quantum entanglement transformations via local operations … · 2016. 11. 17. · entanglement...
TRANSCRIPT
Quantum Entanglement
Transformations via Local Operations
and Classical Communication
Student’s name: Cheng Guo
Supervisor: Prof. Runyao Duan
Centre for Quantum Computation and Intelligent Systems
Faculty of Engineering and Information Technology
University of Technology, Sydney
CERTIFICATE OF ORIGINAL AUTHORSHIP
I certify that the work in this thesis has not previously been submitted for a
degree nor has it been submitted as part of requirements for a degree except as
fully acknowledged within the text.
I also certify that the thesis has been written by me. Any help that I have
received in my research work and the preparation of the thesis itself has been
acknowledged. In addition, I certify that all information sources and literature
used are indicated in the thesis.
Acknowledgments
Firstly, I would like to thank my supervisor Prof. Runyao Duan. He gives
me a deep impression with his patience and wisdom. I met him at Tsinghua
University, when I was looking for an interesting topic for Bachelor graduation
thesis. He introduced Quantum Information Theory to me. Then, I came to
University of Technology, Sydney. He has been a tremendous mentor for me.
His advices both on my research as well as on my career have been priceless. He
offers his continuous advice and encouragement throughout this thesis. I thank
him for his systematic guidance and great effort that he put into training me in
the scientific field.
I would like to thank Prof. Mingsheng Ying, Prof. Chengqi Zhang, Prof.
Yuan Feng, Prof. Sanjiang Li, Prof. Min-Hsiu Hsieh, Prof. Enrico Carlini and
Prof. Eric Chitambar for all their help and encouragement in these years.
Many thanks to Dr. Nengkun Yu for his pragmatic advices. I learned many
original ideas and techniques in our discussions. Thank Dr. Youming Qiao, Dr.
Ching-Yi Lai, Dr. Yangjia Li and Dr. Shenggang Ying for their kind help and
support.
Finally, I would like to thank my parents for everything they have done for
me. Words cannot express how grateful I am to my parents for all of the sacrifices
that you have made on my behalf.
AbstractThe primary goals of this thesis are two-fold: i) calculating the optimal entan-
glement transformation rate between two multipartite pure states via stochas-
tic local operations and classical communication (SLOCC), and ii) showing the
properties of a common resource for a set of multi-partite pure state via local
operations and classical communication (LOCC) or SLOCC.
We introduce a notion of entanglement transformation rate to character-
ize the asymptotic comparability of two multi-partite pure entangled states
under SLOCC. For two well known SLOCC inequivalent three-qubit states:
Greenberger-Horne-Zeilinger (GHZ) state and W state, we show that the entan-
glement transformation rate from GHZ state to W state is exactly 1. We then
apply similar techniques to obtain a lower bound on the entanglement transfor-
mation rates from an N -partite GHZ state to a class of Dicke states.
Then, we discuss the common resource for a set of pure states. We have com-
pletely solved the bipartite pure states case by explicitly constructing a unique
optimal common resource state for any given set of states via LOCC. In the
multi-partite setting, the general problem becomes quite complicated, and we
focus on finding non-trivial common resources for the whole multi-partite state
space of given dimensions. We show that |GHZ3〉 = (1/√
3)(|000〉+|111〉+|222〉)is a nontrivial common resource for three-qubit systems via LOCC. We also ob-
tain a number of interesting properties of non-trivial common resource states for
two N -qubit pure states and multi-partite systems via SLOCC.
Key words: Entanglement Transformation, LOCC, GHZ state, W state,
Tensor Rank, Common Resource States.
Contents
I. Introduction 1
A. Main topics and motivation 2
B. Overview of the thesis 4
II. Preliminaries 7
A. Linear Algebra 7
B. Postulates of Quantum Mechanics 9
C. Tensor rank, Schmidt decomposition, majorization and their
applications 13
III. Asymptotic Rate of State Transformation 21
A. Previous results about R(|GHZ〉, |W 〉) 23
B. R(|GHZ〉, |W 〉)=1 27
C. Generalization to Dicke states 31
D. Alternative proof 34
E. Computational complexity of matrix permanent 36
IV. Common resource via local operations and classical
communication 39
A. Optimal common resource of bipartite pure states 41
B. Optimal common resource of a special kind of multi-partite pure states44
C. A non-trivial common resource of 3-qubit system: |GHZ3〉 46
V. Common resource via stochastic local operations and classical
communication 55
A. Tripartite entangled pure states 57
B. Multi-partite qubit pure states 63
VI. Conclusion 69
References 71
I INTRODUCTION
I. INTRODUCTION
Quantum Mechanics was established in early 20th century. Entanglement
correlation in Quantum Mechanics is one of the most significant differences com-
pared with classical correlation in Classical Mechanics. Such quantum entangle-
ment correlation may exist in multi-subsystems even if they are far away from
each other, and random events taking place on one subsystem can affect the
state’s other part. Historically, the concept of entanglement first appeared in
Schrodinger “cat state” papers [1].
Due to the distances among the different subsystems of quantum entangled
states, we have the following two reasonable restrictions: (1) (local operations)
each party can only perform operations on his/her own subsystem and global
operations on the composite system are not allowed, and (2) (classical commu-
nication) each party can tell other parties his/her measurement outcomes by
classical channels in order to communicate and coordinate. This class of oper-
ations is called “local operations and classical communication”, abbreviated as
LOCC. The fundamental problem of quantum state transformation is to ask, for
a given state, what kind of states can be obtained from this one via LOCC. For
an arbitrary mutli-partite entangled state, it has been proven this problem is
hard except for bipartite pure states [2]. If the success probability of the trans-
formation is positive (but usually strictly less than 1), the protocol is named
“stochastic local operations and classical communication” (SLOCC). There are
many interesting open problems about SLOCC transformations.
In this thesis, we are mainly interested in two topics. The first topic is to
calculate the asymptotic rate of a state transformation from Greenberger-Horne-
1
I INTRODUCTION
Zeilinger (GHZ) state to W state. The second one is to study the properties of
common resource states of a set of multi-partite pure state via LOCC or SLOCC.
A. Main topics and motivation
The problem of transforming one entangled state to another by LOCC is of
central importance in quantum entanglement theory. In order to make quantita-
tive comparison between different types of quantum information resources, the
following fundamental entanglement transformation problem arises: whether an
N -partite state can be transformed into another given N -partite state via LOCC.
In bipartite case, a necessary and sufficient condition for entanglement transfor-
mation was reported by Nielsen [2]. In tripartite case, this condition doesn’t hold.
Dur, Vidal, and Cirac observed that within three-qubit systems, there exist two
distinct equivalence classes of genuinely tripartite entangled states: GHZ-type
states and W-type states [4].
What is the asymptotic rate of state transformation from GHZ state to W
state? This has been a longstanding open problem. We improve the upper bound
of tensor rank of the n-copies W state these years [5] [6] [7]. Finally, we find the
answer: though we all know W and GHZ cannot be transformed to each other
via LOCC or even SLOCC, it is interesting that when we have sufficiently large
number of copies of GHZ state, we can obtain nearly the same number of W
state via SLOCC.
The problem of common resource state naturally arises when we need to pre-
pare a number of different pure states. Due to the practical limitation, one eco-
nomic strategy is to prepare minimal number of standard quantum pure states,
2
I INTRODUCTION
then transform these standard states into the desired pure state via LOCC or
SLOCC. In other words, it is of great importance to find one state which can be
transformed to all other pure states in a given set. We can imagine the following
scene: a quantum laboratory staff is responsible for the quantum state prepa-
ration of a company. Everyday he receives a number of different preparation
requests of multi-partite pure states. If the number of these requests are large
enough, it is impossible to keep adjusting the quantum states preparation equip-
ments. Fortunately, these produced multi-partite quantum states are shared by
his colleagues. These colleagues can communicate with each other in a classical
way and each person can make quantum operations on their own subsystems.
Now the question becomes the following: assume all required quantum states are
known, can that quantum laboratory staff only prepare one state which can be
transformed into any required quantum state via LOCC or SLOCC?
In fact, there exist a trivial kind of entangled pure states in some special state
space which can be transformed into any other pure state in the same state space
by LOCC. These kinds of states are called maximally entangled states, and they
exist in spaces if and only if the dimension of one subsystem is no less than
the product of dimensions of all other subsystems [8]. Bell states are one such
example, which can be transformed into any pure state in two-qubit systems.
For tripartite and more than three parties case, “the product of dimensions of
all other subsystems” grows fast. Based on this fact, we can immediately know:
the maximally entangled states do not exist in tripartite and more than three
parties usually. For any given multi-partite pure state set (possibly a space), can
we find a common state to prepare them via SLOCC? Such a state, if exists, will
3
I INTRODUCTION
be a common resource state for the given set of states. How much local rank of
its each subsystem should be? If these local ranks cannot be less, the state is
named an “optimal common resource ” (OCR) sate. We will present a number
of nontrivial properties of OCR states in Section V.
B. Overview of the thesis
In the next section, we review the necessary background knowledge including
Linear Algebra, Postulates of quantum mechanics, and some properties of the
local operations and classical communication protocols.
In Section III, we introduce research history about asymptotic rate of state
transformation from |GHZ〉 to |W 〉 and tensor rank of |W 〉⊗n. Then, we prove
the result that this rate is just 1. We also discuss some related problem: compu-
tational complexity of matrix permanent. Part of section III has been published
in Physical Review Letters [18].
Section IV is about how to find a common resource for a set of pure states via
local operations and classical communication. This is a new topic in quantum
information theory. We show that |GHZ3〉 = (1/√
3)(|000〉+ |111〉+ |222〉) is a
nontrivial common resource for 3-qubit systems via LOCC.
Section V is about common resource problem with stochastic local operations
and classical communication. We show some properties of the non-trivial com-
mon resource of two N -qubit pure states via SLOCC. We also study the common
resource in multi-partite space system via SLOCC.
Manuscripts about the results in Section IV and Section V are in preparation
4
I INTRODUCTION
for submitting to journals [46][47]. A brief conclusion is drawn in the last section.
5
II PRELIMINARIES
II. PRELIMINARIES
In this section, we will review the necessary background knowledge in this
research area, include the postulates and definitions about quantum mechan-
ics, theorems in linear algebra and some properties of the local operations and
classical communication protocols. Most of definitions and postulates are from
Nielsen and Chuang’s influential textbook “Quantum Computation and Quan-
tum Information” [3].
A. Linear Algebra
The study of quantum mechanics is based on a solid grasp of elementary linear
algebra. The basic objects of linear algebra are vector space. The space of all
n−tuples of complex numbers (z1, · · · , zn) is denoted as Cn.We will use the standard notation (Dirac notations) of quantum mechanics
for concepts from linear algebra. The standard quantum mechanical notation for
a vector in a vector space is as following:
|ψ〉 =
u1
u2...
un
, |φ〉 =
v1
v2...
vn
.The conjugate transpose of a vector is written as
〈ψ| = [u∗1 u∗2 · · · u∗n ] ,
where u∗ denotes the complex conjugate of a complex number u.
7
II PRELIMINARIES
Outer product of |φ〉 and |ψ〉 is |φ〉〈ψ| =
v1u
∗1 v1u
∗2 · · · v1u∗n
v2u∗1 v2u
∗2 · · · v2u∗n
......
. . ....
vnu∗1 vnu
∗2 · · · vnu∗n
,and inner product is 〈φ|ψ〉 =
∑nj=1 v
∗juj.
For a matrix A = (aj,k)m×n, T denotes the matrix transpose, say
AT = (ak,j)n×m.
The tensor product, denoted as ⊗, is a way of putting vector spaces together
to form larger vector spaces. For a matrix B = (bj,k)p×q,
A⊗B =
a11B · · · a1nBa21B · · · a2nB
.... . .
...
am1B · · · amnB
.By definition the tensor product satisfies the following basic properties, where
V and W are linear spaces:
(1) For an arbitrary scalar z and elements |v〉 of V and |w〉 of W ,
z(|v〉 ⊗ |w〉) = (z|v〉)⊗ |w〉 = |v〉 ⊗ (z|w〉).
(2) For arbitrary |v1〉 and|v2〉 in V and|w〉 in W ,
(|v1〉 ± |v2〉)⊗ |w〉 = |v1〉 ⊗ |w〉 ± |v2〉 ⊗ |w〉.
(3) For arbitrary |v〉 in V , |w1〉 and |w2〉 in W ,
|v〉 ⊗ (|w1〉 ± |w2〉) = |v〉 ⊗ |w1〉 ± |v〉 ⊗ |w2〉.
8
II PRELIMINARIES
For vectors |ψ〉 in system V ⊗ W , |v〉 of V , and identity matrix IW of W ,
〈v|ψ〉 denotes (〈v| ⊗ IW )|ψ〉, and 〈ψ|v〉 denotes 〈ψ|(|v〉 ⊗ IW ).
Suppose IV and IW are identity matrices on system V and W , we have:
(IV ⊗ EW )(EV ⊗ IW ) = EV ⊗ EW = (EV ⊗ IW )(IV ⊗ EW ).
The trace of a matrix M is defined as:
Tr (M) =∑j
〈j|M |j〉,
where |j〉 is an orthonormal basis.
B. Postulates of Quantum Mechanics
There are four postulates of quantum mechanics in quantum information the-
ory. They are the basics of the whole theory.
Postulate 1 is about how to characterize a quantum state to in a certain
system. Mathematically, it shows that we can use a unit vector to represent a
pure state.
Postulate 1: Any physical system is associated to a complex Hilbert vector
space, so-called the state space of the system. The quantum state of an isolated
system is completely described by its state vector, which is a unit vector in the
system’s state space.
A density operator of a pure state |ψ〉 is defined by ρ = |ψ〉〈ψ|. We usually
use |i〉 to denote an orthonormal basis of a Hilbert space H. For instance,
9
II PRELIMINARIES
in 2-dimensional Hilbert space H2, we usually use |0〉 and |1〉 to denote vectors
[1, 0]T and [0, 1]T . A density operator is just a positive semi-definite matrix with
trace 1.
Mathematically, Postulate 2 shows that the evolution, in the same system
after a certain time, is just a linear transformation over the state space, no
matter what the initial status of a quantum state is. The state after the
evolution is to apply this linear operator to the original state. Postulate 2
further indicates this linear operator should be unitary.
Postulate 2: The evolution of a closed quantum system is described by a
unitary transformation. That is, the state |ψ〉 of the system at time t1 is related
to the state |ψ′〉 of the system at time t2 by a unitary operator U which depends
only on the times t1 and t2,
|ψ′〉 = U |ψ〉.
In practice, we need to observe the quantum system at the right time in
order to obtain relevant information. This requires the introduction of quantum
measurement or quantum observation.
Postulate 3: Quantum measurements are represented as a set of measure-
ment operators Mm, where∑
mM†mMm = I. These operators act on the state
space of the system being measured. The index m is the ordinal number of the
measurement output after the experiment. If the state of the quantum system
is |ψ〉, then the probability of result m taking place is given by
p(m) = 〈ψ|M †mMm|ψ〉
10
II PRELIMINARIES
and after the measurement, the finial state of the system should be
Mm|ψ〉√〈ψ|M †
mMm|ψ〉.
For example, in an H2 system, all following sets are quantum measurements:
M0 = |0〉〈0|, M1 = |1〉〈1|,M0 = |0〉〈0|+ cos θ|1〉〈1|, M1 = sin θ|1〉〈1| and
M0 = (1/2)|0〉〈1|+ (√
2/2)|0〉〈0|+ (1/2)|1〉〈1|,M1 = (1/2)|0〉〈1| − (
√2/2)|0〉〈0|+ (1/2)|1〉〈1|.
In fact, for any unitary U , U †U = I and 〈ψ|U †U |ψ〉 = 〈ψ|ψ〉. Generally, we
can treat a unitary evolution as a quantum measurement with one outcome.
All above discussions are only about one sone quantum system. When more
than one subsystems have been combined as a whole system, we name those
states in this system as multi-partite states. The following postulate is the char-
acterisation of a multi-partite quantum state.
Postulate 4: The state space of a composite physical system is the tensor
product of the state spaces of the subsystems.
The simplest example is a product state. If we have n subsystems and each
subsystem has a pure state |ψj〉, 1 ≤ j ≤ n, then the composite state in the
whole system is |ψ1〉 ⊗ |ψ2〉 ⊗ · · · ⊗ |ψn〉, denoted as ⊗nj=1|ψj〉.From Postulate 4, we notice that there are not only product states in the com-
posite systems but also some other pure states: entangled states. For example,
a Bell state (1/√
2)(|00〉+ |11〉) is not a product state.
11
II PRELIMINARIES
All pure states can be denoted in this way. For example,
|GHZ〉 = (1/√
2)(|000〉+ |111〉) = [1/√
2 0 0 0 0 0 0 1/√
2]T
and
|W 〉 = (1/√
3)(|001〉+ |010〉+ |100〉) = [0 1/√
3 1/√
3 0 1/√
3 0 0 0]T .
Furthermore, we can define partial trace and reduced density operator.
Suppose |ψ〉 is a multi-partite pure state combining with subsystem A,B, · · · .Partial trace of |ψ〉 on A subsystem can be defined as
∑j〈jA||ψ〉〈ψ||jA〉, denoted
as Tr A(|ψ〉〈ψ|), where |jA〉 is an orthonormal basis for subsystem A. If |ψ〉 is a
composite state with two subsystem A and B, the reduced density operator for
subsystem A is ρB = Tr A(|ψ〉〈ψ|).The local rank of a pure state |φ〉 in some space is the rank of the result after
taking the partial trace on all the other spaces, i.e. the local rank of a 3-qubit
pure state |φ〉 in space A, denoted by rA, is the matrix rank of Tr BC(|φ〉〈φ|).Without loss of generality, we will denote the state with local rank m, n and p
as an m⊗ n⊗ p state.
For instance, |W 〉 and |GHZ〉 are both 2⊗ 2⊗ 2 states.
A Bell state (√
2/2)(|00〉+ |11〉) is a 2⊗ 2 state.
|W 〉⊗2 = (1/3)(|003〉 + |030〉 + |300〉) + (1/24)((|0〉 + |1〉 + |2〉)⊗3 + (−|0〉 −|1〉+ |2〉)⊗3 + (−|0〉+ |1〉 − |2〉)⊗3 + (|0〉 − |1〉 − |2〉)⊗3) is a 4⊗ 4⊗ 4 state.
12
II PRELIMINARIES
C. Tensor rank, Schmidt decomposition, majorization and their
applications
In this subsection, we will introduce tensor rank, Schmidt decomposition, and
majorization.
Consider an N -partite quantum state space such that the i-th subsystem
is described by a di-dimensional Hilbert space Hi, where i = 1, . . . , N . The
tensor rank rk(|ψ〉) of a state |ψ〉 ∈N⊗i=1
Hi, is defined as the smallest number
r of product states N⊗i=1
|φiα〉α=1···r, |φiα〉 ∈ Hi, whose linear span contains |ψ〉.
If we ignore the coefficient, this can also be written as rk(|ψ〉) = min r such
that |ψ〉 =r∑
α=1
N⊗i=1
|φiα〉 [9]. Then a tensor Ai,j,k corresponds to a tripartite state
|A〉 =∑
ijk aijk|ijk〉.The tensor rank of a multi-partite pure state (or tensor) characterizes the
minimal number of product states (decomposable tensors) required to linearly
represent the given state. This quantity, as a natural generalization of matrix
rank, has been playing a significant role in mathematics and computer science.
It is the key notion in determining the communication complexity and algebraic
computational complexity. For bipartite states, their tensor ranks are also named
Schmidt rank. Recently it has been gradually realized that the tensor rank is
also an indispensable tool in many problems of quantum information theory.
Indeed, the tensor rank provides a useful criterion for deciding the feasibil-
ity of entanglement transformation: to decide whether one entangled state can
be transformed into another state by using SLOCC. In certain important cases,
the tensor rank can fully characterize the feasibility of SLOCC transformations.
13
II PRELIMINARIES
This property has been investigated actively, and has led to a new entanglement
measure, namely Schmidt measure, to quantify the amount of entanglement re-
sources contained in a state. However, it was well known that the tensor ranks
of general multipartite pure states are computationally difficult to determine,
and in fact belong to so called NP-hard problems. On the other hand, it is still
possible and of great interest to determine or estimate the tensor ranks of several
classes of special quantum states that frequently appear in various applications
in quantum information theory.
For any bipartite pure state, tensor rank is also named as Schmidt rank and
is essentially reduced to matrix rank. Let us introduce Schmidt decomposition.
Theorem 1. (Schmidt decomposition) [3] Suppose |φ〉 is a pure state of compos-
ite system HA ⊗ HB. Then there exist orthonormal states |iA〉 for system A
and orthonormal states |iB〉 for system B, such that
|φ〉 =r∑i=1
√λi|iA〉|iB〉,
where λi > 0 andr∑i=1
λi = 1, these λj are so-called Schmidt values of |φ〉. Then
r is the Schmidt rank of |φ〉.
The existence of a Schmidt decomposition in this case is simply due to the
fact that any bipartite state |φ〉 can be isomorphically mapped to a matrix.
After taking the singular value decomposition of matrix, we will get a Schmidt
decomposition of the state . This result is very useful. Many important properties
of quantum systems are completely determined by the eigenvalues of the reduced
density operator of the system, so for a pure state of a composite system such
properties will be the same for both subsystems.
14
II PRELIMINARIES
Let |ψ〉 ∈ HA⊗HB⊗HC . We use ρAB to denote the reduced density operator
of |ψ〉〈ψ|, which means we take partial trace on |ψ〉 in HC . As ρAB is a positive
operator, it has a spectral decomposition ρAB =m∑k=1
pk|ψk〉〈ψk| where 0 < pk ≤ 1.
The vector span of |ψk〉 : 1 ≤ k ≤ m is called the support of ρAB and denoted
by supp(ρAB). The following lemma is very useful for determining the tensor
rank of many multipartite pure states.
Lemma 2. [5] Suppose |ψ〉 ∈ HA⊗HB⊗HC. The tensor rank of |ψ〉 is equal to
the minimum number of product states in HA ⊗HB whose linear span contains
supp(ρAB) = supp(Tr C(|ψ〉〈ψ|)).
The SLOCC transformation is of great importance. Firstly, the solution to
this problem are helpful for us to understand the weird nature of quantum en-
tanglement. Secondly, this problem plays a crucial role in understanding the
structures of LOCC and SLOCC protocols. Actually, the results will tell us what
kinds of states are useful resources for quantum (long-distance) communications.
Theorem 3. |ψ〉 can be transformed to |φ〉 via SLOCC, if and only if |ψ〉 and
|φ〉 have the relationship:
|φ〉 = ⊗Ni=1Ei|ψ〉,
where the linear operator Ei acts on space Hi only.
Let us review this fact briefly. It is easy to prove:
(IA ⊗ EB)(EA ⊗ IB) = EA ⊗ EB = (EA ⊗ IB)(IA ⊗ EB). (1)
If |φ〉 = ⊗Ni=1Ei|ψ〉, because all quantum operators Ei denote the successful local
operations, we can construct the operators Ei,√I − E†iEi one by one, which
15
II PRELIMINARIES
means, with some positive non-zero probability, we get |φ〉 from |ψ〉 by acting
Ei ⊗j 6=i Ij on |ψ〉. Also, if the SLOCC transform can be done from |ψ〉 to |φ〉,what we have to do in each space Hi is just some local operations on that space
only. We know if we let Ei denote the whole operations on space Hi, for Eq. (1),
the SLOCC transformation from |ψ〉 to |φ〉 can be denoted by |φ〉 = ⊗Ni=1Ei|ψ〉.It is easy to see that the tensor rank is an SLOCC monotone: if |ψ〉 can be
transformed into |φ〉 via SLOCC, then rk(|ψ〉) ≥ rk(|φ〉). In general the converse
is not true. However, a GHZ-equivalent state |ψGHZ〉 can be SLOCC transformed
into |φ〉, if and only if rk(ψGHZ) ≥ rk(φ).
Every LOCC protocol is just a special SLOCC protocol with success prob-
ability 1. It is a generally hard problem to determine whether a multi-partite
state can be transformed into another state. However, after proving a state |ψ〉cannot be transformed into a state |φ〉 via SLOCC, we always know that the
state |ψ〉 cannot be transformed into the state |φ〉 via LOCC.
Before jumping into the study of entanglement transformation, let us first
acquaint ourselves with a few relevant facts about majorization [2]. Majorization
is an ordering on d-dimensional real vectors intended to capture the notion that
one vector is more or less disordered than another. More precisely, suppose
x = (x1, · · · , xd) and y = (y1, · · · , yd) are two d-dimensional vectors. We use
the notation x↓ to mean x re-ordered in decreasing order, x↓1 ≥ x↓2 ≥ · · · ≥ x↓d,
for example, x↓1 is the largest component of x. x ≺ y denotes “y majorizes x”,
ifk∑j=1
x↓j ≤k∑j=1
y↓j , for k = 1, · · · , d, with equality instead of inequality when
k = d.
The connection between majorization and entanglement transformation is eas-
16
II PRELIMINARIES
ily stated yet rather surprising. Given a bipartite pure state |ψ〉, λψ denotes a
probability vector whose entries are in descending order of the Schmidt coeffi-
cients of |ψ〉. For instance, if |ψ〉 = 1√2|0〉|0〉+ 1√
6|1〉|2〉+ 1√
3|2〉|2〉, λψ = (1
2, 13, 16).
Nielsen established the following fundamental result:
Theorem 4. [2] A bipartite pure state |ψ〉 can be transformed to another pure
state |φ〉 by LOCC if and only if λψ ≺ λφ.
It is interesting that the tensor ranks of some symmetric states are related
to the polynomial ranks of homogenous polynomials, Strassen’s algorithm, and
multiplications in complex number [10] [11]. These problems may help us in
many other areas including algebraic computational complexity and quantum
information theory.
For computing certain quantum states, we mainly focus on the state |W 〉⊗n.
Also, we know some properties of states |GHZ〉⊗n.
As direct corollaries of the definition, we can readily verify the following simple
properties of tensor rank.
1. The tensor rank is subadditive under tensor product: rk(|φ〉 ⊗ |ψ〉) ≤rk(|φ〉)× rk(|ψ〉).
2. The tensor rank is nondecreasing: rk(|φ〉 ⊗ |ψ〉) ≥ maxrk(|φ〉), rk(|ψ〉).
These properties are very weak, and only can give rough estimation of tensor
ranks. In order to obtain better bounds on the tensor rank, many powerful
techniques have been introduced.
Let us now consider a more general class of symmetric states. We will in-
troduce a bijection between symmetric states and homogenous polynomials, and
17
II PRELIMINARIES
establish an one-to-one correspondence between the symmetric tensor rank and
the polynomial rank. We say anN -partite state Φ is a symmetric state if the state
is invariant if we exchange arbitrary two subsystems. For instance, |00〉+ |11〉 is
a symmetric state over H2⊗H2. Clearly, both |GHZ〉⊗n and |W 〉 are symmetric.
Suppose Psym(⊗nk=1|k〉⊗ak) is the sum of all permutation symmetric prod-
uct states of ak |k〉s in space ⊗Nj=1Hj, where ak are natural numbers and
N =∑n
k=1 ak. The Dicke state is simply a symmetrization of a product vec-
tor from a computation basis:
|D(j1 . . . jd)〉 :=
(N
j1 · · · jd
)−1/2Psym
(|1〉⊗j1 ⊗ · · · ⊗ |d〉⊗jd
). (2)
The symmetric tensor rank of symmetric states |ψ〉 ∈ ⊗Ni=1Hi, denoted by
srk(|ψ〉), is the smallest number r, such that |ψ〉 =∑r
i=1 |ψi〉⊗N , where |ψi〉 is in
Hilbert space Hi. So, for all symmetric states |ψ〉,
rk(|ψ〉) ≤ srk|ψ〉.
Let us define the asymptotic tensor rank of |φ〉 via the following way:
rk∞(|φ〉) = infn≥1
rk(|φ〉⊗n)1/n,
which equals to the number x when we use xn to approximate rk(|φ〉⊗n). One can
easily prove that “inf” can be replaced by “lim” in the above definition. From
rk(|W 〉n) ≥ 2n+1 − 1 [6], we have limn→+∞
rk(|W 〉⊗n)1/n ≥ 2.
Border rank can be defined as follows: border rank of a quantum state |φ〉 is
the minimum r such that there exists a sequence of quantum states whose limit
is |φ〉 and each of the state in the sequence with rank no more than r.
18
II PRELIMINARIES
For example, we have the following interesting equation
|WN〉 = limε→0+
(|0〉+ ε|1〉)⊗N − |0〉⊗N
ε.
Thus the border rank of |WN〉 is 2, which is much smaller than the tensor rank N .
It is interesting that the above construction can be used to provide upper bounds
for of the border ranks of Dicke states by simply taking higher derivatives.
Obviously, the border rank is always not more than the ordinary tensor rank,
that is,
brk(|φ〉) ≤ rk(|φ〉), ∀|φ〉.
So srk(|φ〉) is an upper bound of rk(|φ〉) and brk(|φ〉) is a lower bound of rk(|φ〉).The border rank of |WN〉⊗n is 2n, so we have
rk(|WN〉⊗n) ≥ brk(|WN〉⊗n) = 2n.
Note that border rank is not multiplicative, which means for |φ〉 and |ψ〉,
brk(|ψ〉 ⊗ |φ〉) ≤ brk(|ψ〉)× brk(|φ〉)
and the inequality could be made strict (consider the matrix multiplication ten-
sor).
19
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
III. ASYMPTOTIC RATE OF STATE TRANSFORMATION
Multi-partite entanglement has been widely studied [12][13][14][15] since it
is a proven asset to information processing and computational tasks. In order
to quantitatively compare between different types of quantum information re-
sources, the following fundamental entanglement transformation problem arises:
whether a pure N -partite state |ψ〉 can be transformed into another given N -
partite state |φ〉 via LOCC. Nielsen solved all bipartite cases [2]. After that,
multiple-copy entanglement transformation was studied: Duan et al. proved that
entanglement-assisted transformation of bipartite case is asymptotically equiv-
alent to multiple-copy transformation [16]; in a multipartite setting, Ji et al
showed that the entanglement transformation rate between any two genuinely
entangled states are positive, that is, it is always feasible to exactly transform a
genuinely N -partite entangled pure state with sufficient many but a finite num-
ber of copies to any other N -partite state by LOCC [20], where a multipartite
pure entangled state is said to be genuine if it is not in a product form between
any bipartite partition of the parties.
One of the major difficulties in evaluating the entanglement transformation
rate of multipartite case is that the class of LOCC is still not satisfactorily
understood. Another one is the richness of multipartite entanglement. Gen-
erally, there exist incomparable states, even in three-qubit systems. It is still
unclear how to determine whether one multipartite state can be transformed
to another by LOCC. To partially remedy these obstacles, we relax the restric-
tion of LOCC and consider SLOCC [4, 9, 17, 19, 21]. The ability to trans-
form a state |ψ〉 to another state |φ〉 with SLOCC is symbolically expressed as
21
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
|ψ〉 SLOCC−→ |φ〉. The physical meaning of SLOCC operations is that they can be
implemented by LOCC operations with nonzero probability. In fact, SLOCC has
been used to study entanglement classification [4, 27] and entanglement trans-
formation [5–7]. The whole multi-partite state space can be divided into SLOCC
equivalence classes. For instance, Dur et al observed that within three qubit sys-
tems, there exist two distinct equivalence classes of genuinely tripartite entangled
states,|GHZ〉 = 1√2(|000〉+ |111〉) and |W 〉 = 1√
3(|100〉+ |010〉+ |001〉) [4].
Comparing with LOCC entanglement transformation, SLOCC entanglement
transformation of pure states has a much simpler mathematical structure that
can be directly characterized. In order to consider the asymptotic SLOCC entan-
glement transformation between pure states, we only need to deal with SLOCC
equivalence classes. By employing the concept of tensor rank, which is defined
as the smallest number of product states whose linear span contains the given
state, many interesting results are obtained. For three-qubit systems, it was
showed that 3 copies of GHZ-state can be transformed into 2 copies of W -state
[5]. In [6], we proved that |GHZ〉⊗m SLOCC−→ |W 〉⊗2n is valid if 2m ≥ 7n. Later,
it was demonstrated that SLOCC protocol can transform 4 copies of GHZ-state
to 3 copies of W -state [7]. These increasing lower bounds reflect both the rich-
ness of entanglement and the difficulty of obtaining asymptotic results. These
progresses motivate us to introduce a useful notion of the SLOCC entanglement
transformation rate in the following way:
R(|ψ〉, |φ〉) = supmn
: |ψ〉⊗n SLOCC−→ |φ〉⊗m.
This quantity intuitively characterizes the optimal number of copies of |φ〉 one
can obtain from a single copy of |ψ〉 under SLOCC, in an asymptotic setting.
22
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
Therefore, it is of great interest to determine this value by studying the highest
possible rate for the multi-copy transformations. Unfortunately, R(|ψ〉, |φ〉) is
not easy to calculate, even for the simplest non-trivial case, R(|GHZ〉, |W 〉),whose exact value was conjectured to be 1 [6, 7].
A. Previous results about R(|GHZ〉, |W 〉)
By the definition of asymptotic tensor rank and the fact that |GHZ〉⊗n can
be transformed to any pure state with lower tensor rank via SLOCC, we have:
Theorem 5.
R(|GHZ〉, |φ〉) = (log2 rk∞(|φ〉))−1.
Proof. |GHZ〉⊗n SLOCC−→ |φ〉⊗m, if and only if rk(|GHZ〉⊗n) ≥ rk(|φ〉⊗m).
rk(|GHZ〉⊗n) = 2n and rk∞(|φ〉) = infn≥1 rk(|φ〉⊗n)1/n (the definition of
asymptotic tensor rank, see Chapter II).
R(|GHZ〉, |φ〉) = supmn
: |GHZ〉⊗n SLOCC−→ |φ〉⊗m
= supγ : 2m/γ ≥ rk(|φ〉⊗m)
= supγ : 21/γ ≥ rk(|φ〉⊗m)1/m
= supγ : γ ≤ (log2 rk(|φ〉⊗m)1/m)−1
= (log2 infm≥1
(rk(|φ〉⊗ m)1/m)−1
= (log2 rk∞(|φ〉))−1
ut
23
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
For multi-partite cases (more than three parties), Schmidt decomposition of
general states also does not exist. The calculation of tensor becomes extremely
complicated. We have to consider very special states.
rk(|WN〉) = N , where |WN〉 is an N -qubit generalization of |W 〉, i.e.,
|WN〉 = |0〉⊗N−1|1〉+ |0〉⊗N−2|1〉|0〉+ · · ·+ |1〉|0〉⊗N−1.
We can prove a stronger version:
Lemma 6. For any complex number λ, rk(|WN〉+ λ|0〉⊗N) = N .
Before proceed to the proof, let us point the a very useful fact: |WN〉 and
|WN〉+ λ|0〉⊗N are SLOCC equivalent. Hence they have the same rank. As it is
clear that rk(|WN〉) ≤ N , we also have
rk(|WN〉+ λ|0〉⊗N) ≤ N.
In the following we only need to show that N is a lower bound.
Proof. For N = 1, 2, the result trivially holds. Assume that we have proven the
case of N ≤ T , say rk(|Wm〉 + |0〉⊗m) = m for any m ≤ T . Let us consider the
case of N = T + 1.
Let us prove that T + 1 is a lower bound by contradiction. If there exist
product states |φj〉 ∈ H1 ⊗ · · · ⊗ HT+1, 1 ≤ j ≤ T , s.t.
|WT+1〉+ λ|0〉T =T∑j=1
|φj〉.
More precisely, assume |φj〉 = |ψj1〉 ⊗ · · · ⊗ |ψj(T+1)〉 and |ψji〉 ∈ Hi, then
|WT+1〉+ λ|0〉⊗T+1 =T∑j=1
|ψj1〉 ⊗ · · · ⊗ |ψj(T+1)〉.
24
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
After checking the space HT+1, we can see there exists j0, s.t. |ψj0(T+1)〉 6= |0〉.There is no harm for us to assume that j0 = T . Then, ∃|ΦT+1〉 ∈ HT+1, s.t.
〈ΦT+1|ψT (T+1)〉 = 0 and |ΦT+1〉 6= |1〉. So, 〈ΦT+1|0〉 6= 0. Now,
〈ΦT+1|(|WT+1〉+ λ|0〉⊗T+1〉) = 〈ΦT+1|0〉|WT 〉+ (λ〈ΦT+1|0〉+ 〈ΦT+1|1〉)|0〉⊗T .
Because of the hypothesis, the rank of this state should be T. However,
〈ΦT+1|(|WT+1〉+ λ|0〉⊗T+1〉)
= 〈ΦT+1|(T∑j=1
|ψj1〉 ⊗ · · · ⊗ |ψj(T+1)〉)
=T∑j=1
〈ΦT+1|ψj(T+1)〉|ψj1〉 ⊗ · · · ⊗ |ψj(T )〉
=T−1∑j=1
〈ΦT+1|ψj(T+1)〉|ψj1〉 ⊗ · · · ⊗ |ψj(T )〉.
This tensor rank of the above state is strictly less than T , thus contradicts
our hypothesis for the case of T . That completes the proof. ut
For Lemma 6, we have rk(|W3〉) = 3. Hence,
R(|GHZ〉, |W 〉) = (log2 rk∞(|φ〉))−1 ≥ (log2 3)−1 ≈ 0.63093.
Let us abbreviate |W3〉 to |W 〉. We also know that
rk(|W 〉⊗2) = 7.
This is because |W 〉⊗2 = |007〉+ |070〉+ |700〉+1
4((|0〉+ |1〉+ |2〉)⊗3
+ (|0〉 − |1〉 − |2〉)⊗3 + (−|0〉+ |1〉 − |2〉)⊗3 + (−|0〉 − |1〉+ |2〉)⊗3)and rk(|W 〉⊗n) ≥ 2n+1 − 1 [6].
This implies
R(|GHZ〉, |W 〉) ≥ (log2
√7)−1 ≈ 0.7124.
25
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
Let us show rk(|W3〉⊗3) ≤ 16. We have
|W3〉⊗3
= Psym(|0〉⊗2|7〉) + Psym(|1〉|2〉|4〉)+ Psym(|0〉|1〉|6〉) + Psym(|0〉|2〉|5〉) + Psym(|0〉|4〉|3〉)= Psym(|0〉⊗2|7〉) +
1
4(|1〉+ |2〉+ |4〉)⊗3
+ Psym(|0〉 (|3〉+ |5〉) (|2〉+ |4〉 − |1〉)) +1
4(|2〉+ |4〉 − |1〉)⊗3
+ Psym(|0〉 (|3〉+ |6〉) (|1〉+ |4〉 − |2〉)) +1
4(|1〉+ |4〉 − |2〉)⊗3
+ Psym(|0〉 (|6〉+ |5〉) (|2〉+ |1〉 − |4〉)) +1
4(|2〉+ |1〉 − |4〉)⊗3.
Since we have the following nice result from [32]:
rk(Psym(|a〉|b〉|c〉) +1
4|a〉⊗3) ≤ 4,
we can readily figure out that
rk(Psym(|0〉 (|3〉+ |5〉) (|2〉+ |4〉 − |1〉)) + (|2〉+ |4〉 − |1〉)3) ≤ 4.
Hence, we obtain the following result from [7]:
rk(|W3〉⊗3) ≤ 16.
This means
R(|GHZ〉, |W 〉) ≥ (log23√
16)−1 = 0.75.
By applying the techniques of polynomial ranks, we can further obtain the
following results reported in [7]:
rk(|WN〉⊗n) ≥ (N − 1)2n −N + 2, rk(|WN〉⊗2) = 3N − 2.
26
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
B. R(|GHZ〉, |W 〉)=1
In this section, we prove the validity of the above conjecture by constructing an
SLOCC transformation from n+o(n) copies of GHZ-state to n copies of W -state,
that is, one can obtain 1 copy of W -state, from 1 copy of GHZ-state by SLOCC,
asymptotically. To reach our goal, we introduce a class of tripartite states which
an n-copiesW state can be written into the sum of at most n2 items of. Each state
of this class can be obtained by applying SLOCC operations on an n-copiesGHZ-
state. Then, we show that R(|GHZ〉N , |W 〉N), the entanglement transformation
rate, is also 1 forN -partite state |GHZ〉N = 1√2(|00 · · · 0〉+|11 · · · 1〉) and |W 〉N =
1√N
(|0 · · · 01〉+ · · ·+ |10 · · · 0〉).Our first result is the following,
Theorem 7. For three-qubit state system, we have
R(|GHZ〉, |W 〉) = 1.
That is, for sufficient large n, one can transform n + o(n) copies of GHZ-state
to n copies of W -state by SLOCC. An immediate consequence is that the GHZ
state is asymptotically stronger than the W state under SLOCC, although they
are incomparable at the single copy level.
Generally, for N-partite systems,
R(|GHZ〉N , |W 〉N) = 1.
Again one can obtain |W 〉N from |GHZ〉N at a rate 1 by SLOCC.
For convenience, in the following discussions we omit an unimportant normal-
ized factor and denote directly |W 〉 = |100〉 + |010〉 + |001〉. Before proving the
27
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
validity of Theorem 7 for three-qubit systems, we first introduce a class of tri-
partite states, |[a, b, c]〉n for any triple (a, b, c) of nonnegative integers such that
a+ b+ c = n. Let B be the following set
B := (a, b, c) : a+ b+ c = n, 0 ≤ a, b, c ≤ n.
For any (a, b, c) ∈ B, one can define an unnormalized state
|[a, b, c]〉n =∑
i⊕j⊕k=(11···1)n,i∈A(a),j∈A(b),k∈A(c)
|i〉|j〉|k〉,
where ⊕ is the bitwise addition modulo 2, (11 · · · 1)n stands for the n−bit string
with ‘1’ in all n positions, and A(·) represents the set of the n−bit strings with
the same Hamming weight, i.e.,
A(l) = i : h(i) = l, i ∈ Zn2,
where Z2 = 0, 1 and the Hamming weight h(i) of an n-bit string i simply
represents the number of ‘1’s in the string.
We can verify the following equation,
|W 〉⊗n =∑
(i,j,k)∈S
|i〉|j〉|k〉 =∑
(a,b,c)∈B
|[a, b, c]〉n, (3)
where i, j, k are n-bit strings, and S is the subset of Zn2 × Zn2 × Zn2 ,
S = (i, j, k) : i⊕ j ⊕ k = (11 · · · 1)n, itjtkt = 0 for any t,
with it the tth bit of i. Namely, S is the set of (i, j, k) such that on each 1 ≤ t ≤ n,
there is only a single ‘1’ in the t-th bit of i, j, and k. The first equality in Eq. (3)
28
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
follows by calculating |W 〉⊗n in computational basis, while the second equality
follows by observing that
S = (i, j, k) : i⊕ j ⊕ k = (11 · · · 1)n, h(i) + h(j) + h(k) = n.
The following property of tripartite states |[a, b, c]〉n is extremely useful in
proving Theorem 1,
Lemma 8. Any state |[a, b, c]〉n can be obtained from |GHZ〉⊗n by SLOCC.
Proof:— To see the validity of this lemma, we need the following identity,
|[a, b, c]〉n =1
2n
2n−1∑l=0
(∑i∈A(a)
(−1)l·i|i〉)⊗ (∑j∈A(b)
(−1)l·j|j〉)⊗ (∑k∈A(c)
(−1)l·k|k〉).
Here i = (11 · · · 1)n ⊕ (i1i2 · · · in), and l · i is the bitwise inner product of two
n−bit strings l and i. The above identity can be verified by a direct calculation.
We construct three 2n × 2n matrices
E =2n−1∑l=0
∑i∈A(a)
(−1)l·i|i〉〈l|,
F =2n−1∑l=0
∑j∈A(b)
(−1)l·j|j〉〈l|,
G =2n−1∑l=0
∑k∈A(c)
(−1)l·k|k〉〈l|,
Now, it is direct to verify that
(E ⊗ F ⊗G)|GHZ〉⊗n = 2n/2|[a, b, c]〉n,
29
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
where we have assumed that
|GHZ〉⊗n =1
2n/2
2n−1∑l=0
|l〉|l〉|l〉.
That is, E ⊗ F ⊗G transforms |GHZ〉⊗n to |[a, b, c]〉n.
Back to the proof of Theorem 7,
Proof of Theorem 7:— Let |B| be the cardinality of B. Noticing that
|B| =(n+ 2
2
)= O(n2),
we conclude that
|GHZ〉⊗m SLOCC−→ |W 〉⊗n,
holds for all 2m ≥(n+22
)2n. Therefore,
R(|GHZ〉, |W 〉) = 1.
This method can also be used to show that for any N > 1,
|GHZ〉⊗mNSLOCC−→ |W 〉⊗nN ,
holds when 2m ≥(n+N−1N−1
)2n.
This lead us to the fact that
R(|GHZ〉N , |W 〉N) ≥ 1.
On the other hand, it is known that the tensor rank of |W 〉⊗nN is no less than
(N − 1)2n −N + 2 [7], which impies
R(|GHZ〉N , |W 〉N) ≤ 1.
Combining with these results, we know that R(|GHZ〉N , |W 〉N) = 1.
30
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
C. Generalization to Dicke states
Both |W 〉 and |GHZ〉 are symmetric states, i.e., those invariant under any
permutation of its parties. Studying the entanglement measure of such states
has attracted a lot of attention[22, 28]. The entanglement transformation prop-
erties of such states are studied [7, 41] and some families of multi-qubit SLOCC
equivalent states are realized by using symmetric states [42, 43].
In order to generalize Theorem 1 to Dicke states, we need evaluate the entan-
glement transformation rate of Dicke states by SLOCC. A general lower bound
is given as follows.
Theorem 9. For Dicke state |D(j1, · · · , jd)〉 with j1 ≥ · · · ≥ jd and N =∑d
i=1 ji,
R(|GHZ〉N , |D(j1, · · · , jd)〉) ≥ (d∑i=2
log2 (ji + 1))−1.
The bound is tight for Dicke state with j1 ≥∑d
i=2 ji.
Proof. The proof of the lower bound part is the direct generalization of Theorem
7.We prove that:
|GHZ〉⊗mNSLOCC−→ |D(j1, j2, · · · , jd)〉⊗n
holds if
2m ≥d∏
k=2
(njk +N − 1
N − 1
)(jk + 1)n.
Before presenting the proof, we introduce some notations: T is used to denote
|1〉, |2〉, · · · , |d〉⊗n, where the tensor product S1 ⊗ S2 of two sets S1 and S2 is
31
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
defined as |s1〉 ⊗ |s2〉 : |si〉 ∈ Si. Now we associate any |α〉 ∈ T with a d − 1
dimensional vector of natural numbers ~v = (l2, l3, · · · , ld) iff the number of |k〉appearing in |α〉 is lk for all 2 ≤ k ≤ d, ~v is called the characteristic vector of
|α〉, written as C(|α〉) = ~v.
Now we divide T into disjoint subsets A~v according to their characteristic
vector: A~v = |α〉 : C(|α〉) = ~v.Define the B as the following set
B = (~v1, ~v2, · · · , ~vd) : ~v1 + ~v2 + · · ·+ ~vd = (nj2, nj3, · · · , njd),
where ~lk are all vectors of natural numbers.
Now we decompose |D(j1, j2, · · · , jd)〉⊗n according to the computational basis,
and rearrange the elements according to characteristic vectors as follows
|D(j1, j2, · · · , jd)〉⊗n =∑
(~v1,~v2,··· ,~vd)∈B
|[~v1, ~v2, · · · , ~vd]〉n.
Note that here |[~v1, ~v2, · · · , ~vd]〉n is not the superposition of all |α1〉⊗|α2〉 · · ·⊗|αd〉with C(|αk〉) = ~vk, we also require that |α1〉 ⊗ |α2〉 · · · ⊗ |αd〉 does appear in the
decomposition of |D(j1, j2, · · · , jd)〉⊗n.
By noticing that |B| is polynomial of n, we only need to show for any
(~v1, ~v2, · · · , ~vd) ∈ B with 2m ≥d∏
k=2
(jk + 1)n
|GHZ〉⊗mNSLOCC−→ |[~v1, ~v2, · · · , ~vd]〉n.
This is proved by verifying the following equation with wt := e2πit , µ(L) :=
d∏k=2
w∑ni=1 lk,i
jk+1 , and f(α,L) :=n∏i=1
wlsi,ijsi+1 for |α〉 = |s1〉 ⊗ |s2〉 · · · ⊗ |sn〉 with si ∈
32
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
1, 2 · · · d,|[~v1, ~v2, · · · , ~vd]〉n = 1
d∏k=2
(jk+1)n
∑jdld,1=0···
ld,n=0
· · ·∑j2
l2,1=0···
l2,n=0
µ(L) ·
(∑
C(|α1〉)=~v1f(α1, L)|α1〉)⊗ · · · ⊗ (
∑C(|αd〉)=~vd f(αd, L)|αd〉). ut
In order to show the tightness of the above bound for Dicke state with j1 ≥Πdi=2(ji + 1), we regard |D(j1, · · · , jd)〉 as a bipartite state |ψ〉 by arranging the
first r parties of |D(j1, · · · , jd)〉 into a single party, say Alice, and the rest N − rparties into another single party, Bob, where r = bN/2c. From the definition,
we know that the tensor rank of |D(j1, · · · , jd)〉 is not less than that of |ψ〉.Now, we apply local operator M⊗N on |ψ〉, where M maps |α1〉|α2〉 · · · |αr〉 into
|1〉⊗µ1|2〉⊗µ2 · · · |d〉⊗µd , where µi are the multiplicity of i among α1, α2, · · · , αr.The definition of N is similar.
Observe that the tensor rank of (MA ⊗NB)|ψ〉 equals to f which is the car-
dinality of the following set,
(β1, β2, · · · , βd) : 0 ≤ βi ≤ ji,d∑i=1
βi = r.
For Dicke state with j1 ≥∑d
i=2 ji, we observe that j1 ≥ N/2 ≥ r. Thus the
constraint 0 ≤ β1 ≤ j1 in the above set is automatically satisfied and the cardi-
nality is totally determined by other βi such that i ≥ 2. By a simple counting
arguments we know the cardinality is given by Πdi=2(ji + 1), which is also the
tensor rank of the bipartite state. Noticing that the tensor rank of bipartite
pure state is multiplicative, and tensor rank is a strictly non-increasing quantity
33
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
under SLOCC, we conclude that
R(|GHZN〉, |D(j1, · · · , jd)〉) ≤ (d∑i=2
log2 (ji + 1))−1,
for j1 ≥ Πdi=2(ji + 1). Thus, (
∑di=2 log2 (ji + 1))−1 is the tight bound for such
Dicke state.
By applying the above technique to the N−partite state |D(1, · · · , 1)〉, we
have the following result about the SLOCC transformation,
Lemma 10. The tensor rank of |D(1, · · · , 1)〉 is not less than(
NbN/2c
). Thus,
|GHZN〉⊗mSLOCC−→ |D(1, · · · , 1)〉 =⇒ 2m ≥
(N
bN/2c
),
Together with the known result that the tensor rank of |D(1, · · · , 1)〉 is not
more than ≤ 2N−1 from [33], we can conclude that its tensor rank is 2N(1+o(1)).
D. Alternative proof
In fact, there exists a easier way to prove R(|GHZ〉, |W 〉) = 1. The following
result was essentially by Bini [35].
Theorem 11. [35] If an N-partite tensor|ψ〉 can be written as |ψ〉+s∑j=1
λj|φj〉 =
b∑j=1
|f(j, λ)〉⊗N , then
rk(|ψ〉⊗n) ≤ (1 + s n)brk(|ψ〉)n.
Let us use this theorem in our case and show the decomposition of |W3〉⊗n in
detail.
34
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
Theorem 12. rk (|W3〉⊗n) ≤ (n+ 1)2n.
Proof. ∀λ 6= 0, |W3〉+ λ2|111〉 = 12λ
[(|0〉+ λ|1〉)⊗3 − (|0〉 − λ|1〉)⊗3].Due to non-zero Vandermonde determinant, for any nonzero pairwise distinct
λk, 1 ≤ k ≤ n+1, we can always have the solutions µj, 1 ≤ j ≤ n+1, such that1 1 · · · 1
λ21 λ22 · · · λ2n+1
......
. . ....
λ2n1 λ2n2 · · · λ2nn+1
µ1
µ2
...
µn+1
=
1
0...
0
.Then, let us show show the decomposition of |W3〉⊗n:
|W3〉⊗n
= |W3〉⊗n + 0×n−1∑j=0
Psym(|W3〉⊗j|111〉⊗n−j)
= |W3〉⊗nn+1∑k=1
µk +n−1∑j=0
Psym(|W3〉⊗j|111〉⊗n−j)n+1∑k=1
µkλ2n−2jk
=n+1∑k=1
n∑j=0
µkλ2n−2jk Psym(|W3〉⊗j|111〉⊗n−j)
=n+1∑k=1
µk(|W3〉+ λ2k|111〉)⊗n
=n+1∑k=1
µk 12λk
[(|0〉+ λk|1〉)⊗3 − (|0〉 − λk|1〉)⊗3]⊗n.
Hence, rk (|W3〉⊗n) ≤ (n+ 1)2n. ut
For Dicke state |Ψ〉 = |D(j1, · · · , jd)〉 with j1 ≥ · · · ≥ jd and N =∑d
i=1 ji,
brk(|D(j1, · · · , jd)〉) ≤∏d
i=2(ji + 1) [32]. We can also use Theorem 11 to obtain
that an upper bound of rk(|Ψ〉⊗n) is O((∏d
i=2(ji + 1))n).
35
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
E. Computational complexity of matrix permanent
The motivation for studying the tensor rank of |D(1, · · · , 1)〉 is the connec-
tion between its tensor rank and that of matrix permanent, a homogeneous
polynomial. It is worth noting that tensor rank of homogeneous polynomial has
already been extensively studied in algebraic complexity theory [30, 32, 36–40].
A homogeneous polynomial is a multi-variables polynomial whose nonzero terms
(monomials) all have the same degree.The tensor rank of a homogeneous poly-
nomial P (x1, · · · , xn) is defined as the smallest number r such that P (x1, ...xn)
can be written as the sum of r terms of Li(x1, · · · , xn), where each Li(· · · ) is the
product of d homogeneous linear forms with d the degree of the polynomial.
The permanent of matrix X = (xi,j)N×N is defined as
perm(X) =∑σ∈Sn
N∏i=1
xi,σ(i).
The sum here extends over all elements σ of the symmetric group SN ; i.e. over
all permutations of the numbers 1, 2, · · · , N .
It is easy to see that the tensor rank of matrix permanent is defined as the
minimum number rk such that
perm(X) =rk∑j=1
N∏i=1
Li,j(X),
where Li,j(X) is a linear function of X and Li,j(0) = 0.
The tensor rank of matrix permanent is still unknown, and it relates to the
central problem of computational complexity theory–circuit lower bounds. One
possible direction to study this problem is to restrict the form of∏N
i=1 Li,j(X), for
36
III ASYMPTOTIC RATE OF STATE TRANSFORMATION
instance, assume that∏N
i=1 Li,j(X) satisfy multilinear property, see [38–40]. In
this case, it is known that the tensor rank of matrix permanent is lower bounded
by 2NΩ(1)
, where Ω(1) is some non-zero constant.
For a more restricted form, each Li,j(X) only depends on the i-th row of X
and j, we can obtain a better lower bound as follows.
Theorem 13. Let X be an N ×N matrix with permanent
perm(X) =
k(N)∑j=1
N∏i=1
N∑k=1
a(j)i,kxi,k.
Then k(N) ≥(
NbN/2c
).
Proof:— Indeed, since each Li,j(X) only depends on the i-th row of X and j,
we can easily verify that
|D(1, · · · , 1)〉 =
k(N)∑j=1
N∏i=1
(N∑k=1
a(j)i,k |k〉).
The proof is completed by applying Lemma 10.
Most part of Section III is the results form our publication [18].
37
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
IV. COMMON RESOURCE VIA LOCAL OPERATIONS AND
CLASSICAL COMMUNICATION
Given a set of multipartite entangled states, can we find a common state to
prepare them by local operations and classical communication? Such a state, if
exists, will be a common resource for the given set of states. We completely solve
this problem for bipartite pure states case by explicitly constructing a unique
optimal common resource state for any given set of states. In the multipartite
setting, the general problem becomes quite complicated, and we focus on finding
nontrivial common resources for the whole multi-partite state space of given
dimensions. We show that |GHZ3〉 = 1/√
3(|000〉+ |111〉+ |222〉) is a nontrivial
common resource for 3-qubit systems. We also show some properties of the non-
trivial common resource of two N -qubit pure states. We will also study the
common resource in multipartite space system.
The problem of transforming one entangled state to another one by LOCC
is of central importance in quantum entanglement theory. The majorization
characterization can be extended to a class of multipartite pure states having
Schmidt decompositions [2] [31]. Unfortunately, the Schmidt decomposition for
a generic multipartite pure state does not exist. It is still an open problem to
determine whether a general multi-partite pure state can be transformed into
another one by LOCC.
Despite the overall complexity of multipartite entanglement transformations,
we can often find entangled states that can be transformed into any other state
in the same state space by LOCC. These kind of states are called maximally
entangled states, and they exist in spaces if and only if the dimension of one
39
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
subsystem is no less than the product of dimensions of all other subsystems [8].
Bell states are one such example, which can be transformed into any pure state in
two-qubit systems. As a simple corollary of the dimensionality bound, there is no
maximally entangled state in three-qubit systems. In fact, it is well-known that
three-qubit states can be entangled in two different ways, one class consisting of
so-called W-type states and other consisting of GHZ-type states [4].
In this section, we generalize the notion of maximally entangled states
with respect to LOCC transformations. The problem we study can be best
described through the following scenario. Assume that Alice and Bob are
going to implement a series of quantum information tasks, each one requiring a
different entangled state to perform. However, instead of sharing a multitude of
different states, they wish to share only one type of entangled state and then
transform this state into a different form as needed. So the question is: for a
given set of pure entangled states, is there a certain state which can be locally
transformed into all of them by LOCC? Below, we give a complete solution
to this problem for bipartite systems. In addition, we study the case when
the set of target states is the entire state space for some given dimensions.
When the dimension of one subsystem in this target space is not smaller than
the product of all the other subsystem dimensions, there always exists some
(perhaps higher-dimensional) state that can obtain all states in the target
space [8]. However, this dimensionality condition turns out not to be necessary.
Interestingly, we find a non-trivial state |GHZ3〉 which can be transformed into
any three-qubit pure state by LOCC.
40
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
Let us now formulate our problem more precisely. Let S = |ψ1〉, |ψ2〉, · · · be a set of (multipartite) states, possibly infinite. A common resource state |ψ〉to S can be transformed into any state in S by LOCC. We say |ψ〉 an optimal
common resource (OCR) if for any other common resource |φ〉 we have either |φ〉can be transformed into |ψ〉 by LOCC, or |φ〉 and |ψ〉 are not comparable under
LOCC.
A. Optimal common resource of bipartite pure states
In general, it is a hard problem to find OCR for a set of multipartite states.
For bipartite pure states, majorization characterizes the LOCC transformation
between two pure state [2].
Nielsen’s result together with the properties of majorization leads us to an
explicit construction of the “unique’’ OCR of a set of bipartite pure states.
Theorem 14. Let S = |φi〉, i ∈ I be a set of d⊗ d pure states, where I is an
index set (finite or infinite). Assume that the Schmidt coefficient vector of |φi〉is given by λφi = (x
(i)1 , · · · , x
(i)d ). Then if I is a finite set, the OCR state |ψ〉 for
S always exists and λψ is unique. |ψ〉 is given by λψ = (y1, · · · , yd), where
yk = mini∈I
k∑j=1
x(i)j −min
i∈I
k−1∑j=1
x(i)j .
Furthermore, if I is infinite, the min sign in above equations should be replaced
with inf.
Let us consider an example to demonstrate the application of the above
theorem. Let the d−dimensional bipartite target set be Sa = |φ〉|λφ1 ≥ a
41
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
, where a ≥ 1/d. Then an OCR |ψ〉 for Sa can be chosen as |ψ〉 such that
λψ = (a, 1−ad−1 ,
1−ad−1 , · · · ,
1−ad−1). The maximal entangled state 1√
d
d−1∑k=0
|kk〉 is always
a common resource but usually not optimal.
According to Nielsen’s majorization criterion for entanglement transforma-
tion, Theorem 14 is essentially due to the following lemma:
Lemma 15. Suppose that X(k) = (x(k)1 , x
(k)2 , · · · , x(k)d are a set of d–
dimensional vectors where x(k)1 ≥ x
(k)2 ≥ · · · ≥ x
(k)d . There always ex-
ists an optimal vector Y = (y1, · · · , yd) such that Y ≺ X(k) for
any k. Furthermore, Y = (y1, · · · , yd) can be chosen as yk =
min(k∑j=1
x(1)j ,
k∑j=1
x(2)j , · · · ,
k∑j=1
x(n)j ) − min(
k−1∑j=1
x(1)j ,
k−1∑j=1
x(2)j , · · · ,
k−1∑j=1
x(n)j ). Further-
more, if n is infinite, yk = inf(k∑j=1
x(h)j |h = 1, 2 · · · )− inf(
k−1∑j=1
x(h)j |h = 1, 2 · · · ).
Proof. Let us first consider the case that n is finite.
We will complete the proof via the following 3 steps:
Step 1: yk ≥ yk+1.
Suppose min(k∑j=1
x(1)j ,
k∑j=1
x(2)j , · · · ,
k∑j=1
x(n)j ) =
k∑j=1
x(t)j . We have,
yk+1
= min(k+1∑j=1
x(1)j ,
k+1∑j=1
x(2)j , · · · ,
k+1∑j=1
x(n)j )−min(
k∑j=1
x(1)j ,
k∑j=1
x(2)j , · · · ,
k∑j=1
x(n)j )
≤ x(t)j+1 ≤ x
(t)j
≤ min(k∑j=1
x(1)j ,
k∑j=1
x(2)j , · · · ,
k∑j=1
x(n)j )−min(
k−1∑j=1
x(1)j ,
k−1∑j=1
x(2)j , · · · ,
k−1∑j=1
x(n)j )
= yk.
Notice that:k∑j=1
yj = min(k∑j=1
x(1)j ,
k∑j=1
x(2)j , · · · ,
k∑j=1
x(n)j ),
42
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
so ∀k,∃f(k),k∑j=1
yj =k∑j=1
x(f(k))j .
Step 2: ∀m, Y ≺ X(m).
It is obvious to see thatk∑j=1
yj ≤k∑j=1
x(m)j .
Step 3: If ∀k, Z ≺ X(k), then Z ≺ Y.
Otherwise, if ∃k0,k0∑j=1
zj>k0∑j=1
yj.
k0∑j=1
zj>k0∑j=1
yj =k0∑j=1
x(f(k0))j .
This is a contradiction against with Z ≺ X(f(k0)).
If n is infinite, we modify our proof as follows:
Notice that: ∀k,m,k∑j=1
yj = inf(k∑j=1
x(h)j ) ≤
k∑j=1
x(m)j .
Step 1: yk ≥ yk+1.
Suppose inf(k∑j=1
x(h)j ) = lim
h→∞
k∑j=1
x(g(h))j , where
k∑j=1
x(g(h))j is in descending or-
der. We can also find a sub–sequence (f(h)) ⊂ sequence(g(h)), such that
inf(k∑j=1
x(h)j ) = lim
h→∞
k∑j=1
x(f(h))j , also lim
h→∞x(f(h))k and lim
h→∞x(f(h))k+1 exist. This is be-
cause any infinite bounded sequence must have a monotonic convergent sub–
sequence with a limit.
yk+1 = inf(k+1∑j=1
x(h)j )− inf(
k∑j=1
x(h)j ) = inf(
k+1∑j=1
x(h)j )− lim
h→∞
k∑j=1
xf(h)j
≤ limh→∞
k+1∑j=1
xf(h)j − lim
h→∞
k∑j=1
xf(h)j
≤ limh→∞
xf(h)k+1 ≤ lim
h→∞xf(h)k
≤ limh→∞
k∑j=1
xf(h)j − inf(
k−1∑j=1
x(h)j )
43
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
= inf(k∑j=1
x(h)j )− inf(
k−1∑j=1
x(h)j )
= yk.
Step 2: ∀m, Y ≺ X(m).
Step 3: If ∀k, Z ≺ X(k), then Z ≺ Y.
Otherwise, if ∃k0,k0∑j=1
zj>k0∑j=1
yj.
k0∑j=1
zj>k0∑j=1
yj = inf(k∑j=1
x(h)j ).
This contradicts the fact that ∀k, Z ≺ X(k). ut
B. Optimal common resource of a special kind of multi-partite pure
states
Xin and Duan pointed out that Nielsen’s majorization result can be extended
to a special class of multi-partite pure states with a generalized Schmidt decom-
positions in 2007 [31].
Theorem 16. [31] Suppose n people,Alice, Bob, ..., and Dana, share an n-partite
pure state |ψ〉 which has a generalized Schmidt decomposition as follows:
|ψ〉 =r∑
k=1
√xk
n⊗j=1
|k〉j, (4)
where for any 1 ≤ j ≤ n, |k〉j|1 ≤ k ≤ r is an orthonormal basis in the j-th
subsystem, and λ = (x1, . . . , xr) represents the Schmidt coefficient vector with
non-increasing order. These people want to transform |ψ〉 to the following state
44
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
|φ〉 by LOCC:
|φ〉 =r∑
k=1
√yk
n⊗j=1
|k′〉j. (5)
|ψ〉 can be transformed to |φ〉 via LOCC, if and only if λψ ≺ λφ.
We further generalize this theorem to the following case: for a⊗n
j=1 dj quan-
tum system, we say an orthogonal ensemble of states E = |Ψi〉N−1k=0 is locally
permutation invariant if there exists local unitary representation⊗n
j=1 U(g)j
of the permutation group SN such that
eiφgkn⊗j=1
U(g)j |Ψk〉 := |Ψσg(k)〉 ∈ E , ∀g ∈ Sn and arbitrary phases φgk.
Theorem 17. Consider two tripartite states |ψ〉 =∑N−1
k=0
√µk|k〉|Ψk〉 and
|ψ〉 =∑N−1
k=0
√λk|k〉|Ψk〉, where the |Ψk〉 is an orthogonal locally permuta-
tion invariant ensemble. Then |ψ〉 can be transformed into |φ〉 by LOCC iff
(uk) ≺ (λk).
Proof. Let ρλ (resp. ρµ) denote Alice’s reduced state in |ψ〉 (resp. |φ〉). Pro-
ceeding analogously as above, the condition (uk) ≺ (λk) implies that
N−1∑k=0
µj|k〉〈k| =∑g∈SN
pgPg
(N−1∑k=0
λk|k〉〈k|
)P †g =
∑g∈SN
pgPgρλP†g , (6)
where the Pg are permutation matrices acting on the computational basis
and pg is some distribution. Define the set of Kraus operators Mg =
45
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
√pgρ
1/2λ P †g ρ
−1/2µ g∈SN . Then
Mg ⊗ I⊗ I|ψ〉 =√pg
N−1∑k=0
(ρ1/2λ P †g |k〉)|Ψk〉
=√pg
N−1∑k=0
(ρ1/2λ |σ
−1g (k)〉)|Ψk〉
=√pg
N−1∑k=0
ρ1/2λ |k〉|Ψσg(k)〉
=√pg
N−1∑k=0
√λk|k〉
n⊗j=1
U(g)j |Ψk〉
=√pg I
n⊗j=1
U(g)j |ψ〉. (7)
Necessity of the condition (uk) ≺ (λk) follows from the bipartite theory. ut
For the set of states which are all orthogonal locally permutation invariant
ensemble, we can use Theorem 14 to find its optimal common resource.
C. A non-trivial common resource of 3-qubit system: |GHZ3〉
We shall now move to multipartite setting, and consider the problem “what
is the common resource of the whole system”. For N -partite quantum system
d1⊗· · ·⊗dn where d1 ≥ d2 ≥ · · · ≥ dn, the maximal entangled state exists, in the
sense that all other states in the system can be obtained from the state by LOCC,
if and only if d1 ≥n∏i=2
di [8]. For instance, the state 12(|000〉+|101〉+|210〉+|311〉)
is an OCR of tripartite H4⊗H2⊗H2 system. Interestingly, the OCR exists even
46
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
if any sub-system’s dimension is less than the product of other sub-system’s
dimensions.
Theorem 18. |GHZ3〉 = 1√3(|000〉 + |111〉 + |222〉) is a common resource of
3-qubit system.
If |ψ〉 can be transformed into |φ〉 via LOCC, we denote this as
|ψ〉 LOCC−→ |φ〉.
If two pure states can be transformed into each one via stochastic local op-
erations and classical communication (SLOCC), they are in the same SLOCC
equivalence class, namely, they are SLOCC-equivalent. Three-qubit pure states
can be divided into product states, W-type and GHZ-type states [4]. GHZ
(1/√
2)(|000〉+ |111〉) state can be transformed into any product states.
Hence, we divide the proof into two cases according to the target states are
W-type or GHZ-type states. The case of GHZ can be further divide into two
sub-cases: orthogonal GHZ and non-orthogonal GHZ.
The following lemma has been shown in [31]. This also is our step i) in all
the cases.
Lemma 19. ∀z0, z1, z2,2∑
k=0
|zk|2 = 1,
|GHZ3〉LOCC−→ z0|000〉+ z1|111〉+ z2|222〉.
Proof. Alice takes the following measurement and sends the result to both Bob
and Charlie,
M1 = z0|0〉〈0|+ z1|1〉〈1|+ z2|2〉〈2|,
47
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
M2 = z0|0〉〈1|+ z1|1〉〈2|+ z2|2〉〈0|,M3 = z0|0〉〈2|+ z1|1〉〈0|+ z2|2〉〈1|.
The, Bob and Charlie make some unitary operations based on Alice’s mea-
surement outcome, which transform the state to z0|000〉+ z1|111〉+ z2|222〉.If Alice’s outcome is 1, Bob and Charlie do nothing.
If Alice’s outcome is 2, both Bob and Charlie take a unitary operation U =
|0〉〈1|+ |1〉〈2|+ |2〉〈0| on their own subsystems.
If Alice’s outcome is 1, Bob and Charlie take a unitary operation U = |0〉〈2|+|1〉〈0|+ |2〉〈1| on their own subsystems.
ut
a. Protocol of entanglement transformation from |GHZ3〉 to W-type states
If the target state |φ〉 is SLOCC–equivalent to W state, |φ〉 can be written as
x0|000〉+x1|100〉+x2|010〉+x3|001〉, where xk are all positive real numbers and3∑
k=0
x2k = 1 [25]. We can transform |GHZ3〉 to |φ〉 by three following steps:
i) |GHZ3〉LOCC−→
√x20 + x21|000〉+ x2|111〉+ x3|222〉,
ii)√x20 + x21|000〉+x2|111〉+x3|222〉 LOCC−→
√x20 + x21|100〉+x2|010〉+x3|001〉,
iii)√x20 + x21|100〉+ x2|010〉+ x3|001〉 LOCC−→ |φ〉.
Step i) is lemma 19 which z0 =√x20 + x21, z1 = x2 and z2 = x3.
Step ii) :√x20 + x21|000〉+ x2|111〉+ x3|222〉 LOCC−→
√x20 + x21|100〉+ x2|010〉+ x3|001〉
48
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
Alice takes the measurement:
M1 = (|1〉〈0|+ |0〉〈1|+ |0〉〈2|)/√
2, M2 = (|1〉〈0|+ |0〉〈2| − |0〉〈1|)/√
2.
Bob takes the measurement:
M1 = (|1〉〈1|+ |0〉〈0|+ |0〉〈2|)/√
2, M2 = (|1〉〈1|+ |0〉〈0| − |0〉〈2|)/√
2.
Charlie takes the measurement
M1 = (|1〉〈2|+ |0〉〈1|+ |0〉〈0|)/√
2, M2 = (|1〉〈2|+ |0〉〈1| − |0〉〈0|)/√
2.
Alice transmits her result to Bob.
Bob transmits his result to Charlie.
Charlie transmits its result to Alice.
If they get result 1, they continue. Otherwise, if the gotten result is 2, he or
she should take a Z−operation, Z = |0〉〈0| − |1〉〈1|.Now, the state is
√x20 + x21|100〉+ x2|010〉+ x3|001〉.
Step iii) :√x20 + x21|100〉+ x2|010〉+ x3|001〉 LOCC−→ |φ〉.
(Step iii) is also researched by Kintas and Turgut in 2010 [25].)
Alice takes the measurement:
M1 = 1√2(|0〉〈0|+ x1√
x20+x
21
|1〉〈1|+ x0√x2
0+x21
|0〉〈1|),
M2 = 1√2(|0〉〈0|+ x1√
x20+x
21
|1〉〈1| − x0√x2
0+x21
|0〉〈1|).Now, the state is (±x0|000〉+ x1|100〉+ x2|010〉+ x3|001〉).If Alice’s result is 1, we already get the target.
If it is 2, Alice transmits 2 to Bob and Charlie. Bob and Charlie make unitary
operation Z = |0〉〈0| − |1〉〈1| on their own part.
49
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
Then, Alice makes an unitary operation: −Z = −|0〉〈0|+ |1〉〈1|.Finally, the state is x0|000〉+ x1|100〉+ x2|010〉+ x3|001〉.
b. Protocol of entanglement transformation from |GHZ3〉 to GHZ-type
states
Without loss of generality, a GHZ-type pure states |φ〉 can be written
as x|000〉 + y|φAφBφC〉, where a0 = 〈0|φA〉, a1 = 〈1|φA〉, b0 = 〈0|φB〉, b0 =
〈0|φB〉, c0 = 〈0|φC〉, c0 = 〈0|φC〉. a0, a1, b0, b1, c0 and c1 are all real numbers.
GHZ-type pure states can be divided into two kinds: orthogonal or non-
orthogonal. |φ〉 is an orthogonal GHZ state, if a0b0c0 = 0. Otherwise, if a0b0c0 6=0, |φ〉 is an non-orthogonal GHZ state.
The LOCC protocols from |GHZ3〉 to orthogonal GHZ state can be divided
into two steps (suppose c0 = 0, |φC〉 = |1〉, thus, |x|2 + |y|2 = 1):
i) |GHZ3〉LOCC−→ |GHZ2〉,
ii) |GHZ2〉LOCC−→ |φ〉.
Step i) is for lemma 19 with z0 = z1 = 1/√
2 and z2 = 0. This is a protocol
|GHZ3〉LOCC−→ |GHZ2〉.
Step ii) can be farther divided as following:
(Step ii is also researched by Turgut, Gul and Pak in 2010 [24].)
|GHZ2〉LOCC−→ x|000〉+ y|111〉 LOCC−→ x|000〉+ y|φA11〉
50
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
LOCC−→ x|000〉+ y|φAφB1〉 = |φ〉.
For |GHZ2〉LOCC−→ (x|000〉+ y|111〉) :
Alice takes the measurement: M1 = x|0〉〈0|+y|1〉〈1|, M2 = x|1〉〈1|+y|0〉〈0|.If output is “2”, Alice take an X−operation, X = |1〉〈0|+ |0〉〈1|.If output is “1”, they continue.
For x|000〉+ y|111〉 LOCC−→ x|000〉+ y|φA11〉 :
Alice takes the measurement:
M1 =√22
(|0〉〈0|+ |φA〉〈1|),M2 =√22
(|0〉〈0| − |φA〉〈1|).If output is “2”, Alice transmits the result to Charlie and Charlie takes an
Z−operation, Z = |0〉〈0| − |1〉〈1|.If output is “1”, they continue.
The protocol for x|000〉 + y|φA11〉 LOCC−→ x|000〉 + y|φAφB1〉 is similar. Bob
takes the measurement: M1 =√22
(|0〉〈0|+ |φB〉〈1|), M2 =√22
(|0〉〈0| − |φB〉〈1|).If output is “1”, finish. If output is “2”, Bob transmits the result to Charlie and
Charlie takes an Z−operation, Z = |0〉〈0| − |1〉〈1|.
The LOCC protocols from |GHZ3〉 to non-orthogonal GHZ state can be
divided to 4 steps: (this is the most complicated part of the whole protocol)
i) |GHZ3〉LOCC−→ |Ψ1〉,
ii) |Ψ1〉LOCC−→ |Ψ2〉,
iii) |Ψ2〉LOCC−→ x|00φC〉+ y|φAφB0〉,
51
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
iv) x|00φC〉+ y|φAφB0〉 LOCC−→ x|000〉+ y|φAφBφC〉,where |Ψ1〉 =
√|xc0 + ya0b0|2 + |ya1b0|2 |000〉+ |xc1||111〉+ |yb1||222〉
and |Ψ2〉 =√|xc0 + ya0b0|2 + |ya1b0|2 |000〉+ |xc1||101〉+ |yb1||210〉.
Step i) is for lemma 19, with z0 =√|xc0 + ya0b0|2 + |ya1b0|2,
z1 = |xc1| and z2 = |yb1|.Step ii) |Ψ1〉
LOCC−→ |Ψ2〉:Bob takes the measurement:
M1 =√22
(|1〉〈2|+ |0〉〈0|+ |0〉〈1|),M2 =√22
(|1〉〈2|+ |0〉〈0| − |0〉〈1|).Charlie takes the measurement:
M1 =√22
(|1〉〈1|+ |0〉〈2|+ |0〉〈0|),M2 =√22
(|1〉〈1| − |0〉〈2|+ |0〉〈0|).They transmit their results (1 or 2) to Alice. Suppose Bob gets β and Charlie
gets α. Then, Alice takes the unitary operation
M = |0〉〈0| − (−1)β|1〉〈1| − (−1)α|2〉〈2|.Now, the state is |Ψ2〉.Step iii): Alice does the measurement
M1 = ( (xc0+ya0b0)|0〉+ya1b0|1〉√|xc0+ya0b0|2+|ya1b0|2
〈0|+ xc1|xc1| |0〉〈1|+
yb1|yb1|(a0|0〉+ a1|1〉)〈2| )/2,
M2 = ( (xc0+ya0b0)|0〉+ya1b0|1〉√|xc0+ya0b0|2+|ya1b0|2
〈0|+ −xc1|xc1| |0〉〈1|+
yb1|yb1|(a0|0〉+ a1|1〉)〈2| )/2,
M3 = ( (xc0+ya0b0)|0〉+ya1b0|1〉√|xc0+ya0b0|2+|ya1b0|2
〈0|+ xc1|xc1| |0〉〈1|+
−yb1|yb1| (a0|0〉+ a1|1〉)〈2| )/2,
M4 = ( (xc0+ya0b0)|0〉+ya1b0|1〉√|xc0+ya0b0|2+|ya1b0|2
〈0|+ −xc1|xc1| |0〉〈1|+
−yb1|yb1| (a0|0〉+ a1|1〉)〈2| )/2
The resulting states are all LOCC-equivalent to ((xc0 + ya0b0)|0〉 +
ya1b0|1〉)|00〉+ xc1|001〉+ yb1(a0|0〉+ a1|1〉)|10〉 = x|00φC〉+ y|φAφB0〉.It is also LOCC-equivalent to x|000〉 + y|φAφBφC〉, which can be done in next
step.
52
IV COMMON RESOURCE VIA LOCAL OPERATIONS AND CLASSICALCOMMUNICATION
Step iv): Charlie takes a local unitary operation: |φC〉〈0|+ (c1|0〉 − c0|1〉)〈1|.Finally, the state is x|000〉+ y|φAφBφC〉.
In conclusion, we introduce a notion of optimal common resource for a set of
entangled states, and explicitly construct it for any bipartite pure state set. We
also show that |GHZ3〉 state is a nontrivial common resource for 3-qubit sys-
tem, and conjecture its optimality. We hope this problem will stimulate further
research interest in entanglement transformation theory.
53
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
V. COMMON RESOURCE VIA STOCHASTIC LOCAL
OPERATIONS AND CLASSICAL COMMUNICATION
Let S = |ψ1〉, |ψ2〉, · · · be a set of (multipartite) states, possibly infinite. A
common resource state |ψ〉 to S is the state which can be transformed into any
state in S by SLOCC. We say |ψ〉 an optimal common resource (OCR) if for any
other common resource |φ〉 we have either |φ〉 can be transformed into |ψ〉 by
SLOCC, or |φ〉 and |ψ〉 are not comparable under SLOCC.
Without loss of generality, a⊗b denotes b subsystems with dimension a. Let
me ignore unimportant normalization coefficients because we will only talk about
SLOCC protocols in this section. We find that some 3 ⊗ 2⊗3 state is optimal
for two four-qubit symmetric states. For two general N -qubit states, we need a
4⊗ 2⊗N−1 state or a 3⊗N−2 ⊗ 2⊗ 2 state.
In fact, if two N -qubit pure states |ψ1〉 and |ψ2〉 can be SLOCC transformed
by the same state |ψ0〉 and state |ψ0〉 is a pure state with local rank 3⊗ 2⊗N−1,
the intersection between their support in H⊗N−12 space should not be empty.
Theorem 20. The common resource |ψ0〉 of two N−qubit pure states |ψ1〉 and
|ψ2〉 is with local rank 3⊗t ⊗ 2⊗N−t if and only if there exists a (N − t)-qubit |φ〉which is SLOCC-equivalent to a state in the support of |ψ1〉 and also to a state in
the support of |ψ2〉, where partial trace of these two supports are taken in those
H⊗t3 subsystems.
Proof. If the common resource |ψ0〉 of two N -qubit pure states |ψ1〉 and |ψ2〉is with local rank 3⊗t ⊗ 2⊗N−t, without loss of generality, suppose |ψ1〉 =⊗t
j=1Ej1|ψ0〉 and |ψ2〉 =⊗t
j=1Ej2|ψ0〉. Because |ψ1〉 and |ψ2〉 are two N -qubit
55
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
pure states. All Ej0 and Ej1 are rank 2. In H3 system, for any two rank 2 matri-
ces Ej0 and Ej1, there always exists |µj〉 and |νj〉 such that E†j1|µj〉 = E†j2|νj〉 6= 0
(notice that we ignore unimportant positive coefficients before E†j1|µj〉).
(t⊗
j=1
〈µj|)|ψ1〉 = (t⊗
j=1
〈µj|Ej1)|ψ0〉 = (t⊗
j=1
〈νj|Ej2)|ψ0〉 = (t⊗
j=1
〈νj|)|ψ2〉. (8)
Denote this (N−t)-qubit entangled state as |φ〉. |φ〉 is in the support |ψ1〉 and
also in the support of |ψ2〉, where partial trace of these two supports are taken
in those H⊗t3 subsystems.
The proof of another direction is as following.
For convenience of reading, use |j〉 to denote a t-qubits product state. The
number m at the k-th bit of j’s binary form means |m〉 on the k-th qubit sub-
system of |j〉 respectively. For example, when t = 4, |0〉 = |0000〉, |6〉 = |0110〉and |13〉 = |1101〉.
For two N−qubit pure states |ψ1〉 and |ψ2〉, if there exists a (N − t)-qubit |φ〉which is SLOCC-equivalent to a state in the support of |ψ1〉 and also to a state in
the support of |ψ2〉, where partial trace of these two supports are taken in those
H⊗t2 subsystems, without loss of generality, suppose |ψ1〉 = |0〉|φ〉+∑2t−1
j=1 |j〉|ψ1,j〉and |ψ2〉 = |0〉|φ〉 +
∑2t−1j=1 |j〉|ψ2,j〉, where |ψ1,j〉 and |ψ2,j〉 are all (N − t)-qubit
states.
Suppose E = |0〉〈0| + |2〉〈1|. Put operator E⊗t in the front t subsystems onto
|ψ2〉. We are going to prove that |ψ0〉 = |ψ1〉 + E⊗t ⊗ I⊗N−t|ψ2〉 − |0〉|φ〉 is a
common resource of two N−qubit pure states |ψ1〉 and |ψ2〉, alos |ψ0〉 with local
rank 3⊗t ⊗ 2⊗N−t.
56
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
Suppose A1 = |0〉〈0| + |1〉〈1| and A2 = |0〉〈0| + |1〉〈2|. A⊗t1 and A⊗t2 are acted
on those rank 3 subsystems.
We finish the whole proof after checking |ψ1〉 = A⊗t1 ⊗ I⊗N−t|ψ0〉 and
|ψ2〉 = A⊗t2 ⊗ I⊗N−t|ψ0〉. ut
For any two N -qubit pure states |ψ0〉 and |ψ1〉, one can always find some
4 ⊗ 2⊗n−1 state as their common resource, because |ψ0〉 + M ⊗ I⊗N−1|ψ1〉 is
their 4⊗ 2⊗n−1 common resource state, where M = |2〉〈0|+ |3〉〈1|. We can find a
different recourse: a 3⊗m⊗ 2n−m state. We show our research case by case based
on number of qubits.
Any two 2-qubit entangled pure states are SLOCC-equivalent.
A. Tripartite entangled pure states
There are 25 SLOCC equivalence classes in 2⊗ 3⊗ 6 system. This result is in
Ref. [44]. Because of the property of SLOCC, without loss of generality, we use
states to denote their equivalence class and ignore the unimportant coefficients
of these states.
Product state with local rank 1 ⊗ 1 ⊗ 1 are all SLOCC-equivalent, namely,
|000〉. In 1⊗2⊗2 system, there exists a SLOCC-equivalence class: |000〉+ |011〉.Similar for 2⊗ 1⊗ 2 system and 2⊗ 2⊗ 1 system.
In 2⊗ 2⊗ 2 system, there exist two more SLOCC-equivalence classes:
GHZ: |000〉+ |111〉 (ABC − 1) and W: |001〉+ |010〉+ |100〉 (ABC − 2).
In 2⊗ 2⊗ 3 system, there exist two more SLOCC-equivalence classes: |000〉+|011〉+ |102〉 (ABC − 3) and |000〉+ |011〉+ |101〉+ |112〉 (ABC − 4).
57
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
In 2⊗ 2⊗ 4 system, there exists one more SLOCC-equivalence class:
|000〉+|011〉+|102〉+|113〉 (ABC−5) which is so called the maximally entangled
state in 2⊗ 2⊗ 4 system.
In 2⊗ 3⊗ 2 system, there exist two more SLOCC-equivalence classes: |000〉+|011〉+ |120〉 (ABC − 6) and |000〉+ |011〉+ |101〉+ |121〉 (ABC − 7).
In 2⊗ 3⊗ 3 system, there exist six more SLOCC-equivalence classes:
|000〉+ |011〉+ |122〉 (ABC − 9),
|000〉+ |011〉+ |111〉+ |122〉 (ABC − 8),
|000〉+ |011〉+ |022〉+ |101〉 (ABC − 10),
|001〉+ |100〉+ |111〉+ |022〉+ |122〉 (ABC − 11),
|001〉+ |012〉+ |100〉+ |111〉+ |122〉 (ABC − 12) and
|000〉+ |011〉+ |102〉+ |121〉 (ABC − 13).
In 2⊗ 3⊗ 4 system, there exist five more SLOCC-equivalence classes:
|000〉+ |011〉+ |022〉+ |103〉 (ABC − 14),
|001〉+ |100〉+ |112〉+ |123〉+ |023〉 (ABC − 15),
|001〉+ |013〉+ |100〉+ |112〉+ |123 (ABC − 16)〉,|001〉+ |012〉+ |100〉+ |111〉+ |123〉 (ABC − 17) and
|001〉+ |012〉+ |023〉+ |100〉+ |111〉+ |122〉 (ABC − 18).
In 2⊗ 3⊗ 5 system, there exist two more SLOCC-equivalence classes:
|001〉+ |013〉+ |100〉+ |112〉+ |124〉 (ABC − 19) and
|001〉+ |013〉+ |024〉+ |100〉+ |112〉+ |123〉 (ABC − 20).
In 2⊗ 3⊗ 6 system, there exists one more SLOCC-equivalence class:
|000〉 + |011〉 + |022〉 + |103〉 + |114〉 + |125〉 (ABC − 21) which is so called the
maximally entangled state in 2⊗ 3⊗ 6 system.
58
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
The relationship of them is as following:
Figure 5.1. Partial order relation of SLOCC-equalvence classes in 2 ⊗ 3 ⊗ 6
system [44].
59
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
In following discussion, a m ⊗ n ⊗ p state |ψ〉, m,n, p ≥ 2, denotes the
tripartite state with local rank m,n and p. By checking the above map, we can
prove the following lemmas.
Lemma 21.
1) For any 2⊗ 2⊗ 3 entangled pure state |ψ〉 and any pure states |φ〉 in 3-qubit
system, the SLOCC protocol from |ψ〉 to |φ〉 always exist.
2) Any 2 ⊗ 2 ⊗ 3 entangled pure states can be transformed from a 2 ⊗ 3 ⊗ 3
state in the following SLOCC−equivalence class: |000〉 + |011〉 + |111〉 + |122〉,|000〉+ |011〉+ |022〉+ |101〉+ |122〉 and |000〉+ |011〉+ |022〉+ |101〉+ |112〉.3) Any 2⊗ 3⊗ 3 entangled pure states can be transformed from a 2⊗ 3⊗ 5 state
in the following SLOCC−equivalence class: |001〉+ |013〉+ |100〉+ |112〉+ |124〉and |001〉+ |013〉+ |024〉+ |100〉+ |112〉+ |123〉.
No 2⊗ 3⊗ 4 state is a common resource for 2⊗ 3⊗ 3 system.
4) Any 2 ⊗ 3 ⊗ 4 entangled pure states can be transformed from a 2 ⊗ 3 ⊗ 6
maximally entangled state.
No 2⊗ 3⊗ 5 state is a common resource for 2⊗ 3⊗ 4 system.
Proof. 1) we see both ABC − 3 and ABC − 4 can be transformed to either
ABC − 1 or ABC − 2 via SLOCC.
2) ABC−8, ABC−11 and ABC−12 can be transformed to either ABC−3
or ABC − 4 via SLOCC.
3) Neither ABC − 14 nor ABC − 16 can be transformed into ABC − 8 via
SLOCC. Neither ABC − 15 nor ABC − 18 can be transformed into ABC − 10
via SLOCC. ABC − 17 cannot be transformed into ABC − 13 via SLOCC.
We notice that if a state |ψ〉 can be transformed into ABC−15 and ABC−16
60
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
via SLOCC, |ψ〉 will be a common resource for all 2⊗ 3⊗ 3 pure state. Hence,
ABC − 19 and ABC − 20 can be transformed to any 2⊗ 3⊗ 3 states.
4) 6 = 2 × 3, so ABC − 21 is a maximally entangled state. Of cause, it is a
common resource for 2⊗ 3⊗ 4 system.
ABC − 19 cannot be transformed into ABC − 18 via SLOCC and ABC − 20
cannot be transformed into ABC − 14 via SLOCC. Hence, no 2⊗ 3⊗ 5 state is
a common resource for 2⊗ 3⊗ 4 system. ut
In fact, we also have the following lemmas.
Lemma 22. A tripartite entangled pure state |φ〉 can be transformed to a 2⊗2⊗3,
2⊗ 3⊗ 2 or 3⊗ 2⊗ 2 entangled pure state via SLOCC, if and only if a local rank
of |φ〉 is at least 3.
Proof. Suppose local rank of the third subsystem in |φ〉 is at least 3, denoted by
rC(|φ〉) ≥ 3. Take the Schmidt decomposition on AB − C separation:
|φ〉 =∑
j |φj〉AB|φj〉C , where |φj〉C are orthogonal to each other and |φj〉AB are
also orthogonal to each other. As rA(|φ〉) ≥ 2 and rB(|φ〉) ≥ 2, without loss of
generality, we can suppose∑3
j=1 |φj〉AB|φj〉C , denoted by |φ〉, is the state which
local ranks are all at least 2 and rC(|φ〉) ≥ 3.
We can also take the Schmidt decomposition on A−BC or B−AC separation.
Hence, we can suppose the local ranks of the tripartite entangled pure state |φ〉is at most 3.
If |φ〉 is already a 2 ⊗ 2 ⊗ 3, 2 ⊗ 3 ⊗ 2 or 3 ⊗ 2 ⊗ 2 entangled pure state, we
finish the proof. Because of relationship in SLOCC equivalence class of 2⊗ 3⊗ 6
system (section“Previous result”), if a 2⊗ 3⊗ 3, 3⊗ 2⊗ 3 or 3⊗ 3⊗ 2 entangled
pure state, we also finish the proof [44].
61
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
If |φ〉 is a 3⊗ 3⊗ 3 entangled pure state, take the Schmidt decomposition on
AB−C separation: |φ〉 =3∑j=1
|φj〉AB|φj〉C , rk(|φ1〉AB) ≥ rk(|φ2〉AB) ≥ rk(|φ3〉AB).
If rk(|φ1〉AB) = 3, take |1〉〈1| + |2〉〈2| on its third subsystem. The result will
be a 3⊗ 3⊗ 2 state. We can transform it into a 3⊗ 2⊗ 2 state [44].
If rk(|φ1〉AB) = 2, then suppose |φ1〉AB = |0〉A|0〉B + |1〉A|1〉B. rB(|φ〉) = 3,
so 〈2|B|φ2〉 6= 0 or 〈2|B|φ3〉 6= 0. Suppose 〈2|B|φ2〉 6= 0. Take |1〉〈1|+ |2〉〈2| on its
third subsystem. The result will be a 3 ⊗ 3 ⊗ 2 or 2 ⊗ 3 ⊗ 2 state. No matter
what the state is, it can be transformed it into a 2⊗ 3⊗ 2 state via SLOCC [44].
If rk(|φ1〉AB) = 1, then rk(|φ2〉AB) = rk(|φ3〉AB) = 1. |φ〉 is SLOCC-equivalent
to |000〉+ |111〉+ |222〉. Take |1〉〈1|+ |2〉〈2| on its third subsystem. The result will
be a 3⊗ 3⊗ 2 state. It can be transformed it into a 2⊗ 3⊗ 2 state via SLOCC
[44]. ut
A tripartite entangled pure state is a tripartite pure state which any of its
local rank is at least 2. This lemma implies the following theorem.
Theorem 23. A tripartite entangled pure state can be SLOCC transformed into
any pure state in 3-qubit system, if and only if one of its local ranks is at least 3.
Proof. Three qubit can be entangled in two different ways, which is GHZ-type
states and W-type states [4]. Namely, they are in the different SLOCC equiva-
lence classes. Hence, in 3-qubit system, there is no common resource of the whole
system. If a tripartite entangled pure state can be SLOCC transformed into any
pure state in 3-qubit system, one of its local ranks is at least 3. Otherwise, it will
be in 3-qubit system, which will cause contradiction.
If one of a tripartite entangled pure state’s local ranks is at least 3, it can be
62
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
transformed into a 2 ⊗ 2 ⊗ 3, 2 ⊗ 3 ⊗ 2 or 3 ⊗ 2 ⊗ 2 entangled pure state via
SLOCC by Lemma 22. Any one of these three type states can be transformed
into any pure state in 3-qubit system by Lemma 21.
ut
B. Multi-partite qubit pure states
For 4-qubit case, we should use a 4⊗ 2⊗ 2⊗ 2 or a 3⊗ 3⊗ 2⊗ 2 state.
Not all two 4-qubit states can transformed by a 3⊗ 2⊗3. Consider |GHZ4〉 =
|0000〉 + |1111〉 and |ψ〉 = |0000〉 + |1100〉 + |0110〉 + |0011〉 + |1001〉. They
cannot be transformed from a 3⊗ 2⊗ 2⊗ 2 state via SLOCC. The reason is, any
3 subsystems’ support of |GHZ4〉 only has product states or GHZ-type states.
However, in any 3-qubit support of |ψ〉 doesn’t have product states or |GHZ3〉, i.
e. span|000〉+ |110〉+ |011〉, |100〉+ |001〉 has no product states or GHZ-type
states.
Based on 4-qubit example, we have examples for more general case.
Suppose |ψ1〉 = |0〉⊗N−3|W 〉+|1〉⊗N−3|W ′〉 and |ψ2〉 = |GHZN〉, where |W ′〉 =
a|100〉+ b|010〉+ c|001〉 and (a− b)(b− c)(c− a) 6= 0. In the last 3 subsystems,
the support of |ψ1〉 only contains W -type and 1 ⊗ 2 ⊗ 2 (also, 2 ⊗ 1 ⊗ 2 and
2⊗2⊗1 ) states. However, the support of |ψ2〉 in the last 3 subsystems contains
|GHZ〉-type states and product states. Hence, the common result of these two
states is not with local rank 3⊗N−32⊗3.
This is a good example for the following theorem:
Theorem 24. No 3⊗N−32⊗3 state is a common resource for the whole N-qubit
63
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
system.
Hence, the common resource in kind 3⊗a2⊗N−a of N -qubit system should have
the property a ≥ N − 2. Before checking whether equality can be achieved, we
introduce a lemma firstly.
Lemma 25. [45] There always exists a product state in the span of any 2-qubit
pure states.
Therefore, we have have this following theorem:
Theorem 26. For any two given N-qubit pure states, one of their SLOCC com-
mon resources is a pure state with local rank 3⊗N−22⊗2.
Before jumping into the disscusion of symmetric states, let us review a useful
tool to check whether a 3-qubit state is a GHZ-type state. A 3-qubit pure state
|ψ〉 can be written as∑7
k=0 ak|k〉, where |k〉 = |jAjBjC〉, k = 4jA + 2jB + jC is
the value of a binary number (jAjBjC).
Theorem 27. [23] A 3-qubit pure state |ψ〉 =∑7
k=0 ak|k〉 is a GHZ-type state,
if and only if (a0a7 − a2a5 + a1a6 − a3a4)2 − 4(a2a4 − a0a6)(a3a5 − a1a7) 6= 0.
We can prove the following lemma directly by using Theorem 27.
Lemma 28. If any two symmetric W -type states are not SLOCC equivalent,
there exists at least one other type pure states in the span of these two states.
Proof. (By contradiction.) Denote two SLOCC in-equivalent symmetric W -type
states as |ψ0〉 and |ψ1〉. By the symmetric property, there exists operator E =
EA = EB = EC , such that EA⊗EB ⊗EC |ψ1〉 = |W 〉. Suppose E ⊗E ⊗E|ψ0〉 =
64
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
b0|000〉+ b1(|001〉+ |010〉+ |100〉) + b2(|011〉+ |101〉+ |110〉) + b3|111〉. According
to Theorem 27 [23], b20b23 + 4b0b
32 + 4b31b3 − 6b0b1b2b3 − 3b21b
22 = 0. Hence, we also
have:
∀λ, b20b23 + 4b0b32 + 4(b1 + λ)3b3 − 6b0(b1 + λ)b2b3 − 3(b1 + λ)2b22 = 0.
If this lemma is not true, b3 = b2 = 0.
E ⊗ E ⊗ E|ψ0〉 = b0|000〉+ b1(|001〉+ |010〉+ |100〉).
This means a product state is in their span or ∃b1, |ψ0〉 = b1|ψ1〉.However, the second possibility is impossible because it is a contradiction
to |ψ0〉 and |ψ1〉 are not SLOCC equivalent. The first possibility also causes a
contradiction.
Therefore, this lemma is true. ut
Now, we can prove a property of N -qubit symmetric entangled pure state
case:
Theorem 29. When N > 3, any 3-qubit support of an N-qubit symmetric en-
tangled pure state cannot just have W -type state in it.
Proof. (By contradiction.) There are at least two states in any 3-qubit support of
an entangled N -qubit pure state. Suppose they are both W -type states, denoted
as |ψ0〉 and |ψ1〉. Because of Lemma 28, in the span of |ψ0〉 and |ψ1〉, there exists
at least one other type pure states in the span of these two states. This is a
contraction. ut
In 3-qubit symmetric system, there are only 3 SLOCC-equavelence classes:
W , GHZ and product states. Hence, any two 4-qubit symmetric entangled
65
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
states will have a common SLOCC-equavelence state in their 3-qubit subsystems’
support by Theorem 29. This means if the optimal common resource of any two
given N -qubit symmetric states is in the system with local rank 3⊗t ⊗ 2N−t, t is
at most N − 3.
Before extending this property to general cases, let’s introduce a lemma.
Lemma 30. For any t < [N−12
], the tensor rank of any state in the following set
span (Psym(|0〉⊗[N2 ]−t, |1〉⊗[N+12
]), Psym(|0〉⊗[N2 ]−t+1, |1〉⊗[N+12
]−1), · · · ,Psym(⊗|0〉[N2 ], |1〉⊗[N+1
2]−t) is at least 3.
Proof. When t < [N−12
], t ≤ [N−32
], so [N+12
] − t ≥ 2 and [N2
] − t ≥ 1. Hence,
|0〉⊗n and states with kind of |0〉n−1|1〉 are not in this set. Hence, any state
|φ〉 in this set can be written as[(N−3)/2]∑
j=2
λjPsym(|0〉⊗[N2 ]−j, |1〉⊗j). This means
〈0|⊗n|φ〉 = 〈1|⊗n|φ〉 = 0, and for any kind of |0〉n−1|1〉 state |ψ〉, 〈ψ|φ〉 = 0.
We have: rk(|φ〉) 6= 1, otherwise |φ〉 = |µ〉⊗n. However, 〈0|⊗n|φ〉 = 0 =
〈1|⊗n|φ〉 means 〈0|µ〉 = 〈1|µ〉 = 0. This is a contradiction with |µ〉 6= 0.
We also have: rk(|φ〉) 6= 2, otherwise, because 〈1|⊗n|φ〉 = 0, without loss of
generality, we suppose |φ〉 =⊗n
k=1(ak|0〉+ |1〉)−⊗n
k=1(bk|0〉+ |1〉). However, for
any kind of |0〉n−1|1〉 state |ψ〉, 〈ψ|φ〉 = 0. This means for any k, ak = bk. It is a
contradiction with |φ〉 6= 0.
Hence, rk(|φ〉) ≥ 3.
ut
Theorem 31. If an N-qubit symmetric system’s optimal common resource is a
kind of 3⊗t ⊗ 2⊗N−t state, t ≥ [N−12
].
66
V COMMON RESOURCE VIA STOCHASTIC LOCAL OPERATIONSAND CLASSICAL COMMUNICATION
Proof. This is more general. Let us check |GHZn〉 = |0〉⊗n + |1〉⊗n and
Psym(|0〉⊗[N2 ], |1〉⊗[N+12
]).
When t < [N−12
], 2⊗N−t support of |GHZn〉 just contains product states and
|GHZN−t〉 state which is a rank 2 state.
However, 2⊗N−t support of Psym(|0〉⊗[N2 ], |1〉⊗[N+12
]) is span (Psym(|0〉⊗[N2 ]−t,
|1〉⊗[N+12
]), Psym(|0〉⊗[N2 ]−t+1, |1〉⊗[N+12
]−1), · · · , Psym(⊗|0〉[N2 ], |1〉⊗[N+12
]−t).
The rank of all the state in this set is at least 3 by Lemma 30. Hence,
|GHZn〉 = |0〉⊗n + |1〉⊗n and Psym(|0〉⊗[N2 ], |1〉⊗[N+12
]) do not have a common
3⊗t ⊗ 2⊗N−t resource.
ut
In summary, we show that two 4-qubit pure symmetric states can be obtained
from some 3⊗2⊗2⊗2 state by stochastic local operations and classical commu-
nication. For two given N -qubit pure states, we can always find some 3⊗N−22⊗2
state is their resource. The number N − 2 is optimal in the condition that we
can just expand the dimensions of the spaces one by one. We also show some
properties of optimal resource of two N -qubit pure states.
67
VI CONCLUSION
VI. CONCLUSION
We introduce a notion of entanglement transformation rate to characterize
the asymptotic comparability of two multi-partite pure entangled states under
stochastic local operations and classical communication (SLOCC). For two well
known SLOCC inequivalent three-qubit states GHZ and W, we show that the
entanglement transformation rate from GHZ to W is exactly 1. That means that
we can obtain one copy of W-state, from one copy of GHZ-state by SLOCC,
asymptotically. We then apply similar techniques to obtain a lower bound on
the entanglement transformation rates from an N-partite GHZ-state to a class
of Dicke states, and prove the tightness of this bound for some special cases
that naturally generalize the W state. A new lower bound on the tensor rank
of matrix permanent is also obtained by evaluating the the tensor rank of Dicke
states.
It is completely solved that how to find an optimal common resource for
bipartite pure states case by explicitly constructing a unique optimal common
resource state for any given set of states via LOCC. In the multi-partite set-
ting, the general problem becomes quite complicated, and we focus on finding
non-trivial common resources for the whole multi-partite state space of given di-
mensions. We show that |GHZ3〉 = (1/√
3)(|000〉+ |111〉+ |222〉) is a non-trivial
common resource for 3-qubit systems via LOCC.
We also show some properties of the non-trivial common resource of two N -
qubit pure states, N -qubit systems, and symmetric systems via SLOCC.
The most interesting open problem is what is the optimal common resource
of 3-qubit system via LOCC. Is it with local rank 2⊗ 2⊗ 3 or 2⊗ 3⊗ 3? As we
69
VI CONCLUSION
know, any common resource of 3-qubit system via LOCC has at least one local
rank not less than 3. However, we have not found any state, with local rank
2⊗ 2⊗ 3 or 2⊗ 3⊗ 3, which is a common resource of 3-qubit system via LOCC.
We are also interested in a conjecture put forward by Comon et al.: for an
arbitrary symmetric tensor, is its symmetric tensor rank always the same as
its tensor rank [30]? This has been an open problem for many years. Many
various special cases have been studied, and confirmed the above conjecture. In
bipartite cases, it is true that symmetric tensor rank is always equal to tensor
tank, by a smart application of the Schmidt decomposition. We find neither the
counterexamples nor the proof in the multi-partite cases.
Another open problem is to evaluate the exact value of the tensor rank of
|W 〉⊗n for n ≥ 3. We have found a few lower bound 2n+1 − 1 and an upper
bound (n + 1)2n. Unfortunately, no definite result is known when n > 2.This
problem has been studied extensively, and a number of partial results have been
obtained. It relates to the problem so-call polynomial rank of homogeneous poly-
nomials, because |W 〉 is a symmetric pure state. For an a degree 3 homogeneous
polynomials, we can solve some special cases [33] [34]. If someday we can solve
the polynomial rank problem completely, even just for degree three, we may
know the answer.
70
VI CONCLUSION
References
[1] E. Schrodinger, The Present Situation in Quantum Mechanics. Naturwis-
senschaften (1935).
[2] M. Nielsen, Conditions for a Class of Entanglement Transformations, Phys. Rev.
Lett. 83, 436 (1999).
[3] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Informa-
tion, Cambridge University Press (2000).
[4] W. Dur, G. Vidal, and J. I. Cirac, Three qubits can be entangled in two inequiv-
alent ways, Phys. Rev. A 62, 062314 (2000).
[5] E. Chitambar, R. Duan, and Y. Shi, Tripartite Entanglement Transformations
and Tensor Rank, Phys. Rev. Lett. 101, 140502 (2008).
[6] N. Yu, E. Chitambar, C. Guo, and R. Duan, Tensor rank of the tripartite state
|W 〉⊗n Phys. Rev. A 81, 014301 (2010).
[7] L. Chen, E. Chitambar, R. Duan, Z. Ji, and A. Winter, Tensor rank and stochastic
entanglement catalysis for multipartite pure states, Phys. Rev. Lett. 105, 200501
(2010).
[8] R. Duan and Y. Shi, When is there a multipartite maximum entangled state?
Quantum Information and Computation 10 (11): 925–935 (2010).
[9] J. Eisert and H. J. Briegel, Schmidt measure as a tool for quantifying multiparticle
entanglement, Phys. Rev. A 64, 022306 (2001).
[10] V. Strassen, Gaussian Elimination is not Optimal, Numer. Math. 13, 354 (1969).
71
VI CONCLUSION
[11] V. Strassen, Rank and optimal computation of generic tensors, Lin. Algebra Appl.
52/53, pp. 645-685 (1983).
[12] C. H. Bennett, H. J. Bernstein, S. Popescu, B. Schumacher, Concentrating partial
entanglement by local operations, Phys. Rev. A 53, 2046 (1996).
[13] C. H. Bennett, S. Popescu, D. Rohrlich, J. A. Smolin, A.V. Thapliyal, Exact and
asymptotic measures of multipartite pure-state entanglement, Phys. Rev. A 63,
012307 (2000).
[14] T. A. Brun, O. Cohen, Parametrization and distillability of three-qubit entangle-
ment, Phys. Lett. A 81(2-3), 88-100 (2001).
[15] J. Kempe, Multiparticle entanglement and its applications to cryptography, Phys.
Rev. A 60, 910 (1999).
[16] R. Duan, Y. Feng, and M. Ying, Entanglement-assisted transformation is asymp-
totically equivalent to multiple-copy transformation, Phys. Rev. A 72, 024306
(2005).
[17] N. Yu, Multipartite W-type state is determined by its single-particle reduced
density matrices among all W-type states Phys. Rev. A 87, 052310 (2013).
[18] N. Yu, C. Guo and R. Duan, Obtaining a W State from a Greenberger-Horne-
Zeilinger State via Stochastic Local Operations and Classical Communication with
a Rate Approaching Unity, Phys. Rev. Lett. 112, 160401 (2014).
[19] F. Verstraete, J. Dehaene, B. De Moor, and H. Verschelde, Four qubits can be
entangled in nine different ways, Phys. Rev. A 65, 052112 (2002).
[20] Z. Ji, R. Duan and M. Ying, Comparability of multipartite entanglement, Phys.
Let. A 330, 418-423 (2004).
[21] M. Hein, J. Eisert, and H. J. Briegel, Multiparty entanglement in graph states,
72
VI CONCLUSION
Phys. Rev. A 69, 062311 (2004).
[22] G. Toth and O. Guhne, Entanglement and Permutational Symmetry, Phys. Rev.
Lett 102, 170503 (2009).
[23] D. Li, X. Li, H. Huang and X. Li, The Simple Criteria of SLOCC Equivalence
Classes, Phys. Lett. A 359, 428- 437 (2006).
[24] S. Turgut, Y. Gul and N. K. Pak, Deterministic transformations of multipartite
entangled states with tensor rank 2, Phys. Rev. A 81, 012317 (2010).
[25] S. Kintas and S. Turgut, Transformations of W -Type Entangled States, J. Math.
Phys. 51, 092202 (2010).
[26] H. Tajima, Deterministic LOCC transformation of three-qubit pure states and
entanglement transfer, Annals of Physics 329, 1-27 (2013).
[27] G. Gour, and N. R. Wallach, Classification of Multipartite Entanglement of All
Finite Dimensionality, Phys. Rev. Lett. 111, 060502 (2013).
[28] R. Hubener, M. Kleinmann, T-C Wei, C. Gonzalez-Guillen, and O. Guhne, Ge-
ometric measure of entanglement for symmetric states, Phys. Rev. A 80, 032324
(2009).
[29] D. Jonathan and M.B. Plenio, Minimal Conditions for Local Pure-State Entan-
glement Manipulation, Phys. Rev. Lett. 83, 1455 (1999).
[30] P. Comon, G. Golub, L. Lim, and B. Mourrain, Symmetric tensors and symmetric
tensor rank, SIAM. J. Matrix Anal. & Appl., 30(3), 12541279 (2008).
[31] Y. Xin and R. Duan, Conditions for entanglement transformation between a class
of multipartite pure states with generalized Schmidt decompositions, Phys. Rev.
A 76, 044301 (2007).
[32] J. M. Landsberg and Z. Teitler, On the ranks and border ranks of symmetric
73
VI CONCLUSION
tensors, Foundations of Computational Mathematics, 10 (3), 339366 (2010).
[33] E. Carlini, M. V. Catalisano, A. V. Geramita, The solution to the Waring problem
for monomials and the sum of coprime monomials, Journal of Algebra 370, 5-14
(2012).
[34] E. Carlini, C. Guo and E. Ventura, Complex and real ranks of reducible cubics,
http://arxiv.org/abs/1502.01051 (2015).
[35] D. Bini, Relations between exact and approximate bilinear algorithms. Applica-
tions, Calcolo Gennaio-Marzo, 17 (1), 87-97 (1980).
[36] P. Burgisser, M. Clausen, M. A. Shokrollahi, Algebraic complexity theory.
Springer-Verlag New York Incorporated, 1997.
[37] J. Haastaad, Tensor Rank is NP-complete, Journal of Algorithms, 11, 644-654
(1990).
[38] R. Raz, Multi-Linear Formulas for Permanent and Determinant are of Super-
Polynomial Size, Journal of the Association for Computing Machinery, 56(2)
(2009).
[39] R. Raz, and Yehudayoff, Lower Bounds and Separations for Constant Depth Mul-
tilinear Circuits, Computational Complexity, 18(2), 171-207 (2009).
[40] R. Raz, Tensor Rank and Lower Bounds for Arithmetic Formulas, Journal of the
Association for Computing Machinery, 60(6) (2013).
[41] T. Bastin, S. Krins, P. Mathonet, M. Godefroid, L. Lamata, and E. Solano, Op-
erational Families of Entanglement Classes for Symmetric N-Qubit States, Phys.
Rev. Lett. 103, 070503 (2009).
[42] T. Bastin, C. Thiel, J. vonZanthier, L. Lamata, E. Solano, and G.S. Agarwal,
Operational Determination of Multiqubit Entanglement Classes via Tuning of
74
VI CONCLUSION
Local Operations, Phys. Rev. Lett. 102, 053601 (2009).
[43] W. Wieczorek, N. Kiesel, C. Schmid, and H. Weinfurter, Multiqubit entanglement
engineering via projective measurements, Phys. Rev. A 79, 022311 (2009).
[44] E. Chitambar, C. Miller and Y. Shi, Matrix Pencils and Entanglement Classifica-
tion, J. Math. Phys. 51, 072205 (2010) .
[45] A. Sanpera, R. Tarrach, and G. Vidal, Local description of quantum inseparability,
Phys. Rev. A 58, 826 (1998).
[46] C. Guo, E. Chitambar, and R. Duan, Common Resource State for Preparing Mul-
tipartite Quantum Systems via Local Operations and Classical Communication,
In preparation (2015).
[47] C. Guo, and R. Duan, Common Resource of Pure States via Stochastic Local
Operations and Classical Communication, In preparation (2015).
75