Universal space-time scaling symmetry in the dynamics of bosons across a quantum phase transition

See allHide authors and affiliations

Science  04 Nov 2016:
Vol. 354, Issue 6312, pp. 606-610
DOI: 10.1126/science.aaf9657

Shaking the lattice uncovers universality

Most of our knowledge of quantum phase transitions (QPTs)—which occur as a result of quantum, rather than thermal, fluctuations—comes from experiments performed in equilibrium conditions. Less is known about the dynamics of a system going through a QPT, which have been hypothesized to depend on a single time and length scale. Clark et al. confirmed this hypothesis in a gas of cesium atoms in an optical lattice, which was shaken progressively faster to drive the gas through a QPT.

Science, this issue p. 606


The dynamics of many-body systems spanning condensed matter, cosmology, and beyond are hypothesized to be universal when the systems cross continuous phase transitions. The universal dynamics are expected to satisfy a scaling symmetry of space and time with the crossing rate, inspired by the Kibble-Zurek mechanism. We test this symmetry based on Bose condensates in a shaken optical lattice. Shaking the lattice drives condensates across an effectively ferromagnetic quantum phase transition. After crossing the critical point, the condensates manifest delayed growth of spin fluctuations and develop antiferromagnetic spatial correlations resulting from the sub-Poisson distribution of the spacing between topological defects. The fluctuations and correlations are invariant in scaled space-time coordinates, in support of the scaling symmetry of quantum critical dynamics.

Critical phenomena near a continuous phase transition reveal fascinating connections between seemingly disparate systems that can be described via the same universal principles. Such systems can be found in the contexts of superfluid helium (1), liquid crystals (2), biological cell membranes (3), the early universe (4), and cold atoms (5, 6). An important universal prediction is the power-law scaling of the topological defect density with the rate of crossing a critical point, as first discussed by T. Kibble in cosmology (4) and extended by W. Zurek in the context of condensed matter (1). Their theory, known as the Kibble-Zurek mechanism, has been the subject of intense experimental study that has largely supported the scaling laws (7). Recent theoretical works further propose the so-called universality hypothesis, according to which the collective dynamics across a critical point should be invariant in the space and time coordinates that scale with the Kibble-Zurek power law (810).

Atomic quantum gases provide a clean, well-characterized, and controlled platform for studying critical dynamics (6, 11, 12). They have enabled experiments on the formation of topological defects across the Bose-Einstein condensation transition (1316) as well as critical dynamics across quantum phase transitions (1723). Recent experiments using cold atoms in shaken optical lattices (2426) have provided a vehicle for exploring phase transitions in spin models (2729).

Here we study the critical dynamics of Bose condensates in a shaken optical lattice near an effectively ferromagnetic quantum phase transition. The transition occurs when we ramp the shaking amplitude across a critical value, causing the atomic population to bifurcate into two pseudo-spinor ground states (28). We measure the growth of spin fluctuations and the spatial spin correlations for ramping rates varied over two orders of magnitude. Beyond the critical point, we observe delayed development of ferromagnetic spin domains with long-range antiferromagnetic correlations due to the bunching of the domain sizes, which is not expected in a thermal distribution of ferromagnets. The times and lengths characterizing the critical dynamics agree with the scaling predicted by the Kibble-Zurek mechanism. The measured fluctuations and correlations collapse onto single curves in scaled space and time coordinates, supporting the universality hypothesis.

Our experiments use elongated three-dimensional (3D) Bose-Einstein condensates (BECs) of cesium atoms. We optically confine the condensates with trap frequencies of Embedded Image Hz, where the long (Embedded Image) and short (Embedded Image) axes are oriented at Embedded Image with respect to the Embedded Image and Embedded Image coordinates (Fig. 1A). The tight confinement along the vertical Embedded Image axis suppresses nontrivial dynamics in that direction (see the discussion on the dynamics in the Embedded Image direction below), which is also the optical axis of our imaging system. We adiabatically load the condensates into a 1D optical lattice (11) along the Embedded Image axis with a lattice spacing of Embedded Imagenm and a depth of 8.86 Embedded Image, where Embedded Image kHz is the recoil energy and Embedded Image is Planck’s constant.

Fig. 1 Ferromagnetic quantum phase transition of bosons in a shaken optical lattice.

(A) A BEC of cesium atoms (spheres) in a 1D optical lattice (pink surface) shaking with peak-to-peak amplitude Embedded Image can form ferromagnetic domains (blue and red regions). The elliptical harmonic confinement has principal axes rotated 45° from the lattice. (B) The transition occurs when the ground band evolves from quadratic for Embedded Image [paramagnetic (PM) phase], through quartic at the quantum critical point Embedded Image, to a double well for Embedded Image [ferromagnetic (FM) phase] with two minima at Embedded Image (28). (C) Kibble-Zurek picture. Evolution of the condensate crossing the phase transition becomes diabatic in the frozen regime (cyan) when the time Embedded Image remaining to reach the critical point is less than the relaxation time. Faster ramps cause freezing farther from the critical point, limiting the system to smaller domains. Sample domain images are shown for slow, medium, and fast ramps.

To induce the ferromagnetic quantum phase transition, we modulate the phase of the lattice beam to periodically translate the lattice potential by Embedded Image, where Embedded Image is the shaking amplitude and the modulation frequency Embedded Image is tuned to mix the ground and first excited lattice bands (fig. S1) (28, 30). The hybridized single-particle ground band energy Embedded Image can be modeled for small quasimomentum Embedded Image byEmbedded Image(1)where Embedded Image is the atomic mass, and the coefficients of its quadratic (Embedded Image) and quartic (Embedded Image) terms depend on the shaking amplitude (Fig. 1B). For shaking amplitudes below the critical value, the coefficient Embedded Image is positive and the BEC occupies the lone ground state at momentum Embedded Image. The quantum phase transition occurs when the quadratic term crosses zero at Embedded Image, where Embedded Image and Embedded Image. At this point, the speed of sound for superfluid excitations, formally studied in (31), drops to zero along Embedded Image but remains nearly constant along Embedded Image and Embedded Image. Even stronger shaking converts the ground band into a double well with Embedded Image, yielding two degenerate ground states with Embedded Image. Repulsively interacting bosons with this double-well ground band are effectively ferromagnetic, having two degenerate many-body ground states with all atoms either pseudo-spin up (Embedded Image) or down (Embedded Image) (28). Notably, transitioning to one of these two ground states requires the system to spontaneously break the symmetry of its Hamiltonian. Describing the dynamics across the critical point presents a major challenge because of the divergence of the correlation length of quasimomentum and the relaxation time (critical slowing).

The Kibble-Zurek mechanism provides a powerful insight into quantum critical dynamics. According to this theory, when the time remaining to reach the critical point inevitably becomes shorter than the relaxation time, the system becomes effectively frozen (Fig. 1C). The system only unfreezes at a delay time Embedded Image after passing the critical point, when relaxation becomes faster than the ramp. At this time, topological defects become visible, and the typical distance Embedded Image between neighboring defects is proportional to the equilibrium correlation length. The Kibble-Zurek mechanism predicts that Embedded Image and Embedded Image depend on the quench rate Embedded Image asEmbedded Image(2)Embedded Image(3)where Embedded Image and Embedded Image are the equilibrium dynamical and correlation length exponents given by the universality class of the phase transition. Although the details of this picture may not apply to every phase transition, the general scaling arguments are very robust, and similar predictions hold for a variety of quench types across the transition (12) and for phase transitions that break either continuous or discrete symmetries (7).

For slow ramps, Embedded Image and Embedded Image diverge and become separated from other scales in the system, making them the dominant scales for characterizing the collective critical dynamics (810). This idea motivates the universality hypothesis, which can be expressed asEmbedded Image(4)indicating that the critical dynamics of any collective observable Embedded Image obeys the scaling symmetry and can be described by a universal function Embedded Image of the scaled coordinates Embedded Image and Embedded Image. The only effect of the quench rate is to modify the length and time scales.

We test the scaling symmetry of time by monitoring the emergence of quasimomentum fluctuations at different quench rates. Here, fluctuations refer to deviations of quasimomentum from zero, which vary across space and between individual samples; fluctuations should saturate to a large value when domains having Embedded Image are fully formed. After loading the condensates into the lattice, we ramp the shaking amplitude linearly from Embedded Image to values well above the critical amplitude Embedded Image nm (32) and interrupt the ramps at various times to perform a brief time-of-flight (TOF) before detection. After TOF, we measure the density deviation Embedded Image (32), which is nearly proportional to the quasimomentum distribution (fig. S2), where Embedded Image is the density profile of the Embedded Imageth Bragg peak and the angle brackets denote averaging over multiple images. This detection method is particularly sensitive near the critical point when the quasimomentum just starts deviating from zero, indicating the emergence of fluctuations in the ferromagnetic phase where the ground states have nonzero quasimomentum. The spin density measurement used later to study spatial correlations is viable only when atoms settle to Embedded Image.

Over a wide range of quench rates, the evolution of quasimomentum fluctuation can be described in three phases (Fig. 2A). First, below the critical point, quasimomentum fluctuation does not exceed its baseline level. Second, just after passing the critical point, critical slowing keeps the system “frozen,” and fluctuation remains low. Finally, the system unfreezes and quasimomentum fluctuation quickly increases and saturates, indicating the emergence of ferromagnetic domains. We quantify this progression by investigating the fluctuation of contrast Embedded Image (Fig. 2B) that tracks quasimomentum fluctuation in our condensates via the fluctuation of Embedded Image (fig. S2), where Embedded Image is the total density and the angle brackets denote averaging over space and over multiple images. For comparison between different quench rates, we calculate the normalized fluctuation Embedded Image, where subscripts i and f indicate the fluctuation at early and late times, respectively (32). We find empirically that the growth of normalized fluctuations is well fit by the function Embedded Image(5)where the time Embedded Image is defined relative to when the system crosses the critical point at Embedded Image, Embedded Image characterizes the delay time when the system unfreezes, and Embedded Image is the formation time over which the fluctuation grows. The measurement of fluctuation over time provides a critical test for both the Kibble-Zurek scaling and the universality hypothesis. First, both Embedded Image and Embedded Image exhibit clear power-law scaling with the quench rate Embedded Image varied over more than two orders of magnitude (Fig. 2C). Power-law fits yield the exponents of Embedded Image and Embedded Image, respectively. The nearly equal exponents are suggestive of the universality hypothesis, which requires all times to scale identically. Indeed, the growth of contrast fluctuation Embedded Image follows a universal curve when time is scaled by Embedded Image (Fig. 2D), strongly supporting the universality hypothesis (Eq. 4). Note that any observable time characterizing the collective dynamics can be chosen as Embedded Image in Eq. 4, including Embedded Image and Embedded Image.

Fig. 2 Growth of quasimomentum fluctuation in quantum critical dynamics.

(A) Sample images show the emergence of nonzero local quasimomentum via the density deviation between Bragg peaks Embedded Image as the system is linearly ramped across the ferromagnetic phase transition; four ramps with increasing quench rates (bottom to top) are shown. Each ramp exhibits three regimes: a subcritical regime before the transition; a frozen regime beyond the critical point where the fluctuation remains low; and a growth regime in which fluctuation increases and saturates, indicating domain formation. Time Embedded Image corresponds to the moment when the system reaches the critical point. (B) Quasimomentum fluctuation for ramp rates Embedded Image 3.6 (triangles), 0.91 (squares), 0.23 (diamonds), and 0.06 nm/ms (circles) arises at a delay time Embedded Image over a formation time Embedded Image. Fluctuation is normalized for each ramp rate to aid comparison (32). The solid curves show fits based on Eq. 5. (C) The dependence of Embedded Image (circles) and Embedded Image (squares) on the quench rate is well fit by power laws (solid curves) with scaling exponents of Embedded Image and Embedded Image, respectively. The inset shows Embedded Image on a linear scale. (D) Fluctuations measured for 16 ramp rates from 0.06 to 10.3 nm/ms collapse to a single curve when time is scaled by Embedded Image based on the power-law fit. The solid curve shows the best fit based on the empirical function (Eq. 5), and the gray shaded region covers one standard deviation. Error bars in (B) and (C) indicate one standard error.

We next test the spatial scaling symmetry based on the structures of pseudo-spin domains that emerge after the system unfreezes. Here, we cross the critical point with two different protocols: The first is a linear ramp starting from Embedded Image, whereas the second begins with a jump to Embedded Image, followed by a linear ramp. We detect domains near the time Embedded Image in the spin density distribution Embedded Image based on the density Embedded Image of atoms with spin up/down (fig. S3). At this time, the spin domains are fully formed and clearly separated by topological defects (domain walls), as shown in Fig. 3A. Furthermore, choosing this time just after domain formation minimizes the time available for nonuniversal relaxation processes. We characterize the domain distribution with the spin correlation function (17, 28) Embedded Image(6)averaged over multiple images (Fig. 3B). Both ramping protocols lead to similar correlation functions, suggesting that the domain distribution is insensitive to increases in the quench rate below the critical point.

Fig. 3 Antiferromagnetic spatial correlations from quantum critical dynamics.

(A) Two sample images at each quench rate exemplify spin domains measured near the time Embedded Image after crossing the phase transition. These images correspond to linear ramps starting at Embedded Image (Embedded Image to Embedded Image nm/ms) and Embedded Image (Embedded Image to Embedded Image nm/ms). (B) Spin correlation functions Embedded Image (Eq. 6) are calculated from ensembles of 110 to 200 images. (C) Cuts across the density-weighted correlation functions Embedded Image are shown for quench rates Embedded Image 1.28 (triangles), 0.45 (squares), 0.16 (diamonds), and 0.056 nm/ms (circles). Solid curves interpolate the data to guide the eye. The typical domain size Embedded Image and the correlation length Embedded Image are illustrated for 0.056 nm/ms by the arrow and dashed envelope, respectively (32). (D) The dependence of Embedded Image (green) and Embedded Image (black) on the quench rate is well fit by power laws (Eq. 3) with spatial scaling exponents of Embedded Image and Embedded Image, respectively. Marker shape indicates linear ramps starting at Embedded Image (squares) or at Embedded Image (circles). The inset shows the results on a linear scale. Error bars indicate 1 SE. (E) Correlation functions for Embedded Image to Embedded Image nm/ms collapse to a single curve when distance is scaled by the domain size extracted from the power-law fit in (D). The solid curve shows the fit based on Eq. 7; the gray shaded area covers 1 SD. (F) The temporal scaling exponents Embedded Image and Embedded Image from Fig. 2C (magenta) and the spatial scaling exponents Embedded Image and Embedded Image from (D) (green) constrain the critical exponents Embedded Image and Embedded Image according to Eqs. 2 and 3 with Embedded Image (dark) and Embedded Image (light) confidence intervals. The cross marks the best values with contours of Embedded Image and Embedded Image overall confidence (32).

The spin correlations reveal rich domain structure that strongly depends on the quench rate. For slower ramps Embedded Image nm/ms, the structures are predominantly one-dimensional and the density of topological defects increases with the quench rate. The tighter confinement and finite speed of sound near the critical point along the Embedded Image and Embedded Imageaxes allow spin correlations to span the gas in those directions. The dynamics thus appear one-dimensional. When the quench rate exceeds Embedded Image nm/ms, defects start appearing along the Embedded Imageaxis, and the domain structures become multidimensional. We attribute this dimensional crossover to the unfreezing time becoming too short to establish correlation along the Embedded Image axis. For the remainder of this work, we focus on the slower quenches and investigate the spin correlations along the Embedded Image axis.

We examine the 1D correlations using line cuts of the density-weighted correlation functions Embedded Image (17, 28). The results exhibit prominent decaying oscillation (Fig. 3C). We extract two essential length scales from the correlation functions: the average domain size Embedded Image, or equivalently, the distance between neighboring topological defects, and the correlation length Embedded Image, indicating the width of the envelope function. These two scales are determined from the position and width of the peak in the Fourier transform of Embedded Image (32).

These length scales enable us to test the spatial scaling symmetry. The lengths Embedded Image and Embedded Image both display power-law scaling consistent with the Kibble-Zurek mechanism (Fig. 3D), with fits yielding exponents Embedded Image for the domain size and Embedded Image for the correlation length. Similarly, the correlations, measured at the same scaled time, collapse to a single curve in spatial coordinates scaled by the domain size Embedded Image (Fig. 3E). This result strongly supports the spatiotemporal scaling from the universality hypothesis (Eq. 4). An empirical curveEmbedded Image(7)provides a good fit to the universal correlation function, yielding Embedded Image = 1.01(1) and Embedded Image = 1.04(1), indicating that the width of the envelope is close to the typical domain size.

The most striking feature of the universal correlation function is the emergence of oscillatory, antiferromagnetic order in the ferromagnetic phase. In thermal equilibrium, ferromagnets are expected to have a finite correlation length but no anticorrelation. The appearance of strong anticorrelation at Embedded Image suggests that domains of size Embedded Image form preferentially during the quantum critical dynamics. A statistical analysis of the topological defect distribution reveals that the domain sizes are bunched with their standard deviation Embedded Image well below their mean, indicating that the topological defects are created by a sub-Poisson process (fig. S4).

Finally, the combined scaling exponents of space and time allow us to extract the equilibrium critical exponents based on the Kibble-Zurek mechanism (33) (Fig. 3F). Solving Eqs. 2 and 3, we obtain the dynamical exponent Embedded Image 1.9(2) and correlation length exponent Embedded Image 0.52(5), which are close to the mean-field values Embedded Image and Embedded Image up to our experimental uncertainty. Note that the dynamical critical exponent Embedded Image results from the unique quartic kinetic energy Embedded Image of our system at the critical point (32).

Direct identification of domain walls presents intriguing possibilities for future studies of the topological defects generated during critical dynamics. These opportunities would be particularly interesting if the shaking technique were extended to higher dimensions in such a way that the transition breaks a continuous symmetry. In addition, the scaling of the correlation functions suggests that the antiferromagnetic order may be a shared feature of quantum critical dynamics for phase transitions in the same universality class, meriting future experiments.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

References (34, 35)

References and Notes

  1. Materials and methods are available on Science Online.
Acknowledgments: We thank L.-C. Ha, C. V. Parker, B. M. Anderson, B. J. DeSalvo, A. Polkovnikov, S. Sachdev, and Q. Zhou for helpful discussions. L.W.C. was supported by a National Defense Science and Engineering Graduate Fellowship and a Nambu Fellowship. This work was supported by NSF Materials Research Science and Engineering Centers (DMR-1420709), NSF grant PHY-1511696, and Army Research Office–Multidisciplinary Research Initiative grant W911NF-14-1-0003. The data presented in this paper are available upon request to C.C. (

Stay Connected to Science

Navigate This Article