1. Introduction and Motivation
We say that a
integer matrix
is a
dilation matrix if
, that is, all the eigenvalues of
are greater than 1 in modulus. An
-
refinable function (or
distribution)
satisfies the
refinement equation:
where
is called a
refinement mask (or low-pass filter) for
. In the sequel, we assume that the refinement mask
a is finitely supported and normalized; i.e.,
and
, where by
we denote the linear space of all sequences
of complex numbers on
and by
we denote the linear space of all sequences
such that the cardinality of
is finite. It is often convenient to use the
(formal) Fourier seris of a sequence
, which is defined to be:
where
for
and
in
. In terms of Fourier series, the refinement Equation (
1) can be also stated in the frequency domain as:
where for a function
, its
Fourier transform is defined to be
, which can be naturally extended to functions in
or tempered distributions.
Let
be a mask. For a nonnegative integer
, we say that
a satisfies the
sum rules of order
with respect to a dilation matrix
(or a lattice
) if,
where
denotes the set of all polynomials in
of (total) degree at most
. Note that (
3) depends only on the lattice
; that is, if two lattices
and
generated by two dilation matrices
and
are the same
, then
a satisfies the sum rules of order
with respect to
if and only if
a satisfies the sum rules of order
with respect to
.
To find the solution for (
1) or (
2), one starts with an initial function
and employs iteratively the
subdivision scheme , where,
We say that the subdivision scheme
converges in
if there exists a function
such that:
. In such a case,
satisfies (
1). It was shown in [
1] that the subdivision scheme
converges in
if and only if,
where
is the
difference operator with
being the
jth coordinate unit vector in
,
is the
Kronecker delta satisfying
and
for all
, and
is the
subdivision operator defined to be:
In this paper, we are interested in fundamental refinable functions associated with quincunx interpolatory masks and quincunx dilation matrices in arbitrary dimensions. A function
is said to be
fundamental if it is continuous and
. A dilation matrix
is said to be a
quincunx dilation matrix if
. For example, in dimension one (
),
is the dyadic dilation factor; in dimension
, such dilation matrices include:
and in dimension
, an example of quincunx dilation matrices is:
It is easy to show that the
quincunx lattice generated by a quincunx dilation matrix
is:
which is independent of the quincunx dilation matrix
. In dimension one, the quincunx lattice is simply the lattice of even integers; while in a higher dimension, it is also called the checkerboard lattice.
One can show that if
is a fundamental refinable function associated with a refinement mask
a and a quincunx dilation matrix
as in (
1), then it is necessary that
a is a
quincunx interpolatory mask:
The subdivision scheme associated with a quincunx interpolatory mask and a quincunx dilation matrix is called a quincunx interpolatory subdivision scheme. If the quincunx interpolatory subdivision scheme converges to , then is called a quincunx fundamental refinable function.
Interpolatory subdivision schemes play a crucial role in computer-aided geometric design (CAGD) [
2], sampling theory, and wavelet/framelet analysis [
3,
4,
5,
6,
7,
8,
9,
10].
In dimension
, Deslauriers and Dubuc [
11] proposed a family of interpolatory subdivision schemes associated with a family
of quincunx interpolatory masks (with respect to the dyadic dilation factor
). Such a family is unique in the following sense:
- (1)
is a quincunx interpolatory mask: , .
- (2)
is supported on .
- (3)
satisfies the sum rules of order with respect to the dyadic dilation .
In fact, the interpolatory mask
can be explicitly written as:
It is easily seen that such a family enjoys many agreeable properties: symmetry, real-value, non-negativity (
), minimal support, and so on. Moreover, the family of Deslauriers and Dubuc’s interpolatory masks
,
, is closely related to the family of Daubechies’ orthonormal masks
in the sense that
and can be obtained by utilizing the Riesz factorization technique [
12].
On the one hand, in dimension
, a very simple extension of the dyadic dilation is to consider the dilation matrix
. For such a dilation matrix, Dynet al. [
13] constructed the so-called butterfly interpolatory subdivision scheme; Deslauriers et al. [
14] obtained several continuous fundamental refinable functions; Mongeau and Deslauriers [
15] obtained several
fundamental refinable functions; using convolutions of box splines, Riemenschneider and Shen [
16] constructed a family of interpolatory subdivision schemes with symmetry; and Han and Jia [
17] constructed a family of optimal interpolatory subdivision schemes with many desirable properties.
On the other hand, in higher dimensions (
), a more natural way of extending the dyadic dilation factor with respect to the lattice of even integers is the quincunx dilation matrix. Han and Jia in ([
18], Theorem 3.3) constructed a family of quincunx interpolatory subdivision schemes associated with a family,
of 2-dimensional unique quincunx interpolatory masks, which can be viewed as the generalization of the family of Deslauriers and Dubuc’s interpolatory masks in dimension one to dimension two in the following sense:
- (1)
is a quincunx interpolatory mask; i.e., (
6) holds with
.
- (2)
is supported on .
- (3)
satisfies the sum rules of order
with respect to the quincunx lattice
defined as in (
5) for
.
The uniqueness of such a family
implies that
is minimally supported among all the quincunx interpolatory masks which satisfies the sum rules of order
. Moreover, the mask
is real-valued and full-axis symmetric (see (
15)).
A natural question is whether the above result is still true in higher dimensions
. It should be pointed out that the results in [
18] are already highly nontrivial. Many of the results are tailored for dimension two. It was not clear whether techniques in that paper can be carried over directly to higher dimensions. In this paper, we develop new techniques that extend many results in [
18] to arbitrary dimensions and give a complete picture of the unique family of quincunx interpolatory masks in arbitrary dimension
. That is, we prove the following main theorem.
Theorem 1. There exists a family,of d-dimensional masks that are unique in the following sense: - (1)
is a quincunx interpolatory mask; i.e., (
6)
holds. - (2)
is supported on .
- (3)
satisfies the sum rules of order with respect to the quincunx lattice defined as in (
5).
In addition, is real-valued and full-axis symmetric.
Note that a -dimensional mask can be regarded as a d-dimensional mask by identifying as a subset in . The above result not only provides a natural generalization of the Deslauriers and Dubuc’s family of interpolatory masks to arbitrary dimensions, but also shows the close connection between low-dimensional and high-dimensional quincunx interpolatory masks: any such quincunx interpolatory mask in can be regarded as a quincunx interpolatory mask in for any and .
The remainder of this paper is devoted to proving the above theorem and is organized as follows. In
Section 2, we introduce some necessary lemmas and definitions in order to simply give the proof of our main result. Some properties of the unique quincunx interpolatory masks in arbitrary dimensions are discussed. In
Section 3, we present the explicit form of the unique quincunx interpolatory masks in dimension
, thereby showing the nonnegativity property of such a family in dimension two. Some remarks are given in the last section.
2. Quincunx Interpolatory Masks in Arbitrary Dimensions
In this section, we prove the main theorem, which relies essentially on multivariate polynomial interpolation. Before proceeding further, we introduce some notation and definitions.
For
and
, we define,
Elements in
are ordered lexicographically so that
is not greater than
, denoted as
, if either
or
,
for
and
for some
. We say that
if
for
. We denote,
as the ordered index subset in
for
. The cardinalities of
and
satisfy
and
. The polynomial space
can be written as
.
Let,
be a set of distinct points in
.
is said to satisfy the
node nonfiguration in if there exist
distinct lines
in
such that
lies on
, …,
lies on
, …, and
lies on
. Note that the cardinality
of
is
. A simple example of an
satisfying the node nonfiguration in
is:
For node configuration in
, we define it recursively as follows. Let,
be a set of distinct points in
.
is said to satisfy the
node configuration in if there exist
distinct hyperplanes
in
such that,
and each set of points,
considered as a set of points in
satisfies the
node configuration in . Again, a simple example of
satisfying the node configuration in
is:
The cardinality of such a in is . One can easily show that the property of node configuration in is invariant under linear transforms. That is, if is an invertible linear transform, then satisfies the node configuration in if and only if satisfies the node configuration in .
Let be a set of distinct points in . Since hyperplanes in dimension one are just points, the above definition of node configuration in for can be also defined recursively based on . For convention, we assign the empty set.
For two index subsets
and
in
, we denote by:
a matrix
of size
. The vector
is then the column vector of
indexed by
while the vector
is the row vector of
indexed by
.
We have the following result (c.f. [
19], Theorem 4) regarding the uniqueness of multivariate polynomial interpolation associated with such a set
.
Lemma 1. Suppose that satisfies the node configuration in and is a polynomial. If vanishes on , then vanishes everywhere. Consequently, the square matrix,is non-singular. Proof. We prove the result by induction on the pair .
- (1)
For
, the set
contains
distinct points in
and the matrix in (
7) is simply the Vandermonde matrix. The result obviously holds.
- (2)
Now suppose the statement holds for the pair for any dimension and .
- (3)
To prove the statement holds for the pair for any , we proceed by induction on n.
- (3.1)
The statement is obviously true for ( is a singleton).
- (3.2)
Suppose the statement holds for with .
- (3.3)
Let
be any polynomial that vanishes on
. Then by the node configuration property of
, there exists an orthogonal transform
so that for all
,
; for some
,
. That is, the orthogonal transform turns the hyperplane
containing
perpendicular to the first coordinate axis. Hence,
is a set of points lies in a hyperplane perpendicular to the first coordinate. Note that the set
has the same cardinality as
. Since the node configuration property is invariant under orthogonal transforms, we see that
satisfies the node configuration in
. Define
. Then, the polynomial
is a polynomial of
variables, has a degree at most
n, and vanishes on the set
. By the induction hypothesis in item (3.2),
vanishes everywhere. Consequently, we must have:
for some
d-variate polynomial
of the total degree at most
. Again, by our induction hypothesis in item (3.2),
vanishes everywhere. Consequently,
and thus
vanishes everywhere.
The statement for for any has been proven.
Therefore, by induction, the statement holds for any integer pair .
Next, we show that the matrix
is non-singular. Suppose this is not the case. Then there exist non-trivial coefficients
such that,
However, it implies that the non-trivial polynomial vanishes on and hence vanishes everywhere. This a contradiction. Therefore, must be non-singular. ☐
Note that , the statement in Lemma 1 also holds for any . Next, we show that for points on a rectangular grid in , they can be extended to be a set that satisfies the node configuration in .
Lemma 2. Let,with being odd. Define: Then, and can be both extended to two sets and , respectively, such that they both satisfy the node configuration in with . Consequently, the matrices,are both of full row rank. Proof. We prove the result for . The proof of the result for is similar.
First, we extend
to a larger set
. Let
and define,
Obviously, we have
. Define
hyperplanes:
for
, as well as sets of nodes
and,
Moreover,
and
. Now define,
which extends
.
We next show that
satisfies the node configuration in
. In fact, from the above definitions,
lies on the hyperplane
and does not intersect with any other hyperplane
for
and
. Moreover, for each
, one can simply use some linear transforms
so that
. In fact, define for
, the linear transforms
as,
Then, we have and for . Note that when regarded as a set in , the index set satisfies the node configuration in . Since each transform is an affine transform, it preserves the node configuration of the set in . Consequently, by the definition of node configuration in , we conclude that satisfies the node configuration in .
A similar approach can be applied to
in order to obtain
. The corresponding hyperplanes in this cases are:
for
, and
with
for
defined as in (
10).
The fact that the two matrices and are of full row rank follows from Lemma 1. We are done. ☐
The matrices and in Lemma 2 have more columns than rows. Hence, the columns of these two matrices are linearly dependent. The following lemma provides a more precise description of the linear dependence of the columns of these two matrices.
Lemma 3. Let , , and be defined as in (
8)
and (
9)
for odd . Define, Then the matrices,are both of full row rank. Proof. Note that the index set
. By Lemma 2, the matrices
,
, are both of full row rank. Consider the index set:
We next conclude the result by claiming that each of the columns for is a linear combination of columns for , , respectively.
In fact, for each
, using long division of polynomials,
with
can be represented as:
where
are polynomials of
d variables and
is a linear combination of
with
. Obviously, we have
. Now, the claim follows from
for any
. ☐
We denote the full-axis symmetry group
by:
We say that a sequence
is
full-axis symmetric if it satisfies,
One can easily show that
defined in (
9) is full-axis symmetric for
; i.e., for each
,
By considering the symmetry property of the set
, we can further reduce the dependence of columns and rows of matrices in (
12).
Lemma 4. Let , , and be defined as in (
8)
and (
9)
for odd . Define,and, Then the matrices,are square matrices and are both non-singular. Proof. First, we show that
. Recalling the definition of
in (
11), we have,
and,
For
with
, we have,
Note that:
and,
where the hyperplane
. Similarly, for
with
, we have,
In view of,
we conclude that
for
.
Next, we show that the matrices
,
, are non-singular. We use contradiction and shall prove the result for
only. The proof for
is similar. Suppose that there exists a sequence
of complex numbers in
with
for some
such that,
By the symmetry of
and that
, we have:
We extend the coefficient set
to
by defining,
where
. Now consider the linear combination
for all
. On the one hand, if
, then,
On the other hand, if
, then there exists
such that
is odd. By the symmetry of the lattice
, we have,
Consequently, the rows of are linearly dependent, in contradiction of the result in Lemma 3. Therefore, must be non-singular. We are done. ☐
Recall in Theorem 1 that with is supposed to be a mask defined on a lattice and satisfies the sum rules of order with respect to the quincunx lattice . Now we are ready to prove our main result in Theorem 1.
Proof of Theorem 1. Let
and
be the even lattice and odd lattice as defined in (
9). We first construct a quincunx interpolatory mask
supported on
.
By Lemmas 3 and 4, we can choose an index subset
so that,
and the square matrix,
is non-singular, where
are defined as in (
17), (
11), respectively. Then we can uniquely solve the following system of linear equations for
:
Note that
must be full-axis symmetric:
for all
and for all
. In fact, by
and (
20), we have for each
,
Hence, we obtain for each
,
Comparing (
20) and (
21), we see that
must be full-axis symmetric.
Then obviously
is a quincunx interpolatory mask supported on
. Note that
for all
. Together with the symmetry property of
, we conclude that
is full-axis symmetric:
The real-value property of is obvious.
Next, we show that
satisfies the sum rules of order
with respect to the quincunx lattice
, which is equivalent to:
Note that by our definition of
, we have,
Moreover, by our construction, we already have,
for all
. Hence, we only need to show that,
We prove it by considering,
as the union of two index sets. For
, there must exist
odd in view of
. By the full-axis symmetry property of
and similar to (
19), we must have,
. For
, by (
13), we have,
which must be 0 since
is a linear combination of
with
. Hence (
22) holds.
Consequently, is a quincunx interpolatory mask supported on and satisfying the sum rules of order with respect to the quincunx lattice .
We finally show the uniqueness of
. Suppose there is another quincunx interpolatory mask
supported on
and satisfying the sum rules of order
with respect to the quincunx lattice
. Then, we have,
By the non-singularity of , we must have for all . Consequently, . We are done. ☐
In view of the proof of Theorem 1, we can have a more generalized result as follows.
Theorem 2. Let , , and be defined as in (
8)
and (
9)
for odd , and , be defined as in (
17), (
11),
respectively. Suppose the sequences of complex numbers in satisfy, Then the system of linear equations,has a unique solution for . Moreover, is full-axis symmetric for . In particular, if , , are of real value, then , are of real value for . Proof. By Lemmas 3 and 4, the matrices
,
, are both of full row rank. Choose any index set
so that
and the square matrix
is non-singular, respectively, for
. Then the following system of linear equations:
has a unique solution
for
. Similar to the proof of Theorem 1, such a sequence
is full-axis symmetric, which implies,
In particular, if are of real value, then obviously the solutions are of real value. The proof of uniqueness is similar to the proof of uniqueness in Theorem 1. ☐
3. Explicit Form of the Bivariate Quincunx Interpolatory Masks
In dimension
, the unique quincunx interpolatory mask
with
is the Deslaruier–Dubuc interpolatory mask, which has the explicit form ([
20,
21]):
In dimension
, Han and Jia [
1] show that the quincunx interpolatory masks
in Theorem 1 for
and
are given by:
In this paper, we give the explicit form of the full family of Theorem 1 in dimension .
Theorem 3. Let satisfy . Then the unique quincunx interpolatory mask in Theorem 1 with is explicitly given as follows:
- (1)
If , then, - (2)
If , then, - (3)
If with odd and , then,with being determined by items (1)
or (2).
Moreover, we have for all .
Before we proceed to the proof of Theorem 3, let us introduce some notation and lemmas to simplify our proof. Define,
and,
Then, by letting
and
, the masks
in (
24) and (
25) can be written as:
First, consider the sum rule definition in (
3). In the case of the quincunx dilation matrix in dimension
, a mask
a satisfies the sum rules of order
with respect to a quincunx dilation matrix
(
), equivalent to:
The following lemma shows that our masks
in (
28) indeed satisfy the sum rules of order at least
with respect to the quincunx dilation matrix.
Lemma 5. Let be defined as in (28). Define: It follows that,that is, satisfies the sum rules of order . Proof. Note that
is the linear combination of terms of the form:
with
for
and
for
. Using the Leibniz rule, one can easily check that,
Hence (
30) holds. That is, at
,
is a polynomial of the form:
Thus, at
, we have,
We are done. ☐
Second, consider the interpolatory condition in (
6). In dimension
, the mask
a being a quincunx interpolatory mask is equivalent to:
We have the following lemma concerning .
Lemma 6. Let be defined as in (
28).
Then the function:is independent of the variable . Proof. It is easy to show that for
, we have:
and for
, we have:
with
. Hence, we only need to show that the right-hand sides of (
33) and (
34) are independent of the variable
y. We next prove (
33). The proof for (
34) is analogous.
One can show that,
where the
function is defined by
and we have,
Recalling the definition of
in (
27), the right-hand side of (
33) is the linear combination of terms of the form:
By changing the order of
j and
ℓ in the above, we obtain,
which is a constant independent of
y. We are done. ☐
Now, we are ready to prove Theorem 3.
Proof of Theorem 3. Obviously, by the definition, is nonnegative. We will show that with satisfies the following properties:
- (1)
is a quincunx interpolatory mask.
- (2)
is supported on .
- (3)
satisfies the sum rules of order with respect to the quincunx lattice .
Item (2) can be checked directly. We first prove item (1). By Lemma 6, we see that,
are both independent of
y. Consequently, by (
32), (
33), and (
34), we see that
is interpolatory is equivalent to:
for any integer pair
with
. We prove (
35) by induction on
n.
It is straightforward to show that (
35) holds when
(
must be 0) for both
and
, since it reduces to the case
for 1-dimensional interpolatory mask
and the case
in [
18].
Suppose that (
35) holds for
with any
and
satisfying
. We next show that (
35) holds for
.
Note that by the independency of
and
with respect to
y, we have (by substituting
),
Using integration by parts, it is easy to show that,
satisfy,
Consequently, by our induction hypothesis, we have,
and,
Therefore, by induction, we conclude that (
35) holds for any integer pair
with
and hence item (1) holds.
It remains to show that item (3) holds. That is
satisfies the sum rules of order
. By Lemma 5, we see that
satisfies the sum rules of order
; that is,
Noting that
is full-axis symmetric, we have
for any
odd. Together with that
is interpolatory, we conclude that,
We are done. ☐