Computing Strategies for Solving Mixed Model Equations

Three approaches to constructing and solving sets of equations of large order are discussed and compared on the basis of number of data elements that must be processed by a computer program during a round of iteration of the solution phase. An indirect approach is proposed where solutions to mixed model equations are eomputed without explicitly setting up the equations, as the data file (or the data arranged by subclasses) is read once per round of iteration. A typical sire evaluation model is used to compare approaches. For this application, the indirect approach required reading fewer data elements than traditional approaches for each round of iteration. Savings in number of records read would be realized for other models and applications from the indirect approach.