Definition
A numerical technique that approximates derivatives by replacing them with algebraic difference quotients evaluated on a discrete grid of points, producing algebraic equations that approximate differential equations.

Principle

Principle
Replace continuous derivatives with discrete difference operators whose truncation error, together with stability properties of the resulting scheme, determine convergence to the continuous solution.

Demonstration

Demonstration
Solve the one‑dimensional heat equation by replacing ∂²u/∂x² with a central second‑difference (u_{i+1}-2u_i+u_{i-1})/Δx² to obtain a system of ODEs or algebraic equations depending on time discretization.

Misapplication

Misapplication
Applying a naive explicit finite difference discretization for a parabolic PDE with too large a time step that violates the CFL/stability condition, producing numerical blow up or nonphysical oscillations.

Consequence

Consequence
Produces sparse algebraic systems or ODE systems whose solution approximates pointwise values of the continuous field; error scales with the chosen difference order and grid spacing.

Reversal

Reversal
Instead of approximating derivatives at points, adopt an integral or weak formulation (finite volume or finite element) that enforces conservation or weak continuity rather than pointwise derivative approximations.

Boundary

Boundary
Applies primarily to problems where domain geometry and boundary conditions are compatible with structured or logically rectangular grids; direct finite difference stencils are less appropriate on highly irregular domains or unstructured meshes without modification.

Semantic Tension

Semantic Tension
Competes with finite volume methods (which emphasize flux balance across control volumes) and finite element methods (which use weak forms); finite difference emphasizes local pointwise derivative approximation and compact stencils.

Synthesis

Synthesis
A straightforward, grid‑based approach to turn differential operators into algebraic difference operators: simple to implement on regular grids, governed by truncation error and stability, but limited by geometry and conservation considerations unless adapted.