Integration-step sensitivity
Check whether a conclusion changes when the numerical step changes.
Research note · November 14, 2026Research question
Is a simulated failure a property of the modeled system or of the integration settings?
Method
- Choose a bounded scenario with fixed initial state and inputs. Run it at a declared sequence of decreasing integration steps while keeping the physical model unchanged.
- Compare quantities relevant to the question, such as peak displacement, contact force or settling time. Define acceptable differences before inspecting the final sweep.
- Retain unstable and failed runs. Inspect whether the apparent result converges as the step changes, and record computational cost alongside the physical outputs.
Records to retain
- Integrator and step settings
- Initial state and forcing
- Per-run output traces
- Failure and cost observations
Source basis
MuJoCo documents numerical integration choices and their interaction with model dynamics. Step-size sensitivity is therefore a necessary question when a result depends on fast contacts, stiff behavior or rapid control changes.
Interpretation & limits
A stable numerical result can still come from an inaccurate physical model. This proposed sensitivity study is not a general validation of a simulator or of Imprint.
Next experiments
- Run a step-size sweep on a simple mechanism.
- Separate numerical divergence from a modeled operating-limit violation.
References
Technical reference
Implementation details supporting this note, drawn directly from the maintained documentation. These are technical references, not additional papers or new experimental results. Expand a topic to inspect its method, fields and limits.