Behavior of the generalized Rosenblatt process at extreme critical exponent values (Q2412672)

The generalized Rosenblatt process was introduced by \textit{M. Maejima} and \textit{C. A. Tudor} [Probab. Math. Stat. 32, No. 1, 167--186 (2012; Zbl 1261.60041)] as \[ Z_{\gamma_1,\gamma_2}(t)=C\int_{\{x_1\not=x_2\}}\int_0^t(s-x_1)_+^{\gamma_1}(s-x_2)_+^{\gamma_2}ds\,dB(x_1)\,dB(x_2) \] for some constant \(C>0\), a Brownian motion \(B(t)\) and exponents \(\gamma_1,\gamma_2\) in the open triangle \(\{-1<\gamma_1,\gamma_2<-\frac12,\,\gamma_1+\gamma_2>-\frac32\}\) which guarantees that \(Z_{\gamma_1,\gamma_2}\) is well defined. The process is non-Gaussian, has stationary increments and is self-similar with Hurst exponent \(H=\gamma_1+\gamma_2+2\in(\frac12,1)\), thus possesses long memory. It is a two-parameter generalization of the usual Rosenblatt process \(Z_{\gamma}(t)=Z_{\gamma,\gamma}(t)\) for \(\gamma\in(-\frac34,-\frac12)\) introduced by \textit{M. S. Taqqu} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 31, 287--302 (1975; Zbl 0303.60033)]. The authors give a full answer to the question of weak convergence in \(C[0,1]\) of the standardized \(Z_{\gamma_1,\gamma_2}\) as the parameters approach the boundary of the triangle and further show, that this convergence cannot be strengthened to \(L^2\)-convergence or convergence in probability. When \(\gamma_1+\gamma_2\to-\frac32\) with \(\gamma_1,\gamma_2>-1+\varepsilon\) for some \(\varepsilon>0\) the limit is Brownian motion \(B(t)\) and the rate of convergence is given in total variation distance. When \((\gamma_1,\gamma_2)\to(-\frac12,\gamma)\) or \((\gamma,-\frac12)\) with \(-1<\gamma<-\frac12\) the limit is \(W\cdot B_H(t)\) for a standard normal random variable \(W\) and an independent fractional Brownian motion \(B_H(t)\) with Hurst parameter \(H=\gamma+\frac32\). The precise rate of convergence is given in Wasserstein distance. When approaching the corners of the triangle, the limit is either a linear combination of the above if \((\gamma_1,\gamma_2)\to(-\frac12,-1)\) or \((-1,-\frac12)\), where \(B(t),W,B_{1/2}(t)\) are independent, or it is \(t\cdot X\) for a linear combination \(X\) of two independent standard \(\chi^2\)-distributed random variables if \((\gamma_1,\gamma_2)\to(-\frac12,-\frac12)\). In both cases, the linear combination depends on the direction in which the corner is approached and the authors give its precise factors.