Develops decision-support tools for optimizing telecommunications networks, suggesting improvements in structural robustness and rapid recovery strategies.
We transform a validated, regularized graph statistical-mechanics model (Paper I, doi:10.5281/zenodo.20982419) into two practical decision-support tools for telecommunications infrastructure, evaluated on 276 real-world network topologies (Internet Topology Zoo). • 𝐓𝐰𝐨 𝐝𝐞𝐜𝐢𝐬𝐢𝐨𝐧-𝐬𝐮𝐩𝐩𝐨𝐫𝐭 𝐭𝐨𝐨𝐥𝐬 are introduced:– A greedy optimization that ranks candidate links by their improvement of discrete (Forman–Ricci) curvature, a validated proxy for structural robustness, under a fixed budget.– A self-calibrating geographic cost term: distances within a network’s historical maximum link length are free, while longer links are linearly penalized. • The geographic term is 𝐟𝐮𝐥𝐥𝐲 𝐝𝐚𝐭𝐚-𝐝𝐫𝐢𝐯𝐞𝐧, requiring no external $/km estimates and adding no measurable computational cost. • Across all 276 networks:– Mean curvature: −3.23– Median curvature: −2.59– Only 3.3% exhibit positive mean curvature, almost exclusively small, near-complete clustersThese values align closely with the synthetic sparse random baseline from the companion model. • In five case studies (Turkey, Croatia, Latvia, Norway, Greece):– The most critical node is the national capital in 4/5 networks.– Its removal fragments the system into 18–57 disconnected components with no immediate recovery.– A secondary critical node allows fast recovery (0.04–0.2 seconds) with realistic link lengths. • In two cases, unconstrained optimization proposed unrealistically long links; the geographic term 𝐞𝐥𝐢𝐦𝐢𝐧𝐚𝐭𝐞𝐬 𝐚𝐥𝐥 𝐯𝐢𝐨𝐥𝐚𝐭𝐢𝐨𝐧𝐬 (0/8 and 0/3) without added cost. • Norway’s Uninett2010 network shows 𝐧𝐨 𝐬𝐢𝐧𝐠𝐥𝐞 𝐜𝐚𝐭𝐚𝐬𝐭𝐫𝐨𝐩𝐡𝐢𝐜 𝐟𝐚𝐢𝐥𝐮𝐫𝐞 𝐩𝐨𝐢𝐧𝐭, highlighting structural resilience. Scope clarification:Distance is explicitly treated as a proxy for capital cost, not a real economic estimate. The proposed methods provide rapid decision support for network operations, not a replacement for packet-level routing protocols. 𝐂𝐨𝐝𝐞 𝐚𝐧𝐝 𝐟𝐮𝐥𝐥 𝐫𝐞𝐩𝐫𝐨𝐝𝐮𝐜𝐢𝐛𝐢𝐥𝐢𝐭𝐲 𝐦𝐚𝐭𝐞𝐫𝐢𝐚𝐥𝐬 𝐢𝐧𝐜𝐥𝐮𝐝𝐞𝐝.Licensed under PolyForm Noncommercial 1.0.0, covering all algorithms and implementations across languages and hardware platforms.
No takes yet. Share an insight, caveat, or question.
Andrés Sebastián Pirolo (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: