Benchmarks

What

Converts sky positions from RA/Dec (equatorial) to ecliptic longitude/latitude, using the obliquity of the ecliptic.

Why

Ecliptic coordinates are natural for solar system objects. The transform depends on the obliquity model used.

Interpretation

Lower separation = better agreement with ERFA's IAU 2006 obliquity model.

Performance contract

Timed scope: obliquity evaluation + rotation per epoch. Input parsing is outside the timed loop.

What

Rotates a direction vector from the ICRS celestial reference frame to the True-of-Date frame using the Bias-Precession-Nutation (BPN) matrix.

Why

This is the fundamental coordinate transformation used in astrometry. Differences indicate how well each library models Earth's axis wobble.

Interpretation

Lower error = closer to ERFA's IAU 2006/2000A model. Siderust uses simpler Meeus precession + IAU 1980 nutation, so some offset is expected.

Performance contract

Timed scope: matrix construction + single matrix-vector multiply per epoch. Input JD conversion and output formatting are outside the timed loop. All adapters construct the full BPN matrix from scratch.

What

Converts celestial RA/Dec to local azimuth/altitude for a ground observer, using sidereal time and spherical trigonometry.

Why

This is the 'where do I point my telescope?' calculation. Accuracy depends on the GAST (sidereal time) model used.

Interpretation

Lower separation = better. Differences mainly arise from GAST model choice. libnova uses a different sidereal-time model from SOFA/IERS and is not rankable for accuracy in this experiment.

Performance contract

Timed scope: GAST computation + spherical trig per epoch+location. Input parsing (JD, RA/Dec, lon/lat) is outside the timed loop.

What

Computes GMST (how far the Earth has rotated relative to the stars) and ERA (the raw rotation angle) at given epochs.

Why

Time-scale conversions underpin all ground-based astronomical observations. Small errors here compound in coordinate transforms.

Interpretation

Lower error = better agreement with ERFA's IAU 2006 polynomial. Libnova uses Meeus formula which differs at the arcsecond level.

Performance contract

Timed scope: GMST/ERA evaluation per epoch pair (JD_UT1, JD_TT). Input parsing and output serialisation are outside the timed loop. Astropy adapter uses public astropy.time APIs with an explicit UT1 override derived from the benchmark input pair.

What

Solves Kepler's equation M = E - e·sin(E) for the eccentric anomaly E, then computes the true anomaly ν. Tests convergence across eccentricities.

Why

Kepler's equation is fundamental to orbital mechanics. Different solvers (Newton-Raphson vs bisection) have different convergence properties.

Interpretation

Lower residual = better convergence. Libnova's bisection method converges to ~1e-6 deg, while Newton-Raphson methods reach ~1e-15 rad.

Performance contract

Timed scope: iterative solver per (M, e) pair. Input parsing is outside the timed loop. This is a numerical convergence benchmark, not an ephemeris accuracy claim.