WorksPhysics and Astronomy
Pseudogap effects in the strongly correlated regime of the two-dimensional Fermi gas
Abstract
The two-species Fermi gas with attractive short-range interactions in two spatial dimensions provides a paradigmatic system for the understanding of strongly correlated Fermi superfluids in two dimensions. It is known to exhibit a BEC-BCS crossover as a function of , where is the scattering length, and to undergo a Berezinskii-Kosterlitz-Thouless superfluid transition below a critical temperature . However, the extent of a pseudogap regime in the strongly correlated regime of , in which pairing correlations persist above , remains largely unexplored with controlled theoretical methods. Here we use finite-temperature auxiliary-field quantum Monte Carlo (AFMC) methods on discrete lattices in the canonical ensemble formalism to calculate thermodynamical observables in the strongly correlated regime. We extrapolate to continuous time and the continuum limit to eliminate systematic errors and present results for particle numbers ranging from to . We estimate by a finite-size scaling analysis, and observe clear pseudogap signatures above and below a temperature in both the spin susceptibility and free-energy gap. We also present results for the contact, a fundamental thermodynamic property of quantum many-body systems with short-range interactions.
Introduction.— Cold atomic Fermi gases are of great interest in diverse areas of physics in part because they provide a well-defined paradigm of strongly correlated Fermi superfluids. They have been the subject of intensive experimental and theoretical studies. Of particular interest is the two-species uniform Fermi gas with attractive short-range interactions, whose strength can be controlled experimentally through a Feshbach resonance.
The interaction strength is characterized by the two-particle -wave scattering length . In three spatial dimensions (3D), this system makes a crossover at low temperatures from a Bose-Einstein condensate (BEC) regime of weakly interacting dimers for to a Bardeen-Cooper-Schrieffer (BCS) regime at ( is the Fermi wavenumber). The physics of the interacting Fermi gas in two spatial dimensions (2D) differs qualitatively from that in 3D. Unlike the 3D case, where a two-particle bound state is formed when the interaction is sufficiently attractive, there is a bound state for arbitrarily weak attractive interactions [1, 2]. There is still a BEC to BCS crossover [3] as a function of the scattering parameter [4], with the BEC and BCS limits corresponding to and , respectively.
While both the 2D and 3D systems undergo a superfluid transition below a critical temperature , in 2D this phase transition does not have a non-vanishing condensate fraction with off-diagonal long-range order as in the 3D case, but instead exhibits a quasi-long-range order with algebraic decay of correlations in the superfluid regime. This 2D superfluid transition is known as a Berezinskii-Kosterlitz-Thouless (BKT) transition [5, 6, 7].
The 2D BEC-BCS crossover and BKT transition have been intensively studied both experimentally [8, 9, 10] [11, 12, 13, 14, 15, 16, 17] and theoretically [18–37]. The strongly correlated regime, , of the 2D BEC-BCS crossover presents a particular theoretical challenge. In this regime there is no controlled analytic approach, and controlled computational methods provide the most reliable results. At zero temperature, the diffusion Monte Carlo studies of Ref. [3] calculated the energy, the pairing gap, and the contact, and Refs. [23–25, 33] addressed pairing correlations. At finite temperature, lattice auxiliary-field Monte Carlo (AFMC) methods were used in Ref. [21] to calculate the pressure, compressibility, and contact of the 2D crossover but without a continuum extrapolation. The BKT critical temperature was recently calculated in Ref. [37] as a function of using AFMC methods on large lattices in the grand-canonical ensemble with a continuum extrapolation.
An open problem is the extent of a pseudogap regime, in which signatures of pairing correlations survive above the critical temperature for superfluidity. Measurements of the spectral function of a harmonically trapped gas [38] indicate a pairing gap at temperatures well above the critical temperature in the strongly correlated regime. More recent radio-frequency spectroscopy experiments in the normal phase revealed the presence of an energy gap in the spectrum, which, at , far exceeds the two-body binding energy [17]. In Ref. [20], the pseudogap regime was studied by calculating the single-particle spectral function using the Luttinger-Ward self-consistent field theory approach. A pronounced depression in the single-particle density of states was found in the normal phase of the strongly correlated regime, whereas the same method applied in 3D to the unitary Fermi gas showed a substantially reduced pseudogap signature [39–41]. While this self-consistent -matrix method compares remarkably well to AFMC for the unitary Fermi gas [42–45], it is nevertheless an uncontrolled approximation and its accuracy in 2D has not been established.
In this work, we calculate thermodynamic properties of the 2D Fermi gas across the superfluid transition in its strongly correlated regime and, in particular, explore pseudogap effects above . We employ lattice auxiliary-field quantum Monte Carlo (AFMC) methods in the canonical ensemble [43, 46, 47] and extrapolate to continuous time and the continuum limit, thus eliminating any systematic errors and yielding results that are accurate up to statistical errors for a given particle number . This approach has the advantage of being a well-controlled computational method.
We estimate the critical temperature of the BKT transition using a finite-size scaling analysis of the largest eigenvalue of the two-body density matrix. We identify two pseudogap signatures above the critical temperature: (i) the suppression of the spin susceptibility above and below a temperature (also called the spin gap), and (ii) the increase in a model-independent free-energy gap with decreasing temperature within the spin gap regime. The calculation of the free-energy gap requires the use of canonical-ensemble AFMC and particle-reprojection [48, 49] methods. We find that the pseudogap regime is broad at a coupling of and its width decreases with the coupling as the system approaches its BCS regime. We also calculate Tan’s contact, a fundamental quantity for quantum many-body systems with short-range interactions that describes the short-distance pair correlations of opposite spin particles [50, 51]. We find that the contact increases as the temperature decreases within the pseudogap regime.
Finite-temperature canonical ensemble AFMC.— We briefly describe the auxiliary-field quantum Monte Carlo (AFMC) method we use in this work; for recent reviews see Refs. [46, 52]. We consider a system of spin 1/2 fermions interacting with a contact interaction of strength on a finite area with periodic boundary conditions. We discretize our system on a lattice of size with points in each direction and a lattice spacing . The corresponding lattice Hamiltonian is then
where and are creation and annihilation operators for fermions with momentum and spin projection and for a quadratic dispersion. Here is the coupling strength, chosen to reproduce the physical scattering length on a lattice, and is the density operator of the fermions at position and spin projection . The first sum on the r.h.s. of Eq. (1) is taken over the complete first Brillouin zone, and the second sum is taken over all lattice sites .
We divide the inverse temperature into time slices of length , and apply a symmetric Trotter-Suzuki decomposition to the imaginary-time propagator
where is the kinetic energy and is the interaction term in the Hamiltonian (1). Rewriting , we use a Hubbard-Stratonovich transformation to decouple the two-body interaction introducing auxiliary fields at each lattice site and time slice . For , the thermal propagator can then be written in the form
where is the integration measure, is a Gaussian weight, and is the time-ordered one-body thermal propagator for a given auxiliary-field configuration, where is the corresponding one-body Hamiltonian at any given time slice . Each integral over a continuous auxiliary field is discretized by a three-point Gaussian quadrature [53].
In canonical-ensemble AFMC we are interested in the calculation of thermal observables at fixed particle numbers
where is the expectation value of for a given field configuration. The traces at fixed particle numbers are calculated by using projection operators , i.e., . For a finite number of single-particle states (for given spin projection ), the particle-number projection can be represented exactly by a discrete finite Fourier sum [54]
where are quadrature points, and is a real chemical potential introduced to stabilize the Fourier sum. Using Eq. (4), the fixed particle number traces on the r.h.s. of Eq. (3) reduce the calculation of the unrestricted traces involving the propagator . Since the latter is a one-body propagator, these traces can be expressed in terms of the single-particle representation matrix of the many-particle propagator .
In canonical-ensemble AFMC, we sample auxiliary-field configurations according to the positive-definite distribution and use them to estimate the thermal canonical expectation values of observables following Eq. (3).
Figure 1. Spin susceptibility (top row) and free energy gap (bottom row) vs. temperature for couplings of (left column), (middle column), and (right column) for particle numbers and . The black dashed line with the gray band is , and the dotted-dashed line is . is measured in units of the free gas zero-temperature susceptibility and is measured in units of the free Fermi gas energy . The insets in the bottom row show the free energy gap on a logarithmic scale (see text).
We take the continuous time limit by performing calculations for several values of , and extrapolation in . We then take the continuum limit by performing calculations for a range of lattices containing as many as points, and performing a linear extrapolation in the filling factor (see the Supplemental Material for detailed fits). The continuum limit requires large lattice calculations, which were made possible through the use of two algorithms: (i) a stable diagonalization method that reduces the scaling of the stabilization of canonical-ensemble AFMC to [55], and (ii) a controlled truncation of the single-particle model space that further reduces the scaling to essentially for most parts of the algorithm [56], offering a dramatic improvement for . A similar method was implemented in Ref. [57].
Results.— We present canonical-ensemble AFMC results for three different couplings in the strongly correlated regime of the 2D BEC-BCS crossover: , , and . For each of these couplings, we carried out calculations for , , , , and particles.
We estimate the critical temperature of the transition to superfluidity for a given using the phenomenological finite-size scaling approach of Refs. [58, 59] which
is suitable for the BKT universality class. In this approach we scale the largest eigenvalue of the two-body density matrix . For details, see the Supplemental Material [60].
We find at , at , and at . These estimates agree within error bars with the results of Ref. [37] obtained by a different method. They are lower than the experimental estimates of in Ref. [11].
(i) Spin susceptibility: We calculate the spin susceptibility using
In the calculation of , we project only on the total number of particles , as opposed to the two-species projection used for other observables.
The spin susceptibility is suppressed by pairing correlations. Our AFMC results for in units of , the zero-temperature susceptibility of the non-interacting Fermi gas, are shown in the top row of Fig. 1 as a function of . We observe the suppression of as the temperature is decreased below . However, we also see moderate suppression of above but below a temperature scale , known as a spin gap. We define to be the temperature at which reaches 95% of its maximal value. This spin gap regime becomes narrower with increasing . We estimate , and for , and , respectively.
(ii) Free energy gap: We define the free energy gap by
where is the free energy for the system at spin-up particles and spin-down particles. To calculate , we rewrite it as
where is the partition function for spin-up particles and spin-down particles. Each partition function ratio can then be calculated using particle-number reprojection [48, 49], e.g.,
where we have introduced the notation with a positive definite weight function for .
We show our results for as a function of in the bottom row of Fig. 1. Using a linear extrapolation at low temperatures, we determine the zero-temperature energy staggering pairing gap [61, 62] from the finite-temperature . For additional details, see the Supplemental Material [60]. A direct AFMC calculation of is challenging due to a sign problem introduced by the spin imbalance.
We observe that the free energy gap is suppressed above . To see more closely the behavior of above , we show in the insets using a logarithmic scale. In the spin gap regime between and we observe an increase in with decreasing temperature due to pairing correlations. We identify this behavior of as a pseudogap signature that correlates with the suppression of the spin susceptibility .
(iii) Contact: The contact describes the short-range correlations between particles of opposite spin and is defined by
where is the two-body correlation function with and the density of spin up and spin down particles, respectively. The contact is of key interest due to Tan’s relations [50, 51], which relate the contact to several different properties of the interacting Fermi gas. In particular, the contact characterizes the tail of the momentum distribution in the limit . It can also be obtained from the
Figure 2. The contact (in units of ) vs. temperature for couplings of , , and for particle numbers , and . The dashed black line with the gray band represents , and the dotted-dashed line is .
derivative of the thermal energy with respect to the coupling parameter
In the lattice simulations we calculate from the thermal expectation value of the potential energy [63]
We show our results for the contact (in units of ) as a function of temperature for the three coupling values in Fig. 2. The contact increases rapidly with lowering in the vicinity of . This increase occurs within a narrower temperature range as the coupling parameter increases. In general, the contact decreases with increasing . In the pseudogap regime between and , we observe a monotonic increase in the contact with decreasing .
Conclusion and outlook.— We explored thermal properties of the two-species spin-balanced Fermi gas with short-range interactions in two spatial dimensions in the strongly correlated regime . We used canonical-ensemble AFMC methods on discrete lattices and extrapolated our results to continuous time and the continuum limit. We estimated the critical temperature for the superfluid transition and calculated several thermodynamic observables across the superfluid phase transition for several values of . We observed pseudogap signatures in a temperature regime , including a suppression of the spin susceptibility and an increase in the free energy gap with decreasing temperature. Our AFMC results for the spin susceptibility and free-energy gap are the first controlled calculation of observables that probe the pseudogap regime in two spatial dimensions and thus provide an accurate benchmark for experiment. We also calculated Tan’s contact and found it to increase monotonically with decreasing temperature in the pseudogap regime. In future AFMC studies, it will be interesting to calculate the spectral function of the 2D strongly interacting Fermi gas across the BEC-BCS crossover as a dynamical probe of the pseudogap regime.
Acknowledgements.— This work was supported in part by the U.S. DOE grants No. DE-SC0019521 and No. DE-SC0020177. The calculations used resources of the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility operated under Contract No. DE-AC02-05CH11231. We thank the Yale Center for Research Computing for guidance and use of the research computing infrastructure.
References
We discuss a number of technical details including the continuum extrapolations and finite-size scaling. We present additional results for the zero-temperature pairing gap, the thermal energy, and the single-particle momentum distribution.
Extrapolations
The lattice AFMC calculations are carried out by dividing into time slices of finite length and for discrete lattices, and it is necessary to take the continuous time limit and the continuum limit (where is the filling factor). In the following we demonstrate how these limits are calculated.
Continuous time limit
There are two systematic errors in . One arises from the symmetric Trotter-Suzuki decomposition of which is of order , and a second arises from using a three-point Gaussian quadrature formula to evaluate the integral over each auxiliary field, which is accurate to order . We eliminate these systematic errors by performing a linear extrapolation in to find the limit . We demonstrate these extrapolations for the contact in Fig. 1. We note that sufficiently small values of are necessary in order to reach the quadratic regime.
For the the maximal eigenvalue of the two-body density matrix and the thermal energy, we find only a weak dependence on for small values of , in which case we perform a flat extrapolation (i.e., take an average value).
Continuum extrapolations
The continuum limit is obtained by taking the limit , and is achieved by considering increasing lattice sizes. Thermal observables behave linearly in for small values of and we use linear extrapolations to extract the values of observables in the limit . Examples of the linear extrapolations in are shown in Fig. 2 for the contact. We find that this observable has a relatively strong dependence on . In Figure 3, we show the contact (in units of ) for and a coupling of as a function of for several lattice sizes . The extrapolated contact in the limit (black solid circles) show significant differences with the finite filling factor results even on a qualitative level. In particular, the contact in the continuum limit above the critical temperature decreases with increasing temperature. In contrast, the spin susceptibility (not shown) exhibits only a weak dependence on .
Figure 3. The contact versus for different finite lattice sizes ranging from () to (). The extrapolated continuum results at are shown by the black solid circles. We notice a strong dependence of the contact on the filling factor.
Finite-size scaling
We estimate the critical temperature of the transition to superfluidity through a finite-size scaling analysis of the largest eigenvalue of the two-body density matrix . We use the phenomenological renormalization group analysis of Refs. [1, 2]. In the ordered phase scales as Eq. (13)
where is the linear size of the system, is the correlation length, and the exponent at [4]. For a BKT transition, the correlation length for , and we expect the scaled curves across different values of to merge for [5] (rather than intersecting in the manner typical of the three-dimensional superfluid phase transition [1]). At constant density, the particle number , and we scale according to for systems with different particle number .
We demonstrate this finite-size scaling analysis in Figure 4, where is shown as a function of for several values of at the three coupling values. We observe the merging of the different curves in the critical regime and use it to estimate and its associated error. We find at , at , and at .
Figure 4. (see text) scaled by as a function of at the three coupling values. We observe the merging of curves with varying particle number in the critical region, marked by the gray bands with a black dashed line at the critical temperature . The gray bands are used to estimate the errors in .
Zero-temperature pairing gap
At the free energy gap coincides with the energy staggering pairing gap. The statistical errors on the free-energy gap are relatively small and at low temperatures we can use a linear extrapolation to determine the energy gap at . In Figure 5 we compare our results for (here is the two-particle binding energy with being the Euler-Mascheroni constant) for several values of the particle number with the diffusion Monte Carlo results of Ref. [6] (solid squares), Ref. [7] (x’s) and Ref. [8] (solid diamonds).
Figure 5. (see text) as a function of the coupling parameter . Our results, based on extrapolations of the AFMC free energy gaps to , are shown by the green, red and blue symbols for , 90 and 114, respectively. The solid squares, x’s and diamonds are, respectively, the diffusion Monte Carlo results of Refs. [6, 7, 8].
Figure 1 (PDF p. 8). The contact (in units of ) versus the dimensionless parameter for a coupling parameter of and particles. The dashed lines are linear extrapolations in to find the limit . Only the points in red below a cutoff of are included in the fit, and the extrapolated results for are shown in black solid circles.
Figure 2 (PDF p. 8). The contact (in units of ) versus the filling factor for (left column), (middle column), and (right column) for various temperatures and particle numbers. The dashed lines are linear extrapolations in to determine the continuum limit at .
Thermal energy
In Figure 6 we show our results for the thermal energy, defined as the expectation value of the Hamiltonian, as
Figure 6. Thermal energy versus for particle numbers ranging from 42 to 162 and for coupling parameters of , , and . We notice that the energy below increases with increasing , in agreement with Ref. [6].
a function of for , 74, 90, 114 and 162. The thermal energy increases linearly with temperature above the critical temperature, and depends weakly on temperature below the critical temperature. Our results are consistent with the ground-state energies calculated in Ref. [6].
Single-particle momentum distribution
Our AFMC results for the single-particle momentum distribution, , are shown in the top row of Figure 7 for and . Of particular interest is the tail of the momentum distribution. According to Tan’s universality relations [9, 10], at large , where is the contact. In the bottom row of Figure 7, we compare with the value of extracted from the average potential energy (dashed lines) and find good agreement for the larger lattices. It is difficult to take the continuum limit extrapolation of the momentum distribution tail because of edge effects. Thus, calculating the contact from is a more effective method of obtaining the continuum results for the contact.
Figure 7. Left column: single-particle momentum distribution for various couplings and lattice sizes at and . Right column: as in the left column but showing versus . The black dashed line is the continuum limit contact obtained from .
References
References
- [3]A. Moreo and D. J. Scalapino, Phys. Rev. Lett. 66, 946 (1991).
- [4]J. M. Kosterlitz and D. J. Thouless, Journal of Physics C: Solid State Physics 6, 1181 (1973).
- [5]T. Paiva, R. R. dos Santos, R. T. Scalettar, and P. J. H. Denteneer, Phys. Rev. B 69, 184501 (2004).
- [6]G. Bertaina and S. Giorgini, Phys. Rev. Lett. 106, 110403 (2011).
- [7]A. Galea, H. Dawkins, S. Gandolfi, and A. Gezerlis, Phys. Rev. A 93, 023602 (2016).
- [8]E. Vitali, H. Shi, M. Qin, and S. Zhang, Phys. Rev. A 96, 061601 (2017).
- [9]S. Tan, Annals of Physics 323, 2952 (2008).
- [10]F. Werner and Y. Castin, Phys. Rev. A 86, 013626 (2012).








