## Abstract

We demonstrated entanglement distribution between two remote quantum nodes located 3 meters apart. This distribution involves the asynchronous preparation of two pairs of atomic memories and the coherent mapping of stored atomic states into light fields in an effective state of near-maximum polarization entanglement. Entanglement is verified by way of the measured violation of a Bell inequality, and it can be used for communication protocols such as quantum cryptography. The demonstrated quantum nodes and channels can be used as segments of a quantum repeater, providing an essential tool for robust long-distance quantum communication.

In quantum information science (*1*), distribution of entanglement over quantum networks is a critical requirement for metrology (*2*), quantum computation (*3*, *4*), and communication (*3*, *5*). Quantum networks are composed of quantum nodes for processing and storing quantum states, and quantum channels that link the nodes. Substantial advances have been made with diverse systems toward the realization of such networks, including ions (*6*), single trapped atoms in free space (*7*, *8*) and in cavities (*9*), and atomic ensembles in the regime of continuous variables (*10*).

An approach of particular importance has been the seminal work of Duan, Lukin, Cirac, and Zoller (DLCZ) for the realization of quantum networks based on entanglement between single photons and collective excitations in atomic ensembles (*11*). Critical experimental capabilities have been achieved, beginning with the generation of nonclassical fields (*12*, *13*) with controlled waveforms (*14*) and extending to the creation and retrieval of single collective excitations (*15*–*17*) with high efficiency (*18*, *19*). Heralded entanglement with quantum memory, which is the cornerstone of networks with efficient scaling, was achieved between two ensembles (*20*). More recently, conditional control of the quantum states of a single ensemble (*21*–*23*) and of two distant ensembles (*24*) has also been implemented; the quantum states are likewise required for the scalability of quantum networks based on probabilistic protocols.

Our goal is to develop the physical resources that enable quantum repeaters (*5*), thereby allowing entanglement-based quantum communication tasks over quantum networks on distance scales much larger than those set by the attenuation length of optical fibers, including quantum cryptography (*25*). For this purpose, heralded number-state entanglement (*20*) between two remote atomic ensembles is not directly applicable. Instead, DLCZ proposed the use of pairs of ensembles (*U*_{i},*D*_{i}) at each quantum node *i*, with the sets of ensembles {*U*_{i}}, {*D*_{i}} separately linked in parallel chains across the network (*11*). Relative to the state of the art in our previous work (*20*), the DLCZ protocol requires the capability for the independent control of pairs of entangled ensembles between two nodes.

In our experiment, we created, addressed, and controlled pairs of atomic ensembles at each of two quantum nodes, thereby demonstrating entanglement distribution in a form suitable both for quantum network architectures and for entanglement-based quantum communication schemes (*26*). Specifically, two pairs of remote ensembles at two nodes were each prepared in an entangled state (*20*), in a heralded and asynchronous fashion (*24*), thanks to the conditional control of the quantum memories. After a signal indicating that the two chains are prepared in the desired state, the states of the ensembles were coherently transferred to propagating fields locally at the two nodes. The fields were arranged such that they effectively contained two photons, one at each node, whose polarizations were entangled. The entanglement between the two nodes was verified by the violation of a Bell inequality. The effective polarization-entangled state, created with favorable scaling behavior, was thereby compatible with entanglement-based quantum communication protocols (*11*).

The architecture for our experiment is shown in Fig. 1. Each quantum node, *L* (left) and *R* (right), consists of two atomic ensembles, *U* (up) and *D* (down), or four ensembles altogether, namely (*LU, LD*) and (*RU, RD*), respectively. We first prepared each pair in an entangled state, in which one excitation is shared coherently, by using a pair of coherent weak write pulses to induce spontaneous Raman transitions |*g* 〉→ |*e* 〉→ |*s* 〉 (bottom left, Fig. 1). The Raman fields (1_{LU}, 1_{RU}) from (*LU, RU*) were combined at the 50-50 beamsplitter *BS*_{U}, and the resulting fields were directed to single-photon detectors. A photoelectric detection event in either detector indicated that the two ensembles were prepared. The remote pair of *D* ensembles, (*LD, RD*), was prepared in an analogous fashion.

Conditioned upon the preparation of both ensemble pairs (*LU, LD*) and (*RU, RD*), a set of read pulses was triggered to map the stored atomic excitations into propagating Stokes fields in well-defined spatial modes through |*s* 〉→ |*e* 〉→ |*g* 〉 with the use of a collective enhancement (*11*) (bottom left, Fig. 1). This generated a set of four fields denoted by (2_{LU},2_{RU}) for ensembles (*LU, RU*) and by (2_{LD}, 2_{RD}) for ensembles (*LD, RD*). In the ideal case and neglecting higher-order terms, this mapping results in a quantum state for the Field 2 fields given by (1) Here, |*n* 〉_{x} is the *n*-photon state for mode *x*, where *x* ∈ {2_{LU},2_{RU},2_{LD},2_{RD}}, and η_{U} and η_{D} are the relative phases resulting from the writing and reading processes for the *U* and *D* pair of ensembles, respectively (*20*). The ± signs for the conditional states *U,D* result from the unitarity of the transformation by the beamsplitters (*BS*_{U}, *BS*_{D}). The extension of Eq. 1 to incorporate various nonidealities is given in the supporting online material (SOM) text.

Apart from an overall phase, the state jψ_{2LU,2RU,2LD,2RD} 〉 can be rewritten as follows: (2) where |*vac* 〉_{2i} denotes |0 〉_{2iU} |0 〉_{2iD}. If only coincidences between both nodes *L,R* are registered, the first two terms (i.e., with *e*^{–iηD}, *e ^{i}*

^{ηU}) do not contribute. Hence, as noted by DLCZ, excluding such cases leads to an effective density matrix equivalent to the one for a maximally entangled state of the form of the last term in Eq. 2. Notably, the absolute phases η

_{U}and η

_{D}do not need to be independently stabilized. Only the relative phase η = η

_{U}– η

_{D}must be kept constant, leading to 1/2 unit of entanglement for two quantum bits (i.e., 1/2 ebit).

The experimental demonstration of this architecture for implementing the DLCZ protocol relies critically on the ability to carry out efficient parallel preparation of the (*LU, RU*) and (*LD, RD*) ensemble pairs, as well as the ability to stabilize the relative phase η. The first requirement is achieved by the use of real-time control, as described in Felinto *et al*. (*24*) in a simpler case. As shown in Fig. 1, we implemented control logic that monitors the outputs of Field 1 detectors. A detection event at either pair triggers electro-optic intensity modulators (IM) that gate off all laser pulses traveling toward the corresponding pair of ensembles, thereby storing the associated state. Upon receipt of signals indicating that the two pairs of ensembles, (*LU, RU*) and (*LD, RD*), have both been independently prepared, the control logic triggers the retrieval of the stored states by simultaneously sending a strong read pulse into each of the four ensembles. Relative to the case in which no logic is implemented, this process resulted in a 19-fold enhancement in the probability of generating this overall state from the four ensembles (SOM text).

The second requirement—stability of the relative phase η—could be accomplished by active stabilization of each individual phase η_{U},η_{D}, as in (*20*). Instead of implementing this challenging technical task (which ultimately would have to be extended across longer chains of ensembles), our setup exploits the passive stability between two independent polarizations propagating in a single interferometer to prepare the two ensemble pairs (*27*). No active phase stabilization is thus required. In practice, we found that the passive stability of our system was sufficient for operation overnight without adjustment. Additionally, we implemented a procedure that deterministically sets the relative phase η to zero.

We also extended the original DLCZ protocol (Fig. 1) by combining fields (2_{LU},2_{LD}) and (2_{RU}, 2_{RD}) with orthogonal polarizations on polarizing beamsplitters PBS_{L} and PBS_{R} to yield fields 2_{L} and 2_{R}, respectively. The polarization encoding opens the possibility of performing additional entanglement purification and thus superior scalability (*28*, *29*). In the ideal case, the resulting state would now be effectively equivalent to a maximally entangled state for the polarization of two photons (3) where |*H* 〉 and |*V* 〉 stand for the state of a single photon with horizontal and vertical polarization, respectively. The sign of the superposition in Eq. 3 is inherited from Eq. 1 and is determined by the particular pair of heralding signals recorded by (*D*_{1a},*D*_{1b}) and (*D*_{1c},*D*_{1d}). The entanglement in the polarization basis is well suited for entanglement-based quantum cryptography (*11*, *25*), including security verification by way of the violation of a Bell inequality, as well as for quantum teleportation (*11*).

As a first step to investigate the joint states of the atomic ensembles, we recorded photoelectric counting events for the ensemble pairs (*LU,RU*) and (*LD,RD*) by setting the angles for the half-wave plates (λ/2)_{L,R} shown in Fig. 1 to 0°, such that photons reaching detectors *D*_{2b} and *D*_{2d} come only from the ensemble pair *U*, and photons reaching detectors *D*_{2a} and *D*_{2c} come only from the ensemble pair *D*. Conditioned upon detection events at *D*_{1a} or *D*_{1b} (or at *D*_{1c} or *D*_{1d}), we estimated the probability that each ensemble pair *U,D* contains only a single, shared excitation as compared with the probability for two excitations by way of the associated photoelectric statistics. In quantitative terms, we determined the ratio (*20*) (4) where *p*_{X,mn} is the probability to register *m* photodetection events in mode 2_{LX} and *n* events in mode 2_{RX} (*X* ={*U,D*}), conditioned on a detection event at *D*_{1}. A necessary condition for the two ensembles (*LX, RX*) to be entangled is that *h*_{X}^{(2)} <1, where *h*_{X}^{(2)} = 1 corresponds to the case of independent (unentangled) coherent states for the two fields (*20*). Figure 2 shows the measured *h*_{X}^{(2)} versus the duration τ_{M} (where M stands for memory) that the state is stored before retrieval. For both *U* and *D* pairs, *h*^{(2)} remains well below unity for storage times τ_{M} < ∼10 μs. For the *U* pair, the solid line in Fig. 2 provides a fit by the simple expression *h*^{(2)} = 1 – *A*exp[–(τ_{M}/τ)^{2}]. The fit gives *A* = 0.94 ± 0.01, where the error is SD, and τ = 22±2 μs, providing an estimate of a coherence time for our system. A principal cause for decoherence is an inhomogeneous broadening of the ground state levels by residual magnetic fields (*30*). The characterization of the time dependence of *h*^{(2)} constitutes an important benchmark of our system (SOM text).

We next measured the correlation function *E*(θ_{L},θ_{R}), defined by (5) Here, *C*_{jk} gives the rates of coincidences between detectors *D*_{2j} and *D*_{2k} for Field 2 fields, where *j,k* ∈ {*a,b,c,d*}, conditioned upon heralding events at detectors *D*_{1a},*D*_{1b} and *D*_{1c},*D*_{1d} from Field 1 fields. The angles of the two half-wave plates (λ/2)_{L} and (λ/2)_{R} are set at θ_{L}/2 and θ_{R}/2, respectively. As stated above, the capability to store the state heralded in one pair of ensembles and then to wait for the other pair to be prepared markedly improves the various coincidence rates *C*_{jk} by a factor that increases with the duration τ_{M} that a state can be preserved (*24*) (SOM text).

Figure 3 displays the correlation function *E* as a function of θ_{R}, for θ_{L} = 0° (Fig. 3A) and θ_{L} = 45° (Fig. 3B). Relative to Fig. 2, these data are taken with increased excitation probability (higher write power) to validate the phase stability of the system, which is evidently good. Moreover, these four-fold coincidence fringes in Fig. 3A provide further verification that predominantly one excitation is shared between a pair of ensembles. The analysis provided in the SOM text with the measured cumulative *h*^{(2)} parameter for this set of data, *h*^{(2)} = 0.12 ± 0.02, predicts a visibility of *V* = 78 ± 3% in good agreement with the experimentally determined *V* ≅ 75%. Finally, one of the fringes is inverted with respect to the other in Fig. 3B, which corresponds with the two possible signs in Eq. 3. As for θ_{L} = 45°, the measurement is sensitive to the square of the overlap ξ of photon wave-packets for fields 2_{U,D}; we may infer ξ_{U,D} ≅ 0.85 from the reduced fringe visibility (*V* ≅ 55%) in Fig. 3B relative to Fig. 3A, if all the reduction is attributed to a nonideal overlap. An independent experiment for two-photon interference in this setup has shown an overlap ξ ≅ 0.90, which confirms that the reduction can be principally attributed to the nonideal overlap. Other possible causes include imperfect phase alignment η ≠ 0 and imbalance of the effective-state coefficients (SOM text).

With the measurements from Figs. 2 and 3 in hand, we verified entanglement unambiguously by way of the violation of a Bell inequality (*31*). For this purpose, we chose the canonical values, θ_{L} = {0°,45°} and θ_{R} = {22.5°,–22.5°}, and constructed the Clauser-Horne-Shimony-Holt (CHSH) parameters (6) (7) for the two effective states |ψ_{2L,2R±} 〉_{eff} in Eq. 3. For local, realistic hidden-variable theories, *S*_{±} ≤ 2 (*31*). Figure 4 shows the CHSH parameters *S*_{±} as functions of the duration τ_{M} up to which one pair of ensembles holds the prepared state, in the excitation regime of Fig. 2. As shown in the SOM text, the requirements for minimization of higher-order terms are much more stringent in this experiment with four ensembles than with simpler configurations (*21*).

Figure 4, A and B, gives the results for our measurements of *S*_{±} with binned data. Each point corresponds to the violation obtained for states generated at τ_{M} ± Δτ_{M}/2 (Δτ_{M} is marked by the thick horizontal lines in Fig. 4). Strong violations are obtained for short memory times—for instance, *S*_{+} = 2.55 ± 0.14 > 2 and *S*_{–} = 2.61± 0.13 > 2 for the second bin—demonstrating the presence of entanglement between fields 2_{L} and 2_{R}. Therefore, these fields can be exploited to perform entanglement-based quantum communication protocols, such as quantum key distribution with, at minimum, security against individual attacks (*11*, *32*).

As can be seen in Fig. 4, the violation decreases with increasing τ_{M}. The decay is largely due to the time-varying behavior of *h*^{(2)} (Fig. 2 and SOM text). In addition to this decay, the *S*_{+} parameter exhibits modulation with τ_{M}. We explored different models for the time dependence of the CHSH parameters, but thus far have found no satisfactory agreement between model calculations and measurements. Nevertheless, the density matrix for the ensemble over the full memory time is potentially useful for tasks such as entanglement connection, as shown by Fig. 4, C and D, in which cumulative data are given. Each point at memory time τ_{M} gives the violation obtained by taking into account all the states generated from 0 to τ_{M}. Overall significant violations are obtained, namely *S*_{+} = 2.21 ± 0.04 > 2 and *S*_{–} = 2.24± 0.04 > 2 at τ_{M} ∼10 μ*s*.

In our experiment, we were able to generate excitation-number entangled states between remote locations, which are well suited for scaling purposes, and, with real-time control, we were able to operate them as if they were effectively polarization-entangled states, which can be applied to quantum communications such as quantum cryptography. Measurements of the suppression *h*^{(2)} of two-excitation components versus storage time explicitly demonstrates the major source that causes the extracted polarization entanglement to decay, emphasizing the critical role of multi-excitation events in the experiments aiming for a scalable quantum network. The present scheme, which constitutes a functional segment of a quantum repeater in terms of quantum state encoding and channel control, allows the distribution of entanglement between two quantum nodes. The extension of our work to longer chains involving many segments becomes more complicated and is out of reach for any current system. For long-distance communication, the first quantity to improve is the coherence time of the memory. Better cancellation of the residual magnetic fields and switching to new trap schemes should improve this parameter to ∼0.1 s by using an optical trap (*30*), thereby increasing the rate of preparing the ensembles in the state of Eq. 1 to ∼100 Hz. The second challenge that would immediately appear in an extended chain would be the increase of the multi-excitation probability with the connection stages. Recently, Jiang *et al*. (*28*) have theoretically demonstrated the prevention of such growth in a similar setup, but its full scalability still requires very high retrieval and detection efficiency, and photon-number resolving detectors. These two points clearly show that the quest of scalable quantum networks is still a theoretical and experimental challenge. The availability of our first functional segment opens the way for fruitful investigations.

**Supporting Online Material**

www.sciencemag.org/cgi/content/full/1140300/DC1

Materials and Methods

SOM Text

Figs. S1 to S3

Table S1

References