SolvSRK-UQ

Release: Production (GA)

Single-solve uncertainty quantification for SolvSRK

What SolvSRK-UQ is

A licensed capability on the same libsolvsrk / solvsrk package — not a separate solver, wheel, or API surface. When your seat carries features: uq (trial or UQ add-on) or features: edge (Edge, which includes UQ), you set emit_uq=True on a normal solvsrk.run() call and receive y(T) ± σ(T) per state component at the terminal time.

What it does

Question	Without UQ	With SolvSRK-UQ
State at horizon	`y_final` — point estimate	`y_final` — best estimate
Confidence	Run Monte Carlo (N re-integrations)	`uq_sigma` — single-pass envelope
Cost at runtime	O(N × integration cost)	O(1 × integration cost)
Certification story	MC sample size debates	Documented `noise_sigma` + receipt

The envelope models uncertainty from noisy early integration at scale noise_sigma — sensor error, batch variation, model mismatch in the front window — not a full Bayesian posterior and not a replacement for domain validation.

Where to use it

Trial evaluation — every 30-day SolvSRK trial includes UQ; test envelopes before you buy
Certification / MRB — defensible terminal bounds without an MC farm in the loop
Real-time go/no-go — reactor trip, flight envelope, medical dosing: y ± σ at T in one call
Desktop simulation — purchased UQ add-on for teams that do not need Edge but need envelopes

SolvSRK-Edge includes UQ — this page is for desktop/sim customers who subscribe to base SolvSRK and add UQ, or who are evaluating UQ during trial.

How to enable it

import solvsrk
 
assert solvsrk.license_has_uq()  # trial, UQ add-on, or Edge seat
 
result = solvsrk.run(
    cfg, ndim, t0, t_end, y0, rhs_fn=rhs,
    emit_uq=True,
    noise_sigma=1e-3,  # tune to instrument / model noise scale
)
# result["y_final"], result["uq_sigma"], result["uq_t_handoff"]

Deep walkthrough: Journey 05 Diagnostics.

Trial vs purchase

Phase	UQ available?
30-day SolvSRK trial	Yes — `features: uq` on trial `.lic`
Purchased SolvSRK (base)	No — integrator only; `emit_uq` → `failure_mode=license`
Purchased SolvSRK-UQ add-on	Yes — reissue `.lic` with `features: uq`
Purchased SolvSRK-Edge	Yes — `features: edge` implies UQ

What UQ is not

Not Monte Carlo — one integration pass, not N re-runs
Not a ROM or surrogate certificate
Not bundled in purchased base desktop SolvSRK (trial and add-on/Edge only)
Not the same product as SolvLRDE UQ on pricing PDEs — different problem class

When to use it

Use UQ when you need defensible bounds for go/no-go decisions, certification evidence, or real-time loops where MC is too slow — reactor kinetics, flight envelopes, PK/PD terminal concentrations, any ODE where input noise in the early window could move the endpoint.

See Examples for how this edition maps to the SolvSRK journey suite.

Who it is for

Teams evaluating SolvSRK who need terminal envelopes during the 30-day trial
Certification-heavy industries needing defensible uncertainty bounds
Real-time decision loops where Monte Carlo is too slow
Risk-quant and regulatory-affairs teams pricing structured products on ODE models

What ships

Same libsolvsrk / solvsrk binary as SolvSRK
Terminal envelope via emit_uq / noise_sigma
Diagnostic trace via uq_ek_handoff_trace()
Entitlement check: solvsrk.license_has_uq()

Related products

Base integrator: SolvSRK. Included on Edge: SolvSRK-Edge.

Install