Limit theorems for self-similar tilings (Q1950369)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Limit theorems for self-similar tilings |
scientific article |
Statements
Limit theorems for self-similar tilings (English)
0 references
13 May 2013
0 references
The authors consider the translational dynamics on the space of self-similar higher dimensional tilings exhibiting finite local complexity under translations (FLC). The mechanism to generate these tilings of \(\mathbb{R}^d\) is a typical tile-substitution rule \(\omega\), that sends each prototile \(T=(A,i)\), with \(A=\text{supp}(T)\subset \mathbb R^d\) and \(i\in\{1,\ldots,m\}\), into an expanded prototile \(\omega(T)\) that can be decomposed into a union of tiles (translated copies of some prototile) with disjoint interiors, and whose support is \(\phi (A)\), for some expanding map \(\phi:\mathbb R^d\to\mathbb R^d\) satisfying \(|\phi(x)|=\lambda |x|\), \(\lambda>1\). The space of self-similar tilings \(X_\omega\) is formed by tilings where each patch is contained in a translation of some power \(\omega^k(T)\) of a prototile \(T\). The substitution \(\omega\) is considered to be non-periodic (that is, for all \(\mathcal T\in X_\omega\) and \(y\in\mathbb R^d\), \(\mathcal T-y=\mathcal T\) implies \(y=0\)) and primitive, meaning that for some \(k\) we have prototiles of all types in \(\omega^k(T)\), for all prototile \(T\) or, equivalently, the substitution matrix \(S=(s_{ij})\) is primitive, where \(s_{ij}=\#\{\text{copies of } T_i \text{ in } w(T_j)\}\). Under the FLC assumption, it is well-known that the dynamical system \((X_\omega,\mathbb R^d)\) is uniquely ergodic. The tiling space \(X_\omega\) can be locally understood as a product of the unstable ``Euclidean leaf'' -- an open set in \(\mathbb R^d\) -- and the transversal (the set of all tilings which agree with some \(\mathcal T\) exactly on some patch containing the origin in its interior) as stable leaf, which is a Cantor set with the structure of a topological Markov chain. The unique invariant ergodic probability measure \(\mu\) is locally the product of the Lebesgue measure on \(\mathbb R^d\) and a Markov measure. This work concerns the deviation of ergodic averages for ``cylindrical test functions'' over balls or cubes of radius \(R\), or even more general increasing families of Lipschitz domains. The main tool is a family of finitely-additive measures associated with the dynamical system. It is known that \(\mu\) is related to the dominant eigenvalue \(\theta_1\) of \(S\) and its left and right eigenvectors. For each eigenvalue \(\theta\) of \(S\) with absolute value larger than \(\theta_1^{\frac{d-1}d}\) we can associate two finitely-additive complex (or real signed) measures: one defined on an algebra of sets in \(\mathbb R^d\) including Lipschitz domains, and another one defined on the transversal. It is interesting to notice that, in a nutshell, for the first type of measures the contribution of the eigenvalues with absolute values in \((1, \theta_1^{\frac{d-1}d})\) to the deviation of the ergodic average, for instance on a ball of radius \(R\), appears concentrated at the boundary of the sets, a phenomena that has no correspondence in the unidimensional case and that here turns out to be bounded from above (generically by \(CR^{d-1}\)). On the other hand, the latter measures yields an invariant finitely-additive measure for the dynamical system, if we take a product (locally) with the Lebesgue measure. In the main result, Theorem 4.3, the ergodic deviations are estimated using the mentioned finitely-additive real signed measures. Moreover, under the additional assumptions that the tiles are polyhedral, the similarity map \(\phi\) is a pure dilation (\(\phi(x)=\lambda x\), \(\lambda>1\)), and the second eigenvalue \(\theta_2\) (ordered by its absolute value in decreasing fashion) is real, simple, and satisfies \(\theta_2> |\theta_3|\), the authors prove in Theorem 6.1 that the deviations of ergodic averages obey a limit law: more precisely, averages on cubes of side \(r\lambda^n\), appropriately normalized, converge in distribution to a non-degenerate random variable. It remains to mention that this asymptotic formulas and limit laws were obtained for \(d=1\) in [\textit{A. Bufetov}, ``Finitely-additive measures on the asymptotic foliations of a Markov compactum'', Preprint, \url{arXiv:0902.3303}], where, contrary to the higher dimensional case, the '``boundary effects'' have no meaning.
0 references
self-similar tilings
0 references
tiling dynamical system
0 references
deviation of ergodic averages
0 references
0 references
0 references