---
title: 'Dicoloring Graphs: Theory & Reconfiguration'
url: https://www.emergentmind.com/topics/dicoloring-graph
type: topic
---

# Dicoloring Graphs: Theory & Reconfiguration

Searching arXiv for foundational and recent papers on dicoloring, dichromatic number, and dicoloring graphs.
A **dicoloring graph** is the reconfiguration graph associated with acyclic colorings of a digraph. For a digraph \(D\) and an integer \(k\ge \vec{\chi}(D)\), its \(k\)-dicoloring graph \(\mathcal{D}_k(D)\) has as vertices the acyclic \(k\)-colorings of \(D\), and two colorings are adjacent when they differ on exactly one vertex [2510.04990]. The notion is inseparable from the underlying theory of **dicoloring**—also called **dichromatic coloring**—in which a color class is required to induce an acyclic subdigraph rather than an independent set [2507.10266]. In that sense, dicoloring is the directed analogue of graph coloring, the dichromatic number \(\vec{\chi}(D)\) is the minimum number of acyclic color classes, and the dicoloring graph encodes the geometry of the space of such colorings under one-vertex recoloring moves [2507.10266], [2510.04990].

## 1. Definitions and formal framework

A dicoloring of a digraph \(D\) is a vertex-coloring
\[
\varphi:V(D)\to [k]
\]
such that each color class induces an acyclic subdigraph; equivalently, for every \(i\in [k]\), the induced subdigraph \(D[\varphi^{-1}(i)]\) contains no directed cycle [2507.10266]. The least such \(k\) is the **dichromatic number**
\[
\vec{\chi}(D),
\]
which plays the role of the chromatic number in directed graph theory [2507.10266].

The reconfiguration object built from these colorings is the **\(k\)-dicoloring graph**
\[
\mathcal{D}_k(D).
\]
Its vertices are the acyclic \(k\)-colorings of \(D\), and two such colorings are adjacent precisely when they differ on one vertex [2510.04990]. Standard reconfiguration terminology is used in this setting: \(D\) is **\(k\)-mixing** when \(\mathcal{D}_k(D)\) is connected, and **\(k\)-freezable** when \(\mathcal{D}_k(D)\) contains an isolated vertex [2510.04990].

This framework differs fundamentally from ordinary graph coloring. In an undirected proper coloring, same-colored adjacent vertices are forbidden. In a dicoloring, same-colored adjacency is permitted unless it participates in a monochromatic directed cycle. Thus the obstruction is global and cyclic rather than purely local. A complete symmetric digraph, in which every pair of vertices forms a digon, behaves like a clique because any same-colored pair would already create a monochromatic directed cycle of length \(2\) [2405.00991].

A useful symmetrization principle links directed and undirected coloring. If \(G\) is an undirected graph and \(\overleftrightarrow{G}\) is obtained by replacing each edge by a digon, then
\[
\tilde{\Delta}(\overleftrightarrow{G})=\Delta(G),\qquad \omega_b(\overleftrightarrow{G})=\omega(G),\qquad \vec{\chi}(\overleftrightarrow{G})=\chi(G),
\]
so ordinary graph coloring is recovered as a special case of dicoloring on symmetric digraphs [2507.10266].

## 2. Core parameters, obstructions, and the directed analogue of clique structure

The modern structural theory of dicoloring relies on directed analogues of classical graph parameters. In the large-degree theory of dicoloring, the principal clique-type parameter is the **biclique number** \(\omega_b(D)\), where a biclique is an induced complete digraph \(\overleftrightarrow{K}_t\) on \(t\) vertices [2507.10266]. This replaces clique number in several directed Brooks- and Borodin–Kostochka-type statements.

Several degree parameters are used. For a vertex \(v\), \(d^+(v)\) and \(d^-(v)\) denote out-degree and in-degree. Global parameters include
\[
\Delta^+(D)=\max_{v\in V(D)} d^+(v),\qquad
\Delta^-(D)=\max_{v\in V(D)} d^-(v),
\]
the geometric-mean degree parameter
\[
\tilde{\Delta}(D)=\max_{v\in V(D)} \sqrt{d^+(v)\,d^-(v)},
\]
and
\[
\Delta_{\min}(D)=\max_{v\in V(D)} \min\{d^+(v),d^-(v)\}
\]
[2507.10266]. In descriptive results on Borel digraphs, the degree notion is often
\[
d^{\max}(x)=\max\{d^+(x),d^-(x)\},
\]
with boundedness assumed pointwise [2405.00991].

A central discovery of recent structural work is that the naive directed analogue of a clique obstruction is incomplete. The digraph
\[
H_k=\vec{C}_3\boxplus \overleftrightarrow{K}_{k-2},
\]
consisting of a directed triangle, a disjoint biclique of size \(k-2\), and all possible arcs in both directions between them, satisfies
\[
\tilde{\Delta}(H_k)=k,\qquad \omega_b(H_k)=k-1,\qquad \vec{\chi}(H_k)=k
\]
[2507.10266]. This makes \(H_k\) an essentially directed obstruction with no undirected counterpart. It shows that one cannot hope for a general implication of the form “\(\omega_b(D)\le \Delta-1\) implies \(\vec{\chi}(D)\le \Delta-1\)” without excluding this configuration [2507.10266].

This obstruction is also conceptually important for dicoloring graphs. Frozen or poorly connected regions of \(\mathcal{D}_k(D)\) often arise from maximal or rigid acyclic color classes, and \(H_k\)-type structures exhibit the broader phenomenon that directed cyclic interactions create rigidity absent from ordinary coloring.

## 3. Large-degree dicoloring theory

A major recent development is the extension of Borodin–Kostochka-type \((\Delta-1)\)-colorability to digraphs. The conjectural directed form states that if
\[
\min(\tilde{\Delta}(D),\Delta^+(D))=\Delta\ge 9,
\]
then
\[
\vec{\chi}(D)\le \Delta-1
\]
unless \(D\) contains either a biclique of size \(\Delta\) or the exceptional configuration \(\vec C_3\boxplus \overleftrightarrow{K}_{\Delta-2}\) [2507.10266]. The same work formulates separate \(\tilde{\Delta}\)- and \(\Delta^+\)-versions and proves large-\(\Delta\) results for both, thereby extending Reed’s large-degree theorem to directed settings [2507.10266].

The first theorem states that there exists \(\Delta_0\) such that for every \(\Delta\ge \Delta_0\), if
\[
\tilde{\Delta}(D)\le \Delta\quad\text{and}\quad \omega_b(D)\le \Delta-1,
\]
then
\[
\vec{\chi}(D)\le \Delta-1
\]
unless \(D\) contains \(\vec C_3\boxplus \overleftrightarrow{K}_{\Delta-2}\) [2507.10266]. A parallel theorem is proved for \(\Delta^+(D)\), with an explicit reduction showing it holds for
\[
\Delta\ge \max(\Delta_0,556)
\]
[2507.10266].

The same paper establishes a third independent large-degree direction using \(\Delta_{\min}\). For sufficiently large \(\Delta\), every digraph \(D\) with
\[
\Delta_{\min}(D)\le \Delta \quad\text{and}\quad \vec{\chi}(D)\ge \Delta
\]
contains a \(\Delta\)-obstruction of a specified bipartite form [2507.10266]. Two corollaries follow. First, if
\[
\omega_b(D)<\frac12(\Delta-1),
\]
equivalently \(\omega_b(D)\le \left\lceil\frac{\Delta-1}{2}\right\rceil-1\), then
\[
\vec{\chi}(D)\le \Delta-1
\]
for sufficiently large \(\Delta\) [2507.10266]. Second, if the underlying graph \(G=\mathrm{UG}(D)\) satisfies \(\omega(G)\le \Delta-1\), then again \(\vec{\chi}(D)\le \Delta-1\) [2507.10266].

These results place dicoloring theory in close formal analogy with high-degree graph coloring, but the analogy is not exact. The exceptional obstruction \( \vec C_3\boxplus \overleftrightarrow{K}_{\Delta-2}\) is intrinsically directed, and its presence is precisely what prevents a direct transplantation of the undirected theory [2507.10266].

## 4. Proof methods and structural decomposition

The principal technical engine behind the large-\(\tilde{\Delta}\) theorem is a **dense decomposition lemma for digraphs**, explicitly presented as a directed generalization of the Molloy–Reed dense decomposition for graphs [2507.10266]. With \(\bar{\Delta}(D)=\Delta\), a vertex \(v\) is called \(d\)-sparse if the digraph induced by its out-neighbourhood contains at most
\[
\Delta(\Delta-1)-d\Delta
\]
arcs; otherwise it is \(d\)-dense [2507.10266].

For every \(0<\varepsilon<1/2\) and every \(\omega(1)\le d\le o(\Delta)\), every sufficiently large digraph with \(\bar{\Delta}(D)=\Delta\) admits a partition
\[
X_1\sqcup \cdots \sqcup X_t\sqcup S
\]
such that each \(X_i\) has size close to \(\Delta\),
\[
\Delta-\frac{3}{\varepsilon}d<|X_i|<\Delta+1+4d,
\]
few arcs leave or enter \(X_i\),
\[
|E^+(X_i)|\le \frac{8}{\varepsilon}d\Delta,\qquad |E^-(X_i)|\le \frac{8}{\varepsilon}d\Delta,
\]
membership is characterized by having almost all out-neighbours inside \(X_i\),
\[
u\in X_i \iff |N^+(u)\cap X_i|\ge (1-\varepsilon)\Delta,
\]
and every vertex of \(S\) is \(d\)-sparse [2507.10266]. The resulting parts behave as “dense modules” close to bicliques.

The coloring argument then combines probabilistic and structural components. Sparse vertices are handled by random coloring and deletion: every vertex is colored uniformly from \([\Delta-1]\), then uncolored if it lies on a monochromatic directed cycle [2507.10266]. Dense parts are treated by a detailed structural analysis of minimal counterexamples. This yields quasi-biclique structure, forbidden-configuration lemmas, and a “saviour” mechanism producing many disjoint local configurations that guarantee extendability after partial random coloring [2507.10266].

The probabilistic estimates use Talagrand concentration for repeated colors in sparse out-neighbourhoods and Azuma concentration for the number of good saviour-tuples, with bad-event bounds
\[
\Pr(A_s)\le e^{-\log^2\Delta},\qquad \Pr(B_i)\le e^{-\Delta/r^3}
\]
for \(r=\log^4\Delta\) [2507.10266]. Since dependencies are confined to distance-\(2\) neighborhoods in the underlying graph, the Lovász Local Lemma yields an extendable partial \((\Delta-1)\)-dicoloring [2507.10266].

The \(\Delta^+\)-theorem is proved by a different route because \(\Delta^+\) alone gives no control on in-degrees. There the proof proceeds by minimal counterexample analysis, auxiliary digraphs, and a discharging-style boundedness argument leading to explicit index-class estimates
\[
|I^\pm|\le 70,\qquad |I^{\le 6}|\le 441,\qquad |I^{>6}|\le 42,
\]
and hence
\[
\Delta\le 2+70+441+42=555,
\]
contradicting \(\Delta\ge 556\) [2507.10266].

A plausible implication is that the modern theory of dicoloring is no longer primarily extremal in the elementary sense; it is increasingly organized around decomposition, concentration, and minimal-counterexample rigidity.

## 5. Reconfiguration: the dicoloring graph \(\mathcal{D}_k(D)\)

The reconfiguration perspective studies not only the existence of acyclic colorings but also the connectivity and diameter of the space of such colorings. In this setting, the dicoloring graph \(\mathcal{D}_k(D)\) is the analogue of the undirected coloring graph \(\mathcal{C}_k(G)\) [2510.04990].

A general theorem states that if a digraph \(D\) is uniquely \(k\)-colorable, then \(D\) is \(k\)-freezable and \(\mathcal{D}_k(D)\) consists of \(k!\) isolated vertices [2510.04990]. This provides a standard mechanism for total reconfiguration failure. From previously known uniquely colorable tournaments, it follows that for each \(k\ge 2\) there exists a tournament whose \(k\)-dicoloring graph has \(k!\) isolated vertices [2510.04990].

There is, however, a universal high-color mixing bound depending on order rather than dichromatic number. Any digraph \(D\) of order \(n\) is \(k\)-mixing for every \(k\ge n-1\), and \(\mathcal{D}_k(D)\) has diameter at most \(2n\) [2510.04990]. This guarantees eventual connectivity once the number of colors is sufficiently large relative to the number of vertices.

By contrast, there is **no** function \(\phi(\vec{\chi})\) depending only on the dichromatic number such that \(\mathcal{D}_k(D)\) is connected for all \(D\) and all \(k\ge \phi(\vec{\chi}(D))\) [2510.04990]. The paper establishing this gives a directed analogue of a theorem of Cereceda in undirected recoloring theory. Its significance is that \(\vec{\chi}(D)\) is too coarse to control reconfiguration geometry: two digraphs with the same dichromatic number may have radically different dicoloring graphs [2510.04990].

This negative result corrects a common misconception. Existence theory for dicolorings often suggests that once \(k\) exceeds the dichromatic number by a bounded amount, the solution space should become connected. The reconfiguration results show that no such principle holds with \(\vec{\chi}(D)\) alone as the governing parameter [2510.04990].

## 6. Circulant tournaments and explicit dicoloring graphs

The most detailed current calculations of dicoloring graphs concern circulant tournaments. Two families are central: the cyclic circulant tournament
\[
\vec{C}_{2n+1}\langle \emptyset\rangle=\vec{C}_{2n+1}(1,2,\dots,n)
\]
and the one-reversed-jump family
\[
\vec{C}_{2n+1}\langle n\rangle
\]
[2510.04990].

For the cyclic family, one has
\[
\vec{\chi}(\vec{C}_{2n+1}\langle \emptyset \rangle)=2,
\]
and every \(2\)-coloring partitions the vertices into two cyclic intervals of the form
\[
\{a,a+1,\dots,a+n\}
\quad\text{and}\quad
\{a+n+1,\dots,a-1\}
\]
for some \(a\in\mathbb{Z}_{2n+1}\) [2510.04990]. This interval structure yields an exact reconfiguration theorem:
\[
\mathcal{D}_2(\vec{C}_{2n+1}\langle \emptyset \rangle)\cong C_{4n+2},
\qquad
\diam=2n+1
\]
[2510.04990]. For every \(k\ge 3\), the same family is \(k\)-mixing and the diameter satisfies
\[
\diam(\mathcal{D}_k(\vec{C}_{2n+1}\langle \emptyset \rangle))
\le
4n+1+\left\lfloor \frac{n+1}{2}\right\rfloor
\]
[2510.04990].

The family \(\vec{C}_{2n+1}\langle n\rangle\) is more rigid. A key local obstruction is the **forbidden triangle**
\[
\{i,i+n,i+n+1\},
\]
which induces a maximal acyclic subtournament [2510.04990]. Acyclic subtournaments have size at most \(n\), and those of size \(n\) have an explicit arithmetic classification [2510.04990]. A major extension lemma shows that any color class that is not a forbidden triangle can be transformed into a canonical interval of length \(n\), which becomes the basis of most connectivity proofs [2510.04990].

For \(n\ge 5\), the \(3\)-dicoloring graph is connected with
\[
\diam(\mathcal{D}_{3}(\vec{C}_{2n+1}\langle n \rangle))
\le
3n+4+\left\lfloor \frac{2n+1}{3}\right\rfloor
\]
[2510.04990]. For \(n\ge 4\), \(k\ge 4\), and
\[
k\neq \frac{2n+1}{3},
\]
one has connectivity and
\[
\diam(\mathcal{D}_{k}(\vec{C}_{2n+1}\langle n \rangle))
\le
4n+2+\left\lfloor \frac{n}{2} \right\rfloor
\]
[2510.04990].

The exceptional case is
\[
k=\frac{2n+1}{3}
\]
when \(2n+1\equiv 0\pmod 3\). Then \(\mathcal{D}_k(\vec{C}_{2n+1}\langle n\rangle)\) is not connected; it contains
\[
3k!
\]
isolated vertices arising from colorings whose classes are all forbidden triangles, together with one additional component whose diameter is at most
\[
2n+1+\left\lfloor \frac{2n+1}{4}\right\rfloor
\]
[2510.04990].

A distinguished special case is the Paley tournament
\[
ST_7=\vec{C}_7(1,2,4)=\vec{C}_7\langle 3\rangle.
\]
For every \(k\ge 3\),
\[
\mathcal{D}_k(ST_7)\text{ is connected and has diameter }8
\]
[2510.04990]. For \(k=3,4,5,6\), computational enumeration gives exact orders, sizes, and the same diameter \(8\); for example, when \(k=3\), \(\mathcal{D}_3(ST_7)\) has order \(504\), size \(1512\), and diameter \(8\) [2510.04990].

## 7. Related directions: planar, measurable, and definable dicoloring

Dicoloring also appears in planar and descriptive settings, where the emphasis shifts from finite extremal structure to acyclic partitions and definability.

In planar digraph theory, a partition into two acyclic sets is exactly a \(2\)-dicoloring. One recent line studies **CAI-partitions**, where the vertex set is partitioned into a connected acyclic subgraph and an independent set [2412.11774]. For Eulerian oriented planar triangulations with tripartition \(I_1,I_2,I_3\), if one of the induced bipartite graphs \(G-I_i\) has a CAI-partition, then the full digraph can be partitioned into two acyclic sets [2412.11774]. This yields a sufficient condition for \(2\)-dicoloring but not a characterization: there exists an Eulerian oriented planar triangulation such that for every tripartition class \(I\), the graph \(G-I\) admits no CAI-partition [2412.11774]. The paper thus proves both positive subcubic results and a limitation theorem for this strategy [2412.11774].

In descriptive set theory, the measurable analogue of Brooks’s theorem has been established for directed graphs. If \(D\) is a Borel digraph on a standard Borel space, \(d^{\max}(x)\le d\ge 3\) for all \(x\), and \(D\) does not contain the complete symmetric digraph on \(d+1\) vertices, then \(D\) admits a \(\mu\)-measurable \(d\)-dicoloring for every Borel probability measure \(\mu\), and a \(\tau\)-Baire-measurable \(d\)-dicoloring for every compatible Polish topology \(\tau\) [2405.00991]. The same work proves a definable Gallai theorem: bounded-degree Borel digraphs with no Gallai-tree component are Borel degree-list-dicolorable [2405.00991].

These results show that the dicoloring paradigm now spans several regimes:

| Regime | Typical question | Representative result |
|---|---|---|
| Finite extremal | How large can \(\vec{\chi}(D)\) be under degree and obstruction constraints? | Large-\(\Delta\) \((\Delta-1)\)-dicoloring theorems [2507.10266] |
| Reconfiguration | Is \(\mathcal{D}_k(D)\) connected, and what is its diameter? | Circulant tournament classifications [2510.04990] |
| Planar decomposition | When does a planar orientation admit a \(2\)-dicoloring? | CAI-based sufficient conditions and counterexamples [2412.11774] |
| Descriptive | Can dicolorings be chosen measurably or Baire-measurably? | Measurable Brooks theorem for digraphs [2405.00991] |

A plausible implication is that “dicoloring graph” has become a nexus term joining three traditions: directed coloring proper, reconfiguration theory, and structural/definable extensions of Brooks- and Gallai-type theorems.

## 8. Significance, limitations, and current outlook

The study of dicoloring graphs has clarified that the directed analogue of coloring inherits some but not all of the structure of ordinary graph coloring. The finite existence theory now includes large-degree Brooks- and Borodin–Kostochka-type theorems with sharp directed obstructions [2507.10266]. Reconfiguration theory has shown that the space of acyclic colorings can be explicitly analyzable on structured families, yet dichromatic number alone does not control its connectivity [2510.04990]. Planar and descriptive variants further demonstrate that acyclic color-class constraints interact fruitfully with topology, planarity, and definability [2412.11774], [2405.00991].

Several misconceptions are now untenable. It is false that biclique obstructions alone explain high dichromatic number in large-degree digraphs, because the exceptional configuration \(\vec C_3\boxplus \overleftrightarrow{K}_{\Delta-2}\) must also be excluded [2507.10266]. It is also false that increasing \(k\) as a function of \(\vec{\chi}(D)\) must eventually make \(\mathcal{D}_k(D)\) connected; no such universal function exists [2510.04990]. And in descriptive settings, a Borel Brooks theorem fails even where measurable and Baire-measurable versions remain valid [2405.00991].

The present state of the subject suggests two broad tendencies. First, structural dicoloring theory is converging on a refined obstruction-based picture, with different degree parameters—\(\tilde{\Delta}\), \(\Delta^+\), and \(\Delta_{\min}\)—supporting distinct but related theorems [2507.10266]. Second, reconfiguration theory is identifying families where \(\mathcal{D}_k(D)\) can be described exactly, including cycle structure, isolated vertices, and explicit diameter bounds [2510.04990]. Together these developments place the dicoloring graph at the center of a rapidly maturing theory of directed coloring.

Source: https://www.emergentmind.com/topics/dicoloring-graph