60-second version
Extremal graph theory asks how dense a graph can be while avoiding a forbidden pattern. A conjecture may sound local—ban one configuration, control a few degrees—but a large graph can distribute its complexity in ways that defeat a neat global description.
OpenAI reports separate bipartite constructions that disprove the compactness and degeneracy conjectures of Erdős and Simonovits, resolving Erdős problems 146 and 180. The lesson is not that all extremal structure is chaotic. It is that a few individually plausible restrictions do not always control the whole family.
Start with a forbidden pattern
Suppose a graph designer wants as many edges as possible while avoiding a particular subgraph. The first instinct is to organize the graph into a clean template: keep neighborhoods similar, or argue that large dense pieces must contain a common structure.
Extremal examples often exploit the gap between “each local view is simple” and “the global graph is simple.” A bipartite graph has no odd cycle, for example, but its two sides can still carry complicated degree patterns and overlapping neighborhoods. The counterexamples in this chapter use that room to make a family-level statement fail.
Definitions that matter
The extremal number of a forbidden graph (H) is the maximum number of edges in an (H)-free graph on (n) vertices. A compactness conjecture proposes that certain local or finite obstructions should control a global extremal phenomenon. A degeneracy condition limits how much complexity remains after repeatedly removing low-degree vertices.
A counterexample is a construction that satisfies the conjecture’s hypotheses but violates its conclusion. Because the claims concern asymptotic families, one graph is not enough; the construction must scale with (n) and keep the failure visible.
The research question
Do compactness and degeneracy principles correctly predict extremal behavior for the relevant graph families? If every local restriction looks weak, must the whole family still have a predictable density or structure?
The reported answer is no for both conjectures. The separate constructions show that two distinct intuitions—finite compactness and a degeneracy-based control—can each fail in extremal graph theory.
What was known before
Extremal graph theory has many successful compactness and regularity principles. Finite configurations often do control asymptotic densities, and degeneracy is a useful way to measure how graphs can be peeled apart. Those successes made the conjectures plausible.
The remaining cases were difficult because the hypotheses allowed a graph to be locally tame while maintaining a large-scale pattern across many parts. Standard averaging can see the density but miss how the forbidden structure is distributed.
Why previous approaches stalled
The construction has to balance several weak restrictions at once. If it becomes too dense, the forbidden pattern appears. If it is too sparse, it no longer challenges the conjectured extremal threshold. The counterexample therefore needs a family of graphs whose local neighborhoods are controlled just enough to satisfy the premise while their combined arrangement breaks the conclusion.
Bipartite constructions are useful because they eliminate some patterns automatically. That clean baseline leaves room to tune degrees and overlaps without spending the entire proof on avoiding odd cycles. The hard part is making the family-level density change persist as the graph grows.
The new result
The manuscript gives separate bipartite graph constructions that disprove the compactness conjecture of Erdős and Simonovits and a degeneracy conjecture of Erdős. The announcement identifies these as resolving Erdős problems 146 and 180.
The visual moves from one forbidden pattern to a family of weak restrictions and finally to a density shift. That transition is the point: a statement can be correct for each local piece while failing for the family-level extremal quantity.
Formal result and proof architecture
The constructions are parameterized graph families. One tracks the forbidden subgraph condition, the relevant local or degeneracy property, and the asymptotic edge density separately. The contradiction appears when the family satisfies the premise but crosses the density or structural boundary predicted by the conjecture.
The Lean certificate formalizes the finite graph lemmas and counting steps that support the stated counterexamples. The manuscript is required for the exact definitions of compactness and degeneracy used in the two chapters; these are not interchangeable labels.
What it could lead to
Counterexamples are maps for future theory. They tell researchers which hypotheses need strengthening and which proof strategies are likely to overreach. The next questions are constructive: can one classify the families that evade compactness, or replace degeneracy with a condition that actually controls density?
For AI-assisted mathematics, extremal graph theory illustrates why a system needs to search for adversarial families, not only elegant positive examples. A useful collaborator should ask where a theorem can fail and build the smallest scalable witness.
What not to conclude
The counterexamples do not invalidate extremal graph theory or imply that density is unpredictable. They disprove two specific conjectures and leave open the stronger hypotheses that may recover compactness or degeneracy control.