Contractor-Expander and Universal Inverse Optimal Positive Nonlinear Control

This paper introduces two novel inverse optimal control frameworks utilizing "contractor and expander functions" to achieve stabilization for general control-affine nonlinear systems constrained to the positive orthant with positive controls, addressing the limitations of conventional methods through asymmetric cost designs and providing explicit universal formulae for applications such as predator-prey models.

The Gist

Here is an explanation of the paper "Contractor-Expander and Universal Inverse Optimal Positive Nonlinear Control" by Miroslav Krstic, translated into simple, everyday language with creative analogies.

The Big Picture: The "Positive" Problem

Imagine you are trying to steer a very special kind of boat.

• Normal Boats: In standard control theory, you can push the boat forward (positive) or backward (negative). If you need to go left, you turn the wheel left; if you need to go right, you turn it right. The "cost" of turning left is the same as turning right.
• This Paper's Boat: This boat is stuck in a river where you can only push forward. You cannot push backward. Furthermore, the boat's speed and position must always be positive numbers (you can't have negative fish or negative money).

This is the world of Positive Systems. It applies to real-life things like:

• Fisheries: You can catch fish (harvest), but you can't "un-catch" them.
• Chemical Reactions: You can add ingredients, but you can't remove them once mixed.
• Bank Accounts: You can deposit money, but you can't have a negative balance (in this specific model).

The author asks: How do we steer this boat to a safe harbor (stability) using only forward pushes, while also making sure we aren't wasting fuel (optimality)?

The Core Challenge: The "Predator-Prey" Boat

To test his ideas, the author uses a classic model called the Predator-Prey system.

• Prey (Fish): They grow on their own.
• Predators (Sharks): They eat the fish.
• The Control: We can only control the harvesting rate of the sharks. We can catch more sharks or catch fewer sharks, but we can never "un-catch" a shark (negative harvesting is impossible).

The Problem:
If the sharks are too many, they eat all the fish, and the system crashes. If there are too few sharks, the fish overpopulate and crash the ecosystem. We need to find the perfect balance.

Standard control methods (like the famous "Sontag formula") assume you can push the boat forward or backward. If you try to use those standard methods here, they often tell you to "push backward" (harvest negative sharks), which is impossible. The paper solves this by inventing a new way to steer that respects the "no backward" rule.

The New Tools: The "Contractor" and the "Expander"

The author introduces two magical functions to solve the steering problem. Think of them as special lenses or gears that change how we look at the problem.

1. The Expander (The Amplifier)

Imagine you are driving a car with a very sensitive gas pedal.

• The Situation: If the sharks are way too many (Predator > Prey), you need to harvest them aggressively.
• The Expander: This function takes the current ratio of Sharks-to-Fish and "expands" it. If the ratio is 2 (twice as many sharks), the Expander might tell you to harvest as if the ratio were 4. It says, "Go harder! Be more aggressive!"
• The Twist: If the sharks are too few, the Expander "attenuates" (softens) your action. It says, "Don't panic, just harvest a little bit."

Why is this cool? It creates a feedback loop that is naturally "inverse optimal." It means the boat reaches the harbor using the least amount of "fuel" (harvesting effort) possible, given the constraints.

2. The Contractor (The Filter)

If the Expander is the gas pedal, the Contractor is the brake pedal logic. It is the mathematical inverse of the Expander.

• The paper shows that if you design your "Expander" correctly, you automatically create a "Contractor" that defines the cost of your actions.
• The Cost: In normal math, the cost of turning left is the same as turning right. Here, the cost is asymmetric.
  • Scenario A: If you harvest too much when sharks are scarce, the "cost" is huge (ecological disaster).
  • Scenario B: If you harvest a little when sharks are abundant, the "cost" is low.
  • The Contractor function calculates this uneven cost, ensuring the controller knows exactly when to be gentle and when to be tough.

The "Universal" Formula: The Magic Steering Wheel

The paper doesn't just solve the shark problem; it creates a Universal Formula.

Think of this like a universal remote control.

• Old Remotes: You had to buy a specific remote for your TV, a different one for your DVD player, and another for your AC. (Specific controllers for specific systems).
• This Paper's Remote: The author creates a single "Universal Formula" that works for any positive system (fish, chemicals, money, traffic) as long as you give it a "map" (a mathematical function called a CLF).

This formula guarantees two things:

1. Stability: The system will always settle down to the safe harbor.
2. Optimality: It will do so in the most efficient way possible, respecting the "no negative values" rule.

The Biological Insight: Why This Matters

The paper explains that this math isn't just abstract; it makes biological sense.

• When Predators Dominate: The system naturally wants to harvest them aggressively. The math confirms this is the "cheapest" way to save the prey.
• When Prey Dominate: The system naturally wants to harvest very little. The math confirms that trying to harvest too much here is "expensive" and dangerous.

The "Expander" mechanism essentially mimics how nature might self-regulate: Amplify the cure when the disease is strong; dampen the cure when the patient is weak.

Summary in One Sentence

This paper invents a new mathematical "steering wheel" that allows us to control systems where you can only push forward (like harvesting fish or adding chemicals), ensuring they stay stable and efficient by using special "Expander" and "Contractor" lenses that turn standard control theory into a tool that respects the laws of nature.

Technical Summary

Here is a detailed technical summary of the paper "Contractor-Expander and Universal Inverse Optimal Positive Nonlinear Control" by Miroslav Krstic.

1. Problem Statement

The paper addresses the inverse optimal stabilization problem for a class of control-affine nonlinear systems where both the state vector $\xi$ and the control input $\omega$ are constrained to the positive orthant ($\xi \in (0, \infty)^n, \omega \in (0, \infty)$).

• System Dynamics: $\dot{\xi} = f(\xi) + g(\xi)\omega$, with an equilibrium at $\xi_e = \mathbf{1}, \omega_e = 1$.
• The Challenge: Classical inverse optimal control (e.g., Sontag's universal formula or $L_gV$ methods) assumes unconstrained inputs and symmetric cost functions (penalties symmetric about zero). These methods fail for positive systems because:
  1. Inputs must be strictly non-negative.
  2. The equilibrium input is often non-zero (e.g., a baseline harvesting rate), requiring asymmetric penalties.
  3. Standard feedback laws may drive the system out of the positive orthant or require negative control inputs to stabilize, which is physically impossible in many applications (ecology, chemistry, pharmacology).
• Goal: To construct feedback laws that are:
  1. Globally asymptotically stabilizing within the positive orthant.
  2. Inverse optimal: They minimize a meaningful, state-dependent, and inherently asymmetric cost function.

2. Methodology

The paper introduces a novel framework based on "Contractor-Expander" functions to redesign stabilizing feedback laws into inverse optimal ones.

A. The Predator-Prey Benchmark

The authors first develop the theory using a predator-prey model (Lotka-Volterra type) with harvesting control:
$$\dot{X} = (1-Y)X, \quad \dot{Y} = (X-U)Y$$
where $X$ is prey, $Y$ is predator, and $U$ is the harvesting rate.

• They utilize a previously designed strict Control Lyapunov Function (CLF): $V(X,Y) = \Omega(1/X) + \Omega(Y/X)$, where $\Omega(S) = S - 1 - \ln(S)$.
• They demonstrate that standard universal formulas fail on this system because there exists a region where the Lie derivative conditions for positive control are not met by simple linear feedback.

B. The Contractor-Expander Mechanism

The core innovation is the relationship between two functions:

1. Expander ($\Sigma$): A strictly increasing function $\Sigma: (0, \infty) \to (0, \infty)$ with $\Sigma(1)=1$. It "expands" the control away from the equilibrium when the state ratio deviates (e.g., amplifies harvesting when predators dominate).
2. Contractor ($\Theta$): The inverse function $\Theta = \Sigma^{-1}$. It satisfies specific conditions (strictly increasing, $\Theta(1)=1$, $\Theta'(1)<1$, and $(\Theta(s)-s)(s-1) < 0$).

The Redesign Process:

• Start with a nominal stabilizing feedback $\omega_0$.
• Apply an Expander to generate the optimal control: $\omega^* = \Sigma(\omega_0)$.
• This transformation ensures that the new control minimizes a cost function $J = \int (q(\xi) + r(\xi)\Psi(\omega)) dt$.
• The Penalty Function $\Psi$ is derived from the Contractor $\Theta$ via an ODE: $\Psi'(s) = \frac{\Psi(s)}{s - \Theta(s)}$. This $\Psi$ is strictly convex, has a minimum at 1, and is asymmetric (unlike quadratic costs).

C. General Frameworks

The paper extends these concepts from the specific benchmark to general systems via two main theorems:

1. Theorem 5 (Expander-based Redesign): If a nominal controller $\omega_0$ and a strict CLF exist such that a sign-compatibility condition holds, then applying an expander $\Sigma$ yields an inverse optimal controller.
2. Theorem 6 (Constructive Direct Design): A more restrictive but fully constructive framework. It allows the user to choose a Contractor $\Theta$ and a weight function $r$ to directly synthesize an inverse optimal controller without needing a pre-existing nominal controller.

3. Key Contributions

1. First Inverse Optimal Characterization for Positive Systems: The paper provides the first rigorous inverse optimal characterization of a strict CLF for positive systems, moving beyond simple stabilization to optimality with asymmetric costs.
2. Contractor-Expander Theory: Introduces a new mathematical mechanism ($\Theta \circ \Sigma = \text{Id}$) to redesign feedback laws. This mechanism preserves Lyapunov stability, enforces positivity, and induces biologically/ecologically meaningful asymmetric costs.
3. Universal Formulae for Positive Systems:
   • Derives a Universal Stabilizer (Theorem 3) that guarantees global asymptotic stability for positive systems (though not necessarily inverse optimal).
   • Derives a Universal Inverse Optimal Formula (Theorem 4) under a stronger CLF condition ($L_{f+g}V < 0$ when $L_gV \ge 0$). This formula is a positive-input analog of Sontag's formula but differs significantly in structure (using $b^2$ instead of $b^4$ under the square root) to handle one-sided constraints.
4. Biological Interpretation: The paper interprets the mathematical results in the context of ecology.
   • Asymmetry: The optimal cost penalizes "over-harvesting" differently than "under-harvesting."
   • Gain Margin: The inverse optimal controller provides a gain margin of $[1/2, \infty)$, meaning it is robust to significant variations in control gain.
   • Ecological Logic: The controller aggressively harvests predators when they dominate (to save prey) but attenuates harvesting when prey are scarce (to allow predator recovery), a behavior that standard symmetric controllers cannot replicate.

4. Results

• Stability: The proposed controllers guarantee Global Asymptotic Stability (GAS) of the equilibrium $\xi = \mathbf{1}$ within the positive orthant $(0, \infty)^n$.
• Optimality: The controllers minimize an infinite-horizon cost functional with:
  • A state cost $q(\xi)$ that is positive definite.
  • A control cost $r(\xi)\Psi(\omega)$ where $\Psi$ is strictly convex and asymmetric.
• Simulation: Simulations on the predator-prey model demonstrate that the inverse optimal controller reduces the total cost (state deviation + control effort) by approximately 20% compared to a nominal backstepping controller in predator-dominant scenarios, while maintaining stability in prey-dominant scenarios.
• Explicit Designs: The paper provides explicit "Sontag-like" universal formulae for positive systems, offering a practical alternative to complex ad-hoc designs.

5. Significance

• Theoretical Advancement: This work bridges a critical gap in nonlinear control theory. It extends the powerful $L_gV$ inverse optimal framework to the constrained domain of positive systems, which are ubiquitous in engineering and science but previously lacked systematic inverse optimal design tools.
• Practical Applicability: The methodology is directly applicable to real-world systems where negative inputs are impossible (e.g., drug infusion rates, traffic flow, chemical concentrations, population management).
• Design Flexibility: By introducing the "Contractor" and "Expander" functions, the paper offers a parametric family of controllers. Engineers can tune the "aggressiveness" of the control and the shape of the cost function by selecting different Contractor functions ($\Theta$), allowing for trade-offs between performance and control effort.
• Biological Relevance: The results provide a mathematical justification for why certain ecological management strategies (asymmetric harvesting) are optimal, offering a new perspective on bio-ecological control.

In summary, Krstic's paper establishes a rigorous, constructive, and flexible framework for designing inverse optimal controllers for positive nonlinear systems, solving a long-standing problem in the field by introducing novel "Contractor-Expander" transformations that respect the physical constraints of positivity while ensuring global stability and optimality.