The Ambro-Kawamata effective non-vanishing conjecture is a long-standing open problem in algebraic geometry. It predicts that if you have a projective algebraic variety (a geometric object defined by polynomial equations) and a divisor D (roughly, a way of measuring "positivity" on that variety) satisfying certain natural positivity conditions relative to the variety's canonical class (a geometric invariant capturing its curvature), then the space of global sections of D must be non-trivial. In plain terms: if D is "positive enough" compared to the geometry of the variety, there should exist at least one non-zero global function or section associated to D.
The authors construct explicit counterexamples showing this conjecture is false. They build a four-dimensional projective variety X with mild singularities (called terminal singularities, which are the tamest kind allowed in modern birational geometry) and a positively curved divisor D satisfying all the conditions the conjecture requires, yet the space of global sections of D is zero, meaning the conjecture's conclusion completely fails. They also construct a smooth (singularity-free) four-dimensional variety with similar properties, which is even more striking because smooth examples are rarer and harder to build. The constructions work over algebraically closed fields of most characteristics, including characteristic zero, which is the classical setting where the conjecture was originally posed.
These counterexamples matter because the effective non-vanishing conjecture sits at the heart of the Minimal Model Program, a major research framework aimed at classifying algebraic varieties by simplifying them step by step. Many results in this program assume or hope that global sections exist in situations like the one described, and those sections are used to construct the geometric maps that drive the classification. By showing the conjecture fails, this paper forces researchers to either impose stronger hypotheses in related theorems or seek fundamentally different approaches to proving results that were expected to follow from it.