On the uniqueness of maximizers of Markov-Gaussian processes (Q1805961): Difference between revisions

The paper starts with a very nice characterization of unimodal random processes: A stochastic process \(Y= \{Y(t): a\leq t\leq b\}\) with continuous sample functions attains its maximal value at precisely one point with probability one if and only if \[ \int^b_a P\Biggl( \sup_{a\leq t\leq x} Y(t)= \sup_{x\leq t\leq b} Y(t)\Biggr) \lambda(dx)= 0, \] where \(\lambda\) denotes the Lebesgue measure. Using this characterization, the following theorem, which is closely related to a well-known result found by \textit{M. A. Lifshits} [Theory Probab. Appl. 27, 600-607 (1982); translation from Teor. Veroyatn. Primen. 27, 559-566 (1982; Zbl 0495.60045)] is proved: Let \(Y= \{Y(t): a\leq t\leq b\}\) be a non-constant Markov-Gaussian process with a.s. continuous functions. If the set \(D= \{t\in[a, b]: \text{Var}(Y(t))= 0\}\) has Lebesgue measure zero, then the process \(Y\) and the reflected process \(|Y|\) are unimodal almost surely. Thus for instance, the Brownian bridge and the reflected Brownian bridge are unimodal almost surely.