Schmidt's game, badly approximable matrices and fractals (Q1026881)

The author extends a result of \textit{W. M. Schmidt} [J. Number Theory 1, 139--154 (1969; Zbl 0172.06401)] on badly approximable systems of linear forms. A system of \(M\) linear forms in \(N\) variables is given by an \(M \times N\) matrix \(A\) with real entries. The system is said to be badly approximable if there is a constant \(C > 0\) which may depend on \(A\), such that \[ \text{dist}(A x, \mathbb{Z}^M) > C \| x \|^{-N/M} \] for any non-zero integer vector \(x \in \mathbb{Z}^N\). Denote the set of such matrices by \(BA(M,N)\). \textit{W. M. Schmidt} [loc. cit.] showed that this set has maximal Hausdorff dimension. This was accomplished by proving that the set \(BA(M,N)\) has the property of being a winning set for a certain game, the Schmidt game. In the present paper, the methods of Schmidt's paper are extended to prove that \(BA(M,N) \cap {\mathcal K}\) is winning for the Schmidt game whenever \({\mathcal K}\) supports a finite, absolutely friendly Borel measure. The technical definition of an absolutely friendly measure is found in the paper. It includes attractors of iterated function systems of contracting similarities satisfying the Open Set Condition, such as the Cantor middle third set and the Sierpiński gasket. In the latter cases, the winning property implies the maximal dimension property, and the result of the paper extends results of \textit{S. Kristensen}, \textit{R. Thorn} and \textit{S. Velani} [Adv. Math. 203, No. 1, 132--169 (2006; Zbl 1098.11039)] and independently by \textit{D. Kleinbock} and \textit{B. Weiss} [Isr. J. Math. 149, 137--170 (2005; Zbl 1092.28004)], where the case \(N=1\) is dealt with. The main theorem is proved by adapting Schmidt's methods to the present setup. The main difference is that Schmidt's construction of a winning strategy was more explicit. This cannot be accomplished in the more general setup, as the support of an absolutely friendly measure is in general rather wild. However, the decay properties of an absolutely friendly measure ensure that a strategy can still be constructed. In addition to the main theorem of the paper, the author shows that the more refined measure of size known as the winning dimension of a set cannot attain its maximal possible value of \(1/2\) for the present setup. This is done by proving that the winning dimension of \(BA(1,1)\) intersected with the Cantor middle third set is at most \(1/3\). Finally, an example of a set \(E\) supporting an absolutely friendly measure is given for which \(BA(1,1) \cap E\) is in fact equal to \(E\).