ebook img

Interacting Bosons beyond the Gross-Pitaevskii Mean-Field PDF

0.25 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Interacting Bosons beyond the Gross-Pitaevskii Mean-Field

9 0 InteractingBosonsbeyondtheGross-PitaevskiiMean-Field 0 2 M.HeimsothandM.Bonitz n Christian-AlbrechtsUniversit¨atzuKiel,Institutfu¨rTheoretischeundAstrophysik,24098Kiel,Germany a J 3 1 Abstract ] l SystemsconsistingofcoldinteractingbosonsshowinterestingcollectivephenomenasuchasBose-Einsteincondensationor l a superfluidityandarecurrentlystudiedincondensedmatterandatomicphysics.Ofparticularinterestarenonidealbosons h whichexhibitstrongcorrelationandspatiallocalizationeffects.Hereweanalysetheground-stateofatwo-dimensionalBose - systemwithaHartree-FocktypeapproximationthatwasfirstintroducedbyRomanovskyetal.[Phys.Rev.Lett.93,230405 s e (2004)].Weapplythismethodtoaonedimensionalsystemofchargedbosonsandanalyzethebehaviouratstrongcoupling. m . t Keywords: bosesystems,Hartree-Fock a m - d 1. Introduction tionstrength. n Inthemany-bodytheory,Hartree-Fockisastandard o c With the first experimental realisation of a Bose- method to analyse systems of few to many fermions. [ Einsteincondensatein1995[1],thetheoryofultracold Therelatedequationscanbederivedinseveraldiffer- bosonicsystemsbecameafieldofexceptionalinterest. ent ways. One is obtained by expanding the reduced 1 v Inthefirstexperiments,theatomswereweaklyinter- two-particle density matrix in terms of one-particle 2 acting,andforsuchsystemstheGross-Pitaevskiimean densitymatricesinordertoobtainaneffectivesingle- 5 fieldapproximation(GP)isanadequateapproachfor particlehamiltonian.Theresultingsolutionstothisap- 8 theanalysisofthecondensate[2].However,recentex- proach are many-particle states consisting of one sin- 1 perimentssucceededinstronglyincreasingtheinterac- gleanti-symmetrisedproduct-state.Anothercommon . 1 tion strengths of the investigated atomic systems [3]. wayproceedsintheoppositedirectionbytheassump- 0 With this tuning of the coupling, it was possible, to tion, that the many-particle state is given by a single 9 0 produce a phase transition from the superfluid to the anti-symmetrised product state (Slater determinant). : Mott insulator phase [4]. Also, for charged bosons in By applying the Ritzian principle to this ansatz, one v traps crystallization and inhomogeneous distribution obtainstheHartree-Fockequations. i X ofthesuperfluiddensityhasbeenpredicted[5].Fora In contrast to Fermi systems, for bosonic particles r theoretical description of these effects an approxima- the two approaches lead to different systems of equa- a tion beyond GP is required [6,7,8]. In this paper we tions.EventheansatzassumingasingleSlaterperma- analyseanapproximationmethodforbosonicsystems nent can be implemented in various ways differing in which is based on a Hartree-Fock type factorization thechoiceofsupplementaryconditions[6,7]. andisexpectedtobevalidforawiderangeofinterac- Thebestknownapproximationforanultracoldnon- ideal dilute Bose gas is the Gross-Pitaevskii approxi- PreprintsubmittedtoPhysicaE 13January2009 mationinwhichitisassumed,thatallparticlesremain 2. UnrestrictedbosonicHartree-Fockansatz in the same one-particle orbital. Thus the GP equa- (UBHF) tiondescribesonlythecondensatefractionofthesys- temwithoutanyinteractionwiththeremainiggas.By The UBHF method is derived from the following definitiontheGPapproximationcannotdescribephe- ansatzforthemany-particlestate nomena such as condensate depletion and fragmenta- N 1 X O tionorMott-insulatorphasetransitionsofcoldatomic |Φ(cid:105)=|12...N(cid:105):= √ |π(s)(cid:105), (1) gasesonalattice.Theresultsbecomeevenworseifthe N!π∈SN s=1 interactions increase or if systems with small particle whichmeansthateveryparticleremainsinacertainor- numbers are considered. As we will see later, this ap- bitalandtheunderlyingmany-particlestateisasym- proximationcannotdescribeanumberoffundamental metrisedproduct-state.Dependingontheimposedad- properties, such as the localisation of the particles at ditional restrictions, this ansatz contains, as limiting highcouplingstrengths. cases, well-known approximations. In particular, the Anothercommonnumericalmethodtodealwithin- GP approximation is obtained by the additional re- teractingsystemsistheConfigurationInteraction(CI) striction, that all orbitals are identical [8], thus the inwichnoapproximation,exceptthelimitationofthe many-particlestateisassumedtobetotallyBosecon- chosenbasisset,ismade,e.g.[9].Withnowadayscom- densed. A more general approximation would be ob- puters and highly optimised programs CI can be ap- tainedbytheassumption,thattwoorbitalsareeither pliedtosystemswithupto10particles. equalororthogonaltoeachother.Thiskindofansatz In this paper we want to analyse an approximation hasbeenintroducedandanalysedextensivlybyCeder- with substantially less numerical effort than CI, that baumetal.[6,12,13,14]. provideshighqualitynumericalresultsforstronglycor- In contrast, the UBHF approximation is the most related bosons in traps. This approximation, the Un- generalcaseofthisAnsatzwherenofurtherrestriction restrictedbosonicHartree-Fockmethod(UBHF),was to the underlying one-particle orbitals |1(cid:105),...,|N(cid:105) is first introduced by Romanovsky et al. [10,7,11] and imposed. We only require the many-particle state to is less restricted than the related ansatz proposed by benormalised,sothetotalenergyisgivenby Cederbaum [6]. While Romanovsky et al. used an ex- E =(cid:104)Φ|Hˆ|Φ(cid:105). (2) plicit analytical ansatz for the single-particle orbitals (discplaced Gaussians) we develop a completely gen- ThusminimisingE leadstothefunctional eralschemewithoutanysuchrestriction.Thishasthe E(|1(cid:105),...,|N(cid:105),E)=(cid:104)Φ|Hˆ|Φ(cid:105)−E((cid:104)Φ|Φ(cid:105)−1), (3) advantagethatourmethodisapplicabletoanyinter- whereE isaLagrangemultiplyerforthenormalisation actingBosesystem. of |Φ(cid:105). The needed equations to determine the one- Furthermore, our goal is to compare the UBHF re- particle orbitals are obtained by applying the Ritzian sults with CI results to obtain a quantitative conclu- principletothisAnsatzviathesearchedorbitals.This sion about the accuracy of this method. To this end ansatzhasbeenproposedandimplementedforthefirst weperformUBHFandCIcalculationsforN =2...6 timebyRomanovskyetal.[11,10],however,theyused particlesinaharmonictrapusingthesamebasissets the additional assumption that every orbital is a dis- for both cases. Our comparisons concentrates on sys- placedGaussian.Inthepresentpaper,thisrestriction temswithacouplingparameterintherangefromzero willbedropped. tofiveandshowstheexcellentqualityofthisapproxi- Theexpectationvalueoftheenergyofanormalised mation. singlepermanentmanyparticlestatewithinteracting particlesisgivenby[15] (cid:104)Φ|Hˆ|Φ(cid:105)= √1 X XN “Y(cid:104)s|π(s)(cid:105)(cid:104)l|hˆ|π(l)(cid:105) N! π∈SN l=1 s(cid:54)=l 1X Y ” + (cid:104)s|π(s)(cid:105)w . (4) 2 kl,π(k)π(l) k(cid:54)=ls(cid:54)=k,l 2 The norm of such a productstate is proportional to 25 UBHF the permanent of the Gramian matrix of the orbitals CI GP |1(cid:105),...,|N(cid:105): 20 N =4 N (cid:107)Φ(cid:107)2 = √1 X Y(cid:104)s|π(s)(cid:105). (5) ω 15 N! h¯ π∈SNs=1 Ein 10 N =3 It is advantageous to stay in the abstract representa- tion of the orbitals to perform the variational deriva- 5 N =2 tive. By applying the rules given in Appendix A one obtainsthefollowingequation,bydifferentiating(3) 0 ( 0 1 2 3 4 5 X P hˆ|π(n)(cid:105)+XP (cid:104)l|hˆ|π(l)(cid:105)|π(n)(cid:105) λ n nl π∈SN l(cid:54)=n ) Fig. 1. Comparison of the energies obtained with UBHF, +XP Jˆ |π(n)(cid:105)+1 X P w |π(n)(cid:105) GPandCIfordifferentcouplingparametersλandfor2,3 nl lπ(l) 2 nlk kl,π(k)π(l) and4particles. l(cid:54)=n k,l(cid:54)=n k(cid:54)=l X =E Pn|π(n)(cid:105), foralln, (6) xisgiveninoscillatorunitsx =p(cid:126)/mω,andλisthe π∈SN 0 ratiooftheCoulombandconfinementenergy whereweintroducedthesubscriptedentitywithavari- ationalnumberofsubscripts q2 λ= . (11) Pi1...in =P(π)i1...in = Y (cid:104)s|π(s)(cid:105), (7) 4π(cid:15)x0(cid:126)ω s(cid:54)=i1...in Thusλ=0correspondstoanidealsystem.Theshield- andthedoublysubscriptedHartreeoperator ing parameter κ is needed in 1D to make the two- Z Jˆ = dxdyϕ∗(x)w(x,y)ϕ (x)|y(cid:105)(cid:104)y|. (8) particle integrals convergent. For the limit κ −→ 0 ij i j the system becomes fermionised [16]. For all calcula- Bymultiplyingequation(6)fromtherightwith(cid:104)n|, tionsinthispaper,wechooseκ=0.1.Asbasisforthe oneobtainsaclosedexpressionfortheLagrangemul- single-particle orbitals we chose the eigenfunctions of tiplyer theidealsystemwithatotalnumberofbasisstatesof E = (cid:104)Φ|Hˆ|Φ(cid:105), (9) nb = 15. We will compare the results with the exact (cid:104)Φ|Φ(cid:105) solutionobtainedwithCI[9].TheCIcalculationsare i.eE isthetotalenergyoftheSystem.Thekeyequation doneinexactlythesamebasis,thusthedifferencesin (6)canbesolvedwithamultidimensionalminimisation theresultsariseexclusivelyfromtheUBHFansatz. routinewithoutmakinganyassumptionontheexplicit analyticalformoftheorbitals. 3.1. Totalenergy 3. Numericalresults InFig.1theenergiesobtainedbybothmethodsare compared.Interestingly,thedifferencebetweentheex- We applied the UBHF approximation to few inter- act method and the UBHF approximation becomes a actingbosonsinaone-dimensionalharmonictrapde- constant shift for high interaction values, λ(cid:38)2. This scribedbythehamiltonian shift still depends on the considered particle numbers N. As one can see in 1, the shift grows with N. Note Hˆ =X12“x2i−∂∂x22”+Xp(x −λx )2+κ2, (10) thatforGPtheenergydivergesrapidlyfromtheexact i i i<j i j resultalreadyforλ>0.5.ThusFig.1isaconvincing whichhasbeenmadedimensionlessbyusingstandard evidenceforthegoodqualityoftheUBHFapproxima- oscillatorlengthandenergyscales:thespatialvariable tion. 3 0.9 Coulomb interacting trapped bosons. As another ex- CI 0.8 UBHF ample, in Fig. 4 we show the density of six bosons GP 0.7 ϕ (x) fortwovaluesofthecouplingparameter.Again,with 1/2 0.6 increasing λ the overlap of the orbitals decreases and thesystem,asawholeexpands. y 0.5 t si n 0.4 e d 3.3. Delocalisation and non-orthogonality of the 0.3 orbitalsforsmallλ 0.2 0.1 Thereareseveralpossibilitiestoanalysetheoverlap 0 3 2 1 0 1 2 3 of orbitals in a quantitative way. One that works for − − − xinx allparticlenumbersistoconsidertheGramiandeter- 0 minantofthegivensetoforbitals, Fig.2.ParticledensityofasystemwithN =2andλ=0.9 for the three different approximation methods. The two X YN orbitalsobtainedwithUBHFarealsogiven. G(|1(cid:105),...,|N(cid:105))= sign(π) (cid:104)s|π(s)(cid:105) π∈SN s=1 0.8 CI ˛˛(cid:104)1|1(cid:105) (cid:104)1|2(cid:105) ... (cid:104)1|N(cid:105)˛˛ 0.7 UBHF ˛ ˛ GP ˛ ˛ 0.6 ϕ1/2(x) ˛˛(cid:104)2|1(cid:105) (cid:104)2|2(cid:105) ... (cid:104)2|N(cid:105)˛˛ =˛ ˛ (12) 0.5 ˛˛ ... ... ... ... ˛˛ ˛ ˛ nsity 0.4 ˛˛˛(cid:104)N|1(cid:105) (cid:104)N|2(cid:105) ... (cid:104)N|N(cid:105)˛˛˛ e d 0.3 Thisentityisalwayspositiveandapproachesthevalue 0.2 1,iftheorbitalsformanorthonormalset(thisismost 0.1 easilyseenforthecaseoftwoorbitals).Geometrically, 0 theGramiandeterminantisthesquareofthevolumeof 3 2 1 0 1 2 3 − − − theparallelepipedspannedbythevectors|1(cid:105),...,|N(cid:105) xinx 0 [21]. In Fig.5 the λ-dependency of G is shown. Obvi- Fig.3.SameasFig.2,butforλ=1.5 ously,inthelimitoflargeλtheorbitalsbecomepair- wiseorthogonal.Withincreasingparticlenumberthis 3.2. Localisationoftheorbitalsforλ(cid:29)1 limitisreachedforlargervaluesofthecouplingparam- eter. The UBHF approximation scheme offers the possi- Letusnowconsidertheoppositelimitofsmallcou- bility to analyse the delocalisation of the interacting pling. As Fig. 5 shows, for all N, the determinant G particles in a special way. As can be seen in Figs. 2, monotonicallydecreaseswhenλisreduced.SinceGis 3 and 4, the overlap of the orbitals vanishes with in- a determinant it is obvious that it vanishes, if one or creasingλ.Itiscrucialtonotice,thatthedensityfrom moreorbitalsarecollinear.Inthepresentcase,alsothe UBHF shows a localisation of the particles which, by oppositeistrue:vanishingofGisanindicationofthe definition, is missing in the GP model. Finally, the orbitals becoming collinear. This is observed for λ → othertwocurvesinFigs.2and3whichresembleGaus- 0.Duetothesymmetryofthesystemnotonlytwoor- siansarethetwoorbitalsobtainedbyUBHF.Itisin- bitalsbutallorbitalsarebecomingcollinear,i.e.they teresting to see that the localisation of the particles areidentical.ButthisisjustthelimitofanidealBose emerges already in each of the orbitals. This can be gas where we expect that all particles Bose condense understoodasaprecursoroftheclassicalstronglycor- inthegroundstate.Itisaremarkablepropertyofthe related limit of the system where the particles form a present UBHF ansatz which does not impose any re- fullylocalizedcrystal-likearrangement[17,18]. strictionsontheorbitalsthatityieldsthecorrectBEC This trend is typical for the present system of limitofidenticalHartreeFockorbitalsforallparticles. 4 1.0 λ=0.5 chargedbosonsintrapswhenthedimensionlessinter- x) action strengths λ exceeds 0.5. For its resulting ener- φ( 0.5 giesaresubstaniallyhigherthantheexactonesandit cannotreproduceimportantqualitativefeaturesofthe 0.0 λ=1.0 systemsuchasthelocalisationofthetotalparticleden- φx() 0.5 sity.InthelimitofvanishingλtheUBHFapproxima- tionyieldssingleparticleorbitalswhicharecollinear– 0.0 inotherwords,weautomaticallyrecovertheGPmodel. Wementionthatthecomputertimerequiredtosolve x) 1.0 ρ( 0.5 λ=0.5 theequations(6),withinourcurrentimplementation, λ=1.0 hasanunfavorabledependenceontheparticlenumber 3 2 1 0 1 2 3 N scaling as N ·N!. This arises from performing the − − − xinx sumoverallpermutationsin(6).However,itispossi- 0 ble to reduce the calculation of the entities that turn Fig.4.Connectionoftheobtainedorbitals(uppertwofig- up in this equation to the calculation of permanents. ures) and the resulting total particle density of a system By using the Ryser algorythm, the N-dependency of withN =6particlesandλ=0.5and1.0. the complexity becomes of order N ·2N [19]. This is 1 stillarapidlygrowingcalculationtime,butitismuch 0.9 betterthanthefirstversion.Thecalculationtimealso 0.8 dependsonthesizeofthechosenbasissetnb.Forboth 0.7 implementationsitisofordern4.Thisarisesfromthe b 0.6 calculation of the two-particle-integrals in each step G 0.5 [20]. 0.4 Presentlyweareworkingonfurtheroptimizingthe N 0.3 2 implementationofthisansatzinordertoextenditto 0.2 3 systems with higher particle numbers and higher di- 4 0.1 5 mensions. Furthermore, the development of anexten- 6 0 siontothetimedependentregimeisinprogress. 0 1 2 3 4 5 λ Fig. 5. Gramian determinant of the set of orbitals for dif- ferentparticlenumbersversusλ Acknowledgements We thank K. Balzer for useful 4. Discussion discussions. In this paper we have studied a Hartree-Fock ap- proximationforinteractingbosonsinatrapwhichhas proposed by Romanovsky et al. In contrast to their Appendix A. Variational differentiation of workwehavenotusedanyassumptionontheexplicit abstractHilbertspacevectors formoftheorbitalsbutobtainedthemselfconsistently. WehavetestedtheUBHFschemeindetailbycompar- In order to obtain the key equations for the UBHF ing the resulting energies and densities with the ones approximation,onehastoperformavariationalderiva- obtained with CI. Due to the identical choice of the tive with respect to all N orbitals. In this appendix basis set for both approximations, the differences in wederivesomedifferentiationrulesthatareneededto theresultsariseexclusivelyduetotheapproximation deduce(6).Inthefollowingthewavefunctionofthen- made in UBHF. The comparison shows that this is a th orbital will be denoted ϕ (x). Let us now consider n very accurate model. In contrast, it is confirmed that the variational derivative of the matrix element of an the GP approximation is not applicable to nonideal arbitraryoperatorOˆ: 5 δ δ Z (cid:104)i|Oˆ|j(cid:105)= ϕ∗(z)O(x,y)ϕ (y)dxdy δϕ∗(x) δϕ∗(x) i j n n Z =δ O(x,y)ϕ (y)dy=δ (cid:104)x|Oˆ|j(cid:105) References in j in “ δ ” [1] M.H.Anderson,J.R.Ensher,M.R.Matthews,C.E. =(cid:104)x| (cid:104)i|Oˆ|j(cid:105) , (A.1) δ(cid:104)n| Wieman, and E. A. Cornell. Science, 269(5221):198, 1995. wherethelattertransformationcanberegardedasthe [2] Franco Dalfovo, Stefano Giorgini, Lev P. Pitaevskii, definitionofthederivativewithrespecttoanabstract and Sandro Stringari. Rev. Mod. Phys., 71(3):463, Hilbertspacevector: 1999. δ [3] S. L. Cornish, N. R. Claussen, J. L. Roberts, E. A. δ(cid:104)n|(cid:104)i|Oˆ|j(cid:105):=δinOˆ|j(cid:105). (A.2) Cornell, and C. E. Wieman. Phys. Rev. Lett., 85(9):1795,2000. Toverifythatthisdefinitionmakessenseweconsider [4] Markus Greiner, Olaf Mandel, Tilman Esslinger, thesamederivativebutinanarbitraryrepresentation. Theodor W. H¨ansch, and Immanuel Bloch. Nature, Therefore, we differentiate the given matrix element 415(6867):39,2002. withrespecttoc∗ –theγ-thexpansioncoefficientof [5] A. Filinov, J. B¨oning, M. Bonitz, and Yu. Lozovik. nγ then-thorbital Phys.Rev.B,77(21):214527–5,2008. [6] L. S. Cederbaum and A. I. Streltsov. Phys. Lett. A, ∂ (cid:104)ϕ |Oˆ|ϕ (cid:105)= ∂ Xc∗ O c 318(6):564,2003. ∂c∗ i j ∂c∗ iα αβ jβ nγ nγ αβ [7] I. A. Romanovsky. PhD thesis, Novel properties of X ∂c∗ X interactingparticlesinsmalllow-dimensionalsystems, = iα O c =δ O c ∂c∗ αβ jβ in γβ jβ GeorgiaInstituteofTechnology2006. αβ nγ β |{z} [8] L. Pitaevskii and S. Stringari. Bose-Einstein =δinδαγ Condensation. Oxford University Press, New York, “ δ ” =δin(cid:104)γ|Oˆ|j(cid:105)=(cid:104)γ| δ(cid:104)n|(cid:104)i|Oˆ|j(cid:105) . (A.3) 2003. [9] Attila Szabo and Neil S. Ostlund. Modern Quantum Forthedifferentiationofthetwo-particleintegrals,we Chemistry: Introduction to Advanced Electronic canusethesameideastoderivethefollowingequation StructureTheory. DoverPublications,1996. [10]Igor Romanovsky, Constantine Yannouleas, and Uzi δ (ij|wˆ|kl)=δ Jˆ |k(cid:105)+δ Jˆ |l(cid:105) (A.4) Landman. Phys.Rev.Lett.,93(23):230405,2004. δ(cid:104)n| in jl jn ik [11]Igor Romanovsky, Constantine Yannouleas, Leslie O. =δ Kˆ |l(cid:105)+δ Kˆ |k(cid:105). in jk jn il Baksmaty, and Uzi Landman. Phys. Rev. Letters, 97(9):090401,2006. Thus we have some freedom to define the remaining [12]O.E.Alon,A.I.Streltsov,andL.S.Cederbaum. Phys. operator.Thefirstonethatappearsinthisequationis Lett.A,347(1-3):88,2005. defined above Eq. (8) whereas the exchange operator Kˆ isgivenby [13]Ofir E. Alon, Alexej I. Streltsov, and Lorenz S. jk Cederbaum. Phys.Lett.A,362(5-6):453,2007. Z Kˆ = dxdyϕ∗(x)w(x,y)ϕ (y)|y(cid:105)(cid:104)x|. (A.5) [14]Alexej I. Streltsov, Ofir E. Alon, and Lorenz S. jk j k Cederbaum. Phys.Rev.A,73(6):063626,2006. Withtheserulesforthefunctionalderivativeofmatrix [15]John W. Negele and Henri Orland. Quantum Many- particleSystems. WestviewPress,USA,1998. elements,togetherwiththeproductrule, [16]M.Girardeau. J.Math.Phys.,1(6):516,1960. δ `F [...,(cid:104)n|,...]·F [...,(cid:104)n|,...]´ [17]A. Filinov, M. Bonitz and Yu. Lozovik, Phys. Rev. δ(cid:104)n| 1 2 Lett.,86:3851,2001 “δF ” “δF ” = 1 ·F +F · 2 , (A.6) [18]M.Bonitzetal.,Phys.Plasmas,15:055704,2008 δ(cid:104)n| 2 1 δ(cid:104)n| [19]H.J.Ryser. CarusMath.Monograph,14,1963. one derives equations (6) by differentiating the func- [20]D.L.HuestisJ.H.HurleyandW.A.Goddard. J.Phys. tional(3). Chem.,92:4880,1988. 6 [21]G. Fischer Lineare Algebra. Vieweg, Brauschweig/Wiesbaden,2005 7

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.