ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO Séria I: PRACE MATEMATYCZNE X IX (1976)
P. P
ych(Poznan)
The D en joy integral
in some approximation problems, IV
1. Preliminaries. Denote by В* the space of all 27r-periodic functions / integrable in the Denjoy-Perron sense on the interval <0, 27
t). Consider the points æ for which
h
(1) (B*) J (f{œ + u ) —f{œ))dv, = o(h) {h->0 + ),
- h
and write
w(f,oc-, h)D = sup {—
0< t < h l 41
for any f e B*. Evidently, the function w(f,cc', h)D is non-decreasing in h and, by (1), w { f , x ; h
)D - > 0( h 0 + ) almost everywhere.
Let 0 < t < 7u and let
t
(2) f x(t) =/(<» + « )+ /(я» -« )~ 2 /(ж ), Fx(t) = (B*) j f x(u)du.
0
(-£>*) J (f{<D + u)-f{a}))du
We have then
t
(3) \Fx(t)\ = |(D*) f (f{a> + u ) - f ( x ) ) d u \ < 2tw(f, X) t)D.
-t
In this paper we consider some linear operators defined by I)*-integral.
We give estimates of the rate of pointwise and mean convergence of these operators. In order to show th a t our results cannot be essentially improved we introduce the class B(Q) of all functions f e B * such th a t
w ( f , æ : h ) D
sup --- < 1 at a fixed w, o</i=ot В (Ji)
where Û is a non-negative and increasing real-valued function defined
on <0, 7i> with i2(0) = 0.
For 27t:-perio(lic and Lebesgue-integrable functions, in symbols fe L, the corresponding estimates will be given in terms of the expression
t
\ 1 Г
J i ) l = SUP Ьтг \f(x + u ) - f ( x ) \ d u 0 <t^h I J
used in [1]. Clearly, w{f, x\ h
)L - + 0(h—> 0 + ) at every point x for which
h
f \f(x-\-u)—f( x ) \ d u = o ( h ) ( h 0 + ) . - h
Moreover, w(f, x\ h)D ^ w ( f , x; h)L.
In the sequel M x, M 2, ... will denote the suitable positive constants.
2. Pointwise polynomial approximation. Given any
f eD*, we consider trigonometric polynomials
UJx-, f )
71
= - ( - # * ) f /(«){* +
7Г J
— 7Г
'S
1A^ao&k(t — x)}dt,
*=i
where the factors Яр* (k = 1, 2, ..., щ n = 1 , 2 , ...) are some real numbers.
W rite
ЯР = 1 (* — 0 , 1 , 2 , ...),
z u r } = Я Р -Я [% (к = 0 , 1, 2, ..., л - 1 ) , л я р = яр,
^ 2я р = М и) - M i l (* = о , 1 , . . . , и - 1 ) , ^1ЧЯ) = 4 4 Я 1 *яр = А Ч ^ - А Ч ^ и (к = 0 , 1 , . . . , ^ — 1 ), 21 «яр = 4 И).
We start with the following
T
heorem1. Let f e D*. Suppose that for all n П
(4)
and fC— “ 0
(6) (» + 1 ) 2 '( * + 1)M*4*,I
fc=l
Then
(6) |17„ ( « ; / ) • iH3
W !/
—/(#)| < --- > w ( /, a?; —
^ Л w + 1 Z
jV
м’ к k= 0 x P ro o f. Retain the notation (2) and set
(n = 0, 1 ,2 , .. .) .
ФкУ) ^ {(sin\ ( k + 1 ) «/sin\ t Y ).
Applying twice the Abel transformation, we get
ТС П
#»(®i Я - / 0 » ) = ^;(-z>*) / ^ л {к )со$ЩМ
&=i l
2
tc
Tf n
0 *=o ' л '
By partial integration ([3], p. 244) 1 n
P„(®! / ) - / ( * ) = — A4 % \m n \(k + l)TzY-
k=0
1 Tt n
- j
^ f F x(t) £ А Ч Р ъ (t)dt.
* О * = 1
I t is easy to see th a t
f e W K S T c f i + l ^ r 1 if «e(0,7ü/(» + l)> ,
and
n 3 » n
y 1 и ! 4я)1 + ~ j ^ ( f c + l)/J 4 » > sin(fc + l ) f
f t = l fc= 1 A := l
if ïe<7r/(w + l ) ,
тс). These inequalities and (3) lead to '
79, IT 99.
1 2
tc< w
|J»(")I ] ? \ A 2W \ + ~ ^ j |.F«(<)l|J^24“V*№
Л;=0 О &=1
W 7Г n
( / , * ; / *«>(/,®j « ь Ц > + 2Я(Л>*(г)
fc=o о fc=l !
n n n/(w+l)
<w(/,®î 7t)2,^,M !AP| + 2^l(ü + l)2M 24”)| J »(/,»; tb<ff +
A=0 &= 1 0
n 7Г
+ 7Г2^ И2я(^| J t~2w{ f , X ‘, t)Ddt +
к — 1 тт/(?г4-1)
л n
+ ^ J ас, Од I ^ (fc + 1) 42Aj^sin(fc + l)t
Tt/(M+1)
= I 4- Y + Z + 1 7 . n I n 1 n ^ " n
In view of (4),
г < М г тсЬ < т г т т ^ +
М 1\ 1 \ / />®; - r r r ’ ^ + 1 /п
ТС\
- У ( й + 1)2И*Л?>|«;(/,®;
ю + 1 \ n + l JD
п + 1 27Г
ft=l
I тс \ 2тсМг
v^iП
/тс
\<2тг JfjW / , я; — —- ^ У.ю />®; т г г ,
\ w + 1 /Х> ^ + 1 \ D
п n+l I \ тс М
П+1 / \Zn = 71 ^ *=i И24и)! J wl/,a?; 1 ' y ) 'D г ' тИ '
Т с М г V I / тг \
< —--- > w l f . æ : ---.
п + 1 Z
jГ А; + 1 I d
Applying the Abel transformation to the sum in Wn we obtain
n _ n
(lc + l ) A
4Psm(lc + l)t = (w + l ) J 2A<?> J T s i n ( i + l ) i +
*=i
+ ^ { ( к +
1) ЛЧ ^ - ЛЧ ^ } ^ sin (г+ 1) if
*=1 г=1
37Г
T
71 — 1
k = \ k = 1
for #€<7т/(»г + 1), 7т). Therefore, by (4) and (5), Зт с Ц М .+ М ,)
w m<
2(w + l ) / t w ( f , a?; t)Ddt
n/(n+l)
3tc( M 1 - \ - M 2
2(w + l ) 7 K(f’e>rl Зтс(М1-\-312)
v n / тсdt ^ 27 > w / , a?; -— - 2(^ + 1) ^ V fc + l / p Collecting the results we get the desired estimate (6) with Жа
= Н (2 + 9тс) Ж^ + З
тгЖ,}.
Consider now the class D(Q). I t follows immediately from (6) that,, under the assumptions of Theorem 1,
sup I Un(ar, / ) —/(я?)I <
felJ(Q)
П n
-\-1k = о
The last inequality, for some operators Un(æ;f), cannot be improved.
This is a consequence of our next result.
T
heorem2. Suppose that zJ2A ^ > 0 {k = 0 , 1 and that П
1 У (Й + 1)М 24 ”>> Mt > 0 ( « = 0 , 1 , 2 , . . . ) . n -f 1'
k= 0
Then for all 2 we have sup I UJæ-, f ) - f ( œ )I >
feD(O) П + - 1 k= 2
Ё“Ш
' '( Ms< м г).
P ro o f. Let g(t) = Q(\t — x j) if \t — æ\ < iz and g(t +
2iz) = g(t). Then g
e I)*. Moreover, for
0< t < h < тс we have
« t
(D*) J [д{со + и) — g (æ)Jdu = J Q(\u\)du ^
2t û ( h ) .
Consequently, w{g,ec\ h)D^ Q(h) and geD(Q). Next, sup \Un(œ; f ) - f ( œ ) \ > \Un(æ’, g ) - g ( æ )| = | ü n(ar, g)\
VD(Ü)
where
= i f a W ÿ A n t f ^ + ^ î m
tz J \ sm*t
0 k = 0 x 1 '
ü(t)t 2(sin|(fc+ l)/)2d/
k = 0 n/(n + 1)
g+hn
n n+1
f йт*~г(»™№+Щ 2 <я)
k = 0 г=1 in
n+l
n n
4 т г ^ ( и + l ) 2 J T + 4 “ ’ | У f i P ” ( * + i r 2 ^ q ,
k=0 i = l ' ' '
(г+1)я
n + l
A U = f (sin|(/i: + l ) t f
(г + 1)(/с+1)тг
2(n+l)
dt k-\-l l(fc+l)TI / sin
21dt
~2(n + l)
(*+Ря
4(n + l)
k +
1sin 2tdt > tz ( A: -(-1 )2
12 (n + 1)3
Since for all n > 2
И + 1
_ / г7С \ „ 7C /* 1 л ,
£ --- (г + 1 Г 2> ---- --- — Q(t)dt
U + 1 / 6(n + l) , J 1W
t2
(see [1], p. 28), we find th at
d-i)rc
n + 1
snp \Un(w, f) - f ( c c ) \ > 2
/£
г»(
й) 15тг(п + 1)2
157
t:2(^ + 1)
^ ( f c + l ) 2/ !2^ J
± û ( t ) d t
fc=0 n/(n+l)
w M+l . .
-^(fc + l)2zl24“> J fl(y)<B
-Мд У
й7
157т2(тг + 1) ^ \г-| -1/
.-.and this completes the proof.
R e m a rk . If a function / is of class L, then ЗЖ, \~1 / П 7C \
Я - / И К —T T Л ад( ^ а?;1ГГТ' (я- = 0 ,1 , ...),
™ + l \ 7 c +
1I l
provided th a t assumption (4) holds. In this case Theorem 2 remains valid for the class L{Q) consisting of all functions f e L such th a t
w (/, ®ï h)L . _ sup ---—--- < 1.
0<Л<тт iQ(^)
3. Approximation by Abel means. As well known, the Abel means -of Fourier series of fe D* can be w ritten in the integral form
7T
-Р Д * ;/) = — (-»*) f f M j z
1
—Г
22
rcos(t — æ)+r - dt
2(0 < r < 1).
We first give an estimate for the difference P r (o>; / ) —/(&) at the Lebesgue-Denjoy points æ defined by (1).
T
heorem3. I f fc D* and n = [1/(1 — r)] — 1, then
n l \
((7) \Pr{œ; f ) - f { x ) \ ^ M
6{ l - r ) ^ w ^ f , æ i (0 < r < 1).
The inequality cannot be improved.
P ro o f. Using the symbols (2) and integrating by parts we obtain
1 - r 2 r . 1
РЛ *; /)-/< » > = ( * ) f f M j z 2r cos i +r , о
Tzl(n+1) n
r ( l —r2) j Г Г \
+ -( I + f ) * ’ m W = dt
1
— r 27
c(1 + r)
1
—r
sin t
nl(n+l)
((1 — r f + 4rsin2|tf)2 dt 27
t(1 + r)
In view of (3)
1
— r
F X(K) + A(r) + B(r).
2
n{ l +r )
( TZ \
(1 - r ) w ( f , X ‘, 7C)jr><(l- r ) nnd
W l+ r) r „
u ( r ) K 7T (1 —Г)3 J ,
n/(n+l)j i p » ( * ) i td*
2 r(l + r)
7Г ( 1 —
r)S
n/(n+l)
4
tc2
<
3(1 —r) (w + l) 64тг2
wi f , a?; 47Г2
n + l J D 3(1 —г) (тг + 1)
J* t
2w( f , a?;
Ê ” ( f t æ ;
i t ï )
*=o n
a - o J T ' - j / , . ; ^
whenever 0 < r < 1. Moreover, B(r) = 0 if 0 < r < % and
2 r(l — r2) r . . tz
^ :zr(i — r*) r
\ B ( r ) \ < —-—- J «>(/, 7
<
7T/(W + 1) (1 —Г2)тС3 ,
((1 — r)2 + 4rsin2^ ) 2
8r / t w( f , a?; i^dtf rc/(n+l)
(1 — Г2) TC2
8
r
П
+ 171
J d £ < - ^ 2( l - r ) ^ w | / , œ,
1 ^ k
=0 Dprovided th a t % < r < 1. Collecting the results we get the estimate (7) with M
6= (6 + 131712 )/ 6 .
In order to show th a t inequality (7) cannot be essentially improved we consider the class D( ü) . By (7),
sup IPr (®; f ) - f { x )I + M 6(l — r)
f*D { Q)
*=0 '
1
/On the other hand, using the function g defined in Section 2, we get s u p \Pr(x) f ) - f { x ) \ > IP r(x-, g)-g(®)l
feD(O)
1 — r 2 /■ 1
= --- O(t )---
7Z
J 1 —2rcccos t + r
2dt
1— r — гй r
— m
TV J
0 < n/(n+l)
Since (1 — r) < l/(n-\-1) < t, we have
(1 — r)2-f rt2
n + 1
dt,
sup
UD(Q)
\Рт(щ } ) - № \ > f r *Q( t ) M = A ( l - r ) j q ( j ) dt 7T/(n+l)
> --- (1 —r) У q [—^ — and this completes the proof.
B ern a rk . In the case of f e L we can get the following estimate
I-Pr ( ® ; / } - / ( » ) I < 1 2 ( l - r )
X )when 0 < r < 1 and n = [1/(1 —r)] —1.
Following DzvarseiSvili [2], let us introduce in the space D* the norm
2 T Ï
\\ f\\D = sup J (D*) Г («)/(»)<&» I,
<PeV 1 o'" 1
where V denotes the class of all 27r-periodic functions of bounded variation in <0,
2n) and such th a t
sup \<p{x) \к 1 and var 9 ? ( # ) < 1 . 0<а:<2тг
Denote by Ev ( f ) D the constants of the best approximation of fe D * defined by the formula
Д .(/)л = in fj|/(-)-^ (* )lln (v =
0, 1 , 2 , ...),
where the infimum is extended over all trigonometric polynomials tv(x) of degree not exceeding v.
Arguing as in [4], we easily obtain
T
heorem4. Let feJD * and let n = [1 /( 1 — r)] — 1. Then П
\\p, ( •, / ) - / ( • )llz> « 94 ( 1 - »■) £ E, (J)D (0-< r < 1).
v = 0
References