Two-Grid Weak Galerkin Methods

• Two-grid weak Galerkin methods are numerical techniques that combine WG discretizations with a coarse–fine splitting to reduce computational cost while preserving accuracy.
• They are applied to nonlinear, eigenvalue, and diffusion problems using varied WG formulations and stabilization strategies to achieve optimal convergence rates.
• Key insights include optimal mesh coupling, rigorous error estimates, and lower-bound preservation for eigenvalues, ensuring robust performance across different applications.

The two-grid weak Galerkin method denotes a family of algorithms that combine weak Galerkin (WG) discretization with a coarse–fine decomposition in order to reduce the cost of nonlinear, eigenvalue, or linear algebraic computations while retaining the structural advantages of WG formulations. In the arXiv literature, the term covers at least three technically distinct constructions: a coarse nonlinear solve followed by a fine linearized WG solve for quasi-linear elliptic equations of non-monotone type, a coarse WG eigenproblem followed by a fine WG source problem and Rayleigh quotient for eigenvalue approximation, and a two-level auxiliary-space preconditioner for linear WG systems (Zhu et al., 2022, Zhai et al., 2017, Lu et al., 4 Aug 2025, Li et al., 2014, Chen et al., 2014).

1. Conceptual scope

Within WG theory, “two-grid” does not refer to a single canonical algorithm. The common principle is that the expensive part of the computation is confined to a coarse space, while a fine-space correction recovers most of the accuracy otherwise obtained by a full fine-grid solve.

Variant	Coarse component	Fine component
Nonlinear elliptic solver	Nonlinear WG solve on mesh $\mathcal{T}_\tau$	Linearized WG solve on $\mathcal{T}_h$
Eigenvalue acceleration	WG eigenproblem on mesh $\mathcal{T}_H$	WG linear source problem on $\mathcal{T}_h$ plus Rayleigh quotient
Auxiliary-space two-level method	Conforming $P_1$ auxiliary problem	Smoothing and coarse correction for the fine WG system

A recurrent misconception is that two-grid WG always means multigrid preconditioning. The literature shows a sharper distinction. In the nonlinear and eigenvalue settings, two-grid is an approximation strategy for the discrete PDE or eigenproblem itself; in the diffusion literature, it is a solver or preconditioner for the linear WG system generated on the fine mesh (Zhu et al., 2022, Zhai et al., 2017, Lu et al., 4 Aug 2025, Li et al., 2014, Chen et al., 2014).

Another important distinction concerns the role of the coarse space. In the nonlinear and eigenvalue papers, the coarse space is associated with a genuinely coarser mesh with sizes $\tau>h$ or $H>h$. In the auxiliary-space preconditioning papers, the “coarse” space is often the conforming piecewise linear space on the same mesh, used as an auxiliary solver space rather than as a separate physical discretization level (Chen et al., 2014, Li et al., 2014).

2. Weak Galerkin structure

The WG framework replaces strong continuity requirements by weakly defined differential operators acting on pairs of interior and boundary unknowns. A generic weak function has the form $v=\{v_0,v_b\}$, where $v_0$ is an element-interior polynomial and $v_b$ is an edge or face polynomial. In stabilized polytopal formulations for quasi-linear elliptic and eigenvalue problems, one commonly uses spaces such as

$\mathcal{T}_h$0

with homogeneous boundary conditions enforced through the trace component. In RT/BDM-based diffusion formulations, the local choices are instead expressed through triples $\mathcal{T}_h$1, where the weak gradient lives in Raviart–Thomas or Brezzi–Douglas–Marini spaces (Zhu et al., 2022, Zhai et al., 2017, Li et al., 2014).

The core operator is the discrete weak gradient. In several of the cited formulations it is defined locally by

$\mathcal{T}_h$2

for all local polynomial test vectors $\mathcal{T}_h$3. For elasticity, the same pattern yields weak strain and weak divergence through $\mathcal{T}_h$4 and an elementwise weak divergence operator (Lu et al., 4 Aug 2025).

Stabilization is formulation-dependent. For the non-monotone quasi-linear elliptic problem, the stabilization is

$\mathcal{T}_h$5

while the Laplace and elasticity eigenvalue papers use weighted stabilizers of the form $\mathcal{T}_h$6 or $\mathcal{T}_h$7 on the boundary mismatch. By contrast, the auxiliary-space preconditioning paper for second-order diffusion studies a WG bilinear form without an additional stabilization term, namely

$\mathcal{T}_h$8

This difference is structural rather than cosmetic: the particular stabilization enters both the spectral theory and the two-grid error analysis (Zhu et al., 2022, Zhai et al., 2017, Lu et al., 4 Aug 2025, Chen et al., 2014).

Several norm equivalences underwrite the theory. In the nonlinear setting, the mesh-dependent seminorm

$\mathcal{T}_h$9

is equivalent to the broken $\mathcal{T}_H$0-like norm $\mathcal{T}_H$1. In the diffusion-preconditioning literature, one similarly has $\mathcal{T}_H$2, together with inverse and discrete Poincaré inequalities. These identities are the bridge between WG trace variables and classical energy estimates (Zhu et al., 2022, Chen et al., 2014).

3. Two-grid WG for quasi-linear elliptic problems of non-monotone type

For the quasi-linear model

$\mathcal{T}_H$3

the coefficient satisfies $\mathcal{T}_H$4 with bounded derivatives up to second order and

$\mathcal{T}_H$5

The weak formulation is to find $\mathcal{T}_H$6 such that

$\mathcal{T}_H$7

The corresponding WG scheme seeks $\mathcal{T}_H$8 with $\mathcal{T}_H$9 on $\mathcal{T}_h$0 and

$\mathcal{T}_h$1

where

$\mathcal{T}_h$2

The existence theory is based on Brouwer’s fixed point theorem applied to a map constructed from the discrete Fréchet derivative of the nonlinear operator. A discrete Gårding inequality for the linearized bilinear form $\mathcal{T}_h$3 provides the stability input, and the fixed-point map is shown to map a closed convex ball centered at $\mathcal{T}_h$4 into itself; for sufficiently small $\mathcal{T}_h$5 it is also a contraction, which yields a unique WG solution in that ball (Zhu et al., 2022).

The single-grid error analysis gives

$\mathcal{T}_h$6

and

$\mathcal{T}_h$7

These estimates use the projection commutativity relation $\mathcal{T}_h$8, explicit consistency terms, and nonlinear remainder bounds derived from averaged derivatives $\mathcal{T}_h$9 and $P_1$0 (Zhu et al., 2022).

The two-grid algorithm introduces a coarse mesh $P_1$1 and a fine mesh $P_1$2 with $P_1$3. First, one solves the full nonlinear WG problem on the coarse grid: $P_1$4 Second, one freezes the coefficient at the coarse solution and solves a single linear WG problem on the fine grid: $P_1$5 The output is $P_1$6. The main energy-norm estimate is

$P_1$7

with $P_1$8 depending on $P_1$9 and $\tau>h$0, but not on $\tau>h$1 or $\tau>h$2. For $\tau>h$3, the choice $\tau>h$4 yields the optimal rate in the energy-like norm (Zhu et al., 2022).

The analysis is driven by a consistency identity comparing the fine-grid nonlinear solution $\tau>h$5 and the two-grid solution $\tau>h$6, together with a splitting

$\tau>h$7

After Taylor expansion and use of the single-grid $\tau>h$8 estimates, the coarse-grid error contributes at order $\tau>h$9, while the genuine fine-grid discretization contributes at order $H>h$0 (Zhu et al., 2022).

4. Two-grid WG for eigenvalue problems

For eigenvalue problems, the two-grid WG method replaces a fine-grid generalized eigenproblem by a coarse-grid eigenproblem and a fine-grid linear correction. This strategy appears in both Laplace and linear elasticity settings.

For the Dirichlet Laplace eigenproblem, Zhai, Xie, Zhang, and Zhang consider a WG discretization with

$H>h$1

where

$H>h$2

The standard WG eigenproblem seeks $H>h$3 with $H>h$4 such that

$H>h$5

Its single-grid approximation is asymptotically a lower bound: for sufficiently small $H>h$6,

$H>h$7

and

$H>h$8

The two-grid algorithm computes $H>h$9 on $v=\{v_0,v_b\}$0, solves

$v=\{v_0,v_b\}$1

and then sets

$v=\{v_0,v_b\}$2

With

$v=\{v_0,v_b\}$3

one obtains

$v=\{v_0,v_b\}$4

For $v=\{v_0,v_b\}$5 and the favorable dual-regularity regime, this yields the characteristic doubling from a coarse-grid eigenvalue error of order $v=\{v_0,v_b\}$6 to a two-grid contribution of order $v=\{v_0,v_b\}$7. A sufficient condition for preserving the asymptotic lower-bound property is

$v=\{v_0,v_b\}$8

although the reported computations show that lower bounds may persist even when this condition is not strictly enforced (Zhai et al., 2017).

The elasticity eigenvalue version developed by Lu and Zhai follows the same template but uses weak strain and weak divergence: $v=\{v_0,v_b\}$9 with

$v_0$0

After solving the coarse WG eigenproblem and the fine WG linear problem

$v_0$1

the fine-grid Rayleigh quotient defines $v_0$2. Writing

$v_0$3

the theory gives

$v_0$4

The scaling $v_0$5, with the examples using $v_0$6, is the practical regime identified for the “doubling” effect. Lower-bound preservation is obtained if

$v_0$7

which ensures $v_0$8 for sufficiently small $v_0$9 (Lu et al., 4 Aug 2025).

In both eigenvalue strands, the decisive observation is that the fine-grid computation is no longer a generalized eigenproblem but a symmetric positive definite linear solve driven by coarse spectral data. This is the principal source of the acceleration (Zhai et al., 2017, Lu et al., 4 Aug 2025).

5. Auxiliary-space two-level algorithms and preconditioners

A distinct branch of the literature interprets two-grid WG as a preconditioning framework for linear systems arising from WG discretizations of diffusion equations. Here the objective is not a new discrete approximation but rapid solution of the fine-grid WG algebraic system.

Li and Xie analyze a two-level algorithm for the WG discretization of

$v_b$0

on quasi-uniform simplicial meshes. The WG bilinear form is

$v_b$1

with block operator representation

$v_b$2

They prove the norm equivalence

$v_b$3

and consequently

$v_b$4

The auxiliary space is the conforming piecewise linear space

$v_b$5

with prolongation $v_b$6 defined by element and face moment matching. A two-level preconditioner $v_b$7 is constructed from pre-smoothing, left coarse-grid correction, right coarse-grid correction, and post-smoothing. Its analysis rests on an extended Xu–Zikatanov identity,

$v_b$8

which yields a mesh-independent contraction bound and does so without any elliptic regularity assumption (Li et al., 2014).

Chen, Wang, Wang, and Ye develop an auxiliary-space multigrid preconditioner for a WG method for second-order diffusion equations. Their fine WG space is

$v_b$9

with $\mathcal{T}_h$00 or $\mathcal{T}_h$01, and the auxiliary space is the conforming piecewise linear space

$\mathcal{T}_h$02

The prolongation is the $\mathcal{T}_h$03 projection/inclusion $\mathcal{T}_h$04, and the multiplicative two-grid preconditioner is characterized by

$\mathcal{T}_h$05

Under smoother bounds, coarse-solver bounds, prolongation stability, and the existence of a stable quasi-interpolant $\mathcal{T}_h$06, the preconditioned operator satisfies

$\mathcal{T}_h$07

The paper also derives a reduced edge/face system by eliminating interior unknowns and proves the same type of $\mathcal{T}_h$08-independent condition-number bound for the reduced operator $\mathcal{T}_h$09 (Chen et al., 2014).

The conceptual difference between these preconditioning papers and the approximation-oriented two-grid papers is substantial. In the preconditioning setting, the coarse space is part of an auxiliary-space decomposition of a single fine-grid operator; in the nonlinear and eigenvalue settings, coarse and fine meshes define distinct discrete problems whose solutions are combined to approximate the PDE or spectrum itself (Li et al., 2014, Chen et al., 2014).

6. Numerical behavior and methodological significance

The numerical literature consistently reports that the two-grid strategy preserves much of fine-grid WG accuracy while reducing either nonlinear/eigenvalue cost or Krylov iteration complexity.

Setting	Representative observation	Source
Quasi-linear elliptic, $\mathcal{T}_h$10	At $\mathcal{T}_h$11 on rectangular grids, single-grid WG took $\mathcal{T}_h$12 s with $\mathcal{T}_h$13; two-grid WG took $\mathcal{T}_h$14 s with $\mathcal{T}_h$15	(Zhu et al., 2022)
Auxiliary-space preconditioner	PCG iterations were essentially independent of $\mathcal{T}_h$16; in 2D unit-disk tests, the original WG system required about $\mathcal{T}_h$17 iterations and the reduced system about $\mathcal{T}_h$18	(Chen et al., 2014)
Two-level diffusion solver	With a multigrid coarse solver, average error-reduction factors per iteration were about $\mathcal{T}_h$19, $\mathcal{T}_h$20, and $\mathcal{T}_h$21 for $\mathcal{T}_h$22 smoothing steps	(Li et al., 2014)
Laplace eigenproblem	For $\mathcal{T}_h$23, the error $\mathcal{T}_h$24 decayed from $\mathcal{T}_h$25 to $\mathcal{T}_h$26 to $\mathcal{T}_h$27, with observed order about $\mathcal{T}_h$28	(Zhai et al., 2017)
Elasticity eigenproblem	For two-grid WG, reported eigenfrequency orders were around $\mathcal{T}_h$29–$\mathcal{T}_h$30 for $\mathcal{T}_h$31 and about $\mathcal{T}_h$32 for $\mathcal{T}_h$33, with lower bounds observed when $\mathcal{T}_h$34	(Lu et al., 4 Aug 2025)

Several conclusions are common to these results. First, the WG weak-gradient machinery is compatible with coarse–fine acceleration without sacrificing the discrete structure that supports stability and error estimates. Second, the most favorable mesh couplings are problem-dependent: $\mathcal{T}_h$35 is singled out for the $\mathcal{T}_h$36 quasi-linear elliptic case, whereas $\mathcal{T}_h$37 is the standard coupling in the eigenvalue literature (Zhu et al., 2022, Zhai et al., 2017, Lu et al., 4 Aug 2025). Third, lower-bound preservation for eigenvalues is not automatic; it requires a quantitative relation between the coarse-grid correction error and the fine-grid lower-bound error, even though computations often remain lower-bounding outside the strict sufficient regime (Zhai et al., 2017, Lu et al., 4 Aug 2025).

A further methodological point is that WG formulations themselves are not uniform across applications. The cited papers use different polynomial pairings, different stabilizers, and in the diffusion-preconditioning setting even a formulation without an extra stabilization term. Two-grid analysis is therefore inseparable from the particular WG discretization chosen (Chen et al., 2014, Li et al., 2014, Zhu et al., 2022).

Taken together, these works establish the two-grid weak Galerkin method as a broad computational paradigm rather than a single algorithm. Its unifying feature is the use of WG trial spaces and weak differential operators together with a coarse–fine decomposition that transfers the dominant computational burden away from the full fine-grid problem, while rigorous theory quantifies when that transfer preserves optimal convergence, asymptotic lower bounds, or mesh-independent conditioning (Zhu et al., 2022, Zhai et al., 2017, Lu et al., 4 Aug 2025, Li et al., 2014, Chen et al., 2014).