Regular Uniform Hypergraphs, $s$-Cycles, $s$-Paths and Their largest Laplacian H-Eigenvalues
Liqun Qi, Jiayu Shao, Qun Wang
Introduction
Let and . A -uniform hypergraph has vertex set , which is labeled as , and edge set . By -uniformity, we mean that for every edge , the cardinality of is equal to . If , we have an ordinary graph.
The largest Laplacian eigenvalue of a graph plays an important role in spectral graph theory . A natural definition for the Laplacian and signless Laplacian tensors of a -uniform hypergraph , where , was introduced in . It was shown there that the largest Laplacian H-eigenvalue of is always less than or equal to the largest signless Laplacian H-eigenvalue of , while the latter is always less than or equal to , where is the largest degree of . In , the odd-bipartite hypergraph was introduced. In , it was proved that the largest Laplacian H-eigenvalue of a connected -uniform hypergraph is equal to its largest signless Laplacian H-eigenvalue if and only if is odd-bipartite. This result extended the classical result in spectral graph theory .
In this paper, we show that the largest signless Laplacian H-eigenvalue of a connected -uniform hypergraph , where , reaches its upper bound , if and only if is regular. Thus, the largest Laplacian H-eigenvalue of , reaches the same upper bound, if and only if is regular and odd-bipartite.
We then turn our attention to -paths and -cycles.
Researchers in hypergraph theory have studied loose cycles, loose paths, tight cycles and tight paths extensively .
Let be a -uniform hypergraph. Suppose . According to , if such that is an edge of for , then is called an -path. In , is called a loose path if , and a tight path if . In , is also called a loose path for and a tight path for . To avoid confusion, in these two cases, as in , we call a generalized loose path and a generalized tight path respectively. According to , if such that is an edge of for , where vertices for any , then is called an -cycle. According to , if , is called a loose cycle, if , is called a tight cycle. We call a generalized loose cycle for , and a generalized tight cycle for . For an -cycle, in this paper, we assume that . In this way, each pair of consecutive edges in the -cycle will have exactly common vertices. In the next section, we will discuss this in details.
We show in this paper that an -cycle , as a -uniform hypergraph, where , is regular if and only if is a multiple of .
The Laplacian H-eigenvalues of loose paths and loose cycles were studied in . In , power hypergraphs and cored hypergraphs were introduced. Loose paths and loose cycles are power hypergraphs. Power hypergraphs are cored hypergraphs. Even-uniform cored hypergraphs are odd-bipartite. As cycles are symmetric, their largest signless Laplacian H-eigenvalues can be identified directly. Thus, the largest Laplacian H-eigenvalues of odd-bipartite cycles can be identified directly. In , the largest Laplacian H-eigenvalue of an even-uniform loose cycle was identified directly.
According to , the largest Laplacian H-eigenvalue of -uniform hypergraph is always greater than or equal to the largest degree of that -uniform hypergraph. By , when is even, equality cannot hold, but when is odd, equality may hold in certain cases. It was proved in that equality holds for odd-uniform loose paths and loose cycles.
It was observed in that if , then an -path or an -cycle is a cored hypergraph, but not a power hypergraph in general.
These results raised several questions. First, if is even and , are some -paths and -cycles still odd-bipartite, though they are not cored hypergraphs? Second, can we identify the largest Laplacian H-eigenvalues of even-uniform odd-bipartite -cycles directly? Third, when is odd and , are the largest H-eigenvalues of -paths and -cycles equal to the corresponding largest degrees? We will study these questions in this paper.
We give some basic definitions in the next section.
In Section 3, we prove the result about regular uniform hypergraphs mentioned before, and identify regular -cycles.
In Section 4, we show that if is even, all the -paths and all the non-regular -cycles are odd-bipartite. We prove that a regular -cycle with is odd-bipartite if and only if is a multiple of , where is the number of edges in , and for some integers and .
In , several classes of hypergraphs were shown to be odd-bipartite. But only in this paper, some regular -cycles are shown to be not odd-bipartite. To show that a hypergraph is odd-bipartite, one only needs to give an adequate odd-partition for the vertex set of that hypergraph. Such an odd-partition is not unique in general. To show that a hypergraph is not odd-bipartite, one needs to prove that there is no such an odd-partition for the vertex set of that hypergraph. Hence, in general, it is not a trivial task to show that a hypergraph is not odd-bipartite.
In Section 5, we identify the largest Laplacian H-eigenvalues of even-uniform -cycles directly when . These include all the even-uniform non-regular -cycles, and those odd-bipartite regular -cycles.
We introduce supervertices for hypergraphs in Section 6, and show there that the components of an H-eigenvector of an odd-uniform hypergraph are equal if such components corresponds vertices in the same supervertex, and the corresponding Laplacian H-eigenvalue is not equal to the degree of the supervertex. Using this property, in Section 7, we show that the largest H-eigenvalue of an odd-uniform generalized loose -cycle is equal to , the maximum degree of that -cycle.
In Section 8, we show that the largest Laplacian H-eigenvalue of a -uniform tight -cycle is at least , if the number of vertices is even and for some nonnegative integer . Note that in this case . We show that equality holds here if and . When , and , we show that the largest Laplacian H-eigenvalue is no more than .
Some final remarks are made in Section 9.
Preliminaries
H-eigenvalues are real numbers, by Definition 2.1. By , we have that the number of H-eigenvalues of a real tensor is finite. By , we have that all the tensors considered in this paper have at least one H-eigenvalue. Hence, we can denote by as the largest H-eigenvalue of a real tensor .
For a subset , we denoted by its cardinality.
Consider a -uniform hypergraph with vertex set , which is labeled as , and edge set . For a subset , we denote by the set of edges . For a vertex , we simplify as . It is the set of edges containing the vertex , i.e., . The cardinality of the set is defined as the degree of the vertex , which is denoted by . If two vertices and are in the same edge , then we denote . Two different vertices and are connected to each other (or the pair and is connected), if there is a sequence of edges such that , and for all . A hypergraph is called connected, if every pair of vertices of is connected. A hypergraph is regular if . A hypergraph is complete if consists of all the possible edges. In this case, is regular of degree .
For the sake of simplicity, we mainly consider connected hypergraphs in the subsequent analysis. By the techniques in , the conclusions on connected hypergraphs can be easily generalized to general hypergraphs.
The following definition for the Laplacian tensor and signless Laplacian tensor was proposed by Qi .
Let be a -uniform hypergraph. The adjacency tensor of is defined as the -th order -dimensional tensor whose -entry is:
Let be a -th order -dimensional diagonal tensor with its diagonal element being , the degree of vertex , for all . Then is the Laplacian tensor of the hypergraph , and is the signless Laplacian tensor of the hypergraph .
By , zero is always the smallest H-eigenvalue of , and we have , where is the maximum degree of . For , the polynomial system in Definition 2.1 has the form
For , the polynomial system in Definition 2.1 has the form
In the following, we define cored hypergraphs.
Let be a -uniform hypergraph. If for every edge , there is a vertex such that the degree of the vertex is one, then is called a cored hypergraph. A vertex with degree one is called a core vertex, and a vertex with degree larger than one is called an intersection vertex.
The notion of odd-bipartite even-uniform hypergraphs was introduced in .
Let be a -uniform hypergraph. Then is called odd-bipartite if is even and either it is trivial (i.e., ) or there is a partition of the vertex set as such that and every edge in intersects with exactly an odd number of vertices.
In the introduction, we claim that , the number of vertices in an -cycle needs to satisfy the condition such that each pair of consecutive edges has exactly common vertices. We now discuss this in details below.
Let be a k-uniform -cycle with vertices and edges, where and . Then each pair of consecutive edges of G contains exactly common vertices if and only if .
Proof. By the definition, we may assume that , and may agree that when and are both viewed as vertices of . Also, we have , where
Necessity. If each pair of consecutive edges of G contains exactly common vertices, then . On the other hand, we have . So we have
Sufficiency. Now suppose that . Then it is easy to verify that
Since implying , we can also verify that
Regular Uniform Hypergraphs and Regular s𝑠s-Cycles
We now establish the following theorem for a connected -uniform hypergraph .
Suppose that is a connected -uniform hypergraph with and maximum degree . Then if and only if is regular. Furthermore, if and only if is regular and odd-bipartite.
where is the degree of vertex . To make this equality hold, we must have and as long as . Applying the same augment for all such with , we have and as long as . As is connected, we see that for all . Thus is regular.
The last conclusion of this theorem follows from the above conclusion and [9, Theorem 5.1].
Clearly, an -path cannot be regular. We now consider regular -cycles.
Let be a -uniform -cycle, with , , , such that is an edge of for , where vertices for any . Then is regular if and only if for some positive integer . In this case, we have , where .
Proof. If , then we see that . Hence is regular in this case. On the other case, suppose , where . Then we see that and . Thus, cannot be regular in this case.
The conclusions of this proposition follow now.
Since , we see that . For a tight cycle, , we see that is also regular with in this case. Thus, we have the following corollary.
Odd-Bipartite s𝑠s-Paths and s𝑠s-Cycles
We assume that is even in this section, as odd-bipartite hypergraphs are only for even .
Our first proposition in this section shows that when is even, all the -paths are odd-bipartite.
Assume that is even. Let be a -uniform -path, where . Then is odd-bipartite.
Proof. According the discussion at the beginning of this paper, we may assume that such that is an edge of for . Let and . Then we see that is odd-bipartite as each edge has exactly one vertex in .
2 Odd-bipartite Non-Regular s𝑠s-Cycles
Our second proposition in this section shows that when is even, all the non-regular -cycles are odd-bipartite.
Assume that is even. Let be a -uniform non-regular -cycle, where . Then is odd-bipartite.
Proof. When , is a loose cycle, thus a power hypergraph . When , has at least one core vertex, thus is a cored hypergraph . In both cases, is odd-bipartite as long as is even, as observed in .
Now, by Proposition 3.1, the remaining case, after excluding regular -cycles, are that , , where . We may assume that , and note that vertices for all . If is odd, let and . Then each edge has exactly vertices in . If is even, let and . Then each edge has exactly vertices in . In both cases, each edge has odd number of vertices in . Thus, is odd-bipartite as long as is even.
3 Odd-bipartite Regular s𝑠s-Cycles
We now give a sufficient and necessary condition for a regular -cycle to be odd-bipartite.
Let be a k-uniform -cycle with vertices and edges, where , is even and . Assume that there is an integer such that (Thus is regular by Proposition 3.1). Write for some nonnegative integers and . Then is odd-bipartite if and only if is a multiple of .
Proof. We assume that (the set of integers modulo n). Namely, we agree that , when and are both viewed as vertices of .
Also, the edges of are as follows:
where each edge consists of k cyclicly consecutive vertices of .
Sufficiency. Suppose that . Let . Then we have . Let
be the set of all the multiples of in the set .
Since and are both multiples of , we see that each set of k cyclicly consecutive elements in contains exactly elements which are multiples of . Thus each edge of contains exactly vertices in . Hence is odd-bipartite.
Necessity. We write if the two integers and have the same parity. Let
Then we have and . Also we agree that (as subsets of ).
Now suppose that is odd-bipartite with the bipartition . Let
Then by the definition of odd-bipartition, all are odd.
On the other hand, since we also have
Let be the greatest common divisor of and . Then for some integers and . So by (7) and (8) we have
Now let . Then , so by (9) we have
Since is odd, is also odd, which implies that is a multiple of . Thus is also a multiple of .
Figure 1 indicates an odd-bipartite regular -cycle with and . We see that is odd-bipartite with and .
The smallest non-odd-bipartite even-uniform regular -cycle may be as follows: and (thus . Using Matlab, we find that the Laplacian H-eigenvalues of this -cycle are and only. Thus, in this example. This confirms Theorem 4.1.
In Theorem 4.1, if is even and is odd, then is odd-bipartite.
If , then is a tight cycle. We have the following corollary.
Let be a -uniform tight cycle, i.e., . Then is regular. Assume that is even. We may write for two nonnegative integers and . Then is odd-bipartite if and only if is a multiple of .
The Largest Laplacian H-eigenvalue of Odd-Bipartite s𝑠s-Cycles
In this section, we identify the largest signless Laplacian H-eigenvalues of -cycles in all possible cases. When these -cycles are odd-bipartite, these values are also their largest Laplacian H-eigenvalues.
This is the case that . Suppose is such an -cycle. Then for each edge, there are core vertices, and intersection vertices.
Now we take be a positive vector with if is a core vertex, and if is an intersection vertex. Suppose that is an H-eigenvector of corresponding to the H-eigenvalue . Note that the degree of a core vertex is and the degree of an intersection vertex is . By (3), we would have
Since and , (10) has a root . Let . Then . By , since is a cored hypergraph, we have if is even.
We conclude this discussion as the following theorem.
Suppose that is an -cycle with and . Then , where is the unique root of (10) in . When is even, we have too.
Note that in this case and we have . This confirms Theorem 3.1 and [17, Corollary 6.2].
2 Regular s𝑠s-Cycles
By Proposition 3.1 and Theorem 4.1, we have the following proposition.
Suppose that is an -cycle with . Then is a regular hypergraph and . Assume further that is even. If either is odd or for a positive integer and a nonnegative integer , and is a multiple of , then we have too.
It will be a further research topic to find the value of for a non-odd-bipartite regular -cycle.
3 Non-Regular Generalized Tight s𝑠s-Cycles
In this subsection, we consider a generalized tight -cycle , which is not a regular -cycle. We may assume that and , where . Then for each edge, there are vertices with degree , and vertices with degree .
Now we take be a positive vector with if is a vertex with degree , and if is a vertex with degree . Suppose that is an H-eigenvector of corresponding to the H-eigenvalue . By (3), we should have
Since and , (11) has a root . Let . Then . If is even, then is odd-bipartite. By , we have in this case.
We conclude this discussion as the following theorem.
Suppose that is an -cycle with and , , with . Then , where is the unique root of (11) in . When is even, we have too.
In this case and we have . This also confirms Theorem 3.1 and [17, Corollary 6.2].
Note that all -cycles are covered by the discussion in these three subsections.
Supervertices
We now define supervertices for a -uniform hypergraph.
Let be a -uniform hypergraph. Let . The vertex set
is called a supervertex of . Clearly, any vertex in the same supervertex has the same degree. We call this degree the degree of that supervertex. In particular, if a supervertex contains a core vertex, then all vertices in that supervertex are core vertices. We call such a supervertex a core supervertex. Otherwise, we call it an intersection supervertex.
For example, in Figure 1, there are four intersection supervertices , . For a loose cycle or a generalized loose -cycle with , there are core supervertices and intersection supervertices, where is the number of edges in that -cycle. Each core supervertex has cardinality . Each intersection supervertex has degree and cardinality .
Suppose that is a -uniform hypergraph with . Let be a supervertex of , with degree and cardinality . Suppose that is a Laplacian H-eigenvalue of , . Let be a Laplacian H-eigenvector of , corresponding to . Suppose . Then . If is odd, then .
As , we have . The conclusions follow from this equality.
Note that Lemma 3.1 of is a special case of this theorem.
Odd-Uniform Generalized Loose s𝑠s-Cycles
In this section, assuming that is odd, using Theorem 6.1, we show that the largest Laplacian H-eigenvalue of an odd-uniform generalized loose -cycle is equal to , the maximum degree of that -cycle. This result extends the result on odd-uniform loose cycles in Subsection 4.1 of . Since is odd, the case that is not included. Thus, we have . The -cycle has always core vertices.
Suppose that is an -cycle with and is odd. Then . If is even, then the only Laplacian H-eigenvalue of , satisfying , is .
Proof. Suppose that has edges. Then has core supervertices and intersection supervertices , displayed as , such that the edges of are . For . Furthermore, assume that the vertices of are , where . Then , }, for .
Suppose that , is a Laplacian H-eigenvalue of . Let be an Laplacian H-eigenvector corresponding to . By Theorem 6.1, we may assume that for , for , for . Let , , .
If is even, by (12), we must have for all , otherwise we would have for some , contradicting (12). This implies that for at least one . By (13), this implies that , a contradiction. This proves that when is even, implies . The conclusion for the case that is even is proved.
From now we assume that is odd and . Then by (12), we see that implies that .
(i). First assume that for all . By (12), we have for . Thus, must be even, otherwise we get a contradiction by the rule of alternating signs of . Assume that is even. By (13), we have
Since for , we get a contradiction, as should have the same sign as , which is nonzero. The conclusion follows now.
(ii). We now assume for all . This implies that for at least one . By (13), this implies that , a contradiction.
(iii). Finally, we assume that for some and for other . Without loss of generality, we may assume that and . By (12), we have . By taking in (14), we have
Now we use induction to show that for . The case follows from and . We assume and . Then since implying . From (14) we have (since and are both odd):
which implies and . But also implies , so we obtain and thus complete the inductive proof. Taking in , we obtain , a contradiction.
Thus, when is odd, we cannot have . This implies that .
Odd-Uniform Tight s𝑠s-Cycles
In this section, we assume that is odd and . Then we have tight -cycles. We will see that the results on the largest Laplacian H-eigenvalues here are very different from those in the last section.
Suppose that is a tight -cycle with and for a nonnegative integer . Then . When , the number of vertices, is even, we have .
Proof. Let be defined by and for . Also, we assume that for any . From this we see that the sum of any consecutive components of is zero, and the product of any consecutive components of is . Thus we have
Now multiplying the both sides of (2) by , we obtain
This shows that and satisfy this system, i.e., is an H-eigenvalue of . As , the conclusion follows.
This is the second example that when is odd. The first example for this is the -uniform complete hypergraph, given in . By using supervertices, we may generalize this result to -uniform regular -cycles, with , for some , where and are even.
We do not know what kind of result can be established for . But for , we can get the exact value of when , and an upper bound of for all .
Suppose that is a tight -cycle with and . Then . When , we have . When , we have .
Proof. When and , (2) has the form:
Since , we have . Combining with the conclusion of Proposition 8.1, we see that .
When , and , (2) has the form:
for . Summing up it for from to , we have
Since , we have . Thus, in this case.
Using Matlab to solve (16), we find that when and , the -cycle has only three distinct H-eigenvalues: , and .
The second conclusion of Proposition 8.2 is not sharp in the proof. Actually, our Matlab computation shows that when and , the -cycle has only three distinct H-eigenvalues: , and ; when and , the -cycle has only four distinct H-eigenvalues: , , and . Thus, we have the following conjecture.
Suppose that and . Then . When is even, we have . When is odd, we have .
Final Remarks
In this paper, we showed that the largest signless Laplacian H-eigenvalue of a connected -uniform hypergraph , where , reaches its upper bound , where is the largest degree of , if and only if is regular, and that the largest Laplacian H-eigenvalue of , reaches the same upper bound, if and only if is regular and odd-bipartite. We proved that an even-uniform -path and an even-uniform non-regular -cycle are always odd-bipartite. Theorem 4.1 characterized odd-bipartite regular -cycles. We identified the largest signless Laplacian H-eigenvalue of an -cycle. When the -cycle is odd-bipartite, this gives the largest Laplacian H-eigenvalue of that -cycle. We then introduced supervertices and showed that the largest Laplacian H-eigenvalue of an odd-uniform generalized loose -cycle is , the maximum degree of that -cycle. We also showed that the largest Laplacian H-eigenvalue of a -uniform tight -cycle is not less than the maximum degree of that -cycle, plus one, if the number of vertices is even and . It will be a further research topic to prove or to disprove Conjecture 8.1, and to identify the largest Laplacian H-eigenvalue of an -path or a general non-odd-bipartite -cycle, for . It will be interesting to see if one may use the tensor eigenvalue theory to study other research topics related with -paths and -cycles.