02.07.2013 Views

proceedings

proceedings

proceedings

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

FORUM PROCEEDINGS<br />

Mr. Ferber: If you want to solve a small set of equations<br />

with seven or eight unknowns, the machines aren't<br />

adapted to working with such a small number of cards<br />

and one comes to the point where it is quicker to do them<br />

some other way. Then again, since the work goes up very<br />

rapidly with increasing order, you soon come up against a<br />

blank wall in that the work is prohibitive. How do you<br />

handle the problem?<br />

Mr. Bell: We do enough matrix work to keep the basic<br />

602 panels wired up ; so we need not allow time for that.<br />

Our operators are familiar with the job, and they handle<br />

it efficiently even with only seven or eight unknowns.<br />

Everything is kept ready for them. To give you some idea<br />

of time, it requires between one and two hours to evaluate<br />

an eighth order complex determinant. I f there are enough<br />

problems, you can split them up so that some are mUltiplying<br />

while others are being sorted and gang punched, with<br />

no idle machines.<br />

We have tried to handle special cases by special procedures,<br />

but have lost on it. I feel it is better not to use<br />

special procedures. Instead we stick to methods that all of<br />

our people understand; then we are more certain of coming<br />

out with a good answer.<br />

Mr; Harman: I was wondering if you could get around<br />

the division step in your formula by the expansion of a<br />

second-order minor, which involves the difference between<br />

two multiplications.<br />

Mr. Bell: We have tried doing that, and here is the.problem<br />

we got into: the numbers change tremendously in<br />

size, and it is necessary to stop for inspection. That takes<br />

longer than the regular method.<br />

Dr. Tukey: The thing is to get away from division by<br />

small numbers.<br />

Mr. Bell: Yes. 1\1:ost of our work is brought in to us.<br />

Sometimes we know what is behind the problems, and<br />

sometimes we do not. Most of the work is in engineering<br />

fields where extreme accuracy is not required. But we do<br />

have to calculate with expanded accuracy occasionally.<br />

Dr. Tttkey,' As I understand the situation, the pessimists<br />

thought-and I was one of them-that you lost, roughly<br />

speaking, a constant number of significant figures for each<br />

new equation. Now it has been proved! that it doesn't go<br />

like that; the loss goes like the logarithm of the order.<br />

If you are thinking of big equations that is a tremendous<br />

improvement. I think that size ought not to be taken as<br />

grounds for pessimism.<br />

REFERENCE<br />

1. J. VON NEUMANN and H. H. GoLDSTINE, "Numerical Inverting<br />

of Matrices of High Order," Bull. Amer. Math. Soc., 53 (1947),<br />

pp. 1021-99.<br />

31

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!