ANNALES SOCIETATIS MATHEMATICAE POLONAE Series I : COMMENTATIONES MATHEMATICAE X I I I (1969)
J. Kopeć (Szczecin)
Some criterions of Norlund summability
Let 8 be the class of all real-valued functions bounded and inte- grable in each interval <0, a > (a > 0). We shall consider the linear method
oo
of summability Ta{ank} of the series (*) un whose coefficients are defined by the formula:
n = 0
(1) ank = exp j — j a{t)dĄ, n = 0 , 1 , & = ere 8 .
n—k
Write
П s
^ uk — the тг-th partial sum of series (*),
(2) k=0
tn = ankuk — the w-th Ta-mean of series (*).
k=0
Series (*) is said to be Ta-summable if the sequence {tn} is convergent.
It is clear, that for each positive sequence {Pn} there exists a function П
a (x )e8 such that Pn — P 0exp(J a{t)dt) (e.g. o{x) = L '(x), where у = L{x) о
is the equation of the polygonal line joining the points (n—1 , k i P ^ ) , П
(n, In P n), n = 1 , 2 , . . . ) . Hence if P n = p k > 0 {p0 = 1), then Pa-sum- k =0
inability and Norland (Ж, p n)-summability (where ank = Pn~klPn) are equivalent. In this note we shall show, that using form (1) of Norlund method it is possible to simplify several proofs and to get some generali
zations of known theorems.
1. Regularity. It is well known (see e.g. [7], p. 127) that the (W, pn)~
method is regular if and only if the following conditions are satisfied:
lim Pn/Pn+i = 1, Ш = 0 {P n).
7c=0
(3)
I t follows from (3) that for TCT-method the following conditions of regularity hold.
1.1. Suppose that a{x)€S and write a_ (x) = min [(/(a?), 0]. I f
x-\-l со
lim J o(x)dx = 0 and f \a_{x)\dx < oo, then T„-method is regular.
x—нэо x 0
n +1
P ro o f. Since P n/Pm+1 = exp(— f cr(t)dt), the first condition (3) is П
satisfied. Writing a+{x) — тах[сг(ж), 0], we have for Tc = 1, 2 , . . .
к x
\Pk\ = \Pk~ Pft_il = I J o'(a?) exp (J a(t)dtj dx |
k-l о
к x
< J [<r+(a?)+|<r_(a?)|]exp(J a+(t)dt\ dx,
k-l 0
n n . Ж o o »
J" cr+ (a?)exp a+(t)dt)j dx-\- J |o"_ (a?)|da?exp|J* a+ (t)dtj
fc=l о ‘ о о о
n w.
= o j e x p j j a +(t)dt^ = 0 {e x p (J 0 (Pn).
о 0
П
1.2. I f Ta -method is regular, Йетг liminf J a{t)dt > — oo.
те—>oo 0
vn vn
Indeed, if lim J a{t)dt — — oo, then lim P Vji = 0 and £ \Pk\ > 1
и—>oо 0 n—>00 k —0
(p0 = 1). Therefore conditions (3) cannot be satisfied.
00
In particular if Pa-method is regular and J a+{t)d t< oo, then
oo 0
J |<7_ (£)|d< < oo.
0
2. Auxilliary lemmas. Suppose that a e S and that q is an arbitrary real number.
2.1. I f g ^ O and limsup(a?+l)1+ecr(a?) < oo, then there exists a con-
x—>oo
stant C > 0 {depending on q) such that
1—e x p ( — f a(t)dt)^ G '--- (0 1 J k 1 ( n + l ) ( * + l - f c ) e ^ Proof. I t follows from Bernoulli inequality that
1 - [ 1 - fc/(w+l)]g < 1 - [ 1 - Щ п + l)][g]+1 < — - - . n -\-1 Consequently, we have for q > 0
1Ш +.Р * -
«_* £ (w— J c+ lf q (w+l)(w—& + l)e
Therefore applying the mean-value theorem we obtain
ak \ к
1 —exp ( — J cr(/)dfój< 1 — expl ■I
n—k ' (n-\-l)(n—k-Ą-l)e < 0
( n + l) ( n — & + l)e and since a > 0, C is also a positive constant (depending on q). If q = 0, then a{x) < 5 ( # - j- l )-1 (В > 0) and
* n i -j } В
exp|— j a{t)dt^^\l- к
n - k ' '
n
Hence 1 —exp(— J a(t)dt) < (/^/(w+l).
n —k
2.2. I f q > 0 and liminf (x-\-l)l+e o(x) > — oo,
— >oo
&
1 —exp( — J a(t)dt^ > — Ox-
■ (n + l){n -\ -l-T c)e The proof is similar to that of 2.1.
2.3. I f limiiif (x-\-l)a(x) > then
(С г> 0).
n r
exp|— J o(t)dĄ = 0 \
n—k L
n —k-\- I T
---- — — ( k ^ n ) . n + 1 J
Proof. It follows from the assumption that there exists a constant X = X{fi) such that a(x) > jaf(x-j-l) for x > X. Let be n > X and 0 < к < n —X , thus
exp|— J a(t)dłj
n—k
n —k + i v n + 1 / * If n - X < к ^ щ then
Tb Jl lb / y’ i -1 \t*
exp|— J сг(£)<й| < exp|J \a{t)\dłj exp( — J a(t)dtj <
(Cx = (7x(-3^ “hi)** for /л > 0, (7X — Ox for [j, < i 0).
3. Numerical series. In this section we shall prove some theorems of tauberian type.
3.1. Assume that the function o(x )eS satisfies the following conditions:
(4) limsup (x -f 1) a (x) < o o, liminf (ж-f- 1)1+^<т(а?) > — oo
SC—>00 —>OQ
for some real /3 > 0.
n oo
I f lim (w+1)-1 £h\uk\ — 0 and the series £ un is Ta-summable, then
n—yoo к=1 n=0
it is convergent.
Proof. From (2) and (1) we have:
n n n
\tn—*nl < \ank—1| \uk\ = ^ | l — e x p (— J
k= 0 k=0 n—k (
Hence from 2.1 for q — 0 and from 2.2 for 0 < q < /3 we obtain П
\tn—sn\ = 0 { { n + l) - 1 ^ h \ u k\).
k—0
n
E e m a r k . The equality lim (w +l)-1 huk = 0 is the necessary and
łl—>-oo &=0 oo
sufficient condition of convergence of the (C, l)-summable series £ un
n = 0
(see [2], p. 524).
OO
3.2. I f <y(x) satisfies conditions (4), series (*) uk is Ta-summdble
oo к =0
and £ k u k < oo, then series (*) is convergent.
k = l
Proof. Cauchy inequality yields: t
n W 1 И
— Y'lc\uk\ = — — У &1 uk\-\--- — У Tc\uk\
4-1 jLj n -\-1 ^ n -\-1 .
n 4-1 k = l k = 1 k = N+ 1
1/2
L 1 \fc=A7+ l ’ ^ -*=V+1wj-1 ’ J
Since
n
П
Ш
k = N+ 1. 4 1/2 ^ V21 and ^ Jc u l < e k = N+ 1for N sufficiently large, the assumptions of 3.1 are satisfied and series (*) is convergent.
E e m a r k . Theorems 3.2 and 3.1 are generalizations of Lemma 5 of [4].
4. Summability of orthogonal series. Let the series
OO
( 5 ) ^ C n<pn(x)
n = 0
with coefficients {cn} e l2 be orthogonal in the interval < 0 , 1>. Moreover, let sn(x) be the n-th partial sum and let tn(x) be the w-th Ta-mean of series (5).
4.1. Suppose that the function a(x)ęS and that conditions (4) hold.
I f the sequence {nk} of indices satisfies the condition of lacunarity nk+1/nk
^ q > 1 (fc = 0 , 1 , ...), then the series OO
(A) — tnk(oc) ] 2
k=1 is convergent a.e. in <0 , 1> (x).
Proof. Prom (2) and (1) it follows
1 nk nk nk
j [8nk {so)- tnk(oc)]4x = £ [1 - ankijy c} = ^ [l — exp ( - j о {t)dtffc}.
0 j = 0 ? =0 nk- j
In the same way as in 2.1 we obtain
(A') J r * J [snfc {so) — tnk (ж)]2dx k=1 0
oo nk
'fc=1 7= l Since
2 PI(nk +\Y < у «*/(«*+*+ 1)* < У 3’ 2'
series (A') is convergent and according to Levi’s theorem series (A) is convergent a.e.
4.2. I f the function o(x )eS satisfies the conditions
(7) liminf{х + 1)в{х) > 1/2, limsup(a?+l)2|<T(a;)— o {x —1)| < o o,
Ж—^ОО «Г ) 00
then the series
OO
(B) ] ? n [ t n(x) — tn_ x{x)Y
n = 1
is convergent a.e.
Proof. From (2) we have
1 i n П— 1 4
J \tn{x) — tn_1{x)Ydx = § { ^ a nkCk<pk{x)— an-i,kCk<pk{x)f dx 0 * = 0
n—l
k=0
— ^ ‘ ®п_1)к^ ск-\-anncn.
k=о
(x) It is a generalization of Theorem 3.1 of [3], where it was assumed that o{% )> 0 instead of the second condition (4).
Write в(х) = a {x —1) — a{x). It follows from (1) that
iV 71— 1 П
[ank — an_ 1>fc]2 = exp( —2 J cr(tf)dtf)|l —exp ( — J a(t)dt.+ f
n—k n—k—1 n—k
n n
= exp| —2 j a(t)dtj[l — e x p J 0(£)dż)J2.
n—k n—k
If [z< liminf(х-\-1)а(х) and limsnp(a?+l)1+e\в(х)\< oo > 0), then
X—K50 0С-УОО
from 2.3, 2.1 and 2.2 (for в instead of a) we obtain
\0/nk a n- i , k ] 2 —
Consequently
k 2(n — &+1)2/*—2q
(W + l),2 ^ + 2 < n = 0 {(1 I+1)-2"}.
(8)
1 n
J lAO») — tn_x{x)Ydx = o j j T
о 'fc=l
k2( n - k + l f * ~ 2e (w+1)2^+2 4 l Putting q = 1 we have
(BO (•>]•«*» = o { £
4
n = l о v 4 = 1 k = l \ > / '
= О
oo 00
\ Z * * I
4 = 1 n = k+ 1
( n - T c + l f * - 2) ( n + l)2M+1 J For 1/2 < p < ± the function Фк{х) = х2/л 2/(a? + ft)2'“+1 is decreasing-
~ 1
for all к > 0 and x > 0 and J 0 k(x)dx = уу В(2/л—1, 2). Hence series
0 ™
(B') is convergent and series (B) is convergent a.e. in <0,1>.
B e m a r k . The assumptions of 4.2 are satisfied e.g. if e{x) = const
> 0, or if a(x) — C—lIx. In these cases the TCT-method is not regular..
X
If we suppose additionally that the condition of regularity lim f a{u)du
X-+OQ x —l
= 0 holds, then Theorems 4.2 and 3.2 of [3] are equivalent. Indeed, suppose П
that conditions (7) are satisfied, P n —exp(J a{t)dt), B n = pn/Pn = (Pn—
n 0
—P n-i)IP n = 1 — exp (—J a(t)dt). In [3] it was shown, that n(pn —
n—1 n
—Vn-i)lVn = n 2(P n—B n ^ In R n + n R n ^ , B n—Bn_i = pn f d(u)du and
n n n— 1
fin = —exp(— f o(t)dt—ftn f 6(t)dt) (0 < # я < 1).
n -1 П—1
According to (7) and the additional assumption, we have 1 — exp (— J a(u)duj\\ = a?|J [o (t—1)—<r(<)]exp j a{u)du^dt
X — l X t— 1
oo
= 0|жJ |(т(<—1) — cr(i)|cfój = 0 ( 1 ) .
Therefore lim (3n = —1, w2 |i2n— j — 0(1), nBn = 0(1). More-
Ю—>-CO Ж
over, liminf > 1/2 since liminf# fa(u )d u > 1/2 . Hence {pn}e9JT and
П-УОО Х — УОО X — 1
a > 1/2 (see [4]). Conversely, if {рп}еЭДГ, у = L(x) is the equation of the polygonal line with tops in points (n, ln P n) (n = 0 , 1 ,. . .) , then putting a(x) — L' (x), n = [#] we have
( # + l)2\a{x—1) — a(x)\ < (тг + 2)2 In-Ł П+1Х 7)P P P
-L m
{n-\- 2)2 In(-f*n “Ь^тг+ l ) (-^re P%
~ "p2-1 -и.
(n-f 2)2 In 1
0 (w-f 2)[j?n+1--;pJ (tt + 2)j»n+1
Pn+l Pn + (w + 2 )i?n+1(w +2 )j?,
P fi+l Pn PnPn + 1
Pr> P i
= 0(1).
I t is known ([3]) that in this case liminf (x-\-l)a(x) > a. Therefore
x->-oo
conditions (7) are satisfied.
4.3. Suppose that creS and liminf(#-f-l)cr(#) > 1/2, limsup(#-|-l)cr(#)
x—>oo X—>oo
< oo, limsup(# + l ) 2\a{x—1 ) — o(x)\< oo. Series (5) is Ta-summable a.e.
X —M X)
i f and only i f there exists an increasing sequence of indices {nk} such that 1 < q < nk+ilnk < r (q, r — constants) and the sequence {sn&(#)} is con
vergent a.e. in <0 , 1>.
It follows from the assumptions that 4.1 and 4.2 hold. I t is well known ([4] and [3]) that 4.3 is a consequence of these theorems.
4.4. Suppose that a(x) satisfies the assumptions of 4.3 and that the increasing sequence o f indices {nk} satisfies the condition r > nk+ljnk > q > 1 (h = 0, 1 , ...). Moreover, let pn = max (it: nk < n] (k = 0 , 1 , . . . ; n0 = 1).
OO
I f the series JŚ/ c2nln2pn is convergent, Йе?г series (5) is Ta-summable in
71= 1
<0 , 1> a.e. (2)
Indeed, from the definition of pn it follows that pn+l > pn and
nk+1 °°
Pnk = k. Therefore, writing Ak = ( ^ d)lj2 we have £ Aklnzk =
v = nk +1 fe=2
(2) It is a generalization of Theorem 2 of [4].
oo nk + \ °o
— Cv ^ 2 f c2v\t\.2[i v < oo. Hence the sequence {snA.(a?)}
fc = 2 »»=n^. + l v= n 2+ l
is convergent a.e. in <0 , 1> (see [1 ], p. 79, Satz 2.3.4) and according to 4.3 series (5) is Ta-summable a.e.
4.5. Suppose that the function a{x )eS satisfies conditions (7) and let tfn{x) be the n-th Ta-mean of series (5). I f a1(x)eS satisfies conditions (4)
OO
and the series £ \fn{x) — C -i(#)] is Tai-summable a.e., then series (5) is
* П— 1
Ta-summable a.e.
Proof. Write rn(x) = tn{x) — tan_г{х), r0(x) = 0,й. = 1 , 2 , . . . It fol-
OO
lows from 4.2 that the series m2n(x) is convergent a.e. Moreover, the
oo n = 1
series 21 r n{ % ) is PCTl-summable a.e. Theorefore, according to 3.2 it is
n =1
convergent a.e. Consequently series (5) is T0-summable a.e.
OO
5. Absolute Ty-summability. The series (*) 2! un is said to be
0 oo
absolutely Ta-summable (\Ta\-summable) if the series 2 \tn— tn_i\ is con-
n = l
vergent (tn is the w-th Тст-теап of series (*)). Suppose, that r < 1 is real and that the function ereS satisfies the following conditions:
lim inf(x-\-l)o(x) > 1/2 —r, Iimsup(a7 4- l ) 2_r|cr(a7—1) — a(x)\ < oo.
*C— ^oo 32——►OO
We shall consider the [TCT|-summability of the orthogonal series (5).
2W+1 oo
5.1. Write Am — ( 2 Tc2c l f 12. I f the series 2 2m(r l)Am is convergent, then series (5) is \Ta\-summable a.e.
Proof. It follows from (2 ), (1) and (8) for 1/2— r < liminf(a? + X-+CO -i-1) cr (at?) and q = 1 — r that
oo 2W+ 1 1 oo 2m + l 1
у 2 f l("(®)-*» -> (*) I < У {2” 2 11/2
m—0 n = 2m+ l 0 W =2m + 1
2 ^ + 1
= o { y [ 2” У у )ć2(«- ’>(n+ l) -2“- !d]1'2}
m=o n=2«t_|_i fc= 1
oo 2m+ * n
= 0 I ^ [ 2_<2P+1)“ 2
У ( » - ' И - 1 ) 2('‘+’'“ 1)* 24 ] 1'2}m—0 n=2m+l k = l
oo 2m+ 1 2m+ 1
= o j y [2- (2в+1>”* ( У й2с|) ( У
т=о fc = i n= 1
оо 2Ш-)-1 00
= о { у 2"<'-1) ( y k2cl)v) = 0 { у 2га<г-1Ыт} •
ft=l
and
o o 1 o o
(9) 8 = M®)-«^,(®)|d® = o j ^ 2m(r- 1)A m).
n = 2 0 m=Q
oo
Hence the series £ \tn(®) — tn-i(%)\ is convergent a.e.
n=2
2 ^ 4 ’ 1 o o
5.2. Write A'k = ( £ c2)1/2. I f the series £ is convergent,
7 = 2 fc+ l n = 1
tfrsw senes (5) is |Ta|-s^mma&Ze «.e.
Proof. Since
m 27 ' + l m m
^ = {o?+ % у v Ą yz« |Cli+ ( у 2и+! у « * Г < м + % 2f+lA't>
f—° * = rf+l 7 = ° * = 2? + l t=0
it follows from (9) that the following relation holds
o o oo m
8 = 0 { £ M + £ £ 2 f+1A'i\.
m = о m = o / = 0
Since r < 1, the first series is convergent. Changing the order of summation in t h e ' second sum of 8 we obtain
OO o o
8 = 0 { E 2>A'< = 0 { ^ 2 ,rA',}.
7 = 0 m > ? 7 = 0
E e m a r k . If {pn}eM a, a > 1/2, then the convergence of the series
OO
£ A'n is the necessary and sufficient condition of |JV, p n | - summability
n = l
proved in [5] (3).
References
[1] G-. A le x its , Eonvergenzprobleme der Orthogonalreihen, Budapest 1960.
[2] K. K n o p p , Szeregi nieskończone, Warszawa 1956.
[3] J . K o p e c, On some classes o f Nórlund means, Bui. Acad. Polon. Sci. 16 (2) (1968), pp. 93-98.
[4] J . Me der, On the N órlund summability o f orthogonal series, Ann. Polon. Matli.
12 (1963), pp. 231-256.
[5] — Absolute N órlund summability and orthogonal series, ibidem 18 (1966), pp. 1-13.
[6] — Some theorems on the N órlund summability o f orthogonal series, Bull. Acad.
Polon. Sci. 16 (3) (1968), pp. 201-207.
[7] K. Z e lle r, Theorie der Lim itierungsverfahren, Berlin 1958.
(3) In the note [6] Prof. Meder proved later that this theorem is valid also if { Pn} е$Ша and 0 < lim sup
n —> oo
n \p n — p n -l\
Pn — в < a — liminfnpn