1. Introduction
Connectivity of a graph is widely used to evaluate network reliability [1], focusing on the overall connectedness of a graph (defined by the minimum number of vertices whose removal disconnects the graph). Chartrand et al. [2] proposed the concept of generalized
-connectivity, which extends the notion of path connectivity between two vertices to tree connectivity among any
vertices. Since then, generalized
-connectivity has been studied extensively for various graphs and interconnection networks. Chartrand et al. [3] discussed rainbow trees and generalized connectivity in networks; Li et al. [4] [5] determined the generalized connectivity of complete bipartite graphs and complete equipartite 3-partite graphs; Li and Yu [6] studied the generalized 3-connectivity of star graphs and bubble-sort graphs; Li et al. [7] investigated the generalized connectivity of the line graph and the total graph of the complete bipartite graph. As a closely related reliability parameter, the component connectivity of hypercubes and of Cayley graphs generated by transposition trees has been studied in [8] [9].
For a graph
and a
-vertex subset
, an
-tree is a subtree of
that contains all vertices of
. Two
-trees
and
are internally disjoint if 1)
and 2)
. In other words, the two trees share no internal vertices and no edges; they may overlap only on the terminals in
. We denote by
the maximum number of internally disjoint
-trees in
, and the generalized
-connectivity of
is defined as
Clearly,
is a generalization of the classical connectivity
in the sense that
.
Let
be a subset of
. If
is disconnected, then
is a vertex cut of
. Let
be a vertex cut of
. We call the largest component of
a large component, and the other components are small components. A cycle with
vertices is called an
-cycle, and a path with
vertices is called an
-path. Two paths are internally disjoint if they have no internal vertices in common.
Let
and
be the permutation group on
. For a permutation
of
, we define
The
-dimensional pancake graph, denoted by
, is a graph whose vertex set consists of all permutations of
, where two vertices
and
are adjacent if and only if
for some
. The pancake graphs
,
and
are shown in Figure 1. The connectivity and the generalized 4-connectivity of
are given by Akers and Krishnamurthy in [1] and by Kanevsky and Feng in [10], respectively.
Figure 1. The pancake graphs
,
,
.
Proposition 1. [1] The
-dimensional pancake graph
(
) is an
-connected graph.
Proposition 2. [10] The generalized 4-connectivity of the
-dimensional pancake graph
(
) is
.
The
-dimensional hierarchical pancake graph, denoted by
(
), is the graph with vertex set
, where two vertices
and
are adjacent if and only if one of the following three conditions is satisfied: 1)
and
for some
; 2)
,
and
,
; 3)
,
and
,
.
An
-dimensional hierarchical pancake graph
consists of
modules and each module is an
-dimensional pancake graph
, denoted by
, where
.
is an
-regular graph with
vertices and
edges. The modules are connected by additional edges, called crossing edges. For each vertex
in
,
represents the module in which the vertex is located and
identifies the position of
inside its module. For two adjacent vertices
and
in
,
is called the outside neighbor of
if
. Moreover, we denote
for
and
for
. The hierarchical pancake graph
is shown in Figure 2.
Figure 2. The hierarchical pancake graph
.
The hierarchical pancake graph
is a class of hierarchical interconnection network topologies constructed based on pancake graphs
. They are formed by connecting
pancake graph modules through crossing edges and exhibit excellent properties such as
-regularity,
-connectivity, and good fault-tolerance, with promising application prospects in multiprocessor systems [11] [12]. The component (edge) connectivity of pancake graphs was studied by Hua and Yang [13]. For other hierarchical networks, Gu et al. [14] analyzed the fault-tolerance and diagnosability of hierarchical star networks, and Zhao et al. [15] determined the generalized 4-connectivity of hierarchical cubic networks. In [16], the authors determined the connectivity of
.
Proposition 3. [16]
(
) is an
-connected graph.
In this paper, we determine the generalized 4-connectivity of the hierarchical pancake graph
and show that
for
. That is, for any 4-vertex subset
, there exist at least
internally disjoint
-trees in
.
2. Main Results
Before proving the main result, we establish two structural facts about
that will be used throughout.
Lemma 4. The module graph of
, whose vertices are the modules
(
) and whose edges correspond to crossing edges between modules, is the complete graph
. Consequently,
is connected for any module
.
Proof. For any two distinct permutations
, the vertices
and
are adjacent by a crossing edge (condition (2) in the definition of adjacency). Thus, every pair of modules is joined by a crossing edge, and the module graph is complete.
Since the module graph is
, removing any one vertex (module) leaves a complete graph on
vertices, which is connected. Each remaining module
is itself connected (because
is a connected Cayley graph). Therefore
is connected: every module is internally connected, and the inter-module adjacency graph among the remaining modules is connected. □
Lemma 5. (Simultaneous routing lemma.) Let
be a 4-vertex subset with
, where
denotes the number of distinct modules containing vertices of
. Then
.
Proof. Let the modules containing vertices of
be
. By Lemma 4, the module graph is
, which is
-connected. Since
for
, we can choose
internally vertex-disjoint routes in the module graph that connect the relevant modules, where each route uses a distinct set of intermediate modules not containing any vertex of
.
More precisely, for each
, choose a route
in the module graph that connects the modules containing
in a suitable order, passing through intermediate modules
such that: a) the routes
are internally vertex-disjoint in the module graph; b) no intermediate module is one of the modules containing
.
This is possible because the module graph is complete on
vertices, and we only need
disjoint routes among at most 4 terminal modules; there are
spare modules to route through.
For each route
, select the crossing-edge endpoints at each module it passes through. Inside each module on the route, connect the entry endpoint to the exit endpoint by a path. Since each module
is
-connected (Proposition 1), and the entry/exit vertices for different routes
lie in different modules (by the internal disjointness of the routes in the module graph), these internal paths can be chosen independently and are internally disjoint across different
.
Inside the modules that contain terminals of
, connect each terminal vertex to the appropriate route endpoint by an internal path. Again, by the
-connectivity of
, these attachment paths for different routes can be chosen to be internally disjoint.
For each
, the union of the inter-module crossing edges, the internal paths along the route, and the terminal attachments forms a tree
that contains all vertices of
. The family
is internally disjoint: inter-module routes are vertex-disjoint by construction in the module graph, and internal paths within each module are disjoint by the connectivity of
and the fact that different routes use different vertices in shared modules.
Therefore
. □
In this section, we prove two lemmas showing that
for
.
Lemma 6. For the hierarchical pancake graph
,
for
.
Proof. Let
be the hierarchical pancake graph for
and
with
. We denote by
the number of distinct modules in
that contain vertices of
. Clearly,
. We verify
for each case.
Case 1.
.
That is, all vertices of
lie in a single module. Suppose
for some module
. We construct
internally disjoint
-trees as follows.
First, inside the module
, by Proposition 2,
. Hence there exist
internally disjoint
-trees
entirely contained in
.
It remains to construct one more
-tree
that is internally disjoint from
.
Let
be the outside neighbor of
for
. By the adjacency rules of
, each vertex has exactly one outside neighbor, and
. By Lemma 4, the graph
is connected.
Since
is connected and contains
, there exists a subtree
in
that contains all four vertices
. Such a tree exists: in a connected graph, for any finite set of vertices, there is a minimal connected subgraph containing them, which is a tree.
Define
We claim that
is an
-tree. Indeed,
is a tree containing
. Adding the four edges
and the vertices
attaches each
as a leaf to
. The resulting graph is connected and acyclic:
is acyclic, and each new edge connects a new leaf vertex to
, so no cycle is created. Thus
is a tree containing all vertices of
.
We now verify internal disjointness.
Vertex-disjointness: The internal vertices of
all lie in
(outside
). The internal vertices of
all lie in
(since
and the only vertices of
in
are
). Hence
for
.
Edge-disjointness: The edges of
all lie in
. The edges of
are either edges of
(lying outside
) or the crossing edges
(which are not inside
). Hence
for
.
Therefore
are
internally disjoint
-trees.
Case 2.
.
When
, or 4, Lemma 5 (the simultaneous routing lemma) directly gives
. The detailed construction proceeds by routing
internally disjoint tree families through the complete module graph, using distinct intermediate modules for each tree, and attaching the terminals within each terminal module by internally disjoint paths made possible by the
-connectivity of
.
We illustrate the construction for
; the cases
and
are analogous. Suppose
with
. By Lemma 4, the module graph is
, so there exist
internally vertex-disjoint paths between the module set
by way of distinct intermediate modules. For each such path, connect the terminals within
and
to the crossing-edge endpoints by internally disjoint paths (possible because
is
-connected). The result is
internally disjoint
-trees.
In conclusion, for every possible distribution of the 4-vertex subset
(i.e.,
), we have
. By the definition of generalized 4-connectivity,
This completes the proof of Lemma 6. □
Lemma 7. For the hierarchical pancake graph
,
for
.
Proof. It is a standard upper bound for generalized 4-connectivity that for any graph
with minimum degree
,
(see, e.g., [2] [3]). We briefly recall the argument. Let
be a minimum-degree vertex and choose three of its neighbors
. Set
. In any family of internally disjoint
-trees, each tree uses at least one edge incident to
. Since the trees are edge-disjoint at
(internal disjointness includes edge-disjointness), there can be at most
such trees. However, the edges
are internal to
, and in any tree where
is a leaf connected via, say,
, the vertices
must connect to the rest of the tree without passing through
. This forces the terminal structure to consume one additional edge beyond the
incident edges, yielding the sharper bound
. Minimizing over
gives
.
The hierarchical pancake graph
is
-regular, so
. Applying the standard bound,
For completeness, we also give a direct argument specific to
that confirms this bound. Fix a module
. Choose a vertex
and three of its neighbors within the module, say
. Let
.
Consider any family of internally disjoint
-trees. Each tree must use at least one of the
edges incident to
. Since the trees are internally disjoint (including edge-disjointness), at most
trees can be formed. To see that at most
are possible, note that among the
edges incident to
, three are the internal edges
within
. In the extremal case of
trees,
would be a leaf in every tree, each using a distinct incident edge. But then the tree using
must contain
connected to the rest of the tree without passing through
, requiring those vertices to have spare internal edges. Counting the degree budget at the terminal vertices
(each of which also has
incident edges, some of which are consumed by their role as terminals in the various trees) shows that one edge of
must remain unused across the whole family, leaving at most
trees.
Thus
, and by definition,
This completes the proof of Lemma 7. □
By Lemma 6 and Lemma 7, we directly obtain the main result of this paper.
Theorem 8. For the hierarchical pancake graph
with
,
3. Conclusion
Combining the lower bound and upper bound results, we arrive at the exact value of the generalized 4-connectivity of
:
Our result shows how the modular structure of
determines its generalized 4-connectivity. For
, we obtain
, which is consistent with the fact that
is 3-connected; this confirms that the generalized 4-connectivity is a natural and useful extension of the classical connectivity to networks with four terminals.
Future research could study the generalized
-connectivity of
for
, as well as the generalized 4-connectivity of other hierarchical networks such as hierarchical star graphs and hierarchical cube-connected cycles.
Acknowledgements
The author would like to thank the anonymous reviewers for their valuable comments and suggestions that improved the quality of the article. Sincere thanks to the members of JAMP for their professional performance.