Asynchronous and Distributed Tracking of Time-Varying Fixed Points

Abstract

This paper develops an algorithmic framework for tracking fixed points of time-varying contraction mappings. Analytical results for the tracking error are established for the cases where: (i) the underlying contraction self-map changes at each step of the algorithm; (ii) only an imperfect information of the map is available; and, (iii) the algorithm is implemented in a distributed fashion, with communication delays and packet drops leading to asynchronous algorithmic updates. The analytical results are applicable to several classes of problems, including time-varying contraction mappings emerging from online and asynchronous implementations of gradient-based methods for time-varying convex programs. In this domain, the proposed framework can also capture the operating principles of feedback-based online algorithms, where the online gradient steps are suitably modified to accommodate actionable feedback from an underlying physical or logical network. Examples of applications and illustrative numerical results are provided. arXiv:1804.09768v2 [math.OC]

Real-Time Feedback-Based Optimization of Distribution Grids: A Unified Approach

“…We next analyze the proposed algorithm under the assumption of synchronous updates in steps [S1] and [S2] above. The analysis of the asynchronous case can be carried out similarly on expense of heavier notation and further assumptions; see for example [45].…”

Model-Free Primal-Dual Methods for Network Optimization with Application to Real-Time Optimal Power Flow

“…Zero-order (or gradient-free) optimization has been a subject of interest in the optimization, control, and machine learning communities for decades. The seminal paper of Kiefer and Wolfowitz [25] introduced a one-dimensional variant of approximation (6); for a d-dimensional problem, it perturbs each dimension separately and requires 2d function evaluations. The simultaneous perturbation stochastic approximation (SPSA) algorithm [34] uses zero-mean independent random perturbations, requiring two function evaluations at each step.…”

“…It is particularly true for fast communication and control. In fact, the asynchronicity can be modeled as an additional noise source, and can be analyzed similarly to, e.g., [6]. In the next section, we numerically evaluate the sensitivity of our approach to different levels of noise.…”

Technical Report: A Totally Asynchronous Algorithm for Tracking Solutions to Time-Varying Convex Optimization Problems

“…This paper is also closely related to a growing body of research on feedback optimization, which often includes time-varying optimization problems. Relevant work develops prediction-correction algorithms for problems with timevarying objective functions in both the centralized [25], [26] and distributed cases [27], [28]. Specifically, [27] considers an objective function that is the sum of locally available functions at each node and uses higher-order information to perform the prediction step.…”

“…In [28,Remark 3], it is noted that distributed time-varying optimization by agents with different computational abilities is subject to ongoing research. To the best of our knowledge, this problem remains open, and we present a solution to it by allowing agents to both compute and communicate totally asynchronously.…”