ON TWO THEOREMS OF F. CARLSON AND S. WIGERT.

BY

G. H. HARDY
NEw COLL~.~r~ OX~OaV.

I. In this short note I have united a number of remarks relating to two

theorems due in part to Wm~RT and in part to C~RLSO~. ~ The theorems belong
to the same region of the theory of functions, and it is natural to consider
them together.

I.
2. I write z----~ + iy=re~O. Then the first theorem is as follows. ~

The manuscript of the note (then e n t i t l e d 'On two t h e o r e m s of Mr. 8. Wigert') was sent
to Prof. MITTiG-LEFFLER in 19I 7. I was at that t i m e unaware of the existence of Mr. CxaLsoN's
dissertation ('Sur une classe de s6ries de Taylor', Uppsala, I914). This dissertation was g i v e n to
me by Prof. MITTAG-LEFn,~ in S e p t e m b e r 1919; and I found at once that Mr. CARLSO~ had
anticipated not only Mr. WIG~.RT'S theorem of I916, referred to in w 2, but my own generalisa-
tion of this t h e o r e m and indeed the substance of most t h a t I had to say.
The note, however, contains s o m e t h i n g in substance, and a good deal in presentation,
that is new; and I have t h e r e f o r e agreed to Prof. MITTAG-LEFFLER'Ssuggestion that it should
still appear. Except as regards ~ I--2, I h a v e left it substantially in its original form.
W~QER~ ('Sur un th6or~me concernant les fonctions enti~res', Arl6v ]~r Matematik, vol.
1i, ~916, no. 22, pp. I--5) proves a theorem w h i c h is less general in t h a t (I) t h e angle is sup-
posed to cover t h e whole plane and (2)f(z) is supposed to vanish for all positive and n e g a t i v e
integral values of z. CxRnso~ (l. c., p. 58) proves a theorem w h i c h contains t h e present theo-
r e m as a particular case (but is in fact substantially e q u i v a l e n t to it). His m e t h o d of proof is
similar to that of the first two proofs g i v e n here.
Wm~ (l. e.) refers to previous and only partially succesful attempts to prove his theo-
rem, and gives a proof based on a theorem of PHRXGM~ ('Sur une extension d'un th~or~me
classique de la th6orie des fonctions', Acta Mathematiea, vol. 28, i9o4 , pp. 351--369). H e deduces
as a corollary a result relating to the case in w h i c h f(z) vanishes only for positive integral
values of z; in this the n u m b e r n is replaced by the less favourable n u m b e r 2I . I may add

that a similar result, in w h i c h I ~ is replaced by the still less favourable n u m b e r I, was found
2
i n d e p e n d e n t l y by P6n-~& ('(~'ber ganzwertige ganze Funktionen', Rendieonti del Circolo Materna~ieo
328 G.H. Hardy.

z!
1
(I) !(z) ia regular at all points in,ide the angle - - a < ~ < c~, where a ~ ~eg;

(2) ]](z)] < A e ~r, where lc < ~ , throughout this angle;

(3) l(n) = o lor n = x, 2, 3 , . . . ;

then ](z) is identically zero.

3. I t seems most natural to deduce this theorem from those proved b y

P~rl~AG~t~I and LII~DELS~ in P a r t III of their well-known memoir in Vo]. 31 of
the Acts Mathematiea. ~ Let us suppose that l(z) is not always zero, and write,
with PtIRAGI~I~ and LINDELOF,

h(O) = l i r a sup log II(re e) l,

~" - - a - 00 r
so that
h(O)<k, ( - - u < 0 < a).

Then h(O) is oontinuous for - - a < O < a. 2

Now let
/(z)
P(z) = sin ~ z '

so that F(z) also is regular inside the angle of the theorem; and let H(O) be
formed from F(z) as h(fl) is from /(z). Then it is o b v i o u s that

(I) H (fl) = h ( O ) - - z [ s i n O] < k - - z ] s i n 0],

except possibly for 0 = o.

If 0----o, z = x is 'real. We write

l ( x ) = u ( x ) + iv(~), _F(x)= U(x) + iY(x).

Let us suppose that x is not an integer, and that n is the integer nearest to x.
Then
di -Palermo, vol. 40, 19~, pp. 1--i6). I t should be added that this result of PeLYA appears
only incidentally as a corollary of theorems of a somewhat different character and of the
highest interest.
a E. PmRAG~S and E. Li~nm~6~, 'Sur une extension d'un principe classique de l'analyso
et sur quelques .proprietbs des fonctions monog~nes dans le voisinage d'un point singulier',
Aeta Mathematics, vol. 3r, x9o8, pp. 38I--4o6.
2 This follows from the argument of pp. 404--4o 5 of PrI~Ao~,I~ and LI~Dm~OF'S memoir.
This argument presupposes that the value of h(O) is not always - - o r , a possibility ex-
cluded by the theorem of p. 38~.
 On two theorems of F. Carlson and S. Wigert. 329

U(x)= z--n u'ff,), V(~r)- z - - n v'(~,),

sin ~rx sin zcz "

where ~ and ~z are numbers between n and x. I t follows that

IF(x)l< cr
where C is a constant and ~ , is the maximum of I/'(x)l for n - - I < z < _ n + I.
But

/'e~=:-L- f. /(~ d~,

" " 2,r ~ J ( z - - x)'

where the contour of integration is the circle [ z - - ~ [ : I; and so

[/'(z)l < A e kc'+l~, ~v. < AekC,'+s~.

Thus
l~'(x)I< Be ~,
where B is a constant.
I t follows that H(o)<k. H e n c e H(O) is continuous for 0==o, and (z)
holds for all values of 0 between - - a und a. Thus H(8)< o for # < 8 < ~, and

--7 < 0 <- #, # being the positive acute angle whose sine is ke g ' and 7 the lesser
of a and z r - - # .

B u t it is easily proved that this is impossible. Suppose first that a > ~ z.

Then H(O) cannot be negative for - - a <fl < a, since the length of this interval
is greater than z. 1 There is therefore, inside this interval, an interval in which
H(fl) is positive.' This last interval must form part of the i n t e r v a l - - # < 0 < ~ ;
and its length is therefore less than z. And this, finally, is impossible:
I I
Secondly, suppose that a ==-~r. If H(O) is not negative f o r - - - z ~ < 8 < ~z,
2 2 2

we obtain a contradiction in the same w a y as before. If H(O) is negative for

I I
z<0<-~, it must be of the form L c o s 0 + M s i n 0 . ~ It is plain that L
2 2

must be negative and M zero, and that H(O) must tend to zero as we approach
an end of the interval, which is not the case.

* PHRAo~ and LISV~.LOF, /. C., p. 400,

/bid., p, 399.
' /b/&, p. 399"
Ibid., p. 403.
Ac~a mathe'ma~ica, 42, Imprim~ le 3 mars 1920. 49
330 G.H. Hardy.

We have therefore arrived in a n y case at a contradiction, and the theorem

is proved. It is easy to see, by considering a function of the form e-c~ sin ~z,

t h a t it ceases to be true when a < i

2

4. I shall now give an alternative proof of the theorem based on entirely

different ideas. This proof is less elementary t h a n the first, but seems to me
to be of some instrinsic interest.
Suppose t h a t w is positive, p a positive integer, o < u < i, and p < ~ < p + T.
Then it is clear that, under the conditions of the theorem, we have

~[(n)(--w) "~= 2 ~ i ; si-i-ff~z ] ( z ) w " d z - I . ~ . z

. /(z)w"dz,
9 2~ Jsln zz
1

the paths of integration being rectilinear.

L e t us suppose now t h a t ~ = p + I and t h a t p - - ~ . Then

2

'+'~sin ~vz =
fl( 9
cosh ~:y
)1- d y .
--" ~Oo

Also

where B is a constant; and so

;L+iw

where C is another constant. Thus the integral tends to zero if w is sufficiently

small. We have therefore

9 (w) = + w l(3) . . . . . .

for sufficiently small positive values of w. This formula is of course well-known,

and shows t h a t ~(w) is an analytic function regular for all positive values of w.
 On two theorems of F. Carlson and S. Wigert. 331

Suppose now that --i<s<o. Then we c a n choose p and v so t h a t

o<v<--s</,<I. A n d we h a v e 1

1 1 /t+i~

(2) j sin 7z'z

o 0 ,~-'.~r

(3) w'-XO(w)dw-= uf- dw z ufdz,

provided only that these integrals are convergent.

The double integral
1 ~t+i~

is c o n v e r g e n t , as m a y be seen a t once b y c o m p a r i s o n w i t h t h e i n t e g r a l

1 co

ffwl~+8-1~(~k)|ll|dwdy.
o ~oa
H e n c e t h e i n t e g r a l (2) is c o n v e r g e n t , a n d m a y be c a l c u l a t e d b y i n v e r s i o n of t h e
order of i n t e g r a t i o n . T h e s a m e a r g u m e n t s m a y be a p p l i e d to t h e i n t e g r a l (3).
I n v e r t i n g t h e o r d e r of i n t e g r a t i o n , a n d c o m b i n i n g t h e results, we o b t a i n

w "-1 ~ ( w ) d w = Sin~rz z + s sin ~rz z + s

t.
0 / * ~ , 0o v~oo

or
0o

(4) w~-lO(w)dw ~ sin

0

T h i s f o r m u l a h a s b e e n p r o v e d for - - x < s < o. I t is e q u i v a l e n t t o

This artifice is due to MsLIal% from whose work the ideas of the proof are borrowed.
See HJ. MmmIN, 'Ober die fundamentale Wichtigkeit des Sstzes yon Cauchy fiir die Theorien
der Gamma und der Hypergeometrischen Funktionen; Acta 8oeietati* Fennicae, vol. 21, no. i.
x896, pp. x--xll (pp. 37 et se,q.).
332 G.H. Hardy.
0o

(5) w t-1 (ao ~ a l w + a , w ~ . . . . ) d w ----"~ a - t ,

0

where o < t < z and a , = a ( z ) is an analytic function of z subject to certain

conditions. In this form the formula was communicated to me s o m e years
ago b y Mr S. RA~AZ~UJA~, in ignorance of M~LLIN'S work.
So far we have made no assumption as to the values o f / ( z ) for integral
values of z. It is plain that, if / ( n ) = o for n ~-I, 2, 3, . . . . we obtain

(--i <s<o),
so t h a t / ( z ) is always zero.

II.

5. Suppose t h a t / ( z ) is an integral function of z such t h a t

(,) I/(z) l < e("+*"
for every positive e and all sufficiently large values of r,

for every positive e and for a corresponding sequence of values of r whose limit
is infinity. Then, following PRI~GSH~IM) I shall call l(z) an integral function
of order i and type 7. This being so, the second theorem which I wish t o
discuss m a y be stated as follows.
T h e necessary and su//ieient condition that

/ ( z ) .ffiao + a~z + a2z ~ + ...

should be an i n t e g r a l / u n c t i o n el x is that there should be an i n t e g r a l / u n c t i o n

a(z), of order I a n d type o, which takea the valuea a~, a2, as, . . . /or z ~ I, 2, 3 . . . .
I may insert here a few references to the rather extensive literature connected
with this theorem. The complete theorem is due to Wm~.RW2; but the second half of
it, asserting the au//iciency of the condition, was discovered almost simultaneously,
and b y a quite different method, b y L~. R o y . n Another proof of this part of the
i A. Pm~QSHEIM, ' E l e m e n t a r e Theorie der ganzen transcendenten F u n k t i o n e n yon end-
licher Ordnung', Mathematische Annalen, col. 58, 19o4, pp. 257--342.
S. WIa~sr, 'Sur les fonctions enti~res', Ofversikt af K. Vet.-Ak. F6rhandlingar, Arg. 57,
I9oo, pp. l o o I w l O I I .
8 E. T.E RoY, 'Sur les s6ries d i v e r g e n t e s et les fonctions donnges par un d g v e l o p p e m e n t
de Taylor', Aunales de la Facult~ des Sciences de Toulouse, ser. 2, col. 2, i9oo , pp. 317--43o (pp
350--353).
 On two theorems of F. Carlson and S. Wigert. 333

theorem has been given by LINDELSF, 1 who deduces it, along with many other
important theorems, from the '/ormules somrnatoires' of the calculus of residues.
The whole theorem was rediscovered at a slightly later date by FAB~.R,s whose
proof does not differ in principle from Wm~.RT'S. It has been further discussed
by Pm~GS~.m, 8 who presents the whole proof in a particularly simple and
elementary form. And, as explained in the footnote to w i, it is a special case
of much more general theorems to be found in CARLSON'S dissertation.

6. The proof which I give here stands in closest connection with the ideas
of L~. RoY. I begin by proving the following lemma.
I n order that a(z) should be an integral/unction o/order i and type o, it i8
necessary and sufficient that a(z) should be o/the ]orm

(x)

where C i8 a simple contour enclosing the origin, and ~ { u ) = r an integral

function o / w .
This lemma is extremely easy to prove and very useful, but I do not
remember having seen it stated explicitly. In the first place the condition is
sufficient. For we may replace the contour by a circle whose centre is the
origin and whose radius is ~, and then

l a(z) I = l a(re i~ I < e M e ' ,

I /T/I
where M is the maximum of I ~ { J [ on the circle.
| /--/ |

In the second place the condition is necessary. For if

Oo

is a function of the required type, then ~

I E. LINDEL6F, Le calcul des r&idus, Paris, I9o~, p. I27. See also 'Quelques applications
d'une formule sommatoire g6n6rale', Acta Societatis Fennicae, sol. 31, no. 3, I9o2, PP. 1--46.
G. FA~ER, 'Ober die Fortsetzbarkeit gewisser Taylorscher Reihen', Mathematischv Anna-
lot, sol. 57, ~9o3, pp. 369--388.
a A. PRINGSHmM, 'Uber einige funktionentheoretische Anwendungen der Eulerschen Rei-
hen-Transformation', Mi~nehener Sitzungsberichte, I9x2 , pp. 11--92 (pp. 4o--45).
' See, e. 9., PmNGSHmM,l. C., p. 38.
334 G. H. Hardy.

f/n!le.-----i-- o;
so that
~OCW) -~-- 2 n ! (~nWn4"l
o

is an integral function of w. Thus

du.
0 ) )
o c

7. We have now to show that [(z) is an integral function of --~-- if and

I--Z
only if there exists a function of the form (i) which assumes the values at, a~,...
for z - ~ I , 2 , . . .
In the first place, if such a function exists, we have

~o

(I) [l(Z) =/(z) - - ao = ~ a n Z " = 2 ~r I --~ee ~ q~ du,

1

if C is a contour enclosing the origin and ] z e ~ ] < I at all points of C. These

conditions will be satisfied if I z ] < i and G lies entirely to the left of the line

]
The only singularities of the integrand, other than the origin and infinity,

are the various values of log ~ ; and it follows, b y a familiar argument due in
principle to HADAMARD,1 that the only possible singularities of ] l ( z ) a r e the

values of z for which log x is zero or infinite, that is to say the values 0, I,
z
and oo.
L e t us draw a cut in the plane of z from x to ~ , say along the positive
real axis. Then there is a branch ft(z) of /l(z), the so-called 'principal' branch,
which is one-valued and regular in the cut plane and vanishes at the origin.
If finally we can show that /~(z) is one-valued i n the neighbourhood o/ z = i , it
will follow that ft (z) is the only branch of /~(z), and so t h a t / ~ (z) is a one-valued

J. HADAMARI),'Essai sur l'~tude des fonctions donn~es par leur d~tveloppement de Tay-
lor', Journal do math~mat'iques, ser. 4, vol. 8, 1892, pp. ioi--i86.
 On two theorems of F. Carlson and S. Wigert. 335

function with z = x as its sole finite singularity, that is to say an integral func-
tion of i

Suppose then, to fix our ideas, that z tends to I through positive values
less than I, that C is a circle whose centre is the origin and whose radius is
less than log z ' and that is a concentric circle whose radius is greater than

l o g l ~ I. Then ~

h(z) = i_ze,,q~ du
C

log c,

where log x denotes, of course, the value of the logarithm wbich vanishes for
z

z=x. I t is plain that the last integral representents a function regular for
Z~-~ I . As

is one-valued in the neighbourhood of z = I, so also is /l(z). Thus one half of

the theorem is proved.

8. Secondly, let us suppose that /(z) is a function of the type prescribed.

We have

(I)
f /(z) dz '
an-~ 2 z i f l z n+l

the path of integration being a closed contour surrounding the origin, but
excluding the point z = I . I t is evident that, if n > x , * we may deform this
contour into that formed by (z) the right hand half of the circle

Izl=e-z,
1 C o m p a r e G. H. HARDY,'.~k m e t h o d for d e t e r m i n i n g t h e b e h a v i o u r of a f u n c t i o n repre-
s e n t e d b y a p o w e r series n e a r a s i n g u l a r p o i n t on t h e circle of c o n v e r g e n c e ' , _Prec. London
Math. ~oe., ser. 2, vol. 3, I9o5, pP. 38Im389.
T h e a r g u m e n t fails for n----o unless f ( Q v ) = o. C o m p a r e W m ~ T ' s paper.
336 G.H. Hardy.

7 being any positive number, and (2) the parts of the imaginary axis which
stretch from the ends of this semicircle to infinity.

Let us now effect the transformation z----e -u, u = log 1, where that value

of the logarithm is chosen whose imaginary part lies between --rr and ~. The
contour in the z-plane becomes a contour in the u-plane formed b y three sides
of an infinite rectangle whose vertices are

I
V+ ~i, ~ o 0 + 2I-~i, --co - - - 2I~ i , - - 7 - - I ~2i

and we have

the integration being effeeted along this contour. It is obvious that the contour
may now be deformed into any simple closed contour which encloses the origin
b u t lies entirely inside the circle ]u]-----2~c. Finally, ](e -~') is plainly regular
except for u = o and u = + 2 k z i (k-----i, 2 , 3 , . . . ) , and is therefore of the form

where q0(w) is an integral function of w and ~V(u) is a power series whose radius
of convergence is at least 2 ~c. And

9/
an = 2 ~ri e 9 du,

which proves the theorem.

9- The preceding proof of WmERT'S theorem is of course less elementary

than (for example) PRI~mSHEI~'S. I t seems to me interesting none the less on
account of its almost intuitive character. It has the further advantage of
lending itself v e r y readily to generalisation, as I shall proceed to show.
In the first place, the lemma of w 6 m a y be at once generalised as follows:
I n order that a(z) should be an integral ]unction o/ order I and type 7, it is
necessary and su//ieient that a(z) should be o] the ]orm

a(z)=2~, 1 -
 On two theorems of F. Carlson and S. Wigert. 337

where C is a contour which includes the circle l u l l 7 , and qD(w) is a /unction

regular /or I w I < -7 but not /or I w l < ? 9

To prove this we observe that a(z)~--Y, cnz n will be a function of the t y p e
required if, and only if,

lira sup l/n! [c.[----7,1

so that the radius of convergence of 2 n ! c n z ~ is precisely r- The proof is then

practically the same as that of w 6.
We can now express

l,(z) - ao - - a.z"
1

in the form (I) of w 7- The possible singularities of ]~(z) are now o, ~ , and
the values of z for which log z < 7 . The latter values cover the interior of

the curve defined by the equation

We suppose that o < 7 < ~ r and - - ~ < / 9 < ~ r . Then the curve defined b y
(I) has the polar equation

(2) (log ~)' = ~ ' - - O',

and consists of a single loop, enclosing the point r-----x, 0 = o, and cutting the
unit circle where 0 ~ - t - 7 . The function /(z) is regular outside this curve, and
has a branch regular at the origin and at infinity. The curve m a y be the
boundary of existence of the function: in this case the function is one-valued.
B u t in other circumstances the function m a y have other branches of which o
or ~o are singular points. ~

Io. I shall now suppose that ](z) is a branch of an analytic function, one-
valued and regular throughout the region exterior to the curve (5), including

i See pp. 337--342 of PaI~csH~.m's p a p e r in t h e Mathematische Annalen q u o t e d above.

a T h e s u b s t a n c e of t h e s e r e s u l t s is c o n t a i n e d in t h e work of L~ Re7 a n d LI~DELOt~. Cf.
L~ RoY, L c., and LI~D~LOF, Ze calcul des residus, pp. 135--136. A less c o m p l e t e result is
g i v e n by PRX~GSm~XM:see p. 46 of his p a p e r in t h e Miincl~ner 8itzungsberichte already r e f e r r e d to.
Acta mathernatlea. 42. Imprim6 le 5 mars 1920. 4~
338 G.H. Hardy.

infinity; and I shall show that in these circumstances there is an integral func-
tion a(z), of oder I and t y p e 7, which assumes the values a , a=, a3 . . . . for
Z - - I , 2, 3 , ' . -
We start from the formula (i) of w 8, and deform the path of integration
into one of the same general character as that used in w 8, b u t so constructed
as to leave the curve (2) entirely on its right. We m a y take the contour, for
example, to be formed b y part of the circle r = e- ~ where ~ > Z, and parts of
the radii ~ 4-;t, where 7 < ~ < or.This contour transforms, as in w 8, into a
quasi-rectangular contour, which now lies entirely outside the circles [ u l = 7
and [ u 4 - 2 k ~ i i = y ( k = I , 2 , . . . ) . We thus obtain

a=2-if
where ](e -u) is regular in the region exterior to the circles just referred to. We
can express ](e-u) in the form of a LAURE~T'S series

the first series being convergent for I u [ > r and the second for [ u I < 2 z - - 7;
and plainly

where now the path of integration is any closed contour at all points of which
lul>r.
We have thus proved the following theorem.
The necessary and su//icient condition that ](Z)= Za~z~ should be a one-valued
branch o] an analytic ]unction, regular in the region exterior to the curve

log = 7 9 - - 82 (o < 7 < ~),

and including in]inity, but not in any more extensive region o] the same character,
is that there should be an integral ]unction a(z), o] order I and type 7, which assu-
mes the values a 1, a2. . . . ]or z-~ i, 2 , . . .
The function a(z) is, in virtue of Wmv.RT's first theorem, unique. It is
plain t h a t the theorem ceases to be true if 7 > z . The critical curve then is
 On two theorems of F. Carlson and S. Wigert. 339

no longer a simple loop surrounded by an open infinite region, and there are
infinitely m a n y different functions which have the properties specified.
The proof of the theorem which I have given seems to be that which is
most in conformity with the general ideas of this note. But it can also be
proved by an argument more on the lines of WIO~.RT'S note and depending
upon the properties of the functions

d)" I
~,,(z)= I"z § z"z' + 3"zs + .... z~-~ z --z"

Chelsea, London, August I917.