Stabilizing Explicit Solvers for Stochastic Differential Equations with Superlinear Growth

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2109 papers180 conceptsfrozen 2026-10-04

In numerical analysis, the standard "workhorse" for simulating stochastic differential equations (SDEs) is the Euler-Maruyama method. It is simple, explicit, and computationally cheap. However, for a vast class of realistic models in finance, biology, and physics, this workhorse is prone to "numerical explosions"—simulations where the values suddenly diverge to infinity, even when the underlying physical system is perfectly stable.

The research direction synthesized here focuses on taming and projecting these explicit numerical schemes to handle systems with non-globally Lipschitz (superlinear) coefficients, providing the stability of expensive implicit methods at the cost of simple explicit ones.

The Failure of the Global Lipschitz Assumption

Most classical convergence proofs for SDE solvers rely on the global Lipschitz condition. Intuitively, this assumes that the "forces" driving the system (the drift and diffusion) do not grow faster than a linear function of the state. If the state doubles, the force at most doubles.

However, as Desmond Higham, Andrew M. Stuart, and Xuerong Mao demonstrated in Exponential Mean-Square Stability of Numerical Solutions to Stochastic Differential Equations, explicit methods often fail when this assumption is violated. In systems with polynomial growth—such as the cubic drift found in the FitzHugh-Nagumo neuron model or the stochastic Burgers equation—a single large stochastic increment can push the numerical solution into a region where the drift is so powerful that the next discrete step flings the solution toward infinity.

Arnulf Jentzen and Martin Hutzenthaler formalized this in Numerical approximations of stochastic differential equations with non-globally Lipschitz continuous coefficients, showing that the moments of standard Euler approximations can diverge to infinity in finite time, even if the true solution remains bounded.

The Mechanism of Taming

To fix this without resorting to computationally heavy implicit solvers (which require solving a nonlinear equation at every time step), researchers introduced the concept of taming.

The intuition behind taming is to "dampen" the drift function when it becomes too large. If the original drift is $f(x)$, a tamed version might look like: $$\bar{f}(x) = \frac{f(x)}{1 + h^\alpha |f(x)|}$$ where $h$ is the step size. When $|f(x)|$ is small, the denominator is nearly 1, and the scheme behaves like standard Euler. But if a noise spike pushes $x$ into a high-growth region, the denominator grows, effectively capping the step and preventing the "explosion."

István Gyöngy, David Siska, and Sotirios Sabanis proved in Convergence of tamed Euler schemes for a class of stochastic evolution equations that these schemes are strongly convergent for stochastic evolution equations with super-linear drift. This allows for stable simulations of complex phenomena like fluid turbulence or neural signals using fast, explicit updates.

Projection and Structure Preservation

An alternative to taming is the projected Euler method. Instead of modifying the function, this method modifies the state: if a numerical step lands outside a predefined "safe" ball (whose radius depends on the step size), the solution is projected back onto the surface of that ball. Xiaojie Wang and Xiao-Ming Wu established in Strong convergence rates for long-time approximations of SDEs with non-globally Lipschitz continuous coefficients that such projected methods maintain a robust convergence rate of order 1/2 even over infinite time horizons.

Beyond simple stability, these modified schemes are often designed to preserve qualitative properties of the original system:

Why It Matters: From Finance to SPDEs

The impact of this research direction is most visible in high-stakes simulation environments. In quantitative finance, where models often feature super-linear growth, these stabilized schemes are essential for techniques like Multilevel Monte Carlo (MLMC). As Desmond Higham notes in An introduction to multilevel Monte Carlo for option valuation, without stabilized discretization, the variance reduction benefits of MLMC cannot be guaranteed because the "coarse" paths would frequently diverge.

In the realm of Stochastic Partial Differential Equations (SPDEs), these methods have enabled the first explicit, strongly convergent solvers for the stochastic Burgers equation and the stochastic Allen-Cahn equation. By preventing numerical blow-ups, researchers can now rigorously simulate the dissipative dynamics of phase transitions and fluid flows that were previously computationally prohibitive.

Papers in this version

Showing 60 of 109 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 64 papers
  2. Version 1 · 2026-10-04 · 109 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_superlinear_drift_divergence_a9616be2,
  title = {superlinear drift divergence},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/superlinear-drift-divergence/rdr_a9616be2ffb7444497d73b45c4992b65},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_stabilizing_explicit_solvers_for_stochas_a9616be2_v1,
  title = {Stabilizing Explicit Solvers for Stochastic Differential Equations with Superlinear Growth},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/stabilizing-explicit-solvers-for-stochastic-differential-equations-with/rdr_a9616be2ffb7444497d73b45c4992b65/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}