The paper develops a new geometric technique for producing large collections of rational curves (roughly, curves that can be parameterized by simple polynomial or fractional functions) inside higher-dimensional algebraic varieties. The classical "bend-and-break" method, going back to Mori in the 1970s, shows that when you continuously deform a curve while fixing some points, the curve can degenerate and break into simpler pieces, often producing rational curves in the process. The authors extend this idea to extract rational curves systematically from families of curves with more complicated topology, giving a construction that works at a larger scale and in more general settings than previous approaches.
A central theme is the relationship between rational curves and a geometric measure of how "positively curved" a variety is, captured by something called the anticanonical divisor and the Fujita invariant. Loosely, varieties that are more positively curved tend to be richer in rational curves. The paper makes this intuition precise by proving that varieties with a large Fujita invariant must contain large families of rational curves. The authors also construct so-called free rational curves on certain singular Fano varieties, a class of positively curved spaces that are particularly important in classification programs for algebraic varieties. Free rational curves are especially well-behaved ones that move around in families without obstruction.
As a concrete payoff, the paper settles a dimension-counting question about curve families on Fano hypersurfaces. A Fano hypersurface is a solution set of a single polynomial equation in projective space that satisfies a positivity condition, and its index measures roughly how positive that curvature is. For a generic such hypersurface of index at least 3, the authors prove that the space parameterizing curves of any fixed topological type (genus g) but sufficiently high degree has exactly the size that a naive parameter count would predict. This expected-dimension result confirms that no unexpected geometric coincidences force these curve families to be larger or smaller than they should be.