Characterization of ε-nearest points in space with asymmetric seminorm

Abstract


In this note we are concerned with the characterization of the elements of \(\varepsilon\)-best approximation (\varepsilon\)-nearest points) in a subspace \(Y\) of space \(X\) with asymmetric seminorm. For this we use functionals in the asymmetric dual \(X^{b}\) defined and studied in some recent papers [1], [3], [5].

Authors

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

Keywords

Asymmetric seminormed spaces; ε-nearest points; characterization.

Paper coordinates

C. Mustăţa, Characterization of ε-nearest points in space with asymmetric seminorm, Rev. Anal. Numer. Theor. Approx. 33 (2004) no. 2, 203-208.

PDF

About this paper

Journal

Revue d’Analyse Numer.Theor. Approx.

Publisher Name

Publishing House of the Romanian Academy

Print ISSN

2501-059X

Online ISSN

2457-6794

google scholar link

[1] Borodin, P. A., The Banach-Mazur theorem for spaces with an asymmetric norm and its applications in convex analysis, Mat. Zametki, 69, no. 3, pp. 193–217, 2001.
[2] De Blasi, F. S. and Myjak, J., On a generalized best approximation problem, J. Approx. Theory, 94, no. 1, pp. 54–72, 1998.
[3] Cobzas, S. and Mustata, C., Extension of bounded linear functionals and best approximation in spaces with asymmetric norm, Rev. Anal. Num´er. Th´eor. Approx., 33, no. 1, pp. 39–50, 2004.
[4] Cobzas, S., Separation of convex sets and best approximation in spaces with asymmetric norm, Quaest. Math., 27, pp. 1–22, 2004.
[5] Garcia-Raffi, L. M., Romaguera S. and Sanchez-Perez, E. A., The dual space of an asymmetric normed linear space, Quaest. Math., 26, no. 1, pp. 83–96, 2003.
[6] Garcia-Raffi, L.M., Romaguera S. and Sanchez-Perez, E. A. , On Hausdorff asymmetric normed linear spaces, Houston J. Math., 29, no. 3, pp. 717–728, 2003 (electronic).
[7] Krein, M. G. and Nudel’man, A. A., The Markov Moment Problem and Extremum Problems, Nauka, Moscow 1973 (in Russian). English translation: American Mathematical Society, Providence, R.I., 1997.
[8] Li, Chong and Ni, Renxing, Derivatives of generalized distance functions and existence of generalized nearest points, J. Approx. Theory, 115, no. 1, pp. 44–55, 2002.
[9] Mabizela, S., Characterization of best approximation in metric linear spaces, Scientiae Mathematicae Japonica, 57, 2, pp. 233–240, 2003.
[10] Mustata, C. , On the best approximation in metric spaces, Mathematica – Revue d’Analyse Numerique et de Theorie de l’Approximation, L’Analyse Numerique et la Theorie de l’Approximation, 4, pp. 45–50, 1975.
[11] Mustata, C. , On the uniqueness of the extension and unique best approximation in the dual of an asymmetric linear space, Rev. Anal. Numer. Theor. Approx., 32, no. 2,
pp. 187–192, 2003.
[12] Ni, Renxing, Existence of generalized nearest points, Taiwanese J. Math., 7, no. 1, pp. 115–128, 2003.
[13] Pantelidis, G., Approximations theorie f¨ur metrich linear R¨aume, Math. Ann., 184, pp. 30–48, 1969.
[14] Rezapour, Sh., ε-pseudo Chebyshev and ε-quasi Chebyshev subspaces of Banach spaces, Technical Report, Azarbaidjan University of Tarbiot Moallem, 2003.
[15] Rezapour, Sh., ε-weakly Chebyshev subspaces of Banach spaces, Analysis in Theory and Applications, 19, no. 2, pp. 130–135, 2003.
[16] Schnatz, K., Nonlinear duality and best approximation in metric linear spaces, J. Approx. Theory, 49, no. 3, pp. 201–21, 1987.
[17] Singer, I., Best Approximation in Normed Linear spaces by Elements of Linear subspaces, Publishing House of the Academy of the Socialist Republic of Romania, Bucharest; Springer-Verlag, New-York-Berlin, 1970.
[18] Singer, I., Caracterisations des ´el´ements de la meilleure approximation dans un espace de Banach quelconque, Acta Sci. Math., 17, pp. 181–189, 1956.

Paper (preprint) in HTML form

2004-Mustata-Characterization of ε-nearest points-Jnaat

CHARACTERIZATION OF ε ε epsi\varepsilonε-NEAREST POINTS IN SPACES WITH ASYMMETRIC SEMINORM*

COSTICĂ MUSTĂŢA ^(†){ }^{\dagger}Dedicated to professor Elena Popoviciu on the occasion of her 80th anniversary.

Abstract

In this note we are concerned with the characterization of the elements of ε ε epsi\varepsilonε-best approximation ( ε ε epsi\varepsilonε-nearest points) in a subspace Y Y YYY of space X X XXX with asymmetric seminorm. For this we use functionals in the asymmetric dual X b X b X^(b)X^{b}Xb defined and studied in some recent papers 1, 3, 5.

MSC 2000. 41A65.
Keywords. Asymmetric seminormed spaces, ε ε epsi\varepsilonε-nearest points, characterization.

1. INTRODUCTION

Let X X XXX be a real linear space. A functional p : X [ 0 , ) p : X [ 0 , ) p:X rarr[0,oo)p: X \rightarrow[0, \infty)p:X[0,) with the properties:
(1) p ( x ) 0 p ( x ) 0 p(x) >= 0p(x) \geq 0p(x)0, for all x X x X x in Xx \in XxX,
(2) p ( t x ) = t p ( x ) p ( t x ) = t p ( x ) p(tx)=tp(x)p(t x)=t p(x)p(tx)=tp(x), for all x X x X x in Xx \in XxX and t 0 t 0 t >= 0t \geq 0t0,
(3) p ( x + y ) p ( x ) + p ( y ) p ( x + y ) p ( x ) + p ( y ) p(x+y) <= p(x)+p(y)p(x+y) \leq p(x)+p(y)p(x+y)p(x)+p(y), for all x , y X x , y X x,y in Xx, y \in Xx,yX,
is called asymmetric seminorm on X X XXX, and the pair ( X , p X , p X,pX, pX,p ) is called a space with asymmetric seminorm.
The functional p ¯ : X [ 0 , ) p ¯ : X [ 0 , ) bar(p):X rarr[0,oo)\bar{p}: X \rightarrow[0, \infty)p¯:X[0,), defined by p ¯ ( x ) = p ( x ) , x X p ¯ ( x ) = p ( x ) , x X bar(p)(x)=p(-x),x in X\bar{p}(x)=p(-x), x \in Xp¯(x)=p(x),xX is another asymmetric seminorm on X X XXX, called the conjugate of p p ppp.
The functional p s : X [ 0 , ) p s : X [ 0 , ) p^(s):X rarr[0,oo)p^{s}: X \rightarrow[0, \infty)ps:X[0,), defined by
p s ( x ) = max { p ( x ) , p ( x ) } , x X p s ( x ) = max { p ( x ) , p ( x ) } , x X p^(s)(x)=max{p(x),p(-x)},x in Xp^{s}(x)=\max \{p(x), p(-x)\}, x \in Xps(x)=max{p(x),p(x)},xX
is a seminorm on X X XXX. If p s p s p^(s)p^{s}ps satisfies the axioms of a norm, then p p ppp is called an asymmetric norm on X X XXX. It follows that p p ppp satisfies the properties (1), (2), (3), and
(4) p ( x ) = 0 p ( x ) = 0 p(x)=0p(x)=0p(x)=0 and p ( x ) = 0 p ( x ) = 0 p(-x)=0p(-x)=0p(x)=0 imply x = 0 x = 0 x=0x=0x=0.
The asymmetric seminorm p p ppp on X X XXX generates a topology τ p τ p tau_(p)\tau_{p}τp on X X XXX, having as a basis of neighborhoods of a point x X x X x in Xx \in XxX the open p p ppp-balls
B p ( x , r ) = { x X : p ( x x ) < r } , r > 0 . B p ( x , r ) = x X : p x x < r , r > 0 . B_(p)^(')(x,r)={x^(')in X:p(x^(')-x) < r},r > 0.B_{p}^{\prime}(x, r)=\left\{x^{\prime} \in X: p\left(x^{\prime}-x\right)<r\right\}, r>0 .Bp(x,r)={xX:p(xx)<r},r>0.
The family of closed p p ppp-balls
B p ( x , r ) = { x X : p ( x x ) r } , r > 0 B p ( x , r ) = x X : p x x r , r > 0 B_(p)(x,r)={x^(')in X:p(x^(')-x) <= r},r > 0B_{p}(x, r)=\left\{x^{\prime} \in X: p\left(x^{\prime}-x\right) \leq r\right\}, r>0Bp(x,r)={xX:p(xx)r},r>0
generates the same topology. This topology τ p τ p tau_(p)\tau_{p}τp could not be Hausdorff (see [5]), and could not be linear (the multiplication by scalars is not continuous in general, see [1].
Let R R R\mathbb{R}R be the set of real numbers and u : R [ 0 , ) , u ( a ) = max { a , 0 } u : R [ 0 , ) , u ( a ) = max { a , 0 } u:Rrarr[0,oo),u(a)=max{a,0}u: \mathbb{R} \rightarrow[0, \infty), u(a)=\max \{a, 0\}u:R[0,),u(a)=max{a,0}, a R a R a inRa \in \mathbb{R}aR. Then the function u u uuu is an asymmetric seminorm on R R R\mathbb{R}R and, for a R a R a inRa \in \mathbb{R}aR, the intervals ( , a + ε ) , ε > 0 ( , a + ε ) , ε > 0 (-oo,a+epsi),epsi > 0(-\infty, a+\varepsilon), \varepsilon>0(,a+ε),ε>0, form a basis of neighborhoods of a R a R a inRa \in \mathbb{R}aR in the topology τ u τ u tau_(u)\tau_{u}τu. The conjugate asymmetric seminorm of u u uuu is u ¯ : R [ 0 , ) u ¯ : R [ 0 , ) bar(u):Rrarr[0,oo)\bar{u}: \mathbb{R} \rightarrow[0, \infty)u¯:R[0,), u ¯ ( a ) = u ( a ) , a R u ¯ ( a ) = u ( a ) , a R bar(u)(a)=u(-a),a inR\bar{u}(a)=u(-a), a \in \mathbb{R}u¯(a)=u(a),aR, and u s ( a ) = max { u ( a ) , u ( a ) } = | a | u s ( a ) = max { u ( a ) , u ( a ) } = | a | u^(s)(a)=max{u(a),u(-a)}=|a|u^{s}(a)=\max \{u(a), u(-a)\}=|a|us(a)=max{u(a),u(a)}=|a| is a norm on R R R\mathbb{R}R. Consequently, u u uuu is an asymmetric norm on R R R\mathbb{R}R.
Let φ : X R φ : X R varphi:X rarrR\varphi: X \rightarrow \mathbb{R}φ:XR be a linear functional. The continuity of φ φ varphi\varphiφ with respect to the topologies τ p τ p tau_(p)\tau_{p}τp and τ u τ u tau_(u)\tau_{u}τu is called ( p , u p , u p,up, up,u )-continuity, and it is equivalent to the upper semicontinuity of φ φ varphi\varphiφ as a functional from ( X , τ p ) X , τ p (X,tau_(p))\left(X, \tau_{p}\right)(X,τp) to ( R , | | ) ( R , | | ) (R,|*|)(\mathbb{R},|\cdot|)(R,||).
The linear functional φ : ( X , τ p ) ( R , u ) φ : X , τ p ( R , u ) varphi:(X,tau_(p))rarr(R,u)\varphi:\left(X, \tau_{p}\right) \rightarrow(\mathbb{R}, u)φ:(X,τp)(R,u) is ( p , u ) ( p , u ) (p,u)(p, u)(p,u)-continuous if and only if it is p p ppp-bounded, i.e. there exists L 0 L 0 L >= 0L \geq 0L0 such that
φ ( x ) L p ( x ) , for all x X . φ ( x ) L p ( x ) ,  for all  x X . varphi(x) <= Lp(x),quad" for all "x in X.\varphi(x) \leq L p(x), \quad \text { for all } x \in X .φ(x)Lp(x), for all xX.
The set of all ( p , u p , u p,up, up,u )-continuous functionals is denoted by X p b X p b X_(p)^(b)X_{p}^{b}Xpb. With respect to pointwise addition and multiplication by real scalars, the set X p b X p b X_(p)^(b)X_{p}^{b}Xpb is a cone, i.e. λ 0 λ 0 lambda >= 0\lambda \geq 0λ0 and φ , ψ X p b φ , ψ X p b varphi,psi inX_(p)^(b)\varphi, \psi \in X_{p}^{b}φ,ψXpb imply φ + ψ X b φ + ψ X b varphi+psi inX^(b)\varphi+\psi \in X^{b}φ+ψXb and λ φ X p b λ φ X p b lambda varphi inX_(p)^(b)\lambda \varphi \in X_{p}^{b}λφXpb.
The functional . ∣: X p b [ 0 , ) . ∣: X p b [ 0 , ) ||.∣:X_(p)^(b)rarr[0,oo)\| . \mid: X_{p}^{b} \rightarrow[0, \infty).∣:Xpb[0,) defined by
φ | p = sup { φ ( x ) : x X , p ( x ) 1 } , φ X p b φ p = sup { φ ( x ) : x X , p ( x ) 1 } , φ X p b || varphi|_(p)=s u p{varphi(x):x in X,p(x) <= 1},quad varphi inX_(p)^(b)\|\left.\varphi\right|_{p}=\sup \{\varphi(x): x \in X, p(x) \leq 1\}, \quad \varphi \in X_{p}^{b}φ|p=sup{φ(x):xX,p(x)1},φXpb
satisfies the properties of an asymmetric seminorm, and the pair ( X p b , | p X p b , p X_(p)^(b),||*|_(p)X_{p}^{b}, \|\left.\cdot\right|_{p}Xpb,|p ) is called the asymmetric dual of the asymmetric seminormed space ( X , p X , p X,pX, pX,p ) (see [5]). Some properties of this dual are presented in [1], [3, [5]. If there is no danger of confusion we shall use the notation X b X b X^(b)X^{b}Xb and φ φ ||varphi∣\| \varphi \midφ instead of X p b X p b X_(p)^(b)X_{p}^{b}Xpb and φ | p φ p || varphi|_(p)\|\left.\varphi\right|_{p}φ|p, respectively.
Let ( X , p X , p X,pX, pX,p ) be an asymmetric seminormed space and Y Y YYY a subspace of X X XXX. Let Y b Y b Y^(b)Y^{b}Yb be the asymmetric dual of ( Y , p ) ( Y , p ) (Y,p)(Y, p)(Y,p).
The following result is the analog of a well known extension result for linear functionals in normed spaces.
Theorem 1 (Hahn-Banach). Let ( Y , p Y , p Y,pY, pY,p ) be a subspace of asymmetric seminormed space ( X , p X , p X,pX, pX,p ). Then for every φ 0 Y b φ 0 Y b varphi_(0)inY^(b)\varphi_{0} \in Y^{b}φ0Yb there exists φ X b φ X b varphi inX^(b)\varphi \in X^{b}φXb such that
φ | Y = φ 0 φ = φ 0 . φ Y = φ 0 φ = φ 0 . {:[ varphi|_(Y)=varphi_(0)],[||varphi∣=||varphi_(0)∣.]:}\begin{aligned} \left.\varphi\right|_{Y} & =\varphi_{0} \\ \| \varphi \mid & =\| \varphi_{0} \mid . \end{aligned}φ|Y=φ0φ=φ0.
Proof. We consider the functional q : X [ 0 , ) , q ( x ) = φ 0 p ( x ) q : X [ 0 , ) , q ( x ) = φ 0 p ( x ) q:X rarr[0,oo),q(x)=||varphi_(0)∣*p(x)q: X \rightarrow[0, \infty), q(x)=\| \varphi_{0} \mid \cdot p(x)q:X[0,),q(x)=φ0p(x), x X x X x in Xx \in XxX. Obviously q q qqq is subadditive and positive homogeneous and for every y Y y Y y in Yy \in YyY we have
φ 0 ( y ) φ 0 p ( y ) = q ( y ) φ 0 ( y ) φ 0 p ( y ) = q ( y ) varphi_(0)(y) <= ||varphi_(0)∣*p(y)=q(y)\varphi_{0}(y) \leq \| \varphi_{0} \mid \cdot p(y)=q(y)φ0(y)φ0p(y)=q(y)
i.e. φ 0 φ 0 varphi_(0)\varphi_{0}φ0 is majorized by q q qqq on Y Y YYY.
By Hahn-Banach extension theorem it results that there exists the linear functional φ : X R φ : X R varphi:X rarrR\varphi: X \rightarrow \mathbb{R}φ:XR with properties:
φ | Y = φ 0 and φ ( x ) φ 0 p ( x ) , for every x X . φ Y = φ 0  and  φ ( x ) φ 0 p ( x ) ,  for every  x X . {:[ varphi|_(Y)=varphi_(0)quad" and "],[varphi(x) <= ||varphi_(0)∣*p(x)","" for every "x in X.]:}\begin{aligned} \left.\varphi\right|_{Y} & =\varphi_{0} \quad \text { and } \\ \varphi(x) & \leq \| \varphi_{0} \mid \cdot p(x), \text { for every } x \in X . \end{aligned}φ|Y=φ0 and φ(x)φ0p(x), for every xX.
It follows φ | φ 0 | φ φ 0 ||varphi| <= ||varphi_(0)|:}\left\|\varphi\left|\leq \| \varphi_{0}\right|\right.φ|φ0|, and, because
φ = sup { φ ( x ) : x X , p ( x ) 1 } sup { φ ( y ) : y Y , p ( y ) 1 } = sup { φ 0 ( y ) : y Y , p ( y ) 1 } = φ 0 φ = sup { φ ( x ) : x X , p ( x ) 1 } sup { φ ( y ) : y Y , p ( y ) 1 } = sup φ 0 ( y ) : y Y , p ( y ) 1 = φ 0 {:[||varphi∣=s u p{varphi(x):x in X","p(x) <= 1}],[ >= s u p{varphi(y):y in Y","p(y) <= 1}],[=s u p{varphi_(0)(y):y in Y,p(y) <= 1}],[=||varphi_(0)∣]:}\begin{aligned} \| \varphi \mid & =\sup \{\varphi(x): x \in X, p(x) \leq 1\} \\ & \geq \sup \{\varphi(y): y \in Y, p(y) \leq 1\} \\ & =\sup \left\{\varphi_{0}(y): y \in Y, p(y) \leq 1\right\} \\ & =\| \varphi_{0} \mid \end{aligned}φ=sup{φ(x):xX,p(x)1}sup{φ(y):yY,p(y)1}=sup{φ0(y):yY,p(y)1}=φ0
we have φ | = φ 0 | φ = φ 0 ||varphi|=||varphi_(0)|:}\left\|\varphi\left|=\| \varphi_{0}\right|\right.φ|=φ0|.

2. THE ε ε epsi\varepsilonε-BEST APPROXIMATION IN ( X , p X , p X,pX, pX,p )

Let Y Y YYY be a nonvoid subset of the asymmetric seminormed space ( X , p X , p X,pX, pX,p ).
The problem of best approximation of the element x X x X x in Xx \in XxX by elements in Y Y YYY is: find an element y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y such that
(1) d p ( x , Y ) := inf { p ( y x ) : y Y } = p ( y 0 x ) (1) d p ( x , Y ) := inf { p ( y x ) : y Y } = p y 0 x {:(1)d_(p)(x","Y):=i n f{p(y-x):y in Y}=p(y_(0)-x):}\begin{equation*} d_{p}(x, Y):=\inf \{p(y-x): y \in Y\}=p\left(y_{0}-x\right) \tag{1} \end{equation*}(1)dp(x,Y):=inf{p(yx):yY}=p(y0x)
Let ε > 0 ε > 0 epsi > 0\varepsilon>0ε>0. The problem of ε ε epsi\varepsilonε-best approximation of x X x X x in Xx \in XxX by elements in Y Y YYY is: find y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y such that
(2) p ( y 0 x ) d p ( x , Y ) + ε (2) p y 0 x d p ( x , Y ) + ε {:(2)p(y_(0)-x) <= d_(p)(x","Y)+epsi:}\begin{equation*} p\left(y_{0}-x\right) \leq d_{p}(x, Y)+\varepsilon \tag{2} \end{equation*}(2)p(y0x)dp(x,Y)+ε
Obviously, the problem of ε ε epsi\varepsilonε-best approximation always admits a solution, because for every number n N n N n inNn \in \mathbb{N}nN there exists y n Y y n Y y_(n)in Yy_{n} \in YynY such that p ( y n x ) d p ( x , Y ) + 1 n p y n x d p ( x , Y ) + 1 n p(y_(n)-x) <= d_(p)(x,Y)+(1)/(n)p\left(y_{n}-x\right) \leq d_{p}(x, Y)+\frac{1}{n}p(ynx)dp(x,Y)+1n, so that p ( y n x ) d p ( x , Y ) + ε p y n x d p ( x , Y ) + ε p(y_(n)-x) <= d_(p)(x,Y)+epsip\left(y_{n}-x\right) \leq d_{p}(x, Y)+\varepsilonp(ynx)dp(x,Y)+ε, for n > [ 1 ε ] + 1 n > 1 ε + 1 n > [(1)/(epsi)]+1n>\left[\frac{1}{\varepsilon}\right]+1n>[1ε]+1.
In the following we denote by
(3) P Y , ε ( x ) = { y Y : p ( y x ) d p ( x , Y ) + ε } , x X (3) P Y , ε ( x ) = y Y : p ( y x ) d p ( x , Y ) + ε , x X {:(3)P_(Y,epsi)(x)={y in Y:p(y-x) <= d_(p)(x,Y)+epsi}","x in X:}\begin{equation*} P_{Y, \varepsilon}(x)=\left\{y \in Y: p(y-x) \leq d_{p}(x, Y)+\varepsilon\right\}, x \in X \tag{3} \end{equation*}(3)PY,ε(x)={yY:p(yx)dp(x,Y)+ε},xX
the nonvoid set of the elements of ε ε epsi\varepsilonε-best approximation for x X x X x in Xx \in XxX in Y Y YYY.
The paper [3] contains characterizations, in terms of functionals in X b X b X^(b)X^{b}Xb vanishing on Y Y YYY, of the elements of best approximation of x X x X x in Xx \in XxX by elements in a subspace Y Y YYY of X X XXX. Let us observe firstly that, one can consider also the problem of ε ε epsi\varepsilonε-best approximation by using the conjugate p ¯ p ¯ bar(p)\bar{p}p¯ of p p ppp. In this case, for
(4) d p ¯ ( x , Y ) = inf { p ( x y ) : y Y } (4) d p ¯ ( x , Y ) = inf { p ( x y ) : y Y } {:(4)d_( bar(p))(x","Y)=i n f{p(x-y):y in Y}:}\begin{equation*} d_{\bar{p}}(x, Y)=\inf \{p(x-y): y \in Y\} \tag{4} \end{equation*}(4)dp¯(x,Y)=inf{p(xy):yY}
one looks for y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y such that
(5) p ( x y 0 ) d p ¯ ( x , Y ) + ε . (5) p x y 0 d p ¯ ( x , Y ) + ε . {:(5)p(x-y_(0)) <= d_( bar(p))(x","Y)+epsi.:}\begin{equation*} p\left(x-y_{0}\right) \leq d_{\bar{p}}(x, Y)+\varepsilon . \tag{5} \end{equation*}(5)p(xy0)dp¯(x,Y)+ε.
Let us denote by
(6) P ¯ Y , ε ( x ) = { y Y : p ¯ ( y x ) = p ( x y ) d p ¯ ( x , Y ) + ε } (6) P ¯ Y , ε ( x ) = y Y : p ¯ ( y x ) = p ( x y ) d p ¯ ( x , Y ) + ε {:(6) bar(P)_(Y,epsi)(x)={y in Y:( bar(p))(y-x)=p(x-y) <= d_( bar(p))(x,Y)+epsi}:}\begin{equation*} \bar{P}_{Y, \varepsilon}(x)=\left\{y \in Y: \bar{p}(y-x)=p(x-y) \leq d_{\bar{p}}(x, Y)+\varepsilon\right\} \tag{6} \end{equation*}(6)P¯Y,ε(x)={yY:p¯(yx)=p(xy)dp¯(x,Y)+ε}
the set of ε ε epsi\varepsilonε-best approximation of x X x X x in Xx \in XxX with respect to the conjugate asymmetric seminorm p ¯ p ¯ bar(p)\bar{p}p¯.
In the following we obtain characterizations of elements of ε ε epsi\varepsilonε-best approximation of x X x X x in Xx \in XxX by elements of a subspace Y Y YYY, with respect to the asymmetric seminorms p p ppp and p ¯ p ¯ bar(p)\bar{p}p¯.
Results of this type, for elements of best approximations in a normed space X X XXX, using the elements of dual X X X^(**)X^{*}X are obtained in [17] (see also [3, 9, [1], [13], [16], [18]).
Concerning the characterizations of elements of ε ε epsi\varepsilonε-best approximation in normed space, see papers [14, [15].
Theorem 2. Let ( X , p X , p X,pX, pX,p ) be an asymmetric seminormed space, Y Y YYY a subspace of X X XXX and x 0 X Y x 0 X Y x_(0)in X\\Yx_{0} \in X \backslash Yx0XY, such that d = d p ( x 0 , Y ) > 0 d = d p x 0 , Y > 0 d=d_(p)(x_(0),Y) > 0d=d_{p}\left(x_{0}, Y\right)>0d=dp(x0,Y)>0 and d ¯ = d p ¯ ( x 0 , Y ) > 0 d ¯ = d p ¯ x 0 , Y > 0 bar(d)=d_( bar(p))(x_(0),Y) > 0\bar{d}=d_{\bar{p}}\left(x_{0}, Y\right)>0d¯=dp¯(x0,Y)>0. Then
(a) An element y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y is in P Y , ε ( x 0 ) P Y , ε x 0 P_(Y,epsi)(x_(0))P_{Y, \varepsilon}\left(x_{0}\right)PY,ε(x0) if and only if there exists φ X p b φ X p b varphi inX_(p)^(b)\varphi \in X_{p}^{b}φXpb with the properties:
(i) φ ( y ) = 0 φ ( y ) = 0 varphi(y)=0\varphi(y)=0φ(y)=0, for all y Y y Y y in Yy \in YyY,
(iii) φ | p = 1 φ p = 1 || varphi|_(p)=1\|\left.\varphi\right|_{p}=1φ|p=1,
(iii) φ ( x 0 ) p ( y 0 x 0 ) ε φ x 0 p y 0 x 0 ε varphi(-x_(0)) >= p(y_(0)-x_(0))-epsi\varphi\left(-x_{0}\right) \geq p\left(y_{0}-x_{0}\right)-\varepsilonφ(x0)p(y0x0)ε.
(b) An element y 0 y 0 y_(0)y_{0}y0 is in P ¯ Y , ε ( x 0 ) P ¯ Y , ε x 0 bar(P)_(Y,epsi)(x_(0))\bar{P}_{Y, \varepsilon}\left(x_{0}\right)P¯Y,ε(x0) if and only if there exists ψ X p b ψ X p b psi inX_(p)^(b)\psi \in X_{p}^{b}ψXpb with the properties:
(j) ψ ( y ) = 0 ψ ( y ) = 0 psi(y)=0\psi(y)=0ψ(y)=0, for all y Y y Y y in Yy \in YyY,
(jj) ψ | p = 1 ψ p = 1 || psi|_(p)=1\|\left.\psi\right|_{p}=1ψ|p=1,
(jjj) ψ ( x 0 ) p ( x 0 y 0 ) ε ψ x 0 p x 0 y 0 ε psi(x_(0)) >= p(x_(0)-y_(0))-epsi\psi\left(x_{0}\right) \geq p\left(x_{0}-y_{0}\right)-\varepsilonψ(x0)p(x0y0)ε.
Proof. Let x 0 X Y x 0 X Y x_(0)in X\\Yx_{0} \in X \backslash Yx0XY and Z = Y + x 0 Z = Y + x 0 Z=Y+(:x_(0):)Z=Y+\left\langle x_{0}\right\rangleZ=Y+x0 be the direct sum of Y Y YYY with the space generated by x 0 x 0 x_(0)x_{0}x0. Consider the functional φ 0 : Z R φ 0 : Z R varphi_(0):Z rarrR\varphi_{0}: Z \rightarrow \mathbb{R}φ0:ZR defined by
φ 0 ( z ) = φ ( y + λ x 0 ) = λ , φ 0 ( z ) = φ y + λ x 0 = λ , varphi_(0)(z)=varphi(y+lambdax_(0))=-lambda,\varphi_{0}(z)=\varphi\left(y+\lambda x_{0}\right)=-\lambda,φ0(z)=φ(y+λx0)=λ,
where z Z z Z z in Zz \in ZzZ, and z z zzz is uniquely represented in the form z = y + λ x 0 z = y + λ x 0 z=y+lambdax_(0)z=y+\lambda x_{0}z=y+λx0.
The functional φ 0 φ 0 varphi_(0)\varphi_{0}φ0 is linear on Z Z ZZZ.
Observe that φ 0 Y = 0 φ 0 Y = 0 varphi_(0)∣Y=0\varphi_{0} \mid Y=0φ0Y=0, and for every λ > 0 λ > 0 lambda > 0\lambda>0λ>0 we have
p ( y λ x 0 ) = λ p ( 1 λ y x 0 ) λ d = d φ 0 ( y λ x 0 ) . p y λ x 0 = λ p 1 λ y x 0 λ d = d φ 0 y λ x 0 . p(y-lambdax_(0))=lambda p((1)/(lambda)y-x_(0)) >= lambda d=d*varphi_(0)(y-lambdax_(0)).p\left(y-\lambda x_{0}\right)=\lambda p\left(\frac{1}{\lambda} y-x_{0}\right) \geq \lambda d=d \cdot \varphi_{0}\left(y-\lambda x_{0}\right) .p(yλx0)=λp(1λyx0)λd=dφ0(yλx0).
It follows that
φ 0 ( y λ x 0 ) 1 d p ( y λ x 0 ) , φ 0 y λ x 0 1 d p y λ x 0 , varphi_(0)(y-lambdax_(0)) <= (1)/(d)*p(y-lambdax_(0)),\varphi_{0}\left(y-\lambda x_{0}\right) \leq \frac{1}{d} \cdot p\left(y-\lambda x_{0}\right),φ0(yλx0)1dp(yλx0),
for every λ > 0 λ > 0 lambda > 0\lambda>0λ>0.
Because the last inequality is also valid if φ 0 ( y t x 0 ) = t 0 φ 0 y t x 0 = t 0 varphi_(0)(y-tx_(0))=t <= 0\varphi_{0}\left(y-t x_{0}\right)=t \leq 0φ0(ytx0)=t0, it follows
φ 0 | p 1 d , and consequently φ 0 Z p b . φ 0 p 1 d ,  and consequently  φ 0 Z p b . ||varphi_(0)|_(p) <= (1)/(d)," and consequently "varphi_(0)inZ_(p)^(b).\|\left.\varphi_{0}\right|_{p} \leq \frac{1}{d}, \text { and consequently } \varphi_{0} \in Z_{p}^{b} .φ0|p1d, and consequently φ0Zpb.
Now, let ( y n ) n 1 y n n 1 (y_(n))_(n >= 1)\left(y_{n}\right)_{n \geq 1}(yn)n1 be a sequence in Y Y YYY such that p ( y n x 0 ) d p y n x 0 d p(y_(n)-x_(0))rarr dp\left(y_{n}-x_{0}\right) \rightarrow dp(ynx0)d, for n n n rarr oon \rightarrow \inftyn, and such that p ( y n x 0 ) > 0 p y n x 0 > 0 p(y_(n)-x_(0)) > 0p\left(y_{n}-x_{0}\right)>0p(ynx0)>0 for every n N n N n inNn \in \mathbb{N}nN. Then
φ 0 | p φ 0 ( y n x 0 p ( y n x 0 ) ) = 1 p ( y n x 0 ) 1 d , φ 0 p φ 0 y n x 0 p y n x 0 = 1 p y n x 0 1 d , ||varphi_(0)|_(p) >= varphi_(0)((y_(n)-x_(0))/(p(y_(n)-x_(0))))=(1)/(p(y_(n)-x_(0)))rarr(1)/(d),\|\left.\varphi_{0}\right|_{p} \geq \varphi_{0}\left(\frac{y_{n}-x_{0}}{p\left(y_{n}-x_{0}\right)}\right)=\frac{1}{p\left(y_{n}-x_{0}\right)} \rightarrow \frac{1}{d},φ0|pφ0(ynx0p(ynx0))=1p(ynx0)1d,
and, consequently, φ 0 | p = 1 d φ 0 p = 1 d ||varphi_(0)|_(p)=(1)/(d)\|\left.\varphi_{0}\right|_{p}=\frac{1}{d}φ0|p=1d.
By Theorem 1, there exists φ 1 X b φ 1 X b varphi_(1)inX^(b)\varphi_{1} \in X^{b}φ1Xb such that
φ 1 | z = φ 0 , φ 1 | = φ 0 | p = 1 d φ 1 z = φ 0 , φ 1 = φ 0 p = 1 d varphi_(1)|z=varphi_(0),quad||varphi_(1)|=||varphi_(0)|_(p)=(1)/(d):}\varphi_{1}\left|z=\varphi_{0}, \quad\left\|\varphi_{1}\left|=\| \varphi_{0}\right|_{p}=\frac{1}{d}\right.\right.φ1|z=φ0,φ1|=φ0|p=1d
Then, the functional φ = d φ 1 φ = d φ 1 varphi=d*varphi_(1)\varphi=d \cdot \varphi_{1}φ=dφ1 satisfies the properties: φ X p b , φ Y = d φ 1 | Y = 0 φ X p b , φ Y = d φ 1 Y = 0 varphi inX_(p)^(b),varphi∣Y=d*varphi_(1)|_(Y)=0\varphi \in X_{p}^{b}, \varphi \mid Y= \left.d \cdot \varphi_{1}\right|_{Y}=0φXpb,φY=dφ1|Y=0,
φ ( x 0 ) = φ ( y 0 ) + φ ( x 0 ) = φ ( y 0 x 0 ) p ( y 0 x 0 ) p ( y 0 x 0 ) ε . φ x 0 = φ y 0 + φ x 0 = φ y 0 x 0 p y 0 x 0 p y 0 x 0 ε . {:[varphi(-x_(0))=varphi(y_(0))+varphi(-x_(0))],[=varphi(y_(0)-x_(0))],[ >= p(y_(0)-x_(0))],[ >= p(y_(0)-x_(0))-epsi.]:}\begin{aligned} \varphi\left(-x_{0}\right) & =\varphi\left(y_{0}\right)+\varphi\left(-x_{0}\right) \\ & =\varphi\left(y_{0}-x_{0}\right) \\ & \geq p\left(y_{0}-x_{0}\right) \\ & \geq p\left(y_{0}-x_{0}\right)-\varepsilon . \end{aligned}φ(x0)=φ(y0)+φ(x0)=φ(y0x0)p(y0x0)p(y0x0)ε.
Conversely, if y 0 Y y 0 Y y_(0)in Yy_{0} \in Yy0Y and there exists φ X p b φ X p b varphi inX_(p)^(b)\varphi \in X_{p}^{b}φXpb with the properties (a) (i)-(iii), then for every y Y y Y y in Yy \in YyY we have
p ( y 0 x 0 ) φ ( x 0 ) + ε = φ ( y x 0 ) + ε φ | p p ( y x 0 ) + ε p ( y x 0 ) + ε p y 0 x 0 φ x 0 + ε = φ y x 0 + ε φ p p y x 0 + ε p y x 0 + ε {:[p(y_(0)-x_(0)) <= varphi(-x_(0))+epsi],[=varphi(y-x_(0))+epsi],[ <= || varphi|_(p)*p(y-x_(0))+epsi],[ <= p(y-x_(0))+epsi]:}\begin{aligned} p\left(y_{0}-x_{0}\right) & \leq \varphi\left(-x_{0}\right)+\varepsilon \\ & =\varphi\left(y-x_{0}\right)+\varepsilon \\ & \leq \|\left.\varphi\right|_{p} \cdot p\left(y-x_{0}\right)+\varepsilon \\ & \leq p\left(y-x_{0}\right)+\varepsilon \end{aligned}p(y0x0)φ(x0)+ε=φ(yx0)+εφ|pp(yx0)+εp(yx0)+ε
Taking the infimum with respect to y Y y Y y in Yy \in YyY, one finds
p ( y 0 x 0 ) d p ( x 0 , Y ) + ε ; p y 0 x 0 d p x 0 , Y + ε ; p(y_(0)-x_(0)) <= d_(p)(x_(0),Y)+epsi;p\left(y_{0}-x_{0}\right) \leq d_{p}\left(x_{0}, Y\right)+\varepsilon ;p(y0x0)dp(x0,Y)+ε;
so that y 0 P Y , ε ( x 0 ) y 0 P Y , ε x 0 y_(0)inP_(Y,epsi)(x_(0))y_{0} \in P_{Y, \varepsilon}\left(x_{0}\right)y0PY,ε(x0).
Similarly, defining ψ 0 : Z = Y + x 0 R ψ 0 : Z = Y + x 0 R psi_(0):Z=Y+(:x_(0):)rarrR\psi_{0}: Z=Y+\left\langle x_{0}\right\rangle \rightarrow \mathbb{R}ψ0:Z=Y+x0R by ψ ( z ) = ψ 0 ( y + λ x 0 ) = λ , y Y ψ ( z ) = ψ 0 y + λ x 0 = λ , y Y psi(z)=psi_(0)(y+lambdax_(0))=lambda,y in Y\psi(z)=\psi_{0}\left(y+\lambda x_{0}\right)= \lambda, y \in Yψ(z)=ψ0(y+λx0)=λ,yY and λ R λ R lambda inR\lambda \in \mathbb{R}λR, and proceeding in the same way, one obtains the claim (b) of the theorem.
Theorem 2 has the following consequence:
Corollary 3. In the hypothesis of Theorem 2 we have:
(a') M P Y , ε ( x 0 ) M P Y , ε x 0 M subP_(Y,epsi)(x_(0))M \subset P_{Y, \varepsilon}\left(x_{0}\right)MPY,ε(x0), if and only if there exists φ X b φ X b varphi inX^(b)\varphi \in X^{b}φXb verifying (a) (i)-(ii) and the condition:
φ ( x 0 ) p ( u x 0 ) ε , for all u M ; φ x 0 p u x 0 ε ,  for all  u M ; varphi(-x_(0)) >= p(u-x_(0))-epsi,quad" for all "quad u in M;\varphi\left(-x_{0}\right) \geq p\left(u-x_{0}\right)-\varepsilon, \quad \text { for all } \quad u \in M ;φ(x0)p(ux0)ε, for all uM;
(b') M P Y , ε ( x 0 ) M P Y , ε x 0 M subP_(Y,epsi)(x_(0))M \subset P_{Y, \varepsilon}\left(x_{0}\right)MPY,ε(x0) if and only if there exists ψ X b ψ X b psi inX^(b)\psi \in X^{b}ψXb with properties (b) ( j ) ( j j ) ( j ) ( j j ) (j)-(jj)(j)-(j j)(j)(jj), and verifying the condition:
ψ ( x 0 ) p ( x 0 u ) ε , for all u M . ψ x 0 p x 0 u ε ,  for all  u M psi(x_(0)) >= p(x_(0)-u)-epsi,quad" for all "quad u in M". "\psi\left(x_{0}\right) \geq p\left(x_{0}-u\right)-\varepsilon, \quad \text { for all } \quad u \in M \text {. }ψ(x0)p(x0u)ε, for all uM

REFERENCES

[1] Borodin, P. A., The Banach-Mazur theorem for spaces with an asymmetric norm and its applications in convex analysis, Mat. Zametki, 69, no. 3, pp. 193-217, 2001.
[2] De Blasi, F. S. and Myjak, J., On a generalized best approximation problem, J. Approx. Theory, 94, no. 1, pp. 54-72, 1998.
[3] Cobzas, S. and Mustata, C., Extension of bounded linear functionals and best approximation in spaces with asymmetric norm, Rev. Anal. Numér. Théor. Approx., 33, no. 1, pp. 39-50, 2004. 줓
[4] Cobzas, S., Separation of convex sets and best approximation in spaces with asymmetric norm, Quaest. Math., 27, pp. 1-22, 2004.
[5] Garcia-Raffi, L. M., Romaguera S. and Sánchez-Pérez, E. A., The dual space of an asymmetric normed linear space, Quaest. Math., 26, no. 1, pp. 83-96, 2003.
[6] Garcia-Raffi, L.M., Romaguera S. and Sánchez-Pérez, E. A., On Hausdorff asymmetric normed linear spaces, Houston J. Math., 29, no. 3, pp. 717-728, 2003 (electronic).
[7] Krein, M. G. and Nudel'man, A. A., The Markov Moment Problem and Extremum Problems, Nauka, Moscow 1973 (in Russian). English translation: American Mathematical Society, Providence, R.I., 1997.
[8] Li, Chong and Ni, Renxing, Derivatives of generalized distance functions and existence of generalized nearest points, J. Approx. Theory, 115, no. 1, pp. 44-55, 2002.
[9] Mabizela, S., Characterization of best approximation in metric linear spaces, Scientiae Mathematicae Japonica, 57, 2, pp. 233-240, 2003.
[10] Mustăţa, C., On the best approximation in metric spaces, Mathematica - Revue d'Analyse Numérique et de Théorie de l'Approximation, L'Analyse Numérique et la Théorie de l'Approximation, 4, pp. 45-50, 1975. 지
[11] Mustăta, C., On the uniqueness of the extension and unique best approximation in the dual of an asymmetric linear space, Rev. Anal. Numér. Théor. Approx., 32, no. 2, pp. 187-192, 2003. 준
[12] Ni, Renxing, Existence of generalized nearest points, Taiwanese J. Math., 7, no. 1, pp. 115-128, 2003.
[13] Pantelidis, G., Approximations theorie für metrich linear Räume, Math. Ann., 184, pp. 30-48, 1969.
[14] Rezapour, Sh., ε ε epsi\varepsilonε-pseudo Chebyshev and ε ε epsi\varepsilonε-quasi Chebyshev subspaces of Banach spaces, Technical Report, Azarbaidjan University of Tarbiot Moallem, 2003.
[15] Rezapour, Sh., ε ε epsi\varepsilonε-weakly Chebyshev subspaces of Banach spaces, Analysis in Theory and Applications, 19, no. 2, pp. 130-135, 2003.
[16] Schnatz, K., Nonlinear duality and best approximation in metric linear spaces, J. Approx. Theory, 49, no. 3, pp. 201-21, 1987.
[17] Singer, I., Best Approximation in Normed Linear spaces by Elements of Linear subspaces, Publishing House of the Academy of the Socialist Republic of Romania, Bucharest; Springer-Verlag, New-York-Berlin, 1970.
[18] Singer, I., Caracterisations des éléments de la meilleure approximation dans un espace de Banach quelconque, Acta Sci. Math., 17, pp. 181-189, 1956.
Received by the editors: June 11, 2003.

  1. ^(**){ }^{*} This work has been supported by the Romanian Academy under Grant GAR 13/2004.
    ^(†){ }^{\dagger} "T. Popoviciu" Institute of Numerical Analysis, P.O. Box 68-1, Cluj-Napoca, Romania, e-mail: cmustata@ictp.acad.ro.
2004

Related Posts