Summary
In any iteration scheme, such as v k=f(Qv k−1), where a fixed matrix multiplies a vector that depends on the iteration number, Winograd's method for computing inner products can be used in a straightforward manner to reduce the number of multiplications required at the cost of more additions. The key observation is that certain quantities required by Winograd's method have to be computed only at the first iteration. In the Jacobi method for solving systems of linear equations, f is linear. Gauss-Seidel iteration often converges faster than Jacobi iteration, but it cannot be put in the above form. A simple trick is necessary to apply Winograd's method in an efficient recursive manner. Our proposed method is better than the naive method when it is faster to add than to multiply. Versions of Jacobi and Gauss-Seidel iteration appropriate for optimization (as in Markov decision problems) are presented. The analysis specializes easily to the linear equation case.
Similar content being viewed by others
References
Denardo, E. V.: Contraction mappings in the theory underlying dynamic programming. SIAM Review 9, 165–177 (1967)
Faddeeva, V. N.: Computational methods of linear algebra. New York: Dover 1959
Fox, B. L.: Accelerating LP algorithms. Comm. ACM 12, 384–385 (1969)
Fox, B. L.: Numerical calculation of the transient behavior of Markov renewal processes. Université de Montréal, Département d'Informatique, Report No. 119, February 1973
Kushner, H.: Introduction to stochastic control. New York: Holt, Rinehart, and Winston 1971
Varga, R. S.: Matrix iterative analysis. Englewood Cliffs (N.J.): Prentice-Hall 1962
Winograd, S.: A new algorithm for inner product. IEEE Trans. Computers C-17, 693–694 (1968).
Author information
Authors and Affiliations
Additional information
This research was supported by NRC grant A8565.
Rights and permissions
About this article
Cite this article
Fox, B.L. Reducing the number of multiplikations in iterative processes. Acta Informatica 3, 43–45 (1973). https://doi.org/10.1007/BF00288651
Received:
Issue Date:
DOI: https://doi.org/10.1007/BF00288651