A robot mapping its environment while localizing itself — the SLAM problem — produces a graph. Each node is a pose (position and orientation). Each edge is a constraint from sensor data: this pose and that pose are related by this transformation. The optimization problem is to find the set of poses that best satisfies all constraints simultaneously.
Kohler et al. noticed that this optimization, when linearized, produces sparse matrices with the same structure as finite element discretizations of elastic bar networks. Each constraint between poses acts like a bar connecting two nodes. The stiffness of the bar encodes the confidence in the measurement. Solving the SLAM problem is, mathematically, finding the equilibrium configuration of a deformable structure.
This is not metaphor. It is structural identity. The matrices are the same. The sparsity patterns are the same. Which means forty years of domain decomposition theory — Schwarz preconditioners, developed for parallel computing of structural problems — apply directly. The robot's map can be decomposed into overlapping spatial subdomains, each solved independently, then stitched together. The number of iterations required for convergence stays bounded regardless of how large the map grows.
The through-claim: the robot trajectory IS a structure. Not like a structure. Not analogous to a structure. The same mathematical object. Pose-graph SLAM and linear elastic bar networks differ in physical interpretation but share algebraic structure. The preconditioner doesn't care whether the nodes are finite element mesh points or robot poses. It sees only the graph, the sparsity, the coupling.
This is an inheritance of solved problems. The SLAM community had been developing its own solvers for its own matrices. The structural engineering community had already solved those matrices decades ago. The insight is recognizing the equivalence — that the new problem is an old problem wearing different clothes.