Leonhard Euler's Integral: A Historical Profile of the Gamma Function: In Memoriam: Milton

Abramowitz
Author(s): Philip J. Davis
Source: The American Mathematical Monthly, Vol. 66, No. 10 (Dec., 1959), pp. 849-869
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2309786 .
Accessed: 31/07/2013 22:45
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

19591

LEONHARD EULER'S

INTEGRAL

849

FIG. 2: p=3, q=1, k=2a.

LEONHARD EULER'S INTEGRAL:

A HISTORICAL PROFILE OF THE GAMMA FUNCTION
IN MEMORIAM: MITON

ABRAMOWITZ

PHILIP J. DAVIS, National Bureau of Standards,Washington,D. C.

Many people thinkthat mathematicalideas are static. They thinkthat the

ideas briginatedat some time in the historicalpast and remain unchanged for
all futuretimes. There are good reasons forsuch a feeling.Afterall, the formula
forthe area of a circlewas 7rr2in Euclid's day and at the presenttime is still irr2.
But to one who knows mathematicsfromthe inside, the subject has ratherthe
feelingof a living thing. It grows daily by the accretion of new information,it
changes daily by regardingitselfand the world fromnew vantage points, it
maintains a regulatorybalance by consigningto the oblivion of irrelevancya

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

850

LEONHARD EULER S INTEGRAL

[December

fractionof its past accomplishments.

The purpose of this essay is to illustratethis process of growth.We select
one mathematicalobject, the gamma function,and show how it grewin concept
and in content fromthe time of Euler to the recent mathematical treatise of
Bourbaki, and how, in this growth,it partook of the general development of
mathematicsover the past two and a quarter centuries.Of the so-called "higher
mathematicalfunctions,"the gamma functionis undoubtedly the most fundamental. It is simple enough for juniors in college to meet but deep enough to
have called forthcontributionsfromthe finestmathematicians.And it is sufficiently compact to allow its profileto be sketched within the space of a brief
essay.
The year 1729 saw the birthof the gamma functionin a correspondencebetween a Swiss mathematicianin St. Petersburgand a German mathematician
in Moscow. The former:Leonhard Euler (1707-1783), then 22 years of age, but
to become a prodigious mathematician,the greatest of the 18th century.The
latter: Christian Goldbach (1690-1764), a savant, a man of many talents and
in correspondencewith the leading thinkersof the day. As a mathematicianhe
was somethingof a dilettante,yet he was a man who bequeathed to the future
to prove that
a problemin the theoryof numbersso easy to state and so difficult
a
challenge.
as
horizon
mathematical
even to this day it remainson the
The birthof the gamma functionwas due to the mergingof several mathematical streams. The firstwas that of interpolationtheory, a very practical
subject largelythe product of English mathematiciansof the 17th centurybut
which all mathematiciansenjoyed dipping into fromtime to time. The second
streamwas that of the integralcalculus and of the systematicbuildingup of the
formulasof indefiniteintegration,a process which had been going on steadily
formany years. A certain ostensiblysimple problem of interpolationarose and
was bandied about unsuccessfullyby Goldbach and by Daniel Bernoulli (17001784) and even earlierby JamesStirling(1692-1770). The problemwas posed to
Euler. Euler announced his solution to Goldbach in two letterswhich were to
be the beginningof an extensive correspondencewhich lasted the duration of
Goldbach's life.The firstletterdated October 13, 1729 dealt with the interpolation problem, while the second dated January 8, 1730 dealt with integration
and tied the two together.Euler wroteGoldbach the merestoutline,but within
transcendenta year he published all the details in an articleDe progressionibus
ibus seu quarum terminigeneralesalgebraicedari nequeunt.This article can now
be foundreprintedin Volume I14 of Euler's Opera Omnia.
Since the interpolationproblem is the easier one, let us begin with it. One
of the simplest sequences of integerswhich leads to an interestingtheory is 1,
. These are the triangularnumbers,so called
*
1+2, 1+2+3, 1+2+3+4,
because theyrepresentthe numberofobjects whichcan be placed in a triangular
array of various sizes. Call the nth one T.. There is a formulafor T. which is
learned in school algebra: T.= In(n+1).
What, precisely,does this formulaaccomplish? In the firstplace, it simplifies
.

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

851

LEONHARD EULER S INTEGRAL

computationby reducinga large numberof additions to threefixedoperations:

one of addition,one of multiplication,and one of division.Thus, instead of adding the firsthundredintegersto obtain T0oo0
we can compute Tloo= I(100)(100+1)
= 5050. Secondly, even though it doesn't make literal sense to ask for,say, the
sum of the first51 integers,the formulafor T. produces an answer to this. For
whatever it is worth,the formulayields T51= (5 )(5I+1) = 17k. In this way,
the formulaextends the scope of the originalproblem to values of the variable
other than those forwhich it was originallydefinedand solves the problem of
interpolatingbetween the known elementaryvalues.
This type of question, one which asks foran extensionof meaning,cropped
up frequentlyin the 17th and 18thcenturies.Consider,forinstance,the algebra
of exponents.The quantity am is definedinitiallyas the product of m successive
a's. This definitionhas meaningwhen m is a positive integer,but what would a5l
be? The product of 5' successive a's? The mysteriousdefinitionsa'=1, am/n
= /am,a-" = l/am which solve this enigma and which are employed so fruit-

fullyin algebra were writtendown explicitlyfor the firsttime by Newton in

1676. They are justifiedby a utilitywhich derives fromthe fact that the definition leads to continuous exponential functionsand that the law of exponents
am an =am+n becomes meaningfulforall exponentswhetherpositive integersor
not.
Other problems of this type proved harder. Thus, Leibnitz introducedthe
notation dn forthe nth iterate of the operation of differentiation.
Moreover,he
identifiedd-l withf and d-n withthe iteratedintegral.Then he tried to breathe
some sense into the symboldn when n is any real value whatever.What, indeed,
is the 5'th derivative of a function?This question had to wait almost two centuries fora satisfactoryanswer.

THE FACTORIALS

n:
n !:

1
1

2
2

3
6

4
24

5
120

6
720

7
5040

8
40,320

...

FIG. 1
INTELLIGENCE TEST

Question:What numbershouldbe insertedin the lowerline halfway betweenthe upper5

and 6?
Euler's Answer:287.8852*v. Hadamard'sAnswer:280.3002* v

But to returnto our sequence of triangularnumbers.If we change the plus

signs to multiplication signs we obtain a new sequence: 1, 1 *2, 1 *2.3, * * * . This

is the sequence of factorials.The factorialsare usually abbreviated 1!, 2!, 3!, .

and the firstfiveare 1, 2, 6, 24, 120. They growin size veryrapidly.The number
100! if writtenout in full would have 158 digits. By contrast, T1oo=5050 has a

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

852

LEONHARD EULER S INTEGRAL

[December

mere four digits. Factorials are omnipresentin mathematics; one can hardly
open a page of mathematicalanalysis withoutfindingit strewnwith them. This
being the case, is it possible to obtain an easy formulaforcomputingthe factorials? And is it possible to interpolatebetween the factorials?What should
5 ! be? (See Fig. 1.) This is the interpolationproblem which led to the gamma
function,the interpolationproblem of Stirling,of Bernoulli, and of Goldbach.
As we know, these two problemsare related, forwhen one has a formulathere
is the possibilityof insertingintermediatevalues into it. And now comes the
surprisingthing.There is no, in fact therecan be, no formulaforthe factorials
which is of the simple type foundfor Tn. This is implicitin the very title Euler
progreschose forhis article.Translate the Latin and we have On transcendental
sions whosegeneraltermcannotbeexpressedalgebraically.The solutionto factorial
interpolationlay deeper than "mere algebra." Infiniteprocesses were required.
In orderto appreciate a little betterthe problemconfrontingEuler it is useful to skip ahead a bit and formulateit in an up-to-datefashion: finda reasonably simple functionwhich at the integers1, 2, 3, - * - takes on the factorial
values 1, 2, 6, . - a . Now today, a functionis a relationshipbetween two sets of
numberswhereinto a numberof one set is assigned a numberof the second set.
What is stressedis the relationshipand not the natureof the ruleswhichserve to
determinethe relationship.To help students visualize the functionconcept in
its fullgenerality,mathematicsinstructorsare accustomed to draw a curve full
of twistsand discontinuities.The moreof these the moregeneralthe functionis
supposed to be. Given, then, the points (1,1), (2, 2), (3, 6), (4, 24), * * * and
adopting the point of view wherein"function"is what we have just said, the
problemof interpolationis one of findinga curve whichpasses throughthe given
points. This is ridiculouslyeasy to solve. It can be done in an unlimitednumber
of ways. Merely take a pencil and draw some curve-any curve will do-which
passes throughthe points. Such a curve automatically definesa functionwhich
solves the interpolationproblem. In this way, too freean attitude as to what
constitutes a functionsolves the problem trivially and would enrich matheIn the early 18thcentury,a function
matics but little.Euler's task was different.
and by a formulawas meant an
formula,
a
with
less
synonymous
was more or
manipulationswithaddition,
elementary
from
be
derived
expressionwhichcould
exponentials, logarithms,
roots,
powers,
division,
subtraction, multiplication,
came fromthe ordinary
one
which
infinite
i.e.,
series,
integration,
differentiation,
called an expressio
a
was
formula
Such
processes of mathematical analysis.
if he could, an
was
to
task
find,
Euler's
analytica, an analytical expression.
which
of
mathematics
the
from
corpus
analytical expressionarising naturally
would
but
which
was
inserted,
would yield factorialswhen a positive integer
still be meaningfulforothervalues of the variable.
It is difficultto chronicle the exact course of scientificdiscovery. This is
particularlytrue in mathematics where one traditionallyomits from articles
and books all accounts of false starts,of the initial years of bungling,and where
one may develop one's topic forwardor backward orsideways in orderto heighten

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

853

LEONHARD EULER S INTEGRAL

the dramatic effect.As one distinguishedmathematicianput it, a mathematical

result must appear straightfromthe heavens as a deus ex machina forstudents
to verifyand accept but not to comprehend.Apparently,Euler, experimenting
withinfiniteproductsofnumbers,chanced to noticethat ifn is a positiveinteger,
(1)

3
1[()

[G)n
2+2 A1 1[GY

n+ 2J

L\Jn+

+31

n
n+3-

Leaving aside all delicate questions as to the convergenceof the infiniteproduct,

the reader can verifythis equation by cancelling out all the common factors
which appear in the top and bottom of the left-handside. Moreover, the lefthand side is defined (at least formally)for all kinds of n other than negative
integers.Euler noticed also that when the value n -= is inserted,the left-hand
side yields (aftera bit of manipulation) the famous infiniteproduct of the Englishman JohnWallis (1616-1703):

(2)

(1-)

(3i)

)*

* = 7r/2.

With this discovery Euler could have stopped. His problem was solved.
Indeed, the whole theoryof the gamma functioncan be based on the infinite
product (1) which today is writtenmore conventionallyas
m!(m + 1)n

lim

(3)

rs-)o (n + 1)(n

+ 2) **-(n +rn)

However, he went on. He observed that his product displayed the following
curious phenomenon:forsome values of n, namely integers,it yielded integers,
whereas for another value, namely n = 2, it yielded an expressioninvolving 7r.
Now 7r meant circles and their quadrature, and quadratures meant integrals,
and he was familiarwith integralswhich exhibited the same phenomenon. It
thereforeoccurredto him to look fora transformationwhichwould allow him to
express his product as an integral.
He took up the integralfoxe(1-x)ndx. Special cases of it had already been
discussed by Wallis, by Newton, and by Stirling.It was a troublesomeintegral
to handle, forthe indefiniteintegralis not always an elementaryfunctionof x.
Assumingthat n is an integer,but that e is an arbitraryvalue, Euler expanded
found that
(1 -x) by the binomial theorem,and withoutdifficulty
(4)

r1
j-

X)dx

(e+ 1)(e+

1-2 ... n
2) ... (e+n+

1)

Euler's idea was now to isolate the 1-2 ... n fromthe denominatorso that he
would have an expressionforn! as an integral.He proceeds in this way. (Here
we followEuler's own formulationand nomenclature,markingwith an * those

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

854

LEONHARD EULER'S

[December

INTEGRAL

formulaswhich occur in the original paper. Euler wrote a plain f forft.) He

substitutedf/g fore and found
r1
g"+1
1-2
-n
+
I xflx(1-x)ndx= o)dx
f + (n + 1)g( f + g)(f + 2-g) * (f + n*g)

(5)

And so,
(6)*

1-2 (f + g)(f + 2-g) *

f + (n+)g
(f + n-g)

gf+1

xflgdx(1-X)

n.

He observed that he could isolate the 1-2 ... n if he set f = 1 and g=O in the
left-handmember,but that ifhe did so, he would obtain on the rightan indeterminate formwhich he writesquaintjy as

- x)n
rxll0dx(l
d

(7)*

n+1

He now proceeded to findthe value of the expression (7) *. He firstmade the

in place of x. This gave him
substitutionxgl(f+g)
(8)*

9g xfI(u+f)dx
f+g

in place of dx and hence, the right-handmemberof (6) * becomes

(9)*

n+(

+g

dx(l

XgI(f+))n.

Once again, Euler made a trial settingoff = 1, g = 0 having presumablyreduced this integralfirstto
(10)
(10)

+ (n+
~ ~~~f
(f +g)n+l

1)g
Jog(fg
r

;)

nx
dx,

and this yielded the indeterminate

(l )*

X0)n

dx

He now considered the related expression (1 -xz)/z, forvanishingz. He differentiated the numeratorand denominator,as he says, by a known (l'Hospital's)
rule,and obtained
-

(12)*

x-dzlx

(lx = log x),

dz

which forz=0 produced -lx. Thus,

(13)*

(1

x)/

=-

Ix

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

855

LEONHARD EULER S INTEGRAL

and
(1 - X?)8On=

(14)*

(-Ix)".

He thereforeconcluded that
nI

(15)

(-log

x)ndx.

This gave himwhat he wanted, an expressionforn! as an integralwhereinvalues

other than positive integersmay be substituted. The reader is encouraged to
formulatehis own criticismof Euler's derivation.
Students in advanced calculus generally meet Euler's integral firstin the
form
(16)

r(x) = f

00

e-ttx-ldt,

2.71828

This modificationof the integral (15) as well as the Greek r is due to Adrien
Marie Legendre (1752-1833). Legendre calls the integral (4) with which Euler
started his derivation the firstEulerian integraland (15) the second Eulerian
integral.The firstEulerian integralis currentlyknownas the Beta functionand
is now conventionallywritten
(17)

B(m, n)

xm1(1 -x)'-ldx.

With the tools available in advanced calculus, it is readily established (how

easily the great achievementsof the past seem to be comprehendedand duplicated!) that the integralpossesses meaningwhen x > 0 and thus yields a certain
functionr(x) definedforthese values. Moreover,
(18)

r(n + 1)

= n!

whenevern is a positive integer.*It is furtherestablished that forall x>0

(19)

xr(x) = r(x + 1).

This is the so-called recurrencerelation for the gamma functionand in the

years followingEuler it plays, as we shall see, an increasinglyimportantrole
in its theory. These facts, plus perhaps the relationshipbetween Euler's two
types of integrals
(20)

B(m, n)

= r(m)r(n)/r(m
+

n)

and the all importantStirlingformula

* Legendre'snotationshiftstheargument.
a notation7r(x)freeofthisdefect.
Gaussintroduced
to plaguemanypeople.The notationsr, r, and ! can
Legendre'snotationwonout,butcQatinules
all be foundtoday.

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

856

LEONHARD

EULER'

S INTEGRAL

[December

P(x)
'e$xxl2V/(27r),

(21)

which gives us a relatively simple approximate expression for r (x) when x is

large, are about all that advanced calculus students learn of the gamma function. Chronologicallyspeaking,this puts them at about the year 1750. The play
has hardly begun.
Justas the simpledesireto extendfactorialsto values in betweenthe integers
led to the discoveryof the gamma function,the desire to extend it to negative
values and to complex values led to its furtherdevelopmentand to a moreprofound interpretation.Naive questioning, uninhibitedplay with symbols may
have been at the very bottom of it. What is the value of (-5g)! ? What is the
value of V/(- 1)! ? In the early years of the 19thcentury,the action broadened
and moved into the complex plane (the set of all numbersof the formx+iy,
where i = \/(-1)) and thereit became part of the general development of the
theory of functionsof a complex variable that was to formone of the major
chapters in mathematics.The move to the complex plane was initiatedby Karl
Friedrich Gauss (1777-1855), who began with Euler's product as his starting
point. Many famous names are now involved and not just one stage of action
but many stages. It would take too long to record and describe each forward
step taken. We shall have to be contentwith a broader picture.
Three importantfacts were now known: Euler's integral,Euler's product,
and the functionalor recurrencerelationshipxr(x) = r(x+ 1), x>0. This last
is the generalization of the obvious arithmeticfact that for positive integers,
(n+ 1)n! = (n+ 1)! It is a particularlyusefulrelationshipinasmuch as it enables
us by applying it over and over again to reduce the problemof evaluating a factorial of an arbitraryreal numberwhole or otherwiseto the problemof evaluating the factorialof an appropriate number lying between 0 and 1. Thus, if we
writen = 41 in the above formulawe obtain (41+ 1)! = 5(42)! If we could only
findout what (4D)! is, then we would know that (52)! is. This process of reduction to lower numberscan be kept up and yields
(5-)!

(22)

(3/2)(5/2)(7/2)(9/2)(11/2)(1/2)!

from(1) and (2), we can now compute our answer.

V/7r
and since we have (2) ! = 41
Such a device is obviously very important for anyone who must do calculations with the gamma function. Other informationis forthcomingfrom the
recurrencerelationship.Though the formula(n + 1)n! = (n + 1)! as a condensation of the arithmetic identity (n+1)-1-2 . . . n=1-2 - - - n-(n+1) makes
sense only forn = 1, 2, etc., blind insertionsof other values produce interesting
things. Thus, insertingn = 0, we obtain 0! = 1. Insertingsuccessively n =-2
n=-42

(23)

and reducingupwards,we discover

(-52)!

= (2/1)(-2/1)(-2/3)(-2/5)(-2/7)(-2/9)(1/2)!

Since we already know what (i)! is, we can compute (-5kL)! In this way the
recurrencerelationshipenables us to compute the values of factorialsof negative

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

LEONHARD EULER S INTEGRAL

857

numbers.
Turning now to Euler's integral,it can be shown that forvalues of the variable less than 0, the usual theoremsof analysis do not sufficeto assign a meaning to the integral,forit is divergent.On the other hand, it is meaningfuland
yields a value if one substitutes for x any complex number of the forma+bi
where a>0. With such substitutionsthe integral thereforeyields a complexvalued functionwhichis definedforall complex numbersin the right-halfof the
complex plane and which coincides with the ordinarygamma functionforreal
values. Euler's product is even stronger.With the exception of 0, -1, -2, * *
any complex numberwhatever can be insertedforthe variable and the infinite
product will converge,yieldinga value. And so it appears that we have at our
disposal a numberof methods,conceptually and operationallydifferentforextending the domain of definitionof the gamma function. Do these different
methodsyield the same result?They do. But why?
The answer is to be foundin the notion of an analytic function.This is the
focal point of the theoryof functionsof a complex variable and an outgrowth
of the older notionof an analytical expression.As we have hinted,earliermathematics was vague about this notion, meaning by it a functionwhich arose in a
natural way in mathematicalanalysis. When later it was discoveredby J. B. J.
Fourier (1768-1830) that functionsof wide generalityand functionswith unpleasant characteristicscould be produced by the infinitesuperpositionof ordinary sines and cosines,it became clear that the criterionof "arisingin a natural
way" would have to be dropped. The discoverysimultaneouslyforceda broadeningof the idea of a functionand a narrowingofwhat was meant by an analytic
function.
Analytic functionsare not so arbitraryin their behavior. On the contrary,
they possess strong internal ties. Defined very precisely as functionswhich
possess a complex derivative or equivalently as functionswhich possess power
- - - they exhibit the remarkable
series expansions ao+a,(z-z0)+a2(z
-z0)2+
phenomenon of "action at a distance." This means that the behavior of an
analytic functionover any interval no matter how small is sufficientto determine completelyits behavior everywhereelse; its potential range of definition
and its values are theoreticallyobtainable fromthis information.Analytic functions, moreover,obey the principle of the permanence of functionalrelationships; ifan analytic functionsatisfiesin some portionsof its regionof definition
a certain functional relationship,then it must do so wherever it is defined.
Conversely, such a relationshipmay be employed to extend its definitionto
unkndwnregions.Our understandingof the process of analytic continuation,as
this phenomenon is known, is based upon the work of Bernhard Riemann
(1826-1866) and Karl Weierstrass (1815-1897). The complex-valued function
which resultsfromthe substitutionof complex numbersinto Euler's integralis
an analytic function.The functionwhich emerges fromEuler's product is an
analytic function.The recurrencerelationshipfor the gamma functionif satisfiedin some regionmust be satisfiedin any other regionto which the function
.

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

858

LEONHARD

EULER

[December

S INTEGRAL

can be "continued" analyticallyand indeed may be employed to effectsuch extensions. All portions of the complex plane, with the exception of the values
0, -1, -2, * * * are accessible to the complex gamma functionwhich has become the unique, analytic extensionto complex values of Euler's integral (see
Fig. 3).
THE GAMMA FUNCTION

XX

14S

FIG. 2*

To understand why there should be excluded poilntsobserve that r7(x)

-r'(x+1)/Ix, and as x approaches 0, we obtainr 1(0) = 1/0. This is + 00or - 0
depend'ingwhether0 is approached throughpositive or negative values. The
* From: H. T. Davis, Tables of the HigherMathematicalFunctions,vol. I, Bloomington,
1933.
Indi_ana,

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

LEONHARD EULER'S

859

INTEGRAL

functionalequation (19) then, induces this behavior over and over again at
each of the negative integers.The (real) gamma functionis comprised of an
infinitenumberof disconnectedportionsopening up and down alternately.The
portions correspondingto negative values are each squeezed into an infinite
strip one unit in width,but the major portionwhich correspondsto positive x
and whichcontains the factorialsis of infinitewidth (see Fig. 2). Thus, thereare
excluded points forthe gamma functionat which it exhibitsfromthe ordinary
(real variable) point of view a somewhat unpleasant and capricious behavior.
THE ABSOLUTE VALUE OF THE COMPLEX GAMMA FUNCTION, EXHIBITING THE POLES AT
THE NEGATIVE INTEGERS

-7

-3

-2

X,

FIG. 3*

But fromthe complex point of view, these points of singularbehavior (singular

in the sense of Sherlock Holmes) meritspecial study and become an important
part of the story.In picturesof the complex gamma functiontheyshow up as an
infiniterow of "stalagmites,"each of infiniteheight (the ones in the figureare
truncatedout of necessity)which become more and more needlelikeas they go
out to infinity(see Fig. 3). They are known as poles. Poles are points wherethe
functionhas an infinitebehavior of especially simple type, a behavior which is
akin to that of such simple functionsas the hyperbola y = 1/x at x = 0 or of
y = tan x at x = r/2. The theory of analytic functionsis especially interested
* From:E. Jahnkeand F. Emde,TafeinhohererFunktionen,
4thed., Leipzig,1948.

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

860

[December

LEONHARD EULER S INTEGRAL

in singularbehavior, and devotes much space to the study of the singularities.

Analytic functionspossess many types of singularitybut those with only poles
are known as meromorphic.There are also functionswhich are lucky enough to
possess no singularitiesfor finitearguments.Such functionsforman elite and
are known as entire functions.They are akin to polynomials while the meromorphicfunctionsare akin to the ratio of polynomials.The gamma functionis
meromorphic.Its reciprocal, 1/r(x), has on the contraryno excluded points.
There is no troubleanywhere.At the points 0, -1, -2, * * * it merelybecomes
zero. And the zero value which occurs an infinityof times, is stronglyreminiscent of the sine.
In the wake of the extension to the complex many remarkable identities
emerge,and though some of them can and were obtained without referenceto
complex variables, they acquire a fardeeper and richermeaningwhen regarded
fromthe extended point of view. There is the reflectionformulaof Euler

r(z)r(1 - z) = lr/sinirz.

(24)

It is readily shown, using the recurrencerelation of the gamma function,that

the product r(z)r(1 -z) is a periodic functionof period 2; but despite the fact
that sin irzis one of the simplestperiodic functions,who could have anticipated
the relationship (24)? What, after all, does trigonometryhave to do with the
sequence 1, 2, 6, 24 which started the whole discussion? Here is a fineexample
of the delicate patternswhich make the mathematicsof the period so magical.
From the complex point of view, a partial reason for the identitylies in the
similaritybetween zeros of the sine and the poles of the gamma function.
There is the duplication formula
(25)

r(2z)

+ -1)
(2ir)-1I222112Fr(z)r(z

discovered by Legendre and extended by Gauss in his researcheson the hypergeometricfunctionto the multiplicationformula
(26) r(nz) = (2ir)12(1-n)nnz-l12Pr(z)

)r

(z +1)

(Z

f-

1)

There are prettyformulasforthe derivativesof the gamma functionsuch as

(27)

d2 log r(z)/dz2

1
-

z2

1
(Z +1) 2

+
(z +2) 2

This is an example of a type of infiniteseries out of which G. Mittag-Leffler

(1846-1927) later created his theoryof partial fractiondevelopmentsof meromorphicfunctions.There is the intimaterelationshipbetween the gamma function and the zeta functionwhich has been of fundamentalimportancein studying the distributionof the prime numbers,

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

861

LEONHARtD EULER'S INTEGRAL

1959]

t(z) = t(-

(28)

z)r(1

z)2irz-1sin .6z,

where
(29)

~(Z)

2z

3z1

This formulahas some interestinghistoryrelated to it. It was firstproved by

Riemann in 1859 and was conventionallyattributedto him. Yet in 1894 it was
discoveredthat a modifiedversionof the identityappears in some workof Euler
which had been done in 1749. Euler did not claim to have proved the formula.
However, he "verified"it forintegers,for 2, and for3/2. The verificationfor 2
is by directsubstitution,but forall the othervalues, Euler workswithdivergent
infiniteseries.This was more than 100 years in advance of a firmtheoryof such
series, but with unerringintuition,he proceeded to sum them by what is now
called the method of Abel summation. The case 3/2 is even more interesting.
There, invokingboth divergentseries and numerical evaluation, he came out
with numericalagreementto 5 decimal places! All this work convinced him of
the truthof his identity.Rigorous modern proofsdo not require the theoryof
divergentseries,but the notions of analytic continuationare crucial.
In view of the essential unityof the gamma functionover the whole complex
plane it is theoreticallyand aesthetically importantto have a formulawhich
worksforall complex numbers.One such formulawas supplied in 1848 by F. W.
Newman:
(30) i/r(z) = ze't{ (1 + z)e-z} { (1 + z/2)ez12}

wherey= .57721 56649 . X

*,

This formulais essentiallya factorizationof 1/r(z) and is much the same as a

factorizationof polynomials. It exhibits clearly where the functionvanishes.
Setting each factor equal to zero we findthat 1/r1(z)is zero forz =0, z = -1,
z = -2, * - . In the hands of Weierstrass,it became the startingpoint of his
particulardiscussion of the gamma function.Weierstrasswas interestedin how
functionsotherthan polynomialsmay be factored.A numberof isolated factorizations were then known. Newman's formula (30) and the older factorization
of the sine
.

(31)

sin rz = rz(1

Z2)(1

are anaong them. The factorizationof polynomialsis largelyan algebraic matter

but the extensionto functionssuch as the sine which have an infinityof roots
required the systematic building up of a theory of infiniteproducts. In 1876
Weierstrasssucceeded in producingan extensivetheoryof factorizationswhich
included as special cases these well-knowninfiniteproducts, as well as certain
doubly periodic functions.
In addition to showing the roots of 1/P(z), formula (30) does much more.

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

862

[December

LEONHARD EULER S INTEGRAL

It shows immediatelythat the reciprocalof the gamma functionis a much less

difficultfunctionto deal with than the gamma functionitself. It is an entire
function,that is, one of those distinguishedfunctionswhich possesses no singularitieswhatever forfinitearguments.Weierstrasswas so struckby the advantages to be gained by startingwith 1/r(z) that he introduceda special notation
forit. He called 1/r(u+1) thefactorielleof u and wrote Fc(u).
The theoryof functionsof a complexvariable unifiesa hotch-potchof curves
and a patchwork of methods. Within this theory, with its highly developed
studies of infiniteseries of various types, was broughtto fruitionStirling's unsuccessful attempts at solving the interpolation problem for the factorials.
Stirling had done considerable work with infiniteseries of the form

A+Bz+Cz(z-1)+Dz(z-1)(z-2)+

This series is particularlyuseful for fittingpolynomialsto values given at the

integersz=0, 1, 2, * - - . The method of findingthe coefficientsA, B, C, . . .
was well known. But when an infiniteamount of fittingis required,much more
than simple formalwork is needed, for we are then dealing with a bona fide
infiniteseries whose convergencemust be investigated.Startingfromthe series
1, 2, 6, 24, .*. .Stirling found interpolatingpolynomialsvia the above series.
The resultantinfiniteseries is divergent.The factorialsgrowtoo rapidly in size.
Stirlingrealized this and put out the suggestionthat ifperhaps one started with
the logarithmsof the factorials instead of the factorials themselves the size
forone to do something.There the matterrested
mightbe cut down sufficiently
until 1900 when Charles Hermite (1822-1901) wrotedown the Stirlingseries for

logr(i +z):
(32)

log r( + z) =

z(z-1)

_____l)(z-2

log 2 +

1z2-3

(log3-21lg2)+

and showed that this identityis valid wheneverz is a complex number of the
forma +ib with a>0. The identity itself could have been writtendown by
Stirling,but the proofwould have been another matter.An even simplerstarting point is the function41(z)= (d/dz) log r(z), now known as the digamma or
psi function.This leads to the Stirlingseries

(33)

log r(z)
dz

(z

1)

(z

(z1) (z -2)
+
.!
2*2

1)(z- 2)(z3.3!

3)

which in 1847 was proved convergentfor a>0 by M. A. Stern, a teacher of

Riemann. All these mattersare today special cases of the extensivetheoryof the
convergenceof interpolationseries.
Functions are the buildingblocks of mathematicalanalysis. In the 18th and
19th centuriesmathematiciansdevoted much time and loving care to develop-

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

LEONHARD EULER S INTEGRAL

863

ing the properties and interrelationshipsbetween special functions. Powers,

roots, algebraic functions,trigonometricfunctions,exponentialfunctions,logarithmicfunctions,the gamma function,the beta function,the hypergeometric
function,the elliptic functions,the theta function,the Bessel function,the
Matheiu function,the Weber function,Struve function,the Airy function,
Lame functions,literally hundreds of special functionswere singled out for
scrutinyand their main featureswere drawn. This is an art which is not much
cultivated these days. Times have changed and emphasis has shifted.Mathematicians on the whole prefermore abstract fare. Large classes of functionsare
studied instead of individual ones. Sociology has replaced biography.The field
of special functions,as it is now known, is left largely to a small but ardent
group of enthusiastsplus those whose work in physicsor engineeringconfronts
them directlywith the necessityof dealing with such matters.
The early 1950's saw the publication of some very extensive computations
of the gamma functionin the complex plane. Led offin 1950 by a six-place table
computed in England, it was followedin Russia by the publicationof a very extensive six-place table. This in turnwas followedin 1954 by the publication by
the National Bureau of Standards in Washingtonof a twelve-placetable. Other
publications of the complex gamma functionand related functionshave appeared in this country,in England, and in Japan. In the past, the major computations of the gamma functionhad been confinedto real values. Two fine
tables, one by Gauss in 1813 and one by Legendre in 1825, seemed to answerthe
mathematical needs of a century.Modern technologyhad also caught up with
the gamma function.The tables of the 1800's-were computed laboriously by
hand, and the recentones by electronicdigital computers.
But what touched offthis spate of computational activity? Until the initial
labors of H. T. Davis of Indiana Universityin the early 1930's, the complex
values of the gamma functionhad hardly been touched. It was one of those
curious turns of events whereinthe complex gamma functionappeared in the
solution of various theoreticalproblemsof atomic and nuclear theory. For instance, the radial wave functionsforpositive energystates in a Coulomb field
leads to a differential
equation whose solutioninvolvesthe complexgamma function. The complex gamma functionenters into formulasfor the scatteringof
chargedparticles,forthe nuclearforcesbetweenprotons,in Fermi's approximate
formulaforthe probabilityof :-radiation, and in many other places. The importanceof these problemsto physicistshas had the side effectof computational
mathematicsfinallycatchingup withtwo and a quarter centuriesof theoretical
development.
As analysis grew,both creatingspecial functionsand delineatingwide classes
of functions,various classificationswere used in orderto organize them forpurposes of convenientstudy. The earliermathematiciansorganizedfunctionsfrom
without,operationally,asking what operations of arithmeticor calculus had to
be performedin orderto achieve them.Today, thereis a much greatertendency
to look at functionsfromwithin,organically,consideringtheirconstructionas

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

864

LEONHARD EULER S INTEGRAL

[December

achieved and asking what geometricalcharacteristicstheypossess. In the earlier

classificationwe have at the lowest and most accessible level, powers,roots,and
all that could be concocted from them by ordinary algebraic manipulation.
These came to be knownas algebraic functions.The calculus, with its characteristic operation of taking limits, introduced logarithmsand exponentials, the
latter encompassing,as Euler showed, the sines and cosines of trigonometry
which had been available fromearlier periods of discovery.There is an impassable wall between the algebraic functionsand the new limit-derivedones. This
wall consistsin the fact that tryas one mightto construct,say, a trigonometric
functionout of the finitematerial of algebra, one cannot succeed. In more
technical language, the algebraic functionsare closed with respect to the processes of algebra, and the trigonometricfunctionsare foreverbeyond its pale.
(By way of a simple analogy: the even integersare closed with respect to the
operations of addition, subtraction,and multiplication;you cannot produce an
odd integerfromthe set of even integersusing these tools.) This led to the concept of transcendentalfunctions.These are functionswhich are not algebraic.
The transcendentalfunctionscount among their members,the trigonometric
functions,the logarithms,the exponentials,the ellipticfunctions,in short,practicallyall the special functionswhichhad been singledout forspecial study. But
such an indescriminatedumping produced too large a class to handle. The
transcendentalshad to be split furtherforconvenience.A major tool of analysis
equation, expressingthe relationshipbetween a functionand
is the differential
its rate of growth.It was foundthat some functions,say the trigonometricfunctions,althoughtheyare transcendentaland do not thereforesatisfyan algebraic
are algeequation, nonethelesssatisfya differentialequation whose coefficients
braic. The solutions of algebraic differentialequations are an extensive though
not all-encompassingclass of transcendentalfunctions.They count among their
members a good many of the special functionswhich arise in mathematical
physics.
Where does the gamma functionfitinto this? It is not an algebraic function.
This was recognizedearly. It is a transcendentalfunction.But fora long while
it was an open question whether the gamma functionsatisfied an algebraic
differential
equation. The question was settled negativelyin 1887 by 0. Holder
(1859-1937). It does not. It is of a higherorder of transcendency.It is a socalled transcendentallytranscendentfunction,unreachable by solvingalgebraic
equations.
equations, and equally unreachable by solving algebraic differential
in
Alexander
interested
the
and
1925
has
years
through
many
people
The subject
Switzerland,
gave
Professor
Emeritus
of
the
of
Basel,
now
University
Ostrowski,
theorem.
of
an alternate proof H6lder's
Problems of classificationare extremelydifficultto handle. Consider, for
instance, the following:Can the equation X7+ 8x + 1 be solved with radicals?
+1) be found in termsof specifiedelemenIs ir transcendental?Can fdx/X/(x3
Can
the
differential
equation dy/dx= (1/x) + (l/y) be resolved
tary functions?
The
with quadratures?
general problemsof which these are representativesare

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

LEONHARD

EULER'S

865

INTEGRAL

even today far fromsolved and this despite famous theories such as Galois
Theory, Lie theory,theoryof Abelian integralswhich have derived fromsuch
simple questions. Each individual problemmay be a one-shotaffairto be solved
by individual methods involvingincredibleingenuity.
HADAMARD'S FACTORIAL FUNCTION
6

FIG. 4

Thereare infini'tely
whichproducefactorilals.
The function
manyfunctions
=
F(x)
(6/r(1

x)) (d/dx) log {r((

xA
x)/2)/r(

is an ent'ireanalyticfunctilon
whichcoincideswiththe gammafunctionat the pos'itiveintegers.
It satisfiesthefunctional
equationF(x +1) = xF(x) +(1 /rJi(- x)).

We returnonce again to our interpolationproblem. We have shown how,

stri'ctlyspeaking,thereare an unlimitednumberof solutionsto thi'sproblem.To
drive this point home, we mightmention a curious solution given in 1894 by
). Hadamard found a relativelysilmpleformula
Jacques Hadamard (1865involvingthe gamma functionwhich also produces factorialvalues at the positive integers.(See F'igs. 1 and 4.) But Hadamard's function
(34)

1
d
(iog=(lr-x)

in strongcontrastto the gamma functionitself,possesses no singularitiies

anywhere in the finitecomplex plane. It is an entire analytic solution to the interpolation problem and hence, fromthe functiontheoretic point of view, is a
simplersolution. In view of all this ambiguity,why then should Euler's solution

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

866

INTEGRAL

LEONHARD EULER'S

[December

be consideredthe solution par excellence?

From the point of view of integrals,the answer is clear. Euler's integral
appears everywhereand is inextricablybound to a host of special functions.Its
frequencyand simplicitymake it fundamental.When the chips are down, it is
the very formof the integraland of its modificationswhich lend it utilityand
importance. For the interpolatorypoint of view, we can make no such claim.
We must take a deeper look at the gamma functionand show that of all the
solutions of the interpolationproblem,it, in some sense, is the simplest.This is
partiallya matterof mathematicalaesthetics.
FUNCTION
A PSEUDOGAMMA

lo
82

_-

. __

-_

- _

_ _--

-_

FIG. 5

equationof the gamma

satisfiesthe functional
producesfactorials,
The functionillustrated
and is convex.
function,

We have already observed that Euler's integral satisfiesthe fundamental

recurrenceequation, xr(x) = r (x+1), and that this equation enables us to compute all the real values of the gamma functionfromknowledge merelyof its
values in the intervalfrom0 to 1. Since the solutionto the interpolationproblem
is not determineduniquely, it makes sense to add to the problem more conditions and to inquire whetherthe augmented problem then possesses a unique
solution. If it does, we will hope that the solution coincides with Euler's. The
recurrencerelationshipis a natural conditionto add. If we do so, we findthat
the gamma functionis again not the only functionwhichsatisfiesthis recurrence

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

LEONHARD EULER' S INTEGRAL

867

relation and produces factorials.One may easily constructa "pseudo" gamma

functionrs(x) by definingit between, say, 1 and 2 in any way at all (subject
only to rs(l) = 1, rP(2) = 1), and allowingthe recurrencerelationshipto extend
its values everywhereelse.
If, for instance, we let rs(x) be 1 everywherebetween 1 and 2, the recurrence relation leads us to the function(see Fig. 5).

(35)

rs(x) = l/x

O < x < 1;

rs(x) = 11
rs(x) = x -1,

1 < x :! 2;
2?x<3;

rs(x)= (x - 1)(x - 2), 3 < x <4;**-.

We mightend up with a fairlyweird result,depending upon what we start

with. Even if we require the final result to be an analytic function,there are
ways of doing it. For instance, take any functionwhich is both analytic and
periodic with period 1. Call it p(x). Make sure that p(1) =1. The function
1 +sin 27rxwill do for p(x). Now multiplythe ordinarygamma functionrI(x)
by p(x) and the result r(x)p(x) will be a "pseudo" gamma functionwhich is
analytic, satisfiesthe recurrencerelation, and produces factorials! Thus, we
still do not have enough conditions.We must augment the problemagain. But
what to add?
By the middle of the 19th century it was recognized that Euler's gamma
functionwas the only continuous functionwhich satisfiedsimultaneouslythe
recurrencerelationship,the reflectionformulaand the multiplicationformula.
Weierstrasslater showed that the gamma functionwas the only continuoussoluforwhich{r(x +n) } /{ (n - 1)nx} ->1 forall
tionof the recurrence
relationship
x. These conditionsadded to the interpolationproblemwill serve to produce a
unique solutionand one whichcoincideswithEuler's. But theyappear too heavy
and too much like Monday morningquarterbacking.That is to say, the added
conditionsare hardly "natural" forthey are tied in with the deeper analytical
propertiesof the gamma function.The search went on.
Aesthetic conditionswere not to be found in the older, analytic considerations, but in a newer, inner, organic approach to functiontheory which was
developing at the turn of the century. Backed up by Cantor's set theoryand
an emergingtheoryof topology,the new functiontheorylooked not so much
at equations and identitiesas at the fundamentalgeometricalproperties.The
desired condition was found in notions of convexity. A curve is convex if the
is true of it: take any two points on the curve and join them by a
follow,ing
straightline; then the portion of the curve between the points lies below the
line. A convex curve does not wiggle; it cannot look like a camel's back. At the
turn of the century,convexitywas in the mathematicalair. It was foundto be
intrinsicto many diverse phenomena. Over the period of a generation,it was
sought out, it was generalized, it was abstracted, it was investigated for its
own sake, it was applied. Called to attentionby the work of 1I. Brunn in 1887

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

868

LEONHARD EULER S INTEGRAL

[December

and of H. Minkowskiin 1903 on convex bodies and given an independentinterest in 1906 by the work of J. L. W. V. Jensen,the idea of convexityspread and
established itselfin mean value theory,in potential theory,in topology, and
most recently in game theory and linear programming.At the turn of the
centurythen, an application of convexityto the gamma functionwould have
been natural and in order.
The individual curves which make up the gamma functionare all convex.
A glance at Figure 2 shows this to be true. If, as in the previous paragraph, a
pseudogamma function satisfyingthe recurrenceformula were produced by
introducingthe ripple I+sin 2-rxas a factor,it would no longer be true. It
must have occurred to many mathematiciansto findout whetherthe gamma
functionis the only functionwhich yields the factorialvalues, satisfiesthe recurrencerelation,and is convex downwardforx>0. Unfortunately,this is not
true. Figure 5 shows a pseudogamma functionwhich possesses just these properties. It remained until 1922 to discover a correctformulation.But it was not
at too fara distance. The gamma functionis not only convex, it is also logarithmically convex. That is to say, the graph of log r(x) is also convex down for
x>0. This fact is implicitin formula(27). Logarithmicconvexityis a stronger
conditionthan ordinaryconvexityforlogarithmicconvexityimplies,but is not
implied by, ordinaryconvexity. Now Harald Bohr and J. Mollerup were able
to show the surprisingfact that the gamma functionis the only functionwhich
satisfiesthe recurrencerelationshipand is logarithmicallyconvex. The original
proofwas simplifiedseveral years later by Emil Artin,now professorat Princeton University,and the theorem togetherwith Artin's method of proof now
constitutethe Bohr-Mollerup-Artintheorem.Its precise wordingis this:
The Euler gammafunctionis theonlyfunctiondefinedfor x > 0 whichis positive,is 1 at x = 1, satisfiesthefunctionalequationxr(x) = r(x + 1), and is logarithmicallyconvex.
This theoremis at once so strikingand so satisfyingthat the contemporary
synod of abstractionistswho write mathematical canon under the pen name
of N. Bourbaki has adopted it as the starting point for its exposition of the
gamma function.The proof:one page; the discovery: 193 years.
There is much that we know about the gamma function.Since Euler's day
more than 400 major papers relatingto it have been written.But a fewthings
remainthat we do not know and that we would like to know. Perhaps the hardest of the unsolved problems deal with questions of rationalityand transcenwhich appears in
dentality.Consider, forinstance,the number y=.57721 ...
expressions
form'fula
(30). This is the Euler-Mascheroniconstant. Many different
can be given forit. Thus,

(36)

y=

(37)

7 =

dr(x)/dxjX1,
ne

1+ - - +

~2

n.
n/ log

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions

1959]

EQUATIONS WITH CONSTANT COEFFICIENTS

869

Though the numericalvalue of yis known to hundredsof decimal places, it is

not known at the time of writingwhethery is or is not a rational number.Another problem of this sort deals with the values of the gamma functionitself.
Though, curiouslyenough, the product F(1/4)/1Yw can be proved to be transcendental, it is not known whetherF(1/4) is even rational.
George Gamow, the distinguishedphysicist,quotes Laplace as saying that
when the known areas of a subject expand, so also do its frontiers.Laplace
evidently had in mind the picture of a circle expanding in an infiniteplane.
Gamow disputes this forphysicsand has in mind the pictureof a circle expanding on a sphericalsurface.As the circleexpands, its boundary firstexpands, but
later contracts. This writeragrees with Gamow as far as mathematics is concerned. Yet the record is this: each generationhas found somethingof interest
to say about the gamma function.Perhaps the next generationwill also.
and
and criticism
C. Truesdellforhishelpfulcomments
The writerwishesto thankProfessor
Dr. H. E. Salzerfora numberofvaluablereferences.
References
Leipzig,1931.
in die TheoriederGammafunktion,
1. E. Artin,Einffuhrung
Book IV, Ch. VII, La FonctionGamma,Paris,
de Mathematique,
2. N. Bourbaki,D16ments

1951.

Indiana,
3. H. T. Davis, Tables oftheHigherMathematicalFunctions,vol. I, Bloomington,
1933.
1924.
4. L. Euler,Operaomnia,vol. I14,Leipzig-Berlin,
Math6matiqueet Physiquede QuelquesC6lbbresGe6S. P. H. Fuss, Ed., Correspondance
1843:
du XVIIIiemeSiecle,Tome I, St. Petersbourg,
mbtres
6. G. H. Hardy,DivergentSeries,Oxford,1949,Ch. II.
Leipzig,1951.
Funktionen,
7. F. L6schand F. Schoblik,Die Fakultatundverwandte
Leipzig,1906.
8. N. Nielsen,HandbuchderTheorieder Gammafunktion,
NationalBureau of Standards,
9. Table of the Gamma FunctionforComplexArguments,
by HerbertE. Saizer.)
1954.(Introduction
AppliedMath. Ser. 34,Washington,
and G. N. Watson,A CourseofModernAnalysis,Cambridge,1947,Ch.
10. E. T. Whittaker
12.

LINEAR DIFFERENTIAL OR DIFFERENCE EQUATIONS WITH

CONSTANT COEFFICIENTS
H. L. TURRITTIN, Instituteof Technology,Universityof Minnesota

1., Introduction.*Solutions of a system of linear differentialor difference

equations with real constant coefficientsai1, such as
(1)

dxd/dt=

jl1

aijxj and xi(t + h) =E

ji-

aixj(t),

underUSAF contractNo. AF 33(038)22893

* This paperwas prepared
in partwhileworking
underQOR contractNo. DA-11-022-ORD-2042.
and in partwhileworking

This content downloaded from 128.111.121.42 on Wed, 31 Jul 2013 22:45:17 PM

All use subject to JSTOR Terms and Conditions