Arxiv:2104.08622V1 [Quant-Ph] 17 Apr 2021 Set by an External Magnetic ﬁeld

Critical dynamics and phase transition of a strongly interacting warm spin-gas

Yahel Horowicz,1, ∗ Or Katz,1, 2, ∗ Oren Raz,1 and Ofer Firstenberg1 1Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot 76100, Israel 2present address: Department of Electrical and Computer Engineering, Duke University, Durham, North Carolina 27708, USA† Phase transitions are emergent phenomena where microscopic interactions drive a disordered system into a collectively ordered phase. Near the boundary between two phases, the system can exhibit critical, scale-invariant behavior. Here, we report on a second-order phase transition accompanied by critical behavior in a system of warm cesium spins driven by linearly-polarized light. The ordered phase exhibits macroscopic magnetization when the interactions between the spins become dominant. We measure the phase diagram of the system and observe the collective behavior near the phase boundaries, including power-law dependence of the magnetization and divergence of the susceptibility. Out of equilibrium, we observe a critical slow-down of the spin response time by two orders of magnitude, exceeding ﬁve seconds near the phase boundary. This work establishes a controlled platform for investigating equilibrium and nonequilibrium properties of magnetic phases.

Investigation of new phases of matter has long cap- light [42–44]. Forston et al. have characterized the spin tured much scientific interest [1–13]. The macroscopic state by measuring the hysteresis appearing for slow vari- phase of interacting particles is determined by an inter- ation of the light polarization. However, the critical and play between the energy and the entropy of the system. collective behavior at the conditions where spontaneous When the entropy dominates, the system becomes dis- polarization occurs has not been systematically studied. ordered whereas, for energy dominated system, the or- Here, we report on the observation of critical behavior dered phase dominates as it minimizes the free energy of strongly interacting, warm cesium vapor. We mea- [14, 15]. Various phases are commonly described by sure the power-law dependence of the macroscopic mag- macroscopic order-parameters such as density, conduc- netization on both the light intensity and gas density, tivity, or magnetization. The transition between different as well as divergence of the susceptibility to an exter- phases is commonly accompanied by non-analytic behav- nal spin imbalance. We identify these phenomena as a ior of some properties of the system, e.g., the suscepti- second-order magnetic phase transition and measure the bility, correlation length, and time. These non-analytic phase diagram of the system. We observe divergence of properties are often associated with critical exponents, the spin response time up to a few seconds when cross- which can be categorized into universality classes that ing the phase transition, two orders of magnitude longer depend only on robust properties, such as the dimen- than the 20 millisecond spin lifetime. This accessible, sionality and symmetries of the system [14–16]. new platform paves the route for investigating spatial Spin systems are used as key examples for magnetic and out of equilibrium properties of optically-engineered phase transitions, as they are often relatively easy to spin systems. study theoretically [17, 18], numerically [19, 20], and ex- Figure 1 presents the physical system. The cesium perimentally [21–25]. The specific system we study in atoms, enclosed in a glass cell at near-ambient temper- this manuscript is a warm vapor of alkali atoms. These ature, are unpolarized in the absence of optical fields. atoms have a nonzero spin at their electronic ground level In this magnetically disordered phase, the cesium spins and could thus sustain steady magnetization. The spin, equally populate all 16 sub-states of the electronic ground a composite of the electronic and nuclear spins, can be level. To stimulate the transition into an ordered phase, prepared, controlled, and monitored by optical means we introduce linearly-polarized pumping light and in- utilizing the strong spin-orbit coupling provided by the crease the atomic density. The quantization axis zˆ is single valence electron [26, 27]. Frequent spin-exchange arXiv:2104.08622v1 [quant-ph] 17 Apr 2021 set by an external magnetic field. We tune the optical collisions between pairs of atoms in the vapor manifest frequency of the pumping light near the D1 optical tran- a local spin-dependent interaction. This interaction of- sition from the lower hyperfine manifold (Fg = 3) and set ten leads to decoherence and relaxation [28] but can also its polarization xˆ perpendicular to zˆ. This configuration enable coherent coupling and facilitate optical pumping aligns the spins along zˆ [43]. It preferentially and sym- and sensing in various applications [29–41]. In particu- metrically populates the two maximally-polarized states lar, a pioneering work by Forston et al. has demonstrated (with spin projection mF = ±4 along zˆ, in the up- the emergence of spontaneous spin-polarization and mag- per hyperfine manifold Fg = 4), marked by triangles in netic bi-stability upon absorption of linearly-polarized Fig. 1(b). It does so at a rate I linearly proportional to the intensity of the optical field (see Methods). We maintain a constant temperature T , which we can vary in the range of 55−120 ◦C. The temperature sets the ∗ These authors contributed equally to this work. vapor pressure, originating from a reservoir (droplet) of † Corresponding author: [email protected] cesium atoms, allowing us to control the atomic density 2 and thus to determine the rate of spin-exchange collisions The cesium vapor becomes magnetized in the experi- J. During a collision, the electronic spins of the two ment for a range of collision rates J and optical pumping colliding atoms experience random, mutual precession, (spin alignment) rates I. The spins then end up point- which conserves the total spin. The collisions change the ing either at the direction of the magnetic field (M > 0) internal atomic spin states, generate correlations between or opposite to it (M < 0), with the sign varying ran- the atoms, and repopulate the lower hyperfine manifold. domly between experimental realizations. The measured They comprise the microscopic inter-atomic interaction absolute magnetization of the vapor |M| at steady state in our system necessary for the formation of an ordered is shown in Fig. 2a as a function of the pumping power phase. The interplay between optical pumping and spin- and collision rate, which are respectively proportional to exchange collisions, leading to an alignment of the spins, I and J. We find a well-defined region in which the spins is schematically illustrated in Fig. 1c for the case of two are ordered and |M| > 0. We reach a magnetization atoms. It is important to note that the system does not as high as |M| = 0.45 in the ordered phase, where unity reach equilibrium, but rather a non-equilibrium steady |M| = 1 corresponds to maximal magnetization (all spins state, as there is a constant flow of energy from the pump- in the vapor maximally oriented along ±zˆ). ing light through the system to the surrounding environ- For other values of I and J, the spins remain in a dis- ment, which breaks the detailed balance condition. ordered phase with vanishing net magnetization in each The cell also contains buffer gas that renders the realization. Notably, the disordered phase at elevated atomic motion diffusive, yielding an average spin relax- atomic densities, i.e., for large J, results from atten- ation rate Γ = 58 s−1 that is limited by collisions with uation of the pumping light along the medium, which the walls (see Methods). We monitor the macroscopic decreases the actual (spatially averaged) spin-alignment magnetization of the vapor M using Faraday rotation rate below the critical value (see Methods and Supple- measurements of off-resonant probe light [26, 45]. mentary Fig. S1e). In Figs. 2c and 2d, we present the magnetization as a function of Γ/I and Γ/J along two contours crossing the phase boundary (marked by dashed a b lines in Fig. 2a). We observe a critical dependence of the disordered magnetization near the transition between the disordered 푥ො phase M=0 and ordered phases. The continuous but sharp transition 푧Ƹ 푦ො ordered of the magnetization, which acts as an order parameter, phase M >0 indicates that the process is associated with a second- order phase transition. The data fits well to power-law monitor 퐵푧Ƹ c optical functions with critical exponents βI = 0.53 ± 0.04 and I pumping βJ = 0.49 ± 0.02 (see Methods). To further explore the critical behavior near the 푥ො spin- pumping J boundary between the two phases, we measure the depen- 휎ො+ or 휎ො− exchange bias (optional) collision dence of the magnetization on an external bias. The bias towards positive or negative M is introduced by an auxil- Figure 1. Experimental system and coupling scheme. iary optical beam with circular polarization σˆ+ or σˆ−, re- (a) Cesium atoms contained in a glass cell experience frequent spectively, see Fig. 1a. The magnitude of the bias is given local spin-exchange collisions and are illuminated by linearly- by the optical pumping rate H, which is linear in the polarized pumping light. The pumping, with rate I, aligns bias beam’s intensity. In Fig. 3a, we present the steady each spin symmetrically either parallel or anti-parallel to the absolute magnetization as a function of H/Γ near the magnetic field Bzˆ. The collisions, with rate J, enable the gen- phase boundary at (I,J) = (1.5, 2.3)Γ (orange circles) eration of local correlations that lead to the ordering of the compared with the disordered phase at (I,J) = (0, 2.3)Γ spins at high densities. A weak, far detuned, monitor beam (blue circles). In the disordered phase, the steady mag- is used to measure the total atomic magnetization M along netization is determined by M = H/(H + Γ) and, for zˆ, which acts as an order parameter. For measurements of the magnetic susceptibility, an auxiliary, circularly-polarized, a weak bias, grows linearly as M ≈ H/Γ (blue line). light beam is introduced to bias the optical pumping toward Importantly, the linear dependence of the magnetization positive M (σˆ+ bias beam) or negative M (σˆ− bias beam). on the bias beam’s intensity is a universal property of (b) Atomic level structure. In the zˆ quantization basis, the weak resonant optical-pumping of uncorrelated atoms, pumping light comprises σˆ+ and σˆ− components (red arrows). independent of the transition strength or the particular It does not excite the maximally polarized states (N and H, atomic species. In contrast, near the phase boundary, corresponding to M = ±1). In conjunction with spontaneous we find a critical dependence of the magnetization on emission (orange arrows), the pumping drives the unpumped the bias beam, with a much sharper response. We fit atoms (|M| < 1, represented by ) symmetrically to both the data to the power-law function M = (H/Γ)1/δ and directions. (c) Illustration of the bifurcation mechanism for find the critical exponent δ = 2.65 ± 0.09. The deviation two atoms. The coaction of symmetric optical pumping (driv- from the standard optical-pumping relation indicates the ing to and ) and spin-exchange collisions (allowing the N H emergence of correlations between the atoms. pumping cycle to continue if the system arrives at HN or NH) renders the maximally-polarized states (NN and HH) the only Next, we measure the susceptibility function χ = basins of attraction. 3

Figure 2. Magnetic phase diagram and power-law dependence. a. Measured steady absolute magnetization |M| of the gas as a function of the atomic density and the power of the linearly-polarized pumping light. The atomic density sets the local spin-exchange collision rate J, while the pumping power sets the spin-alignment rate I. At moderate J and elevated I, the spins gain order and align either parallel or anti-parallel to the magnetic field, manifesting a macroscopic ordered phase. The presented data are linearly interpolated; the raw data are shown on a scatter plot in Supplementary Figs. S1(a) and S1(c). b. Simulated diagram using a nonlinear mean-field model (see Methods). c-d. Second-order phase transition, observed along the horizontal (black) and vertical (white) contours in a. Shown is measured magnetization as a function of Γ/I (c, with J = 3.8Γ) and Γ/J (d, with I = 4.5Γ), where Γ is the average spin-relaxation rate. We fit the data to a power law (black line) and find the critical exponents βI = 0.53 ± 0.04 and βJ = 0.49 ± 0.02 (See Methods). dM/dH near H = 0, which determines the response of of τ = 7.8 ± 1.3 seconds. Between the pumping pulses the system to a small external bias. Figure 3b presents (white areas), the magnetization rapidly vanishes, with a χ as a function of I at J = 2.3. We observe a strik- decay rate corresponding to the spin lifetime in the dark ing increase of the susceptibility by more than an order (T1 = 17ms at that temperature), as shown in Fig. 4b. of magnitude near the phase transition. Fitting χ to Importantly, as absorption of photons contributes to the a divergent power-law function, we find the critical ex- spin relaxation, the response time of uncorrelated spins ponent γ = 0.94 ± 0.10 (see Methods). The power-law by resonant optical pumping is always shorter than their dependence of the magnetization and the divergence of lifetime in the dark T1. In contrast, here near the phase the susceptibility near the phase boundary testify for the boundary, τ becomes longer than T1 by a remarkable fac- critical behavior of a magnetic, second-order phase tran- tor of 460. The observed response time varies between sition. the pulses due to the critical nature of the phenomenon, Second-order phase transitions exhibit critical, scale- and nevertheless, it is always at least two orders of mag- invariant behavior even out of equilibrium. Here we ex- nitude larger than T1. plore the dynamical transition from a disordered to an The slow down of the spin response time is a critical ordered phase by temporally varying I in a pulsed wave- phenomenon associated with the phase transition. Fig- form. We add no bias and use a fixed J = 3.7Γ, for which ure 4c shows the dependence of τ on I along a contour the critical value of I is 1.6Γ. For I > 1.6Γ, a magne- crossing the phase boundary (dashed line in Fig. 2a and tization initially at M = 0 builds up to a finite steady Fig. 4d). We observe a divergence of τ near I = 1.6Γ and value, randomly in either of the two directions. Figure find the critical exponent zν = 0.86 ± 0.07 by fitting to 4a shows an example of the measured magnetization sub- a divergent power-law function (see Methods). The spin ject to a periodic sequence of pumping pulses (gray areas) response time diverges at the entire phase boundary, as tuned slightly above I = 1.6Γ. We observe a slowdown of shown in Fig. 4d. The observation of a critical divergence the polarization process, with an average response time of the response time (Fig. 4d) near the boundary of the 4

disordered Exponent Measured value Mean-field value Relation 0.2 critical −βI βI 0.53 ± 0.04 0.5 M ∝ I −βJ βJ 0.49 ± 0.02 0.5 M ∝ J 0.1 δ 2.65 ± 0.09 3 M ∝ H1/δ γ γ 0.94 ± 0.10 1 dM/dH ∝ I 0 zν 0.86 ± 0.07 1 τ ∝ I−zν 0 0.025 0.05 Table I. Measured critical exponents compared with the exponents of the mean-field universality class. Simplified power- law relations near the phase boundary are presented for con- disordered phase text, with M denoting the magnetization, I the optical spin- 101 alignment rate, J the spin-exchange collision rate, H the bias pumping rate, and τ the dynamical response time. Full ex- pressions of the critical functions are given in Methods. The fits of the power-law functions in a logarithmic scale is pre- 100 0.8 1 1.2 1.4 1.6 sented in Supplementary Fig. S2.

them to those of mean-field models. While essentially Figure 3. Critical behavior in the presence of a weak the spin-exchange interaction is local, rapid diffusion of external bias. a. The magnetization as a function of the the atoms (at the short time scale ∼ T ) leads to in- bias strength, quantified by the optical-pumping rate H in- 1 duced by an auxiliary, circularly-polarized beam. The spin- teraction between initially distant atoms, thus rendering exchange rate J = 2.3Γ is fixed. For I = 0 (disordered phase) the interaction range effectively longer. Higher buffer-gas the magnetization increases linearly with the bias (blue cir- pressures would slow down the diffusion and potentially cles), as expected for standard optical-pumping of spins. In localize the interaction, which could enable the forma- contrast, for the critical value I = 1.5Γ at the phase boundary, tion of distinct spatial domains. Other approaches for the system becomes critical and the magnetization sharply realizing interacting magnetic domains can exploit multi- increases (orange circles). Solid lines are fits to linear (blue) mode patterns of the pumping beam in a single-cell or and power-law (orange) functions, the latter providing the multiple-cell realizations with engineered spin-dependent critical exponent δ = 2.65 ± 0.09. b. The susceptibility to couplings. The latter approach can exploit, for example, an external bias dM/dH (purple circles), diverging near the the Faraday rotation of a probe beam in one cell to bias phase boundary (gray area). The fit to a divergent power-law (black line) provides the critical exponent γ = 0.94 ± 0.10 for the spin of another cell. the disordered phase. The ability to control and manipulate the spin interactions and magnetic field with a relatively simple setup ordered phase (Fig. 2a) attests to the collective nature of at ambient conditions has many potential applications, the phase transition. starting from demonstrating and better understanding To model the observed phenomena, we employ a non- the mechanism behind anomalous thermal relaxations in linear mean-field model that describes the evolution of magnetic systems [19, 20, 46–48] to the realization of the density matrix of the mean cesium spin in the vapor. Ising machines that can serve to compute a wide class The theoretical model, described in Methods, captures of optimization problems [49–51]. the main features of the emergent phase transition, including the bi-stability, the power-law dependencies, and the divergence of the susceptibility and the spin response ACKNOWLEDGMENTS time. In Figs. 2b and 4e, we present the numerically cal- culated phase diagram and the spin response time, which We acknowledge financial support by the estate of agree with the measured results reasonably well. Emile Mimran, the Israel Science Foundation, the US- In summary, we observe a magnetic phase transition Israel Binational Science Foundation (BSF) and US Na- of warm cesium atoms stimulated by linearly-polarized tional Science Foundation (NSF), the Minerva Founda- light and by frequent spin-exchange collisions. We char- tion with funding from the Federal German Ministry acterize the phase diagram, reveal a critical power-law for Education and Research, the Shlomo and Michla behavior near the phase transition, and observe a critical Tomarin career development chair, the Abramson Fam- slow-down of the spin response time. The substantial de- ily Center for Young Scientists, and the Laboratory in viation of our observations from standard experiments, Memory of Leon and Blacky Broder. This research was in which the atoms are uncorrelated, testifies for the col- facilitated by the Talpiot Program 18-19-30-36. lective nature of the ordered phase and the transition. The measured critical exponents fall into the universality class of mean-field models [14]. In Table I, we summa- rize the measured values of the exponents and compare 5

Figure 4. Critical slow-down out of equilibrium. a. Measured magnetization in response to optical-pumping pulses, with I = 1.6Γ and J = 3.7Γ close to the phase boundary. During the pumping on times (gray shaded areas), the magnetization builds up to |M| & 0.25 in a random direction, at an average response time of τ = 7.8 ± 1.3 seconds (time for the magnetization to reach 63% of its steady value). b. During the off times, the system becomes disordered, and |M| rapidly decays with the standard spin-lifetime T1 = 17 ms. The data shown in b is marked by a green arrow in a. c. Divergence of the spin response time near the phase boundary and a fit to a power-law divergence (black line). Blue points are data obtained along the horizontal (dashed) contour in d, as a part of an automatic 40×330 measurement, whereas the red point corresponds to the data in a, obtained by manually fine-tuning around the critical point. The bound τ . T1, valid for uncorrelated spins, is violated by a large factor. d. Measured spin response time in a logarithmic scale. The divergence appears at the phase boundary, where the spins are strongly correlated. The presented data is linearly interpolated; the raw data are shown in Supplementary Figs. S1(b) and S1(d). e. Simulated spin-response time using a nonlinear mean-field model (see Methods). 6 Methods by an acusto-optic modulator and a commercial intensity noise-eater. A λ/4 waveplate renders the polarization of the beam circular, and it is combined with Detailed experimental setup the probe beam using a non-polarizing beam splitter, as shown in Fig. 1. The beam has a Gaussian profile We use a cubic borosilicate glass cell of length 15 mm with 1/e2 radius of 1 cm. It is 1.2-GHz blue-detuned containing cesium vapor, and a buffer-gas mixture of from the Fg = 3 → Fe = 4 transition for the critical N2 and neon, 16.5 Torr each. These gases slow down behavior experiments, and it is set to resonance with the the diffusion of cesium atoms to the cell walls, and the Fg = 4 → Fe = 3 for calibration experiments. N2 additionally enables the non-radiative decay (quench- Our measurements are found sensitive both to the de- ing) of electronically-excited cesium atoms. The buffer tuning of the pumping beam and to the magnitude of gases broaden the cesium optical lines to γc = 137 MHz the magnetic field. Drift in the detuning predominantly (HWHM), yet the four optical transitions Fg = {3, 4} → affects the rate I, while drift in the magnetic field was Fe = {3, 4} of the D1 line remain resolved. We heat the found to vary the spin response time. Therefore, both cell using high-frequency electrical current at 390 kHz quantities are monitored and kept constant during the flowing through high-resistance twisted-pair wires in a experiment. custom oven. We control the magnetic field in the cell using three pairs of Helmholtz coils and set a constant magnetic field of B = 1 G along zˆ. The coils are located Experimental calibrations within four µ-metal layers, shielding the cell from external magnetic fields. The spin lifetime T1 = 1/Γ is determined by measuring The linearly-polarized pumping beam originates from the decay rate of spins oriented along the magnetic field a free-running DBR diode laser at 895 nm. The laser in the absence of resonant optical fields (measurement in frequency is blue-detuned by ∆ = 700 MHz from the the dark). In this measurement, the spins are first opti- Fg = 3 → Fe = 4 optical transition. We control the cally pumped by two circularly polarized beams resonant beam power using a commercial noise-eater consisting with the Fg = 3 → Fe = 4 and Fg = 4 → Fe = 3 tran- of a liquid crystal, polarizing beam splitter, and a pho- sitions along yˆ, then rotated to the zˆ axis by a magnetic todetector. The beam then passes through a mechan- field pulse, and finally measured with the off-resonant ical shutter and a high-quality linear polarizer, which probe while the pumping beams are off. The relaxation sets the polarization of the beam to be linear along xˆ, rate has a small, linear dependence on the cell’s tem- and finally through a λ/4 waveplate mounted on a pre- ◦ perature, satisfying Γ(T ) = Γ0 + 0.35(T [ C] − 75) in cision, computer-controlled rotating mount. For most of the tested range (55 − 120 ◦C), predominantly due to the the experiments, the fast axis of the waveplate is care- temperature dependence of the diffusion coefficient of the fully aligned with the direction of the linear polariza- cesium atoms. In the entire analysis and figures, we use tion, unaffecting the linear polarization of light. Nev- −1 the constant value Γ = Γ0 = 58 s . Furthermore, in the ertheless, when needed, rotation of the waveplate en- measured and simulated response-time data (Figs. 4d,e), ables rapid calibration of the maximal magnetization we set τ = T1 for all measurements where the spin re- within the experimental sequence (as detailed below). sponse was smaller than 10−3 of the maximal measured 2 The Gaussian beam is then expanded to a 1/e radius magnetization. of 1 cm to cover the entire cell area and enter the cell The spin-alignment rate I is independently determined in the yˆ direction. We measure the magnetization in a by measuring the excess decay rate of the spins in the set of 330 × 40 measurements varying the vapor density presence of the pumping beam at T = 75◦C. We find a 11 −3 13 −3 in the range of 7 × 10 cm ≤ n ≤ 5 × 10 cm and linear dependence on the intensity of the beam Φ, with the optical intensity of the pumping field in the range of a ratio I/Φ = 457 cm2/J. Note that the local spin- 2 mW ≤ Φ ≤ 40 mW . alignment rate varies across the profile of the pumping We monitor the magnetization using a yˆ-polarized, 3- beam, and therefore we always refer to an average rate mW probe beam that propagates along zˆ. The beam across the beam. The pumping beam is also attenuated originates from another free-running DBR diode laser at along the propagation direction (yˆ) due to absorption 895 nm. It has a Gaussian profile with 1/e2 radius of by the atoms. The y-dependent spin-alignment rate is 1.5 cm, covering much of the atoms in the cell. The therefore given by I exp(−nσey) where n(T ) is the vapor beam is blue-detuned by 100 GHz from the D1 lines to density and σe(∆) is the absorption cross-section of the avoid photon absorption. It probes the z-component of beam. the electron spin of the gas via Faraday rotation. The The spin-exchange rate J(T ) is independently deter- beam then goes into a balanced polarimetry setup, which mined by a measurement of the relaxation of the spins outputs a signal proportional to the magnetization along transverse to the magnetic field, in the absence of reso- zˆ. nant optical fields. In this measurement, we weakly pump The bias beam originates from a third free-running the spins along yˆ, apply a magnetic field B along xˆ, and DBR diode-laser at 895 nm, whose power is controlled monitor the precession of the spins, decaying at a deco- 7 herence rate Γ2(B). At each temperature, we determine measured susceptibility in Fig. 3b to the function χ = −γ the spin-exchange rate J = [Γ2(B) − Γ2(0)]/q from the χ0(I0/I − 1) in the disordered phase for I < I0. As measured relaxation at high magnetic field B = 2 G and with the critical exponent β fitting, we first estimate the by subtracting the effect of other field-independent re- initial guess for I0 and γ by fixing the value of one pa- laxations Γ2(0) [31, 52]. Here q = 4.57 is the numerical rameter and fitting for the other and use those results slow-down factor that accounts for the reduction of the for the final fit. We also exclude points that are close to rate by coupling to the nuclear spin I = 7/2 [31]. the maximal measured one and use weights w = (Γ/I)3 The bias rate H is independently determined by mea- in order to compensate for the finite values of the data suring the magnetization as a function of the intensity ΦH around the critical point compared with the divergent of the bias (circularly-polarized) beam. We fit the mea- values of the model. For J = 2.3Γ, we find the critical sured magnetization to the function M = aΦH/(aΦH +Γ) exponent γ = 0.94 ± 0.10 and the critical spin-alignment with a = 1.179 ± 0.005 and determine the linear coeffi- rate I0 = 1.393 ± 0.036Γ in the disordered phase. 2 cient H/ΦH = 99 cm /J. We determine the critical exponent zν by fitting the As we increase the temperature of the cell to increase spin response time τ in Fig. 4c to the power-law function −zν J, the vapor density, and therefore the number of gaseous τ = τ0(1 − I0/I) . We fit the functions to 11 measure- spins in the cell, increase as well. We thus calibrate for ments in the range of exchange rates J = 3.47 − 3.96 Γ the maximal polarization of the vapor at each tempera- (equivalent to T = 87 ± 1 ◦C) in order to comprehend ture to properly determine the magnetization M, which the deviation of the model parameters. As before, we describes the portion of polarized spins in the gas. For use a three-step fitting scheme, exclude points that are that, we use two strong circularly polarized beams that close to the maximal measured one, and use weights. We cover the entire cell and optically pump the spins along then find the critical exponent zν = 0.86 ± 0.07 and the yˆ. We then turn the beams off and apply a magnetic field critical spin-alignment rate I0 = 1.599 ± 0.004Γ. along Bx that stimulates the precession of the spins. The precession amplitude corresponds to the maximal signal obtained by our probe beam, which is identified as the maximal polarization and used to calibrate the magneti- Spin-alignment by linearly polarized light zation in the experiments. The residual circularity of the polarization of the The linearly-polarized pumping light resonantly inter- pumping field is automatically zeroed within each ex- acts with the optical transition Fg = 3 → Fe = 4. As perimental sequence by applying the same technique as the electric field is perpendicular to the magnetic field, in the spin-exchange rate J(T ) calibration for varying upon absorption of a photon, the atom is excited and its λ/4 waveplate angles around the known optimal point. spin projection along the magnetic field mF changes by Since the waveplate is mounted on a precision, computer- either +1 or −1. Importantly, this interaction preferably controlled rotating mount, we automatically repeat this increases the absolute spin projection |mF | [42]. An atom process until the procession amplitude is minimal. This with mF > 0 would preferably be excited while increas- ensures the beam has a minimal circular polarization. ing its spin to mF + 1, and, symmetrically, an atom with mF < 0 would preferably decrease its spin to mF − 1. This preference is quantified in Table II, which presents Critical behavior the probability for changing the ground-level spin projection by absorption of a linearly-polarized photon. The Near the phase boundary, the data exhibit power-law on-average increase of |mF | drives the spins towards ei- dependence with critical exponents. Here we describe the ther mF = 4 or mF = −4, thus generating symmetric fitting procedure we use to determine these exponents. alignment. We determine the critical exponent βI by fitting the data in Fig. 2c to the function M(I > I ) = M (1 − 0 0 |mF | p|mF |→|mF |+1 p|mF |→|mF |−1 ∆p βI I0/I) and M(I < I0) = 0. For proper fitting, we 0 1/2 1/2 0 first estimate the initial guess for I0 and βI by fixing the value of one parameter and fitting for the other and 1 15/21 6/21 9/21 use those results for the final fit. For J = 3.8Γ (black hor- 2 7/8 1/8 6/8 izontal dashed line in Fig. 2a), we find the critical expo- 3 28/29 1/29 27/29 nent βI = 0.53±0.04 and the critical spin-alignment rate I0 = 1.639 ± 0.006Γ. Similarly, we determine the critical Table II. Transition probabilities for changing the spin pro- exponent βJ by fitting the data in Fig. 2d to the function jection |mF | of cesium atoms in Fg = 3 upon absorption of βJ the pumping beam. The pumping tends to increase |m |, as M(J > J0) = M0(1 − J0/J) and M(J < J0) = 0. For F indicated by the positive ∆p = p −p , I = 4.5Γ (white vertical dashed line in Fig. 2a), We find |mF |→|mF |+1 |mF |→|mF |−1 thus driving the spins towards the maximally-polarized states the critical exponent βJ = 0.49 ± 0.02 and the critical mF = ±4 in Fg = 4. spin-exchange rate J0 = (2.393 ± 0.004)Γ. We determine the critical exponent γ by fitting the 8

Theoretical model destruction of the electron spin in the excited-level manifold by collisions with buffer gas atoms at a rate γp. As We implement a mean-field model describing the dy- an approximation, our numerical calculation assumes a namics of the mean density matrix ρ¯ of a single atom in quasi-steady-state solution of ρ¯e in Eq. (2), which adia- the vapor, following the model by Happer et al. [26]. The batically follows the dynamics of the ground-level density density matrix matrix ρ¯g. We then numerically solve only the dynamics of the density matrix ρ¯g. ! ρ¯ ρ¯ ρ¯ = e eg (1) The dynamics of the ground-level density matrix ρ¯g is ρ¯ge ρ¯g modeled by consists of the 16 spin state in the ground level Fg = 2γq † 3 {3, 4} denoted by ρ¯g, the 16 spin states in the excited Lg(¯ρ) = 3D2 D ρ¯eD − Γ 4 ρ¯g − Fρ¯gF . (4) level Fe = {3, 4} denoted by ρ¯e, and the optical-coherence 3 − qJ 4 ρ¯g − Sρ¯gS + M · (¯ρgS + Sρ¯g − 2iS × ρ¯gS) matrices between the two ρ¯ge and ρ¯eg. We describe the evolution of the density matrix by solving the non-linear Liouville equation The first term describes the repopulation of the ground level by quenched excited atoms, where D is the ampli- i ∂tρ¯ = − H + VL, ρ¯ + L(¯ρ). (2) tude of the dipole moment of the optical transition. The ~ second term describes the destruction of the total spin F = I + S at a rate Γ, predominantly due to diffusion H = Hg + He is the spin Hamiltonian of an alkali atom. to the cell walls. The last term describes the effect of Both ground (Hg) and excited-level (He) Hamiltonians consist of the hyperfine interaction AI·S and the interac- spin-exchange collisions, affecting the electron spin at a tion with a magnetic field, predominantly of the electron rate qJ. Importantly, the spin-exchange interaction has spin gB · S. Here S denotes the electronic spin oper- a nonlinear (quadratic) dependence on the density ma- ator and I denotes the nuclear spin operator. We use trix, since the magnetization of the vapor is given by M = Tr(¯ρ S). The quadratic dependence enables the A = 2.3 GHz and g = 2.8 MHz/G in Hg, and we use g emergence of bi-stable steady solutions and manifests the A = 290 MHz and g = 0.9 MHz/G in He. The second term in Eq. (2) describes the atom-photon interaction correlations induced by different atoms via the exchange interaction. It is therefore crucial for the emergence of VL = −E · D, coupling the oscillating electric field of the i(k·v−ω t) the ordered phase and the observed critical phenomena. optical pumping field E = E0e with the atomic dipole operator D. The term k · v denotes the Doppler We numerically calculate the steady state and the shift due to the finite velocity v of that atom. The third buildup of the magnetization of the vapor by solving term L(¯ρ) describes the coupling of the spins to other Eqs. (2-4) using an initially unpolarized state. We use degrees of freedom via radiative or collisional channels γc = 1.86 GHz, γp = 219 MHz, γq = 265 MHz, q = 4.57, ◦ in the ground Lg(¯ρ) and excited levels Le(¯ρ), as well as ∆ = 700 MHz, and Γ(T ) = Γ0 + 0.35(T [ C] − 75) dephasing of the optical coherences at a rate γc. to model the parameters of the experiment [26]. We Rapid collisions with buffer-gas atoms increase γc sig- run the simulation for a range of 70 vapor densities nificantly (pressure broadening) with respect to both the 7×1011 cm−3 ≤ n ≤ 5×1013 cm−3 and 270 optical inten- spontaneous emission rate and the Rabi frequency of the sities 2 mW ≤ Φ ≤ 40 mW . These in turn determine the optical pumping fields. In a frame rotating near the light spin exchange rate J = nhσexviv and spin alignment rate frequency at a detuning ∆, the optical coherence ρ¯ 2 −10 3 −1 eg I = s E0 exp(−OD). Here hσexviv = 7 × 10 cm s , is hence maintained in a quasi-steady state, satisfying s = 220 MHz (mW/cm2)−1, and the optical depth OD = −1 ρ¯eg = wρ¯g − ρ¯ew. The operator w = E (E0 · D) v nσey is given by the measured value at each vapor den- denotes the fraction of optical coherence, with E = sity. He − Hg + ~(k · v − ∆ − iγc) accounting for the optical detuning and dephasing of the different transitions. To alleviate the numerical complexity of the model, we zero the rapidly oscillating hyperfine coherences of ρ¯g at The operation v denotes thermal averaging using the Maxwell-Boltzmann distribution, to properly account for every step of the simulation. This approximation cor- Doppler broadening of the optical transition. responds to the rapid decay of these coherences due to Atoms in the excited level experience rapid relaxation, spin-exchange relaxation in the experiment. In the sim- which we model by ulations presented in Fig. 2b and Fig. 4e, we further zero the Zeeman coherences of ρ¯g at every step of the sim- 3 ulation to speed up the calculation. Indeed in practice, Le(¯ρ) = −γqρ¯e − γp 4 ρ¯e − Sρ¯eS (3) spin-exchange collisions at a large magnetic field partially The first term denotes de-excitation (quenching) of the relax these coherences. We demonstrate in Supplemen- population from the excited level to the ground level at tary Fig. S3 that the effect of these coherences is lim- a rate γq, including both spontaneous emission and colli- ited to variations of the spin response time at the phase sions with N2 molecules. The second term describes the boundary at lower densities. 9

[1] F. Yang, S. F. Taylor, S. D. Edkins, J. C. Palmstrom, https://science.sciencemag.org/content/339/6127/1582.full.pdf. I. R. Fisher, and B. L. Lev, Nematic transitions in iron [14] N. D. Mermin, Introduction to Phase Transi- pnictide superconductors imaged with a quantum gas, Na- tions and Critical Phenomena. H. Eugene Stan- ture Physics 16, 514 (2020). ley. Oxford University Press, New York, 1971. [2] R. Harris, Y. Sato, A. J. Berkley, M. Reis, F. Al- xx, 308 pp., illus. International Series of Mono- tomare, M. H. Amin, K. Boothby, P. Bunyk, C. Deng, graphs on Physics, 4034 (American Association for C. Enderud, S. Huang, E. Hoskinson, M. W. John- the Advancement of Science, 1972) pp. 502–502, son, E. Ladizinsky, N. Ladizinsky, T. Lanting, R. Li, https://science.sciencemag.org/content/176/4034/502.1.full.pdf. T. Medina, R. Molavi, R. Neufeld, T. Oh, I. Pavlov, [15] S. Ma, Statistical Mechanics (WORLD SCIENTIFIC, I. Perminov, G. Poulin-Lamarre, C. Rich, A. Smirnov, 1985). L. Swenson, N. Tsai, M. Volkmann, J. Whittaker, [16] C. D. Schneider, G. M. and M. S. G. (Hrsg.), Phase and J. Yao, Phase transitions in a programmable Transitions and Critical Phenomena, 5 (Academic quantum spin glass simulator, Science 361, 162 (2018), Press Inc., London, New York, 1973) pp. 380–380, https://science.sciencemag.org/content/361/6398/162.full.pdf. https://onlinelibrary.wiley.com/doi/pdf/10.1002/bbpc.19730770517. [3] D.-S. Ding, H. Busche, B.-S. Shi, G.-C. Guo, and C. S. [17] R. J. Glauber, Time-dependent statistics of the ising Adams, Phase diagram and self-organizing dynamics in model, Journal of mathematical physics 4, 294 (1963). a thermal ensemble of strongly interacting rydberg atoms, [18] A. Nava and M. Fabrizio, Lindblad dissipative dynamics Phys. Rev. X 10, 021023 (2020). in the presence of phase coexistence, Physical Review B [4] S. Choi, J. Choi, R. Landig, G. Kucsko, H. Zhou, J. Isoya, 100, 125102 (2019). F. Jelezko, S. Onoda, H. Sumiya, V. Khemani, and et al., [19] M. Baity-Jesi, E. Calore, A. Cruz, L. A. Fernandez, J. M. Observation of discrete time-crystalline order in a dis- Gil-Narvión, A. Gordillo-Guerrero, D. Iñiguez, A. Las- ordered dipolar many-body system, Nature 543, 221–225 anta, A. Maiorano, E. Marinari, et al., The mpemba effect (2017). in spin glasses is a persistent memory effect, Proceedings [5] J. Zhang, P. W. Hess, A. Kyprianidis, P. Becker, A. Lee, of the National Academy of Sciences 116, 15350 (2019). J. Smith, G. Pagano, I.-D. Potirniche, A. C. Potter, [20] A. Gal and O. Raz, Precooling strategy allows exponen- A. Vishwanath, and et al., Observation of a discrete time tially faster heating, Physical review letters 124, 060602 crystal, Nature 543, 217–220 (2017). (2020). [6] C. G. Wade, E. Levi, M. Marcuzzi, J. M. Kondo, [21] I.-D. Potirniche, A. C. Potter, M. Schleier-Smith, I. Lesanovsky, C. S. Adams, and K. J. Weatherill, A A. Vishwanath, and N. Y. Yao, Floquet symmetry- terahertz-driven non-equilibrium phase transition in a protected topological phases in cold-atom systems, Phys. room temperature atomic vapour., Nature communica- Rev. Lett. 119, 123601 (2017). tions. 9, 3567 (2018). [22] S. Gopalakrishnan, B. L. Lev, and P. M. Goldbart, Atom- [7] C. Carr, R. Ritter, C. G. Wade, C. S. Adams, and K. J. light crystallization of bose-einstein condensates in mul- Weatherill, Nonequilibrium phase transition in a dilute timode cavities: Nonequilibrium classical and quantum rydberg ensemble, Phys. Rev. Lett. 111, 113901 (2013). phase transitions, emergent lattices, supersolidity, and [8] K. W. Post, A. S. McLeod, M. Hepting, M. Bluschke, frustration, Phys. Rev. A 82, 043612 (2010). Y. Wang, G. Cristiani, G. Logvenov, A. Charnukha, [23] C. Huang, Some experimental aspects of spin glasses: A G. X. Ni, P. Radhakrishnan, M. Minola, A. Pasupathy, review, Journal of magnetism and magnetic materials 51, A. V. Boris, E. Benckiser, K. A. Dahmen, E. W. Carl- 1 (1985). son, B. Keimer, and D. N. Basov, Coexisting first- and [24] M. Gaudesius, Y.-C. Zhang, T. Pohl, R. Kaiser, and second-order electronic phase transitions in a correlated G. Labeyrie, Phase diagram of spatiotemporal instabil- oxide, Nature Physics 14, 1056 (2018). ities in a large magneto-optical trap, Phys. Rev. A 103, [9] M. Marcuzzi, E. Levi, S. Diehl, J. P. Garrahan, and L041101 (2021). I. Lesanovsky, Universal nonequilibrium properties of dis- [25] H. Miyake, G. A. Siviloglou, G. Puentes, D. E. Pritchard, sipative rydberg gases, Phys. Rev. Lett. 113, 210401 W. Ketterle, and D. M. Weld, Bragg scattering as a probe (2014). of atomic wave functions and quantum phase transitions [10] R. Toskovic, R. van den Berg, A. Spinelli, I. S. Eliens, in optical lattices, Phys. Rev. Lett. 107, 175302 (2011). B. van den Toorn, B. Bryant, J.-S. Caux, and A. F. Otte, [26] W. Happer, Y.-Y. Jau, and T. Walker, Optically pumped Atomic spin-chain realization of a model for quantum atoms (John Wiley & Sons, 2010). criticality, Nature Physics 12, 656–660 (2016). [27] M. Auzinsh, D. Budker, and S. Rochester, Optically [11] K. Janzen, A. Engel, and M. Mézard, The lévy spin glass polarized atoms: understanding light-atom interactions transition, EPL (Europhysics Letters) 89, 67002 (2010). (Oxford University Press, 2010). [12] D. Lips, A. Ryabov, and P. Maass, Nonequilibrium trans- [28] I. Novikova, R. Walsworth, and Y. Xiao, Electromagnet- port and phase transitions in driven diffusion of interact- ically induced transparency-based slow and stored light in ing particles, Zeitschrift für Naturforschung A 75, 449 warm atoms, Laser & Photonics Reviews 6, 333 (2012). (2020). [29] W. Chalupczak, R. M. Godun, P. Anielski, A. Woj- [13] J. Zhang, C.-Z. Chang, P. Tang, Z. Zhang, X. Feng, ciechowski, S. Pustelny, and W. Gawlik, Enhancement K. Li, L.-l. Wang, X. Chen, C. Liu, W. Duan, of optically pumped spin orientation via spin-exchange K. He, Q.-K. Xue, X. Ma, and Y. Wang, Topology- collisions at low vapor density, Physical Review A 85, driven magnetic quantum phase transition in 043402 (2012). topological insulators, Science 339, 1582 (2013), [30] M. Romalis, Hybrid optical pumping of optically dense 10

alkali-metal vapor without quenching gas, Physical review markovian mpemba effect and its inverse, Proceedings of letters 105, 243001 (2010). the National Academy of Sciences 114, 5083 (2017). [31] W. Happer and A. C. Tam, Effect of rapid spin ex- [49] T. Inagaki, Y. Haribara, K. Igarashi, T. Sonobe, S. Ta- change on the magnetic-resonance spectrum of alkali va- mate, T. Honjo, A. Marandi, P. L. McMahon, T. Umeki, pors, Phys. Rev. A 16, 1877 (1977). K. Enbutsu, et al., A coherent ising machine for 2000- [32] O. Katz, R. Shaham, and O. Firstenberg, Quantum inter- node optimization problems, Science 354, 603 (2016). face for noble-gas spins (2019), arXiv:1905.12532 [quant- [50] P. L. McMahon, A. Marandi, Y. Haribara, R. Hamerly, ph]. C. Langrock, S. Tamate, T. Inagaki, H. Takesue, S. Ut- [33] I. Kominis, T. Kornack, J. Allred, and M. V. Romalis, sunomiya, K. Aihara, et al., A fully programmable 100- A subfemtotesla multichannel atomic magnetometer, Na- spin coherent ising machine with all-to-all connections, ture 422, 596 (2003). Science 354, 614 (2016). [34] E. Babcock, I. Nelson, S. Kadlecek, B. Driehuys, L. W. [51] K. Takata, A. Marandi, R. Hamerly, Y. Haribara, Anderson, F. W. Hersman, and T. G. Walker, Hybrid D. Maruo, S. Tamate, H. Sakaguchi, S. Utsunomiya, and spin-exchange optical pumping of 3He, Phys. Rev. Lett. Y. Yamamoto, A 16-bit coherent ising machine for one- 91, 123003 (2003). dimensional ring and cubic graph problems, Scientific re- [35] O. Katz and O. Firstenberg, Light storage for one sec- ports 6, 1 (2016). ond in room-temperature alkali vapor, Nature Communi- [52] O. Katz, M. Dikopoltsev, O. Peleg, M. Shuker, J. Stein- cations 9, 2074 (2018). hauer, and N. Katz, Nonlinear elimination of spin- [36] K. Mouloudakis and I. K. Kominis, Spin-exchange colli- exchange relaxation of high magnetic moments, Physical sions in hot vapors creating and sustaining bipartite en- review letters 110, 263004 (2013). tanglement, Phys. Rev. A 103, L010401 (2021). [37] J. Kong, R. Jiménez-Martínez, C. Troullinou, V. G. Lucivero, G. Tóth, and M. W. Mitchell, Measurement- induced, spatially-extended entanglement in a hot, strongly-interacting atomic system, Nature communica- tions 11, 1 (2020). [38] O. Katz, O. Peleg, and O. Firstenberg, Coherent coupling of alkali atoms by random collisions, Phys. Rev. Lett. 115, 113003 (2015). [39] F. Ripka, H. Kübler, R. Löw, and T. Pfau, A room- temperature single-photon source based on strongly interacting rydberg atoms, Science 362, 446 (2018), https://science.sciencemag.org/content/362/6413/446.full.pdf. [40] T. Wang, D. F. J. Kimball, A. O. Sushkov, D. Aybas, J. W. Blanchard, G. Centers, S. R. O. Kelley, A. Wick- enbrock, J. Fang, and D. Budker, Application of spin- exchange relaxation-free magnetometry to the cosmic ax- ion spin precession experiment, Physics of the Dark Uni- verse 19, 27 (2018). [41] O. Katz, R. Shaham, E. S. Polzik, and O. Firstenberg, Long-lived entanglement generation of nuclear spins using coherent light, Phys. Rev. Lett. 124, 043602 (2020). [42] N. Fortson and B. Heckel, Spontaneous spin polarization and bistability in atomic vapors by optical pumping with unpolarized light, Phys. Rev. Lett. 59, 1281 (1987). [43] W. M. Klipstein, S. K. Lamoreaux, and E. N. Fortson, Observation of spontaneous spin polarization in an optically pumped cesium vapor, Phys. Rev. Lett. 76, 2266 (1996). [44] A. Andalkar, R. B. Warrington, M. V. Romalis, S. K. Lamoreaux, B. R. Heckel, and E. N. Fortson, Experimen- tal and theoretical study of spontaneous spin polarization and hysteresis in cesium vapor, Phys. Rev. A 65, 023407 (2002). [45] B. S. Mathur, H. Y. Tang, and W. Happer, Light propagation in optically pumped alkali vapors, Phys. Rev. A 2, 648 (1970). [46] Z.-Y. Yang and J.-X. Hou, Non-markovian mpemba effect in mean-field systems, Physical Review E 101, 052106 (2020). [47] I. Klich, O. Raz, O. Hirschberg, and M. Vucelja, Mpemba index and anomalous relaxation, Physical Review X 9, 021060 (2019). [48] Z. Lu and O. Raz, Nonequilibrium thermodynamics of the 11

Figure S1. Diﬀerent representations of the phase-diagram data. a-b. Raw data of the measured steady magnetization |M| and spin-response time τ. c-d. Raw data of |M| and τ as a function of the normalized inverse rates Γ/J and Γ/I. e-f. Interpolated |M| and τ data as a function of Γ/J and Γ/hIi, where the spatially-averaged optical spin-alignment rate is given L by hIi= 0 I exp(−nσey)dy, with n(T ) the vapor density and σe(∆) the absorption cross-section for the pumping ﬁeld. ´ 12 a d 10-1 10-1

10-2 ordered phase

10-3 10-2 10-1 disordered 10-2 critical

b 10-1 10-3 10-2

10-2 ordered phase e disordered phase 10-3 10-2 10-1

c 101

102

ordered phase 100 100 10-3 10-2 10-1 10-2 10-1

Figure S2. Critical exponents ﬁtting in a logarithmic scale. a. The critical exponent βI . b. The critical exponent βJ . c. The critical exponent zν. d. The critical exponent δ obtained by ﬁtting the data of the critical value I = 1.5Γ (orange), compared to standard optical-pumping with I = 0 (blue). e. The. critical exponent γ. 13

Figure S3. Dependence of the spin-response time on the Zeeman coherence. a. Simulated spin-response time with zeroed Zeeman coherences. b. Simulated spin-response time at a small, ﬁnite magnetic ﬁeld of 0.1mG. For the experiment carried at B = 1 G, the Zeeman coherences are relaxed by rapid spin-exchange collisions.