1. Introduction
A compact convex set having interior points is called a , whose and are denoted by and , respectively. Let be the set of convex bodies in , and let be the set of convex bodies in that are symmetric about the . If A is a set in , we denote the of A as . Several classical open problems from discrete and convex geometry can be considered via studying certain functionals that are uniformly continuous on . Take Hadwiger’s covering conjecture for example. Let be the least number of translates of needed to cover K.
Conjecture 1 (Hadwiger’s covering conjecture).
For each , we have the equality holds if and only if K is a parallelotope. This conjecture has been completely resolved only for the case of
(cf. [
1]) and remains open for
. For further information on this conjecture, we refer to [
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14].
In fact,
equals the minimum number of
(sets having the form
with
and
) of
K needed to cover
K. Chuanming Zong introduced a quantitative program to tackle this conjecture via estimating
where
. It is evident that
The map
is called the
covering functional with respect to m. For each
,
is an affine invariant. More precisely,
where
is the set of non-degenerate affine transformations from
to
. It is natural to use the
(multiplicative) Banach–Mazur metric to measure the distance between two convex bodies, where, for each pair of
,
is defined as
Set
It is well known that
is a compact metric space (convex bodies that are affinely equivalent are regarded as the same element, of course).
Let
It is clear that
Since
is continuous on the compact metric space
, there exists
such that, for each
with
, the inequality
holds.
Let
be a positive number, and let
be
convex bodies in
, where
is an integer depending on
n and
. If, for each
, there exists a corresponding
satisfying
then, we call
an
ε-net for
. If, for a given
, we can construct an
ε-net
and verify that
holds for each
, then, by (
1), we have
Therefore, Zong proposed the following four steps in [
15] to attack Hadwiger’s covering conjecture:
- (1)
Obtain a good guess of by estimating for special classes of convex bodies;
- (2)
Choose a suitable ;
- (3)
Construct an ε-net of ;
- (4)
For each , verify that .
Clearly, constructing a suitable
-net
is crucial in this program. Recently, three algorithms have been designed to estimate
[
2,
3,
4].
Zong showed in [
15] that
where
c is a universal constant. When
n is large and
is small, the right-hand side is huge. Moreover, one should not expect the cardinality of an
-net to be small, since there are known lower bounds on the cardinality of such nets, see, e.g., [
16,
17].
In [
18], Shenghua Gao et al. provided a possible way of constructing
-nets for
based on finite subsets of
theoretically. However, there is still no practical algorithm for this purpose. In this paper, we propose an algorithmic method for the first time for constructing
-nets with a reasonable cardinality in the two-dimensional case. Furthermore, we construct a
-net for
consisting of 30,306 sets. This is the first concrete
-net for
.
2. A General Framework
Definition 1. Let and . Ifthen, K is called a convex lattice set
. Ifthen K is said to be a lattice set convex in the
i-th coordinate.
Clearly, if K is a convex lattice set, then, for each , K is convex at the i-th coordinate.
Proposition 1. If is a convex lattice set and H is a closed halfspace of , then is also a convex lattice set.
Proof. We only need to show that
. Actually,
□
Theorem 1. Let , , and . There exists a convex lattice set contained in such thatMoreover, if , we may require further to be symmetric with respect to o. Proof. The construction is based on the proof of Theorem 6 in [
18]. By applying a suitable affine transformation if necessary, we may assume that
Thus
, which implies that
Set
Then
For each
, there exists a point
such that
. Thus,
. It follows that
which implies that
and that (since
)
. Therefore,
Next, we show that
is a convex lattice set. Clearly,
. For each point
, by Carathéodory’s theorem, there exist
, a set of
m points
, and a set of numbers
such that
For each
, there exists a point
such that
. It follows that
Since
is convex, we have
Therefore,
Hence,
.
In the rest, we consider the case where
. Suppose that
. Then, there exists a point
such that
, which implies that
. Thus,
which shows that
. This completes the proof. □
Theorem 2 (cf. [
19]).
Let . Then, the John ellipsoid E (the unique ellipsoid of maximum volume contained in K) of K, whose center is c, satisfies Let
; by a suitable affine transformation if necessary, we may assume that the Euclidean unit ball
is the John ellipsoid of
K. From Theorem 2, we have
or, equivalently,
Thus,
which implies that
Since
, we have
where
is the set of non-degenerate linear transformations from
to
. Then
Therefore,
Hence,
It follows that
It is proved in [
20] that
Thus,
.
Corollary 1. Let , and .
- (1)
There exists a convex lattice set contained in - (2)
There exists a centrally symmetric convex lattice set contained in
Proof. We only need to prove
(1), and
(2) can be proved in a similar way. Clearly,
. By applying a suitable affine transformation if necessary, we may assume that
Let
M be the least positive integer such that
. Then,
. As in the proof of Theorem 1, there exists
such that
Moreover,
, which implies that
For each
, since
implies that
, we have
has the desired properties. □
Theoretically, to construct an -net for (or ), one needs to enumerate all convex lattice sets contained in and containing for some positive integers . Then, one needs to take the quotient set of the resulting set with respect to the affine equivalence. Each of these steps is not easy. In the following, we focus on the planar case.
3. Construction of -Nets for
Let
G be the group of transformations whose matrices are
Proposition 2. Let and be two subsets of . If there exists such that , then Proof. It is clear that, for each
,
. From Theorem 3.16 in [
5],
□
Let and be two positive integers satisfying . Let and be the closed halfplanes of bounded by the line span and containing and , respectively.
Let K be a convex lattice set contained in and containing . Then, both and are convex lattice sets. Moreover, can be viewed as , where is another convex lattice set with the desired properties.
Based on these observations, we can construct -nets for with the following steps.
Step 1. Enumerate all subsets
G of
that are convex in the 2nd coordinate and satisfy
Let
. Set
Case 1.
. In this case, the intersection of the vertical line passing through
and the convex lattice set could be ∅, or a singleton, or a set of the form
with
. In this case, we set
where
stands for the collection of all subsets of
A consisting of
i elements.
Case 2.
. In this case, the intersection of the vertical line passing through
and the convex lattice set is a set of the form
where
and
.
Hence, we set
and
We do not consider the situation
, since we take the union of the resulting set with
later.
For each member
we can recover a lattice set that is convex in the 2nd coordinate in the following way.
denotes the cardinality of a set
A.
Let .
Case 1. .
If , then we set .
If has the form , then we set .
Case 2. . If , we set ; otherwise, let be the 2nd coordinate of the point in . Similarly, if , we set ; otherwise, we let be the 2nd coordinate of the point in . Let .
The set
is a lattice set convex in the 2nd coordinate, see Algorithm 1.
Algorithm 1 Convert P to set . |
- Require:
- Ensure:
, a lattice set convex in the 2nd coordinate generated by P. - 1:
Initialize a set - 2:
fordo - 3:
▹ denotes the i-th elements of P - 4:
- 5:
if then - 6:
- 7:
else - 8:
▹ Second coordinate of the point in - 9:
end if - 10:
if then - 11:
- 12:
else - 13:
▹ Second coordinate of the point in - 14:
end if - 15:
- 16:
- 17:
end for - 18:
for do - 19:
- 20:
if then - 21:
- 22:
else if then - 23:
- 24:
- 25:
- 26:
- 27:
end if - 28:
end for - 29:
return
|
Step 2. Obtain all convex lattice sets satisfying
The case for : Let and be an empty list. For each , if is a convex lattice set, then we add to ; otherwise, we ignore P. Let . If is a convex lattice set, then we add to .
The case for : Let be a new empty list. For each , if is a convex lattice set, we add to ; otherwise, we ignore P.
For each
, Algorithm 2 is used to check whether
is a lattice set convex in the 1st coordinate. Thereafter, Algorithm 3 is used to obtain all convex lattice sets in
.
Algorithm 2 Checking convexity in the 1st coordinate for a lattice set. |
- Require:
- 1:
fordo - 2:
- 3:
if then - 4:
- 5:
- 6:
if then - 7:
return False - 8:
end if - 9:
end if - 10:
end for - 11:
return True
|
Algorithm 3 Obtain all convex lattice sets in . |
- Require:
- 1:
- 2:
fordo - 3:
▹ Algorithm 1 - 4:
if is_x_convex▹ Algorithm 2 - 5:
- 6:
if then - 7:
- 8:
continue - 9:
end if - 10:
end if - 11:
end for - 12:
return
|
Step 3. Let
be the list obtained in the last step. For each set
, remove every element
for which there exist
and
such that
to obtain a new list
, see Algorithm 4.
Algorithm 4 Removing redundant members in . |
- Require:
- 1:
- 2:
- 3:
for do - 4:
Compute the centroid of - 5:
- 6:
if then - 7:
continue - 8:
end if - 9:
- 10:
Set - 11:
- 12:
end for - 13:
return
|
Step 4. Constructing the -net.
The updated list is a desired -net.
From [
15] (Theorem A), if
holds for some positive number
, then
This implies that, if one can construct an
-net
for
with
then,
In [
21], M. Lassak proved that
holds for each planar convex body
K, and this bound can be attained.
Therefore, to confirm Hadwiger’s covering conjecture for , we need to construct an ε-net of for some .
If we take and , then, from Corollary 1, the set of all convex lattice sets contained in and containing is a -net for . In the following, we construct such a net.
For
, we have
For
, we have
and
Since
and
, we have
Thus,
. By Step 2, there are 434,849 members in
that are convex lattice sets. After applying Algorithm 4, there are 30,306 sets left. Finally, we obtain the convex hull of the Minkowski sum of these convex lattice sets and
as in Step 4. The family of these convex lattice sets is a
-net for
.
In [
4], Han et al. proposed an algorithm based on CUDA for estimating covering functionals of two-dimensional convex polytopes. On a computer equipped with an AMD Ryzen 9 3900X 12-core processor, and the NVIDIA A4000 graphics processor, and taking
for example, this algorithm takes approximately 0.217 s to obtain an estimation of
that is no greater than
. Thus, we can complete the verification of the 30,306 two-dimensional convex bodies in this
-net for
within approximately 2 h and confirm that
.
Remark 1. An ε-net for or is said to be reduced
if no two members in are affinely equivalent. Most likely, the -net we obtained is not reduced. Clearly, for each member L of this -net, is the convex hull of a convex lattice set. There are several recent results for deciding whether two convex lattice sets are affinely equivalent; see, e.g., [22]. To the best of our knowledge, there is no very efficient algorithm for this purpose yet.