ON THE KIEMANN-CESARO SUMMABILITY

OF SERIES AND INTEGRALS

C. T. RAJAGOPAL

(Received May 8, 1957)

1. Introduction and notation. Otto Szasz ([5], p. 1139, Theorem 4 and

p. 1223, Theorem 1) has proved Tauberian theorems for series, involving the
passage from Abel or (A) summability to each of the Riemann summabilities
(R, 1) and (RO, included in the statement:
CO

THEOREM A. If 2 a
^ 2 S summable (A) to a finite value I, and
n 2n

(TΛ) either ^k\ak\ = O(n), or 2 (k.1 - <fc) = 0(1), n -xχ>,

a
then 2 k i$ summable (R, 1) to I and also summable (Rα) to I.

It is known ([5], p. 1139, Lemma 1) that the second alternative of condi-

tion (TA) along with the summability (A) of Σ ak implies the first alternative
of (TΛ); and so Theorem A need be stated with only the first alternative of
(TΛ) which Szasz uses in his proof of Theorem A without however, explicitly
mentioning it as an alternative hypothesis. The main object of this paper
is to establish two results : (i) Theorem Γ(A) at the end, which is a gene-
ralization of Theorem A with the first alternative of hypothesis (TA), for the
Riemann-Cesaro summability (R,p,a) recently defined by Hirokawa ([2], § 1)
whose case p = 17 a - — 1 is summability (R71) and case p = 1, a = 0 is
summabity (R}), (ii) an integral analogue of Theorem Γ (A) stated as Theorem
I (A) in the last section. υ
The notation and the definitions used in Theαrem ί(A) and other integral
theorems are as follows. For a real function a(u) bounded and integrable' 0
1) It must be borne in mind that the parallelism between series and integrals
is destroyed to some extent by instances of theorems for series, such as the limi-
tation theorem for series summable (C,α;), a> — 1 ([1], Theorem46), which have no
integral analogues. Thus one of Hirokawa's general theorems ([2], Theorem 3) has
no integral analogue which can be proved by his method since it depends on the
limitation theorem referred to. On the other hand, a theorem for integrals, such
as Theorem I (A) of this paper, may present additional complications when we try
to adapt its proof to obtain its analogue for series. It may be added here that
analogous theorems or formulae for integrals and for series, wherever they occur in
this paper, bear the same number, unaccented (e.g.I, 1 etc.) or accented (e.g.Γ,Γ
etc.), according as the theorems or the formulae are for integrals or for series.
2) As in Hardy [1], integrability is in the Lebesgue sense and every integralj
is defined in the Cauchy-Lebesgue sense as lim f* .
248 C. T. RA JAGOPAL
in every finite interval of u^O, the Cesaro sum of order a > — 1 and the
Cesaro mean of order a ^ — 1 are defined by the integrals

X
(1.1) s*{u) - + Ύ)
J (u - xf a{x) dx if a > -1, s^(u) = a{u\ u ^ 0,

nd

(1.2) σ*(y) - Γ(α + 1) ^ - if a > - 1 ? O-.^II) = «s-i(»), if ^ 07

respectively. The Riemann-Cesaro transform of a(u), of positive integral
order p with index α: > — 1, is defined by
CO

/ \ TU

where
i r ^_ (either —
(1.4) - -1,
0

1 if a= - 1 .
and the Laplace-Abel transform of a(u) by

tu
(1.5) **> = / e~ a(u)du, t > 0.
0

The integrals in (1.3) and (1.5) are each 'supposed to exist for every fixed
t in the interval noted against each. There is a similarity between them in
that the existence of j(t) for every t > 0 enables us to express it as an
absolutely convergent integral in the form ([6], Chapter II, proof of Theorem
8.1):

(1.6) j{t) = tΛ^ J e-*» sΛ{u) du, a > 0.

The different kinds of summability of s(u) ~ I a(x) dx required in Theorem

o
I(A) and elsewhere are the following, / in each (as everywhere in this paper)
denoting a finite number: —
summability (C,a), a > — 1, to /: lim σΛ{u) = /,
written s{u)-+l (C,oc),
(1.7) summability (A) to /: lim j{t) = I,
written
 RIEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 249

summability (R,p,a) to I: lim ρ(p,a,t) = I,

3)
written s{u) -> I (R, py a).
An additional definition needed in the case of a particular result for s(u),
namely, Corollary 1(1), is that of (C, oo) summability which follows.
lim sup σa(u), lira inf σ»{u)
whether finite or not, are defined for every a > ~ 1 and they are, for
varying a > 0, the first a monotonic decreasing function of a and the second
a monotonic increasing function of a. Consequently
σ« = lim lim sup σΛ(u), σ«> = lim lim inf σa(u)

exist, and, whether they are finite or not, cr« ^ σoo. We define for s(u)

(1.8) summability (C, oo) to / : σ w ^σ^ = /, writing s{u) -> / (C, oo).
CO

βfe
For a series 2 supposed to be real in this paper, there are the usual
Jfc = l

analogues of definitions (1.1) (1.8). For instance,

(1.Γ) ^ =2 ( w - J + α ) α s forα>-l, tf ~ a*,

and

(1.2') σ « = -g = s- /(w +α ) for α > - 1 , σ - ' = m? «

are the Cesaro sums and the Cesaro means respectively of order a > — 1 of
a
2 k while the Riemann-Cesaro transform, of positive intgral order p and
index a > —1, is
oo

(1.3') />(ί, α, t) = C-'α ί« +1 2

3) TΛe method of summability (R,ρ, <κ) /or ί(«) « regular if either (ϊ) p>a-hl
^ 1 , or (ii) ^>>l>α: + 1^0. For case (i) the proof is that, since α^O, s(u)-+l(u->
oo) implies <ra(u)^l and the existence of
oo / sin & \^
/ ( ) u**»(μlt) du, t>0,
0 \ « /
and finally implies P (^,CB,O^^ as t-++Q by Lebesgue's theorem of dominated conver-
gence. In case (ii), we argue similarly with
! ^(^y c/o , ,
identifying this integral with p(/>,α,ί) by an integration by parts in which we use
(Ta + Ί O ) -W (M^oo).
4) The special definition of σ^1 is not adopted by Hirokawa and others for the
reason that it is inconsistent with the definition of σ* for.a>—1. However, this
definition of σ^1 enables us to unify in Theorem I'(A) the two cases <x—~l and at
>—1 although the first case often requires more elaborate treatment than the second.
 250 C.T.RAJAGOPAL

where CP>a is the constant in (1.4). The different kinds of summability of

a
2 'Λ explicitly figuring in this paper are
summability (C, a), a > — 1 to /: liαi σj = /,
CO

α ι C α
written 2 * = ( ^ )»
1
CO

a
summability (A) to /: lim 2 ^ χ1c — h
5)
(1.70
written = z (A)>
summability (R,A α) to /: lim ρ{p, a, t) =

=
written 2 ^ ^ (#£>α )

2. Lemmas. In the rest of the paper, the notation and the definitions
of the preceding section are used without further explanation. The lemmas
which follow are for integrals, and they have series-analogies which are
omitted for the reason that there is no special difficulty in either formulating
or proving these analogues. In fact, series-analogues of even the subsequent
results for integrals are taken for granted unless there is some such difficulty.
The first two lenmas have obvious proofs which are left to the reader.
LEMMA 1. If fiu) is non-negative and integrabϊe in every finite interval
o/u>0, S>0, q±Q, then

j f(x) dx = O(uq~8) implies j f(x) x~q dx = O (u~6), u ~> oo.

u u

LEMMA 2. If ί' > 0, u > 0, then

siatx dx
/ tu

LEMMA 3. If k <1, fίu) is continuous in every finite interval of u > 0,

F(u) = / f(x) dx, then f(u) - ku^Fiμ) = 0(1) implies f(u) = O{1\ u - ' OO.

PROOF. Adapting a method of Hardy's ([1], p. 107), we solve the differ-

ential equation for F(u) given by
5) The infinite series involved in the definitions of summability (A) and sum-
mability (R,/>,α), in (1.7')» are of course supposed to be convergent for the values
of x and ί,i.e. I Λ ; | < 1 and O<Z<£o> which makes the limiting operations of the defini-
tions possible. Another point to be noted is that the method of summability (R,ρ,aΐ)
for 23fc is known to be regular for either />>OJ+-1^1 or £ > l > α : f l ^ 0 ([2], Corollary
1 under Theorem 1) but not so for ρ = l^ct + \>β ([2], Theorem 2).
 RIEMANN-CESARO SUMMAB1LITY OF SERIES AND INTEGRALS 251

F{u) - ku~Ψ(u) = g(u)

and obtain
ι~kF{u) = * x~kg{x) <&r -f a constant,

whence, as a result of our assumption g(u) = 0(1), we get successively

u"F(u) = O(l) -h O(u-^\ f(u) = te-^F^) + 0(1) =
LEMMA 4. If a > — 1,

= 0(u) implies σrt+1(/ί) = O(l),

ϋ

If a— — 1, ί&£ r^sw/ί zs 5ί«7/ ίrw^ provided στ(u) = 0(1).

PROOF. For α: > — 1, we have by an integration by parts:
u u

/
<τcc(x) dx = - — — cτa +1 ( « ) H ^ — •- • I σβ+i(ΛΓ) ώf,
ϋ o
from which we get the required result by using Lemma 3 with
k = — CLt f(u) = σcc+i («).
The case a = — 1 is obvious since it is simply that s(«) = 0(1) when we
^ assume the two conditions:
u
S(u) - σi(tt) = — / ΛW(Λ?) dx = O(l), σi(«) = O(l).
u
LEMMA 5. If a > — 1, ί/tew
S(M) -> /(A). σ« +1 («) - Ox(l) O S(M) -+ /(C,« + 2)
w^ic^ means that the two hypotheses on the left side, separated by a stop, lead
to the conclusion on the right side.
OL(T) on the left may be replaced by OR(1), or a fortiori, by O(l).
PROOF. The case a = — 1 is classical ([1], p. 154, integral analogue of
Theorem 94). If a > — 1, one method of proving the result is to note that
s{u)^l (A) implies, by (1.6),

+a tu
lim ί* / e- sa+J {u)du = /,
o
and then appeal to a theorem due to Hardy and Littlewood (e.g. [4], pp.256-
7, proof of Theorem t} where r0 is used instead of a 4- 1).
LEMMA 6. If a> —1, then the two hypotheses

j s(«) -• / (A), u -> oo,

 252 C.T.RAJAGOPAL

together imply that

σ-β +iO) = 0(1), σa+2(u) - • /,

PROOF, if a > — 1, the conclusion follows at once from Lemmas 4,5.

if a = — 1, our first hypothesis is that

S(U) - cΓ! («) = -™ J Λ»(ΛΓ)rfΛ?- 0(1).

0

Therefore, by a well-known theorem of Szasz's ([5], p. 635, Theorem 1 ; [4],

case r0 = 0 of Lemma III), the second hypothesis ensures that σι{u)^Ί and
hence also that s(«) = 0(1).
LEMMA 7. If s(u)-*l (C,oo), then there is an a0 such that, for a l>oc0,

The lemma is known (e. g. [4], Theorem t).

LEMMA 8. If a >. — 1 tfw</
0), σ«+z(u

<Γβ+1 ( l l ) = 0(1)
PROOF. It is obvious from σa(u) > — H that σa+i(u) = Ox(l). On the
other hand, σ-α+2(«*) == Oi?(l) implies the existence of a constant Jί > 0 such
that, for all large x,

+ o r
i. e. .fiΓ > g«ΐa~Γ(α + 2) lσ*+ύχ/2) ^Q» o Λ+1(xf2) = OΛ(1) as
so that finally <τa+1(u) = 0(1) as
LEMMA 9. //" «s(«) > — H{H> Q), ύtwrf ^/ί/^r 5 ( M ) ^ / ( C , * ) , £ > 0,
s(w)->/(A),
then

ru
s(u)-+l, J x\σ{x)\dx^ Θ{u).
 RIEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 253

PROOF. The first conclusion, that s(u)-ϊl, is the result of a classical

Tauberian theorem, and it implies

I xa(x) dx = o{u).
ϋ

The second conclusion follows from the above estimate together with the
estimate
ru
I x[\a(x)\-a{x)]dx<2Hu
o
resulting from the relations

LEMMA 10. If a > — 1 and s(u) is summable ίC, a + ϊ | , i.e.

I |Λr β + i(Λ)| < oo,

0

then

f \<τ«(x)\dx=O(u).
o
If a— — 1, the result is the familiar one that

I \a(u)\ du < oo implies I x\a(x)\ dx = o(u).

0 0

PROOF. For a > — 1, we have successively

{a + I ) " 1 # - ^ Γo-β+iOr)] = σ-αiW — σ«+iW for almost all x,

r r r
I I σJx) I dx <Ξ / \σa+i{X)\ dx ~j- {a + I ) " 1 / x\dσa + Λtf)\
J J J
0 0 li

1
S / \<r<*+i(x)\ dx+(a + I ) " Mj \dσ*+i(x)\ =
o o

3. Tauberian theorems connecting Cesaro summability Λvith

Riemann-Cesaro summability. Theorems f, Γ of this section are similar
to the following theorem which is mainly due to Obreschkoff ' and Hirokawa
([2], Theorem 1) and which, in fact, has a case in common with Theorem Γ.
THEOREM B. IfO<B<l^p^a positive integer,
 2 54 C. T. RAJAGOPAL

then, for a such that — lS<x<p — 8 — 1,

REMARKS ON THEOREM B. (i) The theorem is due to Obreschkoff in the

cases a = - 1, p ^ 1 ([3], Satz 1), α: = 0, p > 2 ([3], Satz 4) and due to
Hirokawa in the remaining cases.
(ii) The conclusion of Theorem B can be proved for p = 1 <zwύ? α: = 0; Λ
ease m/#i w ί in the enunciation, by first using Hirokawa's arguments with
p = i} a = — 1, to show that
2 - - sinn^ [ exists for * > 0,
w
"" Π ί ** ί ί j I tends to Z as * -• + 0,
as a result of our two hypotheses and the fact

), ί>0,

which is analogous to Lemma 2 and now serves in lieu of the relation

I sin kt/kt\ = t'ιO{k~ι) used by Hirokawa. The proof can then be completed,
as pointed out by Szasz in a similar context ([5]7 pp. 1221-2, converse of
Lemma 1), by the observation that

κ
" fci nfc fci
1-δ
because sk — O(# ) = o(k\ k-^-co, in consequence of the hypothesis σ
(iii) In every case in which Theorem B holds good, it can be restated
with only (TB) changed to

the proof of the changed theorem requiring an obvious modification of but

two steps numbered (3. 3) and (3.4) in Hirokawa's paper [2]. The cases a =
— 1, p = 1 and Λ = Q,p = 1 of the theorem thus changed are due to Szasz
([5], p. 1159, Theorem A and p. 1228, Theorem 5),
Of our first pair of analogous results for series and for integrals, the
one for series, Theorem I', is stated before its analogue for integrals, Theorem
I, but proved after, so as to facilitate the comparison of Theorems B, I' and
present first the simpler of two proofs. Our second pair of analogues, of
which the one for integrals is Theorem Π and the other for series is taken
for granted, can be deduced from the first pair, and the Tauberian condition
in them is one-sided.

THEOREM Γ, If a + 1 ^ 0,
 RIEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 255

(3.10 2 <** = ι (c> a + 2\

(3.2') Σ K I =°(M).
Jfc-1-

, for a positive integer p> a + \

1=1

This conclusion is true in the case a = 0, p = 1 {ruled out above) if (3.10

(3.20 #r# assumed for a = — 1.
THEOREM I. #* α -f 1 i> 0,
(3.1)

(3.2) J [ cra(x) I JJC = O(»), u

/br β positive integer p > a + 1,

conclusion remains true for a = 0; ^ = 1 provided that (3.1), (3.2)

assumed for a = — 1.
PROOF OF THEOREM I. In all cases except the case a: = 0; p = l, the
transform /o(/>, c*;, /) of (1. 3) exists as an absolutely convergent integral for
fixed t > 0 . For, if the integral in (1.3) is taken from a positive U to oo
and the integrand is replaced by its absolute value, we get
co p CO

/
. , v, sin/w , 1 Γ |crα(^)|
by (3.2) and an appeal to Lemma 1 with f{u) = | σα(«) |, q — p — a, 8 =
ί — oc — 1.
If α: = 0, ί = 1, the existence of p'£, α, ί) for fixed t > 0 is proved as
follows from (3.1) and (3.2) each with a = — 1. The integral

s
(3.3) — Γ a(x)φ*(tx)dx=— [ a(x)dx [ ^^-du, t>0,
7Γ J 7Γ J J U
0

is absolutely convergent, since

J
by an appeal to Lemma 2, followed by (3.1) with α = — 1, and an appeal
to Lemma 1 with /(#) - x\a(x)\, q — 2, δ = 1. Since, as a result of (3.1) and
(3.2) each with a = — 1, s(w) = 0(1) by the case α; = — 1 of Lemma 4, we
 256 C.T.RAJAGOPAL

get from (3.3), by an integration by parts,

(3.4) — I a(x)φ*(tx) dx = — / s(x) ? i r 1 ^ ^ = p(l, 0, t), t > 0.

ϋ ϋ

We proceed to show that in all the cases considered above, ιp(p, a, t)->l
as t ->• + 0.
Case: p > a + 1 > 0. Writing φ(u) = (sin u)p/up, we get by two integrations
by parts, for fixed t > 0 and # > 0,
+1
pip, a, t) = C"i ** ( J + J y*(u) φ(tu) du
0 xlt
ίt-a /f

?a;+ J (w) φ ( / « ) — 5OJ + 2 ( W ) — — φ(/»)

4- C-^ ί*+1 Γ J 5 Λ + 2 («) - ^ φ(tu)du + J s«ίw) φCί«) <ίw 1

(3.5) - C-V7, + 72 + I3 + h)
where

We shall suppose (as we may) that t < 1 and ήx x subject to certain conditions
to be specified presently. To begin with, (3.2) and Lemma 4 in the case
a > — 1 show that <r*+i (#) = O (1) as ^-> oo and that there is a constant M
which makes
(3.6) 1/χl <Mx«+i\φ(x)\
for all large x; while (3.2) and Lemma 1 with f(u) = |.σ*(«j|, a-p—a,
$ —p — a —I, show that there is also a constant iV making

(3 7)
J
for all large ΛΓ. On the other hand, (3.1) is o-<*+>2(u)^l and it gives, when
x is fixed and t -> + 07
 RIEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 257

so that, by the definition of CP,a in <Ί. 4),

(3.8) l 2 + l 3
- C p
^

Given any small 8 > 0, x can be chosen so large that the right-hand members
of (3.6), (3.7), (3.8) are each less than C ^S/S in absolute value. With this
choice of x, (3.5) gives when taken along with (3.6), (3.7) and (3.8):
lim s u p I p(p, a,t) — l\<S, i . e . lim pip, a, t) = L
t-»+0 ί-^ + 0

Case: p > 1, a = — 1. (3.5) will now be replaced by

(3.5i) pip, - 1, t) = Itl) + It1* + A'n + Itτ)
where 7{-]>, I^\ I^l\ l\~ι) are Iu L>, I3, I4 respectively with a = -1, so
that
\l[-l>\ <M\φ(x)\, \2l-»\<Nχ-»,
for all large x and when ί -> + 0, Λ: remaining fixed,

/(-υ _ Ixφ'(x), 73c-]) -> Γ /«φ"(«) du = /Λφ'(Λ?) - /φ(Λ?) 4- /φ( + 0).

o
Substitution of these estimates for I\-χ\ l\~λ\ /J"", /S" υ inί3.5 x ) leads to our
conclusion in the penultimate from
lim sup \p(p, - 1J) - Iφ( -f 0)[ < Λf IΦ(ΛΓ)| + Λ^r-p + I/φte)| < θ
since evidently ΛΓ may be supposed to have been fixed so as to satisfy the
last inequality.

Case: p = 1, a = 0. We see from (3.4) that the treatment of this case is

like that of the case p - 1, a = —1 with the difference that we have now
2π~1φ*(u) instead of φ(u). Hence our conclusion will now assume the penul-
timate form

lim sup I |p.Ί, 0,t) φ*(0) I < M\φ*(x)\ + Nx'1 —Φ*(χ)

7Γ

a choice of x satisfying the last inequality being passible by Lemma 2.

PROOF OF THEOREM Γ. The case α + 3 ^ ^ > α : - h l > 0 i s reducible to

Theorem B with its two hypotheses replaced by the single stronger hypothesis
σζ'δ'ι-^l as n~+co. For, (3.2r) implies, by the series-analogue of Lemma 4,
σ j + i Ξ σ ; - ^ O [ l )
which together with (3.V) .'leads to σl~&~Ύ ->• / for every
positive δ < 1 by a well-known theorem ([1], p. 127, Theorem 70). The case
p-2,a~—l follows from the case p = 1, a = — 1, since we have, assuming
the result for p = 1, a = — 1,
258 C.T.RAJAGOPAL

*=1 fc-1 fc=l * - l

it being well-known that (R,l) c£>(R,2) f[l], Appendix III). Hence we may
suppose that tt + 3 * i > t t + 1 ^ 0 .
Cΰts^: α + 3 Φ ί > α : + l > 0 . After establishing the existence of the p(p,
a, t) in (1.3') as in the proof of Theorem I, we proceed to the step correspo-
nding to (3.5) there, which is now obtained by two Abel or partial-summation
transformations, in the form

p(p, a, t) = C;,« * α + 1 s Γ 1 Φ(.mt) + / α + 1 s*ΐί Δφ(?^ - lί)

ί
" I " Ίr J^^Λ k
Jc - l fc=>m+l

sa
= Ci!JA + ϊι + ί + ^)> y>
where m = the integral ["part of xft, x > 0, t > 0. By (3.2') and the series-
analogue of Lemma 4, σ-*+1 = Oil) as ^->co ? so that, as in the proof of
Theorem I, there is a constant M such that
(3.6r) \K\ < Mxa+I\φ(x)\
for all large x. By (3.2') and the series-analogue of Lemma 1, there is also
a constant N such that
(3.7') \Ϊ4\ >NxΛ+'ί'p
for all large x. lί x is fixed, using the hypothesis (3. Γ) that σ%+2 ->• / as
k~+co} we can show that
(3.8') Km+ sup I/; + / 3 - CPt*i\
CO

|Z|
*"- ^r^ + 1 !Φ^)l + ' ' l

where KΛ:) is a certain positive function of x tending to 0 as #-*oo. After

this we can complete the proof like that of Theorem I, postulating that x
has been chosen so as to make right-hand members of (3.6'), (3.7') and (3.8')
each less than C-^θ/3.
It remains to show how (3.8') is reached. Fix) having been defined as
in (3.18') below and x > 1 having been chosen as indicated above, we can
state the hypothesis (3. Γ) in the form
+2 +2 +2 2
s% = lA» + 8tiAΐ , 8*1Aϊ+ !&*+*-*§ as &-+00,
and find mQ so that, for k^nh + 1,
+2 A
(3.9') s£ = IA£+' + 8*2Aϊ+z, \SJC lA%+2\ < χ-n«+3-*\ ka+'\
6)
With this choice of m0 and with m = the integral part of x/t > m0 + 2 in (S. 5') ,
6) Since finally ί->+0, the supposition regarding m, viz., that m^/w»-ι-3, is
valid.
 RIEMANN-CESARO SUMMABIHTY OF SERIES AND INTEGRALS 259

we can write
Ztl Aφ{m - If) + t«+1l£m^ A%t:{ Aφ{m - It)

k=ι

(3.100 = 4 + 4 + 4 + 4 + 4, say.
Since there is a constant K such that |Δaφ(*f)| < KP-pk~p for all ί > 0 and
^ > 1 ([3], p. 443), we first get, using (3.9') in /^

Since there is evidently also a constant Kx such that 2 ^α+2"P < KιmΛ+z'p,

the above step gives us

[41 < KKάtm)***-*χ-*ι«+s-Pi
where tm-+x as / ^ + 0. Thus
(3.1Γ) lim sup | / 3 J $ AXΊ ^»+3-i»-2|*+3-jpi < j g y ^ι«+3-pi β

Secondly, there is a constant L such that \Aφ(kt)\ < Ltι-pk~* for all ί > 0
and k > 1 and so we find, recalling that

whence follows, exactly like (3.1Γ),

f
(3. Ί2 ) lim sup \ϊ^\ ^LX*+2-P-2\*+*-P\ ^Lx-i-\*+3-p\m
2
Thirdly, \A φ(kt)\ = O(if ) as ί-* + 0 where the constant implied by O(t2) can
2

be chosen to be the same for £=1,2,3, , mo; while /£fcAb*+a/#*+i*is

bounded for & ^ 1. Hence there is a constant Li such that, for all sufficiently
small t,

(3.13') 14) < Lr f+* 2 ** + ! l , i e. lim J41 = ®

Lastly, we have, by two Abel or partial-summation transformations,

m
(3.14') / a + / ; = f-* 2 ; -
+

fc-1
where, since mt-^x as £ -* + 0,

and further, as shown by an argument elsewhere ([3], proof of Lemma 3),

(3.16') t«^A%
 260 C.T.RAJAGOPAL

From (3.14'), (3.15'), (3.16') it follows that

>
£5 «•

(31r c
> • " " -

Using (3.11'), (3.120, (3.130, (3.170 in (3.100, we finally obtain (3.80 with
(3. 18 ) F{X) = KKτ *H*+3-p| + /^l-lβH-s-*!

where K,KΛi L are absolute constants. This completes the proof as already
explained.

Case: a + 1 = 0, 2 * ^ > 1 . We merely put a + 1 = 0 in all the steps of

the preceding proof except (3.80, (3.140 (3.17'). The excepted steps, after
the changes obviously necssary, culminate in the following steps instead of
(3.170 and (3.80:
lim {ϊ.λl -f TΆ) = / — lφ(x), lim sup \ΪQ -f /' — l\ <J \lφ(x)\ -h Fix),

Otherwise the proof is as before.

Case: a = 0, ^ = 1. The passage to this case from the case a = — 1,
^ — 1 is as in the proof of Theorem I with the necessary modifications for
series indicated in Remark (ii) following Theorem B.

DEDUCTIONS FROM THEOREM I.

COROLLARY 1(1). If s(u) -> / (C, oo), then, for all large a, say a > aOt s{u)
-> / (R, py oi) if P is a positive integer chosen corresponding to each a, so that

PROOF. By Lemma 7, the hypotheses (3.2), (3.1) of Theorem ί now obtain

in the stronger form
aju) = 0(1), σa + ι(u)-+l,
Corollary 1(1) invites comparison with the known result a > ct0. ([4],
Lemma II) that, if s(u) ->• / (C, oo) then, for all a ^ oc,h

W, t -> + 0.

COROLLARY 1(2). If a^ —l,s(u) is summable \C,a + 1| to I, then s(u) is

summable (C,p, a) to I, if p is a positive integer such that p > a + 1.

The conclusion is true for p = 1, a ~ 0 on the hypothesis that \ a(u) du

is absolutely convergent.
PROOF. The two hypotheses (3.1), (3.2) of Theorem Γ are ensured, the
 RΪEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 261

first one obviously and the second one by Lemma 10 and so every case of
Corollary 1(2) is reducible to the corresponding case of Theorem I.
THEOREM II. If a > —1, H > 0, then, for a positive integer p > a + 1,
s(u) -* I (C, a + 2). σ-cc(u) > -H φ s(u) -+ / (R, A α).
The above result is additionally true with a = — 1 in the hypothesis (left
side) and a = 0, p ~ 1 in the conclusion (right side).
PROOF. If a > —1, Lemma 8 shows that σ-«+i(«) = 0(1) and that con-
sequently
u u

/ WJp) + H\dx= ~^~-[σa+ι(u) + H]+ - 7 T - / l<rm+1(x) + Htfx = (XI).

a. -\- 1 a +1 uJ
0 ϋ

Thus Theorem II is simply Theorem I with σa(x) 4- H instead of <ra(x) in (3.1)

and corresponding changes elsewhere, in particular, with the conclusion, for
p > a + 1 > 0,

J „ uΛ Ί/sintuγ (exists for ί > 0,

Γ(Λ + 1 ) J V tu ) ^ 1
o
This is the conclusion sought since the terms in H can be removed from both
sides.
If a — —1, Lemma 9 gives

= O(u)
/
which together with the hypothesis s(u) ->• / (G, 1) shows that the case p >
cc + 1 = 0 of Theorem II is included in the same case of Theorem I.
Finally, the additional result follows from the case a = 0, p = 1 of
Theorem I.

COROLLARY II. For s(u) > — H, summabilities (C, 2) #M</ (R3) are equivalent.
PROOF. For the class of s(u) in question, we have, taking a = 0, /> = 2
in Theorem I,
(C?2)C>(R3)
while, by a known theorem ([1], p. 305, Theorem 237),

4. Tauberian theorems connecting Abel summability w i t h Rie-

mann-Cesaro summability. The theorems of this section are derived from
those of the last section.

THEOREM KA). Theorem I can be restated with the hypothesis

(C, a -h 2) of (3.1; changed to s(u) -* / (A) and no other change.
PROOF. From (3.2) and s(u) ->- KA), it fellows, by an appeal to Lemma 6,
 262 C.T.RAJAGOPAL
that s(u)->l(C}a + 2). Theorem Γ(A) is thus reduced to Theorem I.
The integral analogue of Szasz's Theorem A at the outset is generalized
in the following corollary obtained by taking a = — 1 and a = 0 in Theorem
KA).
COROLLARY I(A),

either

THEOREM II(A). Theorem II «s /f we «;#/& sv'») -• /(C, a + 2) replaced by s(u)

PROOF. For α: :> —1, the hypothesis crΛ>) > —77 gives σα+i(M) = OΛ1)
which, in conjunction with the hypothesis s(u) -> / (A) leads to s(u) -> /(C, α: +
2) by Lemma 5. And so Theαrem fI(A)is reducible to Theorem II.
Whether for series or for integrals, the known theorems ([1], Appendix
III) connecting each of the summabilities (R, 1), (R,2), (R2) with summability
(A) are that
(R, 1) - ^ (R, 2) -++ (A), (R,) ^ ^ (A).
There are non-trivial converses of these theorems for integrals, under a one-
sided Tauberian condition, contained in Theorem Π(A) and explicitly stated
below.
COROLLARY Π(A).

s(u) -> I(A). s(u) > -H-+^ s(u)

The series-analogues of all the results in this section can be stated and
proved on the same lines as these results. In particular, Theorem I(A) has
the series-analogue which is given below and deduced from Theorem Γ ex-
actly as Theorem I(A) from Theorem I.
THEOREM Γ(A;. In Theorem V, the hypothesis S ^ = /(C, a + 2) of (3. V)
can be changed to 2 #& = #A) without any other change.
In conclusion, I wish to thank Professor V. Ganapathy Iyer for generously
setting apart time to serutinize the manuscript of this paper.
7) On the hypotheses of Corollary I(A), s(u) is summable (C,δ) for every δ>0
(cf. [5], p. 1141, Lemma 3), but not necessarily convergent. The negative part of
this statement has been demonstrated -by Szasz by means of an example for the
analogous case of series ([5], p. 1145). In this analogous case, there is also an ad-
ditional conclusion proved by Hirokawa ([2], Theorem 4), namely, that the series
considered is summable (R,l,αO, - 1 < O J < 0 .
 RIEMANN-CESARO SUMMABILITY OF SERIES AND INTEGRALS 263

REFERENCES

|Ί] G.H. HARDY, Divergent series (Oxford, 1949).

[2] H.HIROKAWA, Riemann-Cesaro methods of summability, Tδhokα Math. Journ.,
(2), 7(1955), 279-295.
[3] N. OBRESCHKOFF, Uber das Riemannsche Summierungsverfahren, Math. Zeits.,
48(1942-3), 441-454.
[4] C. T. RAJAGOPAL, On Riesz summability and summability by Dirichlet's series,
Amer. Journ. Math., 76(1954), 252-258.
[5] O.Szasz, Collected mathematical papers (Cincinnati, 1955).
[6] D.V.WlDDER, The Laplace transform (Princeton, 1946).

RAMANUJAN INSTITUTE O F MATHEMATICS, MADRAS ( I N D I A ) .