The empirical process on Gaussian spherical harmonics. (Q1879938)

Let \(Y_{m,l}(\theta,\varphi)\) be the spherical harmonics, defined by \[ Y_{m,l}(\theta,\varphi)= \begin{cases} \sqrt{{2l+1\over 4\pi}{(l- m)!\over (l+m)!}} P_{l,m}(\cos\theta)\exp( \operatorname{im}\varphi)\quad &(m> 0),\\ (-1)^m Y^*_{l,-m}(\theta,\varphi)\quad &(m< 0),\end{cases} \] where the asterisk denotes complex conjugation and \(P_{l,m}(\cos\theta)\) denotes the associated Legendre polynomial of degree \(l\), \(m\). Consider a random field \(T(\theta,\varphi)\) indexed by the unit sphere \((S^2)\) with the spectral representation \[ T(\theta,\varphi)= \sum^\infty_{l=1} \sum^l_{m=-1} a_{l,m} Y_{l,m}(\theta,\varphi), \] where \(\{a_{l,m}\}\) is a set of random coefficients. Let \[ y_{l,m}= 1-\exp\Biggl(- {l| a_{l,m}|^2\over\sum^l_{k= 1}| a_{l,k}|}\Biggr)\qquad (m= 1,2,\dots, l) \] and define the doubly indexed empirical process \[ \widehat K_L(\alpha, r)= {1\over\sqrt{L}} \sum^{[Lr]}_{l=1} \Biggl\{{1\over\sqrt{l}} \sum^l_{m=1} (I(y_{l,m}\leq \alpha)- \alpha)\Biggr\}\qquad (0\leq \alpha,r\leq 1). \] Proving the existence of \(\lim_{L\to\infty} E\widehat K_L(\alpha, r)\) \((=2\sqrt{r}b(\alpha)\) (say)), the authors establish weak convergence of \(\widehat K_L(\alpha, r)- 2\sqrt{r}b(\alpha)\) and show that this result is applicable to statistical inference procedures and in particular to various tests for Gaussianity.