Shared solar wind noise could skew space-based gravitational-wave searches

A future network linking the Laser Interferometer Space Antenna, or LISA, with Taiji could search for a persistent gravitational-wave background by comparing their measurements. A study by Mengfei Sun and colleagues finds that solar wind fluctuations could bias those searches if sufficiently large plasma structures affect both observatories.

The result is conditional, not a detection of contamination. The study combines measured electron-density variations with assumed spatial correlations and detector models. Its largest reported shifts remain below the corresponding forecast statistical uncertainties.

The work was first submitted to arXiv on July 6, 2026, at 17:46:03 UTC, and the version examined here was submitted on November 6. Although available as an arXiv manuscript, the source record identifies this revision as the accepted version and lists publication in Physical Review D, volume 114, article 083507.

Why shared noise matters

A stochastic gravitational-wave background is a combined signal from many unresolved sources or processes in the early universe. LISA and Taiji would probe the millihertz range, with one millihertz corresponding to one oscillation every thousand seconds.

Each observatory would use laser links between three spacecraft to measure tiny changes in distance. In the configurations studied, LISA’s arms span about 2.5 million kilometres and Taiji’s about 3 million kilometres. Their centres are separated by roughly 0.684 astronomical unit, with one astronomical unit being approximately the Earth–Sun distance.

Researchers can look for a background using cross-correlation: comparing two detectors for a shared pattern. Independent instrumental disturbances tend not to build up a consistent correlation over long observations, whereas a common gravitational-wave signal can.

Shared environmental noise can also survive that comparison. The solar wind is a stream of plasma containing free electrons. Changes in electron density slightly alter the refractive index along a laser beam’s path, producing apparent optical-path changes. If related density structures influence both observatories, their disturbances could partly resemble the common signal being sought.

From electron measurements to detector forecasts

The authors used 38,198,016 valid electron-density measurements from the Wind spacecraft’s Solar Wind Experiment, covering August 16, 2002, to February 26, 2026. The retained data cover about 63% of that interval, with a median sampling interval of approximately 12.3 seconds.

Because observations are unevenly spaced and contain gaps, the team used Lomb–Scargle spectral estimation. This method fits sinusoidal variations directly to irregularly timed measurements rather than first filling the gaps. They fitted the resulting fluctuation spectrum with separate power laws in two frequency bands.

They then calculated how density fluctuations accumulate along the laser links. Crucially, different parts of a long arm need not experience a fluctuation in phase, so contributions can partly cancel. Their model also uses Taylor’s frozen-flow approximation: plasma structures are treated as being carried past the measurement paths by the solar wind without changing internally during that passage.

The link disturbances were propagated through first-generation time-delay interferometry, which combines delayed laser measurements to suppress laser-frequency noise. The analysis focused on two combinations called the A and E channels and included how network geometry changes the response to gravitational waves.

Two alternative Taiji orientations were considered for the plasma cross-correlation calculations. These are separate possible network configurations, not three observatories operating together.

Large structures make the difference

The central uncertainty is how far solar wind density fluctuations remain correlated. Wind’s measurements at one location cannot establish that across the proposed network.

The authors therefore compared two assumed regimes. In the smaller-coverage case, disturbances can affect one constellation but correlate extremely weakly across the network. The resulting parameter biases were negligible.

In the larger-coverage case, structures can span both observatories. Plasma noise still remained below the reference instrumental noise in the individual-detector comparisons, but its shared component produced larger parameter shifts. Earlier observations by multiple spacecraft motivate considering extended solar wind structures, but they do not establish the particular correlation lengths assumed in this calculation.

For a power-law gravitational-wave background, whose strength varies with frequency according to a slope, the largest reported slope bias was 12.73% of its forecast statistical uncertainty. That result used the LISA–Taijim orientation, ten years of observation and frequencies from 0.0001 to 0.01 hertz, with the background’s amplitude and slope estimated jointly.

The team also tested two cosmic-string background models, M2 and M3. Cosmic strings are hypothetical structures that could generate gravitational waves. The largest reported bias in the natural logarithm of their dimensionless tension parameter was 19.26% of its forecast uncertainty. This occurred for the M3 model in the LISA–Taijim configuration, with a dimensionless tension of 10⁻¹⁸, ten years of observation and frequencies from 0.0001 to 0.05 hertz.

These percentages are not errors of 12.73% or 19.26% in the physical parameters themselves. They represent about 0.13 and 0.19 times the corresponding statistical error estimates.

A conditional warning

Those uncertainties come from a Fisher-matrix approximation, a local forecast rather than a full analysis of future observations. Some cosmic-string cases have very broad uncertainties, limiting how literally the local estimates should be interpreted. The calculations also do not jointly fit astrophysical foregrounds and environmental parameters or incorporate full time-dependent mission orbits.

The assumed large correlation lengths are especially important: they were not measured from the Wind data. Multipoint solar wind observations are needed to constrain them, including how they vary with frequency, solar activity and distance from the Sun.

Within the model, longer observations reduce statistical uncertainty while leaving the absolute systematic shift approximately unchanged. The practical lesson is therefore specific: low plasma noise in each detector does not, by itself, establish that their shared noise is harmless.