Paired Many-to-Many 2-Disjoint Path Cover

• The paper’s main contribution is the formulation of a spanning path-cover problem where two vertex-disjoint paths connect preset endpoints and together cover all vertices even in faulty settings.
• The methodology relies on recursive decompositions and splicing techniques in symmetric networks like burnt pancake graphs, balanced hypercubes, and Johnson graphs.
• Key results establish fault-tolerant guarantees through Hamiltonian connectivity, parity constraints, and sharp degree conditions, with open questions on optimal fault thresholds.

Searching arXiv for the cited paper and closely related work on paired many-to-many 2-disjoint path covers. Searching for (Dvořák et al., 2023). Searching for "Paired 2-disjoint path covers of burnt pancake graphs with faulty elements". Paired many-to-many 2-disjoint path cover is a spanning path-cover problem in which, for two disjoint 2-vertex sets $S=\{u,x\}$ and $T=\{v,y\}$, one seeks two vertex-disjoint paths $P_{uv}$ and $P_{xy}$ with endpoints $(u,v)$ and $(x,y)$ such that $V(G)=V(P_{uv})\cup V(P_{xy})$ (Liu et al., 21 Jul 2025). In the terminology used for burnt pancake graphs, a paired 2-disjoint path cover is a set $\{P[u,v],Q[x,y]\}$ of two vertex-disjoint paths whose vertex sets partition the graph, and in faulty settings the paths are required to lie in the residual graph after deleting faulty vertices and edges (Dvořák et al., 2023). The topic is closely connected to Hamiltonian connectivity, Hamiltonian laceability, linkage, and fault-tolerant routing, and it has been studied in interconnection-network families such as balanced hypercubes, burnt pancake graphs, Johnson graphs, and BCube, as well as in directed and shortest-path routing variants (Ma et al., 2024).

1. Definition and variant structure

In its standard undirected form, the paired many-to-many 2-disjoint path cover problem fixes the pairing in advance: $u$ must be joined to $v$, and $T=\{v,y\}$0 must be joined to $T=\{v,y\}$1 (Liu et al., 21 Jul 2025). This distinguishes it from unpaired many-to-many path cover problems, where a permutation of the terminal sets may be chosen by the construction. The balanced-hypercube literature makes this distinction explicit: a paired $T=\{v,y\}$2-DPC uses the identity permutation, whereas an unpaired $T=\{v,y\}$3-DPC allows an arbitrary permutation $T=\{v,y\}$4 between the two endpoint sets (Lü et al., 2019).

The fault-tolerant formulation deletes a set $T=\{v,y\}$5 of faulty elements and asks for the same two-path vertex partition in $T=\{v,y\}$6. For burnt pancake graphs, the residual graph is defined by removing faulty vertices, faulty edges, and all edges incident with faulty vertices; a fault-free paired 2-DPC is then a paired 2-DPC in that residual graph (Dvořák et al., 2023). In edge-fault-only models, such as the balanced hypercube and BCube settings, the same cover condition is imposed after deleting the faulty edges (Lü, 2018).

In bipartite interconnection networks, endpoint placement is constrained by parity. For $T=\{v,y\}$7, the graph is bipartite, and the literature requires $T=\{v,y\}$8 to lie in one partite set and $T=\{v,y\}$9 in the other; the same parity condition underlies the many-to-many constructions in balanced hypercubes more generally (Lü, 2018). Directed analogues replace undirected paths by directed paths. In the digraph formulation, a paired many-to-many 2-DDPC is a pair of vertex-disjoint directed paths $P_{uv}$0 such that $P_{uv}$1 is an $P_{uv}$2–$P_{uv}$3 path, $P_{uv}$4 is an $P_{uv}$5–$P_{uv}$6 path, and $P_{uv}$7 (Ma et al., 2024).

A related global property is paired 2-coverability: a graph is paired 2-coverable if every choice of two disjoint 2-vertex sets $P_{uv}$8 and $P_{uv}$9 admits such a cover (Liu et al., 21 Jul 2025). This property is stronger than the existence of a single Hamiltonian path and stronger than unpaired many-to-many coverage guarantees.

2. Structural prerequisites and global degree principles

A general structural principle is that paired 2-coverability implies Hamilton-connectedness. For Johnson graphs, it is stated explicitly that if a graph is paired 2-coverable, then it must be Hamilton-connected, but the reverse is not true (Liu et al., 21 Jul 2025). The same paper gives a counterexample to the converse: a Hamilton-connected graph obtained from the 3-dimensional hypercube $P_{xy}$0 by adding the edges $P_{xy}$1 and $P_{xy}$2 is not paired 2-coverable for the terminal choice $P_{xy}$3, $P_{xy}$4, $P_{xy}$5, $P_{xy}$6 (Liu et al., 21 Jul 2025). This separates spanning two-path decompositions from ordinary Hamiltonian connectivity.

Many existence proofs use Hamiltonian connectivity or Hamiltonian laceability as a backbone. In balanced hypercubes, Hamiltonian laceability is the paired 1-disjoint path cover and serves as a local ingredient for paired 2-DPC and paired 3-DPC constructions (Guo et al., 2019). In burnt pancake graphs, Kaneko’s results are used in the form: if $P_{xy}$7, then $P_{xy}$8 is Hamiltonian, and if $P_{xy}$9, then $(u,v)$0 is Hamiltonian-connected (Dvořák et al., 2023). In Johnson graphs, Alspach’s Hamilton-connectedness theorem is the starting point for the proof that all $(u,v)$1 are paired 2-coverable (Liu et al., 21 Jul 2025).

For digraphs, the sharp global condition currently available is Ore-type. If $(u,v)$2 is a digraph of order $(u,v)$3 and $(u,v)$4 for each missing arc $(u,v)$5, then $(u,v)$6 is paired many-to-many 2-coverable (Ma et al., 2024). The proof augments the digraph with one auxiliary vertex $(u,v)$7, applies the Overbeck-Larisch Hamiltonian-connectedness criterion in the augmented digraph, and then splits a Hamiltonian path at $(u,v)$8 into the required two spanning directed paths (Ma et al., 2024). The bound is sharp: there exists a digraph on $(u,v)$9 vertices with $(x,y)$0 for each missing arc that is not paired many-to-many 2-coverable because every $(x,y)$1–$(x,y)$2 path and every $(x,y)$3–$(x,y)$4 path are forced through a single vertex $(x,y)$5 (Ma et al., 2024).

3. Burnt pancake graphs and mixed faulty elements

The most detailed mixed-fault theorem currently in the supplied literature concerns the burnt pancake graph $(x,y)$6, the Cayley graph of the hyperoctahedral group $(x,y)$7 generated by signed prefix reversals $(x,y)$8 (Dvořák et al., 2023). A signed permutation $(x,y)$9 has $V(G)=V(P_{uv})\cup V(P_{xy})$0, and

$V(G)=V(P_{uv})\cup V(P_{xy})$1

The graph $V(G)=V(P_{uv})\cup V(P_{xy})$2 has

$V(G)=V(P_{uv})\cup V(P_{xy})$3

It admits a canonical decomposition into $V(G)=V(P_{uv})\cup V(P_{xy})$4 induced subgraphs $V(G)=V(P_{uv})\cup V(P_{xy})$5, $V(G)=V(P_{uv})\cup V(P_{xy})$6, each isomorphic to $V(G)=V(P_{uv})\cup V(P_{xy})$7. For $V(G)=V(P_{uv})\cup V(P_{xy})$8, the inter-subgraph edge set $V(G)=V(P_{uv})\cup V(P_{xy})$9 forms a matching, and

$\{P[u,v],Q[x,y]\}$0

Every vertex $\{P[u,v],Q[x,y]\}$1 has a unique out-neighbor $\{P[u,v],Q[x,y]\}$2 outside $\{P[u,v],Q[x,y]\}$3 via its $\{P[u,v],Q[x,y]\}$4-edge (Dvořák et al., 2023).

The main theorem states that if $\{P[u,v],Q[x,y]\}$5 and $\{P[u,v],Q[x,y]\}$6 with $\{P[u,v],Q[x,y]\}$7, then for any two pairs $\{P[u,v],Q[x,y]\}$8 and $\{P[u,v],Q[x,y]\}$9 of distinct vertices there exist vertex-disjoint paths $u$0 and $u$1 in $u$2 with $u$3 and $u$4 (Dvořák et al., 2023). Faulty elements may be vertices and/or edges, so the bound $u$5 is a mixed-fault guarantee.

The construction is inductive. If faults are spread so that each $u$6 has at most $u$7 faults, the induction hypothesis supplies local 2-DPCs in each subgraph, and the matchings $u$8 permit global splicing. In the worst-case concentration, one subgraph $u$9 may contain exactly $v$0 faults while all others and all out-edges are fault-free; the proof then performs a detailed case analysis over all terminal distributions across one, two, three, or four subgraphs (Dvořák et al., 2023). A key lemma ensures out-neighbor separation: if two vertices lie in the same $v$1 and have distance at most $v$2, then their out-neighbors are distinct; if they lie in different $v$3, distance at most $v$4 also forces distinct out-neighbors (Dvořák et al., 2023).

The impossibility threshold is not fully closed, but a sharp lower obstruction is known at $v$5 faults. For every $v$6, there exists a set of $v$7 faulty edges or a set of $v$8 faulty vertices such that no fault-free paired 2-DPC exists in $v$9 for some terminal pairs (Dvořák et al., 2023). The explicit counterexample chooses $T=\{v,y\}$00 and $T=\{v,y\}$01 at distance $T=\{v,y\}$02 with common neighbor $T=\{v,y\}$03, then faults the $T=\{v,y\}$04 edges at $T=\{v,y\}$05 other than $T=\{v,y\}$06 and $T=\{v,y\}$07, or instead faults the corresponding adjacent vertices. Any fault-free path through $T=\{v,y\}$08 is then forced to pass both $T=\{v,y\}$09 and $T=\{v,y\}$10, which prevents two disjoint covering paths (Dvořák et al., 2023). The status at $T=\{v,y\}$11 faults remains open: the paper asks whether the bound $T=\{v,y\}$12 can be improved to $T=\{v,y\}$13 for all $T=\{v,y\}$14 (Dvořák et al., 2023).

The proofs are constructive. They specify how to choose edges or vertices, cut Hamiltonian paths or cycles, and splice the resulting pieces through out-edge matchings. Given implementations of Hamiltonian path and cycle routines in the subgraphs, the overall procedure builds the two paths in a number of steps proportional to the number of subgraphs visited; the output size is $T=\{v,y\}$15, so the construction is linear in the output length and polynomial in $T=\{v,y\}$16 (Dvořák et al., 2023).

4. Established graph families and exact guarantees

Several graph classes now admit exact paired many-to-many 2-disjoint path cover theorems.

Graph family	Guarantee	Reference
Balanced hypercube $T=\{v,y\}$17	Fault-free paired 2-DPC for any $T=\{v,y\}$18, $T=\{v,y\}$19; with faulty edges, the bound $T=\{v,y\}$20 is optimal	(Lü, 2018)
Johnson graph $T=\{v,y\}$21	$T=\{v,y\}$22 is paired 2-coverable for all $T=\{v,y\}$23 and $T=\{v,y\}$24	(Liu et al., 21 Jul 2025)
Layered Johnson family $T=\{v,y\}$25	If $T=\{v,y\}$26 and $T=\{v,y\}$27, then $T=\{v,y\}$28 is paired 2-coverable	(Liu et al., 21 Jul 2025)
BCube $T=\{v,y\}$29	Under $T=\{v,y\}$30-PEF, for any source set $T=\{v,y\}$31 and target set $T=\{v,y\}$32, there exist two disjoint covering paths $T=\{v,y\}$33	(Cai et al., 4 May 2025)

For balanced hypercubes, the fault-free result had already been established before the fault-tolerant version. The edge-fault theorem states that for $T=\{v,y\}$34, if $T=\{v,y\}$35 and $T=\{v,y\}$36, then for any $T=\{v,y\}$37 and $T=\{v,y\}$38, there exist two vertex-disjoint simple paths $T=\{v,y\}$39 with $T=\{v,y\}$40, $T=\{v,y\}$41, $T=\{v,y\}$42, and $T=\{v,y\}$43 (Lü, 2018). The upper bound $T=\{v,y\}$44 is optimal: with $T=\{v,y\}$45 faulty edges concentrated around a common neighbor $T=\{v,y\}$46 of symmetric vertices $T=\{v,y\}$47, any cover would force both paths through $T=\{v,y\}$48, contradicting vertex-disjointness (Lü, 2018). The later paired 3-disjoint path cover theorem for $T=\{v,y\}$49, valid for $T=\{v,y\}$50, explicitly treats paired 2-DPC as a previously known special case and extends the same recursive decomposition machinery to three prescribed pairs (Guo et al., 2019).

For Johnson graphs, the theorem is unconditional and fault-free. The proof partitions $T=\{v,y\}$51 into the vertices $T=\{v,y\}$52 not containing $T=\{v,y\}$53, so $T=\{v,y\}$54, and the vertices $T=\{v,y\}$55 containing $T=\{v,y\}$56, so $T=\{v,y\}$57, then performs a five-case analysis according to how many endpoints lie in $T=\{v,y\}$58 (Liu et al., 21 Jul 2025). Splicing relies on neighbors obtained by swapping one element and on Hamiltonian paths inside the two induced halves. The same paper proves that the layered family $T=\{v,y\}$59, built from levels $T=\{v,y\}$60 with vertical edges $T=\{v,y\}$61 whenever $T=\{v,y\}$62, is also paired 2-coverable whenever $T=\{v,y\}$63 and $T=\{v,y\}$64 (Liu et al., 21 Jul 2025).

For BCube, the relevant fault model is partitioned by dimension. In the logical graph $T=\{v,y\}$65, the $T=\{v,y\}$66-PEF constraints are $T=\{v,y\}$67 and, for $T=\{v,y\}$68,

$T=\{v,y\}$69

Under this model, the paper proves Hamiltonian connectivity and then the existence of a paired 2-DPC for arbitrary prescribed source and target pairs (Cai et al., 4 May 2025). The per-dimension budgets grow exponentially in $T=\{v,y\}$70, which is the paper’s “exponential fault tolerance” statement (Cai et al., 4 May 2025).

A higher-rank generalization appears in bipartite transposition-like graphs. The main theorem there is not restricted to $T=\{v,y\}$71: every rank $T=\{v,y\}$72 bipartite transposition-like graph admits a paired $T=\{v,y\}$73-to-$T=\{v,y\}$74 disjoint path cover for all choices of $T=\{v,y\}$75 and $T=\{v,y\}$76, provided $T=\{v,y\}$77 lies in one partite set and $T=\{v,y\}$78 in the other (Coleman et al., 2024). The paper explicitly notes that establishing “paired many-to-many 2-disjoint path covers” in the stronger sense of two internally disjoint $T=\{v,y\}$79–$T=\{v,y\}$80 paths per pair is an open direction for this class (Coleman et al., 2024).

5. Related formulations and neighboring problems

Several neighboring lines of work relax, generalize, or reinterpret the paired many-to-many 2-disjoint path cover problem.

The hypercube literature contains a closely related but unpaired theorem. In $T=\{v,y\}$81 with faulty vertices, for $T=\{v,y\}$82, if $T=\{v,y\}$83, $T=\{v,y\}$84, $T=\{v,y\}$85, and every fault-free vertex has at least two fault-free neighbors, then there exist two vertex-disjoint fault-free paths whose endpoints are exactly $T=\{v,y\}$86 and whose union covers at least $T=\{v,y\}$87 vertices (Li et al., 2012). The theorem guarantees some pairing between $T=\{v,y\}$88 and $T=\{v,y\}$89, not a pre-specified pairing, so it is an unpaired many-to-many result rather than a paired 2-DPC theorem in the strict sense (Li et al., 2012). Balanced hypercubes also admit a higher-cardinality unpaired result: for $T=\{v,y\}$90, any $T=\{v,y\}$91 and $T=\{v,y\}$92 with $T=\{v,y\}$93 admit an unpaired $T=\{v,y\}$94-DPC, and the upper bound $T=\{v,y\}$95 is best possible (Lü et al., 2019).

A different reinterpretation arises in network design under shortest-path routing. In the cover-by-pairs framework, a customer is covered by two facilities if there exist two disjoint shortest paths from the facilities to that customer. The paper distinguishes pathwise-disjoint coverage (PDFL, relevant with MPLS) from setwise-disjoint coverage (SDFL, relevant with OSPF/IS-IS) and states that the “Paired Many-to-Many 2-Disjoint Path Cover” problem under shortest-path routing is exactly these Set Cover by Pairs formulations with pairwise coverage constraints driven by shortest paths (Johnson et al., 2016). This is not a spanning two-path partition of the entire graph; instead, it is a facility-location optimization problem in which each customer must be served by two disjoint shortest paths from selected facilities (Johnson et al., 2016).

In directed acyclic graphs, the closely related MinPCRP problem asks for a minimum-cardinality path cover such that both vertices of every required pair belong to the same path. The paper proves a sharp computational boundary: deciding whether there exists a solution consisting of at most three paths is NP-complete, whereas deciding whether a solution consisting of at most two paths exists is polynomial-time solvable via reduction to 2-Clique Partition in an auxiliary compatibility graph (Beerenwinkel et al., 2013). This is not the same as fixing two prescribed endpoint pairs, but it addresses the two-path cover regime with pair constraints on path co-membership.

6. Methods, applications, and open questions

Across graph families, the dominant proof pattern is recursive decomposition plus splicing. Burnt pancake graphs decompose into $T=\{v,y\}$96 copies of $T=\{v,y\}$97 linked by matchings of $T=\{v,y\}$98-edges (Dvořák et al., 2023). Balanced hypercubes decompose into four $T=\{v,y\}$99 subcubes, with crossing edges and 8-cycles used to reroute and concatenate local covers (Lü, 2018). Johnson graphs split into two induced Johnson graphs according to containment of a distinguished element $P_{uv}$00, and $P_{uv}$01 uses vertical inclusion edges between consecutive levels (Liu et al., 21 Jul 2025). BCube partitions along one coordinate into $P_{uv}$02 sub-BCubes, each isomorphic to $P_{uv}$03, then lifts local path pieces through non-faulty inter-subgraph edges and Hamiltonian paths in the remainder (Cai et al., 4 May 2025). This recurring architecture suggests that paired many-to-many 2-DPC is especially tractable in recursively defined, highly symmetric graph families with abundant inter-module matchings.

The main application domain is fault-tolerant interconnection networks. For burnt pancake graphs, paired many-to-many 2-disjoint path covers are motivated by full utilization, load balancing, and redundancy: they partition the network’s vertices between two end-to-end routes and remain available under up to $P_{uv}$04 mixed faulty elements (Dvořák et al., 2023). In BCube, the same idea is tied to message passing under large numbers of link failures, with per-level budgets growing exponentially in the dimension index (Cai et al., 4 May 2025). In shortest-path routing networks, the paired-cover viewpoint underlies monitor placement and robust content distribution, with SDFL modeling OSPF/IS-IS and PDFL modeling MPLS (Johnson et al., 2016).

Several open questions remain explicit in the cited literature. For burnt pancake graphs, the central unresolved issue is whether the mixed-fault bound $P_{uv}$05 can be improved to $P_{uv}$06 for all $P_{uv}$07 (Dvořák et al., 2023). In hypercubes with faulty vertices, the theorem for $P_{uv}$08 guarantees some pairing but not a predetermined pairing, and extending the stronger vertex-fault bound to mixed vertex-plus-edge faults is stated as a natural direction (Li et al., 2012). For BCube, the paper focuses on $P_{uv}$09 and states that generalization to $P_{uv}$10 is only suggested by the hierarchical splicing strategy, not proved (Cai et al., 4 May 2025). For bipartite transposition-like graphs, the stronger requirement of two internally disjoint $P_{uv}$11–$P_{uv}$12 paths for each pair is explicitly left open (Coleman et al., 2024).

The current body of work therefore presents paired many-to-many 2-disjoint path cover as a sharply structured spanning problem: stronger than Hamiltonian connectivity, sensitive to pairing and parity, highly compatible with recursive network topologies, and increasingly integrated with fault models, routing constraints, and degree-theoretic conditions across both undirected and directed settings.