Imagine you take a photo of a friend with a very cheap camera — it's blurry. You buy a better camera and the photo looks sharper. You buy an even better one… and now it looks exactly the same as the last one. At that point, buying a more expensive camera is a waste of money — the picture won't improve. A grid independence test in CFD works exactly like this. It answers one simple question: "Is my mesh good enough, or do I need more cells?" This guide explains the whole idea in plain language, so any student can understand it and do it correctly.
Table of Contents
What Is a Grid Independence Test?
In CFD (Computational Fluid Dynamics), we cut our flow region into thousands or millions of tiny boxes called cells — together they form the mesh (or "grid"). The computer solves the flow equations inside every cell.
Here's the key point: the number of cells affects your answer. A grid independence test (also called a mesh independence study or grid convergence study) simply checks whether your result still depends on the mesh. You run the same simulation on meshes with more and more cells, and watch one important number — like drag force, pressure drop, or temperature. When that number stops changing as you add cells, your solution is "grid independent," and you can trust it.
Why Do We Need It?
Every CFD result must pass this test before anyone believes it. There are two dangers it protects you from:
- Too few cells (coarse mesh): the answer is inaccurate — it might be off by 10%, 20%, or more. Your beautiful colourful plots would simply be wrong.
- Too many cells (very fine mesh): the answer is accurate, but the simulation takes hours or days and uses huge memory — often for no extra benefit.
The grid independence test finds the "sweet spot" in between. This is why examiners, journal reviewers and companies always ask for it. If your thesis or paper shows CFD results without a grid independence study, the first question you'll be asked is: "How do you know your mesh was fine enough?" This test is your answer.
The Big Idea (in Simple Words)
Think of measuring the length of a curvy road using straight sticks:
- With a few long sticks, you cut corners and get a rough, short length (inaccurate).
- With shorter sticks, you follow the curve better and get a longer, more accurate length.
- With even shorter sticks, the length barely changes — you've captured the true shape.
The cells in your mesh are those sticks. Beyond a certain fineness, making them smaller doesn't change the answer — you've reached the true value. That's grid independence.
How to Do It — Step by Step
How to Read the Convergence Curve
The easiest way to see grid independence is to plot your monitor value on the y-axis against the number of cells on the x-axis.
Read it like this: the steep part on the left means the mesh is too coarse — the answer is still moving. The flat part on the right means you've reached the true value; adding more cells is a waste. You want to pick a mesh right where the curve becomes flat.
The Grid Convergence Index (GCI)
"Less than 1% change" is a good practical rule, but for a thesis or paper you often want a more rigorous, standard number. That's the Grid Convergence Index (GCI), proposed by Roache and based on a clever idea called Richardson extrapolation.
In simple terms, the GCI does two smart things:
- It estimates how far your result is from the true, perfect-mesh value — even though you never actually build an infinitely fine mesh.
- It adds a safety factor (usually 1.25 when you use three meshes) to give a conservative, trustworthy error estimate.
The method uses three grids to calculate an "order of convergence" (how fast your error shrinks) and then reports the error as a percentage. A GCI below about 1% is widely accepted as grid independent. You don't need to fear the maths — the formula is standard, and tools like the free NASA GCI tutorial walk through every step.
A Simple Worked Example
Suppose you simulate air flow through a pipe and monitor the pressure drop (ΔP). You run three meshes:
| Mesh | Cells | ΔP (Pa) | Change vs previous |
|---|---|---|---|
| Coarse | 10,000 | 240 | — |
| Medium | 80,000 | 262 | +9.2% |
| Fine | 640,000 | 264 | +0.8% |
Read the story in the last column. From coarse to medium, ΔP jumps 9.2% — the coarse mesh was clearly too rough. But from medium to fine, it changes only 0.8% — below our 1% rule. So the solution is grid independent at the medium mesh. You would choose the medium mesh (80,000 cells) for the rest of your study: it gives essentially the same answer as the fine mesh but runs about 8 times faster. That's a smart, defensible choice you can put straight into your report.
Common Mistakes to Avoid
- Using only two meshes. You need at least three to see a trend and to apply the GCI.
- Refining too little between meshes. If meshes are almost the same size, the result barely changes and you get a false sense of independence. Keep the refinement ratio ≥ 1.3.
- Changing other settings. Only the mesh should change between runs — not the solver, model or boundary conditions.
- Ignoring mesh quality. A fine but badly shaped (skewed) mesh can be worse than a coarse clean one. Check quality metrics too.
- Choosing the wrong monitor value. Pick something physically meaningful to your problem, not just any random number.
Frequently Asked Questions
What is a grid independence test in CFD?
It is a check to see whether your CFD result still depends on how fine the mesh is. You run the same simulation on several meshes of increasing cell count and watch a key result like drag or pressure drop. When the result stops changing, the solution is grid independent and can be trusted.
Why is a grid independence study important?
Because a result that changes every time you refine the mesh is not reliable. Too coarse a mesh is inaccurate; too fine a mesh wastes computing time. The study finds the smallest mesh that still gives an accurate answer, which is why examiners and journals expect it.
How many meshes do I need?
At least three — coarse, medium and fine. Three is the minimum to see a clear trend and to apply the Grid Convergence Index. Each mesh should be meaningfully finer than the last, with a refinement ratio of at least 1.3.
What is the Grid Convergence Index (GCI)?
A standard way to put a number on how close your mesh is to grid independence. Based on Richardson extrapolation with a safety factor (usually 1.25 for three meshes), it estimates the discretization error. A GCI below about 1% is commonly accepted as grid independent.
When can I stop refining the mesh?
When your monitored result changes by less than about 1% between the medium and fine meshes — the convergence curve has gone flat. At that point extra cells only add computing time, so you pick the coarser converged mesh.
Conclusion
A grid independence test is one of the most important — and most misunderstood — steps in CFD. But the idea is simple: keep adding cells until your answer stops changing, then use the smallest mesh that still gives that answer. Do it with at least three meshes, watch one meaningful result, plot the convergence curve, and — for extra rigour — report the GCI. Get this right and your CFD results become trustworthy, efficient, and ready to defend in front of any examiner or reviewer.
Master this one habit early, and every CFD project you do afterwards will stand on solid ground.
For more beginner-friendly tutorials on CFD, meshing, FEA and simulation, explore Free CFD Tutorial. If this guide helped you, please share it with your classmates.