It is argued that the analysis of family extinction data that resulted in the claim of a 26-Myr periodicity of mass extinctions was flawed in that it did not allow for the possibility of a symmetric random walk model, which is shown to be perfectly consistent with the data.

A serial production line with an arbitrary number of stations is analyzed using the holding-time model. In contrast to Markovian state models, the holding-time model does not require the service times to be exponentially distributed. The distribution of the interdemand time at the first station is obtained as a function of the distribution of station service times which are of a general nature. The throughput rate of the line is the reciprocal of the mean interdemand time. For service times of phase type, two numerical solution methods, one exact and the other approximate, are presented. Numerical results are given for up to ten stations.

This paper studies consecutive-2 systems on trees. We show that given a set of probability values, the optimal assignment for binary trees depends on the particular values, whereas for k−regular trees, which are closely related to binary trees, it depends only on their relative ordering. In our proof, we introduce the notion of first-term-invariance, which might have further applications. We also show that the problem of computing the failure probability of consecutive-2 system is #P-complete in general, though the problem on trees is shown to be strongly polynomial.

This paper deals with a two-unit reliability model having one operating unit and one cold standby unit and with ample repair facilities. An approximate method is given for the calculation of the probability distribution of the proportion of time that the system is available in a given time period. The approximate method first computes the mean-time-to-failure and the meantime-to-repair and next approximates the up-and-down process of the two-unit reliability system by an alternating renewal process having exponentially distributed on and off periods. Numerical investigations show a satisfactory performance of the approximation. Also, a sensitivity analysis is given.

This paper considers a dam (or storage) model of the GI/G/I type with a finite capacity K. An arriving input being larger than the unfilled capacity of the dam causes an overflow where the excess amount is lost. Important performance measures for this system are the overflow probability and the long-run fraction of input that is lost. We give asymptotic expansions for these measures for large K both for the case of a load factor less than 1 and for the case of a load factor larger than 1. Also, related results are obtained for the impatient customer model of the M/G/l type.

In an influential and controversial paper, Raup and Sepkoski [2] defined an event of mass extinction to have occurred in any time period (of roughly 6.4 million years) for which the data value for that time period (equal to the proportion of the families existing at the beginning of that period that went extinct during the period) exceeded that of its immediate neighbors. In this article we analyze the occurrence of such events when the data are randomly generated from a continuous distribution.

Using data obtained by the general capture/recapture sequential sampling process, an exact analytical expression for the maximum likelihood (ML) estimate of the population size, N, is introduced. As a consequence, it is shown that bounded likelihood functions have at most two maxima. For the simple one-by-one case the ML estimate is unique.

We address the problem of assigning multiple copies of n independently developed versions of a program to a set of m(m > n) possibly heterogeneous processors to maximize system reliability. This problem is viewed as a partition and assignment problem. We first partition the set of processors into n clusters or subgroups. A program version is then assigned to be executed on all the processors in the cluster. This means that each processor in the cluster will execute a copy of the assigned version. The cluster's unreliability is the probability of failure of all its processors. Component i of this system is composed of the copies of version i and the assigned cluster of processors.

A large lot is to be sampled by a number of inspectors for the purpose of ascertaining its number of defective units. However, instead of a defective unit always being detected if it is sampled, there are unknown probabilities, depending on the inspectors, that the unit will not be identified as a defective. We offer an approach to estimating the number of defective units in the lot.

The spectral approach to first passage time distributions for Markov processes requires knowledge of the eigenvalues and eigenvectors of the infinitesimal generator matrix. We demonstrate that in many cases knowledge of the eigenvalues alone is sufficient to compute the first passage time distribution.

A sequential procedure to select optimal prices based on maximum likelihood estimation is considered. Asymptotic properties of the pricing scheme and the concommitant estimation problem are examined. For small sample sizes, simulation results show that the proposed procedure has high efficiency relative to the best procedure when the parameter is known.

This paper explores a new, more optimal policy for planning terminal bays for the provisioning of circuits in a telephone operating company. This new policy regularly monitors the terminal bay inventory and orders K units when the spare reaches a certain parameter, R. For particular office characteristics the optimal (R, K) pair which minimizes cost is determined, subject to a performance constraint. A probabilistic circuit demand model is formulated and a combination of simulation and analytic approximation is used to calculate the cost and performance measure. In order to prove that the new policy is more economical, the new policy is compared, via another simulation, to the present policy of a typical telephone operating company.

Two procedures for the group-testing problem based on the Shannon-entropy criteria are proposed. The model considered is that the N units are realizations of N Bernoulli independent and identically distributed (i.i.d.) chance variables with common, known probability q of an arbitrary unit being good and p =1 – q of it being defective. Both the algorithms introduced have low design complexity and yet provide near-optimal result. For N ≤ 5, one of the procedures introduced is optimal for selected values of q.

This paper deals with the MX/G/c queue. Using analytical results for the special cases of the MX/M/c queue and the MX/D/c queue, a two-moment approximation is proposed for the waiting-time percentiles in the general case. This approximation is based on a linear interpolation with respect to the squared coefficient of variation of the service time distribution. Validation experiments indicate that this approximation performs quite well for practical purposes. In particular, the practically important percentiles in the tail of the waiting-time distribution are approximated extremely well.

The lifetimes of the components of a device depend on each other because of their joint dependence on the environmental conditions. We introduce intrinsic age processes, one for each component, to handle such dependence. The data required can be obtained by experiments under controlled laboratory conditions. The computations needed for randomly varying conditions are recursive and can be used for making decisions regarding maintenance and replacement.