An abstract Korovkin type theorem and applications

Abstract

We obtain some Korovkin type theorems for the space \(C(X)\), where \( X \) is a compact metric space (Theorems 2 and 3).

The results are applied to the case when \(X\) is a compact subspace of a prehilbertian space and we obtain bounds for the difference \(  \| B_{n}(f)-f \| \), where \(B_{n}\) is the Bernstein-Lototsky-Schnabl operator.

Authors

Dorin Andrica
Babeș-Bolyai University, Romania

Costica Mustata
“Tiberiu Popoviciu” Institute of Numerical Analysis, Romanian Academy, Romania

Keywords

Paper coordinates

D. Andrica, C. Mustăţa, An abstract Korovkin type theorem and applications, Studia Univ. ”Babeş-Bolyai” XXXIV, fasc. 2 (1989), 44-51.

PDF

About this paper

Journal

Studia

Publisher Name

Univ. Babes-Bolyai Math.

DOI
Print ISSN

1843-3855

Online ISSN

2065-9490

MR # 91j: 41043

google scholar link

[1] De Vore, Ronald A., The Approximation of Con tinuous Functions by Positive Linear Operators, Springer-Verlag, Berlin-Heidelberg-New York, 1972.
[2] Mamedov, R.G., On the order lf the approximation of differentiable funcitons by linear operators (Russian), Doklady SSSR 128 (1959), 674-676.
[3] Shisha, O., Mond, B., The degree of convergence of sequences of linear pozitive operators,  Proc. Nat. Acad. Sci. USA 60 (1968), 1196-1200.
[4] Nishishiraho, T., The degree of convergence of positive linear operators, Tohoku Math. Journal 29 (1977), 81-89.
[5] Nishishiraho, T., Convergence of positive linear approximation processes, Tohoku Math. Journal 35 (1983), 441-458.

Paper (preprint) in HTML form

1989b-Mustata-Studia-UBB-An-abstract-Korovkin-type-theorem-and-applications

AN ABSTRACT KOROVKIN TYPE THEOREM AND APPIICATIONS

D. ANDRICA* and C. MUSTATTA*

Received: January 26, 1989
REZUMAT. - 0 teoremă abstractă de tip Korovkin şi aplicafil. In lucrare se obtin teoreme de tip Korovkin pentru spațiul C ( X ) C ( X ) C(X)C(X)C(X), unde X X XXX este un spaţiu metric compact (Teoremele 2 și 3). Se aplică rezultatele obținute pentru cazul cind X X XXX este o submulțime compactă a unui spațiu prehilbertian şi se dau delimitări ale diferentei B n ( f ) f B n ( f ) f ||B_(n)(f)-f||\left\|B_{n}(f)-f\right\|Bn(f)f, unde B n B n B_(n)B_{n}Bn este operatorul lui Bernstein-LototskySchnabl.
The well known Korovkin's theorem (see e.g. [1]) asserts that if ( L n ) n 1 L n n 1 (L_(n))_(n)oint_(1)\left(L_{n}\right)_{n} \oint_{1}(Ln)n1 is a sequence of positive linear operators, acting from C [ a , b ] C [ a , b ] C[a,b]C[a, b]C[a,b] to C [ a , b ] C [ a , b ] C[a,b]C[a, b]C[a,b] and such that ( L n ( e k ) ) n 1 L n e k n 1 (L_(n)(e_(k)))_(n >= 1)\left(L_{n}\left(e_{k}\right)\right)_{n \geqslant 1}(Ln(ek))n1 converges uniformly to c k c k c_(k)c_{k}ck, for k = 0 , 1 , 2 k = 0 , 1 , 2 k=0,1,2k=0,1,2k=0,1,2, where c k ( t ) = t k c k ( t ) = t k c_(k)(t)=t^(k)c_{k}(t)=t^{k}ck(t)=tk, i [ a , b ] i [ a , b ] i in[a,b]i \in[a, b]i[a,b], then the sequence ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n >= 1)\left(L_{n}(f)\right)_{n \geqslant 1}(Ln(f))n1 converges uniformly to f f fff, for every f ∈∈ C [ a , b ] f ∈∈ C [ a , b ] f∈∈C[a,b]f \in \in C[a, b]f∈∈C[a,b].
This theorem was extended and generalized in many directions. One direction is to replace the above mentioned system of test functions by other systems of functions, which led to the theory of so called Korovkin subspaces. Another direction is to consider functions defined on more general compact spaces than the interval [ a , b ] [ a , b ] [a,b][a, b][a,b], first of all on compact subsets of R m R m R^(m)\mathbf{R}^{m}Rm.
The aim of this paper is to give Korovkin type theorems for the space C ( X ) C ( X ) C(X)C(X)C(X), where X X XXX is a compact metric space. As application, supposing that X X XXX is a compact convex subset of a Hilbert space, one obtains evaluations of the order of approximation by the Bernstein - Lototsky - Schnabl operator, similar to those given in [4].
If ( X , d X , d X,dX, dX,d ) is a compact metric space, denote by C ( X ) = C ( X , R ) C ( X ) = C ( X , R ) C(X)=C(X,R)C(X)=C(X, \mathbf{R})C(X)=C(X,R) the space of all real-valued continuous functions defined on X X XXX and by Lip ( X X XXX ) the subspace of C ( X ) : C ( X ) : C(X)^(:)C(X)^{:}C(X):formed of all real-valued Lipschitz functions defined on λ λ lambda\lambdaλ. Equiped, as usually, with the uniform norm f = sup { | f ( x ) | : x X } , f ∈∈ C ( X ) f = sup { | f ( x ) | : x X } , f ∈∈ C ( X ) ||f||=s u p{|f(x)|:x in X},f∈∈C(X)\|f\|=\sup \{|f(x)|: x \in X\}, f \in \in C(X)f=sup{|f(x)|:xX},f∈∈C(X), the space C ( X ) C ( X ) C(X)C(X)C(X) is a Banach space.
Our first result is a density theorem:
theorem 1. The subspace Lip ( X X XXX ) is dense in C ( X ) C ( X ) C(X)C(X)C(X), with respect to the uniform norm.
Proof. The assertion of the theorem will follow from the Stone-Weicstrass theorem if we shall show that Lip ( X ) ( X ) (X)(X)(X) is a subalgebra of C ( X ) C ( X ) C(X)C(X)C(X) containing the constant functions and separating the points of X X XXX.
If f , g Lip ( X ) f , g Lip ( X ) f,g in Lip(X)f, g \in \operatorname{Lip}(X)f,gLip(X) then
| ( f g ) ( x ) ( f g ) ( y ) | | f ( x ) | | g ( x ) g ( y ) | + | g ( y ) | | f ( x ) f ( y ) | ( f K g + g K f ) d ( x , y ) | ( f g ) ( x ) ( f g ) ( y ) | | f ( x ) | | g ( x ) g ( y ) | + | g ( y ) | | f ( x ) f ( y ) | f K g + g K f d ( x , y ) {:[|(f*g)(x)-(f*g)(y)| <= |f(x)|*|g(x)-g(y)|+|g(y)|*|f(x)-f(y)| <= ],[ <= (||f||*K_(g)+||g||*K_(f))*d(x","y)]:}\begin{gathered} |(f \cdot g)(x)-(f \cdot g)(y)| \leqslant|f(x)| \cdot|g(x)-g(y)|+|g(y)| \cdot|f(x)-f(y)| \leqslant \\ \leqslant\left(\|f\| \cdot K_{g}+\|g\| \cdot K_{f}\right) \cdot d(x, y) \end{gathered}|(fg)(x)(fg)(y)||f(x)||g(x)g(y)|+|g(y)||f(x)f(y)|(fKg+gKf)d(x,y)
for all x , y X x , y X x,y in Xx, y \in Xx,yX, where K f K f K_(f)K_{f}Kf and K g K g K_(g)K_{g}Kg are Lipschitz constants for f f fff and g g ggg, respectively. Therefore f g Lip ( X ) f g Lip ( X ) f^(**)g in Lip(X)f^{*} g \in \operatorname{Lip}(X)fgLip(X) and since Lip ( X ) Lip ( X ) Lip(X)\operatorname{Lip}(X)Lip(X) is a subspace of C ( X ) C ( X ) C(X)C(X)C(X) it follows that Lip ( X ) ( X ) (X)(X)(X) is a subalgebra of the algebra C ( X ) C ( X ) C(X)C(X)C(X).
  • As the constant functions are obviously in Lip ( X ) ( X ) (X)(X)(X) to finish the proof we have only to show that the algebra Lip ( X ) Lip ( X ) Lip(X)\operatorname{Lip}(X)Lip(X) separates the points of X X XXX. For x , y X , x y x , y X , x y x,y in X,x!=yx, y \in X, x \neq yx,yX,xy let f : X R f : X R f:X rarrRf: X \rightarrow \mathbf{R}f:XR be defined by f ( z ) = d ( z , y ) , z X f ( z ) = d ( z , y ) , z X f(z)=d(z,y),z in Xf(z)=d(z, y), z \in Xf(z)=d(z,y),zX. Then
| f ( z 1 ) f ( z 2 ) | = | d ( z 1 , y ) d ( z 2 , y ) | d ( z 1 , z 2 ) , z 1 , z 2 X , f z 1 f z 2 = d z 1 , y d z 2 , y d z 1 , z 2 , z 1 , z 2 X , |f(z_(1))-f(z_(2))|=|d(z_(1),y)-d(z_(2),y)| <= d(z_(1),z_(2)),quadz_(1),z_(2)in X,\left|f\left(z_{1}\right)-f\left(z_{2}\right)\right|=\left|d\left(z_{1}, y\right)-d\left(z_{2}, y\right)\right| \leqslant d\left(z_{1}, z_{2}\right), \quad z_{1}, z_{2} \in X,|f(z1)f(z2)|=|d(z1,y)d(z2,y)|d(z1,z2),z1,z2X,
which shows that f f fff is in Lip ( X ) , f ( y ) = d ( y , y ) = 0 ( X ) , f ( y ) = d ( y , y ) = 0 (X),f(y)=d(y,y)=0(X), f(y)=d(y, y)=0(X),f(y)=d(y,y)=0 and f ( x ) = d ( x , y ) > 0 f ( x ) = d ( x , y ) > 0 f(x)=d(x,y) > 0f(x)=d(x, y)>0f(x)=d(x,y)>0. Theorem is proved.
A Markov operator L L LLL on C ( X ) C ( X ) C(X)C(X)C(X) is a positive linear operator L : C ( X ) C ( X ) L : C ( X ) C ( X ) L:C(X)longrightarrow C(X)L: C(X) \longrightarrow C(X)L:C(X)C(X) such that L ( c 0 ) = e 0 L c 0 = e 0 L(c_(0))=e_(0)L\left(c_{0}\right)=e_{0}L(c0)=e0, where e 0 ( x ) = 1 , x X e 0 ( x ) = 1 , x X e_(0)(x)=1,x in Xe_{0}(x)=1, x \in Xe0(x)=1,xX, i.e. L L LLL preserves the constant functions.
In the following we shall need the following simple lemma:
I.EMMA 1. If L L LLL is a Markov operator acting on C ( X ) C ( X ) C(X)C(X)C(X) then L = 1 L = 1 ||L||=1\|L\|=1L=1.
Proof. Taking into account the positivity of L L LLL and applying L L LLL to the inequalities - f e 0 f f e 0 f e 0 f f e 0 ||f||*e_(0) <= f <= ||f||*e_(0)\|f\| \cdot e_{0} \leqslant f \leqslant\|f\| \cdot e_{0}fe0ffe0, we obtain - f e 0 L ( f ) f e 0 f e 0 L ( f ) f e 0 ||f||*e_(0) <= L(f) <= ||f||*e_(0)\|f\| \cdot e_{0} \leqslant L(f) \leqslant\|f\| \cdot e_{0}fe0L(f)fe0, so that L ( f ) f L ( f ) f ||L(f)|| <= ||f||\|L(f)\| \leqslant\|f\|L(f)f, for all f C ( X ) f C ( X ) *f in C(X)\cdot f \in C(X)fC(X). As L ( c 0 ) = c 0 = 1 L c 0 = c 0 = 1 ||L(c_(0))||=||c_(0)||=1\left\|L\left(c_{0}\right)\right\|=\left\|c_{0}\right\|=1L(c0)=c0=1 it follows L := 1 L := 1 ||L:=1\| L:=1L:=1. Lemma is proved.
If ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geqslant 1}(Ln)n1 is a sequence of Markov operators acting on C ( X ) C ( X ) C(X)C(X)C(X), let
(1) α n ( x ) = L n ( d ( . , x ) ; x ) , β n ( x ) = I n ( d 2 ( . , x ) ; x ) , (1) α n ( x ) = L n ( d ( . , x ) ; x ) , β n ( x ) = I n d 2 ( . , x ) ; x , {:[(1)alpha_(n)(x)=L_(n)(d(.","x);x)","],[beta_(n)(x)=I_(n)(d^(2)(.,x);x)","]:}\begin{align*} & \alpha_{n}(x)=L_{n}(d(., x) ; x), \tag{1}\\ & \beta_{n}(x)=I_{n}\left(d^{2}(., x) ; x\right), \end{align*}(1)αn(x)=Ln(d(.,x);x),βn(x)=In(d2(.,x);x),
for all x X x X x in Xx \in XxX and n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,.
Our first Korovin type theorem is the following:
THEOREM 2. Let ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geqslant 1}(Ln)n1 be a sequence of Markov operators acting on C ( X ) C ( X ) C(X)C(X)C(X). If ( α n ( x ) ) n 1 α n ( x ) n 1 (alpha_(n)(x))_(n >= 1)\left(\alpha_{n}(x)\right)_{n \geqslant 1}(αn(x))n1 converges to zero, uniformly with respect to x X x X x in Xx \in XxX, then ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n >= 1)\left(L_{n}(f)\right)_{n \geqslant 1}(Ln(f))n1 conorges uniformly to f f fff, for all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X).
Proof. Let f Lip ( X ) f Lip ( X ) f in Lip(X)f \in \operatorname{Lip}(X)fLip(X) and let K f 0 K f 0 K_(f) >= 0K_{f} \geqslant 0Kf0 be a Lipschitz constant for f f fff, i.e.
| f ( x ) f ( y ) | K f d ( x , y ) , | f ( x ) f ( y ) | K f d x , y , |f(x)-f(y)| <= K_(f)*d(x,y^(')),|f(x)-f(y)| \leqslant K_{f} \cdot d\left(x, y^{\prime}\right),|f(x)f(y)|Kfd(x,y),
for all x , y X x , y X x,y in Xx, y \in Xx,yX. This inequality can be rewritten in the form:
K f d ( . , x ) f ( ) f ( x ) c 0 K f d ( . , x ) , K f d ( . , x ) f ( ) f ( x ) c 0 K f d ( . , x ) , -K_(f)*d(.,x) <= f(*)-f(x)*c_(0) <= K_(f)*d(.,x),-K_{f} \cdot d(., x) \leqslant f(\cdot)-f(x) \cdot c_{0} \leqslant K_{f} \cdot d(., x),Kfd(.,x)f()f(x)c0Kfd(.,x),
for all x X x X x in Xx \in XxX. Applying to these inequalities the operator L n L n L_(n)L_{n}Ln and taking into account the positivity of L n L n L_(n)L_{n}Ln and the notations (1), one obtains:
K f α n ( x ) I n ( f ; x ) f ( x ) K f α n ( x ) K f α n ( x ) I n ( f ; x ) f ( x ) K f α n ( x ) -K_(f)*alpha_(n)(x) <= I_(n)(f;x)-f(x) <= K_(f)*alpha_(n)(x)-K_{f} \cdot \alpha_{n}(x) \leqslant I_{n}(f ; x)-f(x) \leqslant K_{f} \cdot \alpha_{n}(x)Kfαn(x)In(f;x)f(x)Kfαn(x)
for all x X x X x in Xx \in XxX, or equivalently,
(2) | L n ( f ; x ) f ( x ) | K f | α n ( x ) | , (2) L n ( f ; x ) f ( x ) K f α n ( x ) , {:(2)|L_(n)(f;x)-f(x)| <= K_(f)|alpha_(n)(x)|",":}\begin{equation*} \left|L_{n}(f ; x)-f(x)\right| \leqslant K_{f}\left|\alpha_{n}(x)\right|, \tag{2} \end{equation*}(2)|Ln(f;x)f(x)|Kf|αn(x)|,
for all x X x X x in Xx \in XxX. Since, by the hypothesis of the theorem the sequence ( α n ( x ) ) n 1 α n ( x ) n 1 (alpha_(n)(x))_(n >= 1)\left(\alpha_{n}(x)\right)_{n \geqslant 1}(αn(x))n1 tends to zero, uniformly for x X x X x in Xx \in XxX, the inequality (2) implies that ( L n ( f ) ) n > 1 L n ( f ) n > 1 (L_(n)(f))_(n > 1)\left(L_{n}(f)\right)_{n>1}(Ln(f))n>1 tends uniformly to f f fff.
By Theorem 1 the space Lip ( X ) Lip ( X ) Lip(X)\operatorname{Lip}(X)Lip(X) is deuse in C ( X ) C ( X ) C(X)C(X)C(X) with respect to the uniform norm on C ( X ) C ( X ) C(X)C(X)C(X) and by Lemma 1 , L n = 1 , n = 1 , 2 , 1 , L n = 1 , n = 1 , 2 , 1,||L_(n)||=1,n=1,2,dots1,\left\|L_{n}\right\|=1, n=1,2, \ldots1,Ln=1,n=1,2, so that by the
Banach-Steinhaus theorem, the sequence ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n >= 1)\left(L_{n}(f)\right)_{n \geqslant 1}(Ln(f))n1 tends uniformly to f f fff, for all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X). The theorem is proved.
THEOREM 3. Let ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geq 1}(Ln)n1 be a sequence of Markov operators acting on C ( X ) C ( X ) C(X)C(X)C(X). If β n ( x ) β n ( x ) beta_(n)(x)\beta_{n}(x)βn(x) is defined by (1) and the sequence ( β n ( x ) ) n 1 β n ( x ) n 1 (beta_(n)(x))_(n >= 1)\left(\beta_{n}(x)\right)_{n \geqslant 1}(βn(x))n1 tends to zero, uniformly fw. 5 5 5 5 _(5)^(5){ }_{5}^{5}55 respect to x X x X x in Xx \in XxX, then the sequence ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n >= 1)\left(L_{n}(f)\right)_{n \geqslant 1}(Ln(f))n1 tends uniformly to f f fff, for all ( C ( X ) C ( X ) in C(X)\in C(X)C(X).
If f Lip ( X ) f Lip ( X ) f in Lip(X)f \in \operatorname{Lip}(X)fLip(X) then, furthemore
(3) L n ( f ) f K f β n , (3) L n ( f ) f K f β n , {:(3)||L_(n)(f)-f|| <= K_(f)*sqrt(||beta_(n)||)",":}\begin{equation*} \left\|L_{n}(f)-f\right\| \leqslant K_{f} \cdot \sqrt{\left\|\beta_{n}\right\|}, \tag{3} \end{equation*}(3)Ln(f)fKfβn,
for all n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,.
Proof. We have L n ( e 0 ) = e 0 L n e 0 = e 0 L_(n)(e_(0))=e_(0)L_{n}\left(e_{0}\right)=e_{0}Ln(e0)=e0 and
0 L n ( ( t f c 0 ) 2 ) = t 2 L n ( f 2 ) 2 t L n ( f ) + e 0 . 0 L n t f c 0 2 = t 2 L n f 2 2 t L n ( f ) + e 0 . 0 <= L_(n)((t*f-c_(0))^(2))=t^(2)L_(n)(f^(2))-2t*L_(n)(f)+e_(0).0 \leqslant L_{n}\left(\left(t \cdot f-c_{0}\right)^{2}\right)=t^{2} L_{n}\left(f^{2}\right)-2 t \cdot L_{n}(f)+e_{0} .0Ln((tfc0)2)=t2Ln(f2)2tLn(f)+e0.
for all t R t R t in Rt \in RtR, implying
[ L n ( f ) ] 2 L n ( f 2 ) , L n ( f ) 2 L n f 2 , [L_(n)(f)]^(2) <= L_(n)(f^(2)),\left[L_{n}(f)\right]^{2} \leqslant L_{n}\left(f^{2}\right),[Ln(f)]2Ln(f2),
for all f ˙ C ( X ) f ˙ C ( X ) fin^(˙)C(X)f \dot{\in} C(X)f˙C(X). Applying this mequality to the function f = d ( . , x ) f = d ( . , x ) f=d(.,x)f=d(., x)f=d(.,x), one obtains:
(4) ( L n ( d ( . , x ) ; x ) ) 2 L n ( d 2 ( . , x ) ; x ) , (4) L n ( d ( . , x ) ; x ) 2 L n d 2 ( . , x ) ; x , {:(4)(L_(n)(d(.,x);x))^(2) <= L_(n)(d^(2)(.,x);x)",":}\begin{equation*} \left(L_{n}(d(., x) ; x)\right)^{2} \leqslant L_{n}\left(d^{2}(., x) ; x\right), \tag{4} \end{equation*}(4)(Ln(d(.,x);x))2Ln(d2(.,x);x),
for all x X x X x in Xx \in XxX. Taking into account the notations (1), it follows that the sequence ( α n ( x ) ) n 1 α n ( x ) n 1 (alpha_(n)(x))_(n >= 1)\left(\alpha_{n}(x)\right)_{n \geqslant 1}(αn(x))n1 converges to zero, uniformly for x ˙ X x ˙ X x^(˙)in X\dot{x} \in Xx˙X, provided that the sequence ( β n ( x ) ) n 1 β n ( x ) n 1 (beta_(n)(x))_(n >= 1)\left(\beta_{n}(x)\right)_{n \geq 1}(βn(x))n1 converges to zero uniformly for x X x X x in Xx \in XxX. The first assertion of the theorem follows now from Theorem 2.
The inequality (2), obtained in the proof of Theorem 2, implies
L n ( f ) f K f α n , L n ( f ) f K f α n , ||L_(n)(f)-f|| <= K_(f)*||alpha_(n)||,\left\|L_{n}(f)-f\right\| \leqslant K_{f} \cdot\left\|\alpha_{n}\right\|,Ln(f)fKfαn,
for all f Lip ( X ) f Lip ( X ) f in Lip(X)f \in \operatorname{Lip}(X)fLip(X). By the inequality (4), α n β n α n β n ||alpha_(n)|| <= sqrt(||beta_(n)||)\left\|\alpha_{n}\right\| \leqslant \sqrt{\left\|\beta_{n}\right\|}αnβn, so that
L n ( f ) f K f β n , L n ( f ) f K f β n , ||L_(n)(f)-f|| <= K_(f)*sqrt(||beta_(n)||),\left\|L_{n}(f)-f\right\| \leqslant K_{f} \cdot \sqrt{\left\|\beta_{n}\right\|},Ln(f)fKfβn,
which ends the proof of the theorem.
Now, let H H HHH be a real pre-Hilbert space with inner product . , . , (:.,\langle.,.,. . F o r t H . F o r t H :).Fort in H\rangle . For t \in H.FortH fixed let the function c t : H R c t : H R c_(t):H rarrRc_{t}: H \rightarrow \mathbf{R}ct:HR be defined by c t ( x ) = x , t , x H c t ( x ) = x , t , x H c_(t)(x)=(:x,t:),x in Hc_{t}(x)=\langle x, t\rangle, x \in Hct(x)=x,t,xH, and let e : H R e : H R e:H rarrRe: H \rightarrow \mathbf{R}e:HR be defined by e ( x ) = x , x = x 2 , x H e ( x ) = x , x = x 2 , x H e(x)=(:x,x:)=||x||^(2),x in He(x)=\langle x, x\rangle=\|x\|^{2}, x \in He(x)=x,x=x2,xH.
theorem 4. Let X X XXX be a compact subset of the pre-Hilbert space H H HHH and let ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geqslant 1}(Ln)n1 be a sequence of Markov operators acting on C ( X ) C ( X ) C(X)C(X)C(X). If ( L n ( e ) ) n 1 L n ( e ) n 1 (L_(n)(e))_(n >= 1)\left(L_{n}(e)\right)_{n \geqslant 1}(Ln(e))n1 converges uniformly to e e eee and the sequence ( L n ( e x ; x ) ) n 1 L n e x ; x n 1 (L_(n)(e_(x);x))_(n >= 1)\left(L_{n}\left(e_{x} ; x\right)\right)_{n \geq 1}(Ln(ex;x))n1 converges to c ( x ) c ( x ) c(x)c(x)c(x), uniformly for x X x X x in Xx \in XxX, then the sequence ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n rarr1)\left(L_{n}(f)\right)_{n \rightarrow 1}(Ln(f))n1 converges uniformly to f f fff, for all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X).
Proof. We have
t x 2 = c ( t ) 2 e x ( t ) + e ( x ) . t x 2 = c ( t ) 2 e x ( t ) + e ( x ) . ||t-x||^(2)=c(t)-2e_(x)(t)+e(x).\|t-x\|^{2}=c(t)-2 e_{x}(t)+e(x) .tx2=c(t)2ex(t)+e(x).
Considering x x xxx fixed and t t ttt variable, applying the operator L n L n L_(n)L_{n}Ln to this equality and evaluating at the point t = x t = x t=xt=xt=x, one obtains:
β n ( x ) = L n ( x 2 ; x ) = L n ( c ; x ) 2 L n ( e x ; x ) + e ( x ) = (5) = L n ( e ; x ) e ( x ) 2 [ L n ( e x ; x ) c ( x ) ] . β n ( x ) = L n x 2 ; x = L n ( c ; x ) 2 L n e x ; x + e ( x ) = (5) = L n ( e ; x ) e ( x ) 2 L n e x ; x c ( x ) . {:[beta_(n)(x)=L_(n)(||*-x||^(2);x)=L_(n)(c;x)-2L_(n)(e_(x);x)+e(x)=],[(5)=L_(n)(e;x)-e(x)-2[L_(n)(e_(x);x)-c(x)].]:}\begin{gather*} \beta_{n}(x)=L_{n}\left(\|\cdot-x\|^{2} ; x\right)=L_{n}(c ; x)-2 L_{n}\left(e_{x} ; x\right)+e(x)= \\ =L_{n}(e ; x)-e(x)-2\left[L_{n}\left(e_{x} ; x\right)-c(x)\right] . \tag{5} \end{gather*}βn(x)=Ln(x2;x)=Ln(c;x)2Ln(ex;x)+e(x)=(5)=Ln(e;x)e(x)2[Ln(ex;x)c(x)].
Taking into account the hypotheses of the theorem it follows that the sequence ( β n ( x ) ) n 1 β n ( x ) n 1 (beta_(n)(x))_(n >= 1)\left(\beta_{n}(x)\right)_{n \geqslant 1}(βn(x))n1 converges to zero uniformly for x X x X x in Xx \in XxX, and Theorem 4 follows from Theorem 3.
Remark. If f Lip ( X ) f Lip ( X ) f in Lip(X)f \in \operatorname{Lip}(X)fLip(X) then
(6) L n ( f ) f K f a n 2 b n , (6) L n ( f ) f K f a n 2 b n , {:(6)||L_(n)(f)-f|| <= K_(f)sqrt(||a_(n)-2b_(n)||)",":}\begin{equation*} \left\|L_{n}(f)-f\right\| \leqslant K_{f} \sqrt{\left\|a_{n}-2 b_{n}\right\|}, \tag{6} \end{equation*}(6)Ln(f)fKfan2bn,
where a n ( x ) = L n ( e ; x ) e ( x ) a n ( x ) = L n ( e ; x ) e ( x ) a_(n)(x)=L_(n)(e;x)-e(x)a_{n}(x)=L_{n}(e ; x)-e(x)an(x)=Ln(e;x)e(x) and b n ( x ) = L n ( e x ; x ) e ( x ) b n ( x ) = L n e x ; x e ( x ) b_(n)(x)=L_(n)(e_(x);x)-e(x)b_{n}(x)=L_{n}\left(e_{x} ; x\right)-e(x)bn(x)=Ln(ex;x)e(x), for x X x X x in Xx \in XxX and n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,.
COROLLARY 1. (Korovkin's theorem). If ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geqslant 1}(Ln)n1 is a sequence of Markov opcrators acting on C [ a , b ] C [ a , b ] C[a,b]C[a, b]C[a,b] such that L n ( e 1 ) u e 1 , L n ( e 2 ) u e 2 L n e 1 u e 1 , L n e 2 u e 2 L_(n)(e_(1))rarr"u"e_(1),L_(n)(e_(2))rarr"u"e_(2)L_{n}\left(e_{1}\right) \xrightarrow{u} e_{1}, L_{n}\left(e_{2}\right) \xrightarrow{u} e_{2}Ln(e1)ue1,Ln(e2)ue2, where e 1 ( x ) = x e 1 ( x ) = x e_(1)(x)=xe_{1}(x)=xe1(x)=x and c 2 ( x ) = x 2 , x [ a , b ] c 2 ( x ) = x 2 , x [ a , b ] c_(2)(x)=x^(2),x in[a,b]c_{2}(x)=x^{2}, x \in[a, b]c2(x)=x2,x[a,b], then ( L n ( f ) ) n 1 L n ( f ) n 1 (L_(n)(f))_(n >= 1)\left(L_{n}(f)\right)_{n \geqslant 1}(Ln(f))n1 converges uniformly to f f fff, for all f C [ a , b ] f C [ a , b ] f in C[a,b]f \in C[a, b]fC[a,b].
Proof. In Theorem 4 take H = R , X = [ a , b ] H = R , X = [ a , b ] H=R,X=[a,b]H=\mathbf{R}, X=[a, b]H=R,X=[a,b] and the inner product be the usual multiplication in R , x , y = x y R , x , y = x y R,(:x,y:)=x*y\mathbf{R},\langle x, y\rangle=x \cdot yR,x,y=xy. Then e ( x ) = x 2 = e 2 ( x ) , e 1 ( x ) == t x = t e 1 ( x ) e ( x ) = x 2 = e 2 ( x ) , e 1 ( x ) == t x = t e 1 ( x ) e(x)=x^(2)=e_(2)(x),e_(1)(x)==t*x=t*e_(1)(x)e(x)=x^{2}=e_{2}(x), e_{1}(x)= =t \cdot x=t \cdot e_{1}(x)e(x)=x2=e2(x),e1(x)==tx=te1(x) and L n ( e t ; x ) = t L n ( e 1 ; x ) L n e t ; x = t L n e 1 ; x L_(n)(e_(t);x)=t*L_(n)(e_(1);x)L_{n}\left(e_{t} ; x\right)=t \cdot L_{n}\left(e_{1} ; x\right)Ln(et;x)=tLn(e1;x). By hypothesis L n ( e ) = L n ( e 2 ) u e 2 = e L n ( e ) = L n e 2 u e 2 = e L_(n)(e)=L_(n)(e_(2))rarr"u"e_(2)=eL_{n}(e)=L_{n}\left(e_{2}\right) \xrightarrow{u} e_{2}=eLn(e)=Ln(e2)ue2=e. The corrollary will follow from Theorem 4 if we show that L n ( e x ; x ) x 2 L n e x ; x x 2 L_(n)(e_(x);x)rarrx^(2)L_{n}\left(e_{x} ; x\right) \rightarrow x^{2}Ln(ex;x)x2 uniformly for x [ a , b ] x [ a , b ] x in[a,b]x \in[a, b]x[a,b]. By hypothesis L n ( e 1 ) n e 1 L n e 1 n e 1 L_(n)(e_(1))rarr"n"e_(1)L_{n}\left(e_{1}\right) \xrightarrow{n} e_{1}Ln(e1)ne1, so that if ε > 0 ε > 0 epsi > 0\varepsilon>0ε>0 is given, there exists n ε N n ε N n_(epsi)in Nn_{\varepsilon} \in NnεN such that | L n ( e 1 ; x ) x | < ε / M L n e 1 ; x x < ε / M |L_(n)(e_(1);x)-x| < epsi//M\left|L_{n}\left(e_{1} ; x\right)-x\right|<\varepsilon / M|Ln(e1;x)x|<ε/M for all n n ε n n ε n >= n_(epsi)n \geqslant n_{\varepsilon}nnε and all x [ a , b ] x [ a , b ] x in[a,b]x \in[a, b]x[a,b], where M = max ( | a | , | b | ) M = max ( | a | , | b | ) M=max(|a|,|b|)M=\max (|a|,|b|)M=max(|a|,|b|). Consequently | L n ( e 2 ; x ) t x | = | t | L n e 2 ; x t x = | t | |L_(n)(e_(2);x)-tx|=|t|\left|L_{n}\left(e_{2} ; x\right)-t x\right|=|t||Ln(e2;x)tx|=|t|. - | L n ( e 1 ; x ) x | < ε L n e 1 ; x x < ε |L_(n)(e_(1);x)-x| < epsi\left|L_{n}\left(e_{1} ; x\right)-x\right|<\varepsilon|Ln(e1;x)x|<ε, for all n n ε n n ε n >= n_(epsi)n \geqslant n_{\varepsilon}nnε and all x x xxx and t t ttt in [ a , b ] [ a , b ] [a,b][a, b][a,b]. In particular for l = x l = x l=xl=xl=x, one obtains | E n ( e x ; x ) x 2 | < ε E n e x ; x x 2 < ε |E_(n)(e_(x);x)-x^(2)| < epsi\left|E_{n}\left(e_{x} ; x\right)-x^{2}\right|<\varepsilon|En(ex;x)x2|<ε, for all n n c n n c n >= n_(c)n \geqslant n_{c}nnc and all x [ a , b ] x [ a , b ] x in[a,b]x \in[a, b]x[a,b], which shows that the sequence ( L n ( e x ; x ) ) n 1 L n e x ; x n 1 (L_(n)(e_(x);x))_(n >= 1)\left(L_{n}\left(e_{x} ; x\right)\right)_{n \geqslant 1}(Ln(ex;x))n1 converges to e ( x ) e ( x ) e(x)e(x)e(x), uniformly for x [ a , b ] x [ a , b ] x in[a,b]x \in[a, b]x[a,b]. The corollary is proved.
If L n = B n L n = B n L_(n)=B_(n)L_{n}=B_{n}Ln=Bn, where B n B n B_(n)B_{n}Bn denotes the Bernstein polynomial operator defined by
B n ( f ; x ) = k = 0 n ( n k ) ( 1 x ) n k x k f ( k n ) , x [ 0 , 1 ] , f C [ 0 , 1 ] , B n ( f ; x ) = k = 0 n ( n k ) ( 1 x ) n k x k f k n , x [ 0 , 1 ] , f C [ 0 , 1 ] , B_(n)(f;x)=sum_(k=0)^(n)((n)/(k))(1-x)^(n-k)x^(k)f((k)/(n)),x in[0,1],f in C[0,1],B_{n}(f ; x)=\sum_{k=0}^{n}\binom{n}{k}(1-x)^{n-k} x^{k} f\left(\frac{k}{n}\right), x \in[0,1], f \in C[0,1],Bn(f;x)=k=0n(nk)(1x)nkxkf(kn),x[0,1],fC[0,1],
then
B n ( e 1 ; x ) = e 1 ( x ) and B n ( e 2 ; x ) = e 2 ( x ) + e 1 ( x ) e 2 ( x ) 2 . B n e 1 ; x = e 1 ( x )  and  B n e 2 ; x = e 2 ( x ) + e 1 ( x ) e 2 ( x ) 2 . B_(n)(e_(1);x)=e_(1)(x)" and "B_(n)(e_(2);x)=e_(2)(x)+(e_(1)(x)-e_(2)(x))/(2).B_{n}\left(e_{1} ; x\right)=e_{1}(x) \text { and } B_{n}\left(e_{2} ; x\right)=e_{2}(x)+\frac{e_{1}(x)-e_{2}(x)}{2} .Bn(e1;x)=e1(x) and Bn(e2;x)=e2(x)+e1(x)e2(x)2.
The delimitation (6) gives
B n ( f ) f K f 1 2 n , B n ( f ) f K f 1 2 n , ||B_(n)(f)-f|| <= K_(f)*(1)/(2*sqrtn),\left\|B_{n}(f)-f\right\| \leqslant K_{f} \cdot \frac{1}{2 \cdot \sqrt{n}},Bn(f)fKf12n,
for all f Lip [ 0 , 1 ] f Lip [ 0 , 1 ] f in Lip[0,1]f \in \operatorname{Lip}[0,1]fLip[0,1].
Applications. 1 1 1^(@)1^{\circ}1. In the Hilbert space R m R m R^(m)\mathbf{R}^{m}Rm consider a compact convex set X X XXX with nonvoid interior. For f C 1 ( X ) f C 1 ( X ) f inC^(1)(X)f \in C^{1}(X)fC1(X) (the space of all real-valued continuously differentiable functions on X X XXX ) and u R m u R m u inR^(m)u \in \mathbf{R}^{m}uRm, denote by f ( u ) f ( u ) grad f(u)\nabla f(u)f(u) the gradient vector of f f fff at the point u u uuu, i.e.
f ( u ) = ( f x 1 ( u ) , , f x m ( u ) ) f ( u ) = f x 1 ( u ) , , f x m ( u ) grad f(u)=((del f)/(delx_(1))(u),dots,(del f)/(delx_(m))(u))\nabla f(u)=\left(\frac{\partial f}{\partial x_{1}}(u), \ldots, \frac{\partial f}{\partial x_{m}}(u)\right)f(u)=(fx1(u),,fxm(u))
lemma 2. If f C 1 ( X ) f C 1 ( X ) f inC^(1)(X)f \in C^{1}(X)fC1(X) then f Lip ( X ) f Lip ( X ) f in Lip(X)f \in \operatorname{Lip}(X)fLip(X) and K f = max u X f ( u ) K f = max u X f ( u ) K_(f)=max_(u in X)||grad f(u)||K_{f}=\max _{u \in X}\|\nabla f(u)\|Kf=maxuXf(u).
Proof. Let x , y X , x y x , y X , x y x,y in X,x!=yx, y \in X, x \neq yx,yX,xy. The mean value theorem implies the existence
of a point u X u X u in Xu \in XuX (which is an internal point of the segment joining x x xxx and y y yyy ) such that
f ( x ) f ( y ) = i = 1 n f x i ( u ) ( x i y i ) = f ( u ) , x y . f ( x ) f ( y ) = i = 1 n f x i ( u ) x i y i = f ( u ) , x y . f(x)-f(y)=sum_(i=1)^(n)(del f)/(delx_(i))(u)*(x_(i)-y_(i))=(:grad f(u),x-y:).f(x)-f(y)=\sum_{i=1}^{n} \frac{\partial f}{\partial x_{i}}(u) \cdot\left(x_{i}-y_{i}\right)=\langle\nabla f(u), x-y\rangle .f(x)f(y)=i=1nfxi(u)(xiyi)=f(u),xy.
Applying now the Schwarz inequality, one obtains
| f ( x ) f ( y ) | = f ( u ) x y ( max u X f ( u ) ) x y . | f ( x ) f ( y ) | = f ( u ) x y max u X f ( u ) x y . |f(x)-f(y)|=||grad f(u)||*||x-y|| <= (max_(u in X)||grad f(u)||)*||x-y||.|f(x)-f(y)|=\|\nabla f(u)\| \cdot\|x-y\| \leqslant\left(\max _{u \in X}\|\nabla f(u)\|\right) \cdot\|x-y\| .|f(x)f(y)|=f(u)xy(maxuXf(u))xy.
corollary 2. If ( L n ) n 1 L n n 1 (L_(n))_(n >= 1)\left(L_{n}\right)_{n \geqslant 1}(Ln)n1 is a sequence of Markov operators acting on C ( X ) C ( X ) C(X)C(X)C(X), where X X XXX is a compact convex subset of R m R m R^(m)\mathbf{R}^{m}Rm with non-void interior, then
(7) | L n ( f ; x ) f ( x ) | max u X f ( u ) L n ( x 2 ; x ) , (7) L n ( f ; x ) f ( x ) max u X f ( u ) L n x 2 ; x , {:(7)|L_(n)(f;x)-f(x)| <= max_(u in X)||grad f(u)||*sqrt(L_(n)(||*-x||^(2);x))",":}\begin{equation*} \left|L_{n}(f ; x)-f(x)\right| \leqslant \max _{u \in X}\|\nabla f(u)\| \cdot \sqrt{L_{n}\left(\|\cdot-x\|^{2} ; x\right)}, \tag{7} \end{equation*}(7)|Ln(f;x)f(x)|maxuXf(u)Ln(x2;x),
for all f C 1 ( X ) f C 1 ( X ) f inC^(1)(X)f \in C^{1}(X)fC1(X).
Proof. By Lemma 2, the inequality (7) is a consequence of the inequality (3) (see also (1) for the definition of β n β n beta_(n)\beta_{n}βn ).
2 2 2^(@)2^{\circ}2. The Bernstein-Lototsky-Schnabl operator. If X X XXX is a compact space, S S SSS is a subspace of C ( X ) C ( X ) C(X)C(X)C(X) such that c 0 S c 0 S c_(0)in Sc_{0} \in Sc0S (remind that c 0 ( x ) = 1 , x X c 0 ( x ) = 1 , x X c_(0)(x)=1,x in Xc_{0}(x)=1, x \in Xc0(x)=1,xX ), L L LLL is a Markov operator on C ( X ) C ( X ) C(X)C(X)C(X) and x x xxx is a point in X X XXX then a Radon probability measure v x v x v_(x)v_{x}vx on X X XXX is called an L ( S ) L ( S ) L(S)L(S)L(S) - representing measure for x x xxx if
L ( f ; x ) = x f d y x L ( f ; x ) = x f d y x L(f;x)=int_(x)fdy_(x)L(f ; x)=\int_{x} f d y_{x}L(f;x)=xfdyx
for all f S f S f in Sf \in SfS.
Suppose from now on that X X XXX is a compact convex subset of a pre-Hilhert space H H HHH and let A ( X ) A ( X ) A(X)A(X)A(X) be the space of all real-valued continuous atine functions defined on X X XXX. Let V = ( V n ) n 1 V = V n n 1 V=(V_(n))_(n >= 1)V=\left(V_{n}\right)_{n \geqslant 1}V=(Vn)n1 be a sequence of Markov operators on C ( X ) C ( X ) C(X)C(X)C(X) and let M ( V ) = { v x , n : n 1 , x X } M ( V ) = v x , n : n 1 , x X M(V)={v_(x,n):n >= 1,x in X}M(V)=\left\{v_{x, n}: n \geqslant 1, x \in X\right\}M(V)={vx,n:n1,xX} be a set of Radon probability measures on X X XXX such that v x , n v x , n v_(x,n)v_{x, n}vx,n is an V n ( A ( X ) ) V n ( A ( X ) ) V_(n)(A(X))V_{n}(A(X))Vn(A(X)) - representing measure for x x xxx, for all x X x X x in Xx \in XxX and n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,. Suppose further that the family M ( V ) M ( V ) M(V)M(V)M(V) is such that the functions E n : X R E n : X R E_(n):X rarrRE_{n}: X \rightarrow \mathbf{R}En:XR defined by E n ( x ) = v x , n ( e ) , x X E n ( x ) = v x , n ( e ) , x X E_(n)(x)=v_(x,n)(e),x in XE_{n}(x)=v_{x, n}(e), x \in XEn(x)=vx,n(e),xX, are continuous for all n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,. Let P = ( p n , j ) n , j 1 P = p n , j n , j 1 P=(p_(n,j))_(n,j >= 1)P=\left(p_{n, j}\right)_{n, j \geqslant 1}P=(pn,j)n,j1 be a lower triangular stochastic matrix i.e. an infinite matrix such that p n , j 0 p n , j 0 p_(n,j) >= 0p_{n, j} \geqslant 0pn,j0 for all n , j 1 n , j 1 n,j >= 1n, j \geqslant 1n,j1,
j = 1 n p n , j = 1 and p n , j = 0 j = 1 n p n , j = 1  and  p n , j = 0 sum_(j=1)^(n)p_(n,j)=1" and "p_(n,j)=0\sum_{j=1}^{n} p_{n, j}=1 \text { and } p_{n, j}=0j=1npn,j=1 and pn,j=0
for all j > n j > n j > nj>nj>n. If ρ = ( ρ n ) n 1 ρ = ρ n n 1 rho=(rho_(n))_(n >= 1)\rho=\left(\rho_{n}\right)_{n \geqslant 1}ρ=(ρn)n1 is a sequence of continuous functions ρ n : X [ 0 , 1 ] ρ n : X [ 0 , 1 ] rho_(n):X rarr[0,1]\rho_{n}: X \rightarrow[0,1]ρn:X[0,1], n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,, define
v x , n , 0 ( V ) = ρ n ( x ) v x , n + ( 1 ρ n ( x ) ) ε x V n , v x , n , 0 ( V ) = ρ n ( x ) v x , n + 1 ρ n ( x ) ε x V n , v_(x,n,0)^((V))=rho_(n)(x)v_(x,n)+(1-rho_(n)(x))epsi_(x)@V_(n),v_{x, n, 0}^{(V)}=\rho_{n}(x) v_{x, n}+\left(1-\rho_{n}(x)\right) \varepsilon_{x} \circ V_{n},vx,n,0(V)=ρn(x)vx,n+(1ρn(x))εxVn,
where ε t ε t epsi_(t)\varepsilon_{t}εt denotes the Dirac measure on X X XXX centered at t X t X t in Xt \in XtX. I.et also π n , P :: X n X π n , P :: X n X pi_(n,P)::X^(n)rarr X\pi_{n, P}: : X^{n} \rightarrow Xπn,P::XnX be defined by
for ( x 1 , x 2 , , x n ) X n x 1 , x 2 , , x n X n (x_(1),x_(2),dots,x_(n))inX^(n)\left(x_{1}, x_{2}, \ldots, x_{n}\right) \in X^{n}(x1,x2,,xn)Xn.
π n , P ( x 1 , x 2 , , x n ) = j = 1 n p n , j x j , π n , P x 1 , x 2 , , x n = j = 1 n p n , j x j , pi_(n,P)(x_(1),x_(2),dots,x_(n))=sum_(j=1)^(n)p_(n,j)*x_(j),\pi_{n, P}\left(x_{1}, x_{2}, \ldots, x_{n}\right)=\sum_{j=1}^{n} p_{n, j} \cdot x_{j},πn,P(x1,x2,,xn)=j=1npn,jxj,
The Bernstein-Lototski-Schnabl operator with respect to M ( V ) , P M ( V ) , P M(V),PM(V), PM(V),P and ρ ρ rho\rhoρ is defined by
B n ( f ; x ) = x n f π n , p d 1 j n v x , j ( V ) B n ( f ; x ) = x n f π n , p d 1 j n v x , j ( V ) B_(n)(f;x)=int_(x^(n))f@pi_(n,p)d⨂_(1 <= j <= n)v_(x,j)^((V))B_{n}(f ; x)=\int_{x^{n}} f \circ \pi_{n, p} d \bigotimes_{1 \leqslant j \leqslant n} v_{x, j}^{(V)}Bn(f;x)=xnfπn,pd1jnvx,j(V)
for all x X x X x in Xx \in XxX and all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X). It follows that B n B n B_(n)B_{n}Bn is a Markov operator on C ( X ) C ( X ) C(X)C(X)C(X) and straightforward calculations (see [5]) show that
(8) B n ( e y ; x ) = j = 1 n p n , j p j ( x ) x , y + j = 1 i n p n , j ( 1 p j ( x ) ) x , y = x , y = e y ( x ) (8) B n e y ; x = j = 1 n p n , j p j ( x ) x , y + j = 1 i n p n , j 1 p j ( x ) x , y = x , y = e y ( x ) {:[(8)B_(n)(e_(y);x)=sum_(j=1)^(n)p_(n,j)*p_(j)(x)(:x","y:)+sum_(j=1)^(in)p_(n,j)],[(1-p_(j)(x))(:x","y:)=(:x","y:)=e_(y)(x)]:}\begin{gather*} B_{n}\left(e_{y} ; x\right)=\sum_{j=1}^{n} p_{n, j} \cdot p_{j}(x)\langle x, y\rangle+\sum_{j=1}^{i n} p_{n, j} \tag{8}\\ \left(1-p_{j}(x)\right)\langle x, y\rangle=\langle x, y\rangle=e_{y}(x) \end{gather*}(8)Bn(ey;x)=j=1npn,jpj(x)x,y+j=1inpn,j(1pj(x))x,y=x,y=ey(x)
for all x , y X x , y X x,y in Xx, y \in Xx,yX and
(9) B n ( c ; x ) = j = 1 n p n , j 2 [ ρ j ( x ) v x , j ( e ) + ( 1 ρ j ( x ) ) V j ( e ; x ) ] + ( 1 j = 1 n p n , j 2 ) e ( x ) , (9) B n ( c ; x ) = j = 1 n p n , j 2 ρ j ( x ) v x , j ( e ) + 1 ρ j ( x ) V j ( e ; x ) + 1 j = 1 n p n , j 2 e ( x ) , {:(9)B_(n)(c;x)=sum_(j=1)^(n)p_(n,j)^(2)[rho_(j)(x)*v_(x,j)(e)+(1-rho_(j)(x))*V_(j)(e;x)]+(1-sum_(j=1)^(n)p_(n,j)^(2))e(x)",":}\begin{equation*} B_{n}(c ; x)=\sum_{j=1}^{n} p_{n, j}^{2}\left[\rho_{j}(x) \cdot v_{x, j}(e)+\left(1-\rho_{j}(x)\right) \cdot V_{j}(e ; x)\right]+\left(1-\sum_{j=1}^{n} p_{n, j}^{2}\right) e(x), \tag{9} \end{equation*}(9)Bn(c;x)=j=1npn,j2[ρj(x)vx,j(e)+(1ρj(x))Vj(e;x)]+(1j=1npn,j2)e(x),
for all x X x X x in Xx \in XxX. Here, the functions e y , c : X R e y , c : X R e_(y),c:X rarrRe_{y}, c: X \rightarrow \mathbf{R}ey,c:XR are defined as above by e y ( x ) == x , y e y ( x ) == x , y e_(y)(x)==(:x,y:)e_{y}(x)= =\langle x, y\rangleey(x)==x,y and c ( x ) = x , x , x X c ( x ) = x , x , x X c(x)=(:x,x:),x in Xc(x)=\langle x, x\rangle, x \in Xc(x)=x,x,xX.
As a consequence of the general convergence theorems one obtains the following result:
THEOREM 5. If
(10) lim n j n p n , j n p j ( x ) ( v x , j ( e ) V j ( c ; x ) ) + j = 1 n p n , j 2 ( V j ( c ; x ) e ( x ) ) = 0 , (10) lim n j n p n , j n p j ( x ) v x , j ( e ) V j ( c ; x ) + j = 1 n p n , j 2 V j ( c ; x ) e ( x ) = 0 , {:(10)lim_(n rarr oo)sum_(j)^(n)p_(n,j)^(n)p_(j)(x)*(v_(x,j)(e)-V_(j)(c;x))+sum_(j=1)^(n)p_(n,j)^(2)(V_(j)(c;x)-e(x))=0",":}\begin{equation*} \lim _{n \rightarrow \infty} \sum_{j}^{n} p_{n, j}^{n} p_{j}(x) \cdot\left(v_{x, j}(e)-V_{j}(c ; x)\right)+\sum_{j=1}^{n} p_{n, j}^{2}\left(V_{j}(c ; x)-e(x)\right)=0, \tag{10} \end{equation*}(10)limnjnpn,jnpj(x)(vx,j(e)Vj(c;x))+j=1npn,j2(Vj(c;x)e(x))=0,
uniformly jor x X x X x in Xx \in XxX, then ( B n ( f ) ) n > 1 B n ( f ) n > 1 (B_(n)(f))_(n > 1)\left(B_{n}(f)\right)_{n>1}(Bn(f))n>1 converges uniformly to f f fff, for all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X). If f I , ip ( X ) f I , ip ( X ) f in I,ip(X)f \in I, \mathrm{ip}(X)fI,ip(X) then furthermore:
(11) :: l n ( f ) f K f j = 1 n p n , j 2 p j E j V j ( e ) + j = 1 n p n , j 2 ( V j ( e ) e ) 1 / 2 , (11) :: l n ( f ) f K f j = 1 n p n , j 2 p j E j V j ( e ) + j = 1 n p n , j 2 V j ( e ) e 1 / 2 , {:(11)::l_(n)(f)-f|| <= K_(f)||sum_(j=1)^(n)p_(n,j)^(2)p_(j)E_(j)-V_(j)(e)+sum_(j=1)^(n)p_(n,j)^(2)(V_(j)(e)-e)||^(1//2)",":}\begin{equation*} :: l_{n}(f)-f\left\|\leqslant K_{f}\right\| \sum_{j=1}^{n} p_{n, j}^{2} p_{j} E_{j}-V_{j}(e)+\sum_{j=1}^{n} p_{n, j}^{2}\left(V_{j}(e)-e\right) \|^{1 / 2}, \tag{11} \end{equation*}(11)::ln(f)fKfj=1npn,j2pjEjVj(e)+j=1npn,j2(Vj(e)e)1/2,
where E j ( x ) = v x , j ( c ) , x X E j ( x ) = v x , j ( c ) , x X E_(j)(x)=v_(x,j)(c),x in XE_{j}(x)=v_{x, j}(c), x \in XEj(x)=vx,j(c),xX.
Proof. The convergence result follows from Theorem 4. Indeed, by (9),
B n ( c ; x ) c ( x ) = j = 1 n p n , j 2 [ p j ( x ) v x , j ( c ) V j ( c ; x ) ] + j = 1 n p n , j 2 ( V j ( e ; x ) e ( x ) ) , B n ( c ; x ) c ( x ) = j = 1 n p n , j 2 p j ( x ) v x , j ( c ) V j ( c ; x ) + j = 1 n p n , j 2 V j ( e ; x ) e ( x ) , B_(n)(c;x)-c(x)=sum_(j=1)^(n)p_(n,j)^(2)[p_(j)(x)v_(x,j)(c)-V_(j)(c;x)]+sum_(j=1)^(n)p_(n,j)^(2)(V_(j)(e;x)-e(x)),B_{n}(c ; x)-c(x)=\sum_{j=1}^{n} p_{n, j}^{2}\left[p_{j}(x) v_{x, j}(c)-V_{j}(c ; x)\right]+\sum_{j=1}^{n} p_{n, j}^{2}\left(V_{j}(e ; x)-e(x)\right),Bn(c;x)c(x)=j=1npn,j2[pj(x)vx,j(c)Vj(c;x)]+j=1npn,j2(Vj(e;x)e(x)),
and the condition (10) of the theorem implies that ( B n ( e ; x ) ) n , 21 B n ( e ; x ) n , 21 (B_(n)(e;x))_(n,21)\left(B_{n}(e ; x)\right)_{n, 21}(Bn(e;x))n,21 converges to e ( x ) e ( x ) e(x)e(x)e(x), uniformly for x X x X x in Xx \in XxX. The equality (8) gives for y = x , B n ( e x ; x ) = x , x = e ( x ) y = x , B n e x ; x = x , x = e ( x ) y=x,B_(n)(e_(x);x)=(:x,x:)=e(x)y=x, B_{n}\left(e_{x} ; x\right)=\langle x, x\rangle =e(x)y=x,Bn(ex;x)=x,x=e(x), for all x X x X x≐Xx \doteq XxX and n = 1 , 2 , n = 1 , 2 , n=1,2,dotsn=1,2, \ldotsn=1,2,. The hypotheses of Theorem 4 are all fulfilled and, consequently, the sequence ( B n ( f ) ) n 1 B n ( f ) n 1 (B_(n)(f))_(n >= 1)\left(B_{n}(f)\right)_{n \geqslant 1}(Bn(f))n1 converges uniformly to f 2 f 2 f_(2)f_{2}f2. for all f C ( X ) f C ( X ) f in C(X)f \in C(X)fC(X).
The equalities (5) (for L n = B n L n = B n L_(n)=B_(n)L_{n}=B_{n}Ln=Bn ), (8) and (9) give:
β n ( x ) = B n ( e ; x ) 2 B n ( e x ; x ) + e ( x ) = = j = 1 n p n , j 2 [ ρ j ( x ) v x , j ( e ) + ( 1 ρ j ( x ) ) V j ( e ; x ) ] + + ( 1 j = 1 n p n , j 2 ) e ( x ) 2 e ( x ) + e ( x ) = = j = 1 n p n , j 2 [ ρ j ( x ) v x , j ( e ) + ( 1 ρ j ( x ) ) V j ( e ; x ) ] j = 1 n p n , j 2 e ( x ) = = j = 1 n p n , j 2 ρ j ( x ) [ E j ( x ) V j ( e ; x ) ] + j = 1 n p n , j 2 [ V j ( e ; x ) e ( x ) ] . β n ( x ) = B n ( e ; x ) 2 B n e x ; x + e ( x ) = = j = 1 n p n , j 2 ρ j ( x ) v x , j ( e ) + 1 ρ j ( x ) V j ( e ; x ) + + 1 j = 1 n p n , j 2 e ( x ) 2 e ( x ) + e ( x ) = = j = 1 n p n , j 2 ρ j ( x ) v x , j ( e ) + 1 ρ j ( x ) V j ( e ; x ) j = 1 n p n , j 2 e ( x ) = = j = 1 n p n , j 2 ρ j ( x ) E j ( x ) V j ( e ; x ) + j = 1 n p n , j 2 V j ( e ; x ) e ( x ) . {:[beta_(n)(x)=B_(n)(e;x)-2B_(n)(e_(x);x)+e(x)=],[=sum_(j=1)^(n)p_(n,j)^(2)[rho_(j)(x)v_(x,j)(e)+(1-rho_(j)(x))*V_(j)(e;x)]+],[+(1-sum_(j=1)^(n)p_(n,j)^(2))e(x)-2e(x)+e(x)=],[=sum_(j=1)^(n)p_(n,j)^(2)[rho_(j)(x)v_(x,j)(e)+(1-rho_(j)(x))V_(j)(e;x)]-sum_(j=1)^(n)p_(n,j)^(2)e(x)=],[=sum_(j=1)^(n)p_(n,j)^(2)rho_(j)(x)[E_(j)(x)-V_(j)(e;x)]+sum_(j=1)^(n)p_(n,j)^(2)[V_(j)(e;x)-e(x)].]:}\begin{aligned} \beta_{n}(x) & =B_{n}(e ; x)-2 B_{n}\left(e_{x} ; x\right)+e(x)= \\ & =\sum_{j=1}^{n} p_{n, j}^{2}\left[\rho_{j}(x) v_{x, j}(e)+\left(1-\rho_{j}(x)\right) \cdot V_{j}(e ; x)\right]+ \\ & +\left(1-\sum_{j=1}^{n} p_{n, j}^{2}\right) e(x)-2 e(x)+e(x)= \\ & =\sum_{j=1}^{n} p_{n, j}^{2}\left[\rho_{j}(x) v_{x, j}(e)+\left(1-\rho_{j}(x)\right) V_{j}(e ; x)\right]-\sum_{j=1}^{n} p_{n, j}^{2} e(x)= \\ & =\sum_{j=1}^{n} p_{n, j}^{2} \rho_{j}(x)\left[E_{j}(x)-V_{j}(e ; x)\right]+\sum_{j=1}^{n} p_{n, j}^{2}\left[V_{j}(e ; x)-e(x)\right] . \end{aligned}βn(x)=Bn(e;x)2Bn(ex;x)+e(x)==j=1npn,j2[ρj(x)vx,j(e)+(1ρj(x))Vj(e;x)]++(1j=1npn,j2)e(x)2e(x)+e(x)==j=1npn,j2[ρj(x)vx,j(e)+(1ρj(x))Vj(e;x)]j=1npn,j2e(x)==j=1npn,j2ρj(x)[Ej(x)Vj(e;x)]+j=1npn,j2[Vj(e;x)e(x)].
It follows that the delimitation (11) is a consequence of the delimitation (3) from Theorem 3.
COROLLARY 3. If v x , j = v x v x , j = v x v_(x,j)=v_(x)v_{x, j}=v_{x}vx,j=vx for j = 1 , 2 , j = 1 , 2 , j=1,2,dotsj=1,2, \ldotsj=1,2, and ρ j ( x ) = 1 , x X , j == 1 , 2 , ρ j ( x ) = 1 , x X , j == 1 , 2 , rho_(j)(x)=1,x in X,j==1,2,dots\rho_{j}(x)=1, x \in X, j= =1,2, \ldotsρj(x)=1,xX,j==1,2, then the condition (10) from Theorem 5 reduces to
(12) lim n j = 1 n p n , j 2 = 0 (12) lim n j = 1 n p n , j 2 = 0 {:(12)lim_(n rarr oo)sum_(j=1)^(n)p_(n,j)^(2)=0:}\begin{equation*} \lim _{n \rightarrow \infty} \sum_{j=1}^{n} p_{n, j}^{2}=0 \tag{12} \end{equation*}(12)limnj=1npn,j2=0
and the delimitation (11) takes the form.
(13) B n ( f ) f K f ( j = 1 n p n , j 2 E e ) 1 / 2 , (13) B n ( f ) f K f j = 1 n p n , j 2 E e 1 / 2 , {:(13)||B_(n)(f)-f|| <= K_(f)(sum_(j=1)^(n)p_(n,j)^(2)||E-e||)^(1//2)",":}\begin{equation*} \left\|B_{n}(f)-f\right\| \leqslant K_{f}\left(\sum_{j=1}^{n} p_{n, j}^{2}\|E-e\|\right)^{1 / 2}, \tag{13} \end{equation*}(13)Bn(f)fKf(j=1npn,j2Ee)1/2,
where E : X R E : X R E:X rarrRE: X \rightarrow \mathbf{R}E:XR is defined by E ( x ) = v x ( c ) , x X E ( x ) = v x ( c ) , x X E(x)=v_(x)(c),x in XE(x)=v_{x}(c), x \in XE(x)=vx(c),xX.
Proof. The first assertion of the Corollary follows from the following delimitation for the expression involved in the condition (10), for ρ j ( x ) = 1 ρ j ( x ) = 1 rho_(j)(x)=1\rho_{j}(x)=1ρj(x)=1 and v x , j ( e ) = v x ( e ) = E ( x ) v x , j ( e ) = v x ( e ) = E ( x ) v_(x,j)(e)=v_(x)(e)=E(x)v_{x, j}(e)=v_{x}(e)=E(x)vx,j(e)=vx(e)=E(x) :
| j = 1 n p n , j 2 [ E ( x ) V j ( e ; x ) ] + j = 1 n p n , j 2 [ V j ( e ; x ) e ( x ) ] | j = 1 n p n , j 2 E V j ( e ) + j = 1 n p n , j 2 V j ( e ) e ( E + 2 V j ( e ) + e ) j = 1 n p n , j 2 ( E + 3 ) j = 1 n p n , j 2 . j = 1 n p n , j 2 E ( x ) V j ( e ; x ) + j = 1 n p n , j 2 V j ( e ; x ) e ( x ) j = 1 n p n , j 2 E V j ( e ) + j = 1 n p n , j 2 V j ( e ) e E + 2 V j ( e ) + e j = 1 n p n , j 2 ( E + 3 ) j = 1 n p n , j 2 . {:[|sum_(j=1)^(n)p_(n,j)^(2)[E(x)-V_(j)(e;x)]+sum_(j=1)^(n)p_(n,j)^(2)[V_(j)(e;x)-e(x)]| <= ],[ <= sum_(j=1)^(n)p_(n,j)^(2)||E-V_(j)(e)||+sum_(j=1)^(n)p_(n,j)^(2)||V_(j)(e)-e|| <= ],[ <= (||E||+2||V_(j)(e)||+||e||)sum_(j=1)^(n)p_(n,j)^(2) <= (||E||+3)sum_(j=1)^(n)p_(n,j)^(2).]:}\begin{gathered} \left|\sum_{j=1}^{n} p_{n, j}^{2}\left[E(x)-V_{j}(e ; x)\right]+\sum_{j=1}^{n} p_{n, j}^{2}\left[V_{j}(e ; x)-e(x)\right]\right| \leqslant \\ \leqslant \sum_{j=1}^{n} p_{n, j}^{2}\left\|E-V_{j}(e)\right\|+\sum_{j=1}^{n} p_{n, j}^{2}\left\|V_{j}(e)-e\right\| \leqslant \\ \leqslant\left(\|E\|+2\left\|V_{j}(e)\right\|+\|e\|\right) \sum_{j=1}^{n} p_{n, j}^{2} \leqslant(\|E\|+3) \sum_{j=1}^{n} p_{n, j}^{2} . \end{gathered}|j=1npn,j2[E(x)Vj(e;x)]+j=1npn,j2[Vj(e;x)e(x)]|j=1npn,j2EVj(e)+j=1npn,j2Vj(e)e(E+2Vj(e)+e)j=1npn,j2(E+3)j=1npn,j2.
The delimitation (13) follows immediately from (11), taking ρ j = 1 ρ j = 1 rho_(j)=1\rho_{j}=1ρj=1 and E j = E E j = E E_(j)=EE_{j}=EEj=E.

REFERENCES

  1. De Vorc, Ronald A., The Approximation of Continuous Functions by Positive Linear Operators, Springer-Verlag, Berlin-Heidelberg-New York, 1972.
  2. Mamedov, R. G., On the order of the approximation of differentiable functions by linear operators (Russian), Doklady SSSR 128, (1959), 674-676.
  3. Shisha, O., Mond, B., The degree of convergence of sequences of linear positive operators, Proc. Nat. Acad. Sci. USA 60, (1968), 1196-1200.
  4. Nishishiraho, T., The degree of convergence of positive linear operators, Tohoku Math. Journal 29 (1977), 81-89.
  5. Nishishiraho, T., Convergence of positive linear approximation processes, Tóhoku Math. Journal 35 (1983), 441-458.

    • University of Cluj-Napoca, Faculty of Malhematics and Physics, 3400 Cluj-Napoca, Romania
1989

Related Posts