Multi-Orientation Edge-Minimum Repair (MOEM)

• Multi-Orientation Edge-Minimum Repair (MOEM) is a method that reconstructs fault-pruned broadcast trees by reconnecting healthy components with exactly c–1 repair edges.
• It uses a deterministic, constant-size family of orientations to optimize repair depth and minimize external repair edges in dense Gaussian and Eisenstein–Jacobi networks.
• Re-rooting-assisted formulations further improve repair performance under mixed node/link failures while preserving internal healthy tree structures.

Multi-Orientation Edge-Minimum Repair (MOEM) is a repair paradigm for non-redundant one-to-all broadcasting in which, after faults, one evaluates multiple orientations of a broadcast tree, contracts the fault-pruned tree into healthy components, and reconnects those components using the minimum possible number of external component-crossing repair edges. In dense Gaussian networks and dense Eisenstein–Jacobi (EJ) networks, the repaired structure is a rooted spanning tree of the healthy subgraph, so each healthy node receives the message exactly once and no faulty node is used; for a chosen orientation with $c$ fault-pruned components and a connected healthy component graph, the optimal repair count is exactly $c-1$ (Albader, 16 Jun 2026, Albader, 18 Jun 2026). Re-rooting-assisted formulations extend the same idea to joint node/link failures and runtime-discovered single-link faults by treating root selection, and in the EJ case joint root-orientation selection, as part of the repair problem (Albader, 19 Jun 2026, Albader, 19 Jun 2026).

1. Formal definition and optimization criterion

MOEM begins from a fault-free diameter-level broadcast tree and a fault set, then removes faulty elements to obtain a forest of healthy components. The core repair object is the component graph: one vertex per healthy component, with an edge whenever the underlying healthy network contains a healthy cross-component link. The Gaussian MOEM paper denotes this graph by $\mathcal{C}_\theta$, the Gaussian re-rooting paper by $H_r$, and the EJ re-rooting paper by $K_{r,\theta}$. In each case, the repair objective is identical: preserve the internal structure of the healthy components of the pruned tree and add the fewest possible external component-crossing edges needed to recover a rooted spanning tree (Albader, 16 Jun 2026, Albader, 18 Jun 2026, Albader, 19 Jun 2026, Albader, 19 Jun 2026).

The fundamental optimality statement is a component lower bound. If the pruned tree has $c$ components, any non-redundant repaired tree preserving those internal components needs at least $c-1$ external component-crossing edges. The matching upper bound is constructive: if the healthy component graph is connected, a spanning tree of that component graph maps to exactly $c-1$ healthy cross-component edges, yielding a repaired broadcast tree that is connected, acyclic, and non-redundant. This is the sense in which MOEM is “edge-minimum”: the minimum is exact for the selected orientation, not merely asymptotic or heuristic (Albader, 16 Jun 2026, Albader, 18 Jun 2026).

The “multi-orientation” part is equally central. Different orientations induce different fault-pruned forests under the same fault set, hence different values of $c$, different component graphs, and different repaired depths. The direct Gaussian and EJ MOEM papers fix the source and search over constant-size orientation families; the re-rooting-assisted papers enlarge the search space by changing the root, and in the EJ case by jointly varying root and orientation. This suggests that MOEM is best understood as a two-level optimization: choose a favorable orientation class, then solve the per-orientation edge-minimum repair problem exactly (Albader, 16 Jun 2026, Albader, 18 Jun 2026, Albader, 19 Jun 2026, Albader, 19 Jun 2026).

2. Dense Gaussian MOEM

In dense Gaussian networks, the underlying graph is $G_k$, generated by $c-1$0, with

$c-1$1

The node set is $c-1$2, and edges satisfy

$c-1$3

The network is degree four. A coordinate model uses the bijection

$c-1$4

from the Manhattan ball

$c-1$5

onto the node set, so graph distance from the source is the Manhattan radius $c-1$6, and the diameter is at most $c-1$7 (Albader, 16 Jun 2026).

The Gaussian MOEM paper considers static permanent node faults. A coordinate-reduction orientation assigns each non-source node a parent whose Manhattan radius is smaller by $c-1$8, producing a diameter-level broadcast tree. MOEM does not rely on a single such rule: it evaluates a constant-size family $c-1$9 of $\mathcal{C}_\theta$0 deterministic orientations, including $\mathcal{C}_\theta$1-first, $\mathcal{C}_\theta$2-first, larger-coordinate-first, smaller-coordinate-first, and sign-preference variants. For each $\mathcal{C}_\theta$3, MOEM constructs the fault-free tree $\mathcal{C}_\theta$4, deletes the faults to obtain the forest $\mathcal{C}_\theta$5, builds the component graph $\mathcal{C}_\theta$6, and, when $\mathcal{C}_\theta$7 is connected, reconnects the components with exactly $\mathcal{C}_\theta$8 repair edges. Candidate repairs are ranked lexicographically by

$\mathcal{C}_\theta$9

with success preceding all failures (Albader, 16 Jun 2026).

Depth control is handled through a repair-certificate framework. For an attachment edge $H_r$0 that connects an already repaired part to a detached component $H_r$1, the predicted depth is

$H_r$2

The paper proves that for every one- or two-fault placement, the $H_r$3-orientation family contains a repair with depth at most $H_r$4. The proof combines a certificate framework, an explicit four-case off-axis analysis, and a five-component orthogonal-axis certificate. Exhaustive validation for $H_r$5 and large-scale validation through $H_r$6 confirm the implementation; the exhaustive runs report maximum repair edge count at most $H_r$7 for one fault and at most $H_r$8 for two faults, while the large random experiments show that random two-fault repairs use approximately two external repair edges (Albader, 16 Jun 2026).

3. Dense Eisenstein–Jacobi MOEM

In dense EJ networks, the graph is generated by $H_r$9, where

$K_{r,\theta}$0

and is naturally represented by the hexagonal ball

$K_{r,\theta}$1

The network has

$K_{r,\theta}$2

vertices, degree $K_{r,\theta}$3, and EJ distance

$K_{r,\theta}$4

so the diameter is $K_{r,\theta}$5. The six unit directions are

$K_{r,\theta}$6

and a node label can be written as

$K_{r,\theta}$7

in the direct MOEM paper, or as

$K_{r,\theta}$8

in the re-rooting runtime paper’s quotient-lattice presentation (Albader, 18 Jun 2026, Albader, 19 Jun 2026).

EJ-MOEM uses coordinate-reduction trees defined by priority orderings of the six directions. The direct EJ-MOEM paper evaluates a constant-size family of $K_{r,\theta}$9 orientations: $c$0 cyclic lists $c$1, $c$2 reverse-cyclic lists $c$3, and $c$4 alternating lists $c$5. For a given fault set $c$6, it builds $c$7, removes the faulty nodes, contracts the resulting components, and reconnects them through healthy EJ edges. As in the Gaussian case, edge-minimality is exact: if the component graph $c$8 is connected, the repair uses exactly $c$9 external component-crossing edges, which is the minimum compatible with preserving the healthy internal components of the fault-pruned tree (Albader, 18 Jun 2026).

The depth theory is sharper than in the Gaussian direct-MOEM formulation. The EJ paper proves a depth-certificate theorem for coordinate-reduction trees: every one-fault placement admits a repair of depth at most $c-1$0, and every two-fault placement admits a repair of depth at most $c-1$1. The proof uses the three-strip representation of EJ hexagons, a sector-suffix attachment lemma, a non-adjacent-sector separation lemma, and a six-direction shielding classification for paired cuts. Extended validation includes exhaustive one- and two-fault enumeration for $c-1$2, up to $c-1$3 and $c-1$4 two-fault placements at $c-1$5, structured theorem-critical tests through $c-1$6, and large random tests through $c-1$7, all with $c-1$8 success and no violation of the theorem (Albader, 18 Jun 2026).

4. Re-rooting-assisted MOEM under node and link failures

The re-rooting papers generalize the MOEM idea from static node faults to mixed node/link failures and runtime repair. In the dense Gaussian case, the method is described as choosing among multiple orientations, understood there as multiple roots of a broadcast tree, so that after faults one can repair with the minimum possible number of new edges. The paper states that it does not implement a generic MOEM framework under that name, but effectively is a MOEM scheme specialized to dense Gaussian broadcast networks. In the dense EJ runtime setting, the fundamental unit of analysis is the selected triple $c-1$9: a chosen root, a chosen EJ coordinate-reduction orientation, and the healthy component graph induced by that choice (Albader, 19 Jun 2026, Albader, 19 Jun 2026).

For Gaussian networks, re-rooting exploits the fact that a node at distance $c-1$0 from the root is always a leaf in the diameter-level tree. If all failed nodes can be moved to the distance-$c-1$1 boundary by root choice, they become leaves, pruning them does not disconnect any other node, and the repair edge count drops to $c-1$2. The paper proves that for any one- or two-node fault set there is a root with all faults at distance $c-1$3, using a constant-size boundary-intersection primitive based on at most $c-1$4 signed linear systems, with $c-1$5 arithmetic operations in a word-RAM model. It also proves deterministic single-link repair: if a failed link is not used by the selected tree, no repair is needed; if it is used, pruning creates exactly two components and exactly one repair edge is necessary and sufficient. A constant-time parent-coordinate test determines whether a candidate root uses a failed link (Albader, 19 Jun 2026).

For EJ networks, the runtime paper proves a stronger global condition. Let

$c-1$6

be the healthy EJ graph. Then a repaired broadcast tree over $c-1$7 exists if and only if $c-1$8 is connected. When $c-1$9 is connected, the component graph $c$0 is connected for every retained $c$1, and the repair phase succeeds with exactly $c$2 external component-crossing repair edges for that selected pruned tree. Deterministic guarantees include: one or two faulty nodes can always be placed on the distance-$c$3 boundary by re-rooting; a single failed link is either avoided or repaired by exactly one crossing edge; and under shallowest-layer entry selection the repaired depth satisfies $c$4 (Albader, 19 Jun 2026).

The candidate-selection objectives in the runtime formulations are explicitly lexicographic. In Gaussian networks the selector is essentially

$c$5

where $c$6 is the number of external repair edges, $c$7 the repaired depth, and $c$8 the number of candidate roots inspected. In the EJ runtime paper, valid repaired candidates are chosen by lexicographically minimizing

$c$9

This suggests that MOEM, in its re-rooted form, is a fault-aware search over root-orientation pairs whose first-order criterion is minimum repair-edge count, with depth treated as a secondary objective (Albader, 19 Jun 2026, Albader, 19 Jun 2026).

5. Validation, complexity, and systems behavior

The empirical record for MOEM and MOEM-like repair spans exhaustive enumeration, structured stress tests, large random validation, and simulator replay.

Paper	Validation scale	Headline result
(Albader, 16 Jun 2026)	Exhaustive $G_k$0; large-scale through $G_k$1	Depth at most $G_k$2; random two-fault repairs use approximately two external repair edges
(Albader, 18 Jun 2026)	Exhaustive $G_k$3; structured tests through $G_k$4; random through $G_k$5	$G_k$6 success and no theorem violation
(Albader, 19 Jun 2026)	$G_k$7, up to $G_k$8 nodes, $G_k$9 static trials, $c-1$00 transient trials	$c-1$01 recovery for deterministic and bounded regimes, $c-1$02 for multi-link faults, $c-1$03 for heuristic regimes
(Albader, 19 Jun 2026)	$c-1$04-trial campaign up to $c-1$05	$c-1$06 recovery when $c-1$07 is connected and substantial repair-edge reduction over fixed-source repair

The direct Gaussian and EJ MOEM papers both report linear-time behavior in network size because the orientation families are constant-size. For Gaussian MOEM, total running time remains $c-1$08 because the orientation set has size $c-1$09, component discovery and crossing-edge enumeration are $c-1$10, and the repair loop is dominated by the linear scans over a degree-four graph. For EJ-MOEM, the same structure yields $c-1$11 total work over the $c-1$12-orientation family, with small component counts in the one- and two-fault regimes (Albader, 16 Jun 2026, Albader, 18 Jun 2026).

The re-rooting Gaussian paper adds a systems-level evaluation through patched Gaussian-link Noxim replay. The experiments show that re-rooting reduces average repair edges by $c-1$13–$c-1$14 versus fixed-source repair. Patched Gaussian-link Noxim replays confirm packet-complete execution and show that re-rooting reduces repair edges, components, and depth. The same paper also distinguishes structural repair benefit from end-to-end latency: completion cycles depend on relocation, scheduling, delivery tail, and selector objective, so the stated claim is edge-minimum repair rather than universal completion-cycle dominance (Albader, 19 Jun 2026).

The EJ runtime paper compares hybrid re-rooting-assisted repair with fixed-source repair and global BFS rebuild. For $c-1$15-node faults averaged over all modes and sizes, fixed-source repair uses $c-1$16 repair edges and depth $c-1$17, whereas hybrid repair uses $c-1$18 repair edges and depth $c-1$19, a $c-1$20 reduction in repair edges. The global-BFS audit shows that BFS succeeds whenever $c-1$21 is connected and keeps depth near $c-1$22, but changes parent pointers for roughly every node; hybrid repair preserves the internal healthy tree components and changes only $c-1$23 crossing edges. This places MOEM in a distinct implementation regime: minimal forwarding-state disturbance rather than global tree recomputation (Albader, 19 Jun 2026).

6. Related orientation problems, scope, and limitations

A conceptually related but distinct problem is the orientation problem studied in “On finding orientations with fewest number of vartices with small out-degree” (Khoshkhah, 2014). There the task is to orient the edges of an undirected graph so as to minimize the number of vertices with out-degree at most $c-1$24, called poor or $c-1$25-light vertices. The paper gives a polynomial-time algorithm via a single maximum cardinality matching in an auxiliary graph with $c-1$26 vertices and $c-1$27 edges, and extends the result to a weighted version via maximum weight matching for nonnegative vertex costs. The connection to MOEM-like tasks is conceptual: it isolates a base case in which the only repair action is orientation choice, whereas broadcast MOEM must preserve rooted-tree structure, satisfy reachability from a designated source, and minimize external component-crossing repair edges rather than the number of low-out-degree vertices (Khoshkhah, 2014).

The current MOEM literature is topology-specific and theorem-specific. The direct Gaussian MOEM depth theorem is proved for $c-1$28 node faults, and the direct EJ-MOEM depth theorems are likewise limited to one- and two-fault placements. In the Gaussian runtime paper, deterministic guarantees cover one- and two-node faults, single-link faults, and certain bounded mixed regimes, while higher-order node/link cases are explicitly treated as best-effort heuristic regimes. In the EJ runtime paper, successful repair is characterized exactly by the connectivity of $c-1$29, but global optimization over all roots and orientations is not claimed; the search is ranked and capped rather than exhaustive (Albader, 16 Jun 2026, Albader, 18 Jun 2026, Albader, 19 Jun 2026, Albader, 19 Jun 2026).

Several open directions are explicit. The Gaussian runtime paper states that the link-avoidance exclusion test does not yet yield a constant-size closed-form selector for link faults. The direct Gaussian MOEM paper notes that the $c-1$30-orientation family is proved sufficient, not minimal, and that richer certificate libraries are needed for $c-1$31. The direct EJ-MOEM paper proves a worst-case $c-1$32 depth bound for two faults, but its exhaustive tests found no overhead larger than $c-1$33 for $c-1$34, prompting the open conjecture that depth $c-1$35 might suffice for all two-fault placements when $c-1$36. More generally, the Gaussian runtime paper proposes that a MOEM-style generalization beyond Gaussian networks would require symmetry or reorientation freedom, efficient per-orientation repair, an effective orientation selector, and resilience to local obstructions (Albader, 16 Jun 2026, Albader, 18 Jun 2026, Albader, 19 Jun 2026).

Taken together, these results define MOEM as a rigorous repair framework for structured interconnection networks: choose among multiple orientations, preserve the healthy internal forest induced by the chosen orientation, reconnect components with the exact minimum number of new edges, and control depth well enough to remain close to the diameter of the fault-free network.