Abstract
We prove that the spherical mean value of the Dunkl-type generalized translation operator \(\tau ^y\) is a positive \(L^p\)-bounded generalized translation operator \(T^t\). As applications, we prove the Young inequality for a convolution defined by \(T^t\), the \(L^p\)-boundedness of \(\tau ^y\) on radial functions for \(p>2\), the \(L^p\)-boundedness of the Riesz potential for the Dunkl transform, and direct and inverse theorems of approximation theory in \(L^p\)-spaces with the Dunkl weight.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
During the last three decades, many important elements of harmonic analysis with Dunkl weight on \(\mathbb {R}^d\) and \({\mathbb {S}}^{d-1}\) were proved; see, e.g., the papers by Dunkl [14,15,16], Rösler [40,41,42,43], de Jeu [24, 25], Trimèche [52, 53], Xu [54, 55], and the recent works [1, 11, 12, 19, 20].
Yet there are still several gaps in our knowledge of Dunkl harmonic analysis. In particular, Young’s convolution inequality, several important polynomial inequalities, and basic approximation estimates are not established in the general case. One of the main reasons is the lack of tools related to the translation operator. Needless to say, the standard translation operator \(f\mapsto f(\cdot +y)\) plays a crucial role both in classical approximation theory and harmonic analysis, in particular, by introducing several smoothness characteristics of f. In Dunkl analysis, its analogue is the generalized translation operator \(\tau ^y\) defined by Rösler [40]. Unfortunately, the \(L^p\)-boundedness of \(\tau ^y\) is not obtained in general.
To overcome this difficulty, the spherical mean value of the translation operator \(\tau ^y\) was introduced in [28] and was studied in [42], where, in particular, its positivity was shown. Our main goal in this paper is to prove that this operator is a positive \(L^p\)-bounded operator \(T^t\), which may be considered as a generalized translation operator. It is worth mentioning that this operator can be applied to problems where it is essential to deal with radial multipliers. This is because by virtue of \(T^t\) we can define the convolution operator that coincides with the known convolution introduced by Thangavelu and Xu in [48] using the operator \(\tau ^y\).
For this convolution, we prove the Young inequality and, subsequently, an \(L^p\)-boundedness of the operator \(\tau ^y\) on radial functions for \(p>2\). For \(1\le p\le 2\) it was proved in [48].
Let us mention here two applications of the operator \(T^t\). The first one is the Riesz potential defined in [49], where its boundedness properties were obtained for the reflection group \(\mathbb {Z}_2^d\). For the general case see [21]. Using the \(L^p\)-boundedness of the operator \(T^t\) allows us to give a different simple proof, which follows ideas of Thangavelu and Xu [49]. Another application is basic inequalities of approximation theory in the weighted \(L^p\) spaces. With the help of the operator \(T^t\) one can define moduli of smoothness, which are equivalent to the K-functionals, and prove the direct and inverse approximation theorems. For the reflection group \(\mathbb {Z}_2^d\), basic approximation inequalities were studied in [11, 12].
The paper is organized as follows. In the next section, we give some basic notation and facts of Dunkl harmonic analysis. In Sect. 3, we study the operator \(T^t\), define a convolution operator, and prove the Young inequality. As a consequence, we obtain an \(L^p\)-boundedness of the operator \(\tau ^y\) on radial functions. The weighted Riesz potential is studied in Sect. 4. Section 5 consists of a study of interrelation between several classes of entire functions. We also obtain multidimensional weighted analogues of Plancherel–Polya–Boas inequalities, which are of their own interest. In Sect. 6, we introduce moduli of smoothness and the K-functional, associated to the Dunkl weight, and prove equivalence between them as well as the Jackson inequality. Section 7 consists of weighted analogues of Nikol’skiǐ, Bernstein, and Boas inequalities for entire functions of exponential type. In Sect. 8, we obtain that moduli of smoothness are equivalent to the realization of the K-functional. We conclude with Sect. 9, where we prove the inverse theorems in \(L^p\)-spaces with the Dunkl weight.
2 Notation
In this section, we recall the basic notation and results of Dunkl harmonic analysis, see, e.g., [43].
Throughout the paper, \(\langle x,y\rangle \) denotes the standard Euclidean scalar product in d-dimensional Euclidean space \(\mathbb {R}^{d}\), \(d\in \mathbb {N}\), equipped with a norm \(|x|=\sqrt{\langle x,x\rangle }\). For \(r>0\) we write \(B_r=\{x\in \mathbb {R}^d:|x|\le r\}\). Define the following function spaces:
-
\(C(\mathbb {R}^d)\) the space of continuous functions,
-
\(C_b(\mathbb {R}^d)\) the space of bounded continuous functions with the norm \(\Vert f\Vert _{\infty }=\sup _{\mathbb {R}^d}|f|\),
-
\(C_0(\mathbb {R}^d)\) the space of continuous functions that vanish at infinity,
-
\(C^{\infty }(\mathbb {R}^d)\) the space of infinitely differentiable functions,
-
\(\mathcal {S}(\mathbb {R}^d)\) the Schwartz space,
-
\(\mathcal {S}'(\mathbb {R}^d)\) the space of tempered distributions,
-
\(X(\mathbb {R}_+)\) the space of even functions from \(X(\mathbb {R})\), where X is one of the spaces above,
-
\(X_\mathrm {rad}(\mathbb {R}^d)\) the subspace of \(X(\mathbb {R}^d)\) consisting of radial functions \(f(x)=f_{0}(|x|)\).
Let a finite subset \(R\subset \mathbb {R}^{d}\setminus \{0\}\) be a root system; \(R_{+}\) a positive subsystem of R; \(G(R)\subset O(d)\) the finite reflection group, generated by reflections \(\{\sigma _{a}:a\in R\}\), where \(\sigma _{a}\) is a reflection with respect to hyperplane \(\langle a,x\rangle =0\); \(k:R\rightarrow \mathbb {R}_{+}\) a G-invariant multiplicity function. Recall that a finite subset \(R\subset \mathbb {R}^{d}\setminus \{0\}\) is called a root system if
Let
be the Dunkl weight,
and \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(0<p<\infty \), be the space of complex-valued Lebesgue measurable functions f for which
We also assume that \(L^{\infty }\equiv C_b\) and \(\Vert f\Vert _{\infty ,\mathrm{d}\mu _{k}}=\Vert f\Vert _{\infty }\).
Example
If the root system R is \(\{\pm e_1,\dots ,\pm e_d\}\), where \(\{e_1,\dots ,e_d\}\) is an orthonormal basis of \(\mathbb {R}^{d}\), then \(v_{k}(x)=\prod _{j=1}^d|x_j|^{2k_j}\), \(k_{j}\ge 0\), \(G=\mathbb {Z}_2^d\).
Let
be differential-differences Dunkl operators and \(\Delta _k=\sum _{j=1}^dD_j^2\) be the Dunkl Laplacian. The Dunkl kernel \(e_{k}(x, y)=E_{k}(x, iy)\) is a unique solution of the system
and it plays the role of a generalized exponential function. Its properties are similar to those of the classical exponential function \(e^{i\langle x, y\rangle }\). Several basic properties follow from an integral representation [41]:
where \(\mu _x^k\) is a probability Borel measure, whose support is contained in
the convex hull of the G-orbit of x in \(\mathbb {R}^d\). In particular, \(|e_k(x,y)|\le 1\).
For \(f\in L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), the Dunkl transform is defined by the equality
For \(k\equiv 0\), \(\mathcal {F}_{0}\) is the classical Fourier transform \(\mathcal {F}\). We also note that \(\mathcal {F}_k(e^{-|\,\cdot \,|^2/2})(y)=e^{-|y|^2/2}\) and \(\mathcal {F}_{k}^{-1}(f)(x)=\mathcal {F}_{k}(f)(-x)\). Let
Let us now list several basic properties of the Dunkl transform.
Proposition 2.1
-
(1)
For \(f\in L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(\mathcal {F}_{k}(f)\in C_0(\mathbb {R}^d)\).
-
(2)
If \(f\in \mathcal {A}_k\), we have the pointwise inversion formula
$$\begin{aligned} f(x)=\int _{\mathbb {R}^{d}}\mathcal {F}_{k}(f)(y)e_{k}(x, y)\,\mathrm{d}\mu _{k}(y). \end{aligned}$$ -
(3)
The Dunkl transform leaves the Schwartz space \(\mathcal {S}(\mathbb {R}^d)\) invariant.
-
(4)
The Dunkl transform extends to a unitary operator in \(L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\).
Let \(\lambda \ge -1/2\) and \(J_{\lambda }(t)\) be the classical Bessel function of degree \(\lambda \) and
be the normalized Bessel function. Set
The norm in \(L^{p}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda })\), \(1\le p<\infty \), is given by
Define \(\Vert f\Vert _{\infty }={\text {ess\,sup}}_{t\in \mathbb {R}_{+}}|f(t)|\).
The Hankel transform is defined as follows:
It is a unitary operator in \(L^{2}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda })\) and \(\mathcal {H}_{\lambda }^{-1}=\mathcal {H}_{\lambda }\) [2, Chap. 7].
Note that if \(\lambda =d/2-1\), the Hankel transform is a restriction of the Fourier transform on radial functions, and if \(\lambda =\lambda _k=d/2-1+\sum _{a\in R_+}k(a)\), of the Dunkl transform.
Let \(\mathbb {S}^{d-1}=\{x'\in \mathbb {R}^d:|x'|=1\}\) be the Euclidean sphere and \(\mathrm{d}\sigma _k(x')=a_kv_k(x')\,\mathrm{d}x'\) be the probability measure on \(\mathbb {S}^{d-1}\). We have
We need the following partial case of the Funk–Hecke formula [55]
Throughout the paper, we will assume that \(A\lesssim B\) means that \(A\le C B\) with a constant C depending only on nonessential parameters.
3 Generalized Translation Operators and Convolutions
Let \(y\in \mathbb {R}^d\) be given. Rösler [40] defined a generalized translation operator \(\tau ^y\) in \(L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) by the equation
Since \(|e_k(y, z)|\le 1\), we have \(\Vert \tau ^y\Vert _{2\rightarrow 2}\le 1\). If \(f\in \mathcal {A}_k\) (recall that \(\mathcal {A}_k\) is given by (2.1)), then, for any \(x, y\in \mathbb {R}^d\),
Note that \(\mathcal {S}(\mathbb {R}^d)\subset \mathcal {A}_k\subset L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\). Trimèche [53] extended the operator \(\tau ^y\) on \(C^{\infty }(\mathbb {R}^d)\).
The explicit expression of \(\tau ^yf\) is known only in the case of the reflection group \(\mathbb {Z}_2^d\). In particular, in this case \(\tau ^yf\) is not a positive operator [39]. Note that in the case of symmetric group \(S_d\), the operator \(\tau ^yf\) is also not positive [48].
It remains an open question whether \(\tau ^yf\) is an \(L^p\) bounded operator on \(\mathcal {S}(\mathbb {R}^d)\) for \(p\ne 2\). It is known [39, 48] only for \(G=\mathbb {Z}_2^d\). Note that a positive answer would follow from the \(L^1\)-boundedness.
Let
We have \(\lambda _k\ge -1/2\) and, moreover, \(\lambda _k=-1/2\) only if \(d=1\) and \(k\equiv 0\). In what follows, we assume that \(\lambda _k>-1/2\).
Define another generalized translation operator \(T^t:L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\rightarrow L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(t\in \mathbb {R}\), by the relation
Since \(|j_{\lambda _k}(t)|\le 1\), it is a bounded operator such that \(\Vert T^t\Vert _{2\rightarrow 2}\le 1\) and
This gives \(T^t=T^{-t}\). If \(f\in \mathcal {A}_k\), then from (2.3) and (3.1) we have (pointwise)
Note that the operator \(T^t\) is self-adjoint. Indeed, if \(f, g\in \mathcal {A}_k\), then
Rösler [42] proved that the spherical mean (with respect to the Dunkl weight) of the operator \(\tau ^y\), i.e., \(\int _{\mathbb {S}^{d-1}}\tau ^{ty'}f(x)\,\mathrm{d}\sigma _k(y')\), is a positive operator on \(C^{\infty }(\mathbb {R}^d)\) and obtained its integral representation. This implies that \(T^t\) is a positive operator on \(C^{\infty }(\mathbb {R}^d)\) and, moreover, for any \(t\in \mathbb {R}\), \(x\in \mathbb {R}^d\),
where \(\sigma _{x,t}^k\) is a probability Borel measure,
and the mapping \((x, t)\rightarrow \sigma _{x,t}^k\) is continuous with respect to the weak topology on probability measures.
The representation (3.3) gives a natural extension of the operator \(T^t\) on \(C_b(\mathbb {R}^d)\); namely, for \(f\in C_b(\mathbb {R}^d)\) we define \(T^tf(x)\in C(\mathbb {R}\times \mathbb {R}^d)\) by (3.3), and, moreover, the estimate \(\Vert T^tf\Vert _{\infty }\le \Vert f\Vert _{\infty }\) holds.
Note that for \(k\equiv 0\), \(T^t\) is the usual spherical mean
Theorem 3.1
If \(1\le p\le \infty \), then, for any \(t\in \mathbb {R}\) and \(f\in \mathcal {S}(\mathbb {R}^d)\),
Remark 3.2
-
(i)
The inequality \(\Vert T^tf\Vert _{p,\mathrm{d}\mu _k}\le c\Vert f\Vert _{p,\mathrm{d}\mu _k}\) was proved in [48] for \(G=\mathbb {Z}_2^d\).
-
(ii)
\(\mathcal {S}(\mathbb {R}^d)\) is dense in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1\le p<\infty \), so for any \(t\in \mathbb {R}_+\) the operator \(T^t\) can be defined on \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) and estimate (3.6) holds.
-
(iii)
If \(d=1\), \(v_k(x)=|x|^{2\lambda +1}\), \(\lambda >-1/2\), inequality (3.6) was proved in [7]. In this case the integral representation of \(T^t\) is of the form
$$\begin{aligned} T^tf(x)=\frac{c_{\lambda }}{2}\int _0^{\pi }\{f(A)(1+B)+f(-A)(1-B)\} \sin ^{2\lambda }\varphi \,\mathrm{d}\varphi , \end{aligned}$$where, for \((x,t)\ne (0,0)\),
$$\begin{aligned} c_{\lambda }=\frac{\Gamma (\lambda +1)}{\sqrt{\pi }\Gamma (\lambda +1/2)},\quad A=\sqrt{x^2+t^2-2xt\cos \varphi },\quad B=\frac{x-t\cos \varphi }{A}. \end{aligned}$$(3.7)If \(\lambda =-1/2\), i.e., \(k\equiv 0\), then \(T^tf(x)=\frac{1}{2} (f(x+t)+f(x-t))\).
Proof
Let \(t\in \mathbb {R}_+\) be given and the operator \(T^t\) be defined on \(\mathcal {S}(\mathbb {R}^d)\) by (3.3). Using (3.2), we have
and \(T^t\) can be extended to the space \(L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) with preservation of norm; moreover, this extension coincides with (3.2). Furthermore, (3.3) yields
Since the operator \(T^t\) is self-adjoint, by (3.8),
Hence, \(T^t\) can be extended to \(L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) with preservation of the norm such that this extension coincides with (3.2) on \(L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\cap L^{2}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\).
By the Riesz–Thorin interpolation theorem we obtain
Let \(2<p<\infty \), \(1/p+1/p'=1.\) As for \(p=1\), we get
\(\square \)
For any \(f_0\in L^{p}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda })\), \(1\le p\le \infty \), \(\lambda >-1/2\), let us define the Gegenbauer-type translation operator (see, e.g., [34, 35])
where \(c_\lambda \) is defined by (3.7). We have that \(\Vert R^t\Vert _{p\rightarrow p}\le 1\) and \(\mathcal {H}_{\lambda }(R^tf_0)(r)=j_\lambda (tr)\mathcal {H}_{\lambda }(f_0)(r)\), where \(f_0\in \mathcal {S}(\mathbb {R}_+)\). Taking into account (2.3) and (3.2), we note that for \(\lambda =\lambda _k\) the operator \(R^t\) is a restriction of \(T^t\) on radial functions; that is, for \(f_0\in L^{p}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\),
We also mention the following useful properties of the generalized translation operator \(T^t\).
Lemma 3.3
Let \(t\in \mathbb {R}\).
-
(1)
If \(f\in L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), then \(\int _{\mathbb {R}^d}T^tf\,\mathrm{d}\mu _k=\int _{\mathbb {R}^d}f\,\mathrm{d}\mu _k\).
-
(2)
Let \(r>0\), \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1\le p<\infty \). If \({\text {supp}}f\subset B_r\), then \({\text {supp}}T^tf\subset B_{r+|t|}\). If \({\text {supp}}f\subset \mathbb {R}^d\setminus B_r\), \(r>|t|\), then \({\text {supp}}T^tf\subset \mathbb {R}^d\setminus B_{r-|t|}\).
Proof
Due to the \(L^p\)-boundedness of \(T^t\) and the density of \(\mathcal {S}(\mathbb {R}^d)\) in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), we can assume that \(f\in \mathcal {S}(\mathbb {R}^d)\).
-
(1)
Let \(s>0\). By integral representation of \(j_{\lambda _k}(z)\) (see, e.g., [2, Sect. 7.12]) we have
$$\begin{aligned} T^t(e^{-s|\cdot |^2})(x)&=R^t(e^{-s(\cdot )^2})(|x|)= c_{\lambda _k}\int _{0}^{\pi }e^{-s(|x|^2+t^2-2|x|t\cos \varphi )}\sin ^{2\lambda _k} \varphi \,\mathrm{d}\varphi \\&=e^{-s(|x|^2+t^2)}c_{\lambda _k}\int _{0}^{\pi }e^{2s|x|t\cos \varphi } \sin ^{2\lambda _k}\varphi \,\mathrm{d}\varphi \\&=e^{-s(|x|^2+t^2)}j_{\lambda _k}(2is|x|t), \end{aligned}$$and, in particular,
$$\begin{aligned} T^t(e^{-s|\cdot |^2})(x)\le e^{-s(|x|^2+t^2)}e^{2s|x|t}=e^{-s(|x|-t)^2}\le 1. \end{aligned}$$Using the self-adjointness of \(T^t\), we obtain
$$\begin{aligned} \int _{\mathbb {R}^d}T^tf(x)\,e^{-s|x|^2}\,\mathrm{d}\mu _k(x)=\int _{\mathbb {R}^d}f(x)T^t(e^{-s|\cdot |^2})(x)\,\mathrm{d}\mu _k(x). \end{aligned}$$Since for any \(t\in \mathbb {R}\), \(x\in \mathbb {R}^d\),
$$\begin{aligned} \lim \limits _{s\rightarrow 0}e^{-s|x|^2}=\lim \limits _{s\rightarrow 0}T^t(e^{-s|\cdot |^2})(x)=1, \end{aligned}$$by Lebesgue’s dominated convergence theorem we derive (1).
-
(2)
If \({\text {supp}}f\subset B_r\) and \(|x|>r+|t|\), then, in light of (3.4) and (3.3), for \(z\in {\text {supp}}\sigma _{x,t}^k\) and \(g\in G\), we have that
$$\begin{aligned} |z|\ge |gx|-|z-gx|=|x|-|z-gx|>r \end{aligned}$$and \(f(z)=0\), which yields \(T^tf(x)=0\).
If \({\text {supp}}f\subset \mathbb {R}^d\setminus B_r\), \(|x|<r-|t|\), then, for \(z\in {\text {supp}}\sigma _{x,t}^k\) and \(g\in G\), we similarly obtain \(|z|\le |gx|+|z-gx|=|x|+|z-gx|<r\), \(f(z)=0\), and \(T^tf(x)=0\). \(\square \)
Let g be a radial function, \(g(y)=g_0(|y|)\), where \(g_0(t)\) is defined on \(\mathbb {R}_+\). Note that by virtue of (2.2),
By means of operators \(T^t\) and \(\tau ^y\), define two convolution operators:
Note that operator (3.10) was defined in [48], while (3.11) was investigated in [48, 53].
Thangavelu and Xu [48] proved that if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1\le p\le \infty \), and \(g\in L^{1}_\mathrm {rad}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), then
and if \(1\le p\le 2\), \(g\in L^{p}_\mathrm {rad}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), then, for any \(y\in \mathbb {R}^d\),
Lemma 3.4
If \(f\in \mathcal {A}_k\), \(g_0\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\), \(g(y)=g_0(|y|)\), then, for any \(x, y\in \mathbb {R}^d\),
Proof
which gives
If \(g\in \mathcal {A}_k\), then, by (3.1),
and hence the first equality in (3.14) and the second equality in (3.15) are valid for \(g\in \mathcal {A}_k\).
Assuming that \(g_0\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda })\), \((g_n)_0\in \mathcal {S}(\mathbb {R}_+)\), \(g_{n}\rightarrow g\) in \(L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), and taking into account (3.8)–(3.11) and (3.13), we arrive at
Thus, the first equality in (3.14) holds.
Finally, using (3.1), we get
and the second part in (3.14) is valid. \(\square \)
Let \(y\in \mathbb {R}^d\) be given. Rösler [42] proved that the operator \(\tau ^y\) is positive on \(C^{\infty }_\mathrm {rad}(\mathbb {R}^d)\), i.e., \(\tau ^y\ge 0\), and moreover, for any \(x\in \mathbb {R}^d\),
where \(\rho _{x,y}^k\) is a radial probability Borel measure such that \({\text {supp}}\rho _{x,y}^k\subset B_{|x|+|y|}\).
Theorem 3.5
If \(1\le p\le \infty \), then, for any \(x\in \mathbb {R}^d\) and \(f\in \mathcal {S}(\mathbb {R}^d)\),
Proof
Let \(x\in \mathbb {R}^d\) be given. Let an operator \(B^x\) be defined on \(\mathcal {S}(\mathbb {R}^d)\) as follows (cf. (3.2) and (3.3)): for \(f\in \mathcal {S}(\mathbb {R}^d)\),
Let \(p=2\). We have
and
This, Hölder’s inequality, and the fact that the operators \(\mathcal {H}_{\lambda _k}\) and \(\mathcal {F}_{k}\) are unitary imply
which yields inequality (3.17) for \(p=2\). Moreover, \(B^x\) can be extended to the space \(L^{2}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\) with preservation of norm, and, moreover, this extension coincides with (3.2).
Let \(p=1\). By (3.14) and (3.16), we obtain
which is the desired inequality (3.17) for \(p=1\). Moreover, \(B^x\) can be extended to \(L^{1}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\) with preservation of norm such that the extension coincides with (3.2) on \(L^{1}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\cap L^{2}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\).
By the Riesz–Thorin interpolation theorem we obtain (3.17) for \(1<p<2\).
If \(2<p<\infty \), \(1/p+1/p'=1\), then by (3.14) and (3.13),
Finally, for \(p=\infty \), (3.17) follows from representation (3.3). \(\square \)
We are now in a position to prove the Young inequality for the convolutions (3.10) and (3.11).
Theorem 3.6
Let \(1\le p, q\le \infty \), \(\frac{1}{p}+\frac{1}{q}\ge 1\), and \(\frac{1}{r}=\frac{1}{p}+\frac{1}{q}-1\). We have that, for any \(f\in \mathcal {S}(\mathbb {R}^d)\), \(g_0\in \mathcal {S}(\mathbb {R}_+)\), and \(g\in \mathcal {S}_\mathrm {rad}(\mathbb {R}^d)\),
Proof
Since for \(g(y)=g_0(|y|)\) we have
it is enough to show inequality (3.18). The proof is straightforward using Hölder’s inequality and estimates (3.6) and (3.17). For the sake of completeness, we give it here. Let \(\frac{1}{\mu }=\frac{1}{p}-\frac{1}{r}\) and \(\frac{1}{\nu }=\frac{1}{q}-\frac{1}{r}\), then \(\frac{1}{\mu }\ge 0\), \(\frac{1}{\nu }\ge 0\), and \(\frac{1}{r}+\frac{1}{\mu }+\frac{1}{\nu }=1\). In virtue of (3.17), we have
Using (3.6), this gives
\(\square \)
Theorem 3.7
Let \(1\le p\le \infty \) and \(g\in \mathcal {S}_\mathrm {rad}(\mathbb {R}^d)\). We have that, for any \(y\in \mathbb {R}^d\),
Remark 3.8
Since \(\mathcal {S}(\mathbb {R}^d)\) is dense in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1\le p<\infty \), the operator \(\tau ^{y}\) can be defined on \(L^{p}_\mathrm {rad}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) so that (3.20) holds.
Proof
In the case \(1\le p\le 2\), this result was proved in [48]. The case \(p=\infty \) follows from (3.16).
Let \(2<p<\infty \). Since \(\mathcal {F}_k(g)\) is a radial function and
using (3.19) for \(r=\infty \), \(q=p\), we obtain
\(\square \)
Now we give an analogue of Lemma 3.4 for the case when \(f\in L^{p}\).
Lemma 3.9
Let \(1\le p\le \infty \), \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\cap C_b(\mathbb {R}^d)\cap C^{\infty }(\mathbb {R}^d)\), \(g_0\in \mathcal {S}(\mathbb {R}_{+})\), and \(g(y)=g_0(|y|)\). Then, for any \(x\in \mathbb {R}^d\),
and, in the sense of tempered distributions,
Proof
First, in light of (3.6) and (3.18), we note that the convolution (3.10) belongs to \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\). Moreover, (3.3) implies that it is in \(C_b(\mathbb {R}^d)\).
Taking into account that \(g\in \mathcal {S}(\mathbb {R}^d)\) and \( (-\Delta _k)^re_k(\cdot , z)=|z|^{2r}e_k(\cdot , z), \) we have
Let us show that the integral converges uniformly in x. We have
where \(G\in \mathcal {S}_\mathrm {rad}(\mathbb {R}^d)\) is such that \(\mathcal {F}_k(G)(z)=|z|^{2r}\mathcal {F}_k(g)(z)\). Using Hölder’s inequality and (3.20), we get
Thus, convolution (3.11) belongs to \(C^{\infty }(\mathbb {R}^d)\).
By Lemma 3.4, the equality in (3.21) holds for any function \(f\in \mathcal {S}(\mathbb {R}^d)\). If \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(f_n\in \mathcal {S}(\mathbb {R}^d)\), and \(f_n\rightarrow f\) in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), then Minkowski’s inequality and (3.6) give
while Hölder’s inequality and (3.20) imply
By (3.23), there is a subsequence \(\{n_k\}\) such that \((f_{n_k}*_{ \lambda _k}g_0)(x)\rightarrow (f*_{ \lambda _k}g_0)(x)\) a.e., therefore the relation \((f*_{ \lambda _k}g_0)(x)=(f*_{ k}g)(x)\) holds almost everywhere. Since both convolutions are continuous, it holds everywhere.
To prove the second equation of the lemma, we first remark that Lemma 3.4 implies that (3.22) holds pointwise for any \(f\in \mathcal {S}(\mathbb {R}^d)\). In the general case, since \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(\, (f*_{ \lambda _k}g_0)\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), and \(\mathcal {F}_k(g)\in \mathcal {S}(\mathbb {R}^d)\), the left- and right-hand sides of (3.22) are tempered distributions. Recall that the Dunkl transform of tempered distribution is defined by
Let \(f_n\in \mathcal {S}(\mathbb {R}^d)\) and \(f_n\rightarrow f\) in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(\varphi \in \mathcal {S}(\mathbb {R}^d)\). Then
and
Thus, the proof of (3.22) is now complete. \(\square \)
4 Boundedness of the Riesz Potential
Recall that \(\lambda _k=d/2-1+\sum _{a\in R_+}k(a)\). For \(0<\alpha <2\lambda _k+2\), the weighted Riesz potential \(I_{\alpha }^kf\) is defined on \(\mathcal {S}(\mathbb {R}^d)\) (see [49]) by
where \(d_k^{\alpha }=2^{-\lambda _k-1+\alpha }\Gamma (\alpha /2)/\Gamma (\lambda _k+1-\alpha /2)\). We have, in the sense of tempered distributions,
Using (2.2) and (3.2), we obtain
To estimate the \(L^p\)-norm of this operator, we use the maximal function defined for \(f\in \mathcal {S}(\mathbb {R}^d)\) as follows [48]:
where \(\chi _{B_r}\) is the characteristic function of the Euclidean ball \(B_r\) of radius r centered at 0.
Using (2.2), (3.2), and (3.14), we get
It is proved in [48] that the maximal function is bounded on \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1<p\le \infty \),
and it is of weak type (1, 1); that is,
Theorem 4.1
If \(1<p<q<\infty \), \(0<\alpha <2\lambda _k+2\), \(\frac{1}{p}-\frac{1}{q}=\frac{\alpha }{2\lambda _k+2}\), then
The mapping \(f\mapsto I_{\alpha }^kf\) is of weak type (1, q); that is,
Remark 4.2
In the case \(k\equiv 0\), inequality (4.4) was proved by Soboleff [44] and Thorin [50] and the weighted inequality was studied by Stein and Weiss [46]. For the reflection group \(G=\mathbb {Z}_2^d\), Theorem 4.1 was proved in [49]. The general case was obtained in [21]. We give another simple proof based on the \(L^p\)-boundedness of \(T^t\) given in Theorem 3.5 and follow the proof given in [49] for \(G=\mathbb {Z}_2^d\).
Remark 4.3
In Theorem 4.1, dealing with (4.4), we may assume that \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1<p<\infty \), while proving (4.5), we may assume that \(f\in L^{1}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\).
Proof
Let \(R>0\) be fixed. We write (4.1) as sum of two terms,
Integrating \(J_1\) by parts, we obtain
Here we have used that
since
In light of (4.7), we have
To estimate \(J_2\), we use Hölder’s inequality, the relation \(\frac{1}{p}-\frac{1}{q}=\frac{\alpha }{2\lambda _k+2}\), and (3.17):
for any \(R>0\). Choosing \(R=\left( M_kf(x)/\Vert f\Vert _{p, \mathrm{d}\mu _k}\right) ^{-q/(2\lambda _k+2)}\) implies the inequality
for any \(1\le p<q\). Integrating (4.9) and using (4.2), we have
Finally, we use inequality (4.3) for the maximal function and inequality (4.9) with \(p=1\) to obtain
\(\square \)
5 Entire Functions of Exponential Type and Plancherel–Polya–Boas-Type Inequalities
Let \(\mathbb {C}^d\) be the complex Euclidean space of d dimensions. Let also \(z=(z_1,\dots ,z_d)\in \mathbb {C}^d\), \(\mathrm {Im}\,z=(\mathrm {Im}\,z_1,\dots ,\mathrm {Im}\,z_d)\), and \(\sigma >0\).
In this section, we define several classes of entire functions of exponential type and study their interrelations. Moreover, we prove the Plancherel–Polya–Boas-type estimates and the Paley–Wiener-type theorems. These classes will be used later to study the approximation of functions on \(\mathbb {R}^d\) by entire functions of exponential type.
First, we define two classes of entire functions: \(B_{p, k}^\sigma \) and \(\widetilde{B}_{p, k}^\sigma \). We say that a function \(f\in B_{p, k}^\sigma \) if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) is such that its analytic continuation to \(\mathbb {C}^d\) satisfies
The smallest \(\sigma =\sigma _{f}\) in this inequality is called a spherical type of f. In other words, the class \(B_{p, k}^\sigma \) is the collection of all entire functions of spherical type at most \(\sigma \).
We say that a function \(f\in \widetilde{B}_{p, k}^\sigma \) if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) is such that its analytic continuation to \(\mathbb {C}^d\) satisfies
Historically, functions from \(\widetilde{B}_{p, k}^\sigma \) were basic objects in the Dunkl harmonic analysis. It is clear that \(\widetilde{B}_{p, k}^\sigma \subset B_{p, k}^\sigma \). Moreover, if \(k\equiv 0\), then both classes coincide (see, e.g., [29]). Indeed, if \(f\in B_{p, 0}^\sigma \), \(1\le p<\infty \), then Nikol’skii’s inequality [31, 3.3.5]
and the inequality [31, 3.2.6]
imply that, for \(z=x+iy\in \mathbb {C}^{d}\),
i.e., \(f\in \widetilde{B}_{p, 0}^\sigma \).
In fact, the classes \(B_{p, k}^\sigma \) and \(\widetilde{B}_{p, k}^\sigma \) coincide in the weighted case (\(k\ne 0\)) as well. To see that, it is enough to show that functions from \(B_{p, k}^\sigma \) are bounded on \(\mathbb {R}^d\).
Theorem 5.1
If \(0<p<\infty \), then \(B_{p, k}^\sigma =\widetilde{B}_{p, k}^\sigma \).
We will actually prove the more general statement. Let \(m\in \mathbb {Z}_+,\)\(\alpha ^1,\dots ,\alpha ^m\in \mathbb {R}^d\setminus \{0\}\), \(k_0\ge 0,\)\(k_1,\dots ,k_m>0\), and
be the power weight. The Dunkl weight is a particular case of such weighted functions. The weighted function (5.1) arises in the study of the generalized Fourier transform (see, e.g., [3]).
Let \(L^{p,v}(\mathbb {R}^d)\), \(0<p<\infty \), be the space of complex-valued Lebesgue measurable functions f for which
Let \(\varvec{\upsigma }=(\sigma _1,\dots ,\sigma _d)\), \(\sigma _1,\dots ,\sigma _d>0\).
Again, let us define three anisotropic classes of entire functions: \(B^{\varvec{\upsigma }}\), \(B_{p,v}^{\varvec{\upsigma }}\), and \(\widetilde{B}_{p,v}^{\varvec{\upsigma }}\).
We say that a function f defined on \(\mathbb {R}^d\) belongs to \(B^{\varvec{\upsigma }}\) if its analytic continuation to \(\mathbb {C}^d\) satisfies
We say that a function \(f\in B_{p,v}^{\varvec{\upsigma }}\) if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) is such that its analytic continuation to \(\mathbb {C}^d\) belongs to \(B^{\varvec{\upsigma }}\).
We say that a function \(f\in \widetilde{B}_{p,v}^{\varvec{\upsigma }}\) if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) is such that its analytic continuation to \(\mathbb {C}^d\) satisfies
We will use the notation \(L^p(\mathbb {R}^d)\), \(\Vert \cdot \,\Vert _{p}\), \(B_p^{\varvec{\upsigma }}\), and \(\widetilde{B}_p^{\varvec{\upsigma }}\) in the case of the unit weight, i.e., \(v\equiv 1\).
Theorem 5.2
If \(0<p<\infty \), then
-
(1)
\(B_{p,v}^{\varvec{\upsigma }}\subset B_{p}^{\varvec{\upsigma }}\),
-
(2)
\(B_{p,v}^{\varvec{\upsigma }}=\widetilde{B}_{p,v}^{\varvec{\upsigma }}\),
-
(3)
\(B_{p,v}^{\sigma }=\widetilde{B}_{p,v}^{\sigma }\).
Remark 5.3
- (i)
-
(ii)
Note that in some particular cases (\(k_0=0\) and \(p\ge 1\)) a similar result was discussed in [23].
Parts (2) and (3) of Theorem 5.2 follow from (1). Indeed, the embedding in (1) implies that \(B_{p,v}^{\varvec{\upsigma }}\subset B_{p}^{\varvec{\upsigma }}\subset B_{\infty }^{\varvec{\upsigma }}\). Hence, a function \(f\in B_{p,v}^{\varvec{\upsigma }}\) is bounded on \(\mathbb {R}^{d}\) and then \(f\in \widetilde{B}_{p,v}^{\varvec{\upsigma }}\), which gives (2). Further, \(B_{p,v}^{\sigma }\subset B_{p,v}^{\varvec{\upsigma }}\) holds, where \(\varvec{\upsigma }=(\sigma ,\dots ,\sigma )\in \mathbb {R}_{+}^{d}\) since \(|z|\le |z_{1}|+\dots +|z_{d}|\). Hence, similar to the above, we have \(B_{p,v}^{\sigma }\subset B_{\infty }^{\varvec{\upsigma }}\) and (3) follows. Thus, to prove Theorem 5.2, it is sufficient to verify part (1).
The main difficulty to prove Theorem 5.2 is that the weight v(x) vanishes. In order to overcome this problem, we will first prove two-sided estimates of the \(L^p\) norm of entire functions in terms of the weighted \(l_p\) norm, \(\left( \sum _{n} v(\lambda ^{(n)})|f(\lambda ^{(n)})|^p\right) ^{1/p}\), \(0<p<\infty \), where v does not vanish at \(\{\lambda ^{(n)}\}\subset \mathbb {R}^d\).
Such estimates are of their own interest. They generalize the Plancherel–Polya inequality [33, 6, Chapt. 6, 6.7.15]
where \(\lambda _k\) is an increasing sequence such that \(\lambda _{k+1}-\lambda _k\ge \delta >0\), and f is an entire function of exponential type at most \(\sigma \); and the Boas inequality [5, 6, Chapt. 10, 10.6.8],
where, additionally, \(\left| \lambda _k-\frac{\pi }{\sigma }\,k\right| \le L\) and the type of f is \(<\sigma \).
We write \(\varvec{\upsigma }'=(\sigma _1',\dots ,\sigma _d')<\varvec{\upsigma }=(\sigma _1,\dots ,\sigma _d)\) if \(\sigma _1'<\sigma _1,\dots ,\sigma _d'<\sigma _d\). Let \(n=(n_1,\dots ,n_d)\in \mathbb {Z}^d\) and \(\lambda ^{(n)}:\mathbb {Z}^d\rightarrow \mathbb {R}^d\). In what follows, we consider the sequences of the following type:
where \(\lambda _{i}^{(n)}=\lambda _i(n_1,\dots ,n_i)\) are sequences increasing with respect to \(n_i\), \(i=1,\dots ,d\) for fixed \(n_1,\dots ,n_{i-1}\).
Definition 5.4
We say that the sequence \(\lambda ^{(n)}\) satisfies the separation condition\(\Omega _\mathrm {sep}[\delta ]\), \(\delta >0\), if, for any \(n\in \mathbb {Z}^d\),
Note that if the sequence \(\lambda ^{(n)}\) satisfies the separation condition \(\Omega _\mathrm {sep}[\delta ]\), then it also satisfies the condition \(\inf _{n\ne m}|\lambda ^{(n)}-\lambda ^{(m)}|> 0\).
Definition 5.5
We say that the sequence \(\lambda ^{(n)}\) satisfies the close-lattice condition\(\Omega _\mathrm {lat}[\mathbf {a},L]\), \(\mathbf {a}=(a_1,\dots ,a_d)>0\), \(L>0\), if, for any \(n\in \mathbb {Z}^d\),
We start with the Plancherel–Polya-type inequality.
Theorem 5.6
Assume that \(\lambda ^{(n)}\) satisfies the condition \(\inf _{n\ne m}|\lambda ^{(n)}-\lambda ^{(m)}|> 0\). Then for \(f\in B^{\varvec{\upsigma }}_p\), \(0<p<\infty \), we have
Proof
For simplicity, we prove this result for \(d=2\). The proof in the general case is similar.
The function \(|f(z)|^p\) is plurisubharmonic, and therefore for any \(x=(x_1, x_2)\in \mathbb {R}^2\), one has [38]
where \(\rho _1,\,\rho _2>0\). Following [31, 3.2.5], for \(\delta >0\) and \(\xi +i\eta =(\xi _1+i\eta _1, \xi _2+i\eta _2)\), we obtain that
The separation condition implies that for some \(\delta >0\), the boxes \([\lambda _1^{(n)}-\delta , \lambda _1^{(n)}+\delta ]\times [\lambda _2^{(n)}-\delta , \lambda _2^{(n)}+\delta ]\) do not overlap for any n.
Since
where \(f^{(k)}\) is a partial derivative f of order \(k=(k_1,k_2)\), \(k!=k_1!\,k_2!\), and \((iy)^k=(iy_1)^{k_{1}}(iy_{2})^{k_{2}}\), by applying Bernstein’s inequality (see [31, 3.2.2 and 3.3.5] and [37]), we derive that
Using this and (5.4), we derive that
\(\square \)
Theorem 5.7
Let the sequence \(\lambda ^{(n)}\) of form (5.3) satisfy the conditions \(\Omega _\mathrm {sep}[\delta ]\) and \(\Omega _\mathrm {lat}[\varvec{\upsigma },L]\). Assume that \(f\in B^{\varvec{\upsigma }'}\), \(\varvec{\upsigma }'<\varvec{\upsigma }\), is such that \(\sum _{n\in \mathbb {Z}^d}|f(\lambda ^{(n)})|^p<\infty \), \(0<p<\infty \). Then \(f\in L^p(\mathbb {R}^d)\) and
Remark 5.8
For \(p\ge 1\), a similar two-sided Plancherel–Polya–Boas-type inequality was obtained from [32].
Proof
For simplicity, we consider the case \(d=2\). Integrating \(|f(x_1,x_2)|^p\) at \(x_1\) and applying inequality (5.2), we get, for any \(x_2\),
Since by (5.2), for any \(n_1\),
we then have
\(\square \)
Using Theorems 5.6 and 5.7 we arrive at the following statement:
Theorem 5.9
Let the sequence \(\{\lambda ^{(n)}\}\) of form (5.3) satisfy the conditions \(\Omega _\mathrm {sep}[\delta ]\) and \(\Omega _\mathrm {lat}[\varvec{\upsigma },L]\). If \(f\in B^{\varvec{\upsigma }'}\), \(\varvec{\upsigma }'<\varvec{\upsigma }\), then, for \(0<p<\infty \),
We will need the weighted version of the Plancherel–Polya–Boas equivalence. We start with three auxiliary lemmas.
Lemma 5.10
[18] If \(\gamma \ge -1/2\), then there exists an even entire function \(\omega _\gamma (z)\), \(z\in \mathbb {C}\), of exponential type 2 such that, uniformly in \(x\in \mathbb {R}_+\),
where \(k=[\gamma +1/2]\) and [a] is the integral part of a. In particular, we can take
Lemma 5.11
Let \(m\in \mathbb {N}\), \(j=1,\dots ,m\), \(b^j=(b_1^j,\dots ,b_d^j)\in \mathbb {R}^d\setminus \{0\}\), and either \(|b_i^j|\ge 1\), or \(b_i^j=0\), \(i=1,\dots ,d\). Then there exists a sequence \(\{\rho ^{(n)}\}\subset \mathbb {Z}^d\setminus \{0\}\) of the form (5.3) such that, for any \(j=1,\dots ,m\) and \(i=1,\dots ,d\),
Proof
To construct a desired sequence
we will use the following simple remark. If we throw out m points from \(\mathbb {Z}\), then the rest can be numbered such that the obtained sequence will be increasing and (5.5) holds.
Let \(J_1=\{j:b_1^j\ne 0,\, b_2^j=\dots =b_d^j=0\}\). If \(J_1=\varnothing \), then we set \(\rho _1(n_1)=n_1\). If \(J_1\ne \varnothing \), then \(\rho _1(n_1)\) is an increasing sequence formed from \(\mathbb {Z}\setminus \{0\}\). In both cases (5.5) is valid, and, moreover, for \(j\in J_1\) and any \(\rho _2(n_1,n_2),\dots ,\rho _d(n_1,\dots ,n_d)\), one has (5.6) since \(|\langle b^j, \rho ^{(n)}\rangle |=|b_1^j\rho _1(n_1)|\ge 1\).
Let \(J_2=\{j:b_2^j\ne 0,\, b_3^j=\dots =b_d^j=0\}\), \(n_1\in \mathbb {Z}\). If \(J_2=\varnothing \), then we set \(\rho _2(n_1, n_2)=n_2\). Let \(J_2\ne \varnothing \). If \(j\in J_2\) and \(b_1^j\rho _1(n_1)+b_2^jt_j=0\), then \(t_j=l_j+\varepsilon _j\), \(l_j\in \mathbb {Z}\), \(|\varepsilon _j|\le 1/2\). Here \(l_j\) is the nearest integer to \(t_j\). Note that if \(\rho _2\ne l_j\), then \(|b_1^j\rho _1(n_1)+b_2^j\rho _2|=|b_2^j(\rho _2-l_j-\varepsilon _j)|\ge 1/2\).
Let \(\rho _2(n_1, n_2)\) be an increasing sequence at \(n_2\) formed from \(\mathbb {Z}\backslash \{l_j:j\in J_2\}\). For this sequence (5.5) holds and, for \(j\in J_2\) and any \(\rho _3(n_1,n_2,n_3),\dots ,\rho _d(n_1,\dots ,n_d)\), one has
that is, (5.6) holds as well.
Assume that we have constructed the sets \(J_1,\dots ,J_{d-1}\), and the sequence \((\rho _1(n_1), \rho _2(n_1, n_2),\dots ,\rho _{d-1}(n_1,\dots ,n_{d-1}))\in \mathbb {Z}^{d-1}\).
Let \(J_d=\{j:b_d^j\ne 0\}\), \((n_1,\dots ,n_{d-1})\in \mathbb {Z}^{d-1}\). If \(J_d=\varnothing \), then we set \(\rho _d(n_1,\dots ,n_{d-1},n_d)=n_d\). Assume now that \(J_d\ne \varnothing \). If \(j\in J_d\) and
then \(t_j=l_j+\varepsilon _j\), \(|\varepsilon _j|\le 1/2\). Note that if \(\rho _d\ne l_j\), then
Let \(\rho _d(n_1,\dots ,n_d)\) be an increasing sequence in \(n_d\) formed from \(\mathbb {Z}\setminus \{l_j:j\in J_d\}\), \(\rho ^{(n)}=(\rho _1(n_1), \rho _2(n_1, n_2),\dots ,\rho _d(n_1,\dots ,n_d))\). For the sequence \(\rho _d(n_1,\dots ,n_d)\), inequality (5.5) holds, and, for \(j\in J_d\), one has \(|\langle b^j, \rho ^{(n)}\rangle |\ge 1/2\).
Thus, we construct the desired sequence since, for any \(j\in \{1,\dots ,m\}\) and some \(i\in \{1,\dots ,d\}\), \(b^j\in J_i\) holds. \(\square \)
An important ingredient of the proof of Theorem 5.2 is the following corollary of Lemma 5.11:
Lemma 5.12
If \(\mathbf {a}>0\), \(\alpha ^1,\dots ,\alpha ^m\in \mathbb {R}^d\setminus \{0\}\), then there exists a sequence \(\lambda ^{(n)}\) of the form (5.3) such that for some \(\delta ,\,L>0\) the conditions \(\Omega _\mathrm {sep}[\delta ]\), \(\Omega _\mathrm {lat}[\mathbf {a}, L]\), and \(\xi _{j}(\lambda ^{(n)})\ge \delta \), \(j=0,1,\dots ,m\), \(n\in \mathbb {Z}^d\), hold, where
Indeed, for \(m\ge 1\), it is enough to define
where \(\rho ^{(n)}\) is the sequence defined in Lemma 5.11. For \(m=0\) in (5.8), we can take \(\{\rho ^{(n)}\}=\mathbb {Z}^{d}\setminus \{0\}\).
We are now in a position to state the Plancherel–Polya–Boas inequalities with weights.
Theorem 5.13
Let \(f\in B^{\varvec{\upsigma }}\) and \(\lambda ^{(n)}\) be the sequence satisfying all conditions of Lemma 5.12 with some \(\mathbf {a}>\varvec{\upsigma }\). Then, for \(0<p<\infty \),
Proof
Recall that \(v(x)=\prod _{j=0}^mv_j(x)\), where \(v_j(x)=\xi _j^{k_j}(x)\), \(j=0,1,\dots ,m\) (see (5.1) and (5.7)).
By Lemma 5.10, we construct an entire function of exponential type
where \(w_0(z)=\omega _{\gamma _{0}}(|z|)\), \(w_j(z)=\omega _{\gamma _j}(\langle \alpha ^j, z\rangle )\), \(j=1,\dots ,m\), and
For \(j=0,1,\dots ,m\), we have \(w_j\in B^{2\varvec{\upmu }^j}\), where
and \(w\in B^{2\varvec{\upmu }}\), \(\varvec{\upmu }=\sum _{j=0}^m\varvec{\upmu }^j\). Moreover, for any \(j=0,1,\dots ,m\),
Let \(f\in B^{\varvec{\upsigma }}_{p,v}\), \(0<p<\infty \), \(\varvec{\upsigma }<\mathbf {a}\), and \(\lambda ^{(n)}\) be the sequence satisfying all conditions of Lemma 5.12. Then, for some \(s>0\) such that \(\varvec{\upsigma }+2s\varvec{\upmu }<\mathbf {a}\), we have that \(f(x)w(sx)\in B^{\varvec{\upsigma }+2s\varvec{\upmu }}\).
Using Theorem 5.6 and properties (5.9), we derive
Let \(\delta >0\), \(J\subset J_{m}:=\{0,1,\dots ,m\}\) or \(J=\varnothing \),
Since \(f(x)\prod _{j\in J}w_j(sx)\in B^{\varvec{\upsigma }+2s\varvec{\upmu }}\), using Theorems 5.6 and 5.7 and properties (5.9) for \(\delta \) from Lemma 5.12, we obtain
where we have assumed that \(\prod _{j\in \varnothing }=1\). \(\square \)
Proof of Theorem 5.2
Recall that it is enough to show that \(B_{p,v}^{\varvec{\upsigma }}\subset B_{p}^{\varvec{\upsigma }}\), and the latter follows from \(B^{\varvec{\upsigma }}_{p,v}\subset L^{p}(\mathbb {R}^d)\).
Let \(f\in B^{\varvec{\upsigma }}_{p,v}\), \(0<p<\infty \), \(\mathbf {a}>\varvec{\upsigma }\), and \(\lambda ^{(n)}\) be the sequence satisfying all conditions of Lemma 5.12. Using Theorem 5.6 and properties (5.9) as in Theorem 5.13, we have
\(\square \)
By the Paley–Wiener theorem for tempered distributions (see [25, 53]) and Theorem 5.1, we arrive at the following result.
Theorem 5.14
A function \(f\in B_{p, k}^\sigma \), \(1\le p<\infty \), if and only if \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\cap C_b(\mathbb {R}^d)\) and \({\text {supp}}\mathcal {F}_k(f)\subset B_\sigma \).
The Dunkl transform \(\mathcal {F}_k(f)\) in Theorem 5.14 is understood as a function for \(1\le p\le 2\) and as a tempered distribution for \(p>2\).
We conclude this section by presenting the concept of the best approximation. Let
be the best approximation of a function \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) by entire functions of spherical exponential type \(\sigma \). We show that the best approximation is achieved.
Theorem 5.15
For any \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), \(1\le p\le \infty \), there exists a function \(g^{*}\in B_{p, k}^\sigma \) such that \(E_{\sigma }(f)_{p,\mathrm{d}\mu _k}=\Vert f-g^{*}\Vert _{p,\mathrm{d}\mu _k}\).
Proof
The proof is standard. Let \(g_n\) be a sequence from \(B_{p, k}^\sigma \) such that \(\Vert f-g_n\Vert _{p,\mathrm{d}\mu _k}\rightarrow E_{\sigma }(f)_{p,\mathrm{d}\mu _k}\). Since it is bounded in \(L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\), it is also bounded in \(C_b(\mathbb {R}^d)\). A compactness theorem for entire functions [31, 3.3.6] implies that there exist a subsequence \(g_{n_k}\) and an entire function \(g^{*}\) of exponential type at most \(\sigma \) such that
and, moreover, convergence is uniform on compact sets. Therefore, for any \(R>0\),
Letting \(R\rightarrow \infty \), we have that \(g^{*}\in B_{p, k}^\sigma \). In light of
we have \(\Vert f-g^{*}\Vert _{p,\mathrm{d}\mu _k}\le E_{\sigma }(f)_{p,\mathrm{d}\mu _k}\). \(\square \)
6 Jackson’s Inequality and Equivalence of Modulus of Smoothness and K-Functional
6.1 Smoothness Characteristics and K-Functional
We define the r-th power of the Dunkl Laplacian as a tempered distribution:
The Dunkl Laplacian can also be written in terms of the Dunkl transform
Let \(W^{2r}_{p, k}\) be the Sobolev space, that is,
equipped with the Banach norm
Note that \((-\Delta _k)^rf\in \mathcal {S}(\mathbb {R}^d)\) whenever \(f\in \mathcal {S}(\mathbb {R}^d)\) and \(\mathcal {S}(\mathbb {R}^d)\) is dense in \(W^{2r}_{p, k}\). Indeed, if \(f\in W^{2r}_{p, k}\), defining
we obtain that \((f*_{ k}\Phi _{\varepsilon })\in \mathcal {S}(\mathbb {R}^d)\) and (see [48])
Define the K-functional for the couple \((L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k}), W^{2r}_{p, k})\) as follows:
Note that for any \(f_1,f_2\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) and \(g\in W^{2r}_{p, k}\), we have
and hence,
If \(f\in W^{2r}_{p, k}\), then \(K_{2r}(t, f)_{p,\mathrm{d}\mu _k}\le t^{2r}\Vert (-\Delta _k)^rf\Vert _{p,\mathrm{d}\mu _k}\) and \(\lim _{t\rightarrow 0}K_{2r}(t, f)_{p,\mathrm{d}\mu _k}=0\). This and (6.2) imply that, for any \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\),
Another important property of the K-functional is
Let I be an identical operator and \(m\in \mathbb {N}\). Consider the following three differences:
Differences (6.5) and (6.6) coincide with the classical difference for the translation operator \(T^{t}f(x)=f(x+t)\) and correspond to the usual definition of the modulus of smoothness of order m. Difference (6.7) can be seen as follows. Define \(\mu _s=(-1)^s\left( {\begin{array}{c}m\\ s\end{array}}\right) \), \(s\in \mathbb {Z}\). Then the convolution \(\mu *\mu \) is given by
Note that \(\nu _s\ne 0\) if \(|s|\le m\). Moreover, if \(k\equiv 0\), then
where the operator \(S^{t}\) was given in (3.5) and the averages
were defined by Dai and Ditzian in [9].
Definition 6.1
The moduli of smoothness of a function \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\) are defined by
Let us mention some basic properties of these moduli of smoothness. Define by \(\Omega _m(\delta , f)_{p,\mathrm{d}\mu _k}\) any of the three moduli in Definition 6.1. Using the triangle inequality, estimate (3.6) reveals
and
If \(f\in \mathcal {S}(\mathbb {R}^d)\), then, by (3.2),
where \(\lambda _k=d/2-1+\sum _{a\in R_+}k(a)>-1/2\),
and
These formulas alow us to prove the following remark, which will be important further in Theorem 6.6.
Remark 6.2
The functions \(j_{\lambda _{k},m}(t)\) and \(j^{**}_{\lambda _{k},m}(t)\) have zero of order 2m at the origin, while the function \(j^{*}_{\lambda _{k},m}(t)\) has zero of order \(m+1\) if m is odd and of order m if m is even.
Indeed, first we study \(j_{\lambda _{k},m}(t)=(1-j_{\lambda _{k}}(t))^{m}\). Since, for any t,
we get \(j_{\lambda _{k},m}(t)\asymp t^{2m}\) as \(t\rightarrow 0\). Second, since
(see [36, Sect. 4.2]), using (6.14), we obtain that \(j_{\lambda _{k},m}^*(t)\asymp t^{2[(m+1)/2]}\). Finally, taking into account
(see [36, Sect. 4.2]) and using again (6.14), we arrive at \(j^{**}_{\lambda _{k},m}(t)\asymp t^{2m}\). Some of these properties were known (see [9, 34, 35]).
Remark 6.3
In the paper [9], the authors obtained that \(j^{**}_{\lambda _{k},m}(t)>0\) for \(t>0\).
6.2 Main Results
First we state the Jackson-type inequality.
Theorem 6.4
Let \(\sigma > 0\), \(1\le p \le \infty \), \(r\in \mathbb {Z}_+\), \(m\in \mathbb {N}\). We have, for any \(f\in W_{p, k}^{2r}\),
where \(\Omega _m\) is any of the three moduli of smoothness (6.8)–(6.10).
Remark 6.5
-
(i)
For radial functions, inequality (6.15) is the Jackson inequality in \(L^{p}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\). In this case it was obtained in [34, 35] for moduli (6.8) and (6.9). For \(k\equiv 0\) and the modulus of smoothness (6.10), inequality (6.15) was obtained by Dai and Ditzian [9], see also the paper [10].
-
(ii)
From the proof of Theorem 6.4, we will see that inequality (6.15) for moduli (6.8) and (6.10) can be equivalently written as
$$\begin{aligned} E_{\sigma }(f)_{p,\mathrm{d}\mu _k}&\lesssim \frac{1}{\sigma ^{2r}} \left\| \varDelta _{1/\sigma }^m ((-\Delta _k)^rf) \right\| _{p,\mathrm{d}\mu _k},\\ E_{\sigma }(f)_{p,\mathrm{d}\mu _k}&\lesssim \frac{1}{\sigma ^{2r}} \left\| {^{**}}\varDelta _{1/\sigma }^m ((-\Delta _k)^rf) \right\| _{p,\mathrm{d}\mu _k}. \end{aligned}$$
The next theorem provides an equivalence between moduli of smoothness and the K-functional.
Theorem 6.6
If \(\delta > 0\), \(1\le p\le \infty \), \(r\in \mathbb {N}\), then for any \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\),
Remark 6.7
If \(k\equiv 0\), the equivalence between the classical modulus of smoothness and the K-functional is well known [8, 26], while the equivalence between modulus (6.10) and the K-functional was shown in [9]. For radial functions, a partial result of (6.16), more precisely, an equivalence of the K-functional and moduli of smoothness (6.8) and (6.9), was proved in [34, 35].
Remark 6.8
One can continue equivalence (6.16) as follows (see also Remark 6.16):
We give the proof for the difference (6.7) and the modulus of smoothness (6.10). We partially follow the proofs in [27, 34, 35], which are different from those given in [9]. For moduli of smoothness (6.8) and (6.9), the proofs are similar and will be omitted here (see also [34, 35]). The proof makes use of radial multipliers and is based on boundedness of the translation operator \(T^t\). Note that by (6.2) and (6.11), the K-functional and moduli of smoothness depend continuously on a function. Moreover, the best approximations also depend continuously on a function, and therefore one can assume that functions belong to the Schwartz space.
6.3 Properties of the de la Vallée Poussin Type Operators
Let \(\eta \in \mathcal {S}_\text {rad}(\mathbb {R}^d)\) be such that \(\eta (x)=1\) if \(|x|\le 1\), \(\eta (x)>0\) if \(|x|<2\), and \(\eta (x)=0\) if \(|x|\ge 2\). We write
where \(\mathcal {F}_{k}(\eta _r)\) is a tempered distribution. If \(t=|x|\), \(\eta _0(t)=\eta (x)\), and \(\eta _{r 0}(t)=\eta _r(x)\), then \(\mathcal {F}_{k}(\eta _r)(y)=\mathcal {H}_{\lambda _k}(\eta _{r 0})(|y|)\).
Lemma 6.9
We have \(\widehat{\eta }_{k,r}\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\), where \(r>0\).
Proof
It is sufficient to prove that \(\mathcal {H}_{\lambda _k}(\eta _{r 0})\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\). In the case \(r\ge 1\), this was proved in [35, (4.25)]. We give the proof for any \(r>0\).
Letting \(u_j(t)=(1+t^2)^{-j}\) and taking into account that
we obtain, for any \(M\in \mathbb {N}\) and \(t\ge 0\),
For any \(r>0\), we have \(\mathcal {H}_{\lambda _k}(u_r)\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\) (see [35, Lemma 3.2], [47, Chapt 5, 5.3.1], [31, Chapt 8, 8.1]); therefore \(\mathcal {H}_{\lambda _k}(\psi _1)\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\). Because of \(\psi _2\in \mathcal {S}(\mathbb {R}_+)\), \(\mathcal {H}_{\lambda _k}(\psi _2)\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\). Thus, we are left to show that, for sufficiently large M, \(\mathcal {H}_{\lambda _k}(\psi _3)\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\).
Let \(M+r> \lambda _k+1\), \(t\ge 1\). Since \(\frac{\Gamma (j+r)}{\Gamma (j+1)}\lesssim j^{r-1}\), we have
and
Thus, \(\psi _3\in L^{1}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda _k})\), \(\mathcal {H}_{\lambda _k}(\psi _3)\in C(\mathbb {R}_{+})\), and \(\mathcal {H}_{\lambda _k}(\psi _3)\in L^{1}([0,2],\mathrm{d}\nu _{\lambda _k})\).
Recall that the Bessel differential operator is defined by
Using \(\psi _3\in C^{\infty }(\mathbb {R}_{+})\), we have, for any \(s\in \mathbb {N}\), \(\mathcal {B}_{\lambda _k}^s\psi _3\in L^{1}([0,2],\mathrm{d}\nu _{\lambda _k})\).
If \(t\ge 2\), then \((1-\eta _0(t))u_{j+r}(t)=u_{j+r}(t)\) and
This gives
By induction on s,
and then, for \(t\ge 2\),
and \(\mathcal {B}_{\lambda _k}^s\psi _3\in L^{1}([2,\infty ),\mathrm{d}\nu _{\lambda _k})\). Thus, we have \(\mathcal {B}_{\lambda _k}^s\psi _3\in L^{1}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\) for any s. Choosing \(s>\lambda _k+1\) and using the inequality
we arrive at \(\mathcal {H}_{\lambda _k}(\psi _3)\in L^{1}([2,\infty ),\mathrm{d}\nu _{\lambda _k})\). Finally, we obtain that \(\mathcal {H}_{\lambda _k}(\psi _3)\in L^{1}(\mathbb {R}_+,\mathrm{d}\nu _{\lambda _k})\). \(\square \)
For \(m, r\in \mathbb {N}\) and \(m\ge r\), we set
Since
and
boundedness of the operator \(T^s\) in \(L^{1}(\mathbb {R}^d,\mathrm{d}\mu _k)\) and Lemma 6.9 imply that
Lemma 6.10
Let \(m, r \in \mathbb {N}\), \(m \ge r\), \(1\le p\le \infty \), and \(f\in \mathcal {S}(\mathbb {R}^d)\). We have
and
Proof
Combining (3.15), (6.1), (6.12), and (6.17), we obtain that
Then (6.18) follows from (3.11) and Lemma 3.4. Inequality (6.19) follows from (6.17), (6.18), Lemma 3.4, and (3.12). Note that a constant in (6.19) can be taken as \(\Vert g_{m,r}\Vert _{1,\mathrm{d}\mu _{k}}\). \(\square \)
Remark 6.11
Since \(\mathcal {S}(\mathbb {R}^d)\) is dense in \(W_{p, k}^{2r}\), in light of (6.11), inequality (6.19) holds for any function from \(W_{p, k}^{2r}\).
Let \(f\in \mathcal {S}(\mathbb {R}^d)\). We set \(\theta (x)=\mathcal {F}_k(\eta )(x)\) and \(\theta _\sigma (x) = \theta (x/\sigma )\). Then \(\theta \), \(\theta _\sigma \in \mathcal {S}(\mathbb {R}^d)\). The de la Vallée Poussin type operator is given by \(P_\sigma (f)= f*_{ k}\theta _\sigma \). By Lemma 3.4,
Lemma 6.12
If \(\sigma >0\), \(1 \le p\le \infty \), \(f\in \mathcal {S}(\mathbb {R}^d)\), then
-
(1)
\(P_\sigma (f)\in B_{p, k}^{2\sigma }\) and \(P_\sigma (g)=g\) for any \(g\in B_{p, k}^\sigma \);
-
(2)
\(\Vert P_\sigma (f)\Vert _{p,\mathrm{d}\mu _{k}} \lesssim \Vert f\Vert _{p,\mathrm{d}\mu _{k}}\);
-
(3)
\(\Vert f-P_\sigma (f)\Vert _{p,\mathrm{d}\mu _{k}} \lesssim E_{\sigma } (f)_{p,\mathrm{d}\mu _{k}}\).
Remark 6.13
Property (3) in this lemma means that \(P_\sigma (f)\) is the near best approximant of f in \(L^p(\mathbb {R}^d,\mathrm{d}\mu _{k})\).
Proof
(1) We observe that \({\text {supp}}\eta (\cdot /\sigma )\subset B_{2\sigma }\) and then \({\text {supp}}\mathcal {F}_k(P_\sigma (f))\subset B_{2\sigma }\). Theorem 5.14 yields \(P_\sigma (f)\in B_{p, k}^{2\sigma }\). If \(g\in B_{p,k}^\sigma \), then by Theorem 5.14, \({\text {supp}}\mathcal {F}_k(g)\subset B_{\sigma }\) and \(\mathcal {F}_k(P_\sigma (g))(y)=\eta (y/\sigma )\mathcal {F}_k(g)(y)=\mathcal {F}_k (g)(y).\) Hence, \(P_\sigma (g)=g\).
(2) In light of (3.12),
(3) Using Theorem 5.15, there exists an entire function \(g^*\in B_{p, k}^\sigma \) such that \(\Vert f-g^*\Vert _{p,\mathrm{d}\mu _{k}}=E_\sigma (f)_{p,\mathrm{d}\mu _{k}}\). Then using \(P_\sigma (g^*)=g^*\) implies
\(\square \)
In the proof of the next lemma we will use the estimate
which follows, by induction on n, from the known properties of the Bessel function [2, Chap. 7]
Lemma 6.14
If \(\sigma >0\), \(1 \le p\le \infty \), \(m\in \mathbb {N}\), \(r\in \mathbb {Z}_+\), \(f\in \mathcal {S}(\mathbb {R}^d)\), then
for some \(a=a(\lambda _k, m)>0\).
Proof
We have
where
Setting \(j^{**}_{\lambda _{k},m}(t)= 1- \tau _0(t)\), in light of (6.13) and (6.20), we observe that \(j^{**}_{\lambda _{k},m}(t)\rightarrow 1\) as \(t\rightarrow \infty \). Then we can choose \(a>0\) such that \(|\tau _0(t)| \le 1/2\) for \(|t| \ge a/2\). For such \(a=a(\lambda _k, m)\), we have that \(\varphi (y)=0\) for \(|y|\le 1/2\), \(\varphi (y)>0\) for \(|y|> 1/2\), and \(\varphi \in C^{\infty }(\mathbb {R}^d)\). Moreover, the derivatives \(\varphi ^{(k)}(y)\) grow at infinity not faster than \(|y|^{a_k}\), which yields \(\varphi \in \mathcal {S}'(\mathbb {R}^d)\).
We will use the following decomposition:
where
and
First, we show that \(\mathcal {F}_k(\varphi _1(|\cdot |))\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\). Since for a radial function we have
and, for \(|t|\le 1/2\),
then, by (6.13) and (6.20), we obtain
Hence, for a fixed \(N\ge 2+2/(2\lambda _k+1)\), we have \(\Delta _k^s\varphi _1(|y|)\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\), where \(s\in \mathbb {Z}_+\). Applying (6.1) we derive that
Setting \(s>\lambda _k+1\) yields \(\mathcal {F}_k(\varphi _1(|\cdot |))\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\).
Second, let us show that \(\mathcal {F}_k(\varphi _2(|\cdot |))\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\) for \(r\in \mathbb {N}\). Let
Boundedness of the operator \(T^t\) in \(L^{p}(\mathbb {R}^d, \mathrm{d}\mu _k)\) implies
and
Then for \(p=1\), taking into account Lemma 6.9, we have
Thus, \( \mathcal {F}_k(\varphi ) \in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\). Combining Lemma 3.4, relations (3.12), (6.22), (6.23), and the formula \(\Vert \mathcal {F}_k(\varphi (\cdot /\sigma ))\Vert _{1,\mathrm{d}\mu _k}=\Vert \mathcal {F}_k(\varphi )\Vert _{1,\mathrm{d}\mu _k}\), we obtain inequality (6.21) for \(r\in \mathbb {N}\).
Now let \(r=0\). Define the operators \(A_1\) and \(A_2\) as follows:
and
Since \(\mathcal {F}_k(\varphi _1(|\cdot |))\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\),
We are left to show that
We have
Since \(B^{a/\sigma }g\in \mathcal {S}(\mathbb {R}^d)\), using Lemma 6.12 and inequality (6.24), we get
Using (6.25) and (6.26) with \(g= {^{**}}\varDelta _{a/\sigma }^m f\), we finally obtain (6.21) for \(r=0\). \(\square \)
Lemma 6.15
If \(\sigma >0\), \(1 \le p\le \infty \), \(m\in \mathbb {N}\), \(f\in \mathcal {S}(\mathbb {R}^d)\), then
where \(a=a(\lambda _k, m)>0\) is given in Lemma 6.14.
Proof
We have
where
Since \(j^{**}_{\lambda _{k},m}(a|y|/2)/|y|^{2m}>0\) for \(|y|>0\), we observe that \(\varphi \in \mathcal {S}(\mathbb {R}^d)\) and \(\mathcal {F}_k(\varphi )\in L^{1}(\mathbb {R}^d, \mathrm{d}\mu _k)\). Then estimate (6.27) follows from Lemma 3.4, Young’s inequality (3.12), and \(\Vert \mathcal {F}_k(\varphi (\cdot /\sigma ))\Vert _{1,\mathrm{d}\mu _k}=\Vert \mathcal {F}_k(\varphi )\Vert _{1,\mathrm{d}\mu _k}\). \(\square \)
6.4 Proofs of Theorems 6.4 and 6.6
Proof of Theorem 6.6
In connection with Lemma 6.10 and Remark 6.11, observe that, for \(f\in \mathcal {S}(\mathbb {R}^d)\) and \(g\in W_{p, k}^{2r}\),
Then
On the other hand, \(P_\sigma (f)\in W_{p, k}^{2r}\) and
In light of Lemma 6.14,
Further, Lemma 6.15 yields
Setting \(\sigma =a/(2\delta )\), from (6.29)–(6.30) we arrive at
Proof of Theorem 6.4
Using property (6.4) and inequalities (6.28) and (6.31), we obtain
Remark 6.16
The proofs of estimates (6.31) and (6.32) for the difference (6.7) are based on the fact that the parameter a in Lemmas 6.14 and 6.15 is the same. It is possible due to the fact that \(j^{**}_{\lambda _{k},m}(t)>0\) for \(t>0\), see Remark 6.3. This estimate is valid for the difference (6.5) as well, since \(j_{\lambda _{k},m}(t)=(1-j_{\lambda _{k}}(t))^m>0\) for \(t>0\).
Therefore, the moduli of smoothness (6.8) and (6.10) in inequalities (6.15) and (6.16) can be replaced by the norms of the corresponding differences (6.5) and (6.7). For the modulus of smoothness (6.9), this observation is not valid since \(j^{*}_{\lambda _{k},m}(t)\) does not keep its sign.
Remark 6.17
Properties (6.3) and (6.4) of the K-functional and the equivalence (6.16) imply the following properties of moduli of smoothness:
7 Some Inequalities for Entire Functions
In this section, we study weighted analogues of the inequalities for entire functions. In particular, we obtain Nikolskii’s inequality ([31], see Theorem 7.1 below), Bernstein’s inequality ([31], Theorem 7.3), Nikolskii–Stechkin’s inequality ([30, 45], Theorem 7.5), and Boas-type inequality ([4], Theorem 7.7).
Theorem 7.1
If \(\sigma >0\), \(0<p\le q\le \infty \), \(f\in B_{p, k}^\sigma \), then
Remark 7.2
Observe that the obtained Nikolskii inequality is sharp, i.e., we actually have
and an extremizer can be taken as
for sufficiently large \(m\in \mathbb {N}\).
Proof
Let \(f\in B_{p, k}^\sigma \), \(p\ge 1\), \(q=\infty \). By Theorem 5.14, we have \({\text {supp}}\mathcal {F}_k(f)\subset B_\sigma \), and then
Lemma 3.9 implies
Taking into account that
Hölder’s inequality and Theorem 3.5 yield
i.e., (7.1) holds.
Let \(f\in B_{p, k}^\sigma \), \(0<p<1\), \(q=\infty \). By Theorem 5.1, f is bounded and \(f\in B_{1, k}^\sigma \). We have
Using (7.1) with \(p=1\) and \(q=\infty \),
which gives
Thus, the proof of (7.1) for \(q=\infty \) is complete.
If \(0<p\le q<\infty \), we obtain
\(\square \)
Theorem 7.3
If \(\sigma >0\), \(r\in \mathbb {N}\), \(1 \le p\le \infty \), \(f\in B_{p, k}^\sigma \), then
Proof
It is enough to consider the case \(r=1\). As in the previous theorem, we use (7.2) to obtain
where \(\varphi _0(t)=t^2\eta _0(t)\in \mathcal {S}(\mathbb {R}_+)\). Combining Lemma 3.9, inequality (3.12), and \(\Vert \mathcal {F}_k(\varphi _0(|\cdot |/\sigma ))\Vert _{1,\mathrm{d}\mu _k}=\Vert \mathcal {F}_k (\varphi _0(|\cdot |))\Vert _{1,\mathrm{d}\mu _k}\), we arrive at
\(\square \)
The next result follows from Lemma 6.10, Remark 6.11, and Theorem 7.3.
Corollary 7.4
If \(\sigma ,\,\delta >0\), \(m\in \mathbb {N}\), \(1 \le p\le \infty \), \(f\in B_{p, k}^\sigma \), then
where constants do not depend on \(\sigma , \delta ,\) and f.
Theorem 7.5
If \(\sigma >0\), \(m\in \mathbb {N}\), \(1 \le p\le \infty \), \(0<t\le 1/(2\sigma )\), \(f\in B_{p, k}^\sigma \), then
Remark 7.6
By Remark 6.8, this inequality can be equivalently written as
Proof
We have
Since for \(0<t\le 1/(2\sigma )\),
we obtain that
where
Using
and
and combining Lemma 3.9 and inequality (3.12), we have
\(\square \)
Theorem 7.7
If \(\sigma >0\), \(m\in \mathbb {N}\), \(1 \le p\le \infty \), \(0<\delta \le t\le 1/(2\sigma )\), \(f\in B_{p, k}^\sigma \), then
Remark 7.8
Using Remark 6.8, Theorem 7.5, and taking into account that \(\delta ^{-2m} K_{2m}(\delta , f)_{p,\mathrm{d}\mu _k}\) is decreasing in \(\delta \) (see (6.4)), inequality (7.5) can be equivalently written as
Proof
We have
where \(\theta =(\delta /t)^{2m}\in (0, 1]\),
Using Lemma 3.9 and estimate (3.12), we arrive at inequality (7.5):
provided that the function \(n(\theta )=\Vert \mathcal {F}_k(\varphi _{\theta })\Vert _{1,\mathrm{d}\mu _k}\) is continuous on [0, 1]. Let us prove this.
Set \(\varphi _{\theta }(y)=\varphi _{\theta 0}(|y|)\), \(r=|y|\), \(\rho =|x|\). Then
The inner integral continuously depends on \(\theta \). Let us show that the outer integral converges uniformly in \(\theta \in [0,1]\). Since [2, Sect. 7.2]
integrating by parts implies
where
This and (6.20) give
and, for \(s>\lambda _k+3/2\),
completing the proof. \(\square \)
Remark 7.9
Combining (7.1) and (7.3), the following Bernstein–Nikolskii inequality is valid:
Remark 7.10
For radial functions, Nikolskii inequality (7.1), Bernstein (7.3), Nikolskii–Stechkin (7.4), and Boas inequality (7.5) follow from corresponding estimates in the space \(L^{p}(\mathbb {R}_{+},\mathrm{d}\nu _{\lambda })\) proved in [34].
8 Realization of K-Functionals and Moduli of Smoothness
In the nonweighted case (\(k\equiv 0\)) the equivalence between the classical modulus of smoothness and the K-functional between \(L^p\) and the Sobolev space \(W_p^{r}\) is well known [8, 26]: \(1\le p \le \infty \), for any integerrone has
where
Starting from the paper [13] (see also [17, Lemma 1.1] for the fractional case), the following equivalence between the modulus of smoothness and the realization of the K-functional is widely used in approximation theory:
where g is an entire function of exponential type 1 / t.
Let the realization of the K-functional \(K_{2r}(t, f)_{p,\mathrm{d}\mu _k}\) be given as follows:
and
where \(g^*\in B_{p, k}^{1/t}\) is a near best approximant.
Theorem 8.1
If \(t> 0\), \(1\le p\le \infty \), \(r\in \mathbb {N}\), then for any \(f\in L^{p}(\mathbb {R}^{d},\mathrm{d}\mu _{k})\),
Proof
By Theorem 6.6,
where we have used the fact that \(B_{p, k}^{1/t}\subset W_{p, k}^{2r}\), which follows from Theorem 7.3.
Therefore, it is enough to show that
Indeed, for \(g^*\) being the best approximant (or near best approximant), the Jackson inequality given in Theorem 6.4 implies that
Using the first inequality in Theorem 7.5 and taking into account (8.1), we have
Using again (8.1), we arrive at
completing the proof.
The next result answers the following important question (see, e.g., [22, 51]): when does the relation
(or similar relations with concepts in Theorem 8.2) hold?
Theorem 8.2
Let \(1\le p\le \infty \) and \(m\in \mathbb {N}\). We have that (8.2) is valid if and only if
Proof
We prove only the nontrivial part that (8.3) implies (8.2). Since, by (6.4), we have \(\omega _m(nt, f)_{p,\mathrm{d}\mu _k} \lesssim n^{2m} \omega _m(t, f)_{p,\mathrm{d}\mu _k}\), relation (8.3) implies that
This and Jackson’s inequality give
Moreover, Theorem 9.1 below implies
or, in other words,
Using again (8.4), we obtain
Taking into account monotonicity of \( E_j(f)_{p,\mathrm{d}\mu _k}\) and choosing l sufficiently large, we arrive at (8.2). \(\square \)
9 Inverse Theorems of Approximation Theory
Theorem 9.1
Let \(m, n \in \mathbb {N}\), \(1\le p\le \infty \), \(f \in L^{p}(\mathbb {R}^d, \mathrm{d}\mu _k)\). We have
Remark 9.2
By Remark 6.8, \(K_{2m}\left( \frac{1}{n}, f\right) _{p,\mathrm{d}\mu _k}\) in this inequality can be equivalently replaced by \(\omega _m\left( \frac{1}{n}, f\right) _{p,\mathrm{d}\mu _k}\), \({^{**}}\omega _m\left( \frac{1}{n}, f\right) _{p,\mathrm{d}\mu _k}\), and \({^{*}}\omega _l\left( \frac{1}{n}, f\right) _{p,\mathrm{d}\mu _k}\), \(l=2m-1,2m\).
Proof
Let us prove (9.1) for \(\omega _m \left( \frac{1}{n}, f\right) _{p,\mathrm{d}\mu _k}\). By Theorem 5.15, for any \(\sigma >0\), there exists \(f_\sigma \in B_{p, k}^\sigma \) such that
For any \(s\in \mathbb {Z}_+\),
Using Lemma 6.10,
Then Bernstein inequality (7.3) implies that
Thus,
Taking into account that
we have
Choosing s such that \(2^s\le n<2^{s+1}\) implies (9.1). \(\square \)
Theorem 9.1 and Jackson’s inequality imply the following Marchaud inequality.
Corollary 9.3
Let \(m \in \mathbb {N}\), \(1\le p\le \infty \), \(f \in L^{p}(\mathbb {R}^d, \mathrm{d}\mu _k)\). We have
Theorem 9.4
Let \(1\le p\le \infty \), \(f \in L^{p}(\mathbb {R}^d, \mathrm{d}\mu _k)\), and \(r\in \mathbb {N}\) be such that \(\sum _{j=1}^{\infty } j^{2r-1}E_j(f)_{p,\mathrm{d}\mu _k} <\infty .\) Then \(f \in W^{2r}_{p, k}\), and, for any \(m,n\in \mathbb {N}\), we have
Remark 9.5
We can replace \(K_{2m}\left( \frac{1}{n}, (-\Delta _k)^rf\right) _{p,\mathrm{d}\mu _k}\) by any of moduli \(\omega _m\left( \frac{1}{n}, (-\Delta _k)^rf\right) _{p,\mathrm{d}\mu _k}\), \({^{*}}\omega _l\left( \frac{1}{n}, (-\Delta _k)^rf\right) _{p,\mathrm{d}\mu _k},\) and \({^{**}}\omega _m \left( \frac{1}{n}, (-\Delta _k)^rf\right) _{p,\mathrm{d}\mu _k}\), \(l=2m-1,2m\).
Proof
Let us prove (9.3) for \(\omega _m \left( \frac{1}{n}, (-\Delta _k)^rf\right) _{p,\mathrm{d}\mu _k}\). Consider
By Bernstein’s inequality (7.3),
Therefore, series (9.4) converges to a function \(g\in L^{p}(\mathbb {R}^d, \mathrm{d}\mu _k)\). Let us show that \(g=(-\Delta _k)^rf\), i.e., \(f\in W^{2r}_{p, k}\). Set
Then
where \(\varphi \in \mathcal {S}(\mathbb {R}^d)\). Hence, \(\mathcal {F}_k(g)(y) =|y|^{2r}\mathcal {F}_k(f)(y)\) and \(g=(-\Delta _k)^rf\).
To obtain (9.3), we write
The first term is estimated as follows
Moreover, by Corollary 7.4,
Using (9.2) and choosing N such that \(2^N\le n<2^{N+1}\) completes the proof of (9.3). \(\square \)
References
Anker, J.-P.: An introduction to Dunkl theory and its analytic aspects. arXiv:1611.08213
Bateman, H., Erdélyi, A.: Higher Transcendental Functions, vol. 2. MacGraw-Hill, New York (1953)
Ben Saïd, S., Kobayashi, T., Ørsted, B.: Laguerre semigroup and Dunkl operators. Compos. Math. 148(4), 1265–1336 (2012)
Boas, R.P.: Quelques généralisations d’un théoréme de S. Bernstein sur la dérivée d’un polynôme trigonométrique. C. R. Acad. Sci. Paris 227, 618–619 (1948)
Boas, R.P.: Integrability along a line for a class of entire functions. Trans. Am. Math. Soc. 73, 191–197 (1952)
Boas, R.P.: Entire Functions. Academic Press, New York (1954)
Chertova, D.V.: Jackson theorems in \(L_2({\mathbb{R}})\)-space with power weight. Izv. Tul. Gos. Univ. Estestv. Nauki 3, 91–107 (2009). (in Russian)
DeVore, R., Lorentz, G.: Constructive Approximation. Springer, New York (1993)
Dai, F., Ditzian, Z.: Combinations of multivariate averages. J. Approx. Theory. 131, 268–283 (2004)
Dai, F., Ditzian, Z., Tikhonov, S.: Sharp Jackson inequalities. J. Approx. Theory. 151, 86–112 (2008)
Dai, F., Xu, Yu.: Approximation Theory and harmonic Analysis on Spheres and Balls. Springer, New York (2013)
Dai, F., Xu, Yu.: Analysis on \(h\)-harmonics and Dunkl transforms. CRM Barcelona. Birkhauser, Springer, Basel (2015)
Ditzian, Z., Hristov, V., Ivanov, K.: Moduli of smoothness and \(K\)-functional in \(L_p\), \(0<p<1\). Constr. Approx. 11, 67–83 (1995)
Dunkl, C.F.: Differential-difference operators associated to reflection groups. Trans. Am. Math. Soc. 311, 167–183 (1989)
Dunkl, C.F.: Integral kernels with reflection group invariance. Can. J. Math. 43, 1213–1227 (1991)
Dunkl, C.F.: Hankel transforms associated to finite reflections groups. Contemp. Math. 138, 123–138 (1992)
Gogatishvili, A., Opic, B., Tikhonov, S., Trebels, W.: Ulyanov-type inequalities between Lorentz–Zygmund spaces. J. Fourier Anal. Appl. 20(5), 1020–1049 (2014)
Gorbachev, D.V., Ivanov, V.I., Veprintsev, R.A.: Optimal argument in the Sharp Jackson’s inequality in the space \(L_2\) with the hyperbolic weight. Math. Notes 96(6), 904–913 (2014)
Gorbachev, D.V., Ivanov, V.I., Tikhonov, S.Yu.: Sharp Pitt inequality and logarithmic uncertainty principle for Dunkl transform in \(L_2\). J. Approx. Theory 202, 109–118 (2016)
Gorbachev, D.V., Ivanov, V.I., Tikhonov, S.Yu.: Pitt’s inequalities and uncertainty principle for generalized Fourier Transform. Int. Math. Res. Not. 23, 7179–7200 (2016)
Hassani, S., Mustapha, S., Sifi, M.: Riesz potentials and fractional maximal function for the Dunkl transform. J. Lie Theory 19(4), 725–734 (2009)
Ivanov, V.I.: Direct and inverse theorems in approximation theory for periodic functions in S.B. Stechkin’s papers and the development of these theorems. Proc. Steklov Inst. Math. 273(1 (Suppl)), 1–13 (2011)
Ivanov, V.I., Liu, Y., Smirnov, O.I.: Some classes of entire functions of exponential type in \(L_p({\mathbb{R}}^d)\)-spaces with power weight. Izv. Tul. Gos. Univ. Estestv. Nauki 4, 26–34 (2014). (in Russian)
de Jeu, M.F.E.: The Dunkl transform. Invent. Math. 113, 147–162 (1993)
de Jeu, M.F.E.: Paley–Wiener theorems for the Dunkl transform. Trans. Am. Math. Soc. 358, 4225–4250 (2006)
Johnen, H., Scherer, K.: On the equivalence of the K-functional and moduli of continuity and some applications. In: Constructive Theory of Functions of Several Variables (Proceedings of the Conference at the Mathematical Research Institute, Oberwolfach 1976), Lecture Notes in Mathematics, vol. 571, pp. 119–140. Springer, Berlin (1977)
Li, I.P., Su, C.M., Ivanov, V.I.: Some problems of approximation theory in the spaces \(L_p\) on the line with power weight. Math. Notes 90(3), 362–383 (2011)
Mejjaoli, H., Trimèche, K.: On a mean value property associated with the Dunkl Laplacian operator and applications. Integral Transform. Spec. Funct. 12, 279–302 (2001)
Nessel, R.J., Wilmes, G.: Nikolskii-type inequalities for trigonometric polynomials and entire functions of exponential type. J. Aust. Math. Soc. Ser. A 25(1), 7–18 (1978)
Nikol’skiǐ, S.M.: A generalization of an inequality of S. N. Bernstein. Dokl. Ak. Nauk U.S.S.R. 60(9), 1507–1510 (1948). (in Russian)
Nikol’skiǐ, S.M.: Approximation of Functions of Several Variables and Embedding Theorems. Springer, New York (1975)
Pesenson, I.: Plancherel–Polya-type inequalities for entire functions of exponential type in \(L_p({\mathbb{R}}^d)\). J. Math. Anal. Appl. 330, 1194–1206 (2007)
Plancherel, M., Pólya, G.: Fonctions entières et intégrales de Fourier multiples. Comment. Math. Helv. 9, 224–248 (1937); 10, 110–163 (1938)
Platonov, S.S.: Bessel harmonic analysis and approximation of functions on the half-line. Izv. Math. 71(5), 1001–1048 (2007)
Platonov, S.S.: Bessel generalized translations and some problems of approximation theory for functions on the half-line. Sib. Math. J. 50(15), 123–140 (2009)
Prudnikov, A.P., Brychkov, YuA, Marichev, O.I.: Integrals and Series, Elementary Functions, vol. 1. Gordon & Breach Science, New York (1986)
Rahman, Q.I., Schmeisser, G.: \(L_p\) inequalities for entire functions of exponential type. Trans. Am. Math. Soc. 320, 91–103 (1990)
Ronkin, L.I.: Introduction in the Theory of Entire Functions of Several Variables. AMS, Providence (1974)
Rösler, M.: Bessel-type signed hypergroups on \(\mathbb{R}\). In: Heyer, H., Mukherjea, A. (eds.) Probability Measures on Groups and Related Structures XI, Proceedings, Oberwolfach, vol. 1994, pp. 292–304. World Scientific, Singapore (1995)
Rösler, M.: Generalized Hermite polynomials and the heat equation for Dunkl operators. Commun. Math. Phys. 192, 519–542 (1998)
Rösler, M.: Positivity of Dunkl’s intertwinning operator. Duke Math. J. 98, 445–463 (1999)
Rösler, M.: A positive radial product formula for the Dunkl kernel. Trans. Am. Math. Soc. 355, 2413–2438 (2003)
Rösler, M.: Dunkl operators. Theory and applications. In: Orthogonal Polynomials and Special Functions. Lecture Notes in Mathematics, vol. 1817. Springer, pp. 93–135 (2003)
Soboleff, S.: On a theorem in functional analysis. Dokl. Akasd. Nauk U.S.S.R. 20, 5 (1938). (in Russian)
Stechkin, S.B.: A generalization of some inequalities of S. N. Bernstein. Dokl. Ak. Nauk U.S.S.R. 60(9), 1511–1514 (1948). (in Russian)
Stein, E.M., Weiss, G.: Fractional integrals on \(n\)-dimensional Euclidean space. J. Math. Mech. 7, 503–514 (1958)
Stein, E.M.: Singular Integrals and Differentiability Properties of Functions. Princeton University Press, Princeton (1970)
Thangavelu, S., Xu, Y.: Convolution operator and maximal function for Dunkl transform. J. Anal. Math. 97, 25–55 (2005)
Thangavelu, S., Xu, Y.: Riesz transform and Riesz potentials for Dunkl transform. J. Comput. Appl. Math. 199, 181–195 (2007)
Thorin, G.O.: Convexity theorems generalizing those of M. Riesz and Hadamard with some applications. Commun. Semin. Math. Univ. Lund 9, 1–58 (1948)
Timan, A.F.: Theory of Approximation of Functions of a Real Variable. Pergamon Press, MacMillan, New York (1963)
Trimèche, K.: The Dunkl intertwining operator on spaces of functions and distributions and integral representations of its dual. Integral Transform. Spec. Funct. 12, 349–374 (2001)
Trimèche, K.: Paley–Wiener theorems for the Dunkl transform and Dunkl translation operators. Integral Transform. Spec. Funct. 13, 17–38 (2002)
Xu, Y.: Integration of the intertwining operator for h-harmonic polynomials associated to reflection groups. Proc. Am. Math. Soc. 125, 2963–2973 (1997)
Xu, Y.: Dunkl operators: Funk–Hecke formula for orthogonal polynomials on spheres and on balls. Bull. Lond. Math. Soc. 32, 447–457 (2000)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Yuan Xu.
D. V. Gorbachev and V. I. Ivanov were supported by the Russian Science Foundation under Grant 18-11-00199. S. Yu. Tikhonov was partially supported by MTM 2017-87409-P, 2017 SGR 358, and by the CERCA Programme of the Generalitat de Catalunya.
Rights and permissions
About this article
Cite this article
Gorbachev, D.V., Ivanov, V.I. & Tikhonov, S.Y. Positive \(L^p\)-Bounded Dunkl-Type Generalized Translation Operator and Its Applications. Constr Approx 49, 555–605 (2019). https://doi.org/10.1007/s00365-018-9435-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00365-018-9435-5