PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of scirepAboutEditorial BoardFor AuthorsScientific Reports
 
Sci Rep. 2016; 6: 28362.
Published online 2016 June 21. doi:  10.1038/srep28362
PMCID: PMC4914841

An interpolatory ansatz captures the physics of one-dimensional confined Fermi systems

Abstract

Interacting one-dimensional quantum systems play a pivotal role in physics. Exact solutions can be obtained for the homogeneous case using the Bethe ansatz and bosonisation techniques. However, these approaches are not applicable when external confinement is present. Recent theoretical advances beyond the Bethe ansatz and bosonisation allow us to predict the behaviour of one-dimensional confined systems with strong short-range interactions, and new experiments with cold atomic Fermi gases have already confirmed these theories. Here we demonstrate that a simple linear combination of the strongly interacting solution with the well-known solution in the limit of vanishing interactions provides a simple and accurate description of the system for all values of the interaction strength. This indicates that one can indeed capture the physics of confined one-dimensional systems by knowledge of the limits using wave functions that are much easier to handle than the output of typical numerical approaches. We demonstrate our scheme for experimentally relevant systems with up to six particles. Moreover, we show that our method works also in the case of mixed systems of particles with different masses. This is an important feature because these systems are known to be non-integrable and thus not solvable by the Bethe ansatz technique.

Understanding the properties of low-dimensional systems is not merely an academic pursuit. Technologically promising systems such as nanotubes, nanowires, and organic conductors have one-dimensional nature1,2,3, while there is much evidence that high-temperature superconductors owe their spectacular properties to an effective two-dimensional structure4,5. However, in the case of interacting particles in one dimension there are still many outstanding fundamental issues in their quantum mechanical description. An avenue within which these problems can be studied is that of cold atomic gases6,7 where experiments in one-dimensional (1D) confinement can be performed with tunable interactions for systems of bosons8,9,10,11,12,13,14 or fermions15. Most recently, 1D Fermi systems have been constructed with full control over the particle number16 and thus engineered few-body systems are now available. This provides opportunities to study pairing, impurity physics, magnetism, and strongly interacting particles from the bottom up17,18,19,20,21.

The role of strong interactions in important quantum phenomena such as superconductivity and magnetism drives research into the regime of strong interactions also for 1D systems. In this respect, new theoretical approaches to confined quantum systems with strong short-range interactions have been proposed in the last couple of years22,23,24,25,26,27,28,29. Within these new developments it has become clear that the strongly interacting limit has an emergent Heisenberg spin model description. While this was realized some time ago for the case of a homogeneous system30, in the presence of confinement the Heisenberg model obtained has non-trivial nearest-neighbour interactions that depend on the confining potential. This can be exploited for tailoring systems to have desired static and dynamic properties25. In the opposite limit where we have vanishing interactions, confined 1D systems are trivially solved as the single-particle Schrödinger equation can be solved numerically to any desired level of accuracy. The natural question to ask is how to describe 1D systems with interaction strengths that are somewhere in the region between the two extreme limits?

In this paper, we propose a deceptively simple ansatz that linearly combines our knowledge of the weakly and strongly interacting limits. As an ansatz containing just two wave functions it is much easier to work with as compared to entirely numerical approaches, which typically have the solution represented on a large basis set with many non-zero contributions.

To describe the main idea we will focus on two-component Fermi systems of N particles with spin projection up and N particles with spin projection down. The Hamiltonian for the N = N + N system is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m1.jpg

where m is the mass and the sums run over the number of particles, i = 1, …, N, xi is the coordinate, and pi the momentum of the ith particle. The external confinement, Vext(x), is assumed to be the same for all particles. Furthermore, we will assume that Vext(x) is parity invariant and has at least N bound states. The interaction strength g is positive for repulsive interactions and negative for attractive interactions. We note that since the parity operator commutes with the Hamiltonian, parity is conserved when g changes. While we only discuss the two-component Fermi system below, the case of bosons or Bose-Fermi mixtures is similar in spirit and in formalism and we return briefly to this extension in the outlook. For the majority of the discussion we assume all particles have equal mass, m, but we also discuss the important extension to mass-imbalanced systems. Notice that in the case of two-component Fermi systems, particles of the same spin projection will not interact due to the antisymmetry required by the Pauli principle which implies that the zero-range interaction will vanish for similar spin projections. For this reason we may use the general form of the Hamiltonian in Eq. (1) with the sum over i < j also for fermionic systems.

Consider now an N-body system and assume that we have solved the problem for g = 0 with energy eigenstate |γ0right angle bracket and for 1/g = 0 with eigenstate |γright angle bracket. Now form the linear combination

An external file that holds a picture, illustration, etc.
Object name is srep28362-m2.jpg

This interpolatory ansatz is motivated by the intuitive idea that the wave function of a system with intermediate-strength interactions contains a mixture of qualities from the wave functions with weak and strong interactions. We may compute left angle bracketγ|H|γright angle bracket as a function of α0 and α and look for optimum values. As we shall demonstrate in this paper, |γright angle bracket provides a simple yet accurate description of the system for any value of g. The only exception is deeply bound states to which we return below.

We will demonstrate that the interpolatory ansatz in Eq. (2) can capture the qualitative features of the eigenstates of the Hamiltonian in Eq. (1) and is quantitatively accurate at the level of a few percent. Furthermore, we show that the expression for the optimum energy of |γright angle bracket may be modified slightly to make it perturbatively correct in both limits of the interaction strength. With this modification, the interpolatory ansatz provides an approximation to the eigenenergy with an accuracy that is comparable to state-of-the-art numerical methods, though the ansatz is far simpler.

We provide a proof of principle by considering some important examples from the few-body limit that are experimentally relevant at the moment. For this purpose we restrict to a harmonic confinement, i.e. An external file that holds a picture, illustration, etc.
Object name is srep28362-m3.jpg, throughout (here ω is the angular frequency of the oscillator).

The results we present here are:

  • Analytical expressions for the ansatz parameters α0 and α that depend exclusively on two matrix elements in |γ0right angle bracket and |γright angle bracket (as well as the eigenenergies of these limiting states).
  • The case N = 2 where one has the exact solution available31. Here we show that our method is accurate for both repulsive and attractive interactions to less than a few percent.
  • The N = 3 case where no analytical solutions are known for general g. Here our ansatz provides very accurate results for all g. The results are compared to exact numerical diagonalisation utilising a unitary transformation of the interaction Hamiltonian.
  • Energies for the impurity limit with N = 1 and N = 1–5 which we compare to experiments and find excellent agreement. We also discuss the Anderson orthogonality catastrophe for this system, which is related to the coefficient α0.
  • An extension of the method to systems with particles of different mass. The examples we discuss are three-body systems and we compare to exact numerical results based on the correlated Gaussian method. This is the first application of the correlated Gaussian method to mass-imbalanced systems in 1D that we are aware of.

The ansatz we propose can be used to get very simple expressions for different observables as one needs to compute only a few matrix elements between the g = 0 and 1/g = 0 states of interest. Our method is directly extendable to bosonic systems or mixed systems, as long as one has access to the two limiting wave functions, and while we have focused on harmonically confined systems it can be straightforwardly extended to any other form of confinement. Furthermore, one may systematically improve the ansatz by adding more states.

Results and Discussion

Let |γ0right angle bracket be an energy eigenstate in the non-interaction limit (that is, g = 0) with eigenenergy E0. We now adiabatically change the interaction strength from g = 0 to |g|  ∞ and in turn we adiabatically change |γ0right angle bracket into a new state denoted |γright angle bracket with energy E. The wave function of the state |γright angle bracket vanishes whenever the position coordinates of any two particles coincide, and |γright angle bracket is thus unaffected by the interaction potential

An external file that holds a picture, illustration, etc.
Object name is srep28362-m4.jpg

As our ansatz we now construct the trial state

An external file that holds a picture, illustration, etc.
Object name is srep28362-m5.jpg

where α0 and α are real parameters. Assuming |γ0right angle bracket and |γright angle bracket are normalised, the energy of the trial state is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m6.jpg

where we let ΔE [equivalent] E  E0.

We use a variational approach to solving the Schrödinger equation by identifying stationary points of the trial state energy functional Eq. (5). We thus select the values of α0 and α that optimise the energy of the trial state for a given value of g. This will yield energies and eigenstates that, although approximate, turn out to be extremely accurate as discussed below.

Before presenting the results of the variational calculation, we shall briefly examine Eq. (5) in the limiting cases of the interaction strength: If we require that α/α0  0 for g  0 such that the trial state approaches |γ0right angle bracket in the non-interacting limit, Eq. (5) gives a first-order term in g of the form left angle bracketγ0|V|γ0right angle bracket, in agreement with first order perturbation theory. We caution, however, that requiring α0/α  0 (i.e. |γright angle bracket  |γright angle bracket) for 1/g  0, does not automatically ensure that the first order expansion of Eq. (5) in 1/g is equal to that of the exact eigenstate. We will come back to this point later on.

The coefficients that yield stationary points of Eq. (5) are given by (see the Supplementary Materials for details)

An external file that holds a picture, illustration, etc.
Object name is srep28362-m7.jpg

This gives the energy

An external file that holds a picture, illustration, etc.
Object name is srep28362-m8.jpg

The denominator in Eq. (7) is positive because |left angle bracketγ0|γright angle bracket|  1 by the Cauchy-Schwarz inequality. Hence, An external file that holds a picture, illustration, etc.
Object name is srep28362-m9.jpg is the energy maximum while An external file that holds a picture, illustration, etc.
Object name is srep28362-m10.jpg is the energy minimum. To ensure the correct energies in the limits g = 0 and 1/g = 0, we use the energy minimum, An external file that holds a picture, illustration, etc.
Object name is srep28362-m11.jpg, to approximate the eigenenergy whenever g > 0, while for g < 0 we use the energy maximum, An external file that holds a picture, illustration, etc.
Object name is srep28362-m12.jpg.

While the interpolatory ansatz is extremely simple, it has a shortcoming in the 1/g  0 limit where it does not reproduce the slope of the energy. As is shown in the following, we may, however, modify the ansatz slightly to correct this behaviour. Letting q [equivalent] −1/g, the first-order expansion An external file that holds a picture, illustration, etc.
Object name is srep28362-m13.jpg has the slope

An external file that holds a picture, illustration, etc.
Object name is srep28362-m14.jpg

where K0 = left angle bracketγ0|V|γ0right angle bracket/g is the corresponding slope of the energy curve in the limit of vanishing interactions. This demonstrates that the important quantity in the slope is the overlap left angle bracketγ0|γright angle bracket. In the original philosophy of the ansatz, we exploit that we know both wave functions in this overlap exactly and thus also the overlap itself; this leaves no unfixed parameters. Realising that this yields a discrepancy we have explored how to modify this assumption in order to improve the approximation.

To this end, we note that the derivation of Eq. (7) actually does not depend on |γright angle bracket being an energy eigenstate. It must be a state with energy E (with respect to the non-interacting Hamiltonian), but beyond that the only requirement of |γright angle bracket is that V|γright angle bracket = 0. Furthermore, Eopt only depends on |γright angle bracket through the squared wave-function overlap left angle bracketγ0|γright angle bracket2. Thus, if we substitute left angle bracketγ0|γright angle bracket2 in Eqs (7 and 8) with some parameter λ, we may regard Eopt and An external file that holds a picture, illustration, etc.
Object name is srep28362-m15.jpg as functions of λ. We can then select λ such that Eopt(λ) becomes perturbatively correct, that is An external file that holds a picture, illustration, etc.
Object name is srep28362-m16.jpg, or,

An external file that holds a picture, illustration, etc.
Object name is srep28362-m17.jpg

where An external file that holds a picture, illustration, etc.
Object name is srep28362-m18.jpg is the slope of the true eigenenergy curve at q = 0 (which is known exactly using the formalism of A.G.V. et al.22). We shall refer to this perturbatively correct modification, Eopt(λ), as the modified ansatz. Note that this modification breaks variational bounds since we have no a priori knowledge of any trial state whose energy is Eopt(λ). By going from the original interpolatory ansatz to the modified ansatz, we lose information about the wave function, but gain the correct slope of the energy at 1/g  0. Below we will see that the modified ansatz increases the accuracy significantly as compared to the original ansatz.

We now proceed to discuss examples of the ansatz. Throughout our discussion, the external potential is taken to be harmonic An external file that holds a picture, illustration, etc.
Object name is srep28362-m19.jpg and the same for all particles. This is the most widely studied and experimentally relevant case at the moment so it will be our focus here. Henceforth, we use natural units An external file that holds a picture, illustration, etc.
Object name is srep28362-m20.jpg = m = ω = 1 such that energies are given in units of An external file that holds a picture, illustration, etc.
Object name is srep28362-m21.jpgω, lengths in units of An external file that holds a picture, illustration, etc.
Object name is srep28362-m22.jpg, and interaction strengths g in units of σAn external file that holds a picture, illustration, etc.
Object name is srep28362-m23.jpgω.

Two particles

The N = 2 case is special as analytical results for any g are available due to the seminal work of Busch et al.31. It is therefore an important benchmark case for our approach. First we note that we will only be interested in the relative energy as the center of mass decouples in the harmonic trap. Note that this decoupling is not an essential assumption of our method and is merely a convenience.

The details on how to construct the ansatz states for the two-particle case are given in the Methods section below. The energy spectrum using the interpolatory ansatz of Eq. (7) is shown in Fig. 1. Only the even-parity solutions are shown on the plot as odd-parity states are unchanged by the zero-range interaction. The figure also includes experimental measurements of the ground-state energy. The experiment has been conducted by S. Jochim’s ultra-cold atoms group in Heidelberg, and the data originates from Wenz et al.19. This data has been corrected for the imperfections of the trap as described in Wenz et al.19. As seen, the experimental data agrees with the interpolatory ansatz well within the experimental uncertainties.

Figure 1
Energy spectrum for the two-particle system.

If we expand the energy in terms of g, the first-order term agrees with the result of ordinary non-degenerate perturbation theory. Similarly, in the limit 1/g = 0, the energy curve (as a function of q) of the true eigenstate has the same slope as that of our interpolatory ansatz.

The inset in Fig. 1 shows a zoom of the energy spectrum on the ground state and compares the energy predicted by Eq. (7) with the exact solution. For q > 0 the energy of the interpolatory ansatz is within 0.05 of the exact energy in the vicinity of q ~ 0.4 and even less elsewhere. The error decreases as we move up in the spectrum. For q < 0 the deviation is less than 0.006 for the ground state; again greatest around q ~ −0.4. On this side, the error increases for the excited states, but we find that it is bounded by about 0.03.

We conclude that the ansatz is extremely accurate for the N = 2 case where we can compare to analytical results31. One may also compare the wave functions and again find extremely good agreement (see Supplementary Fig. S1).

For attractive interactions, a deeply bound molecular state exists that we have so far ignored. However, it turns out that the ansatz of Eq. (7) can be extended to also give extremely accurate results for the deeply bound state. As is shown in the Supplementary Materials, this can be done by including an additional state in the ansatz that has the correct asymptotic behaviour as g  −∞. We stress that this is in fact in complete agreement with the universal philosophy of the ansatz method, i.e. interpolation between (known) extremes. Thus to address deeply bound states one needs a state in the extreme limit of large negative energy. This yields a very precise approximation also for the deeply bound state. This highlights the universal nature of our approach. We will not pursue the deeply bound states any further in this paper.

Three particles

The simplest non-trivial example of a three-body two-component Fermi system has N = 2 and N = 1. The interaction potential is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m24.jpg

and we let x1 be the position of the spin-down fermion, while x2 and x3 denote the position of spin-up fermions. Again the third interaction term will vanish for identical fermions but we keep it for generality.

Eigenstates of the harmonic Hamiltonian are described by two quantum numbers, ν  0 and μ  1, which we shall call the radial quantum number and the angular quantum number, respectively. The eigenenergies are

An external file that holds a picture, illustration, etc.
Object name is srep28362-m25.jpg

The quantum numbers ν and μ can be used to describe the energy eigenstates in both the non-interacting limit and the infinite-interaction limit.

Constructing the ansatz

Recall that |γ0right angle bracket and |γright angle bracket denote eigenstates in the non-interacting limit and the infinite-interaction limit, respectively, and that we are looking for states that are adiabatically connected. As states with different radial quantum number ν are orthogonal, we assume that the adiabatically connected states have the same ν values. Parity p is exactly conserved and thus also the same for two states in the ansatz. The quantity that changes with g is therefore the μ quantum number, and we call the limiting values μ0 and μ, respectively. The angular quantum numbers are related by

An external file that holds a picture, illustration, etc.
Object name is srep28362-m26.jpg

For repulsive interactions, μ0 < μ, and for attractive interactions, μ0 > μ, cf. Eq. (11). The optimum energy of the trial state can now be determined using Eq. (7) with E0 and E given by Eq. (11).

Energy spectrum

Figure 2 shows the energy spectrum for E < 8 (deeply bound states are not considered) both as calculated using the interpolatory ansatz and by exact numerical diagonalisation. The interpolatory ansatz clearly describes the qualitative features of the spectrum well; both for repulsive and attractive interactions. In this energy-range, the ν = 0 states also give a good quantitative match to the numerical results. However, the error seems to grow with ν.

Figure 2
Energy spectrum of the interpolatory ansatz for the three-particle system.

We see in Fig. 2 that trial states with ν = 0 and Δμ = 1 offer a particularly good approximation to the corresponding eigenstates in the repulsive region (see e.g., the first excited state). This is because the slope of the energy curve, that is An external file that holds a picture, illustration, etc.
Object name is srep28362-m27.jpg, at q = 0 is the same as that of the exact eigenenergy. For the trial states with Δμ = 2, this is generally not the case. This result suggests that An external file that holds a picture, illustration, etc.
Object name is srep28362-m28.jpg is an important quantity to reproduce correctly in an attempt such as the present to describe the physics of our problem through a simple ansatz. Note also, that the slope of the energy curve has a discontinuity at 1/g = 0, which contradicts the expectation that the states go smoothly through this region.

If we now enforce the correct slope by a modification of the interpolatory ansatz as proposed in the discussion following Eq. (8), we arrive at the spectrum shown in Fig. 3. We see that the modified ansatz agrees better with the numerical results than the original ansatz; especially for states with ν > 0. There are, however, still deviations on the attractive side of the spectrum, and for high energies, also on the repulsive side. We shall give a quantitative discussion of the quality of the approximation when discussing the impurity system below. We have included the experimental measurements of the ground-state energy19 in the figure and see that both ansatz and modified ansatz agrees very nicely with experiment, although for large g the modified ansatz naturally does better.

Figure 3
Energy spectrum for the three-particle system–modified ansatz.

Mass-imbalanced systems

In the case of different masses, one typically uses another length scale given by An external file that holds a picture, illustration, etc.
Object name is srep28362-m29.jpg where An external file that holds a picture, illustration, etc.
Object name is srep28362-m30.jpg, and m1 is the mass of the spin-down fermion while m2 and m3 are the masses of the spin-up fermions. We now consider the case where m1 = M and m2 = m3 = m.

In Fig. 4 we show the energy spectrum obtained by the interpolatory ansatz for M/m = 1/2 and M/m = 2 and compare this with numerically calculated results using the correlated Gaussian approach (see the Supplementary Materials for details on the numerical methods). The agreement between the ansatz and the numerical results is striking for the low-energy part of the spectrum considered here, and we see that the ansatz can be extended also to mass-imbalanced systems.

Figure 4
Energy spectrum for the three-particle system with different masses.

Impurity systems

We now consider a system of N fermions among which one particle (x1) is spin-down and the N = N  1 remaining particles (x2, …, xN) are spin-up. Taking into account that interactions between identical fermions vanish, we can write the interaction term as

An external file that holds a picture, illustration, etc.
Object name is srep28362-m31.jpg

We restrict the discussion to the ground state with repulsive interactions as this has been a focus of recent experimental attention19. The considerations can be extended to obtain more states in the spectrum, to the attractive side, and to deeply bound states in the same manner as in the previous examples.

We denote by |γAright angle bracket the Slater determinant of the single-particle harmonic eigenstates left angle bracketxk|nright angle bracket = ψn(xk) for k = 1, …, N and n = 0, …, N  1. Here ψn(x) is the single-particle eigenstate of the harmonic oscillator Hamiltonian in coordinate space with quantum number n and argument x. The state |γAright angle bracket is antisymmetric with respect to interchange of any two coordinates, so left angle bracketγA|V|γAright angle bracket = 0. We can write the antisymmetric state as

An external file that holds a picture, illustration, etc.
Object name is srep28362-m32.jpg

if we define

An external file that holds a picture, illustration, etc.
Object name is srep28362-m33.jpg

where x = (x1, …, xN), SN is the symmetric group of order N, and Π(σ) indicates the integration region xσ(1) <…< xσ(N). In each region, Π(σ), the wave function of the ground state in the infinite-interaction limit is proportional to left angle bracketx|γAright angle bracket. The (normalised) ground state in the infinite-interaction limit is then

An external file that holds a picture, illustration, etc.
Object name is srep28362-m34.jpg

for some coefficients a = (a1, …, aN) to be determined by the method of A.G.V. et al.22 (see the Supplementary Materials). The ground state in the non-interaction limit is the single-particle harmonic eigenstate ψ0(x1) multiplied by the Slater determinant for the remaining particles in the states ψn(xk) for k = 2, …, N and n = 0, …, N  2.

For systems of four and five particles, the interpolatory ansatz with these |γ0right angle bracket and |γright angle bracket yields integrals that can be evaluated analytically. For larger systems, the integrals are readily evaluated numerically. The resulting energies for N = 4–6 are plotted as dashed lines in Fig. 5. Here we see a very good agreement between the ansatz and the numerically exact results, and in turn excellent agreement with experimental measurements19. This indicates that the energetics of the system is captured by the interpolatory ansatz with high accuracy.

Figure 5
Energy spectrum of the impurity system.

However, as we discussed briefly above, the ansatz does not generally reproduce the correct first-order energy term at 1/g = 0. Assuming that the non-interacting state should remain untouched, this prompts us to investigate whether another state can be found in the 1/g = 0 limit that can replace |γright angle bracket in Eq. (4) and in turn give the exact result for the slope of the energy at 1/g = 0. As noted above, Eq. (7) remains valid if we substitute the eigenstate |γright angle bracket with any other state with energy E that obeys V|γright angle bracket = 0. In particular, any choice of a in Eq. (16) would work, and Eopt only depends on a through the wave-function overlap left angle bracketγ0|γright angle bracket. Hence, because Eopt is monotonic in left angle bracketγ0|γright angle bracket2, we may optimise the energy with respect to a for all values of g by maximising left angle bracketγ0|γ(a)right angle bracket2 with respect to a. Leaving the details to the Supplementary Materials, the optimum is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m35.jpg

with

An external file that holds a picture, illustration, etc.
Object name is srep28362-m36.jpg

For the ground state of the N = 3–6 systems, however, left angle bracketγ0|γ(amax)right angle bracket2 is very close to the known exact value of left angle bracketγ0|γright angle bracket2, and is not large enough to make the slope of the energy correct in the strongly-interacting limit, that is, left angle bracketγ0|γ(amax)right angle bracket2 < λ with λ given by Eq. (9). This indicates that we cannot find a state in the infinite-interaction limit that both has the correct zeroth-order energy, E, and satisfies the delta-boundary conditions, V|γright angle bracket = 0. However, we caution that this is under the assumptions that the wave function of |γright angle bracket is continuous and has discontinuous derivatives that satisfy the boundary conditions imposed by the zero-range interaction.

We see that the interpolatory ansatz–with whatever choice of |γright angle bracket–does not reproduce the correct energy slope. The modified ansatz, however, does. It is worthwhile to note that the inability to find a state, |γright angle bracket = α0|γ0right angle bracket + α|γright angle bracket, whose expectation value is that predicted by the modified ansatz, does not imply that the modified ansatz cannot be used. The two methods rely on different information about the system: The interpolatory ansatz requires the knowledge of the wave function at g = 0 and 1/g = 0, whereas the modified ansatz uses the knowledge of the energy behaviour. Therefore, one can use the modified anzatz to estimate the energies, and the interpolatory ansatz to approximate the wave functions.

Figure 5 compares the ground-state energy of the modified ansatz with results from an exact numerical diagonalisation as well as experimental data19. Here we see an improved agreement for large g. The deviations of the modified ansatz from exact numerical diagonalisation are shown in greater detail in Fig. 6. The errors are more than an order of magnitude smaller than those of the unmodified ansatz (shown in Supplementary Fig. S4).

Figure 6
Ansatz accuracy comparison.

We plot the error scaled against N instead of E0 or E, because E0 scales as An external file that holds a picture, illustration, etc.
Object name is srep28362-m37.jpg while ΔE = N. As seen in Fig. 5, the maximum in the error moves to larger interaction strength for higher N, i.e. with system size. There seems to be only very little increase in the magnitude of the error for larger N for the system sizes we have studied. Notice that the error is slightly negative for 1/g [similar, equals] 0 in the case N = 3–6. Since we know the exact energies and slopes around 1/g = 0, the likely cause of this is that we are pushing the accuracy of the exact numerical diagonalisation method here.

Our first observation is that the approximation of the ground-state energy offered by the modified ansatz is so good that it can compete and even in some cases beat state-of-the-art numerical exact methods. The drawback seems quite evident, i.e. we cannot be sure that any state exists in the infinite-interaction limit that when used in Eq. (5) would reproduce the energy of the modified ansatz. Thus, the modified ansatz does not immediately give us any information about the wave function in spite of its near perfect approximation of the energy. We also note that the modified ansatz breaks the variational bound on the ground state and can in principle have lower energy as compared to the exact result. However, we have clearly demonstrated that the slope of the energy at 1/g = 0 is an extremely important quantity for these systems as it appears crucial to reproduce in order to capture the energetics. This highlights the important role played by the recently developed theory of strongly interacting confined systems22.

Anderson overlap

Finally, we discuss the so-called Anderson overlap, which is the wave-function overlap between the non-interacting eigenstate, |γ0right angle bracket, and the interacting state, |γright angle bracket, for some value of the interaction strength, g. This quantity is related to the Anderson orthogonality catastrophe32, which states that the Anderson overlap is zero in the thermodynamic limit; that is left angle bracketγ0|γright angle bracket  0 for N  ∞, and in particular left angle bracketγ0|γright angle bracket  0 for N  ∞.

In Fig. 7 we illustrate how the overlap left angle bracketγ0|γright angle bracket2–with |γright angle bracket given by the interpolatory ansatz–decreases as a function of g. Due to the fact that we only consider a finite-sized system, the overlap does not approach zero as g  ∞, but the plot clearly shows that the overlap from our interpolatory ansatz tends to decrease as expected. The fact that the ansatz gives a very accurate approximation for the energy of the system does not immediately imply that this is also the case for the wave function. We leave this question for future studies, and the overlaps presented here are thus predictions based on the ansatz. It should be compared either to elaborate exact numerical calculations or to experimental measurements.

Figure 7
Anderson overlap according to the interpolatory ansatz.

Outlook

We have proposed a simple interpolatory ansatz for approximating the energy eigenstate of a confined, one-dimensional system of interacting particles. The ansatz is a linear combination of known eigenstates in the extreme limits of the interaction strength, g  0 and 1/g  0, respectively. Thanks to recent advances in the description of the eigenstates in the 1/g = 0 limit, both these wave functions are now available. An analytical expression for the optimum energy of this ansatz is presented which is an elementary function of only two matrix elements; the interaction energy of the eigenstate at g = 0, and the wave-function overlap between the eigenstates in the two limits of g  0 and 1/g  0. By focusing on harmonically trapped impurity systems of fermions, we have demonstrated that the ansatz is able to capture the physics of such a system. It gives us a highly accurate approximation for the energy and it also gives us a very simple expression for the wave function.

For both the two- and three-particle systems, we have been able to reproduce the entire energy spectrum with the interpolatory ansatz, save for deeply bound molecular states. The ansatz can be extended to describe deeply bound states as well; this has been shown specifically for the two-particle system. Taking the three-particle system as an example, we have also demonstrated that the ansatz works equally well for mass-imbalanced systems. A future extension of this study might investigate mass-imbalanced systems of four or more particles. It should be noted that the bottleneck here is that there are generally quite few known results about mass-imbalanced systems in the 1/g  0 limit33. It is an open problem to find a general method that yields exact eigenstates in the strongly interacting regime for mass-imbalanced systems.

A drawback of our ansatz is that it is in general not perturbatively correct in the strongly-interacting regime. More precisely, if we take the first order derivative of the energy with respect to 1/g it deviates from the known exact result. We may, however, modify the expression for the energy of the interpolatory ansatz slightly such that it is perturbatively correct to linear order in g for g  0 and to linear order in 1/g for 1/g  0. The modified ansatz has great simplicity and accuracy at a level that is competitive with state-of-the-art numerical methods for obtaining the energy of the ground state for arbitrary g. Due to its simplicity it should provide a very useful tool.

We note that although the results presented here assume that the particles are trapped in an external harmonic trap, the formalism is completely general and can be applied for arbitrary external potentials with at least N bound single-particle states for N-body systems. As we have shown, the relative deviation of the energy obtained from the ansatz grows only very slowly with N, and there should be no problem in extending the technique to even larger systems than considered here. The decisive quantity is the overlap of the g = 0 and 1/g = 0 wave functions. This overlap may be computed using the same methods that have recently been used to compute spin chain models for strongly interacting fermions34 and thus scaling to larger N of order 30 or 40 is certainly within reach.

A future direction would be to consider a generalised version of the interpolatory ansatz where one systematically adds more states at g = 0 and 1/g = 0 in order to gradually improve the comparison. In addition, it is relatively straightforward to apply the interpolatory ansatz in the case of strongly interacting bosons35,36, or mixed systems22,28. The requirements are knowledge of states in the two limits and their overlaps so that the interpolation can be performed. The formulas for the interpolated energy given here still apply. An example could be an impurity interacting strongly with a Tonks-Girardeau gas of hard-core bosons, which is a topic of great recent interest37,38,39.

Methods

In the following, we provide the details of the methods used in applying the interpolatory ansatz to two-, three- and many-particle systems.

Details of the two-particle system

We consider a system of two distinguishable fermions and we define An external file that holds a picture, illustration, etc.
Object name is srep28362-m38.jpg. In the absence of interactions, the energy eigenstates are the harmonic eigenstates denoted |nright angle bracket with integer n  0 (we are only concerned with the motion relative to the center of mass).

Exact solutions of the two-particle problem are available for arbitrary values of g31,40. The energy of an exact energy eigenstate is given indirectly by40

An external file that holds a picture, illustration, etc.
Object name is srep28362-m39.jpg

and the wave function of the state is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m40.jpg

where 1F1 is the confluent hypergeometric function of the first kind.

Constructing the ansatz

Let |γ0right angle bracket = |n1right angle bracket be an energy eigenstate in the non-interaction limit. Here |n1right angle bracket is a single-particle eigenstate of a harmonic oscillator Hamiltonian (in the relative coordinate x) with quantum number n1. Furthermore, let |γright angle bracket be the corresponding eigenstate in the infinite-interaction limit–that is, |γ0right angle bracket  |γright angle bracket as the interaction strength is changed adiabatically from g = 0 to 1/g = 0. By conservation of parity, the states |γ0right angle bracket and |γright angle bracket have the same parity.

If |γ0right angle bracket is odd, left angle bracketx = 0|γ0right angle bracket = 0 and thus left angle bracketγ0|V|γ0right angle bracket = 0. Hence, odd harmonic eigenstates do not change as we introduce a non-zero interaction. The even harmonic eigenstates do, however, change; the correct even eigenstate in the infinite-interaction limit is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m41.jpg

with n2 = n1 + 1 for repulsive interactions (g > 0) and n2 = n1  1 for attractive interactions (g < 0).

Details of the three-particle system

Before we employ the interpolatory ansatz, we first separate out the center-of-mass motion using hyperspherical coordinates35,41. This is done merely for convenience and is not in any way essential for the approach. Defining An external file that holds a picture, illustration, etc.
Object name is srep28362-m42.jpg and An external file that holds a picture, illustration, etc.
Object name is srep28362-m43.jpg, the hyperradius is given by An external file that holds a picture, illustration, etc.
Object name is srep28362-m44.jpg and the hyperangle is defined by tanϕ = y/x. The Hamiltonian of the relative motion becomes

An external file that holds a picture, illustration, etc.
Object name is srep28362-m45.jpg

where [nabla]2 is the Laplacian in polar coordinates (ρ, ϕ) and

An external file that holds a picture, illustration, etc.
Object name is srep28362-m46.jpg

is the interaction.

Limiting cases

The quantum numbers ν and μ can be used to describe the energy eigenstates in both the non-interacting limit and the infinite-interaction limit. The eigenstate wave function in both limits has the general form41

An external file that holds a picture, illustration, etc.
Object name is srep28362-m47.jpg

where An external file that holds a picture, illustration, etc.
Object name is srep28362-m48.jpg is a generalised Laguerre polynomial.

In the non-interacting limit, the (normalised) angular part of the wave function is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m49.jpg

where p is the parity of the wave function. In this limit p = (−1)μ.

In the infinite-interaction limit, the wave function vanishes at the lines ϕ = −5π/6, −π/2, −π/6, π/6, π/2, 5π/6. In the regions between these lines, the wave function solves the Schrödinger wave equation for the harmonic oscillator without interactions.

The eigenenergies in the limit 1/g = 0 is given by Eq. (11) with μ = 3, 6, 9, 12, …. Each allowed eigenenergy is three-fold degenerate (not counting states with a different radial quantum number ν): One of the eigenstates in each energy triplet is a harmonic eigenstate with μ = 3, 6, 9, 12, … which is unaffected by the interactions. Between the two remaining eigenstates in the triplet, one is odd and the other is even.

When μ is even (that is, μ = 6, 12, 18, …) the non-trivial eigenstates in the infinite-interaction limit have the angular wave function

An external file that holds a picture, illustration, etc.
Object name is srep28362-m50.jpg

if p is their parity. The coefficients in the four remaining regions follow by symmetry considerations.

When μ is odd (μ = 3, 9, 15, …),

An external file that holds a picture, illustration, etc.
Object name is srep28362-m51.jpg

Details of the mass-imbalanced system

The transformation into hyperspherical coordinates proceeds along the same lines with a modified Jacobian. Generally, the coordinates x = (x1, x2, x3) are transformed to Jacobi coordinates, x = Jx, through the transformation matrix

An external file that holds a picture, illustration, etc.
Object name is srep28362-m52.jpg

where M23 = m2 + m3, M123 = m1 + m2 + m3 and the ‘reduced’ masses are defined as An external file that holds a picture, illustration, etc.
Object name is srep28362-m53.jpg and An external file that holds a picture, illustration, etc.
Object name is srep28362-m54.jpg. This transformation allows us to separate the center-of-mass motion from the relative motion, the solutions of the former being the well-known harmonic eigenstates. Afterwards, we transform the remaining relative coordinates into hyperspherical coordinates, ρ and ϕ, by An external file that holds a picture, illustration, etc.
Object name is srep28362-m55.jpg and tan(ϕ) = An external file that holds a picture, illustration, etc.
Object name is srep28362-m56.jpg.

From now on, we assume that m1 = M and m1 = m2 = m. The interaction potential can then be written as

An external file that holds a picture, illustration, etc.
Object name is srep28362-m57.jpg

where θ0 = arctan(ζ) and An external file that holds a picture, illustration, etc.
Object name is srep28362-m58.jpg. As for the equal-mass case, the energy is given by Eq. (11) 42.

The μ eigenvalue can be found by using parity symmetry, the Pauli principle and the delta-boundary conditions of the interaction potential. Using these conditions, one can setup an equation from which μ can be obtained in the limits g = 0 and 1/g = 0: For solutions with odd parity, μ solves the equation

An external file that holds a picture, illustration, etc.
Object name is srep28362-m59.jpg

For even-parity solutions, the equation is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m60.jpg

Once μ is found, the wave function is also known cf. Eq. (24) and the energy is given by Eq. (11) 42.

The wave function in the infinite-interaction limit now vanishes along the delta-boundary lines ϕ =  ± θ0, ±π/2, π ± θ0. Only when M = m, is the wave function non-zero in all six regions separated by the delta-boundary lines. In order to illustrate this, we look at two explicit examples where M/m = 1/2 and M/m = 2, respectively. For the case of M/m = 1/2, we have θ0 [similar, equals] 0.421 (or 24.1°), and the lowest-energy solutions of Eqs (30 and 31) at 1/g = 0 have μ [similar, equals] 0.731 (both odd and even) while the second-lowest has μ [similar, equals] 3.75 (odd). The angular part of the wave function for the ground state and the first excited state in the infinite-interaction limit are

An external file that holds a picture, illustration, etc.
Object name is srep28362-m61.jpg

where An external file that holds a picture, illustration, etc.
Object name is srep28362-m62.jpg is the parity of the state. The angular part for the non-degenerate second excited state is

An external file that holds a picture, illustration, etc.
Object name is srep28362-m63.jpg

When M/m = 2, the roles are reversed. The ground state is now non-degenerate with μ [similar, equals] 2.552 at g = ∞ and θ0 [similar, equals] 0.615 (or 35.2°). In addition, the wave function has the same form as Eq. (33), only with different θ0 and μ. For the first and second excited states, the wave function has the form of Eq. (32).

One might think that there would be some continuous crossover from M < m to M = m and then to M > m, but this is not the case. Indeed, M = m is a singular case where the wave function is non-zero in all six regions.

Additional Information

How to cite this article: Andersen, M. E. S. et al. An interpolatory ansatz captures the physics of one-dimensional confined Fermi systems. Sci. Rep. 6, 28362; doi: 10.1038/srep28362 (2016).

Supplementary Material

Supplementary Information:

Acknowledgments

Part of this work is based on the Bachelor thesis of M.E.S.A. The authors thank C. Forssén, D.V. Fedorov, and A.S. Jensen for discussions, as well as C. Forssén and X. Cui for feedback on the manuscript. We thank the experimental team of the S. Jochim group for extended discussions and for sharing their results. This work was funded by the Danish Council for Independent Research DFF Natural Sciences and the DFF Sapere Aude program.

Footnotes

Author Contributions A.G.V., E.J.L. and N.T.Z. devised the project. M.E.S.A. and A.S.D. developed the formalism under the supervision of A.G.V. and N.T.Z. The numerical calculations were carried out by A.S.D. and M.E.S.A. with assitance from A.G.V. and E.J.L. The initial draft of the paper was written by M.E.S.A., A.S.D. and N.T.Z. All authors contributed to the revisions that led to the final version.

References

  • Giamarchi T. Quantum Physics in One Dimension (Oxford University Press Inc., New York, 2003).
  • Bezryadin A. Superconductivity in Nanowires (WILEY-VCH Verlag, Weinheim, Germany, 2013).
  • Altomare F. & Chang A. M. One-Dimensional Superconductivity in Nanowires (WILEY-VCH Verlag, Weinheim, Germany, 2013).
  • Anderson P. W. The resonating valence bond state in La2CuO4 and superconductivity. Science 235, 1196–1198 (1987). [PubMed]
  • Lee P. A., Nagaosa N. & Wen X.-G. Doping a mott insulator: Physics of high-temperature superconductivity. Rev. Mod. Phys. 78, 17–85 (2006).
  • Lewenstein M. et al. Ultracold atomic gases in optical lattices: mimicking condensed matter physics and beyond. Advances in Physics 56, 243–379 (2007).
  • Bloch I., Dalibard J. & Zwerger W. Many-body physics with ultracold gases. Rev. Mod. Phys. 80, 885–964 (2008).
  • Moritz H., Stöferle T., Köhl M. & Esslinger T. Exciting collective oscillations in a trapped 1D gas. Phys. Rev. Lett. 91, 250402 (2003). [PubMed]
  • Stöferle T., Moritz H., Schori C., Köhl M. & Esslinger T. Transition from a strongly interacting 1D superfluid to a Mott insulator. Phys. Rev. Lett. 92, 130403 (2004). [PubMed]
  • Kinoshita T., Wenger T. & Weiss D. S. Observation of a one-dimensional Tonks-Girardeau gas. Science 305, 1125–1128 (2004). [PubMed]
  • Paredes B. et al. Tonks-Girardeau gas of ultracold atoms in an optical lattice. Nature 429, 277–281 (2004). [PubMed]
  • Kinoshita T., Wenger T. & Weiss D. S. A quantum Newton’s cradle. Nature 440, 900–903 (2006). [PubMed]
  • Haller E. et al. Realization of an excited, strongly correlated quantum gas phase. Science 325, 1224–1227 (2009). [PubMed]
  • Haller E. et al. Pinning quantum phase transition for a luttinger liquid of strongly interacting bosons. Nature 466, 597–600 (2010). [PubMed]
  • Pagano G. et al. A one-dimensional liquid of fermions with tunable spin. Nat Phys 10, 198–201 (2014).
  • Serwane F. et al. Deterministic preparation of a tunable few-fermion system. Science 332, 336–338 (2011). [PubMed]
  • Zürn G. et al. Fermionization of two distinguishable fermions. Phys. Rev. Lett. 108, 075303 (2012). [PubMed]
  • Zürn G. et al. Pairing in few-fermion systems with attractive interactions. Phys. Rev. Lett. 111, 175302 (2013). [PubMed]
  • Wenz A. N. et al. From few to many: Observing the formation of a Fermi sea one atom at a time. Science 342, 457–460 (2013). [PubMed]
  • Murmann S. et al. Two fermions in a double well: Exploring a fundamental building block of the Hubbard model. Phys. Rev. Lett. 114, 080402 (2015). [PubMed]
  • Murmann S. et al. Antiferromagnetic Heisenberg spin chain of a few cold atoms in a one-dimensional trap. Phys. Rev. Lett. 115, 215301 (2015). [PubMed]
  • Volosniev A. G., Fedorov D. V., Jensen A. S., Valiente M. & Zinner N. T. Strongly interacting confined quantum systems in one dimension. Nat Commun 5 (2014). [PubMed]
  • Cui X. & Ho T.-L. Ground-state ferromagnetic transition in strongly repulsive one-dimensional Fermi gases. Phys. Rev. A 89, 023611 (2014).
  • Deuretzbacher F., Becker D., Bjerlin J., Reimann S. M. & Santos L. Quantum magnetism without lattices in strongly interacting one-dimensional spinor gases. Phys. Rev. A 90, 013611 (2014).
  • Volosniev A. G. et al. Engineering the dynamics of effective spin-chain models for strongly interacting atomic gases. Phys. Rev. A 91, 023620 (2015).
  • Levinsen J., Massignan P., Bruun G. M. & Parish M. M. Strong-coupling ansatz for the one-dimensional Fermi gas in a harmonic potential. Science Advances 1 (2015). [PMC free article] [PubMed]
  • Yang L., Guan L. & Pu H. Strongly interacting quantum gases in one-dimensional traps. Phys. Rev. A 91, 043634 (2015).
  • Hu H., Guan L. & Chen S. Strongly interacting Bose-Fermi mixtures in one dimension. New Journal of Physics 18, 025009 (2016).
  • Yang L. & Cui X. Effective spin-chain model for strongly interacting one-dimensional atomic gases with an arbitrary spin. Phys. Rev. A 93, 013617 (2016).
  • Ogata M. & Shiba H. Bethe-ansatz wave function, momentum distribution, and spin correlation in the one-dimensional strongly correlated Hubbard model. Phys. Rev. B 41, 2326–2338 (1990). [PubMed]
  • Busch T., Englert B.-G., Rzażewski K. & Wilkens M. Two cold atoms in a harmonic trap. Foundations of Physics 28, 549–559 (1998).
  • Anderson P. W. Infrared catastrophe in Fermi gases with local scattering potentials. Phys. Rev. Lett. 18, 1049–1051 (1967).
  • Dehkharghani A. S., Volosniev A. G. & Zinner N. T. Impenetrable mass-imbalanced particles in one-dimensional harmonic traps. arXiv:1511.01702 (2015).
  • Loft N. J. S., Kristensen L. B., Thomsen A. E. & Zinner N. T. Comparing models for the ground state energy of a trapped one-dimensional Fermi gas with a single impurity. arXiv:1508.05917 (2015).
  • Zinner N. T., Volosniev A. G., Fedorov D. V., Jensen A. S. & Valiente M. Fractional energy states of strongly interacting bosons in one dimension. EPL (Europhysics Letters) 107, 60003 (2014).
  • Massignan P., Levinsen J. & Parish M. M. Magnetism in strongly interacting one-dimensional quantum mixtures. Phys. Rev. Lett. 115, 247202 (2015). [PubMed]
  • Palzer S., Zipkes C., Sias C. & Köhl M. Quantum transport through a Tonks-Girardeau gas. Phys. Rev. Lett. 103, 150601 (2009). [PubMed]
  • Mathy C. J. M., Zvonarev M. B. & Demler E. Quantum flutter of supersonic particles in one-dimensional quantum liquids. Nat Phys 8, 881–886 (2012).
  • Knap M., Mathy C. J. M., Ganahl M., Zvonarev M. B. & Demler E. Quantum flutter: Signatures and robustness. Phys. Rev. Lett. 112, 015302 (2014). [PubMed]
  • Farrell A. & van Zyl B. P. Universality of the energy spectrum for two interacting harmonically trapped ultra-cold atoms in one and two dimensions. Journal of Physics A: Mathematical and Theoretical 43, 015302 (2010).
  • Harshman N. L. Symmetries of three harmonically trapped particles in one dimension. Phys. Rev. A 86, 052122 (2012).
  • Loft N. J. S., Dehkharghani A. S., Mehta N. P., Volosniev A. G. & Zinner N. T. A variational approach to repulsively interacting three-fermion systems in a one-dimensional harmonic trap. Eur. Phys. J D 69, 65 (2015).

Articles from Scientific Reports are provided here courtesy of Nature Publishing Group