Average cost optimal control under weak ergodicity hypotheses: Relative value iterations

Ari Arapostathis; Vivek S. Borkar

Home/Archive/arXiv

arXiv · 1902.01048

Average cost optimal control under weak ergodicity hypotheses: Relative value iterations

Ari Arapostathis·Vivek S. Borkar

View original source ↗Read source document ↗Export metadata ↓

Abstract

We study Markov decision processes with Polish state and action spaces. The action space is state dependent and is not necessarily compact. We first establish the existence of an optimal ergodic occupation measure using only a near-monotone hypothesis on the running cost. Then we study the well-posedness of Bellman equation, or what is commonly known as the average cost optimality equation, under the additional hypothesis of the existence of a small set. We deviate from the usual approach which is based on the vanishing discount method and instead map the problem to an equivalent one for a controlled split chain. We employ a stochastic representation of the Poisson equation to derive the Bellman equation. Next, under suitable assumptions, we establish convergence results for the 'relative value iteration' algorithm which computes the solution of the Bellman equation recursively. In addition, we present some results concerning the stability and asymptotic optimality of the associated rolling horizon policies.

Explore related subjects

math.OC cs.SY eess.SY

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Ari Arapostathis, Vivek S. Borkar. 2023-08-14. Average cost optimal control under weak ergodicity hypotheses: Relative value iterations. https://arxiv.org/abs/1902.01048

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

AdGT: Decentralized Gradient Tracking with Adaptive Per-Agent Stepsizes

In decentralized optimization, gradient-tracking methods typically rely on a single global stepsize. This choice can be conservative when agents have local objectives with different smoothness constants, since the stepsize must remain stable for the agent with the largest smoothness constant. This paper proposes AdGT, a decentralized gradient-tracking method in which each agent adapts its own stepsize using local gradient variation and a single global safety factor. The method reduces fixed-stepsize tuning effort and allows agents to exploit local smoothness information during the iterations. For smooth and strongly convex local objectives over undirected networks, we prove that the analyzed AdGT update converges linearly to the exact consensus optimizer. We also study two adaptive stepsize updates that use changes in the gradient-tracking direction. We characterize when the corresponding candidate determines the stepsize and prove conditional lower and upper stepsize bounds and linear convergence under an additional relative tracking-disagreement condition. Experiments on logistic regression, ridge regression, synthetic quadratic problems, and a linear-regression benchmark against state-of-the-art decentralized solvers show that AdGT often reaches a given accuracy in fewer iterations or gradient evaluations than tuned fixed-stepsize GT and the tested baselines, especially under heterogeneous local smoothness. In the topology experiments, each tested AdGT update uses one common safety factor across all graphs, whereas the fixed GT stepsize is tuned separately for each graph and seed.

math.OC↗

arXiv2026

Brockett cost function for symplectic eigenvalues

The sum of symplectic eigenvalues and corresponding eigenvectors of symmetric positive-definite matrices in the sense of Williamson's theorem can be computed via minimization of a trace cost function under the symplecticity constraint. Optimal solutions to this problem only offer a symplectic basis for the symplectic eigenspace corresponding to the sought symplectic eigenvalues. In this note, we introduce a Brockett cost function and investigate its properties and the connection with the symplectic eigenvalues and eigenvectors of the considered matrix. Specifically, we prove that any stationary point consists of symplectic eigenvectors, characterize the saddle points and global minimizers based on which the trace minimization theorem for the symplectic eigenvalues is re-established, and the nonexistence of local nonglobal minimizers is justified.

math.OC↗

arXiv2026

Riemannian Bilevel Optimization with Gradient Aggregation

We study bilevel optimization on Riemannian manifolds when the lower-level solution set is a positive-dimensional submanifold, so that implicit differentiation fails. We propose Riemannian Bilevel Descent Aggregation (RBDA), which extends bilevel descent aggregation to manifolds. Its inner loop aggregates the lower-level descent direction with the upper-level gradient under a decaying multiplier, and its hypergradient is the reverse-mode derivative of the unrolled loop. Under geodesic convexity and quadratic growth of the lower level, the inner iterates converge to a point of the optimistic solution set at a polynomial rate. Approximate minimizers of the objective with a finite number of inner iterations converge to minimizers of the optimistic value. In the experiments RBDA selects the optimistic solution where the implicit and unrolled estimators remain at the initial point or stop at a larger query loss. While each of its outer steps costs more than that of the unrolled estimator, it attains the highest test accuracy in data hyper-cleaning.

math.OC↗