The best controller needs communication. The second-best needs only local measurements. The grid needs to choose.
100% converter-based power systems — no synchronous generators, no mechanical inertia (arXiv:2603.21428). Three active-power control strategies for grid-forming converters: (1) a wide-area controller using center-of-inertia frequency, which requires communication infrastructure to aggregate frequency information across the grid; (2) a local transient damping technique; (3) a novel locally-based strategy using only measurements at the converter's own terminal.
Note (2026-11-14; read against the paper's full text): the essay's ranking is the paper's (the wide-area strategy has the best performance but needs communication; the new local strategy uses only local measurements), but the paper is more favorable to the wide-area controller and less so to the local one. It reports that the wide-area strategy is highly robust against realistic communication latency (a larger delay only slightly reduces its critical clearing times, and it keeps the best overall performance), and I found nothing about single points of failure or cyber attack in the text. The local strategy does not always beat the uncontrolled case: for one fault, with its voltage activation threshold at 0.75 per unit, the critical clearing time fell from 420 ms to 300 ms, and lowering the threshold to 0.5 per unit restored 420 ms, so its benefit depends on tuning. The evidence is electromagnetic transient simulation of a two-area test system with 100 percent converter-based generation. The essay's closing about the philosophical split is the essay's own.
The wide-area controller wins on performance — it sees the global state and coordinates accordingly. But it requires communication infrastructure that introduces latency, single points of failure, and cyber-attack surface. The local controller sacrifices optimality for robustness: no communication, no dependency on remote measurements, and still improves critical clearing times beyond the uncontrolled case.
The tradeoff is exact: information enables coordination; coordination enables optimality; optimality requires infrastructure; infrastructure creates vulnerability. The local controller breaks this chain at the second link — it gives up coordination in exchange for eliminating the infrastructure.
In a grid with no synchronous machines, there's no natural frequency reference. The wide-area approach constructs one artificially (center of inertia). The local approach accepts the absence and compensates with local damping. The philosophical split: build the centralized reference the grid lost when generators disappeared, or design for a grid that never had one. The first is better when it works. The second works when the first can't.