a

r
X
i
v
:
0
8
0
2
.
0
2
3
5
v
1


[
q
u
a
n
t
-
p
h
]


3

F
e
b

2
0
0
8
Entanglement-breaking channels in innite
dimensions
A. S. Holevo
Steklov Mathematical Institute, Moscow
Abstract
We give a representation for entanglement-breaking channels in sepa-
rable Hilbert space that generalizes the Kraus decomposition with rank
one operatorsand use it to describe the complementary channels. We
also give necessary and sucient condition of entanglement-breaking for
a general quantum Gaussian channel. Application of this condition to one-
mode channels provides several new cases where the additivity conjecture
holds in full generality.
1 Introduction
In the paper [1] we gave a general integral representation for separable states in
the tensor product of innite dimensional Hilbert spaces and proved the struc-
ture theorem for the quantum communication channels that are entanglement-
breaking, which generalizes the nite-dimensional result of Horodecki, Shor and
Ruskai [2].
In what follows H denotes separable Hilbert space; T(H) the Banach space
of trace-class operators in H, and S(H) the convex subset of all density
operators in H. We shall also call them states for brevity, having in mind
that a density operator uniquely determines a normal state on the algebra
B(H) of all bounded operators in H (see e. g. [3]). Equipped with the trace-
norm distance, S(H) is a complete separable metric space.
A state S(H
1
H
2
) is called separable if it belongs to the convex closure
of the set of all product states in S(H
1
H
2
). It is shown in [1] that separable
states are precisely those which admit a representation
=
_
X

1
(x)
2
(x)(dx), (1)
where (dx) be a Borel probability measure and
j
(x), j = 1, 2, are Borel S(H
j
)-
valued functions on some complete separable metric space A.
A channel with input space H
A
and output space H
B
is bounded linear
completely positive trace-preserving map : T(H
A
) T(H
B
). Channel is
called entanglement-breaking if for arbitrary Hilbert space H
R
and arbitrary
1
state S(H
A
H
R
) the state (Id
R
)() S(H
B
H
R
), where Id
R
is the
identity map in T(H
R
), is separable.
It is shown in [1] that the channel is entanglement-breaking if and only if
there is a complete separable metric space A, a Borel S(H
B
)-valued function

B
(x) and an observable M in H
A
with the outcome set A given by probability
operator-valued measure (POVM) M(dx) (which is a measure on A taking
values in the positive cone of B(H
A
), with M(A) = I) such that
() =
_
X

B
(x)m

(dx), (2)
where m

(S) = TrM(S) for all Borel subsets S B(A). This gives a continual
version of the class of channels introduced in [4] and can be regarded as a
generalization of a result in [2] to innite dimensions.
In nite dimensions entanglement-breaking channels form a large class in
which the famous additivity conjecture for the classical capacity holds as shown
by Shor [5]. Generalization of this property to innite dimensions is by no means
straightforward. First, we dene generalized ensemble as arbitrary probability
distribution on the state space S(H
A
) [6], [7]. The average state of the ensemble
is given by the barycenter

=
_
(d). Let / be an arbitrary subset of
S(H
A
), then the /-constrained capacity of the channel is dened as
C

(, /) = sup
: A
_
H ([]; [

]) (d),
where H(; ) is the quantum relative entropy. In case the output entropy is
nite on / this amounts to
C

(, /) = sup
A
_
H ([])

H

()
_
, (3)
where

() = inf
: =
_
H ([]) (d) (4)
is the convex closure of the output entropy H ([]) . When / =S(H
A
) and
runs through ordinary ensembles given by probability distributions with nite
supports, this reduces to the familiar denition of (unconstrained) capacity
C

().
Then, as shown in [8], for an entanglement-breaking channel
1
and arbitrary
channel
2
the additivity conjecture holds in its strongest form
C

(
1

2
, /
1
/
2
) = C

(
1
, /
1
) +C

(
2
, /
2
), (5)
where
/
1
/
2
= S(H
A1
H
A2
) : Tr
2
/
1
, Tr
1
/
2
.
Moreover, the convex closure is superadditive

H
12
(
12
)

H
1
(
1
) +

H
2
(
2
) (6)
2
for any state
12
.
In applications the constraints of the form
/(H, E) = S(H
A
) : TrH E , (7)
where H is positive selfadjoint operator (typically the energy operator, see [6])
are of the major interest. In this case one shows as in [9], that (5) implies
C

(
n
, /(H I +. . . I H, nE)) = nC

(, /(H, E))
and hence
C(, H, E) = C

(, /(H, E)) = sup

:Tr HE
_
H ([]; [

]) (d) (8)
for any entanglement-breaking channel , where C(, H, E) is the classical ca-
pacity of the channel with the input energy constraint as dened in [6].
The paper has two self-consistent parts. In part I (Sec. 2, 3) we give another
representation for entanglement-breaking channels in separable Hilbert space,
that generalizes the Kraus decomposition with rank one operators. We also
nd complementary channels and remark that coherent information for anti-
degradable channel is always non-positive.
Part II (Sec. 4-6) is devoted to Gaussian entanglement-breaking channels.
We give necessary and sucient condition of entanglement-breaking for a gen-
eral quantum Gaussian channel. Application of this condition to one-mode
channels provides several new cases where the additivity conjecture holds in the
full generality.
2 A representation for entanglement-breaking
channels
Here we further specify the formula (2), by employing the representation for
POVM from [10]. Basing on this specication, we give explicit description of
the Stinespring isometry for the channel and of the complementary channel.
Lemma 1. (Radon-Nikodym theorem for POVM) For every POVM M on
A, there exist a positive -nite measure on A, a dense domain T H, and
a countable family of functions x a
k
(x) such that for almost all x, a
k
(x) are
linear functionals on T, satisfying
_
X

k
[a
k
(x)[)[
2
(dx) = ||
2
; T, (9)
and
[M(S)) =
_
S

k
[a
k
(x)[)[
2
(dx); T. (10)
3
Note that a
k
(x) are in general unbounded functionals dened only on T,
nevertheless we nd it convenient to continue use of the bra-ket notation for
such functionals.
Proof: We follow the proof of Radon-Nikodym theorem for instruments from
[10]. By Naimarks theorem [3], there exist a Hilbert space /, and sharp ob-
servable E given by spectral measure E(S); S B(A) in /, and an isometry
W from H to / such that
M(S) = W

E(S)W, S B(A). (11)

According to von Neumanns spectral theorem, / can be decomposed into the
direct integral of Hilbert spaces
/ =
_
X
H(x)(dx) (12)
with respect to some positive -nite measure , diagonalizing the spectral
measure E:
E(S) =
_
S
(x)(dx), (13)
where (x) H(x) are the components of the vector / in the decomposition
(12). Let us x a measurable eld of orthonormal bases e
k,x
in the direct
integral (12) and denote
k
(x) = e
k,x
[(x)), where the inner product is in
H(x) (note that
k
(x) are dened -almost everywhere).
Let H, then the decomposition of the vector W / reads
W =
_
X

k
(W)
k
e
k,x
(dx),
where _
X

k
[(W)
k
(x)[
2
(dx) = ||
2
; H, (14)
since W is isometric. Since W is linear, we have for -almost all x
(W(

j
))
k
(x) =

j
(W
j
)
k
(x), (15)
where
j
H is a xed system of vectors, and
j
are complex numbers, only
nite number of which are non-zero.
Let us now x an orthonormal basis
j
H and let T = lin
j
be its
linear span. We dene linear functionals a
k
(x) on T by the relation
_
a
k
(x)[

j
_
=

j
(W
j
)
0
k
(x),
where (W
j
)
0
k
is a xed representative of the equivalence class (W
j
)
k
. Then
by (15), for any xed T there is a subset A

A, such that (A A

) = 0
and
a
k
(x)[) = (W)
k
(x), for all k, all x A

. (16)
4
Combining (13) and (16), we obtain (10). The normalization condition (9)
follows from (14) and (16 ).
Theorem 1. For any entanglement-breaking channel there exist a com-
plete separable metric space }, a positive -nite measure on }, a dense
domain T H
A
, a measurable function y a(y), dened for almost all y,
such that a(y) are linear functionals on T, satisfying
_
Y
[a(y)[)[
2
(dy) = ||
2
; T, (17)
and a measurable family of unit vectors y b(y) in H
B
such that
[] =
_
Y
[b(y))b(y)[ [a(y)[)[
2
(dy) for = [)[ with T. (18)
This is innite-dimensional analog of the Kraus decomposition with rank
one operators from [2] with the dierence that it is in general continual and the
Kraus operators V (y) = [b(y))a(y)[ are unbounded and only densely dened.
As shown in [1], there are entanglement-breaking channels for which the usual
Kraus decomposition with bounded rank one operators does not exist.
Proof. By making the spectral decomposition of the density operators
B
(x) =

l
(x) [b
l
(x))b
l
(x)[ in (2) and using lemma 1, we nd
[] =
_
X

k,l

l
(x) [b
l
(x))b
l
(x)[ [a
k
(x)[)[
2
(dx); T. (19)
Let } be the space of triples y = (x, k, l), with naturally dened metric and the
countably nite measure dened by
(S k l) =
_
S

l
(x)(dx).
Dene a(y) = a
k
(x) and b(y) = b
l
(x), then (17) follows from (9) and (18) from
(19).
3 Remarks on complementary channels
In general, the Stinespring representation
[] = Tr
E
V V

holds in the innite dimensional case for arbitrary channel, where H

E
is envi-
ronment space and V : H
A
H
B
H
E
is an isometry. The complementary
channel is dened as

[] = Tr
B
V V

.
5
If there exists a channel T : T(H
B
) T(H
E
) such that

= T , then is
degradable [11], and if = T

for some channel T

: T(H
E
) T(H
B
), then
is anti-degradable [12].
Proposition 1. If is degradable (anti-degradable) channel then I
c
(, )
0 (resp. I
c
(, ) 0) for arbitrary density operator such that H([]) <
, H(

[]) < .
Proof. As noticed in [13], there is a formula which relates the coherent
information and the quantity giving the upper bound for the classical infor-
mation. Namely, for arbitrary pure-state decomposition =

j

j

j
I
c
(, ) =
B

E
,
where

B
= H([])

j
H([
j
])
and similarly for
E
. But
B
=

j

j
H([
j
]; []), whence
I
c
(, ) =

j
_
H([
j
]; []) H(

[
j
];

[])
_
.
The assertion then follows from the monotonicity of the relative entropy and
the denition of (anti-)degradable channel.
As observed in [12], anti-degradable channels have zero quantum capacity
Q(). Proposition 1 provides a short proof of this statement. Let be anti-
degradable, then such is
n
, hence I
c
(,
n
) 0. Then the coding theorem
for the quantum capacity implies
Q() = lim
n
n
1
sup

I
c
(,
n
) = 0.
For the entanglement-breaking channel (18) we introduce the environment
space H
E
= L
2
(), and dene the operator V : T H
B
H
E
by the formula
(V ) (y) = [b(y)) a(y)[) .
Then V is isometric by (17) and hence uniquely extends to the isometry H
A
H
B

H
E
. It is the Sinespring isometry for channel namely
Tr
E
V V

= [], for = [)[ with T

by (18). On the other hand, Tr
B
V V

is the integral operator in H

E
= L
2
()
dened by the kernel

(y
2
, y
1
) = b(y
2
)[b(y
1
)) a(y
1
)[) a(y
2
)[),
which completely describes the complementary channel

.
Example. Let H
A
= L
2
(), T =C(}) L
2
(), then the relation
[] =
_
Y
[b(y))b(y)[ [(y)[
2
(dy); for = [)[ with T
6
denes entanglement-breaking channel. This extends to
[] =
_
Y
[b(y))b(y)[ (y, y)dy,
where (y
2
, y
1
) is the kernel of the integral operator in L
2
() (for which the
diagonal value (y, y) is unambiguously dened [14]). The output of the com-
plementary channel is the integral operator in H
E
= L
2
() with the kernel

(y
2
, y
1
) = b(y
2
)[b(y
1
)) (y
2
, y
1
).
In nite dimensions every entanglement-breaking channel is anti-degradable
1
.
In innite dimensions, use representation (18) and dene the entanglement-
breaking channel T

: T(H
E
) T(H
B
) by the formula
T

[] =
_
Y
[b(y))b(y)[ (y, y)(dy),
then [] = T

[]] for = [)[ with T, hence the assertion follows.

4 Gaussian observables
In what follows we shall consider real vector space Z equipped with dierent
bilinear forms , , . . . . For concreteness and convenience of notation we shall
consider Z as the space of column vectors with components in R. Then the forms
are given by matrices which we denote by the same letter, e. g. (w, z) = w
T
z,
where
T
denotes transposition. If is an inner product on Z, then (Z, )
is an Euclidean space, while if is a nondegenerated skew-symmetric form
then (Z, ) is a symplectic space. We call symplectic space (Z, ) standard
if the commutation matrix, corresponding to the symplectic form (z, z

), is
= diag
_
0 1
1 0
_
. Let 2n = dimZ, and let d
2n
z denote element of symplectic
volume in (Z, ). Symplectic Fourier transform and its converse are given by

f(w) =
_
e
i(w,z)
f(z)d
2n
z; f(z) = (2)
2n
_
e
i(w,z)

f(w)d
2n
w.
Quantization on a symplectic space (Z, ) is given by (irreducible) Weyl sys-
tem in a Hilbert space H, which is a family of unitary operators W(z), z Z
satisfying the canonical commutation relations, one of the equivalent forms of
which is
W(z)

W(w)W(z) = exp (i(w, z)) W(w). (20)

By Stones theorem, W(z) = exp (iRz), where R is the row vector of selfadjoint
operators in H satisfying the Heisenberg commutation relations, see [3], Ch. V,
for detail.
1
We are indebted to M.-B.Ruskai for this observation [15].
7
Assume we have two symplectic spaces Z
A
, Z
B
with the corresponding Weyl
systems in Hilbert spaces H
A
, H
B
. Let M an observable in H
A
with the outcome
set Z
B
, given by positive operator-valued measure M(d
2n
z). In what follows
we skip the index
B
so that Z = Z
B
etc. It is completely determined by the
operator characteristic function [16]

M
(w) =
_
e
i(w,z)
M(d
2n
z).
A function (w) is characteristic function of an observable if and only if it
satises the following conditions:
1. (0) = I;
2. (w) is weakly continuous at w = 0;
3. for any choice of a nite subset w
j
Z
B
the block matrix with operator
entries (w
j
w
k
) is nonnegative denite.
In the case |
M
(w)| is integrable on Z
B
, measure M(d
2n
z) has density
p
M
(z), M(d
2n
z) = p
M
(z)d
2n
z, which is a.e. dened function on Z
B
, taking
values in the cone of bounded positive operators in H
A
, such that
_
p
M
(z)d
2n
z =
I. Moreover,
p
M
(z) = (2)
2n
_
e
i(w,z)

M
(w)d
2n
w. (21)
Observable M will be called Gaussian (canonical in [16], [3]) if its operator
characteristic function has the form

M
(w) = W
A
(Kw) exp
_

1
2
(w, w)
_
= exp
_
iR
A
Kw
1
2
(w, w)
_
, (22)
where K : Z
B
Z
A
is a linear operator and is a bilinear form on Z
B
. A
necessary and sucient condition for (22 ) to dene an observable is the matrix
inequality

i
2
K
T

A
K. (23)
Indeed, (22) apparently satises conditions 1,2 and (23) is equivalent to the
condition 3 since for an operator function given by ( 22)
(w
j
w
k
) = W(Kw
k
)

W(Kw
j
) exp
_

i
2
(Kw
j
, Kw
k
)
1
2
(w
j
w
k
, w
j
w
k
)
_
= C

k
C
j
exp
_
(w
j
, w
k
)
i
2
(Kw
j
, Kw
k
)
_
,
where C
j
= W(Kw
j
) exp
_

1
2
(w
j
, w
j
)

, and nonnegative deniteness of ma-

trices with scalar entries (w
j
, w
k
)
i
2
(Kw
j
, Kw
k
), with arbitrary choice of
nite subset w
j
, is equivalent to that for exp
_
(w
j
, w
k
)
i
2
(Kw
j
, Kw
k
)

(see [3], proof of Theorem 5.1, Ch. V).

8
Observable M is sharp if and only if = 0, in which case it is the spectral
measure of commuting selfadjoint operators R
A
K. In a sense opposite is the
following case:
Proposition 2. Let be nondegenerate and K invertible, then observable
M has operator density given by
p
M
(z) = W
A
(K
1
z)
A
W
A
(K
1
z)

1
(2)
n
[det K[
,
where
A
is Gaussian state with zero mean and correlation function (K
1
z, K
1
z

)
and K

=
1
K
T
is symplectic transpose.
Proof. Since is nondegenerate, |
M
(w)| = exp
_

1
2
(w, w)
_
is integrable,
and applying (21) we get
p
M
(z) = (2)
2n
_
e
i(w,z)
W
A
(Kw) exp
_

1
2
(w, w)
_
d
2n
w
= (2)
2n
_
e
i(u,K
1
z)
W
A
(u) exp
_

1
2
(K
1
u, K
1
u)
_
d
2n
u
[det K[
= (2)
2n
_
W
A
(K
1
z)W
A
(u)W
A
(K
1
z)

exp
_

1
2
(K
1
u, K
1
u)
_
d
2n
u
[det K[
=
1
(2)
n
[det K[
W
A
(K
1
z)
__
W
A
(u) exp
_

1
2
(K
1
u, K
1
u)
_
d
2n
u
(2)
n
_
W
A
(K
1
z)

.
Here we used change of variable u = Kw in the second line, the relation ( 20)
in the third line and in the fourth line the term in squared brackets is the Weyl
transform of the characteristic function of the state
A
.
Let us describe an explicit construction of Naimarks dilation of observable
M in the spirit of [3], Prop. 5.1, Ch. II.
Proposition 3. Assume the condition (23) holds, then there exist Bosonic
system H
C
with canonical observables R
C
such that H
B
H
A
H
C
and Gaus-
sian state
C
S(H
C
) for which
M(S) = Tr
C
(I
A

C
) E
AC
(S), S Z
B
, (24)
where E
AC
is a sharp observable in H
A
H
C
given by the joint spectral measure
of commuting selfadjoint operators
X
B
=
1
B
(R
A
K +R
C
K
C
)
T
, (25)
where K
C
: Z
B
Z
C
is operator such that
K
T
C

C
K
C
= K
T

A
K. (26)
Proof. The condition (26) means that K
T
C

C
K
C
+ K
T

A
K = 0, that is
commutativity of operators (25). By adapting the proof of Prop. 8.1 from Ch.
VI of [3], we obtain a symplectic space (Z
C
,
C
), operator K
C
: Z
B
Z
C
and
9
an inner product in Z
C
, given by symmetric matrix
C

i
2

C
such that (26)
holds along with
K
T
C

C
K
C
= .
Then the characteristic function of E
AC
is

EAC
(w
B
) =
_
exp[i(w
B
, z
B
)]E
AC
_
d
2n
z
B
_
= exp iw
T
B

B
X
B
= expi (R
A
K +R
C
K
C
) w
B
= W
A
(Kw
B
)W
C
(K
C
w
B
),
whence
Tr
C
(I
A

C
)
EAC
(w
B
) = W
A
(Kw
B
) exp
_

1
2

C
(K
C
w
B
, K
C
w
B
)
_
= W
A
(Kw
B
) exp
_

1
2
(w
B
, w
B
)
_
=
M
(w
B
),
and (24) follows.
5 Gaussian entanglement-breaking channels
Recall that characteristic function of a state is given by
(z) = TrW(z).
Channel : T(H
A
) T(H
B
) transforming states according to the rule

B
(z
B
) =
A
(Kz
B
)f(z
B
), (27)
where K is a linear map between output and input symplectic spaces (Z
B
,
B
),
(Z
A
,
A
) and f is a complex function, is called linear Bosonic. Necessary
and sucient condition on f is nonnegative deniteness of matrices with scalar
entries
f(w
j
w
k
) exp
_

i
2
(w
j
, w
k
) +
i
2
(Kw
j
, Kw
k
)
_
,
with arbitrary choice of nite subset w
j
Z
B
[17], [18].
If, additionally, f is a Gaussian characteristic function, the channel is Gaus-
sian. Thus for Gaussian channel, transformation of states is described as

B
(z
B
) =
A
(Kz
B
) exp
_
im(z
B
)
1
2
(z
B
, z
B
)
_
. (28)
The triple (K, m, ) is called parameters of the Gaussian channel. Without
loss of generality we assume m 0. Necessary and sucient condition on the
parameters of Gaussian channel is

i
2
_

B
K
T

A
K

. (29)
10
This follows from the condition for the general linear Bosonic channel applied
to Gaussian f similarly to the proof of (23). Importance of this condition in the
matrix form was emphasized in [19].
Theorem 2. Let be quantum Gaussian channel with parameters (K, 0, ).
It is entanglement-breaking if and only if admits decomposition
= +, where
i
2

B
,
i
2
K
T

A
K. (30)
In this case has the representation
[] =
_
ZB
W(z)
B
W(z)

(d
2n
z), (31)
where
B
is Gaussian state with parameters (0, ), and m

(S) = TrM
A
(S), S
Z
B
, is the probability distribution of the Gaussian observable M
A
with charac-
teristic function (22).
Proof. First, assume admits the decomposition and consider the channel
dened by (31); we have to show that

[W
B
(w)] = W
A
(Kw) exp
_

1
2
(w, w)
_
. (32)
Indeed, for arbitrary
Tr

[W
B
(w)] = Tr[]W
B
(w) =
_
ZB
TrW
B
(z)
B
W
B
(z)

W
B
(w)m

(d
2n
z)
=
_
ZB
Tr
B
W
B
(z)

W
B
(w)W
B
(z)m

(d
2n
z)
= Tr
B
W
B
(w)
_
ZB
exp[i(w, z)] m

(d
2n
z)
= exp
_

1
2
(w, w)
_
Tr
MA
(w) (33)
= TrW
A
(Kw) exp
_

1
2
(w, w)
1
2
(w, w)
_
,
whence (32) follows.
Conversely, let be Gaussian and entanglement-breaking. We will use Gaus-
sian version of the proof from [1], generalizing the Choi-Jamiolkowski correspon-
dence to innite-dimensional channels. Fix a Gaussian state
A
in S(H
A
) of full
rank and let [e
i
)
+
i=1
be the basis of eigenvectors of
A
with the corresponding
(positive) eigenvalues
i

+
i=1
. Consider the unit vector
[) =
+

i=1
_

i
[e
i
) [e
i
)
in the space H
A
H
A
, then [)[ is Gaussian purication of
A
. Since is
entanglement-breaking, the Gaussian state

AB
= (Id
A
) [[)[] (34)
11
in S(H
A
H
B
) is separable. As follows from the proof of Proposition 1 in [20],
this implies representation

AB
=
_
ZA
_
ZB
W
A
(z
A
)
A
W
A
(z
A
)

W
B
(z
B
)
B
W
B
(z
B
)

P(d
2m
z
A
d
2n
z
B
),
where
A
,
B
are Gaussian states and P is a Gaussian probability distribution.
One then shows as in [1] that the relation
M
A
(S)
=
1/2
A
_
_
_
ZA
_
S
W
A
(z
A
)
A
W
A
(z
A
)

W
B
(z
B
)
B
W
B
(z
B
)

P(d
2m
z
A
d
2n
z
B
)
_
_

1/2
A
,
where bar means complex conjugation in the basis of eigenvectors of
A
, denes
an observable on Borel subsets S Z
B
, and the representation (31) holds for
the channel with these M
A
and
B
. Let us denote the correlation function
of the state
B
; without loss of generality we can assume its mean is zero. It
remains to show that M
A
is Gaussian observable with the characteristic function
(22) where = . But from (33)

[W
B
(w)] = exp
_

1
2
(w, w)
_

MA
(w)
for any channel with the representation (31), whence taking into account (32),
we indeed get (22) with = .
A necessary condition for the decomposability (30) and hence for the channel
to be entanglement breaking is

i
2
_

B
K
T

A
K
_
. (35)
In general this condition implies that for any input Gaussian state of the chan-
nel Id
A
, the output has positive partial transpose. Indeed, this channel
transforms the correlation matrix of the input state according to the rule
_

11

12

21

22
_

_
I 0
0 K
T
_ _

11

12

21

22
_ _
I 0
0 K
_
+
_
0 0
0
_

AB
.
The right hand side representing the correlation matrix of the output state
satises

AB

i
2
_
I 0
0 K
T
_ _

A
0
0
A
_ _
I 0
0 K
_
+
i
2
_
0 0
0
B
K
T

A
K
_
=
i
2
_

A
0
0
B
_
,
where in the estimate of the second term we used (35) with its transpose. How-
ever, this is necessary and sucient for the output state to have positive partial
12
transpose [20]. As shown in [20], there are nonseparable Gaussian states with
positive partial transpose, therefore the condition (35) is in general weaker than
(30).
The condition of the theorem is automatically fullled in the special case
where
K
T

A
K = 0.
In this case operators R
A
K commute hence M
A
sharp observable given by is
their joint spectral measure and the probability distribution m

(d
2n
z) can be
arbitrarily sharply peaked around any point z by appropriate choice of the state
. Hence in this case it is natural to identify as c-q (classical-quantum) channel
determined by the family of states z W(z)
B
W(z)

[4].
6 The case of one mode
Let us apply theorem 2 to the case of one Bosonic mode A = B, where

A
(z, z

) =
B
(z, z

) = (z, z

) = x

y xy

As shown in [22], by choosing appropriate canonical unitary transformations

U
1
, U
2
, any one mode Gaussian channel with parameters (K, 0, ) can be trans-
formed via

[] = U
2

[U
1
U

1
] U

2
to one of the following normal forms, where N 0, k is real number:
A) K [x, y] = k [x, 0] ; (z, z) =
_
N +
1
2
_ _
x
2
+y
2
_
;
B
1
) K [x, y] = [x, y] ; (z, z) =
1
2
y
2
;
B
2
) K [x, y] = [x, y] ; (z, z) = N
_
x
2
+y
2
_
;
C) K [x, y] = k [x, y] ; k > 0, k ,= 1; (z, z) =
_
N +
[1 k
2
[
2
_ _
x
2
+y
2
_
;
D) K [x, y] = k [x, y] ; k > 0; (z, z) =
_
N +
(1 +k
2
)
2
_ _
x
2
+y
2
_
.
The case B
2
) representing channel with additive classical noise, and C) repre-
senting attenuator/amplier, are of major interest in applications [18], [23].
We have only to nd the form K
T

A
K and check the decomposability (30)
in each of these cases. We rely upon the simple fact that
_
N +
1
2
_
I
i
2

if and only if N 0.
A) K
T
K = 0, hence is c-q (in fact essentially classical) channel;
B) K
T
K = , hence the necessary condition (35) requires i. This
is never fullled in the case B
1
) due to degeneracy of . Thus the channel is
13
not entanglement breaking (in fact it has innite quantum capacity as shown
in [22]). On the other hand, in the case B
2
) the condition (30) is fullled with
= = /2 if and only if N 1, hence is entanglement breaking in this
case;
C) K
T
K = k
2
. It is clear that in this case the decomposability condition
holds if and only if
i
2
(1 +k
2
), which is equivalent to N +
|1k
2
|
2

(1+k
2
)
2
or
N min
_
1, k
2
_
. (36)
This gives the condition for the entanglement breaking (which also formally
includes the case B
2
));
D) K
T
K = k
2
. Again the decomposability condition holds if and only
if
i
2
(1 + k
2
), which always holds, hence the channel is entanglement
breaking for all N 0.
Thus the additivity property (5) holds for one-mode Gaussian channels of the
form A), D) with arbitrary parameters, and B
2
), C) with parameters satisfying
(36). To compare this with previous results, the only case where additivity of
C

(, E, c) with the special energy constraint (E = a

a) was established, is C)
with N = 0, k < 1 (pure loss channel) [24], which does not intersect with our
result. The actual computation of C

(, E, c) is in general an open problem:

there is a natural conjecture that the capacity of quantum Gaussian channel
with quadratic energy constraint is attained on a Gaussian ensemble of pure
Gaussian states, but so far this was only established for c-q channels [4] and the
pure loss channel [24]. If the conjecture is true, then in the cases B
2
), C), D) the
optimal ensemble is the complex Gaussian distribution P(d
2
z) with zero mean
and variance c on the coherent states W
A
(z)
0
W
A
(z)

where
0
is the vacuum
state. Hence C(, E, c) = g(k
2
c +N
0
) g(N
0
), where
N
0
=
_
(k
2
1)
+
+N, case B
2
), C);
k
2
+N, case D).
is the mean number of quanta in the output corresponding to the vacuum input
state and g(x) = (x + 1) log(x + 1) xlog x.
In general entanglement-breaking channels have zero quantum capacity Q() =
0, cf. Sec. 3. In this connection it is notable that the domain (36) coincides
with zero quantum capacity domain obtained in [20] from completely dierent
argument. However in any case this is superseded by the broader domain found
in [23] from degradability analysis.
Acknowledgements. This work was partially supported by RFBR grant
06-01-00164-a and the RAS scientic program Theoretical problems of modern
mathematics. The author is grateful to M. Shirokov, V. Giovannetti and M.
Wolf for discussions.
References
[1] A. S. Holevo, M. E. Shirokov and R. F. Werner, Separability and
entanglement-breaking in innite dimensions; quant-ph/0504204
14
[2] M. Horodecki, P. W. Shor, M.-B. Ruskai, General Entanglement Breaking
Channels, Rev. Math. Phys. 15, 629-641, 2003; quant-ph/0302031.
[3] A. S. Holevo, Probabilistic and statistical aspects of quantum theory.
Nauka, Moscow, 1980 (rus); North Holland Translation - 1982.
[4] A. S. Holevo, Quantum coding theorems, Russian Math. Surveys, 53, N.6,
1295-1331, 1998.
[5] P. W. Shor, Additivity of the Classical Capacity of Entanglement-
Breaking Quantum Channels, J. Math. Phys. 43, 4334-4340, 2002;
quant-ph/0201149.
[6] Holevo, A. S., Classical capacities of quantum channels with constrained
inputs, Probability Theory and Applications, 48, N.2, 359-374, 2003;
quant-ph/0211170.
[7] Holevo, A.S., Shirokov M.E., Continuous ensembles and the -capacity of
innite dimensional channels, Probability Theory and Applications, 50,
N.1, 98-114, 2005; quant-ph/0408176.
[8] Shirokov, M.E., The Holevo capacity of innite dimensional channels,
quant-ph/0408009, 2004.
[9] A. S. Holevo, M. E. Shirokov, On Shors channel extension and constrained
channels. Commun. Math. Phys. 249, 417-430, 2004; quant-ph/0306196.
[10] A. S. Holevo, Radon-Nikodym derivatives of quantum instruments, J.
Math. Phys. 39, 1373-1387, 1998.
[11] I. Devetak, P. Shor, The capacity of a quantum channel for simultaneous
transition of classical and quantum information, quant-ph/0311131.
[12] R. Caruso, V. Giovannetti, Degradability of Bosonic Gaussian channels,
quant-ph/0603257.
[13] B. Schumacher, M. D. Westmoreland, Quantum privacy and quantum co-
herence, Phys. Rev. Lett. 80, 5695-5697, 1998; quant-ph/9709058.
[14] A. S. Holevo, Covariant measurements and imprimitivity systems. Lect.
Notes Math., 1055, 153-172, 1984.
[15] T. Cubitt, Ruskai, M.B., G. Smith, The structure of degradable quantum
channels, 2008.
[16] A. S. Holevo, Statistical decision theory for quantum systems, J. Multivari-
ate Analysis, 3, 337-394, 1973.
[17] B. Demoen, P. Vanheuverzwijn, A. Verbeure, Completely positive quasi-
free maps on the CCR algebra, Rep. Math. Phys. 15, 27-39, 1979.
15
[18] A. S. Holevo, R. F. Werner, Evaluating capacities of Bosonic Gaussian
channels, Phys. Rev. A 63, 032312, 2001; quant-ph/9912067.
[19] J. Eisert, M. M. Wolf, Gaussian quantum channels, quant-ph/0505151.
[20] R.F. Werner, M. M. Wolf, Bound entangled Gaussian states,
quant-ph/0009118.
[21] M. M. Wolf, G. Giedke, J. I. Cirac, Extremality of Gaussian quantum
states, quant-ph/0509154.
[22] A. S. Holevo, One-mode quantum Gaussian channels, Problems of Infor-
mation Transmission, 43, 1-11, 2007; Arxiv quant-ph/0607051.
[23] F. Caruso, V. Giovannetti, A. S. Holevo, One-mode Bosonic Gaussian chan-
nels: a full weak-degradability classication, New Journal of Physics 8, 310,
2006; quant-ph/0609013.
[24] V. Giovannetti, S. Guha, S. Lloyd, L. Maccone, J. H. Shapiro,and H. P.
Yuen, Classical capacity of the lossy bosonic channel: the exact solution,
quant-ph/0308012.
16