ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO Séria I: PRACE MATEMATYCZNE XXVII (1987)
He l e n a Mu sie l a k (Poznan)
1. In [3], there were given converse type inequalities for averaged moduli of smoothness тк and best one-sided approximation £ „ (/) of the function / by means of trigonometric polynomials in generalized Orlicz spaces E2n generated by a convex «^-function </> depending on a parameter, satisfying some condition ( B J .. Analogous results in case of Orlicz spaces generated by a concave, s-convex ^-function q> independent of the parameter were obtained in [4]. Here we shall give direct type inequalities in both cases, thus generalizing some results of [5], Chapter 8, and of [6]. These results together with those of [3] and [4] enable us to conclude some saturation theorems for Lipschitz classes defined by averaged moduli of smoothness in H2n.
Let us recall that given an s-convex (0 < s ^ 1) ф-function depending on a parameter r, q>(t, u), 27t-periodic with respect to t, the modular
where / is measurable and 2^-periodic, defines the generalized Orlicz space E?2n = {/: gv (Af) < oo for some A > 0} with an s-homogeneous F-norm
Jackson type inequalities for averaged moduli of smoothness
M / ) = .f <р(*> l/(0 lM c 0
Il/||j = i n f '« > 0 : e „ ( i r I/!/ K 1!) ■ If s = 1, we shall write Ц/Ц^ in place of ||/|Ц .
Moreover, for any 2n-periodic f one defines
cok {/, x; Ô) = sup {\Akhf (f)|: t, t + k h e Ik0(x) \ ,
where I^(x) = (x — kô/2, xA-kô/2) for Ô > 0, к = 1, 2, ..., and arbitrary real x. If / i s measurable and bounded, then cok(f, •; <5) is also measurable ([5], p.
28). For such functions / the averaged modulus of smoothness xk of order к
in Iî2n is defined by
т * (/;< 5 )= .|К (/, -; а д , <5>о
(see also [4]). The usual modulus of smoothness œk of order к in lf2lt is defined by
ojk(f; Ô) = sup \\\Akhf\%\ 0 < h ^ Ô], S > 0. '
Let Я„ be the set of all trigonometric polynomials of degree ^ n, and let for a measurable, bounded, 2^-periodic function f H + ( /) be the class of all PeH„ satisfying the inequality f (x) ^ P(x) in <0, 2л:), and Hn ( /) be the class of all Q e H n satisfying the inequality f (x) ^ <2(x) in <0, 2л>. Then we denote, as usually,
En( f) = inf III/- Щ : T e H n),
£„(/) = in f 111 P-QWV- Р е Я „+ ( / ) ,0б Я ; ( Л ) .
It is easily seen that for any / е Я 2л and arbitrary n there exists a T *eH„
such that En( f ) = \\f— T*||^. Also, for any bounded and arbitrary n, there exist P * e H + ( /) and Q *eH ~ ( /) such that £ „ ( / ) = \\P* — Q*||* (see [3], Proposition 4 and [4], 2.1).
2. In this section we shall investigate the case of convex <p. We shall assume that (p is an iV-function depending on a parameter (see [1]). Let us recall that the function q> satisfies the condition (В rj > 0, if there exist a set A c <0, 2л) of measure zero, a number c > 0 and a family of non-negative, measurable, 2л-
2 n
periodic functions / ( • , h), satisfying the inequality Sv = sup ( f (t, h)dt < oo IM <0 ô
such that for every и ^ 0, t e <0, 2 л )\А and \h\ ^ ц there holds the inequality (p(t — h, и) ^ <p(t, cu)+ f(t, h) (see [3]). It is easily seen that if q> satisfies (Bn), then it satisfies (ВJ for all ц ^ л with the same constant c > 0 and = Sn. If (p satisfies (BJ, then for / е Я 2л there holds the inequality ||/(• + /î)ll<p
^ 2c max (1, S J H /I^ for any h > 0 (see [2], Theorem 1).
Let us still remark that if К : (0, 2л) x (0, 2л) -> R is measurable in (0, 2л) x(0, 2л), K(-, t)eH 2n for almost every t e ( 0, 2л) and ||K(-, t ) ^ is integrable in (0, 2л), then the following Minkowski inequality holds:
(U II j0 (f)|K (-, 01*11» « 2 J 9 (r)||K(-, Oil»*
0 Ô
for every g ^ 0 integrable in (0, 2л).
Th e o r e m 1. Let tp be an N -function depending on a parameter, satisfying
(B„) with constants c and Sn. Let f be 2n-periodic, absolutely continuous and such that the derivative f'e L ? 2n. Then there holds the inequality
£ „ ( /) ^ - 7 г с т а х ( 1 , Sn)E „ (f).
( 2 ) n+ 1
In case of tp independent of the parameter, we have
(3) n+ 1
P ro o f. Let q e H n. x be such that E „ _ i(/') = \\ f '- q \ \ v . By [5], p. 243- 247, there exist Tl n, t 1>n, R l , R 2eH „ _l such that
2n
i f 2
> t liH(t) in <0, 2tu>, - = —
71 J n
0 and
Я 2 (*) < / ( x ) < /?! (x) in <0, 2tu>, 2 n
Rt (x) - R 2(x) = ~ J (TUn( x - t ) - t Un( x - t ) ) \ f ( t ) - q ( t ) \ d t 0
2n
- I \Ti,n(t) — t l n(t)\ \ f ' ( x + t) — q(x + t)\ dt.
Hence, by inequality (1) applied to g(t) = \T1<n{ t ) - t Un{t)\ and K{x, t)
\f '( x + t) — q(x + t)\, we obtain 2n
W R i - R i W ^ l^i,n(0 —n,n(0l \\f'{' + t ) ~ q { ' + t ) \ \ < p dt.
But, by condition (Bn) we have
IIpO + OII* < 2c m ax(1, SJWpW^ for р е П 2к.
Hence, taking p(u) — f'(u ) — q(u), we get 2n
\\R1 - R 2\ \ * < 2 ^ I |Г1.|1( 0 - П .и(0 |2 ст а х (1 ,5 11) ||/ ' - 9 || , Л
= — cm ax (l, Sn) £ „ _ !(/'). 871 n
This implies inequality (2). If (p does not depend on the parameter, then
\\p(' + t)\\v = IIpOILpand we obtain (3).
Co r o l l a r y 1. Let tp be an N -function depending on a parameter,
satisfying (Bn) with constants c and Sn. Let a In-periodic function f posses absolutely continuous derivative f (k~ X) o f order к — 1, k ^ l , such that f (k)e L 2n.
10 — Prace Matematyczne 27.1
Then
where С = 8л;стах(1, Sn) for (p depending on the parameter, and C = 4n in case of (p independent of the parameter.
P ro o f. If P, Q e H n, Q(x) ^ f{ x ) ^ P(x) for all x, then \\f-Q \\v < \\P - 6IIV- Hence
En{f) = inf II/- T\\v ^ inf W f - Q W ^ inf ||P -G II, = Ё Л Л -
T e H n Q e H n ( / ) Q ^ f ^ P
Thus, applying Theorem 1, we get
Ёп( Л ^ С п - ' Ё „ ( Л , and so by easy induction,
Ë „ ( f ) ^ C k n - kËn( f (k))•
But Ë „ (fik)) ^ \\ f {k)\\v , which gives the corollary.
Th e o r e m 2. Let cp be an N -function depending on a parameter, satisfying
condition (Bn) and such that 1 еТ?2л- Let f be a In-periodic, measurable and bounded function. Then
E„(f) < Cxx {f; 1/w) for « = 1 , 2 , . . . ,
where C = 66(4rc + l)c2 m ax(l, S2). In case of (p independent of the parameter, one may take C — 228.
P ro o f. By [5], p. 249, formulae (8.11H8-13), there exist step functions S„ and Tn such that
/ „ w < / ( * ) < s n{x), 0 ^ Sn{ x ) - I n(x) < col if, x; 3n/n) in <0, 2tt) and
\S'n(x)\ < 2nn~ 1 (Oi {f, x; 4n/n), |/^(x)| < 2nn~ 1 w 1 (/, x; 4n/n)
for all хе(0,2л) which are not points of discontinuity of functions S„ and /„.
Hence
l|S „-/„||v ^ I K ( / , •; 3702)11^ = t i ( / ; 3n/n), USX < 2nn~1xl (/; 4n/n), \\T„\\ ^ 2 w T1 xx (/; 4л/п).
Since satisfy the assumptions of Theorem 1, we have
87Г 8 л
En(Sn) < 77^-7 c max (1, S„)£„(S;) ^ ^7-7с т а х (1, 5Я)||5;||^
/2+ 1 /2 + 1
^ 16cm ax(l, iSJi! (/; 4n/n).
Similarly,
Ë„(I„) ^ 16cm ax( 1, 4n/n).
Now, we shall prove the inequality
(4) En ( /) ^ Ên(Sn) + ||S„ - I X + En(/„).
Let h be a bounded, 27r-periodic, measurable function. Let P(h)eH// (h), Q{h)eH~ (h) be such that
Ê„(h) = \\P(h)-Q(h)\\v . Then
P (f)(x) < P(S„)(x), Q (f)(x) > Q(I„)(x) and
(2(S„)(x) < f ( x ) ^ I n(x) ^ P(/„)(x) for all x.
Hence .
P(f)(x)-QU)(x) ^ P(Sn)(x)-Q(Inm
^ (P (Sn) (x) - Q (Sn) (x)) + (Sn (x) - I n (X)) + (P(In) (X) - Q (/„) (x)).
Thus,
£ * (/) ^ W P i S J - Q i s ^ + WSn-inh + W P i U - Q i i n X
= Ên(Sn) + \\Sn- l X + Èn(In)-
Applying (4) and the previous inequalities, we obtain E „ (f) < (32cm ax(l, S J + l ) i i (f; 4n/n).
But we have
T it/; 4n/n) ^ 2c m ax(1, Sn) (4л + 1)xt (/; 1/n) . (see [3], Theorem 1 (4)). Hence
E „ (f) ^ C M /; 1/n) with the desired constant C.
If q> is independent of the parameter, then
Ën(Sn) ^ 8t! (/; 4n/n), Ё„(1п) ^ 8тх (/; 4тг/п), and so
En( f ) < 16ц (/; 4п/п) + т1 (/; 3n/n) ^ 228^ (/; 1/n).
Let us still remark that Theorem 2 is, up to the constant, a
generalization of [5], Theorem 8.2. Now, applying the method from [5], p.
46-48, we shall prove the following
Le m m a 1. Let <p be an N-function depending on the parameter\nd let
f e И2п, where к is a positive integer, 0 < h ^ 2n/к. Let
fk,h(x) + . . .
hГ ( - h ) ~ k
0J
I —f ( x + h + ■■■ +*к) + ( ^ / •••
о
- + ( - i)kC i ) 4 t \ + . . -+ tk
dt j ... dtk
(cf. [5], p. 47, 2.8). Then
(i) If ( x ) - f kth(x)\ ^ ojk(f, x; 2h), (ü) \\f-fk,h\\<p ^ 2mk(/; h),
(iii) fk,h has an absolutely continuous derivative fk kh~ 1] and / $ еТ?2п with ll/a il, ^ 2(2k)k h-'cor(f; h) for r = 1 ,2 , ..., k.
P ro o f, (i) is shown in [5], p. 47. In order to prove (ii), we first remark that applying the Minkowski inequality (1) with interval (0, 2n) replaced by (0, 2n)k c= R k, we obtain
и/-/* j „ « p - ‘ Î- ■ ■ î и?,,+...+,t„i/wi
dt,... dh\iо 0
« 2 r ‘ J . . J | l 4 1 + ...+Illrt/ ( - ) M r , .... dtk s: 2mk(f\ h).
0 0
In order to show (iii) it is sufficient to apply formula [5], 2.9, p. 47, Minkowski inequality (1) with <0, 2n}k~r c= R k ~ r and estimation as in [5], p.
47 and 48, obtaining
h h
ll/S l ( - /) ) -k d h f ( X + t 1 + • • • + f i c - r )
+ 1 / \/c— 1 ^[k-Dhjkf ( * +к - 1
(t 1 + • • • + f i c - r ) +
•' + ( - 1^ ( k k_ l) kr Ah/kf(KX + tl + ' rj j ^ l ■■■dtk-
h h
<2 A -‘ j . . . j | p ; / ( x + t1+ . . . + f lt. r)||.
0 0
... +
^ 2h~r
^ f [ x + - ~ ' - ' k + t k~ r-
+ ...
àti ... Л*-,
« г ( / ; /l) + ( j ) ( ^ z j ) ® r f / ; - j ~ h ) + •
• + (fc*
^ 2 (2kf h~ 2 cor(f\ h).
Lemma 2. Let tp be an N -function depending on a parameter, satisfying condition (B K) and such that \ e E 2n. Let f be a 2n-periodic, measurable and bounded function and let gn(x) = cok(f, x; m/n), where m ^ 1. Then
E„{gn) ^ Czk(/; 2m/n) for n = 1 , 2 , . . . , where C is given in Theorem 2.
P ro o f. Let g(x) = cok( f x; h), h > 0. Then, estimating as in [5], p. 251, we obtain
Hence
a*! (g, a; <5) ^ cok(f, x; h + ô/k) for Ô > 0.
Ti (»*(/, •; h )\3) ^ zk(f; h + ô/k) ^ zk(f; h + 0).
Taking here h = 3 = m/n we obtain
*1 (0„; l/и) ^ Tj (бг„; m/n) ^ zk( f ; 2m/n).
Thus, applying Theorem 2, we get
En(gn) < C ii (#„; 1/w) ^ Czk{f; 2m/n).
Lemma 3. I f tp is a convex (p-function and f g, oteL?2n, |/(x ) —g(x)| ^ ot(x) for all x , then
£ „ ( /) ^ Ën(g) + 2Ën(x) + 2\\ot\\<p.
P ro o f. Keeping the notation from the proof of Theorem 2, we obtain
£ „ ( / ) < £„(gf)4-2||P(oc)||(?), and the desired result follows form the inequality 0 < P (a )(x )-a (x ) ^ P (a)(x )-()(a)(x ) for all x.
Theorem 3. Let q> be an N -function depending on a parameter, satisfying condition (Bn) and such that \ е П 2п. Then for every positive integer к there exists a positive constant c(k) such that for every 2n-periodic, measurable and
Ë»(f) ^ c{k)zk{ f ; 4/n).
P ro o f. We take functions / M from Lemma 1 with h = i/n. Applying Lemma 3 and property (i) from Lemma 1, we obtain
£ „ (/) « Ën(fl n ^ ) + 2Ë„{o)k(f, -, 2 n - ‘)) + 2 |H ( / , ■; 2n~l)\\v . By Lemma 2 with m = 2,
È„(o)k{ f •; 2n ~ 1) ) ^ Cxk(f; 4/n), whence
(5) Ën( f) ^ Èn(fktH- i ) + 2C4 {f; 4/n) + 2xk(f; 2/n).
bounded function f there holds the inequality
Applying Corollary 1 and Lemma 1, (iii), we obtain
£ „ (/M - i) ^ Ck n~k \\fkin-i\\q> < 2(2Ck)k cofc(/; \/n) ^ 2 (2Ck)k xk (/; 2/n).
Substituting this inequality into (5), we obtain the desired result with c(k)
= 2{(2Ck)k + C + \).
Theorem 3 together with [3], Theorem 3, give easily the following result, generalizing [5], Corollary 8.3, p. 257:
Theorem 4. Let <p be an N -function depending on a parameter, satisfying condition (Bn) and such that 1 e Tf2 and let f be a 2n-periodic, measurable and bounded function. Then
(a) if 0 < a < k, then the condition xk(f; <5) = 0(<5a) as 3 —> 0 is equivalent to the condition En( f) = 0(«~a) as n-* oo,
(b) if a = k, then xk(f; (5) = 0 (3 k) implies Ên( f) = 0 (n ~ k) as n —> oo, and E„{f) = 0 ( n ~ k) as n —>oо implies xk(f; <5) = 0 (3 k |log<5|) as <5-> 0,
(c) if a > k, then xk(f; Ô) = 0 (3 a) as Ô -* 0 implies Ê„(f) = 0 ( n ~a) as n
—>oo, and Ën( f ) = 0 (n ~ a) as n —> oo implies xk(f; 3) = 0{ôk) as 3 —> 0.
3. In this section we shall consider the case of s-convex (^-functions, 0
< s ^ 1, limiting ourselves to (p independent of the parameter. Let /„ be a Ire- periodic step function with discontinuities at Xj = nj/n, j = 0, ± 1 , ± 2 , ..., i.e.,
2 n - 1
Ц х ) = у0+ X ( y j - y j- 1
j= 1 0 X j ( x ) for x e <0, 2re>,
where 9y is 2re-periodic and
0y(x) = j j for 0 ^ x < y, for у ^ x < 2re;
obviously, ln(x) = y j for X E ( x j , x j+ l), j = 0, 1, ..., 2 n — 1. The method
applied here follows the same lines as in [6] (case (p{u) = |u|s). We shall define for ÿ = (y0, yi, y2n- i) the pseudomodular
e l<pn]( y ) = Z < p ( \ y j - y j - i \ )
j= 1
and the respective s-homogeneous F-pseudonorm ilÿllS,"1 = inf {м > 0: $ }{7/ulls) < !}•
We shall also assume that cp satisfies the condition (A2) for all и > 0, and apply the function ф(и) = sup[(p(uv)/(p(v)f Let us remark that from s-
v > О
convexity of the ^-function tp follows strict monotonicity of cp and, supposing (A2), also s-convexity and strict monotonicity of ф. Let cp_x and ф ^ г be the inverse functions to (p and ф, respectively.
Lemma 4. Let cp be a concave, s-convex tp-function without parameter, 0
< s ^ 1, satisfying the condition (A2). Then
M n ) ^
CAs)
Ф~Лс2А) n)ll)C[n]
where C1 (s) and C2(s) are positive constants independent of n and y.
P ro o f. The idea of the proof is the same as in [6]. We write
K m(v) 1 /sin j n v A r
am \ sin it; ) where a„ sin^nw 42r
sin^w dw
for m —{n — \)r + \, n, r are positive integers; then K m is a non-negative trigonometric polynomial of degree < m. Let
u mu m i я
Я
f ( z ) K m{ x - z ) d z ,
ъ
— Я
x eR ;
then Um'[ f ] e H m_ l (see [6]). Let us write
/s in in (x —у + я /2 п )\2(г_1) /sin-|rc(x — у — я/2и)\2(г-1) y,m \u s in i( x — у + Tt/2n)J \n s in i( x — у — n/2n)J
^ п ^ п ( х + я/2п )\2(г_1) /sin^n(x —я/2п) \ 2(r_1)
\n s in i( x + n/2n)J \rcsini(x — n/2n) J U i [0,](x) = Uml9A(x) + ASy>m(x),
U~ [0 J(x ) = Um [9y] (x) — A S 2n_ ym (2k — x).
R. Taberski [6] proved that t/ + [0У] е Я + _ , (ву) and U~ [ 0 J e x (0y) for
sufficiently large A = A(r), depending on r. Hence
Ëm- i ( 9 y) < III/; W - U - [0 J||„ = M (S ,>mO + S2, - , . m(2*--))||,.
We choose
Then
3e(*) =
sin^n(x + a) \ 2(r X) n s in |(x + a) J
f , l/s in in x \ 2(r_1)l , „ f , |/s in i n x \ 2(r" 1)| ,
^ (ej- 2J н и э д r x+2J*{Un3ï) r x-
0 K/n
Hence, taking r > 1 + l/2s, we have for an arbitrary и > 0:
4 5 0 ^21* ( i ) dx+2 M 4 ) '
' M ix ^ ~ ф П0 n/n
n \U 1 Is
with a constant Cv > 0. Hence Pall* <
1 where C' = 1/CS
Г - l ( C > )
Hence the best one-sided approximation of 0y in 1%n, Em- 1 (0у)ф ^ Н З Д ^ {j/ s^ l (Cs n)
with some fy eR , for m = (n— l ) r + 1, for sufficiently large integer r. Now, let us choose an arbitrary n and taking r fixed as above, let us choose an integer q for which qr ^ n < (q + l)r. Then
Еп(@у)ф ^ Eqr(0у)ф ^ SAS 8 A s
for 0 < y e 271,
^ - . ( c ^ + i ) ) Г - Л с »
where C" = C’Jr. There exist trigonometric polynomials Pyn e H„ (ву) and Qy,ne H ~ (9 y) such that
E H( 6 y \ , = \ \ Р у , п ~ й у , п \ \ ф -
Hence, since 0 < Ру>п — ву ^ Py,„ — Qy,„, we have
(6) Similarly, (7)
\\Py,n- e y\ \ i ^ \ \ p y,n- Q yj \ i < SAS Г - Л с ; п )
\\0y-QyJ\i <
8 A s Г - Л с ; п )
Writing (as in [6])
V
= i { \ y j - y j - i \ + y j - y j - i }and
V
= i { \ y j - y j - i \ - y j + y j - i } ,we may express l„(x) in the form
2n—1 2n—1
/„(*) = У0+ Z Z 4Г M X) for XGK.
J= 1 J=1
We take also (see [6]) trigonometric polynomials
2n— 1 2 n - l
Р„М = Уо +
I
Л + ^ , л М - S *j- Q ,j,.(x), P„eH* (/„),j =
1
j =1
2n-l 2/1-1
ô,,(x) = yo + Z V 6 x ,,n W - Z p Xj.n(x), QHe H „ ( 0 -
j =
1 1=1
Applying the fact that cp is subadditive and the definition of ф, we obtain for arbitrary rjn > 0 and и > 0:
2 n
P „ ~ l 2/1— 1
I J
i n ( P x l . A x ) ~ e X j ( x ) ) ^ d x(
0
2n
"h Z ^ Qxj,n (-^))^ dx
0
< Z ч > ( ^ У * Ы Р хрп - о х) )
2/i— 1
• Z
ф|^l/s
j Q ^ i ^ n W x j Q x j , n))•
By inequalities (6) and (7) we have
\\>ln(PXj, n - 0 Xjm < 1 and -\\rin(Ox r QXjJ \ i ^ 1 for
(8) Hence
rjn = Ф- Л С » 81;M
^ fan (*%•.« - 0X) ) ^ 1 and Q+ fa„ (0Xj - QXjJ ) < 1
Thus, for цп given by (8) we have
where y = (y0, yu ..., y 2 n - i ) - Similarly,
for all и > 0 and tjn given by (8). By s-convexity of <p, we have
for every и > 0. Hence
R e m a rk 1. If (p(u) = \u\p, 0 < p ^ 1, then cp_ x (и) = t (w) = u1/p and taking s = p, we obtain ij/L l (C2(s)n) = C2(p)n, and Lemma 4 yields the lemma from [6] given in B2n, 0 < p < 1.
Theorem 5. Let (p be a concave, s-convex (p-function without parameter, 0
< s ^ 1, satisfying the condition (A2), and let f be a In-periodic, bounded and measurable function. Then there holds the inequality
Cj(s) and C2(s) being the constants from Lemma 4.
P ro o f. Taking as in [6] the 2rc-periodic step functions
g„(x) = inf {/(и): Xj ^ и ^ xj+ l}, Gn{x) = sup [/(и): Xj ^ u ^ x J+l) for x e ( Xj , xj+1), j = 0 , ± 1 , ..., we have gn{x) < / ( x ) ^ G „ ( x ) . By inequality (4), we have
E„ ( /) < E„ (G„)+ ||G. - gX + E„ Ы • By Lemma 4,
where
9n = ( в п ( хо ) , U n i xi), g n ( x 2 n - i ) ) , Gn = (G„(x0), •••, G J x ^ ! ) ) . But
sup {\f(u)-f(v)\: u, v e (X j, xj + l)}
< sup \ \ f ( u ) - f ( v ) \ : u, v e < x - x l5 x + x ^ ]
= w l { f , x - , 2 x l) for x e < x j , x j+1>.
Hence, we have for u > 0:
xj+1
0* ( J ' I ) <P y^T/s (/• ^ 2*l> = <?* ^ ^ ---
Since u > 0 is arbitrary, this implies
\\Gn- g n\L ^ Urn, (/, •; 2 х 1 ) | | ц) = x x(/; 2 X i ) .
Hence
£ „ (/) ^ M / ; 2xt)+ Ci(s)
Vs- 1 ( G 2 (s) nт Ж 1]+ ш л
But xj + l —Xj — n/n, and so
I < p { ^ s u p \ \ f { w ) - f ( v ) \ : D ,w e<xj . 1, x j+ 1>}
* j + i
n 2n_1 / 1 \
= 2^ I \ 4> ( y /ï ® i (/. xh 2*i) J dx Xj~ 1
n „
^ 271 !
2n— 1 *j+l/* / j
( TTT^^l ( / ’ 4 x l) )dx xj~ 1
^ JI V •; 4*i) ) < ^/ ( 4 / ^ 1 (/> ’i 4xi) )’\ W whence
1 \ , 1 _rnl / 0n
l/s for all и > 0. Hence
1
,1/s
[n]
< ll<^i if, •; 4*i)ll„ = Ji (/; 4 x j,
and so
Similarly,
Il9.lt”1 Ht, (/; 4x,) $ 2ht, (/; 2x,).
Il6.lt"1 « 2 n t,(/; 2x,).
Thus,
Èn( f К / 4C 1(s)n \ . _ , V + * U ( C 2(s)n
R e m a r k 2. In case of (p(u) = |u|s, 0 < s ^ 1, wè obtain 4C i(s)n _ C,(s)
Г - Л с2(*)п) Cz(s)’
whence Theorem 5 gives Theorem 1 of [6]. Let us still remark that in the general case, boundedness of
1 4Ci(s)n
Ф - i ( C 2 ( s ) n )
is equivalent to boundedness of -\j/(K n 1/s) with some К > 0, and this in turn n
is equivalent to the following condition: there exist positive constant К and C such that the inequality (p(Knl/su) ^ Crup(u) holds for every и > 0.
Theorem 5 together with [4], Theorem 2.2 give easily the following result:
Th e o r e m 6. Let cp be a concave, strongly s-convex (p-function without
parameter, satisfying the condition (d 2), 0 < s ^ 1, and let f be a 2n-periodic, measurable and bounded function. Then
(a) i/0 < a < s, then En( f) = 0 ( n ~ a) as n —*• oo implies i j ( /; <5) = 0 (S a) as Ô —> 0, and (/; <5) = 0 (ô a) as Ô -* 0 implies
£ „ (/) = o :+
Ф -1( c 2 (s) n) as n-+ оо ,
(b) if a = s, then En( f) = 0 (n s) as n-> oo implies (/; S) = 0((5s|log<5|) as Ô -* 0 and t i (/; <5) = 0(<5S) as Ô -* 0 implies
En{f) = o ( n ~ s + - П( г (л v) as n-* oo, V Ф- i (C2(s)n)J
(c) if a > s, then En( f ) = О (п~л) as n -> oo implies (/; <5) = 0(<$s) as Ô - 0 .
References
[1] H. H u d z ik , A. K a m in s k a , Equivalence o f the Orlicz and Luxemburg norms in generalized Orlicz spaces LM(T), Functiones et Approx. 9 (1980), 29-37.
[2] H. M u s ie la k , On some inequalities in spaces of integrable functions, Proc. Intern. Confer, on Constructive Theory of Functions, Sofia 1984, 629-633.
[3] —, On the x-modulus o f smoothness in generalized Orlicz spaces, Comment. Math. 25 (1985), 285-293.
[4] — , Converse type inequalities for averaged moduli o f smoothness in Orlicz spaces generated by concave functions, ibidem 26 (1986) (to appear).
[5] Bl. S e n d o v , V. A. P o p o v , Averaged moduli o f smoothness (in Bulgarian), Sofia 1983.
[6] R. T a b e r s k i, One-sided trigonometric approximation in metrics of the Fréchet spaces LP (0
< p < 1), Math. Nachr. 123 (1985), 36-46.