| Sbornik: Mathematics |
|
|
|
|
|
This article is cited in 6 scientific papers (total in 6 papers) Uniformly and locally convex asymmetric spaces I. G. Tsar'kovab a Faculty of Mechanics and Mathematics, Lomonosov Moscow State University, Moscow, Russia b Moscow Center of Fundamental and Applied Mathematics, Moscow, Russia
Bibliographic databases:
     § 1. IntroductionIn the present paper we consider linear spaces with asymmetric norm ‖⋅|, which are more general than classical normed linear spaces. An asymmetric norm on a linear space X is defined by the following axioms: 1) ‖αx|=α‖x| for all α⩾, x\in X; 2) \| x+y|\leqslant \| x |+\| y| for all x,y\in X; 3) \|x|\geqslant 0 for all x\in X, and 3a) \|x|= 0 \Longleftrightarrow x=0. An asymmetric norm is generated by the Minkowski functional of some (generally speaking, asymmetric) set containing the origin in its kernel. In parallel with asymmetric norms {\|\cdot|} it is often convenient to consider the symmetrization norms defined by \|x\|:=\max\{\|x|,\|-x|\} (x\in X). In general, a space with an asymmetric norm (an asymmetric space) satisfies only the T_1 separation axiom (that is, for any points a, b \in X, one can find neighbourhoods O(a) and O(b) of these points such that a\notin O(b) and b\notin O(a)), and can fail to be Hausdorff. We will also consider asymmetric seminorms {\|\cdot|}, which are defined by axioms 1)–3), while axiom 3a) is replaced by the condition \|x|= 0=\|-x| \Longrightarrow x=0. An asymmetric space (X,{\|\cdot|}) equipped with an asymmetric seminorm will simply be called a seminormed space. The case of seminormed spaces will always be explicitly mentioned; otherwise a space will be assumed to be equipped with an asymmetric norm. Among the asymmetric spaces, we will consider the subclass of uniformly convex spaces. The main purpose of this paper is to find a definition of a uniformly convex space for which most properties of uniformly convex spaces (in the normed linear space setting) can be carried over to essentially asymmetric spaces (a space is said to be essentially asymmetric if its asymmetric norm is not equivalent to the symmetrization norm). Given a linear asymmetric normed space or a seminormed space X=(X,{\|\cdot|}), B(x,r)=\{y\in X\mid \|y-x|\leqslant r \} and \mathring{B}(x,r) =\{y\in X\mid \|y-x|< r\} are, respectively, the ‘closed’ and open balls of radius r with centre x. Note that the ‘closed’ ball B(x,r) can fail to be closed relative to the topology generated by the subbase of open balls. The sphere of radius r with centre x is defined by S(x,r) = \{y\in X\mid \|y-x |=r\}. The unit sphere in X is denoted by S=S(0,1), and the dual unit sphere in the dual space X^* of X is denoted by S^*=S^*(0,1). Given an arbitrary subset M of an asymmetric normed space X or a seminormed space X , the distance of a point y\in X to the set M is defined by \varrho(y,M) := \inf_{z\in M}\|z-y|. The set of nearest points in M to some x\in X is defined by P_Mx :=\{y\in M\mid { }{\|y-x|=\varrho(x,M)\}}. Given x\in X and \delta>0, we also consider the metric \delta-projections P_M^\delta x := \{y\in M\mid \|y-x|\leqslant\varrho(x,M)+\delta\}=M\cap B(x,\varrho(x,M)+\delta) and \mathring{P}_M^\delta x := \{y\in M\mid \|y-x|<\varrho(x,M)+\delta\}=M\cap \mathring{B}(x,\varrho(x,M)+\delta). For more on asymmetric spaces, see the survey [1]. Various approximative properties in asymmetric spaces were discussed in [2]–[12]. The novelty of our paper is in the application of the new definition of a uniformly convex asymmetric normed space to classical problems of geometric approximation theory. The paper is organized as follows. In § 2 we obtain conditions under which the intersection of any nested family of nonempty convex closed sets is nonempty (Theorem 2). Then we present conditions under which the set of points of approximative uniqueness for an arbitrary nonempty closed subset of an essentially asymmetric uniformly convex space is dense in this space (Theorem 3). Problems of the existence and uniqueness of Chebyshev centres in asymmetric uniformly convex spaces are discussed in § 4 (see Theorem 7). In the concluding section, § 5, we consider the relationships between \gamma-suns, suns and the existence of best approximation (Theorems 8 and 9); we also find a sufficient condition for the radial \delta-solarity of a set in terms of the convergence to zero of its modulus of uniform approximative continuity (Theorem 10). Given a uniformly convex symmetric norm \Psi(\,\cdot\,{,}\,\cdot\,)\colon \mathbb{R}^2\to \mathbb{R}, and 1<p,q<\infty, the space L_{p,q,\Psi}(\Omega) is defined as the space of equivalence classes of measurable functions f\colon \Omega\to \mathbb{R} with finite L_p(\Omega)- and L_q(\Omega)-norms. It is equipped with the asymmetric norm \|f| defined by \Psi(\|f_+\|_p,\|f_-\|_q), where \|\varphi\|_\theta:=\|\varphi\|_{L_\theta(\Omega)}; see [12]). This space, which serves as a model example in the study of asymmetric uniformly convex spaces, originates from the classical L_p-spaces, which are frequently useful in classical approximation of functions (and, in particular, in problems of approximation by trigonometric polynomials). Numerous general applications of asymmetric norms include numerical mathematics, economics, geometrical optics and differential equations, in which asymmetric convex positively homogeneous functionals (asymmetric norms, in our language) are useful. Asymmetric norms seem to date back to Minkowski (the Minkowski functional; see [13]). M. Krein was the first to use asymmetric norms in infinite-dimensional analysis (see [14]); he also gave them the name of asymmetric norms in 1938 (see [14]). Krein, as a single author and with co-authors, used asymmetric norms when dealing with extremal problems related to the Markov moment problem. The importance of sublinear functionals for several problems in convex analysis and calculus was pointed out by König (see [15] and [16]). Asymmetric distances also appear naturally in approximation theory (in particular, in problems of best one-sided approximations). In this regard, Babenko [17] noted that approximations in spaces with asymmetric norms serve as a ‘bridge’ between best approximations and best one-sided approximations (in this connection we mention the papers [18] and [19] by Dolzhenko and Sevast’yanov and the books [20] (§ I.9.E) by Collatz and Krabs and [1] by Cobzaş). From the point of view of approximation theory it seems interesting to apply the machinery of asymmetric norms to problems of complexity analysis in computer science (see [21]). In geometric approximation theory and convex analysis, asymmetric distances were considered by Borodin [22], [23], Ivanov and Lopushanski [24], [25], Alimov [4]–[6], and by this author, on his own [7]–[12] and in collaboration with Alimov [26], [27]. Let us now give the definition of a uniformly convex asymmetric space. We set Definition 1. An asymmetric space X=(X,{\|\cdot|}) is called uniformly convex if for all \varepsilon>0 and a\in (0,1] there exists \delta>0 such that for all f,g \in X, \|f|=\|g|=1, the condition \Delta(a)<\delta implies that f \in B(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. It follows from this definition that for any \varepsilon>0 there exists \delta>0 such that, for all f,g \in X, \|f|=\|g|=1, the condition \|(f+g)/2|\geqslant 1-\delta implies that \|f -\mu g|\leqslant\varepsilon for some \mu\in [1-\varepsilon,1].
\begin{equation*}
0\leqslant \Delta(a_0)\leqslant \Delta(a) \quad \forall\, a\geqslant a_0>0.
\end{equation*}
\notag
Indeed, by the triangle inequality \|f-a_0g|\leqslant \|f- ag|+(a-a_0)\|g|, which gives Remark 1 implies that if in the definition of uniform convexity we choose \delta>0 for some \varepsilon>0 and a_0\in (0,1], then the same \delta>0 also works for the same \varepsilon>0 and all a\in [a_0,1]. So, in order to show that X is uniformly convex, it suffices to verify that for any \varepsilon>0 and arbitrarily small a>0 (for example, for 0<a\ll\varepsilon), there exists \delta=\delta(a)>0 such that, for all f,g\in X, \|f|=\|g|=1, the condition \Delta(a)<\delta implies that f \in B(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. Remark 2. Note that if X=(X,{\|\cdot|}) is uniformly convex, then so is the mirror asymmetric space X^-=(X,{\|-\cdot|}). Definition 2. An asymmetric space X=(X,{\|\cdot|}) is called left-uniformly convex if for all \varepsilon>0 and a\in (0,1] there exists \delta>0 such that for all f,g\in X, \|f|=\|g|=1, the condition \Delta(a)<\delta implies that g \in B(\mu f,\varepsilon) for some \mu\in [1-\varepsilon,1]. Remark 3. Definitions 1 and 2 can be given not only for normed linear spaces X, but also for seminormed spaces X. § 2. Nested bounded convex closed setsDefinition 3. We say that \{x_n\} \subset X is a Cauchy sequence (an inverse Cauchy sequence) if for each \varepsilon>0 there exists N\in \mathbb{N} such that \|x_m-x_n|< \varepsilon (respectively, \|x_n-x_m|<\varepsilon) for all m\geqslant n\geqslant N. An asymmetric space X=(X,{\|\cdot|}) is called right-complete (left-complete) if for each Cauchy sequence \{x_n\}\subset X there exists a point x\in X such that \|x-x_n|\to 0 (respectively, \|x_n-x|\to 0) as n\to\infty. A right-complete space will be simply called a complete space. An asymmetric space X=(X,{\|\cdot|}) is called inversely right-complete (inversely left-complete) if for each inverse Cauchy sequence \{x_n\}\subset X there exists a point x\in X such that \|x\,{-}\,x_n| \to 0 (respectively, \|x_n\,{-}\,x| \to 0) as n\to\infty. An inversely right-complete space will simply be referred to as an inversely-complete space. A right-complete space will be called a complete space. Remark 4. Note that if X=(X,\|\cdot|) is right-complete (left-complete), then the mirror asymmetric space X^-=(X,{\|-\cdot|}) is inversely left-complete (respectively, inversely right-complete). Remark 5. A set N in an asymmetric space X=(X,{\|\cdot|}) is closed (relative to the topology generated by open balls) if y\in N whenever there exists a sequence \{y_n\}\subset N such that \|y_n-y|\to 0 as n\to\infty. Remark 6. Definition 3 can also be introduced not only for normed linear spaces X, but also for seminormed spaces X. Remark 7. The unit ball B(0,1) of an asymmetric space X=(X,{\|\cdot|}) is closed relative to right-convergence (this convergence corresponds to the topology generated by the left-open balls), that is, y\in B(0,1) whenever there exists a sequence \{y_n\}\subset B(0,1) such that \|y-y_n|\to 0 as n\to\infty. Indeed, if \|y|>1, then there exists a bounded linear functional y^*\in S^* such that y^*(y)=\alpha>1\geqslant y^*(z) for all z\in B(0,1) (see [3]). Then \|y-y_n|\geqslant y^*(y-y_n)\geqslant \alpha -1\nrightarrow 0 as n\to\infty, which is impossible. Definition 4. An asymmetric space X=(X,{\|\cdot|}) is called locally uniformly convex if, for all g\in S, \varepsilon>0 and a\in (0,1], there exists \delta>0 such that for any f\in S the condition \|f-ag|+a\|g|-\|f|<\delta implies that \|f-g|<\varepsilon. If \|g-f|<\varepsilon under the above conditions, then we say that X is a left-locally uniformly convex space. Remark 8. Note that if a linear space X{=}\,(X,{\|\cdot|}) is locally uniformly convex, then so is the mirror asymmetric space X^- = (X,{\|-\cdot|}). For ordinary normed spaces the above definition of locally uniformly convex spaces is equivalent to several well-known characteristic properties of such spaces. The following result is Corollary 1 in [12], proved there for asymmetric normed spaces. Corollary 1. Any uniformly convex linear space with asymmetric seminorm is a locally uniformly convex asymmetric space. Proof. There exists \mu\in [1-\varepsilon,1] such that f,g\in B(\mu g,\varepsilon). Hence and the required result follows. The following result is Theorem 1 in [12]. Theorem A. Let X=(X,{\|\cdot|}) be an asymmetric locally uniformly convex space, and let x\in S and x^*\in S^* be vectors such that x^*(x)=1. Also let \{x_n\}\subset M:=\{z\in X\mid x^*(z)\geqslant 1\}, and let \|x_n|\to 1 as n\to\infty. Then \|x_n-x|\to 0 as n\to\infty. As usual, a space is Hausdorff (or satisfies the T_2 separation axiom) if any two distinct points a,b\in X have disjoint neighbourhoods. We also note that an asymmetric space (X,{\|\cdot|}) is Hausdorff if and only if so is the mirror space (X,{\|-\cdot|}). The following remark is frequently useful. Remark 9. Let X be a locally uniformly convex space with T_2 separation axiom (respectively, T_1 separation axiom), and let \{x_n\}\subset B(0,1) and \|x_n-x|\to 0 (respectively, \|x-x_n|\to 0) as n\to\infty. By the asymmetric version of the Hahn-Banach theorem (see [3], Theorem 4.3) there exists a bounded linear functional x^*\in S^* such that x^*(x)=\|x|. Assume that x^*(x_n)\to 1. Then x\in S. To prove the claim in Remark 9 we consider the Hausdorff case first. Since the sets \{z\in X\mid x^*(z)\geqslant c\} are closed for all c\in \mathbb{R}, it suffices to show that x\in B(0,1). Assume on the contrary that \|x|>1. Let x_0=x/\|x|. The linear functional x^*\in S^* separates the ball B(0,1) and the point x_0. Hence \{x'_n:= x_n/x^*(x_n)\} is a minimizing sequence for the origin in M:=\{z\in X\mid x^*(z)\geqslant 1\}. Now Theorem A implies that \| x'_n- x_0|\to 0 as n\to\infty. In the triangle with vertices 0, x and x'_n the line segments [x_n,x] and [x'_n,x_0] intersect at some point y_n, n\in \mathbb{N}. Hence \|y_n-x|\leqslant \|x_n-x|\to 0 and \|y_n-x_0|\leqslant \|x'_n-x_0|\to 0 as n\to\infty, which is impossible for Hausdorff topological spaces. This contradiction completes the proof in the first case. In the second case (for T_1 spaces), at the last step we have \|x -y_n|\leqslant \|x-x_n|\to 0 and \|y_n-x_0|\leqslant \|x'_n-x_0|\to 0 as n\to\infty, which cannot be the case for T_1 topological spaces. Proposition 1. Let X=(X,{\|\cdot|}) be a Hausdorff asymmetric normed space and let N, \varnothing\ne N\subset X, be such that N\subset B(x_n,\varepsilon_n) for all n\in \mathbb{N}, for some sequence \{x_n\}\subset X and some nonnegative null sequence \{\varepsilon_n\}. Then N is a singleton. Proof. Assume on the contrary that N contains two distinct points a,b\in N. It can be assumed without loss of generality that N=[a,b] . Then \|a-x_n| \to 0 and \|b-x_n|\to 0, that is, the sequence \{x_n\} right-converges to two distinct limit points, which is impossible in the Hausdorff mirror space X=(X,{\|-\cdot|}). Proposition 1 is proved. Theorem 1. Let X=(X,{\|\cdot|}) be a right-complete uniformly convex asymmetric space and let x^* be a norm-one linear functional. Then the origin has a unique nearest point in the set M:=\{z\in X\mid x^*(z)\geqslant 1\} and, moreover, if \{x_n \}\subset M and \|x_n|\to 1=\varrho(0,M) as n\to\infty, then \|x_n -x|\to 0 as n\to\infty. Proof. Let \varepsilon\in (0,1) and let \varepsilon_n:={\varepsilon}/{2^{n+2}}. Since X is uniformly convex, for any \varepsilon\in (0,1) there exists \delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8}) such that for all f,g\in S the condition 0\leqslant \|f-\frac{1}{2}g|+\frac{1}{2}\|g|-\|f|<\delta(\varepsilon) implies that f,g\in B(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1].
We construct the sequence \{f_i\}_{i=0}^{\infty}\subset S recursively. 1. Let \delta_1:=\delta(\varepsilon_1). There exist f_0,f_1\in S such that x^*(f_1),x^*(f_0)\geqslant 1-{\delta_1}/{100}. Setting g=f_0 and f=(f_1+g)/2, we have x^*(f)\geqslant 1-{\delta_1}/{100} and \|f-\frac{1}{2}g|+\frac{1}{2}\|g|=\frac{1}{2}\|f_1|+\frac{1}{2}\|g|=1=\|f_1|. Let \beta_1\geqslant 1 be such that \beta_1 f\in S. Then 1\geqslant x^*(\beta_1 f)=\beta_1 (1-{\delta_1}/{100}), which implies that 1\leqslant\beta_1\leqslant {1}/(1-\delta_1/100). Hence 0\leqslant\beta_1-1\leqslant ({\delta_1}/{100})/(1-{\delta_1}/{100})<{\varepsilon_1}/{4} and
\begin{equation*}
1=\|\beta_1 f|\leqslant \biggl\|\beta_1 f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g| \leqslant \biggl\| f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|+(\beta_1-1)\|f| \leqslant1+\frac{{\delta_1}/{100}}{1-{\delta_1}/{100}}.
\end{equation*}
\notag
Consequently,
\begin{equation*}
\biggl\|\beta_1 f-\frac{1}{2}g\biggr| +\frac{1}{2}\|g|-1\leqslant\frac{{\delta_1}/{100}}{1-{\delta_1}/{100}}.
\end{equation*}
\notag
Therefore, \beta_1 f,g\in B(\widehat{\mu}_1 g,\varepsilon_1) for some \widehat{\mu}_1\in [1-\varepsilon_1,1]. Since [f,(\widehat{\mu}_1/\beta_1)g] is the midline of the triangle with vertices f_1, g and (2{\widehat{\mu}_1}/{\beta_1}-1)g, we obtain
\begin{equation*}
\biggl\|f-\frac{\widehat{\mu}_1}{\beta_1}g\biggr| \leqslant \frac{\varepsilon_1}{\beta_1}\quad\text{and} \quad \|f_1-\mu_1g|\leqslant 2 \frac{\varepsilon_1}{\beta_1} \leqslant 4\varepsilon_1, \quad\text{where } \mu_1=2\frac{\widehat{\mu}_1}{\beta_1}-1.
\end{equation*}
\notag
Hence f_1\in B(\mu_1 g,4 \varepsilon_1).
2. Let \delta_k:=\min_{m=0,\dots,k}\delta(\varepsilon_{m+1}). Assume that we have already constructed points f_0,\ldots,f_{n-1}\in S such that
\begin{equation*}
x^*(f_k)\geqslant 1-\frac{\delta_{k+1}}{100}, \qquad k=0,\dots,n-1.
\end{equation*}
\notag
There exists a point f_n\in S such that x^*(f_{n}) \geqslant 1-{\delta_{n+1}}/{100}\geqslant 1-{\delta_{n}}/{100}. Setting g=f_{n-1} and f=(f_{n}+g)/2 we have x^*(f)\geqslant 1-{\delta_n}/{100} and \|f-\frac{1}{2}g|+\frac{1}{2}\|g|=\frac{1}{2}\|f_n|+\frac{1}{2}\|g|=1=\|f_n|. Let \beta_{n}\geqslant 1 be such that \beta_{n} f\in S. Then 1\geqslant x^*(\beta_{n} f)=\beta_{n} (1-{\delta_{n}}/{100}), and therefore 1\leqslant\beta_{n}\leqslant {1}/(1-{\delta_{n}}/{100}). Hence 0\leqslant\beta_{n}-1\leqslant ({\delta_{n}}/{100})/(1-\delta_{n}/100)<{\varepsilon_n}/{4} and
\begin{equation*}
1=\|\beta_{n} f|\leqslant \biggl\|\beta_n f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g| \leqslant \biggl\| f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|+(\beta_{n}-1)\|f| =1+\frac{{\delta_{n}}/{100}}{1-{\delta_{n}}/{100}}.
\end{equation*}
\notag
Consequently,
\begin{equation*}
\biggl\|\beta_n f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|-1\leqslant\frac{{\delta_n}/{100}}{1-{\delta_n}/{100}}.
\end{equation*}
\notag
Therefore, \beta_{n} f,g\in B(\widehat{\mu}_{n} g,\varepsilon_{n}) for some \widehat{\mu}_{n}\in [1-\varepsilon_{n},1]. Thus,
\begin{equation*}
\biggl\|f-\frac{\widehat{\mu}_n}{\beta_n}g\biggr| \leqslant \frac{\varepsilon_n}{\beta_n}\quad\text{and} \quad\|f_n-\mu_ng|\leqslant 2 \frac{\varepsilon_n}{\beta_n} \leqslant 4\varepsilon_n, \quad\text{where } \mu_n=2\frac{\widehat{\mu}_n}{\beta_n}-1.
\end{equation*}
\notag
Hence f_{n}\in B(\mu_{n} g,4 \varepsilon_n).
3. Note that at each step we can select \varepsilon_{n+1} sufficiently close to zero so that the product \prod_{k=1}^\infty\mu_k is convergent. In this case, of course, P_m:=\prod_{k=m+1}^\infty\mu_k\to 1 as {m\to\infty}. Setting \widehat{f}_k:=P_kf_k, k\in \mathbb{N}, we have
\begin{equation*}
\|\widehat{f}_{k+1}-\widehat{f}_k|=P_{k+1}\|f_{k+1}-\mu_{k+1} f_k|\leqslant \|f_{k+1}-\mu_{k+1} f_k|, \qquad k\in \mathbb{N}.
\end{equation*}
\notag
It follows that
\begin{equation*}
\begin{gathered} \, 1\geqslant \|\widehat{f}_k|\geqslant x^*(\widehat{f}_k)=P_k x^*( {f}_k)\geqslant P_k \biggl(1-\frac{\delta_{n}}{100}\biggr)\to 1, \qquad k\to\infty, \\ \sum_{k= m}^{n}\|\widehat{f}_{k+1}-\widehat{f}_k |\leqslant 4\sum_{k= m}^{n}\frac{\varepsilon}{2^{k+2}}\to 0, \qquad n\to\infty. \end{gathered}
\end{equation*}
\notag
So \{\widehat{f}_n\} is a Cauchy sequence in the complete space X. Therefore, there exists x\in X such that \|x -\widehat{f}_n|\to 0 as n\to\infty. Now, x\in S and x^*(x)=1 by Remark 9.
We claim that \{g\in S\mid x^*(g)=1\}=\{x\}. Indeed, if there exists g\in S such that x^*(g)=1, then the line segment [g,x]\subset B(0,1)\cap M lies in the sphere S and consists of nearest points in M to the origin. By uniform convexity there exists a sequence \{\mu_n\}\subset [0,1] such that \lim_{n\to\infty}\mu_n=1 and \|g-\mu_n x|\to 0 as n\to\infty. Hence
\begin{equation*}
\|g- x|\leqslant \|g-\mu_n x|+\|\mu_n x-x|=\|g-\mu_n x|+(1-\mu_n)\|-x|\to 0, \qquad n\to\infty.
\end{equation*}
\notag
Therefore, g=x.
By Corollary 1 and Theorem A, if \{x_n\}\subset M:=\{z\in X\mid x^*(z)\geqslant 1\} and \|x_n|\to 1 as n\to\infty, then \|x_n-x|\to 0 as n\to\infty. Theorem 1 is proved. Corollary 2. Let N be a nonempty closed convex subset of an asymmetric uniformly convex right-complete space X=(X,{\|\cdot|}). Then each x\in X has a unique point y\in N in N (that is, \|y-x|=\varrho(x,N)) and, moreover, if \{y_n\}\subset N is a minimizing sequence in N for x (that is, \|y_n-x|\to \varrho(x,N) as n\to\infty), then \|y_n-y|\to 0 as n\to\infty. Proof. It can be assumed without loss of generality that x\in X\setminus N, x=0 and \varrho(x,N)=1. By the asymmetric Hahn-Banach theorem (see [3], Theorem 4.3, there exists a bounded linear functional x^*\in S^* separating the ball \mathring{B}(0,1) and the set N, that is, N\subset M:=\{y\in X\mid x^*(y)\geqslant 1\}. By Theorem 2 there exists a unique point y\in M such that \|y-x|=\varrho(0,M) and, moreover, \|y_n-y|\to 0 as n\to\infty for any sequence \{y_n\}\subset N\subset M such that \|y_n-x|\to \varrho(x,M)=1=\varrho(x,N) as {n\to\infty}. Since N is closed, y lies in N. This proves Corollary 2. Remark 10. The existence of a nearest point in M (in Theorem 1) and in N (in Corollary 2) can also be verified in seminormed spaces. In this case, a nearest point is unique if the unit sphere S does not contain nondegenerate line segments. For asymmetric normed spaces this result easily follows from local uniform convexity, which, in turn, is secured by Corollary 1. The following result is Corollary 2 in [12]. Theorem B. Let N be a nonempty closed convex subset of a Hausdorff asymmetric uniformly convex left-complete space X=(X,{\|\cdot|}). Then each x\in X has a unique nearest point y\in N in N (that is, \|y-x|=\varrho(x,N)) and, moreover, if \{y_n\}\subset N is a minimizing sequence in N for x (that is, \|y_n- x|\to \varrho(x,N) as n\to\infty), then \|y_n-y|\to 0 as n\to\infty. Theorem 2. Let X=(X,{\|\cdot|}) be a Hausdorff left-complete uniformly convex asymmetric space, and let \{N_k\} be a nested sequence of nonempty convex closed bounded sets. Then the intersection N:=\bigcap_{k=1}^\infty N_k is nonempty. Proof. It can be assumed without loss of generality that 0\notin N_1. By assumption the sequence \{\varrho_k:=\varrho(0,N_k)\} is bounded and monotone nondecreasing, and therefore it converges to a finite limit \varrho. It can be assumed without loss of generality that \varrho=1. By Corollary 2 in [12] a nearest point in N_k to the origin exists and is unique.
Let \varepsilon\in (0,1) and let \varepsilon_n:={\varepsilon}/{2^{n+2}}. Since X is uniformly convex, for all \varepsilon\in (0,1) there exists \delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8}) such that for any f,g\in S the condition 0\leqslant \|f-\frac{1}{2}g|+\frac{1}{2}\|g|-\|f|<\delta(\varepsilon) implies that f,g\in B(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. We can choose a subsequence \{N_{k_n}\}_{n=0}^\infty such that \varrho_{k_n}\geqslant 1-{\delta(\varepsilon_n/2)}/{200}. For convenience we denote this subsequence and the corresponding subsequence \{\varrho_{k_n}\} by \{N_n\} and \{\varrho_n\} again. For each n\in \mathbb{Z}_+ there exists a bounded linear functional \{x^*_n\}\subset S^* that separates the ball B(0,1-{\delta(\varepsilon_n/2)}/{100}) and the set N_n. We construct the sequence \{f_n\}_{n=0}^{\infty}\subset S recursively. 1. Let \delta_1:=\delta(\varepsilon_1/2) and let \psi_0 and \psi_1 be nearest points to the origin in the sets N_0 and N_1, respectively, that is, \|\psi_0|=\varrho_0 and \|\psi_1|=\varrho_1. Setting f_0:=\varrho_0^{-1}\psi_0 and f_1:=\varrho_1^{-1}\psi_1\in S, we have x^*_0(f_1),x^*_0(f_0)\geqslant 1-{\delta_1}/{100}. We also put g=f_0 and f=(f_1+g)/{2}. Then we have
\begin{equation*}
x^*_0(f)\geqslant 1-\frac{\delta_1}{100}\quad\text{and} \quad \biggl\|f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|=\frac{1}{2}\|f_1|+\frac{1}{2}\|g|=1=\|f_1|.
\end{equation*}
\notag
Let \beta_1\geqslant 1 and \beta_1 f\in S. Then 1\geqslant x^*_0(\beta_1 f)=\beta_1 (1-{\delta_1}/{100}), and so 1\leqslant\beta_1\leqslant {1}/(1-{\delta_1}/{100}). Therefore, 0\leqslant\beta_1-1\leqslant (\delta_1/100)/(1-\delta_1/100)<\varepsilon_1/4 and
\begin{equation*}
1=\|\beta_1 f|\leqslant \biggl\|\beta_1 f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g| \leqslant \biggl\| f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|+(\beta_1-1)\|f| \leqslant 1+\frac{{\delta_1}/{100}}{1-{\delta_1}/{100}}.
\end{equation*}
\notag
Hence
\begin{equation*}
\biggl\|\beta_1 f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|-1\leqslant\frac{\delta_1/100}{1-\delta_1/100}.
\end{equation*}
\notag
Consequently, \beta_1 f,g\in B(\widehat{\mu}_1 g,\varepsilon_1) for some \widehat{\mu}_1\in [1-\varepsilon_1/2,1]. Since [f,(\widehat{\mu}_1/\beta_1)g] is the midline of the triangle with vertices f_1, g and (2{\widehat{\mu}_1}/{\beta_1}-1)g, we obtain
\begin{equation*}
\biggl\|f-\frac{\widehat{\mu}_1}{\beta_1}g\biggr| \leqslant \frac{\varepsilon_1}{\beta_1}\quad\text{and} \quad\|f_1-\mu_1g|\leqslant 2 \frac{\varepsilon_1}{\beta_1} \leqslant 4\varepsilon_1, \quad\text{where } \mu_1=2\frac{\widehat{\mu}_1}{\beta_1}-1\in [1-\varepsilon_1,1],
\end{equation*}
\notag
Hence f_1\in B(\mu_1 g,4 \varepsilon_1). As a result,
\begin{equation*}
\|f_1-(1-\varepsilon_1)g|\leqslant \|f_1-\mu_1g|+\|\mu_1g-(1-\varepsilon_1)g|\leqslant 4 \varepsilon_1 + \varepsilon_1=5\varepsilon_1,
\end{equation*}
\notag
and f_1\in B(\mu_1 g, 5\varepsilon_1), where we take \mu_1=1-\varepsilon_1. Therefore,
\begin{equation*}
\psi_1\in B(\mu_1\varrho_1 g,5 \varepsilon_1\varrho_1)= B\biggl(\frac{\mu_1\varrho_1}{\varrho_0} \psi_0,5 \varepsilon_1\varrho_1\biggr).
\end{equation*}
\notag
2. Let \delta_k:=\min_{m=0,\dots,k}\delta(\varepsilon_{m+1}/2), k=0,\dots,n. Assume that
\begin{equation*}
f_0,\ldots,f_{n-1}\in S \ \ \text{are such that}\ \ x^*(f_k)\geqslant 1-\frac{\delta_{k+1}}{100}, \quad k=0,\dots, n-1.
\end{equation*}
\notag
Let \psi_n be a nearest point to the origin in the set N_n, that is, \|\psi_n|=\varrho_n. Setting f_n:=\varrho_n^{-1}\psi_n \in S we have
\begin{equation*}
x^*_n(f_{n}) \geqslant 1-\frac{\delta_{n+1}}{100}\geqslant 1-\frac{\delta_{n}}{100}.
\end{equation*}
\notag
We also set g=f_{n-1} and f=(f_{n}+g)/2. Then
\begin{equation*}
x^*_{n-1}(f)\geqslant 1-\frac{\delta_n}{100}\quad\text{and} \quad \biggl\|f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g| =\frac{1}{2}\|f_n|+\frac{1}{2}\|g|=1=\|f_n|.
\end{equation*}
\notag
Let \beta_{n}\geqslant 1 be such that \beta_{n} f\in S. Then 1\geqslant x^*(\beta_{n} f)=\beta_{n} (1-{\delta_{n}}/{100}), and therefore 1\leqslant\beta_{n}\leqslant {1}/(1-{\delta_{n}}/{100}). Hence {0\leqslant\beta_{n}-1\leqslant ({\delta_{n}}/{100})/(1-{\delta_{n}}/{100})<\varepsilon_n/{4}} and
\begin{equation*}
1=\|\beta_{n} f|\leqslant \biggl\|\beta_n f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g| \leqslant \biggl\| f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|+(\beta_{n}-1)\|f| \leqslant 1+\frac{\delta_{n}/100}{1-\delta_{n}/100}.
\end{equation*}
\notag
Thus,
\begin{equation*}
\biggl\|\beta_n f-\frac{1}{2}g\biggr|+\frac{1}{2}\|g|-1 \leqslant\frac{\delta_n/100}{1-\delta_n/100}.
\end{equation*}
\notag
Therefore, \beta_{n} f,g\in B(\widehat{\mu}_{n} g,\varepsilon_{n}) for some \widehat{\mu}_{n}\in [1-\varepsilon_{n}/2,1]. Hence
\begin{equation*}
\biggl\|f-\frac{\widehat{\mu}_n}{\beta_n}g\biggr| \leqslant \frac{\varepsilon_n}{\beta_n}\ \ \text{and} \ \ \|f_n-\mu_ng|\leqslant 2 \frac{\varepsilon_n}{\beta_n} \leqslant 4\varepsilon_n, \quad\text{where } \mu_n=2\frac{\widehat{\mu}_n}{\beta_n}-1\in [1-\varepsilon_{n},1].
\end{equation*}
\notag
As a result, f_{n}\in B(\mu_{n} g,4 \varepsilon_n). Proceeding as at step 1 and taking the number 1-\varepsilon_n as \mu_n we see that f_{n}\in B(\mu_{n} g,5\varepsilon_n). Therefore,
\begin{equation*}
\psi_n\in B(\mu_n\varrho_n g,5 \varepsilon_n\varrho_n)= B\biggl(\frac{\mu_n\varrho_n}{\varrho_{n-1}} \psi_{n-1},5 \varepsilon_n\varrho_n\biggr).
\end{equation*}
\notag
Note that by the choice of \varrho_n the quantity {\mu_n\varrho_n}/{\varrho_{n-1}} is smaller than 1 (here we recall that \mu_{n}=1-\varepsilon_n).
3. Note that at each step we can take \varepsilon_{n+1} to be sufficiently close to 0 so that the product \prod_{k=1}^\infty\mu_k is convergent; of course, in this case we have P_m:={\prod_{k=m+1}^\infty{\mu_k\varrho_k}/(\varrho_{k-1})\to1} as {m\to\infty}. Setting {\widehat{\psi}_k:=P_k\psi_k}, k\in \mathbb{N}, we obtain
\begin{equation*}
\|\widehat{\psi}_{k+1}-\widehat{\psi}_k| =P_{k+1}\biggl\|\psi_{k+1}-\frac{\mu_{k+1}\varrho_{k+1}}{\varrho_{k}} \psi_k\biggr| \leqslant \biggl\|\psi_{k+1}-\frac{\mu_{k+1}\varrho_{k+1}}{\varrho_{k}} \psi_k\biggr|, \qquad k\in \mathbb{N}.
\end{equation*}
\notag
It follows that
\begin{equation*}
\sum_{k= m}^{n}\|\widehat{\psi}_{k+1}-\widehat{\psi}_k |\leqslant 5\sum_{k= m}^{n}\frac{\varepsilon}{2^{k+2}}\to 0 \quad\text{as } n\to\infty.
\end{equation*}
\notag
So \{\widehat{\psi}_n\} is a Cauchy sequence in the left-complete space X. Therefore, there exists a point x\in X such that \|\widehat{\psi}_n -x|\to 0 as n\to\infty. Hence \|{\psi}_n -x \|\leqslant \|{\psi}_n-P_n{\psi}_n|+\|P_n{\psi}_n-x|= (1-P_n)\|{\psi}_n|+\|\widehat{\psi}_n-x|\to 0 as n\to \infty.
The set N:=\bigcap_{k=1}^\infty N_k is closed, so x\in N. Consequently, N is nonempty. Theorem 2 is proved. Remark 11. If, in addition to the hypotheses of Theorem 2 regarding the space X , we assume that the unit ‘closed’ ball of X is closed (in this case X is a fortiori a Hausdorff space; see [2]), then \psi_n (the nearest point to the origin in the set N_n) tends to x as n\to\infty, which is a nearest point to the origin in N:=\bigcap_{k=1}^\infty N_k. Of course, the same holds not only for the origin, but also for any point y\in X, that is, \|P_{N_n} y-P_Ny|\to 0 as n\to\infty. A similar result about nearest points is valid if the above additional property is replaced by the condition that \varrho(y,N_n)\to \varrho(y, N) as {n\to\infty}. The same result also holds if at least one set in this family is left-strongly closed (see Definition 5 below). § 3. Points of existence and points of approximative uniquenessLemma 1. Let X be a uniformly convex asymmetric seminormed space, let {\Delta\in (0,1)}, M\subset X, g_0\in X\setminus M and r=\varrho(g_0,M)>0. Then for any \varepsilon\in (0,1) there exists \delta_0\in (0,\varepsilon/8) such that if \widetilde{f}_0\in M, \|\widetilde{f}_0-g_0|<r+\delta_0, and f\in M, \|f-g_1|<\varrho(g_1,M)+\delta_0, where g^1:=(\widetilde{f}_0-g_0)/\|\widetilde{f}_0-g_0| and g_1:=g_0+\Delta g^1, then f\in B(\mu(\widetilde{f}_0-g_0)+g_0,\varepsilon) for some \mu\in [1-\varepsilon,1]. Proof. It can be assumed without loss of generality that g_0=0 and r=1. Since X is uniformly convex, for any \varepsilon\in (0,1) and \Delta\in (0,1) there exists \delta=\delta(\varepsilon,\Delta)>0 such that if f_0,f_1\in S and \|f_1-\Delta f_0|+\Delta\|f_0|-\|f_1|<\delta, then \|f_1-\mu f_0|<\varepsilon for some \mu\in [1-\varepsilon,1]. Let \delta_0:=\frac{1}{2}\min\{\delta(\varepsilon/2,\Delta),\varepsilon/8\}. Consider an arbitrary point \widetilde{f}_0\in M such that \|\widetilde{f}_0|<1+\delta_0. Setting g^1:=\widetilde{f}_0/\|\widetilde{f}_0| and g_1:=g_0+\Delta g^1=\Delta g^1, we have
\begin{equation*}
\varrho(g_1,M)\leqslant \|\widetilde{f}_0-g_1|=\|g^1\cdot \|\widetilde{f}_0|-\Delta g^1|\leqslant \|g^1(\|\widetilde{f}_0|-\Delta)| <1+\delta_0-\Delta
\end{equation*}
\notag
and 1-\Delta=\varrho(0,M)-\|g_1|\leqslant \varrho(g_1,M). Given an arbitrary f\in M, we have
\begin{equation*}
\|f-g_1|<\varrho(g_1,M)+\delta_0\leqslant 1-\Delta+2\delta_0\leqslant 1-\Delta +\delta\biggl(\frac\varepsilon2,\Delta\biggr).
\end{equation*}
\notag
Hence \|f-g_1|+\|g_1|\leqslant 1-\Delta +\delta(\varepsilon/2,\Delta)+\Delta= 1 +\delta(\varepsilon/2,\Delta). As a result, \|f\,{-}\,g_1|+\|g_1|-\|f|<\delta(\varepsilon/2,\Delta), and therefore f\in B(\mu' g^1,\varepsilon/2) for some \mu'\in [1-\varepsilon/2,1]. We have
\begin{equation*}
1\geqslant\frac{\mu'}{\|\widetilde{f}_0|}\geqslant\frac{1-\varepsilon/2}{1+\delta_0}\geqslant \frac{1-\varepsilon/2}{1+\varepsilon/8}\geqslant 1-\varepsilon,
\end{equation*}
\notag
so setting \mu={\mu'}/{\|\widetilde{f}_0|} we find that f\in B(\mu'{\widetilde{f}_0}/{\|\widetilde{f}_0|},\varepsilon/2)=B(\mu {\widetilde{f}_0} , \varepsilon/2). Lemma 1 is proved. Definition 5. A subset M of an asymmetric seminormed space X =(X,{\|\cdot|}) is called left-strongly closed if x\in M and \|x_n-x\|:=\max\{\|x_n-x|,\|x-x_n|\}\to 0 as n\to\infty, whenever \{x_n\}\subset M and \|x_n-x|\to 0 as n\to\infty. Lemma 2. Let X be a left-complete uniformly convex asymmetric seminormed space with closed unit ball and M\subset X be a left-strongly closed nonempty subset. Then any neighbourhood of an arbitrary point x\in X\setminus M contains an existence point for M (that is, a point that has a nearest point in M) at which the distance function \varrho(\cdot,M) is continuous. Proof. By assumption the ball B(0,1) is closed, hence any ‘closed’ ball is too. It can be assumed without loss of generality that \varrho(x,M)=1. We claim that any neighbourhood O_\sigma(x) of x contains a required existence point v_0\in X\setminus M. Let \Delta\in (0,\min\{1/3,\sigma/3\}). We construct sequences \{y_n\}\subset M and \{x_n\}\subset X recursively. For sufficiently small \varepsilon\in (0,1) we set \{\varepsilon_n={\varepsilon}/{2^n}\}, \delta_0=\varepsilon and \sigma_0=\Delta/2. We also let x_0=x.
1. We use Lemma 1. Let \delta_1\in (0,\varepsilon_1/8), let y_1\in M satisfy \|y_1-x|<\varrho(x,M)+\delta_1, and let x_1\in [x,y_1] satisfying \|x_1-x|=\Delta_1:=\Delta/2 (in this case \|y_1-x_1|<\varrho(x_1,M)+\delta_1) be such that, for any point f\in M such that \|f-x_1|<\varrho(x_1,M)+3\delta_1 we have f\in B(\mu_1(y_1-x)+x,\varepsilon_1) for some \mu_1\in [1-\varepsilon_1,1]. Next, let \sigma_1\in (0,\sigma), \sigma_1<\delta_0/2, be such that \varrho(x_1,M)\leqslant \varrho(z,M)+\delta_0/2 for all z\in B(x_1,\sigma_1). 2. Suppose that the points \{y_k\}_{k=1}^{n}\subset M, \{x_k\}_{k=1}^{n}\subset X and numbers \{\delta_k\}_{k=1}^{n}, \{\mu_k\}_{k=1}^{n}, \{\sigma_k\}_{k=1}^{n}, n\geqslant 2, have already been constructed so that, for all k ,
\begin{equation*}
\begin{gathered} \, \|y_{k}-x_{k-1}|<\varrho(x_{k-1},M)+\delta_{k}, \\ x_{k}\in [x_{k-1},y_{k}] \ \ \text{and}\ \ \|x_{k}-x_{k-1}|=\Delta_{k}:=\frac{1}{2}\min\{\Delta_{k-1},\sigma_{k-1},\delta_{k-1}\} \end{gathered}
\end{equation*}
\notag
(in this case \|y_{k}-x_{k}|<\varrho(x_{k},M)+\delta_{k}) and, in addition, for an arbitrary point f\in M, \|f-x_{k}|<\varrho(x_{k},M)+3\delta_{k} the inclusion f\in B(\mu_{k}(y_{k}-x_{k-1})+x_{k-1},\varepsilon_{k}) holds for some \mu_{k}\in [1-\varepsilon_{k},1]. Then y_{k+1}\in B(\mu_{k}(y_{k}-x_{k-1})+x_{k-1},\varepsilon_{k}), k<n. We also have \sigma_{k}<\delta_{k-1}/2, and, moreover, \varrho(x_{k},M)\leqslant \varrho(z,M)+\delta_{k-1}/2 for any z\in B(x_{k},\sigma_{k}).
By Lemma 1 there exist a number \delta_{n+1}\in (0,\min_{k=1,n}\delta_k), a point y_{n+1}\in M such that \|y_{n+1}-x_{n}|<\varrho(x_{n},M)+\delta_{n+1}, and a point
\begin{equation*}
x_{n+1}\in [x_{n},y_{n+1}]\colon \|x_{n+1}-x_{n}|=\Delta_{n+1}:=\frac{1}{2}\min\{\Delta_{n},\sigma_n,\delta_{n}\}
\end{equation*}
\notag
(in this case \|y_{n+1}-x_{n+1}|<\varrho(x_{n+1},M)+\delta_{n+1}) such that for an arbitrary point {f\in M} such that \|f-x_{n+1}|<\varrho(x_{n+1},M)+3\delta_{n+1} the inclusion f\in B(\mu_{n+1}(y_{n+1}- x_{n})+x_{n},\varepsilon_{n+1}) holds for some \mu_{n+1}\in [1-\varepsilon_{n+1},1]. Moreover, y_{n+1}\in B(\mu_{n}(y_{n}-x_{n-1})+x_{n-1},\varepsilon_{n }). Let \sigma_{n+1}\in (0,\min_{k=1,n}\sigma_k), \sigma_{n+1}<\delta_{n}/2, be a number such that \varrho(x_{n+1},M)\leqslant \varrho(z,M)+\delta_{n}/2 for any z\in B(x_{n+1},\sigma_{n+1}).
3. We have
\begin{equation*}
\begin{aligned} \, &\biggl\|\biggl(\prod_{k\geqslant n+1}\mu_k\biggr)y_{n+1}-\biggl(\prod_{k\geqslant n}\mu_k\biggr)y_n\biggr|= \biggl(\prod_{k\geqslant n+1}\mu_k \biggr)\|y_{n+1}-\mu_n y_n| \\ &\qquad\leqslant \|y_{n+1}-(\mu_{n}(y_{n}-x_{n-1})+x_{n-1})| +\|\mu_{n}(y_{n}-x_{n-1})+x_{n-1}-\mu_{n}y_{n}| \\ &\qquad\leqslant \varepsilon_{n}+\|(1-\mu_n)x_{n-1}|\leqslant 2 \varepsilon_{n}. \end{aligned}
\end{equation*}
\notag
Hence \{\widehat{y}_n:=(\prod_{k\geqslant n}\mu_k)y_n\} is a Cauchy sequence. Since X is left-complete, there exists a point y\in X such that \|\widehat{y}_n-y|\to 0 as n\to \infty. We have
\begin{equation*}
\|{y}_n-y|\leqslant \|{y}_n-\widehat{y}_n|+\|\widehat{y}_n-y|= \biggl(1-\prod_{k\geqslant n}\mu_k \biggr)\|y_n|+\|\widehat{y}_n-y|\to 0, \qquad n\to\infty.
\end{equation*}
\notag
This implies that y\in M because M is closed.
By construction \{x_n\} is a Cauchy sequence, and so there exists a point x^0\in X such that \|x_n-x^0|\to 0 as n\to\infty. Consequently,
\begin{equation*}
\varrho(x^0,M)\leqslant\|y_n-x^0|\leqslant \|y_n-x_n|+\|x_n-x^0|=\varrho(x_n,M)+\delta_n+\|x_n-x^0|
\end{equation*}
\notag
for all n\in \mathbb{N}. Since x^0\in B(x_n,\sigma_n) (here we have used the fact that the ball B(x_n,\sigma_n) is closed and \{x_k\}_{k\geqslant n+1}\subset B(x_n,\sigma_n)), we see that \varrho(x^0,M)+{\delta_n}/{2}\geqslant \varrho(x_n,M), and therefore \|y_n-x^0|,\varrho(x_n,M)\to \varrho(x^0,M) as n\to\infty. By assumption M is a left-strongly closed set, hence \|y-y_n|\to 0 as n\to\infty. Therefore,
\begin{equation*}
\varrho(x^0,M)\leqslant \|y-x^0|\leqslant \|y-y_n|+ \|y_n-x^0|\to \varrho(x^0,M), \qquad n\to\infty,
\end{equation*}
\notag
that is, y\in M is a nearest point to x^0 in M. Lemma 2 is proved. Theorem 3. Let X=(X,{\|\cdot|}) be a complete (or complete with respect to the symmetrization norm) locally uniformly convex asymmetric space, M be a nonempty closed set such that the set E of points of existence for M is dense with respect to the asymmetric norm of X (respectively, with respect to the symmetrization norm). Then the set of points of approximative uniqueness for M is of second category with respect to the asymmetric norm of X (respectively, with respect to the symmetrization norm). Proof. Let x\in E be such that \varrho(x,M)>0. It can be assumed without loss of generality that \varrho(x,M)=1 and x=0. By assumption the space X=(X,{\|\cdot|}) is locally uniformly convex. Hence for an arbitrary nearest point y\in P_Mx to x and arbitrary \varepsilon>0 and \Delta\in (0,1] there exists \delta=\delta(\varepsilon,\Delta)\in (0,1) such that, for any f\in S and \Delta'\in [\Delta/4,1], the condition \|f-\Delta' y|+\Delta'\|y|-\|f|\leqslant 3\delta implies that \|f-y|<\varepsilon. Moreover, here it can be assumed that \delta<\varepsilon. Let z\in [x,y], \|z-x|=\Delta. Then for any point w\in O^{\mathrm{sym}}(z)=O^{\mathrm{sym}}_\delta(z):=\{t\mid \|z-t\|<\delta\}, which yields \varrho(w,M)\leqslant \delta+1-\Delta. Correspondingly, if s\in M, \|s-w|\leqslant \varrho(w,M)+\delta, then
\begin{equation*}
\begin{aligned} \, \|s-x| &\leqslant \|s-z|+\|z-x|\leqslant \|s-w|+\|w-z\|+\|z-x| \\ &\leqslant \varrho(w,M)+\delta+\delta +\Delta\leqslant \delta+1-\Delta+2\delta +\Delta=1+3\delta. \end{aligned}
\end{equation*}
\notag
Note that then 1\leqslant \|s|\leqslant 1+3\delta. Setting s':=(s-x)/\|s-x|+x=s/\|s| we have \|s-s'|\leqslant {3\delta}/(1+3\delta)\|s|\leqslant 3\delta. Let z_1\in [x,z] be a point such that the line segments [z_1,s_1] and [z,s] are parallel. The triangles \triangle xz_1s_1 and \triangle xz s are similar with magnification ratio {\|s-x|}/{\|s_1-x|}\leqslant 1, hence
\begin{equation*}
\|z_1-x|+\|s_1-z_1|\leqslant \|z -x|+\|s -z |\leqslant 1+3\delta,
\end{equation*}
\notag
which implies that \|s_1-y|<\varepsilon. Thus,
Given points x\in E, where \varrho(x,M)>0, and y\in P_Mx and numbers \varepsilon\in (0,1) and \Delta\in (0,\|y-x|), consider the set A(x,y,\varepsilon,\Delta) defined as the union of the neighbourhoods O^{\mathrm{sym}}(z). The set A(x,y,\varepsilon,\Delta) is open with respect to the symmetrization norm. The closure of B(\varepsilon,\Delta):=\bigcup_{x,y}A(x,y,\varepsilon,\Delta) in X with respect to the symmetrization norm contains E. It is easily seen that \bigcap_{\varepsilon,\Delta}B(\varepsilon,\Delta) is a set of second category with respect to the asymmetric norm of X (or with respect to the symmetrization norm), and by construction it consists of points of approximative uniqueness. This proves Theorem 3. Theorem 4. Let X=(X,{\|\cdot|}) be a complete asymmetric space (or a left-complete uniformly convex T_2 asymmetric space), and let \{A_k\} be a sequence of closed sets without interior points with respect to the asymmetric norm (respectively, with respect to the symmetrization norm \|\cdot\|:=\max\{{\|\cdot|},{\|-\cdot|}\}). Then the complement of the set A:=\bigcup_{k=1}^\infty A_k is dense in X with respect to the asymmetric norm of X (respectively, with respect to the symmetrization norm of X). Proof. The arguments for the symmetrization norm are the same as for the asymmetric one. The only difference is that at the end of the proof we use the fact that, under the hypotheses of the theorem, the intersection of any nested bounded sequence of nonempty closed convex sets is nonempty (see Theorem 2). So we give a proof only for an asymmetric norm. Let B(x,R)\supset B(y,r). Then \|y-x|+r\leqslant R, that is, \|y-x|\leqslant R-r.
Let B (x_0,r_0) be a ball in X. We construct recursively a nested sequence of balls
\begin{equation*}
\{B(x_n,r_n)\}\subset B(x_0,r_0) \quad\text{such that }B(x_n,r_n)\cap A_n\neq \varnothing \text{ for all }n\in \mathbb{N}.
\end{equation*}
\notag
As a result, \|x_m-x_n|\leqslant r_n-r_m for all m\geqslant n, which implies that \{x_n\} is a Cauchy sequence. Since X is complete, there exists a point x\in X such that \|x-x_m|\to 0 as m\to\infty. Hence
\begin{equation*}
\|x-x_n|\leqslant \|x-x_m|+\|x_m-x_n|\to r_n-\lim_{m\to\infty}r_m\leqslant r_n, \qquad n\leqslant m\to\infty.
\end{equation*}
\notag
As a result, x\in B(x_n,r_n). Therefore, x\notin A_n for all n\in \mathbb{N}, which implies the required result. Theorem 4 is proved. Corollary 3. Assume that each nested sequence of nonempty closed convex sets in X=(X,{\|\cdot|}) has a nonempty intersection. Then X is complete with respect to the symmetrization norm. Proof. Note that any ball B_{\mathrm{sym}}(x ,r ):=\{y\in X\mid \|y-x\|\leqslant r\} is a closed set. It is known that if an arbitrary nested sequence of balls \{B_{\mathrm{sym}}(x_n,r_n)\} has a nonempty intersection, then the normed (with respect to the symmetrization norm) linear space is complete. Now the required result follows.
§ 4. Existence and uniqueness of Chebyshev centres for bounded setsIn this section we consider the classical problems of the existence and uniqueness of Chebyshev centres for asymmetric spaces. For a survey of similar problems in normed spaces, see [26]. Definition 6. Let X=(X,{\|\cdot|}) be an asymmetric space. A set M\subset X is said to be bounded if M\subset B(z,C) for some z\in X, C>0. A set M is left-bounded if M\subset B^-(w,C):=\{y\in X\mid \|w-y|\leqslant C'\} for some w\in X and C'>0. A set M is said to be bibounded if it is both bounded and left-bounded. Note that M is bibounded if and only if M is bounded with respect to the symmetrization norm. Theorem 5. Let X=(X,{\|\cdot|}) be a complete uniformly convex asymmetric space and let M \subset X be a bibounded set. Given C > 0, consider the nonempty set M_C\,{:=}\,\{z \in X\,|\, B(z,C) \supset M\}. Then M has a left-relative Chebyshev centre with respect to \overline{M_C}, that is, there exists a point x\in \overline{M_C} such that r(M,M_C):=\inf_{z\in M_C}\sup_{y\in M }\|z-y| is equal to \sup_{y\in M }\|x-y|. Proof. It can be assumed without loss of generality that r(M,M_C)=1 and 0\in M. Since X is uniformly convex, for any \varepsilon\in (0,1) there exists \delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8}) such that for all f,g\in S the condition 0\leqslant \|f-\frac{1}{2}g|+\frac{1}{2}\|g|-\|f|<\delta(\varepsilon) implies that f,g\in B(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. We set \varepsilon_n:={\varepsilon}/{2^n}, where \varepsilon\in (0,1) and {n\in \mathbb{N}}. Consider the nonincreasing number sequence \delta_n:=\min_{k=1,\dots,n}\{\delta(\varepsilon_k/2)\}, n\in \mathbb{N}.
We construct the sequence \{x_n\}_{n=1}^{\infty}\subset M_C using recursion. We take x_1\in M_C such that M\subset B^-(x_1,1+{\delta_1}/{10^4}) , and for n>1 we take x_n\in M_C such that M\subset B^-(x_n,1+{\delta_n}/{10^4}) . There exists a point y\in M such that \|(x_n+x_{n-1})/2-y|\geqslant 1-{\delta_n}/{10^4}. Setting f_{n}:=x_{n}-y and f_{n-1}:=x_{n-1}-y, we have
\begin{equation*}
\frac{x_n+x_{n-1}}2-y=\frac{f_n+f_{n-1}}2, \qquad \|f_{n}|\leqslant 1+\frac{\delta_n}{10^4}\quad\text{and} \quad \|f_{n-1}|\leqslant 1+\frac{\delta_{n-1}}{10^4}.
\end{equation*}
\notag
Next,
\begin{equation*}
\frac{1}{2}\|f_{n}|+\frac{1}{2}\|f_{n-1}|\geqslant \biggl\|\frac{f_n+f_{n-1}}2 \biggr|\geqslant 1-\frac{\delta_n}{10^4},
\end{equation*}
\notag
so that
\begin{equation*}
\|f_{n}|\geqslant 1-\frac{2\delta_n}{10^4}-\frac{\delta_{n-1}}{10^4}>1-\frac{\delta_{n-1}}{10^3}, \qquad \|f_{n-1}|\geqslant 1-\frac{2\delta_n}{10^4}-\frac{\delta_{n}}{10^4}>1-\frac{ \delta_{n-1}}{10^3}.
\end{equation*}
\notag
Let \alpha_n and \alpha_{n-1} be numbers such that f'_{n}:=\alpha_nf_n\in S and f'_{n-1}:=\alpha_{n-1}f_{n-1}\in S. Since
\begin{equation*}
1-\frac{\delta_{n-1}}{10^3}<\|f_n|,\|f_{n-1}|<1+\frac{\delta_{n-1}}{10^3},
\end{equation*}
\notag
we have
\begin{equation*}
\alpha_{n-1},\alpha_{n}\in \biggl[\biggl(1+\frac{\delta_{n-1}}{10^3}\biggr)^{-1}, \biggl(1-\frac{\delta_{n-1}}{10^3}\biggr)^{-1}\biggr].
\end{equation*}
\notag
If \alpha_n\geqslant \alpha_{n-1}, then
\begin{equation*}
\begin{aligned} \, &\biggl\|\frac{f'_{n}+f'_{n-1}}{2}\biggr| =\biggl\|\alpha_{n-1}\frac{f_{n}+f_{n-1}}{2} +\frac{\alpha_n-\alpha_{n-1}}{2}f_n\biggr| \\ &\qquad\geqslant \alpha_{n-1}\biggl\|\frac{f_{n}+f_{n-1}}{2}\biggr| -\frac{(1-{\delta_{n-1}}/10^3)^{-1}-(1+{\delta_{n-1}}/10^3)^{-1}}{2} \biggl(1+\frac{\delta_n}{10^4}\biggr) \\ &\qquad \geqslant \biggl(1+\frac{\delta_{n-1}}{10^3}\biggr)^{-1} \biggl(1-\frac{\delta_n}{10^4}\biggr) -\frac{{2\delta_{n-1}}/{10^3}}{2(1-{\delta_{n-1}}/{10^3})(1+{\delta_{n-1}}/{10^3})} \biggl(1+\frac{\delta_n}{10^4}\biggr) \\ &\qquad \geqslant\biggl(1-\frac{\delta_{n-1}}{10^3}\biggr)\biggl(1-\frac{\delta_n}{10^4}\biggr) -\frac{{\delta_{n-1}}/{10^3}}{(1-{\delta_{n-1}}/{10^3})} \\ &\qquad \geqslant 1-\frac{2 \delta_{n-1}}{10^3}-\frac{\delta_{n-1}}{10^3} \biggl(1+2\frac{\delta_{n-1}}{10^3}\biggr) \geqslant 1-\frac{\delta_{n-1}}{250}. \end{aligned}
\end{equation*}
\notag
In the case when \alpha_{n-1}\geqslant \alpha_n we obtain
\begin{equation*}
\begin{aligned} \, &\biggl\|\frac{f'_{n}+f'_{n-1}}{2}\biggr|=\biggl\|\alpha_{n}\frac{f_{n}+f_{n-1}}{2} +\frac{\alpha_{n-1}-\alpha_n}{2}f_{n-1}\biggr| \\ &\qquad \geqslant \alpha_{n}\biggl\|\frac{f_{n}+f_{n-1}}{2}\biggr| -\frac{(1-{\delta_{n-1}}/{10^3})^{-1}-(1+{\delta_{n-1}}/{10^3})^{-1}}{2} \biggl(1+\frac{\delta_{n-1}}{10^4}\biggr) \\ &\qquad \geqslant\biggl(1-\frac{\delta_{n-1}}{10^3}\biggr)\biggl(1-\frac{\delta_n}{10^4}\biggr) -\frac{{2\delta_{n-1}}/{10^3}}{2(1-{\delta_{n-1}}/{10^3})(1+{\delta_{n-1}}/{10^3})} \biggl(1+\frac{\delta_{n-1}}{10^4}\biggr) \\ &\qquad\geqslant 1-\frac{2 \delta_{n-1}}{10^3}-\frac{\delta_{n-1}}{10^3} \biggl(1+2\frac{\delta_{n-1}}{10^3}\biggr) \geqslant 1-\frac{\delta_{n-1}}{250}. \end{aligned}
\end{equation*}
\notag
We have \|(f'_{n}+f'_{n-1})/2-\frac{1}{2}f'_{n-1}|+\frac{1}{2}\|f'_n|=1. Hence, if \beta_n\geqslant 1 is such that \beta_n(f'_{n}+f'_{n-1})/2\in S, then
\begin{equation*}
0\leqslant \beta_n-1\leqslant \frac{{\delta_n}/{250}}{1-{\delta_n}/{250}}<\frac{\varepsilon_n}{4}
\end{equation*}
\notag
and
\begin{equation*}
\begin{aligned} \, \biggl\|\beta_n\frac{f'_{n}+f'_{n-1}}{2}-\frac{1}{2}f'_{n-1}\biggr| +\frac{1}{2}\|f'_n|-1 &\leqslant\biggl\|(\beta_n-1)\frac{f'_{n}+f'_{n-1}}{2}\biggr| \\ &\leqslant (\beta_n-1)\biggl(1+\frac{\delta_{n-1}}{10^4}\biggr)<\frac{1}{2}\delta_n. \end{aligned}
\end{equation*}
\notag
Since X is uniformly convex, there exists a number \widehat{\mu}_{n-1}\in[1- \varepsilon_{n-1}/2,1] such that \beta_{n}(f'_n+f'_{n-1})/2,f'_{n-1}\in B(\widehat{\mu}_{n-1}f'_{n-1},\varepsilon_{n-1}). Therefore,
\begin{equation*}
\begin{gathered} \, \biggl\|\frac{f'_n+f'_{n-1}}{2}-\frac{\widehat{\mu}_{n-1}}{\beta_n}f'_{n-1}\biggr| \leqslant \frac{\varepsilon_{n-1}}{\beta_n}\quad\text{and} \quad \|f'_n-\mu_{n-1}f'_{n-1}|\leqslant 2 \frac{\varepsilon_{n-1}}{\beta_n} \leqslant 4\varepsilon_{n-1}, \\ \text{where } \mu_{n-1}=2\frac{\widehat{\mu}_{n-1}}{\beta_n}-1\geqslant 1-\varepsilon_{n-1}, \end{gathered}
\end{equation*}
\notag
which implies that f'_{n}\in B(\mu_{n-1} f'_{n-1},4 \varepsilon_{n-1}). Hence \|f'_n-\mu_{n-1} f'_{n-1}|\leqslant 4\varepsilon_{n-1}, that is,
\begin{equation*}
f_n\in B\biggl(\frac{\mu_{n-1}}{\alpha_n} f'_{n-1},\frac{4\varepsilon_{n-1}}{\alpha_n}\biggr)= B\biggl(\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} f_{n-1},\frac{4\varepsilon_{n-1}}{\alpha_n}\biggr).
\end{equation*}
\notag
Proceeding as in the proof of Theorem 2 and taking \mu_{n-1}=1-\varepsilon_{n-1} we see that {\mu_{n-1}\alpha_{n-1}}/{\alpha_n} <1 and
\begin{equation*}
\begin{aligned} \, 0 &\leqslant 1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} =\frac{\alpha_n-(1-\varepsilon_{n-1})\alpha_{n-1}}{\alpha_n} \\ &\leqslant\frac{(1-{\delta_{n-1}}/{10^3})^{-1}-(1-\varepsilon_{n-1})(1+{\delta_{n-1}}/{10^3})^{-1}} {(1+{\delta_{n-1}}/{10^3})^{-1}} \\ &=\frac{{2\delta_{n-1}}/{10^3}}{1-{\delta_{n-1}}/{10^3}}+\varepsilon_{n-1}<2\varepsilon_{n-1}. \end{aligned}
\end{equation*}
\notag
Therefore,
\begin{equation*}
\begin{aligned} \, \frac{5\varepsilon_{n-1}}{\alpha_n} &\geqslant \biggl\|f_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} f_{n-1}\biggr| =\biggl\|x_n-y-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}(x_{n-1}-y)\biggr| \\ &=\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}- \biggl(y-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}y\biggr)\biggr| \\ &\geqslant \biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr| \biggl(1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}\biggr)\|y| \\ &\geqslant\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr| -\biggl(1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}\biggr)d, \end{aligned}
\end{equation*}
\notag
where d:=\sup_{z\in M}\|z|<+\infty. Hence
\begin{equation*}
\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}|\leqslant \varepsilon_{n-1}\biggl(\frac{5}{\alpha_n}+2d\biggr) \leqslant (8+2d)\varepsilon_{n-1}.
\end{equation*}
\notag
We set P_n:=(\prod_{k\geqslant n}{\mu_{k}\alpha_{k}}/{\alpha_{k+1}}) and \widehat{x}_n:=P_nx_n, n\in \mathbb{N}. Then
\begin{equation*}
\|\widehat{x}_{n}-\widehat{x}_{n-1}|=P_n\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr| \leqslant (8+2d)\varepsilon_{n-1}, \qquad n\in \mathbb{N},
\end{equation*}
\notag
and therefore \{\widehat{x}_n\} is a Cauchy sequence. Since X is complete, there exists a point x\in X (in the actual fact, x lies in the closure of M_C) such that \|x-\widehat{x}_n|\to 0 as n\to \infty, and therefore \|x-x_n|\leqslant \|x- \widehat{x}_n|+\|\widehat{x}_n-{x}_n|=\|x-\widehat{x}_n|+(1-P_n)\|-x_n|\leqslant \|x-\widehat{x}_n|+(1-P_n)(d+C)\to 0 as n\to \infty. Moreover, \|x-z|\leqslant \|x-x_n|+\|x_n-z|\leqslant \|x-x_n|+ 1+{\delta_{n-1}}/{10^4}\to 1=r(M,M_C) for all z\in M. Theorem 5 is proved. Remark 12. A similar analysis proves the existence of both left- and right-relative Chebyshev centres if X=(X,{\|\cdot|}) is a complete uniformly convex asymmetric space, M is a bounded subset of X with respect to the symmetrization norm, and A\subset X is a bounded closed set with respect to the symmetrization norm. In other words, for such M \subset X and A\subset X there exists a relative Chebyshev centre x\in A (a left-relative Chebyshev centre x'\in A), that is, Definition 7. A seminormed space is said to be inversely uniformly convex (or inversely right-uniformly convex) if for all \varepsilon\in (0,1) there exists \delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8}) such that for all f,g\in S the condition 0\leqslant \|f-\frac{1}{2}g|+\frac{1}{2}\|g|-\|f|<\delta(\varepsilon) implies that f \in B^-(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. The definition of an inversely left-uniformly convex space is similar; the only difference is that the last conclusion is replaced by the following one: g \in B^-(\mu f,\varepsilon) for some \mu\in [1-\varepsilon,1]. Note that if (X,{\|\cdot|}) is an inversely right-uniformly convex (inversely left-uniformly convex) linear space, then so is the mirror space (X,{\|-\cdot|}). Theorem 6. Let X=(X,{\|\cdot|}) be a left-complete inversely uniformly convex asymmetric seminormed space, and let M\subset X be a right-bounded set. Then M has a right-Chebyshev centre in X, that is, there exists a point x\in X such that Proof. The proof is similar to the proof of Theorem 5, the only difference is that the sequence \{x_n\}_{n=1}^{\infty} is taken in X, rather than in M_C.
It can be assumed without loss of generality that r(M,X)=1 and {0\in M}. Since X is inversely uniformly convex, for any \varepsilon\in (0,1) there exists a number {\delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8})} such that for all f,g\in S the condition 0\leqslant \|f-\frac{1}{2}g|+{\frac{1}{2}\|g|-\|f|<\delta(\varepsilon)} implies that f \in B^-(\mu g,\varepsilon) for some \mu\in [1-\varepsilon,1]. We set \varepsilon_n:={ }{\varepsilon}/{2^n}, where \varepsilon\in (0,1) and n\in \mathbb{N}, and choose a nonincreasing number sequence \delta_n:=\min_{k=1,\dots,n}\{\delta(\varepsilon_k/2)\}, n\in \mathbb{N}. We construct the sequence \{x_n\}_{n=1}^{\infty}\subset M_C recursively. We consider x_1\in X such that M\subset B(x_1,1+{\delta_1}/{10^4}) and then select x_n\in X so that M\subset B(x_n,1+{\delta_n}/{10^4}) for n>1. There exists a point y\in M such that \|y-(x_n+x_{n-1})/2|\geqslant 1-{\delta_n}/{10^4}. Setting f_{n}:=y -x_{n} and f_{n-1}:=y -x_{n-1} we have
\begin{equation*}
y-\frac{x_n+x_{n-1}}2=\frac{f_n+f_{n-1}}2, \qquad \|f_{n}|\leqslant 1+\frac{\delta_n}{10^4}\quad\text{and} \quad \|f_{n-1}|\leqslant 1+\frac{\delta_{n-1}}{10^4}.
\end{equation*}
\notag
Since
\begin{equation*}
\frac{1}{2}\|f_{n}|+\frac{1}{2}\|f_{n-1}|\geqslant \biggl\|\frac{f_n+f_{n-1}}2 \biggr|\geqslant 1-\frac{\delta_n}{10^4},
\end{equation*}
\notag
we have
\begin{equation*}
\|f_{n}|\geqslant 1-\frac{2\delta_n}{10^4}-\frac{\delta_{n-1}}{10^4}>1-\frac{\delta_{n-1}}{10^3}\quad\text{and} \quad \|f_{n-1}|\geqslant 1-\frac{2\delta_n}{10^4}-\frac{\delta_{n}}{10^4}>1-\frac{\delta_{n-1}}{10^3}.
\end{equation*}
\notag
Let \alpha_n and \alpha_{n-1} be such that f'_{n}:=\alpha_nf_n\in S and f'_{n-1}:=\alpha_{n-1}f_{n-1}\in S. Then
\begin{equation*}
1-\frac{\delta_{n-1}}{10^3}<\|f_n|,\|f_{n-1}|<1+\frac{\delta_{n-1}}{10^3},
\end{equation*}
\notag
and so
\begin{equation*}
\alpha_{n-1},\alpha_{n}\in \biggl[\biggl(1+\frac{\delta_{n-1}}{10^3}\biggr)^{-1}, \biggl(1-\frac{\delta_{n-1}}{10^3}\biggr)^{-1}\biggr].
\end{equation*}
\notag
Arguing as in the proof of Theorem 5 we can show that
\begin{equation*}
\biggl\|\frac{f'_{n}+f'_{n-1}}{2}\biggr| \geqslant 1-\frac{\delta_{n-1}}{250}.
\end{equation*}
\notag
We have \|(f'_{n}+f'_{n-1})/2-f'_{n-1}/2|+\|f'_n|/2=1, and so, if \beta_n\geqslant 1 is such that \beta_n(f'_{n}+f'_{n-1})/2\in S, then
\begin{equation*}
0\leqslant \beta_n-1\leqslant \frac{{\delta_n}/{250}}{1-{\delta_n}/{250}} <\frac{\varepsilon_n}{4}
\end{equation*}
\notag
and
\begin{equation*}
\begin{aligned} \, \biggl\|\beta_n\frac{f'_{n}+f'_{n-1}}{2}-\frac{1}{2}f'_{n-1}\biggr|+\frac{1}{2}\|f'_n|-1 &\leqslant \biggl\|(\beta_n-1)\frac{f'_{n}+f'_{n-1}}{2}\biggr| \\ &\leqslant (\beta_n-1)\biggl(1+\frac{\delta_{n-1}}{10^4}\biggr)<\frac{1}{2}\delta_n. \end{aligned}
\end{equation*}
\notag
Since X is inversely uniformly convex, there exists a number \widehat{\mu}_{n-1}\in[1-\varepsilon_{n-1}/2,1] such that \beta_{n} (f'_n+f'_{n-1})/2 \in B^-(\widehat{\mu}_{n-1}f'_{n-1},\varepsilon_{n-1}). Hence
\begin{equation*}
\biggl\|\frac{\widehat{\mu}_{n-1}}{\beta_n}f'_{n-1}-\frac{f'_n+f'_{n-1}}{2}\biggr|\leqslant \frac{\varepsilon_{n-1}}{\beta_n}\quad\text{and} \quad \|\mu_{n-1}f'_{n-1}-f'_n|\leqslant 2 \frac{\varepsilon_{n-1}}{\beta_n} \leqslant 4\varepsilon_{n-1},
\end{equation*}
\notag
where Thus, f'_{n}\in B^-(\mu_{n-1} f'_{n-1},4 \varepsilon_{n-1}). As a result, \|\mu_{n-1} f'_{n-1}-f'_n|\leqslant 4\varepsilon_{n-1}, that is,
\begin{equation*}
f_n\in B^-\biggl(\frac{\mu_{n-1}}{\alpha_n} f'_{n-1},\frac{4\varepsilon_{n-1}}{\alpha_n}\biggr) = B^-\biggl(\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} f_{n-1},\frac{4\varepsilon_{n-1}}{\alpha_n}\biggr).
\end{equation*}
\notag
Proceeding as in the proof of Theorem 2 and taking \mu_{n-1}=1-\varepsilon_{n-1} we see that {\mu_{n-1}\alpha_{n-1}}/{\alpha_n} <1 and
\begin{equation*}
\begin{aligned} \, 0 &\leqslant 1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} =\frac{\alpha_n-(1-\varepsilon_{n-1})\alpha_{n-1}}{\alpha_n} \\ &\leqslant\frac{(1-{\delta_{n-1}}/{10^3})^{-1} -(1-\varepsilon_{n-1})(1+{\delta_{n-1}}/{10^3})^{-1}}{(1+{\delta_{n-1}}/{10^3})^{-1}} \\ &=\frac{{2\delta_{n-1}}/{10^3}}{1-{\delta_{n-1}}/{10^3}}+\varepsilon_{n-1}<2\varepsilon_{n-1}. \end{aligned}
\end{equation*}
\notag
Therefore,
\begin{equation*}
\begin{aligned} \, \frac{5\varepsilon_{n-1}}{\alpha_n} &\geqslant \biggl\|\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n} f_{n-1}-f_n\biggr| =\biggl\|\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}(y -x_{n-1})-(y- x_n)\biggr| \\ &=\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha-_n}x_{n-1}- \biggl(y-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}y\biggr)\biggr| \\ &\geqslant \biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr| -\biggl(1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}\biggr)\|y| \\ &\geqslant\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr| -\biggl(1-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}\biggr)d, \end{aligned}
\end{equation*}
\notag
where d= \sup_{z\in M}\|z|<+\infty. Hence
\begin{equation*}
\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr|\leqslant \varepsilon_{n-1}\biggl(\frac5\alpha_n+2d\biggr)\leqslant (8+2d)\varepsilon_{n-1}.
\end{equation*}
\notag
Setting P_n:=\prod_{k\geqslant n} ({\mu_{k}\alpha_{k}}/{\alpha_{k+1}}) and \widehat{x}_n:=P_nx_n, n\in \mathbb{N}, we have
\begin{equation*}
\|\widehat{x}_{n}-\widehat{x}_{n-1}| =P_n\biggl\|x_n-\frac{\mu_{n-1}\alpha_{n-1}}{\alpha_n}x_{n-1}\biggr|\leqslant (8+2d)\varepsilon_{n-1},\qquad n\in \mathbb{N},
\end{equation*}
\notag
and therefore \{\widehat{x}_n\} is a Cauchy sequence. Since X is left-complete, there exists a point x\in X such that \|\widehat{x}_n- x|\to 0 as n\to \infty, and so \|x_n- x|\leqslant \|x_n- \widehat{x}_n|+\|\widehat{x}_n- {x}|= (1-P_n)\|x_n|+\|\widehat{x}_n- {x}| \to 0 as n\to \infty. Moreover, \| z-x|\leqslant \|z- x_n|+\|x_n-x|\leqslant \|x_n-x|+ 1+{\delta_{n-1}}/{10^4}\to 1=r(M,X) for all z\in M. Hence x is a right-Chebyshev centre in X for M. Theorem 6 is proved. Remark 13. A result similar to in Theorem 6 also holds for left-Chebyshev centres. Namely, let X=(X,{\|\cdot|}) be a left-complete inversely uniformly convex asymmetric seminormed space and M\subset X be a left-bounded set. Then there exists a left-Chebyshev centre for M in X, that is, there exists a point x\in X such that Definition 8. A set M\subset X is called left-compact if it is compact relative to the topology generated by the subbase of left-open balls. Theorem 7. Let X=(X,{\|\cdot|}) be a complete locally uniformly convex asymmetric space, M\subset X be a left-compact set and x be a left-Chebyshev centre of M. Then \|x-x_n|\to 0 as n\to\infty whenever \{x_n\}\in X is a sequence such that Proof. Assume the contrary. Passing to a subsequence if necessary we can assume that \|x-x_n|> \delta, n\in \mathbb{N}, for some \delta>0. It can also be assumed without loss of generality that r=1. Since X is locally uniformly convex, for all {g\in S} and {\varepsilon\in(0,1)} there exists \delta=\delta(\varepsilon)\in (0,{\varepsilon}/{8}) such that for any f\in S the condition 0\leqslant \|f-\frac{1}{2}g|+\frac{1}{2}\|g|-\|f|<\delta(\varepsilon) implies that \|f-g|<\varepsilon. We set \varepsilon_n:={\varepsilon}/{2^n}, n\in \mathbb{N}, for some \varepsilon\in (0,1). Consider the nonincreasing number sequence \delta_n:=\min_{k=1,\dots,n}\{\delta(\varepsilon_k/2)\}, n\in \mathbb{N}. Passing to a subsequence again we can assume that M\subset B^-(x_n,1+{\delta_n}/{10^4}) for all n . The left-norm is upper semicontinuous with respect to the topology generated by it, hence the function \|x-\cdot| assumes its maximum on any nonempty left-compact set. Therefore, there exists a point {y_n\in M} such that \|(x_n+x)/2-y_n|\geqslant 1. Since M is left-compact, there exists y\in M such that \|y-y_{n_k}|\to 0 as k\to\infty. Moreover, this subsequence contains a new subsequence such that \|y-y_{n'_k}|\leqslant {\delta_k}/{10^4}. For convenience we also denote this subsequence by \{y_k\}. Then
Setting f_{n}:=x_{n}-y and f:=x-y, we have (x_n+x)/2-y=(f_n+f)/2 , \|f_{n}|\leqslant 1+{\delta_n}/{10^4} and \|f|\leqslant 1. Let \alpha_n and \alpha be numbers such that f'_{n}:=\alpha_nf_n\in S and f' :=\alpha f \in S. We also take \beta_n\geqslant 1 such that \beta_n(f'_{n}+f')/{2}\in S. Arguing as in the proof of the previous theorem we obtain
\begin{equation*}
0\leqslant \beta_n-1\leqslant \frac{{\delta_n}/{250}}{1-{\delta_n}/{250}}<\frac{\varepsilon_n}{4}
\end{equation*}
\notag
and
\begin{equation*}
\begin{aligned} \, \biggl\|\beta_n\frac{f'_{n}+f'}{2}-\frac{1}{2}f'_n\biggr|+\frac{1}{2}\|f'_n|-1 &\leqslant \biggl\|(\beta_n-1)\frac{f'_{n}+f'}{2}\biggr| \\ &\leqslant (\beta_n-1)\biggl(1+\frac{\delta_{n-1}}{10^4}\biggr)<\frac{1}{2}\delta_n. \end{aligned}
\end{equation*}
\notag
Now, since X is locally uniformly convex, we show as in the proof of Theorem 5 that \|f'- f'_n|\leqslant 4\varepsilon_{n-1}, that is, f\in B(({1}/{\alpha}) f'_n,{4\varepsilon_{n-1}}/{\alpha})= B(({\alpha_{n}}/{\alpha})f_n,{4\varepsilon_{n}}/{\alpha}). Hence
\begin{equation*}
\biggl\|\frac{\alpha_{n}}{\alpha} f-\frac{\alpha_{n}}{\alpha} f_n\biggr|\leqslant \biggl\|\frac{\alpha_{n}}{\alpha} f-f\biggr|+ \biggl\|f-\frac{\alpha_{n}}{\alpha} f_n\biggr| \leqslant\biggl|\frac{\alpha_{n}}{\alpha}-1\biggr|\|f\|+\frac{4\varepsilon_{n}}{\alpha}\to 0, \qquad n\to\infty,
\end{equation*}
\notag
and therefore
\begin{equation*}
\|x-x_n|=\|f-f_n |=\frac{\alpha}{\alpha_{n}} \biggl\|\frac{\alpha_{n}}{\alpha} f-\frac{\alpha_{n}}{\alpha} f_n\biggr| \to 0, \qquad n\to\infty,
\end{equation*}
\notag
which contradicts the original assumption. This proves Theorem 7. § 5. Properties of sunsDefinition 9. Let \varnothing \ne M\subset X. We say that x\in X\setminus M is a solar point for M if there exists a point y\in P_Mx\ne \varnothing (called a luminosity point) such that y\in P_M((1-\lambda)y+\lambda x) for all \lambda\geqslant 0 (geometrically, this means that there is a ‘solar’ ray emanating from y and passing through x such that y is a nearest point in M to any point on the ray). We say that x\in X\setminus M is a strict solar point if P_Mx\ne\varnothing and each y\in P_Mx is a luminosity point for x. A set M is a sun (a strict sun) with respect to a set K\subset X\setminus M if any point in K is a solar (strict solar) point for M. For K= X\setminus M we say in this case that M is a sun (a strict sun, respectively). Definition 10. A set M of an asymmetric seminormed space X=(X,{\|\cdot|}) is called a \gamma-sun if for any \delta>0 the ball B(x_0,r_0-\delta), r_0=\varrho(x_0,M), can be put in some ball B(x,R) of arbitrarily large radius R which is disjoint from M. Theorem 8. Let X=(X,{\|\cdot|}) be a locally uniformly convex asymmetric space. Then any proximinal \gamma-sun M\subset X in X is a Chebyshev sun. Proof. First we show that M is a strict sun. Consider arbitrary points {x\in X\setminus M} and y\in P_Mx. It can be assumed without loss of generality that {\varrho(x,M)=1} and x=0. We claim that y is a nearest point in M to any point on the ray \ell:=\{y+t(x-y)\mid t\geqslant 0\}. Let z\in \ell and \|y-z|=R>1. Consider sufficiently small \delta\in (0,1/2). We set k=(R-\delta)/(1-\delta), a:=(1-\delta)/(R-\delta) (\geqslant a_0:={1}/(2R+1)), g':=y and y':=y(1-\delta). Let us construct a function {\varepsilon=\varepsilon(\delta)>0} that tends to zero as \delta\to 0 and such that for all \varphi\in S(0,1) the condition \|\varphi-ag'|+\|ag'|-1\leqslant \delta implies that \|\varphi-g'|<\varepsilon(\delta). Since M is a \gamma-sun, there exist z_\delta\in X and R'>R+3\delta such that B(z_\delta,R')\cap M=\varnothing and B(z_\delta,R')\supset B(0,1-\delta). Recall that the condition B(u_1,r_1)\subset B(u_2,r_2) is equivalent to \|u_1-u_2|\leqslant r_2-r_1. Let w be the point of intersection of S(0,1-\delta) and the ray \{ t(0-z_\delta)\mid t\geqslant 0\}. We have B(z_\delta,R')\supset B(z_\delta,\|w-z_\delta|)\supset B(0,1-\delta), and therefore B(z_\delta,\|w-z_\delta|)\cap M=\varnothing. Let z'\in [z_\delta,w] be a point such that R-\delta=\|w-z'|. In this case B(z_\delta,\|w-z_\delta|)\supset B(z',\|w-z'|)\supset B(0,1-\delta) and \|-z'|=R-1. Let y',y''\in [0,y] be such that \|y'|=1-\delta and y''\in S(z',R-\delta). In this case y''\in [y',y].
Let f':={\omega}/(1-\delta), f'':=ag'+(1-a)f' and \theta:=\|f''|<1. Assume that, for some null sequence \delta=\delta_n, n\in \mathbb{N}, we have \theta_n\leqslant \theta_0<1, n\in \mathbb{N}, for the corresponding quantities \theta=\theta_n (constructed from the \delta_n). Then B(f'',1-\theta_0)\subset B(0,1), and therefore
\begin{equation*}
B\bigl(ay'+(1-a)w,(1-\theta_0)(1-\delta)\bigr)=B\bigl(f''(1-\delta),(1-\theta_0)(1-\delta)\bigr) \subset B(0,1-\delta).
\end{equation*}
\notag
A homothety with centre w and ratio k yields B(y',k(1-\theta_0)(1-\delta))\subset B(z',R-\delta). However, this inclusion is impossible when k(1-\theta_0)(1-\delta)>\delta, since in this case the ball B(z',R-\delta) does not meet M, and therefore does not contain the point y.
Consider the case when \theta_n\to 1 as n\to \infty. In this case we have \|\theta_n^{-1}f'' -ag'|+\|ag'|-1\leqslant\|\theta_n^{-1}f''-f''|+\|f'' -ag'|+\|ag'|-1=(\theta_n^{-1}-1)\|f''|\leqslant \theta_n^{-1}-1. Hence \|\theta_n^{-1}f''-g'|< \varepsilon(\theta_n^{-1}-1)=:\varepsilon_n\to 0 as n\to \infty, and so
\begin{equation*}
(1-a)\|f'-g'|=\|\theta_ng'-g'|+\|f''-\theta_ng'|\leqslant (1-\theta_n)\|-g'|+\varepsilon_n\theta_n\to 0, \qquad n\to\infty.
\end{equation*}
\notag
Consequently, for the sequences y'=y'_n and w=w_n, which are constructed from the \delta_n, we have \|y-y'_n|\to 0 and \|w_n-y|\to 0 as n\to\infty. Next, for the sequence z'=z'_n , which is also constructed from the \delta_n, we have \|z-z'_n|\to 0 as n\to\infty. Hence \|q-z'_n|\leqslant \|q-z|+\|z-z'_n|\leqslant R-\sigma+\|z-z'_n|\to R-\sigma for any \sigma\in (0,R) and q such that \|q-z|\leqslant R-\sigma. Therefore, for all n exceeding some number, q lies in \mathring B(z'_n,R-\delta_n) and does not lie in M. Hence the ball B(z,R) supports the set M, and y is a nearest point to z in M. Since R is arbitrary, it follows that y is a nearest point in M to any point on the ray \ell . This proves Theorem 8. Definition 11. Let X=(X,{\|\cdot|}) be an asymmetric seminormed space and let M\subset X. We say that M is left-approximatively closed at x\in X if for each minimizing sequence \{y_n\}\subset M for x (so that \|y_n-x|\to\varrho(x,M) as n\to\infty) that converges to some y\in X (that is, \|y_n-y| \to 0 as n\to\infty) we have y\in M and {\|y-x|=\varrho(x,M)}. We say that M is left-approximatively closed if it is left-approximatively closed at any point x\in X. The definition of a right-approximatively closed set is similar. Remark 14. Each left-approximatively closed set is closed. Conversely, if a nonempty subset M of an asymmetric normed space is closed and the ball B(0,1) is closed, then M is approximatively closed (that is, M is both left- and right-approximatively closed). Theorem 9. Let X=(X,{\|\cdot|}) be a left-complete uniformly convex asymmetric seminormed space and M\subset X be a left-approximatively closed \gamma-sun. Then M is a set of existence. Proof. Let x\in X\setminus M. It can be assumed without loss of generality that \varrho(x,M)= 1 and x=0. For any sequence \{\delta_n\}\subset (0,+\infty) there exists a sequence \{R_n\}\subset (1,+\infty) such that \delta_n\to 0 and R_n\to +\infty as n\to\infty, and B(x_n,R_n)\supset B(0,1-\delta_n) for n\in \mathbb{N}. We can assume that R_n=\|-x_n|+1-\delta_n, n\in \mathbb{N}.
By Lemma 1, for the sequence \{\varepsilon_n={\varepsilon}/{2^n}\} (\varepsilon\in (0,1)) there exists a null sequence \{\delta_n\} such that \delta_n\in (0,\varepsilon_n/16) for n\in \mathbb{N}, and for an arbitrary point f_{n-1}\in M such that \|f_{n-1}|<1+\delta_{n-1} and any f \in M, \|f |<1+\delta_{n}, we have f\in B(\mu_{n-1} f_{n-1},\varepsilon_n) for some \mu_{n-1}\in[1-\varepsilon_{n-1},1]. Here, as \Delta and \delta_0 from Lemma 1 we take, respectively, \Delta_n={\|-x_n|}/{R_n} and 2\delta_n. Completing our recursive construction of the sequence \{f_n\}, we take f_n\in M such that \|f_n |<1+\delta_{n}. In this case f_n\in B(\mu_{n-1} f_{n-1},\varepsilon_{n-1}). We have
\begin{equation*}
\biggl\|\biggl(\prod_{k\geqslant n+1}\mu_k\biggr)f_{n+1}-\biggl(\prod_{k\geqslant n}\mu_k\biggr)f_n\biggr|= \biggl(\prod_{k\geqslant n+1}\mu_k\biggr)\|f_{n+1}-\mu_n f_n|\leqslant \varepsilon_{n+1}.
\end{equation*}
\notag
Hence \{\widehat{f}_n:=(\prod_{k\geqslant n}\mu_k)f_n\} is a Cauchy sequence in X and, next, since X is left-complete, there exists a point y\in X such that \|\widehat{f}_n-y|\to 0 as n\to \infty. As a result,
\begin{equation*}
\|{f}_n-y|\leqslant \|{f}_n-\widehat{f}_n|+\|\widehat{f}_n-y|=\biggl(1-\biggl(\prod_{k\geqslant n}\mu_k\biggr)\biggr)\|f_n|+\|\widehat{f}_n-y|\to 0, \qquad n\to\infty,
\end{equation*}
\notag
and, moreover, y\in M and \varrho(0,M)=\|y| since M is approximatively closed. So y is a nearest point to x in M. Now the conclusions of the theorem follows since x is arbitrary. Theorem 9 is proved. Definition 12. Let \varepsilon\geqslant 0 and M\subset X. We say that \varphi\colon X\to M is an additive (multiplicative) \varepsilon -selection (of the operator of near-best approximation) if, for all x\in X we have ( \varphi(x)\in P_M^{\varepsilon\varrho(x,M)} x, respectively). An \varepsilon -selection on E\subset X is defined similarly.Let \Xi(x,\varepsilon,\tau)=\Xi_M(x,\varepsilon,\tau) = \{\varphi\colon B(x,\tau) \to M \mid \|\varphi(y)-y| \leqslant \varrho(y,M)+\varepsilon \} be the class of additive \varepsilon-selections onto a set M defined on the ball B(x,\tau). The modulus of uniform approximative continuity (at a point x), where x\in X, is defined for \tau>0 by
\begin{equation*}
uc_x(\tau)=\inf\biggl\{ \frac{\omega_x(\varphi,\tau)}{\varrho(x,M)}+\frac{\varepsilon}{\tau}\biggm| \varepsilon>0,\,\varphi\in \Xi(x,\varepsilon,\tau) \biggr\},
\end{equation*}
\notag
and the oscillation of a mapping \varphi at a point x with step \tau is defined by \omega_x(\varphi,\tau)=\sup_{\|x-y|\leqslant \tau}\|\varphi(y)-\varphi(x)|. We recall the definition of a \delta-sun (see [27]). Definition 13. Let M be a nonempty subset of an asymmetric seminormed space X=(X,{\|\cdot|})). A point x_0\in X such that \varrho(x_0,M)>0 is called a \delta-solar point for M if there exists a sequence of points \{x_n\}, \|x_0-x_n|\to 0 as n\to\infty, such that
\begin{equation*}
\frac{\varrho(x_n,M)-\varrho(x_0,M)}{ \|x_0-x_n|}\to 1, \qquad n\to\infty.
\end{equation*}
\notag
A set M is called a \delta-sun if any x \in X satisfying \varrho(x ,M)>0 is a \delta-solar point for M.
\begin{equation*}
\delta_x^r(\tau):=\inf\biggl\{\frac{\|v'-x|-\varrho(v ,M) }{\|x -v|}\biggm| v'\in M,\, v\in X\colon \|x -v|=\tau,\, x\in [v',v]\biggr\}.
\end{equation*}
\notag
A point x\in X\setminus M is a radial \delta-solar point for M if there exists a sequence \{\tau_n\} such that \tau_n\to 0+ as n\to\infty and \delta_x^r(\tau_n) = o(1) as n\to\infty. Theorem 10. Let M be a nonempty subset of an asymmetric seminormed space X=(X,{\|\cdot|}). Assume that uc_x(\tau)=o(1) as \tau\to0+. Then \delta_x^r(\tau)=o(1) as {\tau\to0+}, that is, x\in X\setminus M is a radial \delta-solar point for M. Proof. It can be assumed without loss of generality that \varrho(x,M)=1 and \tau\in (0,1). Indeed, the condition that uc_x(\tau)=o(1) as \tau\to0+ implies that there exists a map \varphi=\varphi_{x,\varepsilon,2\tau}\in \Xi(x,\varepsilon,2\tau) such that \omega_x(\varphi,2\tau)=o(1) and \varepsilon=o(\tau) as \tau\to0+. Let
\begin{equation*}
y=\varphi(x), \qquad v=x+\tau\frac{x-y}{\|y-x|}, \qquad v'=\varphi(v)\quad\text{and} \quad v''=x+\tau\frac{x-v'}{\|y-x|}.
\end{equation*}
\notag
Then \|v'-y|=\|\varphi(v)-\varphi(x)|\leqslant \omega_x(\varphi,\tau)=o(1) ,
\begin{equation*}
\biggl|\frac{\|v'-x|}{\|y-x|}-1\biggr|=\biggl|\frac{\|v'-x|-\|y-x|}{\|y-x|}\biggr| \leqslant\biggl|\frac{v'-x}{\|y-x|}-\frac{y-x}{\|y-x|}\biggr|=\frac{\|v'-y|}{\|y-x|}=o(1)
\end{equation*}
\notag
and
\begin{equation*}
\|v-v''|=\tau\biggl\|\frac{x-v'}{\|y-x|}-\frac{x-y}{\|y-x|}\biggr\|=o(\tau), \qquad \tau\to0+.
\end{equation*}
\notag
Since
\begin{equation*}
\varrho(v,M)\leqslant \|v'-v|\leqslant \varrho(v,M)+\varepsilon=\varrho(v,M)+o(\tau)
\end{equation*}
\notag
we have the asymptotic equality Next,
\begin{equation*}
\|v'-v''|-\|v' -v| \leqslant \|v-v'' | =o(\tau)\quad\text{and} \quad \varrho(v,M)-\varrho(v'',M) \leqslant \|v''-v \| =o(\tau),
\end{equation*}
\notag
hence
\begin{equation*}
\begin{aligned} \, \|v'-x|&=\|v'-v''|-\|x-v''|=\|v'-v''|-\tau\frac{\|v'-x|}{\|y-x|} =\|v'-v''|-\tau(1+o(1)) \\ &= \varrho(v,M)+o(\tau)-\tau+o(\tau)\leqslant \varrho(v'',M)+o(\tau)-\tau. \end{aligned}
\end{equation*}
\notag
Now,
\begin{equation*}
\varrho(v'',M)-\|v'-x|\leqslant \|v'-v''|-\|v'-x|=\|x-v''|=\tau\frac{\|x-v'|}{\|y-x|}=\tau(1+o(1)),
\end{equation*}
\notag
and so we have
\begin{equation*}
\frac{\varrho(v'',M)-\|v'-x|}{\tau}\to 1, \qquad \tau\to 0+.
\end{equation*}
\notag
Hence \delta_x^r(\tau)\to 0, that is, x is a radial \delta-solar point for M. This proves Theorem 10. Remark 15. If there exists a sequence \{\tau_n\} such that \tau_n \to 0+ as n \to \infty and uc_x(\tau_n)=o(1) as n\to \infty, then there exists a sequence \{\tau'_n\} such that \tau'_n\to 0+ as {n\to\infty} and \delta_x^r(\tau'_n)=o(1) as n\to\infty. Hence x\in X\setminus M is a radial \delta-solar point for M. Remark 16. Since \varrho(v'',M)\leqslant \varrho(x,M)+\|x-v''|=\varrho(x,M)+\tau, we have
\begin{equation*}
\frac{\varrho(v'',M)-\|v'-x|}{\tau}\leqslant \frac{\varrho(v'',M)-\varrho(x,M)}{\tau}\leqslant 1,
\end{equation*}
\notag
and so the radial \delta-solarity of M implies that M is a \delta-sun. Remark 17. Alimov proved that in a left-complete asymmetric normed space any \delta-sun with continuous distance function is a \gamma-sun. It follows from this result and Theorem 9 that in a left-complete uniformly convex asymmetric space any left-approximatively closed \delta-sun with continuous distance function is an existence set, which, moreover, is a Chebyshev sun in view of Theorem 8. 
Citation in format AMSBIB
\Bibitem{Tsa22}
This publication is cited in the following 6 articles:
|
|
||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Contact us: |
Terms of Use
|
Registration to the website |
Logotypes |
|