9512.net

甜梦文库

甜梦文库

当前位置：首页 >> >> # The Universal Composable Security of Quantum Key Distribution

The Universal Composable Security of Quantum Key Distribution

Michael Ben-Or1,4,6 , Michal Horodecki2,6 , Debbie W. Leung3,4,6 , Dominic Mayers3,4, and Jonathan Oppenheim1,5,6

benor@cs.huji.ac.il, ?zmh@univ.gda.pl, wcleung@cs.caltech.edu, dmayers@cs.caltech.edu, & J.Oppenheim@damtp.cam.ac.uk Institute of Computer Science, The Hebrew University, Jerusalem, Israel Institute of Theoretical Physics and Astrophysics, University of Gda?sk, Poland n 3 Institute for Quantum Information, California Institute of Technology, Pasadena, USA 4 Mathematical Science Research Institute, Berkeley, USA 5 Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, UK 6 Isaac Newton Institute, University of Cambridge, Cambridge, UK

2 1

arXiv:quant-ph/0409078v1 14 Sep 2004

(February 1, 2008) The existing unconditional security de?nitions of quantum key distribution (QKD) do not apply to joint attacks over QKD and the subsequent use of the resulting key. In this paper, we close this potential security gap by using a universal composability theorem for the quantum setting. We ?rst derive a composable security de?nition for QKD. We then prove that the usual security de?nition of QKD still implies the composable security de?nition. Thus, a key produced in any QKD protocol that is unconditionally secure in the usual de?nition can indeed be safely used, a property of QKD that is hitherto unproven. We propose two other useful su?cient conditions for composability. As a simple application of our result, we show that keys generated by repeated runs of QKD degrade slowly.

1

Introduction

Quantum cryptography di?ers strikingly from its classical counterpart. On one hand, quantum e?ects are useful in the construction of many cryptographic schemes. On the other hand, dishonest parties can also employ more powerful quantum strategies when attacking cryptographic schemes. The security of quantum key distribution One of the most important quantum cryptographic applications is quantum key distribution (QKD) [1, 2, 3]. The goal of key distribution (KD) is to allow two remote parties, Alice and Bob, to share a secret bit string. Classically, KD cannot be unconditionally secure (i.e. secure against all possible classical attacks) (see Sec. 2). Furthermore, the security of existing KD schemes is based on assumptions in computation complexity or limitations of the memory space of the adversary, Eve. In contrast, QKD is based on an intrinsic property of quantum mechanics, “extracting information about an unknown quantum state inevitably disturbs it,” [4] which allows eavesdropping activities to be detected in principle. Indeed, QKD can be unconditionally secure, i.e., against Eve whose capability is only limited by quantum mechanics [5, 6, 7, 8, 9, 10, 11]. Furthermore, QKD remains secure even if the quantum states are sent through a noisy quantum channel, as long as the observed error rates are below certain threshold values. In what sense is QKD secure? We will describe the assumptions and security de?nitions more formally in Sec. 2. In QKD, Alice and Bob are assumed to start with a small initial key Ki (for authentication purposes). They have access to uncorrelated randomness that is not controlled by Eve. They may exchange quantum and classical messages in both directions via channels that are completely under the control of Eve, and may perform local quantum operations and measurements. Based on their measurement outcomes, Alice and Bob either abort QKD or generate their respective keys KA , KB . Correspondingly, we say that the QKD test is failed or passed, and the events can be described as M =0 or M >0, where M is the length of the key generated. Eve also obtains quantum and classical data (her “view” or “transcript”) from which she extracts classical data KE via a measurement. What happens during a speci?c run of QKD depends on Eve’s strategy as well as the particular outcomes of the coins and quantum measurements of all the parties. However, the security of QKD can still be captured by requiring that (1) the conditional mutual information I(KE : KA , KB |M ) is negligible and (2) for all eavesdropping strategies with nonnegligible Pr(M >0), KA ,

2 KB are near-uniform and Pr(kA = kB ) is negligible. Throughout the paper, we use capitalized letters KA , KB , KE , and M to denote the random variables, and uncapitalized letters to denote speci?c outcomes. The security problem of using QKD Proofs of security of QKD (in the sense described above) address all attacks on the QKD scheme allowed by quantum mechanics. The problem is that QKD is not the only occasion for attack — further attack may occur when Alice and Bob use the keys generated. In particular, Eve may never have made a measurement during QKD to obtain any KE . Eve’s transcript is a quantum state. She could have delayed measurements until after more attack during the application, a strategy with power that has no classical counterpart. In other words, security statements in QKD that revolve around bounding I(KE : KA , KB |M ) is not applicable if the key is to be used! The limitations of mutual-information-based security statements were known as a folklore for some time (for example, see Sec. 4.2 in [11]). One of the earliest known security problems in QKD is the following [12]: QKD requires a key for authentication, which may in turns come from a previous round of QKD. Since each run of QKD is slightly imperfect, repeated QKD produce less and less secure keys. A conclusive analysis on the degradation has been evasive, since joint attacks over all runs of QKD have to be considered. As it turns out, there are many other occasions in which joint attacks on QKD and the subsequent use of the generated key have to be considered. For example, suppose Alice and Bob perform QKD to obtain a key, and then use the key to encrypt quantum states [13, 14]. Eve eavesdrops during both QKD and encryption and performs a collective measurement on the two eavesdropped states. It is well-known that such a collective measurement may yield more accessible information than the sum of information obtained in two separate measurements [15]. Our current study is further motivated by the results in [16, 17], which show that there are ensembles of quantum states that provide little accessible information on their own, but can provide much more information when a little more classical data is available. The extra information can be arbitrarily large compared to both the initial information and the amount of extra classical data. Such strange property reveals a new, unexpected, inadequacy of mutual-information-based statements. In particular, in the context of QKD, the usefulness of bounding the initial accessible information of Eve becomes very questionable, if Eve delays her measurement until further data is available during the application of the key — the security of the key is questionable even in classical applications! The goal of the current paper is to study the security of using a key generated by QKD, i.e., the composability of QKD. The universal composability approach Composability is an active area of research that is concerned with the security of composing cryptographic primitives in a possibly complex manner. The simplest example is the security of using a cryptographic primitive as a subroutine in another application. Our paper will follow the universal composability approach. For a speci?c task (functionality), a primitive that realizes the task is said to be universal composable if any application using the primitive is about as secure as using the ideal functionality. A security de?nition that ensures universal composability was recently proposed by Canetti [18], and was extended to the quantum setting by some of us [19, 20]. Such universal composable security de?nitions are useful because they are in terms of the ideal functionality only, without reference to the potential application. The security of a complex protocol can then be analyzed in terms of the security of each individual component in a systematic and error-proof manner. In the quantum setting, universal composability provides the only existing systematic technique for analyzing security in the presence of subtleties including entanglement and collective attacks. We will see in this paper that universal composability provides the precise framework for proving the security of using the keys generated from QKD, a problem that appears intractable at ?rst sight.

3 We note that an alternative approach to achieve universal composability in the classical setting was obtained in [21], with a generalization to the quantum setting studied in [22]. Main Results We have pointed out a serious potential security problem in using the keys generated from QKD. We will address the problem in the rest of the paper. We derive a new security de?nition for QKD that is universal composable. The essence is that QKD and certain ideal KD should be indistinguishable from the point of view of potential adversaries. Then, we prove that the original mutual-information-based security de?nition implies the new composable de?nition. Other simple su?cient conditions for the composable security of QKD will be discussed. One of these conditions, high singlet-?delity, has always been an intermediate step in the widely-used “entanglement-based” security proofs of QKD. We show that high singlet-?delity is much more closely related to composable security than the usual security de?nition, and we obtain much better security bounds for known QKD schemes. We thus prove the security of using a key generated by QKD in various ways, and provide simple criteria for future schemes. As a corollary, we answer the long standing question concerning the extent of key-degradation in repeated use of QKD [12]. Our work also has non-cryptographic applications in the study of correlations in quantum systems. The various security conditions are tied to correlation measures in quantum systems. Each derivation for the composable security for QKD is based on relating a pair of correlation measures. Related work Since the current result was initially presented [23, 24], various related results were reported. The composable security of generic classes of QKD schemes were proved in [25, 26], following a di?erent approach of showing the composable security of certain privacy ampli?cation procedures against quantum adversaries [25]. These related works share the concerns raised in this paper, with results complementary to ours. Organization of the paper We end this section by introducing some basic elements in the quantum setting. We review QKD in Sec. 2, stating our de?nitions and assumptions more formally. In Sec. 3, we review the quantum universal composability theorem. We will restrict ourselves to the much simpler case concerning unconditional security. We start describing our main results in Sec. 4, which contains a derivation of a simple criteria for the universal composable security for QKD. In Sec. 5, we prove that the usual security de?nition for QKD implies the universal composable security. In addition, we demonstrate two other su?cient conditions for composable security. One is based on bounding the Holevo information of Eve on the key. The other is based on bounding the singlet-?delity in security proofs using entanglement-puri?cation. The latter implies much better security of existing QKD protocols than is generically implied by the usual security de?nition. We conclude with lessons learnt from the current results. Frequently used notations and some complicated information theoretic quantities are listed in the appendix. Basic elements of quantum mechanics A quantum system or register is associated with a Hilbert space H. We only consider ?nite dimensional Hilbert spaces. Let B(H) and U(H) denote, respectively, the set of bounded operators and the unitary group acting on H. We loosely refer to the system as H also. A composite quantum system is associated with the tensor product of the Hilbert spaces associated with the constituent systems. The state of H is speci?ed by a positive semide?nite density matrix ρ ∈ B(H) of unit trace. A density matrix is a convex combination of rank-1 projectors (commonly called pure states) and represents a probabilistic mixture of pure states. Up to an overall-phase that is not physically observable, pure states can be represented as vectors in H. |ψ and |ψ ψ| denote the vector and rank-1 projector respectively. A measurement M on H is de?ned by a POVM, which is a decomposition of the identity into a set of positive semide?nite operators {Ok }, i.e., k Ok = I. If the state is initially ρ, the measurement M yields

We mention two distance measures for quantum states. The ?rst is the trace distance ρ1 ? ρ2 1 between the density matrices. It can be interpreted as the maximum probability of distinguishing between the two states. The second measure is the ?delity, F (ρ1 , ρ2 ) = max|ψ1 ,|ψ2 | ψ1 |ψ2 |2 where ρ1,2 ∈ B(H), |ψ1,2 ∈ H?H′ are “puri?cations” of ρ1,2 (i.e., TrH′ |ψ1,2 ψ1,2 | = ρ1,2 ), and ·|· is the inner product in H. We refer our readers to the excellent textbook by Nielsen and Chuang [27] for a more comprehensive review of the quantum model of information processing.

The most general evolution of the state is given by a trace-preserving completely-positive (TCP) linear map E acting on B(H). Any such E can be implemented by preparing a pure state in some ancillary system H′ , applying a joint unitary operator U ∈ U(H ? H′ ), and discarding H′ (i.e., a partial trace over H′ ).

√ √ outcome k with probability Tr(Ok ρ) and changes the state to Ok ρ Ok /Tr(Ok ρ). M is said to be along a basis {|k } if {Ok } = {|k k|}. Measuring an unknown state generally disturbs it.

4

2

Quantum Key Distribution

The goal of key distribution (KD) is to allow two remote parties, Alice and Bob, to share a secret bitstring such that no third party, Eve, will have much information about the bitstring. KD is impossible unless Alice and Bob can identify one another and detect alterations of their communication. In other words, the task of message authentication is necessary for KD. There are unconditionally secure methods for authenticating a classical message with a much shorter key [28]. Thus, KD uses authentication as a subroutine, and achieves key expansion (producing a key using a much shorter initial key). Classically, unconditionally secure KD between two remote parties is impossible. Classical physics permits an eavesdropper to have exact duplicates of all communications in any KD procedure without being detected. In contrast, while quantum key distribution (QKD) cannot prevent eavesdropping, it can detect eavesdropping. This allows Alice and Bob to avoid generating compromised keys with high probability. The usefulness of QKD is to avoid Alice and Bob being fooled into having a false sense of security. It is worth emphasizing what QKD does not o?er. First, QKD does not promise to always produce a key, since Eve can cause QKD to be aborted with high probability with intense eavesdropping. Second, there is a vanishing but non-zero chance that Eve is undetected, so that one cannot make simple security statements conditioned on not aborting QKD. How and why QKD works, through an example Various QKD schemes have been proposed and we only name a few here: BB84 [1], E91 [2], B92 [3], and the six-state scheme [29, 30]. We illustrate the general features and principles behind QKD by describing the class of prepare-&-measure schemes. Recall that Alice and Bob are given secure local coin tosses. Step 1: Alice ?rst generates a random bitstring, encodes it in some quantum state ρA , and sends ρA to Bob through an insecure quantum channel controlled by Eve. During this time, Eve can manipulate the message (system A) in any way allowed by quantum mechanics. Eventually, she will have to give some quantum message ρB to Bob for QKD to proceed. Mathematically, Eve’s most general operation can be described as attaching a private system E in the state |0 0|E , applying a joint unitary operation U to produce a joint state ρ = U (ρA ?|0 0|E ) U ? , and passing system A to Bob (relabeled as system B). Thus, Bob and Eve share the joint state ρ, and ρB := TrE ρ, ρE := TrB ρ are their respective reduced density matrices. Meanwhile, Bob measures ρB (according to his coin tosses). Step 2: Bob acknowledges to Alice receipt of the quantum message. Step 3: Only after Alice hears from Bob will further classical discussion be conducted over a public but authenticated channel. Step 4: At the end, based on their measurement outcomes and discussions, Alice and Bob either abort QKD (m = 0), or generate keys KA and KB (m > 0), and they announce m. Eve will have access to all the classical communication between Alice and Bob, besides the state ρE . She can measure ρE at any time to obtain a classical string KE , though it is to her advantage to wait until after she receive the classical communication. See Figure 1 for a schematic diagram

5 for the class of prepare-&-measure QKD schemes.

Eve |0 Step 1 rA Alice Step 2 Step 3 Step 4 kA m-bit ρA U E ρE

? measure s

ρB rB

kE

? A?

d ? dB

measure

c

' ' '

“received” Bob “discussion” “m”

E E

kB m-bit

Figure 1: Schematic diagram for the class of prepare-&-measure QKD schemes. The classical messages, represented

by double lines, are available to Eve. Eve can make her measurement any time after step 1. Dashed boxes represent private laboratory spaces. Outcomes of Alice and Bob’s local coins are represented by rA , rB .

The principle behind QKD is that, in quantum mechanics, one can only reversibly extract information from an unknown quantum state if the state is drawn from an orthogonal set [4]. Thus in the prepare-&measure scheme described above, if Alice encodes her message using a random basis chosen from several nonorthogonal possibilities, and Eve is to obtain any information on the outcomes of KA , KB , then ρB = ρA . To detect the disparity, Bob measures some of the received qubits (the “test-qubits” chosen randomly to avoid Eve tailoring her attack) and discusses with Alice to check if his measurement outcomes are consistent with what Alice has sent. This intuition can be turned into a provably secure procedure. Alice and Bob estimate various error rates on the test-qubits. If the observed error rates are below certain threshold values, it is unlikely that the untested qubits have much higher error rates. Error reconciliation and privacy ampli?cation are applied to extract bitstrings kA and kB for Alice and Bob respectively. If the observed error rates are above the thresholds, Alice and Bob abort QKD. QKD remains secure whether the observed noise is due to natural channel noise or due to eavesdropping. General features of any QKD scheme There are other QKD schemes besides prepare-&-measure schemes, for example, the entanglement-based QKD schemes (see [2, 7, 31]). Unless otherwise stated, our discussion applies to all QKD schemes. The basic ingredients are still secure local coins, completely insecure quantum communication, and authenticated public classical communication between Alice and Bob. In the most general QKD scheme, the ingredients may be used in any possible way. Alice and Bob still obtain some bitstrings as the output keys, kA and kB , of certain length m. Eve’s view is still given by some quantum and classical data, denoted collectively by ρE,kA ,kB , with explicit dependence on kA , kB . (Her view is a draw from an ensemble.) We emphasize a limitation in QKD. It is possible for Eve to be “lucky,” for example, to have attacked only the untested qubits, or to have attacked every qubit without causing inconsistency in Alice and Bob’s measurements. Thus, it is unlikely, but still possible, for Eve to have a lot of information on the generated key without being detected. No QKD protocol can make the promise “conditioned on passing the test, the keys KA , KB will be so-and-so.” With the above limitation of QKD in mind, there are several approaches to a proper security statement. The approach that is most commonly used in existing security proofs is to bound the probability that Alice and Bob generate bitstrings that are not equal, uniform, or private. We will use a more compact statement in the following. Let n be a security parameter in QKD (for example, the number of qubits transmitted from Alice to Bob). Fix an arbitrary eavesdropping strategy. The attack induces a distribution Pr(M =m) on the key length

6 M . The average value of M is typically a small fraction of n. The outcome m in a particular run of QKD depends on the outcome of the coins and measurements by Alice and Bob. We can assume that m is made public at the end of QKD. Recall m > 0 if the QKD test is passed and m = 0 if QKD is aborted. Let p qkd denote the distribution of KA , KB generated in QKD conditioned on |KA | = |KB | = m, i.e., p qkd (kA , kB ) = Pr(KA = kA , KB = kB |M = m) . Let pideal be the following distribution over two m-bit strings, p ideal (l, l)

(m) (m) (m) (m) (m)

(1)

= 2?m if l = l′ .

p ideal (l, l′ ) = 0

(2)

Let V denote the set of exponentially decaying functions of n. With these notations, a simple statement for the security condition can be made. Usual security de?nition for QKD: A QKD scheme is said to be secure if the following properties hold for all eavesdropping strategies. ? Equality-and-uniformity: ??1 ∈ V s.t.

m=0

Pr(m) p ideal ? p qkd

(m)

(m)

1

≤ ?1

(3)

? Privacy: ??2 ∈ V s.t.

m=0

Pr(m) × I(KE : KA , KB | M = m ) ≤ ?2

(4)

where I above denotes the mutual information [32] between KE and KA , KB conditioned on M = m. Using the equality condition, we only need to focus on kA =: k in Eq. (4). In particular, ? Privacy: ??′ ∈ V s.t. 2

m=0

Pr(m) × I(KE : K | M = m ) ≤ ?′ 2

(5)

The above security conditions revolve around expressions that can be interpreted as deviations from the desired properties, averaged over m. The product in each summand precisely capture the security requirement that large deviations from the desired properties should be a low probability event. Note that the (m) (m) m = 0 terms do not contribute, as p ideal ? p qkd 1 = 0 and I(KE : KA , KB | M = 0 ) = 0.

3 Quantum Universal Composability Theorem

Cryptographic protocols often consist of a number of simpler components. A single primitive is rarely used alone. A strong security de?nition for the primitive should thus re?ect the security of using it within a larger application. This allows the security of a complex protocol to be based only on the security of the components and how they are put together, but not in terms of the details of the implementation. A useful approach is to consider the universal composability of cryptographic primitives [18, 19, 20]. The ?rst ingredient is to ensure the security of a basic composition. We need a security de?nition stated for a single execution of the primitive that still guarantees security of composition with other systems. This de?nition involves a description of some ideal functionality of the primitive (i.e. the ideal task the primitive should achieve). More concretely, we want a security de?nition such that, if σ is a secure realization of an ideal subroutine σI , and a protocol P using σI , written as P+σI , is a secure realization of PI (the ideal

7 functionality of P), P+σ is also a secure realization of PI . Throughout the paper, we denote the associated ideal functionality of a protocol by adding a subscript I, and we denote a protocol P calling a subprotocol σ as P+σ (this last expression stretches the meaning of P a little bit to refer to the module of P calling σ). The second ingredient is a universal composability theorem stating how a complex protocol can be built out of secure components. It is simply a recipe on how to securely perform basic composition recursively. The simpli?cations in analyzing the composable security of QKD Our goal is to analyze the unconditional security of QKD by using the quantum universal composability results in [19, 20]. The setting for QKD is simpler than that considered in [19, 20] in two important aspects. First, we are only concerned with unconditional security. Second, in QKD, Alice and Bob are known to be honest, and Eve is known to be adversarial, and there is no unpredicted corruption of any party. The formal corruption rules are not used in our derivation of a composable security de?nition for QKD. The following simpli?ed model is su?cient for our derivation of a universal composable security de?nition for QKD. The simpli?ed model We ?rst describe the model for quantum protocols and other concepts involved in the quantum composable security de?nition. We base our discussion on the (acyclic) quantum circuit model (see, for example, [33, 34]), with an important extension [20] (see also the endnotes [35]). Throughout the paper, we only consider circuits in the extended model. 1. Structure of a protocol A (cryptographic) protocol P can be viewed as a quantum circuit in the extended model [20, 35], consisting of inputs, outputs, a set of registers, and some partially ordered operations. A protocol may consist of a number of subprotocols and parties. Each subprotocol consists of smaller units called “unit-roles,” within each the operations are considered “local.” For example, the operations and registers of each party in each subprotocol form a unit-role. Communications between unit-roles within a subprotocol represent internal communications; those between unit-roles in di?erent subprotocols represent input/output of data to the subprotocols. A channel is modeled by an ordered pair of operations by the sender and receiver on a shared register. The channel available to perform each communication determines its security features. 2. The game: security in terms of indistinguishability from the ideal functionality Let PI denote the ideal functionality of P. Intuitively, P is secure (in a sense de?ned by PI ) if P and PI behave similarly under any adversarial attack. “Similarity” between P and PI is modeled by a game between an environment E and a simulator S. These are sets of registers and operations to be de?ned, and they are sometimes personi?ed in our discussion. In general, P and PI have very di?erent internal structures and are very distinguishable, and the simulator S is added to PI to make an extended ideal protocol PI +S that is less distinguishable from P. E consists of the adversaries that act against P and an application protocol that calls P as a subprotocol. At the beginning of the game, P or PI +S are picked at random. E will call and act against the chosen protocol, and will output a bit Γ at the end of the game. The similarity between P and PI +S (or the lack of it) is captured in the statistical di?erence in the output bit Γ. 3. Valid E: The application and adversarial strategy of E are ?rst chosen (the same whether it is interacting with P or PI +S). E has to obey quantum mechanics, but is otherwise unlimited in computation power. If P is chosen in the game, E can (i) control the input/output of P, (ii) attack insecure internal communication as allowed by the channel type, (iii) direct the adversarial parties to interact with the honest parties in P. E+P has to be an acyclic circuit in the extended model [20, 35]. 4. Valid PI and S: If PI +S is chosen in the game, E (i) controls the input/output of PI as before. However, the interaction given by (ii) and (iii) above will now occur between E and S instead. (S is impersonating or simulating P.) The strategy of S can depend on the strategy of E. PI should have the same input/output structure as P, but is otherwise arbitrary. (Of course, the security de?nition is only useful if PI carries the security features we want to prove for P.) In particular, PI may be de?ned with internal channels and

8 adversaries di?erent from those of P. S can (ii′ ) attack insecure internal communication of PI and (iii′ ) direct the adversarial parties to interact with the honest parties in PI . Thus, PI exchanges information with S, and this can modi?ed the security features of PI . To E, S acts like part of PI , “padding” it to look like P, while to PI , S acts like part of E. It is amusing to think of S as making a “man-in-the-middle” attack between E and PI . Finally, E+PI +S has to be an acyclic circuit in the extended circuit model [20, 35]. See Figure 2 for a summary of the game and the rules. E

E EΓ

T

T

E

T

T 'ii ′E

′

E EΓ

i

c'

ii,iii

i

c

ii,iii

c

P

PI

iii

S

Figure 2: The game de?ning the composable security de?nition. The curved region in E represents the adversaries against P, and the curved region in S represents the adversaries against PI . We label the types of interactions as described in the text. With a slight abuse of language, the symbols P and PI +S are also used to denote the respective events of their being chosen at the beginning of the game. We can now state the universal composable security de?nition. De?nition 1: P is said to ?-securely realizes PI (shorthand P ?-s.r. PI ) if ?E ?S s.t. Pr(Γ=0|P) ? Pr(Γ=0|PI +S) ≤ ? . (6)

We call ? in Eq. (6) the distinguishability-advantage between P and PI . This security de?nition (in the model described) is useful because security of basic composition follows “by de?nition” [19, 20]. We have the following simple version of a universal composability theorem. Theorem 1: Suppose a protocol P calls a subroutine σ. If σ ?σ -s.r. σI and P+σI eP -s.r. PI , then P+σ ?-s.r. PI for ? ≤ ?P +?σ .

Theorem 1 can be generalized to any arbitrary protocol with a proper modular structure. An example of an improper modular structure is one with a security deadlock, in which the securities of two components are interdependent. Proper modular structures can be characterized as follows. Let P+σ1 +σ2 + · · · be any arbitrary protocol using a number of subprotocols. This can be represented by a 1-level tree, with P being the parent and σ1,2,··· the children. Each of σ1,2,··· may use other subprotocols, and the corresponding node will be replaced by an appropriate 1-level subtree. This is done recursively, until the highest-level subprotocols (the leaves) call no other subprotocols. These are the primitives. It was proved in [20] that more general modular structures, represented by an acyclic directed graph, can be transformed to a tree. The following composability theorem relates the security of a protocol P to the security of all the components in the tree. Theorem 2: Let P be a protocol and TP the associated tree. Let M be the protocol corresponding to any node in TP with subprotocols Ni . Suppose ?M, M+N1I +N2I , · · · , ?M -s.r. MI . If M ?M ≤ ?, then P ?-s.r. P I , where the sum is over all nodes in TP .

Theorem 2 is obtained by recursive use of theorem 1 and the triangle inequality. The idea is to replace subprotocols one-by-one by their ideal functionalities at the highest level, and proceed recursively to lower levels toward the root. The distinguishability-advantage between P and PI is upper bounded by the sum of all the individual distinguishability-advantages between pairs of protocols before and after each replacement. See Figure 4 for an example of TP that describes repeated QKD.

9 Note that the composable security de?nition for QKD derived in the simpli?ed setting will remain applicable in the general setting considered in [19, 20]. However, when applying Theorem 2 to analyze the security of an application using QKD, one should use a setting appropriate for that particular application. In the next section, we analyze QKD in the composability framework. This is part of our main result, and an example to illustrate the composability framework.

4

Universal composable security de?nition of QKD

We ?rst describe a general QKD scheme in the composability framework. Then, we tailor an ideal functionality for KD that resembles QKD. Finally, we express the universal composable security de?nition of QKD as a simple distinguishability criteria.

4.1

QKD in the game de?ning security

Our discussion relies on the existence of authentication schemes that are universal composable in the quantum setting. Furthermore, the authentication scheme should use a key much shorter than the message to be authentication (so that QKD indeed expands a key). For example, the scheme in [28] satis?es such conditions (composability is proved in [36]). Let α denote any such authentication scheme and let αI denote ideal authentication. Let κ+α denote QKD using authentication scheme α and let κI denote an ideal KD protocol to be de?ned. By theorem 1, we can focus on the security of κ+αI , i.e., QKD using perfectly authenticated classical channels. The initial key requirement is embedded in the subroutine αI . In this case, QKD has no input. It outputs some bitstrings kA , kB of certain length m to Alice and Bob, with m = 0 if and only if QKD is aborted. (We can assume that m is publicly announced, and consider m as an output of QKD.) Eve’s view (including both quantum and classical data) is given by the state ρE,kA ,kB . We now turn to the game de?ning the composable security de?nition of QKD. Eve is an adversary that is part of the environment E. Following the discussion in Sec. 3, E will ?x an arbitrary strategy. Since there is no input to QKD, the optimal application in E is simply to receive the output keys from κ+αI or κI . E will also consist of the action of Eve and other circuits that compute Γ. A schematic diagram is given in Figure 3.

E m kA kB Eve ρEk k

E E

E Γ m kk

f 2 i

TT

i

f 1 iii

T

AB

f 3 i i

AB

TT

Eve ρ|m| ?

E E

Γ

f 1 iii

′ A B′ ?m

T

A B

TT '

iii′ f 2

?

κ+αI

κI

S

Figure 3: The game de?ning the composable security de?nition of QKD, with our choice of ideal KD and simulator. An ordering of the interactions is given in circles. We also label the types of interactions (see rules 3 and 4 in Sec. 3) explicitly. Upon an input m, the checkered box generates a perfect key of length m to Alice and Bob. If E is interacting with κ+αI , E will: (i) receive the output bitstrings kA , kB , and m = |kA | = |kB |, and (iii) obtain ρE,kA ,kB which depends on Eve’s strategy and kA , kB . Altogether, E will be in possession of the state ρqkd =

kA ,kB

Pr(kA , kB ) |kA , kB kA , kB | ? ρE,kA ,kB

(7)

10 in which ρE,kA ,kB and kA , kB can be correlated. We have omitted an explicit register for m, because the information is redundant given kA , kB . See Figure 3 for a schematic diagram for QKD, and how it interacts with the environment.

4.2 Ideal KD and the simulator

Our ideal KD functionality κI has to model both the possibility to generate a perfect key, and the possibility for Eve to cause QKD to be aborted. Besides Alice and Bob, κI has a box that accepts a value m from an adversary “Devil” and outputs a perfect m-bit key K to Alice and Bob (m = 0 means abort). When κI is run, Devil sends m to the box, which sends K to Alice and Bob. This formulation of κI satis?es the security conditions Eqs. (3) and (5) perfectly (?1 , ?2 = 0). See Figure 3 for a schematic diagram.

We now de?ne the ideal functionality for QKD. In general, when formulating an ideal functionality, one need not be concerned with how the functionality is realized. What is important is to impose the essential security features while mimicking the analyzed protocol from the point of view of E.

Consider the following simulator S. S runs a “fake QKD” with fake Alice′ and Bob′ . They interact with Eve (in E) and run veri?cation procedure as in QKD. A value m is announced for the fake QKD, but the fake output keys are unused and kept secret in S. The Devil in S then sends m to the box in κI , which generates a perfect m-bit key string k to Alice and Bob in κI , who forward their outputs to E. Let ρm = ?

kA ,kB :|kA |=|kB |=m

Pr(kA , kB |M =m) ρE,kA ,kB .

(8)

Then, at the end of the game, E will be in possession of the state ρideal =

k

? Pr(M =|k|) 2?|k| |k, k k, k| ? ρ|k| .

(9)

See Figure 3 for a schematic diagram for how κI + S interacts with E.

4.3 Universal composable security de?nition and simple privacy condition

Recall that at the beginning of the game, one of κ and κI +S is chosen at random to interact with E. The distinguishability-advantage is upper bounded by the trace distance of the two possible ?nal states of E right before Γ is computed, Pr(Γ=0 | κ) ? Pr(Γ=0 | κI +S) ≤

1 2 1 2

≤

ρqkd ? ρideal ρqkd ? ρqi1

1

1

(10)

1 2

+

ρqi1 ? ρqi2

1

+

1 2

ρqi2 ? ρideal

1

, (11)

where ρqi1 and ρqi2 are hybrid, intermediate, states between ρqkd and ρideal de?ned as ρqi1 =

k

Pr(M =|k|) 2?|k| |k, k k, k| ? ρE,k,k , ? Pr(M =|k|) 2?|k| |k, k k, k| ? ρ|k| , ρE,k,k .

k:|k|=m

(12) with (13) (14)

ρqi2 =

k

ρm = ?

1 2m

The sum of the ?rst and the last terms in Eq. (11) can be bounded by ?1 in the equality-and-uniformity

11 condition (Eq. (3) in Sec. 2) as follows. Using Eqs. (7) and (12),

ρqkd ? ρqi1

1

=

kA =kB

Pr(kA ,kB ) |kA ,kB kA ,kB | ? ρE,kA ,kB +

k

Pr(k,k)?Pr(|k|)2?|k| |k,k k,k| ? ρE,k,k

1

≤ ?1 .

Using Eqs. (9) and (13), ρqi2 ? ρideal

1

≤

Pr(M =m)

m

? ρ ?m ? ρm

1

≤ ?1

(m) (m) ? ? where we have used ρm = kA ,kB pqkd (kA , kB ) ρE,kA ,kB , and the kA ,kB pideal (kA , kB ) ρE,kA ,kB , ρm = equality-and-uniformity condition Eq. (3) for the last inequality. The remaining term in the composable security condition Eq. (11) is given by

1 ρqi1 ? ρqi2 2

1

= ≤

1 2 1 2

k

? Pr(M =|k|) 2?|k| |k,k k,k| ? ρ|k| ? ρE,k,k Pr(M =|k|) 2?|k| ρ ?|k| ? ρE,k,k

1

1

,

(15)

k

which can be interpreted as a new privacy condition. We have thus compartmentalized the composable security de?nition for QKD, Eq. (10) or Eq. (11), into two parts: the original equality-and-uniformity condition Eq. (3) and a new privacy condition Eq. (15), which we loosely call a “composable privacy condition” for QKD. Once Eq. (15) is bounded by some ?? , 2 QKD using ideal authentication κ+αI ?κ -securely realizes the ideal KD κI , if ?1 + ?? ≤ ?κ . Following 2 Theorems 1 and 2, one can use the key “as if it were perfect.” Proving such a bound on Eq. (15) is relatively straightforward, as compared to a direct proof of the security of using a slightly imperfect key from QKD (without the composability theorem). In the following section, we prove several bounds for Eq. (15). First, we show that for any QKD scheme satisfying the usual privacy condition Eq. (5), Eq. (15) can be bounded as well, albeit with a potentially large but manageable degradation. Second, we prove a tighter bound on Eq. (15) assuming a privacy condition in terms of Eve’s Holevo information on the key. Finally, we propose a new, tight, su?cient condition for bounding Eq. (10) (the full composable security condition) (bypassing Eq. (5) and automatically incorporating all of equality, uniformity, and privacy) based on the singlet-?delity considered in most existing security proofs for QKD. As an application, we obtain sharp upper bounds for Eq. (10) for existing QKD schemes.

5 5.1

Universal composability of QKD Usual privacy condition implies composable privacy condition

Bound 1:

We ?rst mention a loose upper bound for ρqi1 ? ρqi2 1 . It is upper bounded by: Pr(M =m)

m k:|k|=m

? 2?m |k,k k,k| ? ρ|k| ? ρE,k,k

1

and according to Lemma 1 of [16], each trace distance is upper-bounded by (2m +1)2 (we use the shorthand m for M = m in the mutual information). Thus

2(ln 2) I(KE :K|m)

12 ρqi1 ? ρqi2 ≤ Pr(M =m)(2m +1)2

m

1

2(ln 2) I(KE :K|m) Pr(M =m) I(KE :K|m)

m

≤ (2max(m) +1)2 2(ln 2) ≤ (2 ≤ (2

max(m)

+1) +1)

2

2(ln 2) √ 2(ln 2) ?2

m

Pr(M =m) I(KE :K|m)

1 2

max(m)

2

where the second last line is obtained by the Cauchy-Schwarz inequality. Typically, max(m) is a small fraction of n, the security parameter such as the number of qubits communicated. Recall that ?2 ∈ V, the set of exponentially decaying functions of n. With a limit on the key rate m/n (based on how fast ?2 vanishes), ?κ ∈ V also. We now derive a slightly better bound.

Bound 2: The second bound of Eq. (15) requires two lemmas. The Shannon distinguishability [37] of two quantum states ?0 and ?1 , SD(?0 , ?1 ), is de?ned as the accessible information on C obtained by measuring a specimen of ?C , where C is a coin toss (see [37]).

m

Proof of Lemma 1: De?ne random variables C, X1 , X2 and Y as: 1. C is a coin toss. 2. X1 = x with probability qx , 3. If C = 0, X2 = x, else X2 = x′ with probability qx′ . 4. Y is the outcome of measuring ?X2 .

Lemma 1: Let Iacc be the accessible information of an ensemble {qx , ?x }2 of ?nite dimensional states i=1 (i.e. Iacc is the maximum information obtained on X by measuring a single specimen of ?X , where Pr(X = x) = qx ). Let ? = x qx ?x . Then, ?x, qx SD(?x , ?) ≤ Iacc .

These random variables are de?ned so that for each x, I(Y : C|X1 = x) is the information gained on whether a randomly drawn state is ?x or ? by measuring the state. Also, Pr(X2 = x) = qx and ?X2 is simply a draw from the initial ensemble. Note that Y depends on the measurement. For the measurement attaining SD(?x , ?), I(Y : C|X1 ) =

l

ql I(Y : C| X1 =l) ≥ qx SD(?x , ?) ,

(16)

whereas for any measurement, I(Y : C|X1 ) = I(Y : X1 C) ? I(Y : X1 )

≤ I(Y : X1 C) ≤ I(Y : X2 ) ≤ Iacc

(17)

where the three inequalities are respectively due to the Chain rule, the fact X1 C → X2 → Y is a Markov chain, and the optimality of Iacc . We also use the following relation between the trace distance and the Shannon distinguishability, readily obtained from Eq. (47) and Fig. 1 of [37]. Lemma 2: ??0 , ?1 , ?0 ? ?1

1

≤ 2 SD(?0 , ?1 ).

m

Proof of bound 2: 1 and 2 imply

For each key length m, de?ne Fm to be the ensemble {2?m , ρE,k,k }|k|=m. Lemmas ρ|k| ? ρE,k,k ?

1

≤ 2 2 +1

I(KE : K| m)

13 and we can bound Eq. (15) using the above, the Cauchy-Schwarz inequality, and the usual privacy condition Eq. (5): ρqi1 ? ρqi2

1

≤ ≤

Pr(M =|k|) 2?|k|

k

ρ|k| ? ρE,k,k ?

1

Pr(M =m) 2 2 +1

m

m

I(KE : K| m)

1 2

≤ 2max(m)/2+1 ≤ 2

max(m)/2+1 √

Pr(M =m) I(KE : K| m)

m

?2

Once again, the key length is a fraction of n, and if appropriate limits on the key rate are imposed (depending on ?2 ), the above still vanishes exponentially with n.

5.2

Small Holevo information implies composable privacy

Suppose, instead of the usual privacy condition Eq. (5) in terms of the accessible information, we have ? Privacy: ??′ ∈ V s.t. 2

m

Pr(M =m) × χ(Fm ) ≤ ?′ 2

(18)

where χ is the Holevo information [38], and Fm = {2?m , ρE,k,k }|k|=m is as de?ned before. Eq. (18) is more stringent than Eq. (5) since the Holevo information is an upper bound for the accessible information. It was proved in [39] that the Holevo information for an ensemble is the average of the relative entropies S(· ·) of the states in the ensemble to the average state. Applying this fact to Fm , χ(Fm ) = 1 2m ? S(ρE,k,k ρm ) .

k:|k|=m

Furthermore, the relative entropy is related to the trace distance [40], ? ρE,k,k ? ρm Thus Eq. (15) can be bound as ρqi1 ? ρqi2

1 2 1

? ≤ 2 (ln 2) S(ρE,k,k ρm ) .

(19)

≤ ≤ ≤ ≤ ≤

Pr(M =|k|) 2?|k|

k

ρ|k| ? ρE,k,k ? ρ|k| ? ρE,k,k ?

?|k|

1 2 1 1 2

Pr(M =|k|) 2?|k|

k

2 (ln 2)

k

Pr(M =|k|) 2

? S(ρE,k,k ρ|k| )

1 2

1 2

2 (ln 2)

m

Pr(M =m) χ(Fm )

2 (ln 2) ?′ 2

14 which does not have an overhead exponential in the length of the key generated.

5.3

A new su?cient condition for composable security

We can easily analyze the composable security of any QKD scheme that has a security proof based on entanglement puri?cation protocol. All existing QKD schemes have such security proofs. The ?nal keys KA , KB are outcomes of Alice and Bob’s measurements on a shared state ρm for some m, and ρm is supposed to AB AB 1 be Φ?m in the absence of eavesdropping. Here, m is again the key length and Φ = 2 (|00 +|11 )( 00|+ 11|). The usual privacy condition Eq. (5) is obtained by showing the following. ? High ?delity: ??′′ ∈ V s.t. 2

m

Pr(m) 1 ? F (ρm , Φ?m ) ≤ ?′′ AB 2

(20)

(See Sec. 1 for the de?nition of F .) The above turns out to provide a sharp bound on Eq. (15), as shown below. Let ρm be the state held by Alice, Bob, and Eve right before the ?nal measurements of Alice and Bob. ABE m m We only need to consider m > 0. Let |ψ1 be a puri?cation of ρm on systems A, B, E and X. |ψ1 is ABE m . By Ulhmann’s Theorem [41], there exists a puri?cation |ψ m over systems A, also a puri?cation of ρAB 2 B, E and X such that

m m F (|ψ1 , |ψ2 ) = F (ρm , Φ?m ) . AB m m By construction of |ψ1 and |ψ2 , measuring A and B and tracing X results in ρm and ρm respectively. qkd ideal But measuring and tracing can only increase the ?delity of two states. Thus

F (ρm , ρm ) > F (ρm , Φ?m ) . AB qkd ideal Finally, we use the fact to obtain ρm ? ρm ideal qkd ρm ? ρm ideal qkd

2 1 1

≤ 2 1 ? F (ρm , ρm ) qkd ideal .

≤ 4 1 ? F (ρm , Φ?m ) AB

Putting all these together, we can bound Eq. (10) as 1 ρqkd ? ρideal 2

1

≤ ≤

1 2

m

Pr(M =m) ρm ? ρm ideal qkd

1 1

1 2 2 (21) Pr(M =m) ρm ? ρm ≤ ?′′ . ideal 1 qkd 2 2 m Eq. (20) is a good new su?cient condition for composable security, being part of the standard QKD proof and a tight bound on Eq. (10) simultaneously. It also implies both equality-and-uniformity and privacy (unlike a bound on Holevo information or mutual information which only implies the composable privacy condition).

15

6 Discussions and applications

We have motivated this work with a discussion of the potential gap between the desired security of using a key generated by QKD and the security promised by the privacy condition Eq. (5) used in the study of “unconditional security” of QKD. Then, we apply the universal composability theorem to obtain a new security condition that will guarantee the security of using a key generated from QKD. We propose a new privacy condition Eq. (15) that is composable, and useful su?cient conditions such as Eq. (18) or Eq. (20). Most interesting of all, we show that a bound on the singlet-?delity Eq. (20) directly implies the composable security condition Eq. (10). These are our main contributions (in the context of cryptography). We also provide a proof that the existing privacy condition Eq. (5) does imply Eq. (15), albeit with a degradation factor in the security exponential in the key size. This ensures the security of using a key generated from any QKD scheme that has been proved secure, provided the key rate is limited accordingly. Despite the existence of such connections, we emphasize that future research should address Eq. (10), Eq. (15), Eq. (18), or Eq. (20) directly. We also provide a sharp bound on Eq. (15) based on Holevo’s information Eq. (18) or singlet-?delity Eq. (20). We are glad to ?nd that the existing security proofs for QKD imply sharp bounds on Eq. (10), when bypassing the usual privacy condition Eq. (5). Outside the context of cryptography, these connections between various privacy conditions can be useful for the study of correlations in quantum systems. It is open whether the degradation of the security (that is exponential in the generated key size) when going from Eq. (5) to Eq. (15) is necessary. However, it is a tempting conjecture, as suggested by the pathologies of the accessible information exhibited recently [16, 17]. As a ?nal application, we analyze the security of repeating QKD t times, without assuming the availability of an authenticated classical channel. (Note that t is a ?xed parameter that does not grow with the problem size.) Each run of QKD κ calls a composable authentication scheme α as a subroutine, and each run of α requires a composably secure key, which is provided by the previous round of κ (as a subroutine to α). Call the t rounds of QKD our protocol P. The associated tree for P, and the ideal realization PI are given in the far left and right of Figure 4. P κ ?α

c c c c c

κ

?κ

α c · · · κ κ α

α c · · · κ

c c c

c

κ

?α

α c · · · κ

c c

c

κ

?κ

α

α c κ0

κ c αI

α c κI

κ c αI κI

c

α c · · ·

c

c

κ

?κ

κI κI

α c · · ·

c c

c

κI

PI

····

· · · c κI κI

c

c

Figure 4: Associated tree for t rounds of κ in the left. κ0 represents some initially shared key. The arrows point from parents to children. Each tree to the right is obtained by replacing one node by its ideal functionality. The distinguishability-advantage of each pair of consecutive schemes is marked between their trees near the roots. Authentication is omitted in the ideal functionality PI . If κ+αI ?κ -s.r. κI (as in Eq. (10)) and if α+κI ?α -s.r. αI , P t(?κ +?α )-s.r. PI . In other words, each additional around of QKD degrades the security parameter by an additive constant (?κ + ?α ). The same result can be obtained by using Theorem 2, or conversely, this simple exercise illustrates the idea behind Theorem 2.

16

7 Acknowledgements

We thank Charles Bennett, Daniel Gottesman, Aram Harrow, and John Smolin for interesting discussions on the security concerns of using a key obtained from QKD. We also thank Dominique Unruh and J¨rn o M¨ller-Quade for interesting discussions on their alternative framework of composability. u Part of this work was completed while MH and JO were visiting the MSRI program on quantum information, Berkeley, 2002. MH is supported by EU grants RESQ (IST-2001-37559), QUPRODIS (IST-200138877). DL acknowledges the support from the Richard Tolman Foundation and the Croucher Foundation. DL and DM acknowledge support from the US NSF under grant no. EIA-0086038. JO is supported by EU grant PROSECCO (IST-2001-39227) and a grant from the Cambridge-MIT Institute.

A

Notations

We gather most of the notations used in the paper, roughly in the order of ?rst appearance: ? KD: key distribution ? QKD: quantum key distribution ? Alice and Bob: two honest parties trying to establish a common key ? Eve: an active adversary ? A, B, E: subscripts labelling objects related to Alice, Bob, and Eve respectively A, B, E: labels of their respective quantum systems ? Capitalized letters denote random variables and the corresponding uncapitalized letters denote particular outcomes ? KA , kA , KB , kB : output keys for Alice and Bob ? K, k: k := kA when kA = kB ? M , m: publicly announced key length at the end of QKD. M = 0 i? QKD is aborted. ? KE , kE : classical data possibly extracted by Eve at the end of QKD by measuring her quantum state ? Pr(·): probability of the event “·” ? log: logarithm in base 2 ? H(X), I(X : Y ), I(X : Y |Z), and I(X : Y |Z=z) for random variables X, Y , Z: H(X) := ? x Pr(x) log Pr(x) is the entropy of X I(X : Y ) := H(X) + H(Y ) ? H(XY ) is the mutual information between X and Y I(X : Y |Z=z) is the mutual information between X and Y conditioned on Z = z I(X : Y |Z) := z Pr(z)I(X : Y |Z=z) is the conditional mutual information ? ρ: generic symbol for a density matrix ? |· , |· ·|: |· denotes a vector in a Hilbert space, with label “·”. |· ·| denotes the “outer-product” of |· and ·| or the projector onto the subspace spanned by |· . ? Tr(·): the trace

17 ? TrH1 (·): the partial trace over the system H1 . Let ρ12 be the density matrix for a joint state on H1 and H2 . TrH1 (ρ12 ) is the state after H1 is discarded. ? ·

1:

the trace distance, which can be taken as the sum of the singular values

? F : the ?delity. For two states ρ1 , ρ2 in H, F (ρ1 , ρ2 ) = max|ψ1 ,|ψ2 | ψ1 |ψ2 |2 where |ψ1,2 ∈ H?H′ are “puri?cations” of ρ1,2 (i.e., TrH′ |ψ1,2 ψ1,2 | = ρ1,2 ), and ·|· is the inner product. ? ρE,kA ,kB : Eve’s view (both quantum and classical data) when the key outputs to Alice and Bob are kA , kB . ? n: security parameter such as the number of qubits communicated in QKD ? p qkd : the distribution of KA , KB generated in QKD conditioned on |KA | = |KB | = m, i.e., p qkd (kA , kB ) = Pr(KA = kA , KB = kB |M = m).

(m) (m) (m)

? pideal : the distribution over two m-bit strings de?ned as p ideal (l, l′ ) = 0 if l = l′ , p ideal (l, l) = 2?m . ? V: the set of exponentially decaying functions of n ? σ, P, σI , PI : σ and P are generic labels for protocols, with σ possibly used as a subroutine. The symbol of a protocol with a subscript I denotes the ideal functionality of the protocol. P+σ: a protocol P calling a subroutine σ. ? E, S: the environment and the simulator. These are sets of registers and operations and they are sometimes personi?ed in our discussion. ? Γ: output bit of E ? ?-s.r. : P ?-s.r. PI is a shorthand for P ?-securely realizes PI (see mathematical de?nition in Eq. (6)). ? is called the distinguishability-advantage between P and PI . ? TP : the associated tree for a protocol P ? α, αI : universal composable authentication with negligible key requirement and its ideal functionality ? κ+α, κ+αI , κI : QKD using authentication α, QKD using ideal authentication αI , and ideal KD de?ned in Sec. 4.2 ? Devil: an adversary that determines the key length m generated by κI ? ρqkd : state possessed by E after interacting with κ+αI , see Eq. (7) ? ρideal : state possessed by E after interacting with κI , see Eq. (9) ? ρqi1 , ρqi2 : hybrid, intermediate, states between ρqkd and ρideal , see Eqs. (12) and (13) ? ? ρm : Eve’s state when M = m, averaged over KA , KB . See Eq. (8) ? ? ρm : uniform average of ρE,k,k for |k| = m. See Eq. (14) ? Ensemble: a distribution {qx }x of quantum states ?x denoted by {qx , ?x }x ? Iacc : accessible information of an ensemble {qx , ?x }x , i.e., the maximum mutual information between X and outcome Y obtained from measuring a specimen ?x ? SD(?0 , ?1 ): Shannon distinguishability of ?0 and ?1 , de?ned as Iacc of the uniform distribution of {?0 , ?1 }.

(m)

(m)

18 ? Fm : the ensemble {2?m , ρE,k,k }|k|=m ? χ({qx , ?x }): Holevo information of an ensemble, given by S( Tr(· log(·)) is the von Neumann entropy

x q x ?x )

?

x qx S(?x )

where S(·) =

? ρm : state on which measurements by Alice and Bob output KA , KB in QKD-security-proofs based AB on entanglement puri?cation ? Φ: a perfect EPR pair 1 (|00 + |11 )( 00| + 11|) 2 ? Singlet ?delity: F (ρm , Φ?m ). Note that “singlet” usually refers to a state that is only unitarily AB equivalent to Φ, but we borrow the term in this paper.

References

[1] C. H. Bennett and G. Brassard. Quantum cryptography: Public key distribution and coin tossing. In Proceedings of IEEE International Conference on Computers, Systems and Signal Processing, pages 175–179, New York, 1984. IEEE. Bangalore, India, December 1984. [2] A. K. Ekert. Quantum cryptography based on Bell’s theorem. Phys. Rev. Lett., 67(6):661–663, 1991. [3] C. H. Bennett. Quantum cryptography using any two nonorthogonal states. 68(21):3121–3124, 1992. Phys. Rev. Lett.,

[4] C. H. Bennett, G. Brassard, R. Jozsa, D. Mayers, A. Peres, B. Schumacher, and W. K. Wootters. Reduction of quantum entropy by reversible extraction of classical information. Journal of Modern Optics, 41(12):2307–2314, 1994. [5] D. Mayers. Quantum key distribution and string oblivious transfer in noisy channels. In Advances in Cryptography–Proceedings of Crypto’96, pages 343–357, New York, 1996. Springer-Verlag. [6] D. Mayers. Unconditional security in quantum cryptography. J. Assoc. Comp. Mach, 48:351, 2001. arXive e-print quant-ph/9802025. [7] H.-K. Lo and H. F. Chau. Unconditional security of quantum key distribution over arbitrarily long distances. Science, 283:2050–2056, 1999. arXive e-print quant-ph/9803006. [8] E. Biham, M. Boyer, P. O. Boykin, T. Mor, and V. Roychowdhury. A proof of the security of quantum key distribution. In Proceedings of the 32nd Annual ACM Symposium on Theory of Computing (STOC), pages 715–724, New York, 2000. ACM. arXive e-print quant-ph/9912053. [9] P. W. Shor and J. Preskill. Simple proof of security of the bb84 quantum key distribution protocol. Phys. Rev. Lett., 85:441–444, 2000. arXive e-print quant-ph/0003004. [10] K. Tamaki, M. Koashi, and N. Imoto. Unconditionally secure key distribution based on two nonorthogonal states. Phys. Rev. Lett., 90:167904, 2003. arXive e-print quant-ph/0212162. [11] G. Gottesman and H.-K. Lo. Proof of security of quantum key distribution with two-way classical communications. IEEE Transactions on Information Theory, 49(2):457–475, 2003. arXive e-print quant-ph/0105121. [12] C. Bennett and J. Smolin ?rst suggested the key degradation problem to one of us, and A. Harrow has obtained partial results.

19 [13] A. Ambainis, M. Mosca, A. Tapp, and R. de Wolf. Private quantum channels. In IEEE Symposium on Foundations of Computer Science (FOCS), pages 547–553, 2000. arXive e-print quant-ph/0003101. [14] P. O. Boykin and V. Roychowdhury. Optimal encryption of quantum bits. arXive e-print quantph/0003059. [15] A. Peres and W. K. Wootters. Optimal detection of quantum information. Phys. Rev. Lett., 66:1119– 1122, 1991. [16] D. P. DiVincenzo, M. Horodecki, D. Leung, J. Smolin, and B. M. Terhal. Locking classical correlation in quantum states. Phys. Rev. Lett., 92:067902, 2004. arXive e-print quant-ph/0303088. [17] P. Hayden, D. Leung, P. W. Shor, and A. Winter. Randomizing quantum state: constructions and applications. arXive e-print quant-ph/0307104. [18] R. Canetti. Universal composable security: A new paradigm for cryptographic protocols. In Proceedings of the 42nd IEEE Symposium on Foundations of Computer Science (FOCS), pages 136–145. IEEE, 2001. [19] M. Ben-Or and D. Mayers. Composability theorem. Part I of presentation by D. Mayers, QIP 2003, MSRI, Berkeley. See http://www.msri.org/publi-cations/ln/msri/2002/qip/mayers/1/index.html . [20] M. Ben-Or and D. Mayers. Composing quantum and classical protocols. arXive e-print quantph/0409062. [21] M. Backes, B. P?tzmann, and M. Waidner. A general composition theorem for secure reactive systems. In Proceedings of the 1st Theory of Cryptography Conference (TCC), pages 336–354, Springer-Verlag, 2004. [22] D. Unruh. Relating formal security for classical and quantum protocols. Presentation at the Special week on Quantum crytography, Isaac Newton Institute for Mathematical Sciecnes, September 2004. Available at http://www.unruh.de/DniQ/publications. [23] M. Ben-Or, M. Horodecki, D. Leung, D. Mayers, and J. Oppenheim. Composability of QKD. Part II of presentation by D. Mayers, QIP 2003, MSRI, Berkeley. See http://www.msri.org/publications/ln/msri/2002/qip/mayers/1/index.html . [24] M. Ben-Or, M. Horodecki, D. Leung, D. Mayers, and J. Oppenheim. Composability of quantum proocols – applications to quantum key distribution and quantum authentication. Part II of presentation by D. Leung, QIP 2004, IQC, University of Waterloo. See http://www.iqc.ca/conferences/qip/presentations/leung-.pdf. [25] R. Renner and R. K¨nig. Universally composable privacy ampli?cation against quantum adversaries. o arXive e-print quant-ph/0403133. [26] M. Christandl, R. Renner, and A. Ekert. A generic security proof for quantum key distribution. arXive e-print quant-ph/0402131. [27] M. A. Nielsen and I. L. Chuang. Quantum computation and quantum information. Cambridge University Press, Cambridge, U.K., 2000. [28] M. N. Wegman and J. L. Carter. New hash functions and their use in authentication and set equality. Journal of Computer and System Sciences, 22:265–279, 1981. [29] D. Bruss. Optimal eavesdropping in quantum cryptography with six states. Phys. Rev. Lett., 81:3018– 3021, 1998.

20 [30] H. Bechmann-Pasquinucci and N. Gisin. Incoherent and coherent eavesdropping in the 6-state protocol of quantum cryptography. arXive e-print quant-ph/9807041. [31] D. Deutsch, A. Ekert, R. Jozsa, C. Macchiavello, S. Popescu, and A. Sanpera. Quantum privacy ampli?cation and the security of quantum cryptography over noisy channels. Phys. Rev. Lett., 77:2818, 1996. arXive e-print quant-ph/9604039. [32] T. M. Cover and J. A. Thomas. Elements of Information Theory. John Wiley and Sons, New York, 1991. [33] A. C. Yao. Quantum circuit complexity. Proc. of the 34th Ann. IEEE Symp. on Foundations of Computer Science, pages 352–361, 1993. [34] D. Aharonov, A. Kitaev, and N. Nisan. Quantum circuits with mixed states. arXive e-print quantph/9806029. [35] An acyclic circuit is a partially ordered set of gates. However, associating the circuit with constraints on the timing of the adversarial attack is a delicate issue. Suppose the circuit contains conditional gates controlled by random public classical registers. The gates on the target may or may not be applied depending on the values of the control registers – and when the gates are not applied, the associated time-constraints of the adversarial attack disappear. In the extension to the usual acyclic circuit model, we consider all possible values of the control registers and the resulting sets of nontrivial partially ordered operations, and the corresponding constraints on the adversarial attack. [36] P. Hayden, D. Leung, and D. Mayers. On the composability of quantum message authentication and key recycling. [37] C. A. Fuchs and J. van de Graaf. Cryptographic distinguishability measures for quantum mechanical states. IEEE Transactions on Information Theory, 45(4):1216–1227, 1999. arXive e-print quantph/9712042. [38] A. S. Holevo. Information-theoretical aspects of quantum measurement. Problemy Peredachi Informatsii, 9(2):31–42, 1973. [A. S. Kholevo, Problems of Information Transmission, vol. 9, pp. 110-118 (1973)]. [39] B. Schumacher and M. Westmoreland. Relative entropy in quantum information theory. arXive e-print quant-ph/0004045. [40] M. Ohya and D. Petz. Quantum entropy and its use. Texts and monographs in physics. SpringerVerlag, Berlin, 1993. [41] A. Uhlmann. The transition probability in the state space of a *-algebra,. Reports on Mathematical Physics, 9:273–279, 1976.

赞助商链接

- Security of EPR-based Quantum Key Distribution using three bases
- Semi-device-independent security of one-way quantum key distribution
- Security of EPR-based Quantum Key Distribution
- Quantum key distribution
- Quantum key distribution network with wavelength addressing
- The Security of Practical Quantum Key Distribution
- A Proof of the Security of Quantum Key Distribution
- SECURITY AND IMPLEMENTATION OF DIFFERENTIAL PHASE SHIFT QUANTUM KEY DISTRIBUTION SYSTEMS
- On the Key-Uncertainty of Quantum Ciphers and the Computational Security of One-way Quantum
- Improvement of quantum key distribution protocols
- A simple proof of the unconditional security of quantum key distribution
- Outline of the SECOQC Quantum-Key-Distribution Network in Vienna
- Security of the Quantum Key Distribution with Blind Polarization Bases against Impersonatio
- Security of Quantum Key Distribution with Realistic Devices
- Summary of the Theory Component of Quantum Key Distribution and Quantum Cryptography A Quan

更多相关文章：
更多相关标签：