Animal Model Evaluation in Finland: Experience with Two Algorithms

¹ Department of Animal Breeding, Agricultural Research Center, SF-31600 Jokioinen, Finland

Two algorithms to solve large-scale animal model equations were implemented in Finland using iteration on data. In the first procedure, the breeding values were solved using second-order Jacobi iteration, and, in the second, the breeding values were solved by successive overrelaxation iteration. In both algorithms, the iteration was by Gauss-Seidel for effects other than breeding values.

The second-order Jacobi algorithm was modified to be 20% more memory efficient. The successive overrelaxation algorithm for breeding values was implemented to use less memory than the second-order Jacobi algorithm. Correlations between solutions after 300 rounds and intermediate solutions during the iteration were calculated to compare the algorithms. The algorithm with successive overrelaxation of breeding values converged to the final solutions faster. However, the second-order Jacobi iteration took less time per round than the successive overrelaxation algorithm.

Key Words: breeding value prediction • animal models • iterative methods

Submitted on October 7, 1991
Accepted on January 27, 1992