Strict Dissipativity and the Turnpike Property in Optimal Control

22 concepts11 papers4 proposals

When planning a long-term journey across a continent, your optimal route rarely resembles a straight line from start to finish. Instead, regardless of where you begin or where you end, you almost always steer toward the nearest major highway, spend the vast majority of your time cruising at a steady speed along that highway, and only exit at the very end to reach your specific destination.

In optimal control theory, this striking phenomenon is known as the turnpike property. For a wide class of sequential decision-making and dynamic optimization problems, the optimal trajectory over a long time horizon spends most of its duration resting near a specific, static optimal steady state—the "turnpike."

But why do optimal trajectories naturally favor these steady states? The answer lies in a deep energetic and structural property of dynamical systems known as strict dissipativity.

Strict Dissipativity as a Cost Driver

To understand why the turnpike property emerges, we must look at how a system stores and expends its "cost energy." A system is said to be strictly dissipative if we can construct a supply rate and a storage function—analogous to an energy landscape—such that any deviation from the optimal steady state strictly penalizes the system's long-term cost-to-go.

Strict dissipativity acts like a gravitational pull toward the steady state. As established by Timm Faulwasser et al. in On turnpike and dissipativity properties of continuous-time optimal control problems, there is a fundamental mathematical equivalence between these concepts: if an optimal control problem exhibits turnpike behavior, it inherently implies that the underlying system is dissipative and operates optimally at that steady state. Strict dissipativity functions as the mathematical engine that forces optimal paths toward equilibrium.

This connection is rooted deeply in the geometry of optimization. In Integral and measure-turnpike properties for infinite-dimensional optimal control systems, Emmanuel Trélat demonstrates that strict strong duality in optimization theory directly implies strict dissipativity. This duality guarantees that the system's cost landscape naturally penalizes off-equilibrium behavior, providing a unified explanation for why optimal trajectories across infinite-dimensional systems exhibit the turnpike property.

The On-Ramp Problem: The Necessity of Controllability

Strict dissipativity explains why the turnpike is desirable, but a system cannot exploit the turnpike if it cannot physically get there. The highway is useless without an on-ramp.

Therefore, the emergence of the turnpike property is tightly constrained by the system's controllability and stabilizability. As Borjan Geshkovski points out in Turnpike in optimal control of PDEs, ResNets, and beyond, the ability to steer a system to a desired state is a more fundamental prerequisite for turnpike behavior than the system's inherent free-dynamics stability. If a system exhibits poor controllability, the mathematical links between dissipativity and the turnpike property begin to break down.

This necessity becomes highly practical in frameworks like economic Model Predictive Control (eMPC), where controllers optimize performance metrics online over a rolling horizon. David Angeli et al. prove in Additional material to the paper "On necessity and robustness of dissipativity in economic model predictive control" that local controllability at the optimal steady state is a strict requirement for guaranteeing convergence. Local controllability ensures that the system can transition to any nearby state within a finite number of steps, allowing the controller to reliably steer the system back to its optimal operating regime despite external disturbances.

Scaling to High Dimensions and Multi-Agent Systems

Does this intuition hold when a system grows from a single trajectory to millions of interacting components, such as robot swarms, traffic flows, or neural networks?

Remarkably, strict dissipativity ensures that the turnpike property scales perfectly. In The turnpike property for mean-field optimal control problems, Martin Gugat et al. show that when moving from individual particle systems to continuous density models (the mean-field limit), strict dissipativity continues to act as a structural constraint that penalizes deviations from the steady state. The authors extend this principle to discrete-time settings in The turnpike property for high-dimensional interacting agent systems in discrete time, proving that high-dimensional multi-agent systems inherit these exact stability characteristics regardless of the number of agents or the discretization time-step.

Furthermore, this convergence is not sluggish. Yizhou Zhou proves in Exponential turnpike property for particle systems and mean-field limit that for large-scale particle systems, both the optimal path and the control effort converge exponentially fast to the target equilibrium. This rapid convergence remains invariant even as the number of particles approaches infinity, justifying why complex, large-scale dynamic control problems can often be safely simplified into static optimization problems.

Applications in Predictive Control and Reinforcement Learning

Understanding the interplay between dissipativity and steady-state convergence allows engineers to design predictive controllers with strong performance guarantees. For instance, Aditi Anand et al. highlight in Data-Driven Predictive Control and MPC: Do we achieve optimality? that systems combining smoothness and dissipativity are far more likely to achieve near-optimal performance because their trajectories naturally settle into stable, well-behaved steady-state distributions. When managing multiple competing objectives, Lars Grüne et al. demonstrate in Multiobjective strict dissipativity via a weighted sum approach that a weighted sum of strictly dissipative stage costs preserves this structural stability, ensuring that multiobjective MPC formulations remain stable and optimal.

This framework also bridges classical control theory with modern reinforcement learning and Markov Decision Processes (MDPs), where future rewards are typically discounted over time. In A New Dissipativity Condition for Asymptotic Stability of Discounted Economic MPC, Mario Zanon introduces a novel "Strong Discounted Strict Dissipativity" condition. By incorporating the optimal value function directly into the dissipativity bound, this approach drastically reduces the conservatism of stability proofs, guaranteeing that discounted systems converge precisely to the optimal steady state even under heavy discounting.

Traditional Dissipativity:   Storage Function Bounds ──> Neighborhood Convergence
Zanon's Refinement:          [Storage Function + Optimal Value Function] ──> Exact Steady-State Convergence

This connection between finite-horizon tracking and infinite-horizon stability proves highly versatile. Dheeraj Narasimha et al. leverage the dissipativity framework in Model predictive control is almost optimal for restless bandits to show that MPC policies achieve near-optimal performance for restless multi-armed bandit problems. Finally, this duality extends even to stochastic survival scenarios: Amarjit Budhiraja et al. demonstrate in Quasistationary Distributions and Ergodic Control Problems that the principal Dirichlet eigenvalue (the decay or extinction rate of a stochastic process) is mathematically identical to the optimal average cost in a corresponding ergodic control problem.

Ultimately, whether managing a fleet of autonomous agents, stabilizing an economic predictive controller, or training a reinforcement learning agent, strict dissipativity provides the underlying energy landscape that makes long-term optimal steady-state operation not just possible, but inevitable.

Key Foundational Work

On turnpike and dissipativity properties of continuous-time optimal control problems (2017) — contributed 10 concept elements to this direction.
Additional material to the paper "On necessity and robustness of dissipativity in economic model predictive control" (2015) — contributed 3 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_strict_dissipativity_38956df3,
  title = {strict dissipativity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/strict-dissipativity/rdr_38956df33d8341838c358f3f08b0718b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_strict_dissipativity_38956df3_v1,
  title = {strict dissipativity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/strict-dissipativity/rdr_38956df33d8341838c358f3f08b0718b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake