Lie–Trotter Splitting

• Lie–Trotter splitting is an operator splitting method that decomposes evolution equations into sequential subproblems to achieve first-order accuracy.
• It alternates the solution flows of sub-operators, with convergence and stability analyzed via commutator estimates and regularity conditions.
• This method underpins advanced schemes such as higher-order, low-rank, and constrained splittings applied to deterministic, stochastic, and delay systems.

The Lie–Trotter splitting is a foundational operator splitting technique for the temporal discretization of evolution equations whose dynamics can be decomposed into formally separate flows. Given a system whose right-hand side is a sum of operators or vector fields, the Lie–Trotter splitting alternates sequential solutions of the corresponding subproblems. It has rigorous first-order accuracy for deterministic, stochastic, semilinear, or constrained PDEs and SDEs, and forms the algorithmic backbone of many higher-order and optimized splitting methods. The scheme's convergence, stability, and error constants are directly determined by both the regularity of the underlying flows and the algebraic structure (commutativity, regularity of noise, and boundary conditions) of the split components.

1. Abstract Framework and the Lie–Trotter Formula

Suppose one seeks to solve an abstract evolution equation on a Banach or Hilbert space $\mathcal{H}$: $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$ where $A$ and $B$ generate strongly continuous (possibly unbounded) semigroups $e^{tA}$ and $e^{tB}$. The Lie–Trotter splitting exploits the Trotter product formula: $$e^{t(A+B)} = \lim_{n\to\infty}\bigl( e^{tA/n} e^{tB/n} \bigr)^n.$$ For practical discretization with step size $h$, one advances via

$u^{n+1} = e^{hA} e^{hB} u^n,$

exchanging the order as dictated by splitting convention and physical context (Iserles et al., 2024, Childs et al., 2019). Rigorous error representation and operator norm estimates (involving commutators $[A,B]$) serve as the basis for both deterministic and stochastic error analyses, allowing application even when $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$0 and $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$1 are unbounded or domain issues arise.

2. Convergence Theory: Error Bounds and Order Estimates

The Lie–Trotter splitting achieves first-order global accuracy, with local error $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$2 and global error $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$3 under appropriate assumptions. The fundamental commutator expansion yields

$\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$4

with the sharpness of the $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$5 term proven via exact Duhamel-variation-of-constants arguments (Iserles et al., 2024, Childs et al., 2019), and further confirmed in non-commutative settings (Barthel et al., 2019). Operator-norm error bounds also account for semigroup growth: $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$6 with prefactor depending on the semigroup logarithmic norms.

For stochastic semilinear evolution equations in Hilbert space (abstract Itô SPDEs), the strong convergence rate deteriorates with noise regularity. For the model

$\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$7

with drift $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$8 and diffusion $\frac{du}{dt} = (A + B)u(t), \qquad u(0) = u_0,$9 Lipschitz and $A$0 analytic, the splitting achieves

$A$1

where $A$2 is the spatial noise regularity: $A$3 for trace-class noise (recovering classical $A$4-order), $A$5 for rougher noise (Padgett et al., 2019). For stochastic PDEs with multiplicative noise, $A$6 is the maximal order attainable by splitting or Euler-type schemes under standard regularity.

3. Lie–Trotter Splitting Algorithms for Deterministic and Stochastic Evolution Systems

The method is instantiated by alternately solving (possibly exactly or approximately) the sub-problems corresponding to $A$7 and $A$8 over a time step $A$9. In the abstract, for $B$0:

• Apply the $B$1-flow: $B$2,
• Apply the $B$3-flow: $B$4.

For stochastic semilinear SPDEs, a typical splitting at each step involves:

• Linear semigroup: $B$5,
• Drift nonlinear step via ODE: $B$6 solving $B$7,
• Diffusion (noise) step: $B$8 via stochastic ODE $B$9, leading to the full step: $e^{tA}$0 The analysis requires Lipschitz continuity, analytic semigroups, and noise regularity constraints (Padgett et al., 2019). For spatially discretized problems (e.g., via FEM or spectral discretization), analogous bounds hold with spatial truncation error determined by the solution's Sobolev regularity.

In delay SDE/PDEs, the splitting alternates delay and non-delay flows, or, in the case of splitting stochastic correlated noise, separate Euler–Maruyama and exact SDE steps, with convergence rates controlled by both time-step and noise correlation parameters (Kelly et al., 23 Mar 2026, Bátkai et al., 2010).

For constrained evolution or boundary-coupled systems, the splitting decouples the constrained linear part (applied with exact constraint enforcement) and the unconstrained nonlinear or reaction part (Altmann et al., 2016, Csomós et al., 2020). The order remains one, with logarithmic correction factors when boundary regularity is weak.

In degenerate or low-regularity settings (e.g., logarithmic nonlinearity, filtered Boussinesq), the convergence order depends continuously on Sobolev index $e^{tA}$1; in these cases, the error is $e^{tA}$2, and careful discrete functional analysis is needed (Zhang et al., 2024, Ji et al., 2024).

4. Error Analysis in Stochastic and Low-Regularity Regimes

Stochastic or degenerate regimes significantly influence the attainable order of convergence. When the driving noise is trace class, analytic smoothing permits the optimal strong order $e^{tA}$3. For rougher or spatially inhomogeneous noise, the strong order is $e^{tA}$4 with $e^{tA}$5 the effective regularity parameter.

The techniques underpinning these results are:

• Stability estimates for the splitting operator, leveraging Lipschitz and stochastic regularity (Padgett et al., 2019, Berg et al., 2020).
• Local consistency via Taylor–Itô expansions, showing that the one-step mean-square defect is $e^{tA}$6.
• Discrete Gronwall inequalities to promote one-step error to a global error bound, making essential use of smoothing from analytic semigroups and Itô isometries.

In the presence of multiplicative, degenerate, or highly oscillatory noise (e.g., stochastic Manakov system or Dirac equation in the nonrelativistic regime), order $e^{tA}$7 accuracy is achieved in expectation or with high probability, with order reduction in the presence of resonances, correlation, or insufficient regularity (Berg et al., 2020, Bessaih et al., 2013, Bao et al., 2019). Notably, for stochastic delay systems with correlated Brownian noises, the $e^{tA}$8 strong order is lost when the noise correlation parameter $e^{tA}$9, with a $e^{tB}$0 error floor that cannot be removed by time refinement (Kelly et al., 23 Mar 2026).

In low-regularity deterministic PDEs (e.g., LogSE or "good" Boussinesq), the error analysis leverages discrete Sobolev/Bourgain spaces to obtain rate $e^{tB}$1 for $e^{tB}$2 with $e^{tB}$3 (Zhang et al., 2024, Ji et al., 2024).

5. Extensions: High-Order, Low-Rank, and Constrained Operator Splittings

While classical Lie–Trotter is strictly first order, systematic extensions to higher-order splittings can be constructed via symmetric or optimized compositions (Strang, Suzuki, Forest–Ruth, Hall-basis minimizations) (Barthel et al., 2019). For dissipative or irreversible equations (reaction-diffusion, Ginzburg–Landau), affine combinations of Lie–Trotter steps at varying sub-step sizes achieve higher order without negative steps (Leo et al., 2013). For matrix ODEs and semilinear stiff PDEs, Lie–Trotter splitting combined with low-rank projector splitting yields robust and efficient methods, provably stable even in the presence of small singular values and allowing time-steps independent of stiffness (Ostermann et al., 2018, Zhao et al., 2020).

Constrained systems and systems with dynamic boundary conditions are naturally suited to splitting: the Lie–Trotter method isolates the constraint-enforcement step and thereby enhances stability and tractability, especially for PDAEs and evolution equations with dynamical boundary coupling (Altmann et al., 2016, Csomós et al., 2020).

6. Applications and Practical Algorithms

Lie–Trotter splitting is fundamental to time integration schemes for a wide class of systems:

• Quantum dynamics (lattice Hamiltonians, quantum circuits) via exponentiation of non-commuting Hamiltonian components, exploiting the splitting for efficient simulation and matching theoretical error bounds with optimized circuits (Childs et al., 2019, Barthel et al., 2019).
• Stochastic evolution equations, including Navier–Stokes, nonlinear Schrödinger/Manakov, semilinear SPDEs, and delay systems, with guaranteed $e^{tB}$4 strong order in many practical regimes (Bessaih et al., 2013, Berg et al., 2020, Padgett et al., 2019, Bátkai et al., 2010).
• Constrained reaction–diffusion, mechanical systems with holonomic constraints, and dynamical boundary conditions (Altmann et al., 2016, Csomós et al., 2020).
• Matrix differential equations, Lyapunov or Riccati equations, and fractional Ginzburg–Landau via low-rank or projector-splitting Lie–Trotter schemes (Ostermann et al., 2018, Zhao et al., 2020).

Implementation typically involves sequential application of sub-solvers (ODE, PDE, SDE, or DDE solvers) over each subflow, plus any necessary projections or correction steps for invariance and constraint enforcement. Numerical experiments corroborate the theoretical error rates in a wide range of settings.

7. Optimality, Limitations, and Advanced Developments

The first-order accuracy (or $e^{tB}$5 order in mean-square for stochastic equations) is optimal for the Lie–Trotter splitting in the presence of multiplicative noise, non-commuting components, or low regularity, unless commutativity or special structure is present (Padgett et al., 2019, Childs et al., 2019). For non-resonant step sizes or higher regularity, improved rates may occur, but these are not typical in general settings (e.g., see the non-resonant improvement in the nonlinear Dirac case (Bao et al., 2019)).

Limitations arise in strongly degenerate or correlated noise, in PDEs with rough data, and for systems with insufficient analytic smoothing: in all such contexts, the attainable order of convergence is dictated by the weakest link (noise regularity, Sobolev index, boundary smoothness). The composition, projection, and domain-invariance requirements, as well as cost of subflow resolution, remain the critical practical and theoretical considerations in deployment.

Recent work also emphasizes the use of optimized multi-stage, composition, or symmetrized Lie–Trotter splittings ("Suzuki decompositions", higher-order extrapolation) for quantum simulation, irreversible systems, and more, allowing systematic error control at increased implementation complexity (Childs et al., 2019, Barthel et al., 2019, Leo et al., 2013, Banjara et al., 21 Jun 2025).

---

References:

• (Padgett et al., 2019) Convergence of an operator splitting scheme for abstract stochastic evolution equations
• (Iserles et al., 2024) An elementary approach to splittings of unbounded operators
• (Childs et al., 2019) A Theory of Trotter Error
• (Barthel et al., 2019) Optimized Lie-Trotter-Suzuki decompositions for two and three non-commuting terms
• (Altmann et al., 2016) Splitting methods for constrained diffusion-reaction systems
• (Bátkai et al., 2010) Operator splitting for dissipative delay equations
• (Berg et al., 2020) Lie-Trotter Splitting for the Nonlinear Stochastic Manakov System
• (Bessaih et al., 2013) Splitting up method for the 2D stochastic Navier-Stokes equations
• (Ostermann et al., 2018) Convergence of a low-rank Lie–Trotter splitting for stiff matrix differential equations
• (Zhao et al., 2020) A low-rank Lie-Trotter splitting approach for nonlinear fractional complex Ginzburg-Landau equations
• (Banjara et al., 21 Jun 2025) Operator Splitting Methods: Numerical Solutions of Ordinary Differential Equations via Separation of Variables
• (Zhang et al., 2024) Low regularity estimates of the Lie-Totter time-splitting Fourier spectral method for the logarithmic Schrödinger equation
• (Ji et al., 2024) Filtered Lie-Trotter splitting for the "good" Boussinesq equation: low regularity error estimates
• (Bao et al., 2019) Uniform error bounds of time-splitting methods for the nonlinear Dirac equation in the nonrelativistic regime without magnetic potential
• (Kelly et al., 23 Mar 2026) On splitting strategies for the numerical solution of stochastic delay differential equations with correlated noises
• (Csomós et al., 2020) Operator splitting for abstract Cauchy problems with dynamical boundary condition
• (Leo et al., 2013) High order methods for irreversible equations