Effect of more stringent convergence criterion of estimated breeding values on response to selection

Abstract

1 Introduction In the analysis of large amounts of data to obtain BLUP, large sets of mixed model equations must be solved iteratively, which can involve considerable computing time. In real life, solutions are required only periodically for breeders to choose the best individuals, so that computing time is not usually a limiting factor. In simulation studies involving evaluation of individuals by BLUP, many rounds of evaluation are required for each simulated population. Since several or many replicates are usually required to obtain an accurate result from stochastic simulations, computing time can become a limiting factor. One of the factors that can drastically affect computing time in iterative methods is the criterion for ceasing iteration, or convergence criterion (CC). With iterative methods, a disadvantage can be that the rate of convergence can be slow, or under certain circumstances not converge at all. Nevertheless, when the system converges, the more stringent the CC, the more accurate the solutions. The more stringent the CC the more iterations and hence more computing time is required. The objectives of this study were to investigate how much response to selection is affected by the stringency of the CC and how much reranking occurs among selected individuals at different levels of the convergence criteria. These explorations provided a profile analysis of the computing time spent for each of the major subroutines in the BBSIM program.