Quantum information theory (MAT4430) Spring 2021

Lecture 18: The decoupling method and the LSD theorem

Lecturer: Alexander Müller-Hermes

In this lecture, we will prove the decoupling theorem and use it to obtain coding schemes
for entanglement generation over a quantum channel T .

1 The coherent information and the LSD theorem

We will start with a definition:
Definition 1.1 (Coherent information). For a quantum channel T : B(HA ) → B(HB ) and
a quantum state σ ∈ D(HA ) we define the coherent information of σ through T as

 √
√
† 
Ic (σ; T ) = H (T (σ)) − H idB(HA ) ⊗ T vec σ vec σ ,

and the maximum coherent information of T as

Ic (T ) = max Ic (σ; T ) .
σ∈D(HA )

Recall that quantum channels T : B(HA ) → B(HB ) and S : B(HA ) → B(HE ) for
complex Euclidean spaces HA , HB and HE are complementary if there exists an isometry
V : HA → HB ⊗ HE such that
h i h i
T (X) = TrE V XV † and S(X) = TrB V XV † .

We will sometimes denote by T c a (standard) complementary channel obtained from a Stine-

spring dilation. The following lemma gives a useful alternative form of the coherent infor-
mation in terms of complementary channels:
Lemma 1.2. For any quantum channel T : B(HA ) → B(HB ) and any quantum state
σ ∈ D(HA ) we have
Ic (σ; T ) = H(T (σ)) − H(S(σ)),
for any quantum channel S : B(HA ) → B(HE ) complementary to T .
Proof. Consider an isometry V : HA → HB ⊗ HE such that
h i h i
T (X) = TrE V XV † and S(X) = TrB V XV † ,

and define the vector

√

|vABE i = (1A ⊗ V ) vec σ ∈ HA ⊗ HB ⊗ HE .

Note that

 √
√
† 
H idB(HA ) ⊗ T vec σ vec σ = H (TrE [|vABE ihvABE |])
= H (TrAB [|vABE ihvABE |]) = H(S(σ)),

where we used that |vABE i is a pure state and that reduced density operators of the same
pure state have the same spectra. The statement of the lemma follows by inserting the
previous equation into the definition of Ic (σ; T ).

1
 The previous lemma allows a neat physical interpretation of the coherent information
of σ through the quantum channel T : The coherent information quantifies how much more
information about the input state at A arrives at the receiver B compared to the environment
system E. Intuitively, it makes sense that this should be related to the task of quantum
communication since sending a pure state with low error means that the environment system
E can only be very weakly correlated with the receiving system B and hence has almost no
information about the state that was send. We will see in the next sections, that this intuition
can be used to prove the following capacity theorem:

Theorem 1.3 (Lloyd, Shor, Devetak). For any quantum channel T : B(HA ) → B(HB ) we
have
1  
Q(T ) = lim Ic T ⊗k .
k→∞ k

We will spend the rest of this lecture proving this result.

2 The decoupling approach

To prove the capacity theorem, we can focus on the task of entanglement generation for
which the achievable rates coincide with the achievable communication rates for quantum
information (see previous lecture). To find efficient entanglement generation schemes we will
apply the so-called decoupling approach. The main idea behind this strategy is that a decoder
generating entanglement for some given pure input |φRA i ∈ HR ⊗ HA state can be directly
obtained from Uhlmann’s theorem. The final error of this entanglement generation scheme is
given by the distance of the RE marginal of the joined quantum state, obtained from sending
the A system of |φRA i through the Stinespring isometry of the quantum channel under
consideration, from a product state. Intuitively, this says that if after the application of the
quantum channel the reference system R and the environment system E are approximately
decoupled, then we can approximately generate entanglement between R and the output
system B of the channel (see Figure ....). The next theorem makes this intuition precise:

Theorem 2.1 (Decoupling implies code). Consider a quantum channel T : B(HA ) →

B(HB ) with Stinespring isometry V : HA → HE ⊗ HB and a pure state |φRA i ∈ HR ⊗ HA
for some complex Euclidean space HR . For the pure state

|ψ REB i = (1R ⊗ V )|φRA i ∈ HR ⊗ HE ⊗ HB

we assume that
1R
k TrB |ψ REB ihψ REB | − ⊗ τ E k1 ≤ ,
 
dim(HR )
for some τ E ∈ D(HE ). Then, there exists a quantum channel D : B(HB ) → B(HR )
satisfying
1
F (ωR , (idR ⊗ D ◦ T ) |φRA ihφRA | ≥ 1 − .


2
Proof. Exercises.

The decoupling theorem concerns the following situation:

To find coding schemes to share entanglement (and hence to communicate quantum
information) we will need the following theorem:

Theorem 2.2 (The decoupling theorem). Consider complex Euclidean spaces HA and HE ,
and a (not necessarily normalized) vector |φAEB i ∈ HA ⊗ HE ⊗ HB . Furthermore, consider

2
a projection P : HA → HA such that HR = Im (P ) ⊂ HA and define the (unnormalized)
vectors s
dim(HA )
|ψUREB i = (P U ⊗ 1E ⊗ 1B ) |φAEB i,
dim(HR )
for each unitary U ∈ U (HA ). Then, we have
Z r
P h i
TrB |ψUREB ihψUREB | ⊗ τE dim(HR ) dim(HE ) Tr (τ AE )2 ,
 
− dη(U ) ≤
U (HA ) dim(HR ) 1

where we used the reduced operators

τ E = TrAB |φAEB ihφAEB |

 

and
τ AE = TrB |φAEB ihφAEB | .
 

We will need a few properties of Haar-integrals:

• By a result from an earlier lecture, we have

Z
1H
U XU † dη(U ) = Tr [X] .
U (H) dim(H)

• We also need the formula for the U U -twirling channel TU U : B(H ⊗ H) → B(H ⊗ H)
given by
Z
TU U (X) = (U ⊗ U )X(U ⊗ U )† dη(U )
U (H)
Psym Pasym
= hPsym , XiHS + hPasym , XiHS ,
Tr [Psym ] Tr [Pasym ]

for
Psym =
1
2
(1H ⊗ 1H + ) F and Pasym =
1
2
F
(1H ⊗ 1H − ) .

We can now prove the decoupling theorem:

Proof. Note first that

Z
|ψUREB ihψUREB |dη(U )
U (HA )
Z
dim(HA )  
= (P U ⊗ 1E ⊗ 1B ) |φAEB ihφAEB | U † P ⊗ 1E ⊗ 1B dη(U )
dim(HR ) U (HA )
P
= ⊗ τ EB ,
dim(HR )

where
τ EB = TrA |φAEB ihφAEB | .
 

3
As a consequence, we find that
Z
P 2
TrB |ψUREB ihψUREB | − ⊗ τ E dη(U )
 
U (HA ) dim(HR ) 2
Z
 REB REB   REB REB 
= hTrB |ψU ihψU | , TrB |ψU ihψU | i
U (HA )
P
⊗ τ E , TrB |ψUREB ihψUREB | i
 
−h
dim(HR )
P
− hTrB |ψUREB ihψUREB | , ⊗ τ Ei
 
dim(HR )
P P
+h ⊗ τ E, ⊗ τ E idη(U )
dim(HR ) dim(HR )
Z h 2 i 1 h
2 i
Tr TrB |ψUREB ihψUREB | dη(U ) − Tr τ E

= .
U (HA ) dim(HR )

Let us focus on the first term in the last line. Note that

F F
h
2 i
Tr ρAE = Tr ( A ⊗ E )(ρAE ⊗ ρAE ) ,
 

F F
where A ∈ U (HA ⊗ HA ) and E ∈ U (HE ⊗ HE ) are flip operators exchanging the two A
or E systems, respectively. Using that
 dim(HA )
TrB |ψUREB ihψUREB | = (P U ⊗ 1E )τ AE (U † P ⊗ 1E ),

dim(HR )
we find that
Z h 2 i
Tr TrB |ψUREB ihψUREB | dη(U )

U (HA )

F F
Z
Tr ( A ⊗ E ) TrB |ψUREB ihψUREB | ⊗ TrB |ψUREB ihψUREB | dη(U )
    

=
U (HA )
dim(HA )2
FA (P U ⊗ P U ) FE
Z h   
i
= Tr U †P ⊗ U †P ⊗ τ AE ⊗ τ AE dη(U )
dim(HR )2 U (HA )

=
dim(HA )2
dim(HR )2

Tr (TU U ( FR ) ⊗ FE ) τ AE ⊗ τ AE


,

where we used the U U -twirling channel TU U from a previous lecture, and introduced the
operator
F
R = (P ⊗ P ) A (P ⊗ P ). F
It is easy to compute that

hPsym , FR iHS = 12 (Tr [FR ] + Tr [P ⊗ P ]) = 12 dim(HR ) (dim(HR ) + 1) =: cs,

and

hPasym , FR iHS = 21 (Tr [FR ] − Tr [P ⊗ P ]) = 12 dim(HR ) (1 − dim(HR )) =: ca.

Using the formula for the U U -twirl from a previous lecture, we find that

TU U ( FR ) = cs TrP[Psym Pasym
+ ca
sym ] Tr [Pasym ]

FA.
   
1 cs ca 1 cs ca
= + 1A ⊗ 1A + −
2 Tr [Psym ] Tr [Pasym ] 2 Tr [Psym ] Tr [Pasym ]

4
An easy (but tedious) computation reveals that
   
1 cs ca dim(HA ) dim(HA ) − dim(HR ) 1
+ = ≤ ,
2 Tr [Psym ] Tr [Pasym ] dim(HR ) dim(HA )2 − 1 dim(HR )
and    
1 cs ca dim(HA ) dim(HR ) dim(HA ) − 1
− = ≤ 1.
2 Tr [Psym ] Tr [Pasym ] dim(HR ) dim(HA )2 − 1
Combining the previous equations shows that
dim(HA )2
dim(HR )2

F F
Tr (TU U ( R ) ⊗ E ) τ AE ⊗ τ AE




F F
≤ Tr ( A ⊗ E ) τ AE ⊗ τ AE +

 1
dim(HR )

Tr (1A ⊗ 1A ⊗ FE ) τ AE ⊗ τ AE



h 2
i 1 h
2 i
= Tr τ AE Tr τ E


+ ,
dim(HR )
and combining this with the computation from before we have
Z h 2 i h
2 i
Tr TrB |ψUREB ihψUREB | dη(U ) ≤ Tr τ AE

.
U (HA )

Finally, we can use the equivalence between the k · k2 -norm and the k · k1 -norm together with
the fact that the square root is concave to obtain
Z
P
TrB |ψUREB ihψUREB | − ⊗ τ E dη(U )
 
U (HA ) dim(H R ) 1
Z s
P 2
dim(HR ) dim(HE ) TrB |ψUREB ihψUREB | −
 
≤ ⊗ τ E dη(U )
U (HA ) dim(HR ) 2
s Z
P 2
TrB |ψUREB ihψUREB | −
 
≤ dim(HR ) dim(HE ) ⊗ τ E dη(U )
U (HA ) dim(HR ) 2
r h i
≤ dim(HR ) dim(HE ) Tr (τ AE )2 ,

and the proof is finished.

3 Some technical lemmas

To prove that rates close to the coherent information are achievable for quantum commu-
nication over a quantum channel, we will need some technical lemmas. The first lemma
summarizes some properties of typical projections of pure states:
Lemma 3.1. For complex Euclidean spaces HA , HB and HE let |ψi ∈ HA ⊗ HB ⊗ HE be
some pure state and set
ρA = TrBE [|ψihψ|]
ρB = TrAE [|ψihψ|]
ρE = TrAB [|ψihψ|] .

Moreover, let Πn,δ ⊗n n,δ ⊗n

and Πn,δ ⊗n




A ∈ Proj HA , ΠB ∈ Proj HB E ∈ Proj HE denote pro-
jections onto the δ-typical subspaces with respect to the marginal states ρ⊗n
A , ρ ⊗n ⊗n
B and ρE ,
respectively, and define the (unnormalized) vector
|ψδn i = (Πn,δ n,δ n,δ
A ⊗ ΠB ⊗ ΠE )|ψi
⊗n ⊗n
∈ HA ⊗n
⊗ HB ⊗ HE⊗n
Then, the following statements hold:

5
 1. We have
k|ψihψ| − |ψδn ihψδn |k1 → 0,
as n → ∞.
2. We have h i
Tr Πn,δ
E ≤ 2nH(ρE )+nδ .

3. We have h i
Tr (TrAn E n [|ψδn ihψδn |])2 ≤ 2−nH(ρB )+nδ

The proof of this lemma will use the operator-inequality

21HA ⊗ 1HB ⊗ 1HC + ΠA ⊗ ΠB ⊗ ΠC
≥ ΠA ⊗ 1HB ⊗ 1HC + 1HA ⊗ ΠB ⊗ 1HC + 1HA ⊗ 1HB ⊗ ΠC , (1)
which holds for any triple of projections ΠA ∈ Proj (HA ), ΠB ∈ Proj (HB ) and ΠC ∈
Proj (HC ). To prove it, note that both sides of the inequality commute and it is therefore
enough to check the inequality on an orthonormal basis of joined eigenvectors. Such a basis
is easily constructed from eigenbases of the individual projection operators.

Proof of Lemma 3.1. By (1) we have

h i h i h i
hψ ⊗n |(Πn,δ
A ⊗ Π n,δ
B ⊗ Πn,δ
E )|ψ ⊗n
i ≥ Tr Π n,δ ⊗n
A Aρ + Tr Πn,δ ⊗n
B Bρ + Tr Π n,δ ⊗n
E Eρ −2
−→ 1 as n → ∞.
Then, we can use an identity from exercise sheet 6 to compute
|ψ n ihψ n |
k|ψihψ|⊗n − |ψδn ihψδn |k1 = k|ψihψ|⊗n − cn δ δ k1
cn
p
= (1 + cn )2 − 4c2n → 0,
as n → ∞, and where we used
cn = hψδn |ψδn i = hψ ⊗n |(Πn,δ n,δ n,δ
A ⊗ ΠB ⊗ ΠE )|ψ
⊗n
i.
The second statement is a basic property of the typical projection, which we showed previ-
ously. For the third statement note that
TrAn E n [|ψδn ihψδn |] ≤ Πn,δ ⊗n n,δ
B ρB ΠB ≤ 2
−nH(ρB )+nδ n,δ
ΠB . (2)
Here, the first inequality in (2) follows from the fact that
Tr [XB TrA [(ΠA ⊗ 1B )YAB (ΠA ⊗ 1B )]] = Tr [(ΠA ⊗ XB )YAB ] ≤ Tr [(1A ⊗ XB )YAB ] = Tr [XB YB ] ,
for any projection ΠA ∈ Pro(HA ), any XB ∈ B(HB )+ and any YAB ∈ B(HA ⊗ HB )+ .
Showing that
TrA [(ΠA ⊗ 1B )YAB (ΠA ⊗ 1B )] ≤ YB .
The last inequality in (2) follows from a bound in lecture 9. Using again basis properties of
typical projections, we find that
h i h i
Tr (TrAn E n [|ψδn ihψδn |])2 ≤ (2−nH(ρB )+nδ )2 Tr Πn,δ
B ≤ 2−nH(ρB )+nδ ,

since for operators X, Y ∈ B(H) we have

h i
Tr (X + Y )2 = Tr X 2 + 2 Tr Y 1/2 XY 1/2 + Tr Y 2 ≥ Tr X 2 ,
       

whenever Y ≥ 0.

6
 The next lemma is a simple fact about the trace distance between quantum states:

Lemma 3.2. For every pair of quantum states ρ, σ ∈ D(H) and any c ∈ R, we have
kρ − σk1 ≤ 2kcρ − σk1 .

Proof. We have

kρ − σk1 ≤ kρ − cρk1 + kcρ − σk1 = |1 − c| + kcρ − σk1 .

The statement of the lemma follows by realizing that

|1 − c| = | Tr [cρ − σ] | ≤ kcρ − σk1 .

Finally, we need an estimate of a certain Haar-integral, which we will prove in the exer-
cises:

Lemma 3.3. Consider a selfadjoint operator H ∈ B(HA ⊗ HB )sa and a projection P :

HA → HA with Im (P ) ⊆ HC ⊂ HA . Then, we have
Z
dim(HC )
k(P U ⊗ 1HB )H(U † P ⊗ 1HB )k1 dη(U ) ≤ kHk1 .
Tr [P ] U (HC )

Proof. Assume for each U ∈ U (HC ) that YU ∈ B(HA ⊗ HB ) satisfies kYU k∞ = 1 and
h i
k(P U ⊗ 1HB )H(U † P ⊗ 1HB )k1 = Tr YU† (P U ⊗ 1HB )H(U † P ⊗ 1HB ) .

Note that kYU k∞ = 1 implies that −1HA ⊗ 1HB ≤ YU ≤ 1HA ⊗ 1HB implying that

−(U † P U ⊗ 1HB ) ≤ (U † P ⊗ 1HB )YU† (P U ⊗ 1HB ) ≤ (U † P U ⊗ 1HB ),

for each U ∈ U (HC ). By integrating these inequalities we find that

Z
dim(HC )
−(1HA ⊗ 1HB ) ≤ (U † P ⊗ 1HB )YU† (P U ⊗ 1HB )dη(U ) ≤ (1HA ⊗ 1HB ).
Tr [P ] U (HC )

This shows that

Z
dim(HC )
k (U † P ⊗ 1HB )YU† (P U ⊗ 1HB )dη(U )k∞ ≤ 1,
Tr [P ] U (HC )

and that
Z
dim(HC )
k(P U ⊗ 1HB )H(U † P ⊗ 1HB )k1 dη(U )
Tr [P ] U (HC )
"Z #
dim(HC )
= Tr (U † P ⊗ 1HB )YU† (P U ⊗ 1HB )dη(U ) · H ≤ kHk1 .
Tr [P ] U (HC )

7
4 Achieving rates close to the coherent information
Now, we can proceed with the main theorem in this lecture:
Theorem 4.1. Let T : B(HA ) → B(HB ) denote a quantum channel and σ ∈ D(HA ) a
quantum state. Then, any rate
0 ≤ R < Ic (σ; T ) ,
is achievable for entanglement generation over the quantum channel T .
Proof. Throughout the proof, we will fix a δ > 0, and we will show that any rate

0 ≤ R < Ic (σ; T ) − 3δ,

is achievable for entanglement generation over the quantum channel T . Let V : HA →

0
HE ⊗HB denote a Stinespring isometry of T and let |φAA i ∈ HA ⊗HA denote the purification
of σ ∈ D(HA ) given by √

0
|φAA i = vec σ .
Next, consider the reduced quantum states
h 0 0
 i h 0 0
i
τ A = TrEB (1A ⊗ V ) |φAA ihφAA | 1A ⊗ V † = TrA0 |φAA ihφAA |
h 0 0
 i
τ E = TrAB (1A ⊗ V ) |φAA ihφAA | 1A ⊗ V † = T c (σ)
h 0 0
 i
τ B = TrAE (1A ⊗ V ) |φAA ihφAA | 1A ⊗ V † = T (σ),

of the pure quantum state

0
|τ AEB i = (1A ⊗ V )|φAA i.
For brevity, let us set

⊗n ⊗n
|τ n i = |τ AEB i ∈ HA ⊗ HE⊗n ⊗ HB⊗n
.

For every n ∈ N
we consider the projections Πn,δ
A ∈ Proj HA⊗n

n,δ
, ΠB ∈ Proj HB⊗n


and
n,δ ⊗n

n,δ ⊗n n,δ ⊗n n,δ ⊗n
ΠE ∈ Proj HE onto the δ-typical subspaces HA ⊂ HA , HB ⊂ HB and HE ⊂ HE
⊗n ⊗n ⊗n
with respect to the marginal states τA , τB and τE , respectively. Then, we define the
(unnormalized) vectors

|τδn i = (Πn,δ n,δ n,δ n,δ n,δ n,δ

A ⊗ ΠE ⊗ ΠB )|τn i ∈ HA ⊗ HE ⊗ HB ,

and we denote its marginal by

E
τn,δ = TrAn B n [|τδn ihτδn |] .
 
n,δ
Moreover, for any unitary U ∈ U HA we define the (unnormalized) vectors
s
n,δ 
dim(HA ) 
⊗n ⊗n ⊗n
|ψUn i = n) P U ⊗ 1H⊗n ⊗ 1H⊗n |τ n i ∈ HR ⊗ HE ⊗ HB ,
dim(HR E B

and s
n,δ 
dim(HA ) 
|ψUn,δ i = n) P U ⊗ 1 n,δ ⊗ 1 n,δ
HE HB
⊗n
|τδn i ∈ HR n,δ
⊗ HE n,δ
⊗ HB ,
dim(HR
for some fixed projection P : HA ⊗n
→ HA ⊗n
with HR n = Im (P ) ⊂ Hn,δ being some subspace
A
of dimension dim(HR n ) = 2nR . Finally, we define the pure quantum state

1 ⊗n ⊗n ⊗n
|ψ̃Un i = p n n |ψUn i ∈ HR ⊗ HE ⊗ HB ,
hψU |ψU i

8
 N and each U ∈ U HAn,δ .  
 
for every n ∈
Note that for every n ∈ N and U ∈ U HA
n,δ
we have

 
|ψ̃Un i = 1Rn ⊗ V ⊗n |φn,U
RA0 ihφ n,U
RA0 | ,

for the pure quantum states |φn,U n ⊗n

RA0 i ∈ HR ⊗ HA arising from normalizing the vectors
s
n,δ 
dim(HA )  0
n) P U ⊗ 1 ⊗n
HA |φAA i⊗n .
dim(HR

N
 
n,δ
We will now show that for each n ∈ , there exists a unitary Un ∈ U HA such that
h i P
k TrB n |ψ̃Unn ihψ̃Unn | − nR ⊗ (τ E )⊗n k1 → 0,
2
as n → ∞. Using Theorem 2.1 and identifying HR
⊗n
C
n with ( 2 )⊗Rn , we find a sequence of
n
quantum channels Dn : B(HB ) → B(HR ) satisfying
 
n,U n,U
F (ω2⊗Rn , (id⊗Rn
2 ⊗ D n ◦ T ⊗n
) |φ RA 0 ihφ RA 0 | ) → 1,

as n → ∞. We conclude that the rate R is achievable for entanglement generation using T .

To finish the proof, we first apply Lemma 3.2 to estimate
h i P P
k TrB n |ψ̃Un ihψ̃Un | − nR ⊗ (τ E )⊗n k1 ≤ 2k TrB n [|ψUn ihψUn |] − nR ⊗ (τ E )⊗n k1 ,
2 2
N
 
n,δ
for every n ∈ and each U ∈ U HA . To estimate the right-hand side in the previous
inequality, we use the triangle inequality to obtain
P h i
k TrB n [|ψUn ihψUn |] − nR ⊗ (τ E )⊗n k1 ≤ k TrB n [|ψUn ihψUn |] − TrB n |ψUn,δ ihψUn,δ | k1
2
h i P
+ k TrB n |ψUn,δ ihψUn,δ | − nR ⊗ τn,δ E
k1
2
P P
+ k nR ⊗ τn,δ E
− nR ⊗ (τ E )⊗n k1 .
2 2
We will now derive upper bounds on the Haar-integrals of the three summands in the last
inequality. First, note that by monotonicity of the trace-distance under partial traces, Lemma
3.2 and 3.1 we have
P P
k nR ⊗ τn,δ E
− nR ⊗ (τ E )⊗n k1 ≤ kτn,δ
E
− (τ E )⊗n k1
2 2
≤ k|τδn ihτδn | − |τ n ihτ n |k1 → 0,
as n → ∞. Furthermore, by Lemma 3.3 we have that
Z h i
n n n,δ n,δ
  k TrB n [|ψU ihψU |] − TrB n |ψU ihψU | k1 dη(U )
n,δ
U HA

≤ k TrB n [|τ n ihτ n |] − TrB n [|τδn ihτδn |] k1 → 0,

as n → ∞. Finally, we note that by Theorem 2.2 we have
Z h
n,δ n,δ
i P E
 k TrB n |ψU ihψU | − nR ⊗ τn,δ k1 dη(U )
2

n,δ
U HA
s  2 
nR n,δ B
≤ 2 dim(HE ) Tr τn,δ
p
≤ 2n(R−H(T (σ))+H(T c (σ))+3δ) → 0,

9
as n → ∞ under the assumptions on the rate R from above. We conclude that
Z
n n P E ⊗n
 k TrB n [|ψU ihψU |] − nR ⊗ (τ ) k1 dη(U ) → 0,
2

n,δ
U HA

 N
and hence there exists a sequence (Un )n∈N ∈ U HA
n,δ
satisfying

P
k TrB n |ψUnn ihψUnn | − nR ⊗ (τ E )⊗n k1 → 0,
 
2
as n → ∞. Finally, we can combine this with the estimates from above to see that
h i P
k TrB n |ψ̃Unn ihψ̃Unn | − nR ⊗ (τ E )⊗n k1 → 0,
2
as n → ∞. By the argument from before, this finishes the proof.

In the previous lecture, we have seen that the entanglement generation capacity coincides
with the quantum capacity, and together with the previous theorem we conclude that

Ic (σ; T ) ≤ Q(T ),

for any quantum channel T : B(HA ) → B(HB ) and any quantum state σ ∈ D(HA ). Op-
timizing over σ ∈ D(HA ) and applying the resulting bound for the quantum channel T ⊗k
instead of T implies1 the following corollary:

Corollary 4.2. For any quantum channel T : B(HA ) → B(HB ) we have

1  
lim sup Ic T ⊗k ≤ Q(T ).
k→∞ k

1
by using that Q(T ⊗k ) = kQ(T ).

10