Spectral criteria, SLLN's and a.s. convergence of series of stationary variables (Q2563935)

Let \(X=\{X_t\}\) be a zero-mean weakly stationary homogeneous random field indexed by \(R^d\). Then \(X_t=\int_{R^d} e^{it\lambda}Z(d\lambda)\), where \(Z\) is a random measure. The usual averages are defined by \(B_t= c_dt^{-d}\int_{|s|\leq t}X_s ds\), where \(c_d\) is a constant and \(|\cdot|\) is the Euclidean norm. \textit{V. F. Gaposhkin} [Theory Probab. Appl. 22(1977), 286-310 (1978); translation from Teor. Veroyatn. Primen. 22, 295-319 (1977; Zbl 0377.60033)] proved that \(B_t\to Z(\{0\})\) a.s. as \(t\to\infty\) if and only if \(Z(\{\lambda:|\lambda|\leq 2^{-k}\})\to Z(\{0\})\) a.e. as \(k\to\infty\). These conditions describe the strong law of large numbers denoted by \(X\in\text{SLLN}\). Let \(S^r_d\) be the \((d-1)\)-dimensional sphere of radius \(r\) in \(R^d\) and let \(\sigma^r_d\) be the unique rotationally invariant normalized measure on \(S^r_d\). Define \(A_t= \int_{S^r_s} X_s\sigma^t_d(ds)\). Assume that \(d\geq 3\). The authors prove that \(A_t\to Z(\{0\})\) a.s. if and only if \(X\in\text{SLLN}\). The restriction on dimension is sharp.