The paper studies self-maps of smooth projective varieties, which are geometric spaces defined by polynomial equations. Specifically, it looks at a class of maps called "int-amplified endomorphisms," meaning the map stretches the geometry of the space in a robust way in every direction. The central objects of interest are "totally invariant cycles," which are geometric subsets (like curves, surfaces, or higher-dimensional pieces sitting inside the variety) that get mapped back onto themselves, possibly after applying the map several times. The main question is: how many such invariant pieces can there be, and how large can they be in total?
The authors prove a sharp upper bound on the total size of these invariant pieces, measured using a natural notion of degree coming from an ample line bundle, which is essentially a way of measuring geometric complexity. Their bound is expressed in terms of something called a Hilbert function, which is a classical algebraic tool that counts how many independent polynomial equations of a given degree are needed to describe a geometric object. The key achievement is that this bound is explicit and works in every codimension, meaning it controls invariant pieces of all possible dimensions simultaneously.
For the most concrete and classical case, projective space of dimension n, the result says the number of totally invariant prime cycles of a given dimension is at most a binomial coefficient, specifically "n+1 choose r" where r relates to the codimension of the cycles. Crucially, the authors show this bound is optimal, meaning there exist maps that actually achieve it. This gives a complete and tight picture of how many invariant geometric pieces a self-map of projective space can have, which has implications for understanding the dynamics and structure of algebraic maps more broadly.