro et al., 2018). Every omnitonal graph is bipartite, and every tree is omnitonal; for bipartite amoebas, the threshold satisfies

$X$3

for all sufficiently large $X$4 (Caro et al., 2018). Balanceable graphs are the analogous class for balanced red-blue distributions, and the later work on the evolution of unavoidable bichromatic patterns shows that for every $X$5 there are graphs $X$6 with $X$7, while there are also graphs with $X$8, with the latter class characterized exactly (Caro et al., 2022). This suggests a spectrum ranging from constant thresholds to near-quadratic thresholds depending on the structural flexibility of the target graph.

4. Specific extremal pattern families

The D-coloring problem provides a concrete recent example of a proper edge-coloring problem with a forbidden pattern. For maximum degree $X$9, the D-chromatic index satisfies

$\gamma:E(X)\to[r]$0

the conjectured sharp bound is

$\gamma:E(X)\to[r]$1

and this conjecture is verified for $\gamma:E(X)\to[r]$2 (Wang, 5 Jun 2026). Complete graphs are extremal, since

$\gamma:E(X)\to[r]$3

and the planar case is isolated by a separate conjecture with sharp examples (Wang, 5 Jun 2026).

Rainbow-cycle avoidance in complete bipartite graphs yields a different extremal family. For every $\gamma:E(X)\to[r]$4, if $\gamma:E(X)\to[r]$5, then any proper $\gamma:E(X)\to[r]$6-edge-coloring of $\gamma:E(X)\to[r]$7 contains a multicolored $\gamma:E(X)\to[r]$8; in contrast, when $\gamma:E(X)\to[r]$9 is odd, there exists a proper $[r]$00-edge-coloring of $[r]$01 forbidding multicolored $[r]$02 (Fu et al., 2014). The case of $[r]$03 is completely determined: $[r]$04 where $[r]$05 denotes the pairs $[r]$06 for which some proper coloring of $[r]$07 forbids multicolored $[r]$08 (Fu et al., 2014).

The PRCF condition gives yet another family of forbidden rainbow patterns. The Hoffman–Singleton graph is PRCF-bad, which answers positively the question whether PRCF-bad graphs of girth greater than $[r]$09 exist (Noble, 2021). The same paper shows that sufficiently large generalized polygons yield PRCF-bad graphs of girth $[r]$10, $[r]$11, $[r]$12, and $[r]$13, with the split Cayley hexagon and the Ree–Tits octagon providing the girth-$[r]$14 and girth-$[r]$15 cases (Noble, 2021). A plausible implication is that forbidding rainbow cycles in a proper edge-coloring becomes increasingly rigid on sparse, high-girth incidence structures.

5. Complexity, extensions, and forbidden-color lists

In the homomorphism framework, the relation between uncolored and partially colored instances is governed by colored determiners. Under Assumption 3.1—existence, for each color, of an $[r]$16-safe, $[r]$17-remote colored determiner—the problems $[r]$18 and $[r]$19 are polynomial-time equivalent (Barsukov et al., 25 Jul 2025). For obstruction families consisting of monochromatic odd cycles and cliques, possibly together with all colorings of a fixed $[r]$20, the same paper proves a P/NP-complete dichotomy: $[r]$21 is in P in a trivial clique-free case and NP-complete otherwise (Barsukov et al., 25 Jul 2025).

The forbidden-transition model yields a parameterized complexity picture for local color-sequence constraints. Finding a compatible path in a forbidden-transition graph is W[1]-hard when parameterized by the distance to a linear forest, and therefore also by treewidth or pathwidth (Bellitto et al., 2020). By contrast, compatible paths are fixed-parameter tractable by treecut-width, and an algebraic rank-based dynamic program decides whether an edge-colored graph of treewidth $[r]$22 has a properly colored Hamiltonian cycle in deterministic time

$[r]$23

where $[r]$24 is the number of colors (Bellitto et al., 2020). This separates arbitrary forbidden-transition systems from the specific “no consecutive equal colors” pattern.

A different restriction model assigns each edge a list of forbidden colors. If $[r]$25 is $[r]$26-edge colorable and $[r]$27 is a $[r]$28-sparse list assignment, then for

$[r]$29

there exists a proper $[r]$30-edge coloring of $[r]$31 that avoids $[r]$32 (Pham, 2019). The same paper proves a stronger sparse-support statement: if $[r]$33 for every edge and the set of edges with $[r]$34 forms a distance-3 matching, then $[r]$35 is avoidable (Pham, 2019). These results treat forbidden colors themselves as the obstruction family and use the presence of many 2-colored $[r]$36-cycles as a local recoloring resource.

6. Enumeration and asymptotic optimization

Counting pattern-avoiding colorings produces a parallel extremal theory. For a pattern $[r]$37, written as a partition of $[r]$38 into at most $[r]$39 color classes, $[r]$40 denotes the number of $[r]$41-edge-colorings of $[r]$42 avoiding $[r]$43 (Benevides et al., 2016). When $[r]$44, there always exists a complete multipartite graph on $[r]$45 vertices that is $[r]$46-extremal, and for every non-monochromatic pattern except the small exceptional patterns $[r]$47 and, for $[r]$48, $[r]$49, every $[r]$50-extremal graph is complete multipartite (Benevides et al., 2016). For the rainbow triangle pattern $[r]$51, any graph with near-maximal number of 3-edge-colorings avoiding $[r]$52 must be almost complete (Benevides et al., 2016).

For forbidden monochromatic cliques, the Erdős–Rothschild-type quantity $[r]$53 counts $[r]$54-edge-colorings in which color $[r]$55 contains no $[r]$56, and $[r]$57 is the maximum over all $[r]$58-vertex graphs (Pikhurko et al., 2016). For every $[r]$59 and $[r]$60, at least one extremal graph is complete multipartite, and a finite optimization problem $[r]$61 satisfies

$[r]$62

together with a stability theorem describing near-extremal complete multipartite graphs (Pikhurko et al., 2016). This places asymptotic counting inside a finite-template framework closely analogous to reduced graphs in extremal graph theory.

The color-class family viewpoint produces further asymptotic regimes. For every $[r]$63,

$[r]$64

are all asymptotically

$[r]$65

where $[r]$66, $[r]$67, $[r]$68, $[r]$69, and $[r]$70 denote local-rainbow, strong-odd, strong-parity, class-parity, and local-parity thresholds, respectively (Caro et al., 2024). By contrast, conflict-free and odd-colored copies satisfy $[r]$71 for every graph $[r]$72 (Caro et al., 2024). A plausible conclusion is that forbidding patterns globally inside each color class tends to produce quadratic Turán-type behavior, whereas vertex-local uniqueness or oddness conditions are often linear.