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.

Differential Structure and Flow equations on Rough Path Space

Geometric rough paths on infinite dimensional spaces

Calabi Flow and Projective Embeddings

Rough differential equations in the flow approach

Pathwise Itô Calculus for Rough Paths and Rough PDEs with Path Dependent Coefficients

Some family of q-vector fields on path spaces

A new definition of rough paths on manifolds

Constrained rough paths

A Transfer Principle for Branched Rough Paths

A gradient flow on control space with rough initial condition

Rough flows

Rough differential equations with unbounded drift term

Differential Graded Structures on Path Categories

Topology of spaces of smooth functions and gradient-like flows with prescribed singularities on surfaces

Some Aspects of Differential Topology of Subcartesian Spaces

A Version of Hörmander's Theorem for Markovian Rough Paths

On Differential and Riemannian Calculus on Wasserstein Spaces

Topologies on unparameterised rough path space

Schwarzian derivatives, projective structures, and the Weil–Petersson gradient flow for renormalized volume

Fragment-wise differentiable structures

Rough path theory