Maximally recoverable codes are error-correcting codes that recover from every erasure pattern that any code with the same structural constraints could possibly recover from. They are optimal by construction — no code with the same architecture can do better. The question is: which erasure patterns are correctable?
Brakensiek & Gopi (2026, arXiv 2602.22042) show that the answer comes from structural rigidity — a branch of geometry originating in Maxwell's 1864 analysis of mechanical frameworks. A correctable erasure pattern in a maximally recoverable tensor code corresponds exactly to a bipartite rigid embedding of a graph. An erasure pattern E is correctable if and only if the corresponding graph can be rigidly embedded into orthogonal spaces. The theorem is exact, not approximate: correctability is rigidity.
The connection is surprising because the two fields have entirely different origins and intuitions. Error-correcting codes are engineering tools — they protect data in storage systems by adding structured redundancy. Structural rigidity is classical geometry — it asks when a framework of bars and joints is stiff versus floppy. But both are asking the same underlying question: given a set of constraints, is the remaining structure sufficient to reconstruct the whole? In rigidity, the constraints are bar lengths and the reconstruction is the joint positions. In coding, the constraints are the surviving symbols and the reconstruction is the erased data. Same combinatorial skeleton, different physical clothing.
The general principle: when two problems from unrelated fields turn out to have the same mathematical structure, the equivalence is not a metaphor — it is a shared constraint. Results proven in one field transfer exactly to the other. Bernstein's characterization of correctable patterns, proved using tropical geometry in the rigidity context, now applies directly to code design. The bridge doesn't suggest similarity. It provides theorems.