Comment on“Witnessedentanglement andthe geometricmeasure ofquantum discord” Swapan Rana∗ and Preeti Parashar† PhysicsandAppliedMathematicsUnit, IndianStatistical Institute, 203BTRoad, Kolkata, India (Dated:January10,2013) Inarecentarticle[Phys.Rev.A86,024302(2012)],theauthorshavederivedsomehierarchyrelationsbe- tween geometric discord and entanglement (measured by negativity and itssquare). Wepoint out that these resultsareincorrectandgiveanalyticcounterexamples. Wealsodiscussbrieflythereasonforsuchviolations. PACSnumbers:03.67.Mn,03.65.Ud WestartwiththedefinitionofGDandnegativityfromRef. m/(m−1)>1andhencethiscasealsoincludesthecasewhen 3 [1]. Foranm⊗n(m≤n)stateρ,theyhavedefined D isnotnormalizedandEq. (2a)istakenasthedefinition (2) 1 ofnegativity. Nowwewillgiveananalyticexampletoshow 20 D(p)(ρ)=mξ∈iΩnkρ−ξk(pp) (1) thattherearestatesviolatingeventheweakerrelation[6] n where Ω is the set of zero-discord (or classical-quantum) 2 9 Ja sptantoersmgigvievnenbbyyξkX=k(Pp) p=i|{iiThri[|X⊗†ρXi]pa/n2d}1/kpX.kT(ph)einsetghaetiSvcithyawtteans mm−1D(2) ≥ (mN−21)2 = "λPi<(0mλ−i(ρ1T)A2)# (4) definedby ] Considerthem⊗mWernerstategivenby h p N(ρ)=kρTAk(1)−1 (2a) m−z mz−1 t- =max{0,− min Tr(Wρ)} (2b) ρw = m3−mI+ m3−mF, z∈[−1,1] (5) n 0≤WTA≤I a whereF = |kihl|⊗|lihk|andsetm = 8,z = −1. Now,ifwe u where ρTA is the partial transposition of ρ with respect to A q andW isanyoptimalentangledwitness. Attemptingtoprove considerthePmatrixformofρw(incomputationalbasis)asthe [ stateofa2⊗32system,thelefthandsideofEq. (4)becomes aconjecturemadein[2], theyhaveclaimed(Eq. 17therein) 1/49whiletherighthandside becomes25/784. Thoughwe 1 thatallbipartitestatessatisfy haveusedtheformulaforD developedinRef.[7](whichis v (2) 1 N2 exactfor2⊗nstates),ameasurementincomputationalbasis 6 D(2) ≥ (m−1)2 (3) willyieldtheresult. Anyvalueofz ∈ [−1,−34/43)willalso 9 workwell. We note thatforlargeenough n, the Refs. [3, 8] 1 Wefirstobserveatypothatthetwodefinitionsofnegativity giveenoughintuitionforviolationofthisrelation. Nonethe- 1. inEq. (2)arenotequal,asthequantityinEq. (2b)isthesum less,theanalyticcounterexamplemakesitmoreexplicit. 0 ofabsolutevaluesofnegativeeigenvaluesofρTA whereasthat The authors of Ref. [1] also proposed to take D(1) as a 3 inEq. (2a)isjustdoubleofit. propermeasure of geometric discord and derived the hierar- 1 As a first gapin theirderivation(which could be taken as chyrelation(Eq. 27therein) : v anothertypo,though),wenotethattheyhavenotnormalized Xi D(2) to have maximum value unity, as has been done in the D(1) ≥ N (6) originalpaper[2]. As a result, if we take Eq. (2a) as defini- ar tionofnegativity,Eq. (3)isnotnecessarilysatisfiedevenby However,ascanbeseeneasily,thisresultisalsonotcorrect. thetwo-qubitmaximallyentangledstate(anyoneoftheBell Itiswellknownthatthetracedistancesatisfieskρ−σk(1) ≤2, states has D(2) = 1/2 whereas N = 1). The importance of forallρandσ[9]. HencewemusthaveD(1) ≤ 2whereasN normalizationcouldbefoundinRefs. [3,4]. Nowwewilltry cantakevalueupto(m−1)/2(forexampleconsidertheBell toremoveallthese(possible)typosandshowthattherelation state in m⊗m). Taking the identity matrix as the classical- (3), whether normalized or not, is always violated by some quantumstate (neednotbeoptimal),weseethattherelation states. isviolatedby4⊗4Bellstates. Letusfirstconsiderthecasewhen D isnormalized(tak- As has been pointed out in Ref. [3], the violation for (2) ing D = mm−1D(2)) and Eq. (2a) is taken as the defini- D(2) stemsfromthefactthattheHilbert-Schmidtnormisnot tion of negativity. Then it becomes the original conjecture monotone—D(2)couldbeincreasedordecreasedbyaddingor (D ≥ N2/(m − 1)2) made in Ref. [2], which we have re- removingafactorizedlocalancilla.Wewouldliketomention futed recently in Ref. [5]. Note that the normalizing factor thatthetracenormbeingmonotone,doesnotsufferfromthis problem.ThustheproposaloftakingD asagoodmeasure, (1) is interestingand mightbe worth investigating. However,as we pointedouthere, establishinganyinterrelationshould be ∗[email protected] donecarefully.Anotherpointofconcernregardingtheuseof †[email protected] the trace norm is thatits analytic calculationis verydifficult 2 and hence the main spirit of usual geometricdiscord will be relation[D > N2/(d−1)]willthenbecorrectforNPTstates (2) lost. (thePPTstatestriviallysatisfyequality),animportantpointis Note added: After submission of this comment to jour- that it can no longer be used to compare geometric discord nal,wefoundanerratumpostedonarXive[10]inwhichthe andentanglement. Wenotethat,bothN andN2 areentangle- authors have tried to fix the errors, based on our criticism. mentmonotones,butN/d(ingeneralN/f(d))isnotanentan- Thoughtheerratumisoutofpurviewofthepresentcomment, glementmonotone,asitmightbeincreasedwithremoval(or wewouldliketopointoutthatsomefurthermodificationsare addition)oflocalancillarysystems. Thisobservationapplies needed. First of all, there is nostate ρ forwhich n = d−1 equallywelltotheinterrelationbetweenD andN/d. (1) holds;sotherelationmustbeastrictinequality.Althoughthe [1] T. Debarba, T. O. Maciel and R. O. Vianna, (4)automaticallyimpliesaviolationofEq.(3). Phys.Rev.A86,024302(2012). [7] S.RanaandP.Parashar,Phys.Rev.A85,024102(2012). [2] D.GirolamiandG.Adesso,Phys.Rev.A84,052110(2011). [8] T. Tufarelli, D. Girolami, R. Vasile, S. Bose and G. Adesso, [3] M.Piani,Phys.Rev.A86,034101(2012). arXiv:1205.0251 [4] E.Chitambar,Phys.Rev.A86,032110(2012). [9] The trace norm k.k , being a norm, must satisfy the tri- (1) [5] S.RanaandP.Parashar,Phys.Rev.A86,030302(R)(2012). angular inequality: kρ − σk ≤ kρk + kσk = 2. For (1) (1) (1) [6] Thisrelationisnot well-justifiedforcomparison, because the many properties and interpretations of the trace norm, we left hand side varies from 0 to 1 whereas the right hand side refer to A. Gilchrist, N. K. Langford and M. A. Nielsen, from0to1/4(i.e.,notnormalized).NotethataviolationofEq. Phys.Rev.A71,062310(2005). [10] T.Debarba,T.O.MacielandR.O.ViannaarXiv1207.1298v3.