Spinor self-ordering of a quantum gas in a cavity

Ronen M. Kroeze,1, 2, ∗ Yudan Guo,1, 2, ∗ Varun D. Vaidya,1, 2, 3 Jonathan Keeling,4 and Benjamin L. Lev1, 2, 3 1Department of Physics, Stanford University, Stanford, CA 94305 2E. L. Ginzton Laboratory, Stanford University, Stanford, CA 94305 3Department of Applied Physics, Stanford University, Stanford, CA 94305 4SUPA, School of Physics and Astronomy, University of St Andrews, St Andrews KY16 9SS UK (Dated: July 16, 2018) We observe the joint -spatial (spinor) self-organization of a two-component BEC strongly coupled to an optical cavity. This unusual nonequilibrium Hepp-Lieb-Dicke is driven by an off-resonant two- Raman transition formed from a classical pump field and the emergent quantum dynamical cavity field. This mediates a spinor-spinor interaction that, above a critical strength, simultaneously organizes opposite spinor states of the BEC on opposite checkerboard configurations of an emergent 2D lattice. The resulting spinor density-wave polariton condensate is observed by directly detecting the atomic spin and momentum state and by holographically reconstructing the phase of the emitted cavity field. The latter provides a direct measure of the spin state, and a spin-spatial domain wall is observed. The photon-mediated spin interactions demonstrated here may be engineered to create dynamical gauge fields and quantum spin glasses.

The strong interaction between quantum or vice-versa) allowing the atoms to superradiantly scat- and provided by cavity quantum electrodynamics ter into the cavity mode. The emergent coherent (QED) provides unique opportunities for exploring quan- field further orders the spins in a self-reinforcing man- tum many-body physics away from equilibrium [1–3]. ner. Cavity dissipation stabilizes the driven, emergent Discovering and classifying the properties of nonequilib- spin order, and the phase of the cavity emission locks to rium quantum phase transitions is an active field [4–6], either 0 or π relative to the pump phase depending on with potential application to the engineering of quan- the symmetry-broken state. Superradiant cavity emis- tum devices, such as those with superconducting cor- sion of a spin-1 Dicke transition was observed with ther- relations [7, 8]. One particularly rich setting in which mal atoms coupled to a cavity [32, 33]. to explore such physics is provided by systems realiz- ing the driven-dissipative (Hepp-Lieb) Dicke model of Both pseudospin organization and superradiant emis- two atomic states strongly coupled to an optical cavity sion have been observed in an alternative form of the field [1, 3]. We present the observation of a nonequilib- nonequilibrium Dicke transition [34–36]. In this version, rium Dicke superradiant phase transition involving the a Bose-Einstein condensate (BEC) matter wave is cou- spontaneous ordering of coupled atomic spin and spatial pled to a cavity, where two different motional states play motion, as has been analyzed in Ref. [9]. The cavity pho- the role of up and down spin components. The atoms oc- tons mediate an effective position-dependent spin-spin cupy either the black or white checkerboard sites (spaced interaction; the resulting transverse Ising model that is λ-apart) of the emergent 2D lattice. The pseudospin realized opens future directions toward the study of ar- organization was detected by observing Bragg peaks at tificial quantum spin glasses and neural networks in a a momentum consistent with a checkerboard lattice to- driven-dissipative setting [10–19]. Moreover, with mi- gether with detection of the relative phase locking of the nor modification, this system could manifest dynamical pump and superradiant cavity emission [35]. The orga- gauge fields [20–25], resulting in topological superfluids nized state may be called a ‘density-wave polariton con- and exotic quantum Hall states. densate’ in recognition of the joint light–matter-wave na- ture of the quasiparticles in the macroscopically occupied As originally proposed [26], the nonequilibrium Dicke and coherent density-photon mode [37]. Roton instabil- model describes an Ising (Z2) symmetry-breaking tran- ities and the extended Bose-Hubbard model have been sition of a spin-1/2 system coupled to a single cavity realized [38–40], and similar systems employing a few de- mode. The phase transition of the nonequilibrium Dicke generate cavity modes have created a supersolid [41], an arXiv:1807.04915v1 [cond-mat.quant-gas] 13 Jul 2018 model is closer to a classical than a quantum transition, intertwined spatial order [42], and supermode-density- though distinct from both [3, 27–30]. Experimentally, the wave polariton condensates [37]. Highly degenerate cav- nonequilibrium Dicke model could be realized by freez- ities have been used to engineer tunable-range photon ing the spins in a 2D lattice of period λ/2, where λ is mediated atom-atom interactions [43] that may lead to the wavelength of both the pump and cavity fields [31]. liquid crystalline states [44]. A superradiant motional The spins are disordered below the transition threshold transition also occurs in cavities with spinless thermal and the cavity field is in a near-vacuum state. Above a atoms [45–47]. Self-organization of cold thermal gases pump threshold, the spins order in a λ-periodic checker- and arrays due to optical feedback from a single board pattern (either up/down on the black/white sites mirror have also been observed [48–52]. 2

(a) (b) What type of nonequilibrium phase transition arises when the pump and cavity fields couple atomic mo- tion and spin? Reference [9] describes such a system as a nonequilibrium spin-spatial Dicke superradiant phase transition in which atomic spins can flip while scatter- ing photons into the cavity, picking up recoil momen- tum in the process [53]. This creates a spin-decorated checkerboard lattice, whose state is a ‘spinor density- wave-polariton condensate.’ The spinor density wave is ˆ y,-g described by the superposition of spinor operators ψ↑,↓(r) described below, and arises due to a spinor-spinor inter- z,B ˆ† 0 ˆ† ˆ 0 ˆ action proportional to ψ↑(r )ψ↓(r)ψ↓(r )ψ↑(r). We note x that this scenario is distinct from an emergent texture of a two-component BEC recently observed in a miscible–

immiscible transition created by a state-dependent op-

(c) ± ± ± tical lattice arising from a nonequilibrium Dicke tran-

sition [54]. In this experiment, the cavity mediated a

0 ± ±

density-density interaction ρ (r)ρ (r ) between two ± +1 −1

Zeeman states m = ±1 of a BEC and the two-component

texture emerged above a critical ratio of the relative y ±

± ± scalar and vector polarizabilities of the light fields. x We now describe the experimental system before re- porting our observations of the superradiant spinor phase

(d) ky ky ± ± k k transition. Figure 1(a) shows the experimental con- ± r ± r figuration; see previous work for technical details of

the cavity and the intracavity BEC production appa-

k k ratus [43, 55]. We trap within the cavity a BEC x x 5 87 kr kr of 4.1(3) × 10 Rb atoms in the |F, mF i = |1, −1i state. The BEC is confined in a crossed optical dipole

trap (ODT) formed by a pair of 1064-nm laser beams

± ±

± ± propagating alongx ˆ andz ˆ, resp. Using ODT shap- Black sites White sites ing techniques [56], we create a trap with frequencies (ωx, ωy, ωz) = 2π × [58(1), 63(1), 47(1)] Hz that con- FIG. 1. (a) Experimental setup and detection techniques. tains a BEC with Thomas-Fermi radii (Rx,Ry,Rz) = The two Raman pump beams (red and blue), polarized along [10.3(1), 9.4(1), 12.8(2)] µm. These are all smaller than the cavity axis, are combined and retroreflected off the same the w0 = 35 µm waist of the TEM0,0 cavity mode [57]. mirror to create a phase-stable lattice (purple). The cav- To engineer the spinor Dicke Hamiltonian, we couple ity mode (green), imaged onto a EMCCD camera, interferes two internal states of 87Rb, |F, m i = |1, −1i ≡ |↓i and with a local oscillator at an angle (also green). This provides F the spatial heterodyne signal (blue lines) for the holographic |F, mF i = |2, −2i ≡ |↑i, through two cavity-assisted two- reconstruction of the cavity field amplitude and phase. Mo- photon Raman processes; see Fig. 1(b). A bias magnetic mentum of the BEC (scarlet) is absorption-imaged in time- field of ∼2.83 G is applied along +ˆz, the direction of of-flight (scarlet beam). (b) Double Raman scheme for cou- the quantization axis, resulting in an energy difference pling two 87Rb Zeeman states, see text. (c) Real-space car- ωHF ≈ 6.829 GHz between |↑i and |↓i due to hyper- toon of the transition from randomly positioned atoms below fine splitting and Zeeman shifts. The Raman processes threshold (left) to a checkerboard spinor order in an emergent are created by the cavity and transversely oriented pump 2D optical lattice above threshold (right). Atoms are in az ˆ (ˆx) spin-polarization state below (above) threshold. Dashed fields. The cavity field is that of the TEM0,0 mode at fre- (solid) lines in left panel are the nodes of the emergent cavity quency ωc with coupling strength g = g0Ξ(x, z), where g0 (pump) field. Solid lines in the right panel are the nodes of is the maximum single-atom coupling rate and Ξ(x, z) is the above-threshold 2D optical lattice. (d) Momentum-space the transverse mode profile. The pump beams have fre- cartoons of the transition for atoms on black (left) versus quency ω± such that ω+ = ω−+2(ωHF+δ), where δ is the white (right) sites. The state |±, bi (|∓, wi) emerges in black two-photon Raman detuning. Each pump field is far de- (white) checkerboard sites after sequential photon recoils from tuned from the atomic excited state by ∆ with coupling the pump and cavity fields. Arrow colors depict the optical ± transition pathway shown in panel (b). strengths Ω±. Their mean frequencyω ¯ = (ω+ + ω−)/2 is detuned by ∆c =ω ¯ − ωc from the cavity. The pump beams are retroreflected off the same mirror to create a phase-stable lattice. See Ref. [58] for a schematic of rela- 3 ) tive field frequencies, their generation and cavity spectra. (a) b c r ) This coupling realizes the interaction Hamiltonian be- -1 tween the two components of the spinor state ψˆ(r) = ˆ ˆ [ψ↑(r), ψ↓(r)]| given by

Z depth ( E lattice Total † (10 s rate Count Hint = dr 2ησˆx(r)(ˆa +a ˆ ) cos krx cos kry, (1) Time (ms) (b) (c) where the coupling strength η is equal for both Raman transitions,a ˆ is the annihilation for the intra- ˆ† ˆ ˆ† ˆ cavity field, andσ ˆx(r) = [ψ↑(r)ψ↓(r) + ψ↓(r)ψ↑(r)]/2. See Refs. [9, 58] for derivations and discussions of this y y k k model. Given the initial state |↓i, and within the sin- x r x r gle recoil scattering limit [59], the spinor components 0.5 0.0 0.5 ˆ ˆ OD OD take the form ψ↓(r) =c ˆ↓ψ0(r) and ψ↑(r) =c ˆ↑ψ1(r), † † with the total atom number N =c ˆ↑cˆ↑ +c ˆ↓cˆ↓. The FIG. 2. (a) Cavity emission detected by single photon coun- zero- and one-recoil wavefunctions equal ψ0 = 1 and ters versus time plotted with the concomitant linear increase ψ1(r) = 2 cos krx cos kry, with the recoil momentum in lattice depth (proportional to pump intensity). The su- perradiant transition threshold is at t ≈ 0.55 ms. (b,c) Spin- ~kr = 2π~/λ. The form of ψ1(r) is due to the 2D op- tical lattice emerging from the crossed pump and cavity sensitive absorption images of the atomic cloud in time-of- flight reveal the optical density (OD) of the momentum dis- standing-wave fields. tribution of both spin states at the times indicated in panel Performing the spatial integral and defining (a). (b) All atoms are in |↓i below threshold and either at zero ˆ † † pseudospin-1/2 operators as Jz = [ˆc↑cˆ↑ − cˆ↓cˆ↓]/2 momentum, or at k = (±2kr, 0) due to pump-lattice diffrac- ˆ † tion. (c) Above threshold, atoms have undergone a spin flip and J± =c ˆ↑↓cˆ↓↑, we arrive at the spinor Dicke-model Hamiltonian [58][60]: to |↑i accompanied by a k = {(±kr, ±kr); (±kr, ∓kr)} mo- mentum kick. The resulting Bragg peaks are spin-colored in the same pattern as in Fig. 1(d). † ˜ ηD † HD = −∆˜ caˆ aˆ+(2ωr −δ)Jˆz + √ (Jˆ+ +Jˆ−)(ˆa+ˆa ). (2) N

The Jˆ operate on the coupled pseudospin-1/2 spin-spatial creates a superposition of the atoms’ 2 degree of freedom. The recoil frequency is ωr = ~kr /2m, initial zero-momentum-|↓i state and the ±|↑i state cou- ˜ ˜ ∆c is ∆c minus the dispersive light shift,√δ = δ − ωs, pled to a momentum-recoil state comprised of the four where ωs is the ac Stark shift, and ηD = Nη/2. The superimposed k = {(±kr, ±kr); (±kr, ∓kr)} states [62]. first two terms account for the bare cavity energy and the We now present the observation of this organized energy shift between the spinor pseudospin states, resp. spinor state in momentum space. As in previous The organized system exhibits a nonzero order parame- work [32, 35, 37], superradiant cavity emission heralds R ter Θ ≡ dr cos krx cos kryσˆx(r)/N above a critical cou- the nonequilibrium Dicke phase transition; see Fig. 2(a). ˜ 1/2 pling strength ηD > ηth, where ηth = [∆˜ c(2ωr − δ)] /2 We first demonstrate of the model by lin- and Θ = ±1 in the Z2-symmetry-broken state. As shown early increasing the power in the Raman beams through in Fig. 1(c), the organized state is one of the |±, bi+|∓, wi the superradiant threshold with ∆c = −4 MHz and states of a spin-decorated λ-periodic checkerboard, where δ = −10 kHz; see Ref. [58] for Raman-coupling-strength |±i = |↓i ± |↑i are theσ ˆx eigenstates and |b/wi are calibration [63]. the black/white checkerboard sites. The Z2 broken- We then use spin-selective absorption imaging to de- symmetry is reflected in the choice between |+i or |−i tect the momentum distribution for each spin species in- residing on black sites. dependently during time-of-flight expansion of the gas. Though staggered, the spinor pseudospin state is ferro- This method records the momentum of both spin compo- magnetic. This can be seen by integrating out the cavity nents in a single realization of the experiment, allowing field and rewriting Eq. 1 as an Ising Hamiltonian [61]: for observation of the spinor state associated with the spin-spatial self-ordering [58]. The spin dependent time- X i j HIsing ∝ Jij cos krxi cos krxj cos kryi cos kryjσˆxσˆx. of-flight images are overlain in Figs. 2(b) and (c) [64]. Be- (3) low threshold, Fig. 2(b) shows only |↓i, zero-momentum The cosine terms can be incorporated into theσ ˆx through atoms (and Bragg peaks from the pump lattice), while a local gauge rotation. This results in a ferromagnetic, above threshold, Fig. 2(c) shows that spin-decorated infinite-range Jij coupling of the locally rotated spin op- Bragg peaks appear in a fashion expected from Fig. 1(d). i erators σ¯ˆx; see Ref. [58]. The absence of |↑i atoms at k = 0 and |↓i atoms at the Figure 1(d) presents the momentum-space cartoon of 1st-order momentum peaks indicates that spinor order the transition. Above threshold, coherent two-photon has emerged in the form of a λ-periodic checkerboard 4

w w0 w0 (a) 0

z z x x z

x 0.2 (b) OD

0.0 FIG. 3. Fringe amplitude factor χ as function of local oscil- y lator frequency detuning δLO. The cavity is pumped above OD threshold at a detuning ∆c = −4 MHz from the TEM0,0 cavity resonance. The EMCCD camera integration time is kr 2 ms. The insets show the spatial heterodyne signal—with x 0.2 local oscillator field subtracted for clarity—for both a maxi- mal χ and at δLO = 3 kHz where fringes average out. The FIG. 4. (a) The cavity field amplitude and phase, measured error bars represent one standard deviation of the mean over through holographic reconstruction, for a cavity locked near five repetitions. the TEM1,0 mode whose spatial profile Ξ(x, z) exhibits a sign- flip at x = 0. The phase of the right-hand lobe is defined as 0 with respect to the local oscillator. The phase shows a jump of exactly π across the cavity center, demonstrating the fixed rel- pattern in the |±i basis. ative phase difference between the Θ = ±1 states with respect Above threshold, the frequency of the superradiant to the local oscillator phase. (b) Observed spin-density struc- cavity emission should be locked atω ¯ [26]. Moreover, ture factor. The small-k transverse-mode-structure appears the phase of the emission should lock to either 0 or π as a node in the 1st-order Bragg peaks. The combination of (depending on the Z broken-symmetry) with respect to atomic and photonic observations indicates the existence of a 2 domain wall in the spinor. a local oscillator (LO) field at ωLO =ω ¯ + δLO. This field is coherently generated from one of the pump fields. To establish that both effects occur, we measure the phase of the cavity field emission in a spatially resolved fash- emission pattern of the TEM0,0 mode. ion using holographic reconstruction. Details of the fre- We now present a measurement of the relative phase quency stabilization and spatial heterodyne methods are locking of the cavity and pump fields. This is determined in Ref. [58]. Briefly, the LO is shone at an angle onto both by observing a π phase change of the superradiant the same EMCCD camera detecting the cavity emission, emission across an induced spinor domain wall and by ob- as depicted in Fig. 1(a). If the LO has the appropriate serving a nodal structural factor in the 1st-order atomic frequency (i.e., δLO = 0), the phase locking between the Bragg peaks caused by this domain wall. To create superradiant emission and the pump beam results in spa- adjacent spinor domains with opposite order parameter tial interference fringes on the camera, realizing a spatial Θ, the above experiment is repeated, but with the cav- heterodyne measurement. Spatial Fourier demodulation ity frequency tuned near the 1st-order transverse mode analysis of the fringes reveals both the spatial dependence TEM1,0;ω ¯ is set to ∆c = −1 MHz, see Ref. [58]. The of the cavity field phase and amplitude [58][65]. field profile Ξ(x, z)1,0 of this mode changes sign across The fringe amplitude factor χ, defined in Ref. [58], is the x = 0 nodal line in the x − z plane. The node ap- pears in the superradiant cavity emission amplitude and plotted in Fig. 3. A distinct peak appears at δLO = 0, as expected, while a significant averaging-out of fringe con- phase are shown in Fig. 4(a). The spinor order compen- trast is manifest for detunings larger than 1/T , where sates for this sign change in the cavity field by flipping the T = 2 ms is the EMCCD integration time, due to a non- Z2-symmetry-broken state from Θ = ±1 to ∓1 across the zero fringe phase velocity. This demonstrates a unique nodal line. That is, the spin-spatial checkerboard pattern feature of the spinor Dicke model: cavity emission is shifts by λ/2. The system does so to allow all the atoms detuned exactly halfway between the transverse pump to superradiantly emit into the cavity in phase, thereby beams, not at either or both of their frequencies. The minimizing the organization threshold. This effect has been discussed for purely spatial organization [37]. high contrast fringes at δLO = 0 shows that the phase is both stable and spatially constant over the superradiant Holographic reconstruction of the emitted cavity field 5 reveals the existence of this π phase shift on either side [6] L. M. Sieberer, S. D. Huber, E. Altman, and S. Diehl, of the nodal line; see Fig. 4(a). The line defect also ap- “Dynamical Critical Phenomena in Driven-Dissipative pears in the momentum distribution of the atoms shown Systems,” Phys. Rev. Lett. 110, 195301 (2013). in Fig. 4(b). A node in the 1st-order Bragg peaks appears [7] D. Fausti, R. I. Tobey, N. Dean, S. Kaiser, A. Dienst, M. C. Hoffmann, S. Pyon, T. Takayama, H. Takagi, due to the structure factor in the spinor organization [37]. and A. Cavalleri, “Light-Induced Superconductivity in Together with the phase flip of π, the nodal structure a Stripe-Ordered Cuprate,” Science 331, 189 (2011). factor implies a spinor domain wall along (0, z). In [8] M. Mitrano, A. Cantaluppi, D. Nicoletti, S. Kaiser, degenerate-mode cavities, such as the adjustable-length A. Perucchi, S. Lupi, P. Di Pietro, D. Pontiroli, M. Ricc`o, near-confocal cavity system of Refs. [37, 55], interference S. R. Clark, D. Jaksch, and A. Cavalleri, “Possible light- among modes could lead to topological spin-defect tex- induced superconductivity in K3C60 at High Tempera- tures and local spin-spin interactions [10, 43]. ture,” Nature 530, 461 (2016). [9] F. Mivehvar, F. Piazza, and H. Ritsch, “Disorder-Driven We have observed a spinor nonequilibrium Dicke su- Density and Spin Self-Ordering of a Bose-Einstein Con- perradiant phase transition among spinful atoms in a densate in a Cavity,” Phys. Rev. Lett. 119, 063602 BEC coupled to a cavity. Moreover, the intracavity pho- (2017). tons mediate a spin-spin Ising interaction. This leads [10] S. Gopalakrishnan, B. L. Lev, and P. M. Goldbart, “Frustration and Glassiness in Spin Models with Cavity- to a phase transition into a ferromagnetic state at a Mediated Interactions,” Phys. Rev. Lett. 107, 277201 critical transverse field value associated with the two- (2011). photon pump intensity. By observing both the photonic [11] P. Strack and S. Sachdev, “Dicke Quantum Spin Glass and atomic manifestations of the polaritonic system, we of Atoms and Photons,” Phys. Rev. Lett. 107, 277202 demonstrated joint spin-spatial self-organization. Using (2011). a higher-order transverse mode of the cavity and holo- [12] S. Gopalakrishnan, B. L. Lev, and P. M. Goldbart, “Ex- graphic reconstruction, we demonstrated the ability to ploring models of associative memory via cavity quantum electrodynamics,” Philos. Mag. 92, 353 (2012). create and image signatures of a domain wall. Strong [13] M. Buchhold, P. Strack, S. Sachdev, and S. Diehl, Ising-type interactions, as realized here, may enable the “Dicke-model quantum spin and photon glass in optical study of quantum spin glass physics [10, 11, 13], which cavities: Nonequilibrium theory and experimental signa- in turn may lead to quantum dissipative neuromorphic tures,” Phys. Rev. A 87, 063622 (2013). computing devices [12, 14–19]. Lastly, a simple reconfig- [14] P. L. McMahon, A. Marandi, Y. Haribara, R. Hamerly, uration of the pump fields will enable the generation of C. Langrock, S. Tamate, T. Inagaki, H. Takesue, dynamical spin-orbit coupling and gauge fields [20–25]. S. Utsunomiya, K. Aihara, R. L. Byer, M. M. Fejer, H. Mabuchi, and Y. Yamamoto, “A fully-programmable We thank S. Goldman and K.-Y. Lin for experimen- 100-spin coherent Ising machine with all-to-all connec- tal assistance and acknowledge funding support from the tions,” Science 354, 614 (2016). Army Research Office, the National Science Foundation [15] T. Inagaki, Y. Haribara, K. Igarashi, T. Sonobe, S. Ta- under Grant No. CCF-1640075, and by the Semicon- mate, T. Honjo, A. Marandi, P. L. McMahon, T. Umeki, K. Enbutsu, O. Tadanaga, H. Takenouchi, K. Aihara, ductor Research Corporation under Grant No. 2016-EP- K.-i. Kawarabayashi, K. Inoue, S. Utsunomiya, and 2693-C. J. K. acknowledges support from SU2P. H. Takesue, “A coherent Ising machine for 2000-node op- timization problems,” Science 354, 603 (2016). [16] V. Torggler, S. Kr¨amer, and H. Ritsch, “Quantum an- nealing with ultracold atoms in a multimode optical res- onator,” Phys. Rev. A 95, 032310 (2017). ∗ R.K. and Y.G. contributed equally to this work. [17] P. Rotondo, M. Marcuzzi, J. P. Garrahan, I. Lesanovsky, [1] H. Ritsch, P. Domokos, F. Brennecke, and T. Esslinger, and M. M¨uller,“Open quantum generalisation of Hop- “Cold atoms in cavity-generated dynamical optical po- field neural networks,” J. Phys. A: Math. Theor. 51, tentials,” Rev. Mod. Phys. 85, 553 (2013). 115301 (2018). [2] L. M. Sieberer, M. Buchhold, and S. Diehl, “Keldysh [18] E. Fiorelli, P. Rotondo, M. Marcuzzi, J. P. Garrahan, field theory for driven open quantum systems,” Rep. and I. Lesanovsky, “Quantum accelerated approach to Prog. Phys. 79, 096001 (2016). the thermal state of classical spin systems with applica- [3] P. Kirton, M. M. Roses, J. Keeling, and E. G. D. tions to pattern-retrieval in the Hopfield neural network,” Torre, “Introduction to the Dicke model: from equi- (2018), arXiv:1806.02747. librium to nonequilibrium, and vice versa,” (2018), [19] V. Torggler, P. Aumann, H. Ritsch, and W. Lechner, “A arXiv:1805.09828. Quantum N-Queens Solver,” (2018), arXiv:1803.00735. [4] S. Diehl, A. Tomadin, A. Micheli, R. Fazio, and P. Zoller, [20] Y. Deng, J. Cheng, H. Jing, and S. Yi, “Bose- “Dynamical Phase Transitions and Instabilities in Open Einstein Condensates with Cavity-Mediated Spin-Orbit Atomic Many-Body Systems,” Phys. Rev. Lett. 105, Coupling,” Phys. Rev. Lett. 112, 143007 (2014). 015702 (2010). [21] L. Dong, L. Zhou, B. Wu, B. Ramachandhran, and [5] A. Polkovnikov, K. Sengupta, A. Silva, and M. Vengalat- H. Pu, “Cavity-assisted dynamical spin-orbit coupling in tore, “Colloquium: Nonequilibrium dynamics of closed cold atoms,” Phys. Rev. A 89, 011602 (2014). interacting quantum systems,” Rev. Mod. Phys. 83, 863 [22] B. Padhi and S. Ghosh, “Spin-orbit-coupled Bose- (2011). Einstein condensates in a cavity: Route to magnetic 6

phases through cavity transmission,” Phys. Rev. A 90, [40] R. Landig, L. Hruby, N. Dogra, M. Landini, R. Mottl, 023627 (2014). T. Donner, and T. Esslinger, “Quantum phases from [23] F. Mivehvar and D. L. Feder, “Synthetic spin-orbit inter- competing short- and long-range interactions in an opti- actions and magnetic fields in ring-cavity QED,” Phys. cal lattice,” Nature 532, 476 (2016). Rev. A 89, 013803 (2014). [41] J. L´eonard,A. Morales, P. Zupancic, T. Esslinger, and [24] C. Kollath, A. Sheikhan, S. Wolff, and F. Brennecke, T. Donner, “Supersolid formation in a quantum gas “Ultracold Fermions in a Cavity-Induced Artificial Mag- breaking a continuous translational symmetry,” Nature netic Field,” Phys. Rev. Lett. 116, 060401 (2016). 543, 87 (2017). [25] W. Zheng and N. R. Cooper, “Superradiance Induced [42] A. Morales, P. Zupancic, J. L´eonard, T. Esslinger, and Particle Flow via Dynamical Gauge Coupling,” Phys. T. Donner, “Coupling two order parameters in a quan- Rev. Lett. 117, 175302 (2016). tum gas,” Nat. Mater. (2018), 10.1038/s41563-018-0118- [26] F. Dimer, B. Estienne, A. S. Parkins, and H. J. 1. Carmichael, “Proposed realization of the Dicke-model [43] V. D. Vaidya, Y. Guo, R. M. Kroeze, K. E. Ballan- quantum phase transition in an optical cavity QED sys- tine, A. J. Koll´ar,J. Keeling, and B. L. Lev, “Tunable- tem,” Phys. Rev. A 75, 013804 (2007). Range, Photon-Mediated Atomic Interactions in Multi- [27] D. Nagy, G. Szirmai, and P. Domokos, “Critical expo- mode Cavity QED,” Phys. Rev. X 8, 011002 (2018). nent of a quantum-noise-driven phase transition: The [44] S. Gopalakrishnan, B. L. Lev, and P. M. Gold- open-system Dicke model,” Phys. Rev. A 84, 043637 bart, “Emergent crystallinity and frustration with Bose- (2011). Einstein condensates in multimode cavities,” Nat. Phys. [28] E. G. D. Torre, S. Diehl, M. D. Lukin, S. Sachdev, and 5, 845 (2009). P. Strack, “Keldysh approach for nonequilibrium Phase [45] A. T. Black, H. W. Chan, and V. Vuleti¸c,“Observa- Transitions in : Beyond the Dicke model tion of Collective Friction Forces due to Spatial Self- in optical cavities,” Phys. Rev. A 87, 023831 (2013). Organization of Atoms: From Rayleigh to Bragg Scat- [29] F. Piazza, P. Strack, and W. Zwerger, “Bose–Einstein tering,” Phys. Rev. Lett. 91, 203001 (2003). condensation versus Dicke–Hepp–Lieb transition in an [46] P. Domokos and H. Ritsch, “Collective Cooling and Self- optical cavity,” Ann. Phys. 339, 135 (2013). Organization of Atoms in a Cavity,” Phys. Rev. Lett. 89, [30] S. Sch¨utz,S. B. J¨ager,and G. Morigi, “Thermodynamics 253003 (2002). and dynamics of atomic self-organization in an optical [47] K. J. Arnold, M. P. Baden, and M. D. Barrett, “Self- cavity,” Phys. Rev. A 92, 063808 (2015). Organization Threshold Scaling for Thermal Atoms Cou- [31] Note that the difference between the cavity and pump pled to a Cavity,” Phys. Rev. Lett. 109, 153002 (2012). frequencies is much smaller than their mean, and so we [48] M. Nixon, E. Ronen, A. A. Friesem, and N. David- take λ as the wavelength for both. son, “Observing Geometric Frustration with Thousands [32] Z. Zhiqiang, C. H. Lee, R. Kumar, K. J. Arnold, S. J. of Coupled ,” Phys. Rev. Lett. 110, 184102 (2013). Masson, A. S. Parkins, and M. D. Barrett, “Nonequilib- [49] V. Pal, C. Tradonsky, R. Chriki, A. A. Friesem, and rium phase transition in a spin-1 Dicke model,” Optica N. Davidson, “Observing Dissipative Topological Defects 4, 424 (2017). with Coupled Lasers,” Phys. Rev. Lett. 119, 013902 [33] Z. Zhang, C. H. Lee, R. Kumar, K. J. Arnold, S. J. Mas- (2017). son, A. L. Grimsmo, A. S. Parkins, and M. D. Barrett, [50] G. Labeyrie, E. Tesio, P. M. Gomes, G. L. Oppo, W. J. “Dicke-model simulation via cavity-assisted Raman tran- Firth, G. R. M. Robb, A. S. Arnold, R. Kaiser, and sitions,” Phys. Rev. A 97, 043858 (2018). T. Ackemann, “Optomechanical self-structuring in a cold [34] D. Nagy, G. K´onya, G. Szirmai, and P. Domokos, “Dicke- atomic gas,” Nat. Photon. 8, 321 (2014). Model Phase Transition in the Quantum Motion of a [51] G. R. M. Robb, E. Tesio, G. L. Oppo, W. J. Firth, Bose-Einstein Condensate in an Optical Cavity,” Phys. T. Ackemann, and R. Bonifacio, “Quantum Threshold Rev. Lett. 104, 130401 (2010). for Optomechanical Self-Structuring in a Bose-Einstein [35] K. Baumann, C. Guerlin, F. Brennecke, and Condensate,” Phys. Rev. Lett. 114, 173903 (2015). T. Esslinger, “Dicke quantum phase transition with a [52] G. Labeyrie, I. Kresic, G. R. M. Robb, G.-L. Oppo, superfluid gas in an optical cavity,” Nature 464, 1301 R. Kaiser, and T. Ackemann, “Magnetic Phase Diagram (2010). of Light-mediated Spin Structuring in Cold Atoms,” [36] H. Keßler, J. Klinder, M. Wolke, and A. Hem- (2018), arXiv:1806.09966. merich, “Steering Matter Wave Superradiance with an [53] The order-by-disorder aspect of the transition proposed Ultranarrow-Band Optical Cavity,” Phys. Rev. Lett. in Ref. [9] is not directly relevant to our system due to 113, 070404 (2014). the initial spin polarization. [37] A. J. Koll´ar,A. T. Papageorge, V. D. Vaidya, Y. Guo, [54] M. Landini, N. Dogra, K. Kroeger, L. Hruby, T. Don- J. Keeling, and B. L. Lev, “Supermode-density-wave- ner, and T. Esslinger, “Formation of a Spin Texture in polariton condensation with a Bose-Einstein condensate a Quantum Gas Coupled to a Cavity,” Phys. Rev. Lett. in a multimode cavity,” Nat. Commun. 8, 14386 (2017). 120, 223602 (2018). [38] R. Mottl, F. Brennecke, K. Baumann, R. Landig, T. Don- [55] A. J. Koll´ar,A. T. Papageorge, K. Baumann, M. A. Ar- ner, and T. Esslinger, “Roton-Type Mode Softening in a men, and B. L. Lev, “An adjustable-length cavity and Quantum Gas with Cavity-Mediated Long-Range Inter- Bose-Einstein condensate apparatus for multimode cav- actions,” Science 336, 1570 (2012). ity QED,” New J. Phys. 17, 43012 (2015). [39] J. Klinder, H. Keßler, M. R. Bakhtiari, M. Thorwart, [56] T. A. Bell, J. A. P. Glidden, L. Humbert, and I. Bloch, and A. Hemmerich, “Observation of a Superradiant Mott “Experimental demonstration of painting arbitrary and Insulator in the Dicke-Hubbard Model,” Phys. Rev. Lett. dynamic potentials for Bose-Einstein condensates,” New 115, 230403 (2015). J. Phys. 11, 043030 (2009). 7

[57] A. E. Siegman, Lasers (University Science Books, 1986). [58] See Supplemental Material for information on experimen- tal details, theory, and data analysis. [59] This limit is valid under the condition of minimal deple- tion of the zero-momentum condensate; Fig. 2(c) indeed shows that the k = 0 peak is much more populous than the scattered momentum peaks. [60] We note that all parameters in this version of the Dicke model are tunable, providing access to a wealth of dynamical phenomena such as chaos and limit cy- cles [3, 33, 66–70]. [61] Writing Eq. 1 as an Ising Hamiltonian is possible because, unlike in Ref. [32], we operate in the large-∆c dispersive regime. [62] Note, unlike in Refs. [20–23], our Raman scheme does not produce (dynamical) spin-orbit-coupling because this excited motional state has zero net momentum. [63] This long-lived superradiance we observe is distinct from the single-beam spin-flip situation [33]. [64] Spherical scattering halos are due to two-body contact interactions. [65] See also Ref. [71] for another recent use of this technique. [66] C. Emary and T. Brandes, “Chaos and the quantum phase transition in the Dicke model,” Phys. Rev. E 67, 066203 (2003). [67] M. J. Bhaseen, J. Mayoh, B. D. Simons, and J. Keeling, “Dynamics of nonequilibrium Dicke models,” Phys. Rev. A 85, 013817 (2012). [68] A. Altland and F. Haake, “ and effective thermalization,” Phys. Rev. Lett. 108, 073601 (2012). [69] A. Altland and F. Haake, “Equilibration and macroscopic quantum fluctuations in the Dicke model,” New J. Phys. 14, 073011 (2012). [70] F. Piazza and H. Ritsch, “Self-Ordered Limit Cycles, Chaos, and Phase Slippage with a Superfluid inside an Optical Resonator,” Phys. Rev. Lett. 115, 163601 (2015). [71] N. Schine, M. Chalupnik, T. Can, A. Gromov, and J. Simon, “Measuring Electromagnetic and Gravita- tional Responses of Photonic Landau Levels,” (2018), arXiv:1802.04418. [72] O. Morsch and M. Oberthaler, “Dynamics of Bose- Einstein condensates in optical lattices,” Rev. Mod. Phys. 78, 179 (2006). 8

SUPPLEMENTAL MATERIAL: seed 1 is used to stabilize the science cavity using the SPINOR SELF-ORDERING OF A QUANTUM Pound-Drever-Hall (PDH) technique. The cavity res- GAS IN A CAVITY 2 onance frequency ωc is detuned from the 5 S1/2|2, −2i 2 to 5 P3/2 transition by 154 GHz. The resulting Raman Cavity spectrum pump beams have atomic detunings of 160 and 147 GHz, resp. Using a fiber electro-optic modulator (EOM), side- The length of our cavity can be adjusted in situ us- bands at ωHF are added onto the ω− Raman beam in a ing a slip-stick piezo [55]. The length in this work is set separate path to derive the cavity probe (for use in tak- such that the TEMl,m modes within l +m = const. fami- ing the data in Fig. 5) and local oscillator beam. The lies [57] are resolvable but far separated in frequency from drive signal to the EOM is split off from the same source other mode families. Figure 5 shows the cavity spectra that locks the seed lasers. We further isolate the correct for the two experiments discussed in this paper. For ex- periments using the TEM0,0 mode, the cavity detuning is ∆c = −4.00 MHz, while ∆˜ c = −2.39 MHz due to the dispersive shift (see the section on derivation of cavity- mediated spin-spin interaction below). Similarly, for the TEM1,0 mode, ∆c = −0.96 MHz and ∆˜ c = −0.79 MHz. That is, the detuning is blue of the TEM0,1 mode, though red of the TEM1,0 mode. We observe dominant coupling to the TEM1,0 mode and no instability from proximity to the blue of the TEM1,0 mode. The splitting of approxi- mately 50 MHz between adjacent families of modes is at AOM AOM AOM least an order of magnitude larger than these detunings.

Frequency content Filter Cavity

The frequency content of the laser beams is schemati- cally summarized in Fig. 6. Both 780-nm Raman beams EOM are derived from frequency-doubled 1560-nm light. The relative frequency between the two 1560-nm seed lasers PPLN PPLN are stabilized with respect to a stable frequency source SHG SHG calibrated via microwave spectroscopy to oscillate at the Fiber Fiber frequency difference ωHF between |1, −1i and |2, −2i. Amplifier Amplifier ωHF includes the Zeeman shift associated with the ap- 1560 nm plied magnetic field. This frequency difference is con- PD 1560 nm trolled using a proportional-integral loop filter with feed- Seed 1 Seed 2 back applied on seed 2. Additional 1560-nm light from PI

FIG. 6. Schematic for the laser system used in this experi- ment. Green lines represent electrical signals. The two 780- nm Raman beams are derived using second harmonic gener- ation (SHG) from two 1560-nm fiber lasers, whose relative frequency is stabilized at ωHF with a beat-note lock referenc- ing seed 2 to seed 1. After SHG, the frequencies of the two

Cavity transmission (arb.) transmission Cavity doubled light beams are separated by 2ωHF. AOMs placed in the path of the beams allow for additional frequency adjust- Detuning (MHz) ments and intensity control. The science cavity is stabilized at ωc using 1560-nm light from seed 1 through the Pound- FIG. 5. Transmission spectra of the cavity in use, at the Drever-Hall (PDH) technique. The same rf source used to l + m = 0 family and the l + m = 1 family. The lock lock the fiber lasers is used to drive an electro-optic modula- points are indicated by dashed lines, giving ∆c = −4 MHz tor (EOM) for the purpose of generating the local oscillator and −0.96 MHZ for the experiments involving TEM0,0 and beam at ωLO. The correct sideband is isolated by a filter TEM1,0 respectively. cavity. 9 sideband from the EOM output using a filter cavity; the both amplitude and phase fluctuations in space. In the resulting beam is at the mean frequencyω ¯ of the two Ra- most general case, this field may be expressed as Ec(r) = iφc(r) man beams and phase stable with respect to the cavity. |Ec(r)|e . The amplitude and phase of this field is Additional acousto-optic modulators (AOMs) provide in- measured using a holographic technique; see Ref. [71] for tensity stabilization and additional frequency-shifting ca- another recent demonstration of this technique. A large pabilities to symmetrically adjust the Raman detuning δ. local oscillator (LO) beam at frequencyω ¯ + δLO is in- All rf signals used in the experiment are stabilized with cident on the EMCCD camera with a wavevector ∆k respect to the same 10-MHz Rb clock. relative to the cavity emission. This LO beam is derived from the output of the filter cavity in Fig. 6 and the AOM provides a controllable frequency shift δLO. The interfer- Holographic reconstruction ence between the cavity emission Ec(r) and the LO field ELO(r) produces an image with an intensity Ih(r) on the Above threshold, the superradiant cavity emission at EMCCD camera; see Fig. 7(a). This may be expressed frequencyω ¯ observed on our EMCCD camera can have as

2 2 Ih(r) = |Ec(r)| + |ELO(r)| + 2χ(δLO)|Ec(r)ELO(r)| cos (∆k · r + ∆φ(r)) , (4)

where ∆φ(r) = φc(r) − φLO(r) is the phase difference wavevector ∆k. between the cavity and LO wavefronts. Both |Ec(r)| Finally, the phase of the cavity field may be extracted and φ (r) are inferred from the amplitude and phase of c from ∆φ(r) by correcting for phase variations φLO(r) of the fringes produced by the oscillatory term of Eq. 4. the local oscillator wavefront. The TEM0,0 mode of the Reduction of fringe contrast is characterized by the fac- cavity is used to calibrate these variations since it has tor χ(δLO). Several factors contribute to this reduction. a uniform phase over its transverse profile. Measuring For example, mismatch in the spatial and polarization- φLO(r) in this manner allows us to calculate the phase of mode overlap of the cavity and LO reduces contrast. The the cavity wavefront as φc = ∆φ+φLO and consequently contrast can also appear smaller due to a frequency dif- ference between the LO and cavity emission: the fringe signals spatially average during the EMCCD camera’s 2- ms integration time because the fringes have a non-zero phase velocity. This spatial averaging effect allows us to determine the cavity emission via measuring fringe con- trast versus δLO, as shown in Fig. 3. Noise in the relative z frequency between the cavity emission and LO also leads to spatial averaging. x In order to accurately extract |Ec(r)| and φc(r), the image must first be corrected to account for inten- sity and phase variations of the LO beam. An inde- pendent measurement of the local oscillator intensity 2 ILO(r) = |ELO(r)| allows us to create a corrected field image Ecorr(r) whose fringe amplitude solely depends on 2 |Ec(r)|: 0 Normalized |E| 1 I (r) − I (r) E (r) = h LO FIG. 7. Holographic reconstruction of cavity fields. (a) Cam- corr p era image I (r) generated by the interference between cavity ILO(r) h emission from a TEM0,0 mode and the local oscillator beam. |E (r)|2 c (b) The corrected image Ecorr(r) generated from (a) using = + 2χ(δLO)|Ec(r)| cos (∆k · r + ∆φ(r)) . 2 |ELO(r)| the procedure described in Eq. 5. (c) The intensity |Ec(r)| (5) of the cavity emission extracted from (b). (d) The phase of the cavity emission extracted from (b). (e) A visualization See Fig. 7(b) for plot of Ecorr(r). Assuming the cavity of the complex electric field constructed using the amplitude field varies slowly over the fringe wavelength 2π/|∆k|, from (c) and phase from (d). (f) Color legend for panels (c) and (e). Color wheel is for panel (e) while color bar is for we may extract |Ec(r)|, shown in Fig. 7(c), and ∆φ(r), panel (c). shown in Fig. 7(d), by demodulating Ecorr at the fringe 10 visualize the complex electric field of higher-order modes states. The Clebsch-Gordon coefficients c(F, m → as shown in Fig. 4(b). F 0, m + q) are the relative strengths of the transitions. We apply a bias magnetic field alongz ˆ. Both pump beams are linearly polarized√ along the cavity axis. The Spin-selective imaging additional factor of 1/ 2 for the Rabi frequency of the two pump beams Ω+ and Ω− comes from the fact that At the end of the experimental sequence, atoms can be the beams couple to both σ+ and σ− transitions, though in either |↓i, |↑i, or a superposition of the two. To selec- only one is close to resonant for each beam due to Zeeman tively detect these states, we perform absorption imag- shifts. 2 ing on the cycling transition between 5 S1/2|2, −2i and The spatial profile of mode Ξ results in a spatially 2 5 P3/2|3, −3i. Only the atoms in |↑i are imaged due to dependent single-photon Rabi frequency g0Ξ(r)/Ξ0,0(0), the absence of repumping light. We verified that there is where Ξ0,0 is the profile of a TEM0,0 mode. Given the negligible depumping with circularly polarized light that large detunings of the pumps from the atomic excited drives purely σ− transitions. Following this initial imag- states compared to the excited-state hyperfine splittings, ing pulse, an intense pulse of light resonant with the all the excited states are assumed to be at the same en- transition is applied, which results in the expulsion of ergy ωa. In the ground states, the Zeeman shift pushes the |↑i population from view. Next, the atoms in |↓i are |F = 2, mF = 0i out of resonance, so we only con- 2 transferred to 5 S1/2|2, −2i using microwave adiabatic sider the spin components |F, mF i = |1, −1i ≡ |↓i and rapid passage and imaged using the same cycling tran- |F, mF i = |2, −2i ≡ |↑i of the atom’s hyperfine states sition. These atoms are subsequently also removed from as the coupled two-level system. All energy levels are the field of view, after which ‘bright’ and ‘dark’ images defined with respect to the energy of |↓i, and the bare are taken for completing the absorption imaging process. energy splitting ωHF (hyperfine splitting plus additional The extracted optical densities from the first and second Zeeman shift) between |↑i and |↓i is set by the bias imaging pulse are then overlaid to produce spin-full ab- magnetic field alongz ˆ of ∼ 2.82 G. We use microwave sorption images such as those presented in Figs. 2(b), (c) spectroscopy to calibrate the field and estimate a field and 4(b). fluctuation-induced frequency noise of 2.4 kHz on ωHF. To obtain the effective Hamiltonian, we transform Eq. 6 into a rotating frame defined by the unitary trans- Derivation of cavity mediated spin-spin interaction formation Uˆ = exp(−iHˆtt), where Z The Hamiltonian of a single cavity mode a with spatial ˆ 1 † 1 3 ˆ† ˆ Ht = (ω+ + ω−)ˆa aˆ + (ω+ − ω−) d r ψ↑(r)ψ↑(r). profile Ξ(r) interacting with atoms can be written as in 2 2 Ref. [33]: (9) Here, the coupled spin-spatial atomic states are repre- † ˆ ˆ ˆ | H = ωcaˆ aˆ + Hatom + Htrap + Hkinetic + Hint, (6) sented by the spinor ψ(r) = [ψ↑(r), ψ↓(r)] . Before writ- ing the resulting Hamiltonian, we define the detunings where ωc is the optical frequency of the cavity mode, ∆+ and ∆− from the atomic excited state for each of the Hatom is the energy of the atomic internal states, and Raman transitions, the detuning ∆c of the mean pump Htrap and Hkinetic capture the potential and kinetic en- frequency from the cavity frequency ωc, and the two- ergy of atoms in different internal states. Hint describes photon detuning of the cavity-assisted Raman transition the coupling introduced by the pump beams (with optical resonance δ as: frequency ω+ and ω−) and cavity: ∆+ = ω+ − ωa Z 1   3 −iω+t −iω−t Hint = d r√ Ω+(r)e + Ω−(r)e ∆− = ω− + ωHF − ωa 2   1 X ˆ+1 ˆ−1 ω¯ = (ω+ + ω−) × AFF 0 (r) − AFF 0 (r) 2 FF 0 1 Z ∆c = (ω+ + ω−) − ωc 3 X 0 2 + d g Ξ(r)ˆa Aˆ 0 (r) + H.c., (7) 0 FF 1 FF 0 δ = (ω − ω ) − ω . (10) 2 + − HF where We set δ ≈ −10 kHz, while the detuning for other al- ˆ(q) X 0 ˆ† ˆ lowed Raman processes, e.g., the coupling between |↓i AFF 0 (r) = c(F, m → F , m + q)ψF 0,m+q(r)ψF,m(r) m and |F = 2, mF = 0i, is on the order of a few MHz due (8) to Zeeman splitting. After adiabatically eliminating the is the atomic raising operators connecting different hy- atomic excited states and ignoring the s-wave interaction ˆ ˆ perfine levels of the ground ψF,m and excited ψF 0,m+q and external harmonic trapping potential, the resulting 11

Hamiltonian H = H↑ + H↓ + Hcavity + HRaman in the rotating frame is given by

Z 2  Ω2 Ω2  2 3 ˆ† h pˆ + − 2 [g0Ξ(x, z)] 2 † i ˆ H↑ = d rψ↑(r) − + + cos (krx) + cos (kry)ˆa aˆ − δ ψ↑(r) 2m 6(∆+ + ωHF) 6∆− ∆+ Z 2  Ω2 Ω2  2 3 ˆ† h pˆ + − 2 [g0Ξ(x, z)] 2 † i ˆ H↓ = d rψ↓(r) − + + cos (krx) + cos (kry)ˆa aˆ ψ↓(r) 2m 6∆+ 6(∆− − ωHF) ∆− † Hcavity = −∆caˆ aˆ √ √ Z 3 h 3g0Ξ(x, z)Ω+ ˆ† ˆ † 3g0Ξ(x, z)Ω− ˆ† ˆ † i HRaman = d r ψ↑(r)ψ↓(r)ˆa cos(krx) cos(kry) + ψ↓(r)ψ↑(r)ˆa cos(krx) cos(kry) + H.c. , 12∆+ 12∆− (11)

where we have separated out the longitudinal dependence We have ignored the small Stark shift term proportional of the cavity mode and kr = 2π/λ. In the regime of large to 1/∆+,− due to the cavity field. Our system therefore cavity detuning ∆c, the dynamics of the cavity mode is realizes a transverse-field Ising model of the form faster than other dynamics, and therefore we adiabati- X i j cally eliminate the cavity mode to obtain an atom-only HIsing ∝ Jij cos krxi cos krxj cos kryi cos kryjσˆxσˆx Hamiltonian. To do so, we define local spin operators i +hσˆz, (16) h i with direct spin-spin interaction mediated through the σˆ (r) = ψˆ†(r)ψˆ (r) − ψˆ†(r)ψˆ (r) /2 z ↑ ↑ ↓ ↓ cavity mode. h ˆ† ˆ ˆ† ˆ i Mapping to the Dicke model σˆx(r) = ψ↑(r)ψ↓(r) + ψ↓(r)ψ↑(r) /2. (12) To understand the threshold at which organization The two cavity-assisted Raman couplings are set to have occurs, it is useful to map our system onto a Dicke the same strength model [34, 35]. The experiment begins with a conden- √ √ sate in |↓i and when the cavity-mediated Raman process 3g Ω 3g Ω causes a spin flip, a momentum kick is also imparted onto 0 − = 0 + ≡ η, (13) 12∆− 12∆+ the atoms. Within the single-recoil limit, the dynamics can be captured by two atomic modes allowing the effective Hamiltonian to be written as ˆ ψ↓ =c ˆ↓ψ0 Z 2 ψˆ =c ˆ ψ , (17) 3 3 0 η 0 0 ↑ ↑ 1 Heff = d rd r Ξ(x, z)Ξ(x , z )× ∆c 0 0 0 where, for simplicity, cos(krx) cos(krx ) cos(kry) cos(kry )ˆσx(r)ˆσx(r ) Z 3 ψ0 = 1 + d r(Hˆk − δ)ˆσz(r), (14) ψ1 = 2 cos(krx) cos(kry). (18) where Since the pump lattice potential is retained in the Hamil- tonian, the differential Stark shift on ψ↑ and ψ↓ due to 2  Ω2 Ω2  the lattice beams has been taken into account. Taking ˆ pˆ + − 2 Hk = − + + cos (krx) advantage of the λ = 2π/kr periodicity along both the 2m 6(∆+ + ωHF) 6∆− pump and cavity direction, shifting the energy of the c  2 2  ↓ Ω+ Ω− 2 − + cos (krx). (15) mode to zero, and performing the integrals, the Hamil- 6∆+ 6(∆− − ωHF) tonian is evaluated to be 12

  2 2   2 2   † 3 Ω+ Ω− 1 Ω+ Ω− † H = − ∆caˆ aˆ + 2ωr + + − + − δ cˆ↑cˆ↑ 4 6(∆+ + ωHF) 6∆− 2 6∆+ 6(∆− − ωHF) √ √ " # 2 2 3 g0Ω+ † 3 g0Ω− †  †  g0 † † 3g0 † † + cˆ↑cˆ↓ + cˆ↓c↑ aˆ +a ˆ + cˆ↓c↓aˆ a + cˆ↑c↑aˆ a. (19) 24 ∆+ 24 ∆− 2∆− 4∆+

The bare energy of the c↑ mode is shifted due to the Dicke model, and the usual threshold expression applies: differential Stark shift q 1 2  2 2   2 2  ηc = (−∆c + Ng0/∆−)(2ωr + ωS − δ). (24) 3 Ω+ Ω− 1 Ω+ Ω− 2 ωS = + − + , Lattice calibration and Raman coupling balancing 4 6(∆+ + ωHF) 6∆− 2 6∆+ 6(∆− − ωHF) (20) which is a dynamic quantity during the linear ramp of the We calibrate the lattice depth of pump beams by per- Raman-beams’ power. The Raman detuning δ is there- forming Kapitza-Dirac diffraction of the BEC [72] pre- pared in either |↑i or |↓i. This also allows us to char- fore chosen such that the bare energy of thec ˆ↑ mode is always positive during the experiment sequence. acterize the differential Stark shift in the experiment. As mentioned above in Eq. 13, the Raman couplings The retroreflection mirror shared by the pump beams are chosen to be equal, and in anticipation of standard is mounted on a translation stage. Measuring the lattice Dicke model notation [3], we define this coupling as depth of the combined pump beams, we adjust the trans- lation stage to match the phases of the pump lattices at √ √ the position of the atoms. We note that the beat length 3N g0Ω+ 3N g0Ω− = ≡ η , (21) of the two pump lattices (separated in optical frequency 24 ∆ 24 ∆ D + − by 13.6 GHz) is ∼5 mm, much larger than the atomic cloud size; therefore, small mechanical fluctuations from where N =c ˆ† cˆ +c ˆ† cˆ is the total number of atoms. We ↑ ↑ ↓ ↓ the mirror mount will not cause the lattice to become now define the collective pseudospin-1/2 operators out-of-phase at the atoms. 1 To match each beam’s coupling strength, we linearly Jˆ = (ˆc† cˆ − cˆ† cˆ ) ramp up the Ω (Ω ) beam intensity for atoms prepared z 2 ↑ ↑ ↓ ↓ + − in |↓i (|↑i) while monitoring cavity emission. Above a ˆ † J+ =c ˆ↑cˆ↓ certain pump strength, atoms are transferred to the other ˆ † spin state with an accompanying brief cavity emission J− =c ˆ↓cˆ↑, (22) pulse. The critical Raman coupling strength at which where the Jˆ operate on the coupled pseudospin-1/2 spin- this pulse occurs is given by: spatial degree of freedom. The Hamiltonian can then be v u "  0 2# rewritten as u γκ −∆c + 2ωr + ω − δ η = t 1 + S , (25) c,single 2N γ + κ  2  Ng0 † H = − ∆c + aˆ aˆ + (2ωr + ωS − δ)Jˆz 2∆− where κ is the cavity decay rate and γ is a phenomeno- ηD † + √ (Jˆ+ + Jˆ−)(ˆa +a ˆ) logical parameter describing the collective spin decay N 0 rate [33]. Since only one beam is involved, ωS denotes the  2 2  N(2ωr + ωS − δ) 3g0 g0 † † differential Stark shift on |↑i and |↓i due to a single pump + + − cˆ↑c↑aˆ a.ˆ 2 4∆+ 2∆− beam. Matching the threshold for a single-beam spin-flip (23) then balances the two Raman coupling processes. We perform the calibration with ∆˜ c = −1.4 MHz. This is The first term in the third line is simply a energy off- significantly larger than the two-beam Stark-shift contri- 0 set, while the second term can be ignored as long as the bution, 2ωr +ωS −δ ≈ 10 kHz. Therefore, the additional † population ofc ˆ↑cˆ↑ is small, which is consistent with the Stark shift from the simultaneous presence of both beams single-recoil limit. The Hamiltonian therefore realizes the does not alter the matching condition considerably.