A set of techniques for constructing a low-dimensional surrogate model that approximates the input–output behavior of a high-dimensional dynamical or parametric system while retaining essential features relevant to analysis, control, or optimization.

Principle

Replace a high-dimensional operator or state space by a lower-dimensional representation that preserves dominant modes, transfer characteristics, or input–output maps so that computation is cheaper but predictions remain accurate within a specified tolerance.

Demonstration

Constructing a reduced-order model for a fluid-structure interaction problem by projecting the full finite-element discretization onto a basis of empirical modes obtained from simulation snapshots, then using the reduced system to evaluate design changes orders of magnitude faster.

Misapplication

Using an ad hoc low-dimensional projection without checking representativeness of the basis, leading to large errors when the system is driven outside the snapshot set or when neglected modes become excited.

Consequence

When applied correctly, model order reduction yields surrogate models that speed up simulation, enable real-time control and parameter studies, and reduce storage and communication costs while providing quantified approximation error.

Reversal

The inverse concept is model order expansion or full-order fidelity where no dimensional reduction is applied and the complete high-dimensional model is retained for simulation and analysis.

Boundary

Applies to linear and nonlinear, time-dependent and steady systems where a dominant low-dimensional structure exists or can be approximated; excludes problems where dynamics are inherently high-dimensional across all relevant inputs or where reduction destroys essential constraints (e.g., conservation laws) unless structure-preserving reductions are used.

Semantic Tension

Competes with surrogate modeling via purely data-driven regression: both reduce cost but differ in whether they preserve system-theoretic structure (e.g., linear operators, symmetries) versus fitting input–output maps empirically.

Synthesis

Model order reduction is the disciplined replacement of a computationally expensive, high-dimensional system by a compact surrogate that preserves the system's critical input–output behavior and mathematical structure sufficiently for the intended analysis or control tasks.