about 7 hours. In the case of Data Sets 2A and 3C the L,{r, s) were punched 

 into cards, the conversion from tape to cards being checked by summation. 

 In the earlier computation of Data Sets 2 and 3, due to an error made with 

 derivation of the equations in Part 8, the values of L*(r, s) were computed 

 and punched into cards. As a check SL<*(r, s) was compared with Q(0,0) /(1.016) , 

 to which it should be equal. This check was made by hand for Data Sets 2 and 3 

 and by machine for 2A and 3C. In all cases there was agreement to four signi- 

 ficant figures, the values for Data Sets 2, 3, 2A, and 3C being 4.614, 4.299, 

 4.105, and 4.049, respectively. 



For Data Sets 2 and 3 the final smoothing was performed incorrectly, due 

 to the above mentioned error, on the Li*{r, s) matrix rather than the L(r, s) 

 matrix. This error was corrected by the time Data Sets 2A and 3C were run, 

 and correct procedures are described in Part 8. The L(r,s) matrix on the 

 drum was bordered to provide the proper values for r = -1, 21 and s = -^21, 21. 

 A final computation of approximately 15 minutes per data set provides the 

 U(r, s). For Data Sets 2 and 3 U(r, s) (incorrect in the two outer columns 

 and rows because of the use of L*(r, s)) was punched on tape and converted 

 to cards. Many of these values were checked by hand, and no errors were 

 discovered. For Data Sets 2A and 3C U(r, s) (correctly computed from lj(rj s)) 

 was used to obtain U*(r, s) (the borders being multiplied by 1/2), punched out 

 on the tape, and converted to cards. For both of these sets as a check SL,*(r, s) 

 i was compared to SU*(r, s), agreement being obtained to seven significant 



1 figures. 



93 



