A note on non-regular martingales (Q956384)

For random variables \(Y_0,Y_1,\dots\) let \[ Y^*_n=\sup_{0\leq k\leq n}Y_k\quad\text{and}\quad Y^*=\sup_{k\geq 0}Y_k. \] \textit{A. G. Pakes} [J. Appl. Probab. 24, 768--772 (1987; Zbl 0636.60047)] showed that if \((Z_n)\) is a nonnegative submartingale such that \[ Z_0\equiv a>0,|E[Z_n\log Z_n]|<\infty(n\geq 1)\quad \text{and}\quad E[Z_n\log Z_n]\to\infty(n\to\infty), \] then \[ \limsup_{n\to \infty}\frac{E[Z_n^*]} {E[Z_n\log Z_n]}\leq a. \] In the case where \((Z_n)\) is a nonnegative martingale, the authors prove (using slightly simpler arguments) the following (slight) generalization of the above result: Theorem. Let \(Z_0, Z_1, \dots\) be a nonnegative martingale such that \[ Z_0\equiv a>0,\;E[Z_n\log^+Z_n]< \infty(n\geq 1)\quad\text{and}\quad E[Z_n\log^+Z_n]\to\infty(n\to\infty). \] Then \[ \limsup_{n\to\infty}\frac {E[Z_n^*]}{E[Z_n\log^+Z_n]}\leq a.\tag{*} \] The authors construct examples which show that, in general, there does not hold equality in (*).