Observation of anomalous diffusion and fractional self-similarity in one dimension

Observation of anomalous diffusion and fractional self-similarity in one dimension


We experimentally study anomalous diffusion of ultra-cold atoms in a one dimensional polarization optical lattice. The atomic spatial distribution is recorded at different times and its dynamics and shape is analyzed. We find that the width of the cloud exhibits a power-law time dependence with an exponent that depends on the lattice depth. Moreover, the distribution exhibits fractional self-similarity with the same characteristic exponent. The self-similar shape of the distribution is found to be well-fitted by a Lévy distribution, but with a characteristic exponent that differs from the temporal one. Numerical simulations suggest that this is due to long trapping times in the lattice and correlations between the atom’s velocity and flight duration.

51.20.+d, 05.40.Fb,37.10.De, 37.10.Gh

Diffusion is a phenomenon encountered in almost every branch of physics. Its ubiquitousness stems from the central limit theorem, which states that a sum of random variables is distributed normally as the number of addends increases. It holds, however, only when the distribution of the variables has a finite variance. When this assumption does not hold, i.e. for heavy-tailed distributions with asymptotic power-law behavior with an exponent with , the sum converges instead to a Lévy distribution . A diffusion process which results in such non-Gaussian spatial distribution is usually regarded as anomalous Bouchaud and Georges (1990). Heavy-tailed distributions are found in many fields, from animals foraging strategies Humphries et al. (2010), to the prices of commodities and stocks Mandelbrot (1963). In physics they emerge in situations where there is no characteristic length scale, such as near phase transitions Ohtsuki and Kawarabayashi (1997), in turbulent flow Solomon et al. (1993), in quantum phase diffusion Geisel et al. (1985) or when a system is out of thermal equilibrium Bardou et al. (2002). In fact, anomalous transport properties are intimately linked to non-linear chaotic dynamics which naturally appears in many physical systems Shlesinger et al. (1993); Argonov and Prants (2007).

A simple diffusion model we have in mind is that of particles in real space, each having a velocity which fluctuates in time due to interaction with a bath. After some time the particles’ position is distributed as . The characteristic width of the ensemble, e.g. the full width at half the maximum (FWHM), usually scales as a power-law . Anomalous diffusion can arise when the distribution of velocities has heavy-tails, and almost always results in Bouchaud and Georges (1990). The theoretical challenge is to connect the microscopic physics to the evolution of the distribution . The experimental challenge is to measure these distributions in a well controlled and isolated environment. In this work we meet this challenge by measuring anomalous diffusion of laser cooled atoms in a polarization optical lattice Dalibard and Cohen-Tannoudji (1989); Castin et al. (1991); Lutz (2003); Marksteiner et al. (1996). In this system the steady state atomic velocity distribution was shown both theoretically and experimentally to follow a power-law, with an exponent that depends on the lattice depth Castin et al. (1991); Lutz (2003); Douglas et al. (2006). It was also predicted that the real space diffusive motion of the atoms in such a lattice is anomalous for a wide range of lattice parameters Marksteiner et al. (1996). The onset of anomalous transport characteristics was observed with a single trapped ion Katori et al. (1997). However, a measurement of the distribution and its anomalous dynamics was not reported to date.

Figure 1: a, The experimental setup as seen from above. The atoms are released from a small crossed dipole trap into a second far-off-resonance trap with a very elongated aspect ratio which confine their motion to a single direction (tube trap). This trap is superimposed with the two counter-propagating Sisyphus lattice beams which induce the diffusive motion. After a chosen time, the atomic density is recorded using absorption imaging technique. Each image gives the column two dimensional density, and we integrate over the radial directions to obtain the axial atomic distribution. b, Atomic distributions after different diffusion times in a deep lattice. Each measurement is repeated 20 times and averaged to improve the signal to noise ratio.

Here we report such a measurement in one dimension with an ensemble of ultra-cold Rb atoms. By setting out with a very small atomic cloud and recording the longitudinal density distribution after different waiting times, we are able to directly measure anomalous dynamics in the lattice. We find that the width of the distribution exhibits a power law time dependence, from which a characteristic exponent can be extracted. The value of this exponent depends on the lattice depth. Furthermore, we show that the density distribution at different times exhibits self-similarity with the same characteristic exponent. The self-similar shape of the distribution is found to be very well fitted by a Lévy function. However, the characteristic exponent extracted from this fit is significantly smaller than the exponent extracted from the dynamics. We investigate this point using classical and quantum Monte-Carlo simulations and find that it originates from correlations between the atom’s velocity and the corresponding time it spends un-trapped in a particular lattice site, combined with heavy-tailed distribution of the durations in which it is trapped by the lattice.

The apparatus is depicted schematically in figure 1a. In each experiment, atoms are prepared in a cigar shaped crossed dipole trap with an aspect ratio of 1:3.9 and a radial oscillation frequency of Hz (for more details regarding the apparatus see Ref. Sagi et al. (2010)). The Sisyphus lattice is created by two counter-propagating lattice beams with identical wavelength and orthogonal linear polarizations Dalibard and Cohen-Tannoudji (1989). These beams originate from an external cavity diode laser whose frequency is locked to an atomic transition and detuned MHz relative to the transition between states and . The atomic cloud has its long axis aligned parallel to the lattice beams. The lattice depth is calculated from the measured beam’s intensity and waist of , and using a saturation intensity of . Before switching on the lattice, the atomic cloud has a temperature of and a maximum phase space density of . To counteract gravitation and improve the signal to noise ratio, the motion of the atoms is confined to the longitudinal direction by superimposing the lattice with a “tube” trap, namely a Gaussian beam with a waist of such that the Rayleigh range is much larger than typical diffusion distances we record. Both the crossed dipole trap and tube trap originate from a single frequency Ytterbium fiber laser at a wavelength of , but their frequencies are shifted relative to each other by more than MHz to prevent standing waves. The power of the tube trap beam is .

The lattice is turned on before the crossed dipole trap is turned off, during which the atoms equilibrate with the lattice. The average rate of photons scattered by each atom from the lattice is , much larger than the initial maximal average elastic collisions rate of . This is important since elastic collisions lead to an undesirable thermal equilibrium. We define as the time at which the crossed dipole trap is switched off and the atoms start diffusing in the lattice. The initial size of the cloud is , much smaller than the typical diffusion distances. We take a series of absorption images after different waiting times, an example of which in a deep lattice is depicted in Figure 1b.

A convenient and useful theoretical framework which describes a broad range of anomalous diffusion processes is the fractional diffusion equation (FDE) Metzler and Klafter (2000):


where is the atomic distribution at time , is the Weyl operator describing a fractional derivative in space and similarly is the fractional time derivative. is a generalized diffusion constant having the dimensions of . For and this equation reduces to a normal diffusion equation. corresponds to long spatial jumps (also referred to as Lévy flights), whereas corresponds to long dwelling times between jumps. The solution for the kernel of this equation can be written in terms of Fox functions Mainardi et al. (2005). A general property of the kernel is its time and space scaling Metzler and Klafter (2000); Mainardi et al. (2005):


where is the reduced kernel function.

Figure 2: A log-log graph of the FWHM squared as a function of the time squared, for different lattice depths. To facilitate the comparison of the slopes, all graphs were shifted in the y-direction to cross at a single point. The linear behavior establishes the power-law time scaling of the width of the atomic distribution. The FWHM is found by fitting the atomic distributions to a general function from which we extract the width. In very shallow lattices (below ) we notice an excess density in the center of the distribution which is probably due to corrugation in the tube trap potential. To avoid systematic errors due to this effect, we exclude in this range of lattice depths the central part of the distribution when fitting and extracting the FWHM.

One conclusion that can be immediately drawn from Eq. 2 is that the typical width of the distribution should scale as , with a dynamical diffusion exponent given by . To test this, we plot in figure 2 on a log-log scale the FWHM extracted from the data as a function of time for different lattice depths. The curves are approximately linear, indicating that indeed the width scales as a power-law in time. Furthermore, the slope of each curve is different, showing that the diffusion exponent depends on the lattice depth, as predicted Castin et al. (1991); Lutz (2003). We fit each of these curves with a line, and from its slope we extract the diffusion exponent. The results are plotted as squares in figure 3, and demonstrate that the whole range of fractional diffusion exponents is accessible in our experiment by changing the lattice beams power. Note that since in the radial direction there is no cooling, atoms are eventually lost from the trap due to spontaneous emission. Owing to the large Rayleigh range of the tube trap, there is almost no mixing between the axial and radial velocity distributions and therefor and therefore the radial loss does not distort the axial spatial distribution. Nevertheless, we use the data only as long as at least of the atoms remain in the trap. Increasing this number to changes only the 3 largest points in figure 3 by up to .

Figure 3: The diffusion exponent as a function of the lattice depth. The exponent is extracted by three different methods; The (blue) squares are obtained by fitting the FWHM scaling to a power-law (the errorbar represents a confidence level). The (red) circles are obtained from the measure of the self-similar transformation. The (black) triangles are found by fitting the distributions taken after with a Lévy function.

Another important conclusion from Eq. 2 is that the distribution should exhibit a self-similar scaling with respect to , regardless of its exact shape. An example of this property is shown in the inset of figure 4. When using the appropriate , all experimental data taken at different times collapse to the same curve when the x and y axes are re-scaled according to Eq. 2. This property can be used to extract the dynamical exponent; we employ an -type measure of the self-similarity:


where are re-scaling of the series of distributions measured at different times, and is their average. In figure 4 we depict this measure as a function of for three lattice depths. For each of these curves there is a single minimum whose position is shifted in accordance with the lattice depth. We interpret the minimum as the most probable value for the diffusion exponent. This value is plotted as a function of the lattice depth in figure 3. The diffusion exponents obtained by the self-similarity method and by fitting the FWHM to a power-law agree to within the uncertainty.

Figure 4: The measure of the self-similarity, , as a function of the diffusion exponent, , for three different lattice depths. Each point is calculated from 13 distributions taken after of diffusion and normalized such that their integral is unity. The inset shows an example of the re-scaling transformation with in a deep lattice.

Up to this point we have analyzed the temporal behavior of the atomic distribution, and now we turn to study its shape. Motivated by the fact that for the solution for the kernel of Eq. 1 is the Lévy stable law Metzler and Klafter (2000); Mainardi et al. (2005), we use the latter as a fitting function. In figure 5 we depict the shape exponent, , extracted from these fits as a function of the diffusion time, for 3 characteristic lattices. In all three lattices the shape exponent converge to an asymptotic value on a timescale of . The inset shows the fits after of diffusion. The Lévy distributions fit the data very well with an average r-square of 0.96 and an uncertainty in the fitted at a confidence level of . The shape exponent is depicted as triangles in figure 3. At very shallow lattices it approaches 2; this is expected since for any finite observation time there are lattices which are too weak for the atoms to reach equilibrium. In deeper lattices, on the other hand, we consistently find that the shape exponent is significantly smaller than the dynamical exponents, and in particular smaller than 1. Similar results are obtained if the tail of the spatial distributions is fitted with a power-law instead of with a Lévy distribution.

Figure 5: The atomic distributions after different diffusion times are fitted with a Lèvy function. The resulting exponent is plotted as a function of the diffusion time for three characteristic lattices. In the inset the distributions after 30ms of diffusion and their corresponding fits are shown.

In order to better understand why the shape exponent is different than the two dynamical exponents, we have written both classical and quantum simulations EPA (). Based on their results we attribute the discrepancy between the exponents to the combination of two factors: in the FDE and correlations between the particle’s velocity and flight duration. The quantum simulation we have performed is a Monte Carlo Wave-Functions (MCWF) of atoms in one dimensional polarization lattice with angular momentum to transitions Mølmer et al. (1993). The atomic evolution in the light field is treated quantum mechanically and photon scattering is described by quantum jumps. In general, we find a good qualitative agreement between experiments and the simulation despite differences in the level structure and lattice detuning which we introduce to simplify the numerics EPA (). The temporal evolution of the simulated spatial distributions, as well as their shape and dependence on the lattice depth, exhibit the same properties that were discussed above. In particular, similarly to figure 3, we find that the spatial Lévy exponent is consistently smaller than the dynamical exponent for deep lattices. The simulations also establish a clear correlation between the velocity and flight-duration of the atoms. In other words, if an atom acquires a large momentum, it is more likely not to be trapped in a single lattice site for longer times. We have also ran classical Monte-Carlo simulations where the velocities, flight-durations and dwelling times in the lattice are all drawn from Levy distributions. Though this simulation shows that a shape exponent smaller than 1 can be obtained by for shallow lattices, it is necessary to include the correlations between the velocity and flight duration in order to sustain this result for increasingly deeper lattices. Both these factors were indeed found in the MCWF simulations EPA ().

To summarize, we have presented measurements and analysis of spatial anomalous diffusion of ultra-cold atoms in a 1D polarization lattice. We find that a complete description of this process goes beyond the FDE and must include the effect of correlations between the motion variables. Future extensions of our work include the study of anomalous diffusion in the presence of an external force and diffusion in very shallow lattices where there is hope to observe super-ballistic diffusion Sagi et al. (2011). Also of great interest is to measure the time-evolution of the velocity distribution which is also expected to exhibit anomalous diffusion Kessler and Barkai (2010); Hirschberg et al. (2011).

We thank Rami Pugatch for stimulating discussions. This work was partially supported by MIDAS, MINERVA, ISF and DIP.


  1. J.-P. Bouchaud and A. Georges, Phys. Rep. 195, 127 (1990).
  2. N. E. Humphries et al.Nature 465, 1066 (2010).
  3. B. Mandelbrot, The Journal of Business 36, 394 (1963).
  4. T. Ohtsuki and T. Kawarabayashi, J. Phys. Soc. Jpn. 66, 314 (1997).
  5. T. H. Solomon, E. R. Weeks,  and H. L. Swinney, Phys. Rev. Lett. 71, 3975 (1993).
  6. T. Geisel, J. Nierwetberg,  and A. Zacherl, Phys. Rev. Lett. 54, 616 (1985).
  7. F. Bardou, J. Bouchaud, A. Aspect,  and C. Cohen-Tannoudji, Lévy Statistics and Laser Cooling (Cambridge University Press, 2002).
  8. M. F. Shlesinger, G. M. Zaslavsky,  and J. Klafter, Nature 363, 31 (1993).
  9. V. Y. Argonov and S. V. Prants, Phys. Rev. A 75, 063428 (2007).
  10. J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B 6, 2023 (1989).
  11. Y. Castin, J. Dalibard,  and C. Cohen-Tannoudji, “Light induced kinetic effects on atoms, ions, and molecules,”  (ETS Editrice, 1991) Chap. The limits of Sisyphus cooling.
  12. E. Lutz, Phys. Rev. A 67, 051402 (2003).
  13. S. Marksteiner, K. Ellinger,  and P. Zoller, Phys. Rev. A 53, 3409 (1996).
  14. P. Douglas, S. Bergamini,  and F. Renzoni, Phys. Rev. Lett. 96, 110601 (2006).
  15. H. Katori, S. Schlipf,  and H. Walther, Physical Review Letters 79, 2221 (1997).
  16. Y. Sagi, R. Pugatch, I. Almog,  and N. Davidson, Phys. Rev. Lett. 104, 253003 (2010).
  17. R. Metzler and J. Klafter, Phys. Rep. 339, 1 (2000).
  18. F. Mainardi, G. Pagnini,  and R. K. Saxena, J. Comp. Appl. Math. 178, 321 (2005).
  19. See supplementary material at http://link.aps.org/supple mental/… for more information regarding the MCWF and classical simulations.
  20. K. Mølmer, Y. Castin,  and J. Dalibard, J. Opt. Soc. Am. B 10, 524 (1993).
  21. Y. Sagi, R. Pugatch, I. Almog, N. Davidson,  and M. Aizenman, Phys. Rev. A 83, 043821 (2011).
  22. D. A. Kessler and E. Barkai, Phys. Rev. Lett. 105, 120602 (2010).
  23. O. Hirschberg, D. Mukamel,  and G. M. Schütz, Phys. Rev. E 84, 041111 (2011).
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description