Parameter-related strong convergence rates of Euler-type methods for time-changed stochastic differential equations

This paper proposes an Euler-type framework with equidistant step sizes for time-changed stochastic differential equations, establishing that both the standard and truncated Euler–Maruyama methods achieve strong convergence rates close to $\alpha/2$ under global Lipschitz and relaxed Khasminskii-type conditions, respectively, which contrasts with the classical $1/2$ order found in methods using random step sizes.

The Gist

Here is an explanation of the paper using simple language, analogies, and metaphors.

The Big Picture: Modeling "Sticky" Time

Imagine you are watching a drop of dye spread through a glass of water. In a normal world (standard physics), the dye spreads out smoothly and predictably. Mathematicians have a perfect tool to describe this: Stochastic Differential Equations (SDEs). It's like a GPS that predicts where the dye will be next, accounting for the random jiggles of water molecules.

But in the real world, things aren't always smooth. Sometimes, the dye gets "stuck" in a sponge, or a particle gets trapped in a muddy patch. It waits there for a long time, then suddenly jumps again. This is called anomalous diffusion (or sub-diffusion). The standard GPS (standard SDEs) fails here because it assumes time flows at a constant speed.

To fix this, scientists use Time-Changed SDEs. They imagine that the "clock" for the particle isn't a normal clock, but a broken, erratic clock. Sometimes the clock ticks fast; sometimes it stops completely for a while. This "broken clock" is mathematically called an Inverse Subordinator.

The Problem: How to Simulate the Broken Clock?

The paper tackles a specific problem: How do we write a computer program to simulate these "sticky" particles accurately?

Previous methods tried to fix the broken clock by forcing it to tick in a predictable, random way. They essentially said, "Let's pretend the clock stops at random moments, but we know exactly when." This worked well, but it hid the true nature of the problem. It gave a standard accuracy (let's call it a "Grade B" result), regardless of how "sticky" the environment actually was.

The authors of this paper asked a bold question: "What if we don't try to fix the clock? What if we let the computer simulate the clock exactly as it is—erratic and unpredictable—and see what happens to our accuracy?"

The Solution: The "Equidistant Step" Method

The authors developed a new way to simulate these systems using Equidistant Steps.

• The Old Way (Random Steps): Imagine trying to measure the distance a hiker walks, but you only check their position whenever they stop to rest. Since the hiker stops at random times, your measurements are messy.
• The New Way (Equidistant Steps): Imagine checking the hiker's position every single minute, no matter what they are doing. If they are stuck in mud for 10 minutes, you record them in the same spot 10 times. If they run fast, you record them moving fast.

By checking the position at fixed, regular intervals (like a metronome), the authors could capture the true "roughness" of the time-change process.

The Big Discovery: The "Alpha" Factor

Here is the magic of the paper. In these "sticky" systems, there is a number called $\alpha$ (alpha) that describes how sticky the environment is.

• If $\alpha$ is close to 1, the system is almost normal (not very sticky).
• If $\alpha$ is close to 0, the system is very sticky (lots of trapping).

The authors proved that the accuracy of their new method depends directly on this $\alpha$.

• Old Methods: Always gave an accuracy of roughly 0.5 (like a standard ruler).
• New Method: Gives an accuracy of roughly $\alpha / 2$.

The Analogy:
Imagine you are trying to guess the temperature of a room.

• If the room is stable (normal), your guess is very good.
• If the room has a broken thermostat that randomly freezes and thaws (the time-change), your guess gets worse.
• The authors found that the worse the thermostat is (the lower the $\alpha$), the slower your computer simulation converges to the right answer. Specifically, the error shrinks at a rate of $\alpha/2$.

This is a huge deal because it proves that the "roughness" of time itself dictates how hard it is to do the math.

Handling the "Explosive" Coefficients

There was a second hurdle. In many real-world scenarios (like population growth or chemical reactions), the forces acting on the particle can get huge very quickly (super-linear growth). Standard computer methods often "explode" or crash when numbers get too big.

The authors introduced a Truncated Method.

• The Metaphor: Imagine a video game character who can run infinitely fast. If they run too fast, the game crashes. To fix this, the developers put a "speed cap" on the character. If the character tries to go faster than the cap, the game forces them to slow down just enough to stay safe, then lets them go again.
• The Math: They created a mathematical "speed cap" (truncation) that prevents the numbers from blowing up, while still keeping the simulation accurate enough to get the right answer.

The Results

The paper concludes with computer simulations that act as the "proof of concept."

1. They tested the method on simple, smooth systems. The results matched the theory perfectly.
2. They tested it on "sticky" systems with different levels of $\alpha$. As predicted, the accuracy dropped as the system got stickier (lower $\alpha$), exactly following the $\alpha/2$ rule.
3. They tested it on "explosive" systems. The "speed cap" (truncation) worked, keeping the simulation stable and accurate.

Summary in One Sentence

This paper introduces a new, smarter way to simulate particles moving through "sticky" time, proving that the accuracy of the simulation is directly tied to how "sticky" the time is, and providing a safety net to prevent the math from crashing when the forces get too strong.

Technical Summary

Here is a detailed technical summary of the paper "Parameter-related strong convergence rates of Euler-type methods for time-changed stochastic differential equations" by Ruchun Zuo.

1. Problem Statement

The paper addresses the numerical approximation of Time-Changed Stochastic Differential Equations (TC-SDEs) of the form:
$$dX(t) = f(X(t))dE(t) + g(X(t))dB(E(t))$$
where:

• $E(t)$ is the inverse subordinator (specifically an inverse $\alpha$-stable subordinator with stability index $\alpha \in (0, 1)$).
• $B(E(t))$ is a time-changed Brownian motion.
• The process models subdiffusion and anomalous diffusion phenomena characterized by memory effects and trapping events.

The Core Challenge:
Existing numerical methods for TC-SDEs typically rely on random step sizes derived from the discretization of the inverse subordinator. These methods often preserve the classical strong convergence order of $1/2$ (similar to standard SDEs) because the randomness of the time change is effectively "absorbed" into deterministic increments within the scheme.
The paper investigates whether it is possible to develop equidistant-step Euler-type methods where the convergence rate explicitly reflects the unique properties of the time-change process (specifically the stability index $\alpha$).

2. Methodology

The authors propose and analyze two numerical schemes using equidistant time steps ($\Delta t$), preserving the intrinsic stochasticity of the subordinator increments within the numerical scheme.

A. Standard Euler–Maruyama (EM) Method

• Scheme: Defined recursively as $X_{n+1} = X_n + f(X_n)\Delta E_n + g(X_n)\Delta B_n$, where $\Delta E_n = E(t_{n+1}) - E(t_n)$.
• Assumptions: The drift $f$ and diffusion $g$ coefficients satisfy global Lipschitz conditions and linear growth conditions.
• Analysis Tool: The authors utilize a uniform bound on the maximum increment of the inverse subordinator, $\Xi_h = \sup_{0 \le u \le T-h} (E(u+h) - E(u))$. They prove that the $n$-th moment of this increment scales as $O(h^{n(\alpha-\epsilon)})$.

B. Truncated Euler–Maruyama Method

• Motivation: The global Lipschitz assumption is too restrictive for many practical applications involving super-linear growth coefficients (where standard EM methods diverge).
• Scheme: A truncated version where coefficients are modified via a truncation mapping $\pi_\Delta(x)$ to ensure boundedness. The coefficients are $f_\Delta(x) = f(\pi_\Delta(x))$ and $g_\Delta(x) = g(\pi_\Delta(x))$.
• Assumptions: The coefficients satisfy Khasminskii-type conditions (one-sided Lipschitz and polynomial growth) rather than global Lipschitz conditions.
• Analysis: The proof involves establishing moment bounds for the truncated solution and analyzing the error between the exact solution, the piecewise constant approximation, and the continuous-time interpolation.

3. Key Contributions

1. Equidistant-Step Framework: The paper introduces a rigorous framework for applying equidistant-step Euler methods to TC-SDEs, contrasting with previous random-step approaches.
2. Parameter-Dependent Convergence Rates: The authors prove that the strong convergence rate is not the classical $1/2$, but is instead determined by the stability index$\alpha$ of the subordinator.
   • For both the standard and truncated schemes, the strong convergence order is shown to be arbitrarily close to $\alpha/2$.
   • Specifically, for any small $\epsilon \in (0, \alpha)$, the error bound is $O(\Delta t^{(\alpha-\epsilon)/2})$.
3. Relaxed Conditions: The extension of these results to the truncated EM method under Khasminskii-type conditions allows for the numerical treatment of TC-SDEs with super-linear growth coefficients, a significant advancement over previous literature.
4. Theoretical Justification for Anomalous Diffusion: The work provides a rigorous mathematical foundation for using equidistant-step methods in simulating fractional Fokker–Planck equations (FFPEs) associated with anomalous diffusion.

4. Key Results

• Theorem 3.1 & 3.2 (Standard EM): Under global Lipschitz conditions, the strong error satisfies:
  $$E\left[\sup_{t \in [0, T]} |X(t) - \tilde{X}_t|^p\right] \le C \Delta t^{\frac{p(\alpha-\epsilon)}{2}}$$
  This implies a convergence rate of approximately $\alpha/2$.
• Theorem 4.1 (Truncated EM): Under Khasminskii-type conditions and super-linear growth, the truncated scheme achieves the same convergence rate:
  $$E\left[\sup_{t \in [0, T]} |X(t) - \bar{X}(t)|^{\bar{p}}\right] \le C \Delta t^{\frac{\bar{p}(\alpha-\epsilon)}{2}}$$
• Limiting Case: The paper notes that if the subordinator has a positive linear drift (making the inverse subordinator Lipschitz continuous), the convergence rate reverts to the classical $1/2$, consistent with existing literature.
• Numerical Validation:
  • Simulations on coupled linear TC-SDEs (Example 5.1) confirmed the theoretical rates for $\alpha = 0.6$ and $\alpha = 0.8$.
  • Simulations on systems with super-linear coefficients (Example 5.2) using the truncated method successfully demonstrated the $\alpha/2$ convergence rate, validating the theoretical extension.

5. Significance

• Bridging Theory and Practice: The results clarify the relationship between the statistical properties of the time-change process (the stability index $\alpha$) and the accuracy of numerical methods.
• Efficiency in Subdiffusion Modeling: By establishing that the convergence rate is $\alpha/2$ (which is less than $1/2$for$\alpha < 1$), the paper warns practitioners that finer discretization is required for subdiffusive systems compared to standard Brownian motion systems to achieve the same accuracy.
• Robustness: The development of the truncated scheme ensures that these methods can be applied to a broader class of realistic physical and financial models where coefficients grow super-linearly, preventing the numerical divergence often seen in standard Euler methods.
• Foundation for Future Work: The rigorous error analysis provides a solid basis for developing higher-order methods or adaptive step-size strategies specifically tailored for time-changed processes.