PEIM: Quantifying Edge Impact in Networks

• PEIM is a matrix-valued framework that quantifies the global impact of adding or modifying network edges using first-order sensitivity analysis.
• It leverages derivatives and physically motivated scores to rank potential improvements in controllability, communicability, entropy, and spectral metrics.
• By avoiding costly combinatorial searches, PEIM enables scalable and near-optimal edge selection for network augmentation and system-level optimization.

A Potential Edges Importance Matrix (PEIM) is a matrix-valued framework to quantify, for all possible (existing or non-existing) edges in a network, the prospective impact of adding or modifying each edge on a global system-level performance metric. PEIMs are deployed across a range of complex networked systems—including dynamical, communication, and power networks—to inform optimization tasks such as edge selection for structural improvement, control enhancement, or resource allocation. The construction of a PEIM is tailored to the specific dynamical or structural quantity of interest (e.g., controllability, communicability, entropy, spectral radius), but always involves efficiently assembling, for all potential edges, a matrix of first-order (gradient-based) or physically motivated importance scores. This methodology enables scalable, principled, and near-optimal ranking and modification of network edges without recourse to costly combinatorial search.

1. Mathematical Formulations and General Construction

PEIMs formalize edge-level sensitivity analysis. For a given system descriptor (e.g., adjacency matrix $A$, Laplacian $L$) and scalar global network metric $\varphi$, the $(i,j)$ entry of the PEIM is typically the first-order directional derivative $\frac{\partial \varphi}{\partial w_{ij}}$, where $w_{ij}$ is the weight of edge $i \to j$, evaluated at the current network state (possibly with $w_{ij}=0$ if the edge is absent). The general paradigm is:

• For linear time-invariant (LTI) or power systems dynamics, PEIM quantifies the Gramian-based controllability sensitivity to edge modifications $E_{ij}=\frac{\partial f(W_c)}{\partial w_{ij}}$, where $f$ is a metric on the controllability Gramian $L$0.
• For global diffusion or communication processes, PEIM encodes the response of total communicability, von Neumann entropy, or spectral complexity to edge perturbations via analytic perturbation theory.
• For spectral or synchronization-related performance, PEIM ranks edges by the change in leading eigenvalue (dynamical importance).

In all cases, the resulting $L$1 (generally symmetric or structured) matrix encodes the full landscape of potential edge contributions.

2. PEIM in Dynamical Systems and Controllability

For continuous-time LTI systems $L$2, the PEIM (also called the Edge Centrality Matrix, ECM) centralizes first-order variations in system controllability metrics due to infinitesimal edge modifications. The steps are:

1. Compute the infinite-horizon controllability Gramian $L$3, solving $L$4.
2. For each possible edge $L$5, compute $L$6 via the Lyapunov sensitivity equation $L$7, with $L$8.
3. Calculate the gradient of the desired scalar metric (e.g., $L$9, $\varphi$0) and take $\varphi$1 as the PEIM entry.
4. Aggregate the results into a matrix $\varphi$2 (the PEIM); entries are large where edge $\varphi$3 appreciably increases controllability, reduces required control energy, or improves damping.

This PEIM approach enables near-optimal edge selection for controllability enhancement in power networks and more general LTI systems, outperforming purely structural indices and efficiently scaling to large graphs (Bahavarnia et al., 15 May 2025, Chanekar et al., 2021).

3. PEIM for Communicability and Spectral Metrics

In complex network analysis, PEIMs are constructed for metrics such as total communicability and spectral radius:

• For total communicability $\varphi$4, the PEIM entry is the Fréchet derivative $\varphi$5, efficiently approximated with Krylov subspace or finite-difference techniques. This quantifies the marginal gain in global communicability from adding edge $\varphi$6 (Noschese et al., 2024).
• For the leading eigenvalue $\varphi$7 (dynamical importance), the PEIM entry for $\varphi$8 not in the current edge set is $\varphi$9 where $(i,j)$0 is the dominant eigenvector. This exact form provides closed-form prioritization for synchronization or epidemic thresholds (Young et al., 2024).

These formulations can be tuned for undirected, directed, or weighted networks, and can incorporate edge absence or presence.

4. Entropic and Redundancy-Based PEIM: von Neumann Entropy and Routing Design

PEIMs have been adapted for entropy-driven and redundancy metrics in applications such as:

• von Neumann entropy (VNE): For a Laplacian $(i,j)$1 and transport operator $(i,j)$2, the PEIM entry approximates the change in VNE from edge removal/addition using first-order spectral perturbation, scaling as $(i,j)$3 per edge after an initial $(i,j)$4 eigen-decomposition (Kazimer et al., 2022). PEIM entries isolate which edges most influence global information/diffusion complexity across diffusion timescales.
• Optical satellite networking: For dynamic ISL assignment, the PEIM $(i,j)$5 normalizes and sums hop-count reduction and new shortest path count for each potential link $(i,j)$6. Higher $(i,j)$7 indicates a link that maximally reduces average hop length and increases redundancy, enabling optimized topology construction (Yang et al., 2023).

These domain-specific PEIM instantiations drive edge selection in high-connectivity, low-delay, or high-redundancy network design.

5. Unified Methodology and Algorithmic Considerations

The PEIM paradigm is characterized by:

• General recipe: Construct a performant global scalar metric; differentiate with respect to each (potential) edge weight; assemble sensitivities into a matrix to enable edge ranking; select and optimize over top-ranked edges.
• Computational tractability: All core PEIM forms are designed to avoid combinatorial edge searches; leading-order approximations, sparsity exploitation, and subspace projection methods ensure scalability (e.g., $(i,j)$8 for Krylov-based communicability PEIM, $(i,j)$9 for VNE-based PEIM after eigendecomposition).
• Flexibility: By switching the core metric, the PEIM framework addresses structurally distinct objectives—reachability, robustness, synchronizability, redundancy.
• Potential edge inclusion: PEIMs naturally extend to hypothetical edge additions, not just existing link improvements, thereby supporting network augmentation scenarios.

A summary table of representative PEIM constructions follows:

Metric	PEIM Entry Formula	Domain/Reference
Gramian Trace/Logdet	$\frac{\partial \varphi}{\partial w_{ij}}$0	(Bahavarnia et al., 15 May 2025, Chanekar et al., 2021)
Total Communicability	$\frac{\partial \varphi}{\partial w_{ij}}$1 (Fréchet/Krylov)	(Noschese et al., 2024)
Dynamical Importance	$\frac{\partial \varphi}{\partial w_{ij}}$2	(Young et al., 2024)
von Neumann Entropy	$\frac{\partial \varphi}{\partial w_{ij}}$3	(Kazimer et al., 2022)
Connectivity/Redundancy	$\frac{\partial \varphi}{\partial w_{ij}}$4	(Yang et al., 2023)

6. Empirical Performance and Optimization Use Cases

Across application domains, PEIM-based edge selection and modification routines consistently yield near-optimal, scalable improvements:

• Power systems: Edge selection by ECM/PEIM drives substantial improvements in reachability, minimum control energy, and damping ratio (e.g., $\frac{\partial \varphi}{\partial w_{ij}}$5 improvement in negated trace-inverse for $\frac{\partial \varphi}{\partial w_{ij}}$6 interventions in standard IEEE models), consistently outperforming static centrality measures (Bahavarnia et al., 15 May 2025).
• Satellite networks: PEIM-guided link assignments directly lower average hop-count (by $\frac{\partial \varphi}{\partial w_{ij}}$7 and $\frac{\partial \varphi}{\partial w_{ij}}$8 over ACT and Greedy, respectively), and reduce wavelength demand ($\frac{\partial \varphi}{\partial w_{ij}}$9+ resource savings) without loss of connectivity (Yang et al., 2023).
• Entropy and diffusion complexity: Tuning timescale $w_{ij}$0 in VNE-PEIM reorders edge rankings, pinpointing context-specific high-impact links for local, mesoscopic, or global diffusion complexity (Kazimer et al., 2022).
• Controllability augmentation: PEIM edge ranking enables budget-constrained optimization (via nonlinear or SDP relaxations) over the most critical links, achieving near-optimal designs at dramatically reduced computational cost relative to brute-force enumeration (Bahavarnia et al., 15 May 2025, Chanekar et al., 2021).

7. Significance, Scope, and Connections

PEIM unifies network-theoretic sensitivity analysis, system-theoretic optimization, and applied network design into a coherent, first-principles methodology. By enabling the systematic identification and ranking of potentially critical edges, PEIM frameworks bridge the gap between purely structural (topological) centrality indices and metrics that capture genuine system-level trade-offs inherent in networked dynamical processes.

The PEIM framework is applicable to a diverse set of domains: power grids, communication/satellite constellations, transportation, social and epidemiological networks, neural/brain connectivity, and more. It facilitates network augmentation, control-system enhancement, robust design, and resource allocation with rigorous performance guarantees and computational transparency.

Recent research establishes that, irrespective of the system-level performance metric, a PEIM-based (first-order or physically motivated) ranking reliably guides interventions in large-scale networks toward maximal and interpretable impact (Bahavarnia et al., 15 May 2025, Noschese et al., 2024, Kazimer et al., 2022, Young et al., 2024, Yang et al., 2023, Chanekar et al., 2021).