To control an active particle in a trap, you can vary the trap stiffness, or the activity (how hard the particle kicks), or both. Optimizing them independently gives you a good protocol. Optimizing them jointly gives you something unexpected: the optimal solution spontaneously breaks a symmetry that the independent solutions preserved.
Note (2026-11-13; read against the paper's full text): the essay says that joint optimization of stiffness and activity makes the minimum-work protocol asymmetric and that the symmetry was only an artifact of optimizing one knob at a time. In the paper the asymmetry about the midpoint of the protocol is found for the work-optimal persistence (activity) protocols, while the heat-optimal persistence protocols are symmetric, and it traces the difference to the mathematical structure of the work and heat cost functions: the compression of the trap imposes an arrow of time, and to avoid a late spike in the variance the persistence must be reduced as late as possible, giving a skewed profile. So the paper explains the asymmetry by the compression, and finds the symmetric pulse for heat; it does not say the symmetry was always broken. The paper's other finding that the essay reports, that naively combining independently optimized controls costs only slightly more work than the full multi-parameter solution, is in its abstract.
The paper finds that when trap stiffness and activity are coupled in the optimization, the minimum-work protocol for a cyclic process becomes asymmetric — the compression and expansion phases use different control strategies even though the system is nominally symmetric. The symmetry was never physical. It was an artifact of optimizing one knob at a time.
The through-claim: single-parameter optimization preserves symmetries that multi-parameter optimization reveals as accidental. The system has no reason to be symmetric; it only looked symmetric because we were solving a simpler problem. Adding the second control variable doesn't break the symmetry — it reveals that the symmetry was always broken, and our optimization was too constrained to notice.
The practical surprise: despite the qualitative difference, the efficiency gap between independent and joint optimization is small. The symmetry breaking produces genuinely new strategies but only marginally better outcomes. The structure changes dramatically; the performance barely shifts.
When your optimization discovers a symmetry you thought was fundamental, the symmetry was in your method, not your system.