Eigenvalues of a Fredholm integral operator and applications to problems of statistical inference (Q1358739)

The eigenvalue problem of interest is \[ \lambda_p\Phi_p(t)=\int\limits_0^1 K_p(s,t)\Phi_p(s)ds, \] where the kernel is ``almost separated'' \[ K_p(s,t)=\min(s,t)+\sum\limits_{m=0}^ph_m(t)h_m(s). \] The problem was formulated by \textit{I. B. MacNeill} [Ann. Stat. 6, 422-433 (1978; Zbl 0375.62064)] while looking for the characteristic function of the stochastic integral given by \[ \int\limits_0^1B^2_p(t)dt, \] where \(\{B_p(t), t\in [0,1]\}\) is a continuous functional defined on the standard Brownian motion. He found explicitly the eigenvalues of (1) for \(p=1\). The objective of the present paper is to find explicit values of (1) for any general \(p\). The characteristic function of the stochastic integral is also found and can be used to obtain the distribution function of (2). The result has, as claimed, a direct application to the change--point problem in statistical inference.