Professional Documents
Culture Documents
OF FUNCTIONS
Jerome Spanier
The Claremont Graduate School
Claremont. California. U.S.A.
Keith B. Oldham
Trent University
Peterborough. Ontario. Canada
Bibliography: p.
Includes indexes
1. Oldham. Keith B. D. Title.
QA331.S695 1997, 515 86.18294
ISBN 0.89116-573-8 Hemisphere Publishing Corporation
Preface lx
0 General Considerations I
The majority of engineers and physical scientists must consult reference books containing information on a variety of
mathematical functions. This reflects the fact that all but the most trivial quantitative work involves relationships that are
best described by functions of various complexities. Of course, the need will depend on the user, but all will require
information about the general behavior of the function in question and its values at a number of arguments.
Historically. this latter need has been met primarily by tables of function values. However, the ubiquity of computers
and programmable calculators presents an opportunity to provide reliable, fast and accurate function values without the
need to interpolate. Computer technology also enables graphical presentations of information to be made with digital
accuracy. An Atlas of Functions exploits these opportunities by presenting algorithms for the calculation of most functions
to more than seven-digit precision and computer-generated maps that may be read to two or three figures. Of course, the
need continues for ready access to the many formulas and properties that characterize a specific function. This need is
met in the Atlas through the display of the most important definitions, relationships, expansions and other properties of
the 400 functions covered in this book.
The Atlas is organized into 64 chapters, each of which is devoted to one function or to a number of closely related
functions; these appear roughly in order of increasing complexity. A standardized format has been adopted for each
chapter to minimize the time required to locate a sought item of information. A description of how the chapters are
sectioned is included in Chapter 0. Two appendixes, a references/bibliography section and two indexes complete the
volume.
It is a pleasure to acknowledge our gratitude to Diane Kaiser and Charlotte Oldham who prepared the typescript and
to Jan Myland who created the figures. Their skill and unfailing good humor in these exacting tasks have exceeded all
expectations. We would also like to record our indebtedness to our families for their support and patience during the
many years that the preparation of the Atlas took time that was rightfully theirs. To them, and especially to Bunny and
Charlotte, we dedicate this book.
Jerome Spanier
Keith B. Oldham
ix
CHAPTER
T
GENERAL CONSIDERATIONS
In this chapter are collected some considerations that relate to all, or most, functions. The general organization of
the Atlas is also explained here. Thus, this could be a good starting point for the reader. However, the intent of
the authors is that the information in the Atlas be immediately available to an unprepared reader. There are no
special codes that must be mastered in order to use the book, and the only conventions that we adopt are those
that are customary in scientific writing.
Each chapter in the Atlas is devoted to a single function or to a small number of intimately related functions.
The preamble to the chapter exposes any such relationships and introduces special features of the subject function.
0:1 NOTATION
The nomenclature and symbolism of mathematical functions are bedevilled by ambiguities and inconsistencies.
Several names may be attached to a single function, and one symbol may be used to denote several functions. In
the first section of each chapter the reader is alerted to such sources of confusion.
For the sake of standardization, we have imposed certain conventions relating to symbols. Thus, we have
eschewed boldface and similar typographical niceties on the grounds that they are difficult to reproduce by pencil
on paper, or with office machines. We reserve the use of italics to represent numbers (such as constants. c: coef-
ficients, a a,, a3, ... and function arguments. x) and avoid their use in symbolizing functions. Another instance
of a convention intended to add clarity is our distinction in using commas and semicolons as separators, as in
F(a.b;c;x). Elements in a string may be interchanged when they are separated by commas but not where a semicolon
serves as separator.
The symbol z(=x + iy) is reserved to denote a complex variable. All other variables are implicitly real.
0:2 BEHAVIOR
Some functions are defined for all values of their variable(s). For other functions there are restrictions, such as - I
< x s I or n = 1, 2, 3...., on these values that specify the range of each variable and thereby the domain of
the function. Likewise, the function itself may be restricted in range and may be real valued, complex valued or
each of these in different domains. Such considerations are discussed in the second section of each chapter.
This section also reveals how the function changes in value as its variables change throughout their ranges.
thereby exposing the general "shape" of the function. This information is conveyed by a verbal description, sup-
1
0:3 GENERAL CONSIDERATIONS 2
plemented by a graphical "map." All diagrams have been computer generated and are precisely drawn. Each graph
has been scaled so that the graticule spacing is either ten or fifteen millimetres (2.00 or 3.00 mm between adjacent
dots), which permits interpolation to an accuracy of two or three significant digits. Much greater precision is
available from the algorithms of Section 8.
A univariate function (i.e., a function depending only on a single variable, its argument) is generally illustrated
by a simple graph of that function versus its argument. Bivariate functions of two continuous variables are often
diagrammed by contour maps whose curves correspond to specified values of the function and whose axes are the
argument and the second variable.
In discussing the behavior of some functions it is convenient to adopt the terminology quadrant. There are
four quadrants defined as illustrated in Figure 0-1.
FIG 0-1
x <0 x>0
f (x) >0 f (x) >0
second first
quadrant quadrant
........................................
third fourth
0
quadrant quadrant
f(x)<0 f(x)<0
X <0 x>0
0:3 DEFINITIONS
Often there arc several formulas relating a function to its variable(s), although they may not all apply over the
entire domain of the function. These various interrelationships are listed in the third section of each chapter under
the heading Definitions even though, from a strictly logical viewpoint, some might prefer to select one as the
unique definition and cite the others as "equivalences."
Several types of definition are encountered in Section 3. For example, a function may be defined:
(a) by an equation that explicitly defines the function in terms of "simpler" functions and algebraic operations;
(b) by a formula relating the function to its variable(s) through a finite or an infinite number of arithmetic or
algebraic operations;
(c) as the derivative or indefinite integral of a simpler function;
(d) as an integral transform of the form
where g is a function having one more variable than f. to and t, being specified limits of integration;
3 GENERAL CONSIDERATIONS 0:5
(e) through a generating function, G(xt), that defines a family of functions f;(x) via the expansion
0:3:3 F(f(x)) = x
Some definitions generate two or more values of a function from a single value of the argument. The Atlas
minimizes this complication by concentrating, wherever possible, on single-valued functions.
0:5 INTRARELATIONSIIIPS
An equation linking the two functions f(x) and g(x) is an interrelationship between them. In contrast, one speaks
of an interrelationship in case there exists a formula that provides a link between values of a single function at
two or more values of one of its variables, for example, between f(x,) and f(x:). In this Atlas intrarelationships
will be found in Section 5 of each chapter, interrelationships mainly in Sections 3 and 13.
An equation expressing the relationship between f(-x) and f(x) is called a reflection formula. Less commonly
there exist reflection formulas relating f(a - x) to f(a + x) for nonzero values of a.
A second class of intrarelationships are translation formulas: these relate f(x + a) to f(x). The most general
translation formula, in which a is free to vary continuously, becomes an argument-addition formula that relates
f(x + y) to f(x) and f(y). However, many translation formulas are restricted to special values of a such as a = I
or a = nw; the relationships are then known as recurrence relations or recursion formulas. Such relationships are
common in bivariate functions; a recursion formula then normally relates f(v;x) to f(v - I;x) or to both f(v - I;x)
and f(v - 2;x).
A very general argument-addition formula is provided by the Taylor expansion:
0:6 EXPANSIONS
The sixth section of each chapter is devoted to ways in which the function(s) may be expressed as a finite or infinite
array of terms. Such arrays are normally series, products or continued fractions.
Notation such as
is used to represent a convergent infinite series, where g is a function of j and x. Unless otherwise qualified, 0:6:1
implies that, for values of x in a specified range, the numerical value of the finite sum
0:6:2 g(0;x) + g(l;x) + g(2;x) + + g(j:x) + + g(J;x)
can be brought indefinitely close to f(x) by choosing J to be a large enough integer.
Frequently encountered are convergent series whose successive terms, for sufficiently large j, decrease in mag-
nitude and alternate in sign. We shall loosely call such series alternating series. A valuable property of such
alternating series that is often used in this Atlas enables the remainder after a finite number of terms are summed
to be estimated in terms of the first omitted term. When l (- I)'g(j;x) is used to represent an alternating series,
this result:
which is termed a concatenation or nested sum. Such nested sums will also frequently be found in algorithms
appearing in Section 8.
The infinite product notation
approaches f(x) indefinitely closely as J takes larger and larger integer values.
The notation
0:0.8 ao+---...
aI a, a,
Rt+ 132t P3+
is a standard abbreviation for the continued fraction
GENERAL CONSIDERATIONS 0:7
Q1
ao +
P, + a,
0:6:9 P2+ a3
P3 + a.
P.+...
in which each as and P3 may denote constants or variables. Continued fractions may be infinite, as denoted in 0:6:8,
or finite (or "terminated"):
0:6:10 as+
P,+ P2+ P3+
---...-- a,
P,_,+ P,
Of great utility in working with continued fractions is the equivalence
0:6:11
(1, a, a3
... a ao
Yia, YiY2a2 Y2Y3a3
P,+ P:+ 03+ P. - 7101+ Y2P2+ Y,P,+ Y.P.
Demonstrating the interchangeability of continued fractions and power series is the identity
X.
I x x2 1 aox a,x a2x a,-,x
0:6:12 - + - + - + +
ao aoa, a0a,a2 a0a,a2 a. a0- a, + x- a2 + x- a3 + x a+x
0:7:2 -
df
dx dr
d'f
0, - (x,,) > 0 f(x,) = minimum of f(x)
and
df d:f
0:7:3 (s M) = 0 , dx2 (xM ) < 0 ( .r M ) = maximum of f(x)
dx
A zero of a function is a value of its argument at which the function vanishes; that is, if
0:7:4 f(r) = 0 then r = a zero of f(x)
Equivalent to the phrase "a zero of f(x)" is "a root of the equation f(x) = 0." A double zero or a repeated root
occurs at a value r of the argument such that
df
0:7:5 f(r) (r) = 0 r = a double zero of f(x)
dx
The concept extends to multiple zeros; thus, if
df d'f
0:7:6 f(r)=-(r)-
dx dx"
(r)=0 r = a zem of f(x) of multiplicity n + I
A value r of x that satisfies 0:7:4 but not 0:7:5 corresponds to a simple zero or a simple root.
0:8 GENERAL CONSIDERATIONS 6
(a) provision for loading the argument and other input parameters, if any:
(b) indication of the restrictions, if any, on the input values:
(c) a list of the constants and variables used by the algorithm;
(d) a list of easily understood sequential commands, such as "Setf = 0" or "Replace g by N g/(1 + x)";
(e) labels in the form of parenthesized numbers, such as "(2)," preceding certain commands;
(f) commands in the form "if . . . go to (#)" where (#) is a label and . . . is some condition, it being understood
that if the condition is not met, one proceeds to the next command in sequence;
(g) provision for output; and
(h) test values to be used in confirming that a program derived from the algorithm is functioning correctly.
The test values supplied with each algorithm have been carefully chosen to exercise the algorithm maximally.
Test outputs are usually reported with two or more terminal digits italicized. A correctly programmed algorithm
should always generate a test output that reproduces at [east those digits that are not italicized. On the other hand,
italicized digits may fail to be reproduced by a correctly programmed algorithm. This is because changes in these
digits correspond to errors that lie within our permitted range of ±6 x 10-8. For example, our algorithm from
Section 43:8 lists the test value f(0.6) = 1.48919225; however, an output of 1.48919228, having a relative error
of only 2 x 10-r, is no cause for concern.
7 GENERAL CONSIDERATIONS 0:10
The techniques employed in the design of the algorithms for this Atlas are so diverse that it is inappropriate
to attempt to summarize them here. However, each algorithm is accompanied by at least a brief synopsis of its
mode of operation.
Throughout the Atlas there are also a number of general-purpose algorithms. These arc recapitulated in Ap-
pendix A, which also contains several additional utility algorithms of wide applicability.
0:9 APPROXIMATIONS
Entries in this section enable function values to be calculated for stated ranges of the argument. Often, the precision
is specified to be "8-bits," which implies
f(x)
(x) - f(x )
0:9:1
-2_B
c "
1(x)
(i.e., a relative error of no more than about ±4 x 10-'). where i(x) is the value given by the approximation and
f(.r) is the value of the function.
0:10:1 -
dx
df
exists and is reported in Section 10. The derivatives of function combinations such as f(.r)g(x), f(x)/g(x) or f(g(.r))
are seldom reported because they may be readily evaluated via the product rule:
0:10:2
d dg
- f(x)g(x) = 1(x) - + g(x)
dx dx
df
d.r
-
--_----
the quotient rule:
d f(x) 1 df f(x) dg
0:10:3 g--(x)
d.r g(.0 g(x) dx dx
or the chain rule:
d df dg
0:10:4 f(g(x)) =
dx d dx
1?
g = g(x)
If f(x) is differentiated n times, the nth derivative
d"f
0:10:5
d.t"
is generated. and expressions for this result are sometimes reported. The nth derivative of a product is given by
the Leibni: theorem:
0:10:6
d"
- f(x)g(x) = f(x)
dx"
d"g
dx
+1!n-dfdx-d"-'g
- + n(n2!- 1) d2f
dx"
d='-g
dx' dx"-'
-
+ - + n1!- d"-'f - + - g(.r)
dg d"f
dx"-' dx dx"
_ n d"-'f d'g
- J/ dx'-' dx'
0:11 GENERAL CONSIDERATIONS S
where (,) denotes a binomial coefficient (Chapter 61 and n! the factorial function [Chapter 2]. Another useful
operation bearing the name of Leibniz gives a rule for differentiating an integral:
(I U) W
d d dt d
0:10:7
dx f,
Ns,
f(x;t)dt =
J
`' - f(x;t)dt + dx f(x;t,(x)) -
dx
f(x;to(x))
Reference books often express the results of indefinite integration in a form such as
where 4b(x) is a function that gives f(x) on differentiation (i.e., d4s/dx = f(x)), and c is an arbitrary constant. To
achieve closer unity with the representation
0:10:9 I , f(t)di
JJ.o
for the with differintegral of f(x) with respect to x, using x0 as the lower limit. The value of v is unrestricted: it may
be positive or negative, integer or noninteger. Differintegration, as its name suggests, generalizes the operations
of differentiation and integration so that 0:10:1. 0:10:5 and 0:10:10 are the v = 1. v = n and v = -1 instances
of 0:10:12. Note that the diffcrintegral 0:10:12 is independent of the lower limit xo only when v = 0, I. 2.....
For further information on differintegration, the reader is referred to a monograph on the subject (Oldham and
Spanier]. The symbol 0:10:12 collapses to
d"f
0:10:13
d
when xo is chosen to be zero.
The most important noninteger order instances of differintegration are generated when v or v = -; and
are termed semidifferentiation and semiintegration, respectively. Some examples of semiderivatives and semiin-
tegrals are reported in Sections 10, and most often with a lower limit of zero or minus infinity.
An algorithm that generates approximate values of the differintegral 0:10:12 may be found in Section A:7.
In Sections 11, however, the effect of replacing the real argument x by the imaginary iy or the complex x +
iy is often reported. The real and imaginary components of f(x) or f(x + iy) are sometimes exposed as well.
0:12 GENERALIZATIONS
Functions can be arranged in a hierarchy in which lower members are subsumed by more general higher members.
In the chapter dealing with the f(x) function, special cases of f(x)-that is, functions lower in the hierarchy-are
reported in Section 4. Conversely, functions that occur higher than f(x) in some hierarchy are reported in Section
12; that is, the functions reported here are ones of which f(x) is a special case.
1:1 NOTATION
Constants are also known as invarianrs and are represented by a wide variety of symbols, mostly letters drawn
from the early members of the English and Greek alphabets.
1:2 BEHAVIOR
The graphical representation (see Figure 1-1) of the constant function fix) = c is a horizontal line, extending
indefinitely in the x - =x directions.
............................. 0
1:3 DEFINITIONS
The constant function is defined for all values of its argument x and has the same value. c, irrespective of x.
11
1:4 THE CONSTANT FUNCTION c 12
1:5 INTRARELATIONSHIPS
Being relations between function values at different values of the argument, intrarelationships are of no consequence
for the constant function.
1:6 EXPANSIONS
A constant may be represented as a finite sum by utilizing the formulas for an arithmetic series:
1:6:1 a=
c -b
J-o
a geometric series:
j c(R - 1)
I.62 c=a+aR+aR2+...+aR+='nRJ 01=
or an arithmetic-geometric series:
1:6:3
In these formulas R and b may take any values and J may be any positive integer.
Any constant greater than ! may be expanded as the infinite geometric sum
1:6:5 [1+(Cc
J
Similarly, a constant is expansible as the infinite continued fraction
a a a
1:6:6 C = - - - ...
R+ R+ R+
[see Section 0:6] in 8 variety of ways, some of which are indicated in Table 1.6.1.
Table 1.6.1
0 9 Constraint
c 1 - c - I <- c< I
c=
0<cs<I
C
+c c2:
r-c c c:
1:7:5 i -In(n))=0.5772156649
The latter is also known as Mascheroni's constant and is often denoted by the symbol C. There are many alternative
formulations of these four constants [see Gradshteyn and Ryzhik. Chapter 0. for some of these]. Also of widespread
occurrence throughout this Atlas is the ubiquitous constant
1:7:6 U = common mean [Section 61:81 of I and 1/f = 0.8472130848
The most important family of rational constants is the family of natural numbers, n = I. 2, 3. ... [see Section
1:141. Other families that occur principally in coefficients of series expansions are the factorials n! [Chapter 2],
Bernoulli numbers B. [Chapter 41 and Euler numbers E, [Chapter 5].
1:9 APPROXIMATIONS
None are needed for the constant function.
1:10:2 cdt = cx
and
1:10:3
f cdt=c(x,-xo)
respectively.
The results of semidifferentiation and semiintegration [Section 0:101 with a lower limit of zero are
d1/' c
1:10:4 C
dx'l: VIIX
and
d_'/2 x
1:10:5 dx-ii: c = 2c -
Differintegration [see Section 0:101 with a lower limit of zero yields
d' _
cx-'.
1:10:6 c
dx' f(1 - v)
where r is the gamma function [see Chapter 431. In fact, equations 1:10:1, 1:10:2, 1:10:4 and 1:10:5 are the v =
1, -1, 1 and -i instances of 1:10:6.
A'+B'
:
A+iB
Powers may be handled making use of the identity
1:11:5 (a + i13)"'B = exp[(A + iB)Ln(a + i13)]
where exp and Ln are the exponential and logarithmic functions treated in Chapters 26 and 25, respectively. Ap-
plying equations 25:11:1, 25:11:2 and 25:11:3 produces the multiple-valued function
1:11:6 (a + i13)4 = (a2 + 132)A'2[cos(d) + i sin(4)]exp[-B(8 + 2kar)]
where k is any integer and
sgn(13)arccot(a/I13I)
R*0
1:11:7
13=0
Setting k = 0 in equations 1:11:6 and 1:11:8 defines a single-valued function. When this is done and when
A + iB = v, a real number, the result is
1:11:9 (a + ip)' = (a' + p'),12[cos(v9) + i sin(ve)[
1:12 GENERALIZATIONS
The constant function is the special b = 0 case of the linear function discussed in Chapter 7.
Whereas the constant function has the same value for all r, the related pulse function is zero at values of the
argument outside a "window" of width h. and is a nonzero constant, c, within this window (see Figure 1-2):
y 0A 1100.%
1t0
FIG 1-2 :
C
p (ct he x-a
it
0 x<a--
2
h
x>a+-
2
The value of a establishes the "location" of the pulse. while c and it are termed the pulse height and pulse width,
respectively. The pulse function can be expressed as
1:13:2
1/ a+hhl-ux-a-h)
p(c;h;.r-a)=c ul.r -
/ h
+1b
i`+, iy+ry
C2
FIG 1-3 :
C1
C9
............. Co
1:14:6
(n 1)/2 n=1.3.5,...
j-1 -n12 n = 2, 4, 6, ...
Factorials occur widely in function theory: for example. in the power series expansions of many algebraic and
transcendental functions. They also play an important role in combinatorics [see Section 2:141.
2:1 NOTATION
The factorial function of n is symbolized Lit in some older literature. The symbol fl(n) is occasionally encountered.
2:2 BEHAVIOR
The factorial function is defined only for nonnegative integer arguments and is itself a positive integer. Figure
2-I (with a logarithmic vertical scale) and the table of rounded values (Table 2.2.1) demonstrate the explosive
increase of n! as n increases beyond n = 2.
c
f c (I
' <o oa yh
(I <
'o s^
'00 400
10
FIG 2-1 Table 2.2.1
:..............:....:..............:....:....: 105
0 I
10 4 x 10'
.:....:....:....:....:....:....*... ...: ...: 109 20 2 x 10"
30 3 x l0`
:....:....:... :.......nl+t .........:....:....:.100 40 8x I0'
30 3 x I0°'
60 8 x 10"
:....:......... :....:....
....
:....:....... 70 I x loll
10 80 7x 10"'
90 IxIO'°
100 9 x 10,!
FE
2:3 THE FACTORIAL FUNCTION n! AND ITS RECIPROCAL 20
2:3 DEFINITIONS
The factorial of the positive integer n equals the product of all positive integers up to and including n:
2:3:1 n=1,2,3,...
J.,
This definition is supplemented by the value
2:3:2 0! = 1
conventionally accorded to factorial zero.
The exponential function [Chapter 261 is a generating function [see Section 0:3] for the reciprocal of the factorial
t"
2:3:3 exp(r) = i -
-0n!
2:5 INTRARELATIONSHIPS
The most important property of the factorial function is its recurrence
2:5:1 (n+ l)!=n!(n+1) n=0,1,2,...
which may be iterated to produce the argument-addition formula
2:5:2 (n+m)!=n!(n+ 1)T n,m=0. 1,2....
where (n + I) is a Pochhammer polynomial [see Chapter 18]. Formula 2:5:2 leads to an expression for the ratio
of two factorials. An alternative expression is
n.
2:5:3
(n - m)! i-0
where is a Stirling number of the first kind, of which a short table is included in Chapter 18.
Because reciprocals of factorials occur frequently as coefficients of power series, many expressions exist for
infinite sums involving these reciprocals. For example:
2:5:4 -+-+-+
I
0!
1
1!
1
2!
1
i-=exp(1)=2.718281828
J-0 j!
(1) /l
2:5:5
!
2
+I-) +II \111/
11 Z
21
r
+==1(2)=2.279585302
i-o (j!)'
2:5:6 + + + _ = cosh(l) = 1.543080635
0! 2! 4! ra (2#
1 1 1 1
2:6 EXPANSIONS
Stirling's formula provides an asymptotic expansion [see Section 0:61
12n
1
288n'
139
51840n'
J
n--
for the factorial n! as n tends to infinity. This formula and a similar one for the logarithm of n!:
2:6:2
1
ln(n!)-nln(n)-n+-ln(2nn)+
2
--
12n
I I
360n' 1260n'
1
-
n--. c
based on definition 2:3:1, is precise, but may be tediously slow. Alternatively, one may use a second algorithm
which is based on exponentiating 2:6:2, truncation after the n-' term and rounding to the next higher integer. This
algorithm has a precision [see Section 0:8] better than 6 x 10-". Because of identities 2:12:1 and 2:12:2, the
algorithms of Sections 18:8 and 43:8 may also be used to calculate factorials.
2:9 APPROXIMATIONS
Based on the asymptotic expansion 2:6: I is the approximation
2:12 GENERALIZATIONS
The factorial is a special case of Pochhammer's polynomial W. [Chapter 18]:
2:12:1 n!=(I)" n=0,1.2....
and of the gamma function [Chapter 43]:
2:12:2 n! = r(n + I) n=0,1.2....
while for odd n it may be expressed in terms of factorials, or as a gamma function (Chapter 43J or as a Pochhammer
polynomial [Chapter 18]
2:13:3 n!!=
n. =2"/' -r1I+-J=2 n(21" n=1.3,5....
2
Of frequent occurrence [for example in Sections 6:4, 32:5, 61:6 and 62:121 is the ratio (n - 1)!!/n!! of the double
factorials of consecutive integers. For odd n the ratio is expressible by the integral
I)!]z
2:13:4
n!!
=-In! LLL \ 2 /
! =
o
sin"(t)dt n = 1,3.5,...
2 I evenn -rx
2:13:6
(n-I)!!_. V rn 4n32.22 ]
n!! n
2n
1--+--
4.
1
32n
1
J
Finite sums of such ratios obey the simple rule
1 3 15 (2n - I)!! (2j - 1)!! (2n + I)!!
2:13:7 1 + - + - + - + + _ ' = n = 0, 1, 2,...
2 8 48 (2n)!! j=o (2j)!! (2n)!!
and there is the related infinite summation due to Ross:
1 3 15 105 (2j - I )!!
2:13:8 + + = In(4)
2 + 16 144 + 1536 ' " = j(2j)!!
The triple factorial is defined analogously
I n=-2.-1,0
X ! ! n(n - 3) x (n - 6) x x 7 x 4 x I n = 1, 4, 7,...
2:13:9 n...
n(n - 3) x (n - 6) x . x 8 x 5 x 2 n = 2, 5, 8,...
and finds application, for example, in connection with Airy functions [Chapter 56J. The extension to a quadruple
factorial n!!!! is obvious; it is useful in Sections 43:4 and 59:7.
2:14:4 v." i = -
;_a (m - Yj!
2:14 THE FACTORIAL FUNCTION n! AND ITS RECIPROCAL 24
Table 2.14.1
I 0 0 0 0 0 0 0 0 0 0 0 0 0 0
m= I o 1 1 1 I I 1 I 1 I 1 I 1 I 1
Some examples of Stirling numbers of the second kind are given in Table 2.14.1. Others may be calculated by the
exact (but rather slow) algorithm
As detailed in Section 3:14, the four number families !;(n). X(n). 11(n) and [3(n) occur as coefficients in many power
series expansions. In this context it is only positive integer orders. n = 1. 2. 3, .... that are encountered, and this
chapter therefore emphasizes these cases. However, one is able to extend the definitions of all four functions to
accept noninteger and negative orders, and these possibilities are also addressed here.
The first three functions are interrelated by the simple proportionalities
3:0:1
f(v) - >,(v) - r)(v)
2" 2' - 1 2'-2
and by the consequential identity
3:0:2 i;(v) + q(v) = 2X(v)
but there are no corresponding relationships involving 13(v). Because they are so easily related to 7,(v) via 3:0:1.
few formulas for M v) and r1(v) are exhibited in this chapter.
3:1 NOTATION
The numbers I,(n). Mn). lI(n) and p(n) do not appear to have acquired definitive names: we shall call them :eta
numbers, lambda numbers, eta numbers and beta numbers. When the order is unrestricted, we adopt the symbolism
f,(v), A(v), q(v) and (3(v) and the names :eta function, lambda function, eta function and beta function. The last
should not be confused with the complete beta function [Section 43:13] or the incomplete beta function [Chapter
62) to which it is totally unrelated.
The symbols o and L have been used, respectively, in place of 11 and 0. Riemann's name is associated with
the zeta function, which is often known as "Riemann's function" or 'Riemann's zeta function." We avoid these
names to prevent confusion with the bivariate function of Chapter 64, with which Riemann's name is also commonly
associated.
3:2 BEHAVIOR
The zeta and lambda functions are infinite only at v = I, whereas rl(v) and 13(v) are always finite. Figure 3-I maps
the behaviors of the four functions. All approach unity rapidly as the order v increases. For v c 0. the four functions
u
3:3 THE ZETA NUMBERS AND RELATED FUNCTIONS 26
a j yeti O
4 i '1 -, 9
,
i~
.......................... ..................3
are oscillatory, with ever-increasing amplitudes as v -. -x. The zeros of Z(v), x(v) and q(v) occur when v is an
even negative integer, whereas 13(v) = 0 at v = -1. -3. -5.....
3:3 DEFINITIONS
The four functions may be defined by the definite integrals
1 t"-'dt
3:3:1 ;(v) = v>1
r(v) Jo exp(r) - I
v>I
3:3:2 X(v) _
2r(v) J t""' csch(t) dr
3:3:3 q(v) = r(v)
I
f 1-1d,
exp(r) + I
v>0
27 THE ZETA NUMBERS AND RELATED FUNCTIONS 3:5
and
3:3:5 Y(v)=lim
I I I ll =limYj"
'
J-= 2" 3 J -=
and apply generally for v > I in the cases of i(v) and A(v), or for v > 0 in the q(v) and p(v) cases. However,
when sufficient additional terms are included in series 3:3:5, to give
I l 1 1 1 v v(v + 1)(v + 2)
3:3:6 ;(v) = lim I+ + + + + + +
J-= 2" 3` (J - 1)" (v - 1)J" ' 2J" 121"'' 720J""
the limit provides a definition of the zeta function for any order v: positive, negative or zero. The general expression
for the kth appended term in 3:3:6 involves functions from Chapters 2, 4 and 18 and is (v)j_1B5/(k!Jr*'-'). It is
necessary to include only those appended terms for which v + k - I is negative or zero, but a few extra will speed
the approach to the limit. The corresponding definition of the beta function is
3:3:7 p(v)=lim
J- =
I--+--...-
3" 5"
1
+-+
I
6J"3
+...1
the kth appended term being (v)t_,2'(2' - I)BR/(2k!J'-"-') in this case, J being a number equal to 4n + I where
n is a natural number.
3:5 INTRARELATIONSHIPS
The zeta and beta functions satisfy the reflection formulas
2r(v)4(v) V'Tr
3:5:1 (1 - v) = cost -
(2a)" 2
involving the gamma function [Chapter 43) and the functions of Chapter 32.
With f(n) representing any one of the four numbers I,(n), Mn). 11(n) or (3(n). one may sum the infinite
series E(-1)"f(n)/n, as well as the series of complements E[I - f(n)], E(-1)"(1 - f(n)]. E[I - f(n)]/n and
F(-1)"[1 - f(n)]/n. With the lower summation limit taken as n = 2, these sums are as shown in Table 3.5.1.
These sums involve the logarithmic function [Chapter 25] of various constants. Archimedes number it (Section
1:7), Euler's constant y (Section 1:7) and the ubiquitous constant U = rr(.'-)/V ar = 0.8472130848 [see Section
1:7). Also listed in Table 3.5.1 are the sums E(- I)"f(n). Strictly, these particular series do not converge, the
tabulated entries being the limits
1
Table 3.5.1
f=; 1 Y -1 -2
Y- 1 1- IM2) - Y
7 2 _
f=A In(2) + In (-) In(2) - I - - In(2)
Y
- In(\n) - I I-In4)-Y
2 \'n. 2 2 \\n 2
In (4 3 8
In(4) - I In(4) - 1 z - In(4) In(n) - I I - In(-)
f-d
r,
4
-In(V2)
n
--
4
V_`
In(-I
U
-+In(\2)-I
IT
4
-+In(\2)--n4
I
2
-+In(-)-I
n
4
n
.\'8U
1-In(V2U)--4
n
3:6 EXPANSIONS
The series
3:6:1 C(v)+2 +3 1 1
v>I
1 1
3:6:2 v>1
3" 5"
_
3:6.3 TI(v)=1_' 1 1
v>0
v>0
] 1
3:6:4
3'' 5'
are the most useful representations of the four functions and serve as definitions of the zeta and lambda numbers,
l(n) and X(n), for n = 2, 3, 4, ... as well as for the eta and beta numbers, q(n) and [3(n), for n = 1. 2. 3, ....
Zeta and lambda functions are expansible as the infinite products
1 1 1 I 1
3:6:6 v>1
I -3-''1 -5-"1 -7-"l - 11-
where a; is the jth prime number.
Table 3.7.1
WY) 0
5
0 --1
0
1 rr
G
rr'
32 A(4)
- 2 d
51 by equation 3:13:2. For negative integer orders, the zeta and beta functions are related much more simply to
the Bernoulli and Euler numbers; thus:
and
The second algorithm is much more sophisticated, being designed to calculate P(v), 4(v), a(v) or q(v) according
to the value given to a code c that is input prior in v. Because C(l) = a(1) = ±-, the algorithm returns the value
10" when c = 0 or I and v = 1. Otherwise, the absolute algorithmic accuracy is better than t6 x 10-1 for all
four functions and for all orders. For v ? i, the algorithm utilizes equation 3:3:7 (or the X-analog of 3:3:6, together
3:8 THE ZETA NUMBERS AND RELATED FUNCTIONS 30
with 3:0:1) with J set equal to 13 and with appended terms up to k = 9. For v < ;, P(1 - v) or M(1 - v) is
computed and this value is transformed into 0(v) or a(v) (and hence to 4(v) or n(v) via 3:0:11 by use of equation
3:5:2 or the k-analog of 3:5:1. The gamma function r(I - v) needed for this transformation is computed by a
routine described in Section 43:8. To avoid "divide by zero" errors, special measures are adopted if v = 0 or 1.
If v - go to (3)
Replace v by I - v
Set w = v
ifc<0goto(I) Input restriction: c is a code
Replace f byf[(2 - 2")/(2" - 1)] cos(90v)
Go to (2)
that must equal - I , 0, 1 or
2, as follows:
(1) Set f = 2 sin(90v) code, c function
(2) Replace f by f/w -I (3(v)
Replace w by w + I 0 ?;(v)
If w<3 oto(2) I k(v)
2 I 2 q(v)
Setg= 1+-7w'
(3%-
g
fl/ 30w2
Replace g by + wlIn(w) - I]
12w
Replace f by [f e%p(g)\ 2*w]/n"
(3) If c < 0 go to (4)
Set c = 1
(4) Set J = 13
Set k = I1
Set g = 0
(5) Replace g by 8g/(297 - 92.6k + 13k2 - 0.6k')
Replace k by k - 2 Test values:
3:9 APPROXIMATIONS
The formulas
3:9:1 i(v) = 1+ 2-'+ 3- 8-bit precision v? 5
3:9:2 X(v) - I + 3 8-bit precision v? 4
3:9:3 TI(v) = 1 - 2 8-bit precision v? 5
3:9:4 0(v) - I - 3-" 8-bit precision va4
are based upon the truncation of expansions 3:6:1 through 3:6:4 and apply for large positive orders. For large
negative orders
v /va`
3:9:5 {(-v) _ -2 sins 2 I
(V)'141
\ sin(Z
/
3:9:6 a(-v) rl(2 v)
_
V -
eXp(-v) l V-.
vl". exp(-v)
(Vir 2)
3:9:7 0(-v) - 2 1 v -. x
In the vicinity of v = I. the zeta function is well approximated by
1 1
but not as established functions. At v = 0 the derivative of the zeta function equals -InV 22rr.
3:12 GENERALIZATIONS
The four functions of this chapter are special cases of the Hurwitz function of Chapter 64. Thus:
3:12:1 {(v) = (v:I)
3:13 THE ZETA NUMBERS AND RELATED FUNCTIONS 32
1)
3:12:2 M(v) = 2-% v, 2
1
3:12:3 *v) = 2 " v: 2) - ;(v:I) = J(v:1)
The last pair of equations shows that the bivariate eta fu nction [see Section 64:13] is also a generalization of the
eta and beta functions.
-1<x<1
7rx "_,
3:14:4 sec(
)-
3:14:5 coth(trx) =
1rx
- ?1 E (-1)"{(2n).rzn
az "_,
- l <x< I
3:14:6 csch(ax) _ -
-arxI + 7rx2 "_, (-1)"i(2n)x2w -1 <x< I
33 THE ZETA NUMBERS AND RELATED FUNCTIONS 3:14
3:14:7 tank
(ax)
2
=
--4
Trx "_,
(-1)"a(2n).iy' -1 <x< I
and
Bernoulli numbers constitute a family of rational numbers that occur principally as coefficients in power series.
4:1 NOTATION
There are two systems for indexing and assigning signs to Bernoulli numbers. Unfortunately, both systems are in
widespread use, although the one we adopt is more generally used than its rival. Table 4.1.1 compares the two
systems. Some authors employ both systems, using B. for one set of Bernoulli numbers and some modified sym-
bolism such as B: or B" for the other. These latter are sometimes called auxiliary Bernoulli numbers.
4:2 BEHAVIOR
We define B. for all nonnegative integers n. The accompanying map, Figure 4-1, plots the first sixteen Bernoulli
numbers.
In our notation, all Bernoulli numbers with odd index, except B,, are zero. All B. for which the index n is a
multiple of 4, except Bo, are negative rational numbers. All other even-indexed Bernoulli numbers are positive
rational numbers. The absolute values of the Bernoulli numbers of even index. IB,,,I, acquire a minimum value of
,A when 2n = 6. Table 4.2.1 shows that the magnitude of even-indexed Bernoulli numbers increases rapidly with
increasing even n.
Ba is the only Bernoulli number that is a nonzero integer but
2(n + 10.
4:2:1 = odd integer n = 2, 4, 6...
(n/2)!2"1'
4:3 DEFINITIONS
The Bernoulli numbers are defined through the generating function
x _ e
4:3:1
exp(x) - l - B" n!
m
4:4 THE BERNOULLI NUMBERS, B. 36
Table 4.1.1
Our Rival
Value system system
Bo
-1
B,
C 0 C *C Table 4.2.1
....:....:......1.0
B, B, .*
6
Bernoulli Rounded
0 B, FIG 4-1 ' : number value
-I
B. -B: O. S B," 8 l0-:
30 B -5 x 10
0 B 6 x 10'
B,
*
B. g -2 x 10'
B. B, B. 8x10"
42 B. -2 x 10"
0 B; B. 3 x 10-
B., -2 x 10"
B.,, 4 x 10'°
30
.............. -0.5 B,,,, -3 x 10"
4:5 LNTRARELATIONSHIPS
Bernoulli numbers satisfy the recursion formula
-1 B,
4:5:1 B"=-n!Z n=2,3.4....
i-a j!(n - I - j)!
with Ba = 1. Recursion 4:5:1 is easily derived from the more compact expression
4:6 EXPANSIONS
Even-indexed Bernoulli numbers are expansible as
2n r
4:6:1 B. = (-1)a+z12 (22)" 1+ + + (-1)1;212 2n! i (2!'27) n = 2, 4, 6, ...
1 2'
- 3" J=1
If the multiplier (-1)1"+2)2 is replaced by - cos(nw/2), the expansion is valid for all values of n except 0 and 1.
-1 1
0
-I I -i 5 -691 7 -5617 43867 -17"611
0 0 -0 0 0 0 0 0 0
2 6 xr 42 m 66 2770 6 510 798 330
4:9 APPROXIMATIONS
Based on expansion 4:6:1 the approximation
1 1 +212
227.1
4:9:1 B. = ( I + 2" 8-bit precision n = 6, 8, 10, ...
(227)"
is useful for all even n exceeding 4. It may be supplemented by values drawn from the table in Section 4:7. The
approximate recursion formula
4:9:2 B,-k k- 2 1
B2A_2/1T2 8-bit precision k = 5, 6, 7, ...
The Bernoulli number B. has been defined only for real integer n.
4:12 GENERALIZATIONS
Bernoulli numbers are the special x = 0 cases of the Bernoulli polynomials B.(x) (Chapter 191
4:12:1 B.=B,(0) n=0,1,2,...
Apart possibly from sign, the Bernoulli numbers are also the x = I values of the Bernoulli polynomials
4:12:2 B,=(-1)"B,(I) n=0.1.2....
Bernoulli numbers are closely related to Euler numbers [Chapter 51 and to the number families discussed in Chap-
ter 3.
In addition to their roles as coefficients in power series expansions of trigonometric and hyperbolic functions (see
Chapters 28-34] Bernoulli numbers occur as coefficients in the F_uler-Maclaurin formula, which provides a valuable
link between a definite integral and a sum. To exhibit this formula, let x°, x x:...., x, be evenly spaced arguments
withx,_1 -x,=handj=0. 1,2,...,J- 1. Then
J-1 1,
h h2 df df
h f(x;) - J a f(1) dr - (f(xj) -f(xo)] + [dx (x,) - d ( X 0 )
12
Euler numbers occur as coefficients in the power series expansions of the secant [Chapter 331 and hyperbolic secant
[Chapter 291 functions.
5:1 NOTATION
As with Bernoulli numbers (Section 4:1 J, there are (at least) two notational systems in use for Euler numbers.
Table 5.1.1 explains the differences between the two systems. Some authors use both systems and introduce sup-
plementary notation such as E, or E. to distinguish from E. The former symbols are sometimes called aueiliary
Euler numbers.
5:2 BEHAVIOR
The Euler number E, is defined for all nonnegative integer index n. All Euler numbers are themselves integers.
Euler numbers of odd index are invariably zero. Even-indexed Euler numbers are positive integers, or negative
integers, according as n is, or is not, a multiple of 4. In the accompanying map, Figure 5-I. either E. (red points)
0
.........
<1 (11
Table 5.1.1
1030
. -E *.
.........*........1010
-1
o
E;
E,
-E,
5 E. E;
0 E,
* -61 F, -E,
0 E.
**** ............1 1385 F, E.
39
5:3 THE EULER NUMBERS. E. 40
or -E" (green points) is plotted logarithmically versus n for even n values. Note the very rapid increase in the
absolute value with increasing even n.
The least significant digit of each negative Euler number is a "1"; that is:
1
5:2:1 frac(-E"/10) = 10 n = 2, 6, 10, ...
where frac denotes the fractional value function [Chapter 9]. The least significant digit of each positive Euler
number, except E0, is a "5"; that is:
1
5:3 DEFINITIONS
5:3:1 sech(t) E. r
may be used to define the Euler numbers. Here sech is the hyperbolic secant discussed in Chapter 29.
An integral definition of even-indexed Euler numbers is
5:5 INTRARELATIONSHIPS
where (;) is the binomial coefficient treated in Chapter 6, gives rise to the recursion formula for even-indexed
Euler numbers
E,, n
5:5:2 E"=-n!
t-o (n - 2j)!(2j)!
n=2.4.6, J=--1
2
with Ev = 1.
41 THE EULER NUMBERS, E. 5:11
5:6 EXPANSIONS
It follows from equations 5:4:1 and 3:6:4 that Euler numbers of even index may be expanded as
l"
/ } rr t (-I)1
5:6:1 E"=(-1)"/22n!(
L1-3, i+5 1)"/2_nii R+ n=0,2,4,...
For n = 2, 4, 6, ... the accompanying algorithm begins by using equation 5:4:1 with S(n + 1) replaced by unity.
This produces a very crude estimate of E", which is then refined by making use of rules 5:2:1 and 5:2:2. The final
output has a precision better than 6 X 10-s.
Setf=0
Input n >> p»»
If frac(n/2) + 0 go to (1)
Storage needed: n, f
1
5:9 APROXIMATIONS
Based on 5:6:1, one is led to
(-1)"n 2n! (2/a)"+i n = 4, 6, 8, ...
5:9:1 E. = 8-bit precision
5:12 GENERALIZATIONS
Euler polynomials E"(x) [Chapter 20] represent a generalization of Euler numbers, to which they are related by
5:12:1 E" = 2E.(1/2) n = 0. 1.2... .
Binomial coefficients occur widely throughout mathematics: for example, in the expansions discussed in Section
6:14 and in the Leibniz theorem, equation 0:10:7.
6:1 NOTATION
When v is the positive integer n, the binomial coefficient (;) is sometimes denoted C. or C". These symbols have
their origins in the role played by the binomial coefficient (:) in expressing the number of combinations of m ob-
jects selected from a group of n different objects [see Section 2:14].
6:2 BEHAVIOR
The lower index m of the binomial coefficient (;) is invariably a nonnegativc integer, whereas the upper index v
may take any value. Positive integer values of the upper index are common, however, and we shall write (a) as
the general expression for these cases.
The binomial coefficient (.) is always a positive integer when m = 0, 1, 2, ..., n and zero when m =n r 1,
n + 2, n + 3, .... Values of (;) for n values up to 16 and m up to 8 are tabulated in Section 6:8 (Table
6.8.1). For a given n. (e) assumes its maximum value when m = n/2 if n is even. or when m = (n ± 1)/2 if n
is odd; this is illustrated in Figure 6-1.
When v is not a positive integer or zero, the binomial coefficient (;) may be positive or negative, integer or
noninteger. Its values are zero when v = 0, 1, 2, ..., m - I and, as illustrated in Figure 6-2, arc close to zero
for all other v values in the range -k < v < m - i The binomial coefficient (;,) increases without limit (except
when m = 0) asv - x.
6:3 DEFINITIONS
The binomial coefficient is defined by the finite product
a
6:3:1 I
1
1 m -I
43
6:4 THE BINOMIAL COEFFICIENTS (;) 44
of a binomial coefficient with zero lower index. Binomial coefficients may also be defined by the generating
function [Section 0:3]
/
6:3:3
m
-1<I<I
For positive integer upper index the expression
6:3:4
m/ = m!(n - m)!
nzm
\A/
in terms of factorials (see Chapter 2], also serves as a definition.
6:5 INTRARELATIONS H I PS
There exist reflection formulas for the upper
6:5:1
rnn - 1)
(n+ ) = (-1)^(
and lower
6:5:2
n m)
(n - (m)
indices, as well as recursion formulas
6:5:3
(v m 1/ - \m/ +
(v - m) y
6:5:4
mI)
V+ (m+1)(m)
for each index. The addition formula
6:5:5
(v mW) -,-0 (J) (--j)
known as Vandermonde's convolution, applies to the upper index.
6:6 THE BINOMIAL COEFFICIENTS (;,) 46
There are a number of intrarelationships involving sums of binomial coefficients and including
+ +m _ n+ 1 n>m
6:5:6 GI m 1
6:5:7 i,,(-I)i(j)
i-o
=(-l)I(v-
m
1
and
(J+vH J+ I +v
6:5:8
`J+mJ+m J
Formula 6:5:5 provides an expression for a sum of products of binomial coefficients. a second such formula is
J?m
6:5:9
\m/ \J ! v/ - \mm v/ J J t m v/
When the upper index is an integer, finite or infinite series of binomial coefficients frequently have simple
expressions; examples include
(n)
6:5:10 =Y Jan
j
(,)6:5:12 nsJ=2.4.6.
and
6:5:15
(n)(2n) Jan
J n
6:6 EXPANSIONS
A binomial coefficient may be expanded as a power series in its upper index by the formula
6:6:1
(v/
m I iS''D
m! '.a
in which S'';' is a Stirling number of the first kind [see Section 18:6). As well, definition 6:3:1 constitutes the
expansion of (;,) as a finite product.
When v, but not in. is large in magnitude, the asymptotic expansion
6:6:2
vv rr
(rn) = - I i -
m(m - 1) m(m - 1)(m - 2)(3m - 1) + ...
v-- 'x
M! 2v + 24v2 I
47 THE BINOMIAL COEFFICIENTS (.') 6:8
6:6:3
(v
\m) ' (-I)"
m"' (1(-v)
[1+
v(v+ 1)
2m +
v(v+ 1Hv +2)(3v+ 1)
24m'
+ ... m-s 70
6:6:4
\n/2/
n 2
an
1-+-+
4n
1
32n'
1
even
Similarly, the asymptotic expansion for the maximum value of Q) when n is odd is
6:6:5
(n/2 ± 1/2)
n 2" V--2
[
1-3+ 2,
4n 32n2
urn J
oddn
6:7:1
and
6:7:2 =v
are valid for all values of the upper index. For integer values of this index, as well as for v = ±12. there are
additional particular values of the binomial coefficient as given in Table 6.7.1.
Table 6.7.1
m-0 I I 1 I 1 1
M - I 1 -1 0 2 I is
(2n)!!
(- IY(2m - 1)!! 0 -(-Ir(2m - 3)!!
m- n+ I.n+2,... (- 0' 0 0
(2m)!! (2m)!I
6:9 THE BINOMIAL COEFFICIENTS (;,) 48
Table 6.8.1
(,) (,) () (,) (,) (.,I (:,) (,j) (,) l,) ('"1 f!,=) (;,')
1 1 I I I I 1 I I I I I I I I I I m-0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 m1
0 0 3 6 10 15 21 28 36 45 55 66 78 91 IOS 120 m2
m3
I
Recursion 6:5:3 provides a useful way of evaluating binomial coefficients. In a construction known as Pascal's
triangle. binomial coefficients are listed in such a way that each entry is the sum of the two above.
0 1 0
0 1 1 0
0 1 2 1 0
0 1 3 3 1 0
0 1 4 6 4 1 0
0 1 5 10 10 5 I 0
0 1 6 15 20 15 6 1 0
Input v >>
f= I I Storage needed: v, f. m and j I
Input m >:
If m = 0 go to (2) Input restrictions: rn must be a nonnegative integer;
Setj=m v is unrestricted
Replace m by m - v
/(I
(1) Replace fbyf -ml
\ j/
Replace j by j - 1 Test values:
Ifj*0goto(1) (u) = I
(2) Output f (9')=-715
(';') = 0.0234375
6:9 APPROXIMATIONS
When both indices are small, the binomial coefficient is so easily calculated that no approximations are needed.
When only one of v and m is large in magnitude, the asymptotic expansion 6:6:2 or 6:6:3 may be truncated to
provide an approximation to (;).
When v is large. (;) is very large for values in the vicinity of v/2. Values of such binomial coefficients are
approximated by the Laplace-de Moivre formula
which finds statistical applications [see Sokolnikoff and Redheffer, pages 623-626]. Even for v as small as 10,
the approximation leads to small absolute errors, as illustrated in Figure 6-3.
111 m/
6:12 GENERALIZATIONS
The reciprocal of the complete beta function B (Chapter 43] generalizes the binomial coefficient to noninteger lower
index because of the relationship
6:12:1
(m) mB(m,vi m+1)
m#0
between the two functions. If the upper index is an integer, this generalization simplifies to
I -n!sin(irµ)
6:12:2
µB(µ.n - µ + 1) a(-µ).-1
where denotes a Pochhammer polynomial (Chapter 18). This formulation was used to draw the curves
linking the binomial coefficient points in Figure 6 1.
A generalization in a different direction is provided by mulrinomial coefficients which are the integer coefficients
that arise in such expansions as
6:12:3 (x+y+Z)5=[x5+y5+z5]+5[x4y+x'z+y'.r+z'y+y'z+z4y]
where the summation embraces all combinations of nonnegative integer m's that satisfy
6:12:5 =N
The multinomial coefficient is given by
N!
6:12:6
(m,)!(mz)!(m3)! ... (m,)!
and the number of terms multiplied by that coefficient is
n!
6:12:7 # of terms =
(W -d)!(W(mz))!(W(m3W ... 44-0)!
where It(m) is the number of occurrences of a specific value of m;. (Thus, the denominator in 6:12:7 is unity if
all the in values are distinct.) Tables 6.12.1 and 6.12.2 give a small number of trinomial coefficients (as exemplified
in 6:12:3) and quadrinomial coefficients: see Abramowitz and Stegun [pages 831-8321 for a more comprehensive
tabulation.
6:13:1
(v)(v-m+1),,
1m m!
and
v _ r(v + I) r(v + 1)
6:13:2
m m!r(v-m+ l) F. 1l)r(v-m+1)
relate the binomial coefficient to the Pochhammer polynomial [Chapter 18] and to the gamma function [Chap-
ter 43J.
6:14:5 ( l ±x)312= 1 =
3
2
X
3 1 3
16x3+ 128x4
_ 3
256
-x5+' -I x I
51 THE BINOMIAL COEFFICIENTS (;) 6:14
Table 6.12.2
No. of
N m,. MI. MI. M, terms
0 0.0.0,0 I 1
No. of 2 2,0.0.0 1 4
N m m m, terms 2 1. 1. 0. 0 2 4
0 0,0,0 1 1 3 3. 0, 0. 0 1 4
3 2, 1. 0. 0 3 12
1 110.0 1 3 3 1. 1. 1. 0 6 4
2 2.0,0 1 3 4 4.0.0.0 1 4
2 1. 1. 0 2 3 4 3.1,0.0 4 12
4 2, 2, 0, 0 6 6
3 3,0.0 1 3 4 2, 1. 1, 0 12 12
3 2. 1. 0 3 6 4 1. 1. 1. 1 24 1
3 1. 1. 1 6 I
5 5.0.0.0 1 4
4 4,0,0 1 3 5 4. 1.0.0 5 12
4 3. 1, 0 4 6 5 3, 2.0.0 10 12
4 2, 2. 0 6 3 5 3,1,1.0 20 12
4 2. 1, 1 12 3 5 2. 2. 1, 0 30 12
5 2.1,1,1 60 4
5 5.0,0 I 3
5 4, 1. 0 5 6 6 6.0.0.0 I 4
5 3.2.0 10 6 6 5, 1, 0, 0 6 12
5 3. 1. 1 20 3 6 4.2.0,0 15 12
5 2. 2. 1 30 3 6 4, 1, 1, 0 30 12
6 3, 3. 0, 0 20 6
6 6.0,0 I 3 6 3, 2. 1. 0 60 24
6 5. 1. 0 6 6 6 3. 1. 1. 1 120 4
6 4, 2. 0 15 6 6 2. 2. 2. 0 90 4
6 4. 1. 1 30 3 6 2, 2, 1, 1 180 6
6 3. 3. 0 20 3
6 3, 2, 1 60 6 7 7.0,0.0 I 4
6 2. 2. 2 90 I 7 6, 1.0.0 7 12
7 5. 2, 0, 0 21 12
7 7.0.0 1 3 7 5. 1. 1, 0 42 12
7 6. 1. 0 7 6 7 4, 3. 0. 0 35 12
7 5. 2, 0 21 6 7 4, 2. 1. 0 105 24
7 5.1.1 42 3 7 4,1,1.1 210 4
7 4, 3. 0 35 6 7 3, 3. 1, 0 140 12
7 4, 2, 1 105 6 7 3. 2, 2, 0 210 12
7 3, 3. 1 140 3 7 3. 2. 1. 1 420 12
7 3. 2. 2 210 3 7 2. 2, 2, 1 630 4
(1 ±x)'12 I,
I
6:14:6 = I ±-x--x-r5+...
I
2 8
I
16
5
128
7
256
-1 K r 1
6:14:7 (l+x)t:=1+mo:
x4 -
1
2
3
x+-
8
x' _ 16
5 x3 35
+ 128
63
256
x5+... -1<±x<1
6:14:8 (1 ±x)-' = I Z x+x' -Px3+x'+x5 + -1<x<I
6:14:9 (1 =x)-112= 1 w 2x+ 8 3 15 x'+ 35 315 693
x5+... -1 <x< I
16x3+ 128 x+256
6:14 THE BINOMIAL COEFFICIENTS (;,) S2
Many relationships in science and engineering take the form f(x) = bx + c. and many others can be manipulated
into this form by a redefinition of the variables. Data analysis problems can therefore often be reduced to the
determination of the coefficients b and c from pairs (x.f(x)) of experimental values. The least squares method for
performing this analysis is presented in Section 7:14.
7:1 NOTATION
The constant c of the linear function bx + c is termed the intercept, whereas b is known as the slope or gradient.
These terms derive, of course, from the observation that, if the linear function bx + c is plotted, its graph is a
straight line that cuts the vertical axis at altitude c and whose inclination from the horizontal is characterized by
the number b [see Figure 7- I ]. Occasionally the word slope is used to mean arctan(b) [see Chapter 35]. The symbol
m frequently replaces b.
33
7:2 THE LINEAR FUNCTION bx + c AND ITS RECIPROCAL 54
The name inverse linear function is sometimes given to the function 1 /(bx + c). Throughout this Atlas. how-
ever, we reserve the phrase inverse function for the relationship described in Section 0:3. The name reciprocal
linear function unambiguously denotes 1/f(x) where f(x) = bx + c.
7:2 BEHAVIOR
The linear function bx + c is defined for all values of x and (unless b is zero) itself assumes all values. The same
is true of the reciprocal linear function I /(bx + c) which, unlike the linear function, is discontinuous at x =
-c/b. The accompanying Figure 7-1 is a map showing typical behavior of the linear function and its reciprocal
when both b and c are positive. Figure 7-2 illustrates how the linear function is positioned with respect to the axes
when b and/or c changes sign.
+*O
i
7:3 DEFINITIONS
The arithmetic operations of multiplication by b and addition of c define f(x) = bx + c for all values of x. The
function f(x) = 1l(bx + c) is defined by the same operations followed by division into unity. The linear function
is completely characterized when its values, f, and f2. are known at two (distinct) arguments, x, and x2. The slope
and intercept may be inferred from the formula
7:5 INTRARELATIONSHIPS
Both the linear function and its reciprocal obey the reflection formula
7:5:1 f(-x)=-flx- b 1
55 THE LINEAR FUNCTION bx + c AND CPS RECIPROCAL 7:6
b2 b2c2 b2x I b2 l
The inverse function [Section 0:3] of the linear function f(x) = bx + c is another linear function
As explained in Section 16:3, the difference of two reciprocal linear functions that share a common b parameter
produces a reciprocal quadratic function. Finite and infinite sums of certain reciprocal linear functions may be
evaluated in terms of the digamma or Bateman's G functions [Chapter 441 as follows:
`s
7:5:7 1+ +
I
+...+ I
=L, 14j
c x+c 2x +c Jx+c ;_oJx+c
1
x x x
7:5:8
c x+c 2x+c Jx+c ;G0Jxx+c 2xIlL` (2+(2x1 20
x(2+2+2x)J
G(J + 1 + C)
2x J
7:5:9
c
1
- - - +
1
x+c
1(-1)'
where the upper signs are taken if J is even, the lower if J is odd, and
2x + c
-
;_o .x + c
1
2x
1
2
+
c
2x)
c
2x)
_ I
2x G X
(c)
1
7:6 EXPANSIONS
The reciprocal linear function is expansible as a so-called geometric series for either small
I I bx 62x2 63x3 I bx ' c
7:6:1 1 i<
bx + c c c' c3 c' c \ c b
or large
1 _ 1 c c' c 1 c c
7:6:2 J.r1 >
bx + c bx b2x2 b'x3 63x3 bx bx b
7:7 THE LINEAR FUNCTION be + c AND ITS RECIPROCAL 56
as an infinite series of Bessel coefficients [Chapter 521 of odd order, although this representation is rarely employed.
The reciprocal linear function may also be expanded as an infinite product
c c-bx 11 bx
IILL I + \bx)2i]
I\ /I J -I<-<I
J-
7:6:4 = i,l
bx+c bx-,c111r f
b2x- L
+l\bx bx >1
I C
7:9 APPROXIMATIONS
For small absolute values of x, the reciprocal linear function may be approximated by a linear function
I bx
4x1 <Icl-
I
7:9:1 =-- 8-bit precision
bx + c c c 16[b1
7:9:2
1 - bx - c 8-bit precision xJ >
161c-
bx + c b''x' JbI
is useful for large arguments of either sign. These two results follow directly from expansions 7:6:1 and 7:6:2,
respectively.
and
I Inl I1 +bxll
7:10:4
Iu bt+c dt I
b \ c J
respectively. If 0 < (-c/b) < x. the integrand in 7:10:4 (and in some of the integrals below) encounters an infinity;
in this case the integral is to be interpreted as a Cauchy limit, as explained later in this section. Other important
--}
integrals include
dt _ bx
7:10:5
1
In 1 +- 1 + Cj/ BC *bC
o (Bt + C)(bt + c) bC - Bc c
Bt+C Bx bC - Bc bx
7:10:6 dt = + In ( 1+
U bt+c b b2 \ c
and
7:10:7
fu
edt
br + c
- (-c)"
b"'
In I+ bx-
c
j\/] +
,-If
lr bxl
Semidifferentiation and semiintegration [see Section 0:10] with lower limit zero give
c
n=0,1,2....
d1f2 2bx + c
7:10:8 dx'/2(bx+c)
Vnx
and
/ \\
7:10:9 d;7Fi(bx+c)= \43x+2c1
7:10:10
dV2 I
=
t - Vx
dx"2 bx + c \(br + c)
d-1/2 1
26
7:10:11 =
dx '2bx+c yzrb
when applied to the reciprocal linear function, where the form of the function d depends on the magnitude of
c/b, as follows:
aresin( bx/c) c
-x < - < -X
U-x - (c/b) b
-bx/c) c
7:10:12 6_i arcosh( -x<-<0
b
V.r + (c/b)
arsinh(Vbx/c) c
b
V
In formulas 7:10:8-7:10:12 it is understood that x > 0.
7:11 THE LINEAR FUNCTION bx + c AND ITS RECIPROCAL 58
7:10:18
JY
f(t)r dt- -s
encounters an infinity at x = s and therefore the definition of the Hilbert transform requires a so-called Cauchy
7:10:19 J i f(t) -
limit interpretation of the integral in 7:10:16;t(that is:
t-s = lima
'-°
(1
J' s , f(t) -
t-s +
r
I
f(t)
Occasionally in this Atlas [e.g.. in Section 22:10] a Hilbert transform will be encountered with integration limits
dt
t - s)
l
} e>0
is then intended. The Hilbert transforms of many functions are tabulated in Erdelyi, Magnus, Oberhettinger and
Tricomi [Tables of Integral Transforms, Volume 2, Chapter 15]. For example, the Hilbert transform of a pulse
function [Section 1:13] is a logarithm [Chapter 25]
7 :10: 2 1
1J ( p( 1;2;r)
dt I
In
( s 1
a t-s a s+I
7:12 GENERALIZATIONS
The linear function is the special a = 0 case of the quadratic function discussed in Chapter 16.
A linear function is a member of most of the families of functions discussed in Chapters 17 through 24; these
polynomials may therefore be regarded as generalizations of the linear function.
FIG 7-3
............ ......:..............
........... .......... f(xJ,t)
Usually a piecewise-linear function has discontinuous derivatives at all the interior values x,. x,, x,...., .t"_
The alternative interpolation formulas discussed in Section 17:14 do not suffer from this defect.
The reciprocal linear function is related to the function discussed in Section 15:4 because the shape of each is
a rectangular hyperbola. Thus, counterclockwise rotation through an angle tr/4 of the curve 1/(bt - c) about the
point x = -c/b on the x axis produces a new function
7:13:2 ,Jlx+bl'3
that is a rectangular hyperbola of the class discussed in Section 15:4.
and are entirely associated with the measurement off (that is. the x values are exact), then the formulas
7:14:1 b =
nIxf - ExIf
n E xz - (Et)z
and
c
_ fxzIf - ExExf Ef- b.x
7:14:2
nExz-(Fx)z it
7:14:3
is used in the formulas of this section. The term "best' is employed here in the sense of producing the minimum
sum of squared deviations E(bx + c - f)z, and the procedure is known as least squares fitting or linear regression
[see also Appendix A. Section A:4J. Figure 7-4 shows an example of five data pairs and the "best straight line"
through them.
A measure of how well the data obey the linear relationship is provided by the correlation coefficient, given
by
The significance to be attached to these statements is that the probability is approximately 68% that the true slope
lies between b - d and b + d. Similarly, there is approximately a 68% probability that the true intercept lies
between c - e and c + e. The formulas
d- nXf'-(Ef)' nTxf-ExTf =b I I
7:14:7
and
2 1.6
Output b 3 4.4
b = slope 5 K««
Set d = bV (I /r) - I n=5
4
0.3
5.5
8.0
Output d
d = slope ««<
error
c
Replace c by - - be
n
7:14 THE LINEAR FUNCTION be + c AND ITS RECIPROCAL 62
Output c utput:
I
c = intercept <<<< r=0.9134
Replace e b% d\ e b - d = 17.9 - 3.3
Output e c-e=2.31=0.57
intercept <<<<
error
A related, but simpler, problem that occurs frequently is the construction of the best straight line through the
points (x f, ), (x2, f2). (x3. f3), .... (x,,. f) with the added constraint that the line must pass through the point (xo, fo)
In practical problems this obligatory point is often the coordinate origin. The slope and intercept in this case are
given by
E(x - Xo)(f - A)
7:14:9 b =
E(x - xo)2
and
7:14:10 c = fo - bxo
Equations 7:14:1-7:14:8 are based on the assumption that all data points are known with equal reliability, a
circumstance that is not always valid. Variable reliability can be treated by assigning different weights to the points;
thus, if the value off, is more reliable than f2, one assigns a larger weight w3 to the data pair (xr, f,) than the weight
w2 assigned to pair (x2, f_). The weights then appear as multipliers in all summations, so that, for example, equations
7:14:1 and 7:14:2 become replaced by
lwlwXf - lwxIwf
7:14.11 b =
1WYWX - (I14'x)2
and
This chapter concerns the unit-step functions u(x - a) and u(x), the signum function sgn(x) and the absolute value
function lxl. These functions are closely interrelated through the simple formulas
and
8:0:4
8:0:5
and
8:0:6 If(x)I
8:1. NOTATION
The unit-step function is also known as the Heaviside function, or Heaviside's step function, and is sometimes
denoted by H(x - a) or S,(x). It should not be confused with the unity function [Section 1:4]. The notation u(x)
replaces u(x - a) when a is zero.
The signum function is also called the sign function and symbolized sign(x) or sg(x).
Alternative names for the absolute value function are the magnitude of f(x) or, especially when f(x) is complex,
its modulus. The term "modulus" should not be confused with "modulo" [see Section 9:13]. The symbolism ABS(x)
sometimes replaces lxl, especially in computer applications.
63
8:2 THE UNIT-STEP u(x - a) AND RELATED FUNCTIONS 64
8:2 BEHAVIOR
Figures 8-1 through 8-4 illustrate the effect of modifying f(x) to each of the expressions 8:0:3-8:0:6. The modi-
fication is shown in red and the original (arbitrary) function f(x) in green. In Figures 8-3 and 8-4, r is a zero of
the function f. Figure 8-4 relates to a function f that is real because, as explained in Section 8:11, a special
significance attaches to jf(x)j when f is complex.
A 6
FIG 8-1
u (x-o) f (x)
.............p
8:3 DEFINITIONS
The unit-step function is defined by
(0 x<a
8:3:1 u(x-a)= 1/2 x=a
I 1 x>a
and therefore the effect of multiplying a function by u(x - a) is to nullify the function for arguments less than a
and leave it unchanged for arguments greater than a:
0 x<a
8:3:2 u(x - a)f(x) = f(x)/2 x = a
f(x) x>a
The signum function extracts the sign of its argument so that
f-I f(x)<0
8:3:3 sgn(f(x)) 0 f(x) = 0
l 1 f(x) > 0
65 THE UNIT-STEP u(x - a) AND RELATED FUNCTIONS 8:3
if -,*1C
The absolute value function equals its argument when the latter is positive and equals the negative of its ar-
gument when the latter is negative. Hence
8:4 THE UNIT-STEP u(x - a) AND RELATED FUNCTIONS 66
and
f(x) f(x) c 0
8:3:6 If(x) =
f(x) f(x) z 0
with the requirement that, for 8:3:5 to be valid, f(x) must be real.
The unit-step function may be used to construct many special discontinuous functions. For example, the pulse
function (Section 1:13] can be represented by a difference of two unit-step functions:
/ / 11
are examples of the staircase and sawtooth functions, respectively [see Section 9:13].
Acting on a sinusoidal function (Chapter 32], the signum function generates a `square wave" such as sgn(sin(x)).
while the absolute value function effects "full rectification" as in Icos(x)I. See Section 36:14 for an explanation of
both of these concepts.
8:5 INTRARELATIONSHIPS
8:6 EXPANSIONS
If the function f(x) is expansible as a power series (Section 11:14], the unit-step function may be distributed through
the series
x< 0 x- 0 x> 0
Ulx) 0 1
1
s$n(x) -1 0
xl -x 0 X
8:9 APPROXIMATIONS
The unit-step and signum functions may be approximated by a variety of continuous functions. For example, as b
takes ever larger positive values, the approximations
1
8:9:1 u(x - a) -(1 + tanh(bx - ba))
and
8:10:3
d
-sgn(f(x))=0
dx
x*r n= 1.2.3,...
8:10:4
d
fgxl) =
I
-
df
(x) x>0
dx
- -df(-x) x<0
8:11 THE UNIT-STEP u(x - a) AND RELATED FUNC71ONS 68
and
8:10:5 -
d
dx
If(x)I = sgn(f(x)) - (x)
df
dx
x * r"
where r,. r2. r ... are the zeros of f(x), that is, f(r") = 0, n = 1, 2. 3, .... The delta function, 8. occurring in
formula 8:10:1 is the Dirac delta function [see Chapter 10].
Integration is also easily accomplished:
8:10:6
f u(t-a)f(t)dt=Jtf(t)dt a
xo<asx
8:10:7dt=2sgn(f(xo))
Jsgn(f(r))
1
Z+r,-r,+r3--(-1)"(r"-f(xo)*0
rr JJ'
f(t)dt
8:10:8
(
f f(I:I)dt=Jxf(t)dt-sgn(xo)
o J
8:10:9 f. If(t)I di = f(r) dt J f(t) dt + J f(t) dt + + J f(r) dr
f
I I ^
where r r,, r3, ..., r" are those zeros of f(t) that lie between x0 and x.
Differintegration with a lower limit x0 gives
d" [x - a]
8:10:10 u(x-a)= xo<a<x
[d(x r(1 - v)
and
d' d"
8:10:11 [u(x - a) f(x)] = u(x - a) f(x) xo < a < x
[d(x - xo)]' (d(x - a)]'
This last result is a generalization of 8:10:1 and 8:10:5.
8:12 GENERALIZATIONS
The function defined by
8:12:1 [u(x - x0) - u(x - x,)] f(x) xo < x,
is equal to f(x) inside the "window" xo < x < x, and to zero outside. Such a window function is exhibited in Figure
8-5.
40 44%
4
.0
69 THE UNIT-STEP u(x - a) AND RELATED FUNCTIONS 8:13
The pulse function (Section/1:131 is the simplest example of such a window function:
xo x,
8:12:2 plc;xi - x0;x - 2 I = ((x - -to) - u(x -
8:12:3 (u(x-u(x-x,.1)lf;(x)
which constitutes a piecewise-defined function, as illustrated in Figure 8-6. The properties of such a function may
be deduced in a straightforward fashion from those of u(x - a)f(x) as discussed throughout this chapter.
;0 y44% if 4'
9:1 NOTATION
The integer-value function is also symbolized [x], E(x) or int(x). The fractional-value function is sometimes sym-
bolized (x). See Section 9:13 for the very similar integer-pan and fractional-part functions; these are sometimes
also denoted Int(x) and frac(x).
9:2 BEHAVIOR
Figures 9-1, 9-2, 9-3 and 9-4 depict maps of f(Int(x)), f(frac(x)), Int(f(x)) and frac(f(x)). They are sketched for
a function f that is monotonic and continuous in the region graphed: a monotonic function has no maximum or
minimum. Several of the formulas in this chapter are restricted to values of x for which f is monotonic.
The functions f(lnt(x)) and Int(f(x)) are piecewise constant [Section 1:13]. Because its segments are of equal
width, the f(Int(x)) function has a graph that resembles a histogram.
The f(frac(x)) and frac(f(x)) functions assume values in restricted ranges. Thus, provided f is monotonic in
the 0 s x s 1 range, f(frac(x)) takes values between f(O) and f(l). Similarly, 0 s frac(f(x)) < 1.
9:3 DEFINITIONS
The integer-value function extracts from its argument the largest integer that does not exceed the argument; that
is:
9:3:1 Int(x)=n n_x<n+1 n=0,±1,±2,...
71
9:3 THE INTEGER-VALUE lnt(x) AND FRACTIONAL-VALUE frac(x) FUNCTIONS 72
4* 4i 1 440 4*
FIG 9-1
FIG 9-2
ff (froc (x))
f(x)
so that
9:3:2 f(Int(x)) = f(n) nsx<n+ I n=0, ±1, ±2,...
and
and
9:3:6 frac(f(x)) = f(x) - n f(x) - n f monotonic 0:5 f(x) - n < 1 n = 0, =1, t2, ...
.....................-3
frac(f(x).).
:..... ... 0
9:5 THE INTEGER-VALUE Int(x) AND FRACTIONAL-VALUE frac(x) FUNCTIONS 74
kw
FIG 9-5
9:5 INTRARELATIONSHIPS
It follows from equation 9:0:1 that Int and frac are complementary functions in the same sense that
9:5:1 f(Int(x) + frac(x)) = f(x)
and
9:6 EXPANSIONS
When f is a monotonic function, lnt(f(x)) may be expanded in terms of the alternative unit-step function [Section
8:13]
df
9:6:2 Int(f(x)) = (I - v(x - F(j))] - v(x + F(-j)) f monotonic <0
).u i- dx
where F is the inverse function of f, that is, F(f(x)) = x.
Because f(frac(x)) is a periodic function of x, it may be expanded by the methods described in Chapter 36.
may be used to find frac(x). The difference x - frac(x) then gives Int(x). This algorithm uses functions from
Chapters 32, 34 and 35 with the identity
ax p<x<-
9:8:1 arctan(tan(a[n + x])) _ 2 n=0,±I, t2,...
a(x-1) 2<x51
The commands shown in green guard against overflow; they may be omitted with many computing devices.
9:9 APPROXIMATIONS
There are none that are sufficiently useful to be included here.
-1 11 'x-N
X-A
df
(
{Xr-Aa- E F(A + j) f monotonic
dx
>0
9:10:7 Int(f(t))d1 - A-x d.
J
Xx - Aa + F(X + j) f monotonic <0
i_ dx
where A = Int(f(a)). X = Int(f(x)) and F is the inverse function of f. Finally, by virtue of 9:5:2
9:12 GENERALIZATIONS
The function f(frac(x)) is an example of a periodic function (see Chapter 36]. Its period is unity. Likewise f(Int(x))
and Int(f(x)) are examples of piecewise-constant functions [see Section 1:131.
Occasionally the symbol n(m) is encountered and the name modulus is used to describe m.
77 THE INTEGER-VALUE Int(x) AND FRACTIONAL-VALUE frac(x) FUNCTIONS 9:14
The integer-value and fractional-value functions occur widely in number theory. In addition they lie at the heart
of such devices as analog-to-digital convertors that digitize measured signals of various kinds. They also play roles
in the conversion of a number x from one number system to another; from the decimal system to the binary, for
example, or from the hexadecimal to the decimal.
Any positive number can be represented as
9:14:1 X 0"
where n is an integer and each N. takes one of the integer values 0, 1, 2, ..., (p - 1), where P is the base of the
number system and where N. * 0. Such a representation is termed floating point or, if 0 = 10, scientific notation.
An alternative is the fixed point representation
9:14:2
where it is a sufficiently large integer and, as before, each N is an integer drawn from the set 0, 1. 2. ,
(R- 1).
The algorithm below digitizes any positive number x by generating a sequence of integers. The first integer is
n, a decimal integer of either sign. The other integers are digits of the base 02 number system. These may be
interpreted as the No, N_ N_2, ... digits of the floating point representation 9:14:1, or as the N. N,_,,
digits of the fixed point representation 9:14:2.
The second algorithm accepts a number in base R, representation and converts it to a decimal number x. The input
is in the form of the signed decimal number n followed by a sequence of digits in base 31. All these digits must,
of course, be positive. Following the entry of the last true digit a negative digit is input as a cue that the data entry
is complete. This negative digit is ignored in the computation of x.
The two algorithms are designed so that they can be combined, with the second preceding the first. The con-
joined routine will then convert any positive number from base B, to base S2.
9:14 THE INTEGER-VALUE Int(x) AND FRACTIONAL-VALUE frac(x) FUNCTIONS 78
This function, strictly speaking, is not a function at all because it violates rules that would be valid if it were a
function. For example, while it takes the value zero at all arguments but one, its integral has the value unity.
Nevertheless, the Dirac delta 'function" is a useful concept that has found widespread application, especially in
classical and quantum mechanics, where it occurs mainly as a multiplier of an integrand. as in
10:1 NOTATION
The impulse function or unit impulse function is an alternative name for S(x - a).
Dirac's name is associated with this function to avoid confusion with the Kronecker delta function [Section
10:13].
10:2 BEHAVIOR
The Dirac delta function cannot be graphed: S(x - a) is zero for all values of x except x = a, where it is infinite.
10:3 DEFINITIONS
The Dirac delta function may be defined in terms of a limit in many ways, for example:
1
10:3:1 S(x - a) = limp -;h;x - a
e-o h
79
10:4 THE DIRAC DELTA FUNCTION 8(x - a) Be
8(x - a) = li m [L2c
a)]
10:3:3 -L h=\x
c
[see Chapter 29). All of these definitions-and many others-describe a function peaked at x - a that, as the limit
is approached, becomes infinitely high and infinitesimally wide but whose area remains constant and equal to unity.
Figure 10-1 illustrates the progress toward the limit in the case of definition 10:3:3. It follows that
d
10:3:5 8(x-a)-u(x-a)
of the unit-step function [Chapter 8]. Yet another representation is as the definite integral
which is important in the evaluation of certain Fourier transforms [see Section 32:14].
10:5 INTRARELATIONSHIPS
One may derive the reflection formula
10:5:1 8(a - x) = 8(x - a)
10:6 EXPANSIONS
There arc none.
10:9 APPROXIMATIONS
No approximation of 8(x - a) itself is of any use. However, expressions 10:3:1-10:3:3 can sometimes be of value
in approximating integrals in which 8(x - a) appears as a factor of the integrand.
Differentiation of the Dirac delta function produces the unit-moment function discussed in Section 10:12. Integration
gives the unit-step function (Chapter( 81
It is this last relationship, known as its sifting properh', that renders the Dirac delta function so useful.
The notation f(x) * g(x) and the definition
relate to the so-called convolution of the functions f and g. The convolution of two Dirac delta functions obeys the
rule
10:10:5 6(x- a) * S(x - b) = S(x -a - b)
10:11:1 J J &(x+yi-a-bi)dxdy= I
as well as other relations that parallel its behavior as a function of a real argument.
10:12 GENERALIZATIONS
By utilizing the concept of differintegration [Section 0:10) it is possible to define a continuum of functions of which
the unit-step function and the Dirac delta function are respectively the v = 0 and v = 1 instances. The general
definition is
d"
10:12:1 u(x - a)
[d(x + x)]°
which evaluates to
.
u(x-a)(x-a)
10:12:2 v * 1,2,3....
r(1 - v)
except when v is a positive integer. The v = 2 case, symbolized S'(x - a), may be regarded as the limit
h\\ / hll
S x-a--l-8(x-a+2/
2 \
10:12:3 S'(x - a) = lim // /
*-a h
and is named the unit-moment function. It satisfies the integral identity
10:12:4
f
r
J a S'(r - a)f(t)dt = -F(a) _ - dx(a)
df
The Dirac delta function may be regarded as a bivariate function of the variables x and a, each of which can adopt
any real value: it is zero except when these two variables are equal. The Kronecker delta function b(n,m) or 5,,,,,
is an analogous bivariate function but its two variables are restricted to integer values. It is defined by
10:13:1 S(n,m) =
0 n*m
1 n=m
CHAPTER
11
THE INTEGER POWERS (bx + c)" AND x"
With n = 0, ±1, ±2, ... this chapter concerns the function (bx + c)" and its special b = 1, c = 0 case. The
powers 1, x, x2, ... and the reciprocal powers 1, x-1, x-2,
... are the units from which most expansions are built;
such expansions are the subjects of Section 11:14.
11:1 NOTATION
The two symbolisms (bx + c)-" and 1/(bx + c)" are equivalent in all respects. The powers x2 and x' are termed
the square and the cube of x, respectively.
In the general notation P°, [i is known as the base and a as the power or exponent. In this chapter [and in
Chapter 13] the family of functions in which the base is the primary variable is treated: In contrast, Chapter 19 is
concerned with functions in which the exponent varies and the base is held constant.
11:2 BEHAVIOR
The power function is defined for all values of x and for all integer n except that x" is undefined when both x and
n are zero. Figures 11-1 and 11-2 illustrate the behavior of x" for n = 0, ± 1, ±2, ±3, ±4, ±7 and ±12. Notice
the contrasting behavior of the positive and negative powers. Note also how the reflection properties depend on
whether it is even or odd.
11:3 DEFINITIONS
The function (bx + c)" for n = 1, 2, 3, ... is the product of n factors, each equal to bx + c:
83
11:4 THE INTEGER POWERS (bx + ci' AND f 84
'L O 0 CO ti
A, A e ?, ea
:....:....:....:....:....:.........:.-0. 4
FIG 11-1
11:5 INTRARELATIONSHIPS
The function x" obeys the simple reflection formula
x n = 0, ±2, ±4, ...
1-x" n = ±1, ±3, ±5....
as THE INTEGER POWERS (bx + cy' AND x' 11:5
11:5:2 +c+xJ+c]
/ \ 1
The recurrences
11:5:3 (bx + c)" = (bx + c)(bx + c)"-'
and
(bx + c)-'
11:5:4 (bx + 0-' =
bx + c
apply, as do the laws of exponents (x")(x') = x"", x"/x' = x"-", (x')' = x"" and their extensions
11:5:5 (bx + c)'(bx + c)' = (bx + c)""
11:5:6 (bx + c)"/(bx + c)' = (bx + c)"-'
and
x"ty(x+y) 7
1
2jn =
11:5:8 11 IX2 :t 2xy cos( +y n= 1,3,5,...=2J+ 1
ri n
11:5:9 z +}
n
)+}xI n=2,4.6,...=2J
2z
11:5:10 x"-y"=(x + y)(x- y) fl[x- 2xy cos(
n/)+, 1
n=2,4,6,... =2J+2
constitute function-addition and function-subtraction formulas for integer powers, cos being the cosine function of
Chapter 32. Simple instances are x2 - y2 = (x + y)(x - y), x3 t y3 = (x t y)(x + xy + y2), x` + y` _ (x +
f2ry+y2)(x- V2xy+y2)andx4-y` = (x + y)(x - y)(x2 +y2).
Finite sums of positive or negative powers may be evaluated as the geometric sums
11:6 EXPANSIONS
If n is positive, (bx + c)" may be expanded binomially as the finite sum
11:6:1 (bx+c)"=c"+nc"'bx+
n(n2.
1)
c j (J)(bx)i
n=0.1,2,...
for all x, where (')is the binomial coefficient of Chapter 6. If n is negative, the sum is infinite and takes the form
cl
n=-I,-2,-3.... 1< b
or
j-n-11 c
11:6:3
2! bx
c
n=-1,-2,-3.... [rj>
b
where o 4 is a Stirling number of the second kind [Section 2:14], or in terms of Chebyshev polynomials of the
first kind (Chapter 22]:
87 THE'INTEGER POWERS (bx + c)" AND x" 11:9
The coefficients y"' are zero whenever j and n have unlike parities. The coefficient yo' equals unity, while y." is
zero for m > n. The numerical values of all other coefficients may be calculated from the recursions Yo' _
y('-0/2, y(') = (,y('-')/2) + yo-" and for m ? 2 y." = (y."_-." + Expansions similar to 11:6:5, but
involving the Chebyshev polynomials of the second kind U"(x), also hold. In fact, such expansions exist with any
set of orthogonal polynomials [Chapters 21-241.
X --= x
c-1
b
X _-cb- x -I -b-c x . x
(bx+c)".n- 1.3.5.... x -I 0 1 x
Ibx + c)", n = 2. 4. 6, ... = 1 0 1 =
11:9 APPROXIMATIONS
Ifn*0:
/
11:9:1 (bx + c)" = c"I l + .rl 8_bit precision xI < Icl
c. 12 n(n - 1)IbI
11:10 THE INTEGER POWERS (bx + c)" AND j 88
nc 12 'v n( Icy
11:9:2 (bx + c)" = b"" I\ I + 8-bit precision xt > ibi
bx)
b(n + 1)
n=0.1.2....
(bt + c)"dt =
11:10:2
x n=-1.-2.-3.
r x n = -1, 0. 1. 2.
Or + c)'dt = (bx + c)
11:10:3
J
b(-n - 1)
n-2.-3.-4,
(bx, + c)"'' - (bxo + c)"
n = 0, 1, ±2,±3....
b(n + 1)
11:10:4 J" (bt + c )'dt =
1 bx, + c
b bxo+c) n= -1
(n)()2)k(n )(-?)
where each upper limit in the summation is an integer chosen to make the final binomial coefficient (:) or
Some examples are
11:11:2 (x + iy)2 = (x2 - y2) + i(2xy)
An example is
3y' 5v' /2
11:11:6
1
(x + iy)-x = (X-2 - x +
xb
xr - ss + jr,
- ... I
11:12 GENERALIZATIONS
In the present chapter the power is restricted to be an integer. This restriction is removed in Chapter 13.
of terms of which ax' is typical, j being a nonnegative integer and a, a constant. Such sums are known as power
series and examples will be found in Section 6 of most chapters of this Atlas. In the notation of Chapter 17, power
series are polynomial functions of infinite degree, and could be symbolized p.(x). The somewhat more complicated
series
11:14:2 aox° + a,x°'B + a,X'-' + a1.t
-o
may also be treated as a power series because a redefinition of the argument and isolation of the factor x' relates
11:14:2 to the 11:14:1 function
11:14:3 Za,x°la=xf(xe)
i-o
Depending on the values of the coefficients a,, power series may converge for all x, only over a certain range
of argument values. or may represent an asymptotic expansion near some x = xo value [see Section 0:6]. Provided
that a power series converges for argument x, it may be differentiated term by term:
d
11:14:4 1)a;.,x,
dx,-o
or integrated:
or divided:
a1x
a 1'
11:14:8 x, '-a "X' where ci = bi-icA.
bo bo k-0
bi
1-0
If f is given by the power series Ea,x', then the inverse function (see Section 0:3] is another power series of
argument If - ao)/a namely
so
11:14:9 c1(f
a2
where c, = a,, c2 = -a,a,, c, = 2ala22 - alla,, c. = Sa a2a, - Sa,a; - a3la., c, = 14a,a2 + 3aja3 +
21 a;a';a, - a;a, and ce = 84a;a, + 7o;a,a. + 7a;a a, - 42a,a2 - 28a3,a - 28a3a2'a, - a;ab. The procedure
of converting a power series for f(x) into a power series for x is known as reversion of series.
Any function that can be repeatedly differentiated may be expanded as a power series by utilizing the Maclaurin
series (the special y = 0 case of the Taylor series 0:5:1]:
df d2f (0) + = x' dJf (0)
11:14:10 f(x) = f(0) + x (0) + - T2
dx 2 1_o j! dx'
Whether or not such a series is convergent, it provides an asymptotic representation of f(x) as x -a 0. Functions
expansible as power series may also be represented as series of Bessel functions [see Section 53:14] or as continued
fractions [Section 0:61.
CHAPTER
12
THE SQUARE-ROOT FUNCTION
bx + c AND ITS RECIPROCAL
Functions involving noninteger powers are known as algebraic functions. The square-root function V and the
reciprocal square root 1/Vx are the simplest algebraic functions. In this chapter, as in the previous one, we gen-
eralize the argument of these simplest algebraic functions and consider mainly the bx + c and 1/vi"" + c func-
tions.
A graph of 'V bx + c versus x generates a curve known as a parabola and the adjective parabolic is therefore
appropriately applied to the (bx + c)'r function. Some geometric properties of the parabola are noted in Section
12:14.
12:1 NOTATION
Especially in computer applications V' is sometimes denoted SQRT(x) or SQR(x). The notation lx is also oc-
casionally encountered.
The symbols x"' and V x- often interpreted as defining equivalent functions but in this Atlas we make a
distinction between the two. If x is positive x 12 has two values, one positive and one negative. The square-root
function f, however, is single valued and equal to the positive of the two x"= values. Hence the relation between
the two functions is
12:1:1 bx+c=I(bx+c)"'1
or
Si m ilarl y
I
12:1:3 = Ox + c)-'"I
bx+c
and
c)-1"2
12:1:4 (bx + =
91
12:2 THE SQUARE-ROOT FUNCTION bx --c AND ITS RECIPROCAL 92
12:2 BEHAVIOR
Figure 12-I is a map of the functions V box + c and I /V bT under standard conditions, that is, when b and c
are both positive. The orientation of bx + cchanges to those shown in Figure 12-2 when b and/or care negative.
The Vbx + c function is not defined for values of x more negative than -c/b. The function itself takes all
nonnegative values. As Figure 12-I shows, V b x+ is zero at x - -c/b, at which point it has an infinite slope.
Likewise, 1/\'bx + c is defined only for bx > -c and takes all positive values. Figure 12-1 reveals that the
reciprocal square-root function steadily declines from an infinite value at -c/b toward zero as x - x.
FIG 12-1
12:3 DEFINITIONS
The square-root function is defined as the inverse of the square function (Chapter I11. Thus, V'bx + c is the
nonnegative number whose square equals bx + c; that is:
A parabola is defined geometrically as constituting all points P whose distance PF from a fixed point F (called
the focus of the parabola) equals the shortest distance from P to a straight line DD (the directrix of the parabola).
If the x-axis of a Cartesian coordinate system is placed along DF. the shortest line joining the directrix to the focus
(see Figure 12-3), and if d is the length of that shortest line, then the equation of the parabola is f = bx + c
where b = 2k and c = k2 + lay, y being the distance from the focus to the coordinate origin.
93 THE SQUARE-ROOT FUNCTION bx T -c AND ITS RECIPROCAL 12:5
12:5 INTRARELATIONSHIPS
The multiplication and division of square-root functions generate root-quadratic functions [see Section 15:121:
12:5:1 b,x + c, byr + r. _ (b,.,, - c,)(b,x + c,) = b,b x1- + (bic: + b.ci)x + c,c,
and
%lb b x
, , ' + (b c. + b . c )x + c c ,
bix+ci/ b.r+c,_
, ,
+c,)/(b,x+c,)_
,
12:5:2 (b,x
bx+c,
while the rule
12:6 EXPANSIONS
12:6:1
/ bx
2c
62x2
8c'
b3x 3
16x-
l
c ,io \ j
1
1/ 2) bx '
`e
_\1cc(2j-3)!' bx)' -c5bxsc
,-i (2j)!! cI
[see Chapter 6 for (;,) and Section 2:13 for n!!] and
12:6:2
I
bx+c Vc
1
--+
2c
z - z16r+
bx 3b'x'
8c
Sb3x 3
_
1
j
1/2) bx
cJ
1
j-o
(2j - 1)!! (-bx)'
(2j)!! c
- csbxsc
j
are valid for small values of (bx/cI, whereas
// c c= (c
12:6:3 bx + c = V rbs l l -vrbx (I/2 -bxscsbx
+ 2bx 862x'+ \ j / \bx/
and
12:6:4
I / c 3c' l 1 / c\
c'
c Vb, 2bx (-1/2) bx)
apply when jbx/cj is large, bx being positive.
The square-root function can also be expanded as a ratio of two exponential series [see Section 27:13):
For b * 0
12:7:1 bx+c=0 x= b
12:7:2 bx+c= I =1 x=
bx+c
and
12:7:3 =0
bx+c
95 THE SQUARE-ROOT FUNCTION bx + c AND ITS RECIPROCAL 12:10
Computer languages provide instructions for implementing the square-root operation. Programmable calculators,
indeed, even nonprogrammable ones, usually have keys that generate N /X.
12:9 APPROXIMATIONS
If p is an approximate value of Vx, then (p2 + x)/2p is a better approximation. This is the basis of Newton's
-
method for estimating square roots.
The first four equations of Section 12:6 lead directly to the approximations
bx c
12 :9 : 1 bx + c =V-c + 8 - bi t prec i s ion 1bx1 <
2 c
12 9 2
: :
' '- rI - 1\
bx)
8- bit prec ision 1bx1 <
10
c
V bx+c Vc
12 9 3
: : bx + Vb. + 8 -bi t prec is ion bx > 71 c 1
2Vb;
_ l (
12 : 94
:
1
d 1 -b
12:10:2
dx br + c 2(bx + c)3i:
and
dr 2
12:10:4 bx+c
f-"I-'Vbt +c b
and
12:10:6 dt 7 arcosh
bx
c>0
c/bf bl+C
f 2
C
arccos
Vbx
C < 0
are two examples drawn from a large class of integrals of the general function (bx + c)" 2(Bx + C)"/2 in which at
least one of the integers n and N is odd. A list of such integrals may be found in Bronshtein and Semendyayev
(pages 423-4241 and a longer list in Gradshteyn and Ryzhik [Sections 2.21-2.24].
Semidifferentiation using a lower limit of zero yields
d'/= c b
12:10:7 bx+c= + I d, >0
dx''2 ax Tr
and
dl/' I 1
2 :1 0 8 c> 0
bx+c ,a
1 :
dx12 bx+c
The function d' in 12:10:7 is given by
and appears also in the expressions for the semiintegrals of the two functions, again with zero lower limit:
d- '1,2 cx bx + c
1 2 :1 0:1 0 bx + + c >0
ds 7F
d-'/2 1 2
12:10:11 c>0
dx-'/2 bx + c sIbl
When the real argument x of the square-root function is replaced by a complex argument x + iy. the function
becomes complex valued:
+X +)'* -X
12 : 11 : 1 x + iy = + sgn(y) i 2
2 )r
Similarly:
1 x +y +s 1777
12:11:2 - sgn(y)i
x + iy 2(x2 +?) 2(x2 + y 2)
97 THE SQUARE-ROOT FUNCTION bx + c AND ITS RECIPROCAL 12:14
12:12 GENERALIZATIONS
The functions discussed in Chapter 13, being fractional powers of x, generalize the functions x-rn, which are the
simplest fractional powers.
,0
44 FIG 12-4
The curved portion of the perimeter of the shaded area is found from
(d
12:14:2 curved perimeter = 2 1+ bt + c I dt
f,.,b `- /
2\/c b _\/b
b )
b
This length is to be increased by 2V'bx + c, the length of the x ordinate, to give the total perimeter of the shaded
region in Figure 12-4.
CHAPTER
13
THE NONINTEGER POWERS x"
Relationships in science and engineering frequently involve fractional exponents. In this chapter we deal with the
power function x', where v is unrestricted.
13:1 NOTATION
When v is the reciprocal 1/n of a positive integer other than one, the symbol is sometimes used to mean x'
or k'i, but we employ only the latter notations in this Atlas. The names cube-root, fourth-root, fifth-root, etc., are
given to x1/3, x1/4, xit3, etc. The symbols x-' and 1/x' represent identical functions.
13:2 BEHAVIOR
The behavior of x' is affected by whether or not the number v is rational. If v is irrational (that is, incapable of
being expressed as the ratio m/n of two integers), x" is defined only for x ? 0 and takes positive values only. In
other words, a graph of x" versus x lies entirely within the first quadrant [see Section 0:2 and Figure 0-1 for a
definition of the four quadrants). Two irrational examples, x° and x°t4, are plotted in Figure 13-1. Notice by
comparing this figure with Figure 11-2 that the x° curve ties roughly where one would expect: between the x3
and X-4 curves, but closer to the former.
If v is a rational nonzero number it may, by definition, be equated to m/n where n is a positive integer and
m is a nonzero integer of either sign. When, by cancellation of any common factors, the m/n fraction has been
reduced to its lowest terms, it can be classified into three cases according to the parities of m and n: even/odd,
odd/odd and odd/even (any even/even fraction must be reducible by dividing the numerator and denominator by
2 a sufficient number of times). We use 1, i and j as typical examples of these three classes of v values and display
the corresponding x" graphs in Figure 13-2.
When m is even but n is odd, x'"l" is defined for all arguments x but takes positive values only since it is the
square of the real function x'1'". Hence the graph of x" versus x occupies the first and second quadrants.
The function x" = x'1" with m and n both odd is defined for all values of x and takes the same sign as x.
Hence, this type of x" graph ties wholly within the first and third quadrants.
When m is odd and n even, the x''l' function is restricted to nonnegative arguments but adopts both positive
and negative values; that is, it is a two-valued function. Accordingly, its graph occupies the first and fourth quad-
rants.
99
13:2 THE NONINTEGER POWERS x" 100
101 THE NONINTEGER POWERS x' 13:5
13:3 DEFINITIONS
When v is the reciprocal of an integer n greater than unity, the definition of x" relies on the concept of an inverse
function [Section 0:3]. Thus, for each x, P. equals any real number that generates x when raised to the power n;
that is:
13:3:1 x11" =f where f" - x n =2,3,4,...
This condition is satisfied by a unique real value off when n is odd. (In addition it is satisfied by other n - I
complex values off, as explained in Section 13:14.) When n is even. 13:3:1 is satisfied by exactly two real numbers
f (equal in magnitude but opposite in sign) when x is positive, but by no real number at all when x is negative.
(Once again a supplementary set of complex values off exists, giving a total of n solutions to 13:3:1.)
Rational powers other than x'/" are all expressible as x"/" and are defined as the m" power [as in Chapter I I ]
of x'1"; thus:
13:3:2 X.I. = (.x'/")" m = ±2, ±3, ±4, ...
This definition applies irrespective of the parities of m and n.
When v is irrational, the definition of x' presents greater difficulties because it must be expressed in terms of
a limiting operation. Let m,/n m2/n,, m3/n7, ..., m//n/, ... be fractions that constitute ever more accurate ap-
proximations to the irrational number v. Then x' is defined as
13:3:3 .r' = lim X"""
For example, x' could be defined as the limit of a sequence of positive numbers of which the initial members are
311159/100000 .--
x3 x31110 x157/50, x157/500' X3927/130.
13:5 LNTRARELATIONSHIPS
If v is irrational, the function a' is defined only for x ? 0 [as is explained in Section 13:21 and no reflection
principle is possible. The same is true when v = m/n and n is even. For odd n, however, one has
13:5:1 (-x)"/"=(-1)"x"/" m=±1.±2. ±3. ... n= 1.3.5,...
The following laws of exponents hold for all values of v and µ:
13:5:2 X'T' = x''"
13:6 EXPANSIONS
The series
13:6:1
x'=[I+(x-l)1'=I+v(x-l)+v(v2
1) (x-1)`* 0<x<2
where (,) is the binomial coefficient [Chapter 61 terminates only if v is a nonnegative integer.
Unless v is a negative integer jr' may be expanded as an infinite series of Bessel functions via 53:3:1 and
53:14:5.
v 0 (-1)" I 0
n ln=1,3,5,...
of Jn
v - l -l, -2. -3,
a n=2.4.6.... undef undef = _1 0
v
nfm =1,2.3_] under undef 0 ±1
n2,4,6.... J
Al
v=-m n-1,3,5,...
n1 . 2, 3. ... (-1)"s (-1)" 0 1
13:9 APPROXIMATIONS
Differentiation gives
13:10:1 X. = vx'-
dx
13:10:2
r
I
t'dt=-
x"
v+l v> - l x >0
and
-( -.t)" i
c-'.' bx
b""
B v+ l;-µ-v+1; bx+c
b>0 0<bx .< c
c>0
13:10:6 t"(bt + c)"dt =
c
'
B
v+l;µ+):-bx
b<0 0<-bz <c v>- 1
(-b)"" c
13:10:7
d'"' _ r(v + 1) x,._v>-1 x>0
5T'n -r(v+'lz)
or
respectively. Here r denotes the gamma function [Chapter 43). Similarly, semiintegration produces
d r(v + 1)
x""1' v>-1 x>0
dx-'/- x' = r(v + 3h)
For example. provided s > I, the Mellin transform of I /[exp(x) - I) is r(s)C(s) [see Chapters 26, 43 and 64 for
the exp, r and { functions]. A tabulation of over 250 Mellin transforms is given by Erdelyi, Magnus, Oberhettinger
and Tricomi [Tables of Integral Transforms, Volume 1, Chapter 6], together with some general transformation
rules. Via the substitution jin(x)I = t, one can convert definition 13:10:13 into
and thereby use tables of Laplace transforms [Section 26:14] to calculate many Mellin transforms. For example,
the Mellin transform of the function f(x) = In'(x)u(I - x) [where the In and u functions are those treated in Chapters
25 and 8) may be obtained by noting that, for 0 < t < -. f(exp(t)) = 0 while f(exp(-t)) = (-t)". Equation 13:10:14
then shows that fM(s) is simply the Laplace transform of (-t)', namely -n!/(-s)"".
105 THE NONINTEGER POWERS x' 13:14
where Ln is the multivalued logarithmic function discussed in Section 25:11, expresses an arbitrary real power of
a complex number. When v is a nonnegative integer, 13:11:1 reduces to 11:11:1. If v = m/n where in is an integer
and n = 1, 2, 3, ..., formula 13:11:1 defines n complex numbers. The m = 1. y = 0 case is elaborated in Section
13:14.
When v is irrational, the power function (x + iy)" adopts infinitely many complex values. Making use of
relationship 13:11:1 and equations 25:11:1. 25:11:2 and 25:11:3 dealing with the logarithmic function of complex
argument, one finds
13:11:2 (x + iy)" = (x' + y-)'''-[cos(6) + isin(4)]
where
13:11:3 46 =v(0+2ka) k=0,±1, ±2,...
and
sgn(Y)arccot(x/IIyI) Y#0
13:11:4 it
2-[1 - sgn(x)] y = 0
13:12 GENERALIZATIONS
The functions (bx + c)' and (ax' + bx + c)' generalize x", and their properties can sometimes be deduced by
appropriate substitutions.
from which these values may be calculated, is derived by setting y = 0, v = 1/n in equation 13:11:1, utilizing
the expression Ln(x) = ln(IxI) + i[2kir + ar(I - sgn(x))/2] (which derives from 25:11:1) and finally replacing the
exponential function via Euler's formula, 26:11:1.
13:14 THE NONINTEGER POWERS x' 106
(1 V/3\1
18'"R1cos(-a/3) + i sin(-a/3)] = 212 + rl 2 IJ - 1 - V3-i
(1 V /
1
13:14:3 (-8)'"' = 181/31[cos(a/3) + i sin(a/3)] = 212 + it 2 I = 1 + V-3i
Together with the parabola [Chapter 12] and the hyperbola [Chapter 15]. the ellipse discussed in this chapter
completes the family of curves of the second degree. These are also known as conic sections because they are
generated by intersecting a cone with various planes.
14:1 NOTATION
x2)-i' I/b
As explained in Section 12:1, the notations b,(a2 - x')1721 and 1(a2 - are equivalent to b a - x and
l /(b a - x ), respectively. The parameters Jal and lab] are known as the semiaxes of the by a s function,
the larger being the semima'or axis and the smaller the semiminor axis.
A graph of tb a - x- versus x is termed an ellipse, and the adjective semielliptical is therefore appropriately
applied to the b a - x function. Some geometric properties of the ellipse are discussed in Sections 14:3 and
14:4.
14:2 BEHAVIOR
The functions W-7 - x- and 1 /(b a - x) are defined (i.e., take real values) only for -la] K x < Jal. They
adopt the sign of b. Map 14-1 shows typical shapes of the two functions when b is positive and less than unity.
Typical shapes of b\/a2 - x- when b = 1 and b > I are depicted in Map 14-2.
14:3 DEFINITIONS
The algebraic operations of squaring [Chapter I l] and takin the square root [Chapter 12], together with arithmetic
operations, fully define the semielliptic function b a2 - x' and its reciprocal.
A parametric definition [Section 0:3] of the f = ±b a' - x function in terms of trigonometric functions
[Chapter 32] is provided by
14:3:1 f = ab sin(w) x = a cos(w)
An ellipse may he defined geometrically in two distinct ways. The first defines an ellipse as the locus of all
points P whose distance PF from a fixed point F (called a focus of the ellipse) and whose distance PD from a
107
14:3 THE bV a - s' FUNCTION AND ITS RECIPROCAL 108
1
b al b>1
blal
FIG 14-1
b<1
Let the x-axis of a cartesian coordinate system be chosen to lie along the shortest straight line joining the focus to
the directrix and let X be this shortest distance between F and DD. Create the origin of the coordinate system at
point 0 (see Figure 14-3] such that the length of the line OF = >\e2/(1 - e'). Then the equation of the ellipse is
f=tb a- x' where b -e- anda=>,e/ l-E
The second geometric definition of an ellipse is as the locus of points P such that the sum PF + PF' of the
distances from P to two points F and F' (the foci of the ellipse) obeys the relationship
14:3:3 PF + PF' = a constant = o
Of course, a must exceed the interfocal distance FF' = 4. If a cartesian coordinate system is erected with its
109 THE 6V.' -72 FUNCTION AND ITS RECIPROCAL 14:7
origin 0 at the midpoint of the line FF', which is chosen as the x-axis, then the equation of the ellipse is
again ±b a- - x- where a = a/2 and b = I - ((b/a)'.
14:5 INTRARELATIONSHIPS
The function f(x) = W a - x is an even function, that is, the reflection relationship
14:5:1 f(-x) = f(x)
is satisfied.
The multiplication formula
14:5:2 f(vx) = by (a/v)' - x'
shows that multiplication of the argument of a semielliptic function by a constant generates another semiellipse,
one semiaxis being unchanged. Choosing v = 1/b generates a semicircle of radius jabl.
14:6 EXPANSIONS
Two useful infinite expansions are
x
a'-x==la1b2a'1 -----
'16
x
- 128x°
14:6:1
X2 8a' 16a°
\x'1
and
14:6:2
I
ba' - xZ
=
I
jajb
1+-+-+-+-+ = l
x'
2a'
3x'
8a'
5x°
16a°
35x°
128x"
1
Ialb J=u \
-1/2
J
lI ax1 -a<x<a
lalb
14:8 THE bVa' - x' FUNCTION AND ITS RECIPROCAL 110
14:9 APPROXIMATIONS
Based upon truncation after the first two terms, expansions 14:6:1 and 14:6:2 yield
b(2a= - x=)
14:9:1 W;' x2 =
21al
8-bit precision
I tar+x: x
14:9:2 8-bit precision < 0.10
b a' - x2 2bla' a
14:10:1bVa'- -bx
ax2
d l x
14:10:2
dx bti a b V(.2 -!x2)'\
14:10:3
b't'dt?a-+ laresinlIQ.I -lalsxal
J \\\
and
r dt I x
14:10:4 = b aresin (am) -lal s x s lal
Jo b1 a //////
The integrals
14:10:7
dt dt
0 r<x<la
-0f f--.,lt ar-lr+J. t a'-t'
Among other important integrals are
dr
14:10:8 F(p;aresin(x/a))
o a 2 - tZ a' - p'r' a
Ill THE b a= x2 FUNCTION AND ITS RECIPROCAL 14:11
and
a' - p2at
14:10:9 p2P = a E(p;arcsin(x/a))
Ja v7=1
which serve as definitions of the incomplete elliptic integrals of the first and second kinds [see Chapter 621.
Semidiffcrentiation produces
1/2
dxb2=b {2E(P;O) 0<x<a
dl/2 1 I 1
=
1
0<x<a
dx'n b a' - x' b(a + x) 2a a-z a ax
The functions F(p;8) and E(p;O) that occur in the above semiderivatives are incomplete elliptic integrals [Chapter
621 of argument 8 = aresin( 2x/(a + x)) and parameter p = (a + x)/2a. The same functions occur in the
expressions
u2
b r
14:10:12 ax-1J1 b I (a - x)F(p;8) + 2xE(p;8) - (2x - a) 0<x<a
3Y ?a]
and
d U2 1
14:10:13 F(p;8) 0< x< a
dx b
14:11:1 b a2-(x+iy)2= _ b
(a+x) +y +y +a2_x2+y2
ib sgn(xy) y2
VV5 +
v1
or
where the sgn(.ry) function [Chapter 81 equals the sign of the xy product, or is zero if either x or y is zero.
When the argument is purely imaginary, equation 14:11:1 reduces to
14:11:3 b a''-(iy)2=b1/'+y2
which corresponds to a semihyperbola [see Chapter 15].
14:12 THE b\/ a= - x' FUNCTION AND ITS RECIPROCAL 112
14:12 GENERALIZATIONS
The ellipse, b(a2 - x2)' 2, is the special n = 2 case of the more general function b(a" - x")'"". When n is even
and greater than 2, the curves obtained by plotting b(a" - x")"" versus x have been called superefiipses: the cases
n = 2. 4 and 8 are depicted in Figure 14-4. As n approaches infinity the superellipse shape becomes more and
more rectangular.
FIG 14-4
n=8
The reader is referred to Section 15:12 for a discussion of the circumstances under which the function
ax + bx + c is a generalization of the semiellipse W a- - x--
f a z o-z
while the entire semicircumference has a length ira. An important property of a semicircle is that the angle APB
in Figure 14-5 is a right angle for any point P on the semicircle. Thus the triangles APB, AQP and PQB are similar.
As explained in Section 14:3. the ellipse ±bV a - r- for b < I has foci F and F' located on the x-axis
at x = waV I b . The sum of the distances to each focus from any point P on the ellipse is a constant. 2a.
Moreover, the lines PF and PF' make equal angles with the ellipse itself at point P, as is illustrated in Figure 14-
6. This means that any wave motion that originates at one focus F' and is reflected at the surface of the ellipse
will be "focused" at the other focus F and all the reflected waves will arrive at F at the same time, though from
a variety of directions [Figure 14-61. This property of an ellipse is exploited in furnace design and in several acoustic
and optical devices.
FIG 14-6
A segment of an ellipse, such as that shaded in Figure 14-7, has an area that can be determined with the help
of integral 14:10:3 as
area of r rr x1 xV a' -- x 1
14:14:3
shaded segment
= 2
J Q fdi = a'1bj
[ 2 + aresinl I + a I f=b a'-r=
a
O0
FIG 14-7
The entire ellipse thus has an area ira`Ibf. The curved perimeter of the shaded area in Figure 14-7 may be evaluated
in terms of an incomplete elliptic integral by utilizing equations 14:10:1 and 14:10:13:
curved = 2 1 - br; 2 //
14:14:4 1+ y dr = 2aLE 1 - b2;aresin (- f/ /I - EI\
perimeter Ja JJ
Because [see Chapter 621 E(p;7r/2) = -E(p;-a/2) = E(p) where E(p) denotes the com lp ete elliptic integral
[Chapter 611 of p. the entire perimeter of an ellipse with semiaxes ial and Jab equals 4aE( I -- bZ).
CHAPTER
15
THE b\/x2 + a FUNCTION AND ITS RECIPROCAL
Al hou h the properties of the b x + a function depend on the sign of the constant a, a graph of
b x + a is a curve called a hyperbola whether a is positive or negative, the sign of a determining only the
orientation of the hyperbola.
15:1 NOTATION
The notations bl(x2 + a)'I2I and I(xz + a)-"21/b are equivalent to b x -+a and I /(b x + a), respectively. Be-
cause the curve ±b\ +x a is a hyperbola, the adjective semihyperbolic is appropriately applied to the
b x + a function.
15:2 BEHAVIOR
When a is positive, bVx- a and its reciprocal are defined for all values of the argument x and both functions
share the sign of b. Figure 15-1 shows typical behaviors of these functions when b is positive.
115
15:3 THE bV x- + a FUNCTION AND ITS RECIPROCAL 116
When a is negative, neither b x -+a nor its reciprocal is defined in the interval -V a < x < V - a and
therefore each of these functions has two branches, as illustrated in Figure 15-2 for positive b.
As x -+ x, the function b x` + a acquires values ever closer to bx. The linear function bx is said to be an
asymptote of the bV +x function. Similarly, -bx is an asymptote of b x' + a as x -> -x. These asymptotes
are shown in Figures 15-I and 15-2.
15:3 DEFINITIONS
Algebraic and arithmetic operations define b x + a and its reciprocal in a straightforward fashion.
A hyperbola may be defined geometrically as the locus of all points P such that the length of the line PF
connecting P to a fixed point F (a focus of the hyperbola) obeys the relationship
PF a constant _ the eccentricity _
15:3: I
PD = greater than unity of the hyperbola - e 1<E<x
with respect to the length PD of the shortest line from P to the straight line DD (called a directrix of the hyperbola).
The eccentricity of the hyperbola ±bV +x is 1 + b-. An illustration of relationship 15:3:1 is shown in Figure
15-3. Note that the hyperbola of Figure 15-3 has two branches and that there exists a second directrix D'D' and
a second focus F' symmetrically positioned with respect to a line parallel to both directrices and midway between
them.
An alternative geometric definition of a hyperbola is as the locus of all points P such that
15:3:2 IPF - PF'I = constant
where F and F' are the foci.
If a function g(x), interpreted as a cartesian graph of g(x) versus x, is rotated through a positive angle 0 in a
counterclockwise direction about the point (X,G) then a new function f = f(x) is created that is related to g(x) by
the general rotation formula
Similarly, the equation 0 = -x ± V(f)2 a, which arises from setting g(x) to ±V +x a and 0 to -e/2 in
equation 15:4:2. reduces to
15:4:4 f= ± x'"-a
Thus, the three functions x V +x a, a/2r and ±\/x a all have exactly the same shape and differ only in their
orientations, as illustrated in Figure 154. The function -a/2r is also a member of the quartet of identically shaped
rectangular hyperbolas.
It ma be apparent from Figure 15-4 that the curve f(x) = ±Vlx' - a can be obtained from g(x) _
± x' + a not only by rotation but also by reflection in their common asymptote. A very general result states that
the reflection of the function g(x) in the straight line bx + c generates the function f = f(x) where
2bf + x(1 - b2] - 2bc
15.4.5 11-b2[f=2bx+2c-II+b2[g(v) y= 1+62
When c = 0 and b = 1. this general reflection formula collapses to
15:4:6 .x = g(f(x))
showing that the reflection of a function g(x) in the straight line x generates the inverse function [see Section 0:31
of g(x). Hence, the rectangular hyperbola ±V a is the inverse function of the rectangular hyperbola
and conversely.
15:5 THE W x* * a FUNCTION AND ITS RECIPROCAL 118
The f(x) = b x ++ a function is even and therefore satisfies the reflection formula
15:5:1 f(-x) = f(x)
The multiplication formula
15:5:2 f(vx) = bpvl + a/(v'')
shows that multiplication of the argument x of the semih perbolic function by a constant generates another semi-
hyperbolic function. The inverse function of ±b x + a is also another hyperbola
±1
15:5:3 f= b
x2-ab' where x= tbVj'+a
The two hyperbolas tb -x + a and ±bV? - a are said to be conjugate to each other. As shown in Section
15:4, they share the same asymptotes.
15:6 EXPANSIONS
when x is large. It is this latter equation that describes the approach of the semihyperbola to its linear asymptote.
119 THE b x= a FUNCTION AND ITS RECIPROCAL 15:9
- I ( a 3a 5a' 35a'
15:6:4
b x= + a i ix- 8x' 16x° + 128x' j X ) ,
-
x--
I
a x= a
b' b'
X= ab-, s I x=- V iaj s= 0 ab' s I .c
b+a
a>0
1
0 sgn(b)
I
bV 2a
I
bVn b I
sgn(b) 0
15:10:2 -dxW;'+ .
d I
_
b
-x
(x'+a)'
The simplest formulas for indefinite integration must employ differing limits according to the sign of a. Thus:
ab x
15:10:3
1.
b
bx
t2+adt= 2 z2+a+ 2 arsinh V; a>0
15:10:5
dt I
_ b arsinhl
/x\ I
Jo b t -+a b a ->O
dr
15:10:6 = I arcosh(' x 1 a< O
J .-e b r'+a b V-a///
15:10:7
f t'+a
dt
dt
'arcsch(
1 // `
a>0 x>0
I a
15:10:8 a<0
-,V,' +a -a arcsecl V-a/
where the functions arsinh. arcosh. arcsch and arcsec are discussed in Chapters 31 and 35. These six indefinite
integrals are simple members of the class ft"(ti7 -+a)" dr where n = 0. ± I. ±2. ... and m = t 1, ±3, ±5, ...:
a long list of such integrals will be found in Gradshteyn and Ryzhik [Section 2.27]
If the argument in 15:11:1 is purely imaginary, the function reduces to a semiellipse (Chapter 14]
15:11:3 b (ry)' + a = b (V)' - y'
provided that a is positive.
15:12 GENERALIZATIONS
The function b (x - a) + a differs from b x- + a only by being translated a distance a alon the x-axis; it is,
therefore, a semihyperbola centered at x = a rather than at x = 0. The function ax + bx + c may similarly
represent a translated semihyperbola; however, this so-called root-quadratic function may represent any conic sec-
tion (or degenerate instances of the conic sections) according to the magnitudes of the coefficients a, b and c. The
situation is summarized in Table 15.12.1.
Table 15.12.1
-
a>0 Va Ix+ -I( Vb
ff
11
b 11b'
2a
4ac
./
(Vax+Vc) V.a
\ 2a/
lx+bl + 4ac - b'
4a=
a*1 Semihyperbola of two branches Two straight lines Isee Chapter 7] Semihyperbola of one branch
centered at x = -b/2a [see this centered at x = -b/2a (see this
chapter] chapterl
// ``
IIII z / /
a=] Ix+21- l d -e) --(x+V) Ix+ +Ic
`` 2/ 4
Rectangular semihyperbola (see Two straight lines of slope unity Rectangular semihyperbo[a of one
Section 15:41 of two branches branch [see Section 15:41 centered
centered at x = -b/2 at x = -b/2
a=0 'VV_-C "ire
Semiperabola [see Chapter 12] Constant [see Chapter I] Not possible
\
centered at x - b/2 [see Section
14:4]
V 'b'
a<0 Ix 1
- + 2a)
a * -1 Semiellipse with semiaxes of Not possible Not possible
V b- - dac (-2a) and
Vc - centered at .r =
-b/2a [see Chapter 141
is shown shaded in Figure 15-5 and may be evaluated directly from the integral 15:10:4 as
15:14:1
shaded
= 2 L Ibt+ a Idt = b x+ a + aIbIarcosh
rx 1
J a<0
area
15:14:2
rimeter
perimeter
=2 f' 1+ (b/')iJ 2 -a - ab2
IT Z F( I+b
.40)
/ I xZ+a 1 +62
- EI I b ,0 1 + X sin(0)1 4o = arctanl b ) a<0
b t2+a
FIG 15-5
Also shown in Figure 15-5 are the asymptotes OS and OT of the hyperbola and the tangent MN to the hyperbolic
branch at an arbitrary point P (M and N being the points at which the tangent meets the asymptotes, as depicted
in Figure 15-5). Two remarkable properties of the hyperbola are that P bisects the line MN, so that MP PN,
and that the area of the triangle MNO equals JabI and is independent of the position of P along the hyperbola.
CHAPTER
16
THE QUADRATIC FUNCTION
axe + bx + c AND ITS RECIPROCAL
Next to the constant function [Chapter 11 and the linear function [Chapter 71. the quadratic function is the simplest
of the large class of polynomial functions; others are discussed in the next eight chapters.
16:1 NOTATION
The parameters a, b and c are called the coefficients of the quadratic function ax' + bx + c. The value of the
quantity
16:1:1 A=4ac-b'
called the discriminant of the quadratic function, strongly influences the properties of the function and its reciprocal.
Some authors define the discriminant to have the opposite sign to that in 16:1:1.
16:2 BEHAVIOR
Graphs of ax' + bx + c are shown in Figure 16-1 for a > 0 and a < 0. The function is defined for all x and, for
positive a, assumes values that are not less than c - (b=/4a). Whether the value zero is part of the range of the
quadratic function is determined by the sign of its discriminant.
The behavior of the reciprocal quadratic function is profoundly affected by the sign of J. Figure I6-2 shows,
for a > 0. typical graphs of 1/(ax' + bx + c) when the discriminant is positive, zero or negative.
16:3 DEFINITIONS
The quadratic function is defined by the arithmetic and algebraic operations signified in the expression ax' + bx
+ c. The reciprocal quadratic function is defined as the quotient of unity by the quadratic function.
A reciprocal quadratic function may also be defined as the difference of two reciprocal linear functions [see
123
16:3 THE QUADRATIC FUNCTION ax2 + bx - c AND ITS RECIPROCAL 124
21
FIG 16-2 $
ox +bx+c
40
4ob
:4oc<0
16:3:1 =
bx+C bx+c b / Cc
c-C\1x+1
C1x''+b(
c- c-C c- C
Not every reciprocal quadratic function may be defined in this way, however, since the decomposition
125 THE QUADRATIC FUNCTION ax' + bx + c AND ITS RECIPROCAL 16:5
16:3:2
ax'+bx+c b
b'-4acx+2c+2a( b2-4ac-b)
I
b
b2-4acx-2c+-( b'-4ac+b)
is possible (using real arithmetic) only if the discriminant 4ac - b'- is negative.
16:5 INTRARELATIONSHIPS
Unless A = +a. the sum or difference of two quadratic functions
16:5:1 (ax2+bx+c)±(Ax'+B.x+C)=(a±A)x'+(b-B).r+c_C
is another quadratic function, while their product
16:5:4
16:6 THE QUADRATIC FUNCTION axe + bx + c AND ITS RECIPROCAL 126
16:6 EXPANSIONS
j
c2 c3 cs
x - r; x - r, -
-b+Vbb-4ac
2a 2a
x=- when b' > 4ac x-0 when b' > 4ac x . x
(4ac - b')/4a
min if a> 0
ax' + bx + c x sin(a) 0 0 58n(a)
$ max if a < 0
4a/(4ac - b')
0 x rmax if. > 01 i
0
ax'+bx - c Sl
min if a < 0 1 c
The values of x that make the quadratic function equal to zero are known as the zeros of the quadratic function or
as the roots of the quadratic equation axe + bx + c = 0. Note that "root" is used here in a sense different from
that of Chapter 13. There are two zeros
-b+ b2-4ac
and r; _
-b- b'-4ac
16:7:1 r1 _
2a 2a
16:7:2 r =-b
-
2a
if 4ac = b2 and no (real) zeros if 4ac > b2. The reciprocal quadratic function becomes infinite at the zeros of the
quadratic function, the so-called poles of the reciprocal quadratic function.
16:9:1
I c-bx Icf
z
8-bit precision Ixl <
16 ibz - ac + abxi
b' -ac+ -I
be
16 I
ax -b x
8-bit precision jxj >
axz +bx+c a:x3 lal
Differentiation produces
d
16:10:1 d (axz + bx - c) = tax + b
and
d I -tax - b
16:10:2
dx ax' + bx + c (ax'' + bx + c)z
while integration results in
7nr3 + 1hr2 + frr
16:10:3 (at' + bt + c)dt =
u 6
and
(2ax b)
2 + = 4ac - b' > 0
dt
16:10:4
J. ,atz+bt+c -2 r2ax+bi -b
artanh ` I <0 x<
I ax'+bx+c l b x
l
nIr 2a
In
/ C
arctan
bx + 2c1
9=4ac-b'>0
aVa
16:10:6 _
u at-+bt+c
,
I h&+bx+c\ b /xv-1
Inl I- artanhl I 1<0
2a ` c / a \bx + _c/
and others of the general form f t"(aY + bt + c)'dt, where n and m are integers, will be found in Gradshteyn and
Ryzhik [Section 2.171.
16:11 THE QUADRATIC FUNCTION ax' + bx + c AND ITS RECIPROCAL 128
Replacement of the real argument x of the quadratic and reciprocal quadratic functions by the complex argument
x + iy leads to
16:11:1 a(x + iv)' + b(x + iv) + c = (ax- - ay' + bx + c) + i(2axy + by)
16:12 GENERALIZATIONS
The quadratic function is a special can of the functions treated in Chapter 17. A quadratic function is a member
of all the function families addressed in Chapters 18-24.
The root-quadratic function V axx + bx + c and its reciprocal are functions of some importance. It is demonstrated
in Section 15:12 that the root-quadratic function is equivalent to various conic sections depending on the magnitudes
of the a. b and c coefficients. Some integrals involving the reciprocal root-quadratic function are
16:13:1
r'
di
Lv- I Vat' +bt+c
dr =
l -l
1= arcosh l
Ic
(
Iarsinh 1
a
aresin
tax + b\
2ax + bl
tax + b\
V
a>0
a>0
a<0
A=4ac-b'>0
1<0
A<0
tax + b
1 a
V-b
s 1
aI+br+c 'va
16:13:2
J \ V_A ) 2a
bx + 2c
vcarsinh
l
a>0 0=4ac-b'>0
xV
16:13:3
Jzelbt atbt+c 1
arcsin
b x + 2c b x + 2c
y-c (x ) xV-O
and
arcoshrbx - 2c
c, > 0 2c
16:13:4
got at+bt+c \
_ -1 p<0 x0 =
0b
Others are given in Gradshteyn and Ryzhik [Section 2.26].
v2sin(2e)
The function a"x" + a"_,x'-1 + - - + a,x + ao, in which the a multipliers arc specified constants, is known as
a polynomial function of degree n. This chapter treats such functions generally and pays particular attention to the
it = 3 case with, for added simplicity, a3 taken as unity.
17:1 NOTATION
A polynomial function is often simply called a polynomial; sometimes the term integral function is used to describe
the same family.
We shall use the general notation
to denote a polynomial of argument x and of degree n. The constants a,,, a"_,, - - . , a,, ao are called the coefficients
of the polynomial function; there are it -1 such coefficients, some of which may equal zero.
Special names are given to polynomials of degrees 5, 4, 3, 2, t or 0; they are known as quintic functions,
quartic functions, cubic functions, quadratic functions, linear functions or constant functions.
The quantity
17:1:2 D = Qt - P3 where P=
a2 b
-
and Q=-----
ab c a`
9 27
2
is known as the discriminani of the cubic function r3 + ax= + bx + c. Its value affects the number of (real) zeros
of the function as explained in Section 17:7.
17:2 BEHAVIOR
Polynomial functions are defined for all values -- < x < x of their arguments. Those of odd degree acquire all
real values, whereas polynomials of even degree have a restricted range.
Figure 17-1 shows the three possible shapes that the cubic function x3 + are + br + c may exhibit according
to the sign of a2 - W. Notice that an inflection [see Section 0:71 occurs at x = -a/3 in all three cases. A local
maximum and minimum occur on either side of this inflection if, but only if, a' > 3b. (i.e, if P, given in 17:1:2,
131
17:3 THE CUBIC FUNCTION x' - ax= + bx - c AND HIGHER POLYNOMIALS 132
is positive). The cubic function always acquires the value zero at least once: that is, there is at least one real root
to the equation p3(x) = 0. Section 17:7 discusses the circumstances under which a second or third mat root exists.
The behavior of a polynomial of higher degree is determined in a complicated fashion by its coefficients and
by its degree. Generally, a graph of a polynomial p (x) exhibits a number of inflections (at the zeros of the poly-
nomial of degree n - 2 that is the second derivative of p (x)-see Section 0:7), and local maxima and minima (at
the zeros of the polynomial dp,(x)/dx of degree n - 1). Some general rules concerning the number of zeros of
p,(x) (i.e., the number of real roots of the equation 0) are given in Table 17.2.1.
17:3 DEFINITIONS
The operations of addition, multiplication and raising to a power fully define a polynomial. Written as a conca-
tenation
133 THE CUBIC FUNCTION x' + ax, - bx + c AND HIGHER POLYNOMIALS 17:5
Table 17.2.1
17:3:2 1 1 (bx + c)
defines a polynomial of degree n, but not every polynomial function of degree it may be defined in this way.
However, every polynomial of degree n can be defined as a product of 1 linear functions and (n - 1)/2 quadratic
functions [Chapter 16]:
r K
n f
17:3:3 P,(x)=Fj (b,.r+c) F1 (AkxZ+Bkx+Ck) K=Z -Z
.l k=1
where I = N the number of (real) zeros of p,(x). The identification of a suitable set of linear and quadratic
coefficients {b,,c,,Ak,B,,C5} from a given set of polynomial coefficients a, will generally require iteration (or in-
spiration!).
17:5 INTRARELATIONSHIPS
The cubic function p3(x) = x3 + axt + bx + c obeys the reflection formula
/ a 2ab
17:5:1 p31 3 x = 4a3 a
27 - 3 + 2c - p, 3 + x
but no comparable formula exists for higher polynomials.
The polynomial function obeys the simple argument-multiplication formula
showing that multiplying (or, of course, dividing) the argument of any polynomial by a constant generates another
polynomial function of the same degree.
The addition or subtraction of two polynomials of degrees n and m yields a polynomial of degree equal to the
larger of n and m (unless m = n and the highest degree coefficients cancel); thus:
17:5:3
' ,
J
yields a polynomial function of degree n + in. The quotient of two polynomials is termed a rational function and
is discussed in Section 7:13.
The converse of 17:5:4-the factoring of a polynomial into two polynomials of strictly lower degrees-
is always possible for n a 3 because of 17:3:3. Thus, a cubic function may be factored into the product of a linear
function and a quadratic function
17:5:5 x3+ax2+bx+c=(x-r)[x2+(a+r)x-c/r]
where r is a (real) zero of the cubic function. (Further factorization of the quadratic function into a pair of linear
functions is possible if its discriminant [see Chapter 16] is nonpositive, i.e., if it has two real zeros). This factoriza-
tion of higher polynomials into ones of lower degree is the key to the decomposition of reciprocal polynomials
[see Section 17:13].
17:6 EXPANSIONS
A polynomial function may be expanded as a continued fraction
ao a,x aoa,x a,ax
17:6:1 01x1 =
,_o 1- ao + aix- a, + a,x- a2 + a3x- an_I + a,x
A polynomial function may be written as the product of n linear functions with complex coefficients
ipi-PI+P2+...+P.-I+P.=
-a,-,
17:6:3
a
17:6:4
n n n
i
k-1, l
PiPk'PIP2+PIPS+PIP4+...+P.-2P.-i+P.--P-
a,_2
-an-)
17:6:5 P,PsPs = P,P2P3 + PIP2P. + PIP2P3 + + P.-A.-IN-I + P.-2P.-&. _ a.
=1 t.k-I
135 THE CUBIC FUNCTION x' + axe + bx + c AND HIGHER POLYNOMIALS 17:7
I -(-1)"a2
+PIP3P4P.+P2P3P.=FT PJI-=
pl k=1 PA a.
(-1) a0
17:6:7 P,P2PI " ' P. _ I1 P, _
-I a.
The functions occurring on the left-hand sides of equations 17:6:3-17:6:7 arc the so-called elementary symmetric
functions of p1, p2, p3, ..., p. [see Kom and Korn, page 11]. An advantage in expressing the polynomial as in
17:6:2 is that its reciprocal can be decomposed thereby into a sum of n terms, known as partial fractions:
17:6:8 C a. iC'
p.(x)
=
J-,x - p,
where C,=
e=1Pi-Pt
1
t.,
Alternatively, C, may be equated to a./p;,(p,) where p.(x) denotes the derivative of p.(x) [see 17:10:1 or 17:10:41.
This decomposition must be modified if any of the zeros of p.(x) are repeated.
The polynomial displays an inflection at any argument x, that satisfies the equation
The numbers of inflections, minima, maxima and zeros depend not only on the degree of the polynomial but on
the values of the coefficients; see Table 17.2.1.
For the cubic function x3 + axe + bx + c, an inflection occurs at x = -a/3 [see Figure 17-I]. Provided a2
exceeds 3b. a maximum and a minimum occur at arguments (- a - 3b - a)/3 and ( a- - 3b - a)/3, respec-
tively. This cubic function has the single zero
if the discriminant D [defined in equation 17:1:2] is positive, and three distinct zeros
r, = 2VPI cos(4/) -
3
)
17:7:6 r_ = -2V IPI cosl d' + where ¢ = 3 autos I - I D<0
3 3
rr = 2V cos 0-
3) 3 J
if the discriminant is negative. The constants P and Q are defined in equation 17:1:2. When the discriminant equals
zero, r_ and r3 of 17:7:6 become identical and there are then two zeros:
_ a
r,-2Q
3
17:7:7 D=O
_ a
-Q
but these coalesce to a single value if a' = 3b.
Formulas exist [see Abramowitz and Stegun, pages 17-18, for example] that permit the zeros of a quartic
function to be determined algebraically, but no such formulas are available for quintics or higher polynomials.
Numerical methods must therefore be used in these cases and, in practice, iterative numerical methods are applied
for these higher polynomials and, in fact, for quartics and even cubits as well.
A popular method for determining the zeros of a polynomial function p (x) = Ea .r! requires that each zero
first be located approximately (often from a crudely drawn graph of the polynomial versus x). The approximation
is then improved iteratively. The accompanying algorithm may be used to carry out the iteration. it requires values
of the n + 1 coefficients of the polynomial as well as a value of the degree n and ro, the approximate zero.
Improvement ceases when two successive approximants to the sought zero differ by less than one part in 10s. The
algorithm may fail to locate a multiple zero [see Section 0:21 and will fail to locate a zero at x = 0. It is based
on Newton's formula that gives
(3) Replace j by j - 1
Replace f by fr + ai
Ifj * 0 go to (3)
Setj=n
Setg=jq
(4) Replace j by j - I
Replace g by gr + ja,
Ifj*0goto(4) Test values: Input n = 4. a4 = 2, al = 5. a2 = 2, a,
Replace r by r[1 - (f/g)) = -9 and ao = -18. There are two zeros: r, = -2
if If I a 10-1 go to (1) and r2 = i. Any input ro < 0.5 will give r,; any input
Output r ro > 0.6 will give r2.
r <<<<< <<<<
These are easily calculated for the polynomial function, especially with the aid of the concatenation formula 17:3:1.
The simple algorithm
17:9 APPROXIMATIONS
One may construct a polynomial p (x) to approximate a polynomial p,(x) of higher degree over a specified interval
xo s x s x, by a process known as economizarion. The procedure is fully described in Section 22:14 and a versatile
algorithm is presented there for calculating the economized coefficients. The degree m of the economized poly-
nomial is determined by the acceptable error e of the approximation
17:9:1 Ip.(x) - P.,(x)I < a xo <- x <- x,
The economization procedure has five steps, each of which would usually be carried out numerically.
First, one replaces the argument of p,(x) by a new argument y = (2x - x, - xo)/(x, - xo)
(x, - xo/'
17:9:2 P.() a1x' _ b,y' where k= (x'
)-0 x, + Xo k=J \k/
f 2 x") ak
so that the region of interest comes to lie in the standardized range - I < y <- 1.
17:10 THE CUBIC FUNCTION x' - ax' + br + c AND HIGHER POLYNOMIALS 138
Second, the given polynomial of degree n is replaced by a sum of Chebyshev polynomials T, (y), where j takes
values up to n. The replacement uses equation 11:6:5 and leads to
and where y,°' is the coefficient of T, (y) in the expansion of ?. About half of these coefficients equal zero.
Now, because for -I s y :s 1 the absolute value of T,(y) never exceeds unity, it follows that
17:9:4 Ec,T,(y) s 7Ic,IIT,(y)I < Y1c,J -I s y s I
for any set of j values. Therefore. if m is the smallest integer that satisfies
17:9:5 14 <e
The third step in the economization procedure consists of finding the smallest integer m that satisfies 17:9:5.
The fourth and fifth steps achieve the reverse of the second and first steps, using the economized polynomial.
Thus, in the fourth step, one enumerates the coefficients B, defined by
L-0
where p)l"' denotes the coefficient of v' in the polynomial expansion of TR(y) [see Section 22:51. Finally, in the
fifth step, one determines
17:10:1
dx 7 ax' = F7ax'-'
i=o
f
i-)
17:10:2
and
f ,-0
('dr -
a,
1-0i + I
x-'
d' ax, r
a,
17:10:3 j
dx' _o
_ j 7
.,r(j-v+l) x' `
follow directly from the results of Section 11:10. The cubic function provides an exemplary subject for the three
operations, yielding 3x2 + 2az + b on differentiation, (3x' + 4ax3 + 6bx2 + 12cx)/12 on integration and (((3x/
(3 - v) + a)2x/(2 - v) + b)x/(l - v) + c)/x"f(l - v) on differintegration.
Operations of the calculus may also be carried out on the polynomial function and on its reciprocal via the
expansion 17:6:2. Thus:
d I I I
17:10:5
dx p,(x) p.(x)
1 x - pt p' .(x)
InI )
17:10:6
dt
C ,
P, p,
x
l + P, P)
where C, _
I
as in 17:6:8. Similarly, we have the Laplace inversion formula [see Section 26:14]:
1 =I exp(p,t)
17:10:7 exp(-st)dt
P.(s) Jo ,-i P.(p,)
sometimes known as the Heaviside expansion theorem. Recall that. in general, the p, need not be real. The four
equations above are useful only if each p, is a simple zero. This limitation does not affect the alternative methods
of handling reciprocal polynomial functions that are discussed in Section 17:13.
Historically, it was the need to integrate the root-cubic v p3(x) and root-quartic functions as well as
their reciprocals. that led to the discovery of elliptic integrals [Chapters 61 and 62). Section 62:14 may be consulted
for further details and for explicit formulas for fdt/\1±p3(t).
If the argument of the polynomial function is complex, the coefficients remaining real, then
+ (i)ax'-'y' + ...J
3 (,5
In the case of the cubic:
17:11:2 (x+iy)'+a(x+iy)2+b(x+iy)+c=x'+ax2+bx+c-y2(3x+a)+iy(3x2-y'+2ax+b)
A zero p, of the polynomial Ea,x' may have real and imaginary components even when the argument
x and the coefficients a, are real. If p, _ X + iµ where k and p. are real, then, invariably, there will also be a zero
pr = h - iµ. The complex numbers (X iµ) and ()L - iµ) are said to conjugate to each other.
17:12 GENERALIZATIONS
The power series discussed in Section 11:14 may in many respects be regarded as polynomials of infinite degree.
Sums of negative or fractional powers can often be expressed as polynomial functions of altered argument,
possibly with a power multiplier. For example:
and
17:12:2
-Uµ + aox 1/4 = x-"4P.(Vx-)
17:13 THE CUBIC FUNCTION x3 + ax' + bx + c AND HIGHER POLYNOMIALS 140
If m > n the rational function is said to be improper and may be resolved into a polynomial and a proper rational
function (i.e., one whose numeratorial degree is less than its denominatorial degree) by the procedure of algebraic
division (see Bronshtein and Semendyayev, pages 150-151].
To simplify a proper rational function, one first factorizes its denominatorial polynomial into linear and qua-
dratic functions, thus:
where each r; is a zero of p,(x) and the discriminant 4q, - pp of each quadratic factor is positive. Next, the r,
values are inspected to see if there are duplicates. Similarly, the p;, q, values are searched to see if there are any
repeated complex zeros. If all zeros of p,(x) are simple (i.e., of multiplicity one) then the proper rational function
may be decomposed into (n + 1)/2 so-called partial fractions as follows:
Ax'
17:13:3
=E a, +Ei "-t+ _n I
m<n
,.,x-r, -,x - pie+q, 2 2
The new constants in this relationship (the a,, P,, y of which there are n all told) are then determined, completing
the simplification- Most often the method of undetermined coefficients [see Bronshtein and Semendyayev, pages
151-153 for details and examples] is used to identify the constants, although other methods can be employed.
Equation 17:13:3 is appropriate only if there are no repeated zeros of p,(x). U two r values, say r, and r,. are
identical, then in 17:13:3 the terms
a, a, a, a2
17:13:4 + should be replaced by +
x - r, x - r2 x - r, (x - r,)2
If the three r values r, r, and r, are identical, that is, if p,(x) has a triple zero at r, [corresponding to d2p,(r,)/dx'
= dp,(r,)/dx = p,(r,) = 0, see Section 0:2]. then the terms
x-r, +x-r,
- + x-r,
a, a, a, a, a, a,
17:13:5 should be replaced by + +
x - r, (x-r,)2 (x-r,)S
Similarly, if p, = p2 and q, = q2, 17:13:3 is to be modified so that
,x+ y, x+ y, _ R,x' + -1, + 02x + 72
17:13:6 + should be replaced by
x2 - p,x + q, x' - p,x + q2 x2-Pox+q, (x2-p,x+q,)'
Otherwise the partial fraction decomposition is unchanged.
With R (x) defined as in 17:13:1, the infinite series R.'(0) + R.'(1) + R.(2) + can be summed provided
n ? m + 2. The procedure is described in Section 44:14.
A rational function frequently serves as a convenient approximation to a transcendental function. For specified
numeratorial and denominatorial degrees, m and n, the Pade approximant is the best such approximation for small
values of the argument x. If the transcendental function f(x) can be represented by the power series EA,x', then
the Ra Pade approximant is simply the sum of the first m + I terms in this series
17:13:7 g=A,x'
Below is shown an assemblage of Pade approximants to the exponential function [exp(x), see Chapter 26].
141 THE CUBIC FUNCTION x' + ax' + bx + c AND HIGHER POLYNOMIALS 17:13
X,
Ra(z)=I+x Rl(x)I+x+-2 R'a(x)=I+x+26
X2
Such an assemblage, extended indefinitely, is known as a Pade table [see Wall, Chapter XX]. Though they may
not always be optimal, the diagonal Pad' approtimants Ra(x). R:(x), RZ(x), ..., Rk(x), ... will be our especial
concern [see Section A:6]: we will report a general procedure for determining successive members of this family.
For the exponential function example. the diagonal Padd approximants are given by the formula
P`(x) ( 2 k - j)'xi
17:13:8 RU(x) = where pt(x) = Y
;-I (k - j)!j!
For the present purpose it is necessary to augment the Pad6 table by including entries corresponding to
;(x), where m and n are odd multiples of i, These new entries are not approximations to f(t), though they are
rational functions. Continuing with our f(x) = exp(x) example, we list below part of the augmented Pad' table:
R111;(x) = 0 It".'a(x) = 0 R "121AX) = 0
x'-+ i-
Rx)=l R,(x)=I+x Ra(x)'1+x+- Ro(x)=I+x+
2 2 6
1 2 6 24
R;!'(x) R;'(x) = R; ,`.'(x) = Ri;'(x) = z
x' s'
2-x 6+4x+ x' 24-I8x+6r+x'
RI(x) _ Ri(x) =
2-x 6-2x 24-6x
- 12 + 12x - 2x' -72 + 48x - 6r -480 4- 240x - 24x'
R3', ' -
t(x) Rii(x) - R (x)
x' x' X'
12+bx-.x' 60+36x+9x'+x'
R'.(x) = R'.(X)
12-6x+.r 60 - 24x + 3.r
Note that the identity
17:13:9 R"_'I1z(x) = 0
is true for all functions. Use of this rule and rule 17:13:7 enables the top two rows of the augmented Padi table
to be completed for any function f(x) whose power series is known. Every other entry in the augmented table may
then be found from those immediately above it by use of the rule [see Macdonald for details]
31 3
17:13:10 R"(x) = R;_1(x) +
1
m=-, 1,-.... 1
n= -. 1,-,...,m
RR-i12(x) - R, ; (x) 2
We will use the phrase "Padd operation on R;(x)" to describe the procedure, indicated by equation 17:13:10, by
which R; is constructed from R;_,. R,'-','/122 and
By carefully sequencing Pad6 operations, it is possible to generate an unlimited number of diagonal Pade
approximants corresponding to any power series. If we assign the verb "papoo" to mean "perform a Padi operation
on" then our goal may be achieved economically by the following sequence: load R. load K. null R' ; JZ, papoo
Rii, load R;,, null R-,/z, papoo R3,;;, papoo R load K, null 1111/_, papoo Ri, , papoo R;, papoo R3j , load R.
null R papoo papoo papoo R. papoo papoo R,"2!,
1/2 Papoo R; 2z, papoo R;, papoo Riii, papoo R, papoo R'/2, papoo k',
papoo R', papoo R55/;, load R,, null R-"/2,
etc.
17:13 THE CUBIC FUNCTION x' + ax' + bx + c AND HIGHER POLYNOMIALS 142
Because the number of papoo steps increases (0,3,7,11,15,...) for each successive diagonal Pade approximant,
the algebra soon becomes daunting. imposing a limit to the usefulness of this particular procedure for the evaluation
of expressions for R}(x). On the other hand, the procedure is extremely useful for calculating numerical values of
the diagonal PadE approximants, for some argument x of interest, because, unlike the algebra, the arithmetic may
be performed efficiently by a computing device. The algorithm below calculates and outputs numerical values of
K. R;, R22, .... R; from input values of to, t,, t2, t2j, ... 12J-
The algorithm makes use of two running indices, k and j, that have simple interpretations in terms of the augmented
PadE table. It may be noticed that the entries in the augmented table define two sets of diagonals. Each diagonal
in one set is composed of R; members such that m + n is a constant. this constant equals 2k. The other set comprises
diagonals that are infinitely long, each diagonal being characterized by a constant difference m - n between the
degrees of the table members. this difference equals the index j. Thus, the identities
be equal (either coincidentally or by loss of arithmetic significance); often these five commands may be omitted.
The diagonal Pad approximants Ro(x), R:(x), R?(x), ... converge toward f(x) much more rapidly than do fo(x),
fi(x), f2(x), ..., where f,(x) is the partial sum defined by 17:13:7. This is illustrated by the test values that accompany
143 THE CUBIC FUNCTION x' + axe + bx + c AND HIGHER POLYNOMIALS 17:14
Table 17.13.1
J f'(x) k
the algorithm. Here. t, = 1/j! and so Rt is an approximant to exp(1), equal to 2.718281828. The output data show
that Rk has converged to nine significant figures already by k = 5. It requires a partial sum of twelve terms, to +
t, + t2 + ... + ill to achieve comparable accuracy. This advantage of the PadE approach is largely illusory,
however, because terms up to r1o were needed to calculate the Rs approximant! In fact, for the summation of
convergent series [Section 0:61 such as this, the Padd procedure will seldom be worth the effort.
The great virtue of the Pade procedure comes from its ability to handle asymptotic series [see Section 0:6],
and the Atlas takes advantage of this in a few of the algorithms in Sections 8 [for instance in Section 51:8]. A
demonstration of the efficacy of this procedure is shown in Table 17. 13. 1, which lists approximations to the zero-
order Basset function [Chapter 51 ] of argument 3:
17:13:13 Ko(3) = 0.03473950439
calculated as partial sums of the asymptotic series 51:6:8 and as the corresponding diagonal Pad@ approximants.
Evidently, whereas the former procedure can never give a precision better than about three significant figures, the
latter has achieved nine significant figures of convergence by k = 7.
Even more remarkable is the effect of the Pad6 procedure on series that are not even asymptotic. Consider the
Gauss function [Chapter 60J F(;,=;1;-2) expanded via 60:6:1:
17:13:14 2,2[(2j
F(11;1;-2)
=o L
(2j- 1)!!12
(2j)!! I
(-2) = 1 -
I 9
2 + 16
225 11025
288 + 9216 -
This series is so wildly divergent that 17:13:14 would not usually be considered a valid representation of F(2,21;1;-2).
Nevertheless, the diagonal Pade approximants converge to 0.745749189, a value that can be identified with
(1/f)F(J,1;1 3) [or with (2/trV3)K( 2/3), where K denotes the complete elliptic integral of Chapter 61], to
which F(1,1;1;-2) is equivalent via transformation 60:5:3.
Table 17.14.1
X. A
X, f
points. In the second situation, error is assumed to contaminate the f(xj) data and a (usually rather simple) rela-
tionship f(x) is sought that does not exactly reproduce f(x) at the x = xo, x,. X2, .... Xj points but is close. Such a
procedure is known as regression. Polynomials are commonly employed for both interpolation and regression.
There exist many interpolative schemes, but one that provides a smooth interpolation without undue complexity
is the sliding cubic, or Lagrange four-point interpolate. In this technique a cubic expression is fitted to the data
at x,- x,, x, and x;.2 but is used to describe the relationship only between the middle two points, x; and x;.,.
The cubic is
17:14:2 f(x) - f, +
X, -
_Xl
/ 6
. f
2 JJ
2
-
3 J
N xl
/
J/-I
2
f+ f
j-
2
+rx=x:1 Lf r f +j J x;szsz.,
h /L6 2 2 6J
which is convenient for computation. When the argument passes from the x, s x s x;., range into the x;., s x s
x;.2 range, the cubic "slides" with it and the interpolation now uses the f,, f,-,, f,,.2, f,, 3 data.
An alternative, and popular, interpolation scheme is provided by the cubic spline. This is also a piecewise
cubic polynomial that is, in general, even "smoother" than the sliding cubic. For this reason it is preferred for
treating data that display discontinuities. For further information on the cubic spline, reference may be made to
Hamming [Section 20.9].
The remainder of this section will deal with the technique of polynomial fitting in which a K-degree polynomial
17:14:3 f(x) = aKx` + aK_ix" ' + - + a,x + au
is fitted to the evenly spaced data (xo,fo),(xi,ji)..... (x,_ ,,f_,), (xj,fj). If K is chosen to equal J. the fit is exact
and the procedure is one of interpolation. If K is chosen to be less than J, the procedure minimizes the sum of
squares
17:14:4 (f - f(xj)12
j-o
and is said to produce a least-squares regression. The integer K is restricted to values not exceeding J since, with
145 THE CUBIC FUNCTION x' + ax' - bx + c AND HIGHER POLYNOMIALS 17:14
K = J, the fit is already exact. Figure 17-3 shows an example of the least-squares cubic function fitted to a set of
nine data pairs.
An algorithm follows that permits the coefficients a5, aK_,. ..., a,. an to be found. There is no limit to
J + 1, the number of data that may be input, The degree K of the fitted polynomial is limited only by the registers
available and by the need for K not to exceed J. There is no lower limit to K: K = I will return the coefficients
a, = b, ao = c of the best straight line fit (see Section 7:14; K = 0 will return the constant that is the average of
all the f, values.
The algorithm is lengthy because it has five phases.
The first phase simply allows the input of J, xo, .r, and K. The x,) and x, parameters are not stored as such but
as (x0 + x,)/2 = u and (x, - xo)/2 = h. The independent variable x is subsequently regarded as having been
replaced by y = (u - x)/h so that all the data lie in the range l ? y a - I and are characterized by y, = I -
(2j/J)
In the second phase x, is output to prompt the operator to load f,. These input values are each multiplied by
values that the algorithm calculates for tt(y) for k = 0. 1. 2...., K. and are accumulated as the sums
in registers bo, b,, b2, .... bK. Here t,(y) denotes the discrete Chebyshev polynomial of degree k and argument y;
its definition and properties are given in Section 22:13. The algorithm uses recursion formula 22:13:8 to calculate
t5(yj).
As explained in Section 22:13, the 'best" Kdegree polynomial through the points (fo,yo). (ff,)y), ..., (f,-i,yj-J
(f,,y,) is given by
K (2k + 1)(J - k + 1),
17:14:6 PK(y) _ d,t,(y) where dA =
(J +
and where (J - k + 1), and (J + 1),., are Pochhammer polynomials [see Chapter 18). The third phase of the
algorithm is devoted to replacing each e, value by the corresponding d,.
The fourth phase effects the replacement of the stored d, values by b, values where
K K
17:14:7 Ib,y4=Zdt,(y)
k_o
A second set of registers ao, a,, a2, ..., aK is employed for this purpose. For j = 0, 1, 2, ..., K the registeris
set equal to c' k", the coefficient of y' in the polynomial expansion of t,(y). These coefficients are zero for j < k
and ifj and k have unlike parities. They are constructed using the recursion formula
k-0 (-
Input J >
Input X ," »»» Set h=xo
Storage needed: As written,
the algorithm requires 2K +
12 registers: for the param-
i
Input xf eters J, h, u, K, k, bo, b b2,
Set u = (xj + h)/2 .. , bK. j, r, s. f r, ao, a,,
Replace h by u - h a2, ..., ax. However, the min-
Input K -, imum number of registers re-
Set k=K quired for K z 3 is 2K + 8
(1) Set bk = 0 because r and ao may share the
Replace k by k - 1 same register, as may s and
If k a 0 go to (1) ---------------------- I
a,, f and a2 and r and a3.
Setj=-1 A
(2) Replace j by j + I
Ifj>Jgoto(4)
Set k = r = 0
Set s = 1
x, (as cue)
Input f >
««« Output [(2j/J) - 1)h + u
ta
a
Set f=f
Replace b, by b, + f
(3) Replace k by k + I
If k > K go to (2)
Sett= r+ (sJ- rJ - 2sj)(2k - l)/[k(J- k + 1))
Replace bk by bk + if
Setr=s
Set s = r
Go to (3) ----------------------------
147 THE CUBIC FUNCTION x3 + ax' + bx + c AND HIGHER POLYNOMIALS 17:14
are calculated via a binomial expansion of the (x - u)k term. The expression used is
u' k
17:14:10 at _ I
k, .bk +
K b1j!
kr .(-h)k (j - k)! \h
The at coefficients are output in the sequence aK, air-,, ..., a,, ao and also remain in the at registers. They may
be used in conjunction with the Section 17:8 algorithm to compute values of f(x).
CHAPTER
18
THE POCHHAMMER POLYNOMIALS (x)n
This family of polynomials has very simple recursion properties [Section 18:5] and plays an important role as
coefficients in the series that define hypergeometric functions, discussed in Section 18:14.
18:1 NOTATION
These polynomial functions were studied in 1730 by Stirling and later by Appell, who used the symbolism (x,n).
The name Pochhammer polynomial recognizes the inventor of the now conventional (x), notation. An alternative
name is shifted factorial function.
18:2 BEHAVIOR
The Pochhammer polynomial W. is defined for all arguments x and for all nonnegative integer values of n [but
see Section 18:12 for a generalization]. In common with other polyl nial functions of degree n, the Pochhammer
polynomial adopts all real values when n is odd. but has a restricted range for even n.
The Pochhammer polynomial (x) has exactly Intl(n - 1)/21 maxima, lnt(n/2) minima and n zeros, these latter
occurring at x = 0, -1, -2, ..., -n + I as illustrated in Figure 18-1. Outside the range -n < x < 1, (x), generally
increases rapidly in magnitude.
18:3 DEFINITIONS
The Pochhammer polynomial is defined as the n-fold product
18:3:1 n=1,2,3,...
-o
supplemented by the definition
18:3:2
of the Pochhammer polynomial of zero degree.
149
18:4 THE POCHHAMMER POLYNOMIALS W. ISO
An alternative definition
18:3:3 n=0.1.2....
relates Pochhammer's polynomial to a factorial function [Chapter 2] and a binomial coefficient [Chapter 61.
The exponential function [Chapter 26] is used to define a generating function [see Section 0:31 for (x),:
18:4:2 (x), = x
18:4:3 (x)2 = x2 + x
18:5 LNTRARELATIONSHIPS
2"(x).,z(x + 1 n=0,2,4,
.r.
18:5:2 (2x). =
1
2"(x),.-na(x + - I n = 1. 3. 5....
2/
Formulas similar to 18:5:2 may be derived for (3x)., (4x). and generally for (mx), where m is an integer.
Equation 18:5:2 may be reformulated as a duplication formula for the degree of the Pochhammer polynomial
x2) I+x
18:5:3 ll
)
and similarly
4"x( 2 .r).(I
18:5:4 (xh.-i = l l ( l
J + 2/. 2~.'(2) \I 2 x)
Likewise, formulas for (x)3., (x),,.1, (x)3.,2, (x),., etc., may be derived.
Simple recursion formulas exist for both the argument
18:5:5 (x + 1). _ (1 + n) W.
X
for the quotient of two Pochhammer polynomials of common argument. Likewise. 18:5:5 may be generalized to
the formulas
(x+m)._(x+n)
18:5:8 m=0.1.2....
W. W.
18:6 THE POCHHAMMER POLYNOMIALS W. 152
and
is a simple consequence of the definition 18:3:1 (or of the recursion formula 18:5:6).
18:6 EXPANSIONS
For further information and an extended table see Abramowitz and Stegun [Chapter 24].
Table 18.6.1
m 0 1 0 0 0 0 0 0 0 0 0 0
M = 1 0 1 -1 2 -6 24 -120 720 -5040 40320 -362880
2 0 0 1 -3 11 -50 274 -1764 13068 -109584 1026576
m=3 0 0 0 1 -6 35 -225 1624 -13132 118124 -1172700
m 4 0 0 0 0 1 -10 85 -735 6769 -67284 723680
M- 5 0 0 0 0 0 1 -15 175 -1960 22449 -269325
M6 0 0 0 0 0 0 I -21 322 -4536 63273
m 7 0 0 0 0 0 0 0 1 -28 546 -9450
153 THE POCHHAMMER POLYNOMIALS W. 18:9
1_n (-m). 1
n=0 1 I 1 I I I I I I I I
Setf= 1
Input n >> Input restriction: The degree it must be a nonnegative
Ifn=0goto(2) integer.
(1) Replace f by fx
Replace x by x + 1
Replace it by n - 1 Test values:
Ifn*0goto(1) (17)0 = I
(2) Output f [ (-0.1)7 = -55.9122291
f - W. <'
18:9 APPROXIMATIONS
When x or it or both are of large magnitude, relationship 18:12:1, together with results from Section 40:9, may be
used to derive asymptotic expressions for Pochhammer polynomials. Thus, for large it and modest x, the asymptotic
approximation
n`-'n! I I + x(x - 1) + x(x - 1)(x - 2)(3x - 1) + ... i
18:9:1 W. _ n -- x
f (x) 2n 24n2 J
18:9:3
follows from 18:7:7, providing a link between what are probably the three most important irrational numbers: a,
f and e.
For large it and x close to -n/2, the Pochhammer polynomial approximates a sine function [see Chapter 321:
n' e2p(`
18:9:4 (x)n - n) sin('nx) I + 2 << 2 -. x
The development of this sinusoidal behavior is evident in Figure 18-1, even for n as small as 4.
18:10 THE POCHHAMMER POLYNOMIALS W. 154
18:10:1
d
dx
1
W. = W. j-0 x + j = (x)"[4(n + x) - t[r(x))
Because of 18:6:1, an expression for (x + iy)" may be derived from equation 17.11:1 on replacement of ai by the
Stirling number (-
1 8:12 GENERALIZATIONS
Pochhammer polynomials can be expressed as the quotient of two gamma functions [Chapter 43)
C(n + x)
18:12:1 W. =
1700
whose arguments differ by the nonnegative integer n. Hence, the quotient fly + x)/f(x), with y unrestricted, is
a generalization of the Pochhammer polynomial. A more restricted generalization, permitting n to be a negative
integer, may be developed from recursion formula 18:5:4.
Factorial functions [Chapter 2), binomial coefficients [Chapter 61, gamma functions [Chapter 43], complete beta
functions [Chapter 43) and Pochhammer polynomials are all closely interrelated.
A factorial polynomial, as defined by Tuma [Section 1.03):
18:13:1 n=1.2.3,...
is a Pochhammer polynomial of changed argument. One has (x)" = X'";(x) and X'(x) = (-h)"(-x/h)". The symbol
x<"' is also used for the factorial polynomial defined as
18:13:2 xl"1=x(x-
Yet another confusing symbolism, due to Kramp, is
18:13:3
None of these notations is employed in this Atlas.
Table 18.14.1
where x is the argument. a,, a, a3, ..., ax are prescribed numeratorial parameters and c c2, c3, .... cL are
prescribed denominatorial parameters. If one of the numeratorial parameters is a negative integer, the 18:14:1
series may terminate, but generally hypergeometric functions are represented by infinite series.
If the number of denominatorial parameters exceeds the number of numeratorial parameters, that is. if L > K,
the hypergeometric series necessarily converges for all finite values of x. If L = K, convergence is generally limited
to the argument range (xl < 1. If L < K the series diverges (unless it terminates) for all nonzero arguments, but
it may nevertheless usefully represent a function asymptotically [see equation 37:6:5 for an example].
The name "hypergeometric" arises because the 18:14:1 series can be regarded as an extension of the geometric
series [see equation 1:6:5J to which it reduces when K = L = 0. Other small values of K and L give rise to certain
well-known families of functions, discussed elsewhere in this Atlas, as well as a number of generic classes of
functions. Table 18. 14.1 summarizes these types of functions. Notice that the so-called generalized hypergeomerric
function is not, in fact, fully general because one of its denominatorial parameters is constrained to be unity.
Hypergeometric functions are important because most of the so-called special functions of mathematical physics
(i.e., functions that arise as solutions to differential equations of practical importance) are instances of this class.
The cylinder functions of Chapters 49-56 provide examples, all being K at 0, L = 2 hypergeometric functions.
Cases with L = 4 and K = 0, 2 or 4 occur in the theory of elasticity and are encountered in Sections 47:11, 53:6
and 55:6 of this Atlas.
Functions that are expressible as hypergeometric series may be evaluated by exploiting convenient features of
expansion 18:14:1. If the abbreviation
(a, + j)(a2 +j)(a3 + j) ... (aK + j)
18:14:2 G=
(Cl + j)(cr + j)(cs + j) ... (cr. + j)
is adopted, then a hypergeometric function of argument X may be represented as
18:14:3
where R, is the remainder when the hypergeometric series 18:14:1 is truncated after the j = J - I term. If J is
chosen to be larger in magnitude than any of the a or c parameters, and also to exceed IXIIIIL-10. then the dominant
term in G, is JK-L. Unless L < K, the R, remainder will become steadily smaller in magnitude as J increases. For
large enough J, the R, term may be well approximated by a geometric series
18:14:4 +G,X+GIG, IX2+G,G,.,G,.2X'+
Go
= GoG,GI ... fJ,-,X'[1 + jK-LX + j2,1-21X2 + JM-XX) + ...] .
I - JK-LX
18:14 THE POCHHAMMER POLYNOMIALS (x), 156
With approximation 18:14:4 incorporated when appropriate, expression 18:14:3 forms the basis of a universal
hypergeometric algorithm that may be used to calculate numerical values of a wide variety of functions. The
algorithm, which is detailed below, requires the following inputs:
Go to (4)
(3) Replace t by t/(b; + J) Test values:
(4) Replace i by i - I input: K = 1, L = 1;
Ifi>Kgoto(3) parameters: 1, Z; X = 4
Ifi=0goto(5) output: 1.57229437, 1.57081694,
Replace t by t(b; + J) 1.57079671,1.57079633,
Go to (4) 1.57079633(=Tr/2), ...
(5) Replace t by tX
Replace f by f + t input: K = 0, L = 2;
Replace jbyj+l parameters: 21, 1; X = 1
If frac(j/5) * 0 go to (7) output: 3.76219871, 3.76219569.
If I >- j L-K/X go to (6) 3.76219569(=cosh(2)), ...
Set i = r/(1 - (jL_x/X)1
(6) Output f - i input: K= 1,L=0;
Approximate parameter: 1; X = 0.04
f(x) value
(7) Seti=K+L+I output: 1.021340580, 1.021340743
1.021340745,1.021340744
Go to (4)
1.021340744(=10 daw(5)), ...
Because the performance of the algorithm depends on many factors, it is not possible generally to predict the
value of J required to achieve a sought precision in f(X). Accordingly. the universal hypergeometric algorithm sets
J = 5. 10. 15, ... and generates an output for each of these J values. leaving it to the operator to observe when
convergence has been achieved, or to judge when the output value has sufficient accuracy for his purpose. In some
cases, convergence may occur but be tediously slow. Note that when L < K, the algorithm may fail to produce
an acceptable value unless X is sufficiently small; in this case the output values may initially tend toward the correct
answer but later diverge.
Table 18.14.2 lists some of the functions that may be evaluated using the universal hypergeometric algorithm.
For many of these functions alternative algorithms will be found in the appropriate chapter of this Atlas. Such
"custom" algorithms are generally more satisfactory than the universal algorithm, but the latter has the supreme
advantage of versatility.
1S7 THE POCHHAMMER POLYNOMIALS W. 18:14
Table 18.14.2
No. of Chapter
Puamnen or
Parameter values secti on
K L b.. b,. b,.... b,,.,. X sought run"" ftxi Restrictions
3
1: 1:7
2 2
1 I < 1:7
2 1 1 3 3
2 I G 1:7
22 2 2
I. . . .. I : 2... .2 -1 'g.) n - 1. 2. 3.... 3
I 3 3
n
1
-1 fi(n) n - I. 2. 3.... 3
2 22 2
-1<x<1
-
r.I %x (I m x)° 6:14
-bx r
0 0 kl < I.-Al 7
c b4+c
I I n; I -r
bx
(bx + cP/b'.1 141 > k/bl II
1; 1 -r
bx
V bx + r/V lu :ri > Ic/bl 12
z
Nc
1>k/M 12
21 '1 bl Vb4+
I_x -1 <x<3
2 2 +I: 1. 1
21
2
2 2 -4. -n; -2n, I n!P.(x)/12n - wu - w x < -1 or x > 3 21
1-x
n 1 +n I
2 2 1 (-[)*"n!!P,Ix)/(n - U!! IW l < 1; n = 0. 2. 4.... 21
2. 2 :Z.
1_n 2+1
n-1 `!!p tx)/
2 2
2 2
: I.
3
2
r !!
< I: n = I. 3. 5.... 21
I I
2 2 .!P.4.)/(,- - I i!!.1 (x1 > 1 21
2 2 2 1
2 2 -n, 8;2 1 Z
1 x
T,(x) -1 < r< 3 22
I I I_1 / 2
2 2
2 x+1 1+.x
2 2 -n.n + 2: 1.i 1x 2
+I -1<x<3 22
18:14 THE POCHHAMMER POLYNOMIALS W. 158
No. of Qnpler
parxmetrn ttf
Parameter value .ection
K L b,. b:, b,, b,., X Sough function fW Restrictions no.
1. 3 F-1 2
2 2 x-1 I+x 22
2 2 -n - v. - n. 1, l+v 1 22:12
2 2 -n.n+211:
2
Z I.-1- +2A-1
22:12
2 -n: 1. 1 x L.(x) 23
2 1 n: 1. I -x exp(-x)L.(x) 23
L;'fr1/(n p)
2 -n: I. I+ p x 23:12
n
2
n
-. I
I
2
r 1Glrl/f-2146fn - w! n=0.2.4.... 24
1-n 3
un!.
2 N. 1.3.5..-. 24
2
a -n -2
22
I
1 Hedx)/f2x)' 24:13
1; 2 L-2 Itxx) 25
x r-1
Ils) - -f - InOlnfxN)
1.2.2 Infx) 0<x+1 25.37
In(s)
In(x)Iilx)
0 positive x, large or small 25. 37
Infx)
x_1
1 1 I;I v -(- In,(rl
xs-1 25:12
thin(s)
2 2 1, 1; 2. 2 1-x 0<s 2 25:13
s-1
trim(s)
3 3 1. 1, 1.2.2.2 1-x :-1 0<s 2 25:13
0 1 I x expel 26
0 -a nIe.(x)/t 26:13
Y
0 2 -. 1 corMF1 28
2 A
No. 0( Chapter
pvatnetm oe
Paranveter values section
K L b,. b, b,. .-. b,_, X Saug1 function f(x) Rcstncsioos
rsm6lx)
2
2
I
22
! I.-2 ; -.r -I<xsI 31
2 2 1. 2 ;
3
2. 2 - 4r[arsinhtxl - IM2x11 x>1 31
3 1
I 1 3 -1
x atcsclKx) I!cl > I 31
2 2 2
ananbul
1
2 2
3
r -I<<1 31
1 I
I I -- - x arcothlx) A > 1 31
2 . x
3 -t vial.,)
-
I
0 2 I. 32.33
2 4 x ocsc(x)
I -r 1
0 2 -, 1 tos(s) - - 32. 33
2 4 sects)
3 _ .'
0 2 I. - sinclxl 32:13
2 4
I Y
0 2 1. +n Vi/2 (2n - x > 0: n = 0. I. 2.... 32:13
2 4
1 -r
0 2 I. U!! x > 0: n - 0. 1. 2.... 32;13
2 4
12
1 0<x<4 3314
1
I -
-r wcx) -15xu1 35
2 2
mint.,)
2 2
22
I
1.2
3
.r -I <x<I 35
1 3 1
2 2 I. x atccsctxl >I 35
2. 2: 2
I 12-
2 2
- x arccot(xl ld a l 35
1 1 3 2x_I
2 2 -. I. - atcsec(x) I 35
22 2 Zr 2x - 2
Einl:)
1 2 1: 2. 2 -x 37
33 -x° SO)
3 38
i
2 2 2 4 x
33 Shifxl
1 3
2 2 2 4 x
18:14 THE POCHHAMMER POLYNOMIALS W. 160
So of or
p4rameters
Parameter values .sawn
K L b1. bj, b), ..., bx-i x Sought function Rx) Reunctions no.
3 1;2.2,2
3
4
-Al
4 Cin(x)/r 38
3 r
3 I: 2. 2 4 Chio(x)/r 38
2 4
-4
0 .1 x fi(x) large x 38:13
2 r
3 -4
2 0 1. 2 r ghx) large x 38:13
3. 3 7 -x
I 3 - 1. - - .
Vrt/'_ Ax)/x? 39
4 2 4 4
1 5
-, 1.-
I
I 3 \`n/2 Clxl/x 39
42 4 4
I 3 -:1.-,-
1+V
2
3 3-v
2 2 4
1+v
IslO:v) - S(svll v<1.x>0 39:12
I 3
V
22
-.1.-
1 +V
2
S
4
IC(O:Y) - Ctrv)) v<I.x>0 39:12
1-v 2-v -4
v < 1. large x
2 2 r sirt)Cfx:v)] 39:12
v 3-v -4
2 0 - (aualxlSlx:v) + cc41x)C(+o)) Y < 1. large x 39:12
2 2 1-v
0 2
3 -5 _i 4
ring + 4) - \r Cmet(x)J /
1
2. 39:13
44 4
0 2 57
4 4
-4
3v'. \2 Fres(x) - crnlr+
// n-)
/4x' 39:13
3 -4
- V x Fralx)
1
2 0 large x 39:13
4 4 i
3 S -4
2 0 \'8a Si Ores(x) large x 39:13
44 7
I 2
I
2 1.2
3
-r V'; aft,)/2x 40
3
0 1 - \a exp0r`wrffx)/2x 40
2
1 -I
\e x exp(rxrktix) large x 40
2
2 4:2.1
I I
-r V i I: + terfelx)] 40:13
_ I -r \'4 rI I 11
11
2 -3 iterfdx)J 40:13
2'2 x `4 + 2
1 2 3 ! :
2
. I r 6V a
x
4 +
x
3
+ i'ce fdx)
1
4 0:1 3
3 -1
2\a r exp(.r)berfc(x) large x 40:13
r
I
No. of
parameters
Parameter values
K L b,, b. b . , be., x Suu1N function f(x1 Restnctioru
3 d..(.)
0 -r 42
I 0 2x daw(x) large x 42
2
1 3 explx)
1 2 -; I. - x _ daw(V x) 42
2 2 V.v
2 2 I- v. x: I. 1+ x I r B(x.yl v> 0 43:13
2 1.v:1+v.2 I
v-I Iirtv)+7) 44
2 2 I- x. I +s 1 x'b'(sl 44:12
I I v, v+ l -I 4G(.)/2 44.13
0 1 1+ r x .e 'exp(xMv;xl 45
I 2 r. I. 1 v -x rx-'7(v;x) 43
I 0 1- v -I x-'expxlrfv:x) large x 45
1 2 v: I. I +u -x vx-'7(vsl large v 45
1 2 2vI; : , I
r= r(
-d)exp( r
1
ID,(-x) + D,(x)1/2V ` 2'a v * I. 3, s. 46
4
_ I.- r
'
I v 3
1D.I-x) - ',r v*0.2.4.... 46
4
_
2 2 2
I
2
v I
:2. 1
2 r(.
2
1
v
exp( D,Ix)I/2V2`z * 1. 3.5.... 46
2 1 2v, 12 v: I large x 46
a; c. A Mta:c:x) 47
1 2 I
I 2 c - a; c, l -x exp(-x)M(ar;x) 47
_I large x 48
r"Uta:r:xl
x
2 +I+2µ.I 46:13
I, I
r Ivfx) 49
4
0 2 I, 2 I,Lv/ 49
4
18:14 THE POCHHAMMER POLYNOMIALS (x), 162
No. of
Parameters
Parameter values
K L b1. bl. b,, ..., br.t X Sough function f(xl Resu.ctions
_I 3-: I
2 1 I V'2trx exp(-x')1i(x) latgt x 49
2 '_ 2x
2 1 -2 -
1
v 2
- + v, I -
2s
V2wxexp(-x)1,(x) Urge 30
3 x7
0 2 1.-+a
2 4
(2R+I)!!x-"itxl m1.0.1.2.... 50:4
1
1
- - v. - - v: 1
2
1
2
-
-1 +
txplx)ILIx) large x 51
0 2 I. I + n
-r n!(2/x)"1,(x) n - 0, 1. 2, ... 52
4
-x'
0 2 1. 1+ r(1 + vK2/x)9,Ix) v -1. -2, -3.... 53
3
-4R
-x (2n + I )!!x""J,(x) R - -1. 0. 1. 2....
0 2 1. 53:4
I 3 1 -1 3 v
4 2 a I - a. - - a; -. I Ptn:x1 with large x 53:6
2 -
4 -2
4 2
22
3 3
a.a+-.--a.2-a:1.- 1
-
-1 2xQ(v:x)
wither= ---
4
3 v
2
I
va: -. larger 53:6
I2, I -x'
0 4 =. 1. 1 bes(xl 55
236
I. I. -.
33 -i 4
0 4
2 256 r bei(xl 55
-
2 16
1+v
t+-
-x
0 4 r(I + vl( Fe,(xl 55:6
2 2 2 256 x
163 THE POCHHAMMER POLYNOMIALS (x). 18:14
No. o( Chapter
parameters /lr
Parameter values section
K L bi, b;. b,..., bx.t X Sough (union r(s) Restrictions tto
v3-v
12
J
0 4 r12 + v)t.-) OeJsl 55:6
2 2 2 x
3"'r(*) Btu)
0 2 1
9
(airs) - - 1 - + Ailx) 56
3 2 LLL V3
0 2 1. I - Aifx) 36
3 9 2r L V3
0 _
45
-33- -9 ?hai(s> 2
- = - (Hi(s) - faifx) - gai(s1I
s' e
56:13
I 5. -3
2 I I 2 ><Vi expl 3 x':2 IAi(xl large positive x 56
6 6 4s". 2 11
1 5 3
: I VrV r expl Bilsl large positive x 56
1
66 4x1 l
Q
1 5 7 11 I I -9 2 a1
4 2 IJAI(-xl + cBif -x11.: large positive x 56
12 12' 12 12 2 4x' a
4 -
7 II
12' 12 12 12
13 17
I. -
3
2
-
-9
4s'
*Vx (sBN-s) - cAif-x11. c - cm large positive x 56
1 1 5 -9
3 1 6.2.6 I large positive s 56
4x'
0 2
1 3 -r 57
Z 2 4
0 2
3 3 -r
- 57
2 2 4
JS -.r 57
0
2 2 4
0 2
33
-2 2 v
-
-xt
4
-r(-+,11-)
2 x
,.
3
2
>0 57
13 4
2 0 -- - (Y .4x1 - A.,(x)l Iarge x 57
I I -4
2 0 -- (hi(s) - Y{.rll large x 57
2 2 5-
-1 1 -4 A
0 - (hdx) - ' 4x11 large x 57
2,2
2 0 I- I -4
xl
Y+r r(2 * (1i.1x1 - YJx)I large x $7
33 x' Va 3 /
r(2+oil-I
11
2'
2 0
2
- v' 2 r V; r(I2 * o)(x) ILJ.r) - 13x)( larger 57:13
2 2
I-x P.lx) -I<s<3 59
1* 3 I 11(5- e)
2 2 -.1+-:
2
I. -2 +r 2
_
VvrU + v)
x>1 59
18:14 THE POCHHAMMER POLYNOMIALS (x) 164
No. of Chapter
ot
Parameters
Parameter values section
K L b,, b, b,..... b- X Sought function fix) Restrictions no.
2 2 -r -: -. 1 I +Y 1
X lefts) -1 <1 59:6
1_V. v; 3
- kW.)
1
1.I
2 2 .r lef,"Y:1 - I <x < 1 39:13
2 2 a. b: . I Z
I+b)/[rl-)r-)J
vsr(2
'llllll
a
b a + b s -1. -3, -5....
a W7
2 0. 1
I
rlr)/ 2' r( a*r )r(l+r-a ) .2.
0.1111 607
d+ I a....., 1. q cr 60:13
1,
2 2 1. 1 Kw -1 61
22
2
2 2 -1 1:1.1 r' -E1x1 -1 cx< 1 61
2 2 n
1 x' 2VI
2 2 K(x) <x<1 61
22 7-1 2
3
2 2 1.1 Elx) -1 <x<1 61
22 aft - x-1
3 2
3 3
1 I 1
3 3 - - -: 1, 1. -
1
2'2'2 2
1 3
x' -
nx
Ewdr 0sxs1 61:10
Operations of the calculus are easily applied to functions that are expressible as hypergeometric functions. For
-
example, values of the derivative of any function f(x) that appears in Table 18.14.2 may be calculated by taking
advantage of the rule
of the indefinite integral of R.O. The situation is more complicated than for differentiation because the form adopted
by the hypergeometric series for the integral depends on the relationship between .x, the argument of the function.
and X, the argument of the algorithm. The most commonly encountered relationships are X = ±x, X = bx" (b is
a constant such as I or -2; is is a positive integer greater than unity), X = b/.t^. X = b/x and X = b(I - x). These
five possibilities are all addressed in Table 18.14.3. which itemizes the changes needed to enable the universal
hypergeometric algorithm to generate values of an indefinite integral of fix). For example, using the Jj(x) entry
from Table 18.14.2 and the fourth row of Table 18.14.3, one sees that
I
18:14:6 J' t)dt
Xo
Other operations of the calculus may also be applied to functions expressible as hypergeometric functions.
Table 18.14.3 reveals the effects of semidifferentiation and semiintegration; see Oldham and Spanier for other cases
of generalized differintegration. Moreover, the Laplace transform [Section 26:141 of the hypergeometric function
18:14:1 is yet another hypergeometric function
IS:14:7
S)
The table suggests that operations of the calculus invariably increase the number of parameters. This is not
necessarily so because, for example, a "new" numeratorial parameter may equal an "old" denominatorial parameter,
permitting the two to be "cancelled."
Table 18.14.3
No. of
paruneurs Parameter values x Algonthm evaluates
I
K+I L+I , u u;...., ur: AMU
e n A
n_I 2n-1 b (n - IU''
K +I L+1 , a, I. a; + I. aK + r, r, + I, r.+I.-... rt + 1. [An IIds
n n S bG " I
1
MI - x)
-,
-IJ fields
_ d"
K+1 t. *I I. a,. a2. .... aK; c,. r_..... s,- bx 71x)
1 n d-2
X I L+1 1, u a....., a,, c'. n.... . re. 2- x ds-,:.
f7x1
2
CHAPTER
19
THE BERNOULLI POLYNOMIALS
Included among the useful properties of the Bernoulli polynomials is their utility for expressing sums of powers
of the integers [see Section 19:141.
19:1 NOTATION
B,(x) is the usual symbol for a Bernoulli polynomial of degree n and argument x, but 13,(x) is sometimes used by
authors who adopt the "rival notation" [see Section 4:11 for Bernoulli numbers. The name "Bernoulli polynomial"
and the symbol 4A,(x) have been used for the quantity that we represent by B,(x) - B., where B. is the n"Bernoulli
number [Chapter 4].
19:2 BEHAVIOR
Bernoulli polynomials are defined for all nonnegative degrees it and all real values of the argument x. However,
0 s x s 1 is the most important range of the argument for B,(x).
For even it, B,(x) never exceeds the Bernoulli number B, in magnitude in the range 0 x s 1:
19:2:1 IB,(x)1s1B,1 0<xsl n=0.2.4....
and equals B. at both ends of this range. Apart from Bo(.x), all Bernoulli polynomials of even degree display a
local extremum [see Section 0:7] at x = z, as exemplified in Figure 19.1. Bernoulli polynomials of odd degree
(except B1(x)) vanish at x = 0. x = I and x = 1.
The magnitude of B,(x) for 2 is n <_ 11 and 0 s x s 1 is small, never exceeding }. Figure 19-2 shows the
typical behavior of Bernoulli polynomials outside these ranges.
19:3 DEFINITIONS
The Bateman manuscript [Erd6lyi, Magnus, Oberhettinger and Tricomi, Higher Transcendental Functions, Volume
1, pages 38 and 391 gives several integral representations of the Bernoulli polynomials, but their usual definition
is via the generating function
t exp(xt) t"
19:3:1 T, B,(x) -
exp(t) - I =u n!
167
19:3 THE BERNOULLI POLYNOMIALS 168
:....:..... 0
:.............................:....:....:.-(L 10
...........I... 5
B7 (x)
..................-10
Another definition
19:3:2 i ( n)B,x"-' n = 0. 1. 2, ..
J-0 )
relates the polynomial coefficients to binomial coefficients Chapter 61 and Bernoulli numbers [Chapter 41.
169 THE BERNOULLI POLYNOMIALS B^(x) 19:5
19:4:3 B2(x)=x2-x+6
3x2 x
19:4:4 B3(x)=X3- +
2 2
1
19:4:5 B4(X)=x4 -2x3+x2-
30
5x' 5x3 x
19:4:6 Bs(x) = x3 -
2 +3 6
5x4 x2 1
19:4:7 B6(x)=x6-3x3+ +
----
2 2 42
and
7x6 7x3 x
19:4:8 B7(x) = x7 -
2
+
2
- 7x3
6
+
6
Coefficients of all terms in B^(x) for n 5 15 are listed in Abramowitz and Stegun (Table 23. 1J.
19:5 INTRARELATIONSHIPS
The Bernoulli polynomials have even or `odd symmetry about x = 1 /2
and
(m )
i
and
r
19:5:6 B^(2x) = 2^-'I B*(x + 2),
19:6 THE BERNOULLI POLYNOMIALS B.(x) 170
19:6 EXPANSIONS
in (x - 1):
/ 1 n(n - 1) r 1
7 (n 1l
B"(x) = 1 x -
19:6:2
2 24 \x 2/" + 240 4)(x 2J" -
r 1 n(n - 1)
l( Il I 1
In these series many terms vanish because the Bernoulli number B. [see Chapter 4] is zero when n = 3, 5, 7, .. .
Expansion 19:6:2 can be written concisely in terms of the eta numbers (Chapter 31 if q(0) is interpreted as 1:
I
A Fourier expansion [see Section 36:6] of a Bernoulli polynomial leads to the general formula
or to
Table 19.7.1
n-8
"E-
0 U
t Y' 1
B,
F-
"B.
B. - a. B. B. B. -8. n ' B.
(Y -')B. .
The algorithm uses expansion 19:6:5 with the infinite summation limit replaced by an empirical function of n that
ensures an absolute algorithmic accuracy of 6 x 10-B for n ? 5. The relative error may exceed 6 x 10-1, especially
near the zeros of the Bernoulli polynomials. The colored portions of the algorithm make use of the recursion 19:5:5,
or the equivalent
19:8:1 B"(x - 1) = B,(x) - n(x - 1)"-'
to bring an arbitrary argument into the range of applicability of the expansion.
19:9 APPROXIMATIONS
Approximate values for many Bernoulli polynomials in the range -0.1 5 Bn(x) 5 0.1 can be read directly from
Figure 19-1. For large n and arguments between zero and unity, the leading term of expansion 19:6:5 provides the
approximation
19:10 THE BERNOULLI POLYNOMIALS B,(x) 172
-2n! na
19:9:1 B,() = - cos 21Tx - - n -° 0:5X5 1
(2,rr)" 2
For x values close to 0, i or 1, approximation formulas are readily derived from expansions 19:6:1-19:6:4.
For x values of magnitude large in comparison with n, the approximation
d"
19:10:1 B"() = n'
dx"
d
19:10:2 B,(x) = nB"_,(x)
dx
B,_1(x) - B,_1
19:10:3 B"(t)dt =
ro n+l
19:10:4 nx*0
B"(t)dr = fo1
n=0
19:10:5 I '
0
n= 1,2,3,...
m!W ! B.."
19:10:6 f B"(t)B.(t)dt = m,n= 1,2,3,...
Ju (m + n)!
19:12 GENERALIZATIONS
A class of functions defined by the generating function
t' exp(xt)
19:12:1 2,...
is known as Bernoulli pohnomials of order m [see Korn and Kom, page 824]. They represent a generalization of
the Bernoulli polynomials discussed in this chapter, which are of first order. A still more general set of higher
order Bernoulli polynomials is discussed by Erdelyi, Magnus, Oberhettinger and Tricomi (Higher Transcendental
Functions, Volume 1. pages 39 and 40].
173 THE BERNOULLI POLYNOMIALS B.(x) 19:14
or a moiety:
1) 2"B"+,(m) (2" 1)B".,
19:14:3 1" + 3" t 5" + - + (Zm - 1)" =
n + 1
n=0.1,2,...
Further simplification occurs if is = 2. 4. 6. ... because B"_ r is then zero. Some examples of these finite summations
are presented in Section 1:14. See Section 20:14 for cases of alternating signs.
CHAPTER
20
THE EULER POLYNOMIALS En(x)
Certain series involving the natural numbers can be conveniently expressed using Euler polynomials, as explained
in Section 20:14.
20:1 NOTATION
Euler polynomials of degree n and argument x are generally denoted E (x), although the symbol E,(x) is occasionally
encountered. The E (x) symbolism is also used to denote Schldmilch functions, which are quite unrelated to Euler
polynomials. The Schldmilch functions are briefly discussed in Section 37:14, but they are not used elsewhere in
this Atlas.
20:2 BEHAVIOR
The Euler polynomials are defined for all nonnegative integer n and for all real values of x. Figure 20-1
shows the behavior of the first few Euler polynomials in the range 0 < x t 1, which is the most important range
for these functions. Over this range the inequality
20:2:1 0 EE , (x )
n 2 0 5 x s 1 n= 0, 2, 4, ...
is valid for even n, E. being the nth Euler number [Chapter 51.
As is evident from the diagram, all Euler polynomials of even degree (except Eo(.r)) are zero at x = 0 and
x = I and exhibit a local extremum [see Section 0:7) at x = 1. In complementary fashion, all Euler polynomials
of odd degree are zero at x = 1 and (except for E,(x)) display a local maximum or minimum at x = 0 and x = I.
Figure 20-2 demonstrates the behavior of typical Euler polynomials outside the 0 rt x -S I range.
20:3 DEFINITIONS
Euler polynomials are defined by the generating function [Section 0:31
175
p,
44 44 44 44 4r
0144 a44O*44 O44O44O*44O44a 44
00 44 44
. . . . .
. . . . . '
1. 1
. . . . .
. . . . .
. . . . .
1.0
j .............
0. 3
177 THE EULER POLYNOMIALS E.(x) 20:4
E.(.x)=2 (n)(2x-1)"-'E,
2"''
E.(x)n+1
_-
(x
B.,1 ( 'I) - B.,,
2
(-x)
n = 1. 2. 3, ...
and
20:3:4
n+1
1B,-,(x) - 2`1313.,, (2)J n = 1, 2. 3, ...
relating Euler polynomials to Bernoulli polynomials (Chapter 19] also serve as definitions.
20:4:2 E1(x)=x-2
20:4:3 E2(x) = xz - x
3x2 1
20:4:4 E3(x)-x3- +4
2
20:4:7 E6(x)=x6-3x5+5x3-3x
20:5 THE EULER POLYNOMIALS E"(x) no
and
Coefficients of all terms in E"(x) for n 5 15 are listed by Abramowitz and Stegun [Table 23.11.
20:5 INTRARELATIONSHIPS
The Euler polynomials possess even or odd symmetry about x = Z:
20:5:1
20:5:5 E"(I2+xl
1
-x" !) n E,
(2x)'
Because all Euler numbers of odd index are zero, about half of the terms in the last summation [which is equivalent
to 20:3:21 vanish.
The argument-multiplication formula takes different forms:
20:5:6 E"(mx)=Z(-1)'B".,Ix+-
n 1 ;_(1 `
l
m//
m=2,4,6,...
, (
20:5:7 l m= 1,3,5,...
m
according to whether the multiplier is even or odd. Definition 20:3:3 is a special case of formula 20:5:6.
20:6 EXPANSIONS
Definition 20:3:2 leads immediately to the polynomial
20:6:1
(
E"(x) = 1 x -
1 n(n - 1)
8 x - 2!" 1 5n(n - 1)(n - 2)(n - 3)
384 (x - 2
+
n(n - 1) E_2
2 2"-' \
lx -- I
2
+-lx--l -
nE"_,
2"-' \
I
2/
+
E.
2"
20:6:2
E"(x)=x._2x. ,+4(3)x" ,_2(5)x" `+ 171x" z+
8
179 THE EULER POLYNOMIALS E,.(x) 20:8
or x - I :
reduces to 2
20:6:5 E"O=
x
'=o (2j + t)
when n is odd.
Table 20.7.1
"=0
r--' - 2 -I 2-r-I 2^ - 2 r-'-2
a." -2 B..I 2
".I 2' - "+1
0 E.
For arguments in the range 0 s x s 1, the portions of the algorithm shown in green may be omitted. The
algorithm is based on the Fourier expansion 20:6:4. truncated at j = 1 + Int(30/n). This choice ensures that the
absolute error in E"(x) does not exceed 6 x 10' for n > 5, although the relative error may be greater than this,
particularly near a zerb of E"(x). When the green portions of the algorithm are included, arguments of any magnitude
can be accommodated. Recursion 20:5:4, or its equivalent:
20:8:1 E"(x - 1) = 2(x - 1)" - E"(x)
is employed to shift the argument into the range covered by the Fourier expansion.
20:9 APPROXIMATIONS
For small n and arguments between zero and unity, Figure 20-1 can provide approximate values of E"(x). For large
n and arguments between zero and unity, the leading term of expansion 20:6:4 provides the approximation
4n.r nn
20:9:1 E"(x) = sin i r x - n-. x 05xs1
2
Approximation formulas may be derived from 20:6:2, 20:6:1 or 20:6:3 for x values close to 0. i or 1, respec-
tively. For x values of large magnitude, the approximation
20:9:2
1
E"(x) _ I x - Z I
(
8-bit precision Ix - 12
> 4(n - 1)
In addition to results that follow from the general formulas of Section 17:10, there are several special results from
applying the differentiation and integration operators to the Euler polynomial. These include
d"
20:10:1 E"(x) n!
dx"
d
20:10:2 E"(x) = nE"-1(x)
dx
r E"-3(x) n1,3,5, xa =0
20:10:3
n+1 n+1
in 0,2,4,... x0 =-
2
181 THE EULER POLYNOMIALS E"(x) 20:14
and
20.10:4
f
J0
, E"(t)E",(t)dt = 1)
4m!n!
(m+n+2)!B° .*_
20:12 GENERALIZATIONS
Euler polynomials may be generalized in the manner described in Erdelyi, Magnus, Oberhettinger and Tricomi
[Higher Transcendental Functions, Volume 1, page 43J.
E,(.r) ± E"(m + x)
20:14:1 =
where the upper/lower signs correspond to odd/even m. Setting x = I in this relationship yields the useful result
20:14:2 1
The first of these results is presented in Chapter 1 as equation 1:14:9. and equations 1:14:6-1:14:8 are special
cases.
CHAPTER
21
THE LEGENDRE POLYNOMIALS Pn(x)
The Legendre polynomials arise in several branches of applied mathematics (as, for example, in describing elec-
trostatic or other fields), especially in problems involving spherical symmetry. The Legendre polynomials are among
the simplest of the families of orthogonal polynomials [see Section 21:14].
21:1 NOTATION
The symbol P,(x) is standard for the Legendre polynomial of degree n and argument x, although the P is often
italicized. The name spherical polynomial is also encountered. Sometimes normalized Legendre polynomials are
specified: the polynomial of this chapter is normalized through multiplication by (2n + 1)/2 [see Section
21:141. Yet another name is zonal surface harmonic [see Section 59:141.
The symbol P,(x), where v is not necessarily an integer, denotes a member of the function family known as
Legendre functions of the first kind, which are a subject of Chapter 59. When v takes any integer value, positive,
negative or zero, P,(x) reduces to a Legendre polynomial
21:1:1 Po(x) = P,(x) v = n = 0, 1. 2...
21:1:2 P,(x) = v = -n = -1, -2, -3, ...
The polynomials of the present chapter are thus special instances of Legendre functions of the first kind and some-
times are so identified. The reader is referred to Sections 21:12 and 21:13 for explanations of the related notations
P:(x), P,'(x). P.*,"(x) and P"(x).
For -1 s x s I the argument x is often replaced by the cosine of a subsidiary variable 0:
21:1:3 0 = arccos(x)
as a reflection of the fact that Legendre polynomials often arise naturally in scientific and engineering applications
with arguments that are cosines of salient angles.
21:2 BEHAVIOR
In this chapter we restrict the degree n to be a nonnegative integer; n = 0, 1, 2..... Although the argument x of
the Legendre polynomial P (x) may take any real value, the most important range is -1 s x 5 1; this is the only
range accessible when x is replaced by cos(9). Within this range the polynomials never exceed unity in magnitude
183
21:2 THE LEGENDRE POLYNOMIALS P,(x) 184
21:2:1 sl -1 5 x 5 1
as illustrated in Figure 21-1. The Legendre polynomial has exactly n real zeros, all of which lie in the interval
-1<x<I.
Outside the - I s x s I range, the Legendre polynomial P (x) increases in magnitude without limit (except
,o ,o
rAj
o
'L
ti
o titi
o
,o oo o h o-0 o1 o0 o
4 4 4 4 4G 4 4 4 4 49 4y 46 4
' _% O
44..4
ti
1.2
185 THE LEGENDRE POLYNOMIALS P"(x) 21:3
for Po(x)) as x - --. This behavior is exemplified in Figure 21-2 and conforms to the rules
21:2:2 1 =Po(x)<PA)<P2(x)< x>1
21:2:3 1=Po(x)<-P1(x)<P2(x)<-PA(x)< x<-1
21:3 DEFINITIONS
Legendre polynomials may be defined by means of the generating function [see Section 0:31
1 P"(x)t" -1<t<1 -I sx - I
=o
21:3:1
l-2xt-t=
"=o
21:3:3
-1)..x"_J
rr
j=0.2.4,...,n--±-
j!!(n - j)! 2 2
shows that Legendre polynomials contain (n + 2)/2 terms if n is even and (n + 1)/2 terms if n is odd. An equivalent
21:4 THE LEGENDRE POLYNOMIALS P,(x) 186
R
sin Rn + 2)t dt
V2
P,(cos(8)) =
21:3:5
it I
a cos(8) - cos(t)
and
The most general solution of this equation is c,P,(x) + c2Q,(x). where c, and c2 are arbitrary constants and Q,(x)
is the function discussed in Section 21:13.
Figure 21-3 illustrates a geometric definition that explains how Legendre polynomials arise in certain appli-
cations. The triangle shown has two of its sides, one of unity length and one of length r, enclosing the angle e.
By the cosine law [see Section 34:14], the length z of the third side equals 1 - 2r cos(9) + r-. In some physical
problems it is necessary to expand z-' as a power series in r-1. Formula 21:3:1 shows this series to be
1 Po(cos(8)) P,(cos(8)) P2(cos(6))
21:3:8 + + +
z r r= r'
if r > 1, or a similar series if r < 1.
3x' - I 1 * 3 cos(29)
21:4:3 P2(x) = 2 = 4
187 THE LEGENDRE POLYNOMIALS P.(x) 21:5
21:5 INTRARELATIONSIIPS
P.(x)=
2n-1 n-1 n=2.3,4,...
21:5:2
is
XP._1(x)--P._2(x)
n
X4 l 8
21:5:5 = Po(x) + 4 P2(x) + P.(-V)
5 7 35
Other well-bchavcd functions of x can be expressed as infinite series of Legendre polynomials P.(x), and a collection
of such formulas is presented by Gradshteyn and Ryzhik [Section 8.92J. Many such series expansions may be
obtained by the procedure explained in Section 21:14.
Among summation formulas for Legendre polynomials arc
21:5:6 (2n - 1)P._1(x) + (2n - 3)P._2(x) + (Zn - 5)P"_3(x) + - + 3P,(x) + Po(x)
n(P.-i(x) - P.(x)]
= x*I
I - x
and
21:5:7 (2n - 1)P._;(x) - (2n - 3)P._2(x) + (2n - 5)PP_3(x) - ± 3P,(x) + Po(x)
_ n[P._,(x) + P.(x)]
x + -1
1+x
21:6 THE LEGENDRE POLYNOMIALS P"(x) 188
21:6 EXPANSIONS
Definition 21:3:3 leads to the finite series
(n - 1)!! r n(n + 1) (n - 2)n(n + 1)(n + 3) X4- 1
x2+ n=0,2,4,...
(n/2)! 2! 4!
n!! (n - 1)(n + 2)zt (n - 3)(n - 1)(n + 2)(n + 4)
21:6:1 P"(x)
_U12
+ xa - ...]
n-1 3! 5!
2
n= 1,3,5.
that are valid for all arguments x. The infinite Fourier series
21:6:2 P"(cos(8)) = ?
G (+ 1)" sin[(n + I - 2j)01
it j_0 (j + Y2)".,
involves coefficients that are ratios of Pochhammer polynomials (Chapter 18]. A third expansion is
(n+j)!
21:6:3 P"(x)=1E
2 2 11
Other expansions can be found in Gradshteyn and Ryzhik [Section 8.9111 and yet others may be derived from the
formulas in Section 59:6 with v = n, or by combining equations 21:12:1-21:12:4 with the expansions given in
Chapter 60.
n - 0 I 1 1 I I
-x 1 0 1
(- 1)*"(" - 1)^
utilizes recurrence 21:5:2 and is exact although, of course, rounding may cause large relative errors close to the
zeros of P.(x).
21:10:2 J Pjr)dt =
P.-1(X) -+ P1.-i(x) n = 1. 2. 3....
i
21:10:3 J P.(t)dr-0
TI
P.(t)dt _ N/"8-
21:10:4
T , 2n+I
f 1 12)
(µ-n+2)(µ-n+4)(µ-n+6).. tµ- 2±
21:10:7 1 t"P.(t)dt =
/ \ µ> -2 ± 2
(µ+n+1)(µ+n-1)(µ+n-3)Iµ+2+ZI
21:11 THE LEGENDRE POLYNOMIALS P"(x) 190
where the upper/lower signs apply according as n is even/odd. Gradshteyn and Ryzhik devote several pages [Sec-
tions 7.22-7.251 to a listing of yet other definite integrals involving Legendre polynomials.
The integral of a product of P,(x) with some other function f(x) is frequently evaluable by utilizing Rodrigues'
formula 21:3:2 and parts integration ([Section 0:10]. Whenf the integration limits are ±1. the simple formula
1 d"f(t) (1
21:10:8 ' f(t)P,(t)dt = - tz)"dt
J 2'n! dt'
emerges.
The integral
0 mn
21:10:9
f P,(t)P"(t)dr = 2
2n + 1
in It
establishes the orthogonalirv of Legendre polynomials over the interval - I <- x s 1, a valuable property that is
discussed in Section 21:14.
The Hilbert transform [Section 7:10] of a Legendre polynomial
I ' P"(t)dt -2 OJs)
21:10:10 =
WT, r-s n
produces a Legendre function of the second kind [see Section 21:13]. a result known as Neumann's formula.
21:12 GENERALIZATIONS
As explained in Section 21:1, Legendre polynomials are the special v = integer cases of the Legendre functions
of the first kind, dealt with in Chapter 59. However, Legendre polynomials may be generalized in a number of
other directions.
Legendre polynomials generalize to give associated Legendre functions of the first kind, P,, '(x) (see Section
59:13], of which they are the v = integer. µ = 0 instances. Likewise. Legendre polynomials result as the v = µ
= 0 instances of the Jacobi polynomials P,` "'(x) that are discussed in Section 22:12.
Legendre polynomials are also special cases of the Gauss function [Chapter 60]. the simple identity
P,(x)=FI-n.n+1:I; 1-x
/
21:12:1 2
/
being known as Murphy's formula. Other relationships between Legendre polynomials and Gauss functions are
(2)7 2
- 1)!! (x - 1)"F( -n, -n;
-2n:
21:12:2 P,(x) _
n! 1 -X ,
21:12:3 P,(x) _ x.
n! 2 2 2
(n-1)!! (/-n n + I 1 \
n!!
F(
2 2 2
x'1 n=0.2.4....
``
21:12:4 P,(x) _
n!! l-n n+2 3 .r')
xF(- . n = I. 3. 5....
(n - I)!! '2
2 2 '
and still other formulas may be developed using the relationships of Chapter 60.
191 THE LEGENDRE POLYNOMIALS 21:13
and
5x'-3x 15x2-4
21:13:6 Q3(x) = 2 artanh(x) - 6 -1 <x< 1
where artanh is the inverse hyperbolic tangent function [see Chapter 311. The general case is
2n - 2j + I
21:13:7 Q(x) = P . ( - r ) artanh(x) - 2 1 j = 1, 3, 5, .... n - # }
J(2n - J - l)
'L 1, a 6 0 0 'L a 0 0 0 1. U 6
i°A" y®O
0.8
3x) p (x) : D 0 C):
:... .. .. .. -0.2
..:....1... ..X:.7..:.r .:.... :.... :.... :.............. :........... -0.4
: FIG 21-4 :
.
. .
..... ..... :A.: .... A...:1.4..:....:... . . . ...:.............. . .... :.... . .
-0.6
l 3 (x):
21:14 THE LEGENDRE POLYNOMIALS P,(x) 192
for - I < x < 1. For j > 1, the artanh function is to be replaced by arcoth. Many of the properties of P,(x) are
mimicked by Q,(x); for example, the recurrence formula
21:13:8
2n-I xQ,-i(x) n-1
- -Q,-2(x) n = 2, 3, 4, ...
n n
is obeyed. However, the symmetry properties of .(x) differ from those in 21:5:1 because Q,(x) is an even function
when n is odd, and vice versa.
21:14:1 0 m*n
fj
but
21:14:2
f f2 s 0
The family is said to be orthonormal if 21:14:1 and 21:14:2 hold and if f2, = 1, n = 0. 1. 2. .... The function
w(x) is known as the weight function; it is required to be nonnegative on the interval xo s x s x,. For example,
integral 21:10:7 shows that the Legendre polynomials constitute an orthogonal family on the interval - I x :s I
with a weight function of unity.
The orthogonality property permits the functions 'Po(x), 'P,(x), 'P2(x), ... to be used as a basis set for the
expansion
21:14:3 f(x) = c0'P0(x) + c,W,(x) + c=Yr2(x) +
of a wide range of functions f(x). The constants c0, c,. C2. ... in this so-called orthogonal series or generalized
Fourier expansion may be evaluated by the integral
21:14:4 C. = z, J
Chebyshev polynomials of the first kind, T (x), constitute an orthogonal family [Section 21:141 on the interval -1
s x 5 1 with a weight function I / 1 -x2. Similarly, the second kind of Chebyshev polynomials, U,(x), are
orthogonal on the same interval with weight function 1 - x2. Chebyshev polynomials are extensively used in
"fitting" techniques, such as those discussed in Section 22:14.
22:1 NOTATION
Alternative transliteration from the Cyrillic alphabet leads to spellings ranging from Chebyshev to Tschebischeff
for the names of these polynomials.
Unfortunately, mathematicians differ in their definitions of Chebyshev polynomials, and there is not even una-
nimity on which family constitutes the first and which the second kind. The notations L(x) and Ujx) may be
encountered with meanings equivalent to the T,(x)/2'-' and nU,_,(x) of this Atlas. Confusingly, the symbol U,(x)
and the name Chebyshev polynomials of the second kind are commonly applied to functions defined by our
V'1 -x i U,_i(x) for n = 1, 2, 3.... and by arcsin(x) for n = 0, despite these not being polynomial functions at
all.
Several supplementary notations are encountered. T; (x) and U! (x) are so-called shifted Chebyshev polynomials
equal to our T,(2x - t) and U,(2r - 1). The symbols C,(x) and S,(x) have been used for 2T,(x/2) and U,(x/2),
respectively. Chebyshev functions. T.00 and .(x) are defined identically to the corresponding polynomials, but n
is not restricted to be an integer. None of these supplementary notations is employed in this Atlas.
22:2 BEHAVIOR
The degree n of a Chebyshev polynomial may take any nonnegative integer value, although U0(x) is not a member
of the orthogonal family [see Section 22:101. Interest in Chebyshev polynomials is confined to the - I s x S I
range, as portrayed in Figures 22-1 and 22-2, although several of the definitions remain valid outside this argument
range.
Chebyshev polynomials of each kind, T,(x) and U,(x). have exactly n zeros, Int(n/2) minima and lnt[(n - 1)/
2] maxima, all of which lie within the -I < .r < I range. The T,(x) polynomial takes the value -I at each
minimum and + I at each maximum, but no comparable rule applies to the Chebyshev polynomials of the second
kind. The polynomials are confined to the ranges
22:2:1 -1 T,(x)S1 -1SxSI
193
22:3 THE CHEBYSHEV POLYNOMIALS AND 194
0 O M ti 1. O 0
44' 14' 44r 44' 440' 440'
440'
440'
T4(x) \ ,TS(z)
FIG 22-1:
and
22:2:2 -n- 1 1 -1 sx I
22:3 DEFINITIONS
Chebyshev polynomials are given by the trigonometric definitions
22:3:1 T,(x) = cos(nO)
0 = arccos(x) - 1 5 x s 1
22:3:2 csc(0) sin[(n + 1)01
0 0 0 ti ti o 0 0
22:4 THE CHEBYSHEV POLYNOMIALS T"(x) AND U"(x) 1%
1
22 : 3 : 5
1-21x+r2 "=o
or by the Rodrigues formulas
d" "
22:3:6 T"(x) (1 - x')"
(2n - 1)!! dx
(-1)"(n+ 1) d" (I -x2)"-ir
22:3:7 U, (z)=
(2n + if
For n = 1. 2, 3. the most general solution of Chebpshev's differential equation
df df
22:3:8 (1-x2)--x
dx2
-+nf
dx
0
is f = c,T"(x) + c2 1 - z2 U"_,(x). where c, and c2 are arbitrary constants. For n = 0 the solution is f = c, +
C2 arcsin(x).
22:4:3 U,(x) = 2x
22:4:4 T2(x) = 2x2 - 1
22:4:5 U2(x) = 4x' - I
22:4:6 T3(x) = 4x 3 - 3x
22:4:7 U3(x) = 8x3 - 4x
22:4:8 T4(x) = 8x4 + 8x2 + 1
22:5 INTRARELATIONSHIPS
Chebyshev polynomials are even or odd
22:5:1 f,(-x) _ (- I)"f.(x) f = T" or U.
according to the parity of the degree n. The recurrence formula
22:5:2 f.(x) = 2xf _,(x) - f._2(x) n = 2, 3, 4....
also applies equally to both kinds of Chebyshev polynomials.
The formulas
22:5:3 T.(x) = U.(x) - xU._,(x) n = 1, 2, 3, ...
and
xT._,(x) - T.(x)
22:5:4 U (z) = n=1,2.3....
1 -x2
permit one kind of Chebyshev polynomial to be expressed in terms of the other.
Positive integer powers can be expressed as finite series of Chebyshev polynomials of either kind. For example:
1 I I l
22:5:5 x= = 2 T0(x) + 2 T2(x) = Uo(x + U2(x)
4 4
22:5:6 x'=
3
3
4
T,(x)+ I4 T,(x)= 4 U1(x)+ I U;(x)
1 1
and generally
where 0 if j > n or if j and n have unlike parities. The simplest coefficient yo' equals unity, and others can
be calculated by means of the recursions y;"' _ 2 y;=;" + 2 y;"-i" except -y"' = Yo 1) + 2 yz and Yo
yi"-11
There are formulas relating the product of two Chebyshev polynomials to (usually the sum of two) other Che-
byshcv polynomials:
1 1
22:5:8 T.(x)T,.(x) = 2 T".,.() + - T._ (x) n L, m
T_(x) -
22:5:9 U,(z)U,.(x) = nsm
2(1 - x2)
1 I
1 1
as well as the so-called formulas for sums of products of Chebyshev polynomials of different
arguments:
1 T..j(x)T,(y) - T,(x)T,.dy)
22:5:11 T; (x)T; (y) _
2 x-y
22:5:12 I -
+7 U,(x) U ,(y) - U,.,(x)U,(y) - U.(x)U,.,(y)
-
22:6 THE CHEBYSHEV POLYNOMIALST.0) AND U.(x) 198
x - T..2(x)
22:5:16 U1(x) + U3(x) + U5(x) + + U.(x) = n = 1. 3. 5....
2(1 - x2)
22:6 EXPANSIONS
Explicit expressions for the two Chebyshev polynomials are
and
rn + I) - (n I) `
22:6:4 U"(x) = - x 2) + 1 n + I I x"-4(l - x')= - .. .
X. 3
If t;," represents the coefficient of x1 in the expansion of T.(x) so that
22:6:6
((n+-J
then tw1' = 0 when j and n are of unlike parity. All nonzero values of t," are integers given by
j-
(-1)4.-))/2n(
2
1 2J-1/ \n
)!J! j = 1, 2. 3..... n
except that t`,° equals unity.
n=0 I I I
n-I
1 1 1
1 ,3 . 5 , -I 0 1 -n- 1 0
n=2.4.6.... i 1 f-1Y'' 1 n+l (-I) n+I
199 THE CHEBYSHEV POLYNOMIALS T,(x) AND U,(x) 22:10
Setf=g={2x
If n = l go to (2) Input restrictions: n must be a nonnegative integer. x
Setf=h=j= 1 may adopt any value.
If n = 0 go to (2)
(1) Replace j by j + I
Replace f by 2xg - h Test values:
If j = it go to (2) To(any x) = 1
Replace h by g U1(x)=2x
Replace g by f T12(0.2) = -0.748302037
Go to (1) U,(0.5) = 0
(2) Output f 1.239131013
l
f= U,(x))
22:9 APPROXIMATIONS
Figures 22-1 and 22-2 can provide approximate values of T,(x) and U,(x) for degrees up to 5. For large values of
the argument, the following approximations are valid:
d 1
22:10:2
dx
U.(x) =
7- i [(n + 1)U.-i(x) - nxU.(x)]
The Chebyshev polynomial of the second kind may be integrated indefinitely
r' 1 - T.- I (x)
22:10:3 U , (r)dt =
J n+1
but we know of no corresponding integral for T.W.
The definite integral
f0 m*n
f' Tj:)T,(t)
22:10:4 dr= J ir m=n=0
IT
m=n=1,2,3,...
2
establishes that the polynomials T0(x), T,(x), T,(x), ... are orthogonal (see Section 21:14] on the interval -1 S x
S I with respect to the weight function I / 1 - x2. Likewise, the integrals
0 mm #=n or m3,=..n.=0
22:10:5 1 - t2 dt = aa
1 2 n=1 ,2,
2
establish the orthogonality of the polynomials U,(x). U2(x), U3(x).... on the same interval with respect to the
weight function V 1- x2.
T
T"(r)dr =
4n2-22:10:6
Many other definite integrals are listed by Gradshteyn and Ryzhik [Sections 7.34 to 7.36]. including
4n2 - 1
f' t T.(t) trr(v + 1)
22:10:7
0 1-r2
dr =
2 ,rlv n+llr(vn}I
\
2 v>-1
(-1)./2a
22:10:8 2 b>0 n=0,2,4,...
J.
0 1 - rr
and
sin(bt)T (22:10:9
f 1-t2 2
dr= 2 b>0 n=1,3,5,
where r denotes the gamma function [Chapter 43] and J. the Bessel coefficient [Chapter 52] of order it.
On Hilbert transformation (Section 7:10] the product of one kind of Chebyshev polynomial and its weight
function gives the other kind:
1 T,(t)dr
22:10:10 J = U.-I(s) -1 <s< I
a 1
12 (tS)
r +_ dr _
22:10:11 J -T.- ds)
r
77 i t - s)
22:12 GENERALIZATIONS
Chebyshev polynomials may be generalized to Gegenbauer polynomials, which are themselves special cases of
Jacobi polynomials. This section briefly discusses both of these generalizations.
The Jacobi polynomial or hypergeometric polynomial is a quadrivariate function, its value being de-
pendent on: the argument x, which may take any real value with interest being concentrated on the -1 s x <- 1
range; the degree n, which takes nonnegative integer values; and two distinct parameters, v and µ, both of which
exceed -1. These polynomials may be defined by the Rodrigues expression
1 d"
22:12:1 PA":w(X) = [ (1 - X)"' "( I + x)"
(-2)"n!(1 - x)"(1 + x)" dx"
They obey the reflection and recurrence formulas
22:12:2 P;,"(-x) _ (-1)"P;; `'(x)
pA.1:w(x)=(2n+v+µ- 1)((2n+v+µ-2)(2n+v+µ)x+v2-µ(X)
22:12:3
2n(n+v+µ)(2n+v+IL -2)
(n+v-1)(n+µ-I)(2n+v+ µ) n(n+v+µ)(2n+v+µ-2)P"i(x)
and
+ 2)x + µ
22:12:5 P;" "'(x) = (v + Z
2
A binomial coefficient (Chapter 6] expresses the value of the Jacobi polynomial at unity argument
!n + v)
22:12:6 P'":w(1) =
It
Jacobi polynomials are orthogonal [Section 21:14] on the interval - I s x s 1 with weight function (I - .r)'
(I + x)"
0 m+n
22.12:7
f (I - t)"(I + 1)"P" w(t)P (t)dt = 2 "''l'(n + v + I)f(n + µ + 1)
(2n+v+µ+l)n'f(n+v+µ+ 1) m=n
For more information about Jacobi polynomials, the reader is referred to more advanced sources. particularly Szego.
Jacobi polynomials can be expressed in several ways as Gauss functions [Chapter 60]. For example:
P;,""'(x)=1nn
)F(-n,n+v+µ+1;v+1;12x
22:12:8
n //I \ /J
and numerical values are therefore calculable using the algorithm of Section 18:14.
Gegenbauer polynomials, or ultraspherical polynomials, are related to those Jacobi polynomials in which the
two parameters are equal: v = A. They may be defined by
22:12:9
(2),)"
(X + '/2)"
P'a-,rza- 1120(x)
-! < 1, + 0
2
Legendre polynomials [Chapter 21 ] and each kind of Chebyshev polynomial are special cases of Gegenbauer
polynomials
22:12:11 P (x) = C°'21(x) = V0.91(X)
n o (2n)!! -112-v2)
22:12:12 T.(x) = - C(x) = (x)
2 (2n - 1). P'
(2n + 2)!' Pn/2:1/2)(x)
22:12:13 U (x) = C;'(x) =
2(2n + I)'!
while Hermite polynomials [Chapter 24) are generated from Gegenbauer polynomials by the limiting process
22:12:14
may be satisfied by a finite set of polynomials `l'o(y), 4'1(Y), 'Y,,(y), known as discrete orthogonal
polynomials.
The simplest of these are the discrete Chebyshev polynomials to(y), t (y), t2(y), ..., tt(y), which correspond
to weights w(y) in 22:13:1 equal to unity and should not be confused with the coefficients discussed in Section
22:6. With the variable y confined to the range -1 :5 y s 1, the orthogonality condition is
0 m n
22:13:2
j.0
where
G^=(J+n+l)!(J-n)!- 1 (J+I)..1
22:13:3
(2n+1)(J!)2 2n+1(J-n+1)
Values of the discrete Chebyshev polynomials generally depend on J as well as on the argument y. The first
few air:
22:13:4 to(y) = 1
22:13:5 tt(y) = y
3Jy2-(J+2)
22 : 13 :6 t2(y) =
2(J - 1)
5J2y3-(3J2+6J-4)y
22 : 13 :7 tr0')
2(J-1)(J-2)
and others may be calculated via the recursion formula
(2n - 1)
22:13:8 t.0') - n(J - n + I) [yt,.-i(y) - t.-2(y)1 + t.-2(y)
The practical importance of discrete Chebyshev polynomials depends on the role that they play in the fitting
of low-degree polynomials to equally spaced data points [see Section 17:14]. Let a K-degree polynomial pk(y) be
sought that passes close to the points (yo,fo), (yi.fi). (y2.f2), , (yj,fj), as shown in Figure 22-3, so as to satisfy
203 THE CHEBYSHEV POLYNOMIALS T"(x) AND U"(x) 22:14
1 O 1
< .o!9 . .. ..
A ..:..:..:..:...:...... .
..:....:.........:....+!....:....:....:..
22:13:9
1=0
J
[f,-pK(Y1)12 Y,-1--J 2%
where
1 J
22:13:11 as=-j4(Y1)
f1k J-0
These relationships are exploited in the algorithm presented in Section 17:14.
Despite their name, discrete Chebyshev polynomials have less in common with Chebyshev polynomials than
they do with Legendre polynomials, to which they reduce as J approaches infinity:
22:13:12 1im t"(y) = P"(Y)
J-- "
whose argument has been defined so that the region of interest is -1 S y S 1. The second phase uses the binomial
relationship
j, [
22:14:4 b,= h'
k )akuk-'_ h ' [aiJI+ akk!u-'/(k-j)!
i k-j- I I
to calculate the new coefficients and stores them in place of the original set.
The third phase computes the coefficients co. c,, c2, .... c, such that
22:14:5 1""}1
i-o i-o i=0 k-0
by first exploiting the fact that, because only the T,(y) polynomial contains a term in y", c" = b,12"'
(if n * 0). Using expression 22:6:6, c,t'1 is now subtracted from each stored b, other than b" so that the degree
of the I b,v polynomial is reduced to n - 1. The procedure is then repeated until all c, coefficients have been
calculated and their values used to replace the now-redundant b, coefficients.
To start the fourth phase of the algorithm, the maximum permissible error a is input. The smallest integer m
that satisfies
22:14:6 i 10 se
i-.-1
is then calculated by the algorithm, output, and used to replace n in storage. Thereafter, no further use is made of
coefficients c,.,, c,,.2, c,.
The fifth phase reverses the process of the third phase, replacing I c,Tj(y) by the polynomial E d,y' of degree
m. The properties of tk' (see Section 22:61 are used in the identity
22:14:7 c,
tik,vk t` tii c, ttil +, ) ckt1j1
= Lr LJ
J.0 1=0 I'0 k-0 a 1 k
where in the final summation k takes the values j + 2, j + 4, j + 6, ..., (m or m- 1) to establish that
22:14:8 dj = 2'-'c, +
(-1)'k-iU2kck(k + j
L \k 2 ,l/ !j !
2
- 1 !2i-'
l
except that, when j = 0, the 2'-'c1 term is to be doubled. The algorithm replaces co, c1, c2, ..., c in storage by
do, d1, d2, .. , d,,.
The sixth phase of the algorithm reverses the effect of the second phase in that it converts I d, y' to E e,x' and
replaces the m + I stored values of d, by the corresponding ej coefficients. The binomial identity
e1xidj(x-u)j(-u)'-idkk!
22:14:9
i.0 j.0 h hll J-0 j ! k-j (k - j )!hk
shows that
(_u) k-i
22:14:10 ei` 1,jli!di(kdkk
h
k-i,1 j)! h
1
1
and this is the relationship used by the algorithm.
The seventh phase simply outputs the economized coefficients in the order e0, e e2, ..., e,,.
Input xo »»» Storage needed: n + 7 registers are
Setu=xa required for u, h, n, j, k, e, m and
Input x, »»» f o r the coefficients ao, a a2, .
Set At = (x, - u)/2 a,,. Note that the n register may be
Replace u by u + h used to store m and that the a reg-
Input n >>>>>>
Setj = 0 I ister is used for temporary storage
twice by the algorithm. The coef-
(1) Output j ,4 ficients denoted bo, b,, b2, ..., b,:
j(as cue) <<<<<< co, ca, C2, - , c,; do,d1,d2, , d.,
Input a; >>>>>> and e0, e,, e2, ..., e,, in the text are
Replace j by j + I stored in the same registers as ao,
Ifj :5ngoto(I) a a2, .... (a or a,,) and are de-
noted a, in the algorithm.
Seth=0
(2) Replace a; by a,j!
Ifj = n go to (4)
Setk=j+ I
(3) Replace a, by aj + akk!uk-'/(k - j)! °
Replace k by k + I c
Ifk:5 n go to (3)
(4) Replace aj by a,h'/j!
Replace j by j + I
Ifj:5 n go to (2)
Set kn
(5) Set j = k
Replace a, by a,/21-'
(6) Replace j by j - 2
Ifj<0goto(7) 2
bete=Vc-j)/e
Replace a, by a, - (-I)'kak(E + j - 1)!2'-'/(e!j!)
Go to (6)
(7) Replace k by k - 1
If k * l go to (5)
Setj=n
Input a »»»
(8) Replace a by e - Ia,I
Replace j by j - 1
Ife>0goto(8)
Setm=j+1 0
Output m
M <<<<<<<<<<
-----------------------------------------
Replace ao by 2ao
Setj= -I
(9) Replace j by j + I
Ifj>mgoto(11) II
Replace a, by a, 2'-'
Setk=j
(10) Replace k by k + 2
Ifk>mgoto(9)
Set s = (k - j)/2
Replace a, by a, + (-1)'kak(e + j - l)!2'-'/(e!j!)
Go to (10)
----------------------------------------
205
22:14 THE CHEBYSHEV POLYNOMIALS T,(x) AND U,(x)
(II) Setj=0
(12) Replace a, by aj!
If j = m go to (14)
Set k = j + 1
(13) Replace a, by a, + q,k!(-u/h)'-'/(k - j)!
Replace k by k + I
If k :5 m go to (13)
(14) Replace a, by a,/(hJ!) Test values: xo = 0, x, = 1, n =
Replace j by j + 1 9.ao= 1, a, _ -1,a2= 1/2,a,
if jsmgo to (12) = -1/6, a, = 1/24, a, = 1/120,
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - --- a6 = 1/720, a, _ -1 /5040, as =
Set j = 0 1/40320, a9 = 1/362880 [i.e..
(15) Output a, f(x) = exp(-x)], e = 1/256
.5 Output: m = 2, eo = 0.99658, e,
Replace j by j + 1 = -0.93532, e2 = 0.30963
1
Because of the very large number of arithmetic operations executed by the algorithm, rounding errors may
accumulate and the values of the economized coefficients e0, e,, e2, ..., e generated by a program may not be
quite the "best." To evaluate this effect, the magnitude of which will depend on the characteristics of the particular
computing device employed, one may repeat the calculation with e = 0. Then m will equal n and the coefficients
eo, e,, e2, ..., e should, but will probably not exactly, equal ab, at, a2, ..., a,.
Apart from these rounding errors, the economized series E e,x' is guaranteed to differ from the original F a,x'
series by no more than a over the xp s x s x, range. However, the actual absolute error 1E a,xJ - E e,x'I will
often be less than e. This actual economization error E is found in the following algorithm by calculating the
maximum value of
22:14:11 i1.0Axj 1
at one hundred points in the xo < x s x, range. The input coefficients A, are given by
22:14:12
a, - e, j =0, 1,2,...,In
A' a, j = m + 1,m+2,m+3,...,n
Input xa > >>>>> Storage needed: n + 8 registers are required for Ao,
Setx=x0 A,. A2, ..., A,,, x, h, j. k, n, E and f.
Input x, > >>>>>
Set h = (x, - x)/99
Setj=k=E=0
Input n >> »»
(1) Output j
j(as cue) < ««
Input A, > >>>>>
Replace j by j + I
If j5ngo to (I)
(2) Set f = 0
Replace j by n + 1
(3) Replace j by j - 1
Replaccfbyfx+A,
Ifj*0goto(3)
If E> If go to (4)
207 THE CHEBYSHEV POLYNOMIALS T,(x) AND U,(x) 22:14
The Lagucrre polynomials are orthogonal [see Section 21:14) on the interval 0 s x s x with a weight function
exp(-x) [see Chapter 261. They arise in a variety of scientific applications, for example, in solutions to the wave
equation of quantum physics.
23:1 NOTATION
Though the L (x) symbolism is used for both, Laguerre polynomials are defined in two different ways. Some
authorities use definitions that differ by a multiplicative factor of n! from those adopted in this Atlas. The termi-
nology Laguerre polynomial is sometimes applied to the more general class of polynomials that we discuss in
Section 23:12 under the name associated Laguerre polynomials.
Although this Atlas avoids the ambiguity, the symbols L (x) and L,.(x) are sometimes used to denote the hy-
perbolic Strove function [see Section 57:131.
23:2 BEHAVIOR
While interest is usually confined to positive arguments, the Laguerre polynomial L (.r) is defined for all values
of the argument x and all nonnegative integer degrees n. Laguerre polynomials of odd degree adopt all values,
whereas those of even degree have a restricted range. Figure 23- I shows the behavior of the first few members
of this family.
The Laguerre polynomial L (x) has exactly n zeros. lnt(n/2) minima and Int[(n - 1)/21 maxima, all of which
are located in the range 0 < x < 2n + I + 4n2 + 4n + (5/4). For x > 2n + I + V57+ 4. L (x)
increases or decreases monotonically according to whether n is even or odd, remaining bounded by
23:2:1
-exp ( --2) s L,(x) K exp (;) x >_ 0
23:3 DEFINITIONS
Laguerre polynomials are defined by the generating function
23:3:1 1
1 r exp1 xr!) - L,(x)t" -I<t<I
209
it" t t4vb
? ?' I
.5
.4
L4'
•0
—1
—2
—3
•
. —5
FIG 23—1
211 THE LAGUERRE POLYNOMIALS L (x) 23:5
or by Rodrigues' formula
exp(x) d" 1 1
23:3:2 L,(x) _ x" exp(-x) n = 0, I, 2, ...
n! dx"
23:3:3 L,(x)=limP;°'" I-
links the Laguerre polynomials to Jacobi polynomials [Section 22:121, while the integral representation
exp(x)
23:3:4 L,(x) = J ; t' exp(-t)Jo(2\ tx)dt
n! u
23:3:5 xaf
dx2
+(I -x)af +nf=0
dx
is f = cL,(x), c being an arbitrary constant.
23:4:2
23:4:3
23:4:4
x5 5x' 5x'
23:4:6 L5(x) 5x - 5x + l
= - 120 + 24 - 3 +
x° x`5x' lox' 15x22
23:4:7 Lb(x)720 20+ 8 - 3 + 2 -6x+1
x' 7x° 7x5 35x' 35x' 21x'
23:4:8 L,(x) - 7x + 1
5040 + 720 40 + 24 - 6 + 2
23:5 INTRARELATIONSHIPS
We know of no reflection or recursion formulas that express L,(-x) or L,(x + 1) in terms of L,(x). but there exists
the recurrence relation
2n-i-x n-I
23:5:1 L,(x) = L,_,(x) - - L,_2(x) n = 2. 3, 4...
n n
relating a Laguerre polynomial to its congenitors of lower degree but identical argument.
23:6 THE LAGUERRE POLYNOMIALS L,(x) 212
where L°' is an associated Laguerre polynomial [Section 23:12]. A series of Laguerre polynomials also occurs in
the multiplication formula
23:5:4 L"(bx)=(1-b)"1
while the addition formula
23:5:5 L(x + )) = ( ll
(n) H,,(vc)H"-t,(\)
.a i
involves a sum of products of Hermite polynomials [Chapter 241. The formula for Laguerre
polynomials is
23:5:6
n+l
, L (x)L (v) _ - IL"(x)L".,(y) -
x-y
23:6 EXPANSIONS
An explicit expression for the Laguerre polynomial is
23:6:1 L,(x)=(-I)" +
n(n - 1)x"-2 - +(-1)"
1
n! 1!(n - 1)1. 2!(n - 2)!
_l
i_o
n=0 1 1 1
We know of no exact expressions for the arguments of the zeros, minima or maxima of the Laguerre poly-
nomials, beyond the first few.
If n = 0 go to (2)
Input m >>
Input x >> t»»Replace f by I + m - x Input restrictions: n must be a
nonnegative integer. x and m are
unrestricted.
If n = I go to (2)
(1) Replace j by j + I
Set g = f
Replace f by [(2j + m - I - x)g - (j + m - I)hl/j
Replace h by g Test values:
Ifj<ngoto(I) l4any x) = 1. L,(9) = -8
(2) Output f L6(8.5) = 12.6229384
f = L`-(x) L'(1) = 2.33333333
23:9 APPROXIMATIONS
For large x one can approximate
23:9:2 L.(x) = + 2N
WN J
where N = 4n + 2. For further details of the asymptotic expansion 23:9:2 and for approximations to L,(x) as n
- z for other ranges of x, see Erddlyi. Magnus, Oberhettinger and Tricomi [Higher Transcendental Functions.
Section 10.151.
and
23:12 GENERALIZATIONS
In previous sections of this chapter we refer to the family of associated Laguerre polynomials L; "(x) of which the
Laguerre polynomial is the m = 0 case
23:12:1 L'.Q1(.x) = L"(x)
exp(x) d"
23:12:3
n!x" dx
2-rI
23:12:4 L;,'`(x) = lim I-
µ-= µl11
exp(x)
23:12:5
n
)
23:12:7
( n-J 1!
and the early members are listed in Table 23.12.1. Numerical values of L'. '(x) are given by the algorithm in Section
23:8 if the portions printed in green are included. Analogs exist of most of the formulas of Sections 23:5, 23:9
and 23:10, including the following recursion, differentiation and orthogonality formulas:
In +m- I -x n+m - 1 x-n L'.7- "(x) + n+,n- I L' 1 "(x)
23:12:8 L;'(x) = L;;_',(x) -
n n x x
Table 23.12.1
x' i.c
m-0 -x+ l 2-2x' 1
x' 3x'
- 6 + 2 - 3x+ 1
24
-+3x'-4x+ I
3
m=0
X2
X
m=1 -x-2 - -3.x+3 - x'6-2x'-6x+4 24 6
,+Sx'- lox+5 m = 1
X2 7x' 21x'
m=3 -x+4 2
-5x+I0 --+3x=-15x+20
6 24
- + - - 35x + 35
6 2
m=3
-a+--21x+35
7x' x' 4x'
m=4 -x+5 2
-6x+I5 6 2 24 6
+ 14x' - 56x + 70 m=4
215 THE LAGUERRE POLYNOMIALS L (x) 23:14
which definition leads to properties radically different from those of the polynomials defined by 23:12:2-23:12:6.
According to Sneddon [Section 44], the definitions adopted in this Atlas are generally favored by pure mathema-
ticians, whereas 23:12:11 is preferred by applied mathematicians. Although interest concentrates on nonnegative
integer values of the order m. definitions 23:12:2-23:12:7 may remain valid for negative integer m. and even for
noninteger values provided m > - 1. Polynomials so defined are often named generalized Laguerre polynomials.
the simplest instances are
23:12:12
2 4
n! x)
and
1 H2,(Vx)
23:12:13
4) " n
where H is the Hermite polynomial [Chapter 241.
The Kummer function [Chap(er 47) is a broader generalization of the Laguerre polynomial and of its associated
counterparts. The relationship is
(n ml M(-n;m + l;x)
23:12:14 L'`(x) = y J
should be added.
The generalized Laguerre function is addressed in Section 47:1. Alternatively, the expression
/ xl dam"
23:14:2 (n + 1)! expl L"'(x)
2 / x'- --j
23:14 THE LAGUERRE POLYNOMIALS 216
is defined as the Laguerre function. It is a solution to the radial portion of the Schrlidinger equation for a hydrogen-
like atom
dZj+2dj_ I n+f(I+ 1)
23:14:3 Jf=0 n?1+ I
dx2 x dx 4 x x
where it and I are the appropriate quantum numbers [see Sneddon. Section 44].
CHAPTER
24
THE HERMITE POLYNOMIALS H,t(x)
The Hermite polynomials are orthogonal [see Section 21:14] on the interval -x < x < x with a weight function
exp(-.x2) [see Chapter 271. They occur in several applications to physical problems, for example, in the solution
of the Schrodinger equation [see Section 24:141.
24:1 NOTATION
The notation H.(x) is in general use for the Hetmite polynomial of degree n and argument x. An alternative Hermite
polynomial, symbolized is discussed in Section 24:13; unfortunately, it also is sometimes denoted
especially in texts on mathematical statistics.
Occasionally, Hermite polynomials of odd degree are defined with the opposite sign to that adopted in this
Atlas.
Although this Atlas avoids the ambiguity, the symbol H (x) is sometimes used for the Struve function (see
Chapter 57).
24:2 BEHAVIOR
The Hermite polynomial has exactly n zeros, Int(n/2) minima and Int((n - 1)/2) maxima. These features
are symmetrically disposed about x = 0. and all occur in the region -V2n < x < V -2n. Outside this range IH,(x)l
increases monotonically. except for Ho(x), remaining bounded globally by
24:2:1 IH,(x)I < I .09 2" n! exp(x=)
The range spanned by the oscillations of H.(x) increases so dramatically with increasing n that Figure 24-1
portrays the behavior of H.(x)/n!. rather than that of the Hermite polynomials themselves.
24:3 DEFINITIONS
Hermite polynomials are defined by the generating function
_o n ,
217
....:....: ...: ...
: H4(x)/41:
219 THE HERMITE POLYNOMIALS 24:5
24:3:5 --2x-+2nf=0
is f = cH,(x), where c is an arbitrary constant; a second solution is a closely related infinite series [see Murphy.
page 321, but note that his definition of H,(x) differs from ours]. Another important differential equation
df
24:3:6 = (x' - 2n - 1)f
dx '
is solved by f = c exp(-x2/2)H"(x) [see Section 24:141.
24:5 INTRARELATIONSHIPS
The reflection formula
24:5:1 H,(-x) _ (-I)"H,(x)
shows that a Hermitc polynomial is an even or odd function according to whether its degree is even or odd. The
addition formula
H,(x + y) = 2-"'
(n)
24:5:2 H,(xV2)H,.. (yV2)
24:6 THE HERMITE POLYNOMIALS H.(x) 220
H.(x) HS(x)
24:5:5 He(x) - = + - - ... = e sin(2x)
3! 5!
where e is the base of natural logarithms [Chapter 11, Series akin to 24:5:4 and 24:5:5 in which the signs do not
alternate sum to (cosh(2x)]/e, and [sinh(2x)]/e, respectively.
The recursion formula
24:5:6 H,(x) = 2xH,_,(x) - 2(n - 1)H,_2(x) n = 2, 3, 4, ...
together with H0(x) = 1 and H1(x) = 2x, permits the polynomial form of any Hermite polynomial function to be
evaluated.
24:6 EXPANSIONS
The explicit expansion of the Hermite polynomial may be written in several equivalent ways; for example:
n(n - 1) n(n - 1)(n - 2)(n - 3) "_,
24:6:1 H,(x) = (2x)° - (2x).-2 +
(2x)
1! 2!
n!
. (-I)J(2x)"- zr J = Int(n/2)
1-0 j!(n - 2j)!
n-0 r r I
0
(-2Y" (n - 1)'
Because of relationship 24:10:1, certain characteristic values of the argument correspond to features in a number
of consecutive Hermite polynomials. For example, the four values ±V(3 ± V)/2 are the zeros of H4(x), generate
maxima or minima of H5(x) and correspond to inflection points of H6(x).
The asterisks in Figure 24-2 show the exact locations of the zeros for H1(x) through H,2(x). The colored lines
in this diagram are explained in Section 24:9.
Input n >> »»
Ifn=0goto(2)
Input x >> »»
Set f = 2x Input restrictions: n must be a nonnegative integer; x
lfn=Igoto(2) is unrestricted.
(1) Replace j by j + 1
Set g = f
Set f = 2[xg + (1 - j)hJ Test values:
Set It = g Ho(any x) = 1
Ifj<ngoto(I) H,(-12) -24
(2) Output f H7(4.2) = 1436292.99
f=H"(x)< ««
utilizes equation 24:5:6 to generate exact values of the Hermite polynomial. Hermite polynomials may also be
enumerated via the universal hypergeometric algorithm of Section 18:14.
holds. For large it, the Hermite polynomial is approximated by the oscillatory function
(-2r'2(n - 1)!! cxp('-/2) cos(x 2n + 1) even n -, x
24:9:2 H"(x) = 2
(-2)`" "/' - n!! exp(x2/2) sin(x 2n + 1) odd n x
Vn
This approximation is best for small I and fails hopelessly for lxI > N/:);;, that is, when the Hermite polynomial
leaves its oscillatory region [see discussion in Section 24:2J.
440' %.
C4
441:Iro+4 ej
44 + i4 i4 44 44 44 44 441 0 44 0 44ti
\n14
.. a ..,t x\. x:.. a 12
*n=10
.....:....:. *: *. *.... ..nab
*\:
FIG 24-2
n-2
:.........:.... :....:....:....:.... ....
. . . . .
..' n=0
24:10 THE HERMITE POLYNOMIALS H"(x) 222
The lines drawn on Figure 24-2 connect the zeros predicted by this approximation, blue for even n and green for
odd n. Evidently, the approximation is good for the inner zeros but worsens the further the zero is from x = 0.
d
24:10:2 [exp(-x')H"(x)] = -exp(-x')H"-,(x)
dx
( H".,(x)
n=0.2,4,...
2(n + 1)
24:10:3 rr H It'-it =
J1 H" ,(x) - (-2)". "n
n= 1.3,5,
2(n+ 1)
and
exp(-x2)H".1(x) n = 2, 4, 6, .. .
24:10:4 J exp(-t')H"(t I = i
o (-2)'"-'11'(n - 2)!! - exp(-s-)H"_,(x) n = I, 3. 5, .. .
Three important definite integrals are
24:10:7
f. r" exp(-t')H"(bt)dt = \n!P"(b)
where P. is the Legendre polynomial of degree n (Chapter 21], and many others are listed by Gradshteyn and
Ryzhik [Sections 7.37 and 7.38].
The orthogonality of Hermite polynomials is established by
r 0 m*n
24:10:8
J = exp(-t2)H"(t)H,"(t)dt = jV'n 2"n! m=R
24:12 GENERALIZATIONS
Hermite polynomials are special cases of the parabolic cylinder function [Chapter 46J in which the order is a
nonnegative integer:
Because the parabolic cylinder function itself generalizes to the Tricomi function [Chapter 481 or the Kummer
function [Chapter 471 these latter functions may also be regarded as generalizations of Hermite polynomials
/-n 1 .\
24:12:3
-(-2)`" 1) n!! M/1-n 3 x'
2 ; 2; n = 1. 3. 5....
The logarithmic function may be regarded as the simplest transcendental function, that is, ln(x) is the simplest
function of x that cannot be expressed as a finite combination of algebraic terms. Prior to the advent of electronic
calculators, logarithms [especially decadic logarithms. see Section 25:14] were extensively used as aids to arithmetic
computation, both in the form of tables and in the design of devices such as slide rules.
In addition to its treatment of the logarithmic function, this chapter includes brief discussions of the logarithmic
integral li(x) and the dilogarithm diln(x).
25:1 NOTATION
The name logarithm of x is used synonymously with logarithmic function of x to describe ln(x). Alternative no-
tations are log(x) and log,(x). although the former is also used to describe a decadic logarithm- The initial letter
is sometimes written in script type to avoid possible confusion with the numeral "one." See Section 25:11 for the
significance of Ln(:).
To emphasize the distinction from logarithms to other bases [Section 25:141, In(s) is variously referred to as
the natural logarithm, the Naperian logarithm, the hyperbolic logarithm or the logarithm to base e.
25:2 BEHAVIOR
The logarithm In(x) is not defined as a real-valued function for negative argument [but see Section 25:11] and takes
the value -x when x is zero. As shown in Figure 25-1, In(x) is negative for 0 s x < I and positive for x > 1.
The slope dln(x)/dx continuously decreases as x increases so that, even though ln(x) increases without limit
as x tends to infinity, its rate of increase becomes ever more leisurely. In fact, In(s) increases more slowly than
any positive power of x:
25:2:1
In(x)
-.0 x-,x v>0
x''
25:3 DEFINITIONS
As illustrated in Figure 25-2, the logarithmic function is defined through the integral
1
25:3:1 In(s) -dt x>0
us
25:4 THE LOGARITHMIC FUNCTION ln(x) 226
4 4 4 4 4 4 4 4 4 4 4
:....:....:....:....:....:....:....:....:....:....:....:.... ....:..-
4 4 4 4 4
5
.:....:....:........:........:....:....:....:1...:....:....:....:.. 4
25:5 INTRARELATIONSHIPS
The logarithmic function of any argument in the range zero to unity may be related to ln(x) where I < x < x by
the reciprocation formula
227 THE LOGARITHMIC FUNCTION In(x) 25:8
The logarithmic functions of products, quotients and powers may be expressed by the formulas
25:5:2 ln(xv) = ln(x) + ln(y) X>0 y > 0
25:6 EXPANSIONS
The logarithmic function may be expanded as a power series in a variety of ways, of which the following are
representative:
x2 x3 (-x)'
25:6:1
I
25:6:2 ln(x) x_-- + 1 (x - 1)'
+
(x - 1)'
+ _
(x - 1)'
ra-
1
x 27 3.r' jx' 2
X2
-1 (x' - W (x2 - 1)S I r' - 1)7j"
25:6:3 ln()= x + I + + x>0
3(x + 1)' 5(x' + 1)5 i=o 2j + I l
25:9 APPROXIMATIONS
A number of approximations, including
x-I 3 4
25:9:1 ln(x) = 8-bit precision - s x :S -
x 4 3
and
25:9:2 ln(x) (x - I)
(6)3f5 8-bit precision -sx52
1 + 5x 2
8-bit precision x z 31
X j.1 J
valid for large arguments.
I:/
25:10:1
b)
[-In(x)1'
25:10:4 In"(t)dt = (- I)"n!x n = 1.2.3....
fl " J=r) j!
+
In(x) - v * -1
v
f
l1
dt _ li(bx + c)
25:10:6
J-/bin(bt+c) b
t''dr
25:10:7
,0 In(t) - {In(ln(x)) v = - 1
The Ii function in the last two formulas is the logarithmic integral function [see Section 25:131. In addition to
definite integrals that can be evaluated as special cases of the last five formulas, the following are of interest:
229 THE LOGARITHMIC FUNCTION ln(x) 25:12
25:10:8 In(-In(t))dr = -y
J0
ln(r)dt (-mj/12
25:10:9
'- r l -a'/6
25:10:10
( In(t)dr _ (-G
J o t ± t- 1
2 - (ir'/12) - In(4)
25:10:11 In(t)ln(1 t r)dt =
f2 - (a'/6)
Others are given by Gradshteyn and Ryzhik [Section 4.2-4.41. The constants appearing in these expressions are
defined in Section 1:7.
Semidifferentiation and semiintegration, with a lower limit of zero, yield
d'r ln(4bx)
25:10:12
in In(bx)=-
and
d1/` x
25:10:13 T7, ln(bx) = 2 (ln(4bx) - 21
25:12 GENERALIZATIONS
We defer to Section 25:14 a discussion of the generalization of ln(x) "to other bases." Polylogarithms are addressed
in Section 25:13.
25:13 THE LOGARITHMIC FUNCTION ln(x) 230
25:12:1
1 x-1 x}1
j+v x J 2
which has been termed the generalized logarithmic function and is denoted ln,.(x) [see Oldham and Spanier, Section
10.51. This generalized function is also representable as
25:12:2
I
ln,(x) = B v; 0;
x-1
x
in terms of an incomplete beta function [Chapter 58).
The logarithmic function is closely related to its inverse, the exponential function [Chapter 261. and to the inverse
hyperbolic functions [Chapter 311.
The logarithmic integral, defined by
25:13:1 li(x) = f A
In(t) = x In(x)at In(t)
is a related function with importance in number theory. Figure 25-3 illustrates its definition, and its behavior is
portrayed in Figure 37-1. Because of the identity
25:13:2 li(x) = Ei(ln(x)) x>0
the properties of the logarithmic integral follow from those of the exponential integral of Chapter 37. Numerical
values of li(xjmay be generated via the algorithm of that chapter.
231 THE LOGARITHMIC FUNCTION In(s) 25:13
25:13:3 diln(x)
f tl
ln(t)
dt =
In(t)
ft- 1
dt - -
6
wr'
which occurs in radiation theory, among other applications. Its definition is illustrated in Figure 25-4 and its be-
havior in Figure 25-I. The dilogarithm is also expressible as the sum
.
25:13:4 diln(x) = (I- -x)'
, 0:S x:5 2
from which the particular values diln(O) = -4(2) _ -a'/6 and diln(2) = q(2) = w2/12 follow. Here 4 and q
signify zeta and eta numbers [Chapter 31. Confusingly, there are a number of alternative definitions and notations
for the dilogarithm. Series 25:13:4 is rapidly convergent for 1/2 S x < 1 and advantage of this fact is taken in
the algorithm
ti
4 '#4
FIG 25-4
by replacing the infinite upper limit in 25:13:4 by 2 + Int(4/x2). The algorithm calculates dilogarithms of arguments
in the 0 < x < ; range by exploiting the reflection formula
a=
25:13:5 diln(1 - x) = ln(x)ln(1 - x) - diln(x) - 0SxS1
6
25:14 THE LOGARITHMIC FUNCTION In(s) 232
and those in the range x > I are accessed via the reciprocation formula
In= (x)
25:13:6 diln(I/x) = - diln(x) x > 0
2
The algorithm generates values with 24-bit precision (relative error s6 x 10'8) for all positive arguments. The
universal hypergeoretrc algorithm [Section 18:141 may also be used to evaluate the dilogarithm in the range 0 s
xs2.
One may similarly define a trilogarithm
25:13:7 triln(x)=J
r diln(t)
e- I
dt =- (I
- J'x)'
it occurs in distribution theory [Section 27:14] and is also evaluable. for 0 <_ x 2. by the universal hypergeometric
algorithm of Section 18:14. The logarithm, dilogarithm and trilogarithm generalize to the polylogarithm
(1 - s)'
25:13:8 polnjx)
,- J,
where v is not necessarily an integer. The polylogarithm is encountered also in Section 64:12 and, in the alternative
notation -F(I - x,v). some of its properties were reported by Bateman [see Erdilyi, Magnus, Obcrhettinger and
Tricomi, Higher Transcendental Functions, Volume 1, pages 30 and 311.
Logarithms to base 10 are called common logarithms. Briggsian logarithms or decadic logarithms. The sub-
script 10 is often omitted so that log,o(x) is written log(s) or even lg(x). The identities
In(s)
25:14:4 log,o(x) = (0.4342944819) In(s) =
2.302585093
obtain. The decadic logarithm of a number in scientific notation [Section 9:141 is easily found because
25:14:5 1og10(no . n_,n_2n_3 ... X I0'") = N + log,o(no - n_,n..n_3 ...)
The final term in 25:14:5, which necessarily lies in the range .000... through .999..., is called the mantissa of
the logarithm, while N is its characteristic. Chemists use the p notation
25:14:6 px = log1o(1/s) = -1og1o(x)
For example, pH = -log,o(H). where H is the activity (or concentration) of hydrogen ions. The name cologarithm
of x is sometimes applied to px.
CHAPTER
26
THE EXPONENTIAL FUNCTION exp(bx + c)
Many natural and manmade assemblies (populations of bacteria, bank balances, collections of radionuclei, arrays
of lightbulbs) increase or decrease at a rate that is proportional to their size. This characteristic, described by
equation 26:3:4. leads to exponential growth or exponential decay, whose functional dependences are expressed
by exp(bx + c) with b positive for growth and negative for decay.
The function exp(bx + c), or sometimes the special cases exp(±x), are considered in this chapter. The ex-
ponential functions of more complicated arguments are deferred to the next chapter.
26:1 NOTATION
The symbolism e', where e is the base of natural logarithms [Section 1:41, often replaces exp(x). The name "natural
antilogarithm" and the symbol In-'(x) are occasionally encountered. Colloquially, exponential functions are often
called "exponentials."
26:2 BEHAVIOR
The exponential function accepts arguments of either sign and any magnitude, but itself adopts only positive values.
As illustrated in Figure 26-1, exp(x) rapidly increases while exp(-x) decreases towards zero, as the magnitude of
x increases.
The exponential function shares with the unity function the remarkable property f(x) f(-x) = 1.
Figure 26-1 also displays a map of the self-exponential function x`. This exhibits a minimum value of
exp(- I /e) = 0.69220 at x = exp(-1) = 0.36787 - .
26:3 DEFINITIONS
The exponential function is defined as the inverse of the logarithmic function; that is:
26:3:1 f = exp(x) where x = In(f )
Alternatively, the definition as a power of the number e [see Section 1:71
26:3:2 exp(x) = e' = (2.7182818284 .)'
233
26:4 THE EXPONENTIAL FUNCTION cxp(bx + c) 234
may be used, or expansion 26:6:1 may serve instead. The limiting operation
26:3:3
/
exp(x) = liml I + -
x
n
is yet another definition of the exponential function, albeit not a very useful one in practice.
The differential equation
26:3:4 of
dx
= bf
is solved by f = exp(bx + c) where c is an arbitrary constant. Thus, exp(x + c) is the only function (other than
the zero function) that remains unchanged on differentiation.
26:5 INTRARELATIONSHIPS
The remarkably simple reflection, addition, subtraction and involution formulas
235 THE EXPONENTIAL FUNCTION cxp(bx + c) 26:6
26:5:1 exp(-x) _
exp(x)
exp(x)
26:5:3 exp(x - y) =
exp(y)
26:5:4 exp(x) = exp(vx)
partly explain the widespread utility of the exponential function.
Infinite series of exponential functions of negative argument that may be summed geometrically [see equation
1.6:51 or as hyperbolic functions [Chapters 29 and 30] include
exp() 2
26:5:7 ex p(-x) + ex p (-3x) + ex p(-5x) + . = = 1 csch(x) x>0
exp(2x) - 1 2
exp(x) I
26:5:8 exp(-x) - exp( -3x) + exp(-5x) - =
exp(2x) + 1
_
2
sech(x) X>0
Similar sums in which the exponential arguments involve the squares of the natural (or the odd) numbers lead to
theta functions [see Section 27:131
26:6 EXPANSIONS
The expansion
bx + c (bx + c)2 (bx + c)'
26:6:1 exp(bx + c) = I + + + =
21 j1
1-o
is valid for all arguments. Another expansion is in terms of hyperbolic Bessel functions [Chapter 491:
x x x x x
26:6:3 exp(x)1- = - - - - - ---
1
1+ 2- 3+ 2- 5+
and
26:6:4 exp(x)=1- 2+
Ix +-----...
x x x
3- 2+ 5-
X
0 t/e 1 e
26:9 APPROXIMATIONS
Relaxing the n -p x condition in definition 26:3:3 leads to
130
exp(bx + c) - exp(c)
26:10:8 exp(bt + c)dt =
Jo b
n! exp(bx + c)
26:10:9 r" exp(bt + c)dt = [exp(-bx)-a"(-bz)J n=0.1.2....
26:10:10
exp(bt c)dt = -exp(bx+c)dawV b>0
exp(b t + c)
26:10:11 J dt = exp(c) Ei(bx) b>0
=
exp(br + c ) dz _ (nb"
26:10:12 I) exp(bx + c)Lexp(-bx) Ei(bx)
0! 1! (n - 2)!
n=2,3,4, . b>0
bx (bx)r (bx)"-'
the a", daw and Ei functions being explained in Section 26:13, Chapter 42, and Chapter 37. See Table 37.14.1
for a compilation of indefinite integrals of .r"exp(x) for commonly encountered values of v.
Equations 26:10:7-26:10:9 apply also when b is negative, but the general expressions
exp(c)
26:10:13 I; v > -1 b<0
L
I exp(c)
26:10:14 J t"exp(bt+c)dt r(v+ I;-bx) v<---I b<0
26:10:15
26:10:16
1
in terms of incomplete gamma functions [Chapter 42] are then more useful. Special cases include
exp(bt + c)
f
exp(bt + c
I
n
dt = V -b exp(c) erf V -bx
26:10:17
fexP(b: + c)
t"
=
bexp(bx + c)
(n - 1)!
0!
bx
+ - 1!
(bx)Z
+
l
+ exp(-bx) Ei(bx)l n - 2. 3, 4, ... b<0
(bx)"
26:11 THE EXPONENTIAL FUNCTION exp(bx + c) 238
and specific expressions for C r" exp(-t)dt will be found in Table 37.14.1 for a variety of v values.
Some other important indefinite integrals include
dt x a + exp(bx + c)
26:10:18
f a + exp(br + c) a
1
- ab ln I a + exp(c)
and
1 /ezp(bx)
b''Va arctan a>0
r dt
26:10:19
ja exp(bt) + a exp(-b:) I
bV artan h (!Z/ Pa )
1
F
a< 0
Some thirty pages are devoted by Gradshteyn and Ryzhik [Sections 3.3 and 3.4] to definite integrals of ex-
ponential functions. Here we cite only four general examples:
n!
26:10:20 1 t" exp(-xt)dr = exp(-x) e,(x) n = 0. 1, 2.... X>0
n!
26:10:21 J t"exp(-xt)dt = -, [exp(x)e"(-x) - exp(-x)e"(x)] n = 0. 1, 2.... X>0
r exp( -xt) ((n -x )""' 0! 1!
26:10:22 dt = - Ei(-x) + exp(-4 - - Z
J r" - 1)! I x x
+...+(-1)"(n-2)!Jj
n=2,3.4,... x>0
and
26:11:2 exp(ti1r) _ - I
26:11:3 exp =i 2 1 tt
are noteworthy.
26:12 GENERALIZATIONS
The function B. where 0 is a positive constant other than unity or e, is known as the general exponential function
or antilogarithm to base P. It is equivalent to an exponential function with changed argument
26:12:1 W = exp(bx) b = ln((3)
This formula enables the properties of B` to be deduced simply from those of exp(bx + c). The function 10' is
frequently encountered and is the common antilogarithm of x.
Generalization in a different direction is provided by the concept of hypergeometric functions introduced in
Section 18:14. In the nomenclature of that section, exp(x) is the K = 0. L = I hypergeometric function in which
the sole parameter equals unity. Generalizing this parameter to v gives the function C(v) exp(x)y*(v - 1;x) (see
Chapter 45 for the 'y* function], which could therefore claim to be a generalized exponential.
provides a convenient method of evaluating the exponential polynomial if its degree n is not too large. Section
18:14 shows how the universal hypergeometric algorithm may be employed to calculate n!x-e,(x). For large n the
asymptotic approximation
n:z -'
26:13:3 e,_2(x) - exp(x) - n-x
n!(n - x)
is useful.
The self-exponential function x` was mentioned in Section 26:2 and is graphed in Figure 26-1. The related
function x"' displays a maximum value at x - e. The inverse function (Section 0:3] of the latter function is the
limit of the sequence
x""I"I",
26:13:4 x, x' x"`' x""",
corresponding to infinitely repeated exponentiation. Figure 26-2 maps these interesting functions.
Those Basset functions [see Chapter 51] K,(x), in which v is an odd multiple of i, are related in a very
straightforward way to the exponential function exp(-x). Thus, we have
and further members of the sequence may be constructed via the recurrence formula
2v - 2
26:13:8 K,(x) = K. -2(x) + K,._,
x
Maps of the functions K,(x) for v = , and are displayed in Figure 26.3. The symbol and the name
modified spherical Bessel function of the third kind are sometimes applied to the function a/2x K.. ,,2(x).
241 THE EXPONENTIAL FUNCTION exp(bx + c) 26:14
fficuh fkile
Final solution Laplace Solution to the transformed
involving the problem involving
inversion
variable r the variable s
In this section we present some general rules for Laplace transformation and inversion; the classic book by Churchill
may be consulted for further information. A short table of transform pairs is also included.
In the following we use If and g to represent functions of t whose Laplace transforms ft, and IL are functions
of the variable s. which is sometimes called the dummy variable. Equation 26:10:25 shows how fL(s) is related to
f(t), but it is often more economical to use the operator notation
to represent this relationship. In most applications t is a real variable whereas s is often complex. Here, however.
we shall not pursue the complex nature of the dummy variable.
It should be recognized that not all functions have Laplace transforms. To be transformable, a function f(t)
must at least be defined as a real function over the entire range 0 s t s -. Any function that is finite throughout
this semiinfinite range has a Laplace transform, although fL(s) is not necessarily a named function. Moreover, many
functions that encounter infinities in the 0 c t s - range may nevertheless be transformed. For example, an infinity
at t = 0 will not impede Laplace transformation provided that Of(f) approaches zero as r tends to zero from positive
values.
Later in this section a table [Table 26.14. 11 of Laplace transforms is presented. Its utility may be greatly
enhanced by making use of certain general properties that obtain under Laplace transformation. Some of these
general rules are explored in the next few paragraphs.
The weighted sum or difference of two (or more) functions may be Laplace transformed by using the linearity
property
26:14:2 L{ctf(t) ± c;g(:)} = c,fL(s) ± c.gL(s)
Provided that both series converge, this rule may be extended to the infinite summation Ec,f,(t), which thereby
gives Eci f; .(s) on transformation.
The scaling property
is generally less useful than the corresponding rule [equation 26:14:2\3] for Laplace inversion. Of course. if f is
a function (such as sin or exp) for which an argument-addition formula [Section 0:51 exists, transformation of
f(bt + c) may be accomplished by that route.
26:14 THE EXPONENTIAL FUNCTION exp(bx + c) 242
With a > 0 and u representing the unit-step function [Chapter 81. the function u(t - a)f(t - a) may be obtained
by nullifying the t < 0 portion of f(t) and then translating the residue along the t-axis as portrayed in Figure 26-
4. Such a function may be Laplace transformed by the delay property
26:14:5 L{u(t - a)f(t - a)} = exp(-as)fL(s) a>0
For certain simple functions g(t), the composite function f{g(t)} may be Laplace transformed as the definite
integral
where G(v) is a bivariate function, some examples of which are shown in Table 26.14.1. The most general chain
rule for Laplace transformation is
Table 26.14.1
g(r) G(sx)
1 (2\ sr)
exP(_
4r /
V exp(-r)
exp(r)- I
I'(I+s)
sinh(r) J,(r)
243 THE EXPONENTIAL FUNCTION exp(bx + c) 26:14
iterate to
which corresponds to differintegration [Section 0:10] with respect to -s. the lower limit being -x.
If f(t) is the derivative of a function F(t), then L{f(t)} = sFL(s) - F(0) and if f(t) is an indefinite integral of
ti(t), then L{f(t)} = [IL(s) + f(0)J/s, Both these rules of Laplace transformation are special cases of the general
differintegral transformation property [see Oldham and Spanier, Section 8.11
i
If
26:14:15 Lif (t) } = s fc(s) - Z s'
( l
' tit, (0) J = -Int(-v)
This rule applies for all is, but the summation is empty if v s 0.
The periodicity property permits the Laplace transform of a periodic function [Chapter 361 to be replaced by
an integral over a single period
I2 - I2 cothl Ps 1
I
/ `26:14:16
Many important periodic functions are symmetrical in the sense that translation by a half-period changes the sign:
that is, if ger(t) is such a symmetrical periodic function of period Q, qer[t + (Q/2)J = - Ber(t). In such cases
Laplace transformation may be accomplished by the formula
in which integration is over a half-period. The half- and fully rectified analogs [Section 36:131 of such a symmetrical
periodic function have transforms equal to [Z - coth(Qs/4)] gerL(s) and coth(Qs/4) gerL(s). respectively, where
gerL(s) is given by 26:14:17.
The rules above, in conjunction with Table 26.14.2, permit a very large number of Laplace transforms to be
26:14 THE EXPONENTIAL FUNCTION exp(bx - c) 244
Table 26.14.2
Chapter Chapter
or or
section section
no fm no
10 r8( t)
1:13 p(c:h:t - a)
2c hsj
exp(-as)sinh1 `1 28
i
steepening
8 £u4t - Wit) j - 0. 1. 2.... staircase
- {(sl I
s
cs+b
7 bt + c I I
s'
-1 cs cs I (b Euler's
7 cxp b) Ei . 37
b b / cs rs, function
10:12 bs +c 7
b
26 7
bs,c
8 u(t - a) a >0 - exp(-asl 26
Staircases
8 2£u(r -(j + )'7r-) j = 0. 1. 2.... with 840:s) 27:13
9
1 + 2!u)r - j:w')
c - bt
Int(bt)
j = 1, 2. 3....
b>0
staircase
lengthening
treads
2b
s
-c
exp(-)
b Mhr s)
+ cs - b
g,(Os)
b
-
-
s'
1J
27:13
26
30
2s 12b.
r 1 ( ) 1
9 fret sawtooth + I - coth Ps 30
P/ 2Ps s 2 J
9 Int(bt - 1) staircase
?t csch(-s) 29
Chapter Chapter
or or
section section
23 n = 0. 1. 2. 11
12 12
Vr
12 13
12 - + -2 16 ex - b)
41
12
b1+ bsexp(b) erfcl bJ 41
1°C cE
12
exp(Cs) erfcl 41
1 b1+c b. `. b
12 7 (1 - Wt - b)j ,J-erf(VbJ) 40
Vr YY J
12 R erfclV bsl 40
Rb
26 12
bs+c
41.42 exp(-bf) erf(V -br) or daw(V br)
bg0 12
VR
41 a cxp(drl erfc(a V 1) 12
V Rt
41 exp(a-r) erfc(aVr) 12
V,I VJ +a)
40 erf(Vbf)
1b
s s-b
12
13 rl o /s, 13
rl vl
13 lt-aY-'u(t-a) v>0 exp(-as) 26
b""f' /
13 v>-1 r(I + 45
bt\\+ c b. \ b
cs\
13 b'P-'01 +cr-' v>0<It r(v)c""-'u(v;v + µ; 48
26:14 THE EXPONENTIAL FUNCTION exp(bx + c) 246
Chapter Chapter
or or
section section
no. lr) no.
13
1bsl bs - c
V2it
Ibr n
46 13
4 (V 2s + \/ y
211-u(r- a))
14 Was) - (das) 57
WNa' -r
15 a >0 ho(sVal - 57
IS Vr - a a>0
Va +
I 1c
1!5 exp(as)erfc(Vas)
- 41
2 v
u(r - b)
K.(bs (
15
V :-b 51
Vs2 - a
a§0 15
Val(Vs= a s)
50. 53 I,uV a) or 3,AVa) Ir s'+a a§0 15
L J
-bi) sinlN/A- v-
32 2 expl Sac-h'=A>0 16
\\` 2a \, 2a as'-bs+c
V
28. 32 sinhuV-ai or sin(rVa) ar0 16
J"+a
28. 32 cosh(rV -a) or cos(tV o) a S 0 16
u0 - b) - u(r - 3b)
16:13 W bs) 49
V-r=+4br-3b=
247 THE EXPONENTIAL FUNCTION exp(bx + c) 26:14
Chapter Chapter
or or
section section
no. no.
\ abt v __1 a
49. 52 exP or f 4ac - b-' A 3: 0 16:13
2a Jo t 2a as'+bs+c
17-24 Tat' a, = b/%' b, - Jta, 17-24
s 2b`
r(1 + v)
52 t'hd) v> -1
b1
s... exp 4s / 1'(4s
b
23:14
r \J 1I
25 In(bt + c) s I In(c) - a%P( b/ Et( bs)J
37
ct
37 In(c) - E - In(bs + c) 25
b
Ci(bt) - 1Mb)
2(
38 25
S
27 eexp(-,:'It) K,.(2af1 51
R2 + 20 o' a
27 (20"exp-af) a>0 r"' exp 8s- D.;,_; J
v > -1 46
\ 3s
I _ba.`
27 b"t" 'expl 2 I v> 0 r(v)exp(4b) D "\b) 46
(bl
50. 53 `[I,(2Vbo or J.(2V_-W)1 v> -1 exp\s/ b 3: 0 27
c
41 exp(br + c) erfc(V bt + exp 27
2Vb-t
46
f xP(Q El.
s
(2s)", CXP(- V'2-b;) 27
I
27:13 sech(\/-S) sinh(Vs - 2vVs) 28. 29
Vs
Chapter Chapter
or or
section section
no. f(r) no.
/bt y..(b)
29 21x-br)"exPl 2) csch(b) n0.I.2... 44:12
22
29 1 G(s
sech(br)
2b
+ bl
l 2b
44:13
fll+v) I 1 + v: s+h)
29 (2hr1' csch(br) v>0 b t 2b
64
tanh(br)
l G12 I_1
30 44:13
2b b s
ar 1
31 2 arstnh(br) [h.(!)
b.``
-Y,(s)J
lb. J 57
32 -
y'r
sinr
l2rl
l or
2r) s
expl- 32
r(
0 * v > -I
arccot(-J)
I /6 \ I i-
55 bc02V br) or bei(2Vbr1 cos - or-sin
I g 32
S s S S
32:13
I
b sinc
)-J
w
1
= I sin(bi)
s
arccat - I = arcten
b
b
s.
- 35
35 2 arctan(V)
rl
exp\6/ e1fc1 bl
1 41
5
249 THE EXPONENTIAL FUNCTION exp(bx + c) 26:14
Chapter Chapter
or or
section section
no. f(r) US) no.
35 arctan(br)
cos(S
I- Si(-) MsCirbl
u/b) 38
IlL l1 /J
I
57:13 srC,lbt) 2 35
V7 \b' s/
s)
br s'
42 2 daw(2 b
Ei(b' ) 37
56 2 }}
r(3/ (2)"'Jexpr \b/ r( 3:
l r l 45
f(I
47 r MIa;2c br) c > -1 + c) FI a, I + c:2c: b1 I 60
50
r(c)'vbr"-"'
1._ ,(2V'
rbr)
f(a) I b\
-s"M ax; - 47
00' $
s
48 t'"'exp(-Br)U(c - ba + a - 6:811 B''" F a,b:c: 60
B
2b sinc(vtb - s)'-'tb + s)-"
48:4
k
B( I - v; I + v:
2b
49 s K(-) 61
49 bin(br)l,lbr) sl I - II E(26) 6I
LLL
50
r
2
51 KdV 'b,) a b 1f
exp(8s/ K.(±) 51
51 K,;.."(br) -I S v t 0 a csc(va)P.(b) $9
YYYar
determined. The table lists pairs f(t), fL(s) of functions and their Laplace transforms. To use this table for trans-
formation, first locate the chapter or section of the Atlas in which the f function appears- Then scan the numerical
index that constitutes the first column of the table to find entries containing the f function. Of course, there exist
many Laplace transform pairs that are not listed in Table 26.14.2 and that cannot be conveniently deduced via
rules 26:14:2-26:14:18. We recommend the comprehensive tabulation by Roberts and Kaufman as a source of
several thousand varied transform pairs.
A useful check on the accuracy of an f(t), fL(s) pair is provided by their limiting forms as the arguments
approach zero and infinity. These limiting properties are
26:14:18 L {lim{f(t)}} = lim{sfL(s)} and L f lim{f(r)}} = lim{sir(s)}
,-.o ,-u
An even simpler check is provided by the requirement that the inverse Laplace transform of any continuous function
f(t) must tend to zero as s -. X.
Section 18:14 demonstrates that the majority of the functions in this Atlas may be expressed as hypergeometric
functions. Let f(t) be such a function and let us adopt (a,_x), as an abbreviation for the product of K Pochhammer
polynomials
(a )j(w),(a3); ... (ax), (a;-.k); P
26:14:19 f(t) -
=o (c1);(c.) (c3); ... (CL), ,_o (ci-L),
Then fL(s) is also a hypergeometric function, namely that given by expression 18:14:7. Moreover, many functions
closely related to f(t) may also be Laplace transformed to hypergeometric functions with one or more extra nu-
meratorial parameters. Thus:
r(1 + v)
26:14:20 L{t"f(br)} =
S
i
,=u
(1 + v) (ai_x)
'
(cl.L)j
' -v > -1
s
(b)'
and
` s'
'
of Laplace inversion incorporates the scaling property,, which`` is the c instance of 26:14:23.
The most general chain rule for Laplace inversion gives the inverse transform of the composite function fL{g(s)}
as the definite integral of the product F(ru)f(u), where f(t) and F(ru) are respectively the inverse transforms of
fL(s) and exp{-ug(s)}. That is:
251 (THE EXPONENTIAL FUNCTION exp(bx + c) 26:14
26:14:27
//bsr
`111
I
cothl -2) gi(s) = Li g(t) + 21 u(t - jb)g(r - jb) }
l /t
J = intl -)
\b
J.1 1
To Laplace invert the product s"gL(s) involves operations of the calculus applied to the inverse transform. The
simplest cases
l (r l
26:14:28 k(s) = L j g(t)dt
s J0
and
d
26:14:29 sgc(s) = L (t) +9(0)8(1)
ddtt
iterate to
( l
26:14:30
AL(s)
= L{ r ... J0g(t)(dt)" } n = 2. 3.4....
lJ -' ; JJJ
and
Here 8'`(t) is the n" derivative of the Dirac delta function (Section 10:131 at t = 0. away from the immediate
vicinity of t = 0 all such functions are zero. Ignoring the contributions from these delta derivatives, the inverse
Laplace transform of S' L(s) is the differintegral d'g/dt' for all values of v.
A number of useful rules exist for the Laplace inversion of the quotients ft{g(s)}/g(s) and fL{g(s)}/h(s) where
g(s) and h(s) are such simple functions as f, I/s, s- . o-. etc. Inversion usually leads to definite or indefinite
Table 26.14.3
g(s) F(ru)
In(s)
rim)
exp
Q
AiIIII
s (3r'')"' \(3:t'
26:14 THE EXPONENTIAL FUNCTION cxp(bx + c) 252
integrals of the product of f(t), the inverse transform of 0s), with other functions. Examples include
26:14:32
fL\s/
= Li
(
cj cos(2 tu)f(u)du }
l
fL( s'+a)_ ( 1
26:14:33 LS i Jo( ar - aue)f(u)du i a?0
s+a = l 1
with Io( au2 - ate) replacing Jo( at - au2) if a is negative; Roberts and Kaufman [pages 171-1741 may be
consulted for others.
If fL(s) is the hypergeometric function
I j (a)/(a2), ... (aK)/ I / 1 (a .r );
26:14:34 k(s) _
s ,-o (e1,(c., ... (CL), s s (c,_L)/ (S')
then inverse transformation of fL(s), and a number of related functions, give hypergeometric functions of t. The
results
/t e t` (a1~x),(±b't /4)'
26:14:36 S. fLl bs I = L C(1 + v) + V),( V\' v > -1
J-0 (I
Many of the results of this chapter follow by combining the properties of the functions discussed in Chapters 12,
13 and 26. It is appropriate to devote a chapter to the functions exp(-ax"), however, because they are of such
widespread importance, particularly when a is positive. For example, the temperature dependence of many physical
properties obeys this functionality with v = - 1. Random events often involve the v = 2 instance and lead to
important distributions, as discussed in Section 27:14.
27:1 NOTATION
No special notation or symbolism need be considered beyond that addressed in Sections 12:1 and 26:1.
27:2 BEHAVIOR
As with the functions discussed in Chapter 13, the range of exp(-ax") depends in a rather detailed way on the
characteristics of the number v. In this chapter, however, attention will be confined to the range x >- 0. except
when v is an integer.
Figures 27-1 and 27-2 are maps of the functions exp(x') and exp(-.r") for assorted values of v. The important
graph of exp(-x2) versus x, known as a Gauss curve, displays a maximum value of unity at x = 0 and points of
inflection [see Section 0:71 at x = ±1/V2. Similarly, exp(-1/x) has an inflection at x = #.
27:3 DEFINITIONS
With x replaced by -ax', any of the definitions in Section 26:3 can serve to define exp(-ax').
0 0 ti v
iX10
O R A
+NO
'r0' p 4ti b+pN +p 00 11. CO
i 4 4 i. 4 4 ,4 4 4 4 4
4~
4 4
O 0 Co
if if
a ti
i4
a
06 O.0
4f if if
tiCj
4f
ti ti
a 4b
if if
v
. . . . .
1.0
27:5 INTRARELATIONSHIPS
The relationships of Section 26:5 hold when x is replaced by -ax".
Summation formulas for series of exponentials of I/x include
/ 4 9) az ) rtx - 1
= 2 93(0;x) - 2 = + \ exp(_ r2x)
X1
\ \ /
and ///
255 EXPONENTIALS OF POWERS exp(-ax') 27:9
27:5:3 ex
P( -
xl) + ex - +
p\ x /
e.p(-
p(x5)
-+ _ -4 B. (0;-
4/ 4
- 2exp 4=x/
X>0
in addition to variants of equations 26:5:5-26:5:12. The B functions are explained in Section 27:13. The final
approximations improve as x increases but are valid to better than I part in 109 for x z 1.
27:6 EXPANSIONS
27:6:2
is rapidly convergent.
exp(V)=1+
2V
2 - + 2+
--
x/3 x/15 x/35 x/(4j' - 1)
2+ 2+ 2+
X = 0 x = t x = x
exp(x')
Jx v < 0` v<0
I v>0 {x1 v>0
exp(-x')
0 v<0 1 v<Ol
{ 1 v>0 {0 v>0
27:9 APPROXIMATIONS
Though crude, the approximation
27:9:1 exp(-x')-
I-
0
- -
IrI ;!: V-1r
is never in error by more than 0.09. This triangular approximation to the Gauss curve has an area of V, equal
to that under the curve.
27:10 EXPONENTIALS OF POWERS exp(-(xx') 256
27:10:1 exp(-ax")
Indefinite integration of the general expressions cxp(-at') and t" exp(-at') can be accomplished by making the
substitution -at' _y and utilizing the general formulas contained in equations 26:10:7. 26:10:13 and 26:10:14.
Below are listed some important indefinite integrals:
27:10:5
f exp(-dt
t
= x exp(1
x
Ei11f
x
x>0
27:10:7
f exp(-V)dt= 2(f + 1) exp(-Vx)
27:10:8
f exp(:'
71
dt=(x-V)exp(-:1
N/_X -Ei1Vzl
x>0
27:10:9
t
1
I exp(-) dt=xexpl-)+Ei -I 1
x///
1
x/
x>0
27:10:10
J! /B2- 4ay
6 exp(-at' - At - y)dt = 2
1
expl erfc l
tax +
a >0
J 4a 2Va
Chapters 42, 37 and 40 are devoted to the daw, Ei and eerfe functions.
Some important definite integrals include
ll
v J
-(D+
27:10:11 j:exP(_w)dt = 1> 0 a> 0
and
and
27:11:2 exp(
/ -a
I = expl
-ax \
x2 I
/ ay
cost = 2 ) + i sin
y1 J
\x + iy/ \ +Y x +y (Fa+? )
27:12 GENERALIZATIONS
The function exp(-ax") may be generalized "to other bases" as described in Section 26:12 so that evaluation of
0'-°"' is possible.
I
27:13:4 94(v;x) exp[-(v +'/2 +j)2/xl = 93(v
but the reader should be alert to the wide variety of symbolisms in use in the literature. Thus, the quantity defined
in equation 27:13:1 would be denoted 9,(nv,exp(-n2x)) by Abramowitz and Stegun [Sections 16.27-16.30J.
It is evident that these four functions are dependent periodically on v so that 9(v + 2;x) _ -9(v + I:x) _
0(v:x) for 9, and 92, whereas 9(v + ';X) = 9(v;x) for 93 and 94. This periodicity is brought out more clearly in
the alternative representations
1
27:13:11 93(v;4x) = 03 -:x /I + \ 2' x 1
941
2 2
27:14 EXPONENTIALS OF POWERS cxp(-ax') 258
1 I-- 1 1 1- 1
27:13:12 9.(v;4x) = x) + x)
2 e3 2 4 2 4
that may be rephrased in several alternative ways.
The series in 27:13:1-27:13:4 converge so rapidly if x K 1/ir that it is rarely necessary to use Iii > 3 to
compute very accurate values of the theta functions. On the other hand, if x ? I /-rr. the series in 27:13:5-27:13:8
converge even more rapidly. This means that accurate computation of theta functions is always a simple matter,
making these functions useful in a variety of applications. Their utility is aided by the widespread occurrence of
theta functions in Laplace transformation [see Section 26:141. As examples
r csch(Vs)
27:13:13 93(v;q exp(-sr)dt = cosh[(2v - 1 0
1
27:13:14 93(0;s) = J I I+2 u(t - j'a2)J exp(-st)dr s>0
s o L=i
but for a more comprehensive listing see Roberts and Kaufman, in which the notation for theta functions is almost
identical with that adopted in this Atlas.
The identity of expressions 27:13:5-27:13:8 with definitions 27:13:1-27:13:4 permits a reformulation of series
of exponential functions, as follows:
I (2j 41)'n'x
27:13:15 I 1 + 21 (-1)' cxp\( ll = 92(0;x) = 27-
exp-
x>0
L x/J J=o (
2
1
Theta functions with v = 0 are important in their own right. Several of the representations coalesce when x
leading to the particular values
27:14:1 µ = J jt f(t)dt
f 27:14
Jf
27:14:2 v= = j (I - p.)'f(t)dt = -12 + t2f(t)dt ? 0
The limits, xo to x,, on the integrals demarcate the range that is accessible to x, often -x to x or 0 to x. The
mean of a distribution provides a measure of the average value of the property x, while the variance describes the
dispersion of the distribution about the mean. If the distribution has too great a dispersion, no finite variance exists.
The Lorentz distribution [see Table 27.14.11 is a case in point: for this particular distribution, depicted in Figure
27-3 for a = 8, the integral in 27:14:2 diverges. Even integral 27:14:1 diverges for the Lorentz distribution, but
the so-called Cauchy principal value of the integral, namely
c
g. g xa
(z3
:....: 0.8
....: X. ....:........... 0.8
:....:....: 0.4
FIG 27-3
is as tabled.
Accompanying each distribution is a cumulative function or distribution function, given by
where, by definition of f(x), F(x,) = 1. The value of F(x) necessarily lies in the range 0 s F(x) s 1 and expresses
the fraction of the measurements that (for N -. x) will lie in the interval between ro and x. Statisticians speak of
percentiles or percentage points: the p'" percentile is the value x,1100 of x such that
The fiftieth percentile .r1j2 is also called the median of the distribution and satisfies the condition
27:14:6 f(t)dt =
f f(t)dt = 2
Though with symmetrical distributions they are the same, the mean and the median are not, in general, identical,
nor does either necessarily correspond to the peak that often exists in a graph of f(x) versus x. Where such a peak
exists, its s-coordinate is known as the mode or most probable value of the distribution.
Distributions occur widely, especially in statistics and physics. Examples from both of these fields have been
assembled into Table 27.14.1, and a selection is displayed graphically. The archetypal distribution is the normal
distribution shown in Figure 27-4 for a = 3. The Gauss distribution is the p. = 0 instance of this. Notwithstanding
27:14 EXPONENTIALS OF POWERS exp(-axe) 260
Table 27.14.1
Llmm Dense/ luncuon Mph Yarulwc Cunwlwn< fuacu0n
D01nMron ,n In al n+r 0 Fn
U. l*m
nnanpularl
an c,l
Gws.
2v
- dl\¢al
.1.,110 Gems!
`!scap{- I.r-411 . (. Cl17\ain-µ1I
Raylcl r
010 tar cap - an' I I -cap -.5 1
circular nonnel 0 a.
0 n= I - ecpl
1 spa.. - yl
tapl-0p - µll II - c.M-=1 - rlll
Bcllxmann 1¢
010= o tap-0r! 1 - eap-nn
e%p0ntnltall
v tMl - Bl
n0 dan,J -Br - IMB' cap,,')
Fenru Dlrac o lnl a - B,
IM1 - BXB - enpvnl 01a1 01 In. I 'Bl
nhwI -0.
-v0 dlln! l -BI - Inkapvnl - 01
Burt-Eimmn Inl1 -Bl
MI I - B) n lafl - B1
r In! i - BI
ae w 010 =
,rln/21
tw(-v
0.- r101,
µ
v0
B01a 0.-
Blv.B!
Sntdaa s+
w>11 ow.
a1rU.
all1.0agh
IfeQel
a rc s µ - Cawk *0 LMt\ ul, - on
-C .. by, pnwipal 2 A
valve
fkll.
la CwMll
-a10% -0) µ
1..
In-II' f
S[ 12//=11!l!'
4, 6 6. 2\rn-2l"`
I 1 _
na i°-1.: -aacuw,/\ nl
SIu m,-r la - II!! ]
-sa:
I1
1 0
n=3.5.7... r\iln - 21!' 121 f'
121 W
261 EXPONENTIALS OF POWERS exp(-ax") 27:14
,, 6 ,'L6 ,6 x6 xti6 x
4 4 ig 4q 44 ig.t
iqt
:....:....:...0.4
0. 2
rL6 , x6 x10
It
q'} 4
:....:....:....:..............:....:....:.1.0
.:....:....:....:.0.6
0. 2
: FIG 27-5 :
:.:....:.............. :....:....:-.. 0
,6 6
it 4 : ig it
......... ......:....:....:.0.e
.:....\ .. Z ....: ....................0.6
FIG 27-6
:....:....: ...................0.2
r .f Cx) .
27:14 EXPONENTIALS OF POWERS exp(-ar") 262
its name, the normal distribution is encountered in practice rather seldom. The asymmetrical log-normal distri-
bution, in contrast, finds more frequent application. An example of this latter distribution in which µ = 2v is
illustrated in Figure 27-5. Even less symmetrical is the Boltzmann distribution depicted in Figure 27-6: such a
distribution describes, for example, the variation of the numbers of molecules with height in the earth's atmosphere.
The shape of the Weibull distribution is a very strong function of a (which must be positive) and this distribution
therefore finds widespread empirical application in failure analysis.
CHAPTER
28
THE HYPERBOLIC SINE sinh(x)
AND COSINE cosh(x) FUNCTIONS
This chapter and the next two chapters address the six so-called hyperbolic functions. The present chapter deals
with the two most important of the six: the hyperbolic sine and the hyperbolic cosine. These two functions are
interrelated by
28:0:1 cosh''(x) - sinh'(x) = I
and by each being the derivative of the other [see equations 28: 10:1 and 28:10:2].
28:1 NOTATION
The names of these functions arise because of their complex algebraic relationship [see Section 32:111 to the sine
and cosine functions. Their association with the hyperbola is explained in Section 28:3.
The notations sh(x) and ch(x) sometimes replace sinh(x) and cosh(x). Although they cause confusion, the sym-
bolisms Sin(x) and Cos(x) are occasionally encountered.
28:2 BEHAVIOR
Both functions are defined for all arguments but, whereas the hyperbolic sine adopts all values, the hyperbolic
cosine is restricted in range to cosh(x) ? 1. Figure 28-1 shows the behavior of the functions for rather small
arguments. For arguments of large absolute magnitude, both functions tend exponentially towards infinite values.
28:3 DEFINITIONS
The hyperbolic sine and cosine functions are defined in terms of the exponential function of Chapter 26 by
exp(x) - exp(-x)
28:3:1 sinh(x) = 2
exp(x) + exp(-x)
28:3:2 cosh(x) = 2
263
28:4 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 264
ti 440 44
To provide a geometric definition of the hyperbolic functions, consider the positive branch of the rectangular
hyperbola [Section 15:41 x - 1 depicted in Figure 28-2. The green area is bounded by the hyperbola and by
a pair of straight lines OP and OP' through the origin with slopes that are equal in magnitude but opposite in sign.
Let a denote this shaded area: it can take values between zero (corresponding to points P and P' coinciding with
A) and infinity (corresponding to lines OP and OP' having slopes of +I and -1 and constituting the asymptotes
of the hyperbola). The lengths PQ and OQ may then be regarded as functions of a and are, in fact, the hyperbolic
sine and cosine of a
28:3:3 PQ = sinh(a) = sinh (green area)
28:3:4 OQ = cosh(a) = cosh (green area)
The second-order differential equation
d2f
28:3:5 = b2x
dx2
has the general solution f = cl sinh(bx) + c2 cosh(bx), where b, cl and c2 are constants.
28:5 INTRARELATIONSHIPS
The hyperbolic cosine function is even
28:5:1 cosh(-x) = cosh(x)
265 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 28:5
generalize to
and
where the T. and U. Chebyshev polynomials are discussed in Chapter 22, as are the Chebyshev coefficients
De Moivre's theorem
28:5:9 cosh(nx) ± sinh(nx) _ [cosh(x) ± sinh(x)]" = exp(±nx)
is also useful.
Equations 28:5:3 and 28:5:4 may be regarded as special cases of the argument-addition formulas
28:5:10 cosh(x ± y) = cosh(x) cosh(y) ± sinh(x) sinh(y)
28:5:11 sinh(x ± y) = sinh(x) cosh(y) ± cosh(x) sinh(y)
x2-1
28:5 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 266
and
cosh(2) - I
28:5:13 sinhl 2 I = sgn(x)
and
cosh(2x) - 1
28:5:15 sinh'(x) =
2
n=2,4,6,...
and
1
28:5:23 sinh(x) cosh(y) = sinh(x + y) + 2 sinh(x - y)
2
1 1
28:5:24 cosh(x) cosh(y) = - cosh(x + y) + - cosh(x - y)
2 2
28:6 EXPANSIONS
The hyperbolic sine and cosine functions may be expanded as infinite series
x3
x5 x2/*r
28:6:1 sinh(x)=x+-+-+
3! 5! ,=o (2j + 1)!
x2 x4 x2j
28:6:2 cosh(x) = 1 + - + + _ i -
2! -
4! /-o (2j)!
or as infinite products
28:6:3 sinh(x) = x I + _)
x2
zr
//
I
\\
1+ Z)
x2
4n
=xnI+
;_1
x2
jn
22
sinh(x)
I e2
0
e-1
2e 2e
I+ e'
coah(x) 1
2e 2e
28:9 APPROXIMATIONS
The hyperbolic sine and cosine may be approximated by polynomials; for example,
x3
28:9:1 sinh(x) = x + 8-bit precision jxj < 0.84
6
z
/
28:9:2 cosh(x) = 1 I+ 4) 8-bit precision jxl < 0.70
exp2lxi)
28 : 9 : 4 cosh(x) = 8 -bi t prec i s i on l xi > 2 . 78
cosh(bx) - I
28:10:3 I sinh(bt)dt = b
0
sinh(bx)
28:10:4 . cosh(bt)dr =
Jo b
The general formulas
-id:
n!!
cosh2'(xl n=1. 3. 5 , ...
" _o (2j)!!
28:10:5 cosh (t)dt = w12 _ r
o (n - I)u rr (2j)!!
Lx + sinh(x) cosh'-'"(x) n = 2, 4, 6, .. .
n, .o (2j + 1)!!
r"-1W2
E (- Iy (2j(2')r
- 1)!r
(-1)u-1)/2(n
- 1)!! cosh(x)
sinh2'(x) n = 1, 3, 5, ...
n!!
nn .r
r"(2j)!!
.
28:10:6 sink"(t)dt = '_o
(-1)"'2(n - 1)!! 1
[x+cosh(x)
;_o (2j + 1)!!
(-1)'sinh2'+'(x)n!!
n = 2, 4, 6, ...
permit the indefinite integration of integer powers of the hyperbolic sine and cosine functions. Alternative expres-
sions may be derived by integration of equations 28:5:16 and 28:5:17. Noninteger powers are treated in Section
58:14. Other important classes of indefinite integral include
n!sinh(bx)
b"'
-+
(bx)"
M!
(b.)" -2
(n - 2)!)
+...+1J
1
+- n!
n=1.3,5,...
a!cosh(bx) 1(bx)" + (bx)" 2
b0+1 +.+1
n! (n - 2)! ]
- n! cosh(bx) r (bx)"-' (bx)"-,
+ ... i. bx
j;77- L (n - I)! + (n - 3)!
n!
28:10:8 Jt" sinh(bt)dr = - b-+, n = 2, 4, 6. ...
0
- 1)!
+
(bx)°''
(n - 3)!
n=1,3,5,I(
...
269 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 28:13
The bracketed series in the above integrals may be expressed as [e (bx) ± e.(-bx)J/2, where the e function is
discussed in Section 26:13. Similar indefinite integrals for is = -1, -2, -3, ... are listed by Gradshteyn and
Ryzhik [Section 2.4751; they involve the chi and shi functions defined in Chapter 38.
A large number of indefinite and definite integrals involving the hyperbolic sine and cosine functions exist.
The reader is referred to Chapters 2.4 and 3.5 of Gradshteyn and Ryzhik.
28:12 GENERALIZATIONS
The Jacobian elliptic functions nc(x;p) and nd(x;p) may be regarded as generalizations of cosh(x), to which they
reduce when p = 1. Likewise, sc(x;p) and sd(x;p) reduce to sinh(x) when p = I and therefore generalize the
hyperbolic sine. See Chapter 63 for all these Jacobian elliptic functions.
) 1 + csch(x) Icoth(x)I
28:13:2 cosh(x) = '1+ sinh2(s) =
sech(x) Icsch(x)l 1 - tanh2(x) Vcoth2(x) -1
relate the hyperbolic sine and cosine to the other hyperbolic functions (see Chapters 29 and 301.
The hyperbolic sine and cosine functions are closely related to those hyperbolic Bessel functions I.(x) in which
v is an odd multiple of ± . Examples are
2
28:13:4 1_112(x) = cosh(x)
Trx
sinh(x)1
28:13:5 I,,,(x) = I cosh(x) -
V
VVVVTrxL X
These functions, some of which are graphed in Figure 28-3. share the properties of all hyperbolic Bessel functions
as discussed in Chapter 50. The name modified spherical Besse! function and the symbol i,(x) is sometimes given
to the function V/Tr/2r I,,,i,:(x).
28:13 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 270
28:14:2 bL = sinh(bh)
271 THE HYPERBOLIC SINE sinh(x) AND COSINE cosh(x) FUNCTIONS 28:14
titer
..............:..............1.p
FIG 28-4
CHAPTER
29
THE HYPERBOLIC SECANT sech(x)
AND COSECANT csch(x) FUNCTIONS
Of the six hyperbolic functions, the two treated in this chapter are perhaps the least frequently encountered. The
property
29:0:1 (1 - sech'(x))(t + csch2(x)) =
interrelates the two. Two features of this chapter-sec Sections 29:3 and 29:8-deal with all six hyperbolic func-
tions.
29:1 NOTATION
The notation cosech(x) is sometimes used for the hyperbolic cosecant. Some authors admit only four hyperbolic
functions, using 1/cosh(x) and 1/sinh(x) to represent the secant and cosecant.
29:2 BEHAVIOR
Figure 29-1 shows the behavior of the two functions. Both approach zero as their arguments tend to ±x. Both
functions accept any argument but, whereas csch(.r) adopts all values, the hyperbolic secant is restricted in range
to 0 : sech(x) s 1.
29:3 DEFINITIONS
The relationships
2
29:3:1 sech(x) _
exp(x) - exp(-.x) cosh(x)
and
2
29:3:2 csch(x) = _
exp(x) - exp(-x) sinh(x)
are the usual definitions of the hyperbolic secant and cosecant.
273
29:3 THE HYPERBOLIC SECANT sech(x) AND COSECANT csch(x) FUNCTIONS 274
O h
41 if if if if i?
cech (x>
....:.. 1.5
sech(x)
:.........:..............:....:.-0.5
/ X
/ each (x)
of
FIG 29-2
Figure 29-2 depicts three similar right-angled triangles. If the sides that are drawn as dashed lines are of unit
length, then each of the other six sides equals one of the hyperbolic functions as labeled. Pythagorean and similarity
properties thus enable the interrelationship between any two hyperbolic functions to be deduced and may serve as
the definition of any one function in terms of any other. Note that the argument x is shown in the diagram, but it
is not related to the triangles by any simple construction. It is, in fact, related to the angles in the triangles via the
gudcrmannian function (see Section 33:14).
The differential equations
d
29:3:3 + oz-_j'-0
dx
are satisfied respectively by f = a csch(ax) and f = a sech(ax).
275 THE HYPERBOLIC SECANT sech(x) AND COSECANT csch(x) FUNCTIONS 29:6
29:5 INTRARELATIONSHIPS
sech(x) csch(x)
29:5:4 csch(2x) = 2
Other relationships may be derived via the equations of Section 28:5. but these are generally more complicated
and less useful than are the intrarclationships of the hyperbolic sine or cosine.
29:6 EXPANSIONS
and
29:6:2 csch(x)= I1
x
-+ --
x
6
7x2
360
31x5
15120
+ I
x
+
2 = TI(2j)
x
x=
a'
' (2-4)
2
(2j).
Byx2 ' -a<x<a
The 0 and t coefficients are defined in Chapter 3 and the Euler E and Bernoulli B numbers in Chapters 5 and 4.
respectively.
Expansions as exponentials take the forms
and
x= -x x = -1 x= 0 x- l x= z
- 2e 2e
csch(xl 0 0
e' - 1
2e 2e
sech(x( 0 0
r' + I e.+1
29:9 APPROXIMATIONS
Indefinite integration of the square roots of the hyperbolic secant and cosecant functions generates special cases of
the incomplete elliptic integral [see Chapter 62] in which the parameter equals I /V2.
r / 1
I csch(x) - I
29:10:8 csch(t) dt = F V2: d 1 cos((b) =
csch(x) + I
Equations 28:11:1 and 28:11:2 may be employed to evaluate csch(x + iv) and sech(x + iv), respectively. For purely
imaginary argument
29:11:1 sech(iv) = sec(v)
29:11:2 csch(iy) = -icsc(y)
where sec and esc are the functions to which Chapter 33 is devoted.
29:12 GENERALIZATIONS
The Jacobian elliptic functions cn(x;p) and dn(.x:p). discussed in Chapter 63. are generalizations of sech(x). while
cs(xp) and ds(x;p) similarly generalize csch(x).
29:13 THE HYPERBOLIC SECANT sech(x) AND COSECANT csch(x) FUNCTIONS 278
1 I )csch(x)( coth'(x) - 1
29:13:1 sech(x) _ _ _ = VI - tanh'(x) _
till - sinh'(r cosh(x) I + csch'(x) Icoth(x)I
relate the hyperbolic secant and cosecant to the other hyperbolic functions (Chapters 28 and 30).
CHAPTER
30
THE HYPERBOLIC TANGENT tanh(x)
AND COTANGENT coth(x) FUNCTIONS
The two functions of this chapter are the reciprocals of each other
30:0:1 tanh(.r) coth(x) = I
and are closely related to the other hyperbolic functions [Chapters 28 and 291.
30:1 NOTATION
The symbolism th(x) sometimes replaces tanh(x); cth(x) or ctnh(x) sometimes replaces coth(x). The misleading
notations Tan(x) and Cot(x) are occasionally encountered.
30:2 BEHAVIOR
These functions are defined for all values of the argument x, but both are restricted in range. The hyperbolic tangent
takes values only in the range - l to + I, whereas coth(x) assumes all values except those between - I and + 1.
As shown in Figure 30-1, both functions lie exclusively in the first and third quadrants [see Section 0:21, and
both approach + I as x -. x and - I as x -i -x. The diagram also shows the coth(x) - I /x function, which also
approaches ± I at its limits, and which is cited in Section 30:13.
30:3 DEFINITIONS
The hyperbolic tangent and cotangent may be defined in terms of the exponential function
exp(2x) - 1 _ 1 - exp(-2r) _ exp(:r) - exp(-.r)
30:3:1 tanh(x) =
exp(2.0 + 1 l + exp(-2r) exp(.x) + exp(-.r)
30:3:2
co(h(x) exp(2r) + I
- I + exp(-20 exp(x) + exp(-x)
exp(2r) - I I - exp(-20 exp(x) - exp(-x)
or in terms of the other hyperbolic functions
279
30:4 THE HYPERBOLIC TANGENT tanh(x) AND COTANGENT coth(x) FUNCTIONS 280
sinh(x) sech(x) I
30:3:3 tanh(x)
cosh(x) csch(x) coth(x)
cosh(x) csch(x) I
30:3:4 coth(x)
sinh(x) sech(x) tanh(x)
Figure 28-2 may be used to provide a geometric definition of the hyperbolic tangent, namely
30:3:5 RA = tanh(a) = tanh (shaded area)
The hyperbolic tangent and cotangent each satisfy the differential equation
30:3:6
df-_ 1 -f _
dx
30:5 INTRARELATIONSHIPS
Both functions are odd:
30:5:1 f(-x) _ -f(x) f = tanh or cosh
The duplication formulas
2 tanh(x) _ 2
30:5:2 tanh(2x) =
T+ tanh2(x) tanh(x) + coth(x)
281 THE HYPERBOLIC TANGENT tanh(x) AND COTANGENT coth(x) FUNCTIONS 30:6
and
1 ± coth(x) coth(y)
30:5:5 coth(x ± y) =
coth(x) ± coth(y)
The equations in this paragraph may be used to build formulas for tanh(3x), coth(4x), etc.
The half-argument formulas
I- 1 - tanh2(x)
30:5:6
tanh(x- )
and
(x I+ 1 - tanh2(x)
30:5:7 cothl - I = coth(x) + sgn(x) coth2(x) - I = = coth(x) + csch(x)
tanh(x)
are useful.
The sums and differences of the functions of this chapter may be expressed in terms of hyperbolic sines and
cosines:
sinh(x ± y)
30:5:8 tanh(x) ± tanh(y) =
cosh(x) cosh(y)
cosh(x ± y)
30:5:9 coth(x) ± tanh(y) =
sinh(x) cosh(y)
sinh(x -!- y)
30:5:10 coth(x) ± coth(y) =
sinh(x) sinh(y)
30:6 EXPANSIONS
The functions tanh(x)/x and x coth(x) may be expanded as power series in x2:
x' 2x5 17x' -2 -4x2 Y _ I m (41 - l)Bu(4x2)' a
30:6:1
3 15 315 x
X(2!') 2 x 1 (2j)!
--<x<-
2
n
2
30:6:2
1 x x3 2x5 1 2 ( x21
I =- By(4x2)' -ar<x<ar
x 3 45 945 x x J.1 \ ar / x ;_o (2j)!
where the lambda and zeta functions are discussed in Chapter 3 and the Bernoulli numbers in Chapter 4.
The hyperbolic tangent and cotangent may alternatively be expanded as series of exponentials. The tanh(x)
expression
tanh(x) =
x x' x' x'
30:6:7
1+3+5+7+
holds for the hyperbolic tangent.
0 x= 9 x - I X -
_1
1- e' 1- e e- 1 e' - 1
tanh(x) I
0
I+e' l+e a+1 e'+1
1
The hyperbolic tangent and cotangent functions may be evaluated via formulas 30:3: I and 30:3:2 or by the algorithm
in Section 29:8.
30:9 APPROXIMATIONS
1 x'
30:9:2 coth(x) = X 1+ 8-bit precision Ixj 5 0.65
3
are valid. For large arguments, one may use
30:9:3 tanh(x) = sgn(x)[I - 2 exp(-2Ixj)j 8-bit precision 1.6
30:9:4 coth(x) = sgn(x)[1 + 2 exp(-2kl)j 8-bit precision [xI z 1.6
283 THE HYPERBOLIC TANGENT tanh(x) AND COTANGENT coth(x) FUNCTIONS 30:11
r tanh(bx)
30:10:8
0 b
30:10:11 rsf(x) =
d 112
I+tanh
(-I ''//;
(-1) VJexp(jx)
rn V X- AX + 2r)
X
(d (x + x )] 1/2 2 =-
where X = (21 - 1)2a- + x2. It is related to Lerch's function [Section 64:121 by the identity rsf{In(.r)} = r(D(-.r:
-1/2;1).
30:12 GENERALIZATIONS
The Jacobian elliptic functions sn(xp) and ns(xp) ]see Chapter 63] may be regarded as generalizations of tanh(x)
and coth(x), respectively. Asp - 1, sn(x;p) -> tanh(x) and ns(x;p) -> coth(x).
and
relate the tangent and cotangent to other members of the hyperbolic family. Figure 29-2 is useful in expressing
these relationships.
The function coth(x) - (1/x), which occurs in the theory of dielectrics, is known as the Langevin function. It
is mapped in Figure 30-1 and its integral is given in 30:10:5. The Langevin function can be expanded via 30:6:6
and its reciprocal as
3 -
30:13:3
1
=-+2x 2
1
where r,(1) denotes the j ° positive root of the equation tan(y) _ y 1see Section 34:7].
CHAPTER
31
THE INVERSE HYPERBOLIC FUNCTIONS
The six functions of this chapter are interrelated by the following permutations of argument:
1 1 z
31:0:1 arsinh(x) = sgn(x) arcosh( 1 + x2) = sgn(x) arsech = aresc i - = artanh
l -+X1 x + x2
l + x2
= arcothl
x
(1 ( I 1/x2 - 1
31:0:2 arcosh(x) = arsinh( x2 - ) = arsechl xJ =
- areschl xZ x
) =artanh)
x
= arcoth x >_ 1
x2- 1
X
x2 I = artanh( 1 - x')
WI
7,-
I
=arcoth 05xs I
x2
x1/
31:0:4 aresch(x) = arsinh - 1 = sgn(x) arcosh( 1 + 1/x2) = sgn(x) arscch( x2/(1 + x2))
1-x2 1
285
31:1 THE INVERSE HYPERBOLIC FUNCTIONS 286
31:0:6 arcoth(x) = sgn(x) arsinh + . ) = sgn(x) arcosh( x2/(x2 - 1)) = sgn(x) arsech( 1 - 1/x2)
x-- 1
I `
= sgn(x) arcssh( x= 1) = artanh I Gxi > 1
x/
31:1 NOTATION
The prefix `ar" means "area," and its pertinence can be appreciated by reference to Figure 28-2. The inverse
hyperbolic sine of x, for example, denotes that area which. in the Figure 28-2 construct, has a hyperbolic sine of
a. That is:
31:1:1 arsinh(x) = a = shaded area if x = sinh(a) = PQ
The symbolisms arcsinh(x), agsinh(x), arsh(x) and sinh-'(x) are all used in place of arsinh(x). Corresponding
variants are encountered for the other inverse hyperbolic functions. The same symbolisms with a capitalized initial
letter-Arcsinh(x), Arsh(x), Sinh-'(x), etc.-are sometimes used synonymously with arsinh(x). but more often
they denote the multiple-valued functions that we discuss in Section 31:12.
31:2 BEHAVIOR
The behaviors of the six inverse hyperbolic functions are shown in Figure 31-1. Notice that all functions exist in
the first and third quadrants (see Section 0:2) only.
The inverse hyperbolic sine has a behavior that is simpler than the other five. It accepts all arguments and
adopts the sign of its argument.
The inverse hyperbolic cosine is normally defined only for arguments x ? I although some authors extend its
domain of definition to W a I via arcosh(-x) = arcosh(x). It adopts only positive values.
The inverse hyperbolic secant is likewise normally defined only for 0 s x s 1 although its definition may be
extended to 0 s i s I by means of arsech(-x) = arsech(x). The values of arsech(x) are always positive.
The inverse hyperbolic cosecant has two branches, as mapped in Figure 31-1. It adopts the sign of its argument
and exhibits a discontinuity at x = 0.
The inverse hyperbolic tangent is defined only for -1 5 x 5 1 and approaches ±x as x -+ ± 1.
The inverse hyperbolic cotangent has two branches. For I s x <_ x, arcoth(x) is positive; for -x s x s -1,
arcoth(x) is negative. The function is not defined in the -I < x < 1 gap.
31:3 DEFINITIONS
Indefinite integrals define each of the six inverse hyperbolic functions. Small diagrams, Figures 31-2 through 31-
7, illustrate these definitions.
dt
31:3:1 arsinh(x) =
f 1 -+r
dr
31:3:2 arcosh(x) = t2 - 1 x >- I
J
ri
dr
31:3:3 arsech(x) = 1 0sx51
/l + t2
dt
-=t 1+t2
287 THE INVERSE HYPERBOLIC FUNCTIONS 31:3
O h O h
4 *` i, A'
y yo
h O
4f
0
4f
O
+
4
l,P I,
31:3 THE INVERSE HYPERBOLIC FUNCTIONS 288
FIG 31-3
cosh (x)
aresch(z)
FIG 31-4
31:3:5 artanh(x) = -I s x s 1
J
< x c x
r=- 1
31:3:6 arcoth(x) _
dt x5X5 - i
t'
Another approach to inverse hyperbolic functions is to use definitions as the logarithms of algebraic functions
of the argument x. Thus:
289 THE INVERSE HYPERBOLIC FUNCTIONS 31:5
31:3:7
31:3:8
arsinh(x) = ln(x +
arcosh(x) = hi(x +
_
x2
x2 - 1)
+l)
x>I
( rI
31:3 : 11 artanh(x) = In -1 - X - 1
. V 1-+ xx )
/ x+1\
31:3:12 arcoth(x) = lnI -- ) W?I
VX
The definition from which the names of these functions arise must be treated with caution. The arsinh, arcsch,
artanh, and arcoth functions are straightforward inverses of the corresponding hyperbolic functions
31:3:13 f = arsinh(x) where x = sinh(f)
31:3:14 f = aresch(x) where x = csch(f)
31:3:15 f = artanh(x) where x = tanh(f)
31:3:16 f = arcoth(x) where x = coth(f)
For the arcosh and arsech functions, however, a proviso is needed:
31:3:17 f = arcosh(x) where x = cosh(f) and f >- 0
31:3:18 f = arsech(x) where x = sech(f) and f ? 0
31:5 INTRARELATIONSHIPS
No reflection formulas exist for the arcosh or arsech functions [although see Section 31:2], but the other four inverse
hyperbolic functions are odd:
31:5:1 f(-x) = -f(x) f = arsinh, arcsch, artanh, arcoth
A number of function-addition formulas exist for the inverse hyperbolic functions. These include
31:5:2 arsinh(x) ± arsinh(y) = arsinh(x 1 + y2 ± y vi -+",)
31:5:3 arsinh(x) ± arcosh(y) = arsinh(xy ± (x2 + 1)(y2 - 1))
±y
31:5:5 artanh(x) ± artanh(y) = artanh x
I±xy/I
31:5:6 artanh(x) ± arcoth(y) = artanh 1
xy
(.X ±y
31:5:7 arcoth(x) ± arcoth(y) = arcoth(1 ± q)
x±y
31:6 THE INVERSE HYPERBOLIC FUNCTIONS 290
and many others may be constructed using the equivalences listed in formulas 31:0:1-31:0:7. Use of these formulas
requires, of course, that each argument lie within the domain of its function.
Special cases of the relationships of the last paragraph may be converted into such formulas as
( VI+x2-1
31:5:8 arsinh(x) = 2 sgn(x) arsinhl
\ 2
31:6 EXPANSIONS
Series expansions exist for the inverse hyperbolic sine and tangent
Replacing x by 1/x and restricting I > 1, these same series depict arcsch(x) and arcoth(x), respectively. We also
have
/2l x2 3x4 5x6 (2j - I)" x2j
31:6:3
4 16 96 ,_. (2j)!! 2j
0<x<l
Replacing x by 1/x and restricting x to exceed unity converts this series into an expression for ln(2x) - arcosh(x).
The similar series with alternating signs is
3 + 5 (2j - I)!'
31:6:4 sgn(x) arsinh(x) - 1n(2x) W L, I
4x2 16x' 96x6 ., 2j(2j)!!(-x2)'
The inverse hyperbolic sine 'anted tangent may each be expressed as continued fractions
arsinh(V x) I I x 2x I x 2x 3 x 4x 3 x 4x 5 x 6x
31:6:5
I +x I+ 3+ 5+ 7+ 9+ 11+
and
Table 31.7.1
x= -x X = -I x=0 x = 1 x =
aresch(x) o In(\ + 1) o
The algorithm below, which is exact, uses definitions 31:3:7-31:3:12 to calculate values of the inverse hyperbolic
functions. The universal hypergeometric algorithm [Section 18:14) may also be used to evaluate these functions
for suitable arguments.
31:9 APPROXIMATIONS
31:9:3 artanh(x)
x
8-bit precision - 2I < x S ` I
is useful.
31:10:1 arsinh(x) =
& 1+x
d 1
d
_ 0Sx<-I
- arsech(x) X 1 x''
d -1
- arcsch(x) _
dx xI V l + x'
d
- artanh(x) _ -1 x 1
dx 1 -x
d
31:10:6 arcoth(x) = I LrI a I
dx 1 - x'
The corresponding indefinite integrals are
+I
31:10:7
.
f arsinh(t)dt = x arsinh(x) -
31:10:8 Jaosh(tz x arcoch(x) - x I
31:10:12
J
` arcoth(t)dt = x arcoth(x) + In l r27
\\l
I <x<x
The semiderivative
31:10:13 , artanh(V ) =
2 I-x
links the inverse hyperbolic tangent to a simple algebraic function.
and
293 THE INVERSE HYPERBOLIC FUNCTIONS 31:13
2-x2- I (1+x2+y2)2-4y2
31:11:4 Y = y + sgn(xy)
2
The resulting explicit expressions in terms of x and y are rather complicated. However, several useful special cases
of such formulas arise upon selecting k = 0 and by restricting the argument to be purely imaginary. This produces
31:11:5 arsinh(iy) = i aresin(y)
31:11:6 arcsch(iy) _ -i arccsc(v)
31:11:7 artanh(iy) = i arctan(v)
31:11:8 arcoth(iv) = - i arccot(y)
where the functions appearing on the right-hand sides are inverse trigonometric functions [see Chapter 35].
31:12 GENERALIZATIONS
The inverse hyperbolic tangent is a special case of the generalized logarithmic function [Section 25:12]
sgX)
31:12:1 artanh(x) = 1 1 1 x-
while the inverse hyperbolic sine is an instance of the Gauss function [Chapter 601
The generalization of arsinh to the multivalued Arsinh function is discussed in Section 31:11. The multivalued
Arcosh, Arsech, Arcsch, Artanh and Arcoth functions may be constructed analogously.
The functions of this chapter are of paramount importance: they are the units from which all periodic functions
[Chapter 361 can be built. The sine and cosine functions are interrelated by
32:0:1 sin2(x) + COs-I(x) = 1
and by
32:1 NOTATION
The symbolism sin(x) and cos(x) is universal. The notations Sin(x) and Cos(x) refer to the hyperbolic functions of
Chapter 28, not to those presently under consideration. Sometimes the names circular sine and circular cosine are
employed to emphasize the distinction from the corresponding hyperbolic functions.
In this Atlas we generally use x to represent the argument of a function, and this convention is retained in the
present chapter. As explained in Section 32:2, however, the arguments of the sine and cosine functions [and of
the functions addressed in Chapters 33 and 34] are often regarded as angles, rather than simply as numbers to
which no special geometric significance is to be attached. We shall write sin(O) and cos(O) [as well as sec(O), tan(d),
etc. in Chapters 33 and 34] when we particularly want to emphasize the angular interpretation of the argument.
The sine and cosine functions, and any "mixture" of them c, sin(x) + c2 cos(x), are known as sinusoidal
functions or sinusoids.. Via equatjon 32:5:26, such a "mixture" may be expressed as c; -+c:, sin(x + (b) where
the angle 4, is known as the pharb of the sinusoid andr +VT
Z as its amplitude.
Collectively, the functions of Chapters 32, 33 and 34 are often known as the trigonometric functions [see
Section 34:14].
32:2 BEHAVIOR
The sinusoids are periodic functions with period equal to 2a, that is, their values at x ± 2w, x ± 4n, x ± 6w, .. .
exactly equal their values at argument x. This is evident from Figure 32-1.
295
32:2 THE SINE sin(x) AND COSINE cos(x) FUNCTIONS 296
In each period the sinusoidal functions display two zeros, one maximum and one minimum. The sine function
has maxima of sin(x) = I at x = a/2. -3w/2, 5,r/2. -7tr/2, ... and minima of sin(x) = -1 at x = -a/2,
3,r/2, -5w/2, 7w/2, ..., a zero being located midway between each adjacent maximum-minimum pair (i.e., at
x = 0. ±w, ±2ar, ...). The behavior of the cosine function is similar. cos(x) = I maxima occur at x = 0, ±2w,
±4n, ...; cos(x) = -I minima at x = ±w, ±3w, ±5w, ...; and zeros at x = ±ir/2, ±3tr/2, ±5w/2.....
The periodicity of the sine and cosine functions can be appreciated perhaps more easily when the argument is
regarded as an angle. Since the angle 0 is coincident with the }nglcs 0 ± 2a, 0 ± 4w, 0 ± trrr, ..., it comes as
no surprise that
32:2:1 sin(o ± 2zr) = sin(0 ± 4ir) = sin(6 ± 6w) _ = sin(0)
and similarly for the cosine [and, in fact, for the other four circular functions discussed in Chapters 33 and 341.
Angles in the ranges
It
32:2:2 2ka<0<-+2kw k=0, ±1, w2, ...
2
are said to be in the first quadrant. Similarly the second, third and fourth quadrants encompass the angles
32:2:3 (ir/2) + 2ktr < 0 < w + 2k second quadrant k = 0, ±1, ±2, .. .
32:2:4 it + 2kw < 0 < (37r/2) + 2kw third quadrant k = 0, ± 1, ±2, .. .
and
32:2:5 (3w/2) + 2ku < 0 < 2(k + ow fourth quadrant k = 0, ±1, ±2, ...
This usage of the team `quadrant' is related to, but differs from, that described in Section 0:2. Some simple rules
respecting the signs of the trigonometric functions in the four quadrants are assembled in Section 33:2.
297 THE SINE sintx) AND COSINE cos(x) FUNCTIONS 32:3
32:3 DEFINITIONS
The trigonometric definitions of the sine and cosine function are illustrated in Figure 32-2. Consider the point P
to have arrived at its present location by having moved from A along the circular path AP such that the distance
from point 0 has remained constant and equal to unity. Then the sine and cosine functions are defined as the lengths
32:3:1 sin(x) = PQ [negative if P lies below line 0A1
32:3:2 cos(x) = OQ [negative if Q lies to the left of 01
PQO being a right angle. The argument x may be interpreted in three distinct ways: first, as the length of the arc
AP, this being assigned a negative value if the rotation of OP was clockwise; second, as the angle POA. measured
in radians, this again being considered negative if P rotated clockwise to arrive at its present location; and third,
as the area (shaded in Figure 32-21 enclosed by the arc AP, the line OP and a mirror-symmetrical construct below
the OA line. It is this third interpretation of x that parallels the one used in Section 28:2 as a geometric definition
of the hyperbolic sine and cosine.
Expansions 32:6:1 and 32:6:2 may serve as definitions of the cosine and sine functions. Alternatively, these
functions may be defined via the formulas
exp(ix) - exp(-ix)
32:3:3 sin(x) _
2i
exp(ix) - exp(-ix)
32:3:4 cos(x) = 2
and
is solved by the general sinusoid f = -(c/b) + c, sin(xV b) + c2 cos(xV b), where c, and c2 are arbitrary
constants. Similarly, for suitable values of the coefficients, the differential equation
32:5 INTRARELATIONSIIIPS
The cosine function is even
32:5:1 cos(-x) = cos(x)
whereas the sine is an odd function
32:5:2 sin(-x) _ -sin(x)
The duplication and triplication formulas
32:5:3 cos(2x) = cos2(x) - sin2(x) = 2 cos2(x) - I = 1 - 2 sin 2(X)
generalize to
and
.-I
32:5:8 sin(nx) = sin(x)U.-,(coa(x)) = csc(x) 7, It,'_'2 coa(x) - !'1 co0x)1
J-o
where T. and U. are Chebyshev polynomials discussed in Chapter 22, as are the Chebyshcv coefficients t '.
Equations 32:5:3 and 32:5:4 are special cases of the argument-addition formulas
32:5:9 cos(x ± y) = cos(x) cos(y) + sin(x) sin(y)
for the cosine and sine of half argument. as well as the formulas
I - cost 2x)
32:5:15 cos'(.r) =
2
1 - cos(2rl
32:5:16 sin'(x) =
2
I (n)
Ij
T_ 71
(n)
i
cos((n - 2j)xJ
n=2.N+I=1.3.5.
32:5:17 cos"(x) =
" Y
0 1 /
cos((n - 21,x1 =
In - 1)!: _ I
V-1
cos((n - 2j)xl
n!! 2"
n= 2N=2_,4.6.
and
32:5:18
-
sin"(.r) = { (- 1)'.v
-0
(-1)'( nj Icos[(n-2j41=
in - I)!'
n
n=2N- I = 1.3.5.
1)-v
(n) - 21
2"-
-u
for positive integer powers of the cosine and sine functions.
The cosine and sine functions satisfy the function-addition formulas
`ry y
32:5:20 COs(y) - cos(y) = 2 cos h ) cos( 2)
y\ kr v
32:5:21 cos(x) - cos(y) _ -2 sin (X I sin )
2 (x
by which any sinusoid may be expressed as a sine or as a cosine. The arctan and arccot functions occurring in
32:5:26 are discussed in Chapter 35.
Infinite series of the forms
or 2 2
32:5:28 s, sin(x) + s2 sin(2r) + s3 sin(3x) + _ I s; sin(jx)
i-j
or sometimes of the combination zc0 + I c, cos(jx) + s, sin(jx), with the c, and s, being specified coefficients,
are termed Fourier series and are discussed in Chapter 36. Here we merely quote two examples with relevance to
the present chapter. If v is any number other than an integer, then
32:5:29 Cos(vx) _ -
2v
sin(wrt)
fI +
cos(x) - cos(jx) + cos(3x) - 1
J
2v2 I-v2 9-
rI (-1),cos(jx)1
=
it 2v r, j 2 - v2
2 sin(x) 2 sin(2x) sin(3x)
32:5:30 sm(vx) sm(vtr) +3
IT 1-v2 4-i? 9-v2
_2 (-1)2j sin(jx)
sin(wn) r
-,n < x < n
-)f
The infinite series
32:5:31
s(Fr) n=1,3,5,... and n=2,4,6....
i-r J* i=I
may be summed in terms of the Hurwitz function (1 - n;x/21r) [see equation 64:6:21. The sum E cos(jx)/j equals
ln{csc(x/2)/2}, while T sin(jx)/j2 is the integral of this function, known as Clausen's integral [see Abramowitz
and Stegun, Section 27.8].
32:6 EXPANSIONS
Taylor series [see equation 0:5:1] exist for the cosine and sine functions and for their logarithms. Those for the
functions
X2 x` x6 (-x2))
32:6:1 cas(t) = 1 - - + - - - + _
2! 4! 6! 1-0 (2j)!
301 THE SINE sin)x) AND COSINE cosCxt FUNCTIONS 32:8
32:6:2 sin(x)=.r--+---
3! 5! 7! ,=0 (2j+1)'
converge for all .x, but more rapidly for small arguments. The logarithmic series have limited domains of conver-
gence:
X 'to
X-
32:6:3 ln(cos(r))=
2
- 12 45
IT IT
21(21)!
-<x<-
2 2
\\\\\\
}
where the r values are the zeros of f(x). Yet another expansion as an infinite product is
sin(x) r r 11 r
32:6:8 = 0.4-!) cos( 4 cos(8 ... = IJ cos( r )
x // /
The cosine and sine functions may be expressed as infinite sums of Bessel functions [Chapter 52]
Table 32.7.1
e o :IP Ir X22' :0 *3e :is` :Sr +6o' -67t' n' '73° t9o
6 3 5
f
12 to 8 5 4 to B 12 2
l+l
f S+VS
a
2+
2
_
2
+\5
B \r
t \r
9 i
\r-V2
2
IIVS
e
VS-I
VM
6
, .vi vi
f.\.
21 3. \15 +1 mt \r xt a Vi zt ±t
6
t B 5 6 \ + 8 2 B
S1
devices often offer the user a choice of "radian mode" or `degree mode" when evaluating trigonometric or inverse
trigonometric functions; be aware that, because of rounding errors inherent in the internal routines of the device.
the "degree thode" is usually the more precise of the two modes. (For example. sin(1440°) will be given as zero.
whereas sin(8ar) will often fail to be calculated as zero.)
The universal hypergeometric algorithm [Section 18:141 permits cos(x) and sin(x)/x to be evaluated for any x.
32:9 APPROXIMATIONS
For small values of the argument, the approximations
'
32:9:1 sin(x) - x I - - 8-bit precision x S 1.1
2X2
and
/ z`` I
32:9:2 coa(x) c 1 I- 3 1 8-bit precision -0.9:% :S 0.9
are useful. \ //
I - cos(bx)
32:10:4 sin(bt)dt =
0 b
sinItx)
cos(bt)dt -
32:10.5
303 THE SINE sin(.r) AND COSINE cos(-v) FUNCTIONS 32:10
arcosh(a) -
a I LLL \\\
V a' - 1
dt I bx a
32:10:8
f I+ a sin(bt) - b
[tan( 2+
4J
1 11
2 /a tan(bx/2)`
arctanl ) - arcsin(a)
bV l \ a'
va-cot(-))
2 (a ' l bx
fl dtb
artanh al
bV'a'-I -I 2
2
-
ran a = 1
32:10:9 -Ib l
;1 -a br
arctan l tan , -I a<I
bVI -a' )t' I +a 2
Indefinite integrals of the sine or cosine of the quadratic function ar' + br + c may be expressed in terms of
Fresnel integrals (Chapter 391
-sin
b'-4uc
ICI
ear-b ) a>0
a
0 ,
r cos(bt)dt 2n
32:10:13 = C( )
J
With Si and Ci denoting the sine and cosine integrals [Chapter 381, we also have
sin(bt)
32:10:14 dr = Si(bx)
o t
and
32:10:15
f
J
(¢ oos(bt)
dt = -Ci(bx)
r
The technique exemplified in 34:10:19 is useful for evaluating many integrals of sine and cosine functions.
32:10:16 fds) =
o
Jo
¢ f(t) exp(ist)dt
With a lower limit of -, differintegration to order v generally causes the addition of va/2 to the argument
b'f(l
d'
32:10:19
ld(x + 00)1'
f(bx) =
\ bx + m
2 ///
v > -1 f = sin, cos
and multiplication by the va power of the argument multiplier. Equations 32:10:1-32:10:3 are instances of this
rule; another example is
d-'n 1 arl sin(bx) - oos(bx)
32:10:20 sin(bx) _ sin( bx - b>0
[d(x + 4/
With a lower limit of zero, semidifferentiation and semiintegration generate auxiliary Fresnel integrals (see Chapter
391
d1/2
32:10:21 sta(bs) = V b sin( bx + 4 I - V Tb tms(V bx)
d 1/2
32:10:22 cos(bx) _ + cos bx + 4/ - V '2b Fres(\)
dTll art
305 THE SINE sin(.o) AND COSINE cos(x) FUNCTIONS 32:12
32:10:23
dx
sin(bx) sins bx - -l
4!
+ Fres(Ni' br)
\/h b
d I (2
a
cos(bx) =
yb cos(bx - 4) - Grestibx)
32:10:24
\b
That the functions I = cos(0.x). sin(x). cos(x). sin(2r), cos(2x). sin(3x).... form an orthogonal family [see
Section 21:14] on the interval 0 :s x :s 27r (or alternatively on -a <- x s n) is established by the integrals
(0 m*n
32:10:26 1 * cos(nt) cos(mt)dt = 2a m=n=0 n, m = 0, 1, 2, .. .
o n m=n+0
32:10:27 1 sin(nt) sin(mt)dt = JO m nl n, m = I, 2, 3,...
1"
In m=n
On the interval 0 s x i- it. the functions cos(Ox), cos(.x), cos(2r), - or the functions sin(x). sin(2x), sin(3x).
are orthogonal. but the conjoined families are not.
32:12 GENERALIZATIONS
As explained in Chapter 36, any periodic function can be regarded as a generalized sinusoid.
The Jacobian elliptic functions sntp;.r) and sd(p:x) generalize the sine function, to which they reduce when p
is zero:
32:12:1 sn(O:x) = sd(O:x) = sin(x)
Likewise. because
32:12:2 cn(0:.r) = cd(0:x) = cos(.x)
the functions cnl p;x) and cd(p;x) are generalizations of the cosine function. The Jacobian elliptic functions are the
subject of Chapter 63. Finally, the sine function is a special case of the incomplete elliptic integral of the second
kind [Chapter 62]:
32:12:3 sin(x) = E(x;l)
32:13 THE SINE sin(x) AND COSINE cos(x) FUNCTIONS 306
I a3 csc'(x) - I a4 a_ cot(x)
32:13:2 cos(x) = 94V I - sin (x)
sec(x) csc(x) V I + tan2(x) cot2(x) + I
where the multipliers a2, a, and a, take the values - I or - I according to the quadrant (sec Section 33:2] in which
x (interpreted as an angle) lies. All multipliers are positive in the first quadrant. or, also in the second, a3 in the
first and third, (r. in the first and fourth. The explicit formulas
32:13:3 a2 = (-1)'° '/"'
32:13:4 a3 = 1)'"'12r/.r
32:13:5 a4 = (-1)W42,+II)/2.I =
ata .
permit the multipliers to be calculated for any x value. Figure 33-2 provides a geometric interpretation of the
interrelationships among the circular function.
The versine function vers(x), coversine function covers(x), and haversine function hav(x), defined by
32:13:6 vers(x) = I - cos(x)
32:13:7 covers(x) = I - sin(x)
and
are rather archaic functions seldom encountered nowadays. The function sinc(x), sometimes known as the samp
function, is important in spectral theory. It is usually defined by
sin(rrx)
32:13:9 sinc(x) =
ax
although the definition sin(x)/x is also encountered. Figure 32-1 includes a graph of sinc(x), and equations 32:6:2,
32:6:4. 32:6:5, 32:6:7 and 32:6:8 are readily adapted to provide information on this function.
Bessel functions [Chapter 53] and Neumann functions [Chapter 54] of orders equal to one-half of a (positive
or negative) odd integer are related in a very simple way to sines and cosines. The simplest cases are
32:13:14 W"(x) _
(n
+ 2 ).r
2j)' r-I lI = Int;-a
For example, W1(x) = 1, V1(x) = 2, W2(x) = 1 - 3x-2 and V2(x) = 6. Of course, all the general properties of
the Bessel and Neumann functions [Chapter 53 and 541 apply to these spherical functions, the first few of which
are mapped in Figure 32-3. The symbols and y"(x) and the names spherical Bessel function of the first kind
and spherical Bessel function of the second kind, respectively, are sometimes applied to the composite functions
Vir/2x and The significance of the adjective "spherical" will be evident from Section
59:14.
. . . . . . . . . . . . . . .
.. 0.7
<x>
32-3
:....: f._:....:.\ ,n
1` ..................... FIG .:....... 0.6
:....:....:.... ...0.4
Y, (x). .
:....... 0.3
which converts a function f of t into a function F of s and is known as the exponential Fourier transform. In
general, f and F may be complex, but t and s are usually real. Properties of the transform and examples for specific
32:14 THE SINE sin(s) AND COSINE tos(s) FUNCTIONS 308
f(t) functions are tabulated [for example, by Beyer, Handbook of Mathematical Sciences, page 597, who, however,
uses a definition slightly different from 32:14:1]. the function f(t) may be regenerated by the inversion formula
[i, 2 k
32:14:4 1, = - r sinl
we used to transform these data into a set of real and imaginary components of the variable F. The subscript k is
intimately related to the frequency variable s: in fact, R. + i!, = Ft is the value of F sampled at s = k/T. The
highest k value that is accessible corresponds to the Nyquist frequency N/2T; see Hamming for further discussions
on this topic.
Equations 32:14:3 and 32:14:4 are the fundamental formulas used for numerical Fourier transformation. When
N is large, these equations call for vast computational resources, and to alleviate this demand the fast fourier
transform has been developed. This invention Ifor full details, Section 2.3.2 of Stoer and Bulirsch. or the original
work of Cooley and Tukey, may be consulted) takes advantage of the symmetry properties of the sine and cosine
functions to abbreviate the volume of arithmetic. Such abbreviation is especially pronounced when N, the number
of data pairs, is a power of 2 (e.g., N = 27 = 128 or N = 210 = 1024) and this is the condition incorporated into
the algorithm below, which performs fast fourier transformation.
The algorithm first requires the input of the integer p. equal to log2(N). followed by the sampled data ro. io,
r,, is, r2. ---. iN-2, rN-,, iN- ,. The ultimate output is in the order Ro, 10, R,, !,, R2, !N-2, RN- 1N-1.
The two portions of the algorithm shown in green are virtually identical and could therefore be programmed
as a subroutine. They are used to set register M equal to the bit-reversed complement of the integer K. This means
that the bit pattern of the number K written in binary notation (e.g.. 0101001 if K = 41 and p = 7) is reversed in
M (e.g., 1001010 = 74 = M).
Even though the algorithm makes provision for imaginary components of the sampled input function, many
applications of Fourier transformation utilize real data only. Then, of course, one enters the values io = is = i2 =
= iu-, = 0 into the fast fourier transform algorithm. In such applications, only the first half of the output has
useful information content because the second half merely duplicates the first in magnitude: RN-k = Rk,1N-k = -1k.
I = It « ««<
Ifp=0goto 10 To test inversion, input p = -4 followed by the Ro.
Set M = R2 + !2 lo, R,, ..., I,s values listed above. Output values: those
Output M listed above as to, io, r , ,. .., il,.
M=Pk< ««<
(10) If k < m go to (9)
The physically significant result of the transformation is usually the real quantity ti R which, as a function
of the frequency k/t, is known as the power spectrum of the input data. The algorithm also outputs this quantity,
Pk.
Many applications of Fourier transformation require that the transformed data be processed in some way in the
frequency domain and then inverted back to the time domain. The equations describing the inversion are
32:14:5 r, =
N-I r
I Rk cos)
kj 2irkj\1
(1 - /t sing N
ij
rrL /2arkj /2nkj/
32:14:6 _E L/k cosl
32:14:44
N
+ Rk sinl N
and are so similar to equations 32:14:3 and that the same algorithm may be used to effect inversion as
transformation. Thus, the algorithm presented above is actually a fast fourier transformation/inversion algorithm.
To select inversion, one merely inputs -p instead of p, followed by Ro, Io, R1. I,. R,...., IN-2- RN_ IN_,. The
output is to, b, r1, il, r2, ... iv-2, rN-1, iv-1-
CHAPTER
33
THE SECANT sec(x) AND COSECANT csc(x) FUNCTIONS
These functions are the reciprocals of those discussed in Chapter 32. The secant and cosecant functions are inter-
related by
33:0:1 [sec2(x) - II[csc2(x) - 11 = I
and by
33:1 NOTATION
The symbol cosec(x) sometimes replaces csc(x). To avoid possible confusion with the functions of Chapter 29. the
names circular secant and circular cosecant may be used. Because of their applicability to triangles [see Section
34:141. the functions of Chapters 32-34 are known collectively as trigonometric functions.
33:2 BEHAVIOR
Figure 33-1 shows that sec(x) and csc(x) adopt all values except those between - I and + 1. The two functions
may receive any argument, but sec(x) is ill defined when x = ±tr/2. ±31r/2. ±5w/2.... and csc(x) is similarly
indefinite at x = 0, '-n, ±21r, .... Both functions are periodic with period 2n. With the argument interpreted as
an angle, the signs acquired by the sec and csc functions depend on the quadrant [Section 32:21 in which the
argument lies, as illustrated in Table 33.2.1. which includes the other four trigonometric functions.
33:3 DEFINITIONS
The secant and cosecant functions may be defined as the reciprocals of the functions of Chapter 32:
33:3:1 sec(x) _
cos(x)
311
33:3 THE SECANT sec(x) AND COSECANT csc(x) FUNCTIONS 312
FIG 33-1 : : :
33:3:2 csc(x) =
sin(x)
Three similar right-angled triangles are shown in Figure 33-2. If the sides represented by the dashed lines are
all of unity length, then the other six sides have lengths that define the six trigonometric functions as depicted in
the diagram. The argument x of these functions may be interpreted either as the marked angle or as the length of
the arc of a unity-radius circle subtended by the angle, as diagrammed. Besides serving to define the six functions,
many interrelations between them may be derived by applying similarity and Pythagorean relationships to the tri-
angles. For example:
I sec(x) csc(x)
33:3:3
cos(x) I cot(s)
follows by equating the ratio of the red to the blue sides in the three triangles, while
33:3:4 csc2(x) = I + cot2(x)
is a consequence of applying Pythagoras' theorem to the third triangle.
Table 33.2.1
As well, the secant and cosecant functions may be defined in terms of exponential functions of imaginary
argument
2 exp(ix)
3 3: 3 : 5 sec(x) _
exp(2ix) + I
2i exp(ix)
33:3:6 csc(x) =
exp(2ix) - I
_ 2 r'`I °dt -a a
33:3:7 sec(x) < x< 2
n 0 tZ + 1 2
33:5 INTRARELATIONSHIPS
Whereas the secant function is even:
33:5:1 sec(-x) = sec(x)
the cosecant is an odd function:
33:5:2 csc(-x) = -csc(x)
33:6 THE SECANT sec(x) AND COSECANT csc(x) FUNCTIONS 314
csc(x) it = 0. 4. 8, .. .
csc(x
nn
+ 2)
- _ ±se(x) n=1,5,9....
-csc(x) it = 2, 6, 10.
-;: sec(x) n=3,7,11,...
parallel those of the cosine and sine functions, but them are no simple formulas, akin to 32:5:9 or 32:5:10, to
express sec(x ± y) or csc(x ± v). Such expressions-as well as those for sec"(x), csc(x) ± esc(v), etc.-are best
evaluated via the formulas of Section 32:5, making use of identities 33:3:1 and 33:3:2.
The secants and cosecants of double argument and half argument may be evaluated through the formulas
scC2(x)
33.55 sect) x) =
33:5:6) =
2 - sec2(x)
2 2 csc (x) - I
(x1 2 sec(x) /n+\
33:5:7 sect I = (-1)" 1 + c(x) m = lntl 21
and
( xl sec(x) (w,
33:5:8 csc = (-1)'" m = bit J
2 sec(x) - I 2zr
33:6 EXPANSIONS
The power series expansion of the secant
33:6:2
I
csc(x) _ - +
x
x
--
6
+
7x5
360
+
31x5
15120
+ _
p(4' - 2)B22{,r2j-l
(2j)!
-n<x<9r
1 2`
+ - E *I(2j)(x/ir)2i
X x r1
involves the Bernoulli numbers [Chapter 4) or the eta numbers [Chapter 3]. Series for the logarithms of sec(x) and
of x csc(x) invoke the lambda and zeta numbers [Chapter 31
k(21) 2x\' -.a Ir
33:6:3 1n(sec(x)) 2 <X<-
,-I J (7)
315 THE SECANT sec(x) AND COSECANT csc(x) FUNCTIONS 33:9
x;
/x
33:6:4 ln(x cec(x)) I - 1 -W < x < n
;-1 1 \
The secant and cosecant functions may be expanded as partial fractions
4ir (-1)'(2j + 1)
33:6:5 sec(x) _
rr2 - 4x2
- 97r2l21r- 4x2 + 25.r20w
- 4x2
- =a
(i + )zirx - s2
I
CSC(x)+ 2x 2x 2x I (-1)'
- 2x
'Rx -x x- 4az -X
33:6:6 +
X 9'tr2-x- x ;°1 )
-2'R2
- xz
as may their squares
4 4 4 4 1
33:6:7 sec-,(x) = + + + + _
(w - 2x)2 (n + 2x)2 (3Tr - 2x)2 2x)2 ._ [x + (j + 2)a]2
33:6:8 csc-(x)
1
+
I
+
I
+
1
+
1
+ _
' 1
33:9 APPROXIMATIONS
For small values of the argument, the approximations
-3/2
/ x-\\
33:9:1 sec(x) - I i- 3 8-bit precision -0.9 5 X:5 0.9
I x 2 2
33:9:2 cse(x) = - + - 8-bit precision - -:S X:5 -
are valid.
Table 33.7.1
6 r -1r -1r 22I' -xr -J6° -ar -5a' -°V -67' X72'' -7r -6t?
2
12 10 8 6 5 10 3 8 5 12
1 V6-V3 2-
V5
Va-VS -2
V3
V6-V20 V2 2--vs
=V; V1 -V6-V20 -V4- V5 -2 2--' ' 5
-V'2 .V6-VIO\'1- -Va-VS \
,5 sV0=V2
33:10 THE SECANT sec(x) AND COSECANT csc(x) FUNCTIONS 316
and
d -csc2(x)
33:10:2 csc(x) = -csc(x) cot(x) =
dx sec(x)
whereas integration leads to
The formulas
coth(y) + i tan(x)
33:11:1 sec(x + ry) =
sec(x) sinh(y) + coc(x) cech(y)
coth(y) - i cot(x)
33:11:2 csc(x + iy) =
sin(x) csch(y) + csc(x) sinh(y)
reduce to
33:11:3 sec(iy) = sech(y)
33:11:4 csc(iy) _ -i cscb(y)
when the argument is purely imaginary.
33:12 GENERALIZATIONS
The Jacobian elliptic functions [see Chapter 631 nc(p;x) and dc(p:x) are generalizations of sec(x). to which they
reduce as p - 0. Similarly, csc(x) is the p = 0 limit of the ns(p;x) and ds(p;x) functions-
317 THE SECANT sec(xt AND COSECANT csc(x) FUNCTIONS 33:14
provided that the arguments x and y are suitably related. The relationship involves the gudermannian function
33:14:7 x = gd(y)
or by the integral
The same function may be realized as a special case of the incomplete elliptic integral of the first kind [Chapter
62]:
33:14:13 invgd(x) = F(l;x)
As Figure 33-3 demonstrates, both the gudermannian and its inverse are odd functions. The gudermannian
function approaches -7r/2 as its argument approaches ±¢. Conversely, invgd(x) approaches =x as x -* w tr/2.
The derivatives of gd(x) and invgd(x) are sech(x) and sec(x), respectively.
The power series of gd(x) and invgd(x) involve Euler numbers [Chapter 51 and are remarkably similar:
xs x' 61x1 Etjxx'"
33:14:14 gd(x) = x - - + -1 <x< 1
-6 24 5040 + l
_ 71(2 + 1 )!
JEjWj a
33:14:15 invgd(x) = x +
xr
6
+
x}
24
+
61x'
5040
+
_
Several other expansions exist for the gudermannian function and its inverse [see Beyer, Handbook of Mathematical
Sciences, pages 323-325].
CHAPTER
34
THE TANGENT tan(x) AND COTANGENT
cot(x) FUNCTIONS
34:1 NOTATION
The alternative notation tg(x) is occasionally encountered for the tangent function, while cotan(x) or ctg(x) some-
times replaces cot(x). To emphasize the distinction from the functions addressed in Chapter 30. the names circular
tangent and circular cotangent are often applied to tan(x) and cot(.t). As in Chapters 32 and 33, we use tan(N) and
cot(H) as alternatives to tan(g) and cot(x) whenever we wish to stress the angular interpretation of the argument [A.
B and C are used as angular arguments in Section 34:141.
34:2 BEHAVIOR
Like the other four trigonometric functions, the tangent and cotangent are periodic functions, but unlike the other
four the period is it, not 21r. This is evident in Figure 341. The signs acquired by tan(O) and cot(9) in the four
quadrants [as defined in Section 32:21 are tabulated in Section 33:2.
The tangent function has zeros at .r = 0. t1r, =21T, ... but is undefined at x = z7r/2, =3a/2, t5n/2, ....
Conversely, cot(s) has zeros at x = tir/2, t3a/2, t5w/2.... but is undefined at x = 0. =ir, t2tt
34:3 DEFINITIONS
Geometric definitions of the tangent and cotangent functions are possible with the aid of Figure 32-2. namely
34:3:1 tan(x) _
319
34:3 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 320
34:3:2 cot(x) =
-
The functions may also be defined in terms of the sine and cosine functions:
sin(x)
34:3:3 tan(x) =
cos(x)
cos(x)
34:3:4 cot(x) =
sin(x)
2i
34:3:6 cot(x) +i
exp(2ix) - I
also define the tangent and cotangent functions, as do the definite integrals
2 it
tZ - 1 dt x<2
34:3:9 tan(x) =
- 0
2 t-1-dt
34:3:10 cot(s) = x<7t
n J t-t
For appropriate values of a, 6 and c, the differential equation
34:3:11
d afZ+bf+e b2 < 4ac
is solved by
When b2 > 4ac, a solution is f = -(b/2a) - 4ac/2a) tanh(xVb c4ac4a /2) or -(b/2a) -
(Vb- - 4ac/2a) [see equation 30:3:6 for a special case]. When b2 = 4ac. a solution is f =
-(1 + bx)/4ax.
34:5 INTRARELATIONSHIPS
The tangent and cotangent functions are both odd:
34:5:1 f(-x) = -f(x) f = tan or cot
The argument-addition formulas
tan(x) ± tan(y)
34:5:2 tan(x ± v)
I = tan(x) tan(y)
and
cot(x) cot(y) 7- 1
34:5:3 cot(x ± y) =
cot(y) ± cot(x)
have the special cases
2 tan(x) 2 cot(x) I
34:5:4 tan(2x) _
T- tan2(x) cot-(x) - I cott2x)
and
34:5:5 tan(3x) =
I - 3 tan-(x) coe(x) - 3 cot(x) cot(3.r)
.44:5 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 322
tan(x) n=0.4,8....
tan(x) ± 1
n= 1.5.9,...
34:5:6 tan( x +/ = I r tan(x)
`` 4 -cot(x) n=2,6,10,...
tan(x) = 1
n=3,7,11....
1 ± tan(x)
and
cot(x) n=0.4,8,...
cot(s) w 1
/ n= 1.5.9,...
34:5:7 cotl x t 4 _ I = cot(x)
-tan(x) n=2,6,10....
cot(x) ± I
-+ cot(x)
n=3,7.11....
I
((x 1 cos(x)
34:5:8
T -tan 2, = (-1)T 1 + cos(x)
is a useful quantity because it is related in a very simple fashion to the following trigonometric functions of Jr
2.7
34:5:9 sin(x)
1+ r
1-T-
cos(x) =
34:5:10
-
T +- : T '
I +T2
34:5:11 sec(x) =
7---Ti
I+ r2
34 : 5 : 12 cec(x) =
2T
1 1
34:5:13 tan(x)
1-T I+T
T'
34:5:14 cot(x) _
2T
2
34:5:15 I + cos(x) =
l + T2
I+T
34:5:17 sec(x) + tan(x) =
I -T
I- T
3 4: 5 : 18 sec(x) - tan(x) =
I + T
323 THE TANGENT tan(x) AND COTANGENT coUx) FUNCTIONS 34:6
These equations apply irrespective of the magnitude of x. that is, in all four quadrants, as defined in Section 32:2.
The series
34:5:21
r
T+-+-+ T3
'r1 ' 1
invgd(x)
3 5 2j + l 2
cos(x - y) + cos(x _ y)
34:5:27 cot(x) cot(y) =
cos(x - y) - cos(x + y)
sin(x + y) + sin(x - y)
34:5:28 tan(x) cot(y) =
sin(x + Y) - sin(x - y)
complete our list of intrarelationships.
34:6 EXPANSIONS
Power series corresponding to the tangent and cotangent functions and their logarithms are
x3 2.r' 17x' 2 7 (4x')`
34:6:1 tan(s) = z + + + + -
3 15 315 x tr-,
4'(4'-I)IB:i x'3 -a IT
(2j)! 2 2
34:6:3 In
-l
[can(s)]
x
-
-ln[xcot(x)]
-r
3
7x
90
62x6
2835
1 -:-<x<-
4'(4) - 2)IB_j
2j(2j)!
n
x
=±rl(21)(
i rr 12T 2
where the X. C and Tj numbers are considered in Chapter 3 and the Bernoulli numbers in Chapter 4.
34:7 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 324
except that the roots ±ro(b) exist only if b a 1. The positive members of this set may be found by utilizing the
following simple algorithm, which is based on inversion of equation 34:7:1 to
34:7:3 x = Arctan(bx) = jtr + arctan(bx) x = r,(b)
where the Arctan and am-tan functions are discussed in Chapter 35. Starting with a crude estimate of r,(b). namely
fir (or A/4 for j = 0), repeated application of 34:7:3 converges to the exact value of the j' positive root. The
algorithm halts when two successive estimates of r,(b) differ by less than 10-9.
Table 34.7.1
6 0. :13' =10' =221' 230' x36' x44 234' 560' =67" :72' x73' :9P
0
tt xt as xt x :3t tx :3t alt x3n
12 10 t 6 5 4 10 3 a 5 12 -
0 =n-\5I 1-
7 3
: s-v20 si ; : =(N,5 1) = =c:
tQ+f, a,rl--
r 0
2
5
i evs -720
7 ( T42-V-3 0
325 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 34:9
With j taking values from 1 through x, the sums Er; "(1) occur in certain problems and have the values ,a, .4 and
s for n = 2, 4 and 6. The values of r,(I) occur in the expansion of the Langevin function [Section 30:13] and
correspond to the zeros of the spherical Bessel functions of the first kind [see Section 53:7]:
Similarly, the roots of the equation
34:7:5 cot(x) = bx -x<b<x
which we denote
34:7:6 x = ±pi(b), ±p2(b), ±p3(b), ..., ±pib), ... where f r - it < p,(b) <jir
are determinable via the recursive algorithm
34:9 APPROXIMATIONS
For small values of the argument, the tangent and cotangent functions can be approximated by
X
34:9:1 tan(x) 8-bit precision -0.6 ts x < 0.6
3
34:10 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 326
.and
x2l'
( 1 6/
34:9:2 cot(x) = 8-bit precision -0.5 x 0.5
x
From expansion 34:6:6 it is evident that I /(x - x,) is an approximation to cot(x) or -tan(x) whenever x is
close to one of the arguments, x,, at which the function is infinite. In fact:
-x
= ±a/2, ±3a/2, t5tr/2, ... 8-bit precision -0.1 s x, - x s 0.1
1
34:9:4 cot(x) _ x, = 0, ±n. =2Tr, ... 8-bit precision -0.1 sx-x;50.1
x - x,
r 1
34:10:3 tan(t)dr = ln(sec(x)) In(1 + tan2(x))
J0
r
34:10:4 cot(t)dt = ln(csc(x)) = 2 ln(1 + cot-'(x))
Z/
0
34 10 7
: :
f tan'(tw
J p
ca'(t)r = 2 0 (
v1
I
and
q 0W4 N
34:10:8 ln( tan(t))dt ln(cot(t))dr = G
0
327 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 34:13
and involve the secant function [Chapter 33], the beta function [Chapter 3] and Catalan's constant [Section 1:51.
The use of formulas 34:5:9-34:5:20, together with the differential identity
2dT
34:10:9 dt= T=tan(2)
can often aid the integration of expressions involving trigonometric functions by converting the integrand to a purely
algebraic function. As a simple example
/n \ /
34:10:10 = -r (1 + Tr)di = 2 tan (x2/I + 6 tan3l 2/ l
fx [ 1 + (r)]z
sin(2x) - i sinh(2y)
34:11:2 cot(x + iy) =
cosh(2y) - cos(2x)
and
34:11:4 cot(iy) = -i coth(y)
34:12 GENERALIZATIONS
As periodic functions, the tangent and cotangent are special cases of the functions of Chapter 36.
Respectively. tan(x) and cot(x) generalize to the Jacobian elliptic functions sc(p:x) and cs(p;x) that are dis-
cussed in Chapter 63. When p = 0, we have
34:12:1 sc(O;x) = tan(x)
The tangent and cotangent functions are related to the other circular functions [those addressed in Chapters 32 and
33] by
34:13:1 tan(x) =
a4 sin(x)
I - sin (x)
az 1 - Cos (z)
cos(x)
a3 sec (x) - l =
a3
csc (.r) - 1
_-I
cot(x)
34:13:2 cot(x) _ _ _
0`3
= a3 CSCix) - I = -
sin(x) 1 - cos'(x) sec (x) - 1 tan(x)
where a2, a3 and a4 take the values + I or -1 according to the magnitude of x. as discussed in Section 32:13.
The inverse functions of the tangent and cotangent are among the functions that are the subject of the next
chapter.
34:14 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 326
a a
Table 34.14.1
a b C A B Area
ab
d+b' arctan(a/b) arctan(b/a)
2
a cot(A) a cw(A)
r.
2-A
-2 ca(A)
a 42
a tan(g) a sec(B) --B tan(g)
2 2
V-P atccos(b/c) aresin(b/c) - 2 b'
v b'
b tantA) J b sec(A) 2--A 2
ton(A)
n b'
b cor(B) b cee(8) __H ,
- cot(B)
A f
c sin(A) c coa(A) 2 A sin(2A)
4
c'
C ca(B) c ara(B) Z- B sin(2B)
4
329 THE TANGENT tan(s) AND COTANGENT cot(s) FUNCTIONS 34:14
Table 34.14.2
C = 2 arctan
c'-(a-b)'
(a+b) -c'
Two sides and the angle c = V a + b' cos(24) - 2b cos(A) a - sin-(A)
opposite the longer of [ b sin (A)1
those two sides. e.g.. B = aresin J
a, b, A where a a b a esin(A)
cas(A)
a'
- stn (A )
rrb b' 11
J
C = arccasl - sin'(A) + cos(A) 1-
- sin'(A)J
For a triangle that is not necessarily right angled, there are six parameters, and three of these (including at
least one side length) must be known in order for the triangle to be determined. For example, if the side a is of
known length and the angles B and C are also known [see Figure 34-3J, sides b and c are calculable by the formulas
given in Table 34.14.2, as is the angle A. Notice that prescribing the magnitudes of a, b and B does not fully
delineate the triangle if a exceeds b. The line of length b has two alternative positions, in that case, as illustrated
in Figure 34-3 by the full green line and the dashed green line, and consequently the magnitudes of c, A and C
each have two alternative values. The alternative magnitudes of these quantities are given in Table 34.14.2, as are
the two possible areas. The formulas in this table are based on the cosine law
34:14:1 2bc cos(A) = b' + c' - a2
the sine law
sin(A) sin(B) sin(C)
34:14:2
a b c
34:14 THE TANGENT tan(x) AND COTANGENT cot(x) FUNCTIONS 330
where s is the semiperimeter, s = (a + b + c)/2. Another useful rule in the mensuration of triangles is the tangent
rule
tanf(A - B)/2] a-b
34:14:4
tanl(A + B)/2] a+b
Three constructions that are important in trigonometry are shown in Figure 34-4. The altitude AO. that is the
line through A perpendicular to BC, has a length
34:14:5 AO = b sin(C) = c sin(B) (angle BOA) = a/2
A line through a vertex of the triangle that bisects the angle there is known as a bisector: the bisector of angle A
in Figure 34-4 has a length
2bcb s(Ac /2) A
34:14:6 AN = (angle BAN) = (angle NAC) =
2
Line AM. which bisects the line BC, is termed a median and has the length
t a_
34:14:7 AM =2 b`--c'+2bccos(A) BM=MC=2
CHAPTER
35
THE INVERSE TRIGONOMETRIC FUNCTIONS
The fundamental interrelationships between the six subjects of this chapter are
and
J X \ xz+ t\
35:0:3 arctan(x) = aresinl I = arccsc
1 I
arccot - - n = -arccos = -arcsec( r'' + 1) x<0
1 -z' I
arccotl 1 = arccos( I - x') = aresecl 0<x<I
l -x' 1
331
35:1 THE INVERSE TRIGONOMETRIC FUNCTIONS 332
35:0:5 arccsc(x) =
(sn(x)) =
(I) >
x
/ x \/ I + xx
35:0:6 arccot(x) = arccosl
\ I+xz ) = aresecl
\ X
arstanl J
Q
= aresinl
\ 1+x
1 I = arccsc(1 x'I+ ) z>0
__X t 1
x-1 x
n - atctan) = rr + aresin 1 = a + arccsc x < -1
x x'
Many more interrelationships may be constructed by combining equations 35:0:3-35:0:8 with those in 35:0:2. For
example:
/ ` / x'
35:0:9 arctan(x) = 2 - arccosl
r
\I/ x
+ x/
I=
-2
- aresecl
l
\
r
x
l
= sgn(x)I 2 - aresinl 11_+X1) sgn(x)I 2 - arccsc( x' + 1)J
follows from combining 35:0:2 with Lthe right-hand members of equation 35:0:3.
35:1 NOTATION
An alternative collective title for the functions of this chapter-the inverse tangent, the inverse sine, the inverse
cosecant, the inverse cotangent, the inverse cosine, and the inverse secant-is the inverse circular functions. The
333 THE INVERSE TRIGONOMETRIC FUNCTIONS 35:2
symbolism tan-'(x), sin-'(x), etc., often replaces arctan(x), arcsin(x), etc. Variants such as arctg(x). arcctg(x) and
arccosec(x) are occasionally encountered. The origin of the "arc" prefix is made evident in Section 35:3.
The notation arccot(x) is sometimes used to denote a function defined, for negative x, somewhat differently
than here, being equal to our arccot(x) - n. Similar discrepancies may be encountered for other inverse trigono-
metric functions.
The multivalued functions Arctan(x), Arcsin(x), etc., are discussed in Section 35:12, but, confusingly, these
are often also denoted arctan(x), aresin(x), etc.
We use arctrig(x) to represent any one of the six inverse trigonometric functions.
ie
O
i i h O g g O g O to
arccec (x) - -
......... ......... ..................... -0.5
arcein(x)-
.. : .......:....:....: ..............-1.0
...... ...................................
Figure 35-1 depicts the behavior of the six functions. Note that both the inverse cosecant and inverse secant func-
tions have two branches. The domains and ranges of all the functions are summarized in Table 35.2.1.
35:3 THE INVERSE TRIGONOMETRIC FUNCTIONS 334
Table 35.2.1
35:3 DEFINITIONS
Indefinite integrals define each of the six inverse trigonometric functions
dt
35:3:1 arctan(x) =
" 1 + t1
dt
35:3:2 aresin(x) _ -1 x51
0 1 -t=
dt
=t t-- 1
35:3:3 arccsc(x) =
dt
1sx<x
t t'1
dt
35:3:4 arccot(x) =
1+
dt
35:3:6 atrsec(x) = x a 1
t t t2 - 1
Diagrams illustrating these definitions, for x > 0, are included as Figures 35-2, 35-3 and 35-4.
Geometric definitions of the inverse trigonometric functions may be illustrated by reference to Figure 35-5 for
arctan(x). A line of length x is made part of a right-angled triangle, the side shown dashed being of unity length.
Thus, x is the tangent of the marked angle. The arc of a unity-radius circle subtended by this marked angle, shown
in black on the diagram, is then defined as the inverse tangent of x. Alternatively, the angle itself may be identified
with arctan(x).
The names "inverse trigonometric functions" imply that these functions are the inverses of the functions of
Chapters 32, 33 and 34. This is true, however, only if appropriate restrictions, namely
35:3:7 f = arctan(x) where x = tan(f )
ar
35:3:8 f= arcsin(x) where x= sin(f) -Zsfs2
35:3:9 f = arccsc(x) where x = csc(f )
335 THE INVERSE TRIGONOMETRIC FUNCTIONS
....................
35:4 THE INVERSE TRIGONOMETRIC FUNCTIONS 336
If f is an inverse trigonometric function, then f(x) ± f(y) may lie within the range of f, or above it, or below
it. These three possibilities are incorporated into the following function-addition formulas by allowing the integer
k to adopt one of three values:
35:5:7
35:5:8
arctan(x) ± arctan(y) = krr + arctanl x
I1
I - x) k = 0, ±1
x9
arccsc(x) ± arccsc(y) = ka + arccsc
35:5:9
(y--r± xt-I
The appropriate k value may be found from the formula
/
k = Int 2 +
1 f(x) ± f(y)1
J f = arctan, arcsin or arccsc
35:5:10
x'-1 yz-1
(1,;:
337 THE INVERSE TRIGONOMETRIC FUNCTIONS 35:8
where
35:6 EXPANSIONS
There are three basic power series expansions for inverse trigonometric functions:
Table 35.7.1
x=-x x=-V2
-I
V2
x=0 x=-
V2
1
computes any one of the six inverse trigonometric functions by utilizing a built-in arctan function. The algorithm,
which is exact, uses relationships 35:0:1 and 35:0:2.
-ts-well, the universal hypergeometric algorithm (Section 18:141 may be used to calculate inverse trigonometric
functions.
35:9 APPROXIMATIONS
Among the approximations to inverse trigonometric functions are
3x
35:9:1 arctan(x) = 8-bit precision -0.45<x50.45
3 + x2
3x
35:9:2 arccot(x) = 3X2 + 1 8-bit precision x Z 1.8
35:10:5
J ` arccot(bt + c)dt = (x + I arccot(bx + c) + b In VI + (bx + c)Z
b
b -(X+bc\I
c
35:10:6
I
Jc aresin(bt+c)dt=Ix+-)aresin(bx+c)--+
\/
1
i -c-lsbxsI-c
r /
35 :10:7 c arccos(bt + c)dt = x + b arccos(bx + c) + - tiZ - t x +b I -c - 1 s bx < 1 - c
J /b ( b
35:10:8
J
f j arccoc(br + c)dt = (x + b) arccsc(bx + c) + b arccos(bx + c) bx > I - c
\
35:10:9 j arzsec(bt + c)dt = I x + 1 aresec(bx + c) - - arccos(bx + c) bur > 1 - c
J 0 -c)/b 6 b
Indefinite integrals of the form ftarctrig(bt + c)dt include
r I :: b c+
35:10:10
J t/bt arctan(bt + Odt = + z - arctaa(bx + c) - 2b,
t l0 1 + (bx + c)2
1 +62x2-C2 2bx+2c-ir c 2
35:10:11 Jtarccot(b:+c)dJ= arccot(bx + c) +
- bile I + (bx + c)
262 462
BYxt _ 2c'
2b c
35:10:12 t arcsin(bt + c)dt = - 1 arcsin(bx + c) +
r/b 4b bt
bx - 3c
+ I - (bx + c)2 -c - I s bx s I - c
462
r 22_ x
I -
35:10:13 I c
111 /b
t arccos(bt + c)dt = x
4b - arccos(bx + c) +
-
bx - 3c
462
I-(bx+c)2 -c-Isbx51-c
but none of ft-'arctrig(bt + c)dt has been evaluated as a finite number of terms. See Spiegel [pages 82-841 for a
long list of indefinite integrals of the form ft="arctrig(t)dt for integer n. Gradshteyn and Ryzhik [Section 2.81 list
similar integrals and include additional entries such as farctrig"(t)dt and f(bt + c)-'arctrig(t)dt.
Gradshteyn and Ryzhik [Section 4.5] also present over 100 definite integrals involving inverse trigonometric
functions;
35:10:14
acctan(t) dt - f arccot(t) dt = G
J
and
r' aresin(t) a
35:10:15 I r dt = 2 ln(2)
0
d1/2 1 aretan(1x)
35 : 10: 16
dx'n V
dr 'n atrcot(V x) Vx 1 + x)
A variety of ways may be used to extend the inverse trigonometric functions to complex argument. One might
couple the definitions 35:3:1-35:3:6 with integration in the complex plane or, alternatively, permit power series
such as 35:6:1 and 35:6:3 to accept a complex argument. Another route makes use of the relationships 35:13:1-
35:13:6, which are valid for complex argument as well. The results, however one proceeds, are rather formidable
looking. The most useful of the resulting multivalued functions are
35:11:1 Arcsin(x + iy) = kir + (-1)4arcsin(Y) + (-1)ri In[X + X' --I]
35:11:2 Arccos(x + iy) = 2k7r ± {arccos(Y) - i In[X + XZ - ]}
2
35:11:4 X=1
2
(x+l)2+y2+
2
(x-1)2+yr y=!2 (x+1)2+y^-2 (x-1)Z+y2
The logarithmic function In occurring in the above equation is the single-valued logarithm defined in Chapter 25.
35:12 GENERALIZATIONS
Because the trigonometric functions we periodic, their unrestricted inverses have infinitely many values for each
acceptable argument. These multivalued inverse functions are distinguished from the single-valued functions treated
elsewhere in this chapter by having their initial letter capitalized. The definitions are encompassed by
35:12:1 Arctrig(x) = y when trig(y) = x
With k = 0, ±1, ±2, ..., the relationships
35:12:2 Aretrig(x) = arctrig(x) + 2kir arctrig = aresin, arccos, aresec, arcsec
35:12:3 Arctrig(x) = arctrig(x) + for arctrig = arctan, arccot
hold.
The inverse tangent and inverse sine functions are each special cases of the Gauss F function [see Chapter 60):
Also, the inverse sine and cosine are special cases` of the incomplete beta function of Chapter 58:
341 THE INVERSE TRIGONOMETRIC FUNCTIONS 35:13
r l 12;x'
35:12:6 aresin(x)= 1 2B 2;
!1 1
35:12:7 arccos(x) = B 2, 2;1 - xr
2
36
PERIODIC FUNCTIONS
Periodic functions play an important role in the solutions of many difficult problems in applied mathematics. Also,
periodic functions, or nearly periodic functions, form the medium by which a good many information transfers
take place. For example, beams of light, sound waves, and a variety of telecommunications signals are instances
of periodic functions, often "modulated" in some way.
36:1 NOTATION
We shall use per(x) to represent any periodic function, and occasionally qer(x) to represent a second periodic
function. Throughout this chapter P will denote the period of per(.r). The quantity 27r/P is sometimes known as
the frequency of the periodic function and is often denoted by w.
36:2 BEHAVIOR
Apart from their characteristic of indefinitely repeating, periodic functions share no common behavior. Periodic
functions may be continuous or discontinuous, simple or complicated. Examples of periodic functions are shown
in Figures 36-1 through 36-7, exhibited later in this chapter.
36:3 DEFINITIONS
....... .........1.........r..4
This particular periodic function, which is mapped in Figure 36-1, has a period of 2. As here, synthetic periodic
functions often exhibit a discontinuity at one point, at least, in the 0 rt x < P range.
36:5 INTRARELATIONSHIPS
The fundamental intrarelationship of periodic functions is their periodicity, which constitutes a recurrence formula
36:5:1 per(x + P) = per(x)
If two periodic functions per(x) and qer(x), with periods P and Q, are added, subtracted, multiplied or divided,
the resultant function will be periodic only if the quotient P/Q is rational. If P/Q is rational, the period of per(x)
± qer(x), per(x)ger(x) and per(x)/qer(x) will normally be the least common multiple of P and Q. There are, how-
ever, frequent exceptions to this rule; for example, sin(x) and cos(x), each of period 2w, have products and quotients
of period it, not 2w as predicted by the rule.
If the argument x of a periodic function per(x) is replaced by a linear function of x, to give per(bx + c),
periodicity is maintained but the period is altered from P to P/b. Replacement of the argument by any other
aperiodic function of x destroys the periodicity; the functions per(ax2 + bx + c) and per(] /x), for example. are
not periodic. Multiplication or division by, or addition or subtraction of, any aperiodic function other than a constant
likewise destroys the periodicity.
On the other hand, composites of periodic functions, such as a per(bx) + b per(x) + c, exp(per(x)) or 1/[c +
per(x)] are themselves periodic. Such functions have the same period P as per(x) or, occasionally, a submultiple
of it such as P/2.
345 PERIODIC FUNCTIONS 36:6
36:6 EXPANSIONS
In a process known as harmonic analysis, any periodic function per(x) of period P may be represented, exactly or
approximately, as the Fourier series
(2jirx) (2jlrx) _ per()
36:6:1 F(x) _ - + Ccos + s s in x
i-
The c and s coefficients in this series, known as Fourier coefficients, are given by the so-called Euler formulas
36:6:4 c; + s; =
2
J
°. pee(r)dt
2+ P
o
provided that per(x) is everywhere finite. A brief tabulation of Fourier coefficients for a variety of periodic functions
is included in Section 36:14.
The Fourier series 36:6:1 may be written in the alternative form
co (2J1rx
36: 6:5 F(x) = z + + arctan(si/ci) I
P
We shall not discuss the convergence of Fourier series in detail [see Hamming, Chapter 321. Suffice it to state
that the Fourier series of a finite continuous periodic function converges to the function, whereas the Fourier series
of a finite discontinuous periodic function converges to the function except at the points of discontinuity. If the
periodic function per(x) has a discontinuity at x = d (as well as at x = d t P, x = d ± 2P, etc., of course), such
that
36:6:8 lim per(d - e) = per(d-) e>0
a-o
and
then the Fourier series converges to [per(d-)+per(d+)[/2 at x = d. For example, the periodic function defined in
36:3:2 and mapped in Figure 36-1 has the Fourier series
4 4
36:6:10 - + 4- sin(jwx)
3 j'tr' .PR
This evaluates, for x = 0, to s + 4;(2)/7rr = 2, which is the mean of the values, 4 and 0. on either side of the
discontinuity at per(0).
36:7 PERIODIC FUNCTIONS 346
P co
2
36:7:2 per
4
=
2
+si-c.-sy+c.+ss-ce-...
P co
36:7:3 per 2 =2-cl+c_-ey+c.-cy+c6-...
per(3P) = co
36:7:4
4 2
-si-cr+sy+c.-sy-ce+...
36:9 APPROXIMATIONS
The derivative of a periodic function per(x) is another periodic function of identical period. Discontinuities in per(x)
will generate Dirac delta functions [Chapter 10] in d per(x)/dx.
The co/2 Fourier coefficient represents the average value of per(x) over the period, that is:
I co
per(t)dt = x arbitrary
36:10:1
P f o Z
Only if this coefficient is zero will indefinite integration of per(x) give rise to another periodic function.
After subtraction of co/2, differintegration with lower limit minus infinity converts per(x) into another periodic
function of identical period. In terms of its Fourier series, the new function is
36:10:2
d"
[d(x + -)]'
[per(x)
co]
2J (P /
2jrrl
[c, cos( P + 2l /2
+ s sinl p vu 1
+ 2)
(2!)'
2!) 2jirx 2jax`
= -P I/ ,.,
f(c,C + s,S)cos(
P/
I + j"(s,C - c;S)sin(
36:12 GENERALIZATIONS
Apart from the double periodicity cited in Section 36:11. no generalization of the concept of a periodic function
has been made.
Figure 36.2 shows an example of each. Full- or half-rectification of a periodic function maintains its periodicity
and, generally, the period remains unchanged, Sometimes, as in the full-rectification example shown in Figure 36-
2, rectification decreases the period (increases the frequency) by a factor of 2 (or more).
FIG 36-2
sin(2nx/P)
7 half rectification
Table 36.14.1
C St
Name of (O
(unction Figure pert.)
4
S q uare wave
363 1-1)b"iSie 0
(odd) Jn
r I _l] _4
Triangular
wave
36.6 t-,) ' 2rs'frac( 2+ P 2 : r
0 0
P
- sint -- j
-k
h P
4c rlrrh
- sinll-
P.
0 0
-
Is
Full-rectified
/ xx -4
36-2 sinl 0 0 0
sine wave `p (j: _ I M
Ixsr. 2a r ut2-J)
!2 zin l P) + !2
Half-rectified l 1
36.2 zi ng/ 0 0
sine wave P r
:....;....:FIG 36-3:....:..p
:....:....:FIG 36-4;......p
;Q{s,
44Q 4+Q 4
;... 0. 5
FIG 365;......0
349 PERIODIC FUNCTIONS 36:14
;4 ti
y4Q A?
ti
44,Q, y440
y4p +4e +4q?
:FIG 36-7 :
CHAPTER
37
THE EXPONENTIAL INTEGRAL Ei(x)
AND RELATED FUNCTIONS
Indefinite integrals of the form ft-'exp(±t)dt cannot be expressed in terms of elementary functions for n = I. 2,
3..... The exponential integral function Ei(x) fills this deficiency. A closely related function, the entire exponential
integral, is also discussed in this chapter, it is related to Ei(x) by an identity
37:0:1 Ein(x) = y + ln(kl) - Ei(-x) y = 0.5772156649
involving Euler's constant y [Section 1:71 and the logarithm of the absolute value of the argument x.
The logarithmic integral 11(x) is discussed in Chapter 25 but, because of its simple relationship
37:0:2 li(x) = Ei(ln(x))
to the exponential integral, many of its properties are given in this chapter.
37:1 NOTATION
Notations abound for the exponential integral function: Ei*(x), Es(x), Ei(x), i(x), E`(x) and E_(x) have all been
used to denote Ei(x) or some very similar function. When one of these symbols is encountered, care is needed to
ascertain whether its definition differs from that which is employed for Ei(x) in this Atlas.
When x is negative, the expression -E,(-x) is often used to replace Ei(x) in view of relationship 37:13:7. The
E,(x) symbol here represents a Schl6milch function [see Section 37:13] and should not be confused with the identical
symbol used for Euler polynomials [Chapter 20]. To add to the confusion, the notation ei(x) is sometimes en-
countered for the Schli milch E,(x) function.
37:2 BEHAVIOR
Figure 37-1 maps the behaviors of the exponential integral Ei, entire exponential integral Ein and logarithmic
integral li functions. All three functions increase without limit as x - x but, following an inflection [Section 0:7]
at x = 1, the increase is dramatic for the exponential integral function (e.g., Ei(10) - 2 x 10' and Ei(100) - 3
X 10"). Note also that Ei(x) rapidly approaches zero as x -. - x and exhibits a discontinuity at x = 0. Observe
also that li(x) is discontinuous at x = 1 and is defined only for x ? 0.
351
FIG 37-1 Black
37:3 THE EXPONENTIAL INTEGRAL Ei(x) AND RELATED FUNCTIONS 352
37:3 DEFINITIONS
1 exp(t)
37:3:1 Ei(x) = J dt
a I
for all x, as illustrated in Figure 37-2. At zero argument the integrand in 37:3:1 encounters an infinity so that for
x > 0 the integral is to be interpreted as the Cauchv limit
is illustrated in Figure 37-3. Its relationship, equation 37:0:1, to the exponential integral Ei permits several alter-
native definitions of the Ein function.
Definitions of the logarithmic integral function were presented in Section 25:13. Some authorities regard the
logarithmic integral as defined only for arguments exceeding unity, but in this Atlas li(x) exists for 0 s x < 1 with
a Cauchy limit interpretation, similar to 37:3:2. serving to extend the definition to .t > 1.
1A
37:5 THE EXPONENTIAL INTEGRAL Ei(x) AND RELATED FUNCTIONS 354
37:5 INTRARELATIONSHIPS
Relating the exponential integral and its entire analog is the identity
37:5:1 Ei(x) + Ein(-x) = Ei(-x) + Ein(x)
that follows from equation 37:0:1. or through definitions 37:3:1 and 37:3:4.
We know of no reflection, recurrence, addition or multiplication formulas for the exponential integral functions
that involve a finite number of elementary functions. However the reflection formulas
37:5:2 Ei(-x) = Ei(x) - 2 Shi(x)
and
37:6 EXPANSIONS
Two alternative power series:
xr x3 X + (-X)'
37:6:1 Ein(x)=x-+18 -
4 96 i- j!j
and
are available to express the entire exponential integral function. In 37:6:2, 0 is the digamma function described in
Chapter 44.
The exponential integral function is expansible in terms of Lagucrre polynomials [Chapter 23J:
L;(-x)
(-x)
37:6:3 Ei(x) = -exp(x) <0
i-a j+1
as the continued fraction:
exp(x) 2 3
37:6:4 Ei(x) _
X-
-1
- -2 - 1
-
1- x- 1- x- I-
or as the asymptotic series
valid for large argument x. For small arguments, combination of expansion 37:6:1 with relation 37:0:1 leads to
x3
37:6:6 X-2
4 18
Table 37.7.1
s=-x x = 0 x = 1 x=x
Ei(s) 0 -= 1.895117816 =
Ein(x) 0 0.7965995993
(((x) undef 0 -s x
37:8:1 Ein(-x) =
J.r
(J + 1)'
+1
(J - I)x
J2
+1
)
(J - 2)x
(J - 1)=
+ +1 -
2r
9
+1
x
-+
4
1(-x)-
1
Set f = -10w
Input x >> D»» Storage needed: x, j and f
Replace x by 1n(x)
If x- 101 2:12 go to (2)
Ifx=0goto(4)
Set j = Int(IO + 2Gx1)
Set f I/(j+ 1)2
(I) Replace f by (fjx + I)/j1 Input restrictions: For the calculation of
Replace j by j - I must be positive.
Ifj*0goto(1)
Replace f by fx + In( I.78107241 Slxl )
Go to (4)
(2) Set j = Int(5 + 20/Ixi)
Setf = x
(3) Replace f by [ 1 /(1 If - I /J)] + x
Replace j by j - I Test values:
Ifj+0goto(3) Ei(25) = 3.00595092 x 109
Replace f by [exp(x)]/f Ei(-1.5) = -0.100019585
(4) Output f li(7) = 4.75705176
f= JEi(x)1
li(x) J
with J assigned the empirical value int(10 + 21x1), and then employs interrelation 37:0:1 to generate Ei(x). For x
s -2 or x >_ 22, the algorithm employs the continued fraction 37:6:4 terminated at (J - 2)/((1 - (J - 1)/(x -
J/(1 - J/x))) where J = Int(5 + The algorithm will generate li(x) when the additional instruction shown
in green is included. Because Ei(0) = Ii(l) = -x. the algorithm returns -1099 at these arguments.
The universal hypcrgcomctric algorithm [Section 18:141 also permits Ei(x), Ein(x) and li(x) to be determined
for suitable values of the argument x.
37:9 THE EXPONENTIAL INTEGRAL Ei(x) AND RELATED FUNCTIONS 356
37:9 APPROXIMATION
Based on the identity 37:0:1 and expansion 37:6:1, the approximation
1.05 s x < 0
36x + x=
37:9:1 Ei(x) ^ + ln(1.78[xl) 8-bit precision 0 < x s 0.36
36 - 8x
0. 38 5 x s 1.25
37:9:2 Ei(x) =
exp(x)
8-bit precision
X:5 -9
X? 15
based on expansion 37:6:5, is useful for arguments of large magnitude. The very simple approximation to the entire
exponential integral
36x - xr
37:9:3 Ein(x) = 8-bit precision -1.1 :5x:5 1.4
36 + 8x
may be used near the origin.
dx In(x)
vf0
I - cxpbbx + c)
37:10:4 Ei(br + Odr = (X + 1 Ei(bx + c) +
b
1 - exb bx-c)
37:10:5 Ein(br+c)dt= Ix+b) [Ein(bx+c)- 11+
b \\\ l
// J
B+b=0
When n = -1, the indefinite integral fI"Ei(bl + c)dl cannot be evaluated in a finite number of terms, but for n
= 0, 1, 2, 3, ... we have
357 THE EXPONENTIAL INTEGRAL Ei(x) AND RELATED FUNCTIONS 37:11
37:10:8
J-pie
t Ei(bt + e)dt
I
Ei(bx + c)
n+l J
x"+' -
while the n = -2, -3. -4, ... cases can be evaluated via the lengthy expression
(l c
\b/
i
l +
n!
(-b)"+' j
_o
c"-'[exp(bx + c)e,(-bx - c)- l)
(j + l)(n - j)!
Ei(bt + c) Ei(bx + c) cl l
37:10:9 J `is 1-1 dt = n(-c/b)" [I - (bx/ J
+
exp(c)e"_,(-c r
Ei(-c) - Ei(bx) + j!
[exp(bx) - exp(-c) 1J [ - I
e,(-c) 11
Jf
n(-c/b)" )_o (bx)j*' (-c)'*'
in which e,(x) denotes the exponential polynomial [Section 26:13].
Included among the definite integrals tabulated by Gradshteyn and Ryzhik [Section 6.21-6.231 are
37:10:12
I 0ln(Bt) Ei(-bt)dt = b b>0 B>0
-fY1 v)
37:10:13 I t' Ei(-bt)dt = + bz0 v>-I
,10 (1 + v)b'-"
37:10:14
Io
z t"-'exp(-Bt) Ei(-bt)dt = BI 0.v; b
\
B B I
/
b+B>0 v> 0
' - In(2 + v)
37:10:15 J t" li(t)dt =
I+v v> -2
o
37:10:16
f
J
f li(t)
t"
dt =
-ln(v - 2)
v- I
The r and B functions occurring in integrals 37:10:13 and 37:10:14 are, respectively, the gamma function [Chapter
v>2
where Ln is the multivalued logarithmic function discussed in Section 25:11. However, as explained in that section,
one usually selects a principal value of the Ln function, which leads to the single value
1
37:11:4 Ei(x+iy)=y+2ln(x'+y)-Ein(-x-iy)+iO
where 0 is the angle defined in Section 25:11. Notice that, according to 37:11:4, the exponential integral may have
a complex value even when its argument is real. In fact:
37:11:5 Ei(x + Oi) = y + In(x) - Ein(-x) + in x>0
The imaginary term in this expression is ignored in the rest of the chapter. Abramowitz and Stegun [Tables 5
and 5.7] provide tables from which the real and imaginary parts of Ei(x + iy) may be determined. For purely
imaginary argument, one has
n
37:11:6 Ei(iy) = Ci(y) + i Si(y) - sgn(y)
2
37:12 GENERALIZATIONS
The exponential integral function is a special case:
37:12:1 Ei(x) = -f(0;-x)
of the complementary incomplete gamma function [Chapter 45J and of the Tricomi function [Chapter 481
37:12:2 Ei(x) _ -exp(x) U(l;l;-x)
The entire exponential integral function is a hypergeometric function [Section 18:14]:
(I)i(-x) '
37:12:3 Ein(x) = x
-o (2),(2),
The first member po(x) = 2[sinh(x)]/x and all members are reducible, via the expression
n!
37:13:5 [3"(x) [exp(x)e"(-x) - exp(-x)e"(x)] n = 0, 1, 2. ...
enable the properties of E"(x) to be deduced from those of Ei(-x). The general relationship is
37:13:9 E"(x)=
-(-x)" ' rEi(
(n - 1).
-x) +
exp(-x) I
X
I --+ +
x
I 2!
X2
(n - 2)!1 ) 1
(-x)
I n=2,3,4,...
Some familial properties of Schldmilch functions are
37:13:10 E"(0)=-
n-1
1
n=2,3,4,...
d
37:13:11 E"(x) E. (x) n = 1. 2, 3....
dx
37:13:12 E"(x) = x"-'1'(1 - n;x) n = 0, 1, 2....
and the asymptotic expansion
I n n(n + 1) (n),
37:13:13 E"(.r) exp(-x) Is - + x)I,i
z' Jr,
The function
Very general expressions for indefinite integrals of the form f t`exp(bt)dt were presented in Section 26:10. Never-
theless. because of the very widespread occurrence of the integrals
1 3
37:14:1 t"exp(tt)dt v = 0, t 2, t I, t 2, ...
J
in applied mathematics, Table 37.14.1, containing such indefinite integrals, may be useful. Integrals of the forms
37:14 THE EXPONENTIAL INTEGRAL Ei(x) AND RELATED FUNCTIONS 360
Table 37.14.1
lI 2 exp(-x) rr 1 11 4V,
3
cxp(x)Ir d.-(N/;) Lx-21 3 erfNl`l
L 3Vx L
exp(x I exp(-x)
-2 -x Ei(x) - Ei(-x) +
X A
2 exp(-x)
X. 2 exp(x) 2 daw(V) - - 2 V a erfc(Vl)
Vx Vx
-x B(x) )
0 2 exp(x) daw(Vx)
exp(x)
0 exp(x)UV -daw(Vx)l
v'
- erfc(V/X) + V. exp(-x)
2
(x - I) exp(x) (x 1 1) exp(-x)
0
rr 3Vr 3 \r - erfc(v;) + VT x + - cxp(-x)
2 2 34A ( 2)
(x2 - 2x + 2) exp(x) (x= + 2x + 2) exp(-x)
CHAPTER
38
SINE AND COSINE INTEGRALS
This chapter concerns functions defined in terms of indefinite integrals of sin(s)/x. cos(x)/x and their hyperbolic
analogs. Whereas the definitions of the hyperbolic sine integral Shi(x) and the sine integral Si(x) present no dif-
ficulty, the corresponding integrals of cosines diverge at zero argument. Accordingly, in addition to the hyperbolic
cosine integral Chi(x) and the cosine integral Ci(x), it is useful also to define an entire hyperbolic cosine integral
Chin(x) and an entire cosine integral Cin(x). These entire integrals are related by
38:1 NOTATION
The initial letter of Shi and Chi is not always capitalized. The "h" that is used to identify the hyperbolic integrals
may occur elsewhere in the function's symbol. For example, Sih(x) sometimes replaces Shi(x). and the anagram
Cinh(x) is often used instead of Chin(x).
Some authors use ci(x) synonymously with Ci(x), but others employ it to denote -Ci(x). The notation si(x) is
usually encountered with the meaning
n
38:1:1 si(x) = Si(.r) - -
361
38:2 SINE AND COSINE INTEGRALS 362
38:2 BEHAVIOR
Figure 38-1 maps the Shi, Chi and Chin functions. Note that for large positive arguments Shi(x) and Chi(x) converge
and approach the value ; Ei(x), where Ei is the exponential integral [Chapter 37].
The damped oscillatory behavior of the Si and Ci functions is evident in Figure 38-2. As x -+ ± x. Si(x)
approaches ±0r/2) and Ci(x) approaches zero, whereas Cin(x) approaches infinity logarithmically via a series of
plateaus.
FIG 38-1 :
..:.... :.... :.... :.... :.... :.... :.... : J.:.. 3
38:3 DEFINITIONS
In addition to their definitions as integrals, the hyperbolic sine and cosine integrals may be defined in terms of the
functions of Chapter 37:
363 SINE AND COSINE INTEGRALS 38:3
-.1
4 47
'L -A 6 0 ti0 1'L
4 ti 4
+ sinc(t)dt = - - J sin(:)
It
2
-t dt
1 J
38:4 SINE AND COSINE INTEGRALS 364
cos(t)
38:3:5 Ci(x) _ -
and
r 1- COW)
38:3:6 Cin(x) =I dt
o t
Some other definitions may be constructed by employing relationships 38:0:1 and 38:0:2.
38:5 INTRARELATIONSHIPS
All cosine integrals, like the cosine itself, are even functions:
38:6 EXPANSIONS
The sine integral may be expanded as the power series
38:6:1 Si(x)=x--+--
x3 x3
600
x7
35280
18
= (-x Z
,_o (2j + 1)(2j + 1)!
The similar series without alternating signs, namely E.r2J"/(2j + 1)(2j + I)!. represents Shi(x). Likewise, re-
placement of all the - signs by + in the expansion
x4 x6 (-x2)'
38:6:2 Cin(x)X24 = - - - 4320
+x6 - - 322560 !_0 23(2j)!
96
of the entire cosine integral produces the series F.xz'/(2j)(2j)!, which represents Chin(x).
The sine integral is also expansible in terms of spherical Bessel functions of the first kind [Section 32:13J:
(l + 16[
(2/l = 4-sins\2! Il / llz
+5s6L\1-121sin12J-2cos(2) Jz+...
365 SINE AND COSINE INTEGRALS 38:7
More rapidly convergent even than 38:6:1 and 38:6:2 are the composite power series
2 - cos(x) sin(x) x x3 x5 (-xY
38:6:4 SO) -
x + x2
=
3
- 180 + 12600 -a (2j + 1)(2j + 3)!
and
38:6:5 Cin(x) +
sin(x)
+
I - cos(x) 3 x' xx xb - ;' (-x')'
x x2 2 24 1440 120960 )-1j(2j + 2)!
Asymptotic expansions of Ci(x) and of Si(x) - (a/2) may be constructed by combining expressions 38:5:3
and 38:5:4 with the expansions 38:13:5 and 38:13:6.
n minimum
38:7:1 Si(±2na) _ 2 + fi(2na) maximum
of Si(.t)
it maximum
38:7:2 Si(±(2n - l)ir) _ ± ± fi((2n - 1)a) minimum
n= 1.2,3....
minimum
38:7:3 Cii±l 2n - 1)- I = W . gil 1 2n - D
2 2 -) maximum
of Ci(x)
maximum
38:7:4 CiI±(2n-2)a1 = +gtl l,2» 2 1rI
minimum.
At x = :2a, ±4a, a horizontal inflection is displayed by Cin(x), that is, d Cin(x)/dx and d' Cin(x)/
d 2 are both zero at these values of the argument [see Section 0:71.
The zeros of the Chi(x) function occur at x = ±0.52382257. The first zeros of the Ci(x) function are found
at =0.61650549, and others occur close to the points of inflection of the function, which are given by
d2
38:7:5 ar Ci(x) = 0 .r = ±p,(-l) j = 1, 2. 3, ...
Table 38.7.1
SM.) 0
CMtx)
Chin(.) 0
Replace J by J - 2
if J * 0 go to (I)
X
Go to (5)
(2) Set f = x exp(-x2/18)
Go to (6)
/
(3) Set J = 2 Intl 2 + 6)
Set f = g = [(J + 2)//x]'
(4) Replace g by 1 - g(J + IV/x2
Replace f by I - fJ(J - 1)/x2
Replace J by J - 2
IfJ*0goto(4)
Replace f by f cos(x) - g sin(x) Test values:
Si(n) = 1.85193705
(5) Replace f by f/x Si(0.1) = 0.0999444641
(6) Output f Si(-20) -1.54824170
f=Si(x)< K««
The algorithm above permits Si(x) to be calculated to 24-bit precision (i.e., the relative error never exceeds 6
X 10-') for any argument. For values of x in the range -0.14 < x < 0.14. the approximation formula 38:9:1 is
used. When j lies in the 0.14 S 14 range, the algorithm uses expansion 38:6:4 in the concatenated and
truncated form
/((((( (( l -x2
l I -x2 1 -x2
38:8:1 Si(x) =\\\\ "\\(J+2)'J-1/J(J+1)+J-3/(J-2)(J-1)
+
-x' \
+. +3/4x5+1/2x31(x)
1
+ 12 - cos - sin(x)
1
x
x2 1
38:8:2 SO) =
W
-
(((
(((
s
+1)J(J-1)+1
-s2
4x3+1\2x Icos(x)
(J - 2)(J - 3)
-x
Go to (6 or 5)
(2) Setf = 1x2 exp(-x'/24)1/4
Go to (6 or 5)
(3) Set J = 2 1ntc2 +
Self = g = 1(J + 2)/-x12
(4) Replace g by I - g(J + I )J/x2
Replace j by l - fJ(J - I)/x2 Test values:
Replace J by J - 2 Cin(ir) = 1.64827764
If J * 0 go to (4) Cin(0.l) = 0.00249895856
sin(x) cos() Cin(20) = 3.5285212
Replace j by f - g X2 Ci(2) = 0.422980829
x
Goto(5or6) Ci(-0.15) = -1.3'-552405
(5) Replace f by ln(1.781072418k1) - f 0(20) = 0.0444199210
(6) Output f
f = Cin(x)
or Ci(.r)
When the alternate commands shown in green are substituted, the algorithm generates values of Ci(x) instead of
Cin(x). However, the fractional error in the calculated values of Ci(x) may exceed 6 x 10-1, especially near the
zeros of the cosine integral.
Also, values of Shi(x), Chin(x), Si(x) and Cin(x) are calculable via the universal hypergeometric algorithm of
Section 18:14.
38:9 SINE AND COSINE INTEGRALS 368
38:9 APPROXIMATIONS
Close to x = 0, the formulas
38:9:4 Ci(x) =
x sin(x)
x-+2
- x'+6
cos(x)
large x
may be used.
d sinh(bx + c)
38:10:1
dx
Shi(bx + c= x + c b
d cosh(bx + c
38:10:2 Chi(br + c) =
dx x + c b)
d cosh(bx + c) - 1
38:10:3 Chin(bx + c) =
dx x + (c/b)
d sin(bx + c_ b
38:10:4 Si(bx + c= sinc(irbx + ac)
dx x + (c/b)
d cos(bx + c
38:10:5 Ci(bx + c) =
dx x + (c/b)
d 1 - cos(bx + c)
38:10:6 Cin(bx + c)\=
dx x + (c/b)
r cosh(bx6+ c) - 1
38:10:7 Shi(bt + c)dt = I x + bc) Shi(bx + c) _
/
c`
/ sinh(bx + c)
38:10:8 Chi(bt + c )d<=1x+blChi(bx+c)- b
Jc
sinh(bx + c)
38:10:9 Chin(bt + c)dt = I x + Chin(bx + c)) - b
J/b \/
1 - co bbx + c)
38:10:10 Si(bt + c)dt = I x + b) Si(bx + c) -
c/b
369 SINE AND COSINE INTEGRALS 38:11
sin(bb + c)
38:10:11 c Ci(bt + c)dt = (x + b) Ci(bx + c) -
/b
sin(bxb + c)
38:10:12 Cin(bt + c)dt = I x + b) [Cin(bx + c) - 1] +
Other indefinite integrals are listed by Gradshteyn and Ryzhik [Section 5.3]. The most useful of these arc probably
the integrals of products of sinusoidal functions with the sine and cosine integrals. We have the results
+ sin C -
Bc
b
Si(+) - Si(-)
2B
- cos C --
Bc Ci(+) - Ci(-)
2B
r Bcb/l Si(+) + Si(-)
38:10:15 sin(Bt + C) Ci(bt + c)dt = cos(Bx + C) Ci(br + c) + sin(C -
J B
-lC-bcos(
Ci(+) + Ci(-)
2B
r -I / Bc Si(+) + Si(-)
38:10:16
``
cos(Bt + C) Ci(bt - c)dt = 8 sin(Bx + C) Ci(bx + c) + cost C - b
J
Bc Ci(+) + Ci(-)
-sin C -
b 2B
where Ci(±) is an abbreviation for Ci[(B t b)(x + c/b)i and Si(±) similarly abbreviates Si[(B t b)(x + c/b)) -
(w/2) sgn(B -t b).
Definite integrals of the sine and cosine integral functions include
With argument x + iy. values of the functions may be calculated from those of the Chapter 37 functions via
the identities
1
38:11:3 Chi(x + iv) _ [Ei(x + iy) + Ei(-x - iy)]
zr 1
38:11:5 Si(x + iy) = + - [Ei(y - ix) - Ei(-y + ix)]
Z
Ci(x+iy)=2[Ei(y-ix)+Ei(-y+ix))
1
38:11:6
1
38:11:7 Cin(x + iv) = 2 [Ein(y - ix) - Ein(-r + ix)]
i-i
38:11:11 Ci(iv) = Chi(y) +
2
38:12 GENERALIZATIONS
The sine integral is the special a = 0 instance of the more general function
38:12:1
sin( a + t')dt
a'+1'
-I sin(t)dt
f -a'
(see Erddlyi, Magnus, Oberhettinger and Tricomi. Higher Transcendental Functions, Volume 2, page 147].
The Shi, Chin, Si and Cin functions are all instances of hypergeometric functions, as discussed in Section
18:14.
Boehmer integrals [Section 39:12) also generalize the sine and cosine integrals. We have
and
a
38:12:3 Si(x) = Z - S(x;0)
38:13:1
and
f sin(:)
fi(x)J t+xdt -JO r+1
_ rr exp(-xt) dt
371 SINE AND COSINE INTEGRALS 38:13
cos(t) r exp(-xt)
38:13:2 gi(x)
I I
(t + x)
dt
I
t2+ I
dt
There appears to be no standard notation. Abramowitz and Stegun [Section 5.21 use the f(x) and g(x) symbolism.
Figure 38-3 maps the two functions, which are defined for positive real argument only.
and
r l
38:13:4 gi(x) = sin(x)f - Si(z)J - cos(x) Ci(x)
to the sine and cosine integrals. The simple asymptotic formulas, valid for large argument
X-'
6
- ;
r
+ xs
120
-e
5040
x
+ ... +
(2j + I)!
x'(-x')'
+ x -
permit convenient calculation of the auxiliary functions and are incorporated into the hypergeometric representation
[Section 18:141.
The differential identities
d
38:13:7 r(.r) = -80)
dx
38:13 SINE AND COSINE INTEGRALS 372
and
d I
38:13:8 gi(x) = fi(x) -
dx x
relate the two auxiliary integrals, which therefore satisfy the differential equations
df + f = - 1
and
39
THE FRESNEL INTEGRALS S(x) AND C(x)
Among other applications, Fresnel integrals occur in theories of physical optics [see Section 39:141. The so-called
auxiliary Fresnel integrals, discussed in Section 39:13, are also of some importance.
39:1 NOTATION
The symbols S(x) and C(.r) are almost universally adopted for the Fresnel sine-integral and Fresnel cosine-integral,
respectively. However, there is a wide variation in the definitions chosen for these functions by various authors.
Thus, in addition to our definition 39:3:1, each of the following integrals is cited as the "Fresnel sine-integral" by
at least one authority:
39:1:2 dt=S(V.r)
V2a Vt
Ir r sin(t)
39:1:4 dt = S(-V,) x}0
2Vx o Vt VY
The right-hand member of each of the above equations is the expression for the left-hand integral in the notation
of this Atlas. In all cases the definitions adopted for the Fresnel cosine-integral involves straightforward replacement
of sin in the integrand by cos. Some authors adopt multiple definitions of these functions and utilize subscripts (as
in S,(x). C:(x), etc.) to distinguish the options. There is, however, no greater unanimity in this secondary notation
than in the primary.
The reader will note that the definitions of S(x) and C(x) adopted in this Atlas agree with those used by Grad-
shteyn and Ryzhik, and by Spiegel, but differ from those selected by Abramowitz and Stegun. and by a number
of other authors.
373
39:2 THE FRESNEL INTEGRALS S(x) AND C(x) 374
39:2 BEHAVIOR
Figure 39-1 maps the behavior of the two functions. Both S(x) and C(x) display damped oscillations that decrease
in both period and amplitude as x - is approaching the values ±i in these limits.
.....................................................0.8
- C (x)
............ ....... ..
.... ............... .
:Free (x).
..... .... .............:....:....... ......0.1
Zras (Z)
0
:........... . .... . :
For x ? 0, the S(x) function lies entirely between the curves ; - Fres(x) - Gres(x) and ; + Fres(x) + Gres(x)
where Fres and Gres are the functions discussed in Section 39:13. The same bounds apply for C(x). and a similar
relationship holds for x s 0. For all arguments
!r2+
39:2:1 sgn(x) - Fres(jxI) - Gres(kf) <_ f(x) <_ !9! Gres(jxJ) f= S or C
2
39:3 DEFINITIONS
The Fresnel sine-integral is defined by either of the equivalent representations
sgn(x) I sin(r)
39:3:1 S(x) = ? sin(r)dl = dt
IT Js V2,r o Vt
375 THE FRESNEL INTEGRALS S(x) AND C(x) 39:6
39:5 INTRARELATIONSIIIPS
Both Fresnel integrals are odd functions
39:5:1 f(-x) = -f(x) f= S or C
They may be expressed in terms of the auxiliary Fresncl integrals (Section 39:13] by
I
39:5:2 S(x) = sgn(x) - cos(x') sin(x2) Gres(lxl)
2
and
39:6 EXPANSIONS
Power series expansions of the Fresnel integrals are
4 +-
j
2 .r r s
39:6:1 S(x)
ir 3 42 1320
and
39:6:2 C(x) =
,2 xs
x - - + - -
x° x (_x')
\ 10 216 / a (4j + 1)(2j)'
More rapidly convergent than these are the composite expansions
x'
39:6:3 - S(x) +
2
sin(.r2) + 2x' cos(_r')
4.r3 4x
3
+24x3- - 1120
- + 73920
x'1
- _
-33
4.r (4j - 1)(2j + I)!
and
39:6:4 - C(x) +
2
cos(x) - 2r= sin(x2)
4x3
_---+--
4x3
I
+
3.r
8
xs
160
x"
8640
_
-3 (-x')'
4.r3,=o (4j - 3)(2j)!
Via relationships 39:5:2 and 39:5:3, one can utilize the various expansions of the auxiliary Fresnel integrals
[Section 39:13] to produce the summations
39:7 THE FRESNEL INTEGRALS Six) AND Clx) 376
1
39:6:5 S(x) = xV [sln(X2) (-4x2')' - 2x2cos(x') (-4x2`)
-o (4j + 1)!! (4j + 3)!!
and
39:6:7 S(x) -
sgn(x) - cos(x') r l - 3 + 105 +
(4j )!! + ...l
2 xV L 4x2 16x2 (-4x')' J
and
39:6:8 C(xl-
sgn(x)
2
+
sin(x2) r
x
ILI -
3
Z;
+--...+ (4j - 1)!!
105
16x"
i +... (-4x2 )1
1
cos(x2) 1
1 --+--...+ (4j + 1)!! +...
15
x2°
945 1
xx
The Fresnel integrals are expansible in terms of spherical Bessel functions [see Section 32:131
Included in the Table 39.7.1 are the values at which the Fresnel integrals exhibit maxima, minima and points of
inflection. The functions also inflect at the origin x = 0. In this table n represents any positive integer, n = 1, 2,
3, ....
Table 39.7.1
As x - ±z, the values of the maxima and minima converge towards ±= according to the formulas
2/ minimum
FresV2na - 2
1
39:7:4 C( 2na - J= ± maximum
2 ±2 /
where the Fres function is the auxiliary Fresnel cosine-integral discussed in Section 39:13. The values at the points
of inflection converge more rapidly towards ±j' and arc easily calculated from
39:7:5
44 ma-2)J
m= 1,2.3,...
C(±V) = ± 2+(_lrGres(V rt)
1
39:7:6
In Section 18:14 we show how the universal hypergeometric algorithm may be utilized to generate values of the
Fresnel integrals (as well as those of the auxiliary Fresnel integrals) in several different ways.
Because applications of Fresnel integrals usually require values of both S(x) and C(x), the algorithm presented
below generates numerical values of the Fresnel sine-integral and the Fresnel cosine-integral simultaneously. The
algorithm has 24-bit precision (i.e., the relative error never exceeds 6 x 10"). For very small absolute values of
x. approximations 39:9:1 and 39:9:2 are used by the algorithm. For intermediate values of W, the composite ex-
pansions 39:6:3 and 39:6:4 are used in the truncated and concatenated forms.
39:8:1
SW ((((".(((2J-1)(J+1)J+2J1 5J (J-I)(J-2)+2J` 9/(!-3)(J-4)
1 -x4 I -x4 I -3 sin(x-) 1
cos(x-)1
7 5x4 3 3x2 -1 2 2x- /x 2a
-x4 _xe I -x'
C(.c)((r
1
39:8:2
.. (2J-3)J(J-1)+2J-7J (J-2)(J-3)+2J-11/(1-4)(J-5)
+ +
1 -x + I a
-x + a I _
-3 - 2x- sin(x-) - COs(t) I
5 4x3 1 2x 1 -3 1 1) x'V8n
39:8 THE FRESNEL INTEGRALS S(x) AND C(x) 378
with J given the empirical value 2 Int(5 + &2/5). Similarly, the concatenated expansions
-x II -x 1
39:8:4 1` 1\ -x'
X 12
+ I)
-' x
+ 1 JI
-x
X2
+...+I1 -x, +1/ X. +
are used to evaluate the two asymptotic series in 39:6:7 and 39:6:8, and these latter are then employed to evaluate
S(x) and C(x) for large values of jx4.
(4) Replace f by I - f I J
4\)\
g cos(x2)
Replace g by + J +
Test values:
(5) Replace f by f/V 2 S(0.2) = 0.00212744901
Output f C(0.2) = 0.159551382
f = S(x) < S(-n) = -0.616486872
Replace g by g/V 2 C(-,r) = -0.451358120
Output g S(5) = 0.421217048
g - C(x) < C(5) = 0.487879892
This algorithm is readily adapted to generate values of the auxiliary integrals Fres(x) and Gres(x) [see Section
39:13].
39:9 APPROXIMATIONS
For arguments of small magnitude one has
The approximation
[C(X)
r 1= I2 1
39:9:5 S(x) - 2] + - 2] = tax' 8-bit precision x 2:3
shows that, for a small range of x values greater than 3, a cartesian graph of the Fresnel sine-integral versus the
Fresnel cosine-integral is roughly a circle, centered at the point (,) and with radius I/xV 2a. The precise shape
of such a graph is discussed in Section 39:14.
d
39:10:1 S(bx) = b
- sin(b-x2)
d
39:10:2 C(bx) = b -ITcos(b-.,6
dr
I - cos(b'x2)
39:10:3 I S(br)dt = x S(bx) -
Jo 6 2n
Among the definite integrals listed by Gra((dshteynl and(Ryzhik [Section 6.32) are
1 r 1 /
39:10:5 I ti -S(br)Jdt= -I <v<2 b>0
0 0 2 V 2w(1 + v)b'-'
f\ 2 1 sin(-)
39:10:6
Jo tC(bt)-1Jdr=
1 2 \(l + v)b"
-1<v<2 b>0
and
r a
39:10:7 3 sin(Bt'') S(bt)dt cos(Br') C(bt)dt =
0<B <b'
32B
J0 0
0 B > b:
--i-l
and
39:11:2 C(x+iy)=
1-i erf x-S
4 ti +i- +-e
x+y I+i
4
x+y
V2
x-y
V2/
to the error function of complex argument [Section 40:11).
39:12 GENERALIZATIONS
Fresnel integrals may be generalized to Boehmer integrals, also known as generalized Fresnel integrals, and defined
by
it
39:12:3 S(x:0) = Z - Si(x)
and reduction to the Fresnel integrals occurs when the parameter is a moiety:
1
39:12:7 S(x v) =-
C(X,v + 1)
V
- x' sin(x)
V
S(x;v + 1) x" cos(x)
39:12:8 C(x;v)
V V
permit any indefinite integral of t-"12sin(t) or t-"ncos(t) to be evaluated for n = 1, 2. 3. ... in terms of the special
cases 39:12:3-39:12:6. Boehmer integrals have the particular values
var
39:12:10 S(0;v) = ['(v) sin(
2
at zero argument.
Convergent expansions in power series
(I - v). 1 r (2 - v)(3 - v)
- + + + (I - v)x" 2 sin(x) L I -
X-1
(1 - v) 1 2 r (2 - v)(3 - v)
(-x2) x-
(2-v)(3-v)(4-v)(5-v) (2-v);,
x' (-x2)'
valid as x x, hold for the Boehmer integrals. These expansions permit the numerical evaluation of S(x;v) and
C(x;v): for example, via the hypergeometric algorithm of Section 18:14.
or as the nonperiodic component of the semiintegral and semiderivative, respectively, of the sine function
d-v2
39:13:3 sin(x) = sinx 4d + \ Fres(f)
v2 f
39:13:4 &T5 sin(x) =sin l\x +4-
with zero lower limit. Their relationships to the ordinary Fresnel integrals
39:13:8 -
1r
sin(x2) + cos(x2)
1
- Gres(x)
rr ILx
4.r' 16x°
V (-4,e)'
2 Va 15 + 945 - a,=o (4j + 1)!!
39:13:9
39:13:10
Fres(x) --
Gres(x)-
x 1
1
3 + 105
-+
15
16x°
945
+ (4j - 1)!! +
+
(-4x')'
(4j + I)!!
4x` 16x' (-4x`)1
Fresnel integrals play a paramount role in the theory of the diffraction of light. Indeed, it was his studies in physical
optics that led Augustin Fresnel to construct the integrals that now bear his name.
A very useful construct in diffraction theory is Cornu's spiral, also known as the clotoid curve. This double
spiral is defined parametrically by
0
10,
1 IO0 h a 'b ti
a %
ti 'b a h m A 0
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
FIG 39-2:
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
........................................................... ............ ...... 0.2
S(r). x-C(w)
...... :.... :.... :.... :.... :.... :.... :.... :.... :.... :.... :.... :.... :.... :...... -0.3
:.... :.... :.... :.... ..... ..... ..... .... .... .. -0.5
dzf
dx2
39:14:2 T/2
I l + (dx/
39:14:3
J0 1+
dt) dt
between the origin 0 and point P. These two quantities are, in fact, w y tar and wv 2%a, respectively, where w
is the paramctcr dcfincd in 39:14:1.
CHAPTER
40
THE ERROR FUNCTION erf(x)
AND ITS COMPLEMENT erfc(x)
and occur widely in problems of heat conduction and similar instances of the diffusion of matter or energy. The
name of the function arises from its importance in probability theory [see Section 40:141.
40:1 NOTATION
The function erf(x) is also known as the probability integral and is sometimes denoted H(x) or 4)(x). The related
notations (b1(x), 4);(x), etc., then denote successive derivatives of the error function
40:2 BEHAVIOR
Figure 40-1 includes maps of erf(x) and erfc(x). both of which are sigmoidal functions that rapidly approach limits
as x -, _w. These limits are ± I for erf(x) and 0 or 2 for crfc(x).
40:3 DEFINITIONS
The most useful definitions of these functions are as the indefinite integrals
sgn(x) f ezp(-t)
40:3:1 erf(.r) _ - J exp(-t')dt = J dt
ao Vtr o Vr
395
40:4 THE ERROR FUNCTION erf(x) AND ITS COMPLEMENT erfc(x) 386
............................................... 1.2
and
40:5 INTRARELATIONSIIIPS
The error function is an odd function
40:5:1 erf(-x) = -erf(x)
while its complement obeys the reflection formula
40:5:2 erfc(-x) = 2 - erfc(x)
40:6 EXPANSIONS
The error function may be expanded in many ways, including
Table 40.7.1
x e 0 x a 0.476936276
erf(x) -1 0 0.682689492 1
erfc(x) 2 1 0.317310508 0
40:9 THE ERROR FUNCTION erf(x) AND ITS COMPLEMENT erfc(x) 388
40:9 APPROXIMATIONS
For small arguments, one can use
3.372x
-0.96 < x s 0.96
40:9:1 erf(x)
+3x 8-bit precision
Differentiation gives
d" d" 2
40:10:2 -axe erf(bx) = erfc(bx) (-b)" n = 1, 2, 3, ...
dx" y'a
Indefinite integration gives
where the icrfe function is discussed in Section 40:13. The same section discusses repeated integrals of the error
function complement.
Of widespread occurrence are indefinite and definite integrals of the product of an exponential function with
an error function or complementary error function. Such integrals will be found in Section 41:10 or may be evaluated
via the table of Laplace transforms in Section 26:14. Other definite integrals are listed by Gradshteyn and Ryzhik
[Sections 6.28-6.311 and include
rl 1 + -)
40:10:5 J t` erfc(bt)dt = \ b>0 v>-I
o (l+V) b
1 - expp ( Bz 14b)
z
VaI
in terms of four integrals that are not evaluable in simpler terms. When the real part of the argument is either zero
or infinity, however, reduction occurs to the simple results
2i
40:11:2 erf(iv) _ - exp(v') daw(y)
Vii
40:11:3 erf(- + iy) = I - i erf(y)
where daw is Dawson's integral, discussed in Chapter 42.
Confusingly, the function generally known as the error function of complex argument, and denoted by the
symbolism W(x + iy), is not that given by 40:11:1 but is instead defined by equation 41:11:1 of the next chapter.
40:12 GENERALIZATIONS
The error function and its complement generalize to the incomplete gamma functions of Chapter 45, of which they
are the v = = cases
40:13 THE ERROR FUNCTION erf(x) AND ITS COMPLEMENT erfc(x) 390
2
40:13:4 i 'erfc(x) _ exp(-x2)
The ierfc and i"erfc functions are known as the complementary error function integral and the repeated integrals
of the error function complement, respectively. The behavior of some of these functions is shown in Figure 40-1.
Because values for x < 0 are seldom of importance in practical applications, these are excluded from the diagram.
The power series expansion
1 (-24'
40:13:5 i'erfc(x) _
1
2 1-0;!r(l +
2,)
shows that the functions have the particular values
1
40:13:6 ierfc(0) =
2T(I
+ 2/
of which the first few are shown in Table 40.13.1. For n a 1 i'erf/c(x) approaches - as x -o -a and approaches
zero asymptotically as x -+ « according to the asymptotic expansion
391 THE ERROR FUNCTION erf(x) AND ITS COMPLEMENT erfc(x) 40:13
Table 40.13.1
M i'erfoo)
and
it being the latter property that explains the name of the function family. It is possible to generalize the family to
i'erfc(x), where v is not necessarily an integer. by using procedures of the fractional calculus [see Oldham and
Spanier. Section 10.5]. The differential equation
40:13:14 a
d-+x--nf=0
2f df a>0
dx' dx
is solved by f = c, i'erfe(x/V 2a) + c2 i"erfc(-x/V 2a), where c, and c, are arbitrary constants.
A useful definite integral is
b
40:13:15 exp(Bt)i"erfe(bt + c)di = i"erfc(c) + exp(Bt)i"-'erfe(bt + c)dt
f B 8
- u
40:14 THE ERROR FUNCTION erf(x) AND ITS COMPLEMENT erfc(x) 392
40:14:1 f(x) = 1 e (x - 10
ay'2ar\ 2a'
in terms of the mean µ and variance o= of thedistribution, the corresponding cumulative function being
40:14:3 f(x) ex P(
ti 2i 2
and
40:14:5 x, +
//N\ //2w
2az InI nf coal N n," I
``
where N is a large integer and n, and ni. , are each random integers in the range 0 < n s N. Knuth recommends
N = 199017 and the formula
40:14:6 n,., = 1 + (24298n, + 99991) mod(199017) j = 0, 1, 2, ...
as a means of creating successive pseudorandom integers. The "seed" no is selected arbitrarily in the range 0 <
no s 199017, and the integers n, are confined to the same range.
An algorithm follows for generating a sequence of J successive numbers that are normally distributed about a
mean s with a variance ox. It uses formulas 40:14:5 and 40:14:6, taking 2J as the seed.
Although the functions considered in this chapter are composites of those discussed in Chapters 12, 26 and 40,
their practical importance warrants a separate treatment. Moreover, many of the properties of these functions tran-
scend those of their component factors.
Primarily, this chapter is devoted to the following functions, each of which involves the product of an expo-
nential function and an error function complement:
41:0:1 exp(x) erfc(V)
41:0:2 exp(x) erfc(-V) = 2 exp(x) - exp(x) erfc(\)
41:0:3 \ rrxexp(x)erfe(f)
and
41:1 NOTATION
Symbol variants such as (2/V)e Erfc(P'') and [I - 4r(V )] cxp(x) occur, as explained in Sections 12:1, 26:1
and 40:1. In addition, a unitary notation is sometimes adopted; for example, erc(x) and eerfc(x) have both been
used as abbreviations for exp(xr) erfc(x).
41:2 BEHAVIOR
Most of the functions of this chapter are defined as real functions only for arguments x ? 0, and this is the only
range shown in Figure 41-1.
The function exp(x) increases rapidly in magnitude as x increases, whereas erfc(Vx) suffers a dramatic decrease.
The competition between these two effects causes exp(x) erfc(Vx) to diminish in a very leisurely fashion toward
395
41:3 THE exp(x) erfc(f) AND RELATED FUNCTIONS 396
O'L O
R
OOCO
0 IV 11 Y 'll
...................:........................:....... 1.0
FIG 41-1 :
zero as x -. x, whereas \x exp(x) erfc() slowly approaches a unity asymptote in the same limit. These trends
are mapped in Figure 41-1.
No such compensation influences the functions exp(x) erfc(- V) or (I / V) + ex,Lx) erfc(- V x-), both of
which increase exponentially as x - x as depicted in Figure 41-2. The function (1/Vex) + exp(x) erfc(-V)
displays a minimum value of about 3.1 close to x = 0.20. The same argument corresponds to a point of inflection
(Section 0:71 in exp(x) erfc(-f), as reported in Section 41:7.
41:3 DEFINITIONS
The exp(x) erfc(V) function may be defined as the indefinite integral
1
z exp(x - t)dt 2
41:3:1 exp(x) erfc(V) _ cxp(x - ts)dt
eJ "7
or via the definite integrals
2
t+x
2\/X
if o
exp(-tt)dt
tr+x
41:3:3 exp(x) erfc(\) -
I exp(-,2 - 2tf )dt
Jo
and
x exp(-xt)dt I exp(-xt)
41:3:4 exp(x) erfc(\)
Jo r+1 A Jo (f + 1)
The integrals in 41:3:4 may be regarded as Laplace transforms [Section 26:14], in applications of which the func-
tions of this chapter play a distinguished role.
397 THE exp(x) erfc(V) AND RELATED FUNCTIONS 41:3
Functions 41:0:1-41:0:5 all satisfy rather simple first order differential equations, namely
df
41:3:5 - f=: +1 f = cxp(x)[crfc(--V) + cl
dx Trx
41:3:6 - -1 f = V exp(x)[erfc(Vx) + c)
dx \ 1 + 2x/f =
y O y
iP tt t
FIG 41-2!:! 1 :
41:4 THE exp(x) erfc(f) AND RELATED FUNCTIONS 398
41:3:7
df f- -1
f= + exp(x) [erfc(- V) + c]
dx 2 ax'
where c is an arbitrary constant. Thus, the various functions may be defined as the special c = 0 cases of the
solutions to these differential equations.
The fractional calculus may be employed to define some of the functions of this chapter. Thus, with lower
limit zero, semiintegration of the exponential function gives
d- 1/2
41:3:9 dx_j/, exp(x) = exp(x) erf(V) = exp(x) - exp(x) erfc(f)
41:5 INTRARELATIONSHIPS
The reflection formula
41:5:1 exp(x) erfc(- V) = 2 exp(x) - exp(x) erfc(f)
applies. The addition formula
exp(y) , (2j - 1)!! y(j + I;y)
41:5:2 ex (x + y) erfc( x +
41:6 EXPANSIONS
A power series for exp(x) erfc(±V)
4 ' x= VXY
41:6:1 +x+-
V
X 3 a 2
'-0r 1+?
involves the gamma function [Chapter 43]. Alternate terms may be summed separately:
xz r 2r 4x2
41:6:2 exp(x) erfc(±Vx) = 1 + x + Z +r 2 a l+ + 15 + J
3
x' +T (2x)i
= exp(x) + exp(x) erf(1rx)
- j' + 2 V/ n (2j + 1)! i
Of course, replacement of f by x in 41:6:1 gives the power series expansion of exp(x2) erfe(±x).
The exp(x) erfc(f) function may also be expanded as a continued fraction:
399 THE cxp(x) erfc(Vx) AND RELATED FUNCTIONS 41:8
2/n 1 2 3 4
41:6:3 exp(x)
V2x+ V2-x+ V -2x+ V2r+
Truncated versions of the asymptotic series
1 3 (2j - 1)!!
41:6:4 Irx exp(x) 1- + + + ... X --O >o
2x 4xZ (-2xY
are useful for large x. The truncation error is less in absolute value than the first neglected term and of the same
sign.
It follows f r o m 41:6:2 that the function (l / V nx) + exp(x) erfc(- Vx) is expansible as the series
I - I fz rV-ulr
11 F j+Il
2
1 (2x)
= exp() +
V rcx =u (2j-1)!!
with uniformly positive terms.
which is based on expansion 40:6:1, truncated with J = 3 + 1nt(9V). For x > 2.25 a continued fraction expansion
analogous to 40:8:1 is utilized.
Replace x by -x
Set g = 2 ex[(-v)
(I) If 1.5<Vxgo to (3)
Set i = 3 + Int(9V )
Setf= 1
(2) Replace f by I + fx('12 - j)/[i('/2 + j)] Input restrictions: When x 2: 0 is input, exp(x)
Replace j by j - I erfc(Vx)is
V_x) output. When -x 5 0 is input. exp(x)
41:9 APPROXIMATIONS
The approximation 40:9:2 is easily adapted to provide an estimate of the exp(x) erfc(V) function for large x. A
more accurate approximation, valid for all arguments, is
2
41:9:1 V,-rx exp(x) erfc(V) 10-bit precision
Cl + x exp( 5x/7)/
+ +
arx
41:10:1
d
exp(bx) erfc(-- Vbx) = b exp(bx) erfc(±V )
V
dx ax
it
41:10:2 exp(bx) erfc(V') = (bx + '/2) exp(bx) erfc(V ) - Vb
41:10:3
d
dc \
I
irbx
+ exp(bx) erfc(-))J = ax
+ b exp(bx) erfc(- Nlrb-x) -
2
1
bx
d b
41:10:4 exp(axs) erfc(bx) = 2 exp(ax')[ax erfc(bx) -
& bnexp(-b-x')
401 THE exp(x) erfc(f) AND RELATED FUNCTIONS 41:10
Repeated differentiation of exp(x) erfc(±f) regenerates the original function together with a finite series of al-
gebraic terms
41:10:5
d"
dx-" exp(x)erfc(±f)=exp(x)erfc(-f)+
I I
2V ax 4
3
-
V ax
1 (2j - I)!!
= exp(x) erfc(±f)
=o (-2x)i
The series is identical with the leading members of the asymptotic expansion 41:6:4.
Indefinite integrals include
41:10:6
r
, exp(-b212) erfc(
+
br)dt = 4b [1 - erfc'-(±bx)]
J0
- - cxp(46: - bc c-
2b
f- J
2b J
/
41:10:8
(
J exp(-Br) erfc
\
-
,VYJ t/Idr =
exp(2V aB)
2B erfc Yx + x )+ exp(-2 aB)
erfc,\ x - /Bx
0
ex Bx)
erfcl x) B>0 a>0 x>0
B
41:10:12
o
r" exp(at2) erfc(btk(t =
2V
V ar(1 + v)b""
/
\
-
Frv+ I
2
I /(b-B+
v+2 v+3
2 '-2
a
' b2
b-bB)
l b2 > a
B <0
v > -1
d'/2 1
41:10:15 exp(bx) Vbx) _ 'ffX z V exp(bx) erfc(±)
and
d-1/2 +1
41:10:16 dx_in
exp(bx) erfc(±\, s) _
7 (1 - exp(bx) erfc(±V)]
41:11 COMPLEX ARGUMENT
The so-called error function of complex argument is. in fact, the function exp(x) erfe(-\) with x replaced by
-(x + iv)2. It is defined by
rr 2i
41:11:1 W(z) = exp(-22)1 1 + - J exp(r')dt = exp(-:) erfc(-:z) . = x + iy
o I
This is an important complex-valued function, the real and imaginary portions of which arise in practical problems.
Accordingly, its values have been tabulated (for example by Abramowitz and Stegun, Table 7.9). or, alternatively,
numerical values may be calculated via expansions 41:6:2 and 41:6:5. which remain valid when x is replaced by
-(x + w2-
The real and imaginary parts of W(x + iy), defined by
41:11:2 W(x + iv) = Re{W(x + iv)} + i lm{W(x + iy)}
are expressible as the integrals
exp(-t2)dt
41:11:3 Re{W(x T ry)} _ (x - t)2 + Y2
7F
and
I (x - r) exp(-t2)dt
41:11:4 Im{W(x + iv )} _ -
tr (x-t)2+y1
When the argument of the W function is real (i.e., y = 0), the function has real and imaginary parts given by
2i
41:11:5 W(x) = exp(-x2) + daw(x)
but when the argument is purely imaginary (i.e., x = 0), the function is real
41:11:6 W(iy) = exp(y2) erfc(v)
403 THE exp(x) erfc(V) AND RELATED FUNCTIONS 41:13
41:12 GENERALIZATIONS
The exp(x) erf(V) function is a special case of the Kummer function [Chapter 471
a I a
41:12:1 M(1 ;1;x) _ - jX exp(x) erf(f) = [exp(x) - exp(x) erfc(\)1
2Y
The asymptotic representation 41:6:4 of the \ exp(x) erfc(Vx) function generalizes to the simple hyper-
geometric function [Section 18:14]
(a),
41:12:2
(-x)'
of which it is the a = special case.
and
In addition, explicit reference to the cxp(x) erf(f) function is made in equations 41:3:9.41:6:2 and 41:13:1. Figure
17-2 includes a graph of exp(x2) erf(x).
An expansion in terms of hyperbolic Bessel functions [Section 28:13] and particular values of the exponential
polynomial [Section 26:131 is
41:13:3 exp(x) erf(f) _ e,(-1)(V)' 1,.x,2(2\ r)
=o
while a more conventional expansion is
expx c x-
Jx [
J-
V L
I+-+-+.
)x
3
4x2
15 J
--2 -xw,o (2j(2z)'
+ 1)!r
The latter is useful for calculating values of cxp(x) erf(f ); for example via the universal hypergeometric algorithm
of Section 18:14.
CHAPTER
42
DAWSON'S INTEGRAL
The functions of this chapter have much in common with those of the previous chapter. Dawson's integral daw(x)
is related to (V'/2) exp(-x2) erf(x) via the identity
42:0:1 daw(x) =
-iV exp(-x') erf(ix)
2
in much the same way that sin(x) is related to sinh(x) [see 28:11:31.
42:1 NOTATION
Because no standard notation exists, this Atlas uses the symbol daw(x) for Dawson's integral of argument x. Abra-
mowitz and Stegun [Chapter 71 use F(s) and, in the light of relationship 42:0:1, Gradshteyn and Ryzhik adopt a
symbolism equivalent to -i(V /2) exp(-x') erf(ir).
Sometimes the name "Dawson's integral" is given to exp(x2)daw(x) and Erfi(x) or erfi(x) is emplod to denote
this function. A rescaling of the argument often leads to simpler formulas; the function Vi daw(Vx/2) has been
symbolized D(x).
42:2 BEHAVIOR
The functions daw(Vx), daw(x) and x daw(x) are all important and all three are mapped in Figure 42-1. Each of
these functions is zero at x = 0, displays a maximum in the vicinity of x = 1 and then slowly declines, approaching
s asymptotically in the case of x daw(x) and zero in the other two cases.
42:3 DEFINITIONS
The usual definition of Dawson's integral is through the indefinite integral
r sgn(x) `exp(: - xI)dt
42:3:1 daw(x) =
0
r exp(r J- xj)dt =
2
J
o V 't
405
42:4 DAWSON'S INTEGRAL 406
h O h O h O h
ids 44' 44' 1.4' 1.4' .,P : i*
:....:.... .. .. .............: 0.6
xdaw (x) :
.. ......
dart.....' 0.2
FIG 42-1
:.... ..:. ..:....:....:..............:....:....:....:....:....:....:....:....:.0.1
but it may also be defined as an inverse Laplace transform [see integral 42:10:7], in terms of the error function of
complex argument [Section 41:11] or as the semiintegral
42:3:2
42:3:3 of+bxf=c
dx
where b and c are constants. The general solution is
2c Ibs\ bx'
42:3:4 f= b dawl 2 I+ (constant) exp 2
42:5 INTRARELATIONSHIPS
Dawson's integral is an odd function:
42:5:1 daw(-x) = -daw(x)
42:6 EXPANSIONS
The power series expansion
i
42:6:1 daw(x) = x - + 15 - ... = x E (-2x')
-u(2j+ 1)"
407 DAWSON'S INTEGRAL 42:7
42:6:2 daw(V x) nx
i (-x)= V'_ i (-x)'
2 _o r(j + 1) J=O (1)f
or as the continued fraction
= x/2 x 3x Sx
42:6:3 daw(12)
1+ 3-x+ 5-x+ 7-x+
The expansion
'
42:6:4 daw(V x) = exp( x) + -x3+ - +
3 10
_ V exp(-x)
j! (2j + 1)
xs J
may be developed into
x = -x x = -1 x=0 .r = I x = =
The values of the maxima and the inflection of each of the functions are related in a simple way to the cor-
responding arguments:
Maximum Inflection(s)
2x, + I
0.854032657 = xM 1.84365320 = .r;
4Vc
0.924138873 = Vru 1.50197527 - x,
2VxM 2.r - I
X 10 5956772781
.
x' - I
1.50197327 = x,
2r'-I 11.95562141 J 2xt - 3
The particular arguments xN and x which correspond respectively to the maximum and inflection point of the
daw(f) function, are cited in Section 37:14 as the lower limits of important indefinite integrals.
42:8 DAWSON'S INTEGRAL 408
(I (...((
O - n)f n
l z-n + ... + -; -z
+tl
1
+ I)
-x' + 1)
42:8:2 daw(x) = +1 1)
-x= / -x' / -X= -x2 2V
For GxI ? 5, the asymptotic expansion 42:6:6 is employed, being suitably terminated and evaluated by the same
set of commands that are used by 42:8:2.
42:9 APPROXIMATIONS
For small and large arguments, one may use the approximations
and
2x x
42:9:2 daw(x) - +I 8-bit precision k i z 3.9
4x3
respectively.
42:10:2 daw(V) = 2
Repeated derivatives involve Hermite polynomials [Chapter 241
12j 2J+1=1,3,5,
d^ ,_o (-2)1(2j)!
42:10:3 - daw(x) _ (- l)" H^(x) daw(x) -
dx'
2
-Q (-2)'(2j + 1)!
n=2J=2,4,6,...
Indefinite integrals include
x daw(y nx)
42:10:4 -
and
j d aw( )dt=
bx - daw(bx)
42:10:5 J r daw(bt)dt =
o 2b2
The indefinite integral jdaw(t)dt is not expressible in terms of any named function. It is, however, quite a simple
hypergeometric function [see Section 18:141
L(1)J
42:10:6 ; daw(t)dt = x (-x=)'
J0 2
The first three of the definite integrals listed below may be regarded as Laplace transforms [Section 26:141 but
are included here because of the importance of these particular integrals:
(
42:10:8
j daw(bt) exp(-st)dt =
46 exp \4b'1 Ei(_. )
b>0 s>0
With lower limit zero, the semiderivative and semiintegral of daw(Vx) are
due
42:10:12 -jr , daw(\) = V-
2 exp(-x)
and
d-'I2
42:10:13 ue/2
d aw( Vx ) = 2 1 1 - exp( - x)1
42:12 GENERALIZATIONS
Dawson's integral is a special case of the incomplete gamma function [Chapter 45]
The gamma function is unusual in the simplicity of its recurrence properties. It is because of this that the gamma
function (and its special case, the factorial) plays such an important role in the theory of other functions. The
reciprocal 1/r(x) and the logarithm ht(r(x)) are also important and are discussed in this chapter, as is the related
complete beta function B(x.y), which is addressed in Section 43:13.
Formulas involving the gamma function often become simpler when written for argument 1 + x rather than x,
and we have sometimes taken advantage of this fact. Because r(x) = r o + x)/x, a change of argument is readily
achieved.
43:1 NOTATION
The gamma function is also known as Euler's integral of the second kind. r(1 + x) is sometimes symbolized x!
or n(x) and termed the factorial function or pi function, respectively. To avoid possible confusion with the functions
of Chapter 45, r is distinguished as the complete gamma function.
43:2 BEHAVIOR
The behavior of r(x) for -5 < x < 6 is shown on the accompanying map, Figure 43-1; it is complicated. For
positive argument, the gamma function passes through a shallow minimum between x = I and x = 2 and increases
steeply as x = 0 or x = x is approached. On the negative side, r(x) is segmented: it has positive values for -2
<x<-1.-4<x<-3,-6<x<-5,...butnegative values for-1<x<0.-3<x<-2,-5<.x<
-4, ..., with discontinuities at x = 0, -1, -2, .... The gamma function never takes the value zero, but it comes
very close between consecutive large negative integers.
The reciprocal 1/r(x) has no discontinuities. It rapidly approaches zero as x -b x and equals zero at x = 0,
-1. -2, .... As shown on the map, its oscillations become increasingly violent as the argument becomes more
and more negative.
The logarithm In(r(x)) is usually only considered for x > 0. This is the convention adopted in drawing the
map, which shows that ln(r(x)) is a positive function except between x = l and x = 2, where it is briefly and
slightly negative.
411
,
43:3 THE GAMMA FUNCTION r (x) 412
4t
4g0 'L
FIG 43-1
43:3 DEFINITIONS
Although it is restricted to positive arguments, the most useful definition of the gamma function is the Euler integral
The gamma function may be expressed as a definite integral in many ways apart from the one given above.
Gradshteyn and Ryzhik [Section 8.311 give a long list of which the following are representative:
(!
x>-1
1
43:3:4 r(1 +x)= J In'I-Idt
o t
Alternatively, in 43:3:6 sec and cos may be/replaced by csc and sin, respectively.
Likewise, there are several ways of representing the logarithm of the gamma function by means of integrals.
One is
( ` -
43:3:7 In(r(l+x))=I -II
-xJ x>-1
0 I t In(t)
and others may be found in Gradshteyn and Ryzhik [Section 8.34].
or
43:5 INTRARELATIONSHIPS
and
generalize to
43:5:5 r(n+x)=x(1 1 +x)r(x)=(x)"r(x) n=0. 1,2,..
and
43:5:6 r(x - n)
r(x) (- I)" r(x) n=0.1.2....
(x- (1 -x)"
in terms of the Pochhammer polynomial [Chapter 18].
The duplication and triplication formulas
4
43:5:7 r(2x) = r(x) r(# + x)
2V
and
(27).
43:5:8 r(3x) = r(x) r(# + x)r(i + x)
2rm V3
2a n°
43:5:9 r(nx) _ H r(n + zI n = 2. 3. 4... .
n i.n
which applies for any positive integer multiplier n.
From 43:5:5 and 43:5:6, one may derive the expressions
415 THE GAMMA FUNCTION r(x) 43:6
r(n + x)
43:5:10 =(x) n=1,2,3,...
r(x)
and
r(x - n) _ (-1)"
43:5:11 n = 1, 2, 3, ...
r(x) (1 - x)
for the ratio of the gamma functions of two arguments that differ by an integer. These formulas may be used even
when the individual gamma functions are infinite (e.g., r(x - 3)/r(x) -. -6 as x -. 0).
Because of the frequent occurrence of the reciprocal gamma function in power series expansions of transcen-
dental functions, particular values of the latter functions often serve as sums of infinite series of reciprocal gamma
functions. For example, on account of expansion 41:6:5, we have
43:5:12
1
r(s)
+-++
r(1)
1 1
r(3)
1
r( J/2)
1
17 +eerfc(-l)=5.573169664
43:6 EXPANSIONS
The power series expansions for the gamma function and its reciprocal are
where ao = 1 and
43:6:2 a_,
ya.;_, + (=1),
1 J
I
k-o
k )..k j = 1, 2, 3, ... Y= 0.5772156649
Here y denotes Euler's constant [Chapter I I and ;(n) is the n'" zeta number [Chapter 3). Numerical values of ao,
a-,, a_2, .... a_2s are listed by Abramowitz and Stegun [page 256). The power series expansion of the logarithm
of the gamma function is less complicated:
43:6:3 -1<xr= l
12 ,.2 J
More rapidly convergent is the similar series
43:6:6 r(x)
2,rr
exp(-x)x,(1 +
I + 1 - 139 1-
X \ 12x 288x' 51840x
43:7 THE GAMMA FUNCTION F(x) 416
where B_, denotes a Bernoulli number [Chapter 4 1. Though exact only in the x limit, these expansions are
remarkably accurate for modest values of the argument [see 43:9:1, for example]. Relationship 43:6:7 is a special
case of Barnes' asymptotic expansion
2
(-
2ir
\x
+
6c' - 6c + 1
12x
43:6:9
C(z + c)
F(x)
-x` 1+
c(c - 1)
2x
+
c(c - l)(c - 2)(3c - I)
24x'
+
c2(c - 1)2(c - 2)(c -3)
48x3
+ x-+x
for the ratio of two gamma functions of large, but not very different, arguments. For example:
43:6:10
V;F(x)
-1-X+-+---- + X=--+O
F(x + 1) X2
2
5X3
2
21X'
8
399X`
8
1
8x
-
s
X -3 -2 -1 0 1 2 3 4
-8Va 4V' _
f (x) t: IS
ws
3
2x -2Vr z- VT I
V'n
2
I
3V+r
4
2
15Vn
8
6
The arguments yielding local maxima or minima of r(x) correspond to the zeros of the digamma function [see
Section 44:7].
Input x >>
(1) Ifx=0goto(2)
Replace g by gx
Replace x by x + I
Ifx<3rrgoto(I)
\1
Setf =ll-?7x' II-
\
230x //
3x'1/
1-f
LLLLL
Replace f by
?x + x[ln(x) - 1]
exp(f) Test values:
Replace f by 1'(0.6) = 1.48919225
8 ILX r(7.4) = 1541.33619
(2) Output f R-4.2) = -0.164061051
f = r(x) <
43:9 APPROXIMATIONS
From 43:6:5. the Stirling approximation
+ 12x
-I 1
an expression derived from 43:6:9 that permits the gamma function to be expressed in terms of the factorial function
[Chapter 2].
Close to its zeros, the reciprocal gamma function is approximated by
1
43:9:3 = (-1)"n!(x + n)[ 1 - (x + n)ti(n + I )] x = -n = 0, - 1. -2. .. .
r(x)
The gamma function with complex argument has real and imaginary components:
1j + x1
43:11:1 r(x + iy) = [cos(O) + i sin(0)1Ir(x)I
i-o yr+(j+x)2
Y
where 0 = V40) + - arctanl( ,
,-oj+x \j+x
Here dj is the digamma function [Chapter 44]. Tables from which r(x + iv) may be evaluated are given by Abra-
mowitz and Stegun (pages 277-2871. It follows from 43:11:1 that the product r(x + iv)r(x - iv) is real; special
cases are
43:11:2 r(I + iy)r(1 - iv) = pry csch(arv)
and
43:11:3 r(12 + iy)r( - iv) = ir
For purely imaginary argument
IT Y
43:11:4 r(iv) = 1 / [sin(6) - i cos(9)] e = -yy + - arctanl I
43:12 GENERALIZATIONS
The gamma function is a special case of the incomplete gamma function -y(v;x)
43:12:1 r(v) = y(v;x)
both of which are discussed in Chapter 45. Moreover, these two functions sum to give the gamma function
43:12:3 y(v;x) + r(v;x) = r(v)
for all values of X.
The functions discussed in Chapters 2, 6. 18. 44 and 45 are all closely related to the gamma function.
Another important function that is intimately related to the gamma function is the complete beta function B(x,y).
Also known as Eider's integral of the first kind or simply as the beta function, it is defined by the Euler integral
419 THE GAMMA FUNCTION r(x) 43:11
= r,
43:13:1 B(x,Y) t'-'(1 - t)'-'d: x>0 y>0
J0
43:13:4 B(x,y) = 2
f r"-'(I - t2)'`'dt = 2
0
cosr*-'(t)dt
and extensive lists may be found in Gradshteyn and Ryzhik [Section 8.38J and in Magnus, Oberhettinger and Soni
x> 0 >0
[page 71. The integral representations of B(.r.y) apply only when both arguments are positive, but relationship
43:13:2 extends the definition to any pair of real arguments, as does the infinite-product expression
43:13:5 B(xY) _
(j l)(x + Y + j)
;_a (x+j)(Y+j)
If the two arguments of the complete beta function are equal in magnitude, or sum to a moiety, we have the
special cases
4' ( I rl _ (2j _ 1)!!
43:13:6 B(.rx) = cos(t<x)B x, - - x) = B x) J
2 2 2 '-o (2j)!!(j + x)
and
(-1)`/x x = 0, t 1, t2, .. .
43:13:7 B(x.-x) =
0 otherwise
Important special cases of 43:13:6 are B(;,;) = 21r/U, B(].]) = it and B(y,y) = 2U, where U is the ubiquitous
constant (Section 1:7J. When one (or both) of the arguments is a positive integer, the reciprocal of the complete
beta function reduces to a binomial coefficient (Chapter 61
43:13:8 =x
n+x-1
B(x.n) n-I
Intrarelationships of complete beta functions, such as
x x
43:13:9 B(.t + 1,y) = - B(x.y) _ - B(x,v + I)
x+v v
respectively, in Figure 43-2. The areas shaded in red on this map show where B(x,y) is positive: blue shading
indicates that the complete beta function is negative.
x, _ X>0
43:14:1
(1)i }tJo(2V-[) X250
XI
43:14:2 exp(X)
I_o (1),
I
43:14:3
I-X
-1<X<I
f-o
and the asymptotic representation
43:14:4 1
X') Ei(') IXI small
can claim a distinguished role in the theory of transcendental functions. The Bessel functions to and J. occurring
in 43:14:1 are treated in Chapters 49 and 52, respectively, while the exponential integral Ei is the subject of Chapter
37. These four have been termed basis hypergeomerric functions because they can be employed to generate almost
any hypergeometric function (i.e., any of the functions tabulated in Section 18:14), as will now be demonstrated.
Functions 43:14:1-43:14:4 have been expressed in terms of (I) the Pochhammer polynomial [Chapter 18] of
421 THE GAMMA FUNCTION r(x) 43:14
unity argument, but since (1), = j! = C(1 + j), each of these functions can be written alternatively as EX'11(1
+ j) with n = 2, 1, 0 or -1. Here X represents the argument x of interest, or some simple variant of it such as
-2x or x2/4. Now, let µ represent the following sequence of operations: (a) division by X1'(1 - µ), (b) differ-
integration [see Section 0:101 to order v - p. with respect to X using lower limit zero and (c) multiplication by
X"1'(1 - v). Then, making use of equations 13:10:11 and 43:5:10, we find for the v = 0 case
= X' 0 (1 - µ),X'
43:14:5
(1), I ;-a (I)j--
=
43:14:7
(1), µ ,-a (1),(1 - V),
In the terminology of Section 18:14, the operation is seen to preserve the difference L - K between the
numbers of denominatorial and numeratorial parameters but to alter the values of some of those parameters. Hence.
starting with the appropriate basis hypergeometric function, one may "synthesize" a sought hypergeometric function
by one or more v operations. Examples are
la(x) (x2/4)' :
43:14:8 -
0
= cosh(x) x = 2NX
1Y _a (1),(i),
43:14:9
exp()x)
_ ( )
-2 O', (2r1' 2
exp(x) 11(x) x=-X
(1), (1),(3), x
1 (X2), O(x'Yp 2
43:14:10
- xr ,=o
1 (I)
i (1); = - E(x)
x = NX
and
+...++..
x'
-2 sdaw( s) x=-X
where cosh, I E and daw are the functions of Chapters 28, 49. 61 and 42, respectively. Further information on
this topic will be found in Oldham and Spanier [Chapter 9J.
CHAPTER
44
THE DIGAMMA FUNCTION 4j(x)
The digamma function is derived by differentiation of the gamma function of Chapter 43. Multiple differentiation
yields the polygamma functions that are addressed in Section 44:12. Digamma and polygamma functions are useful
in summing certain algebraic series, a subject discussed in Section 44:14.
44:1 NOTATION
The digamma function is also known as the psi function. Some authors employ a translated argument and denote
our 4r(x) by ir(x - 1).
44:2 BEHAVIOR
Figure 44-1 includes a map showing the global behavior of the digamma function. The function 4o(x) has an infinite
discontinuity at x = 0 and at each negative integer argument. Otherwise, dir(x)/dx is always positive, this gradient
decreasing steadily as x increases through positive values. As x -+ x, y(x) approaches infinity logarithmically.
Figure 44-2 shows the detailed behavior of *(x), and some other functions, for 0.5 s x s 5.5.
44:3 DEFINITIONS
The digamma function is the derivative of the logarithm of the gamma function (Chapter 431
d I d
44:3:1 4Y(x) = ln(r(x)) = r(x)
dx r(x) dx
A second definition is as the limit
423
44:3 THE DIGAMMA FUNCTION $(x) 424
425 THE DIGAMMA FUNCTION Xx) 44:5
t` exp(-t) - exp(-x:)
44:3:3 x) _ -y I- f dt = -y + dt x>0
+ l0 1-t l0 1 - exp(-t)
where y is Euler's constant [Section 1:71. The integrals
44:3:4 fix) = j [
exp(-t) exp(-xt) f exp(-t) - (I + t)-` dt x>0
t 1 - exp(-t)Jdt - t
and
tdt
44:3:5 4.(x) = ln(x) - I - 2 X>0
to (t'' + x2)[exp(21rt) - 1]
are attributed respectively to Gauss, Dirichlet and Binet. For other definitions of 41(x) as a definite integral, see
Erddlyi, Magnus. Oberhettinger and Tricomi [Higher Transcendental Functions, Volume 1, Section 1.7.2.).
I \\ I
44:5:2 y 2 - s l = Jr - + x - it tan(irrx)
//
1
44:5:4 01+x)=4.(x)+-x
may be generalized to
44:5:5 n= 1,2.3,...
;=o+x
Similarly, the duplication formula
generalizes to
44 : 5 : 8 b(x) - W v) _ (x -
t Y) 5'
1
;-o (j + x)(j + Y)
or as a hypergeometric function [Section 18:141 of unit argument:
44:5:9
=x_y(1);(x-y+l); Y>0
fi(x)-4'(y)
x (2);(x+ l),
Infinite sums of quotients of digamma functions by gamma functions include
01) y(2) _ 4r(3) I)*j) = Ei(1)
44:5:10
F(j)
= 0.697174883
r(1) + r(2) r(3) + e
an d
where Ei is the exponential integral function [Chapter 37) and Yo is the zero-order Neumann function [Chapter
54]. The corresponding series without alternating signs Eilr(j)/r"(j) sum to -eEi(-1) = 0.596347362 and KO(2)
= 0.113893873 for n = I and 2, respectively, Ko being the zero-order Basset function [Chapter 51].
44:6 EXPANSIONS
The Taylor expansion
involves the zeta numbers [Chapter 3] and is valid only for arguments between 0 and 2. Of wider validity are the
expansions
a
44:6:4 I[2(x)=-y-
x2
1 -x2
-cot(az)
2
l
x 12
-- +t)x6J1 I
-2<x<2 k=3,5,7,...
An asymptotic expansion for the digamma function, valid for large argument, is
I I I I B,
44:6:5 ln(x) - x-- x
252x6 + ... - Js' +
120x4
2x 12x'' +
where B, is a Bernoulli number [Chapter 41.
Euler's constant, y = 0.5772156649, is a component of most particular values of the digamma function. Table
44.7.1 includes particular values of the trigamma function (Section 44:121 and of Bateman's G function [Section
44:131, as well as those of the digamma function fi(x).
By exploiting formulas 44:4:1 and 44:5:4, it is possible to evaluate the digamma function of any rational
argument x = m/n. For small values of n, rather simple expressions result. In addition to those incorporated into
equation 44:5:3, we have the following examples:
44:7:4
(3
+J) - 2 n_ 40(l 1
Z
'
- JI+=-y-In(8)+2 4
l=0,1,2, ...
4 4 4J _ 1
r-1
The digamma function encounters zeros at argument values of + 1.461632145, -0.504083008, -1.57349847,
-2.61072087, .... Denoting these zeros by r0, r1, r2, r3,/... then the following approximation holds:
The algorithm below has an absolute accuracy of better than I X 10-6, except when x is a nonpositive integer, in
which case the algorithm returns 1099 instead of ±x. For x z 5, the algorithm uses the expansion 44:6:5 in a
Table 44.7.1
6
q,l _= 3-Y-M4l 2--In14) -T-mi4) -1 --7-.41 1-1 3-y-I.041
3
--y `
Y 40 S
VU) s w J J I 0
2 0 2 2 6 2 6 2 9 6 4
truncated and compensated version. For smaller arguments, the algorithm uses the recursion 4o(x) = y(x + 1) -
(I/x) a sufficient number of times until the argument is brought to 5 or above.
I )/ lox-'
[
Replace f by - + 1) /2x] + In(x) - g Test values:
L Qx dr(7) = 1.87278434
(2) Outputf 4(1.461632145) = 0
f = 4L(x) <. ir(-1.05) = 20.2897713
44:9 APPROXIMATIONS
For large x the approximations
and
1
44:9:2 s() -in x- 2 8-bit precision xz4
are useful.
44:10:3 f(r&=1n(x)
F' x>0
and
429 THE DIGAMMA FUNCTION $(x) 44:12
44:10:5
r
,(t) sin(nzrt)dt =
0 n1
J0 2n n=2,4,6,..
are listed by Gradshtcyn and Ryzhik [Sections 6.46 and 6.471.
44:11:1
x
x +Y +G2 +i L.,-.(i+x)z +y'
;=.J[(i+x)z +Y1
y
Tables permitting the evaluation of 4,(x + iy) are presented by Abramowitz and Stegun [Table 6.81.
For purely imaginary argument
+F
44:11:2 y-
,..1(1=+Yz)
+ it
-
2y
+ - coth(ay)J
2
For the related special case
It y it
44:11:3 q'(z + iy) = -y - 2 + -1 tanh(ay) + : + - tanh(ay)
T+ 4y z 4y [(zj + I) + Y21 2
the imaginary part of the digamma function is seen to contain a hyperbolic function [see Chapter 301.
44:12 GENERALIZATIONS
One differentiation of In([ (x)) gives the digamma function; multiple differentiations generate the family of functions
known as the polygamma functions and denoted %W"(x). One has
The names trigamma function and tetragamma function, respectively, are given to 4, and +'2': -': these particular
polygamma functions are more commonly denoted y' and V. The functions 111th and 4,101 are known as the pen-
tagamma function and hexagamma function. .4,14),
For odd n the polygamma functions 4, are restricted to positive values, but 1,", 4+161, ... take values
of both signs. As illustrated in Figures 44-3 through 44-5. the polygamma functions are infinite at x = 0, -1,
-2..... By combining the recurrence formula
44:12:5 V) 1
=(-2)"'In!A(n+ 1) n= 1,2,3,...
2l/
44:12 THE DIGAMMA FUNCTION $r(x) 430
1 14V
-if 41
431 THE DIGAMMA FUNCTION 4(x) 44:12
involving zeta C and lambda A numbers (Chapter 3], one may derive the expressions
44:12:6 yr"1(n=(-l)"'n!IOn
pl j-! \ 3
+1)-n=1,2,3,...
r
L
/_1 1 1
!-1 1
1
44:12:7 ,yl.l(1 - (-2)"*'n!IA(n
+ l) - (2j - 1)-" J = n! J I, 2. 3,
2 G-l Jl
for the polygamma functions of positive integers and half-integers. Some such values are included in Table 44.7.1
for the n = I case.
For large positive values of their arguments, the polygamma functions are well approximated by
Yet more refined approximations are provided by 44:12:12 for large x. As is evident in Figures 44-3-44-5, for
negative x values of any magnitude, the values of qs``(x) are largely determined by the fractional value [Chapter
9] of x and hardly at all by its integer value. This is a consequence of the fact that, in the reflection formula
the final term for x > 0 is generally negligible compared with the periodic derivative term.
Among the relationships for the polygamma functions that can be derived by differentiation of the corresponding
formulas for the digamma function are the following expansions:
(n + j)!
44:12:10
(-x)"-1
n!
t;(n+j+l)x' n=1.2.3.... -1<x<I
j_0 J
44:12:11 n=1,2,3,...
j-0 !+x
44:12:12
"(x) -(n-I)!+n+n(n+1)-n(n+l)(n+2)(n+3)
(-x)" 2x 12x' 720x4
Ifx<5goto(1)
Set f = [I - (n + 4)(n + 5)/45x2]/60x'
Replace f by (n + 1)[1 -f(n + 2)(n + 3)1/6x
-I1 +f+ lJ
Replacefby
n 2x
Set j = n
(3) Replace f by fj/x
Replace j by j - 1
Ifj*0goto(3) Test values:
Replace f by f + g x'(5.5) = 0.199342388
(4) Output f %r(-0.5) = -0.828796644
f = '"'(x) 403)(9) = 0.00323439609
The algorithm does not generate iIr(x) when n = 0 is input. The algorithm utilizes equation 44:12:3 together with
a truncated, compensated and concatenated version of 44:12:10; it parallels the algorithm of Section 44:8 in its
manner of operation. Some approximate values of 4,'"'(x) may also be read from maps in this chapter.
44:13:3 G(x)=2(=2-n(l;x)
j I-0 x +
Some properties of Bateman's G function are
2
44:13:4 G(1+x)=--G(x)
x
and
1
44:13:7 G(x) = - 1 2x1 + csc(,rrx) + 2 11 - -n(2f + 1)]x22'
i-o
where -n(j) is the eta number of Chapter 3. The behavior of G(x) is mapped in Figures 44-1 and 44-2; Section
18:14 shows how the universal hypergeometric algorithm may be used to calculate its values. Some particular
values of G(x) will be found in Section 44:7.
433 THE DIGAMMA FUNCTION y(x) 44:14
Many properties of the successive derivatives G'"(x) of Bateman's G function are given by Erddlyi, Magnus,
Oberhettinger and Tricomi [Higher Transcendental Functions, Volume 1, Section 1.16]. We mention only
I I I I = Z(-1)'
1
44:14:2
c b+cf 2lb+c\3b+c Jb-b+c -obj+c
/ c/J
IIGI--b I -(-1)'G`Jbb
Because the G function approaches zero as its argument a pproaches infinity, the sum to j = x of (-I)'/(bj + c)
is simply G[(b + c)/b]/2b. Special cases of these results are presented as equations 1:14:5 and 1:14:10.
Similarly, infinite series of reciprocal powers have the sums
and
(b + c)
-2bc -) V1-
44:14:4 -- (b + c)"
C.
+ -
(2b + c)" (3b + c)"
+
_0 (jb + c)" z (-2b)"(n - I)!
2b
n= 1,2,3,...
in terms of polygamma functions. A corresponding finite series may be expressed as the difference between two
infinite series and hence may be summed via equation 44:14:3 or 44:14:4.
Consider the summation of a series of rational functions [Section 17:131 of the summation index k
A k' + A.,_,k-' + - - + Aik + Ao -
44:14:5 S= pT(k)
k.o p.(k) k=o a.k" + o._,k" ' + ... + ask + ao
where p," and p" are polynomial functions of degrees m and n. If n a m + 2. the series will generally converge,
and, in these circumstances, S may be expressed in terms of digamma and/or polygamma functions, as we now
demonstrate.
In Section 17:6 it is shown that 1 /p.(k) may be expressed as the sum of n partial fractions, each of form C;/
[a.(k - pi)] where pi is the ja zero of p. and Ci is a constant. A similar fractionation of p.,(k)/p.(k) is possible so
that
C'
44:14:6 S= =-1 CJd-P,)
a. ,..1 k=o k - pi a. j-,
44:14 THE DIGA.MMA FUNCTION y(x) 434
Because m s n - 2, it follows that IC, = 0, and this fact is used, together with the first of the 44:6:2 expansions,
to establish 44:14:6. Some (or all or none) of the zeros p, may be complex, in which case the corresponding C,
-
constants will also be complex.
The foregoing presupposes that the zeros P. p,, ... p. are all distinct. When this is not the case, one follows
the procedures outlined in Section 17:13. For example, if the zeros p Pr and p, (but no others) were coincident,
one would replace 44:14:6 by
I- j r Ci
+
G
+
C. 1
J +
I C,
44:14:7 S=
k - Pi (k - pi)- (k-P,)' a.,-44-0k-p,
C
Cid(-PI) + I"(-Pi) - 33 d"(-Pi) - i C,,b(-p,)
1. 22 i-4
equation 44:12:9 having been used.
CHAPTER
45
THE INCOMPLETE GAMMA 'y(v;x)
AND RELATED FUNCTIONS
Many of the functions discussed in preceding chapters are special cases of the functions considered in this chapter.
Besides the incomplete gamma function y(v;x) itself, we shall make frequent use in this chapter of two closely
related functions: the complementary incomplete gamma function r(v;x) and the entire incomplete gamma function
y'(v;x). The interrelationships
y(v;x) f(v;x)
45:0:1 fy(v;x) = = I -
r(v) r(v)
link the three, where f(v) denotes the (complete) gamma function (Chapter 43]. The entire incomplete gamma
function remains useful for negative x, whereas y(v;x) and r(v;x) are then often ill defined.
45:1 NOTATION
The variables v and x are termed, respectively, the parameter and the argument of the functions. The adjective
"incomplete" reflects the restricted ranges of integration in definitions 45:3:1 and 45:3:2 compared with that in
Eulei s integral of the second kind, 43:3:1.
Alternative symbolisms abound. fAv) has been used for y(v:x); P(b,x) and Q(v,x) have been used for y(vx)/
r(v) and r(v;x)/r(v). respectively. The notation EV(x) or is sometimes adopted for x"-'r(1 - v;x).
45:2 BEHAVIOR
Because, for negative arguments, the functions y(v;x) and r(v;r) are often complex and/or multivalued, we restrict
our discussion of these functions to x ? 0. Figure 45-1 maps typical behaviors of these two functions for a variety
of values of the parameter.
For x > 0, the y(v;x) function is well defined except for v = 0, - 1, -2, ... when, like the complete gamma
function [Chapter 431, it encounters infinite discontinuities viewed as a function of its parameter. The y(v;x) func-
tion is always a monotonically increasing function of x. For v > 0, -y(v;x) is zero at x = 0 and positive elsewhere.
For -1 < v < 0, -3 < v < -2, -5 < v < -4, etc., the function is invariably negative, infinitely so at x = 0.
For -2 < v < -1, -4 < v < -3, etc., y(v;x) equals -- at zero argument, then increases monotonically, crossing
zero to reach the positive value r(v) at x = x.
435
45:3 THE INCOMPLETE GAMMA y(v;x) AND RELATED FUNCTIONS 436
Y(ZIx)
....... ....
FIG 45-1
.....................-2
The complementary incomplete gamma function is invariably positive and decreases monotonically with x,
approaching zero asymptotically as x -+ x.
The entire incomplete gamma function y*(v;x) is single valued, real and finite at all values of v and x that are
real and finite. Its behavior is comprehensively described by the contour map Figure 45-2. Notice that y* has no
zeros for v > 0. has one zero (at a negative value of x) for -1 < v < 0, -3 < v < -2, etc., and two zeros (one
at positive and one at negative arguments) for -2 < v < -1, -4 < v < -3, etc.
45:3 DEFINITIONS
The y(v;x) and r(v;x) incomplete gamma functions are usually defined via the integrals
Definition 45:3:1 may be extended to negative v by utilizing the recursion formula 45:5:2. The definite integrals
x" exp(-x) t-' exp(-t)
45:3:3 r(v;x) =
r(1 - v) f0 t+x
dt x>0 v<I
and
open up other ranges of the parameter and argument. The more complicated integral representations
link the incomplete gamma functions to Bessel [Chapter 531 and Basset [Chapter 51 ] functions.
Often the series 45:6:1 is taken as the definition of the incomplete gamma function. As well, the entire in-
complete gamma function may be defined succinctly by the differintegration [Section 0: 10] formulation
45:3:7
-
dx'
d'
exp(±x) = x-' exp(±x) y*(-v;±x)
45:4:6 y*(,;-x) =
72
exp(x) daw(f )
439 THE INCOMPLETE GAMMA y(v;x) AND RELATED FUNCTIONS 45:6
Special cases of such integrals are tabulated in Table 37.14.1 for v = -3, -j, -2, ..., 2. Thus, this table serves
to provide a listing of the special cases r(v;x) for v = -2. 9, -I, ..., 3.
45:5 INTRARELATIONSHIPS
The incomplete gamma function and its complementary cohort sum to the (complete) gamma function [Chapter
43]
45:5:1 Y(v;x) + r(v;x) = r(v)
Recurrence formulas for the three functions are
45:5:2 Y(v + l;x) = vY(v;x) - x'exp(-x)
45:5:3 r(v + l;x) = vr(v;x) + x"exp(-x)
and
Y*(v;x) exp(-x)
45:5:4 Y*(v + Ix) x xr(1 + v) x
The first of these may be generalized to
I. -' x' 1
45:5:5 Y(v + n;x) = (v).LY(v;x) - x" exp(-x) 2 -J = x"p._i(x) exp(-x)
45:6 EXPANSIONS
A large number of expansions exist for the incomplete gamma functions. The most important of these are
x"
Y(v;x) = - - - +
x"' -
x"`'
+ = x"
(-x)'
45:6:1
v I+v 2(2 + v) 6(3 + v) J.Oj!(j + v)
and
45:6:6 r(v;x) --
exp(-x) r
x1
I -1-- v+ (I - v)(2 - v) - (I - v)(2 - v)(3 - v)
x
X.
x3
(-x)'
valid for large x.
Expansion 45:6:6 applies when x, but not v, is large. Conversely, there is an asymptotic expansion
45:6:7 y(v;x)--x - -
IV
1
v+I
x
+
x'
2(v + 2)
- 6(vx'+ 3) +- +
(-x)'
j!(v + j)
+ v-'x
that is valid when the parameter, but not the argument, is large.
Table 45.7.1
x=o
v<o v>o v<o v>o
-0;x) -x 0 r(v) 174 V)
r(v:x) undef r( Y) 0 0
y9v:x) 0
f(l+v) r(I+v)
+ x"-')(n - I + v) + x"-'
These are the formulas used in the algorithm for y'(v;x) that is presented below. The integer n is chosen to be at
45:8:3 '
least I and such that n + v > 2. The infinite series in 45:8:1 is evaluated via the truncated concatenation
i
0(n+v),, (((
r(r r
\ n+v+J-x +1lIn+v+J - l
+1
n +v+J-2
441 THE INCOMPLETE GAMMA y(v;x) AND RELATED FUNCTIONS 45:10
n+v+2
x x
n+v+l
+1n+vl -
By use of the empirical assignment J = Int(5 + (3 + kxl)/2) one ensures that approximation 45:8:3 has 24-bit
precision (the relative error is less than 6 X l0-"). The final segment of the algorithm is concerned with calculating
the multiplier exp(-x)/r(n + v), using a procedure similar to that employed in Section 43:8.
Input v >>
Setg=p= I Storage needed: v, g. p. x, j and f
Input x >>
(1) Replace v by I + v
Ifv>2goto(2)
Replace g by gx
Replace p by pv + g
Go to (1)
(2) Set j = lnt(5(3 + !x1)/2)
Set f = 1 /(j + v - x)
(3) Replace j by j - I
Replace f by (fx + 1)/(j + v)
Ifj * 0 go to (3)
Replace p by p + fgx
g= 1- 2 (1 \1
_ 2 I] /30v2
Set
7v'3v'
Replace g by g - v[In(v) - I]
12v Test values:
Set f = p exp(g - V)v/2a y*(#;-0.49) = 1.34326134
Output f y*(-3.9;0) _ -0.521258841
f = y*(v;x) y*(I;a) = 0.304554469
Of course, numerical values of y(v;x) and r(v:x) may be computed from y*(v;x) by using equations 45:0:1.
Alternatively, the universal hypergeometric algorithm of Section 18:14 may be used in several versions.
45:9 APPROXIMATIONS
Differentiation of the incomplete gamma function and its complementary analog give
d d
45:10:1 y(v;x) F(v;x) = x°'exp(-x) = y(v;x) - (v - 1)-y(v - I;x)
dx dx
The differintegration [Section 0:10] formula for the entire incomplete gamma function
d"
45:10:2 If exp(x)y*(v;x)] = x"-" exp(x)y*(v - µ;x)
dx"
and
(' r(v + µ)
45:10:4 J r"-'r(v;bt)di= µbµ v+ IL >0 p.>0
0
and many others are listed by Gradshteyn and Ryzhik [Section 6.45).
45:12 GENERALIZATIONS
The three incomplete gamma functions are special cases either of the Kummer function [Chapter 471
x" x"
45:12:1 y(v;x) _ - exp(-x) M(1;l + v:x) _ - M(v;l + v;-x)
V V
I
Incomplete gamma Parabolic cylinder
function, etc. [451 function (46]
I
Dawson's Error function, Exponential
integral [421 etc. [40, 411 integral (391
Constants [11
CHAPTER
46
THE PARABOLIC CYLINDER FUNCTION
This function arises in the solution to many practical problems that are conveniently expressed in parabolic cylin-
drical coordinates. This coordinate system, and others. are discussed in Section 46:14.
46:1 NOTATION
The parabolic cylinder function is also known as the Weber function or the Weber-Hermite function. The name
"Whittaker's function' is also encountered, but confusion should be avoided with the identically named functions
of Section 48:13. An alternative to the usual D,(.r) notation is U(-v - k,x).
The variables v and x are known, respectively, as the order of the parabolic cylinder function and its argument.
The origin of the name "parabolic cylinder" function is made evident in Section 46:14.
46:2 BEHAVIOR
The D,(x) function is defined for all real values of v and x. Figure 46-1 is a contour map showing some values of
the parabolic cylinder function.
For v less than about -0.20494, D,(x) is a monotonically decreasing positive function of x. For larger orders,
the parabolic cylinder function displays a number of zeros, maxima and minima in the -x < x < x range, as
shown in Table 46.2.1.
Table 46.2.1
-0.20494 < v s 0 0 1 0
0<vs1 1 1 0
I<vs2 2 I
2<vs3 3 2 I
etc.
46:3 DEFINITIONS
The parabolic cylinder function is defined by a number of integrals, including
-xt)-
42
This is the n = 0 case of a general expression for a parabolic cylinder function of negative integer order, which
may be written as either
r / \ / 1l
46:4:3 D_ (x) expl( l Lexpl 2,/ erfcl c IJ n = 0. 1, 2....
n! 2 4ds"
or
in terms of either successive derivatives or repeated integrals [see Section 40:131 of the error function complement.
When the argument of a parabolic cylinder function is positive and its order is an odd multiple of 2, it may
46:5 THE PARABOLIC CYLINDER FUNCTION D.(x) 448
be expressed in terms of one or more Basset functions (Chapter 511 of orders that are odd multiples of #. The
simplest cases are
46:5 INTRARELATIONSHIPS
The parabolic cylinder function satisfies the recursion formula
46:5:1 D,.,(x) = xD,(x) - YD,-,(x)
and the argument-addition formulas
/2xy 4+y2I 2xy - y2 v.,y;D._,(x)
46:5:2 D,(x + y) = exp D,.j(x) = expl 4 11
=Q j! \ ) f=o \
The sum or difference D,(x) t Dr(-x) can be expressed in terms of Kummer functions [Chapter 47]
r(12v\
exp(- I Ml - - - I =
4/ 2 2 2/ r112v\
expl - I MI
\4/ \ 2
-
2 2
l / x2 /1-v 3x2 _ -
46:5:4 D,(x) - D,.(-x) =
;/'eJ
r(_v xexpl 4 ) MI 2 .2. 2 r(-v 3n xezp\4/
x2
2yJ
M\1
v 3_ x2
+ -; 2; 2
2) \ ` 2)
46:6 EXPANSIONS
A number of power series expansions exist for exp(±x2/4) D.(x). Thus, one has
r- zv f (-xV2)'
P('
46:6:1 2a 2r(/ v) '
\
2.
2i2
H2,(x/ V 2) (I '
46:6:4 D,(x) = exp ( ) x>0
r( /2) \ 4 '/ j, (j 4 }
and
46:6:5 D,(x) =
\\\ 2
expl - j!
! 2
(__!) x>0
D.(0) D,(-)
0<v<1,2<v<3.4<v<S.
v=0.1,2.... 0
v<0
(2) If x = 0 go to (3)
Replace g by gx
Replace x by x + 1
Ifx:S 3 go to (2)
Setb=2r/x2
/ \1
+7b
Replace fLbyf-l\
12x
- x[l/n(x) - 11
Replace f by g V-./2 exp(f)
(3) If a = 0 go to (4)
Seta=0
Set w = f
Go to (I)
(4) Set a = w
Set b = f
Input x >>»»
Set G = -b.\/
Set J = 3 + 61x1 + (I(x + 2)(2 - v)I/5) + (x2/2)
Replace x by x 2/2
//
Setw= explvln\2-x-x
Replace G by``Gw //
SetfF=aw
Set g=G
Set d = (F + G) exp(x)
Set w = (v + 1)/2
Setj=0
(5) Replace j by j + I
// 1 1
For large arguments, the parabolic cylinder function is best determined via expansion 46:6:6 (see also Section
48:8].
46:9 APPROXIMATIONS
If x is small and of either sign
v - v' 1
46:9:2 D.(x) _ .r" expl -x'J I I +
4 L 2x'
The latter approximation is valid even for moderate positive x if v is close to one of the integers 0, I. 2 or 3.
46:10:3
46:10:4
(aY
tp exp - D.(t)dt =
-4 ) (a+p r(p+l) p+
2"a
\ ( v
F
2
1
- l +p -;
-v
2
- va-l l
2a+I
r( 2 + 11
p> -1 a>0
yields a Gauss function (Chapter 601.
46:12 GENERALIZATIONS
The Kummer function [Chapter 47] and the Tricomi function [Chapter 48] may be regarded as generalizations of
the parabolic cylinder function. The relationships are
46:12:1 D.(x) =
'
V
/'
r n expl
-r
4/
J
M 2
v
r(l 2 v)
I; x
2 2
V tr
MI
/1- v 3 r'
r('
2 ? 2
1+ v x/ 1 I /v 3 r
/=
x 2. 2 2 \ 22 2
2"a exp 4 1
l - x?0
vl
l r(( l
rr 1
2 2v/
46:13 THE PARABOLIC CYLINDER FUNCTION D,(x) 452
U(I
46:12:3 D,(x) = 2`00/2 eXp(4) x?0
2 v; 23; 2)
y=4
Y-2
FIG 46-2
y=2
y=1
y=o
y=- 1
ya-2
FIG 46-3
y-U
r.. y=-1
rot
..... ..... y a-2
.: y=-3
rsA
sinh(r) sin(8)
46:14:4 x= -x < r < x -V < 8 <_ V
cosh(r) - cos(8) cosh(r) - cos(O)
to cartesian coordinates. As Figure 46-6 shows, the equipotentials are mostly circular cylinders; their axes are the
straight lines (y = 0, x = coth(r)) and their radii are Icsch(r)I. Streamlines are arcs of circles whose centers lie on
the line (x = 0, z) at y = cot(O) with radii equal to Icsc(e)I.
The spherical coordinate system employs one length r and two angles 8 and 4s as its three coordinates. Because
it is inherently three dimensional, we do not include a diagrammatic representation of this system, which is related
to cartesian coordinates by the equations
IT
46:14:5 x= r sin(8) cos(d,) y= r sin(8) sin(4) z= r cos(8) 0s r< -- <- 8< --n < 4 K a
2 2
The generator is a single point at the origin; equipotentials are spheres centered on the origin, and streamlines are
straight lines radiating in all directions from the origin.
There are a number of other orthogonal coordinate systems that we shall not discuss in detail. The generator
of the paraboloidal coordinate system is the half-line (x = 0, y = 0, z > 0). The generator of the prolate spheroidal
coordinate system is the line segment (x = 0, y = 0, - I < z < 1). The generator of the oblate spheroidal coordinate
system is the disc (x2 + y2 < 1, z = 0). The generator of the toroidal coordinate system is the hoop (x2 + v= _
1, z = 0). For details of these systems, and three others, see Spiegel (pages 126-I30J.
With k sometimes positive, sometimes negative and sometimes zero, the equation
46:14:6 V2F=kF
often known as the Helmholtz equation, is ubiquitous throughout all of science. Special instances are the wave
equation for the propagation of vibrations, Laplace's equation describing electric fields, the Fourier equation of
i$, ig i4 1
b A
i' i O 44 4gp 44
2
i4,O
455 THE PARABOLIC CYLINDER FUNCTION D,.(x) 46:14
heat conduction, the Schrodinger equation of quantum physics and Fick's second law of diffusion. In all these
cases F is some scalar property whose value depends on the spatial coordinates (and often on time also, although
here we shall ignore the temporal variation of F). The so-called Laplacian operator V performs double differen-
tiation with respect to the spatial coordinates, the precise form of the operator depending on the coordinate system
adopted. Table 46.14.1 lists the forms of 6''F for a variety of orthogonal coordinate systems.
Table 46.14.1
Coordinate system VF
a'F d'F d'F
Cartesian
ax= ay' d:
-+--+--+-
--
a'F 1 aF I a-F a'F
Cylindrical
ar' r dr r' aB'
I a'F a'F1I a"F
Parabolic cylindrical + +
r + q' ay; aq'/ a-
I a'F d'F a'F
Elliptic cylindrical
sinh'(r) + sin'(6) a ., a9' aY
(a'F d-F d'F
Bipolar lcosh(r) - cos(O)1'
\ dr + d9' +
Spherical -+--+--+-+--
d2F
ar'
r'-
2 aF
r ar
cot(B) dF
r' ae
I a'F
ae'
csc'(B) a-F
r' ab'
46:14 THE PARABOLIC CYLINDER FUNCTION D,(x) 456
it . i4, it
t
tip, i4'L
Because the cartesian coordinate system is undoubtedly the simplest, both conceptually and in the form of its
Laplacian, one might wonder why other systems are ever used. One answer is that, for a physical body that matches
the corresponding generator, the Helmholtz equation is most easily solved in the appropriate orthogonal coordinate
system. We shall illustrate the method of solution by reference to the parabolic cylindrical coordinate system.
The function F that is a solution of the Helmholtz equation
VF=
I (d=F+J=F\+dF=kF
46:14:7
r2 + q' I\dr' dq'/I Bz'
will generally be a function of all three coordinates: r, q and z. Assume, however, that F = R(r) Q(q) Z(z) where
----
R is a function of r, but not of q or z, and similarly for Q and Z. Then equation 46:14:7 is easily transformed to
(1 d'R
46:14:8
1
A consequence of our assumption is that the left-hand side of this equation is a function of r and q, but not of z.
whereas the right-hand side is a function of z, but of neither r nor q. It follows that each side must equal the same
constant, a so-called separation constant, say c4, so that
and
I d 2Q
46:14:10 R- -cr2=cq'- Q dq
457 THE PARABOLIC CYLINDER FUNCTION D,(x) 46:14
Once again, we can argue that, because the left-hand side of 46:14:10 does not depend on q and the right-hand
side does not depend on r. each side must equal another separation constant. It is convenient to represent this new
constant by c2(2v + 1), whence
1 d'Q
46:14:11 I)Q
c dq'
and
1 d'R
(C2
46:14:12 OR
c dr'
Thus, the Helmholtz equation has been decomposed into three ordinary differential equations. It requires only a
redefinition of the independent variable to convert equation 46:14:11 to 46:3:3. showing that a parabolic cylinder
function can be a solution of the Helmholtz equation expressed in parabolic cylindrical coordinates.
Section 59:14 presents a second example of the solution of the Helmholtz equation.
CHAPTER
47
THE KUMMER FUNCTION M(a;c;x)
The Kummer function is one of the most important instances of the hypergeometric function [Section 18:14] and
is closely related to the Tricomi function of Chapter 48.
47:1 NOTATION
The symbol $(a;c:x) often replaces M(a;c;x). The variables a and c are termed the numeratorial parameter and
denominatorial parameter, respectively; x is the argument.
Alternative notations for the Kummer function make recourse to a generalization of the Laguerre function
[Section 23:14]
M(-v:g + l;x)
47:1:1 L"'(x) =
r(v + I)
or to a simple instance of the generalized hypergeometric function [Section 60:121
47:1:2 1Ft(a;c;x) = M(a;c;x)
Collectively, the Kummer function and the Tricomi function (Chapter 48] are known as confluent hypergeo-
metric functions or as degenerate hypergeometric functions. These puzzling names have their origins in definitions
47:3:2 and 48:3:5.
47:2 BEHAVIOR
The Kummer function is defined for all real values of a and x and for all values of c except c = 0, -1, -2, ...
Being trivariate. M(a;c;x) is incapable of having its behavior comprehensively described in a two-dimensional
map. Some examples of the wide variety of behaviors are depicted in Figure 47-1. In its dependence upon x, the
Kummer function may, or may not, encounter zeros, maxima or minima, depending on the values of the two
parameters. Figure 47-2 may be used to find, for specific values of a and c, the number of negative zeros of the
Kummer function, that is, the number of times that M(a;c;x) acquires the value zero in the argument range -x <
x < 0. Similarly, Figure 47-3 can be used to find the number of positive zeros of M(a;c;x). In the regions in which
the Kummer function lacks zeros, it is invariably positive.
For the behavior of the Kummer function as its argument approaches infinity in magnitude, see Section 47:7.
lID
47:2 THE KUMMER FUNCTION M(a;c;x) 460
FIG 47-1
<.: .... : .............. : .... :.1
M (-2. 6s 1. 21 x)
.. .. .. -1
H(1.6s-0.41x)
'1
04 0'% 0 e
c-2
c=1
C-0
c--2
c--3
461 THE KUMMER FUNCTION M(a:c;x) 47:3
47:3 DEFINITIONS
The Kummer function may be defined as a hypergeometric function (Section 18:14) with one numcratorial parameter
and two denominatorial parameters, one of the latter being equal to unity:
47:3:1 M(a;c;x) _ x,
J-o (c)/(l );
It may also be defined in terms of the Gauss function [Chapter 601 by means of the limiting operation
x
47:3:2 M(a;c;x) = llim F a,b;c;b
and
d''f df
47:3:6 z -+ (c - x)- - of= 0
dx dx
has the solution f = c,M(a;c;x) + c2x'-`M(a - c + 1;2 - c;x) provided that c is not an integer; c, and c_ are ar-
bitrary constants.
The Kummer function is obtained by differintegrating the function xa-' exp(x) with respect to x, using a lower
limit of zero:
r( d°
47:3:7 M(a;c;x) = [X,-, exp(x)]
f (a) xl ` dx°-`
Thus, in the notation of Section 43:14, the synthesis
I-c
47:3:8 exp(x) -- M(a;rx)
1-a
gives rise to the Kummer function.
If the denominatorial parameter exceeds the numeratorial parameter by unity, the Kummer function reduces to an
incomplete gamma function [Chapter 45]:
x
When the denominatorial parameter is twice the numeratorial parameter, c = 2a, the Kurnmer function
specializes to the product of the exponential function and the hyperbolic Bessel function (Chapter 50) of order
a-4:
r(z + 4)exp
in terms of the entire incomplete gamma function or, alternatively, if x is positive, in terms of the incomplete
gamma function.
Except perhaps when a is also a negative integer, setting c = 0. - 1, -2, ... causes M(a;c;x) to be undefined
[but see Section 47:12). When c = 1, reduction occurs to a Laguerre function [Section 23:141.
47:4:7 M(a;l;x) = L_o (x)
Use of recurrence formula 47:5:3 then permits the special cases M(a:2;x), M(a;3;x), etc., to be expressed as La-
guerre functions also.
With a denominatorial parameter equal to ;, the Kummer function becomes the sum of two parabolic cylinder
functions [Chapter 461
1
2°-'r(a + z) exPI
2 1[D_z(V tar) + D-.(-VT,)) x20
a
47:4:8 M a;-;x
2 r(1 - a)
expl 2 I Dm-,(- )) x:5 0
2°
The difference of two other parabolic cylinder functions is the special c = ; case of the Kummer function:
3 --
-2"r(a - 1)
V s2ax
/x
expl 2 J1D,-<(\) - x>0
47:4:9 M a;;z
2 r(1 - a) /x\\ /
expl 2 I[Dm_z(V -lx) - D2.-,(- V -2x)) x<0
\/
Bessel functions [Chapter 53) and their hyperbolic counterparts (Chapter 51) arise from the limiting operations
ym r -xl r(1 + e)
47410 lim M a;c+ I; - = J zz0
a x`/'-
47:5 INTRARELATIONSHIPS
Known as Kummer's transformation, the important identity
47:5:1 M(a;c;-x) = exp(-x) M(c - a;c;x)
constitutes a reflection formula for the Kummer function.
Recurrence formulas may be written interrelating three Kummer functions whose parameters differ by unity,
as follows:
2a - C +X c - a
47:5:2 M(a + l;c;x) = M(a;c;x) + - M(a - l;c;x)
a a
a - c+l c-I
M(a;c;x) + - M(a;c - l;x)
a a
c(c - I + x) c(c - 1)
47:5:3 M(a;c + l ;x) =
(c - a)x
M (a;c;x) - (c - a)x
M (a;c - l ;x) = xc- M(a;c;x) - x M (a - l ;c;x)
c-
47:6 THE KUMMER FUNCTION M(a;c;x) 464
47:5:5 M(a;c;x +
(a)')'
i-o c iir
and others are given by Erd6lyi, Magnus, Oberhettinger and Tricomi [Higher Transcendental Functions, Volume
I, Section 6:14]. By setting y = (v - 1)x, 47:5:5 becomes an argument-multiplication formula.
A special linear combination of Kummer functions:
r(l - c) r(c - I)
47:5:6 M(a;c;x) + r(a i M(a - c + 1;2 - c:x) = U(a;c;x)
r(a - c + 1)
generates the Tricomi function [Chapter 481. Among the summablc infinite series of Kummer functions is
5
47:5:7 M(a;c;x) + M(a - 1;c;x) + M(a - j;c;x) = c s 1 M(a;c - l;x) c>
-o 2
and several others are reported by Erdelyi, Magnus, Oberhettinger and Tricomi [Higher Transcendental Functions.
Volume 1, Section 6.15. ].
47:6 EXPANSIONS
The power series expansion
ax a(a + 1)x2 a(a + 1)(a + 2)x' (a)ix'
47:6:1 M(a;c;x) = I +
c
+ 2!c(c + 1) + 3!c(c + 1)(c + 2) + - =o
known as Kummer's series, is convergent for all arguments and for all parameter values except c = 0. -I.
2, .... Using v as an abbreviation for c - a - _, a series expansion in hyperbolic Bessel functions [Chapter 50]
is
/4 x1 (-l)j(2v) (c - 2a) (x\
47:6:2 M(a;c;x) = r(v)1-) expl - I ' ' 1,; I
X 2 jl(c),
i-o 2
Expansions may also be made in Bessel functions of order c - I + j: see Abramowitz and Stegun [Section 13.3]
for the coefficients involved. Another expansion is in terms of generalized Laguetre polynomials (Section 23:12]:
(a),
47:6:3 M(a;c;-x) L"-" (x) x>0
2'(c)1
The Kummer function may be written as the sum of two terms involving Tricomi functions [Chapter 481
(-1)° r(c) (- I)°-` r(c)
47:6:4 M(a;c;x) = r(c - a) U(a;c;x) + r(a) exp(x) U(c - ac;-x)
where powers of -1 are to be interpreted, as explained in Section 13:11, as the (generally complex) number
47:6:5 (-1)` = cos(tar) + i sin(g)
Because the Tricomi function has a simple asymptotic expansion [see 48:6:1], an expansion of the Kummer function
valid for large x may be constructed using 47:6:4.
465 THE KUMMER FUNCTION M(a;c;x) 47:8
Figure 47-5 similarly depicts the values of M(a;c;+x). Notice the exclusion from both diagrams of c = 0,
-1, -2, ... where M(a;c;x) is undefined.
Because the Kummer series, equation 47:6:1, converges rapidly, it constitutes a convenient method for calculating
values of M(a;c;x). If x is negative, the terms in the expansion
(a),x1
47:8:1 M(a;c;x) T,
ultimately constitute an alternating series (see Section 0:61. One can show that, for all values of j greater than
some integer J, the series T, + T,. I + T,.2 + + T) + - is alternating provided that IT,I < rr,_,j and that
J > 21al + lei + I. These considerations form the basis of the algorithm presented below, which exploits the unique
advantage, namely
47:8 THE KUMMER FUNCi7ON M(a;c;x) 466
I
J1 1
Setj=g=0
Ifx<0goto(I) Input restrictions: The c parameter cannot take values
Set f = exp(x)
Replace x by -x
Replace a by c - a
(1) Replace g by g + t
Set p = (a + j)x/(c + j)(j + 1)
Replace j by j + I
Replace t by tp Test values:
If 111 - 10-'IgI go to (1) M(0.7;0.6;2) = 8.94061153
If IpI > I go to (1) M(-4;2;-1) = 4.175
If )s2Ial+Icl+2go to (I) M(0.5:l;-2) = 0.465759608
r
Replace f by f g + 2
467 THE KUMMER FUNCTION M(a.c:x) 47:10
Output f
f - M(a;c;x) < <<
47:9 APPROXIMATIONS
If either a or x, but not c, is small:
ax
47:9:1 M(a;c;x) = I + -
c
If one of a. c or x is large, the other two remaining modest in magnitude, the following approximations hold:
xl,._,(
47:9:3 M(a;c;x) r(c) ax - ex2) exp 4ox - 2cx) a large and positive
/ xl
47:9:4 M(a;rx) I-- c large, either sign
\\\ cI
r(c)
c - a * 0, -1, -2, .. .
r(c - a) (-x)'
47:9:5 M(a;c;x) x large and negative
r(c )r(o)
47:9:6 M(a:c;x) = xe-'exp(x) x large and positive a * 0, -1. -2, .. .
47:10:3 [ o
M(a;c;r)dr = c
a-1
1 (M(a - 1:c - I;.r) - I) a#I c*1
With a lower limit of zero, differintegration of the Kummer function gives a hypergeometric function [Section
18:14]; the formula
d" M(a.c.x) = x - (a) x
47:10:6
dx" r(1 - V) J.0 (c);(I - v);
represents a generalization of 47:10:2. The differintegration formula
d° J 'X
r(c)
47:10:7 x`-'M(a;c:x) = M(a:c - v;x)
dx' r(c - v)
is equivalent to the integral transform 47:10:4. Hence, in the notation of Section 43:14, one Kummer function can
be synthesized from another by
47:10:8
I-C
M(a:c;x) -+ M(a;C;x)
I - C
Similarly:
1 -a
47:10:9 M(a;c;x) - M(A;c;x)
1-A
2
l 2/(I),\2)
(
47:12 GENERALIZATIONS
M(a:c:x)
47:12:1
r(c)
is a modest generalization of the Kummer function designed to embrace those values of the denominatorial param-
eter (c = 0,-1,-2,...) that are excluded from the definition of the Kummer function itself. Unlike M(a;c;x), the
entire Kummer function encounters no discontinuities as c -. 1 - n = 0, -1, -2, ..., having the finite values
M(a;l - n:x) (a),x"
47:12:2 M(a + n;l + n;x) n = 1, 2. 3,...
r(1 - n) = nr
469 THE KUMMER FUNCTION M(a;c;x) 47:13
at these points. It is unfortunate that the entire Kummer function is not widely used because its properties are
significantly simpler than those of M(a:c;x).
As well, all hypergeometric functions with L = K + 1. K >_ I [see Section 18:14] may be regarded as gen-
eralizations of the Kummer function.
Together with the Kummer function, the Tricomi function is a solution of an important differential equation. These
two functions constitute the so-called confluent hvpergeomerric functions. The closely related Whittaker functions
are discussed in Section 48:13.
48:1 NOTATION
The symbol ib(a;c;x) often replaces U(a;c:x), and Tricomi used the G symbol. On account of asymptotic expansion
48:6:1, the notation x-°,F0(a.l + a - c:-1/x) is sometimes used for the Tricomi function.
The variable x is the argument of the Tricomi function. The variables a and c are the parameters but, in contrast
to their roles in the Kummer function, they cannot be separately assigned as numeratorial and denominatorial
parameters. The composite variable I + a - c is important in determining the properties of U(a;c:x) and will
sometimes be denoted by b.
48:2 BEHAVIOR
The Tricomi function is defined for all values of a, c and x. However, when x is negative. U(a;c;x) is generally
complex and we therefore exclude the x < 0 range from most of the discussion of the Tricomi function.
As a function of a and c, Figure 48-I shows the number of positive zeros of U(a;c;x), that is, the number of
times the Tricomi function acquires the value zero in the range 0 < x < x. Where zeros are absent, U(a;c;.r) is
invariably positive, being a monotonically increasing function if a < 0 and monotonically decreasing for a > 0.
The behavior of the Tricomi function when its positive argument is close to zero or infinity can be deduced
from the approximations given in Section 48:9.
48:3 DEFINITIONS
In terms of the Kummer function [Chapter 471, the Tricomi function is defined as
r(1 - c) r(c - I)
48:3:1 U(a;c;x) = M(a;c;.t) 4 M(I + a - c;2 - c;x)
r(1 + a - c) r(a).X
471
48:3 THE TRICOMI FUNCTION U(a;c;x) 472
where b = I + a - c and (-1)` = cos(4rc) + i sin(ac). This formula may be rewritten in a large number of
equivalent forms. Thus, either or both of the Kummer functions may be transformed via 47:5:1. Again, any of the
gamma functions may be subjected to the reflection formula 43:5:1, and the resulting cosecant function may be
combined with the trigonometric terms in (-1)`.
The Tricomi function is also defined by the limiting operation
performed on the Gauss function [Chapter 601 or as the Laplace transform (Section 26:141
If the c parameter exceeds a by unity, the result is a simple power function (Chapter 13):
48:4:2 U(a;a + l;x) = x-"
Using these two formulas, and the recursion formulas in Section 48:5, one may deduce expressions for U(a;a ±
n;x) with n = 1, 2, 3, ....
When the c parameter is twice the a parameter, the Tricomi function specializes to a function involving the
exponential and a Basset function [Chapter 5 I ]:
48:4:3
U(a;2a;x) =
a- K,_,n (2)
1/21
ax
Further specialization may occur [see Section 51:4).
With the a parameter equal to a negative integer, the Tricomi function becomes a generalized Laguerre poly-
nomial [Section 23:12]
48:4:4 U(-n;c:x) _ (-1)"n!L;`-"(x)
Simplifications of the Tricomi function for a = 0 or I are
48:4:5 U(O.c;x) = I
and
48:4:6 U(l ;c;x) = x'-` exp(x) r(c - l;x)
The latter function has an asymptotic expansion
48:4:7
U(l;c;x)--1(2-c),= I l-c)J xyx
X ,.O (-x)j c - 1 J=i (-x)i
that represents the simplest of the K = L + I family of hypergeometric functions [Section 18:14). With the aid of
equations 48:4:5 and 48:4:6, recursion 48:5:2 permits expressions for U(2;c;x), U(3,c;x), etc., to be derived.
Definition 48:3:2 can be simplified somewhat when c = 2 and, employing 48:5:1, the result can be transformed
to
i
48:4:8 U(a;O;x) = ((j! x 11 + j 1n(x) + j 4)(a + j) - 2j 4'(j )]
f(1 + a)
This special case of the Tricomi function is related to Bateman's k function:
With the c parameter equal to 1 or ?, the Tricomi function is a parabolic cylinder function (Chapter 46]:
Clearly, these last two expressions can be used in conjunction with recursion 48:5:3 to express U(a;n + 2;x) where
n is any integer.
The limit operation
48:5 INTRARELATIONSHIPS
The important transformation
48:5:1 U(a;c;x) = x'-` U(I + a - c;2 - c;x)
relates two Tricomi functions of common argument.
Recurrence formulas may be written interrelating three Tricomi functions whose parameters differ by unity.
Examples are
2a-c+x I I 1
and
all + a - c) x
48:5:4 U(a:c;x) = U(a + I:c;x) + - U(a;c + l;x)
a+x a+x
I
U(a - l;c;x) -
I+a-c U(a;c - l;x)
a - I+x a-I+x
Analogous to equation 47:5:5 is the argument-addition formula
48:5:5
=u j
U(a;c;x + Y) = Z (a)'( y) U(a + j;c + j;x)
J LvI < IxI
48:6 EXPANSIONS
The function U(a;c;x) may be written as a power series in x by combining definition 48:3:1 with expansion 47:6:1.
The Tricomi function may be expanded asymptotically as
ab a(a + 1)b(b + 1) - (a), (b) 1
48:6:1 U(a;c;x) = x 1 - - + .. + f + x -. x
I x 2x' j!(-x)j
if x is large, where b = I + a - c.
48:7:1 U(a;c;0)
r(1 - c)/r(i + a - c) c<I
c?1
475 THE TRICOMI FUNCTION U(a;c:x) 48:9
m a< 0
48:7:2 U(a;c x) = 1 a=0
0 a> 0
Output f
1
\
If U(,; 1: 3n) = 0.310662341
U(4;3;5) = 0.101573807
f - U(a;c;x) U(1;1;20) = 0.0477185455
where T, = (a),(b);/j!x' and J is the positive integer such that T, is smaller in magnitude than any other T;. No
particular accuracy is claimed for this algorithm, but the precision will increase with x, rarely being acceptable for
x<5.
48:9 APPROXIMATIONS
For small positive values of its argument, the behavior of the Tricomi function depends dramatically on the c
parameter. This leads to a multiplicity of approximation formulas:
r(1 - c) ar(-c)x
48:9:1 U(a;x)
r(I+a-c) - r(1+a-c) c --0
48:9:2 U(a;0;x) = c = 0
r(I a) - r(-a)
r(1 - c) + _ r(c)x'
48:9:3 U(a:c:x) = 0 Gc<I
r(l + a - c) (I - or(a)
-In(x) 2y - kP(a)
48:9:4 U(a;l;x) = r(a) r(a) c
r(c - l)
48 : 9 : 5 U (a;c;x) r(a)x`_'
- (c - 1)r(2r(1- c)
+ a - c)
I< c <2
48:10 THE TRICOMI FUNCTION U(a;c;x) 476
I I
for large positive argument. Abramowitz and Stegun [Section 13.5] give formulas valid when a (but not c or x) is
of large magnitude, or when all three variables are large.
and
d"
48:10:2 ix; U(a;c;x) _ (- I )"(a) U(a + n;c + n;x)
and
f U(a;c;t)dr =
(a-I)r(l+a-c) a>1 c<2
48:10:5
(('
r"U(a;cbr)dr=
r(1 +v)r(a- l-v)r(2+v-c) b>0 0<1+v<a c<2tv
0 b""r(a)r(l+a-c)
The Tricomi function generally is complex valued when its argument is complex or negative. The latter case was
discussed in connection with definition 48:3:3. The case of complex argument is not treated in this Atlas.
48:12 GENERALIZATIONS
Via its asymptotic representation, equation 48:6:1, the Tricomi function generalizes to the unconstrained K = 2,
L = I hypergeometric function [see Section 18:14)
(a), (b), (x)'
48:12:1 6= I +a-c
J-o (Y)i
477 THE TRICOMI FUNCTION U(a;c;x) 48:13
Although they are simply the v = 0 and v = 1 cases of the general function addressed in Chapter 50, the :ero-
order hyperbolic Bessel function la(x) and the first-order hyperbolic Bessel function l (x) are sufficiently important
in their own right to warrant separate consideration. The present chapter also contains some results that relate to
["(x), the hyperbolic Bessel function of arbitrary integer order.
49:1 NOTATION
The names modified Bessel function of the first kind of order zero and modified Bessel function of the first kind of
order one are also applied to la(x) and I1(x). respectively.
The adjective "hyperbolic" indicates that 10(x) and 1,(x) are related to the Bessel functions J0(x) and J,(x) [see
Chapter 521 in the same way that the functions of Chapters 28. 29 and 30 are related to those of Chapters 32, 33
and 34.
49:2 BEHAVIOR
Figure 49-1 shows that both functions increase rapidly in magnitude as the argument x increases in magnitude.
That the rate of increase is somewhat less than exponential is evident from Figure 49-2. The latter map also includes
a graph of I/ 2TCr, the common asymptote to which the product of exp(-x) with any hyperbolic Bessel function
(including those of noninteger order (see Chapter 501) tends as x -. Z.
49:3 DEFINITIONS
Hyperbolic Bessel functions of integer order are defined by the generating functions
f 1
49:3:1 Z t'1i(x) = lo(x) + ( t + ) 10) + r' + I>(t) +
49:3:2 exp(x cos(:)) = 1a(r) + 2 cos(t) I,(x) + 2 cos(2t) 1.(x) + =la(x) + 2 cos( jt)h (x)
i-
379
49:3 THE HYPERBOLIC BESSEL FUNCTIONS lo(x) AND I,(x) 480
49:3:3 exp(x sin(r)) = lo(x) + 2 sin(t) II(x) + 2 (- I)' {cos(2jt) I.,, (x) + sin(r + 2jr) I.,_,(x)}
f
Zf
=2 f exp(tx sin(t)Wt =
IT
J cosh(x sin(t))dt
0
49:3:6 lo(x) =
2
-IT J
o
'--r
cosh(xt)dt
V1
II(X) 2x ('
49:3:7 _ if- J \ cosh(xt)dt
o
481 THE HYPERBOLIC BESSEL FUNCTIONS 10(x) AND 1,(x) 49:3
may be employed to define the hyperbolic Bessel functions of orders zero and unity. Some of these latter definitions
may be generalized:
49:3:11 x-+--f=0
df
d2f
dx
dx2
f=c,Io(2\[x)+c2Ko(2>/)
d'2f dj
49:3:12 xdx2
- - -dx
- xf = 0 f= CIA(x) + c2xK,(x)
:. :....:....:....:....:....:. 0.5
nxp(-z)Io(z)
0. 4
0.3
0.2
49:4 THE HYPERBOLIC BESSEL FUNCTIONS lo(x) AND 1,(x) 482
49:3:13 x-f= 0
1('-f
1(x2
f= c, Vx1,(2V)+c2\K,(2V)
where c, and c, are arbitrary constants and Ke and K, denote the Basset functions (Chapter 521 of orders zero and
unity.
Hyperbolic Bessel functions of orders zero and unity may be generated by the operations of semiintegration
or semidifferentiation acting upon functions involving exponentials [see Chapter 261:
d-"' exp(2x)
49:3:14 Irx
49:5 INTRARELATIONSHIPS
The reflection formula
49:5:1 I,(-x) = n =0.'1. ±2,...
shows that 10(x) is an even function, whereas 11(x) is odd. Hyperbolic Bessel functions of negative integer order
are identical with their counterparts of positive order.
49:5:2 1_,(x) = 1,(x) n= 1.2.3....
but this is not true for noninteger orders [see Chapter 50).
The multiplication formulas
I bIx-x
49:5:3 10(bx) _ I (x)
r0 j. 2
and
1 b2x x) ,
49:5:4 11(bx) = b i 1'.1(x)
!-0l. 2
483 THE HYPERBOLIC BESSEL FUNCTIONS lo(x) AND 1,(x) 49:6
generate series of hyperbolic Bessel functions of integer order. Other series of such functions have the sums
I exx
49:5:5 10(x) + I,(x) + l.(x) + I3(x) + ... = l 1j(x) =
2 2 2
1 l ,(x)= exp(-x)
2
49:5:6 210(x)-lA)+1:(x)-130x)+...=2
sinh(x)
49:5:8 10) + 1,(x) + 16(x) + =
2
1 1
49:5:9
2 2
/
- /\g
\ ``
49:5:14 14(x) = (1 + 24 110(x) 4J ((x)
\\\\\\
x'/ \x + X3
49:6 EXPANSIONS
The hyperbolic Bessel functions of orders zero and one may be expanded as
/ !
49:6:1 10(x)=1+ ++
4
-4
2304
6
+...=
=o
(x'4)
(j!)2
and
_, y 7
49:6:2 li(x)-+s
2 16
+ 384 "
49:7 THE HYPERBOLIC BESSEL FUNCTIONS 10(x) AND I,(x) 484
\x2/ (x2/4)'
49:6:3 l ,(X)
)_a j!(j + n)!
Asymptotic expansions of lo(x) and 1,(x), valid for large argument. are
exp(x) I + 1 9 + 225
+ ... + [(2j).]' + ...J
49:6:4 10(x) _ + x
8x 128x'' 3072x' (j!)'(32x)'
and
49:6:5 I ,(X) -
exp(x)
tax
1- ---
8x
3
- 128x2
15 315
3072x'
- (2j - 3)!!(2j + 1)!!
j!(8x)'
The asymptotic expansion of the general hyperbolic Bessel function l,(x) of integer order is given by equation
-
11
1
x-+x
x- -x x=0 x=x
Idx) x 1 s
1.(x), 6(x). Idx)._. 0
t¢.d. L(x). tAx).... = 0
49:8:2 R,_, _
1
j=J,J-1,...,3,2,1
R, + 2i
x
Replace j by j - 1
Ifj * 0 go to (2)
Replace f by exp(x)/[ I + 2rf )
Output f Test values:
f-10(x)< 10(6.9) = 153.698996
Replace f by frs 1,(6.9) = 142.079028
Output f lo(-3) = 4.88079259
f='I,(x)< 1,(-3) = -3.95337022
[which follows from 49:5:101 is used to calculate R, values for successfully smaller values of j.
The algorithm is designed to produce values of 10(x) and 1,(x) with a relative error of less than 6 x 10-" for
any nonncgative argument x. When those portions of the algorithm shown in green are included, negative arguments
are also admissible.
There are algorithms in Sections 50:8 and 51:8 that may also be used to generate values of 10(x) and 1,(x). As
well, the universal hypergeometric algorithm of Section 18:14 can perform these tasks.
49:9 APPROXIMATIONS
For small arguments, the algebraic approximations
.c
49:9:1 10(x) I+ 1 8-bit precision -1.5 s x s 1.5
d 1"_,(x) + I".,(x) n n
49:10:3 1"(x) = z 1 (x) = 1"-A) + U-0)
dx 2 .r
The formulas
49:10:8
f tl,(t)dt = n(x) eo(x) - lo(x) ((x)1
Here a-,, Co and e, denote hyperbolic Struve functions [Section 57:13]. The indefinite integrals
(x-y2
4 ixv)I
21,
_ x' - y- x - 6x'y2 + y'
49:11:1 I0(x+iv)=I 1+ +
i-o (j!) 2
4 64
(xvx3v_xv))
x: v2 ixvll'
49:11:2
11(x+iv)=x+iy
a t 21 =(X+ x3-3xy +xs-IOx3y2+5xy'+
2 -o j!U + 1)! 2 16 384
487 THE HYPERBOLIC BESSEL FUNCTIONS lo(x) AND 11(x) 49:13
+i(y+3x'y-y'+5xy- 1Qx'y'+y'+...
\2 16 384
For purely imaginary arguments, these formulas reduce to
49:11:3 1o(iy) =1o(Y)
49:11:4 1,(iY) = iJ,(y)
which generalize to
i
1.0
x t ix (iix'/2)'
49:1 1:7 1,(x ± ix) _ = bei,(x V 2) + i ber,(xV2)
2 _o j!(j + 1)!
where ber" and bei" denote Kelvin functions, as discussed in Chapter 57.
49:12 GENERALIZATIONS
The next chapter discusses hyperbolic Bessel functions of arbitrary order.
The function I,(x) is defined for all values of its order v. Many physically important functions arise when v acquires
special values.
50:1 NOTATION
The names modified Bessel function of the first kind and Bessel function of imaginary argument are commonly
applied to l,(x), but these titles are less informative than the name hyperbolic Bessel function that we adopt. The
adjective "hyperbolic" indicates that the relationship [see 50:11:21 between I,(x) and the Bessel function itself [J,(x),
see Chapter 531 resembles that between hyperbolic functions [Chapters 28-30] and circular functions (Chapters
32-34). See Section 50:4 for the i,(x) symbol.
The parameters v and x are named the order and argument, respectively, of the function.
50:2 BEHAVIOR
Unless v is an integer, the hyperbolic Bcsscl function I,(x) is complex when its argument is negative. Accordingly,
this chapter concentrates on positive values of x. and this is the only range depicted in the contour map, Figure
50-1.
When the order v is an integer, the functions I,(x) and 1_,(x) are identical, but these two functions are also
close to each other in value whenever x exceeds jvl. Moreover, for large enough argument, the hyperbolic Bessel
function I,(x) becomes almost ind ndent, not only of the sign of its order, but also of the magnitude of v, being
well approximated by exp(x)/V 2ax.
Whereas 1,(x) is generally complex for negative x. the composite function x-"I,(x) remains real for all real
arguments and is, in fact, even so that
The evenness is evident on writing 50:2:1 as 2"C,(-x2/4), in terms of Clifford's notation [see Section 53:2].
489
59:2 THE GENERAL HYPERBOLIC BESSEL FUNCTION I,(x) 490
491 THE GENERAL HYPERBOLIC BESSEL FUNCTION I,(x) 50:4
50:3 DEFINITIONS
The hyperbolic Bessel function is defined by a number of definite integrals. including
Substituting t = cos(h) in 50:3:1 gives a second definition and some others are listed by Gradshteyn and Ryzhik
[Section 8.4311.
A Kummer function [Chapter 471 whose denominatorial and numeratorial parameters are in a two-to-one ratio
is related to a hyperbolic Bessel function by
(x/2)" exp(-x)
50:3:2 1,(x) = M( + v;l + 2v;2x)
r(l + v)
The operations discussed in Section 43:14 can generate any hyperbolic Bessel function from any other hyper-
bolic Bessel function and hence from any of the functions discussed in Chapter 49. For example, in the notation
of Section 43:14
lox)
-v r(+ v) 21`
50:3:3 (x/2), Mx) x=
0
A solution of the modified Bessel equation
d'f df
50:3:4 .r-+x--(x'+ir)f=0
dx' d.r
is c,I,(x) + c2unless v is an integer [see Section 51:3 for that ease]. The terms c, and c2 arc arbitrary constants.
Other differential equations that are satisfied by hyperbolic Bessel functions are
and
dZf df
50:3:6 x + (v + 1) -f = 0 f = c,(2\)-" 1.(2V) + c.,(2V)'I_.(2V )
dx- dx
unless v is an integer.
50:4:4 I2 J1(x) =
ax
sinh(x)
I cosh(x)
[
x
ax
x-!
tax
exp(x)
tax
2ax
+ s
(x+l) exp(-x)
2ax
50:5 THE GENERAL HYPERBOLIC BESSEL FUNCTION 11(x) 492
and others may be constructed via recursion formula 50:5:1. Some of these functions are mapped in Figure 28-3.
The s} mbol i (x) and the name modified spherical Bessel function of the first kind are sometimes applied to the
Ni it/2x I., 112W function.
Hyperbolic Bessel functions of order ± are closely related to Airy functions [Chapter 561. One has
n
50:4:5 4X [Bi(X) = V /Ai(X)I X= (2x
Hyperbolic Bessel functions of order ±14 are expressible as parabolic cylinder functions [Chapter 46] of order
-j. The relationships are
13,14-2N6) + D_112(2\)
50:4:6 I-114(x) _
x1/4
and, hence, with the help of recursion formula 50:5:1, so may 45/4(x), etc.
50:5 INTRARELATIONSHIPS
Hyperbolic Bessel functions satisfy the recursion relationship
2v
50:5:1 I,. (x) = I,_,(x) - -11(x)
X
50:5:2 11.(bx) =
b''
r- - ) 11.;(x)
2 j!
Setting b = I + (y/x) converts 50:5:2 into an argument-addition formula, while setting b = i = generates
the summation formula
The hyperbolic Bessel functions 1,(x) and I_,(x) are identical if v is an integer; otherwise:
2
50:5:4 I_,(x) = 1,(x) + - sin(vir) K,(x)
it
where K, is the Basset function [Chapter 51]. This equation may be regarded as an order-reflection formula, in-
terrelating I_,(x) and I,(x). There are similar order-reflection formulas for the product of two hyperbolic Bessel
functions whose orders sum to ± 1. namely
2
50:5:5 1_,(x) 1,_1(x) = I,(x) 1_,1,(x) + - sin(vtr)
ax
and
2
50:5:6 1-,-112(x)1,-112(x) = 1112..(x) 112_,(x) + - cos(va)
'nx
493 THE GENERAL HYPERBOLIC BESSEL FUNCTION I,(x) 50:8
50:6 EXPANSIONS
(;-v))(;+v);
j!(2x)'
+
1
xc
which is valid if x, but not jvi, is large. This series terminates when v = ±; ±? ±i .. but the expansion is
not exact even under those circumstances.
44c) v>0
b(x)
I,(x) -I <v<0.-3<v<-2, -5<v<-4, ...
1,(x) v=-1,-2.-3,.
Ux) -2<v<-1.-4<v<-3.-6<v<-5....
As is evident from the contour map, Figure 50-I, the hyperbolic Bessel function 1,(x) encounters a zero, for
.r > 0. only if the order is negative and in one of the ranges -2 < v < - I, -4 < v < -3, -6 < v < -5, etc.
For other negative values of v, I,(x) generally displays a minimum.
z'
X2
+...+I/8(2+v)4(1+v)+1 (x/2)''
i(1+v)
when x is less than 8 + (v2/12). The I /17(1 + v) multiplier is calculated by a routine that is essentially the one
used in Section 43:8, but special measures are taken if v = 0 or if v is a negative integer. The integer J in 50:8:1
is calculated by an empirical expression to ensure 24-bit precision (the fractional error in the hyperbolic Bessel
function does not exceed 6 x 10-"). If x exceeds 8 + (v2/12), the hyperbolic Bessel function is calculated via the
formula
2)' - v2 l (J - )2 - V2 (J - 1)2 - v2
50:8:2 l,(z) = l l
2A +I/ 2(J - 1)x + I / 2(J - 2)x
v22 1 (2)2 - v2 exp(x)
4x 2x 2arx
which is a concatenated version of the asymptotic expansion 50:6:4, truncated by an empirical choice of J, again
to ensure 24-bit precision.
A simpler alternative is available in Section 49:8 for calculating hyperbolic Bessel functions of integer order.
Several versions of the universal hypergeometric algorithm Isce Section 18:141 are also useful for determining
numerical values of l,(x).
50:9 APPROXIMATIONS
If v is not a negative integer, the hyperbolic Bessel function of small positive argument is approximated crudely
by
2ax\ -
8
is valid and shows that increasing its argument makes the hyperbolic Bessel function increasingly independent of
its order and increasingly closer to exp(x)/ti x.
Making use of the reflection property [Section 43:5] of the gamma function, approximation 50:9:1 shows that
sin(vrr)
50:9:4 I,(x) 1_,(x) -
I
r(1 + v) r(1 - v) va
x-0
a result that remains valid for all orders. It may be combined with equation 50:5:4 to produce the approximation
50:9:5 ----2vK,(x)
1
I,(x) I_,(x)
1
x-.0 v*0.±1, ±2, ...
50:10:2 fL(z)ch= 2
i=o
(- I,(x)
and not in closed form. On the other hand, the indefinite integrals of the following products:
r( +,u
50:10:3 j t`,(t)dt = [I,(x) C,_1(x) - 1,_i(x)l.(x)] v > -
0 2
50:11 THE GENERAL HYPERBOLIC BESSEL FUNCTION 1.(x) 4%
2-"
50:10:5 t-'I,. 1(t)dt = x-`1.(x) -
r. r(1 + v)
x, exp(+z) 21-' 1
50:10:6 C' exp(I 1) I.(t)dt = -" ±I._I(x) - l.(x) + v*
J0 2v - I - I (2v - I) r(v) 2
are straightforward. The f functions in 50:10:3 are instances of the hyperbolic Struve function discussed in Section
57:13, while r denotes the gamma function [Chapter 43).
Among formulas for definite integration is
n
r l
/ z\
50:10:7 J exp(-B'tz) !"(bt)dt = 2B exP\8B2J !"/zI ii2 I v > -1
o
Replacing the argument x by 2N/-x in the first of the two Rayleigh formulas
1d
50:10:8 {x dx} [x"I.(x)) = x"-"I.-.(x)
50:10:9
f x dxf [x_"I"(x)J = x-"-"1...(x)
for multiple differentiation leads to a relationship that generalizes to the very simple expression
d" [xvn-I.(2f )J = xr"-" r_4
50:10:10
dx"
Here the operator d"/dx" signifies differintegration with lower limit zero to an arbitrary (positive or negative, integer
or noninteger) order [see Section 0:10).
where 4 = (2j + v) (8 + 2ka), 0 and k having retained their significances from Section 13:11.
When the argument is purely imaginary, the hyperbolic Bessel function becomes a Bessel function [Chapter
531 of real argument:
r / \ / l
50:11:2 I.(iy) - i"J"(y) = I cosy 2 I + i2 )j My)
(2 )
] when its argument is negative. We
Unless its order is an integer, the hyperbolic Bessel function is complex
have
50:12 GENERALIZATIONS
The function
(x "," . (x2/4)'
50:12:1
\2) 0 r(1 + µ + j)r(1 + v + j)
497 THE GENERAL HYPERBOLIC BESSEL FUNCTION l,tr) 50:13
closely related to the Kummer function [Chapter 47], is a generalization of the hyperbolic Bessel function, which
is the µ = 0 instance of 50:12:1. The hyperbolic Struve C function [Section 57:131 is also a particular instance of
the general function 50:12:1.
Because the Basset function of noninteger order is related so simply [see 51:3:51 to the hyperbolic Bessel function
of the previous chapter, this chapter concentrates on the Basset functions Ko(x), K,(x). K2(x), ... of integer order.
51:1 NOTATION
Alternative names for the Basset function are the modified Bessel function of the third kind, Bessel's function of
the second kind of imaginary argument, Macdonald's function and the modified Hankel function.
We use v generally to represent the order of a Basset function but replace this symbol by n to specify integer
order.
51:2 BEHAVIOR
The Basset function K,(x) is infinite for x = 0 and complex for x < 0 [see equation 51:11:31. Accordingly, we
restrict attention to x > 0 here and generally throughout this chapter.
For all v, K,(x) is a positive and monotonically decreasing function of its argument x, approaching zero as x
- %, in accordance with expression 51:9:6, in a manner increasingly independent of the order v. However, the
approach to infinity as x -. 0 is a strong function of v as detailed in Section 51:9.
For a constant positive argument, K,(x) is an even function of its order v, as evidenced by equation 51:5:1.
Moreover, for constant positive argument, K,(x) invariably increases as wj increases.
Figure 51-1 shows maps of K,(x) for n = 0. 1, 2, 3. 4. 5 and 6. The curves for noninteger order Basset
functions smoothly interpolate between these mapped curves: for example, Ka12(x) lies intermediate between K4(x)
and K5(x).
51:3 DEFINITIONS
Basset functions may be defined via those Tricomi functions [Chapter 481 in which the a parameter is a moiety of
the c parameter.
51:3:1 K,(.,) _ \(2x)" exp(-x) U(12 + v; I + 2v; 2x) v >- 0
499
51:3 THE BASSET FUNCTION K.(x)
40
e1, U O
00 y s $4'O U
*
t '.0 O ti b
,V r0 0 0
4.,y. rtj.'L
ifL
,y0 Rr 4 Nf V y'1.
if i if
2.4
..:.. I .:... .......... .......: .... ............ .... :.... .... ............. . 2.2
2.0
0.2
or approaches infinity:
Equivalent to 51:3:1, but somewhat simpler, is the definition of a Basset function as a special case of Whittaker's
W function [see 48:13:61.
Among definite integrals that define the Basset function are
51:3:3 K,(x) _
exp(-x/)dt v) - 1
F( , + v) (2x)'
2 f (r' - 1) 2
and
c
r(3 + v) cos(t)dt I
51:3:4 K,(x) _ (
)" f. & + v 2
and a large number of alternatives were assembled in the Bateman manuscript (see Magnus, Oberhettinger
and Tricomi, Higher Transcendental Functions, Volume 2. pages 82 and 831. Because of relationship 51:5:1. the
sign of v may be changed in definitions 51:3:3 and 51:3:4.
Basset functions of noninteger order may be defined in terms of hyperbolic Bessel functions by
501 THE BASSET FUNCTION K,(x) 51:5
Tr[l_,(x) - l,(x))
51:3:5
2 sin(vrt)
v*0,±1,-'2,...
but this expression must be written as a limit:
IT 1.(x)1
51:3:6 K(x)2limSl
sin(m) Jr
51:3:7
x. d/+xdf-(.Y+n')f=0 n=0,±1,±2....
dx' dx
is solved by f = c,l (x) + c;K (.r). Other differential equations whose solutions involve the Ko or K, functions are
listed as 49:3:10-49:3:13.
Semiintegration with lower limit zero is a powerful method of generating Basset functions. Examples include
51:3:8 K.
I _,v'rrexp\2x
2x
(1 d.r-"2 xeap d- uz 1
r
xvrt '1
51:3:9 K, ( l 2
exp (
7
l
51:3:10
I
Kv< 2x> = x V to exp
' /- 1 d- vx I
_
x=) d.r ''= eXp .r
51:5 INTRARELATIONSHIPS
With respect to its order, the Basset function is even
51:5:1 K_,(x) = K,.(x)
51:6 THE BASSET FUNCTION K,(x) 502
and
1
51:5:3 K,.,r(x)1,(x) _ - - K,(x) l,.r(x)
X
The zero-order Basset function satisfies the argument-addition and -subtraction formulas
By sufficient applications of formula 51:5:2, any Basset function K"(x) of integer order may be expressed in
terms of and K,(x). The first two examples are
2
51:5:5 K,(x) = K((x) + - K,(x)
x
4 / \
51:5:6 K3(x) = Kd(x) + 1 I + 8, I K,(x)
x \ x'/
and expressions for K,(x). K5(x) and Ke(x) are analogous to 49:5:13-49:5:15. but with uniformly positive signs.
51:6 EXPANSIONS
The Basset function of noninteger order is expansible as the sum of two convergent series
--I
//+ ,
I(v) (xr (x'/4)1 r(-v, Ix\ (x=/4)'
51:6:1 K,(x) = + I
=p j!(l + v),
v*0,'_1,±2....
2 2 _O j!(1 - v), 2 2
that coalesce only if v is an odd multiple ofFor integer order the series are
r /x\ x\\ 11 (x- 4) I x 11 (x' 4)-'
51:6:2 K0(x) y - In{ 2 I + +l- ln(2 1J (1 , + [-,y + I + 2 In(2)J (2 ).
\/
51:6:3 K1(x)
-y + 1 + l
=--
x
1
--l
2
2
-y + I -
+
2
I
3
+I+2+6-1n\2.,J
HI
\1
- ln(x
\2/
ln(x
\2/
I
(x /4)
(3!)2
(x2/4)r12
0!I!
+ ...
- ---
(x'/4)3''
-y + I +
2
1
4
In
x
2
(x-'14)312
l!2!
2!3!
and generally
51:6:4 K"(x)=-(l 1
2 \x/ ;=o j!
r
(n -j- 11.
x
\4 / L
Vb(l+j) dr(I+n+j)
2
+
2
+I - x2) j!(n (x/2)''"
+ j)!
In these formulas r denotes the gamma function [Chapter 43], s the digamma function [Chapter 441, (1 - v), is
ln(J rr
a Pochhammer polynomial [Chapter 181 and y is Euler's constant (Section 1:71. Alternatively. Basset functions of
integer order may be expanded as Neumann series in terms of hyperbolic Bessel functions:
I ' (1 + 2j)
51:6:6 K,(x) _ - -y + I - 10) + - lo(x) - It+2,(x)
2l x J=i J(1 + J)
-I(x)+-j x I(.r)+(-I)LLL
and generally
K,(x) -
v' (1-y')(1v') - (11 -v')(i-y')(1-v')
51:6:8 - exp(-x) f I - 4 2x
+
8xz 48x'
-v),(1+v)i
0 l
j!(-2x)' 1
is generally asymptotic but it does terminate when of is an odd multiple of 1, and under these circumstances the
expansion is exact. If vj is not an odd multiple of 1 and lies between the numbers n - ; and n + 1, where n is
an integer, then the terms in series 51:6:8 corresponding to j = O. j = 1, j = 2..... j = n are uniformly positive
(we are treating only positive x), whereas the j = (n + I)m term is negative; thereafter, the terms alternate in sign.
Accordingly, it follows from the properties of alternating series [see Section 0:5] that if the series 51:6:8 is truncated,
terms after j = J being ignored, then the partial sum is related to the true value of K,(x) by one of the inequalities
x - 0 x * x
K, x) 0
than an empirically chosen integer J. Recurrence 49:8:2 then serves to calculate all other R, values. The hyperbolic
Bessel function 1,(x) is calculated by the algorithm as R010(x) and thence the Basset function K,(x) as R0{[I/xl,(x)]
- Ko(x)}, an identity that follows from 51:5:3. Many problems in applied mathematics require values of the func-
tions 10(x) and 1,(x) as well as those of K0(x) and K,(x); the algorithm below delivers all four functions if the
portions shown in green are included.
Replace j by j -
R+
X
accuracy may be impaired.
2j
SetR= 1 Ir+
Replace j by j -
Ifj*0goto(I)
Replace I by [exp(x) + exp(-x)1/12 + 4IrR]
Output I
10(x)< ««
Replace K by I14KrR - In(.890536209x)] Test values:
Output K 10(1) = 1.26606588
K = K0(x) «« K0(l) = 0.421024438
Replace I by IR 1,(1) = 0.565 1 5 9 /04
Output I K1(1) = 0.601907230
<<<<< K0(3) = 0.0347395044
Replace K by R[(1/xl) - K] K,(3) = 0.0401564311
Output K K((5) = 0.00369109833
K - K,(x) «« K1(5) = 0.00404461344
Although the accuracy of this algorithm is inherently high, it may not deliver precise values of K0(x) and K,(x)
unless x is small. The difficulty arises because equation 51:8:1 calls for the differencing of two terms whose values
become increasingly close as x increases; therefore, significance is lost in their subtraction. The seriousness of this
effect depends on the number of significant bits utilized by the computing device. Changes to the algorithm cannot
ameliorate the problem.
The second algorithm utilizes expansion 51:6:8 and overcomes its asymptoticity by invoking the Pad6 technique
as described in Section 17:13. An empirical formula is used to determine the order J, up to a maximum of 30, of
the diagonal Padd approximant Rjj that is generated. Within the permitted ranges of x and v, K,.(x) is generally
returned with a precision better than 24 bits.
Setq=0
Set k=j
SOS THE BASSET FUNCTION K.(x) 51:10
51:9 APPROXIMATIONS
From expansions 51:6:1-51:6:4, the following two-term approximations, valid as x -' 0, may be derived:
r(v)x-" r(-v)x'
51:9:2 K,(x) = + 0<v<I small x
2' 2"
51:9:3 K,(x) =
I
+ x InrV = I small .r
X 2 (x2)
/2 l
51:9:4 K,(x) = r(wl)l w, > I smallx
x/ [2 - 81vl-8,
More accurate than 51:9:4, for ivl > 2, is the approximation
J
For large x, the approximation
d K,(x)
51:10:3 K,(x) = -K0(x) -
dx x
and lead to the derivatives
d x_'
51:10:4 K"(x) = -x" K,,,(x)
dx
The formulas
rx
51.10:5 Ko(t)dr = [Ko(x) f-,(x) + K,(x) fo(x)]
0 2
and
=
for the indefinite integrals of the zero-order and unity-order Basset functions may be obtained as special cases of
the general expressions 51:10:7 and 51:10:9 below. In equations 51:10:5 and 51:10:7 f is the hyperbolic Struve
function discussed in Section 57:13. Closed-form expressions are available for the following indefinite integrals of
the Basset function multiplied by a power.
i v>-2 1
x` exp(x
J t-` cxp(t) K,(t)dt = IK,(x) * K,_,(x)] v>-
2v- 1 2
The important King's integral is the v = 0 instance of 51:10:10 with upper signs selected.
For vi > I the definite integral fK,(t)dt between limits of r = 0 and t = x diverges. The convergent cases
are the µ = 0 instances of the general formula
For purely imaginary argument, the Basset function is related to the functions of Chapters 53 and 54 by
When the real and imaginary components of the argument' of a Basset function are equal in magnitude, we ha
51:11:2 K,(x ± ix) = i'` [ker,(x y 2) t i kei,(x V G)]
where ker and kei are Kelvin functions [Chapter 55] and i'" is to be interpreted as cos(vw/2) ± i sin(vir/2).
The Basset function is complex for negative real argument. The real and imaginary parts are given by
is
51:11:3 K,(-x) = cos(vir)K,(x) - Z [I,(x) + 1 _,(x)] v * 0, ± 1, ± 2, .. .
51:12 GENERALIZATIONS
The Basset function may be generalized to Whittaker's function [Section 48:131
52
THE BESSEL COEFFICIENTS Jo(x) AND JI(x)
Those Bessel functions [Chapter 53) in which the order is a nonncgative integer are known as "Bessel coefficients"
and are the subject of this chapter. Emphasis is placed on the most important of the functions: those with n
= 0 and 1.
52:1 NOTATION
Bessel coefficients are also known as Bessel functions of the first kind of nonnegative integer order.
52:2 BEHAVIOR
All Bessel coefficients are oscillatory functions whose oscillations become increasingly damped as their arguments
approach large values of either sign. As Figure 52-1 shows, all Bessel coefficients except J0(x) are zero at x = 0.
Moreover, as n increases, J,(x) retains near-zero values over an increasing range in the vicinity of x = 0; for
example, J9(x) does not exceed 0.01 in magnitude in the range -5.4 5 x 5 5.4.
Some regularities are apparent in Figure 52-1. Notice that each local maximum or minimum of J0 corresponds
to a zero of J, but this rule does not extend to larger orders. On the other hand, for n z 1, each local maximum
or minimum of J. corresponds to a point of intersection of the and J,,,, curves. Moreover, at each zero of J.
1,_, and 1,., have equal magnitudes but opposite signs.
For x s 0, each Bessel coefficient encounters its first (and largest) maximum at an argument that, for small
n, is close to nir/2. This rule fails for larger n, however, and for very large order, the Bessel coefficient
attains its first maximum close to x = n. Subsequent, and ever-smaller, maxima are encountered with a spacing
of approximately 2a. Local minima occur approximately midway between consecutive maxima. For x > 0 the first
minimum is the largest and subsequent minima decrease progressively in size. A zero is encountered between each
minimum and its adjacent maxima so that the spacing of the zeros is approximately IT.
52:3 DEFINITIONS
Bessel coefficients are defined by a number of generating functions. Those of even order arise from
52:3:1 cos(x cos(t)) = Jo(x) - 2J,(x) cos(2t) + 2J,(x) cos(4t) - = Jo(x) + 2 (- 1), J,t(x) cos(2jt)
so
52:3 THE BESSEL COEFFICIENTS Jo(x) AND 11(x) 510
. . . . . . . . . . . . . . . .
1.0
. . . . . . . . . . . . . .
:Jo(x - - - - -
........... .........................................................................
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . .
FIG 52-1 :
:J.(x):
:....:.. /.....:. \..:. !.:.. :... .... J ():
Jz (x): - .Tw (z>' 0.4
and if the signs in this series are made uniformly positive, the terms are generated by cos(x sin(s)). Bessel coef-
ficients of odd order arise from
52:3:2 sin[x cos(t)] = 2J,(x) cos(t) - 213(x) cos(3t) + 215(x) cos(5t) - - = 2 E (-I)'J,,,,(x) cos(t+2jt)
i-o
or from the corresponding series for sin[x sin(r)], which differs from 52:3:2 in that sines replace cosines, and all
terms are positive. The\ generating /function
52:3:3 expl xt=2r x I = 10(x) + I r - J1(x) + r= + r. lz(x) + t' - Ily 33(x) + NJ; (z)
The following are among the integral definitions of the general Bessel coefficient
52:3:7
d2f df
x- + - + xf = 0 f = c,J0(x) + c2Yo(x)
dx' dx
52:3:8 xdzf+df+f=0
dx
f=c,Jo(2V)+c2Yo(2Vx)
dx
df
52:3:9 x dzf
dx2
- dx + xf = 0 f= c,4J,(x) + c2xY,(x)
52:3:10 -+f=0
x-dz '
f=c, VxJo(2Vx)+c2VxYo(2Vx)
'
dX2
Here c, and c2 are arbitrary constants and Yo and Y, are Neumann functions [Chapter 541 of zero and unity orders.
The Bessel coefficients JO and J, are generated by applying the operations of semiintegration or semidiffer-
entiation, with lower limit zero, upon certain functions involving sinusoids. Examples include
52:3:11
52:3:12
Jo(x) =
sec(x)
d-
dx'ir
Jo(V x)
()co s(2x)
2
V. dx'2
1
d"1
=
d-"
'
csc(x) d-'/' sin(2r)
dx-"2
(sin(V x))
()
52:3:13 1,(1f) _ 'rrX
dx uz
(sin(
l 1 - JxdxI
d 12 (sin(l/\)/
52:3:14 Vrx 1\ x
Jo\
52:5 INTRARELATIONSHIPS
Bessel coefficients are even or odd
Many formulas exist for infinite sums of Bessel coefficients. Some of these, such as
1
52:5:3 J,(x) - J3(x) + J5(x) - Jr(x) + = 2 sin(s)
1 1
52:5:4 2 J0(x) - J,(x) + J,(x) - J6(x) + = 2 cos(x)
and
1 1 - 1
52:5:5 J0(x) + J2(x) + l1(x) + J6(.r) + = J,(x) =
2 2 2
may be derived as special cases of formulas 52:3:1-52:3:3. Others, for example:
52:5:9 + 1)!(n + 4)
(n- I) !nJ (x) + On + 2) J,,.,(x) + (n
2!
1,.2,(x)
_-
x
i-o j! 2
follow from the properties of Neumann series (see Section 53:14]. Yet others, such as
52:5:11 J0(x) J,(x) - Jdx) J2(x) + J2(x) J3(x) - ... _ (- I )` J,{x) J;.i(x) J,(2x)
;=o
may be derived from Neumann's addition formula
52:5:12 n=0,±1,±2,...
Included among still other types of expansions are
x
52:5:13 1,(x) + J2(2x) + J3(3x) + _ J(ix) _
,_, ' 2(1 - x)
and
52:5:14 121x)+1;(2x)+1;(3x)+...=iJ;(jx)=2[ 1 - 1J
that are examples of Kapteyn series [see Erd6lyi, Magnus, Oberhcttinger and Tricomi, Higher Transcendental
Functions, Volume 2, pages 66-68).
513 THE BESSEL COEFFICIENTS J6(x) AND J,(x) 52:6
52:5:19 JS(x) =
I394 - 72
Zx +
1
I) J,(x) - l 3
( x192 - 121
-J Jo(x)
x
52:6:2 lo(x)=I-+
4
.r- x6 +
64 2304
X2
x
147456
B
-...= (-xz/4);
and
52:6:3 x
JI(x)=- x + X5
- xt +...= x (-x/4);
2 16 384 18432 2 j!(j + 1)!
Expansions as infinite products take the form
52:6:5 J,(x)=zll-
/ z
fIl
z
J'. :z
-Z
J,a
z
=
r-, JI.*
where j,x denotes the k'° zero of the Bessel coefficient J,(x) [see Section 52:7].
52:7 THE BESSEL COEFFICIENTS Jo(x) AND 11(x) 514
By making use of formula 52:5:22, it is possible to expand a Bessel coefficient J,(x) in terms of the Bessel
coefficients J J,,,,, J,-,, ... at any fixed nonzero argument. For example, choosing 2 for this fixed argument
gives
(z\ ` 1/ -x
52:6:6 J,(x) I 1 4 J,",(2) n = 0, 1, 2, ...
2J -old
X -' x
52:6:11 J()-
1
sin(x)
I +-+--
3
8x
15
128x=
315
3072x3
cos(x)
1 -+-+3
8x
15
128x
315
3072x3
x= -x x= 0 x= s
1,(x) 0 I 0
1i(x), J,(x), J ,(X), ... 0 0 0
Those positive argument values that cause 1,(x) to acquire the value zero are denoted j ,,
is called the km zero of the na Bessel coefficient. The value of the derivative d1,(x)/dx at the zero is a quantity
whose value is needed in certain practical problems [see Section 52:141. Such a derivative is known as the associated
value of the zero, and the usual notation is 1' .0.,.). although two alternative notations follow from the identity
52:7:1 J,(J,:A) = J,-1(jn:k) _ - 4*1(1.:A)
515 THE BESSEL COEFFICIENTS J0(x) AND 1,(x) 52:7
Table 52.7.1
which is a consequence of 52:10:1. Approximate values of the first seven zeros, and their associated values, for
n = 0 and 1. constitute the left-hand side of Table 52.7.1. Other approximate values of may be read from the
map in Section 53:7 (Figure 53-2). More precise values of j,.k and J,( j,:k) are available from the algorithm that
follows.
The algorithm operates in two modes according to the relative values of n and k. If k >: n - 3, the algorithm
adopts x, = a(2k + n - 1)/2 as an approximate value of and then uses Newton's method [Section 17:7] to
generate the improved approximation
J,(x1) x,J,(x,) _ x,
52:7:2 x,
JR(xi) - r' n - x,R.
nJ,(xi) - x,J,,.1(x,)
and this procedure is repeated until two successive approximants differ by less than l0-'. The method used to
compute R,,, the Bessel coefficient ratio, is described in Section 52:8. When k < n - 3, the algorithm first calculates
n + 2n'"' [see 53:7] as a crude approximation to j,,, and refines this approximation by a single application of 52:7:2.
A crude approximation to j.4, is then calculated by adding 4 to the j,,., approximant and this, in turn, is refined by
a single application of Newton's method. The procedure of adding 4 and refining is repeated until a crude value
of j,;k is reached. This is then improved by repeated applications of formula 52:7:2. The second half of the algorithm,
which calculates -J,.,(j,;k), is essentially the same as the algorithm in Section 52:8.
If k > n - 4 go to (4)
Set x = n - 4 + 2n'13 [orn-4-0.8n"'[
(I) Replace x by x + 4
Set I = Int n2 + 2(x + 6)2
Replace J by J + 1
Go to (6)
121 If J * I go to (3)
Replace.r by.r - (n114)
(3) IfJ<kgoto(1)
Set J = 0
(4) Set L = 2 + 2 Int [n'/4] + [(.r + 7)2/2] Input restrictions:
(5) Set ! = L n = 0. 1, 2....
(6) Set R = 0 k 1.2.3....
21
(7) If 21 - R go to (8)
X
52:7 THE BESSEL COEFFICIENTS Jo(x) AND J,(x) 516
21
Replace R by I AX- - R)
Replace I by I - I
If I > n go to (7)
n
If - =R go to (8)
X
SetD= I/(nx - R)
Replace D by x'/[D(n' - n - .r' + _rR)[
Replace x by x - D
If J * 0 go to (3) (or to (2))
If IDS > 10-7 go to (5)
,,, (8) Output x
x or } <
SetD=J= I
If x=0go to (II)
(9) Replace R by I /(- x
- R + 10-'S
Test values:
IfL - n> 1 [or>0)goto(10) j,-, = 8.41724414
Rep lace J by JR J:021) = 0.271382590
(L )
(10) Replace D by DR + 2 frac .ho-s = 28.8873751
2 J;0(j10s) = 0.143812670
Replace L by L - I jig, = 0.000000000
If L *0goto(9) Joljv.,) = 1.000000000
Replace J by J/(1 - 2D) [or by J/(2D - 1)) j., = 6.70613319
J
((J'(j,;d
or
J,0.Al
} ««
(11) Output J 1,(j__) = -0.3 1 3 5 304 45
jiu5 = 27.18202/5
J,n(jios) = 0.158631021
Also important in a number of physical applications are those nonnegativc values of the argument that cause
J (x) to attain an extremum (a local maximum or minimum): we call these values extrema of the n' Bessel coefficient
and denote the k' such value by j;,,. Some approximate values of the extrema of J0 and J, occupy the right-hand
side of the table above, together with the corresponding associated values of the extrema of the n a Bessel coefficient
These associated values obey the relationships
x
52:7:3 J.(x) J..,(x) = X- x=
n n
Note also that ja;A.' = j and that Jo(jo,,-,) = J;(j,;k). Figure 53-2 is useful for finding approximate values of the
extrema, and accurate values of both j,;,A and are given by the algorithm above if the additions and modi-
fications shown in green are instituted. The procedures adopted by the modified algorithm generally resemble its
unmodified counterpart. but the Newtonian improvement formula is
x;J,,.,(x,) _ x,(n -
52:7:4
n' - n - x2 + x,R
in this case.
Whether used for finding zeros or extrema, the algorithm of this section is slow, particularly for large n and/
or k.
517 THE BESSEL COEFFICIENTS Jo(x) AND J,(x) 52:9
R,-,=-
1
52:8:1 1=1,2,3,...
--R,
x
and this formula has the useful property that an error in R, generally induces a smaller error in R,_,. This means
that the concatenation
The algorithm of Section 53:8 may also be used to calculate J.(x) for x > 0 and the universal hypergeometric
algorithm (Section 18:14) can perform this calculation as well.
52:9 APPROXIMATIONS
For small arguments, the first two Bessel coefficients are well approximated by
(( : 2
52:9:6 j. A
{ K= (k+)w
3 3 1179
+ --
8K 128K3 5120K5 // 3
52:9:7 )U i K Ik-4 it
52:9:8 j;A-K-8K
7 431
384K'
29893
15360K5
_
K= k-4/1\
for multiple differentiation will be recognized as binomial coefficients [Chapter 61 of alternating sign.
Indefinite integrals of Besse] coefficients include
arx
52:10:4 Jo(i)dr = xJo(x) + [J,(x) ho(.r) - Jo(x) h,(x)) x>0
2
and some others may be obtained by specialization of formulas in Section 53:10. The ho and h, functions in formula
52:10:4 are discussed in Chapter 57.
Among important definite integrals are
J br
52:10:7 I n= 1,2,3,...
r
0 B<b
52:10:8 ¢ cos(Bt) J0(bt)dt =
J0
1
62_b2
B>b
r ' /- 1 62
52:10:9 exp(-Bt) Jo(b V r)dt = B exp(48
Jo
along with many others that follow from formulas in Section 53:10.
With n = 0 and 1, Erdblyi, Magnus, Oberheuinger and Tricomi [Tables of Integral Transforms, Volume 2,
pages 9-21] list many definite integrals of the form
and
d "2 J0(V) d"2 sin(V x)
52:10:12
dx 'r 2 - dx' 2 V;
52:11 COMPLEX ARGUMENT
We discuss specific instances of iv), but not the general case.
When the argument of the Jr Bessel coefficient is purely imaginary, one has a function that is real or imaginary
according to the parity of n:
l(y) n0,4.8,...
52:11:1
n 1,5,9,...
_ I (y) n = 2, 6, 10...
-A,(y) n=3,7,11,-..
Here I is the hyperbolic Bessel function of Chapter 49.
When the real and imaginary components of its argument are equal in magnitude, the real and imaginary
components of a Bessel coefficient correspond to Kelvin functions [Chapter 551
52:11:2 J .(x t ix) _ (- V2) i V l))
52:12 GENERALIZATIONS
The generalization of Bessel coefficients to arbitrary order is the subject of the next chapter.
52:13 THE BESSEL COEFFICIENTS J0(x) AND 11(x) 520
52:14:1 J.(Rt)dr =
02 - a -
Jo [[aJ(a)] + (a2 n2)[J.(a)12
3 = a
2a2
shows that the functions J.(j.;Ix), J.(j,.,x), ... are orthogonal (Section 21:14) on the interval 0 s x s I
with weight function x. For, if we abbreviate J.(j.;kx) _ *k(x), then
0 1 k
52:14:2
f t V (t) `l',(t)dt IJ.(j.:k)l'
2 /k
Hence, any function f(x) which exists in the 0 S x 1 range can be expanded in the orthogonal series
where
2 1I
52:14:4 Ca = tf(r) J.Ij.,r)dr
IJ.(J.:k)1" o
Note that the orthogonal series may be based on a y Bessel coefficient: in practice. J. is often the most suitable.
It is equally evident from 52:14:1 that
r, 0 1k
52:14:5 r Yt(t) *;(r)dr = (n l l IJ.(j.:.)I'
J
where 'Y,(x) abbreviates J, (j;,,x). Hence, the functions 41;(x). W;(x). 41;(x), ... are also orthogonal. In fact, one
may generalize the orthogonality of Bessel coefficients more widely, as discussed by Spiegel (page 144).
CHAPTER
53
THE BESSEL FUNCTION Jt,(x)
Some authors classify transcendental functions (those that cannot be expressed by a finite number of algebraic
operations) as either "elementary" or 'higher." The Bessel function is generally regarded as the most important
member of the latter category, the 'higher transcendental functions." This function was invented in the context of
planetary astronomy but finds application in a wide variety of practical problems.
53:1 NOTATION
The symbolism J,(x) is almost universal, v being termed the order and x the argument of the Bessel function. The
name cylinder function is sometimes applied to J,(x), but this name is also used to describe collectively the broad
class of functions that this Atlas treats in Chapters 49-57.
A change of argument to 2V often simplifies formulas, and Clifford's notation
x)
53:1:1 Cx) = x-"/=J,(2 V
is sometimes useful in this context.
53:2 BEHAVIOR
Except when v is an integer, the Bessel function J,(x) is complex for negative argument. Therefore, in this chapter
we deal exclusively with x a 0, and this is the only range depicted in the contour map, Figure 53-I.
The map suggests that an infinite number of bands of alternating positive and negative values cross the r, v
plane, and this is, in fact, the case. This means that the Bessel function, viewed either as a function of x or of v,
displays an infinite number of zeros, separated by local maxima and minima.
For noninteger v less than zero, J,(x) approaches ±x as x -. 0, while for positive v, J,(x) remains bounded.
For v z 0, J,(x) never exceeds unity.
At constant small v, J,(x) is oscillatory throughout the entire x > 0 domain, whereas if the order takes large
values of either sign there is a range of the argument. approximately 0 < x < v, during which J,(x) increases (if
v > 0) or decreases (if v < 0) in magnitude prior to breaking into oscillations. Once they are established, the
oscillations steadily attenuate, as x increases, remaining centered about zero and eventually conforming to the
asymptotic expression
$21
53:2 THE BESSEL FUNCTION ],(x) 522
523 THE BESSEL FUNCTION J,(x) 53:3
m
53:2:1 J,(x) -- 2
arx
cosx -
\ 2
-4 x-+ m
If the argument x takes some constant value greater than zero, Figure 53-1 also illustrates how J,(x) varies with
v. As v takes increasingly positive values, the oscillations of 1,(x) sooner or later cease and the Bessel function
becomes a monotonically decreasing positive function that eventually obeys the asymptotic expression
I rexl
53:2:2 1,(x) - V-. = x = constant > 0
2av 2v/II
On the other hand, as v acquires ever-more-negative values, 1,(x) remains oscillatory, and the oscillations become
increasingly amplified and eventually satisfy
V/2 /-ex'"
53:2:3 J,(x) sin(vir) v-- -x x = constant > 0
avl\2v
An advantage of Clifford's notation [equation 53:1:1 ] is that whereas J(x) is generally defined as a real function
only for x a 0, C,(x) is real for all values of its argument. In fact, for negative x. C,(x) becomes related to the
hyperbolic Bessel function of Chapter 50:
53:2:4 C,(x) _ (-x)-"121.(2\) x !5 0
53:3 DEFINITIONS
The function (x/2)" is a generating function for the Bessel functions J,(x), J,,,(x), Thus:
(x/2)"
53:3:1
r(1 + v) k_o k'
Bessel functions are defined by a number of definite integrals, including
and
53:3:4 J.(x) -
2(2/x)"
df1 _
d d'-f df
53:3:5
xdx
xdx -x-dx=+xdx--x')f (v'
is f = c,J,(x) + cJ_,(x) if v is not an integer, c, and c_ being arbitrary constants. Equation 53:3:5 is known as
Bessel's equation but Bessel functions also satisfy the simpler Bessel-Clifford equation
djax.
+ (I + v)
dj
53:3:6 x +j= 0 j= x-`2[c,J..(2N6) +
dx
as well as many other differential equations [see Murphy]. See Section 52:3 for the corresponding solutions when
v is an integer.
The differintegration formula
2 d"'
53.3:7 J,(VA) _ sin(Vx)
53:4 THE BESSEL FUNCTION J,(x) su
is valid for all v and illustrates the close relationship of the Bessel function to the sine. Differintegration can also
be employed [see 53:10:10] to convert a Bcsscl function of any order to a Bessel function of any other order.
Hence, a Bessel function of arbitrary order may be regarded as derived from the Jo Bessel coefficient [Chapter 52]
by the operation
d-'
53:3:8 J,(2Vx) =x '12 -Ju(2\Ix)
dx'
or, in Clifford's notation [see 53:1:11:
d
53:3:9 C.(x) = x- - CO(x)
dx'
j,(x)=
n=0,±1,±2,n
5342
53:5 INTRARELATIONSHIPS
Bessel functions satisfy the recursion relationship
2v
53:5:1 J, 1(X) _ -J,(x) - J,_J(x)
X
53:6 EXPANSIONS
The product of two Bessel functions of common argument is expressible as a hypergeometric series with two
numeratorial and four denominatorial parameters:
/1+v+µ
(x/2),+"
= I
2
1+(-x')'
L+±)
2,
53:6:3 J x 44)
°O r(1+v)r(l+u.)j(1),(l+v),(I+µ),(t+v+µ);
From this it follows by setting µ = tv that 12.(x) or J,(x) J_,(x) may each be expressed as a K = I, L = 3 (see
Section I8:14) hypergeometric series.
The Bessel function may be expanded as the infinite product
53:6:4 J,(x) =
r(1 + v) `
l --
I--11-J(1..'=
x'`
r:l/ \\\
1-
x- \f\l
j;J i ill
(x/2)`
r(I + v),=
1I-
l- x2
Z.
where j,;t denotes the k" zero of J, [see Section 53:71. A similar infinite product holds for the derivative J,(x) _
dJ,(x)/dx
J,_,(x) - J,.,(x) (x/2)"'' x
53:6:5 2 = 1;.(x) =
2r(v)
nI- v>0
Although they are of greatest utility when the order is an integer, equations 52:6:6-52:6:9 apply to Bessel
functions generally, on replacement of n by v. Also, see 50:5:3 for an expansion of 1,(x) in terms of hyperbolic
Bessel functions.
When x, but not lvi, is large, the asymptotic expansion
2 ( v tr al , .
53:6:6 J,(x) = I P(v;x) cost x - 2 - - Q(v;x) sin1 z -
ax 4 2 - 4)]
holds, where, as x -. x:
53:6:7 P(v;x)--I-
(i - v2)(i - v2)
2!(2x)2
+
(19
- v2)(i - v2)(2 4 - V2)(1
412x)'
4 - v2)
- +
(2- V)2,(1 + v)2i
(2j)!(-4x2)'
+
and
(, - v)
4
+
(4 - V2)(2
4 - v2)(i - v2)
-
(iV - y2)(Y - y2)( - V2)(s - V2)(1 - v2)
+
53:6:8 Q(v;x)
1!(2x) 3!(2x)' 5!(2x)'
_ (_ - 1):r16 + y)2j_, ...
2(2j + l)!x(-4x2)'
Each of the functions P(v;x) and Q(v;x) is expressible as a K = 4, L = 2 hypergeometric function, as tabulated
in Section 18:14. Although they are generally valid only for sufficiently large x, the series 53:6:7 and 53:6:8
terminate when v is an odd multiple of ± 12, and in this circumstance an exact expression for J,.(x) is provided for
all x.
x=0 x=x
J.(x) v>0 0 0
J,(x) I 0
J,Jx) -I <v<0,-3<v<-2,-5<v<-4.,.. 0
l_.(x) n = 1. 2, 3,... 0 0
J.(x) -2<v<-1,-4<v<-3,-6<v<-5. ... -= 0
The values of x that cause J,(x) to acquire the value zero are denoted by j,,,, j,.j, j,..3, ..., and the name k"'
zero of the Bessel function J,(x) is given to j,x. This name is appropriate for is > -I but may be misleading for
is < -1 because not all these zeros then exist (i.e., as real numbers). As an example, for -4 < is < -3 there
exists the zeros j,:,, js, j,b, ... but there is no real J,,.,, j,, or j,3. For nonnegative integer is. values of j,.,, are
discussed in Section 52:7, and an algorithm is presented there for their calculation. For order - I we have j-..! _
j,:,. J-5! = ju, etc., and, generally:
53:7:1 J-.A =J,- k=n+l,n+2,n+3,...
For orders ±-! we have the simple rules j,1u = ka and j_112., _ (k - '/2)a. The zeros of order ; are given by
j3,2s = q(l) where the latter is discussed in Section 34:7. The same section discusses, and provides an algorithm
for. p,(b), which is related to the zeros of order by j_3na. = p,_,(-1). For small is of either sign j,,, is given
by the approximation formula
2 1
53:7:2 smalljvl
2
as is evident from Figure 53-2, in which the red lines display values of is as a function of is for I S k
s 9, permitting approximate values of the zeros to be found easily. Fur large positive orders, the first zero of J,(x)
is given approximately by
527 THE BESSEL FUNCTION J,(x) 53:8
holds except for k = 1. Figure 53-2 contains graphs of (2/n) j;;, - v versus v for the first nine extrema. Green
lines on this graph represent maxima, blue lines minima. The approximation
53:7:6 j.;, - v + 0.8086v''3 + 0.072v-'13
is valid for large v.
where to = I and t, = -x2:R_,/4k(k + v), which is a truncated version of expansion 53:6:1. Provided K > -v,
series 53:8:1 eventually alternates and the properties of alternating series [see Section 0:61 then guarantee that the
fractional error in approximation 53:8:1 is less than e/2 if ItK(2/x)" f(1 + v) J,(x)I < e. The algorithm increments
K until it exceeds -v and until tK/(to + t, + + tx_, + 2! 1 ',)1 < 10', thus ensuring 24-bit precision in J,(x)
except in the immediate vicinity of a zero of the Bessel function. The multiplier (x/2)`/t(1 + v) is calculated by
a modified version of the algorithm in Section 43:8, but special measures are adopted if v is a negative integer or
zero.
............ .... _y 17
15
14
13
............. :...
. 2Jy e ;y........... . 11
528
529 THE BESSEL FUNCTION J"(x) 53:10
Set h = I 1+ 7v (3v= -1
l] /30v3
v[l/+
For arguments greater than 7.5 + 0.3jvI, equations 53:6:6-53:6:8 are used in the form
K
/( Vir
53:8:2 J"(x)--EticoslkZ -x+ +4/}
2
where to = V2/ar and other tQ values obey the recurrence tk = [(k - 1)' - v2]t4_,/2kx. The number of terms
included in this asymptotic series is determined by three criteria. First, to ensure that the terms in the P(v:x) and
Q(v;x) series alternate in sign, we include terms in 53:8:2 up to at least k = Ivi - I. Second, we cease adding
terms to the cumulative sum f when the magnitude of tt becomes less than (5 x l0 -")Ifs. Third, irrespective of
size of the contribution of t, to f, we never allow k to reach the critical value x - ? + VT x + v2 after which
t4 increases in magnitude and the series diverges.
53:9 APPROXIMATIONS
The first three terms of expansion 53:6:1 are reproduced exactly by the approximation
1.
4 X X3
53:9:1 J"(x) small .r
xT(1 + v) 12 8(1 + v)(2 + v)]
and subsequent terms are close when v is large. An asymptotic approximation, valid for large x, is given by 53:2:1.
The expressions
and
r (VI +b'-b)"
53:10:6 exp(-bt) J.(t)dt = v>-I
J0 I + b'
cos(v arcsin(b))/ 1- O r b< 1
cos(bt)1.(t)dt = -sin(rv/2)
53:10:7
1 b1(b+ b'-Ib>l
Isin(v arcsin(b))/ 1 b' 0<b<I
53:10:8 sin(bt) J"(t)dt = cos(rv/2) v > -2
bI[b+ b'-II" b> I
and many others are listed by Gradshteyn and Ryzhik (Sections 6.5-6.7). Definite integrals of the form
are known as Hankel transforms and tables off V st f(t)1,.(sr)d1 are given by ErdElyi, Magnus. Oberhettinger and
Tricomi [Tables of Integral Transforms, Volume 2. Chapter 8).
The operations of differintegration are particularly simple when Clifford's notation [see 53:1:1) is adopted.
Thus:
d"
53:10:10 - x"C,J.t) = x'""C._"(x)
dxA
The Bessel function adopts complex values when its argument is complex and, unless the order is an integer, even
when the argument is negative. We report the expressions
53:11:1 J.(-x) = (-I)"J,(x) = [cos(vir) + i sin(vr)) J.(x)
and
for negative and imaginary argument, but exclude the case of general complex argument. I is the hyperbolic Bessel
function of Chapter 50. With n replaced by v, equation 52:11:2 applies for the special case of complex arguments
in which the real and imaginary parts have equal magnitudes.
53:12 GENERALIZATIONS
As a hypergeometric function [Section 18:14], the Bessel function may be written
(x/2)" (-x2/4)'
53:12:1 J,(x) = G
r(1) r(1 +v)_omill +v);
so that K = 0, L = 2 and one of the denominatorial parameters is constrained to be unity. Removal of this constraint
generates a function
(x/2)" (-x2/4)'
53:12:2
r(I +µ)r(1+v-p.)j_o (1+µ);(1 +v-p.);
that thus represents a generalization of the Bessel function. The µ = 12 instance of this generalized function is the
Struve function h,-,(x) of Chapter 57.
Here v is arbitrary except that it may not be a negative integer. The coefficient b; is given by
4'r(J+j+v)
53:14:3 bt=Y a.; 2J=k=0,2,4,...
J.o (J - j)!
when k is even, and by
53:14:4 b,=2Z
' 4'r(J+j+I+v) a.., 2J + I = k = 1,3.5....
;_o (J - j)!
for odd k. Equations 52:5:3-52:5:7 are all examples of Neumann series.
Replacing f(x) in 53:14:2 by the unity function leads to the expression
r(k + v)
53:14:5 2' (2k + v) k!
Js.,(.x) v * - I. -2. -3, .
A_o
as an expansion of an arbitrary power. Equations 52:5:6. 52:5:8 and 52:5:9 are special cases.
53:14 THE BESSEL FUNCTION J,.(x) 532
As an alternative to the 53:14:2 expansion. the function f(x) may be expanded as the modified Neumann series
(k-j)!
a,
Equations 53:5:2 and 53:5:4 are examples of modified Neumann series. Replacing f(x) in 53:14:6 by unity leads
to expression 53:3:1.
CHAPTER
54
THE NEUMANN FUNCTION Y (x)
In linear combination with a Bessel function [Chapter 531, the Neumann function satisfies Bessel's equation, and,
accordingly, it appears in the solution to many physical problems, especially those involving cylindrical symmetry.
54:1 NOTATION
The symbol N,(.r) is a common alternative to Y,(.r) to denote the Neumann function of order v and argument x.
The names Bessel function of the second kind and Weber's function are also in use.
54:2 BEHAVIOR
Because the Neumann function is complex for negative argument. we restrict attention to x >: 0. Behavior in this
range can be appreciated best by consideration of the contour map, Figure 54-1.
Viewed as a function either of x, or of v. the Y,(x) function displays an infinite number of oscillations. For a
constant value of v not very different from zero, the oscillations commence close to x = 0, and their amplitudes
steadily diminish as the argument increases, eventually conforming to the asymptotic expression
0.2
Yy(x) y - 2
54:2:3 V -0 JC x = constant > 0
VI WV ex
Comparison of Figures 53-I and 54-1 reveals considerable similarities. The major point of distinction occurs
for positive v as x- 0. Whereas the Bessel function approaches zero in this limit, the Neumann function approaches
--. Another difference concerns the phase of the oscillations, which generally differ by a/2 or thereabouts. This
is evident on comparison of equation 53:2:1 with 54:2:1 and of equation 53:2:3 with 54:2:2.
54:3 DEFINITIONS
For noninteger order, the Neumann function is defined in terms of Bessel functions
cos(VIT) J,(x) - J_,.(x)
54:3:1 Y,(x)=
sin(vrr)
vs0,±I,±2....
For integer order, this definition is to be replaced by its limit
54:3:2 Y,(x) = lim (cot(vir) J,(x) - csc(vrr) J_,(x)} n = 0, ±1, ±2, ...
The combination c,J,(x) + provides a solution to Bessel's equation 53:3:5, for integer or noninteger
V. Similarly, c,J,(2Vx) + c,_Y,(2Vx) satisfies the Bessel-Clifford equation 53:3:6 for all v. The coefficients c, and
c2 are arbitrary constants.
Some integral representations of the Neumann function are given by Gradshteyn and Rhyzik (Section 8.4151.
The close relationship of the Neumann function of zero order to the cosine function is evidenced by the semiintc-
gration formula
54:4 THE NEUMANN FUNCTION Y,(x) S36
/
Y'V x.
d-'.1'
1 1 ar 1
54:3:3 a; i s cost X
54:5 INTRARELATIONSHIPS
The order-reflection formula
54:5:1 n=0.±1,±2....
applies only for integer order, but the recurrence formula
2v
54:5:2 Y- ,W -Y,(x) - Y,.-,(x)
x
applies for all v. Because the form of 54:5:2 is identical to that of equation 52:5:15, expressions analogous to
52:5:16-52:5:20 relate Y,(x), Y3(x), Y4(x), Y5(x) and Y6(x) to Yo(x) and Y,(x).
A number of relationships interconnect Neumann and Bessel functions:
m
54:5:3 Y,(x) + Y-,(x) = cot 2 fJ,.(x) - J_,(x)1
2
54:5:6 J,,.1(x) Y,(x) - J,(x) Y,,,,(x) _ -
ax
54:6 EXPANSIONS
Using definition 54:3:1 and expansion 53:6:1, Y,(x) may be expressed as the sum of two infinite series. Alterna-
tively, by exploiting the properties of the gamma function [Section 43:51 and of the trigonometric functions, the
two infinite series may be conjoined into a polynomial function and a single infinite series:
Bs(-Ivl)
+'=O j!(1 +
Cs(Ivl-n)(x/2)"
(n + J)!(1 + it - jvj)) ( 4/
xz V
l v#0,±1,±2,...
where we use the abbreviation s(p) _ -(2/x)p r(p)/n. When v is positive, the coefficients A and C are unity and
B is cos(va); when v < 0, A = C = cos(vn) and B = 1. Here n may take any positive integer value, but the
expansion is most useful when n is the integer closest to jvi, for then the B and C terms in 54:6:1 have similar
magnitudes, permitting the summation to be evaluated with little loss of precision.
Both of the expansions discussed above become indeterminate when the order of the Neumann function is an
integer. For nonnegative integer order
54:6:2 Y,(x) _
2 x
J (x) -
12 ; (n - j - 1)!
IT
xz i
x)= i-v J! 4
54:6:3
7r
2
Y0(X) -J
2Y ° x+-+
In(x)+
O 4
3x'
128
11x6
13824 (
1
2
I
3
1(-xz/4)'
!/
r +
where y is Eulei s constant, 0.5772156649. Neumann functions of negative integer order can be expanded as in
54:6:2 by taking advantage of relationship 54:5:1.
Neumann functions of integer order are also expansible as an infinite series of Bessel coefficients [Chapter
52]. The general case is
(x/2)j 2
54:6:4 J,(x) + I ln(-) J .(X)
it x ,_0(n - j)j! IT 2
2,,(-1)'(n+ 2j)
ir ,=I I(n + J)
and the n = 0 case reduces to
A =0 X = m
Y,tc) v> - 0
0 0
Y,(x) -<v<
-3 -1 -7
2, .
<v< -5 -II <v < -9
Z,
2....
2
+. 0
Y,(x) -<<-,-<v<-,-<v<-....
2
-3 -9
3
2 2
-7 -13
2 2
-11
2
- 0
54:8 THE NEUMANN FUNCTION Y,(x) 538
The values y,..,, y,2, y,;3.... of the argument x that cause Y,(x) to acquire the value zero are known as the
zeros of the Neumann function, whereas the values y:.,,, y,2, y;,;,.... of the argument that cause Y,(x) to exhibit
a local maximum or minimum are called extrema of the Neumann function. Qualitatively, y,., and y..A are very
similar to j,,, and j;.j so that Section 53:7 contains useful information on these zeros and extrema. Numerically,
the relationships
54:7:1 v`j:.,<y..i<Y..1 <j <J:,_<Y,2<Y:.2<J..2<j 3<...
v
54:7:3 y;,,=j,k= - +k --4)it
1
largek
For large positive orders, the first zero of the Neumann function is approximated by
54:7:4 y.;, - v + 0.9316v"' + 0.260v-"' large v
while the first extremum, a maximum, obeys the approximation
54:7:5 y;., = v + 1.821 lv"' + 0.940v-"large v
Storage needed: v, f, h, A, B, C, t, p, n, q,
Set f = -109° s,jandJ
Set h = x/2
Iffracly+21 0 go to (16)
Ifh+0goto(1)
Ifv>0goto(19)
Replace f by f cos [180 Int(2 - v l
Go to (19) \ ////
(1) If h 4 go to (2)
If h > 0.35(8 + Jul) go to (17) Input restrictions: Order v may take any value
(2) If frac(v) = 0 go to (11) but argument x may not be negative.
Set B = cos(I80v)
SetA=C=f=t=l
If v > 0 go to (3)
Set A = C = B
Set B = 1 Use degree mode or change 180 to it and 90
Replace v by -v to a/2.
(3) Set p = v
Set n=Int (2 + V)
(4) Set q = h-°
(5) Replace q by q/p
Replace p by p + I
Ifp<3goto(5)
1l
Sets= [I+7p=\3p=-IIJ/30p2
L I-s
Replaces by +p(In(p)-I]
Zp
Replace s by -q exp(s) 2/ap
Iff<0goto(6)
Iff=0goto(7)
Replace A by As
Set f = -1
Setp-v
Go to (4)
(6) Replace B by Bs
Setf=j=0
Setp=v - n
Go to (4)
(7) Replace C by Cs
If n = 0 go to (9)
(8) Replace j by j + I
Replace f by f + r
Replace C by Ch/j
Replace t by -the/j(j - v)
Ifj+ngoto(8)
(9) Replace A by Af
Set j = f = 0
(10) Replace f by f + (B + C)
54:8 THE NEUMANN FUNCTION Y,(x)
Replace j by j + I
Replace B by -Bh2/j(j + v)
Replace C by -Ch2/(j + n)(j + n - v)
If j < 4 + 4h go to (10)
B+C
Replace f by f + A +
2
Go to (19)
(II)Setfq=1
Ifv>0goto(12)
Set f = cos(I 80v)
Replace v by -v
(12) Set A = 2 ln(I.78107242h)
Set B = 1 /2
Set C = 0
Ifv=0 go to (14)
Replace A by A - 1
Replace B by Bh
Set C = 1/(2h)
If v = I go to (14)
Set j 1
SetI=C+ C'1-AB
Sets= 901E+4h1
\ a
Set t = 1/Vh
Set p = q = j = 0
(18) Replace p by p + t cos(90j)
Replace q by q - t sin(90j)
Replace j by j + l Test values:
Replace A by A + I Y,13(4.93837580) = -0.163661435
Replace B by B + I Y6(0) = -x
Replace r by tAB/4hj Y3(1) = -5.82151761
If AB<0goto(18) Y-,(0) = -
Set f = p sin(s) + q cos(s) Yso(I) _ -2.19114281 x 10"
If If]> 2 x 10'11] go to (19) Y_1(5) _ -0.147863143
If j < J go to (18) Y-9/t(0) = 0
(19) Output f Y512(2.6) = 0.530168510
f - Y,(x) Y,(50) = 0.0959120278
The third routine is used by the algorithm when x s 8 or when x :5 5.6 + 0.71v], except when v is an odd
multiple of ± Z. It employs expansion 54:6:1 with the infinite series terminated empirically at j = lnt(4 + 2x).
The required gamma functions are calculated by a procedure that is essentially that utilized in Section 43:8.
54:9 APPROXIMATIONS
For a range of integer orders and moderate arguments, the polynomial term in expansion 54:6:2 is dominant. This
leads to the approximation
54:12 GENERALIZATIONS
In light of the limiting operation
-sr
54:12:1 lim {X"Q1j, (cos(kx))} = - Y_,(x) v> 0 x> 0
the Neumann function can be regarded as a special limiting case of the second Legendre function [Chapter 59].
Of the four Kelvin functions, ber,(x), bei,(x), ker,(x) and kei,(x), those with zero order, v = 0, are most important
and are emphasized in this chapter.
55:1 NOTATION
These functions are named after Lord Kelvin, the title adopted by William Thomson (1824-1907) on his elevation
to the peerage. Accordingly, the name Thomson function is given to these functions by those with distaste for the
aristocracy.
The initial letter, b or k (after Bessel and Kelvin), is sometimes capitalized. The terminal letter, r or i, refers
to "real" or "imaginary," as in definitions 55:3:3-55:3:6. The absence of a subscript, as in ber(x) and kei(x),
indicates an order of zero:
55:1:1 ker(x) = kero(x), bei(x) = beio(x). etc.
55:2 BEHAVIOR
For unrestricted values of v and x, the functions ber,(x). bei,(x), ker,(x) and kei,(x) display rather complicated
behaviors, and a comprehensive verbal description wil not be attempted. Moreover, we shall restrict attention to
arguments x 2: 0, although her, and bei, are real for negative arguments if v is an integer.
The four Kelvin functions are oscillatory functions of x when v is in the neighborhood of zero, but the onset
of oscillations becomes increasingly delayed as'vj increases. Once established, the periods of the oscillations ap-
proach \ir for all four functions, but while the amplitudes increase rapidly for ber,(x) and bei,(x), those of ker,(x)
and kei,(x) decline as x -, -.
The magnitudes acquired by ber(x), bei(x), ker(x) and kci(x) arc so diverse that it is impractical to plot the
four zero-order Kelvin functions on the same map, except for arguments in the rather narrow range displayed in
Figure 55-I. For larger arguments we make use of the asymptotic properties of the Kelvin functions by mapping,
in Figure 55-2, the curves labeled A = 2ax exp(-x/V) ber(x), B = V 2ax exp(-x/V) bei(x), C =
2x/a exp(x/'V-2) ker(x) and D = 2x7 exp(x/\/2) kei(x). As is evident from this diagram, or from equations
55:9:15-55:9:18, these four products approach pure sinusoids (Chapter 32) at large arguments.
543
55:3 THE KELVIN FUNCTIONS 544
1. 4
FIG 55-1
0. 8
0.8
.:. 0.4
0.2
55:3 DEFINITIONS
The general her, and bei, functions are defined in terms of Bessel functions of comlex argument, as
55:3:1
2 -x\ -
l
ber(x) = J, ix l +
2 -x\ +V/
1
J,
ix l
55:3:2
V2- N2)
real x.
t
Although it is not immediately apparent, each of hese formulas defines that is wholly real for nonnegative
If we use the symbols Re{ } and Im{ } to signify "the real part or and "the imaginary part of," so that x =
Re{x + iy} and y = Im{x + iiy},/we may define
([cos(!!) /
55:3:3 ber,(x) = Re{ + i sinl 2 IJ I,1 +
\ /
f /\f >
fr \(!!)1
\ I/
+ i sinII,I x + V2
\ I55:3:4
oos(2
bei,(x) m{ Jl
l \
+
7/ = Im{ I
l 111 \\\72
I}
in terms of the Bessel function [Chapter 531 or hyperbolic Bessel function (Chapter 50] of argument whose real
and imaginary parts are of equal magnitude. Similar formulas involving the Basset function [Chapter 51] define
548 THE KELVIN FUNCTIONS 55:4
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
1.2
FIG 55-2 :
_8
55:5 THE KELVIN FUNCTIONS 546
X
55:4:1 ker112(x) exp ( sin(72
2x \
55:4:2 kei112(x)
it
exp(
`V2 \ J 81l
cos(
1 x x- a 1
berln(x) = exp cos + +
55:4:3
taxx 8 - kei1,2(x)
55:4:4 bei1/2(x) =
7(--)
t (I
ax
exp "/
V2
x
sin
V2
x
+
a
8 )
+
1
- ker,,2(x)
The Kelvin functions ber, and bei, become somewhat simpler when v is a multiple of 3. Only for these orders
is ber,(x) or bei,(x) expressible as a simple power series. When the order is an even multiple of 3 (i.e., a multiple
of ;), the power series contains terms of alternating signs + - + - + - ; when the order is an odd multiple
of 3, signs occur in the sequence + + - - + + - - . . Equations 55:6:1-55:6:4 provide details.
55:5 INTRARELATIONSHIPS
2
55:5:1 ber_,(x) = cos(-) ber,(x) + sin(va) bei,(x) + - sin(vt) ker,(x)
it
2
55:5:2 bei_,r(x) = -sin(va) ber,(x) + cos(va) bei,(x) + - sin(va) kei,(x)
IT
all reduce to
55:5:5 f_"(x) _ (-1)" f;,(x) f = ber, bei, ker or kei n = 0, ± 1. ±2,
when the order is an integer.
Negative arguments generally render the Kelvin functions complex. Exceptions are ber, and bei, for integer
order, to which the argument-reflection formula
55:5:6 f,(-x)_(-1)"f(x) f=berorbei n=0,±1,±2,.
applies.
The recurrence relationships may be written
-v V [
55:5: 7 f f fe i f I fer = b er k er
-0/2
55 : 5 : 8 fe i ,,1(x) _ ( fe i ,(x) + fer,(x)1 - fe i ,_1(x) fe i = be i or ke i
x I
547 THE KELVIN FUNCTIONS 55:6
55:6 EXPANSIONS
All four Kelvin functions are expansible in terms of two series for which we shall use the abbreviations
(-x`/16)' Vjr(x/4)' (-x'/256)'
55:6:1 Fe,(x) _(\2/l _o (2j)!f(I + 2j + v) f\((2+2f(1+2)
v ;_o v
1 1 v
(1) 2,(1);(2+2)'11+2)'
are valid for all v, but a restriction to noninteger v is required to validate the expansions
55:6:5 ker,(x) =
n csc(vtr) r
2
/3va\
Lcosl 4 I Fe-.(x) + sin( 4) Ge_,(x) - cost/vn\
/3vir /va
4 I Fe,(x) -sin( 4) Ge,(x)
v#0. +1,±2....
ac
55:6:6 kei,(x) = sin - Fe_.(x) + cos - Ge_,(x) + sin - Fe,(x) -Ge,(.r)
cos__)
2(m) I 134 / \34 4) 4 J
v*0.±1.--2....
For nonnegative integer order, ker,(x) and kei,(x) are given by the expansions
+- (2)' =
2 =o
k4(1 + j) + y(I + It + j)
j!(n + j)!
cos (-+-j7r\I\fx2
(3na
4 2 4
n=0.1,2,...
2 n I (n 3nar a x'
55:6:8 kei(x)=In )bei.(.r)-4ber(x)-2IXIn1 J! sin(
4 +J2)(-)
Irrr 4,0+j)+y(I+n+ J') 3n7r n x'' n=0.1.2,--
+2 \2 )^ -o j! (n+ j)! ll 4)\4)
J
55:6 THE KELVIN FUNCTIONS 548
55:6:9 ber,(x)
exp(x/ V [)
tax ,=o
(2 - v),Q
j!(2x)'
+ v)' cos
( l x +'7 -1n -
2 4 8J
x-. x
55:6:10 bei,(x) -
exp(x / V 1)
2Trx ,moo
(i - v), ( + v),
j!(2.)1
sin
\\
rx +
va
2
X_%
IT x) (1-v),(!+v), /x m + 4 + -1
a
55:6:12 kei,(x) - -
- exp( sit) + X _X
with suitably chosen upper limits to the summation index j. Although the series in 55:6:9 does, indeed, asymp-
totically represent ber,(x) for large x, it more closely represents ber,(x) + [sin(2va) ker,(x) + cos(2v,r) kei,(x)]/
I. Similarly, a better approximation to the series in 55:6:10 is bei,(x) - [cos(2var) ker,(x) - sin(2vir) kei,(x)]/a.
When v = 0, the various expansions simplify to
a / 16Y
55:6:13
64 147456 [(2j)!]2
_
s° x1o
+ x'(-z'/16)' + ...
55:6:14 bei (x) = x2 +
4 2304 14745600 4((2j + 1)!]2
I
55:6:15 ker(x) = [ln() - yJ ber(x) +'r bei(x) + 1 1+ -+ + + 1 (-x`/16)'
x 4 -, ` 2 3 2j1 1(2j)! I'
I3 I \ (-x'/16)'
55:6:16 kei(x) = [In()
x
- beix) - 7 ber(x) + sI
4 4;=u ` 2 2j + I J [(2j + 1)!]2
\
55:6:17
ber
bei(.r)-
exp(x/N/2) "
G [(2j)!]2 cos
(_L_ - !?r4 --8 J = lnt(2x) x-. x
2wx io (j!)2(32x)'sin
As well, ber(x) and bei(x) are expansible in terms of the functions of Chapters 49 and 52:
7) J2,(x)
(_I))17J(
549 THE KELVIN FUNCTIONS 55:7
55:7:2 -ker,,(0) =
2
a
4
55:7:3 kei(0)
Figure 55-3 shows what value represents the limit of the four Kelvin functions of order v as x -. 0. The color
coding used in this diagram is that green represents zero, red represents +x, blue represents -x and black represents
one of the values given in equations 55:7:1-55:7:3. Notice the very elaborate behavior of ber,(0) and bei,(0) for
negative P. In this region these functions have a periodicity of 8 in their orders; that is:
55:7:4 f,(0) = G-00) = f,_16(0) = f = her or bei v < 0
As x approaches infinity, ber,(x) and bei,(x) oscillate with ever-increasing amplitude, but ker,(x) and kci,(x)
converge towards zero:
55:7:5 ker,(x) = kei,(x) = 0
'
All four Kelvin functions have an infinite number of zeros. For a sufficiently large x, these are given by
3
55:7:6 ber,(r) = 0 r = V 2rr(k - 2 - 8 k = large positive integer
v 18)
55:7:7 bei,(r) - 0 r - V to k - + k - large positive integer
2
5
55:7:8 ker,(r) = 0 r "- V k - 2 - g) k =large positive integer
The first few zeros for v = 0 are included in Table 55.7.1 together with argument and function values of early
extrema, as well as the function values at the x = 0 origin. The final line in Table 55.7.1 provides an approximation
formula for calculatintthe k'" zero of each Kelvin function; it uses the abbreviations a = N/2(8k - 3)sr. b =
N/2-(8k + 1)a, c = V2(8k - 5)ir and d = V (8k - Since the approximations are excellent, even for k as
small as 4. the table permits all the zeros of the zero-order Kelvin functions to be estimated.
1 ,4 _ci obi
.y=5
.. y.4
550
551 THE KELVIN FUNCTIONS 55:8
Table 55.7.1
0 1.000000 0 0 0 0 -0.7853982
2.848918 0 3.772674 2.346147 1.718543 0 3.914668 0
6.038711 -8.864036 5.026224 0 2.665845 -0.07102369 4.931811 0.01121607
7.238829 0 8.280989 -36.16540 6.127279 0 8.344225 0
10.51364 153.7818 9.455406 0 7.172120 0.001956681 9.404051 0.0003574233
11.67396 0 12.74215 670.1602 10.56294 0 12.78256 0
14.96845 -2968.681 13.89349 0 11.63219 -6.705969 x 10'' 13.85827 1.280427 x 10-'
16.11356 0 17.19343 -13305.52 15.00269 0 17.22314 0
.7.i
a
8
+-+-
I
aa'
0
b
8
+b- +b'-
I
0
C
8
+---
I
c
0 -+ ---
d
8
I
d
V '32
d2
0
-n)
1
Input x >>
Ifx<8goto(I)
Set r =90
(1\- 41 \
Sets=-r-45
Set t = l /Vs
Go to (4)
(1) SetB=r=1
Set s = 1099
Ifx=0goto(2)
Set s = 0.115931516 - ln(x)
(2) Replace j by j + I
Replace t by t(x/2j)2
Replace r by r + (I /j)
Replace B by B + t cos(90j)
Replace K by K + rt cos(90j) Use degree mode or alter 90 to ir/2, 45 to 7r/4 and
Replace b by b + t sin(90j) 135 to 37r/4.
Replace k by k + rt sin(90j)
If j < Zx + 4 go to (2)
Replace K by K + sB + irb/4
Replace k by k + sb - 7rB/4
Go to (5) Input restriction: x > 0
(3) Replace r by 8x 14j + j
Replace j by j
+\\I
+ l
1
Replace r by r - 45
Replace s by s + 135
(4) Replace B by B + t cos (r)
Replace b by b + t sin(r)
55:9 THE KELVIN FUNCTIONS 552
Replace K by K + r cos(s)
Replace k by k + t sin(s)
70
Ifj<3+- go to (3)
x
Set t = (1 //v) exp(x/ / 2)
Replace K by K/t
Replace k by kit
k
Replace B by Bt - -
it
K
Replace b by br + -
IT Test values:
(5) Output B ber(2) = 0.751734183
B - bcr(x) G«« bei(2) = 0.972291627
Output b ker(2) = -0.0416645140
b = bci(x) kei(2) = -0.202400068
Output K ber(10) = 138.840467
K - ker(x) bei(10) = 56.3704592
Output k ker(10) = 0.000129466329
k - kci(x) kei(10) = -0.000307524569
55:9 APPROXIMATIONS
For small arguments, the following approximations generally hold:
sin(3va/4)(lY
2
cos(3vrr/4) x x
55:9:1 ber,(x) _ v + -1, -2, -3, ...
r(1 + v) 2)Y r(2 + v) 2)
sin(3va/4) cos(3va/4) (x
55:9:2 bei,(x) = + v * -1, -2, -3, ...
r(1 + v) (2x), r(2 + v) 2)
r(v) cos2(3va/4) \x\` - r(v - 1) 2n(3vsr/4) (x)`-2
55:9:3 ker,(x) - v>2
I axe
55:9:4 ker,(x) = :t2
2 - 16
r(v) cos(3va/4) (x2) r(-v) cos(va/4x `
55:9:5 ker,(x) = (2) -2 < v < 2 v + 0, ±1
2 + 2
55:9:10 kei,(x) = x +
[1.(X2)
- y + 2] v s ±2
4
553 THE KELVIN FUNCTIONS 55:10
55:9:11 kei,(z) =
-f(v) sin(3vir/4) (X2)'
2 \- f(-v) sin(va/4) x
2 (2) -2 < v < 2 v * 0, ±1
and these may be supplemented by using equation 55:5:5 for cases of negative integer order. Generally, equations
55:9:1-55:9:14 give the first two terms in expansions of the Kelvin functions in ascending powers of the argument
x. For certain special values of v, however, some of these terms may vanish. The leading terms in equations 55:9:1.
55:9:3 and 55:9:5, for example, become zero when the order adopts one of the values 3, 2, IF, i, 6, ....
For large arguments, the approximations
x va
55:9:15 ber,(x) =
1
exp
x
coal + 2 - a8 large x
tax %J Vt
55:9:16 bei,(x) -
I /x / x+va
explV-2-)s nl -L 2 - ag
i large x
tax
va a
55:9:17 ker"(x) = exp (x cos(x + + large x
Zx 2 8 )
55:9:18 kei,(x) expl ) sinl + + large x
VVVV2x \V-2 `,r2 2 8
and
d 1
while the following definite integrals establish links with the functions of Chapters 32 and 38:
55:10:6
('
/
exp(-i/x) ker(2\ t)dt = 2z (l cos(x) Ci(x) + sin(x) r Si(x) - 2]r j
{
L 111
10 L 2
Other indefinite integrals are given by Abramowitz and Stegun [Section 9.9] and other definite integrals by Grad-
shteyn and Ryzhik (Section 6.87).
55:12 GENERALIZATIONS
If the complex variable z = x + iy is represented by the polar equivalent z = r exp(i(l), one sees from definitions
55:3:3 and 55:3:4 that ber,(r) and bei,(r) are the real and imaginary parts of J,.(z) with 0 = 3a/4. Thus, one may
generalize these Kelvin functions by allowing 0 to adopt an arbitrary angular value.
so Kelvin functions of the third kind are defined by a similar relationship but with the Hankel function [Section
54:13] replacing the Bessel function. Thus:
One has the identities her,(x) _ (2/a) kei,(x) and hei,(x) = (-2/n) ker,(x).
CHAPTER
56
THE AIRY FUNCTIONS Ai(x) AND Bi(x)
The Airy functions Ai(x) and Bi(x) are related to Bessel functions of order 3 and -3, with resealed arguments. The
two so-called auxiliary Airyfunctions are also important and are briefly discussed in this chapter.
56:1 NOTATION
In formulas involving the Airy functions, the auxiliary argument
56:1:1 X = 3 (IxI)`/'
is frequently more convenient than the usual argument x. Note that dX = tix I dx.
There appears to be no standard notation for the auxiliary Airy functions. This Atlas employs fai(x) and gai(x)
for the functions that Abramowitz and Stegun [Section 10.41 denote by f(x) and g(x).
56:2 BEHAVIOR
The functions Ai(x) and Bi(x) exist for all real arguments, but their behaviors depend crucially on the sign of x,
as illustrated in Figure 56-I.
For x z 0, both Airy functions are positive, but whereas Bi(x) increases rapidly as x --> x, Ai(x) steadily decays
towards zero. For negative arguments, Ai(x) and Bi(x) are oscillatory functions with oscillations whose frequencies
gradually increase and whose amplitudes gradually decrease as x -+ -x.
The auxiliary Airy functions are mapped in Figure 56-2. For positive arguments, both increase exponentially
as x -> x, while for negative arguments each function exhibits oscillations similar to those of the Airy functions
themselves.
56:3 DEFINITIONS
The Airy integral
iA i0
44 44
.I ,
44 44 44 44
i',
IF
.1
44
.1
it 44~
FIG 56-1 : : :
: .. .. .. .... ....:....:.. p
defines the Ai function for all arguments, while the similar definite integral sum
3
('
56:3:2 Bi(x) _ ! J expl xt - 3 )d, + if s sin(xt + 3 Idr
serves the same purpose for Bi.
Using the 56:1:1 definition of an auxiliary argument, the two Airy functions may be defined in terms of hy-
perbolic Bessel functions [Chapter 501, or the Basset function [Chapter 51], for positive arguments. The Bessel
functions [Chapter 531, or Neumann functions [Chapter 541, provide the corresponding definitions when the ar-
gument is negative:
1.
[I-it3(X) - 11113M) = K,n(X) x> 0
3 a 3
56:3:3 Ai(x) _
-x x
[1-1p(X) + 1.,3(x) = [Y-113(X) - Y.1 3(X)1 X < 0
3 3
A linear combination of Ai(x) and Bi(x) functions satisfies the Airy differential equation
SS7 THE AIRY FUNCTIONS Ai(x) AND Bi(x) 36:3
drf
56:3:5 xf f = c,Ai(x) + c_Bi(x)
dx' =
where c, and cr are arbitrary constants.
The auxiliary Airy functions are defined by
32/3r(23) [Bi(x)
I _,/3(X) x>0
56:3:6 fai(x) = 2 + Ai(x)
J_,n(X) x<0
31/3
X = 1301)",
3113r(i) [Bi(x)
x>0
2 - Ai(x) _
J,/3(X) x<0
3zp
where r denotes the gamma function [Chapter 431. In the notation of Section 43:14. these auxiliary functions may
56:4 THE AIRY FUNCTIONS Ai(x) AND Bi(x) 558
Syntheses similar to 56:3:8 and 56:3:9, but involving the Bessel J0 coefficient instead of its hyperbolic counterpart.
provide definitions of fai(x) and gai(x) for negative arguments.
56:5 INTRARELATIONSHIPS
The formulas
gai(x)
56:5:1 Ai(x) = Ai(0) fai(x) -
2irBi(0)
and
gai(x)
56:5:2 Bi(x) = Bi(0) fai(x) +
2iAi(0)
relate Airy functions to their auxiliary counterparts. See Section 56:7 for values of Ai(0) and Bi(0).
56:6 EXPANSIONS
The auxiliary Airy functions are expansible as the convergent series
x' x6 x9 (3j + lf!
56:6:1 fai(x)=1+-+(1x4)-+(1x4x7)-+
3! 6! 9! i=o (3j + I)!
x 3i
X4
x7 x10 (3j + 2)!!! 3i+
56:6:2 x
4! 7! 10! r0 (3j + 2)!
in which the triple factorial [see Section 2:13] occurs. Notice that fai and gai are two members of a trio of functions,
the third of which
x11
x2 x! x6
(3J)1tlxrj-2
56:6:3
2! 5! 8! 11! i_o (3j + 2)! i=o (3j + 3)!
is encountered in Section 56:13.
Expansions 56:6:1 and 56:6:2 may be used in conjunction with relationships 56:5:1 and 56:5:2 to generate
power series expansions of the Airy functions Ai and Bi. Terms involving x'', xs, xr, ... are absent from such
expansions.
Asymptotic expansions of the Airy functions take different forms according to whether the large argument is
positive or negative:
559 THE AIRY FUNCTIONS Ai(x) AND Bi(x) 56:8
exp(-X) ai a2 as
2a X X2 X' Xi
56:6:4 Ai(x) - costX+-
sinlX+
' 41 \ a,
nl
a2 aq
f a, as
a0 + X2X+-.. X--* -x
LX XsX + s
exp(X) a, a2 a, a,
Vx X X'' X' X'
56:6:5 Bi(x) a a
cosX + sin X +'--.L
x-s -x
-V - w ap X' X
a + a` + 7t V / [Xa, X
a3 + Xa s -
where the auxiliary argument X is given by 56:1:1, a0 = 1, and other coefficients obey the definition and recursion
(6j-1)!!
56:6:6 a,=
j!(23 - 1)!'(216)'
!1+ 36j 5
a-1
2
j=1,2.3,...
An asymptotic expansion, namely
1155 (6j - 1)!!
56:6:7 W
Ai2(x) + BOW ) -
1
1 + }5 + s
+ + + x -. -x
a 32x 204gx j!(96x')'
describes the sum of the squares of the Airy functions as the argument acquires very negative values.
x=-z x=0 x =.
T(i)
-
Ai(x) 0 = 0.3550280539 0
3''r(i) 2a3,n
r 31n1'd)
Bi(r) 0 0 6149266276
-
.
3u T(i) 2m
fai(s) 0
gai(x) 0 0 x
Table 56.7.1 records the first few zeros and extrema of Ai(x) and Bi(x), all of which occur at negative argu-
ments. The final tabular entries, which use the abbreviations a = [31r(4k - I )J21' and b = (31r(4k - 3)1"', provide
excellent approximations for the k" zeros of the Ai and Bi functions, for k > 5.
Table 56.7.1
Bi(x)
cos(X) a, a2 a3 a4
56:8:1
Ai(x)
so
± +
X3
Bi(x) 2aM X X2 X4
+
sin(X) f mapa + a2- - X3
- - X'
a3
-+ a4
2;%I--x X X2
90
If j < 3 + X go to (4)
If q < 0 go to (5)
Set B = of exp(X) Test values:
Set A = t/[f exp(X)) Ai(4) = 9.516074 x 10-'
Go to (6) Bi(4) = 83.8470714
(5) Set B = u cos(X) - t sin(X) Ai(-4) _ - 0.0702655329
Set A = t cos(X) + u sin(X) Bi(-4) = 0.392234706
(6) Output A Ai(6) = 9.94769437 x I0-6
A = Ai(x) Bi(6) = 6536.44608
Output B Ai(-6) = -0.3 29 1 45 1 74
B - Bi(x) Bi(-6) = -0.146698376
The universal hypergeometric algorithm [Section 18:14) permits values of fai(x) and gai(x) to be determined.
56:9 APPROXIMATIONS
For small arguments, the approximations
56:9:1 Ai(x) - 0.355 - 0.259x + 0.059x3 - 0.022x' small
exp(2x"'/3)
56:9:4 Bi(x) _
-
V. N77
large positive x
so that the product Ai(x) Bi(x) = I/2a\/x. For large negative argument
cos(X) + sin(K) 2(-x)112
56:9:5 Ai(x) = X= 3 large negative x
cos(X) - sin(X)
56:9:6 Bi(x) y X= 3 large negative x
;V --x
so that in this range Ai2(x) + Bi'(x) = 1/itV . a result that is also a consequence of expansion 56:6:7.
56:10:1
d
dx
Ai(x) =
-
x
[I-xn(X) - 1212(X)] _
-c K213(X) x>0
x
[1-2n(X) + 1213(X)] x>0
d V3
56:10:2
dx x
[J-,n(X) + x<0
J2/3(X)]
56:10:4 f
irrespective of sign of the argument. Second derivatives obey the simple relationship
d2
dx'-
= xf f = Ai(x) or Bi(x)
Indefinite integrals of the Airy functions are expressible in terms of their derivatives and that of the function
Hi(x), discussed in Section 56:13:
d
56:10:6 J Bi(t)dt = J Bi(t)di ='rr1 Bi(x) Hi(x) - Hi(x) d Bi(x)J
=
I L dx dx
These relationships are valid for either sign of the argument x. Three useful definite integrals result from setting x
= 0 in 56:10:5 and 56:10:6.
56:12 GENERALIZATIONS
Inasmuch as Airy functions are Bessel functions of order ±3, they may be generalized to other fractional orders.
56:13:1 Hi(x) =
1
exp ( xr
;)r6)
- 3 dt = 32
(j - 2)!!!
x1 =
2Bi(0)
[fai(.r) + gai(x) + hai(r)]
IT o 0 1 3
( / t'\ 27 \ 1
f sinl xr + 3 IdtBi(0)
= - [fai(x)
1
56:13:2 Gi(x) = + 2 I gai(x) - 2haix)J
it 2irBi2(0) -
Of course, Hi(x) + Gi(x) = Bi(x). / \
CHAPTER
57
THE STRUVE FUNCTION
The Struve function h,(x) has many analogies with the Bessel function of Chapter 53. Likewise, the hyperbolic
Struve function Q x), which is discussed briefly in Section 57:13. is analogous to the hyperbolic Bessel function
of Chapter 50.
57:1 NOTATION
The usual notation for the Struve function is H,(x) with the capital H printed in bold type to avoid confusion with
Hermite polynomials (Chapter 241. However, because they are difficult to reproduce by hand or on an office
machine, this Atlas avoids abnormal fonts. Thus, we adopt h,(x) for the Struve function and e,(x), instead of the
more usual boldface L,(x), for its hyperbolic counterpart.
We use h,(x) to denote the Struve function of argument .r and unrestricted order v, but h (x) is employed when
the order is constrained to be an integer.
57:2 BEHAVIOR
Like the Bessel function, the Struve function is real for negative argument only when its order is an integer.
Accordingly. the contour map Figure 57-1 is confined to x ? 0.
Also like the Bessel function, h,(x) is an oscillatory function of its argument, the amplitude of the oscillations
decreasing with increasing x. Unlike the Bessel function, however, the damped oscillations of the Struve function
do not generally occur symmetrically about x = 0. For v > , this asymmetry is so pronounced that h,(x) is uniformly
positive when its argument is positive.
57:3 DEFINITIONS
2(-r/2)"
57:3:1 h,(x) = (t - t1),- in
sin(.rr)dt v>- 2
V;-r(-V+ -21) 0 2
$63
57:3 THE STRUVE FUNCTION S64
545 THE STRUVE FUNCTION 57:5
or its equivalent on replacement of t by sin(O) or cos(6), defines the Struve function for orders exceeding - ;.
Another integral
provides a representation of the difference between the Struve and Neumann functions of common order and ar-
gument. This time tan(O) or sinh(u) may replace t, but the restriction to orders in excess of -; again applies.
Using the notation of Section 43:14. the Struve function may be synthesized by
57:3:3 J o(-r) -
-' -V - 1
0 0
-
4 F(12 + v)
2(.r/2)' " h"(.r) X
-x2
4
from the zero order Bessel coefficient. This, then, constitutes yet another way of defining h"(x).
57:4:3 h)/,(x) = -2
ax
I - cos(x)
x
- sin(x) +
x-
57:5 INTRARELATIONSHIPS
The recurrence relationship
(.T/2)" + 2v h.(x) - h"_,(x)
57:5:1 h" ,(X) _
X
Reflection formulas exist only for integer orders. For argument-reflection, we have
57:5:3 h"(-x) = -(-1)" h"(x) n = 0, t 1, t2, ...
so that the Struve function of even order is odd, and vice versa. Order-reflection introduces an n-term polynomial
function
57:6 EXPANSIONS
For all finite values of v, the expansion
fx (-x2/4Y _ 2(x/2)"" (-x2/4)'
r(i +;)r(j+v+
'
57:6:1 h,(x)=I-i
2 ,=a f r(23 + v) j-o ( ', (2 +v),
converges, provided x is real and positive; negative arguments are admissible if the order is an integer. The most
- ---
Tr 3 45 (2j - 1)!!(2j + I)!!
57:6:4 hi(x) =
2 x2
3
- 45 + X4 X6
1575
-
99225
x"
+ +
(-x2)'"
(2j + 1)!! (2j + 3)10
+
Integer order Struve functions are expansible in terms of Bessel coefficients (Chapter 52]; for example:
(x) _ 4 x) +
J=,(x)
+ -+
Js(x) 4 J,,, i(x)
57:6:5 (J )=
tr 3 5 rr ,_0 2j + 1
4 (1 - Jo(x) J,(-t) JI(x) J6(x) J'' (x)
57:6:6 he(x) _ + = + + + +
it 2 3 15 35 4j2 - I
An asymptotic expansion exists for the difference between Struve and Neumann functions of common argument
and order:
57:6:7 h,(x) -
(x/2)"-' (1 - '
1 - 2v
+
3!!(l - 2v)(3 - 2v) - +
(2j - 1)!! (2' - v),
2 +
V >r r( + v) x x (-x /2)) /
x- x
See Section 54:6 for an asymptotic expansion of Y,(x) that may be used in conjunction with the 57:6:7 series. The
important cases of orders -1. 0 and I are
57:6:8 h_dx)-.(x)
45 1575
I2 _X,+s3 _X6+XB
1
_...+ (2j + 1)!!(2j - 1)" +... r-,x
(-x2))-'
2(1 [(2j - I)"]2
- -+;;-
9 225 1
57:6:9 h0(4-Y6(x)--
rr x .e
1
xr x(-x2)l
J x-,x
567 THE STRUVE FUNCTION 57:8
and
.r° (-x2)
J x
x= -x x =0 x=x
h,(x) n - 3. 5, 7.... 0
h,(x) n = 2, 4. 6,... 0
h,(x) I <v*2,3,4,... undef 0
h,(x) 2/v 0 2/v
h.(r) I > v > - I: v 0 undef 0 0
h.(x) 0 0 0
h-.(x) 0 2/v 0
-3
h,(x) -1 > v > 2- undef x 0
-3 -5 -7
h.(x) 2 .... undef 0 0
2 2
-3 -5 -7
h.(x)
-
>v>-2,-2>v> -, - >v>-4. undef -x 0
h,(x) 0 ±x 0
-7 -9
h,(x) ->v>-3.-3>v--,->v>-5,..
-5
2 2 2
undef x 0
h,(x) 0 x 0
Apart from its x = 0 value, the Struve function has no zeros for v > ;. For v ~= 1, however, there is an infinite
set of zeros that, for large arguments, correspond to the roots of the equation
/ \
1=(I'(v)+J
r'2
v<-2
57:7:1 sinlr - 2 + h,(r)=0 larger
4 r - ir(2k + v + #)/2 where k is a large positive integer.
and therefore ultimately occur close to
The algorithm computes h,(x) as the sum to + ti + t_ + - - - + t; + - + p where to, equal to 2(vj/2)'+'/
I'(2 + v). is evaluated by a modified version of the algorithm of Section 43:8, t, is calculated as (-x2/4)tj_,/
(, + j)(_s + j + v) and J is given by an empirical function of x and v designed to ensure that the series has
converged adequately. Special measures are taken if v has one of the values -. -;, -;..... This algorithm will
also compute the hyperbolic Struve function f,(x), discussed in Section 57:13, when -x is input in place of x.
57:9 APPROXIMATIONS
Expansions 57:6:1-57:6:4 can readily provide approximations to the Struve function that are valid for small ar-
guments. For large arguments, one may similarly use the asymptotic expressions 57:6:7 and 54:6:6. For the cases
of orders equal to - 1. 0 or 1, theVWX leading asymptotic terms are
/
57:9:1 h_,(x)= sin(.,, l
32rx'
cosx+2 4/// Irx
largex
57:9:2 hu (x) _ -s
2 -
in x - - + - -
J 2 1
cos x - -- J
I arge x
ax 4 arx 32ax' ( 4
57:9:3 h0) = - +
2
sin x -3sr)
-+ 3
cos z - -
-) large x
'Ti ax 4 32nx' 4
57:10:1 - h,.(x) +
dr 2V. 2
for the derivative of the Struve function simplifies to dh.(x)/dx = h_ (x) for zero order. One also has
569 THE STRUVE FUNCTION 57:12
d
57:10:2 (x'h.(x)) = xh.-.(x)
dz
1
57:10:4
f t'+%(: t = -x"h. .(x) v<-
2
,
2x
57:10:5 tl-'h.(t)dt =
J0 2" ' + v)
but the corresponding definite integral
v µ
-2"r Z+ 2+2
µn
57:10:6
o
t"h,(t)dt=
Il
cot(-+-l
va
\2 2/
µ<-
1
2
-2<v+µ<0
((v
\2 2 2/
holds for a small range of µ values. Selected indefinite and definite integrals are
(
57:10:8 fo
1h0(t)di = x ? - h_i(x)) = xh1(x)
a
ho(t)
57:10:9 1
0 t
dt=-2
With a redefinition of the argument, equations 57:10:2 and 57:10:3 are the special µ = 1 and µ -1 cases
of the general differintegration formula
57:10:10
'/
- {x` 'h.(2Vx)} = x' "'12h._"(2Vx)
dx"
- v>--
l
2
We present here only the case of the Struve function with a \purely imaginary argument:
57:11:1
I
h.(iy) = i'-`e.( v) _ 1sinl 2 + i cosl Z )JE.(v)
57:12 GENERALIZATIONS
The hyperbolic Struve function is to the Struve function as the hyperbolic Bessel function I,(x) is to the Bessel
function J,(x). It may be defined by the definite integral
I
2(x/2)"
57:13:1 e"(x) = (1 - t=)"-'l: sinh(xt)dt v>-
V r(, + v) o 2
V it r(# + v) O
2
n = 1,2,3,...
for example, (_1(x) = e,(x) + (2/ir) and a-.(x) = fi(x) + (2x/31r) - (2/,rrx).
571 THE STRUVE FUNCTION 57:13
For orders of -!, -;, ..., we have the identity where the latter functions are discussed
in Sections 50:4 and 28:13.
The expansion of e,(x) mirrors that given for h,(x) in 57:6:1 except that the negative signs should be expunged.
Similarly, if the alternating signs on the right-hand side of 57:6:7 are replaced by uniformly positive signs, the
series is an asymptotic representation if I_,(x) - e,(x). Again, it is only by distinctions in sign that the recursion
formula
2v
57:13:4 e,+i(x) = e,_i(x) - e,(x) - (x/2)'
x V++r(v+i)
for the hyperbolic Struve function differs from 57:3:1 for h,-,(x).
Because the Struve function and the hyperbolic Struve function are generally complex for negative argument,
the algorithm in this chapter will accept only positive arguments. Accordingly, the input of a negative x into the
Section 57:8 algorithm is interpreted as an instruction to compute t,(x), instead of the h,(x) that is calculated when
the input x is a positive number.
The differentiation formula
e,-,(x) + f,.,(x) + (x/2)'
57:13:5
-
d
e,(x) = 2
2Vr(v+ )
produces d to(x)/dx = e_i(x) = ei(x) + (2/n) as a special case, and, apart from a sign, is analogous to the first
equation in Section 57:10. Similarly, allowing for changes of sign, all the other relationships in that section have
their analogs for the hyperbolic Struve function.
CHAPTER
58
THE INCOMPLETE BETA FUNCTION B(v;µ;x)
Many simpler functions are special cases of the incomplete beta function. For example, as described in Section
58:14, the indefinite integral of any trigonometric or hyperbolic function, raised to an arbitrary power, is expressible
as an incomplete beta function.
58:1 NOTATION
Alternative symbolisms for B(v;µ;x) are B,(v,µ) and the product 1,(v,µ) B(v,p) where B(v. s) is the complete beta
function discussed in Section 43:13. The 1,(v,µ) function, which has statistical applications [see Section 27:141, is
the incomplete beta function ratio, equal to the quotient
B(v;is,x) _ r(v + µ) B(v;µ;x)
58:1:1 I,(v.µ) =
B(v,µ) r(v) r(µ)
58:2 BEHAVIOR
Being trivariate, the incomplete beta function has a behavior sufficiently complicated that it cannot be adequately
described either in words or planar maps.
While allowing µ to adopt any real value, we shall require v to be positive. For otherwise unrestricted values
of the v and µ parameters, B(v;µ x) is defined only for arguments in the range 0 s x < 1. However, if v is a
positive integer, the formula
573
58:3 THE INCOMPLETE BETA FUNCTION B(v; s;x) 574
58:2:2 B(v;m;x+1)=(m-1)
(v),,,
-(-1),B(m;l-v-m;l+x x?0 m=1,2,3...
Here (v),.. is a Pochhammer polynomial [Chapter 181.
58:3 DEFINITIONS
t' dt
58:3:2 B(v:µ;x) = 0sT=X<
'
T
and
rr
58:3:4 B(v;A;x) = 2 tanh2 '(r) sech"(t)di 0 T = artanh(V x) <
J0
As well, the incomplete beta function may be defined as a number of definite integrals, the simplest being
By sufficient applications of the recurrence formula 58:5:2, this case can also provide expressions for B(2;µ;x),
B(3;µ;x). etc. Formula 58:4:1 becomes indeterminate if µ is zero, but, by considering the p. -* 0 limit, one con-
cludes that B(1;0;x) = -ln(1 - x).
When the µ parameter equals unity:
This formula may be used in conjunction with recursion 58:5:3 to produce the polynomial expression
/ 1- xf
1
58:4:4 B(v;0; x) + In,1 = x"$(x; I ;V)
,-0j v
where In, is the generalized logarithm discussed in Section 25:12 and denotes Lerch's function [see Section
64:12J. Some important instances of this result are
X.
x2 x' x'"
58:4:5 B(n + 1;0;x) = -In(1 - x) - x - - - - - - - = - n = 0, 1, 2,...
2 3 n J=. i +1
xs
r V.r_
i j+
3 5
u2
- X%
n=0,1,2,...
and
58:4:7 B(; -- 14;0;x) = 2 artanh(jx"j) + 2 arctan(Ix'''1)
Negative integer µ cases can be deduced from 58:4:4 by one or more applications of the recurrence (µ - 1) B(v;µ
- 1;.r) = (v - 1) B(v - 1;µ:x) - x'-'(1 - x)"-', which is a restatement of 58:5:2.
Whenever v and µ are both positive multiples of ;, the incomplete beta function reduces to an elementary
function. There are four general cases. When both v and µ are even multiples of i (i.e., both are integers), the
general formula is
\m1 ` (-x)i
58:4:8 B(n + lm + Ix) 1"(1 - t)dt =.r-' n, m = 0, 1, 2, ...
o to
I
n+j+I
When v is an odd but µ an even multiple of z, the general formula is
Vi
(-x)/
58:4:9 B(n + ;;m + Ix) = 2 t2'(1 - t)"dt = x""/2 n, m = 0, 1, 2, ...
0 ).o (-j) n+j+4
When v is an even but p. an odd multiple of s, the general formula is
' n\ 1-(1-x)'" '112
n,m=0,1,2....
58:4:10
fvi_. -0 j
I
m+j+2
The final case, when both v and µ are odd multiples of ;, is more complicated and invariably leads to an
arcsin(V) term. The general formula can be obtained by combining
('r /
58:4:11 B(n + l;m + Z;x) = 2 J sine"(t) cos2T(t)dt = 2 i (-1), I -j) sin'"'=1(t)dt n, m = 0. 1. 2...
o i-o \ IT
where T = aresin(\), with the indefinite integral 32:10:6. The simplest instance is B(# l;x) = 2 arcsin(V), and
other simple cases can be obtained from this result and recursion formulas 58:5:2 and 58:5:3.
When the two parameters differ only in sign, there is the simple result
58:5 INTRARELATIONSHIPS
The formula
r(µ)r(v)
58:5:1 B(A;v;x) = B(v;µ;I - x)
r(v + µ) -
58:6 THE INCOMPLETE BETA FUNCTION B(v;µ;x) 576
shows the effect of interchanging the parameters. As well, it provides an argument-reflection formula.
The recursion formulas
x (I - x) r
58:5:2 B(v + 1;µ;x) B(v;µ + );x) -
µ 4t
x`(1 x)
58:5:3 B(v;µ + l:x) = B(v + 1;µ:x) +
V v
link two incomplete beta functions. There is a large number of interrelationships connecting three incomplete beta
functions, of which
58:5:4 B(v;µ;x) = B(v + I;µ:x) + B(v;µ + Ix)
is the simplest.
58:6 EXPANSIONS
The important power series
1+ v+µ x+(v+1+)(v+µ+1)x2+._. =(1-x) (v+µxf+.
58:6:1 B(v;µ;x)=x(1-x)"
Y v(v + 1) v(v + l)(v + 2) v (1 + v);
converges for 0 s x < I. When x is close to unity. convergence may be slow and the alternative series
r(µ) 1"(Y) x" (v + µ)
58:6:2 B(v;µ;x) = - (1 +
f(µ + v) µ
may then be preferable.
X = 0 x = 1
r(v) r(o
B(v;µ;x) 0
r(v+µ)
but for certain interrelationships between the v and µ parameters, the incomplete beta function of moiety argument
also acquires particular values. For example:
r2(v) V; no
58:7:1 B(v;v;#)
2r(2v) 4°r(v + ;)
1
58:7:2 B(v;-v;z) =
V
and
1 + (v + 1)/J
+x`
, ( 1 + (v + p. + 1)/J][I + (v + p.)/J1
[1 + (v + 2)/J] 11 + (v + 1)/J]
+
and approximately, for large J, by
//x 2xJ2 3xJ'
58:8:3 Rt=tt (l+x+ + )+smallerterms
I
I (p. -1)x -x)-...)
ti I -x+Al +
If µ = I. this evaluation of RJ is exact, even for J = 1, and B(v;µ;x) is calculated as fo + to + ti/(1 - x). Oth-
erwise, J is determined by adding terms into series 58:8:1 until t;(µ - 1)x/[jj(l - x)2J is less than 10-'. with f)
= fo + to + tj + + t;_1, whereupon J = j and the output value is B(v;µ;x) = fr + tt(l + (µ - 1)x/J(1 -
X)1/0 - x).
If µ is zero or a negative integer, the procedure of the last paragraph is adopted irrespective of the magnitude
of x. However, the recurrence B(v;p.;x) _ -x`-'(1 - x)"/µ + ((v - 1)/µ]B(v - 1;µ + l;x) is first applied as
many times as necessary until B(v + R, O;x) is reached. The sought incomplete beta function is then evaluated as
fo + kB(v + µ;O;x) wherefo and k result from updating at each recursion stage.
If x 0.7 and µ is not a negative integer, the reflection formula 58:5:1 is utilized. The incomplete beta function
B(µ;v;l - x) is calculated as described above, and the three gamma functions are evaluated by a routine similar
to that described in Section 43:8.
Generally, the algorithm is of 24-bit precision (i.e., the fractional error does not exceed 6 x 10-8). Its imple-
mentation may be slow, especially when the argument is close to unity and the µ parameter equals zero or a negative
integer.
Storage needed: v, f, j, µ, g, t. x
Setf=j=0 and a
Set g=t=
If µ + Int(lµI) * 0 go to (2)
(1) IfA=0goto(7)
Replace f by f - x"-'(I - .r)"/µ
Replace t by (v - 1) t/µ
Replace v by v - I
Replace p. by µ + I
Go to (1)
(2) If x < 0.7 go to (7)
Seta=v+µ
If a + Int(jaI) * 0 go to (3) Input restrictions: 0 <- x < 1, v >
Set:=-1 0. R unrestricted but if µ = -1,
Go to (6) -2, -3, ... then (v + µ) > 0
(3) Replace g by g/a
Replace a by a + I
58:9 THE INCOMPLETE BETA FUNCTION B(v;µ;x) S78
If a < 3 o to (3)((
Setf= I+7;3a:-I)
Replace f by [(f - I)/12a] - a[ln(a) - 1]
Replace g by g exp(-f) 2./a
Ift<0goto(5)
If r = 0 go to (4)
Replace g by I/g
Sett=0
Set a=v
Go to (3)
(4) Set r = -1
Set a=lt
Go to (3)
(5) Set f = g
(6) Set a = µ
Set p.=v
Set y=a
Replace x by I - x
(7) Replace t by rx"(I - x)°/v
Set a = 10a(1a - 1)x/(1 - x)2
(8)Replace fbyf+t
Replace t by tx(v + µ + j)/(v + I + j)
Replace j by j + 1 Test values:
If j < jail go to (8) 8(3;3;1) = 0.0166666667
1 1 B(;:0:0.8) = 5.09468017
Replace fbyf+t Q +
108j I xJ B(2;-1;0.2) = 0.0268564487
Output f B(1;-1.5;0.9) = 24.0000000
f = B(v; µ;x) <<<< B(1:1;0.75) = 2.09439510
58:9 APPROXIMATIONS
For small enough values of the argument, the approximation
1 1
58:12 GENERALIZATIONS
The Gauss function of Chapter 60 represents a generalization of the incomplete beta function. The incomplete beta
function may be expressed as a Gauss function with a unity numeratorial parameter
1
Table 58.14.1
f v µ
1+A -A
sinh
2 2
1 -A
cosh
2 2
1 A
sech
2 2
1-A A
csch
2 2
I + A
tanh 0
2
I - A
cosh 0
2
I+A I
sin
2 2
I I-S
cos
2 2
1-A
0 2
I
I-A 1
csc
2 2
I-A 1-A
Inn
2 2
I - A I + A
cot
2 2
58:14 THE INCOMPLETE BETA FUNCTION B(v;µ;x) 580
x?0
58:14:1
I f'(t)dt = I B(v; s;tanh2(x))
for hyperbolic functions and
IT
58:14:2 f f"(t)dt = 2 B(v;µ;sin'(x)) 0< x<
0
for trigonometric functions. The parameters v and .s of the incomplete beta functions depend on a as shown in
Table 58.14.1. For the integral to be finite, the v parameter of the corresponding incomplete beta function must
be positive: this limits the range of X in certain cases; for example, it requires A < I for convergence of fcsch'(t)dt
with zero lower limit.
Interchanging the roles of the v and µ parameters produces expressions for certain complementary indefinite
integrals. Thus, for all hyperbolic functions, except tanh and coth, one has
(' I
58:14:3 J f"(f)dt = 2 B(µ;v;sech=(x))
The same table still applies, but the restriction, if any, on A is now governed by the requirement that p. be positiv
CHAPTER
59
THE LEGENDRE FUNCTIONS P (x) AND Qt,(x)
The functions of this chapter arise in several physical contexts, notably to describe the properties of the surface of
a sphere [see Section 59:14). Most of the chapter is concerned with the bivariate P,(x) and Q,(x) functions, but
attention is given in Section 59:13 to the associated Legendre functions P' "(x) and Q,which are trivariate.
59:1 NOTATION
P,(x) is known as the Legendre function of the first kind and Q,(-r) as the Legendre function of the second kind.
The variables v and x are termed the degree and argument, respectively, of the Lcgcndre functions. Frequently,
the argument is expressed as a cosine so that the symbols P,(cos(O)) and P,(cosh(X)) are often encountered.
Sometimes the title "Legendre function" is taken to embrace the associated Legends: functions that we discuss
separately in Section 59:13. The third variable, µ, in these functions is termed the order of P; '(x) or Q,"(x) and
is zero for P,(x) and Q,(x).
As explained in Section 59:11, slightly different definitions for the Legendre functions may be adopted ac-
cording to whether the argument is regarded as an unrestricted complex variable or as a real number confined to
- I < x < I. Slightly different notations sometimes reflect this distinction. Other notational conventions are sum-
marized by Abramowitz and Stegun [Chapter 8).
59:2 BEHAVIOR
Unless the degree is an integer, the Legendre functions adopt complex values when the argument is real and less
than - 1. Accordingly, we restrict consideration to the range .r a - 1. Perhaps it is better to consider two subranges,
- l < x S 1 and 1 K x < x, inasmuch as the behaviors of the Lcgendre functions are substantially different in
the two. In practical applications - I < x < I is the more important subrange and accordingly it is emphasized in
this chapter.
The contour maps Figures 59-1 and 59-2 depict the rather elaborate behaviors of the Legendre functions. Notice
that these diagrams segment naturally into rectangular domains. The boundaries of these domains are x = ± I and
v = 0, 1 for the P,(x) function; for the Q,(x) function they are x = ±1 and v = -1, -2, -3..... Section 59:7
details the values adopted by the Legendre functions along these boundaries.
The number of zeros of the Legendre function of the first kind is given by
581
59:2 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 582
IP
\ C4
is
4b
4g
,o'o
44
p Ati4?o
4a
yaoti +oo +y +o
ao+o 11,
+ ;a %.
°` pro 0 vo
' + ' +h ' + +
v
:...: /7...:'tuipl--l \ .. \ .... \ .. . y-s
583 THE LEGENDRE FUNCTIONS P"(x) AND Q"(x) 59:3
on n - I<vsn
59:2:1 I0 lsvs0 n=1,2,3,...
n - n - Isv< - n
and all these zeros lie in the range -1 < x < 1. For the Legendre function of the second kind, the formula
3 1
n--<vsn--
2 2
-1 <v5 - - 1
2
59:2:2 N,(Q,) =
n+I+ls _2 -n<v< - n
1
-n- I<v:s - 2 I
specifies the number of zeros as a'function of the degree v. Most of these zeros of Q. lie in the - l < x < I range,
but the asterisk in 59:2:2 indicates that sometimes a single zero is located in the I < x < x range of real arguments.
Note that for -'I < v < -1, -' < v < -2, ... one zero of Q,(x) is located so close to x = - I that it cannot be
properly portrayed on Figure 59-2.
The values v = -1, -2, -3, ... are omitted from formula 59:2:2 [and also from the table of particular values
of Q,(x) in Section 59:7] because Q,(x) is poorly defined for these degrees. Consider Q_r(x) and refer to Figure
59-2. As v approaches - 3 from more positive values, Q,(x) -+ -x for -(1 f) < x < (i /V) whereas Q,(x) -y
x for -1 < x < -(I /V) and (I /V) < x < 1. On the other hand, the approach of v to -3 from more negative
values leads Q,(x) toward +x within the ±1/V range and toward -x outside. At the values x = ±l/V (these
are the zeros of P-3, the first kind of Legendre function), there is a flip in the signs of the infinities so that all the
contour lines crowd to these points and, as it were, penetrate the Q_3(x) = ±x barrier there.
Behaviors for very large argument, and for x in the vicinities of ±1, may be deduced from the formulas
presented in Section 59:9.
59:3 DEFINITIONS
Several definite integrals, including
(
59:3:2 Q,(x) = J Y [x + it - 1 cosh(t)J-"-'dr x? I v> -1
0
tdt
P,(x) _ -sin(rnr) -1 <x< 1 - I <V<0
59:3:3
IT f 1 + 2xt + t'
Q.(x) = r' (1 - r'-;-dt
59:3:4 x>I V>-1
FT;
I (
if[t+ itvI-x2x2]-'-'-[.r-it 1-x'1-'-"}
dt
-l<xsl v > -l
t- - 1
0 0 ',
,yea 40 Oti Ob90'co O'0 0
O U
O 'L co O 'L
4ICJ* A* 4ti' y1 yti
4111 4 4 4 4 4 4 4 4 4 4
!....x.../...:/...1.1.. 11111 .1 A.v .. \.:\.
4 4
.
4
;. 4
V-5
2
585 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 59:4
V-2 rx cosh((v + r)
59:3:7 P,(cosh(X)) = - s) dt cosh(X) = x > 1
'a J o cosh([) - cosh(X)
1 exp(-(v + )t)
59:3:8 Q,(cosh(X)) _ -
I cosh(:) - cosh(X)
x > 1
and
cos((+ 1)r)
59:3:9 P,(cos(e)) _ dt 0<6<-rr
Jo cos(t) - cos(9)
the last being known as the Mehler-Dirichlet formula.
The formulas of Section 59:12, and especially equations 59:6:7 and 59:6:8. often serve to define Legendre
functions as special cases of the Gauss function.
With c, and c2 as arbitrary constants, a linear combination of the two Legcndre functions generally satisfies
Legendre's differential equation:
d
In the notation of Section 43:14, the Legendre function of the first kind may be synthesized from the binom
function (1 + x)` by the procedure
rl +xl 0 X=
59:3:11 P,(x)
2 -v
When v is the nonnegative integer n, the Legendre function P,(x) reduces to the Legendre polynomial discussed
in Chapter 21. The same polynomials result when n is a negative integer, on account of the identity
59:4:1 polynomial of degree n n = 0, 1, 2, ...
The Legendre function of the second kind becomes an inverse hyperbolic function [Chapter 31 ] when its degree
is zero:
Qtr) artanh(x) -1 <x< 1
59:4:2
arcoth(.x) [ri > I
Other integer degrees lead to the following special cases:
{P(x)Qox) - polynomial of degree (n - 1 ) n = 1, 2. 3... .
59:4:3
-x n=-1,-2, -3, ...
See equation 59:5:9 for details and Section 21:13 for examples.
When the degree v is an odd multiple of _4, Legendre functions of both kinds reduce to complete elliptic
integrals [see Chapter 61]. The simplest cases are
59:4:4 P-,12(x) =
2
K1 2
1 r/ -1 5 x 5 1
2 2 x 1
K x >I
it x+1 \ x+l/
59:5 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 586
1+
xs
59:4:5 Q_ 112(x) _
K
F2 -1 2
z?
+I
K(x+l? z >I
4
E( P_,,2(x) -1 5 x 5 1
n 2
59:4:6 Pu2(x) = P-3/,(X) _
2
x+E x- x+
2 x' I x> I
/
Kl
Iyx / l+x\
2-2E11 -lax _1
\ YY
59:4:7 Q12(x) = Q_3;2(x) _ \
/ 2
2x + 2 EI I x>I
x+l
and others may be deduced with the aid of recursion formula 59:5:5.
59:5 INTRARELATIONSI{IPS
The argument-reflection formulas
59:5:1 (-1), P,(x) n = 0. 1. 2....
59:5:2 -(-1)" Q.(x) n = 0, 1, 2. .. .
are valid only for integer degree, whereas the degree-reflection formulas
59:5:3 P-..-I(x) = P,(x)
59:5:4 a cot(vvr) P,(x)
have general validity for real arguments.
The same recursion formula
59:5:5 f=PorQ
is obeyed by Legendre functions of both kinds.
The equations
n
59:5:6 Q,(±x) = Z [cot(vrr)P,(±x) - csc(va) P,(+x)] v * 0, ± 1, t2, ... -1 <x< 1
-2
59:5:7 P,(±x) _ - [cot(vir) Q,(tx) + csc(vu)
n
v*0, t1,±2,... -1<x<I
(-v);(1 t v)
59:5:8 Qv(x) = P,(x)[Qo(x) - 4.(v + 1) - y] + [y+ 4i (j+ )1 (_2V 1
(j!)2
-t 0, =I, -t 2, ...
interrelate the two kinds of Legendre function when the degree is not an integer. See 59:4:2 for Q0(x), Chapter 44
for J and y. When the degree is a positive integer:
59:5:9
QO(x)-1--I
2
I
-3 ---1+ n
(n+ j)!
In - j)!(jp2
I+
1
2
+
1
3
+ +
1
j
x -I
2
J
n= 1,2,3,...
S87 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 59:6
59:6 EXPANSIONS
As detailed in Section 59:12, Legendre functions may be expressed as Gauss functions in a multitude of ways.
Because each of these latter functions may be expanded as the power series 60:6;1. the Legendre functions have
very many expansions. Here we present only those that are the most useful computationally.
If. for -I < x < I, we define the auxiliary Legendre functions
r vl (I +v
r v 1+v 1 l1_ (2x)''
59:6:1 lef,(x) = FI
2, 2 2 xZ 1- \2 2 /j (2j)!
and
then
I r(l 2 vv) rl I + ZJ
I /va Ivn\
59:6:3 P,(x)
= 7 r(I+
/ cost - f lef,(x) +
2 1'
\
vl
sinl 2 I leg, (x)
z
-
2 (2,
whereas
-rl l 2 l)
59:6:4 Q,(x) = / sin(7) lef,(x) +
r\ 1 + 2/ /vrr\
cos( - I
2 V
rl I + 2 I rI 2 I +v
For the same argument range, one may use the trigonometric series
r(1 + v) x (1 + v),(4);
59:6:6 Q,(cos(O)) = cos[(1 + v + 2j)91 0<0<n
2v))(1 2 -3 -5 -7
59:6:9 Q.(x) _
r(i - v)(2X)
. E
,-o I!(i - v)jx2'
= Q- -1(x) v=
2 2
2,.. x>I
in conformity with 59:5:4.
x=-1 x0 x = 1 x = z
P,(xl
2.4.6...
Slv=-3.-5.-7....=-1-2n
2n i (2n - 1)"
2)'n'
I+ v
2
P.(c)
1 <v<2, 3<v<4.5<v<6.... ( l
3<v<-2. 5<v<-4.-7<v<-6, }
Var(I _!)
v= 1.3.5....
P,Ix I 0
r(1
P,[x)
10<v<1,2.<v<3.4<v<S.- -2<v<-I,-4<p<-3.-6<v
-x cos
5, }
fr I+ -v) 2
2
Po(x). P.,(x)
P.(x) -I <v<0 0
1 5 9 1
Q,(x) 2n+ 0
2 2 2 2 (4n + 1)!!!!
Q,(x) v = 0, 2, 4. 0 0
-1 3711 1-IT1?(4n-3)"!!
2'2'2' -2n--
22
I -V.-r(1v/
Q.(x) x 0
2 2U(4n - 1)!!!'
- I. I3 5 z
2 2 v<,
7 vn\
Q.(x) - 1 <v< < 2 2 <v< 2 sin 0
2
2v
V-. r
Q,Ax) -< v<-),- <v<-3,- <v<-5,
-3
2
-7
2
-11
2
)
cos(z -x
2r (1-v)
2
Q,4x) v = 2.
_3 _7 _II
2. 2
-3
2- 2n
I
2
_n -4-1)'0(4n- I)!'!!
(4n + 1)!^!
0
589 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 59:8
x= 0 x= I x= x
r( 2v) 'n
-3
2,-4<1,-7<2,-6<v<-,...
-11
Q,(x) -2<v< v)
roll
2r(1
2
V.r'
-5 -9 -13
0.(x) yV) cosI
2 2 2 /
2r(
2
-5 -9 -13 -I W (-1)',r(4n-3)r'!!
V -2n = 0
QJx) 2
2 2. 2 2 2 2U(4n - WT
ar(
5 -9 -13 2v! m
Q,tx) -3<v< 2,-5<v< 2.-7<v< 2 .... x cos -
2
2r(1-v)
The quadruple factorial (4n - I)!!!!! that, with its analogs, appears in several of the tabular entries is defined in
Section 2:13 and equals 3 x 7 x I I x x (4n - 5)(4n - I). The constant U [Section 1:71 equals 0.8472130848;
note that P1,2(0) = 2U/a, P_,12(0) = 1/U, Q,,2(0) = U and Q_,,2(0) = 'rr/2U.
Input x >>
Setg=u=2x
Setf=t= I
Setk - '
Set X = I + [10"/(1 - x2)]
(4) Replace t by tx2(k2 - w)/[(k + 1)2 - 11
4
Replace k by k + I
Replace f by f + t
Replace u by u--, (k2 - w)/[(k + 1)2 - 411
Replace k by k + I
Replace g by g + u
If k < K go to (4)
If (XtI > Ifi go to (4) Test values:
Replace f by f + [x2t/(1 - x2)] P,(0.5) = 0.223144533
Replace g by g + [xIu/(I l2)] Q7(0.5) = 0.343915293
Set p = I(sgR) + (cf/R)]/V a P_2(-I/a) _ -0.318309886
Output P Q_ (-1/rr1 = 2x
p - P,-W< IK«« P3/2(0) = -0.393446867
Set q = aN/7rl(cgR) - (sf/R)1/2 Q72(0) = -0.618024892
Output q P,(8/9) = 0.888888889
q = Q,(x) Q1(8/9) = 0.259205931
For J > 2 w2 vl the infinite series t, + t,_, + t,_. + is well approximated by the geometric series t,(I
-
+ x2 + x4 + x° + ) and the sum is therefore approximately t,/(1 - x''). Hence, the fractional error in truncating
the infinite sum to + t, + t, + at the t,_, term is approximately [t,/(l - x2)1/Ito + It + + t,_, + t,(I -
x)-']. The algorithm makes use of this principle in deciding when to truncate, thereby ensuring that lef, and leg,
are computed to a precision of about 6 X 10-°. Close to the zeros of P,(x) and Q,,(x), this precision may not be
carried over to the Legendre functions themselves.
The algorithm is valid only for arguments in the range -1 < x < I and is extremely slow close to the boundaries
of this range.
59:9 APPROXIMATIONS
Close to x = I, the Legendre function of the first kind is approximated by the linear function
sin(ir) Inr t 2+ xl
I
Here 'h is the digamma function [ Chapter 44] and y is Euler's constant [see Section 1:4].
591 THE LEGENDRE FUNCTIONS P,(.r) AND Q,(x) 59:10
59:9:5 P,(x) =
V.- r(1+v)
v> -1
r(21 + v)(2x)' '
IT (-2v - 2)!! -3 -5 -7
59:9:6 Q.(x)
N/2 v)!
(4x)" v = 2 , 2 , 2 , ... x- z
V'nr(i -v)
r(-v)
cot(var)(2x) " ' - 1
>v* Z,3 Z,5 27
..
r -1
fj'(x) -1 < x < 1
59:10:1 ; U X) = S 1 } f= P or Q
I
59:10:2
f P,.(r)dr = 1 - x= P;-"(x) -1 <x< 1
when the integration limits are - I < t < 1. These formulas are indeterminate when w = v; in that event the three
integrals give [I - 2/1r2) sin2(vrr) ib'(I + v)]/(} + v), -sin(2vrr) qi'(1 + v)/[(I + 2v)tr] and {(w2/2) - (1 +
cos2(v'rt)]y'(I + v)}/(l + 2v), respectively. Similarly, the integral between limits 1 < t < x:
4r(1 + w) - ir(1 + v )
59:10:9 J Q.(t) Q.(t)dt =
f (I+v+w)(w-v)
has y'(1 + v)/(1 + 2v) as its special w = v case. but
I
59:10:10 F P,(t)Q (t)dr = w>V>0
Ji (I+v+w)(w-v)
diverges if the degrees are equal.
Since Pa(t) = 1. P,(t) = t, { P,(r) + } Po(t) = r', etc. [see Section 21:5], the above formulas may be adapted
to give many definite integrals of the form ft'f,.(t)dt for f, = P, or Q,. Very many other definite integrals are listed
by Gradshteyn and Ryzhik [Sections 7.1 and 7.2].
With the real argument x replaced by x + iv. equation 59:12:1 serves to define the complex-valued function P,(x
+ iy) as F(-v,l + v;l;(l - x - iy)/2). The corresponding formula
it 1 - x - iy / I + x + iy1
59:11:1 Q,(x + iy) = 2 [cot(vrr) z i] F -v,1 + v:l; 2 - 2
csc(vrr) FI -v,1 + v;l; 2
provides a valid definition of the Legendre function of the second kind for all combinations of x and y except when
y = 0 and > 1. The alternative signs in 59:11:1 apply accordingly as y > 0 (upper sign) or y < 0 (lower sign).
For y = 0 and - I < x < 1, definition 59:11:1 yields a function that is multiple valued and complex. To avoid
this difficulty, it is usual to redefine Q,(x) as the average value of the two limits Q,(x + iv) and Q,(x - iy) as y
-+ 0. This removes the i from 59:11:1, which then reduces to 59:5:6. This convention has been followed in all
equations of this chapter (except 59:11: 1), so that our Q,(x) is always real for real argument.
59:12 GENERALIZATIONS
Legendre functions may be generalized to the Gauss function of the next chapter. Included among the ways in
which Legendre functions may be expressed as a single Gauss function are
+ v;1; .2
_
I-x -1 <x < 3
59:12:1 V,
)
/
I+v 1-v ,l+Z;l;l-x'
v
59:12:3 P,(x)=Flv 2, 2 ;1;1-x=)=xF( 0<x<V2
1 2 -1
59:12:4 1,14) = [x + x= I J'' F v, :1;
-2 x+ z'- 1
(x - r' - 111/21. 1 2V7- I
+ x- I `2 x+ xr- 1
593 THE LEGENDRE FUNCTIONS P,(x) AND Q"(x) 59:12
2 x'- 1
59:12:5 P"(x) _ [x - x' 11" F -v, 2;1;
'-1-
[x + V77-111/2-
2 x' - 1 3
_
1
V; r(I + v) 2
59:12:7 Q"(x)= F(I+v,1+v,2+2s'; x>2+I
r(i + v)(2x 2)'-" \ 1 ± x/
Var(1+v) /1+v v3
59:12:8 Q"(x) =
[(,' + v)(2x)''"
F(-,l 2 22
1
x
x>I
59:12:9 Q"(x) _
r(? + v)(2Vx' - t)i_
- F lfl+v 1+v3
r(I+v)
2 '2
- + v;
I
1 - x-
)
1 x>
59:12:10 Q"(x)=
=
r(1 + v)
r(,+v)
V (r(I
V
IT(x -
2V'-
z
" j 3
F( I+v,l+v,-+v;
1
1
F\2'2'2+v'
(1
3
2l
x' -
V
.rl
/
x>- 3
59:12:13 P"(x)
r(-v - D
FI
/I+v I+v3- + v; 1
-
2Vnx
FI
1-v ,1 +-;-;z'
v3
-1 <x< 1
r
(l + v1
2
I2/
r (-v} \ 2 22
59:13 THE LEGENDRE FUNCTIONS P,(x) AND Q,(x) 594
r(-v-1) / l+v'I v3 1 1
59:12:15 P,(x) = F + -; - + v; ;
1` r(-v)(2x)' " 2 22 x
+
z + v)(2x)' /-v
FI`- Iv I - v; -/1 I
x>1
V'nr(l +v) 2 2 '2 X.
59:12:16 P,(s) =
C(- v - (x -
F
(1 1 3
+ v; - -
1 Vc J
r(-v) 2a x- 1 222 2 2 x-
r(+ v) (x- ' (1 11 x 3
F I
r(1+v) 2n x'-l \-,2:2
r(-v-1.
2 2V77- -1 Vis
3
59:12:17 P,(x) = Fr-1 + v; - + v;
fir(-v)(x+ 12 2 + x=- 1
r( + v)(x + 1
+ F -,-v: - - v: x>1
r(I + v) 2 2 x + x' - 1
v)
r( l
2
J F( Zv, 1
2 v:
59:12:18 Q,(x) = sinl
zr(l + z) 2 2
r(i+ZI/
- \\
v\
cosva- xF 1-,- Iv + -;v223
-;x- -1 < x < 1
z 2
r(1
2 /I
Many of these formulas become invalid, or require modification, if v is an integer. For x > I the substitution x
= cosh(X), x= I = sinh(X), s t V = exp(tX/2) often yields simplifications.
[see 59:14:4 for a trigonometric equivalent of this equation]. Replacement of x by I - 2x and then f by (x= -
x)"Rf leads to an example of the Gauss equation. 60:3:4, and accordingly associated Legendre functions are in-
stances of the Gauss function of Chapter 60.
Generally, these functions are complex even when their arguments are real. However, it is conventional to
adopt redefinitions similar to that discussed in Section 59:11 to ensure that P;"'(x) and q."(x) are real for real x
between - I and + 1. Here we discuss this range exclusively. Moreover, we shall emphasize cases in which v and
µ are nonnegative integers, using n and m to represent the degree and order, respectively, in these cases.
Calculation of values of the associated Legendre functions is possible via the definition
595 THE LEGENDRE FUNCTIONS P,(x) AND Q(x) 59:13
(l+V+µ\
2 \
I II vtr + µa\
P;°'(x) = cos I Ief,'(x)
r( I + vµ)
59:13:2
V.-(1 - x'-)"r_
2
2
r(i+v+µ\
2 /I
sin Ileg;,"'(x)
r(l+v- 2
\ 2 J
rll+v2IL
/ /
rrl+v-µl cos(m + µ J Ieg;1"(x)
+
1\ 2
Fx-
2 J
59:13:4
59:13:5
Ief;"r(x) =
Ieg;"'(x) = 2xF
(-V - µ 1+v-µ 1
l-v-µ +v-;-µ
;x 3I_ - v - µv-I +µ(2x)"='
i yI+v-) µI (2x)2J -
The algorithm of Section 59:8 may be modified to use these equations, although four distinct gamma function
evaluations are required instead of the single gamma function ratio that sufficed with equations 59:6:1-59:6:4.
Some interrelationships between associated Legendre functions are
59:13:6 W
Pt'-1W W x= 11."W ()
59:13:7 (tan(") - tan(µa)] Q"0_,(x) _ [tan(var) + tan(An)1 Q."(.0 - TrPV`(x)
" r(1 + v - µ) ", 2 sin(µa) "
59 : 13 : 8 P (x) = [cos(µa) P (x) - Q, (x)1
r(1 + v + µ) it
:_" (x) r(1 + v - µ) ir sin(µ-rr)
59 : 13 : 9 Q = r(1 + v + µ) Icos(µ'n) Q :"'(x) + 2 P;"' (x)1
59 : 13 : 11 '' -
Q ( x) _ -cos(av + aµ) Q;"'(x) - a sin(wv + nµ)
2 ;
P ' (x)
5x - 3x'`
59:13:14 P':''(x) = 3(1 - x'-) Q," (x) = + 3(1 -xartanh(x)
1 -x'
59:13:15 P;°'(x)=(-I)"(2n- 1)!!(1 -x22)"' n=0.1.2,...
and others may be obtained via the differentiation
or recursion
(1 + 2rt)x n + in
59:13:17 (x) _ f' '(x) + f'; ',(x)
1 +n - m 1 +n - m
- 2mx f=PorQ
59:13:18 f," (x) = f"'(x) - (I + n - m)(n + m) f'"-"(x)
71=-=T
formulas. Formulas 59:13:12-59:13:15 demonstrate the inappropriateness of the name "associated Legendre poly-
nomials" sometimes given to these functions. Be aware that the (- 1)" factor in 59:13:16 is often omitted so that
associated Legendre functions of odd order may be encountered with signs opposite to those employed in this Atlas.
The associated Legendre functions of the first kind satisfies the orthogonality relationships
10 N+n
59:13:19 f P, '(t) P;,"' Will = 2 (n + m)!
J N=n
' 2n + l (n - m)!
but no such relationship holds for the second kind.
59:14:1
r2 d'R
R dr2
2r dR
R dr
I a'Y cot(o) ay csc2(o) d'Y
- - + - - - kr- = h = - - - - - - -
Y ao' Y ao Y a8'
-
where R is a function only of the radial coordinate r, Y is a function of the two angular coordinates, 0 and +, and
h is a separation constant. With m2 denoting a second separation constant. the second equality in 59:14:1 may be
further decomposed into
sin2(o) d26 sin(20) d6 I d2I
59:14:2 + h sin'(o) = m =
A do' + 29 do 11.$12
where S",,, and C",,, are redefined constants. The components of this solution are known as spherical harmonics or
surface harmonics and, in light of orthogonality relationships 59:13:19, 32:10:25 and 32:10:26. they are usually
defined in the normalized versions
2n +
59:14:8 Y";"(8:m) I I (n + m)! f(mm) P;,"'(cos(8))
21T
///f=sinorcos n=0,1,2,...
However, normalization conventions sometimes differ from author to author.
As their name suggests, spherical harmonics play a vital role in describing oscillatory behavior in systems of
spherical symmetry: for example, in the quantum mechanics of atomic electrons. One distinguishes tonal surface
harmonics when m = 0, sectoral surface harmonics when in = n and tesseral surface harmonics when m = I, 2,
Finally, let us return to the first equality in 59:14:1 with n(n + I) now substituted for the separation constant
h. Replacement of the radial coordinate r by x/Vk (or by .r/' k for negative k) and the dependent variable R
by then leads to
d'f df
59:14:9 x--+x--[(n+?)-'-x'Jf=O
dx- dx
n=0,1,2....
where the alternative upper/lower signs apply as k is positive/negative. This is the (hyperbolic Bessel)/(Bessel)
equation 50:3:4/53:3:5. and the lower alternative is satisfied by
f A.
59:14:10 R = = ] + J. 1/2_(x) = a"j"(x) + b"y"(x)
where J, represents the Bessel function [Chapter 53J; A", B", a" and b, arc arbitrary constants; and j, and v" are
spherical Bessel functions [see Sections 32:13, 53:4 and 54:4J.
CHAPTER
60
THE GAUSS FUNCTION F(a,b;c;x)
A surprisingly large number of simple intrarelationships make the quadrivariate Gauss function unusually flexible.
It embraces many of the functions discussed in previous chapters and may be further generalized as explained in
Section 18:14.
60:1 NOTATION
The Gauss function is also known as the hypergeometric function or as the Gauss hypergeomerric function. The
subscripted symbolism ;F1(a,b;c;x) is sometimes encountered; Section 60:13 contains an explanation of the "2"
and "1" numerals.
As usual, the x variable in the Gauss function is its argument. Because of their locations in expansion 60:6:1,
the a and b variables are known as numeratorial parameters, while c is a denominatorial parameter. There is a
second denominatorial parameter, equal to unity, whose presence is not explicitly displayed in the F(a,b;c;x) no-
tation but is brought out in the first of the following alternative symbolisms:
C(c) f a- 1,b- 1 a,b;x
60 1 1 J = 2F11 = F(a,b;c;-r)
5a)1'(b) Lx 0, c c
60:2 BEHAVIOR
Unless one of the four quantities a, b, c - a or c - b is a nonpositive integer (in which event, see 60:4:10, 60:4:11
or their analogs), the Gauss function is defined only for real values of its argument in the range -- < x < 1. The
domain may be extended to embrace x = I provided that c > a + b.
When c is a nonpositive integer the Gauss function adopts infinite values [unless a or b is also an integer such
that a - c or b - c equals 0, 1, 2, .... so that 60:4:11 applies[- Nevertheless, significance can always be attributed
to the ratio F(a,b;c;x)f[(c), even when c = I - n = 0, -1, -2, ..., because of the limiting operation
I F(a,b;c;x) (a)"(b)"x"
60:2:1 lim r(c) = (b F(n + a,n + b;n + l;x) n = 1, 2, 3, ...
Because it is quadrivariate, the Gauss function displays such a wide variety of behaviors that it is impractical
to depict graphically.
60:3 THE GAUSS FUNCTION F(a,b;c;x) 600
60:3 DEFINITIONS
Expansion 60:6:1 provides the usual definition of the Gauss function for - I < x < 1. The transformation 60:5:3
permits extension to -x < x < ;.
The Euler hypergeometric integral
r(c) ?-'di
60:3:1 F(a,b;c;x) =
r(b) r(c - b) f (1 - t)1° `(1 - xt)°
serves as a definition, as do the equivalent definite integrals
c>b>0 x<I
r(c) (r 1)-°-'r°-dt
60:3:2 F(a,b;c;x) = I + a > c > b x<l
r(b) r(c - b) (r - x)°
an d
r(c) t° 'dt
60:3:3 F(a,b;c;x) = c> b> 0 x< 1
r(b) r(c - b) JG (I + :)`-°(l + x - xt)'
Because of identity 60:5: 1, the two numeratorial parameters may be interchanged in any of these definitions.
A linear combination of two Gauss functions satisfies the h}pergeometric differential equation
df
60:3:4 x(1 - x) --- +
dx
[ I - Y-(1+a+0kldx-aRf=O
f= c,F(a,P:I - y;x) + cx'F(a + y,3 + y;I + y;x)
60:3:5 (1 - x)'°
b
1-c F(a,b;c;x) -1 < x < I
60:4:2 F(a,,a,;(,,;x)
i-o (l)t(c2);
A glance at Table 18.14.2 will reveal the large number of instances in which important functions conform to eithe
60:4:1 or 60:4:2.
If the two numeratorial parameters of the Gauss function are interrelated by any one of the three relationships
60:4:3 6=a-;,a+;orl -a
or if the denominatorial parameter is related by any one of
12l+a+b
60:4:4 c= 2a,2b,a+b-,a+b+ I +a - boil + b - a
2
601 THE GAUSS FUNCTION F(a,b;c;x) 60:4
to the numeratorial parameters, then the Gauss function reduces to an associated Legendre function [Section 59:131
or to one of the special cases (a Legendre function, a Legendre polynomial or a complete elliptic integral, among
others) of the latter. Of these 10 possibilities, seven are detailed in Table 60.4.1 by listing the appropriate values
of "factor," µ, v and X in the identity
60:4:5 F(a,b;c;x) = (factor)Pr"'(X)
The three omissions are easily derived by interchanging a and b.
When the denominatorial parameter equals i (or ), reduction occurs to the sum (or difference) of two associated
Legendre functions; thus:
60:4:6 F(a.b;l + ;;x) = (factor)[P!,`(X) :t PY1`(-X)1
These cases arc also included in Table 60.4.1.
Another trivariale function that is a common special case of the Gauss function is the incomplete beta function.
If either of the numeratorial parameters of the Gauss function equals unity, or is less by unity than the denominatorial
parameter, we have reduction as follows:
to instances of the function discussed in Chapter 58. Powers, polynomials, logarithms and inverse hyperbolic or
trigonometric functions may arise by further specialization of the incomplete beta function, as explained in Section
58:4.
When the denominatorial parameter c equals one of the numeratorial parameters, say b, we have
Table 60.4.1
Parameter
relation Factor µ v X Restriction
where P?"(x) is a Jacobi polynomial [Section 22:12). A similar polynomial combined with a binomial function
[Section 6:14) is generated if a - c or b - c is a nonnegative integer; thus, for example:
n (c - a)I
60:4:11 F(a,c + n;c;x) = (1 - x)-"-" (-x)
,=o J (c)J
60:5 INTRARELATIONSHIPS
The Gauss function is symmetrical with respect to interchange of the two numeratorial parameters
60:5:1 F(b,a;c;x) = F(a,b;c;x)
The transformation
F(c - a,c - b;c;x)
60:5:2 F(a,b;c;x) =
(I - x)"ra-,
which may be regarded as a reflection formula for both numeratorial parameters, provides a relationship between
two Gauss functions of common argument. On the other hand, the transformations
X / x
FI a,c - b;c; - I
x-1///
60:5:3
(1 - x)" (I - x)b
relate a Gauss function with argument in the range 0 t x s 1 to ones with arguments in the -x <_ x <_ 0 range.
The six functions F(a ± I,b;c;x), F(a,b ± I:c;x) and F(a.b:e ± I:x) are said to be contiguous to the Gauss
function F(a,b;c;x). The function F(a,b;c;x) is linearly related to any pair of its contiguous functions. Thus, there
are 15 contiguity relationships. Three of these are
60:5:4 (2a - c + bx - ax) F(a,b;c;x) _ (a - c) F(a - l,b;c;x) + (a - ax) F(a + l ,b;c;x)
60:5:5 (a - b) F(a.b;c;x) = aF(a + l,b;c;x) - bF(a,b + I ;c;x)
and
and the remaining 12 may be found in Erdilyi, Magnus, Oberhettinger and Tricomi (Higher Transcendental Func-
tions, Volume 1, Section 2.8, formulas (33)-(44)1. These contiguity relationships constitute recursion formulas,
enabling F(a + l,b;c;x), for example, to be expressed in terms of F(a,b:c;x) and F(a - I ,b;c;x).
603 THE GAUSS FUNCTION F(ab:c;x) 60:5
The 15 contiguity relationships provide intrarelationships linking three Gauss functions of common argument,
but there also exists a plethora of relationships between trios of Gauss functions with dissimilar arguments. Three
typical relationships of this class are
/ \
F(X.l +A - c;I + 2X-a-b:1
r(c) r(a + b - 2a) x
60:5:7 F(a,b;c;x) _ X<0
r(a+b-\)r(c-k) (-x)"
and
The E notation in these three equations indicates that their right-hand members consist of two terms that differ only
by virtue of A adopting either of the two indicated values. These relationships may fail if the argument of one of
the numeratorial gamma functions equals a nonpositive integer see Abramowitz and Stegun [equations 15.3.10-
15.3.141 for expansions that apply in these exceptional cases.
Because they interrelate Gauss functions whose arguments are linked by a linear relationship, formulas 60:5:2,
60:5:3 and 60:5:7-60:5:9 are examples of linear transformations; the Bateman manuscript [see ErdElyi, Magnus,
Oberhettinger and Tricomi, Higher Transcendental Equations. Volume I. Sections 2.9 and 2.11] contains a catalog
of these transformations. Apart from the exceptional cases noted in the previous paragraph, linear transformations
of Gauss functions are valid for all values of the numeratorial and denominatorial parameters. In contrast, the set
of so-called quadratic transformations, in which a quadratic relationship exists between the Gauss function argu-
ments, holds only if one of the following restrictive conditions apply:
I I
a + b = 1,c+-,c--or 2c- 1
lorc+ l1
13
c=2, 2, 2a or 2b
These 12 conditions are precisely those detailed in Section 60:4 that cause reduction of the Gauss function to
or two associated Legendre functions. Accordingly, the table in that section (Table 60.4.1) may be used to deduce
quadratic transformation formulas. For example, the first two tabular entries may be separately reformulated as
2 F
l+v
2 1+
v2 -:I-µ:l-1
µ1 `
I
60:5:11 P;" (X) = ( X= >0
It - X2) r(1 - µ) x M
and
1-X
ax
F v, 1 + v; I - µ;
2
60.5.12
p"(X)-(I+X)w
1-X rU - µ)
X=I-2x>-I
The equality of the right-hand members of 60:5:11 and 60:5:12 constitutes a quadratic transformation, valid for X
> 0. The Baseman manuscript [see reference above] includes a comprehensive listing of quadratic transformations
and information on cubic transformations.
60:6 THE GAUSS FUNCTION F(a,b;c;x) 604
60:6 EXPANSIONS
(0 a<0,b<0
60:7:3 F(a,b;c;-x) _ {lx 1 (a or b) = 0. (b or a) < 0
(aorb)>0
These are the only particular values that can be formulated without placing severe restrictions on the parameters.
Some particular values under such restrictive conditions include
V; r(1 +a-b)
60:7:4 F(a,b;l+a-b;-1)=
/ ``
a-b*-I,-2.-3, ...
2'TII+Z-b\\ Ir(I2aJ
/
r/I+a+bl
V I\
1+a+b 1 l _ 2 /J
60:7:5 F a,b;
2/
a+b*-1,-3, -5, ...
2 (l + a } (l + bl
r\ 2 / r\ 2
and
r
F a,1 - a;c;
1- 1
I=
2' r(r) c + 0, -1. -2... .
60:7:6
2 2'rla+9(1+2C -a)
2
605 THE GAUSS FUNCTION 60:9
No particular accuracy is claimed for this algorithm, but the precision will generally be better than 24 bits.
The algorithm will often be intolerably slow if x > 0.9 or if x < -9. The linear transformations 60:5:9 or 60:5:8.
respectively, are useful in these circumstances.
60:9 APPROXIMATIONS
multiple differentiation:
and integration:
c - l
60:10:3
o
F(a,b;c;r)dr =
(a - 1)(b - 1) [F(a - I.b - l;c - l:x) - 1]
modify each of the three parameters equally. One may, however, selectively alter a single parameter of the Gauss
function by such operations as
d"
60:10:4 x'-b - {x"-1+h F(a,b;c;x)} _ (b)" F(a,b + n;c;x)
dx"
-1-b do x"-1+,-b
X LL-
60:10:5 F(a,b;c;x) (c - b)" F(a,b - n;c;x)
=
60:10:6
I
(1 - x)`
=
(c - a). (c - b)
(c)"
F(a,b;c + n;x)
d"
60:10:7 x'-"" {x-' F(a,b;c;x)} _ (c - n)" F(a,b;c - n;x)
--
dx'
Differintegration of the product of a Gauss function and a power obeys the rule
and generally creates a K = L = 3 hypergeometric function [Section 18:14]. By suitable choices of v and µ,
however, the generated function may be of the K = L = 2 or even the K = L = I class. The notation of Section
43:14 enables these conversions to be expressed succinctly; for example:
I-a
60:10:9 F(a,b;c;x) -' F(A,b;c;x) X = x
I - A
1 -b
60:10:10 F(a,b;c;x) -+ (I - x)-" X = x
I-c
60:12 GENERALIZATIONS
Because
(a),(b),X;
60:12:1 F(a,b;c;.r)
izo (1)r(e),
607 THE GAUSS FUNCTION F(a,b;c;x) 60:13
60:13:3 i
,=u (c),(c:), ... (c,)i
(a1)i(a>1.
. (axx)
irrespective of whether one of the denominatorial parameters happens to be unity. Moreover, the integers K and
L, which we use to characterize the numbers of numeratorial and denominatorial parameters, include any unity
parameters. Thus, the Fd function generally has K = n and L = I + d. In the F. notation, function 60:13:3 would
be symbolized K.,F,(I,al,a_, ... aK;CI,C,, ... CL;x).
CHAPTER
61
THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E(p)
The complete elliptic integrals of the first and second kinds are simple. although important, univariate functions.
More complicated. but less important, is the complete elliptic integral of the third kind, discussed in Section 61:12.
All these functions arise in the canonical representation of integrals of certain rational functions that do not reduce
to simpler expressions.
61:1 NOTATION
Throughout this Atlas we normally use the name "argument" and the symbol x to represent the single variable of
a univariate function. To achieve uniformity with Chapter 62. we here violate this rule and use the name modulus
(module is also encountered) and the symbol p to represent the variable of K(p), the complete elliptic integral of
the first kind, and of E(p), the complete elliptic integral of the second kind. Similarly, we denote the complete
elliptic integral of the third kind (Section 61:121 by f1(v:p).
An important auxiliary variable is the complementary modulus. defined by
61:1:1 q=
The symbols k and k' frequently replace p and q. As well, the role of the modulus is often taken over by the
modular angle a or the parameter in, where
6 1 : 1 : 2 p = k = sin(s) ' V In
61:2 BEHAVIOR
Figure 61-1 illustrates the simple behaviors of the complete elliptic integrals. They acquire values n/2 s K(p) 5
= and n/2 ? E(p) ' 1 over their usual 0 - p - I domain. The inequality
61:3 THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E(p) 610
IT
61:2:1 1.57 = z a K(p) + In(q) >- In(4) - 1.39 0- 5P : 51
shows that K(p) + ln(q) varies little. The behaviors of K(p) and E(p) asp -+ 0 or p -. I can be deduced from
the expansions presented in Section 61:6.
In many applications the complete elliptic integrals of complementary modulus are almost as important as K( p I
and E(p). For this reason. Figure 61-1 includes maps of K(q) and E(q).
Normally, interest in complete elliptic integrals is confined to the zero to unity range of modulus. When p
exceeds unity. K(p) and E(p) are complex valued; this circumstance is discussed in Section 61:11. The evenness
of the functions, which is evident from equations 61:3:1 and 61:3:2, permits extension to negative moduli. Both
complete elliptic integrals acquire real values when their moduli are imaginary numbers of any magnitude: again,
see Section 61:11.
61:3 DEFINITIONS
The complete elliptic integrals are defined by
61:3:1 K(p)=1
I -p'sin'(0)
dB dt
1` (1 + r)(1 + q=t')
and
t 1 _ P::: 1_
61:3:2 E()
p- Jo
1 - ' sin2(0) dO =
p" -
f 1 - r"
dr =
j1, (1 + r-
q
dt
the two integrals that involve the angle 0 being illustrated in Figures 62-2 and 62-3.
Alternative definitions are as special cases of the Gauss function [Chapter 601:
n 1
K(p)=2F( 2.2:l;p'
1 - n F( 1
61:3:3
3
61:3:4 F-4 p) = 2 F(2 , 2:1;p') = 2
1 1
FI
2.
1
2;l;p')
or via Legendre functions of moiety order [see Section 59:4).
The perimeter of the ellipse [see Section 14:14] shown in Figure 61-2, which has semiaxes of q and 1, equals
4E(p). The common mean [Section 61:81 of the two semiaxes is 'rr/12K(p)].
In the notation of Section 43:14, the complete elliptic integrals may be synthesized as follows:
61:3:5
On" ,-0
K(p) i 2 V t- p 0 E(p) X= p'-
61:5 INTRARELATIONSIIIPS
Both complete elliptic functions are even:
61:5:1 f(-p)=f(p) f=KorE
The four complete elliptic integrals K(p), E(p), K(q) and E(q) are linked by the remarkable Legendre relation
611 THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E(p) 61:5
FL. a lb %
Quo
. . .
. .
Qp ..............
.
f Qp Qp
. .
;~
2.8
61:6 THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E(p) 612
IT
61:5:2 K(p) E(q) + K(q) E(p) _ + K(p) K(q)
Complete elliptic integrals of modulus p are related to those of modulus -1Vp/(1 + p) by the transformations
61:5:3
p1
KI1P)=(1+p)K(p) p)
2E(p) - 9'K(P) q= = 1 - p=
61:5:4 EI
\\1+P I +P
and to those of modulus (I - q)/(1 - q) b
61:6 EXPANSIONS
Alternatives to the power series
61:6:1 K(p)=-
it
I+
1
l p + (1 X 3 pa +
\1 X 3 x 5
p+
ry
r (2J- - 1),
pr
2 \2 x 4 2 x 4 x 6/ =2 (2j)!'
( (1 2p' 1x3l2p' Ix3x51p5 - 1 (2j-1)! 1
61:6:2
2 \, 1 2 x. 4, 3 2x4x6 5 2 ,--1[
_o2j (=i1P
P,J
J
may be constructed by use of transformations 61:5:2-61:5:5. Whereas expansions 61:6:1 and 61:6:2 are most useful
for moduli close to zero,/ the pair
r.3q_1
61:6:3 K(P)=1n(q)+Ilnlq)
\ Ix2 ][1q] 1x2 3x4 2x4
L ` , 2 9
1
J
2 2 2 (2j - 1)!! q) +...
+ In4- ----..._
qJ Ix2 3x4 (2j - 1)(2j) (2j)!! J
and
2 2][Iq]-+4[ln(4)
61:6:4 (l
F(P) = I+ If Lln\q)I x
3 l
q1 x2 3x 4][2 x 4
q=1
J-
2
2' 4
[ln\q/ _ 2
2
2 1 1)(2j)][(2i
2j - 1
(2j -
1)!
IX (2j - 3)(2j --) (21)!! q +.
\/
[2'j!]Fo
is more valuable as p 1.
A trigonometric expansion is provided by
Table 61.7.1
IT IF
Kip)
2 2U
U A
Up)
2 2 4U
Klq) 1 1 1
2 V3 V2 1 0
Kip) N2 V3 2
of the "nome" [Section 61:14). The constant U that appears in this table is the ubiquitous constant [Section 1:7),
equal to 0.8472130848.
This is the relationship that is exploited in the algorithm below. The complete elliptic integral of the second kind
is evaluated via the expression
K(p) F_ . _ 1 w. successive geometric (or arithmetic)
7
means of q and I
Replace e by ek/2
lfn=0goto(3)
Output 4
k = K(p)
Output e
e-E(p)<
Set it = k
Set e=p
Go toll)
131 Replace r: he exp(-ak/nl
Output k Test values:
K(1/2) = 1.68575035
Output e E(1/2) = 1.46746221
Kt\ ? 2r = 2.156515r_._
Output n E(\ 3/2! - 1211056(1;
of 1 "i = 0 01797_'3$7!,
The algorithm ceases to compute successive values of the geometric and arithmetic means when sufficient have
been calculated to ensure 24-bit precision in K(p) and E(p). If the green portions of the algorithm are included.
values of K(q), E(q) are also evaluated, as well as that of the nome [Section 61:141 N(p).
On account of the definitions 61:3:3 and 61:3:4, the universal hypergeometric algorithm of Section 18:14 may
also be used to compute elliptic integrals of the first and second kinds. As well, the algorithm in Section 62:8
generates values of K(p) and E(p).
61:9 APPROXIMATIONS
61:9:1 K(p) _ -
sr
2
/.1
For rather wide ranges of small values of the modulus, the approximations
16 - 5p
16-
J 8-bit precision 0 s p fs 0.67
it 16 - 7p2
61:9:2 E(p) -- 8-bit precision 0 s p :s 0.71
]6 - 3p2)
61:9:3 K(q) = I I + 4
]In( 4) - 4 8-bit precision 0 <p S 0.47
P
In(4)
61:9:4 E(q) - I + -4 8-bit precision 0 <p :5 0.35
P/
2
are valid, whereas for moduli close to unity we have
61:9:5 K(p)
5-8 pz
10 (1
- z- 1 -4 p
16 z
8-bit precision 0.88 5 p < 1
P
1 -P2 1 _ pz
61:9:6 E(p) _ 3 + pz In 8-bit precision 0.945p<I
4 4 16
117++9pz)
5pz
61:9:7 K(q) - T 8-bit precision 0.73 5 p 5 I
2\
(19+
61:9:8 E(q)
8-bit precision 0.65 S p :s
+73Pz)
2 3
61S THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E(p) 61:10
These results derive from expansions 61:6:1-61:6:4, from which more refined approximations we available. Be-
cause of their rapid convergence, the expansions in Section 61:14 give accurate approximations even when severely
truncated.
61:10:2
P-d E(p) _ P P
dp p
Indefinite integration of the complete elliptic integrals yields K = L = 3 hypergeometric functions [Section 18:141
ap
61:10:3
o
K(t)dt =
2
2-'u (Ot(i),(i)'
(I),(I)j!2),
p'; 3
E(1)d( = -
o 2 -, I1),(1);(3),P
fv
61:!0:6
(1 +p:)Elp) - (1 -pK(p)
tE(t)dt =
0 3
are simpler.
Some definite integrals lead to Catalan's constant [Section 1:71
1 K(tIdt _ 1 a2
61:10:9 K
V'1-t \Y'_I 4U:
61:10:10 E(t)dl
_/ I r a L'
E
o V ' I - t- V'j) 2 8U2 2
( 6.14 and 6.15].
and more are listed by Gradshteyn and Ryzhik [Sections
The operations of semidifferentiation and scmiintegration with respect to p-, when applied to complete elliptic
integrals, produce elementary functions. Examples include
dir- ('I- )
61 : 10 : 11 (dpz)i. E (p) =
p
61:11:1 K(ip) =
N P
1
'p
K
V'YP
p
and
61:11:2 E(ip) =V I + pr E( p
1 +p'
are valid for all real p. For example, K(i I = (l /\/2) K(l /\,2) and E(i) = V2 E(1 /V'2).
The complete elliptic integrals acquire complex values when their real moduli exceed unity. The transformation
formulas
61:11:3
p P p
F(P)Pfi\P,-Cp
61:11:4
P
permit the evaluation of the real and imaginary parts in this circumstance. They show, for example, that
V2 V2
i
[2E(
1
\2
1^ 1
AV2
II
=fl-i1-=
V2
Sri
2K(\ 2)
indicating that the real and imaginary parts are equal for a modular value of V2.
61:12 GENERALIZATIONS
Allowing the upper limit to vary in the integral definitions 61:3:1 and 61:3:2 gives rise to the incomplete elliptic
integrals of the next chapter.
In the remainder of this section we discuss the complete elliptic integral of the third kind f1(v;p) defined by
61:12:1 II(v;p) =
dt - f tZ dB
....................... 5
TT(-0. 91p) :
Tr(-(L 85op)
. rr<-20,p) .
-2
617
61:13 THE COMPLETE ELLIPTIC INTEGRALS K(p) AND E4 p) 618
sin(6)
tiv -p2<s'<0
P
E( V---;)
61:12:5 fI(-p2;p) = p), = v = -P
P 1+v
v n
61:12:6 fl(v;p) = K(p) + [K(p) - E(p)) F(q:O - K(p) E(q.d,)
(1 + v)(P2 ' v) {2
Vt--v
sin(k) _ I <v<-p'`
q
61:12:7 fl(- 1;P) = ± - V = -1
n
v
1
61:12:8 II(v;p) _ {E(p) F(p:d,) - K(p) E(p: )} sin(Or) = v < -1
(1 + v)(p + v)
where q = and F(p;d,) and E(p;d)) are incomplete elliptic integrals whose numerical values are calculable
using the algorithm in Section 62:8.
For characteristics of magnitudes less than unity, the expansion
I(
II(v;p)=1-2\v p2 3 vpg
P411
_5( v2p2 3vp' Sp!)
61:12:9 2 +8Iv 2 + 8/ 16 L 2+ 8 16J-
(2j - Il! (2k - ''
1)!!-p,
may provide useful values of the complete elliptic integral of the third kind.
o I - p2 sin'(e) P
61:13:2
f.n to's2(9)de
sin2(O) E(p) - g2K(P)
P2
1 - p2
.n
Sing(e) CAS (e)de 2 - P2 2
=
61:13:3 C(P) =
J [1 - p2 sin2(6)]3n i K(p) - P, E(p)
although not in this Atlas.
ezPaK(q)
61:14:1 N(p) = K(p) ) 4=Vt p zZ
Os ps l
and its utility arises from the large number of functions that can be expressed as infinite sums or products of
algebraic functions of N(p). Except in the immediate vicinity of p = 1, the nome is a small positive number (0
N(p) s ; for 0 5 p s 0.999995; N(p) s 1 for 0.999995 <- p <_ 11, so these infinite sums and products
generally converge very rapidly.
The functions that can be expressed in terms of the note include the elliptic modulus:
If
11 Sp(p) [ I I - N(p)) ] i - N'(p)J L 1 - N°(p),' ...
and second:
61.14:6
K(p) E(p:b) - xE(p) N(p) sin(arx/K(p)J - N-(p) sin[2n.r/K(p)1 + N'(p) sinj31rr/K(p)J
2tr I - N2(p) I - N4(p) 1 - N°(p)
where x = F(p:(b). the elliptic amplitude [Section 63:31:
am(p;x) ax N(p) N'(p) sin[2ax/K(p)J N3(p) sin[3.r/K(p)]
61:14:7
2 4K(p) I + ti=(p) + 2[1 + N4(p)] + 3[1 + N6(p)] +
all four of Neville's theta functions [Section 63:81:
61 : 14:8 H,ip:.r)
2'R VN( )
Isin
az
I - N-(p) sin(
3ax
2 ---- l =
N"(p) sin(._'K(pl
5-rr
- ...
p9K(p 2K(p)
24\ N(p) ar I, 31" Sax
61:14:9 6,P-..T) _ - cos + N'( p ) cost N'( P ) cos
17K(p) 2K1 p)) . 2K(p) 2Krp)
I ax 27tt Sax
-
Kip),K(p).
1
61:14:15 A I (I-p21S
2-r 211-p)
0sp<1
will serve, but otherwise one must resort to iterative or numerical methods: the extended algorithm of Section 61:8
provides one possibility. Some particular values of the nome are easily found by application of definition 61:14:1
to the values of K(q)/K(p) listed in Table 61.7.1.
CHAPTER
62
THE INCOMPLETE ELLIPTIC INTEGRALS
AND
The incomplete elliptic integrals of the first and second kinds are important bivariate functions. Together with more
elementary functions and the incomplete elliptic into teal of the third kind [see Section 62:12]. they may be used
to express any indefinite integral of the form fR(t,V p(t)dt (where p is a polynomial of degree 3 or 4 and R is a
rational function) [see Section 62:14].
62:1 NOTATION
These functions are sometimes named Legendre's elliptic integrals. The adjective "incomplete" is often omitted:
it refers to the fact that the upper limits in the defining integrals 62:3:1 and 62:3:2 are usually smaller than those
in the corresponding integrals 61:3:1 and 61:3:2 that define the 'complete" elliptic integrals. The symbols F and
E are in general use for the incomplete elliptic integrals of the first and second kinds, but there is no unanimity
on the ways in which the variables are specified.
This Atlas uses the notation F(p;¢) for the incomplete elliptic integral of the first kind and we describe d as
the amplitude of the function and p as its modulus (k is often used). Other symbolisms are based on the parameter
m = p` or the modular angle a = aresinl pl. so the notations F(4)lm), F(d)\a). as well as F(k..b), F(.b,k) and even
F(dt) are encountered. Since dt is often interpreted as an angle, it may be expressed in degrees, and occasionally
sin((O) is treated as the variable instead of 4 itself.
The situation is even more confused for the incomplete elliptic integral of the second kind for which the Atlas
uses the E(p:4)) symbol. In addition to analogs of the variants discussed in the last paragraph, one also encounters
E(x), where x represents the incomplete elliptic integral of the first kind, F(p:d,). As well, ambiguities can arise
because the same E symbol is adopted for both the incomplete and complete (Chapter 611 elliptic integrals of the
second kind.
62:2 BEHAVIOR
The incomplete elliptic integrals are sometimes discussed only for 0 s p < I and 0 s d' n a/2 but they may be
defined for all moduli and amplitudes. They are real and finite, for all real p and 6, except that F(I :(.) is infinite
when 41 a it/2. In fact. F(pab) and E(p:4b) may be real even when p is imaginary, as discussed in Section 62:11.
Figure 62-1 shows graphs of F(p:e) and f o r p = 0. 0.5. 0.7, 0.9 and 1.
621
62:2 THE INCOMPLETE ELLIPTIC INTEGRALS F(p:e) AND E(p;4)) 622
623 THE INCOMPLETE ELLIPTIC INTEGRALS F(p;4') AND E(p;4) 62:3
The incomplete elliptic integrals possess a high degree of symmetry. They are even with respect to p and odd
with respect to 46. The differences
240
62:2:1 F(p;(b) - K(p) and E(P;4)) - - E(P)
It a
are periodic (but not sinusoidal) with period it [see Chapter 36). Moreover, each of these difference functions
displays odd reflection symmetry about amplitude values of 0, ±ir/2, trr, -t31r/2, ..., as evident in equation
62:5:2 and from the examples mapped in Figure 62-2.
......................................................0.2
01
-0.2
62:3 DEFINITIONS
The indefinite integrals
de dt
f
qz = 1
62:3:1 F(p;dt) ==
= -P 2
62:3:2 E(p:e) =
6
1 - p' sin'(O)d0 = f
r IMmI I-pt dt =
_ 2 61
l+q -t
, }dt
=
Z
q=1-p
,
I V,
0 0 V -r U
1+ .
define the incomplete elliptic integrals of the first and second kinds. The trigonometric integrals are illustrated by
Figures 62-3 and 62-4. which also demonstrate the relationship to the complete elliptic integrals [Chapter 611.
The incomplete elliptic integrals may be expressed as inverse functions [Section 0:31 of the Jacobian elliptic
functions [Chapter 631. Thus, for example, if d, is the angle defined, in terms of variables p and x, by
04411.
4
le
...........................:...... 1/q
11-p2ein2(0)
IIIIIIIIII
111111113
FIG 62-3
At .......................................D
111111111 unuu
L11111u I MI"HILL11L1L1rn
1-pzein2 (6)
.........................:........ q
E (pt 0)
E ) FIG 62-4
U .........................:...... p
have coordinates (1,0) and (I/q,a/2), respectively, where q is the complementary modulus of F(p4). Now con-
struct an ellipse [Section 14:141 with 01 and OP as its semiminor and semimajor axes. In polar coordinates the
equation of this ellipse is r = I / l - p2 sin2(8). Let db be the amplitude and consider the average length f of the
radius vector from the origin to the ellipse over the segment 0 s 8 s $. Then the product Pd; is defined as the
incomplete elliptic integral F(p;$). That this definition is equivalent to 62:3:1 is evident by evaluating f as follows:
1 I d8 F(p;ib)
62:3:5 P= rd8 =
`Y 0 Jo 1 - p2 sin2(6) It,
A geometric definition of the second kind of incomplete elliptic integral is illustrated in Figure 61-2 of the
previous chapter. This shows a portion of the ellipse -q'V I --t 22 marked in green and delineated by the ordinates
i = 0 and r = sin(4). The length of this green arc defines the elliptic integral E(p:4)). The entire perimeter of the
ellipse is 4E(p). four times the complete elliptic integral of the second kind.
625 THE INCOMPLETE ELLIPTIC INTEGRALS F(p;4) AND E(p:¢) 62:4
See Section 63:7 for Ftp:(b) and Elp;,b) when 4 takes particular values. When the amplitude and modulus are
interrelated such that cotta) = V q = i - p=)"', we have
1
I
Cl
62:5 THE INCOMPLETE ELLIPTIC INTEGRALS F(p:A) AND E(p:4) 626
r
q -2
1 A
62:4:9 E(p; arccot(Vq)) = E(V -cot'(1)4)=ZII-q+E(p)J
62:5 INTRARELATIONSHIPS
The incomplete elliptic integrals are even with respect to their modulus but odd with respect to their amplitude:
62:5:1 fl-p:d,) = ftp:d,) = -f(p:-,b) f= ForE
The latter equality is a special case of the formula
/
62:5:2 flp:
171T
-dt2nf(p;)-f(p: nn' Td>} n=0,±1,±2,
for reflection of the amplitude across any multiple of r/2. In the equation above. and that below. fi p,,-U12) is the
complete elliptic integral K(p) or E(p). Both incomplete elliptic integrals obey the recurrence formula
Then the corresponding incomplete elliptic integrals of the first kind are interrelated by
2 (I - 1 po)
E(P-A,-1) [E(po:dro) + \ P F(po:db)I -
and
62:5:7 E(Po:4J 1 - Po F( Po: d+o I
1 + po 2
PO sin(40)IPo cos(4) + 1 - pu sin2(db)1
Going from elliptic integrals of variables po, dte to those with variables p-,, 4b_, is known as descending Landen
transformation, while the conversion from variables po. dro to P. dr, is called ascending Landen transformation.
The adjectives 'descending" and "ascending" recognize the fact that, for 0 < po < 1. the parameter P-, is smaller
than po, whereas p, is larger. Although it is not immediately apparent from 62:5:4, the numbers p_ : pr,: P, form
627 THE INCOMPLETE E t tv'FIC INTEGRALS F(p*) AND E(p;4) 62:7
a sequence, that is, the rule by which po is constructed from p_, is identical with the rule by which p, is formed
from po. Similarly, though more obscurely, (b-1; din; (01 form another sequence, as do F(p-,;4o-,); F(Po;400); F(pl;4iJ
and E(p-1;4_i); E(po;0o); E(p,;dij). It is evident that each of these sequences may be extended indefinitely in both
directions. In one popular method of computing numerical values of elliptic integrals, one or other of the Landen
transformations is implemented repeatedly, starting with po = p. 4o = di, and generating a sequence of moduli
(either po; p-1; p-2; .. .; P-, or PO; Pi; p2; p,) that satisfy the inequalities
.
62:5:8 P < pi_, < ... < P-2 < P-1 < Pa < P1 < P2 < ... < P.-I < P.
The limits as n - x of p-, and p are 0 and 1, respectively; hence, if the transformation is carried out a sufficient
number of times, it becomes possible to approximate and E(p..pbt.) by use of equations 62:4:1-62:4:3.
In this way F(p;(b) and E(p;di), as well as K(p) and E(p) are calculable. The "common mean' technique used in
Sections 61:8 and 62:8 is, in fact, an adaptation of the descending Landen transformation.
Incomplete elliptic integrals with moduli exceeding unity may be related to ones in the standard 0 < p < I
range by the transformations
p 1
62:5:9 FI 1:(b = PFI p;aresin{
P \ I. P /I
(1 I ( sin(di) 1-p sin(tb)
62:5:10 E[ P:40) = p EI p;aresin - P ) - p F p;aresin p p s
62:6 EXPANSIONS
For small values of the amplitude and modulus. there exist expansions of which
2K(p) [2KI p) - r4K(P) _ 2
62:6:1 F(P 4b) =
it
4) -
IT
1
1J sin($) cos((b) - IL
31T
3
P2 1
6 Jsin3( ) cos(Qr) -
2E(p) 2E(p) '' 4E(p) p2
62:6:2 E(p; (b) = di + I- 11
Jsin3(di) cos((b) +
it a I3 - 6
are the leading terms. Similarly, when p is close to unity:
3K(q 'K( ) sinl<b) 4K(q)
9 2 q2]
62:6:3 F(p; 4)) =it ! mvgd(di) - IT
-1
cos'(rb) 31T
-3- 6 cos'(di)
1 - cos((i)N,'l - p' sin'ldi) 2K(q) - 2E(q) [invgd(40 sin((i) 1 q2 sin(di)
+
62:6:4 E(p: di) =
sinl(b)
+
it cos'(di) J - 2 cos2(di) T
where invgd is the inverse gudermannian function [Section 33:14]. Gradshteyn and Ryzhik [Sections 8.117 and
8.1181 may be consulted for the general terms in these expansions.
Table 62.7.1
-nn
6= -= 6=- 6= 0 6'-nn6= x
9
h p: 6) -_ -nK(pi 0 nK(p)
FA p: 61 -_ -nE(p) 0 n6( p)
The algorithm of this section produces values of F(p:(b) and E(p:6) using an extension of the "common mean"
procedure that is described in Section 61:8. Because applications of incomplete elliptic integrals often require values
of the complete integrals as well, the algorithm also generates K(p) and E(p).
Ut A, and G. retain their significances from Section 61:8 and let a set of their values, for a given p. be generated
by recursions 61:8:1. Further, let a set of angles d,,, 4;, d>,.... be defined by the recurrence formula
with db = 4), the amplitude of the sought elliptic integrals. Then. in addition to the results K(p) = a/(2G,) and
62:8:2 E(P)=K(2)2-p'-Y2'(A;-C)-
LLLLL1
-1 J ro
which were given in the last chapter. we have
62:8:3 F(p; 2K(p)4,,/,rr = 4)x./G
and
E(p)F(p:d.) I
62:8:4 E(p;dr1 = + VA - G,2 sin(2'd,;) (A, - G,) sin(2'-'d),.
K(p)
These equations are the results of applying Landen's descending transformation an indefinite number of times. In
practice, the angles are calculated via the formula
((A, - C,) tan(2'd,;l
62:8:5 4),,4' - 1
17'
are-
A + G tan'(2'dt, l j
and converge toward d): as rapidly as A, and G, converge toward their common mean G.
Output k
k-K(p)< Test `values::
Replace e by re/(8g) KI//2\I = FI/2' 2`I = 1.68575036
Output e
e = E(p) <-
Set F = 4b/g /
EI 2 I = EI`2 2 /I = 1.46746221
Output F
F-F(P;46) «G< FI 2' 4 I =\0.80411366101
Replace E by (E/2) + 24e/a) -;
Output E El 4 I = 0.767 1 95 986
E-E(p;4) <GG< 2
The algorithm is designed to operate over the range IpI < 1 for all 4. The formulas of Section 62:5 may be
employed for moduli outside this range. For 1,01 s it/2, values of E(p;$) and F(p;4) generally are computed to
24-bit precision. However, F(1;tr/2) = 00, and when both input variables are very close to these critical values,
less accuracy may be produced. In this circumstance, transformations 62:5:11 and 62:5:12 may be applied with
advantage.
62:9 APPROXIMATIONS
Based on expansions 62:6:1 and 62:6:2. the approximations
P
62:9:1 F(p:4,) _ ( ) 124, - sin(21b)1 + - sin(26) jpj 0.59 ? :4,
2
Efp) I
62:9:2 E(p4.) = [24, - sin( 261] - 2 sin(26) pj = 0.72 > iii
n
are valid to 8-bit precision when both the modulus and amplitude fall within the specified range of small magnitudes.
As p approaches unity, the approximations
2K(q)
62:9:3 F(p:o) _ [ln{tan(4,) + sect(b)} - tan((b)sec(,b)) + tan(6) secOd.) p-. I
IT
and
and
62:10:3
a
- F(P;cb) _
E(p;d,)
pq
_ - F(p;4) - p sin(d)) cos(h)
aP P qV I - p' sm'(dt)
Indefinite integrals of incomplete elliptic integrals include
arcsin(p sin(4))
62:10:4 sin(g) F(p:B)dQ = - cos(y,) F(p:(b)
Jn P
arcsin(p sin(d,)) sin(do)
62:10:5 sin(B)E(p:B)dO = + V I - p stn (d,) - cosld,) E(p:d,)
r.
0 F(p:B)d0 _ F'(p:d,)
62:10:6 f
t( V1 - p` sin2(B) 2
0
E'(p:4)
62:10:7 I -p sm'(B) E(p:B)dO =
-'
See Gradshteyn and Rvzhik [Sections 5.12 and 6.11-6.13] for a few other indefinite integrals and many definite
integrals.
1 + p" tan(dr). However, for imaginary amplitude. the incomplete elliptic integrals are themselves
where tan(J.) =
imaginary and involve the gudermannian function [Section 33:14) as well as hyperbolic functions (Chapters 30 and
28):
62:11:3 F(p;id,) = IF(q;gd(d)) q=V
62:11:4 E(p;id,) = i [F(q;gd(d,)) - E(q; gd(d,)) + tanh(d,) I + p2 sinh'(d,)]
For a complete list of similar transformations, see Gradshteyn and Ryzhik [Section 8.1271.
62:12 GENERALIZATIONS
The incomplete elliptic integral of the third kind:
dO
62:12:1
o (I + v sin2(0)] 1 - p '" sin2(0)
is a generalization of the first kind because 17(0;p;4,) = F(p;d.) The notation is as confused as for the corresponding
complete integral [Section 61:121. Some other special cases include
631 THE INCOMPLETE ELLIPTIC INTEGRALS F(p;4) AND E(p;m) 62:13
1 I
62:12:2 HI(- I;p;(W = F(P4) - - E(p;40) + - tan(+) 1 - p2 sin=(+) q = V1 -p'
q q
I arctan{(1 ± p) tan(+)/ 1 - p' sio=(.6)}
62:12:3 f1(}p:p:4P) = 2 F(p:4) +
2 t
( arctan{ 1 + v ttm(*)}
-v 1-1
62:12:5 l(v;0;O) = 1' tan(4,) V = -1
artanh{ -v - 1 tan(+)}
v< -1
-v- 1
invgd(tb) - V; artaah{ V v sin(.0)}
v L. 0
I+v
62:12:6 II(v;1A )) _ v = -1
invgd(4) - V -v artanh{ V -v sin(4i)}
I 0 ? V * -1
I +v
and
Because it is trivariate, numerical values of this third kind of elliptic integral are not easy to calculate. For
small magnitudes of the characteristic, v, the expansion
(2j - 1)!! (-v)1 (2k - 1)!!
()5 +1 sine+i(4)
62:12:8 11(v;p:iO)
[ e - cos(d) 1-0(2 + 1)!!
1
]
may be useful. The double factorial ratio [see Section 2:131 occurs repeatedly in this expression. When jvt > 1,
62:12:8 does not converge, and one must resort to rather complicated computational methods that are described by
Mime-Thomson [see Abramowitz and Stegun, Section 17.7, but note that his n equals our -vj.
Here R; (t) is a rational function oft as discussed in Section 17:13, and its indefinite integral is easily found via
the partial fraction decomposition discussed in that section.
The first three integral types in 62:14:2 may be expressed as incomplete elliptic integrals of the first and/or
second and/or third kinds. We shall illustrate how this occurs, using the first member of 62:14:2, with p(t)
--
±p3(t) as our example.
As discussed in Section 17:7. the cubic function p3(1) = r' + ar' + bt + c will have one, two or three distinct
real zeros according to whether the discriminant D is positive, zero or negative. The D = 0 case is easily treated
because the cubic function must then be of the form p,(t) = (t - r,)(t - r)2, where r, and r: are the zeros, leading
f --
to
62:14:3
dt dt
=
2
arcoth
x-r,
V'P ) (t- r2) t- r,Vr;-r,
if x > r2 > r,, or to some equally elementary result otherwise.
If the cubic has a single (real) zero, r, then the sought integral involves the constants
1 2 - Or + a)h=
62:14:4 h' = and P, = I - 4, = 4
3r
and takes one of the following forms:
dt
62:14:5 J = hF(p:2 arccot(h))
P3(t)
r-x
62:14:6 hF(p;2
V p3lrJ
62:14:7 = hF(q;2
dt
x5r
62:14:8 hF(q;2 arccot(h V r - x))
-- V -p3(t)
62:14:9 h2 = 4 and
2 -q2=h2(r2-r,)r2-r,
r3-r, 4 r3-r,
Then the following integrals are valid:
633 THE INCOMPLETE ELLIPTIC INTEGRALS F(p;sb) AND E(p;+) 62:14
=
62:14:10
J ' = hF1 p, arrcosl
r3 sx
r = hF(p;aresin
62:14:11
x-r ,
62:14:12 = hF q;arccosl
ry - rZ
J
62:14:13
rj dt r
= hF`q,aresin-
(1 z ))
JJ
I q
62:14:14 at
= hFl p;arccosl p 11
J ,2 \ \ /
r, <xsr2
dt ( r
62:14:15 I = hFkp;aresinl I I
I Vpslt) ` r2-ri////
dt
JI-V-p(t)= hF`q;aresin`
62:14:17
r,-x
We refer the reader elsewhere [for example to Erddlyi. Oberhettinper. Magnus and Tricomi. Higher Tran-
scendental Functions. Volume 2, Section 13.5] for a discussion of fdt/V p4(t) and for treatments of the other forms
in 62:14:2.
CHAPTER
63
THE JACOBIAN ELLIPTIC FUNCTIONS
These 12 functions have several interesting properties. One is their ability to bridge the gap between circular func-
tions [Chapters 32-341 and hyperbolic functions [Chapters 28-301. Another, discussed in Section 63:11, is their
double periodicity. The close interrelationship of the twelve elliptic functions is detailed in Table 63.0.1.
The a terms in this table take the values + I or - 1 according to the value of the quantity
63:0:1 w = frac(4K(p))
63:1 NOTATION
Jacobi's name is not always attached to these functions. Alternatively, only the sn. en and do functions may be
associated with Jacobi, the other nine being attributed to Glaisher, who invented their notation. Individual elliptic
functions do not usually carry a name, although sinus amplitudinis has been applied to sn. cosinus amplitudinis to
cn and delta amplitudinis to dn.
The Jacobian elliptic functions are bivariate, but one frequently encounters notations, such as sn(x), that suggest
only a single variable. The second variable is then implied and is regarded as a constant. We shall avoid this
oversimplification in the Atlas and write, for example, sn(p;x) where p is the modulus and x the argument of the
function. Commonly k replaces p and a replaces x, as well, the order of citation is often reversed so that sn(u.k)
may be encountered. Rarely in is used for sc.
A number of supplementary univariate and bivariate functions arise in discussions of elliptic integrals. These
include the complementary modulus q = V l - p2, the complete elliptic integrals [Chapter 611 K(p) and K(q), the
elliptic amplitude
l - dn ' (p,r)
63:1:1 am(p;.r) = arcsin{sn(p;x)} = arccos{cn(p;x)} = aresin } _
p
63:2 THE JACOBIAN ELLIPTIC FUNCTIONS
Table 63.0.1
I- f- f- I- I- I- 1- r- r- -
a'lpll dlp.,I sll f.+ a? ,dIO.1 <af per: dY p,i anp:+! dm p.,r nst p:, ,cIp , dIDsI
aJ
V'I-c7' V 14 rr= a.: a,f a,y 7'-I \II-&f
a,7 f VI-f%
VI-f: Vf:_ 11
VIII VI.d1- VI-I° VI-p l' a1r atpf
V'I - 9-f"
o,f
77 VI-f: \l-/- f: -
V.I - v=fl
7-7
.alr... -
VI.r11 v:`!: \Y:-oL( r ! n,pl \/:-r \p"a')' a,p
1 V 7--;-r-- V'a: ": o,a V' y" ' Pr- o,yr a,Df \ 77-7 a.p
a_\ r I
aol p1I -
VI'df
VI-f°
vl- PTT
Vr+F v
\%oI yfI
V'r
f
r
ff
Vr=-f' f
VI=-r
J
VI:+*IP
U
VI+f=
mff.,i ,
I f I - fI V'1- '7°- 1
.,V <
V1.-n ldP 7 fI 7'P nrp a,r a,pr
Vl-a:r: VI T aCI vl 7-7
771 77- V.I f: Vfp% f f
dlp,l -
\'I. dfl 1111
and the incomplete elliptic integrals [Chapter 621 F(p;cb) and E(p;4). The square root \'I - p2 sin2(b) _
I - p2 sn2(p;x) occurs often in the theory of elliptic functions and is frequently abbreviated to A(p;cb) although
we use dn(p;x).
The 12 Jacobian elliptic functions may be classified into four groups, each with three members, according to
the second letter of the function's name. Thus, the cs, ds and ns functions are said to be copolar: they all possess
a pole of "type s."
We shall use ef(p;x), fe(p;x), ge(p;x), hg(p;x), etc., to represent arbitrary Jacobian elliptic functions, it being
understood that e, f, g and h are letters selected from the quartet s, c, d and n. Often we shall reserve e to represent
the pole type.
63:2 BEHAVIOR
The Jacobian elliptic functions display interesting properties when the modulus and/or the argument is imaginary
or complex. In this section, however, we restrict p and x to real values. Moreover, as in Table 63.0.1, we shall
Table 63.0.2
O s W G t W C t W G S W< I
Qi +1 +1
+1
a, *1 I -1 +1
637 THE JACOBIAN ELLIPTIC FUNCTIONS 63:3
consider p to lie in the range 0 5 p 5 1. though equations 63:5:2 and 63:5:14-63:5:16 show that this restriction
is not mandatory.
Except for some of the functions when p = 0 or 1, all Jacobian elliptic functions are periodic, their periods
being either 2K(p):
63:2:1 fe(p;2K(p) + x) = fe(p;x) fe = sc, cs, do or nd
or 4K(p):
63:2:2 fe(p;4K(p) + x) = fe(p;x) fe = sd, sn, cd, cn, ds, dc, ns or nc
This is supported by Figures 63-1 and 63-2, which together map all the elliptic functions forp = 0.9. This particular
value of the modulus was chosen to emphasize the nonsinusoidality of the sn, cn, ds, dc, do and nd functions. For
smaller positive values of p, these six functions become increasingly sinusoidal [see Section 32:1 1 in shape, while
the shapes of the other six Jacobian functions come to resemble those of the functions of Chapters 33 or 34. As
p - x, the periods increase indefinitely, and most of the functions degenerate into hyperbolic functions as explained
in Section 63:5.
The ranges of the sc and es functions are unrestricted, but the other 10 elliptic functions are constrained as
follows, for jpj s 1:
-1 1
63:3 DEFINITIONS
The Jacobian elliptic functions are defined via the inverse of the incomplete elliptic integral of the first kind [Chapter
621. Thus, if
63:3:1
then
x = F(p:rb) _
J., Ip dO
Jhe other I I elliptic functions can then be defined via the interrelationships reported in Table 63.0.1. For example:
63:3:3 cn(p;x) = 1 - sn'(p;x) = cos(4)) = cos(am(p:.r))
and
63:3:4 dn(p;x) = 1 p' ,n2(p;.r) = I - p' sin'ob) _ l - p' sin:i amt p:.r))
63:3 THE JACOBIAN ELLIPTIC FUNCTIONS 638
639 THE JACOBEAN ELLIPTIC FUNCTIONS 63:3
63:3 THE JACOBIAN ELLIPTIC FUNCTIONS 640
..:....:....:....:............Q.9
: F 1 6 6 3 - 3 ::....:....:..0.6
-/t
.:....:....: en(p tx)
v-x/K (p)
The remainder of this section is devoted to making a geometric construction that may be used to define all the
Jacobian elliptic functions. In Sections 29:3 and 33:3 the six hyperbolic and the six trigonometric functions are
defined in terms of the lengths of the sides of three similar right-angled triangles. In strict analogy- six of the
Jacobian functions may be defined as the nonunity lengths of the sides of the triangles OAC. OGI and OJL as
depicted in Figure 63-6. Here the angle d, equals the elliptic amplitude am(p:x). and those sides of the triangles
that are dashed have unity lengths.
Further construction is needed to provide definitions of the other six Jacobian elliptic functions. First. super-
impose the three triangles OAC, OGI and OJL, as shown in Figure 63-7. Then select the point K on line JL such
that length LK equals the complementary modulus q. Join points 0 and K by a straight line and denote its points
of intersection with lines AC and GI by B and H, respectively. Three more elliptic functions may now be defined
in terms of this new line, namely
63:3:5 length OB = dn(p;x) length OH = dc(p;x) length OK = ds(p;x)
Yet more construction is required to define the remaining three functions. Measure unity length from point 0
toward K, and so create point E. Next draw line DEF through E perpendicular to OL to cut lines OJ and OL at D
and F.
Then
components are illustrated in Figure 63-8. on which diagram all 16 lengths are identified. To better accentuate the
similarities, we have adopted the notation
63:3:7 nn(p;x) = dd(p;x) = cc(p;x) = ss(p;x) = 1
for those lines in the diagram that have unity length. This notation is consistent with Glaisher's.
Similarity and pythagorean rules applied to Figure 63-8 lead immediately to the relationships in Table 63.0.1 .
The dismemberment has segregated the elliptic functions into copolar groups. Note that the triangle of pole type
c is always larger than that of d, which, in turn, is necessarily larger than the triangle of pole n. The size of the
s triangle, however, need not be larger than that of c or d, though it must exceed n.
nd(l;x) cosh(x)
63:4:2 ns(l;x) = _ = coth(x)
Sd(l:x) Slnh(.C)
40
,a
ti
: 49O 4 Jga
b,
4*
o 0
t 1.0
FIG 63-4 :
cn(pIx) ;....;..0. 8
v-x/K (p)
......................
.............. :....:. 0. 6
......................
..
..:....:....: 0.4
i:....:..0.2
63:5 THE JACOBIAN EWYrIC FUNCTIONS 642
"O Oti A
sRe
0
.t
p'0
FIG 63-5 :
:.... ; .... ;
t in (pt x) ....:....:.. 0.8
:....4....X.\. 0.a
09
:....:...:...V V..\ .... :...:.... :.... .... .. :..0.4
63:5 INTRARELATIONSHIPS
Table 63.0.1 gives explicitly the relationship between any two Jacobian elliptic functions. If e, f and g represent
letters drawn from the quartet s, c, d and n, then
fe(p;x)
63:5:1 = fg(p;x) e, f, g = s, c, d, n
ge(p;x)
As in 63:3:7, ff(p;x) is interpreted as unity.
Table 63.2.1
p K(p)
o 1.571
0.8 1.995
0.9 2.281
0.95 2.590
0.99 3.357
0.999 4.496
0.9999 5.645
1 - 10' 11.401
643 THE JACOBIAN ELLIPTIC FUNCTIONS 63:5
9
63:5 THE JACOBIAN ELLIPTIC FUNCTIONS
Ce (Pt X)
and
Table 63.4.1
P=0 P - I
1 + dn( z)
Again, rule 63:5:1 may be used to extend these formulas to any fg2(p:x/2). Because of the interrelationships cited
in Section 63:0. all such formulas may be expressed in a vast number of alternative ways.
The so-called Jacobi real transformations:
I x
63:5:14 sn p;x =psn p;p
1 - N I -p. 2Vpn
63:5:17 P E = and p, =
I-vI-p 1-p
and
l I + P,,).,[,,
63 5 18
: : r_ I = ( 1 + ` I - po)x0 an d .r, _
then
63:5:19 dn(p_,:.r-,) =
1p- and dn(p,:x,) =
dn(pn:-r0) - V dn-(p,,:-rn) - I + po
(1 + v l - dn(po:.ro) I + Pn
Table 63.5.1
A popular technique for evaluating elliptic integrals of modulus p and argument x is to set p = p(, and x = x,, in
63:5:17 and 63:5:18 and then sequentially transform dn(p(,:x6) via 63:5:19 into dn(p_I:x_1), dn(p_2:s_2).._. (or
into dn(p,:x1), dn(p2:x2)....) until p" has become so close to zero (or p so close to unity i that the approximation
63:5:20 _a) = 1
2
(or
63:6 EXPANSIONS
Expansions of the Jacobian elliptic functions exist under three circumstances: as a power series in x. valid for small
.t: as a power series in p involving trigonometric coefficients, valid for small p: and as a power series in q involving
hypcrbolic coefficients, valid when p is close to unity. The leading terms in these three expansions are given in
Table 63.6.1. In addition to the elliptic functions. the table lists expansions of the elliptic amplitude am p:x). See
Section 61:14 for expansions in terms of the -norrte.-
Table 63.6.1
r-.0 P-0 P- 1
+ 0101x1 - - - cscha) -
1 6 360 4 u11(1) an(x) 4 slnhl..1
1+(1-pyre+(5- 6p' 4p')i +.. 1001x) - I+ q
+---
2 24 4 colt) COSIt) 2 00&m
x' (+ ) 1- 1001$)x) +
Isi41N2x0 + 2s) qt
2 24 4 -No B coth(I) 0046(.4 )
I(1+pi x+(7-22p'+lip) 0401x) +
[2. - mo(2o)] p'
coil(s) -
)sinh(2r) - 2.1 q'
t 6 360 8 tank) sing)
t2 (5 - 4p')4" ]2x - 1 (2t)] 0
2 24 80ota)0osa)
(2p' - p)t
+ P} +
2 24 4 cocko
p. r
[21 - 104)214] pt
8
6 120 8 coshlt;
647 THE JACOBIAN ELLIPTIC FUNCTIONS 63:8
I 1 -1 -1
1+q 1+q -I +q
0
777 0
.n V9 0 -Vq Zx Vq 0 -Vq
ca P:-t)
1
1 I+q
qq
0
0 -
-1
I+q
I Qq
-I
-I
Vl
II-1
q
0
0
r_
V I + q
I
4
q
-i -\ 1- i q
_
x
V
`'t
9
q
V f I I
nd( p.. )
Vq q V'q Vq q Vq
2K(P) IrK(P)
/o(o. trK(p)) =
I
0a(P:K(p) - x)
Here, as usual. K(p) and K(q) denote the complete elliptic integrals (Chapter 6l1 of the modulus and complementary
modulus, respectively.
63:8 THE JACOBLAN ELLIPTIC FUNCTIONS 648
Every one of the 12 Jacobian elliptic functions is expressible as the quotient of two distinct Neville theta
functions, the literal subscripts serving as a mnemonic to identify the numerator and denominator of the quotient:
thus:
N,( p:x) p:x)
63:8:5 cs(p:x) = nc(p:x) = etc.
O,i. p:x) OJ p:x)
By exploiting this principle, the algorithm becomes compact and rapid.
A shortened version of the algorithm in Section 61:8 is first used to calculate r/[2K( p)] and the nome [see
Section 61:14] N(p). The user then supplies the argument x, followed by a two-digit code. The algorithm decodes
the latter to identify the denominatorial Neville function then the numeratorial Neville function. If N, is called for.
the algorithm computes \ qK(p)/2r 8,(p:x) via equation 61:14:8. Similarly. if 0 is needed. \ qK( p)/2r 0,(p:x)
is evaluated by use of 61:14:11. The second equality in 63:8:2 or 63:8:4 is utilized whenever Ii, or 8,, is required.
p)/2r/2r 0 values are divided to produce the sought Jacobian function.
Finally. the two VqK(
The series 61:14:8 and 61:14:11 are used truncated to only the terms actually listed in Section 61:14 so that
the largest neglected term is of order N°7i p). Nevertheless, these series converge so rapidly that 24-bit precision
is virtually assured provided p does not exceed 0.99.
Because intermediate numbers are stored. one need on]% input the new argument .t and the new code to calculate
a second Jacobian function of unchanged modulus.
63:9 APPROXIMATIONS
The expansions in Section 63:6 provide raw material from which many useful approximations may be constructed.
For example:
.r'
63:9:1 cn(p:x) = I - - 8-bit precision Ixi <
L 32 - 3128p' J'/
63:9:2 ds(p;x) = csc(x) 8-bit precision ipi <
8 1 - x cot(x)
1
63:9:3 nd(p;x) = cosh(x) 8-bit precision Iqi <
8 sinh2(x) + x tanh(x)
where a is independent of x. Values of a depend on fe and are included in Table 63.10.1. The derivative with
respect to the modulus similarly involves a and the two copolar cohorts but has an additional factor equal to lxq'
- 3 sn(p:x) cd(p:x) - E(p:(WJ/pq2. where 41 = am(p;x) and 0 is listed in Table 63.10.1. Thus, the derivative
with respect to p of a Jacobian function of pole type c is
[xq- - 3 sn(p:.t) cd(p.x) - E(pxbiJ
63:10:2 - fe(p:x) = a get p:-r) he( p:.r
aP pq-
Integrals of the types
exist and are in Table 63.10.1. For functions having type s poles, however. these integrals diverge, and it is the
complementary alternatives
rKipi
Kip,
63:10:4 1, = fs(p: tklt and L=I fs p: t)d( f= c. d or n
J
that arc listed in the table. The tabulated integrals are valid for 0 < p < " . 1 and 0 < x < Ki pl but not necessarilc
for variables outside these ranges.
Rule 63:10:1 is the key to the evaluation of a great many indefinite integrals. Thus. we have
Table 63.10.1
a p
I E(pA6)
cd -q' 0 - In(ndl P: x) + psd( p:x)) - + an(p: x) cd(p. r l -
P p. P-
q'x
cn - artxas{dn(p.x)}
P P P:
do 0 In(sc(p:xl - nc(p:x)) x- sn(p;x, dc(p:ti
do -p- do-(P;x) aresin{sn(p:x)} = d Elp.d)
E(p:4) p'sn(p;x)cd(p:xl
nd P. &1Ptl - arccas{cdf p;xl)
q 4 9
I dclp:x) - q nop:xl
fC 1 p- - In - Isn(p: c) dc( p:.r) - EI p Am )1
q I - q q"
E(pAAI .t sn(p:xl dcl p:xl
sd p-dc=( p:x) -a sin{ pq nd(p:x) - pq cdl p:.r)
pq P'q' P-
1 ln' dn(p:x) - p Cal p:x) x - E(p:A)
1
p I-p
-4
Cs In E(Ph) - cn(p:x) ds(p:x) - E(p)
ns(p.x) - ds(p,x !
I
ds In(ns(p;x) q'Klp) - E(p) Elp;d,l - qr + cn(p:.t) ds(p:x)
-cs(p:x))
9
ns In\ds(p:x) Kip) - E(p( - - E(p,tt - cn(p:.n 64 p:.0
- cs(p:x)
permit evaluation of these integrals via 63:10:5 or 63:10:6. Similarly, such transformations as
gh(p:t)dr ge(p:t)he(p:t) _ I dfe(p;t)
63:10:8 gf(p;r) he(p;r)dr =
f
J lh(p;t) eh(p;t) J fe(P;1) a J fe(p;t)
lead to results involving the logarithm (1/a) In{fe(p;x)} and find many applications. In the formulas of this para-
graph, the a constant is that appropriate to the fe(p;x) function.
For the sake of completeness we include the following derivatives of functions commonly associated with the
Jacobian elliptic functions:
a a+ dn(p;x)
63:10:10 am(p;x) = = Ig2x + p 2 sn(p;x) cd(p;x) - E(p;4t))
()P aP P42
a l
63:10:11 dn2(p;x)
ax
(a p cn2(p;x)
63:10:12 ( dp E(P;4) 1 = 42 lsc(p;x) dn(p;x) - E(p;d) - q 'x sc' (p;x)1
\ /// x
651 THE JACOBIAN ELLIPTIC FUNCTIONS 63:12
The transformations 63:11:4-63:11:7 illustrate that elliptic functions have an imaginary period as well as the
real period discussed in Section 63:2. For example, one finds sc(p;x + iv + 4iK(q)) = sc(p;x + iy). Table 63.11.1
summarizes the periods for each of the 12 Jacobian functions and explains why these functions are known as doubly
periodic functions.
We have been discussing imaginary values of the argument. but one can also have an imaginary modulus.
Jacobian elliptic functions of imaginary modulus are, in fact. real. The appropriate transformations are
63:12 GENERALIZATIONS
There are none.
Table 63.11.1
Real Imaginary
period period
icl p:.r + iv). csi p: x - iv I. dn(p: c + ty 1. ndf p:x + iv) 2KI p l 4i K(j)
sd(p;x + ii, cnlp;x + W. ds(p;i - iv), nc(p:x + w) 4Kip) 4iKIq)
snl p:x + n:1). cd(p:.r + tv1. dc(p:.r rv). nsA p:x iv) 4Kipi IKigI
63:13 THE JACOBIAN ELLIP71C FUNCTIONS 652
63:13:1 p= q e,-e2+e;=0
e, - e, e, e3
Likewise, the role of determining the two (real and imaginary) periods is taken over by two new variables
2K(p) 2iK(q)
63:13:2 w, =
N /e, - e. V e, - e_
The principal Weierstrdssian elliptic function. usually symbolized by P(:) with some fancy typographical renderin
of the P, is then expressible as
63:13:3 e, + (e, - e3) cs2(p;:Ve, - e3) = e: + (e, - e3) ds'(p:zVe, - e,) = e3 + (e, - e3) sn(p;:Ve, - e3)
in terms of Jacobian elliptic functions.
Both the Weicrstrass and Jacobian elliptic functions are intimately related to the theta functions [Section 27:131
but, beyond reporting the relationships given in Section 63:8, we do not explore these connections in the Atlas.
Further to complicate a topic already overburdened by redundant symbolism, we should mention Jacobi's eta
functions and Jacobi's theta functions, which are linked to ordinary theta functions and to Neville's theta functions
[Section 63:81 by the equivalences
z K(q))
63:13:6 ei(P;x) = 63 ei(P;0)Od(p:x)
2K(p) 1TK(p)/
Closely related to the functions of Chapters 3 and 44, the Hurwitz function plays an invaluable role in the differ-
integration of periodic functions, a topic discussed in Section 64:14. Along with the related bivariate eta function
[Section 64:13] and Lerch's function [Section 64:121, the Hurwitz function provides a means of summing many
series whose terms involve arbitrary powers.
64:1 NOTATION
The symbol C followed by two parenthesized variables is standard, but the name generalised seta function is often
encountered. Sometimes {(v;u) is called "Riemann's function" or "Riemann's zeta function" but we eschew these
names to avoid confusion with the ;(v) function of Chapter 3.
We shall refer to v and u as the order and parameter, respectively, of the Hurwitz function. It is to achieve
unity with the notation for Lerch's function [Section 64:121 that we resist the temptation to replace the symbol u
and name "parameter" by the symbol x and the name "argument."
64:2 BEHAVIOR
Unless the order is an integer, the Hurwitz function becomes complex when u < 0. Accordingly, this range of
parameter is excluded from our discussion, except in Section 64:11. The zero order function C(0,u) bccomcs ill
defined as a approaches zero; similarly, ;(l .u) is ill defined as u approaches infinity.
The Hurwitz function displays an infinite discontinuity at v = 1. and, as Figure 64-I shows, the behaviors for
v > I and Y < I are quite distinct.
For v > I, 4(v:u) is uniformly positive. When both the order and the parameter exceed unity, the Hurwitz
function declines in value, asymptotically approaching zero, as either v or u increases. Conversely, l,(v;u) - x as
v approaches unity from positive values or as u approaches zero from positive values.
For the most part, the Hurwitz function assumes negative values for v < 1; however, there are "lobes" of
positivity as mapped on the contour diagram. As a function of u. {(v;u) displays a small, finite number of zeros,
mostly in the 0 < u < I range, for v < 1.
653
64:2 THE HURWTTZ FUNCTION {(v;w) 654
655 THE HURWITZ FUNCTION ;(v;u) 64:4
64:3 DEFINITIONS
The Hurwitz function is defined by the integral transforms
I t'-' exp(-ut)dt _ I In'-'(t)dt
64:3:1 Yv;u) = v>l u>0
r(v) Jo 1 - exp(-t) ['(v) tit - 1)
or by Hermite's integral
u- u'-
64:3:2 (v;u)=-++2
v-1
2 o
sin(v arctan(t/u))dt
(u' + t2)'/2[exp(2irt) - 11
u>0
Expansion 64:6:1 provides a definition of;(v;u) for v > 1, and Hurwitz's series 64:6:2, supplemented by recursion
64:5:1, can fill a similar role for v < 0.
Notice that none of these definitions is valid in the important domain 0 < v < 1. Contour integration [see
Erdelyi, Magnus, Oberhettinger and Tricomi, Higher Transcendental Functions, Volume 1, pages 25 and 261 can
define the Hurwitz function in this region. More practically, the series 64:6:4 can be employed for all orders and
parameters.
The difference i(u) - t,(u - t) of two digamma functions [Chapter 441 provides a generating function:
64:4:3 I(0;u)=--u
-1 +6u-6u2
64:4:4 >;(-l:u) =
12
-u+3u2-2u3
64:4:5 (-2;u) = 6
Although the Hurwitz function is infinite for v = I, the quotient;(v;u)/r(1 - v) and the product [v - 1]i;(v;u)
both remain well behaved in the vicinity of v = 1. This is evident from the limiting expressions
VV* = _1
64:4:6 lim
-I r(1-v)
64:5
THE HURWITZ'FUNCTION Vv;u)
656
64:4.7
1 I_
v J
Vu)
where 4' is the digamma function (Chapter 44).
64:5 INTRARELATIONSHIPS
The recursion formula
64:5:1
1'(v;u + 1) = >;(v;u) -U.- u Z. 0
may be iterated to
64:5:2
I,(v;u + J) + Z (j + u)-` u a 0
J=O
and, provided v > I, becomes expansion 64:6:I when J = x. The duplication
formula
64:5:3
Uv,2u) = 2 "LZ(v;u) + rl v; + u1
may be generalized to 2
64:5:5 C('; 3
+ u) = 2"1'(v:2u) - V; I + ul
4
\\\
64:5:6
(2;u) + C(3u) + 4(4;u) + ... _
(n;u) = 1 u > I
U
64:5:7
i'(2;u) - 4(3;u) + Z(4;u) - ... _ (- l)%(n;u) _ u>0
.=2 u
_ + - +
Z(2;u) ,(4;u)
64:5:8
2 3 4 + 2
.-I
4(n:u)
n
=40)-In( u-1) u>I
and
64 : 5 : 9
4(2;u)
2
1'(3;u)
3 +C(4;u)
4 _ ... = L, (-1)" In(u) - tr(u) u>0
,=2 n
may be combined in several ways: for example, to sum the series I{(k;u) where k is
restricted to a single party
64:6 EXPANSIONS
The important series
64:6:1 I I I
k` (I + U), (2 + u)` po
657 THE HURWITZ FUNCTION Ov;u) 64:7
often serves as a definition of the Hurwitz function for orders exceeding unity. Hurwit_ formula:
/ av
sin ( 2jau + 2/ 1
64:6:3 u( v+l;u+l ) I
u U. 2u""
v
+
v(v + I)
l2u;;'
- v(v + I)(v + 2Xv + 3)
720u`+4
(v+
F.-
kBk
k!uk'"
u-.x
where (v)k denotes a Pochhammer polynomial [Chapter 18] and B, is the k'" Bernoulli number [Chapter 41. Odd
members of this latter family, except B i, are zero, and even members, except Bo, may be expressed as zeta numbers
[Chapter 31. Thus also making use of 64:5:2, we may reformulate 64:4:3 as
K
1 + l + (J + U)]-. I (v I) A(2k)
64:6:4 I-2Z
i-0 2(J ; L k_[ [-4az(u + J)z]k J
where the first sum is "empty" if J - 0 (such sums are to be interpreted as zero). While this expansion technically
remains asymptotic (i.e., valid only in the u + J -* x limit), it may be made arbitrarily exact by making J large
enough and choosing K at least to exceed (I - v)/2. Yet a further development leads to
2(2x)"-'4* + ui . r3
64:6:5
1
-2 (v - l)-[C(2k) - I]
]J
2u" V- I v - 1
zsk
where (* is related to the Boehmer integrals of Section 39:12 by
64:6:6 c* = cos(21ru) S(2au;2 - v) - sin(2au) C(21tu;2 - v)
1ka
(2au)k sin 2 - 2au
// Ira \\
=r(2-v)sinl-+2au1 -(2au)'"G
k!(k + 2 - v)
v*2.3,4,.,.
\\ 2 1 k=o
64:7:1 Y,(v;0) _
Ix v>0
(v) .<O
64:7:2 ;(V; l,(v)
and
Similarly, the Hurwitz function of a parameter that is an odd multiple of a moiety is related to the lambda
function of Chapter 3. The simplest case is
64:7:6 Ov;Z) = 2"a(v) = (2. -
a(v)1
64:7:8 Yv; i) = LL
2
re(v) - (3(v)]
64:7:9
Adding equations 64:7:8, 64:7:6. 64:7:9 and 64:7:2 leads to Yv;',) + Y OZ) + {(v;,) + 4(v;l) = 4`I(v), which is
the J = 4 case of the general result
64:7:10v;j1J";(v) J=1.2,3,...
As u increases, the following values are approached
0 v>l
64:7:11 {(v;x) = -x v<1
The algorithm returns 1090, instead of infinity, if v = I or if v a 0 = u. Otherwise. the algorithm generally
has a precision of 24 bits although this may be degraded near the zeros of the Hurwitz function or at extreme values
(say Ivj > 30) of the order.
64:9 APPROXIMATIONS
A useful approximation to the J = 0 version of expansion 64:6:4 is provided by its first three terms:
v0
I 2u I (v2 - I)(v2 + 2v)
64:9:1 f,(v;u) 8-bit precision u> I 3
-u" [v - 1 + I + 6u]
Bear in mind, however, that the expansion is asymptotic and is not itself always accurate.
64:10:10 ftrt =
u(v - l;u) -(v - 1)
I-v
(v - 2u) -(v - 2)
(1 - v)(2 - v)
v<2
and a similar operation generates the Boehmer cosine integral from cos(2nt) ;(v;t). These results relate closely to
the discussion in Section 64:14.
=;(v;m-u)+(cos(m)+isin(vw))Y I M- I <u<m
-0 (u -J) `
64:12 GENERALIZATIONS
In this section we briefly discuss the trivariate Lerch's function 4(.r;v;u), of which the Hurwitz function is the
special case of unity argument, x = l:
661 THE HURWITZ FUNCTION ;(v;u) 64:12
64:12:2 4)(.r;V;u) =
I r" exp(-ut) dt - x
1'(v) jo I - x exp(-r) - (j + u)°
although these are generally real and convergent only for limited ranges of x. v and u. A third definition:
d-" exp(ux)
64:12;3 = exp(ux) (Nexp(x):v:u} x<0
[d(t + x)] '° I - exp(x)
utilizes the concept of differintegration [Section 0:101 with a lower limit of -X.
The recurrence formula
64:12:4 '(x;v:u) + U) +
can clearly be iterated indefinitely often. When the order is a positive or negative integer, Lerch's function can be
expanded in terms of the logarithm of its argument: thus:
-1
64:12:14 4)(x;2;1) diln(l - x) [Section 25:131
x
64:13 THE HURWITZ FUNCTION Uv;u) 662
-1
64:12:15 40m1) - x) [Section 25:13]
x
64:12:16
l
(x;v;)2 =
1
x) -
-2"
poln,(1 - V.) [Section 25:131
64:12:17 ( -x;- I ;1 =
1
rsf(ln(x)) [Section 30:10)
c(-; 2U)
64:13:4 +1(v;u) = &;u)
2-1
Special cases of the bivariate eta function include the eta function of Chapter 3:
64:13:5 I(v;0) =
{-(v) v<0
00 v>0
64:13:6 Y1(v;1) = 11(V)
64:13:7 T1(v;2) = I - TI(v)
the beta function of the same chapter:
-(-1)"
64:13:10 rl(v;u) = G")(u) n= 1,2,3,...
2(n - I)!
The very simple algorithm
Input v >>
Input u >> b»» Storage needed:
v. u, j and f
Set j = 2 Int`
r10u+2
2 Input restrictions:
u > 0; v must be large and positive.
Set f=-(u+j)"
1
is based on definition 64:13:1 and uses the properties of alternating series [Section 0:61 to ensure an absolute
accuracy of 6 x l0-8. Although the algorithm is valid for all positive v and u, it is tediously slow unless v is large.
For orders that are not large and positive one may use identity 64:13:3 or 64:13:4 and the algorithm in Section
64:8.
64:14:5
d'/:(_ I
[(IZ: fracl P/ I- -; frac( - +
P 2/ / J
and Weyl semiintegral
d-/x(-I)t4
64:14:6
[d(x + x)]-'/2 V Ir
[C(-
2
fracI-ll-;I-
\P
1:fracI-+l
2 P 2)
of the square wave function that is mapped in Figure 36-3 and formulated in Section 36:14. The shapes of these
last two periodic functions are displayed in Figure 64-2, together with the square wave itself.
yQ1ti ydQ
yQ
b4
APPENDIX
A
UTILITY ALGORITHMS
Most of the algorithms in the chapters of this Atlas are devoted to generating numerical values of particular func-
tions. However, a few of the algorithms scattered throughout the chapters perform operations that are more general
than evaluating specific functions. This appendix provides a glossary of the latter kind of algorithm and includes
a few additional instances of such general-purpose algorithms.
Section 9:14 contains two algorithms designed to convert numbers from one base to another. The first converts a
decimal number to any other base (to binary or hexadecimal, for example). The second algorithm performs the
reverse conversion: it converts from any base into decimal.
The two algorithms may be conjoined to permit conversion from any number base to any other.
Sometimes one needs to replace a decimal number v by the quotient n/d of two integers. Even if v is ostensibly
a rational number, rounding errors in its decimal representation will usually prevent the equality v = n/d from
being exact. Instead, there is a fractional error
v - (n/d)
A:2:1
V
"S
A:2 UTILITY ALGORITHMS 666
The commands shown in green perform the interchanges N .= n and D 1~ d; with many devices there are simpler
ways of achieving this.
The algorithm first finds the two integers between which v lies so that this number is known to be bracketed
by the fractions
n N
A:2:2 5v5
d D
with d = D = 1. The ratio
N - vD
A:2:3 r=
vd - n
is then calculated. If this exceeds unity, n/d is accepted as a valid approximant to v. and the numerator n and
denominator d of the (proper or improper) fraction n/d are output, along with a rounded value of the fractional
error e. Moreover, if r lies between the integers m and m + 1, then it follows from A:2:3 that
(m+1)n+N 1)d+D<v:5 mn+N
A:2:4 I
(m+ md+D
667 UTILITY ALGORITHMS A:4
On the other hand, if r 5 1, then N/D is accepted as the approximant and is output, followed by the corresponding
fractional error in the approximation. From the definition of r, one can show that
n+mN n + (m + I )N
A:2:5 d+mDSV<d+(m+ m-r<m+
1
1)D
in the r <_ I case. Thus, either A:2:4 or A:2:5 provides a narrower bracketing of v than did the original A:2:2.
The algorithm continues this process of bracket narrowing until the computing device interprets r as either infinity
or zero.
An algorithm in Section 40: 14 generates a succession of J numbers that are pseudorandomly distributed according
to the "normal" distribution [see Section 27:14]. The mean and standard deviation of the distribution are selected
by the user.
A subset of this same algorithm, namely
generates a set of pseudorandom decimal numbers uniformly distributed on the interval r. < x s .r,. The arbitrary
"seed" is taken as J, the number of sought outputs.
The Atlas contains three algorithms that are designed to process large data sets.
In Section 7:14 we present an algorithm for the linear regression analysis of equally spaced and equally weighted
data. This procedure gives the slope b and intercept c of the best straight line fit to data points (x1,f1), (x2,f2),
(x,, f,)...., (x,,f) where x, - x,_, = x,_2 = . = x: - x, and 2 S n < x. Optionally, the algorithm
also generates the correlation coefficient of the fit and the standard errors of the slope and intercept. Although
algorithms are not included, the same section discusses the linear regression analysis of data that are not equally
spaced and not equally weighted.
A straight line may be considered a first degree polynomial function. In Section 17:14 we show how to fit
pK(..r), a polynomial of any degree K, to a set of J + I data pairs, (x,,, fo), (x,, f,), ..., (x;, f)..... (x,, f,), provided
that J ? K. This polynomial fitting algorithm is also restricted to data that are equally spaced and equally weighted.
Several popular methods of processing data involve Fourier transformation, Fourier inversion, or both. Al-
though this Atlas does not discuss such data-processing methods, it does include, in Section 32:14, an algorithm
for fast fourier transformation/inversion. Once again, the algorithm assumes equally spaced and equally weighted
data, moreover, the number of data must be a power of two.
A:5 UTILITY ALGORITHMS 668
A:5:1
/-
estimate. That algorithm uses Newton's method which utilizes the fact that
j(r)
dx
(r,)
is usually a better approximation to the local zero of f(x) than is r,. Newton's method may be used to find the zeros
[but not double zeros-see Section 0:7] of any continuous function (or discontinuous function away from its dis-
continuities) provided that the initial estimate r5 is close enough to the sought zero.
However. Newton's method is sometimes inconvenient in that it requires numerical values of two functions:
f(x) and its derivative. We present below an algorithm that needs values of f(x) only. It is based on approximating
the derivative df(r,)/dx in equation A:5:1 by the difference quotient [f(r,) - f(r,_,)]/(r, - r;_,] and employs the
resulting formula
r_,f(r,) - rf(r,_,)
A:5:2
f(r) - f(r,_,)
The algorithm needs two (different) estimates rp and r, of the zero, but these can often be wild guesses and r, need
not be the closer of the two. In a kindred method [known as regula jalsi; see Bronshtein and Semendyayev. page
1701, the values of r and r, must lie on either side of the zero, but this is not necessary here. Double zeros are
accessible by this method. though convergence is slower than for single zeros.
The algorithm repeatedly outputs the improving values to. r, r., r, ... until two successive values of f(r) are
identical or until the user intervenes.
The segment shown in green must be provided by the user; it could, for example, be one of the algorithms from
Section 8 of the chapter dealing with f(x).
Our algorithm has the advantage of simplicity, but it will seldom be optimal and it may occasionally fail. There
exist very many alternative procedures for locating zeros (or root finding as the process is often called); Ruckdeschel
[Volume I]. Chapter 61 discusses some of these. The same reference [Chapter 7] gives methods for finding the
complex :giros of functions.
The algorithm is easily modified to provide information about features other than the values of the zeros. Thus.
to find the value of x at which f(x) equals, say. it. one need only replace the green segment by a routine to calculate
f(x) - IT. Again, to find the extrema (local maxima or minima) of the function, one simply changes the green
segment to "Routine to calculate j = df(x)/dx from x."
669 UTILITY ALGORITHMS A:7
Such a series may always be reformulated as a concatenation [see equation 0:6:5] or as a continued fraction [via
equation 0:6:12]. These reformulations frequently permit the numerical evaluation of f(x), for specific arguments
of interest, to be carried out more expeditiously, or more accurately, than by the original series A:6:1. As well,
there are a couple of more profound ways in which power series may be manipulated as an aid to the numerical
evaluation of f(x).
If interest is confined to a small range xo <_ x s x, of arguments and if a restricted accuracy, If(x) - [(x)l t
e, is acceptable, then the economized polynamial
A:6:2
J-0
-nay serve as a convenient means of calculating numerous approximate values of the f(x) function. Such economized
iolynomials are the basis of many computer library subroutines for functional evaluation [see Ruckdeschel]. Be-
:ause Chebyshev polynomials are utilized in the procedure, this Atlas's algorithm for calculating values of the
Jegree m and the coefficients eo, e,, e:, ..., e of the economized polynomial will be found in Section 22:14. Note
hat coefficients generated by the Chebyshev procedure are "best" in the sense of minimizing the worst error
)etween f(x) and f(x) rather than in the "least squares" sense.
If the series A:6:1 is asymptotic [Section 0:6], accurate values of f(x) can be calculated by its use only for
ufficiently small values of the argument x. However, by using the procedure detailed in Section 17:13, it is possible
o convert the sequence (ao); (a0 + a,x): (a0 + a,x + a_x); ...; (a0 + a,x + + ajx1); ... of truncated power
cries into a sequence R$(x); RI(x); R=(x); ...; Rj(x); ... of diagonal Pads' approximants. The latter sequence often
tas an advantage over the former in converging to f(.r) when A:6: I is asymptotic or even divergent. An algorithm
n Section 17:13 exploits this advantage in evaluating numerical values of functions from their asymptotic series
is the Padt operation.
he derivative of any function f(x) discussed in this Atlas can always, and its integral can usually, be expressed
terms of some other function(s) of the Atlas. Such expressions will be found in Section 10 of the relevant chapter.
cases of difficulty, the properties of hypergeometric functions [Section 18:141 or Laplace transformation [Section
i:14] can sometimes permit the operations of the calculus to be applied to otherwise intractable functions. Never-
:eless, there will remain certain functions whose integrals are evaluable only by numerical methods. Such an
aluation is known as a numerical quadrature and we here present a simple algorithm for the numerical quadrature
F a function f:
7:1 f(t)dt x0 * ±x * x,
twcen two finite limits. We treat only the case in which f(x) is known (i.e., is calculable or infinite) at all points
the xo s x <- x, range. Not considered is the technically important situation in which the integral is sought of
function whose values are known at a finite set of (usually equally spaced) arguments only.
The algorithm utilizes the trapezoidal approximation
xo
7:2 I f(t)dt ° 1, = x1 f1 xo + L (x, - .to)) J = 1. 3. 5..... 2n - I
h n given the successive values 9, 27, 81, .... The advantage of this threefold increase in the number of
A:7 UTILITY ALGORITHMS 670
summands in the approximation 1, to the integral is that all the summands that are needed as components of I. will
be reused in forming the 13,,. 19,,, 1.,,, ... sums. In the algorithm that follows, the green portion represents the
commands necessary to calculate values of the function f(x) from a value of x:
a=J f b
f(r)dr ««
1
Test values:
with f(x) = I/x: inputs: x9 = 1, x, = 2;
(5) Replace r by r + 1 successive outputs:
(6) Set I = s 0.693144600,
Replace n by 3n 0.693147174.
Go to (1) 0.693147181....
The algorithm does not output 1, values as such, but instead makes use of two levels of Richardson extrapolation
[see Ruckdeschel. Volume Il, page 3171 and returns values of (7291 - 901,,,3 + 1,,,9)/640. For suitable functions
f. this double extrapolation compensates so effectively for the finiteness of n that an accurate approximation to the
integral is soon achieved. The "test values" illustrate such a case: the fourth output (corresponding to n = 243)
already has converged to a value that reproduces the correct value of the integral
21
A:7:3 - dt = ln(2) = 0.693 1 47 1 806
JI t
A:7:4
f sin(tii)
r dt=JO
1" dt r dt
t I=2a_ I
Vt - sin(V)
r dt
into analytically integrable functions and ones that are everywhere finite over the xo <- x <- x, interval. A similar
strategy often can aid the numerical quadrature of functions that encounter an infinity within the x9 < x < x,
671 UTILITY ALGORITHMS A:7
range, as exemplified by
rs
= 2 I dt - It =
- In(2) - I (3r ?
t- 1 - Si
A:7:5 Ids
t l 3 t 1 J r 1 3 3 r't'l 1
Our algorithm, as it stands. cannot be used if either (or both) of the integration limits is infinite. A suitable
redefinition of the integration variable, however, will effect the conversion
A: 7:6
w
` f(t)dt
j g(u)du
and often render the function amenable to the numerical quadrature algorithm. Table All gives examples of
suitable transformations; there are many alternatives.
The preceding algorithm is designed to evaluate definite integrals. If one wanted to use it to study the behavior
of the indefinite integral
A:7:7 f(t)dt
of f as a function of x, one would need to access the algorithm repeatedly, replacing the upper limit x, by different
.r values until the range of interest had been scanned. Such a procedure would usually be impossibly tedious. The
algorithm below, however, is designed for just such a survey: it generates rather crude values of the integral A:7:7
at x values equal to xo + (x, - so)/n..r) + 2(x, - .ro)/n, ro + 3(x, - x.)/n, ..., xo + (n - 1)(.r1 - .xo)/n, x
where x, and n are values provided by the user.
Actually the algorithm is much more versatile than we have just suggested. Indefinite integration is just one
example (the v = - I instance) of the generalized differintegration operator of the calculus [see Section 0:10], and
our algorithm generates approximate values of
d"f(x)
A:7:8 at x=xo+J(x,-x0) for j=1.2.3,....n
[d(x - .ro)]" n
for any value of v [including v = -I when A:7:8 reduces to A:7:7]. The algorithm is based on the improved
Grumvald definition [Oldham and Spanier. pages 56-57]
Table A.7.1
L, Ii g(u)
-x x
A:7 UTILITY ALGORITHMS 672
It should be emphasized that there exist many better algorithms for specific orders of differintegration but that the
one given
x), here has the advantage of versatility. Comparison of the listed test values with those of (V r/2)
J.( the true semiderivative of sin(V) [see 52:3:13], reveals errors of up to 2%, which is typical for this
algorithm using it = 10. Accuracy increases with increasing n, but with a concomitant speed penalty.
APPENDIX
W
SOME USEFUL DATA
Since 1960 Le Syst2me International d Unites has been the official system of units throughout most of the world.
The SI system, as it is usually abbreviated, is used universally in science and increasingly in engineering and
medicine. This appendix contains information about the SI system and lists important physical constants and con-
version factors.
673
B3 SOME USEFUL DATA 674
Table B.2.2
Very many other SI units arise by combining specially named units with the primary units or with each other.
A few examples are listed in Table B.2.3.
B:3 PREFIXES
There is a set of 16 prefixes to the primary and special-name SI units that are used to create multiples and sub-
multiples. These are listed in Table B.3.1.
With one exception, these prefixes are attached to the primary or special-name Si units, as in the examples
B:3:1 l03 in s-' = kilometre per second = km s-'
m-2
B:3:2 10-12 F m-2 = picofarad per square metre = pF
The exception is with mass units, for which the prefix is attached to the gram, rather than the kilogram, unit; for
example:
Table B.2.3
Table B.3.1
There are a few units, not part of the SI system, that are officially condoned. These include
B:4:1 litre (L) = 10-3 m3 = dm3
B:4:2 tonne (t) = 103 kg = Mg
B:4:3 day (d) = 8.64 x 104 s
B:4:4 hectare (ha) = I0` m2 = hm2
a
B:4:5 degree (°) = rad
180
B:4:6 atomic mass unit (u) = 1.660566 x 10-27 kg
B:4:7 clectronvolt (eV) = 1.602189 x 10-19 J
and the degree Celsius (°C), for which see B:7:11 below.
The choice of a fundamental set of universal constants is, to a certain extent, arbitrary. One selection, in SI
units, is
B:5:1 gravitational constant = G = 6.6720 X 10-" N m2 kg-'
B:5:2 speed of light = c = 2.9979258 X 108 m s-'
B:5:3 Planck constant = h = 6.62618 x 10-74 J Hz-'
B:5:4 Boltzmann constant = k = 1.38066 x 10-231 K-1
B:5:5 Avogadro constant = L = 10-3 kg mol-' u-' = 6.02245 x 102' mol-'
B:5:6 elementary charge = q. -1.602189 x 10-19 C
B:5:7 electron rest mass = m, = 5.4858026 X 10-' u = 9.109534 X 10-" kg
B:5:8 proton rest mass = mp = 1.007276470 u = 1.672649 x 10-27 kg
Uncertain digits are italicized. Other constants, related to those above, or arbitrarily defined, include
B:5:9 permeability of free space = pv = 41r x 10-7 H m-'
B:5:I0 permittivity of free space = ea = 1/µac2 = 8.85418782 x 10-12 F m-'
B:6 SOME USEFUL DATA 676
5
B:7:11 (TK - 273.15) K = (Tc)°C = (Tp - 32) °F
9
677 SOME USEFUL DATA B:8
Abramowitz, M., and 1. A. Stegun (editors), Handbook of Mathematical Functions, National Bureau of Standards
(Applied Mathematics Series #55), Washington, D.C. (1964 and subsequent revisions).
Bartsch, H. S., Handbook of Mathematical Formulas, Academic Press, New York (1973).
Bell, W. W., Special Functions for Scientists and Engineers, Van Nostrand, London (1967).
Beyer, W. H., Handbook of Mathematical Sciences, CRC Press, Boca Raton, Fla. (1964 and subsequent editions).
Beyer, W. H., Handbook of Tables for Probability and Statistics, CRC Press, Boca Raton, Fla. (1966 and sub-
sequent editions).
Bronshtein, 1. N., and K. A. Semendyayev, A Guidebook to Mathematics. Hatri Deutsch, Frankfurt (1971 and
subsequent editions).
Carlson, B. C., Special Functions of Applied Mathematics, Academic Press, New York (1977).
Carslaw, H. S., and J. C. Jaeger, Conduction of Heat in Solids, Oxford University Press, London and New York
(1947).
Churchill, R. V., Operational Mathematics, third edition, McGraw-Hill, New York (1972).
Cooley, J. W., and J. W. Tukey, Mathematics of Computation. 19, 297 (1965).
Erd6lyi, A., W. Magnus, F. Oberhettinger and F. G. Tricomi (editors), Higher Transcendental Functions (Bateman
Manuscript), three volumes, McGraw-Hill, New York (1955).
Erd6lyi, A., W. Magnus, F. Oberhettinger and F. G. Tricomi (editors), Tables of Integral Transforms (Bateman
Manuscript), two volumes, McGraw-Hill, New York (1955).
Fletcher, A., J. C. P. Miller and L. Rosenhead, An Index of Mathematical Tables, McGraw-Hill, New York (1946).
Gradshteyn, I. S., and 1. M. Ryzhik, Table of Integrals Series and Products, second English edition, Academic
Press, York (1980).
Hamming, R: W., Numerical Methods for Scientists and Engineers, second edition, McGraw-Hill, New York
(1973).
Hart, J. F., E. W. Cheney, C. L. Lawson, H. J. Maehly, C. K. Mesztenyi, J. R. Rice, H. G. Thacher, Jr. and
C. Witzall, Computer Approximations, Wiley, New York (1968).
Hastings, C., Jr., Approximations for Digital Computers, Princeton University Press, Princeton, N.J. (1955).
Knuth, D. E., The Art of Computer Programming, Volume 2, Addison-Wesley, Reading, Mass. (1969).
Kom, G. A., and T. M. Kom, Mathematical Handbook for Scientists and Engineers, second edition, McGraw-
Hill, New York (1968).
Luke, Y. L., Algorithms for the Computation of Mathematical Functions, Academic Press, New York (1977).
Luke, Y. L., Mathematical Functions and their Approximations, Academic Press, New York (1975).
Luke, Y. L., The Special Functions and their Approximations, two volumes, Academic Press, New York (1969).
Macdonald, J. R., Journal of Applied Physics, 35, 3034 (1961).
679
REFERENCES AND BIBLIOGRAPHY 680
Magnus, W., F. Oberhettinger and R. P. Soni, Formulas and Theorems for the Special Functions of Mathematical
Physics, third edition, Springer-Verlag, New York (1966).
Murphy, G. M., Ordinary Differential Equations and their Solutions, Van Nostrand and Company, Princeton, N.J.
(1960).
Oldham, K. B., and 1. Spanier, The Fractional Calculus, Academic Press, New York (1974).
Reichel, A., Special Functions, Science Press, Sydney (1968).
Roberts, G. E., and H. Kaufman, Table of Laplace Transforms, Saunders, Philadelphia (1966).
Ruckdeschel, F. R., Basic Scientific Routines, two volumes, Byte/McGraw-Hill, New York (1981).
Sneddon, I. N., Special Functions of Mathematical Physics and Chemistry, Oliver and Boyd, Edinburgh (1956).
Sokolnikoff, I. S., and R. M. Redheffer, Mathematics of Physics and Modern Engineering, McGraw-Hill, New
York (1966).
Spiegel, M. R., Mathematical Handbook, McGraw-Hill, New York (1968).
Stoer, J., and R. Bulirsch, Introduction to Numerical Analysis, Springer-Verlag, New York (1980).
Szego, G., Orthogonal Polynomials, American Mathematical Society Colloquium Publications #23, Providence,
R.I. (1959).
Tuma, J. 1., Engineering Mathematics Handbook, McGraw-Hill, New York (1979).
Wall, H. S., Analytic Theory of Continued Fractions, Chelsea Publishing, Bronx, New York (1948).
Weast, R. C. (editor), CRC Handbook of Chemistry and Physics, CRC Press, Boca Raton, Fla. (annual Ikditions).
SUBJECT INDEX
References are to chapters and sections, not pages. Where several references are cited, the first is the prime ref-
erence; others are in numerical order. A citation such as "Sections 9" implies that the subject is encountered in
that section of most chapters. Where a chapter is cited, the subject occurs repeatedly throughout that chapter.
681
SUBJECT INDEX 682
Bernoulli number. Chapters 4 and 19, 1:14, 3:3. 3:7. Chemists' p. 25:14
3:13, 13:5, 20:7 Chi-square distribution, 27:14
Bernoulli polynomials. Chapter 19, 1:14, 64:4 Christoffel-Darboux formula for Hermite polynomi-
Bernoulli polynomials of order m. 19:12 als. 24:5
Bessel coefficients, Chapter 52, 43:14 Christoffel-Darboux formula for Laguerre polynomi-
Bessel function. Chapter 53 (see also Chapter 52), als, 23:5
47:4 Christoffcl-Darboux formulas for Chebyshev poly-
Besse] function of the second kind (see Neumann nomials. 22:5
function) Circular functions. Chapters 32-34
Bessel functions of imaginary argument (see hyper- Circular normal distribution, 27:14
bolic Bessel function or Basset function) Clausen functions, 18:14
Bessel functions of moiety orders, 32:13 Clausen's integral, 32:5
Bessel functions of the third kind, 54:13 Clifford's notation, 53:1
Bessel's equation. 53:3 Clotoid (see Comu's spiral)
Bessel-Clifford equation, 53:3 Coefficients of polynomial functions. 17:1
Best straight line, 7:14. A:4 Coefficients of quadratic function, 16:1
Beta distribution, 27:14 Cognate functions, Sections 13
Beta exponential integral. 37:13 Cologarithm, 25:14
Beta function, Chapter 3, 64:7, 64:13 (see also com- Combinations, 2:14, 6:1
plete beta or incomplete beta function) Combinatorics. 2:14
Beta number, Chapter 3. 5:4 Commands, 0:8
Binary logarithm, 25:14 Common antilogarithm, 26:12
Binomial coefficient, Chapter 6, 0:10, 4:5, 5:5, 43:13 Common logarithm (see decadic logarithm)
Binomial function, 6:14. 18:14 Common mean. 61:8. 1:7. 61:3. 62:5. 62:8
Binomial theorem, 6:14 Complementary error function integral, 40:13
Bipolar coordinate system, 46:14 Complementary function. 42:3
Bisector of angle in triangle. 34:14 Complementary incomplete gamma function, Chapter
Bit-reversed complement, 32:14 45
Bivariate eta function, 64:13, 3:12, 64:12 Complementary modulus, 61:1, 63:1
Bivariate function, 0:2 Complete beta function, 43:13, 6:12
Boehmer integrals, 39:12, 64:6, 64:10 Complete elliptic integral of the third kind, 61:12
Boltzmann distribution, 27:14 Complete elliptic integrals (of the first and second
Bose-Einstein distribution. 27:14 kinds), Chapter 61, 59:4, 62:8
Briggsian logarithm (see decadic logarithm) Complete gamma function (see gamma function)
Complex argument, Sections I1
Calculus operations, 0:10, Sections 10 Complex arithmetic, 1:11
Cartesian coordinate system, 46:14 Complex variable, 0:1
Catalan's constant, 1:7, 3:7, 61:10 Complex zero of polynomial function, 17:6
Catenary, 28:14 Complex zeros, A:5
Cauchy distribution, 27:14 Composite function. 26:14
Cauchy limit, 7:10, 14:10, 37:3 Concatenation, 0:6, A:6
Cauchy principal value, 27:14 Confluent hypergeometric equation, 47:3
Causal distribution. 27:14 Confluent hypergeometric functions (see Kummer
Center of circle, 14:4 function or Tricomi function)
Chain rule for differentiation, 0:10 Conic section, 14:0
Chain rule for Laplace transformation and inversion, Conjugate hyperbolas, 15:5
26:14 Conjugate pair of complex zeros, 17:6
Characteristic of elliptic integrals of the third kind, Constant function, Chapter I
61:12. 62:12 Constant of integration, 0:10
Characteristic of logarithm, 25:14 Contiguity relationships of Gauss functions. 60:5
Chebyshev functions, 22:1 Continued fraction. 0:6, 1:6, 17:6, 25:6. 27:6, 34:6,
Chebyshev polynomials (of the first kind and second 35:6, 37:6, 41:6, 42:6, 45:6, A:6
kind), Chapter 22, 11:6, 17:9, 28:5, 32:5 Conventions. figures, 0:2
Chebyshev's differential equation, 22:3 Conventions, symbolism, 0.1
'$3 SUBJECT INDEX
Convergent series, 0:6 Digamma function, Chapter 44, 1:14, 6:10, 7:5, 18:10,
Conversion factors, B:7 37:6
Convolution of Dirac delta functions, 10:10 Dilogarithm, 25:13, 64:12
Convolution property of Laplace inversion, 26:14 Dirac delta function, Chapter 10, 8:10, 26:14
Convolution, Vandermonde's, 6:5 Directrix of ellipse, 14:3
Copolar elliptic functions, 63:1 Directrix of hyperbola, 15:3
Comu's spiral, 39:14 Directrix of parabola, 12:3
Correlation coefficient, 7:14 Discontinuous function, 8:4
Cosecant function, Chapter 33, 3:14. 43:5 Discrete Chebyshev polynomials, 22:13, 17:14
Cosine function, Chapter 32 Discrete orthogonal polynomials, 22:13
Cosine integral, Chapter 38 Discriminant of cubic function, 17:1
Cosine law, 34:14, 21:3 Discriminant of quadratic function, 16:1
Cosine transformation, 32:10 Distribution function, 27:14
Cosinus amplitudinis, 63:1 Distributions, 27:14
Cotangent function, Chapter 34, 3:14 Domain of function, 0:2
Coversine function, 32:13 Double factorial function, 2:13, 6:4, 43:4
Cube function, 11:1 Double zero. 0:7
Cube-root, 13:1 Doubly periodic function, 63:11
Cubic function, Chapter 17, 62:14 Dummy variable. 26:14
Cubic spline, 17:14 Duplication formula for gamma function, 43:5
Cumulative function, 27:14
Cumulative standard normal probability function, 40:14
Curvature, 39:14 Eccentricity of ellipse, 14:3
Curve of the second degree. 14:0 Eccentricity of hyperbola, 15:3
Cylinder function (see Bessel function) Economization, 22:14, 17:9, A:6
Cylindrical coordinate system, 46:14 Eigenvalues, 59:14
Elementary symmetric functions, 17:6
Ellipse, 14:14, 61:3, 62:3
Dawson's integral, Chapter 42, 37:14, 40:11, 45:4, Elliptic amplitude, 63:3, 61:14, 62:1, 63:1, 63:10
45:14 Elliptic cylindrical coordinate system, 46:14
de Moivre's theorem, 1:11, 28:5, 32:11 Elliptic functions (see Jacobian elliptic functions or
Decadic logarithm, 25:14 elliptic integrals)
Definite integrals, 0:10, Sections 10, A:7 Elliptic integrals, 17:10 (see also complete elliptic in-
Definitions, Sections 3 tegrals or incomplete elliptic integrals)
Degenerate hypergeometric functions (see Kummer Elliptic modulus, 61:1. 63:1
function or Tricomi function) Elliptic parameter, 61:1. 62:1
Delay property of Laplace transformation, 26:14 Entire cosine integral, Chapter 38
Delta amplitudinis, 63:1 Entire exponential integral, Chapter 37
Delta distribution, 27:14 Entire hyperbolic cosine integral. Chapter 38
Delta function (see Dirac delta function or Kronecker Entire incomplete gamma function, Chapter 45
delta function) Entire Kummer function. 47:12
Denominatorial parameter, 18:14 Equilateral hyperbola, 15:4
Density function. 27:14 Equipotential, 46:14
Derivatives, 0:10, Sections 10 Error, 40:14
Diagonal Pad6 approximant, 17:13, A:6 Error function, Chapter 40, 8:9, 45:4. 45:14
Diagram conventions, 0:2 Error function complement, Chapter 40, 37:14, 46:4
Differentiation, 0:10, Sections 10 Error function of complex argument. 41:11, 40:11
Differentiation of integrals, 0:10 Eta function, Chapter 3. 64:13
Differintegral property of Laplace transformation, 26:14 Eta number, Chapter 3. 44:13
Differintegration, 0:10, 43:14 Euler function, 43:14
Differintegration algorithm, A:7 Euler hypergeometric integral, 60:3
Differintegration of periodic functions, 64:14 Euler integral definition of gamma function, 43:3
Differintegration of sinusoid. 64:14 Euler number, Chapter 5. 3:7, 3:13. 19:7, 20:7, 33:14
Differintegration of square wave function, 64:14 Euler polynomials, Chapter 20, 1:14
SUBJECT INDEX 684
Stirling number of the first kind, 18:6, 2:5, 6:6 Ultrespherical polynomials, 22:12
Stirling number of the second kind, 2:14, 11:6 Undetermined coefficients, 17:13
Stirling's formula for factorials, 2:6 Uniform distribution, 27:14
Stirling's formula for gamma function, 43:6 Unit-impulse function (see Dirac delta function)
Streamline, 46:14 Unit-moment function, 10:12
Struve function, Chapter 57, 53:12 Unit-step function. Chapter 8
Student's-t distribution, 27:14 Units, B:l, B:2, B:3
Subroutines, A:6 Unity function, 1:4
Successive derivatives of Bateman's G function, 44:13, Universal constants. B:5
64:13 Universal hypergeometric algorithm, 18:14
Successive derivatives of error function, 40:1, 40: 10, Utility algorithms, Appendix A
46:4
Sums of powers, 1:14, 19:14, 20:14 Vandennonde's convolution, 6:5
Superellipse, 14:12 Vandermonde's theorem, 18:5
Surface harmonics, 59:14 Variance, 27:14
Symbolism, Sections 1 Variations, 2:14
Symbolism, conventions adopted, 0:1 Versine function, 32:13
Symmetrical periodic function, 26:14
Synthesis of functions, 43:14 Wallis' formula. 2:13
Wave equation. 46:14
Tangent function, Chapter 34, 3:14 Weber function (see parabolic cylinder function or
Tangent rule, 34:14 Neumann function)
Taylor expansion, 0:5, 6:14 Weber-Hermite function (see parabolic cylinder func-
Telescoping of series, 22:14 tion)
Terrestrial constants, B:6 Weibull distribution, 27:14
Tesseral surface harmonics, 59:14 Weierstrass infinite product definition of gamma
Test values, 0:8 function, 43:3
Tetragamma function, 44:12 Weierstrassian elliptic functions, 63:13
Theorem of Gauss, 44:4 Weight function of an orthogonal polynomial, 21:14
Theta functions, 27:13, 26:5 Weight of a datum, 7:14
Thomson functions (see Kelvin functions) Weighted least squares, 7:14
Toroidal coordinate system, 46:14 Weyl differintegration, 64:14
Trajectory, 16:14 Whittaker functions, 48:13, 51:12
Transcendental function, 53:0 Window function, 8:12
Translation formula, 0:5
Trapezoidal approximation. A:7 Zero order hyperbolic Bessel function. Chapter 49
Triangles, 34:14 Zeros of Airy functions, 56:7
Tricomi function. Chapter 48, 18:14, 24:12. 40:12, Zeros of Besse) coefficients, 52:7, 52:9
45:14, 46:12 Zeros of Bessel function, 53:7
Trigamma function, 44:12 Zeros of functions, Sections 7, 8:10
Trigonometric functions. Chapters 32-34 Zeros of Kelvin functions, 55:7
Trilogarithm. 25:13, 18:14 Zeros of Legendre functions. 59:2
Trinomial coefficient, 6:12 Zeros of Neumann function, 54:7
Triple factorial function, 2:13, 43:4, 56:6 Zeros of quadratic function, 16:7
Triplication formula for gamma function, 43:5 Zeros of Struve function, 57:7
Zeta function, Chapter 3, 64:7
Ubiquitous constant, 1:7, 3:5, 27:13, 43:4, 59:7, 61:7, Zeta number, Chapter 3. 44:6
61:10 Zonal surface harmonics, 59:14, 21:1
SYMBOL INDEX
References are to the chapter and/or section in which the symbol is defined or introduced. It is intended that this
index will provide a complete glossary of all the notations used in the Atlas. but temporary symbolism that is
defined locally is not indexed. Use of the word 'see" following a symbol implies that the notation is identified in
the referenced section but is not used elsewhere in the Atlas; such a symbol has a meaning identical (or related)
to the notation that follows "see." This listing may also be used as a function index. the final section has been
added for this purpose.
Q.(x) Legendre function (of the second kind) of integer order n and argument x.
21:13
R gas constant, B:5
Re{ } real part of, 55:3
R(x,y ) rational function of arguments x and Y. 26:14
R. ratio of two consecutive hyperbolic Bessel functions, 49:8
R. ratio of two consecutive Bessel functions of common argument, 52:5
R1 remainder after truncating power series to polynomial of order J - 1. 18:14
R Rydberg constant. 13:5
R,'(x) the rational function Po(x)/P.(x). 17:13
diagonal Pade approximants, 17:13
S(x) Fresnel sine integral of argument x. Chapter 39
S(x;v) Boehmer sine integral of degree v and argument x, 39:12
Shi(x) hyperbolic sine integral of argument x. Chapter 38
Si(x) sine integral of argument x. Chapter 38
Sin(x) see sinh(x). 28:1
S,(x). S.(x) see S(x). 39:1
Sa(x) see u(x - a). 8:1
see U,(x), 22:1
S Stirling numbers (of the first kind). 18:6
Tah(x) see tanh(x). 30:1
T,(x), Chebyshev polynomials (of the first and second kinds) of degree n and ar-
gument x, Chapter 22
see 22:1
ubiquitous constant, 1:7
Tricomi functions of parameters a, c and argument .r. Chapter 48
function cognate to parabolic cylinder function, 46:13
finite sums related to spherical Besse] functions, 32:13
error function of complex argument :. 41:11
Whittaker function of parameters v, µ and argument x, 48:13
auxiliary argument of Airy function, 56:1
auxiliary argument of hypergeometric function. 18:14
factorial polynomial, 18:13
Neumann function of order v and argument x. Chapter 54
spherical harmonics of two angular coordinates 0 and $, 59:14
Z the zeta number 4(3). 3:7
GREEK LETTERS: CAPITAL AND LOWERCASE (see Section B:8 for names)
a fine-structure constant. B:5
a modular angle, 61:1, 62:1
SYMBOL INDEX 698
NONALPHABETIC SYMBOLS
z, f(x) approximate value of x, f(x), 0:9
see n! 2:1
n! factorial function of the number n, Chapter 2
X! see I'(x), 43:1
n!!, n!!!, n!!!! double, triple, quadruple factorial function, 2:13
f(x)*g(x) the convolution of two functions, 10:10, 26:14
x square root of x, Chapter 12
.Nrx see x'i", 13:1
V2 Laplacian operator, 46:14
(r) a surface specified by the value of the r coordinate, 46:14
(x) see frac(x), 9:1
(:) binomial coefficient of upper index v and lower index in, Chapter 6
(q,p) a line (space curve) specified by the values of the q and p coordinate, 46:14
(x,n) see (x),,, 18:1
(r,q,p) a point specified by the values of the three coordinates r, q and p, 46:14
n(m) see n(mod rn), 9:13
n(mod m) number n modulo number m, 9:13
[x] see Int(x), 9:1
xt"' factorial polynomial, 18:13
W. Pochhammer polynomial of argument x and degree n, Chapter 18
(x) rounding function, 9:13
absolute-value function of argument x, Chapter 8
CI..L an abbreviation for the product c,c2c3 cL_,cL, 26:14
(a HK)/ an abbreviation for the product (a,)J(a2)i(a3)/ . . 26:14
SYMBOL INDEX 700
Finally, and again with the motive of having this symbol index serve also as a comprehensive function index,
we list functions that are described in the Atlas but for which it was not necessary to adopt any special notation.
These are listed in alphabetical order, followed by a section reference: Clausen function (18:14), cumulative stan-
dard normal probability function (40:14), elementary symmetric functions (17:6), Hankel functions (54:13). Hermite
function (24:14), Heuman's lambda function (62:13), higher-order Bernoulli polynomials (19:12), hypergeometric
function (18:14), Jacobi's zeta function (62:13), Langevin function (30:13), piecewise-constant function (1:13),
piecewise-defined function (8:12), piecewise-linear function (7:13), standard normal density function (40:14), win-
dow function (8:12).