{
  "id": "pulsar-timing-array",
  "title": "Pulsar Timing Array",
  "lang": "en",
  "source_lang": "en",
  "summary": "A pulsar timing array uses precise observations of millisecond pulsars to detect nanohertz gravitational waves through correlated changes in pulse arrival times.",
  "category": "science",
  "subcategory": "astronomy",
  "type": "concept",
  "aliases": [
    "PTA",
    "pulsar timing network"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/pulsar-timing-array/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/pulsar-timing-array/",
      "json": "https://aiwiki.page/zh/pulsar-timing-array.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A pulsar timing array (PTA) is a collection of precisely timed pulsars used to search for extremely low-frequency gravitational waves. Rather than measuring the separation of laboratory test masses, it measures changes in the arrival times of radio pulses from rapidly rotating neutron stars distributed across the sky. A passing gravitational wave changes the propagation of these pulses, producing a characteristic pattern of timing fluctuations among different pulsars. The effective detector therefore extends across a substantial part of the Milky Way, although its measurements are made with radio telescopes on or near Earth.\n\nPTAs are most sensitive to waves with periods of years to decades, corresponding approximately to nanohertz frequencies. This band contains signals expected from pairs of supermassive black holes orbiting in merging galaxies, as well as possible contributions from processes in the early universe. In 2023, several collaborations reported evidence for a gravitational-wave background through the spatial correlations that distinguish gravitational waves from many competing sources of timing noise.\n\nThe method combines long-term astronomical observation, relativistic modeling, precision timekeeping, and statistical inference. Its central challenge is not simply to measure a pulsar accurately, but to identify a weak common signal among numerous imperfect astronomical clocks while accounting for propagation effects, instrumental changes, and errors in terrestrial reference systems."
    },
    {
      "heading": "Pulsars as astronomical clocks",
      "text": "A pulsar is a rotating, magnetized neutron star whose emission beams sweep across the observer’s line of sight. Individual radio pulses can vary substantially in brightness and shape, but averaging hundreds or thousands of rotations usually produces a stable mean pulse profile. The arrival phase of this profile can be measured against a template, allowing an observation to be assigned a pulse time of arrival. The clock-like behavior arises from the predictable rotation of a compact object with a large moment of inertia, not from perfectly identical individual pulses.\n\nMost PTA observations concentrate on millisecond pulsars, which rotate hundreds of times per second. Many acquired their rapid rotation through the transfer of matter and angular momentum from a companion star. Compared with ordinary young pulsars, they generally exhibit slower secular evolution and less irregular rotational behavior. The first millisecond pulsar, PSR B1937+21, was discovered in 1982 with a rotation period of about 1.56 milliseconds. Another important timing source, PSR J0437−4715, has a period of about 5.76 milliseconds and is unusually bright at radio wavelengths.\n\nTiming precision depends on more than rotation rate. Useful sources have high radio brightness, narrow and stable profile features, relatively quiet rotation, and propagation paths that can be modeled accurately. A narrow feature provides a better phase reference than a broad, smooth pulse with the same total signal strength. Some of the best observations achieve arrival-time uncertainties of tens to hundreds of nanoseconds, but this performance is not typical of every pulsar or every observing epoch. The uncertainty assigned to one arrival time must also be distinguished from the stability of the pulsar over many years.\n\nA PTA is not an array in the usual radio-interferometric sense. Its pulsars are separate celestial sources, and their radio electric fields are not combined coherently to form an image. Instead, independently measured timing records are compared statistically. Multiple terrestrial telescopes can observe the same pulsar, while one telescope can monitor many members of the array. The essential geometry is the distribution of pulsar directions across the sky.\n\nNo pulsar is a perfect clock. Rotational irregularities, variable propagation through ionized gas, and changes in the observed pulse profile can all perturb its timing. Consequently, an array cannot be constructed merely by selecting the single most stable source. Many well-distributed pulsars are necessary to distinguish a gravitational-wave correlation pattern from the behavior of individual objects and from errors that affect the entire observing system."
    },
    {
      "heading": "From radio observations to timing residuals",
      "text": "The basic observable is a pulse arrival time referenced to a terrestrial time standard. Radio receivers record a finite band of frequencies, and digital processing corrects the frequency-dependent smearing caused by propagation through free electrons. The data are then folded at the predicted rotation period to build an average profile. Comparing that profile with a template yields a phase offset and hence a time of arrival. Modern analyses may instead fit frequency-resolved profiles directly, retaining information that would otherwise be compressed into a single arrival time.\n\nObservatory timestamps cannot be compared directly with a simple model of pulsar rotation. Earth moves around the Sun, rotates about its axis, and occupies a changing position within the Solar System. Timing software therefore transforms observations to a suitable relativistic time coordinate associated with the Solar System barycenter, its center of mass. This requires observatory coordinates, clock corrections, and a Solar System ephemeris giving the positions and masses of relevant bodies. The light-travel correction associated with Earth’s orbit can reach roughly 500 seconds, vastly exceeding the nanosecond-to-microsecond effects of interest.\n\nThe timing model predicts the pulsar’s rotational phase as a function of time. A simplified expansion is\n\n\\[\\phi(t)=\\phi_0+\\nu_0(t-t_0) +\\frac{1}{2}\\dot{\\nu}_0(t-t_0)^2+\\cdots,\\]\n\nwhere \\(\\nu_0\\) is the spin frequency and \\(\\dot{\\nu}_0\\) its derivative at reference epoch \\(t_0\\). Additional terms describe position, proper motion, parallax, propagation delays, and, when relevant, binary orbital motion. Relativistic effects in binary systems may include gravitational time delay and orbital precession. Maintaining a phase-connected solution means accounting for every rotation across the observing span, even when observations are separated by weeks.\n\nA timing residual is the difference between an observed arrival time and the corresponding model prediction. Residuals are the primary data products used in many PTA searches, but they are not unprocessed measurements of gravitational waves. Fitting the timing model absorbs some possible signals. For example, fitting spin frequency and its derivative removes components resembling linear and quadratic trends, while position errors produce annual structures that interact with searches near a one-year period.\n\nThis fitting process makes the response dependent on cadence, observation span, and the parameters included in the model. Analysis must therefore propagate timing-model uncertainty rather than treating the fitted residuals as an unaffected record. A signal with a period much longer than the observations may appear mainly as a slowly varying trend and be largely removed. The lowest useful frequencies are consequently set not only by the nominal data duration but also by the way the timing solution is constructed."
    },
    {
      "heading": "Response to gravitational waves",
      "text": "In general relativity, gravitational waves are propagating perturbations of spacetime. A wave passing through the Earth–pulsar system changes the observed frequency of the pulse train. Integrating this fractional frequency shift over time gives a timing residual. The effect is therefore different from a simple fixed delay: it is a changing discrepancy between the observed pulsar clock and the timing model.\n\nFor a plane wave, a common expression for the response of pulsar \\(a\\) is\n\n\\[z_a(t)= \\frac{1}{2} \\frac{p_a^i p_a^j}{1+\\hat{\\Omega}\\cdot\\mathbf p_a} \\left[ h_{ij}(t_{\\rm E})-h_{ij}(t_{\\rm P}) \\right], \\qquad r_a(t)=\\int^t z_a(t')\\,dt'.\\]\n\nHere \\(\\mathbf p_a\\) points from Earth toward the pulsar, \\(\\hat{\\Omega}\\) specifies the wave’s propagation direction, and \\(h_{ij}\\) is the metric perturbation. Repeated spatial indices are summed. The overall sign of \\(z_a\\) depends on whether the fractional shift is defined using frequency or pulse period; consistent conventions give the same physical residuals.\n\nThe expression contains an Earth term, evaluated when the pulse is received, and a pulsar term, evaluated when it was emitted. Their separation in wave phase depends on the pulsar distance and direction. For a pulsar thousands of light-years away, these terms sample the source at epochs separated by thousands of years. The Earth terms provide the principal correlated response across an array. Pulsar terms usually have effectively unrelated phases for different pulsars in a stochastic-background search, although they remain physically important and can carry information about individual sources.\n\nA useful scale follows from integrating an approximately sinusoidal fractional frequency shift. For a wave of strain amplitude \\(h\\) and frequency \\(f\\), the induced residual has characteristic size\n\n\\[r\\sim \\frac{h}{2\\pi f},\\]\n\napart from angular and polarization factors. A strain of \\(10^{-15}\\) at \\(10^{-8}\\) hertz therefore corresponds to a timing scale of roughly 16 nanoseconds. Small strain can produce a measurable timing perturbation because the effect accumulates over a long wave period.\n\nA data set lasting \\(T\\) years has its lowest Fourier scale near \\(1/T\\). Twenty years corresponds to about 1.6 nanohertz. Observations every few weeks nominally sample higher frequencies as well, but sensitivity is shaped by irregular cadence, measurement noise, and the steep spectra expected for major source populations. A PTA is thus neither a detector operating at one frequency nor a rigid instrument with equal sensitivity across a sharply bounded band."
    },
    {
      "heading": "Spatial correlations and the Hellings–Downs curve",
      "text": "A shared low-frequency process is not, by itself, sufficient evidence for gravitational waves. Errors in terrestrial time standards or planetary ephemerides can also produce signals in many pulsars. The decisive information is how timing fluctuations correlate between pairs of pulsars as a function of their angular separation. For an isotropic, unpolarized background of the transverse tensor waves predicted by general relativity, this dependence is described by the Hellings–Downs curve.\n\nFor two distinct pulsars separated by angle \\(\\zeta\\), one widely used normalization is\n\n\\[\\Gamma(\\zeta)= \\frac{1}{2} +\\frac{3}{2}x\\ln x-\\frac{1}{4}x, \\qquad x=\\frac{1-\\cos\\zeta}{2},\\]\n\nwith \\(x\\ln x\\) understood to approach zero as \\(x\\) approaches zero. The correlation tends to one-half for two different pulsars in the same direction, becomes negative over an intermediate range of angles, and rises to one-quarter at 180 degrees. Under the corresponding convention, an additional self-correlation contribution gives unit autocorrelation for a pulsar. These normalization details matter when comparing formulas and estimated amplitudes.\n\nThe curve is often called quadrupolar, although that description should not be interpreted as a pure elementary quadrupole function. It results from averaging the directional responses of pulsar pairs over gravitational-wave propagation directions and tensor polarizations. Its mathematical form is specific to the assumed background and theory. An anisotropic background, additional gravitational-wave polarizations, or correlations involving pulsar terms can require a more general description.\n\nCompeting common signals have different idealized angular signatures. A clock error introduces the same timing offset into every pulsar and therefore produces a monopolar pattern. An error \\(\\delta\\mathbf r\\) in the assumed position of the Solar System barycenter gives an approximate residual\n\n\\[\\delta t_a \\simeq \\frac{\\delta\\mathbf r\\cdot\\mathbf p_a}{c},\\]\n\nwhere \\(c\\) is the speed of light. Its directional dependence is dipolar. These distinctions allow the array to separate physical hypotheses, although uneven sky coverage and limited precision can cause leakage between their inferred patterns.\n\nAn array of \\(N\\) pulsars supplies \\(N(N-1)/2\\) distinct pairs: 50 pulsars yield 1,225 pairs. These are not equally precise or statistically independent measurements, because pairs share pulsars and have different observing spans and noise properties. Nevertheless, a larger and more widely distributed array samples the correlation curve more fully. Detection therefore depends on both timing quality and geometry, rather than on simply increasing the duration of observations of one exceptional pulsar."
    },
    {
      "heading": "Noise, propagation, and instrumental systematics",
      "text": "PTA noise is commonly divided into approximately white components and temporally correlated, or red, components. Radiometer noise comes from the finite signal-to-noise ratio of an observation. Pulse jitter arises from stochastic differences among individual pulses and can remain important even for a highly sensitive telescope. Because jitter can affect multiple simultaneous frequency channels together, it is not always well represented by independent errors assigned to each channel.\n\nIntrinsic rotational noise produces longer-timescale fluctuations in pulsar phase. It is often modeled with a power-law spectrum, but real behavior need not follow a single stationary process. Discrete changes, long-term profile evolution, or inadequately modeled binary effects can introduce additional structure. A red spectrum in one pulsar is not evidence for a gravitational-wave background; the required distinction is between individual red processes and a spatially correlated contribution across the array.\n\nRadio propagation through the ionized interstellar medium introduces a delay approximately proportional to the dispersion measure divided by the square of observing frequency. The dispersion measure is the integrated free-electron density along the line of sight,\n\n\\[\\mathrm{DM}=\\int n_e\\,dl.\\]\n\nAs the pulsar, Earth, and intervening gas move, this integral changes. Observing at several radio frequencies helps estimate and remove the associated delay. The solar wind adds another time-dependent electron column, especially for sightlines passing relatively close to the Sun.\n\nOther propagation effects are more complicated than a simple inverse-square law. Multipath scattering broadens the pulse and can shift its apparent arrival time. Interstellar scintillation changes the relative signal strength across frequency, which can bias a measurement if pulse shape also evolves across the band. Wideband timing methods address these effects by fitting frequency-dependent pulse profiles and propagation parameters together. The gravitational-wave timing perturbation is achromatic to leading order, making chromatic information a major tool for separating it from plasma effects.\n\nInstrumental consistency is equally important. Receiver replacements, changes in digital backends, polarization calibration errors, and differences between observatories can introduce offsets or evolving biases. Timing models often include fitted offsets between observing systems, but such parameters cannot automatically correct every systematic effect. Accurate calibration, overlapping observations during equipment changes, and detailed records of the signal-processing chain help preserve long-term continuity.\n\nThese issues create a fundamental trade-off in noise modeling. Models that are too restrictive can misidentify unmodeled disturbances as an astrophysical signal; models that are too flexible can absorb genuine signal power. Analyses therefore examine alternative noise assumptions, sensitivity to particular pulsars and instruments, and the stability of inferred correlations. The objective is not simply to obtain small residuals, but to characterize which structures remain and how reliably they can be attributed."
    },
    {
      "heading": "Statistical inference and spectral characterization",
      "text": "A stochastic background is usually sought through a model for the joint covariance of all pulsar timing data. Schematically, for pulsars \\(a\\) and \\(b\\),\n\n\\[C_{ab}=C^{\\rm individual}_{ab} +\\Gamma_{ab}C^{\\rm GW},\\]\n\nwhere the first term includes measurement noise and pulsar-specific processes, and the second combines a common temporal covariance with the angular correlation \\(\\Gamma_{ab}\\). Real analyses also incorporate timing-model parameters, instrumental terms, and sometimes additional common processes. Irregular sampling means that a simple Fourier transform of the residuals is generally not a sufficient analysis.\n\nA standard spectral quantity is the characteristic strain,\n\n\\[h_c(f)=A \\left(\\frac{f}{f_{\\rm yr}}\\right)^\\alpha, \\qquad f_{\\rm yr}=1\\ {\\rm yr}^{-1}.\\]\n\nHere \\(A\\) is the strain amplitude at the reference frequency and \\(\\alpha\\) is the strain spectral index. With a commonly used one-sided convention, the timing-residual power spectral density is\n\n\\[P_r(f)=\\frac{h_c^2(f)}{12\\pi^2 f^3}.\\]\n\nThus low frequencies receive extra weight in residual space. For an ideal population of circular binaries evolving through gravitational radiation, \\(\\alpha=-2/3\\), giving a residual spectrum proportional to \\(f^{-13/3}\\). Some publications describe this using a positive red-noise index \\(\\gamma=13/3\\).\n\nBayesian inference compares models by integrating their likelihoods over parameter distributions. Models may contain only independent pulsar noise, a common-spectrum process without specified cross-correlations, or a common process following the Hellings–Downs pattern. A Bayes factor expresses relative support for two specified models and depends on their assumptions and prior ranges. It is not automatically interchangeable with a frequentist false-alarm probability.\n\nFrequentist methods include cross-correlation estimators, often called optimal statistics, that weight pulsar pairs according to their expected sensitivity. Significance can be investigated with simulations and with procedures that disrupt the expected signal structure, such as phase shifts or alternative sky configurations. Each procedure has assumptions about the relevant null distribution. Agreement between independent analysis methods provides an important check, but does not eliminate the need to understand shared noise models and data.\n\nSeveral distinct claims must be separated. A common-spectrum process means that multiple pulsars favor red noise with similar spectral properties. Evidence for spatial correlations is stronger because it tests the relationship between different sightlines. Identifying a particular source population requires further information about spectrum, anisotropy, non-Gaussian behavior, or individual sources. Uncertainty in these steps includes not only measurement noise but also the fact that the observed universe contains a finite realization of an underlying source population."
    },
    {
      "heading": "Sources in the nanohertz band",
      "text": "The leading astrophysical explanation for the observed nanohertz background is a population of supermassive black hole binaries. When galaxies merge, their central black holes can become bound after losing orbital energy through interactions with surrounding matter. Systems with masses of roughly millions to billions of solar masses can emit in the PTA band at separations much larger than those probed by ground-based gravitational-wave detectors. For a nearly circular binary, the dominant gravitational-wave frequency is twice the orbital frequency.\n\nFor a slowly evolving circular binary, the strain scales approximately as\n\n\\[h\\propto \\frac{(G\\mathcal M_z)^{5/3}(\\pi f)^{2/3}} {c^4D_L},\\]\n\nwhere \\(G\\) is the gravitational constant, \\(D_L\\) is luminosity distance, and \\(\\mathcal M_z\\) is the redshifted chirp mass. The source-frame chirp mass is\n\n\\[\\mathcal M= \\frac{(M_1M_2)^{3/5}} {(M_1+M_2)^{1/5}}.\\]\n\nThe strong mass dependence makes massive binaries especially important. Their combined background is sensitive to galaxy merger histories, black hole masses, and the time binaries spend at each orbital frequency.\n\nThe familiar \\(f^{-2/3}\\) characteristic-strain spectrum assumes circular orbits and evolution dominated by gravitational radiation. At wider separations, interactions with stars or gas can change the rate of orbital shrinkage. Eccentricity distributes radiation among multiple harmonics instead of concentrating it near twice the orbital frequency. Both effects can reshape the low-frequency spectrum. Measurements can therefore constrain how black holes progress from galaxy-scale mergers to close binaries, but translating a spectral feature into one unique environmental mechanism is generally difficult.\n\nThe background need not be perfectly smooth or Gaussian. At some frequencies, especially toward the higher-frequency end of the accessible band, a small number of nearby or exceptionally massive binaries may dominate. A sufficiently strong system could appear as an individually resolvable continuous wave. Its phase evolution may be extremely slow during a human observing span, while the pulsar terms sample a much earlier stage of the same binary. Source localization and parameter estimation are complicated by uncertain pulsar distances and by the uneven sensitivity of the array.\n\nOther proposed signals include cosmic strings, primordial gravitational waves, and backgrounds generated during early-universe phase transitions. A rapid change in the gravitational field of a violent event can also produce gravitational-wave memory, which would appear as a persistent change in the timing trend after the wave reaches Earth. These possibilities motivate searches beyond a single binary-population model. However, observing a background with the expected tensor correlation pattern does not by itself establish an early-universe origin or distinguish every astrophysical explanation."
    },
    {
      "heading": "Development of observing programs and the 2023 results",
      "text": "The use of pulsar timing to search for gravitational waves developed after the discovery of pulsars in 1967. Mikhail Sazhin in 1978 and Steven Detweiler in 1979 described important aspects of their timing response to gravitational waves. In 1983, Ronald Hellings and George Downs derived the characteristic angular correlation for an isotropic tensor background and applied the idea to the available timing observations. The discovery of millisecond pulsars made much higher long-term timing precision possible, although building a useful array required decades of sustained observations.\n\nThree major regional programs became central to the field: the European Pulsar Timing Array (EPTA), the Parkes Pulsar Timing Array (PPTA), and the North American Nanohertz Observatory for Gravitational Waves (NANOGrav). EPTA combines observations from major European facilities, including Effelsberg, Lovell, Nançay, Sardinia, and Westerbork. PPTA developed around the Parkes 64-metre telescope, now also known as Murriyang. NANOGrav assembled long timing records using facilities including the Green Bank Telescope and, before its 2020 collapse, the Arecibo telescope.\n\nThese programs coordinate through the International Pulsar Timing Array (IPTA), which develops combined data sets and common analysis practices. Combining observations increases the number of pulsars, extends time baselines, and provides independent measurements of some sources. It also requires careful treatment of different receivers, backends, frequency bands, observing schedules, and calibration histories. Further regional efforts include the Indian Pulsar Timing Array and the Chinese Pulsar Timing Array.\n\nIn June 2023, several collaborations released results supporting a nanohertz gravitational-wave background. NANOGrav’s 15-year data set included 67 pulsars and showed evidence for Hellings–Downs spatial correlations. Under a power-law model with the strain spectral slope fixed to the circular, gravitational-radiation-driven binary expectation, its inferred characteristic amplitude at one inverse year was approximately \\(2.4\\times10^{-15}\\). The amplitude depends on the assumed spectrum and analysis model and should not be treated as a model-independent measurement at every frequency.\n\nEPTA, including a joint analysis with Indian PTA observations, and PPTA reported related evidence in their data. The Chinese Pulsar Timing Array reported evidence for the expected correlations using 57 pulsars observed over about 3.4 years with the Five-hundred-meter Aperture Spherical radio Telescope, or FAST. The different programs had different time spans, sensitivities, noise models, and statistical measures, so their quoted significances are not directly interchangeable. Some also observed overlapping pulsar sets. Their broad agreement nevertheless provided a major advance over earlier common-spectrum findings that lacked comparably strong spatial-correlation evidence."
    },
    {
      "heading": "Detector development and wider scientific uses",
      "text": "PTA sensitivity improves through longer time spans, better timing of existing sources, and the discovery and monitoring of additional pulsars. These improvements are complementary rather than equivalent. A longer baseline opens access to lower frequencies and allows more of a slowly varying signal to emerge from fitted trends. Greater collecting area reduces radiometer noise. Adding pulsars increases the number and angular diversity of correlations, improving discrimination among gravitational waves, clock errors, and ephemeris errors.\n\nMore sensitive radio facilities do not remove every limitation. FAST and MeerKAT provide high-quality timing observations, while the Square Kilometre Array is intended to expand pulsar discovery and precision timing substantially. For bright sources, however, pulse jitter and intrinsic or propagation-related noise may set limits that cannot be overcome simply by collecting more radio photons in the same observing interval. Broad frequency coverage and stable instrumentation remain essential. Discovering a moderately bright, exceptionally stable pulsar in a poorly sampled sky region can be particularly valuable.\n\nAs the signal becomes better characterized, analyses can move beyond a single isotropic power law. Targets include the angular distribution of background power, deviations from Gaussian statistics, changes in spectral slope, and individually resolvable binaries. Anisotropy could trace the nonuniform distribution of nearby galaxies or the dominance of a small number of sources. Continuous-wave searches can be combined with electromagnetic observations of candidate host galaxies, although periodic variability in a galaxy alone does not establish the presence of a gravitational-wave-emitting binary.\n\nPTAs also support tests of gravitational physics. Alternative theories may predict scalar or vector wave polarizations with angular correlations different from the Hellings–Downs pattern. Propagation modified by a nonzero graviton mass or other departures from general relativity can alter frequency and distance dependence. Such tests require consistent treatment of pulsar terms, source assumptions, and noise; an apparent mismatch with one simplified background model is not automatically evidence for modified gravity.\n\nThe same timing records can constrain non-wave phenomena. Ensembles of pulsars can be used to construct pulsar-based timescales and to test terrestrial time standards. Their timing response can help assess Solar System ephemerides and constrain planetary-system masses. Searches also consider gravitational disturbances or oscillating signals associated with particular dark matter models. These applications use different predicted temporal and spatial signatures from the tensor background search.\n\nPTAs occupy a frequency range complementary to terrestrial laser interferometers and planned space-based interferometers. Ground-based instruments observe much higher-frequency signals, often from stellar-mass compact binaries, while space missions are designed to access the millihertz region. Nanohertz timing observations probe a different stage and mass scale of gravitational dynamics. Their scientific reach depends on a distinctive form of experimental continuity: maintaining, calibrating, and reanalyzing records that can span several generations of radio instrumentation."
    }
  ],
  "links": [
    "pulsar",
    "gravitational-wave",
    "neutron-star",
    "milky-way",
    "earth",
    "black-hole",
    "moment-of-inertia",
    "angular-momentum",
    "measurement-uncertainty",
    "sun",
    "solar-system",
    "center-of-mass",
    "stellar-parallax",
    "general-relativity",
    "spacetime",
    "speed-of-light",
    "solar-wind",
    "scattering",
    "calibration",
    "bayesian-inference",
    "cosmic-string",
    "gravity",
    "dark-matter"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}