Multifractal analysis of superprocesses with stable branching in dimension one (Q888542)

The paper under review concerns the multifractal analysis of superprocesses with stable branching in dimension one. The \((\alpha, d, \beta)\)-superprocess \(X_t\) in \(\mathbb{R}^d\) is a finite measure-valued process related to the log-Laplace equation \[ \frac{du}{dt}= - (-\Delta)^{\alpha/2} u + a u -b u^{1+\beta}, \] where \(a\in \mathbb{R}\) and \(b>0\) are fixed constants and \(0<\alpha \leq 2, \beta \in (0, 1]\). The fractional Laplacian \(\Delta_{\alpha} = - (-\Delta)^{\alpha/2}\) determines a symmetric \(\alpha\)-stable motion in \(\mathbb{R}^d\) of index \(\alpha \in (0, 2]\) and its continuous-state branching mechanism \(v\mapsto -av +bv^{1+\beta}\) for \(v\geq 0\). \textit{K. Fleischmann} [Math. Nachr. 135, 131--147 (1988; Zbl 0655.60071)] showed that the process \(X_t\) is almost surely absolutely continuous with respect to the Lebesgue measure for \(d<\frac{\alpha}{\beta}\) and \(a=0\); \textit{N. Konno} and \textit{T. Shiga} [Probab. Theory Relat. Fields 79, No. 2, 201--225 (1988; Zbl 0631.60058)] showed that there is a continuous version of the density of \(X_t\) for all \(\alpha > 1\) and \(\beta =1\); \textit{K. Fleischmann} et al. [Ann. Probab. 38, No. 3, 1180--1220 (2010; Zbl 1207.60055)] showed that there is a continuous version of the density of \(X_t\) if \(d=1\) and \(\alpha > 1+\beta\), otherwise the density is unbounded on open sets of positive \(X_t(dx)\)-measure, and the optimal index for local Hölder continuity of \(X_t\) is determined as \(\eta_c=\frac{\alpha}{1+\beta}-1 \in (0, 1)\). The main result of the paper under review is Theorem 1.2 (Multifractal spectrum) that the Hausdorff dimension \(D_U(\eta)\) with probability one is given by \((1+\beta)(\eta - \eta_c)\) for any open set \(U\subset \mathbb{R}\) and \(\eta \in [\eta_c, \overline{\eta}_c)\setminus \{1\}\) whenever \(X_t(U) > 0\), where \(\overline{\eta}_c = H_X(x) = \frac{\alpha - \beta}{1+\beta}\). The result fails in the case \(\eta=1\) due to the critical borderline between differentiable and nondifferentiable functions. Section 2 starts with some estimates for the transition kernel of the \(\alpha\)-stable motion and a bound for stable processes. By the martingale decomposition of the \((\alpha, 1, \beta)\)-superprocess \(X\), the density function of \(X_t(dx)\) can be decomposed into \(Z^1(x)=\mu \star p_t^{\alpha}(x)\) (twice differentiable), \(Z^2(x) =\int_{(0, t]\times R}M(d(s, y))p^{\alpha}_{t-s}(y-x)\) (jumps of the martingale measure \(M\) try to destroy smoothness of \(X_t\) and \(p^{\alpha}\) tries to make \(X_t\) smoother) and \(Z^3(x)= a\int_{(0, t]\times \mathbb R}I(d(s, y))p^{\alpha}_{t-s}(y-x)\) (Hölder continuous of index \(\alpha 1_{\overline{\eta}_c >1} +1_{\overline{\eta}_c \leq 1}\)). Lemma 2.6 shows that \(\frac{dZ^3(x)}{dx}\) is \(P\)-a.s. Hölder continuous with any exponent \(\eta < \alpha -1\) and \(\beta < (\alpha -1)/2\). Section 3 is devoted to the proof of the upper bound for the Hausdorff dimension in Proposition 3.1, which depends on the construction of a random set \(S_{\eta}\) such that \(\dim S_{\eta} \leq (1+\beta) (\eta - \eta_c)\) \(P\)-a.s. in Lemma 3.3. Lemma 3.12 shows Hölder continuity with index \(\eta - 2 \alpha \gamma\) for \(Z^2\) on the domain \((0, 1)\setminus S_{\eta}(J)\). Section 4 is devoted to the proof of the lower bound for the Hausdorff dimension for \(Z^2\) with \(\eta \in (\eta_c, \overline{\eta}-c)\setminus \{1\}\) almost surely on \(X_t(0, 1)>0\) in Proposition 4.1. The proof of Proposition 4.1 is more involved than the proof of the upper bound. The strategy is to derive some uniform estimate on the masses of \(X_s\) of dyadic intervals at time \(s\) close to \(t\) (Section 4.1), then to construct a set \(\tilde{J}_{\eta, 1}\) with \(\dim \tilde{J}_{\eta, 1} \geq (1+\beta) (\eta - \eta_c)\) on those big jumps of \(X\) that occur close to time \(t\) (Section 4.2), to link those jumps with the ones of the auxiliary processes \(L^+\), and to further understand the compensation of the impact of the jumps of \(X\) encoded in \(L^+\). The most difficult part is to show that a compensation is possible on a set of Hausdorff dimension strictly smaller than \((1+\beta) (\eta - \eta_c)\) and hence does not influence the dimension lower bound in Proposition 4.1. Hence Theorem 1.2 follows. The multifractal spectrum results in this paper for \(1<d<\frac{\alpha}{\beta}\) would be interesting to investigate next. The authors also conjecture that their result holds for \(\overline{\eta}_c \geq 1\).