Generalized Taylor–Hood FEM Pair

• Generalized Taylor–Hood finite element pair is a family of mixed finite element constructions that preserves the Taylor–Hood degree offset while adapting to nonstandard geometries and multiphysics environments.
• Key methodologies include phase-wise space duplication, Nitsche coupling, ghost penalty stabilization, and divergence-preserving Fortin operators to maintain robust inf-sup conditions.
• Extensions to surface formulations, weighted axisymmetric problems, and anisotropic meshes demonstrate practical insights for achieving optimal approximation and solver efficiency in complex simulations.

to=arxiv_search.search ￣影音先锋json {"query":"all:\"generalized Taylor-Hood\" OR ti:\"Taylor-Hood\" AND (Stokes OR surface OR unfitted OR anisotropic)", "max_results": 10, "sort_by": "relevance", "sort_order": "descending"} to=arxiv_search.search 天天彩票中奖json {"query":"generalized Taylor-Hood finite element pair", "max_results": 10, "sort_by": "relevance", "sort_order": "descending"} to=search_arxiv 天天中彩票是不是json {"query":"generalized Taylor-Hood finite element pair", "max_results": 10, "sort_by": "relevance", "sort_order": "descending"} to=arxiv.search 天天中彩票无法json {"query":"Taylor-Hood Stokes generalized finite element", "max_results": 10} The generalized Taylor–Hood finite element pair denotes a family of mixed finite element constructions that preserve the defining Taylor–Hood degree offset—higher-order approximation for the primal vector field and one-order-lower approximation for the pressure or pressure-like Lagrange multiplier—while extending the classical fitted Stokes setting to unfitted interfaces, embedded surfaces, weighted axisymmetric formulations, anisotropic meshes, nonconforming or reduced spaces, and related multiphysics systems (Olshanskii et al., 2021, Lee et al., 2012, Jankuhn et al., 2020, Reusken, 2024). Taken together, these works suggest that “generalized Taylor–Hood” is best understood as a design principle rather than a single element family: the polynomial hierarchy of Taylor–Hood is retained, but geometry representation, conformity, stabilization, norms, and even the ambient PDE class may be altered to preserve inf-sup stability, consistency, and optimal approximation.

1. Classical template and the meaning of “generalized”

In its classical conforming form, the lowest-order Taylor–Hood pair consists of continuous piecewise quadratic velocities and continuous piecewise linear pressures. One formulation writes

$$V_h=\mathcal L_2^1(\mathcal T;\mathbb R^d)\cap W^{1,1}_0(\Omega;\mathbb R^d), \qquad Q_h=\mathcal L_1^1(\mathcal T)\cap L^2_0(\Omega),$$

and establishes a constructive Fortin operator in any spatial dimension $d\ge 2$ (Diening et al., 2021). On tensor-product meshes, the corresponding family is the continuous $\mathcal Q_k-\mathcal Q_{k-1}$ pair, $k\ge 2$, on quadrilateral or hexahedral meshes (Jodlbauer et al., 2022). In both cases, the essential structural feature is the degree shift between velocity and pressure.

What changes in the generalized setting is not this degree pattern but the analytical and geometric context in which it is embedded. The phrase is used explicitly for the unfitted two-phase pair $\mathbf P_{k+1}-P_k$, $k\ge 1$, with phase-wise duplicate spaces, Nitsche coupling across an interface, viscosity-weighted pressure normalization, and ghost-penalty stabilization on cut patches (Olshanskii et al., 2021). It is also used for trace and surface formulations, where bulk Taylor–Hood spaces are restricted to an embedded surface or lifted to a high-order surface approximation, and for weighted axisymmetric Stokes problems, where the usual $P^{k+1}/P^k$ structure is analyzed in weighted Sobolev spaces on graded meshes (Lee et al., 2012). A closely related terminology appears in “Taylor–Hood-like,” “reduced Taylor–Hood-type,” and “surface analogue of Taylor–Hood” constructions (Hansbo et al., 2022, Liao et al., 2021, Demlow et al., 25 Jun 2025).

A representative summary of the main variants is as follows.

Setting	Discrete pair	Additional mechanism
Unfitted two-phase Stokes	$\mathbf P_{k+1}-P_k$ or $\mathbf Q_{k+1}-Q_k$	Phase-wise fields, Nitsche coupling, ghost penalty (Olshanskii et al., 2021)
Surface trace FEM	$P_k$-$d\ge 2$0 traces from bulk spaces	Tangential penalty, normal-derivative stabilization (Jankuhn et al., 2020)
Axisymmetric weighted Stokes	$d\ge 2$1	Weighted spaces and graded meshes (Lee et al., 2012)
Anisotropic meshes	$d\ge 2$2, $d\ge 2$3	Generalized Verfürth trick on edge and corner patches (Barrenechea et al., 2017)
Penalty-free surface method	$d\ge 2$4 on $d\ge 2$5	Exact tangentiality, Piola mapping, Gauss–Lobatto edge nodes (Demlow et al., 25 Jun 2025)

This classification also clarifies a useful boundary. Some works employ the standard Taylor–Hood pair inside a larger stabilized or multiphysics algorithm without altering the element family itself. One explicit example states that no new generalized Taylor–Hood element is introduced; instead, the classical $d\ge 2$6 pair is used within a stabilized Newton–Galerkin method for Darcy–Brinkman–Forchheimer flow (Yoon et al., 4 Jan 2025).

2. Unfitted and interface-coupled generalizations

A central line of development concerns interface problems in which the physical interface cuts through a fixed background mesh. In the two-phase Stokes problem with slip between phases, the generalized Taylor–Hood pair is

$d\ge 2$7

with $d\ge 2$8 built from phase-wise $d\ge 2$9 velocities and $\mathcal Q_k-\mathcal Q_{k-1}$0 from phase-wise $\mathcal Q_k-\mathcal Q_{k-1}$1 pressures, and with the corresponding $\mathcal Q_k-\mathcal Q_{k-1}$2 variant admitted on quadrilateral or hexahedral meshes (Olshanskii et al., 2021). Because the method is formulated on the background mesh with separate phase fields, discrete components may be multivalued in the overlap region $\mathcal Q_k-\mathcal Q_{k-1}$3. The interface conditions are imposed weakly through Nitsche terms, while ghost-penalty stabilization controls cut-patch pathologies.

The principal theoretical result is a robust inf-sup theorem. Under the assumptions that the mesh is sufficiently fine, the cut region is quasi-uniform, exact integration is used on cut cells, and the averages are chosen as $\mathcal Q_k-\mathcal Q_{k-1}$4, $\mathcal Q_k-\mathcal Q_{k-1}$5 with $\mathcal Q_k-\mathcal Q_{k-1}$6, the method satisfies

$\mathcal Q_k-\mathcal Q_{k-1}$7

with $\mathcal Q_k-\mathcal Q_{k-1}$8 independent of the viscosity ratio, slip coefficient, interface position relative to the background mesh, and mesh size $\mathcal Q_k-\mathcal Q_{k-1}$9 (Olshanskii et al., 2021). Coercivity of the velocity form is proved with the same parameter robustness, and the resulting error estimate is

$k\ge 2$0

again with $k\ge 2$1 independent of $k\ge 2$2, $k\ge 2$3, $k\ge 2$4, and the interface position. For $k\ge 2$5, the reported two- and three-dimensional experiments show approximately third order in velocity $k\ge 2$6 and second order in weighted $k\ge 2$7-velocity and weighted $k\ge 2$8-pressure (Olshanskii et al., 2021).

A closely related earlier unfitted construction for stationary Stokes interface problems starts from the classical $k\ge 2$9 pair, then duplicates the spaces phase-wise: $\mathbf P_{k+1}-P_k$0 so that the discrete velocity can represent kinks and the pressure can represent jumps across the interface (Lederer et al., 2016). A Nitsche-type formulation weakly enforces continuity and traction balance, and a ghost penalty on pressure derivatives supplies an inf-sup stable variational formulation with a constant independent of the interface position relative to the mesh. The numerical comparison in that work isolates two distinct ingredients: velocity enrichment is needed to recover interface approximation quality, and high-order parametric geometry mapping is needed to recover full convergence order (Lederer et al., 2016).

These unfitted results make clear that the generalized character of the pair is not merely geometric. The velocity-pressure spaces are altered by duplication, the pressure normalization is reweighted, and the underlying inf-sup proof is rebuilt around ghost penalties, cut-cell trace inequalities, and interface-consistent Nitsche couplings (Olshanskii et al., 2021, Lederer et al., 2016).

3. Surface and trace variants

For the surface Stokes equations, generalized Taylor–Hood pairs arise because one retains the Taylor–Hood degree hierarchy while discarding the fitted planar mesh paradigm. In the trace FEM formulation on a smooth closed surface $\mathbf P_{k+1}-P_k$1, bulk spaces on a tetrahedral mesh are restricted to the surface: $\mathbf P_{k+1}-P_k$2 with particular emphasis on the trace $\mathbf P_{k+1}-P_k$3–$\mathbf P_{k+1}-P_k$4 pair (Olshanskii et al., 2019). Since the surface cuts arbitrarily through the background tetrahedra, pressure stabilization by a volume normal derivative term is essential. The main theorem for $\mathbf P_{k+1}-P_k$5 states a discrete inf-sup estimate with a constant independent of both $\mathbf P_{k+1}-P_k$6 and the position of the surface relative to the bulk mesh. The same work shows that without pressure stabilization the trace $\mathbf P_{k+1}-P_k$7–$\mathbf P_{k+1}-P_k$8 pair loses inf-sup stability under refinement, whereas normal volume stabilization restores stability; full-gradient stabilization is also stable but produces larger consistency error and is suboptimal for the analyzed method (Olshanskii et al., 2019).

Higher-order trace formulations generalize this construction to

$\mathbf P_{k+1}-P_k$9

using a parametric geometry map $k\ge 1$0, penalty enforcement of tangentiality, and normal-derivative volume stabilization for both velocity and pressure (Jankuhn et al., 2020). The central discrete inf-sup bound is

$k\ge 1$1

with $k\ge 1$2 independent of $k\ge 1$3 and of how $k\ge 1$4 cuts the background mesh. The resulting energy-norm estimate is optimal of order $k\ge 1$5, while the quantified geometric consistency error is one order higher and therefore non-dominant (Jankuhn et al., 2020). A parallel numerical study of the same family, phrased as higher-order trace FEM with a consistent and an inconsistent penalty formulation, recommends the isoparametric choice $k\ge 1$6, the stabilization scalings $k\ge 1$7, $k\ge 1$8, and a more accurate penalty normal of order $k\ge 1$9; for the consistent method $P^{k+1}/P^k$0, while for the inconsistent method $P^{k+1}/P^k$1 (Jankuhn et al., 2019).

The fitted surface finite element analogue uses the classical surface FEM of Dziuk–Elliott combined with a Hood–Taylor pair on a high-order surface approximation $P^{k+1}/P^k$2. The spaces are

$P^{k+1}/P^k$3

with $P^{k+1}/P^k$4, $P^{k+1}/P^k$5, $P^{k+1}/P^k$6, and a penalty parameter chosen as $P^{k+1}/P^k$7 (Reusken, 2024). The main error bound has the form

$P^{k+1}/P^k$8

so that for sufficiently accurate geometry, $P^{k+1}/P^k$9, the optimal order $\mathbf P_{k+1}-P_k$0 is obtained in the energy norm (Reusken, 2024).

A more recent surface construction removes penalization entirely. It defines a tangential, $\mathbf P_{k+1}-P_k$1-conforming but $\mathbf P_{k+1}-P_k$2-nonconforming velocity space on $\mathbf P_{k+1}-P_k$3, keeps the pressure in the standard continuous $\mathbf P_{k+1}-P_k$4 space, and proves the inf-sup condition

$\mathbf P_{k+1}-P_k$5

Its optimality depends on locating edge degrees of freedom at Gauss–Lobatto nodes; standard equispaced Lagrange edge nodes lead to a loss of one order for $\mathbf P_{k+1}-P_k$6 (Demlow et al., 25 Jun 2025). This result sharply illustrates a recurring theme in generalized Taylor–Hood theory: once conformity or geometry is altered, seemingly secondary implementation choices may become mathematically decisive.

4. Weighted, singular, and anisotropic adaptations

The axisymmetric Stokes equations in polygonal meridian domains furnish a distinct generalization in which the polynomial pair remains classical, but the functional framework is not. After reduction from the three-dimensional axisymmetric domain to a meridian polygon $\mathbf P_{k+1}-P_k$7 in $\mathbf P_{k+1}-P_k$8-coordinates, the problem involves singular coefficients $\mathbf P_{k+1}-P_k$9, $\mathbf Q_{k+1}-Q_k$0, and the weighted divergence $\mathbf Q_{k+1}-Q_k$1 (Lee et al., 2012). The natural weak formulation is posed in

$\mathbf Q_{k+1}-Q_k$2

and the regularity theory is developed in weighted Kondrat’ev-type spaces $\mathbf Q_{k+1}-Q_k$3.

The discrete pair is still the Taylor–Hood pair,

$\mathbf Q_{k+1}-Q_k$4

with

$\mathbf Q_{k+1}-Q_k$5

but its analysis depends on weighted norms, weighted interpolation operators, and $\mathbf Q_{k+1}-Q_k$6-graded meshes near singular vertices (Lee et al., 2012). The grading parameter is chosen as

$\mathbf Q_{k+1}-Q_k$7

and the resulting error estimate is

$\mathbf Q_{k+1}-Q_k$8

The reported $\mathbf Q_{k+1}-Q_k$9-$P_k$0 experiments show rates close to the optimal value $P_k$1 on sufficiently graded meshes and substantial degradation on quasi-uniform meshes (Lee et al., 2012).

A different adaptation concerns anisotropic refinement. For the lowest-order pairs $P_k$2 and $P_k$3 on two-dimensional anisotropic meshes containing refined edge and corner patches, uniform LBB conditions are proved with constants independent of the aspect ratio of thin elements (Barrenechea et al., 2017). The local edge-patch theorem gives

$P_k$4

where $P_k$5 is independent of the size and aspect ratio of the elements in the patch. The proof uses a generalized Verfürth decomposition, explicit low-dimensional pressure splittings, and bubble-function constructions tailored to the anisotropic geometry (Barrenechea et al., 2017).

These two strands indicate different meanings of generalization. In the axisymmetric case, the pair is generalized by weighted spaces and graded meshes adapted to singular coefficients (Lee et al., 2012). In the anisotropic case, the pair itself is unchanged, but the stability theory is extended from shape-regular meshes to a structured anisotropic mesh class (Barrenechea et al., 2017). Both preserve the Taylor–Hood degree hierarchy while substantially altering the analytical environment.

5. Reduced, nonconforming, and Fortin-based reinterpretations

Generalization also occurs through modification of the velocity space while keeping the pressure space in a Taylor–Hood pattern. A foundational result is the construction of a Fortin operator for the lowest-order Taylor–Hood element in arbitrary dimension using tangential edge bubble functions (Diening et al., 2021). The operator is written as

$P_k$6

where $P_k$7 is Scott–Zhang type and $P_k$8 is a divergence-correcting operator built from edge bubbles. The key property is discrete divergence preservation against all $P_k$9, which yields the mesh-size-independent inf-sup condition for the standard $d\ge 2$00-$d\ge 2$01 pair in every dimension $d\ge 2$02 (Diening et al., 2021).

The same construction leads to an alternative reduced stable pair,

$d\ge 2$03

where $d\ge 2$04 is spanned by tangential edge bubbles (Diening et al., 2021). In three dimensions, the total number of unknowns on a standard uniform simplicial partition of the cube is noted to be roughly halved compared to MINI. The same paper also provides a sharp negative result: on an explicit octahedral mesh in three dimensions, the $d\ge 2$05–$d\ge 2$06 and the lowest-order augmented Taylor–Hood pairs are not inf-sup stable, so no divergence-preserving Fortin operator can exist for them (Diening et al., 2021). This is an important corrective to the common assumption that any enrichment or reduction adjacent to Taylor–Hood inherits stability automatically.

A separate nonconforming reduction is obtained with a rotated $d\ge 2$07-type tetrahedral velocity space and continuous piecewise linear pressure: $d\ge 2$08 The local scalar space is

$d\ge 2$09

and the global space enforces continuity only at interior edge midpoints (Hansbo et al., 2022). The authors describe the resulting approximation as similar to the well known continuous $d\ge 2$10–$d\ge 2$11 Taylor–Hood element but with fewer degrees of freedom. Stability is proved by a Verfürth-type argument, including a mesh-dependent norm

$d\ge 2$12

and the method additionally satisfies Korn’s inequality, making it stable for the strain or stress form of the Stokes equations relevant to free-surface flow (Hansbo et al., 2022).

In these reduced and nonconforming variants, the generalized Taylor–Hood idea ceases to mean “use $d\ge 2$13 literally” and instead becomes “retain Taylor–Hood-type pressure coupling and inf-sup structure while redesigning the velocity space.” The Fortin-operator viewpoint makes this especially transparent: what matters is not only polynomial degree but the existence of a divergence-preserving projection with local stability and approximation properties (Diening et al., 2021).

6. Multiphysics extensions, solver implications, and limits of the term

Several works transfer the Taylor–Hood pattern beyond the classical Stokes system. In poroelasticity, a reformulation with two pseudo-pressures splits the model into a generalized Stokes-type system for $d\ge 2$14 and a diffusion equation for $d\ge 2$15, and a Taylor–Hood mixed finite element method is used for the Stokes subproblem together with a conforming $d\ge 2$16 method for diffusion (Feng et al., 2014). The fully discrete scheme satisfies a discrete energy law, converges optimally in the energy norm, and avoids the locking phenomenon reported for direct displacement-pressure discretizations (Feng et al., 2014). In nearly incompressible strain gradient elasticity, a bubble-enriched nonconforming displacement space of order $d\ge 2$17 is paired with a continuous $d\ge 2$18 pressure space, yielding a Taylor–Hood-like family robust in both the incompressible limit and the singular perturbation parameter $d\ge 2$19 (Liao et al., 2021).

A monolithic fluid–porous structure interaction method uses a standard $d\ge 2$20 Taylor–Hood discretization for fluid-type variables in both fluid and porous subdomains, together with $d\ge 2$21 spaces for structure velocity and porous flux (Lozovskiy et al., 2021). The coupled formulation is posed in ALE coordinates, includes an energy-consistent modified interface stress condition, and in the reported three-dimensional benchmark remains computationally manageable with $d\ge 2$22 degrees of freedom (Lozovskiy et al., 2021). A fully decoupled GSAV-BDF discretization of the incompressible Boussinesq equations likewise uses a Taylor–Hood-based $d\ge 2$23-conforming spatial discretization with $d\ge 2$24 velocity, $d\ge 2$25 pressure, and $d\ge 2$26 temperature; the numerical results support second-order temporal convergence and demonstrate applicability to large three-dimensional problems with around $d\ge 2$27 spatial unknowns per time step (Wagner et al., 17 Apr 2025).

The generalized Taylor–Hood pair also has direct solver consequences. For the continuous $d\ge 2$28 quadrilateral or hexahedral family, matrix-free monolithic geometric multigrid solvers based on scaled Chebyshev–Jacobi smoothers are analyzed for Stokes and generalized Stokes systems (Jodlbauer et al., 2022). The discrete problem yields the symmetric indefinite saddle-point system

$d\ge 2$29

and the analysis establishes smoothing and approximation properties in a framework based on Schöberl–Zulehner. This underscores a practical point: once Taylor–Hood is generalized to high-order tensor-product meshes, matrix-free sum-factorization and multigrid preconditioning become integral to its viability at scale (Jodlbauer et al., 2022).

At the same time, the literature sharply separates generalized element design from generalized algorithmic context. The steady Darcy–Brinkman–Forchheimer work explicitly states that it does not introduce a new generalized Taylor–Hood element; it uses the standard $d\ge 2$30 pair together with Newton linearization, Grad–Div stabilization, an augmented-Lagrangian pressure preconditioner, and adaptive refinement (Yoon et al., 4 Jan 2025). This distinction is substantive. A method may be Taylor–Hood-based without altering the pair itself, just as a pair may be generalized without changing the outer solver. The term therefore properly applies when the velocity-pressure spaces or their stability framework are extended, not merely when the standard pair is embedded in a more elaborate computational pipeline.

Across these developments, a common invariant remains visible: the Taylor–Hood degree offset organizes the mixed approximation, but the decisive mathematical work lies in adapting inf-sup stability, coercivity, and approximation theory to nonstandard geometry, weighted norms, nonconforming continuity, or coupled physics (Olshanskii et al., 2021, Lee et al., 2012, Jankuhn et al., 2020, Diening et al., 2021).