Complete Edge Relaxation Overview

• Complete edge relaxation is a multifaceted concept that exhaustively treats edge elements—from network rewiring and label propagation to LP formulation and quantum equilibration—until a limiting state is reached.
• It underpins randomized network dynamics by transforming maximally clustered graphs into equilibrium ensembles and provokes phase transitions in connectivity with precise analytical trajectories.
• The concept extends to non-adaptive algorithmic strategies, enhanced optimization relaxations in binary polynomial problems, and experimental demonstrations of energy equilibration in quantum Hall edge states.

Searching arXiv for the cited papers and topic variants to ground the article. Complete edge relaxation is a polysemous technical term whose meaning depends strongly on disciplinary context. In the arXiv literature it denotes, at minimum, four distinct but structurally related ideas: the relaxation of a maximally clustered graph to an unclustered ensemble under repeated rewiring; repeated full passes of edge relaxations in non-adaptive shortest-path algorithms; an extended LP formulation for the multilinear polytope in binary polynomial optimization; and the complete equilibration of quantum Hall edge-state energy distributions. This suggests a common motif—exhaustive treatment of edge degrees of freedom until a limiting ensemble, exact relaxation, or thermal state is reached—but the mathematical objects, observables, and correctness criteria are domain-specific (Klaise et al., 2017, Eppstein, 2023, Pia et al., 17 Jul 2025, 0907.2996).

1. Network rewiring and relaxation from maximal clustering

In network science, complete edge relaxation refers to the long-time rewiring of an initially maximally clustered, degree-regular graph into the equilibrium ensemble associated with the chosen edge dynamics. The initial graph is a disjoint union of $N/(\langle k\rangle+1)$ cliques of size $\langle k\rangle+1$, so every node has degree $\langle k\rangle$, every connected triple is closed, and the clustering coefficient is $C(0)=1$. Every edge initially has multiplicity $m=\langle k\rangle-1$, with $q_{\langle k\rangle-1}(0)=1$ and $q_m(0)=0$ otherwise (Klaise et al., 2017).

Two canonical dynamics are analyzed. In the degree-preserving double-edge swap, two edges $(a,b)$ and $(c,d)$ are rewired to either $(a,c),(b,d)$ or $\langle k\rangle+1$0, subject to simple-graph constraints; the equilibrium ensemble is the configuration model on the fixed degree sequence. In the single edge replacement dynamic, one existing edge is deleted and replaced by a random non-edge; this preserves $\langle k\rangle+1$1 but not node degrees, and converges to the Erdős–Rényi ensemble with fixed $\langle k\rangle+1$2 and $\langle k\rangle+1$3. Time is normalized as $\langle k\rangle+1$4 and $\langle k\rangle+1$5, so $\langle k\rangle+1$6 means that, on average, each edge has been modified once. Complete relaxation is the limit $\langle k\rangle+1$7, equivalently $\langle k\rangle+1$8, where $\langle k\rangle+1$9 and clustering becomes asymptotically negligible (Klaise et al., 2017).

The central observables are the degree distribution $\langle k\rangle$0, the edge-multiplicity distribution $\langle k\rangle$1, and the clustering coefficient

$\langle k\rangle$2

A mean-field treatment yields

$\langle k\rangle$3

for both rewiring schemes. Under double-edge swaps, the degree distribution is invariant, so

$\langle k\rangle$4

Under single edge replacement, the degree distribution evolves toward Poisson with mean $\langle k\rangle$5, with

$\langle k\rangle$6

and

$\langle k\rangle$7

At large $\langle k\rangle$8, $\langle k\rangle$9 (Klaise et al., 2017).

A further result is a continuous phase transition in connectivity. Starting from disjoint cliques, a giant connected component appears when the average number of external edges between initial cliques exceeds one. The critical normalized times are

$C(0)=1$0

For ER replacement, a revised estimate accounting for repeated rewiring of the same edge is

$C(0)=1$1

Thus complete edge relaxation is not merely triangle destruction: it is a time-resolved randomization process with analytically tractable trajectories for clustering, multiplicity, degree heterogeneity, and percolation onset (Klaise et al., 2017).

2. Non-adaptive shortest-path relaxation as repeated full edge passes

In algorithmic graph theory, complete edge relaxation is not a formal term of art in the source paper, but it corresponds to the regime in which one repeatedly relaxes all edges in a fixed order. The model is single-source shortest paths on a directed graph $C(0)=1$2 with source $C(0)=1$3, tentative distances $C(0)=1$4, and edge relaxations

$C(0)=1$5

The relevant class is non-adaptive relaxation algorithms, where the sequence of relaxed edges is determined only by the graph structure and not by weights or prior outcomes. Bellman–Ford in round-robin form is the canonical example: it performs $C(0)=1$6 rounds, each a complete pass over $C(0)=1$7 in the same order (Eppstein, 2023).

The main contribution is a set of lower bounds showing that this repeated full-pass strategy is, in the non-adaptive model, essentially optimal. On complete directed graphs with $C(0)=1$8 vertices, any deterministic non-adaptive relaxation algorithm must perform at least $C(0)=1$9 relaxation steps, and any randomized non-adaptive algorithm that succeeds with high probability must perform at least $m=\langle k\rangle-1$0 steps. Since non-adaptive Bellman–Ford uses $m=\langle k\rangle-1$1 relaxations on dense graphs, it is optimal up to constant factors there (Eppstein, 2023).

For general $m=\langle k\rangle-1$2 and $m=\langle k\rangle-1$3 with $m=\langle k\rangle-1$4, there exists a directed graph on $m=\langle k\rangle-1$5 edges and $m=\langle k\rangle-1$6 vertices requiring $m=\langle k\rangle-1$7 relaxations for any deterministic or high-probability randomized non-adaptive algorithm. When $m=\langle k\rangle-1$8, this improves to $m=\langle k\rangle-1$9. These bounds hold even when all edge weights are $q_{\langle k\rangle-1}(0)=1$0 and $q_{\langle k\rangle-1}(0)=1$1. Interpreted in terms of complete edge passes, they imply $q_{\langle k\rangle-1}(0)=1$2 rounds on complete graphs and $q_{\langle k\rangle-1}(0)=1$3 rounds on sparse graphs, so the $q_{\langle k\rangle-1}(0)=1$4 passes of Bellman–Ford are near-optimal even away from dense regimes (Eppstein, 2023).

The proof architecture is adversarial. For complete graphs, the shortest-path structure is forced to be a single directed path, and correctness requires the schedule to contain path edges in a sufficiently constrained order. Deterministic bounds are obtained via a telescoping estimate

$q_{\langle k\rangle-1}(0)=1$5

while randomized bounds use Yao’s principle together with a random permutation path distribution. For incomplete graphs, the construction separates “even-position” edges, which generate many choices, from “odd-position” connectors realized through rearrangeable non-blocking networks (Eppstein, 2023).

This usage differs sharply from the network-randomization meaning above. Here, relaxation does not mean stochastic rewiring toward an ensemble; it means deterministic or randomized propagation of shortest-path labels through exhaustive edge scans. The shared phrase marks completeness of the edge schedule, not equilibration of graph structure.

3. Complete Edge Relaxation in binary polynomial optimization

In binary polynomial optimization, Complete Edge Relaxation is a named object: an extended formulation for the multilinear polytope associated with a hypergraph $q_{\langle k\rangle-1}(0)=1$6. With binary variables $q_{\langle k\rangle-1}(0)=1$7 and monomial variables $q_{\langle k\rangle-1}(0)=1$8 for $q_{\langle k\rangle-1}(0)=1$9, the multilinear polytope is

$q_m(0)=0$0

The standard linearization imposes, independently for each monomial, the usual upper and lower envelope inequalities; Complete Edge Relaxation strengthens this by enforcing the exact local convex hull on every maximal hyperedge and sharing subedge variables globally across overlaps (Pia et al., 17 Jul 2025).

Let $q_m(0)=0$1 denote the maximal hyperedges and let $q_m(0)=0$2 be the family of all subedges of original edges of size at least $q_m(0)=0$3. For each $q_m(0)=0$4, CER works in the space $q_m(0)=0$5 and imposes the Sherali–Adams RLT inequalities

$q_m(0)=0$6

for all $q_m(0)=0$7. The resulting relaxation is

$q_m(0)=0$8

Geometrically, this is “complete on each maximal hyperedge”: within every $q_m(0)=0$9, all monomials supported on $(a,b)$0 are described by the full convex hull, and consistency across different maximal edges is enforced by shared subedge variables (Pia et al., 17 Jul 2025).

Its principal exactness theorem is structural. CER is an extension of the multilinear polytope if and only if the hypergraph is alpha-acyclic. The relevant characterization is the running-intersection property: the maximal hyperedges can be ordered $(a,b)$1 so that for every $(a,b)$2 there exists $(a,b)$3 with

$(a,b)$4

This is substantially more general than the exactness condition for the standard linearization, which holds if and only if the hypergraph is Berge-acyclic. The contrast is explicit: CER is exact for the most general standard acyclicity notion used here, whereas standard linearization is exact only for the most restrictive one (Pia et al., 17 Jul 2025).

The relaxation is also strictly stronger than several established alternatives. It dominates the standard linearization, implies the flower relaxation, and is stronger than the intersection of all recursive McCormick relaxations. For alpha-cycles of length three, the framework yields new facet-defining inequalities generalizing the triangle inequalities of the Boolean quadric polytope. If $(a,b)$5 form such a cycle, one family is

$(a,b)$6

together with its cyclic permutations and a fourth inequality

$(a,b)$7

plus all switchings. In the quadratic case these reduce exactly to Padberg’s classical triangle inequalities (Pia et al., 17 Jul 2025).

Because each maximal hyperedge contributes $(a,b)$8 inequalities, CER has size $(a,b)$9 when the rank $(c,d)$0 is bounded. The paper reports that for degree $(c,d)$1–$(c,d)$2 instances from image restoration and error-correcting codes, optimizing over CER is efficient and often produces binary solutions. Here complete edge relaxation is therefore neither dynamic nor thermal: it is a polyhedral completion of all edge-local convex structure.

4. Quantum Hall edge relaxation as full energy equilibration

In quantum Hall physics, complete edge relaxation refers to the spatial progression of a non-equilibrium edge-state energy distribution toward a fully equilibrated downstream state. The precise mechanism depends on the microscopic model. In the bosonization framework for an integer quantum Hall edge capacitively coupled to a linear external circuit, a single-electron excitation propagates through an interaction region where edge magnetoplasmons scatter into circuit photons. The high-frequency admittance is

$(c,d)$3

with $(c,d)$4 the edge plasmon transmission amplitude. Decoherence and energy relaxation are controlled by the same scattering data: extrinsic decoherence from photon emission is

$(c,d)$5

Complete relaxation corresponds to strong suppression of the single-electron coherence and vanishing quasiparticle peak weight $(c,d)$6 in the outgoing momentum distribution (0907.2996).

At filling factor $(c,d)$7, non-equilibrium bosonization yields a different hierarchy of regimes after a quantum point contact creates an initial double-step distribution. At short distances, non-Gaussian current noise produces an asymmetric distribution whose derivative is a Lorentzian with width

$(c,d)$8

At intermediate distances, dispersion-induced overlap suppresses higher cumulants, and the derivative becomes a symmetric Lorentzian centered at $(c,d)$9 with width

$(a,c),(b,d)$0

At long distances, weak non-integrable processes drive a Fermi-like local equilibrium with width

$(a,c),(b,d)$1

In this formulation, complete edge relaxation is the eventual arrival at the Fermi-Dirac regime after the non-Gaussian and Gaussian bosonic stages (Levkivskyi et al., 2011).

Other models emphasize microscopic scattering channels. In clean, translationally invariant integer quantum Hall edges, two-body collisions are kinematically ineffective, so hot-electron relaxation proceeds through three-body collisions. Edge reconstruction qualitatively alters the rates: spin reconstruction enhances the velocity mismatch between branches, while charge reconstruction introduces counter-propagating modes and allows relaxation even at $(a,c),(b,d)$2. The characteristic relaxation length is $(a,c),(b,d)$3, and complete edge relaxation means $(a,c),(b,d)$4, so that downstream distributions are close to stationary or thermal forms (Karzig et al., 2012).

When translational invariance is weakly broken by disorder, non-momentum-conserving two-particle collisions between co-propagating edge states produce a kinetic description that reduces, in the small-$(a,c),(b,d)$5 regime, to coupled Fokker–Planck equations. The injected high-energy peak drifts downward in energy with velocity

$(a,c),(b,d)$6

broadens diffusively, and heats the coupled edge states. Full relaxation occurs around

$(a,c),(b,d)$7

after which both channels approach equal-temperature Fermi functions with

$(a,c),(b,d)$8

and

$(a,c),(b,d)$9

This formulation makes complete edge relaxation a controlled kinetic limit governed by drift, diffusion, and conservation laws (Lunde et al., 2016).

5. Experimental criteria for completeness in quantum Hall edges

Experiments operationalize complete edge relaxation by measuring when a downstream probe sees a single equilibrated distribution rather than a multi-step or overheated one. A QPC-based local-voltage spectroscopy experiment at $\langle k\rangle+1$00, $\langle k\rangle+1$01, and $\langle k\rangle+1$02 distinguishes two mechanisms. In a tunneling-relaxation configuration, the slope $\langle k\rangle+1$03 remains approximately $\langle k\rangle+1$04 from $\langle k\rangle+1$05 to $\langle k\rangle+1$06, implying an inter-edge tunneling relaxation length $\langle k\rangle+1$07. In an exchange-only configuration, the same observable decays exponentially with relaxation length $\langle k\rangle+1$08. For hotspot relaxation generated by large bias across the injector QPC, the extracted length is $\langle k\rangle+1$09. In that setting, near-complete equilibration by exchange occurs within several microns, whereas tunneling alone remains ineffective on tens of microns (Otsuka et al., 2013).

A later experiment uses partitioned thermal noise to define a quantitative relaxation parameter $\langle k\rangle+1$10. Edge modes are emitted from a heated floating reservoir at temperature $\langle k\rangle+1$11, and a downstream QPC measures local effective temperature through the excess partition noise

$\langle k\rangle+1$12

for the single-mode case. Here $\langle k\rangle+1$13 is the bath temperature, $\langle k\rangle+1$14 corresponds to no relaxation, and $\langle k\rangle+1$15 corresponds to complete relaxation to the bath, $\langle k\rangle+1$16. A deliberately inserted massive floating contact provides an experimental realization of the $\langle k\rangle+1$17 limit (Rosenblatt et al., 2020).

The measured relaxation strength depends strongly on the edge phase. At $\langle k\rangle+1$18, particle-like states show mild relaxation over $\langle k\rangle+1$19 and clearer relaxation over $\langle k\rangle+1$20: for $\langle k\rangle+1$21, $\langle k\rangle+1$22 at $\langle k\rangle+1$23 and $\langle k\rangle+1$24 at $\langle k\rangle+1$25; for $\langle k\rangle+1$26, the corresponding values are $\langle k\rangle+1$27 and $\langle k\rangle+1$28. By contrast, the hole-conjugate state $\langle k\rangle+1$29 relaxes much faster, with $\langle k\rangle+1$30 at $\langle k\rangle+1$31 and $\langle k\rangle+1$32 at $\langle k\rangle+1$33, consistent with strong equilibration mediated by counter-propagating neutral modes. Raising the base temperature to $\langle k\rangle+1$34 causes severe relaxation already at $\langle k\rangle+1$35 for all tested states (Rosenblatt et al., 2020).

These experiments sharpen the meaning of “complete.” In quantum Hall devices it is not enough for energy to be redistributed internally; complete edge relaxation is reached only when the local downstream probe is consistent with the fully relaxed benchmark, either a single equilibrated Fermi function in the spectroscopy-based language or $\langle k\rangle+1$36 in the partition-noise language.

6. Comparative structure and sources of ambiguity

The phrase “complete edge relaxation” therefore spans three kinds of completeness. In stochastic network dynamics, completeness is asymptotic ensemble convergence under repeated local rewiring, with observables such as $\langle k\rangle+1$37, $\langle k\rangle+1$38, and giant-component onset (Klaise et al., 2017). In non-adaptive shortest-path algorithms, completeness is exhaustive coverage of the edge set in repeated passes, with complexity measured by total relaxation steps or rounds (Eppstein, 2023). In binary polynomial optimization, completeness means enforcing the full convex hull on each maximal hyperedge in an extended formulation, with exactness characterized by alpha-acyclicity (Pia et al., 17 Jul 2025). In quantum Hall physics, completeness is thermodynamic or kinetic: loss of quasiparticle character, equilibration of edge-state distributions, or cooling to the bath temperature depending on the model and observable (0907.2996, Levkivskyi et al., 2011, Karzig et al., 2012, Lunde et al., 2016, Otsuka et al., 2013, Rosenblatt et al., 2020).

A recurrent misconception is to treat these usages as interchangeable because they share the words “edge” and “relaxation.” They do not. The “edge” may be an edge of a graph, a hyperedge, an edge channel in a quantum Hall system, or an edge in the sense of an algorithmic adjacency. The “relaxation” may mean Markov-chain randomization, Bellman–Ford label correction, LP relaxation, or non-equilibrium thermalization. Only the binary polynomial formulation uses Complete Edge Relaxation as a formal capitalized name; the shortest-path and quantum Hall usages are descriptive or interpretive extensions of terminology present in the cited works (Pia et al., 17 Jul 2025, Eppstein, 2023).

What remains common is methodological rather than semantic. Each literature studies what happens when edge-level structure is treated exhaustively rather than approximately: triangles are destroyed until clustering vanishes; shortest-path labels are propagated by complete scans until exact distances are forced; local hyperedge convex hulls are imposed until the relaxation matches the multilinear polytope on alpha-acyclic instances; and edge-channel excitations evolve until only an equilibrated state remains. That family resemblance is real, but it is an analogy across domains rather than a single unified definition.