Chaos, Solitons and Fractals 42 (2009) 2080–2094
Contents lists available at ScienceDirect
Chaos, Solitons and Fractals journal homepage: www.elsevier.com/locate/chaos
Fractional Korovkin theory George A. Anastassiou Department of Mathematical Sciences, University of Memphis, Memphis, TN 38152, USA
a r t i c l e
i n f o
a b s t r a c t
Article history: Accepted 30 March 2009
In this article we study quantitatively with rates the weak convergence of a sequence of finite positive measures to the unit measure. Equivalently we study quantitatively the pointwise convergence of sequence of positive linear operators to the unit operator, all acting on continuous functions. From there we derive with rates the corresponding uniform convergence of the last. Our inequalities for all of the above in their right hand sides contain the moduli of continuity of the right and left Caputo fractional derivatives of the involved function. From our uniform Shisha–Mond type inequality we derive the first fractional Korovkin type theorem regarding the uniform convergence of positive linear operators to the unit. We give applications, especially to Bernstein polynomials for which we establish fractional quantitative results. In the background we establish several fractional calculus results useful to approximation theory and not only. Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction ln this paper among others we are motivated by the following results. Theorem 1 [10], p. 14. Let ½a; b be a closed interval in R and ðLn Þn2N be a sequence of positive linear operators mapping Cð½a; bÞ into itself. Suppose that ðLn f Þ converges uniformly to f for the three test functions f ¼ 1; x; x2 . Then ðLn f Þ converges uniformly to f on ½a; b for all functions f 2 Cð½a; bÞ. Let f 2 Cð½a; bÞ and 0 6 h 6 b a. The first modulus of continuity of f at h is given by
x1 ðf ; hÞ ¼ sup jf ðxÞ f ðyÞj: x;y2½a;b jxyj6h
If h > b a, then we define x1 ðf ; hÞ ¼ x1 ðf ; b aÞ. Another motivation is the following. Theorem 2 [15]. Let ½a; b R a closed interval. Let fLn gn2N be a sequence of positive linear operators acting on Cð½a; bÞ into itself. For n ¼ 1; . . ., suppose Ln (1) is bounded. Let f 2 Cð½a; bÞ. Then for n ¼ 1; 2; . . ., we have
kLn f f k1 6 k f k1 kLn 1 1k1 þ kLn 1 þ 1k1 x1 f ; ln ;
where
1
ln ¼ Ln ðt xÞ2 ðxÞ
2
1
and k k1 stands for the sup-norm over ½a; b. E-mail address:
[email protected] 0960-0779/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.chaos.2009.03.183
ð1Þ
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
2081
One can easily see, for n ¼ 1; 2; . . .
l2n 6 Ln t2 ; x x2 1 þ 2ckLn ðt; xÞ xk1 þ c2 kLn ð1; xÞ 1k1 ; where c ¼ maxðj a j; j b jÞ. Thus, given the Korovkin assumptions (see Theorem 1) as n ! 1 we get ln ! 0, and by (1) that kLn f f k1 ! 0 for any f 2 Cð½a; bÞ. That is one derives the Korovkin conclusion in a quantitative way and with rates of convergence. One more motivation follows. Theorem 3 (see Corollary 7.2.2, p. 219 [1]). Consider the positive linear operator
L : C n ð½a; bÞ ! Cð½a; bÞ; n 2 N: Let
ck ðxÞ ¼ Lððt xÞk ; xÞ; k ¼ 0; 1; . . . ; n;
1 ba dn ðxÞ ¼ Lðjt xjn ; xÞ n ; cðxÞ ¼ max ðx a; b xÞ cðxÞ P : 2 Let f 2 C n ð½a; bÞ such that x1 ðf ðnÞ ; hÞ 6 w, where w; h are fixed positive numbers, 0 < h < b a. Then we have the following upper bound:
jLðf ; xÞ f ðxÞj 6 jf ðxÞjjc0 ðxÞ 1j þ
n ðkÞ X f ðxÞ jck ðxÞj þ Rn : k! k¼1
ð2Þ
Here
Rn ¼ w/n ðcðxÞÞ
n
dn ðxÞ w h n ¼ hn d ðxÞ; cðxÞ n! cðxÞ n
where hn
h ¼ n!/n ðuÞ=un ; u
with
/n ðxÞ ¼
Z 0
jxj
t ðjxj t Þn1 dt h ðn 1Þ!
ðx 2 RÞ:
de is the ceiling of the number. Inequality (2) is sharp. It is approximately attained by w/n ððt xÞþ Þ and a measure lx supported by fx; bg when x a 6 b x, also approximately attained by w/n ððx tÞþ Þ and a measure lx supported by fx; ag when x a P b x : in each n n n ðxÞ n ðxÞ and dcðxÞ , respectively. case with masses c0 ðxÞ dcðxÞ Using the last method and its refinements one derives nice and simple results for specific operators. For example from Corollary 7.3.4, p. 230 [1], we obtain: let f 2 C 1 ð½0; 1Þ and consider the Bernstein polynomials
ðBn f ÞðtÞ ¼
n X n k k t ð1 tÞnk ; f n k k¼0 l
t 2 ½0; 1;
1 ffiffi pffiffi x1 ðf 0 ; p then kBn f f k1 6 0:78125 Þ: So Bn f ! f as n ! 1 with rates. 4 n n In this article we study quantitatively the rate of weak convergence of a sequence of finite positive measures to the unit measure given the existence and presence of the left and right Caputo fractional derivatives of the involved function. That is in the right hand sides of the derived inequalities appear the first moduli of continuity of the above mentioned fractional derivatives, see Theorem 25 and Corollary 26. Then via the Riesz representation theorem we transfer Theorem 25 into the language of quantitative pointwise convergence of a sequence of positive linear operators to the unit operator, all operators acting from Cð½a; bÞ into itself, see Theorem 27, Corollary 28 and Theorem 30. From there we derive quantitative results with respect to the sup-norm k k1 , regarding the uniform convergence of positive linear operators to the unit. Again in the right hand side of our inequalities we have moduli of continuity with respect to right and left Caputo derivatives of the engaged function. For the last see Theorem 32, a Shisha–Mond type result. From there we derive the first Korovkin type convergence theorem at the fractional level, see Theorem 33. We give plenty of applications of our fractional Shisha–Mond and Korovkin theory, see Corollaries 35–38. In the background section we present many interesting fractional results which by themselves have their own merit and appear for the first time.
2082
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
This is the first in literature article studying the quantitative convergence of positive linear operators to the unit at the fractional level. In approximation theory the involvement of fractional derivatives is very rare, almost nothing exists. The only fractional articles that exist so far are of Dzyadyk [5] of 1959, Nasibov [11] of 1962, Demjanovic [3] of 1975, and of Jaskolski [9] of 1989, all regarding estimates to best approximation of functions by algebraic and trigonometric polynomials [8].
2. Background We give Definition 4. Let v P 0, n ¼ dv e (de is the ceiling of the number), f 2 AC n ð½a; bÞ (space of functions f with f ðn1Þ 2 ACð½a; bÞ, absolutely continuous functions). We call left Caputo fractional derivative (see [4], p. 38 [7], [14]) the function
Dva f ðxÞ ¼
1 Cðn v Þ
Z
x
ðx tÞnv 1 f ðnÞ ðtÞdt
ð3Þ
a
8x 2 ½a; b, where C is the gamma function Cðv Þ ¼ We set
D0a f ðxÞ
R1 0
et t v 1 dt,
v > 0.
¼ f ðxÞ; 8x 2 ½a; b.
v ¼ 12, then n ¼ 1 and f ðtÞ ¼ tb 2 Cð½0; 1Þ, 0 < b 6 12, t 2 ½0; 1. See that f 0 ðtÞ ¼ btb1 2 L1 ð½0; 1Þ.
Example 5. Take
We observe that 1 b D20 f ðxÞ ¼ 1
C
2
Z
x
1
ðx tÞ2 tb1 dt:
ð4Þ
0
By setting t ¼ xs, dt ¼ x ds, 1 b D20 f ðxÞ ¼ 1
C
2
Z
1
1
ðx xsÞ2 xb1 sb1 xds ¼
0
Cðb þ 1Þ b12 x : C b þ 12
1
Let 0 < b < 12, then D20 f ð0Þ ¼ þ1: Let b ¼ 12, then
pffiffiffiffi
1
p
D20 f ð0Þ ¼
2
>0
ð5Þ
a positive real number! Conclusion: In general for Dva f ðaÞ we do not know what it is, it could be infinite, or finite non-zero, or zero! (see next). Lemma 6. Let
v > 0, v R N, n ¼ dv e, f
2 C n1 ð½a; bÞ and f ðnÞ 2 L1 ð½a; bÞ. Then Dva f ðaÞ ¼ 0.
Proof. By (3) we get
Z
v D f ðxÞ 6 a
1 Cðn v Þ
v D f ðxÞ 6 a
ðnÞ f 1 ðx aÞnv ; Cðn v þ 1Þ
x
f
nv 1 ðnÞ
ðx tÞ
ðtÞdt 6
a
ðnÞ f 1 ðx aÞnv ; Cðn v þ 1Þ
i.e.
That is Dva f ðaÞ ¼ 0.
8x 2 ½a; b:
ð6Þ
We need Definition 7. (see also [7,6,2]) Let f 2 AC m ð½a; bÞ, m ¼ dae, a > 0. The right Caputo fractional derivative of order a > 0 is given by
Dab f ðxÞ ¼
ð1Þm Cðm aÞ
Z
b
ðf xÞma1 f ðmÞ ðfÞdf
x
8x 2 ½a; b. We set D0b f ðxÞ ¼ f ðxÞ. Lemma 8. Let f 2 C m1 ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae, a > 0. Then Dab f ðbÞ ¼ 0. Proof. As in Lemma 6.
ð7Þ
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
Lemma 9. Let f 2 AC m ð½a; bÞ, m ¼ dae, a > 0;
2083
l is a positive finite measure on the Borel r-algebra of ½a; b, x0 2 ½a; b. Then
Z
Z m1 ðkÞ X f ðx0 Þ Ex0 :¼ f ðxÞdlðxÞ ðx x0 Þk dlðxÞ k! ½a;b ½a;b k¼0 (Z Z x0
1 ðf xÞa1 Dax0 f ðfÞ Dax0 f ðx0 Þ df dlðxÞ ¼ CðaÞ ½a;x0 x ) Z x Z
a1 a a þ ðx fÞ Dx0 f ðfÞ Dx0 f ðx0 Þ df dlðxÞ :
ð8Þ
x0
ðx0 ;b
Proof. From [4], p. 40, we get by left Caputo fractional Taylor formula that
f ðxÞ ¼
Z x m1 ðkÞ X f ðx0 Þ 1 ðx fÞa1 Dax0 f ðfÞdf ðx x0 Þk þ k! C ð a Þ x 0 k¼0
ð9Þ
for all x0 < x 6 b. Also from [2], using the right Caputo fractional Taylor formula we get
f ðxÞ ¼
Z x0 m1 ðkÞ X f ðx0 Þ 1 ðf xÞa1 Dax0 f ðfÞdf ðx x0 Þk þ k! C ð a Þ x k¼0
ð10Þ
for all a 6 x 6 x0 . Consequently we find
Z
f ðxÞdlðxÞ ¼
½a;b
Z
f ðxÞdlðxÞ þ
Z
½a;x0
¼
f ðxÞdlðxÞ
ðx0 ;b
Z m1 ðkÞ X f ðx0 Þ ðx x0 Þk dlðxÞ k! ½a;b k¼0 (Z Z x0
1 þ ðf xÞa1 Dax0 f ðfÞdf dlðxÞ CðaÞ ½a;x0 x ) Z x
Z a1 a þ ðx fÞ Dx0 f ðfÞdf dlðxÞ : ðx0 ;b
ð11Þ
x0
Notice also that Dax0 f ðx0 Þ ¼ Dax0 f ðx0 Þ ¼ 0. The proof of (8) is now complete. Convention 10. We assume that
Dax0 f ðxÞ ¼ 0 for x < x0
ð12Þ
Dax0 f ðxÞ ¼ 0 for x > x0
ð13Þ
and
for all x; x0 2 ½a; b. We mention Proposition 11. Let f 2 C n ð½a; bÞ, n ¼ dv e; v > 0. Then Dva f ðxÞ is continuous in x 2 ½a; b. Proof. We notice that
Dva f ðxÞ ¼
1 Cðn v Þ
Dva f ðyÞ ¼
1 Cðn v Þ
Z
xa
znv 1 f ðnÞ ðx zÞdz
0
and
Z
ya
znv 1 f ðnÞ ðy zÞdz:
0
Here a 6 x 6 y 6 b and 0 6 x a 6 y a. Hence it holds
ð14Þ
2084
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
Dva f ðyÞ Dva f ðxÞ ¼
1 Cðn v Þ
Z
xa
znv 1 f ðnÞ ðy zÞ f ðnÞ ðx zÞ dz þ
0
Z
ya
znv 1 f ðnÞ ðy zÞdz :
ð15Þ
xa
We have that
v D f ðyÞ Dv f ðxÞ 6 a a
" # ðnÞ f 1 ðx aÞnv 1 x1 ðf ðnÞ ; jy xjÞ þ ðy aÞnv ðx aÞnv Cðn v Þ ðn v Þ ðn v Þ # ðnÞ nv f 1 ðb aÞ 1 6 x1 ðf ðnÞ ; jy xjÞ þ ðy aÞnv ðx aÞnv : Cðn v Þ ðn v Þ ðn v Þ
So as y ! x the last expression goes to zero. As a result,
Dva f ðyÞ ! Dva f ðxÞ; proving the claim.
ð16Þ
Proposition 12. Let f 2 C m ð½a; bÞ, m ¼ dae, a > 0. Then Dab f ðxÞ is continuous in x 2 ½a; b. Proof. As in Proposition 11.
We also mention Proposition 13. Let f 2 C m1 ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae, a > 0 and
Dax0 f ðxÞ ¼
Z
1 Cðm aÞ
x
ðx tÞma1 f ðmÞ ðtÞdt
ð17Þ
x0
for all x; x0 2 ½a; b : x P x0 . Then Dax0 f ðxÞ is continuous in x0 . Proof. Fix x : x P y0 P x0 ; x; x0 ; y0 2 ½a; b. Then
Z y0 1 ma1 ðmÞ ðx tÞ f ðtÞdt Cðm aÞ x0 ðmÞ Z
y0 f 1 6 ðx tÞma1 dt Cðm aÞ x0 ðmÞ f 1 ¼ ðx y0 Þma ðx x0 Þma Cðm a þ 1Þ
a a Dx0 f ðxÞ Dy0 f ðxÞ ¼
! 0, as y0 ! x0 , proving continuity of Dax0 f in x0 2 ½a; b.
ð18Þ
Proposition 14. Let f 2 C m1 ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae, a > 0 and
Dax0 f ðxÞ ¼
ð1Þm Cðm aÞ
Z
x0
ðf xÞma1 f ðmÞ ðfÞdf
ð19Þ
x
for all x; x0 2 ½a; b : x0 P x. Then Dax0 f ðxÞ is continuous in x0 . Proof. As in Proposition 13.
We need Proposition 15. Let g 2 Cð½a; bÞ, 0 < c < 1, x; x0 2 ½a; b. Define
Lðx; x0 Þ ¼
Z
x
ðx tÞc1 gðtÞdt
for x P x0 ;
ð20Þ
x0
and Lðx; x0 Þ ¼ 0, for x < x0 . Then L is jointly continuous in ðx; x0 Þ on ½a; b2 . Proof. We notice that Lðx0 ; x0 Þ ¼ 0. Assume x P x0 , then
Lðx; x0 Þ ¼
Z 0
xx0
zc1 gðx zÞdz ¼
Z 0
ba
v½0;xx0 ðzÞzc1 gðx zÞdz;
ð21Þ
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
2085
where v is the characteristic function. Let xN ! x; x0N ! x0 ; N 2 N and assume without loss of generality that xN P x0N . So we have again
Z
LðxN ; x0N Þ ¼
xN x0N
zc1 gðxN zÞdz ¼
Z
ba
0
0
v½0;xN x0N ðzÞzc1 gðxN zÞdz:
ð22Þ
We have that
v½0;xN x0N ðzÞ ! v½0;xx0 ðzÞ a:e:
ð23Þ
v½0;xN x0N ðzÞzc1 gðxN zÞ ! v½0;xx0 ðzÞzc1 gðx zÞ a:e:
ð24Þ
and
Notice that
v½0;xN x0N ðzÞzc1 jgðxN zÞj 6 zc1 kg k1 ;
ð25Þ
which is an integrable function. Thus by Dominated Convergence theorem we obtain
LðxN ; x0N Þ ! Lðx; x0 Þ; asN ! 1:
ð26Þ 2
Clearly now Lðx; x0 Þ is jointly continuous on ½a; b .
We mention Proposition 16. Let g 2 Cð½a; bÞ, 0 < c < 1, x; x0 2 ½a; b. Define
Kðx; x0 Þ ¼
Z
x0
ðf xÞc1 gðfÞdf for x 6 x0 ;
ð27Þ
x
and Kðx; x0 Þ ¼ 0, for x > x0 . Then Kðx; x0 Þ is jointly continuous from ½a; b2 into R. Proof. As in Proposition 15.
Based on Propositions 15 and 16 we derive Corollary 17. Let f 2 C m ð½a; bÞ, m ¼ dae, a > 0, x; x0 2 ½a; b. Then Dax0 f ðxÞ, Dax0 f ðxÞ are jointly continuous functions in ðx; x0 Þ from ½a; b2 into R. We need Theorem 18. Let f : ½a; b2 ! R be jointly continuous. Consider
GðxÞ ¼ x1 ðf ð; xÞ; d; ½x; bÞ; d > 0, x 2 ½a; b. Then G is continuous on ½a; b. Proof. (i) Let xn ! x, a 6 xn 6 x, and 0 < d 6 b x first (The case when xn ! x with xn P x is similar). Then we can write
Gðxn Þ ¼ maxðA; B; CÞ; where
A ¼ sup fjf ðu; xn Þ f ðv ; xn Þj; u; v 2 ½x; b; ju v j 6 dg;
0 < d 6 b x;
B ¼ sup fjf ðu; xn Þ f ðv ; xn Þj; u 2 ½xn ; x; v 2 ½x; b; ju v j 6 dg; C ¼ sup fjf ðu; xn Þ f ðv ; xn Þj; u; v 2 ½xn ; x; ju v j 6 dg: Now, when xn ! x, then A ! GðxÞ; B ! KðxÞ 6 GðxÞ; C ! 0 (since also u converges to v). In conclusion, Gðxn Þ ! maxfGðxÞ; KðxÞ; 0g ¼ GðxÞ. (ii) If d > b x, then x1 ðf ð; xÞ; d; ½x; bÞ ¼ x1 ðf ð; xÞ; b x; ½x; bÞ, a case covered by (i). That is proving the claim.
Theorem 19. Let f : ½a; b2 ! R be jointly continuous. Then
2086
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
HðxÞ ¼ x1 ðf ð; xÞ; d; ½a; xÞ; x 2 ½a; b, is continuous in x 2 ½a; b; d > 0. Proof. As in Theorem 18.
We make Remark 20. Let
Z
l be a finite positive measure on Borel r-algebra of ½a; b. Let a > 0, then by Hölder’s inequality we obtain a !ðaþ1Þ
Z
a
ðx0 xÞ dlðxÞ 6
½a;x0
ðx0 xÞ
aþ1
dlðxÞ
ð28Þ
½a;x0
and
Z
1
lð½a; x0 Þðaþ1Þ ; a !ðaþ1Þ
Z
ðx x0 Þa dlðxÞ 6
ðx0 ;b
ðx x0 Þaþ1 dlðxÞ
1
lððx0 ; bÞðaþ1Þ :
ð29Þ
ðx0 ;b
Let now m ¼ dae; a R N, a > 0, k ¼ 1; . . . ; m 1. Then by applying again Hölder’s inequality we obtain
Z
k
jx x0 j dlðxÞ 6
Z
½a;b
k !ðaþ1Þ
j x x0 j
aþ1
dlðxÞ
aþ1k
lð½a; bÞ ðaþ1Þ :
ð30Þ
½a;b
Terminology 21. Let LN : Cð½a; bÞ ! Cð½a; bÞ, N 2 N, be a sequence of positive linear operators. By Riesz representation theorem (see [13], p. 304) we have
LN ðf ; x0 Þ ¼
Z
½a;b
f ðtÞdlNx0 ðtÞ
ð31Þ
8x0 2 ½a; b, where lNx0 is a unique positive finite measure on rBorel algebra of ½a; b. Call LN ð1; x0 Þ ¼ lNx0 ð½a; bÞ ¼ M Nx0 :
ð32Þ
We make Remark 22. Let f 2 C n1 ð½a; bÞ, f ðnÞ 2 L1 ð½a; bÞ, n ¼ dv e; v > 0; v R N. Then as in the proof of Lemma 6, we have
v D f ðxÞ 6 a
ðnÞ f 1 ðx aÞnv Cðn v þ 1Þ
8x 2 ½a; b:
ð33Þ
Thus we observe
x1 ðDva f ; dÞ ¼ sup Dva f ðxÞ Dva f ðyÞ x;y2½a;b jxyj6d
6 sup x;y2½a;b jxyj6d
6 Consequently
x1 ðDva f ; dÞ 6
! ðnÞ ðnÞ f f nv nv 1 1 þ ðx aÞ ðy aÞ Cðn v þ 1Þ Cðn v þ 1Þ
2f ðnÞ 1 ðb aÞnv : Cðn v þ 1Þ
2f ðnÞ 1 ðb aÞnv : Cðn v þ 1Þ
ð34Þ
ð35Þ
ð36Þ
Similarly, let f 2 C m1 ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae; a > 0; a R N, then
2f ðmÞ 1 x1 ðDb f ; dÞ 6 ðb aÞma : Cðm a þ 1Þ a
ð37Þ
So for f 2 C m1 ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae; a > 0; a R N, we find
2f ðmÞ 1 ðb aÞma Cðm a þ 1Þ
ð38Þ
2f ðmÞ 1 6 ðb aÞma : Cðm a þ 1Þ
ð39Þ
sup x1 ðDax0 f ; dÞ½x0 ;b 6
x0 2½a;b
and a
sup x1 ðDx0 f ; dÞ½a;x0
x0 2½a;b
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
2087
We also make Remark 23. Let LN : Cð½a; bÞ ! Cð½a; bÞ, N 2 N, be a sequence of positive linear operators. Using (31) and Hölder’s inequality we obtain ðx 2 ½a; b; k ¼ 1; . . . ; m 1; m ¼ dae; a R N; a > 0Þ for k ¼ 1; . . . ; m 1 that
k ðaþ1kÞ ðaþ1Þ k aþ1 LN ðj xj ; xÞ 6 LN ðj xj ; xÞ kLN 1k1 aþ1 : 1
ð40Þ
1
Also we observe that
Cð½a; bÞ 3 j xjaþ1 v½a;x ðÞ 6 j xjaþ1
8x 2 ½a; b
ð41Þ
Cð½a; bÞ 3 j xjaþ1 v½x;b ðÞ 6 j xjaþ1
8x 2 ½a; b:
ð42Þ
and
By positivity of LN we obtain
aþ1 aþ1 LN ðj xj v½a;x ðÞ; xÞ 6 LN ðj xj ; xÞ 1
ð43Þ
1
and
aþ1 aþ1 LN ðj xj v½x;b ðÞ; xÞ 6 LN ðj xj ; xÞ : 1
ð44Þ
1
So if the right hand side of each of (43,44) tends to zero, so do the left hand sides of these. We also make Remark 24. Let a > 0; a R N. Take a 6 x 6 x0 , then
ðx0 xÞaþ1 6 ðx0 xÞaþ1 1 þ 0: Similarly, for x0 6 x 6 b, we get
ðx x0 Þaþ1 6 0 þ ðx x0 Þaþ1 1: So we have
j xjaþ1 6 j xjaþ1 v½a;x ðÞ þ j xjaþ1 v½x;b ðÞ; 8x 2 ½a; b:
ð45Þ
Thus, by positivity of LN , we obtain
aþ1 aþ1 aþ1 LN j xj ; x 6 LN j xj v½a;x ðÞ; x þ LN j xj v½x;b ðÞ; x : 1
1
1
ð46Þ
So if both kLN ðj xjaþ1 v½a;x ðÞ; xÞk1 , kLN ðj xjaþ1 v½x;b ðÞ; xÞk1 ! 0, as N ! 1, then kLN ðj xjaþ1 ; xÞk1 ! 0. 3. Main results We present our first main result Theorem 25. Let f 2 AC m ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ; m ¼ dae; a R N; a > 0; r1 ; r2 > 0; algebra of ½a; b; x0 2 ½a; b. Then
l is a positive finite measure on the Borel r-
Z Z m 1 ðkÞ X f ðx0 Þ k f ðxÞdlðxÞ ðx x0 Þ dlðxÞ ½a;b k! ½a;b k¼0 1 1 1 6 lð½a; x0 Þaþ1 þ ða þ 1Þr 1 Cða þ01Þ 1 1 !ðaþ1Þ Z a þ1 a A ðx0 xÞ dlðxÞ x1 @Dx0 f ; r 1 ½a;x0
Z
a !ðaþ1 Þ
ðx0 xÞaþ1 dlðxÞ
½a;x0
0
x1 @Dax0 f ; r2
Z ðx0 ;b
ðaþ1Þ
ðx x0 Þaþ1 dlðxÞ
9 a !ðaþ1 Þ= aþ1 : ðx x0 Þ dlðxÞ ; ðx0 ;b
Z
½a;x0
1 1 þ ðlððx0 ; bÞÞðaþ1Þ þ ða þ 1Þr 2 ! 1 1 A ½x0 ;b
ð47Þ
2088
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
Proof. By (8) we obtain
(Z ! Z x0 1 a1 a a ð f xÞ j Dx0 f ðfÞ Dx0 f ðx0 Þdf dlðxÞ CðaÞ ½a;x0 x ) Z x Z
a1 a a þ ðx fÞ Dx0 f ðfÞ Dx0 f ðx0 Þdf dlðxÞ ¼ ðÞ:
Ex 6 0
ð48Þ
x0
ðx0 ;b
Let h1 ; h2 > 0, then
("Z Z x0
1 x0 f df dlðxÞx1 Dax0 f ; h1 ðf xÞa1 1 þ ½a;x0 h1 CðaÞ ½a;x0 x ) "Z # Z x
f x0 df dlðxÞ x1 Dax0 f ; h2 ; þ ðx fÞa1 1 þ ½x0 ;b h2 x0 ðx0 ;b
ðÞ 6
ð49Þ
i.e.
("Z #
Z x0 1 ðx0 xÞa 1 21 a1 þ ðx0 fÞ ðf xÞ df dlðxÞ x1 Dax0 f ; h1 ½a;x0 h CðaÞ a 1 x ½a;x0 ) "Z #
Z x ðx x0 Þa 1 þ þ ðx fÞa1 ðf x0 Þ21 df dlðxÞ x1 Dax0 f ; h2 ½x0 ;b h2 x0 a ðx0 ;b ! ("Z # a aþ1 1 ðx0 xÞ 1 ðx0 xÞ dlðxÞ x1 Dax0 f ; h1 ¼ þ ½a;x0 h1 aða þ 1Þ CðaÞ a ½a;x0 ! "Z # a aþ1 ðx x0 Þ 1 ðx x0 Þ dlðxÞ x1 Dax0 f ; h2 : þ þ ½x0 ;b h2 aða þ 1Þ a ðx0 ;b
Ex 6 0
ð50Þ
Therefore
(" Z # Z 1 1 1 a aþ1 ðx0 xÞ dlðxÞþ ðx0 xÞ dlðxÞ x1 Dax0 f ; h1 ½a;x0 h1 aða þ 1Þ ½a;x0 CðaÞ a ½a;x0 ) " Z # Z 1 1 : þ ðx x0 Þa dlðxÞ þ ðx x0 Þaþ1 dlðxÞ x1 Dax0 f ; h2 ½x0 ;b h2 aða þ 1Þ ðx0 ;b a ðx0 ;b
Ex 6 0
ð51Þ
Momentarily we assume positive choices of
h1 ¼ r 1
Z
1 !ðaþ1Þ
aþ1
ðx0 xÞ
dlðxÞ
>0
ð52Þ
> 0:
ð53Þ
½a;x0
and
h2 ¼ r 2
Z
1 !ðaþ1Þ
ðx x0 Þaþ1 dlðxÞ
ðx0 ;b
Consequently we obtain
Ex 6 0
a nh 1 1 1 h1 ðlð½a; x0 ÞÞðaþ1Þ þ x1 Dax0 f ; h1 ½a;x0 r 1 ða þ 1Þr 1 Cða þ 1Þ a h 1 1 h2 þ ðlððx0 ; bÞÞðaþ1Þ þ x1 Dax0 f ; h2 ; ½x0 ;b r 2 r 2 ða þ 1Þ
ð54Þ
proving (47). R Next we examine special cases. If ðx0 ;b ðx x0 Þaþ1 dlðxÞ ¼ 0, then ðx x0 Þ ¼ 0, a.e. on ðx0 ; b, that is x ¼ x0 a.e. on ðx0 ; b, more precisely lfx 2 ðx0 ; b : x–x0 g ¼ 0, hence lðx0 ; b ¼ 0. Therefore l concentrates on ½a; x0 . In that case inequality (47) is written and holds as
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
Z m1 X f ðkÞ ðx0 Þ Z k f ðxÞd l ðxÞ ðx x Þ d l ðxÞ 0 ½a;x0 k! ½a;x 0 k¼0 1 1 1 ðlð½a; x0 ÞÞðaþ1Þ þ 6 ða þ 1Þr 1 Cða þ 1Þ 0 1 1 !ðaþ1Þ Z a þ1 a A x1 @D f ; r 1 ðx0 xÞ dlðxÞ x0
½a;x0
Z
ðx0 xÞ
aþ1
½a;x0
½a;x0
9 a !ðaþ1Þ = : dlðxÞ ;
2089
ð55Þ
Since ðb; b ¼ ; and lð;Þ ¼ 0, in the case of x0 ¼ b, we get again (55) written for x0 ¼ b. So inequality (55) is a valid inequalR ity when ½a;x0 ðx0 xÞaþ1 dlðxÞ–0. R If additionally we assume that ½a;x0 ðx0 xÞaþ1 dlðxÞ ¼ 0, then ðx0 xÞ ¼ 0, a.e. on ½a; x0 , that is x ¼ x0 a.e. on ½a; x0 , which means lfx 2 ½a; x0 : x–x0 g ¼ 0. Hence l ¼ dx0 M, where dx0 is the unit Dirac measure and M ¼ lð½a; bÞ > 0. In the last case we obtain that L:H:S ð55Þ ¼ R:H:S: ð55Þ ¼ 0, that is (55) is valid trivially. R Finally let us go the other way around. Let us assume that ½a;x0 ðx0 xÞaþ1 dlðxÞ ¼ 0, then reasoning similarly as before we get that l over ½a; x0 concentrates at x0 . That is l ¼ dx0 lð½a; x0 Þ, on ½a; x0 . In the last case (47) is written and it holds as
Z m1 X f ðkÞ ðx0 Þ Z k f ðxÞdlðxÞ ðx x0 Þ dlðxÞ ðx0 ;b k! ðx0 ;b k¼0 1 1 1 6 ðlððx0 ; bÞÞðaþ1Þ þ ða þ 1Þr 2 Cða þ 1Þ 0 1 1 !ðaþ1Þ Z aþ1 a @ A x1 Dx0 f ; r2 ðx x0 Þ dlðxÞ ðx0 ;b
Z ½x0 ;b
ðx x0 Þ
aþ1
ðx0 ;b
9 a !ðaþ1Þ = : dlðxÞ ;
ð56Þ
If x0 ¼ a then (56) can be redone and rewritten, just replace ðx0 ; b by ½a; b all over. So inequality (56) is valid when
Z
ðx x0 Þaþ1 dlðxÞ–0:
ðx0 ;b
R If additionally we assume that ðx0 ;b ðx x0 Þaþ1 dlðxÞ ¼ 0, then as before L:H:S: ð56Þ ¼ R:H:S: ð56Þ ¼ 0. The proof of (47) now has completed in all possible cases.
lðx0 ; b ¼ 0. Hence (56) is trivially true, in fact
We continue in a special case. In the assumptions of Theorem 25, when r ¼ r1 ¼ r2 > 0, and by calling M ¼ lð½a; bÞ P lð½a; x0 Þ; lððx0 ; bÞ, we get Corollary 26. It holds
Z m1 X f ðkÞ ðx0 Þ Z k f ðxÞd l ðxÞ ðx x Þ d l ðxÞ 0 ½a;b k! ½a;b k¼0 2 0 1 1 !ðaþ1Þ Z 1 1 1 6 @ a aþ1 A a þ1Þ ð 6 þ ðx0 xÞ dlðxÞ M 4x1 Dx0 f ; r ða þ 1Þr Cða þ 1Þ ½a;x0
Z ½a;x0
Z
0
a !ðaþ1 Þ
ðx0 xÞaþ1 dlðxÞ
þ x1 @Dax0 f ; r
Z ½x0 ;b
a 3 !ðaþ1 Þ aþ1 5: ðx x0 Þ dlðxÞ
½a;x0
1 1 !ðaþ1Þ aþ1 A ðx x0 Þ dlðxÞ ½x0 ;b
ð57Þ
½x0 ;b
Based on Theorem 25, Corollary 26 and (31), we obtain Theorem 27. Let f 2 AC m ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae; a R N; a > 0; r > 0, and LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, a sequence of positive linear operators, x0 2 ½a; b. Then
2090
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
m1 X f ðkÞ ðx0 Þ k L ðf ; x Þ ððx x Þ ; x Þ L N 0 N 0 0 k! k¼0 1 1 1 6 ðLN ð1; x0 ÞÞðaþ1Þ þ ða þ 1Þr Cð"a þ 1Þ
1 ðaþ1Þ x1 Dax0 f ; r LN jx x0 jaþ1 v½a;x0 ðxÞ; x0
½a;x0
a ðaþ1 Þ LN jx x0 jaþ1 v½a;x0 ðxÞ; x0
1 ðaþ1Þ þ x1 Dax0 f ; r LN jx x0 jaþ1 v½x0 ;b ðxÞ; x0 ½x0 ;b a ðaþ1 Þ aþ1 LN jx x0 j v½x0 ;b ðxÞ; x0 :
ð58Þ
Corollary 28 (to Theorem 27). It holds
m 1 ðkÞ X f ðx0 Þ k LN ððx x0 Þ ; x0 Þ LN ðf ; x0 Þ k! k¼0 1 1 1 6 ðLN ð1; x0 ÞÞðaþ1Þ þ ða þ 1Þr Cð"a þ 1Þ
1 ðaþ1Þ a x1 Dx0 f ; r LN jx x0 jaþ1 ; x0
½a;x0
#
1 ðaþ1Þ þx1 Dax0 f ; r LN jx x0 jaþ1 ; x0
a ðaþ1 Þ LN jx x0 jaþ1 ; x0 :
ð59Þ
½x0 ;b
We make Remark 29. Let f 2 ACð½a; bÞ, f 0 2 L1 ð½a; bÞ, 0 < a < 1; x0 2 ½a; b; LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, sequence of positive linear operators. Then by Theorem 27 and
jLN ðf ; x0 Þ f ðx0 Þj 6 jLN ðf ; x0 Þ f ðx0 ÞLN ð1; x0 Þj þ jf ðx0 ÞjjLN ð1; x0 Þ 1j;
ð60Þ
we obtain Theorem 30. Let f 2 ACð½a; bÞ, f 0 2 L1 ð½a; bÞ, 0 < a < 1; x0 2 ½a; b; LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, sequence of positive linear operators. Then
1 1 1 ðLN ð1; x0 ÞÞðaþ1Þ þ jLN ðf ; x0 Þ f ðx0 Þj 6jf ðx0 ÞjjLN ð1; x0 Þ 1j þ ða þ 1Þr Cða þ 1Þ "
1 ðaþ1Þ aþ1 a x1 Dx0 f ; r LN jx x0 j v½a;x0 ðxÞ; x0 ½a;x0
a ðaþ1 Þ LN jx x0 jaþ1 v½a;x0 ðxÞ; x0
1 ðaþ1Þ þ x1 Dax0 f ; r LN jx x0 jaþ1 v½x0 ;b ðxÞ; x0 ½x0 ;b a ðaþ1 Þ aþ1 LN jx x0 j v½x0 ;b ðxÞ; x0 :
ð61Þ
We make Remark 31. We observe that
R:H:S: ð58Þ 6
1 1 1 kLN ð1Þkð1aþ1Þ þ ða þ 1Þr Cð"a þ 1Þ
1 ðaþ1Þ a sup x1 Dx f ; r LN j xjaþ1 v½a;x ðÞ; x 1
x2½a;b
a ðaþ1 Þ LN j xjaþ1 v½a;x ðÞ; x 1
1 ðaþ1Þ a þ sup x1 Dx f ; r LN j xjaþ1 v½x;b ðÞ; x x2½a;b
a ðaþ1 Þ LN j xjaþ1 v½x;b ðÞ; x ¼ H: 1
1
½a;x
½x;b
ð62Þ
2091
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
So that
m1 X f ðkÞ ðxÞ Z :¼ LN ðf ; xÞ LN ðð xÞk ; xÞ 6 H: k! k¼0
ð63Þ
1
We further observe that
ðkÞ f ðxÞ k LN ðð xÞ ; xÞ k! k¼1 m1 X f ðkÞ ðxÞ k 6 jf ðxÞjjLN ð1; xÞ 1j þ LN ðð xÞ ; xÞ þ H: k! k¼1
jLN ðf ; xÞ f ðxÞj 6 Z þ jf ðxÞjjLN ð1; xÞ 1j þ
m 1 X
ð64Þ
We have proved the main result, a Shisha–Mond type inequality at the fractional level. Theorem 32. Let f 2 AC m ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ, m ¼ dae; a R N; a > 0; r > 0, and LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, a sequence of positive linear operators, x 2 ½a; b. Then
kLN f f k1 6k f k1 kLN 1 1k1 þ
m1 X k¼1
ðkÞ f 1 k 1 LN ðð xÞ ; xÞ þ k! 1 Cða þ 1Þ
1 1 kLN ð1Þkð1aþ1Þ þ ða þ 1Þr " 1
ðaþ1Þ sup x1 Dax f ; rLN ðj xjaþ1 v½a;x ðÞ; xÞ
a ðaþ1Þ aþ1 LN ðj xj v½a;x ðÞ; xÞ
1
ðaþ1Þ þ sup x1 Dax f ; r LN ðj xjaþ1 v½x;b ðÞ; xÞ
a ðaþ1Þ aþ1 : LN ðj xj v½x;b ðÞ; xÞ
1
x2½a;b
1
x2½a;b
½a;x
½x;b
1
1
ð65Þ
Next we derive the following Korovkin type convergence result at fractional level. Theorem 33. Let a R N; a > 0; m ¼ dae, and LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, a sequence of positive linear operators. Assume u u LN 1 ! 1 (uniformly), and kLN ðj xjaþ1 ; xÞk1 ! 0, as N ! 1: Then LN f ! f ; 8f 2 AC m ð½a; bÞ, f ðmÞ 2 L1 ð½a; bÞ. (The second condiu aþ1 tion means ðLN ðj xj ÞÞðxÞ ! 0; x 2 ½a; b.) Proof. Since kLN 1 1k1 ! 0 we get kLN 1 1k1 6 K, for some K > 0. We write LN 1 ¼ LN 1 1 þ 1, hence
kLN 1k1 6 kLN 1 1k1 þ k1k1 6 K þ 1 8N 2 N: That is kLN 1k1 is bounded. So we are using inequality (65). By assumption kLN ðj xjaþ1 ; xÞk1 ! 0 and (40) we get kLN ðj xjk ; xÞk1 ! 0, for all k ¼ 1; . . . ; m 1. Also by (43) and (44) we obtain that
aþ1 aþ1 LN ðj xj v½a;x ðÞ; xÞ ! 0; andLN ðj xj v½x;b ðÞ; xÞ ! 0; 1
1
as N ! 1. Additionally by (38) and (39) we get that
sup x1 ðDax f ; Þ½a;x ; sup x1 ðDax f ; Þ½x;b 6
x2½a;b
x2½a;b
2f ðmÞ 1 ðb aÞma ; Cðm a þ 1Þ
so they are bounded. Thus based on the above, from (65), we derive that kLN f f k1 ! 0, proving the claim.
ð66Þ
We make Remark 34. Based on Corollary Theorems17–19, given that f 2 C m ð½a; bÞ, we obtain that
ðiÞ
1
ðaþ1Þ sup x1 Dax f ; rLN ðj xjaþ1 v½a;x ðÞ; xÞ 1
x2½a;b
½a;x
1
ðaþ1Þ ¼ x1 Dax1 f ; rLN ðj xjaþ1 v½a;x ðÞ; xÞ 1
½a;x
1 ! 0; asLN ðj xjaþ1 ; xÞ ! 0; as N ! 1
1
ð67Þ
2092
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
for some x1 2 ½a; b. Similarly
ðiiÞ
1
ðaþ1Þ sup x1 Dax f ; r LN ðj xjaþ1 v½x;b ðÞ; xÞ 1
x2½a;b
½x;b
1
ðaþ1Þ ¼ x1 Dax2 f ; rLN ðj xjaþ1 v½x;b ðÞ; xÞ 1
½x ;b
2 aþ1 ! 0; as LN ðj xj ; xÞ ! 0; as N ! 1
ð68Þ
1
for some x2 2 ½a; b. We give Corollary 35. Here LN : Cð½a; bÞ ! Cð½a; bÞ; N 2 N, positive linear operators. Let 0 < a < 1; r > 0; f 2 ACð½a; bÞ, f 0 2 L1 ð½a; bÞ. Then
1 1 1 ðaþ1Þ kLN f f k1 6k f k1 kLN 1 1k1 þ kLN ð1Þk1 þ ða þ 1Þr Cða þ 1Þ "( ) 1
ðaþ1Þ sup x1 Dax f ; r LN ðj xjaþ1 v½a;x ðÞ; xÞ 1
x2½a;b
a ðaþ1Þ LN ðj xjaþ1 v½a;x ðÞ; xÞ 1 ( 1
ðaþ1Þ þ sup x1 Dax f ; r LN ðj xjaþ1 v½x;b ðÞ; xÞ 1
x2½a;b
a ðaþ1Þ LN ðj xjaþ1 v½x;b ðÞ; xÞ :
½a;x
)
½x;b
ð69Þ
1
4. Application Consider f 2 Cð½0; 1Þ and the Bernstein polynomials ðBN f ÞðtÞ ¼ We have BN 1 ¼ 1 and BN are positive linear operators. Here let 0 < a < 1; r > 0 and take f 2 ACð½0; 1Þ, f 0 2 L1 ð½0; 1Þ. Applying Corollary 35 we get
PN
k k¼0 f ðN Þ
N k t ð1 tÞNk ; 8t 2 ½0; 1; N 2 N. k
Corollary 36. It holds
" 1
1 1 ðaþ1Þ 1þ sup x1 Dax f ; r BN ðj xjaþ1 v½0;x ðÞ; xÞ kBN f f k1 6 ða þ 1Þr x2½0;1 Cða þ 1Þ 1 ½0;x a 1
ð a þ1Þ ðaþ1Þ þ sup x1 Dax f ; r BN ðj xjaþ1 v½x;1 ðÞ; xÞ BN ðj xjaþ1 v½0;x ðÞ; xÞ 1
a ðaþ1Þ BN ðj xjaþ1 v½x;1 ðÞ; xÞ
x2½0;1
1
ð70Þ
1
8N 2 N. 1 , that is r ¼ 23. Notice C Next let a ¼ 12, and r ¼ aþ1
½x;1
3 2
pffiffiffi ¼ 2p.
Corollary 37. Let f 2 ACð½0; 1Þ, f 0 2 L1 ð½0; 1Þ, N 2 N. Then
" 2
1 3 4 2 3 kBN f f k1 6 pffiffiffiffi sup x1 D2x f ; BN ðj xj2 v½0;x ðÞ; xÞ 1 ½0;x 3 p x2½0;1 1 2
1 3 3 2 3 3 BN ðj xj2 v½0;x ðÞ; xÞ þ sup x1 D2x f ; BN ðj xj2 v½x;1 ðÞ; xÞ 1 1 ½x;1 3 x2½0;1 1 3 3 BN ðj xj2 v½x;1 ðÞ; xÞ : 1
ð71Þ
Here we have
( j t xj
3=2
v½0;x ðtÞ ¼
ðx t Þ3=2
for 0 6 t 6 x
0
for x < t 6 1
ð72Þ
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
2093
and
( jt xj3=2 v½x;1 ðtÞ ¼
ðt xÞ3=2
for x 6 t 6 1;
0
for 0 6 t < x:
ð73Þ
Consequently for x 2 ½0; 1, we find
3=2
½xN X N k 3 k BN j xj2 v½0;x ðÞ ðxÞ ¼ x x ð1 xÞNk N k k¼0
ð74Þ
3=2
N X N k 3 k BN j xj2 v½x;1 ðÞ ðxÞ ¼ x x ð1 xÞNk : N k k¼dxNe
ð75Þ
and
One further has
3 3 BN j xj2 v½0;x ðÞ ðxÞ; BN j xj2 v½x;1 ðÞ ðxÞ 3=2
N X N k 3 x k 6 BN j xj2 ðxÞ ¼ x ð1 xÞNk N k k¼0 € lder’s inequalityÞ ðby discrete Ho !34
2
N X N k k Nk x x ð1 xÞ 6 N k k¼0
34 1 1 8x 2 ½0; 1: ¼ xð1 xÞ 6 N ð4NÞ3=4
ð76Þ ð77Þ
ð78Þ ð79Þ
We have proved. Corollary 38. Let f 2 ACð½0; 1Þ, f 0 2 L1 ð½0; 1Þ, N 2 N. Then
"
# 3 1 1 22 1 1 2 2 ffiffiffi ffi p ffiffiffi ffi p ffiffiffi ffi : x D f ; þ sup x D f ; sup kBN f f k1 6 pffiffiffiffip 1 1 x x p 4 N x2½0;1 3 N ½0;x x2½0;1 3 N ½x;1
ð80Þ
3=2
Notice 2pffiffipffi 1:59. u So as N ! 1 we derive again that BN f ! f with rates. Discussion 39. From (80), Corollary 17, and Theorems 18, 19 we get that
" #
3 1 1 22 1 1 2 2 ffiffiffi ffi p ffiffiffiffi p ffiffiffi ffi þ x D f ; x D f ; kBN f f k1 6 pffiffiffiffip 1 1 x1 x2 p4N 3 N ½0;x1 3 N ½x2 ;1
ð81Þ
for some x1 ; x2 2 ½0; 1; f 2 C 1 ð½0; 1Þ. That is
"
# 3 1 1 22 1 1 2 2 ffiffiffiffi x1 Dx1 f ; pffiffiffiffi þ x1 Dx2 f ; pffiffiffiffi : kBN f f k1 6 pffiffiffiffip 3 N ½0;1 3 N ½0;1 p4N 1
ð82Þ
1
Further we assume that D2x1 f and D2x2 f are Lipschitz functions of order 1, that is
1 1 2 Dx1 f ðxÞ D2x1 f ðyÞ 6 K 1 jx yj
ð83Þ
1 1 2 Dx2 f ðxÞ D2x2 f ðyÞ 6 K 2 jx yj 8x; y 2 ½0; 1 and K 1 ; K 2 > 0:
ð84Þ
and
Then from (82) we get 3
kBN f f k1 6
22 ðK 1 þ K 2 Þ pffiffiffiffi 3 : 3 pN 4
Assume next that f 0 is a Lipschitz function of order 1, that is
ð85Þ
2094
G.A. Anastassiou / Chaos, Solitons and Fractals 42 (2009) 2080–2094
jf 0 ðxÞ f 0 ðyÞj 6 K 3 jx yj 8x; y 2 ½0; 1 and K 3 > 0:
ð86Þ
Then from Section 1, we get
kBN f f k1 6
0:1953125K 3 N
N 2 N:
ð87Þ
In [12], Popoviciu for f 2 Cð½0; 1Þ proved that
kBN f f k1 6
5 1 1 x1 f ; pffiffiffiffi ¼ 1:25x1 f ; pffiffiffiffi : 4 N N
ð88Þ
If f is a Lipschitz function of order 1, that is
jf ðxÞ f ðyÞj 6 K 4 jx yj
ð89Þ
8x; y 2 ½0; 1 and K 4 > 0, then we have kBN f f k1 6
1:25K 4 pffiffiffiffi : N
ð90Þ
We also notice that
1 1 1 < < N N 34 N 12
for N 2 N n f1g:
ð91Þ
So looking at (87, 85) and (90), we observe that as the used in the estimates differentiability of f increases so the resulting speed of convergence of BN f to f increases, in fact at the used 12-derivative the speed is in between the corresponding speeds 1
1
for f and f 0 . Of course in the last argument we assumed that f, D2x1 f ; D2x2 f and f 0 are all Lipschitz functions. If f 0 is a Lipschitz 1
1
function or just f 2 C 1 ð½0; 1Þ, not necessarily D2x1 f ; D2x2 f are Lipschitz ones. References [1] Anastassiou GA. Moments in probability and approximation theory. Pitman research notes in math, vol. 287. Harlow (UK): Longman Sci. & Tech.; 1993. [2] Anastassiou G. On right fractional calculus. Chaos, Solitons and Fractals [accepted for publication]. [3] Dem’janovicˇ J. Approximation by local functions in a space with fractional derivatives. (Lithuanian, English summaries), Differencial’nye Uravnenija i Primenen, Trudy Sem. Processy 1975;103:35–49. [4] Diethelm K. Fractional differential equations. Available from:http://www.tu-bs.de/~diethelm/lehre/f-dgl02/fde-skript.ps.gz. [5] Dzyadyk VK. On the best trigonometric approximation in the L metric of some functions. Dokl Akad Nauk SSSR (Russian) 1959;129:19–22. [6] El-Sayed AMA, Gaber M. On the finite Caputo and finite Riesz derivatives. Electron J Theoret Phys 2006;3(12):81–95. [7] Frederico GS, Torres DFM. Fractional optimal control in the sense of Caputo and the fractional Noether’s theorem. Int Math Forum 2008;3(10):479–93. [8] Gorenflo R, Mainardi F. Essentials of fractional calculus. Maphysto Center; 2000. Available from: http://www.maphysto.dk/oldpages/events/ LevyCAC2000/MainardiNotes/fm2k0a.ps. [9] Jaskólski Miroslaw. Contributions to fractional calculus and approximation theory on the square. Funct Approx Comment Math 1989;18:77–89. [10] Korovkin PP. Linear operators and approximation theory. Delhi (India): Hindustan Publ. Corp.; 1960. [11] Nasibov FG. On the degree of best approximation of functions having a fractional derivative in the Riemann–Liouville sense. (Russian–Azerbaijani summary), Izv Akad Nauk Azerbai˘dzˇan. SSR Ser Fiz-Mat Tehn Nauk 1962;(3):51–7. [12] Popoviciu T. Sur l’approximation des fonctions convexes d’ordre superieur. Mathematica (Cluj) 1935;10:49–54. [13] Royden HL. Real analysis. 2nd ed. New York: Macmillan; 1968. [14] Samko SG, Kilbas AA, Marichev OI. Fractional integrals and derivatives, theory and applications. Amsterdam: Gordon & Breach; 1993 [English translation from the Russian, Integrals and derivatives of fractional order and some of their applications. Minsk: Nauka i Tekhnika; 1987]. [15] Shisha O, Mond B. The degree of convergence of sequences of linear positive operators. Natl Acad Sci US 1968;60:1196–200.