ebook img

Entanglement negativity after a local quantum quench in conformal field theories PDF

0.67 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 Entanglement negativity after a local quantum quench in conformal field theories

Entanglementnegativityafteralocalquantumquenchinconformalfieldtheories Xueda Wen,1 Po-Yao Chang,1 and Shinsei Ryu1 1DepartmentofPhysics, UniversityofIllinoisatUrbana-Champaign, Urbana, IL61801, USA (Dated:August11,2015) Westudythetimeevolutionoftheentanglementnegativityafteralocalquantumquenchin(1+1)-dimensional conformalfieldtheories(CFTs),whichweintroducebysuddenlyjoiningtwoinitiallydecoupledCFTsattheir endpoints.Wecalculatethenegativityevolutionforbothadjacentintervalsanddisjointintervalsexplicitly.For twoadjacentintervals,theentanglementnegativitygrowslogarithmicallyintimerightafterthequench. After developing a plateau-like feature, the entanglement negativity drops to the ground-state value. For the case 5 oftwospatiallyseparatedintervals,alight-conebehaviorisobservedinthenegativityevolution;inaddition,a 1 long-rangeentanglement,whichisindependentofthedistancebetweentwointervals,canbecreated.Ourresults 0 agreewiththeheuristicpicturethatquasiparticles,whichcarryentanglement,areemittedfromthejoiningpoint 2 andpropagatefreelythroughthesystem. Ouranalyticalresultsareconfirmedbynumericalcalculationsbased g onacriticalharmonicchain. u A I. INTRODUCTION respecttoA2’sdegreesoffreedomisdefinedas 8 ] A. Introduction (cid:104)e(i1)e(j2)|ρTA21∪A2|e(k1)e(l2)(cid:105)=(cid:104)e(i1)e(l2)|ρA1∪A2|e(k1)e(j2)(cid:105), h (3) c where |e(1)(cid:105) and |e(2)(cid:105) are arbitrary bases in H and H , e Recently, it has been recognized that quantum entangle- i j A1 A2 respectively. Thenthelogarithmicnegativityisdefinedas m ment provides us a powerful tool to study quantum proper- - ties of many-body systems in condensed matter physics [1– E :=ln||ρT2 ||=lnTr|ρT2 |, (4) at 4]. When the system is prepared in a pure state |Ψ(cid:105), a good A1,A2 A1∪A2 A1∪A2 t quantity that describes the bipartite entanglement is the von whereT indicatesthepartialtranspositionwithrespecttoA , s 2 2 . Neumannentropy,whichisdefinedas andthetracenorm||ρT2 ||isdefinedasthesumofallthe at absolute values of theAe1i∪gAen2values of ρT2 . Recently, the m S =−Trρ lnρ (1) A1∪A2 A A A logarithmicnegativityhasbeenextensivelyusedtostudyvari- - ousmany-bodysystems,includingone-dimensionalharmonic d n where ρA = TrBρ is the reduced density matrix of subsys- chains[9,10],quantumspinchains[11–17],freefermionsys- o temA,withρ = |Ψ(cid:105)(cid:104)Ψ|. Analternativemeasureofbipartite tems[18],andtopologicallyorderedsystems[19,20]. Inpar- c entanglementinpurestatesistheRenyientropy ticular, the universal features of the entanglement negativity [ inone-dimensionalcriticalsystemshavebeenunderstoodby 2 S(n) = 1 lnTrρn. (2) developingaconformalfieldtheory(CFT)approach[21,22]. v A 1−n A Lateron,thecomparisonofCFTresultsandnumericalcalcu- 8 lations of one-dimensional critical systems were studied in a 6 Theseentanglementmeasureshaveprovedtobeofgreatuse seriesofworks[23–25]. 5 incharacterizingquantumentanglementofmany-bodystates. Althoughmanyworkshavebeendoneontheentanglement 0 However, for a mixed state, neither the von Neumann en- negativity,thereislessunderstandingonthenon-equilibrium 0 tropy nor the Renyi entropy is a good measure of entangle- propertiesoftheentanglementnegativity. Mostrecently,time . 1 mentsincequantumandclassicalcorrelationsarenotclearly evolution of the logarithmic negativity after a global quench 0 separatedinthesemeasures.Nowsupposeweareinterestedin wasstudiedwithCFTapproach[26]. InRef.[27], thenega- 5 theentanglementbetweentwosubsystemsA andA ,which tivityevolutionfortwoadjacentintervalsafteralocalquench 1 1 2 : arenotnecessarilycomplementarytoeachotherand,areem- was numerically studied in a harmonic chain. However, a v bedded in a larger system, the union A ∪A cannot be de- thoroughstudyofthenegativityevolutionafteralocalquan- 1 2 i X scribedbyapurestateafterintegratingoutdegreesoffreedom tum quench is still lacking, and it is appealing to unveil the inthecomplementofA ∪A . Inthiscase,weneedtosearch universalfeaturesofthedynamicalbehavioroftheentangle- r 1 2 a for other quantities that may characterize quantum entangle- mentnegativitypropagation. ment for a general mixed state. Among different proposals Inthispaper, ourmotivationistostudythetimeevolution [5,6],acomputablemeasurementofentanglement,theloga- of the entanglement negativity after a local quantum quench rithmic negativity [7], turns out to be very useful and practi- analytically. Forsimplicity,weconsidera(1+1)-dimensional cal. In particular, it is proved that the logarithmic negativity criticalsystem,whichisphysicallycutintotwopartsthatare is a proper entanglement monotone in Ref. [8]. Following preparedintheirowngroundstates. Thenattimet = 0, we Ref. [7], the negativity can be obtained by first taking a par- join the two parts together at their endpoints, and study the tialtransposition. Tobemoreprecise,givenadensitymatrix time-evolutionoftheentanglementnegativityafterwards. As ρ which describes a bipartite mixed state in a Hilbert shown in Fig. 1, once the two CFTs are joined at the end- A1∪A2 spaceH = H ⊗H ,thepartialtranspositionwith points,theinteractionbetweenthemisintroducedsimultane- A1∪A2 A1 A2 2 intervalτ ∈[0,β]: 1 (cid:90) (cid:89) ρ= [dφ(x,τ)] δ(φ(x,0)−φ(cid:48)(x(cid:48))) Z x (6) (cid:89) × δ(φ(x,β)−φ(cid:48)(cid:48)(x(cid:48)(cid:48)))e−SE, x wheretherowsandcolumnsofthedensitymatrixarelabeled by the fields {φ(x,τ)} at τ = 0 and β respectively, with β beingtheinversetemperature,S istheEuclideanactionand E Z =Tre−βH isthepartitionfunction. Nowweconsidersub- FIG.1: Setupforalocalquantumquench. TwoseparateCFTsde- systems A1 and A2 located in intervals [u1,v1] and [u2,v2], finedontwosemi-infinitelinesarejoinedtogetherattheirendpoints. respectively. ThenthereduceddensitymatrixρA1∪A2 maybe Then quasiparticles, which may be viewed as entangled pairs, are obtainedbysewingtogetherallthepointsalongedgesτ = 0 generatedatthejointingpoint,andpropagatefreelythroughthesys- andτ =βexceptthepointsinA ∪A . Thatis,weleavetwo 1 2 tem. Theentanglementnegativitybetweentwointervalswhichare opencutsat[u ,v ]and[u ,v ]alongτ =0. 1 1 2 2 farfromeachothermaybebuiltwiththehelpofthesepropagating Next,beforewecomputeTr(ρT2 )ne,itisbeneficialto entangledpairs. seehowtocalculateTr(ρ )nA1fi∪rAst2. Inordertocalculate Tr(ρ )n,weconsiderAn1∪cAo2piesofthecutplane,andthen A1∪A2 ously,whichgeneratesquasiparticles(excitations)atthejoint- sewtogetherthecut[ui,vi]jτ=0− withthecut[ui,vi]jτ+=10+ for ing point. These quasiparticles may be viewed as entangled i=1,2andallthecopiesj =1,··· ,n. Notethatforj =n, pairs [28–32] which carry entanglement information. When wesewtogetherthecut[ui,vi]jτ==n0− withthecut[ui,vi]jτ==10+. the entangled pairs arrive at two intervals separately, the en- Inthisway,wedefinean-sheetedRiemannsurfaceRn. The tanglement negativity can be built immediately. Because the traceof(ρA1∪A2)nisthengivenby (1+1) dimensional critical system is Lorentz invariant at the Z lowenergylimit,wecanutilizethepowerofconformalfield Tr(ρ )n = Rn, (7) theoryandunderstandtheuniversalfeatureofthisdynamical A1∪A2 Zn phenomenon. where Z is the partition function for the orbifold CFT on The rest of the paper is organized as follows. In part B of Rn R . Ratherthandealingwiththefieldsonanontrivialmani- n this section, we give a brief review of the path integral rep- fold,itisfoundmoreconvenienttoworkonasinglecomplex resentationoftheentanglementnegativity,andthenintroduce plane. It turns out Eq. (7) can be expressed in terms of lo- theCFTsetupforalocalquantumquenchinpartC.InSection caltwistedfieldsdefinedat(u ,0)and(v ,0)onthecomplex i i II,byusingCFTapproach,wecomputethetimeevolutionof planeasfollows theentanglementnegativityfortwoadjacentintervalsinpart A,andtwodisjointintervalsinpartB.Weconsiderbothcases Tr(ρ )n =(cid:10)T (u )T¯ (v )T (u )T¯ (v )(cid:11). (8) wherethetwointervalsaresymmetricallyandasymmetrically A1∪A2 n 1 n 1 n 2 n 2 located. In section III, we describe the numerical method of Intuitively, the effect of twist fields T and T¯ is shown in n n calculatingtheentanglementnegativityforaharmonicchain, Fig. 2. Winding anticlockwise (clockwise) around the twist based on which we study the local quench of the entangle- fieldT (T¯ ),oncethebranchcutiscrossed,onewillgofrom n n mentnegativity. Thenwecomparethenumericalresultswith layerj tolayerj+1. theCFTresults. InsectionIV,weconcludeourworkandlist With the introduction of twist fields, the expression of someinterestingfutureproblemstobestudied. Tr(ρT2 )n is very straightforward. As discussed in Refs. A1∪A2 [21,22],theeffectofpartialtranspositionwithrespecttoA 2 isequivalenttochangingthetwotwistoperatorsT (u )and n 2 B. Entanglementnegativityinquantumfieldtheory T¯ (v ). Thenonehas n 2 Adetaileddescriptionofpathintegralrepresentationofthe Tr(cid:16)ρT2 (cid:17)n =(cid:10)T (u )T¯ (v )T¯ (u )T (v )(cid:11). (9) entanglement negativity can be found in Ref. [22]. For the A1∪A2 n 1 n 1 n 2 n 2 completenessofthispaper,wegiveabriefreviewhere. If the two intervals [u ,v ] and [u ,v ] are adjacent to each First,asdiscussedinRef.[22],byusingareplicatrick,one 1 1 2 2 other,wesimplysetu →v ,andthenEq.(9)canbewritten canrelatetheentanglementnegativitywiththeintegerpowers 2 1 as ofρT2 as A1∪A2 E = lim lnTr(cid:16)ρT2 (cid:17)ne, (5) Tr(cid:16)ρTA21∪A2(cid:17)n =(cid:10)Tn(u1)T¯n2(u2)Tn(v2)(cid:11). (10) A1,A2 ne→1 A1∪A2 Therefore,fromEqs.(5),(9)and(10),itisfoundthatthecom- wheren isaneveninteger,andthedensitymatrixρmaybe putationoftheentanglementnegativityreducestothecompu- e expressedasa(Euclidean)pathintegralintheimaginarytime tationofexpectationvaluesoftwistfieldsinacomplexplane. 3 FIG.2:Pathintegralrepresentationof(a)Tr(ρ )nand(b)Tr(ρT2)nfortwodisjointintervals. A A C. CFTapproachtoalocalquench ofthepathintegralonamodifiedword-sheet,wherethephys- ical cut corresponds to two ‘walls’ with one extending from Before we study the CFT approach to a local quantum τ = −∞to−(cid:15)andtheotherextendingfromτ = +(cid:15)to+∞ quench,itisbeneficialtocommentonthedifferencebetween in a complex z-plane. No energy nor momentum can flow localquenchesandglobalquenches.Localquenchesaremore through the two ‘walls’, and therefore conformal boundary complicated than global quenches because they are inhomo- conditions are imposed on the wall (As will be shown later, geneous. Forglobalquenches,wechangetheparametersofa theconcreteboundaryconditiondoesnotaffecttheuniversal translationalinvariantHamiltonianglobally,andthereforethe resultweconsider.). Forconvenienceofcalculation,wemap system before and after global quenches are always transla- the z-plane to a right half plane (RHP) with Rew > 0 by tionalinvariant. Inthiscase,asdiscussedinRefs.[31,32],the usingtheconformalmapping initial state can flow to a conformal invariant boundary state underrenormalizationgroup(RG).Forlocalquenches,how- (cid:114) ever,beforewejointhetwodecoupledCFTstogether,theto- z (cid:16)z(cid:17)2 talsystemisapparentlynottranslationalinvariant. Therefore, w = + +1. (11) (cid:15) (cid:15) theinitialstatecannotflowtoaconformalinvariantboundary state. In addition, for global quenches, quasiparticle excita- tionsareemittedfromeverywhereinthebulkofCFT;forlo- Then the local quench problem is reduced to the calculation calquenches, quasiparticleexcitationsareemittedonlyfrom ofcorrelationfunctionsoftwistfieldsintheRHP[33],which thepointwheretwoCFTarejoinedtogether. wewillstudyindetailinthenextsection. Thetimedependentdensitymatrixcanbewrittenasρ(t)= |φ(x,t)(cid:105)(cid:104)φ(x,t)|, where |φ(x,t)(cid:105) = e−iHt|φ (x)(cid:105). In path 0 integralrepresentation,onehas (cid:104)φ(cid:48)(cid:48)(x(cid:48)(cid:48))|ρ(t)|φ(cid:48)(x(cid:48))(cid:105) 1 = (cid:104)φ(cid:48)(cid:48)(x(cid:48)(cid:48))|e−iHt−(cid:15)H|φ (x)(cid:105)(cid:104)φ (x)|e+iHt−(cid:15)H|φ(cid:48)(x(cid:48))(cid:105), Z 0 0 wherethefactore−(cid:15)H isintroducedtodampouthigh-energy modes and make the path integral absolutely convergent. If theCFTarisesasthelowenergylimitofalatticemodel,then (cid:15)maybeviewedasthelatticespacing. Inthestudyofglobal quenches[31–33],|φ (x)(cid:105)maybeconsideredasaconformal 0 invariant boundary state under the RG. For local quenches, as we discussed above, the initial state cannot flow to a con- formal invariant boundary state under the RG. In this case, FIG.3: IllustrationoftheconformalmappinginEq.(11),basedon onemayintroduceboundaryconditionchangingoperators,as whichthez-planeismappedtoarighthalfplane(RHP)withRew> utilizedinRefs.[34,35]. Inthiswork,however,wewillfol- 0.Forlateruse,wealsolabelw(cid:101)i =−w¯i,whichistheimageofwi. low the method proposed by Calabrese and Cardy [33]. As showninFig.3,thedensitymatrixcanbeexpressedinterms 4 Withananalyticalcontinuationτ →it,onecanobtain (cid:10)T2(z )(cid:11)=c˜ (cid:18) a(cid:15) (cid:19)∆(n2). (16) n 1 n 2((cid:15)2+t2) As discussed in Refs. [21, 22], the scaling dimension ∆(2) n dependsontheparityofnas (cid:26)∆ oddn, ∆(2) = n , (17) n 2∆ evenn, n/2 where FIG.4:Configurationsoftwointervalsconsideredinthiswork:two adjacentintervals(up)andtwodisjointintervals(bottom). c (cid:18) 1(cid:19) ∆ = n− . (18) n 12 n II. ENTANGLEMENTNEGATIVITYAFTERALOCAL ThenbyusingtheexpressionsinEqs. (5)and(9),onecanget QUENCH:CONFORMALFIELDTHEORYAPPROACH c a(cid:15) E =− ln +c˜(cid:48), (19) In this section, we calculate the time evolution of the en- 4 2((cid:15)2+t2) 1 tanglement negativity after a local quench in conformal field wherec˜(cid:48) =lnc˜ . AsinRefs.[3,33],theshorttimebehavior theories. We will consider adjacent intervals in part A and 1 1 of E(t) allows us to fix the regulator (cid:15) in terms of the non- disjointintervalsinpartB,respectively. universalconstantc˜ byrequiring 1 c a E(t=0)=− ln +c˜(cid:48) =0, (20) A. Twoadjacentintervals 4 2(cid:15) 1 basedonwhichonegets 1. Semi-infiniteintervals a (cid:15)= e−4c˜(cid:48)1/c. (21) Asawarmup,weconsiderthesimplestcase,i.e.,thetotal 2 systemisbipartitionedintotwosemi-infinitepartsA andA . 1 2 Nowitispossibletoeliminateaandc˜(cid:48) inEq.(19)intermsof Inthiscase,ρ ispure,andthelogarithmicnegativityis 1 A1∪A2 (cid:15),andthenonecanget the same as the Renyi entropy with n = 1/2 [21, 22]. This casewasalsostudiedinRef.[27]. c (cid:15)2 For two adjacent semi-infinite intervals, we only need to E =− ln . (22) 4 (cid:15)2+t2 consider a single twist field T2(z ) in z-plane, which is in- n 1 sertedat Inthelimitt(cid:29)(cid:15),oneendswith c t z =l+iτ. (12) 1 E = ln , (23) 2 (cid:15) By choosing the insertion position at the origin l = 0, i.e., whichwasobservedinthenumericalcalculationsbasedona A ∈ (−∞,0]andA ∈ (0,+∞), onesimplyhasz = iτ. 1 2 1 criticalharmonicchain[27]. TheexpectationvalueofT2(z )canbeexpressedas[33] n 1 (cid:10)Tn2(z1)(cid:11)=c˜n(cid:32)(cid:12)(cid:12)(cid:12)(cid:12)ddwz(cid:12)(cid:12)(cid:12)(cid:12) 2Rea(w )(cid:33)∆(n2), (13) 2. Symmetricfiniteintervals z1 1 Inthispart,weconsiderthecaseofsymmetricfiniteinter- wherec˜ isanonuniversalconstantwhichdependsonthepar- vals with l = l = l, i.e., A ∈ [−l,0] and A ∈ (0,l], as n 1 2 1 2 ticularboundaryCFT,aisanUVcutoff(e.g.,thelatticespac- showninFig.4.Inthiscase,ρ representsamixedstate, A1∪A2 inginaharmonicchain),and∆(2)isthescalingdimensionof andthereisnocorrespondencebetweenthelogarithmicneg- n T2(z). ByusingtheconformalmapinEq.(11),onehas ativityandtheRenyientropies. ByusingEq.(10)anddoing n aconformalmappingontotheRHP,onehas τ 1(cid:112) w1 =i(cid:15) + (cid:15) (cid:15)2−τ2, (14) Tr(cid:16)ρT2 (cid:17)n =(cid:10)T (z )T¯2(z )T (z )(cid:11) A1∪A2 n 1 n 2 n 3 and (cid:12)(cid:12)(cid:12)dw(cid:12)(cid:12)(cid:12) = √ 1 . (15) =i(cid:89)=31(cid:12)(cid:12)(cid:12)(cid:12)ddwz(cid:12)(cid:12)(cid:12)(cid:12)∆zi(i)(cid:10)Tn(w1)T¯n2(w2)Tn(w3)(cid:11)RHP, (cid:12)dz(cid:12) (cid:15)2−τ2 (24) z1 5 3.0 3.0 (a) l=25 (b) l=25 l=50 l=50 2.5 l=75 2.5 l=75 l=100 l=100 y 2.0 2.0 vit gati 1.5 1.5 e N 1.0 1.0 0.5 0.5 0.0 0.0 0 20 40 60 80 100 0 20 40 60 80 100 Time Time FIG. 5: Entanglement negativity E for two symmetric adjacent intervals as a function of time. Here we choose the central charge c = 1, (cid:15)=0.1,l=25,50,75and100,respectively.Shownin(a)istheCFTresult,and(b)isthenumericalcalculationbasedonacriticalharmonic chain. wherethescalingdimensions∆ =∆ =∆ and∆ = whichmaybefurtherexpressedas (1) (3) n (2) ∆(n2). Thethree-pointcorrelationfunctionontheRHPcanbe c (cid:15)2 c w w w expressedas E =− ln − ln 13 1˜2 2˜3, (28) 4 (cid:15)2+t2 4 w w w 1˜3 12 23 (cid:10)T (w )T¯2(w )T (w )(cid:11) wherewehavedefinedw =|w −w |,w =|w −w |and n 1 n 2 n 3 RHP ij i j i˜j i (cid:101)j c˜ η∆(n2)−2∆n1/2 w˜i˜jW=ith|w(cid:101)thie−exw(cid:101)pjr|e,srseisopnesctoifvewly. that are calculated in the Ap- =(cid:81)3i=1|(wi−nw(cid:101)i)/a|∆(i) η11∆,,23(n2)η2∆,3(n2) F({ηj,k}), pfuenncdtiixo,noonfetimcaenaosbfotalilnowthse iejntanglement negativity E as a (25)  c (cid:15)2 c l+t whereη arecrossratioswhichcanbeconstructedfromthe  − 4ln(cid:15)2+t2 − 4lnl−t, t<l i,j E = (29) eansdfoplolionwtsswi(andtheirimagesw(cid:101)i)oftheintervalsintheRHP  − 4c ln(cid:15)2(cid:15)+2t2 − 4c ln4(cid:15)tl2, t>l. (w −w )(w −w ) Inthelimitl,t(cid:29)(cid:15),E canbesimplifiedas η = i j (cid:101)i (cid:101)j (26) i,j (wi−w(cid:101)j)(w(cid:101)i−wj)  c t2+(cid:15)2(l−t) 4ln (cid:15)2 (l+t), t<l with w = −w¯ being the image of w (see Fig.3). The E = (30) nonun(cid:101)iviersalfuncitionF({ηj,k})dependsionthefulloperator  c ln l +const, t>l. 4 2(cid:15) contentoftheCFT.F({η })isusuallydifficulttocalculate j,k and only known for several specific CFTs and BCFTs. But ShowninFig.5(a)istheplotofE withdifferentl. Atthevery it is found that in the limits ηi,j → 0, 1, or ∞, the function beginning of the local quench t (cid:28) l, based on Eq. (30), one F({ηj,k})isjustaconstant,whichfollowsfromthelong-and has short-distanceexpansionsofthecorrelationfunctionsoftwist c t operators[3,32,36–38]. Forsymmetricintervalsinthispart, E = ln , (31) 2 (cid:15) wecalculatethecrossratiosη explicitlyintheappendix. It i,j isfoundthatonealwayshasη =1or0forthecasest(cid:28)l, which agrees with the result of semi-infinite intervals as ij t=l+0−andt>l. Inotherwords,ourresultsareuniversal shown in Eq. (23). This is reasonable because in the limit fortheabovethreecases. t (cid:28) l,thequasiparticlesessentiallypropagatewithoutnotic- ByusingEqs. (5)and(10),andneglectingvariousnonuni- ingthefinitesizeeffect. Fort<l,theentanglementsaturates versalterms,wehave for a certain time. Then for t > l, we get the ground-state valueofE,i.e., E =− c ln (cid:15)2 − c ln η1,3 , (27) E = c ln l , (32) 4 (cid:15)2+t2 8 η1,2η2,3 G 4 2(cid:15) 6 which is also observed in the numerical calculations in Ref. Ontheotherhand,fromthescalingbehaviorinRef.[27],one [27]. Note that in the numerical calculations, E tends to the canfind ground-state value gradually. In our CFT results, E drops to the ground-state value immediately after t = l. This is l 1 l E (t= )= ln +0.13ln3−0.28ln2. (35) because all quasiparticles propagate at the same velocity in numerical 2 2 2 CFTs. In lattice models, however, the dispersion relation is not linear for all momentum vectors, and therefore not all Inthelargellimit,i.e.,l(cid:29)(cid:15),onealwayshas quasiparticlespropagateatthesamevelocity,asdiscussedin detailinsectionIV. l l 1 l Inaddition,thescalingbehaviorofE foraharmonicchain E (t= )(cid:39)E (t= )(cid:39) ln . (36) wasnumericallystudiedinRef.[27].Fort<l,theyproposed CFT 2 numerical 2 2 2 theansatz Beforeweendthispart,wementionthatitisinterestingto tα (l−t)β E ∼ln . (33) checkhowE(t)behavesinotherlatticemodels. Considering l−ρ(l+t)−γ E(t < l) depends on the non-universal functions F({η }), j,k whichvariesfordifferentCFTs,weexpectthatforothercrit- By fitting the numerical results, they found α = 1/2, β (cid:39) ical lattice models such as the critical Ising model one may 0.15, γ (cid:39) 0.13 and ρ = −(β +γ) (cid:39) −0.28. For our CFT observedifferentscalingbehaviorsinE(t<l). resultsinEq.(30),bysettingc=1andtakingthelimitt(cid:29)(cid:15), one has α = 1/2, β = 0.25, γ = −0.25 and ρ = −(β + γ) = 0. On the other hand, in the limit t (cid:28) l, both Eq. (30) and Eq. (33) collapse to E = 1lnt. We attribute the 2 abovedisagreement/agreementtothefollowingfact. Fort ≤ 3. Asymmetricfiniteintervals l, because we neglect the non-universal functions F({η }) j,k whichmaynotbeconstants, ourresultsarenotaccurateand Inthispart,weconsiderthecaseofasymmetricfiniteinter- thereforemaynotobtainthecorrectscalingbehavior.Fort(cid:28) valswithA ∈ [−l ,0]andA ∈ (0,l ],asshowninFig.4. l, however, ourCFTresultsareuniversalandindependentof 1 1 2 2 Withoutlossofgenerality,wesupposel < l . Thecalcula- thespecificCFT.Toreproducethenumericalresultsfort≤l 1 2 tionsaresimilartothesymmetriccase, andweneedtoeval- in Ref. [27], we have to consider the nonuniversal functions uatethethreepointcorrelationfunctionsinEq.(24). First,as F({η }), which is a difficult task, and out of the scope of j,k shownintheappendix,wecalculatethecrossratioη explic- ourwork. ij itly. Itisfoundthatonealwayshasη = 1or0forthecases Nevertheless, by comparing the values of the plateau be- ij t (cid:28) l , t = l +0− and t > l . That is to say, our results tweenCFTresultsandnumericalresultsinFig.5,itisfound 1 2 2 areuniversalintheseregions. Second,byneglectingvarious they are very close to each other. To be concrete, let’s take nonuniversalterms,wearriveatthesameresultasinEq.(28). E(t= l)forexample. BasedonEq.(30),onecanget 2 The difference is that for asymmetric intervals, we have dif- ferentexpressionsofw ,asexplicitlygivenintheappendix. l 1 l ij E (t= )= ln +const. (34) Bypluggingw intoEq.(28),oneobtains CFT 2 2 2(cid:15) ij  c (cid:15)2 c (l +t)(l +t)  − 4c ln(cid:15)2(cid:15)+2t2 − 8c ln(cid:32)(l112−(cid:115)t()l(l22+−lt))(,t+l )(t−l )t2(cid:33) t<l1 E = − ln − ln 1 2 1 1 , l <t<l  − 44c ln(cid:15)(cid:15)22(cid:15)++2tt22 − 44c ln2((cid:15)l1(cid:15)l+ll2)t2, (l2−l1)l12 1 t>l22. 1 2 Inthelimitl,t(cid:29)(cid:15),E canbesimplifiedas Notethatinthelimitl =l ,wecanreproducethesymmetric 1 2 adjacentintervalsresultinEq.(30).TheplotofE fordifferent 8cc ln((ll11+−(ttl))((ll−22+−l )ttl))2(cid:15)tt442, t<l1 (cclaa1ns,elfi.2n)Adicstthusehaltoliwym,nebeianvseoFdliugot.ino6nE(aoq)f..E(F3(i7tr))s,ti,sofnsoiemrctialan(cid:28)rctohmetchinke[lst1hy,amlt2mi]n,eottrhniece E = ln 2 1 1 , l <t<l (37) 84c ln((cid:15)l(1ll+1+l2l2l)()t++clo1n)(stt,−l1) 1 t>l22. 1 2 7 3.0 3.0 (a) l1=25, l2=75 (b) l1=25, l2=75 2.5 l1=50, l2=75 2.5 l1=50, l2=75 l1=75, l2=75 l1=75, l2=75 y 2.0 2.0 vit gati 1.5 1.5 e N 1.0 1.0 0.5 0.5 0.0 0.0 0 20 40 60 80 100 0 20 40 60 80 100 Time Time FIG.6:EntanglementnegativityEfortwoasymmetricadjacentintervalsasafunctionoftime.Herewechoosecentralchargec=1,(cid:15)=0.1, (l ,l )=(25,75),(50,75),and(75,75),respectively.Shownin(a)istheCFTresult,and(b)isthenumericalcalculationbasedonacritical 1 2 harmonicchain. limitt(cid:28)min[l ,l ] 4(b). In this case, we need to consider the correlation func- 1 2 tionoffourtwistfieldsasshowninEq.(9). Byapplyingthe c t E = ln , (38) conformalmapinEq.(11),onehas 2 (cid:15) which shows the lnt behavior again, as expected. For t > (cid:10)T (z )T¯ (z )T¯ (z )T (z )(cid:11) n 1 n 2 n 3 n 4 max[l ,l ],weobtainthegroundstatevalueoftheentangle- mentn1eg2ativity[21,22],i.e., =(cid:89)4 (cid:12)(cid:12)(cid:12)(cid:12)ddwz(cid:12)(cid:12)(cid:12)(cid:12)∆n(cid:10)Tn(w1)T¯n(w2)T¯n(w3)Tn(w4)(cid:11)RHP, E = c ln l1l2 . (39) i=1 zi G 4 (cid:15)(l1+l2) wherethefour-pointcorrelationfunctionontheRHPhasthe Interestingly,itisfoundthatthesuddendropofE happensat form t=min[l1,l2], (40) (cid:10)Tn(w1)T¯n(w2)T¯n(w3)Tn(w4)(cid:11)RHP which is again straightforward to understand based on the c˜2 1 (cid:18)η η (cid:19)∆(n2)/2−∆n = n 1,4 2,3 hveieuwriesdticaspheynstiacnagllepmicetnutrepathirast otfhetwqouaqsiupaanrttaic.lesFomrayt b<e (cid:81)4i=1|(wi−w(cid:101)i)/a|∆n η1∆,2nη3∆,4n η1,3η2,4 min[l1,l2],twoentangledquasiparticlesareinA1andA2sep- ×F({ηj,k}). arately, andcreatetheentanglementbetweenA andA . At (41) 1 2 t = min[l ,l ] = l (herewesupposel < l ),althoughone 1 2 1 1 2 For the nonuniversal functions F({η }), as explicitly cal- quasiparticleisstillinA , theotherquasiparticlepropagates j,k 2 culated in Ref. [36], they are simply a constant in the limit outofA ,andthereforetheentanglementbetweenA andA 1 1 2 l/d(cid:28)1. Inotherwords,whenthetwointervalsarefarapart, decreasessuddenlyatt=min[l ,l ]. 1 2 we do not need the knowledge of F({η }). By using the Beforeweendthispart, weemphasizethatourresultsare j,k definitioninEq.(5),anddroppingvariousmultiplicativecon- universalfortheregimest(cid:28)l ,t=l +0− andt>l . For 1 2 2 stants,wehave thecaseoft ≤ l ,however,similarwiththesymmetriccase, 2 because we neglected the nonuniversal functions F({ηj,k}) E =−c ln(cid:18)η1,4η2,3(cid:19), (42) whichmaynotbeconstants,ourresultsarenotaccurate,and 8 η η 1,3 2,4 onehastocalculateF({η })fordifferentCFTs. j,k whichisalternativelywrittenas c w w w w w w w w B. Twodisjointintervals E =− ln 14 1˜3 23 2˜4 ˜1˜4 ˜13 ˜2˜3 ˜24. (43) 8 w w w w w w w w 1˜4 13 2˜3 24 ˜14 ˜1˜3 ˜23 ˜2˜4 1. Symmetricfiniteintervals By noting that wij = w˜i˜j and wi˜j = w˜ij, Eq. (43) can be simplifiedas In this part, we consider the symmetric disjoint intervals, E =− c lnw14w1˜3w23w2˜4. (44) i.e., A1 ∈ [−d−l,−d] and A2 ∈ [d,d+l], as shown Fig. 4 w1˜4w13w2˜3w24 8 (a) (b) 0.4 0.06 d=140, l=10 d=140, l=10 d=160, l=10 d=160, l=10 d=180, l=10 y 0.3 d=180, l=10 gativit 0.04 e 0.2 N 0.02 0.1 0.0 0.00 100 120 140 160 180 200 220 100 120 140 160 180 200 220 Time Time FIG.7:EntanglementnegativityEfortwosymmetricdisjointintervalsasafunctionoftime.Herewechoosethecentralchargec=1,(cid:15)=1, (d,l) = (140,10), (160,10) and (180,10), respectively. Shown in (a) is the CFT result, and (b) is the numerical calculation based on a criticalharmonicchain. Withtheexplicitformsofw givenintheAppendix,wecan simultaneously. Note that at t = d + l/2, taking the limit ij obtaintheentanglementnegativityE asafunctionoftimeas d(cid:29)l,onehas follows c l 0c ln(2d+l)(d+l−t)(t2−d2) d<t<td<+dl which is independenEtto=fd+th2le(cid:39)dis4talnnce2(cid:15)d,, as also can be(4o6b)- E = 4 (cid:15)dl(d+l+t) served in Fig. 7. That is to say, with the help of entangled  c ln (2d+l)2 t>d+l pairs, we can create a long-range entanglement between two 4 4d(d+l) intervals which are far from each other. Note that this long- (45) range entanglement was also observed in the time evolution Notethatinthestudyofthenegativityevolutionafteraglobal ofmutualinformationI(t)inRef.[36],whereitisfoundthat quench, it was found that E(t) shows the same behavior as I (cid:39) c ln l . t=d+l 3 2(cid:15) the Renyi mutual information apart from the prefactor [26]. 2 Forthelocalquenchstudiedhere,bycomparingourresultin Eq.(45)withtheresultofmutualinformationinRef.[36],it 2. Asymmetricfiniteintervals isfoundthattheexpressionsarealsothesameexceptforthe prefactor. Inotherwords,ourresultsparallelwiththestoryin thenegativityevolutionafteraglobalquench.Therelationbe- Inthispart, weconsidertheasymmetricdisjointintervals. tweentheentanglementnegativityandthemutualinformation Wehavemultiplechoicesasfollows: (i)d1 (cid:54)=d2,l1 =l2,(ii) afteralocalquantumquenchwillbesystematicallydiscussed d1 = d2, l1 (cid:54)= l2 and (iii) d1 (cid:54)= d2, l1 (cid:54)= l2. For simplicity, insectionIV. weconsiderthecase(i),i.e.,A1 ∈ [−d1+l,−d1]andA2 ∈ As shown in Fig. 7(a), we plot the evolution of the entan- [d2,d2+l]. Withoutlossofgenerality,wechoosed1 <d2 ≤ glementnegativitywithdifferent(d,l)accordingtoEq.(45). d1+l. A ‘light-cone’ effect can be observed: For t < d, there is The calculation of the negativity evolution is similar with no entanglement negativity between A and A . At t = d, thesymmetriccase,andweobtainthesameresultinEq.(44). 1 2 theentanglementnegativitybeginstodevelop,andreachesthe The difference is that we should express w in terms of d , ij 1 maximumapproximatelyatt=d+l/2. Att=d+l,theen- d and l, as explicitly shown in the appendix. By plugging 2 tanglementnegativitydecreasessuddenly,whichcorresponds the expressions of w into Eq. (44), one arrives at the time ij totheentangledpairspropagatingoutofintervalsA andA evolutionoftheentanglementnegativity 1 2 9 0.6 (a) (b) d1=150, d2=150, l=15 d1=150, d2=150, l=15 0.10 0.4 d1=150, d2=155, l=15 d1=150, d2=155, l=15 y vit d1=150, d2=160, l=15 d1=150, d2=160, l=15 gati d1=150, d2=165, l=15 d1=150, d2=165, l=15 e N 0.05 0.2 0.0 0.00 100 120 140 160 180 100 120 140 160 180 Time Time FIG.8:EntanglementnegativityEfortwoasymmetricdisjointintervalsasafunctionoftime.Herewechoosethecentralchargec=1,(cid:15)=1, l=15,(d ,d )=(150,150),(150,155),(150,160),and(150,165),respectively. Shownin(a)istheCFTresult,and(b)isthenumerical 1 2 calculationbasedonacriticalharmonicchain. 0 t<d  −− 8cc llnn((dd(cid:15)11(++d1dd+22)+d(2dl)2)(−d(cid:115)2t)+((ddl12−++tll)−(+d2dt)2+()d(t2d)2(−d+2d1−l−+d1dl))1)(d1+l+t)(d2+l+t) d <d1t<<td<+d21l E = 4 2(d +d +l) (d +l−t)(d +l−t)(t2−d2)(t2−d2) 2 1 (47) 1 2 1 2 1 2  −− 84cc llnn[(td21−+((ddd21+++(dd2l2)l)2+−(](ddld1)112+)+(ddd212)+)2(dd22++l)l3−(td21−)(dd211)+d2+2l) d1+l<tt<>dd22++ll 1 2 Onecancheckthatwhend = d = d,theresultinEq.(45) entanglement negativity for a harmonic chain has been nu- 1 2 isreproduced. merically studied in several works [9, 22, 26, 27, 39]. Here AccordingtoEq.(47),weplotE(t)withdifferent(d ,d ) wefollowthemethoddevelopedintheseworks,andapplyit 1 2 inFig.8(a).Comparedtothesymmetriccase,the‘light-cone’ to the local quench problem. We will first introduce the lat- effect is still observed. The difference is that the time when ticemodelandthecovariancematrixinpartA.InpartB,we E(t)increasesquicklynowhappensat introducetheevolutionmatrixandshowhowtocalculatethe entanglement negativity. In part C, we apply the method to t=max[d ,d ], (48) 1 2 thecasesstudiedwithCFTapproach,andcomparetheresults andthetimewhenE(t)decreasesquicklyhappensat accordingly. t=min[d +l,d +l], (49) 1 2 whichisalsoinagreementwiththequasiparticlepicture. A. Harmonicchainandthecovariancematrix TheHamiltonianoftheharmonicchainis III. NUMERICALEVALUATIONOFTHENEGATIVITY FORAHARMONICCHAINAFTERALOCALQUENCH (cid:88)N p2 Mω2 K H = [ n + 0q2 + (q −q )2], (50) 2M 2 n 2 n+1 n In this section, to confirm our CFT results, we study the n=1 timeevolutionofthelogarithmicnegativityafteralocalquan- where N is the number of sites of the chain, M is the tumquenchinalatticemodel,acriticalharmonicchain. The mass scale, ω is the characteristic frequency, and K is the 0 10 nearest-neighborcoupling. p andq denotethemomentum ForDBC,thecorrelatorsare n n and position operators with canonical commutation relations L−1 (cid:18) (cid:19) (cid:18) (cid:19) [p ,p ]=[q ,q ]=0and[q ,p ]=iδ . 1 (cid:88) 1 πkn πkm n m n m n m n,m (cid:104)0|q q |0(cid:105)= sin sin , For periodic boundary condition (PBC), the Fourier trans- n m L Mω L L k k=1 formofthecanonicalvariablesare L−1 (cid:18) (cid:19) (cid:18) (cid:19)  1 (cid:88) πkn πkm  qn =(cid:88)L q(cid:101)k√1Le2πikn/L, (cid:104)0|pnpm|0(cid:105)= L k=1Mωksin L sin L , k=1 (cid:104)0|q p |0(cid:105)=iδ /2. (58) (51) n m n,m  q(cid:101)k =(cid:88)L qn√1Le−2πikn/L, n=1 B. Evolutionmatrixandthelogarithmicnegativity wheren = 1,··· ,L. Forp ,theFouriertransformisidenti- n FromtheHeisenbergequationofmotion,q˙ (t) = 1 p (t) caltoqn. TheHamiltonianisdiagonalizedinthemomentum (cid:101)k M(cid:101)k space andp(cid:101)˙k(t)=−Mωk2q(cid:101)k(t),wehave  1 H =(cid:88)L (cid:18) 1 p2 + Mωk2q2(cid:19), (52) q(cid:101)k(t)= √M(cosωktq(cid:101)k(0)+ωk−1sinωktp(cid:101)k(0)), 2M(cid:101)k 2 (cid:101)k √ k=1 p (t)= M(−ω sinω tq (0)+cosω tp (0)). (cid:101)k k k (cid:101)k k (cid:101)k where Thetime-dependentcanonicalvariablesintherealspaceare (cid:114) 4K πk  1 (cid:88) ωk = ω02+ M sin( L )2, k =1,··· ,L, (PBC()5.3) qn(t)=√×M(cid:0)coks,mωφt∗kq(n()0φ)k+(mω)−1sinω tp (0)(cid:1) k m k k m For the Dirichlet boundary condition (DBC), the Fourier √ (cid:88) tmraentrsyf.orHmowisenvoetr,vtahliedFdouueriteortshienberteraaknisnfogromftcraannsblaetdioenfianlesdymas- pn(t)= M k,mφ∗k(n)φk(m) ×(−ω sinω tq (0)+cosω tp (0)),  L−1 (cid:114) (cid:18) (cid:19) k k m k m qn =(cid:88)q(cid:101)k L2 sin πLkn , where k=1 (54) 1 q(cid:101)k =L(cid:88)−1qn(cid:114)L2 sin(cid:18)πLkn(cid:19), φn(k)= √2Le−2πikn/L, (PBC), n=1 φ (k)= √ sin(πkn/L), (DBC). (59) n L wheren = 1,··· ,L. Forp ,theFouriersinetransformation n isdefinedsimilarly. TheHamiltonianinthemomentumspace Therefore,thetimeevolutionofthecovariancematrixis isidenticaltoEq. (52). Butthefrequencyω hasadifferent k γ(t)=S(t)γ(0)S(t)T, (60) form, wheretheevolutionmatrixis (cid:114) 4K πk ωk = ω02+ M sin(2L)2, k =1,··· ,L−1, (DBC). Sn,m(t)=(cid:88)φ∗k(n)φk(m) (55) k (cid:32) (cid:33) The covariance matrix is constructed from the two-point × √√1M cosωkt √1√Mωk−1sinωkt . − Mω sinω t Mcosω t correlators k k k (61) (cid:18) (cid:19) (cid:104)0|q q |0(cid:105) (cid:104)0|q p |0(cid:105) γ =Re n m n m . (56) n,m (cid:104)0|p q |0(cid:105) (cid:104)0|p p |0(cid:105) The entanglement properties are encoded in the reduced n m n m densitymatrix,whichcanbeextractedfromthethecovariance ForPBC,thecorrelatorsare matrixγ associatedwiththethesubsystemA. Thelogarith- A mic negativity is defined by the partial transposition of the L (cid:20) (cid:21) 1 (cid:88) 1 2πk(n−m) reduceddensitymatrixρ withthesubsystemA = A ∪A (cid:104)0|q q |0(cid:105)= cos , A 1 2 n m 2L Mω L as E = lnTr|ρT2|. We first consider the partial transposition k=1 k A ofγ ,whichcanbeconstructedbyinvertingthesignsofthe A L (cid:20) (cid:21) 1 (cid:88) 2πk(n−m) momentacorrespondingtoA [9]. (cid:104)0|p p |0(cid:105)= Mω cos , 2 n m 2Lk=1 k L γT2 =(cid:18) IlA 0lA (cid:19)·γ ·(cid:18) IlA 0lA (cid:19), (62) (cid:104)0|q p |0(cid:105)=iδ /2. (57) A 0 R A 0 R n m n,m lA A2 lA A2

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.