Handbook of Mathematical Functions
with
Formulas, Graphs, and Mathematical Tables
Edited by Milton Abramowitz and Irene A. Stegun
1. Introduction
The present Handbook has been designed to
rovide scientific investigators with a compre-
Lnsive and self-contained summary of the mathe-
matical functions that arise in physical and engi-
neering problems. The well-known Tables of
Functions by E. Jahnke and F. Emde has been
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
of mathematical properties of %e tabulated
functions. The number of functions covered has
also been increased.
The classification of functions and organization
of the chapters in this Handbook is similar to
that of An Index of Mathematical TaMa by
A. Fletcher, J. C. P. Miller, and L. Rosenhead.*
In general, the chapters contain numerical tables,
graphs, polynomial or rational approximations
for automatic computers, and statements of the
principal mathematical properties of the tabu-
lated functions, particularly those of computa-
2. Accuracy of the Tables
The number of significant figures given in each
table has depended to some extent on the number
available in existin tabulations. There has been
fc
no attempt to ma e it uniform throughout the
Handbook, which would have been a costly and
laborious undertaking. In most tables at least
five
ificant figures have been provided, and
the ta ular’intervals have generally been chosen
to ensure that linear interpolation will yield.four-
or five-figure accuracy, which suffices in most
physical applications. Users requiring higher
1 The most recent the sixth with F. m h added 811 cc-author, waa
pubhhed in 1960 b$ M~0&Hffl, U.S.A. and Teubner, Qermany.
The aecond edition with L J Comrle added aa cc-author w88 pubhbed
In two volumed In 19.62 by Ad&n-WesleY. U.S.A.. and $clentlnc Com-
puting Service LM., QmtBdtPLn.
tional importance. Many numerical examples
are given to illustrate the use of the tables and
also the computation of function values which lie
outside their range. At the end of the text in
each chapter there is a short bibliography giving
books and papers in which proofs of the mathe-
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
tables is to be found in the National Research
Council quarterly Mathematics of Computation
(formerly Mathematical Tables and Other Aids
to Computation).
The mathematical notations used in this Hand-
book are those commonly adopted in standard
texts, articularly HigLer Transcendental Func-
tions, volumes 1-3, by A. Erdblyi, W. M nus,
F. Oberhettinger’andF. 0. Tricomi (Mc raw- 8
Hill, 1953-55). Some alternative notations have
also been listed. The introduction of new symbols
has been kept to a minimum, and an effort has
been made to avoid the use of conflicting notation.
precision in their interpolates may obtain them
by use of higher-order interpolation procedures,
described below.
In certain tables many-figured function values
\
are given at irregular intervals in the argument.
An example is rovided by Table 9.4. The pur-
pose of these ta les is to furnish “key values” for
the checking of programs for automatic computers;
no uestion of interpolation arises.
T l e maximum end-fi ure error, or “tolerance”
in the tables in this If andbook is XO of 1 unit
everywhere in the case of the elementary func-
tions, and 1 unit in the case of the higher functions
except in a few cases where it has been permitted
to rise to 2 units.
n
X INTRODUCTION

3. Auxiliary Functions and Arguments

One of the objects of this Handbook is to pro-
vide tables or computing methods which enable
the user to evaluate the tabulated functions over
complele ranges of real values of their parameters.
In order to achieve this object, frequent use has
been made of auxiliary functions to remove the
infinite part of the original functions at their
singularities, and auxiliary arguments to co e with
infinite ranges. An example will make t e pro-
cedure clear.
The exponential integral of positive argument
is given by
- m cdu
Ei(z)=r U more readily interpolable function for large 2 is
=y+Inz+++-+-+22 z* . .
1 l! 2.21 331
2 fice to produce an interpolable table.
The logarithmic singularity recludes direct inter-
and z-’[Ei(z)--In 2-71,
P
polation near s=O. The unctions Ei(z)--In z
however, are well-
behaved and readily interpolable in this region.
Either will do as an auxiliary function; the latter
was in fact selected as it yields sli htly higher
$
accuracy when Ei(z) is recovered. he function
z-’[Ei(z)-ln 2-71 has been tabulated to nine
E decimals for the range 0 5 x 5 3 . For 45212,
Ei(z) is sufficiently well-behaved to admit direct
tabulation, but for larger values of 2, its expo-
nential character predominates. A smoother and
ze-”Ei(z); this has been tabulated for 2 5 2 5 1 0 .
. Finally, the range 10 1 250)is covered by use of
the inverse argument 2-l. Twenty-one entries of
ze-”Ei(z), corresponding to &=. 1 (- .005)0, suf-

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%+ *

extra row c o n k s the convergence and provides

a valuable check.
(4) Difference formulas. We use the central
difference notation (chapter 25),
and takin the numerical values of the inte ola-
tion coei3kients I%@), E 4 ), Fa@) and
from Table 25.1, we find t at k
XII INTRODUCTION

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)

1 1 With 20=7.9 and x-x,,=.0527 our computations

l~fo+Wil,that is, 3 Ifa-fi-fo+f-iI. are as follows; an extra decimal has been retained
in the values of the terms in the series to safeguard
(5) Taylor's series. In cases where the succes- against accumulation of rounding errors.
sive derivatives of the tabulated function can be
computed fairly easily, Taylor's expansion k f"' (zo)lk! (x-20) y*(zo)/kl
0 .89717 4302 .89717 4302
1 .01074 0669 .00056 6033 3
2 -.00113 7621 -.OOOOO 3159 5
3 .00012 1987 .OOOOO 0017 9
.89773 7194

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)

To estimate the possible error in this answer,

example, b Lagrange's formula to prepare a new we recall that the maximum error of direct linear
table at a &e interval in the neighborhood of the interpolation in this table is Aj=3XlO-". An
approximate value, and then apply accurate approximate value for djldx is the ratio of the
inverse linear interpolation to the subtabulated first difference to the argument interval (chapter
values. Alternatively, we may use Aitken's 25), in this case .010. Hence the maximum error
method or even possibly the Taylor's series in x is approximately 3X lO-"/(.OlO), that is, -0003.
method, with the roles of function and argument (ii) Subtabulation method. To improve the
ap roximate value of x just obtained, we inter-
interchanged.
It is important to realize that the accuracy of P
PO ate directly for p=.70, .71 and .72 with the aid
of Lagrange's 5-point formula,
an inverse interpolate may be very different from
that of a direct interpolate. This is particularly 2 zezEl(z) 6 6,
true in regions where the function is slowly 8. 170 .89999 3683
varying, for example, near a maximum or mini- 1 0151
mum. The maximum precision attainable in an 8. 171 . 90000 3834 -2
inverse interpolate can be estimated with the aid of 1 0149
the formula 8.172 .go001 3983
Inverse linear interpolation in the new table
gives
in which A j is the maximum possible error in the .9- 39999 3683= .6223
function values. '= .00001 0151
Example. Given ze"El(x) = .9, 6nd z from the Hence x=8.17062 23.
table on page X.
(i) Inverse linear interpolation. The formula An estimate of the maximum error in this result
for v is is
P = K7 -.io)/ -fo). V
I
ddxj 1x10-8
Aj/---~-=lX10-'
.010
In the present example, we have
.9-39927 7888 (iii) Aitken's method. This is carried out in the
'=.90029 7306- .89927 7888=1-8= 72 2112 .708367. same manner as in direct interpolation.
 XIV INTRODUCFION
8. Acknowledgments
The roduction of this volume has been the
P
result o the unrelenting efforts of many persons,
all of whose contributions have been instrumental
in accomplishing the task. The Editor expresses
his thanks to each and every one.
The Ad Hoc Advisory Committee individually
and together were instrumental in establishing
the basic tenets that served as a guide in the forma-
tion of the entire work. In particular, special
thanks are due to Professor Philip M. Morse for
his continuous encoura ement and support.
B
Professors J. Todd and A. rddlyi, panel members
of the Conferences on Tables and members of the
Advisory Committee have maintained an un-
diminished interest, offered many suggestions and
carefully read all the chapters.
Irene A. Stegun has served effectively as associate
editor, sharing in each
the volume. Without her
pletion would never have
Appreciation is expressed for the generous
cooperation of publishers and authors in panting
permission for the use of their source material.
Acknowledgments for tabular material taken
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
to obtain formal permission for incluiing their
material, maintained uniformity throughout the
bibliographic references and assisted in preparing
the introductory material.
Valuable assistance in the preparatian, checkin
and editing of the tabular material was receive%
from Ruth E. Capuano, Elizabeth F. Godefroy,
David S. Liepman, Kermit Nelson, Bertha H.
Walter and Ruth Zucker.
Equally important has been the untiring
cooperation, assistance, and patience of the
members of the NBS staff in handling the myriad
of detail necessarily attending the publication
of a volume of this magnitude. Especially
appreciated have been the helpful discussions and
services from the members of the Office of Techni-
cal Information in the areas of editorial format,
graphic art layout, printing detail, preprinting
reproduction needs, as well as attention to pro-
motional detail and financial support. In addition,
the clerical and typing staff of the Applied Mathe-
matics Division merit commendation for their
efficient and patient production of manuscript
cop involving complicated technical notation.
&nally, the continued support of Dr. E. W.
Cannon, chief of the Applied Mathematica
Division, and the advice of Dr. F. L. Alt, assistant
chief, as well as of the many mathematicians in
the Division, is gratefully acknowledged.
M. ABRAMOWITZ.