---
title: Immersed Isogeometric Framework
url: https://www.emergentmind.com/topics/immersed-isogeometric-framework
type: topic
---

# Immersed Isogeometric Framework

Immersed isogeometric framework denotes a class of unfitted or fictitious-domain formulations in which the approximation space is built from spline technologies such as B-splines, NURBS, or truncated hierarchical B-splines on a simple background domain, while the physical geometry is embedded, trimmed, or otherwise imposed independently of the mesh. Across the literature, the common objective is to retain the high-order accuracy and smoothness of isogeometric analysis while avoiding body-fitted volumetric parameterizations, remeshing, or strict patch conformity. The framework appears in several concrete realizations: finite-cell-style immersed IGA, interpolation-based immersogeometric analysis with foreground extraction, generalized Heaviside-enriched XIGA for multi-material problems, scan-based and point-cloud-based pipelines, boundary-conformal hybrid variants, and specialized formulations for fluids, shells, magnetostatics, and topology optimization [2208.14994] [2209.06892] [2509.09236].

## 1. Core definition and methodological position

Immersed isogeometric analysis is formulated on a background spline space defined over a simple computational container, with the physical domain $\Omega$ embedded into that container. In the scan-based flow literature, the approach is described as a fictitious-domain method that keeps the high-order, smooth spline basis of IGA while avoiding a body-fitted volumetric parameterization of complex geometries. Boundary conditions on the embedded boundary $\Gamma$ are imposed weakly, typically by Nitsche’s method, and cut elements are integrated only over their physical portions [2208.14994].

This construction differs from standard fitted IGA, where the geometry is represented by boundary-fitted multi-patch spline parameterizations and the analysis space conforms to $\Omega$. It also differs from conventional immersed boundary methods, because the weak form remains Galerkin-consistent rather than relying on force spreading by delta functions. Relative to finite cell and CutFEM methodologies, immersed IGA shares the unfitted background-mesh philosophy, but specializes it to spline-based spaces and, in several variants, to higher continuity and CAD-compatible geometry handling [2208.14994] [2107.09024].

Several strands of the literature generalize this baseline idea rather than replacing it. Interpolation-based formulations keep the background space on an unfitted mesh but assemble all integrals on a geometry-conforming foreground Lagrange mesh. Boundary-conformal hybrids add a conformal layer near selected boundaries or interfaces while retaining an immersed background in the rest of the domain. In topology optimization, a level-set method combined with an immersed isogeometric framework allows seamless geometry updates without remeshing, and topological derivatives drive topology changes without predefining holes [2209.06892] [2011.01622] [2509.09236].

## 2. Approximation spaces, geometry representations, and enrichment

The approximation core is usually a spline space on an unfitted background mesh. For tensor-product B-splines, the one-dimensional basis is generated by the Cox–de Boor recursion, and multivariate basis functions are formed by tensor products. In hierarchical settings, truncated hierarchical B-splines are built from nested refinement regions and nested spline spaces, preserving partition of unity and locality while enabling local refinement for individual fields or near interfaces [2212.00882].

A characteristic interpolation-based construction introduces a background space
$$
V_b=\operatorname{span}\{N_i\}
$$
and a foreground Lagrange space
$$
V_f=\operatorname{span}\{\phi_j\}.
$$
The nodal interpolation operator is
$$
I_h w=\sum_{j=1}^{\nu} w(x_j)\,\phi_j,
$$
so that the interpolated background basis functions are
$$
\widehat N_i:=I_h N_i=\sum_{j=1}^{\nu} N_i(x_j)\,\phi_j.
$$
With extraction matrix entries $C_{ji}=N_i(x_j)$, the background basis is represented approximately as $B(x)\approx C^T L(x)$, and the immersed system is assembled as
$$
K=C^TAC,\qquad F=C^TB.
$$
This generalized extraction viewpoint is central to non-invasive reuse of standard finite-element assembly routines on foreground meshes [2209.06892].

Geometry is represented in several ways. Level-set descriptions are common in multi-material and scan-based settings. In scan-based immersed IGA, grayscale or binary voxel data are converted into a smooth spline level set
$$
f(x)=\sum_i N_{i,k}(x)\,a_i,\qquad
a_i=\frac{\int_{\Omega_{\mathrm{img}}}N_{i,k}(y)\,g(y)\,dy}{\int_{\Omega_{\mathrm{img}}}N_{i,k}(y)\,dy},
$$
and the physical domain is defined as a super-level set. Local THB-spline refinement is then used to repair topology changes induced by smoothing, with anomaly detection based on the Euler characteristic [2106.12347]. In other settings, the geometry is given by trimmed spline patches, CAD B-reps, point clouds, or multiple level sets used to define up to $2^n$ material phases through a phase map [2202.05697] [2210.13737].

For multi-material problems, enrichment is required to represent discontinuities or disconnected inclusions within the support of a smooth spline basis. A recurring construction is the generalized Heaviside enrichment
$$
u_h(x)=\sum_i\sum_\epsilon \psi_i^\epsilon(x)\,N_i(x)\,d_i^\epsilon,
$$
where $\psi_i^\epsilon$ is the indicator of the $\epsilon$-th materially connected subdomain inside $\operatorname{supp}(N_i)$. In XIGA and related formulations, enrichment is organized per basis-function support rather than per global material label, specifically to avoid spurious coupling and artificial stiffening when a single support intersects multiple disconnected regions of the same material [2501.17853] [2202.05697] [2402.15937].

Boundary-conformal variants modify the geometry handling rather than the approximation principle. In the immersed boundary-conformal method, a conformal layer is extruded from a boundary representation and coupled to a background spline patch through Nitsche’s method. In magnetostatics, two union variants similarly use independent non-conformal patches, with or without conformal layers, to improve interface resolution while reducing the number of patches relative to conventional conformal multi-patch analysis [2011.01622] [2604.07155].

## 3. Weak forms, boundary conditions, interfaces, and stabilization

Immersed isogeometric frameworks usually adopt standard PDE weak forms on the physical domain, augmented by weak boundary and interface terms. For Poisson’s equation, one representative symmetric Nitsche form is
$$
a_N(u,v)=\int_\Omega \kappa \nabla u\cdot\nabla v
-\int_{\partial\Omega_D}(\kappa \partial_n u)\,v
-\int_{\partial\Omega_D}(\kappa \partial_n v)\,u
+\int_{\partial\Omega_D}\frac{\gamma\kappa}{h}\,uv,
$$
with analogous load terms involving $f$ and the Dirichlet datum $g$. Non-symmetric variants are also used [2209.06892].

For linear elasticity, immersed formulations commonly use
$$
a(u,v)=\int_\Omega \varepsilon(u):C:\varepsilon(v),
$$
supplemented by Nitsche-type terms on immersed Dirichlet boundaries, symmetry boundaries, or material interfaces. In multi-material settings, interface conditions are written with jumps and weighted averages. For example, in enriched THB-spline XIGA, interface terms enforce continuity of displacement or temperature and equilibrium of traction or heat flux by combining consistency terms and penalties on $\Gamma^{m,n}_e$ [2212.00882]. In the interpolation-based multi-material framework, the interface residual for heat conduction is written as
$$
\mathcal R_T^I=
\sum_{i=1}^n\sum_{j=i+1}^n
\left[
-\int_{\Gamma_{ij}} [\![T^h]\!]\,\{\kappa\nabla\theta^h\cdot n\}\,d\Gamma
-\int_{\Gamma_{ij}} [\![\theta^h]\!]\,\{\kappa\nabla T^h\cdot n\}\,d\Gamma
+\int_{\Gamma_{ij}}\gamma_T^{ij}[\![T^h]\!][\![\theta^h]\!]\,d\Gamma
\right],
$$
with analogous expressions for elasticity [2402.15937].

More specialized immersed formulations extend the same pattern. For magnetostatics in the scalar potential formulation, Nitsche coupling across non-conformal interfaces uses
$$
a_n(u,v)=
-\int_{\Gamma_{LR}} \{\nu \nabla u\cdot n\}_{\gamma_2}\,[\![v]\!]\,d\Gamma
-\gamma_1\int_{\Gamma_{LR}} \{\nu \nabla v\cdot n\}_{\gamma_2}\,[\![u]\!]\,d\Gamma
+\beta\kappa\int_{\Gamma_{LR}}\nu_{\max}[\![u]\!][\![v]\!]\,d\Gamma,
$$
with $\gamma_1=1$, $\gamma_2=1$, and master-side flux evaluation chosen to avoid trimming-induced instabilities [2604.07155]. In the immersed boundary-conformal method for linear elliptic problems, coupling between the conformal layer and the background region is also enforced by Nitsche, but the flux is taken one-sided from the conformal layer, which removes the need for extra stabilization on the trimmed side in the two-domain setting [2011.01622].

Stabilization is indispensable when the physical boundary cuts basis supports into arbitrarily small fractions. One family of remedies is face-oriented ghost stabilization. In XIGA, the ghost term is
$$
R_G(u_h,v_h)=
\sum_{F\in\mathcal F_G}
\sum_{u=1}^{n_u(E^+)}
\sum_{v\in\mathcal U_{u,E^-}}
\sum_{k=1}^{p}
\int_F \gamma_G^k\,
[\![\partial_n^k u_h]\!]\cdot[\![\partial_n^k v_h]\!]\,ds,
$$
restricted to same-material subphase pairs across ghost faces [2501.17853]. For locally refined THB-spline multi-material analysis, related face-oriented ghost penalties act on the highest-order normal-derivative jumps across cut-adjacent faces [2212.00882].

For binary-fluid flow governed by the Navier–Stokes–Cahn–Hilliard equations, the immersed framework combines generalized Navier-slip, Nitsche enforcement of convective impermeability, and both skeleton and ghost penalties. Representative penalty terms are
$$
j_{\mathrm{skel}}^p(p_h,q_h)
=
\sum_F \gamma_{\mathrm{sk}}^p\frac{h}{\eta}
\int_F
[\![\nabla p_h\cdot n_F]\!]\,
[\![\nabla q_h\cdot n_F]\!]\,ds
$$
and
$$
g^{u}(u_h,v_h)
=
\sum_{F\in\mathcal F_{\mathrm{cut}}}
\gamma_g^u\,\eta\,h
\int_F
[\![\nabla u_h\cdot n_F]\!]:
[\![\nabla v_h\cdot n_F]\!]\,ds,
$$
with corresponding terms for the phase field and chemical potential [2305.19237].

## 4. Integration, extraction, preprocessing, and software architecture

A defining practical difficulty of immersed IGA is numerical integration on cut cells. Classical finite-cell-style formulations use recursive quadtree or octree subdivision, tessellation, or adaptive sub-cell quadrature on $K\cap\Omega$. In scan-based and flow-oriented implementations, cut elements are recursively subdivided until geometry criteria or volume-fraction criteria are satisfied, and standard Gauss rules are applied on the resulting subcells [2208.14994] [1911.11519].

Interpolation-based immersogeometric analysis shifts this burden by introducing a geometry-conforming foreground mesh for all integrals. Background basis functions are sampled at foreground nodes to build extraction matrices; the foreground code assembles conventional matrices $A$ and vectors $B$; and the immersed system is obtained through the congruence transformation $K=C^TAC$, $F=C^TB$. The same paper emphasizes that this permits non-invasive reuse of existing finite-element software, and the reported prototype is implemented on top of FEniCS, DOLFIN, UFL, and PETSc [2209.06892].

A more algorithmically elaborate realization appears in the preprocessing framework for enriched immersed finite element and isogeometric analysis. There, the input is a background mesh together with one or more geometries, and the output is a set of “clusters” that package background elements with custom bulk, side, interface, and ghost quadrature data. Cut background elements are tessellated into a conformal foreground mesh; material labels are assigned during tessellation; subphase graphs are built; background elements are “unzipped” to encode generalized Heaviside enrichment through their IEN arrays; and cluster pairs are generated for interface and ghost terms. The implementation reports approximately linear time and memory efficiency with problem size; for a large problem of 64 unit blocks, baseline times are $t_0\approx 349.5\,\mathrm{s}$ for tessellation, $439.8\,\mathrm{s}$ for enrichment, and $24.71\,\mathrm{s}$ for ghost-facet generation, with memory consumption in the $60$–$80\,\mathrm{GB}$ range [2501.17853].

A contrasting route is quadrature-free immersed IGA. There, element-local $L^2$ projections transform coefficients and loads into polynomial Bernstein form, and successive applications of the divergence theorem reduce cut-cell volume integrals to surface integrals and then to line integrals over spline trim curves. The final one-dimensional polynomial integrals are evaluated analytically in the Bernstein basis:
$$
\int_0^1 \sum_{i=0}^{P} B_i^P(t)\,\tilde t_i\,dt
=
\frac{1}{P+1}\sum_{i=0}^{P}\tilde t_i.
$$
The reported experiments recover optimal $H^1$ and $L^2$ convergence rates for elliptic problems and demonstrate applicability to CAD models of industrial complexity [2107.09024].

Other variants tailor integration to nonstandard geometry input. For point clouds, surface integration required for weak boundary conditions is approximated by pointwise quadrature
$$
\int_\Gamma \varphi(x)\,d\Gamma
\approx
\sum_{i=1}^{|\mathcal P|} a_i\,\varphi(p_i),
$$
where $a_i$ is a geodesic Voronoi area associated with point $p_i$. Inside–outside classification is performed through generalized winding numbers,
$$
w(q,\Gamma)=
\sum_{i=1}^{|\mathcal P|}
a_i\,
\frac{(p_i-q)\cdot n_i}{\|p_i-q\|^3},
$$
with a threshold of $0.5$ used for physical-domain classification [2210.13737].

Adaptive integration has also been formalized by comparing stiffness, load, and Nitsche contributions at successive octree levels. A representative local estimator is the Frobenius norm
$$
\eta_K^2(\ell)
=
\sum_{i,j\in I(K)}
\left|
(K_K^{(\ell+1)})_{ij}-(K_K^{(\ell)})_{ij}
\right|^2,
$$
which is used to drive local refinement of quadrature rather than uniform cut-cell subdivision [1911.11519].

## 5. Major variants and application domains

The framework has been specialized to a wide range of application classes. In solid and structural mechanics, interpolation-based immersed methods reproduce optimal convergence for Poisson, biharmonic, and linear elasticity problems, and have been applied to Kirchhoff–Love shells on trimmed patches. In the shell benchmarks, interpolation-based immersed analysis reproduces a reference displacement on a trimmed square and converges on a curved bending-tab geometry under pseudo-time continuation with damping [2209.06892].

In multi-material and multi-physics analysis, enriched THB-spline formulations support independently refined fields and generalized Heaviside enrichment. The union-mesh THB framework has been demonstrated on two- and three-dimensional linear elastic and thermo-elastic problems, including a polycrystalline microstructure with approximately $1.06\times 10^8$ union-mesh elements, where mesh generation was reported under approximately $50\,\mathrm{s}$ on four 24-core nodes [2212.00882]. The interpolation-based multi-physics extension employs a single foreground mesh for all fields while allowing distinct background spline spaces per field; in an image-based thermo-mechanical composite example, the interpolation-based method used approximately $48{,}191$ temperature DOFs and $60{,}729$ displacement DOFs, whereas a uniformly refined boundary-fitted FEM comparison used approximately $5{,}027{,}580$ and $10{,}055{,}160$ DOFs, respectively [2402.15937].

Scan-based immersed IGA constructs smooth analysis-suitable geometry directly from voxel or grayscale data. A topology-preserving variant uses THB-spline refinement guided by a moving-window Euler-characteristic detector to prevent smoothing-induced loss of small topological features, and then feeds the resulting geometry into a finite-cell-style immersed solver for elasticity or Stokes flow [2106.12347]. A related review of scan-based flow analysis describes spline-based segmentation, stabilized equal-order Stokes discretization, adaptive integration, and THB refinement for small unresolved features in patient-specific geometries [2208.14994].

Point-cloud-based immersogeometric analysis dispenses even with watertight CAD surfaces. The method estimates normals from local PCA or jet fitting, constructs geodesic Voronoi areas for surface quadrature, and solves incompressible, compressible, and thermal flow problems directly on point-cloud-defined objects. The reported industrial-scale example uses a construction-machinery point cloud with more than 12 million points [2210.13737].

Electromagnetic applications motivate non-conformal patch unions rather than purely fully immersed spaces. In two-dimensional magnetostatics, three strategies are assessed: a fully immersed approach, union with non-conformal patches, and union with conformal layers. The numerical results show that the union methods achieve highly accurate solutions, while the fully immersed approach struggles with discontinuities in field gradients across material interfaces. In the horseshoe-magnet example, a 30-patch conformal reference is replaced by union variants using about five to seven patches, with the union-with-conformal-layers configuration improving corner singularities relative to the union-with-non-conformal-patches configuration [2604.07155].

Fluid and fluid–structure interaction problems have likewise produced specialized immersed frameworks. For Navier–Stokes–Cahn–Hilliard flows, optimal-regularity B-splines, Nitsche boundary enforcement, and skeleton-ghost stabilization are combined to model binary-fluid flow through complex porous geometries, including breakup and coalescence [2305.19237]. For air-blast fluid–structure interaction, a monolithic immersed coupling of a background NURBS fluid discretization and a foreground peridynamic solid uses background spline functions together with RKPM-based peridynamic correspondence to handle large deformation, damage, and fragmentation without explicit interface tracking [2108.11265]. A distinct BEM-based line of work uses NURBS boundaries together with immersed cell meshes for nonlinear underground-excavation analysis under heterogeneous and anisotropic ground conditions [2211.16827].

Topology optimization is an additional application rather than a separate geometry technology. In the cited 2025 paper, a level-set method coupled with an immersed isogeometric framework permits seamless geometry updates without remeshing, and topological derivatives create holes without prescribing them in the initial design. The reported numerical examples indicate that higher-degree basis functions improve the accuracy of the state approximation, while linear basis functions remain sufficient for the level-set representation [2509.09236].

## 6. Accuracy, conditioning, adaptivity, and recurring misconceptions

A consistent result across immersed isogeometric formulations is that optimal approximation properties are attainable, but only after geometry treatment, stabilization, and conditioning control are addressed explicitly. In interpolation-based analysis, the background-fitted theory yields
$$
\|u-u_h\|_{H^1(\Omega)}
\le
C\,h^{\hat k}\,
\|u\|_{H^{\hat k+1}(\Omega)},
\qquad
\hat k=\min\{k,\kappa\},
$$
and numerical evidence shows optimal rates also for background-unfitted foreground meshes [2209.06892]. Quadrature-free immersed IGA reports optimal $H^1$ order $p$ and $L^2$ order $p+1$ for $p=1$–$4$ in both two and three dimensions, until CAD-kernel tolerances dominate on very fine meshes [2107.09024]. XIGA obtains optimal rates for planar interfaces in multi-material heat transfer and elasticity, whereas curved interfaces require sufficiently refined geometric integration to recover the expected rates [2202.05697].

A common misconception is that immersing the geometry removes meshing difficulties without introducing new numerical ones. The opposite conclusion is repeated across the literature: tiny intersections and sliver cuts can lead to severe ill-conditioning, and smooth spline supports exacerbate near-linear dependencies when their physical support becomes small. In scan-based, CutFEM-related, and interpolation-based papers, this is the primary reason for ghost penalties, basis removal, agglomeration, or robust preconditioning [2202.08763] [2209.06892]. In connectivity-based additive Schwarz preconditioning, condition numbers of immersed systems are shown to grow like a power of the minimum cut fraction without preconditioning, whereas the CbAS preconditioner makes the conditioning essentially independent of the cut configuration for Poisson, convection–diffusion, Stokes, and Navier–Stokes problems [1708.03519]. Geometric multigrid with Schwarz-type smoothers likewise achieves mesh-independent and cut-element-independent convergence rates for immersed finite elements and immersed IGA, including THB-spline discretizations, at computational cost that scales linearly with the number of degrees of freedom [1903.10977].

Another misconception is that simply increasing spline degree always improves immersed accuracy. The scan-based topology-preserving literature shows that larger spline degree increases smoothing width and can suppress voxel-scale features, thereby changing topology unless local THB refinement is introduced [2106.12347]. The magnetostatics study shows that high continuity across fully immersed material interfaces can be too smooth to resolve jumps in field gradients, and that union strategies or conformal layers may be preferable in multi-material settings [2604.07155]. The boundary-conformal IBCM work makes a related point in a different way: an immersed background supplemented by a conformal layer can provide higher accuracy per degree of freedom, intuitive control of near-boundary resolution, and strong Dirichlet imposition compared with a purely trimming-based discretization [2011.01622].

Adaptivity addresses both approximation and integration errors. Residual-based THB refinement for Laplace and Stokes problems augments standard element indicators with boundary and stabilization contributions tailored to the immersed setting, and is integrated into scan-based workflows for automated, error-controlled analysis [2202.08763]. Adaptive octree integration, derived from Strang-type consistency arguments, refines quadrature only where successive quadrature levels disagree materially, thereby reducing cut-cell integration cost, especially in three dimensions [1911.11519].

The most stable synthesis emerging from these papers is therefore not a single canonical algorithm, but a modular architecture: spline-based approximation on simple background domains; geometry representation by trimming, level sets, scans, point clouds, or patch unions; weak enforcement of immersed boundary and interface conditions; explicit stabilization of cut-induced pathologies; and, when needed, adaptive refinement or specialized integration. This does not eliminate the classical trade-off between geometric fidelity, stability, and computational cost, but it does relocate that trade-off from mesh generation to analysis design in a way that remains compatible with high-order spline technology and complex geometries [2208.14994] [2202.05697] [2305.19237].

Source: https://www.emergentmind.com/topics/immersed-isogeometric-framework