Abstract

Analog optics is very attractive to optical information processing and computing because of its ability to process 2-D data in parallel very rapidly, but its accuracy is low. Digital electronics is much slower but more accurate. An optical-hybrid system would be a compromise. In this paper we present a bimodal optical computer (BOC) capable of determining the eigenvalues and eigenvectors for positive definite matrices with very high accuracy and speed. This system combines the accuracy of the digital processors and the speed of analog optics. The algorithm is an iterative using the inverse power method. The effect of the condition number of the matrix on the convergence of the solution is studied. The effects of the errors in the I/O devices are analyzed and it is found that with a tolerance of 5% in the electrooptical devices solutions still converge. Also the speed of the BOC in solving this class of linear algebra problems is analyzed and compared with the digital computer.

© 1986 Optical Society of America