Dipolar Bose-Einstein condensates in triple-well potentials
DDipolar Bose-Einstein condensates in triple-wellpotentials
Rüdiger Fortanier, Damir Zajec, Jörg Main and GünterWunner
1. Institut für Theoretische Physik, Universität Stuttgart, 70550 Stuttgart, GermanyE-mail: [email protected] , [email protected] Abstract.
Dipolar Bose-Einstein condensates in triple-well potentials are well-suitedmodel systems for periodic optical potentials with important contributions of thenon-local and anisotropic dipole-dipole interaction, which show a variety of effects suchas self-organisation and formation of patterns. We address here a macroscopic sampleof dipolar bosons in the mean-field limit. This work is based on the Gross-Pitaevskiidescription of dipolar condensates in triple-well potentials by Peter et al a r X i v : . [ qu a n t - ph ] O c t ipolar Bose-Einstein condensates in triple-well potentials
1. Introduction
One of the most interesting features of Bose-Einstein condensates (BECs) is the possibilityof the investigation of quantum effects in a macroscopically controlled way. BECs havebeen experimentally realised with atoms sustaining a large magnetic dipole moment suchas Cr [1–3] and, more recently, Dy [4, 5] and Er [6]. Very recently, fast progresstowards the creation of BECs of polar molecules [7], which sustain large electric dipolemoments, has been made. These developments have opened the field of research of effectsgenerated by the dipole-dipole interaction (DDI). A major part of these effects can besummarised under the topics “self-organisation” and pattern formation. These effects areinvolved in the formation of a supersolid quantum phase. Menotti et al [8] have shown bycalculations on the basis of the Bose-Hubbard model that the existence of a supersolid isclosely connected to the appearance of metastable states in an optical lattice. However,the Bose-Hubbard description is only practicable for small atom numbers and cannotdescribe the collapse of the macroscopic wave function. The stability of these metastablestates with respect to the collapse thus has to be analysed by the use of a method – likethe mean-field description – which is able to describe the local divergence of the wavefunction’s amplitude.A minimal system for the analysis of the effects mentioned above is a dipolar BEC ina triple-well (TW) potential. In Ref. [9] the Bose-Hubbard model was used to investigatethe possible ground states for a mesoscopic sample of dipolar bosons up to 18 particles.There, four different ground-state phases have been observed. Peter et al [10] applied amean-field approach to this system, yet, considering a macroscopic BEC. They found noclear separation of phases, and observed that some population distributions predictedby the Bose-Hubbard model are unstable within the Gross-Pitaevskii description. Mostof these calculations were performed on a grid using imaginary-time evolution (ITE).While this is a globally convergent method for the linear Schrödinger equation, the ITEdoes not necessarily converge to the ground state in a nonlinear system such as the GPE.In Ref. [8] this fact has been used to find metastable states.Dipolar BECs in double-well potentials have been investigated in [11, 12]. In thissystem three major phases have been found: In the first both wells are populated equally.The second is the symmetry-broken phase (macroscopic quantum self-trapping, MQST),in which the majority of the particles populates one well, and the third is the unstablephase, where the condensate wave function collapses. It has been pointed out in [12]and [13] that MQST is a dynamical effect arising from the interaction of the particleswhich reflects itself in the nonlinearity of the GPE.In Ref. [14] Zhang et al have performed a mean-field three-mode approximation todipolar BEC in triple-well potentials. They found that the inter-level coupling of thestates changes depending on their inter-site interactions and leads to macroscopic phasetransitions. Furthermore, they show that the long-range nature of the dipole-dipoleinteraction leads to new dynamical effects such as long-range Josephson oscillations,where tunnelling between the outer wells takes place with negligible alteration of the ipolar Bose-Einstein condensates in triple-well potentials et al [10]. There, however, no excited states were calculated which we findplaying a crucial role in the formation of phases and the dynamical behaviour of thecondensate. The purpose of this paper is to show that these states, which sometimesexist simultaneously, are the reason for quantum phase transitions and can be used todetect such by its dynamical characteristics. In particular, we investigate a dipolar BECin an external TW potential on the basis of the extended time-dependent GPE in theform H Ψ( r , t ) = (cid:18) −
12 ∆ + V TW + V dd + V sc (cid:19) Ψ( r , t ) = i ∂ t Ψ( r , t ) , (1)with V TW = − V (cid:88) i =1 exp (cid:18) − x − q ix ) ω x − y ω y − z ω z (cid:19) ,V dd = 3 N a dd (cid:90) d r (cid:48) − ϑ | r − r (cid:48) | | Ψ( r (cid:48) , t ) | ,V sc = 4 πN a | Ψ( r , t ) | , where N is the number of particles and a dd and a denote the dipole and scatteringlength, respectively. The centres of the three individual wells are given by q x = − l , q x = 0 , and q x = l . The dipoles are aligned along the z -axis, so that ϑ is the anglebetween the z -axis and the vector r − r (cid:48) . Here, we have adopted the unit system ofPeter et al [10], which implies measuring all lengths in units of the inter-well spacing l , all energies in units of (cid:126) /ml and time in units of ml / (cid:126) , where m is the particlemass. The TW potential and the orientation of the dipoles is visualised in figure 1 forthe repulsive configuration (the attractive configuration would imply the dipoles to bealigned in x -direction). The widths of the Gaussians for the external trap are chosensuch that changing the polarisation direction does not change the on-site effects of thedipole interaction ( ω x = ω z ) and that the stability of the condensate is higher thanin the spherical case ( ω y > ω x,z ) . Furthermore, ω x has to be set to a value where thedifferent wells are clearly distinct. We further assume the relevant time scale of theinter-well oscillations to be large in comparison with breathing-mode like oscillations.This corresponds to a sufficiently low tunnelling rate. The interesting effects mentionedabove are expected to arise from the interplay between the short-range and the long-rangenature of the interaction. We therefore included both kinds of interaction and assumed N a and
N a dd to be independently adjustable quantities, whereas the results do notexplicitly depend on N due to the scaling properties of the GPE.The paper is organised as follows. In section 2 we will briefly introduce the two ipolar Bose-Einstein condensates in triple-well potentials Figure 1.
Visualisation of the TW potential. The dipoles are aligned in z -direction(repulsive configuration). The potential width L y = ω y / is 4 times larger than L x and L z . The parameter l denotes the distance between the minima of the potentials. methods we used to solve the GPE. In section 3 we will present our results and investigatethe system for two different sets of parameters in sections 3.1 and 3.2 in detail.
2. Methods
A well-known standard method to treat the dipolar GPE is the solution on a grid, andwe use this technique for the computation of ground states and to simulate the real-timedynamics. However, the large number of grid points (and therefore parameters) does notallow us to obtain stationary states by a nonlinear root search. Thus, the accessibility ofexcited states, which do play a crucial role in the TW system, requires a larger effort, asit has been shown e.g. in [8,15]. For this reason we also apply a variational approach withcoupled Gaussian wave packets (GWPs) which has proven to be a full-fledged alternativeto grid calculations [16–18] for the description of the ground and excited states as wellas for real-time dynamics far beyond the stationary solutions.
The propagation of the macroscopic wave function Ψ in real time and the ITE for thecalculation of stationary solutions can be performed on a grid. Particularly, groundstates are calculated by the evolution of an initial wave function in imaginary time ( t = − i τ ) which dampens all other excited states. ipolar Bose-Einstein condensates in triple-well potentials U ( τ ) on an initial state ψ can be investigated by expanding this state in the eigenfunctions φ of the Hamiltonian HU ( τ ) | ψ (cid:105) = e − Hτ (cid:88) i | φ i (cid:105) (cid:104) φ i | ψ (cid:105) = (cid:88) i e − E i τ c i | φ i (cid:105) . (2)Although the ITE dampens all of the eigenstates of the series, excited states vanish fasterthan the ground state. This guarantees the global convergence of the ITE. However,since in the nonlinear GPE the Hamiltonian depends on the actual state, the basis set ofthe expansion changes after each time step. Therefore the damping of the excited statescannot be assured. We thus have to compare the numerical results with the solution ofthe variational approach or choose an initial wave function which is sufficiently similarto the ground state. The latter can be realised by simply mapping the solution of thevariational approach on the grid or by using previous numerical solutions with similarparameters. Even though we have not computed excited states by means of the ITE,such states can still be investigated on the grid when the solution of the variationalapproach is used as an initial wave function for dynamical simulations of the GPE.We use the split-operator method for the grid calculations, where the scatteringpotential V sc and the dipole-dipole potential V dd have to be calculated at each time step.The latter can be evaluated by means of the convolution theorem. Altogether, we haveto perform six Fourier transforms for each time step. A comprehensive presentation ofthis approach is given in [18]. As an alternative to simulations on a grid the condensate wave function can beparametrised by a set of variational parameters, and the time evolution of the stateis given by the time-dependence of the variational parameters. Our variational ansatzconsists of a linear superposition of three GWPs. Each of the GWPs has the form g k = e − (cid:16) ( x T − q k ) T A k ( x − q k ) − i ( p k ) T ( x − q k ) + γ k (cid:17) , (3)where the symbol T denotes the transposition and where in general the time-dependentparameters A k are × complex symmetric matrices, p k and q k are real 3 d vectors, and γ k are complex numbers. We assume that the z -direction (the direction of the dipolealignment) has a strong confinement due to the external trap and ignore translationsand rotations in this direction by setting A kxz = A kyz = p kz = q kz = 0 . However, for theother directions we apply no further restrictions, particularly with respect to positionand movement of the GWPs in the x -direction. It is reasonable to start with one GWPplaced at the centre of each well.To determine the time-development of the variational parameters we make use ofthe time-dependent variational principle (TDVP) in the formulation of McLachlan [19] I = || i φ − H Ψ( t ) || = min , (4) ipolar Bose-Einstein condensates in triple-well potentials φ is varied and set φ ≡ ˙Ψ afterwards. The variational wave function Ψ = (cid:88) k =1 g k (5)is then inserted into (4) yielding the equations of motion (EOM) for the variationalparameters ˙ z k = f (cid:0) z k ( t ) (cid:1) = f (cid:0) A k ( t ) , q k ( t ) , p k ( t ) , γ k ( t ) (cid:1) . (6)Details can be found in [18], where the same method has been used to describe thecollision of quasi-2 d anisotropic solitons. Note that the method for the computation ofthe dipole integrals (cid:104) V dd (cid:105) in (1) slightly differs from the method used in [18]. In that worka strong confinement of the external trap is assumed in the y -direction perpendicularto the alignment of the dipoles, whereas here a strong confinement is assumed in the z -direction parallel to the alignment of the dipoles.The stationary states are the fixed points of (6) and can be determined by a nonlinearroot search (e.g. Newton-Raphson). An alternative to find the real ground state is theapplication of ITE to the EOM. However, as discussed in section 2.1 the ITE does notalways converge to the ground state. In particular, if the initial wave function is close tothat of an excited state, the ITE will stay for a rather long period in imaginary time onan plateau of almost the same mean-field energy. In practice, it is not always possible todistinguish between that case and the convergence to the ground state. To evolve theEOM in imaginary time as well as in real time, a standard algorithm like Runge-Kuttacan be used. For a more detailed description see Ref. [18].The linear stability of the fixed points can be investigated by the calculation of theeigenvalues Λ = Λ r + iΛ i of the Jacobian J = ∂ (cid:16) Re ˙ A k , Im ˙ A k , ˙ q k , ˙ p k , Re ˙ γ k , Im ˙ γ k (cid:17) ∂ (Re A j , Im A j , q j , p j , Re γ j , Im γ j ) , (7)with k, j = 1 . . . . The eigenvalues appear in pairs of opposite sign and correspond toexcitations described by the Bogoliubov-de Gennes equations [16, 17, 20]. If all real parts Λ r = 0 , the fixed point is stable, otherwise it is unstable.
3. Results
Placing the wells in the attractive configuration (dipole alignment →→→ ) enforces theatoms to occupy the centre well and does not show the same diversity of phases as therepulsive configuration ( ↑↑↑ ) does. We will therefore focus on the repulsive configuration,which is visualised in figure 1 and offers the possibility to investigate all effects of interest.For the TW potential V TW in (1) the identical parameters V = 80 (this correspondsto ∼ recoil energies [21]), ω x = ω z = 1 / , and ω y = 4 as given in [10] have beenused. Figure 2 shows the phase diagram for the repulsive configuration obtained by gridcalculations. ipolar Bose-Einstein condensates in triple-well potentials N a dd N a N a dd N a − P c M SU ˜ T T P Figure 2.
Phase diagram for the repulsive configuration. Coloured areas depict regionsof parameter space where the ITE converges, whereas black areas indicate regions wherethe ITE does not converge. The colour bar represents the occupation of both outlyingwells − P c , where P c is the population of the centre well. For details see the discussionin the text. Our aim here is to understand the nature of the distinct regions and the mechanismsbehind the phase transitions. The regions of the parameter space where the ITE convergesare depicted in grey-scale (colour), whereas the black-coloured area U depicts the regionwhere no convergence occurs. The grey-scale (colour) bar on the right-hand side showsthe occupation of the two outlying wells with P c being the occupation of the innerwell. This means that the strip with − P c ≈ . for dipole strengths N a dd (cid:38) . represents states where most of the particles are located in the outlying wells. Areaswith − P c ≈ / represent states where all three wells are equally occupied.The phase diagram shows an interesting feature for N a dd (cid:46) . : At some criticalvalue of the scaled scattering length N a , the state, which the ITE converges to, exhibitsa qualitative sudden change. Here the number of time steps until a given criterion forthe convergence of the ITE is satisfied, reaches a local maximum. Below this criticalscattering length we find an area in the phase diagram, marked as M , where − P c ≈ . viz. almost all of the particles are located in the centre well. Decreasing the scatteringlength even more finally leads to the collapse of the condensate.A similar result was presented by Peter et al [10]. The comparison with the phasediagram in [10], which has been calculated for the same set of parameters N a and
N a dd ,shows the following differences regarding the convergence of the ITE. It does not showconvergence of the ITE in the area M , whereas we do not find an additional white strip ipolar Bose-Einstein condensates in triple-well potentials − P c ≈ ) which should be located right next to the light (yellow) strip ( − P c ≈ . )for values of N a dd (cid:38) . . This is not necessarily an inconsistency, but indicates theexistence of metastable states in the system as will be discussed in detail in section 3.2.For a better understanding of the phase diagram in figure 2 and in particular toclarify the nature of the phases U, S, and M and their transitions, we have performed twodifferent vertical cuts at N a dd = 0 . and N a dd = 0 . , marked by the dashed horizontallines. Along these lines, we calculated the stationary points by the use of the variationalansatz enabling us to investigate ground and excited states and their stability. In orderto provide a deeper insight into the dynamical properties beyond the linear vicinity ofthe stationary states we subsequently performed real-time simulations. N a dd = 0 . The results for
N a dd = 0 . are shown in figures 3 and 4. In figure 3 we immediately seethat more than one state can be found. There are two tangent bifurcations T and T where two states emerge, respectively. We will denote these two pairs of states with S and S , and the symmetry-broken states bifurcating from S (see below) with S . Ifone investigates the linear stability by calculating the eigenvalues of the Jacobian in (7)one finds that at the tangent bifurcations all four born states are unstable. For the statesborn at T this stays true for all values of N a , but not for the states emerging at T .In figure 4 the eigenvalues of the Jacobian are plotted for the S -state with thelower mean-field energy. It passes through a pitchfork bifurcation at P and becomesstable for higher values of N a . It can be seen as the stable ground state from there on.The point P is shifted to lower values compared to the border obtained by numericalgrid calculations. However, the exact position of this border depends on the numericalmethod (e.g. the choice of initial conditions for the ITE; cf. [10]). Furthermore, thequantitative results of the variational solution is limited by the restrictions of the ansatzwhich uses only one GWP per well. We expect the variational results to converge withincreasing number of GWPs [17]. In the bifurcation P two more states S are involved.The energies of these states are degenerate and the wave function breaks the symmetryof the trap (for one of them the left well is populated more than the right well, for theother one v.v.). The second state born in T also passes through a pitchfork bifurcationat P but becomes stable only in one of the eigenvalues while other unstable directionsexist.In figure 3c we plot the quantity I k = (cid:10) g k (cid:12)(cid:12) g k (cid:11) which is a good estimate for thepopulation of the k th well, if we assume a small overlap of the GWPs. The statesemerging at T show that the middle well is hardly populated. Nearly all particles are inthe outer wells. We will call this a split-state. This is not the case for both of the statesemerging at T , where (for the state with the lower mean-field energy) some particles arein the middle well. Note that the split-state is passing through such a symmetry-breakingpitchfork bifurcation at P as well. In fact, all states we investigated show this kind ofsymmetry-breaking behaviour, yet we did not analyse those in all cases. ipolar Bose-Einstein condensates in triple-well potentials -46-45.5-45 E m f T T P P (a) -45.4-45.35-45.3-0.05 -0.04 P -52-50-48-46-44 µ T (b) -45-44-0.06 -0.04 -0.02 T P P P P o pu l a t i o n N a (c) S S S P c (numerical) I ( S ) I ( S ) I ( S ) Figure 3. (a) Mean field energy and (b) chemical potential as functions of the scaledscattering length
N a for dipolar interaction
N a dd = 0 . . The insets show magnificationsof the rectangles. (c) Overlap integral I = (cid:10) g (cid:12)(cid:12) g (cid:11) which shows the population ofthe centre well. All states shown in (a) are represented. The dots show the results ofthe grid calculations obtained by the ITE. The identically scaled absorption imagesshow the qualitative shape of the wave function at the positions, where the arrows arepointing to. An intriguing feature is the occurrence of a region between P and the crossing ofthe mean-field energy of the lower state of S and S , where we find an unstable statewith a lower energy than the stable ground state (see inset in figure 3a). This is onlypossible due to the nonlinearity of the GPE. Nevertheless, it is interesting to performreal- and imaginary-time calculations in this region. It turns out that the ITE convergesfor a wide range of initial conditions to the stable ground state. However, if one startsclose to the unstable state one finds a large plateau in the ITE, which is an indicator forthe existence of metastable states.The real-time evolution of the unstable states reveals that all of them have a small ipolar Bose-Einstein condensates in triple-well potentials -8-4048 Λ r -8-4048 -0.05 -0.04 -0.03 -0.02 -0.01 0 Λ i N aP T Figure 4.
Eigenvalues Λ of the Jacobian (7) for the state S with the lower mean-fieldenergy (see figure 3). The upper panel shows the real parts, the lower panel shows theimaginary parts. The state emerges unstable at the tangent bifurcation T which canbe seen from the non-vanishing Λ r . At the pitchfork bifurcation P the state becomesstable and can be viewed as the stable ground state from there on. life time and mostly end in the collapse of the wave function. Calculations in which wedecreased the scattering length as a function of time from the stable ground state tovalues of N a below P showed a similar behaviour. The dynamical properties becomemore complex at N a dd = 0 . and will be discussed in more detail in section 3.2 wherewe will find a different scenario.The results we have found provide a better understanding of the ITE-behaviour. Inthe calculation of the phase diagram in figure 2 for each N a the result of the ITE atthe previous value of
N a has been taken as the initial wave function. Obviously, thisprocedure gives rise to difficulties if crossings of states and metastable states are involveddue to the small damping of states with almost the same energy as the ground state.
N a dd = 0 . We performed the second cut at
N a dd = 0 . . The results for the mean-field energy,chemical potential and the population of the centre well are shown in figure 5a-c,respectively. There are essentially three states ˜ S , ˜ S , and ˜ S emerging in the tangentbifurcations ˜ T , ˜ T , and ˜ T , respectively. The ˜ S -state with the lower mean-field energyborn at ˜ T is the stable ground state. It is stable for the whole range of N a . The criticalpoint where it becomes unstable therefore is given by the tangent bifurcation ˜ T (andnot by any stability change in a pitchfork bifurcation, as for N a dd = 0 . in figure 3).Furthermore, the energy of the ground state stays the lowest one for all N a , and no ipolar Bose-Einstein condensates in triple-well potentials -50-48-46 E m f ˜ T (a) ˜ T -0.06 -0.03-60-55-50 µ ˜ T (b) ˜ T -0.06 -0.03 0˜ T ˜ T ˜ T P o pu l a t i o n N a (c) ˜ S ˜ S ˜ S P c (numerical) I ( ˜ S ) I ( ˜ S ) I ( ˜ S ) Figure 5. (a) Mean-field energy and (b) chemical potential as functions of the scaledscattering length
N a for dipolar interaction
N a dd = 0 . . (c) Overlap integral I (c.f.figure 3) which shows the population of the centre well. The absorption images show thequalitative shape of the wave function at the positions, where the arrows are pointingto. other state is crossing. The excited states emerging at ˜ T and ˜ T have been omittedfor reasons of clarity. While ˜ S is a split-state for all N a , ˜ S shows some interestingbehaviour when it becomes energetically close to the ground state. It passes throughtwo consecutive tangent bifurcations ˜ T and ˜ T . It can be seen in figure 5c that thisinvolves a qualitative change of the wave function’s nature from a state, where all wellsare populated equally to a split-state.It is a remarkable fact that we find (unstable) states for values of N a where in ipolar Bose-Einstein condensates in triple-well potentials -80-70-60-50-40-30-20-10 0 0.1 0.2 0.3 0.4 0.5 0.6 E m f τψ i : variational solution ψ i : single Gaussian Figure 6.
Mean-field energies as functions of imaginary time τ for two grid-ITEsat N a dd = 0 . , N a = − . , where we have used as initial wave functions ψ i a singleGaussian covering all wells for the calculation plotted as green dashed line and thevariational solution of the state ˜ S plotted as red solid line. figure 2 no ground state could be found, at all. More precisely, these states exist farbelow the critical scattering length at the tangent bifurcation ˜ T . This suggests a relationto the occurrence of regions in the parameter space such as the area M . The variationalcalculations predict that no stable ground state is present in the area M . Although thegrid-ITE seems to converge in this area, this could still be just an effect of metastablestates being present. This is illustrated in figure 6, where grid-ITEs with two differentinitial states Ψ i are shown for N a = − . , which is in an area where no stable groundstate exists. The closer the initial state to the metastable state is, the more pronouncedthe plateau in the ITE becomes. Finally, both calculations diverge, indicating that thestate is unstable.For a deeper insight into the physics of the metastable states we have investigatedthe dynamical properties in these regions. Real-time evolutions in which we change thescaled scattering length N a over time show that the dynamics of the condensate in thisregion of the parameter space differs from the dynamics in other regions.Furthermore, the dynamical tuning of the scattering length can be a procedurein an experiment to access the metastable region. We illustrate this in figure 7a andb for the variational and grid calculations, respectively, where we have calculated theoccupation of the centre well for two different real-time evolutions. The solid blue anddashed green curve represent the real-time evolutions for
N a dd = 0 . and an alterationof N a from − . to − . and from − . to − . , respectively. For the solid (blue) ipolar Bose-Einstein condensates in triple-well potentials I (a)0.40.50.60.70.80.9 0 50 100 150 200 250 300 350 400 450 500 P c t (b) I , N a : − . → − . I , N a : − . → − . P c , N a : − . → − . P c , N a : − . → − . Figure 7.
Real-time evolution for two transitions of
N a . The scaled scattering length
N a is tuned linearly from t = 0 to its final value at t = 200 and kept constant fromthere on. (a) Variational calculation with the overlap integral (introduced in figure 3c)of the middle well. (b) Population of the centre well obtained by the grid calculationsis plotted as a function of time. In both panels (a) and (b) the solid blue line belongsto the calculation, where N a is tuned from − . to − . and the dashed green linebelongs to the transition from N a = − . to N a = − . . curve, the alteration of the scattering strength N a starts and ends in the region S (seefigure 2) of the parameter space. This leads to small oscillations in the occupation of thecentre well as soon as
N a is adjusted to its final value at t = 200 . For the dashed (green)curve small oscillations are visible as soon as we change the scattering length. Here, thealteration of N a ends in regions of the parameter space where the phase diagram givenin [10] and the one presented in figure 2 differ. In this case, during the ramp-down of thescattering length the qualitative shape of the wave function changes significantly andlarge periodic oscillations set in. This indicates the phase transition from region S to Min figure 2. We conclude from these calculations that the metastable states can preventthe condensate from the collapse in some cases where no stable ground state exists.In an experiment absorption imaging of the condensate during the ramp-down ofthe scattering length could reveal the phase transition. For an experimental setup weadopt the experimental parameters of Ref. [9] where a spacing of l = 1 . µ m is suggested.This setup can be realised by the use of Cr atoms where a dd ∼ . nm and a ∼ . nm.The presented transition in parameter space would be equal to a number of 420 Cratoms where the scattering length would be tuned over a time of 475 ms via Feshbachresonances [22] to its final value. This alteration of the scattering length though can be ipolar Bose-Einstein condensates in triple-well potentials
4. Conclusion
We studied the ground and metastable states of dipolar BECs in triple-well potentialsboth with a full-numerical ansatz and a time-dependent variational principle withcoupled Gaussians. Although the triple-well potential constitutes a very simple system,the occurrence of different phases and their stability properties appears to be quitecomplicated. The phase diagram presented depicts a region of the parameter space,where multiple phases occur. This includes states where all wells are equally occupiedas well as states where more particles are located in the centre or the both outerwells. The variational solutions reveal a variety of excited and unstable states includingsymmetry-broken states and unstable states in regions where no stable ground stateexists. Real-time evolutions and the eigenvalues of the Jacobian J , given in (7), bothallow for predictions about the stability of the investigated states. Moreover, the formeris able to show whether the occupation of the wells for metastable states is characterisedby quasi-periodic, chaotic oscillations, a break of symmetry, or whether the instability ipolar Bose-Einstein condensates in triple-well potentials I k P tI (left) I (center) I (right) P l P c P r Figure 8.
Real-time evolution of a metastable state for
N a dd = 0 . and N a = − . .The variational approach has been used for the calculation shown on the upper panel.The figure shows the overlap integrals I k as already defined in figure 3. The lower panelshows the results of the grid calculation for the population P of the left, centre, andright well, respectively. leads to a collapse. We have pointed out that a dynamical stabilisation of the condensateby the interplay of the metastable states is possible in those regions. Therefore, theseregions are best candidates for the observation of a supersolid phase.Further investigations should include multi-well potentials with additional wells anda different arrangement like a triangular or ring-like configuration. Our results shouldstimulate experimental efforts to study dipolar BECs in multi-well potentials. Acknowledgements
We thank David Peter and Tilman Pfau for valuable discussions. R.E. is grateful forsupport from the Landesgraduiertenförderung of the Land Baden-Württemberg. Thiswork was supported by Deutsche Forschungsgemeinschaft.
References [1] Griesmaier A, Werner J, Hensler S, Stuhler J and Pfau T 2005
Phys. Rev. Lett. Phys. Rev. A Rep. Prog. Phys. Phys. Rev. Lett. ipolar Bose-Einstein condensates in triple-well potentials [5] Lu M, Burdick N Q, Youn S H and Lev B L 2011 Phys. Rev. Lett.
Phys. Rev. Lett.
Science
Phys. Rev. Lett. Phys. Rev. Lett.
J. Phys. B: At. Mol. Opt. Phys. Phys. Rev. A Phys. Rev. A Phys. Rev. A J. Phys. B: At. Mol. Opt. Phys. Phys. Rev. A Phys. Rev. A Phys. Rev. A Phys. Rev. A Mol. Phys. J. Phys. B: At. Mol. Opt. Phys. Nat Phys Nature448