Dead zones

Where we lose

We publish where we lose so you don't have to discover it in production. Cases where a competitor wins, or where every arm fails, are listed first. Each row is scoped to one regime — a specific noise level, or clean.

DISADVANTAGE
A baseline wins Use the winning baseline named on the problem page — not SolvSRK.
DEAD ZONE
Everything fails No solver choice rescues this. Reformulate, shorten the horizon, or change tolerances.

Verdicts summarise a whole problem across the noise ladder. Full definitions →

ProblemVerdictσ / scopeNarrative
Arenstorf orbit (under noise)

orbital mechanics · 2026-08

DISADVANTAGE0.1SolvSRK wins clean precision but collapses to 0% survival at σ=0.1, while the SciPy arms hold 100%. On this periodic three-body orbit SolvSRK is the wrong choice under noise.
Lotka–Volterra (under noise)

population dynamics · 2026-08

DISADVANTAGE0.1SciPy is more precise clean and far more robust noisy — SolvSRK survives just 30% at σ=0.1 versus 100% for SciPy. A clear competitor win.
Brusselator, stiff (clean precision)

chemistry · 2026-08

DISADVANTAGEcleanOn clean runs SciPy Radau reaches ~15.5 SCD versus SolvSRK's 12.9. SolvSRK only pulls ahead once noise enters — the clean-precision crown here is Radau's.
Van der Pol μ=10⁴ (clean precision)

oscillator · 2026-08

DISADVANTAGEcleanSciPy Radau leads clean precision (~15.4 vs 12.7 SCD) and both survive noise. There is no SolvSRK advantage to claim on this problem — it is parity at best.
Kepler orbit (clean precision)

orbital mechanics · 2026-08

DISADVANTAGEcleanThe high-order explicit Vern9 integrator is more precise than SolvSRK on clean Kepler (~7.5 vs 6.4 SCD). SolvSRK's advantage is confined to the noisy regime.
Lorenz attractor (clean precision)

chaotic dynamics · 2026-08

DISADVANTAGEcleanChaos caps terminal precision for everyone, and Vern9 edges SolvSRK on clean SCD. Under noise the two are at parity — no exclusive SolvSRK win here.
E5 autocatalytic reaction

chemistry · 2026-08

DEAD ZONEclean + noisyA stiffness ratio near 10^11 defeats every arm. SolvSRK and all 13 baselines fail the acceptance bar at the catalog horizon — a total dead zone, published alongside the wins.

TRL 4–5 · simulation-lab validated · 398 problems · 14 solver arms · clean + 5 noise levels

Freeze: 2026-08-13 · scipy 1.14 · libsolvsrk 2.3.0 · Methodology

Self-reported by Resonix Labs · not independently verified · Verification status