A class of numerical strategies that partition a global computational domain into subdomains, solve (exactly or approximately) boundary-value problems on these subdomains—possibly in parallel—and enforce interface conditions through coupling, iteration, or coarse-grid corrections to obtain the global solution.

Principle

Reduce large-scale problems by spatial partitioning: enforce local solves on subdomains while coordinating across interfaces via transmission conditions, Lagrange multipliers, iterative Schwarz-type exchanges, or global coarse corrections so that local computations assemble into a consistent global solution.

Demonstration

Solving a large elliptic PDE on a distributed-memory cluster: split the domain into overlapping subdomains, on each node solve the local Dirichlet or Neumann problem with given interface data, exchange updated interface traces with neighbors (additive or multiplicative Schwarz), iterate until interface residuals are small, optionally include a coarse problem to accelerate convergence.

Misapplication

Using non-matching interface discretizations without appropriate projection or enforcing incompatible interface conditions that violate conservation or continuity, leading to spurious modes, loss of accuracy, or divergence of the iterative coupling.

Consequence

Domain decomposition yields scalable solvers suited to parallel computing by localizing work and communication; with proper transmission conditions and coarse corrections it attains mesh-independent convergence and efficient use of distributed resources.

Reversal

Monolithic global solve: form and solve the entire global discretized system on one processor or without subdomain partitioning; while avoiding interface iteration, this approach often lacks parallel scalability and can be memory-limited for very large problems.

Boundary

Scope: PDEs and variational problems where domain partitioning preserves essential properties (ellipticity, conservation) and where interface operators can be posed and solved; excludes problems whose physics are inherently global and nonseparable by domain cuts (strong nonlocal kernels), or where communication cost eclipses local solve cost making decomposition inefficient.

Semantic Tension

Tension between subdomain autonomy and interface coupling: larger subdomains reduce interface overhead and improve local accuracy but increase per-subdomain cost and memory; smaller subdomains favor parallelism and locality but require more coordination and robust transmission conditions to ensure global accuracy and convergence.

Synthesis

Domain Decomposition Method is a framework that divides a global problem into subdomain problems solved locally and coupled through interface conditions and iterations (or coarse corrections), enabling parallel, memory-efficient solution of large PDE systems while preserving global consistency when properly designed.