⚡ electrical engineering

Characterization and Computation of Feedback Nash Equilibria in Scalar Discounted N-Player Linear Quadratic Games

This paper investigates feedback Nash equilibria in scalar discounted N-player linear quadratic games by distinguishing between finite-cost and stable equilibria, deriving existence conditions for up to $2^N-2$ solutions in symmetric cases, and proposing numerical methods to compute all such equilibria.

Original authors: Chiara Cavalagli, Alberto Bemporad, Mario Zanon

Published 2026-06-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chiara Cavalagli, Alberto Bemporad, Mario Zanon

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a group of $N$ people trying to steer a single, wobbly cart (the "system") down a long road. Each person has a hand on the steering wheel, and they can all push or pull the cart at the same time. However, they are all selfish: each person wants to minimize their own specific "effort cost" (like getting tired or burning fuel) while keeping the cart from going too far off course.

This paper is about figuring out the perfect balance where no single person can improve their own situation by changing their steering strategy, assuming everyone else keeps doing exactly what they are doing. In game theory, this perfect balance is called a Nash Equilibrium.

Here is a breakdown of the paper's key discoveries, using simple analogies:

1. The "Discount" Factor: Valuing the Present vs. The Future

In many real-world scenarios (like economics or AI), people don't care about the distant future as much as the immediate present. The paper introduces a "discount factor" ( $\gamma$ ). Think of this as a pair of glasses that makes future problems look blurry and less important.

Without the glasses: Everyone worries equally about the cart crashing 100 years from now.
With the glasses: Everyone only really cares about the next few seconds.

The authors found that wearing these "discount glasses" changes the rules of the game. Sometimes, a strategy looks perfect for minimizing immediate effort (a "finite-cost" equilibrium), but it actually sends the cart careening off a cliff in the long run.

2. The Big Discovery: "Good" vs. "Stable" Equilibria

The paper makes a crucial distinction between two types of "perfect balances":

Feedback Nash Equilibrium (FNE): A strategy where everyone is happy with their current effort, and the total "cost" they pay is a manageable number.
Stable FNE: A strategy where everyone is happy, AND the cart actually stays on the road forever.

The Analogy: Imagine a group of drivers trying to park a car.

A non-stable equilibrium is like everyone agreeing to press the gas pedal just hard enough to keep the car moving at a speed that feels "cheap" right now, but the car is actually speeding up uncontrollably and will eventually crash. The cost is finite for now, but the system is unstable.
A stable equilibrium is where they agree on a speed that keeps the car moving safely forever.

The authors discovered that when you use the "discount glasses," you can easily find those "speeding up but cheap" solutions. They proved that just because a solution has a finite cost doesn't mean the system is safe. They provided a specific "safety check" (a mathematical condition) to ensure the cart stays on the road.

3. Finding All the Solutions (The "Map")

Usually, when people try to solve these games, they just look for one solution. But this paper is like a cartographer who wants to draw the entire map of every possible solution.

They developed a method to find every single possible balance point, not just the most obvious one.
They found that depending on the settings, there can be many different ways the group can balance out (multiplicity). It's like finding that there are 10 different ways to arrange the drivers' hands on the wheel where no one wants to move, but only a few of those ways keep the car from crashing.

4. The "Symmetric" Case: When Everyone is Identical

The paper gets even more interesting when all players are identical (they have the same goals and the same "cost" for steering).

The "Mirror" Effect: If everyone is the same, there is always one solution where everyone does the exact same thing (a symmetric equilibrium). The authors found a neat, closed-form formula (a direct recipe) to calculate this specific solution.
The "Twins" Effect: They also found that there are other solutions where the group splits up. For example, in a group of 7, maybe 3 people push left and 4 push right, or 1 pushes hard and 6 push soft. These are called "hyperbolic" equilibria.
The Limit: They proved that in a group of $N$ players, there can be at most $2N - 1$ different ways to balance the game. It's like saying a puzzle with 7 pieces has a maximum of 127 different ways to fit them together perfectly.

5. What the Experiments Showed

The authors ran computer simulations to test their theories:

Heterogeneity (Different Players): When players are very different from each other (some care a lot about cost, others don't), the number of possible "perfect balances" drops. It's harder to find a compromise when everyone wants something different.
The Danger of Discounting: When they turned up the "discount" (making players care only about the immediate moment), they found many solutions that looked good mathematically but were actually dangerous (the cart would crash). This confirms that you must check for "stability" separately from just "low cost."

Summary

In short, this paper is a guide for a group of selfish agents trying to control a system. It warns them: "Just because you found a strategy that minimizes your immediate effort doesn't mean the system won't crash later." It provides a complete toolkit to find every possible strategy, check if it's safe, and understand how the number of options changes based on how much the players care about the future.

Technical Summary: Characterization and Computation of Feedback Nash Equilibria in Scalar Discounted N-Player Linear Quadratic Games

Problem Statement
This paper addresses the problem of finding Feedback Nash Equilibria (FNE) in scalar, infinite-horizon, discounted, $N$ -player Linear Quadratic (LQ) games. The system dynamics are governed by a scalar state $x_t$ influenced by $N$ players' controls $u_{i,t}$ and a fixed parameter $a$ . Each player $i$ seeks to minimize a discounted quadratic cost function involving state and control weights ( $\sigma_i, r_i$ ) and a discount factor $\gamma \in (0, 1]$ .

A critical distinction is made between the convergence of the discounted cost and the stability of the original closed-loop system. While standard LQ game literature often assumes stability, the inclusion of a discount factor allows for equilibria where the cost is finite (convergent) but the closed-loop system is unstable. The paper formally distinguishes between a general FNE and a Stable FNE (where the closed-loop spectral radius is strictly less than 1).

Methodology
The authors develop a parametric characterization of the equilibrium policies to compute the full set of equilibria, moving beyond methods that typically identify only a single solution.

First-Order Conditions and Stability Analysis:
The optimality conditions for a player's best response are derived as a polynomial system. The paper analyzes the roots of these conditions, denoted as $h^{\pm}_i(k_i)$ . It is proven that for the discounted case, only one branch ( $h^-_i$ ) generally yields stable equilibria, provided a specific condition on the cost weights and discount factor holds: $\sigma_i > \frac{(1-\gamma)^2}{4\gamma^2}$ . When this condition is met, all FNE are guaranteed to be stabilizing.
Parametric Characterization (General Setting):
By introducing an aggregate variable $S^* = \sum k_i$ , the system of coupled best-response equations is reduced to a single scalar equation. The authors show that for a fixed aggregate $S^*$ , individual policies $k_i$ can be expressed as hyperbolic roots. This leads to a system where $S^*$ must satisfy a sum of these roots over all possible sign combinations ( $\epsilon_j \in \{-1, 1\}^N$ ).
- Algorithm 1: A numerical method is proposed to compute all FNE by solving the resulting scalar nonlinear equations for each of the $2^N$ sign combinations. The complexity is $O(2^N)$ .
Symmetric Setting Analysis:
In the case where all players share identical cost parameters ( $\sigma_i = \sigma$ ), the analysis is refined:
- Closed-Form Solution: A unique symmetric FNE is derived in closed form (Theorem 4.1).
- Multiplicity and Hierarchy: The paper investigates non-symmetric equilibria where players choose between two hyperbolic policy values, $k_+$ and $k_-$ . The existence of these equilibria depends on the number of players $p$ choosing the $k_+$ strategy.
- Algorithm 2: An alternative numerical scheme is proposed for the symmetric case. It iteratively checks a condition on a scalar function $\Phi_p(k)$ to determine the existence of hyperbolic pairs. This reduces the complexity to $O(\lfloor N/2 \rfloor)$ .
- Multiplicity Bounds: The paper derives conditions for the existence of up to $M \le 2N - 2$ hyperbolic equilibria (plus the symmetric one, totaling $2N - 1$ ).

Key Contributions

Stability Distinction: The paper explicitly identifies that finite-cost equilibria in discounted LQ games do not necessarily stabilize the underlying system. It provides a sufficient condition (Eq. 11) ensuring that all FNE are stabilizing.
Complete Computation: It presents the first complete algorithm capable of computing all FNE for discounted scalar $N$ -player LQ games, addressing the gap in literature where methods usually target a single equilibrium.
Symmetric Closed-Form: A closed-form expression for the symmetric FNE is derived, along with necessary and sufficient conditions for the existence of the maximal number of equilibria in the symmetric case.
Parametric Framework: The work extends the aggregate-value parametrization from undiscounted games to the discounted setting, revealing how the discount factor alters the solution landscape.

Results

Multiplicity Dependence: Numerical experiments demonstrate that equilibrium multiplicity is highest when players have similar cost parameters and decreases as the game becomes more heterogeneous.
Role of Discount Factor: Simulations with $N=7$ show that low discount factors ( $\gamma \in [0.1, 0.5]$ ) can lead to the emergence of finite-cost but non-stabilizing equilibria. As $\gamma$ increases, the set of stable FNE and FNE coincides.
Algorithm Performance: Algorithm 1 is shown to be practical for moderate $N$ despite exponential complexity, while Algorithm 2 offers a more efficient approach for symmetric games.

Significance
The paper claims to provide the first complete algorithmic framework for computing all equilibria in discounted scalar LQ games. It highlights a previously under-recognized phenomenon: the potential for discounted objectives to yield equilibria that are optimal in terms of cost convergence but unstable in terms of system dynamics. By establishing conditions for stability and providing tools to enumerate all solutions, the work extends the theoretical understanding of scalar LQ games beyond the undiscounted case and offers a rigorous basis for analyzing strategic interactions in economic and control settings where future losses are discounted.