Badly approximable points have full dimension on attractors where every previous technique fails.
Note (2026-11-12): re-read at full text (an earlier abstract-level pass held this essay up). The headline matches: the first class of non-linear, non-conformal attractors on which the badly approximable points have full dimension, answering a 2019 question of Das, Fishman, Simmons and Urbanski. The class is narrower than the essay's wording suggests: the paper's attractors are planar non-linear carpets satisfying a coordinate open set condition (the paper restricts to the plane because the dimension result it uses fails in higher dimensions even for self-affine sets). Its dimension result is a variational principle (the Hausdorff dimension is the supremum of the dimensions of ergodic measures), with Lyapunov exponents and Bernoulli measures in the proof; I did not find the essay's description of a pressure function accounting for direction-dependent stretching. The method is Schmidt's game with the lower dimension and a hyperplane-diffusivity condition; the essay's picture of avoiding thin tubes around rational hyperplanes is its own gloss.
A real number is badly approximable if no rational number gets too close to it relative to the denominator's size — there's a positive constant bounding how well rationals can approximate it. The set of badly approximable numbers has full Hausdorff dimension (it's as large as a fractal set can be while having measure zero). The same question for points on fractal attractors has been harder: does the badly approximable set retain full dimension when restricted to the attractor?
Anttila, Fraser, and Koivusalo (arXiv:2603.11822) answer yes for the first class of non-linear, non-conformal attractors — settling a question posed by Das, Fishman, Simmons, and UrbaĆski in 2019. Every previous result required either linearity (the attractor is built from affine maps) or conformality (the maps preserve angles locally). Non-linear non-conformal maps distort both scale and direction in ways that defeat the standard distortion estimates.
The new dimension formula they derive has independent value: it computes the Hausdorff dimension of these non-linear attractors in terms of a pressure function that accounts for the direction-dependent stretching. The formula connects fractal geometry to Diophantine approximation through the observation that badly approximable points, by resisting rational approximation, must avoid the thin tubes around rational hyperplanes — and the attractor's geometry determines whether this avoidance is compatible with full dimension.
The structural point: the property of being badly approximable (a number-theoretic condition) and the property of lying on a fractal attractor (a dynamical condition) interact through geometry. Neither condition implies the other, but they coexist at full dimension — the set of badly approximable points on the attractor is as large as the attractor allows.
Anttila, Fraser, and Koivusalo, "Badly approximable points on non-linear carpets," arXiv:2603.11822 (2026).