Equation & Data Modeling
A versioned laboratory for equations, parameter exploration, data fitting, and model comparison.
The Equation Lab makes mathematical models executable. A user can start from known systems such as Lorenz, Rössler, Kepler orbits, oscillators, interference patterns, and surface models—or write a custom equation and fit a dataset.
The important design choice is versioned state: equations, parameters, layers, data, and camera position can be committed and restored together.
Real-world jobs
- Teach dynamical systems, chaos, oscillation, and parameter sensitivity.
- Test whether a proposed equation can reproduce an observed dataset.
- Compare two mathematical models under the same initial conditions.
- Prototype a simulation before implementing it in a production codebase.
- Preserve a chain of hypotheses and results during exploratory research.
Environment contract
| Layer | Example |
|---|---|
| User | Researcher, engineer, analyst, instructor, or student |
| Inputs | Equations, initial conditions, parameter ranges, and an optional dataset |
| Controls | Model type, parameters, particle count, time rate, layers, and playback |
| Output | Visual trajectories, fitted behavior, parameter values, and saved versions |
| Verification | Known invariants, residuals, analytic cases, or reproduction from a saved commit |
| Boundary | Hypothesis exploration and communication, not proof that a model is physically valid |
Example workflow: investigate chaotic behavior
Question: At what parameter range does the Lorenz system stop converging to a fixed point and begin showing sensitive dependence on initial conditions?
- Load the Lorenz attractor with a small cloud of nearby particles.
- Record the baseline parameters and behavior.
- Change the Rayleigh parameter in small steps.
- Commit representative stable, transitional, and chaotic states.
- Compare how quickly nearby trajectories separate.
Definition of done: the investigation identifies the tested range, preserves reproducible versions, and distinguishes observation from a broader scientific claim.
Example workflow: fit a dataset
Provide a time series with explicit columns and units. Start with a simple model, record its residuals, then add complexity one layer at a time. Keep the simplest model that explains the target behavior within the accepted error bound.
Definition of done: another user can load the same data and version, reproduce the fit, see the residual measure, and understand why the model was selected.
Why version history matters
Exploration becomes unreliable when the only record is the current screen. A useful modeling environment saves the full state behind each conclusion so a reviewer can return to the exact equation, parameters, data, and view that produced it.
This pattern transfers directly to forecasting, pricing experiments, operations research, system identification, and policy simulations.