Preconditioning for near-contacts in large 2D Stokes flows: a locally compressed method of fundamental solutions

This paper introduces a locally compressed method of fundamental solutions combined with a two-body preconditioning strategy to efficiently solve large-scale 2D Stokes flow problems involving dense collections of nearly-touching rigid particles, achieving rapid iterative convergence even in challenging multi-particle configurations with extremely small gaps.

The Gist

Imagine you are trying to simulate how thousands of tiny, rigid coins move through a thick, sticky fluid (like honey) in a flat, two-dimensional world. This is a problem in physics called Stokes flow.

The paper presents a new, clever way to solve the math behind this simulation, specifically when the coins get extremely close to each other—almost touching, but not quite.

Here is the breakdown of the problem and the solution, using everyday analogies.

The Problem: The "Sticky Gap" and the "Math Gridlock"

When these coins move, they push the fluid around them. If two coins are far apart, the fluid flows smoothly, and standard math tools can handle it easily.

However, when two coins get very close (leaving a tiny gap of just 0.001 of their width), two major headaches occur:

1. The Lubrication Spike: The fluid squeezed between the coins has to move incredibly fast to get out of the way. It's like trying to squeeze a thick paste through a pinhole; the pressure and speed spike dramatically. To calculate this accurately, you need a super-detailed map (a "fine grid") of that tiny gap.
2. The Math Gridlock: If you try to solve the whole system at once using a super-detailed map for every coin, the computer gets stuck. The math equations become "ill-conditioned," which is like trying to balance a house of cards on a shaking table. The computer has to try millions of times to find the answer, or it gives up entirely.

The Old Way:
Previously, to handle these close calls, scientists had to make the entire map of the fluid super-detailed everywhere, just in case two coins got close. This is like trying to see a single ant on a football field by zooming in so high that you can't see the whole field anymore. It requires too much computer memory and takes too long.

The Solution: The "Local Fix" and the "Peanut Wrapper"

The authors (Broms, Tornberg, and Barnett) invented a "two-body preconditioning" method. Think of it as a hybrid strategy that combines a rough sketch with a detailed zoom-in, but only where needed.

Step 1: The Rough Sketch (The Coarse Grid)

For the vast majority of the simulation, they use a "coarse" map. They treat each coin as a simple object with a few key points. This is fast and easy to calculate, like looking at a map of a city where streets are just lines.

Step 2: The Local Zoom-In (The Two-Body Fix)

When two coins get dangerously close, the "coarse" map fails. Instead of redrawing the whole city map, the computer pauses and solves a tiny, separate, high-resolution puzzle just for that pair of coins.

• Analogy: Imagine you are drawing a crowd. For most people, you just draw a circle. But if two people are hugging, you zoom in and draw the details of their embrace perfectly. You don't redraw the whole crowd; you just fix that one spot.

Step 3: The "Peanut" Compression (The Magic Trick)

This is the paper's most creative innovation. The high-resolution zoom-in creates a massive amount of data. If you kept all that data, you'd still be slow.

• The Trick: They take that detailed "hug" between the two coins and mathematically "compress" it. They wrap the two coins in a imaginary peanut-shaped shell.
• How it works: They prove that the complex fluid flow inside that peanut shape can be perfectly mimicked by a much simpler, coarser set of points on the outside of the peanut.
• The Result: The computer can throw away the expensive, detailed data and replace it with a simple "coarse" version that acts exactly the same from a distance. This allows the global simulation to stay fast and simple, even though the physics of the close contact are perfectly resolved.

Why This Matters

The paper tests this method on a massive crowd of 10,000 coins packed tightly together (so tightly that the gaps are 1,000 times smaller than the coins themselves).

• Without this method: The computer would likely crash or take days/weeks to solve.
• With this method: The computer solves the problem in 47 steps (iterations) and finishes in 36 seconds on a single computer.

Summary in One Sentence

The authors created a smart math tool that uses a "rough sketch" for the whole crowd but instantly zooms in to solve the tricky physics of near-touching pairs, then magically shrinks that detailed solution back down to a simple form so the computer doesn't get overwhelmed.

Key Takeaway: They didn't just make the computer faster; they changed how the math is structured to handle the "sticky" moments between particles without needing to calculate every single drop of fluid in the entire system.

Technical Summary

Technical Summary: Preconditioning for Near-Contacts in Large 2D Stokes Flows

Problem Statement
The paper addresses two primary computational challenges in simulating the viscous hydrodynamics of large, dense suspensions of rigid particles in 2D Stokes flow:

1. Ill-conditioning: As particle gaps ($\delta$) shrink, the linear systems arising from discretizing the Stokes equations become severely ill-conditioned. Iterative solvers like GMRES require an unbounded number of iterations to reach a fixed accuracy as $\delta \to 0$.
2. Lubrication-driven fine scales: Accurate resolution of the steep velocity gradients and force peaks in narrow inter-particle gaps (lubrication forces) typically requires extremely fine spatial discretization. In global methods, this leads to a prohibitive increase in the number of unknowns, as local refinements interact globally due to the non-local nature of elliptic kernels.

The authors focus on the exterior Stokes boundary value problems for nearly-touching disks, covering both the resistance problem (prescribed velocities, solving for forces/torques) and the mobility problem (prescribed forces/torques, solving for velocities).

Methodology
The proposed solution is a locally compressed two-body preconditioning strategy implemented within the Method of Fundamental Solutions (MFS). The approach is a hybrid of direct and iterative methods, structured as follows:

1. Two-Body Basis Construction:

   • Instead of using a global fine grid, the method constructs a basis that incorporates pairwise corrections.
   • For each particle, a "one-body" basis is first established by solving the Stokes problem for an isolated particle.
   • For particles in near-contact, a two-body correction is computed by solving a local, high-resolution boundary value problem (BVP) for the specific pair. This local solve uses a fine grid to resolve the lubrication peaks.
   • The global flow field is represented as a superposition of these pair-corrected basis functions.

2. Peanut Compression:

   • To prevent the fine-grid resolution of near-contacts from increasing the global system size, the authors introduce a "peanut compression" step.
   • The fine-grid solution for a particle pair is compressed into an equivalent set of coarse sources located on the original coarse source points.
   • This compression is achieved by matching the velocity fields generated by the fine sources and the coarse sources on a "peanut" proxy surface (a separation boundary formed by rolling a unit circle around the pair).
   • This process effectively creates a "scattering matrix" for the pair, allowing the global solver to operate on a coarse discretization while implicitly accounting for the resolved near-field interactions.

3. Implementation Details:

   • MFS Enhancement: The local pairwise BVPs are solved using an image-enhanced MFS. This involves placing Stokeslet sources not only on a standard interior circle but also on elliptical arcs designed to shield the image accumulation points (singularities) that arise in the analytic continuation of the solution.
   • Mobility Constraints: For the mobility problem, the method employs a "recompleted" formulation where force and torque constraints are automatically satisfied, allowing for unconstrained least-squares solves.
   • Solver: The global system is solved iteratively using GMRES, accelerated by a Fast Multipole Method (FMM) for matrix-vector products. The preconditioner is applied via block-diagonal corrections derived from the compressed two-body interactions.

Key Contributions

• General Two-Body Preconditioner: The paper introduces a general framework for two-body preconditioning that can be applied to any boundary-based linear PDE solver, provided a local pairwise solver is available. It generalizes previous pairwise image-sum methods (e.g., Cheng & Greengard) to arbitrary solvers and allows for fast matrix-vector multiplication.
• Local Resolution, Global Coarseness: The method successfully resolves lubrication-driven fine scales locally without increasing the global degrees of freedom. The global system remains coarse, with the number of unknowns scaling linearly with the number of particles, regardless of the minimum gap size.
• First Application to 2D Mobility: The authors apply near-contact MFS image enhancement to the 2D mobility problem for the first time, handling the specific constraints of prescribed forces and torques.
• Stabilization: The approach stabilizes the ill-conditioned linear system, preventing the iteration count from exploding as particle gaps shrink.

Numerical Results

• Convergence: In challenging multi-particle settings, the method demonstrates rapid GMRES convergence. For a random close packing of $P=10,000$ monodisperse disks with a packing fraction $\phi=0.65$ and a minimum separation of $10^{-3}$, the mobility problem was solved in just 47 GMRES iterations.
• Accuracy: The method achieved five digits of accuracy with 72 vector unknowns per body. The relative boundary residual was uniformly below $10^{-5}$.
• Efficiency: Compared to one-body preconditioning, the two-body approach reduced the number of unknowns by a factor of 12.5 and the solve time by a factor of 30 for mobility problems at small gaps. For resistance problems, the speed-up was a factor of 65.
• Scalability: The iteration count for the mobility problem was found to be essentially independent of the total number of particles ($P$), depending primarily on the local number of near neighbors.

Significance and Claims
The paper claims to present the first convergent method that combines the flexibility of the MFS with a preconditioning strategy capable of resolving lubrication effects at physically relevant small gaps (down to $10^{-3}$ of the particle radius) without requiring a global fine discretization. This is highlighted as crucial for simulations involving very large numbers of particles.

The authors emphasize that while the method is demonstrated in 2D, the underlying ideas are readily applicable to 3D. They position the work as a scalable and flexible foundation for large-scale simulations of dense suspensions, noting that the current dominant cost is the precomputation of corrections for near-contact pairs, which remains an area for future optimization. The method is presented as a robust alternative to existing non-convergent effective particle methods and computationally expensive volume-based or global-boundary refinement methods.