Quantum information theory (MAT4430) Spring 2021

Lecture 16: Werner states, twirling and the no-cloning theorem

Lecturer: Alexander Müller-Hermes

Last lecture, we have studied the actions of the permutation group and the unitary group
on the space H⊗N . In this lecture, we will see two applications of these ideas. We will first
study a class of symmetric quantum states, which palys an important role in entanglement
theory, and second, we will prove the so-called no-cloning theorem showing that it is impos-
sible to copy quantum information.

1 Werner states
We will now apply the ideas of this lecture to define a family of symmetric quantum states.

Definition 1.1 (Werner states). A quantum state C C

ρ ∈ D d ⊗ d is called a Werner state

if (U ⊗ U )ρ = ρ(U ⊗ U ) for every U ∈ U d .

C
What makes Werner states so nice, is that general quantum states can be mapped to
this family by a natural symmetrization operation, which is called a twirl. This twirling
operation can be even be performed in practice, and it is the first step of many protocols
in quantum information theory to symmetrize the input in this way. Let us start with the
following general statement about symmetrizing operations:

C C
Theorem 1.2 (Twirling). The U U -twirl TU U : B( d ⊗ d ) → B( d ⊗ C Cd) given by
Z
TU U (X) = (U ⊗ U ) X (U ⊗ U )† dη(U ),
U (C )
d

is a selfadjoint quantum channel and we have

Psym Pasym
TU U (X) = hPsym , XiHS + hPasym , XiHS ,
Tr [Psym ] Tr (Pasym )

where
Psym =
1
2
(1d ⊗ 1d + Fd ) and Pasym =
1
2
(1d ⊗ 1d − Fd) ,
denote the projections onto the symmetric and the antisymmetric subspace of Cd ⊗ Cd,
respectively. Furthermore, note that
1 1
Tr [Psym ] = d (d + 1) and Tr [Pasym ] = d (d − 1) ,
2 2
are the dimensions of these spaces.

Proof. Consider an operator X ∈ B( Cd ⊗ Cd) and note that

(V ⊗ V )TU U (X) (V ⊗ V )† = TU U (X) ,

for every V ∈ U Cd by unitary invariance of the Haar measure. Clearly, this equation

shows that [V ⊗ V, TU U (X)] = 0 for each V ∈ U

C
d and by Theorem ?? we have

Psym Pasym
TU U (X) = a +b ,
Tr [Psym ] Tr [Pasym ]

1
since
span{1d ⊗ 1d , Fd} = span{Psym, Pasym}.
Moreover, note that
(U ⊗ U )Psym (U ⊗ U )† = Psym ,
and
(U ⊗ U )Pasym (U ⊗ U )† = Pasym ,
such that
a = hPsym , TU U (X)iHS = hTU U (Psym ) , XiHS = hPsym , XiHS ,
and
b = hPasym , TU U (X)iHS = hTU U (Pasym ) , XiHS = hPasym , XiHS .
This proves the formula for TU U stated in the theorem. From this formula it is clear that
TU U is a selfadjoint quantum channel since the projectors Psym and Pasym are positive
semidefinite.

Note that the Werner states are invariant under twirling, i.e., we have

TU U (ρ) = ρ,

for any Werner state ρ ∈ D( Cd ⊗ Cd). Since any quantum state ρ satisfies
1 = Tr [ρ] = hPsym , ρiHS + hPasym , ρiHS ,

and
hPsym , ρiHS ≥ 0 and hPasym , ρiHS ≥ 0,
we have the the following corollary:

Corollary 1.3. The Werner states on Cd ⊗ Cd are given by

Psym Pasym
ρW (t) = (1 − t) +t ,
Tr [Psym ] Tr (Pasym )

with t ∈ [0, 1]. For any quantum state σ ∈ D( Cd ⊗ Cd) we have

TU U (σ) = ρW (t),

for t = hPasym , σiHS .

Finally, we will demonstrate how useful the symmetry of the Werner states is, by deter-
mining the parameter region in which they are entangled.

Theorem 1.4. The following are equivalent:

1. The Werner state ρW (t) is separable.

2. The Werner state ρW (t) has positive partial transpose.

3. We have t ≤ 1/2.

Proof. Clearly, the first statement implies the second, and by direct computation it can be
verified that the second statement implies the third. To see how the third statement implies
the first, we will exploit the twirling operation. We start by noting that
1
hPsym , |φihφ| ⊗ |φ⊥ ihφ⊥ |iHS = = hPasym , |φihφ| ⊗ |φ⊥ ihφ⊥ |iHS ,
2

2
for any pair of orthogonal pure states |φi, |φ⊥ i ∈ Cd. We can also check that
hPsym , |φihφ| ⊗ |φihφ|iHS = 1,

and
hPasym , |φihφ| ⊗ |φihφ|iHS = 0.
Using the formula of the U U -twirl we find that
  
ρW (t) = TU U |φihφ| ⊗ 2t|φihφ| + (1 − 2t)|φ⊥ ihφ⊥ | ,

whenever 0 ≤ t ≤ 12 . But for any such t we then have

Cd, Cd
Z    
ρW (t) = U |φihφ|U † ⊗ 2tU |φihφ|U † + (1 − 2t)U |φ⊥ ihφ⊥ |U † dη(U ) ∈ Sep ,
u∈U (Cd )

and the proof is finished.

2 Impossibility of cloning and applications

2.1 The qualitative no-cloning theorem
The term “no-cloning theorem” is commonly used for the fact that unknown quantum states
cannot be cloned or copied. The most common argument derives a contradiction from
assuming a closed system with state space H undergoing some process modelled by a unitary
U ∈ U (H) satisfying
U (|ψi ⊗ |0i) = |ψi ⊗ |ψi, (1)
for all pure states |ψi ∈ H. Indeed such a unitary cannot exist, since otherwise

|hψ|φi|2 = (hψ| ⊗ h0|) U † U (|φi ⊗ |0i) = |hψ|φi|,

for any pair of pure states |ψi, |φi ∈ H. However, this equation can only be satisfied if
|hψ|φi| = 0 or |hψ|φi| = 1 leading to a contradiction since there are pure states satisfying
neither. Note that this argument actually implies a stronger statement: No unitary U ∈
U (H) can satisfy (1) for distinct, non-orthogonal pure states |ψ1 i, |ψ2 i ∈ H. The assumption
of non-orthogonality is crucial here and there are unitaries, e.g., the controlled-not gate,
satisfying (1) for certain orthogonal pure states.
The previous argument might not seem fully general, since there could exist more general
schemes for copying quantum information. The most general such operation would be some
quantum channel T : B(H) → B(H ⊗ H) satisfying



Tr ⊗idB(H) ◦ T = idB(H) ⊗ Tr ◦ T = idB(H) . (2)

To show that such an operation does not exist, we recall that any quantum state ρAB ∈
D (H ⊗ H) satisfying ρA = |ψihψ| and ρB = |ψihψ| for some pure state |ψi ∈ H is necessarily
of the form ρAB = |ψihψ| ⊗ |ψihψ|. Therefore, (2) implies that

T (|ψihψ|) = |ψihψ| ⊗ |ψihψ|,

for every pure state |ψihψ| ∈ Proj (H). We will now show that this property contradicts
linearity of T . Write some pure state |ψihψ| ∈ Proj (H) as

dim(H)2
X
|ψihψ| = λi |ψi ihψi |,
i=1

3
for {λi }i ⊂ R
and using a set of pure states {|ψi ihψi |}i which are linearily independent in
B (H). Then, we note that
X
T (|ψihψ|) = |ψihψ| ⊗ |ψihψ| = λi λj |ψi ihψi | ⊗ |ψj ihψj |
i,j
dim(H)2 dim(H)2
X X
6= λi |ψi ihψi | ⊗ |ψi ihψi | = λi T (|ψi ihψi |) .
i=1 i=1

We conclude that no quantum channel can satisfy (2) for every pure state |ψihψ| ∈ Proj (H)
as quantum channels are always linear.

2.2 The quantitative no-cloning theorem

The previous argument shows that there is no quantum process that “clones” arbitrary pure
states, i.e., implementing the map |φi 7→ |φi ⊗ |φi. What about the map |φi⊗n 7→ |φi⊗m for
m > n? We will now show that it is also impossible to find a quantum channel performing
this task. Moreover, we will quantify how well a quantum channel can approximately succeed
in it.
Theorem 2.1 (The quantitative no-cloning theorem). For any positive and trace-preserving
linear map T : B(H⊗n ) → B(H⊗m ) we have
d[n]
inf F (T (|vihv|⊗n ), |vihv|⊗m ) ≤ ,
|vi∈H d[m]
hv|vi=1

where  
k+d−1
d[k] = ,
k
is the dimension of the symmetric subspace H∨k , where d = dim(H). Moreover, there exists
a quantum channel T attaining this bound.
Proof. Using that T (|vihv|⊗n ) ≤ T (Psym
n ) by positivity, we can estimate

inf hT (|vihv|⊗n ), |vihv|⊗m iHS ≤ inf hT (Psym

n
), |vihv|⊗m iHS
|vi∈H |vi∈H
Z  ⊗m
≤ n
hT (Psym ), U |0ih0|U † iHS dη(U ).
U (H)

By a result from the previous lecture we have

m
Psym
Z  ⊗m
†

m
= U |0ih0|U dη(U ),
Tr Psym U (H)

and we obtain
m n )
 
⊗n ⊗m n
Psym Tr T (Psym d[n]
inf hT (|vihv| ), |vihv| iHS ≤ hT (Psym ), 
m
 iHS ≤ 
m
 = .
|vi∈H Tr Psym Tr Psym d[m]

To attain this bound, we consider the quantum channel T : B(H⊗n ) → B(H⊗m ) given by
d[n] m  ⊗(m−n)

T (X) = Psym X ⊗ 1H m
Psym + h1⊗n n
H − Psym , XiHS σ,
d[m]
where σ ∈ D (H⊗m ) is arbitrary. Note that T is a quantum channel since
h   i h  i d[m] 
m ⊗(m−n) m m ⊗(m−n) n

Tr Psym X ⊗ 1H Psym = Tr Psym X ⊗ 1H = Tr Psym X ,
d[n]

4
where we used that
Z

⊗(m−n)
 d[m] n
m
(U |0ih0|U † )⊗n dη(U ) =


idH ⊗ Tr Psym = d[m] P .
U (H) d[n] sym

Finally, we note that for any pure state |vi ∈ H we have

d[n] h
⊗(m−n)
 i d[n]
hT (|vihv|⊗n ), |vihv|⊗m iHS = Tr |vihv|⊗n ⊗ 1H m
Psym |vihv|⊗m Psym
m
= ,
d[m] d[m]
m |vihv|⊗m P m = |vihv|⊗m .
since Psym sym

For N = 2, the quantitative version of the no-cloning theorem shows that

2
inf F (T (|vihv|), |vihv|⊗2 ) ≤ ,
|vi∈H d+1
hv|vi=1

for any quantum channel1 T : B(H) → B(H ⊗ H). Hence, there are pure states |vihv| ∈
Proj (H) whose image under T is far away from the product |vihv| ⊗ |vihv| in fidelity. If you
look careful at the proof of the previous theorem, you can observe that a slightly stronger
statement holds: Even the average fidelity between the quantum states T (|vihv|) and the
pure states |vihv|⊗2 when averaging over pure states is low (provided that d is large). From
either result we conclude that it is impossible to exactly copy every pure state using a fixed
quantum channel. Remarkably, the theorem also identifies an optimal cloning channel, which
comes as close as possible to a copying device.

3 The Chiribella identity and quantum de-Finetti theorem

There is even more to say about the structure of the symmetric subspace and the no-cloning
problem. In the final part of this lecture, we will discuss a few additional observations which
will lead to the quantum de-Finetti theorem. To prove this result, we will need to consider
a subspace of symmetric operators:

3.1 Bose-symmetric operators

We start with a definition:
Definition 3.1 (Bose-symmetric operators). We will call operators in B H∨N Bose-symmetric.

Sometimes, we will tacitly view the Bose-symmetric operators as a subspace of B(H)∨N .

N and operators of the form |vihw|⊗N
Examples of Bose-symmetric operators include Psym
for |vi, |wi ∈ H. Note that B H∨N ( B(H)∨N since we have, for example, that

1⊗N ∨N
\ B H∨N .


H ∈ B(H)

The following lemma is left as an exercise:

Lemma 3.2. For any complex Euclidean space H we have

B H∨N = spanC {|vihv|⊗N : |vi ∈ H},

and
B H∨N = spanR {|vihv|⊗N : |vi ∈ H}.


sa

Proof. Exercise.
1
and even unphysical positive trace-preserving maps

5
3.2 Cloning and the measure-prepare map
In the quantitative version of the no-cloning theorem we introduced the optimal cloning map,
which we can restrict to the space B(H∨n ): Let clonen→m : B (H∨n ) → B (H∨m ) denote the
map given by
d[n] m  ⊗(m−n)

m
clonen→m (X) = Psym X ⊗ 1H Psym ,
d[m]
where we denote the dimension of the symmetric subspace H∨k by
 
k+d−1
d[k] := ,
k
with d = dim(H). The map clonen→m is the optimal cloning map from Theorem 2.1 restricted
to the space B(H∨n ). It can also be checked that this map coincides (up to a normalization
factor) with the adjoint Tr∗m→n of the partial trace map Trm→n : B(H∨m ) → B(H∨n ).
We will need another linear map on the symmetric operators: The measure-prepare map
MPm→n : B(H∨m ) → B(H∨n ) is given by
Z
MPm→n (X) = d[m] hφ⊗m ⊗m
U |X|φU i|φU ihφU |
⊗n
dη(U ),
U (H)

where |φU i = U |0i for any U ∈ U (H). We will need the following lemma:
Lemma 3.3. For any m ≥ n the maps MPm→n : B(H∨m ) → B(H∨n ) and clonen→m :
B(H∨n ) → B(H∨m ) are quantum channels.
Proof. It is clear that these maps are completely positive. To see that they are trace-
preserving we can compute
Z
hφ⊗m ⊗m
 m 
Tr [MPm→n (X)] = d[m] U |X|φU idη(U ) = Tr Psym X = Tr [X] ,
U (H)

for every X ∈ B(H∨m ). Similarly, for every Y ∈ B(H∨n ) we have

d[n] h
m

⊗(m−n)
i  n 
Tr [clonen→m (Y )] = Tr Psym Y ⊗ 1H = Tr Psym Y = Tr [Y ] ,
d[m]
as in the proof of Theorem 2.1.

There is a remarkable identity connecting the optimal cloning maps to the measurement-
prepare maps:
Theorem 3.4 (Chiribella identity). For any m ≥ n we have
n m n



d[m] X d[n] s s
MPm→n =
clones→n ◦ Trm→s .
d[m + n] d[s] m+n
s
s=0

Proof. Exercises.

Why is the Chiribella identity useful? Let us consider the coefficient of the last term on
the right-hand side (i.e., the term for s = n):
m
d+n n

 
d[m] n
m!(m + d − 1)! n(d + n)
m+n = ≥ 1− ≥1− .
d[m + n] n (m − n)!(m + n + d − 1)! d+m m+d

Since clonen→n = idB(H∨n ) , we conclude that

MPm→n = (1 − m,n,d ) Trm→n +m,n,d R,

6
for some quantum channel R : B(H∨m ) → B(H∨n ) and some

n(d + n)
m,n,d ≤ .
m+d
As a consequence we obtain the following theorem:

Theorem 3.5 (Quantum de-Finetti theorem). For any m ≥ n and any pure quantum state
|ψi ∈ H∨m there exists a quantum state

σ ∈ conv{|vihv|⊗n : |vi ∈ H, hv|vi = 1},

such that
n(d + n)
k Trm→n [|ψihψ|] − σk1 ≤ .
m+d