The Honest Truncation

Parallel simulation of coupled physical systems faces a fundamental tension. Each subsystem can run on its own processor, but the coupling between subsystems requires iterating to convergence. If you stop iterating early — which is the whole point of parallelism, to save time — you introduce artificial energy into the simulation. The physics becomes dishonest.

Wei, Tao, Nie, and Tan (arXiv:2603.16424) resolve this by embedding the coupling in a Douglas-Rachford splitting scheme within wave coordinates. The result: energy conservation (passivity) is guaranteed at any iteration count. Stop after one iteration or a hundred — the simulation never creates or destroys energy. The error from early termination shows up as a well-defined mismatch between subsystem states, tracked explicitly, and it shrinks monotonically with more iterations.

The key insight is treating the interconnection’s losslessness as an orthogonal constraint. In wave coordinates, the energy-conserving coupling maps to a projection onto an orthogonal complement. Each iteration improves the projection’s accuracy, but even a crude projection preserves the orthogonality — and orthogonality is passivity.

This inverts the usual relationship between constraints and computation. Normally, physical constraints are the thing you’re trying to satisfy, and running out of computation means violating them. Here, the constraint is embedded in the computational structure itself: the algorithm cannot violate energy conservation regardless of how many iterations it runs. The constraint is not a target to converge toward but an invariant maintained throughout.

The practical consequence is a simulation that degrades gracefully under computational pressure. When the processor is busy, you get a less accurate simulation — but a physically honest one. The robot being simulated might move slightly wrong, but it won’t spontaneously generate energy. Accuracy is negotiable. Physics is not.