Implicit time integration can remove a strict stability barrier without removing the temporal-resolution requirement. If important structures move too far, oscillate too quickly or change too much during one step, the answer can be damped or phase-shifted even while the run remains stable.
Problem: the transient solution depends strongly on timestep
Typical symptoms include damped pressure oscillations, shifted peak loads, smeared free surfaces, unstable phase fractions, wrong shedding frequency, torque ripple that changes with Δt, or temperature transients that lag measured behavior.
A timestep-independence study should therefore monitor the engineering output itself—not merely the number of inner iterations required per step.
Identify the timescale the simulation must resolve
- Convective transport.
How long does fluid take to cross a relevant cell, gap, jet diameter, passage or wake structure?
- Periodic physics.
Blade passing, vortex shedding, pulsation, wave period or imposed motion sets a frequency that needs multiple samples per cycle.
- Interface motion.
In VOF or moving-front problems, the interface should not jump through too many cells per step.
- Thermal response.
Solid thermal inertia can be slow, while fluid-side convection may be much faster; both may matter depending on the quantity of interest.
- Moving mesh / rotating geometry.
Angular displacement per step determines how accurately rotor-stator interaction and geometry motion are sampled.
Estimate the convective timestep
Use velocity and representative cell size to calculate a starting Courant number and timestep.
Warning signs that Δt is too large
Peak damping
Maximum pressure, force, torque or temperature is lower on coarse timesteps because fast changes are numerically averaged.
Frequency shift
Dominant spectral content changes when timestep is reduced, indicating inadequate temporal sampling.
Phase error
The timing of peaks or wave arrival shifts even if mean values remain similar.
Interface smearing
VOF structures diffuse or chatter because the interface moves too far relative to the local mesh per step.
Inner-iteration struggle
Each physical step requires many corrections because the solution changes too much between consecutive states.
Mesh/timestep coupling
Local refinement decreases cell size, raising the local Courant number unless timestep is reduced accordingly.
How to diagnose timestep sensitivity
- Plot local or maximum Courant number.
Use it as a transport-resolution indicator, especially near jets, interfaces, narrow gaps and rotating interfaces.
- Monitor the engineering signal at every physical step.
Do not output so infrequently that high-frequency behavior is hidden by sampling.
- Compare mean, amplitude, phase and frequency.
Two runs can have the same mean while disagreeing strongly on transient loads.
- Check inner convergence within each timestep.
A small Δt with poorly converged inner iterations can still be inaccurate.
- Inspect the first transient separately from statistically stationary behavior.
Startup sensitivity should not be confused with timestep dependence of the final periodic or statistical state.
A practical timestep-independence study
- Choose a baseline Δt from physics.
Use Courant number, motion increment or samples-per-period—not an arbitrary round number.
- Run a smaller timestep, commonly by a factor such as two.
Keep mesh, schemes and boundary conditions identical.
- Compare the quantity of interest over the same physical window.
For periodic problems, compare equal cycles after transients have decayed.
- Reduce again if the engineering difference is still material.
Stop when further reduction changes the decision-relevant metric by an acceptably small amount.
- Document the chosen Δt and evidence.
This turns timestep from a hidden solver setting into a verified modelling choice.
Common mistakes
- Assuming an implicit solver allows arbitrarily large timesteps.
- Using only global Courant number when a small critical region controls the physics.
- Reducing timestep without increasing output sampling or inner convergence quality.
- Comparing runs over different physical durations or different startup phases.
- Changing mesh and timestep simultaneously, making the sensitivity source impossible to identify.
- Judging timestep independence from residuals instead of the engineering signal.
Related Abecator resources
Need help setting a defensible transient timestep?
Submit the mesh scale, velocity/motion timescale and engineering signal you need to resolve.