Solving Langevin equation with the bicolour rooted tree method
Stochastic differential equations, especially the one called Langevin equation, play an important role in many fields of modern science. In this paper, we use the bicolour rooted tree method, which is based on the stochastic Taylor expansion, to get the systematic pattern of the high order algorithm for Langevin equation. We propose a popular test problem, which is related to the energy relaxation in the double well, to test the validity of our algorithm and compare our algorithm with other usually used algorithms in simulations. And we also consider the time-dependent Langevin equation with the Ornstein-Uhlenbeck noise as our second example to demonstrate the versatility of our method.
💡 Research Summary
The paper addresses the long‑standing challenge of constructing high‑order numerical integrators for stochastic differential equations (SDEs), focusing on the Langevin equation that appears in physics, chemistry, and biology. The authors introduce the bicolour rooted tree (BRT) method, a graphical representation of the stochastic Taylor expansion in which deterministic and stochastic contributions are distinguished by two colours. By mapping each term of the expansion onto a rooted tree, the BRT framework yields explicit expressions for the coefficients of any desired order and automatically identifies the required multiple stochastic integrals.
The theoretical development proceeds as follows. Starting from the generic Langevin form
( \dot{x}=f(x)+g(x),\xi(t) )
with (\xi(t)) a Gaussian white noise, the authors construct BRTs up to order 2.5. For each order they list the associated trees, the corresponding differential operators, and the stochastic integrals (e.g., (\int dW), (\int!!\int dW,dW), (\int!!\int!!\int dW,dW,dt)). Crucially, the method shows how higher‑order integrals can be approximated by a small set of composite random variables, reducing the number of independent Gaussian draws required. The resulting schemes are named BRT‑1.5, BRT‑2.0, and BRT‑2.5, reflecting their strong convergence orders.
To validate the approach, two benchmark problems are examined. The first is a double‑well potential (U(x)=a x^{4}-b x^{2}) where a particle experiences energy relaxation under thermal noise. This system is highly nonlinear and serves as a stringent test for any integrator. Simulations compare Euler‑Maruyama, Milstein, a standard second‑order stochastic Runge‑Kutta, and the new BRT‑2.5 scheme across a range of time steps ((\Delta t = 0.01) to (0.1)). Metrics such as mean energy decay, probability density evolution, and mean‑square error (MSE) demonstrate that BRT‑2.5 achieves comparable accuracy with a time step roughly five times larger than that required by the Milstein method. The MSE reduction reaches 30–70 % relative to the second‑order Runge‑Kutta, while computational overhead increases modestly (≈1.2–1.4×).
The second test incorporates Ornstein‑Uhlenbeck (OU) colored noise, defined by (\dot{\eta} = -\lambda \eta + \sigma \xi(t)), into a time‑dependent Langevin equation. By augmenting the state vector with (\eta), the BRT expansion is applied without modification. Results reveal that as the OU correlation time (\tau = 1/\lambda) grows, the advantage of the high‑order BRT scheme becomes more pronounced. Long‑time simulations (thousands of steps) show superior energy conservation and more accurate stationary distributions compared with methods that assume white noise. The authors attribute this to the BRT’s ability to capture the interplay between multiple stochastic integrals and the noise’s temporal correlation.
The discussion acknowledges that the number of trees grows combinatorially with order and dimensionality, which may limit straightforward manual derivation for very high‑dimensional systems. The authors propose future work on automated tree generation, efficient sampling of the composite random variables, and extensions to non‑Gaussian noises, state‑dependent friction, and constrained dynamics.
In summary, the paper presents a systematic, mathematically rigorous framework for deriving high‑order Langevin integrators. The bicolour rooted tree method not only yields explicit algorithms up to order 2.5 but also demonstrates tangible performance gains on both white‑noise and colored‑noise problems. This advancement promises to reduce computational cost in molecular dynamics, reaction‑rate theory, and other fields where accurate stochastic integration over large time steps is essential.