Difference-of-DP in Graph DP-Coloring

• Difference-of-DP is the gap between the online DP-chromatic number and its offline counterpart, highlighting how adversarial token play forces higher color requirements.
• The study contrasts fractional DP-coloring with classical list coloring by demonstrating that the gap between the fractional DP-chromatic number and the conventional fractional chromatic number can be made arbitrarily large.
• Constructive techniques using intricate gadget constructions and token-counting arguments establish the unbounded nature of DDP, motivating further research in online combinatorics and graph theory.

The term "Difference-of-DP" (abbreviated as DDP) primarily refers to the gaps between chromatic numbers defined in the context of DP-coloring (correspondence coloring) for graphs. In modern combinatorics and graph theory, the difference between the online and offline variants of the DP-chromatic number—also called DDP(G)—has emerged as a significant complexity parameter, with recent work establishing that this gap can grow arbitrarily large. Closely related differences also appear in the study of the fractional DP-chromatic number versus its classical counterparts. This article systematically surveys the mathematical definitions, key theorems, constructive techniques, and the implications of such "difference-of-DP" results, focusing on both the online/offline gap and the fractional/integer gap in DP-coloring.

1. Foundational Definitions of DP-Coloring

An “f-fold DP-cover” H=(L,H) of a finite graph G=(V,E) is constructed by assigning to each vertex v∈V a finite set L(v) of size f(v), with these forming cliques in H, and—for each edge uv∈E—adding an arbitrary matching between the sets L(u), L(v). A set S⊆V(H) is quasi-independent when it contains no cross-edges (edges not contained within any L(v)). A DP-coloring for a k-fold cover (all f(v)=k) is a quasi-independent transversal with |S∩L(v)|=1 for all v.

The offline DP-chromatic number χDP(G) is defined as the least k such that every k-fold DP-cover of G admits a DP-coloring. In the online DP-coloring, a sequential game (the "DP-painting game") is played between Lister and Painter, where covers are revealed incrementally and the Painter must irrevocably color vertices as covers are exposed. The minimum k granting a winning strategy to Painter for every possible play yields the online DP-chromatic number χDP^online(G).

The difference-of-DP, DDP(G), is defined as

$$DDP(G) := \chi^{\text{online}}_{DP}(G) - \chi_{DP}(G).$$

A related difference arises in the fractional regime. The fractional DP-chromatic number χ^*_{DP}(G) generalizes to allow (η,k)-colorings (with partial color sets per vertex) and is contrasted with the classical fractional chromatic number χ^*(G).

2. Unbounded Gap Between Online and Offline DP-Chromatic Numbers

The existence of graphs with arbitrarily large DDP(G) is established in "Separating the online and offline DP-chromatic numbers" (Bradshaw, 2022). For every integer t≥1, one constructs a finite graph G_t such that

$$\chi^{\text{online}}_{DP}(G_t) - \chi_{DP}(G_t) \geq t,$$

demonstrating that DDP(G) is unbounded in general families. The foundational gadget H_t, parameterized by a large integer k=2^{8t^3}, and subsequent assembly via layers of cliques and independent sets yield graphs for which even weaker online parameters (e.g., the list-painting number χP(G_t)) exceed χDP(G_t) by at least t.

The construction leverages the interplay between color set reductions in the online process (where tokens or colors can be systematically exhausted before coverage is possible) and the global flexibility of the offline process (where the entire k-fold cover is known in advance). The results refute the possibility of a general or polynomial bound on DDP(G) in terms of χDP(G) or other basic graph invariants.

3. Fractional DP-Coloring Gap: χ^*_{DP}(G) Versus χ^*(G)

In "Fractional DP-Colorings of Sparse Graphs" (Bernshteyn et al., 2018), Bernshteyn, Kostochka, and Zhu introduce and analyze the fractional analogue of DP-coloring. While for classical list coloring, the fractional list-chromatic number always equals the fractional chromatic number (Alon–Tuza–Voigt theorem), this equality fails for fractional DP-colorings:

$$\chi^*_{\ell}(G) = \chi^*(G) \quad\text{but}\quad \chi^*_{DP}(G) \geq \chi^*(G).$$

Moreover, the gap Δ(G) = χ^*_{DP}(G) − χ^*(G) can be made arbitrarily large via explicit constructions, such as for nearly complete bipartite graphs K_{d,N} (N≫d), where χ^*(K_{d,N})=2, but χ^*{DP}(K{d,N}) grows at least as fast as d/(2 ln d).

Additionally, there exist high-girth, high-chromatic-number graph sequences for which Δ(G) scales as Θ(d/ln d) for average degree d, and even for simple families like even cycles or bipartite graphs with more edges than vertices, χ^*_{DP}(G) exceeds 2 strictly. This separation refutes any analogue of the Alon–Tuza–Voigt theorem for DP-colorings.

4. Techniques and Proof Outlines for DDP Constructions

The existence proofs for large DDP(G) utilize intricate gadget constructions and token-counting arguments. For online/offline coloring, the adversarial role of Lister in the DP-painting game is harnessed to force Painter into dead-ends within certain copies of subgraphs after an exponential number of rounds; combinatorial pigeonhole principles certify uncolorability in the online regime while the offline regime allows global coordination, preserving colorability.

In the fractional setting, probabilistic method and union bound techniques over random covers provide asymptotic lower bounds; stirling-type entropy estimates are combined with explicit greedy and intersection arguments.

In both areas, degeneracy and average degree serve as key structural parameters controlling the extremal behavior of χ^*_{DP}(G), χDP(G), and their differences.

5. Asymptotics, Bounds, and General Implications

Several scaling regimes and inequalities are established:

• χDP(G) = Ω(d/log d) for graphs of average degree d.
• For any G of degeneracy d, χDP^online(G) ≤ d+1 ≤ 2^{(1+o(1)) χDP(G)}, with refined upper bounds χDP^online(G) ≤ (2+o(1)) χDP(G)·log χDP(G).
• For the complete bipartite graph K_{n,n}, χDP^online(K_{n,n}) = O(log n), whereas χDP(K_{n,n}) = Ω(n/log n).

These results underscore that the online–offline gap for DP-coloring is not bounded by any function of classical chromatic invariants, nor does it share the behavior seen in list coloring or ordinary fractional chromatic numbers. The fractional DP gap, being unbounded, further distinguishes the DP framework from traditional colorings and list formulations.

6. Illustrative Examples and Quantitative Behavior

For small t, explicit parameterizations yield the following (with non-optimal but illustrative constants):

t	k (=2^{8t^3})	Offline χDP(G_t)	Online χDP^online(G_t)	DDP(G_t)
1	256	≤255	≥257	≥2
2	2^{64}	≤k-1	≥k+1	≥2

While the constructions are not optimized for parameter size, they demonstrate concretely that DDP(G) is scalable to any prescribed value. In practice, the constants can be prohibitively large, but the underlying combinatorial phenomena hold universally.

7. Broader Impact and Research Directions

Difference-of-DP phenomena introduce new structural complexities into the DP-coloring landscape, motivating further study of online combinatorics, generalized colorability, and fractional structures. They also challenge existing intuition from classical and list coloring, indicating substantial methodological differences in DP-based generalizations.

Open directions include refinement of upper and lower bounds in natural graph classes, better understanding of the behavior in random and geometric graphs, and elucidation of algorithmic complexity regarding DDP parameters. The gap’s sensitivity to covering rules and adversarial manipulations remains a promising ground for future research in combinatorics and theoretical computer science (Bernshteyn et al., 2018, Bradshaw, 2022).