Lie–Trotter Splitting Discretization

• Operator splitting, particularly the Lie–Trotter discretization, is a method that decomposes evolution equations into simpler sub-flows for efficient and robust time integration.
• The scheme achieves first-order convergence with error bounds dictated by operator commutators, making it effective for both linear and nonlinear, deterministic and stochastic systems.
• Its modular design supports practical applications in PDEs, DAEs, SPDEs, and quantum simulations, facilitating stable and efficient numerical solvers.

An operator splitting method is a numerical time-integration strategy for evolution equations whose generator decomposes naturally into a sum of operators, each of which may be exponentiated or solved more efficiently or robustly than the full problem. The Lie–Trotter splitting, also called the Godunov splitting, is a first-order (in time) discretization scheme that replaces the full evolution operator $e^{t(A+B)}$, for (possibly unbounded) operators $A$, $B$, with a product of the sub-flows $e^{tA}$ and $e^{tB}$. This method forms the foundational building block for a wide spectrum of splitting and composition schemes in computational mathematics, with rigorous error and stability analyses available for both deterministic and stochastic, linear and nonlinear, finite- and infinite-dimensional systems. Lie–Trotter splitting arises naturally from the Trotter product formula and is extensively used for PDE, ODE, DAE, and SPDE time discretization as well as in operator-theoretic and quantum simulation contexts.

1. Theoretical Foundations and General Definition

Let $H$ be a (real or complex) Hilbert or Banach space, and let $A$ and $B$ be (possibly unbounded) linear operators on $H$ such that both generate strongly continuous semigroups $e^{tA}$ and $A$0, and their sum $A$1 also generates a strongly continuous semigroup $A$2. The primary theoretical underpinning for Lie–Trotter splitting is the Trotter product formula: $A$3 This expresses the semigroup of the total generator as the strong-operator limit of compositions of the sub-semigroups. On a time interval $A$4, one thus defines, for $A$5, the approximation

$A$6

which, as $A$7, converges in norm (in appropriate function spaces and under natural regularity assumptions) to $A$8. The method extends directly to cases where the right-hand side splits into $A$9 operators: the $B$0-part Lie–Trotter splitting is

$B$1

where $B$2 are the differentiations or Lie derivatives associated with each sub-operator (Spiteri et al., 2024).

For practical time-stepping, one fixes a small step size $B$3 and applies the two sub-flows consecutively: $B$4 This procedure defines the canonical explicit-in-flow, implicit-in-operator splitting integrator.

2. Convergence, Error Analysis, and Commutator Structure

The convergence theory of Lie–Trotter discretization is fundamentally linked to commutator estimates and the noncommutativity of the split operators.

The operator-norm error for $B$5 is (for $B$6, assuming semigroup norms $B$7 for readability): $B$8 where $B$9 is the commutator (Iserles et al., 2024). This is derived via a double Duhamel expansion and explicit nesting of the integrals arising from the noncommutativity. For single-step error over $e^{tA}$0, one finds

$e^{tA}$1

so the local truncation error is $e^{tA}$2 and the global error $e^{tA}$3, establishing first-order accuracy in $e^{tA}$4.

Nonlinear and non-autonomous settings can admit similar analyses, replacing operator commutators with Lie (Fréchet) brackets. In stochastic (e.g., SPDE) or degenerate settings, regularity properties of noise or operator domains limit the maximal convergence order, often to ½ in the strong sense (Padgett et al., 2019).

For $e^{tA}$5-operator splittings, the leading local error is determined by all pairwise commutators: $e^{tA}$6 Again yielding first-order global order for arbitrary $e^{tA}$7 (Spiteri et al., 2024).

3. Implementation, Computational Aspects, and Stability

The practical implementation relies on being able to efficiently compute or approximate the exponentials $e^{tA}$8 and $e^{tA}$9—often corresponding to evolution under sub-systems, each amenable to specialized solvers, analytic formulas, or efficient algorithms, e.g., via spectral or Kronecker structures for high-dimensional PDEs (Hansen et al., 2015).

The stability of the Lie–Trotter split scheme depends on:

• The stability regions of the sub-flows (especially under stiff components).
• The interaction between sub-integration method and operator ordering.
• Potentially, for very stiff dissipative or diffusive terms, the use of implicit methods for those sub-flows. For instance, in reaction–diffusion systems, using an explicit sub-integrator for backward sub-steps can avoid stability “poles” (Wei et al., 4 Jan 2025).

For dimensionally split diffusion (e.g., 2D problems), the action of $e^{tB}$0 and $e^{tB}$1 can often be reduced to independent sets of 1D elliptic solves due to the Kronecker (tensor product) structure arising in the finite element or finite difference discretization (Hansen et al., 2015).

4. Applications Across Problem Classes

Lie–Trotter splitting is widely deployed across linear and nonlinear PDEs, stochastic evolution equations, operator-valued Riccati equations, DAEs, and biochemical/stochastic simulation contexts:

• Linear dissipative evolution: For $e^{tB}$2 maximal dissipative, spatial FEM discretization combined with Lie–Trotter or more sophisticated splittings yields full space-time error bounds $e^{tB}$3 (with $e^{tB}$4 meshsize and $e^{tB}$5 timestep) under mild regularity (Hansen et al., 2015).
• DAEs: Index-1 semi-explicit systems can be split into subsystems, each retaining DAE structure, with rigorous proofs of first-order global accuracy for the Lie–Trotter protocol (Bartel et al., 2023).
• Nonlinear damped wave/Westervelt: Separating nonlinear diffusion and reaction-type sub-flows, Lie–Trotter splitting retains first-order convergence under regularity, and is robust under strong damping (Kaltenbacher et al., 2013).
• Stochastic SPDEs/CLEs: In stochastic settings, the Lie–Trotter splitting converges at rate $e^{tB}$6, with $e^{tB}$7 the noise regularity parameter (Padgett et al., 2019, Zeng et al., 31 Mar 2026). Error decompositions separate commutator, truncation, and discretization errors, with adaptive step control strategies for efficiency.
• Differential Riccati equations: For operator-valued DREs, Lie–Trotter splitting between linear and quadratic terms yields first-order operator-norm convergence as long as initial data or nonlinearity is sufficiently regularizing (Hansen et al., 25 Apr 2025).
• Systems with dynamical boundary conditions: Splittings involving coupled bulk-boundary evolution fall under this framework. The Lie–Trotter sequential scheme yields a global $e^{tB}$8 error, with explicit matrix representations for the sub-semidroups (Csomós et al., 2020).
• General nonlinear ODEs: The Koopman–Lie semigroup viewpoint justifies direct splitting of the vector field into coordinate sub-flows, allowing for efficient high-dimensional computation via sequence of decoupled 1D problems (Banjara et al., 21 Jun 2025).

5. Comparison to Higher-Order and Alternative Splittings

While the Lie–Trotter discretization is always first-order and can be universally applied for arbitrary $e^{tB}$9-splitting, alternative higher-order methods exist but impose additional structure (e.g., palindromic/Strang splitting, Zassenhaus product exponential, complex-coefficient compositions):

• Douglas–Rachford and Peaceman–Rachford: These achieve first and second order, respectively, for maximal dissipative generators, at the expense of more complex operator algebra or additional sub-steps per time step (Hansen et al., 2015).
• Strang and palindromic splitting: The only generally available real-coefficient second-order splitting for $H$0. Extends to $H$1-operators with increased sub-flow count (e.g., $H$2 for Strang) (Spiteri et al., 2024, Iserles et al., 2024).
• Complex-coefficient second-order ($H$3-split) methods: Recent advances show that for arbitrary $H$4, two-stage complex-valued splittings with positive real parts permit second-order accuracy and robust linear stability (Spiteri et al., 2024).
• Zassenhaus expansions: High-order splitting schemes and local corrections built from nested commutators can be combined with iterative splitting for arbitrarily high order at practical computational cost (Geiser, 2012).
• Adaptive multiscale splitting: In highly stiff or multiscale stochastic systems, splitting can be dynamically coupled with an adaptive controller to balance fast/slow discretization accuracy and computational cost (Zeng et al., 31 Mar 2026).

6. Limitations and Conditions for Optimal Convergence

Attaining optimal convergence with Lie–Trotter discretization relies on:

• Sufficient regularity of the initial data and the analytical flows of $H$5 and $H$6, especially for PDE/DAE systems.
• Boundedness (or controlled growth) of commutators $H$7, as error constants depend linearly on their norms (Iserles et al., 2024).
• In stochastic and fractional-dissipation settings, the regularity of noise or lack of commutation can strictly bound the achievable strong order—e.g., Lie–Trotter splitting for SPDEs with trace-class noise converges at most in mean-square with order $H$8 (Padgett et al., 2019).
• In the presence of singular coefficients or potentials (e.g., Coulomb or less regular multipliers), the algebraic convergence rate can degrade below first order, precisely quantified via Favard or fractional Sobolev spaces (Becker et al., 2024).

In nonlinear or infinite-dimensional scenarios, first-order convergence is the universal guarantee, and more sophisticated methods require higher regularity (sometimes not available for rough data or singular coefficients).

7. Summary Table: Key Features Across Contexts

Context	Local Error	Global Order	Spatial Coupling	Remarks
Linear evolution/PDE	$H$9	$A$0	via projection/FE/FV	Error $A$1 $A$2
Nonlinear ODE/SPDE	$A$3	$A$4/$A$5	N/A	Requires noise/solution regularity
DAE, index-1	$A$6	$A$7	coupled algebraic constr.	Flows must satisfy constraints
Riccati equation (operator)	$A$8	$A$9	need domain regularity	Superlinear convergence for Strang
Fast-slow CLE (stochastic)	$B$0	$B$1	adaptive microsteps (fast)	Error split by commutator, discretization

The Lie–Trotter discretization provides the canonical, robust, and widely applicable baseline for operator splitting-based time integration, with error and stability properties that are fully characterized and serve as the reference point for more advanced splitting strategies. Its optimal use occurs in scenarios where sub-operators correspond to physically or numerically natural decompositions, enabling efficient, modular, and structure-preserving computations across deterministic and stochastic, linear and nonlinear, and finite- and infinite-dimensional evolution problems (Hansen et al., 2015, Kaltenbacher et al., 2013, Wei et al., 4 Jan 2025, Iserles et al., 2024, Padgett et al., 2019, Becker et al., 2024).