Lyapunov-Tamed Euler Scheme Overview

• The Lyapunov-Tamed Euler scheme is a numerical integrator that uses Lyapunov functionals to control super-linear or singular behaviors in stochastic differential equations.
• It employs explicit taming of the drift and diffusion components to prevent blow-up, ensuring uniform moment bounds and geometric ergodicity.
• The scheme is widely applied in simulating complex stochastic dynamics, including the stochastic Allen-Cahn equation and systems with multiplicative or jump noise.

The Lyapunov-Tamed Euler (LTE) scheme is a class of explicit and semi-implicit numerical integrators for stochastic differential equations (SDEs) and stochastic partial differential equations (SPDEs) with super-linear or singular drift and diffusion coefficients, designed to ensure long-time stability, geometric ergodicity, and strong convergence rates by incorporating Lyapunov functional control directly into the discretization. The scheme is grounded in the interplay between Lyapunov drift conditions for moment control and explicit "taming" of nonlinearity to preclude blow-up, thereby enabling robust simulation of systems beyond the reach of classical Euler–Maruyama–type methods.

1. Mathematical Formulation and Origin

The LTE scheme addresses stochastic evolution equations, notably SPDEs of the form

$dX(t) = (A X(t) + F(X(t)))\,dt + G(X(t))\,dW(t)$

where $A$ is typically a Laplace-type operator on a Hilbert space $H$, $F$ exhibits super-linear growth, and $W$ is a $Q$-Wiener process. In finite dimensions, SDEs or SODEs with drift $b(x)$ and diffusion $\sigma(x)$ exceeding linear growth, and possibly with singularities, are also covered within the LTE conceptual framework (Liu et al., 26 Feb 2025, Johnston et al., 23 Jan 2026, Liu et al., 2024).

The discretization replaces the standard Euler step by a "tamed" version: $$Z_{n+1} = Z_n + \tau A Z_{n+1} + \tau F_\tau(Z_n) + G_\tau(Z_n) \Delta W_n,$$ with $F_\tau$, $G_\tau$ resulting from explicit taming, e.g.,

$$f_\tau(\xi) = \frac{f(\xi)}{\sqrt{1+\tau|\xi|^{2q}}}, \quad g_\tau(\xi) = \frac{g(\xi)}{\sqrt{1+\sqrt{\tau}|\xi|^{q}}}$$

(where exponents and forms reflect the polynomial degree of nonlinearity or singularity) (Liu et al., 26 Feb 2025, Liu et al., 2024, Gyöngy et al., 2014). For SDEs with even stronger singularity (not locally Lipschitz), the taming adopts a truncation dictated by an underlying Lyapunov function $V$, ensuring suppression near singular points (Johnston et al., 23 Jan 2026).

This approach readily extends to Galerkin-based finite-element full discretization for SPDEs, and encompasses drift-only or joint taming of drift/diffusion according to the noise structure (Liu et al., 26 Feb 2025).

2. Lyapunov Functionals and A Priori Stability

At the core of the LTE methodology is the preservation of a Lyapunov functional $V$ under both the original dynamics and the tamed numerical scheme. A typical example in the parabolic SPDE context is

$V(u) := \|u\|^2 + 2\tau \|\nabla u\|^2,$

for $u$ in an appropriate Sobolev space, ensuring coercivity and dissipativity (Liu et al., 26 Feb 2025). In finite dimensions, $V(x) = \|x\|^2$ or generalizations control polynomial or singular moments (Liu et al., 2024, Sabanis, 2013).

The discretized process $\{Z_n\}$ is shown to satisfy a one-step Lyapunov drift condition: $$\mathbb{E}[V(Z_{n+1})|\mathcal{F}_n] \leq (1-K_1\tau)V(Z_n) + K_2\tau$$ for constants $K_1 > 0$, $K_2$, and sufficiently small $\tau$, mirroring the continuous-time dissipativity (Liu et al., 26 Feb 2025, Bao et al., 2024). This implies uniform-in-time moment bounds, precluding explosion over arbitrarily long horizons—even in the presence of highly non-linear or singular terms.

The Lyapunov functional is also used to tune or localize the taming (e.g., truncation for unbounded coefficients), and it fundamentally justifies the entire structure of the LTE discretization (Johnston et al., 23 Jan 2026).

3. Taming Mechanisms and Variant Schemes

The distinguishing feature of LTE schemes is the explicit modification (taming) of the nonlinearity in both drift and, when required, diffusion terms. Several taming strategies appear in the literature:

• Power-law denominator: For coefficients growing as $|x|^k$, the taming often takes the form

$b_\tau(x) = \frac{b(x)}{\sqrt{1 + \tau |x|^{2q}}},$

ensuring the numerator's growth is suppressed for large $|x|$ (Liu et al., 26 Feb 2025, Liu et al., 2024).

• Truncation via Lyapunov function: For singular coefficients, the drift is replaced by $$b_n(x) = b(x) \mathbf{1}_{V(x)\leq n^{1/2}, |b(x)|\leq n^{1/2}}$$, cutting off excursions near singularities at the boundary of $D$ (Johnston et al., 23 Jan 2026).
• Drift-only or full (drift and diffusion) taming: In pure additive noise regimes, drift-only taming suffices. In multiplicative settings, the diffusion must often be tamed with a similar rule (Liu et al., 26 Feb 2025, Liu et al., 2024).
• Semi-implicit update: The LTE update can be formulated as a linearly implicit solve, such as $(I - \tau A)Z_{n} = \cdots$, crucial for stiff or parabolic operator-dominated problems (Liu et al., 26 Feb 2025).

For SDEs with jumps, analogous taming can be implemented (e.g., modifying the jump-drift component), and "semi-tamed" splitting variants are available (Tambue et al., 2015).

4. Geometric Ergodicity and Long-Time Error

A central result of the LTE framework is the discrete-time analogue of Foster–Lyapunov drift, guaranteeing geometric ergodicity: $$\mathbb{E}[V(Z_n)] \leq e^{-K_1 n\tau}V(Z_0) + \frac{K_2}{K_1},$$ with the resulting Markov chain admitting a unique invariant measure and enjoying exponential convergence in moments and (in finite dimensions) in $L^1$-Wasserstein distance (Liu et al., 26 Feb 2025, Bao et al., 2024).

Recent studies provide explicit quantitative error bounds between invariant measures of the SDE and its tamed numerical counterpart, typically of order $\mathcal{O}(\tau^{1/2})$ in $W_1$ due to coupling arguments and Lyapunov techniques (Bao et al., 2024).

For SDEs with non-globally Lipschitz, singular, or even exploding coefficients, geometric ergodicity is proven under general Lyapunov-based dissipativity, often with sharp rates (Johnston et al., 23 Jan 2026, Bao et al., 2024, Liu et al., 2024).

5. Strong and Weak Convergence Results

LTE-type schemes yield optimal strong and weak convergence rates, provided the underlying SDE/SPDE coefficients fulfill additional one-sided Lipschitz and polynomial-growth assumptions:

• For SODEs/SPDEs with super-linear but polynomial growth:
  • Strong order $1/2$ in $\tau$ for multiplicative noise;
  • Strong order $1$ in $\tau$ (and up to $h^2$ in space) for additive noise provided additional regularity (Liu et al., 26 Feb 2025, Liu et al., 2024, Sabanis, 2013).
• For singular SDEs (coefficients not even locally Lipschitz), rates of strong convergence $1/2$ or 1 (with higher regularity/truncation) are established, with explicit constants depending on Lyapunov functionals and initial data (Johnston et al., 23 Jan 2026).
• Weak error bounds of the form

$$\sup_{n\geq 0} \left| \mathbb{E}[f(x_{t_n})] - \mathbb{E}[f(J_{t_n})] \right| \leq C\left( \tau \|x\|^{2q} + \tau^{1/2} \right)$$

hold uniformly in time for a large class of observables $f$ (Angeli et al., 2023).

Spatial discretization (e.g., Galerkin finite element) can be handled with the same Lyapunov-tamed structure, resulting in joint space–time error regimes, e.g., $\mathcal{O}(h^{1+\gamma} + \tau^{1/2})$ under suitable regularity assumptions (Liu et al., 26 Feb 2025).

6. Applications and Algorithmic Implementation

LTE schemes have been successfully applied in a range of settings:

• Super-linear parabolic SPDEs, including the stochastic Allen-Cahn equation with both multiplicative and additive trace-class noise (Liu et al., 26 Feb 2025, Gyöngy et al., 2014).
• Super-linear SDEs and ODEs with polynomially growing drifts and diffusions (SODEs) (Liu et al., 2024, Sabanis, 2013).
• SDEs with singular/degenerate drift, such as mean-field particle systems with singular (e.g., Coulomb) repulsion (Johnston et al., 23 Jan 2026).
• Systems with Poisson jump noise under non-global Lipschitz and polynomial growth (drift and jumps), using tamed and semi-tamed variants (Tambue et al., 2015).

For SPDEs, practical implementation combines finite-element spatial projection, nodewise evaluation of tamed nonlinearities, and solving a linear system at each time step (see step-by-step prescription in (Liu et al., 26 Feb 2025)). For finite-dimensional or particle systems, the algorithm reduces to explicit steps with tamed drift and/or diffusion (Sabanis, 2013, Liu et al., 2024, Johnston et al., 23 Jan 2026).

7. Theoretical Implications and Comparative Analysis

The LTE approach generalizes the tamed Euler schemes by explicitly tying the taming mechanism to any coercivity/dissipativity controlled by the Lyapunov functional, enabling robust explicit time discretization for both polynomially super-linear and singular SDE/SPDEs. The method is strictly superior to classical Euler–Maruyama in these regimes, as standard explicit methods can display finite-time blow-up or divergence in moments (Liu et al., 26 Feb 2025, Bao et al., 2024).

The Lyapunov-tamed schemes accommodate a variety of noise structures (additive, multiplicative, jump), deliver provable geometric ergodicity and uniform-in-time stability, and achieve optimal asymptotic convergence—often matching those of implicit methods at reduced computational cost (Liu et al., 26 Feb 2025, Tambue et al., 2015).

A plausible implication is that further advances in the discretization of nonlinear stochastic systems with pathologies will likely exploit the systematic interplay between Lyapunov-dissipativity and numerically adaptive taming.

---

Principal References:

• "Geometric Ergodicity and Optimal Error Estimates for a Class of Novel Tamed Schemes to Super-linear Stochastic PDEs" (Liu et al., 26 Feb 2025)
• "Geometric Ergodicity and Strong Error Estimates for Tamed Schemes of Super-linear SODEs" (Liu et al., 2024)
• "A Lyapunov-tamed Euler method for singular SDEs" (Johnston et al., 23 Jan 2026)
• "Geometric ergodicity of modified Euler schemes for SDEs with super-linearity" (Bao et al., 2024)
• "Convergence of tamed Euler schemes for a class of stochastic evolution equations" (Gyöngy et al., 2014)
• "A note on tamed Euler approximations" (Sabanis, 2013)
• "Uniform in time convergence of numerical schemes for stochastic differential equations via Strong Exponential stability" (Angeli et al., 2023)
• "Stability of the semi-tamed and tamed Euler schemes for stochastic differential equations with jumps" (Tambue et al., 2015)