Solid Earth tides + GR Schwarzschild in the high-precision propagator

[ssmichael1]

Summary

Adds two precision force-model terms to the numerical propagator, both default-on, and restructures the propagation documentation around the natural SGP4 / numerical split.

Closes #16. Closes #109.

What's in the force model

• Solid Earth tides (IERS 2010 §6.2.1 Step 1) — frequency-independent Love-number response. ≈99% of the solid-tide signal at ~5% per-step overhead. New tides module with TideModel enum (None / SolidStep1 / SolidFull), solid_tide_deltas(), and tide_accel(). Step 2 (71 frequency-dependent constituents) deferred; SolidFull currently falls back to Step 1.
• GR Schwarzschild (IERS 2010 §10.3 Eq. 10.12, β = γ = 1) — single closed-form acceleration, ~10 flops per call. New relativity module. Lense-Thirring / de Sitter not yet implemented (sub-cm at MEO/GEO).

Both wired into all three integrator force closures (RKV, RODAS4, GJ8). STM partials skipped for both — ∂a/∂r contributions are ≲1e-12 of the J2 partial.

Settings

PropSettings::default() now ships tide_model: TideModel::SolidStep1 and use_relativistic_correction: true. Behavior change for callers with state-vector regression tests: end-state positions will shift. Pass TideModel::None and use_relativistic_correction=false to reproduce pre-0.18 numerics.

Validation

• test_gps: residual vs ESA SP3 truth on a 1-day GPS arc dropped from 6.42 m (no tides, no GR) → 5.71 m (tides) → 1.86 m (tides + GR + refitted v0). Threshold tightened to 2.5 m, with a refitted hardcoded v0 and Cr*A/m.
• New test_solid_tides_perturb_orbit: confirms ~0.34 m position drift at GEO over half a day from solid tides, matching M&G Table 3.1.
• TestRelativisticCorrection and TestSolidTides Python test classes cover bindings + a functional with-vs-without diff.

Python bindings

• satkit.tidemodel enum (none / solid_step1 / solid_full)
• satkit.propsettings.tide_model and satkit.propsettings.use_relativistic_correction fields, exposed via kwarg ctor + getter/setter
• .pyi stubs updated with docstrings on the new enum and fields

Documentation restructure

The monolithic docs/guide/satprop.md was split into three focused theory pages:

• guide/forces.md — modeled forces with a summary table (now including tides + GR)
• guide/integrators.md — RKV / GJ8, step-size selection, tolerances
• guide/satstate.md — state vectors, STM, covariance, impulsive maneuvers

The "Learn" nav now has two top-level subsections — SGP4 Propagation (1 theory + 3 tutorials) and Numerical Propagation (4 theory + 2 tutorials) — instead of the previous flat 8-item list. Dropped the "Un-modeled Forces" section (was already stale — tides + GR just moved out of it).

Notebook updates

• "High Precision Propagation" → renamed to "GPS Example" (it's a GPS-class orbit-determination worked example, not a generic propagation demo).
• Narrative updated for tides + GR default-on: the empirical NTW acceleration is now framed as a catch-all for what isn't yet covered (ocean tides, box-wing SRP, albedo, Lense-Thirring / de Sitter, EOP residuals).
• Re-executed end-to-end; headline residual unchanged (~0.38 m mean fit) but the fitted NTW components shift meaningfully, indicating the empirical now absorbs different physics.

Plus an unrelated fix to the TLE Fitting tutorial that surfaced during testing (fitresults["success"] → fitresults["converged"], the actual key).

Test plan

• cargo test --lib orbitprop:: — 56 tests pass (53 prior + 3 new in relativity/tides modules + new test_solid_tides_perturb_orbit)
• cargo test --lib orbitprop::propagator::tests::test_gps — passes with new 2.5 m threshold
• pytest python/test/ — 97 tests pass (94 prior + 3 new in TestRelativisticCorrection)
• cargo fmt --all -- --check clean
• mkdocs build — no broken links from the restructure
• Notebook re-execution: HPP/GPS Example runs to completion with new defaults
• Reviewer: confirm the v0 refit values in test_gps reproduce the 1.86 m max-axis residual on your machine (small SP3 / EOP drift can shift this slightly)

🤖 Generated with Claude Code