Isogeometric Analysis: CAD-Integrated Simulation

Updated 2 February 2026

Isogeometric Analysis (IGA) is a computational method that uses spline basis functions like B-splines and NURBS to integrate CAD and simulation, achieving exact geometry representation and high continuity.
IGA employs Galerkin discretization with tensor-product spline spaces to solve PDEs efficiently, resulting in superior accuracy per computational degree of freedom compared to traditional FEM.
Adaptive refinement techniques, including multi-patch methods and locally refinable splines, enable optimal error control and high solver performance for complex geometries and multiphysics applications.

Isogeometric Analysis (IGA) is a computational methodology for the Galerkin discretization of partial differential equations (PDEs) that utilizes spline-based basis functions—primarily B-splines and non-uniform rational B-splines (NURBS)—to provide a unified representation of both geometry and solution spaces. The core philosophy of IGA is to integrate Computer-Aided Design (CAD) and numerical simulation by employing the same mathematical objects for both, thereby achieving exact geometry, high-order smoothness, and superior accuracy per computational degree of freedom compared to traditional finite element methods (FEM). Since its introduction by Hughes, Cottrell, and Bazilevs in 2005, IGA has transformed simulation workflows in computational structural mechanics, electromagnetics, acoustics, electronic structure, and geometric PDEs.

1. Spline Spaces and Exact Geometry Representation

The foundation of IGA is the use of B-splines and NURBS as the basis for both geometry mapping and solution approximation. Given a parametric domain, typically the unit interval or unit cube, a knot vector $\Xi$ is specified, along with polynomial degree $p$ . Univariate B-spline basis functions $N_{i,p}(\xi)$ are constructed recursively via the Cox–de Boor formulas. For multi-dimensional problems, tensor-product B-spline or NURBS bases are formed: $R_{i,j,k}(\xi, \eta, \zeta) = \frac{N_{i,p}(\xi)M_{j,q}(\eta)L_{k,r}(\zeta)w_{i,j,k}}{\sum_{\hat{i},\hat{j},\hat{k}}N_{\hat{i},p}(\xi)M_{\hat{j},q}(\eta)L_{\hat{k},r}(\zeta)w_{\hat{i},\hat{j},\hat{k}}}$ where $w_{i,j,k}$ are the positive NURBS weights. The parameter-to-physical domain map is

$\mathbf{x}(\xi, \eta, \zeta) = \sum_{i,j,k} R_{i,j,k}(\xi, \eta, \zeta) \mathbf{P}_{i,j,k}$

with $\{\mathbf{P}_{i,j,k}\}$ the control points (defining both geometry and mesh). This approach ensures that the computational mesh exactly represents CAD geometries, including conics and rational shapes, at all levels of mesh refinement (Nguyen et al., 2012).

Continuity across elements is $C^{p-1}$ for B-splines with simple knots, and knots of multiplicity $k$ reduce continuity to $C^{p-k}$ . High continuity is a key advantage for IGA, especially for higher-order PDEs (Wilhelm et al., 2015).

2. Isogeometric Galerkin Discretization

IGA formulates the Galerkin method where the trial and test spaces are spanned by spline functions mapped from the parametric to the physical domain. For a model elliptic PDE: $-\nabla \cdot (k \nabla u) = f \quad \text{in } \Omega,$ the variational form is approximated by seeking $u_h$ in the isogeometric space $V_h$ such that: $a(u_h, v_h) = \int_{\Omega} k \nabla u_h \cdot \nabla v_h \, dx = \int_{\Omega} f v_h \, dx = l(v_h) \qquad \forall v_h \in V_h.$ The basis functions are constructed by pullback: $\varphi_i = B_i \circ F^{-1}$ where $F$ is the spline or NURBS map from the reference domain to $\Omega$ . Matrix assembly follows standard finite element procedures, with careful mapping of gradients via the Jacobian of $F$ (Riva et al., 2020, Nguyen et al., 2012, Bontinck et al., 2017).

3. Multi-Patch, Overlapping, and Domain Decomposition Methods

Realistic CAD models are generally specified by a collection of patches, each with a local NURBS parametrization. The global computational domain $\Omega = \bigcup_k \Omega_k$ is partitioned into non-overlapping or overlapping patches, each with independent basis functions and geometry maps (Langer et al., 2014, Kargaran et al., 2023).

Interface continuity between patches can be enforced in various ways:

Multi-Patch Matching: Enforce $C^0$ or higher continuity at interfaces; commonly for nontrivial geometries or multiphysics applications (Hofer et al., 2018, Bontinck et al., 2017).
Discontinuous Galerkin (dG) IGA: Weakly enforce transmission conditions via penalization, enabling arbitrarily complex multi-patch assemblies including non-matching meshes (Langer et al., 2014).
Overlapping or Nonmatching Meshes: Domain decomposition strategies such as Additive Schwarz (ASDDM) with overlapping nonmatching meshes allow for local refinement, Chimera-type zooming, and parallelism, enabling efficient coupling of complex patch assemblies and local adaptivity (Bercovier et al., 2015, Kargaran et al., 2023).

A key computational challenge is the design of scalable solvers robust with respect to the spline degree $p$ and mesh size $h$ ; techniques such as subspace-corrected mass smoothers, tailored Schwarz-type preconditioners, and $h$ -multigrid are critical for optimal solver performance for high-order, large-scale problems (Hofer et al., 2018, Riva et al., 2020).

While classical NURBS are limited to global, tensor-product refinement (global $h$ - or $p$ -refinement), modern IGA leverages:

Multi-Patch Refinement: Multi-resolution IGA adapts grid sizes on patches independently, yielding local adaptivity without sacrificing the tensor-product structure essential for efficient assembly and preconditioning (Tyoler et al., 2024). Nested-trace constraints between patches ensure global conformity while enabling $p$ -robust and $h$ -robust error estimates.
Locally Refinable Splines: T-splines, LR-splines, truncated hierarchical B-splines, and PHT-splines support local $h$ -refinement; however, implementing them efficiently in multi-patch settings is often complex (Atroshchenko et al., 2017).
Geometry Independent Field approximaTion (GIFT): GIFT generalizes IGA by decoupling the geometry and solution approximation spaces—keeping the CAD-exact NURBS map fixed while allowing locally refined solution spaces (e.g., T-splines, hierarchical splines), thus overcoming NURBS' limitations in adaptivity without breaking the CAD–analysis link (Atroshchenko et al., 2017).

These strategies yield adaptivity that recovers optimal convergence rates in regions of low regularity and for singularity-dominated problems, as confirmed in benchmarks on L-shaped domains, magnetostatics, and other settings (Tyoler et al., 2024, Langer et al., 2014, Kargaran et al., 2023).

5. Specialized IGA Constructions: Hermite Interpolation and Scaffolds

Element-based B-spline framework enables direct construction of basis spaces using Hermite interpolation at nodes (matching functionals and derivatives), allowing element-level assembly, superconvergent error properties, and explicit imposition of boundary conditions through function and derivative matching—much like standard finite elements but with higher smoothness (Yang et al., 28 Jan 2026).

For complex three-dimensional or biomedical geometries, the solid NURBS conforming scaffolding methodology constructs multi-compartmental, conforming NURBS parametrizations by fitting the domain to a skeleton graph, yielding highly smooth (almost $G^2$ ) control lattices for efficient, accurate volumetric IGA (Moriconi et al., 2022).

6. Applications and Benchmark Performance

IGA has been applied across a broad spectrum of computational mechanics and scientific computing tasks:

Electromagnetics: Construction of curl- and div-conforming spline spaces compatible with de Rham complexes for Maxwell's equations, delivering spectrally correct behavior and improved accuracy per degree of freedom compared to low-order FEM (Bontinck et al., 2017).
Acoustics and Wave Propagation: IGA with infinite elements enables high-fidelity, low-dispersion discretizations for unbounded domain scattering and acoustic-structure interaction, achieving orders-of-magnitude gains in accuracy per DOF (Venås et al., 2022).
Structural Mechanics and Composites: NURBS-based IGA for complex curvilinear anisotropy, optimization of fiber layouts, and thin-walled structures demonstrates substantial reductions in stress concentration and improved time-to-solution relative to standard FEM (Suzuki et al., 2020).
Electronic Structure: Use of globally $C^2$ -continuous splines for Kohn–Sham density-functional theory ensures correct evaluation of Hellmann–Feynman forces and spectral convergence in electronic eigenproblems (Cimrman et al., 2016).
Surface and Geometric PDEs: IGA-based smoothing (IsoGeometric Smoothing—IGS) solves surface-based fourth-order PDEs for denoising, aerodynamic shape fitting, and surface force prediction in industrial applications (Wilhelm et al., 2015).

Numerical results consistently show that IGA attains optimal or even superconvergent rates, often requiring 30–80% fewer degrees of freedom for a given accuracy target compared to equivalent FEM (Bontinck et al., 2017, Suzuki et al., 2020, Bontinck et al., 2017, Cimrman et al., 2016). Efficient high-order solver strategies—parallel multigrid, two-level Schwarz, additive Schwarz preconditioners—further enhance computational scalability and $p$ -robustness (Hofer et al., 2018, Riva et al., 2020).

7. Implementation, Software Libraries, and Best Practices

IGA has inspired robust, extensible open-source and commercial software:

GeoPDEs, G+Smo, igatools, PetIGA, IGA-BEM: Toolboxes for spline evaluation, assembly, preconditioning (including multipatch, hierarchical refinement, curl/div-conforming spaces), and visualization (Bontinck et al., 2017, Langer et al., 2014).
Bézier Extraction: Universal interface between NURBS/T-spline representations and standard FEM assembly, permitting IGA integration into existing FE codes and efficient handling of generalized spline bases (Nguyen et al., 2012, Cimrman et al., 2016).
Multi-core, MPI, Domain Decomposition: Patch-wise parallelization and domain decomposition enable efficient large-scale computations, exploiting data locality and minimizing communication for multiphysics and multi-patch domains (Hofer et al., 2018, Bercovier et al., 2015).

Best practices include maintaining maximal inter-patch continuity, using tensor-product refinement on patches for efficient assembly, exploiting multi-patch decomposition for localized adaptivity, and leveraging existing tensor-product structure for fast preconditioners and solvers. For infinite or unbounded domains, coupling IGA with infinite elements (BGU/PGU) achieves accurate radiation enforcement with reduced DOFs (Venås et al., 2022).

References (arXiv paper IDs):

(Riva et al., 2020, Wilhelm et al., 2015, Hofer et al., 2018, Venås et al., 2022, Bontinck et al., 2017, Cimrman et al., 2016, Kargaran et al., 2023, Bontinck et al., 2017, Langer et al., 2014, Atroshchenko et al., 2017, Moriconi et al., 2022, Bercovier et al., 2015, Suzuki et al., 2020, Nguyen et al., 2012, Yang et al., 28 Jan 2026, Tyoler et al., 2024)