Report

Observation of many-body dynamics in long-range tunneling after a quantum quench

See allHide authors and affiliations

Science  13 Jun 2014:
Vol. 344, Issue 6189, pp. 1259-1262
DOI: 10.1126/science.1248402

Tilting just right makes atoms tunnel

One of the most fascinating phenomena in the quantum world is the ability of particles to go through an energy barrier — a process called quantum tunneling. Meinert et al. studied the dynamics of quantum tunneling in an optical lattice of strongly interacting atoms. When the lattice was suddenly tilted, the atoms, originally each in their own lattice site, tunneled to non-neighboring sites.

Science, this issue p. 1259

Abstract

Quantum tunneling is at the heart of many low-temperature phenomena. In strongly correlated lattice systems, tunneling is responsible for inducing effective interactions, and long-range tunneling substantially alters many-body properties in and out of equilibrium. We observe resonantly enhanced long-range quantum tunneling in one-dimensional Mott-insulating Hubbard chains that are suddenly quenched into a tilted configuration. Higher-order tunneling processes over up to five lattice sites are observed as resonances in the number of doubly occupied sites when the tilt per site is tuned to integer fractions of the Mott gap. This forms a basis for a controlled study of many-body dynamics driven by higher-order tunneling and demonstrates that when some degrees of freedom are frozen out, phenomena that are driven by small-amplitude tunneling terms can still be observed.

Quantum tunneling is ubiquitous in physics and forms the basis for a multitude of fundamental effects (1) related to electronic transport, nuclear motion, light propagation, and superfluidity in lattice systems (2). Whereas for weakly interacting particles tunneling at a rate J will occur as an individual process for each particle, in strongly interacting systems the behavior of each particle is correlated with the behavior of other particles. Such correlated processes are believed to play an important role, for example, in superconductivity of the cuprate systems (35). Second-order tunneling has been observed in cold atom experiments as driven resonances (6) or directly as a dynamical process for pairs of strongly interacting particles in arrays of double-well potentials (7). That process results in an effective nearest-neighbor super-exchange interaction (8, 9), which forms the basis of important forms of quantum magnetism (10), and provides a starting point for the formation of quantum many-body phases. Such tunneling processes have also recently been observed for electrons in systems of quantum dots (11).

Higher-order processes involving correlated tunneling across multiple lattice sites can give rise to longer-range effective interaction terms and more complex many-body critical phenomena (12), as well as marked changes in out-of equilibrium dynamics. Parallels can be drawn between long-range tunneling processes in tilted lattices and multiphoton electron-positron creation in strong electric fields, with connections to relativistic phenomena such as the Sauter-Schwinger effect in tilted Mott insulators (13), and also to long-distance electron transport in molecular systems, for example (14, 15). However, although single-particle tunneling loss via higher-band resonances (16) has been demonstrated, it has been difficult to observe coherent quantum dynamics due to higher-order tunneling processes because the small amplitude driving these terms places challenging upper limits on the energy scales for required temperatures and allowed disorder.

Our experiment is based on an array of one-dimensional (1D) Mott-insulating “Ising” chains of bosons in an optical lattice near zero temperature (1721). We model the system by a single-band Bose-Hubbard (BH) Hamiltonian (22, 23). For large on-site interaction energy U >> J, the many-body ground state is a Mott insulator with unit occupation at commensurate filling (Fig. 1A). This phase is characterized by exponentially localized atoms and highly suppressed tunneling. In addition, we superimpose a linear gradient potential, which introduces a site-to-site constant energy shift E. We perform a quantum quench to a highly nonequilibrium situation by rapidly tilting the initial Mott state to an integer fraction of the Mott gap EU/n. The quench initiates resonantly enhanced long-range tunneling to the nth neighbor for all sites simultaneously (lower part of Fig. 1 A). For n = 1, one couples to nearest-neighbor dipole states and observes strong coherent oscillations in the number of doubly occupied sites (doublons) with a characteristic frequency 4J (20). For n > 1, resonant tunnel coupling occurs across n – 1 intermediate lattice sites. The process involves up to n other particles, giving rise to occupation-dependent nth-order tunneling, with a characteristic rate of atom-pair formation set by Embedded Image in nth order perturbation theory. Here, αn is a proportionality factor that includes the effect of Bose enhancement (23). Because all particles participate in a tunnel process across n sites, one expects the build-up of massive correlations in the interacting many-body system. As we discuss in (23), the quench onto the critical point in the many-body system results in many-body dephasing of oscillations in the atom-pair number, corresponding to a characteristic growth in many-body entanglement (2426) in our numerical simulations.

Fig. 1 Tunneling resonances in a tilted 1D Mott insulator.

(A) Schematic view of the long-range correlations across n sites for a tilt of E = U/n after the quench from the initial 1D one-atom Mott insulator (top) to the tilted configuration (bottom). Here, n = 3. (B) Number of doublons Nd as a function of E at th = 200 ms after the quench. Here, Vz = 10ER and as = 252(5) a0, giving U = 1.077(20) kHz for Vx,y = 20ER. The solid lines are Lorentzian (for E = U) and Gaussian (for E = U/2 and E = U/3) fits to the data to determine the center positions and widths. The insets show matter-wave interference patterns obtained in TOF at E1 = U, E2 = U/2, and E3 = U/3 taken after th = 1 ms, 9 ms, and 28 ms, respectively. (C to E) Integrated line densities of the TOF images shown in the insets in (B). The solid lines are fits according to double-slit interference patterns with Gaussian envelopes (32). Error bars in all figures reflect ±1 SD.

We prepare an ensemble of 1D Mott insulators (20) starting from a 3D Bose-Einstein condensate (BEC) of typically 8.5 × 104 Cs atoms without detectable uncondensed fraction. The BEC is levitated against gravity by a magnetic field gradient of Embedded Image and initially held in a crossed optical dipole trap (20, 27). We load the sample adiabatically into a cubic 3D optical lattice generated by laser beams at a wavelength of λl = 1064.5 nm, thereby creating a singly occupied 3D Mott insulator for a lattice depth of Vq = 20ER (28) in each direction (q = x, y, z) with less than 4% residual double occupancy. Here, Embedded Image is the photon-recoil energy, with kl = 2π/λl and m the mass of the Cs atom. The optical lattice results in a residual harmonic confinement of νz = 11.9(2) Hz in the z direction of gravity. A broad Feshbach resonance allows us to set the atomic scattering length as, and thus U independently of J, by means of an offset magnetic field B (20).

Tunneling resonances are observed by quickly tilting the lattice in the z direction through a reduction of Embedded Image and then lowering Vz to 10 ER within 1 ms, giving J ≈ 25 Hz (22, 28). All dynamics are now restricted along 1D Mott chains with an average length of 40 sites (20). The chains, in total ≈ 2000, are decoupled from each other on the relevant experimental time scales. We let the systems evolve for a hold time th of up to 200 ms in the tilted configuration and then quickly ramp back Vz to its original value and remove the tilt. The ensemble is characterized by measuring the number of doubly occupied sites Nd through Feshbach molecule formation with an overall efficiency of 80(3)% (20). Alternatively, we detect the emergence of momentum-space coherence in time-of-flight (TOF) by quickly turning off all trapping potentials and allowing for 20 ms of free levitated expansion at as = 0 (27) before taking an absorption image.

The experimental result for a specific choice of U = 1077(0) Hz is shown in Fig. 1B. For a hold time of th = 200 ms, the transient response as discussed below has settled to a steady-state value. Besides a broad resonance at E = 1095(2) Hz with full width at half maximum (FWHM) = 172(9) Hz, two narrower resonances at E = 532(1) and 351(1) Hz with FWHM = 44(2) and 27(2) Hz can be seen. Whereas the broad resonance is the result of resonant tunnel coupling to nearest-neighbor dipole states at E1 = U (20), the positions of the narrower resonances are consistent with E2 = U/2 and E3 = U/3, and we hence interpret them to emerge from tunnel processes extending over a distance of two and three lattice sites, respectively. The reduced widths reflect the smaller amplitude of the higher-order tunnel processes. We believe that the resonances are slightly broadened inhomogeneously by the external harmonic confinement. The assignment of the resonance features to tunneling processes over multiple lattice sites is supported by TOF images (insets to Fig. 1B) taken for each resonance En in the course of the transient response. The images clearly exhibit matter-wave interference patterns, indicating delocalization of the atoms during the tunnel processes. The integrated line densities are presented in Fig. 1, C to E. The periodicity of the sinusoidal density modulation, found to be Embedded Image, is in agreement with spatial coherence of the atomic wave function over a distance of n sites.

We now investigate the transient dynamics after the quantum quench. Figure 2, A and C (B and D), shows the on-resonance response of Nd and the fringe visibility V in the TOF images for E1 (E2). The quench to E1 results in large-amplitude oscillations for Nd; calculations show that the decay is due to many-body dephasing, which plays an increased role for larger chain lengths (20). The oscillatory response at E1 is clearly reflected in the dynamics for V, as each local minimum coincides with an extremum for Nd. The dynamics for E2 are, in contrast, highly overdamped and fit to a saturated growth function of the form Embedded Image, with a characteristic rate 1/τ; a simple three-site BH model predicts oscillations at frequency Embedded Image (23). In the experiment, we find a single maximum for V before it decays. The dephasing here results from more complicated dynamics in BH chains longer than three sites (23).

We now focus on the scaling of the resonant doublon growth rate 1/τ with J and U for the resonance E2. Example data sets (Fig. 3A) clearly demonstrate that 1/τ depends not only on Vz but also on U when Vz and thereby J are kept constant. In Fig. 3B, we plot the same data with the time axis rescaled by the energy scale J 2/(U/2) for a second-order tunneling process. The data collapse onto a single curve, demonstrating that indeed second-order tunneling dominates the transient dynamics after the quench. The numerical data for 10 to 30 site BH chains (23) show similar rise characteristics and reveal the same scaling collapse (Fig. 3 C). The values for 1/τ from measurements taken at different combinations of Vz and U have a linear dependence on J 2/(U/2) (Fig. 3D) with a surprisingly large prefactor α2 = 38(2), which we analyze in two ways. First, we compare to the frequency of coherent doublon oscillations in the simple three-site model. The role of many-body dephasing faster than a full second-order tunneling cycle is estimated by assuming Embedded Image as a quarter of the full tunneling period. The value 1/τ ≈ 4 × ν2 is indicated by the solid line in Fig. 3D. Second, we extract a characteristic growth rate from the numerical data, indicated by the dashed regions in Fig. 3, C and D, revealing quantitative agreement with the experiment.

Fig. 2

Comparison of the tunneling dynamics to nearest and second-nearest neighbors. (A and B) Double occupancy Nd and (C and D) fringe visibility V in the TOF images as a function of hold time th after the quench (symbols). Coherent oscillations in Nd at E1 = U in (A) are contrasted to overdamped dynamics at E2 = U/2 (B) for Vz = 10ER and as = 253(5) a0. The evolution of Nd is fitted (lines) by an exponentially damped sinusoid (A) and a saturated growth (B).The solid lines in (C) and (D) are fits to guide the eye based on the modulus of an algebraically decaying sinusoid.

Fig. 3 Second- and third-order tunneling dynamics.

(A) Double occupancy at E2 for Vz/ER, as/a0 = 10, 253(5) (squares), 12, 253(5) (triangles), and 10, 400(5) (circles). (E) Double occupancy at E3 for Vz/ER, as/a0 = 7,253(5) (squares), 9,253(5) (triangles), and 9,175(5) (circles). The solid lines are fits to the data with saturated growth functions. (B and F) Collapse of the data shown in (A) and (E) for rescaled time axes. (C and G) Result of a numerical simulation of the resonant response at E2 and E3, respectively. (D and H) Growth rates 1/τ for E2 and E3, respectively. In (D), the data for Vz = (8,9,10,12,14)ER with fixed as = 253 a0 (squares) and as = (175,253,325,400) a0 with fixed Vz = 10ER (circles) are plotted as a function of J2/(U/2). The solid line gives the prediction from a three-site BH model. In (H), the data for Vz = (7,8,9,10)ER with as = 253 a0 (squares) and for as = 175 a0 at Vz = 9ER and as = 300 a0 at Vz = 7ER (circles) are plotted as a function of J3/(U/3)2. The dashed line is a linear fit to the experimental data. The shaded areas in (C), (D), (G), and (H) indicate the spread in the growth rate extracted from the numerical data with fixed steady-state values from the experiment (23).

A similar behavior in the dynamical scaling of the resonant response at the resonance E3 = U/3 is seen in Fig. 3, E to H. Scaling collapse is observed when rescaling time by J 3/(U/3)2, indicating a third-order tunneling process. Residual oscillations after the initial growth period in the numerical data (Fig. 3G) relate to the finite system size and the lack of averaging over positions in the trap (23). From the linear fit to the growth rate 1/τ in Fig. 3H, we obtain a slope of α3 = 34(2), in good agreement with a characteristic growth rate determined from the numerical data, which we indicate by the dashed region as before. We note that the signature of the third-order process is not masked by the presence of second-order energy shifts (23).

To what extent can one reverse this many-body dephasing dynamics? In Fig. 4A, we show the result of a many-body echo experiment for which we switch the sign of U and E at the E2 = U/2 resonance in the course of the transient response. A clear, although only partial, reversal in the time evolution for Nd can be seen before Nd reaches the same steady-state value as before. It would be interesting to test whether the revival could be improved by switching the sign of J as well. Naively, the second-order process scaling with J 2 should not depend on the sign of J. Switching J by means of modulation techniques (29) may allow a detailed benchmarking of many-body damping versus the presence of mere inhomogeneous broadening in our system.

Fig. 4 Many-body echo and higher-order tunneling resonances.

(A) Double-occupancy Nd as a function of hold time th at E = U/2 for Vz = 8ER and as = –250(5) a0, giving U = –994(20) Hz (squares). Partial time reversal of the many-body dynamics (circles) occurs after switching as to +250(5) a0 and simultaneously reversing E to –E at th = 6 ms within 1 ms (gray bar). For the echo data (circles), a typical error bar is given for the data point at th = 16 ms. (B) Nd as a function of E after th = 200 ms at Vz = 7ER with U = 959(20) Hz, for as = 252(5) a0. The arrows indicate the expected positions of the tunneling resonances at En = U/n. An additional resonance at 2U/3 appears. The inset gives a fine scan of the U/4 and U/5 resonances. The solid line is a fit based on the sum of multiple Gaussians to guide the eye.

Finally, in Fig. 4B, we show resonances corresponding to many-body tunneling across four and five lattice sites. For these data, the lattice depth was reduced to Vz = 7ER to speed up the processes; the system was initially in the Mott-insulating regime. With decreasing Vz, the resonances at U/2 and U/3 slightly broaden, which we attribute to the increase of the second- and third-order tunneling rates. The new resonances at U/4 and U/5 are clearly detectable. We note that these fourth- and fifth-order tunneling processes greatly benefit from substantial Bose enhancement (23) and speculate that even higher-order processes should become accessible when one eliminates residual parabolic energy shifts due to the trapping laser beams.

Our results underline the utility of cold atoms in optical lattices for the investigation of fundamental physical processes driven by small-amplitude terms and specifically higher-order tunneling. By partly freezing the motion in the deep lattice, these sensitive processes can be observed here despite finite initial temperatures (which here are converted into defects and missing atoms in an ensemble of initial states). This will motivate further investigation of quantum phases and critical properties near these higher-order resonances, which are presently unknown, including systems with tilts along multiple axes (19, 30). Our initial studies of parameter reversals also open the door to the study of many-body dephasing and echo-type experiments on a quantum many-body system, as well as investigations into the nature of the many-body dephasing and (apparent) thermalization (31). Parallels can be drawn with arrays of quantum dots, opening further possibilities to model electron tunneling over multiple sites (11) by using fermionic atoms.

Supplementary Materials

www.sciencemag.org/content/344/6189/1259/suppl/DC1

Supplementary Text

Figs. S1 to S5

References (3337)

References and Notes

  1. Materials, methods, and additional theoretical background are available as supporting material on Science Online.
  2. The lattice depth Vq is calibrated by Kapitza-Dirac diffraction. The statistical error for Vq is 1%, although the systematic error can reach up to 5%. We give all energies in frequency units.
  3. The matter-wave interference pattern is fit by Embedded Image with the fringe visibility V, the wave vector k, and a phase ϕ.
  4. Acknowledgments: We are indebted to R. Grimm for generous support and thank J. Schachenmayer for discussions and contributions to numerical code development. We gratefully acknowledge funding by the European Research Council (ERC) under project no. 278417 and support in Pittsburgh from NSF grant PHY-1148957.

Stay Connected to Science

Navigate This Article