Multifractal formalism for almost all self-affine measures (Q1942289)

Let \(T_i: \mathbb{R}^d\to\mathbb{R}^d\) be contractive linear maps and \(a_i\in\mathbb{R}^d\), \(i=1,\dots,m\). The article is devoted to the multifractal formalism for self-affine measures associated with iterated function systems of the form \((\mathbb{R}^d, T_i+a_i: i=1,\dots,m)\). Recall that given a positive probability vector \((p_i)_{i=1}^{m}\), \(p_i>0\), \({\sum}_i p_i = 1\), there exists a unique probability measure \(\mu\), called self-affine, which fulfills the following invariance equation \[ \mu(B) = \sum_{i=1}^{m} p_i\cdot \mu\circ (T_i+a_i)^{-1}(B) \] for all Borel sets \(B\subset\mathbb{R}^{d}\), and moreover, \(A=\bigcup_{i} (T_i+a_i)(A)\), where \(A=\operatorname{supp} \mu\). Let us denote \({\phi}^{0}(T)=1\) and for \(s>0\), \[ \phi^s(T) = {\alpha}_1\cdot\dots\cdot\alpha_{k-1}\cdot {\alpha}_{k}^{s-k+1}, \text{ when } k-1< s \leq k \leq d, \] \[ (\alpha_1\cdot\dots \alpha_d)^{s/d}, \text{ when } s\geq d, \] for nonsingular linear \(T: \mathbb{R}^d\to\mathbb{R}^d\), where \(\alpha_1\geq \alpha_2\geq\dots\geq \alpha_d\) are the positive square roots of the eigenvalues of \(T^{*}T\). Fix \(q\geq 0\). We need also the following notation: \[ S= \left\{s\geq 0: \sum_{k=1}^{\infty} \sum_{(i_1,..,i_k)\in\{1,..,m\}^k} \left(\phi^s(T_{i_1}\circ\dots\circ T_{i_k})\right)^{1-q} \cdot \left( \prod_{j=1}^{k} p_{i_j}\right)^{q} <\infty \right\}, \] \[ D(q) = (q-1)\cdot \operatorname{mum} S, \] \[ \tau(q) = (q-1) \cdot \min \left\{\frac{D(q)}{q-1}, d\right\}, \] where mum means infimum if \(0\leq q <1\), supremum if \(q>1\), and \(D(q) = \tau(q)= 0\) as \(q-1 =0\). The level sets are \[ E(\mu,\alpha) = \{x\in\mathbb{R}^d: \lim_{r\to 0} \frac{\log\, \mu(B_r(x))}{\log r} = \alpha\} \] and \(\underline{E}(\mu,\alpha)\), if \(\lim_{r\to 0}\) is replaced with \(\liminf_{r\to 0}\); \(B_r(x)\) stands for the closed \(r\)-ball at \(x\). The main result (Theorem 1.3) states that for a self-affine measure \(\mu\) associated with \(T_i+a_i\), \(i=1,\dots,m\), \(\|T_i\|< 1/2\), \((a_i)_{i=1}^{m}\in\mathbb{R}^{md}\) (except a set of parameters \((a_i)_{i=1}^{m}\) of the null \(md\)-dimensional Lebesgue measure) the Hausdorff dimension of level sets obeys \[ \dim_{H} \underline{E}(\mu,\alpha) = {\alpha} q - \tau(q) \] in several specific cases: (i) either \(\alpha=D'_{-}(q)\) (left derivative) or \(\alpha = D'_{+}(q)\) (right derivative), and \(0<q<1\), \(D(q)/(q-1) <1\), \({\alpha} q - D(q) \leq 1\); (ii) \(1<q<2\), \(T_i=\operatorname{diag}(t_{i,1},\dots,t_{i,d})\) with \(1/2>t_{i,1}>\dots>t_{i,d}>0\), and \(k<D(q)/(q-1)< k+1\) for some integer \(k\). In case (ii) with the additional assumption \(k=0\), as well as in case (i) we have in fact the stronger assertion: \(\dim_{H} E(\mu,\alpha) = \dim_{H} \underline{E}(\mu,\alpha)\). The paper contains many complementing remarks and results which we omit to stay concise.