Geometric Thermodynamics of Scallop Motion with Two Control Parameters
arXiv:2608.24158
2026
Dynamics
1 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper provides a transferable geometric mechanism: a force-free internal coordinate generates an effective connection, while slow cyclic modulation of two controls produces a geometric phase proportional to Berry-Sinitsyn-Nemenman curvature. The same control loop carries a thermodynamic cost described by a Riemannian metric, creating a quantitative displacement-versus-dissipation trade-off rather than a generic schedule heuristic. A neural analogue is a two-control optimizer whose stochastic internal state is driven around small loops in learning-rate and momentum or noise space, with loop orientation and enclosed area controlling a reproducible net update. The key tests are orientation reversal, vanishing effect for one control, quadratic small-area scaling, and a measurable excess-cost metric.
Ideas from this paper
Unverified
2026
Augment an optimizer with two slowly and periodically modulated controls, such as learning rate and momentum or learning rate and gradient-noise scale. The optimizer state then traces a loop in control space; nonzero curvature can create a net parameter displacement that depends on loop orientation, even when the controls return to their initial values. Use curvature estimates to select loops that produce useful descent while penalizing loops with excessive dissipation.
Useful6/10
Difficulty6/10
Novelty8/10