🔢 mathematics

Boundary-preserving Lamperti--Itô--Taylor approximations for some stochastic differential equations

This paper proposes high-order, boundary-preserving numerical schemes for scalar stochastic differential equations with bounded invariant domains by utilizing the Lamperti transform to convert the problem into one with additive noise, thereby enabling the application of standard high-order methods while guaranteeing convergence and domain preservation.

Original authors: Johan Ulander

Published 2026-07-23

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Johan Ulander

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 you are trying to predict the path of a leaf floating down a river. The river has a current pushing the leaf, but it's also being tossed around by unpredictable gusts of wind. In the world of science, this is modeled by something called a Stochastic Differential Equation (SDE). It's a math tool used to describe everything from how stock prices wiggle to how diseases spread or how chemicals mix. The tricky part is that these equations are usually too messy to solve with a simple formula, so scientists have to use computers to take tiny steps and guess where the leaf will be next.

However, there's a catch. In many real-world situations, the leaf cannot leave the river. It can't float into the dry land or sink into the mud; it is trapped within specific banks. These banks are called "invariant domains." If a computer simulation is too clumsy, it might accidentally calculate a path where the leaf jumps over the bank and ends up on dry land, which makes no sense physically. This is a common headache for scientists: standard computer methods are often great at guessing the path, but terrible at respecting the walls that keep the leaf safe. They treat the wind as if it could blow the leaf anywhere, even into places that are impossible.

This paper, titled "Boundary-Preserving Lamperti–Itô–Taylor Approximations for Some Stochastic Differential Equations" by Johan Ulander, tackles exactly this problem. The author proposes a clever new way to build computer simulations that are not only more accurate but also guaranteed to keep the "leaf" inside the "river" at all times.

The core idea is a bit like changing the map. Imagine the river is winding and narrow, making it hard to draw a straight line for the leaf's path. The author suggests using a mathematical trick called the "Lamperti transform" to stretch and reshape the river into a perfectly straight, wide canal with no curves. In this new, straightened world, the wind (the random noise) behaves much more simply—it just pushes the leaf in a straight line without getting tangled in the river's curves. Because the new map is so simple, scientists can use very advanced, high-precision tools (called "Itô–Taylor schemes") to calculate the leaf's position with incredible accuracy.

But here is the magic: once the calculation is done in the straight, easy world, the author simply reverses the map transformation. Because the math was designed carefully, when you fold the straight canal back into the original winding river, the leaf is guaranteed to still be inside the banks. The paper proves that if the river's rules (the mathematical coefficients) are smooth enough, this method works perfectly. It doesn't just guess that the leaf stays inside; it mathematically guarantees it.

The author tested this new method on three different types of "rivers" that model real-world phenomena: the Allen–Cahn equation (used in physics for things like phase changes), the Nagumo equation (used in biology for nerve signals), and the SIS equation (used in epidemiology for disease spread). In every test, the new method kept the leaf safely within the boundaries, even when the wind was very strong. In contrast, the old, standard computer methods frequently failed, letting the leaf jump over the banks and land in impossible places.

Furthermore, the paper shows that this new method is highly accurate. It achieves high-order strong convergence, meaning the error shrinks very quickly as you add more steps, reaching accuracy levels of 1.5 or even 2.0. The author ran hundreds of computer simulations to prove that the method works as promised, showing that the error drops exactly as the math predicted.

In short, this paper offers a new recipe for simulating random processes that are trapped in a specific area. It combines a clever map-shifting trick with high-precision math to ensure that the computer never makes a "silly" mistake where the object escapes its world. It's a significant step forward for anyone who needs to model things that must stay within their limits, from financial markets to biological cells, ensuring that the simulations remain both realistic and reliable.

Technical Summary: Boundary-Preserving Lamperti–Itô–Taylor Approximations for Some Stochastic Differential Equations

Problem Statement
Stochastic differential equations (SDEs) are widely used to model phenomena in finance, biology, and physics. Many such models possess an invariant domain—a bounded open interval $D = (a, b)$ such that if the initial condition lies within $D$ , the solution remains in $D$ for all time with probability one. Standard numerical schemes, such as the Euler–Maruyama (EM) method, often fail to preserve this property. This failure typically arises because the discretization of the noise introduces random variables with unbounded support, causing numerical approximations to exit the physical domain. While various methods (implicit, truncation-based, geometric) have been proposed to preserve domains, achieving high-order strong convergence while maintaining exact boundary preservation remains a challenge. Existing high-order schemes generally do not preserve invariant domains, and existing boundary-preserving schemes often lack high-order convergence (typically limited to order 0.5 or 1).

Methodology
The paper proposes a class of numerical schemes, termed Lamperti–Itô–Taylor (LIT) schemes, designed to achieve arbitrarily high strong convergence orders while exactly preserving the invariant domain. The methodology proceeds in three steps:

Lamperti Transform: The original scalar SDE, $dX_t = f(X_t)dt + g(X_t)dB_t$ , is transformed using the Lamperti transform $\Phi(x) = \int_{w_0}^x \frac{1}{g(w)} dw$ . This maps the original SDE with multiplicative noise to a new SDE with additive noise:
$dY_t = H(Y_t)dt + dB_t$
where $Y_t = \Phi(X_t)$ . The invariant domain $D$ of the original SDE is mapped to the entire real line $\mathbb{R}$ for the transformed SDE.
Regularity Assumptions: The paper imposes specific regularity conditions on the original coefficients $f$ and $g$ (Assumptions 1–3). These conditions ensure that the drift coefficient $H$ of the transformed SDE is sufficiently smooth (specifically, $H \in C^k_b(\mathbb{R})$ with bounded derivatives up to order $k$ ). This regularity is crucial for the application of high-order Itô–Taylor schemes.
High-Order Approximation and Inverse Transform:
- The solution $Y_t$ of the transformed SDE is approximated using standard Itô–Taylor schemes of strong order $\gamma$ (denoted $IT_\gamma$ ). Because the noise is additive, the construction of these schemes simplifies, and the domain of the approximation remains $\mathbb{R}$ , avoiding boundary violations.
- The final approximation for the original solution $X_t$ is obtained by applying the inverse Lamperti transform: $X_m \approx \Phi^{-1}(Y_m)$ . Since $\Phi^{-1}$ maps $\mathbb{R}$ back to the bounded interval $D$ , the resulting scheme is guaranteed to stay within the invariant domain.

The paper provides explicit recursive formulas for the LIT schemes corresponding to orders $\gamma = 0.5$ (Lamperti–EM), $\gamma = 1.5$ , and $\gamma = 2.0$ . It also details the computation of the necessary iterated stochastic integrals (e.g., $I_{(1,0)}$ and $I_{(1,1,0)}$ ) using Chen's lemma and specific sampling techniques to ensure consistency across different time grids.

Key Contributions

Proposal of Boundary-Preserving High-Order Schemes: The paper introduces a family of LIT schemes that preserve the invariant domain exactly while achieving strong convergence orders greater than 1.5 (specifically orders 1.0, 1.5, and 2.0 are demonstrated).
Theoretical Guarantees: The author proves that under the specified regularity assumptions on $f$ and $g$ , the transformed drift $H$ possesses the necessary bounded derivatives. Consequently, the paper establishes that the LIT schemes are $p$ -strongly convergent with order $\gamma$ and satisfy the boundary-preserving property with probability one (Theorem 3 and Corollary 8).
Novelty in Convergence Order: To the best of the author's knowledge, these are the first boundary-preserving numerical schemes for SDEs with bounded invariant domains that achieve strong convergence orders higher than 1.5.

Results
The theoretical results are validated through numerical experiments on three specific classes of SDEs:

Allen–Cahn type SDE: Defined on $D=(-1, 1)$ .
Nagumo-type SDE: Defined on $D=(0, 1)$ .
SIS (Susceptible-Infected-Susceptible) SDE: Defined on $D=(0, 1)$ .

In these experiments, the LIT schemes (LEM, LIT1.5, LIT2.0) were compared against classical schemes (EM, semi-implicit EM, and tamed Euler).

Boundary Preservation: In all test cases, the proposed LIT schemes remained strictly within the invariant domain for 100% of the 100 sample paths tested, even with high noise scaling parameters ( $\lambda$ ). In contrast, the classical schemes frequently exited the domain, with preservation rates dropping significantly as noise intensity increased.
Convergence Order: Log-log plots of the $L^2(\Omega)$ -error versus the time step $\Delta t$ confirmed the theoretical convergence rates. The LEM scheme achieved order 1.0, LIT1.5 achieved order 1.5, and LIT2.0 achieved order 2.0, aligning with the reference slopes.

Significance and Claims
The paper claims that the primary contribution is the successful combination of the Lamperti transform with high-order Itô–Taylor schemes to solve the specific problem of boundary preservation in SDEs with bounded domains. The work demonstrates that by transforming the problem to one with additive noise and ensuring sufficient regularity of the original coefficients, one can utilize standard high-order methods without sacrificing the physical constraint of the domain. The author notes that while the focus is on strong Itô–Taylor schemes, the underlying strategy (Lamperti transform + high-order scheme + inverse transform) is not limited to this class and could theoretically be applied to other high-order strong or weak schemes. The paper does not propose new applications for the SDEs themselves but rather provides a robust numerical tool for solving existing models (like Allen–Cahn, Nagumo, and SIS) that were previously difficult to approximate with high-order accuracy while respecting their natural bounds.

More like this