Skip to content

Squeezed States as Entangled Modes

Squeezed states are oscillator states in which noise in one quadrature is reduced below the vacuum level while the conjugate quadrature becomes noisier. They are central in quantum optics because they can be prepared, transformed by linear optics, measured by homodyne detection, and used to generate continuous-variable entanglement.

This page connects squeezing to mode entanglement. A single-mode squeezed state is a one-mode nonclassical state, not automatically a bipartite entangled state. A nontrivial two-mode squeezed vacuum is entangled across the two mode factors and is a finite-energy Gaussian approximation to EPR correlations. Squeezing a sufficiently noisy mixed input need not produce entanglement.

Helpful background. Gaussian covariance matrices describe the variances, and creation and annihilation operators describe the pair amplitudes.

Throughout, quadratures are dimensionless:

q=a+a†2,p=a−a†i2,[q,p]=i,q = \frac{a+a^\dagger}{\sqrt2}, \qquad p = \frac{a-a^\dagger}{i\sqrt2}, \qquad [q,p]=i,

so the vacuum variances are

Var⁡0(q)=Var⁡0(p)=12.\operatorname{Var}_0(q) = \operatorname{Var}_0(p) = \frac12.

For a real squeezing parameter rr, define the single-mode squeeze operator

S1(r)=exp⁡ ⁣[r2(a2−a†2)].S_1(r) = \exp\!\left[ \frac r2 \left( a^2-a^{\dagger 2} \right) \right].

With this phase convention,

S1†(r)aS1(r)=acosh⁡r−a†sinh⁡r.S_1^\dagger(r)aS_1(r) = a\cosh r-a^\dagger\sinh r.

Therefore

S1†(r)qS1(r)=e−rq,S1†(r)pS1(r)=erp.S_1^\dagger(r)qS_1(r) = e^{-r}q, \qquad S_1^\dagger(r)pS_1(r) = e^{r}p.

The squeezed vacuum

∣sq(r)⟩=S1(r)∣0⟩\lvert\mathrm{sq}(r)\rangle = S_1(r)\lvert0\rangle

has covariance matrix

Vsq(r)=12(e−2r00e2r).V_{\rm sq}(r) = \frac12 \begin{pmatrix} e^{-2r} & 0\\ 0 & e^{2r} \end{pmatrix}.

Thus

Var⁡(q)=12e−2r,Var⁡(p)=12e2r,\operatorname{Var}(q) = \frac12e^{-2r}, \qquad \operatorname{Var}(p) = \frac12e^{2r},

and the uncertainty product remains minimal:

Var⁡(q)Var⁡(p)=14.\operatorname{Var}(q)\operatorname{Var}(p) = \frac14.

In the number basis, the same state contains only even occupation numbers:

∣sq(r)⟩=1cosh⁡r∑n=0∞(−tanh⁡r)n(2n)!2nn!∣2n⟩.\lvert\mathrm{sq}(r)\rangle = \frac{1}{\sqrt{\cosh r}} \sum_{n=0}^{\infty} (-\tanh r)^n \frac{\sqrt{(2n)!}}{2^n n!} \lvert2n\rangle.

This even-number structure is a signature of pair creation into the same mode. It is nonclassical, but by itself it is still a state of one mode. Entanglement requires a specified split into at least two subsystems or modes.

For two modes AA and BB, define the two-mode squeeze operator

S2(r)=exp⁡ ⁣[r(aA†aB†−aAaB)].S_2(r) = \exp\!\left[ r \left( a_A^\dagger a_B^\dagger-a_Aa_B \right) \right].

Acting on the two-mode vacuum gives

∣TMSV(r)⟩=S2(r)∣0A,0B⟩=1cosh⁡r∑n=0∞(tanh⁡r)n∣nA,nB⟩.\lvert\mathrm{TMSV}(r)\rangle = S_2(r)\lvert0_A,0_B\rangle = \frac{1}{\cosh r} \sum_{n=0}^{\infty} (\tanh r)^n \lvert n_A,n_B\rangle.

This formula shows the physical pair structure: excitations are created in mode pairs. The two modes always have the same occupation number in this ideal state.

The coefficients are normalized because

∑n=0∞(tanh⁡2r)ncosh⁡2r=1.\sum_{n=0}^{\infty} \frac{(\tanh^2 r)^n}{\cosh^2 r} = 1.

The mean occupation of each mode is

nˉ=sinh⁡2r.\bar n = \sinh^2 r.

The transformation of annihilation operators is

S2†(r)aAS2(r)=aAcosh⁡r+aB†sinh⁡r,S_2^\dagger(r)a_A S_2(r) = a_A\cosh r+a_B^\dagger\sinh r,

and similarly with AA and BB interchanged. To check the sign, write G=aA†aB†−aAaBG=a_A^\dagger a_B^\dagger-a_Aa_B. Differentiating A(r)=e−rGaAerGA(r)=e^{-rG}a_Ae^{rG} gives A′(r)=e−rG[aA,G]erG=B†(r)A'(r)=e^{-rG}[a_A,G]e^{rG}=B^\dagger(r), and a second derivative gives A′′(r)=A(r)A''(r)=A(r). The initial values A(0)=aAA(0)=a_A and A′(0)=aB†A'(0)=a_B^\dagger yield the displayed hyperbolic functions.

This is a Bogoliubov transformation: annihilation operators mix with creation operators while preserving the commutators. Here S2S_2 is an active unitary on two fixed mode factors; it creates excitations from their vacuum.

The two-mode squeezed vacuum is already in Schmidt form:

∣TMSV(r)⟩=∑n=0∞pn ∣n⟩A∣n⟩B,\lvert\mathrm{TMSV}(r)\rangle = \sum_{n=0}^{\infty} \sqrt{p_n}\, \lvert n\rangle_A\lvert n\rangle_B,

with

pn=(tanh⁡2r)ncosh⁡2r.p_n = \frac{(\tanh^2 r)^n}{\cosh^2 r}.

For r=0r=0, only p0p_0 is nonzero and the state is the product vacuum. For any r>0r>0, at least two Schmidt coefficients are nonzero, so the state is entangled across the A,BA,B mode split.

Tracing out mode BB gives

ρA=∑n=0∞pn∣n⟩A⟨n∣.\rho_A = \sum_{n=0}^{\infty} p_n \lvert n\rangle_A\langle n\rvert.

This is a thermal oscillator state with mean occupation nˉ=sinh⁡2r\bar n=\sinh^2 r. The entanglement entropy is

SA=(nˉ+1)log⁡2(nˉ+1)−nˉlog⁡2nˉ.S_A = (\bar n+1)\log_2(\bar n+1) -\bar n\log_2\bar n.

The entropy is in bits. Locally this density operator is indistinguishable from a thermal state of the same oscillator: the distinction lies in its correlations with mode BB, not in any measurement on AA alone.

Single-mode squeezing can also be converted into two-mode entanglement by linear optics. For the real balanced transformation qC=(qA+qB)/2q_C=(q_A+q_B)/\sqrt2, qD=(qA−qB)/2q_D=(q_A-q_B)/\sqrt2 and the same transformation for pp, choose input AA squeezed in pp and input BB squeezed in qq, with equal strength rr. Then qC−qD=2qBq_C-q_D=\sqrt2q_B and pC+pD=2pAp_C+p_D=\sqrt2p_A each have variance e−2re^{-2r}: the output is the two-mode squeezed vacuum in our phase convention. If instead both inputs have the same squeeze strength and angle, this beam splitter leaves a product of identical squeezed states. The relative squeezing angle matters.

The two-mode squeeze transformation gives

S2†(r)(qA−qB)S2(r)=e−r(qA−qB),S_2^\dagger(r)(q_A-q_B)S_2(r) = e^{-r}(q_A-q_B),

and

S2†(r)(pA+pB)S2(r)=e−r(pA+pB).S_2^\dagger(r)(p_A+p_B)S_2(r) = e^{-r}(p_A+p_B).

Since the two-mode vacuum has

Var⁡0(qA−qB)=Var⁡0(pA+pB)=1,\operatorname{Var}_0(q_A-q_B) = \operatorname{Var}_0(p_A+p_B) = 1,

the two-mode squeezed vacuum has

Var⁡(qA−qB)=e−2r,Var⁡(pA+pB)=e−2r.\operatorname{Var}(q_A-q_B) = e^{-2r}, \qquad \operatorname{Var}(p_A+p_B) = e^{-2r}.

These are the finite-squeezing versions of ideal EPR correlations. The collective observables commute:

[qA−qB,pA+pB]=0.[q_A-q_B,p_A+p_B] = 0.

As r→∞r\to\infty, these variances tend to zero, but the mean occupation nˉ=sinh⁡2r\bar n=\sinh^2 r diverges. Exact EPR correlations are therefore an infinite-energy idealization in this family, not a laboratory state.

The Gaussian States Preview gives the covariance-matrix version of the same facts. The EPR State Preview explains why the exact EPR state is a generalized, non-normalizable limit.

Squeezed and two-mode squeezed states appear naturally in systems with parametric interactions. Schematic Hamiltonians include

Hsq=iℏκ2(a2−a†2)H_{\rm sq} = \frac{i\hbar\kappa}{2} \left( a^2-a^{\dagger 2} \right)

for degenerate squeezing, so U(t)=S1(κt)U(t)=S_1(\kappa t) with the convention above, and

Htms=iℏκ(aA†aB†−aAaB)H_{\rm tms} = i\hbar\kappa \left( a_A^\dagger a_B^\dagger-a_Aa_B \right)

for two-mode pair creation, so U(t)=S2(κt)U(t)=S_2(\kappa t). Here κ\kappa is real and has units of inverse time. These effective interaction-picture Hamiltonians assume a classical undepleted pump, suitable phase matching or resonance, and omission of rapidly oscillating terms. A different pump phase changes the squeezed quadrature; pump depletion limits indefinitely extrapolating the model.

Applications include:

  • sub-vacuum noise measurements in a chosen quadrature;
  • gravitational-wave interferometry and other precision measurements;
  • continuous-variable teleportation and dense coding protocols;
  • EPR steering and entanglement verification with homodyne detection;
  • microwave and optical parametric amplifiers;
  • pair generation in parametric down-conversion and four-wave mixing.

These applications are not automatic wins. Loss, phase noise, detector inefficiency, finite squeezing, mode mismatch, and excess thermal noise can quickly degrade the useful correlations. A trustworthy claim should state the modes, the squeezing convention, the measured quadratures, and the relevant noise model.

For a pure-loss channel with transmissivity η\eta, each output quadrature is qout=ηq+1−ηqvacq_{\rm out}=\sqrt\eta q+\sqrt{1-\eta}q_{\rm vac}, with an independent vacuum environment. Thus a single squeezed variance becomes

Vq,out=12[ηe−2r+1−η].V_{q,\rm out}=\frac12[\eta e^{-2r}+1-\eta].

If both modes of the squeezed vacuum suffer equal independent pure loss, each EPR variance becomes ηe−2r+1−η\eta e^{-2r}+1-\eta. For r>0r>0 and any η>0\eta>0 this remains below the separable threshold one per variance, though the violation becomes small. Thermal environmental noise can instead remove the violation and eventually the entanglement. These conclusions assume pure loss; they are not universal for arbitrary noise.

Squeezed light develops optical preparation and homodyne detection. Squeezing in metrology explains why reduced noise must be compared with signal response and readout costs, and Bogoliubov quasiparticles applies related transformations to collective excitations.

  • S. L. Braunstein and P. van Loock, “Quantum information with continuous variables,” Reviews of Modern Physics 77, 513–577 (2005), doi:10.1103/RevModPhys.77.513.
  • M. O. Scully and M. S. Zubairy, Quantum Optics, Cambridge University Press (1997), doi:10.1017/CBO9780511813993.
  • D. F. Walls and G. J. Milburn, Quantum Optics, 2nd ed., Springer (2008).
  • C. Weedbrook, S. Pirandola, R. García-Patrón, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, “Gaussian quantum information,” Reviews of Modern Physics 84, 621–669 (2012), doi:10.1103/RevModPhys.84.621.
  1. Single-mode uncertainty product. Use the squeezed variances to show that a single-mode squeezed vacuum remains a minimum-uncertainty state.
Solution

The variances are

Var⁡(q)=12e−2r,Var⁡(p)=12e2r.\operatorname{Var}(q) = \frac12e^{-2r}, \qquad \operatorname{Var}(p) = \frac12e^{2r}.

Their product is

Var⁡(q)Var⁡(p)=14e−2re2r=14.\operatorname{Var}(q)\operatorname{Var}(p) = \frac14e^{-2r}e^{2r} = \frac14.
  1. Two squeezing descriptions. Put b±=(aA±aB)/2b_\pm=(a_A\pm a_B)/\sqrt2. Express S2(r)S_2(r) as single-mode squeeze operators on b+b_+ and b−b_-.
Solution

The generator becomes

aA†aB†−aAaB=12[(b+†2−b+2)−(b−†2−b−2)].a_A^\dagger a_B^\dagger-a_Aa_B =\frac12[(b_+^{\dagger2}-b_+^2)-(b_-^{\dagger2}-b_-^2)].

The two terms commute, so S2(r)=S1,+(−r)S1,−(r)S_2(r)=S_{1,+}(-r)S_{1,-}(r). The same state is product in the +,−+,- mode split and entangled in the A,BA,B split. This equality describes a change of mode factors; implementing a beam splitter on fixed physical ports is an active operation.

  1. Reduced state. Trace out mode BB from the two-mode squeezed vacuum and show that the result is diagonal in the number basis.
Solution

Start from

ρAB=∑m,npmpn ∣m⟩A∣m⟩B⟨n∣A⟨n∣B.\rho_{AB} = \sum_{m,n} \sqrt{p_m p_n}\, \lvert m\rangle_A\lvert m\rangle_B \langle n\rvert_A\langle n\rvert_B.

Taking the trace over BB gives ⟨n∣m⟩B=δmn\langle n\vert m\rangle_B=\delta_{mn}, so all off-diagonal terms in m,nm,n vanish:

ρA=∑npn∣n⟩A⟨n∣.\rho_A = \sum_n p_n \lvert n\rangle_A\langle n\rvert.
  1. Loss calculation. Each mode has e−2r=1/4e^{-2r}=1/4 and experiences independent pure loss with η=0.6\eta=0.6. Find the sum of output EPR variances and compare with the separable bound.
Solution

Each variance is 0.6/4+0.4=0.550.6/4+0.4=0.55. Their sum is 1.10<21.10<2, so entanglement remains certified. For a thermal environment of mean occupation nenvn_{\rm env}, the same calculation instead gives each variance ηe−2r+(1−η)(2nenv+1)\eta e^{-2r}+(1-\eta)(2n_{\rm env}+1); vacuum noise and thermal noise must not be interchanged.

  1. EPR variance. Use
S2†(r)(qA−qB)S2(r)=e−r(qA−qB)S_2^\dagger(r)(q_A-q_B)S_2(r) = e^{-r}(q_A-q_B)

to compute Var⁡(qA−qB)\operatorname{Var}(q_A-q_B) in the two-mode squeezed vacuum.

Solution

In the two-mode squeezed state,

Var⁡TMSV(qA−qB)=Var⁡0(e−r(qA−qB)).\operatorname{Var}_{\rm TMSV}(q_A-q_B) = \operatorname{Var}_{0} \left( e^{-r}(q_A-q_B) \right).

The two-mode vacuum has Var⁡0(qA−qB)=1\operatorname{Var}_0(q_A-q_B)=1, because each vacuum mode contributes 1/21/2 and there is no cross-correlation. Hence

Var⁡TMSV(qA−qB)=e−2r.\operatorname{Var}_{\rm TMSV}(q_A-q_B) = e^{-2r}.