Intensity Interferometry: The Hanbury Brown-Twiss Stellar Diameter Trick

What it is: correlating flicker instead of waves

All light flickers. Because thermal (chaotic) light from a star is a random superposition of many independent wave trains, its intensity fluctuates on the timescale of its coherence time τ_c ≈ 1/Δν, where Δν is the optical bandwidth. For a filter admitting a few nanometers of visible light, τ_c is on the order of 10⁻¹⁴ seconds.

Here is the key insight: those fluctuations are partially correlated between two detectors that see the same star. If the two light buckets sit close together, they catch the same flicker in lockstep. As you pull them apart, the correlation fades. Crucially, the separation at which the correlation dies off is set by the angular size of the source, not by any property of the detectors.

• Detectors record only intensity (photon count rate) — no phase, no interference fringes.
• The two signals are multiplied together and time-averaged by an electronic correlator.
• The residual correlation above random-chance level is the Hanbury Brown-Twiss (HBT) signal.

Because it discards the phase and works with slow electronic signals, the method is astonishingly robust to the atmosphere that destroys ordinary optical interferometry.

The mechanism: second-order coherence and photon bunching

Formally, the technique measures the normalized second-order correlation function:

g²(τ) = ⟨I(t) I(t+τ)⟩ / ⟨I(t)⟩²

For chaotic (thermal) light this obeys the Siegert relation, g²(τ) = 1 + |g¹(τ)|², where g¹ is the ordinary (amplitude) degree of coherence. At zero delay and zero baseline, g² = 2 — photons arrive in bunches rather than at random, a purely statistical consequence of their bosonic nature. Random (coherent-laser) light would give g² = 1.

This is the trick's engine: measuring the square of the visibility, |g¹|², lets you recover the visibility modulus without ever phasing the two beams together. The spatial part of g¹ across a baseline B is governed by the Van Cittert-Zernike theorem — the visibility is the Fourier transform of the source's brightness distribution. So mapping the correlation versus baseline directly maps the star's brightness profile in the Fourier plane.

When Edward Purcell (Nature, 1956) defended the effect against skeptics, and Roy Glauber later built the full quantum theory (Nobel Prize, 2005), photon bunching became textbook physics.

Key quantities: a worked Sirius example

For a uniform circular disk of angular diameter θ, the visibility |g¹| first hits zero when the baseline reaches

B₀ = 1.22 λ / θ

the exact analogue of the Airy-disk formula. Rearranged, θ = 1.22 λ / B₀.

• Sirius (1956): Hanbury Brown and Twiss found correlation falling toward zero near B ≈ 9 m at λ ≈ 443 nm. Plugging in gives θ ≈ 1.22 × (443×10⁻⁹ m) / (9 m) ≈ 6×10⁻⁸ rad ≈ 0.0063 arcseconds — 6.3 milliarcseconds, matching the modern value.
• Resolution scaling: at Narrabri's maximum B = 188 m, the reach was ~0.0004 arcsec, fine enough for hot main-sequence stars only a fraction of a milliarcsecond wide.
• Coherence time: τ_c ~ 10⁻¹⁴ s means the true HBT signal is diluted by the electronic resolving time (~ns) by a factor τ_c/τ_electronic ~ 10⁻⁵ — which is why only bright stars work and why long integrations (hours) are needed.

How it's observed: from Narrabri to Cherenkov arrays

The Narrabri Stellar Intensity Interferometer (NSW, Australia, 1963–1974) was the flagship instrument. Two 6.5-meter mosaic-mirror light collectors rode on a 188-m-diameter circular railway track, letting operators vary the baseline from about 10 m to 188 m while keeping both aimed at the target. Photomultipliers fed a broadband electronic correlator.

Over eleven years the NSII delivered the first catalogue of 32 stellar angular diameters, all hotter than ~7000 K and brighter than about mV = 2.5. From these Hanbury Brown built the first wholly empirical effective-temperature scale for hot stars and even detected limb darkening — the first such measurement on any star but the Sun.

The technique has been revived since 2019 using Imaging Atmospheric Cherenkov Telescopes — huge, cheap gamma-ray light buckets. VERITAS (Arizona) resolved the sub-milliarcsecond stars β Canis Majoris and ε Orionis to better than 5% (Nature Astronomy, 2020), and the H.E.S.S. array in Namibia has performed multi-color intensity interferometry, exploiting baselines of 100+ meters and modern GHz digital correlators.

How it compares to amplitude interferometry and other regimes

The essential trade is phase for robustness. A Michelson/amplitude interferometer (Michelson & Pease measured Betelgeuse this way in 1920; today CHARA and the VLTI carry the torch) combines the actual light beams and records fringes, capturing the complex visibility — amplitude and phase — which enables true image reconstruction. But it demands the two paths be matched to a fraction of a wavelength, so atmospheric turbulence and vibration are relentless enemies and baselines are limited to a few hundred meters at enormous cost.

• Amplitude (first-order): more sensitive, gives phase, reaches faint stars — but fragile and expensive.
• Intensity (second-order): loses phase and sensitivity, but nanosecond electronic delays make it nearly turbulence-proof and cheap to scale to kilometer baselines.
• Speckle interferometry: a third route that freezes turbulence in short exposures — good for close binaries, but diffraction-limited to a single aperture.

Because intensity interferometry needs only cheap, optically crude collectors, it is the natural way to reach ultra-high resolution in the blue/UV, where amplitude interferometers struggle most — a niche it still uniquely occupies.

Significance, famous cases, and open questions

Intensity interferometry's deepest legacy is conceptual: the HBT effect launched quantum optics. The 1956 Sirius result and the ensuing Purcell–Glauber theoretical clarification forced physicists to reckon with photon statistics, and the same correlation technique now probes everything from cold-atom clouds to heavy-ion collisions (where HBT correlations size the fireball) and free-electron beams (which show anti-bunching).

Its astronomical payoff was the empirical hot-star temperature scale and 32 measured diameters that anchored stellar astrophysics for a generation. The modern revival aims higher:

• Resolving stellar surface features, oblateness of fast rotators, and circumstellar disks at sub-milliarcsecond scale.
• Building arrays of many Cherenkov telescopes (the future CTA could give hundreds of simultaneous baselines) to reconstruct genuine images despite the missing phase.

Open challenges remain: recovering phase information from purely intensity data (algorithms borrowed from radio and X-ray crystallography), pushing to fainter targets with photon-counting detectors and larger bandwidths, and combining many baselines fast enough to freeze a star's rotation. The Hanbury Brown-Twiss trick, once dismissed as impossible, is enjoying a genuine renaissance.

Intensity (Hanbury Brown-Twiss) interferometry versus classical amplitude (Michelson) interferometry

Property	Intensity interferometry (HBT)	Amplitude interferometry (Michelson)
Quantity correlated	Intensity fluctuations I₁·I₂ (second-order, g²)	Electric-field amplitudes (first-order, g¹)
Measures	|g¹|² → visibility modulus only (no phase)	Complex visibility (amplitude + phase)
Atmospheric sensitivity	Very low — nanosecond electronic delays tolerate turbulence	Extreme — needs optical-wavelength path stability (λ/10)
Practical baseline	Up to ~km with cheap light buckets (Narrabri 188 m)	Tens to ~330 m (VLTI, CHARA), costly delay lines
Sensitivity limit	Only bright stars (mV ≲ 2–3, historically)	Reaches much fainter targets
Signal timescale	Coherence time ~10⁻⁹–10⁻¹⁵ s; needs GHz electronics	Fringe tracked at optical frequencies