Exponential Decay of Sensitivity in Graph-Structured Nonlinear Programs

  title={Exponential Decay of Sensitivity in Graph-Structured Nonlinear Programs},
  author={Sungho Shin and Mihai Anitescu and Victor M. Zavala},
  journal={SIAM J. Optim.},
We study solution sensitivity for nonlinear programs (NLPs) whose structure is induced by a graph $\mathcal{G}=(\mathcal{V},\mathcal{E})$. These graph-structured NLPs arise in many applications such as dynamic optimization, stochastic optimization, optimization with partial differential equations, and network optimization. We show that the sensitivity of the primal-dual solution at node $i\in \mathcal{V}$ against a data perturbation at node $j\in \mathcal{V}$ is bounded by $\Upsilon \rho^{d_… 

Figures and Tables from this paper

Condensed interior-point methods: porting reduced-space approaches on GPU hardware

This work proposes a novel reduced-space IPM algorithm that condenses the KKT system into a dense matrix whose size is proportional to the number of degrees of freedom in the problem, and adapts their workflow so that the vast majority of computations are accelerated on GPUs.

Near-Optimal Performance of Stochastic Predictive Control

Under the standard stabilizability and detectability conditions, the dynamic regret of SPC is exponentially small in the prediction horizon length, and SPC can achieve near-optimal performance—the expected performance can be made arbitrarily close to the optimal solution—at a substantially reduced computational expense.

An Example Article

  • 2022

Near-Optimal Distributed Linear-Quadratic Regulator for Networked Systems

It is shown that under mild assumptions, including stabilizability, detectability, and a polynomially growing graph condition, the performance between κ -distributed control and centralized optimal control becomes exponentially small in κ .

Exponential Decay of Sensitivity in Dynamic Optimization: A Graph-Theoretic Approach

We study exponential decay of sensitivity (EDS) in dynamic optimization (DO) problems, which include model predictive control (MPC) and moving horizon estimation (MHE). The property indicates that

Controllability and Observability Imply Exponential Decay of Sensitivity in Dynamic Optimization

Decentralized Online Convex Optimization in Networked Systems

This work proposes a novel online algorithm, Localized Predictive Control (LPC), which generalizes predictive control to multi-agent systems and achieves a competitive ratio of 1 + O ( ρ kT ) + ˜ O ( rS ) in an adversarial setting, which is the first competitive ratio bound on decentralized predictive control for networked online convex optimization.

A Consensus-Based Alternating Direction Method for Mixed-Integer and PDE-Constrained Gas Transport Problems

. We consider dynamic gas transport optimization problems, which lead to large-scale and nonconvex mixed-integer nonlinear optimization problems (MINLPs) on graphs. Usually, the resulting instances

Constrained Policy Optimization for Stochastic Optimal Control under Nonstationary Uncertainties

—This article presents a constrained policy opti- mization approach for the optimal control of systems under nonstationary uncertainties. We introduce an assumption that we call Markov embeddability

A Julia Framework for Graph-Structured Nonlinear Optimization

This work presents a Julia framework for modeling and solving graph-structured nonlinear optimization problems and demonstrates the scalability of the framework by targeting a large-scale, stochastic gas network problem that contains over 1.7 million variables.



Perturbed Kuhn-Tucker points and rates of convergence for a class of nonlinear-programming algorithms

This paper establishes quantitative bounds for the variation of an isolated local minimizer for a general nonlinear program under perturbations in the objective function and constraints. These bounds

Strongly Regular Generalized Equations

A regularity condition is introduced for generalized equations and it is shown to be in a certain sense the weakest possible condition under which the stated properties will hold.

Overlapping Schwarz Decomposition for Nonlinear Optimal Control

It is shown that the algorithm exhibits local convergence and that the convergence rate improves exponentially with the size of the overlap, and that this condition holds for nonlinear OCPs under a uniform second-order sufficient condition, a controllability condition, and a uniform boundedness condition.

A Parallel Decomposition Scheme for Solving Long-Horizon Optimal Control Problems

It is proved that linear quadratic problems satisfy the asymptotic decay condition, and it is discussed numerical strategies to determine if the condition holds in more general cases.


  • Mathematics
One of the key components of chemical plant operability is flexibility - the ability to operate over a range of conditions while satisfying performance specifications. A general framework for

Abstract nonlinear sensitivity and turnpike analysis and an application to semilinear parabolic PDEs

An exponential turnpike result is proved and it is shown that perturbations of the extremal equation’s dynamics, e.g., discretization errors decay exponentially in time, can be used for very efficient discretized schemes in a Model Predictive Controller, where only a part of the solution needs to be computed accurately.

Overlapping Schwarz Decomposition for Constrained Quadratic Programs

It is shown that convergence is guaranteed if the overlap is sufficiently large and that the convergence rate improves exponentially with the size of the overlap, and conditions under which this property holds for constrained QPs (as those found in network optimization and optimal control), thus extending existing work that addresses unconstrained QPs.

Diffusing-Horizon Model Predictive Control

We present a new time-coarsening strategy for model predictive control (MPC) that we call diffusing-horizon MPC. This strategy seeks to overcome the computational challenges associated with optimal


Superconvergence of Online Optimization for Model Predictive Control

It is proved that the one-Newton-step-per-horizon, online, lag-$L$, model predictive control algorithm for solving discrete-time, equality-constrained, nonlinear dynamic programs exhibits a behavior that is called superconvergence; that is, the tracking error with respect to the full horizon solution is not only stable for successive horizon shifts, but also decreases with increasing shift order.