86 TECHNICAL NOTES AND SHORT PAPERS
A point of practical importance is that the d's (which are, in fact, correlation
coefficients) do not exceed 1 in magnitude ; the same is true of the r's. For this
reason, the iterative scheme can easily be programmed using fixed-point arith-
metic, though double-precision working is frequently needed in the calculation of
the sines and cosines.
M. J. R. Healy
Rothamsted Experimental Station
Harpenden
Herts., England
1. H. Hotelling, "Relations between two sets of variâtes," Biometrika, v. 28,1936, p. 321-377;
2. A. C. Aitken, "The evaluation, with applications, of a certain triple product matrix,'
Roy. Soc. Edinburgh, Proc, v. 57, 1936, p. 172-181.
3. A. S. Householder, Principles of Numerical Analysis (Sect. 4.115), McGraw-Hill Book
Co., Inc., New York, 1953.
Polynomial Approximations to Bessel Functions of Order Zero
and One and to Related Functions
In the course of preparing a programme for a large digital computer, the
following formulas, which are of use in the calculation of Bessel functions on such
machines, have been obtained. The methods of obtaining the approximations are
those described by Lanczos in NBS AM S 9 [1], with trivial modifications (a
complete set of references is given in [1]). The results enable Jo(x), /i(x), Fo(x),
xFi(x), and Jo(x) to be calculated for all values of x and Ko(x), Ki(x), and
Ko(x) to be calculated for values of x exceeding unity; where we define
Jo(x) = | J0(t)dt
Jo
and
Ko{x) = fX K0(t)dt.
We define auxiliary functions such that
J«(X) = \— yPn(x) COSÍ X - y - ^j - ÇB(X)Sin Í X - y - ~J ) ,
» - 0,1,
F„.(x) = yj— \Pn(x) sin Í x - y - ^j + Qn(x) cos Ix - y - ^ ) ) ,
»-0,1,
yn(x) = yjn(x) log^ - -2 Yn(x)J , n = 0, 1,
Kn(x) = e-*yj^Gn(x), n = 0, 1,
Jo(x) = 1 - \¿.(^o(*) cos (x+l) - #o(x) sin (x + Jj V
£o(x) =\-<rx\yxG»(x)-
TECHNICAL NOTES AND SHORT PAPERS 87
Then with the maximum error stated in brackets in each case, and provided
0 ^ Z^ 1
J0(4Z) = 1.00000,00000 - 3.99999,98721Z2+ 3.99999.73021Z4
- 1.77775.60599Z6+ 0.44435.84263/8 - 0.07092,53492í10
+ 0.00767,71853z12- 0.00050,14415z14(10 X 10"10)
-Ji{At) = 1.99999,99998 - 3.99999.99710Z2+ 2.66666,60544z4
1 - 0.88888,39649Z6+ 0.17775.82922Z8- 0.02366,16773Z1(t
+ 0.00220,69155Z12- 0.00012,89769/» (3 X lO"10)
j.39894,22793 - 0.00175,30620Z2+ 0.00017,34300*4
a/2"" - 0.00004.87613Z6+ 0.00001,73565Z8- 0.00000.37043Z10
(12 X 10-10)
-J—Q0 (-) = - 0.01246,69441 + 0.00045.64324Z2- 0.00008.69791Z4
t V2T \tj + o.00003,42468i8 - 0.00001,42078Z8+ 0.00000,32312Z10
(25 X 10-10)
-i=Pi (-) = 0.39894,22819 + 0.00292.18256/2 - 0.00022.32030Z4
V2ir w + 0.00005,80759Z6- 0.00002,00920Z8+ 0.00000.42414Z10
(15 X 10-1»)
-J— Qi(-) = 0.03740,08364- 0.00063.90400Z2
+ 0.00010,64741z4
t\2ir \t/ - 0.00003.98708/6 + 0.00001,62200Z8- 0.00000,36594/10
(25 X 10-10)
yo(it) = - 0.57721,56649 - 1.69113.74142Z2+ 3.69113,88793**
- 2.23311,02234/« + 0.66943,21484/8 - 0.12141,87561/10
+ 0.01489,99271/12 - 0.00135.08487/14 + 0.00008,91322/1S
(6 X 10-10)
UyiiAt) = 1.00000,00004 - 0.61772.53972/2- 10.76454.72724/4
+ 11.62078,91416/6 - 4.91052,91148/8 + 1.14180,33012z1»
- 0.16910.81720/12+ 0.01699.21876/14- 0.00102,66368/,s
(6 X 10-10)
J-Go (-) = 1-25331,41373- 0.15666,41816/+ 0.08811,12782/2
>2 W _ 0.09139,09546/' + 0.13445.96228/4 - 0.22998.50328/5
+ 0.37924,09730*"- 0.52472,77331/7+ 0.55753.68367/8
- 0.42626,32912/» + 0.21845,18096/10- 0.06680,97672z11
+ 0.00918.93830/12 (10 X ÎO"10)
J-Gi (-) = 1.25331,41373+ 0.46999.27013Z- 0.14685.82957/2
V2 W + 0.12804,26636Z8- 0.17364,31637z4+ 0.28476,18149Z5
- 0.45943.42117Z6+ 0.62833.80681/7 - 0.66322.95430Z8
+ 0.50502.38576Z9- 0.25813,03765Z10+ 0.07880,00118/"
- 0.01082.41775/12 (10 X ÎO"10)
88 TECHNICAL NOTES AND SHORT PAPERS
7.99999,99990 - 42.66666,64204z2+ 102.39999,00866z4
- 130.03159.01993Z6+ 101.13454.38222/8 - 52.95243,71745Z10
+ 19.89838,57672Z12- 5.60245,72363z14+ 1.20146,77449Z16
- 0.18673.71001Z18+ 0.01624,75465Z20(10 X lO"10)
0.79788,45600 - 0.01256,42405z2+ 0.00178,70944z4
- 0.00067.40148Z6+ 0.00041,00676z8 - 0.00025,43955/10
+ 0.00011.07299/12 - 0.00002,26238Z14(8 X ÎO"10)
- 0.06233,47304 + 0.00404.03539Z2- 0.00100.89872Z4
+ 0.00053,66169/6 - 0.00039.92825Z8+ 0.00027,55037Z10
- 0.00012.70039/12 + 0.00002.68482Z14(8 X lO"10)
1.25331,39163 - 0.78323.44963Z+ 1.25733,12033Z2
- 3.09054,43850Z3+ 9.02560.45356Z4- 25.43912,19592Z6
+ 60.46288.82856Z6- 112.80726.52384Z7+ 158.83274,70627z8
- 163.74821.02377Z9+ 119.22659,27008Z10- 57.88900.96515Z11
+ 16.78876,58787Z12- 2.19748,24449/13 (25 X lO"10in K0).
An attempt to derive similar formulas for Jo(x), 7i(x), and 7o(x) by the same
methods proved unsuccessful, since these functions cannot be defined by a
differential equation and a boundary condition at infinity.
A. J. M. Hitchcock
United Kingdom Atomic Energy Authority
Industrial Group Headquarters
Risley, Warrington,
Lancashire, England
1. NBS Applied Mathematics Series, No. 9, Tables of Chebyshev Polynomials S„(x) and Cn(x),
U. S. Govt. Printing Office, Washington, D. C, 1952, Introduction.
An Iterative Method for the Solution of Linear Equations
Based on the Power Method for Proper Vectors
Introduction. A computing machine program for obtaining the largest proper
value and proper vector of an (w + 1, n + 1) matrix A by the power method
(i.e., by the iteration Xv+i = AX,) may be used for solving a system of n linear
equations in n unknowns. In the following note a method of setting up the itera-
tion together with a simple criterion for convergence of the iteration are given.
Method. Let B be an (n + 1, n + 1) matrix with real elements and real proper
value k such that | k | is greater than the moduli of all the remaining proper values
of B. If X is a non-exceptional non-zero column vector, X = col (£i, £2, • • •, £„, £„+i),
with real components, it is well known that the iteration Xo = X, Xi = BXo,
■• •, Xv+i = BXr is such that Xr+i = kXr. (There are exceptional choices of X
for which the iteration will not converge to the largest proper value and vector ;
but round-off error during the iteration will usually make the Xv non-exceptional