1. Introduction
A semiring is a generalization of a ring, where it is not required that each element has an additive inverse. The definition of a semiring varies throughout the literature. In this paper, we use the definition of a semiring in [
1]. The most common example of a semiring is the set of natural numbers, i.e., the set
. Semirings were formally introduced by Vandiver in 1934 (see [
2]) but made its first implicit appearance in the work of authors like Dedekind forty years earlier.
Semirings have been studied extensively over the past few decades, either in an attempt to extend the theory of semigroups and rings, or, to broaden its applications in mathematics, computer science, and data analysis; for example, see [
3,
4,
5]. Bourne [
6] was the first to investigate semirings in topology, and Ahsan, Saifullah, and Khan [
7] initiated the study of semirings in a fuzzy context. The theory of semirings has developed across different branches of pure mathematics, including graph theory, geometry, topology, and linear algebra; for example, see [
8,
9,
10,
11]. Semirings are also used in various applications in computer science; for example, see [
12,
13].
The properties of ideals in semirings do not necessarily coincide with those in rings. For example, it is well known that the quotient structure of a ring modulo an ideal partitions the ring. However, this is not necessarily the case for semirings, making the task of generalizing ring theorems to semirings a non-trivial matter. Authors like Bourne, Henriksen, and LaTorre used equivalence relations to determine quotient structures of semirings using ideals (see [
14,
15,
16]). Using this approach, LaTorre [
16] derived multiple analogs of ring theorems for a large class of semirings. Bourne [
14] and LaTorre [
16] attempted to derive a precise analog of the Fundamental Theorem of Homomorphisms but encountered some difficulty.
Allen [
17] remedied this by defining a special class of ideals, namely partitioning ideals (also called
Q-ideals). A
Q-ideal of a semiring
S is an ideal
I of
S for which there exists a subset
such that
partitions
S. This notion allowed Allen to generalize the Fundamental Theorem of Homomorphisms for a large class of semirings.
The study of
Q-ideals continued. Chaudhari and Ingale [
18] introduced a partitioning ideal for ternary semirings. Sharma [
19] continued the investigation of the partitioning ideals of skew group semirings. Chaudhari, Nemade, and Davvaz [
20] extended some theory of partitioning ideals to
semirings, and could thus generalize the Fundamental Theorem of Homomorphisms to include
semirings.
Partitioning elements of sets according to their properties is central to mathematics and many applications thereof. For example, in computer science, the
k-means clustering method partitions Boolean vectors into
k groups. The method works by iteratively assigning data points to the closest cluster centroid and then updating the centroids based on the assigned points. This partitioning method has wide applications in biology, marketing, medicine, and psychology (see [
21]). In mathematics, a method called matroid partitioning partitions a set of elements into matroids, combinatorial structures that generalize the concept of linear independence in graphs (for example, see [
22]). The origin of the topic of this paper lies in the question of how best to partition site-by-species matrices in ecology. This is related to the partitioning of Boolean vectors, which is prominent in computer science and mathematics.
Boolean rings are closely related to Boolean algebras, in that every Boolean ring can be interpreted as a Boolean algebra by treating its operators as logical operators. Conversely, every Boolean algebra can be interpreted as a Boolean ring by treating its operators ∨ and ∧ as addition and multiplication, respectively. These structures are thus mathematically equivalent, since they describe the same set of elements and have operators that behave in a similar way. Consequently, there is a close connection between ideals of Boolean algebras and ideals of Boolean rings, in that an ideal of a Boolean algebra can be mapped to an ideal of the corresponding Boolean ring.
In this paper, we study the ideals of the Boolean semiring
, where
is the product of the two-element Boolean algebra
with itself
s times. We notice that, since
is also a Boolean algebra, studying its ideals means studying the ideals of its corresponding Boolean ring, a well-known topic. We choose to study
as a semiring and develop partitions thereof using the language of
Q-ideals, thus extending the work of Allen in [
17].
In
Section 2, we give the preliminary material needed for this paper. We construct the Boolean semiring
, the set of all
s-tuples over a particular Boolean algebra. In
Section 3, we propose an alternative construct for the partitioning of
using the ideal
(see Theorem 1) and call this the
Q-ideal partition. We prove that the cells of a
Q-ideal partition have equal cardinality and express this cardinality in terms of the number of non-zero elements of its initial Boolean vector, called the initial seed (Lemma 3). We prove that each seed induces a unique
Q-ideal partition (Lemma 4). Using this novel
Q-ideal, we develop a hierarchical partitioning algorithm (Algorithm 1) and prove that it is nested (Theorem 2). We determine the number of steps after which this algorithm terminates, as well as the number of cells into which
is partitioned at each step of the algorithm (Lemma 5).
As Allen introduced maximal semiring homomorphisms to construct
Q-ideals that partition semirings (see [
1]), we further investigate the relationship between these concepts. While his result is insightful, the method for identifying such homomorphisms remains unclear. To connect the notions of maximal semiring homomorphisms and
Q-ideal partitions, we draw inspiration from one of Allen’s examples to determine the maximal semiring homomorphism corresponding to the
Q-ideal partition (see Lemma 7).
2. Preliminary Material
We start with the definition of a special type of algebra.
Definition 1 ([
23])
. A Boolean algebra
is a tuple , where B is a non-empty set, and + and · are two binary operations on B, such that- (a)
Both + and · are commutative;
- (b)
Each operation is distributive over the other;
- (c)
For all , we have and ;
- (d)
For all , there exists , called the complement of b, such that and .
A
Boolean matrix is a matrix over a Boolean algebra [
23]. The origin of the topic of this paper lies in the question of how best to partition the rows of a subclass of Boolean matrices, called site-by-species matrices. Below, we give the well-known definition of such a matrix.
Definition 2. A site-by-species matrix
of size is a Boolean matrixwith rows representing sites and columns representing species. For each i and j, we have if species is detected at site ; otherwise, .
The following example demonstrates Definition 2.
Example 1. Consider the site-by-species matrixwhere , , and represent three distinct ponds in the Western Cape, and , , and represent three distinct fish species. Since , species is detected at site , and since , species is not detected at site .
In order to achieve this objective, we express the rows of such a matrix as Boolean vectors
where, for each
i, the element
belongs to the well-known two-element Boolean algebra
. Thus, the problem of partitioning the rows of a site-by-species matrix using algebraic tools reduces to partitioning a set of Boolean vectors over
.
The following result is well known.
Lemma 1. If is an algebraic structure, is a partition of A and , then is a partition of X, where for each .
Proof. Let be an algebraic structure, and let be a partition of A. Let , and for each , define . We show that is a partition of X, that is, (1) and (2) for each .
- (1)
If , then since , we have . Since , it follows that for some . Hence, for some , so that for some . By the definition of the set union, we deduce that . Since x was arbitrary, we conclude that . Conversely, if , then for some . Since for each , it follows that . Since x was arbitrary, we conclude that . Therefore, .
- (2)
Suppose with . Suppose, for contradiction, that . It follows by construction that ; thus, is a contradiction, since is a partition of A. From this, we deduce for each .
This completes the proof. □
Our objective is to find partitions of the algebraic structure where the elements are all possible rows over (also termed in ecology, the set of all possible realizations), and then, by applying Lemma 1, to find partitions of the rows of a site-by-species matrix.
Definition 3 ([
23])
. Let denote the set of s-tuples over , called seeds
of length s, where the i-th entry of is denoted by . The weight
of , denoted by , is the number of non-zero entries in , while the complement
of , denoted by , is the distinct element in such that if and only if . We define two binary operators + and · on by making use of the operators on componentwise. That is,andfor all in
.
Notation 1. We denote the seeds and by and , respectively.
It is clear that is a product of with itself s times. In the next result, we use Definition 1 in order to rigorously prove that is also a Boolean algebra.
Proposition 1. is a Boolean algebra.
Proof. Let .
- (a)
If + and · are commutative in
, then
and
- (b)
If + and · are distributive over each other in
, then
and
- (c)
Since
and
for all
v in
, it follows that
and
- (d)
Since, for each
in
, we have that
and
for all
, it follows that
and
It follows by Definition 1 that is a Boolean algebra. □
The partitioning of Boolean algebras using ideals is a well-known topic. We make use of a different approach by viewing as a Boolean semiring.
Definition 4 ([
1])
. A semiring
is a tuple , where S is a non-empty set, and + and · are two associative binary operators on S, such that- (i)
+ is commutative;
- (ii)
· distributes over + both from the left and from the right;
- (iii)
There exists such that and for each .
A semiring is commutative
if the multiplication operator · is commutative. Furthermore, a semiring S is Boolean
if for all s in S, the conditionholds.
There is a distinction between a Boolean semiring and an idempotent semiring. An idempotent semiring is a semiring in which the addition operation has the idempotent property; that is, for each s in S, we have . Since the additive operation in a Boolean semiring B corresponds to set union, and for every set A, it follows that for each b in B, we have . This shows that every Boolean semiring is an idempotent semiring, but the converse is not necessarily true. That is, not every idempotent semiring is a Boolean semiring. This is because the multiplication operation in an idempotent semiring need not have the idempotency property.
We give an example of an idempotent semiring that is not a Boolean semiring.
Example 2. Consider the set of positive real numbers (including zero) equipped with two binary operators: and , where × is the usual multiplication. We verify that is a semiring.
- (a)
For each , we have .
- (b)
For each , we have Suppose w.l.o.g. that , then and , so that A similar method can be used to show that .
- (c)
For each , we have and .
It follows by Definition 4 that is a semiring. Moreover, since for each r in , it follows that is an idempotent semiring. However, since × is not idempotent (for example, ), is not a Boolean semiring.
We give an example of a Boolean semiring.
Example 3 ([
1])
. Let be a well-ordered set where and for all . It follows that is a semiring. To see why S is Boolean, notice that for every a in S, we haveSince S is a Boolean semiring, it is also an idempotent semiring.
Notation 2. From here onwards, to indicate the multiplication of two vectors and , we will write instead of , and for elements a and b that are not vectors, we will write instead of . In cases where two numbers are multiplied, for example, 0 and 1, we will write instead of 01 to avoid confusion.
In ring theory, ideals are significant. The same notion exists for semirings.
Definition 5 ([
1])
. An ideal
in a commutative semiring is a subset such that for any and , we have and .
It is well known that for any ring R and ideal I of R, the quotient of R with respect to I is a partition of R. We give an example to show that this property cannot be generalized to include semirings.
Example 4. Consider the ideal in the semiring . We haveandso that Therefore, does not partition .
A special class of ideals was introduced by Allen to partition a semiring.
Definition 6 ([
1])
. An ideal I of a semiring is a Q-ideal
if there exists a subset , such that the following conditions hold:- (i)
;
- (ii)
For all , we have if and only if .
We give examples of Q-ideals.
Example 5. Consider the set of natural numbers with usual addition and multiplication. Let and . It can be argued that, since is the set of all even natural numbers (including zero), and is the set of all odd natural numbers, we have and . Therefore, is a Q-ideal of .
Example 6 ([
1])
. Let be a well-ordered set where and for all . Fix . Let and . Then, is a Q-ideal of S.
Since is a semiring, partitioning requires constructing Q-ideals of . We adopt the same construction used by Allen in Example 6.
We start by defining the following relation on .
Definition 7 ([
23])
. For any , we say that is related
to and write , if for all , we have that implies that . We say that is unrelated
to , and write , if there exists an such that and .
Proposition 2. The following holds:
- (a)
For all , we have that .
- (b)
For all , we have that if and only if .
- (c)
is an ordered set with the least element and the largest element .
- (d)
For all , we have that is an ideal in .
Proof. - (a)
This follows since, for all , if , then .
- (b)
For any , suppose that , and let be arbitrary. If , then , so that (since ), and thus . Since i was arbitrary, we have that . Hence, whenever . Conversely, suppose that . If , then , so that (since ), and thus . Since i was arbitrary, we have that . Hence, whenever .
- (c)
Suppose and that .
- (i)
If , then . Hence, , so that ≤ is reflexive.
- (ii)
Suppose that and that . If , then (since ), and thus (since ). Hence, , so that ≤ is transitive.
- (iii)
Suppose that and that . If , then (since ). If , then (since ). Hence, for all , so that . Therefore, ≤ is antisymmetric.
It follows from (i), (ii), and (iii) that ≤ is an order relation on , so that is an ordered set. To see why is the least element in , notice that whenever . Hence, . To see why is the largest element of , notice that whenever . Hence, .
- (d)
Suppose and . Let be arbitrary. If , then either or (or both). Suppose w.l.o.g. that , then (since ). Since i was arbitrary, it follows that . Hence, . If , then (and ), so that (since ). Since i was arbitrary, it follows that . Hence, . It follows that is an ideal in .
□
If we were to use the same Q as Allen did in Example 6, then need not be a Q-ideal of . We illustrate this with the following example:
Example 7. If , thenand Since Definition 6(ii) is not satisfied, is not a Q-ideal of .
In the next result, we show under which conditions will be a Q-ideal of for Q as defined in Example 6.
Lemma 2. Let . It follows that is a Q-ideal of , where , if and only if .
Proof. Suppose . If , then Moreover, Definition 6(ii) is satisfied since for all , we have if and only if . If , then Moreover, Definition 6(ii) is satisfied, since . Therefore, is a Q-ideal of whenever .
Conversely, suppose . Since and , it follows that Since (by Proposition 2(a)) and (since is the largest element and ), we have that . Moreover, since , Definition 6(ii) is not satisfied. Hence, is not a Q-ideal in whenever .
This concludes the proof. □
If we plan to use the ideal for partitioning , an alternative construction of Q is needed. We address this in the next section.
3. Constructing -Ideals Using Seeds
Returning to Example 6, we notice that if Q was restricted to then would be a Q-ideal of . This is used as motivation for the new choice of Q constructed in this section. Before proposing this different choice of Q, we express the cardinality of in terms of the weight of .
Lemma 3. Let . For all , we have that Proof. If = 0, then the result follows trivially. Suppose . Define a map by for all . We verify that is bijective.
It follows by construction that
is surjective. To show that
is injective, suppose that
for some
. Then, for all
, we have that
We need to prove that Suppose by way of contradiction that . Then, for some i ∈ {1, …, s}, we have that gi ≠ hi Consider this i-th entry.
Case 1. If
, then
, so that
, contradicting (
2).
Case 2. If
and
, then
, so that
, contradicting (
2).
Case 3. If and , then, since , we have that , and, since , we have that , which implies that . Hence, , a contradiction.
We do not consider the case where , since if , then, since , we have that , and, since , we have that . Hence, , a contradiction.
We deduce that . Therefore, ϕ is injective. Hence, ϕ is bijective, so that
Suppose that , where , and w.l.o.g. that . It follows that the in for which are those for which and for all . Clearly, there are such . If , then , so that , and thus .
This concludes the proof. □
We are now in a position to prove one of our main results. The partition constructed will be called the Q-ideal partition.
Theorem 1. Let . It follows that is a Q-ideal if and only if .
Proof. Let . Suppose that is a Q-ideal in and by way of contradiction that . Then, either there exists a in Q such that , or there exists an in such that (or both).
Case 1. Suppose there exists a
in
Q such that
. Since
, we have that
. Consider
. We notice that
and
(since for all
, if
, then
), and
(since
and
). Thus, we have that
, where
and
. Hence,
, contradicting (
1).
Case 2. If there exists an in such that , then for no in Q, so that for any in Q. Hence, , so that , a contradiction.
From this, we deduce that .
Conversely, suppose that . It is either the case that , or that .
Case 1. If
, then
(since
), so that
, and,
. Since
, Definition 6(i) is satisfied. Also, Definition 6(ii) is satisfied since
. Hence,
is a
Q-ideal of
whenever
.
Case 2. If
, suppose that
, where
, and w.l.o.g. that
We will now verify two properties of what it means for the cosets of a Q-ideal to partition , namely that (i) their unions equals and (ii) they are pairwise disjoint.
- (i)
Their union equals :
It is trivial that . To verify that , suppose that . It is either the case that or that .
Subcase 1. If , then, since and , we have that .
Subcase 2. If , then there exist at least one such that and . Suppose w.l.o.g. that this is true for , where .
For each
, let
Let
. We verify that
. Let
be arbitrary. If
, then by (
4), we have that
, i.e.,
. It follows by (
3) that
, so that
. Since
i was arbitrary, we have that
.
For each
, let
Let
. We verify that
. Let
be arbitrary. If
, then by (
5), we have that
and
, so that
. Since, by assumption,
for all
, and
, we have that
. Hence,
. By (
3), we thus have that
. Since
i was arbitrary, it follows that
. Moreover, we have for each
that
Hence, . Therefore, .
Since was arbitrary, it follows that . Hence, .
- (ii)
They are pairwise disjoint:
Suppose that
. By way of contradiction, suppose that
. Then,
which contradicts the fact that
. We deduce that
.
Conversely, suppose that . By way of contradiction, suppose that . Then, there exists an in such that and . Let .
Subcase 1. If
and
, then
and
. Thus, there exists a
such that
, i.e., for all
, if
, then
(a). Since
(by (
3)), we have that
(since
) and
, so that
, contradicting (a).
Subcase 2. If
and
for some
, then
and
for some
. Thus, there exists a
such that
, i.e., for all
, if
, then
. In particular, for all
, if
, then
(b). Suppose that
and that
. Then,
(by (
3)) and
. Since
and
, it follows that
for some
, contradicting (b).
Subcase 3. If and for some , then and for some . Thus, there exist such that , i.e., for all i ∈ {1,…, s}, we have that + gi = bi + hi (c). Since vi = 0 for all i ∈ {1, …, j} and g, h ∈ Iv, it follows that gi = hi = 0 for all i ∈ {1, …, j}. Also, we have that bi = 0 for an i ∈ {1, …, j} (since b ≠ v′). Consider this i-th entry. We have that = 1, so that + gi = 1 + 0 = 1 ≠ 0 = 0 + 0 = bi + hi, contradicting (c).
Subcase 4. If and for some , then and for some . Thus, there exist such that , i.e., for all i ∈ {1,…, s}, we have that bi + gi = + hi (d). Since vi = 0 for all i ∈ {1, …, j} and g,Iv, it follows that gi = hi = 0 for all i ∈ {1, …, j}. Also, since b ≠ v′, we
have that bi = 0 for some i ∈ {1, …, j}. Consider this i-th entry for which bi = 0 and = 1. It follows that bi + gi = 0 + 0 = 0 ≠ 1 = 1 + 0 = + hi, contradicting (d).
We deduce that whenever .
It follows by Definition 6 that is a Q-ideal of whenever .
This concludes the proof. □
We illustrate Theorem 1 with the following example.
Example 8. Let .
- (a)
If , thenis a Q-ideal, because is a partition of .
- (b)
If , then is not a Q-ideal, because is not a partition of .
The following result assures us that each seed induces a unique partition.
Lemma 4. If , then if and only if .
Proof. Suppose . If , then , and, , so that . This implies that for all . Hence, .
Conversely, if , then for all in there exists a unique in such that . That is, if , then there exists a unique in such that . We show that .
Suppose by way of contradiction that . Then, , a contradiction, since . Therefore, we deduce that . Hence, . Thus, we have that . This concludes the proof. □
Definition 8. If is such that is a Q-ideal partition, then we call the initial seed of this Q-ideal partition.
In light of Theorem 1, we construct the following hierarchical partitioning algorithm (Algorithm 1).
Algorithm 1 Algorithm QP(seed). |
Step 1. Let seed and .
|
Step 2. Compute .
|
Step 3. Compute .
|
Step 4. Compute .
|
Step 5. Return and as a partition of .
|
Step 6. Choose in with .
|
Step 7. If , terminate; else, let seed and go to Step 1.
|
Example 9. Consider with seed . We have the following steps for demonstration:
Step d1. Let and .
Step d2. Thus, .
Step d3. We have that .
Step d4. Thus, .
Step d5. We have that , andis a partition of .
Step d6. Choose in with .
Step d7. Let and go to Step 1.
Step d8. Thus, and .
Step d9. Thus, .
Step d10. We have that .
Step d11. Thus, .
Step d12. We have that , andis a partition of .
Step d13. Choose in with .
Step d14. Since , terminate the algorithm.
We take note of the following:
Remark 1. Consider Example 9.
The hierarchical partitioning of , according to Algorithm 1, terminated at 14 steps.
The semiring is partitioned into
- -
cells of equal cardinality
at Step d5, where
is the chosen seed at Step d1;
- -
cells of equal cardinality at Step d12, where is the chosen seed at Step d6.
We use this observation to deduce the following result:
Lemma 5. In , Algorithm 1 will terminate at steps. Let . If is the chosen seed at Step j, then , and is partitioned into cells each of equal cardinality .
Proof. The conclusion follows immediately from the construction. □
In set theory, a set A is said to be nested within a set B if . We notice the following:
Remark 2. In Example 9, we observe that every cell generated at some step is nested within a cell from the previous step.
We use the above observation to deduce the following result:
Theorem 2. Let and let (respectively, ) be the chosen seed at the j-th (respectively, ( )-th) step. For every in , there exists an in such that is nested within .
Proof. Let , and let (respectively, ) be the chosen seed at the j-th (respectively, -th) step. Suppose . We need to construct an in so that (*).
Since (by Algorithm 1), it follows from Proposition 2(b) that , so that . Since , it is either the case that , or that .
Case 1. If , then, since , we have that . Thus, (*) holds for .
Case 2. Suppose that and let be arbitrary. We need to construct an and a so that .
Since
, there exists an
such that
. Let
g =
+
h, where
for all
. We verify that
. It follows by construction that
. Since
and
, we have that
. Since
is an ideal and
, it follows that
, i.e.,
.
Now, for all
, let
Let
. We verify that
. Let
be arbitrary. If
, then it follows by (
7) that
and
(A). Since
and
, it follows that
, and hence
. We need to show that
. So, by way of contradiction, suppose that
. Then,
. Since
and
, it follows by (
6) that
, which contradicts (A). Therefore, we deduce that
. Hence,
whenever
for an arbitrary
i. From this, we deduce that
.
All that is left to prove is that , i.e., . So, let i ∈ {1,…, s} be arbitrary. We show that qi + hi = ri + + hi. We consider all possible cases. (Note that
we do not consider the cases where ri = = 1, since by (7), if ri = 1, then = 0.)
Subcase 1. If
, then
. Since
and
, it follows from (
7) that
. Hence,
, so that
.
Subcase 2. If
and
, then
. Since
, it follows from (
7) that
. Hence,
, so that
.
Subcase 3. If
and
, then
; by (
6),
, and thus
. Since
, we need
to equal 1 for
to equal 1. Thus, by way of contradiction, suppose that
. Since
i is arbitrary, we have that
implies that
, and thus
, a contradiction. Therefore, we deduce that
, so that
. Hence,
.
Subcase 4. If and , then and (regardless of the value of ), so that .
Subcase 5. If
and
, then
, and, by (
7), we have that
. Hence,
, so that
.
Subcase 6. If and , then and (regardless of the value of ), so that .
Therefore, for all , we have that . Hence, . Since was arbitrary, it follows that . Since was arbitrary, it follows that for all in , there exists an in such that .
This concludes the proof. □
4. Maximal Semiring Homomorphisms
As previously mentioned, the notion of a maximal semiring homomorphism was introduced by Allen [
1] as a means of constructing
Q-ideals.
Definition 9 ([
1])
. A semiring homomorphism is called maximal
if, for every , there exists a unique such that for every , we have , where .
We give an example of a maximal semiring homomorphism.
Example 10 ([
1])
. If R denotes the set of non-negative real numbers with the usual ordering, then , where and for all , forms a commutative semiring. Let , where Let be a map defined byIt can be verified that is a maximal semiring homomorphism.
It is true that there exist semiring homomorphisms that are not maximal.
Example 11 ([
1])
. Consider the set of natural numbers , which is a well-ordered set under the usual ordering of the natural numbers. Thus, can be viewed as a semiring as described in Example 6. Define byWe verify that η is a homomorphism from onto : Let be arbitrary and suppose w.l.o.g. that . Then, and .
Case 1. If , then and . Since and , we haveand Case 2. If and , then and . Since and , we haveand Case 3. If , then and . Since and , we haveand Hence, η is a homomorphism.
Since , we have, for each , that for each . Thus, for each . Similarly, if . Since, for all , we have , it follows that for all . Hence, no in exists such that for all . Therefore, η is not maximal.
Using the notion of a maximal semiring homomorphism, Allen introduced a method by which the quotient structure of a semiring with respect to a Q-ideal can be constructed.
Lemma 6 ([
1])
. Suppose is a semiring homomorphism. If η is maximal, then is a Q-ideal, where .
Even though the above result allowed Allen to generalize the Fundamental Theorem of Homomorphisms for a large class of semirings, the method for identifying these homomorphisms remains unclear.
We imitate the construction of Allen in ([
1] Example 12, p. 415) with the aim of constructing a maximal semiring homomorphism corresponding to the
Q-ideal partition.
Lemma 7. Let . Suppose that , where , and w.l.o.g. that Let be an ordered set with least element , where and for all . Define by for all . Then, η is a maximal semiring homomorphism.
Proof. It is easy to verify that is a semiring homomorphism. We verify that is maximal.
Suppose . We consider all possible cases.
Case 1. : Let
. Suppose
. Then,
. Suppose
. We have that
for some
, and thus
Since was arbitrary, it follows that . Since was arbitrary, it follows that for all . This proves existence.
To prove uniqueness, suppose by way of contradiction that there exists a
such that for every
, we have that
If
, then
. It follows from (
10) that
If
, then
and
by (
11). Thus, there exists an
such that
, so that
which contradicts (
12). This proves uniqueness.
Case 2. : Let
be the element in
with first entry equal to 1, and 0 elsewhere. Suppose
. Then,
and
. Suppose
. We have that
for some
; thus,
Hence, . Thus, and .
For each , let whenever ; otherwise, let . If , then and if ; if ; otherwise, for each . Hence, , so that . This proves existence.
To prove uniqueness, suppose by way of contradiction that there exists a
such that for every
, we have that
If
, then
. It follows from (
13) that
If
, then
and
. It follows from (
14) that
. Thus, there exists an
such that
, so that
which contradicts (
15). This proves uniqueness. A similar approach can be used for other values of
.
To show that corresponds to the Q-ideal partition, we will prove that (a) and (b)
- (a)
If
, then
. Suppose
. If
, then
. By (
9), we have that
. Since
i was arbitrary, it follows that
. Since
was arbitrary, it follows that
. If
, then for all
, if
, then
. Hence,
. Thus,
. Since
was arbitrary, it follows that
. Therefore,
.
- (b)
Suppose
. If
, then
. If
, then
for at least one
i in
. Consider this
i-th entry. By (
9), we have that
, so that
. Since
i was arbitrary, it follows that
. Since
was arbitrary, it follows that
. Now, suppose
. If
, then
. If
, then there exists at least one
i in
such that
. Hence,
. Since
was arbitrary, it follows that
. Therefore,
.
This completes the proof. □
Remark 3. In Lemma 7, we assume that has j zero entries amongst its s entries and without loss of generality that these zero entries lie among the first j entries of . If we do not make this assumption, we assume that the zeros of are “scattered” across its s entries. That is, for each in . Proving that η is a maximal semiring homomorphism follows the same outline as the proof of Lemma 7.
Example 12. Let be a well-ordered set where . We define and . Then, is a semiring. For each , let It follows from Lemma 7 that is a maximal semiring homomorphism. We have that Hence, corresponds to the Q-ideal partition of .
5. Concluding Remarks
We introduced a novel approach for constructing
Q-ideals, resulting in a new class of partitions of a Boolean semiring. This method facilitated the development of a hierarchical partitioning algorithm that uses the weight of selected seeds. We provide corresponding maximal homomorphisms and illustrate the results with low-dimensional examples. While Allen’s [
1] method for creating
Q-ideals relies on maximal homomorphisms, our alternative technique leverages the weight of selected seeds to predict the cell count of the partition and the size of individual cells. As the primary focus of our manuscript is not on maximal semiring homomorphisms, we decided not to delve deeply into this specific topic. The exploration of
Q-ideals in the context of maximal semiring homomorphisms could serve as an interesting direction for future research.
The method proposed in Theorem 1 is designed with theoretical clarity and precision in mind, and we recognize that scalability can be a concern when applied to larger Boolean semirings. In applications, one would have to expand the realized vectors (the rows of a site-by-species matrix) to the power set and then, after partitioning, extract the realized vectors from the partitioned power set. This could potentially limit computational efficiency for larger Boolean semirings, since the expansion and extraction of the realized vectors could increase execution time and memory consumption. For smaller or more manageable semirings, we believe that the proposed method offers valuable insights and practical utility and that its broader applicability to larger systems offers an opportunity for future research.
Since the focus of this paper was to introduce our method and give illustrative low-dimensional examples thereof, comparing its computational efficiency to other methods in this paper is not possible but will rather form part of ongoing research. The focus of our follow-up paper is on the evaluation of the “goodness” of our method by elucidating the relationship between the average within- and between-similarity of the resulting partitions. Our results suggest that our proposed method consistently produces “good” partitions. We believe that comparing its computational complexity to those of other methods is relevant to the scope therein.
The number of ways to partition a set corresponds to the Bell numbers. That is, the number of ways to partition a set of size n equals , where denotes the number of ways to choose k elements from a set of size , where . For example, the number of ways to partition a set of size 8 equals 4140; the number of Q-ideal partitions of a set of size 8 equals 8 (since each seed produces a unique partition; see Lemma 4). Since hierarchical partitioning typically has exponential time complexity, especially when evaluating all possible partitions of a given set (e.g., agglomerative or divisive hierarchical clustering [O()], graph partitioning via minimum bisection [NP-hard], optimal decision tree construction [exponential], K-partitioning of sets via exhaustive search [Bell numbers, that is, super-exponential], and integer linear programming ILP-based partitioning [NP-hard]), we believe that our algorithm mitigates this inefficiency. This is because the number of possible Q-ideal partitions is significantly smaller than the total number of partitions, particularly as n increases.
Throughout this paper, there are no bounds placed on the weight of a selected seed to partition a Boolean semiring, except in Lemma 9, where we assume that the weight of a selected seed of length s is strictly less than s. Although we do not place boundaries on the weight of a selected seed in our proposed method to construct Q-ideals (Theorem 1), we are less interested in the two trivial cases, that is, where the weight equals either 0 or s. This is because the resulting Q-ideal partition will either be the set where each seed belongs to its own cell, or the set where all seeds belong to a single cell, respectively. While we believe that using seed weights can significantly reduce the computational cost compared to evaluating all possible partitions, we considered whether the initial choice of seed weights affects the overall partitioning outcome. This is particularly relevant in situations where the seed weights do not fully reflect the structure of the dataset or if they are selected in a manner that is not optimal (for instance, if a seed of significantly low weight is chosen, but the realized set of vectors all have weights close to s). Identifying the optimal set of seeds for hierarchical partitioning or designing a single seed for “good” partitioning remains an open question, offering opportunities for future research and exploration.