Abstract
For the Dunkl transforms associated with the weight functions \(h_{\kappa }^2(x)=\prod _{j=1}^d |x_j|^{2{\kappa }_j}\), \({\kappa }_1,\ldots , {\kappa }_d\ge 0\) on \({{\mathbb {R}}}^d\), it is proved that if \(p\ge 2\) and \({\delta }>{\delta }_{\kappa }(p):=\max \{(2l_{\kappa }+1) |\frac{1}{2}-\frac{1}{p}|-\frac{1}{2},0\}\), the Bochner–Riesz means \(B_{R}^{\delta }(h_{\kappa }^2; f)\) converges to f at almost everywhere for all \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2dx)\). This extends a well known result of Carbery et al. for the classical Fourier transforms (J Lond Math Soc 38:513–524, 1988).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and main results
In this paper, we consider weight functions
Let \({\sigma }_j\) be the reflection with respect to the coordinate plane \(x_{j}=0\) (\(j=1,\ldots ,d\)). Namely,
Denote by \({{\mathbb {Z}}}_2^d\) the Abelian reflection group generated by \({\sigma }_1,\ldots , {\sigma }_d\). Clearly, \(h^2_{\kappa }(x)\) is invariant under \({{\mathbb {Z}}}_2^d\).
For \(1\leqslant p\leqslant \infty \), let \(L^{p}({\mathbb {R}}^{d}; h_{{\kappa }}^{2})\equiv L^{p}({\mathbb {R}}^{d}; h_{{\kappa }}^{2}dx)\) be the \(L^{p}\)-space defined with respect to the measure \(h_{{\kappa }}^{2}(x)dx\) on \({{\mathbb {R}}}^d\), and \(\Vert \cdot \Vert _{{\kappa },p}\) the norm of \(L^{p}( {{\mathbb {R}}}^d; h_{\kappa }^2)\). For a set \(E\subset {{\mathbb {R}}}^d\), write \({{\,\mathrm{meas}\,}}_{\kappa }(E):=\int _E h_{\kappa }^2(x)\, dx.\) And denote by \(\Vert \cdot \Vert \) and \({\langle }\cdot , \cdot {\rangle }\) the Euclidean norm and the Euclidean inner product on \({{\mathbb {R}}}^d\), respectively. Let \(B(x,r):=\{ y\in {{\mathbb {R}}}^d:\ \Vert y-x\Vert \le r\}\) be the ball with center \(x\in {{\mathbb {R}}}^d\) and radius \(r>0\).
The Dunkl transform \({\mathcal {F}}_{\kappa }f\) of \(f\in L^{1}({{\mathbb {R}}}^d; h_{\kappa }^2)\) is defined by
where \(c_{\kappa }^{-1}=\int _{{{\mathbb {R}}}^d}h_{\kappa }^2(y)e^{-\Vert y\Vert ^2/2}\, dy\), \(E_{\kappa }(-{\mathbf {i}}x, y)=V_{\kappa }\Bigl [ e^{-{\mathbf {i}}\langle x, \cdot \rangle }\Bigr ](y)\) is the weighted analogue of the character \(e^{-{\mathbf {i}}\langle x, y\rangle }\) on \({{\mathbb {R}}}^d\), and \(V_{{\kappa }}:\ C({{\mathbb {R}}}^d)\rightarrow C({{\mathbb {R}}}^d)\) denotes the Dunkl intertwining operator. In the case of \({\kappa }=0\) (i.e., the unweighted case), \(V_{\kappa }\) is simply the identity operator on \(C({{\mathbb {R}}}^d)\), and the Dunkl transform \({\mathcal {F}}_{\kappa }f\) becomes the classical Fourier transform. Furthermore, in the case when \(f(\cdot )=f_0(\Vert \cdot \Vert )\) is a radial function, \({\mathcal {F}}_{\kappa }f\) becomes a Hankel transform of \(f_0\).
It is clear that \(h_{\kappa }\) is invariant under a reflection group. And Dunkl transform is an isometry of \(L^2({{\mathbb {R}}}^d; h_{\kappa }^2dx)\), which has applications in physics for the analysis of quantum many body systems of Calogero–Moser–Sutherland type (see, for instance, [15, 16, 10, Section 11.6], [14]). From the mathematical analysis point of view, the importance of the Dunkl transform lies in that it generalizes the classical Fourier transform, and plays the similar role as the Fourier transform in classical Fourier analysis.
Many properties of Dunkl transform are very similar to those of the classical Fourier transform (see [7, 15, 16]). For instance, if the Dunkl transform \({\mathcal {F}}_{\kappa }f\) of \(f\in L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\) is also in \( L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\), then f can be recovered by \({\mathcal {F}}_{\kappa }f\) via inverse Dunkl transform. However, to check whether \({\mathcal {F}}_{\kappa }f\in L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\) is sometimes difficult. To weaken this condition, a very useful tool is the Bochner–Riesz means of f, which are defined as
where \(R>0\), \(f\in L^{1}( {{\mathbb {R}}}^d; h_{\kappa }^2)\), and \({\delta }>-1\) is the order of the Bochner–Riesz means. As in classical Fourier analysis, \(B_R^{\delta }(h_{\kappa }^2; f)(x)\) can be expressed as an integral
which further extends \(B_{R}^{\delta }(h_{\kappa }^2; f)\) to a bounded operator on \(L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\) for all \(1\le p<\infty \) and \(R>0\).
Summability of the Bochner–Riesz means \(B_R^{\delta }(h_{\kappa }^2; f)\) for the Dunkl transform was studied in the pioneering paper by Thangavelu and Xu [15, Theorem 5.5], where it was proved thatFootnote 1
for all \(1\le p\le \infty \) if and only if \({\delta }>l_{\kappa }: =\frac{d-1}{2}+|{\kappa }|\), where \(|{\kappa }|=\sum _{j=1}^d {\kappa }_j\). Moreover, it was proved in [4, Theorem 4.3] that if \(1\le p\le \infty \) and \(|\frac{1}{p}-\frac{1}{2}|\ge \frac{1}{2l_{\kappa }+2}\), then (4) holds for all \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\) if and only if
This last result extends the celebrated Thomas–Stein theorem for the Fourier transforms [17].
Thangavelu and Xu [15] also studied the maximal Bochner–Riesz operators:
They proved that:
Feng Dai and Wenrui Ye gave a stronger result in [6] under the condition \(\frac{1}{2}-\frac{1}{p}\ge \frac{1}{2l_{\kappa }+2}\). They proved that (6) holds for all \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\) if and only if \({\delta }>{\delta }_{\kappa }(p)\). This result is due to Christ [2]. And by a standard density argument, this result directly implies that if \(\frac{1}{2}-\frac{1}{p}\ge \frac{1}{2l_{\kappa }+2}\), \({\delta }>{\delta }_{\kappa }(p)\) and \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\), then
In [6], it was also claimed, but not proved, the condition \(\frac{1}{2}-\frac{1}{p}\ge \frac{1}{2l_{\kappa }+2}\) for (7) can be dropped. In this paper, we are devoted to give the proof of this fact. Our main result is stated as follows:
Theorem 1.1
If \(p\ge 2\)and \({\delta }>{\delta }_{\kappa }(p):=\max \{(2l_{\kappa }+1) |\frac{1}{2}-\frac{1}{p}|-\frac{1}{2},0\}\), then for all \(f\in L^p({{\mathbb {R}}}^d, h_{\kappa }^2)\),
We point out that in the unweighted case of Fourier transform (i.e., the case of \({\kappa }=0\)), Theorem 1.1 is well known and is due to Carbery et al. [1]. However, the proof of the weighted case is rather involved and more difficult.
This paper is organized as follows. Section 2 contains some preliminary materials on analysis associated with the Dunkl transform. Then, in Sect. 3, we prove our main result, Theorem 1.1. The proof relies on a technical result Theorem 3.3 which will be proved in Sect. 5. And in Sect. 4, we will prove some lemmas as preparations.
2 Preliminaries
In this section, we shall present some preliminary materials on Dunkl analysis associated with the group \({{\mathbb {Z}}}_2^d\) and the weight \(h_{\kappa }^2(x)\) given in (1), most of which can be found in [5, 8,9,10, 14, 15, 16].
2.1 Dunkl intertwining operators
Let \(E_j\) be the difference operator given by \( E_j f(x):=\frac{f(x) -f(x{\sigma }_j)}{ x_j}\). The Dunkl operators \({{\mathcal {D}}}_{{\kappa },j}\) are defined by
These operators are mutually commute, and map \({{\mathbb {P}}}_n^d\) to \({{\mathbb {P}}}_{n-1}^d\), where \({{\mathbb {P}}}_n^d\) denotes the space of homogeneous polynomials of degree n in d variables. A fundamental result in the Dunkl theory states that there exists a linear operator \(V_{\kappa }\) on the space of all algebraic polynomials on \({{\mathbb {R}}}^d\), called the Dunkl intertwining operator, determined uniquely by the following conditions:
It was proved by Yuan Xu [18] that
where \(d\mu _j(s)=c_{{\kappa }_j} (1+s)(1-s^{2})^{\kappa _{j}-1}ds\) with \(c_{{\kappa }_j}:=\frac{ \varGamma ({\kappa }_j+1/2)}{\sqrt{\pi }\varGamma ({\kappa }_j)}\) if \({\kappa }_j>0\), and \(d\mu _j\) is the dirac measure supported at \(s=1\) if \({\kappa }_j=0\). In particular, the formula (11) extends \(V_{\kappa }\) to a positive operator on the space of continuous functions on \({{\mathbb {R}}}^d\).
For the Dunkl intertwining operator associated with a general finite reflection group, we refer to the work of Rösler [12].
2.2 Dunkl transforms
The Dunkl transform \({\mathcal {F}}_{\kappa }f\) of \(f\in L^{1}({\mathbb {R}}^{d}; h_{\kappa }^{2})\) is defined by
where \(c_{\kappa }^{-1}=\int _{{{\mathbb {R}}}^d}h_{\kappa }^2(y)e^{-\Vert y\Vert ^2/2}\, dy\),
and \(J_{\alpha }\) denotes the Bessel function of order \({\alpha }\) of the first kind:
If \({\kappa }=0\), then the Dunkl transform coincides with the usual Fourier transform. And in this paper, we also use \({\widehat{f}}\) to denote the Dunkl transform of f.
The Dunkl transform \({\mathcal {F}}_{\kappa }\) restricted on the class \({\mathcal {S}}({{\mathbb {R}}}^d)\) of Schwartz functions on \({{\mathbb {R}}}^d\) satisfies the Plancherel formula \( \Vert f\Vert _{{\kappa },2}=\Vert {\mathcal {F}}_{\kappa }f\Vert _{{\kappa },2}\). Thus, it extends to an isometric isomorphism on the space \(L^2({{\mathbb {R}}}^d; h_{\kappa }^2)\). As a result, the Dunkl transform can also be defined on the space \( L^1+L^2\supset \bigcup _{1\le p\le 2} L^p\).
The Dunkl transforms have properties similar to those of the classical Fourier transform. We collect some of these properties in the following lemma.
Lemma 2.1
-
(i)
If \(f\in L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\)then \({\mathcal {F}}_{\kappa }f\in C({{\mathbb {R}}}^d)\)and \(\lim _{\Vert \xi \Vert \rightarrow \infty } {\mathcal {F}}_{\kappa }f(\xi ) =0\).
-
(ii)
The Dunkl transform \({\mathcal {F}}_{\kappa }\)is an isomorphism of the Schwartz class \({\mathcal {S}}({{\mathbb {R}}}^d)\)onto itself, and \({\mathcal {F}}_{\kappa }^2 f(x) = f (-x)\)for \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\).
-
(iii)
If f and \({\mathcal {F}}_{\kappa }f\)are both in \(L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\)then the following inverse formula holds:
$$\begin{aligned} f(x)=c_{\kappa }\int _{{{\mathbb {R}}}^d} {\mathcal {F}}_{\kappa }f(y) E_{\kappa }({\mathbf {i}}x, y) h_{\kappa }^2(y)\, dy ,\quad x\in {{\mathbb {R}}}^d. \end{aligned}$$ -
(iv)
If \(f, g\in L^1({{\mathbb {R}}}^d; h_{\kappa }^2)\)then
$$\begin{aligned} \int _{{{\mathbb {R}}}^d} {\mathcal {F}}_{\kappa }f(x) g(x)\, h_{\kappa }^2(x)\, dx =\int _{{{\mathbb {R}}}^d} f(x) {\mathcal {F}}_{\kappa }g(x)\, h_{\kappa }^2(x)\, dx . \end{aligned}$$(14) -
(v)
If \(1\le p\le 2\)and \(\frac{1}{p}+\frac{1}{p'}=1\), then
$$\begin{aligned} \Vert {\mathcal {F}}_{\kappa }f\Vert _{{\kappa },p'}\le \Vert f\Vert _{{\kappa },p}. \end{aligned}$$(15) -
(vi)
Given \(\varepsilon >0\), let \(f_\varepsilon (x) =\varepsilon ^{-(2l_{\kappa }+1)}f(\varepsilon ^{-1} x)\)with \(l_{\kappa }:=\frac{d-1}{2} +|{\kappa }|\). Then \({\mathcal {F}}_{\kappa }f_\varepsilon (\xi ) ={\mathcal {F}}_{\kappa }f (\varepsilon \xi )\).
-
(vii)
If \(f(x)=f_0(\Vert x\Vert )\)is a radial function in \(L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\)with \(1\le p\le 2\), then \({\mathcal {F}}_{\kappa }f (\xi ) = H_{l_{\kappa }} f_0 (\Vert \xi \Vert )\)is again a radial function, where \(H_{{\alpha }}\)denotes the Hankel transform defined by
$$\begin{aligned} H_{\alpha }g(s) = \frac{1}{\varGamma ({\alpha }+1)} \int _0^\infty g(r) \frac{J_{\alpha }(rs)}{(rs)^{\alpha }} r^{2{\alpha }+1}\, dr,\ \ {\alpha }>-\frac{1}{2}. \end{aligned}$$
We will also consider the Dunkl transform on the space \({\mathcal {M}}\) of finite Borel measures on \({{\mathbb {R}}}^d\):
Many identities in this paper are interpreted in a distributional sense. Denote by \({\mathcal {S}}'({{\mathbb {R}}}^d)\) the space of all tempered distributions on \({{\mathbb {R}}}^d\). We shall identify a function f in \( L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\), \(1\le p\le \infty \) with a tempered distribution in \({\mathcal {S}}'({{\mathbb {R}}}^d)\) given by
According to (14), we may define the distributional Dunkl transform \({\mathcal {F}}_{\kappa }f\) of \(f\in \bigcup _{p>2} L^p({{\mathbb {R}}}^d, h_{\kappa }^2)\) via
2.3 Generalized translations
Definition 2.2
[13, 14, 15] Given \(y\in {{\mathbb {R}}}^d\), the generalized translation \(T^y f\) of \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\) is defined by
It was proved in [6, Proposition 2.5] and [15] that the generalized translations are bounded operators on \(L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\) for all \(1\le p\le \infty \). Throughout the paper, we will use (16) as the definition of the generalized translation \(T^y f\) of \(f\in \bigcup _{p\ge 1} L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\).
According to the inverse formula, if \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\) and \(x, y\in {{\mathbb {R}}}^d\), then
Moreover, it was shown in [13, Lemma 2.2] that \(T^y\) is a linear operator from the space \({\mathcal {S}}({{\mathbb {R}}}^d)\) to itself, and for any \(f, g\in {\mathcal {S}}({{\mathbb {R}}}^d)\),
And the property (17) of the generalized translation operators carries over to \(L^p\) spaces as well:
Lemma 2.3
([6, Propsition 2.7]) If \(y\in {{\mathbb {R}}}^d\)and \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\)with \( 1\le p\le \infty \), then (17) holds almost everywhere (a.e.) on \({{\mathbb {R}}}^d\)for \(1\le p\le 2\), and holds in a distributional sense for \(1\le p\le \infty \):
Moreover, the generalized translation has the following integral representation on the space \({\mathcal {S}}({{\mathbb {R}}}^d)\).
Lemma 2.4
([15, Theorem 7.1]) For \(y=(y_1,\ldots , y_d), x=(x_1,\ldots , x_d)\in {{\mathbb {R}}}^d\)and \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\),
Here the operators \(T_{j,y_j}\)are defined by
where \(v_j(x_j, y_j, t):= \sqrt{x_j^2+y_j^2 -2x_jy_j t} \), \(u_j(x, y, t): =\Bigl (x_1, \ldots , x_{j-1}, v_j(x_j, y_j, t), x_{j+1}, \ldots x_d\Bigr )\), \( f_{j,e} (x): = \frac{1}{2} \Bigl ( f(x) + f(x{\sigma }_j)\Bigr )\), \( f_{j,o} (x) = \frac{1}{2} \Bigl ( f(x) - f(x{\sigma }_j)\Bigr ),\) and the measure \(d\mu _j\)is the same as in (11).
According to (21), we have
where \(d\mu _{x,y}\) is a signed Borel measure supported on
2.4 Generalized convolutions
We start with the definition on the space \({\mathcal {S}}({{\mathbb {R}}}^d)\).
Definition 2.5
The generalized convolution of \(f, g\in {\mathcal {S}}({{\mathbb {R}}}^d)\) is defined by
The generalized convolution has the following basic property: for \(f, g\in {\mathcal {S}}({{\mathbb {R}}}^d)\),
Since the generalized translation operators are uniformly bounded on \(L^p\)-spaces with \(1\le p\le \infty \), the following Young’s inequality can be established (see [15, Proposition 7.2]):
where \( 1\le p, q, r\le \infty \) and \(1+\frac{1}{r}=\frac{1}{p}+\frac{1}{q}\). This implies that \(f*_{\kappa }g\) can be defined for \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\) and \(g\in L^q({{\mathbb {R}}}^d; h_{\kappa }^2)\) with \(1\le p, q\le \infty \) and \(\frac{1}{p}+\frac{1}{q}\ge 1\). And is has the following property.
Lemma 2.6
([6, Corollary 2.9]) If \(f\in L^p({{\mathbb {R}}}^d; h_{\kappa }^2)\), \(1\le p\le \infty \)and \(g\in {\mathcal {S}}({{\mathbb {R}}}^d)\), then
holds in a distributional sense.
2.5 Spaces of homogeneous type
A straightforward calculation shows that
This, in particular, implies that the measure \(d\mu _{\kappa }(x) =h_{\kappa }^2(x)\, dx\) satisfies the doubling condition on the space \({{\mathbb {R}}}^d\), and hence \(({{\mathbb {R}}}^d, d\mu _{\kappa })\) is a space of homogeneous type in the sense of Coifman, and Weiss [3]. In this subsection, we will review briefly some basic definitions and facts on homogeneous spaces, most of which can be found in [3, 11].
Definition 2.7
A space of homogeneous type \((X, \rho , \mu )\) is a topological space X endowed with a quasi-metric \(\rho \) and a nonnegative Borel measure \(\mu \) such that
-
(i)
\(\rho \) is continuous on \(X\times X\) and every ball \(B_\rho (x,r):=\{y\in X:\ \ \rho (x,y)<r\}\) is open in X;
-
(ii)
the measure \(\mu \) satisfies the doubling condition:
$$\begin{aligned} \mu (B_{\rho }(x, 2r))\le C \mu (B_{\rho }(x,r)),\ \ \forall x\in X,\ \ \forall r>0; \end{aligned}$$ -
(iii)
\(0<\mu (B_\rho (x,r))<\infty \) for every \(x\in X\) and \(r>0\).
Definition 2.8
A weight w (any nonnegative measurable function) on a homogeneous space \((X,\rho ,\mu )\) satisfies the \(A_p\) condition for \(1\le p<\infty \) if
and in the case of \(p=1\), we replace \(\Bigl (\frac{1}{\mu (B)}\int _B w^{-\frac{1}{p-1}}\, d\mu \Bigr )^{p-1}\) with \(\Vert w^{-1}\Vert _\infty \).
The \(A_p\) classes are increasing with respect to p; that is, \(A_{p_1}\subset A_{p_2}\) for \(1\le p_1<p_2\). As a result, we can define \(A_\infty \) class by \(A_\infty =\bigcup _{p>1} A_p\).
Next, we define a modified Hardy–Littlewood maximal function on the homogeneous space \(({{\mathbb {R}}}^d, d\mu _{\kappa })\) by
where the supremum is taken over all the balls B in \({{\mathbb {R}}}^d\) such that \(x\in \widetilde{B}=\{x\sigma :\ \ x\in B,\ \ \sigma \in {{\mathbb {Z}}}_2^d\}\). Since \(\mu _{\kappa }\) is a \({{\mathbb {Z}}}_2^d\)-invariant doubling measure on \({{\mathbb {R}}}^d\), it follows that
and
2.6 The Hardy–Littlewood–Sobolev inequality
Recall that \({\mathcal {D}}_{{\kappa },j}\), \(1\le j\le d\) denote the Dunkl operators as defined in (9). The Dunkl h-Laplacian \(\varDelta _{\kappa }\) is defined by \(\varDelta _{{\kappa }}=\sum _{j=1}^d {\mathcal {D}}_{{\kappa }, j}^2,\) and it plays the role similar to that of ordinary Laplacian. It is known that
where \(\varDelta =\sum _{j=1}^d \partial _j^2\) and
Furthermore, \((-\varDelta _{\kappa }f)^{\wedge }(x)=\Vert x\Vert ^2 \widehat{f}(x)\) for \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\). As a result, we may define fractional power of the h-Laplacian \(\varDelta _{\kappa }\) in a distributional sense by \({\mathcal {F}}_{\kappa }(-\varDelta _{\kappa })^\beta f(x)=\Vert x\Vert ^{2\beta } \widehat{f}(x)\) for \(\beta \in {{\mathbb {R}}}\).
In the case when \(0<\beta <2l_{\kappa }+1\), the fractional h-Laplacian \((-\varDelta _{\kappa })^{\beta /2} f\) becomes the weighted Riesz potential, which we denote by \( I_{\kappa }^\beta \). It is known ([16, Section 4]) that for \(0<\beta <2l_{\kappa }+1\) and \(f\in {\mathcal {S}}({{\mathbb {R}}}^d)\),
where \(K_\beta (y)=\Vert y\Vert ^{\beta -2l_{\kappa }-1}\) and
Lemma 2.9
[16, Theorem 4.3] Let \(1<p< q<\infty \)and \(0<\beta <2l_{\kappa }+1\)be such that \(\beta =(2l_{\kappa }+1) (\frac{1}{p}-\frac{1}{q})\). Then
In the rest three sections of this paper, we use both \(\widehat{f}\) and \({\mathcal {F}}_{\kappa }f\) to denote the Dunkl transforms of f for convenience.
3 Proof of Theorem 1.1
In this section, we will prove Theorem 1.1. Let \(\xi \) be a \(C^\infty \)-function on \({{\mathbb {R}}}\) satisfying that \(\xi (x)=1\) for \(|x|\le 1\) and \(\xi (x)=0\) for \(|x|\ge 2\), and set \(\theta (x)=\xi ( x)-\xi (2x)\). Clearly, \(\text {supp}\, {\theta }\subset \{x:\ \ \frac{1}{2}\le |x|\le 2\}\), and \(\sum _{j=0}^\infty {\theta }(2^j x)=\xi (x)=1\) for all \(x\in [-1,1]{\setminus }\{0\}\). Thus, we can write, for \(\beta >-1/2\),
where \(m_j^\beta (\xi )=2^{j\beta }\theta \bigl (2^j(1-\Vert \xi \Vert ^2)\bigr ) (1-\Vert \xi \Vert ^2)_+^\beta \). As a result, the operator \(B_R^\beta (h_{\kappa }^2; f)\) can be decomposed as
where the operators \(T_{R,j}^\beta \) are given by \(\widehat{T_{R,j}^\beta f}(\xi )=m_j^\beta (R^{-1}\xi ) \widehat{f}(\xi ) \).
Since \(m_0^\beta , m_1^\beta , m_2^\beta \in {\mathcal {S}}({{\mathbb {R}}}^d)\), by [6, Lemma 4.6],
Definition 3.1
Let \(\varPhi \) be a \(C^\infty \)-function supported on \([-\frac{1}{2}, \frac{1}{2}]\). For a parameter \({\alpha }\in (0,1/2]\), set
and let \(\psi (x)={\widehat{\phi }}(x)\) for \(x\in {\mathbb {R}}^{d}\).
An important tool in the proof of Theorem 1.1 is the maximal operator defined below.
Definition 3.2
For \(t>0\), define the operator \(S_t^{\alpha }\) by \(S_t^{\alpha }f (x) =f*_{\kappa }\widehat{\phi }_{t^{-1}}(x)\), where \(\widehat{\phi }_{t^{-1}}(x)=t^{2{\lambda }_{\kappa }+1} \widehat{\phi }(tx)\). The maximal operator \(S_*^{\alpha }\) is then defined by \(S_*^{\alpha }f(x)=\sup _{t>0}|S_t^{\alpha }f(x)|\).
The proof of Theorem 1.1 can be reduced to the following estimate on the maximal operator \(S_*^{\alpha }\).
Theorem 3.3
For \(p\ge 2\), \({\delta }>{\delta }_{\kappa }(p):=\max \{(2l_{\kappa }+1) |\frac{1}{2}-\frac{1}{p}|-\frac{1}{2},0\}\)and \(0\le \beta <2l_{\kappa }+1\),
where for \(0<t\le 1\),
The proof of Theorem 3.3 will be given in the next two sections. For the rest of this section, we take this theorem for granted and proceed with the proof of Theorem 1.1.
Proof of Theorem 1.1
Let \({\delta }>{\delta }_{\kappa }(p)\) with \(p\ge 2\). Without loss of generality, we may assume that \(0<{\delta }\le {\lambda }_{\kappa }\) since otherwise the result follows directly from (6).
We decompose a function \(f\in L^p({{\mathbb {R}}}^d; d\mu _{\kappa })\) as \(f=f_1+f_2\), where \(f_{1}(x)=f(x)\chi _{\Vert x\Vert \leqslant 1}\), \(f_{2}(x)=f(x)\chi _{\Vert x\Vert >1}\) and we recall that \(d\mu _{\kappa }(x)=h_{\kappa }^2(x)\, dx\). Since \(p\ge 2\), it follows by Hölder’s inequality that \(\Vert f_{1}\Vert _{L^{2}({\mathbb {R}}^{d}, d\mu _{\kappa })} \leqslant C \Vert f\Vert _{L^{p}({\mathbb {R}}^{d}, d\mu _{\kappa })}<\infty \). Hence, (6) and a standard density argument imply that (8) holds with \(f=f_1\).
It remains to show (8) holds with f being replaced by \(f_2\). Since \( (2\lambda _{{\kappa }}+1)(1-\frac{2}{p})<1+2\delta , \) we may choose a constant \(\beta \) such that \((2\lambda _{{\kappa }}+1)(1-\frac{2}{p})<\beta <1+2\delta \). By Hölder’s inequality, we then have
where the third step uses the fact that \( \frac{\beta p}{p-2}-2l_{\kappa }>1\). This means that \(f_{2}\in L^{2}({\mathbb {R}}^{d};\Vert x\Vert ^{-\beta }d\mu _{\kappa })\). Thus, we reduce the proof of Theorem 1.1 to the estimate:
To show (34), we use the decomposition (31) and the estimate (32) to obtain
where \(\widehat{T_{R,j}^{\delta }f}(\xi )=m_j^{\delta } (R^{-1}\xi ) \widehat{f}(\xi ) \), \(m_j^{\delta } (\xi )=\varPhi (2^{j-2}(1- \Vert \xi \Vert ^2))\), \(\varPhi \) is a \(C^\infty \) function supported in the interval \([-\frac{1}{2}, \frac{1}{2}]\). Thus, with the notations of Definition 3.2, we have
Since \(\Vert x\Vert ^{-\beta }\) for \(0\le \beta <2{\lambda }_{\kappa }+1\) is an \(A_1\)-weight on the homogeneous space \(({{\mathbb {R}}}^d, d\mu _{\kappa })\), it follows by Theorem 3.3 that
where the last step uses the fact that \(\beta <2{\delta }+1\). This shows (34) and hence completes the proof of Theorem 1.1. \(\square \)
4 Some lemmas
Lemma 4.1
Let \(0<\beta <2l_{\kappa }+1\)and \(0<\varepsilon <100d\). Set \(u_\beta (x) =\Vert x\Vert ^{\beta -2l_{\kappa }-1}\)and \(E=\{ x\in {{\mathbb {R}}}^d:\ \ ||x|-1|\le \varepsilon \). Then for \(x, y\in E\),
where \({\bar{x}}=(|x_{1}|,\ldots ,|x_{d}|)\)for \(x=(x_{1},\ldots ,x_{d})\in {\mathbb {R}}^{d}\).
Proof
Let \(\psi \in C_c^\infty ({{\mathbb {R}}}^d) \) be a radial function supported in \([\frac{1}{2}, 4]\) such that \(\sum _{k\in {{\mathbb {Z}}}} \psi (2^{k} x)=1, \ \ \forall x\ne 0.\) Then setting \(\varphi (x)=\Vert x\Vert ^{\beta -2l_{\kappa }-1} \psi (x)\), we obtain that for \(x\ne 0\),
where \(\varphi _t(x)=t^{-2l_{\kappa }-1} \varphi (x/t)\). It follows by [6, Lemma 4.6] that for \(\ell >2l_{\kappa }+3\),
For simplicity, we set \(\rho =\Vert {\bar{x}}-{\bar{y}}\Vert \). If \(0<\beta <d\), then
If \(\beta =d\), then
Finally, if \(d<\beta <2l_{\kappa }+1\), then
\(\square \)
Lemma 4.2
For \(0<\varepsilon <100d\)and \(0\le \beta <2l_{\kappa }+1\), we have
where
Furthermore, for any \(M>2l_{\kappa }+1\)and \(\beta > 0\),
Proof
We start with the proof of (37). Without loss of generality, we may assume \(\beta >0\). Set \(E:=\{ x\in {{\mathbb {R}}}^d:\ \ \bigl |1-\Vert x\Vert \bigr |\le \varepsilon \}\), and denote by \(L^2(E; h_{\kappa }^2)\) the subspace of \(L^2(h_{\kappa }^2)\) consisting of all functions supported in the set E. We first claim that (37), is a consequence of the following estimate: for any function \(g\in L^2(E; h_{\kappa }^2)\),
To see this, consider the operator \(T: L^2(\Vert x\Vert ^{\beta }h_{\kappa }^2(x))\rightarrow L^2(E; h_{\kappa }^2)\) given by \(Tf =\widehat{f}\Bigl |_{E}\), and note that for any \(g\in L^2(E; h_{\kappa }^2)\),
This means that \(T^*{\bar{g}} (x)=\Vert x\Vert ^{-\beta }\overline{\widehat{g}(x)}\) where \(T^*: L^2(E; h_{\kappa }^2)\rightarrow L^2(\Vert x\Vert ^\beta h_{\kappa }^2(x))\) denotes the dual operator of T. The claim then follows.
By the standard density argument, it suffices to show (39) for \(g\in L^2(E;h_{\kappa }^2)\cap C_c^\infty ({{\mathbb {R}}}^d)\). Indeed, invoking [16, Lemma 4.1], we obtain that for \(g\in L^2(E;h_{\kappa }^2)\cap C_c^\infty ({{\mathbb {R}}}^d)\),
where
Clearly,
where \(B_\beta \) is the Lebesgue constant of the operator L; that is,
According to Lemma 4.2, if \(0<\beta <d\), then
if \(\beta =d\), then
and finally, if \(d<\beta <2l_{\kappa }+1\), then \(B_\beta \le C \sup _{x\in E}\int _E \, dy\le C \varepsilon .\)
Now we turn to the proof of (38). Note that
Thus, it suffices to show that
(40) is an easy consequence of Lemma 2.9. Indeed, let \(q\ge 2\) be such that \( (2l_{\kappa }+1) (\frac{1}{2}-\frac{1}{q})=\frac{\beta }{2}\). Then using Hölder’s inequality and Lemma 2.9, we obtain
\(\square \)
5 Proof of Theorem 3.3
Definition 5.1
Let \(\ell \in {{\mathbb {N}}}\) and \(0<\beta <\ell \). Define
where
By (22),
where \(d\mu _{x,y,\ell }\) is a signed Borel measure supported in \(\{z\in {{\mathbb {R}}}^d:\ \ \Vert {\bar{z}}-{\bar{x}}\Vert \le \ell |y|\}.\)
Lemma 5.2
For \(\ell \in {{\mathbb {N}}}\)and \(0<\beta <\ell \),
Proof
By definition, we have
where
It is easily seen that the function \(A_\ell (x,y)\) has the following properties:
-
(i)
For \(x, y\in {{\mathbb {R}}}^d\), \(|A_\ell (x,y)|\le 2^\ell .\)
-
(ii)
As \(\Vert y\Vert \rightarrow 0\),
$$\begin{aligned} A_\ell (x,y)= q_\ell (x,y)+O(\Vert y\Vert ^{\ell +1})\end{aligned}$$(42)holds uniformly in \(x\in {\mathbb {S}}^{d-1}\), where
$$\begin{aligned} q_\ell (x,y)&= i^\ell V_{\kappa }\Bigl [ ({\langle }y, \cdot {\rangle })^\ell \Bigr ](x)= c_{\kappa }i^{\ell } \int _{[-1,1]^d} \Bigl ( \sum _{j=1}^d t_j x_jy_j\Bigr )^\ell \prod _{j=1}^d (1-t_j^2)^{{\kappa }_j-1}(1+t_j)\, dt_j.\end{aligned}$$ -
(iii)
For any \(r>0\) and any \(x, y\in {{\mathbb {R}}}^d\), \(A_\ell (rx, y)=A_\ell (x, ry)\).
On the other hand, using Plancherel’s formula, we obtain
where
Thus, it suffices to show that
To this end, let \(\varepsilon =\varepsilon _{d,{\kappa }}\in (0,1)\) be a small constant depending only on d and \({\kappa }\), and set
Clearly,
Thus, for the proof of (43), it is sufficient to show that
Indeed, using (42), we obtain that for \(x\in {\mathbb {S}}^{d-1}\),
where
Thus, (44) is a consequence of the following estimate:
Since \(M_\ell \) is a homogeneous polynomial of degree \(2\ell \) on \({\mathbb {S}}^{d-1}\) and \({\mathbb {S}}^{d-1}\) is compact, it is enough to show that
Assume (45) were not true, that is, there exists \(x=(x_1,\ldots , x_d)\in {\mathbb {S}}^{d-1}\) such that \(M_\ell (x)=0\). This would imply that \(q_\ell (x, y)=0\) for all \(y\in {\mathbb {S}}^{d-1}\). However, this is impossible because for \(y=e_{j_0}\) with \(j_0\in \{1,\ldots , d\}\) satisfying \(|x_{j_0}|=\max _{1\le j\le d}|x_j|\), we have
\(\square \)
Recall that \(\phi _t(x)=\phi (x/t)\) for \(t>0\), and
where \(\varPhi \in C_c^\infty ({{\mathbb {R}}})\) be such that \(\text {supp}\ \varPhi \subset [-\frac{1}{2}, \frac{1}{2}]\).
By Definition 3.2,
Lemma 5.3
For \(f\in L^2(\Vert x\Vert ^{-\beta }h_{\kappa }^2(x))\)and \(0<\beta <d\),
Proof
It is easy to see
where \(K_j(x)=\widehat{\phi }(x)\eta _j(x)\) and \(2^{-i}\sim {\delta }\). By [6, Lemma 6.4], for any \(N\ge 1\),
In particular, this implies that
For \(j\ge i\), we set \(T_j f=f*_{\kappa }K_j\), and claim that
Here and throughout the proof, N is a large positive integer whose exact value is not important. Indeed,
For the first integral \(I_1\), using (47) and Lemma 4.2, we have
For \(I_2\), using (47) and Lemma 4.2 again, we obtain
For \(I_3\), we have
Now combining the above estimates of \(I_1,I_2, I_3\), we deduce the estimate (48).
Second, we show that (48) implies that for \(j\ge i\),
Indeed, the dual operator \(T_j^*: L^2(h_{\kappa }^2)\rightarrow L^2(\Vert x\Vert ^{\beta }h_{\kappa }^2(x))\) of the operator \(T_j: L^2(\Vert x\Vert ^\beta h_{\kappa }^2(x)) \rightarrow L^2(h_{\kappa }^2)\) can be obtained as follows:
where \(T_j^*g(x)=\Vert x\Vert ^{-\beta } T_j g(x)\). The asserted estimate (49) then follows by (48) and duality.
Third, we show that for \(j\ge i\),
To this end, we recall the following local property of the operator \(T_j\): If f is supported in a cube Q of side length \(c 2^{j}\), then \(T_j f\) is supported in the set
We decompose \({{\mathbb {R}}}^d\) as an almost pairwise disjoint union of cubes \(Q_k\), \(k\in {{\mathbb {Z}}}\), where \(Q_0\) is a cube centered at the origin of side length \(2^{\ell }2^j\), and the \(Q_k\), \(k\ne 0\) are cubes of side length \(2^j\). We choose \(\ell \in {{\mathbb {N}}}\) large enough so that \(Q^*_k\cap \frac{1}{2} Q_0=\emptyset \) for all \(k\ne 0\). With this decomposition, we have
Set \(f_k=f\chi _{Q_k}\) for \(k\in {{\mathbb {Z}}}\). Then \(f =\sum _{k\in {{\mathbb {Z}}}} f_k\). For \(k=0\), we use (49) to obtain
For \(k\ne 0\), we use Plancerel’s theorem to obtain
Since \(\Vert x\Vert \sim \Vert x'\Vert \) for all \(x, x'\in Q^*_k\), it follows that for \(k\ne 0\),
On the other hand, since \(\text {supp}\ T_j f_k \subset Q_k^*\) for all \(k\in {{\mathbb {Z}}}\) and \(\sum _{k\in {{\mathbb {Z}}}} \chi _{Q_k^*}(x)\le C_d\), we have
It follows that
This proves (50).
Finally, using (46) and Minkowskii’s inequality, we obtain
\(\square \)
Remark 5.1
Lemma 5.3 and a duality argument imply that for \(0\le \beta <2l_{\kappa }+1\) and \(t>0\),
Note that by Lemma 5.2,
where
and r is the smallest integer bigger than \(\beta /2\). Note that if f is supported in a set \(\widetilde{B}(x_0, t)=\bigcup _{{\sigma }\in {{\mathbb {Z}}}_2^d} B(\sigma x_0, t)\), then \(\triangle ^r_y f(x)\) is supported in the domain \(\widetilde{B}(x_0, t+r\Vert y\Vert )\).
Lemma 5.4
For \(0\le \beta <2l_{\kappa }+1\)and \(k\in {{\mathbb {Z}}}\),
Proof
It suffices to prove the asserted conclusion for \(k=0\). For simplicity, we denote by \(L^2_{x,t}\) the Hilbert \(L^2\)-space defined with respect to the measure \(t^{-1}\Vert x\Vert ^{-\beta }h_{\kappa }^2(x) dtdx\) on \({{\mathbb {R}}}^d\times [1,2]\). Consider the operator \(T: L^2(\Vert x\Vert ^{-\beta }h_{\kappa }^2(x))\rightarrow L_{x,t}^2\) given by \(f\mapsto S_t^{\delta }f(x)\). Its dual \(T^*: L_{x,t}^2\rightarrow L^2(\Vert x\Vert ^{-\beta }h_{\kappa }^2(x))\) can be obtained as follows: for \(\{g_t\}_{1\le t\le 2}\in L_{x,t}^2\),
Thus,
and it is sufficient to show that
Setting \(f_t(x)=\Vert x\Vert ^{-\beta }g_t(x)\), we then conclude that (53) is a consequence of the following estimate:
Note that by (52),
Since for \(t\in [1,2]\), \(\phi _t\) is supported in \(\{z\in {{\mathbb {R}}}^d:\ \ (1-\frac{{\delta }}{2}) t\le \Vert z\Vert \le (1+\frac{{\delta }}{2})t\}\), \(\triangle _y^r ( \phi _t \widehat{f_t})(x)= 0\) unless
which holds either
In the first case, setting \(a_{x,y}=\frac{\Vert x\Vert -r\Vert y\Vert }{1+\frac{{\delta }}{2}}\), we have that
Similarly, in the second case, \(0\le t-a'_{x,y}\le c{\delta }\), where \(a'_{x,y}=\frac{\Vert x\Vert +r\Vert y\Vert }{1+\frac{{\delta }}{2}}\). Thus, for fixed \(x, y\in {{\mathbb {R}}}^d\), the function \(A_{x,y}(t):=\triangle _y^r ( \phi _t \widehat{f_t})(x)\) is supported in the set
Using Hölder’s inequality, we then deduce that
This implies that
which, by (51), is estimated above by
This shows the asserted estimate (54). \(\square \)
Lemma 5.5
For \(0\le \beta <2l_{\kappa }+1\),
Proof
Recall that \((L_j f)^{\wedge } (\xi )=\widehat{f}(\xi ) \varphi (2^j\xi )\), where \(\varphi \in C_c^\infty ({{\mathbb {R}}}^d)\) is supported in [1, 3] and satisfies that \(\sum _{j\in {{\mathbb {Z}}}} \varphi (2^j\xi )=1\) for all \(\xi \in {{\mathbb {R}}}^d{\setminus }\{0\}\). Thus,
where \(((L_j f)*_{\kappa }\widehat{\phi _t})^{\wedge } (\xi ) =\varphi (2^j\xi ) \widehat{f}(\xi ) \phi (t^{-1}\xi ).\) Note that \(\varphi (2^j\xi ) \phi (t^{-1}\xi )\) is identically zero unless
which also implies that
This means that
and hence
It follows by Lemma 5.4 that
Since for \(0\le \beta <2l_{\kappa }+1\), \(w(x)=\Vert x\Vert ^{-\beta }\in A_1\), it follows by the weighted Paley-Littlewood inequality (see [6, Theorem 5.2]) that
This completes the proof. \(\square \)
Let \(\widehat{\phi }(u)={\delta }\phi '(u) u \) for \(u\ge 0\). Define \( \widetilde{S_t^{\delta }} f(x)\) by
where \(\widetilde{\phi }_t (\Vert \xi \Vert )=\widetilde{\phi }(t^{-1}\Vert \xi \Vert )\). It is easily seen that
By Lemma 5.5 and (55), we can easily get the following lemma.
Lemma 5.6
For \(0\le \beta <2l_{\kappa }+1\),
We are now in a position to prove Theorem 3.3.
Proof of Theorem 3.3
For \(t>0\),
where
It follows that
\(\square \)
Notes
The result of [15] holds for more general weights invariant under a finite reflection group as well.
References
Carbery, A., De Francia, J.L.R., Vega, L.: Almost everywhere summability of Fourier integrals. J. Lond. Math. Soc. 38(2), 513–524 (1988)
Christ, M.: On almost everywhere convergence of Bochner–Riesz means in Higher dimensions. Proc. Am. Math. Soc. 95(1), 16–20 (1985)
Coifman, R., Weiss, G.: Analyse harmonique non-commutative sur certains espaces homogénes. (French) Étude de certaines intgrales singulières. Lecture Notes in Mathematics, vol. 242. Springer, Berlin (1971)
Dai, F., Wang, H.P.: A transference theorem for the Dunkl transform and its applications. J. Funct. Anal. 258(12), 4052–4074 (2010)
Dai, F., Xu, Y.: Analysis on \(h\)-harmonics and Dunkl transforms. In: Tikhonov, S. (ed.) Advanced Courses in Mathematics CRM Barcelona. Birkh user/Springer, Basel (2015)
Dai, F., Ye, W.: Local restriction theorem and maximal Bochner–Riesz operators for the Dunkl transforms. Trans. Am. Math. Soc. 371, 641–679 (2019)
de Jeu, M.F.E.: The Dunkl transform. Invent. Math. 113, 147–162 (1993)
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(6), 1213–1227 (1991)
Dunkl, C.F., Xu, Y.: Orthogonal Polynomials of Several Variables, 2nd edn. Cambridge University Press, Cambridge (2014)
Hytönen, T., Pèrez, C., Rela, E.: Sharp reverse Hölder property for \(A_\infty \) weights on spaces of homogeneous type. J. Funct. Anal. 263(12), 3883–3899 (2012)
Rösler, M.: Positivity of Dunkl intertwining 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. Orthogonal Polynomials and Special Functions (Leuven, 2002). Lecture Notes in Math, vol. 1817, pp. 93–135. Springer, Berlin (2003)
Thangavelu, S., Xu, Y.: Convolution operator and maximal function for the Dunkl transform. J. Anal. Math. 97, 25–55 (2005)
Thangavelu, S., Yuan, X., Riesz, : Transform and Riesz potentials for Dunkl transform. J. Comput. Appl. Math. 199(1), 181–195 (2007)
Tomas, P.: Restriction theorems for the Fourier transform. In: Harmonic Analysis in Euclidean space, Proceedings of Symposium Pure Math., vol. 35, Part I, pp. 111–114. Amer. Math. Soc., Providence (1979)
Xu, Y.: Orthogonal polynomials for a family of product weight functions on the spheres. Can. J. Math. 49, 175–192 (1997)
Acknowledgements
This paper is completed under the guidance of my Ph.D. supervisor Dr. Feng Dai. I wish to express my deepest gratitude to him. And this work is supported by the National Natural Science Foundation of China (Project no. 11701082).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Feng Dai.
Rights and permissions
About this article
Cite this article
Ye, W. Almost everywhere convergence of the Bochner–Riesz means for the Dunkl transforms of weighted \(L^{p}\)-functions. Ann. Funct. Anal. 11, 981–1006 (2020). https://doi.org/10.1007/s43034-020-00065-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s43034-020-00065-x