Scaling regimes of the Kuramoto-Sivashinsky equation from the functional renormalization group

What the math gives to ML

The paper provides a constructive functional-renormalization-group treatment of stochastic Kuramoto–Sivashinsky dynamics using a smooth cutoff, avoiding singular momentum derivatives produced by sharp-cutoff Wilsonian RG. Its transferable mechanism is a scale-dependent effective dynamics whose two-point response changes across three measurable regimes: KPZ scaling with dynamical exponent z=3/2, Edwards–Wilkinson scaling with z=2, and an inviscid regime with z=1 caused by vanishing effective viscosity. A neural-network analogue is a smooth spectral, mode-wise optimizer that estimates effective damping of parameter-space modes and changes preconditioning or regularization when fitted scaling exponents cross these predicted regime boundaries.