Quantum Optics
Squeezed Light: Beating the Vacuum Noise Floor
Empty space is never quiet: the electromagnetic vacuum jitters with zero-point fluctuations that set a hard "shot-noise" floor on every optical measurement. Squeezed light cheats this floor — it redistributes the unavoidable ½ℏω of quantum uncertainty out of the observable you care about and dumps it into the one you don't. The current record squeezes one quadrature to 15 dB below vacuum (a factor of ≈32 in variance, Vahlbruch et al. 2016), and since 2019 the LIGO and Virgo gravitational-wave detectors have run permanently on squeezed vacuum, boosting their event rate by roughly 50%.
Formally, a squeezed state is a minimum-uncertainty state of the light field in which the variance of one quadrature X̂ falls below the coherent-state (vacuum) value ¼, at the cost of the conjugate quadrature P̂ rising above it — the Heisenberg product ΔX·ΔP ≥ ¼ stays saturated, but the noise ellipse is rotated and squashed.
- RegimeNonclassical continuous-variable quantum optics
- Key relationΔX = e^-r·½, ΔP = e^r·½; ΔX·ΔP = ¼
- First observedSlusher et al. 1985 (four-wave mixing in Na vapor, ~0.3 dB)
- Record squeezing15 dB below vacuum (×32 in variance), 1064 nm — Vahlbruch 2016
- Realized inχ⁽²⁾ optical parametric oscillators (PPKTP); measured by balanced homodyne
- Matters forGravitational-wave detection (LIGO/Virgo), metrology, CV quantum computing
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What squeezed light is and why the vacuum has a noise floor
Quantize a single mode of the electromagnetic field and it becomes a harmonic oscillator: Ĥ = ℏω(â†â + ½). The stubborn ½ℏω is the zero-point energy, and it is not just an energy bookkeeping term — it manifests as genuine fluctuations of the field amplitude even when no photons are present. Write the field in terms of two quadratures, X̂ = (â + â†)/2 and P̂ = (â − â†)/2i (the in-phase and out-of-phase components, the optical analogues of position and momentum). These obey [X̂, P̂] = i/2, so ΔX·ΔP ≥ ¼ (conventions vary; here vacuum variance = ¼).
A coherent state — the best a laser can do — spreads this minimum uncertainty equally: ΔX = ΔP = ½. Detect its intensity and you see shot noise, the √n̄ Poissonian graininess that limits every precision optical measurement. Squeezed light breaks the equality. It is a state where one quadrature's variance drops below ½ while the other rises to compensate — the noise disk of the vacuum is deformed into an ellipse, and you orient the thin axis onto your signal.
The mechanism: parametric amplification and the squeeze operator
Squeezing is generated by a process that is phase-sensitive: it amplifies one quadrature and de-amplifies the orthogonal one. The workhorse is a χ⁽²⁾ nonlinearity — a nonlinear crystal driven by a strong pump at 2ω. One pump photon splits into two correlated photons at ω (degenerate parametric down-conversion). Because the two daughter photons are created as a pair, their amplitudes are correlated, and the quantum interference between adding and removing pairs sharpens one quadrature.
The interaction Hamiltonian Ĥ ∝ iℏ(ξ*â² − ξ↲) generates the unitary squeeze operator Ŝ(ξ) = exp[½(ξ*â² − ξ↲)], with ξ = r·e^(iθ). Its action is a Bogoliubov transformation: Ŝ†âŜ = â·cosh r − ↷e^(iθ)·sinh r. The â² and ↲ terms create and annihilate photons two at a time — which is why an ideal squeezed vacuum contains only even photon numbers. The parameter r (the squeeze factor) scales the two quadratures oppositely: the balanced act of pair creation and annihilation is what lets one variance shrink while the conjugate one, forced by Heisenberg, must grow by exactly the reciprocal factor.
The key relation: e^(±2r), decibels, and how loss kills squeezing
Under Ŝ(r) with θ = 0 the quadrature variances transform as
Var(X̂) = ¼·e^(−2r), Var(P̂) = ¼·e^(+2r),
so ΔX·ΔP = ¼ is still saturated — squeezed vacuum is a minimum-uncertainty state, just a rotated, squashed one. The noise reduction is quoted in decibels: squeezing (dB) = −10·log₁₀(e^(−2r)) = 20r/ln10 ≈ 8.686·r. Thus 15 dB corresponds to r ≈ 1.73 and a variance reduction of ×31.6; 10 dB is r ≈ 1.15.
The brutal enemy is optical loss. A squeezed state passing through a channel of efficiency η is mixed with vacuum, and the measured variance becomes V_meas = η·e^(−2r) + (1−η). Even with infinite r, 5% loss caps you at 13 dB. This is why reaching 15 dB demanded near-unity everything: custom PPKTP crystals, ppm-level scatter, and an InGaAs photodiode measured at 99.5% quantum efficiency at 1064 nm. The anti-squeezed quadrature, meanwhile, grows unbounded — pump depletion and loss make real states impure, with Var(X̂)·Var(P̂) > ¼.
How it is generated and measured: OPOs and balanced homodyne
The dominant modern source is the sub-threshold optical parametric oscillator (OPO) or amplifier (OPA): a periodically poled crystal (PPKTP, or LiNbO₃) inside a cavity, pumped at 532 nm to squeeze 1064 nm. Below the oscillation threshold the cavity emits squeezed vacuum rather than a bright beam. Early demonstrations used four-wave mixing in atomic vapor — Slusher, Yurke and colleagues at Bell Labs saw the first ~0.3 dB dip in 1985 in sodium — and Wu, Kimble et al. reached ~3.5 dB with an OPO in 1986.
You cannot read a quadrature variance with a power meter; you need a phase reference. Balanced homodyne detection supplies it: the squeezed field is combined on a 50/50 beamsplitter with a strong local oscillator (LO) derived from the same laser, and the two outputs are subtracted. The difference photocurrent measures the quadrature X̂_φ selected by the LO phase φ. Sweeping φ traces out the noise ellipse — the signature is a variance that dips below the independently calibrated shot-noise (vacuum) level and rises above it a quarter-cycle later.
Where it operates: from gravitational-wave detectors to related nonclassical states
The flagship application is gravitational-wave interferometry. LIGO and Virgo are shot-noise-limited at high frequencies, so injecting squeezed vacuum into the interferometer's dark port suppresses that noise directly. Introduced in GEO600 (2010) and Advanced LIGO (from O3, 2019), squeezing cut phase noise by ~2–3 dB early on and 4.6 dB (Hanford) to 5.8 dB (Livingston) in O4 — enough to raise the detection rate by roughly half.
A subtlety: squeezing one quadrature worsens the conjugate one, so pure phase-squeezing lowers high-frequency shot noise but raises low-frequency radiation-pressure noise (the anti-squeezed amplitude kicks the mirrors harder). The cure is frequency-dependent squeezing: a 300 m filter cavity (LIGO A+ upgrade) rotates the squeeze angle with frequency, so shot noise is squeezed above the crossover and radiation-pressure noise below it — broadband quantum enhancement. Distinguish squeezed light from entanglement (two-mode squeezing yields EPR correlations and is the CV analogue of a Bell pair) and from sub-Poissonian light from single emitters.
Applications, significance, and open problems
Beyond gravitational waves, squeezed light is a metrology and quantum-technology primitive. It enhances biological imaging and particle tracking where photodamage caps the usable power, improves magnetometry and atomic-clock spectroscopy, and Vahlbruch's team used 15 dB states to absolutely calibrate photodiode quantum efficiency to 99.5% ± 0.5%. Two-mode squeezed vacuum is the backbone of continuous-variable quantum computing and photonic quantum advantage: Gaussian boson sampling machines (Jiuzhang, Borealis) inject dozens of squeezed modes into interferometers.
Open frontiers remain. Loss is the perennial ceiling — pushing well past 15 dB needs sources and detectors with sub-percent total loss. In gravitational-wave detectors, the next gains come from longer filter cavities, higher intracavity squeezing, and beating thermal/coating noise so quantum noise stays the limit. Deeper questions concern non-Gaussian resources: squeezing alone is a Gaussian operation and is efficiently classically simulable, so photon subtraction/addition and cat-state breeding are needed to reach universal, fault-tolerant CV computation. Squeezing turned the vacuum from an immovable floor into an engineerable resource.
| State | ΔX (squeezed quad.) | ΔP (anti-squeezed quad.) | Uncertainty product | Photon statistics |
|---|---|---|---|---|
| Vacuum |0⟩ | ½ | ½ | ¼ (minimum) | no photons; zero-point only |
| Coherent |α⟩ (ideal laser) | ½ | ½ | ¼ (minimum) | Poissonian, Δn = √n̄ |
| Amplitude-squeezed | < ½ | > ½ | ¼ (if pure) | sub-Poissonian, Δn < √n̄ |
| Squeezed vacuum | e^-r·½ | e^r·½ | ¼ (if pure) | even photon numbers only |
| Thermal | > ½ | > ½ | > ¼ (mixed) | super-Poissonian, bunched |
Frequently asked questions
Does squeezed light violate the Heisenberg uncertainty principle?
No. The product ΔX·ΔP ≥ ¼ is always respected. Squeezing only redistributes the fixed uncertainty budget: it pushes one quadrature's variance below the ¼ vacuum level while forcing the conjugate quadrature above it. An ideal (pure) squeezed state sits exactly on the Heisenberg bound, just as a coherent state does — it is a minimum-uncertainty state with an elliptical, rather than circular, noise distribution.
Why does the record stop at 15 dB — why not 30 or 40 dB?
Optical loss, not the nonlinearity, is the bottleneck. Any inefficiency η mixes vacuum into the beam, giving measured variance η·e^(-2r) + (1-η). Because the (1-η) term is a floor independent of r, even 3% total loss caps you near 15 dB regardless of how hard you pump. The 2016 record required near-unity efficiency everywhere: low-scatter PPKTP, minimal propagation loss, and a photodiode with 99.5% quantum efficiency.
How is squeezing actually detected if a photodetector just measures intensity?
Through balanced homodyne detection. The squeezed field is mixed on a 50/50 beamsplitter with a bright local oscillator from the same laser, and the two output photocurrents are subtracted. The difference current is proportional to the field quadrature X̂_φ selected by the local-oscillator phase φ. Sweeping φ maps out the noise ellipse; the signature is a variance dipping below the separately measured shot-noise (vacuum) reference and rising above it 90° later.
Why does squeezed vacuum contain only even photon numbers?
The squeeze operator is generated by the â² and ↲ terms of the parametric Hamiltonian, which create and annihilate photons strictly in pairs. Acting on the vacuum, it therefore populates only |0⟩, |2⟩, |4⟩, ... The photon-number distribution is that of correlated pairs, and measuring an odd count in an ideal squeezed vacuum is impossible — a direct fingerprint of the pairwise down-conversion mechanism.
How much did squeezed light actually improve LIGO?
Squeezing was introduced in Advanced LIGO's O3 run (2019), reducing quantum shot noise by about 2–3 dB and increasing the observable volume and event rate meaningfully. With the A+ upgrade and frequency-dependent squeezing in O4, noise fell by 4.6 dB at Hanford and 5.8 dB at Livingston, boosting the detection rate by roughly 50%. It is now a permanent, always-on subsystem of the detectors.
What is frequency-dependent squeezing and why is it needed?
Squeezing the phase quadrature lowers high-frequency shot noise but the anti-squeezed amplitude quadrature increases radiation-pressure noise, which dominates at low frequencies. Frequency-dependent squeezing rotates the squeeze angle as a function of frequency so the right quadrature is squeezed in each band. LIGO achieves this by reflecting the squeezed vacuum off a 300 m detuned filter cavity before injection, giving broadband quantum enhancement from tens of Hz to several kHz.