arXiv subjects

Yian Yao

Publications and source records attributed to Yian Yao.

3 recordsLinked to original sources

arXiv2026

The Geometry Behind Diffusion and Flow Matching: Gradient Flows and Geodesics in Wasserstein Space

The space $\mathcal{P}_2(\mathbb{R}^d$) of probability measures with finite second moment carries a natural geometry: the quadratic Wasserstein distance W_2 makes it a complete metric space and, following Otto, a (formal) Riemannian manifold whose geodesics are the optimal-transport interpolations. On this manifold, the gradient flow of the free energy F(rho) = KL(rho || π) is exactly the Fokker-Planck equation, and its implicit-Euler discretization is the JKO scheme. This is the geometry underlying diffusion models: the forward process descends the free energy, and each denoising step realizes one JKO step, which recovers DDPM, DDIM, NCSN/SMLD, and Energy Matching; this is one scheme, not separate theories. The same manifold supports a second variational principle. Its geodesics - the minimum-action curves of the Benamou-Brenier formula - are precisely the optimal-transport paths that Flow Matching learns. Fixing both endpoints and following the geodesic, generation becomes a deterministic ODE along a straight line, hence far fewer sampling steps. Placing both families of models on one manifold makes their relationship exact: diffusion follows a free-energy gradient flow, an initial-value problem; optimal-transport Flow Matching follows a Wasserstein geodesic, a boundary-value problem. The two reach the same endpoints along different paths.

cs.AI↗

arXiv2021

Computing the Invariant Circle and its Stable Manifolds for a 2-D Map by the Parameterization Method: Effective Algorithms and Rigorous Proofs of Convergence

We present and analyze rigorously a quadratically convergent algorithm to compute an invariant circle for 2-dimensional maps along with the corresponding foliation by stable manifolds. We prove that when the algorithm starts from an initial guess that satisfies the invariance equation very approximately (depending on some condition numbers, evaluated on the approximate solution), then the algorithm converges to a true solution which is close to the initial guess. The convergence is faster than exponential in smooth norms. The distance from the exact solution and the approximation is bounded by the initial error. This allows validating the numerical approximations (a-posteriori results). It also implies the usual persistence formulations since the exact solutions of the invariance equation for a model are approximate solutions for a similar model. The algorithm we present works irrespective of whether the dynamics on the invariant circle is a rotation or it is phase-locked. The condition numbers required do not involve any global qualitative properties of the map. They are obtained by evaluating derivatives of the initial guess, derivatives of the map in a neighborhood of the guess, performing algebraic operations, and taking suprema. The proof of the convergence is based on a general Nash-Moser implicit function theorem specially tailored for this problem. The Nash-Moser procedure has unusual properties. As it turns out, the regularity requirements are not very severe (only 2 derivatives suffice). We hope that this implicit function theorem may be of independent interest and have presented it in a self-contained appendix. The algorithm in this paper is very practical since it converges quadratically, and it requires moderate storage and operation count. Details of the implementation and results of the runs are described in a companion paper [YdlL21].

math.DS↗

arXiv2021

Computing the Invariant Circle and the Foliation by Stable Manifolds for a 2-D Map by the Parameterization Method: Numerical Implementation and Results

We present and implement an algorithm for computing the invariant circle and the corresponding stable manifolds for 2-dimensional maps. The algorithm is based on the parameterization method, and it is backed up by an a-posteriori theorem established in [YdlL21]. The algorithm works irrespective of whether the internal dynamics in the invariant circle is a rotation or it is phase-locked. The algorithm converges quadratically and the number of operations and memory requirements for each step of the iteration is linear with respect to the size of the discretization. We also report on the result of running the implementation in some standard models to uncover new phenomena. In particular, we explored a bundle merging scenario in which the invariant circle loses hyperbolicity because the angle between the stable directions and the tangent becomes zero even if the rates of contraction are separated. We also discuss and implement a generalization of the algorithm to 3 dimensions, and implement it on the 3-dimensional Fattened Arnold Family (3D-FAF) map with non-resonant eigenvalues and present numerical results.

math.DS↗