Approximation of invariant probability measures for super-linear stochastic functional differential equations with infinite delay

This paper proposes an explicit truncated Euler-Maruyama scheme with time and space truncation to approximate the invariant probability measures of super-linear stochastic functional differential equations with infinite delay, establishing strong convergence and proving that the numerical invariant measure converges to the exact one in Wasserstein distance with an explicit rate.

The Gist

Imagine you are trying to predict the long-term weather patterns of a very chaotic, complex city. You know that the weather today isn't just about the temperature right now; it's heavily influenced by what happened yesterday, last week, and even last month. This is a system with "infinite delay."

Furthermore, imagine that the rules governing this weather aren't simple. Sometimes, a small change in wind speed causes a massive, explosive storm (this is the "super-linear" part). If you try to use a standard, simple calculator (a standard computer algorithm) to predict this, the numbers might explode to infinity, crashing your computer.

This paper is about building a smart, robust calculator that can handle these chaotic, memory-heavy systems and tell us what the "average" weather will look like after running for a very long time. In math terms, this "average weather" is called an Invariant Probability Measure (IPM).

Here is a breakdown of the paper's journey using simple analogies:

1. The Problem: The "Memory" and the "Explosion"

• The Infinite Delay: Think of a person trying to walk a tightrope. They don't just look at their feet; they look at where they were 10 seconds ago, 20 seconds ago, and so on. Their balance depends on their entire history. In math, this is a system where the future depends on the entire past.
• The Super-Linear Growth: Imagine a snowball rolling down a hill. Usually, it grows steadily. But in these specific equations, the snowball grows so fast that if you try to simulate it with a basic step-by-step method, the snowball becomes the size of the universe in one second. This is why standard methods fail.
• The Old Solutions: Previous attempts to solve this used "Implicit Schemes." Think of this like trying to solve a puzzle where you have to guess the answer, check if it's right, and if it's wrong, guess again, over and over, for every single step. It's accurate but incredibly slow and computationally expensive (like trying to solve a Rubik's cube by feeling every single square).

2. The Solution: The "Truncated" Euler-Maruyama (TEM) Scheme

The authors propose a new method called the TEM scheme. Here is how it works, using a metaphor:

Imagine you are driving a car on a road with a speed limit, but the car has a tendency to accelerate uncontrollably if you press the gas too hard.

• The "Truncation" (The Governor): The TEM scheme acts like a smart governor on the engine. It says, "Okay, you can accelerate, but if you get too fast (too large a number), I'm going to gently cap your speed." It doesn't stop the car; it just keeps the numbers from exploding.
• The "Explicit" Nature: Unlike the old "guess-and-check" methods, this new method is explicit. It's like driving with a clear map. You know exactly where you will be in the next second based on where you are now. No guessing, no waiting. It's fast.
• The "Memory" Trick: Since the car remembers everything (infinite delay), you can't store the whole history in your glovebox. The authors designed a clever way to only store the most recent, relevant "snapshots" of the history. It's like a photographer who only keeps the last few photos needed to reconstruct the scene, rather than keeping every photo ever taken. This saves massive amounts of computer memory.

3. The Results: Proving the Calculator Works

The paper doesn't just build the calculator; it proves it works mathematically.

• Strong Convergence (Short Term): They proved that if you run this new calculator for a short time (say, 10 minutes), the path it draws is almost identical to the real, chaotic path. The error is tiny, and they even calculated exactly how tiny it is (roughly proportional to the square root of the step size).
• Ergodicity (Long Term): This is the big one. They proved that if you let this system run for a very long time, it settles into a stable pattern. Even though the weather is chaotic day-to-day, the average climate is predictable. The TEM scheme finds this average climate correctly.
• The "Water" Distance: To measure how close their "average climate" is to the real one, they use a mathematical ruler called the Wasserstein distance. Think of it as measuring how much "work" it takes to move the dirt from one pile (the real average) to another pile (the calculated average). They proved that as you make your steps smaller, the dirt piles become indistinguishable.

4. Why This Matters

• Efficiency: Because the method is "explicit" (fast) and "truncated" (stable), it can run on standard computers without needing supercomputers to solve complex equations at every step.
• Real World Applications: This isn't just abstract math. These equations model:
  • Finance: Markets that react to long-term trends and can crash violently.
  • Biology: Population dynamics where species interact based on past densities.
  • Physics: Fluid dynamics and heat transfer with memory.

Summary Analogy

Imagine trying to predict the final shape of a piece of clay being molded by a chaotic, invisible hand that remembers every touch from the past.

• Old Method: You try to mold it by feeling the clay, guessing the shape, and reshaping it repeatedly. It takes forever and you might break the clay.
• New Method (This Paper): You build a machine that gently guides the clay, capping the force so it doesn't fly apart, and only remembering the last few touches needed to keep the shape. You prove that this machine will eventually produce the exact same final shape as the invisible hand, and you can do it much faster.

The paper successfully bridges the gap between complex, chaotic reality and efficient, reliable computer simulation.

Technical Summary

Here is a detailed technical summary of the paper "Approximation of invariant probability measures for super-linear stochastic functional differential equations with infinite delay" by Guozhen Li, Shan Huang, Xiaoyue Li, and Xuerong Mao.

1. Problem Statement

The paper addresses the numerical approximation of Invariant Probability Measures (IPMs) for a class of Stochastic Functional Differential Equations (SFDEs) with infinite delay and super-linear drift coefficients.

• The Equation: The study focuses on equations of the form:
  $$dx(t) = f(x_t)dt + g(x_t)dB(t), \quad t \geq 0, \quad x_0 = \xi \in C_r$$
  where $x_t$ represents the segment process (history) in a phase space $C_r$ with fading memory, $f$ and $g$ are drift and diffusion coefficients, and $B(t)$ is a Brownian motion.
• The Challenge:
  1. Super-linear Growth: Standard explicit schemes (like the classical Euler-Maruyama method) fail for super-linear coefficients because their moments diverge to infinity, even if the exact solution's moments are bounded.
  2. Infinite Delay: Unlike finite-delay cases, infinite delay requires storing an infinite history. Standard numerical approaches often require implicit schemes (solving nonlinear equations at each step), which are computationally expensive, or they fail to handle the infinite memory storage efficiently.
  3. Ergodicity Preservation: Constructing an explicit scheme that not only converges in finite time but also preserves the long-term ergodic properties (convergence to a unique stationary distribution) of the exact system is a significant open problem.

2. Methodology

The authors propose a novel Explicit Truncated Euler-Maruyama (TEM) scheme that employs both time and space truncation.

• Truncation Mapping: To handle super-linear growth, a truncation mapping $\Pi_\Delta$ is introduced. This maps the state space to a bounded region defined by an inverse function $\Lambda^{-1}$ of a growth bound function $\Lambda$. This ensures the numerical coefficients remain globally Lipschitz within the simulation.
• Space-Time Truncation:
  • Space Truncation: Limits the magnitude of the drift and diffusion terms to prevent explosion.
  • Time/History Truncation: To address the infinite delay, the scheme approximates the infinite history by storing only a finite number of past discrete time nodes ($k$ steps). The history beyond this window is approximated by the value at the truncation point.
• Auxiliary Processes: The proof strategy relies on a three-step convergence analysis:
  1. Exact Solution ($x_t$) $\to$ Truncated Exact Solution ($x^k_t$): Proving that the solution to the SFDE with infinite delay converges to a solution with finite delay (truncating the memory) as the memory length $k \to \infty$.
  2. Truncated Exact Solution ($x^k_t$) $\to$ Numerical Solution ($X^{k,\Delta}_t$): Proving the strong convergence of the TEM scheme to the finite-delay truncated solution as the step size $\Delta \to 0$.
  3. Triangle Inequality: Combining the above to establish convergence of the numerical solution to the exact infinite-delay solution.

3. Key Contributions

The paper makes four primary theoretical contributions:

1. Construction of an Explicit TEM Scheme:
   The authors construct an explicit scheme that overcomes the super-linear growth issue via truncation and handles infinite delay via finite historical storage. Unlike implicit schemes, this method does not require solving nonlinear algebraic equations at each step, significantly reducing computational cost.

2. Strong Convergence and Rate:

   • They prove the strong convergence of the numerical segment process to the exact segment process on any finite time horizon $[0, T]$.
   • Under polynomial growth conditions, they derive a strong convergence rate arbitrarily close to **$1/2$** (specifically$\Delta^{q/2 - \epsilon}$).

3. Existence and Uniqueness of Numerical IPM:
   Under suitable dissipativity conditions, the paper proves that the discrete-time transition semigroup generated by the TEM scheme admits a unique numerical IPM. Furthermore, this numerical IPM is shown to be exponentially ergodic in the Wasserstein distance.

4. Convergence of Numerical IPM to Exact IPM:
   The most significant result is the proof that the numerical IPM converges to the exact IPM of the underlying SFDE in the Wasserstein distance. The paper explicitly derives the convergence rate for this approximation, linking the error to the step size $\Delta$ and the memory truncation parameter $k$.

4. Key Results

• Boundedness: The numerical segment process is proven to be bounded in $L^p$ moments uniformly over time and step sizes.
• Ergodicity: The TEM scheme preserves the ergodicity of the original system. The numerical IPM exists, is unique, and the system converges to it exponentially fast.
• Convergence Rates:
  • Finite Time: The error between the numerical and exact segments is $O(\Delta^{1/2 - \epsilon})$.
  • Long Time (IPM): The error between the numerical IPM ($\pi_\Delta$) and the exact IPM ($\pi$) in the Wasserstein distance $W_2$ is bounded by $C \Delta^{\bar{\rho}_\epsilon}$, where the exponent depends on the ergodicity rates and the step size.
• Storage Efficiency: A crucial practical result is that the scheme requires storing only $O((kl+1)n)$ historical nodes, where $k$ is the memory truncation length and $l$ is the inverse of the step size. This storage is bounded and independent of the simulation horizon, making long-time simulations feasible.

5. Significance

• Bridging Theory and Practice: This work resolves the open problem of how to construct an explicit numerical scheme for super-linear SFDEs with infinite delay that preserves ergodicity. Previous methods were limited to implicit schemes (high cost) or finite-delay cases.
• Computational Efficiency: By avoiding implicit iterations and managing infinite memory through controlled truncation, the proposed TEM scheme offers a computationally efficient alternative for simulating complex systems with long-term memory (e.g., in physics, biology, and economics).
• Theoretical Rigor: The paper provides rigorous proofs for the convergence of invariant measures, a notoriously difficult task in stochastic numerics, establishing explicit error rates that guide practical implementation.
• Validation: Numerical experiments on a super-linear SFDE and a stochastic Lotka-Volterra model with infinite delay confirm the theoretical convergence rates (close to 0.5) and the existence of the numerical IPM, demonstrating the scheme's robustness even in cases slightly outside the strict theoretical dissipative framework.

In summary, this paper provides a comprehensive framework for the efficient and accurate numerical simulation of the long-term statistical behavior of complex stochastic systems with infinite memory and super-linear dynamics.