The tail of the convolution of densities and its application to a model of HIV-latency time (Q1198583)

Let \(p(x)\) and \(q(x)\) be probability density functions. Under the hypothesis of the regular oscillation of the functions \[ v(x)=-(d/dx)\log p(x)\qquad\text{and}\qquad w(x)=-(d/dx)\log q(x), \] the author obtains the asymptotic form of the convolution \((p*q)(x)\) for \(x\to\infty\). The concept of regular oscillation was introduced by the author [Adv. Appl. Probab. 14, 811-832 (1982; Zbl 0494.60076)]. Recall that a positive continuous function \(g(x)\) is regularly oscillating if \[ \lim_{u,u'\to\infty,u/u'\to 1} g(u)/g(u')=1. \] The results obtained are applied to study the limiting conditional distribution of a random variable, given the sum of itself and another random variable, and to a stochastic model previously proposed by the author [Biometrika 77, No. 4, 733-741 (1990; Zbl 0709.62098)] for the latency time for HIV infection, that is, the time between the moment of infection and the discovery of the presence of the virus in a blood test. The concept of regular oscillation is also used to obtain a new result of independent interest in extreme value theory (Lemma 2.3). Namely, the author gives a new sufficient condition for a distribution function \(F(x)\) to be in the domain of attraction of the extreme value distribution \(\Lambda(x)=\exp(-e^{-x})\).