🔢 mathematics

Steady Incremental Viscosity Splitting Method for solving the stationary Navier-Stokes equation

This paper introduces a novel steady incremental viscosity splitting method for solving incompressible stationary Navier-Stokes equations that decouples the problem into an elliptic velocity PDE and a constant-coefficient pressure system at each iteration, while proving the scheme's boundedness, geometric convergence, and numerical efficiency.

Original authors: Aziz Takhirov, Driss Yakoubi

Published 2026-05-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aziz Takhirov, Driss Yakoubi

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 how a thick, sticky fluid (like honey or oil) flows through a pipe or a room. In physics, this is described by a set of very complicated math rules called the Navier-Stokes equations.

The problem is that these equations are like a tangled knot of spaghetti. If you try to solve them all at once, the math gets so heavy that even the world's most powerful computers can get stuck, run out of memory, or take forever to find an answer.

This paper introduces a new, clever way to untangle that knot. Here is the breakdown of their solution, using simple analogies:

1. The Problem: The "All-or-Nothing" Knot

Usually, to solve these fluid equations, computers try to calculate the speed of the fluid and the pressure pushing it at the exact same time.

The Analogy: Imagine trying to tie a knot while someone else is pulling the rope in a different direction. You have to adjust your hand speed and the tension on the rope simultaneously. It's a messy, unstable dance that requires a massive amount of brainpower (computing power) to get right.

2. The Solution: The "Step-by-Step" Dance

The authors propose a new method called Steady Incremental Viscosity Splitting (SIVS). Instead of trying to solve for speed and pressure all at once, they break the problem into two smaller, easier steps that repeat until the answer is perfect.

Step 1: The "Speed Guess"
First, the computer guesses the speed of the fluid based on what it knows so far, ignoring the pressure for a split second.
- Analogy: Think of a dancer practicing their footwork alone. They focus only on moving their feet correctly without worrying about the music (pressure) yet.
Step 2: The "Pressure Correction"
Next, the computer looks at that speed guess and asks, "Is this fluid flowing smoothly, or is it piling up?" It then calculates the exact pressure needed to fix any "piling up" and corrects the speed.
- Analogy: Now the dancer listens to the music and adjusts their posture to match the rhythm. They fix their stance so everything flows perfectly.

3. The Secret Weapon: The "Stabilizer"

The paper mentions a special ingredient called grad-div stabilization.

The Analogy: Imagine you are trying to balance a stack of blocks. Sometimes, if you just stack them, they wobble and fall. This "stabilizer" is like adding a little bit of glue or a wider base to the blocks. It doesn't change the final shape of the tower, but it makes the process of building it much less likely to collapse. In the math, this prevents the computer from getting confused by tiny errors.

4. Why This is a Big Deal

The authors prove two main things:

It Won't Crash: They mathematically proved that this method will always stay within safe limits and won't blow up, provided the fluid isn't moving too wildly (a condition they call "small data").
It Gets Faster: They showed that with every step of their dance, the answer gets closer to the truth at a steady, predictable rate.

The "Algebraic Splitting" Trick:
The paper also explains that their method is like a smart shortcut in algebra. Instead of solving a giant, complex puzzle (a massive matrix), they break it down into two smaller, simpler puzzles that are much easier for a computer to solve.

Analogy: Instead of trying to move a giant, heavy sofa up a staircase in one go, they break it down into the seat and the backrest, carry them up separately, and put them back together at the top. It's much lighter and easier to manage.

5. The Results: Did It Work?

The authors tested their new method on three scenarios:

A Test Tube: They created a fake fluid flow with a known answer to see if their math was accurate. It was.
A Box with a Moving Lid: They simulated fluid in a square box where the top moves (a classic test). They tested it at different speeds (Reynolds numbers) and found their method worked just as well as older methods but with less computational headache.
A 3D Cylinder: They simulated wind flowing around a cylinder in 3D space. This is a very hard problem for computers. Their method solved it quickly and gave results that matched known scientific benchmarks.

Summary

In short, this paper offers a new, efficient way to teach computers how to simulate fluid flow. Instead of forcing the computer to juggle speed and pressure simultaneously (which is hard), they teach it to take small, steady steps: guess the speed, fix the pressure, and repeat. This makes the calculations faster, more stable, and much easier for computers to handle, especially for complex 3D problems.

Technical Summary: Steady Incremental Viscosity Splitting Method for Stationary Navier-Stokes Equations

Problem Statement
The paper addresses the numerical solution of the stationary, incompressible Navier-Stokes equations with grad-div stabilization. The governing equations are:
$-\nu\Delta \mathbf{u} + \mathbf{u} \cdot \nabla \mathbf{u} + \nabla p - \gamma\nabla(\nabla \cdot \mathbf{u}) = \mathbf{f}, \quad \nabla \cdot \mathbf{u} = 0$
in a domain $\Omega$ , with homogeneous Dirichlet boundary conditions. The authors note that solving the resulting non-symmetric saddle-point linear systems (arising from standard Picard linearization) is computationally expensive, particularly for 3D problems where direct solvers fail due to memory constraints and iterative solvers struggle with the saddle-point structure. While splitting methods are well-established for unsteady flows, their application to genuinely steady problems is less developed.

Methodology
The authors propose a Steady Incremental Viscosity Splitting (SIVS) method. This approach adapts the Incremental Viscosity Splitting (IVS) scheme, previously used for unsteady flows, to the steady-state context. The method operates as an iterative fixed-point scheme:

Step 1 (Velocity Prediction): Given the previous iterate $(\mathbf{u}^{k-1}, p^{k-1})$ , solve an elliptic PDE for an intermediate velocity $\tilde{\mathbf{u}}^k$ :
$a(\tilde{\mathbf{u}}^k, \mathbf{v}) + c^*(\mathbf{u}^{k-1}, \tilde{\mathbf{u}}^k, \mathbf{v}) + b(p^{k-1}, \mathbf{v}) = \langle \mathbf{f}, \mathbf{v} \rangle$
Here, the bilinear form $a(\cdot, \cdot)$ includes the viscous term and the grad-div stabilization term ( $\gamma$ ).
Step 2 (Pressure Correction and Projection): Solve for the updated velocity $\mathbf{u}^k$ and pressure $p^k$ by correcting the intermediate solution:
$a(\mathbf{u}^k - \tilde{\mathbf{u}}^k, \mathbf{v}) + b(p^k - p^{k-1}, \mathbf{v}) = 0$
$b(q, \mathbf{u}^k) = 0$
This step ensures the final velocity is divergence-free and satisfies the boundary conditions.

Algebraic Interpretation
The paper interprets SIVS as an algebraic splitting method. By performing an Incomplete LU (ILU) decomposition on the system matrix, the authors show that SIVS corresponds to a specific block factorization where the Schur complement matrix $S = -B\tilde{A}^{-1}B^T$ (with $\tilde{A}$ representing the linear viscous operator) is spectrally equivalent to the pressure mass matrix. Crucially, unlike standard Picard iterations that require inverting a different Schur complement at every step, the SIVS scheme requires solving a system with a fixed SPD matrix for the pressure variable across all nonlinear iterations. This allows for the pre-computation and reuse of preconditioners, significantly reducing computational cost.

Key Contributions and Theoretical Results

Novel Scheme: Introduction of the SIVS method for steady Navier-Stokes equations with grad-div stabilization, ensuring the discrete velocity field is divergence-free and satisfies full Dirichlet boundary conditions.
Convergence Analysis: The authors prove the uniform boundedness of the iterates and geometric convergence of the scheme under a smallness condition on the data ( $\Lambda < 1/\sqrt{2}$ ).
Algebraic Efficiency: The method is shown to be an algebraic splitting approach that decouples the velocity and pressure solves into simpler linear systems compared to the standard coupled saddle-point system.

Numerical Results
The authors validate the method using FreeFem++ with Taylor-Hood ( $P_2, P_1$ ) elements across three test cases:

Convergence Test: Using a manufactured solution, the method achieved the expected optimal convergence rates for velocity, pressure, and divergence errors. The number of nonlinear iterations remained uniform across mesh refinements.
2D Lid-Driven Cavity: The scheme was tested at Reynolds numbers up to $Re=5000$. Results for velocity profiles and streamlines matched reference data (Ghia et al.) accurately. The iteration counts were comparable to the Incremental Picard-Yosida (IPY) method, but the authors anticipate lower CPU time due to the simpler linear systems.
3D Flow Past a Cylinder: A benchmark problem was solved with $Re \approx 450$ (based on $\nu=10^{-3}$ ). The scheme converged in 11 nonlinear iterations, producing drag, lift, and pressure drop coefficients within the reference intervals provided in the literature.
Parameter Influence: Tests on the grad-div parameter $\gamma$ indicated that larger values of $\gamma$ accelerate convergence to the steady state, though they may deteriorate the conditioning of the linear system in Step 1.

Significance
The paper claims that the SIVS method offers a computationally efficient alternative to standard Picard linearizations for steady incompressible flows. By leveraging the grad-div stabilization and algebraic splitting, the method yields simpler linear systems that are easier to precondition and solve iteratively. This makes the approach particularly suitable for large-scale 3D problems where direct solvers are infeasible. The authors conclude that the method maintains low iteration counts while accurately reproducing standard benchmarks, suggesting its viability for complex multiphysics applications.