Public demonstrations
See it run.
Seven demos. Real benchmarks. Published problems.
Every demo below runs on published literature or public datasets — Lorenz (1963), Van der Pol (1920), Robertson (1966), IEEE EMTP, IEEE-CIS, CelesTrak TLEs. Zero proprietary data. Full citations. The numbers are real.
SolvSRK vs. the field on textbook problems
SolvSRK is our hybrid ODE solver — it handles the messy, real-world simulations where standard solvers crash or slow to a crawl. Think of it as the engine that keeps running when sensors are noisy and equations get wild.
Technical: Benchmarked against SciPy and SUNDIALS baselines at matched tolerances (atol=1e-8, rtol=1e-8). Where noise is applied, it matches real sensor conditions. Every problem comes from published literature (Hairer-Wanner, DETEST, Lorenz 1963).
Lorenz Attractor
The iconic butterfly attractor. Chaotic 3D system with sigma=10, rho=28, beta=8/3. Under additive sensor noise (sigma=0.01), standard solvers collapse step size while SolvSRK filters and integrates.
Source: Lorenz (1963)
| Solver | Steps | Wall time | Max rel. error | Survived |
|---|---|---|---|---|
| SolvSRK | 4,218 | 12.3 ms | 2.1e-7 | pass |
| SciPy BDF | 18,420 | 48.7 ms | 3.8e-7 | pass |
| SciPy Radau | 14,830 | 42.1 ms | 2.9e-7 | pass |
| SciPy RK45 | 6,250 | 8.4 ms | 1.4e-6 | pass |
| SciPy LSODA | 7,890 | 14.2 ms | 8.7e-7 | pass |
| CVODE BDF | 15,600 | 38.5 ms | 3.2e-7 | pass |
Want to see this on your problems?
These demos run on textbook benchmarks and public datasets. The real story is what happens on your system, with your noise, at your tolerances. We offer a 30-day evaluation pilot on your actual problems.
Talk to us