Cubic Hermite Quasi-Interpolation Operators

• Cubic Hermite quasi-interpolation operators are local, spline-based projectors that reproduce function values and first derivatives to yield C^1-smooth approximations.
• They ensure high-order accuracy with sharp error estimates and stability properties, making them effective for approximating functions and derivatives.
• These operators extend to multidimensional settings using tensor-product and Bernstein–Bézier constructions, facilitating adaptive and hierarchical mesh refinements in various applications.

Cubic Hermite quasi-interpolation operators are local, data-efficient spline-based projectors that reproduce cubic Hermite data (function values and first derivatives), achieving high-order approximation while preserving smoothness and computational locality. They form a central tool in numerical analysis, computational geometry, finite element methods, and scientific computing for constructing high-accuracy, low-cost, and stable approximations of functions and their derivatives in one and multiple dimensions, including adaptive and hierarchical mesh settings.

1. Theoretical Foundations and Operator Construction

Cubic Hermite quasi-interpolation operators are defined to interpolate or closely project data comprised of both function values and first derivatives at specified nodes, leading to $C^1$-smooth splines. The operator can be constructed in several mathematically equivalent forms, emphasizing either Hermite spline basis representations, B-spline refinements, or Bernstein–Bézier coordinates, depending on the application and mesh topology.

The core construction in 1D starts by selecting the cubic Hermite finite element space, e.g., on a reference interval $\hat I = [0,1]$, with basis functions $v_1(\hat x),\ldots,v_4(\hat x)$, associated to endpoint derivatives and values. For canonical Hermite interpolation, one defines nodal functionals $N_1(u) = u'(0)$, $N_2(u) = u'(1)$, $N_3(u) = u(1) - u(0)$, $N_4(u) = u(1) + u(0)$, and forms the interpolant $$\hat I_0 u(\hat x) = \sum_{i=1}^4 N_i(u)v_i(\hat x)$$, which satisfies $N_i(\hat I_0 u) = N_i(u)$ (Bonizzoni et al., 2020).

For quasi-interpolation, the nodal functionals are replaced by mollified, weighted analogs, e.g., for an interval $K=[y_\ell,y_r]$, one defines

$\hat I = [0,1]$0

where $\hat I = [0,1]$1 evaluates the original functional at nearby points, and $\hat I = [0,1]$2, $\hat I = [0,1]$3 are cutoff bump functions. This yields the local quasi-interpolation projector $\hat I = [0,1]$4 (Bonizzoni et al., 2020).

Alternative constructions employ B-spline and blossoming methodology, yielding explicit, normalized basis functions $\hat I = [0,1]$5 and local Hermite quasi-interpolants assembled as $\hat I = [0,1]$6, realizing non-negative, partition-of-unity, locally supported, $\hat I = [0,1]$7-continuous operators (Boushabi et al., 2024).

2. Cartesian and Triangular Mesh Extensions

Tensor-product extensions generalize the 1D cubic Hermite quasi-interpolants to higher dimensions. On Cartesian meshes of $\hat I = [0,1]$8, the multivariate quasi-interpolant is constructed via repeated tensor products, producing $\hat I = [0,1]$9-conforming splines with polynomial preservation, $v_1(\hat x),\ldots,v_4(\hat x)$0-boundedness, and commutation with exterior differentiation (Bonizzoni et al., 2020). For example, in 2D, the operator acts separately in the $v_1(\hat x),\ldots,v_4(\hat x)$1 and $v_1(\hat x),\ldots,v_4(\hat x)$2 directions: $v_1(\hat x),\ldots,v_4(\hat x)$3 with local Hermite data $v_1(\hat x),\ldots,v_4(\hat x)$4 collected on a regular tensor grid (Bracco et al., 2016, Bertolazzi et al., 2022).

For triangular meshes, particularly uniform three-direction (hexagonal) triangulations, the cubic Hermite quasi-interpolation operator is specified in Bernstein–Bézier form on each triangle. The spline $v_1(\hat x),\ldots,v_4(\hat x)$5 is expressed as a sum of Bernstein basis polynomials with Bézier coefficients computed as local linear combinations of point and gradient data at a hexagonal stencil of seven mesh points. The operator is constructed to guarantee exactness on polynomials up to degree two, $v_1(\hat x),\ldots,v_4(\hat x)$6 continuity across triangle edges, and explicit stability bounds (Barrera et al., 2024).

Dimensionality	Basis Representation	Mesh Type	Reference
1D	Hermite/B-spline	Uniform/interpol.	(Bonizzoni et al., 2020, Bertolazzi et al., 2022, Boushabi et al., 2024)
2D Cartesian	Tensor-product Hermite/B-spline	Rectangular	(Bonizzoni et al., 2020, Bracco et al., 2016, Bertolazzi et al., 2022)
2D Simplicial	Bernstein–Bézier	Triangular (hex)	(Barrera et al., 2024)
dD (general)	Tensor of 1D Hermite spaces	Cartesian	(Bonizzoni et al., 2020)

3. Algebraic Properties and Approximation Theory

Cubic Hermite quasi-interpolation operators are designed to satisfy polynomial reproduction up to a fixed degree (usually $v_1(\hat x),\ldots,v_4(\hat x)$7 or $v_1(\hat x),\ldots,v_4(\hat x)$8, depending on the setting), locality (data at or near each node determine coefficients on adjacent elements), and commutation with differentiation. For sufficiently smooth $v_1(\hat x),\ldots,v_4(\hat x)$9,

$N_1(u) = u'(0)$0

ensuring preservation of differential structure (Bonizzoni et al., 2020). The operators are stable in $N_1(u) = u'(0)$1 (and sup-norm for splines), with constants independent of the mesh step $N_1(u) = u'(0)$2 (Bonizzoni et al., 2020, Bracco et al., 2016, Boushabi et al., 2024, Barrera et al., 2024).

Error estimates are sharp: for $N_1(u) = u'(0)$3, on each element $N_1(u) = u'(0)$4,

$N_1(u) = u'(0)$5

and, globally, for smooth $N_1(u) = u'(0)$6,

$N_1(u) = u'(0)$7

for both univariate and multivariate settings (Bonizzoni et al., 2020, Bracco et al., 2016, Boushabi et al., 2024, Barrera et al., 2024).

4. Basis Structures and Implementation

A hallmark of cubic Hermite quasi-interpolation is the flexible basis representation. Several options include:

• Hermite element basis: Interpolatory, stable, $N_1(u) = u'(0)$8 functions with compact support, naturally aligned to derivative and value DOFs (Bonizzoni et al., 2020).
• Normalized B-spline-like basis: Non-negative, partition-of-unity, locally supported functions amenable to explicit quasi-interpolation via blossoming, control polynomials, or discrete polarization; particularly for Hermite data with $N_1(u) = u'(0)$9 at each knot (Boushabi et al., 2024).
• Bernstein–Bézier basis: Used on triangles, enabling explicit coefficient expressions for domain (vertex, edge, barycenter) points via local linear combinations governed by precomputed masks; ensures $N_2(u) = u'(1)$0 across edges (Barrera et al., 2024).

Efficient implementation leverages banded linear algebra for B-spline/Hermite systems, FFTs for periodic/cylindrical grids, and modularity across dimensions. The QIBSH++ library provides C++/MATLAB implementations supporting derivative-free options via finite differences, tensor-product extensions, and efficient coefficient computation through stencil-based assembly (Bertolazzi et al., 2022).

5. Adaptive, Hierarchical, and Generalized Constructions

Hierarchical and adaptive mesh strategies are supported through truncated hierarchical B-spline (THB-spline) frameworks. Here, nested tensor-product spline spaces are built at increasing levels of refinement, and a cell marking and subdivision strategy guides local mesh refinement. The hierarchical Hermite quasi-interpolant, constructed by replacing standard B-splines with their truncated hierarchical counterparts and propagating local Hermite functionals to all active basis elements, retains reproduction, locality, stability, and achieves essentially the same error as the full non-hierarchical tensor product but with much fewer degrees of freedom (Bracco et al., 2016).

In 2D and 3D, adaptive strategies sample the function on a fine point set, compute local errors, mark and refine problematic cells, and update the hierarchical quasi-interpolant iteratively. These techniques enable efficient resolution for localized features, singularities, or steep gradients while controlling global approximation error.

6. Practical Applications and Numerical Performance

Cubic Hermite quasi-interpolation operators are used in boundary value problem solvers, $N_2(u) = u'(1)$1-conforming finite element complexes with commuting projections, adaptive spline-based PDE solvers, surface/volume fitting in geometric modeling, and scientific computing tasks requiring stable, high-order smooth reconstructions from scattered Hermite data.

Empirical results consistently confirm the theoretical convergence rates: in experiments with smooth test functions (e.g., $N_2(u) = u'(1)$2, $N_2(u) = u'(1)$3), maximum error decays as $N_2(u) = u'(1)$4 in 1D (Boushabi et al., 2024). In bivariate and hierarchical settings, the error bound in both function and derivatives aligns with $N_2(u) = u'(1)$5 for cubic splines, attaining full approximation order even under adaptive refinement (Bracco et al., 2016, Barrera et al., 2024, Bertolazzi et al., 2022). The operators’ basis positivity confers shape-preserving and non-oscillatory properties beneficial for visualization and geometric modeling (Boushabi et al., 2024, Barrera et al., 2024).

7. Connections, Variants, and Ongoing Developments

Cubic Hermite quasi-interpolation is intimately connected with boundary-value multistep (BS) methods, blossoming theory, control polynomials, and projection frameworks in finite element exterior calculus. It generalizes to arbitrary degrees, non-uniform meshes, periodic and cylindrical domains, and derivative-free scenarios via finite-difference derivative reconstruction (Bertolazzi et al., 2022).

Recent work has yielded explicit closed-form mask coefficients for Bézier-based triangular quasi-interpolation with $N_2(u) = u'(1)$6 continuity, polynomial exactness, and compact support (Barrera et al., 2024), as well as strong error/stability guarantees for B-spline and THB-spline-based schemes on adaptive grids (Bracco et al., 2016). Ongoing research investigates optimal mask selection, superconvergence, and robust implementation in higher-dimensional, non-tensor-product, or anisotropic settings.

---

Key references:

• H1-conforming finite element cochain complexes and commuting quasi-interpolation operators on cartesian meshes (Bonizzoni et al., 2020)
• Normalized B-spline-like representation for low-degree Hermite osculatory interpolation problems (Boushabi et al., 2024)
• Bivariate hierarchical Hermite spline quasi--interpolation (Bracco et al., 2016)
• The Object Oriented c++ library QIBSH++ for Hermite spline Quasi Interpolation (Bertolazzi et al., 2022)
• Construction of 2D explicit cubic quasi-interpolating splines in Bernstein-Bézier form (Barrera et al., 2024)