Abstract:We consider the (sub-Riemannian type) control problem of finding a path going from an initial point $x$ to a target point $y$, by only moving in certain admissible directions. We assume that the corresponding vector fields satisfy the bracket-generating (Hörmander) condition, so that the classical Chow-Rashevskii theorem guarantees the existence of such a path. One natural way to try to solve this problem is via a gradient flow on control space. However, since the corresponding dynamics may have saddle points, any convergence result must rely on suitable (e.g. random) initialisation. We consider the case when this initialisation is irregular, which is conveniently formulated via Lyons' rough path theory. We show that one advantage of this initialisation is that the saddle points are moved to infinity, while minima remain at a finite distance from the starting point. In the step $2$-nilpotent case, we further manage to prove that the gradient flow converges to a solution, if the initial condition is the path of a Brownian motion (or rougher). The proof is based on combining ideas from Malliavin calculus with Łojasiewicz inequalities. A possible motivation for our study comes from the training of deep Residual Neural Nets, in the regime when the number of trainable parameters per layer is smaller than the dimension of the data vector.

Quartic Gradient Flow

On a Complex Hessian Flow

Flow by Gauss Curvature to the Aleksandrov and Dual Minkowski Problems

Flow by Gauss Curvature to the $ L_p $ Dual Minkowski Problem

Geometric structure of pseudo-plane quadratic flows

Determining the Quark Mass with the Gradient Flow

A gradient flow of Spin(7)-structures

The gradient flow in simple field theories

Two novel numerical methods for gradient flows: generalizations of the Invariant Energy Quadratization method

A gradient flow on control space with rough initial condition

Minimal slopes and bubbling for complex Hessian equations

Gradient flows in asymmetric metric spaces and applications

Hyperbolic Gradient Flow: Evolution of Graphs in R^n+1

Short-flow-time expansion of quark bilinears through next-to-next-to-leading order QCD

Novel approach for computing gradients of physical observables

Gradient Flow of the Sinai-Ruelle-Bowen Entropy

Gradient-flowed order parameter for spontaneous gauge symmetry breaking

Applications of the Perturbative Gradient Flow

Gradient flows on graphons: existence, convergence, continuity equations

Construction of unstable concentrated solutions of the Euler and gSQG equations