Momentum-Accelerated Minimization of the Ginzburg-Landau Perimeter Functional
Adds momentum to the gradient-flow minimization of a Ginzburg-Landau diffuse-perimeter energy, replacing the parabolic Allen-Cahn flow with a damped hyperbolic evolution and a convex-splitting FISTA-type time discretization; on Euclidean domains and on weighted graphs it can converge faster at moderately large time steps and recovers the sharp-interface perimeter dynamics in the phase-field limit, with application to semi-supervised classification.
2501.00389
This paper studies accelerating the minimization of a Ginzburg-Landau (diffuse-perimeter) energy - a double-well potential plus gradient penalty whose sharp-interface limit is the perimeter functiona…