Triple Systems and Strong Equivalence

The hierarchical system PSR J0337+1715 places a neutron star and an inner white dwarf in free fall through the gravitational field of a second white dwarf. If gravitational self-energy changes free fall, the two inner bodies acquire a differential acceleration toward the outer companion. That acceleration forces a polarization of the inner orbit at a phase locked to the outer orbit, making the triple a strong-field universality test rather than merely another measurement of orbital decay. The empirical assessment is current through 2026-08-19.

Required background. Probability and Statistical Inference supplies the joint likelihood and covariance needed to separate a differential acceleration from correlated three-body parameters.

Helpful background. Weak-Field Gravity supplies the external-field limit; Relativistic Two-Body Dynamics supplies the nested orbital dynamics; and Gravitational Radiation and Waveforms supplies the distinction between conservative free fall and dissipative radiation.

Differential acceleration in a hierarchical triple

Let $\mathbf g_{\rm out}$ be the outer white dwarf’s field evaluated across the compact inner orbit. A convenient phenomenological parameter is

$$\mathbf a_{\rm NS}-\mathbf a_{\rm WD}=\Delta\,\mathbf g_{\rm out},$$

for the anomalous external-field response, after the ordinary tidal and relativistic three-body accelerations have been included in the baseline model. The full difference of the two bodies’ accelerations is nonzero even in GR because they occupy different positions and attract each other. The null is $\Delta=0$, not equality of those complete accelerations. In an alternative theory, $\Delta$ can depend on the neutron star’s binding energy, its equation of state, and body-specific scalar or vector charges; it is not automatically the weak-field Eötvös parameter measured with laboratory materials.

To see the origin of the signal, write the anomalous part of each external acceleration as $\delta_A\mathbf g_{\rm out}$. Subtracting the two inner equations cancels the universal external acceleration but leaves $(\delta_{\rm NS}-\delta_{\rm WD})\mathbf g_{\rm out}$, so $\Delta=\delta_{\rm NS}-\delta_{\rm WD}$ at this order. In the relative variational equation this is a periodic forcing term. Its radial projection contains the relative inner–outer phase, which is why the timing signature includes combinations of both orbital frequencies. It cannot be modeled faithfully by adding a constant displacement to an otherwise isolated Kepler ellipse.

The perturbation drives a small component of the inner eccentricity vector aligned with the projected outer acceleration. The signal is therefore not an arbitrary extra eccentricity: its amplitude, phase, and secular modulation are correlated with the outer orbit. A model that changes only the inner mass or inclination generally cannot reproduce all three. The full observable is a set of pulse TOAs fitted with an N-body timing model including the two nested orbits, light-travel time, relativistic delays, astrometry, and a parameter such as $\Delta$.

The discovery system and its hierarchy are described by Ransom et al. 2014, pp. 520–524 and Extended Data. The dedicated universality analysis used a numerical three-body timing model and found no required differential acceleration Archibald et al. 2018, pp. 73–76 and Methods. Freire and Wex explain how that phenomenological result maps into strong-field gravity parameters in their 2024 review, §5.1 and Eqs. (46)–(47).

Results and limitations through 2026-08-19

Position or evidence	Result for the test	Limitation or counterevidence
A phase-locked $\Delta$ term is fitted in a relativistic three-body timing model	Published analyses require no violation of compact-body free fall	The result belongs to one hierarchy and depends on the TOAs, clock, ephemeris, dispersion, and numerical model
Independent timing data and implementations give compatible null results Voisin et al. 2020, §§2–4 and §6	Reduces the chance that one cadence or integrator created the null	Shared source physics, reference frames, and some nuisance assumptions prevent complete independence
A longer Nançay dataset admits either achromatic red noise or a circum-ternary-planet model	Both hypotheses keep $\Delta$ consistent with zero Voisin et al. 2025, §4.4, Eq. (7), and §6	The corresponding 95% limits, $\lvert\Delta\rvert<2.29\times10^{-6}$ and $\lvert\Delta\rvert<1.46\times10^{-6}$, show that the current bound is timing-model dependent
Alternative theories predict body-dependent $\Delta$ through sensitivities or charges	Converts the phenomenological limit into theory-specific information	Equation of state, coupling branch, external field, and screening make the conversion model-dependent

The strongest conservative position is that a compact neutron star and a weakly bound white dwarf fall alike in this external field to the sensitivity of the published timing analyses. A competing deviation interpretation remains possible only in theory regions where the two effective accelerations nearly coincide, the extra field is screened or short-ranged, or the sampled neutron-star mass misses a nonperturbative branch. No robust forced polarization is presently required. The 2025 analysis also makes the timing-model choice part of the present uncertainty rather than a negligible after-the-fact correction. It remains unresolved how fully one triple covers the mass- and environment-dependent space of strong-field charges.

Data, calibration, and likelihood

For a numerical use of the result, the dataset version must identify the primary TOA span, telescope and backend, pulse templates, clock corrections, Solar-System ephemeris, dispersion treatment, and the exact N-body timing code or equations. Calibration includes observatory time transfer, backend offsets, profile evolution, astrometric reference frame, and numerical integration tolerance. The selection records removed radio-frequency-interference intervals, TOA quality cuts, observing frequencies, conjunction handling, and which outer-orbit phases were sampled. This single discovered system is not a complete population sample; its discovery and follow-up selection limit any population statement.

With residual vector $\mathbf r(\Delta,\boldsymbol\eta)$ and covariance $C(\boldsymbol\lambda)$, a normalized likelihood has the form

$$p(\mathbf d\mid\Delta,\boldsymbol\eta,\boldsymbol\lambda) =\frac{\exp[-\mathbf r^{\mathsf T}C^{-1}\mathbf r/2]} {(2\pi)^{N/2}\det(C)^{1/2}}.$$

The nuisance vector $\boldsymbol\eta$ contains component masses, inner and outer initial conditions, reference-frame motion, parallax, dispersion, and any post-Newtonian terms; $\boldsymbol\lambda$ contains white and red noise. Priors and the interval convention for $\Delta$ must be stated. Mutual inclination, mass ratios, secular perturbations, Galactic acceleration, clock and ephemeris errors, and unmodeled pulse-profile changes are systematic uncertainties, not independent scatter to add after fitting.

An independent check injects a known $\Delta$ with the predicted outer-orbit phase into synthetic TOAs, recovers it using a separately implemented N-body integrator, and withholds some outer phases from fitting. An adversarial fit then perturbs masses, inclinations, reference-frame drift, dispersion, and red noise. A gravity interpretation is credible only if these alternatives cannot reproduce both the forced amplitude and phase. The rerun boundary is explicit: a prose description cannot reproduce the published numerical limit without versioned TOAs, covariance choices, and code.

What could change the conclusion. The conclusion could change when a new timing release extends the outer-orbit phase coverage, an independent clock or ephemeris correction changes the residuals, a separate N-body implementation disagrees, or a viable theory calculation changes the mapping from $\Delta$ to compact-body sensitivities. A signal must pass blinded injection recovery and the phase-coherence discriminator before it changes the present null interpretation.

Exercise

Why is a nonzero fitted eccentricity of the inner binary insufficient evidence for differential free fall?

Solution

An ordinary eccentricity vector follows the inner orbital dynamics and can be set by formation or Newtonian perturbations. Differential free fall predicts an additional forced component whose direction follows the projected acceleration from the outer white dwarf. Therefore one must compare both amplitude and phase over the outer orbit. Mass, inclination, or initial-eccentricity errors may change the amplitude, but they generally cannot reproduce the prescribed outer-phase locking; that independent phase relation is the essential discriminator.

References

• Archibald, A. M., et al. (2018), “Universality of free fall from the orbital motion of a pulsar in a stellar triple system,” Nature 559, 73–76. DOI.
• Freire, P. C. C., and Wex, N. (2024), “Gravity Experiments with Radio Pulsars,” Living Reviews in Relativity 27, article 5. DOI.
• Ransom, S. M., et al. (2014), “A millisecond pulsar in a stellar triple system,” Nature 505, 520–524. DOI.
• Voisin, G., et al. (2020), “An improved test of the strong equivalence principle with the pulsar in a triple star system,” Astronomy & Astrophysics 638, A24. DOI.
• Voisin, G., et al. (2025), “Explanation of the exceptionally strong timing noise of PSR J0337+1715 by a circum-ternary planet and consequences for gravity tests,” Astronomy & Astrophysics 693, A143. DOI.