Reduced Oseen Discretization Techniques

• Reduced Oseen discretization is a formulation that eliminates pressure and auxiliary variables via static condensation to produce a velocity-only global system.
• It leverages an embedded Trefftz-DG method with local inversion to ensure coercivity, inf-sup stability, and optimal convergence in steady Oseen and Navier-Stokes problems.
• By separating global and local modes, the approach enhances computational efficiency and robustness in simulating incompressible flow phenomena.

Searching arXiv for recent work on reduced Oseen discretization and closely related Oseen elimination formulations. Reduced Oseen discretization denotes a class of formulations for the steady Oseen problem in which pressure variables, or other local auxiliary variables attached to the PDE residual, are eliminated from the global algebraic system so that the final solve is posed in terms of velocity unknowns only. In the embedded Trefftz-DG construction for the Oseen problem, the reduced formulation is obtained from a full discontinuous Galerkin system by local inversion and static condensation; the resulting system is “posed in terms of the velocity unknown only,” and this is identified as “a crucial step in the analysis especially for the nonlinear Navier-Stokes problem in Part II” (Stocker et al., 11 Jun 2026). Closely related Oseen discretizations realize pressure elimination differently, including a pseudostress reformulation in mixed virtual elements and exactly divergence-free H(div)-conforming mixed methods whose velocity estimates are pressure-robust (Lepe et al., 27 Jan 2026, Zhang et al., 27 Nov 2025).

1. Continuous Oseen setting

In the steady Oseen model underlying the reduced Trefftz-DG discretization, one seeks $(u,p)$ satisfying

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$

$\nabla\cdot u = 0 \quad \text{in }\Omega,$

$u = 0 \quad \text{on }\partial\Omega,$

with $\Omega\subset\mathbb{R}^d$ $(d=2,3)$, viscosity $\nu>0$, given convective field $w$, and body force $f$. The weak form is: find $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$ such that for all $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$0,

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$1

where

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$2

This is the linearized incompressible flow problem to which the reduced discretization is applied (Stocker et al., 11 Jun 2026).

Related Oseen formulations in the recent literature enlarge this baseline operator. One steady Oseen model adds an advection field $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$3 and a reaction coefficient $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$4,

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$5

under the condition $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$6 on $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$7 (Zhang et al., 27 Nov 2025). A two-dimensional generalized Oseen problem further includes a permeability term $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$8,

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$9

together with $\nabla\cdot u = 0 \quad \text{in }\Omega,$0, homogeneous Dirichlet data, and $\nabla\cdot u = 0 \quad \text{in }\Omega,$1 (Lepe et al., 27 Jan 2026). These variants clarify that “reduced” refers to the discretization strategy rather than to a simplification of the continuous model.

2. Trefftz-DG spaces and local complement construction

The embedded Trefftz-DG method is built on a shape-regular simplicial mesh $\nabla\cdot u = 0 \quad \text{in }\Omega,$2 of $\nabla\cdot u = 0 \quad \text{in }\Omega,$3, with elementwise diameter $\nabla\cdot u = 0 \quad \text{in }\Omega,$4, global $\nabla\cdot u = 0 \quad \text{in }\Omega,$5, and skeleton $\nabla\cdot u = 0 \quad \text{in }\Omega,$6. The ambient discrete space is

$\nabla\cdot u = 0 \quad \text{in }\Omega,$7

while the test-side “PDE-residual” space is

$\nabla\cdot u = 0 \quad \text{in }\Omega,$8

On each element $\nabla\cdot u = 0 \quad \text{in }\Omega,$9, the local PDE operator $u = 0 \quad \text{on }\partial\Omega,$0 is defined by

$u = 0 \quad \text{on }\partial\Omega,$1

The Trefftz subspace is then

$u = 0 \quad \text{on }\partial\Omega,$2

It is a kernel-in-a-relaxed-sense space: the Oseen operator and divergence constraint vanish against the residual test space on each cell (Stocker et al., 11 Jun 2026).

A second ingredient is a local complement $u = 0 \quad \text{on }\partial\Omega,$3 with $u = 0 \quad \text{on }\partial\Omega,$4. Its pressure part is

$u = 0 \quad \text{on }\partial\Omega,$5

Its velocity part is chosen as

$u = 0 \quad \text{on }\partial\Omega,$6

subject to zero-mean normal flux on $u = 0 \quad \text{on }\partial\Omega,$7 of the $u = 0 \quad \text{on }\partial\Omega,$8 faces, where $u = 0 \quad \text{on }\partial\Omega,$9. One checks

$\Omega\subset\mathbb{R}^d$0

and, under a mild resolution assumption $\Omega\subset\mathbb{R}^d$1, the local operator $\Omega\subset\mathbb{R}^d$2 is bijective on $\Omega\subset\mathbb{R}^d$3 with stable inverse (Stocker et al., 11 Jun 2026).

This decomposition is the structural basis of the reduction. The Trefftz part carries the globally coupled information, while the complement isolates those local modes on which the residual operator is stably invertible. A plausible implication is that the algebraic elimination is not an ad hoc post-processing step but is encoded in the space design itself.

3. Full embedded Trefftz-DG variational problem

The full formulation uses the usual SIPG-Stokes forms on $\Omega\subset\mathbb{R}^d$4. The viscous form is

$\Omega\subset\mathbb{R}^d$5

and the discrete divergence-pressure coupling is

$\Omega\subset\mathbb{R}^d$6

For convection one takes “an upwind- or Temam-modified convection trilinear $\Omega\subset\mathbb{R}^d$7 satisfying continuity (2.21) and $\Omega\subset\mathbb{R}^d$8.” The global bilinear form is

$\Omega\subset\mathbb{R}^d$9

The Trefftz-DG method reads: find $(d=2,3)$0 such that

$(d=2,3)$1

for all $(d=2,3)$2 and $(d=2,3)$3 (Stocker et al., 11 Jun 2026).

Algebraically, this is “an upper-block-triangular system coupling the global Trefftz unknowns with local residual unknowns.” That structure is fundamental: the PDE-residual block is local, whereas the Trefftz block is the part that remains globally assembled. The reduced Oseen discretization is obtained by exploiting exactly this block form.

4. Pressure elimination and static condensation

The reduction proceeds from the splitting $(d=2,3)$4 together with invertibility of $(d=2,3)$5. On each element $(d=2,3)$6, one solves the local problem for $(d=2,3)$7 satisfying

$(d=2,3)$8

After this local solve, one substitutes back into the global equation on $(d=2,3)$9 only. In the formulation summarized in the paper, this “removes all high-order pressure unknowns and the local flux degrees of freedom” (Stocker et al., 11 Jun 2026).

In practice, the elementwise matrix representation of $\nu>0$0 is assembled and inverted, “e.g. via QR or SVD,” to build the static-condensation Schur complement. Writing

$\nu>0$1

the condensed contribution to the Trefftz block is

$\nu>0$2

Here $\nu>0$3 is the discrete pressure lifting $\nu>0$4, so that $\nu>0$5 (Stocker et al., 11 Jun 2026).

The reduction therefore has two simultaneous meanings. First, it is a pressure elimination in the algebraic sense: the global system no longer carries the full pressure block. Second, it is a localization mechanism: the residual-driven complement variables are resolved cellwise and never appear as independent global unknowns.

5. Velocity-only reduced system and analytical properties

The reduced unknown is

$\nu>0$6

Imposing discrete $\nu>0$7 against $\nu>0$8 gives a subspace of $\nu>0$9. On that reduced space, the formulation establishes three key properties: coercivity,

$w$0

continuity,

$w$1

and inf-sup on the residual block $w$2, expressed as a stable pairing $w$3. Consequently, “the fully reduced bilinear form” on the reduced velocity space and local complement is inf-sup stable (Stocker et al., 11 Jun 2026).

After condensation, the matrix system has the form

$w$4

where

$w$5

plus facet terms at $w$6, and

$w$7

The paper characterizes this as “a global SPD-like system for the DOFs of $w$8” (Stocker et al., 11 Jun 2026).

The analytical results are stated in DG norms. With $w$9 denoting the DG norm $f$0, Theorem 3.2 gives local invertibility: under $f$1 small, $f$2 is bijective with

$f$3

Theorem 4.3 states that the coupled system is inf-sup stable on the full space $f$4. Theorem 5.1 gives quasi-optimal error in DG norm: if $f$5 is the exact solution with $f$6, $f$7, and $f$8 is the reduced DG solution, then

$f$9

Using a BDM-interpolant, the corollary recovers the optimal $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$0–$(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$1 convergence (Stocker et al., 11 Jun 2026).

6. Assumptions, implementation, and related pressure-eliminating Oseen formulations

The reduced Trefftz-DG construction assumes a shape-regular simplicial mesh $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$2, polynomial degree $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$3, and a penalty parameter with $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$4 for coercivity of $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$5. The convection form may be “any standard upwind or Temam-modified DG form satisfying (2.21),(2.22),(2.23).” The convective field satisfies

$(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$6

with the resolution bound

$(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$7

All constants in the estimates depend only on shape-regularity, $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$8, and $(u,p)\in [H_0^1(\Omega)]^d\times L_0^2(\Omega)$9, not on $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$00 or $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$01. The implementation is correspondingly local-to-global: on each $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$02, assemble the small local matrix of $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$03, compute its inverse or a QR/SVD-based pseudoinverse, build the condensed element stiffness $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$04 on $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$05, assemble the global reduced system for $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$06, and recover local complements $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$07 by back-substitution if needed (Stocker et al., 11 Jun 2026).

Other recent Oseen discretizations show that pressure elimination is not unique to Trefftz condensation. In a mixed virtual element method for the two-dimensional generalized Oseen problem, the additional unknown is the pseudostress

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$08

which yields

$$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$09

The pressure is therefore eliminated from the first-order system and “recovered element-wise” after solving for $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$10 (Lepe et al., 27 Jan 2026). In the H(div)-conforming finite element method for the steady Oseen equations, the discrete velocity is “exactly divergence-free pointwise,” the pressure enters “only via the constraint and not in the residual or stabilization,” and the velocity estimate contains “no factor $$-\nu \Delta u + (w\cdot \nabla)u + \nabla p = f \quad \text{in }\Omega,$$11 in front of the data error” (Zhang et al., 27 Nov 2025). This suggests that reduced Oseen discretization is a broader design principle: the global approximation may be organized around velocity-dominant unknowns even when the underlying formulations differ substantially.

A common misconception is that a reduced formulation discards pressure. The cited methods indicate a narrower and more precise statement. In the embedded Trefftz-DG setting, the global reduced system is velocity-only, but the pressure is still represented through the discrete pressure lifting and through local back-substitution. In the pseudostress VEM setting, the pressure is recovered from the trace relation. In the H(div) setting, pressure remains part of the mixed problem, but velocity approximation is decoupled from the discrete inf-sup constant in the stated pressure-robust estimate. The common theme is not the absence of pressure, but its displacement from the dominant global approximation mechanism.