🔢 mathematics

Equivalence of mixed and nonconforming methods on general polytopal partitions. Part I: Multiscale and projection methods

This paper establishes the equivalence between mixed and nonconforming methods for variable diffusion problems on general polytopal partitions by analyzing multiscale approaches and providing a practical criterion for the well-posedness and equivalence of projection methods.

Original authors: Simon Lemaire

Published 2026-02-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Simon Lemaire

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to figure out how water flows through a complex, rocky underground aquifer. The rocks (the "polytopal partitions") are irregular shapes—some are triangles, some are weird polygons, and they don't fit together perfectly like a Lego set. You want to know two things:

The Pressure: How hard is the water pushing at any given point? (This is the "Potential").
The Flow: Which way is the water moving, and how fast? (This is the "Flux").

In the world of math and engineering, there are two main ways to solve this puzzle. One way focuses on the Pressure first (Primal method), and the other focuses on the Flow first (Mixed method). For a long time, mathematicians argued about which way was better, or if they even gave the same answer.

This paper, written by Simon Lemaire, is like a master translator. It proves that, under the right conditions, these two completely different ways of solving the puzzle are actually equivalent. They are just two different languages describing the exact same reality.

Here is a breakdown of the paper's main ideas using simple analogies:

1. The Two Languages (Primal vs. Mixed)

Think of the Primal method as a team of architects who only care about the blueprint of the pressure. They build a model where the pressure is smooth and continuous, but they have to guess the flow later.
Think of the Mixed method as a team of hydraulic engineers who only care about the pipes (the flow). They ensure the water flows perfectly through the pipes, but the pressure is a bit "jagged" or broken between sections.

Usually, if you ask the architects and the engineers to solve the same problem, they give you slightly different numbers. This paper says: "Wait, if we build our models correctly, they will give you the exact same answer."

2. The "Hybrid" Trick (The Glue)

The secret sauce in this paper is something called Hybridization.
Imagine you have a room full of people (the cells of your mesh) who are all shouting different things.

The Old Way: You try to force everyone to agree on a single, perfect voice immediately. This is hard and computationally expensive.
The Hybrid Way: You put a "messenger" on the walls between the rooms. The people in the rooms talk to the messengers on the walls, and the messengers talk to each other.
This paper shows that whether you start with the "Architects" (Primal) or the "Engineers" (Mixed), if you use these messengers correctly, you end up with the same final result. It's like showing that two different recipes for a cake (one starting with eggs, one starting with flour) result in the exact same delicious dessert.

3. The "Virtual" Construction (The Magic Blueprint)

The paper introduces a "Virtual Construction." Think of this as a perfect, invisible blueprint that exists in a parallel universe.

In this perfect world, the math works flawlessly. The pressure and flow are perfectly linked.
The problem is, we can't actually build this perfect blueprint in the real world because it involves shapes and functions that are too complex to calculate on a computer.
The Breakthrough: The author shows that even though we can't build the perfect blueprint, we can build a "shadow" of it. By using a technique called Projection, we can take our messy, real-world calculations and project them onto this perfect blueprint to get the right answer.

4. The "Silver Bullet" (When it's Easy)

The paper finds one special situation where everything is easy. If your rocks are simple shapes (like triangles or rectangles) and the ground is uniform, the "Virtual Blueprint" actually becomes a real, buildable object.

This is the Raviart-Thomas method (a famous math tool).
The paper proves that this famous tool is mathematically identical to a "bubble-enriched" version of a simpler method (Crouzeix-Raviart).
Analogy: It's like discovering that a high-tech, expensive Ferrari engine is actually just a very cleverly tuned version of a standard bicycle engine. They are different, but they do the exact same job with the same efficiency.

5. The "Projection" Method (The Shortcut)

What if your rocks are weird shapes and the ground is messy? You can't build the perfect blueprint.

Here, the paper introduces Projection Methods.
Analogy: Imagine you are trying to take a photo of a 3D object, but your camera is broken. Instead of taking a perfect photo, you take a shadow and use math to "project" what the object must look like based on that shadow.
The paper proves that even with this "shadow" approach, you still get the correct answer. It also gives a simple checklist (a "criterion") for engineers to see if their specific math setup will work or if it will fail.

Why Does This Matter?

For Engineers: It means they can choose the math method that is easiest for their computer to handle, knowing it will give the same result as the "gold standard" method.
For Scientists: It connects two different schools of thought (Multiscale methods and Projection methods) that were previously seen as separate. It's like realizing that two different tribes were actually speaking dialects of the same language.
For the Future: This is "Part I" of a series. This part sets up the theory. The next part will tackle even more complex, weird-shaped rocks (general polytopal elements), promising to make these simulations even more powerful for things like oil exploration, groundwater management, and climate modeling.

In a nutshell: This paper is a bridge. It connects two islands of mathematical thought, proving they are actually the same land, and gives us a map (the projection criterion) to travel between them safely, even when the terrain is rough and irregular.

1. Problem Statement

The paper investigates the mathematical equivalence between mixed finite element methods (flux-based) and nonconforming primal methods (potential-based) for variable diffusion problems on general polytopal partitions.

The underlying continuous problem (Problem 0) is the variable diffusion equation:
$\mathbf{b}^{-1}\boldsymbol{\sigma} - \nabla u = 0, \quad -\text{div}\,\boldsymbol{\sigma} = f, \quad u|_{\partial\Omega} = 0$
where $\boldsymbol{\sigma}$ is the flux, $u$ is the potential, and $\mathbf{b}$ is a symmetric, positive-definite mobility tensor field.

The core challenge addressed is that while hybridization (introducing Lagrange multipliers on element faces) is well-understood for conforming methods, establishing a rigorous equivalence between nonconforming primal schemes and mixed schemes on general polytopal meshes (especially with variable coefficients) is complex. Specifically, the paper asks:

Can we construct finite-dimensional spaces where primal and mixed formulations are equivalent?
How does this equivalence hold when the constitutive relation ( $\boldsymbol{\sigma} = \mathbf{b}\nabla u$ ) is preserved exactly versus when it is projected?
How do these theoretical constructs relate to practical multiscale and projection methods?

2. Methodology

The author employs a virtual space construction approach, moving from infinite-dimensional continuous analysis to finite-dimensional discrete analysis.

A. Continuous Level: Hybridization

The paper first establishes a "Rosetta stone" at the continuous level by introducing hybridized formulations:

Hybridized Primal (Problem 3): Relaxes $H^1_0(\Omega)$ conformity, introducing a flux trace on the skeleton.
Hybridized Mixed (Problem 4): Relaxes $\mathbf{H}(\text{div}; \Omega)$ conformity, introducing a potential trace on the skeleton.
Using Lagrange multiplier theory, the author proves that:

Primal $\iff$ Hybridized Primal
Mixed $\iff$ Hybridized Mixed
Crucially: Hybridized Primal $\implies$ Mixed and Hybridized Mixed $\implies$ Primal.
This establishes a complete equivalence diagram (Figure 1) linking all four formulations via the constitutive relation $\boldsymbol{\sigma} = \mathbf{b}\nabla u$ .

B. Discrete Level: Virtual Construction

The paper defines finite-dimensional counterparts using broken polynomial spaces and virtual element spaces:

Flux Space ( $\boldsymbol{\Sigma}_{\ell,k}$ ): Defined locally such that $\text{div}\,\boldsymbol{\tau} \in \mathbb{P}_{\ell-1}$ and normal traces are in $\mathbb{P}_{k-1}$ .
Potential Space ( $U_{\ell,k}$ ): Defined implicitly via the PDE: $v \in H^1(K)$ such that $\text{div}(\mathbf{b}\nabla v) \in \mathbb{P}_{\ell-1}$ and normal flux traces are in $\mathbb{P}_{k-1}$ .
Key Identity: The construction ensures $\boldsymbol{\Sigma}_{\ell,k}(K) = \mathbf{b}\nabla U_{\ell,k}(K)$ . This means the discrete constitutive relation is exactly preserved.

The author proves that the discrete versions of the hybridized problems are equivalent, thereby establishing discrete primal/mixed equivalence for these virtual spaces.

C. Practical Implementation: Projection Methods

Since the spaces $U_{\ell,k}$ are generally non-polynomial (virtual) and not directly computable, the paper investigates two scenarios:

Exact Preservation (Multiscale Methods): In specific cases (e.g., lowest-order Raviart-Thomas on simplices with cell-wise constant $\mathbf{b}$ ), the virtual space reduces to a known polynomial space enriched with "bubbles" (e.g., Crouzeix-Raviart enriched with quadratic bubbles). This recovers known equivalences (e.g., between MsHHO and MHM methods).
Projection (General Case): When the flux space is polynomial but the constitutive relation cannot be exactly preserved (i.e., $\mathbf{b}\nabla U \not\subset \boldsymbol{\Sigma}$ $b \nabla U \neq \subset Σ$ ), the author introduces Projection Methods.
- The primal problem is reformulated using a weighted orthogonal projection of the gradient onto the flux space.
- The resulting scheme (Problem 45) is a "projection finite element method" where the potential space may remain purely virtual (defined only by degrees of freedom), while the flux space is computable.

3. Key Contributions

Unified Equivalence Framework: The paper provides a rigorous proof of equivalence between primal and mixed methods on general polytopal meshes with variable diffusion coefficients. It unifies various multiscale methods (MsFE, MsHHO, MHM) under a single theoretical umbrella.
Virtual Space Construction: It introduces a specific finite-dimensional virtual construction ( $U_{\ell,k}$ and $\boldsymbol{\Sigma}_{\ell,k}$ ) that preserves the constitutive relation $\boldsymbol{\sigma} = \mathbf{b}\nabla u$ exactly. This bridges the gap between multiscale methods and classical finite elements.
Generalization of "Silver Bullet" Results: The paper clarifies that the equivalence between the lowest-order Raviart-Thomas (mixed) method and the bubble-enriched Crouzeix-Raviart (primal) method holds for arbitrary simplices (not just regular ones) and extends to hyperrectangles with diagonal mobility tensors.
Reinterpretation of Self-Stabilized Methods: It demonstrates that self-stabilized Hybrid High-Order (HHO) and Weak Galerkin (WG) methods are essentially projection methods. By viewing them through the lens of the derived equivalence, the paper provides their mixed finite element equivalents, which were previously unknown or unclear.
Practical Well-Posedness Criterion: The paper derives a simple, checkable criterion for the well-posedness and equivalence of projection methods on general polytopal meshes. This criterion relies on the surjectivity of the flux map (mapping fluxes to their divergence and normal traces), which is a reformulation of the LBB condition.

4. Key Results

Theorem (Discrete Equivalence): For the virtual spaces defined, the mixed problem (Problem 6) and the primal problem (Problem 5) are equivalent. The solution to the mixed problem yields a flux $\boldsymbol{\sigma} = \mathbf{b}\nabla u_K$ and a potential $u = \pi_{K}(u_K)$ that solve the primal problem.
Equivalence of Multiscale Methods: The paper confirms the first-level equivalence between the MsHHO (Multiscale Hybrid High-Order) method and the MHM (Multiscale Hybrid Mixed) method, broadening previous results.
Bubble Enrichment: It is proven that the lowest-order Raviart-Thomas method on arbitrary simplices is equivalent to a Crouzeix-Raviart method enriched with a specific quadratic "weak" bubble, without requiring geometric regularity of the mesh.
Projection Equivalence: For higher-order Raviart-Thomas or BDM spaces where the constitutive relation cannot be preserved exactly, the mixed method is equivalent to a primal projection method (Eq. 45). This provides a theoretical foundation for self-stabilized HHO and WG methods.

5. Significance

Theoretical Unification: The work unifies the seemingly disparate fields of multiscale methods, hybrid methods, and projection methods, showing they are different faces of the same underlying variational structure.
Algorithmic Insight: By establishing the equivalence, the paper allows practitioners to use the most convenient formulation. For instance, one can solve a primal system (often easier to precondition) and reconstruct a conforming flux, or vice versa.
Validation of Self-Stabilized Methods: It provides a rigorous justification for "stabilization-free" hybrid methods (HHO/WG) by linking them to well-understood mixed finite element theories. This validates their stability and convergence properties.
Foundation for Part II: This paper sets the stage for Part II, which will address Polytopal Element Methods where the global flux space is not conforming in $\mathbf{H}(\text{div})$ , relaxing the assumptions made here.

In summary, Lemaire's work offers a profound theoretical bridge between primal and mixed formulations on complex meshes, clarifying the role of virtual spaces, bubbles, and projections in modern numerical analysis of diffusion problems.