Fidelity of mixed states of two qubits
Frank Verstraete 1,2 and Henri Verschelde 1
1
Department of Mathematical Physics and Astronomy, Ghent University, Ghent, Belgium
2
Department of Electrical Engineering (SISTA), KU Leuven, Leuven, Belgium
共Received 8 April 2002; published 12 August 2002兲We consider a single copy of a mixed state of two qubits and show how its fidelity or maximal singlet fraction is related to the entanglement measures concurrence and negativity. We characterize the extreme points of a convex set of states with constant fidelity, and use this to prove tight lower and upper bounds on the fidelity for a given amount of entanglement.
DOI: 10.1103/PhysRevA.66.022307 PACS number
共s兲: 03.67.⫺aThe concept of fidelity
关1兴, also called the maximal singletfraction
关2兴, is of central importance in the field of quantuminformation theory. It is defined as the maximal overlap of the state with a maximally entangled
共ME兲 stateF
共兲⫽ max兩典⫽ME具兩兩典
.
共1兲An explicit value for the fidelity has been derived by Horo- decki
关3兴. If one considers the real 3⫻3 matrix R˜⫽Tr(i丢j
) with
兵i, i
⫽1, . . . ,3其the Pauli matrices, then
F
共兲⫽1
⫹1
⫹2
⫺sgn关det共R˜兲兴3
4 ,
with
兵i其the ordered singular values of R ˜ and sgn关det(R˜)兴 the sign of the determinant of R ˜ .
The concept of fidelity appears in the context of entangle- ment distillation
关1,4兴 where it quantifies how close a state isto a maximally entangled one, and in the context of telepor- tation
关5兴, where it quantifies the quality of the teleportationthat can be achieved with the given state. Due to the linearity and the convexity of the definition
共1兲, this measure has verynice properties that make it also possible to derive upper bounds for the entanglement of distillation
关6兴.Despite the importance of the concept of fidelity, no rig- orous comparison appears to have been made before between the value of the fidelity on one side and entanglement mea- sures on the other side. This paper aims at filling this gap and gives explicit tight lower and upper bounds on the fidelity for given concurrence
关7兴 and negativity 关8兴.First we explicitly derive the possible range of values of the fidelity as a function of its concurrence
关7兴 or entangle-ment of formation. Next we show that the states that mini- mize
共maximize兲 the fidelity for given values of the entangle-ment of formation are also extremal for given negativity
关8兴.Following
关9,10兴, we use the following definition of negativ-ity:
N
共兲⫽max关0,⫺2min
共⌫兲兴,with
min the minimal eigenvalue of the partial transpose of
denoted as
⌫.
Theorem 1. Given a mixed state of two qubits
with negativity equal to N and concurrence equal to C, then its fidelity F is bounded above by
F
⭐1
⫹N2
⭐1
⫹C2 .
Moreover, the first inequality becomes an equality if and only if N
⫽C, and this condition is equivalent to the condi-tion that the eigenvector corresponding to the negative eigen- value of the partial transpose of
is maximally entangled.
Proof. The fidelity of a state
is given by max
UA
,U
B苸SU(2)Tr
关共UA丢U
B兲兩典具兩共UA丢U
B兲†
兴⫽
1 2 max
UA
,U
BTr 冋冉 1 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 冊
⫻共UA丢
U
B*
兲†
⌫共UA丢U
B*
兲册 ,
with
兩典⫽(兩00典⫹兩11典)/
冑2. An upper bound is readily ob- tained by extending the maximization over all unitaries in- stead of all local unitaries, and it follows that F
⭐Tr(兩⌫兩)⫽(1⫹N)/2. Equality is achieved if and only if the eigenvec-
tor of
T⌫corresponding to the negative eigenvalue is maxi- mally entangled. As shown in
关10兴, this condition is exactlyequivalent to the condition for N to reach its upper bound C,
which ends the proof.
䊏Note that the upper bound is achieved for all pure states.
A more delicate and technical reasoning is needed to ob- tain a tight lower bound on the fidelity. We will need the following lemma.
Lemma 1. Consider the density operator
and the real 3
⫻3 matrix R˜ with coefficients R˜i j⫽Tr(i丢j) with 1
⭐i, j⭐3. Then
is a convex sum
共i.e., mixture兲 of rank 2density operators all having exactly the same coefficients
R ˜
i j.
Proof. Consider the real 4
⫻4 matrix R with coefficientsR
␣⫽Tr(␣丢), parametrized as
R
⫽冉 y y y 1 1 2 3 x 1 x R ˜ 2 x 3 冊 .
If
is full rank, then a small perturbation on the values
兵x
i其,
兵y
i其will still yield a full rank density operator. Consider a perturbation on x 1
⬘⫽x1
⫹⑀and the corresponding
⬘. As the set of density operators is compact, there will exist a lower bound b
l⬍0 and an upper bound bu⬎0 such that⬘is positive if and only if b
l⬍⑀⬍bu. Call
bl,
buthe rank 3 density operator obtained when
⑀⫽bland
⑀⫽burespec- tively. It is easy to see that
⫽(bubl⫹blbu)/(b
l⫹bu), such that it is proven that a rank 4 density operator can always be written as a convex sum of two rank 3 density operators with the same corresponding R ˜ .
Consider now
of rank 3 and its associated ‘‘square root’’
⫽XX
† with X a 4
⫻3 matrix. A small perturbation of theform
⬘⫽⫹⑀XQX † , with Q an arbitrary Hermitian 3
⫻3matrix Q
⫽兺i9
⫽1q
iG
iand G
igenerators of U
共3兲, will stillyield a state of rank 3. Moreover, there always exists a non- trivial Q such that R ˜ is left unchanged by this perturbation.
This is indeed the case if the following set of equations is satisfied:
兺i
q
iTr
关GiX †
共␣丢兲X兴⫽0for (
␣,
)
⫽(0,0) and␣,
⭓1. It can easily be verified thatthis set of ten equations only contains at most eight indepen- dent ones irrespective of the 4
⫻3 matrix X, and as Q hasnine independent parameters there always exists at least one nontrivial solution to this set of homogeneous equations. A similar reasoning as in the full rank case then implies that one can always tune
⑀such that
can be written as a convex sum of two rank 2 density operators with the same R ˜ , which
concludes the proof.
䊏This lemma is interesting if one wants to maximize a convex measure of a density operator
共such as the entropy oran entanglement monotone
兲 under the constraint that the fi-delity is fixed: indeed, the fidelity is only a function of R ˜ , and by the previous lemma we immediately know that states with maximal entropy for given fidelity will have rank 2.
Note that exactly the same reasoning applies when one wants to maximize a convex measure under the constraint that the Clauser-Horne-Shimony-Holt
共CHSH兲 Bell violation is fixed 关8兴, as this CHSH Bell violation is also solely a function ofR ˜ . This is in exact correspondence with the results derived in
关8兴, where it was proven that the states exhibiting the mini-mal amount of Bell violation for given entanglement of for- mation are rank 2.
We are now ready to prove a tight lower bound on the fidelity.
Theorem 2. Given a mixed state of two qubits
with concurrence equal to C, then a tight lower bound for its fidelity F is given by
F
⭓max冉 1
⫹C4 ,C 冊 .
Proof. A direct consequence of lemma 1 is that to find states with minimal fidelity for given concurrence
共i.e., maxi-mal concurrence for given fidelity
兲, it is sufficient to look atstates of rank 2. Consider therefore a rank 2 state
and associated with it the real 4
⫻4 matrix R with coefficientsR
␣⫽Tr␣丢. As shown in
关8,10–12兴, if R is multipliedright and left by proper orthochronous Lorentz transforma- tions leaving the (0,0) element equal to 1, then a new state is obtained with the same concurrence. Moreover, the fidelity of a state
is variationally defined as
F
共兲⫽min
OA
,O
B苸SO(3)Tr 冋 M 冉 1 0 O 0
A冊 R 冉 1 0 O 0
BT冊册 ,
with M
⫽diag(1,⫺1,⫺1,⫺1) (M is the representation of thesinglet in the R picture
兲. The minimal fidelity for given con-currence can therefore be obtained by minimizing the fol- lowing constrained cost function over all proper orthochro- nous Lorentz transformations L 1 ,L 2 :
K
⫽Tr共ML1 RL 2
T兲⫺ Tr冋 L 1 RL 2
T冉 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 冊 册 .
Note that
is a Lagrange constraint. Without loss of gener-ality we can assume that the lower 3
⫻3 block R˜ of R isdiagonal and of the form R ˜
⫽diag(⫺兩s1
兩,⫺兩s2
兩,⫺s3 ) with
兩s1
兩⭓兩s2
兩⭓兩s3
兩, as this is precisely the form needed formaximizing the fidelity over all local unitary operations. The cost function K can be differentiated over L 1 ,L 2 by introduc- ing the generators of the Lorentz group
共see, e.g., 关8兴兲, andthis immediately yields the optimality conditions (
⫽0,M R M
⫽RT) or (
⫽2, R⫽RT). Note, however, that the above argument breaks down in the case that
兩s2
兩⫽⫺s3 . Indeed, the fidelity cannot be differentiated in this case as for example a perturbation of s 3 of the form s 3
⬘⫽s3
⫹⑀always leads to a perturbation of the fidelity F
⬘⫽F⫹兩⑀兩. In this casethe conditions x 2
⫽y2 , x 3
⫽y3 or x 2
⫽⫺y2 , x 3
⫽⫺y3 van- ish, and if also
兩s1
兩⫽兩s2
兩⫽⫺s3 there are no optimality con- ditions on
兵x
i, y
i其left.
Let us first treat the case with R symmetric and s 1
⭓s2
⭓兩s
3
兩:R
⫽冉 x x x 1 1 2 3
⫺sx 0 0 1 1
⫺sx 0 0 2 2
⫺sx 0 0 3 3 冊 .
The condition that
corresponding to this state is rank 2 implies that all 3
⫻3 minors ofare equal to zero. Due to the conditions s 1
⭓s2
⭓兩s3
兩, it can easily be shown that astate of rank 2
共and not of rank 1兲 is obtained if and only ifx 1
⫽0⫽x2 and x 3
⫽⫾冑(1
⫺s1 )(1
⫺s2 ) and 1
⫺s1
⫺s2
⫹s3
⫽0. In this case the concurrence is equal to C⫽s
2 and the fidelity is given by F
⫽(s1
⫹s2 )/2, and the constraints be- come 1
⭓s1
⭓s2
⭓(1⫺s1 )/2 which implies that C
⭓1/3. Theminimal fidelity for given concurrence occurs when s 1
⫽s2
and then C
⫽F, which gives the lower bound of the theoremin the case of C
⭓1/3.Let us now consider the case where R
⫽MRTM :
R
⫽冉
⫺x⫺x⫺x1 1 2 3
⫺sx 0 0 1 1
⫺sx 0 0 2 2
⫺sx 0 0 3 3 冊 ,
with again s 1
⭓s2
⭓兩s3
兩. Let us first note that, due to thesymmetry, R has a Lorentz singular value decomposition
关11兴of the form R
⫽L1
⌺M˜ M L 1
TM with
⌺ of the formdiag(
兩0
兩,⫺兩1
兩,⫺兩2
兩,⫺兩3
兩) and M˜ of the form diag(1,1,1,1) or diag(1,
⫺1,⫺1,1) or diag(1,⫺1,1,⫺1) ordiag(1,1,
⫺1,⫺1). It follows that Tr(R)⫽Tr(⌺M˜ ), and due to the ordering of the Lorentz singular values, M ˜ has to be equal to the identity if Tr(R)
⭐0. But Tr(⌺) is just ⫺2Cwith C the concurrence of the state, and Tr(R)
⫽2⫺4F withF the fidelity of the state. Therefore it holds that F
⫽(1⫹C)/2 if Tr(R)⭐0, which corresponds to the upper bound
of the fidelity. Therefore only the case where Tr(R)
⬎0 hasto be considered for finding lower bounds of the fidelity. The condition that the state be rank 2
共and not rank 1兲 immedi-ately yields x 3
⫽0, s1
⫹s2
⫺s3
⫽1, and s1
⫹s2
⫽x1
2 /(1
⫺s2 )
⫹x
2
2 /(1
⫺s1 ). If we consider only the case with Tr(R)
⬎0, itholds that s 3
⬍0 and the inequality constraints become (1⫺s
1 )/2
⭐s2
⭐(1⫺s1 )
⭐2/3. The concurrence can again becalculated analytically and is given by C
⫽(1⫺s1
⫺s2
⫺s
3 )/2, and it follows that F
⫽(1⫺C)/2. Note that the in-equality constraints limit C to be in the interval C
苸兵0,1/3
其, and so this bound is less stringent than the one stated in the theorem.
Let us now move to the degenerate case where s 1
⬎s2
⫽⫺s
3 :
R
⫽冉 y y y 1 1 2 3
⫺sx 0 0 1 1
⫺sx 0 0 2 2 x s 0 0 3 2 冊 .
As s 1
⬎s2 , optimality requires x 1
⫽⫾y1 . We first treat the case x 1
⫽y1 . Defining
␣⫽x3 / y 3 , a set of necessary and suf- ficient conditions for being rank 2 is given by
0
⫽x1
⫽y1 ,
0
⫽x2
⫹␣y 2 ,
0
⫽␣2
⫺␣1
⫺ss 1
2
⫹1,
0
⫽共x2
2
⫹x3
2
兲⫺␣s 2
共1⫹s1
兲.Under these conditions the concurrence can again be calcu- lated exactly and is given by C
⫽s2 , while the fidelity is given by F
⫽(1⫹s1 )/4. Note that the above set of equations has a solution only if (1
⫺s1 )/2
⭓s2 , implying that C
⭐1/3.The fidelity will now be minimal when s 2
⫽s1 , and then F
⫽(1⫹C)/4 which is the second bound stated in the theorem.
Let us now consider the degenerate case with s 1
⬎s2
⫽⫺s
3 but x 1
⫽⫺y1 . The rank 2 condition implies that s 1
⫹2s
2
⫽1 and x2
⫽⫺y2 and x 3
⫽y3 . Some straightforward algebra leads to the condition
4 1
⫺s1
1
⫹s1
x 1 2
⫹1⫺s1 2
⫺2x2
2
⫺2x3 2
⫽0.Taking into account the constraints, the concurrence is again given by C
⫽s2
⫽(1⫺s1 )/2 and bounded above by 1/3, while the fidelity is equal to F
⫽(1⫹s1 )/4
⫽(1⫺C)/2. Thisbound always exceeds the previously derived bound F
⭓(1⫹C)/4 for C⭐1/3, and is therefore useless.
It only remains to consider the case where s 1
⫽s2
⫽⫺s3 :
R
⫽冉 1 y y y 1 2 3 x
⫺s0 0 1 1 x 0
⫺s0 2 1 x 0 0 s 3 1 冊 .
Defining
␣⫽x1 / y 1 , the rank 2 constraint leads to the follow- ing set of necessary and sufficient conditions:
0
⫽x2
⫺␣y 2 , 0
⫽x3
⫹␣y 3 , 0
⫽s1
␣2
⫹␣共1⫺s1
兲⫹s1 , 0
⫽␣共x1
2
⫹x2 2
⫹x3
2
兲⫹s1
共1⫹s1
兲.The inequality constraint reads s 1
⭐1/3, and the concurrencecan again be calculated exactly and is given by C
⫽s1 . Therefore the fidelity of these states obeys the relation F
⫽(1⫹C)/4 for C⭐1/3, which is the sharp lower bound. 䊏
It might be interesting to note that all rank 2 states mini- mizing the fidelity for given concurrence are quasidistillable
关2,11兴 and have one separable and one entangled eigenvector.More specifically, the states minimizing the fidelity for C
⭐1/3 are, up to local unitaries, of the form
⫽
冉 1⫹C2 0 0 0 1
⫺C⫹冑⫺1 0 4 0
⫺2C⫺3CC 2 2 1
⫺C⫺冑⫺1 0 4 0
⫺2C⫺3CC 2 2 0 0 0 0 冊 ,
and those for C
⭓1/3 of the form⫽
冉 1
⫺C0 0 0
⫺C/2C/2 0 0
⫺C/2 0C/2 0 0 0 0 0 冊 .
Exactly the same states also minimize the fidelity for given negativity. This leads to the following sharp bounds for the fidelity versus negativity:
F
⭓1 4
⫹1
8
共N⫹冑5N 2
⫹4N兲, F⭓冑2N
共N⫹1兲⫺N, F⭐1
⫹N2 .
The first condition applies when N
⭐(冑5
⫺2)/3 and the second when N⭓(冑5
⫺2)/3. A plot of these bounds is given in Fig.1. One observes that the difference between the lower bound and the upper bound in terms of the negativity becomes very small (
⯝⑀2 /16) for large negativity N
⫽1⫺⑀. Moreover, the fidelity is always larger than 1/2 if the negativity exceeds (
冑2
⫺1)/2.
In conclusion, we derived a tight upper bound for the fidelity for given values of the concurrence and fidelity, and we identified all states for which this upper bound is saturated. Next we characterized the extreme points of the convex set of states with given fidelity, and this enabled us to derive tight lower bounds on the fidelity for a given amount of entanglement.
关1兴 C.H. Bennett, D. DiVicenzo, J. Smolin, and W.K. Wootters,
Phys. Rev. A 54, 3824
共1996兲.关2兴 M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. A
60, 1888
共1999兲.关3兴 Piotr Badzig, Michal Horodecki, Pawel Horodecki, and
Ryszard Horodecki, Phys. Rev. A 62, 012311
共2000兲.关4兴 C.H. Bennett, G. Brassard, S. Popescu, B. Schumacher, J.
Smolin, and W.K. Wootters, Phys. Rev. Lett. 76, 722
共1996兲.关5兴 C.H. Bennett, G. Brassard, C. Cre´peau, R. Jozsa, A. Peres, and
W.K. Wootters, Phys. Rev. Lett. 70, 1895
共1993兲.关6兴 E.M. Rains, IEEE Trans. Inf. Theory 47, 2921 共2001兲.
关7兴 W. Wootters, Phys. Rev. Lett. 80, 2245 共1998兲.
FIG. 1. Range of values of the
fidelity for given concurrence and
negativity.
关8兴 G. Vidal and R.F. Werner, Phys. Rev. A 65, 032314 共2002兲.
关9兴 J. Eisert and M. Plenio, J. Mod. Opt. 46, 145 共1999兲; K.
Zyczkowski, P. Horodecki, A. Sanpera, and M. Lewenstein, Phys. Rev. A 58, 883
共1998兲.关10兴 F. Verstraete, K. Audenaert, J. Dehaene, and B. De Moor, J.
Phys. A 34, 10 327
共2001兲.关11兴 F. Verstraete, J. Dehaene, and B. De Moor, Phys. Rev. A 64,
010101
共R兲 共2001兲.关12兴 F. Verstraete, J. Dehaene, and B. De Moor, Phys. Rev. A 65,