Bulletin of Mathematical analysis and Applications ISSN: 1821-1291, URL: http://www.bmathaa.org Volume 1, Issue 2, (2009), Pages 1-16
Weierstrass transform associated with the Hankel operator ∗
Slim Omri, & Lakhdar Tannech Rachdi
Abstract Using reproducing kernels for Hilbert spaces, we give best approximation for the Weierstrass transform associated with the Hankel transform. Also, estimates of extremal functions are checked.
1 Introduction
The Hankel transform Hµ, µ > −1/2, is defined for all integrable functions on r2µ+1 [0, +∞[ with respect to the measure dr as 2µΓ(µ + 1) Z +∞ 1 2µ+1 Hµ(λ) = µ f(r)jµ(λr)r dr, 2 Γ(µ + 1) 0 where jµ is the modified Bessel function of the first kind and index µ.
Many harmonic analysis results related to the transform Hµ are established in [6, 9, 13, 18, 20, 21].
Our purpose in this work is to define and study the Weierstrass transform Wµ,t associated with the Hankel transform Hµ.
This transform is defined by Z +∞ s2µ+1 Wµ,t(f)(r) = Et(r, s)f(s) µ ds, 0 2 Γ(µ + 1) where Et(r, s), t > 0 is the heat kernel associated with the Hankel transform which will be defined later. This integral transform which generalizes the usual Weierstrass transform [11, 15, 16], solves the heat problem
∂u `µ(u)(r, t) = ∂t (r, t), u(r, 0) = f(r),
∗Mathematics Subject Classifications: 44A15, 46E22, 35K05. Key words: Weierstrass transform, best approximation, Hankel transform, reproducing kernel, extremal function. c 2009 Universiteti i Prishtin¨es,Prishtin¨e,Kosov¨e. Submitted March, 2009. Published May, 2009.
1 2 Weierstrass transform associated with the Hankel operator
where `µ is the singular differential operator defined on ]0, +∞[ by ∂2 2µ + 1 ∂ ` = + . µ ∂r2 r ∂r Building on the ideas of Saitoh, Matsuura, Fujiwara and Yamada [5, 12, 14, 15, 16], and using the theory of reproducing kernels [2], we give a best approx- imation of this transform and nice estimates of the associated extremal function.
r2µ+1 Let L2([0, +∞[, dr) be the Hilbert space of square integrable func- 2µΓ(µ + 1) r2µ+1 tions on [0, +∞[ with respect to the measure dr, and h.|.i its inner 2µΓ(µ + 1) µ product.
µ For ν ∈ R, we consider the Sobolev type space Hν ([0, +∞[), consisting of func- r2µ+1 tions f ∈ L2([0, +∞[, dr) such that the function 2µΓ(µ + 1)
2 ν/2 λ 7−→ (1 + λ ) Hµ(f)(λ), r2µ+1 belongs to the space L2([0, +∞[, dr). Then for ν > µ+1, Hµ([0, +∞[) 2µΓ(µ + 1) ν is the Hilbert space when equipped with the inner product
Z +∞ 2µ+1 2 ν λ hf|giν = (1 + λ ) Hµ(f)(λ)Hµ(g)(λ) µ dλ. 0 2 Γ(µ + 1) Moreover, the kernel
Z +∞ 2µ+1 jµ(λr)jµ(λs) λ Kν (r, s) = 2 ν µ dλ, 0 (1 + λ ) 2 Γ(µ + 1) µ is a reproducing kernel of the space Hν ([0, +∞[), where
+∞ 2µΓ(µ + 1) X (−1)n z j (z) = J (z) = Γ(µ + 1) ( )2n, z ∈ , µ zµ µ n!Γ(µ + n + 1) 2 C n=0 and Jµ is the Bessel function of the first kind and index µ. Using the properties of the Hankel transform Hµ and its connection with the convolution product, we show that the Weierstrass transform Wµ,t is a bounded linear operator from r2µ+1 Hµ([0, +∞[) into L2([0, +∞[, dr) and that for all f ∈ Hµ([0, +∞[), ν 2µΓ(µ + 1) ν
kWµ,t(f)k2,µ 6 ||f||ν . µ Next, for ξ > 0, we define on the space Hν ([0, +∞[), the new inner product by setting hf|giν,ξ = ξhf|giν + hWµ,t(f)|Wµ,t(g)iµ. S. Omri, & L. T. Rachdi 3
µ We show that Hν ([0, +∞[) equipped with the inner product h.|.iν,ξ, is a Hilbert space and we exhibit a reproducing kernel, that is Z +∞ j (λr)j (λs) λ2µ+1 K (r, s) = µ µ dλ. ν,ξ 2 ν −2tλ2 µ 0 ξ(1 + λ ) + e 2 Γ(µ + 1) The last section of this paper is devoted to study the extremal function. More r2µ+1 precisely, for all ν > µ + 1, ξ > 0 and g ∈ L2([0, +∞[, dr), the 2µΓ(µ + 1) infimum of
n 2 2 µ o ξ||f||ν + ||g − Wµ,t(f)||2,µ; f ∈ Hν ([0, +∞[) ,
∗ is attained at one function fξ,g, called the extremal function. We establish also, the following estimates
µ • For all f ∈ Hν ([0, +∞[) and g = Wµ,t(f), ∗ lim fξ,g − f = 0. ξ→0+ ν
µ • For all f ∈ Hν ([0, +∞[) and g = Wµ,t(f), ∗ lim fξ,g(r) = f(r), uniformly. ξ→0+
2 The Hankel transform
We denote by
• dγµ the measure defined on [0, +∞[ by r2µ+1 dγ (r) = dr, µ 2µΓ(µ + 1)
p • L (dγµ) , p ∈ [1, +∞], the space of measurable functions f on [0, +∞[ satisfying
1 Z +∞ p p kfkp,µ = |f(r)| dγµ(r) < +∞, if p ∈ [1, +∞[; 0 kfk∞,µ = ess sup |f(r)| < +∞, if p = +∞. r∈[0,+∞[
2 • h.|.iµ the inner product on L (dγµ) defined by Z +∞ hf|giµ = f(r)g(r)dγµ(r). 0
•C∗,0(R) the space of even continuous functions f on R such that lim f(r) = 0. |r|→+∞ 4 Weierstrass transform associated with the Hankel operator
Let `µ be the Bessel operator defined on ]0, +∞[ by 2µ + 1 ` (u) = u00 + u0, µ r then for all λ ∈ C, the following problem,
2 `µ(u) = −λ u, u(0) = 1, u0(0) = 0,
admits a unique solution given by jµ(λ.), where
+∞ 2µΓ(µ + 1) X (−1)n z j (z) = J (z) = Γ(µ + 1) ( )2n, z ∈ , (2.1) µ zµ µ n!Γ(µ + n + 1) 2 C n=0
and Jµ is the Bessel function of the first kind and index µ [1, 3, 10, 22].
The eigenfunction jµ satisfies the following properties
• The function jµ has the Mehler integral representation, for all x ∈ R
2Γ(µ + 1) Z 1 √ (1 − t2)µ−1/2 cos x t dt, if µ > −1/2; jµ(x) = πΓ(µ + (1/2)) 0 cos x, if µ = −1/2.
• For all n ∈ N and x ∈ R
(n) jµ (x)| 6 1. (2.2)
• The function jµ satisfies the product formula [10, 22], for all r, s ∈ [0, +∞[ Z π Γ(µ + 1) p 2 2 2µ jµ( r + s + 2rs cos θ)(sin θ) dθ, if µ > −1/2; j (r)j (s) = Γ(µ + 1/2)Γ(1/2) 0 µ µ j (r + s) + j (r − s) −1/2 −1/2 , if µ = −1/2. 2 Using the product formula, we will define and study the Hankel translation operator and the convolution product.
µ Definition 2.1 1) For all r ∈ [0, +∞[, the Hankel translation operator τr is p defined on L (dγµ) by Z π Γ(µ + 1) p 2 2 2µ f( r + s + 2rs cos θ)(sin θ) dθ, if µ > −1/2; τ µ(f)(s) = Γ(µ + 1/2)Γ(1/2) 0 r f(r + s) + f(|r − s|) , if µ = −1/2. 2 S. Omri, & L. T. Rachdi 5
1 2) The convolution product of f, g ∈ L (dγµ) is defined by Z +∞ µ f ∗µ g(r) = τr (f)(s)g(s)dγµ(s). 0 The following assumptions hold
• The product formula can be written
µ ∀(r, s) ∈ [0, +∞[×[0, +∞[, τr (jµ(λ.))(s) = jµ(λr)jµ(λs). p µ • For all f ∈ L (dγµ), p ∈ [1, +∞], and for all r ∈ [0, +∞[ the function τr (f) p belongs to the space L (dγµ) and
µ kτr (f)kp,µ 6 kfkp,µ. (2.3) p • Let p, q, r ∈ [1, +∞] be such that 1/p+1/q = 1+1/r. For all f ∈ L (dγµ) and q r g ∈ L (dγµ), the function f ∗µ g belongs to L (dγµ) and we have the following Young inequality kf ∗µ gkr,µ 6 kfkp,µkgkq,µ. (2.4) 1 µ 1 • For all f ∈ L (dγµ) and λ ∈ [0, +∞[, the function τλ (f) belongs to L (dγµ) and we have Z +∞ Z +∞ µ τλ (f)(r)dγµ(r) = f(r)dγµ(r). (2.5) 0 0
1 Definition 2.2 The Hankel transform Hµ is defined on L (dγµ) by [17] Z +∞ Hµ(f)(λ) = f(r)jµ(rλ)dγµ(r), λ ∈ R, 0 where jµ is the modified Bessel function defined by the relation (2.1).
The Hankel transform satisfies the following properties
1 • For all f ∈ L (dγµ) the function Hµ(f) belongs to the space C∗,0(R) and
kHµ(f)k∞,µ 6 kfk1,µ. 1 • For all f ∈ L (dγµ) and r ∈ [0, ∞[ µ Hµ τr (f) (λ) = jµ(rλ)Hµ(f)(λ). (2.6) 1 • For f, g ∈ L (dγµ) Hµ(f ∗µ g) = Hµ(f)Hµ(g).
1 Theorem 2.3 (Inversion formula for Hµ) Let f ∈ L (dγµ) such that Hµ(f) ∈ 1 L (dγµ), then for almost every r ∈ [0, +∞[, we have Z +∞ f(r) = Hµ(f)(λ)jµ(λr)dγµ(λ). 0 6 Weierstrass transform associated with the Hankel operator
Theorem 2.4 (Plancherel theorem) The Hankel transform Hµ can be extended 2 to an isometric isomorphism from L (dγµ) onto itself. In particular for all 2 f, g ∈ L (dγµ), we have (Parseval equality) Z +∞ Z +∞ f(r)g(r)dγµ(r) = Hµ(f)(λ)Hµ(g)(λ)dγµ(λ). 0 0 1 2 Remark 2.5 i) Let f ∈ L (dγµ) and g ∈ L (dγµ), by the relation (2.4), the 2 function f ∗µ g belongs to L (dγµ), moreover
Hµ(f ∗µ g) = Hµ(f)Hµ(g). 2 ii) For all f, g ∈ L (dγµ) the function f ∗µ g belongs to the space C∗,0(R) and we have f ∗µ g = Hµ Hµ(f)Hµ(g) . (2.7)
3 Weierstrass transform associated with the Han- kel operator
In this section, we will define and study the Weierstrass transform associated with Hµ. For this we define some Hilbert spaces and we exhibit their reproduc- ing kernels.
Let ν be a real number, ν > µ + 1. We denote by
µ 2 • Hν ([0, +∞[) the subspace of L (dγµ) formed by the functions f, such that the maps 2 ν/2 λ 7−→ (1 + λ ) Hµ(f)(λ), 2 belongs to L (dγµ).
µ • h.|.iν the inner product on Hν ([0, +∞[) defined by Z +∞ 2 ν hf|giν = (1 + λ ) Hµ(f)(λ)Hµ(g)(λ)dγµ(λ). 0 µ • ||.||ν the norm of Hν ([0, +∞[) defined by p ||f||ν = hf|fiν . Remark 3.1 For ν > µ + 1, the function 1 λ 7−→ , (1 + λ2)ν/2 2 µ belongs to L (dγµ). Hence for all f ∈ Hν ([0, +∞[), the function Hµ(f) belongs 1 to L (dγµ), then by inversion formula 2.3, we have for almost every r ∈ [0, +∞[ Z +∞ f(r) = Hµ(λ)jµ(λr)dγµ(λ). 0 S. Omri, & L. T. Rachdi 7
Proposition 3.2 For ν > µ + 1 the function Kν defined on [0, +∞[×[0, +∞[ by Z +∞ jµ(λr)jµ(λs) Kν (r, s) = 2 ν dγµ(λ), 0 (1 + λ ) µ is a reproducing kernel of the space Hν ([0, +∞[), that is i) For all s ∈ [0, +∞[, the function
r 7−→ Kν (r, s),
µ belongs to Hν ([0, +∞[). µ ii) (The reproducing property) For all f ∈ Hν ([0, +∞[) and s ∈ [0, +∞[,
hf|Kν (., s)iν = f(s).
Proof. i) From Remark 3.1 and the relation (2.2), we deduce that for all s ∈ [0, +∞[, the function
j (λs) λ 7−→ µ , (1 + λ2)ν
1 2 belongs to L (dγµ) ∩ L (dγµ). Then, the function Kν is well defined and by Theorem 2.3, we have j (λs) K (r, s) = H µ (r). ν µ (1 + λ2)ν
By Plancherel theorem, it follows that for all s ∈ [0, +∞[, the function Kν (., s) 2 belongs to L (dγµ), and we have j (λs) H K (., s) (λ) = µ . (3.8) µ ν (1 + λ2)ν
Again, by the relation (2.2) and Remark 3.1, it follows that the function
2 ν/2 λ 7−→ (1 + λ ) Hµ Kν (., s) (λ),
2 belongs to L (dγµ).
µ ii) Let f be in Hν ([0, +∞[). For every s ∈ [0, +∞[, we have Z +∞ 2 ν hf|Kν (., s)iν = (1+λ ) Hµ(f)(λ)Hµ(Kν (., s))(λ)dγµ(λ), 0 and by the relation (3.8), we get
Z +∞ hf|Kν (., s)iν = Hµ(f)(λ)jµ(λs)dγµ(λ). 0 The result follows from Remark 3.1. 8 Weierstrass transform associated with the Hankel operator
The heat equation associated with the Hankel transform is given by
∂ ` u(r, t) = u(r, t), (3.9) µ ∂t where `µ is the Bessel operator defined above.
Let E be the kernel defined by
Z +∞ −tλ2 E(r, t) = e jµ(rλ)dγµ(λ) (3.10) 0 2 e−r /4t = . (2t)µ+1
Then, the kernel E solves the equation (3.9).
Definition 3.3 The heat kernel associated with the Hankel transform is defined by
µ Et(r, s) = τr E(., t) (s) (3.11)
2 2 e−(r +s )/4t irs = j ( ). (2t)µ+1 µ 2t
Then, we have the following properties
i) For all t > 0, Et > 0. ii) From the relations (2.3),(2.6),(3.10) and (3.11), for all t > 0, r ∈ [0, +∞[, 1 the function Et(r, .) belongs to L (dγµ) and for all λ ∈ [0, +∞[, we have