Polynomial Chaos Expansion¶

Represent a finite-variance model response in polynomials orthogonal to the probability law of declared random inputs, enabling coefficient-based uncertainty propagation and moments.

Version: v2 · 2026-09-06 · History
Domain-specific #: 2505
Origin domain: uncertainty quantification
Subdomain: stochastic spectral methods
Aliases: Polynomial chaos, PCE, Generalized polynomial chaos

Core Idea¶

A polynomial chaos expansion represents a square-integrable random output \(Y\) as a series of multivariate polynomials in a declared random input vector \(\xi\). The basis polynomials are orthogonal with respect to the probability measure of \(\xi\):

\[ Y=\sum_{\alpha}c_\alpha\Psi_\alpha(\xi),\qquad \mathbb E[\Psi_\alpha\Psi_\beta]=\gamma_\alpha\delta_{\alpha\beta}. \]

In computation the series is truncated to a finite multi-index set. Coefficients are found intrusively through stochastic Galerkin projection or nonintrusively through quadrature, regression, or model evaluations. The resulting expansion is a stochastic spectral representation and often a surrogate for uncertainty propagation.

Scope of Application¶

PCE is used for uncertainty quantification in differential equations, stochastic finite elements, dynamical systems, reliability, sensitivity analysis, and surrogate modeling. Xiu and Karniadakis generalized classical Gaussian/Hermite chaos to distribution-matched polynomial families and applied it to stochastic differential equations. Related work uses generalized chaos for elliptic and flow problems with uncertain coefficients, forcing, or boundaries.

The method is most effective when the response depends smoothly on a modest number of well-characterized random inputs. Multi-element, sparse, adaptive, and compressed approaches extend the range but do not remove dimensional scaling.

Clarity¶

With an orthogonal basis, the exact projection coefficient is

\[ c_\alpha=\frac{\mathbb E[Y\Psi_\alpha]}{\mathbb E[\Psi_\alpha^2]}. \]

If \(\Psi_0=1\), then \(\mathbb E[Y]=c_0\). For an orthonormal basis, \(\gamma_\alpha=1\) and

\[ \operatorname{Var}(Y)=\sum_{\alpha\ne0}c_\alpha^2. \]

Manages Complexity¶

Repeated evaluation of a costly deterministic model across uncertain inputs can be prohibitive. PCE separates expensive coefficient construction from cheap downstream use. Once an accurate expansion exists, means, variances, Sobol-type sensitivity contributions, and many response evaluations follow directly from coefficients and orthogonality.

The representation also exposes interaction order. A multi-index with nonzero components for two inputs represents a polynomial interaction; grouping squared coefficients by support yields variance contributions under the standard independent-input orthonormal setting.

Abstract Reasoning¶

If the retained coefficients decay rapidly with degree and validation error falls, increasing order can produce rapid convergence for smooth responses. If the response has a discontinuity across input space, global polynomial coefficients often decay slowly and oscillations can appear; domain decomposition or a different surrogate may be preferable.

Knowledge Transfer¶

The exact abstraction transfers among stochastic ordinary and partial differential equations, structural mechanics, fluid flow, and model-based risk when input law, orthogonal basis, coefficient rule, and truncation remain explicit. Hermite, Legendre, Laguerre, Jacobi, and other families are variants selected by measure rather than different abstractions.

The portable core is Basis: independent generating elements support coordinate representation. Approximation and projection also transfer widely. The domain-specific residual is probabilistic orthogonality and uncertainty inference; without the random-input measure and coefficient-to-moment semantics, the object is a generic polynomial basis expansion.

Relationships to Other Abstractions¶

Current abstraction Polynomial Chaos Expansion Domain-specific

Parents (1) — more general patterns this builds on

Polynomial Chaos Expansion presupposes Basis Prime

Basis is the minimal parent because PCE represents a random response through coordinates in an orthogonal polynomial generating family.

Hierarchy path (1) — routes to 1 parentless root

Polynomial Chaos Expansion → Basis → Set and Membership

Neighborhood in Abstraction Space¶

Polynomial Chaos Expansion sits in a sparse region of the domain-specific corpus (70^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (1565 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08