MATHEMATICAL COMMUNICATIONS 331

Math. Commun., Vol. 15, No. 2, pp. 331-346 (2010)

Approximation by Jackson-type operator on the sphere

Feilong Cao
1,
and Xiaofei Guo
1
1
Department of Information and Mathematics Sciences, China Jiliang University,
Hangzhou-310 018, P. R. China
Received November 7, 2009; accepted March 20, 2010
Abstract. This paper discusses the approximation by a Jackson-type operator on the
sphere. By using a spherical translation operator, a modulus of smoothness of high or-
der, which is used to bound the rate of approximation of the Jackson-type operator, is
introduced. Furthermore, the method of multipliers is applied to characterize the satura-
tion order and saturation class of the operator. In particular, the function of saturation
class is expressed by an apparent formula. The results obtained in this paper contain the
corresponding ones of the Jackson operator.
AMS subject classications: 41A17, 41A05, 41A63
Key words: approximation, Jackson-type operator, saturation order, saturation class,
sphere
1. Introduction
A volume of work has been carried out for studying Jackson and Bernstein type
theorems on the sphere during the past decades by Lizorkin and Nikolski [13],
Ditzian [8], Dai [7], Dai and Ditzian [6], Li and Yang [12], Rustamov [16] and
many other mathematicians. In 1983, to prove direct and inverse theorems on a
more special Banach space H
r
p
(S
n1
), Lizorkin and Nikolski [13] constructed an
iterated Boolean sum of Jackson operator. They introduced a Jackson-type operator
(Q
d
k,s
f) (which will be dened in the next section), which is a linear combination
of an iterated Jackson operator and can be reduced to the known Jackson operator
when d = 1. Here we are interested in the approximation properties of (Q
d
k,s
f).
After introducing a modulus of smoothness of high order by using the translation
operator on the sphere, we will give an estimation of the rate of approximation for
the operator (Q
d
k,s
f). We also use the method of multipliers to obtain saturation
theorems of (Q
d
k,s
f), and the expression of the function of saturation class. The
obtained results show that the operator (Q
d
k,s
f) has a higher degree of approximation
and order of saturation. When d = 1, our results coincide with those of the known
Jackson operator.

The research was supported by the National Natural Science Foundation of China, No. 60873206.

Corresponding author. Email addresses: feilongcao@gmail.com (F. Cao),

feifei-198433@163.com (X. Guo)
http://www.mathos.hr/mc c 2010 Department of Mathematics, University of Osijek
332 F. Cao and X. Guo
2. Denitions and auxiliary notations
A function on the sphere S
n1
is in L
p
(S
n1
), 1 p , if
f
p
:= f
L
p
(S
n1
)
:=
__
S
n1
|f(x)|
p
d
_
1/p
< , 1 p <
and
f

:= f
L

(S
n1
)
:= ess sup
xS
n1
|f(x)|, p = ,
where d is the surface measure on the sphere S
n1
given by
S
n1
:= {x = (x
1
, x
2
, . . . , x
n
) R
n
: x
2
1
+ +x
2
n
= 1}.
Let C(S
n1
) be the space of continuous functions dened on S
n1
, which is endowed
with norm
f
C(S
n1
)
:= max
xS
n1
|f(x)|.
We denote by n the north pole of S
n1
. For a xed [0, ] and all x

:= {x : x S
n1
, ( n x) = cos }, if h(x) is unchanged over

, then it is called
a zonal function on the sphere. Clearly, h(x) is isomorphic to a univariate function
f(), 0 . Let n = (1, 0, . . . , 0), x

= (cos , sin, 0, . . . , 0), then for x

,
h(x) = h(x

) = h(cos , sin, 0, . . . , 0) = f(), [0, ]. It is obvious that the space

of all continuous zonal functions C

C(S
n1
) is a Banach space the norm of which
is
f
C

:= sup
0
|f()|.
Correspondingly, the space of all p-th Lebesgue integrable zonal functions is a sub-
space of L
p
(S
n1
), i.e. L
p

L
p
(S
n1
), 1 p < , 2 = n 2, and the norm of
this space is dened by
f
p,
:=
_

n2

n1
_

0
|f()|
p
dm

()
_1
p
,
where dm

() = sin
2
d, and
n1
means the (n 1)-dimensional surface area of
the unit sphere of R
n
. For p = , L

(S
n1
),
f
,
:= ess sup
0
|f()|.
Here and elsewhere, we always use X to denote the space L
p
(S
n1
), for 1 p
or C(S
n1
).
For f L
1
(S
n1
), the translation operator on the sphere is given by (see [1, 18])
S

(f)() :=
1

n2
sin
2

=cos
f(

)d

, 0 < < ,
where 2 = n 2. Here we integrate over the family of points

S
n1
whose
spherical distance from the given points S
n1
(i.e. the arc length between and

on the great circle passing through them) is equal to .

Approximation by Jackson-type operator on the sphere 333
Set
S
(0)

(f)() := f(), S
(j)

(f)() := S

(S
(j1)

(f)()), j = 1, 2, 3, . . . .
We introduce the sphere dierence (see [13]):

(f) :=

(f) := (S
(1)

f f)(),

(f) :=

(
r1

(f)) =
r

j=0
(1)
rj
_
r
j
_
S
(j)

f() = (S

I)
r
f(), r = 1, 2, . . .
at the point S
n1
, where I is the identity operator on L
p
(S
n1
). Then the 2r-th
modulus of smoothness of f L
p
(S
n1
) is dened by (see [16])

2r
(f, t)
p
:= sup
0<t

r

f
p
, 0 < t , r = 1, 2, . . . .
Obviously,
2r
(f, t)
p
can be rewritten as

2r
(f, t)
p
:= sup
0<t
(S

I)
r
f
p
.
Now we state some properties of spherical harmonics (see [18], [10] and [15]).
For integer k 0, the restriction to S
n1
of a homogeneous harmonic polynomial
of degree k is called a spherical harmonic of degree k. The class of all spherical
harmonics of degree k will be denoted by H
n
k
, and the class of all spherical harmonics
of degree k m will be denoted by
n
m
. Of course,
n
m
=

m
k=0
H
n
k
, and it comprises
the restriction to S
n1
of all algebraic polynomials in n variables of total degree not
exceeding m.
For f L
p
(S
n1
), the known Jackson operator on S
n1
is dened by (see [13])
J
k,s
(f, ) :=
1

n2
_
S
n1
f(

)D
k,s
(arccos

)d

, k = 1, 2, . . . ,
where k and s are positive integers,
D
k,s
() := A
1
k,s
_
sin
k
2
sin

2
_
2s
is the classical Jackson kernel, and A
k,s
is a constant connected with k and s such
that
_

0
D
k,s
() sin
2
d = 1, =
n 2
2
.
We observe that
J
k,s
(f, ) =
1

n2
_
S
n1
f(

)D
k,s
(arccos

)d

=
_

0
_
1

n2
sin
2

=cos
f(

)dS
n2
(

)
_
D
k,s
() sin
2
d
=
_

0
S

(f, )D
k,s
() sin
2
d. (1)
334 F. Cao and X. Guo
We now introduce the denition of the Jackson-type operator as follows (see
[13]).
Denition 1. Let f L
p
(S
n1
), 1 p , and let d be a positive integer. The
Jackson-type operator is given by
(Q
d
k,s
f)() := f()
_

0
D
k,s
()(I S

)
d
f() sin
n2
d. (2)
It is clear that for d = 1, the Jackson-type operator (Q
d
k,s
f)(u) coincides with
the known Jackson operator J
k,s
(f, ) and it is a spherical harmonic of total degree
not exceeding (k 1)s. But for d 2, (Q
d
k,s
f)(u) is not a spherical harmonic unless
f is not so (see [13]).
We briey introduce the projection Y
j
() by Gegenbauer polynomials {G

j
}

j=1
( =
n2
2
) for discussion of saturation property of (Q
d
k,s
f)(u).
Gegenbauer polynomials {G

j
}

j=1
are dened in terms of the generating function
(see [17]):
1
(1 2rt +r
2
)

j=0
G

j
(t)r
j
, (3)
where |r| < 1, |t| 1. For any > 0, we have (see [17])
G

1
(t) = 2t (4)
and
d
dt
G

j
(t) = 2G
+1
j1
(t). (5)
When =
n2
2
(see [18]), there holds
G

j
(t) =
(j + 2)
(j + 1)(2)
P
n
j
(t), j = 0, 1, 2 . . . ,
where P
n
j
(t) is the Legendre polynomial of degree j (see [15]).
Particularly,
G

j
(1) =
(j + 2)
(2)(j + 1)
, j = 0, 1, 2 . . . ,
therefore,
P
n
j
(t) =
G

j
(t)
G

j
(1)
. (6)
Besides, for any j = 0, 1, 2, . . . and |t| 1, |P
n
j
(t)| P
n
j
(1) = 1 (see [15]). In
addition, we use the estimate (see [1])
G

j
(cos )
C

= sup
0
|G

j
(cos )| |G

j
(1)| =
_
j + 2 1
j
_
= O(j
21
), j . (7)
Approximation by Jackson-type operator on the sphere 335
We shall rely on the classical statement that every function f() L
p
(S
n1
)
admits a unique Laplace expansion of the form
f() =

j=0
Y
j
(f; )
with respect to the spherical harmonics, where the convergence of this series is meant
in a weak sense (see [13]). Here
Y
j
(f; ) :=
()(j +)
2
+1
_
S
n1
G

j
(

)f(

)d

, j = 0, 1, 2, . . . .
It follows from (4), (5) and (6) that
lim
0
1 P
n
j
(cos )
1 P
n
1
(cos )
=
j(j + 2)
2 + 1
, j = 0, 1, 2 . . . . (8)
Next we give the denition of saturation order of operators (see [1]).
Denition 2. Let () be a positive function with respect to , 0 < < , tending
monotonely to zero as . For a sequence of operators {I

}
>0
if there exists
K L
p
(S
n1
) such that
(i) If I

(f) f
p
= o(()), then I

(f) = f;
(ii) I

(f) f
p
= O(()) if and only if f K,
then I

is said to be saturated on L
p
(S
n1
) with order O(()) and K is called its
saturation class.
In order to get the saturated class of (Q
d
k,s
f)(), we introduce a function class
H, which will be proved to be a saturated class of (Q
d
k,s
f)() later.
Denition 3. Let N
+
be a set of non-negative integers, : N
+
R, and (0) = 0.
For all j N
+
, H(X, (j)) is dened as follows:
H(X, (j)) =
_

_
{f C(S
n1
) : there exists g L

(S
n1
)
such that (j)Y
j
(f; x) = Y
j
(g; x)}
{f L
1
(S
n1
) : there exists M(S
n1
)
such that (j)Y
j
(f; x) = Y
j
(d; x)}
{f L
p
(S
n1
) : there exists g L
p
(S
n1
)
such that (j)Y
j
(f; x) = Y
j
(g; x)}
where M(S
n1
) is the collection of all Borel measures on S
n1
.
Finally, we introduce spherical convolution. In 1955 Calderon and Zygmund
(see [5, p.218]) dened the convolution on S
n1
with the help of a zonal function
in [0, ]. Later Bochner [2] made some introduction, and Hirschman, Jr. [11] gave
some detailed discussion.
336 F. Cao and X. Guo
Denition 4. Let f L
1
(S
n1
) and g L
1

, 2 = n 2. Then
h(x) =
1

n1
_
S
n1
f(y)g(x y)dy (9)
is called a spherical convolution of f and g, denoted by h = f g. And f g can be
expanded by the spherical function:
Y
n
(f g; x) =

n +
g(n)Y
n
(f; x), (10)
where g(n) is the Gegenbauer coecient (see [1, p.207]).
3. Some Lemmas
In this section, we show some lemmas as a preparation for the proof of the main
results.
Lemma 1. For 0, 2s > +n 1, 0 < , and n 3, we have
_

0

D
k,s
() sin
2
d C
,s
k

(11)
where =
n2
2
, and s, n, k are positive integers.
The proof of Lemma 1 is simple (see also [13]).
The following Lemma 2 gives the description of Q
d
k,s
f() by multiplies.
Lemma 2. For f L
p
(S
n1
), 1 p , there holds
(Q
d
k,s
f)() =

j=0

d
k,s
(j)Y
j
(f; ) (12)
where

d
k,s
(j) = 1
_

0
D
k,s
()
_
1
G

j
(cos )
G

j
(1)
_
d
sin
n2
d, j = 0, 1, 2 . . . (13)
and the convergence of the series is meant in a weak sense.
Proof. We know f() =

j=0
Y
j
(f; ). Then from Denition 1,
(Q
d
k,s
f)() =

j=0
Y
j
(f; )
_

0
D
k,s
()(I S

)
d
f() sin
n2
d.
From [13], it follows that
S

(f) =

j=0
G

j
(cos )
G

j
(1)
Y
j
(f; ).
Approximation by Jackson-type operator on the sphere 337
So
(Q
d
k,s
f)() =

j=0
_
_
1
_

0
D
k,s
()
_
1
G

j
(cos )
G

j
(1)
_
d
sin
n2
d
_
_
Y
j
(f; )
=

j=0

d
k,s
(j)Y
j
(f; ).
This nishes the proof of Lemma 2.
The next lemma is useful for determining the saturation order. It can be deduced
by the methods in [1] and [4].
Lemma 3. Suppose that {I

}
>0
is a sequence of operators on L
p
(S
n1
), and there
exists series {

(j)}

j=1
with respect to , such that
I

(f)(x) =

j=0

(j)Y
j
(f)(x)
for every f L
p
(S
n1
). If there exists () 0+ ( ) such that for any
j = 0, 1, 2, . . . ,
lim

0
1

(j)
()
=
j
= 0,
then {I

}
>0
is saturated on L
p
(S
n1
) with the order O(()) and the collection of
all constants is the invariant class for {I

}
>0
on L
p
(S
n1
).
In [9] Dunkl introduced the spherical convolution with respect to the measure
and zonal function. Normally M

is denoted by the set of Borel measures , and it

satises

:=

n2

n1
_

0
|d()| < .
Now, let M(S
n1
) be the space of all the limited regular Borel measures on S
n1
endowed with the norm

M
:=
1

n1
_
S
n1
|d(x)|.
Then it is a Banach space and every M(S
n1
) can be expanded as the following
spherical series:
S(d; x)

j=0
Y
j
(d; x),
where
Y
j
(d; x) =
()(j +)
2
+1
_
S
n1
G

j
(x y)d(y), j = 0, 1 . . . .
From the above we get Lemma 4 (see [9]).
338 F. Cao and X. Guo
Lemma 4. Let f L
p

, 1 p or C

, M(S
n1
). Then
(f d)(x) =
1

n1
_
S
n1
f(x y)d(y) (14)
is in L
p
(S
n1
) or C(S
n1
),
f d
p
f
p,

M
(15)
and
Y
j
(f d; x) =

j +

f(j)Y
j
(d; x). (16)
Moreover, if is a zonal function, so is f .
To get the saturation class, we need to discuss spherical convolution and a sin-
gular integral together (see [1]). Let f X, B

L
1

, and also
c(0, )
_

0
B

(cos ) sin
2
d = 1, (17)
c(0, )
_

0
|B

(cos )| sin
2
d M, (18)
(for > 0, it holds uniformly and also M 1)
lim

sup

|B

(cos )| = 0. (19)
Then
H

(f; x) =
1

n1
_
S
n1
f(y)B

(x y)dy (20)
is a singular integral with kernel B

( > 0), where c(0, ) =

n2
/
n1
. It is easy
to know that the operator (Q
d
k,s
f)() is a singular integral with classical Jackson
kernel D
k,s
.
Now we give the fth lemma (see [1]).
Lemma 5. Let f X. Suppose that the kernel B

of singular integral (20) satises

(17),(18) and
lim

j+

(j) 1

(
)
= (j), (21)
where the meaning of () is the same as that of Lemma 3. Then the following
statements hold:
a) If there exists g X, such that
lim

_
_
_
_
1
()
{H

f f} g
_
_
_
_
X
= 0, (22)
Approximation by Jackson-type operator on the sphere 339
then

j
Y
j
(f; x) = Y
j
(g; x), j = 0, 1, 2, . . . .
Especially, if g = 0, then f is constant.
b) If
H

f f
X
= O(()), ,
then f H(X, (j)).
To represent the functions in the saturation class (see Theorem 3), we dene the
following kernel:
k
d
(1, t) :=
_

_
1
2
_
1
0
1 r
2
r
_
1 r
2
(1 2rt +r
2
)
+1
1
_
dr, d = 1;
1
2
_
1
0
1 r
2
r
2+1
k
d1
(r, t)dr, d = 2, 3, . . .
where |r| < 1,|t| 1,
k
1
(r, t) =
1
2
_
r
0
_
1 r
2
1
(1 2r
1
t +r
2
1
)
+1
1
_
r
2
r
2
1
r
1
dr
1
and
k
d
(r, t) =
1
2
_
r
0
r
2
r
2
1
r
2+1
1
k
d1
(r
1
, t)dr
1
, d = 2, 3 . . . .
Lemma 6 gives some important properties of k
d
.
Lemma 6. Let t = cos . For >
1
2
, 1 q < (2 + 1)/(2 1), we have
k
d
(1, cos())
q,
< . (23)
Furthermore, (23) also holds for =
1
2
, 1 q < .
Proof. From (3) it follows that
1 r
2
(1 2rt +r
2
)
+1
=

j=0
j +

r
j
G

j
(t), (24)
where 1 t 1, 0 r < 1 and >
1
2
, = 0. By (24) we have

j=1
r
j+
1
j +

j
(t) =
_
1 r
2
1
(1 2r
1
t +r
2
1
)
+1
1
_
r

1
, 0 r
1
< 1, 1.
Since the left series converges uniformly, by termwise integration with respect to r
1
on [0, r] we have

j=1
r
j++1
j + + 1
j +

j
(t) =
_
r
0
_
1 r
2
1
(1 2r
1
t +r
2
1
)
+1
1
_
r

1
dr
1
.
340 F. Cao and X. Guo
We rst take = 1 and multiply r
2
, and take = 2 1 in the above formula,
respectively. Then their dierence is given by

j=1
_
1
j

1
j + 2
_
r
j+2
j +

j
(t) =
_
r
0
_
1 r
2
1
(1 2r
1
t +r
2
1
)
+1
1
_
r
2
r
2
1
r
1
dr
1
.
For 0 r < 1 and 1 t 1, we have
k
1
(r, t) =
1
2
_
r
0
_
1 r
2
1
(1 2r
1
t +r
2
1
)
+1
1
_
r
2
r
2
1
r
1
dr
1
=

j=1
1
j(j + 2)
r
j+2
j +

j
(t).
Similarly,
k
2
(r, t) =
1
2
_
r
0
r
2
r
2
1
r
2+1
1
k
1
(r
1
, t)dr
1
=

j=1
1
(j(j + 2))
2
r
j+2
j +

j
(t).
It follows from induction that
k
d
(r, t) =
1
2
_
r
0
r
2
r
2
1
r
2+1
1
k
d1
(r
1
, t)dr
1
(25)
=

j=1
1
(j(j + 2))
d
r
j+2
j +

j
(t). (26)
For 0 r 1, 1 t < 1, 1 2rt +r
2
> 0 holds. So from the integral ex-
pression of k
1
(r, t) we can attain that for any t [1, 1), lim
r1
k
1
(r, t) exists and
k
1
(1, t) = lim
r1
k
1
(r, t) is continuous for t. Furthermore, k
d
(1, t) = lim
r1
k
d
(r, t)
also holds for t [1, 1). Therefore,
k
d
(1, t) =
1
2
_
1
0
1 r
2
1
r
2+1
1
k
d1
(r
1
, t)dr
1
.
Next we prove the boundary of k
d
(1, t). From (25), k
d
(1, t) can be written as
k
d
(1, t) =
1
2
_ 1
2
0
1 r
2
1
r
2+1
1
k
d1
(r
1
, t)dr
1
+
1
2
_
1
1
2
1 r
2
1
r
2+1
1
k
d1
(r
1
, t)dr
1
:= I
1
+I
2
. (27)
Approximation by Jackson-type operator on the sphere 341
Applying (7) and (26), we have
|I
1
| =
1
2

j=1
1
(j(j + 2))
d1
j +

j
(t)

_ 1
2
0
1 r
2
1
r
2+1
1
r
j+2
1
dr
1

1
2

j=1
1
(j(j + 2))
d1
j +

_
j + 2 1
j
__
1
2
j
j

1
(j + 2)2
j+2
_
C
1
()
_
_

j=1
j
2
(j(j + 2))
d1
1
2
j
j

j=1
j
2
(j(j + 2))
d1
1
(j + 2) 2
j+2
_
_
= C
2
() < , (28)
where C
i
()(i = 1, 2, . . . ) are constants depending on . Thus I
1
is uniformly
bounded for t.
Next, we discuss I
2
. For r
1

_
1
2
, 1
_
,
k
1
(r
1
, t) =
1
2
_
r
1
0
_
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
1
_
r
2
1
r
2
2
r
2
dr
2
=
1
2
_ 1
2
0
_
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
1
_
r
2
1
r
2
2
r
2
dr
2
+
1
2
_
r
1
1
2
_
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
1
_
r
2
1
r
2
2
r
2
dr
2
:= B
1
(t) +B
2
(t).
From the above discussion, B
1
is uniformly bounded for t. Also
|B
2
(t)| =

1
2
_
r
1
1
2
_
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
1
_
r
2
1
r
2
2
r
2
dr
2

1
2
_
r
1
1
2
r
2
1
r
2
2
r
2
dr
2
+
1
2
_
r
1
1
2
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
r
2
1
r
2
2
r
2
dr
2
.
It is easy to verify that the rst integral of the righthand side is uniformly bounded.
Hence we consider the second integral on the righthand side. Let
I(t) =
1
2
_
r
1
1
2
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
r
2
1
r
2
2
r
2
dr
2
.
For 1 t <
3
4
, I(t) is bounded uniformly. For
3
4
t 1, we divide I(t) into two
parts, (1 r
2
)
2
1 t and (1 r
2
)
2
1 t, and estimate each part respectively.
Since 1 2r
2
t + r
2
2
= (1 r
2
)
2
+ 2r
2
(1 t), we have the following inequalities (see
[14]):
(1 2r
2
t +r
2
2
)
+1
(1 r
2
)
2+2
,
(1 2r
2
t +r
2
2
)
+1
(1 t)
+1
, r 1/2.
342 F. Cao and X. Guo
It follows from 1 r
2
2
= (1 r
2
)(1 +r
2
+r
2
2
+ +r
21
2
) 2(1 r
2
) that
|I(t)| =
1
2
_
_
_
1(1t)
1
2
1
2
+
_
r
1
1(1t)
1
2
_
_
1 r
2
2
(1 2r
2
t +r
2
2
)
+1
r
2
1
r
2
2
r
2
dr
2
4
_
1(1t)
1
2
1
2
(1 r
2
)
2
dr
2
+ 4(1 t)
1
_
1
1(1t)
1
2
(1 r
2
)
2
dr
2

_
C
2
+C
3
|ln(1 t)| , =
1
2
,
C
2
() +C
3
()(1 t)
+
1
2
, >
1
2
.
Thus as 0+,
|k
1
(r
1
, cos )| =
_
O(ln(sin

2
)), =
1
2
,
O((sin

2
)
2+1
), >
1
2
.
From the above estimations we can see that k
1
(r
1
, cos ) is uniformly bounded for
r
1
. Analogously, by (25), (26) and (28), we have
|k
2
(r
2
, cos )| C
4
() +
_
r
2
1
2
r
2
2
r
2
1
r
2+1
1
|k
1
(r
1
, cos )|dr
1
< .
This implies that for 1 q <
2+1
21
, and > 1/2, or 1 q < and = 1/2, there
holds
k
1
(1, cos())
q
q,
=

n2

n1
_

0
|k
1
(1, cos )|
q
sin
2
d < .
Using (25), (26) and (28), by induction we can prove |k
d
(1, cos )| < . Then by
(27), we can know that for 1 q <
2+1
21
, and > 1/2, or 1 q < and = 1/2,
there holds
k
d
(1, cos())
q
q,
=

n2

n1
_

0
|k
d
(1, cos )|
q
sin
2
d < .
4. Main results and their proofs
In this section, we rst give an estimation of the rate of approximation by the
Jackson-type operator (Q
d
k,s
f)(). Then, we prove the saturation theorem of the
Jackson-type operator on L
p
(S
n1
).
Theorem 1. Let d, s, k be positive integers, 2s > 2d + n 1, and f L
p
(S
n1
),
1 p . For the Jackson-type operator (Q
d
k,s
f)() dened above, we have
Q
d
k,s
f f
p
C
1

2d
_
f,
1
k
_
p
(29)
where the constant C
1
only depends on d and s.
Approximation by Jackson-type operator on the sphere 343
Proof. By Denition 1 we have
(Q
d
k,s
f)() f() =
_

0
(1)
d1
(S

I)
d
f()D
k,s
() sin
n2
d.
Therefore
Q
d
k,s
f f
p

_

0
(S

I)
d
f
p
D
k,s
() sin
n2
d

_

0

2d
(f, )
p
D
k,s
() sin
n2
d

_

0
(k + 1)
2d

2d
_
f,
1
k
_
p
D
k,s
() sin
n2
d

2d
_
f,
1
k
_
p
2d

j=0
_
2d
j
_
C
j,s
= C
1

2d
_
f,
1
k
_
p
.
The last inequality follows from Lemma 1, since 2s > 2d + n 1 > j + n 1,
j = 0, 1, 2, . . . , 2d. This completes the proof of Theorem 1.
Theorem 2. Let d be a positive integer, 2s > 2d+n1. Then (Q
d
k,s
f)() is saturated
on L
p
(S
n1
) with order k
2d
and the collection of constants is their invariant class.
Proof. For j = 0, 1, 2, . . . , we rst prove that
lim
k
1
d
k,s
(j)
1
d
k,s
(1)
=
_
j(j + 2)
2 + 1
_
d
. (30)
In fact, for any 0 < < , there exists such that 2s > + n 1 > 2d + n 1,
then it follows from (11) that
_

D
k,s
() sin
2
d
_

D
k,s
() sin
2
d

_

0

D
k,s
() sin
2
d
C
,s
(k)

.
Since > 2d, then
_

D
k,s
() sin
2
d = o(k
2d
), k . (31)
344 F. Cao and X. Guo
And using (11) again, we have
1
d
k,s
(1) =
_

0
D
k,s
()
_
1
G

1
(cos )
G

1
(1)
_
d
sin
2
d
= 2
d
_

0
D
k,s
() sin
2d

2
sin
2
d

_

0
D
k,s
()
2d
sin
2
d
Ck
2d
.
So we obtain
1
d
k,s
(1) = O(k
2d
). (32)
Let
T

j
(cos ) = 1
_
1
G

j
(cos )
G

j
(1)
_
d
.
We deduce from (8) that
lim
0
1 T

j
(cos )
1 T

1
(cos )
=
_
j(j + 2)
2 + 1
_
d
.
Hence for any > 0, there exists > 0 such that for 0 < <

(1 T

j
(cos ))
_
j(j + 2)
2 + 1
_
d
(1 T

1
(cos ))

(1 T

1
(cos )).
Then

1
d
k,s
(j)
_
j(j + 2)
2 + 1
_
d
(1
d
k,s
(1))

_

0
D
k,s
()
_
_
_
1
G

j
(cos )
G

j
(1)
_
d

_
j(j + 2)
2 + 1
_
d
_
1
G

1
(cos )
G

1
(1)
_
d
_
_
sin
2
d

_

0
D
k,s
()
_
1
G

1
(cos())
G

1
(1)
_
d
sin
2
d
+
_

D
k,s
()
_
1 +
_
j(j + 2)
2 + 1
_
d
_
sin
2
d
O(k
2d
) +o(k
2d
)
= o(k
2d
), k .
Therefore (30) holds and is not equal to zero.
According to Lemma 3, we nally nish the proof of Theorem 2.
Approximation by Jackson-type operator on the sphere 345
From Lemma 5, it is easy to see that the saturation class of (Q
d
k,s
f)() is
H(X, (j(j + 2))
d
).
Next we give an equivalent theorem about the saturation class.
Theorem 3. Let f X. The following statements are equivalent.
(i) f H(X, (j(j + 2))
d
).
(ii) For f L
p
(S
n1
) and g L
p
(S
n1
) (1 < p < )
f(x) = Y
0
(f; x)
1

n1
_
S
n1
k
d
(1, (x y))g(y)dy (33)
holds almost everywhere, and for f C(S
n1
) and g L

(S
n1
), (33) follows too.
For f L
1
(S
n1
) and the measure M(S
n1
)
f(x) = Y
0
(f; x)
1

n1
_
S
n1
k
d
(1, (x y))d(y) (34)
holds almost everywhere, where k
d
(1, t)(1 t 1) is given in Section 3.
Proof. We rst prove that (i) implies (ii). From (26), we obtain the Gegenbauer
coecient of k
d
(1, t):

k
d
(1, j) =
j +

1
(j(j + 2))
d
, j = 1, 2 . . . ,

k
d
(1, 0) = 0.
If f H(X, (j(j + 2))
d
), and X = L
p
(S
n1
), 1 < p < or X = C(S
n1
), then
for any j = 1, 2 . . . , we have
Y
j
(f; x) =
1
(j(j + 2))
d
Y
j
(g; x) =

j +

k
d
(1, j)Y
j
(g; x).
If f H(X, (j(j + 2))
d
) and X = L
1
(S
n1
), then we also have for j = 1, 2 . . .
Y
j
(f; x) =

j +

k
d
(1, j)Y
j
(d; x).
Therefore, by (10), Lemma 4 and the uniqueness theorem of Laplace series ex-
pansion, there hold (33) and (34).
Using (9), (10) and Lemma 4, we can deduce (i) from (ii). This completes the
proof.
Acknowledgement
The author would like to thank the referees for their helpful suggestions.
346 F. Cao and X. Guo
References
[1] H. Berens, P. L. Butzer, S. Pawelke, Limitierungsverfahren von reihen mehrdi-
mensionaler Kugelfunktionen und deren Saturationsverhalten, Publ. Res. Inst. Math.
Sci. 4(1968), 201268.
[2] S. Bochner, Positive zonal functions on spheres, Proc. Natl. Acad. Sci. USA
40(1954), 11411147.
[3] P. L. Butzer, H. Johnen, Lipschitz spaces on compact maifolds, J. Funct. Anal.
7(1971), 242266.
[4] P. L. Butzer, R. J. Nessel, W. Trebels, On summation processes of Fourier expan-
sions in Banach spaces II. Saturation theorems, Tohoku Math. J. 24(1972), 551569.
[5] A. P. Calder on, A. Zygmund, On a problem of Mihlin, Trans. Amer. Math. Soc.
78(1955), 209224.
[6] F. Dai, Z. Ditzian, Jackson inequality for Banach spaces on the sphere, Acta Math.
Hungar. 118(2008), 171195.
[7] F. Dai, Jackson-type inequality for doubling weights on the sphere, Constr. Approx.
24(2006), 91112.
[8] Z. Ditzian, Jackson-type inequality on the sphere, Acta Math. Hungar. 102(2004),
135.
[9] C. F. Dunkl, Operators and harmonic analysis on the sphere, Trans. Amer. Math.
Soc. 125(1966), 250263.
[10] W. Freeden, T. Gervens, M. Schreiner, Constructive approximation on the
sphere, Oxford University Press, New York, 1998.
[11] I. I. Hirschman, Jr., Harmonic analysis and ultraspherical polynomials, Proceedings
of the Conference on Harmonic Analysis, Cornell, 1956.
[12] L. Li, R. Yang, Approximation by Jackson polynomials on the sphere, J. Beijing Nor-
mal Univ. (Natur. Sci.) 27(1991), 112, in Chinese.
[13] I. P. Lizorkin, S. M. Nikolski, A theorem concerning approximation on the sphere,
Anal. Math. 9(1983), 207221.
[14] B. Muckenhoupt, E. M. Stein, Classical expansions and their relations to conjugate
harmonic functions, Trans. Amer. Math. Soc. 118(1965), 1792.
[15] C. M uller, Spherical harmonics, Lecture Notes in Mathematics, Springer, Berlin,
1966.
[16] K. V. Rustamov, On equivalence of dierent moduli of smoothness on the sphere,
Proc. Stekelov Inst. Math. 204(1994), 235260.
[17] E. M. Stein, G. Weiss, An introduction to Fourier analysis on Euclidean spaces,
Princeton University Press, New Jersey, 1971.
[18] K. Wang, L. Li, Harmonic analysis and approximation on the unit sphere, Science
Press, Beijing, 2006.