Immersed Raviart–Thomas Space Overview

• Immersed Raviart–Thomas space is a modified finite element space that retains classical RT structure while integrating geometric, interface, or stability corrections.
• Variants include curved-boundary formulations that shift flux degrees of freedom, unfitted interface methods enforcing transmission conditions, and RT bubble enrichments for divergence control.
• These constructions improve accuracy and stability in mixed formulations by accurately imposing true boundary normals and local conservation constraints.

Searching arXiv for the cited papers to ground the article in current records. Immersed Raviart–Thomas space denotes a family of constructions in which the classical Raviart–Thomas (RT) finite element structure is modified or embedded so that flux unknowns better reflect geometric or interface information that is not represented by a standard fitted simplicial mesh. In recent work, the term is used in at least three technically distinct senses: a Petrov–Galerkin trial flux space whose boundary degrees of freedom are shifted from a polyhedral approximation to the true curved boundary of a smooth domain; a locally modified RT space on unfitted interface elements; and an RT-bubble enrichment embedded into Scott–Vogelius velocity spaces to restore inf–sup stability on general simplicial grids (Ruas, 24 Feb 2026, Ji, 2021, John et al., 2022, Capatina et al., 2 Apr 2026).

1. Terminology and range of meanings

The terminology is not uniform. In the curved-domain mixed formulation of Raviart–Thomas elements on straight-edged tetrahedra, the immersed space is the trial flux space $P_h^k$, whose RT face-normal degrees of freedom on Neumann faces are imposed at points on the true boundary $\Gamma$ rather than on the polyhedral boundary $\Gamma_h$ (Ruas, 24 Feb 2026). In unfitted interface methods, the immersed RT space is a local element space on cut cells, typically piecewise RT on the subcells $T^\pm$ or $T_i$, with additional interface constraints that encode continuity of normal flux and a tangential constitutive condition (Ji, 2021, Capatina et al., 2 Apr 2026). By contrast, in the Scott–Vogelius stabilization literature, “immersed RT” explicitly refers to embedding a small RT bubble subspace into an $H^1$-conforming velocity space; it does not refer to unfitted or immersed boundary/interface discretization (John et al., 2022).

Usage	Defining mechanism	Representative paper
Curved-boundary trial space	Boundary RT DOFs shifted from $\Gamma_h$ to $\Gamma$	(Ruas, 24 Feb 2026)
Unfitted interface space	Piecewise RT on cut cells with interface constraints	(Ji, 2021, Capatina et al., 2 Apr 2026)
Scott–Vogelius enrichment	RT bubbles embedded into $V_k$	(John et al., 2022)

This suggests that “immersed” functions less as the name of a single canonical finite element family than as a design principle: standard RT structure is retained locally, while geometry-, interface-, or stability-critical information is injected through modified constraints, modified degrees of freedom, or embedded bubble components.

2. Classical Raviart–Thomas structure retained under immersion

All recent variants preserve the algebraic core of the classical RT family. On a simplex $K$, the local space is

$\Gamma$0

with

$\Gamma$1

and the usual unisolvent degrees of freedom are face-normal flux moments together with suitable interior moments for $\Gamma$2 (Ruas, 24 Feb 2026, John et al., 2022, Capatina et al., 2 Apr 2026). In the lowest-order triangular case,

$\Gamma$3

so the divergence is constant on each element and the edge-normal flux moments determine the field (Ji, 2021).

The contravariant Piola transform remains central. For an affine map $\Gamma$4 with Jacobian $\Gamma$5, the RT pullback is

$\Gamma$6

which preserves normal fluxes and transforms divergence by

$\Gamma$7

This is exploited both in the curved-domain tetrahedral setting, where the geometry is kept straight-edged and “immersion” is realized only through shifted boundary evaluations, and in RT-bubble enrichment, where elementwise reference-space constructions transfer by simple scaling (Ruas, 24 Feb 2026, John et al., 2022).

What changes from one immersed variant to another is therefore not the local RT polynomial ansatz itself, but the manner in which continuity, boundary data, interface laws, or divergence-surjectivity are enforced.

3. Curved domains with straight-edged tetrahedra

For smooth curved three-dimensional domains, the mixed model considered in recent work is the first-order system

$\Gamma$8

with $\Gamma$9 on $\Gamma_h$0 and $\Gamma_h$1 on $\Gamma_h$2, where $\Gamma_h$3 and $\Gamma_h$4 is allowed. The mixed weak formulation uses

$\Gamma_h$5

and $\Gamma_h$6, with the standard mean-zero condition when $\Gamma_h$7 (Ruas, 24 Feb 2026).

On a straight-edged tetrahedral mesh $\Gamma_h$8 approximating $\Gamma_h$9 by a polyhedron $T^\pm$0, three discrete spaces are introduced. The scalar space $T^\pm$1 consists of piecewise polynomials of total degree at most $T^\pm$2. The test flux space $T^\pm$3 is the standard $T^\pm$4-conforming RT space with $T^\pm$5 on polyhedral Neumann faces $T^\pm$6. The immersed trial flux space $T^\pm$7 is also elementwise $T^\pm$8, but on Neumann faces its boundary DOFs are enforced at points $T^\pm$9 on the true boundary $T_i$0, obtained by shifting designated face quadrature points $T_i$1 along half-lines $T_i$2 to their nearest intersections with $T_i$3. The number of shifted points is

$T_i$4

for $T_i$5 there is one point, the face centroid, and for $T_i$6 there are three symmetric barycentric points $T_i$7, $T_i$8, $T_i$9 (Ruas, 24 Feb 2026).

The resulting Petrov–Galerkin formulation keeps testing on the polyhedral boundary but imposes trial-side normal flux conditions against the true normal $H^1$0 on the true boundary. In operational terms, the local RT shape functions are built on straight tetrahedra and their face traces are evaluated at the shifted points $H^1$1, so no nonaffine curved-element mapping is required. For general Neumann data $H^1$2, the curved-boundary enforcement is represented by

$H^1$3

The principal motivation is the observed accuracy downgrade that occurs when true flux DOFs live on $H^1$4 but are imposed on $H^1$5. The immersed trial space is designed to avoid that boundary shift. Numerical evidence for the two lowest orders supports the design. For $H^1$6, the immersed Petrov–Galerkin method and the “do-nothing” Galerkin method both achieve first order in $H^1$7 for $H^1$8 and its divergence on the unit ball with $H^1$9, but the immersed method is clearly more accurate in $\Gamma_h$0, with $\Gamma_h$1 experimental order of convergence $\Gamma_h$2 versus $\Gamma_h$3 for Galerkin and consistently smaller errors. For $\Gamma_h$4, first-order convergence of $\Gamma_h$5 persists and the immersed method is more accurate across all reported metrics. On a hollow ellipsoid with outer Neumann and inner Dirichlet data, optimal first-order rates are recovered for $\Gamma_h$6 and its divergence, while $\Gamma_h$7 shows near-second-order behavior for both formulations, with the immersed method slightly more accurate (Ruas, 24 Feb 2026).

The same paper also studies the Hermite analog $\Gamma_h$8, where $\Gamma_h$9 is approximated in an incomplete quadratic polynomial space $\Gamma$0 and $\Gamma$1 via the broken gradient. The bilinear form

$\Gamma$2

delivers second-order $\Gamma$3 convergence for $\Gamma$4 and first-order convergence for $\Gamma$5 in $\Gamma$6. In the reported tests, immersed $\Gamma$7 consistently outperforms its Galerkin counterpart for $\Gamma$8, sometimes with superconvergent behavior for $\Gamma$9 (Ruas, 24 Feb 2026).

For $V_k$0 with $V_k$1, a symmetric enriched Petrov–Galerkin variant is introduced: $V_k$2

$V_k$3

This formulation yields SPD linear systems. The reported implementation uses conjugate gradients and static condensation of seven element-internal DOFs, leaving only face-normal flux DOFs globally. In the unit ball test, immersed $V_k$4 achieves EOCs $V_k$5–$V_k$6 for $V_k$7 and $V_k$8, versus $V_k$9–$K$0 for Galerkin, but with significantly smaller errors, especially for $K$1. The ellipsoidal and hollow-ball tests show the same pattern: near-optimal rates with improved accuracy relative to the do-nothing strategy, and no significant order loss from the inner polyhedral approximation. Stability with respect to small $K$2 is also reported down to $K$3 on fixed meshes (Ruas, 24 Feb 2026).

4. Unfitted interface formulations on cut elements

In unfitted interface problems, the immersed RT construction is local to elements cut by the interface. For a polygonal domain $K$4 split by a smooth closed interface $K$5 into $K$6 and $K$7, one writes the mixed system for $K$8 as

$K$9

with interface conditions

$\Gamma$00

where $\Gamma$01 and $\Gamma$02 is the unit tangent (Ji, 2021).

On an interface triangle $\Gamma$03, the local immersed space $\Gamma$04 is defined by piecewise RT fields

$\Gamma$05

subject to three interface constraints: continuity of $\Gamma$06 along the interface chord $\Gamma$07, continuity of $\Gamma$08 at one point $\Gamma$09, and equality of the divergence on both sides. The degrees of freedom remain the three standard edge-normal fluxes

$\Gamma$10

Under the maximum-angle condition $\Gamma$11, these DOFs are unisolvent. The construction reduces to standard RT when the coefficient jump disappears (Ji, 2021).

A central property is the commuting relation

$\Gamma$12

which holds elementwise for the immersed interpolant. Together with piecewise $\Gamma$13 regularity assumptions, it yields the approximation estimates

$\Gamma$14

and

$\Gamma$15

Because the global space $\Gamma$16 is generally nonconforming in $\Gamma$17, a symmetric interior penalty on interface edges is added: $\Gamma$18 The resulting lowest-order immersed mixed method is stable, and for $\Gamma$19 it satisfies an optimal first-order a priori estimate for $\Gamma$20. The same source emphasizes that the penalty is not needed for stability but is needed for optimal convergence; $\Gamma$21 yields suboptimal rates (Ji, 2021).

A related but distinct use appears in conservative flux reconstruction for CutFEM interface discretizations. There the local immersed space on a cut triangle $\Gamma$22 is

$\Gamma$23

while on uncut cells one uses standard $\Gamma$24 (Capatina et al., 2 Apr 2026). The three outer-edge flux moments again determine the local field uniquely. The defining feature here is that the transmission condition $\Gamma$25 is built into the space, so local conservation on cut cells is automatic. For the reconstructed flux $\Gamma$26, one obtains the elementwise relation

$\Gamma$27

The CutFEM paper contrasts this with a naive reconstruction using independent RT fields on $\Gamma$28 and $\Gamma$29, which generally fails to enforce

$\Gamma$30

on the interface. The immersed reconstruction is then used to define a posteriori indicators

$\Gamma$31

together with additional cut-face and interface-jump contributions, and the reliability estimate

$\Gamma$32

with $\Gamma$33 independent of $\Gamma$34, the cut configuration, and the coefficient contrast. The sharpness of the bound relies on the immersed RT space enforcing strong continuity of the normal flux across $\Gamma$35 and exact local conservation on cut cells (Capatina et al., 2 Apr 2026).

5. Raviart–Thomas enrichment of Scott–Vogelius spaces

In the Stokes discretization literature, the RT component is not a standalone flux space for an elliptic mixed problem but an enrichment of the Scott–Vogelius velocity space on arbitrary simplicial meshes. The basic pair is

$\Gamma$36

with $\Gamma$37, so that a stable scheme would yield exactly divergence-free discrete velocities. On general meshes, however, inf–sup stability may fail (John et al., 2022).

The proposed remedy enriches $\Gamma$38 with a small RT bubble space: $\Gamma$39 The enrichment is designed so that

$\Gamma$40

with $\Gamma$41 chosen from interior RT bubbles or, in low order, from $\Gamma$42 plus interior RT bubbles. For $\Gamma$43, one takes

$\Gamma$44

and the divergence map from $\Gamma$45 to $\Gamma$46 is bijective, making the method parameter-free (John et al., 2022).

The explicit constructions are local. On a simplex $\Gamma$47 with barycentric coordinates $\Gamma$48, the lowest-order RT face functions are

$\Gamma$49

and interior bubbles are formed as

$\Gamma$50

Higher-order explicit bubble systems are also given. In 2D, for example,

$\Gamma$51

supplemented by an additional mixed bubble for $\Gamma$52. In 3D, pairwise combinations

$\Gamma$53

provide the required divergence moments (John et al., 2022).

The discrete bilinear form couples the $\Gamma$54- and $\Gamma$55-parts: $\Gamma$56 with $\Gamma$57 present only when $\Gamma$58. A Fortin operator combining an $\Gamma$59-stable projector on $\Gamma$60 with an $\Gamma$61-reconstruction proves the discrete inf–sup condition

$\Gamma$62

with $\Gamma$63 independent of $\Gamma$64. The full discrete velocity is exactly divergence-free, because the RT correction $\Gamma$65 repairs the non-solenoidal part of the $\Gamma$66-conforming velocity (John et al., 2022).

The analysis yields a pressure-robust velocity bound independent of the pressure: $\Gamma$67 and the pressure estimate

$\Gamma$68

For $\Gamma$69, the RT-bubble and higher pressure modes can be eliminated locally, producing a reduced $\Gamma$70 scheme. The reduced system has the form

$\Gamma$71

with $\Gamma$72. Numerical studies in 2D and 3D confirm optimal $\Gamma$73-velocity and $\Gamma$74-pressure convergence, machine-precision divergence, and identical discrete solutions for the full and reduced schemes in the cases reported (John et al., 2022).

6. Conservation properties, implementation patterns, and conceptual distinctions

Across these variants, the principal invariant is flux structure. In curved-domain formulations, the key issue is where boundary-normal flux DOFs are imposed. The immersed strategy uses true normals $\Gamma$75 at shifted points on $\Gamma$76, while the test space still uses polyhedral normals $\Gamma$77 on $\Gamma$78. This preserves elementwise RT conformity and avoids nonaffine geometry maps (Ruas, 24 Feb 2026). In interface IRT methods, the key issue is internal transmission: continuity of normal flux across $\Gamma$79 is imposed inside cut elements, together with a tangential constitutive condition such as $\Gamma$80 or its $\Gamma$81-weighted analog (Ji, 2021, Capatina et al., 2 Apr 2026). In Scott–Vogelius enrichment, the key issue is divergence-surjectivity onto the pressure space, achieved by adding an $\Gamma$82-conforming correction with explicitly constructed RT bubbles (John et al., 2022).

Several misconceptions are addressed directly by the literature. First, immersed RT is not synonymous with unfitted interface discretization; in the Scott–Vogelius setting the term refers to enrichment of a conforming velocity space, not to geometry immersion (John et al., 2022). Second, immersed RT spaces are not uniformly $\Gamma$83-conforming across all formulations. The curved-domain trial space $\Gamma$84 and the RT-bubble enrichment are $\Gamma$85-conforming on the mesh, whereas the mixed immersed finite element space $\Gamma$86 for interface problems is generally nonconforming in $\Gamma$87 and requires interface-edge penalization for optimal convergence (Ji, 2021). Third, the use of straight simplices does not by itself imply an accuracy loss, provided the geometry-sensitive RT DOFs are imposed in a way that respects the true boundary or interface (Ruas, 24 Feb 2026, Capatina et al., 2 Apr 2026).

The implementation patterns are correspondingly different. Curved-boundary immersion requires half-line intersections $\Gamma$88, true normal evaluation $\Gamma$89, and quadrature of boundary terms on the true curved boundary; for $\Gamma$90, static condensation and conjugate gradients are used to keep the symmetric system computationally manageable (Ruas, 24 Feb 2026). Interface IRT on unfitted meshes requires geometric reconstruction of cut cells, subcell quadrature, and either penalty terms on interface edges or multiplier-based flux recovery, depending on whether the goal is a primal mixed approximation or equilibrated flux reconstruction (Ji, 2021, Capatina et al., 2 Apr 2026). RT enrichment for Scott–Vogelius requires local reference-element bubble bases, Piola scaling, and local solves for the divergence inverse $\Gamma$91, after which the global problem may be reduced to a $\Gamma$92 system (John et al., 2022).

The reported limitations are likewise formulation-specific. The curved-domain Petrov–Galerkin method assumes smooth $\Gamma$93, sufficiently fine meshes, accurate quadrature for face DOFs and curved-boundary integrals, and reliable construction of the shifted points $\Gamma$94; extremely strong curvature may challenge the half-line intersection procedure, and for $\Gamma$95 the way the additive constant of $\Gamma$96 is fixed can affect coarse-mesh convergence (Ruas, 24 Feb 2026). The interface IFE analysis assumes a maximum-angle condition on interface triangles and notes that the edge-penalty term is essential for optimal order (Ji, 2021). The CutFEM reconstruction relies on ghost penalties to stabilize small cut fractions and on a straight-segment approximation $\Gamma$97 at the element level (Capatina et al., 2 Apr 2026). The Scott–Vogelius enrichment is parameter-free only for $\Gamma$98; when $\Gamma$99, a mild stabilization on the lowest-order $\Gamma_h$00 part is required (John et al., 2022).

Taken together, these developments show that immersed Raviart–Thomas constructions are best understood as RT-preserving modifications targeted at specific structural defects of standard discretizations: geometry mismatch at curved Neumann boundaries, missing transmission conditions on cut cells, or missing divergence control in high-order incompressible flow discretizations. The precise mathematical object called an immersed Raviart–Thomas space therefore depends on which defect is being corrected.