J. J. SMITH (Schenectady - U. S. A.

HEAVISIDE'S OPERATORS AND CONTOUR INTEGRALS

In recent years the Operational Methods of Heaviside have been given a

considerable amount of attention. At least two books (*) and several papers
have appeared which have for their purpose the development of the subject. A
bibMography is given in each of the books referred to, and hence need not be
given here.
The term « operational methods » as appMed to Heaviside's theory may be
divided into two distinct classes:
1°) Operations involving only whole integrations or differentations ;
2°) Operations involving fractional integrations and integrations.
Of the latter Mttle wül be said in the present paper. It is probably the most
obscure and least understood part of Heaviside. It is curious to note, however,
that Heaviside himself plunges the reader directly into this part of the subject
at the beginning of the second volume in which he takes up the discussion of
operational methods. Carson lays more stress on the appMcation to fractional
differentiation and his discussion of the Expansion Theorem seems to show
that he regards it as an interesting result rather than as something of practical
value. Jeffreys also largely appMes his discussion to fractional operators.
There is, of course, a reason for the greater interest in fractional operators
from the practical point of view.
Consider a transmission Mne of length I initiaMy in a state of rest. Let one
one end be suddenly raised to a potential E while the other end is kept at
zero potential. The flow of electricity in such a Mne can be developed as a series
of normal functions, and in the operational method no impMcit operations of
fractional differentiation arise. If, however, the length of the line is assumed to
become infinite the series of normal functions can be written as a Fourier integral,
and on performing the integration the variable of integration disappears. It also
happens in such cases that the operational method of solving the problem involves
operations of fractional differentiation. In other words, fractional differentiation

(*) CARSON: Electric Circuit Theory and Operational Calculus, McGraw Hill, 1926. —
JEFFREYS: Operational Methods in Mathematical Physics, Camb. Univ. Press, 1927.
310 COMUNICAZIONI

arises in such problems in operational methods, as far as experience has shown

to date, when and only when a boundary or the boundaries are at an infinite
distance.
For the purpose of actual numerical computation for, let us say, the flow
of electricity in a finite Mne, it is often found that the simplest procedure is to
obtain the solution for an infinite Mne. This solution which is generaMy caMed
the wave solution, to distinguish it from the solution in terms of normal functions,
may be used up to the time the first wave reaches the end of the cable. From
the terminal conditions the value of the reflected wave can be computed, and
this togethen with the original wave can be used up to the time at which the
reflected wave reached the other end of the cable, and so on. The use of the
solution for the infinite cable in this manner is, however, purely a question of
a high order of convergence and faciMty in numerical work. It has no advantage,
however, from a formal point of view since both this solution and the solution
in terms of normal functions are made up of infinite series.
It has always seemed to the author that practicaMy aM that is known at
present about the subject of fractional operators may be written in the Heaviside
pnxml\n=xm~n/\m—n

m—n where n mai have any value positive or negative. When operations of
the type n n n n
f(Ìp + a ) or f(ip + A ),
where p denotes differentiation with respect to one variable, A differentiation
with respect to another variable and a denotes a constant, are involved, the
operational methods become so cumbersome that they are in many cases impracti-
cable to use. In such cases it wiM generaMy be found that solutions can be
fairly readily obtained in terms of normal functions when the boundaries are
assumed first to be at a finite distance. The Mmit of such a series may then
be taken as the boundaries approach infinity when the series of normal functions
wiM be transformed into an infinite integral. The evaluation of such an integral
wMl generaMy be found to be at least as simple a matter, if not in many cases
more simple than the evaluation of the fractional operator obtained by assuming
at the outset that the boundaries are at an infinite distance. When a solution
is thus obtained for infinite boundaries it may be appMed for numerical compu-
tation, as described above, in the case where the boundaries are at a finite
distance.
Having pointed out how it is possible, though perhaps not always desirable
from the point of view of brevity, to avoid the introduction of fractional operators
they wiM not be discussed further in this paper. They are very interesting from
many points of view and deserving of much more attention than is given them
at present. Some of the more fundamental aspects of the operational methods
 J. J. S M I T H : Heaviside'S operators and contour integrals 311

which occur in Heaviside arise, however, in connection with whole differentiations

and can, therefore, be discussed more readily in this region.
The mere technique of handling Mnear symbolic operators is not a matter
of great concern, except insofar as the use of these operators may be shown
to have significance. It is a mathematical common-place that the differential
operation — may be treated as a constant with respect to certain rational
operations and in connection with certain types of functions of x. The use of
symboMc operators in the solution of linear differential equations with constant
coefficients is given in most textbooks. Several different appMcations are usuaMy
made, such as the breaking up into partial fractions of a given inverse operator
and the expansion into infinite series of expressions involving the differential
operator in the denominator. These symboMc methods are sometimes caMed into
question by critics of mathematical rigor and most textbooks do not attempt to
estabMsh definitely the conditions under which some of these expansions can
be assured to give a convergent result.
The utüity of operational methods to discover a possible final form for a
proposed answer cannot be questioned. The final form secured by their use may
be tested by direct substitution and in aM of the usual cases, where fractional
differentiation is not involved, is found to provide an actual solution of the
problem. If by sentable, ingenious guesses the same solution should be arrived
at by some other method it would stiM be necessary to substitute the proposed
expression into the given equation.
Many mathematical writers who criticise operational methods adopt the even
less justifiable process of asking the reader to consider a given expression.
This given expression is written down without any hint as to its source. By
various unmotivated steps, the reader is finaMy in the position to verify that
an expression obtained by foMowing the steps laid out by the author satisfies
a differential equation, but the relation between the original expression and the
differential equation is never cleared up.
In Heaviside's method certain further steps are taken. These are for the
purpose of leading in a simple and direct manner to the possible expression
which as in the simpler cases is actuaMy found to satisfy a given equation.
The importance of these methods can best be appreciated perhaps by considering
the character of the boundary conditions as related to a given differential equation.
In the earMer stages of the historic study of mathematical analysis the boundary
conditions were regarded as relatively trivial incidentals in the study of a dif-
ferential equation. What was desired was a so-caMed general solution. The problem
of inserting suitable boundary conditions was left as an obvious exercise for
the reader. The study of these differential equations was supposed to be imme-
diately appMcable to physical problems, for it was pointed out that the functions
required in given physical problems may be shown to satisfy differential equa-
312 COMUNICAZIONI

tions of the type studied. This method has its Mmitations even in the case of
one dimension.
In the case of partial differential equations of physics involving two or more
dimensions the older point of view was soon discovered to be out of harmony
with the physical problems. In the most important physical problems the partial
differential equation is a very simple type. The solution involves infinitely many
parameters. The so-caMed general solution has no physical interest. The entire
difficulty of the mathematical problem in connection with given physical phenomena
consists in incorporating the given boundary conditions into the problem in
such a way as to identify the unique solution desired. The theory of Fourier's
series in its appMcation to the problem of the flow of heat shows a marked
departure from the traditional methods of study of differential equations of the
old school. It is to Heaviside's credit that in connection with the problems arising
from electrical trasmission Mnes he reaMzed the importance and the possibiMty
of a further step, which was to incorporate the boundary conditions in the
practical construction of a solution. Suppose we face the problem of determining
the actual character of the electric current at a given time in a given circuit.
There are two parts to the problem. We may suppose that the condition is a
permanent one and so-caMed transient effects have been damped out. In this
case a solution may be obtained vaMd from t= — oo to £ = + oo. For many
purposes, however, it is much more important to determine the transient effects
which are ultimately damped out and which arise when the current is started
in the circuit. What might be caMed the general solution involves both parts. It
is convenient, therefore, to assume that previous to a given instant of time,
say #=0, the voltage apphed to the circuit is zero. As a result of the given
configuration it is desired to determine the current at some subsequent time.
The conventional procedure of estabhshing a single differential equation of
simple analytical form and seeking its continuous solution is of only partial
value because it wiM give us merely a steady state and wMl give us the transient
only when we use proper boundary conditions. The flow of current is completely
determined by the physical data and it seems natural to seek a direct method
which wiM take into consideration the voltage which is not appMed untM the
instant t==0. Heaviside's method provides the means. Consider the simple function
which is zero from t= — oo to t=0 and which is equal numericaMy to t for t
positive. The function wiM be supposed to be continuous at t=0 and taking on
there the value 0. It has a derivative at every point except # = 0 and its second
derivative, where it exists, is zero. One might be incMned to say that this function
satisfies the equation f"(t) = 0. The function, however, does not satisfy a single
equation of the form f(t)=a + bt. The difficulty occurs on account of the ambi-
guity arising at the point £ = 0 . In this case and in aM of the analogous cases
here discussed the desired solution is 0 for t<0 and behaves in an assigned
 J. J. SMITH: Heaviside's operators and contour integrals 313

fashion for t>0. The problem consists in identifying its behavior for t>0 in
a simple manner.
The expansion of functions in terms of given assigned elementary functions
is standard practice throughout mathematical analysis. In the classical theory
the elementary functions are themselves analytic while the function which is
expanded may or may not have numerous discontinuities. For the present discussion
the elementary functions wiM be taken as having discontinuity for a derivative
of some assigned order. If we take the simple function which is 1 for t positive,
but 0 for t negative and whose behavior at £ = 0 may be discussed later, this
function when multipMed by successive powers of t affords the convenient sequence
of elementary functions for certain simple problems. Any function expanded in
terms of this sequence wiM vanish identicaMy for t negative, and may take on
the character of an assigned analytical function for t positive. In applying
differential methods to these functions it is desirable to introduce other functions
which may be at first regarded as purely symboMc in the sense that, although
they are identified at aM points other than t=0 by definite values, at t=0 the
representation becomes symboMc in the sense that reference may be made to
the genesis of the function. This wiM become clear when we take up the discussion
in detail. We may briefly summarize the present aspect of Heaviside's theory
by saying that linear differential equations with constant coefficients are used.
The solutions of these are expanded in terms of a sequence of elementary functions,
certain of which may be regarded from some points of view as symboMc rather
than actual. The boundary conditions can be incorporated into the problem in
such a way that although the solution wiM be known beforehand to vanish
identicaMy for t negative (in the simple cases) yet the behavior for t positive
wiM be completely determined in the direct process of the work and without
appeal to auxiMary conditions to be imposed by the reader as best he may at
the conclusion of the discussion.
It may be pointed out that although the books of Carson and Jeffreys,
referred to above, have been caMed by their authors works on operational calculus,
they are, strictly speaking, works on how to avoid the operational calculus.
Carson bases his work on the development of a solution of a certain integral
equation. Jeffreys' work is the appMcation of a contour integral to the solution
of problems in mathematical physics. It is very difficult to consider either of
these methods as being the same as the operational method used by Heaviside.
That Jeffreys seems to hold the same view with regard to the contour integral
method may be seen from the foMowing quotations from the introduction to his
book:
« It is often said that it (the operational method) wiM solve no problem that
cannot be solved otherwise. Whether this is true it would be difficult to say;
but it is certain that in a large class of cases the operational method wiM give
314 COMUNICAZIONI

the answer in a page when ordinary methods take five pages, and also that it
gives the correct answer when the ordinary methods, through human faMibüity,
are Mable to give the wrong one ».
And further on we read:
« After aM, the use of contour integrals in this connection was introduced
by Bromwich who has repeatedly declared that the direct operational metod of
solution is the better of the two ».
A distinct Mmitation encountered in using either of these substitutes for the
operational method is that, as far as the author is aware, they only lend themselves
to the solution of problems involving Laplace's equation where but one space
coordinate in addition to the time is involved, or possibly where two space
coordinates and no time coordinate are involved. This is on account of the
complexity of the integrals which occur when more coordinates are present. The
operational method, on the other hand, has proved itself adaptable to cases in
which one, two, three or any number of space coordinates are present in addition
to the time. It can also be appMed with equal faciMty to problems in potential
theory or other branches of mathematical physics in which the time coordinate
does not enter.
In order to get a better conception of the problems of the Heaviside operational
calculus in its relation to the Heaviside unit function, it seems desirable to
compare the operational results obtained by the method of contour integrals
which are stated clearly by Jeffreys (loc. cit.). Certain problems wiM arise in
this discussion on the contour integral method which do not appear to have
been given sufficient serious consideration. While, from a practical standpoint,
a criticism on the score of inconsistency in singular cases may not be very
important, and should in no sense be considered as a rejection of the results
obtained in cases which do not involve these singularities, past history has
shown that an examination of such inconsistencies has, in general, led to impro-
vement in methods and generaMzation of results.
On page 19 of Jeffreys' book the foMowing results are given with sMght
changes in notation :
(D " «P-^)-^/; dx
Z*+i

(where the contour is along a Mne parallel to and on the positive side of the
y axis, the contour being completed by a circle of infinite radius which encloses the
negative half of the plane). The only pole of the integrand is at the origin where
xnn
the residue is x— Hence when n is a positive integer
n!
n
(2) p-H(x)=xn[
also we see that
(3) pnH(x)=0
 J. J. SMITH: Heaviside's operators and contour integrals 315

since the integrand has no singularity in the finite part of the plane; and
(4) p°H(x) = l».
A Mttle later on page 25, Jeffreys shows : « when m is fractional

(5) P-HW-rW+i,
and since when m is a positive integer JH(ra + l ) = m ! and when m is a negative
integer r(m + l) is infinite, the interpretation already adopted (given above) for
integral powers of p are special cases of these ».
xn
When m is a negative integer it is undoubtedly true that p , is zero
provided x 4=0; when x=0, however, the result is by no means obvious. Let
us see what Heaviside has to say about it.
On referring to Proc. Roy. Soc. Vol. LU, p. 520, the foMowing is found
(replacing V in Heaviside by p in order to make the notations similar) :
« 18. - It should be noted that when we say that

(6) *"£ = 1
X°
for aM values of n, the right-hand member is reaMy ^ and means 0 on the
left and 1 on then right side of the origin of x. That is, it is the Mmiting form
x
of the function — when n is infinitely smaM positive. It is convenient in the
treatment of the form F= Yf to have the function f zero up to a certain point,
XQ x—i
and then begin to act. Similarly the expression p — or p i or r—-, although it
has the value zero for aM positive values of x is infinite at the origin. But its
total amount is finite viz. 1. Imagine the unit amount of a quantity spread
along an infinitely long line to become aM massed at the origin. Its Mnear density
wiM, in the limit, be represented by
QO

(7) pl= - cos mxdm.

b
It is zero except at x=0. But its integral is stiM finite, being p°l or 1. If
xn
we draw the curve y= — with n infinitely smaM, consisting of two straight
Mnes, with a rounded corner, the curve derived from it by one differentiation
wiM nearly represent the function pi, being nearly aM heaped up close to the
origin, and of integral amount 1. Similarly p2l means a double infinite point,
p3l a triple infinite point, and so on. But it is the function pi that is most
useful in connection with differentiating operations, while the others are less
prominent ».
xn
Allowing for the doubtful vaMdity of his assumption that — is zero when x < 0,
316 COMUNICAZIONI

it should be evident from the above that Heaviside's view of p i is a very different
conception from that of Jeffreys'. Heaviside's conception, if I understand it correctly,
is that H(x) or 1, which is caMed the
Heaviside unit function, can be repre-
sented by the curve shown in Fig. 1, or
H(x)=0 x<0
Fig. 1. (8) I H(x) = l x>0
0^H(x)^l x=0
Jeffreys' conception seems to be as shown in Fig. 2, or
S H(x)=0 x<0
(9)
I H(x) = l x>0
and the value of H(x) is either unde-
fined or defined as an arbitrary value
at x=0. Fig. 2.
Heaviside's conception of pi can be
represented by the curve shown in Fig. 3, or
pH(x)=0 x<0
(10) pH(x)=0 x>0
0^pH(x)^.oo x=0
in such a way that
(11) j pH(x)dx=

Jeffreys' conception of pi, as I in-

terpret it from his words, can be
represented by the horizontal axis.
Not only do we have
pH(x) = 0 x<0
(12)
pH(x)=0 x>0
but the value of pH(x) is given the
expMcit value at x=0.
Of the two conceptions I beMeve
Fig. 3. that Heaviside's is the only one which
can serve for the purpose intended.
In a series of papers on the Theory of iJ-Functions in the Journal of the
FrankMn Institute Oct., Nov., Dec. 1925 (A), I have discussed Heaviside's con-

(1) These papers will hereafter be referred to as H — F.

 J. J. SMITH: Heaviside's operators and contour integrals 317

ception from a sMghtly different viewpoint, using a modified definition for a

derivative given in those articles. I have been told since that Fréchet has given
a simMar definition for a derivative although I have not seen his work. In
the H—F papers it has been shown
that by using p i as a definite function JC=0
having the properties given in (10) pig. 4.
and (11) above, the derivation of the
Green's Functions in an expansion of normal functions in Mathematical Physics
for almost any type of space becomes an extremely simple operation, and re-
solves itself into a sMght extension of the Heaviside Expansion Theorem. It is
also shown there that to correctly write the Equation of Heat for the instan-
taneous Green's Function in Heat problems or for a similar case in electrical
problems, the function pi, or should appear in this equation. It is im-
possible to go into this latter point
here, but for fuMer details the ori-
ginal papers referred to above should
be consulted.
In Figs. 1 and 3 the curves re-
Fig. 5. presenting Heaviside's unit function
and its derivative have been drawn
as continuous unbroken curves (in Fig. 3 the two Mnes which approach infinity
along the Y axis are shown connected together in the upper portion by dotted
Mnes). The curve representing the Heavi-
side unit function considered by Jeffreys
is discontinuous at the point x=0, and
the value of H(x) at this point has to be
arbitrarily defined. Its first derivative, as 30=U
defined by him, has the value pH(x)=0 Fig. 6.
and is continuous at x=0. Let us con-
sider the consequence of such assumptions in connection with the problem of
integration.
Consider the function f(x) as shown in Fig. 4
3 ( f(x)=0 x^O
K }
I f(x)=0 x^O
The first derivative of f(x) is shown in Fig. 5 and is given by
( f'(x) = 0 x<0
K }
\ f'(x) = l x>0.
The value of f(x) when x=0 cannot be obtained from the usual definition
of a derivative and hence f'(0) must be arbitrarMy defined.
318 COMUNICAZIONI

The value of f(x) is by definition the Mmit as (5 — 0, where it exists, of

*^ ' — , for aM real values of ô different from zero. In this case at x=0,
there is no unique Mmit and hence the Mmit is said not to exist. The differential
quotients obtained from J .A —, di^-0, ô2^>0, but ôL and ô2 not
\ ö i ~r °2Ì $
both zero simultaneously, and the ratio ^ arbitrary, do not however approach
°2
all possible values, but approach aM the
values from 0 to 1 inclusive and these
00 =
V only. If we extend the method of de-
Fig- 7- fining f'(x) as above to include aM pos-
sible limits approached by the latter
differential quotient, we get the function defined by equations (8) and shown
graphicaMy in Fig. 1.
The second derivative of f(x) is shown in Fig. 7 and is given by
( f»(x)=0 x<0
K }
\ f"(x)^0 x>0.
The value of f'(x) when x=0 cannot be obtained from the usual definition of
a derivative and f'(0) must be arbitrarily defined, or an extended definition of
derivative adopted.
Let us assume that it is desired to apply the process of integration to the
function f(x) given by (14) and represented in Fig. 6, starting from some negative
value of x which may be caMed xL, and let the constant of integration be deter-
mined by the relation
(16) i f(x)dx =0 Zi<0

The function f(x) given by (13) and shown in Fig. 5 can be reproduced
up to the point x=0. As the point x=0
is approached we find ourselves in some-
what of a dilemma. The function f'(x)
is either undefined or arbitrarily defined
and there is an opening in the curve
at this point. Are we to add a constant
of integration at this point, or not? If oc~0
a constant of integration is not added, Fig. 8.
f(x) is reproduced for values of x greater
than zero. If a constant of integration C is added a function fL(x) shown in
Fig. 8 is obtained which is given by
fi(x)=0 x<0
(17)
( f^^C+x x>0
and fi(x) is either undefined or arbitrarMy defined at x=0.
 J. J. SMITH: Heaviside's operators and contour integrals 319

While in this particular example it might seem inviting to say that no arbitrary
constant should be added ; such a procedure would lead into difficulties in connection
with f'(x) defined in (15) and shown in Fig. 7.
If f'(x) shown in Fig. 7 is integrated from a negative value of x which
may be caMed Xi, and if the constant of integration is determined by the relation

(18) \(f"(z)& = 0 X±<0

X=X\

the function f(x) defined in (14) and shown in Fig. 6 is reproduced up to the
point x=0. In order to reproduce f(x) when x is greater than zero it is neces-
sary to add an integration constant at the point x=0 to the integral.
It cannot, however, be admitted in general that integration constants can be
added to an integral at any point in the range of integration, for that would
reduce the whole theory of integration to chaos. The singularity which appears
in these functions at x=0 is that they are either discontinuous so that there
is a break in the curve or that they hahe been arbitrarily defined at this point.
It would therefore appear that a good supposition to make is that an integration
constant can be introduced in the range of integration at any point at which
the function is not defined, or is discontinuous or is arbitrarMy defined.
We are accustomed to speak in mathematics of a definite integral of a one
valued function as completely defined, at least in such cases as this, where the
function has only a single discontinuity and where the integrand is assumed
to be completely defined. However, consider the functions defined by (13) and
(17) and shown in Figs. 5 and 8 respectively. The second function in (17)
may or may not be defined for x=0, save that if defined its value or values
are at most restricted to the interval O^y^l. The conventional derivative of
each of these functions is given in (16) and shown graphicaMy in Fig. 6 with
at most ambiguity at the origin.
If differentiation is to be a reversible process with the resulting integral
function completely defined, two methods of attack are open. We may carefuMy
define the character of the integrand at x=0 and so with suitable conventions
identify the integral, or we may adopt the older method of admitting that, where a
break occurs, an opportunity arises for the introduction of an arbitrary constant
of integration.
When the function as defined in (8) and shown in Fig. 1 is considered as
the derivative of the function defined in (13) and shown in Fig. 5, it is evident
that there is no possibiMty of the introduction of an arbitrary constant at x=0,
whether the problem is considered from the viewpoint of theories of integration,
or from the viewpoint of checking it by means of the definition of derivitive
given in H—F. The integration of the U-Function as defined in (10) and shown
in Fig. 3 to give the function as defined in (8) and shown in Fig. 1 is also
 320 COMUNICAZIONI

discussed in that paper. It is there shown that the definition of the function
at x=0
(19) 0^pH(x)^oc x=0

is not a complete definition of its properties and the suggestion is made that
it should at least be written
(20) O^pi^^loo4 x=0

which is another notation for equation (11). Further on in that paper it is

suggested that perhaps the most accurate way to define it is to say that it is
obtained from H(x) by a certain process and to regard (20) as an approximate
definition.
The definition of a function by its past history, so to speak, runs counter
to the mathematical notion that a function should be defined as a one to one
correspondence over a certain range, and that no questions as to its relation to
other functions should enter into the definition. While this may be a desirable
object to strive for, it is by no means certain that, in the practical problem of
discovering the finite form of a solution, symbolic methods requiring a reference to
the derivation of a function should be discarded on the basis of styMstic demands
alone.
As an example of the different results which may be obtained by proceeding
from the historical or non-historical viewpoint, the discussion on the sum of an
infinite series given in H—F may be referred to. It is shown there that if the
functions un(x) are continuous and if the series denoted by
71" 00

(21) %) = 2«»W
has a discontinuity at x=Xi, the sum being taken in the ordinary sense, when
the sum is taken under the rules laid down in the theory of U-Funetions aM
the values between S(xL+0) and Si(xi — 0) and perhaps additional values are
obtained. In the ordinary theory of the sum of the infinite series, the value
of x=xL is first inserted in the series, and the resulting series is summed as a
00
series of pure numbers, the fact that it came from the series 2 un(x) which is
o
its history, being completely ignored. In the theory of iï-Functions x and n are
made to approach their Mmits simultaneously and thus the fact that the sum
has a history is preserved.
A further comment on the definition of pH(x) at x=0 by means of deriving
it by a certain process from H(x) is that regarding functions from a general
viewpoint, the functions which do not require that they be defined at a point
at which they increase without Mmit are the exceptions rather than the rule,
since a function which has no singularities in any part of the plane reduces
 J. J. SMITH: Heaviside's operators and contour integrals 321

to a constant. If, therefore, we are in the position of not having precise notions
about infinity, it would seem justifiable to use the relation of any function at
its infinite points to other functions which do not become infinite at these points,
to define the function at the point at which it becomes infinite.
The notion of infinity as symboMzed by oo is undoubtedly a multivalued
concept. For instance, we write
(22) Lim —î— = oo
x-+ aa-~x
but we also write
(23) ^ i ^ - ~ -
Simüarly the notion of an infinitesimal is also a multivalued concept and we
write
(24) Um(x-a)=0
x-**a
and also
(25) Um(x=a)2=0.
x—+a
From these we obtain
(26) Lim x Lim (x — a) = oo x 0 which is indeterminate
x^aa x
x-+a

(27) Lim -q x Lim (x — a) = ooxO which is indeterminate

x-+a(x — a)- x_^a

(28) Lim- -2 x L i m ( a ; - a ) 2 = ooxO which is indeterminate

and
(29) Lim xLim(a;-ft)2=ooxO which is indeterminate.
x-+ax a
x^a

On the other hand we have

(30) L i m — — x(x-a) =l
x-+ax — a

1
(31) Lim x(x-a) = °o
x-+a Ve a
'
(32) i^__i_><(iC_a)2==1

(33) Lim — — x(z- a)2 = 0.

x-+ax ~~ a

Thus if we were to attempt to define the value of ooxO by agreeing to

assume that the reversal of Mmits was a justifiable process, from the above
examples we should be obMged to admit that it was a multivalued concept. In
the past, however, the problem has been regarded from the opposite point of
view, namely that the reversal of the limits was not a justifiable process. '-
In the so-caMed operational mathematics it seems far more profitable to assume
that Mmiting processes are always reversible, but to note at the same time that
Atti del Congresso. 21
322 COMUNICAZIONI

the concepts we are deaMng with are not always necessarMy single-valued but
may be multivalued. As an example, we may consider a series which represents
a function f(x) and which we shaU
denote by S(x). Let Sf(x) be the deri-
vative with respect to x. At points at
f which S'(z) is uniformly convergent it
represents the derivative of f(x), and at
j points where it is non-uniformly conver-
OO—O gent it may or may not represent f'(x).
Fig. 9. Let us assume that #4 is a point at
which S(x) is non-uniformly convergent
and that its value at that point is not equal to f'(x). We cannot then say

(34) [f'(x)dx= (S'(x)dx

where a<Xi<b so that the range of integration includes the point Xi. If, however,
we regard S'(z) as a multivalued function of x at the point x0, and take aM of
these values when obtaining the integral, and in a simüar manner if we take
aM of the values of f(x) in cases
where it becomes multivalued at xiy
a definition of a derivative being ta-
ken which permits it to do so, then
it would appear that we could defi-
nitely state that

(35) ff'(x)dx=fs'(x)dx

since we have taken aM possible values —

for both of the integrands. F i g 10
It may be profitable to discuss the
possibiMty of evading some of the difficulties which arise in connection with
differentiation of the Heaviside unit function by replacing it with a Mmiting
curve. Assume
(36) H(x)==Umf(x,n)

to be as shown by a Mmiting curve in Fig. 9 for a finite n.

df(x, n)
(37) Lim dx
n-+O0

is shown in Fig. 10 in fuM Mnes for finite n.

 J. J. SMITH: Heaviside's operators and contour integrals 323

Now let us multiply this by &(x) where <&(x) is shown dotted on Fig. 10, obtaining
df(x, n)
(38) Lim <$(x)
dx
shown in Fig. 11 for finite n.
Integrating, we get

(39) Lim [<P(z)dfix

dx
'n) dx
Fig. 11.
as shown in Fig. 12 for finite n.
The actual quantity we desire to obtain is, however,
dH(x)
(40) /*<*): dx
dx

as shown in Fig. 13. In the ordinary theory the reversal of Mmits used in
connection with Lim f(x, n) would be un justifiable since H(x) and its derivatives
n—»-oo
are aM non-uniformly convergent at x=0.
If, however, we agree to assume the Mmiting process is reversible but that
the functions involved may be multivalued then
(41) J * ( , ) ^ &-/*(.) Lim f(x, n) dx
»—»-00

and Fig. 13 foMows immediately from Fig 12.

While it might be objected that uniform convergence is a sufficient condition
but possibly not a necessary condition for the reversal of Mmits above, it must
be remembered that it is the only cri-
terion we have at present and hence we
have Mttle choice in the matter of using
T it. Furthermore, the possibiMty of the
r \$(0) necessary conditions being estabMshed
A seems rather necessary remote.
Fig. 12. In order to see the value of a ca-
reful interpretation of the unit function
and its derivatives we may consider a few appMcations to a simple Mnear differ-
ential equation of the first order with constant coefficients, whMe at the same
time Mnking it with a physical problem. T
In a circuit consisting of inductance \fi(0)
L and resistance r, let the E. M. F. be
zero when t<0, and E when t>0. We K g . 18.
shaM assume that at the instant t=0,
O^E^l, so that the E. M. F. can be denoted by 2?1 where 1 denotes the
Heaviside unit function. We can then write
(42) L^+ri=El (-oc<t<oo)
at
324 COMUNICAZIONI

and this equation is equivalent, save at the point £ = 0 to the equations we

should write in the ordinary theory

(43) lft+ri=0 (_oo<^<0)

di

(44) Ljt+ri==E (0<t<oo).

"Let us seek a solution of (42) in the form

(45) i=ai + C2e L 1.
Then
r r
/An\ di sy dl . ~ —jtdl rC2 —jt
<46> Tt"G^ + Ce
' L
Tt~-Le L L

Substituting in (42) we get

r r r

(47) LC^+LCë"1*^ -rC2e~ï''l + rCil + rC2e~i;t1=M.

Equating the terms containing 1 together and also those containing — we get
(48) rCà=E\
giving
(49) ft-f
and also r

(50) ZftS+LCe-ï'l-a
But — is zero everywhere except at £ = 0 where it has any value between
0 and oo but such that its integral fom 0 — e to 0-j-e is equal to 1. At £ = 0
~-tL
however e is uniformly convergent and has the value unity, hence we may
write
L dl _ dl
(51) e dt ~~~ dt
(45) thus becomes
(52) i=
-?(' — e
We have thus apparently obtained a definite result out of equation (42)
without evaluating any arbitrary constants. Let us examine more closely what
--t
this means. A term Cse L could be added to equation (45). The left-hand side
of equation (47) would then contain the terms
--t --t
L L
—rCde +rC3e
 J. J. SMITH: Heaviside's operators and contour integrals 325

which would cancel out and would not aMow us to determine the value of the
constant C 3 . Equation (52) should then be written
r r

(53) i=c9e~î' + § {l-e~yf

or as it might be written in the usual manner
-~tL
i=C3e (-oo<^0)
(54) k r r
i=C3e~ < " + ä^e~Lt\
I i
(O^t<oc)
It wMl be noted in (53) that the term involving C3 does not have a 1 sign
attacched to it, and thus the part involving C3 must be vaMd when t<0. Since,
however, the voltage E was appMed to £ = 0 it can hardly be consireded to have
anything to do with this part of the current, if we are to hold to notions of
cause and effect. Thus if it is desired to find the current produced by El, the
--*
term CBe L must be rejected on the grounds that it is due to some other cause
entirely independent of El.
The procedure in interpreting the result is somewhat analogous to that used
in functions of the complex variable when an equation such as
(55) u + iv=f(x, y) + ig(x, y)
is broken up into two equations
u
(56) j =f^y\
( v=g(x,y)
u, v, f and g being real.
We interpret the solution of
(57) %+°V-E
to be
(58) i=f(t)+g(t)l
where f(f) denotes that part of the current which corresponds to what was in
the circuit before £ = 0 when El commenced to act, and g(f) denotes the part
of the current due to El. As in general we are interested in the effects produced
by El we may neglect the term f(t) and write

(59) i=g(t)i.
As a second example we may consider that the voltage E acting on the
same circuit for an infinitely long time is suddenly short-circuited at the instant
£ = 0 . The equations then become

(60) L*t +ri=E-El.

326 COMUNICAZIONI

A solution may be taken of the form

r
< 61 > i^Ctl + Cte"1 1 + Ci-
This is simMar to (45) except for the addition of the constant CB which is not
multipMed by 1, and is required to take care of the term E in the right-hand
side of (60). Then
r r T
(62) LC,^ +LC2e~ltft -rC^e'1* l+rC.l+rC^1* l+rC3=E-El.

Equating as before the coefficients of 1, —, and also the terms which do

not contain the 1 symbol we get
(63) rC3=E
(64) rCil=-El
and r

(65) ZGS+Lftrz'g-Q.
Hence we find
(66) & - - G » !
and
(67) Ct — Ct.
Thus
(68) t -_|(/_i) + | a - ^ l

which might be written in the ordinary form

/ . E
t<0
(69) { „ rt
L
| t=^e t>0
r
the first term in (68) being zero when t>0 and the second term being zero
L
when t<0. Again a term de might be added to (68) but could be rejected
if we are only interested in the effects produced by E.
As a third example, consider the equation

< 70 > £%+«-*%.

Let us assume a solution
(71) i=Cil+C2e~ltl.
Then on substituting in (70) we get
r r r
(72) LClft+LCie~lt^t-rC2e~ïtl + rCil+rCie~ïtl=*K^
(73) 7-^1=0
 J. J. SMITH: Heaviside's operators and contour integrals 327

and rt

(74) LClft+LCie-ïft^.
From these
(75) ^=0
K
(76) <
L
and thus
r
_I
(77) »-J« 'i.
K
In this last equation when t=0 + e, i0=T, which affords us a method of
solving such equations subject to an arbitrary initial condition. Therefore in
order to solve the equation
(78) L%+Ti-B
K
subject to the terminal condition i0 = — it might be replaced by the equation
LA

and the latter solved operationaMy as above would give the correct solution.
Jeffreys (loc. cit.) gives a method very similar to this for solving equations for
arbitrary initial conditions, and seems much perturbed when his work leads him
to the point where it seems that aM the results required can be obtained directly in
this manner from the differential equation and there seems to be no room for
the introduction of arbitrary constants of integration.
For instance, he says on page 1 6 :
« Then has p any meaning of its ows ? Since it replaces -=- when we form
the subsidiary equation we may naturally suppose that p means — , and this
is the meaning sometimes actuaMy attributed to it ; but care is needed. We recar
that when the subsidiary equation is formed a term Mke py0 appears on the
right ; but -—• is zero, so that if we pushed this interpretation too far we should
be faced with the alarming result that the solution of the equations do not
depend on their initial values. The fact is that though the operators — and Q
both satisfy the laws of algebra and are freely commutative with constants,
they are not as a rule commutative with each other ».
It wiM be shown later that with proper use of the Heaviside unit function
the operators ax
— and Q can be made commutative with each other, and hence
another explanation for the above must be found. A possible alternative seems
to be in the denial of the fact that - ^ , or as it should be written in the notation
v dl
of the present paper ^~—, is zero for aM values of x.
328 COMUNICAZIONI

It should be clear from the preceding that if we are only interested in

K dl
effects produced by El and — -— then the sum of equations (52) and (77) is
--*
the complete solution. A term Ce L may be added to the equation for i if it
is desired to do so, but since this part of the current existed before the time £ = 0
it cannot be supposed to have been caused by any of the terms on the right-
hand side of (79) since these were equal to zero when t<0.
Although we have used the Mnear differential equation of the first order in
the above for the purpose of iMustrating the points involved in the use of Heaviside
unit function, the place where its usefulness is most apparent is in appMcations
to equations of higher orders.
For instance, consider an equation of the nth degree:
/on\ dny dn-~Ly , dn~2y , dy , ™
(80) J- + a, - ^ + a2 ^ + .... + «„_, J + an=El.
This can be written
E
(81) y- dn dn~i d

and on factoring the denominator in the form

m '(B)-(B+')(S+>)-(B+»)
and spMtting up the operand in the expression (81) into partial fractions we
obtain
E
(83) y=%
a,ß....fi\— + a\f'(—a)
dt

Now any one of these terms in the sum can be evaluated from our previous
work. Let it be required to determine the value of

<84> 7T-^ 1
-

Assume it equal to u, then we have

< 85 > (1+°)"- r^T)1

and on comparing this equation with (42) and its solution it is evident that

(86) „__£_<!_,--,!
and hence from (83)
87
< > v- S 5r^r,( 1 -^) 1
a,ß....ft
 J. J. SMITH: Heaviside's operators and contour integrals 329

which by rearrangement can easMy be shown to be the same result as the

Heaviside Expansion Theorem. In this case the economy obtained by the use
of the unit function is very apparent. NormaMy, with an equation of the nth
order, it would be necessary to determine the value of n constants of integration.
If, however, we are only interested in the results produced by El, they can
be obtained directly by the above method without the evaluation of any such
constants.
As an even more extended example consider the appMcation to a Mnear equation
with constant coefficients of an infinite order.
Consider an electrical cable of length I, with resistance r and capacity C
per unit length initiaMy at zero potential. Let one end be maintained at zero
potential while the other end is raised to a potential E at the time £=0. Thus
we have „ N

(88) d S-O-S-**
where ^p = dt
— and K= Cr.
The general solution of this problem is
(89) v = s i n h ipKx • AL + cosh fpKx • A2
where Ai and A2 are functions of t but not of x.
Since when
(90) x=0, v=0, hence A2=0.
Also when
(91) x=l, v=El,
hence _
(92) A,= £Lr-
sinh YpK • I
and —
sinh ìpK • I
(93) v=sinh^l-xEl
now
(94) sinh 6 - © (l + 5 ) (l + £).... (l + £ ) ....
and thus
(95) sinh i^K • i - 1JK • I ( l + P~).... (l + g g ) . . . .

If the series expressions for the sine is inserted in the denominator of (93)
and the latter expanded in partial fractions, we get

n—oo' 2na sin —— El

96 1 l
< > *= S (-D- pmi+ nw
n—0
the ' signo on the summation denoting that one-half the value is taken for # = 0 .
Each of these fractions may be evaluated in a manner simMar to that used in
preceding work.
330 COMUNICAZIONI

Let

The inserting the value of p= - w e get

and hence
(99) „ n = _ ^ ( i _ e w ji.
Using this in (96) the usual result is readily obtained on rearranging the
terms.
It should be noted that if (93) were written

(100) sinh fpKl. v=sinh ipKx . El

and the series expressions for the sine inserted on the left-hand side, this becomes
a differential equation of infinite order for v, and would thus require the evaluation
of an infinite number of arbitrary constants to solve the problem in the ordinary
manner. It is in problems such as this that the careful use of the Heaviside
unit function wiM be found to have the greatest advantage. The appMcation to
other types of problems such as potential problems wMl be found in H—F and
need not be gone into in further detail here.
In connection with the presence of the fp in (100) it might be thought
that fractional operators appear in this equation. On writing out the expanded
expression for the operators on both sides of this equation, it wiM be found
that there is a common factor fp on each side which therefore cancels out and
hence only whole integrations are involved.
It may be worth while to discuss the use of the unit function in making
differentiation and integration reversible processes in connection with the ordinary
functions used in analysis. On page 17 Jeffreys points out that while
x

(101) |L ff(x)dx=f(x)
Ó
X

(102) [^f(x)dx=f(x)-f(0)
5
and hence unless f(0)=0, the two processes are not reversible.
The same operations may now be considered with respect to f(x) multipMed
by the unit function. Since H(x) is multivalued at x=0 being 0^-H(x)^l
the lower Mmit of integration wMl be taken to be 0—e instead of 0, and thus
 J. J. SMITH: Heaviside's operators and contour integrals 331

ambiguity in the result wiM be avoided, (e is used to denote a smaM quantity

which may be diminished indefinitely). Then
X

(103) £. f f(x)H(x)dx=f(x)H(x) (x>0-e)

0—e
X X

(104) flf(x)H(x)dx=([f'(x)H(x) + f(x)H'(x)]dx (x>0-e)

0—e 6—e

where H'(x) denotes the first derivative of H(x) with respect to x. But since
H(x)=0 when x<0 the integral of the first term becomes zero if x<0, and
becomes x

\f'(x)dx if x>0.
à
Hence x

f f
«HAS M « MTì v \ W- (°)+ (f(x)H'(x)dx when x>0
(104a) J WH*>-
0 - e5 0 L When^<0.

But as has been pointed out before, f(x)H'(x) is zero everywhere except
at x=0 where its value is f(0) multipMed by a quantity whese integral from 0—e
to 0 + e is 1. Hence

dos) /w^-srt^n
Af 0 whenz<0
U—e \
and thus we get
(106) f'MWto-W-M + W-™^'^
ax
•1 [0 when£<;0.
U—E *

It is not difficult to show that if x=0

X

f x H x dx fx
(106a) °^\a\ () () ^ i )-
0-e
But (106) and (106a) mean that
X

(106ft) [| f(x)H(x)dx=f(x)H(x).
0-e

Thus for functions which are multipMed by H(x) or 1 the operators

X X

— / dx and / dx-z-
dx
dx JJ JJ dx
dx
0—e 0—e
332 COMUNICAZIONI

are reversible. It is evident that these operators might also have been written
X X

— / dx and / dx~
dx J J dx

which are useful in deaMng with functions multipMed by H(x + u) which is zero
when x<— u and equal to 1 when x> — u its value at x=— u being given
by 0^H(x + u)^l.
It is evident that if in writing (1) Jeffreys had considered

(107) e**= .... + ^ + ^ - + _ + _ + _ + ....

the series being continued both ways, he would obtain

(108) p»H(x)=^
—n
on page 19 which would have been consistent with his result on page 25.
The objection which might be raised to writing (107) for use in the contour
integral is, of course, that it introduces singularities into eKx at x=0. It would
seem, however, that the consistency obtained by foMowing this procedure, more
than counter-balances the objections which might be raised. By doing so we
h a v e
i x—
(109) ±p"H(x)=p^^H(x)=*H(x).
Where as if
(110) pnH(x)*=0
(HI) ^,p-H(x)^^0

and we can proceed no further without additional data. Of course, it might be

argued that the latter is nothing but a case where the operations are irreversible.
This may be so, but it can easily be shown that the other cases of irreversi-
biMty of operators given by Jeffreys could be made reversible without any serious
difficulty.
In addition, the above series for eKx would make it fit in a formal manner
with the generaMzed series for e£x used by Heaviside and discussed also by
Jeffreys loc. cit. page 91,
(112) «»- 2 (n + r)\
r——oo
where n is a proper fraction. The part of the series corresponding to negative
values of r is always divergent but can be shown to possess an asymptotic
property (see Jeffreys, page 91). To discuss fuMy the difference between the
usual series for eKx and the generaMzed series for eKx would require quite a
lengthy discussion.
 J. J. SMITH: Heaviside's operators and contour integrals 333

The results to be obtained from the use of multivalued functions such as

that given for H(x) in Fig. 1 would have a very far-reaching effect in mathe-
matical theory. It may be weM to give a few examples.
Perhaps one of the most interesting results is that it destroys the vaMdity
of the reasoning by which it is considered that if it is shown that
(113) A>B and A<B
at the same time, we have arrived at an inconsistency. An examination of Fig. 1
wiM show that when x=0, H(x) is greater than, equal to and less than 1/2,
and that there is no inconsistency in the statement. In other words, before we
can assume we have arrived at an inconsistency because A>B and A<B
simultaneously we must show that both A and B are single valued concepts.
It is clear that the law of the excluded middle does not hold at points at
which functions become multivalued, and the method of proof by reductio ad
absurdum becomes in itself absurd if appMed at such points. In this connection
the work of Brouwer (A) is interesting in that he declares the use of the excluded
middle is legitimate only for finite sets. It would seem from the above that
unless we could show first that an infinite set was a single-valued concept that
the law of the excluded middle could not be appMed to it. We shaM not discuss
whether the law of the excluded middle could be appMed to an infinite set, if
it were shown to be a single-valued concept, as we know no way of showing
the latter statement to be true.
A lack of reaMzation of the necessity for considering this aspect of the
problem seems to Me at the root of many of the mathematical paradoxes. For
instance, to refer to one paradox, the set of aM things. There is no inherent
contradiction in such a set containing itself, if such a set is a multivalued
conception. There is possibly a contradiction if such a set is a single-valued
conception. In order, therefore, to show that a paradox has arisen it would be
necessary to prove first that the conception of such a set is a single-valued
conception, and I am not aware that anyone has demonstrated this. It may
have been defined to be such, but that is an entirely different matter.
We may consider the effect upon another region of mathematics by referring
to Carslaw Mathematical Theory of Heat, 1921, page 14, where he treats of
boundary conditions in Heat problems as foMows:
« In the mathematical treatment of the question these surface and initial
conditions are not regarded as conditions which v must satisfy on the surface
itself or at the instant £ = 0 . They are taken as Mmiting conditions and it is
required in the one case that one solution shaM converge to the given surface

(l) See for instance a discussion by Pierpont, Mathematical Rigor, Bulletin Amer. Math.
Soc, XXXIV, 1, Jan.-Feb. 1928, p. 51.
334 COMUNICAZIONI

or initial value, and in the other case that the differential coefficients in the
Mmit as we approach the surface shaM satisfy the corresponding conditions ».
Yet in the next paragraph Carslaw asks us to consider the problem

(114) | j =WH
in the soMd
(115) v=f(x,y,z)
for t=0 in the solid
(116) v=$(x,y,z,t)
at the surface.
The best we can say for the statement of the problem is that it is very
carelessly written if we interpret it according to the quotation in the preceding
paragraph. If we do not interpret it according to the preceding paragraph it
involves a contradiction provided that the solution to be derived is to be single-
valued, unless
(117) 0(x,y,z,O)=f(x,y,z)
at the surface.
If, however, the solution is aMowed to be multivalued there is no contradiction.
In the solutions of such problems which arise in connection with initial
conditions as given above, it wiM be found that the solution can be interpreted
as a multi-valued function which does actuaMy satisfy the conditions laid down.
Such an interpretation, however, involves the rejection of the assumption made
in pure mathematics that if we insert the value of x at any point in an infinite
series, and obtain a definite value for the sum of the series at that point, then
the series has only one value at that point. In the paper on U-Functions
referred to above the opposite viewpoint is taken, namely that the sum of a series
at any point may be multivalued until it is definitely shown to be single-valued.
If it is multivalued then aM possible values which can be found from the series
are taken as the sum. For details, the paper referred to must be consulted. It
is interesting to note some of the most important series which occur in connection
x=l
with Green's Functions, such as ^cosnfa—Xi) are not functions in the ordi-
a;=oo
nary mathematical sense. Considered as multivalued functions they are multi-
valued at every point, but this does not prevent some very useful results being
derived from them.
In general, the impression seems to be that the introduction of multivalued
functions would lead to confusion and serve no useful purpose. Actual experience
with such functions seems to show that they have a large sphere of usefulness
and lead to many interesting results much more directly than ordinary mathematical
methods. It is only natural that they should, because the theory of multivalued
functions is naturaMy very much broader that that of single-valued functions
 J. J. SMITH: Heaviside's operators and contour integrals 335

and includes the latter as a subdivision. It also makes what Hobson caMs the
morbid pathology of the Theory of Functions the most interesting part, for it
wiM generaMy be found where functions exhibit singular behavior and give rise
to difficulties in the ordinary theory that at such points they become multivalued
when the opportunity to do so is presented to them.
Note: It might appear from the comments in the above paper that the
x°
Heaviside Unit function was synonymous with .—. This is, however, not the case
since rrr is not zero when x<0. The reason that r—: which is the derivative
jpO L_ I
of rz can be considered, from one point of view at least, to have an infinity of
values at x=0 is only seen when both the real and imaginary values of it as
a function of x and n (n being the exponent) are considered.