Validation

Disclaimer

Not certified for critical applications

This software is provided as-is and is not certified for flight-critical, safety-critical, or life-critical applications. Independent verification and validation is required before operational use.

General Validation Guidance

When integrating the library into an analysis pipeline:

1. Cross-reference results against trusted external tools such as JPL Horizons before relying on them for operational decisions.
2. Verify kernel coverage — confirm that the loaded SPICE kernels are correct and complete for the analysis interval.
3. Check time-system assumptions — be explicit about TDB vs. UTC and verify conversions at domain boundaries.
4. Account for numerical precision — long-duration or high-sensitivity propagations accumulate integration error. Compare against reference solutions at intermediate epochs, not only at the final epoch.
5. Perform independent V&V — production workflows should include an independent verification step before operational use.

SSA-Specific Validation

For conjunction assessment and CDM workflows:

1. Validate collision-probability behavior against trusted SSA references for the probability method family being used.
2. Inspect EncounterQualityFlags before using Pc values in operational decision support. Flags such as low relative velocity or covariance remediation are applicability warnings.
3. Confirm covariance realism — verify that participant covariances and hard-body radii are realistic and current. The library does not propagate covariance to TCA; see Conjunction Assessment.
4. Schema-validate CDM products using Cdm.ValidateSchema() or Cdm.ValidateSchemaFromXml() before exchanging them with external systems.
5. Treat edge cases carefully — low-relative-velocity encounters may require analysis beyond the standard 2D Pc workflow.

Conformance Tests

The conformance cases come from the conformance-tests suite: inputs, golden outputs produced by an external reference tool, and tolerances. Two harnesses run them:

• the conformance runner IO.Astrodynamics.ConformanceRunner runs every case with the adaptive RK7(8) integrator (absolute and relative tolerances 1e-11) and writes a report, IO.Astrodynamics.Net/conformance-report.json;
• unit tests run the three 24-hour propagation cases with RK7(8) (RK78IntegratorTests.Conformance00x_*) and with Velocity-Verlet at a fixed 1 s step (PropagatorTests.Conformance00x_*), and print the errors they measure.

Both evaluate the final state at the exact end of the window, whose UTC epoch is converted to TDB, like the reference tool's elapsed time, and report the magnitude of the position and velocity errors.

Measured Errors

Measured on 2026-10-04. RK7(8): conformance runner report (IO.Astrodynamics 10.x, the framework commit is recorded in the report). Velocity-Verlet: output of the PropagatorTests.Conformance00x_* tests.

Case	Orbit and force model (inputs)	Golden	RK7(8), 1e-11	Velocity-Verlet, 1 s
propagator_24h_leo_grav10_001	LEO about 6,800 km, 24 h; EGM2008 10x10, Moon, Sun	GMAT R2025a, JGM3 10x10	13.13 m, 14.8 mm/s	294.3 m, 332.0 mm/s
propagator_24h_geo_grav70_002	GEO (INTELSAT 901), 24 h; EGM2008 70x70, Moon, Sun, planets	GMAT R2025a, JGM3 70x70	8.02 m, 0.57 mm/s	8.48 m, 0.61 mm/s
propagator_24h_sso_grav10_003	SSO 700 km, 24 h; EGM2008 10x10, Moon, Sun	GMAT R2025a, JGM3 10x10	3.56 m, 3.84 mm/s	246.4 m, 258.4 mm/s
propagator_1sd_geo_twobody_004	GEO equatorial, one sidereal day; two-body	Kepler's third law	0.08 m, 0.006 mm/s	not run
propagator_1sd_geo_i30_twobody_005	GEO inclined 30 deg, one sidereal day; two-body	Kepler's third law	0.08 m, 0.006 mm/s	not run

The test limits are non-regression guards set a few percent above these values, not accuracy specifications:

Case	RK7(8) limits	Velocity-Verlet limits
propagator_24h_leo_grav10_001	13.5 m, 15.5 mm/s	300 m, 350 mm/s
propagator_24h_geo_grav70_002	8.3 m, 0.6 mm/s	9 m, 0.7 mm/s
propagator_24h_sso_grav10_003	4 m, 4 mm/s	250 m, 260 mm/s

With RK7(8), the remaining few meters come from the differences between the models of the two tools (see Error Budget). With Velocity-Verlet at 1 s, the LEO and SSO errors are almost entirely the truncation error of the integrator: 279.5 m over 24 h on the two-body LEO orbit alone (see Integrators). At GEO the orbital motion is slow enough for a 1 s step to be nearly exact.

Run the conformance runner, next to a clone of conformance-tests:

cd IO.Astrodynamics.Net
dotnet run -c Release --project IO.Astrodynamics.ConformanceRunner -- \
    ../../conformance-tests IO.Astrodynamics.Tests/Data/SolarSystem conformance-report.json

Run the unit tests and see the measured errors:

cd IO.Astrodynamics.Net
dotnet test IO.Astrodynamics.Tests/IO.Astrodynamics.Tests.csproj \
    --filter "FullyQualifiedName~Conformance00" --logger "console;verbosity=detailed"

Error Budget

Known sources of the few-meter residual between IO.Astrodynamics (RK7(8)) and the GMAT goldens. None of them has been isolated yet; each open point is tracked.

Source	IO.Astrodynamics	Golden (GMAT)	Effect	Status
Gravity field	EGM2008, tide-free, truncated to the case degree	JGM3 at the same degree	Not measured; expected to dominate	Open: goldens to regenerate with EGM2008 (conformance-tests#5)
Earth orientation of the gravity field	SPICE ITRF93 with earth_latest_high_prec.bpc, EOP included	GMAT's own Earth-fixed reduction and EOP file	Not measured	Open question
Inertial frame	Frame.ICRF (SPICE J2000, ICRF-aligned)	EarthMJ2000Eq (EME2000)	At most the ~23 mas frame bias, about 0.8 m in LEO and 4.7 m in GEO, depending on how GMAT ties its ephemerides to EarthMJ2000Eq	Open question
Ephemerides	de440s.bsp	de440s.bsp, the same file (identical MD5)	None	Verified
Constants (Earth GM, reference radius of the field)	From the EGM2008 field and the SPICE kernels	GMAT defaults and the JGM3 file; Earth.Mu is not overridden in the scripts of cases 001 to 003	Not measured	Open question
Time scales	UTC window converted to TDB	Elapsed time ElapsedSecs = 86400	Sampling the end of the window in the wrong scale shifts it by about 18 µs, i.e. 0.14 m in LEO (measured: 12.99 m against 13.13 m)	Resolved: tests and runner both use the UTC end converted to TDB
Integration error	RK7(8), tolerances 1e-11	PrinceDormand78, accuracy 1e-13	0.08 m over one sidereal day at GEO (two-body cases 004 and 005)	Measured

Reference Validations

Beyond the conformance cases, the test suite compares the library with published references:

Reference	What is compared	Tests	Tolerance
NASA CARA Pc2D_Foster_UnitTest, Omitron case 01	Foster 2D collision probability	CollisionProbability_NasaOmitronReferenceCase_MatchesPublishedValue; through the entry points, Analyze_NasaOmitronReferenceCase_MatchesPublishedValueThroughEntryPoint and AnalyzeAll_NasaOmitronReferenceCase_MatchesPublishedValueThroughEntryPoint	1e-5 relative to the published Pc
NASA CARA Pc2D_Foster_UnitTest, Alfano case 03	Foster 2D collision probability	CollisionProbability_NasaAlfanoCase03_MatchesPublishedValue, Analyze_NasaAlfanoCase03_MatchesPublishedValueThroughEntryPoint, AnalyzeAll_NasaAlfanoCase03_MatchesPublishedValueThroughEntryPoint	1e-5 relative to the published Pc
Frisbee, maximum Pc with a single covariance	Critical covariance matrix and Table 1 outcomes	SingleCovarianceMaximumPc_FrisbeeCriticalCovarianceMatchesPublishedMatrix, SingleCovarianceMaximumPc_FrisbeeTable1_MatchesPublishedApproximateOutcomes	1e-6 absolute on the matrix; 3 % relative on the four Table 1 cases
IAU SOFA, release 2023-10-11	Frame bias, precession, nutation, CIP, CIO locator, Earth rotation angle, GCRS to CIRS and to TIRS	Iau2006ModelTests, Iau2006FundamentalArgumentsTests, FrameChainIntegrationTests, Iau2006SofaSweepTests (300 epochs, 1990 to 2040), with reference values from the harnesses in tools/sofa_reference	From 1e-14 rad (precession angles) to 5e-9 rad (IAU 2000B against SOFA's IAU 2000A); see Standards & Units

See Also

• High-Fidelity Propagation
• Conjunction Assessment