Professional Documents
Culture Documents
with
Formulas, Graphs, and Mathematical Tables
Edited by Milton Abramowitz and Irene A. Stegun
1. Introduction
The present Handbook has been designed to tional importance. Many numerical examples
rovide scientific investigators with a compre- are given to illustrate the use of the tables and
Lnsive and self-contained summary of the mathe- also the computation of function values which lie
matical functions that arise in physical and engi- outside their range. At the end of the text in
neering problems. The well-known Tables of each chapter there is a short bibliography giving
Functions by E. Jahnke and F. Emde has been books and papers in which proofs of the mathe-
numerica
by
\
invaluable to workers in these fields in its many
editions1 during the ast half-century. The
present volume extends t e work of these authors
. more extensive and more accurate
tables, and by giving la er collections
matical properties stated in the chapter may be
found. Also listed in the bibliographies are the
more important numerical tables. Comprehen-
sive lists of tables are given in the Index men-
tioned above, and current information on new
of mathematical properties of %e tabulated tables is to be found in the National Research
functions. The number of functions covered has Council quarterly Mathematics of Computation
also been increased. (formerly Mathematical Tables and Other Aids
The classification of functions and organization to Computation).
of the chapters in this Handbook is similar to The mathematical notations used in this Hand-
that of An Index of Mathematical TaMa by book are those commonly adopted in standard
A. Fletcher, J. C. P. Miller, and L. Rosenhead.* texts, articularly HigLer Transcendental Func-
tions, volumes 1-3, by A. Erdblyi, W. M nus,
In general, the chapters contain numerical tables,
graphs, polynomial or rational approximations
F. Oberhettinger’andF. 0. Tricomi (Mc raw- 8
Hill, 1953-55). Some alternative notations have
for automatic computers, and statements of the also been listed. The introduction of new symbols
principal mathematical properties of the tabu- has been kept to a minimum, and an effort has
lated functions, particularly those of computa- been made to avoid the use of conflicting notation.
4. Interpolation
The tables in this Handbook are not provided Let us su pose that we wish to compute the
with differences or other aids to interpolation, be- value of zezg1(z)for 2=7.9527 from this table.
cause it was felt that the space they require could We describe in turn the application of the methods
be better em loyed b the tabulation of additional of linear interpolation, Lagrange and Aitken, and
functions. 1 6 ;
dmitte y aids could have been given
without consuming extra space by increasing the
of alternative methods based on differences and
Taylor’sseries.
intervals of tabdation, but this would have con- (1) Linear interpolation. The formula for this
flicted with the requirement that linear interpola- process is given by
tion is accurate to four or five figures.
For a plications in which linear interpolation fP= (l-P)fo+Pfl
is insdciently accurate it is ’intended that
Lagrange’sformula or Aitken’smethod of itera- where fo, f l are consecutive tabular values of the
tive linear interpolation3 be used. To help the function, corresponding to arguments zo, zl, re-
user, there is a statement at the foot of most tables spectively; p is the given fraction of the argument
of the maximum error in a linear inte olate, interval
and the number of function values nee ed in
Lagrange’sformula or Aitken’smethod to inter-
‘fl P= (2-30) /(%I -20)
polate to full tabular accuracy. and jpthe required interpolate. In the present
As an example, consider the following extract instance, we have
from Table 5.1.
fo=.89717 4302 fi=.89823 7113 p=.527
z ze=E,(z) z m’Ei(4
7.5 .89268 7854 8.0 .89823 7113 The most convenient way to evaluate the formula
7.6 .89384 6312 8.1 .89927 7888 on a desk calculating machine is to set and $
7.7 .89497 9666 8.2 ,90029 7306
7.8 .89608 8737 8.3 .90129 60’3 in turn on the keybbard, and carry out t e multi-
7.9 .89717 4302 8. 4 .go227 4695 plications by 1-p and cumulatively; a partial
check is then B b y the multiplier dial
reading unity. T Z k a i n
The numbers in the square brackets mean that f.az.r=(1 -.527)(.89717 4302) +.527(.89823 7113)
the maximum error in a linear inte olate is = .89773 4403.
3X and that to interpolate to the f tabular
accurac five points must be used in Lagrange’s
3 Since it is known that there is a possible error
l
and Ait en’smethods. of 3 X in the linear formula, we round off this
result to .89773. The maximum possible error in
‘ A C Aitkm O n inte OlatlOn b ItemtlOn Of ~ ~ ~ O r t i o narts
a l With.
Xw.
out Be ; ~ ofe dieerenoas, L.
Edingurgh ~ 8 t h . s, 6 7 6 &mi. this answer is composed of the error committed
INTRODUCTION XI
by the last rounding, that is, .4403X10-6, plus The numbers in the third and fourth columns are
3X10-', and so certainly cannot exceed .8X10-6. the h t and second differences of the values of
(2) La ange's formula. In this example, the ze"E,(z) (see below); the smallness of the second
relevant P
ormula is the 5-point one, given by difference provides a check on the three inte
tions. The required value is now obtains";;
f=A-z(p)f-2 +A-i@lf-i +&(p)f,+Ai@).fi linear interpolation:
+A2@)fa
Tables of the coefficientsA&) are ven in chapter .3(.89772 9757) 4- .7(.89774 0379)
25 for the range p=O(.Ol)l. #
fp=
e evaluate the
formula for p=.52, .53 and .54 in turn. Again, =.89773 7192.
in each-evaluationwe accumulate the Ad-p) in the
multi her register spce their sum is wnlty. We
now !l ave the followmg subtable. In cases where the correct order of the Lagrange
polynomial is not known, one of the prelimina
z
7.952
(2)
-89772 9757
mterpolations may have to be performed wit
polynomials of two or more different orders aa a
T
10622 check on their ade uacy.
7.953 .89774 0379
10620
-2
l
(3) Aitken's met od of iterative linear interpola-
tion. The scheme for carrying out this process
7.954 .89775 0999
in the present example is as follows:
n 2, yn=ze*El(z) vo. (I Yo. 1.
(I
vo, 1.1. (I YO.1.9.3, (I 2.-z
3 8 3 . 89823 7113 .0473
1 7. 9 .89717 4302 .89773 44034 -. 0527
2 8. 1 . 89927 7888 . 89774 48264 . 89773 71499 . 1473
3 7. 8 . 89608 8737 2 90220 2394 . 89773 71938 -. is27
4 a2 .goo29 7306 4 98773 1216 16 89773 71930 .2473
5 7. 7 . 89497 9666 2 35221 2706 43 30 -. 2527
Here
6%
6%
#fa
Here
If the quantities z,,-x and zm-x are used as 6fiiz=fi-fo9 afair=fa-fi, . . . ,.
multipliers when forming the cross-product on a 6% = 6fs1n -Sfill =fa -2f1 +fo
desk machine, their accumulation (zn-z) -(zm-s) 6y8/3= gf2- 6Yl=f3- 3f2+ 3fi-fO
in the multiplier register is the divisor to be used 6'ftin -SYa/l=fr- 4fa + 6ja- 4f1+fo
at that stage. An extra decimal place is usually and so on.
carried in the intermediate interpolates to safe- In the present exam le the relevant part of the
guard against accumulation of rounding errors. %
difference table is aa fo o m , the differences being
written in units of the last decimal lace of the
The order in which the tabular values are used
is immaterial to some extent, but to achieve the
maximum rate of convergence and at the same
function, as is customary. The sma ess of the
high differences provides a check on the function
Bn
time minimize accumulation of rounding errors, values
we begin, as in this example, with the tabular z zezEl(z) alf a4f
argument nearest to the given argument, then 7. 9 . 89717 4302 -2 2754 -34
take the nearest of the remaining tabular argu- 8. 0 . 89823 7113 -2 2036 -39
ments, and so on.
The number of tabular values required to Applying, for example, Everett's interpolation
achieve a given precision emerges naturally in formula
the course of the iterations. Thus in the present
example six values were used, even though it was f~=(1-~p)fo+%(~)~'~o+E4(~)6'fo+ . . .
known in advance that five would suflice. The +pf,+F*(P)~fi+F4(P)6%+ *
lOQf.&n=.473(89717
4302)+.061196(2 2754) -.012(34) can be used. We first compute as many of the
+ .527(89823 7113)+ .063439(2 2036) -.012(39)
=E49773 7193.
derivatives j'"'(z,)
as are significant, and then
evaluate the series for the given value of x.
An advisable check on the computed values of the
We may notice in passing that Everett's derivatives is to reproduce the adjacent tabular
formula shows that the error in a linear interpolate values by evaluating the series for x=x-l and xl.
is approximately
=xezEl(z)
Since the maximum value of IE2(p)+Fz(p)1 in the = (l+z-')f(z) - 1
range Op
<l< is %, the maximum error in a linear = (l+x-')f'(z) -z-2f(z)
interpolate is approximately + 22-3f(x).
= (1 -tz-l)f"(z)-22-?f'(2)
5. Inverse Interpolation
With linear interpolation there is no difTerence The desired z is therefore
in principle between direct and inverse interpola-
tion. In cases where the linear formula rovides x =ZO+ =8.1+.708357(.1)= 8.17083 57
an insufEciently accurate answer, two met ods are
available. We may interpolate directly, for
t P ( X ~-20)
8. Acknowledgments
The roduction of this volume has been the bibliographic references and assisted in preparing
P
result o the unrelenting efforts of many persons,
all of whose contributions have been instrumental
the introductory material.
Valuable assistance in the preparatian, checkin
in accomplishing the task. The Editor expresses and editing of the tabular material was receive%
his thanks to each and every one. from Ruth E. Capuano, Elizabeth F. Godefroy,
The Ad Hoc Advisory Committee individually David S. Liepman, Kermit Nelson, Bertha H.
and together were instrumental in establishing Walter and Ruth Zucker.
the basic tenets that served as a guide in the forma- Equally important has been the untiring
tion of the entire work. In particular, special cooperation, assistance, and patience of the
thanks are due to Professor Philip M. Morse for members of the NBS staff in handling the myriad
his continuous encoura ement and support. of detail necessarily attending the publication
B
Professors J. Todd and A. rddlyi, panel members
of the Conferences on Tables and members of the
of a volume of this magnitude. Especially
appreciated have been the helpful discussions and
Advisory Committee have maintained an un- services from the members of the Office of Techni-
diminished interest, offered many suggestions and cal Information in the areas of editorial format,
carefully read all the chapters. graphic art layout, printing detail, preprinting
Irene A. Stegun has served effectively as associate reproduction needs, as well as attention to pro-
editor, sharing in each motional detail and financial support. In addition,
the volume. Without her the clerical and typing staff of the Applied Mathe-
pletion would never have matics Division merit commendation for their
Appreciation is expressed for the generous efficient and patient production of manuscript
cooperation of publishers and authors in panting cop involving complicated technical notation.
permission for the use of their source material. &nally, the continued support of Dr. E. W.
Acknowledgments for tabular material taken Cannon, chief of the Applied Mathematica
wholly or in part from published works are
Fen
on the first page of each table. Myrtle R. Re ing-
ton corresponded with authors and ublishers
Division, and the advice of Dr. F. L. Alt, assistant
chief, as well as of the many mathematicians in
the Division, is gratefully acknowledged.
to obtain formal permission for incluiing their
material, maintained uniformity throughout the M. ABRAMOWITZ.