prompt
stringlengths
0
256
answer
stringlengths
1
256
=1$ with the optimal constant $d=3\cdot 2^{n-1}$, see \cite{BS1}. In the early 2000s the Finiteness Principle was proved by C.~Fefferman for all $k$ and $n$ for regular moduli of continuity $\omega$ (i.e., $\omega(1)=1$), see \cite{F1}. The upper bound for
the constant $d$ in the Fefferman proof was reduced later to $d=2^{ k+n \choose k}$ by Bierstone and P. Milman \cite{BM} and independently and by a different method by Shvartsman \cite{S}. The obtained results (and the Finiteness Principle in general) ad
mit the following reformulation in terms of geometric characteristics of closed unit balls $B_b^{k,\omega}(S)$ of $G_b^{k,\omega}(S)$. Specifically, let $B_b^{k,\omega}(S;m)\subset B_b^{k,\omega}(S)$, $m\in{\mbf N}$, be the balanced closed convex hull of t
he union of all finite-dimensional balls $B_b^{k,\omega}(S')\subset G_b^{k,\omega}(S')$, $S'\subset S$, ${\rm card}\, S'\le m$. \begin{Th}\label{teo1.5} There exist constants $d\in{\mbf N}$ and $c\in (1,\infty)$ such that \[ B_b^{k,\omega}(S;d)\subset B_
b^{k,\omega}(S)\subset c\cdot B_b^{k,\omega}(S;d). \] Here for $k=0$, $d=2$ (-\,optimal) and $c=1$, for $n=1$, $d=k+2$ (-\,optimal) and $c$ depends on $k$ only, for $k=1$, $d=3\cdot 2^{n-1}$ (-\,optimal) and $c$ depends on $k$ and $n$ only, and for $k\ge 2
$, $d=2^{ k+n \choose k}$ and $c=\frac{\tilde c}{\omega(1)}$, where $\tilde c$ depends on $k$ and $n$ only. \end{Th} \subsection{Complementability of Spaces ${\mathbf G_b^{k,\omega}(S)}$} We begin with a result describing bounded linear operators on $G_b^
{k,\omega}({\mbf R}^n)$. To this end, for a Banach space $X$ by $C_b^{k,\omega}({\mbf R}^n;X)$ we denote the Banach space of $X$-valued $C^k$ functions on ${\mbf R}^n$ with norm defined similarly to that of Definition \ref{def1} with absolute values replac
ed by norms $\|\cdot\|_X$ in $X$. Let $\mathcal L\bigl(X_1;X_2\bigr)$ stand for the Banach space of bounded linear operators between Banach spaces $X_1$ and $X_2$ equipped with the operator norm. \begin{Th}\label{te1.6} The restriction map to the set $\{\d
elta_x^0\, :\, x\in {\mbf R}^n\}\subset G_b^{k,\omega}({\mbf R}^n)$ determines an isometric isomorphism between $\mathcal L\bigl(G_b^{k,\omega}({\mbf R}^n);X\bigr)$ and $C_b^{k,\omega}({\mbf R}^n;X)$. \end{Th} Let $q_S: C_b^{k,\omega}({\mbf R}^n)\rightar
row C_b^{k,\omega}(S)$ be the quotient map induced by the restriction of functions on ${\mbf R}^n$ to $S$. A right inverse $T\in {\mathcal L}(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ for $q_S$ (i.e., $q_S\circ T={\rm id}$) is called a {\em linear ex
tension operator}. The set of such operators is denoted by $Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$. \begin{D}\label{def1.5} An operator $T\in Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ has depth $d\in{\mbf N}$ if for all $x\in{\mbf R}
^n$ and $f\in C_b^{k,\omega}(S)$, \begin{equation}\label{equ1.6} (Tf)(x)=\sum_{i=1}^d \lambda_i^x\cdot f(y_i^x), \end{equation} where $y_i^x\in S$ and $\lambda_i^x$ depend only on $x$. \end{D} Linear extension operators of finite depth exist. For $k=0$ (
the Lipschitz case) the Whitney-Glaeser linear extension operators $C_{b}^{0,\omega}(S)\rightarrow C_{b}^{0,\omega}({\mbf R}^n)$, see \cite{Gl}, have depth $d$ depending on $n$ only and norms bounded by a constant depending on $n$ only. In the 1990s bound
ed linear extension operators $C_{b}^{1,\omega}(S)\rightarrow C_{b}^{1,\omega}({\mbf R}^n)$ of depth $d$ depending on $n$ only with norms bounded by a constant depending on $n$ only were constructed by Yu.~Brudnyi and Shvartsman \cite{BS2}. Recently bound
ed linear extensions operators of depth $d$ depending on $k$ and $n$ only were constructed by Luli \cite{Lu} for all spaces $C_b^{k,\omega}(S)$; their norms are bounded by $\frac{C}{\omega(1)}$, where $C\in (1,\infty)$ is a constant depending on $k$ and $
n$ only. (Earlier such extension operators were constructed for finite sets $S$ by C.~Fefferman \cite[Th.\,8]{F2}.) In the following result we identify $(G_b^{k,\omega}(S))^*$ with $C_b^{k,\omega}(S)$ by means of the isometric isomorphism of Theorem \ref
{te1.2}. \begin{Th}\label{teo1.6} For each $T\in Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ of finite depth there exists a bounded linear projection $P:G_b^{k,\omega}({\mbf R}^n)\rightarrow G_b^{k,\omega}(S)$ whose adjoint $P^*=T$. \end{Th} \begin
{R}\label{rem1.7} {\rm It is easily seen that if $T$ has depth $d$ and is defined by \eqref{equ1.6}, then \[ p(x):=P(\delta_x^0)=\sum_{i=1}^d \lambda_i^x\cdot \delta_{y_i^x}^0 \quad {\rm for\ all}\quad x\in{\mbf R}^n. \] Moreover, $p\in C_b^{k,\omega}({\m
bf R}^n; G_b^{k,\omega}(S))$ and has norm equal to $\|T\|$ by Theorem \ref{te1.6}. } \end{R} \subsection{Approximation Property} Recall that a Banach space $X$ is said to have the {\em approximation property}, if, for every compact set $K\subset X$ and e
very $\varepsilon > 0$, there exists an operator $T : X\to X$ of finite rank so that $\|Tx-x\|\le\varepsilon$ for every $x\in K$. Although it is strongly believed that the class of spaces with the approximation property includes practically all spaces wh
ich appear naturally in analysis, it is not known yet even for the space $H^\infty$ of bounded holomorphic functions on the open unit disk. The first example of a space which fails to have the approximation property was constructed by Enflo \cite{E}. Sinc
e Enflo's work several other examples of such spaces were constructed, for the references see, e.g., \cite{L}. A Banach space has the $\lambda$-{\em approximation property}, $1\le\lambda<\infty$, if it has the approximation property with the approximatin
g finite rank operators of norm $\le\lambda$. A Banach space is said to have the {\em bounded approximation property}, if it has the $\lambda$-approximation property for some $\lambda$. If $\lambda=1$, then the space is said to have the {\em metric approxi
mation property}. Every Banach spaces with a basis has the bounded approximation property. Also, it is known that the approximation property does not imply the bounded approximation property, see \cite{FJ}. It was established by Pe\l czy\'nski \cite{P}
that a separable Banach space has the bounded approximation property if and only if it is isomorphic to a complemented subspace of a separable Banach space with a basis. Next, for Banach spaces $X,Y$ by ${\mathcal F}(X,Y)\subset {\mathcal L}(X,Y)$ we den
ote the subspace of linear bounded operators of finite rank $X\to Y$. Let us consider the trace mapping $V$ from the projective tensor product $Y^*\hat{\otimes}_\pi X\to {\mathcal F}(X,Y)^*$ defined by \[ (Vu)(T)={\rm trace}(Tu),\quad\text{where}\quad u\in
Y^*\hat{\otimes}_\pi X,\ T\in {\mathcal F}(X,Y), \] that is, if $u=\sum_{n=1}^\infty y_n^*\otimes x_n$, then $(Vu)(T)=\sum_{n=1}^\infty y_n^*(Tx_n)$. It is easy to see that $\|Vu\|\le \|u\|_\pi$. The $\lambda$-bounded approximation property of $X$ is eq
uivalent to the fact that $\|u\|_\pi\le\lambda\|Vu\|$ for all Banach spaces $Y$. This well-known result (see, e.g., \cite[page 193]{DF}) is essentially due to Grothendieck \cite{G}. Our result concerning spaces $G_b^{k,\omega}(S)$ reads as follows. \begi
n{Th}\label{te1.3} \begin{enumerate} \item Spaces $G_b^{k,\omega}({\mbf R}^n)$ have the $\lambda$-approximation property with \penalty-10000 $\displaystyle \lambda=\lambda(k,n,\omega):=1+C\cdot\lim_{t\rightarrow\infty}\,\mbox{$\frac{1}{\omega(t)}$}$, wher
e $C$ depends on $k$ and $n$ only. \item All the other spaces $G_b^{k,\omega}(S)$ have the $\lambda$-approximation property with \penalty-10000 $\lambda= C'\cdot\lambda(1,n,\omega)$, where $C'$ is a constant depending on $n$ only, if $k=0,1$, and with $\la
mbda=\frac{C''\cdot \lambda(k,n,\omega)}{\omega(1)}$, where $C''$ is a constant depending on $k$ and $n$ only, if $k\ge 2$. \end{enumerate} \end{Th} If $\lim_{t\rightarrow\infty}\omega(t)=\infty$, then (1) implies that the corresponding space $G_b^{k,\omeg
a}({\mbf R}^n)$ has the metric approximation property. In case $\lim_{t\rightarrow\infty}\omega(t)<\infty$, one can define the new modulus of continuity $\widetilde\omega$ (cf. properties (i) and (ii) in its definition) by the formula \[ \widetilde\omega(t
)=\max\{\omega(t),t\},\quad t\in (0,\infty). \] It is easily seen that spaces $C_b^{k,\omega}({\mbf R}^n)$ and $C_b^{k,\widetilde\omega}({\mbf R}^n)$ are isomorphic. Thus $G_b^{k,\omega}({\mbf R}^n)$ is isomorphic to space $G_b^{k,\widetilde\omega}({\mbf
R}^n)$ having the metric approximation property. However, the distortion of this isomorphism depends on $\omega$. So, in general, it is not clear whether $G_b^{k,\omega}({\mbf R}^n)$ itself has the metric approximation property. In fact, in some cases sp
aces $G_b^{k,\omega}(S)$ still have the metric approximation property. E.g., by the classical result of Grothendieck \cite[Ch.\,I]{G}, separable dual spaces with the approximation property have the metric approximation property. The class of such spaces $G
_b^{k,\omega}(S)$ is studied in the next section. \begin{R}\label{k} {\rm It is not known, even for the case $k=0$, whether all spaces $C_b^{k,\omega}({\mbf R}^n)$ have the approximation property (for some results in this direction for $k=0$ see, e.g., \ci
te{K}). } \end{R} At the end of this section we formulate a result describing the structure of operators in ${\mathcal L}(G_b^{k,\omega}({\mbf R}^n);X)$, where $X$ is a separable Banach space with the $\lambda$-approximation property. In particular, it can
be applied to $X=G_b^{k,\omega}(S)$ and $\lambda:=\lambda(S,k,n,\omega)$ the constant of the approximation property for $G_b^{k,\omega}(S)$ of Theorem \ref{te1.3}\,(2). \begin{Th}\label{teor1.10} There exists the family of norm one vectors $\{v_j\}_{j\in{
\mbf N}}\subset X$ and given $H\in {\mathcal L}(G_b^{k,\omega}({\mbf R}^n);X)$ the family of functions $\{h_j\}_{j\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ of norms $\le 32\cdot\lambda^2\cdot\|H\|$ such that for all $x\in{\mbf R}^n$, $\alpha\in{\mb
f Z}_+^n$, $|\alpha|\le k$, \begin{equation}\label{equa1.7} H(\delta_x^\alpha)=\sum_{j=1}^\infty D^\alpha h_j(x)\cdot v_j \end{equation} (convergence in $X$). \end{Th} \begin{R} {\rm If $X=G_b^{k,\omega}(S)$ and $H\in\mathcal L(G_b^{k,\omega}({\mbf R}^n);G
_b^{k,\omega}(S))$ is a projection onto $G_b^{k,\omega}(S)$, then in addition to \eqref{equa1.7} we have \begin{equation}\label{equa1.8} \delta_x^0=\sum_{j=1}^\infty h_j(x)\cdot v_j\quad {\rm for\ all}\quad x\in S. \end{equation} In this case, the adjoint
$H^*$ of $H$ belongs to $Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ and for all $x\in{\mbf R}^n$, $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, the extension $H^*f$ of $f\in C_b^{k,\omega}(S)$ satisfies \begin{equation}\label{equa1.9} D^\alpha (H^*f)
(x):=\sum_{j=1}^\infty D^\alpha h_j(x)\cdot f(v_j). \end{equation} } \end{R} \subsection{Preduals of ${\mathbf G_b^{k,\omega}(S)}$ Spaces} Let $C_{0}^{k,\omega}({\mbf R}^n)$ be the subspace of functions $f\in C_b^{k,\omega}({\mbf R}^n)$ such that \begin{
itemize} \item[(i)] for all $\alpha\in{\mbf Z}_+^n$, $0\le |\alpha|\le k$, \[ \lim_{\|x\|\rightarrow\infty}D^\alpha f(x)=0; \] \item[(ii)] for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|=k$, \[ \lim_{\|x-y\|\rightarrow 0}\frac{D^\alpha f(x)-D^\alpha f(y)}{\omeg
a(\|x-y\|)}=0. \] \end{itemize} It is easily seen that $C^{k,\omega}_0({\mbf R}^n)$ equipped with the norm induced from $C^{k,\omega}_b({\mbf R}^n)$ is a Banach space. By $C^{k,\omega}_0(S)$ we denote the trace of $C^{k,\omega}_0({\mbf R}^n)$ to a closed
subset $S\subset{\mbf R}^n$ equipped with the trace norm. If $\lim_{t\rightarrow 0^+}\,\frac{t}{\omega(t)}> 0$ (see condition (i) for $\omega$ in section~1.1), then clearly, the corresponding space $C^{k,\omega}_0({\mbf R}^n)$ is trivial. Thus we may nat
urally assume that $\omega$ satisfies the condition \begin{equation}\label{omega2} \lim_{t\rightarrow 0^+}\,\frac{t}{\omega(t)}=0. \end{equation} In the sequel, the weak$^*$ topology of $C_b^{k,\omega}(S)$ is defined by means of functionals in $G_b^{k,\om
ega}(S)\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$. Convergence in the weak$^*$ topology is described in section~4.2. \begin{Th}\label{te1.4} Suppose $\omega$ satisfies condition \eqref{omega2}. \begin{enumerate} \item Space $\bigl(C^{k,\omega}_0(
{\mbf R}^n)\bigr)^*$ is isomorphic to $G_b^{k,\omega}({\mbf R}^n)$, isometrically if $\displaystyle \lim_{t\rightarrow\infty}\omega(t)=\infty$. \item If there exists a weak$^*$ continuous operator $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ s
uch that $T\bigl(C_0^{k,\omega}(S)\bigr)\subset C_0^{k,\omega}({\mbf R}^n)$, then $\bigl(C^{k,\omega}_0(S)\bigr)^*$ is isomorphic to $G_b^{k,\omega}(S)$. \end{enumerate} \end{Th} From the first part of the theorem we obtain (for $\omega$ satisfying \eqref{
omega2}): \begin{C}\label{cor1.10} The space of $C^\infty$ functions with compact supports on ${\mbf R}^n$ is dense in $C^{k,\omega}_0({\mbf R}^n)$. In particular, all spaces $C^{k,\omega}_0(S)$ are separable. \end{C} It is not clear whether the condition
of the second part of the theorem is valid for all spaces $C_b^{k,\omega}(S)$ with $\omega$ subject to \eqref{omega2}. Here we describe a class of sets $S$ satisfying this condition. As before, by $\mathcal P_{k,n}$ we denote the space of real polynomials
on ${\mbf R}^n$ of degree $k$, and by $Q_r(x)\subset {\mbf R}^n$ the closed cube centered at $x$ of sidelength $2r$. \begin{D}\label{wm} A point $x$ of a subset $S\subset{\mbf R}^n$ is said to be weak $k$-Markov if \[ \varliminf_{r\rightarrow 0}\left\{\s
up_{p\in\mathcal P_{k,n}\setminus 0}\left(\frac{\sup_{Q_r(x)}|p|}{\sup_{Q_r(x)\cap S}|p|} \right) \right\}<\infty . \] A closed set $S\subset{\mbf R}^n$ is said to be weak $k$-Markov if it contains a dense subset of weak $k$-Markov points. \end{D} The
class of weak $k$-Markov sets, denoted by ${\rm Mar}^*_k({\mbf R}^n)$, was introduced and studied by Yu.~Brudnyi and the author, see \cite{BB1, B}. It contains, in particular, the closure of any open set, the Ahlfors $p$-regular compact subsets of ${\mbf
R}^n$ with $p > n-1$, a wide class of fractals of arbitrary positive Hausdorff measure, direct products $\prod_{j=1}^l S_j$, where $S_j\in {\rm Mar}^*_k({\mbf R}^{n_j})$, $1\le j\le l$, $n=\sum_{j=1}^l n_j$, and closures of unions of any combination of su
ch sets. Solutions of the Whitney problems (see sections 1.2 and 1.3 above) for sets in ${\rm Mar}^*_k({\mbf R}^n)$ are relatively simple, see \cite{BB1}. We prove the following result. \begin{Th}\label{te1.11} Let $S'\in {\rm Mar}^*_k({\mbf R}^n)$ and
$\omega$ satisfy \eqref{omega2}. Suppose $H:{\mbf R}^n\rightarrow{\mbf R}^n$ is a differentiable map such that \begin{itemize} \item[(a)] the entries of its Jacobian matrix belong to $C_b^{k-1,\omega_o}({\mbf R}^n)$, where $\omega_o$ satisfies \begin{equ
ation}\label{equ1.8} \lim_{t\rightarrow 0^+}\frac{\omega_o(t)}{\omega(t)}=0; \end{equation} \item[(b)] the map $H|_{S'}:S'\rightarrow S=:H(S')$ is a proper retraction.\footnote{I.e., $S\subset S'$ and $H|_{S'}(x)=x$ for all $x\in S$, and for each compact
$K\subset S$ its preimage $(H|_{S'})^{-1}(K)$ is compact.} \end{itemize} Then the condition of Theorem \ref{te1.4} holds for $C_b^{k,\omega}(S)$. Thus $G_b^{k,\omega}(S)$ is isomorphic to $\bigl(C^{k,\omega}_0(S)\bigr)^*$ and so $G_b^{k,\omega}(S)$ and $
C^{k,\omega}_0(S)$ have the metric approximation property. \end{Th} \begin{R} {\rm (1) In addition to weak $k$-Markov sets $S\subset{\mbf R}^n$, Theorem \ref{te1.11} is valid, e.g., for a compact subset $S$ of a $C^{k+1}$-manifold $M\subset{\mbf R}^n
$ such that the base of the topology of $S$ consists of relatively open subsets of Hausdorff dimension $> {\rm dim}\,M - 1$. Indeed, in this case there exist tubular open neighbourhoods $U_M\subset V_M\subset{\mbf R}^n$ of $M$ such that ${\rm cl}(U_M)\sub
set V_M$ together with a $C^{k+1}$ retraction $r: U_M\rightarrow M$. Then, due to the hypothesis for $S$, the base of topology of $S':=r^{-1}(S)\cap {\rm cl}(U_M) $ consists of relatively open subsets of Hausdorff dimension $>n-1$ and so $S'\in {\rm Mar}_
p^*({\mbf R}^n)$ for all $p\in{\mbf N}$, see, e.g., \cite[page\,536]{B}. Moreover, it is easily seen that $r|_{S'}$ is the restriction to $S'$ of a map $H\in C_b^{k+1}({\mbf R}^n; {\mbf R}^n)$. Decreasing $V_M$, if necessary, we may assume that $S'$ is co
mpact, and so the triple $(H, S', S)$ satisfies the hypothesis of the theorem. \noindent (2) Under conditions of Theorem \ref{te1.11}, $C_b^{k,\omega}(S)$ is isomorphic to the second dual of $C^{k,\omega}_0(S)$.} \end{R} \section{Proof of Theorem \ref
{te1.2}} By $\delta_x^\alpha$, $x\in{\mbf R}^n$, $\alpha\in\mathbb Z^n_+$, we denote the evaluation functional $D^\alpha|_{\{x\}}$. By definition each $\delta_x^\alpha$, $|\alpha|\le k$, belongs to $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ and has norm $\
le 1$. Similarly, functionals $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y\in{\mbf R}^n$, $x\ne y$, belong to $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ and have norm $\le 1$. \begin{Proposition}\label{p2.1} The closed unit
ball $B$ of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ is the balanced weak$\,^*$ closed convex hull of the set $V$ of all functionals $\delta_x^\alpha$, $|\alpha|\le k$, and $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y\in
{\mbf R}^n$, $x\ne y$. \end{Proposition} \begin{proof} Clearly, $V\subset B$ and therefore the required hull $\widehat V\subset B$ as well. Assume, on the contrary, that $\widehat V\ne B$. Then due to the Hahn-Banach theorem there exists an element $f\in C
^{k,\omega}_b({\mbf R}^n)$ of norm one such that $\sup_{v\in\widehat V}|v(f)|\le c<1$. Since $V\subset \widehat V$, this implies \[ \|f\|_{C^{k,\omega}_b({\mbf R}^n)}\le c<1, \] a contradiction proving the result. \end{proof} Let $X$ be the minimal closed
subspace of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ containing $V$. \begin{Proposition}\label{p2.2} $X^*$ is isometrically isomorphic to $C^{k,\omega}_b({\mbf R}^n)$. \end{Proposition} \begin{proof} For $h\in X^*$ we set $H(x):=h(\delta^0_x)$, $x\in\mat
hbb R^n$. Let $e_1,\dots, e_n$ be the standard orthonormal basis in ${\mbf R}^n$. By the mean-value theorem for functions in $C^{k,\omega}_b({\mbf R}^n)$ we obtain, for all $\alpha\in\mathbb Z^n_+$, $|\alpha|<k$, $x\in{\mbf R}^n$, \begin{equation}\label{eq
2.6} \lim_{t\rightarrow 0}\frac{\delta^{\alpha}_{x+t\cdot e_i}-\delta^\alpha_x}{t}=\delta^{\alpha+e_i}_x \end{equation} (convergence in $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*)$. From here by induction we deduce easily that $H\in C^k({\mbf R}^n)$ and fo
r all $\alpha\in\mathbb Z^n_+$, $|\alpha|\le k$, $x\in{\mbf R}^n$, \[ h(\delta^\alpha_x)=D^\alpha H(x). \] This shows that $H\in C^{k,\omega}_b({\mbf R}^n)$ and $\|H\|_{C^{k,\omega}_b({\mbf R}^n)}\le \|h\|_{X^*}$. Considering $H$ as the bounded linear func
tional on $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ we obtain that $H|_V=h|_V$. Thus, by the definition of $X$, \[ H|_{X}=h. \] Since the unit ball of $X$ is $B\cap X$, \[ \|h\|_{X^*}\le \|H\|_{C^{k,\omega}_b({\mbf R}^n)}\, \bigl(\le \|h\|_{X^*}\bigr). \
] Hence, the correspondence $h\mapsto H$ determines an isometry $I :X^*\rightarrow C_b^{k,\omega}({\mbf R}^n)$. Since the restriction of each $H\in C_b^{k,\omega}({\mbf R}^n)$, regarded as the bounded linear functional on $\bigl(C^{k,\omega}_b({\mbf R}^n)
\bigr)^*$, to $X$ determines some $h\in X^*$, map $I$ is surjective. This completes the proof of the proposition. \end{proof} Note that equation \eqref{eq2.6} shows that the minimal closed subspace $G_b^{k,\omega}({\mbf R}^n)\subset\bigl(C_b^{k,\omega}({\
mbf R}^n)\bigr)^*$ containing all $\delta_x^0$, $x\in{\mbf R}^n$, coincides with $X$. Thus, $\bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ is isometrically isomorphic to $C^{k,\omega}_b({\mbf R}^n)$; this proves Theorem \ref{te1.2} for $S={\mbf R}^n$. \begin{
C}\label{cor2.3} The closed unit ball of $G^{k,\omega}_b({\mbf R}^n)$ is the balanced closed convex hull of the set $V$ of all functionals $\delta_x^\alpha$, $|\alpha|\le k$, and $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y
\in{\mbf R}^n$, $x\ne y$. \end{C} \begin{proof} The closed unit ball of $G^{k,\omega}_b({\mbf R}^n)$ is $B\cap G^{k,\omega}_b({\mbf R}^n)$. Since the weak$^*$ topology of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ induces the weak topology of $G^{k,\omega}
_b({\mbf R}^n)$ and the weak closure of the balanced convex hull of $V$ coincides with the norm closure of this set, the result follows from Proposition \ref{p2.1}. \end{proof} Now, let us consider the case of general $S\subset\mathbb R^n$. Let $h\in \bigl
(G_b^{k,\omega}(S)\bigr)^*$. We set $H(x):=h(\delta_x^0)$, $x\in S$. Due to the Hahn-Banach theorem, there exists $\tilde h\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $\tilde h|_{G_b^{k,\omega}(S)}=h$ and $\|\tilde h\|_{(G_b^{k,\omega}({\mbf R}
^n))^*}=\|h\|_{(G_b^{k,\omega}(S))^*}$. Let us define $\widetilde H(x)=\tilde h(\delta_x^0)$, $x\in {\mbf R}^n$. According to Proposition \ref{p2.2}, $\widetilde H\in C^{k,\omega}_b({\mbf R}^n)$ and $\|\widetilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\tilde
h\|_{(G_b^{k,\omega}({\mbf R}^n))^*}$. Moreover, $\widetilde H|_S=H$. This implies that $H\in C_b^{k,\omega}(S)$ and has norm $\le \|h\|_{(G_b^{k,\omega}(S))^*}$. Hence, the correspondence $h\mapsto H$ determines a bounded nonincreasing norm linear injecti
on $I_S:\bigl(G_b^{k,\omega}(S)\bigr)^*\rightarrow C_b^{k,\omega}(S)$. Let us show that $I_S$ is a surjective isometry. Indeed, for $H\in C_b^{k,\omega}(S)$ there exists $\widetilde H\in C_b^{k,\omega}({\mbf R}^n)$ such that $\widetilde H|_S=H$ and $\|\wid
etilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|H\|_{C_b^{k,\omega}(S)}$. In turn, due to Proposition \ref{p2.2} there exists $\tilde h\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $\widetilde H(x)=\tilde h(\delta_x^0)$, $x\in{\mbf R}^n$, and $\|\wide
tilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\tilde h\|_{(G_b^{k,\omega}({\mbf R}^n))^*}$. We set $h:=\tilde h|_{G_b^{k,\omega}(S)}$. Then $h\in \bigl(G_b^{k,\omega}(S)\bigr)^*$ and $H(x)=h(\delta_x^0)$, $x\in S$, i.e., $I_S(h)=H$ and \[ \bigl(\|h\|_{(G_b^{k,
\omega}(S))^*}\ge\bigr)\, \|I_S(h)\|_{C_b^{k,\omega}(S)}\ge \|h\|_{(G_b^{k,\omega}(S))^*}. \] The proof of Theorem \ref{te1.2} is complete. \section{Proofs of Theorems \ref{teo1.5}, \ref{te1.6}, \ref{teo1.6}} \subsection{Proof of Theorem \ref{teo1.5}} \
begin{proof} According to the Finiteness Principle there exist constants $d\in{\mbf N}$ and $c\in (1,\infty)$ such that for all $f\in C_b^{k,\omega}(S)$, \begin{equation}\label{e3.13} \sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\
le\|f\|_{C_b^{k,\omega}(S)}\le c\cdot\left(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\right). \end{equation} Here for $k=0$, $d=2$ (-\,optimal) and $c=1$, see \cite{McS}, for $n=1$, $d=k+2$ (-\,optimal) and $c$ depends on $k$ onl
y, see \cite{M}, for $k=1$, $d=3\cdot 2^{n-1}$ (-\,optimal) and $c$ depends on $k$ and $n$ only, see \cite{BS1}, and for $k\ge 2$, $d=2^{ k+n \choose k}$ and $c=\frac{\tilde c}{\omega(1)}$, where $\tilde c$ depends on $k$ and $n$ only, see \cite{F1} and \c
ite{BM}, \cite{S}. Considering $f$ as the bounded linear functional on $G_b^{k,\omega}(S)$, we get from \eqref{e3.13} the required implications \[ B_b^{k,\omega}(S;d)\subset B_b^{k,\omega}(S)\subset c\cdot B_b^{k,\omega}(S;d). \] Indeed, suppose, on the c
ontrary, that there exists $v\in B_b^{k,\omega}(S)\setminus c\cdot B_b^{k,\omega}(S;d)$. Let $f\in C_b^{k,\omega}(S)$ be such that \[ \sup_{c\cdot B_b^{k,\omega}(S;d)}|f|<|f(v)|. \] By the definition of $B_b^{k,\omega}(S;d)$ the left-hand side of the pr
evious inequality coincides with $c\cdot\bigl(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\,\|f\|_{C_b^{k,\omega}(S')}\bigr)$. Hence, \[ c\cdot\left(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\right)<|f(v)|\le \|f\|_{C_b^{k,\omega}
(S)}, \] a contradiction with \eqref{e3.13}. \end{proof} \subsection{Proof of Theorem \ref{te1.6}} \begin{proof} We set \begin{equation}\label{e4.14} r_{X}(F)(s):=F(\delta_s^0),\quad F\in \mathcal L(G_b^{k,\omega}({\mbf R}^n); X),\quad s\in {\mbf R}^n. \en
d{equation} Applying the arguments similar to those of Proposition \ref{p2.2} we obtain \[ r_{X}(F)\in C_b^{k,\omega}({\mbf R}^n;X)\quad {\rm and}\quad \|r_X(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}\le \|F\|_{\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)}. \] On th
e other hand, for each $\varphi\in X^*$, $\|\varphi\|_{X^*}=1$, function $r_{{\mbf R}}(\varphi\circ F)\in C_b^{k,\omega}({\mbf R}^n)$. So, since $r_{{\mbf R}}(\varphi\circ F)=\varphi (r_{X}(F))$, \[ \|\varphi\circ F\|_{(G_b^{k,\omega}({\mbf R}^n))^*}=\|r_
{{\mbf R}}(\varphi\circ F)\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\varphi (r_{{\mbf R}^n;X}(F))\|_{C_b^{k,\omega}({\mbf R}^n)}\le \|r_{X}(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}. \] Taking supremum over all such $\varphi$ we get \[ \| F\|_{\mathcal L(G_b^{k,\omega}(
{\mbf R}^n);X)}\le \|r_{X}(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}\, \bigl(\le \| F\|_{\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)}\bigr). \] This shows that $r_{X}:\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)\rightarrow C_b^{k,\omega}({\mbf R}^n;X)$ is an isometry.
Let us prove that it is surjective. Since every finite subset of ${\mbf R}^n$ is interpolating for $C_b^{k,\omega}({\mbf R}^n)$, the set of vectors $\delta_s^0\in G_b^{k,\omega}({\mbf R}^n)$, $s\in{\mbf R}^n$, is linearly independent. Hence, each $f\in C
_b^{k,\omega}({\mbf R}^n;X)$ determines a linear map $\hat f:{\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}\rightarrow X$, \[ \hat f\left(\sum_{j}c_j\delta_{s_j}^0\right):=\sum_j c_j f(s_j),\quad \sum_{j}c_j\delta_{s_j}^0\in {\rm span}\{\delta_s^0\, :\, s
\in {\mbf R}^n\}\, (\subset G_b^{k,\omega}({\mbf R}^n)). \] Next, for each $\varphi\in X^*$, $\|\varphi\|_{X^*}=1$, function $\varphi\circ f\in C_b^{k,\omega}({\mbf R}^n)$ and \[ \|\varphi\circ f\|_{C_b^{k,\omega}({\mbf R}^n)}\le\|f\|_{C_b^{k,\omega}({\m
bf R}^n;X)}. \] Since $r_{{\mbf R}}:\bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*\rightarrow C_b^{k,\omega}({\mbf R}^n)$ is an isometric isomorphism, there exists $\ell_{\varphi\circ f}\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $r_{{\mbf R}}(\ell_{
\varphi\circ f})=\varphi\circ f$. Clearly, $\ell_{\varphi\circ f}$ coincides with $\varphi\circ\hat f$ on ${\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}$ and for all $v\in G_b^{k,\omega}({\mbf R}^n)$, \[ \begin{array}{r} \displaystyle |\ell_{\varphi\cir
c f}(v)|\le \| \ell_{\varphi\circ f}\|_{(G_b^{k,\omega}({\mbf R}^n))^*}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)} =\|\varphi\circ f\|_{C_b^{k,\omega}({\mbf R}^n)}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)}\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}({\
mbf R}^n;X)}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)}. \end{array} \] These imply that $\hat f:{\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}\rightarrow X$ is a linear continuous operator of norm $\le \|f\|_{C_b^{k,\omega}({\mbf R}^n;X)}$. Hence, it extend
s to a bounded linear operator $F:{\rm cl}({\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\})=:G_b^{k,\omega}({\mbf R}^n)\rightarrow X$ such that $r_{X}(F)=f$. Thus, $r_{X}(F):\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)\rightarrow C_b^{k,\omega}({\mbf R}^n;X)$
is an isometric isomorphism. The proof of the theorem is complete. \end{proof} \subsection{Proof of Theorem \ref{teo1.6}} \begin{proof} Without loss of generality we may assume that $T$ has depth $d$ and is defined by \eqref{equ1.6}. Let $T:\bigl(C_b^{k
,\omega}({\mbf R}^n)\bigr)^*\rightarrow \bigl(C_b^{k,\omega}(S)\bigr)^*$ be the adjoint of $T$ and $q_S^*:\bigl(C_b^{k,\omega}(S)\bigr)^*\rightarrow \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$ the adjoint of the quotient map $q_S: C_b^{k,\omega}({\mbf R}^n)\
rightarrow C_b^{k,\omega}(S)$. Clearly, $q_S^*$ is an isometric embedding which maps the closed subspace of $\bigl(C_b^{k,\omega}(S)\bigr)^*$ generated by $\delta$-functionals of points in $S$ isometrically onto $G_b^{k,\omega}(S)\subset \bigl(C_b^{k,\omeg
a}({\mbf R}^n)\bigr)^*$. We define \begin{equation}\label{proj} P:=q_S^*\circ T^*. \end{equation} By the definition of $T_S$, for each $\delta_x^0\in G_b^{k,\omega}({\mbf R}^n)$, $x\in{\mbf R}^n\setminus S$, and $f\in C_b^{k,\omega}(S)$ we have, for some $
y_i^x\in S$, \[ (P\delta_x^0)(f)=\delta_x^0(Tq_S f)=\sum_{i=1}^d \lambda_i^x\cdot f(y_i^x)=\left(\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\right)(f). \] Hence, \begin{equation}\label{equ3.11} P\delta_x^0=\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\qu
ad {\rm for\ all}\quad x\in{\mbf R}^n\setminus S. \end{equation} Since $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$, \begin{equation}\label{equ3.12} P\delta_x^0=\delta_x^0\quad {\rm for\ all}\quad x\in S. \end{equation} Thus $P$ maps $G_b^{k,\o
mega}({\mbf R}^n)$ into $G_b^{k,\omega}(S)$ and is identity on $G_b^{k,\omega}(S)$. Hence, $P$ is a bounded linear projection of norm $\|P\|\le \|T q_S\|\le \|T\|$. Next, under the identification $\bigl(G_b^{k,\omega}(S)\bigr)^*=C_b^{k,\omega}(S)$ for all
closed $S\subset{\mbf R}^n$ (see Theorem \ref{te1.2}), for all $x\in{\mbf R}^n\setminus S$, and $f\in C_b^{k,\omega}(S)$ we have by \eqref{equ3.11} \[ (P^*f)(\delta_x^0)=f(P\delta_x^0)=f\left(\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\right)=\sum_{i=1}
^d \lambda_i^x\cdot f(y_i^x)=(Tf)(\delta_x^0). \] The same identity is valid for $x\in S$, cf. \eqref{equ3.12}. This implies that $P^*=T$ and completes the proof of the theorem. \end{proof} \section{Proofs of Theorems \ref{te1.3} and \ref{teor1.10}} S
ections 4.1 and 4.2 contain auxiliary results used in the proof of Theorem \ref{te1.3}. \subsection{Jackson Theorem} Recall that the {\em Jackson kernel} $J_N$ is the trigonometric polynomial of degree $2\widetilde N$, where $\widetilde N:= \left\lfloor\fr
ac N2\right\rfloor$, given by the formula \[ J_N(t)=\gamma_N\left(\frac{\sin\frac{\widetilde N t}{2}}{\sin \frac t2}\right)^4, \] where $\gamma_N$ is chosen so that $\displaystyle \int_{-\pi}^\pi J_N(t)\, dt =1$. For a $2\pi$-periodic real function $f\in
C({\mbf R})$ we set \begin{equation}\label{jack1} (L_N f)(x):=\int_{-\pi}^\pi f(x-t)J_N(t)\, dt,\quad x\in\mathbb R. \end{equation} Then the classical {\em Jackson theorem} asserts (see, e.g., \cite[Ch.\,V]{T}): $L_N f$ is a real trigonometric polynomial
of degree at most $N$ such that \begin{equation}\label{jack2} \sup_{x\in (-\pi,\pi)}|f(x)-(L_N f)(x)|\le c\,\omega\bigl(f, \mbox{$\frac 1N$}\bigr), \end{equation} for a numerical constant $c>0$; here $\omega(f,\cdot)$ is the modulus of continuity of $f$.\
subsection{Convergence in the Weak$^*$ Topology of ${\mathbf C_b^{k,\omega}({\mbf R}^n)}$} In the proof of Theorem \ref{te1.3} we use the following result. As before, we equip $C_b^{k,\omega}({\mbf R}^n)$ with the weak$^*$ topology induced by means of fun
ctionals in $G_b^{k,\omega}({\mbf R}^n)\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$. \begin{Proposition}\label{prop3.1} A sequence $\{f_i\}_{i\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ weak$\,^*$ converges to $f\in C_b^{k,\omega}({\mbf R}^n)$ if
and only if \begin{itemize} \item[(a)] \[ \sup_{i\in{\mbf N}}\|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}<\infty; \] \item[(b)] For all $\alpha\in\mathbb Z_+^n$, $0\le |\alpha|\le k$, $x\in{\mbf R}^n$ \[ \lim_{i\rightarrow\infty}D^\alpha f_i(x)=D^\alpha f(x). \] \
end{itemize} \end{Proposition} \begin{proof} Without loss of generality we may assume that $f=0$. If $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$, then (a) follows from the Banach-Steinhaus theorem and (b) from the fact that each $\delta_x^\alpha\in
G_b^{k,\omega}({\mbf R}^n)$. Conversely, suppose that $\{f_i\}_{i\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ satisfies (a) and (b) with $f=0$. Let $g\in G_b^{k,\omega}({\mbf R}^n)$. According to Corollary \ref{cor2.3}, given $\varepsilon>0$ there exis
t $J\in{\mbf N}$ and families $c_{j\alpha}\in {\mbf R}$, $x_{j\alpha}\in{\mbf R}^n$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $0\le |\alpha|<k$, and $d_{j\alpha}\in {\mbf R}$, $x_{j\alpha}, y_{j\alpha}\in{\mbf R}^n$, $x_{j\alpha}\ne y_{j\alpha}$, $1\le j
\le J$, $\alpha\in {\mbf Z}_+^n$, $|\alpha|=k$, such that \[ g=\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}+g', \] where
\[ \sum_{j,\alpha}|c_{j\alpha}|+\sum_{j,\alpha}|d_{j\alpha}|\le \|g\|_{G_b^{k,\omega}({\mbf R}^n)}\quad {\rm and}\quad \|g'\|_{G_b^{k,\omega}({\mbf R}^n)}<\frac{\varepsilon}{2M},\quad M:=\sup_{i\in{\mbf N}}\|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}. \] Further,
due to condition (b), there exists $I\in{\mbf N}$ such that for all $i\ge I$, \[ \left|f_i\left(\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\al
pha}-y_{j\alpha} \|)} \right) \right|<\frac{\varepsilon}{2}. \] Also, for such $i$, \[ |f_i(g')|\le \|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}\cdot \|g'\|_{G_b^{k,\omega}({\mbf R}^n)}< M\cdot \frac{\varepsilon}{2M}=\frac{\varepsilon}{2}. \] Combining these in
equalities we obtain for all such $i$: \[ |f_i(g)|<\varepsilon. \] This shows that $\lim_{i\rightarrow\infty}f_i(g)=0$. Thus $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$, as required. \end{proof} \subsection{Proof of Theorem \ref{te1.3}\,(1)} We se
t \[ \mathbb K_N^n:=\bigl\{x=(x_1,\dots, x_n)\in{\mbf R}^n\, :\, \max_{1\le i\le n}|x_i|\le N\bigr\}. \] Let $\rho:\mathbb R^n\rightarrow [0,1]$ be a fixed $C^\infty$ function with support in the cube $\mathbb K_2^n$, equals one on the unit cube $\mathbb K
_1^n$. For a natural number $\ell$ we set $\rho_\ell(x):=\rho(x/\ell)$, $x\in\mathbb R^n$. Then there exist constants $c_{k,n}$ (depending on $k$ and $n$) such that \begin{equation}\label{rhol} \sup_{x\in\mathbb R^n}|D^\alpha\rho_\ell(x)|\le \frac{c_{k,n}}
{\ell^{|\alpha|}}\quad {\rm for\ all}\quad \alpha\in\mathbb Z_+^n\quad {\rm with }\quad |\alpha|\le k+1. \end{equation} Let $f\in C_b^{k,\omega}(\mathbb R^n)$. We define a $8\ell\sqrt n\,$-periodic in each variable function $f_\ell$ on $\mathbb R^n$ by
\begin{equation}\label{eq3.10} f_\ell(v+x)=\rho_\ell(x)\cdot f(x),\qquad v+x\in 8\ell\sqrt n\cdot\mathbb Z^n+\mathbb K_{4\ell\sqrt n}^n. \end{equation} Note that $f_\ell$ coincides with $f$ on the cube $\mathbb K_\ell^n$. \begin{Lm}\label{norm} There exi
sts a constant $C_\ell=C(\ell,k,n,\omega)$ (i.e., depending on $\ell, k,n$ and $\omega$) such that \[ \lim_{\ell\rightarrow\infty}C_{\ell}=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}, \] and \[ \|f_\ell\|_{C_b^{
k,\omega}(\mathbb R^n)}\le C_\ell\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] \end{Lm} \begin{proof} We use the standard multi-index notation. According to the general Leibniz rule, for $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k$, \[ D^\alpha f_\ell=\sum_{\n
u\, :\, \nu\le\alpha}\binom{\alpha}{\nu}(D^\nu\rho_\ell)\cdot (D^{\alpha-\nu}f)\quad {\rm on}\quad \mathbb K_{4\ell\sqrt n}^n . \] From here and \eqref{rhol} we get for $\alpha\in \mathbb Z_+^n$, $|\alpha|\ge 1$, \[ \begin{array}{l} \displaystyle \sup_{x\i
n\mathbb R^n}|D^\alpha f_\ell(x)|\le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left(\sum_{\nu\, :\, 0< \nu\le\alpha}\binom{\alpha}{\nu}\cdot\frac{c_{k,n}}{\ell^{|\nu|}}+1\right)\medskip\\ \displaystyle \le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left(c_{k,n}\cdot\left(\left
(1+\frac 1\ell \right)^{|\alpha|}-1\right)+1\right)\le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left( \left(1+\frac 1\ell\right)^{|\alpha|-1}\cdot \frac{c_{k,n}\cdot |\alpha|}{\ell}+1 \right). \end{array} \] Hence, \[ \|f_\ell\|_{C_b^k(\mathbb R^n)}\le\left( \lef
t(1+\frac 1\ell\right)^{\max\{k-1,0\}}\cdot \frac{c_{k,n}\cdot k}{\ell}+1 \right)\cdot \|f\|_{C_b^k(\mathbb R^n)}=:C_1(\ell,k,n)\cdot \|f\|_{C_b^k(\mathbb R^n)}. \] Similarly, for $\alpha\in\mathbb Z_+^n$, $|\alpha|=k$, and $x,y \in \mathbb K_{2\ell}^n$
using properties of $\omega$ we obtain\smallskip \begin{equation}\label{eq2.9} \begin{array}{lr} \qquad\\ \displaystyle |D^\alpha f_\ell (x)-D^\alpha f_\ell (y)|\medskip\\ \displaystyle \le \sum_{\nu\, :\, \nu\le\alpha}\binom{\alpha}{\nu}\cdot \bigl(| D^\
nu \rho_\ell(x)-D^\nu\rho_\ell(y)|\cdot |D^{\alpha-\nu}f(x) |+| D^\nu\rho_\ell(y)|\cdot |D^{\alpha-\nu}f(x)-D^{\alpha-\nu}f(y)|\bigr)\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}(\mathbb R^n)} \cdot\left(\sum_{\nu\, :\, 0<\nu\le\alpha}\binom{\alpha}{
\nu}\cdot c_{k,n}\cdot\left(\frac{\|x-y\|}{\ell^{|\nu|+1}}+\frac{\|x-y\|}{\ell^{|\nu|}}\right)+\frac{c_{k,n}\cdot\|x-y\|}{\ell}+\omega(\|x-y\|)\right)\medskip\\ \displaystyle = \|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot\left(c_{k,n}\cdot\left(\left(1+\frac
1\ell\right)^{k+1}-1\right)\cdot \|x-y\|+\omega(\|x-y\|\right)\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\cdot\left( c_{k,n}\cdot \left(1+\frac 1\ell\right)^{k}\cdot\frac{4\ell\sqrt n \cdot (k+1)}{\ell\cdot\omega
\bigl(4\ell \sqrt n\bigr)}+1\right)\medskip\\ \displaystyle =:\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\cdot C_2(\ell,k,n,\omega). \end{array} \end{equation} Observe that \begin{equation}\label{equ4.18} \lim_{\ell\rightarrow\infty}C_{2}(\ell
,k,n,\omega)=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}. \end{equation} Next, assume that $x,y\in {\rm supp}\, f_\ell$. Since the case $x,y\in \mathbb K_{2\ell}^n$ was considered above, without loss of generality
we may assume that $x\in \mathbb K_{2\ell}^n$ and $y\in v+\mathbb K_{2\ell}^n$ for some $v\in (8\ell\sqrt n\cdot\mathbb Z^n)\setminus\{0\}$. Then for $y':=y-v\in \mathbb K_{2\ell}^n$ we have $D^\alpha f_\ell(y')=D^\alpha f_\ell(y)$. Also, \[ \|x-y\
|=\|x-y'-v\|\ge \|v\|-\|x-y'\|\ge 8\ell\sqrt n- 4\ell\sqrt n=4\ell\sqrt n. \] Therefore for $\alpha\in\mathbb Z_+^n$, $|\alpha|=k$, \[ \begin{array}{l} \displaystyle |D^\alpha f_\ell (x)-D^\alpha f_\ell (y)|= |D^\alpha f_\ell (x)-D^\alpha f_\ell (y')
|\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y'\|)\medskip\\ \displaystyle \le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(4\ell\sqrt n)\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\m
athbb R^n)}\cdot \omega(\|x-y\|). \end{array} \] Finally, in the case $x\in {\rm supp}\, f_\ell \cap \mathbb K_{2\ell}^n$, $y\not\in {\rm supp}\, f_\ell$, there exists a point $y'\in \mathbb K_{2\ell}^n$ lying on the interval joining $x$ and $y$ such t
hat $f_\ell(y')=0$. Since $\|x-y'\|\le \|x-y\|$ and inequality \eqref{eq2.9} is valid for $x$ and $y'$, a similar inequality is valid for $x$ and $y$. Hence, combining the considered cases we conclude that (cf. \eqref{eq5}) \[ |f_\ell|_{C_b^{k,\omega}(\mat
hbb R^n)}\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] Therefore the inequality of the lemma is valid with \[ C(\ell,k,n,\omega):=\max\{C_1(\ell,k,n), C_2(\ell,k,n,\omega)\}, \] see \eqref{equ4.18}. \end{proof} We set \begin{equa
tion}\label{eq2.10} \lambda:=\frac{4\ell\sqrt n}{\pi}. \end{equation} For a natural number $N$ and $x=(x_1,\dots, x_n)\in\mathbb R^n$ we define \begin{equation}\label{eq2.11} (E_Nf_\ell)(x)=\int_{-\pi}^\pi\cdots\int_{-\pi}^\pi f_\ell(x_1-\lambda t_1,\dots,
x_n-\lambda t_n)J_N(t_1)\cdots J_N(t_n)\, dt_1\cdots dt_n. \end{equation} \begin{Lm}\label{lem2.2} Function $E_Nf_\ell$ satisfies the following properties: \begin{itemize} \item[(a)] $(E_Nf_\ell)(\lambda x)$, $x\in\mathbb R^n$, is the trigonometric polyn
omial of degree at most $N$ in each coordinate; \item[(b)] \[ \|E_Nf_\ell\|_{C_b^{k,\omega}(\mathbb R^n)}\le C_\ell\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}; \] \item[(c)] There is a constant $c_N$, $\lim_{N\rightarrow\infty} c_N=0$, such that \[ \|f_\ell
-E_Nf_\ell\|_{C_b^{k}(\mathbb R^n)}\le c_N\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] \end{itemize} \end{Lm} \begin{proof} (a) If we set $f_\ell^\lambda (x):=f_\ell(\lambda x)$, $x=(x_1,\dots, x_n)\in \mathbb R^n$, then \[ E_Nf_\ell(\lambda x)=(L_N^1\cdo
ts L_N^nf_\ell^\lambda)(x), \] where $L_N^i$ is the Jackson operator \eqref{jack1} acting on univariate functions in variable $x_i$, $1\le i\le n$. This and the properties of $L_N^i$ give the required statement.\smallskip (b) According to definition \eqre
f{eq2.11}, for each $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k$, \begin{equation}\label{eq2.12} \sup_{x\in\mathbb R^N}|D^\alpha(E_Nf_\ell)(x)|=\sup_{x\in\mathbb R^n}|(E_N D^\alpha f_\ell)(x)|\le \|f_\ell\|_{C_b^k(\mathbb R^n)}\le C_\ell\cdot \|f\|_{C_b^k(\ma
thbb R^n)}. \end{equation} In turn, if $|\alpha|=k$ and $x,y\in\mathbb R^n$, then \begin{equation}\label{eq2.13} \begin{array}{l} \displaystyle |D^\alpha (E_Nf_\ell)(x)-D^\alpha(E_Nf_\ell)(y)|\le |(E_N D^\alpha f_\ell)(x)- (E_ND^\alpha f_\ell)(y)|\
medskip\\ \displaystyle \le |f_\ell|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\le C_\ell\cdot |f|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|). \end{array} \end{equation} Thus \eqref{eq2.12} and \eqref{eq2.13} give the required statemen
t.\smallskip (c) For each $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k-1$, using \eqref{eq2.11} and due to \eqref{jack2} one obtains\smallskip \begin{equation}\label{eq2.14} \\ \begin{array}{l} \displaystyle \sup_{x\in\mathbb R^n}|D^\alpha\bigl( f_\ell
-E_Nf_\ell\bigr)(x)|=\sup_{x\in\mathbb R^n}|D^\alpha f_\ell(x)-(E_ND^\alpha f_\ell)(x)|\medskip\\ \displaystyle \le \sup_{x\in\mathbb R^n}\bigl\{|D^\alpha f_\ell(\lambda x)-(L_N^1D^\alpha f_\ell)(\lambda x)|+|\bigl(L_N^1\bigl(D^\alpha f_\ell-L_N^2D^\alph
a f_\ell\bigr)\bigr)(\lambda x)|+\,\cdots \medskip\\ \displaystyle + |\bigl(L_N^1\cdots L_N^{n-1}\bigl(D^\alpha f_\ell-L_N^nD^\alpha f_\ell\bigr)\bigr)(\lambda x)| \bigr\}\le \frac{c\cdot n}{N}\cdot\|f_\ell\|_{C_b^k(\mathbb R^n)}\le \frac{c\cdot n\cdot
C_\ell}{N}\cdot\|f\|_{C_b^k(\mathbb R^n)}. \end{array} \end{equation} Similarly, for $|\alpha|=k$, we have\smallskip \begin{equation}\label{eq2.15} \sup_{x\in\mathbb R^n}|D^\alpha(f_\ell -E_Nf_\ell)(x)|\le c\cdot n\cdot C_\ell\cdot\omega(\mbox{$\frac
1N$})\cdot |f|_{C_b^{k,\omega}(\mathbb R^n)}. \end{equation} Equations \eqref{eq2.14} and \eqref{eq2.15} imply that the required statement is valid with \[ c_N:=c\cdot n\cdot C_\ell\cdot \max\left\{\mbox{$\frac 1N ,\, \omega(\frac 1N)$}\right\}. \]
\end{proof} \begin{proof}[Proof of Theorem \ref{te1.3}\,(1)] For $f\in C_b^{k,\omega}({\mbf R}^n)$ we set \begin{equation}\label{eq3.18} L_{N,\ell}f:=E_Nf_\ell. \end{equation} According to Lemma \ref{lem2.2}, $L_{N,\ell}: C_b^{k,\omega}({\mbf R}^n)\right
arrow C_b^{k,\omega}({\mbf R}^n)$ is a finite rank bounded linear operator of norm $\le C_\ell$. \begin{Lm}\label{lem3.3} Operators $L_{N,\ell}$ are weak$\,^*$ continuous. \end{Lm} \begin{proof} Since $C_b^{k,\omega}({\mbf R}^n)=\bigl(G_b^{k,\omega}(
{\mbf R}^n)\bigr)^*$ and $G_b^{k,\omega}({\mbf R}^n)$ is separable, $C_b^{k,\omega}({\mbf R}^n)$ equipped with the weak$^*$ topology is a Frechet space. Then $L_{N,\ell}$ is weak$^*$ continuous if and only if for each sequence $\{f_i\}_{i\in{\mbf N}}\subs
et C_b^{k,\omega}({\mbf R}^n)$ weak$^*$ converging to $0\in C_b^{k,\omega}({\mbf R}^n)$ the sequence $\{L_{N,\ell}f_i\}_{i\in{\mbf N}}$ weak$\,^*$ converges to $0$ as well. Note that such a sequence $\{f_i\}_{i\in{\mbf N}}$ is bounded in $C_b^{k,\omega}({
\mbf R}^n)$ due to the Banach-Steinhaus theorem. Then $\{L_{N,\ell}f_i\}_{i\in{\mbf N}}$ is bounded as well and according to Proposition \ref{prop3.1} we must prove only that \begin{equation}\label{eq3.19} \lim_{i\rightarrow\infty}D^\alpha (L_{N,\ell}f_
i)(x)=0\quad {\rm for\ all}\quad \alpha\in{\mbf Z}_+^n,\ 0\le |\alpha|\le k,\ x\in{\mbf R}^n. \end{equation} Further, since $D^\alpha (L_{N,\ell}f_i)(x)=(E_ND^\alpha (f_i)_\ell)(x)$ for such $\alpha$ and $x$, $\{D^\alpha (f_i)_\ell\}_{i\in{\mbf N}}$ is a
bounded sequence of continuous functions and $E_N$ is the convolution operator with the absolutely integrable kernel, to establish \eqref{eq3.19} it suffices to prove (due to the Lebesgue dominated convergence theorem) that \[ \lim_{i\rightarrow\infty}D
^\alpha (f_i)_\ell(x)=0 \quad {\rm for\ all}\quad \alpha\in{\mbf Z}_+^n,\ 0\le |\alpha|\le k,\ x\in{\mbf R}^n. \] The latter follows directly from the definition of $(f_i)_\ell$, see \eqref{eq3.10}, the general Leibniz rule and the fact that $D^\alpha f_
i(x)\rightarrow 0$ as $i\rightarrow\infty$ for all the required $\alpha$ and $x$ (because $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$). Thus we have proved that operators $L_{N,\ell}$ are weak$^*$ continuous. \end{proof} Lemma \ref{lem3.3} implies
that there exists a bounded operator of finite rank $H_{N,\ell}$ on $G_b^{k,\omega}({\mbf R}^n)$ whose adjoint $H_{N,\ell}^*$ coincides with $L_{N,\ell}$. \begin{Lm}\label{lem3.5} The sequence of finite rank bounded operators $\{H_{N,N}\}_{N\in{\mbf N}}
$ converges pointwise to the identity operator on $G_b^{k,\omega}({\mbf R}^n)$. \end{Lm} \begin{proof} Let $g\in G_b^{k,\omega}({\mbf R}^n)$. Due to Corollary \ref{cor2.3}, given $\varepsilon>0$ there exist $J\in{\mbf N}$ and families $c_{j\alpha}\in {\
mbf R}$, $x_{j\alpha}\in{\mbf R}^n$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $0\le |\alpha|<k$, and $d_{j\alpha}\in {\mbf R}$, $x_{j\alpha}, y_{j\alpha}\in{\mbf R}^n$, $x_{j\alpha}\ne y_{j\alpha}$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $|\alpha|=k$,
such that \[ g=\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}+g''=:g'+g'', \] where \[ \sum_{j,\alpha}|c_{j\alpha}|+\sum_{
j,\alpha}|d_{j\alpha}|\le \|g\|_{G_b^{k,\omega}({\mbf R}^n)}\quad {\rm and}\quad \|g''\|_{G_b^{k,\omega}({\mbf R}^n)}<\frac{\varepsilon}{2(C_N+1)}, \] see Lemma \ref{norm} for the definition of $C_N$. Next, for each $f\in C_b^{k,\omega}({\mbf R}^n)$, $\|
f\|_{C_b^{k,\omega}({\mbf R}^n)}=1$, we have by means of Lemma \ref{lem2.2},\smallskip \begin{equation}\label{eq3.20} \begin{array}{l} \ \ \ \vspace*{-2mm}\\ \displaystyle \bigl|f\bigl(H_{NN}\,g-g\bigr)\bigr|=\bigl|\bigl(L_{NN}f-f\bigr)(g)\bigr|\le \bigl|
\bigl(L_{NN}f-f\bigr)(g')\bigr|+ \bigl|\bigl(L_{NN}f-f\bigr)(g'')\bigr|\medskip\\ \displaystyle < \bigl|\bigl(E_{N}f_N-f\bigr)(g')\bigr|+\|L_{NN}-{\rm id}\|\cdot\frac{\varepsilon}{2(C_{N}+1)}\le\bigl|\bigl(E_{N}f_N-f\bigr)(g')\bigr|+\frac{\varepsilon}{2}.
\end{array} \end{equation} Let $N_0\in{\mbf N}$ be so large that all points $x_{j\alpha}, y_{j\alpha}$ as above belong to $\mathbb K_{N_0}^n$. Since $f_{N_0}=f$ on $\mathbb K_{N_0}^n$, for all $N\ge N_0$, \[ \bigl(E_{N}f_N-f\bigr)(g')=\bigl(E_{N}f_N-f_N\bi
gr)(g'). \] Hence, due to Lemma \ref{lem2.2}\,(c) for $z=x_{j\alpha}$ or $y_{j\alpha}$, \[ \bigl|\bigl(E_{N}f_N-f\bigr)(\delta_z^\alpha)\bigr|\le \|E_{N}f_N-f_N\|_{C_b^{k}({\mbf R}^n)} \le c_N. \] This implies that for all $N\ge N_0$ \[ \bigl|\bigl(E_{N}f_
N-f\bigr)(g')\bigr|\le c_N\cdot\left(\sum_{j,\alpha}|c_{j\alpha}| +\left(\sum_{j,\alpha}2|d_{j\alpha}|\right)\cdot\max_{j,\alpha}\left\{\frac{1}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}\right\}\right). \] Choose $N_0'\ge N_0$ so large that for all $N\ge N_0'
$ the right-hand side of the previous inequality is less than $\frac{\varepsilon}{2}$. Then combining this with \eqref{eq3.20} we get for all $N\ge N_0'$, \[ \|H_{NN}\,g-g\|_{G_b^{k,\omega}({\mbf R}^n)}<\varepsilon. \] This shows that for all $g\in G_b
^{k,\omega}({\mbf R}^n)$ \[ \lim_{N\rightarrow\infty}H_{NN}\,g=g \] which completes the proof of the lemma. \end{proof} Let us finish the proof of the theorem for $S={\mbf R}^n$. We set \begin{equation}\label{equ4.28} T_{N}:=\left(1+c_{k,n}\cdot 4\sqr
t n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}\right)\cdot\frac{H_{NN}}{C_N}, \end{equation} see Lemma \ref{norm} for the definition of $C_N$. Since $\{C_N\}_{N\in{\mbf N}}$ converges to the first factor in the definition of $T_N$, due t
o Lemma \ref{lem3.5} $\{T_N\}_{N\in{\mbf N}}$ is the sequence of operators of finite rank on $G_b^{k,\omega}({\mbf R}^n)$ of norm at most $\lambda:=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}(1/\omega(t))$ converging pointwise to the
identity operator. In particular, this sequence converges uniformly to the identity operator on each compact subset of $G_b^{k,\omega}({\mbf R}^n)$. This shows that $G_b^{k,\omega}({\mbf R}^n)$ has the $\lambda$-approximation property with respect to the
approximating sequence of operators $\{T_N\}_{N\in{\mbf N}}$. The proof of Theorem \ref{te1.3} for $S={\mbf R}^n$ is complete. \end{proof} \subsection{Proof of Theorem \ref{te1.3}\,(2)} \begin{proof} In the case of $G_b^{k,\omega}(S)$, the required seque
nce of finite rank linear operators approximating the identity map is $\bigl\{PT_N|_{G_b^{k,\omega}(S)}\bigr\}_{N\in{\mbf N}}$, where $T_N$ are linear operators defined by \eqref{equ4.28} and $P: G_b^{k,\omega}({\mbf R}^n)\rightarrow G_b^{k,\omega}(S)$ is
the projection of Theorem \ref{teo1.6}. We have \[ \|PT_N|_{G_b^{k,\omega}(S)}\|\le \|P\|\cdot\|T_N\|=:\|P\|\cdot\lambda(k,n,\omega). \] Choosing here $P$ corresponding to the extension operators of papers \cite{Gl} ($k=0$), \cite{BS2} ($k=1$) and \cite{Lu
} ($k\ge 2$) we obtain the required result. The proof of Theorem \ref{te1.3} is complete. \end{proof} \subsection{Proof of Theorem \ref{teor1.10}} \begin{proof} Due to the result of Pe\l czy\'nski \cite{P} there are a separable Banach space $Y$ with a nor
m one monotone basis $\{b_j\}_{j\in{\mbf N}}$, an isomorphic embedding $T:X\rightarrow Y$ with distortion $\|T\|\cdot\|T^{-1}\|\le 4\lambda$, and a linear projection $P:Y\rightarrow T(X)$ with $\|P\|\le 4\lambda$. For an operator $H\in\mathcal L(G_b^{k,\o
mega}({\mbf R}^n);X)$ we define \[ \widetilde H:=T\cdot H\in \mathcal L(G_b^{k,\omega}({\mbf R}^n);Y). \] Then for each $x\in{\mbf R}^n$, \[ \widetilde H(\delta_x^0)=\sum_{j=1}^\infty \tilde h_j(x)\cdot b_j \] for some $\tilde h_j(x)\in{\mbf R}$, $j\in{\m
bf N}$. Further, consider the family of bounded linear functionals $\{b_j^*\}_{j\in{\mbf N}}\subset Y^*$ such that $b_j^*(b_i)=\delta_{ij}$ (- the Kronecker delta) for all $i, j\in{\mbf N}$. As the basis $\{b_j\}_{j\in{\mbf N}}$ is monotone, $\|b_j^*\|\l
e 2$ for all $j\in{\mbf N}$. Since $b_j^*\circ\widetilde H\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*=C_b^{k,\omega}({\mbf R}^n)$, the functions $\tilde h_j$, $\tilde h_j(x):=(b_j^*\circ \widetilde H)(\delta_x^0)$, $x\in{\mbf R}^n$, belong to $C_b^{k,\ome
ga}({\mbf R}^n)$ and \begin{equation}\label{equ4.32a} \|\tilde h_j\|_{C_b^{k,\omega}({\mbf R}^n)}\le 2\cdot\|T\|\cdot\|H\|\quad {\rm for\ all}\quad j\in{\mbf N}. \end{equation} In particular, $(b_j^*\circ\widetilde H)(\delta_x^\alpha)=D^\alpha (b_j^*\circ
\widetilde H)(\delta_x^0)=D^\alpha \tilde h_j(x)$ for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, $x\in{\mbf R}^n$, $j\in{\mbf N}$. This implies that for all such $\alpha$ and $x$, \begin{equation}\label{equ4.32} \widetilde H(\delta_x^\alpha)=\sum_{j=1}^