Constrained zero-sum LQ differential games for jump-diffusion systems with regime switching and random coefficients

This paper establishes the open-loop solvability and derives a closed-loop representation for a cone-constrained two-player zero-sum stochastic linear-quadratic differential game involving jump-diffusion systems with regime switching and random coefficients, utilizing forward-backward stochastic differential equations and newly proposed multidimensional indefinite extended stochastic Riccati equations with jumps.

The Gist

Imagine you are the captain of a ship navigating through a chaotic, stormy sea. But this isn't just any sea; it's a world where the weather changes instantly and unpredictably, the map itself shifts, and the ship's engine behaves differently depending on the time of day.

Now, imagine there are two captains on this ship, but they have completely opposite goals:

• Captain A wants to keep the ship as safe and stable as possible (minimizing damage).
• Captain B wants to push the ship as hard as possible to reach a destination quickly, even if it risks crashing (maximizing speed/gain).

This paper is about finding the perfect "dance" between these two captains when the rules of the game are tricky.

Here is a breakdown of the paper's concepts using everyday analogies:

1. The Setting: A Ship with a Shifting Map

In the real world, things don't stay the same.

• Regime Switching: Imagine the ocean suddenly changes from "Calm Summer" to "Hurricane Season" without warning. The ship's behavior changes instantly based on this "regime."
• Jump-Diffusion: The ship moves smoothly (diffusion) like a boat on water, but occasionally gets hit by a giant, random wave (a "jump") that throws it off course instantly.
• Random Coefficients: The ship's engine isn't perfectly predictable. Sometimes it sputters, sometimes it roars, and no one knows exactly why until it happens.

2. The Conflict: The Zero-Sum Game

This is a Zero-Sum Game. Think of it like a poker game where the total money on the table never changes. If Captain A loses $10, Captain B wins $10.

• The Goal: The paper tries to find a Saddle Point. This is a special moment where Captain A cannot do any better to protect the ship, and Captain B cannot do any better to push it, assuming the other captain is playing their best strategy. It's a state of perfect, tense balance.

3. The Twist: The "No-Go" Zones (Constraints)

In many math problems, captains can steer the ship in any direction. But in this paper, there are Cone Constraints.

• The Analogy: Imagine the ship is in a narrow canyon. Captain A can only steer "forward" or "slightly left," but never "backward." Captain B can only steer "forward" or "slightly right," but never "backward."
• The Problem: Because they are stuck in these specific directions (cones), the usual math tricks (like the "Four-Step Scheme") that mathematicians use to solve these problems break down. It's like trying to solve a maze where you can only turn left, but the solution requires a right turn.

4. The Solution: The "Magic Mirror" (Riccati Equations)

To solve this, the authors had to invent a new kind of mathematical tool.

• The Old Way: Usually, you solve these problems by looking at the future and working backward. But with the "canyon" constraints, this didn't work.
• The New Way: The authors used a technique called "Completing the Square" combined with a complex set of equations called Indefinite Extended Stochastic Riccati Equations (IESREJs).
  • The Metaphor: Imagine you are trying to balance a stack of plates on a moving truck. The "Riccati Equation" is like a super-complex instruction manual that tells you exactly how to adjust your hands (the controls) at every single millisecond to keep the plates from falling, even when the truck hits a pothole (the jump) or the road tilts (regime switching).
  • "Indefinite": This is the tricky part. Usually, these instructions say "Keep the plates stable." But because one captain wants to crash and the other wants to save, the instructions sometimes say "Push harder" and sometimes "Pull back" depending on who is winning. The math has to handle this contradiction.

5. The Breakthrough

The authors proved two main things:

1. Existence: They showed that a perfect balance does exist. Even with the crazy weather, the shifting maps, and the "canyon" steering limits, there is a specific strategy where neither captain can gain an advantage by changing their mind.
2. The Formula: They didn't just say "it exists"; they wrote down the actual formula (the feedback representation) that tells the captains exactly what to do based on where the ship is right now.

Why Does This Matter?

You might ask, "Who cares about two captains fighting on a math ship?"

• Real World: This applies to Finance. Imagine an investor (Captain A) trying to minimize risk and a speculator (Captain B) trying to maximize profit in a market that crashes suddenly (jumps) and changes economic regimes (bull vs. bear markets).
• The Constraint: Often, investors are legally forbidden from "shorting" (betting against the market) or must keep certain assets. This paper gives them the math to navigate those strict rules in a chaotic world.

In a Nutshell

This paper is a mathematical guide for navigating a chaotic, unpredictable world where two opponents are fighting for opposite goals, and both are forced to play by strict, limited rules. The authors figured out how to calculate the perfect move for both sides, ensuring that even in the worst storm, the game reaches a stable, predictable conclusion.

Technical Summary

Here is a detailed technical summary of the paper "Constrained zero-sum LQ differential games for jump-diffusion systems with regime switching and random coefficients" by Tang, Li, and Xiong.

1. Problem Formulation

The paper investigates a two-player zero-sum stochastic linear-quadratic (SLQ) differential game (denoted as Problem C-ZLQJ) under the following complex conditions:

• System Dynamics: The state process $X(t)$ is governed by a stochastic differential equation (SDE) driven by:
  • A standard Brownian motion $W(t)$.
  • A Poisson random measure $N(dt, dz)$ representing jumps.
  • A continuous-time Markov chain $\alpha(t)$ representing regime switching.
  • Random coefficients: All system parameters ($A, B, C, D, E, F, Q, R, S, G$) are adapted to the filtration generated by both the Brownian motion and the Poisson measure, making them stochastic processes rather than deterministic functions.
• Constraints: The control strategies $u_1$ (Player 1, minimizer) and $u_2$ (Player 2, maximizer) are constrained to lie within closed convex cones $\Pi_1$ and $\Pi_2$ in $\mathbb{R}^{m_1}$ and $\mathbb{R}^{m_2}$, respectively.
• Objective: The players aim to optimize a quadratic cost functional $J(\xi, i; u_1, u_2)$, where Player 1 minimizes and Player 2 maximizes the expected value of the integral of a quadratic form involving the state and controls, plus a terminal cost.

2. Methodology

The authors employ a combination of stochastic control theory, convex analysis, and backward stochastic differential equation (BSDE) techniques:

• Open-Loop Solvability:

  • The Stochastic Maximum Principle (SMP) is established to derive necessary and sufficient conditions for an open-loop saddle point.
  • The Uniform Convexity-Concavity (UCC) condition is imposed to ensure the existence and uniqueness of an open-loop saddle point. This condition guarantees that the cost functional is strongly convex with respect to the minimizer's control and strongly concave with respect to the maximizer's control.
  • The open-loop saddle point is characterized via a system of Forward-Backward Stochastic Differential Equations (FBSDEs).

• Closed-Loop Representation (Feedback Control):

  • Due to the cone constraints, the classical "four-step scheme" (which links FBSDEs to Riccati equations via a linear feedback law) fails because the optimal control is no longer a linear function of the state.
  • To overcome this, the authors utilize Meyer's Itô formula (specifically for the positive and negative parts of the state process, $X^+$ and $X^-$) and the method of completing the square.
  • This approach leads to the derivation of a closed-loop representation for the saddle point, expressed as a feedback law involving the positive and negative parts of the state.

• Riccati Equations:

  • The feedback representation relies on the solution to a new class of equations: Multidimensional Indefinite Extended Stochastic Riccati Equations with Jumps (IESREJs).
  • These equations are indefinite (due to the zero-sum nature where weighting matrices have opposite signs) and extended (due to the cone constraints and jump-diffusion dynamics).
  • The authors prove the solvability of these IESREJs using an approximation technique (truncating the control space) and the comparison theorem for multidimensional BSDEs with jumps.

3. Key Contributions

1. Generalization of System Models: The paper extends existing SLQ game theory to systems with random coefficients, regime switching, and jump-diffusion processes simultaneously. This is more general than previous works that often assumed deterministic coefficients or only Brownian motion.
2. Handling Cone Constraints in Zero-Sum Games: Unlike optimal control problems where constraints usually lead to positive semi-definite weighting matrices, zero-sum games naturally lead to indefinite weighting matrices. The paper successfully derives the feedback form for constrained controls in this indefinite setting, a significant theoretical hurdle.
3. New Riccati Equations (IESREJs): The authors introduce and analyze a novel system of coupled, indefinite, extended stochastic Riccati equations with jumps. They establish the existence of solutions where the first component remains positive, which is crucial for the stability of the game.
4. Explicit Feedback Strategy: The paper provides an explicit feedback-form representation for the optimal strategies:
   $$u^*(t) = \Theta^+(\cdot) X^+(t) + \Theta^-(\cdot) X^-(t)$$
   where $\Theta^+$ and $\Theta^-$ are derived from the solutions to the IESREJs and involve projection operators onto the constraint cones.

4. Main Results

• Theorem 3.2: Under the UCC condition, Problem (C-ZLQJ) admits a unique open-loop saddle point for any initial state.
• Theorem 4.1: Assuming the solvability of the IESREJs, the open-loop saddle point admits a closed-loop (feedback) representation. The optimal cost is given by $E[P_1(0, i)(\xi^+)^2] + E[P_2(0, i)(\xi^-)^2]$.
• Theorem 5.1: Under specific coefficient conditions (simplifying the coupling between players in the Riccati equations), the authors prove the existence of solutions to the IESREJs. This is achieved by constructing a sequence of approximating BSDEs with Lipschitz generators and using monotone convergence and comparison theorems.

5. Significance

• Theoretical Advancement: This work bridges the gap between constrained stochastic control and zero-sum game theory in the presence of complex stochastic environments (jumps and regime switching). It resolves the difficulty of deriving feedback controls when the standard linear-quadratic structure is broken by cone constraints and indefinite cost matrices.
• Mathematical Finance Applications: The framework is highly relevant for financial modeling, particularly in mean-variance portfolio selection with constraints (e.g., no-short-selling) in markets subject to sudden shocks (jumps) and changing economic regimes. The indefinite nature of the Riccati equations reflects the conflicting interests of the two players (or the risk-minimizing vs. risk-seeking nature in a game setting).
• Methodological Innovation: The use of Meyer's Itô formula on $X^+$ and $X^-$ combined with the completion of squares provides a robust template for solving other constrained stochastic differential games where standard dynamic programming or four-step schemes are inapplicable.

In summary, the paper provides a rigorous mathematical foundation for solving constrained zero-sum games in highly realistic, stochastic environments, offering explicit feedback strategies and proving the solvability of the associated complex Riccati equations.