SolvSRK-UQ: Integration

Preconditions

import solvsrk
 
assert solvsrk.license_valid()
assert solvsrk.license_has_uq(), (
    "Need trial .lic, SolvSRK-UQ add-on, or Edge seat — "
    "base purchased SolvSRK returns failure_mode=license"
)

Check entitlement before enabling UQ in production pipelines. Without features: uq or features: edge on the seat file, emit_uq refuses cleanly.

Choosing noise_sigma

noise_sigma is the assumed scale of input noise in the early integration window — instrument error, actuator jitter, batch-to-batch variation. It is not a universal constant:

Start from sensor spec or identified model mismatch
Run Journey 05 with a bracket of values
Use uq_ek_handoff_trace() to audit why uq_sigma grew large or stayed tight

Enable the terminal envelope

import numpy as np
import solvsrk
 
def rhs(t, y):
    return np.array([-y[0], -2.0 * y[1]])
 
cfg = solvsrk.config_defaults()
 
result = solvsrk.run(
    cfg, ndim=2, t0=0.0, t_end=1.0, y0=(1.0, 1.0), rhs_fn=rhs,
    emit_uq=True, noise_sigma=1e-3,
)
print(result["y_final"], result.get("uq_sigma"), result.get("uq_t_handoff"))

Interpretation: treat y_final[i] as the estimate and uq_sigma[i] as a ± band for go/no-go at terminal time only — not a full trajectory cloud.

C API

solvsrk_config_t cfg;
solvsrk_config_defaults(&cfg);
cfg.emit_uq = 1;
cfg.noise_sigma = 1e-3;
solvsrk_run(&cfg, ndim, t0, t_end, y0, rhs_fn, ...);

Diagnostics (audit trail)

trace = solvsrk.uq_ek_handoff_trace(
    cfg, ndim, t0, t_end, y0, rhs_fn, noise_sigma=0.0,
)
# (t, h, y, sigma) samples from the noisy early window — for certification review

Runnable walkthrough: 05 Diagnostics (hub: UQ Examples).

What UQ is not

Not residual-bound ROM certification
Not Monte Carlo — single pass only
Not a substitute for validating the underlying physics model
Not included in purchased base SolvSRK without the UQ add-on or Edge seat

Related products

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

Examples Licensing