Strong convergence of FISTA iterates under Hölderian and quadratic growth conditions

Applied Nanoscience · Springer Nature · 2026-11-02 · DOI: 10.1007/s10107-026-02431-4

Abstract

Introduced by Beck and Teboulle in [1], FISTA (Fast Iterative Shrinkage-Thresholding Algorithm) is a first-order method widely used in convex optimization. Adapted from Nesterov’s accelerated gradient method for convex functions [2], the generated sequence guarantees a worst-case objective decay of $$\mathcal {O}(n^{-2})$$. In this paper, we prove, under mild growth assumptions, that this standard sequence converges strongly to a minimizer, i.e. that the length of the iterates’ trajectory is finite. Unlike approaches that require algorithmic modifications such as Tikhonov regularization, our strong convergence results follow naturally from the intrinsic geometry of the objective. This result applies to globally subanalytic functions with bounded sublevel sets, covering a wide class of problems. This convergence property, which to our knowledge has not been previously established without assuming uniqueness of the minimizer, is associated with improved convergence rates for the function values. The proposed analysis is based on a preliminary study of the asymptotic vanishing damping system introduced by Su et al. [3] to model Nesterov’s accelerated gradient method in a continuous setting. Novel improved convergence results are also established for the solutions of this dynamical system, including the finite length of the trajectory under the aforementioned geometric conditions.

Read at publisher →

Similar articles