🔢 mathematics

A Generalised Curvilinear Coordinate system-based Patch Dynamics Scheme in Equation-free Multiscale Modelling

This paper introduces a novel patch dynamics scheme for equation-free multiscale modelling that utilizes generalised orthogonal curvilinear coordinates and non-uniform, non-rectangular patch configurations to efficiently and accurately simulate unsteady, linear, heterogeneous convection-diffusion-reaction problems on complex geometries, significantly outperforming traditional full-domain simulations in computational efficiency and memory usage.

Original authors: Tanay Kumar Karmakar, Durga Charan Dalal

Published 2026-06-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tanay Kumar Karmakar, Durga Charan Dalal

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 predict the weather for an entire continent. The standard way to do this is to build a massive computer model that simulates every single molecule of air, every drop of rain, and every gust of wind across the whole map, second by second. While accurate, this approach is like trying to count every grain of sand on a beach to measure the beach's size—it takes forever and requires a computer the size of a city.

This paper introduces a smarter, faster way to solve these massive problems, specifically for situations where the "ground" isn't flat and the "weather" changes wildly in certain spots.

Here is the breakdown of their new method, using simple analogies:

1. The Problem: The "Too Big to Fit" Puzzle

In many real-world situations (like heat moving through a complex machine or chemicals flowing through a curved pipe), the area we need to study isn't a perfect square. It might be a ring (like a donut) or a shape with weird curves. Furthermore, some parts of the problem change very slowly, while other tiny spots change incredibly fast (like a thin layer of heat right against a hot wall).

Trying to simulate the entire area with high detail is too expensive and slow. Trying to simulate it with low detail misses the important fast changes.

2. The Old Solution: The "Uniform Tiling" Approach

Previous methods tried to solve this by laying down a grid of identical, square tiles (patches) over the problem area.

The Flaw: If your problem area is a curved donut, square tiles leave ugly gaps or overlap awkwardly. If you need to zoom in on a tiny, fast-changing spot, you have to make every tile tiny, which wastes time on the parts that don't need it.

3. The New Solution: The "Stretchy Rubber Sheet" and "Smart Patches"

The authors propose a new system that combines two clever ideas:

A. The Stretchy Rubber Sheet (Curvilinear Coordinates)
Imagine your complex, curved problem area is drawn on a piece of rubber.

Instead of trying to force square tiles onto this rubber, you stretch and reshape the rubber itself until the curved problem looks like a perfect, flat rectangle.
Now, you can easily lay down your square tiles (patches) on this flat rectangle.
Once the simulation is done on the flat rectangle, you simply "un-stretch" the rubber to see what the solution looks like on the original curved shape.
Why it helps: It turns a messy, irregular shape into a neat, manageable rectangle without losing any of the original geometry's details.

B. The "Smart Patches" (Patch Dynamics)
Instead of simulating the whole flat rectangle at once, the computer only looks at small, isolated islands (patches) scattered across the map.

The Micro-Simulation: On each tiny island, the computer runs a super-detailed, high-speed simulation of the physics (the "micro" view).
The Macro-Extrapolation: The computer then takes the results from these tiny islands and uses a mathematical trick to guess what is happening in the vast empty spaces between the islands. It essentially says, "We know what's happening here and here; let's project that forward to skip the boring middle parts."
The Result: You get the accuracy of a full simulation but in a fraction of the time and memory.

4. Handling the "Fast Spots" (Stretched Grids)

In some problems, the action happens in a very thin strip (like a boundary layer).

The authors' method allows them to "stretch" the grid so that the tiles are tiny and crowded in the fast-changing areas, but large and sparse in the calm areas.
Analogy: Imagine taking a photo of a crowd. Instead of taking a photo where every person is the same size, you zoom in so the people in the front row are huge and detailed, while the people in the back are small. This captures the important details without needing a camera with infinite resolution.

5. What They Tested

The authors proved this works with two specific tests:

The "Hot Wall" Test: A square area where heat moves quickly and creates a sharp, thin layer of change near the edges. They showed that their "stretchy grid" method was much more accurate than standard methods, especially near those sharp edges.
The "Donut" Test: A ring-shaped area (an annulus) where heat diffuses. Because a ring is hard to grid with squares, they used their "rubber sheet" trick to turn the ring into a rectangle, solved it, and mapped it back. The results matched the known perfect answers almost exactly.

6. The Bottom Line

The paper claims that this new "Patch Dynamics" system is:

Faster: It runs much quicker than simulating the whole area.
Lighter: It uses less computer memory.
More Accurate: It handles weird shapes and sharp changes better than old methods.
Versatile: It works on both flat, square areas and curved, ring-shaped areas.

In short, they built a toolkit that lets computers solve complex, messy, real-world physics problems by breaking them into small, manageable pieces and using a "stretchy map" to make the math easier, all while skipping the boring parts of the simulation.

Technical Summary: A Generalised Curvilinear Coordinate System-based Patch Dynamics Scheme in Equation-free Multiscale Modelling

Problem Statement
The paper addresses the computational challenges inherent in multiscale modelling of complex systems, specifically those governed by unsteady, linear, heterogeneous convection-diffusion-reaction (CDR) equations. Traditional numerical schemes often require fine-level simulations over large spatial domains and long time intervals, leading to prohibitive computational costs and memory requirements. While "equation-free" multiscale methods, such as the patch dynamics scheme, have been developed to bridge microscopic and macroscopic scales by simulating only fractions of the space-time domain, existing implementations are largely restricted to rectangular domains with uniform grids and rectangular patches. This limitation hinders their application to real-life problems characterized by irregular geometries, non-uniform grid requirements (e.g., for capturing boundary layers), and complex physical domains.

Methodology
The authors propose a novel extension of the patch dynamics scheme that integrates generalised orthogonal curvilinear coordinates. The core methodology involves:

Coordinate Transformation: The physical domain ( $\Omega_p$ ), which may be non-rectangular or require non-uniform grid spacing, is mapped to a rectangular computational domain ( $\Omega_c$ ) with a uniform mesh via a smooth, one-to-one transformation. This transformation converts the original heterogeneous CDR equation into a form where the mixed derivative term vanishes (orthogonal condition), facilitating computation on a structured grid.
Patch Dynamics Framework: The scheme operates on the computational domain using a combination of the gap-tooth scheme and coarse projective integration.
- Micro-simulation: The microscopic problem is solved only within small, sparse patches centered at macroscopic grid points.
- Lifting and Restriction: A lifting operator reconstructs the microscopic initial conditions within patches from the macroscopic state using bi-quadratic polynomial interpolation. Conversely, a restriction operator averages the microscopic solution over each patch to update the macroscopic state.
- Edge Conditions: Neumann boundary conditions are imposed at patch edges using Lagrange polynomial interpolation based on neighboring macroscopic grid points (a 9-point stencil).
- Projective Integration: The macroscopic time derivative is estimated from short micro-simulations, and the solution is extrapolated forward over a large macroscopic time step using the forward Euler method.
Handling Heterogeneity: The scheme accommodates space-dependent diffusion tensors, convective velocities, and space-time-dependent source terms. It is designed for problems with smooth, non-oscillatory coefficients.

Key Contributions

Generalised Framework: This is the first application of a generalised orthogonal curvilinear coordinate system within the patch dynamics scheme, enabling the handling of both non-uniform grids in rectangular domains and non-rectangular physical domains.
Non-Uniform and Non-Rectangular Patches: The paper introduces the concept of non-uniform and non-rectangular patch configurations in the physical domain, which are mapped to uniform rectangular patches in the computational domain.
Explicit Formulation: An explicit representation of the scheme is provided for two-dimensional, unsteady, linear, heterogeneous CDR problems, including the derivation of transformed coefficients for the governing equations.
Algorithmic Implementation: The authors detail the complete algorithm, including the specific handling of edge conditions, lifting, and projective extrapolation in the context of curvilinear coordinates.

Results
The proposed scheme was validated through two distinct numerical test cases:

Convection-Dominated Heterogeneous CDR on a Stretched Grid:
- A problem with a rectangular domain featuring high gradients (boundary layers) near the top and right boundaries was solved.
- Stretched Grids: Using a stretching parameter $\lambda = 0.1$ , the scheme achieved significantly higher accuracy compared to uniform grids ( $\lambda = 0$ ), particularly in high-gradient regions.
- Performance: The patch dynamics scheme demonstrated superior accuracy (errors up to 13 times lower in specific cases) compared to a full-domain simulation (FDS) using the method of lines. It also required less computational time (approx. 40% reduction) and memory (approx. 10% reduction) while using fewer lattice points.
- Grid Independence: Solutions were shown to be grid-independent at a resolution of $20 \times 20$ macro grids with 8,500 time steps.
Non-Axisymmetric Diffusion in an Annulus:
- A diffusion problem in an annular domain (non-rectangular geometry) was solved using a body-fitted curvilinear coordinate system.
- Accuracy: The numerical results showed excellent agreement with the analytical solution.
- Convergence: The scheme demonstrated convergence as the macroscopic grid and time levels were refined.
- Boundary Conditions: The method successfully handled both periodic boundary conditions (in the angular direction) and Dirichlet boundary conditions (in the radial direction).

Significance and Claims
The paper claims that the proposed scheme offers a robust and efficient alternative to full-domain simulations for complex multiscale problems. Its significance lies in:

Geometric Flexibility: It overcomes the limitation of rectangular domains, allowing for the accurate simulation of systems with complex geometries and localized high-gradient regions without the need for global fine-grid resolution.
Computational Efficiency: By restricting microscopic simulations to small patches and utilizing projective integration, the scheme significantly reduces computational time and memory usage while maintaining high accuracy.
Versatility: The method is applicable to both "Type-A" problems (where microscale details are needed only in specific regions like boundary layers) and "Type-B" problems (where microscale models are required throughout the domain due to the absence of a macroscopic description).

The authors note that the current implementation is an initial development focused on linear, unsteady, heterogeneous CDR equations with smooth coefficients. They acknowledge that further work is needed to extend the framework to problems with highly oscillatory coefficients, where resonance errors from artificial edge conditions might require buffer regions or averaging kernels.