See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235460475

Positive maps that are not completely positive

Article in Physical review A, Atomic, molecular, and optical physics · July 2000
DOI: 10.1103/PhysRevA.62.024302 · Source: arXiv

CITATIONS READS
19 484

1 author:

Sixia Yu
University of Science and Technology of China
130 PUBLICATIONS 2,744 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sixia Yu on 22 May 2014.

The user has requested enhancement of the downloaded file.

 Positive Maps Which Are Not Completely Positive
Sixia Yu
The Institute of Theoretical Physics, Academia Sinica, Beijing 100080, P.R. China
(Dated: February 1, 2008)
The concept of the half density matrix is proposed. It unifies the quantum states which are described by
density matrices and physical processes which are described by completely positive maps. With the help of
the half-density-matrix representation of Hermitian linear map, we show that every positive map which is not
completely positive is a difference of two completely positive maps. A necessary and sufficient condition for a
positive map which is not completely positive is also presented, which is illustrated by some examples.

PACS numbers: 03.67.-a, 03.65.Bz, 03.65.Fd, 03.65.Ca

arXiv:quant-ph/0001053v1 14 Jan 2000

Entanglement has become one of the central concept in ρ can be expressed as

quantum mechanics, specially in quantum information. A
quantum state of a bipartite system is entangled if it cannot be X
L−1
T †
ρ = Trρ − σmn ρ σmn , (1)
prepared locally or it cannot be expressed as a convex combi-
m,n=0
nation of direct product states of two subsystems. This kind √
of state is also called inseparable. Though easily defined, it is where σmn = (|mihn| − |nihm|)/ 2. We see immediately
very hard to recognize the inseparability of a mixed state of a that the transposition is a difference of two completely pos-
bipartite system. itive maps. And this statement will be proved to hold true
for all positive maps which are not completely positive, which
An operational-friendly criterion of separability was pro- will be also characterized by a necessary and sufficient condi-
posed by Peres [1]. This criterion is based on the observa- tion in this Letter.
tion that the partial transposition of a separable density matrix For this purpose we shall first develop an extremely useful
remains positive semidefinite. That the partial transposition tool — half density matrix that unifies the description of the
of a density matrix is not positive semidefinite infers the in- quantum states and physical processes. And then we derive
separability of the density matrix. This provides a necessary a half-density-matrix representation of an arbitrary Hermitian
condition for the separability. There exist entangled states linear map from which our main results are obtained. Along
with positive partial transposition, which exhibit bound en- with the introduction of the concept of half density matrix, its
tanglement [2]. Examples of such kind were first provided in relations to the ensembles and the purifications of mixed states
Ref. [3] and then constructed in Ref. [4] systematically with are clarified and its applications in the field of quantum infor-
the help of unextendible product basis. mation such as quantum teleportation [8] are also presented.
Normally, quantum states, pure or mixed, are described
Later on, by noticing that the transposition is a positive map
by density matrices, positive semidefinite operators (whose
(to be described later in details), a necessary and sufficient
eigenvalues are all nonnegative) on the Hilbert space of the
condition of the separability was proposed in Ref. [5]: A bi-
system. Because of its property of positive semidefinite the
partite state is separable iff it is still positive semidefinite un-
density matrix ρ can always be written as ρ = T T † where
der all positive maps acting on a subsystem. In other words, a
matrix T is called here as the half density matrix (HDM) for a
density matrix of a bipartite system is inseparable iff there ex-
quantum state.
ists a positive map acting on a subsystem such that the image
Obviously, the half density matrix for a given density ma-
of the density matrix is not positive semidefinite. Hence the
trix is not unique. For example T U and T are correspond-
inseparability can be recognized by positive maps which are
ing to the same mixed state ρ = T T † whenever U is uni-
not completely positive.
tary. Generally, the half density matrix T for a mixed state
Completely positive maps, which are able to describe the ρ of an s-level system is an s × L rectangular matrix with
most general physical process [6], are better understood than L ≥ r = Rank(ρ), i.e., a linear map from an L-dimensional
positive maps which are not completely positive. Positive Hilbert space HL to an s-dimensional Hilbert space Hs . The
maps from Hilbert space H2 (two-dimensional) to H2 or H3 rank r of the density matrix equals to the rank of the half den-
are all decomposable [7], which are characterized by transpo- sity matrix T and r = 1 for pure state.
s−1
sition and completely positive maps only. As a result in the Under an orthonormal and complete bases {|mi}m=0 and
L−1
cases of H2 × H2 and H2 × H3 the transposition criterion is {|ni}n=0 of Hilbert spaces Hs and HL , a typical half den-
also a sufficient condition for separability [5]. Therefore fur- sity matrix of dimension s × L can be constructed as Te =
ther understandings of positive maps which are not completely V † (∆s , 0s×(L−s) ) , where ∆s is a diagonal s × s matrix
positive will facilitate the recognition and classification of the formed by all the square roots of the eigenvalues of ρ (the
inseparable mixed states. singular numbers of Te ) and V is an s × s unitary matrix diag-
onalizing the density matrix ρ. Obviously we have ρ = Te Te† .
As a direct calculation will show, under an orthonormal and As a direct result of the singular number decomposition of an
L−1
complete basis {|ni}n=0 the transposition of an L × L matrix arbitrary matrix [9] we have the following
 2

Lemma: Given a density matrix ρ of an s-level system, an |φi12 and |k; li23 = Tkl |Φs i23 denote an orthonormal com-
s × L matrix T is a half density matrix for ρ, i.e., ρ = T T † , plete basis for systems 2 and 3 with HDMs Tkl satisfying
P
if and only if there exists an L × L unitary matrix U such that †
TrTkl Tk†′ l′ = δkk′ δll′ for orthogonality and kl Tkl OTkl =
T = Te U . TrO for completeness. We then have expansion
When written explicitly in the established basis, the rela-
tion ρ = Te Te† results in exactly an ensemble formed by all X
s−1
∗
the eigenvectors V † |mi of the mixed state, which is referred |φi12 |ψi3 = Tφ Tkl |ψi1 |k; li23 . (3)
to as eigen-ensemble here. In this way every half density ma- k,l=0
trix T of a mixed state ρ corresponds to an ensemble of the
mixed state. The above Lemma tells us that every ensemble This describes exactly a quantum teleportation of an unknown
of a given mixed state is related to the eigen-ensemble by a quantum state |ψi from system 3 to system 1 when both Tφ
unitary matrix which has been proved by other means [10]. and Tkl are unitary or state |φi12 and basis |k; li23 are maxi-
Therefore the half density matrix of a density matrix is phys- mally entangled states [13].
ically equivalent to an ensemble of the corresponding mixed The mixed state ρsL of an (s × L) bipartite system can also
state. be equivalently and conveniently characterized by HDMs of
Every mixed state ρ admits a purification [11], a pure state pure bipartite states. Let {|φi i, pi }R
i=1 be an ensembles of ρsL
|φi of a bipartite system including this system as a subsys- we have
tem such that ρ = Tr2 |φihφ|. Under the established basis, a
general pure state in Hs ⊗ HL is X
R X
R
ρsL = pi |φi ihφi | = Ai ML A†i , (4)
X X
s−1 L−1
i=1 i=1
|φi = Cmn |mi1 |ni2 := T |ΦL i. (2)
m=0 n=0 where we have denoted Ai as the half density matrix of the
√ √
PL−1 pure state pi |φi i, i.e., pi |φi i = Ai |ΦL i. Obviously
P
Here pure state |ΦL i = n=0 |ni1 |ni2 lives in Hilbert space HDMs defined by Ãi = j Uij Aj characterize the same den-
HL ⊗ HL and T is a linear map from HL to Hs acting on the sity matrix whenever U is a unitary matrix. And from the
first L-dimensional Hilbert space HL . Under the given bases Lemma we know that given a density matrix this is the only
linear map T is represented by an s × L rectangular matrix freedom that the half density matrices can have.
with matrix elements given by hm|T |ni = Cmn . When the
The density matrix expressed in the form as in Eq.(4) can be
pure state |φi is normalized we have Tr(T † T ) = 1. Alterna-
easily manipulated by local operations. For example the den-
tively, we also have |φi = T T |Φs i with state |Φs i defined in
sity matrix under operation Us ⊗ UL∗ is transformed to density
Hs ⊗Hs similar to state |ΦL i. The linear map T T : Hs 7→ HL
acts on the second Hs and it is represented by the transposi- matrix specified by half density matrices Us Ai UL† . The tilde
tion of T under the established basis. operation ρ 7→ ρ̃ introduced in Ref. [14] to obtain explicitly
Tracing out the second system we obtain a reduced density the entanglement of formation of two-qubit is simply an anti-
matrix of the first subsystem ρs = T T † and similarly ρL = linear transformation Ãi = TrA†i − A†i .
T T T ∗ for the second subsystem. That is to say T is the HDM In the discussions above we have defined the half density
for the reduced half density matrix ρs = TrL |φihφ| of the first matrices for the states of a single system, for pure bipartite
subsystem and its transposition T T for ρL = Trs |φihφ|. Thus states, and for mixed bipartite states. The physical processes
a one-to-one correspondence between a normalized pure state can also be characterized by half density matrices. A general
|φi of a bipartite system, a purification, and a linear map T physical process which can include unitary evolutions, trac-
satisfying Tr(T † T ) = 1, a half density matrix, is established. ing out one system, and general measurements is described
Therefore a half density matrix T is also equivalent to a pu- by trace-preserving completely positive maps [6, 11], which
rification of the mixed state. The linear map T is also referred is a special kind of Hermitian linear map.
to as the half density matrix of a bipartite pure state, which is A Hermitian linear map sends linearly Hermitian opera-
unique by definition. If s = L the polar decomposition of T tors to Hermitian operators that may live in different Hilbert
will result in the useful Schmit-decomposition. spaces. Let L denote a general Hermitian linear map from
The pure bipartite state is separable iff the rank of its half Hilbert space HL to Hs . Because the map L is linear the
density matrix is one. For a pure product state |vis |wiL the map L ⊗ IL is also a Hermitian linear map from HL ⊗ HL
half density matrix is |vihw∗ | where |w∗ i is the index state to Hs ⊗ HL , where IL denotes the identity map on HL . Re-
of state |wi defined by |w∗ i = hw|ΦL i [11]. For later use we calling that the mirror operator ML = |ΦL ihΦL | is defined on
define a mirror operator ML = |ΦL ihΦL | in the Hilbert space HL ⊗ HL , its image
HL ⊗ HL , which has the property hw∗ |ML |w∗ i = |wihw|.
The partial transposition of the mirror operator X = MLT1 is HsL = L ⊗ IL (ML ) (5)
in fact the exchanging (or swapping) operator introduced by
Werner [12] (denoted as V there). is therefore a Hermitian operator in Hs ⊗ HL . Let |ψi+ i =
As an application, we consider a state |φi12 |ψi3 of a tri- Ai |ΦL i (i ≤ i+ ) denote the eigenvectors corresponding to
partite system with all three subsystems 1,2 and 3 being s- the positive eigenvalues of HsL and |ψi− i = Bi |ΦL i (i ≤ i− )
level systems. Let Tφ denote the HDM of the bipartite state to negative eigenvalues of HsL , where i± is the number of the
 3

positive/negative eigenvalues of HsL . We then have map S : HL 7→ Hs and an arbitrary positive integer k the
induced map S ⊗ Ik from HL ⊗ Hk to Hs ⊗ Hk is positive.
X
i+
X
i−
However it is enough to check whether the image HsL =
HsL = Ai ML A†i − Bi ML Bi† , (6) S ⊗ IL (ML ) of the mirror operator ML is positive semidefi-
i=1 i=1 nite or not. If it is positive semidefinite, then the negative part
in the HDM representation Eq.(7) disappears, which yields
in which the norms of the eigenvectors |ψi± i have been taken exactly the operator-sum representation of a CP map [11]
to be the absolute values of corresponding eigenvalues. Be-
cause the eigenvectors corresponding to different eigenvalues X
i+
are orthonormal we have Tr(Ai Bj† ) = 0 for all i and j. In this S(ρ) = Ai ρA†i . (11)
sense two families of half density matrices {Ai } and {Bi } are i=1
orthogonal to each other. P
For a pure state Pw = |wihw| in the Hilbert space HL we If the trace is preserved, we have further i A†i Ai = 1.
have Pw = hw∗ |ML |w∗ i where |w∗ i is the index state of |wi. Therefore the operator-sum representation of a CP map can
As a result we have L(Pw ) = hw∗ |HsL |w∗ i. Taking into also be referred to as a half-density-matrix representation. Es-
account of the linearity of the Hermitian map L, we finally pecially, if HsL equals to the identity matrix Is ⊗ IL , the cor-
obtain responding CP map is simply the trace operation ST (ρ) =
Is Trρ.
X
i+
X
i−
A positive map which is not completely positive (non-CP)
L(H) = Ai HA†i − Bi HBi† , (7) is nonetheless a Hermitian map so that it has a HDM represen-
i=1 i=1 tation as Eq.(7), from which we obtain S = SA − SB where
where H is an arbitrary Hermitian matrix in HL . This is called two CP maps SA,B are represented by HDMs {Ai } and {Bi }
the half-density-matrix representation of a Hermitian linear respectively. Two CP maps SA,B are said to be orthogonal if
map. As one result we have their HDMs are orthogonal to each other, i.e., Tr(Ai Bj† ) = 0
for all i, j. We see that HsL can not be positive semidefinite.
hΦs |Is ⊗ L(ΣsL )|Φs i = Tr(HsL ΣTsL ) (8) Conversely, if the Hermitian matrix HsL has at least one
negative eigenvalue then it determines a non-CP positive map.
for an arbitrary Hermitian matrix ΣsL in Hs ⊗ HL . As an- Let |ψi denote an eigenvector corresponding to one of the neg-
other consequence, a one-to-one correspondence between the ative eigenvalues of HsL and Pψ = |ψihψ|. From identity (8)
Hermitian maps L : HL 7→ Hs and Hermitian matrix HsL we see immediately that Is ⊗ S(PψT ) is not positive semidef-
(an observable) in Hs ⊗ HL can be established inite, i.e., map S is not completely positive. We note that the
eigenspace corresponding to the negative eigenvalues of HsL
L(H) = TrL (HsL H T ) (9) contains no product state because of positivity. To summarize,
we have the following
in addition to Eq.(5). Theorem: Every positive map which is not completely pos-
The HDM representation of Hermitian linear map is not itive is a difference of two orthogonal completely positive
unique. Suppose two integers M ≥ i+ and N ≥ i− and maps; A Hermitian linear map S : HL 7→ Hs is positive
let SU (M, N ) denote the pseudo-unitary group formed by but not completely positive if and only if for all pure product
(M + N ) × (M + N ) matrices satisfying SηS † = η where state Ps ⊗ QL in HL ⊗ Hs we have Tr(HsL Ps ⊗ QL ) ≥ 0
η = IM ⊕ (−IN ) and IM(N ) is the M × M (N × N ) identity while HsL = S ⊗ IL (ML ) is not positive semidefinite.
matrix. If we define a family of HDMs {Ti }M+N
i=1 as Ti = Ai This theorem provides an obvious way to construct a non-
(1 ≤ i ≤ i+ ), Ti = Bi (M + 1 ≤ i ≤ M + i− ) and Ti = 0 CP positive map form HL to Hs . First, we choose a proper
otherwise and take an arbitrary element S of SU (M, N ), a Hermitian matrix HsL in Hs ⊗ HL satisfying the conditions
P
new family of HDMs {T̃i }M+Ni=1 defined by T̃i = j Sij Tj specified in the above theorem. Then a non-CP positive map
represents the same Hermitian linear map S : HL 7→ Hs is determined by S(ρL ) = TrL (HsL ρTL ).
As the first example we consider the the exchanging opera-
X
M X
N tor defined in HL ⊗ HL by X = MLT1 or explicitly
L(H) = T̃i H T̃i† − T̃j H T̃j† . (10)
i=1 j=1 X
L−1
X= |m, nihn, m|. (12)
A positive map is a special Hermitian linear map which m,n=0
maps any positive semidefinite operator to a positive semidef-
inite operator. A Hermitian linear map S : HL 7→ Hs is The exchanging operator X has two eigenvalues ±1 and
T
positive if and only if Tr(S(QL )Ps ) = Tr(HsL Ps ⊗ QL ) ≥ 0 σmn |ΦL i (m > n) are the eigenvectors corresponding to
for all pure product state Ps ⊗QL where HsL = S ⊗IL (ML ). eigenvalue −1. Therefore X is not positive semidefinite
In the following S is always a positive map. and for any pure product states |ppi = |vi|wi we have
A completely positive (CP) map is a positive map which hpp|X|ppi = |hv|wi|2 ≥ 0 as specified by the above theorem.
keeps its positivity when the system it acts on is embedded as In fact the resulting non-CP positive map on HL is exactly the
a subsystem in an arbitrary larger system. That is, for a CP transposition ρT = Tr2 (XρT ). By writing X in its diagonal
 4

form we obtain ρT = ST (ρ) − Sσ (ρ), where the CP map Sσ and 1 ≤ d ≤ sL. Matrix Hǫ is not positive semidefinite since
is represented by HDMs {σmn } and ST is the trace operation. TrHǫ ρ̃ = −ǫdTr(ρ0 ρ̃) < 0 and for an arbitrary pure product
As the second example we consider a Hermitian matrix in state hα|hβ|Hǫ |αi|βi ≥ 0. If we choose ρ0 = Is ⊗ IL /sL
HL ⊗ HL defined by HR = IL ⊗ IL − ML . It is not positive then a non-CP positive map is defined by
semidefinite because hΦL |HR |ΦL i < 0 and for every product
states |ppi we have hpp|HR |ppi = 1 − |hv|w∗ i|2 ≥ 0. Ac-
cordingly, a non-CP positive map is defined on HL as Λ(ρ) = X
S
Trρ − ρ, which provides the reduction criterion [15, 16]: Ev- Sǫ (ρ) = Ti ρTi† − ǫTrρ, (13)
ery inseparable state in HL ⊗ HL which loses its positivity i=1
under map IL ⊗ Λ is distillable and in the distillation pro-
cedure provided in Ref. [15] the HDM of pure bipartite state
serves as the filtering operation. Because Λ(ρ) = Sσ (ρT ), where Ti = |αi ihβi∗ | is the half density matrix of the product
the reduction map Λ is a decomposable positive map, which base |αi i|βi i. In Ref. [17] ρ0 is taken as a maximally entangle
is generally of form Sd (ρ) = S1 (ρ) + S2 (ρT ) with S1,2 being state and d = min(s, L). Positive map Sǫ is indecomposable
two CP maps. because Is ⊗ Sǫ (ρ̃) is not positive semidefinite while Is ⊗
The last example makes use of an unextendible product ba- Sd (ρ̃) is positive semidefinite for any decomposable map.
sis [4], a set of orthonormal product basis {|αi i|βi i}Si=1 of In conclusion, the concept of the half density matrix was
Hs ⊗ HL where S < sL and there is no other pure prod-
studied and its applications to the quantum information are
uct state
P that is orthogonal to this set of basis. If we denote discussed in some detail. Based on the half-density-matrix
P = |αi ihαi | ⊗ |βi ihβi | then ρ̃ = (1 − P )/(sL − S) repre-
representation of a Hermitian linear map, we proved that every
sents an inseparable states with positive partial transposition.
positive map which is not completely positive is a difference
If we define
of two completely positive maps. A necessary and sufficient
ǫ = min hα|hβ|P |αi|βi condition for a non-CP positive map is given, which provides
|αi|βi a way of constructing such kind of maps. Some examples are
also presented. Further applications of the half density matrix
it can be sure that 0 < ǫ ≤ S/sL [17]. Denoting ρ0 as a in the quantum information and other fields can be expected
normalized density matrix in Hs ⊗ HL which has the property and the understandings of positive maps provided here may be
Tr(ρ0 ρ̃) > 0, we define a Hermitian matrix as Hǫ = P − ǫdρ0 helpful the recognition of the inseparable quantum states and
where to the quantification of the entanglement [18].
1 The author gratefully acknowledges the financial support of
= max hα|hβ|ρ0 |αi|βi K. C. Wong Education Foundation, Hong Kong.
d |αi|βi

[1] A. Peres, Phys. Rev. Lett. 77, 1413 (1996). University Press, 1985)
[2] M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. [10] L. P. Hughston, R. Jozsa and W. K. Wootters, Phys. Lett. A 183,
80, 5239 (1998). 14 (1983).
[3] P. Horodecki, Phys. Lett. A 232, 333 (1997). [11] B. Schumacher, Phys. Rev. A 54, 2614 (1996).
[4] C. H. Bennett, D. P. DiVincenzo, T. Mor, P. W. Shor, J. A. [12] R. F. Werner, Phys. Rev. A 40, 4277 (1989).
Smolin, and B. M. Terhal, Phys. Rev. Lett. 82, 5385 (1999). [13] S. X. Yu and C. P. Sun, Phys. Rev. A (accepted).
[5] M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Lett. A [14] W. K. Wootters, Phys. Rev. Lett. 80, 2245 (1998).
223, 1 (1996). [15] M. Horodecki and P. Horodecki, Phys. Rev. A 59, 4026 (1999).
[6] K. Kraus, States, Effects ad Operations: Fundamental Notions [16] N. J. Cerf, C. Adami, and R. M. Gingrich, Phys. Rev. A 60, 898
of Quantum Theory, (Wiley, New York 1991). (1999).
[7] S. L. Woronowicz, Rep. Math. Phys. 10, 165 (1976). [17] B. M. Terhal, LANL e-print quant-ph/9810091.
[8] C. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. [18] V. Vedral, M. B. Plenio, M. A. Rippin, and P. L. Knight Phys.
K. Wooters, Phys. Rev. Lett. 70, 1895 (1993). Rev. Lett. 78, 2275 (1997).
[9] R. A. Horn and C. R. Johnson, Matrix Ananlysis, (Cambridge

View publication stats