Reduced Basis Methods (RBM)

• Reduced Basis Methods (RBM) are model order reduction techniques that construct low-dimensional trial subspaces for rapid, certified solutions of parametrized PDEs.
• RBM significantly reduces computational cost by using snapshots and Galerkin projection, making many-query applications in engineering more efficient.
• RBM incorporates strong a posteriori error control mechanisms to ensure accuracy while streamlining the workflow from snapshot selection to surrogate model construction.

Reduced Basis Methods (RBM) are model order reduction techniques that deliver rapid, certified solutions for parametrized partial differential equations (PDEs), with strong a posteriori error control and substantial savings in computational cost for many-query applications. RBM accomplishes this by constructing a low-dimensional trial subspace from solutions (“snapshots”) at carefully selected parameter values and using Galerkin projection to produce an efficient, accurate surrogate for the high-fidelity model. This article presents the foundational principles, computational workflows, error-certification mechanisms, and selected advanced variants and applications of RBM across physical and engineering domains.

1. Mathematical Foundations and Standard Workflow

The RBM framework addresses a parametrized PDE in variational form:

$$a(u(\mu), v; \mu) = f(v; \mu) \qquad \forall v \in X$$

where $X$ is a Hilbert (trial/test) space, $\mu$ is a $P$-dimensional parameter vector ($\mu\in D\subset\mathbb{R}^P$), and $a$ is a bilinear or nonlinear form depending continuously (and, often, affinely) on $\mu$.

The high-fidelity (“truth”) problem is solved on a large finite-dimensional subspace, yielding expensive systems—for example, in Maxwell's equations for computational lithography, $N\sim3\times10