Operational Quantification of Continuous Variable Correlations
C. Rod\'o, Gerardo Adesso, A. Sanpera

TL;DR
This paper introduces a practical method to quantify quantum and classical correlations in continuous variable systems by measuring correlated bits via quadrature signs, applicable to Gaussian and non-Gaussian states, facilitating experimental entanglement assessment.
Contribution
It proposes an operational measure of correlations based on sign measurements of quadratures, which simplifies entanglement quantification without full state tomography.
Findings
Bit correlations majorize entanglement for Gaussian states.
For non-Gaussian states, bit correlations are monotonic with negativity.
Method enables direct homodyne detection-based entanglement measurement.
Abstract
We quantify correlations (quantum and/or classical) between two continuous variable modes in terms of how many correlated bits can be extracted by measuring the sign of two local quadratures. On Gaussian states, such `bit quadrature correlations' majorize entanglement, reducing to an entanglement monotone for pure states. For non-Gaussian states, such as photonic Bell states, ideal and real de-Gaussified photon-subtracted states, and mixtures of pure Gaussian states, the bit correlations are shown to be a {\em monotonic} function of the negativity. This yields a feasible, operational way to quantitatively measure non-Gaussian entanglement in current experiments by means of direct homodyne detection, without a full tomographical reconstruction of the Wigner function.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Operational quantification of continuous variable correlations
Carles Rodó
Grup de Física Teòrica, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain.
Gerardo Adesso
Grup de Física Teòrica, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain.
Dipartimento di Fisica “E. R. Canianello”, Universitá degli Studi di Salerno, Via S. Allende, 84081 Baronissi (SA), Italy.
Anna Sanpera
Grup de Física Teòrica, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain.
ICREA, Institució Catalana de Recerca i Estudis Avançats, Barcelona, 08021 Spain.
(November 12, 2007)
Abstract
We quantify correlations (quantum and/or classical) between two continuous variable modes in terms of how many correlated bits can be extracted by measuring the sign of two local quadratures. On Gaussian states, such ‘bit quadrature correlations’ majorize entanglement, reducing to an entanglement monotone for pure states. For non-Gaussian states, such as photonic Bell states, ideal and real de-Gaussified photon-subtracted states, and mixtures of pure Gaussian states, the bit correlations are shown to be a monotonic function of the negativity. This yields a feasible, operational way to quantitatively measure non-Gaussian entanglement in current experiments by means of direct homodyne detection, without a full tomographical reconstruction of the Wigner function.
pacs:
03.67.Mn, 42.50.Dv
Quantum information with continuous variables (CVs), relying on quadrature entanglement as a resource, has witnessed rapid and exciting progresses recently, also thanks to the high degree of experimental control achievable in the context of quantum optics books . While Gaussian states (coherent, squeezed, and thermal states) have been originally the preferred resources for both theoretical and practical implementations, a new frontier emerges with non-Gaussian states (Fock states, Schrödinger’s cats, …). The latter can be highly nonclassical, possess in general more entanglement than Gaussian states Wolf06 , and are useful to overcome some limitations of the Gaussian framework such as entanglement distillation NOGO and universal quantum computation Menic06 . Therefore, it is of central relevance to provide proper ways to quantify non-Gaussian entanglement in a way which is experimentally accessible.
At a fundamental level, the difficulty in the investigation of entanglement – quantum correlations – can be traced back to the subtle task of distinguishing it from classical correlations groisman . Correlations can be regarded as classical if they can be induced onto the subsystems solely by local operations and classical communication, necessarily resulting in a mixed state. On the other hand, if a pure quantum state displays correlations between the subsystems, they are of genuinely quantum nature (entanglement). We adopt here a pragmatic approach: if two systems are in toto correlated, then this correlation has to be retrieved between the outcomes of some local measurements performed on them. We, therefore, investigate quadrature correlations in CV states. We are also motivated by the experimental adequacy: field quadratures can be efficiently measured by homodyne detection, without the need for complete state tomography. Specifically, in this paper, we study optimal correlations in bit strings obtained by digitalizing the outcomes of joint quadrature measurements on a two-mode CV system. First, we apply our procedure to Gaussian states (GS), finding that bit quadrature correlations provide a clear-cut quantification of the total correlations between the two modes. They are monotonic with the entanglement on pure states, and can be arbitrarily large on mixed states, the latter possibly containing arbitrarily strong additional classical correlations. We then address non-Gaussian states (NGS), for which the exact detection of entanglement generally involves measurements of high-order moments Shchukin05 . The underlying idea is that for NGS obtained by de-Gaussifying an initial pure GS and/or by mixing it with a totally uncorrelated state, our measure based entirely on second moments is still expected to be a (quantitative) witness of the quantum part of correlations only, i.e. entanglement. We show that this is indeed the case for relevant NGS including photon-subtracted states, photonic Bell states, and mixtures of Gaussian states. Notably, the complete entanglement picture in a recently demonstrated coherent single-photon-subtracted state Ourjoumtsev07 is precisely reproduced here in terms of quadrature correlations only. Our results render non-Gaussian entanglement significantly more accessible in a direct, practical fashion.
Quadrature measurements and bit correlations.— We consider a bipartite CV system of two bosonic modes, and , described by an infinite-dimensional Hilbert space. The probability distribution associated to the measurement of the rotated position quadrature in mode with outcome and uncertainty is given by , where is a single-mode Gaussian (squeezed) state with first moments and covariance matrix . Here is a unitary operator describing a rotation of on mode , corresponding to a symplectic transformation given by notenota . Thus, one can either measure the rotated quadrature on the state (passive view) or antirotate the state and measure the unrotated quadrature (active view). Similarly we define for mode . The probability distribution associated to a joint measurement of the rotated quadratures and , is given by .
We digitalize the obtained outputs by assigning the bits to the positive (negative) values of the measured quadrature. This digitalization transforms each joint quadrature measurement into a pair of classical bits. A string of such correlated bits can be used e.g. to distill a quantum key Navascues05 ; Rodo07 . Let us adopt a compact notation by denoting (at given angles ) , and . The conditional probability that the bits of the corresponding two modes coincide is given by . Correspondingly, the conditional probability that they differ is . Trivially, . If the measurement outcomes display correlations, otherwise they display anticorrelations. Notice that, if the two modes are completely uncorrelated, . For convenience, we normalize the strength of bit correlations as
[TABLE]
so that for a completely uncorrelated state . The interpretation of Eq. (1) in terms of correlations is meaningful if a fairness condition is satisfied: on each single mode, the marginal probabilities associated to the outcomes “” or “” must be the same: , notefair .
For a two-mode CV system, whose state is described by a Wigner function , we define the ‘bit quadrature correlations’ as the average probability of obtaining a pair of classically correlated bits (in the limit of zero uncertainty) optimized over all possible choices of local quadratures
[TABLE]
where is the marginal Wigner distribution of the (rotated) positions, and . After some algebra, we can rewrite Eq. (2) as , where is the ‘sign-binned’ quadrature correlation function, which has been employed e.g. in proposed tests of Bell inequalities violation for CV systems notebell . While this form is more suitable for an analytic evaluation on specific examples, the definition Eq. (2) is useful to prove the following general properties of :
Lemma 1** (Normalization)**
*: .
Proof. It follows from the definition of , as both and the marginal Wigner distribution range between [math] and . *
Lemma 2** (Zero on product states)**
*: .
Proof. For a product state the probabilities factorize i.e. and so , where we have used the fairness condition. Namely , hence the integral in Eq. (2) trivially vanishes. *
Lemma 3** (Local symplectic invariance)**
*: Let be a unitary operator amounting to a single-mode symplectic operation on the local phase space of mode notenota . Then .
Proof. Any single-mode symplectic operation can be decomposed in terms of local rotations and local squeezings (Euler decomposition). By definition Eq. (2) is invariant under local rotations, so we need to show that local squeezings, described by symplectic matrices of the form , also leave invariant. Adopting the passive view, the action of local squeezings on the covariance matrix of each Gaussian state is irrelevant, as we take eventually the limit . The first moments are transformed as , so that . On the other hand, the Wigner distribution is transformed as . Summing up, local squeezings transform into . As Eq. (2) involves integration over the four phase space variables , we change variables noting that , to conclude the proof. *
It follows from Lemmas (1–2) that if , then the state necessarily possesses correlations between the two modes. Lemma 3, moreover, suggests that embodies not only a qualitative criterion, but might be interpreted as bona fide operational quantifier of CV correlations. We will now show that this is the case for various important classes of states.
Gaussian states.— Even though entanglement of GS is already efficiently accessible via their covariance matrix books , we use such states as ‘test-beds’ for understanding the role of in discriminating CV correlations. The covariance matrix of any two-mode GS can be written in standard form as:
[TABLE]
where, without loss of generality, we adopt the convention . The covariance matrix describes a physical state if and , with . The negativity Vidal02 , quantifying entanglement between the two modes, reads , where with . For two-mode GS, Eq. (2) evaluates to:
[TABLE]
where the optimal quadratures are the standard unrotated positions (). First, we notice that describes a product state: for GS, is then necessary and sufficient for the presence of correlations. Second, we observe that for pure GS [reducible, up to local unitary operations, to the two-mode squeezed states characterized by and ], Eq. (3) yields a monotonic function of the negativity (see Fig. 1): is thus, as expected, an operational entanglement measure for pure two-mode GS. Third, we find that for mixed states majorizes entanglement. Given a mixed GS with negativity , it is straightforward to see that , [Eq. (3)], is always greater than , with being a pure two-mode squeezed state with the same negativity . Hence quantifies total correlations, and the difference (where the first term is due to total correlations and the second to quantum ones) can be naturally regarded as an operational measure of classical correlations noteclassic . We have evaluated on random two-mode GS as a function of their entanglement, conveniently scaled to , as shown in Fig. 1. Note that for any entanglement content there exist maximally correlated GS with , and also that separable mixed GS can achieve an arbitrary from 0 to 1, their correlations being only classical.
Non-Gaussian states.— Let us now turn our attention towards NGS, whose entanglement and correlations are, in general, encoded in higher moments as well. We focus on the most relevant NGS recently discussed in the literature and sometimes experimentally realized. Remarkably, we find for all of them a monotonic functional dependence between the entanglement (negativity) and the quadrature correlations .
Photonic Bell states. We consider Bell-like states of the form and (with ), which are nontrivial examples of superpositions of Fock states, entangled with respect to the (discrete) photon number. The negativity of these states is . For the four of them, Eq. (2) reads , showing a perfect agreement between the quadrature CV correlations and the entanglement.
Photon subtracted states. Most attention is being drawn by those CV states obtained from Gaussian states via subtracting photons Ourjoumtsev07 ; Furasawa : they perform better as resources for protocols like teleportation Kitagawa06 ; illu and allow for loophole-free tests of nonlocality bellref . Let us recall their preparation, following Kitagawa06 . The beam () of a two-mode squeezed state is let to interfere, via a beam splitter of transmittivity , with a vacuum mode (). The output is a four-mode Gaussian state of modes . Detection of one photon in each of the two beams and , conditionally projects the state of modes into a pure NGS, given in the Fock basis by , where , and . The limiting case , occurring with asymptotically vanishing probability, corresponds to an ideal two-photon substraction, . For any , the negativity is computable as . It increases with and with : only in the case it exceeds for any , diverging for . For all , the entanglement of eventually saturates, and above a squeezing threshold , the original GS is more entangled than the resulting non-Gaussian one. Following bellref , the explicit expression of the quadrature bit correlations Eq. (2) can be analytically obtained:
[TABLE]
where the ’s are defined above, , and is the Euler function. As depicted in Fig. 2, the behaviour of the entanglement is fully reproduced by which again is a monotonic function of : Eq. (2) is thus measuring truly quantum correlations of this important class of NGS as well.
Experimental de-Gaussified states.
Up to now we considered nearly-ideal non-Gaussian situations. We apply now our definition to the coherent photon-subtracted state recently studied and demonstrated by Ourjoumtsev et al. Ourjoumtsev07 . Referring to their paper for details on the state preparation, we just mention that each of the two beams of the two-mode squeezed state is let to interfere with the vacuum at a beam splitter with reflectivity , and by using an avalanche photodiode a single photon is subtracted from the state in a delocalized fashion. The realistic description of the obtained highly mixed state involves many parameters: we fix all to the values obtained from the theoretical calculations and/or experimental data of Ref. Ourjoumtsev07 , but for the reflectivity and the initial squeezing which are kept free. We then evaluate Eq. (2) as a function of for different values of . Unlike the previous cases, optimal correlations in the state occur between momentum operators (). Also for this realistic mixed case, the correlation measure reproduces precisely the behaviour of the negativity, as obtained in Ourjoumtsev07 after full Wigner tomography of the produced state . In particular, the negativity (and ) increases with the squeezing , and decreases with . Below a threshold squeezing which ranges around 3 dB, the NGS exhibits more entanglement (larger ) than the original two-mode squeezed state. Our results depicted in Fig. 3 compare extremely well to the experimental results (Fig. 6 of Ourjoumtsev07 ) where the negativity is plotted as a function of for different ’s.
Mixtures of Gaussian states etc. Recent papers VanLoock ; mista dealt with mixed NGS of the form , with . They have a positive Wigner function yet they are NGS (but for the trivial instances ). Clearly, the de-Gaussification here reduces entanglement and correlations in general. The negativity of such states reads and is increasing both with and with . The same dependency holds for the bit correlations, , which again is a monotonic function of for any . We further studied other NGS including photon-added and squeezed Bell-like states illu , and their mixtures with the vacuum: for all we found a direct match between entanglement and inprep .
Discussion.— By analyzing the maximal number of correlated bits () that can be extracted from a CV state via quadrature measurements, we have provided an operational quantification of the entanglement content of several relevant NGS (including the useful photon-subtracted states). Crucially, one can experimentally measure by direct homodyne detections (of the quadratures displaying optimal correlations only), in contrast to the much more demanding full tomographical state reconstruction. One can then easily invert the (analytic or numeric) monotonic relation between and the negativity to achieve a direct entanglement quantification from the measured data. Our analysis demonstrates the rather surprising feature that entanglement in the considered NGS can thus be detected and experimentally quantified with the same complexity as if dealing with GS.
Interestingly, this is not true for all CV states. By definition, quantifies correlations encoded in the second canonical moments only. We have realized that there exist also states [e.g. the photonic qutrit state ] which, though being totally uncorrelated up to the second moments (), are strongly entangled, with correlations embedded only in higher moments. The characterization of such states is an intriguing topic for further study inprep . In this respect, it is even more striking that the measure considered in this paper, based on (and accessible in terms of) second moments and homodyne detections only, provides such an exact quantification of entanglement in a broad class of pure and mixed NGS, whose quantum correlations are encoded nontrivially in higher moments too, and currently represent the preferred resources in CV quantum information. We focused on optical realizations of CV systems, but our framework equally applies to collective spin components of atomic ensembles polzik04 , and radial modes of trapped ions referee .
We thank Ph. Grangier, A. Ourjoumtsev, and M. Piani for discussions. We were supported by EU IP Programme SCALA, ESF PESC QUDEDIS, MEC (Spanish government) contract FIS2005-01369, CIRIT (Catalan government) contract SGR-00185, and Consolider-Ingenio 2010 CSD2006-0019 QOIT.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1(1) S. L. Braunstein and P. van Loock, Rev. Mod. Phys. 77 , 513 (2005); G. Adesso and F. Illuminati, J. Phys. A 40 , 7821 (2007).
- 2(2) M. M. Wolf et al. , Phys. Rev. Lett. 96 , 080502 (2006).
- 3(3) J. Eisert et al. , Phys. Rev. Lett. 89 , 137903 (2002); J. Fiurás̆ek, ibid. 89 , 137904 (2002); G. Giedke and J. I. Cirac, Phys. Rev. A. 66 , 032316 (2002).
- 4(4) N. C. Menicucci et al. , Phys. Rev. Lett. 97 , 110501 (2006).
- 5(5) See, e.g. , B. Groisman et al. , Phys. Rev. A 72 , 032317 (2005).
- 6(6) E. Shchukin and W. Vogel, Phys. Rev. Lett. 95 , 230502 (2005).
- 7(7) A. Ourjoumtsev et al. , Phys. Rev. Lett. 98 , 030502 (2007).
- 8(8) Unitary operations U ^ ^ 𝑈 \hat{U} at most quadratic in the creation / annihilation operators amount to symplectic transformations S : S Ω S T = Ω : 𝑆 𝑆 Ω superscript 𝑆 𝑇 Ω S:S\Omega S^{T}=\Omega in phase space, where the symplectic matrix Ω Ω \Omega encodes the canonical commutation relations. Then: ρ ^ ↦ U ^ ρ ^ U ^ † maps-to ^ 𝜌 ^ 𝑈 ^ 𝜌 superscript ^ 𝑈 † \hat{\rho}\mapsto\hat{U}\hat{\rho}\hat{U}^{\dagger} (on the density matrix) corresponds to W ( ξ ) ↦ W ( S − 1
