1. Introduction
Let
,
be the the Dunkl operators (see [
1]) associated to arbitrary finite reflection group
G and non-negative multiplicity function
k. These are differential-difference operators generalizing the usual partial derivatives. The Dunkl kernel
on
associated with
G and
k, introduced by C. F. Dunkl in [
2], generalizes the usual exponential function (to which it reduces in the case
). This kernel is especially of interest as it gives rise to an integral transformation on
associated with
G,
k and to the weight measure
(where
is the weight function given in (
23)). It is called the Dunkl transform
, and it is defined on
by:
where
is the Mehta-type constant given in (
24), and
and
are the Banach spaces consisting of measurable functions
f on
equipped with the norms:
The Dunkl transform
extends uniquely to an isometric isomorphism on
. In particular, if
, then the Dunkl operators are the usual partial derivatives and the Dunkl kernel is the usual exponential function so that the Dunkl transform is exactly the usual Fourier transform
defined by,
This is why the Dunkl transform can be considered one of the deformed Fourier transforms (see e.g., [
3] for other deformed Fourier transforms). Moreover, if
is a radial function on
, then
where
,
, is the Hankel transform (also known as the Fourier-Bessel transform) defined by (see [
4]),
where
is the spherical Bessel function (see [
5]), and
is the gamma function.
In order to present our result, let , and be the weighted Lebesgue spaces, and let , be the Dunkl–Sobolev space on . Then we prove the following new Heisenberg-type uncertainty inequalities for the Dunkl transform.
Theorem 1. Let.
- 1.
If, then there exists a constantsuch that for every nonzero function, - 2.
If, then there exists a constantsuch that for every nonzero function,
The proof of these inequalities is based on Nash’s inequality [
6] (Proposition 3.2) and Carlson’s inequality [
6] (Proposition 3.3) for the Dunkl transform, combined with the following new inequalities:
- 1.
For all
,
and
,
- 2.
For all
and
,
Theorem 1 is a variation of the following uncertainty inequalities (see [
6,
7]):
- 1.
There exists a constant
such that for all
,
- 2.
There exists a constant
such that for all
,
Heisenberg-type uncertainty inequalities (
6) and (
7) are related to Inequality (
11), since in both cases the proof is based on Nash’s inequality and Carlson’s inequality for the Dunkl transform. Unfortunately, these three inequalities are not sharp, and the best constants and extremal functions remained unknown until now. However, the proof of Heisenberg-type uncertainty inequality (
10) can be obtained from either the Faris-type local uncertainty principle [
7] (Theorem A) or from the Benedicks–Amrein–Berthier uncertainty principle [
7] (Theorem B). Moreover, the sharp constant in (
10) is well known only for the special case
, and it is equal to
; equality in (
10) occurs if and only if
for some
, and
(see [
8,
9,
10]).
Our second result will be the following sharp Shannon-type and logarithmic Sobolev inequalities for the Dunkl transform.
Theorem 2. Let.
- 1.
There exists an optimal constantsuch that for all nonzero, - 2.
If, then there exists a positive constantsuch that for all nonzero f belongs to, - 3.
There exists a positive constantsuch that for all nonzero,
Shannon-type inequality (
12) is sharp, and the constant
that appears in (
13) is the optimal constant of the following Sobolev-type inequality for the Dunkl transform (see [
11]): For all
,
More generally (see [
11] (Theorem 1.1)), if
, and if
, then there exists a constant
such that for all
f in the Dunkl–Sobolev space
we have
Using the sharp Shannon’s inequality (
12) and the logarithmic Sobolev inequality (
13) we derive for
, the well known sharp Heisenberg’s uncertainty inequality for the Dunkl transform (see [
9]) is:
Moreover, from Inequalities (
12) and (
14), we give another, short proof of Inequality (
10).
Next, from the Sobolev-type inequality (
16), we prove the following
-logarithmic Sobolev inequality.
Theorem 3. If, then for allsuch that, we havewhereis the constant of the Sobolev inequality (16). Consequently, we obtain the following -Heisenberg-type uncertainty inequality and -Nash-type inequality for the Dunkl transform.
Theorem 4. - 1.
If, then there exists a constantsuch that for all, - 2.
If, then for all, whereis the constant of the Sobolev inequality (
16).
The remainder of this paper is arranged as follows: in the next section we recall some useful results associated to the Dunkl transform. In
Section 3, we prove a sharp Shannon-type inequality, and in
Section 4, we prove some new logarithmic Sobolev inequalities.
Section 4 is devoted to the study of a Nash-type inequality. Connecting the previous results, we give an application of the uncertainty principle, showing some new Heisenberg-type uncertainty inequalities.
3. Sharp Shannon-Type Inequality
Following Shannon (see [
15]), the entropy of a probability density function
on
is defined by
In this section, we continue the study of the sharp Shannon-type inequality in the Dunkl setting started in [
16]. First, we consider the weighted Lebesgue spaces
,
and
by
where
, for
(cf. [
16]). Notice that if
, we have:
Then
f and
belong to
. On the other hand, if
and
are in
, then
Moreover, we have the following result.
Lemma 1. Let and let . Then there exists a positive constant such that for all , Proof. Let
and let
, where
is the open ball of
of radius
r. Then by Hölder’s inequality and (
26),
Minimizing the right-hand side of the last inequality by
we obtain (
44), with
This completes the proof. □
Remark 1. From the previous lemma, if , then for all , In particular, we have if and In the following, we recall the sharp Beckner-type logarithmic inequality proved in [
16] (Theorem 6.1).
Theorem 5. Let. Then for all nonzero,where is the sharp constant, and equality holds if and only if Consequently, by following the same process as in [
17], we derive the following Shannon-type inequality in the Dunkl setting.
Corollary 1. Let . Then for all nonzero ,where Proof. Let
such that
. Then by Inequality (
48) and Jensen’s inequality,
Now let
,
, the function is defined by:
Then
, and by replacing
f with
in Inequality (
53), we obtain
Minimizing the right-hand side of the last inequality with
, we get
Finally, if
f is any nonzero function in
, then by replacing
f by
in (
56), we obtain (
51). □
Moreover, we have the following improvement:
Theorem 6. Let. Then for any nonzero, we havewhereis the sharp constant, and it is attained byup to dilation, whereis given by Proof. Let
be a nonzero function such that
, and let function
satisfy
. Then by Jensen’s inequality,
Now replacing
f by
(which is defined in (
54)) in Inequality (
60), we obtain:
By optimizing (
61) with
, we have
Hence,
where
.
Now, since
gives the equality of (
60), then
with
attains the equality in (
62). Moreover, since
then by taking
in the following equality
we obtain
and this yields
.
Finally, if
f is any nonzero function in
, then by replacing
f by
in (
62), we obtain Inequality (
57). □
Remark 2. The constant in the Shannon-type inequality (51) coincides with the sharp constant of the Shannon-type inequality (57) for large enough . In fact, from Stirling’s approximationwe have for . Moreover, the optimal Shannon-type inequality (
57) can be written as
Now, if , then . Thus by (57), we have for all and all Moreover, from Theorem 5, we have for all and all As a corollary of Theorem 6, we obtain the following:
Corollary 2. Let and . Then for any non-negative function where , and the constant on the right-hand side is the best possible and is given by (
58).
4. Nash-Type and Logarithmic Sobolev-Type Inequalities and Applications
For
, we consider the Dunkl–Sobolev space on
(cf. [
14]) defined by
or equivalently (by Plancherel’s Formula (
33))
Noting that if
, then
and
are in
. Therefore, from (
47),
is in
if
. Thus, it follows from the Riemann–Lebesgue lemma that
f is uniformly continuous on
, vanishing at infinity. If, in addition
(e.g., if
), then by the inverse Dunkl transform Theorem (
32),
f is almost everywhere equal to a function in
. Moreover, we have the following Sobolev’s embedding theorem:
Proposition 1 (Sobolev’s embedding theorem).
If , then for all ,Moreover, there exists a constant such that for all ,where Proof. Let
. Then by the inversion formula for the Dunkl transform, we have
Therefore, by the Cauchy–Schwarz inequality and by (
33),
Now replacing
f by
in the last inequality, we obtain
Minimizing the right-hand side with
we obtain the desired result. □
Remark 3. We recall that the inhomogeneous Dunkl–Sobolev spaces endowed with the normand the homogeneous Dunkl–Sobolev spaces defined in a similar way by replacing with in (77) have been defined and studied in [18]. The fact that is a simple consequence of Plancherel’s identity. When , Sobolev’s embedding theorem in the Dunkl setting follows from Hölder’s inequality and the Hausdorff–Young inequality (34). Indeed, if , then Hence, for and , we have In particular, for and , 4.1. New Heisenberg-Type Uncertainty Inequalities
In this section, we revisit and prove new Heisenberg-type uncertainty principles for the Dunkl transform. The first result is the following well-known uncertainty inequality (see [
8,
9,
10]).
Theorem 7. For every ,with equality if and only if for some and . It is also well known that by using a dilation argument the last inequality is equivalent to the following sharp additive uncertainty inequality (see e.g., [
9] (Theorem 1.1)):
with equality if and only if
for some
.
More generally, we recall the following results (see [
6,
7]):
Theorem 8. Let.
- 1.
There exists a constantsuch that for all, - 2.
There exists a constantsuch that for all,
Notice that if the time dispersion
(or
) or the frequency dispersion
are not finite, then the last inequalities are trivial. Hence we may assume that
in Inequalities (
80)–(
82), and
in Inequality (
83).
The Proof of Inequality (
82) can be obtained from either the Faris-type local uncertainty inequalities [
7] (Theorem A) or from the Benedicks–Amrein–Berthier uncertainty principle [
7] (Theorem B). The Proof of Inequality (
83) can be obtained by combining the following Nash-type inequality [
6] (Proposition 3.2) and Carlson-type inequality [
6] (Proposition 3.3) in the Dunkl setting:
Proposition 2. Let .
- 1.
A Carlson-type inequality: There exists a positive constantsuch that for all, - 2.
A Nash-type inequality: There exists a positive constantsuch that for all,
In [
6], we have proved that the Nash-type inequality (
85) is a key tool in proving uncertainty inequalities for the Dunkl transform. Moreover, from Lemma 1 and the Nash-type inequality (
85), we can give another proof of the Heisenberg-type uncertainty inequality (
82) (only for
).
Corollary 3. Letand. Then
- 1.
There exists a positive constantsuch that for all, the time-dispersion satisfies - 2.
There exists a positive constantsuch that for all, the frequency-dispersion satisfies - 3.
There exists a positive constantsuch that for all
Proof. Let
be a nonzero function. Then from Lemma 1, the function
f belongs in
, and
Then we derive (
86). Moreover, from Nash’s inequality (
85), we have (
87). Combining (
86) and (
87), we obtain (
88). □
Remark 4. The last corollary is valid only for (not for all ), but it is stronger than Inequality (82), since here Inequalities (86) and (87) give separate lower bounds for the values of the time-dispersion and the frequency-dispersion , which give more information than the lower bound of the product between them in Heisenberg’s inequality (
82).
Moreover, we have the following new uncertainty inequalities involving and norms.
Theorem 9. Let.
- 1.
If, then there exists a positive constantsuch that for every nonzero, - 2.
If, then there exists a positive constantsuch that for every nonzero,
Proof. From Inequality (
74), we have
and from (
84), we have
Then Inequality (
90) follows from (
86) and (
92), and Inequality (
91) follows from (
92) and (
93). □
4.2. Logarithmic Sobolev Inequalities and the Uncertainty Principles
First recall the Sobolev-type inequality for the Dunkl transform, recently proved in [
11] (Theorem 1.1, Theorem 6.1).
Theorem 10. Suppose that , and let . Then for all ,where is the sharp constant given by Then we have the following logarithmic Sobolev inequality.
Corollary 4. Let . Then for all nonzero , where is the constant given in (95). Proof. Let
such that
. Then by (
94) and Jensen’s inequality, we have
Finally, if
f is any nonzero function in
, then by replacing
f by
in (
97), we obtain (
96). □
Now, using the asymptote of the sharp Shannon’s inequality (
57) and the sharp Sobolev inequality (
94), we derive the sharp Heisenberg’s uncertainty inequality (
80).
Corollary 5. Let . Then there exists a positive constant such that for all f belonging to , we havewhere Moreover, for , we have the sharp inequality Proof. Let
be a nonzero function. Then, from (
68), for
we have
Moreover, by (
96), we have
Hence,
which implies that
where
. Now using the sharp constants (
58) and (
95), we have
Then by Stirling’s approximation
we have for
Thus for
,
which is the sharp constant in Heisenberg’s inequality (
80). □
Now from [
16] (Theorem 6.2) we have the following Beckner-type inequality for the Dunkl transform, although not with the sharp constant
c.
Theorem 11. For any nonzero , we have This inequality is stronger than Inequality (
96) since it implies the following logarithmic Sobolev inequality.
Corollary 6. Let . Then there exists a constant such that for all nonzero , Proof. For any nonzero function
, set the measure
by
Then, by Jensen’s inequality and Plancherel’s Frmula (
33),
Thus by Inequality (
110), we obtain the desired result, with
. □
The last result allows us to give a new, short proof for Heisenberg’s Inequality (
82).
Corollary 7. Let . Then there exists a positive constant such that for every , we have Proof. Let
be a nonzero function. Then from (
68) and (
111), we have
This implies (
114) with
. □
Following we give another formulation for the logarithmic Sobolev inequality (
96).
Proposition 3. Let For any and , the following inequality holds Proof. Involving (
96) and Plancherel’s Formula (
33), we have
Moreover, if we introduce a parameter
we have
Thus combining the relations (
118) and (
119), we derive the result. □
Remark 5. - 1.
One can chooseand obtain the following inequality Involving the inequalities (
108)
and (
117)
, we derive that for and then the inequality does not depend on the dimension.
- 2.
We can derive (
96)
from (
116).
Indeed, we optimize the parameter appearing in Proposition 3, and then the equivalent form of (
116)
is the inequality (
96).
More precisely, we set and optimize the right-hand side with the parameterSince and the minimum ofis realized bywith then the minimum of the right-hand side is
Now we will study the logarithmic Sobolev inequalities on the space
defined by
The sharp Sobolev inequality implies the following sharp logarithmic Sobolev inequality.
Theorem 12. Let For any , the following inequality holds The constant in the right-hand side is the best possible, and it is attained on the closure of a Weyl chamber by for any .
Proof. Let
, and for simplicity we assume that
By (
96), we have
where
is the best possible constant for the Dunkl–Sobolev inequality given by (
95). Plugging
for
with
and
into (
122) and noticing
and
we see that
By using the Stirling formula:
, as
, we obtain
Letting
, we obtain the sharp inequality (
121).
To see the optimality, we take , and by simple calculations we derive the result. □
Remark 6. Proceeding as in Proposition 3 and Remark 5, we prove that (
121)
is equivalent to The equality is attained on the closure of a Weyl chamber by .
Corollary 8. Let For any non-negative with it holds that The constant in the right-hand side is the best possible, and it is attained on closure of a Weyl chamber by with
Remark 7. The resulting inequality can be seen by the following normalized form: for any non-negative f with and
We end this section by involving the logarithmic Sobolev inequality on the space with the Gaussian measure.
Theorem 13 (The Gaussian logarithmic Sobolev inequality).
Let and 0. For any non-negative function g and any G invariant and satisfying , we havewhere Proof. Substituting
into
in (
124), and since
one can see that
Letting
, one can see that
After applying the
-invariant scaling
, we have
Since
choosing
we obtain the desired result. □
5. -Nash-Type and Uncertainty Inequalities
Let
, and let
. Then from [
11] (Theorem 1.1)), there exists a constant
such that for all
:
This inequality extends naturally to the Dunkl–Sobolev space (see also [
19,
20])
Then we have the following -logarithmic Sobolev inequality.
Theorem 14. Let . Then for all such that , we havewhere is the constant of the Sobolev inequality (
129).
Proof. This proof is inspired by [
21]. Let
s be a real number such that
, and let
be a function such that
, by which
is a probability measure. Then by Hölder’s inequality, we have
where
Then
, and since
, we have
Thus by the Sobolev inequality (
129), we obtain
Now, since (see [
22]),
then by taking the limit of both sides in (
133) as
s approaches
p, we obtain
The density argument completes the proof. □
Remark 8. Instead of the limit in (
133),
we can obtain the result by Jensen’s inequality: Hence, we derive the following -Heisenberg uncertainty inequality for the Dunkl transform.
Corollary 9. Let . Then for all , we havewhere Proof. Let
be a nonzero function such that
. Then, from (
130) and (
57), for
and
, we have
Therefore,
which implies that
Replacing f with in the last inequality, we obtain the result. □
It’s well know that Sobolev’s inequality and Nash’s inequality are equivalent (see e.g., [
11,
23,
24]). The Nash-type inequality (
85) (for
and with the assumption
) can be derived from the Sobolev-type inequality (
94) by using only Hölder’s inequality (see [
11] (Theorem 4.2)). Now we will use the
-logarithmic Sobolev inequality (
130) and the techniques of Beckner in [
24] to derive an
-Nash-type inequality for the Dunkl transform.
Corollary 10. Let . Then for all ,where is the constant of the Sobolev inequality (
129).
Proof. Let
, and let
such that
. Then, by Jensen’s inequality and the
-logarithmic Sobolev inequality (
130),
Now replacing
f with
in the last inequality, we obtain
as desired. □
Remark 9. If instead of the -logarithmic Sobolev inequality (130) we use the logarithmic Sobolev inequality (96) in the last proof, we obtain, with the assumption ,where is the sharp constant in Sobolev’s inequality (96) given by (95). Therefore the optimal constant in the Nash-type inequality (145) should be for . Theorem 15. Let . For any non-negative function such that belongs to , we havewhere is the best possible constant, and it is attained by with Proof. Involving the fact that
Green’s identity in the Dunkl setting (cf. [
1,
14]) and Hölder’s inequality, we get
To see the optimality, we substitute
in (
146), involve the identity (
65), use the fact that
, and, after standard calculations, we derive the result. □
Remark 10. Let . If we assume that , we derive that the inequality (80) is a direct consequence of (146) by setting and into f. We illustrate a different method using the logarithmic Sobolev inequality and the generalized version of the Shannon inequality in
. Firstly, we recall the following results from [
11].
Theorem 16. Assume that . Let . Then for any , we havewhere Moreover, the constant is sharp, and equality is achieved if and only if f is supported on the closure of a Weyl chamber, where it takes the form , where .
Remark 11. Assume that , and let . We have (see [11]) where is the constant defined by (149), and . We proceed as in the proof of Theorem 14, and, involving (
148), we get the following inequality:
Corollary 11. Let For any with where is the constant given by (149). Theorem 17. Let . For any non-negative function such that , we havewhere and are given in (149) and (58), respectively. In particular, Proof. From Theorem 6 with
, we have
where
From the sharp version of the logarithmic Sobolev inequality (
152) with the best constant (
149),
Combining (
155) with (
156) and (
157), we immediately obtain that for any
The proof is complete. □
Let p and q be two positive reals (not necessarily Hölder conjugate exponents). We proceed as in Theorem 15 to derive a generalized version of the uncertainty principle.
Proposition 4. Let and Then for any nonnegative with , it holdswhere and are given in (149) and (58), respectively. Theorem 18 (Generalized uncertainty principle).
Let and For any non-negative function with , it holdswhere and denote and respectively. Proof. Replacing
in (
70) and using
, we get
Similarly substituting
in (
153) and using
, we derive
Combining inequalities (
158) and (
159),
Since
it follows from (
160) that
and thus we have
The proof is complete. □