of time no check on the above computation was made for Sets 2 and 3 other than 

 visual observation of the results for reasonableness. For Sets 2A and 3C, 

 however, a check computation of S Q(p, q)(90 - q)(60 - p) was nnade, two 



p.q 



minor errors being discovered and corrected in the results for Set 3C. This 

 check computation required about 6 1/2 hours for each set and would have re- 

 quired about 11 hours for Data Sets 2 and 3. Since the Logistics Computer 

 does not include division as a basic operation, this had to be subroutined in 

 order to find the values of Q{p, q), a matter of a few minutes. This division was 

 checked by repetition of the program. The values of Q(p, q) were converted 

 from tape to punched cards for listing, the conversion being checked in the case 

 of Data Sets 2A and 3C by comparing the sum of the Q(p, q) on tape with the 

 corresponding card sum. 



The Q(p, q) were fed back into the computer, being doubled before being 

 stored. Each side of the resulting matrix was then multiplied by 1/2, the corT 

 ner elements belonging to two sides, being multiplied twice. In this manner 

 the coyariance surface Q*(p, q) was stored on the drum. 1.21095 cos — 

 (j = 0, 1, • • • , 39) were also stored on tJie drum. The factor 1.21095 x 10"^ 

 was introduced to divide by 800, to locate the decimal point (since, say, 0.311 

 was entered as 311), and to convert from the scale of the stereo planigraph to 

 feet (0.1016 mm = 1 foot). During the computation of tlie spectrum rp + sq 

 was reduced modulo 40 to the least non-negative residue. Because of the 

 ranges of p, q, r, and s, resetting of rp + sq at the lim.it of the range of any 



variable was quite easy. The total computation time for each spectrum was 



92 



