🔢 mathematics

Quantitative Homogenization of a Cahn--Hilliard System with Source Term in Periodically Perforated Domains

This paper establishes the qualitative and quantitative homogenization of a Cahn--Hilliard system with a source term in periodically perforated domains, deriving a homogenized model via the periodic unfolding method and proving an improved $O(\varepsilon^{1/2})$ convergence rate for corrector approximations under $H^2$ -regularity, which matches the optimal rate for second-order elliptic problems.

Original authors: Amartya Chakrabortty

Published 2026-06-30

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Amartya Chakrabortty

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

The Big Picture: Mixing Coffee in a Sieve

Imagine you have a cup of coffee with milk. You want to mix them perfectly. In a normal cup, the milk swirls and blends smoothly. This is like the standard Cahn-Hilliard equation, a famous mathematical model used to describe how two substances (like oil and water, or different phases of a metal) separate or mix over time.

Now, imagine you are trying to mix that coffee, but the cup is actually a sieve filled with tiny, solid obstacles (like a sponge or a porous rock). The liquid can only flow through the tiny holes between the solid bits. This is a perforated domain.

The problem is: The holes are microscopic. If you tried to simulate the mixing by drawing every single tiny hole on a computer, it would take forever. You need a shortcut. You want a "big picture" rule that tells you how the coffee mixes on average, without needing to track every single grain of the sieve. This process of finding the "big picture" rule is called homogenization.

The Problem: A Leaky Sieve with a Pump

This paper tackles a specific, tricky version of this problem:

The Obstacles: The sieve is made of a regular, repeating pattern of holes (periodic).
The Twist (Source Term): Usually, in these mixing problems, the total amount of "stuff" stays the same (conserved). But in this paper, the authors add a source term. Imagine that while the coffee is mixing, someone is secretly pumping more coffee in or draining some out at specific spots. The amount of liquid isn't constant; it's changing due to this "pump."
The Goal: They want to prove that as the holes get infinitely small (the sieve gets finer), the messy, detailed mixing process settles down into a smooth, predictable "averaged" process.

The Main Discovery: A Better Map

The authors did two main things:

1. The Qualitative Map (The "What")
They proved that as the holes get smaller, the chaotic, detailed mixing in the sieve does converge to a single, smooth equation.

The Analogy: Think of looking at a digital photo from far away. The individual pixels (the holes) blur together, and you see a clear image. They proved that the "blurred image" (the homogenized equation) is a valid description of the system.
The Catch: They found that the "pump" (the source term) actually helps the math work out better than expected. Usually, adding a pump makes things unstable, but because their pump is "monotone" (it behaves in a predictable, steady way), it actually acts like a stabilizer, helping the system settle down faster.

2. The Quantitative Map (The "How Fast")
This is the paper's biggest claim. They didn't just say "it works"; they calculated how fast the detailed version matches the smooth version.

The Old Way: Previous research suggested that if you wanted to know how close the detailed sieve-mixing was to the smooth average, the error would shrink at a rate of $\epsilon^{1/4}$ .
- Analogy: If you shrink the holes by a factor of 10, your error only gets 1.78 times smaller. That's slow.
The New Way: This paper proves the error shrinks at a rate of $\epsilon^{1/2}$ .
- Analogy: If you shrink the holes by a factor of 10, your error gets 3.16 times smaller. This is much faster!
Why is this special? This rate ( $\epsilon^{1/2}$ ) is the "natural" speed for these types of problems. The authors managed to reach this speed even though the math involves a very complex, fourth-order equation (which usually makes things harder).

How They Did It: The "Scale-Splitting" Tool

To get this faster speed, they invented a clever way to build a "correction" tool.

The Problem: The smooth average equation misses the tiny, rapid wiggles caused by the holes. To fix this, you need to add a "correction" that accounts for those wiggles.
The Old Tool: Previous methods required the correction to be extremely smooth and perfect (mathematically, $W^{1,\infty}$ ). This was like trying to draw a perfect, smooth curve over a jagged mountain range—it required too much detail and made the math break down near the edges of the cup.
The New Tool: They used a Scale-Splitting Operator.
- Analogy: Imagine you are trying to describe a bumpy road. Instead of trying to draw every single pebble, you draw the general slope of the road, and then you add a "bumpiness factor" that is calculated locally.
- This new tool allowed them to use simpler, less perfect corrections. It meant they didn't need the "perfectly smooth" assumption, which removed the mathematical bottlenecks that were slowing down the error rate in previous studies.

The "Edge Effect" (Why not 100% perfect?)

The authors note that their speed ( $\epsilon^{1/2}$ ) is limited by the edges of the domain (the rim of the cup).

The Analogy: In the middle of the sieve, the holes are perfect and repeating. But right near the rim of the cup, the holes get cut off. They are "incomplete cells." These messy edges create a "boundary layer" of confusion that slows down the convergence.
The Silver Lining: They proved that if you remove the edges entirely (imagine the sieve is a giant, infinite loop like a video game world with no walls, called a "torus"), the error rate improves even further to $\epsilon$ (linear speed). This confirms that the "slowness" in the main result is purely due to the messy edges, not the complexity of the mixing itself.

Summary of Claims

Convergence: A Cahn-Hilliard system with a source term in a porous medium does converge to a smooth, averaged equation as the pores get tiny.
Speed: The error between the real, detailed system and the smooth average shrinks at a rate of $\epsilon^{1/2}$ . This is a significant improvement over the previous best rate of $\epsilon^{1/4}$ for this specific type of problem.
Method: They achieved this by using a "scale-splitting" tool that allows for simpler mathematical corrections, avoiding the need for overly strict smoothness assumptions.
Limitation: The rate is limited by the "messy edges" of the domain. If the domain had no edges (a torus), the rate would be even faster ( $\epsilon$ ).

Note: The paper strictly focuses on the mathematical proof of these rates and the behavior of the equations. It does not claim specific results for clinical uses, industrial applications, or future technologies, other than the general mathematical context of phase separation in porous media.

Technical Summary: Quantitative Homogenization of a Cahn–Hilliard System with Source Term in Periodically Perforated Domains

Problem Statement
This paper investigates the qualitative and quantitative homogenization of a Cahn–Hilliard system featuring a nonconservative, monotone source term within a periodically perforated domain. The microscopic model describes phase separation in a porous medium where the fluid occupies a connected pore space $\Omega^\varepsilon_p \subset \Omega \subset \mathbb{R}^d$ ( $d \in \{2,3\}$ ), obtained by removing a periodic array of solid inclusions from a bounded Lipschitz domain. The system couples the order parameter $\phi^\varepsilon$ and chemical potential $\mu^\varepsilon$ via:
$\begin{cases} \partial_t \phi^\varepsilon - \Delta \mu^\varepsilon + G(\phi^\varepsilon) = 0, \\ \mu^\varepsilon = -\Delta \phi^\varepsilon + F'(\phi^\varepsilon), \end{cases}$
subject to homogeneous Neumann boundary conditions on both the outer boundary $\partial \Omega$ and the pore walls $\Gamma^\varepsilon_s$ . The potential $F$ satisfies polynomial growth and coercivity (e.g., the classical double-well potential), while the source term $G$ is globally Lipschitz, monotone, and satisfies $G(0)=0$ . The primary challenge lies in deriving effective macroscopic equations as the pore scale $\varepsilon \to 0$ and establishing sharp convergence rates for the error between microscopic and homogenized solutions.

Methodology
The analysis employs the periodic unfolding method adapted to perforated domains, utilizing the unfolding operator $T^\varepsilon_*$ to handle the oscillating geometry. The methodology proceeds in two main stages:

Qualitative Homogenization:
- Energy Estimates: Uniform a priori estimates are derived for the microscopic solutions. A key structural observation is that the monotone source term $G$ (with $G(0)=0$ ) contributes a dissipative term in the energy identity. Specifically, testing the chemical potential equation with $G(\phi^\varepsilon)$ yields a positive gradient term, reinforcing dissipation rather than competing with it. This allows for uniform bounds independent of $\varepsilon$ without requiring a Gronwall argument with exponential time factors.
- Two-Scale Convergence: Using the unfolding operator and compactness arguments (Aubin–Lions–Simon lemma), the authors prove the convergence of $(\phi^\varepsilon, \mu^\varepsilon)$ to a homogenized pair $(\phi, \mu)$ . The limit system is a Cahn–Hilliard equation on the macroscopic domain $\Omega$ with an effective diffusion tensor $B_{\text{hom}}$ .
- Cell Problems: The effective tensor $B_{\text{hom}}$ is characterized by scalar Neumann cell problems on the reference pore cell $Y_p$ . The correctors $\chi_i$ belong to $H^1_{\text{per},0}(Y_p)$ .
Quantitative Analysis (Corrector Estimates):
- Scale-Splitting Operator: To achieve quantitative rates, the paper constructs first-order corrector approximations $\Phi^\varepsilon$ and $M^\varepsilon$ using the $Q_1$ scale-splitting operator $Q_\varepsilon$ (from Griso [29, 30]). This operator acts on the fixed macroscopic domain $\Omega$ and is subsequently restricted to $\Omega^\varepsilon_p$ . Crucially, this approach allows the cell correctors to remain in $H^1_{\text{per}}(Y_p)$ , avoiding the restrictive $W^{1,\infty}$ assumptions required by classical formal expansions.
- Variational Residual Method: The error analysis is conducted in a negative-norm framework. The phase error $e^\varepsilon_\phi = \phi^\varepsilon - \Phi^\varepsilon$ is tested against the Neumann inverse Laplacian $N_\varepsilon e^\varepsilon_\phi$ , while the chemical potential error is tested against $e^\varepsilon_\phi$ . This avoids the need for higher regularity of the residuals.
- Consistency Estimates: The proof relies on deriving variational residual estimates (Lemma 4.6) that account for the boundary layer of incomplete cells near $\partial \Omega$ .

Key Results

Qualitative Convergence (Theorem 3.7): The microscopic solutions converge (up to subsequences) to the homogenized system:
$\begin{cases} \partial_t \phi - \text{div}(B_{\text{hom}}\nabla \mu) + G(\phi) = 0, \\ \mu = -\text{div}(B_{\text{hom}}\nabla \phi) + F'(\phi), \end{cases}$
with effective Neumann boundary conditions. The tensor $B_{\text{hom}}$ is symmetric and uniformly elliptic.
Quantitative Rates (Theorem 4.7): Under an additional $H^2$ -regularity assumption on the homogenized solution and well-prepared initial data, the paper establishes an error estimate of order $\varepsilon^{1/2}$ :
$\|\phi^\varepsilon - \phi\|_{L^2(0,T; L^2(\Omega^\varepsilon_p))} + \|\nabla \phi^\varepsilon - (\nabla \phi + \sum Q_\varepsilon(\partial_{x_i}\phi)\nabla_y \chi_i(\tfrac{x}{\varepsilon}))\|_{L^2((0,T)\times\Omega^\varepsilon_p)} + \|\mu^\varepsilon - \mu\|_{L^2(0,T; H^1(\Omega^\varepsilon_p)')} \leq C\varepsilon^{1/2}.$
- The uncorrected order parameter error is controlled in $L^2(0,T; L^2(\Omega^\varepsilon_p))$ .
- The corrected gradient error is controlled in $L^2((0,T)\times\Omega^\varepsilon_p)$ .
- The chemical potential error is controlled in the dual norm $L^2(0,T; H^1(\Omega^\varepsilon_p)')$ .
Boundary Layer Effects: The $\varepsilon^{1/2}$ rate on bounded domains is attributed to the boundary layer of incomplete cells near $\partial \Omega$ . On the flat torus $\mathbb{T}^d$ , where no such boundary layer exists, the rate improves to $O(\varepsilon)$ (Remark 4.10).

Significance and Contributions
The paper makes two primary contributions to the homogenization theory of phase-field equations in perforated media:

Qualitative Extension: It rigorously derives the homogenized Cahn–Hilliard system with a general monotone nonconservative source term, demonstrating that the source term's dissipative structure simplifies the energy estimates compared to general nonconservative forcing.
Quantitative Improvement: It improves the known convergence rate for Cahn–Hilliard systems in perforated domains from $\varepsilon^{1/4}$ $ε^{1/4}$ (established by Schmuck–Kalliadasis [49]) to the natural rate of $\varepsilon^{1/2}$ $ε^{1/2}$ for second-order elliptic problems in perforated domains.
- Methodological Advancement: Unlike previous works requiring cell correctors in $W^{1,\infty}$ and macroscopic solutions in $C^1(0,T; W^{k,\infty})$ with $k \geq 4$ , this approach utilizes the scale-splitting operator $Q_\varepsilon$ . This reduces the required regularity of the macroscopic solution to $H^2$ and keeps cell correctors in $H^1_{\text{per}}$ .
- Variational Consistency: The consistency estimates are derived entirely in variational form, utilizing the weak no-flux condition inherited from the cell problem. This avoids the boundary-term contributions that previously caused the rate degradation to $\varepsilon^{1/4}$ .

The authors note that while the $\varepsilon^{1/2}$ rate is sharp within the current variational framework due to the boundary layer, achieving the optimal $O(\varepsilon)$ rate on bounded domains would likely require a duality argument (as in Kenig–Lin–Shen [40]) or an operator-estimate approach (Birman–Suslina [7–9]), which are left for future work. The paper does not claim to obtain a gradient rate for the chemical potential $\mu^\varepsilon$ at this order, as the current regularity of residuals does not justify the necessary testing.