A numerical strategy to approximate the solution of an evolution equation generated by a sum of operators by sequentially applying the flows or solution operators of the individual parts over substeps.

Principle

Approximate exp((A+B)Δt) by compositions of exp(A Δt_i) and exp(B Δt_j) so that local error is controlled by commutators [A,B]; higher-order schemes are obtained by symmetric compositions and adjusted substep sizes.

Demonstration

Solve advection–diffusion by alternating an advection step (using a conservative transport solver) and a diffusion step (implicit elliptic solver); Strang splitting uses half a diffusion step, a full advection step, then another half diffusion to achieve second-order accuracy.

Misapplication

Splitting stiff or strongly noncommuting operators with large time steps without accounting for commutator error or stability properties, leading to loss of conserved quantities, spurious oscillations, or numerical instability.

Consequence

Permits use of specialized solvers for subproblems, reduces implementation complexity, and often improves efficiency and stability for mixed-physics problems when time steps and ordering are chosen appropriately.

Reversal

A monolithic integrator treats the combined operator directly; it can preserve global invariants and treat coupling implicitly but typically requires solving a more complex coupled system.

Boundary

Applies to time-evolution semigroups or flows where sub-operators generate solvable substeps; not suitable when splitting violates essential constraints (for example positivity, mass conservation) unless corrected.

Semantic Tension

Competes with implicit–explicit (IMEX) time-stepping and fully implicit solvers: IMEX splits stiff vs nonstiff terms inside a single integrator framework while operator splitting composes distinct solvers and may introduce splitting-specific errors.

Synthesis

A composition-based integration technique that decomposes complex evolution operators into simpler subflows, trading a controlled splitting error for modularity and computational efficiency when sub-operators are easier to solve independently.