---
title: Generalised Pancake Graphs
url: https://www.emergentmind.com/topics/generalised-pancake-graphs
type: topic
---

# Generalised Pancake Graphs

Generalised pancake graphs are Cayley graphs of the generalized symmetric group \(S(m,n)=C_m\wr S_n\), equivalently \(\mathbb{Z}_m\wr S_n\), generated by prefix reversals. In the cited literature this family appears as both \(\mathcal{P}_m(n)\) and \(\mathbb{P}_m(n)\). Its vertices are colored permutations, its order is \(m^n n!\), and its two most studied special cases are the classical pancake graph at \(m=1\) and the burnt pancake graph at \(m=2\). Recent work has developed a comparatively coherent theory of these graphs in four directions: recursive combinatorial structure, cycle and Hamiltonian phenomena, spectral analysis via equitable partitions, and topological embeddings via rotation systems [2204.10494] [2509.09425] [2306.11295].

## 1. Algebraic construction and notation

The generalised pancake graph is defined on the wreath product \(S(m,n)=\mathbb{Z}_m\wr S_n\), whose elements are viewed as colored permutations. Two vertices are adjacent when one is obtained from the other by a prefix reversal of the first \(k\) entries, with a possible color change by \(+1\) or \(-1 \bmod m\) on the entries that are flipped. For \(m=1\) this recovers the classical pancake graph, and for \(m=2\) it gives the burnt pancake graph, where reversing a prefix also changes the sign of every involved entry. For \(m\geq 3\), each prefix length yields two types of flips, usually described as positive and negative color flips [2509.09425].

In the undirected formulation, the graph is the Cayley graph of \(S(m,n)\) generated by prefix reversals and their inverses, sometimes called flips and flops. The resulting graph is vertex-transitive. The 2022 cycle paper further records that it is not edge-transitive in general, although the set of edges with the same label is edge-transitive. A complementary group-theoretic viewpoint treats prefix reversals as generators of \(S_n\) and \(B_n\), and relates the orders of products of generators to cycle lengths in the corresponding Cayley graphs [2204.10494] [1803.01760].

## 2. Recursive architecture

A central structural fact is that \(\mathbb{P}_m(n)\) contains \(mn\) non-overlapping copies of \(\mathbb{P}_m(n-1)\), obtained by fixing the last signed-symbol. This hierarchical decomposition underlies most known inductive proofs on cycle lengths, genus bounds, and fault-tolerant path constructions. The same paper also exhibits an explicit base cycle of length \(2mn\), obtained by successively applying flop-flip combinations across these layers [2204.10494].

For the burnt special case \(BP_n\), the decomposition becomes especially rigid: \(BP_n\) can be partitioned into \(2n\) induced subgraphs \(BP_n^i\), each isomorphic to \(BP_{n-1}\), by fixing the last symbol together with its sign. In the cycle literature, edges labeled \(r_n\) connect these copies; in the connectivity literature, the same partition is described as a cluster structure. The data further records that each vertex has exactly one out-neighbour in another cluster, and that edges between two distinct clusters form a matching. These properties are repeatedly exploited in inductive constructions of Hamiltonian paths, disjoint path covers, and internally edge-disjoint Steiner trees [1808.04890] [2211.05619] [2310.18831].

This recursive architecture has methodological consequences. It enables arguments that build global objects by first constructing cycles, paths, or trees inside copies of \(\mathbb{P}_m(n-1)\) or \(BP_{n-1}\), and then patching them through controlled inter-copy edges. A plausible implication is that recursive decomposition, rather than representation theory, is the dominant structural mechanism currently available for this family.

## 3. Cycle structure, girth, and Hamiltonicity

For \(m\geq 3\), the girth of the undirected generalized pancake graph is
\[
\operatorname{girth}(\mathbb{P}_m(n))=\min\{m,6\}.
\]
This complements the known special cases \(m=1\) and \(m=2\). The parity of \(m\) determines the global cycle spectrum. If \(m>2\) is odd, then \(\mathbb{P}_m(n)\) is \(m'\)-pancyclic, where \(m'=\min\{m,6\}\): it contains cycles of every length from \(m'\) up to \(m^n n!\). If \(m>2\) is even, then \(\mathbb{P}_m(n)\) is \(m'\)-panevencyclic: it contains all even cycle lengths from its girth to a Hamiltonian cycle. The base cases are explicit: \(\mathbb{P}_3(n)\) has all cycle lengths, while \(\mathbb{P}_4(n)\) has all even cycle lengths [2204.10494].

| Regime | Girth | Guaranteed cycle lengths |
|---|---:|---|
| \(m=2\) | \(8\) | every \(\ell\) with \(8\leq \ell \leq 2^n n!\) |
| \(m=3\) | \(3\) | every \(\ell\) with \(3\leq \ell \leq 3^n n!\) |
| \(m=4\) | \(4\) | every even \(\ell\) with \(4\leq \ell \leq 4^n n!\) |
| odd \(m>2\) | \(\min\{m,6\}\) | every length from girth to \(m^n n!\) |
| even \(m>2\) | \(\min\{m,6\}\) | every even length from girth to a Hamiltonian cycle |

The proof strategy in the general case is inductive. Cycles inside copies of \(\mathbb{P}_m(n-1)\) are merged through the base cycle, with resulting lengths of the form
\[
\sum_{i=1}^q (\ell(C_i)-1) + (2mn-q).
\]
This formula is used to cover the required interval of admissible lengths [2204.10494].

The burnt pancake graph is especially well understood. It is Hamiltonian and weakly pancyclic: for \(n\geq 2\), \(BP_n\) contains a cycle of every length \(\ell\) with \(8\leq \ell\leq 2^n n!\). Its girth is \(8\), and the 8-cycles admit a complete classification. Every 8-cycle in \(BP_n\) has one of the canonical forms
\[
r_k r_j r_i r_j r_k r_{k-j+i} r_i r_{k-j+i},\qquad
r_k r_j r_k r_i r_k r_j r_k r_i,\qquad
r_k r_i r_k r_i r_k r_i r_k r_i,
\]
with the parameter restrictions stated in Theorem 4.1 of the paper. The proof is constructive and relies on the recursive structure of \(BP_n\) [1808.04890].

## 4. Spectral properties

The recent spectral theory of generalized pancake graphs is centered on equitable partitions and quotient matrices. For a graph with eigenvalues \(\lambda_1\geq \lambda_2\geq \cdots\), the spectral gap is \(\lambda_1-\lambda_2\); in regular graphs, \(\lambda_1\) equals the degree. The 2025 note proves that
\[
\lambda_1(\mathcal{P}_m(n))-\lambda_2(\mathcal{P}_m(n))<
\begin{cases}
1,& m=2,\\
2,& m\geq 3,
\end{cases}
\]
so the burnt pancake graph has spectral gap strictly less than \(1\), while the generalised pancake graphs with \(m\geq 3\) have spectral gap strictly less than \(2\). The same paper also establishes multiplicity lower bounds for integer eigenvalues when \(m\equiv 0\pmod 4\):
\[
\operatorname{mult}_{\mathcal{P}_m(n)}(2k)\geq
\begin{cases}
3,& 1\leq k\leq n-1,\ k\neq \left\lfloor \frac n2\right\rfloor,\\
2,& k=\left\lfloor \frac n2\right\rfloor.
\end{cases}
\]
These results settle two conjectures of Blanco and Buehrle [2509.09425].

The quotient spectra are explicit. For \(m\geq 3\),
\[
\operatorname{Spec}\big(Q(\mathfrak{P}_{m,n})\big)
=
\biguplus_{k=0}^{m-1}
\operatorname{Spec}\!\left(2D_n+2\cos\frac{2\pi k}{m}E_n\right),
\]
and for \(m=2\),
\[
\operatorname{Spec}\big(Q(\mathfrak{P}_{2,n})\big)
=
\operatorname{Spec}(D_n+E_n)\uplus \operatorname{Spec}(D_n-E_n),
\]
where \(D_n\) is diagonal and \(E_n\) is an upper-triangular \(0\)-\(1\) matrix. The proofs combine equitable partitions, quotient-matrix computations, eigenvalue interlacing, and the divisibility of characteristic polynomials. The same paper explicitly notes why this route is needed: although \(\mathcal{P}_m(n)\) are Cayley graphs, their generating sets are not conjugation-closed, so classical group-character techniques do not apply [2509.09425].

For the burnt pancake graph alone, an earlier result proves that the adjacency spectrum contains every integer in
\[
\{0,1,\ldots,n\}\setminus\left\{\left\lfloor \frac n2\right\rfloor\right\}.
\]
That work constructs a \(2n\times 2n\) quotient matrix
\[
M(\mathbb{BP}_n)=
\begin{pmatrix}
A_n & D_n^\top\\
D_n & C_n
\end{pmatrix},
\]
and also shows that the eigenvalue \(n-1\) has multiplicity at least \(2\) [2408.05349].

## 5. Genus and 2-cell embeddings

The topological theory of generalized pancake graphs is comparatively recent. For the undirected generalized pancake graph \(\mathbb{P}_m(n)\), the first general upper and lower bounds for the orientable genus \(\gamma\) were proved constructively in 2023. For \(m\geq 3\) and \(n\geq 2\), the lower bounds are
\[
\gamma(\mathbb{P}_m(n)) \geq
\begin{cases}
m^{n-1}((m-2)n-m)n!+1,& m=3,4,5,\\[4pt]
\frac16 m^n(2n-3)n!+1,& m\geq 6,
\end{cases}
\]
while the upper bounds are
\[
\gamma(\mathbb{P}_m(n)) \leq
\begin{cases}
m^{n-1}(2mn-2m-n-1)n!+1,& m\ \text{odd},\\[4pt]
m^{n-1}(mn-m-n)n!+1,& m\ \text{even}.
\end{cases}
\]
These estimates are asymptotically tight, and in particular
\[
\gamma(\mathbb{P}_m(n))=\Theta(m^n n n!)
\quad\text{for all } m\geq 1,\ n\geq 2.
\]
The paper also records the example \(\gamma(\mathbb{P}_3(2))=1\), so \(\mathbb{P}_3(2)\) is toroidal and not planar [2306.11295].

For \(n=2\), sharper explicit bounds are given. If \(m\) is odd and \(m=3j\), then
\[
\gamma(\mathbb{P}_m(2))\leq m^2-3m+1,
\]
whereas if \(m\) is odd and \(m\neq 3j\), then
\[
\gamma(\mathbb{P}_m(2))\leq m^2-2m+1.
\]
If \(m\) is even and \(m=3j\), then
\[
\gamma(\mathbb{P}_m(2))\leq m^2-\frac72 m+1,
\]
and if \(m\) is even and \(m\neq 3j\), then
\[
\gamma(\mathbb{P}_m(2))\leq m^2-\frac52 m+1
\]
[2306.11295].

The proofs use rotation systems, or Edmonds' permutation technique, together with custom vertex-labeling algorithms ALCYC and ALGRA. For \(m>2\), prefix reversals are not involutions, unlike the cases \(m=1,2\). The construction arranges incident edges at each vertex so that certain cycles become facial boundaries in a 2-cell embedding, and repeated application of a generator \(r_i\) yields a face whose length is the order \(o(r_i)\) [2306.11295].

## 6. Connectivity, fault tolerance, and related directions

The strongest reliability results in the supplied literature concern the burnt special case \(BP_n\). Its generalized 3-connectivity and generalized 4-connectivity are both exactly \(n-1\):
\[
\kappa_3(BP_n)=n-1,\qquad \kappa_4(BP_n)=n-1.
\]
Thus, for any three vertices, or any four vertices, there exist \(n-1\) internally edge-disjoint trees connecting them. The proofs rely on the cluster decomposition into \(2n\) copies of \(BP_{n-1}\), fan arguments, out-neighbour control, and case analyses according to the distribution of the terminals among clusters [2211.05619] [2310.00878].

Fault-tolerant Hamiltonian and path-cover properties are also known. If \(n\geq 4\) and \(BP_n\) has at most \(n-4\) faulty elements, then for any two terminal pairs \(\{u,v\}\) and \(\{x,y\}\) there exist two vertex-disjoint paths \(P[u,v]\) and \(Q[x,y]\) whose vertices partition \(V(BP_n-F)\); for every \(n\geq 3\), there is a set of \(n-2\) faulty edges or faulty vertices for which such a paired 2-disjoint path cover does not exist. Under a hybrid model combining faulty edges with removals of both end-vertices of matching edges, \(BP_n\) is \((n-2)\)-hybrid fault Hamiltonian and \((n-3)\)-hybrid fault Hamiltonian connected, and both bounds are tight [2310.18831] [2412.17236].

These burnt-pancake results are routinely interpreted in the source papers as evidence for the suitability of pancake-type Cayley graphs as interconnection networks. A plausible implication is that the same recursive toolkit—cluster decomposition, matching-like inter-cluster edges, and inductive Hamiltonicity arguments—should remain useful in broader generalized pancake settings, although the supplied corpus proves such robustness explicitly only for \(m=2\).

Several open-ended directions surround the family. The spectral note restates conjectures about the precise spectral gap for large \(n\) and about possible coincidence between the spectral gap of the full graph and that of the quotient matrix [2509.09425]. The paired 2-DPC paper asks whether the \(n-4\) fault bound can be improved to \(n-3\) in \(BP_n\) [2310.18831]. A related fixed-degree variant, the cubic pancake graph \(\mathrm{Cay}(\mathrm{Sym}_n,\{r_a,r_b,r_c\})\), has recently been studied through the problem of characterizing triples of prefix reversals that generate \(\mathrm{Sym}_n\); that work gives full characterizations for triples containing at least one of \(r_2\), \(r_3\), \(r_{n-2}\), or \(r_{n-1}\), and reports computational data on girth, diameter, and Hamiltonicity for small \(n\) [2511.16959].

Source: https://www.emergentmind.com/topics/generalised-pancake-graphs