ebook img

One-sided Device-Independent Quantum Key Distribution: Security, feasibility, and the connection with steering PDF

0.17 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 One-sided Device-Independent Quantum Key Distribution: Security, feasibility, and the connection with steering

One-sided Device-Independent Quantum KeyDistribution: Security, feasibility,andthe connection with steering CyrilBranciard1,EricG.Cavalcanti2,StephenP.Walborn3, ValerioScarani4,5,andHowardM.Wiseman2,5 1 School ofMathematicsandPhysics, TheUniversityofQueensland, StLucia, QLD4072, Australia 2 Centre for Quantum Dynamics, GriffithUniversity, Brisbane QLD 4111, Australia 3 Instituto de F´ısica, Universidade Federal do Rio de Janeiro, Caixa Postal 68528, Rio de Janeiro, RJ 21941-972, Brazil 4 CentreforQuantumTechnologiesandDepartmentofPhysics,NationalUniversityofSingapore, Singapore117543 5ARCCentreforQuantumComputationandCommunicationTechnology,GriffithUniversity,BrisbaneQLD4111,Australia (Dated:January24,2012) WeanalyzethesecurityandfeasibilityofaprotocolforQuantumKeyDistribution(QKD),inacontextwhere 2 only one of the two parties trusts his measurement apparatus. This scenario lies naturally between standard 1 QKD,wherebothpartiestrusttheirmeasurementapparatuses,andDevice-IndependentQKD(DI-QKD),where 0 neither does, and can be a natural assumption in some practical situations. We show that the requirements 2 for obtaining secure keys are much easier to meet than for DI-QKD, which opens promising experimental n opportunities.Weclarifythelinkbetweenthesecurityofthisone-sidedDI-QKDscenarioandthedemonstration a ofquantumsteering,inanalogytothelinkbetweenDI-QKDandtheviolationofBellinequalities. J 3 PACSnumbers:03.67.Dd,03.65.Ud,03.67.Mn 2 ] QuantumKey Distribution (QKD) allows two parties(Al- Intermediate scenarios between S-QKD and DI-QKD re- h ice and Bob) to establish secret keys at a distance, with se- quirelesstrustthantheformerandwillbeeasiertoimplement p curityguaranteedbythelawsofQuantumMechanics[1]. In thanthelatter[11]. Imagineforinstancethatabankwantsto - t standardQKD (S-QKD),the securityistypicallyprovedun- establish secretkeys with its clients; the bank couldinvesta n a der the assumption that Alice and Bob can trust the physi- lot of money to establish one trustworthy measurement de- u cal functioning of their preparation and measurement appa- vice,buttheclientsattheotherendofthechannelwouldcer- q ratuses. For instance, standard security proofsfor the BB84 tainlygetcheap(andinsecure)detectionterminals.Thisleads [ protocol[2]assumethatAlicesendsqubitstoBob,prepared usto studyone-sidedDI-QKD(1sDI-QKD):we consideran 3 in some eigenstatesof theσ orσ Paulioperators, andthat entanglement-based scenario in which Bob’s measurement z x v Bobmeasurestheminoneofthosetwobases. Recentdemon- apparatusistrusted,whileAlice’sisnot.Intheentanglement- 5 strationsofhackingofthedeviceshasshowntheimportance basedsetupthesourceisalsountrusted,althoughwewillalso 3 and weaknessof this assumption[3]. Moreover,since QKD discuss prepare-and-measure(P&M) implementationswhere 4 1 isbecomingcommerciallyavailable,AliceandBobmayend thesourceistrusted. Wepresentasecurityboundagainstco- . upbuyingtheirdevicesfromuntrustedproviders. herentattackswith similar assumptionsas in Refs. [7, 8], in 9 particularthatthedevicesarememoryless,asthisenablesthe 0 1 Remarkably, there are ways to guarantee security with strongest security analysis presently available. Focusing on 1 fewerassumptions.Theminimalsetofassumptionsistheone practicalimplementations,weshowthatthedetectorefficien- v: used in Device-IndependentQKD (DI-QKD) [4, 5]. There, ciesrequiredforapracticalimplementationof1sDI-QKDare i AliceandBobcancertifythesecurityofQKDbasedonlyon muchlower thanfor DI-QKD,makingit feasible with exist- X the observed violation of Bell inequalities [6]: the measure- ingdevices. Beforethat,letusstartbystressingalinkwitha r mentapparatusesareuntrustedblackboxes,withaknobsup- hierarchyoftestsofquantumnonlocality[12]. a posedly related to the measurementsettings. Alice and Bob have only to trust the random number generator with which QKD and quantum nonlocality.— It is known that no se- they vary the positions of the knob, and of course the in- cret key can be extracted in a QKD experiment if the chan- tegrityoftheirlocations. Whilethequalitativeunderstanding nel between Alice and Bob is entanglement-breaking [13]. “Bellviolationimpliessecurity”iscertainlytrue, thederiva- Hence, in order to demonstratesecurity, one must show that tionofquantitativesecurityboundsischallenging. Themost the channel preserves entanglement. The three different as- recent results report on security against the most general at- sumptionsonAlice’sandBob’sdevicesmentionedabovecor- tacks(“coherentattacks”)undertheassumptionthatprevious respondnaturallytothreedifferentcriteriaforquantumnon- measurementsdonotfeedanyinformationforwardtosubse- locality [12] (see Figure 1): S-QKD or DI-QKD require the quent ones [7, 8] (or, to put it simply: that the devices are observed correlations to violate a separability criterion or a memoryless). Inadditiontothese(hopefullytemporary)lim- Bell inequalityrespectively; 1sDI-QKDrequiresthe correla- itationsof the theoreticalstudies, DI-QKD imposesveryde- tionstoviolateanEPR-steeringinequalityasdefinedin[14]. manding requirements on practical demonstrations. In par- Thatis, if oneimaginesthatBob’ssystem has a definite (al- ticular, the Bell test would need to close the detection loop- beitunknowntohim)quantumstate,theprotocolmustprove hole[9],whichrequiresveryhighdetectionefficiencies[10]. thatAlice,byherchoiceofmeasurement,canaffectthisstate. 2 tions. Ontheotherhand,sinceAlice’smeasurementdeviceis (i) untrusted, Eve could control whether her detectors click de- pendingon the state she receives and on her choice of mea- surement setting. We can therefore not simply discard Al- ice’s no-detection events. In case her detectors don’t click, (ii) sherecordsabitvalueofherchoiceastheresultofhermea- surement, keeps track of the fact that her detectors did not click,andtellsBob(sothattheycanlaterpost-selecttheraw keyonAlice’sdetections);Evehasaccesstothatinformation. (iii) We denote by A and B the strings of classical bits Al- i i ice and Bob get from measurements A and B (and where i i Bobgota detection,as discussedabove). Amongthebitsof FIG.1: Linkbetweenthethreeconceptsofquantumnonlocalityas A ,somecorrespondtoactualdetectionsbyAlice,andsome, 1 classifiedin[12]andthethreescenariosofS-QKD,1sDI-QKD(this correspondingto non-detections,were simplychosen byAl- paper)andDI-QKD.Inordertoobtainasecretkey,(i)ifAliceand ice herself. Everyoneknowswhichonesare which. Thede- Bobtrusttheir measurement devices (transparent boxes), thenthey tectedbitsformastringAps(they’llbepost-selectedbyAlice mustnecessarilydemonstrateentanglement; (ii)ifAlice’smeasure- 1 mentdeviceisuntrusted(blackbox),whileBob’sistrusted,thenAl- andBob),whilethosethatwerenotactuallydetectedforma icemustdemonstratesteeringofBob’sstate;(iii)ifbothAlice’sand string Adis (they’ll be discarded for the key extraction), so 1 Bob’s measurement devices are untrusted, then they must demon- that A = (Aps,Adis). Bob’s correspondingbit strings are 1 1 1 strateBell-nonlocality. Inallcases, AliceandBobmusttrusttheir Bps andBdis, resp., so thatB = (Bps,Bdis). We denote 1 1 1 1 1 randomnumbergenerator(RNG),andtheintegrityoftheirlocation. byN thelengthofthestringsA andB ,andbynthatofthe 1 1 stringsApsandBps. 1 1 Security proof and key rate.— Recently, Tomamichel and Renner[20], togetheralso withLim andGisin [22] havede- Thissortofnonlocality,firstdiscussedbyEinstein,Podolsky veloped an approach to QKD based on an uncertainty rela- andRosen[15],wascalled‘steering’bySchro¨dinger[16]. In tionforsmoothentropies,whichenablesonetoprovesecurity Ref.[12]theseconceptsofnonlocalityarosefromconsidering againstcoherentattacksinpreciselythis1sDI-QKDscenario; entanglement verification with untrusted parties. However, notehoweverthatonealsoneeds(asin[7,8])theassumption evenifAliceandBobtrusteachother,asinQKD,theymay that the devices are memoryless [23]. To prove the security nottrusttheirdevices,whichisananalogoussituation. From of our protocolin realistic implementations, we extend their thisperspective,ourscenarioof1sDI-QKDcanthusbeseen analysisbyconsideringimperfectdetectionefficiencies[24]. asapracticalapplicationoftheconceptofquantumsteering. From the n-bit strings Aps and Bps, on which Eve may 1 1 A 1sDI-QKD protocol.— We consider the following 1sDI havesome(possiblyquantum)informationE,AliceandBob versionoftheBBM92entanglement-basedprotocol[17]: Al- canextract,throughclassicalerrorcorrectionandprivacyam- ice andBob receivesome(typically,photonic)quantumsys- plification(fromBobtoAlice),asecretkeyoflength[25] temsfromanexternalsource. Alicecanchoosebetweentwo ℓ Hǫ (Bps E) nh(Qps). (1) binarymeasurements,A1andA2;sinceshedoesnottrusther ≈ min 1 | − 1 measurementdevice,shetreatsitasablackboxwithtwopos- Here Hǫ (Bps E) denotes the smooth min-entropy [26] of siblesettings,yieldingeachtimeoneoftwopossibleoutputs. Bps,comndinition1ed|onquantumsideinformationE;histhebi- Bob, on the other hand, trusts his device to make projective 1 naryentropyfunction:h(Q) Qlog Q (1 Q)log (1 measurements B1 or B2 in some qubit subspace, typically Q);andQpsisthebiterrorra≡te−betwee2nA−psan−dBps. 2 − corresponding to the operators σz and σx respectively. Af- Toboun1dHǫ (Bps E), we willuse the1uncertai1ntyrela- ter publicly announcingwhich measurements they chose for min 1 | tion introduced in [20], which bounds Eve’s information on each system, Alice and Bob will try to extract a secret key B givenAlice’sinformationontheincompatibleobservable 1 fromtheconclusiveresultsofthemeasurementsA1 andB1; B . However,weneedtousethefullstringsB ,B ,aspost- 2 1 2 as explained below, the results of measurements A and B 2 2 selectionmayleadtoanapparentviolationoftheuncertainty willallowthemtoestimateEve’sinformation. relation.Usingthechainrule[25]andthedata-processingin- AliceandBobmightnotalwaysdetectthephotonssentby equality[27]forsmoothmin-entropies,wefirstboundEve’s the source, because of losses or inefficient detectors. Since informationonBpsrelativetoherinformationonB : 1 1 Bobtrustshisdetectors,hetruststhatEvecannotcontrolhis detections. Also,Evecannotgetanyusefulinformationfrom Hmǫin(B1|E) = Hmǫin(B1ps,B1dis|E) (2) Bob’s (null) result if the photons going to him are lost or if Hǫ (Bps BdisE)+log Bdis (3) ≤ min 1 | 1 2| 1 | she keeps them. Cases where Bob gets ambiguous results Hǫ (Bps E)+N n. (4) (e.g. doubleclicks)canbedealtwithusingthetechniquesof ≤ min 1 | − Ref.[18]–seeSupplementalMaterial[19]fordetails. Hence, Now,considerahypotheticalrunoftheprotocolwherethe we cansafelyconsideronlythecaseswhereBobgetsdetec- bitsofA andB wouldbemeasuredinthesecondbasis;we 1 1 3 denotebyA andB thecorrespondinghypotheticalstrings. 1.0 2 2 r Fromthegeneralizeduncertaintyrelationof[20],onehas e at 0.8 r Hmǫin(B1|E) ≥ qN −Hmǫax(B2|A2). (5) key et 0.6 where q is a measure of how distinct Bob’s two measure- cr e s mentsare; fororthogonalqubitmeasurements,q = 1. Here, he 0.4 tHiomǫnaexd(Bon2A|A22.)Itissathtiesfisemsotohtehfmolaloxw-einntgro[p2y2][26]ofB2,condi- ndont 0.2 u o Hǫ (B A ) . Nh(Q ), (6) B max 2| 2 2 0.0 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00 whereQ isthebiterrorratebetweenA andB .Now,since 2 2 2 Alice'sdetectionefficiencyΗ A the choice of basis was made randomly, Q is the same as 2 thebiterrorrateobserved—withoutpost-selection—whenthe FIG.2: Solidcurves: bounds(8)onthesecretkeyrater inatypi- second basis was actually chosen (no matter how rarely) by calimplementationof1sDI-QKD,asafunctionofAlice’sdetection bothAliceandBob. efficiency ηA, for visibilitiesV = 1,0.99,0.98,0.95 (from top to Substituting(4),(5)and(6)inEq.(1),weobtain bottom) and q = 1. Dashed curve: for comparison, bounds (for V = 1)forDI-QKD,obtainedbyadaptingthesecurityanalysisof ℓ & n[1 h(Qps)] N[h(Q )+1 q]. (7) Ref.[7], whenBobhasthesamedetection efficiencyηB = ηA as − 1 − 2 − Alice(seeSupplementalMaterial[19]). In the asymptotic limit of infinite key lengths, the aboveap- proximate inequality becomes exact [21, 22]. The fraction n/N ofphotonswhichAlicedetects,giventhatBobdetected one,willbedenotedη . Thisallowsustowritethesecretkey key from the non-post-selected data. If the key is extracted A rater ℓ/N (thenumberofsecretbitsobtainedperphoton fromthe post-selecteddata (asweconsideredherefor1sDI- detecte≡dbyBob,measuredinthefirstbasis),as QKD), the threshold remains quite high, η > 91.1% (see Figure 2 and SupplementalMaterial [19]). The much lower r η [1 h(Qps)] h(Q ) (1 q). (8) efficiency threshold for 1sDI-QKD compared with DI-QKD ≥ A − 1 − 2 − − is related to the fact that it is much easier to close the de- RelationtoEPR-steering.—Aswerecalledbefore,these- tection loophole in a steering experiment [28–30] than in a cretkeyrateabovecanonlybepositiveifAliceandBobcan Bell test, for whichthereare no photonicdetection-loophole checkthattheyshareentanglement.Inour1sDIscenario,this freedemonstrationstodate. Heraldingefficienciesof 62% amounts to demonstrating quantum steering. Hence, the in- have recently been reported [30], in an experiment d∼emon- equalityηA[1−h(Qp1s)]−h(Q2)−(1−q)≤0canbeunder- strating detection-loophole-freequantumsteering; our 1sDI- stoodasanEPR-steeringinequality[14].IntheSupplemental QKDprotocolcouldbedemonstratedwithaverysimilar(but Material[19],wegiveamoredirectproofofthisclaim,start- slightlyimproved)experimentalsetup. ingfromtheso-calledLocalHiddenStatemodel[12]. Notealsothatinthe1sDI-QKDcase,thelossesbetweenthe Experimental prospects.— We now turn to the feasibil- sourceandBob’slabdonotaffectthesecurityoftheprotocol; ity analysis. Consider a typical experimental setup, where theyonlydecreasethekeyrateproportionallytothedecrease a source sends maximally entangled 2-qubit states to Alice ofBob’sdetectionrate(aslongasthenoiseinBob’sdetectors and Bob, through a depolarizing channel with visibility V, does not become prominent). Hence, long distances can in and where, as in the BBM92 protocol, A = B = σ and 1 1 z principlebereachedifthesourcestaysclosetoAlice;thisis A =B =σ ,withAlice’sdetectionefficiencybeingη as 2 2 x A in contrastto the fully DI-QKD case, where the limit on the above. (WeemphasizethatthisissimplyamodelforAlice’s detectionefficienciesimposesalimitonthealloweddistance measurements,whichareimplementedinablackbox.) betweenAlice andBob (althoughsome proposalshavebeen The secret key rate that Alice and Bob can extractis then suggestedtoovercomethisproblem[31–33]). boundedby(8),withq =1and Comparison of different scenarios.— Key rate boundsfor Qps =(1 V)/2, Q =(1 η V)/2. entanglement-based QKD also apply to P&M schemes, as 1 − 2 − A long as the preparation device can be trusted to produce a Figure2showsthevaluesofthebound(8)asafunctionofη , certainaveragestateindependentofAlice’s(orBob’s,asthe A andfordifferentvaluesofV. ForaperfectvisibilityV = 1, casemaybe)choiceofpreparationbasis(e.g. thecompletely onegetsapositivesecretkeyrateforallη >65.9%. mixedstate,regardlessofwhetherσ orσ isthechosenba- A x z This detection probability threshold is much lower than sis). A preparation device with this property can be envis- those required for DI-QKD. For instance, if Alice and Bob aged as a trusted entanglementsource situated in Alice’s (or have the same detection efficiency η, then they require η > Bob’s) laboratory, with a trusted channel between it and the 94.6%fortheprotocolstudiedin[7],whentheyextracttheir localdetector. In this pictureitstill makessense to consider 4 thecasewherethelocal“hypothetical”detectorisuntrusted, Basedon AD C S C BD Keyratebound Eff.Thresh. asthisisequivalenttosayingthatwecannottrusttheprepa- P&M T T T U T r0(Qp1s,Qp2s) none r(awtihoinchdweveicweiltlodpenreoptearaestηh∗e)doefstihreedhystpaoteth.eHtiecraeldtheeteecftfiorcimenocdy- P&M U T T U T r1(Qp1s,Q2) ηA∗ >65.9% elstheprobabilitythatthepreparationdeviceregisterstothe P&M U U T T T r1(Qp1s,Q2) ηA >65.9% senderwhichofthetwostates(inthechosenbasis)wassent. P&M U U T T U r2(Qp1s,S) ηA >83.3% Forawell-functioningdevicethiscanbeclosetounity;even Entang. T U U U T r0(Qp1s,Qp2s) none ifthepreparationdoesuseaprobabilisticphoton-pairsource Steering U U U U T r1(Qp1s,Q2) ηA >65.9% andadetector,thesendercangeneratemanypairswithinthe Bell U U U U U r2(Qp1s,S) η>91.1% timewindowforeachsystem,andswitchoutthesystem(one photon,ideally)onlywhenits preparationis heraldedbythe TABLEI:BestknownboundsonsecretkeyratesforQKD(secure againstcoherentattacks)withprivacyamplificationfromBobtoAl- detection of the other. A greater experimental challenge is ice and memoryless devices, both P&M and entanglement-based. thelossoftheheraldedphoton(withinthesender’slaboren Thesecond column tellswhich of the components—Alice’s detec- route),whichmustbefactoredintothereceiver’sefficiency. tors(AD),thesource(S),Bob’sdetector(BD)andeachofthechan- Considering all non-trivial permutations of device trust- nels (C) between the source and the detectors—are trusted (T) or worthiness,thereareeightP&Mscenarios,andfourgenuine untrusted(U).ThickverticallinesineachrowseparateAlice’slab, entanglement-based scenarios, whose security can be anal- thechannelopentoEve,andBob’slab. IntheP&Mcases,thede- ysed using the methods of Refs. [20] or [7]. We remove tectorplussourceinalabisaformalmodelforapreparationdevice; theefficiencyη∗inthesecasesmodelstheprobabilitythattheprepa- mirror-imagescenariosbykeepingonlytheversionwhichis rationdeviceregisterswhichstateisprepared, andwould typically better (or equally good)underthe assumption of the privacy beclosetounity(fordetailsseetext). Incolumnthree,thebounds amplification being from Bob to Alice, as shown in Table I. onkeyrates(here,perphotonpairproducedbythesource, i.e.per NotethatP&MbyBobwithafullytrustedpreparationdevice preparationeventintheP&Mcases)aregivenintermsofthefunc- (row 3) does not improve the threshold efficiency required tionsr0(Qp1s,Qp2s) ≡ ηBηA[1−h(Qp1s)−h(Qp2s)]fromS-QKD, forAlice’suntrusteddetector,ascomparedto steering-based r1(Qp1s,Q2) ≡ ηB{ηA[1−h(Qp1s)]−h(Q2)}from Eq. (8), and 1sDI-QKDwithanuntrustedentanglementsource(row6). r2(Qp1s,S) ≡ ηAηB[1−h(Qp1s)]−log2(cid:2)1+p2−(S/2)2(cid:3)(see Conclusion.— We have introduced the scenario of 1sDI- SupplementalMaterial[19]).HereSisthevalueoftheCHSHpoly- QKD and analyzed its security against coherent attacks in nomial[34],whileQ1andQ2arebiterrorrates.Thesuperscriptps a practical situation where losses are taken into account. meanspost-selectiononcoincidentdetections;Q2inr1(Qp1s,Q2)is post-selectedonBob’sdetections,butnotonAlice’s;Smustbeesti- Ouranalysisshowsthattheassumptionsof1sDI-QKDallow matedfromthewholenon-postselecteddata.TheEfficiencyThresh- onetosignificantlylowerthenecessarydetectionefficiencies olds(columnfour)arecalculatedwitheverythingelseperfect.Inrow comparedtofullyDI-QKD,andthattherequirementsforob- 4,weassumedηB∗ →1(thethresholdwequoteisthereforealower tainingsecurekeyratesinanexperimentarewithintherange boundonthethresholdsforηB∗ <1),whileinrow7,ηA =ηB =η. ofcurrenttechnology. Wehavealsostressedthat1sDI-QKDrequirestheviolation of an EPR-steering inequality, in analogy with the require- Noteadded.—Whilefinishingwritingthismanuscript,we mentofviolationofaBellinequalityforsecurityofDI-QKD. becameawareofarelatedwork[35]wheresimilarboundson TherelationbetweentheQKDhierarchyandthenonlocality thedetectionefficiencythresholdsarederived. hierarchyintroducessomeopenquestions: (i)Ithasrecently beenshownthatsteeringcanbedemonstratedwitharbitrarily lowefficiencies[28]. Canonefind(andprovethesecurityof) 1sDI-QKD protocols that would also tolerate arbitrarily low efficiencies?(ii)With2measurementsettingsperparty,steer- [1] V.Scaranietal.,Rev.Mod.Phys.81,13011350(2009). ingcanbedemonstratedforefficienciesη > 50%[28,30]. [2] C.H.Bennett,G.Brassard,Proc.oftheIEEEInternationalCon- A Thereisalargegapbetweenthisandthethresholdof 66% ference on Computers, Systems and Signal Processing, Banga- ∼ lore,India(IEEE,NewYork,1984),p.175. for our 1sDI-QKD protocol. The same situation occurs for [3] L.Lydersenetal.,Nat.Photon.4,686(2010);I.Gerhardtetal., fullyDI-QKD,wherethereisalso agapbetweenthethresh- Nat.Commun.2,349(2011) oldofη >82.8%foraviolationoftheCHSHinequality[34] [4] D.Mayers,A.Yao,Proc.ofthe39thIEEEConf.onFoundations andthatforthesecurityofDI-QKD.Howsmallcanthesebe ofComputerScience(IEEE,LosAlamitos,1998),p.503. madeingeneral? Anothertopicforfurtherresearchistoex- [5] A.Ac´ınetal.,Phys.Rev.Lett.98,230501(2007). tend ourresults to finite keys, forinstance alongthe linesof [6] J.S.Bell,Physics1,195(1964). Ref.[22]. [7] L.Masanes,S.Pironio,A.Acin,Nat.Comm.2,238(2011). [8] E.Ha¨nggi,R.Renner,arXiv:1009.1833. Acknowledgements.— This work was supported by the [9] P.M.Pearle,Phys.Rev.D2,1418(1970). Australian Research Council Programs CE110001027 and [10] S.Pironioetal.,NewJ.Phys.11,045021(2009). DP0984863,theNationalResearchFoundationandtheMin- [11] M.Pawlowski,N.Brunner,Phys.Rev.A84,010302(R)(2011). istry of Education, Singapore, the Brazilian agencies CNPq, [12] H.M.Wiseman,S.J.Jones,A.C.Doherty,Phys.Rev.Lett.98, FAPERJ,andtheINCT-Informac¸a˜oQuaˆntica. 140402(2007). 5 [13] M.Curty,M.Lewenstein,N.Lu¨tkenhaus, Phys.Rev.Lett.92, [23] S.FehrandR.Renner,privatecommunication. 217903(2004). [24] Toavoidanyconfusion,wenotethatherewehavereversedthe [14] E.G. Cavalcanti, S. J. Jones, H. M. Wiseman &M. D. Reid, roles of Alice and Bob compared to that in Refs. [20, 22], in Phys.Rev.A.80,032112(2009). ordertobeconsistentwithearlierworkonsteering[12,14],in [15] A.Einstein,B.Podolsky,N.Rosen,Phys.Rev.47,777(1935). whichAlice’sdetectorsarenottrusted,whileBob’sare. [16] E.Schro¨dinger,Proc.Camb.Phil.Soc.31,555(1935). [25] J.M.Renes,R.Renner,arXiv:1008:0452. [17] C.H.Bennett,G.BrassardandN.D.Mermin,Phys.Rev.Lett. [26] R. Renner and S. Wolf, in Advances in Cryptology - ASI- 68,557(1992). ACRYPT,vol.3788ofLNCS(Springer),199(2005); R.Ren- [18] T.Moroder, O.Guhne, N.Beaudry, M.Piani, N.Lutkenhaus, ner,Int.J.QuantumInf.6,1(2008). Phys.Rev.A81,052342(2010). [27] M.Tomamichel,R.Colbeck,R.Renner,IEEETrans.Inf.The- [19] See Supplemental Material below for more details on practi- ory56,4674(2010). caldetectorsandmultiplecounts, adirectproof ofthesteering [28] A.J.Bennetetal.,arXiv:1111:0739. inequality, and a more detailed comparison of our results with [29] B.Wittmannetal.,arXiv:1111:0760. thoseforDI-QKD. [30] D.H.Smithetal.,Nat.Commun.3,625(2012). [20] M. Tomamichel, R. Renner, Phys. Rev. Lett. 106, 110506 [31] N. Gisin, S. Pironio and N. Sangouard, Phys. Rev. Lett. 105, (2011). 070501(2010). [21] Theapproximate(in)equalitiesinthederivationofEq.(7)ac- [32] M.Curty,T.Moroder,Phys.Rev.A84,010304(2011). countforadditionalcorrectingtermsthatappearintheequations [33] D.Pitka¨nenetal.,Phys.Rev.A84,022325(2011). inthecaseoffinitekeylengths.Onetypicallyfindsthatforraw [34] J.F.Clauser,M.A.Horne,A.ShimonyandR.Holt,Phys.Rev. keylengthslonger than104 −105 bits, significant fractionsof Lett.23,880(1969). theasymptoticsecretkeyratesareobtained[22]. [35] X.Ma,N.Lu¨tkenhaus,arXiv:1109.1203. [22] M. Tomamichel, C. C. W. Lim, N. Gisin, R. Renner, arXiv:1103:4130. 6 Supplemental Material Practicaldetectorsandmultiplecounts.—Inpracticalim- with H(B a˜ ) = P(b a˜ )log P(b a˜ ) and i| i −Pbi i| i 2 i| i plementations with photons, Bob’s detector comprises two H(B λ)= P (b λ)log P (b λ). i| −Pbi Q i| 2 Q i| photon counters, corresponding to his two possible outputs. Now,Bob’sstatistics,conditionedonλ,arebyassumption Becauseofdarkcountsorimperfectstatepreparation(i.e. the givenbyaquantumstate;theymustthereforesatisfyallquan- presence of multi-photonstates) there will be rare occasions tumuncertaintyrelations,andinparticular wherebothcountersclick.Insuchcases,ourprotocolrequires Bobtooutputarandomresult. Thisisastandardwaytotreat H(B1 λ)+H(B2 λ) q, (S4) | | ≥ doubleclicksinthecontextofQKD[S1]. WhenBobimple- whereqquantifieshowdifferentthemeasurementsB andB ments the σ or σ measurements, his detection scheme has 1 2 z x are[S5]. Togetherwith(S3),weobtain a “squashing property” that should allow one to extend the present theoreticalsecurity proof for perfectqubit states (on H(B A˜ )+H(B A˜ ) q. (S5) 1 1 2 2 Bob’sside)toimperfectimplementations[S2]. | | ≥ Denoting Aps for A when s = 1 and Adis for A when i i i i i Direct proof that “r 0” is a steering inequality.— We s = 0, and assuming that P(s = 1) = η independently ≤ i i A obtained in Eq. (8) in the main text a bound on the secret ofA (asshouldbethecaseifitisadetectorefficiency),we i key rate of our 1sDI-QKD protocol. As we argued, a posi- have tivesecretkeyratecanbeobtainedonlyifAliceandBobcan demonstrate steering. That is, if they violate the inequality H(B A˜ )=η H(Bps Aps)+(1 η )H(Bdis Adis). (S6) i| i A i | i − A i | i “r.h.sofEq.(8) 0”,ormoreexplicitly ≤ Now,onecaneasilycheck(usingforinstancetheconcavity ηh(Qp1s)+h(Q2)+(1−η) ≥ q, (S1) ofthebinaryentropyfunction)thatforbinaryvariablesAand B,onehas thentheycanconcludethattheyhavedemonstratedsteering: Eq.(S1)thusdefinesanEPR-steeringinequality[S3]. H(B A) h(Q), (S7) | ≤ Hereweshowinadifferentwaythat(S1)isindeedanEPR- where Q is the bit errorrate relativeto A and B (the proba- steeringinequality,startingmoredirectlyfromtheLocalHid- bilitythatA=B). Applyingthisto(S6)fori=1,itfollows den State model [S3]—a local model whereby the statistics 6 that observedbyAliceandBobcanbedecomposedas[S4] H(B A˜ ) η h(Qps)+(1 η ), (S8) P(a˜i,bj)=XP(λ)P(a˜i λ)PQ(bj λ), (S2) 1| 1 ≤ A 1 − A | | λ whileontheotherhand ewxhpeerreima˜ien=tal(aoiu,tsc)o,mweisthcaoirr=esp±o1ndainndgbtjo=me±as1udreemnoetnintsgAthe, H(B2|A˜2)≤H(B2|A2)≤h(Q2). (S9) i B ,andsdenotingwhetherAlicegetsadetection(s = 1)or Substituting (S8) and (S9) in (S5) we finally obtain (S1), as j not(s = 0);recallthatincaseofnodetection,Alice’sresult desired. a ischosentobeapre-establishedbitinour1sDI-QKDpro- Note that for the purpose of deriving an EPR-steering in- i tocolandwillcontributetothebitstringAdis. Thesubscript equality, there is no need to follow the last steps of the i Q indicates that Bob’s statistics are determined by quantum above procedure where we treat and bound H(B A˜ ) and 1 1 mechanics: P (b λ) = Tr[E ρ ], where ρ is some quan- H(B A˜ ) differently. Rather we can use Eq. (S5)| directly Q j j λ λ 2 2 | | tumstate(inourcase,aqubitstate)and E arethepositive asanEPR-steeringinequality. Undertheusualconditionsfor j { } operatorssummingtounitywhichdescribehismeasurements. QKD as discussed in the main text, q = 1 and Eq. (S8) be- The variable λ fully specifies the quantum state received comesatightinequality,soEq.(S5)thenbecomes byBob. Asaconsequence,Alicecanhavenomoreinforma- tiononaverageaboutBob’sresultsthanwhatonecouldknow ηA[2−h(Qp1s)−h(Qp2s)]≤1. (S10) throughλ. Intermsofconditionalentropies,thisimpliesthat For perfect correlations(Qps = Qps = 0), this will be vio- 1 2 H(Bi A˜i) H(Bi Λ), (S3) lated(indicatingEPR-steering)providedthatηA > 21. Thisis | ≥ | the minimum possible efficiency threshold for EPR-steering where H(B A˜ ) = P(a˜ )H(B a˜ ) and usingtwosettings[S6]. i| i Pa˜i i i| i H(B Λ) = P(λ)H(B λ) are the average en- i| Pλ i| tropies of Bob’s measurement B conditioned on the Comparison with DI-QKD.— In the DI-QKD case, the i knowledge of the result of A˜ and of λ, respectively, strongestsecurityanalysisavailabletodayisthatofMasanes, i 7 Pironio and Ac´ın [S7], who showed the security of a large arethefiguresappearinginTable1inthemaintext. Inboth class ofprotocolsagainstcoherentattacks(underthe techni- cases, butespecially the P&M case, using ourpost-selection calassumptionthatthedevicesarememoryless). Inthesim- techniqueyieldssubstantiallybetterefficiencythresholdsthan plest version, the protocol they consider involves Alice and thenon-post-selectedprotocols.Wenotethatsomewhatlower Bob measuringin bases that enable a CHSH inequality[S8] efficiency thresholds for DI-QKD (η > 92.3% and η > to be tested, where one of Bob’s bases is that which is used 88.9%fortheentanglement-basedschemes,withandwithout (togetherwithadifferentoneofAlice’sbases)toyieldthese- post-selection, resp.) can be obtained using the analysis of cretsharedkey. Letthe CHSH polynomialhavethe valueS Ref.[S9]. However,thisanalysisshowssecurityonlyagainst (suchthat S < 2√2forquantumcorrelations,and S > 2 collectiveattacks (which includes, in particular, the assump- | | | | indicatesaviolationoftheCHSHinequality),andletthenon- tionthatthedevicesarememoryless). post-selectedquantumbiterror rate for measurementsin the In the main text we presented only the post-selected ver- basesthatwillyieldthekeybeQ1. ThenRef.[S7]showsthat sion of 1sDI-QKD. However one can also consider a non- thefinalkeyratecanbeboundedfrombelowby post-selected version of that, in which Alice and Bob try to extract a secret key directly from the non-post-selected data rMPA =1−h(Q1)−log2(cid:2)1+p2−(S/2)2(cid:3). (S11) A1,B1 (rather than from Ap1s,B1ps; note however that we still post-select on a detection by Bob). In this case the ForavisibilityV andinefficienciesη andη ,wehavearaw A B key rate per detection by Bob is bounded from below by sharedrandomkeywithbiterrorrateQ =[1 (1 η )(1 1 A 1 h(Q ) h(Q )forsecurityagainstcoherentattacks.Here − − − 1 2 η ) Vη η ]/2,andaCHSHparameterS =2√2Vη η + − − B A B A B foratypicalimplementationthebiterrorratesarebothgiven − 2(1 η )(1 η ). The latter is obtainedif Alice and Bob − A − B by Qi = (1 VηA)/2, and with V = 1, we get a positive use the same predeterminedlist of random results whenever − rate for η > 78.0%. Recall that when Alice and Bob use A they fail to detect a photon. With V = 1 and η = η = A B the post-selected data, the threshold is η > 65.9%, so the A η, Eq. (S11) yields a positive key rate for η > 94.5%. In improvementisevenmoredramaticthaninthefullyDIcase. thesituationofprepare-and-measure(P&M),wherethereisa trustedsourceinBob’slab,buthisdetectorisstilluntrusted, we replace η by η∗, Bob’s preparation efficiency. Taking B B η∗ 1,Eq.(S11)givesalowerboundη− >89.6%. B → A [S1] N.Lu¨tkenhaus,Phys.Rev.A59,3301(1999);ibid.,61,052304 We now show that Alice and Bob can improve their se- (2000). cretkeyratebyextractingtheirsecretkeyfromthedatapost- [S2] T.Moroder, O.Guhne, N.Beaudry, M.Piani,N.Lutkenhaus, selected oncoincidencedetections, as we suggestedhere for Phys.Rev.A81,052342(2010). thecaseof1sDI-QKD.Indeed,onecanapplyouranalysisto [S3] E.G. Cavalcanti, S.J. Jones, H. M. Wiseman& M. D.Reid, that of Ref. [S7] to find a new secret key rate boundedfrom Phys.Rev.A.80,032112(2009). belowby [S4] H.M.Wiseman,S.J.Jones,A.C.Doherty,Phys.Rev.Lett.98, 140402(2007). r2 =ηAηB[1−h(Qp1s)]−log2(cid:2)1+p2−(S/2)2(cid:3). (S12) [S5]RDev..DDe.u3ts5c,h3,0P7h0y(s1.9R8e7v).;LHe.ttM.5a0a,ss6e3n1a(n1d98J.3)U;fKfin.kK,rPahuys,s.PRhyevs.. Lett.60,1103(1988). Here S is unchanged from above (it must still be estimated [S6] A.J.Bennetetal.,arXiv:1111:0739. from the whole non-postselected data), while Qp1s = (1 − [S7] L.Masanes,S.Pironio,A.Acin,Nat.Comm.2,238(2011). V)/2. WithV = 1andηA = ηB = η, thisyieldsapositive [S8] J.F.Clauser,M.A.Horne,A.ShimonyandR.Holt,Phys.Rev. key rate for η > 91.1%. Considering again the P&M case Lett.23,880(1969). withη∗ 1,weobtainthelowerboundη− >83.3%.These [S9] S.Pironioetal.,NewJ.Phys.11,045021(2009). B → A

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.