Intro to Computational Fluids

Discretization, stability, and convergence in introductory flow simulations.

Conceptual context

Computational fluid dynamics turns a continuous flow into a finite set of cells, variables, and algebraic equations. Each cell carries approximations of $u$, $v$, pressure, and other properties. $h$ determines which flow structures can be resolved and how far $u_num$ lies from $u_ref$. A coarse grid smooths gradients and boundary layers; a fine grid increases cost and may require a smaller $dt$. The central task is not producing an attractive image, but demonstrating that $\epsilon _u$ decreases consistently under refinement without changing the physical problem.

🟢 Essential level

A fluid simulation divides the domain into small cells and computes how the fluid moves through them.

$h$ describes the spatial scale of those cells. Reducing it allows the grid to represent changes in $u$, $v$, and pressure more accurately, especially near walls and inlets. Consequently, $\epsilon _u$ usually decreases under refinement. A finer mesh does not repair incorrect boundaries, an unreliable $u_ref$, or an unsuitable physical model. It can also create instability when $dt$ is not adjusted. The correct practice is to compare several meshes while keeping $Re$, geometry, and boundary conditions fixed. A regular error trend supports convergence; an irregular curve indicates that another mechanism controls the result. Refinement is therefore a test of the numerical model, not merely a request for more detail. It must also be evaluated with a quantity defined consistently across meshes: comparing a pointwise velocity on one mesh with an averaged value on another would create a false impression of improvement.

🔵 Formal level

Basic formulas

Mass conservation in a two-dimensional incompressible flow requires the local defect $div_u$ to vanish:

Continuity for incompressible flow

The dynamic regime is characterized by $Re$, built from $\rho$, $U$, $L$, and $\mu$:

Reynolds number

For transient simulations, advective travel per cell is controlled through $CFL$, which links $dt$ and $h$:

Advective CFL number

The discrepancy between $u_num$ and $u_ref$ is measured by:

Relative velocity error

Real student questions

Is a finer mesh always better? Only when spatial error dominates and the model, solver, and time step are appropriate.

Why must $Re$ remain fixed? Because it defines the physical regime. Changing it means comparing different flows.

What does small $div_u$ mean? The computed field approximately satisfies local mass conservation.

Can a stable simulation still be inaccurate? Yes. A suitable $CFL$ prevents some instabilities but does not guarantee spatial resolution.

What if the reference velocity is near zero? Use absolute error or a global norm because pointwise relative error becomes misleading.

Final synthesis

A trustworthy flow simulation is demonstrated, not assumed. $h$ should reduce $\epsilon _u$ consistently while conservation, regime, and stability remain controlled. One mesh gives a result; several meshes provide evidence.