THE d-PLANE RADON TRANSFORM ON THE TORUS Tn
Ahmed Abouelaz
Abstract
We define and study thed-plane Radon transform, namelyR,on then- dimensional (flat) torus. The transformationRis obtained by integrating a suitable functionf over alld-dimensional geodesics (d-planes in the torus).
We specially establish an explicit inversion formula of R and we give a characterization of the image, under the d-plane Radon transform, of the space of smooth functions on the torus.
MSC 2010: Primary 44A12; Secondary 05B35, 11D04
Key Words and Phrases: d-plane Radon transform on the torus, Fourier transform, inversion formula, characterization of the range.
1. Introduction
The Radon transform is a fundamental topic in the integral geometry.
It plays a preponderant role in many various fields of mathematics, physics and medecine.
This transform was studied and generalized in the case of Grassman manifolds and projective spaces, by many authors for example: I.M. Gelfand, M.I. Graev, and S.I. Shapiro[6],S. Helgason [8],B. Rubin[11],F. Rouvi´ere [10].
c 2011 Diogenes Co., Sofia
pp. 233–246 , DOI: 110.2478/s13540-011-0014-8
We briefly recall the definition of the classicald-plane Radon transform Rc on the Euclidean space Rn, d being an integer such that 1≤d≤n−1, with n≥2.The classicald-plane Radon transformRc is defined by
Rcf(ξ) =
ξ
f(x)dm(x), (1.1)
where f is a complex-valued C∞-function with compact support, dm(x) is the Euclidean measure on the d-plane ξ. The essential problem of the integral geometry is to recover f from Rcf.
It is noted that the inversion formula for the Radon transform is closely related to fractional calculus, see [12], pages 29 and 30, formula (2.4). See also the papers [9], [13], [14], [2], [1].
The goal of this paper is to define an analogue of (1.1) and to give a inversion formula, that is to recover a continuous function f on the n- dimensional torus from the d-plane Radon transformRf.
The integral geometry in the torus was studied by I.M. Gelfand, S.G.
Gindikin and M.I. Graev in their book [7]. These authors have given the inversion formula based on the Poisson formula (see formulas (5.8), (5.13) of [7]). In addition, the inversion formula of Radon transform on the torus given in the book [7] is different from ours (see [5], see also formula 4.1 in Section4). In this work, we mainly use algebraic-arithmetic techniques and also the techniques of Fourier analysis (see Theorem4.1and its demonstra- tion). The origin of this problem was introduced by Strichartz [15], who gave a solution for n= 2. But, his method does not extend in an obvious way to n-dimensional torus (see Remark 3.2 in the paper [5]).
We consider here the n-dimensional (flat) torus Tn = (R/Z)n and the d-plane Radon transform defined by integrating f along all d-dimensional geodesics (d-planes in the torus) of Tn. Arithmetic techniques are used in this paper, as in the case of the Radon transform on Zn, already studied by the author and his collaborators (see [3], [4]). Our purpose of this paper is to establish an explicit inversion formula for the d-plane Radon transform on the torus Tn.
Our paper is organized as follows:
In Section 2, we shall fix some notations and also recall the properties of the parametrization of the discrete d-planes in Zn,see [3] and [4], that will be useful in the sequel of this paper.
In Section 3, we describe a suitable set of parameters for the closed d-dimensional geodesics (d-planes in the torus). The main result of this section is the parametrization’s lemma of the d-planes of the torus Tn (see Lemma 3.2). This lemma allows us to define the d-plane Radon dual of the operator R on the torus Tn.
In Section 4, we establish an inversion formula for the d-plane Radon transform (see Theorem 4.1) by extending the inversion theorem proved in [5]. We note that several classical inversion formulas for the Euclidean d-plane Radon transform involveR∗cRcf.However, in the case of the torus, the sum defining R∗Rf is not convergent in general. We shall therefore introduce a weight function (see Theorem 4.1). This result is the main theorem in this paper.
Section 5 is devoted to a range theorem characterizing the space R(C∞(Tn), whereC∞(Tn) is the space of all complex-valuedC∞-functions on Tn. See Theorem5.1for more details.
2. Notations and preliminaries
In this section, we shall fix once and for all some notations that will be useful in the sequel of this paper. For m ∈ N such that 1 ≤ m ≤n, the set consisting of all integer m×nmatrices is designated by Mm,n(Z).We denote by SL∗(k,Z) (1 ≤k ≤ n) the group of all integer k×k matrices whose determinant is equal to ±1.
It is noted that throughout this paper, every row vectorx= (x1, . . . , xn)
∈Zn is assumed to be identified with its transpose column vector tx given by
tx=
⎛
⎜⎝ x1
... xn
⎞
⎟⎠.
A matrix D= (dij)1≤i≤m,1≤j≤n∈Mm,n (Z) of rankr (1≤r≤m≤n) is called to be diagonal if dii =di for i= 1, . . . , r, and dij = 0 otherwise.
In the sequel we note this matrix by D = diag(d1, . . . , dr,0, . . . ,0), (see [3],[4]).
We begin with the following fundamental theorem which associates to each integer matrixA∈Mm,n(Z) of rankr a diagonal matrixD∈Mm,n(Z) uniquely defined by A and called the Smith normal form ofA.
Theorem 2.1. (See [3],[4]) Let A be a matrix inMm,n (Z) of rank d, with 1 ≤ m ≤ n. Then there exist two matrices L ∈ SL∗(m,Z) and V ∈SL∗(n,Z) such that
LAV =D=diag(l1, . . . , ld,0, . . . ,0),
where li>0for all i= 1, . . . , dand li|li+1 fori= 1, . . . , d−1.
In the special case when m=d(1≤d≤n) the matrixD of the above theorem becomes
D=
⎛
⎜⎜
⎜⎝
l1 0 . . . 0 0 . . . 0 0 l2 . . . 0 0 . . . 0 ... ... . .. ... ... . . . ... 0 0 . . . ld 0 . . . 0
⎞
⎟⎟
⎟⎠. (2.1)
Let D0 be the matrix obtained by replacing, in the above matrix, li by 1 for i= 1, . . . , d. In the following, we denote by Md,n(Z) the set of all matrix of rank d, and Pd,n the subset of all elements A of Md,n(Z) such that A = QD0V, where Q ∈ SL∗(d,Z) and V ∈SL∗(n,Z), see [3], [4] for more details.
Remark that ifA= (aij)1≤i≤d
1≤j≤n ∈ Pd,n,then for all k= 1, . . . , d the row vk = (ak1, . . . , akn) of the matrix A belongs to P, where
P ={a∈Zn\ {0} |d(a) = 1},
with d(a) is the greatest common divisor ofa= (a1, . . . , an). In addition the vectors system (vk)1≤k≤d is linearly independent (see [3],[4]). In the sequel, we will need the following result.
Theorem 2.2. (See [3],[4]) Let (A, b),(A, b)∈ Pd,n×Zd.Then the following two assertions are equivalent:
(1) The two systems of linear diophantine equationsAx=b andAx= b have the same solutions inZn.
(2) There exists a matrix Q∈ SL∗(d,Z) such that A =QA and b = Qb.
For the proof, see [[3], Theorem 3.7, page 205] and also [4].
In the following, we denote by (ek)1≤k≤d an orthonormal basis ofRd. 3. Closed d-dimensional geodesics (d-planes) of the torus For n≥2 let Tn = (R/Z)n be then-dimensional torus equipped with the flat Riemannian metric induced by the canonical Euclidean structure of Rn. In Tn thed-dimensional geodesic from x = (x1, . . . , xn)∈ Tn with (non zero) linearly independent vectors of Rn is
D=
⎧⎨
⎩x+ pr
⎛
⎝d
j=1
tjvj
⎞
⎠|tj ∈Rand 1≤j≤d
⎫⎬
⎭,
where pr :Rn→Tnis the natural projection. Ifd= 1,the setDcoincides with the line l
l={x+ pr (tv)|t∈R},
where the initial speed v ∈ Rn, see [5] for more details. Since v1, . . . , vd are non zero vectors, the setGof all (t1, . . . , td)∈Rdsuch that d
j=1tjvj belongs toZn is a discrete subgroup of Rdthat is of the form
G=Zv1+· · ·+Zvd,
butGcontains the subgroup of all (0, . . .0, ti,0, . . . ,0) such thattivi ∈Zn. By [5] we have τivi = (ai1, . . . , ain)∈ P fori= 1, . . . , d, andτi >0. That
is, ⎧
⎪⎨
⎪⎩
τ1v1 = (a11, . . . , a1n) ...
τdvd= (ad1, . . . , adn) .
For d= 1, we find again the results of the paper [5]. It is noted that τivi ∈ P for all i= 1, . . . , d. Let A= (aij)1≤i≤d
1≤j≤n
be the matrix which is of rank d, since the non zero vectors v1, . . . , vd are linearly independent and τi>0 for alli= 1,2, . . . , d.
Now, we state and prove some results which will be useful in the sequel of this paper. We begin with the following lemma.
Lemma 3.1. Let A= (aij)1≤i≤d
1≤j≤n ∈ Pd,n and (t1, . . . , td)∈Rd.Then (t1, . . . , td)A∈Zn if and only if(t1, . . . , td)∈Zd. (3.1) P r o o f. Ifd= 1,the equivalence (3.1) can be reduces to
t1(a11, . . . , a1n)∈Zn if and only ift1 ∈Z.
The above equivalence has been proved in [5]. Let x = (t1, . . . , td), taking A = (aij)1≤i≤d
1≤j≤n ∈ Pd,n,by Theorem 2.1, there exist Q∈SL∗(d,Z) and V ∈SL∗(n,Z) such that A=QD0V,therefore tAtx= tV tD0tQtx∈ Zn, and this implies that tD0tQtx ∈ Zn because tV ∈ SL∗(n,Z). That is, tQtx ∈ Zd since tD0 =
Id 0
, and finally tx ∈ Zd because tQ ∈
SL∗(d,Z).This proves the lemma. 2
For x ∈ Tn and A ∈ Pd,n, we denote by D(x, A) the d-plane in the torusTn given by
D(x, A) =
x+ pr ((t1, . . . , td)A)|(t1, . . . , td)∈Rd .
The map (t1, . . . , td)∈Rd→x+ pr ((t1, . . . , td)A) induces a bijection from Td onto D(x, A), since we have
pr ((t1, ..., td)A) = pr
t1, ..., td A
if and only if
t1−t1, ..., td−td
∈Zd, by the above lemma. We shall therefore lettj (j= 1, . . . , d) run over [0,1[
only in the sequel.
Lemma 3.2. (Parametrization) Let x, y ∈Tn and A, A ∈ Pd,n.We have D(x, A) =D(y, A) if and only if there exist a matrixQ∈SL∗(d,Z) and (s1, . . . , sd)∈Rd such thaty=x+ pr ((s1, . . . , sd)A) and A =QA.
The set of closed d-dimensional geodesics fromx is therefore in one-to- one correspondence with Pd,n/SL∗(d,Z).
P r o o f. It is clear that
D(x, A) = D(x, QA) =D(x+ pr ((s1, . . . , sd)QA), QA), (3.2) for any (s1, . . . , sd)∈Rdand for allQ∈SL∗(d,Z).Indeed, lety∈D(x, A), y can be written in the form
y = x+ pr ((t1, . . . , td)A)
= x+ pr
(t1, . . . , td)Q−1QA
= x+ pr ((s1, . . . , sd)QA),
where (s1, . . . , sd) = (t1, . . . , td)Q−1. If y ∈ D(x, QA) the element y have the form x+ pr((r1, . . . , rd)QA) which is equal to x+ pr ((t1, . . . , td)A), where (t1, . . . , td) = (r1, . . . , rd)Q. Conversely, assume that D(x, A) = D(y, A) whereA= (aij)1≤i≤d
1≤j≤n
and A= (aij)1≤i≤d
1≤j≤n
.The equality D(x, A)
=D(y, A) is equivalent toD(x, A) =D(x, A).IndeedD(x, A) =D(y, A) shows that x ∈ D(y, A), it follows that y = x −pr (sA) where s = (s1, . . . , sd) ∈ Rd but D(x−pr (sA), A) = D(x, A). Now, assume that D(x, A) =D(x, A). This equality shows that for all t= (t1, . . . , td)∈ Rd there exists (s1, . . . , sd) =s∈Rd such that
tA−sA ∈Zn,
for allt= (t1, . . . , td)∈Rd.But by [3], Proposition 3.10, page 209; see also [4], Proposition 2.11, page 307, we have
tA−sA ∈
b∈Zd
H(A, b) =
b∈Zd
H A, b
(disjoint union), where H(A, b) is ad-plane ofZn.Thus
tA−sA ∈H(A, b(t, s)), and
tA−sA∈H
A, b(t, s) ,
where b(t, s) and b(t, s) are vectors of Zd which depend on t and s. This gives the two systems of linear diophantine equations
⎧⎨
⎩
A(tA−sA) =b(t, s) A(tA−sA) =b(t, s)
tA−sA∈Zn, for all t= (t1, . . . , td)∈Rd.
These two systems have the same solutions, therefore by Theorem 2.2, Section 2, there exists Q ∈ SL∗(d,Z) such that A = QA and b(t, s) =
Qb(t, s),and this proves the lemma. 2
4. An inversion formula
Let f be a continuous function on Tn. We define its d-plane Radon transform Rf as the integral of f over closed d-dimensional geodesics namely (d-planes in the torus) onTn by
Rf(D(x, A)) = 1
0 · · · 1
0 f(x+ pr((t1, . . . , td)A)dt1. . . dtd, (4.1) withx∈Tn,A∈ Pd,n.As noted in the previous section,x+pr((t1, . . . , td)A) run over the wholed-plane whentj varies from 0 to 1,withj∈ {1,2, . . . , d}. Remark 4.1. Let f be a continuous function on Tn and (x, x0) ∈ (Tn)2.Then for all A∈ Pd,n,we have
Rfx0(D(x, A)) =Rf(D(x+x0, A)),
where fx0(x) =f(x+x0).It suffices to replace f by fx0 in formula 4.1.
Let x∈Tn and G(x) be the set of all d-planes ofTn fromx. LetG0 be the set G0=
x∈TnG(x) and Hx∈G(x).Denote by ΛHx the set given by ΛHx ={A∈ Pd,n|D(x, A) =Hx}.
Let Γ be the function defined on G0 by Γ (Hx) = inf
A∈ΛHx||A||2, where
||A||=
⎛
⎜⎜
⎝
1≤i≤d
1≤j≤n
|aij|2
⎞
⎟⎟
⎠
12
,
with A = (aij)1≤i≤d
1≤j≤n
and for a= (a1, . . . , an) ∈ P, ||a|| = n
j=1|aj|212 . Let S(G0) be the Schwartz space defined by a family (TN,x)N∈N, x∈Tn of semi-norms as follows
S(G0) =
F :G0−→C| sup
x∈TnTN,x(F)<∞, for allN ∈N
, with
TN,x(F) = sup
Hx∈G(x)
(1 + Γ (Hx))N|F(Hx)|, for all function F :G0−→C.
Definition 4.1. LetF ∈ S(G0).The dual Radon d-plane transform of F is the complex-valued functionR∗F defined on Tn by
R∗F(x) =
Hx∈G(x)
F(Hx). (4.2)
Remark 4.2. Let F ∈ S(G0) and x ∈ Tn. Show that R∗F(x) has a sense. For this, it suffices to prove that the series
Hx∈G(x)F(Hx) is absolutely convergent. Since F ∈ S(G0) there exist an integer N > d×n and a positive constant C(F, N, x) =TN,x(F)<∞ such that 2
Hx∈G(x)
|F(Hx)| ≤C(F, N, x)
Hx∈G(x)
1
(1 + Γ (Hx))N.
But for all Hx∈G(x), there exists an element A(Hx) of ΛHx such that Γ (Hx) = inf
A∈ΛHx||A||2=||A(Hx)||2, since
||A||2 |A∈ΛHx
is a non empty subset of N.Then
Hx∈G(x)
|F(Hx)| ≤ C(F, N, x)
Hx∈G(x)
1 (1 + Γ (Hx))N
≤ C(F, N, x)
A∈Pd,n
1
1 +||A||2N <∞.
This proves that R∗F(x) has a sense, and therefore the functionR∗F is well-defined. The dual of the d-plane Radon transform is obtained by summing over all closed d-dimensional geodesics (d-planes) through a given point. Where F is a function belonging to S(G0). The transform R∗ is actually dual of R in the following sense
TnR∗F(x)f(x)dx=
A∈Pd,n
TnF(D(x, A))Rf(D(x, A))dx, (4.3) where dx is the canonical invariant measure on the torus Tn. Indeed, by (3.2) and (4.2)
R∗F(x) =
A∈Pd,n
F(D(x−pr ((t1, . . . , td)A), A)),
for any (t1, . . . , td)∈Rd,and the left-hand side of (4.3) can be transformed as follows
TnR∗F(x)f(x)dx =
A∈Pd,n
TnF(D(x−pr ((t1, ..., td)A), A))f(x)dx
=
A∈Pd,n
TnF(D(x, A))f(x+pr ((t1, ..., td)A))dx.
Then (4.3) follows by integration with respect tot1, . . . , td.The calculations are valid whenever for any (t1, . . . , td)∈Rd,
A∈Pd,n
Tn|F(D(x, A))f(x+ pr ((t1, . . . , td)A))|dx <∞.
Several classical inversion formulas for the Radon transform involve the function R∗Rf. However the sum defining this function does not converge here in general. We then introduce a weight function ϕ:Pd,n → ]0,+∞[ given by
ϕ(A) = exp −
inf
||QA||2 |Q∈SL∗(d,Z)
, (4.4)
and ϕi(A) = exp −
inf
||ei(QA)||2|Q∈SL∗(d,Z)
, for all A ∈ Pd,n. It is clear that ϕ(A) = ϕ(QA) for all Q ∈ SL∗ (d,Z). Since the subset
||QA||2 |Q∈SL∗(d,Z)
ofN is non empty, we have inf
||QA||2 |Q∈SL∗(d,Z)
=||Q(A)A||2, (4.5) where Q(A) is an element of SL∗(d,Z) which depends on the matrix A.
Consequently, ϕ(A) = exp
− ||Q(A)A||2
and ϕi(A) = exp
− ||ei(Q(A)A)||2
. In the sequel, we associate to each A ∈ Pd,n an element Q(A) ∈SL∗(d,Z) given by formula (4.5). It is clear that ϕ(A) = d
i=1ϕi(A).Consider the operatorR∗ϕR defined onC(Tn) as follows
R∗ϕRf(y) =
A∈Pd,n
ϕ(A)Rf(D(y, A)) (4.6)
=
A∈Pd,n
ϕ(A) 1
0 · · · 1
0 f(y+ pr ((t1, . . . , td)A))dt1· · ·dtd. It is noted that for all A∈ Pd,n andi∈ {1, . . . , d},the vector eiA∈ P, because eiA= (ai1, . . . , ain) is the ith row of the matrix A,see [3],[4]. As usual we denote the Fourier coefficients of f by
f!(k) =
Tnf(x) exp (−2iπkx)dx, withk∈Zn.
In the following, we state and prove the main result of this paper, which is the inversion theorem for the d-plane Radon transform on the torusTn. Theorem4.1. Letf be a continuous function onTn, n≥2,such that f!∈l1(Zn).Withϕas in (4.4) the d-plane Radon transform R is inverted by
f(x) =
k∈Zn
1 ψ(k)
TnR∗ϕRf(y) exp (2iπk(x−y))dy, (4.7) for anyx∈Tn,whereψdenotes the strictly positive function onZndefined by
ψ(k) =
A∈Pd,n k(e1A)=···=k(edA)=0
" d
#
i=1
ϕi(A)
$
=
A∈Pd,n, Ak=0
" d
#
i=1
ϕi(A)
$
=
A∈Pd,n, Ak=0
ϕ(A). In particular, this theorem applies to any function f ∈Cn(Tn).
P r o o f. LetT be a distribution on Tndefined by
< T, f >=
A∈Pd,n
ϕ(A) 1
0 · · · 1
0 f(pr (−((t1, . . . , td)A)))dt1· · ·dtd, whereT is depending on the functionϕdefined byϕ(A) = di=1ϕi(A) for all A∈ Pd,n, see formula (4.4). Indeed, the estimate
|< T, f >| ≤
A∈Pd,n
" d
#
i=1
ϕi(A)
$
||f||=
⎛
⎝
A∈Pd,n
ϕ(A)
⎞
⎠||f||,
where||f||= supx∈Tn|f(x)|,shows thatT is actually a distribution onTn. Then
R∗ϕRf =T∗f, convolution onTn, (4.8) since by (4.6)
R∗ϕRf(y) = < T(x), f(y−x)> .
The convolution equation (4.8) can be easily inverted by means of Fourier coefficients. From (4.8) we have
R∗ϕRf(k) =T!(k)f!(k), (4.9) with
T!(k) = < T(x),exp (−2iπkx)>
=
A∈Pd,n
ϕ(A)
⎛
⎝#d
j=1
1
0 exp (2iπtj(ejA)k)dtj
⎞
⎠
=
A∈Pd,n
#d j=1
(ϕj(A)) 1
0 exp (2iπtj(ejA)k)dtj
.
The right hand side of the last equality vanishes whenever there exists j0 ∈ {1, . . . , d} such that k·(ej0A)= 0,therefore
T!(k) =
A∈Pd,n k(e1A)=···=k(edA)=0
⎛
⎝#d
j=1
ϕj(A)
⎞
⎠ (4.10)
=
A∈Pd,n, Ak=0
⎛
⎝#d
j=1
ϕj(A)
⎞
⎠=
A∈Pd,n, Ak=0
ϕ(A) =ψ(k).
Now, given an arbitrary k = (k1, . . . , kn) ∈ Zn, we claim that ψ(k) >
0. Indeed ϕj > 0 for all j = 1, . . . , d, see the expression of ϕ given by formula (4.4). Let Pk be the set defined by
Pk = {A∈ Pd,n|k·(e1A) =· · ·=k·(edA) = 0}
= {A∈ Pd,n|Ak= 0}.
We fix some notations which will be useful in the sequel. Let be the function :Z →Zdefined by
(r) =
⎧⎨
⎩
1 if r = 0 r if r= 0.
And let δ be the function defined on Z2 by δ(α, β) =
⎧⎨
⎩
1 if α=β = 0 0 otherwise.
Now, let us show that Pk=∅ for allk∈Zn. For allk= (k1, . . . , kn)∈Zn, we seek a matrix A ∈ Pd,n depending on k such that Ak = 0. Let A =
(arl)1≤r≤d
1≤l≤n
be the matrix defined as follows
arl=
⎧⎪
⎪⎪
⎨
⎪⎪
⎪⎩
arr = kr+1
d(ε(kr+1), ε(kr))+δ(kr, kr+1) ifl=r, r= 1, . . . , d ar,r+1 =− kr
d(ε(kr+1), ε(kr))+δ(kr, kr+1) if r= 1, . . . , d, l=r+ 1
0 otherwise.
We distinguish the following cases:
(1) If ki= 0 for all i= 1, . . . , n, thenPk =Pd,n.
(2) Ifki = 0 for alli= 1, . . . , d+ 1, we have(ki) =ki andδ(ki, ki+1) = 0.In this caseaii= d(kki+1
i,ki+1), aii+1 =−d(kik,kii+1),thusaiiki−ki+1aii+1 = 0.
(3) Ifki= 0 andki+1= 0 for all i= 1, . . . , d, then(ki) = 1, (ki+1) = ki+1 and δ(ki, ki+1) = 0. In this case aii = d(1,kki+1
i+1) and aii+1 = 0 (since ki = 0), it follows thataiiki−aii+1ki+1= 0.
(4) In the simple case whereki= 0 andki+1= 0 for alli= 1, . . . , d,we have aii =aii+1 = 1 and aiiki−aii+1ki+1 = 0 (sinceki =ki+1 = 0). Then Pk is non empty for all k∈Zn.
In particular, this theorem applies to any function f ∈ Cn(Tn), it suffices to use the same technics as in [5]. And this completes the proof of
the theorem. 2
Remark 4.3. (1) LetF be a function defined onTn× Pd,nas follows F!(k, A) =
TnF(x, A) exp (−2iπkx)dx, it is clear that
Rf% (k, A) =
⎧⎨
⎩
f!(k) ifAk= 0 0 ifAk= 0.
(4.11) Indeed,
Rf% (k, A) =
Tnf(x) exp (−2iπkx)dx
⎛
⎝#d
j=1
1
0 exp (2iπk(ejA))dtj
⎞
⎠,
therefore the equality (4.11) follows.
(2) For k∈Zn, letA(k) be an element of Pd,n such that A(k)·k= 0.
By (4.11) we have f!(k) =Rf% (k, A(k)),then
Tn|f(x)|2dx=
k∈Zn
&&
& !f(k)&&&2 =
k∈Zn
&&
& %Rf(k, A(k))&&&2.
5. A range theorem
As in [5], we give a characterization of the space R(C∞(Tn)), where R is thed-plane Radon transform on the torusTn.LetF be a function on Tn× Pd,n,we writeF!(k, A) ='
TnF(x, A) exp (−2iπkx)dx.
LetEbe the space of all functionsFonTn×Pd,nsatisfying the following three conditions:
(1) For any A ∈ Pd,n,the map x → F(x, A) belongs to C∞(Tn) and for any multi-index α ∈Nn,the partial derivative ∂xαF(x, A) is uniformly bounded on Tn× Pd,n.
(2) F!(k, A) = 0 whenever k∈Zn, A∈ Pd,n and Ak= 0.
(3) F!(k, A) =F!(k, B) whenever k∈ Zn, with A, B ∈ Pd,n and Ak = Bk= 0.
Theorem 5.1. The d-plane Radon transform f → F, F(x, A) = Rf(D(x, A)),is a bijection of C∞(Tn)onto E.
P r o o f. The proof is exactly as in ([5], Theorem 4.1). 2 LetGbe the set of alld-planes of the torusTnand K be a compact of Tn.We put
GK ={D(x, A)∈G|D(x, A)∩K=∅}, with (x, A)∈Tn× Pd,n. Remark 5.1. Letf be a continuous function onTn.Then
suppf ⊂K implies that suppRf ⊂GK.
Indeed, letD(x, A) be an element ofGwhich does not belong toGK.Then D(x, A)∩K =∅,this implies that D(x, A)⊂Tn\K.It follows that
|Rf(D(x, A))| ≤
D(x,A)|f(y)|dy ≤
Tn\K|f(y)|dy= 0, since D(x, A)⊂Tn\K and suppf ⊂K.
Acknowledgements
The author is grateful to the referee for his suggestions and comments.
He also thanks professor V. Kiryakova for scientific advice. Finally, he would like to express his deep gratitude to professor F. Rouvi`ere for his suggestions and encouragements.
References
[1] A. Abouelaz, Integral geometry in the sphere Sd. Research Notes in Mathematics 422 (2001), 83–125.
[2] A. Abouelaz and R. Daher, Sur la transformation de Radon de la sph`ere Sd.Bull. Soc. Math. France 121 (1993), 353–382.
[3] A. Abouelaz and A. Ihsane, Integral geometry on discrete Grassman- nians G(d, n) associated to Zn. Integr. Transf. Spec. Funct. 21, No 3 (2010), 197–220.
[4] A. Abouelaz, E. Casadio Tarabusi and A. Ihsane, Integral geometry on discrete Grassmannians in Zn.Mediterr. J. Math. 6, No 3 (2009), 303–316.
[5] A. Abouelaz and F. Rouvi`ere, Radon transform on the torus. To appear in: Mediterranean Journal of Mathematics 9, No 3 (2012).
[6] I.M. Gelfand, M.I. Graev, and S.I. Shapiro, Differential forms and In- tegral geometry. Funct. Anal. Appl.3 (1969), 24–40.
[7] I.M. Gelfand, S.G. Gindikin and M.I. Graev,Selected Topics in Integral Geometry. Transl. of Math. Monographs, AMS, Providence, RI, 2003.
[8] S. Helgason, The Radon Transform. Progress in Mathematics 5, Birkh¨auser, Boston, Inc. MA, Second Ed., 1999.
[9] E. Ournycheva and B. Rubin, On Y. Nievergelt’s inversion formula for the Radon transform.Fract. Calc. Appl. Anal.13, No 1 (2010), 43–56.
[10] F. Rouvi`ere, Transformation aux rayonsX sur un espace sym´etrique.
Comptes Rendus Acad. Sci. Paris, Ser. I,342(2006), 1–6.
[11] B. Rubin, Radon transforms on affine Grassmannians. Trans. Amer.
Math. Soc.356 (2004), 5045–5070.
[12] B. Rubin, Notes on Radon Transforms in integral Geometry, Fract.
Calc. Appl. Anal.6, No 1 (2003), 25–72.
[13] B. Rubin, Inversion of fractional integral related to the spherical Radon transform. In: Fractional Analysis, Vol. 157, No 2 (20 August 1998), 470–487.
[14] B. Rubin, Fractional calculus and wavelet transforms in integral ge- ometry. Fract. Calc. Appl. Anal.1, No 2 (1998), 193–219.
[15] R. Strichartz, Radon inversion - variations on a theme. Amer. Math.
Monthly 89, (June-July 1982), 377–384 and 420–423.
D´epartement de Math´ematiques et Informatique Facult´e des Sciences A¨ın Chock
Route d’El Jadida, Km 8, B.P. 5366 Maˆarif 20100 Casablanca, MOROCCO
e-mail: a.abouelaz@fsac.ac.ma Received: July 7, 2010