Parametric Traversal for Multi-Dimensional Cost-Aware Graph Reasoning
Nicolas Tacheny
TL;DR
The paper addresses the challenge of planning connectivity in incomplete infrastructure graphs where multiple constraints must be considered simultaneously. It introduces a parametric traversal framework that treats gap transitions (potentially deployable connections) as first-class transitions, and separates acceptability domain generation from feasibility predicates, enabling scalable pruning. Traversals carry multi-dimensional accumulation states and are guided by an exploration predicate, allowing policy-driven feasibility reasoning and preservation of non-scalar trade-offs. Demonstrative industrial case studies in datacenter and telco networks show conditional feasibility, non-scalarizable decisions, and policy calibration, highlighting practical applicability beyond classical shortest-path formulations. Overall, the approach provides a programmable, explainable planning tool for infrastructure design and decision-support in uncertain or evolving networks.
Abstract
Classical path search assumes complete graphs and scalar optimization metrics, yet real infrastructure networks are incomplete and require multi-dimensional evaluation. We introduce the concept of traversal: a generalization of paths that combines existing edges with gap transitions, missing but acceptable connections representing links that can be built. This abstraction captures how engineers actually reason about infrastructure: not just what exists, but what can be realized. We present a parametric framework that treats planned connections as first-class transitions, scales to large graphs through efficient candidate filtering, and uses multi-dimensional criteria to decide whether a traversal should continue to be explored or be abandoned. We evaluate the framework through representative scenarios in datacenter circuit design and optical route construction in telecommunication networks, demonstrating conditional feasibility, non-scalarizable trade-offs, and policy calibration capabilities beyond the reach of classical formulations.