Pulsar Timing as a Gravity and Matter Probe

A pulsar supplies a phase-coherent clock, but gravity is inferred only after radio voltages have been calibrated, pulse profiles have been reduced to times of arrival, and propagation, terrestrial-clock, Solar-System, spin, and binary delays have been modeled together. In a relativistic binary, two independent post-Keplerian observables can determine the two masses within a chosen theory; every additional observable then overdetermines that theory. The same timing record can constrain neutron-star matter through masses, glitches, and spin–orbit coupling, although only the last of these directly exposes a moment of inertia and it is not yet a competitive independent measurement.

Required background. Frames, Coframes, and Anholonomy supplies the distinction between observatory, barycentric, and local pulsar measurements.

Helpful background. Flat-Background Hilbert Stress-Energy from Metric Variation supplies radiation and orbital-energy balance; The Einstein Equation supplies relativistic propagation and orbital dynamics; Relativistic Matter and Compact Objects supplies neutron-star structure and moment-of-inertia physics; Computational Relativity supplies likelihood, convergence, and reproducibility checks.

From pulse phase to a timing residual

Let $t_{\mathrm{obs}}$ be a time tag at an observatory. Clock corrections and the observatory position first map it to a Solar-System-barycentric time. A timing model then removes dispersion and the Rømer, Einstein, Shapiro, and aberration delays. The resulting pulsar emission-time variable $\tau$ is inserted in a phase model such as

$$\phi(\tau)=\phi_0+\nu(\tau-\tau_0) +\frac{1}{2}\dot\nu(\tau-\tau_0)^2+\Delta\phi_{\rm glitch}+\cdots .$$

Here $\phi$ is measured in cycles and $\nu$ is the spin frequency in cycles per unit time. Choose a time residual to mean observed time minus predicted time. For pulse number $N$, let $\tau_{\rm obs}$ be its inferred emission time, while $\phi(\tau_{\rm model})=N$. Linearization gives $r_\tau=\tau_{\rm obs}-\tau_{\rm model}\simeq[\phi(\tau_{\rm obs})-N]/\nu$. Propagating this through the same timing model gives the arrival-time residual. Stating the sign avoids confusing a phase residual with a time residual. A smooth residual is not automatically gravitational: an incorrect sky position, frequency-dependent dispersion, pulse-profile evolution, a backend offset, red spin noise, or an unmodeled glitch can imitate a deterministic term. Wide frequency coverage separates the approximately $\nu_{\rm radio}^{-2}$ dispersive delay from achromatic orbital structure, while a phase-connected solution accounts for every rotation across the data span.

The phenomenological Damour–Deruelle construction fits Keplerian and post-Keplerian parameters before assigning them a theory. Its delay terms and parameter logic are developed in Damour and Taylor 1992, §§II–IV and eqs. (2.1)–(3.17). This separation is essential: fitting a general-relativistic mass directly is efficient for parameter estimation, but it cannot by itself constitute a test of general relativity.

Why post-Keplerian parameters overdetermine a theory

Write $n_b=2\pi/P_b$, $M=m_p+m_c$, and $x=a_p\sin i$, with the inherited convention $c=1$ and explicit $G$. At Newtonian order the measured orbital period and projected semimajor axis give the mass function

$$f=\frac{n_b^2x^3}{G} =\frac{(m_c\sin i)^3}{M^2}.$$

General relativity predicts, at leading post-Newtonian order,

$$\dot\omega =\frac{3(GM)^{2/3}n_b^{5/3}}{1-e^2}, \qquad r=Gm_c, \qquad s=\sin i,$$

where $r$ and $s$ are the Shapiro range and shape. The Einstein-delay amplitude $\gamma$ and intrinsic orbital decay $\dot P_b^{\rm int}$ provide different mass-dependent curves. Explicit leading expressions, including the quadrupole-radiation term, appear in Freire and Wex 2024, §3.4.2, eqs. (28)–(34).

Under general relativity, any two nondegenerate post-Keplerian measurements fix $(m_p,m_c)$; a third must cross the same point in the mass–mass plane. With $N$ independently measured post-Keplerian quantities, this supplies up to $N-2$ consistency tests. Independence here is physical, not merely a count of fitted columns: correlated Shapiro parameters, shared astrometry, and a common noise model must remain in the covariance. In an alternative theory, the same phenomenological observables can depend on additional strong-field charges and on the stellar equation of state, so the mapping must be rebuilt rather than reusing the general-relativistic curves.

Timing reaches matter physics in three distinct ways. A precisely measured mass restricts equations of state that cannot support it. Glitch steps and recoveries constrain participating angular-momentum reservoirs, but require a torque and coupling model. In a relativistic binary, the Lense–Thirring contribution to $\dot\omega$ depends on spin orientation and the pulsar moment of inertia. For the Double Pulsar, the second-post-Newtonian and spin–orbit terms are comparable enough to require modeling, yet the present intrinsic $\dot P_b$ uncertainty does not yield a competitive moment-of-inertia constraint; Freire and Wex 2024, § 4.2.2, pp. 37–38 states this boundary.

The named Double Pulsar dataset

Evidence cutoff: 2026-08-18. The primary worked evidence is the 16.2-year PSR J0737−3039A/B analysis by Michael Kramer et al. Its accepted times of arrival run from MJD 52760 to 58667 and combine six observatories. The paper records receiver, backend, frequency, bandwidth, dedispersion, sampling, span, count, and residual information for each stream in Kramer et al. 2021, §§III–IV, Table I, and Appendix A. Two related products served different roles: 4-minute samples constrained astrometry and dispersion variations, whereas 30-second samples resolved the relativistic binary delays. The article describes 199,913 accepted 4-minute times of arrival and 916,648 accepted 30-second times of arrival.

The Australia Telescope National Facility catalogue is useful for discovering systems and literature, but it is not the Double Pulsar likelihood. At the cutoff, the reproducible catalogue identity is Australia Telescope National Facility 2026, PSRCAT version 2.8.1 release-history entry, released 2026-06-18 with 4,393 catalogued pulsars. A catalog-derived population claim must additionally freeze its query, reference codes, survey selection, and download checksum.

Kramer et al. report the observing metadata and analysis procedure, but the article itself does not identify a public machine-readable release containing the full accepted time-of-arrival arrays and covariance-ready configuration. Consequently this page treats the paper as a named analysis dataset, not as a claim that an end-to-end public rerun is presently possible. Reproduction requires either an author-supplied data package or an archive record that contains those precise selected products.

Calibration, selection, and statistical model

For the named analysis, calibration included observatory-specific noise-diode and polarization procedures, radio-frequency-interference excision, coherent or incoherent dedispersion as recorded per backend, frequency-dependent pulse templates, fitted backend and band jumps, and EFAC/EQUAD rescaling. The astrometric stage used Tempo2 and the DE436 Solar-System ephemeris; the binary stage used Tempo with added higher-order terms. A rerun must freeze the time standard, clock-correction files, observatory coordinates, ephemeris, template files, dispersion convention, and exact software revisions—not merely cite the package names.

Selections must preserve the observing span, frequency coverage, phase connection, removed radio-frequency interference and discrepant times of arrival, eclipse treatment, and profile-evolution rules. Kramer et al. excluded Westerbork, Effelsberg, Jodrell Bank, and the Nançay 2.5 GHz stream from the 30-second relativistic fit because their sampling or precision did not serve that analysis. Changing those cuts without rerunning template, noise, and covariance estimation changes the effective experiment.

A transparent reanalysis would write a normalized Gaussian residual likelihood with deterministic timing design matrix and a covariance containing white, red-spin, dispersion, and profile-noise components. It would declare priors on parallax, proper motion, inclination, masses, backend jumps, noise amplitudes, Galactic acceleration, and—when inferred—the moment of inertia and equation of state. It would also state whether intervals are frequentist confidence regions or Bayesian credible regions, give optimizer or sampler settings and convergence diagnostics, and perform posterior-predictive residual checks. No population prior or survey selection is needed for a single-system consistency test; both become necessary when generalizing from catalogued binaries.

The physical orbital decay is not the fitted value alone. Let $\hat K$ point from the Solar-System barycentre toward the pulsar, let $\boldsymbol a_{\rm CM}=\boldsymbol a_{\rm pulsar\ system}-\boldsymbol a_{\rm SSB}$ be the pulsar-system centre-of-mass acceleration relative to the Solar-System barycentre, and let $d$ be the distance. Then

$$\left(\frac{\dot P_b}{P_b}\right)_{\!\rm int} =\left(\frac{\dot P_b}{P_b}\right)_{\!\rm obs} -\boldsymbol a_{\rm CM}\mathbin{\cdot}\hat K-\mu^2d,$$

where the last terms represent differential acceleration and the Shklovskii effect in units with $c=1$. Distance, Galactic-potential model, ephemeris, dispersion and scattering changes, red noise, pulse-shape evolution, inclination, higher post-Newtonian terms, spin-down mass loss, and Lense–Thirring coupling are consequential alternatives. The Double Pulsar analysis explicitly propagates several of them in Kramer et al. 2021, §§IV–VI, Tables II–V, and Appendices A–C.

This page has not downloaded the private or public telescope voltages, reconstructed profiles, fitted times of arrival, or executed Tempo, Tempo2, or a new likelihood. A credible rerun package would contain immutable data and template checksums, all clock and ephemeris files, the parameter file, selection log, covariance and priors, environment lockfile, commands, posterior or least-squares products, and alternate-ephemeris and alternate-noise results.

Check your understanding

Kramer et al. report for the Double Pulsar $\dot P_b^{\rm int}/\dot P_b^{\rm GR}=0.999963\pm0.000063$. How far from unity is this ratio in standard-deviation units, and does it by itself show a significant failure of the quadrupole prediction?

Solution

The standardized residual is

$$z=\frac{1-0.999963}{0.000063} =\frac{0.000037}{0.000063} \simeq0.59.$$

The result lies only $0.59$ standard deviations from unity, so it is consistent with the prediction. This calculation is a diagnostic of the reported corrected ratio, not an independent timing test: its validity still depends on the distance, acceleration correction, timing covariance, and mass inference used to predict $\dot P_b^{\rm GR}$.

References

• Australia Telescope National Facility. “ATNF Pulsar Catalogue.” PSRCAT version 2.8.1, released 18 June 2026. Official version history.
• Damour, Thibault, and Joseph H. Taylor. “Strong-Field Tests of Relativistic Gravity and Binary Pulsars.” Physical Review D 45 (1992): 1840–1868. DOI.
• Freire, Paulo C. C., and Norbert Wex. “Gravity Experiments with Radio Pulsars.” Living Reviews in Relativity 27 (2024): 5. DOI.
• Kramer, Michael, et al. “Strong-Field Gravity Tests with the Double Pulsar.” Physical Review X 11 (2021): 041050. DOI; Open preprint.