Science · 6 min read
What Gravitational-Wave Detectors Measure and How Astronomers Infer the Source
From differential arm strain to probabilistic claims about black-hole masses, distance, and source populations.

The observable is strain, not a picture of a collision
When a gravitational-wave observatory reports two black holes merging, the detector has not photographed an event. Its immediate measurement is a time series related to strain: a fractional difference in length produced as a gravitational wave passes. The astrophysical account—masses, spins, distance, and source type—is inferred by comparing that measured pattern with physical models while accounting for instrumental noise.
This distinction is central to what the evidence supports. General relativity predicts that accelerating, asymmetric distributions of mass can radiate changes in spacetime geometry. Far from the source, those changes propagate as waves. A passing wave alternately stretches one direction and compresses the perpendicular direction. The effect at Earth is extraordinarily small, but a large laser interferometer can compare perpendicular paths precisely enough to detect some sources.
On 14 September 2015, the two Advanced LIGO instruments recorded GW150914, the first direct gravitational-wave detection. The discovery paper, published in Physical Review Letters, reported a transient signal consistent with the inspiral and merger of two black holes. That conclusion rested on measurements in two detectors, tests of local disturbances, statistical analysis of background noise, and agreement with relativistic waveforms.
Interference converts length difference into readable light
Each LIGO detector has two perpendicular, four-kilometre arms. A laser beam is split, sent along both arms, reflected by suspended mirrors, and recombined. The instrument is arranged so that changes in the relative optical path alter the interference pattern at a photodetector. Additional optical cavities make the light interact with the arms repeatedly, increasing sensitivity to differential motion.
Strain is dimensionless: a length change divided by the reference length. The LIGO Scientific Collaboration's detector account of GW150914 explains that the event produced an arm-length difference on the order of 4 × 10^-18 metres. Stating the displacement alone can mislead because the inference depends on a calibrated instrument response across frequency, not a ruler reading one mirror position.
The observatory must isolate mirrors from ground motion, keep large vacuum systems operating, control laser frequency and power, and measure how thousands of instrument and environmental channels behave. Noise changes with frequency. Seismic motion dominates the lowest accessible frequencies; thermal motion in mirror coatings and suspensions matters in intermediate bands; photon counting statistics contribute at higher frequencies. The sensitivity curve is therefore not one number. It describes the smallest strain fluctuations the detector can distinguish across a band.
The detector output also contains brief non-astrophysical transients called glitches. Lightning, ground vibration, electronics, scattered light, and control-system behavior can produce features that resemble part of a signal. Detection requires more than noticing a curve with the expected shape.
A network separates shared sky signals from local disturbances
GW150914 appeared in the LIGO Livingston detector and, about seven milliseconds later, in LIGO Hanford. The sites are roughly 3,000 kilometres apart. A gravitational wave can produce a compatible pattern in both with a delay no larger than the light-travel time between them; most local disturbances cannot. Two sites therefore provided a powerful consistency check.
Coincidence is not by itself sufficient. Analysts searched auxiliary sensors for environmental or instrumental causes. They estimated how often noise alone could produce events as strong and mutually consistent by comparing data at artificial time offsets much longer than the physical travel time. This generated a background distribution without assuming the noise was perfectly Gaussian. The discovery analysis reported a false-alarm probability corresponding to significance greater than 5.1 standard deviations for its modeled search.
That statistic is conditional on the search procedure and available data. It does not mean there was a 5.1-sigma visual match by human judgment, nor that any bump appearing in two instruments is a discovery. Later observing networks added Virgo, KAGRA, and improved LIGO sensitivity, enabling stronger localization and polarization constraints when several detectors are operating. Network geometry matters: different antenna orientations and arrival times help triangulate a source, while downtime or weak sensitivity in one instrument broadens the possible sky area.
Source properties come from model-based inference
A compact binary emits a characteristic “chirp” as its orbit shrinks: amplitude and frequency rise, followed by merger and a settling phase called ringdown. The rate of frequency change constrains a combination of the component masses called the chirp mass. Other features carry information about mass ratio and spins. Overall amplitude, together with orientation and the waveform model, informs luminosity distance.
For GW150914, the collaboration inferred source-frame black-hole masses of roughly 36 and 29 solar masses in its initial analysis, forming a remnant of about 62 solar masses. The difference was radiated largely as gravitational-wave energy, subject to the reported uncertainties. Subsequent waveform improvements and revised calibration can update parameter estimates without reversing the detection.
Inference is probabilistic because parameters trade off. A binary seen face-on can resemble a more distant, differently oriented system in amplitude. Spin can partially mimic mass-ratio effects. Weak signals yield broad posterior distributions. Researchers run waveform families derived from general relativity and numerical simulations against calibrated data, then report credible intervals that also depend on prior assumptions.
Tests of general relativity ask whether portions of the signal disagree with the theory or whether inferred properties from inspiral and post-inspiral stages conflict. Agreement constrains deviations within the sensitivity and frequency range of the observations. It does not verify general relativity in unmeasured regimes or settle every alternative theory.
What the catalog sees is a selected population
Detector catalogs contain the loudest members of an underlying population. More massive binaries radiate strongly but merge at lower frequencies; very massive systems may spend little time in the sensitive band. Orientation affects amplitude. Distance, detector duty cycle, and the search pipeline affect recovery. Observed mass and spin distributions cannot be read as an unbiased census without modeling these selection effects.
The LIGO–Virgo–KAGRA collaboration's peer-reviewed GWTC-3 catalog documents how candidates from a completed observing period were ranked and how source parameters were estimated. Catalog methods evolve as detector sensitivity, waveform models, and noise treatment improve, so counts from different releases require their stated thresholds and observation windows.
There are also signals that current detectors do not resolve individually. Many weak compact-binary events may combine into a stochastic background. Other possible backgrounds could originate in the early universe, but distinguishing components requires spectral information, network correlations, and much more data. Pulsar-timing arrays probe nanohertz frequencies through correlated timing changes, while proposed space interferometers would probe lower frequencies than ground-based instruments. These are complementary bands, not simply more sensitive versions of one experiment.
The public alert attached to a candidate is provisional. Calibration can change, data-quality problems can emerge, and estimates narrow after offline analysis. Catalog papers separate confident events from marginal candidates using stated thresholds. That procedure acknowledges uncertainty rather than treating every automated alert as equivalent evidence.
The measurement-inference boundary remains visible
Gravitational-wave astronomy has established that kilometer-scale interferometers can measure strain patterns from compact-object mergers and that those patterns agree closely with relativistic predictions. Multiple detections have revealed black-hole and neutron-star systems inaccessible through light alone. Joint observation of GW170817 with electromagnetic telescopes showed how one event can connect gravitational and conventional astronomy.
Yet a detector records calibrated motion amid noise, not masses or images directly. Source properties depend on waveform adequacy, instrument calibration, population assumptions, and network coverage. Some parameters remain weakly constrained, and many source classes have not been observed. The strength of the field lies partly in preserving this chain: from light at a photodetector, through a strain estimate, through statistical detection, to a bounded astrophysical inference. Each link answers a different question, and none licenses claims beyond the frequencies, sources, and uncertainties the instruments can test.
