🔢 mathematics

Multicontinuum Generalized Multiscale Finite Element Method (MC-GMsFEM). Theory and applications to upscaling of two-phase flow

This paper introduces a Multicontinuum Generalized Multiscale Finite Element Method (MC-GMsFEM) that unifies multiscale discretization with multicontinuum representations to systematically derive physically consistent macroscopic equations for heterogeneous media, demonstrating its effectiveness in accurately upscaling two-phase immiscible flow where classical models fail.

Original authors: Mohammed Al Kobaisi, Dmitry Ammosov, Yalchin Efendiev, Wing Tat Leung

Published 2026-06-02

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Mohammed Al Kobaisi, Dmitry Ammosov, Yalchin Efendiev, Wing Tat Leung

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 understand the traffic flow in a massive, chaotic city. The city has millions of tiny streets, alleys, and one-way systems (the "fine scale"). If you tried to track every single car on every single street, you would need a supercomputer that never sleeps, and the data would be impossible to manage.

Usually, scientists try to solve this by drawing a simple map with just a few main highways (the "coarse scale"). They say, "Okay, 100 cars go from Point A to Point B." This is fast and easy, but it often misses the messy reality: what if 50 of those cars got stuck in a dead-end alley? Or what if a specific neighborhood has a unique traffic pattern that doesn't fit the main highway rules?

The Problem with Old Maps
Traditional methods are like those simple highway maps. They are great at giving you a rough idea of where traffic is going, but they can't tell you why traffic is jamming in a specific alley, nor can they easily write down a new set of "traffic laws" that explain the jam. They give you a number, but not a story.

The New Solution: MC-GMsFEM
This paper introduces a new method called MC-GMsFEM (Multicontinuum Generalized Multiscale Finite Element Method). Think of this as a smart, layered map that does two things at once:

It gives you a highly accurate simulation of the traffic.
It automatically writes a new set of "macroscopic laws" (equations) that explain the traffic patterns, including the jams.

How It Works: The "Neighborhood" Analogy
Imagine the city is divided into large blocks (coarse grids). Inside each block, there are different types of "neighborhoods":

The Main Roads: Where cars flow freely (mobile phases).
The Dead-Ends: Where cars get stuck in a cul-de-sac and can't move easily (trapped phases or "ganglia").

Old methods treated the whole block as one big mix of traffic. The new method realizes that the "Main Road" traffic and the "Dead-End" traffic behave differently. It creates separate lanes for each type of neighborhood within the same block.

The "Continuum" Concept: The paper calls these separate lanes "continua." It's like saying, "In this city block, we have two distinct worlds: the flowing world and the stuck world."
The Magic Trick: The method uses special mathematical tools (called "basis functions") to look at the tiny details of the city once, and then builds a simplified model that remembers those details. It separates the "average speed" from the "local bumps and bumps."

Why This Matters for Oil and Water
The authors tested this on a problem involving two fluids mixing, like oil and water moving through a sponge (porous media).

In a sponge, some water flows freely through the big holes.
Other water gets trapped in tiny, isolated pockets (like a drop stuck in a sponge's corner).

If you use the old "highway map" method, you assume all the water moves the same way. The paper shows this leads to big errors, especially for the trapped water. It's like assuming a car stuck in a cul-de-sac is moving at highway speeds.

The new method treats the "free-flowing water" and the "trapped water" as two different characters in the story. It calculates how they move separately and then combines them. The result? A much more accurate prediction of where the fluids go.

The Big Takeaway
This paper doesn't just give you a better number; it gives you a better story.

Old Way: "Here is the total amount of fluid." (Accurate number, but no explanation).
New Way: "Here is the total amount, and here are the specific rules for how the free fluid moves versus how the trapped fluid moves." (Accurate number + a clear physical explanation).

The authors prove that by separating these "continua" (the different types of fluid behavior), they can create a model that is both fast enough to run on a computer and smart enough to capture the complex, messy reality of fluids getting stuck in rocks. It's a bridge between the tiny, chaotic details and the big, understandable picture.

Technical Summary: Multicontinuum Generalized Multiscale Finite Element Method (MC-GMsFEM)

Problem Statement
Multiscale problems in heterogeneous media are typically addressed by constructing coarse-scale models that incorporate fine-scale information via multiscale basis functions. While existing methods (such as GMsFEM) efficiently approximate fine-scale solutions on coarse grids, they generally lack a systematic framework for deriving physically consistent macroscopic differential equations. The coarse-scale representation in these methods often involves degrees of freedom that lack spatial continuity, obscuring the connection to underlying continuum models. Furthermore, traditional multicontinuum homogenization techniques struggle with complex systems where the number of interacting continua varies spatially and dynamically, often requiring the solution of special cell problems that are difficult to implement.

This paper addresses these limitations by developing a unified framework capable of constructing conforming multiscale basis functions while simultaneously enabling the systematic derivation of macroscopic governing equations. A primary application considered is the upscaling of two-phase immiscible flow in heterogeneous porous media, where multiple interacting continua (including mobile and trapped phases) arise.

Methodology: MC-GMsFEM Framework
The proposed Multicontinuum Generalized Multiscale Finite Element Method (MC-GMsFEM) integrates multiscale finite element constructions with multicontinuum representations. The core methodology proceeds as follows:

Continuum Identification: The method assumes the identification of continua globally (e.g., via characteristic functions $\psi_i$ ). In the context of two-phase flow, these continua correspond to distinct fluid phases or specific regions such as trapped ganglia.
Construction of Auxiliary Functions:
- Continuum Functions ( $N_i$ ): Local functions representing the $i$ -th continuum are constructed by solving local problems with integral constraints. Specifically, $N_i$ is defined such that it has a unit average over continuum $i$ and vanishing averages over all other continua within a coarse block $K$ : $\int_K N_i \psi_j = \delta_{ij} \int_K \psi_j$ .
- Gradient Correction Functions ( $M_i$ ): To account for fine-scale features and gradients, auxiliary functions $M_i$ (or their discrete counterparts $z_i = M_i \cdot \nabla \phi_0$ ) are constructed. These satisfy local problems with zero boundary conditions on coarse block boundaries and orthogonality constraints with respect to the continua.
Basis Function Representation: The multiscale basis functions are expressed as:
$\phi^j_i = N_i \phi^j_0 + M_i \cdot \nabla \phi^j_0$
where $\phi^j_0$ are standard partition of unity functions. This representation separates coarse variables and their gradients.
Derivation of Macroscopic Equations: The coarse-scale solution is decomposed as $u_H = \sum_i (N_i U_i + M_i \cdot \nabla U_i)$ , where $U_i$ are the continuum variables. By substituting this decomposition into the variational formulation of the governing equations (e.g., elliptic equations or Stokes flow), the authors derive macroscopic equations. These equations naturally include terms involving the gradients of the continuum variables (Brinkman-like terms), which are crucial for capturing effects in trapped regions.
Application to Two-Phase Flow: The framework is applied to coupled Stokes flow and transport equations. The method treats long trapped fluid regions as distinct continua. The incompressibility and momentum equations are derived for both single-pressure and multicontinuum pressure formulations, resulting in macroscopic models that account for transfer terms between continua and incompressibility constraints.

Key Results
The paper presents numerical experiments for elliptic equations in heterogeneous and perforated domains, as well as two-phase flow simulations in perforated media.

Elliptic Problems: The MC-GMsFEM was tested on heterogeneous and perforated microstructures. The method demonstrated high accuracy, with relative errors for upscaled and downscaled solutions generally remaining below 20% across various test cases (Table 1). The visualizations confirmed that the downscaled multiscale solutions closely matched reference fine-grid solutions.
Two-Phase Flow:
- Ganglia Detection: A breadth-first search (BFS) algorithm was used to identify trapped fluid regions (ganglia) and treat them as separate continua.
- Accuracy: In tests involving Target domains 1 and 2 with various finger realizations (Tests A–E), the MC-GMsFEM approach (both global and local variants) produced coarse-scale solutions with small relative errors compared to reference fine-grid averages (Tables 2–16).
- Impact of Gradient Terms: A critical experiment (Test C) compared models with and without the velocity gradient terms ( $M_i \cdot \nabla U_i$ ). The results (Table 17) showed that omitting these terms (reducing the model to a standard multicontinuum Darcy model) significantly increased errors, particularly for the velocities of trapped ganglia (error increased from ~16% to ~96%). This indicates that gradient terms are essential for accurately modeling trapped regions.
- Continuum Separation: The study showed that treating ganglia as distinct continua improves the physical consistency of the model, although the numerical impact on concentration errors was found to be minor in the specific test cases, while the velocity representation was significantly refined.

Significance and Claims
The paper claims that MC-GMsFEM offers a robust and general approach for modeling complex multiscale systems. Its primary significance lies in:

Unified Framework: It provides a systematic pathway to derive macroscopic governing equations directly from multiscale discretizations, bridging the gap between numerical approximation and continuum homogenization.
Flexibility: Unlike previous multicontinuum approaches, this framework handles heterogeneous media where the number of interacting continua varies spatially and dynamically without requiring special cell problem solutions for every configuration.
Physical Consistency: By naturally embedding the method within a standard finite element setting, it allows for the derivation of macroscopic equations that include necessary gradient terms, which are shown to be critical for capturing the dynamics of trapped phases in two-phase flow.
Computational Efficiency: The method enables efficient coarse-grid simulations while preserving fine-scale information, offering a computationally efficient alternative to full fine-scale modeling.

The authors modestly note that while the method offers advantages over existing homogenization techniques, it does not explicitly define the gradient term coefficients in the macroscopic equations in a closed form that can be easily pre-calculated without solving local problems; however, the method's adaptability to varying continua and general heterogeneities is highlighted as its main strength.

More like this