---
title: DP Color Function in Graph Coloring
url: https://www.emergentmind.com/topics/dp-color-function
type: topic
---

# DP Color Function in Graph Coloring

The DP color function, denoted \(P_{DP}(G,m)\), is the correspondence-coloring analogue of the chromatic polynomial for a graph \(G\). For a fixed positive integer \(m\), it records the minimum number of DP-colorings over all \(m\)-fold covers of \(G\), so it is a worst-case counting invariant rather than a direct enumerator of ordinary colorings. Introduced as a counting counterpart to DP-coloring, or correspondence coloring, it refines the comparison between ordinary coloring, list coloring, and edge-dependent correspondence constraints; in particular, it can agree with the chromatic polynomial on some graph classes and remain strictly smaller on others even for arbitrarily large \(m\) [1904.07697][2012.12897].

## 1. Definition and formal framework

DP-coloring generalizes list coloring by allowing the identification of colors to vary from edge to edge. In the standard graph-theoretic formulation, a cover of a graph \(G\) is a pair \(H=(L,H)\) in which the sets \(L(u)\) for \(u\in V(G)\) partition \(V(H)\), each \(H[L(u)]\) is a clique, edges of \(H\) between distinct parts occur only along edges of \(G\), and for each \(uv\in E(G)\) the edges between \(L(u)\) and \(L(v)\) form a matching. If \(|L(u)|=m\) for every \(u\), the cover is \(m\)-fold. An \(H\)-coloring of \(G\) is an independent set in \(H\) of size \(|V(G)|\), equivalently a choice of one vertex from each part \(L(u)\) with no conflicts [1904.07697].

For a fixed \(m\)-fold cover \(H\), the number of \(H\)-colorings is denoted \(P_{DP}(G,H)\). The DP color function is then
\[
P_{DP}(G,m)=\min\{P_{DP}(G,H): H \text{ is an } m\text{-fold cover of }G\}.
\]
This definition is directly parallel to the chromatic polynomial \(P(G,m)\), which counts proper \(m\)-colorings, and to the list color function \(P_\ell(G,m)\), which minimizes over all \(m\)-assignments. The basic comparison chain is
\[
P_{DP}(G,m)\le P_\ell(G,m)\le P(G,m),
\]
because list assignments form a restricted class of DP-covers, and the ordinary coloring situation is recovered from a canonical cover [1904.07697][2012.12897].

A full \(m\)-fold cover with a canonical labeling behaves exactly like ordinary \(m\)-coloring: if the vertices of the cover can be labeled as \((u,j)\) with \(L(u)=\{(u,1),\dots,(u,m)\}\) and the cross-edges between \(L(u)\) and \(L(v)\) are precisely the parallel edges \((u,j)(v,j)\), then
\[
P_{DP}(G,H)=P(G,m).
\]
This implication is one of the main structural bridges between the DP color function and the chromatic polynomial [2210.06000].

## 2. Comparison with chromatic polynomials and eventual behavior

The central structural problem is to determine when \(P_{DP}(G,m)\) eventually coincides with \(P(G,m)\), and when it remains strictly smaller. Two asymptotic classes formalize this distinction. A graph lies in \(DP_\approx\) if there exists \(M\) such that \(P_{DP}(G,m)=P(G,m)\) for all \(m\ge M\), and in \(DP_<\) if there exists \(M\) such that \(P_{DP}(G,m)<P(G,m)\) for all \(m\ge M\). It is not known whether there are graphs outside both classes [2203.07704].

The first general asymptotic gap estimate was
\[
P(G,m)-P_{DP}(G,m)=O(m^{n-2})
\]
for an \(n\)-vertex graph \(G\), and this was later sharpened to
\[
P(G,m)-P_{DP}(G,m)=O(m^{n-3})
\]
for every graph \(G\) on \(n\) vertices. The same work showed that for every graph \(G\), there exists \(N\) such that
\[
P_{DP}(K_1\vee G,m)=P(K_1\vee G,m)\qquad\text{for all }m\ge N,
\]
so adjoining one dominating vertex forces eventual agreement with the chromatic polynomial [1904.07697][2009.08242].

The negative direction is equally important. If a graph has even girth, then there exists \(N\) such that
\[
P_{DP}(G,m)<P(G,m)\qquad\text{for all }m\ge N.
\]
This was strengthened to a local criterion: if \(\ell(e)\), the length of a shortest cycle containing an edge \(e\), is even for some edge \(e\), then \(P_{DP}(G)<P(G)\) in the eventual sense. Conversely, the reverse implication fails with infinitely many counterexamples [1904.07697][2105.11081].

These phenomena explain why the DP color function is neither a trivial perturbation of the chromatic polynomial nor a simple list-color analogue. For list coloring, eventual equality with \(P(G,m)\) is much more common; for DP-coloring, even very sparse parity obstructions can force persistent deviation [2012.12897].

## 3. Exact formulas and parity-controlled classes

A substantial portion of the theory consists of exact formulas for specific graph families. For chordal graphs,
\[
P_{DP}(G,m)=P(G,m)\qquad\text{for all }m,
\]
so the DP color function coincides identically with the chromatic polynomial on that class. Trees are an immediate special case [1904.07697].

For cycles, parity already creates the basic dichotomy. If \(C_n\) is odd, then
\[
P_{DP}(C_n,m)=P(C_n,m)=(m-1)^n-(m-1).
\]
If \(C_n\) is even and \(m\ge 2\), then
\[
P_{DP}(C_n,m)=(m-1)^n-1 < (m-1)^n+(m-1)=P(C_n,m).
\]
Thus even cycles provide a permanent gap between the DP color function and the chromatic polynomial [2107.08154][2412.16790].

Unicyclic graphs admit a similar parity classification. If the unique cycle has odd length, then
\[
P_{DP}(G,m)=P(G,m)\qquad\text{for all }m.
\]
If the unique cycle has even length \(2k+2\), then for \(m\ge 2\),
\[
P_{DP}(G,m)=(m-1)^n-(m-1)^{n-2k-2},
\]
where \(n=|V(G)|\). The same parity principle persists in more complicated families built around a single cycle or around two cycles sharing an edge [1904.07697].

Theta graphs furnish the first family for which the DP color function was determined in full generality. A Theta graph is \(\Theta(l_1,l_2,l_3)\), formed by three internally disjoint paths between the same two endpoints. If \(l_1\) has parity different from both \(l_2\) and \(l_3\), then
\[
P_{DP}(G,m)=P(G,m)\qquad\text{for all }m\in\mathbb N.
\]
In the other parity configurations, explicit closed formulas were obtained, and the function is always eventually polynomial. For generalized Theta graphs \(\Theta(l_1,\dots,l_k)\), eventual equality with the chromatic polynomial occurs exactly when \(l_1\) has parity opposite to every other path length; otherwise \(P_{DP}(G,m)<P(G,m)\) for all sufficiently large \(m\). More generally, if \(G\) has a feedback vertex set of size one, then there exist \(N\) and a polynomial \(p(m)\) such that \(P_{DP}(G,m)=p(m)\) for all \(m\ge N\) [2012.12897].

A compact summary of representative classes is given below.

| Graph class | Behavior of \(P_{DP}(G,m)\) | Source |
|---|---|---|
| Chordal graphs | \(P_{DP}(G,m)=P(G,m)\) for all \(m\) | [1904.07697] |
| Odd cycles | \(P_{DP}(C_n,m)=P(C_n,m)\) | [2107.08154] |
| Even cycles | \(P_{DP}(C_n,m)=(m-1)^n-1\) for \(m\ge 2\) | [2107.08154] |
| Unicyclic graphs | Equality for odd cycle; explicit smaller formula for even cycle | [1904.07697] |
| Theta graphs | Exact parity-dependent formulas | [2012.12897] |
| Graphs with feedback vertex set size one | Eventually polynomial | [2012.12897] |

These results show that parity is not merely an artifact of cycle computations. It is a recurrent organizing principle for the DP color function, especially when the underlying graph has a near-tree structure [2012.12897].

## 4. Structural methods: deletion–contraction, canonical labelings, and extremal bounds

A major methodological advance was the introduction of a deletion–contraction relation for the DP color function. To make deletion–contraction compatible with contraction-generated parallel edges, the theory was extended to multigraphs, together with the dual DP color function
\[
P^*_{DP}(G,m)=\max_H P_{DP}(G,H),
\]
where the maximum is over full \(m\)-fold covers. If \(e\) is an edge of a multigraph \(G\), then the DP-coloring counts of suitable covers satisfy
\[
P_{DP}(G,H)\ge P_{DP}(G-e,H-e)-P_{DP}(G\cdot e,H\cdot e),
\]
and consequently
\[
P_{DP}(G-e,m)-P^*_{DP}(G\cdot e,m)\le P_{DP}(G,m).
\]
This is the DP analogue of the chromatic deletion–contraction formula, although it is generally an inequality rather than an identity [2107.08154].

Canonical labeling is closely tied to equality with the chromatic polynomial, but the converse is subtle. It is well known that if a full \(m\)-fold cover has a canonical labeling, then \(P_{DP}(G,H)=P(G,m)\). However, the converse fails in general: there are a \(3\)-fold cover \(\mathcal H_1\) of \(W_4\) with
\[
P_{DP}(W_4,\mathcal H_1)=P(W_4,3)=6
\]
and a \(4\)-fold cover \(\mathcal H_2\) of \(W_4\) with
\[
P_{DP}(W_4,\mathcal H_2)=P(W_4,4)=72,
\]
yet neither cover has a canonical labeling. By contrast, if \(G\) is unicyclic and \(m\ge 2\), or if \(G\) is a theta graph and \(m\ge 3\), then
\[
P_{DP}(G,\mathcal H)=P(G,m)\Longrightarrow \mathcal H \text{ has a canonical labeling}.
\]
Thus the converse holds for some low-cycle-rank families and fails already on \(W_4\) [2210.06000].

The DP color function also satisfies sharp extremal bounds. For a connected graph \(G\) on \(n\) vertices,
\[
P_{DP}(G,m)\le m(m-1)^{n-1},
\]
with equality for \(m\ge 2\) if and only if \(G\) is a tree. For a \(2\)-connected graph \(G\) on \(n\ge 3\) vertices,
\[
P_{DP}(G,m)\le 
\begin{cases}
(m-1)^n-(m-1), & \text{if }n\text{ is odd and }m\ge 3,\\[2mm]
(m-1)^n-1, & \text{if }n\text{ is even and }m\ge 3,
\end{cases}
\]
with equality exactly when \(G=C_n\). These results are DP analogues of classical chromatic-polynomial extremal theorems and rely on ear decompositions together with exact cycle formulas [2210.06000].

## 5. Graph operations and threshold phenomena

The DP color function is especially effective when coloring behavior is governed by graph operations. For joins with complete graphs, one threshold theorem states that if
\[
P_{DP}(K_p\vee G,m)=P(K_p\vee G,m)\quad\text{for all }m\ge N,
\]
then
\[
P_{DP}(K_{p+1}\vee G,s)=P(K_{p+1}\vee G,s)\quad\text{for all }s>N+1.
\]
A key special case is the cone-reduction lemma: if \(P_{DP}(G,m)=P(G,m)\), then
\[
P_{DP}(K_1\vee G,m+1)=P(K_1\vee G,m+1).
\]
For cycles this yields the exact threshold
\[
T_{DP}(K_p\vee C_n)=p+3
\]
for all \(p\in\mathbb N\) and \(n\ge 3\) [2104.12268].

Vertex-gluings and clique-gluings reveal both positive and negative analogies with chromatic-polynomial product formulas. If \(G\) is a vertex-gluing of \(G_1,\dots,G_n\), then
\[
P_{DP}(G,m)\le \frac{\prod_{i=1}^n P_{DP}(G_i,m)}{m^{\,n-1}}.
\]
For vertex-gluings of chordal graphs and cycles, equality holds:
\[
P_{DP}(G,m)=\frac{\prod_{i=1}^n P_{DP}(G_i,m)}{m^{n-1}}\qquad\text{for all }m.
\]
For \(K_p\)-gluings, the expected DP analogue of the chromatic-polynomial formula holds for edge-gluings (\(p=2\)) but fails for triangle-gluings (\(p=3\)); a relaxed canonical version remains valid for \(p\ge 3\) [2104.12268][2112.05316].

In Cartesian products, the DP color function becomes a threshold parameter. A general result gives
\[
\chi_{DP}(G\square K_{k,t})=\chi_{DP}(G)+k
\quad\text{whenever}\quad
t\ge \bigl(P_{DP}(G,\chi_{DP}(G)+k-1)\bigr)^k.
\]
Thus the smallest \(t\) forcing the upper bound in the Cartesian-product DP-coloring theorem is controlled not only by \(\chi_{DP}(G)\) but by the worst-case number of surviving DP-colorings at level \(\chi_{DP}(G)+k-1\). This suggests that the DP color function is an obstruction-size invariant for product colorability rather than merely a secondary counting parameter [2110.04700].

## 6. Extensions, monotonicity, and current frontiers

Two long-standing questions for the DP color function are whether eventual equality \(P_{DP}(G,m)=P(G,m)\) holds for large \(m\), and whether \(P_{DP}(G,m)\) is eventually polynomial. Theta graphs and generalized Theta graphs supplied exact answers in both directions, and graphs with feedback vertex set size one supplied a larger eventually polynomial family [2012.12897].

The DP color function is not chromatic-adherent. A function \(f\) is chromatic-adherent if \(f(G,a)=P(G,a)\) for some \(a\ge \chi(G)\) implies \(f(G,m)=P(G,m)\) for all \(m\ge a\). For the graphs \(\Theta(2,3,3,3,2)\) and \(\Theta(2,3,3,3,3,3,2,2)\),
\[
P_{DP}(G,3)=P(G,3),
\]
but there exists \(N\) such that
\[
P_{DP}(G,m)<P(G,m)\qquad\text{for all }m\ge N.
\]
This shows that agreement at one value of \(m\) need not persist [2110.04058].

At the same time, the normalized quantity \(P_{DP}(G,k)/k^n\) satisfies a universal monotonicity law. For every \(n\)-vertex graph \(G\) and every \(k\in\mathbb N\),
\[
\frac{P_{DP}(G,k+1)}{(k+1)^n}\ge \frac{P_{DP}(G,k)}{k^n}.
\]
This is the DP analogue of Dong’s shameful inequality and holds for all graphs and all \(k\), unlike the chromatic-polynomial version, which is only known in general for \(k\ge n-1\) [2412.16790].

The surrounding landscape is broader than the graph case. Hypergraph versions of the DP color function have now been introduced. For connected \(r\)-uniform hypergraphs, one has the upper bound
\[
P_{DP}(H,k)\le k^{\,n-(r-1)|E(H)|}(k^{r-1}-1)^{|E(H)|},
\]
with equality if and only if \(H\) is a hypertree. For linear \(r\)-uniform unicyclic hypergraphs, parity of the unique cycle again governs whether \(P_{DP}(H,k)\) equals \(P(H,k)\) or differs from it. A later hypergraph study established that for any linear and uniform hypergraph with even girth, there exists \(N\) such that
\[
P_{DP}(\mathcal H,k)<P(\mathcal H,k)\qquad\text{for all }k\ge N,
\]
while for \(\mathcal H\vee K_p\) with \(\mathcal H\) uniform, there exist \(p\) and \(N\) such that
\[
P_{DP}(\mathcal H\vee K_p,k)=P(\mathcal H\vee K_p,k)\qquad\text{for all }k\ge N.
\]
This suggests that the graph-theoretic dichotomy between even-cycle obstructions and clique-join stabilization has a genuine hypergraph analogue [2503.14879][2602.06747].

The current picture therefore combines exact solvability on several parity-controlled families, robust asymptotic comparison theorems, and persistent open classification problems. The DP color function sits below the list color function and the chromatic polynomial, but its behavior is governed by structural features—especially parity, local cycle geometry, canonicality of covers, and near-tree decompositions—that have no exact analogue in ordinary coloring.

Source: https://www.emergentmind.com/topics/dp-color-function