1.8 KiB
title, description, tags
| title | description | tags |
|---|---|---|
| Differentiation on irregular time grids | See which LawSynth derivative estimators accept non-uniform timestamps, what the validator rejects, and why spectral differentiation needs a regular grid. | irregular grid differentiation, finite differences, cubic spline, derivative estimation |
Differentiation on irregular time grids
LawSynth differentiates a valid non-uniform time series without first resampling it. The irregular-grid entry point validates the time and signal arrays, then delegates to the same three-point Lagrange estimator used by finite differences.
Input contract
The estimator requires:
- at least two aligned time and signal values;
- finite timestamps and finite signal values;
- a strictly increasing time axis, with no duplicate timestamps.
It does not sort timestamps, merge duplicates, impute missing values, or select an interpolation rule. A validation failure is explicit rather than a silent data transformation.
Supported estimators
The local finite-difference estimator uses the adjacent left and right spacings for each interior point. Its three-point formula is exact for a quadratic evaluated on a valid non-uniform grid, but—as with other numerical derivatives—it can amplify measurement noise.
The natural cubic-spline estimator also accepts strictly increasing non-uniform times and returns a derivative at each knot. Because it uses a global spline system, its behavior differs from the local three-point estimate.
Spectral differentiation deliberately rejects irregular grids. Resampling before a spectral method is a separate modeling decision and can introduce interpolation artifacts.
For data preparation, gap handling, and validation advice, see the irregular sampling guide.