Local Taylor-based Polynomial Quasi-Trefftz Space

• Local Taylor-based polynomial quasi-Trefftz spaces are equation-dependent polynomial spaces that enforce vanishing Taylor residuals at a specific point to yield approximate solutions of PDEs.
• They offer explicit basis constructions with reduced local dimensions while preserving high-order approximation properties for both scalar and system equations.
• These spaces are integral in DG and Trefftz-like discretizations, enhancing computational efficiency and stability in high-order numerical methods.

Local Taylor-based polynomial quasi-Trefftz spaces are local, equation-dependent polynomial spaces whose elements are not exact local solutions of a partial differential equation, but local approximate solutions in the precise sense that a truncated Taylor expansion of the residual vanishes at a chosen point. In the scalar setting, the modern formulation identifies such a space as the kernel of the quasi-Trefftz operator, i.e. the composition of Taylor truncation with the differential operator. For scalar linear operators with smooth coefficients, this kernel-based viewpoint yields spaces that are substantially smaller than full polynomial spaces while retaining the same local approximation orders, together with fully explicit basis-construction procedures valid in arbitrary dimension and for operators of arbitrary order (Imbert-Gerard, 24 May 2025).

1. Definition and conceptual setting

Trefftz methods use local discrete spaces made of exact local solutions of the governing PDE. Local Taylor-based polynomial quasi-Trefftz spaces relax exactness: a polynomial is retained if, after applying the differential operator, its Taylor expansion up to the maximal admissible order vanishes at a fixed point, typically an element center or barycenter. For a scalar differential operator $\mathcal{L}$ of order $\gamma$, polynomial degree $p\ge \gamma$, and expansion point $\mathbf{x}_0$, the scalar local space is

$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$

equivalently $\mathbb{QT}_p=\ker(\mathcal{D}_p)$ for

$$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$

The truncation order $p-\gamma$ is the natural one because an operator of order $\gamma$ applied to a degree-$p$ polynomial produces a polynomial of degree at most $\gamma$0 (Imbert-Gerard, 24 May 2025).

A practically important inhomogeneous variant appears for general linear PDEs with smooth coefficients and source term. If $\gamma$1 has order $\gamma$2 on an element $\gamma$3 with center $\gamma$4, the degree-$\gamma$5 local quasi-Trefftz space with source $\gamma$6 is

$\gamma$7

This states that the PDE is satisfied to Taylor order $\gamma$8 at the chosen point. In this formulation, if $\gamma$9, every polynomial of degree at most $p\ge \gamma$0 is trivially quasi-Trefftz, so the construction becomes nontrivial only for $p\ge \gamma$1 (Imbert-Gérard et al., 2024).

The modifier “local” is essential. These spaces are defined pointwise through Taylor data and are therefore elementwise objects. They become usable in numerical discretizations only after a global coupling mechanism—most often discontinuous Galerkin, embedded Trefftz, or LDG coupling—is imposed across mesh interfaces (Lederer et al., 2024).

2. Algebraic structure and explicit basis construction

The central structural fact is that the quasi-Trefftz operator is most naturally analyzed on the graded decomposition of polynomial spaces,

$p\ge \gamma$2

where $p\ge \gamma$3 denotes homogeneous polynomials of degree $p\ge \gamma$4. In this representation, the quasi-Trefftz operator has a block-triangular structure. Its leading block is determined by the principal part of the differential operator at the expansion point,

$p\ge \gamma$5

while all lower-order terms and coefficient variations contribute only to strictly lower blocks. This graded viewpoint is the basis of the systematic scalar theory and is the reason an explicit forward-substitution construction is possible (Imbert-Gerard, 24 May 2025).

For scalar linear equations, the crucial solvability input is that, for every $p\ge \gamma$6, the map

$p\ge \gamma$7

is surjective for every nontrivial linear constant-coefficient homogeneous differential operator. The construction then proceeds degree by degree. The homogeneous components below degree $p\ge \gamma$8 are free. For each higher degree, one solves

$p\ge \gamma$9

so that the residual vanishes at the required Taylor order. The outcome is a fully explicit local basis with minimal computational cost for scalar operators with smooth coefficients (Imbert-Gerard, 24 May 2025).

Earlier constructive formulations already exhibited the same recursive pattern, although without the later fully general kernel formalism. In the general smooth-coefficient setting for a PDE of order $\mathbf{x}_0$0, one may impose a non-degeneracy condition such as

$\mathbf{x}_0$1

choose as free “Cauchy data” the coefficients with first index $\mathbf{x}_0$2, and recover the remaining coefficients recursively from the Taylor constraints. Each admissible choice of Cauchy data determines a unique quasi-Trefftz polynomial, and a basis is obtained by systematically selecting a basis for those data (Imbert-Gérard et al., 2024). A closely related recursive construction appears for diffusion-advection-reaction operators with piecewise-smooth coefficients, where the polynomial coefficients are computed from local Taylor expansions of the PDE coefficients by a simple completion algorithm (Perinati, 2023).

3. Dimension, economy, and approximation properties

The main appeal of local Taylor-based polynomial quasi-Trefftz spaces is that they preserve the approximation order of full polynomial spaces while drastically reducing local dimension. In the scalar kernel formulation,

$\mathbf{x}_0$3

so the dimension is reduced by exactly the number of independent Taylor residual constraints (Imbert-Gerard, 24 May 2025).

For smooth exact solutions, the Taylor polynomial itself belongs to the local quasi-Trefftz space. In the scalar homogeneous case, if $\mathbf{x}_0$4 solves $\mathbf{x}_0$5, then $\mathbf{x}_0$6, and the best local quasi-Trefftz approximation reproduces Taylor accuracy: $\mathbf{x}_0$7 In the inhomogeneous smooth-coefficient formulation, the degree-$\mathbf{x}_0$8 Taylor polynomial of the exact solution at $\mathbf{x}_0$9 belongs to $$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$0, and the local best-approximation estimate matches that of the full polynomial space in $$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$1 seminorms (Imbert-Gerard, 24 May 2025, Imbert-Gérard et al., 2024).

The following representative formulas illustrate the resulting economies.

Setting	Local defining condition	Dimension result
Scalar operator of order $$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$2	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$3	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$4 (Imbert-Gerard, 24 May 2025)
Second-order elliptic scalar PDE	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$5	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$6 in 2D, $$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$7 in 3D (Imbert-Gérard et al., 2024)
Space-time heat equation	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$8	$$\mathbb{QT}_p = \left\{ \Pi \in \mathbb{P}_p \mid T_{p-\gamma}[\mathcal{L}\Pi]=0 \right\},$$9 (Gómez et al., 2024)
Second-order Maxwell system	$\mathbb{QT}_p=\ker(\mathcal{D}_p)$0 and $\mathbb{QT}_p=\ker(\mathcal{D}_p)$1	$\mathbb{QT}_p=\ker(\mathcal{D}_p)$2 (Imbert-Gérard, 29 Aug 2025)

These formulas make precise the statement that quasi-Trefftz spaces are “much smaller” than full polynomial spaces. In the heat equation case, for example, the local dimension scales as $\mathbb{QT}_p=\ker(\mathcal{D}_p)$3, whereas full space-time polynomials scale as $\mathbb{QT}_p=\ker(\mathcal{D}_p)$4 (Gómez et al., 2024). In elliptic diffusion-advection-reaction problems, the same $\mathbb{QT}_p=\ker(\mathcal{D}_p)$5 convergence as full polynomial DG can be obtained with these reduced spaces (Imbert-Gérard et al., 2024).

4. Role in discontinuous Galerkin and Trefftz-like discretizations

Local Taylor-based polynomial quasi-Trefftz spaces are mainly used as building blocks inside discontinuous formulations. For elliptic diffusion-advection-reaction problems with smooth coefficients, polynomial quasi-Trefftz DG restricts a classical SIPG-plus-upwind bilinear form to the quasi-Trefftz trial and test spaces. In the non-homogeneous case, a local quasi-Trefftz particular solution is constructed elementwise and the homogeneous correction is solved for globally. The resulting scheme is stable, achieves high-order convergence, and its stated practical advantage over standard DG is higher accuracy for comparable numbers of degrees of freedom (Imbert-Gérard et al., 2024).

For piecewise-smooth coefficients, a closely related quasi-Trefftz DG method for the homogeneous diffusion-advection-reaction equation proves that the local quasi-Trefftz space has smaller dimension than the full polynomial space of the same degree while yielding the same optimal convergence rates. The method is shown to be well-posed, consistent, stable, and high-order convergent, with numerical experiments in two dimensions exhibiting the expected approximation and convergence behavior (Perinati, 2023).

In space-time wave discretization, the quasi-Trefftz condition is imposed on polynomials whose Taylor coefficients satisfy the variable-coefficient wave operator up to order $\mathbb{QT}_p=\ker(\mathcal{D}_p)$6 at the element center: $\mathbb{QT}_p=\ker(\mathcal{D}_p)$7 The basis is generated recursively from Cauchy data $\mathbb{QT}_p=\ker(\mathcal{D}_p)$8 and $\mathbb{QT}_p=\ker(\mathcal{D}_p)$9, and the DG scheme built from the associated first-order variables is proved stable and high-order convergent for piecewise-smooth coefficients (Imbert-Gérard et al., 2020).

A broader abstract interpretation is supplied by the unified framework for Trefftz-like methods. There the discrete space is decomposed as

$$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$0

with the quasi-Trefftz space appearing as the globally coupled component satisfying local Taylor conditions. A notable analytical point is that the resulting error estimate is formulated in terms of the best approximation over the full discrete space $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$1, not only over the quasi-Trefftz subspace. This framework covers quasi-Trefftz methods for second-order scalar elliptic PDEs and a scalar reaction-advection problem and clarifies their relation to embedded Trefftz DG and other local/global decompositions (Lederer et al., 2024).

For parabolic problems, an inf-sup stable space-time LDG method for the heat equation admits quasi-Trefftz polynomial spaces as local trial spaces. The weaker inf-sup theory applies because quasi-Trefftz spaces generally do not satisfy $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$2, but the method still yields optimal algebraic rates in the natural energy-type norm with substantially reduced local dimension (Gómez et al., 2024).

5. Extensions to wave problems, first-order systems, and Maxwell equations

The polynomial quasi-Trefftz idea first matured in wave-oriented contexts where exact polynomial Trefftz spaces are unavailable or too restrictive. For the 3D convected Helmholtz equation, a polynomial quasi-Trefftz function is defined by requiring the PDE residual to satisfy

$$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$3

which is equivalent to vanishing of the Taylor coefficients of the residual up to degree $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$4. The corresponding coefficient system is linear and triangular by degree, the dimension of the space is $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$5, and the basis is generated by initializing one free monomial coefficient for multi-indices with $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$6 and solving recursively for the remainder. The same work reports that the polynomial basis avoids the ill-conditioning typical of wave-like quasi-Trefftz bases: in one test case, at order $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$7, the condition number was $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$8 for the plane-wave basis and $$\mathcal{D}_p:\mathbb{P}_p\to \mathbb{P}_{p-\gamma}, \qquad \Pi\mapsto T_{p-\gamma}[\mathcal{L}\Pi].$$9 for the polynomial quasi-Trefftz basis (Imbert-Gerard et al., 2022).

For variable-coefficient wave propagation, generalized plane waves (GPWs) provide a nonlinear quasi-Trefftz alternative. A GPW basis function has the form $p-\gamma$0 and is required to satisfy a Taylor-truncated residual condition of the form

$p-\gamma$1

The polynomial quasi-Trefftz space is the linear analog of this construction: it uses the ansatz $p-\gamma$2 and the linear condition $p-\gamma$3. The two constructions are therefore closely related, but polynomial quasi-Trefftz spaces retain a finite-dimensional linear-algebraic structure (Fontana et al., 13 Aug 2025).

The extension from scalar equations to systems is nontrivial. For the second-order time-harmonic Maxwell equation with variable coefficient $p-\gamma$4, the local Taylor-based polynomial quasi-Trefftz space is defined by two simultaneous conditions: $p-\gamma$5 The second, quasi-divergence condition is essential because of the structure of the curl-curl operator. A Helmholtz decomposition of homogeneous polynomial vector fields yields both an explicit dimension formula,

$p-\gamma$6

and a recursive construction procedure for the local basis. This is identified as the first study of local Taylor-based polynomial quasi-Trefftz spaces for a system of PDEs (Imbert-Gérard, 29 Aug 2025).

A complementary first-order development appears for a first-order formulation of the Helmholtz equation in 2D. There the local space consists of triples $p-\gamma$7 satisfying Taylor-truncated first-order residual constraints, and the local dimension is $p-\gamma$8. Two explicit recursive constructions are given: a coupled system algorithm and a decoupled pressure algorithm based on the associated scalar Helmholtz equation (Imbert-Gérard et al., 10 Sep 2025).

6. Distinctions, limitations, and related constructions

Several distinctions are fundamental. First, quasi-Trefftz does not mean Trefftz. In a classical Trefftz space, the local residual vanishes identically in the element; in a local Taylor-based polynomial quasi-Trefftz space, only a finite Taylor jet of the residual vanishes at a point. The two coincide only in special constant-coefficient situations where exact polynomial solutions exist or where the truncated condition is sufficient to reconstruct an exact local polynomial solution (Imbert-Gerard, 24 May 2025).

Second, the local condition is point-centered rather than globally conforming. By itself, a quasi-Trefftz space does not enforce continuity or interface transmission. This is why quasi-Trefftz spaces are predominantly paired with DG-type couplings, embedded Trefftz constructions, or other weak interface formulations (Lederer et al., 2024).

Third, the family is generally not nested in the polynomial degree: for diffusion-advection-reaction quasi-Trefftz spaces one has, in general, $p-\gamma$9. This distinguishes them from standard polynomial hierarchies and affects basis design, $\gamma$0-adaptivity, and conditioning strategies (Perinati, 2023).

A further limitation is exposed by the Helmholtz equation itself. For nonzero wavenumber, no nonvanishing piecewise-polynomial exact Trefftz function exists because the Helmholtz operator contains both second- and zero-order terms. This is one of the original motivations for quasi-Trefftz constructions and for non-polynomial exact Trefftz spaces such as plane waves or generalized harmonic polynomials (Hiptmair et al., 2015).

Finally, not every local PDE-adapted polynomial space should be identified with a Taylor-based quasi-Trefftz space. In the embedded Trefftz DG method for the Oseen problem, the analysis relies on an explicitly constructed local complement space

$\gamma$1

where $\gamma$2 is polynomial, local, and designed so that the Oseen operator is stably invertible on it. The paper explicitly states that for Oseen this construction is polynomial and geometric, not strictly Taylor-based, even though it is closely related to earlier Taylor-based quasi-Trefftz ideas in the constant-coefficient Stokes case (Stocker et al., 11 Jun 2026). This distinction delineates the quasi-Trefftz concept from the broader class of local PDE-tailored polynomial decompositions.

In current usage, the term therefore denotes a precise algebraic and local notion: a polynomial kernel space defined by Taylor-truncated residual annihilation. Its significance lies in combining PDE adaptation, explicit local construction, reduced dimension, and high-order approximation for variable-coefficient problems where exact local Trefftz spaces are unavailable or impractical (Imbert-Gerard, 24 May 2025).