Nanohertz Gravitational Wave Observations

• Nanohertz gravitational waves are low-frequency ripples in space-time detected using pulsar timing arrays that reveal signals from supermassive black hole binaries and early Universe sources.
• Pulsar timing arrays precisely measure timing residuals across millisecond pulsars to extract a stochastic gravitational wave background with distinct spectral and spatial signatures.
• These observations test fundamental astrophysics and cosmology by constraining models of cosmic strings, primordial relics, and the dynamics of galaxy mergers.

Nanohertz gravitational wave (nHz GW) observations constitute a rapidly advancing frontier in astrophysics and cosmology. By exploiting the exceptional timing precision of millisecond pulsars distributed across the Galaxy, pulsar timing arrays (PTAs) operate as a galaxy-scale GW detector most sensitive to years-to-decades–period GWs ($f \sim 1$–$100$ nHz). These frequency bands are uniquely populated by a stochastic background generated primarily by cosmic populations of supermassive black hole binaries (SMBHBs), as well as by potential cosmological sources such as cosmic strings and relics of the early Universe. Recent PTA observations—including the NANOGrav 15-yr dataset—have found highly significant evidence for such a background, with properties that begin to resolve spectral shape, angular correlations, and constraints on nHz GW source populations.

1. Fundamentals of Nanohertz Gravitational Wave Detection

PTAs exploit the stability of millisecond pulsars as galactic clocks. A passing GW perturbs the space-time metric between a pulsar and Earth, thereby modulating pulse arrival times. The GW-induced timing residual for pulsar $a$ is

$$\Delta t_a(t) = \frac{1}{2}\frac{\hat{p}_a^i \hat{p}_a^j}{1 + \hat{\Omega} \cdot \hat{p}_a} \left[h_{ij}(t_p) - h_{ij}(t_e)\right]$$

where $\hat{p}_a$ is the direction to the pulsar, $\hat{\Omega}$ the GW propagation direction, $h_{ij}$ the metric perturbation, $t_e$ the time at Earth, and $t_p = t_e - L_a(1 + \hat{\Omega} \cdot \hat{p}_a)$ the time at the pulsar ("pulsar term"), with $L_a$ the distance. The observable is the timing residual $\delta t$ after subtraction of a deterministic timing model, which sensitively encodes any correlated, low-frequency signals across the PTA (Brazier et al., 2019, McLaughlin, 2013, Burke-Spolaor et al., 2018).

The array’s sensitivity band is dictated by the timespan and cadence: for a data span $T$, the frequency coverage is $f_\mathrm{min} \sim 1/T$ up to $f_\mathrm{max}$ set by typical inter-observation intervals of weeks to months. This accesses the $1$–$100$ nHz regime inaccessible to LIGO, Virgo, KAGRA, or LISA (Burke-Spolaor et al., 2018, Brazier et al., 2019).

2. Astrophysical and Cosmological Gravitational Wave Sources

The dominant expected nHz GW sources are:

• Supermassive Black Hole Binaries (SMBHBs): Galaxy mergers form SMBHBs, which, as they inspiral under GW emission, produce a stochastic background with a characteristic strain spectrum $h_c(f) = A (f/f_\mathrm{yr})^{\alpha}$ where $f_\mathrm{yr} = 1$ yr$^{-1}$, $\alpha=-2/3$ in the GW-driven regime, and $A \sim 10^{-15}$ (Burke-Spolaor et al., 2018, Sesana et al., 21 Dec 2025). The ensemble of unresolved SMBHBs yields a stochastic background, with a few especially nearby or massive systems potentially forming individually resolvable “continuous wave” sources (Arzoumanian et al., 2023, Brazier et al., 2019, Yang et al., 2024).
• Cosmic Strings: One-dimensional topological defects form loops which radiate GWs, generating a broadband stochastic background often parameterized by the string tension $G\mu$. PTA observations constrain such backgrounds at the level $G\mu < 10^{-10}$ (Demorest et al., 2012, Sesana et al., 21 Dec 2025).
• Early Universe relics: Models include inflationary gravitational waves (typically subdominant in the PTA band unless a blue spectrum is invoked), first-order phase transition bubbles, sound waves, and domain walls, all of which create peaked or broken-power-law spectra at nHz–$\mu$Hz frequencies (Sesana et al., 21 Dec 2025, Liu, 2023). Scalar-induced backgrounds from enhanced primordial curvature perturbations or condensate fragmentation are also viable (Liu, 2023).

Notably, cosmological sources that match PTA background amplitudes also predict large density perturbations on small scales. Such models can thus be tested (and in practice strongly constrained) by observing or failing to observe the expected dark-matter ultra-compact mini-halo (UCMH) populations (Liu, 2023).

3. PTA Data Analysis: Statistical Methods and Spectral Inference

PTAs model each pulsar’s timing residuals as

$$\delta t = M\epsilon + n_\text{white} + n_\text{red} + s_\text{gw}$$

where $M\epsilon$ are timing model corrections, $n_\text{white}$ is white measurement noise, $n_\text{red}$ is time-correlated pulsar noise, and $s_\text{gw}$ the GW-induced signal (Arzoumanian et al., 2023, Kimpson et al., 13 Jan 2025). GWs enter as a “common red process” across pulsars.

The stochastic background is statistically extracted using a joint likelihood: $$L(\theta) \propto \frac{1}{\sqrt{\det(2\pi C)}} \exp\left[ -\frac{1}{2} \delta t^\top C^{-1} \delta t \right]$$ with $C$ the full covariance (white+red+common red noise). Bayesian posterior sampling explores the space of common spectral amplitude $A$, spectral slope $\gamma$, and spatial correlation structure (Arzoumanian et al., 2023, Agazie et al., 2023).

The canonical expectation is a power-law spectrum: $$h_c(f) = A (f/f_\mathrm{yr})^{-2/3} \qquad P(f) = \frac{A^2}{12\pi^2} \left(\frac{f}{f_\mathrm{yr}}\right)^{-\gamma} \;\;\text{with}\;\; \gamma=13/3$$ (Sardesai et al., 2024). Spectral estimation can flexibly move beyond power laws through methods such as $t$-process modeling, allowing explicit damping, features, or excesses due to individual binaries (Sardesai et al., 2024).

4. Spatial Correlation: The Hellings–Downs Signature and Beyond

The unique signature of an isotropic, unpolarized stochastic GW background is the quadrupolar spatial correlation predicted by Hellings and Downs: $$\Gamma(\xi) = \frac{3}{2}\, x\ln x - \frac{x}{4} + \frac{1}{2} + \frac{1}{2} \delta_{ab}, \quad x = \frac{1-\cos\xi}{2}$$ for pulsar angular separation $\xi$ (Agazie et al., 2023, Burke-Spolaor et al., 2018). This curve—quadrupolar under general relativity—distinguishes GW-induced correlations from potential instrumental or terrestrial systematics (clock errors: monopolar; solar-system ephemeris: dipolar).

In the NANOGrav 15-yr analysis, Bayesian model comparison yields overwhelming evidence for a common-spectrum red process (Bayes factor $>$ $10^{14}$), with HD-correlated models strongly favored (Bayes factors $200$–$1000$) over uncorrelated ones (Agazie et al., 2023). Harmonic decomposition in Legendre multipoles reveals significant power only in the quadrupole ($\ell=2$), with $c_2/c_2^\mathrm{HD}=1.088^{+0.32}_{-0.45}$, in quantitative agreement with GR expectations (Agazie et al., 2024). There is an unexplained monopolar excess at $\sim$4 nHz, meriting continued investigation (Agazie et al., 2024).

5. Upper Limits and Astrophysical Interpretation

Continuous searches for individual SMBHBs yield stringent upper limits. For the NANOGrav 12.5-yr data:

• Sky-averaged limit: $h_0 < (6.82 \pm 0.35) \times 10^{-15}$ at $f=7.65$ nHz,
• Most sensitive sky location: $h_0 < (2.66 \pm 0.15) \times 10^{-15}$,
• For the SMBHB candidate 3C 66B, $\mathcal{M} < (1.41 \pm 0.02) \times 10^9 M_\odot$ at $f_\text{GW}\approx6.04\times10^{-8}$ Hz.

These null results are consistent with the expected number density of individually resolvable binaries (Arzoumanian et al., 2023). Stochastic background amplitude posteriors from NANOGrav (e.g. $A = 2.4^{+0.7}_{-0.6} \times 10^{-15}$ at $f=1$ yr$^{-1}$ from the 15-yr set) suggest the observed signal matches the optimistic tail of astrophysical SMBHB predictions (Agazie et al., 2023, Sesana et al., 21 Dec 2025).

Large-scale anisotropy in the GW background, characterized via simulations, is $C_1/C_0\sim$ few $\times 10^{-3}$ at $f=1$ yr$^{-1}$—potentially within PTA reach as arrays grow (Yang et al., 2024).

6. Systematics, Multiwavelength Approaches, and Future Prospects

Accurate modeling of per-pulsar noise, including chromatic interstellar medium (ISM) effects and clock or ephemeris errors, is critical to avoid spurious detection claims (Hazboun et al., 2019). Bayesian frameworks and the use of flexible spectral models, such as $t$-processes or free spectra, mitigate false positives. Frequentist and Bayesian analyses are regularly cross-validated.

Gamma-ray PTAs, using Fermi-LAT, provide independent, complementary upper limits on the GWB (e.g., $A_\mathrm{gwb} < 1.05 \times 10^{-14}$) that will approach radio sensitivity over time owing to weaker propagation effects (Ajello et al., 2022). Joint radio and gamma-ray PTA analyses help deconvolve common vs per-pulsar noise and extend frequency coverage.

Looking forward, longer data spans and more precise timing will extend sensitivity to lower frequencies, improve angular resolution of the GW background, refine spectral inferences, and likely enable resolution of individual SMBHB "continuous" sources. SKA-era PTAs (with 100–1000 MSPs at $\lesssim$50 ns precision) are projected to detect anisotropy, non-Gaussianity, and conduct direct multi-messenger identification of GW sources (Sesana et al., 21 Dec 2025).

Bayesian parameter estimation using flexible cosmological templates already constrains string cosmology GW backgrounds (e.g., $f_s = 1.2^{+0.6}_{-0.6}\times10^{-8} \mathrm{Hz}$ and $$\Omega_\mathrm{gw}^{s} = 2.9^{+5.4}_{-2.3}\times 10^{-8}$$; $H_r$ yet unconstrained) (Tan et al., 2024).

7. Implications for Fundamental Physics and Cosmology

nHz GW observations are opening new tests of supermassive black hole binary astrophysics, hierarchical galaxy evolution, and a broad class of early-universe models including cosmic strings, phase transitions, and axion-like dark matter. The measured amplitude and spectrum of the stochastic background favor the SMBHB origin, yet blue-tilted, broken power-law, or peaked cosmological models remain under investigation, with discrimination requiring more precise spectral, temporal, or angular information (Sesana et al., 21 Dec 2025, Liu, 2023). Forthcoming constraints on cosmic string tension, reheating equation of state, primordial black hole mass functions, and dark matter microstructure are anticipated as PTA data and analysis frameworks continue to evolve.

---

References:

(Arzoumanian et al., 2023, Agazie et al., 2023, Agazie et al., 2024, Burke-Spolaor et al., 2018, Arzoumanian et al., 2020, Brazier et al., 2019, Ajello et al., 2022, Sardesai et al., 2024, Sesana et al., 21 Dec 2025, Arzoumanian et al., 2015, McLaughlin, 2013, Tan et al., 2024, Yang et al., 2024, Demorest et al., 2012, Hazboun et al., 2019, Liu, 2023, Xiao et al., 11 Dec 2025, Kimpson et al., 13 Jan 2025)