---
title: Below-Guarantee Graph Coloring
url: https://www.emergentmind.com/topics/below-guarantee-graph-coloring
type: topic
---

# Below-Guarantee Graph Coloring

Below-guarantee graph coloring denotes a family of research directions in which the objective is to improve on a baseline coloring guarantee. In vertex coloring, the baseline is typically a trivial or near-trivial upper bound such as $\Delta(G)+1$ or $\Delta(G)$, and the goal is to prove that $\chi(G)\le \Delta(G)-1$ unless a structural obstruction such as a large clique forces otherwise. In parameterized complexity, the question is whether a graph can be colored with $g-k$ colors when $g$ is an efficiently computable guarantee such as $n$ or $\omega+\overline{\mu}$. In multigraph edge-coloring, the goal is to maximize the size of a $\Delta$-edge-colorable subgraph beyond the fraction that follows from Shannon’s theorem. In exact sampling, the term is also used for algorithms that produce a perfect uniform sample of proper colorings while requiring fewer colors than earlier algorithmic guarantees. These usages are developed, respectively, in "Coloring some $(P_6,C_4)$-free graphs with $\Delta-1$ colors" [2405.18455], "Graph Coloring Below Guarantees via Co-Triangle Packing" [2509.12347], "Beyond the Shannon's Bound" [1309.6069], and "Fewer colors for perfect simulation of proper colorings" [2004.08716].

## 1. Canonical guarantees and the meaning of “below guarantee”

A recurring pattern in the literature is to begin with a universal guarantee and then ask whether one can save colors relative to that guarantee. For finite simple graphs, the basic parameters are the chromatic number $\chi(G)$, the maximum degree $\Delta(G)$, and the clique number $\omega(G)$. The standard inequalities are $\chi(G)\ge \omega(G)$ and $\chi(G)\le \Delta(G)+1$ by greedy coloring. Brooks’ theorem sharpens the latter to
$$
\chi(G)\le \max\{\Delta(G),\omega(G)\}\quad\text{for }\Delta(G)\ge 3.
$$
The Borodin–Kostochka conjecture is the canonical below-guarantee statement in this setting:
$$
\chi(G)\le \max\{\Delta(G)-1,\omega(G)\}\quad\text{for }\Delta(G)\ge 9.
$$
Equivalently, for $\Delta(G)\ge 9$, any graph with $\chi(G)=\Delta(G)$ must contain $K_{\Delta(G)}$ [2405.18455].

In parameterized coloring, the same idea is recast as a decision problem. The graph is trivially colorable with $g$ colors, and one asks whether it is colorable with $g-k$ colors, where $k$ measures how many colors are saved. Two guarantees are emphasized: $g=n$, which yields $(n-k)$-Coloring or Dual Coloring, and the stronger structural guarantee $g=\omega+\overline{\mu}$, where $\overline{\mu}=\mu(\overline{G})$ is the maximum matching size in the complement graph. Every graph is colorable with at most $\omega+\overline{\mu}$ colors, and $\omega+\overline{\mu}\le n$ [2509.12347].

For multigraph edge-coloring, the relevant baseline is Shannon’s theorem:
$$
\chi'(G)\le \left\lfloor \frac{3\Delta}{2}\right\rfloor.
$$
Below-guarantee edge-coloring asks how large a subgraph can be colored with only $\Delta$ colors even when the entire multigraph is not $\Delta$-edge-colorable. Writing $y_\Delta(G)$ for the maximum fraction of edges in a $\Delta$-edge-colorable subgraph, Shannon’s theorem implies
$$
y_\Delta(G)\ge \frac{\Delta}{\left\lfloor \frac{3\Delta}{2}\right\rfloor},
$$
and the research objective is to beat this baseline whenever specific obstructions are absent [1309.6069].

In exact sampling, the guarantee is algorithmic rather than existential. Earlier perfect-sampling methods required $k>3\Delta$ colors to sample uniformly from the proper colorings of a graph. The below-guarantee objective is to lower that threshold while preserving exact uniformity and efficient expected runtime [2004.08716].

## 2. Below $\Delta$ in vertex coloring: the Borodin–Kostochka paradigm on hereditary classes

The paper "Coloring some $(P_6,C_4)$-free graphs with $\Delta-1$ colors" proves the Borodin–Kostochka conjecture for two hereditary subclasses of $(P_6,C_4)$-free graphs:
- $(P_6,C_4,K_7)$-free graphs;
- $(P_6,C_4,C_5^+)$-free graphs [2405.18455].

Here $P_t$ denotes the induced path on $t$ vertices and $C_t$ the induced cycle on $t$ vertices. The graph $C_5^+$ is defined explicitly from an induced $C_5=v_1v_2v_3v_4v_5v_1$ by adding a triangle $xyzx$ and an edge $t_1t_2$ so that
$$
N(x)\cap V(C)=N(y)\cap V(C)=\{v_1,v_2,v_3\},\quad N(z)\cap V(C)=\{v_2\},
$$
$$
N(t_1)\cap V(C)=\{v_4,v_5\},\quad N(t_2)\cap V(C)=\{v_1,v_4,v_5\},
$$
with $N(t_1)\cap\{x,y,z\}=\{z\}$ and $N(t_2)\cap\{x,y,z\}=\varnothing$. The main theorems state that if $G$ is $(P_6,C_4,H)$-free with $H\in\{K_7,C_5^+\}$ and $\Delta(G)\ge 9$, then
$$
\chi(G)\le \max\{\Delta(G)-1,\omega(G)\}.
$$

The $K_7$ case follows from an external structural bound already known for $(P_6,C_4)$-free graphs:
$$
\chi(G)\le \left\lceil \frac{5}{4}\omega(G)\right\rceil.
$$
If $G$ is additionally $K_7$-free, then $\omega(G)\le 6$, hence $\chi(G)\le 8$. This contradicts $\chi(G)=9$ in a minimal counterexample, so the Borodin–Kostochka bound holds immediately in this subclass [2405.18455].

The $C_5^+$ case is substantially more intricate and exemplifies the minimal-counterexample and recoloring paradigm. For a hereditary class $\mathcal{G}$, Catlin–Kostochka-type reductions show that it suffices to prove the conjecture for graphs in $\mathcal{G}$ with $\Delta(G)=9$. A minimal counterexample with $\Delta(G)=9$ is called a relaxed graph; such a graph is $9$-vertex-critical, satisfies $\omega(G)\le 8$, and has $d(v)\in\{8,9\}$ for every vertex. Fixing a $9$-degree vertex $u$ with
$$
N(u)=\{u_1,u_2,\dots,u_7,x,y\},
$$
the proof considers a proper $8$-coloring $\varphi$ of $G-\{u\}$ such that $\varphi(u_i)=i$ for $i\in\{1,\dots,7\}$ and $\varphi(x)=\varphi(y)=8$. This is the canonical setup for recoloring arguments.

Several structural lemmas drive the contradiction. For each $i\in[7]$, the vertex $u_i$ has no missing color among $[8]$ in $G-\{u\}$, so $u_i$ has at most one repeat color in its neighborhood. If $u_i$ and $u_j$ are nonadjacent, then there exists an induced $(i,j)$-alternating $u_i$–$u_j$ path, and internal vertices on such alternating paths have no missing colors. A global constraint shows that in a relaxed graph, at least one of the following must fail: each $u_i$ is nonadjacent to at most two of $\{u_1,\dots,u_7\}$, or $\lvert (N(x)\cup N(y))\cap\{u_1,\dots,u_7\}\rvert\ge 5$. A further structural lemma yields some $i\in[7]$ with
$$
|N(u_i)\cap\{u_1,\dots,u_7\}|\le 3.
$$
This sparseness around one neighbor becomes the seed for the terminal case analysis.

The proof of the $(P_6,C_4,C_5^+)$-free theorem then constructs short alternating $(1,j)$-paths between $u_1$ and $u_j$ for $j\in\{2,3,4\}$ and rules out the possible intersections of internal vertices. The exclusions rely on three ingredients: internal vertices on alternating paths have no missing colors; certain adjacencies force induced $C_4$ or $P_6$; and specific local patterns on vertices around $u$, $u_1$, $u_2$, $u_3$, $u_4$, $x$, and $y$ inevitably induce a $C_5^+$, contradicting the class assumption. The argument is existential and nonconstructive in the algorithmic sense, but it demonstrates a standard mechanism of below-guarantee vertex coloring: identify a local obstruction that sustains a hypothetical $\chi=\Delta$ configuration, then exclude it through alternating-path recoloring and forbidden-subgraph structure [2405.18455].

The same paper also records a corollary for further subclasses. Since the $C_5^+$ construction contains several smaller forbidden subgraphs as induced subgraphs, the Borodin–Kostochka bound extends to $(P_6,C_4,H)$-free graphs with
$$
H\in\{\text{kite}^+,\text{flag}^+,\text{tripod},\text{HVN},K_5-e,\text{crown},\text{butterfly}\}.
$$
It does not, however, settle the full $(P_6,C_4)$-free case; the paper isolates the unresolved regime to $\omega(G)\in\{7,8\}$.

## 3. Below-guarantee parameterizations and co-triangle packing

A distinct algorithmic interpretation appears in "Graph Coloring Below Guarantees via Co-Triangle Packing" [2509.12347]. The basic problem is $\ell$-Coloring: given a graph $G=(V,E)$, decide whether there exists a proper coloring with at most $\ell$ colors. The below-guarantee form asks whether $G$ is colorable with $g-k$ colors, where $g$ is a trivial or structural upper bound.

The paper develops a win–win framework around co-triangles, that is, independent sets of size $3$ or induced $\overline{K_3}$ subgraphs. A co-triangle packing is a collection of vertex-disjoint co-triangles obtained greedily in polynomial time. For Dual Coloring, where $g=n$, let $T$ be a maximal packing and $t=|T|$. If $t\ge k/2$, then the graph is immediately a YES-instance for $(n-k)$-Coloring because one color is assigned to each packed co-triangle and distinct colors to all remaining vertices, using
$$
t+(n-3t)=n-2t\le n-k
$$
colors. If instead $t<k/2$, then the vertices in the packing form a co-triangle modulator $S$ of size $|S|=3t<3k/2$ such that $G[V\setminus S]$ is co-triangle-free. This yields a randomized
$$
O^*(2^{3k/2})
$$
algorithm for $(n-k)$-Coloring, improving the previous $O^*(4^k)$ bound.

The reason co-triangle-free structure is useful is that every color class in $G[V\setminus S]$ has size at most $2$; otherwise an independent triple would appear. This bounded color-class size enables an algebraic modulator algorithm. The principal extension theorem states that if $S$ is a co-triangle-free modulator of size $p$, then $\ell$-Coloring can be solved in randomized
$$
O^*(2^p)
$$
time for arbitrary $\ell$. The proof works in the complement graph, where $\ell$-Coloring becomes $\ell$-Clique Cover. Valid cover types $(t_0,t_1)$ are enumerated, an auxiliary graph $H$ is built so that perfect matchings encode how cliques intersect $V\setminus S$, and a skew-symmetric matrix is formed whose Pfaffian polynomial enumerates perfect matchings. Variables are then replaced by clique-polynomials over a squarefree ring $R=F[Y]/(y^2)$, so that vertex-disjointness in $S$ is enforced combinatorially. Randomized polynomial identity testing, a division-free Pfaffian circuit, and fast subset convolution produce the $O^*(2^p)$ bound.

The same packing principle also applies to the stronger guarantee $g=\omega+\overline{\mu}$, where $\overline{\mu}=\mu(\overline{G})$. Every graph is colorable with at most $\omega+\overline{\mu}$ colors: if $M$ is a maximal matching in $\overline{G}$ and $C$ is the set of vertices not covered by $M$, then $C$ is independent in $\overline{G}$ and hence a clique in $G$ of size at most $\omega$; giving one color to each edge of $M$ and one fresh color to each vertex of $C$ yields a proper coloring with at most $\omega+\overline{\mu}$ colors. Since $\omega+\overline{\mu}\le n$, the parameterization $(\omega+\overline{\mu}-k)$-Coloring is strictly harder than $(n-k)$-Coloring. The paper nonetheless proves that it is solvable in randomized
$$
O^*(2^{6k})
$$
time.

The framework is complemented by negative results. There is no fixed-parameter tractable algorithm for $(\omega-k)$-Coloring unless $P=NP$, and $(\overline{\mu}-k)$-Coloring is W[1]-hard. The paper therefore identifies a narrow tractability frontier: saving colors below $n$ or below the combined guarantee $\omega+\overline{\mu}$ is fixed-parameter tractable via co-triangle packing, but parameterizing below $\omega$ alone or below $\overline{\mu}$ alone is unlikely to be fixed-parameter tractable [2509.12347].

## 4. Edge-coloring below Shannon’s bound

In multigraphs, below-guarantee coloring concerns the edge-chromatic index rather than the vertex chromatic number. "Beyond the Shannon's Bound" studies undirected multigraphs $G=(V,E)$ of maximum degree $\Delta$ and asks how large a subgraph can be edge-colored with only $\Delta$ colors [1309.6069].

Shannon’s theorem gives the worst-case upper bound
$$
\chi'(G)\le \left\lfloor \frac{3\Delta}{2}\right\rfloor.
$$
Selecting the $\Delta$ largest color classes from a Shannon coloring yields the baseline ratio
$$
y_\Delta(G)\ge \frac{\Delta}{\left\lfloor \frac{3\Delta}{2}\right\rfloor}.
$$
The paper improves this baseline by one unit in the denominator, proving that, except for explicit dense three-vertex obstructions, there exists a $\Delta$-edge-colorable subgraph with at least
$$
\frac{\Delta}{\left\lfloor \frac{3\Delta}{2}\right\rfloor-1}\,|E|
$$
edges.

The exceptional configurations are completely characterized. For even $\Delta$, the obstruction is $(\Delta/2)K_3$, the $3$-vertex complete multigraph in which each pair has multiplicity $\Delta/2$. For odd $\Delta$, the obstruction is $((\Delta-1)/2)K_3+e$, a triangle where one edge has one additional parallel copy. These are exactly the dense triangle gadgets that force equality in Shannon-type behavior and prevent improvement beyond the baseline fraction. For $\Delta=3$, earlier work of Kamiński and Kowalik gives a $3$-edge-colorable subgraph of size at least
$$
\frac{7}{9}|E|
$$
unless the graph has a component isomorphic to $K_3+e$; the present paper extends the same below-guarantee phenomenon to all $\Delta\ge 4$ [1309.6069].

The proof is constructive and algorithmic. A partial $\Delta$-coloring $T$ is evaluated by a potential function $Y(G,T)$ that orders colorings lexicographically according to the number of colored edges and the distribution of free components. Local recolorings, called elementary moves, uncolor some edges in the closed neighborhood of a free component and color an equal number of edges inside it. A global charging method assigns one unit of charge from each colored edge to the nontrivial free components that control its endpoints. Lower bounds on the charge received by each free component imply a lower bound on the total number of colored edges. The argument is reinforced by collapsing $k$-collapsible $3$-vertex subgraphs, which reduces local density while preserving the relevant three-vertex thresholds. The resulting algorithm runs in polynomial time; for fixed $\Delta$, the paper gives the bound
$$
O(m^2 n \Delta/2+1),
$$
where $n=|V|$ and $m=|E|$.

These structural results feed directly into approximation algorithms for Maximum $k$-Edge-Colorable Subgraph. The paper derives a
$$
\frac{2k+2}{3k+2}
$$
approximation for every even $k\ge 4$ and a
$$
\frac{2k+1}{3k}
$$
approximation for every odd $k\ge 5$ [1309.6069]. In this multigraph setting, below-guarantee coloring is thus both a structural theorem and an approximation framework: the task is not to color the whole graph with fewer than the universal guarantee, but to recover a provably large colorable subgraph beyond what the universal guarantee alone would imply.

## 5. Fewer colors for exact uniform sampling

A further usage of the same theme appears in exact sampling. "Fewer colors for perfect simulation of proper colorings" studies the problem of sampling exactly uniformly from the set $\Omega(G,k)$ of proper $k$-colorings of a graph $G$ with maximum degree $\Delta$ [2004.08716]. In this context, “below guarantee” refers to reducing the number of colors required by general-purpose perfect-sampling algorithms.

The paper gives a randomized perfect simulator based on the randomness recycler protocol. For every graph with $\Delta\ge 2$, if
$$
k>2.27(\Delta-1),
$$
then the algorithm outputs an exact uniform sample from $\Omega(G,k)$ in
$$
O(n\ln n)
$$
expected steps, where $n=|V|$. This improves the earlier perfect-sampling guarantee of Bhandari and Chakraborty, which required $k>3\Delta$.

The protocol maintains a set $S\subseteq V$ of currently colored vertices and a proper partial coloring $\sigma:S\to[k]$, initially with $S=\varnothing$. Its invariant is that $\sigma$ is uniformly distributed over the proper colorings of the induced subgraph $G[S]$. At each step, an uncolored vertex $v$ is chosen and a proposal color $c\in[k]$ is sampled uniformly. If no colored neighbor of $v$ has color $c$, the proposal is accepted. If conflicts occur, the algorithm defines the conflict set
$$
T=\{u\in N(v)\cap S:\sigma(u)=c\}
$$
and performs a local recycle surgery: $v$ is colored with $c$, the vertices in $T$ are uncolored, and the consumed random bits are kept live rather than discarded. The central claim is that this update is measure-preserving, so the uniformity invariant survives every step.

The analysis is drift-based. Let $\Phi=|S|$ be the number of colored vertices. The raw change after one step is $1-|T|$, so the goal is to bound the expected size of the conflict set. A representative sufficient inequality is
$$
1-\sum_{i=1}^{\Delta-1}\frac{1}{k-i}>0,
$$
and the recycler-specific analysis strengthens this with localized dependency bounds to obtain the linear threshold $k>2.27(\Delta-1)$. A multiplicative-drift argument then yields the $O(n\ln n)$ expected runtime.

This exact-sampling interpretation differs from existential vertex-coloring and fixed-parameter coloring, but the common principle remains the same: a previously accepted color threshold serves as the guarantee, and the contribution lies in proving that fewer colors suffice while preserving a stringent notion of correctness. Here the correctness requirement is exact uniformity rather than merely the existence of a proper coloring [2004.08716].

## 6. Structural themes, limitations, and open directions

Across these formulations, below-guarantee graph coloring is driven by the interaction between baseline bounds and explicit obstructions. In the Borodin–Kostochka setting, the obstruction is a large clique or, in restricted hereditary classes, a specific induced configuration such as $C_5^+$. In co-triangle-packing algorithms, the obstruction is the failure to find enough disjoint co-triangles, which is then converted into a small modulator. In Shannon-type edge-coloring, the obstruction is a dense $3$-vertex multigraph. In perfect sampling, the obstruction is not combinatorial in the same sense; it is the point at which recycling no longer guarantees positive drift under a given palette size. This suggests a broad unifying viewpoint: below-guarantee results typically succeed by isolating a minimal local structure that saturates the baseline guarantee, then proving that all other instances admit either recoloring, packing, or algebraic compression.

The limitations are equally structural. The $(P_6,C_4)$-free Borodin–Kostochka problem remains open in general; the available reduction shows that it would suffice to settle the cases $\omega(G)=7$ and $\omega(G)=8$ [2405.18455]. The co-triangle framework yields randomized algorithms, but removing randomization and obtaining deterministic fixed-parameter algorithms of comparable runtime is open; so is improving the exponents, in particular reaching $O^*(2^k)$ for Dual Coloring and improving the $O^*(2^{6k})$ bound for $(\omega+\overline{\mu}-k)$-Coloring [2509.12347]. In edge-coloring, the paper identifies the exact dense triangle obstructions but also asks what the next bottleneck beyond these obstructions should be, and to what extent stronger local sparsity assumptions on three vertices yield better fractions [1309.6069]. In perfect simulation, pushing the threshold from $2.27(\Delta-1)$ toward $2(\Delta-1)$ or even toward $\Delta+1$ is explicitly left open [2004.08716].

Taken together, these works show that below-guarantee graph coloring is not a single theorem but a research program spanning structural graph theory, fixed-parameter algorithms, multigraph edge-coloring, approximation, and exact randomized algorithms. Its common methodology is to sharpen a universal bound by one of three means: excluding a forbidden local configuration, exploiting a packing-or-modulator dichotomy, or maintaining a measure-preserving process whose drift becomes positive below the previous threshold.

Source: https://www.emergentmind.com/topics/below-guarantee-graph-coloring