Simulation Guide
Knowledge Guides /Does a smaller simulation time step always improve accuracy? How do you check convergence and computational cost?
Last updated: · New Horizon Digital Outsourcing Team
Smaller steps do not necessarily produce trustworthy results. They help assess some numerical errors but cannot repair incorrect boundary conditions or model assumptions. Compare at least one progressively refined series of results under representative conditions and check whether target metrics stabilize, rather than judging animation smoothness alone.
Define the comparison metrics first
- Select peaks, integrated quantities, steady-state values, or event times that matter to the task. Comparing a few curve screenshots can miss narrow peaks and brief events. Fix initial conditions and inputs and record computation time to assess whether accuracy gains justify the extra cost.
Distinguish different discretization scales
- Time steps, spatial meshes, and co-simulation exchange intervals are different concepts. Keep other settings consistent where possible when changing one. Models with frequent collisions or switching events also require checks of event handling; similar averages do not mean critical events match exactly.
How should unstable results be investigated?
- Check units, initial states, conflicting constraints, and energy anomalies in order, then examine the solver and tolerances. If results keep changing with refinement, report nonconvergence rather than selecting a result that meets expectations. For stochastic models, also distinguish sampling variability from numerical error.
Which records should the report retain?
- List discretization settings, main metrics, and computation times for each run, and explain the final settings chosen. Deliver rerunnable input configurations, rather than only final curves. Convergence applies only to the checked metrics and conditions; it does not mean every output has been verified.