ON STRONG GRAPH BUNDLES

F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

Abstract. We study strong graph bundles: a concept imported from topology which gen-
eralizes both covering graphs and product graphs. Roughly speaking, a strong graph bundle
always involves three graphs E, B and F and a projection p : E → B with fiber F (i.e.
p−1 (x) ∼= F for all x ∈ V (B)) such that the preimage of any edge xy of B is trivial (i.e.
p−1 (xy) ∼= K2  F ). Here we develop a framework to study which subgraphs S of B have
trivial preimages (i.e. p−1 (S) ∼= S  F ) and this allows us to compare and classify several
variations of the concept of strong graph bundle. As an application, we show that the clique
operator preserves triangular graph bundles (strong graph bundles where preimages of tri-
This is an author version of a paper published in: DM 340 (2017) pp 3073–3080

angles are trivial) thus yielding a new technique for the study of clique divergence of graphs.

1. Introduction

In topology a fiber bundle is a space which is locally a product of spaces [41]. This concept has
proved to be very important in many fields of mathematics including algebraic geometry,
differential geometry and differential topology. Also, fiber bundles play a central role in
general relativity. Thus, the importance of fiber bundles in mathematics and physics is
difficult to overstate. The analogues of fiber bundles in graph theory, i.e. graph bundles, were
introduced (as reported in [34]) by Pisanski and Vrabec in a 1982 unpublished preprint, and
appeared for the first time (with Shawe-Taylor as an additional author) in [33]. Since there
are several notions of a product in graph theory, there are also several notions of a graph
bundle. Most works on graph bundles focus on Cartesian graph bundles [1–3, 5, 7, 8, 12, 13,
https://doi.org/10.1016/j.disc.2017.07.007

15–23, 30, 33–35, 40, 43–47], where graphs are locally a Cartesian product of graphs, but
there is also research on strong graph bundles [18, 30, 43, 47], tensor graph bundles [14, 18]
and lexicographic graph bundles [30]. The one just given is an exhaustive classification of
all the papers on graph bundles that we could find. Here we shall focus on strong graph
bundles, as the strong product  suits our purposes best.
More specifically, a strong graph bundle always involves three graphs E, B, and F and a
projection p : E → B. Saying that E is “locally a product” means then that the preimage
p−1 (x) of each vertex x ∈ V (B) can be seen as {x}  F in such a way that the restriction
of p is just the first projection, and also the preimage p−1 (xy) of each edge xy ∈ E(B) is
isomorphic to {xy}  F , again in a way equally compatible with the first projection. Mohar,
Pisanski and Škoviera remarked in [30] that a more natural equivalent definition is obtained
by asking that the preimage p−1 (St(x)) of the star of each vertex x ∈ V (B) can be seen as
St(x)  F in such a way that the restriction of p is just the first projection. As far as we
know, this is the only notion of locality employed so far in the literature of graph bundles.

Partially supported by SEP-CONACyT, grant 183210.

2010 Mathematics Subject Classification. Primary 05C76, 05C69; Secondary 05C62, 05C75.
Key words and phrases. triangular graph bundle, strong graph bundle, fiber bundle, clique graph, graph
dynamics.
1
2 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

But just as the concept of a product of graphs is not unique, neither is so that of locality in
a graph, and hence each kind of locality may produce a variant version of graph bundles. We
shall study strong graph bundles where these local subgraphs are indeed vertices and edges
(or stars), but then we shall explore other types of locality by adding triangles, cliques and
closed neighborhoods. It will turn out that all three of them are equivalent (Theorem 3.2),
but not equivalent to the original one involving only vertices and edges (Figure 1a). The
new kind of strong graph bundle introduced here will be called triangular graph bundle.

Our Theorem 2.8 (together with Lemma 2.1) provides a framework in which different versions
of locality for strong graph bundles can be studied and compared. In Theorem 3.2 this
result proves the equivalence of our three definitions of a triangular graph bundle, and in
Corollary 2.9 it also yields the above-mentioned equivalence [30] of the two definitions of
the original strong graph bundles. An interesting and useful tool behind Theorem 2.8 is the
concept of agreement of graph morphisms at a vertex in Definition 2.3.

We also give an application to clique graphs: Theorem 4.1 states that triangular graph
bundles are preserved by the clique operator. This yields, in Theorem 4.2, a new method
for proving clique divergence or clique convergence which generalizes and unifies previously
known results about strong products [24, 31] and triangular covering maps [25].

Let us quickly review the basic terminology now.

Our graphs are simple and finite. The vertex and edge sets of a graph G are denoted by
V (G) and E(G), and |G| = |V (G)| is the order of G. A graph H is a subgraph of G (denoted
by H ≤ G) if V (H) ⊆ V (G) and E(H) ⊆ E(G). An induced subgraph of G is a subgraph H
of G such that whenever x, y ∈ V (H) and xy ∈ E(G), we also have xy ∈ E(H). The union
of the graphs G and H is given by V (G ∪ H) = V (G) ∪ V (H) and E(G ∪ H) = E(G) ∪ E(H),
and their intersection is given by V (G ∩ H) = V (G) ∩ V (H) and E(G ∩ H) = E(G) ∩ E(H).
Two vertices x, y are adjacent-or-equal in G (denoted by x ' y), if x = y or xy ∈ E(G). The
closed neighborhood of x ∈ V (G) is the subgraph NG [x] ≤ G induced by {y ∈ V (G) | x ' y}.

A morphism (or map) f : G → H is a function on the vertex sets f : V (G) → V (H) such that
x ' y ⇒ f (x) ' f (y). In this case we denote the domain, codomain and image of f by Df ,
Cf and If respectively, i.e. Df = G, Cf = H, and If is given by V (If ) = {f (x) | x ∈ V (G)}
and E(If ) = {f (x)f (y) | xy ∈ E(G) and f (x) 6= f (y)}. Note that If ≤ Cf may be non-
induced. Given f : G → H and S ≤ H, the inverse image of S is the subgraph f −1 (S) of G
given by V (f −1 (S)) = f −1 (V (S)) and E(f −1 (S)) = {xy ∈ E(G) | f (x) ' f (y) in S}. Also
f −1 (S) could be non-induced in G. The strong product GH of two graphs is determined by
V (G  H) = V (G) × V (H) and E(G  H) = {(v, w)(v 0 , w0 ) | v ' v 0 in G and w ' w0 in H}.

As usual when studying clique graphs, a complete of G is a complete subgraph of G, and

we reserve the word clique for maximal complete subgraphs. The clique graph K(G) is
the intersection graph of the cliques of G and the operator K is called the clique operator.
Then the iterated clique graphs K n (G) are defined inductively by K 0 (G) = G and K n (G) =
K(K n−1 (G)). If the sequence {|K n (G)|} is bounded (equivalently, if K m (G) ∼= K n (G) for
some m > n), we say that G is K-convergent. On the other hand, G is called K-divergent if
the sequence {|K n (G)|} is unbounded. The K-behavior of G can be either K-convergent or
K-divergent. This dichotomy is a major topic in the theory of clique graphs, and many papers
 ON STRONG GRAPH BUNDLES 3

have appeared providing techniques for determining the K-behavior (e.g. [4, 6, 9, 10, 25–
29, 31, 32, 36, 42]). Applications of the theory of the clique operator include the fixed point
property for posets [11] and loop quantum gravity [37–39].

2. Strong Graph Bundles

Given a graph B, hereinafter referred to as the base graph, a projection over B is a graph
morphism p : E → B which is vertex-surjective and edge-surjective, i.e. Ip = Cp . The domain
E = Dp will be called the total graph of the projection. The fiber of a vertex x ∈ V (B) is the
preimage p−1 (x) ≤ E of the one-vertex subgraph {x} ≤ B. These fibers of p are non-empty
induced subgraphs of E. In fact, if S ≤ B is induced, then p−1 (S) ≤ E is induced. Even
if S is not induced the restriction of p, denoted also by p : p−1 (S) → S, is again a projection.

Any projection p : E → B partitions V (E) into the disjoint union of the vertex sets of its
fibers, so each v ∈ V (E) lies in a unique fiber of p, namely v ∈ V (p−1 (x)) for x = p(v). The
projection p being understood, we say that v lies over x, or that v is a vertex over x. The
formula v = x̃ means both that v ∈ V (E) and p(v) = x, so in what follows x̃ shall always
denote a vertex of the total graph E lying over the vertex x of the base graph B.

If p0 : E 0 → B is another projection over the same base graph B, a morphism from p to p0 is

a graph morphism ϕ : E → E 0 such that p0 ◦ ϕ = p, i.e. the following diagram commutes:

ϕ
E E0
p p0
B.

In this case ϕ clearly sends (distinct) fibers of p into (distinct) fibers of p0 . If ϕ is a graph
isomorphism then ϕ−1 is also a morphism from p0 to p, and in this case one says that p and p0
are isomorphic. Notice that if this is the case ϕ restricts to an isomorphism from p−1 (x) to
(p0 )−1 (x) over each vertex x of B: isomorphic projections are fiberwise isomorphic.

When all the fibers of a projection p : E → B are isomorphic to some graph F we say that p
is a projection with fiber F . There is always a trivial projection with any given base B and
fiber F : the total space is the strong product B  F and the projection is just the first
canonical projection π1 : B  F → B given by π1 (v, w) = v. Any projection isomorphic to
a trivial one is also called trivial. Thus, a projection p : E → B is trivial (with fiber F ) if
there is an isomorphism ϕ : E → B  F such that the following diagram commutes:

ϕ
E BF
p π1

B.
4 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

In this situation, the isomorphism ϕ is called a trivialization of p. Being morphisms to a

product, trivializations are determined by their components as in the following diagram:

τ F
π2

ϕ
E BF ϕ = (p, τ ) = (π1 ◦ ϕ, π2 ◦ ϕ)
π1
p B ϕ(v) = (p(v), τ (v)).

As the first component has to be p, a candidate for a trivialization ϕ : E → B  F is always

given as ϕ = (p, τ ) where τ : E → F can be any morphism, and then ϕ(v) = (p(v), τ (v))
for any v ∈ E, or rather ϕ(x̃) = (x, τ (x̃)) for any x̃ over x as explained above. This
ϕ = (p, τ ) is called the product map of its components p and τ . As ϕ needs to restrict to
isomorphisms from the fibers of p to those of π1 , the restrictions of τ to the fibers of p should
be isomorphisms onto F . This would already imply that ϕ is vertex-bijective, and it would
even be part of the other condition (that is, to be edge-surjective) that ϕ still needs to satisfy
in order to be indeed a trivialization.
Graph bundles won’t be forced to be trivial, they will only have to be trivial over certain
subgraphs of the base graph. Let p : E → B be a projection and take a subgraph S ≤ B.
We say that p is trivial over S (with fiber F ), or that S is (p, F )-trivial, if the restricted
projection p : p−1 (S) → S is trivial with fiber F . In other words, S is (p, F )-trivial if there
exists some isomorphism ϕS : p−1 (S) → S  F for which the following diagram commutes:
ϕS
p−1 (S) SF

p π1

S.
In this case we say that ϕS is a trivialization (of p) over S. The following is immediate:
Lemma 2.1. If T ≤ S ≤ B and S is (p, F )-trivial, then T is also (p, F )-trivial. 

Note that a projection p : E → B has fiber F if and only if all one-vertex subgraphs
of B are (p, F )-trivial. Indeed, if τx : p−1 (x) → F is any morphism, the product map
ϕx = (p, τx ) : p−1 (x) → {x}  F is an isomorphism (and hence trivializes p over x) if and
only if τx = π2 ◦ ϕx is an isomorphism, as π2 : {x}  F → F is an isomorphism in this case.
Definition 2.2. [30]: Let p : E → B be a projection with fiber F . Then (E, B, F, p) is a
strong graph bundle if every edge xy ∈ E(B) (as a subgraph {xy} ≤ B) is (p, F )-trivial.
Since we will only consider strong graph bundles, we will call them just graph bundles. Notice
that (E, B, F, p) is a graph bundle if and only if p : E → B is a projection and every complete
subgraph C ≤ B of order 1 or 2 is (p, F )-trivial.
For the rest of the section we shall need the notions of twinship and agreement.
Two vertices x, y are twins in G (denoted by x ≈ y) if V (NG [x]) = V (NG [y]). Neighborhoods
are induced subgraphs, so this is the same as asking that NG [x] = NG [y]. With this definition
every vertex is a twin of itself, and twinship is an equivalence relation.
 ON STRONG GRAPH BUNDLES 5

Definition 2.3. Two morphisms f, g will be said to agree at a vertex x ∈ V (Df ∩ Dg ) if

f (x) and g(x) are twins (i.e. f (x) ≈ g(x)) in If ∪ Ig .
We also say that two trivializations ϕS and ϕT of our projection p : E → B agree over
x ∈ V (S ∩ T ) if they agree at every vertex x̃ over x.
We use ‘twinship’ instead of ‘equality’ in our definition of agreement since this way our
statements and proofs are easier and smoother. Now we prove that agreement over one
vertex can always be achieved (the proof actually establishes strict agreement here):
Theorem 2.4. If two subgraphs S, T ≤ B are (p, F )-trivial and x ∈ V (S ∩ T ), then for any
trivialization ϕS over S there is a trivialization ϕT over T which agrees with ϕS over x.
Proof. Let ϕS : p−1 (S) → S  F and ϕ0T : p−1 (T ) → T  F be any two trivializations.
We know already that in the following diagram the restrictions of τS and τT0 to p−1 (x) are
isomorphisms, because so are π2 and both ϕ’s. Then there is an automorphism α ∈ Aut(F )
which makes the whole diagram commute, namely, with the restricted τ ’s, α = τS ◦ (τT0 )−1 :
ϕ0T ϕS
{x}  F p−1 (x) {x}  F
τT0 τS
π2 π2
α
F F.
Define τT : p−1 (T ) → F by τT = α ◦ τT0 . Then, restricting to p−1 (x), τT = α ◦ τT0 = τS over x.
Now define ϕT : p−1 (T ) → T  F by ϕT = (p, τT ). Then ϕT = (p, τT ) and ϕS = (p, τS ) agree
over x since τT (x̃) = τS (x̃) for all x̃ over x. Notice that ϕT = (1T  α) ◦ ϕ0T is a composition
of isomorphisms:
ϕ0T 1T  α
p−1 (T ) T F T  F.
ϕT

Therefore ϕT is an isomorphism, and hence it is a trivialization of p over T . 

Now, consider two vertices (x, v), (x, w) of BF , both in the same fiber of π1 : BF → B: as
NBF [(x, v)] = NB [x]  NF [v] and NBF [(x, w)] = NB [x]  NF [w], we see that (x, v) ≈ (x, w)
in B F if and only if v ≈ w in F . Hence we can characterize agreement of the trivializations
ϕS , ϕT in terms of their second components τS and τT :
Theorem 2.5. Assume that S, T ≤ B and that ϕS and ϕT are trivializations over S and T .
Assume further that x ∈ V (S ∩T ). Then ϕS and ϕT agree over x if and only if τS (x̃) ≈ τT (x̃)
in F for all x̃ ∈ V (p−1 (x)).
Proof. Let x̃ be a vertex over x. Note that IϕS ∪ IϕT = (S  F ) ∪ (T  F ) = (S ∪ T )  F ,
and that both ϕS (x̃), ϕT (x̃) ∈ (S ∪T )F are in the same fiber {x}F of the first projection
π1 : (S ∪ T )  F → S ∪ T . Using that ϕS (x̃) = (x, τS (x̃)) and ϕT (x̃) = (x, τT (x̃)) we get, as
observed above, that ϕS (x̃) ≈ ϕT (x̃) in IϕS ∪ IϕT if and only if τS (x̃) ≈ τT (x̃) in F . 

A key property is that twins of adjacent-or-equal vertices are also so: x ≈ y ' z ≈ w implies
x ' w. We next prove that agreement over one vertex implies agreement over its neighbors:
6 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

Theorem 2.6. Assume that S, T ≤ B and that ϕS and ϕT are trivializations over S and T .
Assume further that x, y ∈ V (S ∩ T ) with x ' y in S ∩ T . Then, if ϕS and ϕT agree over x,
they also agree over y.
Proof. Since ϕS and ϕT agree over x, τS (x̃) ≈ τT (x̃) in F for all x̃ over x by Theorem 2.5.
By the same result it suffices to show, given a vertex ỹ over y, that τS (ỹ) ≈ τT (ỹ) in F .
Let z ∈ NF [τS (ỹ)], so z ∈ F and z ' τS (ỹ). The restriction τS : p−1 (x) → F is onto,
so z = τS (x̃) for some x̃ over x. Then τS (x̃) ' τS (ỹ) =⇒ ϕS (x̃) ' ϕS (ỹ) =⇒ x̃ ' ỹ in
p−1 (S). Hence x̃ ' ỹ also in p−1 (T ) and τT (x̃) ' τT (ỹ). It follows that z = τS (x̃) ≈ τT (x̃) '
τT (ỹ), and hence z ' τT (ỹ). Therefore NF [τS (ỹ)] ⊆ NF [τT (ỹ)]. By symmetry we also have
NF [τS (ỹ)] ⊇ NF [τT (ỹ)], and thus τS (ỹ) ≈ τT (ỹ) in F . 

g : V (Df ∪ Dg ) → V (Cf ∪ Cg ) by
F
Given two morphisms f, g we define the glued function f

 G  f (x) if x ∈ Df
f g (x) =
g(x) otherwise.
Notice that the gluing operation is not commutative, but it is associative. Therefore when
we have an ordered list of morphisms f1 , f2 , . . . , fr we can define:
r
G G G G
fi = f1 f2 ··· fr .
i=1
F
In general the glued function f g is not a morphism, but we have the following theorem:
Theorem 2.7. If the graph morphisms f and g agree at every x ∈ V (Df ∩ Dg ), then the
glued function f g : Df ∪ Dg → Cf ∪ Cg is a morphism of graphs.
F

Proof. Let xy be an edge of Df ∪ Dg . We have to prove that (f g)(x) ' (f

F F
g)(y), and
we shall consider four cases:
Case 1: xy ∈ E(Df ).
Then (f g)(x) = f (x) ' f (y) = (f g)(y).
F F

Case 2: Fxy ∈
/ E(Df ), xy ∈ E(Dg ) and x, y ∈ V (DfF∩ Dg ).
Then (f g)(x) = f (x) ≈ g(x) ' g(y) ≈ f (y) = (f g)(y).
Case 3: xy ∈
/ E(Df ), xy ∈ E(Dg ) and (say) x ∈ V (Df ∩ Dg ) 63 y.
Then (f g)(x) = f (x) ≈ g(x) ' g(y) = (f g)(y).
F F

Case 4: Fxy ∈
/ E(Df ), xy ∈ E(Dg ) and x, y ∈
/ V (Df ∩ Dg ).
Then (f g)(x) = g(x) ' g(y) = (f g)(y).
F


We can now show that triviality extends over unions of connectedly intersecting regions:
Theorem 2.8. Assume that S, T ≤ B are both (p, F )-trivial and that S ∩ T is non-empty
and connected. Then S ∪ T is (p, F )-trivial.
Proof. By Theorem 2.4 we can choose trivializations ϕS and ϕT over S and T which agree
over some x ∈ V (S ∩ T ). Then ϕS and ϕT agree over all y ∈ V (S ∩ T ) by Theorem 2.6.
Hence the glued function ϕ = ϕS ϕT : p−1 (S ∪ T ) → (S ∪ T )  F is a morphism of graphs
F
 ON STRONG GRAPH BUNDLES 7

by Theorem 2.7. We shall only have to keep in mind that, for each x̃ ∈ V (p−1 (S ∪ T )), the
definition of ϕ is that ϕ(x̃) = ϕS (x̃) if x ∈ V (S), and ϕ(x̃) = ϕT (x̃) if x ∈
/ V (S). Let us now
prove that ϕ is indeed a trivialization of p over S ∪ T .
Just as ϕS : p−1 (S) → S  F and ϕT : p−1 (T ) → T  F are the product maps of F their
components ϕS = (p, τS ) and ϕT = (p, τT ), it is quite clear that ϕ = (p, τ ) with τ = τS τT ,
so the following diagram certainly commutes:
ϕ
p−1 (S ∪ T ) (S ∪ T )  F

p π1

S ∪ T.
As both τS and τT restricted to fibers of p are isomorphisms to F , and τ acts as τS over
the vertices of S and as τT over those of T \ S, we already know that ϕ is a vertex-bijective
morphism and hence, it only remains to show that it is edge-surjective.
Let (x, v), (y, w) ∈ (S ∪ T )  F with (x, v) ' (y, w). Then x ' y in S ∪ T and v ' w in F .
Take two vertices x̃ and ỹ in p−1 (S ∪ T ) such that ϕ(x̃) = (x, v) and ϕ(ỹ) = (y, w).
Thus we have ϕ(x̃) ' ϕ(ỹ) in (S ∪ T )  F . We shall prove that x̃ ' ỹ in p−1 (S ∪ T ).
As a first case we assume that x ' y in S. Then ϕ(x̃) ' ϕ(ỹ) in S  F because v ' w in F ,
and also ϕ(x̃) = ϕS (x̃) and ϕ(ỹ) = ϕS (ỹ) because x, y ∈ V (S). Now, as ϕS (x̃) ' ϕS (ỹ) in
S  F and ϕS : p−1 (S) → S  F is an isomorphism, x̃ ' ỹ in p−1 (S) ≤ p−1 (S ∪ T ).
Assume then that x 6' y in S, so x ' y in T and ϕ(x̃) ' ϕ(ỹ) in T  F ≤ (S ∪ T )  F
because v ' w in F . We claim now that ϕT (x̃) ' ϕT (ỹ) in T  F . Let z̃ be one of x̃, ỹ. If
z∈ / S, ϕ(z̃) = ϕT (z̃) and, if z ∈ S, ϕ(z̃) = ϕS (z̃) ≈ ϕT (z̃) in (S ∪ T )  F . In both cases,
ϕ(z̃) ≈ ϕT (z̃) in (S ∪ T )  F . Then in (S ∪ T )  F we have ϕT (x̃) ≈ ϕ(x̃) ' ϕ(ỹ) ≈ ϕT (ỹ)
and hence ϕT (x̃) ' ϕT (ỹ) in (S ∪ T )  F . Since x ' y in T , it follows that ϕT (x̃) ' ϕT (ỹ)
also in T  F as claimed. Once again, since ϕT : p−1 (T ) → T  F is an isomorphism, x̃ ' ỹ
in p−1 (T ) ≤ p−1 (S ∪ T ). 

The star of a vertex x ∈ B is the subgraph of B consisting of all edges incident to x (and
their vertices). The following is now immediate:
Corollary 2.9. [30]: Let p : E → B be a projection with fiber F . Then (E, B, F, p) is a
graph bundle if and only the star of each vertex of B is (p, F )-trivial. 

3. Triangular Graph Bundles

We observed previously that (E, B, F, p) is a graph bundle if and only if p : E → B is

a projection and every complete subgraph C ≤ B of order 1 or 2 is (p, F )-trivial. Thus,
enlarging the notion of “locality” by substituting “order 1, 2 or 3” for “order 1 or 2” is just
taking this one step further:
Definition 3.1. A triangular graph bundle is a strong graph bundle (E, B, F, p) for which
every triangle, considered as a subgraph xyz ≤ B is (p, F )-trivial.
Trivial bundles are always triangular because every subgraph of B is (π1 , F )-trivial for the
strong product projection π1 : B  F → B. Triangular covering maps were characterized
8 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

in [25] as being just the local isomorphisms, i.e. those graph morphisms p : G̃ → G for which
all the restrictions p| : NG̃ [x̃] → NG [p(x̃)] are isomorphisms. It is easy to see that triangular
covering maps with connected base graph are the same as triangular graph bundles with
connected base and discrete (i.e. without edges) fiber.
Therefore, triangular graph bundles are a common generalization of both strong products
and triangular covering maps, and we shall see in Section 4 that they are just as well suited
to the study of clique graphs as those particular cases.
We now prove that once triangles are admitted into the notion of locality, complete subgraphs
of all sizes get in, and even closed neighborhoods. Indeed, triangular graph bundles can be
defined by means of several different notions of locality.
Theorem 3.2. For a projection p : E → B with fiber F the following are equivalent:
(1) (E, B, F, p) is a triangular graph bundle.
(2) Every complete subgraph of B is (p, F )-trivial.
(3) Every clique of B is (p, F )-trivial.
(4) Every closed neighborhood in B is (p, F )-trivial.

Proof. That (4) ⇒ (3) ⇒ (2) ⇒ (1) follows from Lemma 2.1. Let us prove (1) ⇒ (4).
Let (E, B, F, p) be a triangular graph bundle, so that any vertex, edge or triangle of B is
(p, F )-trivial. Taking x ∈ V (B), we shall prove that N [x] is (p, F )-trivial. Let T1 = {x} and
let T2 , T3 , . . . , Tr be all the edges and triangles of B which contain x. Clearly N [x] = ∪ri=1 Ti .
If r = 1, N [x] = {x} is indeed (p, F )-trivial. Otherwise, note that for all s ∈ {2, 3, . . . , r}
(∪s−1
i=1 Ti )∩Ts is non-empty and connected. It follows by reiterated application of Theorem 2.8
that ∪si=1 Ti is (p, F )-trivial and, in particular, N [x] = ∪ri=1 Ti is (p, F )-trivial. 

Not all graph bundles are triangular, see Figure 1(a). Some further notions of locality lead
still to triangular graph bundles, but there are others that do not, as for example adding
4-cycles to vertices, edges and triangles, see Figure 1(b).

1 3 1 3

2 2 2 2

3 1 3 1
(a) (b)

Figure 1. Two examples. (Identify vertices with equal labels in each graph).
(a) A non-triangular graph bundle with base C3 and fiber P3 . (b) A triangular
graph bundle with base C4 and fiber P3 where B = C4 is not (p, F )-trivial

As another example of the use of Theorem 2.8 to produce new kinds of locality leading to
triangular graph bundles, we present the following result which will be used in Section 4.
Theorem 3.3. If (E, B, F, p) is a triangular graph bundle, then the union of any three
pairwise intersecting cliques q1 , q2 and q3 of B is (p, F )-trivial.
 ON STRONG GRAPH BUNDLES 9

Proof. If q1 ∩q2 ∩q3 6= ∅, then q1 ∪q2 ∪q3 ≤ N [x] for some x, and q1 ∪q2 ∪q3 is (p, F )-trivial by
Theorem 3.2. Otherwise, q1 ∩q2 ∩q3 = ∅. Note that in this case (q1 ∪q2 )∩q3 is not connected
and hence we can not apply Theorem 2.8 directly. Let x ∈ V (q1 ∩ q2 ), y ∈ V (q2 ∩ q3 ) and
z ∈ V (q1 ∩q3 ) and let T be the triangle induced by {x, y, z}. Then q1 ∪q2 ∪q3 = T ∪q1 ∪q2 ∪q3
is (p, F )-trivial by Theorem 2.8, since all the required intersections T ∩ q1 , (T ∪ q1 ) ∩ q2 and
(T ∪ q1 ∪ q2 ) ∩ q3 are connected. 

4. Clique Graphs

If p : E → B is any morphism and q is a clique of E, p(V (q)) = {p(x) | x ∈ V (q)} induces a

complete subgraph in B. Then we can define a morphism K(p) : K(E) → K(B) by selecting
for every clique q ∈ V (K(E)) a fixed clique K(p)(q) ∈ V (K(B)) containing p(V (q)).
Furthermore, when p is, for instance, a local isomorphism no choosing is necessary as p(V (q))
already induces a clique in B. The same will happen for the projection of a triangular graph
bundle. It is well known, and easy to see, that the cliques of a product B  F are precisely
the products of cliques: V (K(B F )) = {q1 q2 | q1 ∈ V (K(B)), q2 ∈ V (K(F ))}. Therefore,
for the trivial graph bundle with projection π1 : B  F → B the image of a clique is already
a clique. But then the same is true for any triangular graph bundle, as it is trivial over the
image of any clique of the total graph.
It is known that the clique operator preserves triangular covering maps [25] and strong
products [24, 31]. The next theorem extends and unifies both results, and the theorem
following it does the same for the corresponding techniques for proving clique convergence
or divergence.
Theorem 4.1. If (E, B, F, p) is a triangular graph bundle, then (K(E), K(B), K(F ), K(p))
is also a triangular graph bundle.
Proof. Let q be any clique of B, i.e. q ∈ V (K(B)). Let S be any (p, F )-trivial subgraph of B
containing q (which exists by Theorem 3.2) and let ϕS : p−1 (S) → S  F be a trivialization.
Then any clique qE of E satisfying K(p)(qE ) = q must be a subgraph of p−1 (q). Note that
the restriction of ϕS to p−1 (q) is an isomorphism onto its image ϕS : p−1 (q) → q  F . Then
the set of cliques of p−1 (q) is precisely {ϕ−1
S (q  qF ) | qF ∈ V (K(F ))}. All these cliques of
−1
p (q) are also cliques of E. Therefore
V (K(p)−1 (q)) = {ϕ−1
S (q  qF ) | qF ∈ V (K(F ))}.

Given ϕ−1 −1 0 −1 −1 −1 0
S (q  qF ), ϕS (q  qF ) ∈ V (K(p) (q)), we have ϕS (q  qF ) ' ϕS (q  qF ) ⇐⇒
ϕ−1 −1 0 0 0
S (q  qF ) ∩ ϕS (q  qF ) 6= ∅ ⇐⇒ (q  qF ) ∩ (q  qF ) 6= ∅ ⇐⇒ qF ∩ qF 6= ∅ ⇐⇒
0 −1 ∼
qF ' qF in K(F ). Hence K(p) (q) = {q}  K(F ) and therefore {q} is (K(p), K(F ))-
trivial. Note in particular that K(p) is vertex-surjective and that the required trivialization
ϕ{q} : K(p)−1 ({q}) → {q}  K(F ) is ϕ{q} = K(ϕS ).
Now let q1 , q2 , q3 ∈ V (K(B)) with qi ∩ qj 6= ∅ for all i, j, so {q1 , q2 , q3 } induces a triangle ∆
in K(B). Let S = q1 ∪ q2 ∪ q3 , which is (p, F )-trivial by Theorem 3.3. By the previous
argument we know, for i = 1, 2, 3, that:
V (K(p)−1 (qi )) = {ϕ−1
S (qi  qF ) | qF ∈ V (K(F ))}
10 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

Given ϕ−1 −1 0 −1 −1 −1
S (qi  qF ), ϕS (qj  qF ) ∈ V (K(p) (∆)), we have ϕS (qi  qF ) ' ϕS (qj  qF )
0
−1 −1
⇐⇒ ϕS (qi  qF ) ∩ ϕS (qj  qF0 ) 6= ∅ ⇐⇒ (qi  qF ) ∩ (qj  qF0 ) 6= ∅ ⇐⇒ qF ∩ qF0 6= ∅
⇐⇒ qF ' qF0 in K(F ). It follows that K(p)−1 (∆) ∼ = ∆  K(F ). Hence any triangle
∆ ≤ K(B) is (K(p), K(F ))-trivial. We point out, as before, that the required trivialization
ϕ∆ : K(p)−1 (∆) → ∆  K(F ) is ϕ∆ = K(ϕS ).
The case of an edge q1 q2 ∈ E(K(B)) is entirely analogous to the previous one. Here is where
we readily see that K(p) : K(E) → K(B) is edge-surjective. The commutativity property
K(p) = π1 ◦ ϕT in all three cases (when T is a vertex, and edge or a triangle of K(B)) is
trivially true. 

Theorem 4.2. If (E, B, F, p) is a triangular graph bundle, E is K-divergent if and only if

at least one of B and F is K-divergent.

Proof. As p−1 (x) ∼ = F for each x ∈ V (B), it follows that |E| = |B| · |F |. By Theorem 4.1
we also have |K n (E)| = |K n (B)| · |K n (F )| for all n, and the result follows. 

Acknowledgment We thank Professor Luis Montejano Peimbert for asking, after a talk
by one of us about triangular covering maps and clique graphs, whether something as our
Theorem 4.1 was true. We also thank the anonymous referee, whose apt comments helped
us improve the presentation of this paper.

References

[1] I. Banič, R. Erveš and J. Žerovnik. The edge fault-diameter of Cartesian graph bundles. European J. Combin.
30 (2009) 1054–1061. http://dx.doi.org/10.1016/j.ejc.2008.09.004.
[2] I. Banič and J. Žerovnik. Fault-diameter of Cartesian graph bundles. Inform. Process. Lett. 100 (2006) 47–51.
http://dx.doi.org/10.1016/j.ipl.2006.05.010.
[3] I. Banič and J. Žerovnik. Wide diameter of Cartesian graph bundles. Discrete Math. 310 (2010) 1697–1701.
http://dx.doi.org/10.1016/j.disc.2009.11.024.
[4] C.F. Bornstein and J.L. Szwarcfiter. On clique convergent graphs. Graphs Combin. 11 (1995) 213–220.
[5] S. Cooper and S. Prassidis. Zeta functions of infinite graph bundles. Linear Multilinear Algebra 58 (2010)
185–201. http://dx.doi.org/10.1080/03081080801928084.
[6] C.P. de Mello, A. Morgana and M. Liverani. The clique operator on graphs with few P4 ’s. Discrete Appl. Math.
154 (2006) 485–492.
[7] R. Erveš and J. Žerovnik. Mixed fault diameter of Cartesian graph bundles. Discrete Appl. Math. 161 (2013)
1726–1733. http://dx.doi.org/10.1016/j.dam.2011.11.020.
[8] R. Erveš and J. Žerovnik. Improved upper bounds for vertex and edge fault diameters of Cartesian graph bundles.
Discrete Appl. Math. 181 (2015) 90–97. http://dx.doi.org/10.1016/j.dam.2014.08.025.
[9] F. Escalante. Über iterierte Clique-Graphen. Abh. Math. Sem. Univ. Hamburg 39 (1973) 59–68.
[10] M.E. Frías-Armenta, F. Larrión, V. Neumann-Lara and M.A. Pizaña. Edge contraction and
edge removal on iterated clique graphs. Discrete Applied Mathematics 161 (2013) 1427 – 1439.
http://www.sciencedirect.com/science/article/pii/S0166218X13000553.
[11] S. Hazan and V. Neumann-Lara. Fixed points of posets and clique graphs. Order 13 (1996) 219–225.
[12] S. Hong, J.H. Kwak and J. Lee. Bipartite graph bundles with connected fibres. Bull. Austral. Math. Soc. 59
(1999) 153–161. http://dx.doi.org/10.1017/S0004972700032718.
[13] I. Hrastnik Ladinek and J. Žerovnik. Cyclic bundle Hamiltonicity. Int. J. Comput. Math. 89 (2012) 129–136.
http://dx.doi.org/10.1080/00207160.2011.638375.
[14] I. Hrastnik Ladinek and J. Žerovnik. On connectedness and hamiltonicity of direct graph bundles. Math. Commun.
17 (2012) 21–34.
[15] W. Imrich, T. Pisanski and J. Žerovnik. Recognizing Cartesian graph bundles. Discrete
Math. 167/168 (1997) 393–403. 15th British Combinatorial Conference (Stirling, 1995).
http://dx.doi.org/10.1016/S0012-365X(96)00242-7.
 ON STRONG GRAPH BUNDLES 11

[16] D. Kim, H.K. Kim and J. Lee. Generalized characteristic polynomials of graph bundles. Linear Algebra Appl.
429 (2008) 688–697. http://dx.doi.org/10.1016/j.laa.2008.03.023.
[17] H.K. Kim and J.Y. Kim. Characteristic polynomials of graph bundles with productive fibres. Bull. Korean Math.
Soc. 33 (1996) 75–86.
[18] S. Klavžar and B. Mohar. The chromatic numbers of graph bundles over cycles. Dis-
crete Math. 138 (1995) 301–314. 14th British Combinatorial Conference (Keele, 1993).
http://dx.doi.org/10.1016/0012-365X(94)00212-2.
[19] J.H. Kwak and Y.S. Kwon. Characteristic polynomials of graph bundles having voltages in a dihedral group.
Linear Algebra Appl. 336 (2001) 99–118. http://dx.doi.org/10.1016/S0024-3795(01)00301-9.
[20] J.H. Kwak and J. Lee. Isomorphism classes of graph bundles. Canad. J. Math. 42 (1990) 747–761.
http://dx.doi.org/10.4153/CJM-1990-039-3.
[21] J.H. Kwak, J. Lee and M.Y. Sohn. Isoperimetric numbers of graph bundles. Graphs Combin. 12 (1996) 239–251.
http://dx.doi.org/10.1007/BF01858458.
[22] J.H. Kwak, J. Lee and M.Y. Sohn. Bartholdi zeta functions of graph bundles having regular fibers. European J.
Combin. 26 (2005) 593–605. http://dx.doi.org/10.1016/j.ejc.2004.05.002.
[23] J.H. Kwak, Y.S. Park and I. Sato. Weighted complexities of graph products and bundles. European J. Combin.
28 (2007) 228–245. http://dx.doi.org/10.1016/j.ejc.2005.07.008.
[24] F. Larrión and V. Neumann-Lara. A family of clique divergent graphs with linear growth. Graphs Combin. 13
(1997) 263–266.
[25] F. Larrión and V. Neumann-Lara. Locally C6 graphs are clique divergent. Discrete Math. 215 (2000) 159–170.
[26] F. Larrión, M.A. Pizaña and R. Villarroel-Flores. The clique behavior of circulants with three small jumps. Ars
Combinatoria 113A (2014) 147–160.
[27] M.C. Lin, F.J. Soulignac and J.L. Szwarcfiter. The clique operator on circular-arc graphs. Discrete Appl. Math.
158 (2010) 1259–1267. http://dx.doi.org/10.1016/j.dam.2009.01.019.
[28] M. Liverani, A. Morgana and C.P. de Mello. The K-behaviour of p-trees. Ars Combin. 83 (2007) 33–45.
[29] M. Matamala and J. Zamora. A new family of expansive graphs. Discrete Appl. Math. 156 (2008) 1125–1131.
[30] B. Mohar, T. Pisanski and M. Škoviera. The maximum genus of graph bundles. European J. Combin. 9 (1988)
215–224. http://dx.doi.org/10.1016/S0195-6698(88)80012-X.
[31] V. Neumann-Lara. On clique-divergent graphs. In Problèmes combinatoires et théorie des graphes (Colloq.
Internat. CNRS, Univ. Orsay, Orsay, 1976), volume 260 of Colloq. Internat. CNRS, pages 313–315. CNRS,
Paris, 1978.
[32] V. Neumann-Lara. Clique divergence in graphs. In L. Lovász and V.T. Sós, editors, Algebraic methods in graph
theory, Coll. Math. Soc. János Bolyai, vol. 25 Szeged, pages 563–569. North-Holland, Amsterdam, 1981.
[33] T. Pisanski, J. Shawe-Taylor and J. Vrabec. Edge-colorability of graph bundles. J. Combin. Theory Ser. B 35
(1983) 12–19. http://dx.doi.org/10.1016/0095-8956(83)90076-X.
[34] T. Pisanski and J. Žerovnik. Hamilton cycles in graph bundles over a cycle with tree as a fibre. Discrete Math.
309 (2009) 5432–5436. http://dx.doi.org/10.1016/j.disc.2008.11.037.
[35] T. Pisanski, B. Zmazek and J. Žerovnik. An algorithm for k-convex closure and an application. Int. J. Comput.
Math. 78 (2001) 1–11. http://dx.doi.org/10.1080/00207160108805092.
[36] E. Prisner. Graph dynamics. Longman, Harlow, 1995.
[37] M. Requardt. (Quantum) spacetime as a statistical geometry of lumps in random networks. Classical Quantum
Gravity 17 (2000) 2029–2057.
[38] M. Requardt. Space-time as an order-parameter manifold in random networks and the emergence of physical
points. In Quantum theory and symmetries (Goslar, 1999), pages 555–561. World Sci. Publ., River Edge, NJ,
2000.
[39] M. Requardt. A geometric renormalization group in discrete quantum space-time. J. Math. Phys. 44 (2003)
5588–5615.
[40] M.Y. Sohn and J. Lee. Characteristic polynomials of some weighted graph bundles and its application to links.
Internat. J. Math. Math. Sci. 17 (1994) 503–510. http://dx.doi.org/10.1155/S0161171294000748.
[41] E.H. Spanier. Algebraic topology. Springer-Verlag, New York, 1981. Corrected reprint.
[42] J.L. Szwarcfiter. A survey on clique graphs. In B.A. Reed and C. Linhares-Sales, editors, Recent advances in
algorithms and combinatorics, volume 11 of CMS Books Math./Ouvrages Math. SMC, pages 109–136. Springer,
New York, 2003.
[43] J. Žerovnik. On recognition of strong graph bundles. Math. Slovaca 50 (2000) 289–301.
[44] B. Zmazek and J. Žerovnik. On recognizing Cartesian graph bundles. Discrete Math. 233 (2001) 381–391. Graph
theory (Prague, 1998). http://dx.doi.org/10.1016/S0012-365X(00)00254-5.
[45] B. Zmazek and J. Žerovnik. Algorithm for recognizing Cartesian graph bundles. Discrete Appl. Math. 120
(2002) 275–302. Sixth Twente Workshop on Graphs and Combinatorial Optimization (Enschede, 1999).
http://dx.doi.org/10.1016/S0166-218X(01)00296-7.
12 F. LARRIÓN, M.A. PIZAÑA, AND R. VILLARROEL-FLORES

[46] B. Zmazek and J. Žerovnik. Unique square property and fundamental factorizations of graph bundles. Dis-
crete Math. 244 (2002) 551–561. Algebraic and topological methods in graph theory (Lake Bled, 1999).
http://dx.doi.org/10.1016/S0012-365X(01)00106-6.
[47] B. Zmazek and J. Žerovnik. On domination numbers of graph bundles. J. Appl. Math. Comput. 22 (2006) 39–48.
http://dx.doi.org/10.1007/BF02896459.

F. Larrión (paco@math.unam.mx)
Instituto de Matemáticas, Universidad Nacional Autónoma de México.
México 04510 D.F. MEXICO

M.A. Pizaña (map@xanum.uam.mx, http://xamanek.izt.uam.mx/map)

Universidad Autónoma Metropolitana, Depto. de Ingeniería Eléctrica.
Av. San Rafael Atlixco 186. Col Vicentina. Del. Iztapalapa.
México 09340 D.F. MEXICO

R. Villarroel-Flores (rafaelv@uaeh.edu.mx)
Área Académica de Matemáticas y Física,
Universidad Autónoma del Estado de Hidalgo,
Carr. Pachuca-Tulancingo km. 4.5,
Pachuca 42184 Hgo. MEXICO