Geometry Independent Field Approximation (GIFT)

• Geometry Independent Field Approximation (GIFT) is an advanced numerical method that decouples the exact CAD geometry from the solution field, allowing independent and flexible local refinement.
• It supports diverse spline technologies such as PHT-splines, T-splines, and hierarchical structures, optimizing both geometric fidelity and computational efficiency.
• GIFT enhances multi-physics simulations by reducing degrees of freedom and CPU time through adaptive local h-refinement and effective error control strategies.

Geometry Independent Field approximaTion (GIFT) is a generalization of isogeometric analysis (IGA) that removes the requirement for the field approximation space to coincide with the geometry parameterization space. While classical IGA integrates design and simulation through a common spline representation (typically NURBS), GIFT preserves the exact CAD geometry but enables independent and flexible choice of the analysis basis for unknown fields, supporting local $h$-adaptivity and a broader selection of spline technologies—including PHT-splines, T-splines, and other hierarchical structures. This decoupling optimizes both geometric fidelity and computational efficiency, particularly in applications demanding localized mesh refinement and complex multi-physics coupling (Atroshchenko et al., 2017, Marussig et al., 2014, Yu et al., 2018, Videla et al., 2022).

1. Motivation and Conceptual Framework

The central objective of GIFT is to weaken the tight coupling between CAD geometry and solution field approximation inherent to standard IGA. In classical IGA, geometry parameterization and field approximation use the same spline (often NURBS), leading to:

• Preservation of geometric exactness throughout the analysis.
• No natural mechanism for local refinement due to the tensor-product structure of the mesh.
• Unnecessary refinement or elevation in geometry when only the solution space requires enrichment.
• Repeated geometric re-parameterization and associated CAD processing if locally-refined splines are desired for the solution.

GIFT preserves the fixed, exact CAD geometry description but enables an independent choice of solution space, allowing for localized mesh refinement and adaptive analysis without geometry modification. The geometry space $\mathcal G$ is typically a globally-tensorized NURBS basis extracted from the native CAD model, while the field approximation space $\mathcal F$ can be any locally-refinable spline (PHT-splines, T-splines, LR-splines, THB-splines) defined on the same parametric domain but without any requirement of being nested within $\mathcal G$ (Atroshchenko et al., 2017).

2. Mathematical Formulation

2.1 Geometry and Field Spaces

Let $\hat\Omega\subset\mathbb{R}^d$ be the reference (parametric) domain and $\Omega\subset\mathbb{R}^d$ the physical domain. GIFT employs:

• Geometry Mapping: A fixed map $$\boldsymbol F(\boldsymbol\xi) = \sum_{I\in\mathcal{I}} N_I(\boldsymbol\xi)\mathbf{P}_I$$, where $N_I$ are geometry basis functions (NURBS), and $\mathbf{P}_I$ are the CAD control points.
• Field Approximation: The solution is approximated as $$u_h(x) = \sum_{J\in\mathcal{J}} u_J M_J(\boldsymbol\xi)$$ with $M_J$ from the chosen solution basis (possibly PHT-splines or T-splines) and $(\boldsymbol\xi = \boldsymbol F^{-1}(x))$.

Derivatives in the physical domain are mapped via the chain rule with the stationary geometry Jacobian $J_F(\boldsymbol\xi)$, hence physical and computational domains remain congruent irrespective of analysis refinement (Atroshchenko et al., 2017).

2.2 Galerkin and Boundary Element Discretization

In PDE problems, the discrete variational formulation is assembled using the geometry mapping for all geometry-related quantities (Jacobians, boundary normals, integration elements), but solution-space refinement affects only the field basis in the weak form. In boundary element contexts, three spaces are typically kept independent: the geometry mapping (NURBS), the displacement field, and the traction field, with traction often discretized using discontinuous polynomial bases to handle singularities at geometric edges (Marussig et al., 2014).

3. Adaptive Local Refinement and Error Control

A principal advantage of GIFT is its compatibility with adaptive local $h$-refinement, as the mesh and basis for the solution field can be locally enriched without altering the geometry:

• Hierarchical PHT-splines: Common in GIFT implementations, they store analysis meshes as quadtrees (2D) or octrees (3D), support localized $h$-refinement by splitting each cell into $2^d$ children, and update control points and local knot vectors via Bézier extraction.
• A Posteriori Error Estimation: Adaptive refinement is guided by hierarchical or recovery-based error indicators; two-mesh strategies are standard, comparing solutions on a coarse mesh and its refined counterpart to estimate local and global errors. Element marking is typically carried out via Dörfler-marking to ensure control of global error norms (Yu et al., 2018, Videla et al., 2022).

A summary algorithmic workflow is as follows (Atroshchenko et al., 2017):

1. Extract NURBS geometry mapping from CAD.
2. Select and initialize a solution spline space over the parametric domain.
3. Assemble and solve the discrete system.
4. Estimate local errors; mark and refine the solution mesh accordingly.
5. Iterate until desired accuracy is obtained.

4. Patch Test, Convergence, and Theoretical Properties

In GIFT, the geometry and field approximation spaces are generally non-nested, so the classical patch test (exact reproduction of linear solutions) is often not satisfied. However, numerical experiments and isogeometric error theory indicate:

• Optimal $H^1$-norm convergence rates, i.e., $\|u-u_h\|_{H^1(\Omega)} = O(h^p)$ for field basis of degree $p$ and smooth underlying geometry and solution.
• Patch test success is neither necessary nor sufficient for convergence; theoretical and computational studies confirm that the decoupling of spaces does not compromise asymptotic analysis rates when the geometry remains regular (Atroshchenko et al., 2017, Marussig et al., 2014).

In the boundary element method (BEM) context, GIFT enables the independent discretization of the geometry, displacement, and traction fields, maintaining isogeometric exactness and supporting discontinuities and singular behaviors where required. Convergence in $L^2$ or $L^\infty$ norms matches predictions for both primal and dual solutions (Marussig et al., 2014).

5. Numerical Performance and Benchmark Studies

GIFT demonstrates favorable numerical properties and efficiency gains in a variety of benchmark domains:

• Poisson, elasticity, and vibro-acoustics: Benchmarks on quarter-annulus, wedge, cylinder, plate-with-hole, sphere, and circular plate domains confirm optimal $h$-convergence rates for both uniform and adaptively refined PHT-spline or T-spline solution bases (Atroshchenko et al., 2017, Yu et al., 2018).
• Time-harmonic acoustics and shape optimization: Adaptive GIFT with NURBS geometry and PHT-spline solution achieves up to 90% reduction in degrees of freedom and CPU time due to concentrated refinement in high-error regions. Examples include plane-wave scattering by cylinders, acoustic horn design, and sound barrier optimization (Videla et al., 2022).
• Boundary element method efficiency: The adoption of hierarchical matrix (H-matrix) approximation for dense system blocks enables storage and computational complexity reduction from $\mathcal{O}(n^2)$ to $\mathcal{O}(n\log n)$ without loss in accuracy. Large-run simulations (industrial crank-shaft, cantilever beam) demonstrate substantial storage compression and accelerated assembly (Marussig et al., 2014).

Representative results include recovery of modal frequencies in elastodynamics to within 0.05% of reference IGA for moderate DOFs, as well as near-optimal storage and accuracy tradeoffs in geometry-rich industrial simulations.

6. Implementation Aspects and Extensions

GIFT frameworks have been implemented in specialized open-source packages (notably IGAPACK) with the following features:

• Quadtree- or octree-based management of hierarchical PHT-spline meshes.
• Bézier extraction to update analysis basis under local refinement.
• Efficient management of control variables, boundary conditions via dropping corresponding DOFs, and multi-patch NURBS interface coupling via $C^0$ identification.
• Use of direct sparse solvers (e.g., PARDISO) for the field equations, and standard optimization solvers for design-variable iterations in shape optimization.
• Compatibility with partition of unity methods for wave enrichment, and extensibility to 3D domains (Videla et al., 2022).

Limitations include increased complexity for geometric inversion (mapping from physical to parametric space) in highly-distorted or multi-patch domains, and the need to preserve mesh quality under repeated field-space refinement.

7. Applications, Limitations, and Outlook

GIFT is applicable to any simulation scenario benefiting from decoupled geometric and analysis representations. Major application areas include structural mechanics, elastodynamics, time-harmonic acoustics, and shape optimization with design-space defined at the NURBS control points while the solution-space is adaptively enriched via PHT-splines.

Limitations concern situations requiring adaptive geometry updates (with the fixed geometry mapping assumption less suitable), and implementation overhead associated with managing independent geometric and field bases. Potential extension areas include full 3D adaptivity, integration with design-optimization pipelines, and the use of exact shape derivatives for enhanced optimization efficiency (Atroshchenko et al., 2017, Videla et al., 2022).

---

Principal references: (Atroshchenko et al., 2017, Marussig et al., 2014, Yu et al., 2018, Videla et al., 2022).