Imagine you have an infinite grid of cells, some colored black and some white, and you want to understand how complex the pattern can be. One way to measure complexity is to pick a small "window" shape and count how many distinct patterns appear when you slide that window across the entire grid. Nivat's conjecture, proposed in 1997, says that if this count stays surprisingly low (specifically, no more than the number of cells in your window), then the overall grid pattern must be highly structured and repetitive, meaning it must repeat itself in at least one direction. Proving or disproving this conjecture for arbitrary window shapes is a major open problem in the mathematics of symbolic dynamics and combinatorics.
The paper constructs an explicit counterexample showing that Nivat's conjecture fails for certain non-convex window shapes. The authors build a specific 8-cell window shape and a grid coloring pattern such that the coloring has low complexity with respect to that window, yet the pattern is not periodic in any direction, and you cannot even nudge the pattern slightly (in a technical sense called "orbit closure") to find one that is periodic. This is stronger than previously known counterexamples, because earlier failures of the conjecture involved "degenerate" windows, meaning windows whose cells all happened to lie on a coarser sub-grid, making the problem secretly simpler. A researcher named Kari and a collaborator had asked whether all counterexamples must be degenerate in this way. This paper answers that question with a firm no, by producing a genuinely non-degenerate counterexample.
The authors also prove a positive result to show this behavior cannot happen with very small windows. For windows whose cell count is the square of a prime number, the conjecture holds. Combined with an earlier theorem covering windows of prime size, this means no window with fewer than 8 cells can produce this bad behavior, except possibly windows with exactly 6 cells, which remains an open question. The paper therefore precisely locates where the conjecture starts to break down, making 8-cell non-convex windows the minimal known frontier for this kind of complexity-periodicity failure.