Cfl Condition

The Courant-Friedrichs-Lewy condition, almost always called the CFL condition, is the stability rule that governs grid spacing in any explicit time-stepping simulation: the time step must stay small enough that information cannot cross more than one grid cell per step, which ties mesh resolution and time step together rather than letting either be chosen freely.

What Governs Grid Spacing in a Time-Stepped Solver the CFL condition links mesh spacing and time step through wave speed time step Δt grid spacing Δx → fine coarse CFL limit: Δt = Δx / v unstable — step too large stable region chosen operating point Halving grid spacing forces the time step down by the same factor, not independently. That coupling, not accuracy alone, is what actually governs grid spacing in an explicit solver.

The rule itself is a single ratio. For a wave or signal moving at speed v on a grid with spacing Δx, the Courant number C equals vΔt divided by Δx, and an explicit scheme stays stable only while C is at or below 1; push the time step past that line and the scheme does not just lose accuracy, it blows up numerically within a handful of steps.

This is the real reason fine meshes get expensive fast. Halving the grid spacing to resolve a smaller feature does not just double the number of grid points, it also forces the time step down by roughly the same factor to keep the Courant number in bounds, so a 2x spatial refinement in a 3D explicit solver can cost close to an order of magnitude more compute, not 2x.

Pick grid spacing
resolve the smallest feature
Find local wave speed
light, sound, or diffusion speed
Apply the CFL conditioncompute the maximum stable time step
Compute dt_max
dx / v, scaled by dimension
Check Courant number
C = v dt / dx must stay at or below 1
Run stable time steps
no refinement needed mid-run

Different solvers hit the condition differently. Explicit electromagnetic solvers such as Lumerical FDTD and MEEP must satisfy the CFL condition every single step, which is why their grid spacing is chosen around the smallest feature and the shortest wavelength in the design, not the other way around; implicit solvers such as Synopsys Sentaurus TCAD and COMSOL trade that constraint away by solving a coupled system at each step, letting mesh and time step be chosen independently at the cost of a more expensive linear solve.

Solver typeTime-step ruleGrid-spacing cost
Explicit FDTD (Lumerical, Meep)dt bounded by CFL every stepfiner mesh forces much smaller dt
Explicit finite-difference diffusiondt bounded by dx squaredfiner mesh is even more expensive
Implicit solvers (Sentaurus TCAD, COMSOL)unconditionally stablemesh and time step can be chosen independently
Particle-in-cell plasma codesdt bounded by Debye length and CFL togethermesh must resolve the Debye length first

Plasma and device simulations add a second constraint on top. A particle-in-cell plasma solver used for etch and deposition chamber modeling must resolve the Debye length with its grid before the CFL condition even becomes the binding limit, because an under-resolved Debye length lets numerical heating corrupt the physics regardless of how small the time step is — grid spacing in these tools is really governed by two conditions stacked on top of each other, not CFL alone.

Read grid spacing through a *stability-coupling* lens rather than an *accuracy* lens: a finer mesh is not simply "more accurate for free," because the CFL condition chains that mesh choice directly to a smaller time step and a steeper compute bill, which is exactly why production tools default to implicit schemes wherever the physics allows it and reserve explicit, CFL-limited solvers for problems like FDTD electromagnetics where resolving fast wave propagation in time is the whole point.

Take cfl condition further

Ask the copilot about this term, or have our engineers assess it against your process.