Ramsey-Good Graphs: Exact Chromatic Bounds

• Ramsey-Good Graphs are defined by the equality R(G,H) = (|G|-1)(χ(H)-1)+σ(H) when G is connected, showing that chromatic constraints solely determine the Ramsey number.
• They encompass classical examples like trees versus cliques and extend to families analyzed via expansion, bandwidth limits, and probabilistic methods.
• Recent studies refine thresholds, explore chromatic surplus effects, and develop random graph analogues, highlighting new exact phenomena and open problems.

In graph Ramsey theory, “Ramsey-good” most commonly refers to exact attainment of Burr’s lower bound for an off-diagonal Ramsey number: if $G$ is connected, then $G$ is called $H$-good when

$R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$

where $\chi(H)$ is the chromatic number of $H$ and $\sigma(H)$ is the size of the smallest color class in a proper $\chi(H)$-coloring of $H$ (Pokrovskiy et al., 2015). Equivalent formulations in the literature swap the roles of the two graphs and write that a connected graph $H$ is $G$0-good if

$G$1

where $G$2 is the chromatic surplus, i.e. the minimum size of a color class in a proper $G$3-coloring of $G$4 (He et al., 7 Jun 2025). This notion, introduced by Burr and Erdős and developed across exact, asymptotic, multipartite, random-host, and coloring-based variants, isolates graph families whose Ramsey numbers are governed by the chromatic obstruction alone rather than by additional extremal structure (Pokrovskiy et al., 2015).

1. Classical definition and the Burr lower bound

The standard framework begins with the Ramsey number $G$5, the least $G$6 such that every red-blue coloring of the edges of $G$7 contains a red copy of $G$8 or a blue copy of $G$9 (Pokrovskiy et al., 2015). If $H$0 is connected, Burr’s lower bound states that

$H$1

where $H$2 is the smallest color-class size among proper $H$3-colorings of $H$4 (Pokrovskiy et al., 2015). The usual extremal construction is the disjoint union of $H$5 red cliques of size $H$6 together with one further red clique of size $H$7, with all cross-edges blue; this avoids a red connected copy of $H$8 and also avoids a blue copy of $H$9 (Pokrovskiy et al., 2015).

A connected graph $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$0 is $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$1-good if equality holds in this lower bound (Pokrovskiy et al., 2015). The equivalent formulation used in several more papers fixes the first graph and writes, for connected $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$2,

$R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$3

with equality defining $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$4 as $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$5-good (He et al., 7 Jun 2025). Here $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$6, called the chromatic surplus, is the minimum size of a color class in a proper $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$7-coloring of $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$8 (He et al., 7 Jun 2025). The two formulations encode the same phenomenon: the Ramsey number is as small as the chromatic lower bound permits.

The special case $R(G,H)=(|G|-1)(\chi(H)-1)+\sigma(H),$9 recovers the older language of $\chi(H)$0-goodness. Since $\chi(H)$1 and $\chi(H)$2, the formula becomes

$\chi(H)$3

and a connected graph $\chi(H)$4 satisfying this is $\chi(H)$5-good (Fox et al., 2021). This clique-versus-sparse-graph regime historically supplied the first major examples of Ramsey-goodness.

2. Foundational examples and asymptotic general theorems

The model theorem is Chvátal’s result that every tree is $\chi(H)$6-good (Hu et al., 2022). In the language of paths and trees, the theory was sharpened substantially by Pokrovskiy and Sudakov, who proved that

$\chi(H)$7

so sufficiently long paths are $\chi(H)$8-good for every fixed graph $\chi(H)$9 (Pokrovskiy et al., 2015). Their proof proceeds through the stronger statement

$H$0

under the condition $H$1, where $H$2 are the part sizes of a complete $H$3-partite graph (Pokrovskiy et al., 2015). This yields a uniform linear threshold in $H$4, and for balanced $H$5 the same argument gives goodness already for $H$6 (Pokrovskiy et al., 2015).

The same theme extends from paths to bounded-degree trees. Balla, Pokrovskiy, and Sudakov proved that for all $H$7 and $H$8 there exists a constant $H$9 such that any tree $\sigma(H)$0 with $\sigma(H)$1 is $\sigma(H)$2-good whenever

$\sigma(H)$3

and $\sigma(H)$4 (Balla et al., 2016). They also established a stronger many-leaves theorem: if $\sigma(H)$5 has $\sigma(H)$6 leaves and maximum degree at most $\sigma(H)$7, and

$\sigma(H)$8

then $\sigma(H)$9 is $\chi(H)$0-good (Balla et al., 2016). The proof combines a tree dichotomy—many leaves or many disjoint bare paths—with expansion, generalized Haxell-type embedding, and linked systems for long path insertion (Balla et al., 2016).

A different asymptotic direction concerns graphs $\chi(H)$1 and $\chi(H)$2 of the same order. For bounded-degree $\chi(H)$3 with $\chi(H)$4, it was shown that

$\chi(H)$5

so $\chi(H)$6 is asymptotically $\chi(H)$7-good under bounded maximum degree and a small-independence condition (Pei et al., 2014). This is not exact eventual equality, but it identifies a broad same-order regime in which the Burr lower bound is asymptotically correct (Pei et al., 2014).

These results fit into a more structural picture developed by Allen, Brightwell, and Skokan. They showed that bounded degree alone does not force always-good behavior, but bounded degree together with sufficiently small bandwidth does: if $\chi(H)$8 is connected, $\chi(H)$9, and $H$0, then for every fixed $H$1,

$H$2

for all sufficiently large $H$3 in the class (Allen et al., 2010). They also proved the same conclusion without any maximum-degree restriction when the bandwidth is at most $H$4 (Allen et al., 2010). This identifies expansion, rather than bounded degree itself, as the core obstruction to goodness (Allen et al., 2010).

3. Exact families beyond the classical tree-versus-clique setting

A major recent direction studies specific sparse target families against non-complete fixed graphs. One example is the generalized fan $H$5, formed by joining one hub to $H$6 disjoint copies of a fixed connected graph $H$7. Let $H$8 and $H$9. It was proved that $H$0 is $H$1-good if either

$H$2

or

$H$3

improving an earlier Chung–Lin threshold with constant $H$4 in

$H$5

down first to $H$6, then $H$7, and finally to the $H$8-scale in $H$9 (Zhang et al., 2023). In the special case $G$00, so $G$01 is the fan $G$02, the paper gives the sharper bound

$G$03

with $G$04 (Zhang et al., 2023).

Books form another central family. For the $G$05-book $G$06, consisting of a spine $G$07 and $G$08 pages adjacent to the spine and independent from one another, Fox, He, and Wigderson proved that

$G$09

so sufficiently large books are $G$10-good (Fox et al., 2021). This avoids the regularity method and replaces earlier tower-type thresholds with an explicit doubly exponential bound (Fox et al., 2021). The same paper also proves an exact multipartite-versus-book result: if $G$11 with $G$12, $G$13, and $G$14, then

$G$15

for sufficiently small explicit $G$16, with

$G$17

available from their proof (Fox et al., 2021).

A separate line concerns disjoint unions of cliques. For trees $G$18, it was proved that

$G$19

so every tree is $G$20-good (Hu et al., 2022). From this, together with a leaf-deletion criterion, one obtains

$G$21

so every tree is $G$22-good (Hu et al., 2022). The same paper extends goodness from connected components to disconnected unions: if every component of a disconnected graph $G$23 is $G$24-good, then $G$25 is $G$26-good in the corresponding Gould–Jacobson sense (Hu et al., 2022).

Goodness for complete multipartite graphs with one large part has also been classified asymptotically. For

$G$27

where $G$28 parts have size $G$29 and one part has size $G$30, and for fixed $G$31 with $G$32, $G$33, $G$34, the paper “Ramsey goodness of complete multipartite graphs with one large part” gives a full asymptotic if-and-only-if classification for $G$35: $G$36 is $G$37-good for large $G$38 exactly when

$G$39

for every tree $G$40 with

$G$41

where $G$42 is the smallest non-divisor of $G$43 (Mi et al., 26 May 2026). The dependence on $G$44 shows that divisibility, not merely chromatic data, can control multipartite goodness (Mi et al., 26 May 2026).

4. Chromatic surplus, non-clique fixed graphs, and new exact phenomena

Most earlier star-like goodness results assumed the fixed graph had chromatic surplus $G$45. A recent advance breaks this restriction by studying the Hajós graph $G$46, which satisfies

$G$47

and therefore induces the Burr bound

$G$48

for connected $G$49 with $G$50 (He et al., 7 Jun 2025). For the star $G$51, the exact Ramsey number is

$G$52

so $G$53 is $G$54-good if and only if $G$55 is even (He et al., 7 Jun 2025). The lower bound in the odd case comes from the explicit obstruction

$G$56

which has $G$57 vertices, contains no $G$58, and whose complement has maximum degree $G$59, hence contains no $G$60 (He et al., 7 Jun 2025). This gives a genuine parity obstruction to goodness in the surplus-$G$61 setting (He et al., 7 Jun 2025).

The same paper proves that the fan $G$62 is $G$63-good for all $G$64: $G$65 Its proof uses the external input

$G$66

a matching analysis in the large-$G$67 case, and a detailed wheel-based decomposition in the complementary case (He et al., 7 Jun 2025). The threshold $G$68 is stated explicitly as not best possible (He et al., 7 Jun 2025).

Odd cycles against complete bipartite graphs provide another exact recent family. For odd $G$69, the paper “A study of two Ramsey numbers involving odd cycles” proves that

$G$70

whenever

$G$71

showing that $G$72 is $G$73-good in that range (Gupta, 22 Apr 2025). Since $G$74, $G$75, and $G$76, the Burr formula indeed predicts $G$77 (Gupta, 22 Apr 2025). The upper bound rests on common-neighborhood control in $G$78-free graphs, specifically

$G$79

for all $G$80, plus cycle-richness lemmas for 2-connected non-bipartite graphs (Gupta, 22 Apr 2025).

These results collectively show that once one moves beyond fixed graphs with chromatic surplus $G$81, goodness can remain exact but the obstruction landscape changes substantially. This suggests that surplus $G$82 is not a minor perturbation of the classical theory but a distinct regime with its own structural exceptions (He et al., 7 Jun 2025).

5. Random-host analogues and sparse ambient graphs

A different extension asks not for the complete graph $G$83 to force the Ramsey alternative, but for a sparse random host $G$84. In this setting, the relevant statement is that the random host behaves like a Ramsey-good ambient graph up to a natural additive correction.

For paths versus cliques, Moreira showed that

$G$85

and also that

$G$86

(Moreira, 2019). Since the deterministic good threshold is $G$87 for large $G$88, these results show that sparse random graphs on essentially the same number of vertices already force the same pair once $G$89 is above the appropriate threshold (Moreira, 2019). The paper also proves matching negative statements, including that $G$90 is insufficient in the near-critical regime (Moreira, 2019).

For bounded-degree trees, Bucić, Letzter, Sudakov, and Tran proved a random analogue of Chvátal’s theorem. Writing $G$91 for the family of trees with $G$92 edges and maximum degree at most $G$93, they showed that for each $G$94 there exist constants $G$95 such that if

$G$96

then

$G$97

with high probability (Araújo et al., 2020). Since Chvátal’s deterministic theorem gives

$G$98

for every tree $G$99 with $H$00 edges, this is a near-exact random-host analogue with the additive $H$01 term representing the correct sparse correction scale (Araújo et al., 2020). The proof develops a stability theorem showing that any bad coloring must resemble the extremal $H$02-partite obstruction up to an exceptional set of size $H$03 (Araújo et al., 2020).

These random-host results do not redefine goodness, but they transfer the goodness heuristic from complete hosts to sparse pseudorandom ambient graphs. A plausible implication is that the classical Burr–Erdős framework has a robust probabilistic counterpart whenever the deterministic extremal colorings are sufficiently stable (Moreira, 2019).

6. Alternative meanings of “good” and adjacent Ramsey notions

The term “good” is not uniform across Ramsey theory. Several papers use it in related but formally distinct senses.

In mixed Ramsey theory, an edge-coloring of $H$04 is $H$05-good if it contains no monochromatic copy of $H$06 and no rainbow copy of $H$07. The set of achievable numbers of colors in such colorings is the mixed-Ramsey spectrum

$H$08

Axenovich and Iverson proved that if $H$09 is not a star and $H$10, then $H$11 is an interval for every $H$12 (Axenovich et al., 2010). They also exhibited the exceptional finite example

$H$13

showing that interval behavior can fail once $H$14 has a pendent edge (Axenovich et al., 2010). Here “good” describes a coloring, not a graph attaining Burr’s lower bound (Axenovich et al., 2010).

A different chromatic-host notion appears in “Good Graph Hunting.” A tuple $H$15 is called good if every $H$16-edge-coloring of an $H$17-chromatic graph contains a monochromatic $H$18 in color $H$19 for some $H$20 (Garrison, 2015). In this framework, a graph $H$21 is $H$22-good if $H$23 is good, and good if it is $H$24-good for every $H$25 (Garrison, 2015). The paper proves that stars are good, that $H$26 is $H$27-good for $H$28, that $H$29 is good, and that $H$30, $H$31, and $H$32 are $H$33-good (Garrison, 2015). This is a chromatic-host strengthening of ordinary Ramsey theorems, not Burr-type goodness (Garrison, 2015).

A third usage occurs in extremal graph constructions. In “Doubly Saturated Ramsey Graphs,” a graph $H$34 is called $H$35-good if it contains neither a clique of size $H$36 nor an independent set of size $H$37, equivalently

$H$38

(Przybocki et al., 23 Apr 2026). The paper studies doubly saturated $H$39-good graphs, meaning that adding any non-edge or deleting any edge destroys the property, and proves that for all $H$40 there exists a doubly saturated $H$41-good graph on $H$42 vertices (Przybocki et al., 23 Apr 2026). This is a standard extremal-Ramsey usage, but it is unrelated to Burr’s lower bound (Przybocki et al., 23 Apr 2026).

Related but distinct again are Ramsey-minimal graphs, such as minimal $H$43-Ramsey graphs $H$44 with $H$45 but every proper subgraph failing that property (Bikov, 2016). These are not called Ramsey-good in the Burr sense, yet they address a nearby extremal boundary of Ramsey forcing (Bikov, 2016).

Because these meanings coexist, precision about the underlying definition is essential in the literature.

7. Obstructions, sparse Ramsey density, algebraic limits, and open directions

Not every sparse-looking graph family is Ramsey-good. One obstruction comes from expansion. Allen, Brightwell, and Skokan emphasize that bounded degree alone does not imply always-goodness: Graham, Rödl, and Ruciński constructed bounded-degree expanders showing that the diagonal linear constant $H$46 must satisfy

$H$47

for some $H$48, while bounded degree together with sublinear bandwidth restores exact off-diagonal goodness (Allen et al., 2010). Their paper also shows that Burr’s diagonal conjecture fails even for path powers $H$49 (Allen et al., 2010).

Another measure of sparse Ramsey structure is the Ramsey density

$H$50

For even cycles, it was shown that

$H$51

and for complete bipartite graphs $H$52 with $H$53,

$H$54

(Mütze et al., 2011). For paths,

$H$55

(Mütze et al., 2011). These are not goodness results in the Burr sense, but they show that sparse Ramsey hosts can be much sparser than complete graphs, especially for bipartite targets (Mütze et al., 2011).

A further negative perspective comes from algebraic constructions. If an algebraic graph of complexity $H$56 on $H$57 vertices is defined by zero-patterns of boundedly many bounded-degree polynomials in bounded dimension, then it must contain either a clique or an independent set of size at least

$H$58

(Sudakov et al., 2021). Thus bounded-complexity algebraic graphs have polynomial-size homogeneous sets, whereas the Erdős random-graph benchmark has largest clique and independent set of order $H$59 (Sudakov et al., 2021). This shows that any algebraic family with genuinely strong Ramsey-type pseudorandomness must let at least one of the parameters $H$60, $H$61, or $H$62 grow (Sudakov et al., 2021).

Across the recent literature, several open problems recur. The precise smallest $H$63 for which $H$64 is $H$65-good remains open below the current threshold $H$66 (He et al., 7 Jun 2025). For books, the exact Ramsey-goodness threshold between the lower bound $H$67 and the upper bound $H$68 is open (Fox et al., 2021). For bounded-degree trees, the conjectured linear theorem

$H$69

would remove the $H$70 factor from the current best general result (Balla et al., 2016). For mixed Ramsey spectra, it remains open whether there exist graph pairs $H$71 for which non-interval behavior persists for arbitrarily large $H$72 (Axenovich et al., 2010).

Taken together, these works show that Ramsey-goodness is neither a single theorem nor a single class, but a broad organizing principle. In its classical Burr–Erdős form, it identifies graph families whose Ramsey numbers are controlled exactly by chromatic data; in modern developments, it interacts with chromatic surplus, multipartite structure, expansion, random hosts, and several adjacent notions of “goodness” used elsewhere in Ramsey theory (Pokrovskiy et al., 2015).