Note on authorship: Throughout this document, the term “Chapront” is used as a shorthand reference to the collaborative analytical work of Jean Chapront and Michelle Chapront-Touzé at the Bureau des Longitudes in Paris.
No distinction is intended between their respective contributions, and all references to "Chapront", the "Chapront model" or "Chapront series" should be understood to mean the full joint authorship and analytical legacy they share.
For centuries, predicting the Moon’s position required immense numerical labor. Astronomers relied on printed lunar tables derived from the theories of Hansen, Hill, and ultimately E. W. Brown, whose monumental Tables of the Motion of the Moon became the global standard in the early twentieth century. These tables were not simple lookup charts — they encoded thousands of empirical and semi-analytical corrections, refined across generations of observation.
Every eclipse prediction, every occultation timing, every nautical longitude solution depended on evaluating and interpolating those printed numbers by hand. The process was rigorous but fragile: precision depended on arithmetic care, and updating constants required wholesale recomputation.
By the late twentieth century, digital computation made it possible to replace tabular interpolation with direct analytical evaluation. Jean Chapront and Michelle Chapront-Touzé at the Bureau des Longitudes transformed the lunar theory into compact trigonometric series expressed in the Delaunay arguments. Instead of consulting tables, the Moon’s longitude, latitude, distance, and librations could be computed directly from explicit formulas:
This shift was not merely computational convenience. It marked the transition from numerical tabulation to fully deterministic analytical ephemerides — continuous functions of time that could be evaluated, differentiated, and embedded directly into software. The Moon’s motion had moved from printed pages into executable mathematics.
By the early 1990s, Chapront and Chapront-Touzé developed a fully analytical lunar theory—ELP 2000—that unified centuries of empirical terms into a consistent modern framework.
Each lunar quantity—longitude (\(\lambda\)), latitude (\(\beta\)), distance (\(\Delta\)), and the three physical librations (\(l, b, c\))—is represented as a sum of periodic series built from the five Delaunay arguments.
Each term’s amplitude (\(a_j, b_j\)) and argument coefficients (\(\alpha_j … \varepsilon_j\)) were determined from a combination of classical theory, numerical integration, and spacecraft data.
The resulting model reproduces the Moon’s motion to the arcsecond level in position and to a few tenths of an arcsecond in physical libration over the 1900–2150 interval — well within the requirements of traditional almanac and eclipse work.
Unlike earlier tables, Chapront’s series are deterministic and differentiable: each variable is a continuous analytic function of time, not an interpolated dataset. This structure makes the model compact enough for real-time evaluation in browsers or embedded systems, yet faithful to the full physical dynamics encoded in modern ephemerides.
By the 1990s, NASA’s SPICE system (developed at the Jet Propulsion Laboratory) became the authoritative framework for storing and serving planetary and lunar ephemerides. Instead of analytic series, SPICE expresses each quantity—position, orientation, and libration—as a piecewise Chebyshev polynomial, valid over fixed time segments.
For librations, the general form is:
At runtime, SPICE evaluates the polynomial using Clenshaw’s recurrence, a numerically stable algorithm that constructs recursively:
Each segment of coefficients is stored in a binary PCK kernel, typically spanning 32–64 days. During evaluation, SPICE chooses the proper segment, normalizes the time variable τ, and reconstructs the libration or coordinate value with machine-precision accuracy.
This approach eliminates all trigonometric evaluation, greatly reducing runtime cost. SPICE kernels can represent any physical quantity—rotations, orientations, positions, or their time derivatives—within a unified data structure. For the Moon, SPICE achieves residuals of a few milliarcseconds, far exceeding the accuracy of analytic theories.
While SPICE provides the highest-fidelity representation of lunar motion, its binary kernel format and evaluation chain are not ideal for lightweight or browser-based tools. To bridge that gap, we generated a uniformly sampled dataset directly from SPICE, evaluating each quantity and its time derivative at 15-minute cadence over the years 1900–2150 CE.
Importantly, both the function values and the derivatives were obtained directly from SPICE state vectors and orientation models. No finite differences were used in the production archive. This ensures that the interpolant inherits the same dynamical consistency as the underlying kernel realization.
Intermediate values are reconstructed using cubic Hermite interpolation. For any quantity \(f(t)\) with known values and derivatives at endpoints \(t_0\) and \(t_1\), the interpolant is:
For sufficiently smooth underlying motion, the interpolation error scales as \(O(h^4)\), where \(h\) is the sampling interval. Halving the sampling cadence reduces the error by roughly a factor of sixteen.
A quintic Hermite formulation was evaluated during development, but was rejected. Although quintic interpolation matches second derivatives, it introduced greater sensitivity to endpoint curvature and amplified high-frequency numerical noise present in the sampled data. In practice, quintic fits produced larger maximum deviations than cubic fits over the 15-minute grid. Because the underlying SPICE segments are themselves polynomial reconstructions of integrated motion, cubic Hermite proved to be the most stable and numerically consistent transport layer.
In effect:
This dual system preserves SPICE-level smoothness and sub-arcsecond precision while enabling fast evaluation in JavaScript, Python, or other client-side environments without requiring binary kernel parsing.
Even within the 1900–2150 range, small secular drifts are expected because the
lunar rotation and orbit models continue to evolve. Each new SPICE kernel
incorporates refined dynamical constants, updated tidal parameters, and
long-baseline lunar laser-ranging (LLR) fits. These incremental adjustments—
typically a few milliarcseconds per decade in orientation or tens of meters
in position—reflect improved knowledge, not errors in prior kernels.
Consequently, a set of Hermite coefficients generated from one kernel
(for example, moon_pa_de440_2000.tpc) will diverge slightly from
those derived from a later release (de441, de442,
etc.).
For high-precision research, users should periodically refresh their coefficient tables using the latest SPICE data. For visualization, eclipse prediction, and educational tools, the current implementation remains more than adequate: the cumulative drift over two centuries remains well below any observable threshold.
Although both Chapront’s analytical model and SPICE’s numerical kernels describe the same physical motion of the Moon, they differ fundamentally in how that motion is represented.
| Aspect | Chapront (Analytic) | SPICE (Numerical) |
|---|---|---|
| Representation: | Finite trigonometric series in Delaunay arguments | Piecewise Chebyshev polynomials |
| Form: | Explicit functions of time | Stored coefficients per time segment |
| Continuity: | Global and analytic | Local to each segment (continuous, differentiable) |
| Accuracy: | ≈1″ in position, 0.1″ in libration (1900–2150) | Milliarcsecond level, spacecraft-calibrated |
| Dynamical coupling: | Implicit in fitted series coefficients | Explicit n-body numerical integration |
| Storage: | Text-based coefficients (~40 MB total) | Binary PCK (~2 MB for same coverage) |
| Evaluation: | Trigonometric sums | Polynomial recursion (Clenshaw) |
| Best suited for: | Education, browsers, long-term trends | Research, navigation, high-precision timing |
Our implementation allows both systems to be evaluated from a single time input (Julian Date TT). When a user supplies the epoch:
Between the 15-minute SPICE samples, the Hermite interpolator provides continuous values. This enables a live comparison of analytic and numerical librations at arbitrary times without requiring the user to load or execute SPICE kernels directly.
To verify that the Hermite interpolation reproduces SPICE-kernel values at arbitrary times, we performed a dense sampling test between all 15-minute grid points across the 1900–2150 span. For each interval, the analytic Hermite reconstruction was compared to direct SPICE evaluation at 1-minute cadence.
The results show that the interpolated values remain indistinguishable from direct SPICE evaluation to within 0.01″ RMS and 0.05″ maximum deviation, well below the measurement noise of the underlying kernel and far beneath any visual significance in eclipse or libration visualizations. In terms of rotation angles, this corresponds to fractional errors well below visual or eclipse-relevant thresholds, confirming that the Hermite representation preserves positional and angular continuity at practical numerical precision.
This means that for browser or educational use, the interpolation effectively is the SPICE result—delivering the same precision without any kernel dependencies or binary parsing on the client side.
Interpolation accuracy depends on the sampling interval \(h\). To quantify this behavior, multiple cadences were tested by regenerating the SPICE sample grid at different spacings and comparing Hermite reconstructions to direct SPICE evaluation at dense (1-minute) resolution.
With a 15-minute grid, the cubic Hermite reconstruction agrees with direct SPICE evaluation to within ≈ 0.01″ RMS and ≈ 0.05″ maximum deviation over the full 1900–2150 interval.
When the grid spacing is halved to 7.5 minutes, the RMS interpolation error decreases by approximately a factor of sixteen, consistent with the expected \(O(h^4)\) convergence of cubic Hermite interpolation. Conversely, increasing the grid spacing to 30 minutes raises the RMS error by roughly an order of magnitude, but still keeps it below visually or eclipse-relevant thresholds.
These tests confirm that the interpolation layer behaves as a true transport mechanism rather than a re-fitting procedure: its accuracy is governed by mathematical order, not empirical tuning.
SPICE achieves its accuracy by integrating the full n-body equations of motion and fitting to modern spacecraft tracking data. The Chapront model, though purely analytical, remains invaluable for education, legacy reproduction, web presentation, and verification.
Because its form is explicit, it can be differentiated, expanded, or transformed symbolically—something far more difficult with opaque binary kernels. In practical use, Chapront provides intuition and traceability; SPICE provides precision.
Chapront’s analytical outputs \((\lambda,\,\beta,\,\Delta)\) are expressed in the mean ecliptic and equinox of date — the classical reference frame of analytical lunar theory. SPICE, by contrast, provides data in the true equator and equinox of date. To compare them directly, the reference frames must be reconciled by applying nutation in longitude (Δψ) and obliquity (Δε).
The mean obliquity in arcseconds is evaluated as:
where \(t\) is in Julian centuries (TT) from J2000.0.
The IAU 2000A or 1980 nutation model provides Δψ and Δε as trigonometric series in the Delaunay arguments. The leading terms, driven by the regression of the Moon’s node Ω, dominate:
Ω, the Moon’s nodal longitude, appears in both Chapront’s series and the nutation model. It decreases by ≈ 19.35° per year, completing a full retrograde cycle in roughly 18.6 years. This nodal regression drives the dominant term in Δψ and Δε, coupling the Moon’s orbital motion with the Earth’s axial precession.
Therefore,
The corrected ecliptic coordinates \((\lambda',\,\beta)\) are then rotated to the true equatorial frame using the obliquity \(\varepsilon'\):
Finally, the right ascension and declination follow as:
A comprehensive comparison between Chapront’s analytical results and SPICE kernel outputs over the 1900–2150 CE interval shows:
These results demonstrate that the Hermite interpolation preserves the full fidelity of SPICE while maintaining a smooth, differentiable representation ideal for web and real-time visualization.
The present implementation spans 1900–2150 TT, aligning with the interval over which both the Chapront ELP 2000/82 analytical theory and the standard JPL SPICE lunar orientation kernels maintain uniform accuracy. This range encompasses all modern observations while avoiding epochs where historical constants or incomplete tidal models dominate. It also coincides with the period for which Earth–Moon parameters have been most rigorously constrained by lunar laser-ranging data.
Even within these limits, users should expect very small cumulative drifts between successive kernel releases. SPICE is not a frozen model —it is a living framework continuously refined as new retroreflector data, spacecraft tracking, and planetary constants become available. These revisions slightly alter the Moon’s computed orientation and libration history, producing minute differences (on the order of milliarcseconds) between kernel generations.
For typical scientific and visualization work, the current dataset provides a consistent and stable foundation. However, the same framework can be extended seamlessly: by sampling any newer SPICE kernel at 15-minute cadence and regenerating the Hermite coefficients, users can carry the model forward indefinitely. The interpolation and visualization code remain unchanged; only the underlying coefficient tables are replaced. In this way, the system is both self-contained and future-proof, mirroring the continuous refinement that defines modern lunar ephemerides.
The development of lunar ephemerides is not a sequence of discarded models, but a progression of representations of the same physical reality. Each generation preserved what came before while changing the mathematical language in which the Moon’s motion was expressed.
The great nineteenth-century lunar tables transformed observation into predictive structure. Brown’s theory consolidated those efforts into a coherent analytical framework. Chapront and Chapront-Touzé then recast the lunar problem in modern trigonometric series form, making the motion of the Moon compact, deterministic, and computationally explicit.
SPICE represents the next stage of that evolution. Instead of expressing the Moon’s motion as a fixed analytical expansion, it encodes the results of full n-body numerical integration into piecewise Chebyshev polynomials. The physics did not change — only the computational realization.
In this sense, Chapront and SPICE are not competing systems. They are two realizations of the same gravitational dynamics: one analytic and symbolic, the other numerical and integrated. The analytical form provides transparency and mathematical traceability. The numerical form provides dynamical self-consistency and extensibility.
The present implementation intentionally preserves both. Chapront’s series remain invaluable for understanding structure, derivation, and long-period behavior. SPICE provides the operational geometric standard required for eclipse prediction, spacecraft navigation, and high-precision timing.
From handwritten tables to trigonometric series to binary kernels, the objective has remained constant: to describe the Moon’s motion compactly, reproducibly, and with steadily increasing fidelity. This project does not replace one era with another — it demonstrates how they connect.
SPICE Acknowledgment: This work makes use of the NASA Navigation and Ancillary Information Facility (NAIF) SPICE system, developed and maintained by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
The following technical works include both direct sources used in this page and additional papers originally cited by Yuk Tung Liu in his Notes on Libration Calculation. They are repeated here for convenience and continuity.