SCALING OF MULTIFRACTAL MEASURES UNDER AFFINE TRANSFORMATIONS (Q5701545)

Generally a measure which is statistically invariant under a certain group of transformations does not display invariance under other transformations. The authors find one exception to this rule, which has important practical implications. A random measure \(X(\Omega)\), \(\Omega \subset {\mathbb R}^d\), is said to be isotropically multifractal if it satisfies a scale invariance condition: \[ \overline{X}(\Omega) \overset \text{d} {=} A_r \cdot \overline{X}( r \Omega), \qquad r \geq 1, \tag{1} \] in the sense of all finite-dimensional distributions, where \(\overline{X}(\Omega)\) \(=\) \(X(\Omega) / | \Omega |\) is the average measure density in \(\Omega\), \(A_r\) is a nonnegative random variable independent of \(\overline{X}( r \Omega)\) such that \(E(A_r) = 1\). The distribution of \(A_r\) determines the scaling properties of the measure. Let \(\alpha =\) \(( \alpha_1, \dots, \alpha_d)\) be a vector of scaling exponents such that \(\alpha_1=1\) and \(0 \leqslant \alpha_i \leqslant 1\) for \(i > 1\) and denote by \(\Omega_{r, \alpha}\) the region obtained from \(\Omega\) through the affine expansion \[ \Omega_{r, \alpha} = \{ ( r^{\alpha_i} x_i, i=1,2, \dots, d); \;x=(x_1, \dots, x_d) \in \Omega \}, \quad r \geq 1. \tag{2} \] Hence, \(\Omega_{r, \alpha}\) results from stretching \(\Omega\) by \(r\) in the \(x_1\) direction and by \(1 \leqslant r^{\alpha_i} \leqslant r\) in the other directions. The authors are interested in whether (1) implies a scaling property: \[ \overline{X}(\Omega) \overset \text{md} {=} A_{r, \alpha} \cdot \overline{X}( \Omega_{r, \alpha}), \quad r \geq 1, \tag{3} \] in the sense of marginal distributions. One of the main concerns is to determine how the random variable \(A_{r, \alpha}\) in (3) is related to the random variable \(A_r\) in (1). Thus the authors consider multifractal measures \(X\) which are statistically invariant under isotropic contraction of the space and multiplication of the field by an independent random variable, and examine whether these measures possess any statistical invariance under affine space transformations. First of all, the validity of (3) is analyzed by using discrete cascades, and the asymptotic validity of (3) is found when the region \(\Omega\) is highly elongated in one or several coordinate directions. In that case, the distribution of \(A_{r, \alpha}\) is obtained in a simple way from the distribution of \(A_r\) in (1). For non-extremely elongated regions, (3) does not hold exactly. More precisely, the main result asserts that, under affine transformations, the marginal distribution of the measure retains multifractal scaling properties for rectangular regions that are highly elongated in one or several directions. The scaling properties depend on the direction of elongation and have a simple relationship with those under isotropic space transformations. Moreover, the authors use numerical simulation to investigate the transition range between asymptotic scaling regions, when the regions do not have extreme elongation in any one direction. This result has important practical implications as it can be used to guide and legitimize frequently used procedures to infer the scaling parameters of multifractal fields from observations in regions with one or more fixed dimensions.