Both sites, or it didn't happen
A length change smaller than a proton
The mirrors hang in silence, the vacuum holds, and the number that comes out is 10⁻¹⁸ metres. Every decision in the instrument is about earning the right to trust that number.

The claim is an engineering claim first
When LIGO announces a gravitational-wave detection, the headline is always about astrophysics — a binary merger, a distance in megaparsecs, a chirp reconstructed from warped spacetime. What the headline does not say is that the measurement itself, before any physics interpretation begins, is a statement about length: that the difference in the lengths of two perpendicular arms, each four kilometres long, changed by roughly one-thousandth the diameter of a proton. That number is not a model output. It is a reading from an interferometer — an instrument that turns a path-length difference into a fringe shift in laser light — and every subsystem in the building exists to make that reading believable.
The measurement claim is an engineering one before it is a physics one — isolation, optics and vacuum at a scale nothing else needs.
The arms at Livingston, Louisiana and Hanford, Washington are L-shaped beam tubes, each arm four kilometres long, kept at a pressure of around one-trillionth of an atmosphere. The vacuum is not incidental. A photon scattering off a residual gas molecule produces a phase noise that would swamp the signal, so the vacuum vessels at each site collectively hold one of the largest ultra-high-vacuum systems on Earth, with an interior volume of roughly 10,000 cubic metres. Achieving that pressure — and holding it continuously — required years of baking the steel tube walls at elevated temperatures to drive out absorbed gases before the system was ever closed.

Inside that vacuum, the mirrors themselves are among the most precisely figured optical surfaces ever made. Each test mass is a fused-silica cylinder, about 34 centimetres across and 20 centimetres thick, polished to surface irregularities of a fraction of a nanometre. The reflectivity of the mirror coatings must be high enough that photons bounce back and forth inside the Fabry-Pérot cavities formed by the arm mirrors hundreds of times before leaking out — effectively multiplying the four-kilometre arm length by a factor of several hundred when it comes to accumulated phase sensitivity. The coatings are alternating layers of silica and titania-doped tantala, and their internal mechanical loss contributes directly to thermal noise in the measurement band.
Key numbers
The noise floor is a catalogue of physics to be defeated
The sensitivity curve that LIGO publishes — the strain noise as a function of frequency, in units of strain per square root of hertz — is a map of every noise source the collaboration has characterised and partially suppressed. At high frequencies, above roughly a few hundred hertz, shot noise dominates: the quantum randomness in photon arrival times sets a floor that improves with more laser power. At low frequencies, below about 30 hertz, the dominant limit is seismic noise. Between those two regimes, in the band from roughly 20 to 200 hertz where LIGO is most sensitive, thermal noise in the mirror suspensions and coatings is the primary adversary.
The suspension system is the instrument's answer to seismic noise. Each 40-kilogram mirror hangs from a four-stage pendulum — a chain of masses and fibres culminating in fused-silica fibres thinner than a human hair that attach to the mirror itself. A pendulum is a mechanical filter: vibrations above its resonant frequency are attenuated at 40 decibels per decade of frequency per stage, so four stages produce attenuation steep enough that ground motion at a hertz translates to displacement at the mirror far below the measurement floor. The entire suspension assembly sits on an active seismic isolation platform — a table of sensors and actuators that measures the building's motion and cancels it before it reaches the pendulum. The platform, in turn, sits on a passive isolation stack. The layers are deliberate: each removes a different part of the noise spectrum, and none is sufficient alone.

What makes the system particularly demanding is the absence of a single-point failure tolerance at the level of noise coupling. A grounding fault in a single cable, a microseismic storm in the Gulf of Mexico transferring energy to the Louisiana ground, a passing freight train — all of these have, at various times, been traced as contributors to elevated noise floors or temporary losses of lock. "Lock" is the state in which the laser light resonates stably inside all the cavities simultaneously; achieving it requires controlling the mirror positions to picometres while the ground moves by microns, using feedback loops fast enough to hold the system stable. Losing lock means losing data, and the coincidence requirement between the Livingston and Hanford detectors means that simultaneous lock at both sites is necessary for a confident detection.
Engineering layers (what each subsystem defeats)
Calibration is itself a measurement
Every number in a published LIGO result carries a calibration uncertainty, and the calibration is not a procedure done once at commissioning. It is a continuous process. Photon calibrators — auxiliary lasers whose radiation pressure applies a known, precisely computed force to the mirror — provide an independent length reference that can be checked during observing runs. The force from a photon calibrator is calculated from measured laser power and the speed of light; the uncertainty on the absolute strain calibration in the third observing run was a few percent, small enough to not limit most astrophysical inferences but large enough to matter for source parameter precision.

Quantum noise, which sets both the shot-noise ceiling at high frequencies and, through radiation-pressure back-action, a noise contribution at low frequencies, is now addressed through squeezed light injection. A quantum optics technique that redistributes the vacuum fluctuations entering the interferometer — reducing phase uncertainty at the cost of increased amplitude uncertainty — squeezed light has been in routine operation at LIGO since the third observing run and contributes measurable improvement to the sensitivity curve. It is a reminder that the instrument is not finished: each observing run follows a commissioning period that adjusts mirror alignments, suspension tunings, and optical configurations, and the published sensitivity curves from each distinct run reflect those changes.
Chronology of capability
- Initial LIGO (iLIGO)operated early 2000s, no detections
- Advanced LIGO (aLIGO) installation2010–2015, order-of-magnitude sensitivity improvement
- O1 (2015–2016)first gravitational-wave detection, GW150914
- O2 (2016–2017)first binary neutron star detection, GW170817
- O3 (2019–2020)squeezed light in routine operation; calibration uncertainty characterised to few percent
- O4(from 2023): further sensitivity improvements
The engineering claim, then, is not a single number but a defended envelope. LIGO measures what it measures because the vacuum is hard enough, the mirrors are still enough, the suspensions are soft enough at the right frequencies, and the calibration is traceable to known forces. The physics result — the chirp, the merger, the distance — inherits its credibility entirely from the instrument. A length change smaller than a proton is not observed. It is inferred, carefully, from light that has been made to care about it.