Markov or non-Markov property of \(cM-X\) processes (Q1884810)

Denote by \((B_t^{(\mu)})_{t\geq 0}\) one-dimensional Brownian motion with constant drift \(\mu\) and by \(M_t^{(\mu)}:= \sup_{0\leq s\leq t} B_s^{(\mu)}\) its running maximum. It is well known from (obvious variations of) Lévy's and Pitman's theorems that the processes \(-B_t^{(\mu)}, M_t^{(\mu)} - B_t^{(\mu)}\) and \(2 M_t^{(\mu)} - B_t^{(\mu)}\) are again Markov processes, see e.g. \textit{P. Lévy} [``Processus stochastiques et mouvement brownien'' (1948; Zbl 0034.22603)] resp. \textit{J. W. Pitman} [Adv. Appl. Probab. 7, 511--526 (1975; Zbl 0332.60055)]. The aim of the present paper is to show that all other combinations \(cM_t^{(\mu)} - B_t^{(\mu)}\), \(c\neq 0,1,2\), are non-Markovian. As a bonus, the authors give new elementary proofs for the Markovianity in the case where \(c=1,2\). It is mentioned that this result seems to be part of the folklore but, as so often in probability theory, no rigorous proof can be found. Part of the present main theorem was anticipated by \textit{T. Jeulin} [in: Séminaire de probabilités XIII. Lect. Notes Math. 721, 521--531 (1979; Zbl 0422.60028)] who showed that for \(\mu=0\) and \(c>1, c\neq 2\) the process \(cM-B\) is not a time-homogeneous Markov process. In contrast to Jeulin's arguments, the proofs in the present paper are elementary and completely based on explicit calculations of higher-dimensional (joint) distributions of \((M^{(\mu)},B^{(\mu)})\) and \(cM^{(\mu)} - B^{(\mu)}\); only the case \(1<c<2\) is done in detail, the more or less similar case \(c>2\) is left to the reader, and the arguments for \(0<c<1\) are outlined where they differ from the previous cases.