The Speed of Gravity: How a 1.7-Second Delay Pinned c_gw to c

What 'the speed of gravity' actually means

In Newton's theory gravity is instantaneous: move a mass and its field updates everywhere at once, implying an infinite propagation speed. Einstein's general relativity replaced that with a field theory in which disturbances in the spacetime metric travel as gravitational waves at a finite speed, c_gw.

General relativity makes a sharp prediction: gravitational waves are massless and propagate at exactly the speed of light, so c_gw = c in vacuum, with no dispersion (all frequencies travel at the same speed). This is not an add-on assumption but a structural feature — the same Lorentzian metric that carries light also carries gravity.

• If the graviton had mass m_g, high frequencies would outrun low ones, and c_gw would fall slightly below c.
• Many modified-gravity theories (extra fields, Lorentz-violating terms) predict c_gw ≠ c at some level.

Measuring c_gw is therefore a clean, model-independent test of whether gravity really is the geometric, luminal phenomenon Einstein described.

The mechanism: a time-of-flight race across 40 megaparsecs

The measurement is conceptually a stopwatch experiment. Two messengers — a gravitational wave and light — leave the same source and race to Earth. If they travel at different speeds, the lead accumulates over the entire path length D.

The arrival-time difference is Δt = D/c_em − D/c_gw. For a tiny fractional speed difference δ ≡ (c_gw − c)/c, this becomes approximately Δt ≈ −(D/c)·δ. The leverage comes from D being astronomically large: even a minuscule δ is amplified into a measurable delay by the 130-million-light-year baseline.

For GW170817, D ≈ 40 Mpc gives a light-travel time D/c of about 1.3×10⁸ years. The observed gap was only 1.74 s. Dividing seconds by 10⁸ years is what yields the ~10⁻¹⁵ precision:

• δ ≈ −Δt·c/D, so a 1.7 s discrepancy over that baseline maps to δ of order 10⁻¹⁵.
• The dominant uncertainty is astrophysical: nobody knows the exact lag between the merger and the moment gamma rays escaped the ejecta.

Key numbers and the worked bound

GW170817 was detected on 17 August 2017 at 12:41:04 UTC by Advanced LIGO and Advanced Virgo. It was a binary neutron-star inspiral with component masses in the range 1.17–1.60 M_sun (total ≈ 2.74 M_sun) at a luminosity distance of 40 (+8/−14) Mpc in the galaxy NGC 4993. About 1.74 ± 0.05 s later, Fermi-GBM and INTEGRAL's SPI-ACS caught the short burst GRB 170817A.

To turn Δt into a bound, the LIGO/Virgo team made conservative bracketing assumptions:

• Lower edge: assume the GW peak and the first photons left simultaneously, so the whole 1.74 s is a speed difference. Using the nearest allowed distance (26 Mpc, the low end of the 90% distance interval) gives δ ≥ −3×10⁻¹⁵.
• Upper edge: assume the gamma rays were emitted up to 10 s after the merger (jet launch takes time), which allows the GW to have been slightly faster: δ ≤ +7×10⁻¹⁶.

Combined: −3×10⁻¹⁵ ≤ (c_gw − c)/c ≤ +7×10⁻¹⁶ — a fifteen-order-of-magnitude improvement over prior bounds.

How it was observed: multi-messenger astronomy

The result required catching the same event in two fundamentally different channels within seconds — the birth of practical multi-messenger astronomy.

• Gravitational-wave channel: LIGO Hanford, LIGO Livingston, and Virgo tracked the inspiral chirp for ~100 s as the neutron stars spiraled from ~24 Hz up to merger frequencies of several hundred Hz, timestamping the coalescence precisely.
• Gamma-ray channel: Fermi-GBM independently triggered on GRB 170817A; INTEGRAL confirmed it. Their absolute clocks fixed the 1.74 s offset.
• Localization and follow-up: combining GW triangulation with Virgo's near-null shrank the sky area to ~28 deg², letting optical teams find the kilonova AT2017gfo in NGC 4993 within ~11 hours.

Crucially, the two timestamps come from independent instruments with independently calibrated clocks, so the delay is a real physical measurement, not a data-processing artifact. No pre-existing electromagnetic counterpart alert was needed.

How it relates to graviton mass, dispersion, and modified gravity

The c_gw = c result is one member of a family of tests, and it complements the others:

• Graviton mass: A massive graviton makes GWs dispersive (speed depends on frequency). LIGO/Virgo's analysis of the waveform's phase, plus c_gw ≈ c, bounds m_g below about 1.3×10⁻²³ eV/c², corresponding to a graviton Compton wavelength larger than a light-year.
• Binary pulsars: PSR B1913+16 (Hulse–Taylor) confirmed that orbits decay by radiating GWs, matching GR to ~0.2% — indirect proof gravity propagates, but not a direct speed measurement.
• Shapiro/PPN tests: Cassini measured light's extra delay in the Sun's field to γ − 1 ≈ 2×10⁻⁵, probing how gravity bends the paths of massless particles rather than gravity's own speed.

The GW170817 bound was decisive for cosmology: it instantly ruled out or severely constrained large classes of dark-energy and modified-gravity models (e.g. certain Horndeski, Galileon, and massive-gravity theories) that predicted c_gw ≠ c.

Significance, caveats, and open questions

Beyond confirming Einstein, the near-simultaneity underpinned the first standard-siren measurement of the Hubble constant (H₀ ≈ 70 km/s/Mpc from GW170817 alone) and tightened tests of Lorentz invariance and the equivalence principle via the Shapiro delay both messengers share.

Important caveats remain:

• The bound assumes the GW and photons were emitted within ~10 s of each other. That astrophysical prior — not detector precision — sets the width, so the number is a physics-informed limit, not a pure timing measurement.
• The constraint applies to LIGO's ~10–1000 Hz band; exotic frequency-dependent effects outside it aren't fully closed off.
• Only one BNS event supplies this bound; a nearby repeat with better emission modeling could sharpen it further.

Open questions: Is c_gw exactly c at all frequencies and cosmic epochs? Do any surviving modified-gravity models mimic GR here while differing elsewhere? Future detectors (LISA, Einstein Telescope, Cosmic Explorer) will extend the test across new frequency bands and larger distances.

How different probes constrain the speed and nature of gravity

Probe / method	Quantity constrained	Bound or result
GW170817 + GRB 170817A	(c_gw − c)/c	−3×10⁻¹⁵ to +7×10⁻¹⁶
GW dispersion (LIGO/Virgo)	Graviton mass m_g	< 1.3×10⁻²³ eV/c²
Binary pulsar PSR B1913+16	GW energy loss vs GR	Agrees to ~0.2%
Shapiro delay (Cassini)	Light propagation in gravity	γ − 1 = (2.1±2.3)×10⁻⁵
Solar-system ephemerides	Historic 'speed of gravity' estimates	≈ c (weak, model-dependent)