Extremal graph theory asks how many edges you can pack into a network of n nodes before you are forced to create a specific forbidden pattern, called F. The maximum edge count that avoids F is called the extremal number, and the graphs that achieve this maximum are called extremal graphs. Once you add even one extra edge beyond that maximum, copies of F must appear. A natural follow-up question is: how does the number of forced copies of F grow as you keep adding more edges? This is the subject of supersaturation theory.
In 2013, Mubayi proposed a conjecture saying that the growth should be essentially linear and well-behaved. Specifically, he predicted that if you add q extra edges to an n-node graph, you get at least q times as many copies of F as you would get from adding just one extra edge to an extremal graph. The intuition is that each additional edge contributes independently and additively to creating new copies of F, so q edges should produce at least q times the copies that one edge produces.
The authors of this paper show that Mubayi's conjecture is false, and false in a strong way. For every integer d at least 2, they explicitly build a family of graphs F for which adding q edges produces strictly fewer than q times the copies generated by a single extra edge, and this gap holds for a wide range of q values, up to roughly n to the power 4/3. Because counterexamples exist across all values of d and across a broad range of q, the conjecture is not just slightly off in a narrow edge case but is fundamentally wrong. The result means that the number of forced copies of F can grow in a more irregular, sublinear fashion rather than the clean proportional growth Mubayi envisioned.