ON THE NULL SPACE OF THE BACKPROJECTION OPERATOR AND RUBIN’S

CONJECTURE FOR THE SPHERICAL MEAN TRANSFORM

DIVYANSH AGRAWAL∗ , GAIK AMBARTSOUMIAN† , VENKATESWARAN P. KRISHNAN∗ AND NISHA SINGHAL∗

Abstract. The spherical mean transform associates to a function f its integral averages over all spheres.
We consider the spherical mean transform for functions supported in the unit ball B in Rn for odd n,
with the centers of integration spheres restricted to the unit sphere Sn−1 . In this setup, Rubin employed
properties of Erdélyi-Kober fractional integrals and analytic continuation to re-derive the explicit inversion
formulas proved earlier by Finch, Patch, and Rakesh using wave equation techniques. As part of his work,
Rubin stated a conjecture relating spherical mean transform, its associated backprojection operator and
the Riesz potential. Furthermore, he pointed to the necessity of a detailed analysis of injectivity of the
arXiv:2406.15815v1 [math.CA] 22 Jun 2024

backprojection operator as a crucial step toward the resolution of his conjecture. This article addresses
both questions posed by Rubin by providing a characterization of the null space of the backprojection
operator, and disproving the conjecture through the construction of an explicit counterexample. Crucial
to the proofs is the range characterization for the spherical mean transform in odd dimensions derived
recently by the authors.

1. Introduction and statements of main results

The purpose of this article is to address two questions posed by Rubin in his paper [23] for the spherical
mean transform (SMT) in odd dimensions. The first is a conjecture that he suggested about a relation
between SMT and a Riesz potential, and the second question (which is related to his conjecture) is about
the injectivity of the associated backprojection operator. Our first result is that the backprojection op-
erator is not injective; in fact, we give a complete characterization of the null space of the backprojection
operator. Our second result is that the conjecture stated in [23] is not true. We show this through an
explicit counterexample.
As in Rubin’s work, our results specifically deal with the case of odd dimensions. In the recent work [4],
we derived a range characterization for SMT in odd dimensions, which is simpler than what previously
existed in the literature. Our range characterization plays a pivotal role in proving the results described
above.
The spherical mean transform maps a function to its integral averages over spheres with centers in Rn .
A formal dimension count shows that SMT depends on n+1 variables, while the function itself depends on
n variables. One can make the inversion of SMT a formally determined problem by restricting the centers
of the spheres of integration to a hypersurface. Motivated by potential applications in tomography, a
common approach is to consider this hypersurface to be Sn−1 , and functions supported in B, the unit
ball in Rn .
For f ∈ Cc∞ (B), SMT is defined as
Z
1
Mf (p, t) = f (p + tθ) dS(θ),
ωn
Sn−1

where ωn denotes the surface area of Sn−1 and dS denotes the surface measure. Note that due to the
support restriction on f , Mf (·, t) = 0 for t ≥ 2. In fact, Mf (·, t) = 0 for t close enough to 0 and 2.
Thus, M : Cc∞ (B) → Cc∞ (Sn−1 × (0, 2)).
The spherical mean transform arises naturally in several tomographic applications, including, ther-
moacoustic and photoacoustic tomography, radar and sonar, as well as ultrasound reflectivity imaging.
For this reason, the study of SMT in the context of inverse problems attracted significant attention in
the recent past, including, inversion formulas, range characterization and reconstruction algorithms; see
for instance [1–15, 17, 19–23, 25, 26].
Next, we present the explicit inversion formula for the SMT in odd dimensions derived by Finch, Patch
and Rakesh in [14], and (using a simpler argument) by Rubin in [23].

Date: June 25, 2024.

2020 Mathematics Subject Classification. 44A12, 44A15, 44A20, 45Q05, 33C10.
Key words and phrases. Spherical mean transform; backprojection operator; null space; range characterization.
1
2 AGRAWAL, AMBARTSOUMIAN, KRISHNAN, SINGHAL

Theorem 1.1. [14, Theorem 3], [23, Theorem 3.2] Let n ≥ 3 be odd, f be a smooth function supported
in B, and Mf (p, t) be known for all p ∈ Sn−1 and all t ∈ (0, 2). Then
Z


f (x) = c(n)∆ D n−3 tn−2 Mf (θ, t) dS(θ), x ∈ B
t=|x−θ|
Sn−1 (1.1)


= c(n)∆ P D n−3 tn−2 Mf (x) ,
1 d
where D = t dt , and
Z
1
(P F )(x) = F (θ, |x − θ|)dS(θ), x ∈ B.
ωn
Sn−1
The notation in the above statement are slightly different from those of [14, 23]. Furthermore, we
will not pay much attention to the constant c(n), as this does not enter into the analysis that follows.
As already mentioned, Rubin in [23] gave an alternate and much simpler proof than that of [14], using
Erdélyi-Kober fractional integrals and analytic continuation. He then stated the following conjecture.
More precisely, the necessity part was proved in [23] and was required for his proof of the above inversion
formula, while the sufficiency part was stated as a conjecture.
∞ n−1 × (0, 2)) belongs to the range of the operator f 7→ Mf iff
Conjecture 1.
n−3 n−2

[23] A function g ∈ C2c (S
P D t g belongs to the range I [Cc∞ (B)], where I 2 is the Riesz potential defined by

Z
2 Γ n−22 f (y)
(I f )(x) = dy.
4π n/2 |x − y|n−2
B
Rubin suggested that analyzing the injectivity of the backprojection operator P may be a way to
resolve his conjecture. In this paper, we characterize the null space of the backprojection operator P ,
and show that the sufficiency part of Conjecture 1 is not valid by constructing an explicit counterexample.
Here are the main results of our paper.
Theorem 1.2. A function g ∈ Cc∞ (Sn−1 ×(0, 2)) belongs to Ker(P ) if and only if g ∈ Range(D n−2 tn−2 M).
Theorem 1.3. There exists a function g ∈ Cc∞ (Sn−1 × (0, 2)) such that P (D n−3 tn−2 g) = 0, but g ∈
/
Range(M). Consequently, the sufficiency part of Conjecture 1 is not valid.
Both Theorems 1.2 and 1.3 are based on an equivalent representation of the range characterization
results of [4] in an integral form. This is the content of the next theorem. Interestingly, our proof directly
shows the equivalence of the distinct formulas stated in [4] and [15] (comment below [15, Theorem 3]) as
range conditions, that is, without invoking the fact that they are range characterizations for SMT.
Theorem 1.4. A function g ∈ Cc∞ (Sn−1 × (0, 2)) is in Range(M) if and only if P (D n−2 tn−2 g) = 0.
The proofs of these theorems are presented in the next section. We first treat the case of radial
functions, and the general case follows as a consequence with suitable modifications.
Throughout the paper we repeatedly use certain technical results, which are spelled out below for
easy reference. The first one is the Faà di Bruno formula, relating the higher order derivatives of the
composition of two functions and the derivatives of the individual functions. This is a generalization of
the usual chain rule to higher order derivatives (see, for instance, [16]).
Lemma 1.5 (Faà di Bruno formula). Let F and G be two smooth functions of a real variable. The
derivatives of the composite function F ◦ G in terms of the derivatives of F and G are expressed as
Xp
dp
F (G(t)) = F (q) (G(t))Bp,q (G(1) (t), . . . , G(p−q+1) (t)),
dtp q=1
where Bp,q are the Bell polynomials given by
X  x j1  jp−q+1
p! 1 xp−q+1
Bp,q (x1 , . . . , xp−q+1 ) = ··· ,
j1 ! . . . jp−q+1 ! 1! (p − q + 1)!
with the sum taken over all non-negative sequences, j1 , · · · , jp−q+1 such that the following two conditions
are satisfied:
j1 + j2 + · · · + jp−q+1 = q,
j1 + 2j2 + · · · + (p − q + 1)jp−q+1 = p.
 NULL SPACE OF THE BACKPROJECTION OPERATOR FOR SPHERICAL MEAN TRANSFORM 3

Since all main results of this paper use the differential operator D, it will be helpful to reformulate
the above statement in terms of D. Multiplying the standard chain rule by 1t , the D-derivative of the
composition of two functions can be re-written as
D(F (G(t))) = F ′ (G(t))DG(t),
where F ′ denotes the usual derivative of F . The following lemma is then an easy verification.

Lemma 1.6. Let F and G be smooth functions of a real variable. Then the D-derivatives of the composite
function F ◦ G satisfy the relation
p
X
D p F (G(t)) = F (q) (G(t))Bp,q ((DG)(t), · · · , D (p−q+1) G(t)).
q=1

This paper requires only the following special case of the above formula proved in [4].

Lemma 1.7 (Faà di Bruno formula - special case). Let F and G be smooth functions of a real variable
such that D j G = 0 for j ≥ 3. Then the following identity holds
p
X p!
p−q
D p F (G(t)) = p−q
F (q) (G(t)) (DG(t))2q−p D 2 G(t) .
(2q − p)!(p − q)!2
q≥p/2

Two other results employed in the article include the integration by parts formula involving the
operator D (proved by direct verification) and the Funk-Hecke theorem (e.g. see [18, 24]).

Lemma 1.8. For smooth functions F and G, the following identity holds:
Zb "k−1 #b Zb
X
k l k−l l k
∂t D F · G dt = (−1) D F · D G + (−1) ∂t F · D k G dt, (1.2)
a l=0 t=a a

where the sum is interpreted as empty for k = 0.

R1 n−3
Theorem 1.9 (Funk-Hecke). If |F (t)|(1 − t2 ) 2 dt < ∞, then for each η ∈ Sn−1 ,
−1
 
Z Z1
Sn−2 n−2 n−3
F (hσ, ηi) Yml (σ)dS(σ) = n−2
 F (t)Cm (t)(1 − t2 )
2 2 dt Yml (η),
Sn−1
Cm (1)
2
−1
n−2
where |Sn−2 | denotes the surface measure of the unit sphere in Rn−1 , Cm2 (t) are the Gegenbauer poly-
nomials, and Yml for 0 ≤ m < ∞, 1 ≤ l ≤ dm = (2m+n−2)(n+m−3)!
m!(n−2)! , d0 = 1, are the spherical harmonics.

Finally, we recall the range characterization for SMT in odd dimensions proved in [4]. It is used in
the proof of Theorem 1.4 presented in the next section.

Theorem 1.10 ([4]). Let B denote the unit ball in Rn for an odd n ≥ 3, and k := (n − 3)/2. A
function g ∈ Cc∞ (Sn−1 × (0, 2)) is representable as g = Mf for f ∈ Cc∞ (B) if and only if for each
(m, l), m ≥ 0, 0 ≤ l ≤ dm , hml (t) = tn−2 gml (t) satisfies the following two conditions:
• there is a function φml ∈ Cc∞ ((0, 2)) such that
hml (t) = D m φml (t), (1.3)
• the function φml (t) satisfies
[Lm+k φml ](1 − t) = [Lm+k φml ](1 + t), (1.4)
with
m+k
X (m + k + p)! 1 d
Lm+k = (1 − ·)m+k−p D m+k−p and D = .
(m + k − p)!p!2p t dt
p=0
4 AGRAWAL, AMBARTSOUMIAN, KRISHNAN, SINGHAL

2. Proofs
We first prove the radial version of Theorem 1.4 and then move on to the general case. Henceforth,
the letter C is used to denote dimensional constants, whose value may change from line to line in a given
computation. The exact value of the constants can be computed, but does not affect the analysis.
Theorem 2.1. Let g ∈ Cc∞ ((0, 2)). Then g ∈ Range(M) if and only if P (D n−2 tn−2 g) = 0.
Proof. Define h(t) = tn−2 g(t) and let k = n−3
2 . For every x ∈ B, one can compute
Z
P (D n−2 h)(x) = [D n−2 h](|x − θ|)dS(θ)
Sn−1
Z p 
= [D n−2 h] |x|2 + 1 − 2x · θ dS(θ)
(2.1)
Sn−1
Z1 p 
k
=C [D n−2 h] |x|2 + 1 − 2|x|t 1 − t2 dt,
−1
p
using the Funk-Hecke formula in the last step. Making the substitution u = |x|2 + 1 − 2|x|t yields
1+|x|
Z
C
k
P (D n−2 h)(x) = u[D 2k+1 h](u) 4|x|2 − (1 + |x|2 − u2 )2 du.
|x|n−2
1−|x|

Let us denote
A(x, u) = 1 + |x|2 − u2 ,
and
B(x, u) = 4|x|2 − A2 (x, u).
Observe that
A(x, 1 ± |x|) = ∓2|x|,
B(x, 1 ± |x|) = 0.
Due to the above equality, one can apply integration by parts k times without picking up any boundary
terms and obtain
1+|x|
Z
n−2 C
P (D h)(x) = u[D k+1 h](u)D k B k du. (2.2)
|x|n−2
1−|x|
We will transfer another k derivatives using integration by parts, but this time picking up boundary
terms. Before performing that computation, let us find an expression for
D k+l B k for 0 ≤ l ≤ k.
Observe that
DB = 4A,
D 2 B = −8,
and D j B = 0 for j ≥ 3.
Invoking the special case of Faà di Bruno formula (see Lemma 1.7), one gets
k
X k!(k + l)!22i
D k+l B k = (−1)k+l−i A2i−k−l B k−i .
(k − i)!(2i − k − l)!(k + l − i)!
i≥ k+l
2

Substituting this above leads to

 1+|x|
k−1 k
C X X (−1)k+l−i k!(k + l)!22i 
P (D n−2 h)(x) = 2k+1  (−1)l k−l
D h A2i−k−l B k−i 
|x| (k − i)!(2i − k − l)!(k + l − i)!
l=0 k+l
i≥ 2 1−|x|
 NULL SPACE OF THE BACKPROJECTION OPERATOR FOR SPHERICAL MEAN TRANSFORM 5

1+|x|
Z  
k C (−1)k 22k k!(2k)!
+ (−1) ∂t h dt.
|x|2k+1 k!
1−|x|

Since B(x, 1 ± |x|) = 0, only the i = k term survives in the first expression on the right to give
"k−1 #1+|x|
C X k!(k + l)!22k
P (D n−2 h)(x) = Ak−l D k−l h
|x|2k+1 (k − l)!l!
l=0 1−|x|
C 1+|x|
+ 2k+1 22k (2k)![h]1−|x| .
|x|
Writing it out, we have
" k
C X (−1)k−l 2k−l k!(k + l)! h i
n−2 k−l k−l
P (D h)(x) = |x| D h (1 + |x|)
|x|2k+1 (k − l)!l!
l=0
# (2.3)
2k−l k!(k + l)! k−l h k−l i
k
X
− |x| D h (1 − |x|) .
(k − l)!l!
l=0
In other words,
C
P (D n−2 h)(x) = ([Lk h](1 + |x|) − [Lk h](1 − |x|)) , (2.4)
|x|2k+1
where Lk is the linear differential operator of order k, defined by
k
X (k + l)!
Lk = (1 − ·)k−l D k−l .
(k − l)!l!2l
l=0

Using Theorem 1.10, we observe that g ∈ Range(M) if and only if P (D n−2 tn−2 g) = 0. This concludes
the proof of radial case. 
Next we will prove the general case, that is, Theorem 1.4. We will need the following three lemmas.
Lemma 2.2. Let U ∈ Cc∞ ((0, 2)). Then
Z2
s2j+1 U (s) ds = 0 for all 0 ≤ j ≤ n
0

if and only if U = D n+1 V for some V ∈ Cc∞ ((0, 2)).

Proof. We use induction on n. Assume n = 0. The “if” part is obvious. Let us consider the “only if”
part. Since
Z2
s U (s) ds = 0,
0
Rt
the function V (t) = s U (s) ds is compactly supported, smooth, and satisfies DV = U .
0
Next, let us assume the result is true for all 0 ≤ j ≤ n. As in the base case, the “if” part of the
induction step is straightforward, so we prove the “only if” part here. Suppose
Z2
s2j+1 U (s) = 0 for all 0 ≤ j ≤ n + 1.
0
From the induction assumption, it follows that
U = D n+1 V for some V ∈ Cc∞ ((0, 2)).
Combining the last two relations with j = n + 1 leads to
Z2
s2n+3 D n+1 V (s)ds = 0.
0
6 AGRAWAL, AMBARTSOUMIAN, KRISHNAN, SINGHAL

Applying repeated integration by parts yields

Z2
sV (s)ds = 0,
0

which implies that V = D Ve for some Ve ∈ Cc∞ ((0, 2)), i.e. U = D n+2 Ve for Ve ∈ Cc∞ ((0, 2)). 
Lemma 2.3. Let B(r, u) = 4r 2 − (1 + r2 − u2 )2 . Then for any m ≥ 1, Bm can be expressed as
2m
X
B m (r, u) = qj,2m (u2 )r 2j ,
j=0
where qj,2m is a polynomial of degree exactly 2m − j.
Proof. By regrouping the terms of B(r, u) and applying the multinomial theorem we get
X m!
B m (r, u) = (−1)i r 4i 2j r 2j (1 + u2 )j (−1)k (1 − u2 )2k
i!j!k!
i+j+k=m
X 2j m!
= (−1)i+k r 4i+2j (1 + u2 )j (1 − u2 )2k .
i!j!k!
i+j+k=m

One can rearrange the terms of the above sum in the increasing order of powers of r 2 to obtain
2m [α/2]
X X m!
B m (r, u) = (−1)m (−2)α r 2α (1 + u2 )α−2i (1 − u2 )2m+2i−2α
22i i!(α − 2i)!(m + i − α)!
α=0 i=0
2m
X X 1  m m − α + 2i
[α/2]
= (−1)m (−2)α r 2α (1 + u2 )α−2i (1 − u2 )2m+2i−2α
α=0
4i α − 2i m+i−α
i=0
2m
X
= qα,2m (u2 )r 2α ,
α=0
where
X 1  m m − α + 2i
[α/2]
2 m α
qα,2m (u ) = (−1) (−2) (1 + u2 )α−2i (1 − u2 )2m+2i−2α .
4i α − 2i m+i−α
i=0
Observe that qα,2m is a polynomial of degree 2m − α, where the coefficient of the highest degree term is
[α/2]
P 1 m
m−α+2i

(−1)m (−2)α 4i α−2i m+i−α , which is non-zero. This finishes the proof. 
i=0
Lemma 2.4. Let ε > 0 and h be a function such that
Z2−ǫ
uh(u)B k (r, u) du = 0, for some k ≥ 1 and all r ∈ (1 − ǫ, 1).
ǫ
Then,
Z2−ǫ
u2j+1 h(u) du = 0, for all 0 ≤ j ≤ 2k.
ǫ
Proof. Note that the given quantity vanishes as a polynomial in r, hence each of its coefficients must
also vanish. Therefore, using the previous lemma,
Z2−ǫ
qj,2k (u2 )uh(u)du = 0, for each 0 ≤ j ≤ 2k.
ǫ
Since qj,2k is a polynomial of degree exactly 2k − j, (starting with j = 2k and working backwards) we
have
Z2−ǫ

2k−j
u2 uh(u) du = 0, for each 0 ≤ j ≤ 2k.
ǫ
 NULL SPACE OF THE BACKPROJECTION OPERATOR FOR SPHERICAL MEAN TRANSFORM 7

The proof is complete. 

Proof of Theorem 1.4. Let us expand g in spherical harmonics:

∞ X
X dm
g(θ, t) = gml (t)Yml (θ),
m=0 l=1

with gml ∈ Cc∞ ((0, 2)). Define h(θ, t) = tn−2 g(θ, t), and hml (t) = tn−2 gml (t). Then
∞ X
X dm
n−2
[D h](θ, t) = D n−2 hml (t)Yml (θ),
m=0 l=1

1 d
where D = t dt . Applying the backprojection operator to the above expression, one gets
dm Z
∞ X
X
n−2
P (D h)(x) = [D n−2 hml ](|x − θ|)Yml (θ)dS(θ)
m=0 l=1
Sn−1
∞
XXdm Z p 
= [D n−2 hml ] |x|2 + 1 − 2x · θ Yml (θ)dS(θ)
m=0 l=1
Sn−1
( Z1 )  
∞ X
X dm p  n−2
x
n−2 2 (n−3)/2
=C [D hml ] 1 + |x|2 − 2|x|t Cm (t) 1 − t
2
dt Yml ,
|x|
m=0 l=1 −1

where in the last step, we used the Funk-Hecke theorem. Also recall that Cmα (t) are the Gegenbauer

polynomials defined by
1 d
m
m+α− 1
α
Cm (t) = Km (1 − t2 )−α+ 2 m 1 − t2 2
,
dt
where
(−1)m Γ(α + 12 )Γ(m + 2α)
Km = m .
2 m!Γ(2α)Γ(m + α + 12 )
Substituting this above, and letting |x| = r, we get,
(Z
1+r )
C X
∞ dm
X n dm
n−3 o x
n−2 2 m+ 2
P (D h)(x) = Km u[D n−2 hml ](u) 1 − t du Yml .
r m=0 dtm t= 1+r 2 −u2
2r
r
l=1 1−r

A simple application of Faá di Bruno formula gives

  !m+ n−3
2 − u2 2 o
m m 1 + r 2
dm
m+ n−3
(−r) Du 1 − = m 1 − t2 2
2 2 .
2r dt t= 1+r2r−u

Then
(Z
1+r  2 !m+ n−3 )
∞
X dm
X 1 + r 2 − u2 2 x
n−2 m−1
P (D h)(x) = C Km r u[D n−2 hml ](u)Dum 1− du Yml
2r r
m=0 l=1 1−r
(Z
1+r )
C
∞
X dm
X 1
n−3 x
m+
= Km u[D n−2 hml ](u)Dum 4r 2 − (1 + r 2 − u2 )2 2
du Yml .
r 2k+1 m=0
rm r
l=1 1−r
(2.5)
Now suppose g ∈ Range(M). Then by Theorem 1.10, hml = D m φml for φml ∈ Cc∞ ((0, 2)) and φml
satisfies (1.4). Substituting hml = D m φml into (2.5), one gets
(Z 1+r )
C X∞ Xdm
1
n−3 x
m+
P (D n−2 h)(x) = 2k+1 Km u[D m+n−2
φml ](u)D m
4r 2
− (1 + r 2
− u 2 2
) 2
du Y ml .
r rm r
m=0 l=1 1−r
8 AGRAWAL, AMBARTSOUMIAN, KRISHNAN, SINGHAL

As before, we can transfer D derivatives k times in integration by parts without picking boundary terms.
Then
(Z 1+r )
C X
∞ dm
X 1
x
n−2 m+k+1 m+k 2 2 2 2 m+k
P (D h)(x) = 2k+1 Km u[D φ ml ](u)D 4r − (1 + r − u ) du Y ml .
r rm r
m=0 l=1 1−r

The expression on the right is exactly as in (2.2) with k replaced by m + k. Hence if φml satisfies (1.4)
for each m, l, we have that P (D n−2 h)(x) = 0.
Let us now show the other implication. Suppose P (D n−2 h)(x) = 0. Then proceeding as before, we
have from (2.5),
Z1+r

m+k
u[D 2k+1 hml ](u)D m 4r 2 − (1 + r 2 − u2 )2 du = 0.
1−r
Let ǫ > 0 be such that supp(hml ) ⊂ [ε, 2 − ε]. Choosing r to be 1 − ε′ for any ε′ < ε and integrating by
parts m times, one gets
Z2

m+k
u[D m+2k+1 hml ](u) 4(1 − ε′ )2 − (1 + (1 − ε′ )2 − u2 )2 du = 0.
0
The expression on the left is a polynomial in ε′ . Since this polynomial vanishes for all ε′ in an interval,
using Lemma 2.4, we obtain:
Z2
u2j+1 D m+2k+1 hml (u) = 0 for all 0 ≤ j ≤ 2m + 2k.
0

Therefore, as a consequence of Lemma 2.2, one can conclude that D m+2k+1 hml = D 2m+2k+1 φml for some
φml ∈ Cc∞ ((0, 2)). On the space of compactly supported functions, the operator D has a zero kernel, and
therefore
hml (u) = D m φml .
Substituting this in (2.5), and integrating by parts as before, one can see that φml satisfies (1.4). Thus,
g ∈ Range(M) as a consequence of Theorem 1.10. 
Next we use Theorem 1.4 to prove Theorem 1.2. As before, we start with functions g ∈ Cc∞ (Sn−1 ×
(0, 2)) independent of the angular variable.
Theorem 2.5. A function g ∈ Cc∞ ((0, 2)) is such that g ∈ Ker(P ) if and only if g ∈ Range(D n−2 tn−2 M).
Proof. From Theorem 2.1, we know that a function ge ∈ Range(M) if and only if P (D n−2 tn−2 e g) = 0.
Therefore, g ∈ Range(D n−2 tn−2 M) implies P g = 0.
Let us show the reverse implication. Suppose g ∈ Cc∞ ((0, 2)) such that g ∈ Ker(P ). We would like to
prove that there exists an f ∈ Cc∞ (B) such that g = D n−2 tn−2 Mf .
Consider
Z
1
P g(x) = g(|x − θ|) dS(θ)
ωn
Sn−1
Z r 
1 x
= g 1 + |x|2 − 2|x|θ · dS(θ).
ωn |x|
Sn−1
Applying the Funk-Hecke theorem, and writing all the constants as C, we have
Z1 p 
n−3
P g(x) = C g 1 + |x|2 − 2|x|t 1 − t2 2 dt.
−1

The substitution 1 + |x|2 − 2|x|t = u2 yields

1+|x|
Z
C 
2 k
P g(x) = k 2k+1 ug(u) 4|x|2 − 1 + |x|2 − u2 du.
2 |x|
1−|x|
 NULL SPACE OF THE BACKPROJECTION OPERATOR FOR SPHERICAL MEAN TRANSFORM 9

Denoting |x| = r, we get

1+r
Z
C 
2 k
0 = P g(x) = k 2k+1 ug(u) 4r 2 − 1 + r 2 − u2 du. (2.6)
2 r
1−r

Next, let supp(g) ⊂ (ε, 2 − ε). Then for any ε′ < ε, letting r = 1 − ε′ , and using the structure of supp(g):
Z2 
2  k
0= ug(u) 4(1 − ε′ )2 − 1 + (1 − ε′ )2 − u2 du.
0
We then have, using Lemma 2.4,
Z2
u2j+1 g(u) = 0 for all 0 ≤ j ≤ 2k.
0
Based on Lemma 2.2, one has
g(u) = D 2k+1 h(u) for h ∈ Cc∞ ((0, 2)).
Since by assumption, P g(x) = 0, we get,
  
2k+1 2k+1 n−2 1
P (D h(t)) = P D t h(t) = 0.
tn−2
1
Note that due to the support condition on h, tn−2 h(t) ∈ Cc∞ ((0, 2)). By Theorem 2.1, we have that
1 ∞ n−2 Mf (t).
tn−2 h(t) ∈ Range(M). In other words, there exists a radial f ∈ Cc (B) such that h(t) = t
Consequently, g = D 2k+1 tn−2 Mf (t), that is g ∈ Range(D n−2 tn−2 M). This concludes the proof. 
The previous result can be generalized as follows.
Corollary 2.6. Let g ∈ Cc∞ ((0, 2)). For each 0 ≤ l ≤ 2k + 1, g ∈ Ker(P D 2k+1−l ) if and only if
g ∈ Range(D l tn−2 M).
This follows as a straightforward consequence of the above proof. One only needs to observe that D r ,
on the space of compactly supported smooth functions, has a trivial kernel. We skip the proof.
Let us now generalize the previous result to the following. The proof follows by minor modifications
of Theorems 2.5 and 1.4. Note that the case l = 2k + 1 is precisely Theorem 1.2.
Theorem 2.7. Let g ∈ Cc∞ (Sn−1 × (0, 2)). For each 0 ≤ l ≤ 2k + 1, g ∈ Ker(P D 2k+1−l ) if and only if
g ∈ Range(D l tn−2 M).
Finally, we present a counterexample to Conjecture 1.
Proof of Theorem 1.3. We construct a non-trivial g ∈ Cc∞ ((0, 2)) such that g ∈ Ker(P D n−3 tn−2 ), but g ∈
/
Range(M). This would give a counterexample to the conjecture since 0 = P (D n−3 tn−2 g) ∈ I 2 [C ∞ (B)]
(being the image of the 0 function), but g ∈ / Range(M).
Consider a non-trivial ge ∈ Cc∞ ((0, 2)), such that e g ∈ Range(M). Then, it follows from Theorem 2.1
that h = t n−2 ge ∈ Ker(P D n−2 ). Since n ≥ 3, it implies that Dh ∈ Ker(P D n−3 ). A straightforward
computation yields
 
n−4 n−3 ′ n−2 (n − 2)e
g (t) eg ′ (t)
Dh(t) = (n − 2)t g(t) + t
e g (t) = t
e + .
t2 t
Define
(n − 2)eg (t) eg ′ (t)
g(t) = + .
t2 t
Note that g(t) ∈ Cc∞ ((0, 2)) and g(t) is not the zero function. The latter claim follows from the fact that
if g(t) ≡ 0, then e
g (t) solves a homogeneous linear ODE with a zero initial condition, implying that ge ≡ 0
and contradicting our choice of e g. We now have
P (D n−3 tn−2 g(t)) = 0.
Suppose there exists an f ∈ Cc∞ (B) such that Mf = g. Then the inversion formula from [14] (see (1.1))
gives f ≡ 0, and hence g ≡ 0. This contradicts the fact that g 6≡ 0. Therefore g ∈
/ Range(M), thus
showing that the sufficiency part of Rubin’s conjecture is not valid.
10 AGRAWAL, AMBARTSOUMIAN, KRISHNAN, SINGHAL

In fact, the above argument can be generalized as follows. Consider g̃ = g + Mf for g as above and
f ∈ Cc∞ (B). Then g̃ ∈ Cc∞ (Sn−1 × (0, 2)), and P (D n−3 tn−2 g̃) = P (D n−3 tn−2 Mf ) belongs to the range
I 2 (Cc∞ (B)) (by necessity part of Conjecture 1), but g̃ ∈
/ Range(M) (for otherwise g ∈ Range(M)). 

References
[1] M. Agranovsky, P. Kuchment, and L. Kunyansky, On reconstruction formulas and algorithms for the thermoacoustic
tomography, Photoacoustic Imaging and Spectroscopy, 2017, pp. 89–102.
[2] M. Agranovsky, P. Kuchment, and E. T. Quinto, Range descriptions for the spherical mean Radon transform, J. Funct.
Anal. 248 (2007), no. 2, 344–386. MR2335579
[3] M. Agranovsky and L. V. Nguyen, Range conditions for a spherical mean transform and global extendibility of solutions
of the Darboux equation, J. Anal. Math. 112 (2010), 351–367. MR2763005
[4] D. Agrawal, G. Ambartsoumian, V. P. Krishnan, and N. Singhal, A simple range characterization for spherical mean
transform in odd dimensions and its applications, Preprint, arXiv:2310.20702 (2023).
[5] G. Ambartsoumian, R. Gouia-Zarrad, V. P. Krishnan, and S. Roy, Image reconstruction from radially incomplete
spherical Radon data, European J. Appl. Math. 29 (2018), no. 3, 470–493. MR3788452
[6] G. Ambartsoumian, R. Gouia-Zarrad, and M. A. Lewis, Inversion of the circular Radon transform on an annulus,
Inverse Problems 26 (2010), no. 10, 105015, 11. MR2719776
[7] G. Ambartsoumian and V. P. Krishnan, Inversion of a class of circular and elliptical Radon transforms, Contemporary
Mathematics 653 (2015), 1–12.
[8] G. Ambartsoumian and P. Kuchment, A range description for the planar circular Radon transform, SIAM J. Math.
Anal. 38 (2006), no. 2, 681–692.
[9] G. Ambartsoumian and L. Kunyansky, Exterior/interior problem for the circular means transform with applications to
intravascular imaging, Inverse Problems and Imaging 8 (2014), no. 2, 339–359.
[10] G. Ambartsoumian and S. K. Patch, Thermoacoustic tomography: numerical results, Proceedings of SPIE, vol. 6437,
Progress in Biomedical Optics and Imaging 8 (2007), no. 14, 346–355.
[11] Y. A. Antipov, R. Estrada, and B. Rubin, Method of analytic continuation for the inverse spherical mean transform in
constant curvature spaces, J. Anal. Math. 118 (2012), no. 2, 623–656. MR3000693
[12] R. H. Aramyan and R. M. Mnatsakanov, To recovering the moments from the spherical mean Radon transform, Journal
of Mathematical Analysis and Applications 490 (2020), no. 2, 124334.
[13] D. Finch, M. Haltmeier, and Rakesh, Inversion of spherical means and the wave equation in even dimensions, SIAM
J. Appl. Math. 68 (2007), no. 2, 392–412. MR2366991
[14] D. Finch, S. Patch, and Rakesh, Determining a function from its mean values over a family of spheres, SIAM J. Math.
Anal. 35 (2004), 1213–1240.
[15] D. Finch and Rakesh, The range of the spherical mean value operator for functions supported in a ball, Inverse Problems
22 (2006), no. 3, 923.
[16] S. G. Krantz and H. R. Parks, A primer of real analytic functions, Second, Birkhäuser Advanced Texts: Basler
Lehrbücher. [Birkhäuser Advanced Texts: Basel Textbooks], Birkhäuser Boston, Inc., Boston, MA, 2002. MR1916029
[17] L. A. Kunyansky, Explicit inversion formulae for the spherical mean Radon transform, Inverse Problems 23 (2007),
no. 1, 373–383. MR2302980
[18] F. Natterer, The mathematics of computerized tomography, Classics in Applied Mathematics, vol. 32, Society for In-
dustrial and Applied Mathematics (SIAM), Philadelphia, PA, 2001. Reprint of the 1986 original. MR1847845
[19] L. V. Nguyen, A family of inversion formulas in thermoacoustic tomography, Inverse Problems and Imaging 3 (2009),
no. 4, 649–675.
[20] L. V. Nguyen, Range description for a spherical mean transform on spaces of constant curvature, J. Anal. Math. 128
(2016), 191–214. MR3481173
[21] S. J. Norton, Reconstruction of a two-dimensional reflecting medium over a circular domain: Exact solution, The Journal
of the Acoustical Society of America 67 (1980), no. 4, 1266–1273.
[22] S. J. Norton and M. Linzer, Ultrasonic reflectivity imaging in three dimensions: exact inverse scattering solutions for
plane, cylindrical, and spherical apertures, IEEE Transactions on Biomedical Engineering 28 (1981), no. 2, 202–220.
[23] B. Rubin, Inversion formulae for the spherical mean in odd dimensions and the Euler-Poisson-Darboux equation, Inverse
Problems 24 (2008), no. 2, 025021, 10. MR2408558
[24] B. Rubin, Introduction to Radon transforms: With elements of fractional calculus and harmonic analysis, Encyclopedia
of Mathematics and its Applications, vol. 160, Cambridge University Press, New York, 2015. MR3410931
[25] Y. Salman, Recovering functions from the spherical mean transform with limited radii data by expansion into spherical
harmonics, J. Math. Anal. Appl. 465 (2018), no. 1, 331–347. MR3806707
[26] M. Xu and L. Wang, Time-domain reconstruction for thermoacoustic tomography in a spherical geometry, IEEE Trans-
actions on Medical Imaging 21 (2002), no. 7, 814–822.
∗
Centre for Applicable Mathematics, Tata Institute of Fundamental Research, Bangalore, India
E-mail: agrawald@tifrbng.res.in, vkrishnan@tifrbng.res.in, nisha2020@tifrbng.res.in
Orcid: 0009-0003-5125-0640, 0000-0002-3430-0920, 0009-0006-3005-1986
†
Department of Mathematics, The University of Texas at Arlington, Texas, USA
E-mail: gambarts@uta.edu
Orcid: 0000-0002-1462-9964