Multiband Gravitational-Wave Sources

A multiband gravitational-wave source is one physical system measured in separated frequency bands, most notably a stellar-mass compact binary first followed by a space interferometer and later by a ground network. The scientific gain comes from a joint probability model: the early data predict when and where the later signal may arrive, while the long frequency lever arm tests waveform evolution and sharpens correlated parameters. A source detectable in each band separately is not automatically a confidently associated multiband event, and no event yield is independent of mission, population, or selection assumptions.

Helpful background. Probability and Statistical Inference supplies conditional independence, priors, and selection. The gravitational-wave likelihood supplies coherent network inference and evidence. Gravitational Radiation and Waveforms supplies compact-binary evolution, while Black-Hole Astrophysics supplies stellar-mass populations and environmental channels.

How a binary crosses frequency bands

For a quasicircular binary with detector-frame chirp mass $\mathcal M^{\mathrm{det}}$, the leading evolution of the observed gravitational-wave frequency is

$$\dot f =\frac{96}{5}\pi^{8/3} (G\mathcal M^{\mathrm{det}})^{5/3} f^{11/3}.$$

Integrating from $f$ to the formal endpoint gives

$$t_c-t =\frac{5}{256} (G\mathcal M^{\mathrm{det}})^{-5/3} (\pi f)^{-8/3}.$$

This steep dependence is the basis of advance prediction: a small frequency change in the millihertz band corresponds to a long interval, whereas the same source sweeps rapidly through the ground band. The formula uses detector-frame mass,

$$\mathcal M^{\mathrm{det}} =(1+z)\mathcal M^{\mathrm{src}},$$

and detector time. Mixing source-frame mass with observed frequency introduces an erroneous redshift factor.

For propagation over years, the leading expression is a diagnostic, not the prediction engine. Spin effects, eccentricity, higher post-Newtonian terms, external acceleration, environmental torques, and waveform calibration can accumulate phase or timing bias. A valid forecast evolves posterior samples with the same declared time coordinate and reports a distribution for $t_c$, sky position, and phase rather than a point estimate.

For scale, a detector-frame chirp mass of $26M_\odot$ at $10\,\mathrm{mHz}$ has a leading remaining time of about 19.4 years. At $20\,\mathrm{mHz}$ it has about 3.05 years. Detectability in a space band therefore does not guarantee a merger within the same mission or ground observing run; selection must include the crossing time.

One source, two likelihoods

Let $d_s$ denote space data and $d_g$ ground data. If their instrumental noises are conditionally independent given shared source parameters $\vartheta$ and detector-specific nuisance parameters $\lambda_s,\lambda_g$, then

$$p(\vartheta,\lambda_s,\lambda_g|d_s,d_g) \propto \mathcal L_s(d_s|\vartheta,\lambda_s) \mathcal L_g(d_g|\vartheta,\lambda_g) \pi(\vartheta,\lambda_s,\lambda_g).$$

The shared parameterization must be explicit. Masses and spins need consistent frames and reference frequencies; eccentricity needs a reference frequency and definition; sky position, polarization, and epoch need the same coordinates; and propagation or beyond-GR parameters need one sign convention.

Using a space posterior as a ground-analysis prior requires retaining its sampling prior. If space analysis produced

$$p_s(\vartheta|d_s) \propto \mathcal L_s(d_s|\vartheta)\pi_s(\vartheta),$$

then a joint target with prior $\pi_J$ is

$$p_J(\vartheta|d_s,d_g) \propto p_s(\vartheta|d_s) \mathcal L_g(d_g|\vartheta) \frac{\pi_J(\vartheta)}{\pi_s(\vartheta)}.$$

Multiplying two posterior densities without removing one prior double counts prior information. Sequential importance sampling can implement the equation, but a small effective sample size signals that the space posterior did not cover the support selected by the ground likelihood; then a fresh joint run is required.

This update also assumes that detector-specific nuisance parameters have been consistently marginalized and that their prior factorizes appropriately conditional on $\vartheta$. If a nuisance quantity is shared, retain it in the joint parameter vector. For a simple check, two independent unit-variance measurements $y_s,y_g$ with prior $\vartheta\sim N(0,1)$ give posterior precision $1+1+1=3$ and mean $(y_s+y_g)/3$. Multiplying the two individual posteriors instead gives precision four because it uses the prior twice, making the result spuriously narrow.

Association is a model-selection problem. The common-source hypothesis $H_C$ must be compared with an unrelated-source hypothesis $H_U$, including the prior rate for chance coincidences:

$$\mathcal O_{C/U} =\frac{p(d_s,d_g|H_C)}{p(d_s,d_g|H_U)} \frac{p(H_C)}{p(H_U)}.$$

Agreement of median masses and a time window is useful for candidate generation, but it is not this odds calculation.

What the long baseline can add

The two bands emphasize different aspects of the same binary:

• Prediction and scheduling. The space posterior can forecast coalescence time and a sky distribution for ground-network preparation or electromagnetic follow-up. The useful quantity is probability inside a time–sky window after accounting for duty, latency, and ephemeris—not merely an expected lead time.
• Intrinsic parameters. Long inspiral phase constrains chirp mass and slowly varying spin combinations, while merger and higher frequencies constrain mass ratio, precession, and remnant properties. Joint inference can break degeneracies that neither band removes alone.
• Eccentricity and environment. Space-band harmonics can reveal eccentricity that radiation reaction later reduces. A phase anomaly across bands could also reflect acceleration or environmental coupling, but only after waveform and clock alternatives are tested.
• Strong-field and propagation tests. Comparing a low-frequency inspiral prediction with the later merger probes phase evolution over a large frequency interval. The inference remains conditional on the parameterized modification, waveform accuracy, and treatment of selection.

Vitale demonstrated these parameter-estimation gains in simulated multiband analyses Vitale 2016, pp. 2–4. The calculation establishes the mechanism for the assumed detectors and injections; it does not establish an event count or guarantee an electromagnetic counterpart.

Selection couples the two bands

For population hyperparameters $\Lambda$, the multiband selection factor is

$$\xi_{\mathrm{mb}}(\Lambda) =\int p_{\mathrm{sel}}^{\mathrm{mb}}(\vartheta) p(\vartheta|\Lambda)\,d\vartheta .$$

The probability $p_{\mathrm{sel}}^{\mathrm{mb}}$ must encode the actual common-source search, space and ground duty, sky-dependent antenna patterns, forecast window, thresholds, data quality, and association rule. It generally is not safe to replace it with a product of two averaged detection probabilities: sky, inclination, masses, coalescence time, and operations correlate the selections. The reliable estimator is a shared injection campaign replayed through both pipelines.

Early forecasts established that binaries like those found by ground detectors can traverse the space band before merger Sesana 2016, pp. 2–4. Newer population-informed work re-evaluates complementarity with updated populations and detector assumptions Buscicchio et al. 2024, §§2–5. Differences in predicted yield should therefore be traced to source rates, mass and redshift distributions, mission duration, sensitivity, thresholds, and association efficiency—not averaged into a supposedly universal number.

Results and interpretations at the stated cutoff

All space-mission entries below are forecasts or simulations as of 2026-08-18. No LISA flight observation or confirmed LISA–ground multiband event is being summarized.

Question	Strongest supported statement	Conservative interpretation and countercheck
Band crossing	Radiation reaction maps a compact binary from lower to higher frequency with a calculable leading time-to-coalescence scaling	Propagate full waveform posterior samples; vary spin, eccentricity, acceleration, and waveform order
Advance prediction	Space data can constrain a future time–sky probability distribution in controlled injections	Report coverage under gaps, nonstationary noise, foreground, response, and latency; do not equate detectability with alert readiness
Joint inference	Conditionally independent datasets multiply at the likelihood level and can break parameter degeneracies	Preserve interim priors and shared conventions; compare sequential reweighting with a fresh joint run
Common-source association	Coalescence time, sky, masses, and phase evolution can support a common-source odds ratio	Include the unrelated-source rate and search trials; perform time-slide or scrambled-catalog controls
Event yield	Declared populations and detector configurations produce simulated yield distributions	Rate, mission, waveform, and selection uncertainty dominate any single count; publish joint injection efficiency
New physics	A long frequency baseline can constrain parameterized phase or propagation deviations	Repeat with alternative waveform systematics, calibration, clocks, environments, and population priors

The ESA 2024 definition study supplies the reference space-mission configuration ESA 2024, §§2–4. A forecast that uses a different arm length, observing span, duty, or sensitivity must label that configuration rather than inherit the ESA name.

A paired-injection reproduction

A transparent end-to-end benchmark can start from the LISA Data Challenge 2b “Spritz” simulated second-generation $X,Y,Z$ fractional-frequency TDI streams at 5 s cadence LISA Data Challenge 2b, dataset description. Use a declared Spritz interval or its noise realization, inject a catalog of binaries whose coalescences enter the chosen ground band, and generate matching ground-detector strain from the same source records. This creates a paired simulation; it does not turn Spritz or the ground injection into an observation.

The reproduction record should include:

• checksums for all challenge files and injection tables; the ESA configuration, orbit, link timestamps, ordered TDI delays, cadence, band, observing interval, gaps, and space response;
• ground sites, geographic geometry, duty and data-quality mask, sample rate, calibration perturbations, and named power-spectral-density files;
• the exact LALSuite release or commit, waveform approximant, reference frequency, redshift and cosmology convention, and any eccentric or beyond-GR implementation;
• population drawing measure, common-source labels hidden from the search, instrumental and foreground realizations, priors, search statistics, trials, thresholds, association odds, and random seeds; and
• space-only, ground-only, and joint posterior coverage; timing and sky-window coverage; effective sample size for sequential updates; false associations from time slides; and $p_{\mathrm{sel}}^{\mathrm{mb}}$ over the injection population.

Run the benchmark in stages: zero-noise likelihood agreement, independent Gaussian noises, moving space response, foreground and gaps, ground-network duty, and unrelated-source contamination. Refine cadence and waveform tolerances; verify that posterior shifts are smaller than declared numerical error. Refresh conclusions after a new mission configuration, public data challenge, ground-detector sensitivity or observing scenario, population-rate update, or validated waveform change that alters timing or joint selection.

Check your understanding

Linearize the leading time-to-coalescence formula. If independent uncertainties satisfy

$$\frac{\sigma_{\mathcal M}}{\mathcal M}=10^{-4}, \qquad \frac{\sigma_f}{f}=2\times10^{-5},$$

estimate $\sigma_t/t$ and the corresponding uncertainty for a five-year prediction. Ignore every other source of error.

Solution

Because

$$t\propto\mathcal M^{-5/3}f^{-8/3},$$

the linear perturbation is

$$\frac{\delta t}{t} =-\frac53\frac{\delta\mathcal M}{\mathcal M} -\frac83\frac{\delta f}{f}.$$

For independent errors,

$$\frac{\sigma_t}{t} =\left[ \left(\frac53\,10^{-4}\right)^2 +\left(\frac83\,2\times10^{-5}\right)^2 \right]^{1/2} \simeq1.75\times10^{-4}.$$

Five Julian years contain $43{,}830$ hours, so

$$\sigma_t \simeq (43{,}830\ \mathrm{h})(1.75\times10^{-4}) \simeq7.7\ \mathrm{h}.$$

This is only an algebraic uncertainty propagation. Correlations, higher-order phasing, detector response, waveform bias, acceleration, and operations can dominate a real forecast.

References

• Buscicchio, Riccardo, et al. “Stellar-Mass Black-Hole Binaries in LISA: Characteristics and Complementarity with Current-Generation Interferometers.” arXiv:2410.18171 (2024). Open PDF.
• European Space Agency. LISA Definition Study Report. ESA-SCI-DIR-RP-002 (2024). Official PDF.
• LISA Data Challenge. “Challenge 2b: Spritz.” Public simulated dataset and documentation. Dataset.
• Sesana, Alberto. “Prospects for Multiband Gravitational-Wave Astronomy after GW150914.” Physical Review Letters 116 (2016): 231102. DOI. Open PDF.
• Vitale, Salvatore. “Multiband Gravitational-Wave Astronomy: Parameter Estimation and Tests of General Relativity with Space- and Ground-Based Detectors.” Physical Review Letters 117 (2016): 051102. DOI. Open PDF.