The Mass-Distance Degeneracy
Why Le Verrier got Neptune’s longitude right but its distance wrong, and what this teaches about inverse problems.
The surprise is not finding Neptune
The most striking result of the Neptune search is not that the optimizer finds a planet — it is which parameters behave best across different problem setups. In the synthetic benchmark the recovered mean longitude matches the seed value exactly (it has to, by construction). In JPL comparison mode the fitted mean longitude is biased by tens of degrees relative to Neptune’s actual λ at 1781, but that bias is remarkably insensitive to the assumed semi-major-axis prior — across very different a-ranges the longitude barely moves. The distance (semi-major axis), by contrast, wanders freely depending on the bounds you choose. Le Verrier had exactly the same experience in 1846: his predicted longitude was accurate enough for Galle to spot Neptune on the first night of looking, yet his predicted distance of 36 AU was 20% too large.
This is not a failure of the method. It is a fundamental property of the inverse problem.
The degeneracy
The cleanest way to see the degeneracy is to interrogate the optimizer directly. Pin Neptune’s distance at a series of values from 26 to 44 AU and, at each one, let the optimizer refit the mass, eccentricity, and longitude. If distance were well constrained, fit quality would collapse away from 30 AU. Instead the fit stays nearly as good across the whole range — the optimizer simply trades mass for distance along a ridge — while the recovered longitude barely moves.
Note
The figure above traces the ridge deterministically by refitting at fixed distances. For a genuine sampled posterior over all parameters — same ridge, with calibrated uncertainties — see Notebook: Bayesian Inference.
Higher mass compensates for greater distance. The left panel’s ridge is a one-dimensional family of solutions that all but erase the anomaly: the fit RMS creeps from under 0.1 to about 1.3 arcseconds across the ridge, and even that worst point removes over 98% of the ~106-arcsecond no-Neptune signal. Meanwhile the right panel shows the contrast that matters: while the mass changes 11-fold along the ridge, the recovered longitude stays within about ±16° of the seeded truth. The data pin down where Neptune is in the sky far more tightly than how far away it is.
Physical intuition
The reason is rooted in the scaling of gravitational perturbations. At leading order, the angular displacement that a perturber of mass \(m\) at distance \(a\) imposes on Uranus goes as
\[ \Delta \lambda \propto \frac{m}{a^2} \]
This is the familiar tidal scaling: the perturbing force falls off as \(1/r^2\), and the lever arm for angular displacement introduces another factor of distance in the denominator. Any \((m, a)\) pair satisfying
\[ \frac{m}{a^2} \approx \text{const} \]
produces the same amplitude of perturbation and therefore fits the residuals equally well. The ridge traced above is the locus of this constraint.
(Strictly, this \(m/a^2\) scaling applies to a static perturber at large separation. For a co-orbiting planet like Neptune, the coupling involves Laplace coefficients \(b_{3/2}^{(1)}(\alpha)\) where \(\alpha = a_{\text{Uranus}} / a_{\text{Neptune}}\), and the dependence on distance is more complex. The tidal scaling is a useful first approximation that captures the qualitative positive correlation, but it overstates the sharpness of the degeneracy ridge at moderate \(\alpha\).)
The phase of the perturbation — the direction in which Uranus is pulled — is a different story entirely. It depends on Neptune’s angular position in its orbit (the mean longitude), not on how far away Neptune sits. Because the perturbation signal stretches over 65 years of observations and Neptune’s orbital period is 165 years, the data sample enough of the phase curve to lock down the longitude tightly.
In parameter space, longitude is approximately orthogonal to the degenerate mass-distance direction: the ridge runs along combinations of \((m, a)\) that preserve the perturbation amplitude, while the longitude is pinned by the perturbation phase, which is independent of the amplitude. This is why Le Verrier got the longitude right despite getting the distance wrong.
Why Le Verrier got the longitude right
Le Verrier did not have an optimizer to interrogate, but he faced the same degeneracy. His solution was to break it by assumption: he started from the Titius-Bode Law value of 38.8 AU — an empirical spacing rule that had correctly anticipated the positions of Uranus and the asteroid belt — and used it as a prior constraint on the fit. His refined result settled at about 36.154 AU, much closer to Bode than to the true 30.07 AU. With distance effectively pinned down, the remaining parameters (mass, eccentricity, longitude) were well-determined.
We can reproduce this directly. Running the optimizer with three different bounds on the semi-major axis shows how the distance assumption propagates through the solution.
The distance swings by more than 10 AU depending on the assumed bounds, but the longitude barely moves. This is exactly Le Verrier’s experience: his predicted longitude of about 326 degrees was off by only 1 degree, while his predicted distance of 36 AU overshot the true 30 AU by 20%. The Bode’s Law assumption gave him a wrong distance but — crucially — it did not contaminate the longitude.
The lesson for inverse problems
The mass-distance degeneracy is not a quirk of celestial mechanics. It is a textbook example of non-identifiability in inverse problems: when two or more parameters trade off against each other, the data can constrain certain combinations of them far better than it can constrain either one alone.
Degeneracy ridges appear throughout the sciences:
- Seismology. The velocity and depth of a subsurface layer are anti-correlated; travel-time data constrain the product (velocity \(\times\) depth) but not the factors separately.
- Pharmacokinetics. Absorption rate and bioavailability trade off in plasma concentration curves; the peak height constrains a ratio, not the individual values.
- Radial-velocity exoplanets. The classic \(m \sin i\) degeneracy — radial velocity measurements constrain the product of planet mass and the sine of orbital inclination, leaving the true mass undetermined without additional information.
In each case the practical message is the same: some parameters are better constrained than others. Recognizing which parameters live along the ridge and which are pinned down is often more valuable than chasing a single “best fit.” Le Verrier understood this instinctively. He reported a longitude with confidence and treated the distance as provisional — a remarkably modern attitude toward parameter estimation, a century before Bayesian inference entered mainstream science.